Health in Civil Engineering

Mohammed M.
Ettouney • Sreenivas Alampalli
Infrastructure Health
in Civil Engineering
Theory and Components
Volume I
Volume I
Mohammed M. Ettouney
Sreenivas Alampalli
Cover image courtesy of New York State Department of Transportation.
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
© 2012 by Taylor & Francis Group, LLC

CRC Press is an imprint of Taylor & Francis Group, an Informa business
No claim to original U.S. Government works

Version Date: 20110815
International Standard Book Number-13: 978-1-4200-0375-8 (eBook - PDF)
This book contains information obtained from authentic and highly regarded sources. Reasonable efforts have been made to
publish reliable data and information, but the author and publisher cannot assume responsibility for the validity of all materials
or the consequences of their use. The authors and publishers have attempted to trace the copyright holders of all material repro-
duced in this publication and apologize to copyright holders if permission to publish in this form has not been obtained. If any
copyright material has not been acknowledged please write and let us know so we may rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or utilized in any
form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying, microfilming,
and recording, or in any information storage or retrieval system, without written permission from the publishers.
For permission to photocopy or use material electronically from this work, please access www.copyright.com (http://www.copy-
right.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400.
CCC is a not-for-profit organization that provides licenses and registration for a variety of users. For organizations that have been
granted a photocopy license by the CCC, a separate system of payment has been arranged.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identifica-
tion and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at

http://www.crcpress.com
This book is dedicated to Mohammed A. Ettouney, Fatima A. Abaza,
and William Zacharellis, may God be merciful on their souls.
To my wife Sharada and son Sandeep for their love and patience
Sreenivas Alampalli
Contents
Preface.............................................................................................................................................xxi
Acknowledgments...........................................................................................................................xxv
Authors..........................................................................................................................................xxvii
Chapter 1 Introduction................................................................................................................... 1
1.1 Infrastructure Health in Civil Engineering (IHCE)........................................... 1
1.2 General Concepts of Infrastructures Health......................................................5
1.2.1 What Are Infrastructures?....................................................................5
1.2.2 What Is Structural Health?.................................................................... 5
1.2.3 Infrastructure Health in Civil Engineering........................................... 6
1.2.3.1 General Examples.................................................................. 6
1.2.3.2 Objectives of IHCE................................................................8
1.2.4 The 95% Solution!.................................................................................9
1.3 History of Structural Health Efforts................................................................... 9
1.3.1 Overview...............................................................................................9
1.3.2 Fulfilling Defined Objectives.............................................................. 10
1.3.3 Operations and Management............................................................... 12
1.4 Need for Infrastructure Health Efforts............................................................. 12
1.4.1 Overview............................................................................................. 12
1.4.2 Specific Needs..................................................................................... 13
1.4.3 Case Studies for Uses of IHCE........................................................... 15
1.4.3.1 Cost and Bridge Management............................................. 15
1.4.3.2 Benchmark Bridge Behavior............................................... 15
1.4.3.3 Inspection and Rehabilitation.............................................. 15
1.4.3.4 Monitoring Specific Bridge Types....................................... 16
1.5 Analogy with Human Health........................................................................... 18
1.6 Structural Health: A New Field in Civil Engineering?.................................... 21
1.6.1 What Constitute a New Field?............................................................. 21
1.7 Structural Health: Integration of Fields............................................................ 23
1.8 Horizontal Integration: Multihazards...............................................................24
1.8.1 Introduction to Multihazards...............................................................24
1.8.1.1 General................................................................................24
1.8.1.2 What Is Not Multihazards...................................................25
1.8.2 Theory of Multihazards......................................................................25
1.8.3 Multihazards Categories and Structural Health..................................26
1.8.4 Multihazards and Infrastructure Health.............................................. 27
1.8.4.1 General................................................................................ 27
1.8.4.2 Structural Health Monitoring..............................................28
1.8.5 Benefits versus Limitations.................................................................28
1.9 Vertical Integration: Multidisciplinary............................................................. 31
1.9.1 Different Disciplines and SHCE......................................................... 31
1.9.2 Treatment of SHCE Multidisciplinary Needs in Infrastructure
Health Environment�� 31
1.10 SHM and IHCE: A Quick Primer.................................................................... 32
1.11 Challenges Ahead............................................................................................. 33
vii
viii Contents
1.11.1 General................................................................................................ 33
1.11.2 Manual versus Automatic Inspection.................................................. 35
1.11.3 Climate Change................................................................................... 36
1.12 Appendix I: DHS Infrastructure Sectors (as of 2009)...................................... 38
References................................................................................................................... 39
Chapter 2 Elements of Structural Health in Civil Engineering (SHCE)..................................... 41

2.1 SHCE: A New Field in Civil Engineering?...................................................... 41
2.1.1 This Chapter........................................................................................ 41
2.2 SHCE: Baseball Analogy................................................................................. 43
2.2.1 Health Monitoring from Start to Finish.............................................. 43
2.2.2 Home Plate: Problem Definition......................................................... 43
2.2.2.1 First Base: Measurements.................................................... 43
2.2.2.2 Second Base: Structural ID.................................................46
2.2.2.3 Third Base: Damage Detection...........................................46
2.2.2.4 Fourth Base (Home Plate): Decision Making......................46
2.2.3 Example............................................................................................... 48
2.2.4 Bringing All Together......................................................................... 49
2.3 SHCE and DESIGN Paradigms....................................................................... 50
2.3.1 General................................................................................................ 50
2.3.1.1 Statistical Pattern Recognition............................................ 50
2.3.2 Capacity/Demand Design.................................................................... 51
2.3.2.1 General................................................................................ 51
2.3.2.2 Capacity/Demand Ratio as the True Measure of
Structural Health................................................................. 52
2.3.3 Performance-Based Design................................................................. 57
2.3.3.1 Overview.............................................................................. 57
2.3.3.2 Prescriptive versus Performance Design Paradigms........... 57
2.3.3.3 Modern versus Traditional PBD.......................................... 59
2.3.3.4 Elements of PBD.................................................................. 59
2.3.3.5 Future Trends of PBD..........................................................64
2.3.3.6 Role of SHCE and PBD....................................................... 65
2.4 NDT versus SHM............................................................................................. 70
2.4.1 Overview............................................................................................. 70
2.4.2 NDT Tools for Different SHM/SHCE Components........................... 70
2.4.3 NDT, SHCE/SHM Tools and Bridge Hazards.................................... 70
2.4.4 Closing Remarks................................................................................. 75
2.5 Value of Information........................................................................................ 76
2.5.1 Introduction......................................................................................... 76
2.5.2 General Approach to Evaluate Value of Information.......................... 77
References................................................................................................................... 78
Chapter 3 Lives and Deaths of Infrastructure............................................................................. 81

3.1 Overview.......................................................................................................... 81
3.1.1 Enabling versus Triggering Causes..................................................... 81
3.1.2 This Chapter........................................................................................ 81
3.2 Birth of Bridges................................................................................................ 82
3.2.1 Ensuring Healthy Life......................................................................... 82
3.2.1.1 Preliminary Planning.......................................................... 83
Contents ix
3.2.1.2 Design Phase........................................................................ 83

3.2.1.3 Construction Phase..............................................................84
3.2.1.4 Quality Control/Quality Assurance (QA/QC)..................... 85
3.2.2 Role of SHCE...................................................................................... 87
3.3 Why Bridges Live?........................................................................................... 87
3.3.1 Definitions of Healthy Life................................................................. 87
3.3.2 Classification of Healthy Life Causes................................................. 87
3.3.2.1 Enabling Causes of Healthy Life......................................... 87
3.3.2.2 Triggering Causes of Healthy Life...................................... 89
3.3.3 Quantification of Healthy Life............................................................ 89
3.3.3.1 Conventional Metrics........................................................... 89
3.3.3.2 Healthy Life Causes as Metrics........................................... 93
3.3.4 Role of SHM........................................................................................96
3.4 Why Bridges Fail/Die?.....................................................................................97
3.4.1 Definitions of Failure..........................................................................97
3.4.2 Classification of Failure Causes........................................................ 100
3.4.2.1 Triggering Causes.............................................................. 100
3.4.2.2 Enabling Causes................................................................ 102
3.4.2.3 Estimating Bridge Failure Propensity to Causes............... 102
3.4.2.4 SHM/SHCE Role............................................................... 105
3.4.3 Attributes of Failure.......................................................................... 105
3.4.4 Structural Modes of Failure.............................................................. 108
3.4.4.1 Ductile Failure................................................................... 108
3.4.4.2 Brittle Failure..................................................................... 109
3.4.4.3 Global Failure.................................................................... 110
3.4.4.4 Progressive Failure (Collapse)........................................... 110
3.4.4.5 Mitigation Strategies and Failure Modes........................... 110
3.5 Examples of Bridge Failures.......................................................................... 115
3.5.1 Introduction....................................................................................... 115
3.5.2 Tay Rail Bridge, UK.......................................................................... 115
3.5.3 Quebec Bridge Failure...................................................................... 116
3.5.4 Tacoma Narrows, WA....................................................................... 117
3.5.5 Silver Bridge, OH.............................................................................. 121
3.5.6 Mianus River Bridge Collapse.......................................................... 122
3.5.7 Schoharie Bridge Collapse................................................................ 124
3.5.8 I-35 Bridge, Minneapolis, MN.......................................................... 126
3.6 Appendix I: Count of Bridges in the United States by Construction
Material�� 130
References................................................................................................................. 130
Chapter 4 Theories and Principles of IHCE.............................................................................. 133

4.1 Introduction.................................................................................................... 133
4.2 General Theory of Experimentation (GTE)................................................... 135
4.2.1 Introduction....................................................................................... 135
4.2.2 Overview of GTE.............................................................................. 136
4.2.3 Proof of GTE..................................................................................... 137
4.2.3.1 General.............................................................................. 137
4.2.3.2 Corollary # 1: Types of Experiments and α GTE................ 139
4.2.3.3 Corollary # 2: Initial Conditions....................................... 141
4.2.4 Practical Applications....................................................................... 141
x Contents
4.2.4.1 General.............................................................................. 141

4.2.4.2 Identifying Goals............................................................... 142
4.2.4.3 Identifying Algorithms...................................................... 142
4.2.4.4 Needed Type of Measurements (TOM)............................. 142
4.2.4.5 Values of Parameters......................................................... 142
4.2.4.6 Parameter Cost Evaluation................................................ 143
4.2.4.7 Value–Cost (V–C) Diagram.............................................. 143
4.2.4.8 Critical Magnitudes: Vcr and Ccr........................................ 143
4.2.4.9 Summary of the Steps....................................................... 143
4.2.5 Examples of General Theory of SHM.............................................. 144
4.2.5.1 Frequency Analysis........................................................... 144
4.2.5.2 Corrosion FRP Mitigation Efficiency................................ 146
4.3 Special Theory of Experimentation (STE)..................................................... 147
4.3.1 Introduction....................................................................................... 147
4.3.2 Overview of STE............................................................................... 150
4.3.3 Proof of STE...................................................................................... 151
4.3.3.1 Optimality of Sensor Type................................................. 151
4.3.3.2 Optimality of Sensor Number........................................... 151
4.3.3.3 Optimality of Sensor Location.......................................... 153
4.4 Theory of SHM Triangulation........................................................................ 153
4.4.1 Theory Statement.............................................................................. 153
4.4.2 Proof of Theory................................................................................. 154
4.4.3 Discussion and Applications............................................................. 155
4.4.3.1 General Discussion............................................................ 155
4.4.3.2 Case Study......................................................................... 156
4.4.4 Evaluating Detectability of SHM Schemes....................................... 157
4.4.4.1 Numerical Method for Evaluating Detectability............... 158
4.4.4.2 Subjective Method for Evaluating Detectability............... 160
4.4.5 Triangulation Cost–Benefit Analysis................................................ 160
4.4.5.1 Analytical Considerations................................................. 160
4.4.5.2 Case Studies....................................................................... 161
4.4.6 Concluding Remarks......................................................................... 162
4.5 Duality Principle in SHM.............................................................................. 162
4.6 Scaling Principle in SHM............................................................................... 164
4.6.1 Overview........................................................................................... 164
4.6.2 One-Dimensional Scales................................................................... 164
4.6.2.1 Frequency Spectrum.......................................................... 164
4.6.2.2 Sensor Density................................................................... 165
4.6.2.3 Scaling in Structural Identification.................................... 165
4.6.2.4 Scaling in Damage Identification...................................... 166
4.6.2.5 Hazard (Demand) Scale..................................................... 167
4.6.3 Two-Dimensional Scale.................................................................... 167
4.6.4 Scales of Higher Dimensions: Role of Decision Making.................. 167
4.6.5 Benefits of Scale Considerations in SHM......................................... 170
4.7 Serendipity Principle in SHM........................................................................ 170
References................................................................................................................. 172
Chapter 5 Sensors and Infrastructure Health............................................................................ 175

5.1 Introduction.................................................................................................... 175
5.1.1 Sensing and Infrastructure Health.................................................... 175
Contents xi
5.1.2 This Chapter...................................................................................... 175

5.2 Sensor Categorizations................................................................................... 176
5.3 Basics of Sensor Behavior.............................................................................. 178
5.3.1 Voltage-Generating Sensors.............................................................. 179
5.3.1.1 Piezoelectric....................................................................... 179
5.3.1.2 Other.................................................................................. 180
5.3.2 Resistive Sensors............................................................................... 180
5.3.2.1 Wheatstone Bridge............................................................ 181
5.3.3 Change of Capacitance...................................................................... 182
5.3.4 Linear Variable Differential Transformers (LVDT)......................... 182
5.3.5 Optics................................................................................................ 183
5.3.6 Other.................................................................................................. 183
5.3.7 NDT-Specific..................................................................................... 183
5.4 Sensor Measurements in SHM....................................................................... 183
5.4.1 Overview........................................................................................... 183
5.4.2 Strain Sensors.................................................................................... 184
5.4.2.1 General.............................................................................. 184
5.4.2.2 Foil Strain Gages............................................................... 184
5.4.2.3 Demountable Strain Gages/Strain Transducers................. 186
5.4.2.4 Vibrating Wire Gages........................................................ 188
5.4.3 Position.............................................................................................. 189
5.4.4 Accelerometers (Angular and Linear)............................................... 191
5.4.4.1 LVDT-Based Accelerometers............................................ 191
5.4.4.2 Piezoelectric Sensor Accelerometers................................. 192
5.4.5 Velocities (Angular and Linear)........................................................ 193
5.4.6 Displacements (Angular and Linear)................................................ 194
5.4.7 Force.................................................................................................. 194
5.4.8 Inclinometers..................................................................................... 197
5.4.9 Corrosion........................................................................................... 199
5.4.10 Pressure.............................................................................................200
5.4.11 Temperature....................................................................................... 201
5.4.12 Other Sensors in SHM...................................................................... 201
5.4.13 Advanced Sensing Technologies....................................................... 201
5.4.13.1 Microelectromechanical systems (MEMS)
Technology......................................................................... 201
5.4.13.2 Friction-Based Strain Sensing...........................................202
5.4.13.3 Scour Sensing....................................................................202
5.4.13.4 Innovative ETDR Sensing.................................................202
5.5 Emerging Technology: Fiber Optics Sensors.................................................204
5.5.1 Introduction.......................................................................................204
5.5.1.1 General..............................................................................204
5.5.1.2 Time of Flight and FOS.....................................................206
5.5.1.3 Different Modes of FOS....................................................206
5.5.1.4 ToMs of FOS......................................................................206
5.5.2 Local FOS: Bragg Grating................................................................206
5.5.2.1 Applications of FBG..........................................................208
5.5.3 Local FOS: Fabry-Perot....................................................................209
5.5.4 Distributed (Long) FOS: Optical Time Domain
Reflectometry (OTDR)...................................................................... 211
5.5.4.1 BOTDR in SHM................................................................ 212
xii Contents
5.5.5 Distributed (Long) FOS: Brillouin Optical Time Domain

Analysis (BOTDA)............................................................................ 213
5.5.5.1 Smart BOTDA................................................................... 213
5.5.6 Comparison between Local and Distributed FOS............................ 214
5.5.7 FOS Use in NDT versus SHM.......................................................... 214
5.5.8 Value (Cost–Benefit) of FOS............................................................. 216
5.5.8.1 Advantages and Disadvantages of FOS............................. 216
5.5.8.2 Objective Cost–Benefit Analysis of FOS........................... 217
5.5.9 Case Studies—Applications.............................................................. 219
5.5.9.1 Overall SHM Experiment.................................................. 219
5.5.9.2 Monitoring Composite Rebars........................................... 221
5.6 Wireless Sensors............................................................................................. 221
5.6.1 Overview........................................................................................... 221
5.6.1.1 Composition of Wireless Sensor Units.............................. 221
5.6.1.2 Composition of Wireless Sensor Network......................... 222
5.6.1.3 Wireless Sensors and SHM............................................... 223
5.6.2 Case Studies......................................................................................224
5.6.2.1 Trout Brook Bridge............................................................224
5.6.2.2 Geumdang Bridge.............................................................. 225
5.6.2.3 Benjamin Franklin Bridge, PA.......................................... 226
5.6.3 Wireless Sensors–Specific Methods and Techniques....................... 230
5.6.3.1 RFID in Civil Infrastructure............................................. 230
5.6.3.2 Web-Based Systems........................................................... 232
5.6.3.3 Structural Identification..................................................... 232
5.6.3.4 Damage Identification........................................................ 234
5.6.3.5 Remote Programmability.................................................. 236
5.6.3.6 Decision Making............................................................... 236
5.7 Smart Structures............................................................................................. 236
5.7.1 What Is a Smart Structure?............................................................... 236
5.7.1.1 Smart Structures and Smart Materials.............................. 238
5.7.2 Smart Structures and IHCE Components......................................... 239
5.7.3 Emerging Paradigms for Smart Structures....................................... 241
5.8 Optimum Sensor Selection............................................................................. 242
5.8.1 Overview........................................................................................... 242
5.8.2 General Parameters........................................................................... 242
5.8.2.1 Objectives of Sensing/Project............................................ 242
5.8.2.2 Type of Structure............................................................... 243
5.8.2.3 Physical Parameter for Sensing (ToM).............................. 243
5.8.2.4 Monitoring Time Span....................................................... 243
5.8.2.5 Operating Environment..................................................... 243
5.8.2.6 Assembly/Maintenance/Disassembling (Excluding
Labor Cost)........................................................................ 245
5.8.2.7 Cost.................................................................................... 245
5.8.2.8 Instrumentation Type.........................................................246
5.8.2.9 User Knowledge/Experience.............................................246
5.8.2.10 Sensitivity/Range/Resolution.............................................246
5.8.2.11 Sensor Physical Attributes................................................. 247
5.8.2.12 Formalizing Sensor Choice Parameters............................ 247
5.8.3 Decision Tree Method.......................................................................248
5.8.4 Sensor Utility Method....................................................................... 250
5.9 Optimum Sensor Location............................................................................. 252
Contents xiii
5.9.1 Overview........................................................................................... 252

5.9.2 Analytical Treatment......................................................................... 253
5.9.2.1 General.............................................................................. 253
5.9.2.2 OSL Methodology............................................................. 253
5.9.2.3 Stress Ratio–Based OSL.................................................... 255
5.9.2.4 Multiple Loading Conditions—Goal Programming......... 255
5.9.3 Case Studies...................................................................................... 255
5.9.3.1 Steel Transmission Tower.................................................. 256
5.9.3.2 Suspension Bridge............................................................. 258
5.9.4 Closure: Importance of OSL for SHM.............................................. 261
5.10 Step by Step Guide for Choosing Sensors...................................................... 262
5.11 Remote Sensing in SHM................................................................................ 262
5.11.1 Overview........................................................................................... 262
5.11.2 Types of RS....................................................................................... 263
5.11.2.1 Type I: Sensors Attached to Subject.................................. 263
5.11.2.2 Type II: Sensors Attached to Robots.................................264
5.11.2.3 Type III: Remote Sensors, Passive RS...............................264
5.11.2.4 Type IV: Remote Sensors, Active RS................................ 265
5.11.3 Technical Aspects of Remote Sensing.............................................. 269
5.11.3.1 Penetration versus Backscatter.......................................... 269
5.11.3.2 Decision Making............................................................... 271
5.11.4 Current Applications in SHM........................................................... 271
5.11.4.1 Thermography................................................................... 271
5.11.4.2 Ground Penetrating Radar (GPR)...................................... 273
5.11.4.3 Thermoelastic Method....................................................... 274
5.11.4.4 Correlating Load Level with Surface-Wave Laser
Measurements.................................................................... 275
5.11.5 Closing Remarks............................................................................... 275
References................................................................................................................. 276
Chapter 6 Structural Identification (STRID)............................................................................. 281

6.1 Introduction.................................................................................................... 281
6.1.1 Two-Dimensionality of Modern STRID Techniques........................ 281
6.1.2 Definition of STRID.......................................................................... 281
6.1.3 Objectives of STRID......................................................................... 282
6.1.4 General Model of Structural Identification....................................... 282
6.1.5 General Methods of STRID..............................................................284
6.1.6 Static versus Dynamic STRID..........................................................284
6.1.7 This Chapter......................................................................................284
6.2 STRID Processes............................................................................................ 285
6.2.1 Generalized Classifications of STRID.............................................. 285
6.2.2 STRID: Interrelation of Time and Space Modeling Issues............... 285
6.2.2.1 Frequency and Space Scales.............................................. 285
6.2.2.2 Objectives of STRID and Frequency-Space Domain........ 289
6.2.2.3 Methods of STRID and Frequency-Space Domain........... 290
6.3 Modal Identification Methods........................................................................ 291
6.3.1 Theoretical Background.................................................................... 291
6.3.1.1 Overview............................................................................ 291
6.3.1.2 Equations of Motion.......................................................... 293
6.3.1.3 Modal Solution.................................................................. 293
xiv Contents
6.3.1.4 Frequency Response Function (FRF)................................ 294

6.3.1.5 Impulse Response Function (IRF)..................................... 295
6.3.1.6 Pseudo Inverse................................................................... 295
6.3.1.7 Singular Value Decomposition (SVD)............................... 295
6.3.1.8 Modal Assurance Criteria (MAC)..................................... 296
6.3.1.9 Matrix and Vector Norms.................................................. 296
6.3.2 Order of System Models.................................................................... 296
6.3.3 Attributes of Modal Identification methods...................................... 297
6.3.4 Peak Amplitude................................................................................. 298
6.3.5 Eigensolution Realization Method (ERA)........................................ 299
6.3.6 Ibrahim Time Domain (ITD) Methods.............................................300
6.3.7 LSCE (Least Square Complex Exponential).....................................302
6.3.8 Random Decrement........................................................................... 303
6.3.8.1 General.............................................................................. 303
6.3.8.2 Overview of Problem......................................................... 303
6.3.8.3 Theoretical Background....................................................304
6.3.8.4 Infrastructure Application................................................. 310
6.3.8.5 Concluding Remarks......................................................... 311
6.4 Parameter Identification (PI).......................................................................... 313
6.4.1 Problem Definition............................................................................ 313
6.4.2 Eigenstructure Assignment Update................................................... 313
6.4.2.1 Summary of Input/Output Measurements/Testing............ 315
6.4.2.2 Advantages/Disadvantages and Applicability to SHM..... 315
6.4.3 Optimal Matrix Update (OMU)........................................................ 316
6.4.3.1 General Description........................................................... 316
6.4.4 Sensitivity-Based Update.................................................................. 316
6.4.4.1 Sanayei Static-Based Methods........................................... 316
6.4.5 Other Parameter Identification Methods........................................... 319
6.5 Artificial Neural Networks (ANN)................................................................ 320
6.5.1 Introduction....................................................................................... 320
6.5.2 Basic Developments of ANN............................................................ 321
6.5.2.1 Single Layer ANN............................................................. 321
6.5.2.2 Multilayer ANN: Versatile Complex Problems................. 323
6.5.2.3 Theoretical Background of Multilayer ANN.................... 323
6.5.3 Advanced Topics............................................................................... 326
6.5.3.1 ANN versus FEM.............................................................. 326
6.5.3.2 Neural Network, Regression Analysis, and Expert
Systems.............................................................................. 326
6.5.3.3 Training versus Validation................................................. 328
6.5.3.4 Batch and Online Learning............................................... 328
6.5.3.5 Supervised versus Nonsupervised Learning..................... 328
6.5.3.6 How to Design a Neural Network...................................... 328
6.5.4 Applications of Neural Networks in SHM Field............................... 329
6.5.5 Examples and Case Studies............................................................... 330
6.5.5.1 Single Input, Single Output: Linear Neural Network........ 330
6.5.5.2 Nonlinear Spring Behavior................................................ 330
6.5.5.3 Bridge Scour...................................................................... 333
6.6 Other Methods................................................................................................ 335
6.6.1 Frequency Ranges............................................................................. 335
Contents xv
6.6.2 Energy-Based Methods..................................................................... 337

6.6.2.1 Statistical Energy Analysis (SEA)..................................... 337
6.6.2.2 Energy Finite Element Analysis (EFEA).......................... 338
6.6.3 Dispersion Analysis........................................................................... 341
6.6.4 Genetic Algorithms........................................................................... 341
6.7 Modeling Techniques..................................................................................... 342
6.7.1 Overview........................................................................................... 342
6.7.2 Soil-Structure Interaction Role in Structural Identification
Problem............................................................................................. 342
6.7.3 Soil-Structure Interaction Role in Damage Identification
Problem.............................................................................................344
6.7.4 Fluid-Structure Interaction (FSI) in STRID Problem.......................344
6.7.5 Static Condensation Modeling Techniques....................................... 345
6.7.6 Scaling Needs....................................................................................346
6.7.7 Multihazards Considerations and STRID......................................... 347
6.7.7.1 Modal Identification.......................................................... 347
6.7.7.2 Parameter Identification..................................................... 347
6.7.7.3 Neural Networks................................................................348
6.8 Scale Independent Methods (SIM)................................................................. 349
6.8.1 Scale Independent Element............................................................... 349
6.8.1.1 Overview............................................................................ 349
6.8.1.2 Theory................................................................................ 350
6.8.1.3 Efficiency of SIE Method.................................................. 352
6.8.1.4 Examples............................................................................ 353
6.8.1.5 Closing Remarks................................................................ 357
6.9 Case Studies................................................................................................... 358
6.9.1 General.............................................................................................. 358
6.9.2 Generating Baseline Models............................................................. 358
6.9.3 Scour STRID..................................................................................... 358
6.9.4 Foundation Identification.................................................................. 359
6.9.5 Condition Assessment for Existing Bridges......................................360
6.9.6 Signal Processing.............................................................................. 362
6.9.7 Optimal Sensor Locations (OSL)...................................................... 363
6.9.8 STRID as a Component in SHM......................................................364
6.10 Life Cycle Analysis and STRID..................................................................... 365
6.11 Cost–Benefit Analysis of STRID................................................................... 367
References................................................................................................................. 369
Chapter 7 Damage Identification (DMID)................................................................................. 373

7.1 Introduction.................................................................................................... 373
7.1.1 Mathematical Representation of Damage......................................... 373
7.1.2 Identification Adequacy, Utility, and Prioritization of DMID
Methods............................................................................................. 373
7.1.3 This Chapter...................................................................................... 374
7.2 Damage Parameters........................................................................................ 374
7.3 STRID, DMID, and SHM.............................................................................. 376
7.3.1 Overview........................................................................................... 376
7.3.2 Objectives of STRID and DMID...................................................... 378
7.3.2.1 Objectives of STRID......................................................... 378
xvi Contents
7.3.2.2 Objectives of DMID.......................................................... 378

7.3.3 Definition and Classifications of Damage......................................... 378
7.3.3.1 Physical Classifications...................................................... 379
7.3.3.2 Detectability Classification................................................ 379
7.3.4 STRID Methods and Inferred Damage Identification...................... 379
7.3.4.1 Overview............................................................................ 379
7.3.4.2 Factors Affecting Inferred Damage Detection.................. 380
7.3.4.3 Summary........................................................................... 382
7.3.5 Guide for Using STRID as a DMID Tool......................................... 382
7.3.6 Closing Remarks............................................................................... 383
7.4 NDT Techniques............................................................................................. 383
7.4.1 Introduction to NDT Concepts and Methods.................................... 383
7.4.2 Ultrasound Methods.......................................................................... 386
7.4.2.1 Overview............................................................................ 386
7.4.2.2 Theoretical Background.................................................... 387
7.4.2.3 Impact-Echo....................................................................... 387
7.4.2.4 Pulse-Echo (Normal Waves).............................................. 390
7.4.2.5 Guided (Lamb) Waves....................................................... 392
7.4.2.6 Ultrasonic Vibration.......................................................... 395
7.4.2.7 Laser-Based Ultrasound (LBU)......................................... 396
7.4.2.8 Applications in SHM......................................................... 398
7.4.3 Thermography...................................................................................400
7.4.3.1 Overview............................................................................400
7.4.3.2 Categories of Thermography.............................................400
7.4.3.3 Geometric-Based: Heating Methods................................. 401
7.4.3.4 Time-Based Thermal Application Methods......................402
7.4.3.5 Sensing...............................................................................402
7.4.3.6 Applications in SHM.........................................................403
7.4.4 Electromagnetic Methods.................................................................403
7.4.4.1 Overview............................................................................403
7.4.4.2 Eddy Current.....................................................................403
7.4.4.3 Magnetic Particle (MP).....................................................408
7.4.4.4 Static Magnetic Field (SMF)............................................. 411
7.4.4.5 Electromagnetic Waves: Microwaves or Radio Waves...... 412
7.4.5 Penetrating Radiation........................................................................ 413
7.4.5.1 Overview............................................................................ 413
7.4.5.2 Computed Tomography..................................................... 417
7.4.5.3 Computed (Compton) Backscatter..................................... 419
7.4.6 Other Methods................................................................................... 419
7.4.6.1 Visual Methods.................................................................. 419
7.4.6.2 Liquid Penetrant................................................................ 420
7.4.6.3 Optical Methods................................................................ 421
7.5 Acoustic Emission.......................................................................................... 423
7.5.1 Overview........................................................................................... 423
7.5.2 Theoretical Background.................................................................... 423
7.5.2.1 Basics of AE...................................................................... 423
7.5.2.2 Kaiser, Dunegan, and Felicity Behavior of AE................. 424
7.5.2.3 Emission Initiation and Propagation................................. 425
Contents xvii
7.5.2.4 Important Characteristics of AE Wave.............................. 425

7.5.2.5 AE Data Filtering.............................................................. 426
7.5.2.6 Calibration of AE Sensors................................................. 427
7.5.2.7 Source Location................................................................. 427
7.5.2.8 Properties of AE as Related to SHM................................. 429
7.5.3 AE versus Other NDT Methods........................................................ 429
7.5.4 Applications in SHM—General........................................................ 430
7.5.4.1 General.............................................................................. 430
7.5.4.2 Types of AE Tests in Infrastructure.................................. 431
7.5.5 Corrosion Rate Measurements (Continuous AE).............................. 431
7.5.6 Proof Loading (Burst AE)................................................................. 434
7.5.7 Quality Control (QC)........................................................................ 434
7.5.8 Incipient Failure Detection (IFD)...................................................... 435
7.5.9 Failure of High Strength Wires......................................................... 437
7.5.10 Local Area Monitoring (LAM)......................................................... 438
7.5.11 Advantages and Disadvantages of AE.............................................. 441
7.6 Vibration–based Methods.............................................................................. 442
7.6.1 General.............................................................................................. 442
7.6.2 Overview of Methods........................................................................ 442
7.6.2.1 Changes in Modal Properties............................................ 442
7.6.2.2 Utilization of Numerical Formulations FEM/BEM.......... 443
7.6.3 Case Study I: Theoretical Simulation of DMID Procedures............ 445
7.6.3.1 Introduction....................................................................... 445
7.6.3.2 Vibration-Based DMID methods....................................... 445
7.6.3.3 Results................................................................................ 447
7.6.3.4 Conclusions........................................................................ 451
7.6.4 Case Study II: I-40 Bridge Experiment............................................ 452
7.7 Signal Processing and DMID......................................................................... 455
7.7.1 Time Signals...................................................................................... 455
7.7.2 Fourier-Based Transforms................................................................. 457
7.7.2.1 Conventional Fourier Transforms...................................... 457
7.7.2.2 Time and Frequency Transforms....................................... 459
7.7.2.3 Time-Fourier Transforms (TFT).......................................460
7.7.3 Wavelets.............................................................................................460
7.7.3.1 Introduction.......................................................................460
7.7.3.2 Wavelet Functions.............................................................. 463
7.7.3.3 Heisenberg Principle..........................................................464
7.7.3.4 Basic Example of DWT.....................................................464
7.7.3.5 Applications of Wavelets in Damage Identification..........466
7.7.4 Hilbert–Hwang Transform................................................................469
7.7.4.1 General..............................................................................469
7.7.4.2 Instantaneous Frequency...................................................469
7.7.4.3 Sifting Process................................................................... 471
7.7.4.4 Hilbert Spectrum............................................................... 472
7.7.4.5 Use in SHM: Earthquake Applications............................. 474
7.7.4.6 WT versus HHT in Infrastructure Applications................ 475
7.7.5 Signal Processing as an Integrator of DMID Methods..................... 477
7.7.5.1 Use of Correlation and Operational Modal Analysis........ 477
7.7.5.2 Filtering Multihazards Effects........................................... 479
7.7.6 Rules for Using Signal Processing Tools..........................................480
xviii Contents
7.8 Damage Identification in SHM....................................................................... 481

7.8.1 General.............................................................................................. 481
7.8.2 Important Considerations.................................................................. 481
7.8.3 Interactions with Other SHCE Components..................................... 481
7.8.4 Common Bridge Damages and SHM Applications.......................... 482
7.9 Appendix: Lamb Waves................................................................................. 483
7.10 Appendix: Dispersion Curves........................................................................ 483
7.11 Appendix: Helmholtz Equation...................................................................... 489
7.12 Appendix: Angular Spectrum Method........................................................... 490
References................................................................................................................. 490
Chapter 8 Decision Making in IHCE........................................................................................ 495

8.1 Introduction.................................................................................................... 495
8.2 Decision Making Process and Structural Health Components...................... 496
8.2.1 Introduction....................................................................................... 496
8.2.2 Sensors and Measurements: Data Collection.................................... 497
8.2.3 Structural Identification: Vulnerabilities and Mitigation.................. 498
8.2.4 Damage Identification....................................................................... 499
8.2.5 Decision Making...............................................................................500
8.3 Probability and Statistics................................................................................ 502
8.3.1 Introduction....................................................................................... 502
8.3.2 Statistics............................................................................................ 502
8.3.2.1 Data Collection.................................................................. 502
8.3.2.2 Frequency and Histograms................................................ 502
8.3.2.3 Random Variables.............................................................. 502
8.3.3 Probability Distribution..................................................................... 503
8.3.3.1 Discrete versus Continuous Distributions......................... 503
8.3.3.2 Expectations and Moments of Random Variables.............504
8.3.3.3 Uniform Distribution......................................................... 505
8.3.3.4 Normal and Truncated Normal Distribution..................... 505
8.3.3.5 Poisson’s Distribution........................................................506
8.3.3.6 Marginal Distributions......................................................506
8.3.4 Conditional Probabilities: Bayes Theorem........................................506
8.3.5 Confidence Intervals and Chebyshev Inequality............................... 507
8.3.6 Functions of N Random Variables.................................................... 507
8.3.6.1 General..............................................................................507
8.3.6.2 Taylor Series...................................................................... 508
8.3.6.3 Numerical Integration........................................................ 510
8.3.7 Regression and Correlation Analysis................................................ 510
8.3.7.1 Linear Regression.............................................................. 510
8.3.7.2 Correlation Analysis.......................................................... 511
8.4 Traditional Theories of Decision Making...................................................... 511
8.4.1 Elements of Traditional Decision Making........................................ 511
8.4.1.1 Value versus Utility Methods............................................ 511
8.4.2 Utility Method................................................................................... 512
8.4.2.1 Theory of Utility................................................................ 512
8.4.2.2 Types of Utility.................................................................. 513
8.4.2.3 Multidimensional Utilities................................................. 515
8.4.2.4 Expected Utilities.............................................................. 515
8.4.3 Cost–Benefit/Value (Monetary) Method........................................... 516
Contents xix
8.4.4 More on Traditional Decision Making Methods............................... 517

8.4.5 Use in SHM....................................................................................... 518
8.4.6 Decision Tree.................................................................................... 518
8.4.7 Closing Remarks............................................................................... 518
8.5 Reliability....................................................................................................... 519
8.5.1 Introduction: Reliability and SHM.................................................... 519
8.5.1.1 Capacity, Demand, and SHM............................................ 519
8.5.1.2 Reliability (Safety) Index and SHM.................................. 520
8.5.2 Reliability Methods as Applied to SHM........................................... 521
8.5.2.1 Structural Components...................................................... 521
8.5.2.2 Structural Systems............................................................. 521
8.5.2.3 NDT versus SHM Reliability............................................ 524
8.5.3 Capacity Monitoring: Avoiding Catastrophic Failure....................... 524
8.5.4 Integrated Capacity—Demand Monitoring: Virtual Sensing........... 529
8.5.5 Separate Capacity—Demand Monitoring......................................... 532
8.5.5.1 General.............................................................................. 532
8.5.5.2 Proof Load Testing............................................................ 532
8.5.5.3 LRFD and LRFR............................................................... 533
8.6 Risk................................................................................................................. 534
8.6.1 Overview........................................................................................... 534
8.6.2 Components of Risk Management.................................................... 534
8.6.2.1 Risk Acceptance................................................................ 534
8.6.2.2 Risk Assessment................................................................ 535
8.6.2.3 Risk Treatment................................................................... 535
8.6.2.4 Risk Communication......................................................... 536
8.6.3 Other Forms of Risk.......................................................................... 536
8.6.3.1 Relative versus Absolute Risk........................................... 536
8.6.3.2 Deterministic versus Probabilistic Risk............................ 536
8.6.4 Evaluation of Risk and Its Components............................................ 537
8.6.4.1 General Risk Expression................................................... 537
8.6.4.2 Hazards.............................................................................. 537
8.6.4.3 Vulnerability...................................................................... 537
8.6.4.4 Cost—Consequences......................................................... 537
8.6.5 Relative Risk: Prioritization.............................................................. 539
8.6.5.1 General.............................................................................. 539
8.6.5.2 Failure Mode and Effect Analysis..................................... 539
8.6.6 Relative Risk: Matrix Analysis.........................................................540
8.6.6.1 Introduction.......................................................................540
8.6.6.2 Problem Description.......................................................... 541
8.6.7 Absolute Risk: Cost–Benefit............................................................. 545
8.6.7.1 Introduction....................................................................... 545
8.6.7.2 Problem Description..........................................................546
8.6.8 Absolute Risk: Life Cycle Analysis.................................................. 554
8.6.8.1 Introduction....................................................................... 554
8.6.8.2 Problem Description.......................................................... 555
8.7 Stochastic Models........................................................................................... 559
8.7.1 Markov Processes.............................................................................. 560
8.7.1.1 Example: Bridge Paint Status............................................ 561
xx Contents
8.7.1.2 Use of SHM in Markov Processes: Paint States of a

Bridge................................................................................ 563
8.7.1.3 Value of SHM for a Markov Process: Paint States............ 563
8.7.2 Monte Carlo Method......................................................................... 567
8.8 Structural Analysis in Decision Making........................................................ 568
8.8.1 General.............................................................................................. 568
8.8.2 Probabilistic Finite/Boundary Element Methods.............................. 570
8.8.2.1 Theoretical Background.................................................... 570
8.8.2.2 Example............................................................................. 571
8.9 Financial Considerations................................................................................ 572
8.9.1 Introduction....................................................................................... 572
8.9.2 Present Value..................................................................................... 573
8.9.2.1 Example I: Present Value in Decision Making.................. 573
8.9.3 Future Value...................................................................................... 573
8.9.3.1 Example II: Future Value in Decision Making................. 573
8.9.4 Considerations of Uncertainties........................................................ 574
8.9.4.1 Example I, Continued: Present Value and
Uncertainties...................................................................... 574
8.9.4.2 Example II, Continued: Future Value and
Uncertainties...................................................................... 575
8.9.5 Life Cycle Costs/Benefits and Life Span.......................................... 576
8.9.6 Peer Review....................................................................................... 576
8.10 Appendix: Tables for Computing Cost of Bridge Failure.............................. 577
References................................................................................................................. 581
Appendix Unit Conversion.......................................................................................................... 583
Preface
OVERALL
There is a purpose for building infrastructure, and infrastructure owners are responsible for ensur-
ing that the intended purpose is served while achieving maximum benefit at minimal costs. Taking
appropriate timely actions requires a good understanding of the infrastructure—its current and
expected condition in the future. This volume and its companion volume, Infrastructure Health in
Civil Engineering: Applications and Management (CRC Press, 2012), are dedicated to discussion of
these aspects (see Figure 0.1). This volume focuses on providing an overview of the infrastructure
health in civil engineering (IHCE) and associated theories followed by the description of its four
components: measurements, structural identification, damage identification, and decision making.
Decision making aspect is a unique feature and is introduced with an argument that any project that
does not integrate decision making (or cost–benefit) ideas in all tasks cannot be successful. The
companion volume builds on the ideas presented in this volume and deals with the application of
the IHCE and asset management aspects.
PART I: OVERVIEW AND THEORIES OF IHCE

In the first part of this volume, consisting of four chapters, we lay the foundations of IHCE with
a thorough discussion of its basic components as well as related emerging issues. The introduc-
tory chapter covers the history and importance of IHCE followed by the need for it as applied to
structures in general and, in particular, to bridges. Analogies between structural and human health
fields are explored. The need for knowing (or monitoring) the health of the structure throughout its
life (referred to as “95% solution”) is shown compared to current partial (or 5%) solution, which
is applied to structures at construction and rehabilitation phases. IHCE is a complex subject that
encompasses many disciplines and applications; the second chapter introduces the four basic com-
ponents of IHCE to make its study more systematic and relatively easier. We introduce the baseball
analogy concept to IHCE to further illustrate its basic components. This chapter also explores engi-
neering paradigms and their relationships with IHCE.
Chapter 3 provides analogy between human and structural healths from a chronological view-
point. This chapter provides a comprehensive view of the infrastructure birth (construction), life
(operations), and death (failure/decommissioning), with an emphasis on bridge structures. The
importance of the 95% solution as well as the potentially immense contributions of IHCE to the
entire bridge life will emerge in this chapter. Chapter 4 deals with optimization of structural health
monitoring projects with both technical and financial considerations. The importance of “value” in
an emerging field such as IHCE is well recognized; several theories and paradigms related to IHCE
are thus introduced in this chapter. General and special theories of experimentation are presented
in an attempt to help stake holders obtain the maximum value from the IHCE-type experiments at
reasonable costs. These theories link experiments to the concept of “value” in an objective fashion
by integrating all aspects of IHCE and its goals, such as improving safety and reducing costs, in a
concise manner. A third theory on sensing triangulation is presented in this chapter to maximize
sensing efficiency by using different types of sensors within an IHCE experiment. Finally, the chap-
ter introduces three principles—duality, scaling, and serendipity—needed to maximize the effi-
ciency of IHCE efforts. Several examples from the literature that illustrate the use of these theories
and principles are also included in this chapter.
xxi
xxii Preface
Infrastructure Health in Civil Engineering (IHCE)
Volume I Volume II
PART I PART II PART III Part IV

Overview and Components of Applications Asset Management
Theories SHCE
5— 8—Load
Measurments: Natural Hazards
1—Introduction testing/
Sensors and
instrumentations bridge rating
1—Scour
2— 6— 9—Bridge
Components of Structural 2— management
SHCE identification Earthquakes
3—Lives and 10—Bridge life

deaths of 7—Damage cycle analysis
identification Deterioration
infrastructures (BLCA)
3—Corrosion of
4—General conventional
and special 8—Decision R/C bridges 11—Bridge
theories of making security
SHCE 4—Corrosion
of prestressed/
post tensioned
bridges
5—Fatigue
Components/
Advanced
Materials
6—FRP bridge
decks
7—FRP
Wrapping
FIGURE 0.1 General layout of this volume and the companion volume, Infrastructure Health in Civil
Engineering: Applications and Management (CRC Press, 2012).
PART II: COMPONENTS OF IHCE

The second part of this volume comprehensively reviews the four components of IHCE that were
introduced in the second chapter. The first and most recognizable phase of IHCE is the measure-
ments/sensing phase and is discussed in Chapter 5. In this chapter, we discuss sensors, sensing tech-
niques, and instrumentation. Measurement needs for various IHCE situations are identified, and
important aspect of sensors are discussed following the road map shown in Figure 0.2. This chapter
also explores various sensing techniques used in IHCE experiments, including emerging topics
Preface xxiii
Aspects of sensors in IHCE
Categories Case studies
Technical Different
theoretical practical
basis applications
Hardware Examples
basic of actual
components sensors
Advantages
and
limitations
FIGURE 0.2 Sensor topics.
such as fiber optic sensors (FOS), wireless sensing, and remote sensing. Optimum sensor locations
and optimum sensor selection that are intended to maximize the benefits of IHCE efforts are also
explored with practical case studies from the literature.
Chapter 6 explores the second component in the IHCE field “Structural Identification (STRID).”
Objectives of STRID are discussed first, followed by discussions on three STRID methodologies
commonly used: modal identification, parameter identification, and functional methods. A spe-
cial attention is paid to artificial neural network methods while discussing functional methods.
STRID modeling techniques—static, dynamic, linear, or nonlinear analyses—as they pertain
to structural health monitoring (SHM) and IHCE are then discussed, keeping in mind that the
complexity of modeling varies widely, depending on the situation at hand. Even though determin-
istic structural modeling is predominantly discussed, some discussion of probabilistic structural
modeling is also included. Scaling aspects and scale-independent methods that show promise in
the STRID field are presented in the end followed by a discussion of cost–benefit issue as applied
to STRID.
Chapter 7 presents an overview of damage identification (DMID). This perhaps is one of the
most challenging subjects presented in this volume as the subject of damage is not very well defined
in the field of SHM/IHCE. This chapter identifies damages from several view points, including
damage types as they result from various hazards. We also discuss the interrelationship between
STRID and DMID. We explore the vibration-based DMID methods, as well as the use of signal
processing in DMID field. Widely used nondestructive testing (NDT) technologies in the field of
SHM are presented, with special emphasis on their usage and applicability in the SHM field. The
chapter finally explores hazard-specific DMID methods.
The main premise of this volume is the tight integration of structural health monitoring (the first
three components of IHCE) and objective decision making processes that is the final component of
IHCE. We advocate that without a strong decision making support, any SHM project could not ful-
fill its promise and thus objective decision making processes must be the basis for maintaining and
improving structural health. Hence, the last chapter of this volume (Chapter 8) is devoted to decision
making processes and presents numerous useful tools/techniques. This includes basics of probabil-
ity and statistics tools, theory of decision making, structural reliability, and risk tools. In addition,
stochastic modeling, economic tools, and probabilistic structural methods are also covered. In all
situations, pertinent examples from the IHCE field are given to illustrate the use and importance of
decision making tools.
It should be noted that all the opinions and views expressed in this volume are those of the
authors and not necessarily of the organizations they represent.
Acknowledgments
The life of every person is measured, among other things, by the special people who kindly affect
such a life. This volume is a direct result of the many beautiful interactions I have had over the years
with those special people. My late mentor, Dr. Mohammed S. Aggour, taught me the very founda-
tions of structural engineering and design. My childhood friend, Dr. Elsayed A. Elsayed, has always
been with me during good and bad times. My long friendships with Dr. Raymond (Ray) Daddazio,
Michael (Mike) Udvardy, Amr Aly, Dr. Loraine Whitman, Ron (Ronnie) Check, Dr. Hamid Adib,
and Christina (Tina) Plink have had a direct as well as an indirect influence on my writing this vol-
ume. I am grateful for the professional help, encouragement, and support given by Antranig (Andy)
Ouzoonian, Norman (Norm) Glover, Gary Higbee, Dr. Amar Chaker, and Arturo (Artie) Mendez.
The special friendship of and the unparalleled technical and professional help and support given
by Dr. Anil K. Agrawal will always be warmly remembered and deeply acknowledged. The advice
and patience of the editors of CRC Press helped immensely in making this project materialize. For
that I am deeply grateful.
Over the millennia, when darkness descended, human beings used to look up to the heavens for
help and guidance. The countless stars of the Milky Way never disappointed them. The combined
belts of Orion and Sirius have always guided those who needed help through the darkest of the dark
hours. I am no exception: without my own belts of Orion and Sirius this volume could have never
materialized. For that, I shall always be grateful to Milagros (Mila) Kennett and Dr. Sreenivas
(Sreeni) Alampalli. Their friendship, kindness, help, and guidance will always be remembered as
long as there are the belts of Orion and Sirius.
It has been a great pleasure to have worked for the past 20 years with several of my hard-working
and intelligent colleagues at the New York State Department of Transportation (NYSDOT), who
strive to ensure that all users of transportation infrastructure are safe and secure every day. I learned
a lot working with them, and several examples in this volume are the result of working with many
of them. One person requiring special acknowledgment is my mentor, the late Dr. Robert J. Perry,
for his encouragement and support in advancing my professional interests.
During the past 20 years, I was very fortunate to have worked with a variety of people from vari-
ous state and federal agencies, universities, private industry, and professional organizations. This
volume has benefitted from the practical and research experiences gained by working with them.
I am specially grateful to Dr. Glenn Washer and Dr. Hamid Ghasemi for their professional help
and support during these years. I thank my good friend Dr. Anil Agrawal for his support and for
simply being there when needed. Several people have had an influence on me but none as much as
Dr. Mohammed M. Ettouney. I met Dr. Ettouney about 15 years ago and found in him a great friend,
an excellent colleague, and a mentor. I am always amazed at his professional competence and his
ability to look beyond the normal. It has been a privilege writing this volume with him.
Sreenivas Alampalli
xxv
Authors
Dr Mohammed M. Ettouney, Ph.D., P.E., MBA, F. AEI, Dist. M. ASCE, was conferred the
Innovators Award in 2008 by the New Jersey Inventors Hall of Fame after he was nominated to
receive such a great honor by the American Society of Civil Engineers (ASCE). Dr. Ettouney also
received the Homer Gage Balcom lifetime achievement award by the Metropolitan Section of
ASCE (2008). He won the Project of the Year Award, Platinum Award (2008) for the New Haven
Coliseum Demolition Project (ACEC, NY). He is a fellow of the Architecture Engineering Institute
(AEI) and a distinguished member in American Society of Civil Engineers. Among his other recent
achievements are the pioneering work on “Theory of Multihazards of Infrastructures,” “Theory of
Progressive Collapse,” pioneering work in multihazards/multidisciplinary evaluation of risk and
resiliency of buildings, tunnels, bridges, and transit stations, and an innovative green design method
for protecting utilities from demolition/blasting (City of New Haven, CT). He has professional inter-
est in diverse areas of structural engineering as demonstrated by the list of his publications, invited
presentations, seminars, and sessions organized during national/international conferences, besides
his membership in different professional organizations.
Dr. Ettouney is a principal with Weidlinger Associates, based in New York City, NY. He received
his Doctor of Science in structural mechanics from the Massachusetts Institute of Technology (MIT),
Cambridge, MA, in 1976. Since then, his interest in structural engineering has been both as a prac-
titioner and a researcher in multihazards safety of structures; probabilistic modeling of progressive
collapse of buildings; uncertainties in structural stability; blast mitigation of numerous buildings
around the world; innovative concepts such as “Probabilistic Boundary Element Method,” “Scale
Independent Elements,” and “Framework for Evaluation of Lunar-Based Structural Concepts.” He
is a past president and past member of the board of governors of AEI, member of the Board of
Directors of the BSC, member of several technical committees on building/infrastructure security,
earthquake hazards, architectural engineering, and Nondestructive Testing and Structural Health
Monitoring. He was chair of the AEI National Conference 2006 and 2008. He is a member of the
NIBS Advanced Materials Council. He is also Editor of the Journal of Advanced Materials. He
serves on the Board of Multihazard Mitigation Council (MMC).
Dr. Ettouney has authored or co-authored more than 325 publications and reports and contrib-
uted to several books. He introduced several new practical and theoretical methods in the fields of
earthquake engineering, acoustics, structural health monitoring, progressive collapse, blast engi-
neering, and underwater vibrations. He co-invented the “Seismic Blast” slotted connection. More
recently, he introduced the “Economic Theory of Inspection,” “General and Special Theories of
Instrumentation,” and numerous principles and techniques in the field of infrastructure health—all
pioneering efforts that can help in developing durable infrastructures at reasonable costs.
Dr. Sreenivas Alampalli, P.E., MBA, is Director of the Structures Evaluation Services Bureau at
the New York State Department of Transportation (NYSDOT). His responsibilities include manag-
ing structural inspection, inventory, and safety assurance programs at the NYSDOT. Before taking
up the current responsibility in 2003, Dr. Alampalli was Director of the Transportation Research
and Development Bureau. In this position, he managed a targeted transportation infrastructure
research and development program to enhance the quality and cost-effectiveness of transportation
policies, practices, procedures, standards, and specifications. He also taught at Union College and
Rensselaer Polytechnic Institute as an adjunct faculty.
Dr. Alampalli obtained his Ph.D. and MBA from Rensselaer Polytechnic Institute, his M.S.
from the Indian Institute of Technology (IIT), Kharagpur, India, and his B.S. from S.V. University,
xxvii
xxviii Authors
Tirupati, India. His interests include infrastructure management, innovative materials for infra-
structure applications, nondestructive testing, structural health monitoring, and long-term bridge
performance. He co-developed the theory of multihazards and has been a great proponent of it
to integrate all vulnerabilities, including security for effective infrastructure management. Dr.
Alampalli is a Fellow of the American Society of Civil Engineers (ASCE), American Society for
Nondestructive Testing (ASNT), and International Society for Health Monitoring of Intelligent
Infrastructure (ISHMII). He has received several awards, including the prestigious Charles Pankow
Award for Innovation, from the Civil Engineering Research Foundation in 2000; ASNT Mentoring
Award in 2009; and Herbert Howard Government Civil Engineer of the Year Award from ASCE
Metropolitan (NYC) section in 2009. He has authored or co-authored more than 250 technical
publications.
Dr. Alampalli is an active member of several technical committees in TRB, ASCE, and ASNT,
and currently chairs the ASCE Technical Committee on Bridge Management, Inspection, and
Rehabilitation. He served as the Transportation Research Board representative for the NYSDOT
and also as a member of the National Research Advisory Committee (RAC). He is an Associate
Editor of the ASCE Journal of Bridge Engineering and serves on the editorial board of the jour-
nal Structure and Infrastructure Engineering: Maintenance, Management, Life-Cycle Design and
Performance and the journal Bridge Structures: Assessment, Design and Construction.
1 Introduction
1.1 INFRASTRUCTURE HEALTH IN CIVIL ENGINEERING (IHCE)
President Obama in his inaugural speech on January 20, 2009 said, “We will build the roads and
bridges, the electric grids and digital lines that feed our commerce and bind us together” (NRP 2011).
This was an acknowledgment of the importance of the role of infrastructure in modern societies. It
is a multifaceted role, including commerce, communication, mobility, recreation, and comfort. We
observe that there is a need to change conventional approach to infrastructure. As modern societies
evolved in the past century, challenges to infrastructure stakeholders evolved, and many new chal-
lenges are emerging. Some of these challenges are increased costs , changing social expectations and
demands, advent of many modern hazards, and the increased complexities of infrastructure. Take,
for example, the bridge infrastructure sector. There are suspension bridges (Figures 1.1, 1.2, and 1.3)
that offer grace and functionality, yet they are extremely complex constructs. Arch and truss bridges
(Figures 1.4 and 1.5) and conventional multigirder, multispan bridges (Figures 1.6 and 1.7), still pres-
ent immense complexities and demands. Single-span bridges (Figure 1.8) need considerable attention
from bridge managers because of the potential consequences that might occur if they are damaged.
Finally, historical bridges (Figure 1.9) need attention and preservation because of their function and
their historic nature.
Infrastructure is aging, even newly built ones start aging as soon as the construction phase
is over. The aging of infrastructure, given all other just mentioned challenges, would require
diligence on part of owners to ensure that they continue operations efficiently at reasonable
costs.
Fortunately, the state of the profession has also been evolving in many ways that is making it
possible to meet many of these challenges. For example, computing, material, and sensing tech-
nologies have been advancing at a rapid pace. Decision making and management techniques are
evolving and creating new opportunities to owners and managers. New engineering paradigms are
emerging and have the potential of helping stakeholders to meet modern infrastructure challenges
such as multihazards considerations, structural health monitoring (SHM), nondestructive testing
(NDT), performance based engineering (PBD), life cycle analysis (LCA), and many more. Note
that SHM is a subset of IHCE. SHM, as we will consider it in this volume and the companion vol-
ume (Ettouney and Alampalli 2012), contains measurements, structural identification, and damage
identification. IHCE adds decision making component to SHM. Chapter 2 addresses this subject
in more detail.
The new paradigms should be coordinated between stakeholders, carefully considering the inter-
actions between them. These paradigms include, but are not limited to, economics, as built versus
new construction, structural identification (STRID), damage identification (DMID), efficient man-
agement, and decision making (DM).
Perhaps an easy way to deal with all of the above complexities and potential solutions is to note the
similarities of the desired infrastructure healthy performance with human health. We call this conflu-
ence IHCE (infrastructure health in civil engineering): a new and essential field that is a composition
of many disciplines in civil, mechanical, and business communities. It is an inclusive way of thinking
by all stakeholders of civil infrastructure. This volume aims at gathering all the pieces of this new way
of addressing infrastructure and linking them at different and natural points: costs, benefits, design,
analysis, sensing, hazards interaction, decision making, etc.
1
2 Infrastructure Health in Civil Engineering
FIGURE 1.1 Suspension bridge (Brooklyn Bridge, NYC). (Courtesy of Francois Ghanem.)
FIGURE 1.2 Manhattan Bridge, NYC (looking from Brooklyn Bridge). (Courtesy of Francois Ghanem.)
FIGURE 1.3 Old and new Carquinez Bridges, over Suisun Bay, CA.
Introduction 3
FIGURE 1.4 Arch bridge. (Courtesy of New York State Department of Transportation.)
FIGURE 1.5 Truss bridge. (Courtesy of New York State Department of Transportation.)
FIGURE 1.6 Conventional multigirder steel bridge. (Courtesy of New York State Department of Transportation.)
FIGURE 1.7 Multi-span overpass. (Courtesy of New York State Department of Transportation.)
FIGURE 1.8 Single-span road bridge. (Courtesy of New York State Department of Transportation.)
FIGURE 1.9 Historic Wooden bridge. (Courtesy of New York State Department of Transportation.)
Introduction 5
This chapter gives an overview of the main points in the infrastructure health field. We first try
to define what constitutes “infrastructure health.” The history of and the need for structural health
efforts will then be discussed. We offer an analogy between structural health and human health.
The above-discussed need for IHCE is explored in depth later, including the fact that IHCE is a
composition of several disciplines. We then discuss how different hazards and different disciplines
need to be integrated efficiently to form important components of IHCE along with strategic chal-
lenges that require further consideration.
1.2 GENERAL CONCEPTS OF INFRASTRUCTURES HEALTH

1.2.1 What Are Infrastructures?
Perhaps the basic question that needs to be addressed first is what do we mean by infrastructure?
Before we attempt to answer this question, we hasten to say that we concern ourselves in this volume
with civil infrastructure only. American Society of Civil Engineers (ASCE 2009) defined infra-
structure as those that may be built (such as structures, energy, water, transportation, and com-
munication systems), natural (such as surface or ground water resources), or virtual (such as cyber,
electronic data, and information systems). The US Department of Homeland Security (US DHS) has
more specific definitions of infrastructure. As of the writing of this volume, the US DHS subdivided
infrastructure into 17 sectors, US DHS (2009). See Section 1.12 for complete list of the sectors. Of the
17 sector, we can identify civil infrastructure in the transportation (bridges, tunnel, roads, etc.), dams,
and commercial facilities (e.g., office buildings). Many other sectors include civil infrastructure as
subsector components, such as government facilities, public health, or commercial nuclear reactors.
Many of the discussions in this volume utilize bridges as the main subject of the discussion. Most
of the ideas and logics apply to almost all types of civil infrastructure.
1.2.2 What Is Structural Health?

Dictionaries define health as an overall condition at any given time, freedom from abnormalities, or
optimal well-being. Infrastructure is noted as an underlying base or foundation for an organization
or in our context the facilities and installations needed for the functioning of transportation system.
Thus, infrastructure health can be stated as its current ability to provide intended level of service
in safe, secure, cost-effective manner against hazards it is designed for or expected during its life.
Although this statement seems simple, it is not that simple depending on how one perceives it.
Structural health depends on the components it is made of, the network it is in, the characteristics of
the individual structures as well as the network, users needs and expectations, socioeconomic aspects,
expected performance, and approaches taken from the moment the structure is envisioned to its replace-
ment before or after its expected life. Several performance indicators such as safety, mobility, reliability,
and security define structural health. The physical and structural condition is an important factor as
safety is the primary measure that is expected. This aspect of health is more dependent on its compo-
nents, normal loading conditions, current and future environment, and expected hazards during its life.
Operational conditions mostly influence other performance factors of structural health. Thus, all these
factors are variables, and thus it is not easy to deterministically define the structural health. An easier,
deterministic definition of structural health is the current condition or performance of the structure
compared to as-built performance, assuming all other variables are the same as when it was built.
Structures are built for a purpose with certain expectations under in-service as well as extreme
(limit state) loading conditions. Thus, there are two competing factors in ensuring structural health—
the strength at any given time for any given loading or combination of loadings and the loading it
is subjected to at any given time. These loadings can be dead loads that can be predicted with high
probability, live loads that can be predicted with good probability for a long period of time, environ-
mental loads such as wind and ice that can be predicted reasonably well, and other hazards such as
blast and other manmade hazards that cannot be predicted easily. Thus, the health of infrastructure
is not simple to define—as it used to be—as it varies depending on the expected level of service for
a given combination of hazards and time. On the basis of improved knowledge, performance of sys-
tem can vary suddenly as structural strength for a given hazard can change significantly if new fail-
ure mechanisms were discovered and current mechanism are better understood. Thus, probabilistic
approaches are gaining popularity and are being implemented since the last decade.
Another way to define structural health is capacity/demand ratio. As noted above, capacity and
demand vary with time (see Chapter 2). When a structure is designed, the expected demands from
known hazards, based on prevalent design codes and specifications, are considered to make sure that
design capacity is higher than the expected demands. Thus, the capacity is a theoretical value and can be
very conservative due to several simplifications used in the analysis. Demand in its life can vary based
on several factors such as change in specifications, improved knowledge on loading and hazards, change
in network conditions (closure of a bridge can change traffic patterns on another bridge), expecta-
tions of users, socio-economic changes, etc. Similarly, capacity also depends on deterioration, improved
knowledge, rehabilitation and replacement, deterioration, maintenance, damage, etc. Capacity can also
increase due to better evaluation methods such as material testing, improved analysis methods, and test
methods such as load testing that can evaluate the system in a realistic fashion. It should be noted that
similar to human health, component health can vary significantly from the system health.
As noted before, infrastructure performance/health can be structural or operational, that is, a
bridge may have enough capacity to handle the loads but geometry, approach pavement, and other
factors can prevent vehicle capacity and mobility. A network performance can be significantly
reduced due to one small link or segment unless there is redundancy built into the system.
Thus, evaluating and measuring performance is very important to make sure structural health is
maintained through appropriate decision making that include operational decisions, preventive and
corrective maintenance actions, and capital program. Operational performance is relatively easy
to ensure when compared to structural performance. In most cases, structural health/performance
or the degradation effects is measured using a direct physical parameter or an indirect parameter
related to the physical parameter under a given loading. Direct parameters are those that can be cor-
related to structural health directly, whereas indirect parameters (symptoms) are those that give an
indication of structural health. For example, strain can be directly related to the flexural capacity of
a component or the bridge and thus the influence of degradation on its health. Whereas, measuring
the half-cell potentials indicate that the potential to corrosion exists but does not directly indicate
the extent of degradation and its effect on the structure. This is analogous to a blood test where
certain counts may indicate a presence of cancer, whereas change in pulse or blood pressure may
only indicate the symptom of a bigger problem but not the problem itself without further indication.
Sometimes, the measurements can be passive: just observing the state of the system. Some times the
measurements are active, where actuators stimulate a system response that might give indications of
the state of the system (see Figures 1.10 and 1.11).
1.2.3 Infrastructure Health in Civil Engineering

1.2.3.1 General Examples
Numerous authors provided examples and projects related to IHCE. For example, DeWolf et al.
(2002) presented examples showing the remote control/long-term SHM for different bridges.
Another study that is based on the I-40 experiment in New Mexico was performed by Fritzen and
Bohle (2000). They based their study on the use of finite element (FE) model of the bridge. They
utilized several structural and damage identification techniques on the FE model to localize and
then estimate the extent of the damage. They reported high accuracy in achieving their goals. Also
Cheung and Naumoski (2002) presented a general application of SHM: a long-span bridge (smart
bridge) from grounds up.
Introduction 7
(a)
Interaction through Different interactions
mechanical loading with surrounding
Information to observing
(e.g., Gravity loads) environment
decision maker
Monitored structure
Sensor
Interaction through
natural environment
Interaction through (temperature, moisture, etc.) Interaction through
supports (other structural supports (other structural
components) components)
Passive monitoring
(b) Appropriate input

to actuator Different interactions
Interaction through
Information to observing with surrounding
mechanical loading
decision maker environment
(e.g., Gravity loads)
Monitored structure
Sensor Actuator
Interaction through
natural environment
Interaction through (temperature, moisture, etc.)
supports (other structural Interaction through
components) supports (other structural
components)
Active monitoring
FIGURE 1.10 (a) Passive (b) active SHM experiments.
A different paradigm in sensor technology, a component of IHCE, which integrate the fiber
optics sensor properties with the structural behavior, was presented by Glisic and Inaudi (2002). It
is called SOFO and utilizes a finite-element metaphor to detect overall structural behavior with the
local effect of material behavior. In other words, they contrast “long gauge sensors” versus “short
range sensors” to detect structural versus material responses. The concept is investigated and gen-
eralized further in Chapter 6 of this volume via time (frequency) and space scaling. Wang et al.
(2002) offered a general application of SHM to long-span bridges. They described several concepts
of SHM that monitor the bridge as a whole. Oshima et al. (2002) identified three major causes for
damage to aging bridges in Japan as earthquakes, fatigue, and environmental effects. The authors
presented two examples: vibration monitoring and prestressed concrete monitoring.
Sikorsky et al. (2002) linked demand to damage indicators. They enumerated analytical evalua-
tion of stiffness properties. They also offered system identification example. Rochette et al. (2002)
summarized several utilizations of SHM techniques and the information that was gained from them.
Taha et al. (2002) offered an overview on the use of microelectromechanical systems (MEMS),
wireless sensors, and artificial intelligence (AI) in SHM context. Cheng (2004) emphasized the role
of data acquisition in SHM projects. Yamada et al. (2004) related gross vehicle weight from measur-
ing the weigh-in-motion vibrations. This is one of the needed methods for estimating demands on
bridge structures. Savard and Laflamme (2004) presented generic benefits of SHM. Wenger et al.
(2004) offered a general overview of the issues and topics of SHM. Yang and Newhook (2004)
(a)
Damage
Sensor Acoustic emission Sensor

from damage
Data acquisition and

processing unit
Passive monitoring: acoustic emission
(b) Lamb waves from

damage to sensor Damage
Sensor Lamb waves from Actuator

actuator to damage

Generator
processing unit
Active monitoring: ultrasonic lamb wave generation
(c)
Damage
Actuator / Sensor

Generator
processing unit
Active monitoring: electromechanical impedance
FIGURE 1.11 Examples of passive/active. SHM techniques (a) passive monitoring, (b) active monitoring I,
and (c) active monitoring II.
presented the foundations of establishing a general SHM. Note that all these examples cover rela-
tively a small portion of the IHCE and will be explored further in this book.
1.2.3.2 Objectives of IHCE

There are numerous new technologies and construction materials being developed. Obviously, each
of those new technologies (materials, design procedures, analytical methods, etc.) would need to be
observed more closely than conventional technologies after deployment in the field for general pub-
lic use. SHM would help in such observations. Consider, for example, the research by Gangone et al.
(2005). They deployed an experimental platform that can be expanded into an SHM applications for
observing the newly introduced fiber-reinforced polymer (FRP) rebars for use as concrete reinforce-
ment (in lieu of steel rebars). They used long range fiber optics strain sensors that can be attached
to FRP rebars to measure strains along the rebar. As it is a long range sensor, it can measure strains
over the whole length of the rebar. Since one of the design parameters of FRP rebars is concrete
cracks (at a maximum width of 0.01 in, see ACI 2008, it is of importance to ensure that the FRP
strains are within such a limit. The proposed usage of long range fiber optics strain sensor can be
used to assess strains in the FRP rebars, hence concrete crack width along the whole rebar length.
El-Ragaby et al. (2004) presented an example of using two modern technologies to enhance
safety and reduce cost: fiber optics sensing (Fabry-Perot) and glass fiber-reinforced polymer rebars.
They reported a whole real-life SHM experimentation on the Cookshire-Eaton Bridge in Canada.
Introduction 9
Another integrated project that covers all aspects of IHCE was offered by Shehata et al. (2004).
The authors mention that by implementing it, the life of the bridge can be extended, thus saving
costs.
1.2.4 The 95% Solution!

The performance measurement of bridges in the United States and the rest of the world is
predominantly through visual inspection. The data from these visual inspections form the basis for
most of the evaluation and further decision making. Detailed nondestructive testing and analyses
are performed generally on a needed basis to evaluate local or component level data to better under-
stand the structure. Monitoring the structural health on a continuous basis is done only in excep-
tional cases where such data is justified, generally for complex structural systems or structures with
known problems where repairs could be very expensive or are not possible in a very near term and
monitoring is done to closely watch the structure. This is very analogous to the way human health
system works, as explained in later sections of this volume. In either periodic visual inspection or
continuous health monitoring, collection of data is only a minor detail in these authors’ opinion,
that is, a 5% solution. Collecting the right information (data) in the right way, designed to provide
enough information to make appropriate decisions within the time frame required, to make sure
that the structural health is continuously maintained is the important issue, that is, termed as the rest
of the story (95% solution) yielding to complete solution.
1.3 HISTORY OF STRUCTURAL HEALTH EFFORTS

1.3.1 Overview
Since the beginning of structures, structural health evaluation and monitoring existed as structures
that were always visually examined by hearing to the sounds they made in operation or when tapped
on them, and by the touch. Visual inspection probably is the oldest technology used for structural
evaluation and has been used for centuries. It is still the most important technology used to ensure
the integrity of the structure in fulfilling its mission and has come a long way in the last two
decades. Thus, the New York State Department of Transportation describes their bridge visual
inspection process as “Sight, Sound, and Touch.” When accomplished using an educated and expe-
rienced person with commonsense, visual inspection is very important and valuable even with all
the other technology used for SHM because finally a person has to make a decision based on data
from any source or technology. Even though visual inspection offers advantages such as availabil-
ity almost everywhere and is easily understood by everyone, it too has several disadvantages. It is
subjective, that is, depends on the person and his/her capabilities, experience, and understanding of
the structure. Hence, the consistency and quality depends on the entire process, manuals, training
etc. Thus, changing the process based on the changing state-of-the-art and practice is very lengthy,
cumbersome, and often expensive in big organizations. Reliability of the process (probability of
detection) is very hard to quantify.
Nondestructive testing (NTD) is the next phase of evaluating structural health. NDT in some
form has always existed as part of or as a first and foremost supplement to visual inspection. Tapping
on the structure with a hand or a coin or a hammer is probably the most rudimentary and early non-
destructive testing after the visual inspection. This is generally used where visual inspection is not
sensitive or when a structure needs better quantifiable evaluation. As illustrated in the later sections
of this volume, NDT is mostly referred to local area evaluation, and in recent years there have been
several tools that are very reliable, relatively inexpensive, and easily available. These are also well
accepted by the industry, owners, and public in recent years. But, most NDT applications still require
manual operation, the quality of their results depends on training and experience, their applications
require access to the area requiring inspection, and their results are hard to interpret and use. These
limitations of NDT increases as the structure gets complicated or thicker components due to attenua-
tion of waves. Automation is also extremely difficult in most cases. Most of these techniques are also
material dependent and hence knowledge of the structural material and stress patterns is a must.
The structural health monitoring is commonly referred to sensor-based technologies that mea-
sure the structural integrity and are an extension to the common nondestructive instrumentation
used for local evaluation. These are generally meant for global evaluation rather than local evalu-
ation and in most cases are treated as continuous or periodic monitoring context. They offer new
era in structural evaluation and are still in development phase. These offer to measure not only the
issues with damage but also can relate them to structural stresses and strains. They also but have
potential to relate them to loads causing them, environmental conditions influence, and operating
conditions effect. These systems, thus, are generally referred to permanent or semipermanently
attached or embedded sensors used continuously or as needed for monitoring/evaluation of struc-
tural health.
Advances in recent sensor technologies and innovation in both information technology and data
transmission technologies have potential to offer opportunities that were unimaginable or unthink-
able a decade ago. As described in other sections, in more detail, in the future these offer efficien-
cies in all aspects of managing structures—planning, design, analysis, inspection, maintenance,
rehabilitation, and replacement—and thus increased optimization, innovation, and cost savings. As
all these activities currently are more reactive or on schedule, in the future all these have potential
to be predictive in nature and can lead to just-in-time activities, reducing unscheduled activities to
a minimum.
It should be noted that all three, visual inspection, local NDT, and automated SHM, are invalu-
able and one can not be substituted for the other completely. Even though SHM may indicate prob-
lems in the structure they may still have to be followed up with visual inspection and local NDT to
select most appropriate option/decision. All these are tools in a decision tool chest and should be
used together on a need basis so that they complement each other.
The progression of health evaluation explained above, at first, is seen in human health evalu-
ation: physical examination—visual, sound, and touch by a physician, nondestructive evalu-
ation such as X-ray, followed by periodic or continuous monitoring that are in development.
Aerospace industry has evolved in this field significantly, where all these methods have been
used very effectively to meet the demands, to minimize downtimes, and to provide just-in-time
maintenance schedules. Bridge engineering field still relies on visual inspection supplemented
by NDT for local evaluation. See Table 1.1 for changes in bridge inspections nationally and
Table 1.2 for some common NDT methods used. In recent days, there has been a push for
sensor-based SHM, but there are ways to go due to limited or unknown reliability of SHM,
diversity, and complexity of structures, cost-effectiveness, organizational culture, and type of
workforce required.
1.3.2 Fulfilling Defined Objectives

Fulfilling the defined objectives is a key mission of IHCE concept. Weinmann and Lewis (2004)
illustrated the structural health in civil engineering (IHCE) concept, which started with an objective
and fulfilled the objective. The authors presented an example of monitoring carbon fiber composite
cables (CFCC) in transverse and longitudinal direction, CFCC strands as mild reinforcement, and
CFRP grid reinforcement in top slab of a bridge with 12 double-tree girder. Due to the experimental
nature and the lack of design standards, several design assumptions were made during the design
phase. Hence, at first, a full-scale prototype girder was lab tested to failure. Lab test results showed
that the cracking moment was within 7% of the assumed value; and total ultimate mid-span moment
at failure was 60% greater than the corresponding theoretical estimate.
The bridge was instrumented to monitor long-term performance in service. The objectives were
well defined and instrumentation was chosen appropriately. The objectives included measuring
Introduction 11
TABLE 1.1
Highlights of National Bridge Inspections
Before 1916 Ad hoc based inspections
1916 Act Inspections mentioned as part of maintenance work by states and others
1967 Prompted by Silver Bridge collapse, president establishes a task force to
avoid future disasters
1968 FHWA orders initial inventory and reviews of bridges
2-year inspection interval for important structures
1971 Establishment of NBIS
• Defines qualification of personnel
• Defines inspection types
• Detailed reporting format
• Establishment of appraisal, condition, and sufficiency ratings
• 2-year inspection cycles
1988 NBIS revisions
• Varied inspection intervals
• 2-year maximum inspection cycle for fracture critical bridges
• Underwater bridge inspection requirements
2004 NBIS Revisions
• Complex bridges requiring specialized procedures
• QA/QC emphasis
• Follow-up of critical findings
• Refresher training
TABLE 1.2
NDT Methods Used in Bridge Engineering
Routine Visual inspection
Sounding
Chain drag
Ultrasonics
Magnetic-particle
Dye-penetrant
Half-cell potentials
Geophysical methods
Often Radiography
Ground penetrating radar
Eddy current
Vibration monitoring
In development Impact echo
Thermography
Remote sensing
pretension load applied to CFRP tendons, concrete strain distributions, and girder camber and
forces in post-tensioning CFCC tendons during fabrication/construction sequence. The integrity of
longitudinal tendons was monitored during the construction sequence. The load tests indicated that
all three spans exhibited similar load distribution behavior and the load distribution behavior was
considerably better than what was assumed.
1.3.3 Operations and Management

Catbas et al. (2000) define bridge health monitoring as the measurement of the operating and load-
ing environment and the critical responses of a structure to track and evaluate the symptoms of
operational incidents, anomalies, and/or deterioration or damage indicators that may impact opera-
tion, serviceability, or safety reliability. It requires various strategies and specialized diagnostic
tools, many of which are depicted in Figure 1.12. These strategies depend on at what point of time
a facility may be in its lifespan and whether an event or symptom triggers a routine or an in-depth
check-up just as in medical health management.
Authors note that effective bridge management requires accurate estimates of the actual load-
carrying capacity, system reliability and failure mode rather than just a subjective measure of con-
dition. Thus, integration of both operational and SHM of bridges is important to provide objective,
accurate, and sufficiently comprehensive data to serve as a basis for reliable evaluations, projections,
and forecasts. This project showed an initial attempt for a major long-span bridge with a focus to
examine information management and implementation strategies.
The challenges in structural identification and health monitoring of a major cantilevered
through-truss bridge were being explored in Philadelphia (Figure 1.13). The health monitoring
strategy is based on various experiments with controlled input as well as continuous monitoring
of ambient inputs and responses at the global, element, and localized levels. Triggered high-speed
scans of traffic inputs in conjunction with global and member level responses help determine the
live-load stress levels. Catbas et al. (2000) also observed that the important challenge for monitor-
ing of large infrastructure systems as the difficulties in the integration and interpretation of data and
information.
1.4 NEED FOR INFRASTRUCTURE HEALTH EFFORTS

1.4.1 Overview
The main objective of structural health efforts is to make sure that the structure is safe and secure
for its users during the intended use. The second objective is to optimize the level of service and
use provided the structure with minimum disruptions, minimize life cycle costs through effective
decisions through just-in-time operations, avoid catastrophic failures, improve safety, evaluate new/
Bridge health monitoring
Service/operation Maintenance Repair-retrofit
Intrinsic force Damage

Evaluation of
due to fabrication detection
operation & service;
& construction maintenance
effectiveness Condition indices: Influence
Slow speed Continuous
event-based coefficient ensemble
monitoring of: Changes in intrinsic strains
monitoring ambient
traffic
accident
wind
earthquake
Diagnostic Diagnostic Diagnostic

tests, NDE tests, NDE tests, NDE
Damage indices: Displacement (flexibility) influence coefficients, strain

influence coefficients, derivative indices, local response
FIGURE 1.12 Infrastructure health monitoring. (With permission from CRC Press.)
Introduction 13
Views of the principal structural system

13,912′
PA NJ
Steel stringer Deck truss Cantilever truss Deck truss Steel stringer
approach approach approach approach
Dampers
822′ 1644′ 822′
Cantilever truss
Lower chord of main truss Deck truss approach Steel stringer approach
FIGURE 1.13 Commodore Barry Bridge. (With permission from CRC Press.)
innovative practices/materials, better understand performance, and effects of factors influencing

structural/operational performance to continuously make improvements and changes as needed.
In simple terms it can be stated that the need for structural health is to make better and effective
management decisions. This incorporates all the others mentioned above.
Structural health evaluation, through visual inspections or using sensors and instrumentation,
when effectively designed, used, and implemented can lead to improved reliability in expected
performance of the structure in cost-effective manner. In the long run, as will be shown and illus-
trated in this volume, it can lead to system reliability approach than component reliability approach
and lead to multihazard, multidisciplinary, performance-based approach. This means unnecessary
maintenance, inspection, and other operations, both scheduled and periodic, can be avoided to
reduce costs by better planning and execution. This will allow for making appropriate decision at
all levels—component level, individual structure level, and network level. This also will lead to
improving the state-of-the-art and practice through research, and to bridge the gap between both.
In essence, effective structural health evaluation can also lead to innovations both in technology and
operations, can bring effective cultural change in organizations, better quality of life with increased
mobility and reliability, and improve communications and coordination between all stakeholders.
1.4.2 Specific Needs

Structural health in civil engineering principles evaluate, periodically or continuously, the present
health of the structure and compare it to baseline or expected performance to make appropriate deci-
sions on its future. Some of the needs for IHCE activities, as described above, are given below:
A. Safety improvements
B. Assuring security
C. Improve reliability, as shown in Figure 1.14, both at component and system (network)
level
D. Improving decision making on operations
a. No action
b. Posting
c. Decommissioning
d. Determination of appropriate inspection activities
i. Inspection interval
ii. Type and nature of inspection
iii. Required expertise for inspection
iv. Estimation of cost for inspection
e. Appropriate maintenance activity
i. Type
ii. Interval
iii. When and where?
iv. Cost
f. Repair or rehabilitation
i. When?
ii. Type?
iii. Cost
iv. Type of assessment required
g. Replacement
i. When?
ii. What type of structure?
iii. Cost
iv. Type of assessment required
E. Research
a. Durability evaluation
b. Life cycle costs
c. In-service performance
d. Evaluation of new design concepts
e. Evaluation of new materials
f. Evaluation of new construction methodologies
F. Evaluation of in-service environment
a. Hazard characteristics
b. Vehicular characteristics
c. Environmental characteristics
Infrastructure quality of service
Reliability
Cost of operations
Infrastructure life span

No IHCE With IHCE
FIGURE 1.14 Qualitative SHM effects on structural reliability.

Introduction 15
G. Bridge management system

a. Network level analysis
b. Project level analysis
H. Risk analysis: real (observed) versus conceptual (theoretical)
1.4.3 Case Studies for Uses of IHCE

1.4.3.1 Cost and Bridge Management
One of the advantages of IHCE is its potential in helping bridge managers reaching efficient deci-
sions. Sazonov et al. (2004) discussed costs of SHM and potential help to bridge management tools.
Some of the authors’ observations were as follows:
1. Cost of monitoring systems, from literature, varied significantly. For example, Lynch et al.
(2006) utilized 60 accelerometers in California at a reported cost of $300,000 (i.e., $5,000/
sensor) while the Tsing Ma suspension bridge instrumentation with 600-channels had a
reported cost of $16 million (i.e., $27,000/channel).
2. Authors note that the two popular BMS software used, PONTIS 2005 and BRIDGIT (Hawk
and Small 1998), attempt to predict the remaining bridge life through generation of life
cycle curves, using empirical data inputs. In PONTIS, bridge population is represented on
a network level by individual bridge elements with field inspection data providing numeri-
cal condition states for each element. A probabilistic Markovian chain prediction model
is applied for each element to estimate proportion of each bridge element that is expected
to deteriorate in the next inspection cycle. A rank order of element condition states in any
inspection cycle leads to an application at the bridge level. The BRIDGIT also applies a
Markovian prediction model at the element level, but the optimization model is more bridge
specific and thus addresses the element interaction more extensively than PONTIS.
3. On the basis of a comparative study of vibration signal analysis using short-time Fourier
transform and Hilbert-Huang transform (HIHT) spectrum, authors note that HIHT is more
sensitive to the dynamic energy-frequency distribution and is capable of capturing the dif-
ference in the structural response caused by the damage while the short-time Fourier trans-
form indicates the natural frequency changes but fails to capture magnitude changes.
1.4.3.2 Benchmark Bridge Behavior

Four bridges in Connecticut state were instrumented by University of Connecticut, Olund et al.
(2006), to monitor the effects of ambient loading on bridge behavior in a network of bridges criti-
cal for CT state’s infrastructure. A steel multi girder bridge, a segmental post-tensioned box-girder
bridge (see Figure 1.15), curved cast in place post-tensioned box-girder bridge, and a steel box-girder
bridge were instrumented with variety of sensors with specific initial reasons and long-term moni-
toring. This data is also being used to create benchmarks for each bridge and then will be used as
reference for future monitoring.
1.4.3.3 Inspection and Rehabilitation

Fisk et al. (2008) described a method for inspecting the Jamestown Verrazzano Bridge over
Narragansett bay, RI. The bridge features a 4950-ft-long prestressed segmental box-girder main
bridge with 23 spans varying in length from 109 to 636 ft, and a 2402-ft-long trestle structure. The
bridge was open to traffic in 1992 and nondestructive and destructive methods were used to investi-
gate the post-tensioning ducts for the presence of voids.
Ground penetrating radar (GPR) was first used to locate the centerline of the ducts (Figure
1.16). Then, impact echo technique was used to identify voided tendon ducts by running a four-
sensor array just ahead of the impact signal device along the centerline marked out using the
83′–0′′
11'–0′′
36′–0′′
FIGURE 1.15 Typical cross-section of the segmental post-tensioned box-girder bridge. (Reprinted from
ASNT Publication.)
FIGURE 1.16 GPR investigation of box bottom slab using 1500 Hz antenna. (Reprinted from ASNT
Publication.)
GPR. Where voids were identified by impact echo (Figure 1.17), small diameter holes were
carefully drilled through the concrete cover to the surface of the sheet metal duct, which was
carefully peeled away to reveal the duct interior. Where a void was confirmed, a bore scope was
used to document the size and length of the void, and the present state of the tendon. Several
tendon anchorages were investigated with drilling and borescope inspections (Figure 1.18) as
these anchorages are located in thicker concrete diaphragms and anchor blocks, GPR and impact
echo are ineffective with these components.
Over 93,000 linear ft. of nondestructive impact-echo (sonic/ultrasonic) measurements were taken
on the concrete top slab, webs, and bottom slab containing the tendons to evaluate the grouted ten-
don ducts for voids. Of the approximately 1520 tendon ducts tested, 7.5% or 114 tendon ducts were
determined to have voids. Void lengths ranged from 1 ft to over 314 ft. In most cases the tendons
were grout covered but some of the tendons were exposed and exhibited corrosion.
1.4.3.4 Monitoring Specific Bridge Types

Lloyd et al. (2004) offered some basic details of a monitoring system used, experience gained, and
thoughts on how to improve the field based on this experience. The subject bridge is Kishwaukee
Bridge in Rockford, IL. It is a five-span segmental post-tensioned concrete box-Girder bridge and
Introduction 17
FIGURE 1.17 Impact-echo air void detection using four-sensor array. (Reprinted from ASNT Publication.)
FIGURE 1.18 Borescope with video recorder and monitor. (Reprinted from ASNT Publication.)
showed extensive cracking adjacent to the piers in the web immediately after the erection. Thus the
goal was to understand the crack origin and the impact of the cracks on the performance and life of
the structure. Modal testing proved to be unsuccessful in understanding the relation between cracks
and the structural performance due to their inability to connect to the deformational responses to
the loading conditions. But this testing was useful in developing calibrated FEM models.
Monitoring system objectives were to continuously measure several hundreds of sensors to
indicate long-term changes that affect load-carrying capacity, collect temperature and traffic load-
ing conditions, statistical and expert system for postprocessing, and provide meaningful display
tools. The system included a field data acquisition system (provide for continuous operation, data
archiving tasks scheduling, data reduction to minimize data transfer, and allow remote configu-
rability), Integrated Services Digital Network (ISDN) line for data transfer, and an office system
(provides long-term archiving, user interface, and report generation capabilities). A bridge engineer-
ing interface was provided to access data records with appropriate information (sensor information,
data type, structural location, etc.) and statistical measures as needed.
A “confidence methodology” was used for uncertainty analysis of different data collected. This
scheme was used to select appropriate warning thresholds.
Analysis of data showed substantial influence of temperatures and showed the value of tempera-
ture compensation on the various data collected. On the basis of the experience, authors suggest that
compensating for thermo-mechanical effects is important and should be well understood and consid-
ered in data-driven modeling techniques, load-carrying capacity methodologies need refining, and
recommend development of systematic procedures to verify the response under various scenarios.
1.5 ANALOGY WITH HUMAN HEALTH

Most resources are (were) spent on health-care issues in the United States and probably in the
entire world than any single subject. Significant monies are expended for R&D and monitor-
ing to provide new resources to keep human bodies feel better, preventing measures to make
sure future health-care costs are lower, and extending the overall life. This is evident if one
compares any of the following data for now versus 50 or 20 years ago: the average life span,
number of working days lost due to sicknesses, quality of life, and so on. One can learn quite a
bit by comparing the efforts from medical/health industry to structural health in civil engineer-
ing (Figure 1.19).
Let us review the health industry first and then make some observations.
1. The care for a human being starts as soon as it is conceived in a human body. General
medical practitioner (MD) is the one normally consulted first to assure that one is pregnant.
As soon as it is confirmed, mothers are advised not to drink and smoke, to eat well, to take
prenatal vitamins and other medications, advised to make regular visits to special doctors
(gynecologists), etc. If there were any previous complications, they are referred to special-
ists right away. But, features and future cannot be controlled.
The structure is born in similar ways but probably requires specialists to start with—planners,
surveyors, geotechnical engineers, designers, etc. The shape, size, materials, details, aesthetics, fea-
tures, and so on can be controlled, but future can not be controlled like human future and depends
on both controllable (humanistic) and uncontrollable factors (hazards).
2. Once a baby is born, certain data is collected and documented for identification as well
as for use as baseline health data for future use as well as for statistical (census) purposes.
Mandates require that these be done in certain periods. If any of the data is abnormal, more
testing is done. At the same time, some medications/vaccinations are given to avoid serious
problems based on probability of these problems during an expected life span. Figure 1.20
shows the analogy between human and bridge testing scopes.
Once the bridge is built, in the same fashion as a baby, the inventory data is collected for identifica-
tion as well as for future use and statistical purposes. Inspections are performed within 60 days as
mandated by National Bridge Inspection Standards (NBIS) to document its current condition. If any
issues are noticed corrective actions may be taken. At the same time, some measures may be taken
to improve its durability (such as deck sealers, etc.)
3. In the past, most people did not take any preventive medications or go to a physician unless
there is a problem. In some cases, this was too late to cure or very expensive. It was real-
ized that taking simple precautions or taking certain preventive medications help in this
Introduction 19
Human health Bridge health
Baseline/average As-built condition Baseline
General physical
(blood pressure, Routine
Visual inspection inspection
pulse, heart beats,
etc.)
Simple testing:
Blood tests, X-ray, Simple
ultrasonic, tests
etc.
thermography, etc.
Advanced testing:
Advanced testing: Advanced
vibration, acoustic tests
CATSCAN, MRI
emission, etc.
Advanced action: Advanced action:

surgery, retrofit, rehabilitation,
replacements, etc. posting
Remedial
actions
Emergency action:
close, decommision
FIGURE 1.19 Analogy of human and bridge healthcare processes.
aspect. It was also realized that having regular physicals is also useful to check on changes
in some routine indicators. The physical interval may change based on age, health, type of
problems, and other factors.
The same can be said about structural health too. For example, until late sixties there were no
national requirements for bridge inventory and inspection formats and intervals until a major bridge
failure with casualties occurred. Now there is a realization that, just like human health, structures
need inventory and routine inspections to make sure they are safe, did not change drastically from
previous physical (inspection), and to minimize surprises (sudden costs). There are efforts under-
way to have rational inspection intervals for structures too based on their age, condition, and so on.
Figure 1.21 shows the age of bridges in New York State.
4. During a routine physical, doctor does a thorough physical examination, checks for some
vital signs (such as BP, pulse, etc.), provides certain immunizations requiring periodic
shots or booster shots, and conduct routine blood tests.
The same can be said to structures too. If routine inspection reveals issues, further actions are taken
depending on the type of issue, seriousness of the issue, extent of it, and its effect on the person’s life
and quality of life. On the basis of an inspection report and simple tests (such as magnetic-particle
Local Testing
Humans: Bridges:
X-Rays Acoustic Emission
Ultrasound Ultrasonic
Specific Organ Fatigue Monitoring
etc. etc.
Global Testing
Humans: Bridges:
Computerized axial Proof load test
tomography (CAT) scan Modal Identification
General wellness etc.
Exercise
etc.
FIGURE 1.20 Analogy of human and bridge testing scopes.
State highway bridge data

30
Percent of total state highway bridges
25
20
15
10
0
Pre 1900 1900– 1910– 1920– 1930– 1940– 1950– 1960– 1970– 1980– 1990– 2000–
1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 Present
Year built
% Bridges
FIGURE 1.21 Age of bridges in New York State. (Courtesy of New York State Department of Transportation.)
test, sounding, or chain drag), certain recommendations are made. These include cleaning, deck seal-
ing, joint fixing, removing loose concrete, etc. More interim inspections also may be scheduled.
5. If physical examination reveals any problems, physician may conduct some more testing or
order further nondestructive testing such as X-rays, and CAT scans. Depending on diagnosis
of specialists who can interpret these, further analysis or examinations may be scheduled.
If inspector and simple testing can not reveal the problem or extent, then more serious testing may
be recommended. These include ultrasound testing or thermography, detailed analysis or load rating
Introduction 21
based on in-depth inspection, etc. This is an area where more work is probably required, that is,
not many NDT methods are available to detect the extent of problem easily in structures and more
progress probably is needed.
6. If these tests and simple trail-and-error solutions do not work, then serious testing and seri-
ous actions (such as scheduling a surgery) may be recommended.
If the structure requires extensive work, rehabilitation may be done. If not, a new structure is born.
This is one area, where the structure life may be terminated due to cost–benefit analysis, whereas in
human aspects cost is not an issue to save a life (in most cases).
7. With all the knowledge gained from and number of humans involved in the history and all
the care taken and money spent, people do die due to natural and man made hazards. The
same happens to structures too. Even though, the requirements for extending on life and
quality of life are known, based on individual economy and several competing priorities,
people take decisions that may not be in their best interest or have to make a choices where
short-term issues or more important. Same apply to structures and their owners, that is, the
choices are made based on resources at hand and hence solutions taken may not be based
on long-term basis. Just like physicians knowledge and experience dictate interpretation
based on several key factors affecting human health, the bridge owner’s experience, and
so on play a major role in a structure’s life and durability. Mistakes can bring premature
termination of the structure.
8. Decisions in health-care fields normally affect a single individual with effects affecting a
small group of people, that is, in most cases his/her immediate family. But, in the cases of a
structure, the decisions made can have far reaching effects, including affecting few people’
lives to quality of life of an entire community.
9. Redundancy is another factor that differs in structure versus human life. There is little
redundancy in human life/components in most cases. Any redundancy is fixed and can not
be changed, that is, a second heart or a third eye. Design of human life is not in our hands;
where as redundancy of a structure can be controlled. This gives a big edge to owners for
future maintenance and in controlling the structure life, safety, and durability.
10. We finally note that a healthy bridge offers pleasant esthetics, in addition to the optimal
safety and functionality (see Figure 1.22). In the long range, healthy bridges, and healthy
infrastructure in general also will operate at an optimal life cycle cost.
1.6 STRUCTURAL HEALTH: A NEW FIELD IN CIVIL ENGINEERING?

1.6.1 What Constitute a New Field?
Most civil engineering activities are reactive in nature. Even though the life span of the structure is
considered during the design, at any given time, most actions are considered based on current struc-
tural and loading conditions. The structural health concept considers the time dimension and hence
makes the whole field dynamic. Thus, conceptually, if fully implemented SHCE concept considers
past history, present condition, and expected future condition. Hence, it gives the complete story and
hence makes it a new field.
Thus, as illustrated earlier and later in this volume, SHCE concept is purely a decision making
tool based on required level of service to be maintained in its life span. Thus, it has all aspects of the
structure during its life built into it—hazards, performance, damage, capacity evaluation, remain-
ing life, inspection, and preventive and corrective actions. Simply, it is a new and efficient way to
make all these engineering functions making use of concepts such as risk management, benefit-cost
(value) analysis, and evaluation. Let us consider couple of examples here.
FIGURE 1.22 A healthy bridge offers pleasant esthetics as well as optimal safety and functionality:
Court Street Bridge over the Susquehanna River, Owego NY. (Courtesy of New York State Department of
Transportation.)
Currently, the bridges are designed for specific combination of loading based on specifications.
The capacity of the structure thus is not optimized for any hazard it faces. Ideally, using SHCE
concept implementation, in the future, one can estimate the hazards it is expected to face, estimate
probability of failure and consequence, and design the structure effectively to provide appropriate
capacity. This will yield uniform reliability and the better understanding gained in this phase can
be used to develop data needed to collect in the future and thus, can develop appropriate inspection
procedures, and required actions.
Using the SHCE concept, the entire NDE process can be improved based on what decisions have
to be made in the future. Thus, true understanding of the structure, its operating environment, and
loading has to be taken into account in developing the following:
• What decisions has to be made?

• What data is required to answer these questions?
• Is NDE required?
• What methods (for local or global testing and analysis) are required?
• What kind of sensors and equipment are required?
• Where do these sensors go?
• How do these get integrated into structure/network?
• How and when the data is recorded?
• How is data transmitted?
• How and when data is analyzed?
• Processing required at structure and decisions to be taken?
• Do all the above provide answers to the questions in hand?
• Are the resources available to make the above happen?
• If the answer to the above two questions is “yes,” then go to next step. If not, the whole
process should not be done or reformulated.
• What is the value or benefit?
• Is it worth the cost? If the answer is “yes,” then implement the evaluation plan.
This data from the evaluation system coupled with other knowledge base can be used to effectively
maintain and operate the structure with the evaluation system employed. The data from individual
Introduction 23
Structure # 1 Structure # 2
Monitoring and diagnosis system

Historic network data
(structural operating and
and knowledge base
environment data)
Maintenance and other

Network management
operations of structures 1 & 2
FIGURE 1.23 Network management.
structures can be generalized with historic knowledge base and expert systems to optimize the
entire network management (see Figure 1.23).
By effectively implementing this at individual component/structure level, this can be generalized
to and can be used for entire network as data is collected and can be used for future planning—that
is, go toward predictive and proactive civil engineering than the reactive engineering. Thus, the
SHCE is very different, that is, it yields a multihazard multidisciplinary, performance-based engi-
neering centered toward constant reliability, known risk, and value-based decision making process
by combining other relevant fields with civil engineering.
1.7 STRUCTURAL HEALTH: INTEGRATION OF FIELDS

As illustrated in the previous section, the structural health concept is a closed loop or an active
concept, that is, it starts and ends with decision making while integrating all aspects of structural
life and the environment it is part of. Thus, it is a new field in civil engineering with integration of
almost all fields and all disciplines. This section briefly mentions all these fields.
Most structures are built and stay in relatively harsh environment not shielded from natural haz-
ards (such as rain, snow, earthquakes, and wind) as well as man made hazards (such as blast). Thus,
before one can design and maintain the structures, there is a need to understand the magnitude
of hazards and how they behave. Even global warming needs consideration as loads expected on
structures decades ago may be a lot different now and in the future. For example, temperature dif-
ferences and flood loads may need reconsideration periodically. Hence, environmental engineering,
risk engineering, seismic engineering, wind engineering, and blast engineering are a few that are
needed to supply the data required on the environment structures are built. While lot of this data
can be obtained in the literature, IHCE offers a means to verify the models used by collecting the
data at the structural sites and provide a feedback to these experts to refine the models or to adopt
better data for future designs and maintenance. Not only for the data but also for the analysis of
structures, very good understanding of the hazard mechanisms is important to conduct optimum
design any analysis. This is discussed further in multihazards sections later in this volume and in
Ettouney and Alampalli (2012).
Data collection is mentioned above. Data collection again is an integral part of the SHCE.
Collection of data requires understanding of the behavior of structures at system and component
level. Determining system versus component performance needs require good understanding of
the structural performance, deterioration aspects, etc. Material behavior is also very important to
make sure appropriate sensors are used and inspection methods employed are appropriate. Data
needs to be collected, transmitted, analyzed, and appropriate decisions are taken. Thus, materials
engineering, mechanical engineering, sensor and electrical engineering, information technology,

and decision sciences are some of the few disciplines required in this aspect. Innovations in one
field can affect the entire system and need to be accounted for and communicated well. One of the
authors had experience where a change in equipment in a telephone company made an entire moni-
toring system useless for 6 months. Thus, good project management skills also go a long way and
that is again another discipline.
Construction of structures of course requires several aspects. Regulating agencies have to be
dealt with, where changing role of government identities affect regulations and practices. Several
industries and specialties are needed for good construction where quality control and assurance
are very important. Manufacturing engineering, testing for quality control (QC)/quality assurance
(QA) during fabrication and on-site construction, safety specialists, hauling industry, and traffic
engineering disciplines are few that should be considered for moving toward just-in-time approach
that is the goal of SHCE.
Once the structure is constructed, inspection and maintenance is the key for durability and sev-
eral of the disciplines mentioned before and other expertises are required.
Resources are a big factor in any project and management. Due to the advent of internet and other
innovations, the entire world has became relatively small and understanding of global economy and
its implications on funding is very important for planning and other aspects. The increase in cost
of construction materials due to recent Olympics in China is a good example to illustrate this fact.
Since IHCE concept is based on benefit-cost aspects, finance engineering is a critical discipline in
the success of this concept.
Customer expectations and demands drive the need for structures and their performance. Hence
understanding the customers—demographics, urban/rural differences, cultures—and factors influ-
encing them at present and future are very important and needs to be included. For example, a
decade ago road users were happy to have a means to get to one place to other, where as now they
expect reliable travel times. After the 9/11 incident, security became another aspect to be considered
and has implications on all aspects of civil/bridge engineering.
A knowledgeable, competent, and adoptable workforce is very important for the well-being of
future infrastructure. Thus, education and training industry along with human resource manage-
ment fields cannot be forgotten. Thus, the SHCE is not a single field but integration of almost all
the fields available there. For effective bridge/structural management, all these fields should work in
harmony to meet the level of service expected from the bridges already built and that will be built
tomorrow under SHCE umbrella.
1.8 HORIZONTAL INTEGRATION: MULTIHAZARDS

1.8.1 Introduction to Multihazards
1.8.1.1 General
The main responsibility of professional engineers is to provide safety of the customers/public
utilizing the infrastructure. Other factors governing the designs include constructability, service-
ability, durability, maintainability, inspectability, economy, and aesthetics. Most of the designs
are done using the standard specifications or the guidance provided by the regulating agencies,
which state, in most cases, the minimum requirements to ensure the safety of the infrastructure
users. They also provide guidelines to consider the extreme events and dictate which hazards
should be considered as part of the routine design along with the return periods, which may be
significantly greater than the design life. In these cases the structures may be designed for sur-
vival alone or to meet different performance levels based on the importance of the structure or
its use in post-hazard situations. Thus, structures in general are designed to ensure strength and
stability, see Timoshenko and Gere (1961), to resist the significant load combinations they are
Introduction 25
expected to experience during their service/design life, and then checked for their survival under
extreme events.
Engineers routinely deal with natural hazards such as earthquakes, floods, wind, and ice. At the
same time, certain man made hazards, such as blast and impact loading, are also becoming very
important and were not considered seriously until recently. The severity of these hazards can sig-
nificantly increase the costs of construction and maintenance, especially if they are considered for
existing infrastructure requiring rehabilitation. In some cases, some of these severe hazards have
conflicting demands from the system, with the resulting increase in costs being even higher.
The primary reasons for not utilizing a comprehensive analysis under multihazard environment
include (a) some of the hazards were never anticipated or demand on systems due to these hazards
were not considered severe, and (b) lack of adequate computational tools and databases. Recent
improvements in computational tools, database population (that include behavior of systems under
several severe hazards), and SHM, all contributed to the increased consideration of multihazard
approach to systems.
Multihazard analysis and design is gaining popularity in the nation. This approach considers
increasing complexity of the structural systems to meet the demands of the current environment and
takes advantage of the recent developments and innovations in computing, analytical, and sensing
technologies. This concept requires engineers to understand the characteristics of various hazards,
their interactions, and the structural response under various hazards, so that appropriate analysis
can be conducted and structures can be designed to meet the expected performance to each of the
hazards considered. This understanding is very important not only for design purposes but also for
planning purposes, as these can affect the site selection, construction process, and future inspection
requirements.
Finally, it should be noted that there are two very distinct types of multihazard considerations
in the field of infrastructure: development and execution of emergency plans at local, state, and
national level; and engineering and economic considerations. This section deals with the latter type,
that is, engineering and economic considerations.
1.8.1.2 What Is Not Multihazards

Conventionally, infrastructure health professionals used to consider the effects of hazards on infra-
structure in a serial fashion: a one hazard at a time. The potential of more than one hazard affecting
the structure simultaneously is accommodated through load factors and load combinations equa-
tions. We immediately point out that this is not a multihazards consideration. This conventional
method is a serial one-hazard-at-a-time approach. Multihazards considerations accommodate the
effects of hazards, when they occur, on the structural response to other hazards. Several examples
will explore this concept later in this section.
1.8.2 Theory of Multihazards

As noted earlier, a multihazard outlook to systems such as buildings and bridges would provide
great benefits, but such an outlook is hindered by the lack of a perspective that presents a concise
and systematic view of the subject matter. Upon reflection on how physical structures behave, we
then propose the following theory of multihazards:
For a given system that is exposed to multihazards, there exists an inherent multihazards resiliency
within the system. This multihazards resiliency implies an interrelationship between the manners that
the system responds to different hazards.
The theory, as stated above, would help in giving a perspective, which is concise and systematic. In
the next section, we explore some of the ramifications of the multihazards theory. In all, we find that
by adopting the perspective of the theory, we can produce safer and more economical structures.
The theory of multihazards offers two doctrines for systems: inherent resiliency for all hazards
and the interrelationships between the manners in which the system responds to those hazards. By
utilizing these two doctrines, the potential benefits, as illustrated earlier, would be achieved dur-
ing structural analysis, design, life cycle cost analysis, risk analysis, and SHM. In what follows, we
investigate each of these fields and how multihazards considerations can be used to improve accu-
racy, safety, and cost reduction.
1.8.3 Multihazards Categories and Structural Health

Our definition of hazards can be man made or natural. They can be either load-induced (earth-
quakes, wind, flood, collision, etc.) or durability-induced (corrosion, fatigue, freeze-thaw, wear and
tear, etc.). As such, we offer that there are numerous ways to categorize those hazards as shown
in Figure 1.24. The main premise of this section, and the multihazards consideration concept
itself, is that there are interactions between hazards in almost every component that are shown
in Figure 1.24. The only interaction that is currently employed in conventional designs is the like-
lihood interaction. The likelihood interaction is the probability that hazards can occur simultane-
ously, or as a consequence of each other. This type of hazard interaction is well studies, and will not
be considered in this section.
The just-mentioned theory of multihazards deals with interaction between hazards through the
structure: that is the interaction through structural vulnerabilities. This type of multihazards inter-
action is discussed in several parts of this volume: for example, multihazards considerations of seis-
mic and blast hazards. Hazards also interrelate to each other through their consequences. This type
of interaction is also the same as the interaction of hazards through hazard management compo-
nents. For example: if a decision is made to mitigate against corrosion deterioration by using
FRP wrapping solution, this decision could affect the response of the system to other hazards such
as earthquakes.
Interactions of hazards can also be detected by studying how hazards are defined and how they
affect the infrastructure. Any hazard affects the system in a three-parameter space as shown in
Figure 1.25. The time (or frequency) distribution of the hazard is one of those parameters. The other
parameters are the spatial distribution of the hazard and the amplitude of the hazard.
Figure 1.26 shows a qualitative frequency-amplitude distribution of some conventional hazards
that affect infrastructure. Whenever the hazards range of effect intersects with another hazard, we
expect that the two hazards would interact through the system. If the hazard zones do not intersect
Structural health: Hazard viewpoint
Risk Hazard definition Hazard management

Temporal
Likelihood How is the hazard Decision making
applied in time?
Spacial Cost/benefit
Vulnerability How is the hazard
applied in space
Life cycle
Amplitude
Impact
severity of hazard Utility
FIGURE 1.24 Elements of hazards.

Introduction 27
Amplitude,
hazard
level scale
Spatial distribution of hazard;

structural geometry
Frequency
scale
FIGURE 1.25 Three-parameter definition of hazards.

High
Blast
Amplitude scale
Wind
Medium
Seismic
Machine vibration
Acoustic
Low
Very low Low Medium High Very high
Frequency scale
FIGURE 1.26 Frequency-amplitude distributions of hazards.
in that space, we expect the two hazards to act independently. Chapter 4 presents the spatial distri-
bution of some popular hazards.
Figure 1.27 shows some important infrastructure hazards and a qualitative measure of the impor-
tance of their respective interactions.
1.8.4 Multihazards and Infrastructure Health

1.8.4.1 General
According to the theory of multihazards, all hazards interact through the infrastructure. These
interrelationships between the hazards would require attention; it means that an action relating to
one hazard can have an effect on how the infrastructure responds to other hazards. Thus, multi-
hazards considerations are as important to infrastructure health as conventional individual hazards
considerations. Tables 1.3 through 1.11 show some examples of how hazards might interact through
the infrastructure.
Hazard matrix: interrelation
Wear & tear

Earthquakes
Corrosion
Overload
Fatigue
Gravity
Impact
Scour
Wind
Fire
Fatigue NA H L M H H M M H H
Corrosion NA M L L L M H H H
Scour NA L L H H H H H
Fire NA H H M L L L
Wind NA L M L L L
Earthquakes NA M M L L
Impact NA H M M
Overload NA H H
Wear & tear NA H
Gravity NA
FIGURE 1.27 Qualitative hazards interaction table.
1.8.4.2 Structural Health Monitoring

Perhaps a less obvious use of the multihazard theory of this work is the SHM field. However, recall
that the theory considers that a given system comprises numerous multihazard resiliencies that are
interconnected and interdependent, and that ignoring these interconnections and interdependencies
might lead to erroneous consequences. It is clear that the SHM field would have several utilizations
to the multihazard theory. In this volume, we present several of these utilizations such as optimum
sensors locations (OSL), duality principle, and serendipity concept.
1.8.5 Benefits versus Limitations

Given below are some of the benefits of multihazards considerations:
1. Potential for an economic design and construction.

2. A more accurate estimation of inherent resiliency of the system.
3. A more accurate treatment/estimation of life cycle cost of the system.
4. A more accurate analysis of the system
5. The multihazards design philosophy of a particular infrastructure project may affect the
entire network on infrastructure systems. For example, multihazards design of an impor-
tant highway bridge may improve the reliability of the entire transportation network of the
region that the highway bridge serves.
6. Optimization of the SHM to increase experimental efficiency.
Such benefits seem to be fairly obvious, yet there is one limitation that must be considered; it is
the lack of a concise set of tools that can accommodate multihazards considerations. This results
TABLE 1.3
Multihazards Table: Interaction with Fatigue
Corrosion Scour Fire Wind Seismic Impact Overload Wear and Tear Gravity
Introduction
Stress corrosion and Fatigue of Response of Wind-induced Seismic-induced Response of Repeated Repeated Repeated live
hydrogen chemical systems to only cyclic higher cyclic higher systems to overloading- environmental loads-induced
embitterment (in gels (in severe fires can stresses have stresses have demands of high induced cyclic demands’ cyclic higher
Chapter 4 by Chapter 1 by be affected by direct effects direct effects impact amplitudes higher stresses (thermal, stresses have
Ettouney and Ettouney and reduced fatigue on fatigue on fatigue can be affected by have direct freeze-thaw, etc.) direct effects on
Alampalli [2012]) Alampalli capacity of reduced fatigue effects on induced cyclic fatigue
can both have an [2012]) has a systems capacity of fatigue stresses have
effect on fatigue direct effect systems direct effects on
behavior of on scour fatigue
high-strength cables
TABLE 1.4
Multihazards Table: Interaction with Corrosion
Scour Fire Wind Seismic Impact Overload Wear and Tear Gravity
Corrosion in Limited Loss of strength or Loss of strength or Loss of strength or Loss of strength or There is a direct Loss of strength or
submerged interactions ductility (in ductility (in ductility (in ductility (in interrelation ductility (in
foundation can might occur high-strength cables high-strength cables high-strength cables high-strength cables between corrosion high-strength cables
be affected by between due to stress due to stress due to stress due to stress damage and wear due to stress
soil erosion that corrosion corrosion or corrosion or corrosion or corrosion or and tear (in Chapter 4 corrosion or
might happen and fire hydrogen hydrogen hydrogen hydrogen by Ettouney and hydrogen
due to scouring embitterment) can embitterment) can embitterment) can embitterment) can Alampalli 2012) embitterment) can
affect long-term affect long-term affect long-term affect long-term affect long-term
resistance to high resistance to seismic resistance to impact resistance to resistance to gravity
wind effects overload
29
TABLE 1.5
Multihazards Table: Interaction with Scour
Fire Wind Seismic Impact Overload Wear and Tear Gravity
Limited Loss of Loss of Loss of Loss of Lost of Loss of
interactions foundation foundation foundation foundation foundation foundation
might support due support due support due support due support due to support due
occur to scour to scour to scour to scour scour might to scour
between might have might have might have might have accelerate might have
scour and an effect on an effect on an effect on an effect on deterioration of an effect on
fire global global bridge global global affected global
bridge wind seismic bridge bridge foundations bridge
resistance resistance impact overload gravity load
resistance resistance resistance
TABLE 1.6
Multihazards Table: Interaction with Fire
Wind Seismic Impact Overload Wear and Tear Gravity
Fire damage, if Fire damage, if Fire damage, if Fire damage, if Fire damage, if Fire damage, if
not retrofitted not retrofitted not retrofitted not retrofitted not retrofitted not retrofitted
properly, can properly, can properly, can properly, can properly, can properly, can
reduce reduce resistance reduce reduce accelerate wear reduce
resistance to to earthquake resistance to resistance to and tear resistance to
high wind effects impact hazard overloads deterioration gravity loads
TABLE 1.7
Multihazards Table: Interaction with Wind
Seismic Impact Overload Wear and Tear Gravity
Wind and seismic loads has Local nature of impact loads Differences in Wear and tear There are
many similarities makes their interaction temporal and losses can interactions
(including the global with wind loads less likely. spatial affect wind between
nature of both loads). Thus On other hand, damages of distributions of resistance gravity
there is high degree of impact loads, if not treated wind and overload loads and
intersection between the properly, can reduce wind make interaction wind loads
two hazards resistance less likely
TABLE 1.8
Multihazards Table: Interaction with Seismic
Impact Overload Wear and Tear Gravity
Local nature of impact loads makes Differences in temporal and Wear and tear There are interactions
their interaction with seismic loads spatial distributions of losses can affect between gravity loads
less likely. On other hand, damages of seismic and overload make seismic resistance and seismic loads
impact loads, if not treated properly, interaction less likely
can reduce seismic resistance
Introduction 31
TABLE 1.9
Multihazards Table: Interaction with Impact
Overload Wear and Tear Gravity
Impact effects can reduce Wear and tear losses can affect There are interactions between
resistance to overloads impact resistance gravity loads and impact effects
TABLE 1.10
Multihazards Table: Interaction with Overload
Wear and Tear Gravity
Wear and tear losses can affect There are interactions between
overload resistance gravity loads and overload
TABLE 1.11
Multihazards Table: Interaction with Wear and Tear
Gravity
There are interactions between gravity loads and wear and tear
in spotty applications to an otherwise increasingly important subject. Thus, even though the infra-
structure community realizes this value, there are no objective roadmaps to tap on the immense
potential of the multihazards philosophy.
1.9 VERTICAL INTEGRATION: MULTIDISCIPLINARY

1.9.1 Different Disciplines and SHCE
As noted in previous sections, infrastructure or structural health in civil engineering is not simply
structural monitoring but is an integrated field where it is a means to make appropriate decision
from design to manufacturing to maintaining to replacement, that is, complete structural cycle to
assure safety and intended level of service in a cost-effective manner. In essence, the idea is to make
the entire infrastructure network smarter—that includes making structures as well as materials they
are made of are smarter and intelligent. As noted, in previous section, this involves several disci-
plines, from several fields, dependent on each others.
These opportunities also bring challenges in making sure all these personnel and technologies
works together seamlessly to improve the performance measures and decision making process to
achieve increased safety at lowest costs. This may need change in organizational culture, change in
thinking, and change in practices and work flow. Figure 1.28 gives a broader pictorial representa-
tion of the multidisciplinary nature of the SHCE. The broader fields include engineering, decision
makers, businesses, and others.
1.9.2 Treatment of SHCE Multidisciplinary Needs in

Infrastructure Health Environment
Infrastructure process starts with identifying a need by planners, who work with network data
and the level of service needed at global level. They work with finances available, using some
Structural engineer Regulators

Vendors Government
Consultants Owners
Materials Risk managers
Hazard specialist Users
Planners Other...
Decision
Engineering makers
Other Business
Enforcements
Finance
First responders
Marketing
Inspection
Management
Construction
Economists
Research
IT
Other...
FIGURE 1.28 Interaction of multidisciplines during an SHM effort.
preliminary engineering estimates derived based on previous experience. Hence, effective SHCE is
very important to provide these data that are needed to make a decision on to develop or expand or
remove a new highway or a structure, replacement/rehabilitation/maintenance options, do nothing,
etc. based on current and expected performance. Once several options based on social, economi-
cal, political, and engineering aspects are considered, a decision to continue with detailed studies
is made. Environmental and engineering aspects based on regulations, codes, loads, and hazards
the structure may face during its lifespan will shape the behavior of the structure or network along
with financial and other factors. The detailed studies phase needs hazard specialist, risk evaluators,
engineers, material personnel, etc. As the project continues with the construction phase, it requires
quality control aspects that itself is a combination of personnel from several disciplines. After the
construction, come inspection and maintenance, and eventually planning. In case of an unfortunate
incident, the network or system level of service depends on informed users, enforcement, and first
responders. The economy also dictates to the user needs and the finances available to all facets of
this field. This shows the multidisciplinary needs in infrastructure health environment.
Effectiveness of the process depends on how effective the communication is and what data is
available at every step of the process as a decision has to be made. Thus, effective SHCE process
can assist this decision making process by completing the cycle, that is, start with what is an issue
and end with making an optimal decision to solve the issue.
1.10 SHM AND IHCE: A QUICK PRIMER

We argued earlier that there is an analogy between human and infrastructure health. One of the
major components of health of both is the need for diligent monitoring. In case of infrastructure,
this means the need to have a reasonable knowledge of the status of the infrastructure. Infrastructure
status would degrade as a function of passage of time and/or exposure to different hazards. In many
situations, the signs of such degradations may not be visible. In addition, the functional and perfor-
mance demands from infrastructure change with time. Again, in many situations those demands
may need monitoring (e.g., the truck loads on roadways or bridges). The knowledge of the status
of the infrastructure as a result of monitoring is generally referred to as SHM. SHM would result
Introduction 33
Labor
System
(bridge,
Economic tunnel, etc.)
Sensing
Technical Decision Structural

making Monitoring
ID
Other Damage
Social ID
Political
Network
FIGURE 1.29 Combining monitoring and decision making to form IHCE/SHCE.
in estimates of structural properties such as stiffness, strength, mass, damping, or ductility. These
general properties include, but are not limited to, the knowledge of damages that can affect struc-
tural performance. For our current purpose, we define damage as the degradation of a structural
property from its intended state. Damages can be fatigue cracks, corrosion loss of mass, inelastic
performances, or scour holes near bridge foundation. Until recently, the knowledge of structural
state has been obtained on a qualitative basis. The advent of modern sensing and NDT techniques is
changing the methods of monitoring structural states into more quantitative methods.
Clearly, the knowledge of structural states is only part of maintaining the health of the infrastruc-
ture. The remaining question is: what to do with such knowledge? There are many possibilities that
range from doing nothing to replacing the infrastructure completely. Factors that can affect deci-
sions are numerous. These factors can be economical, labor, social, technological, and/or political.
Among other factors that can affect decisions are infrastructure network and uncertainties involved
in all of these factors. After a decision is made, such a decision will affect the state of the infra-
structure on hand (a null decision can also have an effect on the infrastructure). When the process is
combined with SHM, we will call it SHCE/IHCE. Figure 1.29 shows the combined process.
We explore this process and the interrelationships between its components in full detail in
Chapter 2. The remainder of the volume will discuss many of the complex components of SHCE.
1.11 CHALLENGES AHEAD

1.11.1 General
There are several advantages in using infrastructure health in civil engineering concepts. But, to
move forward with more proactive bridge management concepts for effective decision making,
there are several technological, organizational, and cultural challenges ahead.
The major challenge will be the cultural and organizational changes required in the structural
engineering organizations. Most owners of structures such as bridges are public entities with no
profit motivation, rightfully so, to assure safety first. These owners include state transportation
agencies with better resources, experience, and expertise to small local owners, owning one or two
bridges with very limited resources and expertise. Thus, type of structures built, maintenance, and
other operational decision taken vary considerably. Hence, standardization and implementation of
advanced processes is not only difficult but may not be affordable unless it can be justified in a
rational and systematic fashion using risk and cost–benefit analyses.
The second big challenge is the diversity of infrastructure and people involved with building and
maintaining them. They are in general unique as they are designed and constructed to suit the physi-
cal field settings and also to make sure mobility of the network is not affected. Manufacturing of
components and field construction is not limited to very few companies around the world as the case
for aero space industry, as literally thousands of organizations are involved with infrastructure (say
bridges, for example) construction. Thus, standards and regulations are developed conservatively in
most aspects to account for the variability associated with this fact. Most standards are prescriptive
and thus are not performance based. For example, concrete mix specifications are specified by an
owner in such a way that the minimum concrete strength is maintained (say 3000 psi). Thus, the
real concrete strength could be anywhere from 3000 psi after 24 hours of placement to as much as
8000 psi or more. Thus, for most designs nominal conservative values of material properties are
used that in most cases works well and absolves reactive based management procedures. Given, the
resources are not dependent on the needs, this system works well. For a proactive and just-in-time
management approach, there is a need for resources that can be planned and controlled well for the
life of the network. SHCE advocates should be able to articulate this well, for making progress, by
convincing not only the bridge owners but also the regulators and funding sources.
As noted, there are several stakeholders involved with SHCE that will be dealt in next sections.
Understanding each others and what other stakeholders expect, and having a common understanding
is another big challenge. There has also been overselling the automated SHM by several proponents
of these concepts in the past and falling short on deliverable. Through effective communications,
these should be avoided to make progress. Realistic goals should be set with common understand-
ing with all stake holders involved to moving forward with these concepts. All should realize that
automated SHM concepts are complementary to existing tools and optimizing the use of all tools
based on the decisions to be made in a cost-effective manner should be emphasized and understood
by all the stakeholders. Risk and cost–benefit analyses may seem like rudimentary concepts but are
not routinely used in change management when people try to push the SHM technology to own-
ers. These should be emphasized more along with the value of these advances more and forms the
fundamental reason for SHCE.
Besides these challenges, there are several other technological challenges that have to be dealt
with. Unlike, aerospace industry, the civil structures are spread around geographically and cannot
be taken to a site where maintenance and other operations can be performed. This may seem simple
but poses several logistical and technological challenges. Logistically, since these structures are not
reproduced in mass like aeroplanes, one should have specific understanding of not only the mate-
rial behavior but also the global and local structural behavior before appropriate proactive activities
can be performed. There may not be power or telephone or internet connectivity at many of these
sites; accessibility maybe an issue due to the required work zone control, disruptions it may cause to
users, or due to the features under or over the structure. This limits type of technology and resources
that can be used. Hence, for example, there is a big push for passive technology. There is also a need
for portability and other aspects for sensors and other equipment specifications. Replacement or
rehabilitation of components is also not as simple as in aeronautical fields unless future design and
construction practices change significantly as we move forward with IHCE concepts.
Another major challenge is the type of workforce required and thus the change in educational
institutions. At present, most civil engineers leaving the 4-year colleges are only aware of design
Introduction 35
aspects and not other areas of civil engineering. But, in reality, even though the structural under-
standing is a must, very small percentage of graduates will work in design areas. All other areas
such as maintenance, inspection, analysis, planning, management, instrumentation, and other new
technologies are not emphasized or made aware in the colleges. For effective progress, a radical
change in the entire educational system is needed and should be well thought. All stakeholders
should pay considerable attention to this aspect to make good progress in positive direction. Owners
also should consider and change their hiring practices to accommodate the current state of the field.
For example, in most transportation departments, there are no engineers specialized in corrosion
even though corrosion is a major contributor to most problems faced by the bridges. In most cases,
the civil engineers with no background in corrosion principles end up solving corrosion problems
solely because they have some experience in dealing with it as part of their job. The same can be
said about maintenance, research, constructions, risk analysis, and other aspects too. Even many
organizations that have significant authority on the practice and specifications select their mem-
bers based on the position irrespective of the knowledge and experience required to make such
decisions.
In addition to all above challenges, there are other global challenges that can affect infrastructure
health in numerous ways. For example, in the field of energy, energy consumption of (a) structural
health technologies, and (b) manufacturing of construction materials, and construction methods
need to be optimized and addressed. The sustainability issues of infrastructure are of major con-
cern. Lately, efforts of making infrastructure green and environmentally friendly have just started.
Another major challenge is infrastructure security. Security has emerged lately as a concern to
infrastructure decision makers. This subject is discussed in more detail in Chapter 11 by Ettouney
and Alampalli (2012).
The above are a few challenges to consider as we move forward with proactive multihazard per-
formance-based bridge management philosophy using SHCE concepts. But in the authors’ opinion,
all these can be achieved with improved communications between all stakeholders. Infrastructure
owners have a key role to play in this by taking the lead and working with all other stake holders in
an open format.
In the remainder of this section we discuss two major challenges to infrastructure health field:
manual versus automatic inspection and climate change.
1.11.2 Manual versus Automatic Inspection

Manual inspection of infrastructure has been the mainstay of inspection processes since humans
started building infrastructure. As sensing and other technological methods emerge, there has been
slow utilizations of automatic inspection. These utilizations will increase as technology improves.
For example, there are many potential advantages to using automated ultrasonic testing (AUT) for
inspecting welds versus performing manual weld inspection. These include permanent record of
inspection data, higher probability of detecting flaws, faster inspection times, less operator depen-
dence on the results, and very accurate sizing. Disadvantages to AUT inspection are higher equip-
ment costs, increased setup and calibration time, requires special calibration blocks, additional
operator training and experience, and lack of specifications. Johnson and Bell (2008) attempted to
verify the ability of AUT to perform acceptably on a typical AWS type component through a special
qualification block fabricated with intentional weld related defects in the weld and heat-affected
zone. This block contained both longitudinal and circumferential welds (Figure 1.30).
A technique combining TOFD (time of flight diffraction) and pulse echo UT methods were
employed to obtain 100% volumetric inspection of the weld area. Lack of fusion, porosity, inclu-
sions, cracks, lack of penetration, and transverse defects were put in the areas of the weld where
they would be typically generated, including several defects in areas that are known to be difficult
to inspect. This was done to test the limitations of the inspection techniques. According to the
authors, all of the flaws in the weld area were detected with the AUT inspection. And sizing of the
FIGURE 1.30 Qualification block on rotating stand. (Reprinted from ASNT Publication.)
AUT detected flaws was in close agreement with the manual UT measurements and the intended
flaw size.
The main challenge in the field of infrastructure health is to obtain a reasonable balance between
automatic and manual inspections. Such a balance would include cost, technology, and accuracy
balances. This volume will discuss those issues whenever appropriate.
1.11.3 Climate Change

There is a strong relationship between infrastructure health and climate. Since early times, the
humans designed and built their infrastructure with the prevailing climate in mind. Notice the
differences between early building properties in the arctic region and the equator. In addition
to building properties, humans learned the hard lessons of accommodating, or ignoring, loca-
tions and material properties of infrastructure as they relate to climate changes. Two obvious
examples come to attention from classical Egypt. The ancients choose location of the great pyr-
amids on high grounds in the Nile Valley: they survived for more than five millennia. When the
ancients chose to build the magnificent classical Alexandria without regards to location, sizable
section of ancient Alexandria lies now a few feet below the Mediterranean Sea as a result.
More recently, two important changes have occurred in the field of civil infrastructure. First:
there has been a revolution of construction materials: the utilization of modern concrete and steel
has been predominant in the last two centuries. More recently, the expectations of longer life spans
of modern infrastructure became predominant. A parallel ominous change occurred on a global
scale: it is becoming clearer now that earth climate is changing. As a direct consequence of the
confluence of those three factors, the effects of climate change on longer-lived infrastructure is
quickly emerging as a major challenge in the field of infrastructure health. Generally speaking cli-
mate change demands would increase as the capacities of infrastructure decrease (as they age). The
concept is shown in Figure 1.31.
There are numerous manners climate change can affect infrastructure. Some of those are sum-
marized below:
Rising sea water: Melting of polar caps and other environmental effects would result in ris-
ing sea water levels. The effects on bridges and coastal structures can be immense. Those
effects can be structural and architectural.
Corrosion: Increased rates of corrosion of reinforced concrete in submerged columns/foun-
dation (effects can be most noticeable in causeways) in low coastal areas: low areas that are
dry now will be attacked by sea water, with potential acceleration in deterioration rates.
Introduction 37
2-... their capacities

decrease 3-... while the
climate change
demands normally
increase
1-As civil
infrastructure age...
Time
Note: All infrastructures “age,” even newly constructed ones:

They start aging immediately after they are constructed
FIGURE 1.31 Climate change and aging infrastructure dilemma.
FIGURE 1.32 Extent of Hurricane Ike (2008).
Direct pressure loads: These can result in added hydrostatic pressures on retaining walls.
This might result in the need for redesigns of existing systems.
Buoyancy problems: Changes in water levels can result in emerging buoyancy problems on
structural foundations.
Clearance problems: Increased water levels can result in head clearance problems under low
level water crossings.
Direct forces of flood and winds: Climate changes can result in an increased severity of
floods and winds (Figure 1.32). Redesign of systems might be needed. Scour problems
needs to be reevaluated. Storm surge can have negative effects on bridges (Figure 1.33).
Debris impacts during floods might become more severe. Similarly, wind borne objects
can be more severe.
Temperature rise and increased fluctuation ranges: Effects on expansion joints, and inte-
gral bridge designs might have major effects.
Deterioration issues: Potential increased deterioration rates due to sulfur attacks, corrosion, and
so on. Prestressed systems can have additional shrinkage, creep: resulting in loss of prestress.
1-Rotating 2-..generates low pressure at

4-The high surface wind speeds will
hurricane wind.. the water surface
drive the surged water level even
higher
3-The low pressure will
generate a surge in the
Normal open Normal open
surface water level
water surface water surface
Storm surge level records 6-There are two additional factors that affect
-27.8 ft. (Hurricane Katrina, 2005) storm surge levels:
-24.6 ft. (Hurricane Camille, 1969) a-Wave motion of water surface, and
b-Rotation of earth (Coriolis acceleration effect).
-43.0 ft. (Cyclone Mahina, 1899)
Storm surge Inland extent of storm surge damage
level
Normal open
water surface
5-As the surge approached land,

the surge height increases. The
increase is a function of the
slope of the shore lines
FIGURE 1.33 Anatomy of storm surge.
Society
Long term Short term
Hazards Capacities
Stakeholders Demands
Costs
FIGURE 1.34 Integrated approach in meeting climate change challenges.
Detailed discussion of the above climate change effects on infrastructure are beyond the scope
of this volume. However, many of the subjects we discuss later can be applied directly when the
decision makers try to address climate change effects on the health of civil infrastructure. We
note that one of the basic tenets of this volume is the integrated approach in solving infrastructure
health challenges. Same can be said about climate change challenges for infrastructure, as shown
in Figure 1.34.
1.12 APPENDIX I: DHS INFRASTRUCTURE SECTORS (AS OF 2009)

For ease of managing the vast infrastructures inventory of the United States, the U.S. Department of
Homeland Security (DHS) categorized infrastructures into several sectors. Table 1.12 shows those
sectors and the U.S. agency lead for each sector.
Introduction 39
TABLE 1.12
DHS Infrastructure Sectors, US DHS (2009)
Critical Infrastructure/Key Resources Sectors Federal Sector-Specific Agency (SSA) Lead
Agriculture and food Department of Agriculture and Department of
Health and Human Services
Banking and finance Department of the Treasury
Chemical Department of Homeland Security
Commercial facilities Department of Homeland Security
Commercial nuclear reactors, materials and waste Department of Homeland Security
Dams Department of Homeland Security
Defense industrial base Department of Defense
Drinking water and water treatment systems Environmental Protection Agency
Emergency services Department of Homeland Security
Energy Department of Energy
Government facilities Department of Homeland Security
Information technology Department of Homeland Security
National monuments and icons Department of the Interior
Postal and shipping Department of Homeland Security
Public health and health care Department of Health and Human Services
Telecommunications Department of Homeland Security
Transportation systems Department of Homeland Security
REFERENCES
ACI (2008). Building Code Requirements for Structural Concrete and Commentary, American Concrete
Institute, Farmington Hills, MI.
ASCE, (2009), http://ciasce.asce.org/working-definitions, site accessed on March 20, 2009.
Catbas, F. N., Grimmelsman, K. A., and Aktan, A. E., (2000) “Bridge Health-Monitoring for Structural
Performance,” Structural Materials Technology: An NDT Conference, ASNT, Atlantic City, NJ.
Cheng, J., (2004) “Potential and Challenges of Using SHM in the Canadian Energy Industry,” Proceedings of
2nd International Workshop on Structural Health Monitoring of Innovative Civil Engineering Structures,
ISIS Canada Corporation, Manitoba, Canada.
Cheung, M. and Naumoski, N., (2002) “The First Smart Long-Span Bridge in Canada,” Proceedings of 1st
International Workshop on Structural Health Monitoring of Innovative Civil Engineering Structures, ISIS
Canada Corporation, Manitoba, Canada.
DeWolf, J., Mao, J., and Virkler, C., (2002) “Non-Destructive Monitorinh of Bridges in Connecticut,”
Proceedings, NDE Conference on Civil Engineering, ASNT, Cincinnati, OH.
El-Ragaby, A., El-Salakawy, E., and Benmokrane, B., (2004) “Field Performance of a Concrete Bridge Deck Reinforced
with GFRP Bars: Cookshire-Eaton Bridge,” Proceedings of 2nd International Workshop on Structural Health
Monitoring of Innovative Civil Engineering Structures, ISIS Canada Corporation, Manitoba, Canada.
Ettouney, M. and Alampalli, S., (2012) Infrastructure Health in Civil Engineering: Applications and
Management, CRC Press, Boca Raton, FL.
Fisk, P., Abrams, M., Kaufman, S., (2008) “Inspection and Rehabilitation of Jamestown - Verrazzano Segmental
Concrete Bridge,” NDE/NDT for Highway and Bridges: Structural Materials Technology (SMT), ASNT,
Oakland, CA.
Fritzen, C. and Bohle, K., (2000) “Parameter Selection Stratefies in Model-Based Damage Detection,” Proceedings
of 2nd International Workshop on Structural Health Monitoring, Stanford University, Stanford, CA.
Gangone, M., Kroening, R., Minnetyan, L., Janoyan, K., and Grimmke, W., (2005) “Evaluation of FRP Rebar
Reinforced Concrete Bridge Deck Superstructure,” Proceedings of the 2005 ASNT Fall Conference,
Columbus, OH.
Glisic, B. and Inaudi, D., (2002) “Long-Gauge Fibre Optic Sensors for Global Structural monitor-
ing,” Proceedings of 1st International Workshop on Structural Health Monitoring of Innovative Civil
Engineering Structures, ISIS Canada Corporation, Manitoba, Canada.
Hawk, H., and Small, E. P., (1998) “The BRIDGIT bridge management system.” Structural Engineering
International, (4), 309–314. International Association of Bridge and Structural Engineering (IABSE),
Zurich, Switzerland.
Johnson, W., and Bell, R., (2008) “Comparing Manual AWS UT Inspection to Automated TOFD and Pulse
Echo Inspection,” NDE/NDT for Highway and Bridges: Structural Materials Technology (SMT), ASNT,
Oakland, CA.
Lloyd, G. M., Wang, M. L., Wang, X., and Chen, H., (2004) “Components of a Real-Time Monitoring
System for a Segmental Pre-Cast Concrete Box Girder Bridge,” Proceedings, NDE Conference on Civil
Engineering, ASNT, Buffalo, NY.
Lynch, J., and Loh, K., (2006) “A Summary Review of Wireless Sensors and Sensor Networks for Structural
Health Monitoring,” Shock and Vibration Digest, 38(2), 91–128.
NPR (2011) “Transcript: Obama’s State of the Union Address,” http://www.npr.org/2011/01/26/133224933/
transcript-obamas-state-of-union-address, site accessed June 10, 2011.
Olund, J., Cardini, A. J., D’Attilio, P., Feldblum, E., and DeWolf, J., (2006) “Connecticut’s Bridge Monitoring
Systems,” NDE Conference on Civil Engineering, ASNT, St. Louis, MO.
Oshima, T., Mikami, S., Yamazaki, T., and Sato, M., (2002) “Application of Intelligent Monitoring System to
Bridge Health Diagnosis,” Proceedings of 1st International Workshop on Structural Health Monitoring of
Innovative Civil Engineering Structures, ISIS Canada Corporation, Manitoba, Canada.
Pontis 4.4. (2005) “Pontis Release 4.4: User’s Manual,” Prepared for AASHTO by Cambridge Systematics,
Inc., Cambridge, MA.
Rochette, P., Neale, K., Pierre, L., and Demers, M., (2002) “Monitoring FRP-Strengthened Structures : Lessons
from Applications in Quebec,” Proceedings of 1st International Workshop on Structural Health Monitoring
of Innovative Civil Engineering Structures, ISIS Canada Corporation, Manitoba, Canada.
Savard, M. and Laflamme, J-F., (2004) “Long-Term Monitoring of a Reinforced Concrete Bridge,” Proceedings
of 2nd International Workshop on Structural Health Monitoring of Innovative Civil Engineering Structures,
Sazonov, E., Janoyan, K., and Jha, R. (2004) “Sensor Network Application Framework for Autonomous Structural
Health Monitoring of Bridges,” Proceedings, NDE Conference on Civil Engineering, ASNT, Buffalo, NY.
Shehata, E., Haldane-Wilson, R., Stewart, D., Mufti, A., Tadros, G., Bakht, B., and Ebenspanger, B., (2004)
“Structural Health Monitoring of the Esplanade Riel Pedestrian Bridge,” Proceedings of 2nd International
Workshop on Structural Health Monitoring of Innovative Civil Engineering Structures, ISIS Canada
Corporation, Manitoba, Canada.
Sikorsky, C., Stubbs, N., Bolton, R., and Karbharti, V., (2002) “The Application of Structural Health Monitoring
to Evaluate Bridge Strength,” Proceedings of 1st International Workshop on Structural Health Monitoring
Taha, M, Kinawi, H., and El-Sheimy, N., (2002) “The Realization of Commercial Structural Health Monitoring
Using Information Technology Based Techniques,” Proceedings of 1st International Workshop on
Structural Health Monitoring of Innovative Civil Engineering Structures, ISIS Canada Corporation,
Manitoba, Canada.
Timoshenko, S., and Gere, J., (1961) Theory of Elastic Stability, McGraw Hill, New York, NY.
US DHS, (2009), http://www.dhs.gov/xnews/gc_1179776352521.shtm, site accessed on March 20, 2009.
Wang, J., Ni, Y., Ko, J., and Chan, T. (2002) “Damage Detection of Long-Span Cable-Supported Bridges,”
Proceedings of 1st International Workshop on Structural Health Monitoring of Innovative Civil
Weinmann, T., and Lewis, A. (2004) “Structural Testing and Health Monitoring for the Bridge Street Bridge
Deployment Project,” Proceedings, NDE Conference on Civil Engineering, ASNT, Buffalo, NY.
Wenger, L., Zhou, Z., Alwash, M., Siddique, A., and Sparling, B., (2004) “Vibration-Based Damage Detection
on Bridge Superstructures,” Proceedings of 2nd International Workshop on Structural Health Monitoring
Yamada, K., Ojio, T., Fukada, S., and Kajikawa, Y., (2004) “Use of Bridge Weigh-IN-Motion System With
Environmental Monitoring of Viaducts,” Proceedings of 2nd International Workshop on Structural Health
Yang, C., and Newhook, J., (2004) “Load Distribution as a Damage Detection Tool for Steel-Free Bridge
Deck,” Proceedings of 2nd International Workshop on Structural Health Monitoring of Innovative Civil
2 Elements of Structural Health
in Civil Engineering (SHCE)
2.1 SHCE: A NEW FIELD IN CIVIL ENGINEERING?

The field of structural health is a complex field. It includes many components, and cover numer-
ous disciplines, subjects, and concerns. The complexities of the field are also due to the fact
that many of its components and methodologies are fairly new to the civil engineering practice.
For example: wide use of nondestructive testing (NDT), though known in civil engineering
applications for long time, has not been an integral part of the practice. For proper and effi-
cient execution of SHCE, the methodologies of NDT must be utilized and applied. This would
require ingenious applications and perhaps some changes in NDT applications as they have been
practiced in other fields such as aeronautics or mechanical engineering. Similarly, structural
identification (STRID) or damage identification (DMID) methods have been utilized in other
fields, such as mechanical or aerospace fields. Both STRID and DMID methodologies and tools
must be retooled for efficient and successful applications in the SHCE field. In addition, there
are emerging design paradigms and concepts in the civil engineering field such as performance
based engineering, multihazards considerations, life cycle analysis (LCA) and component and
system resiliency. All of those paradigms and concepts need to be included in the SHCE field.
Their inclusion and interaction with SHCE field adds to the complexity and wide reach and
coverage of the field.
Triumvirate Concept and SHCE: It is clear that for successful implementation of a healthy
structure strategy, over the whole length of its life span, all of the above issues, and more, must be
considered. Moreover, the interactions between all of those issues through both the stakeholders and
the structures itself must be adapted properly. The triumvirate concept is shown in Figure 2.1. We
offer that this triumvirate is the basis of the SHCE field. This volume aims at studying the different
SHCE/triumvirate components and the interaction between them.
2.1.1 This Chapter

Since this is a first attempt to address the new field of SHCE, our first task is naturally to try to
define its components or categories. This is needed to ensure adequate, efficient, and successful
coverage of the subject. There are numerous ways to categorize or compartmentalize the a field of
SHCE. We try to use the general components as shown in Figure 2.2, as follows:
Human Health/Baseball Analogy: Chapter 1 introduced an analogy between human and

structural health. Such an analogy can be restated in terms that are more pertinent to struc-
tural health issues: the baseball analogy. The baseball analogy is surprisingly simple, yet
very convenient and encompassing to all the activities of SHCE.
Design Paradigms: Other essential components in the field of SHCE are the different civil
engineering design paradigms. We note that the conventional design paradigm: capacity/
demand (CD) paradigm has been in use from early times. More recently, the performance-
based design (PBD) paradigm is emerging. Both paradigms are related directly to SHCE.
NDT Methods: One of the bases of SHCE is the knowledge of the state of the structure
in real, or near real, time. To achieve this knowledge, field and laboratory experiments
41
Triumvirate concept of SHCE
Stakeholders
SHCE
Interaction
Components
Structure
FIGURE 2.1 Triumvirate concept of SHCE.
Components of SHCE
Value of information
Design Paradigms Physical components NDT methods
Capacity/demand Sensing
All methods and
techniques of NDT
Performance - STRID
based
DMID
Decision making
FIGURE 2.2 Components of SHCE/SHM.
are needed. The well-developed field of NDT must play an integral role in gaining this
knowledge.
Value of Information: The execution of all needed activities pertaining to the field of
SHCE will require additional costs and efforts that are not conventionally incurred. Of
course, there are benefits that will be gained as a result of those costs. The question is
whether those benefits outweigh the costs? In other words: how much value an asset man-
ager will gain from SHCE?
This chapter will discuss the above components of SHCE in detail. Subsequent chapters explore dif-
ferent topics related to each component and potential interactions between these components.
Elements of Structural Health 43
2.2 SHCE: BASEBALL ANALOGY

2.2.1 Health Monitoring from Start to Finish
Monitoring structures such as bridges, dams, and buildings has received significant attention in the
last decade. Several owners and researchers are actively pursuing development of monitoring sys-
tems, with reliable sensor technologies and remote monitoring systems for decision making. Most
of the system identification process thus far is concentrated in three phases, namely data acquisition,
structural identification (SI), and damage detection, and lacks decision making phase. This section
presents a comprehensive view on the engineering of structural health process integrating the deci-
sion making process.
2.2.2 Home Plate: Problem Definition

Both existing and new civil engineering structures and constructed facilities throughout the world
are subjected to continuous deterioration due to several reasons such as corrosion resulting from
exposure to different environmental conditions. Day-to-day service loads can initiate fatigue prob-
lems over the time. Moreover, infrequent actions such as earthquake, fire, and improper use or
maintenance result in acceleration of the structure deterioration. This deterioration causes signifi-
cant degradation of a structure’s mechanical properties and its ability to perform intended functions
without failure. Due to the immense importance and cost of infrastructure, the subject of structural
health and performance during the life span of the structure has emerged lately as a major issue for
the engineering community.
Health monitoring of structures is perhaps the most serious attempt to address this problem. In a
general sense, it involves three distinct phases: the measurement of performance phase, the STRID
phase, and the damage detection phase. Considerable research and several methods in the literature
address each of these three phases. We offer a detailed overview of these phases in Chapters 5, 6,
and 7.
Unfortunately, health monitoring of structures in the present form lacks a major component,
namely the decision making phase. As an example: assume that a certain damage profile has been
detected during a health-monitoring project for a specific structure. With the damage profile infor-
mation, the decision maker is faced with the question of what to do next. Thus, important factors
and necessary tools that could help that decision maker should be readily available so that proper
decision can be made.
In this volume, we propose adding the decision making process to the three health monitor-
ing phases to form an integrated SHCE field. This concept was first proposed by Alampalli and
Ettouney (2004). This field will then be composed of four phases: measurements, identification,
damage detection, and decision making. Figure 2.3 shows a schematic representation of this pro-
cess. Moreover, an integrated approach is proposed to the SHCE phases, meaning that all four
phases of the SHCE process should be considered simultaneously. We will show that failing to do
so may yield inefficient or even erroneous results in many situations.
A more general, and more realistic, paradigm of SHCE is shown in Figure 2.4. This general para-
digm allows for base stealing, that is, it allows for cross-communications between different SHCE
components, instead of the serial paradigm shown in Figure 2.3. Note that in the more general para-
digm, the decision making process is always available at any step of the process.
2.2.2.1 First Base: Measurements

2.2.2.1.1 General
The first phase in SHCE is the measurement phase and forms the basis for the entire SHCE process
(Figure 2.4). Measuring structural behavior for SHCE will have two main goals. First, it is desirable
to numerically identify the structural system, and estimate the state of structural damage, if any.
Second base
Damage Identification: Structural Identification:

• Global vs. Local • Modal Identification
• Type • Parameter Identification
• Extent • Non-Physical (Neural Networks)
• Effect on the system • Other?
Third base First base
Decision-Making: Measurements:
• Do nothing • Sensors
• Maintenance (type, timing)
• Instrumentation
• Repair (simple, extensive)
• Replace (Decommission) • Technologies
Home plate
BASEBALL PARADIGM of SHCE (Structural Health in Civil Engineering)
FIGURE 2.3 SHCM baseball analogy.
Ultimate goals
Cost reduction Improve safety
SHM project
Measurements
(sensors and
instrumentation)
Structural Decision making

identification tool chest
Damage
identification
FIGURE 2.4 General SHCE paradigm.
Often this requires structural measurements. The most obvious is selecting structural parameters,
which represent important structural response measures of interest. These include the following:
• Direct measurements: These relate to the structural performance and give an estimate of
structural strength or failure condition directly. These include, but are not limited to, the
following
• Structural displacements
• Velocities
• Accelerations
• Structural deformation, for example, strains
• Inclinations
• Indirect measurements: These include parameters that give the necessary information, on
structural parameters, indirectly through some correlations
• Measuring the acoustics of cable snapping in suspension bridges, and then finding the
location of the breaks in the cables by tracing the time of arrival of acoustic waves.
• Half-cell potential data from concrete columns can be correlated to corrosion rate.
In all situations, complete knowledge of the structural system is mandatory to collect successful
and efficient measurements. This includes all the structural functions, intended use, design loading,
required service life span, and economics of maintenance.
All of the above factors will affect the measurement phase in SHCE. In any experiment, number
of sensors nc is directly related to the economics of maintaining the structure as well as the struc-
tural function and the intended use. In addition, number of sensors and their locations are directly
related to the type of structure, loading levels, and environment conditions. Any structural measure-
ment scheme not accounting for these interrelationships can lead to inefficient/insufficient data, or
possibly measurement information that is not suited for the intended SHCE use. A brief discussion
of these interrelationships is presented in next sections.
2.2.2.1.2 Number of Sensors and Cost
In any SHCE measurements, the minimum number of sensors no that is needed to insure the required
quality of measurement information qo are related by the number curve as shown in Figure 2.5. In
addition, the cost of obtaining qo is Co. They are related by the cost curve shown in Figure 2.5. Thus,
for a required information quality qo, both cost and number of sensors are uniquely correlated. If
the actual number of sensors in an experiment nc falls below no then the measurement information
can be rendered insufficient. In addition, an upper bound sensor cost ceiling Cc ≥ Co also exists. It
is determined by economic factors that relate to the structure’s intended use and importance. Both
Co and no requirements need to be consistent for optimum SHCE process. If this proves to be infea-
sible, either the cost ceiling need to be raised, or the technical scope of the measurements has to be
reduced. Figure 2.5 illustrates this process.
2.2.2.1.3 Sensor Locations and Number
Another important step in the SHCE measurements is the optimum sensor location problem (OSLP).
See Cobb (1996) for comprehensive review of this important issue. It can be stated as follows: For
Cost of sensors
Number of sensors
Cost of
sensors
Co Number of
sensors
no
qo
Quality of measurement information
FIGURE 2.5 Cost, number of sensors, and quality of information relationships.

any given structural measurement problem, find no locations of nc sensors such that the information
gathered is optimum. Several methods are available for the solution of the OSLP. However, it was
argued (Ettouney et al. 1999) that any OSLP solution has to account for the usage and the environ-
ment of the structure in an integral fashion. Failing to do this may result in sensor locations that do
not yield optimum information. The same authors proposed goal-programming techniques to solve
the general OSLP. Ettouney et al. (1999) studied OSLP for seismic, dead, and live load conditions.
They showed that all loading conditions have to be considered in deciding upon sensor locations;
otherwise, erroneous or inefficient sensor locations might result.
2.2.2.2 Second Base: Structural ID

Once data from measurement phase is obtained, transformation of this data to understand the struc-
tural properties is necessary. This is widely known as SI problem. The structure can be identified
by several modes of information. For example, the structure can be identified by the distribution of
stiffness throughout the structure, that is, the stiffness matrix of the structure. Identifying natural
frequencies and mode shapes of a structure (modal analysis) is one of the popular SI methods.
Detailed discussion of SI methods is beyond the scope of this work. For more information, the
reader is referred to Cobb (1996).
Even with the existence of the large number of studies and methods for SI, it is wise to place SI in
the context of the larger issue of SHCE. A successful SI technique should be capable of addressing
the needs of the SHCE problem. For example, if the SHCE problem on hand is the corrosion of rein-
forcing steel in a reinforced concrete bridge, the employed SI method should be capable of address-
ing the initiation and propagation of corrosion, and structural deterioration due to corrosion.
This discussion also shows that the choice of a particular method of SI should also depend on the
type and degree of damage expected during a structure’s life span.
2.2.2.3 Third Base: Damage Detection

The third phase in the engineering of structural health is detection of damage that can occur in
the structural system. For example, offshore oil platforms need continuous monitoring for struc-
tural damage that might occur below the waterline from extreme sea conditions and ship impact.
Damage monitoring of bridges is increasingly necessary as the effects of corrosion, ship traffic,
earthquakes and, sadly, even terrorist actions continuously threaten the soundness of these vital
structures. During the last two decades, research has been focused on using the vibration char-
acteristics of a structure for structural damage identification (Faarar 1996; Ettouney et al. 1998).
The vibration characteristics of a structure may be defined by its modal parameters (natural fre-
quencies and mode shapes), which generally depend only on characteristics of the structure and
not the excitation, and may be determined from taking measurements at one or more locations on
the structure. Vibration signatures obtained before and after damages may be utilized to locate
the damage and estimate its severity (Alampalli 1998). Chapter 7 explores damage identifica-
tion issues from defining damage to various analytical and experimental methods available in the
infrastructure field.
2.2.2.4 Fourth Base (Home Plate): Decision Making

All data from the above three phases should answer the main reason why the structural health moni-
toring (SHM) was considered in the first place. These includes questions such as
• For a given damaged pattern in a structural component or a complete structural system,

what is the reliability of such a component or a structure?
• What is the remaining service life before failure?
• Is the failure imminent?
• Are the assumptions used in design, construction, and so on are valid?
• What loads are needed to cause failure?
• What are the predominant failure mechanisms under the expected loads of a structure’s life?
• What type of actions (such as maintenance, rehabilitation, or replacement) are necessary
and when?
Let us explore how structures degrade to illustrate the point. When an engineering structure is
loaded it will respond in a manner that depends on the type and magnitude of the load as well as the
strength and stiffness of the structure. Whether the response is considered acceptable, depends on
the requirements that must be satisfied (Melchers 1987). These requirements might include safety of
the structure against collapse, limitation of damage, magnitude of deflection, or any other such cri-
teria. Not meeting these requirements is considered a limiting state violation. Hence, the reliability
of a structure is defined as the probability of occurrence of the limit state violation at any stage dur-
ing its lifetime. This probability can be obtained from measurements of the long-term occurrence
of the violations on similar structures, subjective estimation, or by using small-scale prototypes and
testing them under different conditions. Alternatively, a measure of the structural degradation with
time, such as rate of corrosion, can be used to determine the time at which a structure will cease
to function as desired. The level of degradation at that time is referred to as threshold degradation
level. Elsayed (1996) classifies degradation models as physics-based and statistics-based models.
The physics-based degradation models are those in which the degradation phenomenon is described
by a physics-based relationship such as Arrhenius law, the corrosion initiation equation (Enright
and Frangopol 1998) or experimentally based results such as crack propagation or crack growth
model (Oswald and Schuëller 1983). The statistics-based degradation models are those in which the
degradation phenomenon is described by a statistical model such as regression. The description of
the advantages and limitations of the two types of models are beyond the scope of this study. An
example of the use of a degradation model in SHCE is as follows.
Suppose that a degradation/damaged state database for a structure, such as that shown in
Figure 2.6 can be assembled. The damaged state database of Figure 2.6 was assembled for a rein-
forced concrete bridge that is subjected to corrosion. It was obtained using combined analytical,
experimental, and statistical methods (Ettouney et al. 1999). Using the degradation model that was
6.E+03
0.E+00 1.E+03 2.E+03 3.E+03 4.E+03 5.E+03
Strength (Kip. In)
1
6
11
16
Y 21
ea 26
r
31
36
19
17
15
13
11
9
7
t
Uni
5
3
1
FIGURE 2.6 Deterioration data.

described by (Eghbali and Elsayed 1997), the reliability of the structure can then be estimated as a
function of time, as shown in Figure 2.7. Note that this damage model relies heavily on structural
function and usage.
This structural reliability information can then be used to make proper decisions concerning the
optimum course of action.
2.2.2.4.1 Cost Analysis
Perhaps the most important step in SHCE is the cost analysis phase. After reviewing reliability and
degradation data (similar to those of Figures 2.6 and 2.7), the question facing the owners of the struc-
ture will be; what should be done next? The answer can be obtained by employing a general cost
analysis as was employed by Ettouney and Elsayed (1999). Some cost analysis tools are available for
specific SHCE problems, such as corrosion cost analysis. However, cost analysis should include not
only monetary costs, but also possible social and economical costs. When cost analysis of an SHCE
is performed in such generalized manner, it becomes more specific to the structure under study.
This shows the importance of integration and interdependability of all aspects of SHCE.
2.2.3 Example
As an example of the baseball analogy in an SHM project that can be found in a paper by Helmicki
et al. (2002). The strategy of this SHM was to monitor load rating (global) and several loacalized
issues (ultrasonic pin testing), tesnile coupon testing, and Charpy V-Notch testing. This led the
authors to an estimation of remaining fatigue life. A combination of localized and global SHM
approach is one of the features of this case study. One result was to modify cold weather mainte-
nance procedures for the bridge.
The study involved the Ironton–Russell Bridge, built in 1922, a 2401’ long steel thru truss with
steel plate girders and concrete deck approach slabs. It has gone through various rehabs during its
life and was rated low in 1997 during the biennial inspection due to superstructure deficiencies.
Hence it was decided by the owner to conduct instrumentation-based field test to evaluate the struc-
ture further. The objectives are to evaluate required reduced load posting, near-term retrofits, and
long-term management and planning.
A finite elements model was developed. On the basis of the model, visual inspection data, and
discussions with owners, the sensor locations were chosen sufficient to capture the critical bridge
elements’ response. Strain gages with known truck loading was used for field tests. The field tests
and subsequent analysis was used to select inventory and operating capacity of the structure.
The bridge was monitored for about 20 months for its in-service long-term operating behavior (to
study the effect of its operating environment). The data revealed fixity of the pin and hanger con-
nections, which were supported by physical observations such as pack rust.
s=2500
1
s=3000
0.8
s=3500
Reliability
0.6 s=4000
0.4 s=4800
0.2
0
0 10 20 30 40 50 60
Time (years)
FIGURE 2.7 Reliability versus Time for different strengths (S) of a reinforced concrete beam. Courtesy of
Dr. E. A. Elsayed.
On the basis of above observations, ultrasonic tests were scheduled to obtain more quanti-
tative information on pin and hanger connections. On the basis of this testing and determina-
tion of section loss, retrofit options were considered at select locations to increase structural
redundancy.
Selected material testing was conducted to make sure appropriate properties (Elastic Modulus
and yield strength) were used in the analysis. Charpy V-notch testing was done to evaluate tempera-
ture effect on ductility of steel. On the basis of the testing, it was decided that the material became
brittle at temperatures below 0°F. The bridge used to be closed below 20°F. On the basis of these
tests, it was decided to close the bridge, when temperatures fall below –5°F.
Due to the presence of low category weld details, appropriate members were instrumented to
measure continuous traffic data to observe actual stress ranges these members are subjected to. On
the basis of this data, it was decided that these members had infinite fatigue life due to very low
stress ranges observed.
2.2.4 Bringing All Together

In the above sections, SHCE is compared to a baseball game. Each component of SHCE is similar
to a base in a baseball game. A runner in a baseball game starts at the home plate with a goal to
return back there through all three bases in between to score a run. In SHCE efforts, the investigator
starts with a decision to make and then go through all four components to make the final decision.
If he/she obtains the required information after going through the entire SHCE cycle, then the goal
is fulfilled and a run is scored.
If the player is stranded at intermediate bases and thus cannot return to home base, then the
efforts are of limited or no use. In similar manner, if the SHCE study does not conclude with the
information that cannot aid in making the decision, for which the study is intended for, all the
efforts are of no use or limited value.
The efforts required to achieve or complete each of the components of SHCE can vary signifi-
cantly depending on study at hand. In some cases, amount of time required for sensor selection,
system identification, and damage identification can be relatively less due to a simple or well studied
structure, previous experience, and so on. In such cases, it can be compared to a home run in a
baseball game, where all bases are run through with a single shot. Such an example could be find-
ing an impact factor associated with an easily accessible simple supported single short-span bridge
with couple of girders in a rural environment. This requires installing one or two gages on the
structure and measuring strain under traffic loads. The effort required for all SHCE components is
very limited in such cases and one can obtain the impact factor with relative ease. As structure and
service environment becomes complex, the efforts required for each SHCE component will vary
significantly.
If multiple decisions have to be made using the same SHCE study, information required to make
each decision can be compared to a run. Number of decisions that have to be made can be thought
as number of runs required to win the game and thus the study can be concluded if all decisions the
study is intended to assist are successfully made. If the study produces more information to enhance
the state of the practice or state of the art besides providing the minimal information needed from
it, then the study value is significant. Such studies and the teams involved in planning such studies
can be considered better teams.
Several comparisons such as above, where SHCE is compared to base ball game, can be made.
But, the bottom line is that every SHCE study has to be initiated based on the probability of success
and when benefits outweigh costs (i.e., produces value to the decision making process). If prob-
ability of success or the probable value accomplished is not significant then study should not be
initiated, that is, it is better not to play the game at all. This is one place, probably; the SHCE differs
with the baseball game!
2.3 SHCE AND DESIGN PARADIGMS

2.3.1 General
The dominant civil/structural engineering paradigm has been traditionally the CD paradigm.
Another engineering paradigm has emerged lately, which is the performance-based design (PBD).
It is obvious that SHCE/SHM should be directly related to the paradigms of engineering practices.
See, for example Figure 2.8. The figure shows the relationships between time-dependent hazards
and demands from a given structure.
2.3.1.1 Statistical Pattern Recognition

An integrated SHM paradigm that was called statistical pattern recognition was introduced by
Farrar et al. (2000). The authors considered vibration-based damage detection as the main basis
for SHM. They also recognized that conventional vibration-based damage detection has two main
drawbacks (a) damage is typically a local phenomenon that needs accurate recognition of higher
vibration modes to be detected properly, and (b) the need for vibration-based damage detection to be
performed in an unsupervised mode. The statistical pattern recognition paradigm includes several
topics that aim at utilizing the advantages of vibration-based SHM, while addressing the above two
Demands and hazards (Time scale)
Constant
Sudden Gradual (or very slow
varying)
Earthquakes Normal wear and tear
Gravity loads
(own weight)
Flood Scour
Regular design
loads (traffic,
wind, etc.)
increased traffic
Environmental
Accidental impact conditions for
corrosion
(alkaline,
moisture, salt, etc.)
Environmental
conditions for
FRP (temperature,
moisture, UV, etc.)
FIGURE 2.8 Demands and hazards (time scale).

drawbacks. Some of the topics of the paradigm are as follows:
Operational Evaluation: Any SHM must accommodate different operational needs of the
system on hand. These operational needs include type and significance of damage, opera-
tional and environmental conditions, and limitations of data collection, if any.
Data Acquisition and Cleansing: The method(s) of acquiring and preparing data for use
must be compatible with both operational conditions, above, and selected features, below.
Feature Selection: The authors of the statistical pattern recognition describe in great detail the
different categories of vibration-based damage recognitions. The categories are (a) basic
modal properties, (b) dynamically measured flexibility, (c) updated structural model param-
eters, (d) time-history and spectral pattern methods, (e) time domain methods, (f) frequency
domain methods, (g) time-frequency methods, and (h) nonlinear methods. It should be
noted that the above categories were discussed at even greater details by Farrar et al. (2000).
Statistical Model Development: The final topic is statistical development. The authors
stated that there are five subjects that need to be defined in the damage state in a system:
(a) existence of damage, (b) location of damage, (c) type of damage, (d) severity, or extent,
of damage, and (e) remaining useful life.
Three statistical modeling approaches were recommended by the authors of the paradigm. They
are as follows:
1. Group classification in a supervised learning mode. Group classification would try to clas-
sify results of observations into groups that can be utilized by decision makers; an example
of grouping is damaged/undamaged grouping.
2. Regression analysis in a supervised learning mode. As the name implies, the authors of this
model recognized that utilization of regression analysis tools can help in damage detection.
3. Density information in an unsupervised learning mode. This technique tries to handle situ-
ations when damage information is not complete. By utilizing probabilistic methods the
authors contend that it is possible to identify damage information.
The proposed statistical pattern recognition paradigm is a powerful paradigm that can be of help in
numerous situations. As such, it is an essential approach in the field of SHM.
2.3.2 Capacity/Demand Design

2.3.2.1 General
Monitoring large civil structures such as bridges, dams, and buildings has received significant atten-
tion in the last decade. Several owners and researchers are actively pursuing development of long-
term monitoring systems, with reliable sensor technologies and remote monitoring systems for system
and/or damage identification. This work emphasizes that there are several other aspects to structural
health in addition to structural monitoring. Considerations of all aspects are necessary before any
decision or strategy is made to ensure safe, economic, and reliable structural performance.
The safety and adequate performance of any structural or nonstructural component can be
assured by the well-known CD equation, as follows:
C
≥a
D (2.1)
Where C and D are measures of the capacity of the system and the demand from the system, respec-
tively. The factor α is usually taken as unity. It is referred to as the CD equation.
The process of analysis and design of structures involve the evaluation of the CD equation. In
addition, there is an interconnection between the subject of health monitoring and the CD equation.
Understanding this interconnection is important for the decision making process in the SHM subject.
2.3.2.2 Capacity/Demand Ratio as the True Measure of Structural Health

2.3.2.2.1 General
We offer a general discussion of some characteristics of the capacity of a system and the demand
from the system as they relate to SHCE/SHM in this section.
2.3.2.2.2 Evaluating Structural Capacity

Capacity of a structure, or a smaller component in the structure, can take several forms. These
forms of capacity include structural strength, structural strains, or structural displacements. One of
the most important tasks of any structural design code is describing the process of evaluating the
structural capacity.
We argue here that structural capacity is not a constant entity. It is a time-dependent entity. From
the instant the structure is built, the capacity of this structure, and its components, changes. The
time-dependent capacity can be either gradual, or abrupt. Among the reasons for gradual changes
in structural capacities are the following:
1. Natural deterioration of material properties. Rusting and corrosion are the best known
reasons for this phenomenon.
2. Suddenly applied additional structural demands that might push the structural capac-
ity beyond the linear limit to a nonlinear limit. Thus, after the removal of the additional
demands, the structural capacity will not revert back to its original state, but to a lower
state of capacity. Earthquake events are an example to such a situation.
3. Changes in structural geometry. Spalling of concrete can lead to gradual reduction of
capacity. On the other hand, if a retrofit, or addition, is added to the structure, the resulting
capacity might be increased.
Figure 2.9 shows schematics of the different possible modes of time-dependent capacity of a
structure.
Health monitoring of any structure or a structural component is only a single step in the multi-
steps evaluation of the time-dependent structural capacity as shown in Figure 2.10. The figure also
Capacity
Time
Natural deterioration, e.g., corrosion
Gradual changes in structural geometry
Sudden loss of capacity, e.g., earthquake damage
Retrofits
FIGURE 2.9 Time-dependent capacity of a system.

Qualitative/
Visual
inspection
Step # 1: Structural health Testing/

monitoring sensing/ Past experience
estimate state of system measurements
Analytical/
structural
identification/
damage
identification
Qualitative
Engineering
Step # 2: Estimate capacity of judgment
Hybrid
system
Quantitative/ Design codes

conventional
structural
analysis or
design
FIGURE 2.10 Steps for evaluating structural capacity.
shows the different steps and tasks that are needed to evaluate the time-dependent structural
(or component) capacity. Figure 2.10 shows that to evaluate the state of the structure, at any instant
of time, the health monitoring techniques might be used. Alternatively, purely analytical techniques
might also be used. Each of these two alternatives has its own strong and weak points. Unfortunately,
knowing the state of the system is only a necessary but not sufficient part of the capacity-evaluation
process. An evaluation of the capacity of the system, utilizing the knowledge of the structural state,
is still needed. Figure 2.10 shows that this evaluation can be either qualitative, quantitative, or a
qualitative-quantitative hybrid process. In what follows, some examples of capacity-evaluation pro-
cesses and their relationship with health monitoring techniques are presented.
2.3.2.2.3 Evaluating Demands on Structures

Evaluating the time-dependent demand on a structure, or a structural component, is as important
as evaluating the capacity. The demand on any system can change with time. This change can be
either gradual or sudden. It can either increase or decrease, depending of the functional changes of
the structure. Some examples of changes in demands are as follows:
1. Changes in traffic loads on a highway bridge. This can be either increasing or decreasing
demands. They usually occur gradually and imperceptibly.
2. Changes in design code specified demands. These demand changes tend to be abrupt. They
can either increase or decrease the demand. For example, seismic demands on structures,
as required by seismic codes, have fluctuated over the past several years. See UBC-97
(1997) and IBC-2006 (2006), for example.
Figure 2.11 shows schematically the different modes of time-dependent demand on structural sys-
tems. Note that demands on structures are decreed by the pertinent design codes for the structure
of interest. Unfortunately, most of the current design codes are written for new construction. This
Demand
Time
Slowly decreasing, e.g., demography changes

Slowly increasing, e.g., demography changes
Sudden decrease, e.g., alternate route available
Sudden increase, e.g., loss of alternate route
FIGURE 2.11 Time-dependent demand from a structure.

Non-Exceedance Probabilitiy Levels
pnew
pexisting
Demarcation time at which

a new structure starts to be
considered as an existing
structure. There is no clear
definition of such a
demarcation line
Time
Seismic Demands on New Structures

Seismic Demands on Existing Structures
FIGURE 2.12 Changes of seismic demands for existing structures (as prescribed by seismic design codes).
means that when an engineer is evaluating the demands an existing building, he will have to use the
same demands on a new building. This could lead to overly conservative and unrealistic demands.
Some earthquake design codes have tried to mitigate this conservatism. Figure 2.12 shows sche-
matically the earthquake demand reduction as a function of time as mandated by an existing build-
ing seismic design code. Note that such a code has not specified the needed elapsed time for the
building designation to change from a new to an existing building. Is it one day, one week, one year,
or more? The problem that this hypothetical question signifies will be discussed in more details in
later sections.
2.3.2.2.4 Setting the Structural CD Ratio

The previous section discussed in some detail the concepts of capacity and demand in relationship
to the current application of health monitoring of structures. It was shown that health monitoring, as
it is practiced now, occupies a small part in the time-dependent structural evaluation process. This
process is actually dependent on the CD ratio, as shown in Equation 2.1, not the individual values
of the capacity and demand. In this section, we investigate the evaluation process as it relates to
SHM.
2.3.2.2.4.1 “Optimum” CD Ratio

For new structures, it is desired to have α = 1.0 for all components of the structure, and for the
structure as a whole. Unfortunately, this is not always possible. For practical reasons, the ratio α is
generally designed to be as close to unity as possible but not less than unity such that α ≥ 1.0. The
closer α to unity the more efficient the structure is. In some cases, the design engineer may design a
component such that α < 1.0. This situation happens when his engineering judgment makes him/her
conclude that a CD ratio that is slightly less than unity is still acceptable. There is no quantification
of engineering judgment; it is a qualitative measure that is dependent on the situation on hand, and
the personal experience of the design engineer.
Following the above discussion, it becomes clear that, for new buildings, there is a range of
design for α. This can be stated as αmin ≤ α ≤ αmax. This inequality is such that αmin ≤ 1.0 and 1.0 ≤ αmax.
Both αmin and αmax are desired to be as close to 1.0 as possible. The optimum structural design is
when α = αmin = αmax = 1.0 for the structure and all its components.
Figure 2.13 shows the time-dependent CD ratio after the structure goes into service. Figure 2.13
shows a case when CD changes abruptly. This can happen when either (or both) the capacity or the
demand changes abruptly. It also shows the case when CD changes slowly. This can happen when
both the capacity and the demand changes slowly.
Figure 2.13 also shows schematically the range αmin and αmax of the original design. For the pur-
poses of this study, we will investigate only the situations when α decreases with time.
Studying Figure 2.13 further it appears that there are two main phases that can be identified. The
first phase of the life span of the structure occurs while αmin ≤ CD. This is where the structure, and
its components, is still safe within the bounds of the engineering judgment of the design engineer.
αmax
CD Ratio
Initial α
αmin
Time
Gradually Changing CD Ratio, e.g., Corrosion

Suddenly Changing CD Ratio, e.g., Floods
FIGURE 2.13 Time-dependency of actual versus acceptable CD levels.

The second phase of the life span of the structure occurs while CD ≤ αmin. This is where the CD ratio
for the structure, or its components, starts dipping below the bounds of the engineering judgment
of the design engineer.
The above discussion leads to an intriguing observation; if the structure was designed such that
no engineering judgment were exercised, that is, αmin = 1.0, the structure would be rendered unsafe
almost from the instant it is built. This shows the importance of engineering judgment, either during
the design phase, or during the maintenance phase of the structure.
2.3.2.2.4.2 Target CD Ratio

As was described earlier, some design codes have recognized this dilemma, and tried to pre-
scribe a demand (hence CD ratio) for existing structures. We generalize this approach by
identifying a time-dependent target CD ratio a . A safe and properly performing structure is
operating such thata ≥ a , as shown in Figure 2.14. Note that a = a min for new structures. As
time progress a varies in a realistic manner. This variation should reflect engineering judg-
ment, probabilistic considerations, structural service time already spent, and expected/desired
life of the structure.
2.3.2.2.5 Decision Making Process

Let us assume that the time-dependent CD ratio α is known for a given structure, or a structural
component. This knowledge can be based partly on a health monitoring system that identifies the
state of the structure. This knowledge is also dependent on translating the state of the structure
into structural capacity. Also, the time-dependent demands on the structure on hand is needed for
the full knowledge of α. Let us assume that the target CD ratio α1 is known. Again, this knowledge
can be based on design codes, engineering judgment, or any other mean of evaluation. Figure 2.14
shows the ensuing decision making process. The decision making process would lead finally to
one of the three possible decisions (a) do nothing, (b) retrofit the system, or (c) replace the system.
Deciding to do nothing is a simple decision. However, the decision of retrofitting, and the levels of
retrofit, or replacement can be complex process. Ultimately, cost, social and environmental pro-
cesses would control such a decision.
CD Ratio
Initial CD
Time
Time-Varying CD Ratio
Time-Varying Target CD Ratio
FIGURE 2.14 Time-dependency of actual versus acceptable CD levels.

For a given structural state...
Yes is α ≥ α No
1
System is safe. No
further action is Action is needed
needed
• Economic studies
Stop • Social impact
• Environmental
impact
• Other factors...
Retrofit (level of
Replace
retrofit)
FIGURE 2.15 Decision making process when both actual and target structural states are known.
2.3.3 Performance-Based Design

2.3.3.1 Overview
Performance-based design of buildings has been under development in the earthquake engineer-
ing community for more than a decade. It has been in use in the blast community for much longer.
Communities of other hazards (progressive collapse) and infrastructure (bridges) are considering
the usage of PBD. This article explores different aspects of PBD as it applies to seismic designs.
The concepts are generally applicable to almost any other hazard, or any other infrastructure.
We first compare prescriptive and PBD. The steps of PBD are presented next. Finally, we explore
the interrelations between PBD and other emerging paradigms in the design/infrastructure
community.
2.3.3.2 Prescriptive versus Performance Design Paradigms

There are several types of differences between common prescriptive and emerging PBD paradigms.
The differences can be classified as (a) design objectives, (b) computational underpinnings, and
(c) practical steps of the methods. Figure 2.16 illustrates the differences in design objectives between
these two approaches. Basically, prescriptive design methods call for ensuring that the prescribed
capacity C is larger than prescribed demand D. Design iterations continue until the capacity to
demand ratios are within an acceptable range. In PBD, the objective of the design is to achieve a
level of performance, as correlated to appropriate consequences (measured monetarily, for exam-
ple). The design iteration involves reaching a design that would achieve the desired performance.
The computational underpinnings of prescriptive design are related to the relationship between
capacity and demand, and are based on structural reliability methods. PBD is based on risk meth-
ods that consider hazards (demands), vulnerabilities (capacity), and consequences. The differences
between prescriptive and PBD steps are shown in Table 2.1.
Clearly, PBD involves more efforts by different building stakeholders, and in return it offer many
advantages, such as the potential cost savings in the long run, options of continued operations and
Iterate/improve design
Capacity
Demand
(C) or Is D/C ratio No
(D) or
vulnera- acceptable
hazard
bility
Yes
Prescriptive
or Good
D/C Design design
Iterate/improve design
Vulnerabil Hazard
ity (V) or (H) or Are consequences
capacity demand deterministically and/ No
or probabilistically
acceptable
Conquences
Yes
Performance-based design
Good
design
FIGURE 2.16 Prescriptive versus performance-based paradigms.
TABLE 2.1
Differences in Steps between Prescriptive Design and PBD
Design Considerations Prescriptive Performance
Seismic hazard level Determined by codes Chosen during the process
Structure damage level (acceptance criteria) Specific Chosen during the process
Consequences NA Computed as part of the process (qualitatively
for first generation PBD, quantitatively for
second generation PBD)
Uncertainties (performance probabilities) NA Computed as part of the process (second
generation PBD)
immediate occupancy after events that can be of importance for sensitive facilities, and a clear
quantitative picture on how the facility will perform during a designed event, and what the con-
sequences of such performance would be with no surprises to the stakeholders. As a rule, PBD is
based on concepts of risk, where CD design is based on concepts of reliability. The main difference
is that PBD accounts for consequences (or outcomes) whereas CD design does not. More in-depth
analysis of risk and reliability are provided in Chapter 8.
2.3.3.3 Modern versus Traditional PBD

Seismic PBD was introduced in FEMA 273/274, which were then reissued, with minor modifica-
tions as FEMA 356. It is generally accepted that those efforts constitute the first generation of seis-
mic PBD. An ASCE standard (ASCE 2007) has been issued on the basis of that first generation PBD
effort. There has been an effort by FEMA to generate a second generation of seismic PBD since
2002. That effort is continuing as of the writing of this article. Several differences exist between
the first and second generation of the PBD effort. Among them are the continuous quantitative per-
formance measures compared to the discrete qualitative measures offered by the first generation.
Uncertainties are also addressed quantitatively in the second generation seismic effort. In addition,
the second generation utilizes component and system fragilities and presents more details of ana-
lytical and design techniques.
2.3.3.4 Elements of PBD

2.3.3.4.1 Overview
Performance-based design roughly comprises three steps that are usually performed consecutively:
(a) estimation of hazard, (b) evaluation of vulnerability, and (c) computation of consequences. Figure
2.17 shows schematically these steps, which will be discussed next. Seismic hazard and building
structures are used as an example in this section due to the use of PBD in this field for several years,
as noted in the previous section.
2.3.3.4.2 Hazard Level

The choice of hazard level includes choosing seismic input characteristics for the project. Seismic
input characteristics can be simple, such as the choice of hazard level and the shape of the design
spectra. It can be more involved, such as generating an ensemble of seismic acceleration time his-
tories. It also can be deterministic-based, or probabilistic-based. In most situations, the designer
needs to address issues such as return period (the length of time one earthquake of a given level
is expected to occur), maximum earthquake ground acceleration, and so on. In the second gen-
eration seismic PBD effort, the probability of the chosen seismic hazard is an integral part of the
design input needs. This is needed to compute the expected consequences of the design, as shown
in Figure 2.18. Another feature of second generation seismic PBD is that it can be either based on
a single scenario, such as a unique earthquake level, or be based on multiple earthquake levels,
with varied return periods. This latter approach is obviously more time consuming, since designs
must be performed for each of the scenarios. The advantage of the multiple scenario approach
is that it gives a more complete picture of the total life cycle of the building. Table 2.2 shows
the seismic hazard evaluation differences between prescriptive and performance-based seismic
designs.
2.3.3.4.3 Damage Level (Vulnerability)

After the seismic input is defined, the building design process starts. In general, the process is fairly
similar to the conventional prescriptive design process. There are, however, some important differ-
ences, as shown in Table 2.3.
Computing types, levels, and probabilities of earthquake structural or nonstructural damages
in buildings, or any other type of infrastructure, are not an easy task. An advanced analysis tech-
niques, as well as an experienced analyst are needed. This is one area that is currently undergoing
extensive research and development. An emerging technique for relating earthquake damage to
uncertain inputs, and computing the needed damage uncertainties is the fragility technique. Figure
2.18 shows how fragilities are used in a PBD context. We note that component seismic fragilities
have been under development for some time. Efficient, practical, and general methods for system
level fragility, on the other hand, are just starting to be considered by the community.
60
Sensing:
Structural identification:
Hazard Severe hazard Vulnerability
-Accelerometers Large hazard -Structural properties
(earthquakes)
Medium hazard (strength/stiffness)
n
-Pressures (wind)
-Modeling assumptions
-Velocities (flood waters)
Low hazard (neural networks)
-etc.
-Parameter identification
Acceleration
Choose hazard level -etc.
Force
Period Deformation
No damage
Severe
Damage identification tools: Limited damage damage
-Strains Collapse
-Ultrasonic
-Crack detection
-Tilt
-Soil erosion
Decision making:
-etc.
-Cost estimation tools
-Prioritization tools
Roles of SHCE at -etc.
different steps of PBD
Cost of damage = C(DM)

Damage level (DM)
Steps of
PBD
Cost of damage
FIGURE 2.17 Steps of PBD.

Infrastructure Health in Civil Engineering
Severe Computed
No damage Limited damage damage
vulnerabilities
Collapse
Probability of damage
100%
Fragilities
level occurrence
Set of predetermined curves
that relates seismic hazard
levels with probabilities of
0%
different damage states
Seismic hazard level
Probability P(DM)
Histogram
Given seismic hazard level,
compute probability of each
damage state: From
histograms
Damage level (DM)
Compute consequences
Consequence = Cost * probability
= C(DM) * P(DM)
FIGURE 2.18 Computations of risk.
TABLE 2.2
Seismic Hazard Evaluation
Issue Prescriptive Performance Based
Earthquake amplitude and wave form Prescribed Varies, depending on desired
(time domain), or design spectra performance
Probabilities of occurrence Usually not an issue Needed to compute consequence
TABLE 2.3
Design (Vulnerability) Evaluation
Issue Prescriptive Performance Based
Acceptance criteria Prescribed, mainly to Varies, depending on desired level of earthquake damage to
ensure life safety the building
Analysis techniques Mostly linear Can be linear. However, nonlinear analysis is preferred
Analysis objectives Limited. Only check to Wide range. Check for acceptance limits. Also computes
see if specific acceptance damage types and levels, which are needed for consequence
limits are met analysis
Probabilities of damage/ Usually not an issue Needed to compute consequence
response occurrence
TABLE 2.4
Failure Rates of Buildings
Building Type Probability of Failure during Average Life Span
Apartment 0.3 in 10,000,000
Mixed housing 5 in 10,000
Controlled domestic housing 1 in 100,000
Mixed housing 1 in 1,000
Engineered structures 1 in 10,000
Source: Based on Melchers, R. E.: Structural Reliability Analysis and Prediction. 2002. Copyright
Wiley-VCH Verlag GmbH & Co. KGaA. Reproduced with permission.
TABLE 2.5
Failure Rates of Bridges
Bridge Type Probability of Failure during Average Life Span
Steel railway I in 1000
Large Suspension 3 in 1000
Cantilever and suspended span 1.5 in 1000
General 1 in 100
Source: Based on Melchers, R. E.: Structural Reliability Analysis and Prediction. 2002. Copyright
Wiley-VCH Verlag GmbH & Co. KGaA. Reproduced with permission.
Melchers (2002) gave probability of failure for buildings that vary from one in 10,000 for engineered
structures in Canada to 0.3 per million in apartment buildings in Denmark during their average life
span. Melchers (2002) also gave typical failure rates for bridges between 1900 and 1960 that vary from
1 in 100 for general bridges based on an Australian source to 3 in 1000 for large suspension bridges
in the world. It is interesting to note that the probability of failure is generally an order higher for the
bridges when compared to buildings even though the average life spans are very similar.
Table 2.6 gives the hazard related deaths in United States during a period of 25 years from 1970
to 2004. Similar information including risk of death in various activities can be found in the litera-
ture (e.g., Melchers 2002). Melchers (2002) reported that the risk of death per year due to structural
failures is about 0.1 x 10 –6, that is 10,000 times more than that of cigarette smoking and 1200 times
more than that of boating. Table 2.7 gives generally accepted risk tolerance from infrastructure,
based on values reported in the literature (see, e.g., Melchers 2002). This general criterion assumes
that the risk due to a hazard should not exceed the risk faced by an individual in everyday life, that
is, approximately one in 1,000 chance of annual risk of death.
2.3.3.4.4 Consequences
Computing consequences of seismic event on design of buildings is perhaps the most important dif-
ference between prescriptive design and PBD. In the context of PBD, consequences generally relate
to the owner, or user. The consequences to the neighborhood or other regional effects are beyond
the scope of current PBD efforts. In the current efforts of FEMA, two types of consequences are
considered: monetary and casualty. Figure 2.18 illustrates how consequences are estimated as the
final step of the PBD effort. The probabilities of different types of damages, as estimated by fragil-
ity curves, for example, are combined with predetermined damage level damage costs relationships.
The estimated cost of the earthquake event can then be computed as shown in Figure 2.18. This
TABLE 2.6
Hazard Related Deaths in the United States
during a 25-year period (1970–2004)
Natural Hazard Deaths (%)
Volcano 0.3
Wildfire 0.4
Avalanche and landslides 0.8
Fog 0.8
Earthquakes 1.2
Hurricanes/tropical storms 1.5
Coastal 2.3
Wind 8
Severe storms/thunder storms 10
Lightning 11.3
Tornado 11.6
Flooding 14
Winter weather 18.1
Heat 19.6
Source: Borden, K. A., Natural hazards mortality in the

United States, Ph. D. thesis, University of South
Carolina, SC., 2008.
TABLE 2.7
Risk Tolerance
Category Annual Risk of Death
Broad acceptance (no efforts required in risk mitigation) Less than one in a million
Tolerable (not negligible and mitigation suggested) Between one in 10,000 to one in a million
Unacceptable (mitigation required) Greater than one in 10,000
method of computing cost based on uncertainties is one of the many definitions of risk (Chapter 8).
Thus, we can see that PBD is a risk-based paradigm, as stated earlier.
After the cost (risk) of the seismic event, given the chosen performance levels, is determined,
the stakeholders (owner, architect, engineer, users, insurance companies, etc.) must decide if it is an
acceptable cost (risk). If it is acceptable, the design should go forward as is. If the costs proved to
be too high the performance levels are then changed, and the whole procedure is repeated until an
acceptable cost (risk) is reached (Figure 2.16).
2.3.3.4.5 Uncertainties
One of the main advances that second generation seismic PBD paradigm offer is that it acknowl-
edges the uncertainty present during the process of seismic, or any other hazard, design of buildings,
or any other infrastructure. The uncertainties in defining the seismic hazard, the design process
(vulnerabilities), and estimating consequences are all included within the PBD paradigm. This is in
sharp contrast with prescriptive designs. Admittedly, uncertainties are also accommodated to a cer-
tain extent in prescriptive designs: allowable stress design (ASD) utilizes factors of safety and load
and resistance factor design (LRFD) accounts for load factors and strength reduction factors, as the
name implies. Yet PBD allows for far more freedom in prescribing desired degrees of exceedance
levels and probabilistic levels for the building and events on hand. For example, a particular build-
ing stakeholder might decide that a nonexceedance probability of 95% is needed for the perfor-
mance of the building in a case of a seismic event. Another stakeholder, for another building, might
decide that an 85% nonexceedance probability is more adequate for the building. Such a freedom of
uncertainty level choice is one of the major advantages of PBD.
2.3.3.5 Future Trends of PBD

Performance-based design for seismic and other hazards is gaining interest from the professional
community, for the reasons stated earlier. There are several other emerging paradigms that will
make PBD of more importance to the infrastructure design community. Some of these paradigms
are briefly discussed below.
2.3.3.5.1 Multihazards Considerations

The rising costs of different hazards, seismic, wind, floods, bomb blasts, progressive collapse, and
so on have prompted interest in optimal designs that consider all of those hazards so as to increase
safety and reduce costs, Ettouney et al. (2005). For a given building design, two hazards, seismic
and blast, for example, can both exhibit higher performance levels as the magnitudes of both haz-
ards increase (Figure 2.19). In other building configuration, the performance levels as the building
responds to each of the hazards might decrease (Figure 2.19). The former situation can be costly,
while the latter situation can be unsafe. Careful multihazards considerations are needed as a com-
panion to PBD approach.
2.3.3.5.2 Life Cycle Analysis

Life cycle analysis is a byproduct of PBD. The knowledge of life cycle behavior is of immense
importance to asset managers in their decision making efforts: inspection, prioritizing, budgeting,
maintenance, and so on. Current prescriptive design paradigm does not offer such knowledge.
Performance level
Interaction
performance Beneficial IPS
surface (IPS)
Sei
s
haz mic
t lev ard
as el
Bl zard
a
h el
lev
Performance level
Interaction
performance Nonbeneficial IPS
surface (IPS)
Sei
s
haz mic
t lev ard
as el
Bl zard
a
h el
lev
FIGURE 2.19 PBD and multihazards considerations.

Performance-based design offer numerous advantages as discussed above. It also presents sev-
eral challenges: need for learning, smooth multidisciplinary integration, and added expertise of
professionals. The advantages of PBD make meeting these challenges a worthwhile goal.
2.3.3.6 Role of SHCE and PBD

2.3.3.6.1 Overview
Structural health monitoring is also emerging as an essential paradigm for preserving the health of
infrastructure. There is a natural symbiosis between PBD and SHM. On one hand, SHM produces
the needed plethora of information by PBD. Conversely, PBD techniques provide to SHM profes-
sionals some valuable information about expected damages in the structure that can aid them in
placing sensors and identification of structures and damages. Figure 2.20 shows how the three basic
components of PBD paradigm interact with SHM/SHCE methods and components.
2.3.3.6.2 Fragilities
We discussed earlier the use of fragility concepts in PBD. Since fragilities offer simple way of
describing probabilities of different limit states, their use in infrastructure design and assessment
is gaining popularity. For example Reinhorn et al. (2001) used fragilities in seismic assessment of
structures. FEMA-445 (2006) used fragilities extensively in developing PBD in seismic mitiga-
tion of nonstructural components in buildings. A fragility relationship as used in Figure 2.21 is
defined as
P ( D ≥ Di ) = f ( H ) (2.2)

SHM ROLE IN PERFORMANCE-BASED

ENGINEERING APPROACH
Wind
velocity Corrosion
Traffic
rate
ds
SH
tho
M
me
pro
nd
ced
sa
ure
Hazard
ure
sa
ced
nd
pro
me
M
tho
SH
ds
Three components of
performance-based
FRP engineering Damage
wripping
Mitigation Acceptance ID
limits
Stiffness Strains
Structural
ID Tilt
SHM procedures and methods
FIGURE 2.20 SHM methods as related to performance-based engineering.

1.0
P(D >= Di)
0.0
Hazard level (H)
Damage state: Pristine

Damage state: Moderate
Damage state: Heavy
Damage state: Failure
FIGURE 2.21 Typical fragility curves.
where P(D ≥ Di) is the probability that the damage D of the system, or component, of interest is at
damage state that is equal or more severe than Di due to a hazard level of H. The function 0.0 ≤ f(H)
≤ 1.0 is the fragility function. Numerous authors have studied and produced fragility relationships.
See, for example, Mostafa and Grigoriu (1999) and Dumova-Jovanoska (2000).
When fragility relationships are known, it can help officials in assessing competence of the
system. Also, it is a invaluable step in PBD, as shown in Figure 2.18. The main problem is: how
can fragilities be developed accurately? Fragility is usually developed by analytical techniques.
Common steps are as follows:
1. For a given component, estimate statistical parameters of uncertain parameters. For exam-
ple: if a bridge girder is under consideration, properties such as the moment of inertia,
modulus of elasticity, and so on can be assumed to be uncertain. For each of these uncer-
tain properties, the mean and variance should be estimated.
2. Choose a damage state Di. The damage can be any type of degradation in the system of
interest. For example damage can be deformation level, cracking of any type, settlement
of foundation, or state of paint or corrosion. The damage states can vary from pristine
to complete failure. Note that for a total of damage states of N, i = 1, 2, …, N. Since the
system parameters of # 1 are random, it is obvious that the damage measures, for example
deformation levels, are random.
3. For the hazard of interest, choose an appropriate hazard level Hj with j = 1.
4. Using appropriate probabilistic method, estimate the statistics or the histogram of the
damage level D when the system is subjected to the hazard level Hj. Monte Carlo simula-
tion method is used for this purpose for complex systems. For simple systems, analyti-
cal or semianalytical techniques can be used. See Chapter 8 for more details. From the
histogram, the probabilities that the system is in a given damage state Di due to Hj can be
computed.
5. Change hazard level to Hj+1, and repeat #4 given above.
6. Repeat steps 3 through 5 until all reasonable levels of hazard are considered.
7. The N fragility relationships P(D ≥ Di) = f(H), with i = 1, 2, …, N are formed. A set of N
fragility curves can be plotted as shown in Figure 2.21.
The above-mentioned analytical method in computing fragilities is fairly straightforward. However,

it relies in its accuracy on several assumptions, such as the knowledge of the statistical parameters
of the system properties. Another implicit, yet very important, assumption is that we assumed that
we know the exact relationship between the hazard, and the resulting damage during the analysis
step # 4. Such an assumption does not produce accurate results in many situations.
Structural health monitoring techniques can be used to reduce the effects of some of these
knowledge gaps. For example, on-site sensing can produce enough data points to form a reason-
able data set that can produce accurate statistical estimates of the desired system properties. The
hazard-damage relationship can be accurately computed using many of the SI/damage identification
methods of Chapters 6 and 7. Next, we present two useful uses of fragilities in bridge management
that utilize SHM techniques.
2.3.3.6.2.1 Fragilities and Deterioration Transition Matrices Let us consider next bridge dete-
rioration as a damage measure and pose the question: can we use SHM techniques to produce
bridge, or bridge component, deterioration fragilities? We offer the following method to produce
such curves. We start by using the well-known bridge deterioration data that is generated from
inspection reports; see Agrawal et al. (2008), or DeLisle et al. (2002), for example. We formalize
these curves as
D = D (T ) (2.3)
where D is an estimate of the deterioration and T is the time. A typical deterioration curve is shown
in Figure 2.22. A measure of deterioration can be bridge rating, which is one of the outcomes of a
manual bridge inspection. Obviously, at anytime, the estimation of D is uncertain. As such D can be
considered a random variable. DeLisle et al. (2002) studied this subject and developed a method to
Maximum
rating
Conventional
deterioration Mean + 2.0 σ
Damage (rating)
curves Mean + 1.0 σ

Mean
Mean - 1.0 σ
Mean - 2.0 σ
Minimum
rating
Time (T)
At a given time, the
probabilities of a given
state (rating) can be
transcribed into the
fragility curve format
1.0
Damage state: Pristine

P(D >= Di)
Damage state: Moderate

Damage state: Heavy
Damage state: Failure
Fragility curves
0.0
Hazard level = passing of time (T)
Constructing fragility curves from conventional deterioration curves
FIGURE 2.22 Relationship of deterioration and fragility curves.

estimate the uncertainties of D. Confidence intervals can be produced using that method as shown
in Figure 2.22. Close inspection of Figure 2.22 reveals that the damage, as measured by bridge
deterioration, is an outcome of the passing of time T. Thus we propose that the time T is the hazard
that produces deterioration. On the basis of this we observe that Figure 2.22 actually includes the
three basic ingredients of fragilities:
1. It includes an increased hazard (passing of time)

2. It includes damage states (deterioration levels as represented by bridge, or component,
rating)
3. It includes probabilities of occurrence as represented by confidence intervals
Armed with these observations, we can transpose conventional deterioration curves into fragility
curves using the conceptual method of Figure 2.22.
Bridge, or bridge component, deterioration fragility curves enable the decision maker with a
powerful tool to use PBD techniques to reach accurate decisions. This can be accomplished by
improving the estimates of the probability transition matrices [P] that are commonly used in
Markov processes in bridge management systems (see Chapter 8). The matrix [P] is a square matrix
with size of n that represents the number of possible damage states. When considering the bridge
deterioration as damaged states, it is customary to consider n as the number of bridge rating states,
see Agrawal et al. (2008b). As such, in NYSDOT practice, n = 7. The components of the matrix pij
indicates the probability that the damage state of the system, or the component, will change from
damage state Di to damage state Dj in a given time interval ∆T. The values of pij are computed using
historical records and measurements. They are usually assumed to be stationary, that is, time-inde-
pendent. Such an assumption is not an accurate assumption in many situations. We recognize these
probabilities as the prior estimates of probabilities; we redefine them as pij•PRIOR.
Fragility deterioration curves, as described above, can be used to improve on as well as provide
time dependency of the traditional estimates of pij. Close inspection of the fragility curves (note that
there is a curve for each damage, or rating state) show that the probabilities of the system, or compo-
nent, in a given damage state, at a particular time can be computed from the curves. We define these
sets of probabilities p(T,Di) as the probability that the system, or component, is at a given damage state
Di at time T.
We use Bayes theorem to improve prior probabilities as
pij i p (T , Di )
PRIOR
pij = (2.4)

POSTERIOR
∑
j=n
j =1
pij
PRIOR
(
i p T , Dj )
Note that the expressions of Equations 2.4 are time dependent; they are also improved estimates of
the prior transition probabilities.
2.3.3.6.2.2 Fragilities and Life Cycle Costs Fragility curves can be easily used in estimating
deterioration life cycle costs during a time span of ∆T using the expression
T = T0 + ΔT ⎡ i=n ⎤ (2.5)
LCC ( ΔT ) = ∫
T = T0 ⎢ ∑ p (T , Di ) i C ( Di )⎥ dT
⎣ i =1 ⎦
The initial time is T0. The cost of damage state Di is C(D). Equation 2.5 is useful in estimating costs
during shorter time spans. For longer time spans, a dynamic programming, or Monte Carlo tech-
nique should be used (see Chapter 9 of Ettouney and Alampalli 2012).
2.3.3.6.3 Hazards, Vulnerabilities, and Consequences

SHCE can play an even bigger role in enhancing PBD methods. Table 2.8 shows some examples
of how SHCE techniques can aid in accurate estimations of hazards, vulnerabilities, and conse-
quences for different situations. Judicial use of the techniques can improve safety at reasonable
costs.
TABLE 2.8
SHM Role in Estimating Hazards, Vulnerabilities, and Consequences
Class Application Hazard Vulnerability Consequence
Deterioration Paint Paint systems When paint deteriorates, One of the major consequences
deteriorate as a irregularities on the of loss of paint protection is the
direct response to surface would start formation of rust. Rust
environment. forming. At first, these mitigation can be costly.
Monitoring different irregularities can’t be Ultimately, severe loss of
environmental detected by visual structural resistance due to
factors, such as inspection. A rusting can lead to local, or
temperature, laser-based surface more extensive levels of failure.
freeze/thaw, inspection can help in Decision making techniques can
moisture, and so on detecting such small help in forming optimal safe and
can give an accurate size irregularities. Of cost effective decisions
estimate on how the course, as the size of regarding rust mitigation. See
paint would survive paint irregularities Chapters 3 and 4 in Ettouney
the environmental increases, it becomes and Alampalli (2012) for
onslaught possible to detect them examples of decision making
using visual inspection processes regarding rust and
corrosion
Corrosion See Chapters 3 and 4 in Ettouney and Alampalli (2012)
Fatigue See Chapter 5 in Ettouney and Alampalli (2012)
Settlement Displacement and tilt Strain monitors, located Consequences of ductile
sensors, (Chapter 5 at optimal locations, settlement (gradual and
of this volume), can are most suited to noticeable settlement) are
be used to monitor estimate vulnerability usually manageable, if detected
potential settlement to settlement. Note that early on
STRID methods can Consequences of large and
help in identifying sudden settlements can be large,
accurately potential including large-scale collapse.
damages from Scour is the leading cause for
settlement such sudden settlements. For
detailed discussions on SHM
and scour, see Chapter 1 in
Ettouney and Alampalli (2012)
Hazard Earthquakes See Chapter 2 in Ettouney and Alampalli (2012)
Scour See Chapter 1 in Ettouney and Alampalli (2012)
Live loads Monitor truck loads, Monitor deflections, Consequences of overloads vary
(Truck frequency and strains and dynamic from shorter fatigue life
Weights) dynamics. Also, if amplifications. Ensure (Chapter 5 in Ettouney and
possible, monitor that the locations of Alampalli [2012]) to brittle
the dynamic sensors are in optimal failure (I-35, MN collapse as
bridge-truck positions, see Chapter described in Chapter 3 of this
interaction 5 for more on optimal volume)
sensor locations
2.4 NDT VERSUS SHM

This section explores the interrelationship and differences between methods and techniques of non-
destructive testing (NDT) and SHCE/SHM. NDT offer wealth of methods and techniques that can
compliment and enhances traditional visual inspection methods (see Figure 2.23). These methods
and techniques offer opportunities for the SHM community. We argue that the use of NDT meth-
ods has been successful in SHCE/SHM when hazard-centric needs of SHCE/SHM and physical
demands of these hazards are recognized.
2.4.1 Overview
Nondestructive Testing methods have been in use and in continuous development for more than a
century. Their main use has been to provide better understanding of material and system properties.
There are numerous publications on this subject including several handbooks such as NDT Methods
Handbook of the American Society for NDT. SHCE and its subcomponent SHM are emerging fields
in the infrastructure arena and are intended to provide more information about structural condi-
tion (health) to make appropriate decisions. Even though there are a vast array of NDT methods
and techniques that are available to the SHCE/SHM professionals, the use of NDT methods in the
SHCE/SHM field has been relatively slow.
2.4.2 NDT Tools for Different SHM/SHCE Components

As noted earlier, SHM can be subdivided into three components: sensing/measurements, STRID,
and damage identification. The subject of SHCE contains SHM and decision making. Tables 2.9
through 2.12 show how sensing, SI, damage identification, and decision making relate to both SHCE
and NDT fields, respectively.
2.4.3 NDT, SHCE/SHM Tools and Bridge Hazards

The previous section showed some of the differences between NDT and SHCE/SHM concepts. In
this section, beneficial uses of NDT tools in the field of SHCE/SHM are explored. We investigate
this subject by recognizing that the field of SHCE/SHM aims mostly to counter different hazards
FIGURE 2.23 Visual inspection of bridges. (Courtesy Barton Newton.)

TABLE 2.9
General Comparison of NDT and SHCE/SHM for Sensing/Measurements
NDT SHCE/SHM
Localized in nature Generally covers larger area
Needs good idea as to the location of damage Location of damage is not known before
Mostly a localized event in time Time spans can be fairly long: almost as long as the lifespan of the target
system. Can be continuous monitoring or long-term monitoring
Sensing technology is very advanced Sensing technology is developing (wireless sensing, remote sensing, etc.)
Mostly manual Mostly automatic or semiautomatic
Mostly single type of sensor Mostly multiple types of sensors
TABLE 2.10
General Comparison of NDT and SHCE/SHM for STRID
NDT SHCE/SHM
STRID methods use limited subset of NDT methods, since STRID STRID is an integral part of most SHM projects
deals with global structural behavior rather than localized
structural regions
Need to develop STRID methods that make better utilization of STRID methods are well developed
NDT methods
Most NDT projects do not utilize/need STRID methods Utilize/need STRID methods in many projects
TABLE 2.11
General Comparison of NDT and SHCE/SHM for Damage Identification
NDT SHCE/SHM
Localized identification of damage sources Efficient solutions are still being developed
If type and approximate location of damage is Damage location can be in a wider area; and cannot be detected
known, would give accurate results easily. Still under investigation
Due to localized nature, global structural Requires more information regarding structure (geometry, properties,
information is not needed. etc.)
Simple computational techniques Need extensive analytical methods
Higher reliability. No need for extensive statistical Reliability unknown yet. Need extensive statistical and reliability
and reliability analyses analyses
TABLE 2.12
General Comparison of NDT and SHCE/SHM for Decision Making
NDT SHCE/SHM
Many tools are available for NDE More involved decision making tools and processes are needed
due to the global nature (both in space and time) of SHM
Consequences of global failure are not usually Consequences of global failure must be considered
addressed
Vulnerability of global systems is not an issue due to Global system vulnerability is important
the local nature of the NDT
Due to short time scope of NDT, stochastic modeling is Time scope can be long (as long as the useful life of the target
limited system) and hence, stochastic modeling is important
Traditionally, risk as a tool, is not used. Only reliability Risk as a tool is used in addition to reliability (or probability of
(probability of failure) is used extensively failure) issues
that might affect bridges. Hence, we discuss some NDT technologies as they relate to four of the
common hazards—scour, earthquakes, corrosion, and fatigue. Figure 2.24 shows an experiment to
detect remaining fatigue life in a steel girder bridge. These four hazards affect the bridge in several
manners and thus, the use and efficiency of different NDT methods can vary considerably.
The following commonly used NDT methods are considered to illustrate the above.
Ultrasound/Sonar: Ultrasound methods are based on generating stress waves in the target
system at frequencies in the range of 0.10–35 MHz. The measured response of the system
will reveal information about the defects. Sonar waves also are used for similar purposes.
In a recent NYSDOT project, underwater sonar was used in an SHM setup that calls for
continued monitoring of soil level around bridge piers as shown in Figure 2.25 (Scour
2006). When the soil level falls below a critical scour level, it is detected by the sonar sys-
tem and the bridge officials can be warned.
FIGURE 2.24 Remaining fatigue life experiment. (Courtesy Keith Ramsey.)
FIGURE 2.25 Sonar setup for a scour project, elevation view. (Courtesy of New York State Department of
Transportation.)
Ultrasound Methods: Other ultrasound methods are used extensively in bridge health proj-
ects. Figure 2.26 shows an ultrasound detection process for bridge pin. The accuracy of
the methods is needed due to the importance of bridge pins to the overall well being of the
bridge. Impact echo is a popular and simple ultrasound technique that is also used success-
fully in bridge monitoring.
Acoustic Emission: The damage process produces stress waves that propagate away from the
damage locations, that is, acoustic emissions. The frequencies of these waves are gener-
ally in the range of 20 KHz to ~ 1.2 MHz. Recording and analyzing these stress waves can
assist in damage detection.
Thermography: Infrared thermography is used widely as an NDT technique to detect near-
surface defects. It is based on stimulating the material by heating/cooling it and observing
its surface temperature response to this stimulus. Knowing thermal properties of the mate-
rial, material defects can be detected. Possible use of infrared thermography to inspect
bridge column wrapped with fiber reinforced polymer materials is shown in Figure 2.27.
Figure 2.28 shows results of infrared thermography when detecting possible damage in
a bridge pavement. If the method is well developed, a base line image can be taken after
construction and can be used to monitor the deck periodically during its life.
FIGURE 2.26 Ultrasound monitoring of bridge pins. (Courtesy Barton Newton.)
FIGURE 2.27 Infrared thermography: bridge column. (Courtesy of New York State Department of
Transportation.)
FIGURE 2.28 Infrared thermography: bridge pavement. (Courtesy Keith Ramsey.)
FIGURE 2.29 Monitoring of cable stayed bridge. (Courtesy Keith Ramsey.)
Penetrating Radiation: When high frequency waves, such as X-rays or gamma-rays pen-
etrate solids, their back-scattered or penetrated traces can reveal information about the
interior of the solid, including any defects.
Vibration Monitoring: This technique is also used extensively in STRID and DMID of
bridges. Figure 2.29 shows an SHM project for monitoring cable stayed bridge behavior.
Other Methods: There are several other NDT methods that are used to monitor bridge health.
Some of these methods include magnetic-particle methods, magnetic flux leakage (Figure
2.30), dye penetrant (Figure 2.31) and ground penetrating radar.
Table 2.13 shows a sample of the applicability of these NDT methods for use in SHCE/SHM proj-
ects as they pertain to different hazards.
FIGURE 2.30 Monitoring for cracks using magnetic flux NDT method. (Courtesy Barton Newton.)
FIGURE 2.31 Dye penetrant capability in detecting surface cracks in bridge structures. (Courtesy Barton
Newton.)
2.4.4 Closing Remarks

This section explored the interrelationships between NDT and SHCE/SHM and showed some of
the differences between the two fields. It also illustrates the manner that the well-established NDT
field can help the emerging SHCE/SHM field to achieve their goals of safer infrastructure at reduced
cost. In conclusion, NDT has immense potential in advancing the SHCE/SHM field but should
address the specific and unique needs of SHCE/SHM.
We close this section by observing that for NDT methods to become more valuable in the SHCE/
SHM field, they should be
• Reliable
• Easily understood
• Justified in terms of value (benefit) versus costs
• Easy to use by routine inspectors
TABLE 2.13
Sample Applicability of NDT for SHCE/SHM in the Bridge Hazard Field
NDT Method
Bridge Hazard Ultrasound/Sonic Acoustic Emission Thermography Ground Penetrating Radar
Scour Sound waves in water No clear direct use of the technologies. An indirect use would be to
(Sonar) can identify spot-check structural damages due to scour settlement or failures
critical scour depth
of bridge foundations
See Chapter 1 of
Ettouney and
Alampalli (2012)
Earthquake Identifies material Identifies damage Condition Has been is used extensively
properties and after seismic events. assessment for condition assessment of
condition assessment Useful for most before and after bridge decks. Can be used
for structural construction seismic events. for condition assessment of
identification before materials Mainly for connections after seismic
and after seismic FRP-based events (Housner and Masri
events construction 1996)
Corrosion Detects corrosion Was shown to be Can detect Can be used for condition
presence indirectly, capable of detecting damages in monitoring of bridge decks
through detecting emissions from the FRP-wrapping
cracks or changes in rusting process. May material due to
material properties be used to degradation in
continuously underlying
monitor corrosion concrete
activities material
Fatigue Detects cracks due to Has potential to detect Can detect Can be used for fatigue-
fatigue active fatigue cracks damages due to related condition monitoring
fatigue in of bridge decks
FRP-based
construction
• Easy to implement at network level

• Durable and maintenance free (or close)
• Validated theoretically in the early stages of development
2.5 VALUE OF INFORMATION

2.5.1 Introduction
SHCE is all about gaining information about the structure on hand. Executing any SHCE project
would entail cost expenditure. Costs can be direct financial, labor, social, and so on. In numerous
situations most of the SHCE costs are required upfront or at early stages of the project. In return,
mostly information about the structure is obtained. This brings one of the most important questions
of the subject of this volume:
Are the information obtained through a particular SHCE project worth the costs involved?
Clearly, if the answer to this question is “no,” then the particular project under consideration is not
worth pursuing. If the answer is “yes,” then that project worth pursuing.
To find an accurate answer to the above question, we need to evaluate both the cost of the project
C as well as the benefit of the project B. If
C>B (2.6)
then the project is not worth pursuing, and if
C<B (2.7)
then the project is worth pursuing.

We can simplify the above equations even further by introducing the concept of value V where
B
V= (2.8)
C
So the evaluation of an SHCE project is reduced simply to evaluating the value of the project: if
V > 1 (2.9)
then pursue the project, and if
V < 1 (2.10)
then the project is not worth pursuing.

Since C, B, and V are paramount to the subject of SHCE, we will be addressing those issues
in most of the subjects of this volume, and will be trying to illustrate different methods and tech-
niques that would enable the professionals and decision makers to make accurate decisions regard-
ing specific SHCE projects. Of particular importance is that we need to define both C and B using
same units, to be able to find a numerical expression for V. It is easy to express C in monetary
terms. Thus, it is desired to express B also in monetary terms. Since both C and B are functions of
time their monetary estimates will also be the function of discount rate I throughout the desired
life span T.
The remainder of this section will be devoted to the more specific application of the above ques-
tion. We can restate the question as: can there be a value to a given type and quality of information?
We will offer a general approach to addressing the issue next.
2.5.2 General Approach to Evaluate Value of Information

Cost Estimation: Generally speaking, estimating C of any SHCE project is much easier than esti-
mating B. Elements of costs, as was mentioned earlier, include labor, equipments, analysis, design,
operations, and so on. The cost of the project C can be expressed as
i = NC
C= ∑ C (I ) p
i i (2.11)
i =1
where Ci(I) is the present value of the cost of the ith cost component. The probability of occurrence
of the ith cost component is pi. The total number of cost components is NC.
Benefit (Direct) Estimation: Elements of benefit are much more difficult to enumerate, espe-
cially in monetary terms. For example, some possible benefits of an SHCE project are direct, such
as increasing the life span of the structure, increasing the safety of the users, or reducing the costs
of operations. The direct benefits of the project BD can be expressed as
i = NC
BD = ∑ B (I ) Di pBDi (2.12)
i =1
where BDi(I) is the present value of the direct benefit of the ith direct benefit component. The prob-
ability of occurrence of the ith direct benefit component is pBDi.
Benefit (indirect) Estimation: Some indirect benefits of SHCE are gained knowledge about
particular monitoring technique, evaluation of specific type of sensors, applications of structural or
damage identification methods on an experimental basis. Can we put a monetary estimate to these
indirect benefits?
Indirect benefits are those benefits that might be achieved after several steps have been accom-
plished. For example: if an experiment in the validity of ultrasonic laser in damage identification
(Chapter 7) is under consideration. Since it is an experimental project, there are no direct benefits.
Of course, there are indirect benefits. To estimate those indirect benefits, we establish that as a
direct result of the experiment, there are two possible outcomes: success and failure. Each of those
outcomes has a particular probability of occurrence. In case of success, an experimental, system-
wide project will be established. Such a system-wide experiment will have a potential of success
or failure, each with its own occurrence probability. This will continue until the final effects of the
experiments will reach a self sustaining beneficial use. In this case, the potential use of ultrasonic
laser in identifying damage. It is at this moment that an actual monetary estimate of the final benefit
of the project can be made. The indirect direct benefits of the project BI can then be expressed as
i = NI j = NS
BI = ∑ BIi ( I ) ∏p ij
(2.13)
i =1 j =1
Where BIi(I) is the present value of the indirect benefit of the ith indirect benefit component. The
probability of success of the ith indirect benefit component at the jth step is pij.
The above method of estimating indirect benefit is a simple form of sequential decision tree
analysis. See Hamburg (1977) and/or Benjamin and Cornel (1970) for more details. In the next sec-
tion we present a method of estimating value of information with uncertainties.
Note that BI can be considered as a monetary measure of the benefits of information. We note
also that Equation 2.13 is a special form of Equation 2.12, with NC = 1.
On the basis of the above, the value of the information gained from an SHCE project is
C (2.14)
V=
BI + BD
We can then conclude that as a result of any SHCE project there always exists a finite value V for
the information that result from the project. It behooves the professional to ensure that the value of
such information is well understood and documented.
REFERENCES
Agarwal, A., Kawaguchi, A., and Qian, G., (2008). “Bridge Element Deterioration Rates: Phase I, Report,”
TIRC/NYSDOT Project # C-01–51, NYSDOT, Albany, NY.
Alampalli, S., (1998) “Effects of Testing, Analysis, Damage, and Environment on Modal Parameters.” Proceedings
of Modal Analysis & Testing. NATO-Advanced Study Institute, Sesimbra, Portugal, pp. 427–443.
Alampalli, S. and Ettouney, M., (2004) “Observations, Recommendations and Items of Interest to Health
Monitoring Community,” Proceedings of 2nd International Workshop on Structural Health Monitoring of
ASCE., (2007) Seismic Rehabilitation of Existing Buildings, ASCE / SEI Standard No 41-06, American Society
of Civil Engineers, Reston, VA.
Benjamin, J., and Cornell, A, (1970). Probability, Statistics, and Decision for Civil Engineers, McGraw-Hill,
New York, NY.
Borden, K. A. (2008) “Natural hazards Mortality in the United States,” Ph.D. thesis, University of South
Carolina, SC.
Cobb, R. G., (1996) “Structural damage identification from limited measurement data,” Ph. D. dissertation,
School of Engineering, Airforce Institute of Technology, Wright Patterson AFB, OH.
DeLisle, R., Sullo, P., and Grivas, D., (2002) “A network-level pavement performance model that incorporate
censored data,” TRP preprint Number 03–4281.
Dumova-Jovanoska, E., (2000) “Fragility curves for reinforced concrete structures in Skopje (Macedonia)
region,” Soil Dynamics and Earthquake Engineering, Elsevier, 19, 455–466.
Eghbali, G.H. and Elsayed, E.A., (1997) “Reliability Estimation Based on Degradation Data,” Working Paper
No. 97-117, Department of Industrial Engineering, Rutgers University, Piscataway, NJ.
Elsayed, E. A., (1996) Reliability Engineering, Addison-Wesley, Reading, MA.
Enright, M. P. and Frangopol, D. M., (1998) “Probabilistic Analysis of Resistance Degradation of Reinforced
Concrete Bridge Beams Under Corrosion,” Engineering Structures, 20(11), 960–971.
Ettouney, M. and Alampalli, S., (2000) “Engineering Structural Health” Proceedings of the Structural Engineers
World Congress, Philadelphia, PA.
Ettouney, M., and Alampalli, S., (2008) “Virtual Sensing in Structural Health Monitoring,” XXVI International
Modal Analysis Conference (IMAC), Orlando, FL, February 2008.
Ettouney, M. and Alampalli, S., (2012) Infrastructure Health in Civil Engineering: Applications and
Ettouney, M., Alampalli, S., and Agrawal, A., (2005) “Theory of multihazards for bridge structures,” Journal of
Bridge Engineering, IOS Press, 1(3), 281–291, Amsterdam, Netherlands.
Ettouney, M. and Elsayed, E. A., (1999) “Reliability Estimation of Degraded Structural Components Subject to
Corrosion,” Fifth ISSAT International Conference, Las Vegas, NV.
Ettouney M., Daddazio, R. and Hapij, A., (1998) “Health Monitoring of Complex Structures” SPIE Smart
Structures and Materials Conference, San Diego, CA.
Ettouney, M., Daddazio, R., and Hapij, A., (1999) “Optimal sensor locations for structures with multiple load-
ing conditions” SPIE Smart structures and Materials Conference, San Diego, CA.
Farrar, C. Duffy, T., Doebling, S., and Nix, D., (2000) “A statistical Pattern Recognition Paradigm for Vibration-
Based Structural Health Monitoring,” Proceedings of 2nd International Workshop on Structural Health
Monitoring, Stanford University, Stanford, CA.
FEMA-445., (2006) Next-Generation Performance-Based Seismic Design Guidelines, FEMA Report No 445,
Washington, DC.
Ganji, V., Tabrizi, K. and Vittilo, N., (March 2000) “Project Level Application of Portable Seismic Pavement
Analyzer,” Proceedings of the SMT/NDT Conference, NYS-DOT, NJ-DOT and FHA, Atlantic city, NJ.
Hamburg, M., (1977) Statistical Analysis for Decision Making, Harcourt Brace Jovanovich, Inc., New York, NY.
Helmicki, A., Hunt, V. J., and Lenett, M., (2002) “Non-Destructive Testing and Evaluation of the Ironton-
Russell Bridge,” Proceedings of NDE Conference on Civil Engineering, ASNT, Cincinnati, OH.
Housner, G. and Masri, S., (1996) “Structural Control Issues Arising from the Northridge and Kobe Earthquakes,”
11th World Conference on Earthquake Engineering, Mexico, Paper No. 2009.
IBC., (2006) 2006 International Building Code, International Building Council, Washington, DC.
Melchers, R. E., (1987), Structural Reliability, Ellis Horwood Limited.
Melchers, R. E., (2002) Structural Reliability Analysis and Prediction, John Wiley and Sons, New York, NY.
Mostafa, E. and Grigoriu, M., (1999) “Fragility curves for non-structural systems,” Report No. 99-2, School of
Civil and Environmental Engineering, Cornell University, Ithaca, NY.
Oswald, G. F. and Schuëller, G. I., (1983) “On the Reliability of Deteriorating Structures,” in Reliability Theory
and Its Application in Structural and Soil Mechanics, Editors: Thoft-Christensen, P. and Nijhoff, M. The
Hague, Netherlands.
Pugsley, A., (1962) Safety of Structures, Edward Arnold, London.
Reinhorn, A., Barron-Corvera, R., and Ayala, A., (2001) “Spectral Evaluation of Seismic Fragility of Structures,”
in Structural Safety and Reliability (ICOSSAR 2001), Editors: Corotic et al., Swets & Zeitlinger, Balkema.
“Scour Monitoring Program Manual,” Hardesty & Hanover, prepared for New York State Department of
Transportation, July 2006.
UBC-97., (1997) Uniform Building Code, International Conference of Building Officials, Whittier, CA.
3 Lives and Deaths of
Infrastructure
3.1 OVERVIEW
We explore in more detail the concept of bridges (or any other civil infrastructure) as an organism
that is born, lives and dies (fails). We note that, similar to humans, there are three distinct phases of
bridge life cycle: (a) its birth, (b) its normal existence (life), and (c) its death (failure). We propose
that ensuring health of bridges applies to all phases of their life cycles. We immediately define the
main attributes of healthy bridge during all phases as follows:
• A healthy bridge meets or exceeds stated performance goals

• At reasonable costs
Among performance goals we can state further that the bridge should experience no unwarranted
service interruptions due to distress, and partial or total failure. A healthy bridge should also per-
form its stated functions, such as load-carrying specifications, in a safe and secure fashion. Near the
end of the life cycle of the bridge, it is essential to control the decommissioning process, as opposed
to abrupt failure/distress.
Overall, we aim at discussing categories that would enable decision makers to quantify differ-
ent phases of life cycle of the bridge; we also explore different roles structural health monitoring
(SHM)/structural health in civil engineering (SHCE) during life cycle of the bridge, and how those
roles change as the bridge life span change. Understanding these details can enhance the optimal
health of bridges during all the phases of its existence.
3.1.1 Enabling versus Triggering Causes

To simplify examinations of health concepts of bridges, we categorize causes that affect bridge (or
any other type of structure) into enabling and triggering causes. The categorization was used by
Wardhana and Hadipriono (2003) to explore bridge failures. We extend the definition into other
phases of bridge life span (construction, service life, and ultimate failure). Table 3.1 shows a general
comparison between the two types of causes.
3.1.2 This Chapter

This chapter explores the health of a bridge at different phases of its life cycle as shown in Figure 3.3.
First the health of the bridge during the birth phase is described. The different issues regarding
health during the life span of the bridge is then surveyed. The third phase, bridge failure, or decom-
missioning, is discussed next. In all life cycle phases the triggering and enabling causes of struc-
tural health are defined and explored. The many roles SHM methodologies and techniques play at
different phases are discussed in details. Whenever possible, cost–benefit issues of different SHM
roles are sited. The chapter ends with an in-depth survey of important bridge failure examples in the
United States in the past one hundred years.
81
TABLE 3.1
Enabling versus Triggering Causes
Causes Enabling Triggering
Temporal behavior Occurs over a long period of time (see Figure 3.1) Occurs suddenly (see Figure 3. 2)
Severity Both can have severe effects on bridges
NDT role Can be useful Effective only before and after event
SHM role Long-term monitoring; intermittent monitoring Before, during and after event monitoring
might require different strategies
FIGURE 3.1 Example of enabling cause: corrosion of bridge column. (Courtesy of New York State
Department of Transportation.)
FIGURE 3.2 Example of triggering cause: overload on a bridge. (Courtesy of New York State Department
of Transportation.)
3.2 BIRTH OF BRIDGES

3.2.1 Ensuring Healthy Life
Healthy life of structures starts at “birth.” Several phases of birth include (a) preliminary planning,
(b) analysis, design, and detailing, (c) construction, and (d) quality assurance/quality control (QA/
QC). SHM/SHCE can play many roles during all of those phases. These roles are described next.
Lives and Deaths of Infrastructure 83
Lives and deaths of bridges: A

quantitative outlook
Birth of bridges
Roles of
Bridge
Healthy life of bridges: cycle of
SHM/ Why bridges live? life
SHCE
Death of bridges:
Why bridges fail?
The cycle continues
FIGURE 3.3 Organization of Chapter 3.
3.2.1.1 Preliminary Planning

Preliminary planning of a bridge includes numerous roles of SHM techniques and methods. Perhaps
the most obvious are different decision making (DM) processes. Activities such as cost–benefit
analysis, finance and economics implications, and social studies must precede any bridge construc-
tion phase. Also an exhaustive risk and life cycle studies must be done. Risk analysis would include
different uncertainties that the bridge might encounter during its life span. Life cycle analysis of the
bridge is performed. Implications of different decisions regarding the bridge should be analyzed.
Decisions on optimal attributes of the bridge can then be made. Those DM methodologies are dis-
cussed in detail in Chapter 8.
Other SHM components also have roles in the preliminary planning phase. For example, soil
explorations can affect decisions regarding the bridge, including the very viability of constructing it.
STRID activities that make use of available data, such as soil exploration results can help in ensur-
ing healthy life of the bridge, after it is built.
3.2.1.2 Design Phase

Analysis phase of new bridges can include STRID methods for optimal designs. Since STRID
methods (Chapter 6) need actual test results as an input, simulated test results might be used.
Simulated results can be generated using similar previous tests or any other appropriate technique.
Another important measure can be to prepare base analytic models for future STRID/DMID activi-
ties. Such base analytic models can be for the global bridge structure, or for important local details.
As future actual datasets start accumulating, those analytic models can be updated and improved
to help in different future decisions and activities such as DMID, inspection, maintenance, and
rehabilitation.
Design phase can utilize available test results (such as soil test results) to aid design paradigms,
whether it is the capacity/demand or performance-based designs. Also, the design of future mainte-
nance, rehabilitation, retrofits, and decommissioning activities has to be compatible with the overall
bridge design. The designer should ensure that all such designs are optimal. A comprehensive life
cycle analysis plan should be intertwined with the initial designs of the bridge. Such life cycle analy-
sis plans should be scalable, flexible, and easy to be adjusted to future events.
Details of the bridge SHM projects must be included with the initial structural details. Such
SHM details should be for (a) long-term monitoring, (b) intermittent monitoring, (c) one-time spe-
cialized in situ monitoring such as bridge load testing, and (d) laboratory testing, such as different
NDT techniques. The bridge designer should also try to accommodate conventional visual inspec-
tion by using simple detailing as well as easy-to-inspect detailing such that other expensive options
can be minimized.
3.2.1.3 Construction Phase

Innovative uses of SHM during construction phase of large civil infrastructure can produce many
dividends. An example of such innovations was reported by Ni et al. (2008). The subject of the proj-
ect is the Guangzhou New Television Tower, which is located in Guangzhou China (see Figure 3.4).
It is a 610-m high structure that is constructed of a mix of reinforced concrete and concrete-filled
steel tubes. The Hong Kong Polytechnic University designed the SHM project with two major
goals: (a) to monitor in-construction performance, and (b) to monitor in-service performance.
Both goals are done in real time. In addition, the SHM project has other innovative features such as
(a) a modular design of the SHM system, (b) efficient wireless-sensing array, (c) advanced fiber
optics sensing that is based on Bragg grating technology, and (d) structural identification process-
ing using static and dynamic monitoring data. The project also has a sophisticated system for data
validation and benchmark case studies.
What interests us in this section is the implementation of a during-construction SHM system by
Ni et al. (2008) (see Figure 3.5). They used a total of 527 sensors that were placed at 12 cross-sec-
tions of the tower to monitor performance during construction. The choices of the sensor locations
were obtained by analyzing the structure during construction process and placing them at sections
with higher stress levels. Most of the sensors were strain sensors (total of 416) to monitor strains,
creep, and shrinkage (vibrating wire strain gauges). In addition, large number of temperature sen-
sors was used (total of 96). Some additional weather, wind speed, inclination, leveling of floors and
tower, and displacement sensing were utilized. The use of these SHM sensors enabled the project
construction engineers to ensure higher quality and safe construction process of the tower.
FIGURE 3.4 Guangzhou New Television Tower. (With permission from John Wiley & Sons Ltd.)
INNER TUBE OUTER TUBE

Vibrating wire strain gauge (24)
Vibrating wire strain gauge (12) loop45 Temperature sensor (8)
SUB-STATION (1) 433.2n

loop40 Temperature sensor (8)
SUB-STATION (1) 376.0n Vibrating wire strain gauge (24)
Vibrating wire strain gauge (12) loop38 Temperature sensor (8)
SUB-STATION (1) 355.2n Vibrating wire strain gauge (24)
334.4n loop35 Temperature sensor (8)
SUB-STATION (1)
Vibrating wire strain gauge (12) Vibrating wire strain gauge (24)
SUB-STATION (1) 303.2n loop32 Temperature sensor (8)
loop21 Temperature sensor (8)
SUB-STATION (1) 204.4n
SUB-STATION (1) loop17 Temperature sensor (8)
173.2n
SUB-STATION (1) 121.2n loop11Temperature sensor (8)
Vibrating wire strain gauge (12) 100.4n Vibrating wire strain gauge (20)
loop9
Temperature sensor (8)
SUB-STATION (1)
Weather station (1)
loop3
Temperature sensor (8)
SUB-STATION (1) 32.8n GPS reference station (1)
Digital camera (3)
SUB-STATION (1)
FIGURE 3.5 In-construction SHM sensor placement for Guangzhou New Television Tower. (With permis-
sion from John Wiley & Sons Ltd.)
We should also note that one of the innovations of the SHM project was to integrate the in-
construction monitoring with a long-term in-service monitoring. In addition, 280 sensors were used
for the in-service SHM task. Of these, arrays of fiber optics Bragg grating sensor network were
installed in permanent steel conduits. A total of 120 Bragg grating strain and temperature sensors
were utilized for this task.
3.2.1.4 Quality Control/Quality Assurance (QA/QC)

Bridges, as well as all other major civil infrastructure, involve complex and lengthy construction
process. The importance of performing construction tasks in accordance with the design and detail-
ing specifications is exemplified in a usually rigorous and detailed QA and QC strategies. SHM
methods are emerging as an accurate and inexpensive construction phase QA/QC device. Simple
monitoring of different attributes such as concrete temperature, humidity, ambient temperature,
and/or member alignment, can help in QA/QC processes. Consider, for example, the SHM experi-
ment by Hansen and Surlaker (2006). They aimed at coupling the maturity principle of concrete
during the curing phase and real time SHM activities in providing real time and accurate QA/QC
information that can save costs, and ensure life cycle safety of the system. The maturity principle
simply relates the curing time and the curing temperature to the concrete strength. Thus, by know-
ing the elapsed time, and concrete temperature, it was shown that an accurate estimate of concrete
strength is possible. Two important reasons as to why accurate estimation of cured concrete strength
is important: (a) it permits the removal of temporary forms when the concrete reaches postulated
minimum strength, and (b) it signals the optimal time for post-tensioning in post-tensioned systems.
The authors used a set of embedded radio frequency ID (RFID) sensors in two pilot structures: a
parking garage, and a concrete highway pavement. They reported the following major benefits for
their technique.
• In situ estimation of concrete strength eliminates the need for using laboratory cylinder tests,
thus saving time and cost, while ensuring an even more accurate strength estimation.
• When used for a post-tensioned system, an online estimation of concrete strength would
reveal the exact time when the concrete strength would permit the post-tensioning process.
This optimizes the time utilization during construction, while ensuring safe post-tension-
ing operations.
• By placing RFID sensors across the depth of the pavement, it was possible to monitor the
temperature gradient across the depth. This information can help predict any curling of the
pavement; such curling can lead to a premature surface fatigue cracking and reduction of
the service life of pavements.
Several other benefits of the experiment were reported by Hansen and Surlaker (2006). The authors
provided estimates of cost savings that the use of their technique can provide the owners during the
construction phase. We observe that there are additional savings that can be had during the life span
of the system. Those additional savings result from factors such as improvements to fatigue strength
limits, and reduced maintenance costs (by reducing or eliminating early tensile pavement cracking).
It is clear that such innovative, yet simple and inexpensive use of SHM technologies can help in QA/
QC of bridges during their construction phase.
One of the applications where NDT is very useful is quality control and assurance of bridges
and pavements during construction. Garg et al. (2008) described a portable seismic property ana-
lyzer (PSPA) used to determine the modulus of concrete pavement slabs at FAA’s National Airport
Pavement Test Facility (Figure 3.6).
PSPA is a portable device (see Figure 3.7) and consists of a receiver transducers and a source
transducer; and is based on generating and detecting stress waves in a layered medium. The data
collected is processed by spectral analysis to determine the modulus of the layer. Advantages of
using PSPA include nondestructive in nature, rapid to perform, immediate availability of results,
repeatable, in situ testing of pavement in its natural state, and easy and convenient to operate.
FIGURE 3.6 FAA National Airport Pavement Test Facility. (Reprinted from ASNT Publication.)
Receivers Source
FIGURE 3.7 Portable seismic property analyzer (PSPA). (Reprinted from ASNT Publication.)
PSPA used estimate modulus values from tests on three concrete slabs under three different
support conditions at different times after the concrete pours. The results compared well with the
modulus values measured by free-free resonance tests performed on the concrete beam and cylin-
drical specimens prepared at the same time as the concrete slabs. The same PSPA device was also
used to determine the modulus of hot mix asphalt (HMA) layer during CC2-OL (HMA overlay over
rubblized concrete) project except that the active length of the seismic path was shortened. These
two studies demonstrated the application potential of a NDT device such as PSPA for assessing the
early-age strength of concrete and the modulus of HMA layer.
3.2.2 Role of SHCE

Structural health monitoring/structural health in civil engineering tools and methods can
play major role during early phases of bridge life span. Table 3.2 shows some examples of such
a role.
3.3 WHY BRIDGES LIVE?

3.3.1 Definitions of Healthy Life
Healthy life of bridges can be expressed in three facets: (a) meeting functional goals, (b) adequate
performance, and (c) at reasonable costs. Functional goals include traffic, pedestrian, and commer-
cial needs. Performance metrics include ensuring safety and security of users. This can be accom-
plished, in technical terms, by meeting adequate engineering criteria such as deflection, vibration,
and stress/strain limits. Reasonable costs, or adequate cost benefits, are a major issue that is dis-
cussed in numerous parts of this volume.
This section addresses healthy life of bridges, its causes, and how to quantify a healthy bridge
life. We offer some specific examples as to how SHM/SHCE field can improve healthy bridge life.
3.3.2 Classification of Healthy Life Causes

3.3.2.1 Enabling Causes of Healthy Life
By definition, enabling causes are continuous: they occur nearly all the time and their effects are felt
over the long term. There are many enabling causes that would ensure healthy life of a bridge. To
continue the framework of quantification of healthy life, we assign formal variables to these causes
as discussed in Table 3.3.
TABLE 3.2
Roles of SHM in Early Phases in Bridge Life Span
Ensuring
Healthy Life for
New Bridges Sensing/Measurements STRID DMID Decision Making
Preliminary Soil conditions need to be Conventional Literature searches for • Cost–benefit
planning measured. In some structural analysis damages and methods of analysis
situations, special sensing methods can be DMID for bridges and sites • Risk analysis
might be needed, such as used for that are comparable to the • Life cycle
propensity for liquefaction preliminary bridge and the site that are analysis
in areas with high seismic planning under considerations. The
activities results of these searches can
have two benefits:
• Affects decisions for the
current project
• Affects decisions on any
DMID method to be used
in an SHM setup for the
current project, if any
Design/detailing Placement of future sensors. Base analytical Optimal integration of SHM Optimal decisions
Types, numbers, and STRID models for activities and structural that accommodate
locations of the sensors updating at design of the bridge • Cost–benefit
Optimal NDT methods different stages of • Life cycle
the bridge life cycle analysis
Construction Safe construction process can
be monitored by strain,
displacement or vibration
sensors that are placed in
optimal locations. Those
locations are predetermined
using structural analysis tools
QA/QC Adequate curing of concrete
can be monitored using
embedded RFID
temperature sensors
Quality of welds can be
inspected using different
NDT techniques
TABLE 3.3
Enabling Causes for Healthy Life
Enabling Cause Objective Rating
Sound management 0 ≤ EL1 ≤ 10
Design 0 ≤ EL2 ≤ 10
Detailing 0 ≤ EL3 ≤ 10
Construction 0 ≤ EL4 ≤ 10
Scheduled maintenance 0 ≤ EL5 ≤ 10
Inspection/monitoring 0 ≤ EL6 ≤ 10
Other To be added as needed
We thus establish the enabling measure of a bridge healthy life as

i = NE
∑
HLTH E = WE (WEi i ELi ) (3.1)
i =1
where
WEi = Appropriate weight for the ith parameter
WE = Appropriate weight for the whole enabling causes
NE = Total number of enabling causes
Note that in Table 3.3 NE = 6. The number of enabling causes can change, depending on the situ-
ation on hand. The different weights in Equation 3.1 are assigned based on experience and research
results. The measure HLTHE is thus a relative measure that can be used in DM situations, such as
allocation of budgeting, and so on.
3.3.2.2 Triggering Causes of Healthy Life
Triggering healthy life causes are relatively sudden: they occur suddenly on the time scale of the
bridge life span; their effects are felt suddenly in the form of improved bridge health. There are
many triggering causes that would ensure healthy life of a bridge. Similar to the enabling causes,
we assign formal variables to triggering causes as discussed in Table 3.4.
We thus establish the triggering measure of a bridge healthy life as
i = NT
HLTHT = WT ∑ (WTi i TLi ) (3.2)
i =1
where
W Ti = Appropriate weight for the ith parameter
W T = Appropriate weight for the whole triggering causes
N T = Total number of triggering causes
Note that in Table 3.4 N T = 4. The number of triggering causes can change, depending on the situ-
ation on hand. The different weights in Equation 3.2 are assigned based on experience and research
results. The measure HLTHT is thus a relative measure that can be used in DM situations, such as
allocation of budgeting, and so on.
3.3.3 Quantification of Healthy Life
3.3.3.1 Conventional Metrics
There are numerous methods to quantify the health of a bridge; see, for example, NYSDOT (2008).
Among some of those methods:
Bridge Condition Rating: A simple rating system from 1 (totally deteriorated condition) to N (new
condition). The value of N vary. For example, New York State uses N = 7. The value of the condition
rating indicates the health condition of the bridge. Table 3.5 shows qualitative description of the condi-
tion ratings as used in New York State.
TABLE 3.4
Triggering Causes for Healthy Life
Triggering Cause Objective Rating
Retrofits 0 ≤ TL1 ≤ 10
Rehabilitation 0 ≤ TL2 ≤ 10
Special maintenance 0 ≤ TL3 ≤ 10
Replacement 0 ≤ TL4 ≤ 10
Other To be added as needed
TABLE 3.5
Definitions of Bridge Condition Ratings
Rating Definition
1 Totally deteriorated or failed condition
2 Used for condition between 1 and 3
3 Serious deterioration or not functioning as originally designed
4 Used for conditions between 3 and 5
5 Minor deterioration and is functioning as originally designed
6 Used for conditions between 5 and 7
7 New condition
TABLE 3.6
Categories of Bridge Deficiencies
Condition Rating Deficiency Level Description Potential Remedial Action
<3 Severe Comprehensive serious deterioration Priority to remediate and repair
of the bridge structural elements
3.0 and ≤ 3.999 Moderate Serious deterioration to some of the Comprehensive structural work is
bridge main structural elements likely. Rehabilitation and
replacement options are
generally available
4.0 and ≤ 4.999 Marginal Moderate structural deterioration. Minor rehabilitation/major
Minor deterioration to primary maintenance activities might be
support elements Efficient structural needed
performance is rarely compromised
One common way to quantify the bridge condition rating is described in detail in NYSDOT
(2008). The procedure is summarized as follows: during each general inspection of the bridge,
various components, or elements of each bridge span are rated by the inspector according to the
extent of deterioration and the ability of the component to function structurally, relative to when
it was newly designed and constructed. These element rating values are then combined using a
weighted average formula to compute an overall bridge condition rating value for each bridge. This
formula assigns greater weights to the ratings of the bridge elements having the greatest structural
importance and uses lesser weights for minor structural and nonstructural elements. If a bridge has
multiple spans, each element common to multiple spans is rated on a span-by-span basis; the lowest
individual span element rating is used in the overall condition rating formula.
Deficiency: A deficient bridge can be defined using its condition rating value. For example,
NYSDOT defines deficient bridges as those with a condition rating less than 5 (on a scale from 1
to 7). A deficient condition rating indicates the presence of sufficient deterioration and/or loss of
original function to require corrective maintenance or rehabilitation to restore the bridge to its fully
functional, nondeficient condition. It does not mean that the bridge is unsafe. In most cases, bridges
have enough excess or reserve structural capacity to accommodate some deterioration or degrada-
tion of structural function as indicated by a deficient condition rating. Figure 3.13 shows the number
of deficient bridges in New York State between 1992 and 2006. The trend of the deficient bridge
numbers is decreasing. To correlate state of deficiency with DM, NYSDOT (2001) in a report on
bridge overloading identified further different states of bridge deficiency, their condition rating and
potential remedial actions in Table 3.6.
Posting: When the rating of a bridge is reduced even further to a rating of 3 or less, some pro-
tective measures might be needed. One example is the posting of weight limits that are allowed
on the bridge. Also, the bridge might be closed until it can be repaired, rehabilitated, replaced or
permanently closed. Another qualitative indication of bridge health is red flags. These flags identify
potentially or imminently unsafe structural conditions and require the owner to take prompt, certi-
fied corrective, or protective actions to resolve the flag. These include repair, posting, or closure.
Figure 3.14 shows the number of red flags issued in New York State between 1989 and 2006.
Alampalli and McCowan (2008) described in detail the flagging procedure, which is a practical
measure to describe state of bridge health, in New York State. It is part of the comprehensive bridge
inspection program implemented in New York State to identify serious deficiencies, both structural
and nonstructural, affecting public safety so that owners can take appropriate action in a timely
fashion. The procedure is very robust, safety oriented, and establishes requirements for certifying
that appropriate corrective or protective measures are taken within an appropriate time frame. The
critical inspection findings (flags) can be either structural or safety related. The structural flags are
further subdivided into two categories.
Red structural flags are used to report the failure of a critical primary structural component or a
failure that is likely before the next scheduled inspection (see Figures 3.8 and 3.9).
Yellow structural flags are used to report a potentially hazardous condition which, if left unat-
tended beyond the next anticipated inspection, would likely become a clear and present danger. This
flag is also used to report the actual or imminent failure of a noncritical structural component, where
such failure may reduce the reserve capacity or redundancy of the bridge, but would not result in a
structural collapse by the time of the next scheduled inspection interval (see Figures 3.10 and 3.11).
FIGURE 3.8 Red flag due to serious cracking in abutment stem. (Courtesy of New York State Department
of Transportation.)
FIGURE 3.9 Red flag due to overextended bearing. (Courtesy of New York State Department of Transportation.)
FIGURE 3.10 Yellow flag due to a crack in the primary redundant member. (Courtesy of New York State
FIGURE 3.11 Yellow flag due to loss of bearing area in a pedestal. (Courtesy of New York State Department
of Transportation.)
Nonstructural conditions are reported using a “Safety Flag.” The safety flag is used to report a
condition presenting a clear and present danger to vehicle or pedestrian traffic, but is in no danger
of structural failure or collapse. Safety flags can also be issued on closed bridges whose condition
presents a threat to vehicular or pedestrian traffic underneath the bridge (see Figure 3.12).
The notification and response procedures vary depending on the flag type.
US Federal Rating System: The federal ratings result from an overall condition assessment of
each bridge’s three or four major components and do not require the multielement evaluations man-
dated by other states, such as NYSDOT inspection program. The federal ratings are used to iden-
tify bridges that do not meet contemporary Federal Highway Administration (FHWA) standards.
Those bridges are classified as either “structurally deficient” or “functionally obsolete.” Bridges
are considered “structurally deficient,” according to the FHWA, if significant load-carrying ele-
ments are found to be in poor or worse condition due to deterioration and/or damage, the bridge has
inadequate load capacity or repeated bridge flooding causes traffic delays. The fact that a bridge is
“structurally deficient” does not imply that it is unsafe or is likely to collapse.
A “structurally deficient” bridge, when left open to traffic, typically requires significant main-
tenance and repair to remain in service and eventual rehabilitation or replacement to address the
FIGURE 3.12 Safety flag due to exposed wiring. (Courtesy of New York State Department of
Transportation.)
TABLE 3.7
Count of Bridges by Construction Material
Prestressed
Count of Bridges Concrete Steel Concrete Wood Othersa
Total stock 248,739 188,551 132,033 26,682 3802
Structurally deficient 18,506 38,419 5036 9855 710
Functionally obsolete 28,187 34,678 12,773 3482 685
a Includes: Masonry, aluminum, wrought iron, or cast iron, and other.
deficiencies. To remain in service, structurally deficient bridges are often posted with weight limits.
“Functionally obsolete” refers to a bridge’s inability to meet current standards for managing the
volume of traffic it carries, not its structural integrity. For example, a bridge may be functionally
obsolete if it has narrow lanes, no shoulders, or low clearances.
The federal bridge rating scale addresses both structural condition and functional adequacy, using
different criteria than the above mentioned NYSDOT inspection condition rating scale. Table 3.7
and Figure 3.13 show structurally deficient and functionally obsolete highway bridges in the United
States as categorized by construction material. Table 3.8 and Figure 3.14 show structurally deficient
and functionally obsolete bridges in the United States as categorized by bridge type.
We end this section by observing that a good state inspection practice should include both Federal
ratings as well as the state’s own ratings, if available (Figures 3.15 and 3.16).
3.3.3.2 Healthy Life Causes as Metrics
We can quantify the health of a bridge as
HLTH = HLTH E + HLTHT (3.3)

Note that the health of the bridge measure HLTH is a qualitative measure that should be used for
DM situations. It should be used as a complementary measure to other bridge performance mea-
sures such as bridge ratings. The main difference between HLTH and bridge rating is that bridge
rating evaluated the field condition of bridge. On the other hand, HLTH evaluates other factors that
lead to the current bridge condition. Both measures account for different set of parameters, thus the
consideration of both can give a more complete picture of the bridge health.
Count of bridges (NBI 2007)
250,000
200,000
150,000
100,000
50,000
0
Concrete Steel Prestressed concrete Wood
Material type Structurally deficient Functionally obsolete
FIGURE 3.13 Count of bridges by construction material. (From FHWA, National Bridge Inventory, NBI,
Federal Highway Administration, Washington, DC, 2007.)
TABLE 3.8
Count of Bridges by Structure Type
Stringer/ Girder and Box Beam or Truss Suspension/
Count of Multi Beam Floor Beam TEE Girders (Single or (Deck and Stayed Othersa
Bridges Slab or Girder System Beam Spread/Multiple) thru) Girder (Culvert)
Total stock 79,879 249,238 7432 36,444 55,330 12,608 133 158,726
(126,401)
Structurally 6600 39,911 2594 4770 3189 7161 36 8262
deficient (2944)
Functionally 9612 39,819 1792 7807 6660 2455 44 11,609
obsolete (5543)
a Includes: Frame; orthotropic; arch-deck and arch-thru; movable (lift, bascule, and swing); tunnel; mixed types; segmental
box girder; channel beam; other; and culverts.
Count of bridges (NBI 2007)
250,000
200,000
150,000
100,000
50,000
0
Slab Stringer / Girder & Tee beam Box beam Truss
multibeam floor beam or girder
or girder system
Total stock Structurally deficient

Functionally obsolete
FIGURE 3.14 Count of bridges by structure type. (From FHWA, National Bridge Inventory, NBI, Federal
Highway Administration, Washington, DC, 2007.)
Deficient bridges by number of bridges

60.00%
50.00%
Percent deficient
40.00%
30.00%
20.00%
10.00%
0.00%
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
Year
State Local Total
FIGURE 3.15 Percent of deficient bridges. (Courtesy of New York State Department of Transportation,
NYSDOT 2008.)
Red flags
600
545
524 535
500
451
434 433
420 418
393 388
400 375
350
Number of red flags
349
319
299
288
300
253
253 248
230
202
191 194
200 184 178 181
160 153
147 144
127 123 131 120 175
102
90
100
101 96 62 109
91 87 88 79
59 58 60 63
0
89
90
91
92
93
94
95
96
97
98
99
00
01
02
03
04
05
06
19
19
19
19
19
19
19
19
19
19
19
20
20
20
20
20
20
20
Year
Totals State Locals
FIGURE 3.16 Number of red flagged bridges. (Courtesy of New York State Department of Transportation,
NYSDOT 2008.)
To give an example of the use of HLTH let us consider an example of two bridges. The official
needs to compare the health of the bridges. By making an in-depth study of all enabling and triggering
causes, the different estimated factors were computed (qualitatively) as shown in Tables 3.9 and 3.10.
Notice that the relative weights in both Tables 3.9 and 3.10 are assigned by the official. More studies are
needed for the nature and the values of these weights. The final estimates for HLTH are 695 and 763 for
bridges A and B, respectively. This outcome indicates that bridge B should be healthier than bridge A.
TABLE 3.9
Example of Enabling Causes for Healthy Life of Two Bridges
Enabling Cause Weight Objective Rating of Bridge A Objective Rating of Bridge B
Sound management 10 7 9
Design 8 6 6
Detailing 10 4 7
Construction 10 7 5
Scheduled maintenance 12 8 6
Inspection/monitoring 12 8 9
Total 420 438
TABLE 3.10
Example of Triggering Causes for Healthy Life of Two Bridges
Triggering Cause Weight Objective Rating of Bridge A Objective Rating of Bridge B
Retrofits 10 5 5
Rehabilitation 10 6 5
Special maintenance 15 8 8
Replacement 15 3 7
Total 275 325
3.3.4 Role of SHM

Structural health monitoring/structural health in civil engineering can play a major role in almost
all enabling and triggering bridge healthy life. Some specific examples are shown in Table 3.11.
The project that was described by Kosnik and Hopwood (2008) provides an example of healthy
bridge life that was aided by HSM monitoring. The John F. Kennedy Memorial Bridge, a large
cantilever through truss opened in 1963, carries Interstate 65 across the Ohio River between
Louisville, KY, and Jeffersonville, IN. An overall view of the bridge is shown in Figure 3.17. The
bridge is restrained by pairs of bearing assemblies on each bank of the river. The bearings resist
considerable designed-in uplift forces, particularly on the Indiana side. In 2006, one of the anchor
bolts on the northwest bearing was found to have been fractured (see Figure 3.18). The washer on
the bolt opposite the failed bolt could be spun in place by hand, indicating that there was very little
tension in that bolt. The bearing assembly with the fractured bolt also moved visibly under live
traffic.
Live strain, displacement, and acceleration data were collected on uplift bearing anchor bolts for
a total of approximately 17 hours over several weeks under a variety of weather and traffic condi-
tions and both before and after replacement of a fractured anchor bolt (see Figure 3.19). On the basis
of initial data it was determined that the compromised North West, NW bearing assembly was sub-
ject to large live strains in two of the three remaining anchor bolts. The fractured North West-North
East (NW-NE) anchor bolt was replaced with a threaded rod instrumented with strain gauge arrays,
and strain data on the three original anchor bolts and the replacement bolt were recorded before
and after installation and tightening of the replacement. The live strains in the original anchor bolts
decreased after tightening of the replacement bolt, and no further indications of bending of the
anchor bolts were observed.
TABLE 3.11
SHM Role in Some Bridge Healthy Life
No. Healthy Life Issues SHM Applications
1 Rehabilitation/repair choice, closing/opening/usage- Monitoring and evaluating condition state as is.
restriction of the structure, evacuation for safety STRID and DMID projects
2 Evaluation/implementation of new designs/construction New and existing engineering paradigms
methodology at local or global level
3 Quality of life (improving mobility, etc. —opening Bridge testing and life cycle analysis.
appropriate lanes or redirecting traffic in-time)
4 Maintenance decisions (routine to using just-in-time Bridge management tools and methods Applying
concept—say sending a salt truck to a bridge only when reliability and risk methods for maintenance
you see that such conditions warrant that will help decisions
structural durability, saves money, and improves safety)
5 Safety (users and structural safety) Engineering design and analysis paradigms
6 Evaluation (durability, design/analysis option evaluation, etc.) Deterioration monitoring and decision making tools
7 Research, development, and technology transfer
a. Improving the state-of-the-art SHM/SHCE in civil infrastructure is in its infancy
b. Improving the state-of-the-practice (as of writing of this volume) more research is
c. Bridging the gap between the state-of-the-art and needed in all components of the fields
state-of-the practice
8 Security (relatively new in bridge field, but probably There are many common needs and applications for
common for security applications you deal with) bridge security and SHM (see Chapter 11 of
Ettouney and Alampalli 2012)
FIGURE 3.17 Overall view of the I-65 John F. Kennedy Memorial Bridge. (Reprinted from ASNT Publication.)
3.4 WHY BRIDGES FAIL/DIE?

3.4.1 Definitions of Failure
There are several definitions of bridge failure. Most of those definitions are qualitative. A general failure
definition was provided by NYSDOT (2001), which categorized failure into three groups as follows
Catastrophic: The structure is vulnerable to a sudden and complete collapse of a superstruc-

ture span or spans. This failure maybe the result of partial or total failure of either the
superstructure or the substructure. A failure of this type would endanger the lives of those
on or under the structure.
Northeast pier 6
Northwest pier 6 NE-NW NE-NE
NW-NW NW-NE
Fractured
bolt
NW-SW NW-SE NE-SW NE-SE

Loose
bolt
Southwest pier 1 Southeast pier 1

SW-NW SW-NE SE-NW SE-NE
N
SW-SW SW-SE SE-SW SE-SE
FIGURE 3.18 Anchor bolt identification. (Reprinted from ASNT Publication.)
Anchor bolt
strain guages
FIGURE 3.19 Anchor bolt strain gauge locations. (Reprinted from ASNT Publication.)
Partial Collapse: The structure is vulnerable to major deformation or discontinuities of a

span. (This would result in loss of service to traffic on or under the bridge.) This failure
may be the result of tipping or tilting of the substructure causing deformation in the super-
structure. A failure of this type may endanger the lives of some of those crossing or those
under the structure.
Structural Damage: The structure is vulnerable to localized failures. This failure may be the
result of excessive deformation or cracking in the primary superstructure or substructure
members of the bridge. A failure of this type maybe unnoticed by the travelling public but
would require repair once it is discovered. This type of damage would also make a bridge
more susceptible to overload failures.
Wardhana and Hadipriono (2003) showed the relationship between classes of failure that are
similar to the above classification of NYSDOT and the phasing (construction vs. in-service phases).
The number of failures (up to 2003) according to failure phasing is shown in Table 3.12.
TABLE 3.12
Number of Bridge Failures According to Failure Categories
Types of Failures Construction Service Unknown
Distress 0 17 0
Partial collapse 3 80 13
Total collapse 5 12 21
Unknown 0 277 75
Total 8 386 109
The distress category is fairly similar to the NYSDOT Structural Damage category.
Note that each of the above failure classifications/groups has two attributes: (a) The scale and
extent of the damage, and (b) The consequences of such failure. This makes those definitions to be
closer to a risk-based definition that concerns itself with vulnerabilities, threats, and consequences.
Such general and encompassing definition makes it possible to quantify failure. A general risk-
based approach would be
FR = T i V i C (3.4)

where
FR = Failure severity estimate
T = Threat/hazard level estimate
V = Vulnerability of structure to failure due to T
C = Consequences of failure
We recognize Equation 3.4 as the classical risk equation FEMA 452 (2010) and FEMA (2009).
We now try to make Equation (3.4) more specific for quantification of failure risk by defining
STV = T i V (3.5)
where
STV = Propensity to failure
Thus, a suitable failure severity estimate can be expressed as

FR = STV i C (3.6)

Assuming that the limits of STV and C to be
1 ʺ STV ʺ 10 (3.7)

1 ʺ C ʺ 10 (3.8)
The failure estimate should have limits as
1 ʺ FR ʺ 100 (3.9)

Assuming that the propensity to failure of a given bridge “A” is STV = 4.5, while the consequence
of this failure is STV = 7.5 the failure severity estimate is FR = 33.75. For a different bridge “B,” if
the propensity to failure is STV = 7.3, while the consequence of this failure is STV = 3.0 the failure
severity estimate is FR = 21.9. So, the severity of failure of bridge “A” is much more than the severity
of failure of bridge “B,” even though bridge “B” is more vulnerable to the postulated threats than
bridge “A.” Equations 3.4 through (3.9) can be used to quantify severity of failure across a number
of bridge networks.
3.4.2 Classification of Failure Causes

Failure causes have been classified by Wardhana and Hadipriono (2003) into enabling and triggering
causes. The differences between the two types of causes were discussed earlier; in general, enabling
causes occur over longer period of time, while triggering causes occur suddenly. We discuss different
enabling and triggering failure causes first. Next we try to associate those failure causes with Equations
3.4 through 3.9 for estimating f FR. We finally discuss the role of SHM in evaluating failure causes.
3.4.2.1 Triggering Causes

Triggering causes of failure include hydraulic/flood (scour, debris, drift, etc.), collision/impact,
overload (Figures 3.20 through 3.22), fire, ice, earthquakes (including tsunamis), wind (hurricanes/
tornadoes, etc.), and soil failures. Note that all of these causes occur suddenly, and last relatively
short period of time. Figures 3.23 and 3.24 show percentages of bridge failure rates according to
different triggering causes in the United States and New York State, respectively. Note that this
database is not a comprehensive database of all the failures but only the failures recorded by the
NYSDOT based on data collected from periodic survey of other states.
FIGURE 3.20 Impact failure.
FIGURE 3.21 Overload failure. (Courtesy of International Association of Structural Movers.)

FIGURE 3.22 Traffic overload. (Courtesy of International Association of Structural Movers.)
US bridge failures
Steel/deterioration, 46, 3% Collision, 232, 12% Collision

Concrete/deter., 10, 1% Concrete/deter.
Overload, 223, 13% Construction
Construction, 12, 1%
Misc. deterioration
Misc. deterioration, 63, 4% Earthquake
Nature, 32, 2%
Earthquake, 18, 1% Fire
Misc, 54, 3% Hydraulic
Fire, 50, 3%
Misc
Nature
Overload
Steel/deterioration
Hydraulic, 1021, 57%
FIGURE 3.23 Failure rates (US). (Courtesy of New York State Department of Transportation.)
New York State bridge

failures
Steel/deterioration, 7, 4%
Collision, 37, 19%
Overload, 33, 17% Collision
Concrete/deterioration
Construction
Concrete/deterioration, 4, 2% Misc. deterioration
Nature, 4, 2% Construction, 4, 2% Fire
Hydraulic
Misc, 11, 6% Misc. deterioration, 9, 5% Misc
Nature
Fire, 3, 2%
Overload
Steel/deterioration
Hydraulic, 82, 41%
FIGURE 3.24 Failure rates (New York State). (Courtesy of New York State Department of Transportation.)
3.4.2.2 Enabling Causes

Wardhana and Hadipriono (2003) subdivided enabling causes (they called it principal causes) as
design, detailing (Figures 3.25 and 3.26), construction, maintenance, material related (including
deterioration, normal wear and tear, fatigue, and/or corrosion). Of these, maintenance and material
related causes were the largest causes for collapse or distress. Note that all of these causes occur
over relatively long period of time. Figure 3.27 shows effects of construction material type on bridge
failures. Figure 3.28 shows the effects of bridge construction dates and Figure 3.29 shows bridge
type on failure rates.
3.4.2.3 Estimating Bridge Failure Propensity to Causes

We just presented a plethora of bridge failure causes. For a given bridge, many of those causes might
not be applicable; yet several others might have an effect on the bridge. We present a method for
evaluating total failure severity estimate when there are several potential causes of failure that we
define as FR_TOTAL . A generalized form of Equation 3.4 can be offered as
1/ n
⎛ i = NCAUSE n⎞ (3.10)
FR _ TOTAL = a ⎜ ∑ ( FRi ) ⎟
⎝ i =1 ⎠

where
FRi = Failure severity estimate for ith failure cause
NCAUSE = Number of pertinent failure causes
n = Suitable power number
α = Scaling factor
The form of Equation 3.10 was used in FEMA (2009) to sum the individual risks of uncorrelated
threats; it is reasonable to use to sum the severity of failures due to uncorrelated failure causes in our
current endeavor. Note that n = 2 reduces Equation 3.10 to the popular square root of sum of squares
(a)
(b)
FIGURE 3.25 Failure due to concrete details (a) overview of failure site and (b) failed box girder. (Courtesy
of New York State Department of Transportation.)
FIGURE 3.26 Failure due to steel details. (Courtesy of New York State Department of Transportation.)
65, 5%
233, 17%
Steel
Concrete
Timber/Wood
830, 59% Other
267, 19%
FIGURE 3.27 Failure of bridges according to material of construction. (Courtesy of New York State
US NYS
136 28 12
170 Pre-1990
37 1990–1994
1995–1999
221 2000–2005
104
1197 2005–Present
24
14
FIGURE 3.28 Failure of bridges according to construction date. (Courtesy of New York State Department
of Transportation.)
(SRSS) method. However, we propose to use a much higher value for n, say n = 10, which would
produce more realistic results, see FEMA (2009) for more discussion of this approach. Proceeding
as before
FRi = Ti i Vi i Ci
(3.11)
where
Ti = Threat/hazard level estimate ith failure cause
Vi = Vulnerability of structure to failure due to T ith failure cause
Ci = Consequences of failure of the ith failure cause
36 Arch
128
Steel beam/girder
Concrete beam/girder
369 Timber beam/girder
Covered
Culvert
443 Slab
Steel stringer
Suspension
106
Timber/wood stringer
34 Truss
32 31
8 84 21 Other
27
FIGURE 3.29 Failure of bridges according to bridge type. (Courtesy of New York State Department of
Transportation.)
Also, defining
STVi = Ti i Vi
(3.12)
where
STVi = Propensity to failure due to Ti
Thus, a suitable failure severity estimate can be expressed as
FRi = STVi i Ci
(3.13)
Assuming that the limits of STV and C to be
1 ʺ STVi ʺ 10
(3.14)
1 ʺ Ci ʺ 10
(3.15)
By adjusting the value of α, the total failure severity estimate would have limits as
1 ʺ FR _ TOTAL ʺ 100
(3.16)
The limits in Equations 3.14 through 3.16 are arbitrary. Careful studies should be conducted to
evaluate the effects of those limits, and provide realistic limits, for practical bridge situations.
As an example of estimating failure severity estimate for multiple causes, let us consider two
bridges, “A,” and “B,” with different estimated failure causes, propensity of failures, and conse-
quences of failure and shown in Table 3.13. In this example, the power number is assumed to be
n = 10. The number of pertinent hazards are NCAUSE = 6 and NCAUSE = 5 for bridges “A,” and “B,”
respectively. Bridge “A” does not have hydraulics as a realistic failure cause, since it does not cross a
water body. Bridge “B” is not located in a seismically prone area; it does not have wind as a poten-
tial cause of failure. The failure severity estimates of bridges “A,” and “B,” are 30.15 and 25.65,
respectively. Studying the numbers in Table 3.13, it is clear that the consequences of bridge “A”
failures caused these results even though bridge “A” seem to have lesser threats and vulnerabilities
than bridge “B.”
Equations 3.10 through 3.16 can be used to quantify severity of failure due to any number of
failure cause combinations across a number of bridge networks.
TABLE 3.13
Failure Severity Estimates for Two Bridges
Bridge “A” Bridge “B”
Cause STVi Ci Fri STVi Ci Fri
Construction 2 9 18 4 4 16
Hydraulic NA NA NA 6 5 30
Seismic 4 9 36 NA NA NA
Wind 2 9 18 NA NA NA
Collision/impact 2 7 14 6 3 18
Fatigue 6 4 24 6 3 18
Overload 3 5 15 7 3 21
Ncause 6 5
α 0.84 0.85
Scaled totals 30.15 25.65
TABLE 3.14
SHM Roles in Bridge Failure—General
Causes of Failure Sensing/Measurements STRID DMID
Enabling Design Use sensing to verify design Modal or parameter Using in-field damage
Detailing procedures for unusual identification methods information, such as
geometries that are not within can be used to enhance remaining fatigue life,
the bounds of conventional analytical bridge models, provide for an
codes such as bridges with especially for existing efficient and safe
large skew angles bridges retrofit bridge designs
Construction See Table 3.15
Maintenance NDT methods can compliment Different STRID methods Detecting damage in
conventional visual produce more accurate an accurate and
inspection practices analysis results for timely manner would
maintenance projects focus maintenance
efforts in safer
directions
Material related See Chapters 3 and 4(for concrete deterioration), Chapter 5 (for fatigue) and Chapters 6
and 7 (for fiber reinforced polymers) in Ettouney and Alampalli (2012)
Triggering External events See Table 3.15
3.4.2.4 SHM/SHCE Role

Structural health monitoring plays a major role in all aspects of bridge failure. Some examples of
the different roles each of the three components of SHM can play in enabling and triggering causes
of bridge failure are shown in Table 3.14 through 3.16.
3.4.3 Attributes of Failure

The failure event of a bridge can be used as an information source to help understand failure mecha-
nisms, causes, attributes, and so on. When bridge fails, it might be instructive to collect information
regarding the event as follows:
Bridge Type: The structural type, including superstructure, substructure, foundation, and
soil
Bridge Attributes: Span length, support conditions (before and after the failure)
TABLE 3.15
SHM Roles in Bridge Failure—Triggering Causes
Hazard Sensing/Measurements STRID DMID
Collision/accidents On bridge vibration monitoring Before event and after Global or local DMID
(Figure 3.30) can reveal occurrence and event STRID efforts can methods can aid in
extent of collision events give estimates of extent producing safe and
of global, and perhaps cost-effective retrofit efforts
local, damage
Concrete deterioration See Chapters 3 and 4 in Ettouney and Alampalli (2012)
Construction Sensing behavior of systems Modal identification DMID techniques can show
during construction projects methods can show extent of damages after
would provide an invaluable abnormal dynamic construction accidents
QA/QC insight that helps in behavior during
improving safety during construction. Thus
construction and potential providing additional
clarifications of future behavior safety measures to the
construction process
General deterioration SHM procedures for general deterioration are similar to those procedures in Chapters 3 and 4
in Ettouney and Alampalli (2012)
Fire (Figures 3.31 and After fire events, sensing would Modal identification NDT methods such as
3.32) produce valuable information methods can reveal acoustic emission (local
regarding extent of fire damage potential changes in damage) or thermography
(deflections, tilt, etc.) global bridge behavior (regional damage) can
after fire provide damage information
Hydraulic/scour/flood See Chapter 1 in Ettouney and Alampalli (2012)
Nature See Chapters 1 through 5 in Ettouney and Alampalli (2012)
Overload Capacity and demands can be STRID methods offer a Global or local DMID
correlated in real time by unique opportunity for methods can aid in
measuring bridge response improving modeling producing safe and
(capacity: strains, technique of overloaded cost-effective retrofit efforts
displacements, etc.) and live systems when a system is
load demands (truck loads, overloaded
volume and frequency)
Steel deterioration See Chapters 3, 4, and 5 in Ettouney and Alampalli (2012)
Seismic (Figure 3.33) See Chapter 2 in Ettouney and Alampalli (2012)
Fatigue See Chapter 5 in Ettouney and Alampalli (2012)
Material: Record the material of construction of different bridge components

Abutments and Piers (before and after the failure): Type, height, size, bearing types, foun-
dations, seat widths, and so on.
Age: Year Built, year failed, and if there were any indications of sudden deterioration or loss
of function
Failure Type: Describe the type of failure (as described throughout this section)
Failure Cause: The cause (or causes) of failure (as described throughout this section)
Consequences: This includes direct consequences such as number of Fatalities and/or num-
ber of injuries. Additional consequences might be of interest such as economic, social or
other local/regional consequences
Multimedia: Recording failure aftermath with photographs and/or video can provide valu-
able database. This is particularly important since failure sites are usually reconstructed
soon after the failure event
TABLE 3.16
SHM Role in Bridge Failure—Age of Bridge
Bridge Age Sensing/Measurements STRID DMID
Older Monitoring systems on older Can validate rehabilitation and Different DMID procedures can help
bridges can be costlier than their existing analytical models in detecting damages of older
counterparts on new systems bridges, thus extending their life span
Newer Newer bridges offer Can validate rehabilitation and Newer bridges are less exposed to
opportunities of placing existing analytical models. aging-type damages, such as
monitoring systems that are They can help in validating deterioration or increased traffic
more interconnected and and improving accuracy of demands. However, they can be
consistent with the bridge construction-related analytical vulnerable to abnormal hazards that
system than older bridges models exceed their design limits, such as an
abnormally high flood or abnormally
strong earthquake
FIGURE 3.30 Failure due to accidents. (Courtesy of New York State Department of Transportation.)
FIGURE 3.31 Fire failures: steel girders have deformed extensively. (Courtesy of New York State Department
of Transportation.)
FIGURE 3.32 Fire failure: deck damage. (Courtesy of New York State Department of Transportation.)
FIGURE 3.33 Seismic failure. (Courtesy of New York State Department of Transportation.)
3.4.4 Structural Modes of Failure

Bridges, as well as any other infrastructure can fail only in one of two temporal structural modes:
either brittle or ductile. Spatially, the structures can either fail progressively, or globally. We explore
those structural modes of failure next. At the end of this section, we examine the role and limita-
tions of SHM in each of those modes. Note that we differentiate between damage and failure in
this discussion. Damage would be a local event in spatial terms, while failure is more extensive in
spatial terms.
3.4.4.1 Ductile Failure

Ductile structural failure occurs over long period of time. Signs of distress are usually observable.
As such most conventional SHM techniques can be used to monitor the damages and alert for any
impeding failure. Examples of ductile failure are failures due to foundation settlements, or exces-
sive wind-induced deflections and vibrations. The failures of Tacoma Narrows Bridge (Figure 3.45)
and the Tay Rail Bridge (Figure 3.42) are ductile-type failures. The Tacoma Narrows collapse due
to excessive wind vibrations lasted long enough to prevent any human causality. Unfortunately, the
Tay Rail Bridge collapse occurred suddenly, thus causing great loss in life. We consider the Tay Rail
Bridge failure as a ductile failure since the signs of poor performance were observed long before
the collapse, such as rattling of steel connections, and excessive vibrations when trains crossed the
bridge.
3.4.4.2 Brittle Failure

Brittle structural failure occurs suddenly. Signs of distress are usually subtle, and difficult to
observe. As such care is needed when applying conventional SHM techniques. It is important to
clearly identify the brittle modes of failure that the SHM project aims to uncover, and ensure that
the techniques used are capable of identifying those brittle failure modes. Examples of brittle fail-
ure are failures due to corrosion (Figure 3.34) or fatigue (Figure 3.35) since both types of failure
occur suddenly, even though the damage occurred over a long time period. The failure of the I-35
Bridge in Minnesota (Figure 3.59) is clearly a brittle failure; it occurred suddenly and caused major
catastrophic consequences.
FIGURE 3.34 Corrosion failure. (Courtesy of New York State Department of Transportation.)
FIGURE 3.35 Fatigue failure. (Courtesy of New York State Department of Transportation.)
3.4.4.3 Global Failure

Brittle or ductile failure can lead to global bridge failure. Global failure occurs when the whole
structural system becomes unstable. Global instability is perhaps the most severe failure mode and
the most difficult to detect or monitor. When the bridge is newly constructed, its global stability
condition is usually investigated and ensured. As the bridge ages, this global stability condition
changes: the global stability safety factor starts decreasing. After a suddenly applied abnormal
loading condition, such as an earthquake event, the structural stability safety factor decreases even
further, due to the nonlinear changes in the system. Such nonlinearities would decrease global
stability resilience. Ettouney et al. (2006) explored the effects of such nonlinearities on the global
stability of structural systems. For highly redundant structural systems reductions of global stability
safety factors are slow. For low redundant systems, investigating global structural stability should
be performed often; for such investigations to be beneficial, they should be done using actual bridge
properties, not theoretical properties. An example of the use of SHM in observing stability condi-
tion of a steel truss is given in Chapter 8.
3.4.4.4 Progressive Failure (Collapse)

Progressive collapse condition occurs when the failure starts locally, then progress further away
from the initial location of the failure. Ettouney et al. (2004) presented a progressive collapse theory
that stated that progressive collapse event can lead to one of two outcomes: (a) if the failure front is
arrested, say by reaching a resilient support, or (b) if large portion of the structure have failed such
that the remaining structure becomes globally unstable. The latter scenario was discussed earlier.
We discuss now the first outcome, that is, the progression of failure front. Extent of progressive col-
lapse in bridges can be described in terms of four scales, as follows.
Limited Progression: The collapse progression is limited in scale. Examples are fatigue
cracks, local buckling, and limited corrosion
Local Progression: Collapse progressed to a single and or continuous spans; superstructure
only
Regional Progression: Collapse progressed to a single and or continuous spans; superstruc-
ture, as well as substructure
Global Progression: The collapse would propagate across many spans and over many piers
and supports
Figure 3.36 shows schematics of the failure scales. Clearly, SHM techniques can be of value in
monitoring and assessing progressive collapse at different collapse scales, as shown in Table 3.17.
3.4.4.5 Mitigation Strategies and Failure Modes

Obviously, optimal failure mitigation strategies should depend on the mode of failure. An optimal
strategy should aim at delaying or eliminating the failure under consideration at reasonable costs.
Quantitatively, this means that we should strive to reduce probability of failure. Since in this section
we are concerned with bridge (or system) failure, we recall immediately that an optimal failure miti-
gation strategy should aim at increasing the bridge system reliability. Note that for optimal strategy,
reduction of threat (corrosion prevention measures, for example) might offer a more cost-effective
solution than reducing vulnerability. Thus an optimal mitigation strategy might be defined as the
one that would reduce risk, provide for higher benefit to cost ratio, or produce the optimal life cycle
analysis (costs, benefits and life spans). Chapters 8, 9, and 10 in Ettouney and Alampalli (2012)
discuss those issues in greater detail. In what follows, we explore specific optimal mitigation needs
for different structural failure modes.
Ductile Failure: The goal of ductile failure mitigation should be to eliminate or reduce the chance of
such a failure. Inspection (Figure 3.38) or monitoring can help in such a situation. Regular maintenance
2 - Span-level failure: distress,

partial or total failure of single or
connected spans of
superstructure
1 - Limited failure: cracks,
corrosion, dents, etc.
Roadway
3 - Regional failure: super and

substructure, including
foundations for single or 4 - Global failure: failure extends
connected spans beyond single or connected spans
Scales of bridge failures
FIGURE 3.36 Scales of bridge failures.
TABLE 3.17
SHM Role in Different Progressive Collapse Scales
SHM Role
Failure
Scale Example Sensing STRID DMID
Limited Fatigue crack See Chapter 5 in Ettouney and Alampalli (2012)
Local Barge collision Before event: proximity Before and after event Global DMID methods,
sensors STRID model e.g., thermography or
During event: vibration evaluation can help in more localized DMID
signatures (time damage assessment, methods, e.g.,
histories of accurate retrofit ultrasonic or acoustic
accelerations, design, and decision- emission can aid in
velocities, or strains) making processes detecting presence,
After event: different type, location and
NDT procedures to extent of damage
detect damages
Regional Scour failure of See Chapter 1 in Ettouney and Alampalli (2012)
footing or underlying
soils (Figure 3.37)
Global Failure of single span Placing vibration Simulating progressive NDT methods can help
will propagate to sensors over the whole collapse of any in forensic analysis of
other spans by length of the bridge structural system in an failed structure
dynamics, or catenary (in an optimal manner accurate fashion is not
action as in Ettouney and an easy task. Using
Alampalli [2012]) can different STRID
help in accurate methods, coupled with
understanding of the optimal sensing data,
progressive collapse would result in an
event improved accuracy of
progressive collapse
modeling
FIGURE 3.37 Scour failure. (Courtesy of New York State Department of Transportation.)
FIGURE 3.38 Bridge inspection.
and rehabilitation efforts should be adequate in mitigating ductile failure. The bridge in Figure 3.39 is
being rehabilitated after a general deterioration in its conditions was observed over the course of several
years. Note that the rehab efforts are striving to preserve the historic nature of the bridge.
Brittle Failure: Brittle failure is much more difficult to mitigate. This is due to the difficulty in
continuously assessing the actual global stability condition. We immediately notice two contribut-
ing factors:
• Governing Redundancy Factor: It is essential to identify all pertinent potential failure

modes and the redundancies that control each of these modes. Traditionally, theory of
structures provided some help in assessing some structural redundancies. For example,
configurations of members in trusses (Figure 3.40) or the number of supports (Figure 3.41)
can indicate the degree of redundancy in a given system. Noting that such general assess-
ment rely on the implicit assumption that all structural components are in pristine condi-
tion. Sudden damages or normal wear and tear can alter this situation, reducing what might
FIGURE 3.39 Rehabilitation to avoid ductile failure. (Courtesy of New York State Department of
Transportation.)
Failure of this member can

cause global failure
No internal redundancy
Failure of this member will not

have catastrophic effects
Internally redundant truss
FIGURE 3.40 Internal redundancy.
have been a highly redundant system into a system with low redundancy that is susceptible
to brittle failures.
• Time Factor: As indicated earlier, passing of time can reduce the governing redundancy
of a bridge system. An optimal brittle failure mitigation strategy should account for the
as-built condition, not the designed condition.
An effective mitigation strategy should follow some or all of the following steps
1. Identify all governing redundancies in the system. This should be done using a well-
designed STRID effort
2. Identify and monitor actual conditions of items that control governing redundancies of #1.
A well integrated system of inspection and SHM is needed
3. Use the different SHM principles in this volume: duality, scaling and serendipity to help
identifying low redundancy that might not have been detected in #1 or #2. For example,
a subtle change in local resonance might indicate a reduction in a governing redundancy
Failure of this member can cause

global failure
No external redundancy
Failure of this member will not

have catastrophic effects
Externally redundant truss
FIGURE 3.41 External redundancy.
When embarking on brittle failure mitigation efforts, the decision maker needs to ask the following
questions:
• Can visual inspection alone detect brittle failure potential?

• Can sensing strains alone detect brittle failure potential?
• What kind of sensing can detect brittle failure potential?
The detailed answers to these questions can ensure an effective and optimal brittle failure mitigation
strategy.
Global Failure: Global failure mitigation strategies share same attributes to both ductile and
brittle failure mitigation strategies.
Progressive Failure (Collapse): First, we need to define progressive collapse mitigation strategy
(PCMS). We propose two types of strategies. An active PCMS would try to prevent local failures
that might cause progressive failures from happening in the first place. A passive PCMS would try
to prevent the scale of failure from increasing. Thus, passive PCMS implies accepting the fact that
initial failure might/can occur.
Passive PCMS include the use of two concepts: (a) improving redundancy of the structural sys-
tem, and/or (b) hardening different components of the structural system. Detailed and careful stud-
ies are needed to ensure that
• The PCMS is adequate to meeting the project objectives

• The PCMS does not have side effects that can reduce resiliency during other types of
threats (multihazards effects)
The active PCMS aims at reducing or eliminating local failures that can cause progressive col-
lapse. Different SHM techniques, such as NDT methods (Chapter 6), bridge security methods (see
Chapter 11 in Ettouney and Alampalli 2012) or bridge management methods (see Chapter 9 in
Ettouney and Alampalli 2012) can be used for optimal active PCMS.
Decision Making Role: Mitigation of bridge failure is complex and overreaching subject, as
evident from the above overview. There are numerous problems to resolve, and there are even more
choices for mitigation approaches. Issues such as hardening versus improving redundancy, govern-
ing redundancies, passive PCMS versus active PCMS provide overwhelming choices to the deci-
sion maker. Pairings of problems and solutions are not easy to find. We recommend a utilization of
different quantitative DM techniques in finding such pairings. Not only DM offer potential of using
optimal mitigation solution, it also can be automated (due to its quantitative nature) to any SHM,
or management practice. The cost saving and consistency of decisions are two obvious benefits of
using DM techniques in this field.
3.5 EXAMPLES OF BRIDGE FAILURES

3.5.1 Introduction
Competent personnel using well-established material and structural specifications design bridges.
In recent years, most bridge designs do employ quality control and assurance measures. Similarly,
they are also inspected routinely to identify issues that affect their structural safety to carry the
in-service loads as well as routine overloads imposed on them. Appropriate actions are taken based
on inspection results depending on the urgency, effect of the critical findings on structural safety
and durability, and available resources. But, most of these are reactive in nature and applies to slow
form of deterioration such as corrosion and age-related issues. But, bridges do face several hazards
and are designed to some of these hazards based on a certain return interval while not designed for
other hazards that the structure has a very low probability with low consequence of encountering
in its service life. Due to several uncertainties and due to the limitations associated with specifica-
tions, constructability, design/analysis difficulties and time, variability of materials used, and other
factors, designs are generally very conservative in nature.
Despite all the conservatism, failures do happen often and in some cases with loss of life and sig-
nificant economic impacts, due to several factors. These include design errors, unanticipated level
of loads or hazards, material issues, failure of quality control or assurance, limitations associated
with inspection, communication issues between different groups involved with bridge service life,
neglect. But, very few failures can be attributed to hazard level exceeding the designed hazard level.
In most cases, bridges do not fail due to a single factor but a combination/convergence of a number
of factors. Even though some of these can be detected through inspection process several such as
design errors, items concealed from general inspections, and severe problems that can develop rap-
idly before the next cycle are hard to detect once bridge is in service. This section discusses some of
the historic failures and documented reasons to show that the bridge failures in general are due to
convergence of several factors progressing for several years than a simple, sudden, single reason.
3.5.2 Tay Rail Bridge, UK

The 3.5-km-long Tay Rail Bridge spanned the Firth of Tay in Scotland, between the city of Dundee
and the suburb of Wormit in Fife, replacing an early train ferry. On December 28, 1879, the centre
section known as the “High Girders” (thru truss) collapsed killing 75 people when a train on it went
down with the bridge.
The Tay Rail Bridge was a lattice-grid (truss) design, combining cast and wrought iron and was
opened on June 1, 1878. The bridge had several issues early in the construction that was attributed
to original bedrock surveys. During the construction, it was found that the bedrock near the banks
was deeper than predicted earlier. This led to the redesign of piers and reduced number of piers
making superstructure spans longer that originally planned. The bridge was a combination of deck
truss (“Low Girders”) and thru truss sections.
The official forensic investigations noted that the bridge was “badly designed, badly built and
badly maintained, and that its downfall was due to inherent defects in the structure, which must
sooner or later have brought it down” (Court of Inquiry 1879). Thus several factors contributed to
the failure including the following: (a) Allowance for wind load had been made by the designer
Bouch (Seim 2008) as he was advised that this was unnecessary for girders shorter than 200 ft. The
designer apparently did not make any allowances when the revised design involved longer girders.
The middle section of the bridge was a thru truss to allow a higher clearance for the passage of ships
underneath and was potentially top heavy and very vulnerable to high winds; (b) The cylindrical
cast iron columns supporting the 13 longest spans of the bridge, each 75 m long, were of poor qual-
ity. Many had been cast horizontally, with the result that the walls were not of even thickness, and
there was some evidence that imperfect castings were disguised from the quality control inspec-
tions. In particular, some of the lugs used as attachment points for the wrought iron bracing bars had
been “burnt on” rather than cast with the columns. However, no evidence of the burnt-on lugs has
survived, and the normal lugs were very weak. They were tested for the Inquiry by David Kirkaldy
and were proved to break at only about 20 t rather than the expected load of 61 t. These lugs failed
and destabilized the entire centre of the bridge during the storm; (c) There was clear evidence that
the central structure had been deteriorating for months before the final accident. A few months after
the bridge was opened, the bridge inspector noticed that several joints had loosened making many
of the tie-bars useless for bracing the cast-iron piers; and (d) Recent research also indicates that the
cast iron used to join the columns of the bridge together might have become brittle under great strain
and might have contributed to the failure.
This bridge failure shows the value of preliminary site data collection, proper design, selection of
structural material, proper quality control (during the design, fabrication, construction, and inspec-
tion), and proper maintenance when deficiencies are noted during the bridge inspection. It is inter-
esting to that some of the same errors contributed to the failure of the recent collapse in Minnesota
that failed more than100 years after the failure of the Tay Rail Bridge (Figures 3.42 and 3.43).
3.5.3 Quebec Bridge Failure

One of the oldest failures of major bridges, killing 75 of the 86 people on the bridge, during the con-
struction can be the First Quebec Bridge in August of 1907. Failure of this bridge shows the value of
checking the design calculations after any significant change in structures design, quality control in
the design process, involvement of the design engineer during construction through field visits, and
supervision of a competent reengineer during the construction phase.
During 1907, the Quebec Bridge was supposed to be the bridge with the longest central span
as well as longest cantilever span once built. But, the southern half of the bridge that was near-
ing completion collapsed suddenly. Theodore Cooper, then famous engineer, directed the project.
The engineer soon after his appointment recommended increasing the span from 1600 ft to 1800
ft to reduce the vulnerability of the piers to scour by moving them to shallow water and to reduce
the project time by making it easy to construct. To offset the increase steel costs, he also modified
specifications that would allow for higher unit stresses. Because the St. Lawrence was a shipping
lane, the 2800-ft bridge needed almost 150 ft clearance above the water to allow the ocean-going
vessels to pass. Further, the bridge was to be multifunctional and was required to be 67-ft-wide to
accommodate two railway tracks, two streetcar tracks, and two roadways.
FIGURE 3.42 Tay Rail Bridge after construction.

FIGURE 3.43 Tay Rail Bridge after collapse.
In the rush to provide shop drawings for steel fabrication preliminary weight calculations pre-
pared in early stages were never properly checked when the design was finalized using revised
specification. The real bridge weight exceeded significantly than the preliminary estimated weights
and thus the bridge dead load capacity. Due to the inability of the project director to visit the field,
an engineer with less experience for the major bridge construction was in charge of the field site.
By the time, a material inspector reported the excess weights, south anchor arm, tower, and two
panels of the south cantilever arm were fabricated and six panels of the anchor arm were already
in place. The project director concluded that the increased stresses are within allowable limits and
proceeded with the construction of the bridge. When the bridge was nearing completion during the
1907 summer, the effect of increased dead loads were revealed in the form of distortions of lower
chord compression members that are key structural members. Signs of buckling were observed
in them and also in the splices between some lower chord members. The deflection of one of the
chords of the south anchor arm grew from 0.75 in to 2.25 in. But the work continued as it was felt
that there is no immediate danger and some attributed the bends presence before the installation.
By the time, project director was informed, a decision was made and conveyed to field personnel
to stop the construction, two compression chords of the anchor arm buckled and the subsequently
bridge collapsed in matter of seconds. For detailed discussions about the Quebec Bridge collapse,
see Lienhard (2008), Smith (2008), and Ricketts (2008) (Figure 3.44).
3.5.4 Tacoma Narrows, WA

In 1937, the Washington State legislature created the Washington State Toll Bridge Authority and
appropriated $5,000 to study the request by Tacoma and Pierce County for a bridge over the Narrows.
The first Tacoma Narrows Bridge, a cable-supported bridge, was opened to traffic on July 1, (See
Figures 3.45 and 3.46). It collapsed 4 months later on November 7, 1940, at 11:00 AM (Pacific time) due
to a physical phenomenon known as aeroelastic flutter caused by a 67 km per hour wind in a moderate
windstorm (Ketchum 2011c).
The bridge was the third longest suspension bridge when built with a center span of 2800 ft,
29-ft-wide, and had the greatest ratio of length to width ever built. The false bottom type of cais-
sons were used in sinking of cellular piers in water 120-ft-deep. Total height of west and east piers
was 198, and 247 ft, respectively. Shore anchorages for suspension cables each contained 2500
cubic yards of concrete weighing 52, 500 tons. Two steel towers, 425 ft in height and each weighing
The 640-foot central span of the mighty Quebec bridge in place:

Undismayed by tragic failures in 1907 an 1910, the builders have at last
achieved one of the greatest triumphs in the history of bridge building
The central span, weighing over 5000 tons, being raised by practically the
same method as was employed last year, with certain added precautions
Ready for the hoist of 150 feet: The span is about to be lifted from its
pontoons by means of hydraulic jacks and chains, composed of massive bars
of steel. The raising occupied practically four days
FIGURE 3.44 Collapsed Quebec Bridge: sequence of events that led to collapse.
FIGURE 3.45 Completed bridge, before opening. (Courtesy of University of Washington Libraries, Specail
Collections.)
FIGURE 3.46 Opening of bridge for traffic. (Courtesy of University of Washington Libraries, Specail
Collections.)
1927 tons, surmount the piers and support the suspension cables. The two suspension cables, 17.5
in diameter and more than 1 mile long, each consists of 6308 #6 parallel wires formed into 19
individual strands of 332 wires each. This forms a total of 14,000 miles of wire weighing 3817 tons
(Washington University 2008a and 2008b).
The Tacoma experience taught engineers that wind causes not only static loads on the bridge,
but also significant dynamic actions. It is commonly presented as an example of failure due to reso-
nance when fundamental frequency of the bridge coincides with the external periodic frequency of
the load. A cable-supported bridge is subject to wind-induced drag (the static component), flutter
(the instability that occurred at Tacoma Narrows), and buffeting (where gusts “shake” the bridge).
Adequate aerodynamic performance is required with respect to each of these effects. For modest
span bridges, drag generally controls the strength required to resist wind. Flutter becomes critical
when the wind acting on the structure reaches a critical velocity that triggers a self-excited unstable
condition. The task in design is to assure that the critical wind velocity is high enough so that it
has a very low probability of occurrence. This can be achieved by providing a stiff structure and/
or an aerodynamically streamlined superstructure shape. Buffeting influences fatigue of the bridge
materials as well as users’ comfort. The magnitude of buffeting response under higher probability
wind conditions must be controlled. Addressing these issues in an engineering context requires the
use of wind tunnel models. Current practice is converging on use of such models for the aerody-
namic properties of the bridge shape only. The mechanical properties of the bridge, and the final
wind evaluation, are performed using computer models that incorporate the wind tunnel results
(Ketchum 2008). Figure 3.47 illustrates the sequence of events that led to the bridge collapse.
(a)
Some consequences of the failure of the
Tacoma Narrows Bridge
Demands on bridges:
Bridge geometry:
importance of wind
streamline bridge profiles
demands
Mechanical properties: Lab testing:

increase bridge stiffness, Needs for 3-D wind tunnel
especially torsional stiffness testing of bridges
Analysis / design:
Aerodynamic demands:
limits of ‘deflection theory’
Need to model aerodynamic
(vertical deflection as the
stability
only design requirement)
(b) Chain of events that culminated in the failure of Tacoma Narrows bridge
November 7, 1940, around 10AM Pacific time zone
1-Slip of cable band near
mid-span causes the cable to
separate into two unequal parts. 2-The resulting twist of the deck
causes slight rotation (twist)
motion.
Lower pressure at vortex
produces suction force
3-With slight twist of deck,
vortex shedding creates even
more twisting demands on the
deck.
Bridge cross
section
4-As the bridge twists even

more, the twisting motion
5-The twisting motion becomes in phase with the
eventually became self-induced vortex. Torsion flutter results.
(output energy is more than
dissipating energy). Thus an ever
increasing amplitude results that
culminates in an ultimate bridge
failure.
FIGURE 3.47 Sequence of bridge collapse (a) interrelationship between geometry and increased wind
demands and (b) wind flutter.
3.5.5 Silver Bridge, OH

The Silver-Point Pleasant Bridge (known in history as Silver Bridge) collapsed on December 15,
1967 evening killing 46 people and injuring 9 when 31 of the 37 vehicles on the bridge plunged into
water. Since the bridge on the U.S. Highway 35 connects Point Pleasant, WV and Kanauga, OH
over the Ohio River, a major transportation route connecting both the States was destroyed caus-
ing a major disruption to many and caused enormous sensation across the nation on the safety of
bridges.
The Silver Bridge got its name as it was the country’s first aluminum painted bridge. The 2235-ft-
long Silver Bridge was a two-lane eye-bar suspension type bridge designed according to American
Society of Civil Engineers specifications and constructed in 1928 by the General Corporation and
the American Bridge Company. It was designed with a 22-ft roadway and one 5-ft sidewalk with
“High Tension” eye-bar chains, a unique anchorage system, and rocker towers. At the time of its
construction, the maximum permitted truck gross weight was about 20,000 lb compared to the large
truck limit of 60,000 lb or more in 1967. See Figure 3.48.
The Silver Bridge was the first eye-bar suspension bridge of its type to be constructed in the
United States. The bridge’s eye bars were linked together in pairs like a chain. A huge pin passed
through the eye and linked each piece to the next. Each chain link consisted of a pair of 2 × 12 in
bars and was connected by an 11 in pin. The length of each chain varied depending upon its location
on the bridge. Such bridges had usually been constructed from redundant bar links, using rows of
four to six bars, sometimes using several such chains in parallel. The eye bars in the Silver Bridge
were not redundant as links were composed of only two bars each of high strength steel (more than
twice as strong as common mild steel), rather than a thick stack of thinner bars of modest material
strength “combed’ together as is usual for redundancy. With only two bars, the failure of one could
impose excessive loading on the second, causing total failure—unlikely if more bars are used.
While a low-redundancy chain can be engineered to the design requirements, the safety is com-
pletely dependent upon correct, high quality manufacturing, and assembly.
“Rocker’ towers were used that was also a unique feature of the Silver Bridge. About 131-ft-high
towers allowed the bridge to move due to shifting loads and changes in the chain lengths due to
temperature variations. A curved fitting was placed next to a flat one at the bottom of the piers. The
rocker was then fitted with dowel rods to keep the structure from shifting horizontally. With this type
of connection, the piers were not fixed to the bases. These allow the bridge to respond to various live
loads by a slight tipping of the supporting towers that were parted at the deck level, rather than pass-
ing the suspension chain over a lubricated or tipping saddle or by stressing the towers in bending.
Thus the towers required the chain on both sides for their support, so failure of any one link on either
side, in any of the three chain spans would result in the complete failure of the entire bridge.
The forensic investigations concluded that the heat-treated carbon steel eye-bar (#330, on the
north of the Ohio subsidiary chain, the first link below the top of the Ohio tower) broke due to a
small crack that formed through fretting wear at the bearing. Over the years, partly due to increased
Point of failure initiation
Ohio Tower W.VA Tower
360’ 0” 700’ 0” 360’ 0”
Elevation
FIGURE 3.48 Silver Bridge collapse—before collapse.

Change in weight demands from Silver Bridge over the years

70.00
60.00
50.00
Weight (Kips)
40.00
1928
When the bridge was built
30.00 1967
When the bridge collapsed
20.00
10.00
0.00
Average weight Maximum allowed
of family car (Kips) vehicle gross weight (Kips)
FIGURE 3.49 Changes in demands for Silver Bridge.
loads (Figure 3.49) when compared to design loads, stress corrosion, and corrosion fatigue allowed
the crack to grow, and when became critical, the member broke in a brittle fashion. This placed
undue stress on other members of the bridge. The remaining steel frame buckled and fell due to the
newly concentrated stresses. The entire structure collapsed in the span of a minute.
The visual inspections performed cannot detect such cracks formed in the eye bar without dis-
assembling the eye bar. These defects are still hard to detect even in present days. This failure
prompted President Johnson to initiate a taskforce charged to determine procedures available to pre-
clude future disasters and implement changes, if needed. This led to the establishment of National
Bridge Inspection Standards by the Federal Government and current routine bridge inspections
with qualified inspection personnel. Details of the bridge and its failure can be found at National
Transportation Safety Board (NTSB) (1971) and LeRose (2001). Figure 3.50 illustrate some of the
factors that led to collapse. Figure 3.51 shows the aftermath of the collapse.
3.5.6 Mianus River Bridge Collapse

Mianus River Bridge carried Interstate 95 over the Mianus River near Greenwich, CT. A 100-ft
suspended eastbound span between piers 20 and 21 collapsed on June 28, 1983, killing three and
seriously injuring three others when their vehicles fell into the river with the bridge. The collapse was
attributed to the failure of the pin and hanger connection that was commonly used decades ago as this
offered easier analysis and low construction costs. For more details, see Nationmaster.com (2011d).
The suspended span that collapsed was attached to the bridge structure at each of its four corners.
A pin and hanger assembly attached each corner to the girders of the cantilever arm of an adjacent
anchor span to support the weight of the northeast and southeast corners of the suspended span, The
pin and hanger assembly includes an upper pin attached through the 2 1/2-in-thick web of the girder
of the cantilever arm and a lower pin attached through the 2 1/2-in-thick web of the girder of the
suspended span. One and one half-inch-thick steel hangers connect the upper and lower pins-one
on the inside and one on the outside of the web (NTSB 1984). Figure 3.52 shows a pin and hanger
system that is similar to the one used in Mianus River Bridge.
Visual inspection
The location of the flaw were not accessible visual to inspection NTSB explained the reasons for the
Silver Bridge failure
Testing
There were no adequate testing methods for the flaw without disassembly of the eye
bar
NTSB then proceeded to
Material degradation recommend pertinent needed
research as an outcome of the
Stress corrosion cracking: when tensile stresses are applied to ductile materials in failure
a corrosive environment it can suddenly crack (fail) in what is called stress
corrosion cracking.
a. Identify bridge building materials susceptible
Corrosion fatigue: results from a combination of two issues: alternating stress to slow flaw growth by any of the suspected
cycles and corrosive environments. As can be seen from the figure, corrosive mechanisms;
environment can have a degrading effect on maximum fatigue stresses b. Determine critical flaw size under various
stress levels in bridge building materials;
Comparison between air-based and c. Develop inspection equipment capable of
corrosive environment-based fatigue detecting critical or near critical flaws in
standing bridge structures;
Metal in air
d. Devise analytical procedures to identify
(conventional fatigue)
critical locations in bridge structures which
require detailed inspection;
e. Develop standards which incorporate
Fatigue stress limit
appropriate safeguards in the design and

fabrication of future bridges to ensure
Faigue stress limit Metal in water protection against failures of material such as
(metal in air) occurred in the Point Pleasant Bridge;
f. Develop standards for the qualification of
materials for future bridge structures, using the
Metal in corrosive information disclosed in this investigation;
environment (such as g. Devise techniques for repair, protection, or
sodium chloride salvage of bridges damaged by internal flaws;
solution) and
Number of stress cycles h. E xpand the knowledge of loading history and
life expectancy of bridges.
FIGURE 3.50 Silver Bridge collapse—reasons for collapse.
FIGURE 3.51 Silver Bridge collapse—after collapse. (With permission from National Highway Institute.)
Sometime before the collapse of the suspended span, the inside hanger in the southeast corner
of the span came off of the inside end of the lower pin. This action shifted the entire weight of the
southeast comer of the span onto the outside hanger. The outside hanger gradually worked its way
farther outward on the pin, and over a period of time, a fatigue crack developed in the top outside
end of the upper pin. The shoulder of the pin fractured off, the pin and hanger assembly failed, and
the span collapsed into the river. The NTSB determined that the probable cause of the collapse of
the Mianus River Bridge span was the undetected lateral displacement of the hangers of the pin and
FIGURE 3.52 Typical pin and hanger construction.
FIGURE 3.53 Failed span. (With permission from National Highway Institute.)
hanger suspension assembly in the southeast corner of the span by corrosion-induced forces due to
deficiencies in the State of Connecticut’s bridge safety inspection and bridge maintenance program
(NTSB 1984). Figure 3.53 shows the failed bridge. Figure 3.54 illustrates the sequence of failure.
The investigations cited corrosion from water build up due to inadequate drainage as a cause. The
highway drains had been deliberately blocked during road mending some 10 years before and water
leaked down through the pin bearings, causing them to rust. The outer bearings were safety-critical
and nonredundant, a design flaw of this particular type of structure. The bearings were difficult to
inspect close up, although traces of rust could be seen near the affected bearings.
This highlights the need for thorough inspection and maintenance program, use of nondestruc-
tive procedures when needed, and the value of redundancy.
3.5.7 Schoharie Bridge Collapse

On April 5, 1987, two spans of the five-span Schoharie Creek Bridge collapsed causing five vehicles
falling into the river with 10 fatalities. A third span collapsed 90 minutes later. The bridge was
(a)
4-The relative
deflections between
the pins are 3D
1-The hanger
moves laterally
2-The hanger moves

3-The resulting state vertically
of strains are 3D
(b) Pin Anatomy of failure of a pin and hanger system

Hanger
1-As the Pin-Hanger

assembly failed...
(c)
2-The nonredundant span

collapsed instantly.
FIGURE 3.54 Failure sequence of a pin and hanger system (a) lateral and longitudinal expansion of hangers,
(b) as-designed pin-hanger system, and (c) failure of bridge due to lack of system redundancy.
built in 1954 and owned by the New York State Thruway Authority. It was located in Mohawk val-
ley northwest of Albany, NY. The 540-ft-long and 112-ft-wide bridge carried four lanes of traffic,
median area, and shoulders over Schoharie creek. The average height of the bridge above the creek
was about 80 ft. The bridge was designed to 1949 edition of AASHTO Standard Specifications for
Highway Bridges. It was a five-span steel superstructure with reinforced concrete deck. The piers
consists of two slightly tapered reinforced concrete columns connected by a tie beam near column
tops and supported on a common pedestal and footing. Two of the piers were in the Creek with other
two piers on the creek banks to support the structure along with two end abutments.
The bridge collapsed on April 5, 1987 during the spring flood caused by a rainfall totaling
150 mm combined with snowmelt to produce an estimated 50 year flood (WJE Associates 1987).
The collapse was initiated by the toppling of pier three, which caused the progressive collapse of
spans three and four into the flooded creek. Pier two and span two fell 90 minutes after span three
dropped, and pier one and span one shifted 2 hours after that (Thornton-Tomasetti 1987). The NTSB
suggested that pier two collapsed because the wreckage of pier three and the two spans partially
blocked the river, redirecting the water to pier two and increasing the stream velocity (NTSB 1988).
According to investigations, scouring under piers began shortly after the bridge was built in
1955 as the bridge footings experienced floodwater flows unanticipated in the design of the bridge,
a 100-year flood. It is believed that the majority of the scouring energy was dissipated into moving
the original riprap layer from around the footings. Once the backfill had been exposed, the years of
peak flows removed the backfill material, and the backfill material in turn was replaced by sediment
settling into the scoured. Furthermore, the riprap placed at construction had probably been washed
away during the 1955 flood, and had not been replaced (WJE Associates 1987).
The investigations concluded that the collapse of the Schoharie Creek Bridge was due to the exten-
sive scour under pier three that was affected by four important factors (Thornton-Tomasetti 1987):
(a) the depth of shallow footings, bearing on soil, was not enough to take them below the probable
limit of scour; (b) the foundation of pier three was bearing on erodable soil that allowed high velocity
floodwaters to penetrate the bearing stratum; (c) the as-built footing excavations and backfill could
not resist scour; and (d) riprap protection, inspection and maintenance were inadequate.
Besides scour issues, several factors including two common practices when the bridge was designed
in the 1950s contributed to the severity of the collapse (Thornton-Tomasetti 1987). These were the
bridge bearings allowed the spans to lift or slide off of the concrete piers and the simple spans were not
redundant. Thus, the use of continuous spans, rather than simple spans, would have provided redun-
dancy once pier three failed, and perhaps allowed for the redistribution of forces between the spans.
The failure also shows that it is important for bridge owners to identify the critical features that can
lead to the collapse of a bridge and to ensure that those critical features are inspected frequently and
adequately (NTSB 1988). Although the Schoharie Creek Bridge had been inspected annually or bien-
nially since 1968, an underwater inspection of the piers footings had never been performed. The bridge
was scheduled for an underwater inspection in 1987, but the bridge collapsed before the inspection took
place (NTSB 1988). The failures also emphasize how important it is to design footings deep enough to
avoid loss of support capacity due to scour (Shephard and Frost 1995) (Figures 3.55 and 3.56).
3.5.8 I-35 Bridge, Minneapolis, MN

On August 1, 2007, evening the eight lane, 1907-ft-long I‑35W highway bridge over the Mississippi
River in Minneapolis, Minnesota, experienced a catastrophic failure in the main span of the deck
truss killing 13 and injuring 145 persons. One thousand feet of the deck truss collapsed, with about
456 ft of the main span falling 108 ft into the 15-ft-deep river.
The I-35W Bridge was located about 1 mile northeast of the junction of I-35W with Interstate
94. In addition to spanning the Mississippi River, the bridge also extended across Minnesota
Commercial Railway railroad tracks and three roadways: West River Parkway, 2nd Street, and the
access road to the lock and dam. The bridge was opened to traffic in 1967 with 141,000 ADT in
FIGURE 3.55 Collapsed bridge, from http://ny.water.usgs.gov/projects/scour/text.html

FIGURE 3.56 Scour hole, from http://ny.water.usgs.gov/projects/scour/fig5.html
FIGURE 3.57 I-35 Bridge spanning the Mississippi river. (NTSB 2008.)
2004. The 14-span structure used welded built-up steel beams for girders and truss members, with
riveted and bolted connections. The bridge was 1907-ft-long and carried eight lanes of traffic, four
northbound and four southbound. The 1064-ft-long deck truss portion of the bridge encompassed a
portion of span 5; all of spans 6, 7, and 8; and a portion of span 9. The deck truss was supported by
four piers (see Figure 3.57).
Riveted steel gusset plates at each of the 112 nodes (connection points) of the two main trusses
tied the ends of the truss members to one another and to the rest of the structure. The gusset plates
were riveted to the side plates of the box members and to the flanges of the H members. All nodes
had at least two gusset plates, one on either side of the connection point. A typical I-35W main truss
node, with gusset plates, is shown in Figure 3.58.
The NTSB report (NTSB 2008) attributed the probable cause of the collapse of the bridge
(Figure 3.59) to the inadequate load capacity due to a design error of the gusset plates at the U10
nodes (see Figure 3.60), which failed under a combination of substantial increases in the weight
of the bridge, which resulted from previous bridge modifications, and the traffic and concentrated
construction loads on the bridge on the day of the collapse. Contributing to the design error was
the failure of quality control procedures by the design company to ensure that the appropriate main
truss gusset plate calculations were performed for the I-35W Bridge and the inadequate design
review by Federal and State transportation officials. Contributing to the accident was the generally
accepted practice among Federal and State transportation officials of giving inadequate attention to
FIGURE 3.58 Typical five-member node (two upper chord members, one vertical member, and two diagonal
members) on I-35W Bridge. (NTSB 2008).
FIGURE 3.59 Collapsed bridge. (NTSB 2008).
U10W
L8'W
L13'W L13W L11W

L9W
L8W L8W
Pier 7
West - Pier 6 - East
FIGURE 3.60 Locations of components. (NTSB 2008.)

(b)
(a)
Number of built bridges in 1000s (5 years intervals) Number of built bridges in 1000s (5 years intervals)
0
5
10
15
20
25
30
0
100
200
300
400
500
600
700
To 1900 To 1900
1900–1905 1900–1905
1905–1910 1905–1910
1910–1915 1910–1915
1915–1920 1915–1920
1920–1925 1920–1925
1925–1930 1925–1930
1930–1935 1930–1935
Lives and Deaths of Infrastructure
1935–1940 1935–1940
1940–1945 1940–1945
1945–1950 1945–1950
material, and (b) total number of bridges built.

1950–1955 1950–1955
Year built
Year built
1955–1960 1955–1960
1960–1965 1960–1965
1965–1970 1965–1970
Total number of bridges built
1970–1975 1970–1975
1975–1980 1975–1980
1980–1985 1980–1985
1985–1990 1985–1990
1990–1995 1990–1995
Number of bridges built-sorted by construction material
1995–2000 1995–2000
2000–2005 2000–2005
Steel
Wood
Concrete
Cumulative total
Prestressed concrete
Total built per period
FIGURE 3.61 Count of bridges in the United States: (a) Number of bridges built - sorted by construction
129
TABLE 3.18
Count of Bridge Inventory in the United States
Year Built Concrete Steel Prestressed Concrete Wood
Before 1900 1738 4196 642 731
1901–1905 283 825 126 34
1906–1910 882 2129 222 69
1911–1915 1556 1597 161 47
1916–1920 3248 2899 381 182
1921–1925 5564 2395 403 152
1926–1930 12,121 6529 852 716
1931–1935 11,874 9255 752 1650
1936–1940 13,883 11,566 657 1994
1941–1945 4463 2949 198 898
1946–1950 11,824 9509 418 2121
1951–1955 15,185 11,268 728 2255
1956–1960 24,584 16,684 5278 3090
1961–1965 24,345 18,621 9176 2453
1966–1970 22,017 18,002 11,709 2309
1971–1975 15,333 14,304 12,513 1631
1976–1980 13,801 10,990 12,195 1701
1981–1985 14,186 9321 13,510 1286
1986–1990 15,409 10,742 15,685 1589
1991–1995 14,265 9611 15,580 1534
1996–2000 14,500 8845 15,178 870
2001–2007 14,545 9275 15,977 568
gusset plates during inspections for conditions of distortion, such as bowing, and of excluding gusset
plates in load rating analyses.
3.6 APPENDIX I: COUNT OF BRIDGES IN THE

UNITED STATES BY CONSTRUCTION MATERIAL
The number of highway bridges in the United States is a massive number indeed as shown in
Figure 3.61 and Table 3.18. The primary materials used in bridge construction are concrete, pre-
stressed concrete, steel, and wood, as shown in Figure 3.61a. The utilization of wood as a bridge
construction material peaked in the mid-twentieth century. Around the same time, the utilization of
pre-stressed concrete increased. At present, the use of steel, concrete, and pre-stressed concrete dom-
inate bridge construction. Figure 3.61b indicates the total bridges built in 5-year intervals. As of the
writing of this volume, the total number of bridges in the United States is over 600,000. Note that the
numbers in Figure 3.61b does not account for the decommissioned, failed, or demolished bridges.
REFERENCES
Alampalli, S. and McCowan, P., (2008) “Addressing Critical Findings from Highway Bridge Inspection: New
York State Procedures,” NDE/NDT for Highways and Bridges: Structural Materials Technology (SMT),
ASNT, Oakland, CA.
Court of Inquiry, (1879), “Tay Bridge Disaster,” Report Of The Court of Inquiry, and Report Of Mr. Rothery,
“Upon the Circumstances Attending the Fall of a Portion of the Tay Bridge on the 28th December 1879”
place this as the second reference”. The Railways Archive. Retrieved on 2007-02-05.
Ettouney, M., and Alampalli, S., (2012) Infrastructure Health in Civil Engineering: Applications and
Ettouney, M., Smilowtz, R., Tang, M., and Hapij, A., (2006) “Global System Considerations for Progressive
Collapse with Extensions to Other Natural and Man-Made Hazard,” Journal of Performance of
Constructed Facilities, 20(1), 403–417.
FHWA., (2007) “National Bridge Inventory, NBI,” Federal Highway Administration, Washington, DC.
FEMA., (2009) Handbook for Rapid Visual Screening of Buildings to Evaluate Terrorism Risk, Federal
Emergency Management Agency, FEMA Publication, Washington, DC.
FEMA., (2010) A How-To Guide to Mitigate Potential Terrorist Attacks, Federal Emergency Management
Agency, FEMA Publication, Washington, DC.
Garg, N., Larkin, A., and Ricalde, L., (2008) “Evaluation of the Application Potential of PSPA in determination
of Modulus of Concrete HMA pavements at the FAA’s National Airport Pavement Test Facility,” NDE/
NDT for Highways and Bridges: Structural Materials Technology (SMT), ASNT, Oakland, CA.
Hansen, W. and Surlaker, S., (2006) Embedded Wireless Temperature Monitoring Systems For Concrete Quality
Control, University of Michigan, An Arbor, MI.
Ketchum, M., (2011c) “Tacoma Narrows” http://www.ketchum.org/wind.html [accessed February 20, 2011].
Kosnik, D. and Hopwood, T., (2008) “Monitoring of In-Situ Strains in Bearing Assembly Anchor Bolts on a
Large Through - Truss Bridge,” NDE/NDT for Highways and Bridges: Structural Materials Technology
(SMT), ASNT, Oakland, CA.
LeRose, C., (2001) The Collapse of the Silver Bridge, West Virginia Historical Society Quarterly Vol. XV, No. 4.
Lienhard, J., (2008) “Quebec Bridge Collapse,” http://www.uh.edu/engines/epi220.htm [accessed February 20,
2011].
National Transportation Safety Board (NTSB)., (1988) “Collapse of New York Thruway (1–90) Bridge over
the Schoharie Creek, near Amsterdam, New York, April 5, 1987.” Highway Accident Report: NTSB/
HAR-88/02, Washington, D.C.
Nationmaster.com (2011d) “Mianus River Bridge,” http://www.nationmaster.com/encyclopedia/Mianus-
River-Bridge [accessed February 20, 2011].
Ni, Y., Xia, Y, Liao, W., and Ko, J., (2008) “Technology Innovation in Developing the Structural Health
Monitoring System for Guangzhou New TV Tower,” International Symposium on Structural Control and
Health Monitoring, Taichung, Taiwan.
NTSB., (1971) “Mianus River Bridge Collapse” http://www.ntsb.gov/Publictn/1971/HAR7101.htm, accessed
on September 2, 2008. [accessed February 20, 2011].
NTSB., (1984) “Silver Bridge Collapse” report date 1984 accessed on 5 – 15 – 2008, http://www.ntsb.gov/
publictn/1984/HAR8403.htm [accessed February 20, 2011].
NTSB., (2008) “Highway Accident Report, Collapse of I-35W Highway Bridge Minneapolis, Minnesota,
August 1, 2007,” National Transportation Safety Board, Washington, D.C., NTSB/HAR-08/03, PB2008–
916203, Notation 7975C.
NYSDOT., (2008) SFY 2006–07 Annual Report of Bridge Management and Inspection Programs, New York
State Department of Transportation, Albany, NY.
Ricketts, B., (2008) “The Collapse of the Quebec City Bridge,” accessed from the web on 2008-02-05, http://
www.mysteriesofcanada.com/Quebec/quebec_bridge_collapse.htm [accessed February 20, 2011].
Seim, C., (2008) “Why Bridge Have Failed Throughout History,” Civil Engineering Vol 78 (5): 64–71, 84–87,
http://www.ASCE.org [accessed February 20, 2011].
Shepherd, R. and Frost, J. D., (1995) Failures in Civil Engineering: Structural, Foundation and Geoenvironmental
Case Studies. American Society of Civil Engineers, New York.
Smith, C., (2008) “A Disaster in the Making,” accessed from the web on 2008-02-05, http://www.civeng.
carleton.ca/ECL/reports/ECL270/Prelude.html [accessed February 20, 2011].
Thornton-Tomasetti, P. C., (1987) “Overview Report Investigation of the New York State Thruway Schoharie
Creek Bridge Collapse.” Prepared for New York State Disaster Preparedness Commission, December.
Wardhana, K. and Hadipriono, F., (2003) “Analysis of Recent Bridge Failures in the United States,” ASCE J. of
Performance of Constructed Facilities, Vol 17, No. 3.
Washington University., (2008a) Data sheet on the 1940 Tacoma Narrows Bridge, 1940, http://content.lib.
washington.edu/cdm4/item_viewer.php?CISOROOT=/farquharson&CISOPTR=108&CISOBOX=1&
REC=1
WJE, Associates, Inc., and Mueser Rutledge Consulting Engineers., (1987) “Collapse of Thruway Bridge at
Schoharie Creek,” Final Report, Prepared for: New York State Thruway Authority, November.
4 Theories and
Principles of IHCE
4.1 INTRODUCTION
Conventional experiments in bridge engineering have been mostly of the nondestructive testing
(NDT) type. NDT experiments are mostly local, both in time and space. They are very adapting in
revealing accurate picture of the condition of a local part of the bridge at specific time. Due to such
limited scope of NDT, the value (or benefit-to-cost) of NDT experiments were not a major issue,
since the objectives of the NDT experiments are usually well defined, and the tools of executing
such an experiment are well suited for obtaining the objectives of the experiments. The structural
health monitoring (SHM) paradigm, which purport a bridge monitoring experimentation that is
wider in both locale and time than the conventional NDT experimentation, have been gaining
interest lately. The overall system of SHM/SHCE/IHCE is shown schematically in Figure 4.1.
(Note that SHM is used in this chapter [and in most of this volume] in lieu of SHCE/IHCE without
loss of generality.) SHM projects and experiments are demanding, as shown in Figure 4.2. Due to
the larger space and time range of SHM experimentation, the optimum experiment parameters,
such as number of sensors, type of sensor mix, type of data acquisition, labor costs, and the value
of additional information from the experiment, needs to be carefully studied. Currently, there are
no quantifying methods that help in designing an SHM experiment that studies and optimizes the
overall value of such an experiment.
This chapter introduces several theoretical SHM concepts. We discuss the concepts of cost and
value in SHM as metrics for optimizing different projects. We then introduce a general theory of
experimentation (GTE). The general theory addresses the optimum value of SHM experiment. The
theory lays the groundwork for quantifying the value any SHM experiment, thus providing the deci-
sion maker a subjective tool to justify the SHM project a priori, with clearly quantifiable value of
the results. A special theory of experimentation (STE) is also introduced that concentrates on the
narrower subject of sensors within the SHM field. This special theory addresses the type, number,
and location of sensors in an overall SHM project. It also includes the cost of material, labor, and
the overall expected value of using the sensor topology. The purpose is to optimize such a value,
given the direct and indirect costs. Finally, a theory of SHM triangulation is offered. The theory
addresses the all important concept of sensor, or detection, types. It lays down the theoretical basis
for using different detection schemes during SHM projects. Some applications in the SHM field will
be demonstrated.
The chapter then presents several principles of SHM, namely: the duality, scaling, and serendip-
ity principles. The three principles can help in achieving efficient and cost-effective executions of
SHM projects.
It is believed that the theories and the principles of this chapter will help in providing
qualitative means to enumerate the values of SHM projects. This should, in turn, help in
promoting SHM projects, thus fulfilling the SHM goals of improving safety while reducing
costs.
133
SHM project
Scaling of information Triangulation of

and resources detection schemes
Other Resources:
Data pool
• Bridge
management
systems
• LTBP
• Other Decision making tool box
Optimum decision
Overall organization of SHM system
FIGURE 4.1 Organization of SHM/SHCE System.
Anatomy of an experiment
Physical phenomena
(displacements, strains, tempratures, etc.)
Snesors/
Technology
transducers
Digitizers
(analogue to
digitals)
Graphs/
Signal Cabling/
printouts/
proccessing networking
monitors
Data storage
Algorithms
FIGURE 4.2 Anatomy of SHM, or general, experiments.

Theories and Principles of IHCE 135
4.2 GENERAL THEORY OF EXPERIMENTATION (GTE)

4.2.1 Introduction
We discussed the importance of SHM and the SHCE in general for preserving and enhancing struc-
tural health in the previous chapter. In embarking on an SHM project, one of the most important
concerns that the decision makers face is: how valuable such a project is? We already have seen that
the value of any activity related to costs of such an activity and the benefits that such an activity
would bring. We try in this section to formalize the value versus cost of experiments. Figures 4.3
and 4.4 show the different parameters that might be needed for quantifying cost and value of experi-
ments. We introduce a general theory for experimentation that would help in such formalization. We
then provide two practical examples to show how the theory can be used to ensure optimum value
for the experiment on hand.
Before introducing the theory, we would like to discuss its governing parameters first. In
several other locations of this volume, we identified value as a relationship between costs and
benefits. For simplicity, we opted to relate the value directly with costs in the theory below. By
doing this, we couple benefits and value in a single expression. There are several advantages
for doing that: (a) it simplifies the theory and its applications, (b) it allows for a qualitative
treatment of the concept of value in a nondimensional manner, (c) it avoid the intricacies and
complications of relating value to benefits and costs, and (d) it allows for incorporating non-
quantifiable benefits of experiments in estimating the value (such benefits include, for example,
added redundancies by using additional sensors, or educational benefits). Of course, the con-
cept of benefit can be incorporated to the application of the theory if desired, without any loss
of generality or accuracy.
Parameters affecting cost
Laying and
Sensors Instrumentation connecting
sensors
Direct cost Direct cost Direct cost
Labor cost Labor cost Labor cost
Data acquisition Data analysis Miscellaneous
Direct cost Direct cost Direct cost
Labor cost Indirect cost Indirect cost
FIGURE 4.3 Parameters affecting cost and value

Parameters affecting cost

and value
Type of Type of Type of

Experiment Structure measurment
Static Material
Reinforced Displacements
Dynamic
Concrete
Steel
Strains
Other
T
emprature
Component
Single span
Other
Multi-span Girders
Other Deck
Pavement
Other
FIGURE 4.4 Parameters affecting cost and value.
4.2.2 Overview of GTE

The GTE can be stated as follows:
There exists an intrinsic value for every experiment. The value, as a function of cost, posses a critical
point; beyond this point, any additional expenditure is not cost-effective.
we discuss several aspects of the theory next:

Analytically, the theory can be enumerated by first defining a value for the SHE as
V = V (C ) (4.1)
where
C = Cost of the experiment
Define the critical value point, Vcr such that
Vcr = V (Ccr ) (4.2)

where
Vcr = Critical value of the experiment
Ccr = Critical cost of thhe experiment
According to the theory, when the cost of the experiment C > Ccr it would be logical not to expend
such additional cost since any additional cost beyond and above the critical cost would produce
either a diminishing, or no additional, value.
In addition to the critical point, (Ccr ,Vcr), in the C − V space, it is also beneficial to define an ulti-
mate point, (Cult , Vult), such that
Vult = Ultimate value of the experiment

Cult = Ultimate cost of the experiment
The ultimate value of an experiment Vult is the value beyond which there can be no added value to
an experiment, in spite of any additional cost. The ultimate cost of an experiment Cult is the cost that
corresponds to Vult , such that
Vult = V (Cult ) (4.3)

Formally
dV
=0 (4.4)
dC C = Cult
4.2.3 Proof of GTE

4.2.3.1 General
For the function V = V(C), a small incremental added value of the experiment ∆V can be expressed as
d V (C )
ΔV = ΔC (4.5)
dC
The term dV(C)/dC is the rate of value versus cost, it can be defined as
d V (C ) (4.6)
RV =
dC
Consider now the parameters that affect V as
a1 , a2 , a3 , aNE
where
ai = i th parameter of the experiment
NE = Total number of parametters that affect the experiment
Moreover, the cost C can be subdivided into two parts:
C = CV + CF (4.7)
CV = Costs that are dependent on the parameters ai, as such, they are the costs that directly affect
the value
CF = Costs that are independent on the parameters ai, as such, they are the costs that do not affect
the value
In general, CF represents the fixed costs of the experiment. These costs are the fixed costs that
do not change much for a given experiment. As such, these costs are considered fixed costs. Fixed
costs of an experiment can be data acquisition systems, labor for traffic protection, software for data
collection and analysis, test vehicles, and so on. In most cases, changing these costs does not result
in any meaningful changes in the resultant value of the experiment.
On the other hand, CV represents variable costs, that is, the costs that can affect the value of the
experiment. Examples of these variable costs are costs of sensors, instrumentation, and data acqui-
sition and analysis.
As such, variable costs can be expressed as
CV = CV ( a1 , a2 , a3 , aNE ) (4.8)
The parameters ai represent any/all of the important discrete parameters that affect the value of a
particular experiment. For example, these parameters can represent individual sensors, components
of instrumentation system, data acquisition system(s), and so on. Equation 4.8 can be simplified,
without any loss of generality, to
i = NE
CV = ∑ C (a )
i =1
Vi i (4.9)
Similarly, the value can be expressed as

i = NE
V= ∑ V (a )
i =1
i i (4.10)
From Equations 4.6, 4.9, and 4.10, the rate of value can be expressed as

⎡ ∂ V (a ) / ∂ a ⎤
i = NE
i i i
RV = ∑ ⎢⎢ ∂ C ( a ) / ∂ a ⎥⎥ (4.11)
i =1 ⎣ i i i ⎦
The terms ∂Vi(ai)/∂ai and ∂Ci(ai)/∂ai represents the rates of change of value Vi and cost Ci as
the parameter ai changes, respectively. It is reasonable to conclude that as ai increases, the
magnitude of ∂Vi(ai)/∂ai will eventually decreases. Ultimately, there exists a point beyond which
there would be no added value to the experiment, no matter how much ai is increased. This can
be expressed as
∂ Vi ( ai )
=0 (4.12)
∂ ai ai →∞
On the other hand, it is reasonable to assume that as ai increases, the cost CVI(ai) will always increase:
it will never reach a value of 0. As such
∂ CVi ( ai )
> 0 (4.13)
∂ ai ai →∞
From Equations 4.11, 4.12, and 4.13 it can be stated that
RV → 0, (4.14)
as the cost C ≈ Cult. In other words, at a given level of cost expenditures Cult there can be no addi-
tional value to the experiment.
At the lower end of the magnitude of the parameters ai, it is reasonable to assume that
RV 0 (4.15)
This means that at a lower level of cost, there is a finite magnitude of RV; there is an appreciable
added value to the experiment for additional costs.
Equations 4.14 and 4.15 provide the limits of RV. It indicates that the maximum rate of value
of any given experiment is between very low and high cost levels: there is an ultimate cost point
beyond which there is no additional added value.
We have proved so far the existence of an ultimate value and cost point for every experiment.
How about the critical value and its associated critical cost?
We need to recall that the critical cost of experiment is the cost point beyond which any addi-
tional cost will result in a diminishing, or no, additional value. We can then establish the following
relationships
Vcr ʺ Vult (4.16)
and
Ccr ʺ Cult (4.17)
In addition, we propose the following relationship
Vcr = a GTE Vult (4.18)
From Equations 4.16 and 4.18 it is clear that α GTE ≤ 1.0. Also, it is logical that 0 < α GTE.
Since the ultimate value of an experiment exists, based on Equation 4.18, we can conclude that
the critical value and its associated costs, do exist. This completes the proof of GTE.
4.2.3.2 Corollary # 1: Types of Experiments and α GTE
We differentiated between Vcr and Vult for a very practical reason. Such a reason becomes evident by
categorizing experiments into two types:
1. Type I experiments, where a finite Cult < ∞ and Vult < ∞ exist, or
2. Type II experiments, where only an infinite Cult = ∞ and Vult = ∞ exist. In this type of
experiments there is always an added value for additional costs, albeit with an increasingly
diminishing rate of return.
Figures 4.5 through 4.7 show the two different types of experiments. Note that for type I:
Ccr = Cult (4.19)

Vcr = Vult (4.20)
Thus, for type I
a GTE = 1.0 (4.21)
For Type II experiments, since both Cult and Vult exist only at ∞, the critical point (Ccr , Vcr) is sub-
jectively located on the C − V based on the shape of the curve, usually at a point when there is a
sharp downturn of the magnitude of Rv as shown in Figure 4.7. For practical reasons, Vcr is always
less than Vult.
Critical value, Vcr

Ultimate value, Vult
Rate of value, RV
Value, V
Critical cost, Ccr

Ultimate value, Vult
Cost, C Cost, C
FIGURE 4.5 Type I experiment.
Critical value, Vcr

Rate of value, RV
Value, V
Critical cost, Ccr
Cost, C Cost, C
FIGURE 4.6 Type I experiment.
Critical value, Vcr

Rate of value, RV
Value, V
Critical cost, Ccr
Cost, C Cost, C
FIGURE 4.7 Type II experiment.
This leaves a final question: what is the magnitude of αGTE for type II experiments? Unfortunately,
there is no possible quantitative answer for this question, beyond the knowledge of the limits
0 < αGTE ≤ 1.0. Nor there is any body of knowledge as of the writing of this volume that can help
in answering such a question. We propose that a detailed study of value-cost (V-C) relationship of
experiments are performed, and a reasonable choice of αGTE, and (Ccr , Vcr) be done based on such a
study.
4.2.3.3 Corollary # 2: Initial Conditions

In the C − V space of an experiment, there can be situations where the experiment needs an initial
outlay of funding, before any value is realized. This can be expressed as
V (C ) 0 ≤ C ≤ C =0 (4.22)
min
Here, Cmin is the minimum outlay of funding (cost) that needs to be spent before any value can be
realized. Figure 4.8 shows the C − V space for this situation. Generally speaking, most experiments
would have a finite magnitude of Cmin.
In practice, a reasonable estimate of Cmin must be established before embarking on any experi-
ment. The decision maker must then compare Cmin with the available funding FA and decide if
the executing the experiment is still viable option. Two examples of this process are shown in
Figure 4.8.
4.2.4 Practical Applications

4.2.4.1 General
The existence of an optimal configuration of experiments was discussed in an indirect manner by
Hyland and Fry (2000). They presented an outline of a scheme to optimize the performance of
SHM systems. Their approach constituted two parts: local and global parts. The local part deals
with sensor performance in detecting damage. They recommended an Artificial Neural Network
(ANN) to train the sensors to detect damages. The global part of the scheme is a genetic algorithm
for continued evolution of the overall SHM system. This includes continued evaluation of sensor
Critical value, Vcr
= FA
Value V
Minimum cost, Cmin
Critical cost, Ccr
FA < Cmin Cost C FA >> Cmin

Do not execute the experiment Experiments is feasible economically
Rate of Value RV
Cost C
FIGURE 4.8 Effects of initial costs on experiment.

performance, and making decisions as to ways to improve such performance. The type, number, and
placement of sensors are always evaluated and improved by the genetic algorithm.
The recognition of the existence of optimal sensor types, number, and placement by Hyland and
Fry is, indirectly, supportive of our general and special theories of experimentations. The special
theory deals specifically with the optimal number of sensors for each experiment. This optimal
condition can be found, as the theory imply, by any suitable method, including, but not limited to,
ANN. Also, our GTE indicates that there is an optimal setting for each experiment. Such optimality
is controlled by the objective(s) of the SHM project. Again, the methods of studying the optimal con-
ditions should be done on a case by case basis. Genetic technique is one of the many approaches that
can be used to study and achieve such optimality. In addition, our GTE includes both the technical
performance of the experiment as well as the cost of such performance (value of experiment). The
genetic approach of Hyland and Fry (2000) stresses mostly on technical performance of experiments.
In realistic settings, cost implications must be an integral part of any optimal solutions of an SHM
project.
4.2.4.2 Identifying Goals
In identifying goals of experiment several issues must be considered. The technical goal is fore-
most of those goals to be identified. For example: is the purpose to monitor corrosion progres-
sion, fatigue effects, or load capacity? Is the goal to diagnose a problem that was observed? (For
example: an excessive vibration). Is the goal to validate a particular design or particular analytical
techniques? The length of the experiment (short term or long term) must be decided next, depend-
ing on the goal of the experiment. Length of experiments can be a major factor in evaluating Vcr
and Ccr .
4.2.4.3 Identifying Algorithms
After identifying the goals of the experiment, the algorithms that would be used to achieve these
goals are chosen next. This document includes numerous examples on how to choose appropriate
algorithms to achieve specific goals. Some examples are given in Table 4.1.
4.2.4.4 Needed Type of Measurements (TOM)
Selection of appropriate algorithms is a logical step before choosing the measurements. This will
ensure efficient operations. It is the algorithm that drives the measurements, not the other way
around. Table 4.2 shows some examples of needed TOMs.
4.2.4.5 Values of Parameters
After assigning all pertinent parameters to the experiment, a value should be assigned to each
parameter. The values can be either monetary, or dimensionless. The dimensionless values can
have an arbitrary scale, say from 1 to 10, or 1 to 100. The only obvious condition in choosing the
values is that they need to be consistent for all the parameters. Relative magnitudes of values of
TABLE 4.1
Algorithm Examples
Goal Algorithm
Fatigue at a given point Rain flow algorithm remaining life, etc. (see Chapter 9 of Ettouney and Alampalli 2012)
Model verification Finite elements code, optimization schemes and tolerances, neural networks (see Chapter 6
of this volume)
Modal testing and Modal identification method based on type of excitation source and pertinent analytical
excessive vibrations method (see Chapter 6 of this volume)
Load rating Diagnostic testing, influence lines (see Chapter 8 of Ettouney and Alampalli 2012, etc. and
analysis type)
TABLE 4.2
Needed Measurements
Algorithm Needed Measurements
Remaining life (Fatigue) Strains as a function of time
Model verification and Time histories of displacements or velocities at given points
modal testing
Influence lines Displacements. For dynamic influence lines, need displacements as a function of time
Load rating Depending on the type of load rating analysis: displacements, strains, weights, etc.
each parameter should reflect the relative importance of that parameter to achieving the objectives
of the experiment. The values of the two examples given later in this section are dimensionless, with
a scale from 0.0 to 1.0. Of course, any null value indicates that the parameter has no useful role in
the experiment.
4.2.4.6 Parameter Cost Evaluation

Costs of each parameter should also be assigned next. As discussed earlier, there are two types
of costs: fixed and variable costs. Fixed costs are those costs that are needed for each experiment,
yet they have no effect on the value of the experiment. They play no role in determining Vcr or Vult.
Variable costs need to be estimated for each parameter.
Similar to values, costs can be either monetary or dimensionless. The dimensionless costs can
have arbitrary scales; say from 1 to 10, or 1 to 100. Again, the only condition in choosing parameter
costs is that they need to be consistent for all the parameters. Relative magnitudes of costs of each
parameter should reflect the relative cost of that parameter. The costs of the two examples given
later in this section are dimensionless. Of course, any null cost indicates that the parameter has no
effect on Vcr or Vult; no-cost parameters should be considered as fixed costs.
4.2.4.7 Value–Cost (V–C) Diagram

After assigning values and costs for each parameter two vectors, {V} and {C}, containing the values
and costs, respectively, can be populated. The size of the vectors is NE. Before they can be used
logically, they should be sorted in an ascending order using the value as the controlling index. After
sorting, the cumulative cost and value vectors should be prepared. Using these cumulative results,
the cumulative value-cumulative cost curve can be charted, and this will produce the required
C − V space.
4.2.4.8 Critical Magnitudes: Vcr and Ccr

The C − V space, either type I or type II, as shown in Figures 4.5, 4.6, or 4.7, can be used to deter-
mine Vcr, Ccr. If the C − V space is of type I, it should be a straightforward determination. If the
C − V space is of type II, then a logical decision regarding the value of α GTE is needed, as described
above.
4.2.4.9 Summary of the Steps

The theory can be applied to almost any SHM/SHCE situation. The benefits of applying this theory
are to
1. Get the optimum value for any experiment

2. Gain more understanding to the different issues pertaining to any SHM/SHCE situation
3. For complex/large SHM/SHCE project, applying the theory can shed light to shortcomings
of the proposed project
The practical steps of applying the theory are as follows.
1. Identify the goal of the SHM

2. Identify the algorithms to be used in reaching the above goal
3. Identify the needed measurements for the algorithm
4. From 2 and 3, identify all parameters ai
5. Identify all costs for measuring ai. These costs include both variable and fixed costs
6. Assign relative values for all ai in a descending order, starting from the most essential
parameter down to the least essential parameter
7. Assign costs to all variable costs (for each parameter ai) and fixed costs
8. Evaluate the nondimensional costs for each of the parameters ai
9. From 6 and 8, construct the (V-C) diagram
10. From 9, evaluate Vcr and Ccr
4.2.5 Examples of General Theory of SHM

4.2.5.1 Frequency Analysis
4.2.5.1.1 The Problem
Let us consider a monitoring system for a simple-span bridge/beam that aims at evaluating seismic
vulnerability of the bridge. These natural frequencies might be utilized in a later evaluation of pos-
sible damages in the bridge, or in validating a finite element model for the bridge.
Since only the natural frequencies of the simple-span beam are of interest, a possible algorithm
for the experiment can be one of the several modal identification algorithms of Chapter 6. Any of
those algorithms requires a time history recording of the vertical displacement of the bridge U(t). It
is implied that only vertical modes of vibration are of interest in this example. On the basis of this it
is reasonable to place the sensor at the center of the bridge. Theoretically, only one sensor is needed
to achieve the goal of the experiment. It should be noted that the measured center displacements
would only generate symmetric modes of vibration. Generally speaking, for such a system, sym-
metric modes may be the most important modes in evaluating the behavior of the bridge. Of lesser
importance are the asymmetric modes. If these modes are needed, then placing a second sensor at
quarter span is needed. In addition, sensors might be added for redundancy purposes.
4.2.5.1.2 The Test Setup

The experimental setup is to utilize NE number of vibration detection sensors. These sensors will
be mounted on the center line of the beam. A list of the parameters ai that control the value of this
experiment is shown in Table 4.3. A normalized (dimensionless) costs as well as values are used for
the example. Note that the maximum value is obtained from placing the first sensor. The additional
sensors would add lesser value to the experiment. Also note that the cost of cabling and data acquisi-
tion is independent of the number of sensors. For simplicity, we only included the variable costs of
the project in this example.
4.2.5.1.3 Discussion of Results

We use the information of Table 4.3 to produce a C − V plot, as shown in Table 4.4 and Figure 4.9.
Note that there is no value to the experiment until the cost of the first sensor is spent (C = 5). After
the value of the experiment jumps from 0 to 3, the additional costs of additional sensors produce
lower value.
It is clear that this experiment is type II experiment. The question is what is Vcr and the associ-
ated Ccr. The answer is mostly qualitative. It depends on available funding, quality of sensors, and
experience of personnel. It appears that the most prudent cost-value choice is 7 and 3.6, respectively.
Any additional cost beyond 7 seems to offer negligible additional value.
TABLE 4.3
Variable Costs and Value of Experiment
Parameters Dimensionless Value and Cost
Type Parameter # i Description CVi (ai) Vi (ai)
Variable cost 1 Sensor # 1, midspan 1 1
parameters 1 Cabling 2 1
1 Data acquisition system 2 1
2 Sensor # 2, quarter span left 1 0.3
3 Sensor # 3, quarter span right 1 0.3
4 Sensor # 4, midspan 0.9 .1
5 Sensor # 5, quarter span left .9 .05
6 Sensor # 6, quarter span right .9 .05
TABLE 4.4
Cost-Value Relationship of
Frequency Testing
Dimensionless Dimensionless
I Cost Value
0 0 0
1 5 3
2 6 3.3
3 7 3.6
4 7.9 3.7
5 8.8 3.75
6 9.7 3.8
Value -Cost for frequency analysis experiment

4
3.5
3
Dimensionless value
2.5
1.5
0.5
0
0 2 4 6 8 10 12
Dimensionless cost
FIGURE 4.9 Value–cost relationship for frequency analysis experiment.

4.2.5.1.4 Concluding Remarks

This simple example can be expanded and utilized for the popular SHM vibration testing of bridges.
First, it encourages the testers to study the objectives of the test, i.e., how many resonant frequencies are
needed and what frequency range. This example also shows that problem definition is probably the most
important step, both for the test itself as well as for evaluating the value of the test as a function of costs
(applying GTE). If the objective of the test is just the evaluation of the first mode, one sensor (such as a
strain gage or accelerometer reacting fast enough) is probably enough, as long as it is kept anywhere on
the beam except at supports. At the same time, if the sensor is kept at the center, there is a better chance
of getting good data due to better signal to noise ratio as modal magnitude should be highest there. If
the first two modes are needed, still one sensor may be enough, although it cannot be placed at the mid-
span of the beam. So, number of sensors will become a combination of needed number of frequencies,
frequency range, etc. If mode shapes are needed, then there is a need for more sensors, measurement
locations, or a combination of two to achieve the goal of the test depending on how many nodal points
one will encounter. In other words, the aforementioned simple example problem can be used to resolve
many of the decisions that are needed during vibration testing, and ensure a cost-effective test.
4.2.5.2 Corrosion FRP Mitigation Efficiency
Consider now the Congress Street Bridge experiment (New York State Department of Transportation
[NYSDOT]) that is described in Chapter 7 of Ettouney and Alampalli (2012). The experiment was
to monitor the long-term effects of corrosion mitigation of columns by utilizing an fiber reinforced
polymers FRP wrapping techniques. Let us explore a similar project. Figure 4.10 shows schematics
of the bridge geometry. Let us apply the theory of experimentation of this chapter to identify the
critical cost-value point.
The goal of the experiment is to investigate the efficiency of using FRP layer to control and slow
corrosion rate for the bridge columns. To accomplish this, all fixed costs of the experiment are esti-
mated. This includes agency costs, management costs, as well as data acquisition and analysis costs.
Due to the small number of sensors in this experiment, it is assumed that data acquisition costs are
independent of number of sensors; hence they were assumed to be fixed costs. As was mentioned
earlier, fixed costs of an experiment are spent without any gained value to the experiment in the C-V
charts. For our example, all costs and values will be dimensionless, as was described earlier.
In designing the experiment, each column station is assumed to have three locations on three sides
of the column. Each location will have three types of sensors: corrosion rate sensor, temperature
sensor and humidity sensor. Another sensor will be added at the FRP layer to make sure appropriate
bond exists between the FRP and the concrete it is attached to. All the sensors in a given column sta-
tion will be assigned same value. It is essential for this experiment to assess the correlation between
corrosion rate, temperature and humidity (humidity will indicate if moisture is present for corrosion
process). It should be noted that such a correlation might not be essential in other types of corrosion
experiments; in such a case, different values for each type of sensor might be assumed.
FIGURE 4.10 Illustration of Congress Street Bridge.

Generally speaking, corrosion monitoring situations for columns, no more than three column
stations are needed: near top, near center, and near bottom. For this experiment, the observed dam-
age of corrosion along the height of the column was judged to be near uniform. As such it was
decided not to monitor the center of the columns. The top station monitoring has more value than
the bottom station since the top of the column is near the source of leaking joints. However, the cost
of using the column top as monitoring station is several times more than the cost of monitoring of
the bottom stations. The sensor installation costs as well as the monitoring costs for such elevated
station contribute to such higher costs.
The bridge as shown in Figure 4.10 is built on a series of frames. Each frame has three columns
(North, Center, and South). The relative values of sensors at different columns were assigned based
on the following assumptions
• Assign higher value for the north column

• Next is the value for the center column. The center columns, by contrast, can add a base-
line measure
• Since the south columns are similar to the north columns, they were assigned the lowest
value. It is assumed that the south column information is fairly redundant when the north
column information is available
Given the framing configuration of Figure 4.10, two types of frames can be identified: end frames
and middle frames. It is noted that the columns of the end frames are more susceptible to corrosion
damage, since they are adjacent to the joints, which are the primary source of leaking that would
lead to corrosion. As such, it is decided that the value of monitoring the columns in an end frame
is higher than the value of monitoring the columns in an intermediate frame. Costs of monitor-
ing intermediate frames are slightly less than the cost involved with monitoring end frames. It is
assumed that the management will start by installing sensors at end frames (due to its higher value
for the experiment). Adding sensors to an intermediate frame should cost less; some of the cost can
be shared with the cost of installing sensors of the end frame.
On the basis of the above discussions of relative values and estimated relative costs, Table 4.5
was developed. It shows the parameters of the experiment and the aggregated dimensionless costs
and values assigned to each parameter. Note that there can be several components that can have
identical parameter number. For such components, it is assumed that they are all needed to work
together, that is, they must be treated as a parameter group. As such there were fifty parameters
that were identified for this situation. These parameters were consolidated to only twelve parameter
groups.
Table 4.6 shows the consolidations of Table 4.5. Finally, Figure 4.11 shows the corresponding
C − V curve and C −(dV/dC) curves. The curves show clearly that the critical point (Ccr , Vcr) of this
experiment is achieved with instrumenting only one station at the bottom three columns at the end
frame, total dimensionless cost of 43.1 and dimensionless value of 18.8. All these data are for illus-
tration purposes only. Value analysis is one of the tools in a decision-maker’s tool chest and should
be used wisely.
4.3 SPECIAL THEORY OF EXPERIMENTATION (STE)

4.3.1 Introduction
GTE showed a process of optimizing the value of any experiment. It explored the overall parameters
that affect an experiment. Many of the arrangements of sensors were presumed to be fixed during
the execution of GTE. In this section, we explore the optimality of sensor arrangements that includes
location, number, and type of sensors. The problem is well known in the NDT community. For
example, Annis (2005) posed the important question: what is the optimum number of specimens in
TABLE 4.5
Variable Costs and Values of Experiment
Parameter Parameter Total Total
Column ID #i group # i Description Cost Value Cost Value
1 1 Management, OH, etc. 10 0 19 4
2 Data acquisition System 4 0
N1 - Bottom 3 CP, HP, and TP 1 1
4 CP, HP, and TP 1 1
6 FRP layer 2 1
C1 - Bottom 7 2 CP, HP, and TP 1 0.95 24 7.8
8 CP, HP, and TP 1 0.95
9 CP, HP, and TP 1 0.95
10 FRP layer 2 0.95
S1 - Bottom 11 3 CP, HP, and TP 1 0.9 29 11.4
12 CP, HP, and TP 1 0.9
13 CP, HP, and TP 1 0.9
14 FRP layer 2 0.9
N2 - Bottom 15 4 CP, HP, and TP 0.9 0.7 33.7 14.2
16 CP, HP, and TP 0.9 0.7
17 CP, HP, and TP 0.9 0.7
18 FRP layer 2 0.7
C2 - Bottom 19 5 CP, HP, and TP 0.9 0.6 38.4 16.6
20 CP, HP, and TP 0.9 0.6
21 CP, HP, and TP 0.9 0.6
22 FRP layer 2 0.6
S2 - Bottom 23 6 CP, HP, and TP 0.9 0.55 43.1 18.8
24 CP, HP, and TP 0.9 0.55
25 CP, HP, and TP 0.9 0.55
26 FRP layer 2 0.55
N1 - Top 27 7 CP, HP, and TP 4 1 61.1 22.8
30 FRP layer 6 1
C1 - Top 31 8 CP, HP, and TP 4 0.95 79.1 26.6
32 CP, HP, and TP 4 0.95
33 CP, HP, and TP 4 0.95
34 FRP layer 6 0.95
S1 - Top 35 9 CP, HP, and TP 4 0.9 97.1 30.2
36 CP, HP, and TP 4 0.9
37 CP, HP, and TP 4 0.9
38 FRP layer 6 0.9
N2 - Top 39 10 CP, HP, and TP 4 0.75 115.1 33.2
40 CP, HP, and TP 4 0.75
41 CP, HP, and TP 4 0.75
42 FRP layer 6 0.75
TABLE 4.5 (continued)

Variable Costs and Values of Experiment
Parameter Parameter Total Total
Column ID #i group # i Description Cost Value Cost Value
C2 - Top 43 11 CP, HP, and TP 4 0.7 133.1 36
44 CP, HP, and TP 4 0.7
45 CP, HP, and TP 4 0.7
46 FRP layer 6 0.7
S2 - Top 47 12 CP, HP, and TP 4 0.65 151.1 38.6
48 CP, HP, and TP 4 0.65
49 CP, HP, and TP 4 0.65
50 FRP layer 6 0.65
CP = corrosion rate sensor, HP = humidity sensor, TP = temperature sensor, and FRP Layer = sensor at FRP layer. N = North,
S = South, and C = Center.
TABLE 4.6
Cost-Value Relationship of Corrosion Monitoring
Cumulative Cumulative
Parameter Cost Value Rate of Value
0 0 0 0
0 14 0 0
0 14 0 0.210526316
1 19 4 0.78
2 24 7.8 0.74
3 29 11.4 0.659793814
4 33.7 14.2 0.553191489
5 38.4 16.6 0.489361702
6 43.1 18.8 0.273127753
7 61.1 22.8 0.216666667
8 79.1 26.6 0.205555556
9 97.1 30.2 0.183333333
10 115.1 33.2 0.161111111
11 133.1 36 0.15
12 151.1 38.6 0.144444444
an NDT project? He showed that the answer to such question is to use as many samples as needed to
reduce the uncertainty of the result to an acceptable level. We will now generalize the question in two
ways: (a) From the local confines of NDT to the more global applications of SHM, and (b) We will
consider, in addition to sensor numbers, both sensor locations and sensor types.
So, we ask the question: what is the optimum arrangement (number, location, and type) of sensors
in an SHM project? To find a simple answer to such a complex question, we introduce a STE. The exis-
tence of an optimal sensor arrangement of any experiment (NDT and/or SHM) would encourage pro-
fessionals to attempt finding such an optimal arrangement in an objective fashion (see Figure 4.12).
We note that in addition to answering such an important question, this volume offers methods
and steps for obtaining such an optimal sensor arrangement (see Chapter 5).
Cost value and cost value rate curves for corrosion FRP mitigation of
column
45 0.9
40 0.8
35 0.7
30 0.6
Cummulative value
Rate of value
25 Rate of value 0.5
Value
20 0.4
15 0.3
10 0.2
5 0.1
0 0
0 20 40 60 80 100 120 140 160
Dimensionless cost
FIGURE 4.11 Cost-value and rate of value of corrosion monitoring.
Sensor arrangement
Sensor types Sensor numbers Sensor locations
Components of sensor arrangements

By optimizing the three components, an optimal sensor arrangement would result
FIGURE 4.12 Components of sensor arrangements.
4.3.2 Overview of STE

A STE can be stated as follows:
There exists an optimum arrangement of sensors for any experiment. The sensor arrangement includes
type, number, and location of sensors.
The STE states that the optimal arrangement of sensors in any experiment depends on three fac-
tors: sensor types, number, and locations of sensors. We first need to define a cost function that is to
be optimized. Similar to the GTE, the sensor arrangement need to minimize costs, and maximize
benefits; that is, the maximize value. It seems reasonable then to express STE formally as
Given the function f( )
V = f ( N S , LS , TS ) (4.23)
V = The value of the sensor arrangement

NS = Measure for number of sensors
LS = Measure for location of sensors
TS = Measure for type of detection schemes (type of sensors)
There is an optimal magnitude for V that depends on the measures NS, LS, and TS.
We can prove the theory by first making the reasonable simplifying assumptions
• Depending on the goals of the experiment, there is an optimal NS

• For a given NS, there is an optimal LS (optimal sensor location, OSL), and
• There is an optimal arrangement of detection schemes TS
On the basis of the above assumptions, then it is reasonable to state
(
VOPT = f N S OPT
, LS OPT
, TS OPT ) (4.24)
VOPT = Optimal value of the sensor arrangement

NSOPT = Optimal measure for number of sensors
LSOPT =Optimal measure for location of sensors
TSOPT =Optimal measure for type of detection schemes (type of sensors)
Equation 4.24 proves of the existence of optimal V. But, this requires proof of the above men-
tioned three assumptions. Next section discusses these in more detail.
4.3.3 Proof of STE

4.3.3.1 Optimality of Sensor Type
Using different detection schemes optimally can result in an VOPT. This can be done by triangulat-
ing the detection schemes. A theory of triangulation that proves this concept is introduced later in
this chapter.
4.3.3.2 Optimality of Sensor Number

We prove the optimality concept of NS, that is, the existence of NSOPT by introducing a sensor
demand and sensor capacity regions. We then optimize the interrelationship of these two concepts
and show that such an optimization would produce NSOPT.
Demand Region: Let us define the demand region DR of an experiment as that region of the struc-
ture that the professional need to obtain information about. For example, for an FRP bridge deck,
the professional might need to know the delamination extent in the entire upper surface of the
deck. For a post-tensioned reinforced concrete beam, the corrosion and/or hydrogen embrittlement
extent along the tendons might be of interests. Another example for DR might be the displacement
of a single-span bridge during a load testing experiment. Note that the DR can be 3D (volume), 2D
(area), 1D (length), or a single point. The shape and/or extent of the DR are dependent on the type
and objectives of the experiment. Table 4.7 shows some examples of DR.
We can safely state that there is a DR for every SHM, as well as every NDT experiment. The
extent of DR is one of the several distinguishing factors between SHM and NDT. Generally speaking,
TABLE 4.7
Demand Regions for Different SHM Experiments
Type of DR Example
Point Maximum strain at center of single-span bridge
Linear Extent of damage (corrosion, hydrogen embrittlement) of post-tensioned tendons
Surface/area Delaminations of composite (FRP) plates in an FRP bridge deck
Volume Extent of damage in reinforced concrete bent after seismic event
SHM would be involved with larger DR than NDT. By definition, the DR must be well defined by
the project official (it is advisable to define the DR such that it is consistent with the project goals,
and with an eye on the potential decision making processes that might result as the SHM experi-
ment progresses).
Capacity Region: We define next the capacity region of a sensor CR as the coverage region
of the sensor of interest. For example, local fiber optics, for example, Fabry-Perot, or Fiber Bragg
Grating, has a point CR. Similarly, most mechanical or thin film strain sensors are point CR. Linear
CR can be guided wave ultrasonic transducers that are used for detecting pipeline damages (see
Chapter 7). Similarly, distributed Brillouine Fiber Optic Sensors are linear CR (see Chapter 5).
Area CR sensors are those that have surface applications. For example piezoelectric transducers that
pulses in a composite plate so as to identify damage location (using time of flight (TOF) technique,
in conjunction with other sensors) would have a surface type CR. Another application for surface
CR is using acoustic emission sensors to detect damage within a plate, again in conjunction of a set
of acoustic emission sensors using TOF technique. Similar examples can be used for volume CR.
Units Mismatch: For simplicity, we will assume that CR and DR have compatible units, that is,
point CR sensors will be used for sensing DR, and so on. Sometimes this is not the case. For exam-
ple, Fiber Bragg Grating sensors (which are point CR sensor) might be used for predicting damage
in a surface DR such as FRP bridge decks. Such a situation is more complex and will not be consid-
ered here. It should be mentioned that the added complexity will not detract from the generalization
of the STE.
Optimum Number of Sensors: We can now express the minimum number of sensors for a give
experiment as
DR
nDCR = (4.25)
CR
The fraction in Equation 4.25 is then approximated upward to the next higher integer number NDCR.
Note that for the trivial case (point DR and CR),
CR = DR = 0 (4.26)
N DCR = nDCR → 1 (4.27)
Uncertainties: Equation 4.25 computes nDCR using the deterministic values of DR and CR. Generally
speaking, the value of DR should be deterministic, with little or no uncertainties. Its value is deter-
mined by the goals of the SHM experiment. As such, it is dependent of the needs for cost savings,
or increased safety. In other words, the value of DR should be determined based on economic and/
or social reasons.
Most of the uncertainties in Equation 4.25 are found in estimating CR. We note that an upper
bound of CR might be found in the specifications of the sensor of interest. Using such an upper
bound of CR will result in the minimum number of sensors nDCR. There are several potential uncer-
tainties that can affect CR such as material properties, installation errors, quality of manufactur-
ing (tolerances), environmental conditions, potential interactions between different sensors and the
material they are embedded in/on, and so on. Current practice is to use a reduced value of CR to
evaluate the number of sensors using Equation 4.25. This is due to the fact that the statistical prop-
erties of such uncertainties are impossible to estimate generically. Is it possible to develop a quanti-
tative method to estimate the required value of CR?
Let us assume that the uncertain capacity region cr is uniformly distributed. A truncated normal
distribution might be more accurate, however, the uniform distribution is simpler to express. If the
readers desire they can develop similar expressions for the truncated uniform distribution case. The
maximum value of CR is CRmax, as discussed earlier. We need to define the minimum value of CR
as CRmin. The estimation of CRmin will depend on the experiment itself, as well as the expertise of
the professionals who are involved in the project. Defining X as the probability that a value of CR
will exceed certain value CR X, we can evaluate CR X as
CRX = CRmin + (1 − X ) (CRmax − CRmin ) (4.28)
The optimum number of sensors becomes
DR
nDCR =
CRX (4.29)
As an example, let us assume that in a particular TOF piezoelectric experiment, the total area of the
plate to be monitored is DR = 3000 in2. Each sensor has a CRmax range of 30 in2. The professional
estimated that, based on previous experience, CRmin = 10 in2. Moreover, the professional needed a
90% nonexceedance probability for that experiment. Thus X = 0.90. Applying Equation 4.28 we find
CR X = 12 in2. The optimum number of sensors that accommodates uncertainties of this situation is
250 sensors (utilizing Equation 4.29). By comparison, applying Equation 4.25 that does not account
for uncertainties would have required only 100 sensors. Utilizing such low number of sensors might
have resulted in an inaccurate SHM results.
4.3.3.3 Optimality of Sensor Location

Chapter 5 shows a detailed method of computing optimum sensor locations (OSL) for measuring
structural responses to gravity, seismic, and wind loads. Similar approaches can be used to compute
OSL for any other type of sensor arrangement.
4.4 THEORY OF SHM TRIANGULATION

4.4.1 Theory Statement
Suppose that a particular structural state needs to be detected by the use of some monitoring activi-
ties. Such a structural state can be the overall stability condition of the structure, current scour con-
dition under the bridge foundation, or the extent of damage in post-tensioned tendons. Each of these
conditions can be detected by one or more of SHM techniques. Equally, the detection of each of
these conditions is not an easy task; even with good planning, there is always a finite probability that
successful detection would not be achieved. We immediately recognize that this situation describes
a general situation, which can be formalized as follows:
1. There is an unknown structural state SDT.

2. There is a particular monitoring resource scheme SRi. The subscript i is a counter that
refers to the ith monitoring scheme. We assume that there are N monitoring schemes that
might be used to achieve the task.
3. There is an uncertainty of SRi being able to detect SDT. Such uncertainty can be assigned a
variance Vi.
Obviously, the probability that SDT can be detected, P(SDT) is the same as the probability that SRi
can detect SDT, P(SRi), if only single detection scheme i is used. Suppose that and additional detec-
tion scheme SRj, with i ≠ j, for a total of two schemes, is used. Would that improve the probability
of overall detection?
The problem can be illustrated qualitatively in Figure 4.13. The figure shows that each detection
scheme has a detection range. Whenever the detection ranges intersect, the probability of detection
increases. The qualitative illustration in Figure 4.13 shows three detection schemes: the region they
intersect at has a higher potential of being detected than surrounding ranges.
Accuracy of detecting state of

structure increases when targeted
by several detection schemes
Domain of detection Domain of detection

scheme #1 scheme #2
Detection schemes Domain of detection

include sensing, scheme #3
structural identification,
and damage
identification methods
FIGURE 4.13 Triangulation scheme in SHM.
We recognize the proposal in Figure 4.13 to be a form of triangulation. It is also intuitive that the
answer to the above posed question is “yes.” On the basis of this, we propose the following SHM
triangulation theory: Efficiency of detecting the structural state is improved by using a mix of detec-
tion schemes.
4.4.2 Proof of Theory

Even though the theory as stated above is fairly intuitive, there exists a simple proof by using prob-
ability theory. Let us consider two potential detection schemes SRi and SRj, with i ≠ j. The prob-
ability of detecting the structural state SDT if each of the schemes is used separately is P(SRi) and
P(SRj), respectively. If the two schemes are used together, the probability that SDT can be detected,
P(SDT) can be shown to be
(
P ( SDT ) = P SRi ∩ SR j ) (4.30)
which implies that the desired detection can be achieved by either SRi, or SRj. From elementary
probability theory
( ) ( ) ( )
P SRi ∩ SR j = P ( SRi ) + P SR j − P ( SRi ) P SR j (4.31)
Note that, by definition, 0.0 ≤ P(SRi) ≤ 1.0 and 0.0 ≤ P(SRj) ≤ 1.0. Let us disregard the trivial situa-
tions when P(SRi) = 0.0 or P(SRj) = 0.0, which would revert to the single detection scheme. We can
easily then, by inspection, reach the conclusions that
P ( SDT ) > P ( SRi ) (4.32)

and
( )
P ( SDT ) > P SR j (4.33)
This proves the theory, without any reservation.

The above proof makes the realistic assumption that the two detection schemes are mutually
exclusive, that is, the use of one scheme does not have an effect on the other scheme. Such an
assumption is reasonable in SHM situations. Furthermore, we note that the above proof can be
easily extended to situations where more than two detection schemes are used.
As an example, let us assume that it is desired to detect extent of damage in a small bridge.
The professional is considering two detection schemes: strain measurements and natural frequency
(vibrations) measurements. Previous experience indicates that the probabilities of successful detec-
tion for each of the two schemes, when used separately, are 75% and 60%, respectively. When the
two schemes are used together, by applying Equation 4.31, the probability of detection improves to
90%. The improvement in detection potential is certainly obvious. Such improvements can be even
greater in situations where detection probabilities of individual schemes are low to start with. For
example, if the probabilities in the previous example are 50% and 35% instead, the detection prob-
ability of the combined detection scheme, according to Equation 4.31 is 67.5%. The improvement in
detection probability is noteworthy indeed. Figure 4.14 shows overall triangulation effects for two
detection schemes.
4.4.3 Discussion and Applications

4.4.3.1 General Discussion
According to the theory, for example, if it is desired to monitor potential of brittle failure, it would
be more efficient to use a mix of accelerometers and stain gauges, than to use only one ToM, say
strain gauges. The explanation is simple: strains are direct measure of deformation; they will offer
insight to stiffness, but will not offer a clear insight to an imminent loss of stability. Velocities offer
a direct measure of inertia; careful monitoring would reveal a relationship between stiffness and
effective inertia (Timoshenko 1955). Effective inertia is a direct measure of imminent loss of stabil-
ity. Thus, a careful mix of different ToM can offer more insight to the structural behavior. By careful
use of different ToM more information can be revealed. A step by step application of triangulation
is as follows.
Triangulation effects
100%
90%–100%
90% 80%–90%
70%–80%
Overall probability of detection
80% 60%–70%
50%–60%
70%
40%–50%
60% 30%–40%
20%–30%
50% 10%–20%
0%–10%
40%
30%
20% 90%
10% 60%
0% Probability of detection for
30% scheme # 2
0%
10%
20%
30%
40%
50%
60%
0%
70%
80%
90%
100%
Probability of detection for scheme # 1
FIGURE 4.14 Triangulation probabilities for two-detection schemes situation.

State goals of Monitoring, for example, detecting changes in boundary conditions, or connection
stiffness
1. Identify parameters that cause, or result from, such changes. In the above example, a fro-
zen bearing can cause boundary condition change, while a large strain can result from
boundary condition change
2. Place appropriate tilt and strain gauges to monitor the boundary condition changes
3. Using a mix of predetermined acceptance rules, if observed changes reach a threshold, the
facility owner is informed
In order to generalize the triangulation concept even further, we can qualitatively define structural/
damage identification by a qualitative measure DETEC (for detectability). The detectability of the
structural state as a function of a specific sensor type ST and number SN can be described formally as
DETECi = f ( STi , SN i ) (4.34)
The subscript i is used to identify the specific experiment. It there are two different sensing types
each will result in a monotonically increasing DETEC as shown in Figure 4.15. The needed thresh-
old detection might be reached at a specific combination ST1 and SN1 for the first type or ST2 and
SN2 for the second type. Figure 4.15 shows a situation where DETEC1 reaches the required threshold
of detection, while DETEC2 failed to detect the desired structural state. By using a combination of
ST1, SN1, ST2 ,and SN2 the detectability rate and potential increases, for similar sensing resources, as
shown in Figure 4.15 (DETEC3). The qualitative relationship shown in Figure 4.15 will be studied
and quantified during the course of the project.
The above examples and discussions can be generalized so as to cover most generic SHM situ-
ations. This can be accomplished by following the above steps. Note that in the above example,
we used a mix of only two ToMs. There is no theoretical limit to the number of ToM mix that can
be used to triangulate and identify the desired behavior. For example, temperature, accelerations,
acoustic emission, ultrasound or even radiation waves can be used in the mix. There would be, of
course an optimum number and location of each type of sensors that would satisfy the SHM objec-
tives in an efficient and accurate manner. Table 4.8 shows the four generic steps in the triangulation
method, and the issues that need to be considered for each step.
Let us have a more detailed look at some of the practical situations that were mentioned earlier,
and see how triangulation might be applied to them. Table 4.9 shows practical situations and the
different possible SHM detection schemes that can triangulate the detection process.
4.4.3.2 Case Study

Maser (2008) offered a practical example on the utility of integrating two technologies and how
such an integration can offer increased efficiency and accuracy. In the experiment, ground pen-
etrating radar (GPR) and infrared thermography (IR) have been applied for bridge deck condition
Detection level
DTECT1
DTECT2
DTECT3
Required detection
threshold
Detection resources
FIGURE 4.15 Detection versus resources relationships.

TABLE 4.8
Triangulation Issues in SHM
Triangulation Steps Issues
SHM goals Diagnostics, performance, safety, hazard (earthquake, scour, fatigue, etc.) effects, etc.
Parameters Strains, displacements, modal parameters, tilting, corrosion, normal wear and tear, bridge
rating, etc.
Appropriate sensors Strains, accelerometers, pressure, etc.
Acceptance rules Confidence levels, acceptable risk, reliability (probability of failure), etc.
TABLE 4.9
Triangulation Examples
Problem Triangulation scheme
Structural stability For stability (especially elastic stability) evaluation, both capacity and demand need to
be sensed. For an accurate triangulation effort, use a combination of vibration, strain,
and load sensors.
Scour condition See Chapter 1 by Ettouney and Alampalli (2012)
Damage in post-tensioned Use impact-echo process (Section 7.4.2.3). The triangulation is achieved by applying
tendons the process at uniform grid on the surface of the post-tensions system. The direction of
the detection scheme is perpendicular to the direction of the tendons. This can be
combined with methods such as ground penetrating radar.
assessment. GPR is equally effective on bare concrete decks as well as on decks with concrete and
bituminous overlays. GPR data requires office processing to reveal deteriorated areas. While pro-
ducing good quantity estimates and general locations, it is not strong at precisely locating delami-
nated areas. While GPR is generally effective for deck slabs on girders, it has limitations with
one-way slab type of decks. IR produces a visual delaminations image immediately in the field that
can be checked on the spot using sounding and cores, and thus provides good location accuracy.
However, as a surface temperature method, its detection capability is depth limited. While it is very
effective for non-overlaid decks, its effectiveness can be reduced in the presence of overlays. IR data
collection takes place at 5 mph and requires a moving closure, and solar radiation conditions must
be adequate to produce the required temperature differentials. This project combined the data from
the two methods to improve the data quality by minimizing limitations of each method.
Forty five bridge decks in Wisconsin were tested using both techniques along with under deck
visual surveys, and selective core sampling (see Figure 4.16). GPR rebar depth data was used for
each deck to determine the depth limitations of IR detection. Where the rebar was too deep, IR
was used to identify overlay debonding, while GPR was used to identify rebar depth deterioration.
For decks that were not well suited for GPR, IR provided necessary deterioration information. The
author reports that combination of IR and GPR maximized the capabilities of each method and
compensates for the limitations (see Table 4.10).
4.4.4 Evaluating Detectability of SHM Schemes

In the previous sections of this chapter, we have made extensive use of the probability detection
P(SRi) of a given SHM scheme SRi. In this section, we offer methods of evaluating this probability.
Specifically, we offer a quantitative and a qualitative method. The qualitative method is based on
evaluating previous experiences in a numerical fashion, while the qualitative method depends on the
experiences of the analyst. Obviously, quantitative method is preferable, but it requires more effort.
(b)
(a)
FIGURE 4.16 GPR and infrared equipment. (a) back of the equipment vehicle and (b) front of the equipment
vehicle. (Reprinted from ASNT Publication.)
TABLE 4.10
Summary of Core Results
Detected by Total (#) Debonded (#) Delaminated (#) Correct (%) False alarms (%) Misses # − (%)
IR &/or GPR 105 25 61 83 17 5 5
IR only 64 20 35 86 14 22 26
GPR only 59 9 33 71 29 18 17
TABLE 4.11
Consistent Categories for Histogram
Category Comment
Type of SHM experiment Vibration measurement, remaining fatigue life, scour effects, corrosion extent, etc.
Type of bridge Simple span, multiple spans, construction material, structural system type and other
pertinent bridge attributes
Sensors Relatively similar types of sensors. Number and interconnectivity of sensors should be
consistent. If wireless sensing is used, the wireless network and topology should be
consistent
Instrumentation Instrumentation hardware should be consistent with sensors, bridge type, and experiment
objectives.
4.4.4.1 Numerical Method for Evaluating Detectability

The steps for such an evaluation are as follows:
Identify Suitable SHM Database: This step requires identifying and enumerating all previous
SHM schemes that are consistent with the SHM scheme of interest. Some examples of schemes of inter-
est can be to detect modes of vibration, remaining fatigue life or scour performance. Table 4.11 shows
other consistent categories that need to be considered for proper inclusion in the histogram database.
Identify Adequate Detectability Measurement Scale: A scale should be established for mea-
suring detectability SRi, for the remainder of this section we drop the subscript, without any loss
of accuracy. One possible scale can range from 0 to 10. As shown in Table 4.12. Admittedly, the
descriptions of each value of SR are qualitative. This approach is not uncommon in many engineer-
ing application and thus, we will be adopting it for our current purpose.
TABLE 4.12
Scale of SR
SRSCALE =
Scale of SR Description
0 Complete failure of experiment. No beneficial information were gathered
1 About 30% of the objectives were realized
2 About 30% of the objectives were realized. Some additional information was realized
8 Complete success for all objectives. No additional objectives were realized
9 Complete success for all objectives. Some additional objectives were realized
10 Complete success for all objectives. Many additional objectives were realized
Note that there is an added value in the estimation of SR to realizing additional information (Serendipity
principle).
TABLE 4.13
DRL Scale
DRL Scale Description
0 Trivial case: no resources for the SHM scheme
1 Very limited resources
2 Medium-low resources
3 Medium resources
4 Medium-high resources
5 High level of resources
Using the database of consistent SHM schemes and evaluating the results of those schemes, a
numeric value of SR can now be estimated using tables similar to Table 4.12.
Identify measure of detection resource and detection resource level: Since SHM schemes vary
greatly in regards to the resources used for the experiment, it is logical to assume that detectability
of a given SHM scheme SR is a function of the SHM detection resource level DRL. The DRL is the
horizontal scale in Figure 4.15. The DRL can be described on a sliding scale, say from 0 to 5, with
each level qualitatively defined as in Table 4.13.
The database that is collected in previous step would be categorized further according to their
estimated DRL.
Prepare Histogram: For all the records in the database within a specific DRL, a set of the estab-
lished values of SR form the basis of a histogram. The size of the histogram is NRECORD: the number
of consistent SHM schemes in the database. See Chapter 8 for a description of this process.
Compute Desired Probability: Using the histogram, the probability of detection P(SR) can be
evaluated as the cumulative probability
P(SR) = P( X ≥ SRSCALE ) (4.35)

i
The choice of i depends on the desired standards of management. For example i = 10 is too demand-
ing, while a choice of i = 7 is more realistic and less demanding.
Note that the method as described above can be performed by sending questionnaires to bridge
managers or experts for reporting their experiences with different SHM schemes. By archiving the
results, a probability database can be established in a central location. Such probability database can
be used by the SHM community for future SHM projects.
4.4.4.2 Subjective Method for Evaluating Detectability

When the information needed for the numerical method is not available, the users can rely on their per-
sonal experiences and judgments. The desired probabilities P(SR) can be estimated by the users. Such an
approach is judgmental and is less accurate than the numerical approach. It is simple and less costly.
4.4.5 Triangulation Cost–Benefit Analysis

4.4.5.1 Analytical Considerations
We have been focusing on the methods and benefits of triangulations so far. Of course, there is a
cost for achieving such benefits. Whenever an additional detection scheme is used, there will be
additional costs. The balance of cost–benefit needs to be considered carefully before embarking on
a triangulation technique. The remainder of this cost–benefit section will consider the two-scheme
situation. More schemes can be analyzed in similar manner.
Cost of Triangulation: Cost of triangulation is simple to compute. It is the total cost of adding
another SHM scheme. For example, if the cost of the first scheme is Ci, and the cost of adding
another scheme is Cj, then the total cost of triangulation is
CTriangulation = Ci + C j (4.36)
For ease of computations, we use monetary measures for all costs and benefits.
Benefit of Triangulation: As usual, estimating monetary value of benefits is not an easy task.
We try to devise a technique to avoid this difficulty, without any loss of generality. We assume
first that the monetary benefit of a 100% successful SHM project is BSHM. For the ith and jth SHM
schemes, the probability of detection are P(SRi) and P(SRj), respectively. The probability detec-
tion when using both the ith and jth SHM schemes is P(SDT). The benefit when using ith scheme is
Bi = P ( SRi ) BSHM (4.37)
Similarly, the benefits of the jth and the triangulated SHM schemes are
( )
B j = P SR j BSHM (4.38)
BTriangulated = P ( DT ) BSHM (4.39)
P(DT) is the probability of detection with triangulated scheme. Note that, unlike the cost, the tri-
angulated benefits are not the simple sum of the benefits of the two schemes. Having identified all
costs and benefits, we can move to the benefit cost analysis phase.
Benefit-Cost Analysis: What concern us here is the changes in benefit-cost ratios from the base-
line case (ith SHM scheme only) to the triangulated situation, where both ith and jth SHM schemes
are used. The baseline benefit-cost ratio is
Bi (4.40)
BCi =
Ci
The triangulated benefit-cost is

BTriangulated
BCTriangulated = (4.41)
CTriangulated
Expressing the benefit costs in terms of BSHM
P ( SRi ) BSHM
BCi = (4.42)
Ci
The triangulated benefit cost is
P ( DT ) BSHM
BCTriangulated = (4.43)
Ci + C j
The ratio of the benefit cost of triangulated and baseline schemes is
BCTriangulated P ( DT ) 1
= (4.44)
BCi P ( SRi ) Cj
1+
Ci
The condition the a triangulated scheme would produce a better benefit cost is
BCTriangulated
> 1.0 (4.45)
BCi
This condition is satisfied when

P ( DT ) Cj
> 1+ (4.46)
P ( SRi ) Ci
This equality is an interesting result: even though the probability of detection in a triangulated
SHM setup is always higher than an individual SHM scheme, the benefit cost of triangulation is not
always higher than a single scheme.
Let us continue the previous example of a baseline detection probability and triangulated prob-
ability of 75% and 90%. If the costs of the baseline and additional SHM schemes are $1,000, and
$600, the left-hand side of the equality is 1.2, while the right-hand side is 1.6: the equality is not
satisfied, and triangulation is not benefit-cost effective. If the costs of the two schemes are $1,000,
and $150, the left-hand side of the equality is still 1.2, while the right-hand side is 1.15: the equality
is satisfied, and triangulation is benefit-cost effective.
4.4.5.2 Case Studies

Gray and Gray (2005) discussed the cost benefits of the use of multitudes of nondestructive evalua-
tion (NDE) techniques in a given situation. Since different NDE techniques provide differing data
(information), the combination of those differing data was defined as the data fusion. They argued
that all available NDE data must be considered in the data fusion paradigm. Of course, they raised
important question: is the additional available date worth the additional cost? We can extend the
logic of the data fusion into the SHM realm. The data fusion in SHM is the combination of data
that result from varying SHM projects, or techniques. For example, in an SHM corrosion evalua-
tion project for bridge piers, appropriate data fusion approach would be periodic measurements of
pH levels in concrete, ambient temperature as well as periodic visual inspection. Note that each
of these parameters can result in an estimate of corrosion in concrete individually. However, cor-
relating the data (fusing them) can result in a more reliable and accurate corrosion estimates. Even
though the use of data fusion in any SHM project would increase reliability and accuracy of the
results, the SHM professional should ask the same question suggest by Gray and Gray (2005): is the
cost for the additional data source worth the effort?
Obviously, the answer to such a question is a case-dependent. We can offer the following guide-
lines for efficiently using additional data sources to incorporate a data fusion process in an SHM
project:
1. Utilize a multihazards process, if appropriate

2. Utilize optimum sensors location techniques whenever possible
3. Utilize different analytical techniques for evaluating value and cost–benefit
4. Utilize different decision making tools
4.4.6 Concluding Remarks

This section presented the concept and theory of triangulation as applied to SHM. We proved that
by using a mutually exclusive detection schemes, the odds of successful detection increases. Few
examples showed that there are several other potential applications to the SHM triangulation theory.
Further explorations to these potential applications can be performed by future researchers. We
provided benefit-cost analysis of triangulation, and showed that triangulation can be cost–benefit
effective in certain conditions only. For now, we summarize some benefits of SHM Triangulation as
(a) improved efficiency (both cost and accuracy), (b) improved expandability of SHM efforts, (c) can
aid in objective evaluation of overall SHM schemes, and (c) makes it easier to link SHM results to
other efforts, such as bridge management systems.
4.5 DUALITY PRINCIPLE IN SHM

Structural health monitoring can be considered as field experimentation application: it strives
to monitor the structural behavior through a combination of automatic sensing and visual
observations. The GTE and STE mentioned earlier address the efficiency of experiments by
discussing the value, in a qualitative manner, of both the experiment itself and/or the sensor
arrangement. For us to present the duality principle, we need to define the value in a more
quantitative fashion as
BE
VE = (4.47)
CE
The cost, in today’s dollars, of the SHM experiment is CE. The benefit of the experiment is BE. The
cost CE is relatively easy to define, while benefit BE is more difficult to quantify. For our immediate
purposes, let us define a qualitative benefit scale, as shown in Table 4.14. (Note that the benefits BE
in Table 4.14 are only for demonstration. Magnitudes of BE should be assigned by the users to suit
individual circumstances. Note that the scale and range of these magnitudes should be consistent
with cost scale and range.) The designer of the SHM experiment would assign the appropriate BE
in Equation 4.47, to determine VE. Since we are interested in the relative merits of the experiment
SHM field, this simple definition of BE would suffice.
Having laid down the quantitative foundations for evaluating the value of an SHM experiment, we
turn our attention to the duality issue. We will limit the scope of Equation 4.47 to only those SHM
experiments that handle specific hazard (earthquakes, wind, corrosion, fatigue, scour, excessive
TABLE 4.14
Categories of SHM Experimental Benefit, BE
Description of SHM Experiment Benefit BE
Experimental results would not lead to direct benefit (system 1
behavior, safety concerns, economic concerns)
Somewhat beneficial behavioral information. No immediate, or long 2
term, direct result to safety or economic concerns
Considerable behavioral information. Some long-term results to 3
either safety or economic concerns
Considerable behavioral information. Considerable long-term results 4
to either safety or economic concerns
Considerable behavioral information. Considerable immediate and 5
long term direct results to either safety or economic concerns
vibrations, etc.). Let us assume that a particular SHM experiment is to be performed for hazard i.
The value of such an experiment is
BEi
VEi = (4.48)
CEi
In a multihazards environment, the value Equation 4.48 can be generalized to

j = NH
VEi = ∑V
j =1
Eij (4.49)
BEij
VEij = (4.50)
CEij
The value of SHM experiment for the considerations of the primary hazard (i = j) is V Eii. The
cost of the primary experiment is CEii, while the benefit of the primary experiment is BEii. The
i ≠ j terms in Equation 4.50 can be defined as the multihazards terms. Thus VEAi = Σ jj == 1NH VEij |i ≠ j
represents the added value to the primary experiment if multihazard considerations are consid-
ered. The added cost to the SHM experiment with the ith primary hazard to consider aspects of
the jth hazard is CEiji≠j. The added benefit to the SHM experiment with the ith primary hazard to
consider aspects of the jth hazard is BEiji≠j.
The multihazard SHM value equation can now be represented as
VEI = VEii + VEAi (4.51)
Equation 4.51 represents another manifestation of the multihazards theory; see Ettouney et al.
(2005). An implicit assumption that was made in developing the equation is that the structure would
behave in a way that manifests multiple hazards at once; we call this physical behavior the duality
principle. It behooves the SHM designer to take advantage of that behavior to maximize the value of
the SHM experiment. For example, accelerations resulting from the impact damages of Figures 4.17
and 4.18 might have been detected if there was an earthquake monitoring system on those systems.
Such monitoring could have alerted the officials of the accident in real time. Accelerations, wave
forms and amplitudes could have given a better indication of severity of damage. Figure 4.19 shows
the concept of the principle.
FIGURE 4.17 Impact damage. (Courtesy of New York State Department of Transportation.)
FIGURE 4.18 Truck impact damage. (Courtesy of New York State Department of Transportation.)
4.6 SCALING PRINCIPLE IN SHM

4.6.1 Overview
The categorization of SHM into global and local is well known in the field. The feasibility of inte-
gration of global and local SHM activities is one of the main objectives of this proposal. We argue
that one of the most promising way to integrate global and local SHM efforts is by approaching it
as a multiscale problem, and use both qualitative and quantitative techniques to scale parameters so
as to achieve the desired SHM integration. The scaling process involves different techniques at all
levels and for all components of SHM.
4.6.2 One-Dimensional Scales

4.6.2.1 Frequency Spectrum
Frequency scale, which is shown in the electromagnetic spectrum, as explained in Chapter 7, is
widely utilized in NDT applications. The unit of the scale is in Hz. Sometimes wave length is
Main purpose for monitoring Duality principle in SHM
Monitoring for expected Increase value of

earthquakes experiment by
-Accelerometers monitoring for dual
objectives
-Strains
-etc.
Strain sensors can Accelerometers can

Other? be used to monitor be used to monitor
for -fatigue for
-Truck loads -wind
Dual roles for monitoring
FIGURE 4.19 Duality principle in SHM.
also used as a companion scale. Frequency scale spans most of NDT methods and applications. It
typifies the very concept of scaling: different methods and techniques are applicable at different
frequency scales. When designing an SHM projects, it makes sense to try to utilize different ranges
in the frequency scale so as to achieve the goals of the project. For example, it might be beneficial
to utilize acoustic emission at higher frequency range to detect smaller fatigue cracks, while using
conventional accelerometers at much lower frequency range to detect the global effects of larger
size damages.
4.6.2.2 Sensor Density

Sensor density scale ρ SENSOR can be defined as the average number of sensors N in a given area A.
Formally,
N
rSENSOR =
A (4.52)
As ρ SENSOR increases the capability of detection of the ToM increases. There is an optimal value of
ρ SENSOR that depends on sensor sensitivity, costs, and SHM project objectives. It is prudent to try to
estimate this optimal vale of ρ SENSOR before embarking on the experiment. Before the emergence
of wireless sensing as a primary tool of SHM, ρ SENSOR was fairly small, because of the limitations
imposed by wires. The increasingly wide use of wireless sensing enabled SHM practitioners to use
higher values of ρ SENSOR.
4.6.2.3 Scaling in Structural Identification

Structural scaling problem have been well known in other field, such as acoustics and vibration
fields (Ettouney et al. 1997). Simply stated: low frequency range (0 ~ 100 Hz), medium frequency
range (~ 100 ~ 1000 Hz), and high frequency range (higher than ~ 1000 Hz) all require differ-
ent sets of structural analysis techniques. In problems such as NDT/SHM, immense frequency
range do exists. Some NDT methods such as ultrasound or acoustic emission operate in the KHz or
MHz ranges, while conventional modal identification methods operate in the low frequency range.
There is a need to interrelate information of such a wide frequency range. Scaling techniques offer
potential of such interrelationships. By having a continuous picture of the entire range of the system
in a single data bank (visual, or numerical), decisions as to the state of the structural health can be
easily and accurately analyzed.
Another important structural analysis scaling problem occurs when the structural model is built
to model different hazards. Even though, the physical system accommodates different hazards,
including their interdependencies, numerical analysis of different hazards would require separate
analysis considerations. Ignoring this would produce inaccurate results. This can be simply exem-
plified through multihazard design of bridge piers for seismic and blast load cases. For seismic
loading, the design and analysis model can be a simplified SDOF model. On the other hand, a
higher precision finite element model may be required for analysis and design for blast load cases.
Figure 4.20 shows a high resolution finite element model for studying blast load on bridges. Note
the extremely small finite element sizes that are needed to capture the modeling demands of the
high frequency blast pressures. STRID modeling techniques should accommodate such scaling
principles closely.
We introduce STRID scale STRSCALE. The scale is a qualitative descriptor of the structural ana-
lytical model resolution. One possible formal definition of STRSCALE is
Min ( SELEMENT )
STRSCALE = (4.53)
l MIN
SELEMENT = Descriptor of the size of elements in a finite element model

λMIN = Minimum wave length of interest in the SHM experiment
The minimum element size Min(SELEMENT) should adequately transmit the minimum wave in the
model whose wave length is λMIN.
4.6.2.4 Scaling in Damage Identification

There are numerous methods for detecting damages in structural systems. Each of these methods is
designed to detect specific type of damages. Currently there are no techniques to scale all of these
methods and damage types in a coherent and general fashion. Alampalli and Ettouney (2007) pre-
sented a chart for categorizing and scaling damage in a comprehensive fashion. Such damage scal-
ing organization can be easily used to categorize varying NDT/SHM results (which seems at first
FIGURE 4.20 High resolution numerical finite element model. (Courtesy of Dr. Anil Agrawal.)
glance to be completely unrelated) and input them into a decision making tool box, thus utilizing all
pieces of information that is available from any NDT/SHM effort.
Similar to STRID, we introduce a DMID scale DMSCALE. The scale is a qualitative descriptor of the
damage size as related to a specific size of the system. One possible formal definition of DMSCALE is
Min ( DM SIZE )
DM SCALE = (4.54)
SSYSTEM
DMSIZE = Descriptor of the size of monitored damage in the SHM project

SSYSTEM = Adequate descriptor of size of system
Different NDT or SHM methods are adequate for different ranges of damage sizes. Thus, the
chosen method of monitoring must scale the required range of DMSCALE. If the method can’t scale
the required range, more than one method may be needed (see theory of triangulation in this chap-
ter). Figure 4.21 illustrates the importance of damage scale in SHM.
4.6.2.5 Hazard (Demand) Scale

We concern ourselves with spatial scale of hazards. Hazards of any type can be categorized by its
spatial attribute. Figure 4.25 shows qualitatively spatial values of different hazards as they affect
structures. Hazards with local spatial range, such as fatigue or corrosion (Figure 4.22), will result in
local effects on structure. Hazards with larger spatial range, such as overload (Figure 4.23), wind,
flood (Figure 4.24), or earthquakes will affect the structure both locally and globally. The concept
is shown in Figure 4.25.
Table 4.15 summarizes some scaling issues, for illustration purposes, for different components of
SHM field that this proposal will consider.
4.6.3 Two-Dimensional Scale

We observe that the linear scales described so far do interact together; thus generating what might
be described as two-dimensional scale. The importance of interactions between the different SHM
scales varies in importance. Table 4.16 shows a qualitative estimate of such importance for illustra-
tion purposes. Only the upper side of Table 4.16 is filled, since the table is symmetric.
4.6.4 Scales of Higher Dimensions: Role of Decision Making

We identified several scales in SHM. Some or all these scales need to be covered, either partially
or fully, during SHM projects. We also showed that in most situations those scales are related,
Large impact: Pitting: Extensive corrosion:

Large-scale damage Small-scale damage Medium-scale damage
Reinforced concrete beam
As the damage scale changes, the pertinent

SHM method should change accordingly
FIGURE 4.21 Damage scale concept in SHM.

FIGURE 4.22 Corrosion hazard: localized. (Courtesy of New York State Department of Transportation.)
FIGURE 4.23 Overload hazard: global. (Courtesy of New York State Department of Transportation.)
together, forming 2D or higher dimensions. Clearly, there is no single technology or algorithms that
can handle all of those multiple scales simultaneously. A good management scheme is needed. We
offer the following steps for multiple, or single, scaling situations:
1. Identify pertinent scales

2. Identify technologies that are needed to cover required range of scales for the project on
hand
3. Identify algorithms that would span different scale ranges, first on each linear scale, then
across multiple scales
4. Place adequate decision making rules
FIGURE 4.24 Flood hazard: global. (Courtesy of New York State Department of Transportation.)
Wind
Blast pressure
Different popular hazards
Earthquakes
Impact
Corrosion
Machine vibrations
Progressive collapse
Accoustic
Gravity loads
Intermediate
Local (affects several Global
(affects few components, (affects most components,
subcomponents) subglobal effects) global effect)
Spatial extent of hazard when applied to system
FIGURE 4.25 Spatial scale and multihazards considerations.

TABLE 4.15
Important Scaling Issues in SHM
SHM Component Scaling Issues
Sensing Type of measurement (ToM), space/distance, time (real, intermittent), sampling rates
Structural identification Frequency range, number of modes, mode shapes, standing versus propagating waves, wave
numbers/lengths, element type and size
Damage identification Size, extent, source of damages, multihazards issues
TABLE 4.16
Qualitative Importance of SHM Scale Interactions
Scale Frequency Sensor Density STRID Damage Scale Hazard Scale
Frequency NA M H H L
Sensor density NA M H M
STRID NA H H
Damage scale NA
Hazard scale NA
NA = not applicable, L = low, M = medium, H = high
4.6.5 Benefits of Scale Considerations in SHM

There are immense benefits and potential for having successful scaling of SHM methods as
described above. For example, by scaling sensor arrays, it is possible to measure and detect higher
order mode shapes (thus being able to make a judgment for the reason of a frequency shift as to
whether a bearing is frozen, or loss of composite action). Scaling structural identification methods
would enable the observations of effects of large range of wave numbers (or wave lengths). Another
benefit of scaling SHM techniques would be the smooth integration of current SHM applications
which are geared in general to global behavior with the well-known techniques of NDT, which are
geared to localized behavior.
4.7 SERENDIPITY PRINCIPLE IN SHM

When an SHM experiment is conducted to monitor a hazard, and if by an accident or by chance,
some beneficial information were obtained regarding another hazard, we call this event the seren-
dipity principle in SHM. Serendipity principle, as defined, is definitely a by-product of the multihaz-
ards theory as introduced earlier. It is natural to expect that while investigating a particular hazard,
an un-anticipated bit of information about another hazard might reveal itself due to the interdepen-
dencies that the multihazard theory describes. Figure 4.26 illustrates the serendipity principle in
SHM.
We observe that the serendipity principle does not have a rigorous objective approach to it. As
such, it might be difficult to defend its inclusion in the context of this chapter. However, there are
measures that the owner might follow to maximize the utility of the serendipity principle, and the
chances that unintended, yet beneficial, information might result from an SHM experiment. Some
of these measures are as follows:
• Increase awareness of the system under consideration. This includes construction material,
structural system, load paths, redundancies, support conditions, etc.
Serendipity principle in SHM
Hazard # 1 (Target hazard) Hazard # 2
SHM experiment designed only

for hazard # 1
SHM experiment designed mainly for hazard # 1, with

additional considerations to other hazard possibilities
If targeting hazard #1 If targeting hazard #1

for SHM for SHM
experimenting was experimenting was
erroneous, the erroneous, the
experiment would fail, experiment would
since it was designed succeed, since it still
to monitor hazard #1 can monitor hazard #2
only
FIGURE 4.26 Serendipity principle in SHM.
• Perform careful design of sensors, instrumentation and data collection, and analysis
algorithms
• Ensure the compatibility of the SHM experiment to the intended goal of the experiment
• Keep an open mind of any environments that have not been included in the SHM experi-
ment design phase
• Pay attention to all the observed trends, especially those trends which might not have been
expected; do not be hasty in discounting unexpected trends as “just noise”
The above are just some common-sense rules that would increase chances of success of a SHM
experiment through the serendipity principle.
As an example to the serendipity principle, let us consider the Newburgh-Beacon SHM experi-
ment (Shenton et al. 2002). Before the experiment, field tests have indicated large magnitude stress
spikes that occurred in the bridge hangers. The level of the stress spikes was high enough to prompt
the launching of an SHM experiment in search for the causes of those spikes. In planning for the
experiment, the authors arranged for thirteen active strain gauges. In addition, sensors for measur-
ing wind speed and direction, vertical accelerations, relative bearing displacement, and temperature
were included in the overall instruments. The authors also provided two “dummy” strain gages.
One of those dummy strain gauges (named “zero dummy gauge”) had no lead wires and was used
for control purpose (it was not attached to the structure). The other “dummy” strain gage was not
attached to the structure either, however, it had a 100-ft lead wire that was draped over the catwalk
(hence it was called “100 ft. dummy gauge”). Neither of the “dummy” sensors was expected to reg-
ister any strain spikes during the experimentation.
Spk1197: 11/16/00 4:11 pm

5
(a)
0
–5
Stress –10
–15
Gage 3
–20
0 50 100 150
Time (sec)
Spk1197: 11/16/00 4:11 pm
(b) 5
0
–5
Stress
–10
–15 100ft dummy

–20
0 50 100 150
Time (sec)
FIGURE 4.27 Stress measurements at actual and dummy gauges (a) measurements during weekend, and
(b) dummy gage measurements. (Reprinted from ASNT Publication.)
The researchers made three observations during the experimentation; (a) The active strain gages
recorded spikes only during work days, but had almost no recorded spikes during weekends, see top
graphic in Figure 4.27; (b) All the truss members that intersect in an instrumented joint recorded
compressive spikes; and (c) While the zero dummy gage recorded no spikes, as expected, the 100-ft
dummy gage did record spikes with large strain amplitudes, see bottom graphic in Figure 4.27. The
first observation pointed to the potential that the spikes did not result from human activities, and
the second observation pointed to the fact that the spikes can’t be due to a physical behavior of the
bridge (not all members in a single joint can be compressive at the same time). Finally, the record-
ings of the 100-ft dummy gage pointed the attention to what the researchers suspected: the reason
for the spikes is noise/interferences, not a real stress. In addition, experimentation confirmed this
otherwise unexpected finding.
This experiment shows that careful planning and foresight, and a little extra effort (by adding the
dummy gauges) yielded a successful experiment, which is a good example for the serendipity prin-
ciple: what started as an investigation for sources of overstress spikes in bridge components, ended
in finding completely different source for the spikes: a harmless noise/interference.
REFERENCES
Alampalli, S. and Ettouney, M., (April 2007) “Structural Identification, Damage Identification and Structural
Health Monitoring,” Nondestructive Characterization for Composite Materials, Aerospace Engineering,
Civil Infrastructure, and Homeland Security 2007, SPIE Volume 6531, San Diego, CA.
Annis, C., (2005) “How Many Specimens is Enough?” Proceedings of the 2005 ASNT Fall Conference,
Columbus, OH.
Ettouney, M. and Alampalli, S., (2012), Infrastructure Health in Civil Engineering: Applications and
Ettouney, M., Alampalli, S., and Agrawal, A., (2005) “Theory of Multihazards for Bridge Structures,” Bridge
Structures, 1(3): 281–291.
Ettouney, M., Daddazio, R., and Abboud, N., (1997) “Some Practical Applications of the use of Scale Independent
Elements for Dynamic Analysis of Vibrating Systems,” Computers and Structures, 65(3): 423–432.
Gray, I. and Gray, J., (2005) “Assessing UT, EC, and RT Image Processing and Information Fusion with NDE
Computer Simulation,” Proceedings of the 2005 ASNT Fall Conference, Columbus, OH.
Hyland, D. and Fry, G., (2000) “A Neural-Genetic Hybrid Approach for Optimizing Structural Health
Monitoring Systems,” Proceedings of 2nd International Workshop on Structural Health Monitoring,
Stanford University, Stanford, CA.
Maser, K., (2008) “Integration of Ground Penetrating Radar and Infrared Thermography for Bridge Deck
Condition Evaluation,” NDE/NDT for Highwaya and Bridges: Structural Materials Technology (SMT),
ASNT, Oakland, CA.
Shenton, H., Chajes, M., Finch, W., and Sivakumar, B., (2002) “Long-Term Monitoring of the Newburgh-
Beacon Bridge,” Proceedings, Structural Materials Technology V: An NDT Conference, NYSDOT,
FHWA and ASNT, Cincinnati, OH.
Timoshenko, S., (1955) Vibration Problems in Engineering, Van Nostrand Reinhold Company,
New York, NY.
5 Sensors and
5.1 INTRODUCTION
5.1.1 Sensing and Infrastructure Health
In a general overview, sensing can be defined as the process of knowing specific information about
the system of interest. With this general definition, we can include human inspection as a sensing
activity. A more technical definition of sensing is that it is a process where sensors interrogate
(interact) the system, and produce certain information about the state of the system. We have thus
identified several components in the sensing process: (a) the sensors, (b) the information, (c) the
system. The sensors are those vehicles that will detect the state of the system and produce the appro-
priate information to the user. The information (defined later as type of measurements [ToM]) is
transmitted to end user by the sensors: they describe specific states of the system. The system is the
infrastructure on hand, or any component within it. Figure 5.1 shows this relationship.
The well-being of any infrastructure from its very inception to the last day of its decommission-
ing is a continuous exercise of this system → sensor → information process. In addition, we note
that as an outcome to the process, there will be some kind of a decision making that may affect
the system (a null decision is also a type of decision making). Throughout history such a process
existed, and will always exist as long as there is infrastructure. What is changeable are the (a) type,
quantity, and quality of information; (b) type, shape, and accuracy of sensors; (c) type, size, and
function of infrastructure and (d) type of resulting decisions. As we mentioned earlier, humans can,
and should be, considered as a type of sensor. Human sensing abilities are remarkable. Such abili-
ties can be extremely accurate at times. Unfortunately, there are limits to human sensing abilities.
In such situations, the need for more accurate sensing devices arises.
The interrelationship between the triumvirate system → sensor → information, and their interac-
tion with the decision making processes, is fairly complex. There are immeasurable combinations of
systems, sensors, and information. Because of this, the choice of sensing method, or sensor, should
be done carefully. The situation can be described as an optimization problem: given all the problem
parameters, it is decided to find the optimal sensor and sensing method that would result in the most
efficient resolution. The efficiency metric in this case can be cost effectiveness, accuracy of results,
time savings, or any other suitable metric.
We thus reach an important conclusion: sensors (including humans) are an integral and invalu-
able component in the well-being of infrastructure health. Not only the sensors as separate entities
are important, but also their interrelationships with all other aspects of infrastructure are important.
This chapter will attempt to address sensors both as separate units and as a component within the
complex infrastructure health system.
5.1.2 This Chapter

This chapter will explore categories of sensors that are used in the infrastructure first. We then
review briefly to mechanisms and theories that govern the behavior of major sensing mechanisms.
We believe that some understanding of the basics of sensors is important for optimal utilization of
sensors in any experiment or project. After that we turn our attention to the utilization of sensors
175
The decisions
might affect the System: Infrastructure
state of the system state
Decision making Sensing

(Based on sensing results) (including human sensing,
i.e., inspectors)
STRID and DMID

activities might also Information from
utilize sensing sensors
results
FIGURE 5.1 Sensing paradigm.
in structural health monitoring (SHM). We will balance the theory of sensors with the practical
applications of those sensors. The following two sections will be devoted to two emerging sensing
techniques: fiber optics sensors (FOR) and wireless sensing. These will be followed by a section
discussing smart structures (SS). It is obvious that structures are only smart because they can sense
the surrounding environment. Two practical sections will follow: a section describing parameters
that can affect choices of sensors for a particular project and optimum sensor locations (OSL).
We then offer a guide to help practitioners to choose appropriate sensors. The final section of this
chapter is devoted to another emerging sensing field: remote sensing (RS). We immediately note
that RS is not the same as wireless sensing. Wireless sensing places sensors in contact with the
system and transmit the measurements (information) wirelessly. RS performs the measurements
remotely: with no contact with the system. The measurements of remote system can then be trans-
mitted either in a wired (tethered) mode, or in a wireless mode. Figure 5.2 shows the composition
of this chapter.
5.2 SENSOR CATEGORIZATIONS

Sensing is the first and most utilized component in SHM and infrastructure health in civil engineer-
ing (IHCE). There has been countless number of manuscript, research, and development regarding
sensors, measurements, and their use in SHM as well as civil infrastructure. For our purposes in this
volume, we start exploring the subject of sensing by considering logical sensor categorizations. This
is essential step if we desire to quantify such concerns as optimum sensor selection (OSS), OSL, or
any other class of decision making processes that involve using sensors. We propose, to categorize
sensors into four categories in cascading order as follows (see Figure 5.3):
Scope: This is a fairly qualitative category that we deemed necessary to differentiate between non-
destructive testing (NDT) and SHM fields. Such differentiation is not well defined, however, it does
exist. We propose to subdivide sensing within the SHM field into two categories: a general sensing
and an NDT sensing categories. General sensing categories are used to sense conventional behavioral
metrics such as temperature, displacements, strains, humidity, and so on. They also have the important
Sensors and Infrastructure Health 177
Sensors and measurements in infrastructure
Sensor SHM and Emerging Geometry of

overview sensors technologies sensing
SHM and
Sensor Wireless Selection of
sensors
categorization sensors sensors
(General)
Optimum
Sensor Fiber optic Smart
sensor
mechanisms sensors (FOS) structures
locations
Practical
Remote guide to
sensing sensor
selection
FIGURE 5.2 Composition of sensors chapter.
General sensor categories
Type of
Behavior/ measurements
Scope mechanisms Technology
(ToM)
• Ultrasonic • Electrical • Strains

• NDT/specific • Acoustic • Optical • Stress waves
emission • Etc. • Displacements
• Etc. • Temperature
• Humidity
• Electric • Etc.
• SHM/general resistance
• Voltage
change
• Etc.
FIGURE 5.3 Sensor categories.
distinction in that they usually do not aim to identify damage in the system. As such, they are general
in scope, and fairly limited in objectives. NDT category offer the opposite properties: they are more
limited in scope, but they are more general in objectives. For example: acoustic emission (AE), as an
NDT technology has a very well-defined scope: measure stress wave emissions that emanate from
damages in systems to locate the damage. The objective of AE goes far beyond just measuring stress
emissions: its objective it to identify damage attributes in the system. We note that in SHM field,
a mixture of NDT and general purpose sensing usually offer the most efficient project solutions.
Technology: This level of categorization addresses the technological basis of sensors. As can be
seen from Figure 5.4, there are several technologies that are used for sensing.
Categorization of sensors by technology
Sensor technology Sample application
Mechanical Accelerometers
Optical Fiber optics
Acoustic Acoustic emmision
Chemical Chloride rate
Electromagnetic Magnetic pulse
Thermal Theromgraphy
Damage of stay
X-ray
cables
RF GPS measuremnts
Pattern recognition
Digital video
of damage
Other.. Miscellaneous use
FIGURE 5.4 Categorization of sensors by technology.
Behavior/Mechanism: Within each technology, there exists several ways to sense a physical
phenomenon. For example, electric resistance, and electric voltage change are different mechanisms
within electric circuit technology that can be used for sensing. In the NDT field, electromagnetism
technology offers different sensing mechanisms such as eddy current and magnetic particle. For
wide-ranging discussion of different sensing mechanisms for general purpose sensing, see Northrop
(2005). ASNT (1996) offers in-depth review of NDT-based mechanisms.
Type of Measurement: Another important sensing category is the ToM category. ToMs can bay
any physical metric that is of interest to an SHM project. They can be strains, velocities, displace-
ments, temperature, AE, wave length, or corrosion rates. Figure 5.5 shows some examples of ToMs.
5.3 BASICS OF SENSOR BEHAVIOR

Sensing mechanisms can be categorized into several classes. Among the popular classes are electri-
cal, optical, and chemical, etc. Within each class of sensing mechanisms there are usually several
subclasses. In this section, we discuss electrical-based sensing mechanisms and optical-based sens-
ing mechanisms. Those mechanisms are traditionally not categorized within the NDT field. We
conclude the section by discussing the distinction between the classes of sensing mechanisms that
are presented here and the NDT mechanisms that are explored in Chapter 8.
Categorization of Sensors by ToM
ToM
Strains
Position
Displacement
Velocities
Accelerations
Pressure
Temprature
Force / Reactions
Tilt
pH Level, corrosion
rate
Other.
FIGURE 5.5 Categorization of sensors by type of measurement (ToM).
5.3.1 Voltage-Generating Sensors

In some situations, when a physical environment changes, an electric magnetic field (EMF) is gen-
erated. Perhaps the most popular example is piezoelectric materials that generate EMF when sub-
jected to certain changes. By relating the generated voltages to the properties of these materials,
the physical parameters that caused the change can be measured. We discuss the basic principles of
piezoelectric sensors first, then summarize other types of voltage-generating sensing techniques.
5.3.1.1 Piezoelectric
Figure 5.6 shows the basic concept of piezoelectric strain transducer. When a transducer is in its
baseline stage, its thickness is h. The voltage across the transducer is V0. When a force is applied to
the transducer, it changes its thickness by ∆h. Such a change will generate a change in the voltage
across the transducer to V1. The relationship between the change of thickness of the transducer and
the voltage can be shown to be as follows:
V1 = g E Δh (5.1)

Where g and E are the voltage sensitivity and the modulus of elasticity of the material, respectively.
From Equation 5.1, by measuring V1, the strains in the transducer can be evaluated as
Δh
e= (5.2)
h
Finally, the strain measurement as a function of voltage is
⎛ 1 ⎞
e=⎜ V1 (5.3)
⎝ g E h ⎟⎠
(a)
+
Voltage across h
transducer V0
−
Piezoelectric transducer: not strained
(b)
Applied force When the transducer is
Initial surface strained, the voltage
across it changes
+
∆h
Voltage across
h - ∆h
transducer V1
−
Piezoelectric transducer: after applying a normal force
FIGURE 5.6 Principles of piezoelectric transducers (a) before straining, and (b) after straining.
There are many other uses of piezoelectric materials in SHM. This includes foil strain gauges and
accelerometers. Both are discussed later in this chapter.
5.3.1.2 Other
Among other systems that relate EMF voltages and physical parameters are thermocouples that
generate EMF when subjected to changes in temperatures. Certain accelerometers also generate
EMF voltages when subjected to accelerations. These accelerations can be recorded by observing
the voltage in the circuit. Velocities can also be measured using generated voltages in linear velocity
sensors (LVS).
5.3.2 Resistive Sensors

Resistive type sensors rely on changes in electric resistance of a given circuit for measuring the
desired physical parameter. If the changes in the electric resistance ∆R, is small compared to a
baseline resistance R0 that is, ∆R/R0 ≫ 1, then the resistance changes are linearly proportional to
the voltage change ∆V in the circuit. The basic relationship is
ΔR R0
= (5.4)
ΔV V0
In Equation 5.4, V0 is a baseline voltage. By measuring the voltage changes it is possible to infer
the changes in the desired physical parameter. For example, it is known that electric resistance R is
related to temperature T such that
R = R(T ) (5.5)
A change in temperature ∆T will generate a change, in electric resistance of ∆R. Such a change will
produce a change in voltage ∆V. If the inverse of relationship Equation 5.5 is known as
T = T ( R) (5.6)
then the change of temperature can be measured as
ΔT = T ( R ) = T ( ΔV , R0 ) (5.7)

Resistive strain sensors operate with similar principles. Since strains are related to changes in length,
which can be related to electric resistance, it is possible to develop a relationship between strains and
changes in voltages that can be measured in electric circuits. In addition, resistive sensors are also
used to measure changes in humidity, linear position, and angular position.
Several electric circuit topologies are used for this purpose, among them are Wheatstone and
Kelvin bridge circuits.
5.3.2.1 Wheatstone Bridge

Wheatstone bridges are a type of direct current (DC) electrical circuit that is used to make precise
measurements of sensitive metrics, such as strains in structural components. The basic operational
concept of the Wheatstone bridge is the null concept. In a null concept, the input and output voltages
are balanced so as to produce a null voltage across the bridge. Figure 5.7 shows the basic operational
concept of a null DC instrument.
Utilizing the null DC concept, the Wheatstone bridge (Figure 5.8), is a popular technique that is
used for SHM measurements, for example, strain foil transducers. In a Wheatstone bridge circuit,
the input and output voltages VINP and VOUT, respectively, are balanced so as to produce a null volt-
age across the circuit such that
V0 = VINP − VOUT = 0 (5.8)

From Figure 5.8, the resistance ratio of the bridge is
R1 R2 (5.9)
=
R4 R3

Null DC instrument for measurments
Change in target
Change properties of
behavior (strain level
null instrument until a
changed)
balance (null
measurement)
Voltage across null between input and
instrument changes output
Examples of
Null DC By calibrating the
Instruments: change in the
• Wheatstone instrument properties,
bridges the change in target
• Kelvin behavior can be
bridges quantified
FIGURE 5.7 Operational concept of a null DC instrument.

R1 R4
VINP
Null detector
VOUT
R2 R3
FIGURE 5.8 Wheatstone Bridge configuration.
The relationship between VINP and VOUT can be shown to be
R1 R3 ⎛ Δ R1 Δ R2 Δ R3 Δ R4 ⎞
VOUT = VINP − + − + (5.10)

( R1 + R2 ) ( R3 + R4 ) ⎜⎝ R1 R2 R3 R4 ⎟⎠

Where R1, R2, R3, and R4 are the resistance magnitudes in the Wheatstone bridge components, and
∆R1, ∆R2, ∆R3, and ∆R4 are the incremental changes in the resistance. If one of the components, say
R1, represents a sensor that undergoes a change ∆R1 that is proportional to a strain (see foil strain
sensors section later this chapter), while no other components change their resistances, that is,
Δ R2 = Δ R3 = Δ R4 = 0 (5.11)

Equation 5.10 can be used to obtain the desired strain level. We note that the Wheatstone bridge
can be used in more complex sensing arrangements, such as calibrating temperature drift effects for
some sensors, by careful applications of Equation 5.10.
5.3.3 Change of Capacitance

Capacitance is a property that is a function of temperature and humidity. Because of this, several
types of humidity and temperature sensors are based on measuring changes in capacitance and
detecting corresponding changes in humidity or temperature accordingly, (see Northrop 2005 for
more details).
5.3.4 Linear Variable Differential Transformers (LVDT)

This class of sensor is based on measuring magnetic fields that result from relative movements
between coils that are arranged in certain manner. The relationship between the magnetic field and
the relative motion can then be used to compute the required relative motions. This class of sen-
sors is used to measure displacements, velocities, and accelerations. They are widely used in SHM
projects. For example, they are used to measure deflections of bridge members by converting mea-
sured displacements to equivalent voltages which can be recorded using data acquisition systems.
Creation of a stable support to serve as a datum is a major problem when measuring deflections on
bridges. Field measurement of deflections is not easy, even though they can be related to member
stiffness more directly than other parameters such as stresses, strains, and inclinations. However, in
field testing of bridges this is not always practical and more than often not possible. Some examples
are given in the next section.
5.3.5 Optics
Optics-based sensors can be subdivided into two general categories: FOS, and mechano-optics sen-
sor. Fiber optic sensors (FOS), as the name implies, are based on the passage of light waves through
fibers, and through the interaction of the light with the surrounding media, the desired measure-
ments can be obtained. Due to their importance and emerging wide use in SHM field, FOS will be
described in detail later in this chapter.
The mechano-optics sensors are based on transmitting light onto the object of interest. The reflec-
tion, or transmission, of the light from or through the object will be affected by the motion of the
object. By processing the reflected light, some information regarding the motion of the object can be
inferred. This type of sensing is explored further in the remote sensing section of this chapter.
Another type of mechano-optics sensors is based on transmitting a light beam (laser) on the
object. The light will interact with the object; such an interaction will generate ultrasound waves
within the object. These waves can be body waves, surface waves, or plate waves (in case of thin
plate systems). The waves are then detected by a receiver that would infer properties of the object,
such as presence of defects in the path of the ultrasound waves. This type of sensing is also explored
further in the remote sensing section of this chapter.
5.3.6 Other
There are numerous other types of sensing mechanisms. Some of those are based on chemical or
radiating wave physics. We discuss the chemical sensing for monitoring corrosion process later this
section. Radiating wave sensing is discussed in detail in Chapter 7.
5.3.7 NDT-Specific
There is another large class of sensing mechanisms that are used in SHM projects. These sensing
mechanisms are used in the NDT field. Among these mechanisms are ultrasonic sensing, electro-
magnetic sensing, thermal sensing, radiating wave sensing, and AE. What differentiate these sens-
ing mechanisms, and the whole of NDT field from the type of sensing mechanisms that is discussed
in this section is that the NDT mechanisms is more integrated. Most of the sensor mechanisms of
this section concern themselves only with the ToM they are designed for. For example, an LVDT
accelerometer will measure a given acceleration wave form only. The analysis of the wave form is
left to the tester to perform depending on the objectives of the sensing project. Meanwhile, NDT-
type mechanism measure a particular ToM (such as stress wave form) and tries to analyze it and
detect damages, if any: all in one integrated method.
Recognizing this subtle, but important difference, we decided to discuss many of the impor-
tant NDT sensing and evaluation methods in Chapter 7, on damage identification, instead of this
chapter.
5.4 SENSOR MEASUREMENTS IN SHM

5.4.1 Overview
Previous section presented different types of sensing mechanisms. This section explores sensors
from ToM view point. Important ToMs for civil infrastructure are shown in Figure 5.5. We explore
all of them in this section, including types of sensors used and the underlying physical behavior of
them. We also point to the fact that there are several different ways that each ToM can be measured.
This is important point, since it can, and should, be used when choosing sensors for specific project,
as will be explained in Section 5.8.
5.4.2 Strain Sensors

5.4.2.1 General
Strain gages are widely used to measure strains in such structural components as girders, rebars,
and concrete decks. Measured data can then be employed to calculate physical parameters, evaluate
bridge load capacity, and monitor structural safety. Environmental conditions may play important
roles in bridge testing. For example, self-temperature compensating type gages are generally pre-
ferred when wide temperature variations are expected. Gages are welded or attached with adhesives
to the structure. Two or four arm gages are also favored as they increase the signal to noise ratios,
and several gage combinations are sometimes adopted to measure different components of strain
(such as shear, bending, and axial). Normally, strain gages require DC input in the range 0–15 volts,
and give outputs in the same range. Measured voltages are then converted to strains/stresses using
appropriate engineering relations.
5.4.2.2 Foil Strain Gages

Strain gages are the sensors which are very widely used in civil engineering testing and research
to measure the structural behavior under loads. Figures 5.9, shows schematics of foil strain gage,
Figure 5.10 shows schematics of piezoelectric foil gages Figure 5.11 shows a typical strain gage in
the field. Figure 5.12 shows a strain gauge in an SHM experiment. Since strain is directly related to
stress up to elastic limit and most structures are designed to operate within elastic limits, stress is
very critical to failure of civil structures. Bonded strain gages became very commonly used sensors
in the world. Bonded strain gages become integral to the structure they are bonded to and, if applied
properly, will undergo same strains as the parent structure and gives direct indication of the strains
experienced by the parent structure. Ideal strain gage characteristics depend on the type of applica-
tion and nature of the strain field experienced during the testing. For more on foil strain gauges see
Perry and Lissner (1962) and Hannah and Reed (1991).
The change in electrical resistance in the strain gage is related to strain; and this principle using
simple Wheatstone bridge circuit forms the basis for strain gage operations. Strain gages primarily
form one, two, three, or all legs of the Wheatstone bridge as described in the previous section. In
general, in a balanced bridge, the change in strain generally translates to change in current through
the gage or the change in voltage across the gage resulting in an unbalanced Wheatstone bridge.
Substrate
Etched foil
Solder tabs
Wire leads
Foil strain gauge
FIGURE 5.9 Concepts of foil strain gauges.

Top Electrode
Wire leads
Piezoelectric film
Bottom electrode
Substrate
Elevation view
Substrate
Etched foil
Bottom electrode Top electrode
The piezoelectric
Wire leads film is wedged
between top and
bottom electrodes
Plan view
Foil piezoelectric gauge
FIGURE 5.10 Concepts of piezoelectric foil gauges.
FIGURE 5.11 Typical foil strain gage.
This unbalance can be translated into appropriate strain using the electrical principles and strain
gage characteristics following the relation
ΔR
= Ge (5.12)
R
Where ε is the desired strain, R is the baseline resistance, and ∆R is the measured change in resis-
tance that produce a null balance in the Wheatstone Bridge. The gauge factor G is a constant that
depends on the sensor configurations.
FIGURE 5.12 Foil strain gage in an SHM experiment. (Courtesy of New York State Department of
Transportation)
Advantages
• Foil gages are easy to bond or weld

• Available in small to big sizes. Hence, ideal for measuring stress concentrations as well
as in applications where an average strain is needed (in Reinforced concrete applications).
Can be grouped easily to get tri-axial strains
• Become integral part of the structure and direct indication of the strain in parent structure
• Can give both static and dynamic strains
• Not expensive
• Low mass
Disadvantages
• Needs good understanding of the strain gage principles for good interpretation of data
• Where environmental conditions, such as temperature, rapidly vary, comprehensive under-
standing of strain age principles and the effect of environment on structural behavior are
needed to get good data
• Instrumentation required could be significantly expensive
• Not very suitable for long-term continual use. Drift can be a big issue due to several factors
that cause unbalance of the Wheatstone bridge circuit
• Not reusable
• Needs constant power supply
• Needs expertise for good and reliable installation of gages
• Needs protection from environment for long-term installations
• Needs extensive surface cleaning and preparation for good bonding
5.4.2.3 Demountable Strain Gages/Strain Transducers

Demountable strain gages in general employ number of bonded strain gages, typically with a full
Wheatstone bridge to eliminate dummy gages, into a fixed enclosure such as a ring or a plate. This
makes the gage reusable (see Figures 5.13 and 5.14 for typical BDI transducers, a demountable strain
gage) and hence the name demountable strain gage. These gauges are attached to the structure at
FIGURE 5.13 Demountable strain gauges.
FIGURE 5.14 Close up of a Demountable strain gauges.
two ends, with a known distance between them, and the change in gage length is related to strain
in the structure.
Advantages
• Eliminates the need for extensive surface preparation

• Easy to fix and use
• Reusable
• Needs less expertise for installation and operation
• Eliminates the need for dummy gages and balancing circuits; and hence instrumentation
can be cheaper
Disadvantages
• Since these use strain gages in their assembly, some of the inherent disadvantages with
strain gages are not eliminated
• Reliable calibration is required

• In general, needs big gage lengths and cannot be easily installed at tight locations and
closure to connections, where stress concentrations can be important
• Since, they are not integral to the structure, the structural deformations can create errors in
response, if not calibrated and understood well
• As these are relatively big, it is not easy to install a group of gages together to measure
stresses in several directions at same location
• Use is generally limited to shorter durations
• Mismatch of structural material and gage materials can create issues with temperature and
other environmental factors. Hence, good understanding of structural behavior and sensor
behavior are essential
5.4.2.4 Vibrating Wire Gages

Vibrating wire strain gages use the principle that a wire vibrates when held in tension and a force
is applied. Any relative movement of the fixity at ends (supports) will effect a change in measured
frequency of vibration, and can be used to measure the movement. This principle is used in strain
gages, load cell piezometers, and crack meters, but predominantly used for measuring strains in
infrastructure area.
The natural frequency of a vibrating wire f can be related to strain (ε) as
e = K f 2 (5.13)
where K is a constant factor for the sensor and Gage-Technique (2011) depends on (a) wire density,
(b) gravity constant, (c) wire length, and (d) wire modulus of elasticity. Thus, any relative change in
strain can be measured knowing the frequency at both initial and current states using the following
equation:

(
Δ e = K f12 − f22 ) (5.14)
Many of the vibrating wire gages are designed to be fixed to the surface of the structure requiring
strains using epoxy or welds or embedded inside. These normally have two end blocks with a wire
between them. The end blocks are attached to the surface of interest. When the surface deforms or is
strained, the end blocks move relative to each other. This causes change in the length of the wire and
thus the natural frequency of the wire. The data loggers measure the frequency and thus associated
strain. These gauges are used for long-term static applications due to the advantages they offer. More
details on demountable strain gauges can be found at Perry and Lissner (1962), Hannah and Reed
(1991), Hag-Elsafi et al. (2003) and Gage-Technique (2011) (Figures 5.15 and 5.16).
Advantages
• The response from these gages is analogue . Hence, the resolution is infinite and is limited
by the devise used to read and display the data
• Immune to electrical noise
• Tolerate wet environment
• Can transmit data to long distance without loss of accuracy
• Stable for long-term applications as there is no drift
• Can compensate for temperatures with a built-in temperature measurement
• Relatively rugged and durable
• Easy to install
FIGURE 5.15 Vibrating wire strain gauges (surface mounted).
FIGURE 5.16 Vibrating wire strain gauges.
Disadvantages
• Not suitable for dynamic measurements

• Restively bigger than foil gages
• Data analysis could be tricky, in long-term applications, if structural environment in not
understood well
5.4.3 Position
Linear variable differential transformer gives position of an object they are mechanically attached
to. This converts position into DC voltage and can be read using an appropriate reader; and does
not require physical connection to the extension like a potentiometer. The extension valve shaft (or
control rod) in an LVDT moves between primary and secondary windings of a transformer causing
inductance between the two windings to vary. This varies the output voltage proportional to the
position of the valve extension. The theoretical principles of the LVDT are shown in Figure 5.17.
(a)
Supply coils
Secondary coil #1 Secondary coil #2
Housing
Motion of the core
to be measured from Insulating foam
an arbitrary position X0
Ferromagnetic moving
core
X0
Output coils
Output coils
X0 Supply coils
Schematic cross-section in an LVDT transducer
(b)
Core at X1 Core at X0 Core at X2
Output
voltage
X1
X2 Displacement - X
X0
Relationship between position and voltage
FIGURE 5.17 Principles of LVDT transducers (a) schematics, and (b) basic principles.
Figure 5.18 shows a typical LVDT unit. Figure 5.19 shows an LVDT unit that is installed to measure
displacements during an SHM project. Several online references are available such as Sensorland
(2011), Engineersedge (2011), Singer (2011), and Macrosensors (2011).
Advantages
• Very reliable
• Rare failures
• Can measure very small movements (as small as 0.01 microns)
• Rugged
Disadvantages
• High variation in ambient temperatures can affect quality and hence long-term measurements
• Setup on field structures could be difficult
• Operating ranges could be fixed and limited
• Limited sensitivity
FIGURE 5.18 Typical LVDT unit.
FIGURE 5.19 Field application of LVDT unit for measuring displacement. (Courtesy of New York State
5.4.4 Accelerometers (Angular and Linear)

Accelerometers are sensors that utilize different concepts for measuring accelerations. We explore
two different types of accelerometers: LVDT and piezoelectric types
5.4.4.1 LVDT-Based Accelerometers

Linear variable differential transformers based accelerometers, (see Figure 5.20), are based on the
following equation of motion:
M
x1 + C Δ x + K Δ x = 0 (5.15)

with
Δ x = x1 − x0 (5.16)

Motion of the mass

is = X1 Moving mass Supply coils
X1 Secondary coil #1 Secondary coil #2
Housing
Spring
Ferromagnetic
moving core
Output coils Output coils

X0
Supply coils
Motion of the housing
is = X0
LVDT accelerometer
FIGURE 5.20 LVDT-based accelerometer basics.
The mass, damping and stiffness parameters of Equation (5.15) are M, C, and K, respectively. They
are sensor properties that are assumed to be known. The relative motion between the mass and the
base of the sensor ∆x is proportional to the measured LVDT voltage V as
V = Κ Δx (5.17)
In Equation (5.17), K is a known LVDT factor. Thus, by measuring the output voltage, the accelera-
tion can be obtained from Equations (5.15), (5.16), and (5.17).
5.4.4.2 Piezoelectric Sensor Accelerometers

Piezoelectric elements-based accelerometers are designed to produce electric signal proportional to
forces induced due to vibration of the structures they are attached to. Quartz is a popular piezoelec-
tric element in many commonly used accelerometers due to its long-term stability, and temperature
stability, but requires voltage amplifiers due to the high voltage sensitivities they exhibit. Voltage
amplifiers limit the signal to noise ratios but allow high level of vibration measurements. Polarized
lead zirconate titinate (PZT) is another alternative material used in accelerometers, PCB (2011) but are
limited to low level vibrations. These use quieter microelectronic charge amplifiers and also require
frequent calibration. The selection of an accelerometer depends on the amplitude range, frequency
range, environment or in-service conditions, and duration of use. Figure 5.21 shows the basic concepts
of piezoelectric accelerometer. Figures 5.22 and 5.23 show the use of piezoelectric accelerometers in
lab testing and in-situ testing, respectively. See Northrop (2005) and PCB (2011) for more on piezo-
electric accelerometers.
Advantages
• Easy to use
• Easy to install
• Wide variety
• Relatively low cost
Disadvantages
• Requires thorough understanding of structural vibration theories

• Do not directly correlate to structural health
• Turn-key systems are uncommon for civil applications
Mass is
stationary
relative to the case
Piezoelectric crystal
Housing
Very stiff spring
Motion of the case will induce

acceleration of the mass. The
acceleration of the mass will
induce force in the piezoelectric
crystal, thus generating
proportional electric voltage
Motion of the housing X0

is = X0
Piezoelectric accelerometer
FIGURE 5.21 Concepts on piezoelectric accelerometers.
FIGURE 5.22 Piezoelectric sensor on concrete specimen. (Courtesy of New York State Department of
Transportation.)
• Needs experienced personnel for good quality results

• Long-term installations are hard to maintain
• Can be damaged relatively easily
• Accompanying equipment is relatively expensive
5.4.5 Velocities (Angular and Linear)

Velocities can be measured indirectly by using an LVDT accelerometer as in Figure 5.20. The
accelerations are computed using Equations 5.15, 5.16, and 5.17, then the velocity can be estimated
FIGURE 5.23 Piezoelectric sensor under steel girder.
by integrating the measured accelerations as

x1 = (∫ x dt ) + x
t1
t0 1 1 t0

(5.18)
with the initial velocity as x1 t .

0
A direct linear velocity transducer/sensor (LVT/LVS) method of computing velocity is shown in
Figure 5.24. It is based on slightly different form of Equation 5.17 such that is directly relates the
voltage to the relative velocity between the coil and the magnet � x . The basic equation becomes
V = Κ VEL Δ x (5.19)

where K VEL is a known LVT/LVS factor and
Δ x = x1 − x 0 (5.20)

Thus, by measuring the output voltage, the relative velocity can be obtained from Equations 5.19
and 5.20.
5.4.6 Displacements (Angular and Linear)

Linear variable differential transformers concepts can be easily used to measure displacements.
The relative displacement between a baseline x0 and a displaced object x1 as in Equation 5.16 can be
computed by measuring the LVDT generated voltage as in Equation 5.17.
5.4.7 Force
Load cells are mechanical devices used to measure naturally or mechanically induced loads, the
structure is subjected to (see Figures 5.25 and 5.26). Load cells convert the load or force into elec-
trical output. There are several types of load cells and are characterized by type of output signal
generated (pneumatic, hydraulic, or electric) and the weight detection mechanism (bending, shear,
tension, etc.). These include mechanical load cells (such as hydraulic or pneumatic load cells), strain
gage based load cells (such as shear beam, ring and pancake, bending beam load cells), and other
(a) Housing
Coil
Motion of the core
relative to the coils Insulating foam
will generate voltage
proportional to the
core speed Ferromagnetic
moving core
Housing
V0 (length is about twice
Coils the stroke length of
moving core)
Schematic cross-section in an LVT /LVS transducer
(b)
Core velocity V1 Core velocity V0 Core velocity V2

Output
voltage
V1
V2 Velocity - V
V0
Relationship between velocity and voltage
FIGURE 5.24 Concepts of LVT/LVS instruments. (a) schematics, and (b) basic principles.
FIGURE 5.25 Load cell.

FIGURE 5.26 Load cell cable.
load cells (such as fiber optic or piezo-resistive load cells). For more on load cell sensing, see Perry
and Lissner (1962), Hannah and Reed (1991), Hag-Elsafi et al. (2003), and Omega (2011)
But, in recent days, most commonly used load cells are transducers built based on strain gages
and principles associated with them. For example, if a strain gage is attached to an axial member
(see Figure 5.26), by knowing the modulus of elasticity, cross-section, and the strain the member is
subjected to, the load can be calculated as follows:
P = e ( E A) (5.21)

Note that P is the desired load. The strain, modulus of elasticity, and area of cross-section are ε, E,
and A, respectively. Thus, as long as the load applied keeps the axial member in the linear rage of
modulus of elasticity, the load cell can be used to measure the load by multiplying the strain with a
constant calibration factor. Many load cells also use a full bridge (i.e., four gauges, usually two in
tension and two in compression), to avoid the use of dummy gages to complete the balancing circuit,
to avoid temperature compensation, and to obtain maximum sensitivity. These can also be used
such a way that the electrical output to load can be increased Omega (2011).
In bridges, load cells are normally used in bridge testing to measure naturally or mechanically
induced loads for which structures are being subjected. Figure 5.27 shows a high capacity, hydrau-
lic load cell that utilizes the change in fluid pressure in estimating the applied load level. Again,
conversion of output voltages to loads is direct and is accomplished in a similar manner to that for
inclinometers.
Advantages
• Relatively small in size, as small as a button
• Good linearity within the elastic range
• Can measure tension and compression
• Can be made weather resistant
• Low profile and can be integrated into restricted areas
• Connections can be custom made
• Offer good long-term stability
• Low cost
• Low thermal effects
Force F
When the plate deform

under the applied force,
the fluid pressure Thin plate
Fluid
increases
By measuring the
fluid pressure, the
applied force can
be computed
LVDT accelerometer
FIGURE 5.27 Hydraulic load cell.
Disadvantages
• Issues inherent to strain gages still exists, that is, drift and so on
• Need to be water tight to avoid moisture and humidity as they can impact the bonding
between the strain gages and the material they are affixed to
• Low electrical output associated with strain gages may necessitate external amplification
of the signal (between the gage and the instrumentation)
5.4.8 Inclinometers
Inclinometers are used to measure inclination of various structural elements due to distresses in the
system (see Figure 5.29). They are mainly used to evaluate fixity of bridge girders at supports and
to monitor long-term movements of piers, abutments, and girders for general assessment of struc-
tural integrity. Normally, inclinometers require DC input in the range 5–15 volts and give outputs
in the same range. Measured voltages are then converted to inclinations using their sensitivity.
Furthermore, this class of sensors are used to measure inclination or slope or elevation of elements
they are attached to with respect to gravity. These are also known as tilt sensors or slope sensors or
grade indicators. There are several types of inclinometers. Simple, hydrostatically based inclinom-
eters are based on hydrostatic leveling of water, as shown in Figure 5.28. The angle of incline, θ is
measured as
H1 − H 2
q= (5.22)
L
Advantages
• Gives absolute slope with respect to gravity or reference
• No need of interpolation
• Easy to attach to concrete or metal structures
• Very sensitive
(a) Housing
Air
Water
Level baseline
Inclinometer: at level position
(b)
H1
H2
Inclined leve
l
Inclinometer: at rotated position
FIGURE 5.28 Basis of hydrostatic inclinometers (a) level position, and (b) rotated position.
FIGURE 5.29 Typical electrically based inclinometer.

Disadvantage
• Very sensitive in small range, making it hard to install

• Do not react very fast. Hence, not possible to make dynamic measurements
• Installation method is very important and can be difficult
5.4.9 Corrosion
Corrosion sensors, (see Corrosionsource 2011) are normally used to measure the rate of corrosion in
steel or concrete. These are relatively new to infrastructure arena and are used for long-term mea-
surements, in most cases to monitor anticorrosive performance of new materials. Corrosion monitor-
ing systems/sensors vary in complexity and size. But, potential cost savings from the implementation
of these in the future could be significant to infrastructure arena. The technologies vary from use of
polarization resistance to time domain reflectometry. Size varies from the size of a pebble to a small
box depending on the technology. Many are coupled with temperature and other measurements.
A popular corrosion measuring method is the half cell approach, ASTM 1987. Half cell sensing
is used to measure corrosion levels as well as possibility of corrosion development in reinforcing
steel in concrete structures. It is well known that the passive oxide film with high pH level that cover
steel rebar is meant to protect the rebar from corrosion process. However, when concrete structures
are subjected to environments with high chlorine levels, such as salt water, these chlorine ions pen-
etrate reinforced concrete to breakdown the protective oxide film and initiate the corrosion process
(rust forming) of the now unprotected steel.
The half cell sensing method measures the voltage drop between a known reference point and
any other point at the reinforced concrete surface. Such drop in electric potential can be mapped as
contours. The concentrations of the contour lines indicate areas of high probability of corrosion.
Some known disadvantages of the method are
• Can’t be used for post-tensioned construction, since the tendons do not come into contact
with the concrete itself, rendering the half cell potential readings not indicative of corro-
sion condition of the tendons themselves
• Since the half cell method results are affected with humidity levels, its results can be inac-
curate in dry environment
• Compensation for concrete cover might be needed
• The sensors give localized corrosion rate. Hence, may not represent the entire structure
and care should be given in choosing appropriate locations
• Temperature and humidity can influence the corrosion rates
• Installation is not very easy for some of these
• No standards or formal guidelines for field application
• Many are intrusive
The advantages of the half cell potential are
• Simplicity and ease of implementation

• Can be embedded inside, or mounted on the surface of the structure (see, respectively,
Figures 5.30 and 5.31).
• Can be monitored over long time periods; this feature makes the method very suitable for
SHM projects.
• Gives rate of corrosion directly
• Versatile and reliable
• Continuous monitoring can be accomplished
• Many are passive
• Makes remote corrosion monitoring possible
FIGURE 5.30 Embedded half cell potential sensor. (Courtesy of New York State Department of
Transportation.)
FIGURE 5.31 Half cell potential sensor after installation.
5.4.10 Pressure
There are several methods to measure pressure. Many of the concepts that are discussed earlier in
this section can be used to measure pressure. For example:
• Piezoelectric pressure sensors use the generated voltage that results from the strain ε of a
piezoelectric material to relate the pressure p = P/A to (see Equation 5.21) to the generated
voltage from the material strain.
• The LVDT principle can be used to measure pressure in LVDT pressure sensors. When a
coil is displaced relative to a magnet, the resulting voltage can be measured. The displace-
ment can be related to the pressure of interest.
• Vibrating wire pressure sensors are used for highly transient pressures, such as pressures
from explosives (see Equations 5.13 and 5.14). When a pressure is applied to the sensor, the
natural frequency of the sensor changes, by measuring the change, similar to Equation 5.14,
the magnitude of the pressure can be estimated.
For detailed pressure sensing discussion, see Norton (1989).
5.4.11 Temperature
Temperature measurements are often needed in SHM projects. Corrosion, fatigue, FRP as well
as general thermal expansion of structures are of interest to decision makers. There are numerous
methods of measuring temperature effects. For example:
• A biomaterial temperature sensor that uses two different materials. When the temperature
changes from the baseline temperature, sensor deforms according to the differences of
the thermal expansion coefficient and the length of each material within the sensor. The
deformation can be used to detect the temperature changes.
• Electrical resistance thermometer uses the relationship between the temperature and the
electric resistance of materials to detect temperature changes. It is used in applications such
as underwater submerged temperature measurements as well as pipelines, tanks, and air.
• Thermocouples thermometers use electrical properties of two dissimilar materials to accu-
rately measure temperature. Thermocouples thermostats are widely used in many forms.
• Pyroelectric thermometers are based on using pyroelectric materials that generate electric
charge when subjected to heat.
Other forms of thermometers are fiber optic temperature sensors and infrared (IR) thermometers.
For extensive discussion of thermometers, see Webester (1999).
5.4.12 Other Sensors in SHM

There are several other types of sensors that are used in SHM projects. Some of these sensors and
their use in SHM are shown in Table 5.1.
The reader is referred to more comprehensive sensing references such as Northrop (2005), Norton
(1989), and Webester (1999).
5.4.13 Advanced Sensing Technologies

5.4.13.1 Microelectromechanical systems (MEMS) Technology
MEMS is an emerging sensing technology that holds great promise for SHM in civil infrastructure.
An overview of MEMS technology is offered by El-Sheimy et al. (2004).
TABLE 5.1
Use of Other Types of ToM in SHM
ToM SHM Use
Humidity Corrosion detection and mitigation
UV exposure FRP behavior
Radiation Location of hidden defects in concrete structures
Fluid flow Scour investigation
5.4.13.2 Friction-Based Strain Sensing

Utilization of this new type of strain sensors is described by Ojio et al. (2004). The authors used the
new sensor in monitoring bridge behavior in Japan.
5.4.13.3 Scour Sensing

The scour hazard has been and is still a major concern for bridge owners and it contributes to more
failures than other hazards. Measuring scour depth is still hard and is important where conventional
analyses are not available in several cases (tidal scour is an example). Mercado and Rao (2006)
describe the development of a pneumatic scour detection system (PSDS) designed to monitor scour
depth on an as-needed or real time basis during either normal or extreme flood conditions (see
Figure 5.32). It is based on differential resistance to air (or liquid) flow through an array of stain-
less steel porous filters. This system is intended to eliminate common deficiencies found with other
conventional systems, that is, debris accumulation around piers masking sound signals, and strong
turbulence and impact of large debris during floods.
Pressure versus time data is recorded at filters situated at various depths of the probe driven into
a river bed. Scour depth is determined by measuring the resistance to air flow through the filters as
a function of the nature of the material against which the filter is embedded. Low resistance indi-
cates water, where as high resistance indicates that the filter is sealed against competent soil (see
Figure 5.33).
The system was tested in the laboratory and then was demonstrated in the field with an excavated
pit. The laboratory and simulated field test results showed that system is consistent and produces
clear results.
5.4.13.4 Innovative ETDR Sensing

Electric time domain reflectometry (ETDR) sensing technology is used by Chen (2004) to develop
crack detection sensors. The method is based on the propagation of electromagnetic waves in an elec-
tric cable or a transmission line, which functions both as a signal carrier and sensor. Discontinuity
causes a reflected signal due to an electric property change. Arrival time of the reflected signal and its
amplitude gives an indication of location and degree of discontinuity, respectively, (see Figure 5.34).
A cable sensor embedded in concrete can thus detect both the location and width of the crack.
Two innovative cable sensors (see Figure 5.35) were designed based on the change in topology,
or electric structure, to utilize them as crack sensor. The exact dimensions were finalized based on
available materials in the market and sensitivity analysis.
The operation is illustrated in Figures 5.36 and 5.37. The presence of a partial or complete sepa-
ration between adjacent spirals, which act as the outer conductor of the cable sensor, will force the
return current on the transmission line outer shield to change its flow path.
Water #1
Water layer layer
Mud #2
layer
Soft soil layer
#3
Compact
soil layer #4
Competent soil layer
#5
#6
FIGURE 5.32 Sketch of the pneumatic scour detection probe. (Reprinted from ASNT Publication.)
Rising pressure benchtest

14
11.95 psig
12
Hard Soil
9.52 psig
10
Mud
Pressure (psig) 7.18 psig
08
Water
06
04
02 Hard soil (sealed)

Mud
Water
00
00 20 40 60 80 100 120
Time (s)
FIGURE 5.33 Rising pressure tests in the laboratory. (Reprinted from ASNT Publication.)
Incident voltage step Reflected voltage step
Digital sampling
oscilloscope with a Coaxial cable
SD-24 TDR
sampling head
Distance between points of

monitoring and discontinuity
FIGURE 5.34 Principle of cable sensor measurement. (Reprinted from ASNT Publication.)
(a) Single wire (b)

Air
Rubber tube Solder cover
Spiral wrapping copper tape Commercial stainless
steel spiral
Teflon (ψ = 5.6 mm)
Silver plated, branded
copper wires (ψ = 4.6 mm)
Prototype I Prototype II
FIGURE 5.35 Schematic view of two crack sensors (a) Prototype I, and (b) Prototype II. (Reprinted from
ASNT Publication.)
Current flow path
Partial separation of spiral
FIGURE 5.36 Change of current flow path. (Reprinted from ASNT Publication.)
L-gap
V0+
_
Z0 V0 Z0
FIGURE 5.37 Equivalent transmission line model. (Reprinted from ASNT Publication.)
Static tests and dynamic loading tests were conducted to investigate the sensitivity, signal loss,
and partial resolution. Both types of sensors were found to be sensitive to cracks of various sizes,
and can identify crack locations.
5.5 EMERGING TECHNOLOGY: FIBER OPTICS SENSORS

5.5.1 Introduction
5.5.1.1 General
Fiber optic sensors use light as both a transduction and signal transmission mechanism. Optical
effects that are useful for transduction include intensity modulation, modal domain interference,
Bragg grating wavelength selective transmission and reflection, noncoherent white light interference,
and Brillouin backscatter. Intensity-based FOSs use mechanical effects to modulate transmitted or
reflected light intensity. Light intensity can be detected easily with phototransistors, photoresistors,
and photodiodes, and can be used to make absolute intensity sensors that are useful in SHM. Optical
interference is a useful transduction mechanism for FOSs. When a beam of light is split into two
parts, each of which is sent on a separate path, and then rejoined, interference patterns may form. A
variation in the length of one of the paths causes the interference pattern to shift. Measuring interfer-
ence pattern shifts can give very precise and accurate information about changes in physical param-
eters that affect effective light path lengths. Both coherent (usually laser light) and incoherent (white
or light-emitting diode [LED] light) interferometry are useful for sensing applications. FOS can be
sued for measuring strain, temperature, inclination, and acceleration among others. Ansari (2002)
provided an overview of different FOS usage in SHM, theory, and different types. The author also
described advantages and disadvantages of FOS. Tennyson et al. (2004) presented theoretical study
about validation of FOS in pipelines. A practical application of using FOS in bridge monitoring was
offered by Benmokrane and El-Salkawy (2002). Park et al. (2004) discussed the theoretical basis and
advantages of using long gauge FOS. A landmark paper that introduced the expression “CIVIONICS”
was presented by Rivera et al. (2004). The authors observed that the integration of automatic moni-
toring of civil infrastructure, which relies on sensing, is analogous to automatic sensing in aviations
community that utilizes the expression “AVIONICS.” They also introduced specifications of FOS
systems, which aimed at standardizing the use of FOS in civil infrastructure applications.
A typical construction of FOS is shown in Figure 5.38. The fiber core that transmits light signals
is made of materials such as plastics or glass. The fibers are protected by a cladding tube that helps in
guiding light signals along the fiber. An environmental cover that protects the system is on the outside.
Fiber optics sensing can be categorized in several manners (see Figure 5.39). The mode of sens-
ing can be either intrinsic or extrinsic. There are several sensing methods: light wave phase sensing
(Figure 5.40), time of flight (TOF) sensing, polarity, and intensity are among those methods. The
geometry of sensing can be either local (at discrete points), or distributed (continuously along the
cable). Finally the method FOS is attached to the test article can be either embedded in the interior,
or attached to the exterior. Table 5.2 summarizes some of the classifications of FOS. This section will
explore many of the FOS modes, especially as they are applied to civil infrastructure. We then dis-
cuss cost–benefit of FOS. Several practical applications are offered at the conclusion of the chapter.
Environmental cover
Cladding
needed to
guide the light
Fiber
(Glass, plastic, etc.)
Basic cross-section of fiber optic sensor
FIGURE 5.38 Basic cross-section of fiber optic sensor (FOS).
Fiber optic sensor classifications

(as applied to SHM)
Mode Method of Geometry Location

sensing
Local
Intrinsic Phase Embedded
(material)
Distributed/
Extrinsic TOF Attached
long
(time of flight)
(structure)
Other
(polarity,
intensity, etc.)
FIGURE 5.39 Classifications of fiber optic sensors.
Constructive interference:
When two waves are
in-phase, the net effect is
an increased amplitude
Destructive interference:
When two waves are
out-of-phase, the net effect
is reduced amplitude
Effects of phase: destructive and constructive interferences
FIGURE 5.40 Phase measurement concept in FOS.

TABLE 5.2
Classifications of Some Popular FOS
Bragg Optical Time Domain Brillouin Optical Time
Classification Type Fabry-Perot Grating Reflectometry Domain Analysis
Mode Intrinsic Yes (only some Yes Yes
applications)
Extrinsic Yes Yes
Method of Phase Yes Yes
sensing TOF Yes Yes Yes
Other?
Geometry Local Yes Yes
Distributed Yes (incrementally) Yes (continuous)
Location Embedded Yes Yes Yes Yes
Attached Yes Yes Yes Yes
5.5.1.2 Time of Flight and FOS

One of the concepts in FOS is the TOF, (see Chapter 7). In TOF sensing, a light signal is sent at a
time T0 through the fiber. When the signal encounters a scaterer, it is split into two parts in the form
of forward and back scatter. When the scattered signal is observed at time T1, the distance between
the observation point and the scaterer is computed as
V
D= (5.23)
T1 − T0

The speed of light in the fiber is V.
5.5.1.3 Different Modes of FOS

There are two modes of sensing using Fiber Optics (FO): intrinsic and extrinsic. Intrinsic FOS is
when the sensing occurs totally within the fiber. In this case, the light does not leave the fiber (see
Figure 5.41). Time domain reflectometry (TDR), and Bragg grating are some intrinsic FOS. The
extrinsic FOS occurs by some interaction between the light in the fiber and the exterior environ-
ment. Brillouin and Fabry-Perot are typical examples of extrinsic FOS.
5.5.1.4 ToMs of FOS

In civil infrastructure applications, FOS can sense most needed ToM. These include strains, accel-
erations, displacements (position), rotation, forces, pressure, temperature, pH level, and humidity.
An overview of different uses of FOS to detect temperature, strains, and crack growth was offered
by Tennyson and Morison (2004). Although the main application in that work is pipes, the lessons
can be applicable to other types of infrastructure.
5.5.2 Local FOS: Bragg Grating

Bragg grating is a popular FOS, see, for example, Chuang et al. (2004). It is an intrinsic type-sens-
ing. The basic operations of the sensor are based on defining n as the refractive index
lGLASS
n= (5.24)
l AIR
Transmitter:
Generates laser signals
into the fiber sensor
.. or..
Time Time
Input single Input multiple

laser pulse laser pulses
Electronic Fiber
processing
Detector:
Detects output light waves and generates
adequate spectra
T0 T1 T2 T3 λ0 λ1 λ2 λ3
Time Wave numbers
Different types of output spectra
Basic fiber optic sensing system
FIGURE 5.41 Intrinsic fiber optic sensing basics.
λ is the wave length of light, which is frequency dependent. The Bragg grating governing
equation is
l REFLECTED = 2 n D (5.25)
With D as the distance between the gratings, and λREFLECTED is measured in air.
For a strained fiber grating, the distance between the gratings, the reflected wave length, and the
refracted index will all change as ∆λ, ∆D, and ∆n. They are governed by
Δl ΔD Δ n (5.26)
= +
l D n
The initial and strained wave lengths λ 0 and λ1 = λ 0 + ∆λ are observed during the sensing. The old
and strained refractive indices n and n + ∆n can also be observed by using an external light source.
From the above, the required strain can then be evaluated as
ΔD Δl Δ n
e= = − (5.27)
D l n
Figure 5.42 shows the basic operating principles of Bragg grating FOS.
Input waves with Gratings with Passed waves with

wavelengths other spacing λ0 wavelengths other
than λ0 than λ0
Input waves
with wavelength λ0
Reflected waves λ0
off gratings with
Original length
wavelength λ0
Baseline geometry
Input waves with Gratings with Passed waves with

wavelengths other spacing λ1 wavelengths other
than λ0 than λ1
Input waves
with wavelength λ0
Reflected waves λ1
off gratings with Stretched length
wavelength λ0
Stretched geometry
FIGURE 5.42 Fiber Bragg gratings (FBG) optic strain sensor.
5.5.2.1 Applications of FBG

Demountable Sensors: Metje et al. (2004) explained how a demountable sensor is developed,
calibrated, and tested before put in operation. A sensor is developed using fiber rods along
with multiple fiber optic Bragg grating sensors for monitoring movements in tunnels for
safety, maintenance operations, and long-term monitoring. It is meant to measure settle-
ment, rotation, and distortion. Several connected fiber rods will be introduced into tunnel
lining for monitoring purposes. The new sensor is based on the principle that strain will be
measured at discrete points; and by knowing the geometry of these fixed positions strains
can be converted into displacements, (see Figure 5.43).
At first, the authors tested the ring with regular strain gages to understand its behavior, fibers
were tested for understanding their mechanical properties then tested with Bragg grating
sensors incorporated in to the rods, and then testing them under simple conditions that can
be interpreted with relative ease.
Soil Settlement: The settlement of soil under concrete slab road will cause road damage;
and a gap is formed between the slab and the soil with air is trapped in between, (see
Figure 5.44). Since thermal properties of the air and the soil are different, thermal dispar-
ity between them was used by Cai et al. (2008) to address the issue of soil settlement under
the concrete slab using fiber Bragg grating (FBG) sensor for temperature monitoring, (see
Figure 5.45). It was concluded that by measuring the daily temperature variation under the
bridge approach slab, the settlement and erosion of the soil beneath can be conceptually
monitored and predicted.
10 mm
9 µm 125 µm
Core Bragg grating: Periodic variation

of refraction index in optical
Cladding fibre core.
FIGURE 5.43 Schematics of Bragg grating and smart rod. (Reprinted from ASNT Publication.)
Heat
Concrete Gap resulted

slab from soil
FBG subsitence
temperature
sensor
Soil
FIGURE 5.44 Soil settlement monitoring method. (Reprinted from ASNT Publication.)
Stainless
Optic fiber
tubing FBG sensor
Adhesive bonded on the rod
Optic fiber
Stainless Adhesive
rod
FIGURE 5.45 Fiber Bragg grating (FBG) temperature sensor design. (Reprinted from ASNT Publication.)
5.5.3 Local FOS: Fabry-Perot

The reflected wave from the near and far edges of the Fabry-Perot gap will interfere either construc-
tively or destructively, depending on the phase between the different waves. For wave numbers with
phase of multiple of 2π, the interferences will be constructive. The length of the Fabry-Perot gap
can then be expressed as
L1 =
(j2 − f1 ) l1 l2
(5.28)
4p ( l2 − l1 )

where λ1 and λ2 are any constructive reflected wave numbers and (ϕ 2 − ϕ1) is the phase difference
between these two waves. For two consecutive waves, such a phase difference is 2π. The above
relationship reduces to
1 l1 l2
L1 = (5.29)
2 ( l2 − l1 )

For the strained sensor, the relationship becomes
1 l1 l2
L2 = (5.30)

(
2 l2 − l1 )
where l1 and l2 are the strained constructive reflected wave numbers. The required strain of the
sensor is
e=
( L2 − L1 )
(5.31)
L1

Figure 5.46 shows the basic operating principles of Fabry-Perot FOS.
Input waves reflected Fabry-Perot gap Reflecting fiber

from near edge of
Fabry-Perot gap
Input waves reflected

from far edge of
L1
Fabry-Perot gap Sensor casing
Fabry-Perot gap original length
Baseline geometry
Input waves reflected Fabry-Perot gap Reflecting fiber

from near edge of
Fabry-Perot gap
Input waves reflected

from far edge of
L2
Fabry-Perot gap Sensor casing
Fabry-Perot gap original length
Stretched geometry
FIGURE 5.46 Fabry-Perot FOS.

5.5.4 Distributed (Long) FOS: Optical Time Domain Reflectometry (OTDR)

Monitoring the times of arrival of reflected light pulses, the following relationship can be
established
Li − L j
Tij = Ti − T j = (5.32)
VPULSE

where Ti and Tj are the times of arrival of the ith and jth pulses, respectively. The distances Li and Lj
are the distances of the ith and jth partial reflectors, while VPULSE is the speed of laser pulse in the
fiber. For the strained condition, the above relationship changes to
Li − L j
Tij = Ti − T j = (5.33)
VPULSE

Where Ti and T j are the new times of arrival of the ith and jth pulses, respectively. The distances Li
and L j are the distances of the ith and jth partial reflectors in the strained configuration. The average
axial strain between the ith and jth station is
Tij − Tij
e= (5.34)
Tij

OTDR sensing is appropriate for long uniform constructs such as wires, cables, pipelines, and so on.
Figure 5.47 shows the basic operating principles of OTDR.
Partial reflectors
Fiber
Transmitter
T0
L01
Time Detector L02
Input laser pulse L03
Electronic processing
T0 T1 T2 T3
Time
Output spectrum
Fiber optic sensor: optical time domain reflectometry (OTDR)
FIGURE 5.47 OTDR basic operating principles.

5.5.4.1 BOTDR in SHM

Wu and Xu (2002) found that distributed BOTDR can be effective. The bonding length, if short,
can affect the strain accuracy. They recommended some work-around for this problem in the con-
clusions of their work. Yamaura et al. (2000) showed the applicability of using Brillouin Optical
Time Domain Reflectometer (BOTDR) to measure strains along steel plats and reinforced concrete
beams. Both experiments show good accuracy of the BOTDR when compared to conventional
strain gauges, up to the point of crack development. The accuracy of the BOTDR dropped slightly
after the formation of the crack in the beam. As the applied load increased, the conventional strain
gauge did break, while the BOTDR kept on functioning. We can conclude that the BOTDR can
develop an accurate strain measurements along the length of the specimen of interest up to develop-
ment of cracks, beyond which the accuracy of the BOTDR might drop; the overall measured strain
trends, are still of use up to the point of ultimate strains.
Another comparison of the adequacy of BOTDR of sensing strains along a distance is given
by Oka et al. (2000). Figure 5.48 shows the predicted axial strains when the actual FOS itself is
subjected to an axial tensile force. The predicted strains agreed well with expected strains. The
authors reported that the sensor was able to predict strains as high as 1%. In a more realistic testing
configuration, they subjected a 0.3-m beam to two points loading. The deflections of the beam were
measured by integrating measured strains from the BOTDR sensors, conventional strain gauges,
and from a displacement meter. Figure 5.49 shows a comparison of the three measurements. The
BOTDR-based displacements agreed well with the displacement meter results.
1.2
1.0
Strain from displacement-meter
0.8
Strain (%)
0.6
0.4
0.2
0.0
–0.2
0 5000 10,000 15,000 20,000
Distance (mm)
FIGURE 5.48 Axial strain distribution. (With permission from Dr. Fu-Kuo Chang.)
0
–5 Displacement-
meter
Deflection (mm)
–10
Strain-gauges
–15
BOTDR
–20
–25
–30
–35
0 2600 5200
Distance (mm)
FIGURE 5.49 Displacement comparison. (With permission from Dr. Fu-Kuo Chang.)
5.5.5 Distributed (Long) FOS: Brillouin Optical Time Domain Analysis (BOTDA)
Brillouin frequency shift vB is directly proportional to acoustic wave velocity of the material such
that (see Texier and Pamukcu 2003, and Cho and Lee 2004)
2 nVa
vB = (5.35)
lPUMP

where n is the refractive index, λPUMP is the wave length of the incident laser wave (pump), and Va is
the local acoustic wave velocity. The refractive index n and the incident wave number λPUMP are con-
trolled during the experiment. Thus by measuring the Brillouin frequency shift vB, the local acous-
tic wave velocity can be measured. When the material undergoes compressive, or tensile, strains the
changes in local Va can be measured. Such changes are directly related to the local normal strains in
the material. By combining a series of pulses, and observing times of flights, the strains at different
lengths along the fiber can be computed. Equation 5.35 shown above can be simplified as
vB = Ae e (5.36)

where Aε and ε are a factor of strain proportionality and normal local strains, respectively. It should
be mentioned that the Brillouin shift can also be directly related to temperature, see Hecht (2006)
and OZOptics (2006). The Brillouin shift is generalized to
vB = Ae e + AT T (5.37)

where AT and T are a factor of temperature proportionality and local temperature, respectively.
Equation 5.37 shows the power of Brillouin sensing: it can sense both temperature and strains along
the whole length of the fiber. In addition, BOTDA is capable of distributed sensing at high spatial
resolution. This makes it suitable for infrastructure applications such as prestressed, post-tensioned
tendons (corrosion), and measuring hydrogen embrittlement for high strength wires, complex rein-
forced concrete or connection behavior. The locations of potential damage in these applications are
not known a priori, which makes BOTDA a suitable sensing technique. Figures 5.50 and 5.51 show
the basic operating principles of BOTDA.
Many authors studied the use of BOTDA in infrastructure applications. For example, Zeng et al.
(2002) described using distributed Brillouin scattering fiber optics in measuring temperatures. They
argued for the use of reference temperature measurements to ensure accurate results. Zou et al.
(2004) showed the interdependence between strains and temperature measurements.
5.5.5.1 Smart BOTDA

Bastianinin et al. (2006) explored a technique using “smart” FRP that smears the measured strains
along a crack, thus avoiding the step functions effect of conventional FOS. The technique uses
Brillouin FOS, which features a distributed strain sensing, that is, capability to simultaneously
measure the strain level and locate the strained point along the sensor. To mitigate the issues com-
monly observed with traditional optical fiber cables (easy breakage or insulation rendering it not
useful as a strain sensor), the following Brillouin FOS-specific solutions have been developed (see
Figure 5.52):
1. Sensor featuring an inner FRP core hosting the optical fiber and an outer molded coating
that carries helical grooves intended to enhance adhesion to concrete
2. A “Smart” composite obtained by adding one or more optical fibers to the glass or carbon
or aramid weft of a woven textile
3. A thermoplastic ribbon featuring multiple optical fiber for strain for strain sensing and
thermal compensation
Fixed
The beam is not
end
loaded
Cantilever beam with
Time Fiber uniform properties
Input laser pulses

to fiber Free end
Fiber
Strain distribution
along the length of
ave
the beam is
t r ain ties h hift
null o s per in s
e is n pro illou r
r ic r e
the ust o B fib
i n ce l aco d), n g the
S itia nge lon
(in t cha ted a
Signal strength (dB)
no etec
is d
ber
num
ve
Wa
Dis
the tance
bea alon
m( g
fibe
r)
Brillouin optical time domain analysis (BOTDA)

non-strained beam
FIGURE 5.50 BOTDA: no-strain state.
The “Smart” FRP sensor has an advantage over traditional sensor in the sense that it smears the
measured strains along a crack, thus avoiding the step functions effect of conventional FOS. Thus,
these can maintain sufficient accuracy even in the case of cracked substrates, and hence may be a
good solution for reinforce concrete (RC) structures susceptible for cracking.
These sensors were used on a RC bridge and a large steel bridge. The load test data form RC
bridge showed that the peak strains expected by analysis matched closely with the mart sensor,
where as the traditional gages and traditional FOS showed low values. This was attributed to the
inability of these sensors in the presence of cracks and crack openings were observed during the
load tests. The data from steel bridge showed good comparison to those obtained using analytical
tools. Figure 5.53 shows measured strain profiles along two main girders of the test bridge at two
load passes (bold lines) compared with theoretical strains from finite element analysis (FEA) and
LRFD design provisions (AASHTO).
5.5.6 Comparison between Local and Distributed FOS

A comparison between local and distributed FOS is shown Table 5.3; for more details, see ISIS
Canada (2006), and OZOptics (2006)
5.5.7 FOS Use in NDT versus SHM

When it comes to the use of FOS in civil infrastructure, the distinctions between NDT and SHM
is reduced to a question of time and space. For temporary tests/observations, the use of FOS would
Uinform vertical
Fixed load on the beam
end
Cantilever beam with
Time Fiber uniform properties
Input laser pulses

to fiber Free end
Fiber
Strain distribution
along the length of
the beam f
ft o ve al
shi wa ion n
u in ered port catio
llo tt ro lo
Bri ksca s is p ber
fi
bac mber s at
nu strain
Signal strength (dB)
to
er
u mb
v en
Wa
Dis
the tance
bea alon
m( g
fibe
r)
Brillouin optical time domain analysis (BOTDA)

strained beam
FIGURE 5.51 BOTDA: strained state. (Reprinted from ASNT Publication.)

Thermoplastic Strain sensing
Epoxy matrix Temperature sensing polymer optical fibre
Resin coating ERP core Sensing fiber optical fibre
Flexible
fibers (weft)
Sleeve
Structural Strain sensing
fibers (warp) optical fibre Temperature sensing
optical fibre
FIGURE 5.52 Brillouin SHM sensor concepts. (Reprinted from ASNT Publication.)
x [ft] x [ft]
0 50 100 150 200 250 300 0 50 100 150 200 250 300
–400 –400
FEM
–200 –200
FEM AASHTO
Strain [µm]
0 0
BOTDR
Girder 1
200 200 AASHTO Pass D
BOTDR
400 Girder 5 400
Pier 1
Pier 2
Pier 1
Pier 2
Joint
Joint
Joint
Joint
Joint
Joint
Pass C
600 600
0 20 40 60 80 100 0 20 40 60 80 100
x [m] x [m]
FIGURE 5.53 Measured strain profiles. (Reprinted from ASNT Publication.)

TABLE 5.3
Local and distributed FOS
Type of FOS Local Distributed
Optimum use Material behavior General structural behavior
Measuring resolution High Medium
Application Single point Very long distances
Fiber configuration Multiple fibers Can be either single or
loop-type fibers configuration
resemble a classical NDT. While for a permanent, or semipermanent, installation of FOS, the proj-
ect would be an SHM project. Spatially, we observe that local FOS can be considered either an NDT
or an SHM tool. For distributed FOS, it is perhaps more appropriate SHM tool, since it covers larger
range, without loss of accuracy. Finally, it is safe to state that an embedded FOS should be consid-
ered as an SHM application, due to its permanent setting. Attached FOS, on the other hand, can be
either NDT or SHM application, depending on other factors, as discussed above.
5.5.8 Value (Cost–Benefit) of FOS

This section we present a qualitative summary of FOS in civil infrastructure monitoring. We follow
it by introducing an objective method to evaluate costs-benefits and value of using FOS.
5.5.8.1 Advantages and Disadvantages of FOS

Disadvantages of FOS include the following:
Cost: Relatively speaking, FOS can cost more than conventional sensors. As newer manufac-
turing techniques emerge, the relative cost of FOS is reduced.
Experience: FOS is relatively new when compared with other conventional sensors. Body of
experience is thus not as large as conventional sensors.
Durability: Some FOS might not be adequately rugged, specially when used in demanding
civil infrastructure environment (heat, humidity, abnormal loads and responses, etc.).
Size: Some FOS can be bulky (conduits, jackets, etc.).
Advantages of FOS include the following:
Miniaturization: Some FOS can be miniaturized, see Newton (1999).

Embedment: Capability of FOS for embedment in solids is attractive when measuring inter-
nal responses and environments, such as strains of reinforcing steel rebars in concrete.
Sensitivity and range: FOS can sense wide range of ToMs, see, for example, Kao and Taylor
(1996).
Nonelectrical: FOS operations depend on light transmission. Lack of electrical or electro-
magnetic use or need can be attractive for certain material applications, as well as when
used in certain environmental conditions. In addition, there is no need for electrical power
since the light can travel through the fiber for long distances.
Can be inexpensive: Careful planning and efficient utilization can produce adequate cost–
benefit and life cycle utilization.
Distributed sensing: Some FOS is capable of continuous (distributed) sensing such as Brillouin
FOS. This capability offers potential sensing of whole fields of strains, for example.
Geometrically efficient: Signal can be transmitted to long distances along the fibers. This
provides for efficiency, and potential cost savings.
Multifunctional: In nondistributed FOS applications, several measurement points along one

cable is possible. Bragg grating FOS is a typical example of this usage.
Other advantages: Other potential advantages of FOS are as follows (a) it can provide both
relative and absolute measurements, (b) provides ease of installations: they can be bonded
or embedded easily, (c) capable of both static and dynamic sensing, and (d) can be used for
long-term applications.
5.5.8.2 Objective Cost–Benefit Analysis of FOS

We are now faced with the problem of quantifying the value of FOS for a particular SHM project.
We use a utility-type approach (see Chapter 8) in quantifying the costs and the benefits. The value
can then be established by associating the costs and the benefits. Since the approach is based on
qualitative estimates, it can be used mainly for comparison purposes, such as comparing FOS with
other sensors.
Estimate of Benefits: Potential benefits of using FOS can be estimated qualitatively by the pro-
fessional as BFOS_i subjected to
1 ʺ BFOS _ i ʺ 10 (5.38)

With i = 1,2…9. The descriptions of BFOS_i are shown in Table 5.4. For a given use of FOS, the
magnitudes of BFOS_i are estimated according to the particular conditions and demands of the
experiment.
The total benefits of using FOS is
i=9
BFOS = ∑ WBi BFOS _ i (5.39)
i =1
The weights WBi should be assigned by the professional to reflect relative importance of different
benefits of Table 5.4.
Estimate of Costs: Potential costs of FOS can be estimated qualitatively by the professional as
CFOS_i subjected to
1 ʺ CFOS _ i ʺ 10 (5.40)

TABLE 5.4
Benefit Factors
i BFOS_i
1 Importance of miniaturization
2 Embedment need/potential
3 Sensor sensitivity
4 Sensor measurement range
5 Importance of nonelectrical properties of sensor
6 Importance of electrical power cable and
topology
7 Importance of distributed sensing
8 Advantage of using cables, rather than discrete
conventional sensors; multifunctionality
9 Other factors
TABLE 5.5
Cost Factors
i CFOS_i
1 Out of pocket cost
2 Lack of experience
3 Potential durability damages
4 Potential size disadvantages
5 Other cost/disadvantage factors
with i = 1, 2, … 5. The descriptions of CFOS_i are shown in Table 5.5. For a given use of FOS
the magnitudes of CFOS_i are estimated according to the particular conditions and demands of the
experiment.
The total cost of using FOS is
i=5
CFOS = ∑ WCi CFOS _ i (5.41)
i =1
The weights WCi should be assigned by the professional to reflect relative importance of different
costs/disadvantages of Table 5.5. It is important to assign the cost weights WCi so as to be consistent
with the benefit weights WBi. Such consistency is needed for ensuring accurate estimates of the FOS
value.
Value of FOS: The value of using FOS can now be computed as
BFOS
VFOS = (5.41)
CFOS

Comparison with other types of sensors: Obviously, as benefits outweigh the costs, the value of
using FOS is assured. In some situations, other sensing technologies might be available, and a com-
parison between values of competing solutions might be needed. In such a situation, and to ensure
a reasonable comparison, The costs, benefits, and value of the other technology can be evaluated
using similar equations from Equation 5.38 through Equation 5.41 such that
i=9
BOTHER = ∑ WBi BOTHER _ i (5.42)
i =1
i=5
COTHER = ∑ WCi COTHER _ i (5.43)
i =1
BOTHER
VOTHER = (5.44)
COTHER

The benefits and costs for the other technology have assigned relative weights for different param-
eters as BOTHER_i and COTHER_i. These assignments are consistent with the definitions of Table 5.4
and Table 5.5.
The relative values of FOS and any other competing technology can now be compared using
Equations 5.41 and 5.44.
Example: Let us illustrate the methodology by considering the situation where an SHM project man-
ager considered the use of FOS and a competing technology. The manager assigned relative weights for
the situation as in Table 5.6. Clearly, the cost parameter is a major consideration in the decision. Also,
the benefits from using FOS were not in need for that project. The resulting two values for FOS and the
competing technology reflects those issues. The value of the competing technology is higher than FOS.
Another situation, where benefits of FOS are judged to be important to the project, and the cost
is relatively similar between the competing technologies, is shown in Table 5.7. The value of using
FOS in this case is superior to the competing technology.
5.5.9 Case Studies—Applications

5.5.9.1 Overall SHM Experiment
Example of using Fiber Optics in an overall SHM experiment was offered by Neubauer et al. (2004).
They described a monitoring system for a bridge in Iowa with high performance steel and integral
abutments. It had some typical and atypical fatigue sensitive details such as web gaps (can lead to
distortion induced cracking), stiffeners welded to tension flanges, and weld concentrations at web
and flange splices. Thus the long-term bridge performance was being evaluated using monitoring
system. All sensors are fiber Bragg grating sensors to accommodate long-term monitoring without
bias, drift, and so on. Figure 5.54 shows the schematics of the SHM system.
The three main objectives of the monitoring and evaluation portion of this project are as follows:
• Evaluate bridge performance continuously
• Monitor for deterioration of the bridge over time and develop a baseline record for identify-
ing structural performance changes
• Conduct a detailed fatigue evaluation
Both global performance (distribution factors, neutral axis, and restraint ratios) and local per-
formance were considered in designing these systems and sensor locations. Local effects of traffic,
TABLE 5.6
Cost, Benefit, and Value of FOS and Competing Technology, Case 1
Sensor Wt.
Parameter Competing
Item i Parameter Wt. FOS Tech.
Benefits 1 Importance of miniaturization 1 5 4
2 Embedment need/potential 1 7 4
3 Sensor sensitivity 7 6 6
4 Sensor measurement range 8 7 7
5 Importance of nonelectrical properties of sensor 2 8 8
6 Importance of electrical power cable and 2 8 6
topology
7 Importance of distributed sensing 3 9 1
8 Advantage of using cables, rather than discrete 1 9 4
Costs 1 Out of pocket cost 10 8 3
2 Lack of experience 7 4 2
3 Potential durability damages 5 6 6
4 Potential size disadvantages 5 5 6
Total Benefits 178 141
Total Costs 163 104
Value 1.09 1.36
TABLE 5.7
Cost, Benefit, and Value of FOS and Competing Technology, Case 2
Sensor Wt.
Parameter Competing
Item i Parameter Wt. FOS Tech.
Benefits 1 Importance of miniaturization 1 5 4
2 Embedment need/potential 7 7 2
3 Sensor sensitivity 7 6 6
4 Sensor measurement range 8 7 7
5 Importance of nonelectrical properties of sensor 9 8 3
6 Importance of electrical power cable and topology 9 8 3
7 Importance of distributed sensing 9 9 1
8 Advantage of using cables, rather than discrete 5 9 4
Costs 1 Out of pocket cost 10 8 3
2 Lack of experience 7 4 2
3 Potential durability damages 5 6 6
4 Potential size disadvantages 5 5 6
Total Benefits 422 199
Total Costs 163 104
Value 2.59 1.91
Client
Data acquisition sub-system Gateway sub-system

Wireless access point Wireless access
point & router Modem
Internet
Laser swept interrogator
Data collection server Data storage/processing sub-system

& web server Video Server
Fiber optic gratings
Elementary data checks

temperature elimination
global behavior
rainflow counting
FIGURE 5.54 Block diagram of SHM system. (Reprinted from ASNT Publication.)
FIGURE 5.55 Fiber optic sensor use for sensing composite rebars—overview. (Courtesy ISIS Canada.)
web gap regions, and stress concentration areas were instrumented along with temperature sen-
sors. Some gages were not attached to the structure and hence they give the effects of temperature
and can be used for compensating for the temperature. Rainflow algorithm is used to estimate the
remaining life of the details and the susceptibility of details.
The system allows the client to view real-time video of traffic crossing the bridge and the cor-
responding real-time strain data at various bridge locations.
5.5.9.2 Monitoring Composite Rebars

The use of composite rebars in reinforced concrete systems has been gaining interest since they are
not as susceptible to corrosion problems as steel rebars. Because of their recent introduction to civil
infrastructure systems, continuous or intermittent monitoring their behavior is of importance. ISIS
Canada (2006) reported a monitoring project that utilizes the attributes of FOS to monitor compos-
ite rebars, (Figures 5.55 and 5.56).
5.6 WIRELESS SENSORS

5.6.1 Overview
This section explores briefly the wireless sensor units and components. Finally, the utility of wire-
less sensors in SHM field are presented.
5.6.1.1 Composition of Wireless Sensor Units

Wireless sensors can be divided into two general categories: passive and active, as shown in
Figure 5.57. In passive sensing, as the name implies, the sensor will measure a particular ToM
(displacement, acceleration, pressure, etc.) by responding passively to the conditions of the sen-
sor environment. Passive wireless sensors are defined, in a generic manner, by four issues: (i)
the data acquisition specifications, (ii) embedded computing specifications, (iii) wireless channel
specifications, and (iv) the physical attributes of the sensor itself. In active sensing, the sensor
will generate signals in a controlled manner, and then sense the response of the system to these
signals. An additional unit is included in active sensors that would generate the signal is included
in active sensors. A comprehensive discussion of wireless sensor unit was offered by Lynch and
Loh (2006).
PVC conduit from

junction box to cable
termination site
Junction box
Flexible conduit from sensor

location to junction box
Fibre optic sensors on

composite reinforcement
bars
FIGURE 5.56 Fiber optic sensor use for sensing composite rebars—detail. (Courtesy ISIS Canada.)
Components of wireless sensor unit
Active wireless unit
Passive wireless unit
Data acquisition Embeded Wireless channel Physical Actuation

specifications computing specifications attributes specifications
specifications
• Number of • Type of • Type of • Number of
radio • Physical
channels processor dimensions D/A
• Sample rate • Bus size • Frequency channels
band • Power
• A/D • Clock speed • Power • Sample rate
resolution • Program • Wireless • D/A
standard source
• Number of memory resolution
digital • Data • Spread • Voltage
inputs memory spectrum outputs
used?
• Outdoor
range
• Indoor range
• Data rate
FIGURE 5.57 Composition of wireless sensor unit. (Based on Lynch, J. and Loh, K., Shock and Vibration
Digest, 38(2), 2006. With permission.)
5.6.1.2 Composition of Wireless Sensor Network

What differentiate wireless sensing from conventional sensing is exemplified when studying the
components of wireless sensors, as shown in Figure 5.58. First is the sensing unit itself, which can
be active or passive. The wireless radio protocol that offer communications between the sensors
as well as the project data processing center is another component. The local data processing and
Components of wireless sensors technology
Sensors Wireless radio Data processing/ Power

computational
capabilities
Propriety Operating Conventional:
Passive
protocols systems batteries
Active Open source Signal RFID

processing
IEEE 802.15.4 Ambient

Damage
energy
identification
sources
techniques
Data structure Vibrations,

methods solar,
thermal, etc.
FIGURE 5.58 Wireless sensor components. (Based on Lynch, J. and Loh, K., Shock and Vibration Digest,
38(2), 2006. With permission.)
TABLE 5.8
Importance of Components of Wireless Sensors to SHM
Wireless Components Applicability to SHM
Data acquisition specs. Frequency limits would affect detection frequency ranges. Sensitivity of sensors
would affect measured amplitudes
Embedded computing specs. Can be extremely efficient in producing real time or near real time results.
Wireless channel specs. Can affect amount of transmitted data
Physical attributes Size and ruggedness of the sensors can affect longevity of the SHM project
Actuation specs. (for active sensors) Higher sensitivity can produce more accurate experiment
computational capabilities can play a major role in the efficiency of the whole monitoring system.
By processing the data locally (e.g., by generating frequency spectra from the measured temporal
data) and reaching preliminary, or final, conclusions the computational demands on the central data
processing resources are reduced. Another important component of wireless units is power source.
Several power sources can be used by wireless sensing systems. This includes conventional batteries,
radio-frequency identification (RFID), ambient energy sources, solar, vibration, thermal, and so on.
The different components of wireless sensors are offered in detail by Lynch and Loh (2006)
5.6.1.3 Wireless Sensors and SHM

This section discusses the different modes of interrelationships between wireless sensing and SHM.
As discussed above, wireless sensing contains several components. The importance of these com-
ponents in an SHM project is discussed in Table 5.8. The utility of wireless sensors in SHM project
is discussed in Table 5.9. Table 5.10 compare tethered and wireless sensors from SHM view point.
Finally, Table 5.11 shows the applicability of wireless sensing and different NDT/SHM technolo-
gies. Clearly, the utilization of wireless sensing enables the use of NDT methods in the more general
and demanding SHM projects.
TABLE 5.9
Use of Wireless Sensors in SHM Projects
Component Issue SHM Applicability
Data processing/ Operating systems
computational Signal processing Applications to wavelet and Fourier Transform methods are well proven
capabilities (see Pendat and Piersol 2010). Need to develop wireless-sensing-specific
methods that take advantage of potential of large number of sensor network
(see discussion of this subject later in this section)
Structural Applications to conventional structural identification methods are well proven
identification (see Lynch and Loh 2006). Need to develop wireless-sensing-specific
methods that take advantage of potential of large number of sensor network
(see discussion of this subject later in this section)
Damage Need to develop wireless-sensing-specific methods that take advantage of
identification potential of large number of sensor network (see discussion of this subject
later in this section)
Power Batteries Use of conventional batteries might be adequate for short term and small
number of sensors in an SHM project. For long term and/or large number
of sensors, a new generation of longer lasting batteries is needed
RFID A promising technology for utilization in SHM projects Deployments in
realistic and large scale SHM projects, such as bridges, have not been
demonstrated yet
Ambient energy Promising technologies that might have a great impact on the wider usage
sources of wireless sensors should be further explored
TABLE 5.10
Comparison between Wireless and Tethered Sensors
Issue Tethered Wireless
Connecting hardware Extensive demands of wiring, safety, and Limited demands
expense
Power Power is supplied via wires Power is provided via batteries, or via
innovative methods such as solar, or ambient
vibrations
Sensors (Type) Most sensors can be used in either mode
Sensors (Number) Number of sensors can be limited due to Higher number of sensors can be used
tethering demands wirelessly
Labor Maintaining the tethering equipments requires higher labor effort
Overall cost Cost per sensor is lower for wireless sensing
Accuracy Tethered sensors provide accurate sensing Wireless sensing accuracy has been improving
results
5.6.2 Case Studies

5.6.2.1 Trout Brook Bridge
Gangone et al. (2006) showed in a real-life experiment some of the important attributes of wire-
less sensing. The experimental bridge is located in St. Lawrence County in New York State. It is a
56-ft-span integral abutment bridge. The reinforced concrete deck is supported by four steel beams
that are cast into the abutment. Ten accelerometers and ten strain gauges were mounted wirelessly
on the steel beams, as shown in Figure 5.59. Figure 5.60 shows the actual layout of the sensors.
Figure 5.61 shows the wireless arrangement of both strain and acceleration sensors. The measured
accelerations were processed and STRID analyses were performed to find the experimental natural
TABLE 5.11
Applicability of Wireless Sensors to Different NDT/SHM Technologies
NDT/SHM
Technology Applicability of Wireless Sensors to NDT/SHM Technology
Ultrasound Fairly suited to wireless sensing
Acoustic emission
Corrosion
Microwave Due to security need, wireless sensing might not be suitable for use with NDT microwave
Radiography processes
Fiber optics A combination of wireless and fiber optic sensing might not be feasible. However, such
combination might not be very efficient
Vibrations Fairly suited to wireless sensing
Thermography A combination of wireless and thermography might be feasible
Eddy current Eddy currents require manual expertise, as such it might not be suited to wireless sensing
Strain (S)
28 Ft Acceleroneter (A)
25 Ft 31 Ft
Beam 3
AS S10 A10
(Bottom)
1.5 Ft 7 Ft 13 Ft 19 Ft 25 Ft 31 Ft 37 Ft 43 Ft 49 Ft 54.5 Ft
Beam 2
S2 (Top) A2 A3 A4 A1 A6 A7 A8 A9 (Top) S9
S3 (Bottom) S4 S1 S6 S7 (Bottom)S8
(Bottom) (Bottom) (Bottom) (Bottom)
56 Feet
FIGURE 5.59 Sensor locations. (Courtesy Dr. Jerome Lynch.)
frequencies and mode shapes. These frequencies and mode shapes were compared with analytical
(FEA) results as shown in Figure 5.62. A good comparison shows the adequacy of wireless sensing
in this SHM project.
5.6.2.2 Geumdang Bridge

Some of the objectives of this project by Lynch et al. (2006) were (a) to investigate different aspects
of cost-effective wireless accelerometer sensors, and (b) to compare the wireless sensor results to
conventional tethered sensors. The sensors utilized in the project are shown in Figure 5.63. Figure
5.64 shows the assembled unit. The test bridge is the Geumdang Bridge in Korea. It is a single box
girder concrete bridge (Figure 5.65). The wireless and tethered sensor arrangements are shown
in Figure 5.66. A typical installation of a wireless sensor is shown in Figure 5.67. The collected
accelerations from both wireless and tethered sensors were processed, and a follow-up STRID pro-
duced the frequencies and mode shapes of Figures 5.68 and 5.69 for the wireless and tethered
3
4
9
1 10 7
5 6 8
FIGURE 5.60 Wireless sensors layout on bridge. (Courtesy Dr. Jerome Lynch.)
Wireless mote
Accelerometer
Strain transducer
FIGURE 5.61 Details of wireless arrangements. (Courtesy Dr. Jerome Lynch.)
measurements, respectively. The good agreements between the two systems indicated the accuracy
of the wireless system.
5.6.2.3 Benjamin Franklin Bridge, PA

The Benjamin Franklin Bridge—connecting Philadelphia, PA and Camden, NJ—was opened to
traffic in 1926. It supports seven lanes of traffic and two outboard tracks that support a transit rail
service. On the basis of biennial inspection and follow-up structural evaluations, two cantilever
floor beam members were suspected to have potential fatigue problems under the train loading of
the rail service. The analysis performed by Rong and Cuffari (2004) indicated that stresses were
near the fatigue threshold and hence it was decided to use monitoring to ensure that the actual field
stresses are below fatigue threshold.
The authors described an SHM system. The system data collection rate, sensor locations, num-
ber of sensors, and triggers for data collection (to collect only the data needed when strain exceeds
certain level) were designed based on the structural analysis and loading analysis (train speed, type
of data, etc.). Actual sensors (strain gages) were wired and were collected to data loggers (nodes)
Mode 1: Experimental data

1 FEM analysis: Mode 1
1
0.5
0
0.5
40
20 0
f = 9.56 Hz 0 18.6 37.2 56
0 0 5 Distance (ft)
1
0.5 1 FEM analysis: Mode 2
0
–0.5
0
40
20 –1
f = 11.04 Hz 0
18.6 37.2 56
0 0 5 Distance (ft)
0 FEM analysis: Mode 3
0
–0.5
–1
–0.5
40
20 –1
f = 17.49 Hz 0 18.6 37.2 56
0 0 5 Distance (ft)
FIGURE 5.62 Measured and analytical results. (Courtesy Dr. Jerome Lynch.)
4-channel sensor 8-bit atmel

interface microcontroller 128 kB SRAM
6.4 cm (2.5 in)
16-bit ADC
10 cm (4 in)
FIGURE 5.63 Wireless sensor printed board. (Courtesy Dr. Jerome Lynch.)
FIGURE 5.64 Assembled wireless unit. (Courtesy Dr. Jerome Lynch.)

(a)
38M 46M 38M
North South
S=+0.4457%
P.C Box Gr, L=38.0+46.0+38.0=122.00 M
B=33.00 M
EL.73.700 EL.75.300
EL.72.000
(b) 12.600 (c)
S=–2%
2.600
FIGURE 5.65 Views of Geumdang Bridge (a) elevation, (b) cross section, and (c) general view of bridge.
(Courtesy Dr. Jerome Lynch.)
within 20 ft of the sensors. The data from these nodes were transmitted to two base stations wire-
lessly and to personal computer from base stations in similar fashion.
The measured strain data was about 60% lower than the predicted values and thus it was deter-
mined that the fatigue is not an issue. The project illustrates the importance of stating an objective,
and designing the process around such an objective. It also illustrates how wireless sensing can be
used effectively in collecting project data.
15°
(a) 5.75 m 5.75 m 5.75 m 5.75 m 5.75 m 5.75 m 5.75 m 5.75 m
19 m 19 m 19 m 19 m
1 2 3 4 5 6 7 8 9 10 11 12 13
Tethered
Pier
Pier
Pier
server
6
Wireless
4
14 15 16
server
(b)
9.5 m 9.5 m 9.5 m 9.5 m 11.5 m 11.5 m 11.5 m 11.5 m 9.5 m 9.5 m 9.5 m 9.5 m
Wireless
1 2 3 4 5 6 server 7 8 9
Abuth
Tethered
Pier
Pier
Pier
server
ment
5
6
4
10 11 12 13 14 15 16 17 18
Wireless sensor
Tethered sensor
Data sensor
FIGURE 5.66 Wireless and tethered sensor locations. (Courtesy Dr. Jerome Lynch.)
FIGURE 5.67 Typical installation of wireless sensor assembly. (Courtesy Dr. Jerome Lynch.)
Mode 1 Mode 2
2.98 Hz 4.35 Hz
1 1
0 0
–1 –1
25 25
20 20
15 15
10 10
5 150 5 150
0 50 100 0 50 100
Mode 3 Mode 4
5.03 Hz 7.03 Hz
1 1
0 0
–1 –1
25 25
20 20
15 15
10 10
5 100 150 5 100 150
0 50 0 50
FIGURE 5.68 First four modes—wireless sensors. (Courtesy Dr. Jerome Lynch.)
Mode 1 Mode 2
3.15 Hz 4.54 Hz
1 1
0 0
–1 –1
25 25
20 20
15 15
10 10
5 5
50 100 150 50 100 150
0 0
Mode 3 Mode 4
5.25 Hz 7.32 Hz
1 1
0 0
–1 –1
25 25
20 20
15 15
10 10
5 100 150 5 100 150
0 50 0 50
FIGURE 5.69 First four modes—tethered sensors. (Courtesy Dr. Jerome Lynch.)
External reader antenna
IC chip <Electromagnetic wave>

External reader
(Reading and writing equipment)
RFID tag
Substrates
Antenna circuit
FIGURE 5.70 Basic system of RFID techniques. (Reprinted from ASNT Publication.)
5.6.3 Wireless Sensors–Specific Methods and Techniques

5.6.3.1 RFID in Civil Infrastructure
Satou et al. (2006) discussed wireless sensors using passive RFID techniques that are perceived
to be very useful for bridge management purposes, in the future, and this chapter investigates the
fundamental properties of an RFID tag to be attached to a reinforcing bar within the concrete (see
Figure 5.70). It also explores some of the possible applications for this technology.
The results indicate the following:
1. If RFID tag is placed inside the structure behind a reinforcing bar, it shortens the commu-
nication distance significantly as the bar absorbs and reflects (see Figure 5.71) electromag-
netic wave and also changes the external reader’s resonant frequency (see Figure 5.72)
2. Concrete has little influence on the communications distance with the RFID tag embed-
ded inside it (see Figure 5.73)
The authors reported that the RFID technology was successfully used, experimentally, in measur-
ing the concrete temperature during the hardening process and detection of duct grouting.
30
Lattice number : 1
De at the Max. communication distance

25 Lattice number : 19
Lattice number : 15
20 In the case of no
obstruction for the
radio wave
15
(cm)
10
0
0 10 20 30
Depth from level surface of lattice
to RFID-tag : Dt (cm)
FIGURE 5.71 Influence of lattice frame with reinforcing bars on communication distance. (Reprinted from
ASNT Publication.)
15
Resonance frequencey range of
external reader antenna (MHz)
14.5
14
13.5
13.56 MHz line
13
0 10 20 30
Distance from reinforcing bar to external
reader antenna (cm)
FIGURE 5.72 Influence of reinforcing bar on resonance frequency of external reader antenna. (Reprinted
from ASNT Publication.)
Max. communication distance (cm)
12
After placing concrete
10
0
Before Just after 6 hours 24 hours 48 hours
placing placing
Measurement time
FIGURE 5.73 Influence of concrete on communication distance. (Reprinted from ASNT Publication.)
Another system concept that is based on radio frequency (RF) interrogation was presented
by Thompson and Bridges (2002). The sensor is wireless, and require no power. It can measure
strains, and can be embedded for continuous monitoring. It belongs to the “smart pebble” family
of sensors.
5.6.3.2 Web-Based Systems

Lynch et al. (2009) presented an overview of the potential of using cyber technologies in SHM. The
authors advocated the integration of different sensing technologies to collect pertinent data and
streaming it through the internet where data are processed. Decision making operations can then be
performed to help bridge managers in reaching appropriate decisions.
Shenton et al. (2004) provided an efficient and cost-effective system that is made from off-the shelf
components for quick short-term field testing using strain gages. The system known as in-service
bridge monitoring system (ISBMS) is a small battery-operated system to measure live loads strains in
a girder or slab for up to 3 weeks unattended. It can be accessed via the web for downloading the data
using an interface. The system has three modes of operation: peak detect (measures peak strains that
exceed a user specified trigger threshold), waveform (dynamic strain history of strain of a triggered
event such as passing vehicle), and rainflow (count cycles of randomly varying strain). The system
can be set up in hours and utilizes BDI full bridge strain gage or a quarter bridge foil strain gage.
12-V-rechargeable battery powers the modem and a 2-V-battery powers the strain gage. Figure 5.74
shows the ISBMS as mounted to a bridge. Figure 5.75 shows a sample recorded time history.
This approach can be very useful for cost-effective fatigue evaluation, permit vehicle presence
checking, and load rating analysis.
5.6.3.3 Structural Identification

One of the potential advantages of wireless sensing over tethered sensing is that wireless sensing
offers an almost unlimited potential in the number of sensors that are used in any particular SHM
experiment. This brings the question of whether there is really no limit to the number of wireless
sensors in experiments, such as the Geumdang Bridge or the Trout Brook Bridge cases that were
discussed above. To explore the presence of such limit, let us describe the overall problem first.
Assume that the purpose of the experiment is similar to the two cases presented earlier: to measure
dynamic responses of the bridge and identify its modal behavior (mode shapes and natural frequen-
cies). For simplicity, let us assume that the bridge of interest can be assumed to be a simple supported
beam. As such, we can express the nth natural frequency of the bridge as (Timoshenko 1955).
p n2 EIg
fn = (5.45)
2 2 Ag

With E, I, ℓ, A, and γ representing the modulus of elasticity, moment of inertia, length, area, and unit
weight, respectively. Next, we assume that the sampling frequency of each of the sensors is f SAMPLE.
This implies that the highest usable frequency f USABLE of the project is approximately
fUSABLE = fSAMPLE / 2 (5.46)
Note that we used a factor of 4 to relate usable and sampling frequencies. Such a factor might be
increased if higher measuring reliability is required. Using Equations 5.45 and 5.46 above, we
can compute the theoretical upper limit of natural frequency/mode that can be identified in this
project as
2 Ag (5.47)
nMax = fSAMPLE
p EIg
FIGURE 5.74 Photograph of the ISBMS mounted to a bridge. (Reprinted from ASNT Publication.)
Time history 4
90
80
70
60
50
Microstr
40
30
20
10
0
0 0.5 1 0.5 2 2.5
–10
Time (sec)
FIGURE 5.75 Sample time history recorded by the ISBMS using the time history program from Bridge
I-704. (Reprinted from ASNT Publication.)
To get realistic values on the theoretical limit on the usable number of modes and the sensor’s
properties (sampling frequency) let us consider a rectangular simply supported reinforced concrete
beam. The relations between sampling frequency and nMax are shown in Table 5.12. We reached an
important observation that relates sensors and STRID: the theoretical limit of the number of identi-
fied modes is governed by the sensor properties.
To continue this line of development further, following Timoshenko (1955), the first few natural
modes of a simply supported beam can be illustrated as in Figure 5.76. We propose, rather uncon-
servatively, that we need one sensor for each natural mode to be identified. We can then reach an
otherwise interesting and important conclusion: the maximum number of sensors that can be used
efficiently for this experiment is nMax. Any sensor that is used in excess of nMax sensors can’t be uti-
lized efficiently, since it can’t provide additional useful information to the experiment. Of course,
additional sensors can improve the accuracy of the identified results.
We should note that the above developments and example applies for both tethered and wireless
sensors. However, since the practical limit of the number of tethered sensors is much smaller than
the practical limit of the number of wireless sensors, the upper limit that is described by Equation
5.47 will apply mostly to wireless sensors.
Note that the above development was based on assuming that the maximum usable frequency is half
of that of the sampling frequency. In practice the ratio might be as larger, that is, the maximum usable
TABLE 5.12
Theoretical Limits on Identified Modes
fSAMPLE (KHz) Width (in) Depth (in) A (in2) γ (lb/in3) E (psi) I (in4) g (in/sec2) ℓ (in) nMax
50 18 36 648 0.086806 3.00E + 06 69984 386.4 240 20
100 18 36 648 0.086806 3.00E + 06 69984 386.4 240 28
200 18 36 648 0.086806 3.00E + 06 69984 386.4 240 39
Simple supported
Supported beam
Beam
Mode # 1
Mode # 2
Mode # 3
Mode # 4
Location of sensors needed to identify this mode accurately
FIGURE 5.76 Sensors needed for modal identification.
frequency might be taken as one-third of one-fourth the sampling frequency, to ensure better accuracy.
Doing so will reduce the maximum efficient number of sensors in the experiment. Another factor is the
number of additional sensors needed for each additional frequency. In the simple beam example, we
assumed that a single additional sensor is needed for each additional mode shape. For complex struc-
tural geometries the additional number of sensors for additional frequencies might be larger than one
sensor. This will result in an increase of the maximum efficient number of sensors in the experiment.
However, the main conclusion remains that there is an upper bound of the number of sensors that can
be used to identify the structure using modal identification methods. Any additional sensor beyond that
optimum number will not provide any useful additional information. Figure 5.77 shows the steps that
can be used to evaluate this maximum number of usable sensors for general geometry.
5.6.3.4 Damage Identification

As wireless sensing offers an immense potential for increasing the number of sensors, question
regarding spatial resolution of those sensors and their arrangements arise. We observe that for every
sensor, there is a zone in which the sensor can monitor its ToM. We call this zone as the active
sensing zone (ASZ). For example, the ASZ can be global, for a displacement sensor, or local for a
strain or AE sensors. In addition, we observe that there is an active damage zone (ADZ). The ADZ
is the zone at which a damage, or response, can be felt. The presence of the damage outside an ADZ
will not be felt. For large damage sizes the ADZ is large and the reverse is true. Figure 5.78 shows
Maximum number of usable sensors in a modal

identification experiment
Analytical, or
numerical
Sampling
(finite
frequency
element)
methods
Estimate the
Estimate the
maximum
maximum
number of
usable
modes that can
frequency
be identified
Estimate number of Evaluate the

additional sensors that maximum
are needed for each number of
additional mode usable sensors
FIGURE 5.77 Maximum number of usable sensors for structural identification SHM experiment.
Damage "A"
Sensors can detect
damage “A”, since they
are within ADZ
Sensors can detect

damage “B”, since they
are within ASZ
Damage “B”
Sensors can’t detect Sensors can't detect

damage “A”, since they damage “B”, since they
are outside ADZ are outside ASZ
Sensors in a dense wireless network
Active Damage Zone (ADZ)
Active Sensing Zone (ASZ)
FIGURE 5.78 Active sensing zone (ASZ) and active damage zone (ADZ).
conceptually the ASZ and ADZ. For accurate sensing, the sensor network needs to consider the
interrelationships between ASZ and ADZ as shown in Figure 5.78.
Hay et al. (2008) report that AE technique has been used to monitor several bridge locations as
identified in Figure 5.79. Authors propose a conceptual wireless AE monitoring system to replace
wired AE continuous monitoring system. The field systems have 16-channel data acquisition
Hanger connection Link pin connection
Cope and stringer

Welded stiffener
Cope and stringer Welded stiffener
FIGURE 5.79 Representative areas monitored by acoustic emission. (Reprinted from ASNT Publication.)
capability with pattern recognition and AE data interpretation methodologies. The power manage-
ment strategies using hibernation mode is described with wake-up mechanism based on the para-
metric input of a strain sensor.
5.6.3.5 Remote Programmability

Remote programmability of wireless sensors was addressed by Arms et al. (2004). They described a
methodology used to develop modular wireless sensor network capable of wireless reprogramming
using cell phone network (see Figure 5.80).
The authors then give more details of the system used in the previous chapter. Nodes distributed
over the structure report to base station located near the structure. A cellular phone link from the base
station allows the sensing network to be controlled and reprogrammed from an office elsewhere.
5.6.3.6 Decision Making

Chen et al. (2004) presented a simple methodology, assuming the bridge behaves similar to a single
degree of freedom (SDOF) system, to get the stiffness change in the bridge by estimating change in fre-
quency. Once the stiffness is estimated, the load capacity is estimated by multiplying stiffness with the
allowable deflection. A moving car (using wireless drive-by network) is used to induce ambient vibrations
with one sensor to estimate the frequency. For this method to work, initial stiffness should be known and
also considering lot of other factors influence the frequency. In addition, this research explored the value
of utilizing monitoring systems. Figure 5.81 shows sample computational cost results.
5.7 SMART STRUCTURES

5.7.1 What Is a Smart Structure?
“Smart structures” (SS) is a term that has increased in use lately. There are several definitions of the
term. For our purpose, we define SS as those structures that can sense changes in their environments
Sensor inputs
Lithium thionyl
chloride battery
Radio frequency
(RF) transceiver
Sensor singla 8 bit, low power,
conditioning microcontroller
Flash EEPROM for

sensor logging
A/D converter (12-
or 16-bit resolution)
Multiplexer,
programmable gain
instrumentation
amplifier
FIGURE 5.80 Wireless sensor node block diagram. (Reprinted from ASNT Publication.)
Confidence level vs cost (first year)

2,50,000
2,00,000
Cost in dollars
1,50,000 0.5Hrs
1Hrs
2Hrs
3Hrs
1,00,000 4Hrs
50,000
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Confidence level
Recovery for 1 Hr inspection time

2,50,000
2,00,000
CF (0.1)
Recovery (dollars)
1,50,000 CF (0.2)
CF (0.3)
CF (0.4)
1,00,000 CF (0.5)
CF (0.6)
CF (0.7)
CF (0.8)
50,000
CF (0.9)
CF (1)
0
1 2 3 4 5 6 7 8 9 10 11
–50,000
Years
FIGURE 5.81 Cost estimation results. (Reprinted from ASNT Publication.)

(such as loading conditions, environmental conditions, and or damaged conditions). Moreover, a SS

might be able to respond to such changes in environment by alerting the stakeholders of the struc-
ture, or by actively changing some of the structural properties so that the harmful effects of those
changes are mitigated.
Upon closer inspection to the above definition of SS, it becomes clear that a SS is simply a struc-
ture that has an implemented SHM/IHCE system. The only requirement is that the SHM/IHCE
system should have longer-term objectives, rather than short-term objectives.
5.7.1.1 Smart Structures and Smart Materials

Smart composite materials were categorized into three levels of intelligence by Huston (2007) as
shown in Table 5.13. Huston correctly argued that the levels of complexity and capability increases
as the material intelligence level increase. We would like to generalize such levels such that they
are applicable to SS. First, we note that composite materials, as manufactured materials, are fairly
different from other main engineering materials (steel, concrete, masonry) in the following two
main points:
• The mass of composite materials are usually less than the mass of other engineering mate-
rial. This enables miniature actuators (level II) to be effective on material level.
• The chemical and other processes that are used/can be used in material self-healing (level
III) are more feasible for composite materials than for steel, concrete, or masonry.
So, it might not be easy to identify steel, concrete, or masonry as potentially feasible smart material.
However, we need to remember that a structure is a composite tapestry of different components.
There are distinct differences between structural and material behavior; the following are among
them:
• The structure contains several substructure components that are assembled together by a
set of joints and connections. Good part of the structural damage occur at or near those
joints/connections.
• Structural damage also can occur due to geometry (stability problems) rather than material
damage.
• Structural failures were reported also due to damages of supporting ground (see Chapter 1
in Ettouney and Alampalli 2012).
Considering all of those differences, we still can identify three intelligence levels for the structure.
For example, we propose to consider the relatively small effective structural mass, thus the structure
can be made to react to stimuli as in level II. We also propose that a structure can induce self-healing
processes in a way similar to a composite material changing its color. We can introduce Table 5.14
as the structural equivalent to Table 5.13.
TABLE 5.13
Levels of Smart Composite Materials
Intelligence Level Smart Material
Level I Sensing
Level II Sensing + reacting so as to change geometry/shape
Level III Sensing + reacting so as to change properties
(self-healing, color, stiffness, etc.)
TABLE 5.14
Attributes of Smart Structures
Intelligence Level Smart Structure Function SHM Components
Level I Sense Sensors/instrumentations
Level II React to stimuli: active control measures All four components of IHCE are needed
Level III React to stimuli: recommend healing methods All four components of IHCE are needed
Logic of smart systems
Sense damage or
state
System level Material level
Diagnostics
Self-healing Report to
(if possible) decision makers
FIGURE 5.82 Smart systems logic.
What we are suggesting here is that a complete IHCE project (four components) is actually a SS
system (see Figure 5.82). The structural material need not be a smart material for the structure to
be a SS.
We now discuss the revere situation: can every structure that contains smart material be con-
sidered a SS? Let us consider one of the most important problems in structural behavior: structural
stability. It is well known that structure become unstable because of their geometry. Stability prob-
lem is a brittle-type problem: the structure can become unstable suddenly, without major material
damage. In such situation, smart materials might not be capable of detecting the impeding instabil-
ity situation. On the other hand, carefully placed and monitored structural vibration sensors might
be capable of detecting the upcoming instability situations (see Chapter 8).
We now reached an interesting conclusion: use of smart material in a structure does not guaran-
tee a safe structural behavior; only a carefully designed smart structural system, that may, or may
not take advantage of smart material technologies, can provide such a guarantee.
5.7.2 Smart Structures and IHCE Components

Sensing: Sensing is a common issue for all levels of SS. We need first to recall the comprehensive
sensing description of smart material by Huston (2007). In that work, the smart material sensing
main attribute was that it is distributed within the material itself (see Figure 5.83). Huston described
several topologies for such distribution such as (a) linear (fiber optic sensing), (b) grid, and (c) ran-
dom. The sensing technologies described by Huston ranged from electromagnetic to optics. SS
sensing will need those smart materials sensing attributes. In addition, more sensing attributes are
needed to account for the more complex features of structures as compared to materials, as shown
in Table 5.15.
(a) (b)
Distributed fiber obtic sensors Sensors placed on a
Com grid-pattern Com
p osit p os
em ite m
ater ater
ial ial
Damage can be Damage can

triangularized be
by the linear triangularized
Linear system with Grid by the grid
less accuracy system
(c)
Randomly distributed sensors
Com
p osit
em
ater
ial
Damage can
be triangulated
by the random.
Random system
FIGURE 5.83 Geometries of sensors in composite smart materials (a) linearly distributed FOS, (b) grid-
based FOS, and (c) randomly distributed FOS. (Parts (b) and (c) are courtesy of Dr. Dryver Huston.)
TABLE 5.15
Sensing Attributes of Smart Materials and Structures
Sensing Smart Structure:
Attribute Smart Material In Addition to Smart Material, also:
Time Intermittent and/or continuous
Locations Material body Connections, joints, foundations, decks, local, and global
ToMs Material strains/stresses, motions, Behavior of Soil, connection/joint global motions, etc.
humidity, temperature, etc. Also demands such as loadings, seismic motion, wind
speeds, flood levels, etc.
Scope Local in nature Global
STRID: Structural identification tools are needed for levels II and III. Examples of potential
uses of STRID in SS are shown in Table 5.16.
DMID: Damage identification is an integral part of any smart system, albeit smart material or SS.
Several authors categorize different levels of damage identification as (a) existence of damage, (b) location
of damage, (c) extent of damage, (d) effects of damage, and (d) different ways to respond to # 1 through
4. The first four levels are general enough to be applicable to smart material and SSs (we recognize the
fifth level to be a decision making issue that is discussed below). However, identification of damage in SS
can, in a very general way, be considered as a superset of that in smart material. As was discussed above,
identifying material damage is necessary, but not sufficient, in a structural sense. Table 5.17 shows the
differences between smart material and SS when identifying some types of damage.
Decision Making: The type of decision making and its consequences is what differentiate
between smart structures levels II and III. In general, level III decision making is more advanced
and complex than level II.
TABLE 5.16
STRID for Smart Materials and Smart Structures
STRID Method SS Applications
ANN ANN to estimate seismic damages in real time immediately after a seismic event and
reported to officials
Modal identification Use of chaos technique to estimate potential loss of structural stability
Measure natural frequencies to estimate potential loss of structural stability (see Chapter 8)
Parameter identification Use FEM in evaluating critical scour depths and set up a threshold limits at which the
officials will be notified
Note that STRID is utilized for smart structures levels II and III.
TABLE 5.17
DMID for Smart Materials and Smart Structures
Type of Damage Smart Material Smart Structure
Fatigue cracks Yes: require special attention for joints and connection Yes
Wear and tear Yes No
Soil erosion (scour, liquefaction, etc.) No Yes
Corrosion Yes No
Overload: general Yes Yes
Overload: stability conditions No Yes
Note that DMID is utilized for smart structures levels II and III.
5.7.3 Emerging Paradigms for Smart Structures

As structures continue becoming “smarter,” the need for new methods of sensing becomes even
greater. One potential limitation of conventional sensing is the quantification of the reliability of
sensing. Bergmeister (2004) discussed the general behavior of different sensors as well as data
acquisition techniques. The durability of sensors and probabilistic analysis of results were also of a
concern to the author: the reliability of the sensor measurements must be of concern to officials to
ensure accurate decision making process. A novel approach to accommodate several sensing limi-
tations was offered by McNiel and Card (2004). These authors used neural network technique for
data analysis of sensor measurements. Another technique that aimed at minimizing sensing errors
by careful selection of sensor locations was presented by Weber et al. (2004).
The above shows that there are many inherent limitations of conventional sensing. Huston (2002)
explains the limitations of regular sensors when used in field validation. The author then gives some
practical suggestions.
• Owners lack enough data on structures due to lack of good sensing systems. Sensing sys-
tem must be easy to use, reliable, and cost effective.
• Issues of geometry and timescale complicate implementation of a sensing system. Structures
are large, yet often the key damage is localized.
• Many of the items of interest are inaccessible.
• Their life is very large compared to natural hazard’s short time scale, where as some dam-
ages such as fatigue and corrosion can occur most of the structures’ life.
• A dense array of systems can be expensive, overload of nonessential info in most cases,
and hard to maintain.
The author then observes the need for adaptive sensing systems. Sensors with electromagnetic
waves as part of their transduction and transmission system offer the possibility of reconfigurable
sensing because of the freedom with which the geometry of the system can be readily changed. One
approach to adaptive sensing is to arrange sensors to be sensitive to failure modes or other important
behavior of structures. If these modes/patterns are predictable, sensors can be placed at appropri-
ate locations. But, one needs good understanding of structural mechanics for this to be successful.
Thus:
(a) Sensor arrays with local signal processing: Sensor data is processed locally and send only
important/required data/results to central system. An example is passive peak strain sen-
sor, where only the max strain at allocation during a particular timeframe is recorded.
(b) Imaging system with squinting: Here first the whole system is scanned coarsely and then
interesting/suspected areas are scanned at higher resolution. An example is beam crawling
robotic inspection system. GPR and IRT also fall in the same category.
5.8 OPTIMUM SENSOR SELECTION

5.8.1 Overview
This chapter shows a very clear message: there are numerous types of sensors, and there are even
greater numbers of sensors that can perform similar functions. The different technologies of strain
sensors, for example, are very evident. Figures 5.11 through 5.16 show several types of strain sen-
sors. This leaves the professionals in charge of SHM project in a very good position: there are many
options to choose from. Unfortunately, this begets the question: of the many sensor choices that are
available, which sensor is most appropriate? It is clear that every experimental of monitoring situa-
tion is different in scope, objectives and means. The choice of the wrong sensor can lead to a less-
than-optimum result. Currently, choosing sensors is mostly a qualitative process that is based on the
experience of the professional. As usual, such a qualitative process has the advantage of simplicity;
unfortunately it runs the risk of producing an inefficient process and perhaps even missing the goal
of the project.
In this section, we explore this situation. First we enumerate different parameters that can have
an effect on the sensor performance within a given SHM experiment. We then introduce two opti-
mum sensor selection (OSS) methods that utilize those parameters. One of the methods is quantita-
tive, which is based on decision tree analysis. The second OSS method is qualitative that is based on
the judgment of the user. We present examples for the use of both methods
5.8.2 General Parameters

There are numerous parameters and factors that affect the choice of an appropriate sensor for a
given experiment. We will discuss some of the parameters that are general enough to be applicable
to most situations. We observe that there can be more special parameters that are applicable only to
specific monitoring situations. We will also discuss some of those specific situations at the end of
this section. Also, note that we offer more guidelines on sensor selection that are more specific to
bridge load testing in Ettouney and Alampalli (2012).
5.8.2.1 Objectives of Sensing/Project

Objectives of sensing are an important factor since it is not always possible to use the same sensor in
each test. For example, many of the commercially available inclinometers require longer response
time to obtain stable data, therefore, sampling speed or load cycle time is important and should
be given special consideration. Sensor selection depends on loading type and expected structural
response in each test. In many cases, different installation procedures and adhesive types may be
required, depending on whether static or dynamic load effects are measured, so that gages may
survive structural response duration. Because of all of the above considerations objectives of sens-
ing is a common and important parameter in any/all SHM experiments. Different types of possible
objectives are shown in Table 5.18.
5.8.2.2 Type of Structure

Type of structure to be monitored affects the type of sensor as it may have to match the structural
material properties condition. For example, prevailing standards and specifications should be con-
sidered. In steel bridges, drilling holes for gage installations in tension zones of bridge girders may
not be permitted. Strain gages are made with materials to match thermal conductivity of structural
materials to increase accuracy and reliability of measured data.
Table 5.19 shows different structural properties and how they can affect sensor selection
process.
5.8.2.3 Physical Parameter for Sensing (ToM)

Type of physical parameter plays a major role in selection of sensor and all other aspects of testing.
Some parameters such as deflection can be measured directly where as others can be measured only
indirectly. For example, stress in most cases cannot be measured directly where as pressure can be,
even though both are similar (see Table 5.20).
5.8.2.4 Monitoring Time Span

Monitoring time span and number of times a sensor is planned to be used also play a role in the
selection of equipment/sensors, costs, and other factors; these are briefly discussed in Tables 5.21
and 5.22.
5.8.2.5 Operating Environment

Operating environment plays a major role in choosing a sensor. Specially, when one considers the
fact that many of the available sensors are designed for laboratory applications and may not sustain
TABLE 5.18
Objectives of Sensing/Project
Objective Comments
Type of decision (consequence) Examples:
Just a warning to act on at leisure
Warning requiring immediate attention
Avoiding life threatening condition
Research Examples
Behavior of new material in the system
Distribution coefficients on different girders
State of composite action
Validate design assumptions Examples
Deflections of a prestressed concrete bridge
Strains of an innovative connection
Properties (strength/stiffness) of a newly rehabilitated system
Cost implications Examples
Traffic measurements
Corrosion threshold
Deterioration (e.g., extent of cracks)
Hazard-specific (earthquake, Mitigation efforts, evaluation, prediction, etc.
scour, wind, etc.)
Security Deterrence, defense, etc.
TABLE 5.19
Type of Structure
Objective Some Parameters to Consider
Material type (steel, concrete, frp, Material characteristics (thermal conductivity, coefficient of expansion, weldability,
aluminum) adhesive properties, conductivity etc.)
Gage length to accommodate strain distribution
Design type (box beam, I Accessibility, strain distribution across sections, protection from environment,
sections, slab, etc.) structural stiffness affecting sensor range, signal issues
Design life of structure Sensor life, sensor durability, power requirements
Size (large, small) Number of sensors, sensor size, wired vs. wireless
Connections Range of expansion, space requirements, access, sensor protection
Construction type Access, sensor protection, sensor installation phases, wiring sequences, sensor
protection
Underground, over ground Access, sensor protection
Below water or above water Sensor protection, corrosive resistance, wiring protection
Construction equipment used Sensor durability, sensor protection, wired vs. wireless sensors, installation costs
TABLE 5.20
Physical Parameter for Sensing
Objective Comments
Strain Gage type (Foil strain gages, strain transducers, vibrating wire,
strain sensor range, and resolution)
Stresses Mostly measured from strain gages or by measuring loads (forces)
Temperature Thermo couples, thermography
Humidity Humidity sensors
Vibration (acceleration/ Sensor type (accelerometers, velocity transducers, LVDTs), range
velocity/displacement) and resolution, reaction time, frequency range
Pressure Unlike stress, can be measured directly using pressure sensors
Inclination Inclinometers, relatively hard to install
TABLE 5.21
Monitoring Time Span
Objective Comments
Short Sensor response time, wired vs. wireless
Long Sensor durability and type, data range, sensor and equipment
protection, amount of data and storage, power requirements
Continuous Sensor type, data thresholds and triggers, environmental protection,
sensor and equipment protection, data storage, maintenance
Periodic Data continuity, sensor durability, repeatability
TABLE 5.22
Durability
Objective Comments
One time use Durability is not an issue, need low cost sensors
Multiple use Environmental durability, installation issues,
repeatability, documentation
inclement weather conditions without proper protection. However, there are a few rugged sensors
specially manufactured for use in variable or hostile environments that are more suitable in bridge
testing. If these sensors are unattainable, proper weatherproofing (against temperature, dust, water,
wind, vibration, etc.) should be specified to meet necessary requirements for sensor operation.
Even though they are relatively expensive, rugged gages are a must in long-term monitoring. When
weatherproofing is required, the entire measurement system, including connecting cables, should
be properly protected.
Among operating environment issues is the loading conditions: Some sensors do not operate
effectively under certain types of loading. Each sensor takes a brief time (reaction time) to stabilize,
depending on its principle of operation. For example, many LVDTs and inclinometers require longer
reaction time compared to strain gages. If structural response changes faster than the sensor reac-
tion time, accurate data cannot be collected even when high sampling speeds are used. If transient
or dynamic response is to be measured, sensors with short-reaction times should be chosen. Also,
when long-term monitoring is planned, possible instrument drift should be determined.
Table 5.23 summarizes many of the important issues in sensing operating environments.
5.8.2.6 Assembly/Maintenance/Disassembling (Excluding Labor Cost)

Preparation for the test, access requirements, disassembly of setup, etc. also play an important role
in making decision to proceed with the test or not, test costs, and data quality (see Table 5.24 for
some comments on these factors).
5.8.2.7 Cost
Total cost of a test depends on cost of sensors, equipment, processing time, and expertise needed for
data analysis, maintenance of traffic required during tests, and other costs as noted in Table 5.25.
These costs depend on structural type, complexity, structure location, etc. and all these should be
considered from the test planning stages to get more value.
TABLE 5.23
Operating Environment
Objective Comments
Hot Operating environment needs to be within
Cold manufacturers’ specifications. Structural
Hot and cold characteristics in different temperatures
Humidity should be understood and installation
Acid-rain conditions issues can be very different. Operator
Chloride environment experience could become very important in
Traffic type and loads some cases. Sensor covering and sensor
Lab vs. field material play a significant role
TABLE 5.24
Assembly/Maintenance/Disassembling
Objective Comments
Ease of initial erection/assembly Saves time and cost. Gives more time for actual testing and data
verification. Durability of adhesion, etc. can depend of these factors
Ease of disassembling project setup Saves time and cost
Access Ease of access of sensors can affect project performance
Maintenance and protection under traffic Increases cost. Reduces test time and quality
TABLE 5.25
Cost
Objective Comments
Cost of sensor Impacts number of sensors
Cost of additional hardware Impacts costs and need for multiplexing
Cost of initial labor Can be a major portion for short tests and complex structures, good planning can
reduce these costs
Cost of labor during experiment Mostly skilled expertise with understanding of test subject and instrumentation,
critical for test success, can be relatively low for most projects
Other costs Need careful management to make the benefits outweigh costs
TABLE 5.26
Instrumentation Type
Objective Comments
Turn-key Need limited experience and can be used by users
with limited experience
Power Batteries (DC), electric cables (AC), solar, other
Environment needed by instrumentation Have significant impact on operating costs, e.g.,
A/C or watertight enclosures
Memory Higher sensor-level memory capabilities can
enhance certain decision making process
Data collection mode Stored at-site, transmitted to remote site
Interface of sensors with rest of instrumentation Wired, wireless, individual vs. network
Processing capabilities at site Sensor-level processing, onsite, offsite, remote
5.8.2.8 Instrumentation Type

It is very important to select sensors that are software and hardware compatible. The excitation
voltage, output voltage, wiring scheme, overload handling conditions, and resetting time of the
hardware and software should be considered when selecting a sensor. It is possible to obtain sensors
of very high accuracy, but if the software and hardware are less reliable, it may be more economical
to choose sensors with the same order of reliability.
Table 5.26 illustrates other important instrumentation factors that need to be considered when
choosing a sensor.
5.8.2.9 User Knowledge/Experience

Sensor selection also depends on available technical skills. While a large number of sensors can be
installed with relative ease, a few installations need more experience and technical skills to obtain
reliable and consistent data. For example, some strain gages are manufactured and sold with cables
that eliminate the necessity for field soldering and connections. However, they are relatively expen-
sive compared to their counterparts without cables, a trade off for the time saved and the lesser need
for gage installation expertise. These parameters are illustrated in Table 5.27.
5.8.2.10 Sensitivity/Range/Resolution
Range of measurements during the entire test period has to be estimated before the sensor selec-
tion is made. Accuracy of some sensors varies when used to measure data at different ranges. Some
sensors behave linearly only for a small part of their operable range and may behave differently for
TABLE 5.27
User Knowledge/Experience
Objective Comments
Experienced NDT Level III or P.E. type of certifications, several years of both direct
test experience and test project management experience, etc.
Medium NDT Level II, several years of direct test and installation experience, etc.
Low NDT Level I, trainee or limited test experience
TABLE 5.28
Sensitivity/Resolution
Objective Comments
Loading type Static vs. dynamic
Processing type Time domain vs. frequency domain
Decision type Safety vs. security vs. durability
TABLE 5.29
Sensor Physical Attributes
Objective Comments
Size Important to accommodate rapidly varying change in ToM, detail type
and size, ToM type, and installation ease
Weight Not critical for most bridge applications except for some connections
Interaction effects with the structure Important to obtain the effect due to loading condition alone
Ruggedness Makes operations easy and a necessity for many civil applications
positive and negative responses. Hence, the sensor’s operating range and required accuracy should
be chosen with the expected response range in mind.
Resolution is often coupled with data range. Normally, the minimum possible range has to be
used to obtain maximum resolution. Most available instrumentation converts analog response to
digital response using A/D converters into a fixed number of steps. For example, 20 με data range
with 12 bit A/D gives a resolution of 0.078 με. When the same A/D used with 200 με range gives a
resolution of 0.78 με. If expected structural response is less than 10 με, a sensor with a range close to
10με should be chosen. At the same time, when the data range is around 200 με, and if a resolution
of 0.078 με is required, an A/D with more number of steps has to be used.
Table 5.28 shows some parameters that involves sensitivity/resolution of sensors in an
experiment.
5.8.2.11 Sensor Physical Attributes

Physical attributes of sensors can play a major role in an experiment or test, and can affect the accu-
racy of test results and should be considered carefully (see Table 5.29). For example, selecting the
right foil gage that is compensated for the material of the structural member it is adhered to is very
important to get the accurate strain the structural member is subjected to. Similarly, to get a better
idea of average strain in concrete, a longer gage length is required due to the heterogeneous nature
of the material.
5.8.2.12 Formalizing Sensor Choice Parameters

To formalize different methods of sensor choices, let us define the vector{SPj}, with SPj as the
jth parameter that affects sensor choices. The counter j = 1, 2, … NSP and NSP is the total number
TABLE 5.30
Sensor Selection Parameters
jth SPj
1 Objectives of sensing/project
2 Type of structure
3 Physical parameter for sensing
4 Monitoring time span
5 Durability
6 Operating environment
7 Maintenance (excluding labor cost)
8 Cost
9 Instrumentation type
10 User knowledge/experience
11 Sensitivity/resolution
12 Sensor physical attributes
of parameters. Following the discussion of last section, the different values of SPj are shown in
Table 5.30. This list will be used later in the sensor selection methods.
The above-mentioned guidelines do not cover the entire criteria for sensor selection. Several
factors such as cable types, distance between test structure and instrumentation, cable lengths, sur-
rounding environment, and electrical installations are also important in deciding the right sensors
for the test. For example, if the cable lengths between instrumentation and sensors are very long,
then sensors with built-in amplifiers should be considered to reduce noise/signal ratio. In summary,
relying on the above guidelines and good engineering judgment, a variety of sensors can be con-
sidered first, and then the most appropriate ones based on application, accuracy, reliability, repeat-
ability, and available instrumentation can be selected.
5.8.3 Decision Tree Method

We can use the sensor choice parameters to help in quantifying processes of choosing the sensors
for any SHM experiment. One method is the decision tree method. We present it by an example.
Consider an SHM project that calls for sensing strains. The professional found that there are two
types of sensors: A and B that can be used in the experiment. It was found that there are four param-
eters that can affect the decision as follows:
1. Cost: There are two types of costs to be considered in this decision. First: the direct cost of
sensors. This is the cost incurred if the sensor fails. The second type of costs is the cost of
wrong readings by a sensor. Those types of costs are usually much higher than the direct
sensor costs.
2. Environment: Severe environments can have an effect on sensor performance. This is
especially true for long-term SHM experiments.
3. Reliability: Sensors can fail due do environmental effects. They can also fail due to other
factors.
4. Accuracy: Sensors might give wrong results, thus causing the SHM experiment to fail,
with a much higher resulting costs.
Figures 5.84 and 5.85 show the decision trees for sensors A and B, respectively. Each figure also
shows the probabilities of occurrence for each branch of each node of the tree. The decision trees
also show the different costs of failure, and/or inaccurate readings. For simplicity, we are not con-
sidering the benefits of the SHM process in this example: thus, the decision will be based entirely
on the above-mentioned parameters and their consequences.
Reliability Operating environment Correct reading? Consequences

(Cost: $1,000)
Fail 1.5
Fail 1.5
0.2 0.5
0.8 Yes
Not adequate 0.5 Perform
0.2 No
0.3 environment
5.5
0.8 Perform 0.7 Adequate 0.9 Yes
environment
No 5.5
0.1
Cost of failed sensor = $1,500
Cost of wrong readings = Consequences ($4,000) +
Cost of sensor ($1,500)
= $5,500
FIGURE 5.84 Decision tree for choice of sensor A.

(Cost: $1,000)
Fail 3.0
Fail 3.0
0.1 0.4
0.85 Yes
Not Adequate 0.6 Perform
0.15 No
0.3 environment
7.0
environment
No 7.0
0.05
= $7,000
FIGURE 5.85 Decision tree for choice of sensor B.
Using the decision trees approach, the cost of each option is obtained by adding the costs of all
paths. Each path cost will be the product of its probability of occurrence times its cost.
Thus, the cost of sensor A can be shown to be $1,800, while the cost of sensor B is $1,900.
Obviously, consequences of using sensor A will cost less than using sensor B.
Note that in the above example, the consequences of wrong readings were assumed to be $4,000
(which make the cost of wrong readings = $1,500 + $4,000 = $5,500). Such a cost is fairly close to
the cost of the sensors, which biased the total cost towards sensor A, which costs half of that of sen-
sor B. Assume that the consequences of wrong readings were $20,000 instead and that will make it
much higher than the direct costs of the sensors. We also assume that the direct costs of sensors A
and B are same as before. The new corresponding decision trees are shown in Figures 5.86 and 5.87.
Now the cost of using sensor A will change to $3,100. The cost of using sensor B will be $2,900.
Now the consequences of using sensor A will cost more than using sensor B. This is due to the more
reliable performance of sensor B as compared to sensor A, even though the direct cost of sensor B
is twice that of sensor A.
The above example is suitable for use when the number of parameters affecting the sensor choice
decision is small. It is also suitable for quantitative decision making. For qualitative decision mak-
ing, we offer a method that is based on estimating qualitative weights of parameters.

(Cost: $1,000)
Fail 1.5
Fail 1.5
0.2 0.5
0.8 Yes
0.2 No
0.3 environment
21.5
environment
No 21.5
0.1
= $21,500
FIGURE 5.86 Decision tree for choice of sensor A, modified example.

(Cost: $1,000)
Fail 3.0
Fail 3.0
0.1 0.4
0.85 Yes
0.15 No
0.3 environment
23.0
environment
No 23.0
0.05
= $23,000
FIGURE 5.87 Decision tree for choice of sensor B, modified example.
5.8.4 Sensor Utility Method

The sensor utility method in choosing sensors relies on the fact that each SPj has certain rela-
tive value or weight WSPj in the overall project goals. These parameters are dimensionless with
0 ≤ WSPj ≤ 10. As the suitability of SPj to the experiment increases, the value of WSPj increases. If
SPj is not applicable to the experiment WSPj = 0. These weights will differ from one experiment
to another. They are assigned by the official, preferably with the advice of all stakeholders of the
experiment. Note that the weights are dependent on the type of project, and they are independent
from the type of sensors.
We also define another dimensionless factors αij with 0 ≤ αij ≤ 10 as weighting factors that mea-
sure the suitability of the ith sensor for the jth SPj parameter in the experiment on hand.
Finally we introduce the general utility U S i of a given sensor i as
j = N SP
1
US i = ∑ a ij ⋅ WSPj (5.48)
N SP j =1
When comparing different sensors for an experiment, the optimum sensor would be the one with
the maximum utility such that

US OPTIMUM
= MAX U S ( )
i
(5.49)
The advantage of this approach over the decision tree approach is its simplicity. Note that it does
not account for consequences of failure, or different probabilities, in a formal fashion: it account for
them in a qualitative manner in assigning the different values of αij and SPj.
As an example, consider the situation of a load testing of a short-span bridge. The engineer is
interested in choosing appropriate strain sensor at the midspan for the test. Table 5.31 shows the
different SPj and the corresponding weights WSPj that were assigned to the experiment. There were
three possible sensors that were available for use in the experiment; we define them as A, B, and
C, respectively. After studying different attributes of each sensor carefully, the project management
team assigned values of αij to the three sensors as shown in Table 5.32. From Tables 5.31 and 5.32,
and Equation 5.48, the general utilities of the sensors are
US A
= 35.5
U S B = 28.9
US C
= 30.7
It is clear that the three sensors are fairly matched; with the general utility of sensor A is slightly
better (higher) than sensor B or C. From Table 5.32, it appears that sensor A has better utility due to
its better (optimum) resolution and user familiarity, as well as overall higher scores in other issues
even though it costs more than both sensors B and C.
The sensor utility method can be improved by accounting for the inherent uncertainties in some,
or all, weights WSPj and/or αij. Thus, Equation 5.48 becomes a function of random variables that can
be solved using any of the methods of Chapter 8. The uncertainties of sensor utility U S i can then be
accommodated using the resulting moments. Confidence levels of U S i can be used in deciding the
TABLE 5.31
Weights of Sensor Choice Issues
jth SPj WSPj Comments
1 Objectives of sensing/project 9 Objectives of this project is safety related; thus very important
2 Type of structure 3 The structure is fairly typical, thus this factor is not a governing
factor
3 Physical parameter for sensing 3 Strain sensing is fairly typical, thus this factor is not a governing
factor
4 Monitoring time span 2 Limited time span
5 Durability 2 Due to the short time span, this factor is not important
6 Operating environment 4 Due to the short time span, this factor is not as important
However, bad weather can still have an effect
7 Maintenance (excluding labor cost) 3 Due to the short time span, this factor is not important
8 Cost 9 Important factor
9 Instrumentation type 2 Strain sensing is fairly typical, thus this factor is not a governing
factor
10 User knowledge/experience 8 Important factor
11 Sensitivity/resolution 8 Important factor
TABLE 5.32
Factors of Sensor Suitability to Parameters
Factors αij for
Each Sensor
jth SPj A B C Comments
1 Objectives of sensing/ 9 9 9 Generally speaking the sensors must be good match with the
project project objectives
2 Type of structure 9 3 6 Installation problems, or lack of them need to be considered in
this issue. For example: how are the sensors attached to the
structure? What are the effects of wireless sensing?
3 Physical parameter for 10 10 10 Must be a good match
sensing
4 Monitoring time span 8 8 8 Can the sensor accommodate the different needs of intermittent
or long time spans
5 Durability 7 4 6 Durable sensors score higher values. Wireless and remote
sensing can affect this issue
6 Operating environment 8 5 5 This can be scored by investigating manufacturer’s
7 Maintenance 7 5 6 specifications as well as past performances of the sensor What
(excluding labor cost) are the effects of wireless sensing?
8 Cost 2 5 7 Higher costs score lower values
9 Instrumentation type 8 7 How close does the sensor output match the objectives of the
project? Are there any extra information does the sensor
provide beyond the direct project needs (see serendipity
principle, Section 4.7)
10 User knowledge/ 9 4 4 How familiar the user is with the sensor under consideration?
experience
11 Sensitivity/resolution 8 6 6 Does the sensor produce optimum resolution? (not too coarse,
or unwanted refinement?)
optimum sensor for the experiment on hand. Note that for routine tests, these methods may not be
needed, but can be very useful for long-term and/or major projects.
5.9 OPTIMUM SENSOR LOCATION

5.9.1 Overview
One of the basic tools of SHM is different sensors that are located optimally throughout the struc-
ture. In general, the number of these sensors is small when compared with the size and complex-
ity of the structure. Thus, the locations of these sensors have to be chosen optimally, to insure
accuracy of detected damage and efficiency in both computations and cost. Even for situations
with very large sensor arrays, with large number of sensors, an optimal sensor arrangement is still
needed.
The optimal sensor location problem (OSLP) has been studied by many authors. Shah and
Udwadia (1977), and Udwadia (1994) presented an OSL algorithm that is based on the optimization
of the norm of Fisher information matrix. Papadimitriou et al. (2000) presented an entropy-based
OSL formulation. Their method is optimized by the entropy of the information of the structural
system. Cobb (1996) presented an OSL technique that is based on the sensitivities of the eigensolu-
tions of the structural system.
Although the above OSL algorithms vary in their mathematical details, they all have at least two
commonalties. These commonalties are (a) they are usable in structural damage detection schemes,
and (b) they all have two main parts, namely the structural sensitivities and the structural damage.
The structural sensitivities parts can be, for example, the Fisher information matrix as suggested
by Udwadia (1994), or the eigensolutions derivatives, as presented by Cobb (1996). The structural
damage part, which is common to all OSL algorithms, assumes, a priori, that the location and mag-
nitude of the sensitive (damaged) elements are known. It was shown by Shah and Udwadia (1977)
and Udwadia (1994) that the OSL depends on the assumed locations of the sensitive (damaged)
structural elements. Unfortunately, it is not possible to assume this information, a priori, that is,
before measuring the actual damage, which is what the sensors are supposed to do.
In addition to the above apparent causality problem is that the environmental sources of the
structural damage can be varied. For example, structural damage patterns that result from a severe
windstorm can be different from the damage pattern of the same structure that result from corrosive
effects. Each of these damage patterns would require different OSL. Given that only one layout of
sensors is practical, the question is how to resolve this damage profile conflict?
This section presents an attempt to address these issues in a logical and simple manner. That
attempt was presented by Ettouney et al. (1999). They first tried to estimate the expected damage
location and magnitudes in a given structure. Second, the authors attempted to resolve the conflict-
ing OSL demands resulting from different environmental sources. They then related the structural
environment to the OSL algorithms. It was accomplished by using the stress ratio as a damage indi-
cator. Stress ratio is well known and widely used structural design parameters, see McCormac (1992
and 1995). The possible conflicts between environmental sources of damage can then be resolved
by applying the stress ratio concept and the load factor concept, AISC (1989 and 1995), to the goal
programming approach, Ravindran (1987).
Later in the section, two example problems will be presented to demonstrate the applicability
of the solution. The two structures, though realistic, do vary in complexity. The example problems
demonstrate the importance of considering the structural environs and damage sources during the
solution phase of any OSL computations. It should be noted that the methodologies of this study are
applicable to any OSL algorithm. However, the OSL logic proposed by Cobb (1996) was used by
Ettouney et al. (1999). We summarize their work in the remainder of this section.
5.9.2 Analytical Treatment

5.9.2.1 General
The OSL development of this section is based on evaluating the eigenvalue and eigenvector deriva-
tive for a given structure. These derivatives are computed with respect to any desired structural
component. The eigensolutions derivatives will then be used to compute the eigensolutions gradi-
ents for any assumed structural damage profile. The gradients of the eigensolutions can be used to
prioritize the sensor locations, after colinearity check is performed to eliminate any redundancies
in the prioritized sensor locations. For detailed discussion about evaluating eigensolutions deriva-
tives and the use of these derivatives in the sensor prioritization and colinearity checks, the reader
is referred to Cobb (1996).
In this section, we will summarize the governing equation leading to the sensor prioritization
scheme. The stress ratio Sr based OSL approach will be introduced, and the goal programming
methodology will be used to resolve any conflict in the OSL results from different structural load-
ing conditions.
5.9.2.2 OSL Methodology

Consider the eigenvalue problem

([ K ] − l [ M ]) {f} = [0]
i i
(5.49)
where [K] and [M] are the structural stiffness and mass matrices. The ith Eigensolutions is rep-
resented by {ϕ}i and λi. Nelson (1976) showed that the derivatives of the Eigensolutions could be
expressed by
∂ li T ⎛ ∂ [ K ]⎞
= {f}i ⎜ ⎟ {f}i (5.50)
∂ gj ⎝ ∂ gj ⎠

∂ {f}i
= ci {f}i + {V }i (5.51)
∂ gj

ci = − {f}i + [ M ] {V }i
T
(5.52)

The vector {V}i is the solution of the equation

⎡~⎤
{}
~
⎢⎣ E ⎥⎦ {V }i = F
i
(5.53)
~ ~
{}
The matrix [E ] and the vector F i are the rearranged forms of [E] and {F}i where
[ E ] = [ K ] − li [ M ] (5.54)
∂ [K ]
{F }i = ([ M ] {f}i {f}iT − [ I ]) {f}i (5.55)
∂ gj

~ ~
{}
Rearranging [E] and {F}i to obtain [E ] and F i is essential, since [E] is singular, by definition.
Cobb (1996) has developed a rearranging methodology that enables the accurate solution of
Equation 5.53.
The parameter g j is any structural variable that can be associated with damage, that is, changed
its value during the operation of the structure under consideration. It can be an axial stiffness of a
particular element, a joint rotational stiffness or a prestressing force in a bridge element. The deriva-
tive ∂[K]/∂gi can be explicitly evaluated for any parameter g j.
Using ∂λi/∂gi and ∂{φ}i/∂gi Cobb (1996) showed that the gradients of the eigensolutions ∆λij and
∆{ϕ}ij can be evaluated as
∂ li
Δ lij = Δ gj (5.56)
∂ gj

∂ {f}ij
Δ {f}ij = Δ gj (5.57)
∂ gj

The gradients of the expected structural damage ∆g j include information about the type, location,
and magnitude of the expected damage in the structural system. Cobb (1996) presented a sensor
prioritization scheme using ∆λij and ∆{φ}ij for as many parameters g j as required. His method is
based on ordering the vector
{ D} = ∑ ∑ Δ {f}ij (5.58)
i j

in a descending fashion. The priority of the sensor location follows the entries of the vector {D}. The
top n locations will be the OSL for n sensors. To eliminate any redundancy in the resulting sensor
location, Cobb (1996) employed a colinearity check using ∆λij and ∆{φ}ij. If a pair, or more, of the
top n locations are found colinear, this indicates that their measurements are redundant. Only one
sensor of each colinear group is chosen in the final OSL profile. For more details on the colinearity
checks, the reader is referred to Cobb (1996).
5.9.2.3 Stress Ratio–Based OSL

Two basic types of information in the OSL algorithm outlined above are needed. They are the eigen-
solutions sensitivities ∆λij and ∆{φ}ij as well as the profile and amplitude of the expected structural
damage ∆g j. In reality, all OSL algorithms depend on types of available information, structural
sensitivities, and expected damage gradients. The developments of structural sensitivities are well
established, as shown above. Unfortunately, there is no consistent manner to evaluate the damage
gradient ∆g j.
To determine ∆g j, let us ask the question: what are the causes of structural damage? The answer
is that different structural environments could cause structural damage. Thus, if a specific structural
environment can be specified, the damage gradient ∆g j can be determined.
One of the basic damage caused is structural overstressing, due to different loading conditions.
For a given loading condition 𝓁 it is reasonable to assume that the damage gradient due to this
loading condition � g j is proportional to the stress ratio (Sr)ℓj in the jth structural component. This
assumption is consistent with the fact that the stress ratio is the ratio of the actual stress to the allow-
able design stress in the structural components. It should be noted that 0.0 ≤ (Sr) ≤ 1.0. On the basis
of this the damage gradient profile can be accurately estimated as
Δ g j = ( Sr )j (5.59)

The number of elements included in the OSL algorithm can be established by determining a dam-
age threshold below which the structural components are assumed not susceptible to damage. All
the structural elements will be included if the threshold is taken to be 0.0, conversely, the problem
becomes trivial, that is, no elements would be included, if the threshold is taken to be 1.0.
5.9.2.4 Multiple Loading Conditions—Goal Programming

All realistic structures are subjected to multiple environments and loading conditions that promote
structural damage. It was shown in the above section that for a given structural loading 𝓁 there
exists a specific OSL. Suppose that there are L structural loading conditions. It is reasonable to
expect that each of these loading conditions would have its own OSL profile. These 𝓁 profiles are not
necessarily similar. Furthermore, it is not practical or feasible to change sensor locations to accom-
modate each of these optimal locations for each loading condition.
The possible conflicting OSL problem can be related to the goal programming problem. The
solution of a goal programming problem depends on assigning a weighting factor to each of the
loading conditions, and solving the optimization problem. The difficulty here is to find reasonable
weighting factors. In any structural design problem, the relative importance of the multiple loading
conditions is resolved statistically, and a weighting factor is assigned to each loading condition for
design purposes. These statistical factors will be used in the goal programming technique in this
study.
5.9.3 Case Studies

To study the importance of the consideration of structural loading conditions (environmental effects)
on the optimal sensor locations, two realistic structural cases will be presented. First, a simple
structural steel tower is studied, which is subjected to its own weight, as well as, lateral wind loads.
Second, a more complex bridge structure is presented, which is subjected to its own weight, as well
as, lateral seismic loading (SL).
5.9.3.1 Steel Transmission Tower

Consider the tower of Figure 5.88. The tower is a steel structure that is subjected to both its own
weight (dead load [DL]), as well as, wind loading (WL). The DL is acting vertically downward
while the WL is acting laterally. It is required to place sensors on this structure to measure the pos-
sible damage that might result from either of these two loads, or any combination of them. Only
six sensors are available for such a task. First, the OSLs will be estimated independently from the
loading conditions. This is achievable by assuming uniform damage pattern for all structural ele-
ments. Figure 5.89 shows the OSL for this case. Also, Figure 5.89 portrays the prioritization of the
six sensors. Note that the location with the highest priority is at the top of the tower. In addition, no
horizontal sensors are required for this OSL approach.
The OSL were then estimated using the applied loads on the tower. The damage levels and pro-
file is estimated using the DL alone. In addition, the damage level and profile due to WL alone is
estimated. The OSL due to DL alone were computed, as shown in Figure 5.90a. The OSL due to
WL alone was also computed and shown in Figure 5.90b. Finally, the goal programming technique
of this study was applied to the two loading conditions. The weighting factors for each of the two
optimum conditions were the design load factors applicable to each loading condition. The resulting
OSL for the combined case is shown in Figure 5.90c. A comparison between the three solutions
of Figure 5.90 shows that the OSL was not sensitive to the change of the loading conditions. It is
interesting to note that the OSL of Figure 5.90 is different from that of Figure 5.89. The locations of
highest priority are now near the bottom of the tower. In addition, two horizontal sensors are among
the top six prioritized sensor locations. The importance of the lower locations of sensors for this
type of structures has also been addressed by Shah and Udwadia (1977).
3.0
All dimensions
are in feet
60.0
14.0
Hinged support
Roller support
Sensor for horizontal motion
Sensor for vertical motion
FIGURE 5.88 Tower structure.

Hinged support
Roller support
FIGURE 5.89 OSL using a uniform damage profile.
4 4 4
3 3 3
2 6 2 6 2 6
1 1 1
5 5 5
Optimum sensor location Optimum sensor location Optimum sensor location

dead load - based wind load - based combined basis
Hinged support
Roller support
FIGURE 5.90 OSL with a threshold of 0.5: (a) dead load-based OSL, (b) wind load-based OSL, and (c) load
combination-based OSL.
The OSL configurations of Figure 5.90 were obtained using a lower damage threshold of 0.5.
This means that only structural elements with stress ratio Sr ≥ 0.5 were included in the methodol-
ogy of this study. To investigate the importance of this limit, the OSL problem was repeated with
stress ratio threshold of 0.8. The resulting OSL for DL-only, WL-only, and combined conditions are
shown in Figure 5.91. Again, it seems that the OSL is not sensitive to the loading conditions, as in
the case of Sr ≥ 0.5. However, the changing of the stress ratio threshold had an effect on the optimum
locations of three out of the six sensors.
The insensitivity of the OSL to the loading condition for this example problem, either Figures
5.90 or 5.91, is noted. The reason for this insensitivity lies in the fact that the tower is essentially a
vertical cantilever system that is supported at the base. The stress ratio, and hence the damage pro-
file and magnitude, for all studied loads (DL, WL and any resultant combination) increase from top
to bottom. Thus, the computed OSL should be similar for all these loading conditions.
In the methodology presented in this study, each of the assumed damaged structural elements is
assumed to act independently from the other. If the elements are assumed to be damaged simulta-
neously, as in ∆g j = 1.0, the resultant OSL profile is shown in Figure 5.92. This OSL result is less
accurate than the results of Figure 5.90c or Figure 5.91c albeit more computationally efficient.
5.9.3.2 Suspension Bridge
We now investigate the more complex bridge structure of Figure 5.93. This bridge is subjected to
two loading conditions and their combination. One loading condition is the vertically applied DL.
The horizontal seismic load (SL) is the other loading condition. The combination of these two load-
ing conditions is assumed the basis for any damage that might occur in the bridge.
Due to the complexity of this structure, we will assume that 20 sensors are available for damage
detection. As in the case of the tower, the OSL will be estimated independently from the loading
6 6 6
3 3 3
2 4 2 4 2 4
1 1 1
5 5 5
Optimum sensor location Optimum sensor location Optimum sensor location

dead load - based wind load - based combined basis
Hinged support
Roller support
FIGURE 5.91 OSL with a threshold of 0.8: (a) dead load-based OSL, (b) wind load-based OSL, and (c) load
combination-based OSL.
1
5
Hinged support
Roller support
FIGURE 5.92 OSL using a damage-dependent element assumption.

127.2 189.1
932.5 1598.5 932.5
All dimensions are in feet

First natural period = 2.57 s
FIGURE 5.93 Bridge structure.
Hinged support
Roller support
FIGURE 5.94 OSL using a uniform damage profile.
conditions. Figure 5.94 shows the OSL when uniform damage in all the structural elements is
assumed. The numeric prioritization of each of the 20 sensors is not reported for the bridge case,
since it would complicate the illustrations. It should be noted that the prioritization of the sensors
were different for each of the cases studied. Figure 5.94 shows that all the sensors that were com-
puted with the uniformity assumption are vertical sensors.
The effects of the loading conditions on the OSL were then estimated. The damage levels and
profile were estimated using the DL alone. In addition, the damage level and profile due to SL alone
were estimated. The OSL due to DL alone were computed, as shown in Figure 5.95a. The OSL due
to the horizontal SL alone was also computed and shown in Figure 5.95b. The Goal programming
technique of this study was applied to the two loading conditions, where the weighting factors for
each of the two optimum conditions were the design load factors applicable to each loading condi-
tion. The resulting OSL for the combined case is shown in Figure 5.95c. A comparison between
Figures 5.95a, 5.95b, and 5.95c shows that the OSL is extremely sensitive to the change of the load-
ing conditions, unlike the tower case. For the DL-base OSL, Figure 5.95a, all the sensors are in the
vertical direction. For the horizontally based SL, Figure 5.95b, the OSL are all in the horizontal
directions. For the combined loading case, Figure 5.95c, the sensors are all in the horizontal direc-
tion, albeit in different locations than those reported in Figure 5.95b. The overpowering effects of
the SL on the damage in the bridge are obvious, when compared with the DL.
The OSL of Figure 5.95 was obtained using a lower damage threshold of 0.5. This means that
only structural elements with stress ratio Sr ≥ 0.5 were included. The OSL problem was repeated
with stress ratio threshold of 0.8. The resulting OSL for DL-only, SL-only, and combined conditions
are shown in Figure 5.96. The sensitivity of OSL to the loading conditions is evident. The impor-
tance of the SL loading is also clear. Note that varying the stress ratio threshold has an effect on the
optimum locations.
Figures 5.95 and 5.96 depict the sensitivity of the OSL to the loading condition for this example
problem. The reason for this sensitivity lies in the fact that the bridge is a complex structural system
with multiple supports. The stress ratio, and hence the damage profile and magnitude, for all studied
loads (DL, SL and their combinations) have differing profiles and amplitudes. Thus, the computed
OSL should obviously be different for all these loading conditions.
(a)
Optimum sensor locations: dead load-based
(b)
Optimum sensor locations: seismic load-based
(c)
Optimum sensor locations: combined basis
Hinged support
Roller support
FIGURE 5.95 OSL with a threshold of 0.5: (a) dead load-based OSL, (b) seismic load-based OSL, and
(c) load combination-based OSL.
(a)
Optimum sensor locations: Dead load-based
(b)
Optimum sensor locations: Seismic load-based
(c)
Optimum sensor locations: Combined basis
Hinged support
Roller support
FIGURE 5.96 OSL with a threshold of 0.8: (a) dead load-based OSL, (b) seismic load-based OSL, and
(c) load combination-based OSL.
Hinged support
Roller support
FIGURE 5.97 OSL using a damage-dependent element assumption.
As in the tower case, if the elements are assumed to be damaged simultaneously, as in ∆g j = 1.0,
the resultant OSL profile is shown in Figure 5.97. This OSL result is less accurate than the results of
Figures 5.95c or 5.96c, albeit more computationally efficient.
5.9.4 Closure: Importance of OSL for SHM

This section showed the importance of optimally placing sensors in structures, especially complex
structures. It also explored the importance of accommodating different causes of damage when
choosing the location and number of sensors. As of the writing of this volume, there are still many
more issues to be researched. For example, applications of OSL to other types of complex structural
systems in civil engineering, different types of sensors (strains, velocity, etc.), different damage
sources (corrosion, temperature, fatigue, scour), and nonlinear effects are all among the important
subjects that can have profound effects on any OSLP.
5.10 STEP BY STEP GUIDE FOR CHOOSING SENSORS

This section summarizes optimum steps for choosing efficient and cost-effective sensors that
matches particular SHM projects. The steps are
1. Identify Project Objectives: This is perhaps the most important step in an SHM project.
The objective(s) of the project must be well identified and understood by all stakeholders.
Object definitions must be enumerated in as much detail as possible.
2. Identify ToM: Type of measurements can range from strains, accelerations to humidity or
corrosion rates. In most SHM projects, more than one ToM will be utilized.
3. Identify Sensor Technology: As discussed earlier in the chapter, there are numerous tech-
nologies that can be used to sense different ToMs. The advantages and disadvantages of
different technologies and their suitability to the project objectives should be studied care-
fully before settling on a particular technology for a given project.
4. Perform Needed Technical Analysis: Technical analysis might be needed at this stage
to help in choosing appropriate sensors. For example: the number of structural modes that
need to be measured during a modal identification project would affect the chosen sam-
pling limits of accelerometers. Expected seismic drifts will affect limits of location, or
displacement sensors that might be chosen to measure expected seismic event. In the first
example, preliminary dynamic modal analyses of the structure need to be performed. In
the latter, nonlinear seismic analysis of the structure needs to be performed.
5. Perform Needed Decision Making Analysis: Decision making analysis might be needed
to evaluate cost benefits or utilities of different choices that are available to the decision
maker. Examples of decision making for sensor analysis are given in Section 5.8 of this
chapter.
6. Define General Parameters That Might Affect Sensing: Use Section 5.9 of this chapter
to evaluate pertinent parameters that affect sensor choices. If needed repeat steps 4 and 5
above to refine choices.
7. Compare Competing Sensor Models/Types: If there is more than one possible sensor
or sensor technology that can be used, perform prioritization analysis (similar to those in
Sections 5.8.3 or 5.8.4) to help decision making process.
8. Compute Optimum Number of Sensors: Use structural identification techniques to iden-
tify optimum number of sensors for this project. Experience of professionals can also be
used to choose number of sensors.
9. Identify OSLs: Use techniques similar to those of Section 5.9.2 to optimize sensor locations.
10. Estimate Multihazards Needs/Effects: If there are more than one hazard, estimate its
effects and/or needs on the chosen sensor.
Figure 5.98 shows the above steps.

Note that the above steps are for illustration only. The order of executing those steps can change,
depending on the particular project on hand.
5.11 REMOTE SENSING IN SHM

5.11.1 Overview
Emerging asset management concepts, in the field of civil infrastructure, call for extending the
useful life span of infrastructure. This is a direct result of the increased cost of rehabilitation and
Sensor choice: step by step guide
General analysis
Identify project objectives
Analysis
• Structural
modeling
• Damage modeling
• Other hazards? Identify needed
Identify sensor
sensor ToM (Type of
technology measurements)
Decision making
• Cost/benefit
• Utility
Define general parameters that affect sensing
Number of sensors Location of sensors

Compare sensor
candidates
Yes
Multihazards
FINISH
No
FIGURE 5.98 Steps of choosing sensor in an SHM project.
replacement of systems to accommodate modern hazards and needed safety of the public. One of
the rapidly developing tools that can help to extend useful life spans is the field of IHCE and its
subset field of SHM. The SHM field calls for monitoring the performance of systems at reasonable
intervals. The cost implications of such monitoring are a factor that must be accounted for. One pos-
sible way to reduce SHM costs is by the utilization of remote sensing (RS) methods for monitoring
performance. It is desirable and may be essential to sense the structural response remotely during a
posthazard event to assess potential risks, execute appropriate risk-mitigation programs, and coor-
dinate incident responses. Remote structural sensing also is an emerging field and has potential to
be used in conjunction with IHCE.
In this section, we explore different techniques of RS as they pertain to the SHM/IHCE field. We
observe that RS technologies have been developed mostly for the NDT field. There are, however,
some differences between NDT and SHM/IHCE fields as they pertain to RS, as shown in Table
5.33. For the purpose of this chapter, we define the system, or target system, as the structure to be
tested or interrogated. This section will categorize the RS field into four categories, and discuss
each of the categories as they pertain to the SHM/IHCE field.
5.11.2 Types of RS
5.11.2.1 Type I: Sensors Attached to Subject
Traditionally, the term RS is used to refer to situations when sensors are attached on the target
system and the instrumentation (the sensor results) is placed elsewhere. The sensors and the instru-
mentations are connected either by wires (tethered) or by wireless signals. This Type I of RS is the
conventional sensing method. Many of the basic concepts of Type I sensing is discussed in the sec-
tion of wireless sensing of this chapter.
TABLE 5.33
Features of RS System for NDT and SHM
Feature NDT SHM/IHCE
Sensors and system contact Needed
Distance between transmitters, Short distances are Usually need longer
receivers, and system acceptable distances
Overall Geometry Less complex More complex
Exposed surface condition Less complex More complex
Ambient vibrations Usually not a factor Can be an important factor
Size of system Small Large
Accessibility Usually not a main factor Can be a major factor
Type of material Can be demanding
5.11.2.2 Type II: Sensors Attached to Robots

Another type of RS is when robots are used to sense in difficult situations. An example of this situ-
ation is the bomb detecting robots. This type of RS is not within the scope of this volume.
5.11.2.3 Type III: Remote Sensors, Passive RS

Type III of RS is the passive type. This is when a transmitter projects signals onto the target system.
The signals will have minimal interaction with the target system. They will mostly just reflect from
the system. By processing the reflected signals, some information about the target system can be
gathered. Some of the technologies within Type III RS are described next.
Laser Distance Sensing: Laser pulsers have been used for sometime in estimating distances
between the source and the target system. This technique has been used successfully to measure
distances in several fields, such as surveying, mining, mapping, military, construction, and marine.
It also has been used in the vehicle positioning and collision avoidance markets. This technology
is well proven and is based on TOF technique. The tolerances of the measurements are extremely
low and thus make it suitable to measuring static changes on the surface of the target system. For
example, long-term relative motions between both ends of bridge bearings can be observed by this
technique. Small swelling of reinforced concrete beams that might result from hidden rusting of
rebars that can’t be observed by manual inspection may also be measured by this technique. This
particular use can differentiate between harmless concrete discoloration and more harmful hidden
rusting of steel rebars; the concept is shown in Figure 5.99. In this situation, corrective measures
can be taken before the rusting problem becomes more costly. Fuchs and Jalianoos (2008) pre-
formed laser-based experiments that are fairly similar to this concept (Chapter 3 in Ettouney and
Alampalli 2012).
Holography: Use of laser pulsers to measure surface response of systems has seen several appli-
cations recently. The basic concept is to project one or more laser pulses on the system and observe
the reflected pulses. The reflected pulses can be compared to the original transmitted pulses in sev-
eral manners, such as TOF, intensity, and/or phase. By processing the results, information about the
target system emerges. For example, one such technique projects three laser pulses on front and back
of a plate and then observes the resulting ultrasonic waves that are generated on the plate surface. By
processing interferences between different reflecting pulses, the plate surface waves can be observed
in real time. Any defects at or near the surface can easily be observed from this holographic image.
Another application was reported by Telscho and Larson III (2003). They used full-field laser mea-
surements at normal video rates to measure modes for plate frequencies in the range of Hz to MHz.
Radar: The use of laser as an RS tool is a very promising technique for SHM/IHCE. However,
one of the main disadvantages of lasers is that they might have limited use during foggy days at long
3-The increased volume

demands will cause
2-Rust in steel rebar will concrete surface to deforme
increase volume demands outward, in very small
Concrete inside the concrete magnitude
Steel 1-Rust will

rebar start to form
4-With a remote laser

Laser position device, the bulge in
position
the concrete can indicate
device
rust formation in early
stages
FIGURE 5.99 Conceptual use of RS to detect early rusting stages in reinforced concrete.
distances. A promising emerging application in the field of SHM/IHCE is the use of radar waves. It
was shown that Millimeter-wave (MMW) short range radar waves can be used remotely to identify
ground-based objects such as grass, asphalt, and snow. Larger objects such as sign posts and guard-
rails were also included in the study. Evans et al. (1998) showed that it is possible to achieve higher
sensing resolutions by probing in the submillimeter (SMMW) range. For more optimal performance
they proceeded to combine SMMW with MMW.
5.11.2.4 Type IV: Remote Sensors, Active RS

Type IV RS is based on sending signals from a transmitter through a low density medium, such as
air, into the system that needs to be tested. The signals will then interact with the system generating
some type of response. The system response will scatter the incident signal, and the scattered signal
will then travel back (or forward) through the low density medium into a receiver that will analyze
the received signal. Such an analysis will reveal some information about the remote system.
There are two main difficulties with Type IV RS that distinguish it apart from other types of
sensing. At first, there is the impedance mismatch problem. Limiting our interest, for this chapter, to
RS through air, let us assume that the impedance of air to ultrasound waves (as an example of input
signals) is Z1 and that of the system material to be tested is Z2. Then, we can define the strength of
the refracted signal through the system T as
4 Z1 Z 2
T=

( Z1 + Z 2 )2
The value of T is of the order of 10 −4 ~ 10 −5 for conventional structural materials. This indicates that
most of the incident signal will be reflected from the structural surface and limited signal will pene-
trate the structure. The same difficulty is present for other types of air borne signals such as electro-
magnetic waves and laser signals. The other major difficulty of type IV RS is the rapid attenuation
of signals within the system, that make it difficult to detect damages that are deep within the thick
systems, such as steel or concrete. The two difficulties are conceptually shown in Figure 5.100.
Most of the energy is

reflected back, due to
When an air-borne the large impedance
wave impacts a solid mismatch
Air
Limited energy is
refracted
through the solid
The amplitude will
attenuate rapidly as
the depth increases
Solid
FIGURE 5.100 Difficulties of remote sensing type IV.
Air-coupled transducer (Transmitter)
Ultrasonic waves
Testing article
pass through air
gap and propagate
through the solid as
body waves, then passes
through air gap on the
other side until detected
by receiver
Air-coupled transducer (Receiver)
FIGURE 5.101 Air-coupled transmitter and receiver systems (across setting).
On the other hand, Type IV RS offers immense benefits. Among those benefits are the follow-
ing: (a) no need for contact between the instruments and the system, (b) can be used in difficult or
complex conditions, and (c) can detect more types of damages and system properties than Type III
RS. Because of these benefits, there has been interest in developing methods for Type IV RS. Those
methods are based on ultrasound, electromagnetic, radar waves, AE, and laser methods. We sum-
marize some of those techniques next.
Ultrasound RS Methods: Berriman et al. (2003) used noncontact NDT to evaluate density of
plain concrete. They tested a 30-mm-thick specimen, with lateral dimensions of 100 × 100 mm.
The objective of the test was mainly to explore the capability of a noncontact setup in detecting the
material density of the concrete. They used an air-coupled method as shown in Figure 5.101. The
transmitter was a piezoelectric transducer that generated ultrasound waves in the air. These waves
created body waves in the concrete specimen that in turn created air waves on the other side of the
specimen. By utilizing a piezoelectric transducer on the other side as a receiver, it was possible to
detect the ultrasound air waves. By recording the transmitted motions and given the time of arrival
and the thickness of the sample, the longitudinal wave speed of the specimen can be computed.
In addition to the capability of the noncontact testing method, the experiment had an additional
beneficial outcome: the authors observed that due to the noncontact feature of the experiment, the
measured results have no artificial by-product of the interaction effects that the conventional con-
tact transducers might have had on the results. Kommareddy et al. (2005) used a similar setting to
evaluate embedded defects within an FRP plate. An air-coupled noncontact pair of piezo-ceramic
transducers was used in that experiment. The air gap between each of the transmitters and the
receiver, and the FRP plate was 2.5 in. Since the transmission of the signal through the FRP plate
depended on the presence of defects in the plate, it was possible to detect those defects by observing
the transmission of the signals. By performing a C-scan of the plate, the researchers were able to
obtain 3D visual plots of the defects.
In a novel technique to increase the transmitted energy T Solovov and Busse (2006a) used an
oblique angle of incidence of the ultrasonic waves as shown schematically in Figure 5.102. The
slanted transmitter of ultrasound waves through the air would generate ultrasound waves within the
solid. These ultrasound waves can be plate acoustic waves (PAW) or surface acoustic waves (SAW).
The angle of incidence of the waves can be slanted or normal. The incident ultrasound waves will
be either reflected from the surface or transmitted through the solid, if the solid is not too thick. By
measuring the reflected or transmitted waves, the defects within the solid can be detected through
a TOF technique. The authors compensated for low T by enhancing the transmitted (or reflected)
waves through the use of resonance of the air-solid geometry. They showed that by changing the
angle of incidence of the waves, they could produce a resonance in the PAW/SAW that will com-
pensate for the low T. They showed that for different materials—such as paper, aluminum foil, and
thin wood veneer—gains in the range of 10–23 dB can be achieved through this method. They
also showed that the same method can be used for NDT of thick materials through gains of SAW.
Another use of the noncontact ultrasound testing is to evaluate classical painting, see Siddiolo and
Maev (2006).
Laser RS Methods: Another technique is to use lasers as both the transmitting and receiving
signals (see Figure 5.103). Laser RS can generate body waves (shear and bulk) as well as surface
(Rayleigh) and plate (Lamb) waves. They can be used in high-temperature, corrosive, and conduc-
tive environments. They are also very attractive since the transmitter and the receiver can both
be located at long distances from the system of interest. Sanyal (2007) used this technique by
using a laser pulser to generate waves across damaged FRP (fiber reinforced polymer) material. The
response on the backside of the material is then measured by a 2D scanning of a laser vibrometer.
The material properties of the FRP (modulus of elasticity, shear modulus, and Poisson’s ratio) were
measured through this technique. Also, a C-scan of known defects was performed and the defects
were estimated. Both material properties and the presence and extent of defects were shown to have
been estimated accurately by this technique.
Combined Laser and Ultrasound RS Methods: Combined laser-acoustic setup is another
technology that has been used successfully for RS. It is based on using laser-generated beams that
are aimed at the system. Upon impacting the surface of the system, the laser beam will generate
acoustic body and surface waves. These waves will propagate through the system and will be scat-
tered through the defects. The scattered waves will generate surface waves that will interact with
Air-coupled transducer (Transmitter) Air-coupled transducer (Receiver)
Air coupled transducer will

generate body and surface
waves in the solid The waves will travel
through the solid and
are detected through
the air gap by the
air-coupled receiver
Testing article
FIGURE 5.102 Air-coupled transmitter and receiver systems (same side setting).
Laser pulser
Will generate waves in

the solid when the
laser impacts the solid
Testing article
Laser Vibrometer
will detect waves
as they refract
from the solid surface
FIGURE 5.103 All-laser transmitter and receiver setting.
Laser vibrometer Air-coupled transducer (Transmitter)

Will generate waves in
Laser vibrometer will the solid through air
measure as they refract from gap
the solid
The waves will travel
the air gap by laser
Testing article vibrometer
FIGURE 5.104 Combined setting: laser receiver and air-coupled transmitter.
air, which will vibrate accordingly. By measuring the air vibrations by an ultrasonic transducer, it
is possible to detect the location and shape of those defects. Kenderian et al. (2006) reported on a
successful use of this laser-air hybrid ultrasonic technique (LAHUT) for detecting defects in rail
systems (see Figure 5.104). The LAHUT technique was successful on several fronts: (a) It reported
a damage detection success rate of 100% in several configurations; (b) It was capable of perform-
ing well for a variety of surface conditions, including shiny surfaces; (c) The distances between the
transmitters, receivers, and the system were relatively long at 200 mm and higher; and (d) The RS
system was used on typical high-density steel rail material. Such a system has great promise for RS
of near-surface damage detection of homogeneous dense materials, such as steel or concrete. A sim-
ilar study was reported by Scalea et al. (2005). They included in their study a comprehensive SHM
damage identification algorithm that is based on artificial neural network pattern recognition.
Solodov and Busse et al. (2006b) and Kohler (2006) performed the noncontact coupled RS exper-
iments with the air-coupled ultrasound transducers as the transmitters and the laser vibrometers as
the receivers. See Figure 5.105 for this setup.
Radar RS Methods: The use of high frequency radar waves were used in detecting damage
of steel structures (Housner and Masri 1996). The radar waves would penetrate the steel construct
and scatter through cracks in the base or the welding material. The scattered wave signals are then
processed to reveal these defects. The high impedance mismatch between the steel material and
air voids that form though the defects would produce high scattering signals. A signal processing
Laser pulser
Will generate waves in Air-coupled transducer (Receiver)

the solid when the
laser impacts the solid
The waves will travel
the air gap by receiver
Testing article
FIGURE 5.105 Combined setting: laser transmitter and air-coupled receiver.
Electronic microscopic
scanning will detect emission
Defect will emit

acoustic signals
Defect Emission signals will

Solid propagate through the
solid
FIGURE 5.106 Acoustic emission concepts in RS.
scheme, such as those used in computer aided tomography scans (CATSCAN) can help in damage
identification. Since the radar waves have much higher frequencies than, for example, ultrasound
waves, they can penetrate construction materials (such as steel or concrete) deeper than the ultra-
sound waves.
Electromagnetic Methods: Electromagnetic acoustic transducers (EMAT) can generate acous-
tic signals in the system of interest as a response to electromagnetic waves generated by a non-
contact transducer close to the system. Similarly, noncontact EMAT transducers can also receive
signals from the response of the system. EMAT can generate all types of ultrasonic waves within
the tested body. Unfortunately, there are some limitations to the use of EMAT as RS devices. The
tested system must be electrically conductive. Also, the efficiency of the EMAT decreases rapidly as
the air gap between the transducer and the tested system increases. The optimum air gap is usually
in the range of 1 mm. As such, it seems that EMAT may not be a practical technology for SHM/
IHCE field.
Acoustic Emission: AE is a passive NDT technology. It purports to listen to defects. As such,
in the field of RS technologies, there is only a receiver to detect any AE from defects, as shown
in Figure 5.106. The AE technique in the field of RS was used to detect emission signals from the
crack in stainless steel specimens. The signals were emitted in the frequency range of 9–10 MHz.
An electron microscope scanning was performed to detect the AE signals.
5.11.3 Technical Aspects of Remote Sensing

5.11.3.1 Penetration versus Backscatter
All Type IV RS techniques will rely on one of two STRID methods: penetration or backscatter of
waves, as shown in Figures 5.107 or 5.108. In the penetration of waves method, the transmitter will
generate the wave type of interest (can be laser, ultrasonic, electromagnetic, penetrating radar such
as X-rays, etc.). They will impact the structure and penetrate it. It will interact with the structure
Incident pulse
Amplitude
1-The wave
impacts the
subject
Time
Signatures
of multiple 2-The wave
reflections penetrates
and refractions the subject
3-The wave
Amplitude projects out
of the subject
on the opposite end
Time
4-By analyzing the output wave,
some information about
the subject can be inferred
Penetrating remote sensing technique
FIGURE 5.107 Penetrating remote sensing technique.
Incident pulse Signatures of

backscatter
Amplitude
Amplitude
Time Time
1-The wave impacts 3-The scattered wave
the subject bounces back to the
receiver
2-The wave reflects and

scatter from and within
the subject (scatter)
4-By analyzing the output wave,
some information about the
subject can be inferred
Backscatter remote sensing technique
FIGURE 5.108 Backscatter remote sensing technique.
and when the waves exit the structure from the opposite end, it will be detected by the receiver. The
received signals have been modified through reflections and refractions by the structure and any
structural defects. By analyzing the incident and output wave, information about the structure can
be gathered (see Figure 5.109). Perhaps the most popular penetrating RS method is the CATSCAN
in NDT and medical fields. Chapter 8 explores the basics of this method.
Bridge girder
Damaged
area
4-Decision making: Based on

analysis results, a decision is made as
to the severity of damage and what
next steps might be
1-Transmitter: Transmit appropriate
2-Reciever: Receive backscatter waves, such as ultrasound, laser,
information optical, etc.
3-Analysis: Analysis of transmitted

and backscattered information
Concepts of remote sensing after a blast event
FIGURE 5.109 Remote sensing after a blast event.
In the backscatter method (Figure 5.108), as the name implies, both the transmitter and the
receiver are placed on the same side of the structure. When the incident waves impact the structure,
it will reflect and refract on the surface, the refracted waves will further penetrate the structure and
reflect from other structural boundaries. When the structure responds to this complex wave patterns,
it will further affect the reflection/refraction. The process will result in a set of scattered waves from
the structure, and some of these scattered waves will find its way back to the receiver. The analysis
of incident and backscattered waves can reveal useful information about the structure. The NDT
ultrasound methods pitch-catch that was described earlier offer good examples of the backscatter
methods. Table 5.34 shows the differences between penetrating and backscattered methods.
5.11.3.2 Decision Making

It was argued by Ettouney and Alampalli (2000) that the decision making process must be an inte-
gral part of any SHM/IHCE. This is also true for RS. Several decisions must be undertaken by the
officials including whether to embark on an RS approach versus a more conventional sensing system.
These can range from budgetary to technological and also require attention as to the frequency of
the input. For example, assets in large metropolitan areas in seismic zones may need a more frequent
monitoring than those in smaller localities. Clearly, there are several advantages and disadvantages
to the RS system. Some of the advantages are (a) usability in complex and difficult geometries, and
(b) the process is noncontact in nature. Some of the disadvantages are (a) limitations of the current
technologies, and (b) cost disadvantage. As usual in any decision making process, the official should
weigh the costs versus benefits of any method before embarking on a particular project.
5.11.4 Current Applications in SHM

5.11.4.1 Thermography
Given the average age of a bridge in the inventory in the United States being about 44 years, there is
an increasing focus on the condition assessment, repair and rehabilitation of concrete components in
bridges. Quantitative evaluation (location and extent) of the concrete components subjected to spalls and
TABLE 5.34
Penetrating versus Backscattered Methods
Issue Penetrating Waves Backscattered Waves
Location of receiver Opposite side of the transmitter Same side as the transmitter
Structural size Larger Thinner
Type of waves High energy waves, such as X-rays and Gamma rays are used for deeper penetration, or stronger
backscattering of structures. Ultrasonic, electromagnetic and laser waves can be used too, with
lesser penetration or backscattering capabilities
Safety Use of high energy penetrating waves, such as Same concerns for safety issues if high
X-rays or Gamma rays, that can penetrate deep energy waves are used
into the structure can pause safety concerns
Location of defects Any location within the structure, as long as they Usually close to the surface due to the
are within reasonable proximity of the potentially rapid attenuation of the waves
penetrating waves deep inside the structure
(a) (b)
46.1
29.4
FIGURE 5.110 Test block and its IR image (a) visual image, and (b) IR image. (Courtesy of Dr. Glenn
Washer.)
delaminations due to steel corrosion of embedded reinforcing steel is very useful for better bridge man-
agement. Traditional inspection methods require arms-reach access to the surface being inspected. This
frequently requires lane closures or other traffic disruptions, and can be time consuming for the inspec-
tors. The application of IR thermography for detecting deterioration in concrete bridge components
may provide a means for imaging large areas of a structure from a distance, reducing inspection times
and averting traffic disruptions. Washer et al. (2008) discussed experimental testing designed to identify
the optimum conditions for the detection of subsurface features in concrete bridge components.
The authors erected a test concrete block to evaluate the effect of environmental influences upon
a concrete structure (see Figure 5.110). The test block was instrumented with Styrofoam targets of
known thickness, area, location, and depth were embedded in the concrete was used to evaluate the
effect of environmental influences upon a concrete structure to determine optimum conditions for
use of IR method for defect detection. Environmental parameters including wind speed, solar load-
ing, humidity, and ambient temperatures are assessed to determine their influence on the detection
of subsurface features in the block. Figure 5.110 shows the IR images of the same block. Note that
the depth of the targets is qualitatively apparent by the contrast of the IR images. Figure 5.111 shows
the changes in the contrasts of the four targets at different times of the day. The correlation between
solar loading and thermal contrast is evident in Figure 5.112.
Initial results of the study suggest that clear skies, resulting in uninterrupted solar loading, pro-
vide good contrast in thermal images as the total thermal energy introduced into the block on a
(a) (b)
(c) (d)
FIGURE 5.111 Thermographic images at different targets taken at different times during the afternoon
(a) 1:00 p.m., (b) 1:30 p.m., (c) 2:30 p.m., and (d) 4:00 p.m. (Courtesy of Dr. Glenn Washer.)
7.0 12.6
Area of solar loading
6.0 y = 1.87x–0.31 10.8
R2 = 0.75
5.0 9.0
Thermal contrast (°C)
4.0 7.2
3.0 5.4
2.0 3.6
1.0 1.8
0.0 Solar power 0.0

Linear (Solar power)
–1.0 –1.8
–1.0 0.0 1.0 2.0 3.0 4.0
Area of solar loading (kWatt-ht/m2)
FIGURE 5.112 Scatter plot relating thermal contrast and solar loading. (Courtesy of Dr. Glenn Washer.)
clear day can be greater, establishing a greater thermal gradient in the block and therefore greater
contrast in the images.
5.11.4.2 Ground Penetrating Radar (GPR)

Geophysical methods can yield more information than the chain dragging and hammer sounding and
if used carefully may have the ability to quantify the degree of deterioration. GPR has been used by
several on number of bridges for getting more information on bridge decks with the intention of get-
ting information such as deck thickness, reinforcement/rebar depth (concrete cover), and configura-
tion, potential for delamination, concrete deterioration, and estimation of concrete properties.
Gucunski et al. (2008) provided a comparative study of bridge deck condition assessment by high
frequency air (horn) and ground coupled GPR antennas. Horn antennas have been used in the past
to primarily provide a fast overview of the condition of the deck, while ground coupled antennas
provide more detailed imaging and analysis of the deck condition. The primary objective was to
evaluate a new 2.6 GHz ground coupled antenna on bare concrete decks through a comparison with
existing high frequency 1.5-GHz ground coupled and 2.0-GHz air-coupled antennas. Figure 5.113
shows the method of data collection in the field. The results showed that the 2.6-GHz antenna pro-
vides significantly more detail compared to the 1.5 GHz, so that a strong scatter from the aggregate
in the concrete above the rebar level becomes clearly visible. Disadvantages of lower resolution of
images from the air coupled antenna are to some extent compensated by the capability to conduct
surveys of bridge decks at highway speeds.
5.11.4.3 Thermoelastic Method

A thermoelastic method is described by Fuchs and Jalianoos (2008) for identifying full-field stress
distribution on structural members. This method is described as easy to apply, requiring s minimal sur-
face reparation, can be applied at a wide viewing angle, and is relatively immune to vibrations. A ther-
moelastic stress measurement instrument was used to measure the surface stress of a specimen through
thermoelastic effect that takes place when a material’s temperature changes as a result of experiencing a
bulk compression or expansion. For example, if a solid material is subjected to a tensile force, the mate-
rial experiences a temperature change proportional to the sum of the principle stresses.
To measure the temperature changes on a specimen (steel light pole, Figure 5.114) that result
from stress, the authors took infrared (IR) images of the specimen with a thermal camera. Due to
extremely small temperature changes that result from typical stresses and hence the specimens are
subjected to dynamic loads in the order of a few cycles per second. The specimen is coated with a
flat black paint to make its surface a diffuse emitter of IR radiation.
Testing was done on small specimens as well as a steel light pole and a steel I-girder. The data
showed that the color scale in the image is proportional to the sum of the principle stresses in the
specimen. Figure 5.115 shows the tests on steel light pole structures loaded dynamically to induce
fatigue cracks.
(a) (b)
FIGURE 5.113 Data collection by GPR (a) leading instrumentation, and (b) data collection vehicle.
(Reprinted from ASNT Publication.)
Hydraulic actuator Infrared camera
Light pole Light pole base with weld
FIGURE 5.114 Steel light pole measurements with the thermographic system. (Reprinted from ASNT
Publication.)
Light pole base Crack
Strain gage Tape Tape

Strain gage
FIGURE 5.115 Thermographic results of crack at the base of the steel pole. (Reprinted from ASNT
Publication.)
5.11.4.4 Correlating Load Level with Surface-Wave Laser Measurements

Zoëga and Wiggenhauser (2008) investigated the influence of uniaxial load on the propagation
time of surface waves in concrete well below the ultrasonic range. The surface-wave velocity was
measured at different load levels along different directions relative to the direction of the applied
uniaxial load. Results indicate that the surface-wave velocity is influenced by the applied load and
direction; therefore it has a distinct stress-sensitive behavior and it is also possible to determine the
load direction from measurements. The different behavior of the velocity curve at different load lev-
els may potentially be used for determining actual stress level and ultimate strength of the concrete
specimen. Figure 5.116 shows the experimental setup (a) and the experimental concept (b).

This section presented an overview of the field of RS as it can potentially be used in the field
of SHM/IHCE. We categorized the RS field in four types. We then described each of the types.
Clearly, Types III and IV are the most useful RS types.
(a) (b) Load
0.6 meter
0.4
m
Actuator
1m
10°
M 0.1
1
0.
2*36 points
L as
90° er-
vib
3.5
Load m
FIGURE 5.116 Laser measurements of surface waves (a) overview of the test setting, and (b) illustration of
the concept. (Reprinted from ASNT Publication.)
We also showed several applications of RS in the field of SHM/IHCE. Remote monitoring of early
stages of corrosion (rust) of reinforced concrete is one example. Accurate detection of modes is another
example. Also, RS of fatigue cracks, or tilting of piers after severe flooding, or hidden but near-surface
cracks after an earthquake event are all examples of potentially useful RS. Another important applica-
tion is protection of first responders after a catastrophic event, such as a terrorist attack, where damage
could be monitored remotely. Figure 5.9 shows this concept. Thus, the RS field is evolving rapidly and
promises to benefit the SHM/IHCE field and the whole community to a great extent.
REFERENCES
AISC., (1995) Load and Resistance Factor Design, American Institute of Steel Construction, Chicago, IL.
AISC., (1989) Manual of Steel Construction, Allowable Stresses, American Institute of Steel Construction,
Chicago, IL.
American Society of Testing and Materials, ASTM., (1987). Standard Test Method for Half-Cell Potentials of
uncoated Reinforcing Steel in Concrete. ASTM C876.
Ansari, F., (2002) “Fibre Optic Sensors and Systems for Structural Health Monitoring,” Proceedings of 1st
Arms, S. W., Galbreath, J. H., Newhards, A.T., and Townsend, C.P., (2004) “Remotely Reprogrammable Sensors for
Structural Health Monitoring,” Proceedings, NDE Conference on Civil Engineering, ASNT, Buffalo, NY.
ASNT., (1996) Nondestructive Testing Handbook, 2nd Edition, American Society of Non-Destructive Testing,
Columbus, OH.
Bastianinin, F., Matta, F., Rizzo, A., Galati, N., and Nanni, A., (2006) “Overview of Recent Bridge Monitoring
Applications using Distributed Brillouin Fiber Optic Sensors,” NDE Conference on Civil Engineering,
ASNT, St. Louis, MO.
Benmokrane, B. and El-Salkawy, E., (2002) “Design, Construction, and Monitoring of a Bridge Deck Reinforced
with FRP Bars: Wotton Bridge,” Proceedings of 1st International Workshop on Structural Health Monitoring
Bergmeister, K., (2004) “Reliability Based Structural Health Monitoring of Bridges,” Proceedings of 2nd
Berriman, J., Gan, T. H., Hutchins, D. A., and Purnell, P., (2003) “Non-Contact Ultrasonic Interrogation of
Concrete,” Non-Destructive Testing in Civil Engineering International Symposium, Berlin, Germany.
Cai, S., Hou, S., and Ou, J., (2008) “Settlement and Erosion Monitoring Using FBG Sensors for Soil Under
Bridge Approach Slab,” NDE/NDT for Highwaya and Bridges: Structural Materials Technology (SMT),
ASNT, Oakland, CA.
Chen, G., (2004) “Novel Cable Sensor for Crack Detection of RC Structures,” Proceedings, NDE Conference
on Civil Engineering, ASNT, Buffalo, NY.
Chen, S-E., Callahan, D., Jones, S., Zheng, L., Biswas, P., Eldina, P., El Yamak, B., Malpekar, A. D., and
Lokanath, R., (2004) “Design of a Remote monitoring Techniques of Bridge Integrity Using Wireless
Drive-by Network,” Proceedings, NDE Conference on Civil Engineering, ASNT, Buffalo, NY.
Cho, S-B. and Lee, J-J., (2004) “Strain Event Detection Using a Double-Pulse Technique of a Brillouin
Scattering-Based Distributed Optical Fiber Sensor,” Optics Express, 12(18), 4339–4346.
Chuang, J., Brown, B., Rivera, E., and Fletcher, D., (2004) “Design and Implementation of an Integrated Fiber
Bragg Grating Interrogation System Using Distributed Computing Architecture,” Proceedings of 2nd
Cobb, R. G., (1996) “Structural damage identification from limited measurement data,” Ph. D. dissertation,
School of Engineering, Air Force Institute of Technology, Wright Patterson AFB, OH.
Corrosionsource (2011) “Corrosion Sensors,” http://www.corrosionsource.com/technicallibrary/corrdoctors/
Modules/MonitorBasics/Types.htm. Site accessed February 23, 2011.
El-Sheimy, N., Taha, M., and Niu, X., (2004) “Next Generation Low Cost MEMS Based Sensors: Challenges
and Implementation in SHM Systems,” Proceedings of 2nd International Workshop on Structural Health
Engineersedge (2011). “LVDT Sensors,” http://www.engineersedge.com/instrumentation/lvdt_linear_voltage_
displacment_tranducer.htm. Site accessed February 23, 2011.
Ettouney, M. and Alampalli, S., (2000) “Engineering Structural Health,” Proceedings, Structural Engineers
World Congress, Philadelphia, PA.
Ettouney, M. and Alampalli, S., (2012) Infrastructure Health in Civil Engineering, Applications and
Ettouney, M., Daddazio, R., and Hapij, A., (1999) “Optimal Sensor Locations for Structures with Multiple
Loading Conditions,” Proceedings, International Society of Optical Engineering Conference on Smart
Structures and Materials, San Diego, CA.
Evans, K. F., Walter, S. J., Heymsfield, A. J., and Deeter, M. N., (1998) “Modeling of Submillimeter Passive
Remote Sensing of Cirrus Clouds,” Journal of Applied Meteorology, 37(2), 184–205.
Fuchs, P. and Jalianoos, F., (2008) “Thermal Stress Measurements on Structural Defects,” NDE/NDT for
Highwaya and Bridges: Structural Materials Technology (SMT), ASNT, Oakland, CA.
Gage-Technique (2011) “Demountable Strain Gauges” http://www.gage-technique.demon.co.uk/pdf/gti-
straingaugetheory.pdf. Site accessed February 23, 2011.
Gangone, M. V., Whelan, M. J., Fuchs, M. P., Janoyan, K., D., and Minnetyan, L., (2006) “Implementation
and deployment of a structural health monitoring (SHM) system for bridge superstructures” Presented at
Transportation Research Board, TRB, Washington, DC.
Gucunski, N., Rascoe, C., Parrillo, R., and Roberts, R ,(2008) “Comparative Study of Bridge Deck Condition
Assessment by High Frequency GPR,” NDE/NDT for Highways and Bridges: Structural Materials
Technology (SMT), ASNT, Oakland, CA.
Hag-Elsafi, O., Kunin, J., and Alampalli, S., (2003) “In-Service Evaluation of a Concrete Bridge FRP
Strengthening System,” Report 139, Transportation Research and Development Bureau, New York State
Department of Transportation, Albany, NY.
Hannah, R. L. and Reed, S. E., (1991) “Strain Gage User’s Gandbook.” Society for Experimental Mechanics,
Bethel, CT.
Hay, T., Jayaraman, S., Ledeczi, A.,Volgyesi, P., and Sammane, G., (2008) “Wireless Technology for Steel
Bridges,” NDE/NDT for Highways and Bridges: Structural Materials Technology (SMT), ASNT,
Oakland, CA.
Hecht, J., (2006) Understanding Fiber Optics, Prentice Hall, New Jersey, NJ.
Housner, G. and Masri, S. (1996) “Structural Control Issues Arising from the Northridge and Kobe Earthquakes,”
Proceedings of 11th World Conference on Earthquake Engineering, Mexico, Paper No. 2009.
Huston, D., (2002) “Electromagnetic Interrogation of Highway Structures,” Proceedings, NDE Conference on
Civil Engineering, ASNT, Cincinnati, OH.
Huston, D., (2010) Structural Sensing, Health Monitoring and Performance Evaluation, CRC Press, Boca
Raton, FL.
ISIS Canada., (2006), Civionics Specifications, ISIS Canada Corporation, Manitoba, Canada.
Kao, T. and Taylor, H., (1996) “High-sensitivity intrinsic fiber-optic Fabry-Perot pressure sensor,” Optics
Letters, 21(8), 615–617.
Kenderian, S., Djordjevic, B., Cermiglia, D., and Garcia, G., (2006) “Dynamic Railroad Inspection Using the
Laser-Air Hybrid Ultrasonic Technique,” NDT in Rail Industry, Insight, 48(6).
Kohler, B., (2006) “Dispersion Relations in Plate Structures Studied with a Scanning Laser Vibrometer,”
ECNDT 2006, We.2.7.1, pp 1–11.
Kommareddy, V., Peters, J., and Hsu, D., (2004) “Air-Coupled Ultrasonic Measurements in Composites,”
Proceedings, SPIE 3rd International Conference on Experimental Mechanics, Singapore.
Lynch, J. P., Kamat, K., Li, V. C., Flynn, M. P., Sylvester, D., Najafi, K., Gordon, T., Lepech, M., Emami-Naeini,
A., Krimotat, A., Ettouney, M., Alampalli, S., and Ozdemir, T., (2009) “Overview of a Cyber-enabled
Wireless Monitoring System for the Protection and Management of Critical Infrastructure Systems,”
SPIE Smart Structures and Materials, San Diego, CA.
Lynch, J. and Loh, K., (2006) “A Summary Review of Wireless Sensors and Sensor Networks for Structural
Health Monitoring,” Shock and Vibration Digest, 38(2), 91–128.
Lynch, J. P., Wang, W., Loh, K. J., Yi, J-H., and Yun, C-B., (2006) “Performance Monitoring of the Geumdang
Bridge using a dense network of high-resolution wireless sensors” Smart Materials and Structures,
Institute of Physics Publishing, 15, 1561–1575.
Macrosensors (2011), “LVDT Sensors” http://www.macrosensors.com/lvdt_macro_sensors/lvdt_applications/
temperature_effects.html. Site accessed February 23, 2011.
McCormac, J. C. (1992) Structural Steel Design: ASD Method, Harper Collins Publishers, New York.
McCormac, J. C. (1995) Structural Steel Design: LRFD Method, Harper Collins Publishers, New York.
McNiel, D. and Card, L., (2004) “Novel Event Localization for SHM Data Analysis,” Proceedings of 2nd
International Workshop on Structural Health Monitoring of Innovative Civil Engineering Structures,”
Mercado, E. J. and Rao, J. R., (2006) “The Pneumatic Scour Detection System,” NDE Conference on Civil
Engineering, ASNT, St. Louis, MO.
Metje, N., Chapman, D. N., Rogers, C. D. F., Kukureka, S. N., Miao, P., and Henderson, P. J., (2004) “Structural
Monitoring using Optical Fibre Technology,” Proceedings, NDE Conference on Civil Engineering,
ASNT, Buffalo, NY.
Nelson, R. B., (1976) “Simplified Calculations of Eigenvector Derivatives,” AIAA Journal, 14(9), 1201–1205.
Neubauer, S., Hemphill, D., Phares, B., Wipf, T., Doornik, J., Greimann, L., and Monk, C., (2004) “Use of
Fiber Bragg Gratings for the Long Term Monitoring of a High Performance Steel Bridge,” Proceedings,
NDE Conference on Civil Engineering, ASNT, Buffalo, NY.
Newton, R., (1999), “Construction of a Chemical Sensor/Instrumentation Package Using Fiber Optic and
Miniaturization Technology,” NASA Report Number TM-1999-209732, Marshall Space Flight Center,
Alabama.
Northrop, R. B., (2005) Introduction to Instrumentation and Measurements, Taylor and Francis, New York.
Norton, H. N., (1989) Handbook of Transducers, Prentice Hall, Englewood Cliffs, NJ.
Ojio, T., Saito, Y, Yamada, K, and Shiina, S., (2004) “Development of ‘Strain Probe’ By Frictional Type Strain
Gauge,” Proceedings of 2nd International Workshop on Structural Health Monitoring of Innovative Civil
Engineering Structures,” ISIS Canada Corporation, Manitoba, Canada.
Oka, K., Ohno, H., Kurashima, T., Matsumoto, M., Kumagai, H., Mita, A., and Sekijima, K., (2000) “Fiber
Optic Distributed Sensor for Structural Monitoring,” Proceedings of 2nd International Workshop on
Structural Health Monitoring, Stanford University, Stanford, CA.
Omega (2011) “OMEGA Engineering Technical,” http://www.omega.com/prodinfo/loadcells.html. Site
accessed February 23, 2011.
OZOptics., (2006) “Fiber Optic Distributed Brillouin Sensor Applications,” www.ozoptics.com. Site accessed
February 27, 2011, ON, Canada.
Papadimitriou, C., Beck, J. , and Au, S. (2000) Entropy-Based Optimal Sensor Location for Structural Model
Updating, J. Vib. Control, 6(5), 781–800.
Park, H., Jung, H., and Baek, J., (2004) “Safety Monitoring Technique for Steel Beams Using Long-Gauge
Fiber Optic Sensors,” Proceedings of 2nd International Workshop on Structural Health Monitoring of
Innovative Civil Engineering Structures,” ISIS Canada Corporation, Manitoba, Canada.
PCB (2011) “PCB Piezotronics Tech Support Documents,” http://www.pcb.com/techsupport/tech_indaccel.
php. Site accessed February 23, 2011.
Perry, C. C. and Lissner, H. R., (1962) “The Strain Gage Primer.” McGraw-Hill Book Company, 2nd Edition.
Pendat, J. and Piersol, A. (2010) Random Data: Analysis and Measurement Procedures, Wiley, New York.
Ravindran, A., Phillips, D., and Solberg, J., (1987) Operations Research-Principles and Practice, John Wiley
& Sons, New York.
Rivera, E., Mufti, A., and Thomson, D., (2004) “CIVIONICS Specifications for Fibre Optic Sensors for Structural
Health Monitoring,” Proceedings of 2nd International Workshop on Structural Health Monitoring of
Innovative Civil Engineering Structures,” ISIS Canada Corporation, Manitoba, Canada.
Rong, A. Y. and Cuffari, M. A., (2004) “Structural Health Monitoring of a Steel Bridge using Wireless Strain
Gages,” Proceedings, NDE Conference on Civil Engineering, ASNT, Buffalo, NY.
Sanyal, D., (2007) “Laser Induced Ultrasonic Characterization of FRP Composites,” Proceedings, International
Conference on Advanced Materials and Composites, ICAMC.
Satou, T., Kaneda, Y, Nagaoka, S., and Ogawa, S. (2006) “Development of Wireless Sensors Using RFID
Techniques for Concrete Structures,” NDE Conference on Civil Engineering, ASNT, St. Louis, MO.
Scalea, L., Rizzo, P., Coccia, S., Bartoli, I., Fateh, M., Viola, E., and Pascale, G., (2005) “Non-Contact
Ultrasonic Inspection of Rails and Signal Processing for Automatic Detection and Classification,” NDT
in Rail Industry, Insight, 47(6), 346–353.
Sensorland (2011), “LVDT Sensors,” http://www.sensorland.com/HowPage006.html. Site accessed February
23, 2011.
Shah, P. and Udwadia, F. E., (1977) “A methodology for optimal sensor locations for identification of dynamic
systems,” Journal of Applied Mechanics, 45(2).
Shenton, H. W., Jones, R., and Howell, D. A., (2004) “A Web-Based System for Measuring Live Load Strain in
Bridges,” Proceedings, NDE Conference on Civil Engineering, ASNT, Buffalo, NY.
Siddiolo, A. and Maev, R., (2006) “An Air-Coupled Ultrasonic Techniques for NDE of Ancient Paintings,”
ECNDT 2006, Th.2.4.4.
Singer (2011), “LVDT Sensors,” http://www.singer-instruments.com/tutorials/lvdt/lvdt_tut.html. Site accessed
February 23, 2011.
Solodov, I. and Busse, G., (2006a) “Mapping of Elastic Anisotropy with Air-Coupled Phonon-Focusing of
Guided and Surface Waves,” ECNDT 2006, We.2.7.1.
Solodov, I. and Busse, G., (2006b) “New Advances in Air-Coupled Ultrasonic NDT Using Acoustic Mode
Conversion,” ECNDT 2006, We.2.4.2.
Telscho, K. and Larsen III, J., (2003) “Determination of Lateral Mode Dispersion from Full-Field Imaging of
Film Bulk Acoustic Resonator Motioh,” 2003 IEEE Ultrasonics Symposium.
Tennyson, R., Manuelpillai, G., and Cheng, J., (2002) “Application of Fibre Optic Sensors for Monitoring
Pipelines,” Proceedings of 1st International Workshop on Structural Health Monitoring of Innovative
Civil Engineering Structures,” ISIS Canada Corporation, Manitoba, Canada.
Tennyson, R. and Morison, W., (2004) “Fiber Optic Structural Health Monitoring System for Pipelines,”
Proceedings of 2nd International Workshop on Structural Health Monitoring of Innovative Civil
Texier, S. and Pamukcu, S., (2003) “Fiber Optic Sensor for Distributed Liquid Content Quantification in
Subsurface,” International Symposium on Field Measurements in GeoMechanics, NGI, Oslo, Norway.
Thompson, D. and Bridges, G., (2002) “Smart Aggregate,” Proceedings of 1st International Workshop on
Manitoba, Canada.
Timoshenko, S., (1955) Vibration Problems in Engineering, McGraw-Hill, NY.
Udwadia, F. E., (1994) “Optimal Sensor Locations for Structural Identification,” Journal of Engineering
Mechanics, 110(2), 1757–1770.
Washer, G., Bolleni, N., Fenwick, R., and Alampalli , S., (2008) “Environmental Factors for the Thermographic
Inspection of Highway Bridges,” NDE/NDT for Highways and Bridges: Structural Materials Technology
(SMT), ASNT, Oakland, CA.
Weber, B., Paultre, P., and Proulx, J., (2004) “Damage Detection in a Two-Story Concrete Building:
Minimizing the Effect of Measurement Errors,” Proceedings of 2nd International Workshop on
Manitoba, Canada.
Webester, J. G., (1999) The Measurement, Instrumentation and Sensors Handbook, CRC Press, Boca
Raton, FL.
Wu, Z. and Xu, B., (2002) “infrastructural Health Monitoring with BOTDR Fibre Optic Sensing Technique,”
Proceedings of 1st International Workshop on Structural Health Monitoring of Innovative Civil Engineering
Structures, ISIS Canada Corporation, Manitoba, Canada.
Yamaura, T, Inoue, Y, Kino, H., and Nagai, K., (2000) “Development of Structural Health Monitoring System
Using Brillouin Optical Time Domain Reflectometer,” Proceedings of 2nd International Workshop on
Zeng, X., Yu, Q., Ferrier, G., and Bao, X., (2002) “Strain and Temperature Monitoring of A Concrete Structure
Using A Distributed Brillouin Scattering Sensor,” Proceedings of 1st International Workshop on Structural
Health Monitoring of Innovative Civil Engineering Structures, ISIS Canada Corporation, Manitoba,
Canada.
Zoëga, A. and Wiggenhauser. H., (2008) “Propagation Time of Elastic Surface Waves on Concrete Specimens
under Uniaxial Loads,” NDE/NDT for Highways and Bridges: Structural Materials Technology (SMT),
ASNT, Oakland, CA.
Zou, L., Ravet, F., Bao, X, Chen, L., Huang, R., and Khoo, H., (2004) “In-Line Inspection of Pipeline Buckling
by Distributed Brillouin Scattering Sensor,” Proceedings of 2nd International Workshop on Structural
Canada.
6 Structural
(STRID)
Identification
6.1 INTRODUCTION
6.1.1 Two-Dimensionality of Modern STRID Techniques
Analysis of structures has followed a generalized Newton’s 2nd law of motion since early times.
Simply stated, the analyst would estimate some properties of the structure under consideration, and
the exciting entity (generally in form of a force). Using a generalized equilibrium equation (which is
a generalization of Newton’s 2nd law of motion), the structural response is computed. Obviously, the
accuracy of such an approach depends on the accuracy of the estimation of the structural proper-
ties. Given the complexities of modern structures and the great demands of natural and manmade
hazards, it is clear that traditional estimates of structural properties needs to be improved. The high
costs of construction, rehabilitation, and failure of infrastructure highlights the need for an even
higher degree of accuracy of estimating (identification) the structural properties.
A technique of improving the accuracy of estimating structural properties is emerging in the
form of structural identification (STRID) through field testing and monitoring. Thus STRID can be
considered as an expansion of traditional Newton-based structural analysis into an experimental-
theoretical level. In a way, STRID adds another dimension to the traditional structural analysis.
To explain further: the original dimension of the traditional structural analysis as based only on
theoretical identification of structures. Modern STRID techniques add to this another dimension of
experimental analysis (Figure 6.1).
6.1.2 Definition of STRID

The subject of structural identification has been known in several fields such as aerospace and
mechanical engineering for long time. Civil infrastructure community started investigating the sub-
ject in the 1970s. STRID is the process of identifying unknown parameters of a system by using
some experimental results. There have been many more complex definitions; however this definition
is simple, yet comprehensive. We will follow it throughout this volume. The simplest example of a
structural identification process is the single spring construct,
KU = P (6.1)
The structural identification problem here is to identify the structural property K (the stiffness)
where P is the force and U is the displacement. Traditionally, the stiffness K is estimated using some
kind of theoretical basis, resulting in an estimate of K Theory. Modern STRID provide an experimen-
tal approach of estimating K. We can devise a simple experiment where we apply a force Pex on the
system, and measure the resulting displacement Uex. The identified unknown property is then
Pex (6.2)
K ex =
U ex
Obviously, the experimental stiffness is more accurate than the theoretical stiffness: it represents the
real value of the stiffness. When using Equation 6.2 to estimate displacements with the measured
281
Tradit
io
only o nal structu
n theo ra
on prope ry to l analysis
also o rt es re
re lies esults t uctural condit ies when e timate stru lied
T R ID ntal r of str ions ( xecut
in ctural
S e physic N ge
rim uracy ed in s) ewton
’s 2nd quilibrium
expe ve acc t are us aking Law o
r o a f
imp r ties th cision m 1st dim
p e e l
r o
p n an d d enta e
of pro nsion: Th
s i g p e rim ion pertie eoretic
de : Ex stimat s
sion e al estim
d imen timize ation
2nd s to op
lt
resu perties
ro
of p
2-Dimensionality of STRID
FIGURE 6.1 2-Dimensionality of STRID.
stiffness value, the improved displacement estimate would naturally be more accurate. The differ-
ence between the theoretical and experimental estimations provides the benefit of the STRID effort.
Figure 6.2 shows the STRID process.
6.1.3 Objectives of STRID

There are several uses and objectives of STRID, and the following are among them:
Condition Assessment of Structure: This includes design validation of new and existing struc-
tures: For new structures, identifying and verifying new techniques might be needed. For existing
structures, rehabilitation effectiveness can be verified by STRID efforts. Load testing of bridges and
other types of infrastructure sometimes utilize STRID techniques for further verification of bridge
condition.
Construction Support: STRID can aid in verifying construction safety and adherence to plans.
Time-Dependent Changes: These include changes in demands such as load and changes in capac-
ity due to environmental deterioration or changes in governing design codes.
Analytical Model Updating for New and Existing Structures: Modern infrastructure are com-
plex and involved; this is reflected by the computer models that are used to analyze the structure.
STRID can aid in updating these models so that they are up-to-date and accurate.
Damage Identification of New and Existing Structures: Diligent monitoring of structural health
necessitates identifying damages and deterioration as early as possible to reduce costs of rehabilita-
tion, and ensure safety.
Accurate LCA: Life cycle analysis (LCA) is emerging as an essential infrastructure management
tool.
6.1.4 General Model of Structural Identification

There are perhaps hundreds of different quantitative methods and techniques that address structural
identification issue. Moreover, there are additional countless qualitative personal approaches that
handle the issue. One of our objectives of this chapter is to try to have a unified approach to the
structural identification issue, both quantitatively and qualitatively, from SHM vintage point.
Structural Identification (STRID) 283
Numerical
Given structural structural
information identification
algorithm
SHM data that would

affect structural
identification process Validate
Build model model
• Input
• Excitations
• Forces
• Etc.
Iterate until model is

• Output validated
• Responses
• Strains
• Velocities
• Etc. Validation parameters
• Geometry
• Material
• Etc.
Additional validation
parameters from SHM No Acceptable Yes
experiments validation?
An adequate
numerical
model is
obtained
FIGURE 6.2 Process of structural identification with SHM experiments.
Perhaps the very general structural scheme of Figure 6.2 can help in devising a general model for
structural identification issue. It simply states that for any situation, there is an input I, output O, and
a systematic relationship S between them. Formally, the relationship can be written as
SI =O (6.3)
Either of the input or output, or any combination of the two, can be measured during a nondestruc-
tive test (NDT) or SHM experiment. Similarly, they can be estimated by an analytical means.
Structural identification problem can now be defined as
Evaluate the components of S that minimize |∆I| and |∆O|, subjected to set of constraints C(IC, OC, SC).
Δ I = I MEASURED − I COMPUTED (6.4)

Δ O = OMEASURED − OCOMPUTED (6.5)

The measured inputs and outputs, IMEASURED and OMEASURED are the inputs and outputs that are
obtained as results of an NDT or SHM experiments. Similarly, the computed inputs and outputs
ICOMPUTED and OCOMPUTED are the inputs and outputs that are obtained as results of analytical evalu-
ations. It is clear that structural identification problem, as described above, is an optimization prob-
lem. The result of such an optimization problem is the optimized components of S = SOPTIMUM.
It was mentioned earlier that the structural identification problem can be either qualitative, or
quantitative. Qualitative approach to structural identification is usually done by best guess, engi-
neering judgment, or past experiences. It is usually accurate for simple systems, or when very lim-
ited set of measurements are available. When the system becomes more complex, or when the
measured data set is large, quantitative structural identification methods are utilized.
6.1.5 General Methods of STRID

Adding the second dimension of experimental basis to the conventional dimension of theoretical
basis to form STRID methodology created several categories of STRID. The confluence of time
(frequency) and space within the theory and tests helped in creating these STRID categories. They
are as follows
• Modal identification (MI)

• Parameter identification (PI)
• Scale independent elements (SIE)
• Dispersion curves (DC)
• Statistical energy analysis (SEA)
• Artificial neural networks (ANN)
• Other miscellaneous methods
We explore each of these methods in more detail in this chapter.
6.1.6 Static versus Dynamic STRID

There are two major categories for STRID: statics based and dynamics based. As the names imply,
the static based STRID uses static structural principles, mainly static equilibrium principles, to
achieve its objectives. The dynamics based STRID utilize dynamics principles such as modal analy-
sis or frequency domain analysis. Static based STRID is used in day to day bridge management
activities such as bridge load rating and different bridge calibration procedures. This is mainly due
to the simplicity of static based STRID. Dynamics based STRID offers potential for more accuracy
and better insight to the bridge behavior. Of course, there is a case to be made that static based
methods are simply a subset of dynamics based methods. For these reasons, we concentrate in most
of this chapter on dynamics based methods. Discussion of some static based methods is offered in
Section 6.4.4.1. We also note that artificial neural networks (ANN) can be applied to both static and
dynamic STRID problems.
6.1.7 This Chapter

This chapter will discuss general aspects of STRID. These include metrics and methods. Of
importance is, how space and time (frequency) attributes of the structure can affect choice
and accuracy of the appropriate STRID method. We also argue that the metrics of structural
response will vary as time and space ranges vary; thus affecting the STRID methodology. The
following sections of this chapter will present in detail each of the STRID methods. Modeling
techniques and their effects on STRID accuracy are presented next. Finally, decision making
issues as they relate to STRID subject are discussed. Specifically: the importance of STRID to
Structural identification (STRID): Layout of chapter
Overview of methods General methods of Modeling techniques Decision making and

STRID STRID
Classifications of Modal Identification Soil-Structure Life cycle analysis

STRID (MI) Interaction (LCA)
• Metrics
• Methods Parameter Fluid-structure Cost–benefit
identification (PI) interaction aspects of STRID
Effects of space Scale independent Static

and frequency elements (SIE) condensation
ranges on STRID
process
Egnergy-based
methods, SEA, Scaling needs
EFEM
Dispersion Multihazards
relations considerations
Artificial neural
networks (ANN)
Other?
FIGURE 6.3 Structural identification chapter layout.
the emerging issue of LCA is presented. Also, recognizing that STRID efforts can be costly, the
cost–benefit of STRID efforts is discussed. Simple methods are offered that can help the deci-
sion maker in estimating the value (cost vs. benefit) of an STRID effort. The contents of this
chapter are shown in Figure 6.3.
6.2 STRID PROCESSES

6.2.1 Generalized Classifications of STRID
Structural identification methods can be categorized into several general methods. The question, of
course, is how is the structure identified? Each STRID method purports to identify the structure by
using different metrics. Table 6.1 shows STRID general methods and some of the popular metrics
each method uses to identify the structure.
The choice of an STRID method obviously depends on the desired metric. The reverse is true.
In addition, we note that STRID methods are dependent on time and frequency scales, as discussed
next.
6.2.2 STRID: Interrelation of Time and Space Modeling Issues

6.2.2.1 Frequency and Space Scales
6.2.2.1.1 Global Response and Damage
Global response measure of a system can be described by global modes or global metrics such as
displacements, natural frequencies of the system. When STRID method is based on subdividing the
structure into smaller components (such as in finite element method, or statistical energy analysis
method), the size of the components should be related to the frequency range of interest as well as
the objective of the identification. For example, let us assume that a STRID project aims at identi-
fying the degree of exposure of piles due to scour. The mode shapes of the piles that are sensitive
TABLE 6.1
Methods of STRID
STRID Method Identification Metrics
Modal identification (MI) Mode shapes, natural frequencies, modal damping
Parameter identification (PI) Material properties, properties of modeling components
such as springs, masses and joint/connection properties
Scale independent elements (SIE) Material properties, properties of modeling components
such as springs, masses and joint/connection properties
Also, properties of wave mode shapes and wave numbers
Dispersion curves (DC) Frequency/wave velocity relationships
Statistical energy analysis (SEA) Modal density, loss factors, and coupling factors
Artificial neural networks (ANN) Relationships between input and output parameters
B
L
(a) Baseline geometry of plate
Crack
B
L
(b) Damaged plate with crack
FIGURE 6.4 Local damage example (a) baseline geometry, and (b) damaged plate with crack.
most to scour erosion are those that have the same characteristic length as the scour depth. Assume
that the ratio of the total pile height to the exposed pile height is n. For accurate identification the
following must be executed:
• The test and the follow-up identification should be performed up to a frequency range of
about nf0. The first natural frequency of the pile is f0.
• If a structural modeling is used (parameter identification of a finite element model), each
pile should include at least n elements.
Failing to follow the above rules at a minimum will result in an inaccurate identification.
6.2.2.1.2 Local Response and Damage

Let us consider the thin rectangular plate shown in Figure 6.4. Figure 6.4a shows the pristine plate
and Figure 6.4b shows a plate that is slightly damaged by a small edge crack. The crack has a length
of αℓ, where α is a scalar constant and ℓ is a length measure. The plate length is L and the width
of the plate is B. We are interested in studying the modal behavior of the plate and the assumed
crack.
When the two plates of Figure 6.4 vibrate, their vibration modes will be almost identical, with
the exception of the small dynamic effects of the crack. If the sources of vibration are located away
from the crack the dynamic effects of the crack presence will result in a dynamic pressure bulb as
shown in Figure 6.5. The dynamic pressure bulb DPB is defined as
p
DPB = 1 − 1
p0
Where p 0 and p1 are the pressures at a given location for the pristine and the damaged plate,
respectively. The idea of dynamic pressure bulb can be visualized as a generalization of the static
pressure bulb in soils under building foundations: the extent of the effects of the static (or dynamic)
soil-structure interaction (SSI). The DPB can be used to explore the extent of the dynamic effects
of the crack. The size of DPB depends on the frequency of vibration Ω and the wave number λ.
Similarly, we can argue that the effects of the trenches in soils (which is similar geometrically to
the crack problem of Figure 6.5) are dependent also on the size and shape of the trench. From the
above we can claim that
DPB = g ( Ω, l, ) (6.6)

The above relationship is important in the field of structural and damage identification; there are
two reasons for this importance:
• Placing any sensors outside the DPB would not effectively measure the dynamic effects of
the crack (or damage)
• Whenever using modal decomposition in structural identification, the DPB that is active
during the corresponding modal frequency will be the main clue as to the existence of the
damage. As such, it would be of great interest to study the frequency range that maximizes
the size of the DPB
Let us try to estimate the extent of the DPB. It is not an easy task, since there are limited results
that can help in such an estimate. However, we can qualitatively assign a maximum size of a DPB,
LDPB_MAX, to be
LDPB _ MAX = b ( a ) (6.7)

A value of β = 20 seems to be reasonable. Such an upper limit of β should be confirmed on a case
by case basis for important experiments.
For small size cracks, say a 0.1 in, the estimated maximum dynamic effects of such a crack
is about 2.0 in. If such a crack exists in a plate with a depth and length of 25 in and 120 in.,
Crack
Contour line 3.0
Contour line 2.0
Contour line 1.0
Contour line 0.01
B
Dynamic pressure bulb
FIGURE 6.5 Example of dynamic pressure bulb.

respectively, its effects will not be easily discernable in most modes in a modal analysis approach.
The only time where such an effect can be easily observed is when the mode of vibration is mostly
the local damaged area (crack), that is, when the crack vibrates, while the motion of the rest of the
plate is negligible. Such a mode is shown qualitatively in Figure 6.6.
Estimating the frequency range of such a local mode is fairly simple. Let us assume that the mass
density is ρ, the equivalent vibrating width is Wequ, the equivalent vibrating height is Hequ, and the
shear modulus is G. From Figure 6.5 we can estimate the vibrating mass to be
M = r Wequ H equ t (6.8)

and the equivalent stiffness
H equ
K =G (6.9)
2
Thus, the resulting frequency is
1 K
f =
2p M (6.10)
For the same 0.1-in crack length example above, we estimate the equivalent width and the equiv-
alent height of the vibrating mass to be 0.07 in and 0.07 in, respectively. Assuming a unit plate
width, the local frequency is estimated as 1.57 MHz. At crack length of 10 in the local frequency
is 1.57 KHz.
The above example leads to several conclusions
• To detect a local damage experimentally, such as crack, sensors must be within the local
dynamic pressure bulb.
• In addition to the above, the sampling rate of such measurements must be consistent with
the expected frequencies of the crack size.
• To detect a crack through analytical structural identification methods, the employed ana-
lytical or numerical model must have high resolution to accommodate local frequencies.
This includes the size of elements (if a finite element method is utilized) and the modal
frequency range.
• If the resolution (element size or frequency range) is not consistent with demands of the
crack (or damage) size, then the structural identification method that rely on finite element
Vibrating local mass Assumed local

vibrating mode
Assumed crack
length
Crack
FIGURE 6.6 Local vibration of the crack.

modeling and/or modal analysis will result on averaging the effects of the DPB. Such an
averaging might give some clues on the existence of damage. Moreover, in the absence of
the required resolution, it is impossible to accurately estimate the damage location.
6.2.2.2 Objectives of STRID and Frequency-Space Domain

There are many objectives of STRID and some of those objectives were mentioned earlier.
We can summarize those objectives into two basic ones: design verification and damage iden-
tification. The interrelationships between STRID objectives and frequency/space scales are
discussed next.
6.2.2.2.1 Design Verification

Design verification, which includes parameter identification, system response during construction,
or during normal operating conditions, or during abnormal hazards, is a major objective of STRID.
Design parameters can be global (displacements, acceleration, overall stiffness, etc.) or local, such
as local stiffness of a connection or weld and, local material property (modulus or elasticity, crack
length, etc.). Obviously, methods of STRID should change as frequency and space scales change to
identify those design parameters in an efficient and accurate fashion. Figure 6.7 shows the effects of
frequency and space scales on STRID and design verification parameters.
Damage identification (DMID) is also one of the major objectives of STRID. The interrelation-
ship between STRID and DMID depends on the frequency and space scales, as shown in Figure 6.8.
There are three modes of STRID-DMID interrelationship. At lower frequency and larger space
scales, to identify a structural damage, two structural states need to be identified. The damage can
be identified as the difference between the two states. At high frequency and low space scales, the
damage can be identified directly. In these conditions, STRID and DMID coincide. Another cat-
egory at which STRID can be used for DMID is by using virtual sensing paradigm (VSP): The
damage can be related to structural conditions by predetermined set of functions. VSP concepts
are discussed elsewhere in this volume. They are most efficient at mid-frequency and mid-space
scales.
Global design Global design

Decreased
parameters parameters
accuracy
such as global can be
Global
Global
flexural inaccurate
stiffness
Intermediate
Intermediate
Spatial scale
Localized design Decreased Localized

parameters accuracy design
Local
Local
can be parameters such

inaccurate as local E, ρ, k
Low Medium High
Frequency scale
Design verification parameters depend on

frequency and spatial scales; this controls
the type of STRID method used.
FIGURE 6.7 Design parameters in STRID frequency-space domain.

Remote
sensing
1 technology,
Global
1
Global
e.g., laser
3
1 1 2
Intermediate
Intermediate
Spatial scale
2
2 3 3
2
Local 3
Local
2
3
Low Medium High
Frequency scale
DMID-STRID interrelationship depends on frequency and
spatial scales as follows:
1 To identify damage, two states of structure needs to be

identified
2 To identify damage, VSP or ANN can be used
3 Damage can be identified directly
FIGURE 6.8 DMID methods in STRID frequency-space domain.
TABLE 6.2
Methods of STRID
STRID Method Comment
Modal Modal identification is limited to relatively lower frequency range. As number of modes increase,
identification (MI) the accuracy of their identification is reduced. This also limits the method into large space scales
Parameter Parameter identification is limited to relatively lower frequency range due to computational
identification (PI) demands and modeling accuracies. As space scale decreases, those modeling and computational
demands increase this limit and the method moves into larger space scales
Scale independent Scale independent elements method reduces the frequency and scale demands somewhat, thus
elements (SIE) permitting the method to be used at mid-frequency and space scale ranges
Dispersion curves Dispersion curves can traverse into high frequency ranges and simultaneously to low space scales
(DC)
Statistical energy SEA is fairly suited to be used at medium and high frequency scales at almost any space scale
analysis (SEA)
Artificial neural ANN does not aim to model the physical system properties. It relates input to output parameters
networks (ANN) directly. Thus it can be used at any frequency or space scales
6.2.2.3 Methods of STRID and Frequency-Space Domain

The efficiency and accuracy of the general categories of STRID methods are directly dependent
on the frequency and time scales. Table 6.2 describes the frequency dependency of STRID gen-
eral categories. Figure 6.9 shows the relative applicability frequency spectrum of STRID methods.
Clearly, care is needed when using an STRID method to ensure its compatibility of the desired
frequency range.
When we consider the space scale, the interdependency of the STRID methods on the fre-
quency space becomes more intricate as shown in Figure 6.10. As the frequency and space
A.N.N.
S.E.A.
D.C.
S.I.E.
P.I.
M.I.
Low Medium High
Frequency scale
Frequency ranges of STRID techniques
FIGURE 6.9 Frequency ranges of STRID methods.
M.I. S.I.E. S.E.A.

P.I. M.I. D.C.
Global
A.N.N. P.I. A.N.N.
Global
A.N.N.
S.I.E. S.I.E. S.E.A.
Intermediate
Intermediate
M.I. D.C. D.C.

Spatial scale
P.I. A.N.N. A.N.N.

A.N.N.
A.N.N. D.C. S.E.A.

A.N.N. D.C.
Local
Local
A.N.N.
Low Medium High
Frequency scale
STRID techniques vary as frequency and

space ranges change
FIGURE 6.10 STRID techniques in frequency-space domain.
change, the applicable STRID methods change. As of the writing of this volume, the boundar-
ies of applicability has not been sufficiently studied, further understanding of this problem is
needed.
6.3 MODAL IDENTIFICATION METHODS

6.3.1 Theoretical Background
6.3.1.1 Overview
6.3.1.1.1 General
Modal identification (MI) methods aim at utilizing in field measurements to estimate modal values
of the structure. The input to these methods is generally time (or frequency)-dependent excitations.
The excitation can be ambient (no specific input), single source, or multiple sources. The measure-
ments also vary greatly. They can be single or multiple measurements. Figure 6.11 shows general
issues that affect MI efforts. The output of MI efforts will include estimated frequencies, mode
shapes, and modal damping. These modal parameters can then be used directly in analysis, design,
Issues affecting modal structural identification
Measured input Type of experiment Measured output
No input Single input

Time response measurement
measurment
Single input Frequency Multiple input

measurement response measurement
Multiple input
Free vibration
measurement
Forced vibration
FIGURE 6.11 Issues affecting structural identification using modal methods.
DMID, or other decision making efforts. In some situations, the estimated MI parameters are used
in further STRID efforts such as parameter identification.
Among the advantages of MI are that (1) they are well established, (2) they can be used directly
in design and analysis efforts of low frequency problems such as seismic designs, (3) there is large
body of experience of MI methods, and (4) there are large number of MI methods that can be used in
numerous situations. Among the disadvantages of MI methods are the following (1) they are appli-
cable only to low frequency range, (2) nonlinear applications are limited, (3) lack of fluid-structure
interaction applications, and (4) limited use for DMID at early stages of damage formation (at early
stages, damages usually affect high frequency ranges).
6.3.1.1.2 Different Classifications

There are several types of classifications of MI methods, we explore some of those classifications next.
Experimental versus Operational MI: As the name implies, experimental MI would entail artifi-
cially exciting the system on hand by a shaker, for example. Operational MI would utilize the ambi-
ent operating conditions as the exciting mechanism, Brinker and Moller (2006). As such, no special
exciting equipments are needed. Experimental modal analysis have been around much longer, and
thus the MI methods are well established. Lately, however, many operational MI methods have
been developed, and their popularity is increasing. For bridge structures, operational modal analysis
offer several advantages: (1) They affect traffic much less than experimental MI; (2) Large bridges
can be modeled since no mechanical equipment is needed to excite the large bridge mass, (3) They
would be more cost effective, and (4) They can be included as a part of long-term SHM projects. The
details of an SHM experiment that utilize operational MI are presented later this section.
Order of MI Methods: Single degree of freedom dynamic structural equations are second-order dif-
ferential equations. For multiple modes, the matrix size N would relate to the number of modes to be
identified. It is possible to reduce the size of the matrices while increasing the order of the dynamic
TABLE 6.3
Order of Modal Methods
Experimental Conditions
Civil
Spatial Measurements: Temporal Measurements: Engineering
Model Matrix Size N Order r Number of Sensors Inertia Effects Examples
High order Large Low Small Large STRID
Low order Low Large Large or complete Medium to low STRID
Zeroth order Large Zero Large or complete Static Load testing
equation. At the limit, as the dynamic equation is reduced to a scalar equation, the order of such an
equation becomes N. There are different MI solution methods for different order of equations. Also,
the type of experiment depends on the order of the MI solution method as Table 6.3 shows. Detailed
description of model order is explained later.
Frequency versus Time Domain: MI methods can be solved either in the frequency of time
domains. Examples of both approaches are given later in this section.
6.3.1.2 Equations of Motion

Equation of motion of free vibration is expressed by
[ M ]{U} + [C ]{U } + [ K ]{U } = {0} (6.11)
The mass, damping, and stiffness matrices are [M], [C], and . [K], respectively. The displacements,
velocity, and acceleration vectors of the system are {U}, {U}, and {Ü}, respectively. The character-
istic equation using Laplace operator s is
s 2 [ M ] + s [C ] + [ K ] = {0} (6.12)

Both Equations 6.11 and 6.12 are used extensively in MI methods.
6.3.1.3 Modal Solution

For a nontrivial solution of Equation 6.11, it needs to have the harmonic form
{U } = {j} el t (6.13)

Substituting into Equation 6.11
⎡⎣ − l2 [ M ] + il [C ] + [ K ]⎤⎦ {f} = {0} (6.14)

We recognize Equation 6.14 as a quadratic eigenvalue problem, with 2N eigensolution of λi and {𝜙}i,
where i = 1,2…2N. The eigenvalue λi and the eigenvectors {𝜙}i are complex valued, in general.
We note that for an undamped system ([C] = [0]) or for proportional damping, Equation 6.14
is reduced to a linear eigenvalue problem. In such a situation, only N eigensolutions are needed to
solve the problem. These eigensolutions are real, and the well known equalities
[Φ ]T [ M ][Φ ] = I (6.15)

and
[Φ ]T [ K ][Φ ] = w (6.16)
apply.
The identity matrix is 〈 I 〉. The diagonal matrix 〈 ω 〉 contains the natural frequencies of the sys-
tem. The matrix [Φ] contains the vectors of the mode shapes {𝜙}i.
Returning back to the general problem of Equation 6.14, the displacement of a structural system
can be expanded into the series
i = 2N
{U } = ∑ Ai {f}i e li t (6.17)
i =1
where A is an arbitrarily constant that is a function of initial conditions. Some methods incorporate
the constant Ai with the mode shapes {𝜙}i to get the simpler form
i = 2N
{U } = ∑ {f}i el t i
(6.18)
i =1
Complex natural frequencies λk are related to the natural frequency ωk and modal damping by

(
Re al ( lk ) = −xk w k + i w k 1 − x2k ) (6.19)
Im ag ( lk ) = −xk w k − i (w k 1− x )
2
k
(6.20)

where i = −1.
We changed the mode count from i to k in Equations 6.19 and 6.20 to avoid confusion with the
complex constant i = −1. This indicates that by evaluating the complex natural frequency it is
possible to evaluate both the natural frequency and the modal damping. The practice is a popular
practice in MI efforts.
6.3.1.4 Frequency Response Function (FRF)

In forced harmonic vibrations with a driving frequency Ω the equation of motion becomes
⎡⎣ −Ω2 [ M ] + iΩ [C ] + [ K ]⎤⎦ {U } = {P} (6.21)

The harmonic force and displacement vectors are {P} and {U}, respectively. The harmonic displace-
ment is obtained as

{U } = [ F ]{P} (6.22)
with
−1

[ F ] = ⎡⎣ −Ω2 [ M ] + iΩ [C ] + [ K ]⎤⎦
(6.23)
The matrix [F] is called FRF. It relates applied harmonic forces to harmonic displacements of the
system. Its components are frequency dependent and complex valued, in general. The components
of FRF in the ith row and jth fij(Ω) represents the harmonic displacement of the ith degree of freedom
due to unit harmonic force in the direction of the jth degree of freedom. Physically, it represents the
dynamic flexibility of the system. It is directly related to dynamic influence lines (DIL) discussed
in Chapter 8 of Ettouney and Alampalli (2012). FRF is used extensively in MI. Measuring system
responses at different points during experiments due to harmonic forces would generate the FRF
of the system.
6.3.1.5 Impulse Response Function (IRF)

The inverse Fourier transform (see Chapter 7), of the FRF fij(Ω) is a time-dependent signal hij(t).
This signal is called IRF. It can be shown to be the time-dependent displacement response of the
ith degree of freedom of the system due to unit initial velocity in the direction of the jth degree of
freedom. The time-dependent matrix that holds all hij(t) is also used extensively in MI methods. By
applying initial velocity (impulse) to the system and measuring responses, the IRF matrix can be
established.
6.3.1.6 Pseudo Inverse

Consider the linear matrix equation
[ A]N × M { x}M = { y}N (6.24)
If M ≠ N, the matrix [A]N×M can’t be inverted by conventional means. However, there is an easy way
to solve Equation 6.24. First, remultiply by [A]T
[ A]TM × N [ A]N × M { x}M = [ A]TM × N { y}N (6.25)
Rewrite as
[ B]M × M { x}M = [ A]TM × N { y}N (6.26)
with
[ B]M × M = [ A]TM × N [ A]N × M (6.27)
The square matrix [B]M×M can be inverted by conventional method. The required solution {x}M is
obtained by

{ x}M = [ B]−M1× M [ A]TM × N { y}N (6.28)
or

{ x}M = [ A]+M × N { y}N (6.29)
with
[ A]+M × N = [ B]−M1× M [ A]TM × N (6.30)
The matrix [ A]M × N is the pseudo inverse of [A]N×M. The pseudo inverse process is used in many
+
structural identification methods, such as the eigenstructure assignment method.
6.3.1.7 Singular Value Decomposition (SVD)

Sometime expansions of matrix [A]N×M is needed in the form
T
[ A] = [U ] D [V ]

(6.31)
The matrices [U], 〈D〉, and [V]T can be evaluated using popular numerical solutions. They are related
to the eigenvalue properties of [A]N×M. These matrices are used in several MI methods. For addi-
tional information see He and Fu (2001).
6.3.1.8 Modal Assurance Criteria (MAC)

A very useful estimate of the accuracy of estimated mode shapes is the MAC. MAC is a measure
of correlation between the estimated mode shapes {𝜙}Ei and the measured mode shapes {𝜙}Mi. Note
that i = 1,2, … N and j = 1,2, … M. With N and M representing the total estimated mode shapes and
the total measured mod shapes, respectively.
The MAC is now defined as
({f} {f}Mj )
T 2
(
MAC {f}Ei , {f}Mj = ) Ei
(6.32)
({f}TEi {f}Ei ) ({f}TMj {f}Mj )
2 2

Note that MAC({𝜙}Ei, {𝜙}Mj is a scalar constant. Since it is a correlation measure, its range is between
0 and 1.0. If MAC({𝜙}Ei, {𝜙}Mj) is equal to, or close to, 1.0, then the ith estimated mode is well cor-
related to the jth measured mode. If MAC({𝜙}Ei, {𝜙}Mj) is equal to, or close to, 0 then the ith estimated
mode is not correlated to the jth measured mode.
MAC is a popular technique that is used by many practitioners and researchers, see, for example,
Padur et al. (2002).
6.3.1.9 Matrix and Vector Norms

A norm is a measure of the matrix, or vector, magnitude. Since there are many ways to express such
magnitude, there are many ways to define a norm of a matrix or vector. For example, the sum of the
magnitude of a square matrix [C] is the norm
n n
[C ] 1 = ∑ ∑ C ij (6.33)
i =1 j =1
The size of the matrix is n.

Another important norm that are used frequently in structural identification methods is the
Euclidean norm (sometimes is referred to as Frobenius norm) it is expressed as
n n
[C ] E = ∑ ∑ Cij2 (6.34)
i =1 j =1

The reader will recognize that this norm is the basic equation in any least square process. The
Euclidean norm importance in the SHM field is self evident.
Note that the norm expressions for vectors are similar to the above expressions. In case of vec-
tors, replace the double summation operators with a single summation operator.
6.3.2 Order of System Models

Perhaps one of the most important decisions in the structural identification process, as well as the
MI process is the choices of the structural degrees of freedom (DOF) and/or the number of modes
of interest to be estimated. For the purposes of this section, let us assume that the number of desired
modes to be identifies is the same as the DOF of the structural model N. The general matrix equa-
tion of motion of the system is Equation 6.11 with the size of the matrix equation is N. Laplace
transform of the equation of motion is Equation 6.12. Note that the power of the Laplace operator of
this equation is 2. We can express Equation 6.12 as
⎡M M12 ⎤ ⎡C C12 ⎤ ⎡ K11 K12 ⎤

s 2 ⎢ 11 + s ⎢ 11 ⎥ + ⎢K = {0} (6.35)
⎣ M 21 M 22 ⎥⎦ C
⎣ 21 C 22 ⎦ ⎣ 21 K 22 ⎥⎦

Where we expressed the original matrices with size N with individual submatrices with size N/2.
The new equation can be rearranged as
s 4 [C4 ] + s 3 [C3 ] + s 2 [C2 ] + s [C1 ] + [C0 ] = {0} (6.36)

The size of the above equation is N/2 where as the power of the Laplace operator is 4: we reduced
the size of the matrices, while increasing the power of the operator.
We can continue this reduction/increase process. After n processes, the generic characteristic
equation becomes
s 2 n [C2 n ] + s 2 n −1 [C2 n −1 ] + s 2 n −2 [C2 n −2 ] + + [C0 ] = {0} (6.37)

The size of the matrices is N/2n while the power of the Laplace operator is 2n. Ultimately, the char-
acteristic equation becomes scalar, with the power of Laplace operator of N such as
s N cN + s N −1 cN −1 + s N −2 cN −2 + + c0 = 0 (6.38)

The roots of this scalar characteristic equation, or any intermediate reduced equation, are the
same eigensolutions of the original system matrix Equation 6.11.
The power of the Laplace operators in any of the above characteristic equations is also referred to
as the order of the model. For example, the minimum order of the model represented by Equation 6.12
(the original characteristic equation) is 2, the highest order of the model is N, it occurs for scalar
characteristic equations.
There is important significance for the model order representation in structural identification
field. Recall that the structural identification purports to estimate modal properties relying only on
measurements. The number of stations where the measurements are made P is not always equal to
the number of modes of interest. This frequent mismatch necessitates using models with different
order for maximum computational efficiency. For example, when P is too small compared with N,
the methods with high model order might be more efficient than models with low order. There are
numerous MI methods that utilize low- or high-order models.
One of the major challenges to any MI method is the way the characteristic equation is formed,
and the relationship between such formation and the measurements. In later sections some popular
methods will be presented.
We need to state one final note regarding the reduction/increase of matrix size/power of opera-
tors. We demonstrated the concept using Laplace characteristic equations. Similar processes can be
made using the differential operators, for example, Equation 6.11 in the time domain. The charac-
teristic equations in this case are differential equations of higher order.
6.3.3 Attributes of Modal Identification methods

The model order is one of the basic attributes of MI methods. There are several other attributes for
those methods. For example, the solution domain (time domain or frequency domain). But, some
methods do utilize both time and frequency domain solutions. Another attribute is the number of
input sources and number of output stations. Input sources are the locations where the structure is
excited by an external force, usually to generate an IRF. Output stations are where the structural
response is measured, usually in the form of IRFs. If FRFs are needed by the MI method, a sim-
ple transform can be performed. There are several input/output combinations that can affect the
choice of the MI method. Among those combinations are single input source/single output station
(SISO); single input source, multiple output sources (SIMO); and multiple input sources, multiple
output sources (MIMO). We also note that there are some methods that do not rely on an artifi-
cially generated excitation source, such as the random decrement (RD) method see Asmussen,
J.C., (1997). The RD method utilizes random excitation of the structure, which make it ideal for
TABLE 6.4
Attributes of Modal Identification Methods
Method DOF Input-Output Domain Order
Peak amplitude SDOF SISO FD
Least squares SDOF SISO FD
Rational fraction polynomials MDOF SIMO FD High
Ibrahim method MDOF SIMO TD
Least squares MDOF SIMO TD
Random decrement MDOF NA TD
Autoregressive moving average (ARMA) MDOF SIMO TD
Least Squares Complex Exponential (LSCE) MDOF SIMO TD High
Poly-reference MDOF MIMO FD Low
Ibrahim method-multi-input MDOF MIMO TD Low
Poly-reference MDOF MIMO TD Low
Global modal analysis method MDOF MIMO TD
Eigensystem realization algorithm (ERA) MDOF MIMO TD High
heavy bridge structures that are difficult to excite by an artificial means. Hence, this method is
given in detail with a case study on bridges. Table 6.4 shows the different attributes of some pop-
ular MI methods. Some of those methods will be discussed briefly in the next sections. As the
number of input sources and/or output stations increase, the cost of experiment as well as the
accuracy of the results and completeness of information increase. To realize an optimal balance
between cost and benefit of the experiment that aim at MI of particular structure, a study using
the theory of experimentation is recommended. See the pertinent example in Chapter 4.
6.3.4 Peak Amplitude

For the FRF, α(ω) of an SDOF, define the range of interest. Such a range should contain a peak, or
resonance. The peak is defined by α(ω)max within the range of interest. The natural frequency that
corresponds to that resonance is ωn, where n is the nth mode of interest. The damping ratio ξn that
correspond to the nth mode can be estimated using the analytical expression for α(ω) of the SDOF.
The modal damping is related to ω1 and ω 2, which are the two frequencies at both sides of ωn as the
magnitude of α(ω) drops to a (w )max / 2 . Thus,
w 12 − w 22
xn = (6.39)
4 w 2n

The above completely identifies SDOF parameters. If the measurements involve an MDOF, then
the above procedure is applicable for every range of the FRF that contains resonance. In the MDOF
case, the identification of the modal amplitude An is needed. Recalling the frequency domain modal
solution near resonance, the modal amplitude can be defined as
An = 2 a ( w )max xn w n (6.40)

The peak amplitude method is simple, yet can result in very accurate results. For ranges of FRF
with close resonances, it might yield inaccurate results. Also, it might not yield accurate results for
highly damped systems. Such systems are prevalent in civil infrastructure, such as composite steel-
concrete bridges, or soil-structure systems. Nevertheless, due to its simplicity, this method can be
used as an approximate estimate for modal parameters.
6.3.5 Eigensolution Realization Method (ERA)

The ERA is one of the most popular MI methods. As usual, it aims at finding modal shapes, modal
frequencies, and modal damping of a given system, using a set of measurements, usually displace-
ments. We will summarize some of the essential steps of the method. For more details, the reader
should consult other references, such as He and Fu (2001), or Doebling et al. (1999).
We assume that there are N1 and N2 measurement points and external forces, respectively. We
also assume that the system to be identified has N DOF. The basic step in the ERA method is
to form Hankel matrix using available measurement points. Let us define the impulse response
vector {h}mij to be of size M. Also note that i = 1,2, … N1, and j = 1,2, … N2. Note that the vector
represents a time measurement stream at the ith measurement point due to the jth external force.
The m subscript represents the time step at which {h}mij measurements starts. Thus, the kth element
in the vector {hk}mij is the time sample measured at k+m time step at the ith point due to the jth
external force. The size of each vector M should be greater than or equal to 2N. If this condition
is not met, then a time shifting scheme can be used to fill the needed spaces in the vector to make
it at least equal to.
We can now define the basic Hankel matrix [Hm]Base as
⎡ {h}m11 {h}m12 {h}m1N

⎤
⎢ ⎥ 2
⎢ {h}m 21 {h}m 22 {h}m 2 N

⎥
[H ] m Base =⎢ 2
⎥ (6.41)
⎢ ⎥
⎢{h} {h}mN 2 {h}mN N ⎥
⎣ mN11 1 1 2 ⎦ mN × N
1 2
The base Hankel matrix [Hm] is then formed by using a time shifted [Hm]Base as submatrices as
follows
⎡ [ H m ]Base [H ] m +1 Base [H ]
m + m 2 Base
⎤
⎢ ⎥
[H ] [H ] [H ]
[ H m ] = ⎢⎢ m+1 Base m + 2 Base m + m 2 +1 Base ⎥
⎥ (6.42)
⎢ ⎥
⎣[ H m + m1 ]Base [H ] [ H m + m1+ m 2 ]Base ⎥⎦
⎢ m + m1+1 Base
m m1 N1 × m 2 N 2
The constants m1 and m2 are arbitrary. They are used to reduce the effects of measurement noise
on the accuracy of results. Note that the larger m1 and/or m2, the larger the amount of measurement
information in Hankel matrix, thus, it is expected that the resulting modal parameters will be more
accurate. There is a limit, though on how high m1 and/or m2 can be. Additional data might not
improve the accuracy. The values of m1 and/or m2 can be established by experience, or by iterative
processes.
Now Hankel matrix is completely defined, and the realization of the modal properties can pro-
ceed. He and Fu showed that if we decompose Hankel matrix that is evaluated at m = 0, [H0] using
SVD as
[ H ] = [U ] [ Σ ][V ]
T (6.43)
0
Equation 6.43 contains the singular value decomposition components of [H0], see Section 6.3.1.7.
The matrix [Σ] is a diagonal matrix of size 2N and contains the singular values of [H0]. The matrices
[V] and [U] are the right and left singular vectors of [H0], respectively. The row size of [V] is 2N and
its column size is of the same size as the size of the column of [H0]. Also, the row size of [U] is 2N
and its column size is of the same size as the size of the row of [H0].
Upon defining another Hankel matrix at m = 1, [H1], we can define the matrix [A] as
[ A] = [ Σ ]−1/2 [U ] [ H1 ] [V ]T [ Σ ]−1/2 (6.44)
Note that [A] is a square matrix with a size of 2N. The complex 2N eigenvectors of [A] are the
required system mode shapes. Each of the complex 2N eigenvalues of [A] constitutes the ithcomplex
natural frequency si of the system. They are related to the natural frequency and modal damping by
Equations 6.19 and 6.20. For more details, see He and Fu (2001).
6.3.6 Ibrahim Time Domain (ITD) Methods

As one of the popular MI techniques, the ITD is a low order method that uses the modal equation
i = 2N
{U } = ∑ {f}i el t i
(6.45)
i =1
The method is described in several references; see, for example, Ibrahim (1998). We summarize the
basic approach of the method. The method aims at computing the complex eigensolutions (complex
frequencies) λi and the associated complex modes shapes {𝜙}i. The measurements {U(tj)} include
measurements at N locations. Thus, at time tj
i = 2N
{ ( )} = ∑ {f} e
U tj i
li t j

(6.46)
i =1
The initial sample time is T0 = t1. Defining 2N measurements vector of size N each as

[U ]T 0
{ }{ } {
= ⎡⎣ U ( t1 ) U ( t2 ) U ( t2 N ) ⎤⎦

} (6.47)
We can write
[U ]T = [Φ ]T [ Λ ] (6.48)
0 0
The matrix [Φ]T contains the vectors of complex mode shapes {𝜙}i. The components of the matrix
0
[Λ] are

[ Λ ]ij = el t j

(6.49)
Note that the matrix [U]T is of size N × 2N.

0
The method defines three more response equations, each with a time shift of T1, T2 and T1 + T2
from the first response equations, thus

[U ]T 1
{ }{ } {
= ⎡⎣ U ( t1 + T1 ) U ( t2 + T1 ) U ( t2 N + T1 ) ⎤⎦ } (6.50)

[U ]T
2
{ }{ } {
= ⎡⎣ U ( t1 + T2 ) U ( t2 + T2 ) U ( t2 N + T2 ) ⎤⎦

} (6.51)

[U ]T +T
1 2
{ }{
= ⎡⎣ U ( t1 + T1 + T2 ) U ( t2 + T1 + T2 ) U ( t2 N + T1 + T2 ) ⎤⎦

} { } (6.52)
Furthermore, we can define the time-shifted mode shape vectors as
{f}i,T = {f}i el T i 1 (6.53)

1

Defining the matrix [Φ]T to contain the vectors of complex mode shapes {f}i ,T1 the T1-time-shifter
1
response matrix equation can be written as
[U ]T = [Φ ]T [ Λ ] (6.54)
1 1
The responses of T0 and T1 can be combined as
⎡[U ]T ⎤ ⎡[ Φ ]T ⎤
⎢ 0
⎥=⎢ 0
⎥ [Λ ] (6.55)
⎢⎣[U ]T1 ⎥⎦ ⎢⎣[ Φ ]T1 ⎥⎦

In short,
[Y ] = [Φ ][ Λ ] (6.56)
Not that [Y ] and [ Φ ] are 2 N × 2 N .

Repeating for initial sampling times of T2 and T1+T2.
⎡ [U ]T ⎤ ⎡ [ Φ ]T ⎤
⎢ 2
⎥=⎢ 2
⎥ [Λ ] (6.57)
⎢⎣[U ]T1 +T2 ⎥⎦ ⎢⎣[ Φ ]T1 +T2 ⎥⎦

or
⎡ [U ]T ⎤
[Y ]1 = ⎢ U ⎥ = [Φ ] 2
g [Λ ] (6.58)
⎢⎣[ ]T +T ⎥⎦
1 2
The ith component of the diagonal matrix 〈 γ 〉 is
gi = e li T1 (6.59)
We now eliminate the highly nonlinear matrix [Λ] from Equations 6.56 and 6.58
[Y ]1 [Y ]−1 [Φ ] = [Φ ] g

(6.60)
Rearranging, we reach the eigenvalue problem
[ A][Φ ] = g [Φ ] (6.61)

where
[ A] = [Y ]1 [Y ]−1 (6.62)
Upon solving the eigenvalue problem, Equation 6.61, the resulting mode shapes [Φ] will represent
the structural mode shapes. The eigenvalues γi are related to the ith modal damping and natural
frequency. See Ibrahim (1998) for details of extracting the modal damping and natural frequencies
from γi. Techniques for establishing the number of modes that contribute to the response, the mis-
match between number of measurements and the unwanted noise effects are also described in detail
by Ibrahim (1998).
6.3.7 LSCE (Least Square Complex Exponential)

Earlier we identified the characteristic equation of a dynamic system, Equation 6.38, as a high-order
model for structural identification. This characteristic equation is the basis of the LSCE for struc-
tural MI. For convenience, we rewrite Equation 6.38 for an order of 2N as
i = 2N
(6.63)
∑c i si = 0
i=0
The roots of Equation 6.63 sk represent the complex natural frequencies of the system, where k =
1, 2, … 2N. To compute the roots of the characteristic Equation 6.63, we first need to compute its
constants. Consider the solution for the IRF of Equation 6.38.
k=2N
hij ( t ) = ∑ Aijk e sk t (6.64)
k =1
As usual, we assume that the IRF hij(t) is the measured responses at the ith DOF due to a force at the
jth DOF. Assuming t = ℓ ∆t, where ∆t is the sampling time step and introducing zk = e sk Δt , we can
rewrite Equations 6.63 and 6.64) as
i = 2N
∑c i zki = 0 (6.65)
i=0
k=2N
(6.66)
hij = ∑ Aijk zk
k =1
Note that we changed the constant ci into ci in Equation 6.65. Simple multiplications and summa-
tions of Equation 6.66 produce
= 2 N k =2 N =2 N
∑c hij = ∑ A ∑c ijk zk

(6.67)
= 1 k =1 =1
From Equation 6.67, we write the left hand side of Equation 6.65 as
= 2 N
∑c hij = 0 (6.68)
= 1
Dropping the i and j subscripts for simplicity, Equation 6.67 becomes a linear combination of a mea-
sured IRF set of 2N samples. If we assume, arbitrarily without any loss of generality, that c2N = −1.0,
and repeat Equation 6.67 for 2N sets of IRF, we reach the matrix equality
[ H ] {c} = {h} (6.69)

The size of the matrices in Equation 6.69 is 2N. Vector {c} contains the unknown coefficients cℓ.
The vector {h} can be expressed as
⎧ h2 N ⎫
⎪h ⎪
{ } ⎪⎨ 2N −1 ⎪⎬
h = (6.70)
⎪ ⎪
⎪⎩h4 N −1 ⎪⎭

Also
⎡ h0 h1 h2 N −1 ⎤
⎢ h h h2 N ⎥⎥
[ H ] = ⎢⎢ 1 2 ⎥
(6.71)
⎢ ⎥
⎣h2 N −1 h2 N h4 N −2 ⎦

Equation 6.69 can be solved for the unknown coefficients cℓ. The characteristic Equation 6.65 can
be solved to find the characteristic roots zk. Using the relation between zk and the complex natural
frequencies sk, the complex natural frequencies of the system sk can be evaluated. Using expres-
sions of complex natural frequencies in Equations 6.19 and 6.20, the natural frequencies and modal
damping ωk and ζk are computed.
The mode shapes can be computed by evaluating the modal amplitudes Aijk. Since we know now
the values of zk, we can rewrite Equation 6.67 in matrix form as
[ Z ] { A} = {h} (6.72)
The components of the square matrix [Z] are the computed values of zk. The right-hand side vector
{h} contains the first measured 2n − 1 samples. The unknown vector {A} contains the unknown
modal amplitudes Aijk. Upon solving Equation 6.72 to find {A}, the MI of the system is complete. For
a comprehensive description of the LSCE method, see He and Fu (2001).
6.3.8 Random Decrement

6.3.8.1 General
Reliable bridge-condition evaluation and early detection of bridge-component failures are critical
for bridge owners in the United States for better utilization of available resources. Remote bridge
monitoring systems (RBMS) have been perceived to assist periodic evaluation of structures to sup-
plement bridge-management systems with quantitative data, and for examining new design tech-
niques. Many RBMS thus far are based on measured bridge vibration. One of the major issues in
developing RBMS is the lack of reliable methods to obtain modal parameters using traffic excita-
tion. This section summarizes a signal processing method that was offered by Alampalli and Cioara
(2000). The method relies on data measured from one of the RBMS installed in New York State
bridges.
6.3.8.2 Overview of Problem

Bridges are relatively complex structures and their structural integrity must be verified through peri-
odic inspection. Most bridges require inspection at least once in every 2 years, by federal mandate,
but serious failures can occur between the inspections, FHWA (2004). A visual inspection, often,
may not reveal hidden defects, and the quality of the inspection is highly dependent on training, per-
sonnel experience, available equipment, and in-service environment, see Moore et al. (2001), Phares
et al. (2001), and Phares et al. (2004). Thus other nondestructive methods were developed and used
by the inspectors to supplement, as needed, the visual inspection of bridges. See Washer (2002),
Tolbert et al. (2004), Siswobusono et al. (2004), Aref and Alampalli (2001), Moore et al. (2004) and
Fuchs et al. (2004). Continuous health monitoring of bridge structures has also been a major topic
of research due to the aging bridge infrastructure and lack of economic resources to completely
renovate all of them. Thus, for maximizing and efficiently using available economic resources, reli-
able bridge-condition evaluation and early detection of possible failures has been a major subject in
transportation research arena.
Remote bridge monitoring systems, using experimental modal investigation where bridge integ-
rity is correlated with structural modification, have received considerable attention in the last decade
due to their perceived ability to assist in bridge evaluation and bridge management, see Chong et al.
(2001), Alampalli and Ettouney (2003), and Cheung and Naumoski (2002). Recent developments in
sensor technology have yielded durable and reliable sensors making use of such systems possible.
New York realizing the long-term benefits of RBMS, installed monitoring systems on three bridges
on an experimental basis to investigate reliability and performance of these systems for practical
application and also to create its own database. New York State systems were equipped with a vari-
ety of sensors. Data under known loading and normal traffic loading were collected over a period
of 3 years. Reliability and performance of the monitoring systems, sensors, and communication
systems were observed.
Most available or developed system identification methods based on measured vibration response
require input excitation and the system output, Katkhuda et al. (2005). In practice, for bridge struc-
tures, this could be very challenging. If forced excitation, such as an impulse hammer or an electro-
dynamic shaker, is used, the bridge has to be closed fully or partially to vehicular traffic. This poses
public inconvenience, is expensive, and in some cases could be a safety hazard. It is more conve-
nient to use the readily available traffic excitation. But, in these cases, input excitation cannot be
measured with any certainty. This requires using output only structural response and requires pro-
cessing methods that can deal with measured structural vibration due to several sources of unquan-
tifiable excitation. Thus, one of the major issues in developing RBMS has been lack of reliable
methods to obtain modal parameters using traffic excitation. In recent days, there has been several
studies investigating various signal processing methods to obtain modal parameters from measured
ambient vibration data from bridge structures (Katkhuda et al. 2005, Cole 1973, and Ibrahim 1977).
This section addresses one such signal processing method and discusses the results obtained using
the data measured from one of the RBMS installed on New York State bridges.
6.3.8.3 Theoretical Background

The dynamic model of a bridge can be simplified using a differential equation system,
[ M ]{x(t )} + [C ]{ x (t )} + [ K ]{ x(t )} = {F (t )} (6.73)
where x is the bridge motion at a given time t, the column vector F represents the general excita-
tion, and the matrices [M], [C], and [K], respectively represent inertia, damping, and stiffness. For
a general excitation expressed by the column vector {F(t)}, the motion xs(t) of a given point Ps of the
discrete structure, including both deterministic and random components, can be written in a modal
decomposition form,
n n
t
xs (t ) = ∑ e − st t ⎡⎣Cs , Rr cos( pr t ) + Cs , Ir sin( pr t )⎤⎦ + ∑ ∫ Fs ,r (z)hr (t − z)dz (6.74)
0
r =1 r =1

where pr and σr are circular damped natural frequency and decay rate of rth natural mode (r = 1,2,3 . . . n).
Cs, Rr and Cs, Ir are n pairs of constants, which can be determined using initial conditions at t = 0;
n n
(
xs (0) = ∑ Cs ,Rr : vs (0) = x (0) = ∑ pr Cs ,Ir − s r Cs ,Rr ) (6.75)
r =1 r =1
and eigensystem,

([ M ] l + [C ] l + [ K ]) { X } = {0}
2
r r r (r = 1, 2, 3, ... n)

(6.76)
where the eigenvalues are of the form,
lr = −s r + jpr ( j = −1 (r = 1, 2, 3 ... n) (6.77)

and the complex eigenvector is given by,

{ Xr } = { X Rr } + j { X Ir } (r = 1, 2, 3 ... n),

(6.78)
The column vectors {X Rr} and {X Ir} are related to constants Cs,Rr and Cs,Ir by the following
expressions:
1
{ X Rr } = − 2 {C1, Rr , C2, Rr , C3, Rr .......Cn, Rr }
T
(6.79)
1
{ }
T
{ X Ir } = 2 C1,Ir , C2,Ir , C3,Ir .......Cn,Ir

With r = 1, 2, 3, … n.
The above formulation yields only 2 × n independent initial conditions, which can be determined
using known initial displacement xs(t = 0) and velocity vs (0) = x s (t = 0). The first sum in Equation
6.74 is influenced by exponential terms e − sr t decaying in time, and the motion x(t) being modulated
by active force components Fs,r(t) through the integral convolution with unit impulse functions,
1 − sr t
hr (t ) = e sin( pr t ) (6.80)
pr mr
where mr is the rth modal mass, dependent on time evolution of excited forces Fs,r(t).
Application of structural modal parameter estimation, using structural response to a noncon-
trolled excitation such as vehicular traffic on the bridge, requires elimination of unknown excitation
forces’ influence in the measured signal as well as separating the free decay component.
6.3.8.3.1 Selective Random Decrement (SRD) Technique

In the RD technique (Cole 1973; Ibrahim 1977; Asmussen et al. 1999) it is assumed that every
transient response time record contains both free decay and forced components. A typical time
history data from a structure, which includes both components, is shown in Figure 6.12a. SRD
method is used to eliminate the forced component (i.e., the influence of unknown excitation forces)
to obtain the free decay component shown in Figure 6.12b. First, on the time history of the signal
(a) ti tk
xs(t)
xrd
t
(b)
ys(τ)
xrd
FIGURE 6.12 Illustration of RD processing (a) original signal and (b) damped signal. (From Alampalli, S.
and Cioara, T. (2000).)
xs(t) (see Figure 6.12a) a line of constant magnitude xrd is selected, such that it intersects the signal
at N points at discrete times tk. A new time function is defined,
1 N
y(t) = ∑ x (t + t) (6.81)
N k =1 s k
with a time shift τ. Using the modal decomposition, Equation 6.74, and knowing that at every tk
moment the initial shift time is τ = 0, the following can be obtained:
1 N n N n
t
ys (t) = ∑ ∑e − sr t
[Cs(,kRr cos( pr t) + Cs(,kIr sin( pr t)] + ∑ ∑∫ Fs(,kr (z).hr (t − z)dz} (6.82)
N k =1 r =1 k =1 r =1
0

Writing the rth modal component of excitation force at point Ps of the structure in the form,

T
Fs(,rk (t ) = X s ,Rr { X Rs } {F (k
}
(t )
(6.83)
and changing the order between summation and integral operations in Equation 6.82 yields,
n
⎡1 ⎛ N ⎞ 1⎛ N ⎞ ⎤
ys (t) = ∑ e − sr t ⎢ ⎜ ∑ Cs(,kRr ⎟ cos( pr t) + ⎜ ∑ Cs(,kRr ⎟ sin( pr t)⎥
r =1 ⎣ N ⎝ k =1 ⎠ N ⎝ k =1 ⎠ ⎦ (6.84)
n
⎡ T ⎛1 N
t ⎞ ⎤
+ ∑ ⎢ X s , Rr { X Rs } ∫0 ⎜⎝ N ∑ {F ( k (z)}⎟ hr (t − z)dz⎥
r =1 ⎣ k =1 ⎠ ⎦
In the above when

1 N
{ }
∑ F ( k (t) ⇒ {0} ,
N k =1
(6.85)

the function ys(τ) represents a free decay of n modal components vibration signal. (Note that
Figure 6.12b is the time history of a single mode free decay component signal.) Taking into account
the first initial conditions of Equation 6.75, the following can be obtained for τ = 0,
n
⎛1 N ⎞ 1 N ⎛ n ⎞ 1 N
ys (t = 0) = ∑ ⎜ ∑ Cs(,kRr ⎟ = ∑ ⎜ ∑ Cs(,kRr ⎟ = ∑ xs( k (ti ) (6.86)
r =1 ⎝ N k =1 ⎠ N k =1 ⎝ r =1 ⎠ N k −1
If all xs(k (ti) samples are triggered at the same level xrd at every time ti, then
ys (t = 0) = xrd
(6.87)
The triggering can be also performed at level xrd = 0 for a positive or negative slope (i.e., xs(ti +
∆t) > 0 or xs(ti + ∆t) < 0, where ∆t is the sampling time). Mathematically, Equation 6.85 is satisfied
for a Gaussian white noise excitation that requires large number of points (N → ∞). In many cases,
when vibration responses are obtained from structures such as bridges excited by vehicular traffic,
this condition is hard to fulfill (see Figure 6.13).
For a typical multi-span bridge shown in Figure 6.13, with a deck Dk on the metallic girders Gr,
more than one moving vehicle will be on the bridge inducing the transient bridge deformations.
The deck and girders are supported on piers through a number of bearings Be with foundations Fd
Gr mg
Dk
v
Soil
Be Pr
Sp
Be
Soil
Fd
Sa
XS(t)
FIGURE 6.13 Typical bridge structure carrying vehicular traffic.
transferring the loads to the soil. The deck is also connected to the highway to ensure the smooth
traffic flow. Typical damage in the bridge structure, such as a crack on the midspan of one girder,
changes the local area stiffness and modifies the parameters contributing to the static and dynamic
equilibrium of the structure. Hence, predicting the change in the measured structural parameters
such as natural frequencies and localizing the area contributing to the change are essential for dam-
age detection. For the first step, in the case of a bridge structure, the ambient response triggered
by the vehicles passing along the bridge is needed to predict structural modification caused by a
localized damage. This response does not depend on the dynamical characteristics of the moving
loads, but is excited by these loads. SRD algorithm developed above can be used easily to extract
the response to ambient excitation.
A typical time history recorded using an accelerometer Sa on the bridge, six events occur (marked
1 to 6 in Figure 6.14). In the first period of every excitation, unknown forces from moving loads
increase the vibration magnitude, which is dominated by a second component of the general form
of motion xs in Equation 6.74.
At the end of each period, excitation forces drop in magnitude and the dominant first component
of Equation 6.74 is more appropriate to a multi-modal free decay vibration motion. During each
such event, a narrow time window Tw can be selected such that triggering start of the RD func-
tion (signal) ys(τ) cis of length Ts. By positioning the window Tw in the area of the envelope signal
decaying slope, it can be made certain that the level of unknown excited forces drop in magnitude
and the sum in Equation 6.85 converges quickly to zero, for a limited number (N) of averages. For
this particular case, the RD processing can be redefined, more appropriately, as the SRD. The RD
function processed (Figure 6.15) using the data shown in Figure 6.14 clearly shows the free decay
components.
When events occur at close intervals (Figure 6.16), length Ts of the SRD function ys (τ)
overcomes the span between events 2–4 and 5–6. End of the function ys (τ) will be affected
here by deviation from free decay law in processing, that is, the forced component. To over-
come this problem, it is necessary to reduce the time trace length Ts of ys (τ) to less than the
span between the two closed events, and apply an exponential time window for each sequence
processed,
− s0 ( t − t j )
w j (t ) = e
(6.88)
Triggering windows
Xrd
Triggering level
0
Ts Ts Ts
1 2 3 4 5 6
0 5 10 15 20 25 30 t [s]
FIGURE 6.14 Typical vibration signal of a bridge due to vehicular traffic. (The markers 1 through 6 indicate
separate marked events as measured by an accelerometer. Each event marks increase in live load, such as pas-
sage of a truck.)
10
ys(τ)
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

τ [s]
FIGURE 6.15 RD function obtained from the signal shown in Figure 6.14.
Tw
Ts
Xrd Ts
1 2 3 4 5 6
0 5 10 15 20 25 30
t [s]
FIGURE 6.16 Typical vibration time history with closely spaced multiple events. (The markers 1 through
6 indicate separate marked events as measured by an accelerometer. Each event marks increase in live load,
such as passage of a truck.)
at the start of triggering time tj, where σ0 is virtual damping decay rate. In addition, all the modal
decay rates will be estimated from modal parameter estimation using a new modal free decay
signal,
z s (t ) = x s (t )w j (t ) (6.89)
The decay rate σ0 is calculated such that the end of the signal, zs(t = Tos)⇒ 0, where Tos is the new
time length of the signal zs(t). Finally this will yield the following SRD function, ready for modal
parameter estimation.
n
⎡1 ⎛ N ⎞ ⎤ n ⎡1 ⎛ N ⎞ ⎤
Rs (t) = ∑ e − ( s0 + sr ) t ⎢ ⎜ ∑ Cs(,kRr ⎟ cos( pr t)⎥ + ∑ e − ( s0 + sr ) t ⎢ ⎜ ∑ Cs(,kIr ⎟ sin( pr t)⎥ (6.90)
r =1 ⎣ N ⎝ k =1 ⎠ ⎦ r =1 ⎣ N ⎝ k =1 ⎠ ⎦
6.3.8.3.2 Modal Parameter Estimation

The processed free decay signal can be used now for highly accurate estimation of modal param-
eters using a circle-fit algorithm, see Cioara (1988), and Cioara (1999). First, Fourier transform of
the form shown below, is applied to the free decay,
n
⎛ Ur + jVr Ur − jVr ⎞
F (w ) = ∑ ⎜ + ⎟ (6.91)
s
r =1 ⎝ r + j ( w − pr ) s r + j (w − pr ) ⎠
where ω is the circular frequency with harmonic steady state excitation. Note that the Nyquist plot
in the vicinity of rth damped natural frequency pr is approximated by circle Cr (see Figure 6.17).
Experimentally determining the coordinates of the representative points P(ωi), P(ωk), P(ωs), on the
curve Cr and determining the angles βik between the vector radius OrP(ωi) and OrP(ωk), one can
define,
1
Tik =
tan(bik ) (6.92)
to build a system of linear equations,

pr (w k − w i ) + s r [(w s − w k )Tks − (w s − w i )Tis ] = w s (w k − w i )
pr (w s − w k ) + s r [(w s − w i )Tis − (w k − w i )Tis ] = w i (w s − w k ) (6.93)
pr (w s − w i ) + s r [(w s − w k )Tks − (w k − w i )Tik ] = w k (w s − w i )
where the damped natural frequency pr and decay rate of the rth natural mode are unknowns.
Usually the modal test implies the constant increment of the frequency ∆ω and three consecutive
points P(ωi), P(ωi + 1), and P(ωi + 2) with,
w i = (i − 1)Δw (6.94)
Using Equations 6.94 and 6.93), one can obtain
2 pr + (Ti + 2,i +1 − Ti +1,i )s r = 2i Δw (i = ....k − 1, k, k + 1.......) (6.95)
Yk
Im P(ω)
P(pr)
P(ωi) βik
Xk
P(ωk)
β rk
Cr Or
βks
P(ωs)
ω
Re
FIGURE 6.17 Illustration of circle fit.

This can be solved in a least squares manner, since it has more than three equations and only two
unknowns: pr and σr. The algorithm developed by Cioara (1988) and (1999) allows us to determine
mode shapes in terms of modal complex components Ur and Vr .
An illustrative example, to show the high accuracy of this method to estimate modal parameters
estimation, is presented in Figures 16.8 and 6.19. These figures contain FFT spectrum and circle-
fit algorithm of the signal shown in Figure 6.15. The SRD signal was obtained by processing the
ambient vibration data obtained from a bridge (see Figure 6.14). This exhibits two damped natural
frequencies f1 = 9.267 Hz and f 2 = 14.361 Hz. These are indicated as P1 and P2 in the FFT spectrum
(Figure 6.18) and the Nyquist plot (Figure 6.19). Application of the same procedure to two other
signals (recorded simultaneously with the signal shown in Figure 6.14) showed very small devia-
tions (<0.01%) relative to frequency values f1 and f 2, even though the modal components have very
different magnitude levels between the three signals.
6.3.8.4 Infrastructure Application

New York State installed structural monitoring systems (STRUMS) on three bridges (Alampalli
1998b; Alampalli and Fu 1994). These systems include sensors for measuring vibration (acceler-
ometers), stress (strain gages), and slope (clinometers). Each vibration data file contains 16,384 data
points collected using an 8-bit A/D converter. These systems were named STRUMS20, STRUMS30,
and STRUMS31 and data were acquired during a period of 2 years.
The data presented in this chapter is from the bridge located in Rochester, NY, where STRUMS
30 was installed, Alampalli and Fu (1994). It was built in 1963 and carries I490 East Bound over
Conrail Mainline in Monroe County. It is a 3-span structure with seven steel girders carrying con-
crete deck. The structure is about 76-m-long with 44 degree skew. The end spans were supported by
steel rocker bearings and were continuous over the reinforced concrete piers; the middle span was
F(f) P2
P1
0 10 20 30 40 50 60
f [Hz]
FIGURE 6.18 FFT of signal shown in Figure 6.15.
p1 f = 9.267 Hz η1 = 1.105% U1 = 1.389 V1 = 48.33

p f = 14.361 Hz η2 = 1.532% U2 = –30.80 V2 = 6.493
2
+Im
P2
+Re
P1
FIGURE 6.19 Nyquist plot of spectrum shown in Figure 6.18.

connected to the end spans by pin-and-hanger. The bridge carries three lanes of traffic. This bridge
was rehabilitated in 1993 to replace corroded diaphragms, repair concrete deck with a concrete
overlay, install a new joint system, replace steel rocker bearings with elastomeric bearings, and pro-
vide continuity between the spans by splicing with steel plates. This bridge was instrumented with
STRUMS to examine the consistency of measured structural indices to the in-service environment
and changes in the structural conditions.
Structural changes and environmental conditions will influence natural frequencies of bridges
over long periods of time. To study these effects, data acquired during March 93 and April 95 from
several accelerometers (S2-S15) of STRUMS30 were analyzed using the method illustrated above
and results are presented in Figure 6.20 and Table 6.5. The bridge was instrumented with 11 accel-
erometers, and the data was collected simultaneously from all the accelerometers. The subset of the
data, during March 93 and April 95, showing the variation of the first natural frequency is shown in
Table 6.5. The plotted vertical bar in Figure 6.20 represents the average values recorded by various
accelerometers (S2 to S15) simultaneously (average of columns shown in Table 6.5 during a single
event).
To estimate the influence of moving loads (such as a truck) passing on the bridge during an event,
several events were recorded on the same day and are designated with an extension (2t or 3t) for
second and third events. Data reliability is expressed using 95% confidence intervals, plotted using
shadow bars at the top of each frequency bar (Figure 6.20) if more than three corresponding values
exists.
The analyzed data records contain 16,384 points collected using an 8-bit digital conversion
(±128 units). The levels of signals are located in the lower range of ±15 units, and created major dif-
ficulty in data processing using conventional methods. Therefore using dedicated methods such as
those developed in this chapter will yield higher accuracy results. Similar results were obtained for
other higher order modal frequencies of STRUMS30, STRUMS 20, and STRUMS 31.
6.3.8.5 Concluding Remarks

Analysis of data, for natural frequencies, collected during 3 months from STRUMS20 station (15 data
records), 24 months from STRUMS30 station (17 data records), and 20 months from STRUMS31 (12 data
records) showed small standard deviations, indicating the reliability of measured data. The first impor-
tant conclusion from the present study is that specific methods developed for the bridge-structure modal
parameter estimation, using vibration signals monitored from ambient excitation, work very well.
Standard deviation
[Hz]
3.8
95%+σ95%
3.85
3.7
Model frequency
–σ
3.65
3.6
3.55 fm = 3.515 Hz
3.5
3.45
3.4
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
03/11/93
08/12/93
10/15/93
10/29/93
11/30/93
12/07/93
06/05/94
06/07/94
06/12/94
06/13/94
06/27/94
06/28/94
07/06/94
07/13/94
08/27/94
09/13/94
04/22/95
Strums 30 data acquired
FIGURE 6.20 Variation of first mode during 1993–1995.

TABLE 6.5
Variation of First Modal Frequency of the Bridge Structure during 1993–1995
Modal Frequency Values [Hz]
Sensor
No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
S2 3.472 3.629 3.493 3.497 3.489 3.519 3.499
S2-2t 3.480 3.570 3.534
S3 3.749 3.516 3.491
S3-2t 3.477
S4 3.529 3.571 3.473
S4-t 3.451
S5 3.750 3.472 3.478 3.548 3.510 3.484 3.544
S5-2t 3.473 3.482
S6 3.749 3.472 3.465 3.568 3.555 3.488 3.434 3.495 3.504 3.548
S6-2t 3.472 3.484
S7 3.748 3.464 3.485 3.569 3.493 3.502 3.496 3.513 3.516 3.555
S7-2t 3.471 3.482 3.570
S8 3.749 3.472 3.464 3.502 3.567 3.552 3.493 3.510 3.485 3.487 3.513 3.516
S8-2t 3.468 3.485 3.565
S9 3.465 3.567 3.500 3.492 3.483 3.469
S9-2t 3.472 3.468
S10 3.748 3.456 3.547 3.473 3.462 3.482 3.482 3.522
S10-2t 3.434 3.487 3.551
S14
S15 3.475 3.492 3.498
Sensor no. Extension 2t or 3t indicate second or third events of the same data record.
The natural frequencies do not exhibit negative slope with time indicating that they do not corre-
spond to possible local stiffness reduction due to continuous structural degradation. (Note that the
change in frequency between March 1993 and August 1993 was attributed to bridge rehabilitation
that includes replacement of steel rocker bearings with elastomeric bearings.) Thus, these variations
may be attributed to environmental parameters such as temperature and humidity. These environ-
mental conditions modify the mechanical characteristics near bridge boundary conditions such as
foundations (Fd in Figure 6.13). Note that the soil dynamic characteristics and dynamic character-
istics of bridge bearings are also dependent on environmental conditions. The same conclusions
were reached by Alampalli (1998a), Alampalli et al. (1998), and Farrar and Doebling (1998), after
significant experimental efforts.
Comparison of frequencies listed in the columns of Table 6.5, obtained from different events
from the same data record show no notable deviations. This indicates that moving loads have small
influence on dynamic behavior of the bridge structure. Predominant influence of moving load char-
acteristics should lead to noticeable deviations between modal frequencies estimated from consecu-
tive events.
This study also indicates that practical methods for bridge fault diagnostics using the vibration
signals monitored on the bridge structures are necessary for further supplementary research work.
Relative influence of various environmental parameters on dynamic response of bridge structure
should be determined. It is also necessary to find the natural modes, which are most insensitive to
environmental parameter variations, for structural monitoring and damage detection. This can be
determined using dynamic model simulation by modifying the parameters of bearing and boundary
conditions in the ranges of interest.
6.4 PARAMETER IDENTIFICATION (PI)

6.4.1 Problem Definition
The model, or parameter, update problem in structural identification starts with an analytical or
numerical model of the structure. For example, a finite element model of the structure. As such,
estimates of the stiffness [K], mass [M], and damping [C] matrices are assumed to be known. It is
desired to improve the estimates, update, of one, or more, of those matrices. This estimate improv-
ing, or update, is performed using an available set of test results. In our case, it would be results of an
SHM project. Numerous methods for model updating have been developed and presented. For sim-
plicity, those methods are subdivided into three generic categories. These categories are (1) the opti-
mal matrix update, (2) sensitivity-based parameter update, (3) eigenstructure assignment. Overview
of those categories were described by Friswell and Mottershead (1995) and Doebling et al. (1996).
We should observe here that vast majority of the model update methods and categories utilize modal
information that result from MI methods of the previous section. Note that when a method is based
on an assumed composition of the stiffness, mass, and damping matrices, any errors or mistakes in
the initial assumption will result in a major errors that will render the whole procedure inaccurate.
Thus, we summarize the problem of parameter identification as follows:
1. Establish, based on theoretical and personal experience, the system matrices: [K]0, [M]0
and [C]0. The details of the numerical model should be consistent with the goals and objec-
tives of the effort (design assessment, local of global behavior, etc.). If damage identifica-
tion is the goal of the effort, an accurate baseline model should be established first.
2. Select desired parameters to be updated. Note that the analytical model should be consis-
tent in its details with those parameters.
3. Perform testing of the system as appropriate. Collect appropriate data from the test.
4. Apply an appropriate PI method to update the parameters chosen in #2. The updated matri-
ces are matrices: [K]1, [M]1 and [C]1 can now be used to achieve the original objectives of
the effort.
Note that the above steps constitute only a STRID effort. The resulting matrices define the struc-
tural state as tested. No accurate damage identification can be achieved using these steps. The dif-
ferences between the original matrices, with subscript of 0, and the updated matrices, with subscript
of 1, only define the differences between the original estimate of the matrices (based on theory and
personal experiences) and the model that is based on test results. The effects of the damage in the
tested system only reflects in the updated matrices. No information is available in this process that
can help in identifying the already existing damage state in the system. To identify the existing
damages in the system, two approaches are possible: (1) Perform testing of a pristine non-damaged
system, or (2) Perform direct DMID efforts (Chapter 7).
6.4.2 Eigenstructure Assignment Update

The method of Eigenstructure assignment also starts by assuming that the pristine (not damaged)
structural model is known and is correct. The objective of the method is to identify the damage in
that pristine structure. To achieve this objective, a set of measured mode shapes and modal frequen-
cies are obtained, usually through a modal testing and identification procedure. The Eigenstructure
assignment method then adds a set of artificial forces in the pristine structural equilibrium equa-
tion. Through a set of algebraic reassignments least square optimization, the damaged elements (or
DOF) of the structure can then be estimated. There is large body of knowledge and research of the
method. One of the earliest works was by Andry (1983). Some of the more relevant applications of
the method to SHM were by Lim (1995) and Lim and Kashangaki (1994). We summarize the basic
equations of their approach as follows.
Assume that the matrix equation of a given structure models accurately the structure in its pris-
tine condition. We assume that all the system matrices, [M], [K], and [C] are known. Let us assume
that the physical structure is now damaged somehow, and it is desired to obtain updated estimates
of the damaged structure through identifying the location and the magnitudes of the damage. We
also assume that the mode shapes, {𝜙}k and the modal frequencies, λk of the damaged structure are
identified, perhaps through a MI process. For simplicity, we assume that the damage of interest is
manifested mostly in the stiffness matrix, [K] and no damage is present in the damping or the mass
matrices. By assigning the known modal parameters to the known structural matrices through the
structural free vibration, we get
([ M ] l + [C ] l + [ K ]){f} = {R}
2
k k k (6.96)
The vector {R} is a residual vector that results from the changes in [K] due to damage. As such {R}
represents additional forces in the structure that occurred due to the unknown structural damage.
The eigenstructure assignment method utilizes {R} to estimate the locations and magnitude of dam-
ages by adjusting the free vibration equation to accommodate the damage as

[ M ]{U} + [C ]{U } + [ K ]{U } = [Q ] {F } (6.97)
The vector {F} contains artificial control forces, while the matrix [Q] is an influence matrix. The
displacements and control forces can be related by
{F } = − [G ][Q ]T {U } (6.98)
Thus
[ M ]{U} + [C ]{U } + ([ K ] + [Q] [G ] [Q]T ){U} = {0} (6.99)

Note that the change in the stiffness matrix due to the structural damage is expressed by [Q][G][Q]T.
The modal parameters can now be applied as
([ M ]l2k + [C ]lk + ([ K ] + [Q] [G ] [Q]T )){f}k = {0} (6.100)

Rearrange Equation 6.100 to express the damage effects on the right side
([ M ]l2k + [C ]lk + [ K ]){f}k = −([Q] [G ] [Q]T ){f}k (6.101)

Following Lim (1995) assuming that the potential of damage only in the jth element, the influence
matrix [Q] is evaluated as [Q] j. Note that [Q] j is completely defined by the location of the jth element
within the overall matrix composition of Equation 6.101. In addition, the unknown gain matrix is
reduced to an unknown scalar g j. Equation 6.101 is reduced to

([ M ]l 2
k ) ( )
+ [C ]lk + [ K ] {f}k = − g j [Q] j [Q]Tj {f}k (6.102)
Finally,
{f}k = [ L ]kj {J}kj (6.103)

with
[ L ]kj = ([ M ]l2k + [C ]lk + [ K ])−1 ([Q] j [Q]Tj ) (6.104)

and
([ M ]l2k + [C ]lk + [ K ]){f}k = − g j ([Q] j [Q]Tj ) {f}k
(6.105)
{J}kj = − g j {f}k (6.106)

Since there are potential errors in forming the structural matrices, as well as measurement errors,
a new vector, called best achievable vector {𝜙}bkj can be estimated, using a least square approach.
Defining [L]+kj as the pseudo inverse of [L]kj (see previous section), the vector {ϑ}kj is eliminated and
the best achievable vector is related to the measured mode shape as
{f}bkj = [ L ]kj [ L ]+kj {f}k

(6.107)
The relationship between the measured and the best achievable mode shapes as expressed in
Equation 6.107 would be an indication for the damage presence in the jth element. If the damage is
located in the jth element then the {𝜙}bkj and {𝜙}k will be identical; if the damage is not located in the
jth element then {𝜙}bkj and {𝜙}k will be different. A measure for the presence of damage is the angle
αkj between the two vectors
⎛ {f}T {f} ⎞
a kj = cos −1 ⎜
bkj k
⎟
⎜⎝ fbkj fk E ⎟⎠ (6.108)
E
With ||…||E is the Euclidean norm of a vector. The angle αkj measure the presence of damage in the jth
element in the measured kth mode. For more details on the physical interpretations of αkj, the reader
can refer to Lim (1995) and Lim and Kashangaki (1994).
The above approach can be extended to estimate the level of damage in the jth element. The
details of the analytical approach can be found in Andry et al. (1983), Lim (1995), and Lim and
Kashangaki (1994).
6.4.2.1 Summary of Input/Output Measurements/Testing

To utilize the eigenstructure assignment method in an SHM experiment the following steps can be
followed:
1. A modal testing setup should be performed first. As such, adequate number of motion sen-
sors need to be installed on the required structure.
2. Adequate number of modal information should be realized next using any of the MI meth-
ods that were described earlier.
3. Adequate structural matrices (stiffness, damping, if needed, and mass) are composed.
Special care should be given while constructing those structural matrices: their resolution
should be large enough to meet the objectives of the SHM experiment. The finite element
modeling technique is usually utilized in this step, although other numerical techniques
can be used (such as finite difference).
4. Finally one of the many eigenstructure methods can be applied to the results of #2 and
#3, above. The method that was summarized, above that was developed by Lim (1995) or
any other adequate method can be used. For a comprehensive review of available eigen-
structure assignment methods, the reader is referred to Friswell and Mottershead (1995) or
Doebling et al. (1996)
6.4.2.2 Advantages/Disadvantages and Applicability to SHM

Among the advantages of the eigenstructure assignment methods are that they can update all struc-
tural matrices, including damping matrix. Also, by definition, these methods will satisfy exactly
the required measured mode. The disadvantages of the methods are the intensive computational
demands as well as the lack of possible physical interpretation of the updated results, since the
updates are applied to the whole matrices. In simplified systems, such as simple beams, simple
planer pinned trusses, or lumped mass systems, the physical interpretations of the corrected DOF
might be possible to interpret physically. Unfortunately, in a realistic structure, such as complex post-
tensioned reinforce concrete bridge, physical interpretation of damaged tendons using a 3D finite
element model of post-tensioned steel and concrete might be difficult, if not impossible to do.
6.4.3 Optimal Matrix Update (OMU)

6.4.3.1 General Description
Optimal matrix update method is based on updating an assembled system matrices using test results.
As most of the methods in this section, this class of methods also depends on the accurate analytical
modeling of the system on hand. The method starts by assuming a baseline analytical model of the
structure K. On the basis of some test observations, the model information is updated using an opti-
mization scheme. For example, a minimization of the rank of the matrices was used by Zimmerman
and Kaouk (1994) to update K. Doebling (1996) introduced a new technique called minimum rank
elemental updates (MREU) that computes and updates the minimum ranks of the element stiff-
ness parameters. Hemez and Doebling (2001) provided an in-depth review of update methods for
nonlinear dynamic systems. Yang and Liu (2007) compared some matrix update methods using
a simulate truss system. They observed the importance of measurement noise on the accuracy of
matrices updating.

Since the OMU methods rely mostly on analytical/numerical modeling solutions, they are not
demanding computationally. They do not offer, however, in-depth physical insight to the correc-
tions they offer. This is mainly due to the fact that they operate on full structural matrices. This
is a rather limiting disadvantage, especially in case of a complex SHM situation (such as an SHM
project dealing with FRP bridge deck, with all the geometric and material complexities it presents).
What makes many of these methods even less attractive to SHM is the tendency for many of these
methods to spread the updates throughout the entire matrices under considerations as observed by
Hemez (1993). This would result in an inaccurate error, or damage, localization. Such inaccuracies
would reduce the benefits of any SHM project.
6.4.4 Sensitivity-Based Update

6.4.4.1 Sanayei Static-Based Methods
Most structural identification methods are based on dynamic motions of the structures. On one
hand, measuring a stream of time histories provides an abundance of information that would
help in identifying the structure. Of course, identifying dynamic properties of the structure
includes identifying structural stiffness, damping, and inertial properties. Static-based identifi-
cation methods would be far simpler, since they involve only the stiffness of the structure. They
also involve simpler static measurements. As such they would be suitable for parameter identifi-
cation of stiffness matrices, design verification, and assessment for static-based loading. We will
detail the basics of a static-based parameter identification method that was developed by Sanayei
and Saletnik (1996a, 1996b), and Sanayei et al. (1997). The method is elegantly simple, yet fairly
powerful and easy to implement. Many of the basic features of structural identification methods
(both dynamic-based and static-based) are included in this method; hence it is suited for our
immediate task of exploring general characteristics of structural identification methods.
Let us assume that for a given linear elastic structural system the finite element stiffness
matrix [K(p)] is well developed. The order of this matrix is N, which represent the total number
of constrained DOF of the structure on hand. The functional relationship of some parameters
of the structure p are assumed to be known. These parameters include any structural param-
eters that are used to form [K(p)] such as cross-sectional area, different moment of inertia, or
different types of springs. We would like to point out that parameter identification is perhaps
most valuable in identifying spring values in finite element modeling. In practice, engineers
tend to approximate complex components of the structural system by springs. For example,
large swathes of not well-defined soils are approximated by soil springs. Also, steel or concrete
connections are modeled by equivalent springs. Using a method such as the current method to
accurately identify those otherwise not well-defined springs can prove to be a valuable effort.
Some parameters that can yield a highly accurate finite element model when identified are
shown in a typical bridge and abutment example of Figure 6.21. We should note that there is
no theoretical limit to the number of parameters p that need to be identified. In practice, only
limited number of parameters is considered.
The equilibrium equation of the structure can be written as
[ K ( p)] {U} = {F} (6.109)
where {U} and {F} are the displacement and force vectors, respectively. Both vectors are of the
order N. One of the main features of the method is that it defines an elemental strain εi for the ith
finite element. This definition relates the strains in the finite element model directly to measured
strains. Elemental strains can be related easily to the element displacements {U}i by using the strain
displacement relation
ei = {B}Ti {U}i (6.110)
The vector {B}i can be developed easily for any type of finite element.
Bridge and abutments
Soil Soil
Foundations
Physical system
Structural connections
Finite element nodes

Lateral soil stiffnesses
Vertical soil spring
Rotational soil spring
Finite element analytical model
Identifying highly uncertain structural parameters such

as soil springs and/or structural connections can result
in a highly accurate structural model
FIGURE 6.21 Practical example of parameter identification.

Assembling all elemental strains in the system we get
{e} = [ B] {U} (6.111)
where [B] is assembled from the individual {B}i. We eliminate {U} from Equations 6.109 and 6.111,
to get
{e} = [ B] [ K ( p)]−1 {F} (6.112)

Since only those measured strains at locations of measurements are of interest to us, Equation 6.112
above is subdivided such that
{e}a = [ B]a [ K ( p)]−1 {F} (6.113)
where {ε}a is the strain vector whose components are the analytically computed strains at measure-
ment locations. The matrix [B]a is a subset of [B]. Note that Equation 6.113 represent only a single
force distribution. The method is originally developed for multiple force distributions, and the gen-
eralization is straightforward.
If the force distribution {F} is applied to the structure, and the corresponding strains {ε}m are
measured, then it is possible to define the error function
{e( p)} = {e( p)}a − {e}m (6.114)
Note that the computed strains and the error function are both functions of the parameters p. Using
Equation 6.113
{e( p)} = [ B]a [ K ( p)]−1 {F} − {e}m (6.115)
What remains now is to estimate the parameters p that would minimize the error function {e(p)}.
The error function is first expanded in a Taylor series around a base value of p such as
⎡ ∂{e( p)} ⎤
{e( p) + Δ p} ≅ {e( p)} + ⎢ ⎥ {Δp} (6.116)
⎣ ∂ ( p) ⎦
The sensitivity matrix is defined as
⎡ ∂ {e( p)} ⎤
[ S ( p)] = ⎢ ⎥ (6.117)
⎣ ∂{ p} ⎦
The sensitivity matrix can be evaluated numerically, since the functional relationship between
[K(p)] and p is known, in other words, the matrix ∂/∂pj([K(p)]) can be evaluated. Note that pj is the
jth parameter that needs identification.
The error function can be minimized in a least square sense by minimizing the norm
J ( p) = {e( p + Δ p)}T {e( p + Δ p)} (6.118)
Assuming that the required minima occur when the gradient of J(p) is set to zero,
[ S ( p)]T [ S ( p)] {Δ p} = − [ S ( p)]T {e( p)} (6.119)
The above equation can be solved for the required change in parameters {∆p} from the assumed
initial parameter set {p}. There are several solution techniques that were suggested by the authors
of the method. The pseudo inverse method, the singular valued decomposition (SVD) method, and
iterative methods can all be used to solve for {∆p}. Pseudo inverse method and SVD methods are
described elsewhere in this chapter. The computed {∆p} can be used to update {p}. The method is
then repeated in an iterative way until converged. Criteria of convergence have been recommended
by the authors.
6.4.4.1.1 Modal Extension of the Method

The aforementioned methods were generalized to take advantage of potential measured modal val-
ues, see Sanayei and Santini (1998) and Olson (2002). The modal expression for a free vibrating
system is
⎡ K aa K ab ⎤ ⎧Φ a ⎫ 2 ⎡
M aa M ab ⎤ ⎧Φ a ⎫
⎨ ⎬ = vi ⎢ ⎨ ⎬ (6.120)
⎢K
⎣ ba K bb ⎥⎦ ⎩Φ b ⎭i ⎣ M ba M bb ⎥⎦ ⎩Φ b ⎭i

The ith subscript refers to the mode number. The a subscript refers to the measured entity, while the
b subscript refers to the unmeasured entity. A modal-based error function can be expressed using a
condensation scheme as
{es ( p)}i = [( K aa − v i M aa ) − ( K ab − v i M ab )][( K bb − v i M bb ) − ( K ba − v i M ba )]{Φ a }i

2 2 2 2
(6.121)
Note the similarity between the expressions of error functions of Equations 6.114 and 6.121. Similar
developments to Equations 6.115 through 6.119 can be made until optimal unknown parameters are
obtained. Details of the method can be found in Sanayei and Santini (1998) and Olson (2005). This
method of modal-based parameter estimation was used by Olson (2005) to investigate dynamic
evaluation methods for bridges.

Among the advantages of the sensitivity-based methods are the following: (1) They can be physi-
cally interpreted, since the parameter to be updated are selected a priori; (2) Many of these methods
use localized updating techniques, making it easier to control, understand, and interpret the results;
and (3) If formulated properly, they can be integrated with the commercially available finite element
tools. Two disadvantages of these methods are (1) they can be computationally demanding, and (2)
using Taylor series would require the derivation of the derivative of the parameter, which might not
an easy task. Perhaps the most important limitation of this class of method is that the parameters
must be chosen a priori. This would require knowledge of the problem that might negate an SHM
procedure where the cause of the problem to be monitored might not be known a priori. This prior
knowledge of the parameters can be a severe limitation of the use of sensitivity-based methods in
an SHM project.
6.4.5 Other Parameter Identification Methods

There are many other parameter identification methods. For example, Beardsley et al. (2000)
investigated a parameter updating method: Generalized least square (GLS). The method relies on
computing the displacements vector {U (t)} of a nonlinear F.E. model using the initial nonlinear
structural F.E. model
M ( p){U(t )} + K ( p) {U (t )} + {Fi ( p, t )} = ( F (t )) (6.122)

Note that M(p) and K(p) are the mass and stiffness matrices that were derived from an F.E. mod-
eling technique. Both matrices are assumed to be linear, and functions of different parameters p.
The parameters p are those that need updating during this structural method. Note that all dynamic
nonlinear effects were lumped together in the nonlinear internal force vector {Fi(p,t)}. The external
force vector is {F(t)}.
If during an experiment, the measured structural displacements are {U(t)test}, then the updating
problem can be defined as minimizing the objective function J(p) where
J ( p) = R( p, t ) (6.123)

and
R( p, t ) = {U (t )test } − {U (t )}
(6.124)
The above approach is similar to the method of sensitivity-based updates in the fact that it updates
the parameters p by minimizing the differences between measured and computed displacements.
The method, however, applies the minimization scheme (a least square approach) across both the
2-norm spaces (the parameter space p and the time space t).
Another model updating method called principal component decomposition (PCD) method that
addresses the problem of Equation 6.122 was developed by Hasselman et al. (1998). They applied
the SVD on the computed displacement vector {U (t)} to find the displacements, nonsingular modal
components, thus increasing the accuracy of the results, as in
[U ] [Σ] [V (t )]T = SVD([U (t )]) (6.125)

The objective function to be minimized is
J ( p) = WU RU ( p, t ) + WΣ RΣ ( p, t ) + WV RV ( p, t )
(6.126)
where
[ RU ( p, t )] = [U test ]T [U ] − I (6.127)

[ RV ( p, t )] = [Vtest ]T [V ] − I (6.128)

[ RΣ ( p, t )] = [Σ test ] [Σ] (6.129)

It is interesting to note that SVD basis of the PCD method is similar to the SVD basis of the ERA
method for modal realization. Both approaches use the SVD to realize the nonsingular modes of
the time samples. The basic difference, of course is that the ERA methods apply SVD on measured
time samples, while the PCD methods apply SVD on computed time samples.
Beardsley et al. (2000) compared the results of GLS and PCD methods as applied to an impact
testing. They concluded that the two methods produce comparable results. However, they observed
that the PCD approach has two potential advantages when compared with the GLS approach: (1) It
is more sensitive to parameter changes, thus less likely to experience numerical problems, and (2)
Due to its reliance on an SVD process, it will filter away some of the unwanted measurement noise,
or rigid body behavior.
6.5 ARTIFICIAL NEURAL NETWORKS (ANN)

6.5.1 Introduction
Artificial neural network is a numerical method that relates unknown output parameters to known
input parameters. The relationship is evaluated using complex interrelationships between input
and output parameters. To aid in computing these interrelationships (called weights and biases), a
dataset of input-output measurements are used. This makes ANN fairly suited to SHM projects,
where pairs of input-output datasets are known. ANN is also utilized in many other fields such
as finance and investing, Trippi and Turban (1992). Software specifically designed for model-
ing ANN problems is popular; see, for example, Demuth et al. (2007). One of the distinguishing
features of ANN is that it does not follow symbolic algorithms, same as humans: we will not find
any equations inside the human brain. Symbolic algorithms are higher forms of neural algorithms.
Other features of ANN are that the (1) connectivity (interrelationships) between basic units can be
modified, terminated, or generated during the computing process; (2) the computations are mostly
distributed computations, which are suitable for advance computing machines; (3) ANN can be
performed in a supervised or unsupervised setting (which is suited for real time SHM); and finally,
(4) ANN computations are local, not global, a feature that is also suited for modern computing
algorithms. All of those ANN features have parallels with biological neural networks, as shown
in Figure 6.22.
This section explores many features of ANN that are pertinent to SHM field. We also describe
some of the yet untapped immense potential of ANN in the fields of structural and damage identifi-
cation. Finally we present some examples to illustrate some of those features.
6.5.2 Basic Developments of ANN

6.5.2.1 Single Layer ANN
The simplest ANN contains single input measure and single output measure. They are related by a
single relationship unit. Also, there is a single bias unit that can affect the input-output relationship
such that
y = f (wx + b) (6.130)

where x, y represent the input and output. The weight and bias are w and b. The input unit, bias unit,
and output unit are called neurons, in recognition to the biological analogy of the ANN. The single
neuron ANN is shown in Figure 6.23.
For a given input, which is assumed to be known, and a given output (measurement) data-
set pairs, the values of w and b are computed so as to minimize any errors. Such an error is the
Biological neural networks Artifical neural networks

(BNN) (ANN)
Commonalities
• Nonlinear units
• Local/distributed
processing
• Parallel computing
• Can be pruned, or
grafted
FIGURE 6.22 ANN versus BNN.
Neural network with single neuron

Input layer Output layer
Weight (W)
Bias (B)
FIGURE 6.23 Single-neuron ANN.

difference between the computed (using an assumed values for the weight and bias units) and the
measured output. The weights and biases are assumed functional relationships between the input
and computed output. Those functional relationships are called activation functions (sometimes
referred to as transfer functions). Some examples of activations functions are shown in Table 6.6
and Figure 6.24.
Note that sigmoid and hyperbolic activation functions have similar forms. The range and the
average of the two functions vary. The average of the sigmoid function is 0.5, while the limiting
values are 0.0 and 1.0, respectively. The average value of the hyperbolic tangent is 0.0, and the two
limiting values are –1.0 and 1.0, respectively. This makes the sigmoid activation function more
suited for problems relating to average type behavior. The hyperbolic tangent functions are more
suited to problems where the digression from the average is required.
TABLE 6.6
Activation Functions
Activation Function Expression Comment
Identity function f (y) = y Used mostly in linear situations
Step function f (y) = A for 0 < y Not widely used, since the differential of the
f (y) = 0 for 0 ≥ y function do not exist
A is a constant, usually assigned to 1.0
Sigmoid function 1 Popular for nonlinear problems
f ( y) =
1 + e− A y
A is a constant, usually assigned to 1.0
Hyperbolic tangent function 2 Special form of the sigmoid function, A = 2
f ( y) = −1
1 + e− 2 y
f (y) f (y)
+1.0 +1.0
0.0 0.0 +0.5 +0.5

y
0.0 0.0
y
0.0 0.0
Identity (With a slope of 1.0) Sigmoid
f (y) f (y)
+1.0 +1.0 +1.0 +1.0
y y
–1.0 –1.0
0.0 0.0
Step Hyperbolic tangent (a special form of sigmoid)
FIGURE 6.24 Examples of activation functions.

The values of the weights and biases are optimized, generally in least square sense, so as to min-
imize the resulting error. Single neuron ANN has very limited practical use. We will consider more
complex ANN models next.
6.5.2.2 Multilayer ANN: Versatile Complex Problems

The two-layer ANN (Figure 6.25), is fairly suitable to solve linear problems. We drop the bias from
ANN illustrations for simplicity; however, their presence is implied for the rest of this volume. They
can also be used quiet effectively for some classes of nonlinear problems. For example, monotonic
type nonlinear problem where the input-output relationships are expected to behave in a monotonic
fashion, such as a simple nonlinear spring as in Figure 6.26a. Sigmoid or hyperbolic tangent activa-
tion functions can be used effectively in a two-layer network to simulate such a behavior. Consider
now a more complex nonlinear behavior, such as the softening spring of Figure 6.26b. It is obvious
that sigmoid or hyperbolic tangent functions will not be effective in simulating such a behavior in
a two-layer ANN. Because of this, additional layers between the input and the output layers are
introduced. These layers are called hidden layers. There can be as many hidden layers in an ANN
system as needed. We note that one or two hidden layers might be sufficient for most practical SHM
problems.
6.5.2.3 Theoretical Background of Multilayer ANN

One way to handle nonlinear problems is to use a nonlinear base function to connect the input layer
and the output layer. A popular nonlinear activation function is the sigmoid function, shown in
Figure 6.24. In addition to nonlinear activation functions, additional layers in the network are used.
Neural network with single layer

Output layer
Input layer
FIGURE 6.25 Two-Layers ANN model.
(a) (b)
Force
Force
Deformation Deformation
Monotonic nonlinear spring Softening nonlinear spring
FIGURE 6.26 Examples of nonlinear springs (a) monotonic spring, and (b) softening spring.
These additional layers are called hidden layers since they are not seen at the input or output levels.
However, the presence of the hidden layers adds immense versatility to the overall neural network.
Figure 6.27 shows a neural network with the essential input and output layers as well as with two
hidden layers. As usual, the weights in the networks are estimated by minimizing the overall error
between the computed results of the network at the output layer, and the measured, or training, data.
The adjustment of weights based on an error that is computed at the output layer was a straightfor-
ward task for networks with no hidden layers, as was demonstrated above. With hidden layers, the
error estimation process is more complicated. Remember that we have no means to compute errors
at hidden layers: there are no available measurements or data at the hidden layer. Errors (as defined
by a measure of difference between measured and computed data points) can only be estimated
at the output layer. The presence of hidden layers presents the obvious question which is: how to
distribute errors that are computed at the output layer back into the weights that connects to inter-
mediate hidden layers?
There are several methods that distribute errors in a neural network that includes hidden layers.
Among these methods are the delta rule and the generalized delta rule. We will present generalized
delta rule below.
Problem Description: We use matrix notations to describe ANN with hidden layer for simplicity.
Let the vectors {X}i, {Y}i and {Y}mi represent the input, computed output and measured output pairs
for the network, respectively. Note that i = 1…N TRIES, where N TRIES is the number of input-output
pairs that will be used in the training of the network. We note that the size of the vector {X}i is the
number of neurals in the input layer NIN while the size of the vectors {Y}i and {Y}mi is the number of
neurals in the output layer NOUT. For the network topology shown in Figure 6.27, we define the num-
ber of layers, including the input and output layers as NLAYER. For such topology, it is implied that all
neurals in layer J is connected to all neurals in layer J + 1, where J = 1…(NLAYER − 1). The weights
connecting layer J and layer J + 1 is placed in a matrix. [W] ji Similarly, a bias vector is defined as
{B}ji. The size of the bias vector is (NLAYER − 1). The number of rows in matrix [W]mi is equal to the
number of neurals in layer J + 1. The number of columns in matrix [W] ji is equal to the number of
neurals in layer J.
The objective of the analysis is to compute the weights [W] ji and the bias {B}ji that minimize the
differences between the computed and the measured output as defined in a least square sense by
i = NTRIES
1 T
E= ∑ 2
({Y }i − {Y }im ) ({Y }i − {Y }im )
i =1
Neural network with two hidden layers
Hidden layers
Output layer
Input layer
FIGURE 6.27 Neural network with hidden layers.

The steps of the method include feed forward estimations of the responses of different layers, up to
the output level. Estimation of the error at the output layers is computed next. The error gradients
are propagated backward then. From these error gradients, estimates of the changes in the weights
and bias that will minimize the errors are computed. The steps (feed forward, error estimation, and
back propagation) are then repeated until the error reaches an acceptable level. The basic equations
for each step are summarized below.
Feed Forward: The solution begins by assigning arbitrary initial values for the weights and biases.
The response of layer 2 can be estimated as
{H}2 = [W]2{X} + {B}2
Note that we dropped the indicator i for simplicity. The response of next layer is
{H}3 = [W]3{H}2 + {B}3
The feed forward process continues until the last (output) layer
{Y} = [W]NLAYER-1{H} NLAYER-2 + {B} NLAYER-1
Generalized Delta/Back Propagation: The output error at each of the layers (output layers and
hidden layers) can be computed by computing the generalized delta vector {δ}J for each of these
layers. Note that the size of {δ}J is the same as the number of neurons in that layer. It can be shown
that for the output layer {d}N LAYER −1 = { f }N LAYER −1 ⋅ ({Y } − {Y } ).
m
For back layers

{δ} = { f} - [W]J+1 {δ}J+1
The vector {f}J is the derivative of the activation function with respect to the neural response:
∂f {u}J
{ f }j =
∂{u}J
Note that f(u) is the activation function as defined in Table 6.6.

The process continues backward for all layers, until layer number 2. There is no error involving
input layer since the input is exactly equal to the output for that layer.
Gradients for Weights and Bias: The gradients of the errors with respect to the weights can be
computed for the second layer as ∂e/∂[W]2 = {δ}2{X}T
For other layers
∂e
= {d}J {H}TJ +1
∂[W ] j +1

The gradients of the errors with respect to bias can be computed for each layer as
∂e
= {d}J
∂[ B] j

Iteration and Training: The above steps are repeated N TRIES times for all available training input-
output pairs. Reintroducing the i indicator for each learning pair, the total error gradient for each
weight matrix is
i = NTRIES
∂e
G([W ]J ) = ∑ ∂[W ]iJ
i =1
and for bias vectors

i = NTRIES
∂e
G({B}J ) = ∑ ∂{B}iJ
i =1
We can compute the change in each of the weight matrices and bias vectors as
Δ[W ]iJ = h G([W ]J ),
and
Δ{B}iJ = h G({B}J ),
respectively. While η is the learning step.

The weights and bias are updated as
[W ]J = Δ[W ]iJ + [W ]J
and
{B}J = Δ{B}iJ + {B}J
The process is repeated with the newly updated weights and bias, until the error E satisfies the con-
dition E < ∈, where ∈ is a redefined acceptable error limit. It should be noted that there are numer-
ous techniques for the procedure of optimizing the error. Among such procedures is the steepest
decent approach (Figure 6.28).
6.5.3 Advanced Topics

6.5.3.1 ANN versus FEM
Artificial neural networks can be used as a structural identification tool in similar fashion as FEM
method. The analogy between the two methods is shown in Figure 6.29. Table 6.7 shows additional
comparison between ANN and FEM methods as identification, or analysis tools.
6.5.3.2 Neural Network, Regression Analysis, and Expert Systems

The reader will perhaps recall that another method of formalizing a relationship between a given
set of input-output pairs, while using a minimization of errors using a least square approach, is the
regression method (see Chapter 8). In this sense, regression and neural networks are fairly similar.
In particular, the similarities are that both methods depend on a set of functional forms (activation
functions and regression equation for neural network and regression analysis, respectively) and the
Reduce total error

for each training
epoch
An “Epoch” is a complete
iteration training loop over
the whole available data
set. Several epochs can be
used in an iterative manner
to find optimum solution
Converage to optimum solution
FIGURE 6.28 Method of steepest decent for error optimization.

ANN paradigm FEM paradigm
I=WO KU = P
Train the ANN to get optimum W Build FEM, “K”
I=WO KU = P
Then for a new set of input “I,” get Then for a given input “P”
an esstimate for the output “O” get output “U”
FIGURE 6.29 ANN versus FEM methods.
TABLE 6.7
Detailed Comparisons of ANN versus FEM
Issue ANN FEM
Basic component Neurons, weights, and biases Finite elements (physical)
Resolution Number of layers, number of neurons Size of elements
Behavior Transfer (activation) function Stress-strain relationships
Needs Input-output pairings of dataset A single set of inputs and a
structural description
Objectives Relates input-output by evaluating Elements are preestablished by
weights and biases for future use user
Can it be hybrid? Mixture of ToMs Yes, without loss of any accuracy Yes. Special numerical
(displacements, strains, pressures, adjustments might be needed
temperature, etc.)
Can it scale local-to-global behavior in Yes, without loss of any accuracy Numerical and computational
single model? resources can be prohibitive
weights that control those functional forms (weights and bias for neural networks and regression
coefficients for regression analysis). However, a basic difference between the two methods is that
neural networks of the final functional form is not known a priori; it is determined after the network
learning process is completed. For regression analysis, the functional form of the regression model
is predetermined at the start of the analysis, and it does not change throughout the analysis. For both
methods, the weights (for neural networks) and regression coefficients (for regression analysis) are
determined from the process.
Another major difference between neural networks and regression models is that the output is
affected directly by the input during a regression analysis. That is also true for two-layer (input and
output layers) neural networks. However, for neural networks with layers more than two, that is, with
one or more hidden layers, these direct input-output relations are lost. The intermediate hidden lay-
ers interfere with the input-output direct relationships. This would result in a difficult interpretation
of the input-output relationship. This is an interesting balance: added accuracy and complexity from
using hidden layers versus increased difficulty of interpreting the input-output interrelationship.
Obviously, the added accuracy far outweigh the loss of interpretation capability, since, for complex
situations, such interpretation capability is too difficult to begin with.
Another well-known algorithm that has some resemblance with neural networks is the expert
system approach. The basic resemblance is that both methods can relate input-output relationships.
TABLE 6.8
Comparison of Expert Systems and Neural Networks
Expert System Neural Network
Need to know much larger body of knowledge to Do not require as much body of knowledge
make a decision
Very sensitive to presence of noise in data If trained well, can detect noisy data and produce accurate results
The whole expert system will need readjusting If additional data is acquired, only simple retraining of the network
rules, if additional data is acquired is needed
However, there are several differences between expert systems and neural networks, as shown in
Table 6.8.
6.5.3.3 Training versus Validation

It is customary to divide the ANN data set (input-output associations) into two subsets such that
1. Training data subset to realize optimum values (weights and biases) of an ANN
2. Validation data subset to ensure that the ANN performs properly
In general, training data subset is about 67%–75% of the whole data set. For SHM projects that aim
to utilize ANN it is essential to be cognizant of this need since it can affect decisions regarding
database design and collection.
6.5.3.4 Batch and Online Learning

One of the technical properties of ANN is how the learning/optimization process proceeds. There
are two distinct approaches: batch and online learning. The distinction is of importance to SHM
usage. Batch learning operates on the whole training dataset, while the online training operates
consecutively on each input-output record within the dataset. The rate of convergence of each
of the approaches would depend on the size and complexity of the dataset. For simple dataset,
it is expected that batch training would offer faster convergence. For smaller size data set, it is
expected that the online training would provide faster convergence. Thus it is not an easy task
to determine which of the two approaches is more suited for an ANN processing on a generic
basis.
For real time, or near real time SHM effort, it seems that an online training ANN process might
be the logical approach. In such a situation, we note that data records are continually monitored and
added to the main dataset. Such a situation is more suited to an online training than to batch train-
ing. More studies are needed to verify this otherwise logical observation.
6.5.3.5 Supervised versus Nonsupervised Learning

Artificial neural networks can be formulated so as to be updated and changed in a nonsupervised
mode, Trippi and Turban (1992). This is one of the major advantages of ANN when utilized for a
real time, or near real time SHM process. There aren’t many other STRID methods that have such a
feature. Of course, designing a nonsupervised learning process is not an easy task. Again, this topic
(use of nonsupervised ANN in the SHM field) needs further research that is not available as of the
writing of this volume.
6.5.3.6 How to Design a Neural Network

Figure 6.30 shows the basic components that need to be addressed when designing a neural net-
work. The types of activation function need to be chosen first. Table 6.6 can be used as a guide in
this choice. The important rule to remember when choosing an activation function is that it needs to
resemble, as closely as possible to the nature of the problem on hand. Another important issue is the
Basic design issues
Network Activation Error distribution Learning rates

topology functions methods
Connections Identity Delta
Generalized
Number of
Step delta
layers
Sigmoid
Hyperbolic tan
FIGURE 6.30 Neural network design issues.
choice of the error distribution methods. The delta and the generalized delta methods are the most
popular and in the following sections, we investigate the generalized delta method. Another issue
that must be decided is the topology of the network. This includes the number of hidden layers, the
number of neurals in each of the layers and the type of connections between the neural in different
layers. Finally, the rate of learning η needs to be addressed.
6.5.4 Applications of Neural Networks in SHM Field

Neural Networks are flexible tools that relates set of inputs to set of outputs in numerous ways, as
such, it is expected that there can be many types of Neural Networks applications. We can state
two main application types: they are the identification problems and the classification problems.
Figure 6.31 illustrates these two main types, and some specific uses of each type. In this chapter,
we will discuss the identification application of neural networks. Figure 6.31 illustrates the wide
range of identification applications of neural networks. For example, the method can be used to
identify the flexibility behavior model of the structure. Flexibility coefficients in a flexibility model
is the displacement at a given point and direction that results from applying a force at same (or
another) point and direction. Such a model is fairly simple and well identified. Humar et al. (2004)
proposed the use of neural networks to define the structure, then to use a modal type algorithm to
define the damage. A more complex model is the general structural (or modal) identification model
that is discussed extensively in this chapter. A much more complex and not well-defined model is
the damage identification application. Using neural network method for damage identification is
more complex due to the general lack of numerical or analytical damage models. Examples of the
three uses of neural networks are given in this section and the chapter on damage identification.
In Chapter 9 of Ettouney and Alampalli (2012), we will discuss the classification application of
neural networks.
Laurens et al. (2006) applied neural networks in structural identification (they aimed at identi-
fying the electrical resistivity of concrete, i.e., identifying the properties of materials). This shows
the benefits of using neural networks, especially if good database is available for training. The
neural network was trained with an experimental database of electrical resistivity of concrete (see
Figure 6.32) porosity, water content, and chloride content.
Types of neural network applications for srtuctural health

monitoring (SHM)
Classification Identification
Well identified Flexibility matrices

Bridge rating models of structure
Partially Structural
identified models identification
Models are not Damage

well known identification
FIGURE 6.31 Applications of neural networks in SHM field.
I
Wenner probe
V a
Concrete surface
Concrete
Equipotential lines Current

lines
FIGURE 6.32 Wenner device: principle of resistivity measurement. (Reprinted from ASNT Publication.)
6.5.5 Examples and Case Studies

6.5.5.1 Single Input, Single Output: Linear Neural Network
To illustrate the use of ANN in a simple structural identification problem, let us consider the stiff-
ness of a linear SDOF. Assume that during an SHM experiment a dataset of 33 force-displacement
points is observed. An ANN model with single input-single output neurons and no hidden layers is
used to model the observed behavior (Figure 6.23). The ANN model used linear activation func-
tion for simplicity. The error in both the weight and bias during the batch iterations are shown in
Figure 6.33. Clearly, the ANN model converged quickly. After 50 batch (rather than online) train-
ing epochs is found to be sufficient to produce an accurate model. The ANN produced a functional
relationship as shown in Figure 6.34. The figure also shows the ANN functional model after differ-
ent batch iterations. The least square solution is also shown in Figure 6.34. Again, it is clear that 50
batch iterations produced an accurate solution when compared to the least square solution.
6.5.5.2 Nonlinear Spring Behavior

A conventional nonlinear problem in structures is when the force-displacement relationship is a
monotonically increasing function. Suppose that a dataset of 55 force-displacement points is
observed, as in Figure 6.35. An ANN model with single input-single output neurons and three hid-
den layers, each with two neurons is used to model the observed behavior. The ANN model used
Convergence of errors for single neuron example

10
Error gradient -G 0
–5
Weight Bias
–10
–15
–20
–25
–30
0 10 20 30 40 50 60
Number of iterations
FIGURE 6.33 Convergence of errors for single-neuron example.
Neural network with single neuron

4.5
3.5
3 Measurments
1 Try
5 Tries
2.5 10 Tries
Force
15 Tries
2 20 Tries
25 Tries
50 Tries
1.5 Least Squares
0.5
0
0 0.5 1 1.5 2 2.5 3 3.5
Displacement
FIGURE 6.34 ANN behavior versus least squares.
hyperbolic activation function. After 100 online (rather than batch) training epochs, the ANN pro-
duced a functional relationship as shown in Figure 6.35.
A more difficult nonlinear problem in structures is when the force-displacement relationship is a
softening function. Such problem occurs in nonconfined concrete, or in columns that are subjected
to large axial forces beyond their stability limits. Suppose that a dataset of 128 force-displacement
points is observed, as in Figure 6.36. An ANN model with single input-single output neurons and
four hidden layers, each with three neurons is used to model the observed behavior. The ANN
model used hyperbolic activation function. After 100 online (rather than batch) training epochs, the
ANN produced a functional relationship as shown in Figure 6.36. Note that more information is
needed for this ANN model than the previous one due to the higher nonlinearity.
ANN and data set

18
16
14
12
Forces (Normalized)
Data
10
ANN
8
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
Displacements (Normalized)
FIGURE 6.35 ANN versus data for monotonic spring behavior.
Data set vs. ANN function

1.80E+01
1.60E+01
1.40E+01
Spring force (Normalized)
1.20E+01
1.00E+01
ANN
8.00E+00 Data
6.00E+00
4.00E+00
2.00E+00
0.00E+00
0 10 20 30 40 50 60
Displacements (Normalized)
FIGURE 6.36 ANN versus data for softening spring behavior.
The use of ANN in simulating spring behavior can be expanded for even more complex STRID
utilization. Consider, for example, a bridge testing experiment. Assume that trucks with a known
weight P were positioned at N stations along the bridge. For each station, a set of M displacements
were measured at M stations. This produced a load vector{F} and an equivalent displacement matrix
[U]. The order of the load vector is N and the size of the matrix is M × N. If the matrix [U] is rear-
ranged in a vector form {D} where the displacement set for the trucks at each loading station, then
we recognize the two vectors pair {F} − {D} set as a potential single data set in an ANN model.
Let us assume that the experiment is repeated NANN times to produce {F}i − {D}i data sets with i =
1, 2 …NANN. Using ANN, this data set can be used to produce functional stiffness, flexibility, or
hybrid approaches as follows:
ANN-Based Stiffness Approach: If we assume that the input layer for an ANN model is
the forces {F} and the output layer is the displacement {D}, this would make the ANN
model a stiffness model. The input and output layers would have an N and an M neurons,
respectively.
ANN-Based Flexibility Matrix: If we assume that the input layer for an ANN model is
the displacement {D}, and the output layer is the forces {F} this would make the ANN
model a flexibility model. The input and output layers would have an M and an N neurons,
respectively.
ANN-Based Hybrid Approach: Given the versatility of the ANN method, different hybrid
models can be built. Such models can include smaller subsets of either forces and/or dis-
placements as input, or output. This can easily be done without impacting the overall accu-
racy or resolution of the resulting ANN models.
Advantages/Limitations of Using ANN: The advantages of using ANN to produce functional

stiffness/flexibility matrices are as follows
• Accounts as built conditions

• Accommodates any nonlinear effects in an accurate manner
• Resolution can be improved by adding more data sets with minimal efforts (grafting/
pruning)
• Avoids the inherent numerical and size limitations of conventional finite elements
approximations
• Changes in trends can be monitored in real time
• Can model both local and global behavior within the same ANN model
In addition to all of above advantages, we note that there is no reason to expand the vector
{D} such that it includes other ToM than just displacement. For example, in an experiment, the
vector {D} can include displacements, strains, rotations, and stresses. There should be no loss
of accuracy of the ANN model; a similarly hybrid FE model can experience loss of accuracy,
Zienkiewicz (1971).
The main limitation of using ANN as a stiffness/flexibility tool is that it is a functional represen-
tation, that is, the results are difficult to relate to actual physical parameters. Such limitation might
be overcome by taking advantage of the potential of measuring local as well as global responses of
varying ToMs. Further research on this intriguing aspect of ANN is needed.
6.5.5.3 Bridge Scour

An SHM application of the use of ANN is discussed here. The problem involves the identification
of the equilibrium scour depth around bridge piers. Scour is one of the most dangerous hazards
that afflict bridges (see Figures 6.37 and 6.38). Scour problem is discussed extensively in Chapter
1 by Ettouney and Alampalli (2012). The scour problem results when rapid flowing waters around
bridge foundations and piers causes soil erosion. When the degraded soil properties reach a thresh-
old, the foundation/piers fail. It is highly nonlinear and involves several parameters including soil
and foundations properties. It also involves the hydraulics of the water flow around and near the
foundations. Jeng et al. (2005) used ANN to study damaging effects of scour near bridge piers.
The remainder of this section describes their procedure and findings. We note that this study is
more of damage identification study, rather than a structural identification study; it aims directly to
FIGURE 6.37 Scour damage. (Courtesy of New York State Department of Transportation.)
FIGURE 6.38 Scour damage. (Courtesy of New York State Department of Transportation.)
identifying the scour damage, rather than identifying the system properties. The authors identified
five variables that might affect the scour process as follows
U c = Critical flow velocity

D= Pier diameter
U= Average velociity of approach flow (6.131)
Y= Depth of approach flow
d50 = Median grain parameter
Thus, the ANN model has five input neurons. A single ANN output that describe the scour damage
was chosen to be the equilibrium scour depth dse. Figure 6.39 illustrates the problem parameters.
Figure 6.40 shows the ANN model for the problem. The model has five input neurons and one out-
put neuron. There is a single hidden layer with four neurons. A sigmoid transfer (activation) function
was chosen for the problem. The authors chose a 187 data points that relates the five input param-
eters to the desired output dse. A learning rate of 0.4 was chosen. A total of 9000 epochs were used
to formulate the ANN.
D
Pier
Water surface
Y U
dse
FIGURE 6.39 Simplified scour problem.
Neural network for scour problem
Input layer Hidden layer Output layer
Uc
U dse
d50
FIGURE 6.40 Scour ANN model.
Figure 6.41 shows the results of the ANN model when compared with actual measurements of
dse. The ANN model is well behaved. It can be used to predict scour behavior in similar settings.
Such a model can also be used during a real time SHM experiment as follows:
• It can be used in a nonsupervised mode to improve its parameters as more observation data
sets are monitored
• If there is a predetermined threshold limit state of dse, an automatic warning signal can be
sent to bridge officials whenever the ANN model predicts such a threshold in real time
6.6 OTHER METHODS

6.6.1 Frequency Ranges
Earlier in this chapter, we subdivided qualitatively the frequency spectrum into three ranges: low,
medium, and high ranges. We need to discuss this concept further. Low frequency range can be further
defined as the range where wave lengths are longer than the typical length of the component of interest.
Typically, few wave modes exist in the low frequency range. Also, in this range, the coupling between
(4,1)
Epochs = 9000
0.3
LR = 0.4
mc = 0.2 +15%
0.25
–15%
Real values of dse
0.2
RMS = 0.001013
0.15 CC = 0.962696
0.1
0.05
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Training results
FIGURE 6.41 ANN results. (Courtesy of Department of Civil Engineering, University of Sydney: Jeng et al.
[2005].)
Natural modes
High frequency
Low frequency
range
range
At low frequencies, Frequency scale At high frequencies,

the modes are well the modes are tightly
spaced, thus the spaced, thus the
uncertainties in uncertainties in
location and shapes location and shapes
are low are high
FIGURE 6.42 Mode distributions at low- and high-frequency ranges.
the wave modes is not significant, due to the relatively longer frequency steps between the different
wave modes. Thus uncertainties due to coupling between modes are not high. Because of this tradi-
tional finite element methods can be used as the basis of STRID efforts with reasonable accuracy.
High frequency range is defined as the range where wave lengths are shorter than the typical length
of the component of interest. Numerous wave modes will exist in this frequency range. Because of
this, the frequency spacing between the modes becomes small. Thus the coupling between the wave
modes can be significant, due to this short frequency spacing. The uncertainties due to the coupling
between modes are high. Because of this, the accuracy of traditional finite element methods becomes
unacceptable. Intermediate frequency range is the range where some short and long wave modes
exist. Figure 6.42 shows the concept of modal coupling at low- and high-frequency ranges.
Because of these physical uncertainties, there is a need for different STRID methods in the high
and medium frequency ranges. Several methods accommodate those needs. Some of them are dis-
cussed next.
6.6.2 Energy-Based Methods

6.6.2.1 Statistical Energy Analysis (SEA)
Lyon (1975), Lyon and Maidanik (1962), and Keane and Price (1969) observed that at high frequency
range, the vibration energy of a system can be averaged using representations of its modal densities.
The modal density in a particular vibration entity n is an estimate of the number of resonances (modes)
that exist in a prescribed frequency range. Thus, by estimating the energy losses in the system, which
can be shown to relate to modal densities, and the energy input to the system, an energy balance equa-
tion can be established.
In SEA, the system is subdivided into physical components. The physical components in a civil
infrastructure can be plates, beams, shells, or columns. If a physical component can excite more
than one type of waves, energy flow for each wave type need to be considered separately. For exam-
ple, beams need to have four subsystems for each of the four wave types that can propagate within
it (axial, torsional, flexural, and shear). For plates, shear, flexure, and axial waves need to be repre-
sented by three subsystems. The energy flow within each of the subsystems can be characterized by
the component length, shape, and material properties (which include damping). This subdivision is
one of the basic tenets of SEA: evaluating properties of subsystems (either by testing or analytically)
is feasible at high frequencies. Analytical expressions of many of the parameters for the subsystems
are available. Thus, the energy flow between the subsystems can be analyzed in a formal fashion.
As an example, let us consider a vibration of a system that contains several components; each
component contains its own subsystem, as appropriate. The total number of subsystems is N. The
energy balance equation of the system is expressed as follows:
Ω[Η ]{Ε} = {Ρ} (6.132)

The driving frequency is Ω, the energy loss matrix is [H], the subsystem energy vector is {E}, and
the vector of the power input to the system is {P}. The order of Equation 6.132 is N.
The components of the matrices in Equation 6.132 was formalized by Koizumi et al. (2002) as
follows
⎧ Pi1 ⎫
⎪P ⎪
⎪ ⎪
{Ρ} = ⎨ i 2 ⎬ (6.133)
⎪ ⎪
⎪⎩ PiN ⎪⎭

The input power to the jth subsystem is Pij. The subscript i indicates input power. Also:
⎧ E1 ⎫
⎪ ⎪
⎪E ⎪
{Ε} = ⎨ 2 ⎬ (6.134)
⎪ ⎪
⎪⎩ EN ⎪⎭

with
Ej
Ej = (6.135)
nj

The average energy and modal density of the jth subsystem are Ej, and nj, respectively. Energy loss
matrix [H] is expressed as
⎡ h11 h12 h1N ⎤

⎢h h22 h2 N ⎥⎥
[Η ] = ⎢ 21 (6.136)
⎢ ⎥
⎢ ⎥
⎢⎣ hN 1 hN 2 hNN ⎥⎦

Koizumi et al. (2002) proved that the diagonals of [H] are
⎛ i= N ⎞
h jj = n j ⎜ h j + ∑ h ji ⎟ (6.137)
⎝ i≠ j ⎠

and for off-diagonal (i ≠ j) of [H],

hij = − ni hij
(6.138)
The energy loss within the ith subsystem is ηi. The coupling loss factor from ith subsystem to jth sub-
system is ηij.
Equations 6.132 through 6.138 provide the basis of SEA method. The properties of the ith sub-
system, ηi, ni, and the coupling loss factors to the jth, ηij, provide complete description of the system.
For any know input power distribution to the whole system {P}, the energy flow in the whole system
{E} can be estimated. At a subsystem level, the energy can be related to the average velocity vi of
the subsystem as
Ei = Mi vi2
(6.139)
The mass of the subsystem is M.

SEA is a popular method in the fields of automotive, aeronautics, and naval architecture. Its sim-
plicity and applicability at high frequency range account for this popularity. Limited applications
are known in the field of civil infrastructure. Koizumi et al. (2002) used it to explore noise flow
in buildings. One potential use is in estimating remaining fatigue life evaluation. Instead of using
strains as metric for the evaluation, energy content in frequency band is used (see Lyon 1975). Since
incipient damage is felt first by higher modes before it is felt by lower modes, detecting damage
using SEA holds high potential. López-Díez et al. (2005) used SEA to detect damage in spacecraft
structures. However, the use of SEA in the fields of STRID and/or DMID in civil infrastructure
remains largely untapped as of the writing of this volume.
6.6.2.2 Energy Finite Element Analysis (EFEA)

Another energy-based method was developed to model both high and medium frequency ranges
in an analogous manner to traditional finite element method. The EFEA is based on computing an
average of the energy in space and frequency range in a similar manner to the computation of dis-
placements by traditional finite element method. For example, the traditional flexural beam equation
is (Timoshenko 1955)
d4 y (6.140)
EI + my = p( x, t )
dx 4
The displacements and accelerations for the beam are y and y. The flexural rigidity and mass of the
beam are EI and m, respectively. The applied load is p in the space and time variables x and t. The
equivalent average flexural equation can be written as (Zhao and Vlahopoulos 2004)
−cg2 d 2 e
+ h Ω e = Qin (6.141)
h Ω dx 2
The flexural group wave velocity is cg. The hysteretic damping is expressed by η and the driving
frequency is expressed by Ω. The unknown energy density is e . The averaged input power is Qin .
As before, the averaging is over the time (frequency range) and space. Equation 6.141 is expressed
in matrix form for a single ith beam element as (Bernhard and Huff 1999)
[ E e ]i {ee }i = {F e }i + {Q e }i (6.142)

The element matrix is [Ee]i. The input power vector is {Fe}i, and the internal power flow is {Qe}i.
The average nodal energy density vector is {ee}i. The superscript e indicates element level in the
equation. The analogy between Equation 6.142 and traditional element equation is obvious. The
general equation of energy flow of the system under consideration is then assembled using elements
equations similar to Equation 6.142. The power flow between the ith and jth elements are accounted
for using special energy coupling equation in the form (Cho 1993)
e
⎪⎧ Qic ⎪⎫ ⎧ ee ⎫
i ⎪ ic ⎪
⎨ e ⎬ = [J ]j ⎨ e ⎬ (6.143)
⎩⎪Q jc ⎭⎪ ⎩⎪e jc ⎭⎪
The matrix [ J ]ij includes power transfer values between the ith and jth elements. The subscript c indi-
cates common nodes between the ith and jth elements. Finally, the joint energy flow equation between
the ith and jth elements is expressed as (Zhao and Vlahopoulos 2004)
⎛ ⎡[ E e ]i ⎤ i
⎞ ⎪⎧{ee }i ⎪⎫ ⎪⎧{F e }i ⎪⎫
⎜⎢
⎜⎢ e ⎥ + [ JC ] j ⎟⎟ ⎨ e ⎬ = ⎨ e ⎬ (6.144)
⎝⎣ [ E ] j ⎥⎦ ⎠ ⎪⎩{e } j ⎪⎭ ⎪⎩{F } j ⎪⎭

The joint matrix [ JC ]ij is a rearranged version of [ J ]ij to accommodate the nodal sequencing in the
model. The resulting equations can then be used to estimate energy flow at high- and medium-
frequency ranges for structural systems.
Vlahopoulos et al. (1999) used EFEA to evaluate energy flow at high frequency of welded plates
as shown in Figure 6.43. A comparison between test measurements and results from EFEA computa-
tions are shown in Figure 6.44. The accuracy of analytical results at such high frequency is evident.
Energy finite element analysis holds great potential in the field of STRID in civil infrastruc-
ture. It can identify structural properties at high and medium frequencies, thus compliment low
frequency STRID methods. When compared to SEA, the EFEA can model more system details and
EFEA can utilize the vast numerical solution algorithms that are available for conventional FEM,
Vlahopoulos et al. (1999).

There are other energy-based methods for analyzing system behavior at high frequency ranges.
Sarradj (2004) provided a discussion of those methods. High frequency boundary element method
(HFBEM) is based on boundary integral formulation that uses energy instead of displacement vari-
ables. Sarradj (2003) compared the results of SEA, EFEM, and HFBEM. An energy source is
Shaker
location
Measurement
10' area in excited
6.31' 10'
plate
9.75' 37'
21.785'
40'
39'
10.5' 30' 21.75' 10.5'
Measurement
12.625' area in
A 10.375'
receiving plate
36'
FIGURE 6.43 EFEA test setup. (Reproduced with permission from Academic Press.)
1.0
0.8
Energy ratio
0.6
0.4
250 400 630 1000 1600 2500

Frequency
(___) Test Data, (---) Updated EFEA, (......) Baseline EFEA
FIGURE 6.44 EFEA test results. (Reproduced with permission from Academic Press.)
Source (2,3) (2.8,3)

Start of
pos.
path
(2.5, 1.5)
(0,2)
(1,1)
(0.3,0.5)
(0,0) (4,0)
FIGURE 6.45 Steel plate experiment. (Courtesy of Dr. Ing. Ennes Sarradj.)
applied to a large steel plate using a shaker as shown in Figure 6.45. The energy along the path indi-
cated in Figure 6.45 was computed using the three methods at different driving frequencies (1 and
10 K.Hz) and two damping levels (0.1% and 1%). The results are shown in Figure 6.46. The EFEM
and HFBEM show more path-dependent distributions than the SEA, as expected. Also, HFBEM
seems to produce higher fidelity results than EFEM.
10–1
f = 1 kHz, = 0.1%
10–2
Energy density (J/m2)

f = 1 kHz, = 1%
10–3
f = 10 kHz, = 1%
10–4
EFEM
HF-BEM
SEA
10–5
0 1 2 3 4 5 6 7
Path length (m)
FIGURE 6.46 Comparison of energy-based methods. (Courtesy of Dr. Ing. Ennes Sarradj.)
The energy-based STRID methods offer potential in identify systems at high frequencies. Thus
complimenting traditional low frequency MI methods. Their capability in directly identifying small
size damage is an additional obvious advantage.
6.6.3 Dispersion Analysis

When the speed of wave propagation in a system varies as a function of frequency, the system is
called dispersive system. The relationship between wave phase velocities in solid systems and fre-
quencies of vibration is called dispersion relations. Such relationships depend on geometry and
material properties of solid. When a system is dispersive, the wave propagates at different speeds at
different frequencies. By understanding dispersive relations, the behavior of systems can be under-
stood. Numerous studies regarding dispersive systems and the nature of wave propagating through
them. A simple way of illustrating the dispersion relations is by plotting such relationship in a dia-
gram, such diagrams are called dispersion curves. Detailed descriptions and derivations of disper-
sion relations are given by Rose (2004).
Figure 6.47 shows dispersion curves of axial waves in solid infinite cylinder (Rose 2004). The
dispersion relations of circumferential waves in hollow cylinder are shown in Figure 6.48 (Rose
2004). Dispersion curves can be obtained analytically for simple systems. For example, the devel-
opment of the dispersion relations for plates is shown in Section 7.10.
As the system geometry becomes more complex, analytical solutions of dispersion relationships
becomes difficult. Numerical solutions, such as finite element methods can be used. Scale indepen-
dent element (SIE) method can also be used to evaluate dispersion relationships (Chapter 8).
Since dispersion relations are fundamental part of system behavior, we consider their identifica-
tion as an STRID process. The knowledge of a dispersion relation for a given system is similar to the
knowledge of modal properties of a system: both will help in understanding certain aspects of the
system, albeit at different space and frequency scales. In addition, given the applicability of disper-
sion relations to high frequency ranges, it is possible to use their measurements directly to identify
damage presence, location, shape, and extent (see Chapter 8).
6.6.4 Genetic Algorithms

Another class of STRID uses genetic algorithms (GAs). Koh and Htun (2004) described a GA meth-
odology in STRID. As of the writing of this volume GAs are not as widely used in STRID civil
infrastructure applications as other methods described in this chapter.
25
20
cp (km/s) 15
10
0
0 5 10 15 20 25 30 35 40 45 50
f d (MHz-mm)
FIGURE 6.47 Dispersion curves for axial waves in solid cylinder. (Rose, J. Ultrasonic Waves in Solid
Media, Cambridge University Press, Cambridge, UK, 2004. Reprinted with the permission of Cambridge
University Press.)
9
8
7
6
5
cp (km/s)
4
3
2
1
0
0 1 2 3 4 5
f (MHz)
FIGURE 6.48 Dispersion curves for circumferential waves in hollow cylinder. (Rose, J. Ultrasonic Waves in
Solid Media, Cambridge University Press, Cambridge, UK, 2004. Reprinted with the permission of Cambridge
University Press.)
6.7 MODELING TECHNIQUES

6.7.1 Overview
Many times the accuracy of STRID solutions depends on modeling techniques of system components
other than the structural system itself. In particular, modeling of soils and water can significantly
influence results of STRID problems. We discuss next some modeling issues with regards to soils
and water. We conclude the section with discussion of some general modeling issues that can affect
STRID activities such as model scaling, static condensation, and molding problems in a multihaz-
ards environment.
6.7.2 Soil-Structure Interaction Role in Structural Identification Problem

The interaction between the structure and its supporting soil is an important factor in the overall
behavior of the system in many, but not all, situations. Because of this, several authors pointed to the
importance of SSI in STRID problems. In general, whenever we encounter a structural identification
situation, it is important to (1) decide whether the Soil Structure Interaction (SSI) effects are impor-
tant in that particular situation, and (2) the most effective way to account for such an effect.
There are some general rules that can help in assessing the importance of SSI in any structural
identification situation. The importance of SSI can be ascertained based on type of hazard, magni-
tude of hazard, and/or overall relative structure-to-soil stiffness. Some hazards would require the
consideration of SSI in any structural identification problem. For example: it seems logical to con-
sider SSI effects while addressing the effects of scour and earthquakes. On the other hand, SSI effects
are negligible when addressing corrosion effects. Higher levels of some hazards can increase the
importance of SSI effects. Impact of barges on a bridge pier might require the consideration of SSI
effects. Perhaps the most important factor that controls the importance of SSI effects is the relative
stiffness of the structure and the underlying soil. If the structure is supported by a rigid foundation,
the SSI effects are negligible. Such a situation arises when the structure is supported directly by
rock. When the structure becomes stiffer, relative to the supporting soil, the SSI effects need to be
considered. Note that this relative stiffness effects must be considered in a static and dynamic sense,
depending on the kind of problem under consideration. For slowly varying loads, such as a bridge
load testing situation, the condition can be assumed as a static condition. In such situation, the relative
structure-to-soil stiffness can be evaluated using the relevant static stiffness parameters.
As a guide for evaluating the relative importance of SSI effects for the rotation of the bent foun-
dation, if the soil rotational stiffness is KSOIL_𝜙 and if the bent column rotational stiffness is KSTR_𝜙,
then the scalar ratio
K STR _ f
a= (6.145)
K SOIL _ f

should indicate the need for accommodating SSI effects for such a problem.
For dynamic situation, Equation 6.145 must be generalized to account for time, or frequency,
effects. In the time domain:
K (t )STR _ f
a (t ) = (6.146)
K (t )SOIL _ f

and in the frequency domain
K (Ω)STR _ f
a (Ω) = (6.147)
K (Ω)SOIL _ f

The time and driving frequency are t and Ω, respectively. Evaluating 6.146 in the time domain is not
a trivial endeavor, since it requires solution of iterative nonlinear problem. Evaluating Equation 6.147
in the frequency domain is a bit easier. There are some frequency-dependent soil stiffness expres-
sions that are available (Ettouney 1978 and 1979). Also, some frequency-dependent structural stiff-
ness were reported in some cases, see, for example, Ettouney (1979). Note that α alpha in Equation
6.146 is a real function while α in Equation 6.147 is a complex number.
Soil-structure interaction effects need to be evaluated for more than just rotation of foundations.
They need to be evaluated for horizontal, vertical, and tensional motions. In some situations, cou-
pled SSI motions should also be investigated.
We should emphasize that the above are merely general rules. The analyst should study the spe-
cific condition of the problem under consideration, and make the decision on a case-by-case basis.
Let us assume that the analyst decided that SSI effects are important to accommodate in the
STRID problem under consideration. In such a case, the analyst must answer two questions: (1)
What is the acceptable resolution of the soil model, and (2) What type of nonlinear material model
is needed for the soil representation?
TABLE 6.9
Some Soil Modeling Techniques
Model Advantages Disadvantages
Linear spring Simple suited for modal analysis methods Accurate only for static, or low
frequency zone
Nonlinear spring Can simulate certain aspects of soil Amount of energy dissipation is
nonlinearity neither controllable, nor verifiable.
Can produce unconservative results
Frequency-dependent spring Better representation of damping Requires frequency domain analysis
Multiple degree of freedom Suited for modal analysis methods (if soil Analytically demanding
(finite elements) elements are linear)
Accommodates higher modes
Better geometry representation
There are many types of soil models that can be used in a soil-structure coupled modeling
method. Some of the popular models are as follows: (1) single linear spring, (2) nonlinear spring,
(3) single frequency-dependent spring, or (4) multiple DOF (finite elements soil representation). The
advantages and disadvantages of each model are shown in Table 6.9.
The above discussed only coupled soil-structural modeling issues. It must be mentioned that
there are several other methods that are based on decoupled modeling of SSI situations. Decoupled
methods are usually based on certain assumptions. We believe that the magnitude of errors that are
introduced by those assumptions might not be acceptable in a structural identification effort. Until
a detailed study on the adequacy of decoupled soil-structure modeling for STRID problems is per-
formed, it is recommended that only coupled soil-structure methods as discussed above is used for
structural Id situations.
Finally we discuss the nonlinear soil behavior issue. We observe that soil is a nonlinear mate-
rial. We immediately observe that most STRID methods that were discussed in this chapter, and
elsewhere, are based on linear modeling techniques. This poses the following apparent paradox: if
SSI effects are important, and since soil behavior is nonlinear in general, then we should expect to
use nonlinear methods for the STRID problem. However, most of the structure STRID methods are
linear. There are two possible solutions to this apparent paradox: either use linear approximation of
the soil properties, or use a true nonlinear STRID method. Again, nonlinear STRID methods are
not well developed as of the writing of this volume. Further research is needed in this field.
6.7.3 Soil-Structure Interaction Role in Damage Identification Problem

Sometimes, it is desired to utilize STRID techniques to identify damage directly. Olson (2005)
used parameter identification techniques to identify scour damage (as simulated by certain level of
exposure of piles). When soil and structure are components in an STRID effort, attributes such as
soil-structure interface, soil nonlinearities (especially near the interface), and the potential impor-
tance of higher modes need to be considered in such models. If water is present, such as submerged
foundations during a scour process, the water-structure interaction behavior should also be consid-
ered carefully.
6.7.4 Fluid-Structure Interaction (FSI) in STRID Problem

Structural identification for potential use in scour problems will naturally concern itself with local
scour; other types of scour damage occur away from the bridge structure where STRID efforts
would not be of particular value. The distinguishing feature of local scour, from STRID view point,
is that it should involve the interaction of three very distinct systems: the structural foundation, the
Fluid structure
interface needs
to be modeled
Fluid For proper identification,

mode shapes should be of similar
length as scour hole depth
Scour hole
Soil structure
Soil interface needs
to be modeled
FIGURE 6.49 STRID modeling needs for scour problem.
surrounding soil, and the fluid. In a MI test, the frequency characteristics of the combined system
should be considered. We observe that the presence of the fluid would have an effect of an added
mass to the system, assuming low frequency range, see Ettouney and Daddazio (1978). The added
hydrodynamic mass can have large effect on the dry modes of systems, Ettouney et al. (1990).
Similarly, soil effects should be considered in any scour-related STRID effort. In particular, we
note that scour holes near foundations may result in slight changes of low modes (both frequency
and eigenshapes), until the system approaches the state of instability. A way to increase sensitiv-
ity of MI of scour hole presence is to identify higher response modes. Mode shapes with specific
length that is close to the scour hole depth, or shorter, offer better chance of being identified (see
Figure 6.49).
6.7.5 Static Condensation Modeling Techniques

Conventional SSI problems, such as seismic, or scour problems, would require modeling both soil
and structural system. Such a model can be computationally demanding. A popular method is to
subdivide the model into soil and structure parts, as shown in Figure 6.50. The governing matrix
equation for the soil is
[ K ]* {U } = {P}* (6.149)

The soil stiffness is [K]*. The soil displacement vector is {U} and the applied loads on the soil is
{P}*. Equation 6.149 can be subdivided as
⎡ K11 K12 ⎤ ⎧U1 ⎫ ⎧ P1 ⎫

⎨ ⎬=⎨ ⎬

⎢K
⎣ 21 K 22 ⎥⎦ ⎩U 2 ⎭ ⎩ P2 ⎭ (6.150)
The subscripts i = 1 represent the DOF at soil-structure interface. The remainder of soil DOF is
represented by subscripts i = 2. Operating on Equation 6.150 we get
[ K ]* {U1} = {P}*
(6.151)
Foundation
2-Will be attached to
pertinent structural
degrees of freedom
1-Condensed soil
degrees of freedom
Soil
FIGURE 6.50 Static condensation concept.
where
[ K ]* = [ K11 ] − [ K12 ][ K 22 ]−1 [ K 21 ] (6.152)

Equation 6.151 can now be considered as multi-DOF soil spring. The equivalent soil spring [K]∗ can
be added to the pertinent structural foundation DOF. The soil is now completely represented in the
structural model. The resulting soil-structure model can now be either solved using conventional
finite element techniques, or utilized efficiently in an STRID parameter identification problem.
Such an approach in dynamic problems would lead to inaccurate results, since it produces modal
relationships that do not resemble the actual modal relationships. Note that the soil dynamic equa-
tion is
[ M ]{Ü} + [ K ]{U} = {P} (6.153)

The soil mass matrix is [M] and the soil acceleration vector is {Ü}. The condensed matrix [K]*
does not have any information regarding the inertial properties of the soil; thus it is incapable of
reproducing accurate soil modal behavior. It also truncated the number of soil modes arbitrarily, so
that higher soil modes can’t be present in the system. The conservatism, or the lack of it, can’t be
assured in such an approach. In parameter identification, the problem is compound, since the static
condensation matrix [K]* is not a physical matrix: it is a numerical approximation. Trying to iden-
tify any of its components might produce meaningless results: we are trying to update a nonphysical
entity to a physical one.
6.7.6 Scaling Needs

When MI is used for the purpose of STRID, the professional usually tries to verify a numerical
model of the structure by the identified modes. In parameter identification, the numerical model
itself is updated using the experimental results. In both situations the resolution of the numerical
model should be adequate for accurately modeling the frequency range of the dynamic test. For
example, it is common practice to model structural columns by few, usually single, finite elements.
Dynamically, the dynamic range of such an element is limited to single flexural and single axial
modes. Higher modes can’t be captured by single element. For higher modes, additional elements
are required, to ensure accurate dynamic representation. As the dynamic frequency range of interest
is increased, the number of elements increases. At certain level, this can add unacceptable numeri-
cal demands on the model. Scaling techniques, such as scale independent element (described in the
next section), can help in reducing such demands, while preserving needed accuracy.
6.7.7 Multihazards Considerations and STRID

There is a tight relationship between STRID modeling and multihazards considerations. However,
the three major STRID techniques vary with regards to their multihazards interactions, as follows:
6.7.7.1 Modal Identification
Identifying modal information is related to hazards that are dynamic in nature. Chapter 8 pres-
ents qualitative ranges of some important hazards. It is logical to relate the frequency range of
desired identified modes to the range of hazard of interest. If there is a mismatch in the indentified
modal range and the hazard of interest, the value of the identified modal properties is reduced. This
implicit interrelationship can control the type of MI project.
6.7.7.2 Parameter Identification
If the system under consideration is numerically modeled (by finite elements method, for example)
then, according to multihazards theory, such a numerical model can be utilized to analyze all haz-
ards of interest. The physical system accommodates all interdependencies of all the hazards, and
hence an analytical model should be capable of handling all hazards.
Unfortunately, this is far away from the truth. In translating the physical system into an analytical
model, numerous simplifying assumptions must be made and these assumptions would affect how
the analytical model analyzes and estimates the response to different hazards. Specifically, these
assumptions can be classified into three categories: spatial, temporal, and response level (ampli-
tude). Spatial assumptions include the size of discretization (element size in finite element method),
extent of the analytical model and boundary conditions (see Figure 6.51). Temporal assumptions
include frequency range of analytical model and time integration schemes, if applicable. Response
level (amplitude) assumptions include types and levels of nonlinearities. Obviously, the assump-
tions made would depend on the hazard of interest and the expected behavior of the structure to
that hazard. Any hazard has to be described spatially, temporally (frequency), and quantitatively
(amplitude), as shown in Figure 6.51. Figure 1.26 in Chapter 1 shows schematically how different
hazards command different ranges in the response level (amplitude)-temporal (Frequency) space.
Figure 4.25 in Chapter 4 shows spatial characteristics of various hazards.
Amplitude (hazard level) scale
Temporal (frequency) scale

d,
a zar
le: of h etr y
sca n
ti al utio geom
a b
Sp istri tural
–D truc
–S
FIGURE 6.51 Hazards attributes.

Measurements
(SHM projects)
STRID technique
Conventional
STRID
paradigm
Identify structural
parameters
Functional
relationships (e.g., Modal parameters Physical parameters
neural Networks)
FIGURE 6.52 Conventional STRID paradigm (without accommodating multihazards).
For a given defect/damage from an

SHM project
Additonal
STRID technique
multihazards
STRID
paradigm
Identify the hazard,

or hazards that
caused the defect
FIGURE 6.53 Multihazards-based STRID paradigm.
On the basis of the previous discussion, it is clear that different hazards demand different numer-
ical assumptions and numerical analysis techniques to ensure accurate analytical results. So, we
reach the following important conclusion: even though the physical system can accommodate dif-
ferent hazards, including their interdependencies, numerical analysis and modeling of different
hazards would require separate analysis and modeling considerations. Ignoring this would produce
inaccurate results. This can be simply exemplified through multihazard parameter identification of
bridge piers for purpose of seismic or blast load investigations. For seismic hazard, the design and
analysis model can be a simplified beam/column model. On the other hand, a higher precision FEM
model may be required for analysis and design for blast load cases. Figures 6.52 and 6.53 show the
conventional and multihazards-based STRID paradigms, respectively.
6.7.7.3 Neural Networks
By definition, neural networks represent direct informal relationship between input and output.
Different hazards, such as traffic loads, scour, and earthquakes are generally the input to the net-
work. The output can be any desired response function, such as displacements, and scour damage.
Because of the large variations in hazard special and temporal characteristics, having a network that
can span over different hazards is almost impossible.
6.8 SCALE INDEPENDENT METHODS (SIM)

6.8.1 Scale Independent Element
Discrete deterministic methods such as finite elements offer great flexibility in analyzing the dynamic
response of vibrating systems. However, these methods can easily grow beyond available computer
resources as frequencies of interest grow higher. In this chapter we present a new approach for the fre-
quency domain dynamic analysis of structures. A theory is developed for the analysis of systems that
are uniform along a single coordinate axis but otherwise arbitrary in geometry and material composi-
tion. This approach, termed the Scale Independent Element (SIE), is shown to be an accurate, efficient,
and general method for the analysis of vibrating systems. This technique extends the applicability of
discrete deterministic finite element based modeling to higher frequencies and is capable of bridging
the gap to frequency regimes where statistical energy methods become applicable. Ettouney et al.
(1995) offered a detailed description of the technique. We reiterate their development next.
6.8.1.1 Overview
Three frequency regimes are usually distinguished in the analysis of structural vibrations in complex
systems. At very low frequencies, deterministic modal methods (Leung 1988) are well established.
At high frequencies, asymptotic methods such as ray acoustics and frequency response smooth-
ing by geometric mean values (Girard 1990) are applicable. The midfrequency regime, spanning
a substantial range of structural behavior, has several competing solution methodologies. On one
end, discrete deterministic methods such as finite elements offer great flexibility in analyzing prob-
lems that are intractable analytically. These methods can easily grow beyond available computer
resources as frequencies of interest grow higher. At the other end, where resonances are hard to
distinguish, statistical energy methods (Lyon 1975) are appropriate. However, a gap remains in the
midfrequency regime where no method has a clear advantage.
In this chapter, the SIE is introduced as a new super element designed for the frequency domain
dynamic analysis of finite structures (or substructures) that are uniform (or periodic) in the length
direction, say x, and arbitrary in the other directions. The heterogeneity of the cross-section, whether
geometric or material, calls for a discretization technique such as the finite element method. On the
other hand, the uniformity along x calls for some analytical or modal expansion of the solution in
that direction. This is especially desirable in the mid-frequency regime where frequencies of inter-
est are high enough to make the cost of wavelength resolution by discretization substantial, but too
low for asymptotic theories to govern. The SIE approach reconciles these two demands by concep-
tualizing the finite structure of length Nh as an assembly of N cells of length h. For each SIE, a
single cell needs to be modeled by finite elements having for length dimension h. The characteristic
equation resulting from this periodic discrete structure is a quadratic eigenvalue problem. The cell
eigensolutions form the complete basis for the representation of the SIE dynamic stiffness matrix.
Even though, exactly equivalent in its dynamic characterization to a finite element model of the
entire finite structure, the SIE computations only involve cell matrices. In other words, the compu-
tational effort is independent of the length scale N, and thus the name SIE. Furthermore, since the
SIE displacements are expressed in an expansion basis consistent with conventional finite elements,
SIEs can easily be connected to conventional finite elements used to model nonuniform regions.
The SIE approach differs from better known substructuring techniques in two respects:
1. The SIE approach does not truncate the modal basis and therefore maintains the frequency
resolution inherent in the definition of the unit cell. In this, the SIE is akin to an exact
dynamic condensation procedure.
2. The modeling and computational efforts in an SIE formulation are independent of the
length scale. On the other hand, conventional substructuring or super-element techniques
involve at least the meshing of the entire substructure and sometimes manipulation of the
assembled substructure matrices.
6.8.1.2 Theory
The formulation of the SIE, presented below, is based on a series of ideas that have their origin in
the development of the consistent energy transmitting boundary for semi-infinite layered soil strata
by Lysmer (1970), Waas (1972), and Kausel (1974). The concept of an eigenfunction expansion in the
horizontal direction was extended with the “hyperelement” to finite layered continua by Kausel and
Roesset (1977). The “Cloning Algorithm” of Dasgupta (1982) for unbounded homogeneous conti-
nua constitutes a further extension in that the coordinate system inherent to the problem geometry
is no longer assumed separable. Subsequently, the “hyperelement” concept was further generalized
to both semi-infinite and finite truss and beam structural systems by Ettouney et al. (1990). The SIE
concept was introduced by Ettouney et al. (1995a and 1995b). The concept extends the “hyperele-
ment” idea to structural systems that are uniform in one direction but have arbitrary cross-sections.
In what follows, we will summarize the essential equations of the method.
6.8.1.2.1 Equations of Motion

Consider a finite structure, that is uniform in its length direction x. We introduce a finite element
approximation, with uniform discretization h along x of the equation of motion. This translates into
a periodicity along x that we track with a cell counter i. Thus, cell i is bounded by the (i − 1)th and
ith sets of nodes. To each set of nodes i corresponds M degrees of freedom.
The discretized time-harmonic equations of motion for cell i may be written as
⎡ S11 S12 ⎤ ⎧ui −1 ⎫ ⎧ − Pi −1 ⎫

⎢S ⎨ ⎬=⎨ ⎬
⎣ 21 S22 ⎥⎦ ⎩ ui ⎭ ⎩ Pi ⎭
(6.154)
where ui is the displacement solution vector for node set i and S is the cell dynamic stiffness given by
Skl = K kl + jwCkl − w 2 M kl k = 1, 2 , l = 1, 2
(6.155)
Kkl, Ckl, and Mkl are the stiffness, damping, and mass matrices, respectively, ω is the forcing circular
frequency and j = −1 .
The assembly at node set i of contributions from the ith and (i − 1)th cells yields the following
governing stencil
S21ui −1 + (S11 + S22 )ui + S12 ui +1 = 0

(6.156)
It can be shown that the stencil, being a recurrence relation with constant coefficients, admits solu-
tions of the form
ui = Ali = Ae − jkxi , xi = ih
(6.157)
where k is the complex wavenumber.

Substitution of Equation 6.157 into Equation 6.156 yields the following characteristic equation
[ l2 S12 + l(S11 + S22 ) + S12 ] A = 0

(6.158)
6.8.1.2.2 Characteristic Equation: A Quadratic Eigenvalue Problem

Equation 6.158 is an M ⋅ M (M × M) quadratic complex symmetric eigenvalue problem with λ for
eigenvalue. For a given forcing frequency ω, it can be shown that Equation 6.158 admits for solution
an 2M eigensolutions (λs,Vs) representing the full spectrum of waves (including evanescent waves)
that the system admits. It can further be shown that the 2M eigensolutions can be segregated into
two sets of M eigensolutions: the first, denoted by (λs,Vs) for s = 1,…,M, is characterized by |λ s| ≤ 1
and corresponds to waves decaying in the positive x direction (outgoing waves); the second half,
denoted by (l s , Vs ) for s = 1,…,M, is characterized by l s = 1 / ls ≥ 1 and corresponds to waves
decaying in the negative x direction (incoming waves). Furthermore, the eigenvectors of each of the
two sets can be related by
Vs = TVs (6.159)
where T is a simple diagonal transformation matrix that was discussed by Waas (1972) for plane
strain/plane stress problems, and by Kausel (1974) for 3D axisymmetric continua. In the general
case of 2D and 3D structural elements, the expression for T must be generalized to properly account
for rotational degrees of freedom.
To formulate the dynamic stiffness matrix of the finite structure that is represented by the assem-
bly of N identical cells, both of the above sets of eigensolutions will be required (Kausel and Roesset
1977 and Ettouney 1990). This is in contrast to semi-infinite structures that require only one of the
two sets; see, for example, Waas (1972), Kausel (1974), Dasgupta (1982) and Ettouney (1990).
6.8.1.2.3 Dynamic Stiffness Matrix

Consider the finite structure of length L = Nh and consisting of N identical cells. The dynamic stiff-
ness matrix relating forces and displacements at x = 0 and x = L can be derived using the eigensolu-
tions of the characteristic equation. Following Kausel (1977), the displacement vectors u 0 and uN at
node sets i = 0 and i = N, respectively can be expanded as follows
⎧ u0 ⎫ ⎧ Γ0 ⎫
⎨ ⎬ = [ D0 ] ⎨ ⎬ (6.160)
u
⎩ N⎭ ⎩Γ N ⎭

where
[ D0 ] = D0 (V , V , Λ ) (6.161)

with
V = [V1 ,..., VM ] (6.162)

V = [V1 ,..., VM ] (6.163)

Λ = diag[ l1 ,..., l M ] (6.164)

Γi = [Γi1 ... Γim ]T , i = 0 or N (6.165)

where Γ0s and ΓNs are the sth mode modal participation factors for outgoing and incoming waves,
respectively.
Enforcement of equilibrium, by virtue of Equation 6.154, yields for the consistent boundary
force vectors P0 and PN, at x = 0 and x = L, respectively, in terms of the modal participation factors
⎧ P0 ⎫ ⎧ Γ0 ⎫
⎨ ⎬ = [ D1 ] ⎨ ⎬ (6.166)
⎩ PN ⎭ ⎩Γ N ⎭
where
D1 = D1 (V , V , Λ, Skl ) (6.167)

Elimination of the modal participation factors from Equations 6.160 through 6.167 yields the expres-
sion for the dynamic stiffness matrix D of the SIE
⎧ P0 ⎫ ⎧ u0 ⎫ (6.168)
⎨ ⎬ = [ D] ⎨ ⎬
⎩ PN ⎭ ⎩uN ⎭
where
[ D] = [ D1 ][ D0 ]−1 (6.169)

[D] is a complex symmetric matrix of dimension 2M · 2M.
6.8.1.3 Efficiency of SIE Method

In studying the relative efficiency of the SIE methodology as compared to the traditional finite
element technique, we assume that the frequency considerations of the problem require that the
model is subdivided into N identical cells and having M degrees-of-freedom per set of boundary
displacements. A simple one-beam system, as well as a more realistic compound systems will be
studied.
6.8.1.3.1 One-Beam System

Consider the one beam system (Figure 6.54). Table 6.10 shows a comparison of the computational
requirements of both SIE and traditional finite element methodologies.
Note that the formation of the dynamic stiffness matrix of the scale independent element
requires the solution of the quadratic eigenvalue problem, Equation 6.157, which is an oper-
ation of order M3, irrespective of N. Thus, for large N, the reduction in computational cost is
substantial.
1 2 3 N
Number of cells = N
Number of degrees of freedom per cell = M
FIGURE 6.54 One beam system.
TABLE 6.10
Comparison of the Two Methods for
Simple Structural System
Attribute Finite Element SIE
Number of equations (N + 1) M 2M
Band width 2M 2M
Order of computation time NM2 M3
6.8.1.3.2 Compound System: Computational

Now consider a more realistic compound systems. Figure 6.55 shows a planar truss that is built up
from individual beams. The truss has an Ns subspans. All the members of the truss are assumed to
have identical cross-section. Such an assumption is consistent with practical engineering practices.
Table 6.11 shows a comparison of computational attributes of the truss if it is modeled using the
traditional finite element and the SIE methodologies.
The ratio of the computational requirements of the two methods is N3. The computational advan-
tage of the SIE is even greater for more realistic systems than in simple systems.
6.8.1.3.3 Compound System: Modeling

Table 6.11 also shows the relative modeling requirements of both traditional finite element method
and the SIE method for realistic systems. For the structure under consideration, traditional finite
element method requires much more modeling effort than the SIE methodology. Actually, the SIE
method will require the minimum amount of geometric information to represent the structure. Any
added modeling resolution that is needed in the traditional finite element method due to the driving
frequency requirements are not needed by the SIE approach.
It is also of importance to note that the SIE modeling will result in a significant reduction in
memory/storage.
6.8.1.4 Examples
The purpose of this section is to demonstrate some practical applications of the SIE methodology.
The wave mode shapes of a thin walled, box shaped beam will be presented first. The torsional and
bending behavior of a single beam is studied next. Finally, the behavior of a complex framed deck
structure is studied.
6.8.1.4.1 Wave Modes

Understanding the behavior of systems in different frequency ranges is an important aspect of
analysis. The SIE methodology provides an automatic mean to such an understanding. Evaluation
2 4 6 2Ns 2(Ns +1)
1 3 5 2Ns–1 2(Ns +1)
Span = 1 2 2Ns–1 2Ns
FIGURE 6.55 Compound truss system.
TABLE 6.11
Comparison of the Two Methods
Attribute Finite Element SIE
subspans Ns Ns
Additional degrees-of-freedom NM M
per node, n
Number of equations 2NsNM 2NsM
Band width 4NM 4M
Order of computation time M3N3 M3
of the eigenvectors V and V of Equations 6.162 through 6.163 can be used for that purpose. Consider,
for example, the beam with a box cross-section of Figure 6.56. The material of the section is steel. A
unit load is applied symmetrically at one end of the beam. It is of interest to understand the behavior
of the beam when the cross-section is no longer rigid.
Figure 6.57 shows the dominant eigenvector of the cross-section plotted at resonant frequencies
in the range of 60 to 850 Hz. Even at the lowest bending resonant frequency, 60 Hz, the local defor-
mation of the cross-sectional plating is observed. Care should be taken when modeling such a box
cross-section, using conventional beam theory that assumes that the beam cross-section remains
rigid throughout the frequency spectrum.
6.8.1.4.2 Torsion and Bending of Box Beams

Figure 6.58 shows a free standing one-beam system. The beam has a tubular cross-section of 6 × 6 × ¼.
The moment of inertia of the section is 30.30 in4, the cross-sectional area is 5.591 in2, while the shear area
of the section is 3.00 in2. The torsional moment of inertia is 48.5 in2. These specifications have been pub-
lished by The Steel Company of Canada (1966). The thickness of this tubular section is 0.25 in. The radius
of the corners is specified at 0.75 in. The manufacturing tolerance of the corner radius has a limit of 10%.
3/16''
4''
4''
FIGURE 6.56 Box cross-section.
Frequency = 310 Hz Frequency = 370 Hz Frequency = 455 Hz
Frequency = 659 Hz Frequency = 789 Hz Frequency = 819 Hz
FIGURE 6.57 Wave mode shapes of cross-section.

8
Steel block Steel block
A
6 x 6 x 1/4 Output
point
8 8
96
A 2 2
F = 1.0
Section A-A
FIGURE 6.58 Box beam configurations.
(a) (b)
Square corners Rounded corners
FIGURE 6.59 Rounded and square box corners (a) model with square corners, and (b) model with rounded
corners.
To simulate the end joints of the system, heavy steel blocks are welded to the ends of the beam
(Figure 6.58). The system is excited by a vertical eccentric unit force applied at one of the steel
blocks. This application of force on the system is designed to excite both bending and torsional
responses, simultaneously.
Several analytical models were considered for this problem. The SIE methodology was used in
all of those models.
First, the beam was modeled using a Timoshenko et al. (1974) approximation, where the plane
of the tubular cross-section remains rigid during vibrations. The SIE technology was also used to
model the cross-section of the beam as a square tube (Figure 6.59a). Amore refined modeling of the
cross-section, which account for the rounding of the corners, Figure 6.59b was also considered.
Figure 6.60 shows a comparison between the Timoshenko beam modeling and the square cross-
section modeling. Both models simulate the first two bending modes accurately, around ≈100 Hz
and ≈330 Hz, respectively. The Timoshenko model predicts the first torsional mode at ≈430 Hz,
while the more accurate box model predicts the first torsional mode at around 370 Hz A difference
of almost 15%. For higher bending, and torsional modes, the two models predict a progressively
divergent frequencies of response, as expected.
Figure 6.61 shows the effect of accurate modeling of the rounded corners on the first torsional
≈370 Hz and the second bending modes ≈330 Hz of the beam. The rounding of the corners will
result in a slight decrease of bending impedance, while increasing the torsional impedance in a
more pronounced manner.
6.8.1.4.3 Framed Deck-Beam Structure

Consider the structure of Figure 6.62. It is a deck system mounted on four columns. The cross-
section of each of the columns is a wide flange section of HP10 X 42, while the deck system is com-
posed of a deck plating and three supporting stringers, as shown in Figure 6.62. The material of the
1000.0
100.0
Accelerance
10.0
1.0
Beam
Box-7a
0.1
0 200 400 600 800 1000 1200 1400 1800
Hz
FIGURE 6.60 Effects of elements type.
1000.0
100.0
Accelerance
10.0
1.0 Beam
Box-7a
Box-7b
0.1
250 300 350 400 450
Hz
FIGURE 6.61 Effects of corner modeling.
t = 1.5 0.5
0.5
De
8 ck
Ou
12 12 120 C tp
HP olum oin
10 n
x4
t
Deck cross-section 2
12
0 24
FIGURE 6.62 Framed deck system.

(a) (b) (c)

1.0
SIE
de
ck
ele
me
nt
SIE
be tim
am osh
ele enk
Ou
me o tpu
nt tp
oin
t
Traditional finite element model SIE modeling “Equivalent” model
FIGURE 6.63 Different models for the framed deck system (a) traditional model, (b) SIE model, and
(c) equivalent model.
1.E-3
1.E-4
Displacement amplitude
1.E-5
Sie deck model
“Equivalent” model
1.E-6
1.E-7
1.E-8
0 100 200 300 400 500
Hz
FIGURE 6.64 Comparison between SIE and conventional models.
frame is steel, and the loading on the system is assumed to be a unit steady state horizontal force.
The frequency range of interest is 0–500 Hz.
Traditional finite element modeling of this deck-beam system for the frequency range of interest
is shown in Figure 6.63a. The SIE modeling for the same system is shown in Figure 6.63b. An equiv-
alent, simple model of the system is also shown in Figure 6.63c. This equivalent model is based on
simulating the deck by an equivalent Timoshenko beam. Each of the pair of columns at each end of
the frame is simulated by one Timoshenko beam. The modeling and computational efficiencies of
the SIE model when visually compared to the traditional finite element model is clear.
Figure 6.64 shows a comparison between horizontal displacement amplitude of the output point
(Figure 6.62) as computed by the SIE model and the equivalent simple model. As expected, the
equivalent simple model produces good results at lower frequencies. At higher frequency range the
need for the detailed deck modeling become apparent.
6.8.1.5 Closing Remarks

SIE represents a new approach for the frequency domain dynamic analysis of structures. We devel-
oped a theory for the analysis of systems that are uniform along a single coordinate axis but otherwise
arbitrary in geometry and material composition. We showed that this approach, termed the SIE, is an
accurate, efficient, and general approach for the analysis of vibrating structures. This technique extends
the applicability of discrete deterministic finite element based modeling to higher frequencies and is
capable of bridging the gap to the frequency regime where statistical energy methods are applicable.
6.9 CASE STUDIES

6.9.1 General
STRID process is utilized in almost all IHCE projects in varying and innovative ways. To illus-
trate the variety of the utilization of STRID procedures, we offer few case studies in this sec-
tion. Section 6.9.2 illustrated the use of modal testing and 3D finite elements modeling. Similarly,
Sections 6.9.3, 6.9.4, and 6.9.5 showcase the use of different analysis techniques in support of NDT
procedures. Section 6.9.6 illustrates combining signal processing techniques, NDT, and STRID
methods. Section 6.9.7 shows the essential importance of STRID methods (modal identification)
for optimally placing sensors during an experiment. Finally, Section 6.9.8 shows the use of modal
testing methods in some IHCE projects.
6.9.2 Generating Baseline Models

An example of monitoring and generation of a baseline model for Maysville Bridge in Kentucky was
given by Harik et al. (2002). It is 2100-ft-long cable stayed bridge, with two anchor spans, flaking
span, and a center span (1050 ft). A two-plane semi-harped system was utilized in the bridge. Stays
were placed at 50′ intervals along each edge of the deck consisting of four sets with 20 cables in
each set covered by coextruded high-density polyethylene pipes with white outer layers. The outer
layers had small spiral beads around the pipe to break up airflow in cases of light rain and wind thus
preventing cable galloping. Soft neoprene collars were employed to connect damping cables to the
stay cables. A field ambient vibration test was conducted by the authors using two loaded trucks and
data was collected from about 80 locations at 1000 Hz rate. Locations were chosen based on a pre-
liminary FEA model. Two complementary SI methods known as Peak Picking (PP) in the frequency
domain and the stochastic subspace identification (SSI) method in time domain were used. A good
agreement was found in identified frequencies by both methods. Then a 3D finite element model
was developed, calibrated, and analyzed. Test data and FEM results were in good agreement.
Multireference modal testing was conducted on a five-span r/c deck on steel-stringer bridge in
Dayton, Ohio by Kangas et al. (2002). The deck was replaced with an interlocking system of FRP
deck panels. Modal testing was conducted before and after the rehabilitation, and changes in the
bridge characteristics were analyzed using modal flexibility. The changes were attributed the loss
of composite action.
6.9.3 Scour STRID

STRID methods were also used in bridge scour problems. Over 100,000 bridges in National Bridge
Inventory have unknown foundation conditions. Olson (2002) summarized research on surface and
borehole methods, including stress wave, vibration, magnetic, and radar NDE methods. These stud-
ies were part of NCHRP 21–5 and 21–5(2) projects.
Parallel Seismic Method: NCHRP studies found that borehole-based PS to be more accurate and
most applicable NDE method for determination of the depth of unknown bridge foundations for
bridge scour safety evaluation purposes. Thus, it was suggested for any project that at least one PS
test be performed with other methods so that the reliability of other methods can be validated. This
method requires impacting the exposed portion of the substructure and monitoring the response of
the foundation and surrounding soil with a receiver in a cased borehole drilled next to the founda-
tion. As the wave energy is monitored by the receiver at depths parallel to and below the bottom of
the foundation, it becomes typically weaker and slower below the foundation bottom. The change in
the received signals with depth indicates the foundation depth.
The method requires the borehole to be drilled next to the foundation to be tested, preferably
within 5 ft or less, cased 2 in ID PVC casing, and extend at least 15-ft-deeper than the minimum
required foundation depth.
Surface Ultraseismic (US) method: This was developed to overcome difficulties encountered by
sonic echo/impulse response method and the bending wave tests on noncolumnar and complex
columnar bridge substructures. US is a broad application in that the initial arrivals of both com-
pression and bending waves and their subsequent reflections can be analyzed to predict unknown
foundation depth.
US vertical profiling test is used for one-dimensional imaging of the foundation depth by track-
ing the upgoing and downgoing events. Bridge columns or abutment is hit from the top or bottom
(both vertically and horizontally) and the output wave motion is recorded at regular intervals down
the substructure element. Typically, three component record of the wave field is taken to analyze all
types of ensuing wave motion.
For 2D imaging of foundation depth, horizontal profiling (HP) US test is used where there is flat,
horizontal access for testing, such as top of an accessible pier or abutment. The reflection echoes
from the bottom are analyzed to compute the foundation depth. The source and receiver(s) are
located horizontally along the top of accessible substructure, or any accessible face along the side of
the substructure element, and a full survey is taken along the top of the element.
6.9.4 Foundation Identification

There are different applications for validating information regarding foundations, such as slabs,
piles, and so on. These applications are examples of using Condition assessment efforts in support of
STRID. For example, as part of the “Re-use of Foundation for Urban Sites” program, Niederleithinger
and Taffe (2006) reported on a planned test center, with a slab of varying thicknesses and reinforce-
ments and 10 bored piles of varying length, was built as a test bed (see Figures 6.65 and 6.66) for
experimental and validation of methodologies. One of the tasks was to evaluate thickness of slabs
using NDT methods. Some of the conclusions show that
1. Thicknesses of r/c slabs are possible up to 125 cm with about 3% uncertainty. Up to 75 cm

discrimination between lower reinforcement and back wall seems possible
Upper reinforcement
Strip
foundation
Lower reinforcement
Pile foundation
FIGURE 6.65 Illustration of the RuFUS test slab. (Reprinted from ASNT Publication.)
FIGURE 6.66 Detail of the reinforcement and geometrical features of the slab. (Reprinted from ASNT
Publication.)
(a) (b)
FIGURE 6.67 Instrumented downstream truss members: (a) schematic view of sensors, and (b) photograph
of actual sensors placed on bridge.
2. Localization of elements as piles and strips beneath a slab are possible, at least if there are
no horizontal sealing layers
3. Radar (500 mHz to 1.5 GHz) was able to locate upper reinforcement, but not the lower one
or the back wall
4. The IE results have been unclear
Commonly used low strain integrity testing methods (Pile Integrity Testing, hammer impulse
testing, etc.) were also evaluated to find that pile lengths that were under or overestimated in most
cases (accuracy is less than 10%).
6.9.5 Condition Assessment for Existing Bridges

Truck Loads: Hag-Elsafi et al. (2004) discussed instrumentation and deck pour monitoring of
Court Street Bridge in Tioga County, New York. The bridge is a six-span continuous steel structure,
about 338-m-long and 14.45-m-wide, consisting of stringers, floor beams, two trusses (upstream
and downstream), and a light weight concrete deck. The deck was built composite with the string-
ers and the top chords of the trusses. The motive for the instrumentation and monitoring (Figure
6.67) was to determine service load axial forces and secondary moments in the truss main members
during deck pours. The authors concluded that the members’ actual service dead load axial forces
and moments were overestimated by about 20% in the design. Regarding moments, the authors also
concluded that service dead load moments were within 20% of those used in the design. The differ-
ences between actual and theoretical axial forces and moments for service dead load (Figure 6.68)
were attributed to the manner in which the monitoring data was corrected for temperature effects
and the possible presence of construction loads on the deck during the pours monitoring.
Temperature Distribution: Mondal and DeWolf (2004) monitored a segmental concrete bridge for
Connecticut DOT to learn how different bridges behave, performance evaluation, renovation guid-
ance, and show the value of long-term monitoring. The bridge is a large, multi-span, continuous,
post-tensioned box-girder bridge, shows the cross-section of the bridge. It is a 11-span, 250-ft-long,
and 178–275- ft-wide bridge. Figure 6.69 shows the cross-section of the bridge. The designers rec-
ommended monitoring temperatures over time. The bridge experienced significant cracking and
monitoring has offered an opportunity to explore the influence of the temperature variations on the
performance (see Figure 6.70). About 16 thermocouples were used with 14 in one cross-section.
Other two measured the air temperature inside the box and the temperature inside the instrument
cabinet.
Axial force
(a)
5000
4000
3000
2000
y 1.2389x + 393.92
1000
FE analysis (kN)
–5000 –4000 –3000 –2000 –1000 1000 2000 3000 4000 5000
–1000
–2000
–3000
–4000
–5000
Test (kN)
(b)
Secondary moment
175
150
125
y = 1.227x – 11.709
100
FE analysis
75
50
25
0
0 25 50 75 100 125 150 175
–25
Test
FIGURE 6.68 Service dead load axial forces and moments: FE analysis and load test (a) axial force com-
parisons, and (b) secondary moment comparisons.
83'–9'
8'
2 3 10 9
1 8
4 11'–8' 3'
7 8 11
13 8' 5 12
6 14
36'–0'
FIGURE 6.69 Bridge cross-section and locations of thermocouples. (Reprinted from ASNT Publication.)
14
Vertical
12 temperature
Temperature difference (°F)
difference at
10 centerline
Horizontal
8 temperature
difference at
6 top
Horizontal
4 temperature
difference at
2 bottom
0
1 6 11 16 21 26 31
Date
FIGURE 6.70 Vertical and horizontal temperature differences during July 2002. (Reprinted from ASNT
Publication.)
Among the conclusions of that landmark research were (1) vertical variations of temperatures
were observed to be higher than horizontal temperature variations, (2) temperature gradient profile
is similar to AASHTO specifications, even though the magnitudes were different (see Figure 6.71)
and (3) the magnitude of stress-induced temperature gradients were low.
6.9.6 Signal Processing

Signal processing techniques can also be used, in conjunction with other NDT techniques, to iden-
tify structural properties. For example, Algernon and Wiggenhauser (2006) discussed some of the
suggested improvements to data analysis during impact-echo method (see Chapter 7) for nonde-
structive testing of concrete structures. Impact-echo signals from concrete structures can be very
noisy due to geometry effects and may lead to misinterpretation of signals. Hence, autocorrela-
tion methods and Hilbert-Huang transform methods were used to analyze the impact-echo data to
improve the accuracy of the analysis.
Using autocorrelation helps to amplify the effect of multiple reflections and minimizes the effects
of white noise. Cross-correlation is also used to eliminate variations in the excitation pulse, and is
then treated as the original pulse. This was found to be useful in increasing the intensity of the
thickness from the enhanced frequency spectrum.
Use of Hilbert-Huang transform was also explored as it has ability to localize an event in time
and frequency making it possible to identify short transient signals within longer time sweeps. This
was found to be very useful to analyze data from a highway bridge during heavy traffic and obtain a
0 5 10 15 20 25 30 35
0 15 33
11
Depth from bottom surface to top surface (in)

3
Positive
20 3.6 temperature
gradient for
40 0 segmental
bridge under
60 consideration
2.8 AASHTO
80
positive
temperature
100
gradient
120
2.8
0
140 5
FIGURE 6.71 Observed thermal gradient. (Reprinted from ASNT Publication.)
(a) 20 (b) 0.6

0.5
Amplitude / V
10
0.4
Amplitude
0 0.3
0.2
–10
0.1
–20 0.0
0 2 4 6 8 10 12 0 5 10 15 20 25 30
Time (ms) Frequency (Hz)
(c) 1.0
0.8
0.6
Amplitude
0.4
0.2
0.0
0 5 10 15 20 25 30
Frequency (Hz)
FIGURE 6.72 Comparison of different signal processing schemes (a) time signal, (b) Fourier spectrum, and
(c) Hilber-Huang marginal spectrum. (Reprinted from ASNT Publication.)
clear indication of the thickness (see Figure 6.72). Due to its iterative nature, HHT is highly efficient
but should be carefully applied as a theoretical performance evaluation is only partly possible. With
instantaneous frequency it becomes possible to identify short transient signals within the data as
well as frequency fluctuations, which remain hidden for other methods.
6.9.7 Optimal Sensor Locations (OSL)

Papadimitiou et al. (2000) combined an optimum sensor location methodology and a structural
identification algorithm. Their approach is based on minimizing the uncertainties involved in
experiments. Their method addressed only a single hazard (loading condition). They showed that
the optimal location of sensors is dependent on the number of structural modes that are considered
in the experiment, as well as the number of available sensors.
6.9.8 STRID as a Component in SHM

Examples of the use of structural identification methods as an integral part of an SHM as applied
to bridges was reported by Stubbs et al. (1999), Sikorsky et al. (1999), and Bolton et al. (1999
and 2000). Some of the goals of their experiments/structural (modal) identification were (1) assess
instrumentation techniques with minimal interference with the operations of the bridges, (2) assess
testing techniques, (3) apply MI method (MEScope software, version 2.0) on measurements to esti-
mate modal responses (frequencies, damping, and mode shapes), and (4) compare with numerical
solutions such as finite elements.
The authors stated clearly that the whole effort is based on global NDE paradigm, that is, identify
damage from properties of the whole structure. To achieve this, they tested two bridges: a reinforced
concrete and a steel bridge. We summarize their experiments, point to some of the highlights of
their results and show the value of such experiments, with some cautionary notes.
Used two different field tests to analyze the RC 2-span bridge: the differences between the two
tests were mainly on how the data were collected and handled. The first field test collected the data
in two sets: (1) vertical bending and torsional behavior, and (2) lateral bending behavior. A single
axis accelerometers were used for the first field test. The analytical MI were performed separately.
MI of the results of the second field test were performed using unified sets of measurements. This
was accomplished by using triaxial accelerometers to measure combined behavior of the bridge.
Table 6.12 shows the MI results as reported by the authors.
The results of those tests are very illuminating in that they demonstrate the many potential
benefits of using structural identification methods in SHM. For example, note that the estimated
modal damping for different modes vary from 3.5% to 1.01%. This information would be invalu-
able to bridge analyst and designer, since the common practice is to assume a constant modal
damping when analyzing/designing a bridge. If the assumed constant damping is in the range of
3.5%, then the resulting analysis/design will be not safe. On the other hand, if the assume constant
modal damping in analysis/design is in the range of 1.01%, then the resulting analysis/design will
TABLE 6.12
Modal Identification Results
Field Test I Field Test II
Mode # Frequency (Hz) Damping (%) Frequency (Hz) Damping (%)
1 3.09 3.5 3.35 3.51
2 3.21 3.1 4.82 3.30
3 4.42 3.3 6.73 2.34
4 6.78 2.0 8.75 2.57
5 8.32 2.7 10.63 2.10
6 10.58 1.9 12.93 3.65
7 11.64 4.0 14.62 2.33
8 14.45 1.7 20.55 1.69
9 20.72 1.3 20.93 1.30
10 23.35 1.1 21.71 1.58
11 24.42 1.8 24.42 1.87
12 Not reported Not reported 27.04 1.95
13 Not reported Not reported 29.01 1.01
TABLE 6.13
Modal Properties of Steel Bridge
Mode Frequency (Hz) Damping (%)
1 2.21 0.520
2 3.08 0.439
3 5.01 0.407
4 6.18 1.548
5 7.57 0.487
6 8.94 0.403
7 9.86 0.262
8 10.31 0.480
9 11.35 0.279
10 13.01 0.233
11 13.89 0.246
not be economical. Clearly, using a combination of testing/structural identification can result in a

safer yet more economical structure.
Another important observation that was reported by the authors of this experiment is that it
shows the conventional use of FE methods can adequately simulate the estimated mode shapes.
Since the maximum reported acceleration was about 0.050 g, with most of the measured responses
in the range of ±0.025 g, the bridge responded in an elastic fashion. Thus, the conventional assump-
tions of elastic responses while using FE modeling techniques is validated. The high frequency
range of the experiment (≅30 Hz.) would make such results fairly adequate to analyze the response
of the bridge to earthquake motions.
The value of the information gained from such an experiment can easily be determined using
a cost/benefit approach as described in several parts of this volume. For example, we can compute
the benefit from the experiment by computing the cost savings from using a more realistic damping
than either too high or too low damping levels. Computing the cost of the experiment is straightfor-
ward. Thus the value of the experiment is the computed benefit to cost ratio.
We should also point out to the fact that the results of such a successful experiment should
not be extrapolated or interpreted beyond its intended goals. For example, since the experiment
is based on measuring motion (accelerations, velocities, or displacements), using the results to
interpret deformation measures (such as strains, or stresses) would risk leading to erroneous
results.
The author also reported a similar MI experiment on a single span-steel truss bridge. The results
of the MI scheme are shown in Table 6.13. The value of this experiment is similar to the first one. It
is of interest to note that the modal damping is more evenly distributed than that of the reinforced
concrete bridge, and the damping of mode number 4 (1.548%).
6.10 LIFE CYCLE ANALYSIS AND STRID

Life cycle analysis requires large amount of information about the system on hand. As more infor-
mation become available, the accuracy of LCA increase. STRID techniques can be of help in the
process of LCA. Recall that a generic model of estimating life cycle cost (LCC) is
LCC = ∫ ∫ p( H ) c( H ) dH dt (6.170)

The probability of hazard level H is p(H) and the cost of the hazard is c(H). The double integrals in
Equation 6.170 are over the hazard space and the life span of the structure. The cost of hazard is a
function of system properties as they respond to h. These properties include system stiffness K, sys-
tem mass M, system damping β, system strength R, and system response U. Note that all of these sys-
tem properties and responses include all system components that include the structure, the supporting
soil, and the surrounding water, if present. Also, note that K, M, β, R, and U are general functions that
can be scalar, matrices, or analytic functions, as appropriate. As such, we have the relationship
LCC = f ( K , M , b, R, U ,..., H ,..., xi ) (6.171)

The variables xi represent the many other nonstructural related factors that might affect LCC, such
as consequences of failure, or management costs. Some expressions of the function f( ) are discussed
in Chapter 10 of Ettouney and Alampalli (2012).
What concerns us here is the implied relationships in Equation 6.171. Such relationships imply
that the accuracy of LCC depends directly on the accurate estimations of K, M, β, R, and/or U. Since
the main goal of STRID is to improve the accuracy of some or all of those parameters, we can con-
clude that STRID can have a major role in the accurate estimation of LCC. To gain an understanding
of the nature of the relationship, let us consider a discrete form of Equation 6.170 as
LCC = ∑ pi ci (6.172)
i
Note that Equation 6.172 is for only a unit period of time for simplicity. The sum in Equation 6.172
is over the hazard space. The cost of the ith hazard is ci. This cost is a function of the damage due to
the ith hazard, Di such that
ci = Di ci (6.173)

The cost per unit damage is ci, assuming a linear relation between damage and cost. We can further
assume the structural response Ui to Hi can be evaluated using
KUi = Hi (6.174)

The representative stiffness of the system is K. For simplicity, we assume a linear structural perfor-
mance. For nonlinear performance, the logic of this development will not be affected. The final step
in the development is to relate the damage Di to the response Ui such that
Di ∼ Ui (6.175)

Thus
Di ∼ K −1 Hi (6.176)

or
Di ∼ F Hi (6.177)

The flexibility of the system is F.

Equation 6.172 becomes
LCC = ∑ pi F Hi ci (6.178)
i
or
LCC = F A (6.179)
with A being a constant such that

A = ∑ pi Hi ci (6.180)
i
Equation 6.179 proves that there is a direct relationship between the flexibility, or stiffness, measures
of the system, and the estimate of its LCC. In this case, the relationship is linear. For more complex
systems, the relationship can be different; however, the direct correlation will still be there.
Let us now consider the effects of accurate estimation of flexibility on the estimation of LCC.
A conventional LCC is estimated using conventional (not based on experiment) flexibility, F0 such
that
LCC0 = F0 A
(6.181)
Let us assume that experimentally base flexibility (estimated using some form of STRID method)
is F1. The corresponding LCC is
LCC1 = F1 A
(6.182)
It is reasonable to assume that LCC1 is more accurate than LCC0, since it is based on experimental
results. The dimensionless improvements is
LCC1 − LCC0 F1 − F0
Δ LCC = =
LCC0 F0
(6.183)
We reach an interesting conclusion. The dimensionless improvement of LCC due to experimen-

tally based STRID is proportional to the dimensionless improvement of flexibility estimates. In case
of stiffness, the proportionality is an inverse one.
The above shows an important STRID use: the accurate estimation of LCC. As such, STRID
should be considered as an essential infrastructure management tool.
6.11 COST–BENEFIT ANALYSIS OF STRID

The effort of STRID will have costs as well as benefits to asset manager. The costs, as usual, are
fairly easy to estimate. They include the following:
1. Cost of the experiment (sensors, instrumentation, etc.)

2. Cost of management, including costs of decision making as well as owner’s costs
3. Cost of labor
4. Cost of analysis
The first three are the usual costs of SHM. The cost of analysis includes costs of performing differ-
ent STRID methods, data analysis, validation, reporting, and so on.
Estimating benefits of STRID is a difficult task. Usually, the benefits are estimated on a quali-
tative manner. We try for the remainder of this section to present few examples of estimating the
benefits of STRID in a quantitative manner. Thus the value of STRID becomes easier to establish
for decision makers. Our general approach will be to estimate the benefits as the cost savings. For
example, if the costs of certain situation without the STRID effort is estimated to be C0 and the costs
of the same situation, given the information of STRID effort, is estimated to be C1, then the benefit
of the STRID effort is
BSTRID = C0 − C1
(6.184)
Given the above, let us consider some cases, as follows:

Analytical Model Updates: Let us assume that a large analytical model of a bridge structures is
available. The model is formed using analytical estimates and best guesses engineering judgment.
Let us assume further that a major retrofit is planned to mitigate a potential earthquake hazard. The
seismic retrofit is based on a modified model from the original model, again based on best guesses
and engineering judgments. Let us define the differences between the estimated analytical state
and the actual state as ∆S 0. In an actual earthquake event, the cost of the damage C0 is going to be
proportional to ∆S 0 such that
C0 = A ΔS0 (6.185)
with A as a constant that relates the cost of damage to the errors in the model that was used in
design. If an STRID effort is done to update the design-based analytical model, then the differences
between the estimated STRID-based analytical state and the actual state will be ∆S1. In an actual
earthquake event, the cost of the damage C1 is going to be proportional to ∆S1 such that
C1 = A ΔS1 (6.186)
Since it is reasonable to assume that
ΔS0 > ΔS1 (6.187)
The benefit of STRID will be real (positive) in the sense
BSTRID = A(ΔS0 − ΔS1 ) (6.188)
A detailed cost estimate analysis of Equations 6.185 and 6.186 can yield a quantitative estimate of
the benefit of STRID effort. Figure 6.73 shows conceptual benefits of STRID for model updating
efforts.
Updated model
based on
engineering
judgement The potential
and best guesses Resulting difference in cost of
Initial model
earthquake damages of the two
for the structure
Updated model damage scenarios is the
based on STRID Earthquake benefit of STRID
principles event effort
FIGURE 6.73 Benefits of STRID in model updating.

REFERENCES
Alampalli, S., (1998a) Effects of Testing, Analysis, Damage, and Environment on Modal Parameters. Proc.
Modal Analysis & Testing. NATO-Advanced Study Institute, Sesimbra, Portugal.
Alampalli, S., (1998b) “Remote Bridge-Monitoring Systems: New York’s Experience.” Structural Materials
technology: An NDT Conference, San Antonio, TX, SPIE Volume 3400.
Alampalli, S. and Cioara, T., (2000) “Selective Random Decrement Techniques for Bridge Vibration Data,”
Structural Materials Technology: An NDT Conference, Atlantic City, NJ.
Alampalli, S. and Ettouney, M. (Eds.), (2003) Proceedings of the Workshop on Engineering Structural Health.
New York State Department of Transportation, New York.
Alampalli, S. and Fu, G., (1994) “Remote bridge monitoring systems for bridge condition.” Clinet
Report 70, Engineering Research and Development Bureau, NYS Department of Transportation,
Albany, NY.
Algernon, D. and Wiggenhauser, H., (2006) “Impact-Echo Signal Processing,” NDE Conference on Civil
Engineering, ASNT, St. Louis, MO.
Andry, A. N., Shapiro, E. Y., and Chung, J. C., (1983) “Eigenstructure Assignment for Linear Systems,” IEEE
Transactions on Aerospace und Electronic Systems, Vol. AES-19, No. 5.
Aref, A. and Alampalli, S., (2001) “Vibration Characteristics of a Fiber Reinforced Polymer Bridge
Superstructure.” Journal of Composite Structures, Elsevier Science, 52(3–4).
Asmussen, J.C., (1997) “Modal analysis based on the random decrement technique. application to civil engi-
neering structures.” Ph.D. Thesis, Aalborg University, Department of Building Technology and Structural
Engineering, Denmark.
Asmussen, J.C., Brincker, R., and Ibrahim, S.R., (1999) Modal Analysis based on the Random Decrement
Transform. Proc. Modal Analysis & Testing, NATO-Advanced Study Institute, Sesimbra Portugal.
Beardsley, P., Hemez, F., and Doebling, W., (2000) “Updating Nonlinear Finite Element Models in the
Time Domain,” Proceedings of 2nd International Workshop on Structural Health Monitoring, Stanford
University, Stanford, CA.
Bernhard, R. and Huff, J., (1999) “Structural-acoustic design at high frequency using the energy finite element
method”, Journal of Vibration and Acoustics, Vol 221, Academic Press.
Bolton, R., Stubbs, N., Park, S., Choi, S., and Sikorsky, C., (1999) “Measuring Bridge Modal Parameters for
Use in Damage Detection and Performance Algorithms,” Proceedings 17th International Modal Analysis
Confrence & Exposition, Kissimmee, FL.
Bolton, R., Stubbs, N., Park, S., Choi, S., and Sikorsky, C., (2000) “Field Measurements of Modal Parameters
for Non-Destructive Damage Detection Algorithms,” Proceedings of 2nd International Workshop on
Brinker, R. and Moller, N., (2006) “Operational Modal Analysis – a New Technique to Explore,” Sound and
Vibration, Volume 40, No 6, pp. 5–6.
Cheung, M. S. and Naumoski, N., (2002) “The First Smart Long-Span Bridge in Canada – Health Monitoring
of the Confederation Bridge.’ Structural Health Monitoring Workshop, ISIS Canada Research Network,
Manitoba, Canada.
Cho, P., (1993) “Energy flow analysis of coupled structures,” Ph.D. Dissertation, Mechanical Engineering
Department, Purdue University.
Chong, K. P., Carino, N. J., and Washer, G. A., (2001) “Health Monitoring of Civil Infrastructures.” Proceedings
of the SPIE Health Monitoring and Management of Civil Infrastructure Systems, Plenary Paper, Newport
Beach, CA.
Cioara, T. Gh., (1988) “On a Curve-Fitting Algorithm for Modal Parameter Estimation.” Sixth International
Modal Analysis Conference, Kissimmee, FL.
Cioara, T. Gh., (1999) Vibration Monitoring of Mechanical System – Fault Diagnosis and Vibration Level
Reduction. USC-ME-LAMSS-99–101, University of South Carolina.
Cole, H., (1973) On-Line Failure Detection and Damping Measurement of Aerospace Structures by Random
Decrement. NASA CR-2205, Washington, DC.
Dasgupta, G., (1982) “A finite element formulation for unbounded homogeneous continua,” Trans. ASME
Journal of Applied Mechanics, 49, 136–140.
Demuth, H., Beale, M., and Hogan, M., (2007) Neural Network Toolbox 5: User’s Guide, The Mathworks, Inc.,
Natic, MA.
Doebling, S., (1996) “Minimum-rank optimal update of elemental stiffness parameters for structural damage
identification,” AIAA Journal. 34(12), 949–957.
Doebling, S., Farrar, C., Prime, M., and Shevitz, D., (1996) “Damage Identification and Health Monitoring
of Structural and Mechanical Systems from Changes in Their Vibration Characteristics: A Literature
Review,” Los Alamos National Laboratory Report, LA- 13070-MS, Los Alamos, NM.
Ettouney, M., (1978) “Compliance Function of Footing in Two Layered Medium,” ASCE Geotechnical
Engineering Division Specialty Conference, Pasadena, CA.
Ettouney, M., (1979) “Empirical Expressions for Frequency Dependent Soil Springs for Machine Foundation
Design,” Proceedings, the 1979 Annual Convention, American Concrete Institute, Milwaukee, Wisconsin.
Ettouney, M., Benaroya, H., and Wright, J., (1990) “Wave propagation in hyper-structures,” in Space 90:
Engineering, Construction and Operations in Space, Editors: Johnson, S. J. and Wetzel, J. P. ASCE,
New York.
Ettouney, M. and Daddazio, J., (1978) “Hydrodynamic Interaction Between the Walls and the Floor of
Cylindrical Fluid Containers,” Proceedings, Conference or Structural Analysis, Design and Construction
in Nuclear Power Plants, Port Allegre, Brazil.
Ettouney, M., Daddazio, R., and Abboud, N., (1995a) “Scale Independent Elements for Dynamic Analysis
of Vibrarting Systems,” Proceedings Third International Conference on Mathematical and Numerical
Aspects of Wave Propagation Phenomena, Mandelieu - La Napoule, France.
Ettouney, M., Daddazio, R., and Abboud, N., (1995b) “Some Practical Applications Of the Use of Scale
Independent Elements For Dynamic Analysis of Vibrating Systems,” Computers and Structures, Vol 65,
No. 3, Pergamon Press.
Ettouney, M., Daddazio, R., and DiMaggio, F., (1990) “Wet Modes of Submerged Structures,” Technical report
prepared for David Taylor Research Center, Bethesda, MD. Contract No. N00167‑86‑c‑0100, Weidlinger
Assoc., Report No. WA 9008, NY.
Farrar, C. R. and Doebling, S. W., (1998) Damage Detection - Field Application to a Large Structure. Proc.
Modal Analysis & Testing, NATO-Advanced Study Institute, Sesimbra, Portugal.
FHWA., (1995) “Recording and Coding Guide for the Structure Inventory and Appraisal of the Nation’s
Bridges,” Report No. FHWA-PD-96–001, Office of Engineering, Bridge Division, Washington, DC.
FHWA., (2004) National Bridge Inspection Standards (NBIS), Federal Regulations 69 FR74436, Federal
Highway Administration, USA.
Friswell, M. and Mottershead, J., (1995) Finite Element Model Updating in Structural Dynamics, Kluwer
Academic Publishers, Norwell, MA.
Fuchs, P. A., Washer, G. A., Chase, S. B., and Moore, M., (2004) “Applications of Laser Based Instrumentation
for Highway Bridges.” Journal of Bridge Engineering, ASCE, 9(6), 541–549.
Girard, A. and Defosse, H., (1990) “Frequency smoothing, matrix assembly and structural paths: a new approach
for structural dynamics up to high frequencies,” Journal of Sound and Vibration, 137(1), 53–68.
Hag-Elsafi, O., Kunin, J., and Alampalli, S., (2004) “Monitoring of Court Street Bridge for Deck Load Stresses,”
Proceedings, NDE Conference on Civil Engineering, ASNT, Buffalo, NY.
Harik, I., Zatar, W., Herd, D., Goodpaster, S., Hu, J., Givan, G., and Crane, R., (2002) “NDE / NDT of Bridges
in Kentucky,” Proceedings, NDE Conference on Civil Engineering, ASNT, Cincinnati, OH.
Hasselman, T., Anderson, M., and Wenshui, G., (1998) “Principal Component Analysis for Nonlinear Model
Correlation, Updating and Uncertainty Evaluation,” Proceedings, 16th IMAC, Santa Barbara, CA.
He, J. and Fu, Z., (2001) Modal Analysis, Butterworth Heinemann, Boston.
Hemez, F. M., (1993) “Theoretical and Experimental Correlation between Finite Element Models and Modal
Tests in the Context of Large Flexible Space Structures,” Ph.D. Dissertation, University of Colorado,
Department of Aerospace Engineering Sciences.
Hemez, F. and Doebling, S., (2001), “Review and Assessment of Model Updating for Non-Linear, Transient
Dynamics,” Mechanical Systems and Signal Processing, Vol 15, No. 1, Academic Press.
Humar, J., Xu, H., and Bagchi, A., (2004) “Application of Artificial Neural Networks in the Detection of
Structural Damage,” Proceedings of 2nd International Workshop on Structural Health Monitoring of
Ibrahim, S. R., (1977) Random Decrement Techniques for Modal Identification of Structures. Journal of
Spacecraft and Rockets, 14(11), 696–700.
Ibrahim, S., (1998) “Fundamentals of Time Domain Modal Identification,” in Modal Analysis and Testing,
Editors: Silva, J., and Maya, M. Nato Science Series, Kulwer Academic Publishers, The Netherlands.
Jeng, D., Bateni, S., and Lockett, E., (2005) “Neural Network assessment for scour depth around bridge
piers,” University of Sydney, Department of Civil Engineering Research Report No. R855, Sydney,
Australia.
Kangas, S., Helmicki, A., Hunt, V., Shahrooz, B., Brown, D., and Lenett, M., (2002) “Modal Test-Based
Condition Assessment,” Proceedings, NDE Conference on Civil Engineering, ASNT, Cincinnati, OH.
Katkhuda, H., Flores, R., and Haldar, A., (2005) Health Assessment at local Level with Unknown Input
Excitation. Journal of Structural Engineering, ASCE, 131(6), 956–965.
Kausel, E., (1974) “Forced vibrations of circular foundations on layered media,” Research Report R74–11,
Soils Publication No. 336, Structures Publication No. 384, Massachusetts Institute of Technology,
Cambridge, MA.
Kausel, E. and Roesset, J. M., (1977) “Semi analytic hyper element for layered strata,” Journal of the
Engineering Mechanical Division, ASCE, 103(EM4), 569–588.
Keane, A. and Price, W., (1969) Statistical Energy Analysis, Cambridge University Press.
Koh, C. and Htun, S., (2004) “Adaptive Search Genetic Algorithm for Structural System Identification,”
Proceedings of 2nd International Workshop on Structural Health Monitoring of Innovative Civil
Koizumi, T., Tsujiuchi, N., Tanaka, H., Okubo, M., and Shinomiya, M., (2002) “Prediction of the Vibration
In Buildings Using Statistical Energy Analysis,” Proceedings of the 20th International Modal Analysis
Conference (IMAC), Los Angeles, California.
Laurens, S., Shartai, Z. M., Kacimi, S., Balayssac, J. P., and Arliguie, G., (2006) “Prediction of Concrete
Electrical Resistivity using Artificial Neural Networks,” NDE Conference on Civil Engineering, ASNT,
St. Louis, MO.
Leung, A. Y. T., 1(988) “A simple dynamic substructure method,” Earthquake Engineering and Structural
Dynamics, 16, 827–837.
Lim, T. W., (1995) “Structural Damage Detection Using Constrained Eigenstructure Assignment,” Journal of
Guidance, Control and Dynamics, 18(3), 411–418.
Lim, T. W. and Kashangaki, T., (1994) “Structural Damage Detection of Space Truss Structures Using Best
Achievable Eigenvectors,” AIAA Journal, 32(5), 2310–2317.
López-Díez, J., Torrealba, M., Güemes, A., and Cuerno, C., (2005) “Application of Statistical Energy Analysis
for Damage Detection in Spacecraft Structures,” Engineering Materials, Vol 293–294, Trans Tech
Publications, Switzerland.
Lyon, R., (1975) Statistical Energy Analysis of Dynamical Systems, M.I.T. Press, Cambridge, MA.
Lyon, R. and Maidanik, R., (1962) “Power Flow between Linearly Coupled Oscillators,” Journal of the
Acoustical Society of America, 34, 623–639.
Lysmer, J. (1970) “Lumped Mass Method for Rayleigh Waves,” Bulletin of the Seismological Society of
America, 60, 89–104.
Mondal, P. and DeWolf, J. T., (2004) “Long-Term Monitoring of Temperatures in a Segmental Concrete
Box-Girder Bridge in Connecticut,” Proceedings, NDE Conference on Civil Engineering, ASNT,
Buffalo, NY.
Moore, M., Phares, B., Graybeal, B., Rolander, D., and Washer, G., (2001) Reliability of Visual Inspection
for Highway Bridges, Volume I: Final Report. Report No. FHWA-RD-01–020. Federal Highway
Administration, Washington, DC.
Moore, M., Phares, B., and Washer, G., (2004) Guidelines for Ultrasonic Inspection of Hanger Pins. Federal
Highway Administration, Report No. FHWA-HRT-04–042, Washington, DC.
Niederleithinger, E. and Taffe, A., (2006) “Improvement and Validation of Foundation NDT Methods,” NDE
Conference on Civil Engineering, ASNT, St. Louis, MO.
Olson, L., (2002) “Determination of Unknown Depths of Bridge Foundations for Scour Safety Investigation,”
Proceedings, NDE Conference on Civil Engineering, ASNT, Cincinnati, OH.
Padur, D. S., Wang, X., Turer, A., Swanson, J., Helmicki, A., and Hunt, V., (2002) “Non-Destructive Evaluation/
Testing Methods – 3D Finite Element Modeling of Bridges,” Proceedings, NDE Conference on Civil
Engineering, ASNT, Cincinnati, OH.
Papadimitiou, C., Katafygiotis, L., and Yuen, K., (2000) “Optimal Instrumentation Strategies for Structural
Health Monitoring Applications,” Proceedings of 2nd International Workshop on Structural Health
Phares, B., Graybeal, B., Rolander, D., Moore, M., and Washer, G., (2001) Reliability and Accuracy of Routine
Inspection of Highway Bridges. Transportation Research Record, Journal of Transportation Research
Board, National Research Council, Issue Number 1749, 82–92.
Phares, B., Washer, G., Rolander, D., Graybeal, B., and Moore, M., (2004) Routine Highway bridge Inspection
Condition Documentation Accuracy and Reliability. Journal of Bridge Engineering, ASCE, 9(4),
403–413.
Rose, J., (2004) Ultrasonic Waves in Solid Media, Cambridge University Press, Cambridge, UK.
Sanayei, M., Imbaro, G., McClain, J., and Linfield, B., (1997) “Structural Model Updating Using Experimental
Static Measurments,” Journal of Structural Engineering, ASCE, 123(6), 792–798.
Sanayei, M. and Saletnik, M. J., (1996a) “Parameter Estimation of Structures From Static Strain Measurements,
Part I: Formulation,” Journal of Structural Engineering, ASCE, 122(5), 555–562.
Sanayei, M. and Saletnik, M. J., (1996b) “Parameter Estimation of Structures From Static Strain Measurements,
Part II: Error Sensitivity Analysis,” Journal of Structural Engineering, 122(5), 563–572.
Sanayei, M. and Santini, E., (1998) “Dynamic Bridge Substructure Evaluation and Monitoring System,” Report
of subcontract to Tufts University from FHWA grant to Olson Engineering.
Sarradj, E., (2003) “High frequency boundary integral method as an alternative to SEA,” Proceedings of 10th
International Congress on Sound and Vibration, Stockholm, Sweden.
Sarradj, E., (2004) “Energy-based vibroacoustics: SEA and beyond.” 30th German Convention on Acoustics
(DAGA) and 7th Congrès Francais d’Acoustique (CFA).
Sikorsky, C., Stubbs, N., Park, S., Choi, S., and Bolton, R., (1999) “Measuring Bridge Performance Using
Modal Parameter Based Non-Destructive Damage Detection,” Proceedings of the 17th International
Modal Analysis Confrence & Exposition, Kissimmee, FL.
Siswobusono, P., Chen, S., Zheng, L., Yamak, B., Jones, S. L., and Callehan, D., (2004) Dynamic Load Rating
of Rural Bridges. International Modal Analysis Conference XXII, Michigan, MI.
Stubbs, N., Park, S., Choi, S., Sikorsky, C., and Bolton, R., (1999) “A Methodology to Nondestructively
Evaluate the Structural Properties of Bridges,” Proceedings of the 17th International Modal Analysis
Conference & Exposition, Kissimmee, FL.
The Steel Company of Canada Limited., (1966) Hollow Structural Sections, Hamilton, Ontario, ON.
Timoshenko, S., (1955) Vibration Problems in Engineering, Van Nostrand Reinhold Company, New York.
Timoshenko, S., Young. D. H., and Weaver, W. Jr., (1974) Vibration Problems in Engineering, 4th Edition, John
Wiley & Sons, New York, Chapter 5.
Tolbert, R., Zheng, L., and Chen, S., (2004) Modal Testing of a Unique Cantilevered Structure above the Smith
Lake. International Modal Analysis Conference XXII, Michigan, MI.
Trippi, R. and Turban, E., (1992) Neural Networks in Finance and Investing, Probus Publishing Company,
Chicago, IL.
Vlahopoulos, N., Zhao, X., and Allen, T., (1999) “An Approach For Evaluating Power Transfer Coefficients
For Spot-Welded Joints In An Energy Finite Element Formulation”, Journal of Sound and Vibration, Vol.
220, No. 1, 1999, pp. 135–154.
Waas, G., (1972) “Linear Two-dimensional Analysis of Soil Dynamics Problems in Semi-infinite Layered
Media,” Ph. D. Thesis, University of California at Berkeley, CA.
Washer, G. A., (2002) “Developments in NDE for Highway Bridges in United States.” Eight European
Conference on Nondestructive Testing, European Society of Nondestructive Testing, Barcelona, Spain.
Yang, Q. and Liu, J., (2007) “Structural Damage Identification Based on Residual Force Vector,” Journal of
Sound and Vibrations, 305(1–2), 298–307.
Zhao, X. and Vlahopoulos, N., (2004) “A basic hybrid finite element formulation for mid-frequency analysis of
beams connected at an arbitrary angle,” Journal of Sound and Vibration, Vol 269, No 6, pp. 135–164.
Zienkiewicz, O. C., (1971) The Finite Element Method in Engineering Science, McGraw-Hill, New York.
Zimmerman, D. and Kaouk, M., (1994) “Structural damage detection using a minimum rank update theory,”
Journal of Vibration and Acoustics, 116(2), 222–230.
7 Damage Identification (DMID)
7.1 INTRODUCTION
7.1.1 Mathematical Representation of Damage
We start this chapter by trying to define damage. A general definition that would fit our purposes
is as follows: Damage is the change of structural properties; such a change usually results in a
degradation of performance.
Since damage is a change in an initial state, then to identify the damage, we have the following
two approaches:
1. Identify the damage directly, by observing it

2. Identify the change, that is, by observing a baseline state, and a changed state, then quan-
tifying the differences between the two states as the damage of interest
Methods that identify damages directly include most conventional nondestructive testing (NDT)
methods. In most of these methods, no baseline state is needed. Methods that identify the change
in structural state would need a baseline (initial state) and a current state. Both states can be ana-
lytical or experimental. Of course, if both initial and current states are described analytically (by
finite elements method, for example), the DMID becomes an analytical exercise. We recognize that
this practice is followed in many communities in the civil infrastructure field. However, it is not
explored in this volume. For our purpose, we consider only situations where experiments are part of
the overall project.
Methods that identify damage as a change from one state to another would require a minimum
resolution in the method used for detection. Such a resolution can be temporal, spatial, or both.
In this situation, we introduce the analogy of changes of functions of some independent variables.
We recognize immediately that to accurately compute change in a function, in a finite difference
sense, for example, there need to be certain resolutions in the computations of the functions. DMID
as a change of state is similar: there need to be certain resolutions in computing the states (both
initial and current) of the system. Lacking such resolutions would result in an inaccurate damage
definition. When the states (initial and/or current) are evaluated either analytically or experimen-
tally, different types of resolutions are needed, depending on the type of evaluation. These points
will be explored in detail in later sections of this chapter.
7.1.2 Identification Adequacy, Utility, and Prioritization of DMID Methods

Any DMID method needs to be adequate in three issues: (a) technical, (b) utility, and (c) cost–
benefit. Unfortunately, most NDT/structural health monitoring (SHM) experiments concentrate on
technical adequacy, and neglect utility, or cost–benefit issues. All three issues must be considered
carefully, before embarking on any DMID project.
Technical: There are many technical issues that relate to DMID. Among these issues are:
• Scale (size) of damage

• Failure modes
373
• Defect modes
• Temporal considerations, which includes before, during, and after hazard occurrence: also
includes frequency ranges of interest
• Spatial considerations
Utility: Among utility parameters of DMID are:
• Size of equipment
• Simplicity
• Environmental effects
• Maintenance needs
• Labor needs
• In situ versus laboratory settings
• Cost
Methods for computing the utility of any event are discussed in Chapter 8. Extending utility analysis
to DMID methods can be easily done using most of those methods.
Cost–Benefit: Decision making methods provide many methods for computing the cost benefits
of different methods of DMID. Those methods and numerous examples are given in many parts of
this volume. They will not be discussed in this chapter.
7.1.3 This Chapter

This chapter will explore in detail different damage parameters; we will argue that there are several
parameters that define damage and to accurately identify damage, we need to know as much as pos-
sible about these parameters, their causes, and their effects on the structure on hand. Next section
will be devoted to differentiating between DMID and structural identification (STRID). There are
some who consider DMID as an extension of STRID. We show that the two fields are not exten-
sions of each other, and they should be approached with different mindset. We argue that damage is
the rate of structural change, both in time and in space. It is analogous to a given function of some
variables and its partial derivatives with respect to these variables. To obtain the derivatives of the
function, given the function itself, there should be certain numerical rules. We follow such an anal-
ogy and provide rules for DMID from popular STRID methods. If such rules are not followed, the
accuracy of the identified damage would be in question.
We devote the next section to exploring popular NDT methods and their roles in DMID. We always
discuss those NDT methods from SHM in civil infrastructure viewpoint. The following section in this
chapter features the acoustic emission (AE) method. We use AE for an in-depth discussion because of
its many special properties and its larger potential in SHM applications. Vibration-based DMID and
signal processing applications are also discussed. We consider signal processing as an integral part of
DMID methods since most of them are time dependent. This chapter will conclude by presenting the
role of DMID methods and techniques in Structural Health in Civil Engineering (SHCE).
7.2 DAMAGE PARAMETERS

Damages in a bridge structure, or any type of civil infrastructure, vary “depending on the cause of
the damage. In addition to the cause of the damage, several other parameters need to be identified.
We will call these parameters damage parameters.” Generally speaking, the damage parameters
are as follows:
• Cause of damage
• Type of damage
Damage Identification (DMID) 375
• Temporal characteristics of damage

• Location of damage
• Extent of damage
• Severity of damage
The causes of damage vary immensely in a bridge structure. These causes can be natural disasters
such as earthquakes or accidental such as impact. Types of damage include small or large cracks,
degradation of material properties, or loss of structural material. Temporal characteristics of dam-
age can range from very slow such as in corrosion damage to very fast such as damages that result
from accidental impact. Location of damage can be almost anywhere in the vicinity of the bridge;
it is obviously directly related to the cause of the damage. The extent of the damage can range from
a microscopic extent to a global extent. Finally, damages vary in their severity: some damages are
of minor effect, while others have a severe effect, and must be addressed immediately. Table 7.1
describes some hazards and the parameters of damages that result from them.
TABLE 7.1
Examples of Damage Parameters
Temporal Potential Type of Potential
Hazard Type of Damage Characteristics Location Extent Failure
Fatigue Cracks (can be Can be slow or Usually at structural Localized Can be
hidden) sudden processes connections damage catastrophic
Corrosion Degradation of Slow process Where concrete or Can be Due to the slow
properties due to steel is subjected to extensive process, effects
changes in moisture, running can be mitigated
material chemical water, or salt (such
compositions as de-icing materials)
Concrete Cracks, e.g., due Can be slow due to Damage location will Can be Due to the slow
Cracking to freeze and environmental, or depend on structural extensive process, effects
(Other than thaw sudden due to configurations and can be mitigated
corrosion) mechanical loads its response to
environmental or
mechanical effects
Scour Soil erosion Can be slow (sandy Foundation-soil Can be Due to
soil) or sudden interfaces extensive submergence of
(rocks) damage (soil
erosion), the
extent of damages
can vary from
minor to major
Impact Large Sudden due to Structural Ranges from Extent of damage
deformations, mechanical components that are local to global varies from minor
delamination, loading exposed to collision damage effects to major
etc. (e.g., bridge
underpasses)
Deterioration Degradation of Slow processes Damage location will Ranges from Due to the slow
(wear and material depend on structural local to global process, effects
tear) properties configurations and damage effects can be mitigated
its response to
environmental
effects
FIGURE 7.1 Deterioration damage. (Courtesy of New York State Department of Transportation.)
FIGURE 7.2 Impact (collision) damage. (Courtesy of New York State Department of Transportation.)
It is obvious that all of the damage parameters in Table 7.1 should be directly related to the
method of DMID in an SHM experiment. If the DMID method is not consistent with the param-
eters of damage it aims to identify, the efficiency and accuracy of the experiment can be com-
promised. For example, Figure 7.1 shows general deterioration and corrosion damages, while
Figure 7.2 shows impact damages. Clearly, types of DMID methods for each of these two situations
need to be different. Karbhari et al. (2005) studied the pertinence of numerous DMID methods as
they relate to fiber-reinforced polymers (FRP) wrapping. Table 7.2 shows examples of pertinence
of DMID methods in identifying hazard-related damage.
7.3 STRID, DMID, AND SHM

7.3.1 Overview
There are numerous STRID methods and techniques. The purpose of STRID in the civil infrastruc-
ture arena include the following: (a) design validation of new structures, (b) condition assessment of
TABLE 7.2
Relationship of Damage Sources and Damage Identification Methods
Hazard
DMID
Method Fatigue Corrosion Earthquakes Scour
Vibration Usually ineffective, Usually ineffective, Useful due to the Limited use due to the
methods except for very except for very global nature of brittle nature of scour
extensive fatigue extensive corrosion earthquake motion damage
damages damages
Ultrasound Useful when damage Usually not used due Effective only after the Effective only after the
location is known. to the extensive occurrence of an occurrence of a scour
Labor intensive labor needs earthquake (after event). damage (after event)
Strain Useful when Does not measure Effective only after the Effective only after the
measurements damage location corrosion extent, occurrence of an occurrence of a scour
(at selected is known, or measure only the earthquake (after damage (after event)
locations) suspected structural response event)
to damage. As such,
it has limited value
except when
corrosion damage is
extremely large
Penetrating Useful when damage location is known. Labor intensive. Logistical and Not practical
radiation safety issues can be problematic
existing structures, (c) analytical model updating of existing and new structures, and (d) DMID of
existing structures. Many STRID researchers utilized STRID methods for DMID. However, not all
STRID methods are suited for DMID and not all damages are possible to be identified by popular
STRID methods. This chapter investigates the relationship between STRID methods and DMID.
We investigate which STRID method is suited for DMID for the type of damages that may affect
civil infrastructure. It will be shown that some damages can be identified by STRID methods, while
some other damages need specific methods that are not within the conventional realm of STRID
methods. Some guidelines for choosing STRID methods that are appropriate for DMID are given.
Structural health monitoring is an important emerging engineering field for infrastructure appli-
cations. This is due to its potential of saving costs, while improving safety. SHM was shown (see
Ettouney and Alampalli 2000) to involve several fields: measurements, STRID, DMID, and deci-
sion making. Several STRID methods have been presented in the past (e.g., see Doebling et al. 1996;
He and Fu 2001). Many of the STRID methods have the potential of identifying damage in the
structure, thus fulfilling one of the most important objectives of SHM, and bringing to the decision
maker valuable information that can help in managing their assets more effectively. Thus, the use of
STRID methods in DMID is attractive, since there is a plethora of STRID methods that are fairly
familiar to researchers and practitioners.
Unfortunately, the use of STRID methods for DMID must be done with extreme care. Using
the wrong STRID method for a particular DMID situation might lead to an erroneous result, which
might lead the decision maker to make the wrong decision. This chapter studies the interrelation-
ship between STRID and DMID. First, we identify different objectives of STRID and DMID. Next,
different methods of STRID are briefly discussed. We then discuss different pertinent damage clas-
sifications and introduce a unifying damage classification that is appropriate for the purpose of this
chapter. Theoretical issues that can affect the results of STRID methods as tools for DMID are then
presented. Finally, a step-by-step guide to check for the adequacy of STRID methods in detecting
damage is presented. Following such steps will ensure accurate DMID when using different STRID
methods.
7.3.2 Objectives of STRID and DMID

7.3.2.1 Objectives of STRID
STRID in civil engineering applications has several objectives: condition assessment, design valida-
tion, analytical model updating, and DMID. We discuss each of these objectives next.
Condition Assessment of Existing Structures: Condition assessment includes estimating values
of system properties that might be of importance to the performance of the system on hand. For
example, it might be of interest to estimate the effective moment of inertia of a concrete beam,
Poisson’s ratio of underlying soil, or effective axial pile stiffness. Parameter estimation methods can
fulfill such an objective directly.
Design Validation of New Structures: Since design of new structures relies on numerous assump-
tions, validation of some of these assumptions using experimental methods is needed. For example,
modal damping or modal frequencies that are used in modern seismic designs may need verification
or validation. Such validation can improve safety, reduce costs, or both.
Analytical Model Updating for New and Existing Structures: Complex analytical mod-
els for structures are used for purposes other than just design. For example, they can be
used for weight estimations, cost estimation, determining the feasibility of construction
procedures and techniques, and developing repair or retrofit needs. Different STRID meth-
ods would have potential to provide accurate analytical structural models to accomplish
these objectives.
Damage Identification of New and Existing Structures: Numerous STRID methods are offered
as capable of meeting the above objectives and detecting the damage. The capability of STRID
methods in detecting the damage is the focus of this chapter.
All of the above objectives of STRID are of use in SHM. However, our immediate objective is to
evaluate the DMID potential for different STRID methods.
7.3.2.2 Objectives of DMID

Damage identification is one of the basic components of SHM (see Ettouney and Alampalli 2000).
It involves several objectives as given below (see Farrar and Jauregui 1996):
Identify Occurrence: Did the damage occur? For example: the start of corrosion, the start of
soil scour, and so on.
Identify Location: Where did the damage occur? For example: the location of a void in a
post-tensioned tendon.
Identify Extent: How extensive is the damage? For example: the amount of rust in the cross-
section of a steel rebar.
Identify Remaining Useful Life of Structure: Given the above damage properties, what is
the remaining expected service life of the structure?
In addition to the above, we add another potential objective of DMID:
Expected Cost: Given the above damage properties, what are its effects on the life cycle cost
of the structure?
7.3.3 Definition and Classifications of Damage

Before proceeding further into the discussion of the suitability of STRID methods for DMID, we
have to first define “damage.” A simple and encompassing definition would be that the damage is
an unwanted change in the system property or properties that occurs due to a natural or manmade
environment.
7.3.3.1 Physical Classifications

The above definition is too general and would not serve us well in our attempt to explore the inter-
relationship between STRID and DMID. We need to be more specific. We submit that a single
definition of damage, though attractive for its simplicity, might not be well suited to our immediate
task. Because of this, we propose the following physical classifications: cause, material, and extent.
Causes for damage are many and include corrosion, fatigue, impact, fire, and normal wear and tear.
Materials include conventional steel, reinforced concrete, aluminum, post-tensioned or prestressed
concrete as well as modern construction materials such as FRP. Damage extent can range from a
hairline fatigue crack in a welded steel connection to large sudden (or gradual) soil erosion under a
large size foundation.
Due to the large variability in classification and type of damage, it is reasonable to deduct that
no single STRID method can identify all types of damage. This is an important observation, but
not too beneficial. Let us try to address the apparent complex STRID-DMID interrelationship by
introducing new classifications of damage: the detectability classification.
7.3.3.2 Detectability Classification

Instead of classifying the damage itself, we classify the method of identifying the damage. We
propose this classification of DMID method: All DMID methods in any structural system can be
classified in one of two methods: direct or inferred. Direct DMID methods are those methods that
purport to identify the damage directly. Direct methods, generally speaking, do not require base-
line measurements, that is, they can detect the damage using only information from the damaged
state. Inferred DMID methods are those methods that require information from a baseline state in
addition to information from the damaged state. The STRID methods that estimate damage using
inferred methods are the subject of this chapter. Those STRID methods can be categorized in to
three categories: modal identification, parameter identification, and nonphysical (functional) meth-
ods. Detectability classifications of DMID methods are shown in Figure 7.3. We will discuss only
STRID methods that detect damage in an inferred mode.
7.3.4 STRID Methods and Inferred Damage Identification

7.3.4.1 Overview
The first STRID category that can identify damage in an inferred mode is the modal identifica-
tion method category (see He and Fu 2001). This method estimates modal properties by utilizing
dynamic measurements of the structure of interest. Modal properties are mode shapes, natural fre-
quencies, and modal damping. Since that damage is not measured directly, it must be detected by
inference, that is, its properties must be inferred from the identified modal properties.
Parameter identification is another class of STRID that uses inference to detect damage
(see Doebling et al. 1996). The method is based on modeling the structure of interest using a
suitable numerical technique, such as finite elements. Some of the main parameters that are
used in modeling (such as material properties, element properties, etc.) are then identified
using the results of testing. This realization of a numerical model is consistent with tests.
The damage is related to the realized model by inference, since the damage is not measured
or detected directly—only the changes in structural parameters are realized, not the dam-
age itself.
The final class of STRID methods is nonphysical (functional) methods. The most popular tech-
nique in this classification is neural networks (see He and Fu 2001). Neural networks compute
relations between a set of inputs and the corresponding set of output. These relationships are then
evaluated by minimizing an error function in a least square sense. Neural networks are suited for
nonlinear systems. However, the physical interpretations of the realized relationships in the network
are usually difficult. Because of this, damage can only be identified by inference.
Damage identification methods

(detectability classifications)
Inferred Direct
SHM methods NDT methods
Virtual sensing SHM methods
Signal
processing
Structural
identification
Modal Parameter
Neural networks
identification identification
FIGURE 7.3 Damage identification detectability classifications.
7.3.4.2 Factors Affecting Inferred Damage Detection

For any STRID method that is to be used as a DMID tool, there are four factors that need to be
satisfied for successful DMID. These four factors are discussed below.
7.3.4.2.1 Error Scale
One of the basic steps in any STRID method is that it relies on optimization. The objective func-
tions of such an optimization is usually associated with some measure of error, E, as computed by
the difference between the computed results of the estimated structural model Xi and the measured
experimental results Yi where i is a counter that represents the input-output pairing. The error func-
tion can be defined conceptually as

(
E = f X i − Yi )
By definition, the units of such an error function are the same as the measured experimental
results.
Examining a damaged structure using a STRID method that detects damage in an inferred mode
requires a baseline and a damaged state data to detect the damage. The baseline can be experimen-
tal or analytical. Similarly, the damaged state data can be experimental or analytical. We can then
define the evaluated damage D as
D = g (U D − U B )
Or
D = g (Δ U )

where UD and UB represent responses (measured or analytic) of the damaged structure and
the baseline structure, respectively. Note that UD and UB are evaluated with an error of E.
This means that for the estimated damage, the following inequality must be satisfied
O ( D) >> O ( E )

For example, if the error used in identifying a particular structural parameter is 0.1%, then the
effect of the damage must be much larger than 0.1%. A convenient measure of damage effect might
be the damage ratio expressed as ∆U/(UD − UB). Admittedly, this is a strict limitation. However,
if it is not accommodated, then the STRID method of interest is not adequate in identifying the
damage of interest. The error scale is applicable to all STRID methods when used for DMID.
7.3.4.2.2 Dynamics Scale

Any damage of a structure would generate a dynamic effect zone (DEZ) beyond which its effects
would be negligible. The size of the DEZ depends on the type and extent of the damage as well as the
frequency of excitation and the material properties. Furthermore, it is reasonable to state that the DEZ
is largest at specific frequencies of excitation. These specific frequencies are those frequencies where
resonances that the damage causes occur. It is possible to estimate the relationships between all of
those parameters and the DEZ. Unfortunately, this is beyond the scope of this chapter. Qualitatively,
for example, if we can state that for a steel crack of a length of order of 1.0 in, the effective frequency
of excitation would be in the order of 1.0 KHz, or higher. Then, for a STRID method to successfully
identify such damage, it should be accurately validated at 1.0 KHz, or higher.
The error scale is applicable to most STRID methods when used for DMID. The only exceptions
are the statics-based parameter identification methods (see Sanayei and Saletnik 1996a and 1996b).
7.3.4.2.3 Spatial Scale

Damage, as defined earlier, is a change in the structure condition. In an SHM situation, the change
should be small, in spatial sense. If the damage (change) is large, then there is no need for SHM in
the first place: the size of the damage makes it identifiable by manual (e.g., visual) methods. The
small spatial size of the damage would require using smaller size finite elements that are compat-
ible with the expected damage size. For example, if the expected damage is a 1.0-in crack in a steel
girder, it is prudent to have the finite elements in the area of a damage of size that is smaller than
1.0 in. If the element sizes are larger than 1.0 in, then the damage effect would be averaged over a
larger area. Such an averaging might result in erroneous conclusions. Let us consider the example
shown in Figure 7.4. The beams and columns are modeled using conventional axial-flexure ele-
ments. In addition, the beam-to-column connections are modeled by simple springs. The spring
parameters, in a linear analysis parameter identification method would be the spring stiffness. The
axial-flexure parameters would be the area, inertia, and material properties. Such a model will
produce average results, such as a prediction of damage in the spring (reduced spring stiffness).
However, detecting the exact damage type (loss of bolt tightening, weld cracking, slippage, etc.), is
not possible without a more detailed finite element modeling.
The error scale is applicable to most STRID methods when used for DMID. The only exception
is the statics-based parameter identification methods (Sanayei and Saletnik 1996a and 1996b).
7.3.4.2.4 Geometric Modeling

Spatial scale referred to the size or resolution of finite elements that are appropriate for parameter
identification methods when used for DMID. The geometric modeling issue refers to the accuracy
of the model itself. It is essential to accurately model the structural geometry, utilizing appropri-
ate finite elements, in the vicinity of the expected damage. For example, it is popular to model
beams and columns in structures by using the popular axial-flexure element shown in Figure 7.4.
We should remember that such elements are based on the assumption that the cross-section of the
element remains rigid, that is, does not deform except as a plane (Connor 1976). Due to this basic
assumption, if the expected damage is in the form of a partial change along the depth of the element
(a partial crack, or limited corrosion, for example), modeling the system as an axial-flexure element
Structural connections
(Modeled by scalar springs)
Finite element nodes
Rotational soil spring
FIGURE 7.4 Example of spatial scale in parameter identification problem.
TABLE 7.3
Factors Affecting STRID Methods for DMID in Inference Mode
STRID method
Category Error Scale Dynamics Scale Spatial Scale Geometric Modeling
Modal Error tolerances during Frequencies that N.A. N.A.
identification modal or parameter define dynamic
Parameter identification realization effect zone (DEZ) Size and resolution of Geometric and
identification must be smaller than for the damage of finite elements, material details in
damage ratio interest must be especially in the the damage area
within the range of damage area, must must be sufficient
the test and/or be adequate for the to model the
identified expected damage expected damage
frequencies size
Neural Acceptable errors during NA Can be of
Networks network training must importance when
be smaller than damage designing the
ratio network topology
will not be accurate. A detailed 3D element is needed to model such an expected damage. An
example of the need to accurately model the geometry is if the damage of interest is the corrosion
of rebars in a given reinforced concrete beam. In such a situation, modeling the reinforced concrete
beam as an equivalent homogeneous beam is not adequate. The steel rebars need to be modeled
explicitly to be capable of identifying the corrosion damage in the rebars.
Similar to the spatial scale, the geometric modeling issue is applicable only to parameter identi-
fication methods. It is not pertinent to modal identification. For neural networks, it might be appli-
cable to the topology of the network. However, this subject is beyond the scope of this chapter.
7.3.4.3 Summary
Table 7.3 summarizes the different requirements of STRID methods when used as a DMID tool in
an inferred mode.
7.3.5 Guide for Using STRID as a DMID Tool

On the basis of the previous discussions, it is clear that using STRID methods to identify damage in an
inferred mode needs careful consideration. Figures 7.5 through 7.7 show step-by-step guidelines and
help a user to assess the adequacy of three STRID methods as a DMID tool in the inferred mode.
Suitability of modal identification for damage identification:

quick checklist
• Modal testing information

• Damage attributes
No Is the frequency (time)

scale of measurements
of same, or lower order
as the damage scale?
Yes
Is the error in
No identifying the modes
of same, or lower order
than the damage effects
on the structure?
Yes
Modal identification method is Modal identification method is

not appropriate for damage appropriate for damage
detection detection
FIGURE 7.5 Guide to using modal identification methods for damage identification.

We just explored different methods of DMID. Specifically, we were interested in the applicability of
a certain class of STRID methods that are used for DMID. First, we classified DMID methods into
direct and inferred classes. Of the inferred class of DMID, we discussed STRID methods that might
detect damage in inference mode. The four factors that need to be considered, if the STRID method
is to be successful in detecting the damage in an inferred mode, are discussed. We presented a
guideline for three of the STRID methods that need to be followed to ensure a successful DMID.
Similar guidelines need to be developed for other DMID methods. Developing these guidelines will
help professionals to choose the right DMID method for any given SHM project.
7.4 NDT TECHNIQUES

7.4.1 Introduction to NDT Concepts and Methods
Traditional NDT methods have been used extensively for DMID in civil infrastructure. They are
regarded as a mainstay in the emerging civil infrastructure SHM field. Because of this, the decision
makers need to understand their basics, strengths, and weaknesses. Without such an understanding,
those methods can’t be used efficiently in a cost-effective manner in SHM.
Suitability of parameter identification for damage

identification: quick checklist
• Testing information
• Structural model
attributes
No Is the frequency (time) Yes No

scale of measurements Is the size of elements
of same, or lower order adequate for damage
as the damage scale? simulation?
Yes
Is the error in No
identifying the modes Yes Is the damaged area
of same, or lower order adequately modeled?
than the damage effects on Are material Properties
the structure? appropritate?
Yes
No
Parameter identification method Parameter identification method

is not appropriate for is appropriate for damage
damage detection detection
FIGURE 7.6 Guide to using parameter identification methods for damage identification.
Nondestructive Testing methods are numerous, and vary immensely in cost, complexity, scien-
tific basis and techniques; we aim first to find a single basic rule that might help in bring most of
NDT methods into focus. Such rule might be the basic wave length relationship:
c=lf (7.1)
With λ, f, and c representing the wave length, the frequency, and speed of light. The rule is perhaps the
single underlying common rule that governs most popular NDT methods. The wave length changes
inversely as the frequency changes. Different types of waves occur at different wave lengths, as
shown in the wave spectrum (Figure 7.8). The relationship between the frequency and wave length
is shown in Figure 7.9. This figure also shows the spectrum used by various NDT methods.
We note that different NDT methods target different ranges of the wave spectrum. For example,
ultrasonic (UT) methods target lower wave frequency range at higher wave frequency range, ther-
mal radiation (thermography methods) resides in the infrared waves range. Next, we find the visual
light frequency range that is the basis of visual and laser-based NDT testing methods. At higher
frequency range the microwave and radar frequency range form the basis of electromagnetic wave
NDT methods. At even higher ranges we find the penetrating radiation NDT X-rays and gamma
rays. Figure 7. 9 illustrates these relationships.
Suitability of neural networks for damage identification:

quick checklist
• Testing information (input

output)
Is the frequency
No (time) scale of
measurements of same,
or lower order as
the damage
scale?
Yes
Is the error in
No Network learning
process of same, or lower
order than the damage
effects on the
structure?
Yes
Neural Network method is Neural network method is

not appropriate for appropriate for damage
damage detection detection
FIGURE 7.7 Guide to using neural networks methods for damage identification.
Ultraviolet Visible
light
Cosmic Gamma X-rays Infrared

rays rays Microwaves Radio Ultrasonic
waves
Thermal
radiation
10–5 10–4 10–3 10–2 10–1 1 10 10+2 10+3 10+4 10+5 10+6 10+7 10+8 10+9 10+10
Wave length - nanometers
FIGURE 7.8 Wavelength spectrum.
As the wave length and the type of waves change, the properties of these waves change. The inter-
action of the different waves with test objects also changes. The state of the test object (properties,
geometry, or damage) can be detected by observing the interactions of waves of different types and
the test object. NDT methods aim at identifying such interactions in quantitative or qualitative manner.
The NDT technique changes as the type of wave changes. It is the objective of this section to study how
different NDT techniques vary in relationship with the types of pertinent waves, and how the tools,
analysis, and processing of such information can lead to identifying the desired state of the test object.
10–5
Cosmic
Increasing wavelength
10–4 rays
10–3
Gamma
rays
10–2
Penetrating
radiation
10–1
1 X-rays
Wave length - nanometers

10
Ultraviolet
10+2 Optical
Visible light methods
10+3
Thermal
radiation
10+4 Thermography
Infrared
10+5
10+6
Microwaves
10+7 Electromagnetic
10+8 Increasing Frequency

Radio
10+9 waves
10+10
Ultrasonic
Ultrasonic
1-As the 2-The type of wave 3-and the 4-...and the

wavelength changes... frequency pertinent NDT
changes changes method changes
FIGURE 7.9 NDT method dependence on type of waves.
This section will investigate some popular NDT methods such as UT, electromagnetic waves
(MV), thermography, and penetrating radiation. More NDT methods are briefly discussed at the end
of the section (such as visual, liquid penetrant, and optical methods). In discussing all the methods,
we try to present the basic technical grounds of the method and the tools of the method. Also the
advantages and disadvantages of each method are presented. If the method has more than one basic
technique, we will highlight the different techniques. For each of the methods we will relate them to
the field of SHM and explore how different SHM issues are related to NDT techniques.
7.4.2 Ultrasound Methods

7.4.2.1 Overview
Ultrasonic testing is very popular technique in NDT field. It is based on generating UT waves in
the test object. As the waves propagate through the object, it changes its form through reflections,
refraction, and scattering. By using a listening device to the signals, these signals can be processed
and the state of the test object can be detected. There are numerous NDT UT methods such as
impact-echo, pulse echo, sonic vibrations, guided waves, and laser-based UT. We will explore the
basics of these methods in this section. There are other types of sonic/UT testing. Those types of
testing methods are beyond the scope of this document, the reader should consult Birks et al. (2007).
Another major UT NDT method: AE is discussed in more detail elsewhere in this chapter. We note
that the term ultrasound is also used to describe UT NDT methods; we use both terms interchangeably
in this volume.

The basic equations of wave propagation in solids are described in the Appendix in Section 7.9
(Lamb waves), the Appendix in Section 7.10 (dispersion curves), and the Appendix in Section 7.11
(Helmholtz equation). More in-depth discussion of waves in solids can be found in Rose (1999). Table
7.4 discusses the basic technical issues in ultrasound waves in solids as they apply to SHM field.
7.4.2.3 Impact-Echo
7.4.2.3.1 General
This is the most basic of ultrasound NDT techniques; see Fisk (2008). It is based on generating
normal waves (compression or tension) within the testing object by impacting it mechanically. The
signal will travel in the body, and if there are any defects of damages, the resulting sound will reveal
such defects.
7.4.2.3.2 Tools and Instruments

The impacting source can be a hammer or mechanically applied pulses. The responses can be
detected by microphones or other suitable sensors.
7.4.2.3.3 Uses in NDT Applications

Since this type of testing is mostly qualitative, the resolution and accuracy of the test is also rela-
tively very low when compared with other ultrasound tests. The tests can reveal flaws, debonding, or
general qualitative damages. The resolution of the damages is in the order of tens of millimeters.
7.4.2.3.4 Use in Condition Assessment

The traditional approach in condition assessment of bridge decks by impact-echo is based on
review of individual test point records. A new automated approach based on three dimensional
TABLE 7.4
Ultrasonic NDT Technical Issues as Applies to SHM
Ultrasound Technical Effects in SHM
Issues
Phase/particle velocities Can be detected by comparing theoretical dispersion curves with measured frequency-
time spectra
Scattering Can be beneficial if theoretical solutions can be validated with experimental
measurements
Wavelengths Some ultrasonic methods allow for scaling wavelengths, such as ultrasonic vibration
methods. This can be invaluable in SHM experiments
frequency range For successful general SHM experiments, frequency range must be as wide as possible.
This is needed to cover both global and local modes of structural behavior
Attenuation As frequencies increases, waves tend to attenuate faster. This necessitates higher density
of sensors. Some ultrasound methods cover wider areas naturally, such as laser-based
ultrasonics (LBU)
Medium between probe Due to the large impedance mismatch between air and engineering solids (steel,
and test article concrete, etc.) ultrasonic transducers need to be in full contact with the solids. In
certain situation a medium needs to be present between the transducers and the test
object. Such needs can make SHM experiments more complicated. Remote sensing
methods do not have such a requirement
data visualization was presented by Gucunski et al. (2008). The IE test was conducted using
an impact source and a single nearby receiver. The return frequency together with the previ-
ously determined P-wave velocity can be used to measure the depth of the reflector, in this case
the deck thickness using a simple relationship T = Vp/2fr, where fr is the return frequency.
Depending on the extent and continuity of the delamination, the partitioning of energy of elas-
tic waves may vary and different grades can be assigned to that particular section of a deck
as a part of the condition assessment process. In the case of a sound deck (good condition), a
distinctive peak in the response spectrum corresponding to the full depth of the deck, can be
observed (see Figure 7.10). In the case of a delaminated deck, reflections of the P-wave occur
at shallower depths as well as deck bottom, showing two distinct peaks in the response. At the
same time, a progressed delamination is characterized by a single peak at a frequency corre-
sponding to a reflector depth indicating that little or no energy is being propagated toward the
bottom of the deck. In a very severe case of a wide delamination, the dominant response of
the deck to an impact is characterized by a low frequency response of the upper delaminated
portion of the deck.
Good (intact) condition Fair condition

Impact Receiver Impact Receiver
source source
T t
1.2 1.2
Normalized amplitude
1 1
Rt.287 - 2360 - good Rt.287 - 2546 - fair
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 10000 20000 30000 40000 0 10000 20000 30000 40000
Frequency, Hz Frequency, Hz
Poor condition Serious condition

Impact Receiver Impact Receiver
source Delamination source Delamination
t t
T
1.2 1.2
1 Rt.287 - 2550 - poor 1

Rt.287 - 2594 - serious
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 10000 20000 30000 40000 0 10000 20000 30000 40000
Frequency, Hz Frequency, Hz
FIGURE 7.10 Grades for various degrees of deck delamination. (Reprinted from ASNT Publication;
Gucunski, N. et al., Automated interpretation of bridge deck. Impact echo data using three-dimensional visu-
alization, NDE/NDT for Highways and Bridges: Structural Materials Technology (SMT), ASNT, Oakland,
CA, 2008. With permission.)
Using the reflector depth relationship, a unique thickness or depth can be correlated to every
component of the frequency spectrum as illustrated in Figure 7.11 by a frequency spectrum and to
it corresponding thickness spectrum.
Advanced data visualization techniques were then used for data interpretation through 3D trans-
lucent imaging of IE test data and separation of important reflector layers (see Figure 7.12). The
presented visualization and to it associated interpretation algorithms allows both the overall assess-
ment of the condition of the deck and identification of deteriorated zones of the deck.
7.4.2.3.5 Impact-Echo in SHM

Colla and Wiggenhauser (2000) discussed the advantages and disadvantages of IE for detecting
concrete issues. They observed that IE has shown to be effective to evaluate concrete for locating
voids and discontinuities, thickness measurements, duct cover, and delaminations. They noted the
following IE disadvantages/limitations:
1. Results in literature have not always been conclusive and reproducible

2. For complex element geometry, such as multiple ducts and reinforcements layers cross-
ing each other, interpretation of the single frequency spectra is not always unequivocal. In
those cases, numerical modeling is used along with experimental data, which will limit its
use for generalization
1.2 1.2
1 Deck bottom 1 Deck bottom

0.8 0.8
Delamination Delamination
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 10000 20000 30000 40000 0 0.1 0.2 0.3
Frequency, Hz Reflector depth, m
25,000
Delamination - poor Delamination - poor/fair
20,000
Frequency, Hz
15,000 Deck bottom - good Apparent deep

reflector -
10,000 serious
5000
0
0 1 2 3 4 5 6
Length, m
0
–0.05 Delamination - poor Delamination - poor/fair
–0.1 Deck bottom - good Apparent deep
Depth, m
–0.15 reflector -
–0.2 serious
–0.25
–0.3
–0.35
–0.4
0 1 2 3 4 5 6
Length, m
FIGURE 7.11 Frequency and thickness spectra and spectral surfaces. (Reprinted from ASNT Publication;
Gucunski, N. et al., Automated interpretation of bridge deck impact echo data using three-dimensional visual-
ization, NDE/NDT for Highways and Bridges: Structural Materials Technology (SMT), ASNT, Oakland, CA,
2008. With permission.)
Loward distance, it
10
8
6
4
2
0
1 11 21 31 41 51
Longitudinal distance, it
Legend
2 to + 2 to 2.25 2.25 to 2.5 1.5 to 1.75
2.5 to 2.75 2.75 to 3 2 to 2.25 1.25 to 1.5
2.25 to 2.5 2.5 to 2.75 1.75 to 2 1 to 1.25
Good Fair Poor Serious
FIGURE 7.12 Condition assessment of Carter Creek Bridge deck. (Reprinted from ASNT Publication;
Gucunski, N. et al., Automated interpretation of bridge deck impact echo data using three-dimensional visu-
alization, NDE/NDT for Highways and Bridges: Structural Materials Technology (SMT), ASNT, Oakland,
CA, 2008. With permission.)
3. IE is a point testing method. Hence, selection of test points is very important to identify
damage.
4. Output of these punctual readings is single frequency spectra, whose interpretation and
comparison is difficult.
5. Unfavorable factors include variations in spectra peaks amplitude due to uneven coupling
of the transducers, small frequency shift of the peak due to influence from element vibra-
tion, and due to geometrical location of the measurement point on the element.
6. The frequencies of interest may not be excited. Bad selection of the hammer can cause
the concrete surface to crumble and a lower frequency of excitation is achieved due to bad
signal to noise ratios.
In order to avoid some of the above and to eliminate the need for skilled personnel to plan and test
using IE, the use of scanning IE was suggested by the authors instead of point measurements. An
automated procedure was developed for data collection to reduce influence of transducer coupling.
Cross-sections of the element can be plotted. The method showed potential degree of success.
We observe that similar points can be made about most NDT methods, that is, making it a turn-
key system with reliability can lead to increased use of NDT methods due to the cost–benefit it
offers by doing so.
7.4.2.4 Pulse-Echo (Normal Waves)

7.4.2.4.1 General
The main difference between impact-echo and pulse-echo techniques is the source of the signal.
In pulse echo, the signal is generated mechanically, or electrically, using a probe. The signal travels
Source/receiver
Initial pulse
Defect
Amplitude
1 Reflection from
Plate
edge
2
Time
t2 1 3
No damage
Initial pulse
As the pulse source / receiver moves
along the plate surface, the defect 2 Reflection from
presence can be detected by observing defect
Amplitude
the reflected signals across the plate 3 Reflection from
thickness. edge
If the defect is parallel to the sound Time

pulse direction, this method will not
be efficient in detecting the defect. t1
t2
Damaged condition
FIGURE 7.13 Pulse-echo principles.
within the media in the form of compression/tension (normal) wave. A source receives the signal,
and the pertinent information is then processed from the signal. Figure 7.13 shows the principles of
pulse-echo technique.
7.4.2.4.2 Modes of Scanning

There are several techniques to apply the pulse-echo method. They vary in complexity and resulting
amount of information. Figure 7.14 illustrates some of these concepts, which are briefly described
below.
A-Scan: The UT probe is used to inspect the in-depth state of a plate at a single point on the plate
surface. The results of the scan reveal in a 1D scale the size and location of the damages across the
plate thickness.
B-Scan: The probe is moved along a line on the plate surface. Thus the ultrasound waves cover a
surface that is bound by the depth axis and the line traveled during the scan. The results are displayed
in a 2D diagram that shows the size and location of damages in that surface.
C-Scan: The probe is moved along series of parallel lines on the plate surface so as to cover
certain surface area. Thus the ultrasound waves cover a volume that is bound by the depth axis and
the surface that was covered during the scan. The results are displayed in a 3D diagram that shows
the size and location of damages in that volume.
M-Scan: This technology aims to scan moving objects. Biomedical and manufacturing appli-
cations are some of the more popular examples of M-scan technology. Obviously, the technol-
ogy has little importance in SHM applications of civil infrastructure: Civil infrastructure do not
move.
7.4.2.4.3 Use in NDT Applications

Pulse echo is used for most civil infrastructure materials such as steel, concrete, masonry, or FRP.
It can be used to detect properties and state of as built material or component properties. Defects in
materials such as cracks or debonding can be detected using this technique. Also, pulse echo can be
Defects
Section A-A Pulse-echo test setup
There are three different

Plate types of scanning for
defects in plates using the
pulse-echo technique
A - Scan B - Scan C - Scan
The probe is placed at single The probe is moved along a line The probe is moved along a line
location then probe single line while scanning for defects in the grid that cover the whole area of
across the plate below the probe plate cross-section below interest. It scans for defects in a
plane inside the plate
Cross-section Plan of defects

of defect in A-A Section A-A
in scanned
Amplitude
plane
Time Scanned plane
Only a time signal from the Scanned view shows defects The scanned view is a plan of defects
pulse-echo event is available in the cross-section under the on the scanned plane. If the process is
in an A-Scan path of the probe repeated at different planes, a 3D
representation of defects can be created.
FIGURE 7.14 Scanning techniques for pulse-echo method.
used to measure thickness and density of the material. Most of the applications would be plate-like
materials or near surface conditions.
7.4.2.5 Guided (Lamb) Waves

7.4.2.5.1 General
In this technique, also called transmission UTs, the UT waves are lamb waves, which are also known
as plate waves. Thus, the waves are combinations of flexural and or shear waves. Rayleigh surface
waves can also be generated using this technique. Lamb waves travel along the plate axis. Because
of this, their detection capabilities cover much wider ranges than normal waves that are utilized in
the pulse-echo technique. Because of the versatility of guided waves technique, they can be used in
several modes as discussed next.
7.4.2.5.2 Modes of Testing Guided Waves

7.4.2.5.2.1 Pulse-Echo (Transmission Mode)
In this mode, the UT pulse sender would generate a Lamb wave(s) that travels along the plate long
dimensions until it is reflected either a natural boundary condition (plate edge, for example), or a
defect. Upon reflecting back, the receiver would capture the signal that can be analyzed to obtain
pertinent information (see Figure 7.15).
7.4.2.5.2.2 Pitch-Catch
In this mode the sender and receiver are located far apart. This would enable an even longer detection
surface than in the pulse-echo technique. This is due to the fact that the wave does not have to travel
back from the reflection surface. Figure 7.16 shows the concept of pitch-catch of Lamb waves.
Plate
Sender/receiver Sound pulse
Defect
1 2
Initial pulse
Amplitude
Reflection from
edge 2
Time
t2
No Damage
Initial pulse
Reflection from
Amplitude
defect 1 Reflection from

edge 2
Time
t1
t2
Damaged condition
FIGURE 7.15 Pulse-echo concepts for guided waves method.
Plate
Sender Sound pulse Receiver
Defect
2
1
Pulse seen by
the receiver
Amplitude
Time
t1 1 No damage
Pulse seen by
the receiver
(Note attenuation
Amplitude
effects of damage)
Time
2
Damaged
t1 condition
FIGURE 7.16 Pitch-catch concept for guided waves method.

7.4.2.5.2.3 Pulse-Reflection
This method is similar to the pulse-echo method, except that a separate sender and receiver are used
during the experiment as shown in Figure 7.17
7.4.2.5.3 L-Scan and Guided Waves

L-scan technique is based on angular spectrum method in which the guided waves in plates will
result in propagation modes that involve numerous planar waves. This will result from reflections
and refractions from different boundary conditions, as well as any different damages or anomalies
in the system being tested. It is desirable to decompose those 3D waves to a set of planar wave. The
angular spectrum method can help in identifying different planar waves that are present in a com-
plex wave field. See the Appendix in Section 7.12 for angular spectrum method at end of this chapter
for more details on the method.
7.4.2.5.4 Phased Arrays

Phased-array technique is based on producing N UT signal pulses with specific amplitude and time
delays. The pulse amplitudes and time delays are designed so as to maximize the range, angle, and
overall defect detection of the transducers. The number of the probes within a transducer varies. Some
popular transducers include 16, 32, or more probes. The phased-arrays synchronize both the transmit-
ting pulse properties (amplitudes and delays) and the receiving delays and summation laws as shown
in Figure 7.18. Those synchronized delays can produce an efficient pulse echo, or pulse-reflection
methods. There are numerous probe geometries including linear, 2D, or annual (Figure 7.19).
The phased-array transducers offer numerous detection techniques. The simplest of these tech-
niques are the linear scanning method, where the probes are moved in linear motions along the test
object. The dynamic depth focusing (DDF) method changes the focal length of the array along the
depth of the specimen, thus producing a high efficiency (range and SN ratio) pulse-echo mecha-
nism. By changing the focal length at a given angle, the phased-array produces a sector scan S-scan.
When the S-scan is coupled with a B-scan method, a 3D picture of the object can be produced.
Plate
Sender Receiver Sound pulse
Defect
1
2
FIGURE 7.17 Pulse-reflection concept in guided waves method.
#1 #2 #3 #4
Probe #1 Probe #2 Probe #3 Probe #4

Amplitude
Time Time Time Time
Time-delayed pulses from a

4-probe phased arrays transducer
FIGURE 7.18 Time delays in phased arrays.

Probe
Linear phased array probes (8)
Probe
2-D Phased array probes (64)
Annular
probe
Annular phased array probes (8)
FIGURE 7.19 Geometry of some popular phased-array probes.

Guided waves are very attractive in NDT because of their superior range, and versatile theoretical
underpinnings: they can accommodate several types of elastic waves. They are suitable for detect-
ing cracks (as results from fatigue or corrosion) or voids (in FRP laminates, for example). They can
also be used to measure as built geometries and properties, such as material densities, wave veloci-
ties, or plate thicknesses. One potentially attractive use in FRP bridge decks is to investigate degree
of homogeneity of the composite matrix, ASNT (1999). The method can also be used for quality
assurance (QA) purposes in construction. The method can detect damages up to 0.2 mm.
7.4.2.6 Ultrasonic Vibration
7.4.2.6.1 General
Ultrasonic vibrations technique generates harmonic signals with varying frequencies in the test object.
The response of the object is monitored such that any detected system resonances can be identified and
the required system state is inferred. For example, the thickness of a plate can be measured by sweep-
ing UT signals in the plate (see Figure 7.20). When the driving wave length is twice the plate thickness,
the response is maximized. This will indicate the desired plate thickness. The experiment can be used
to estimate as built plate thickness of steel beams, or FRP bridge deck plating systems.
A potential, yet not fully explored use, of the UT vibrations is to monitor voids in a tendon ducts
in a post-tensioned (PT) bridge system. By imparting a sweeping UT harmonics at the surface of the
PT beam, the response of the local system can be monitored. In a nonbonded situation, the resonat-
ing system will react as if its support is stress free, which will have different resonance frequency
than the perfectly bonded system, as shown in Figure 7.21.
We note that the UT vibrations technique can be considered as the higher frequency scale coun-
terpart of the popular global vibration identification techniques that are discussed elsewhere in this
volume. A potential complete-scale frequency SHM technique can potentially include both global
vibration and UT vibration methods.
1-The transducer produces 3-The response is

ultrasonic signals with varying maximum when the
wavelengths ultrasonic wave has a
wavelength twice the
2-The responses of the solid are thickness of the plate
observed
Using ultrasonic vibrations to measure plate thickness
FIGURE 7.20 Thickness measurements using ultrasonic vibrations.
The signal produced by the

transducer resonate at higher Transducer
frequency due to the impedance
of the grout.
Grout Tendon
Void
The signal produced by the

transducer resonate at lower Transducer
frequency due to the presence
of the void.
Post-tensioned beam
Using ultrasonic vibrations to detect voids in a post-tensioned bridge beams
FIGURE 7.21 Ultrasonic vibrations and potential PT damage identification.

The idea behind UT vibrations is the same as conventional forced vibrations, only with at a higher
frequency range. The frequency range can be as high as 20 KHz. Because of this, the frequency
scaling from low frequencies (where lower global modes are excited) to high frequencies (where
high local modes are excited) holds great promise. Spatial resolutions can be as large as the whole
bridge volume and as small as 1.00 mm, which represent smaller size corrosion or fatigue cracks.
A careful design of SHM experiment can reveal general changes in dimensions and geometry. The
existence of damages can also be revealed. To evaluate exact locations of small damages and size
of such damages, pulsers might have to be close in distance to each other, or other types of sensors,
such as AE, might be needed.
7.4.2.7 Laser-Based Ultrasound (LBU)

Conceptually speaking LBU is fairly similar to using piezoelectric transducers in a pulse-echo UT
setup. The main differences are that the UT waves are generated using impacting laser. Thus, LBU
combines two physical phenomena: ultrasound and optics. It is based on generating ultrasound waves
in solids by impacting the solid with a pulsed laser source. The laser source will generate UT waves in
the solid by either (a) heating the impacted portion of the solid, where the sudden thermal expansion
will cause an outward propagating waves, or, (b) the sudden inertial movement caused by the impact-
ing laser would also cause outward propagating waves, ablation process. The UT waves include com-
pression, shear and Rayleigh surface waves in the test object. They can then be received by adequate
receiver, such an adaptive interferometer. The motion can reveal defects in the solid as in conventional
pulse-echo process (Figure 7.22). The efficiency of detection of LBU is much less than similar conven-
tional transducer detection, however, laser is very efficient generator of UT (see Figure 7.23).
Among the potential advantages of LBU in civil infrastructure is that it can scan larger surface
areas than the stationary traditional transducers. The relative costs (both generators and receivers)
are more expensive that the traditional transducers. Note that LBU has the flexibility of applying
the impacting laser; it can be applied as an array with varying energy intensity, in a phased-array
type setting. This will provide flexibility in DMID region. The LBU would generate broadband UT
frequencies up to 100 MHz, or even higher if needed.
Laser pulse Laser

source receiver Reflection from
1-The source sends surface
Reflection
a laser pulse
Amplitude from far end
2-The pulse reflects
through set of mirrors
and impacts the surface. Mirrors Time
It reflects from the surface,
t1
and also passes through it Plate t2
4-Both signals are

3-The pulse then detected by the laser
reflects from the receiver
far end of the plate
FIGURE 7.22 Concepts of laser-based ultrasonic (pulse-echo setup).
Input laser source
Motion detector
Acoustic wave
The impacting laser

will generate acoustic
propagation due to...
Acoustic stresses
due to thermal
expansion
Acoustic stresses
due to inertial
motions
FIGURE 7.23 Laser impact generation of acoustic waves in solids.

TABLE 7.5
Comparison between Some Ultrasound Testing Methods
Impact Guided (Lamb) Laser-based
Issue Pulse Echo Echo Ultrasonic Vibration Waves Ultrasonic (LBU)
Type of damage? Cracks, debonding. Also Cracks, debonding. Cracks, debonding. Also properties
properties such as thickness Also properties such such as thickness or mass density
or mass density as thickness or mass
density. Smaller
frequencies can
detect global changes
in the structure
Size of flaws (mm) 0.01 O (10.00) 1.00 0.2 Less than 0.01
In-situ? Yes
Predominant waves Propagating: Normal Propagating/standing Propagating: Propagating:
during test (compression/tension) Lamb, normal, lamb,
flexural, shear, Rayleigh,
Rayleigh flexural, shear
Applications in civil infrastructure include weld or other defect identification, surface delamina-
tions of FRP wrappings in reinforced concrete (RC) installations. It can also be used for measuring
thicknesses and temperature responses.
Some general advantages of LBU are that they (a) can be applied remotely, (b) can be used in
hostile environment, (c) very flexible for complex geometries, (d) can provide very high resolution,
(e) suitable for real time measurements: for example, can monitor real time responses and damages
due to truck motions on bridges, and (f) can be used for curved surfaces (Table 7.5).
7.4.2.8 Applications in SHM

7.4.2.8.1 General
Ultrasonic NDT techniques can be utilized in many ways in SHM field. Table 7.6 summarizes some
of the potential ways of these applications.
7.4.2.8.2 Choice of Appropriate UT Method

Sometimes in the field of SHM, we are faced with a dilemma: two available choices, with oppos-
ing advantages and disadvantages. The Heisenberg principle as applied to signal processing (where
higher resolution of frequency sampling is only possible with lower time sampling resolution and
higher time sampling resolution is only possible with lower frequency sampling resolution*) is a
good example of such a dilemma. Another example of such a dilemma is in the field of UT measure-
ments of defects (mainly in piping systems).
Consider, for example, the UT system presented by Owens et al. (2005). It is a high frequency
system (up to 1.0 MHz), that is capable of exciting longitudinal, flexural or torsional waves in the
pipes. The echo of those excitations can reveal location of defects along a pipe length of 155 in, as
reported by the authors. The advantage of the use of high frequency is that smaller defects can be
observed. They reported as small defects as 0.36% CSA (cross-sectional area). Higher sensitivity of
detecting defects can safeguard against brittle failures in pipes.
On the other hand, Demma et al. (2005) reported a low frequency (in the range of 0–100 kHz) UT
guided wave system that can detect CSA as low as 5%. The main advantage of such a system is the lon-
ger distance of the pipe that the system can cover. The authors reported a distance as long as 100 m. A
comparison between the two approaches is shown in Figure 7.24. As mentioned above: the dilemma that
* For more discussion, see Section 7.7 on signal processing in this chapter. Also refer to Hubbard (1998).
TABLE 7.6
NDT Ultrasonic Utilization in SHM Civil Infrastructure Field
Virtual Sensing Paradigm
Method Direct Sensing (VSP) Remote Sensing Local vs. Global—Scaling Issues Cost–Benefit Issues
Impact-echo Mostly direct sensing local Results can be easily used NA Potential for efficient scaling from Labor intensive. Cover
capability with other STRID local to global SHM is minimal smaller areas. Mostly local
techniques to provide due to the labor demands of this applications
DMID insights in a VSP technique
Pulse echo setting The techniques can be used In a well-designed experiment or Can provide cost–benefit
Guided waves Provide local and in remote sensing settings SHM setting, the potential of advantages in well-
Damage Identification (DMID)
(General) semilocal direct sensing (see Chapter 5) spatial scaling (local to global) designed SHM settings
capabilities exists
Guided waves A series of phased-arrays probes Given the potential of local
(phased arrays) that are places optimally on (or to global scaling, this
inside) that structure can scale technique can be optimized
time and space demands, thus to provide optimal
providing both local and global cost–benefit returns. In real
identification potential within time, or near real time
single setting SHM settings
Ultrasonic vibrations Works well in a local Results can be easily used No direct uses exist as of The potential of scaling the time If used in a local-global
setting. Can be expanded with other STRID the writing of this volume. (frequency) scale so as to cover setting, the cost–benefit of
to larger domains when techniques to provide The potential of using the both local and global SHM issues this method becomes more
using several online DMID insights in a VSP technology in remote exists attractive
probes. In such a case this setting sensing does exist
technology can scale
from local to global SHM
domains
Laser-based Has the potential of Using laser as an ultrasonic Offer potential of spatial scaling As the costs of equipments
ultrasonic (LBU) covering large area of exciter and high (local-to-Global) SHM settings are reduced, the cost benefit
detection when used in a resolution photography as of this technology becomes
remote sensing setting a sensing mechanism can more attractive
provide remote sensing
capability
399
Transducer
(a)
Inspected
region
Transducer
array
(b)
Inspected
region
FIGURE 7.24 Two piping inspection solutions: (a) high accuracy but short distance, (b) lower accuracy but
longer distance. (From Owens, S. et al., A new high frequency long range ultrasonic guided wave inspection
system for pipes, Proceedings of the 2005 ASNT Fall Conference, Columbus, OH, 2005. With permission.
Reprinted from ASNT Publication.)
faces the decision maker is: shorter distance and higher accuracy, or longer distance, and lower accuracy.
The answer, of course depends on the situation on hand. For (a) brittle pipes, (b) pipes that carry hazard-
ous material, or (c) newer pipes where the defect sizes tend to be small, it would be advisable to use the
higher frequency approach; otherwise, the lower frequency approach would be adequate to use.
Since this document is interested mainly in bridge structures, it is of interest to consider potential
situations where the above examples might be applied to SHM of bridges. We note that piping systems
in general are simple (mostly cylinders) and uniform (mostly single material) constructs, especially when
compared with the complex bridge constructs. The high frequency but short distance UT testing is an
NDT-type solution. Such solutions were discussed in many parts of this document. Their usability in an
SHM project is limited by the fact that the approximate location of the damage must be known in advance.
On the other hand, the low frequency but long distance UT approach is more suited for SHM projects. It
reduces the need for knowing, a priori, the location of the damage. Unfortunately, because of the com-
plexity of the bridge constructs, the needed technology for such a solution is not available. We expect that
if/when such technology becomes available, it can be used first in simpler bridge constructs.
7.4.3 Thermography
7.4.3.1 Overview
Measuring the manner structural temperature changes, or heat flow to, from or through materials or
structures, is the basis of thermography-based NDT. Heat flow is related to material and structural
properties. So, by observing how heat flows, it is possible to obtain information regarding the state
of the structure. Such information can be, but not limited to, surface-connected flaws, internal flaws,
measures of thickness or densities, mechanical or thermal properties, and certain surface properties
such as delamination, see ASNT (1999). Thermography has shown, and holds, great promise in the
field of NDT and SHM for civil infrastructure. This section presents some of the basics of thermog-
raphy, and its uses in SHM for civil infrastructure.
7.4.3.2 Categories of Thermography

There are several was to categorize thermography (Figure 7.25). Methods of heating, methods of
sensing and heat flow mechanisms are three basic categories.
Categories of thermography
Method of heating Method of sensing Heat flow

mechanism
Active Contact
Conduction
Passive Noncontact
Radiation
(remote
sensing)
Convection
FIGURE 7.25 Categorization of NDT thermography.
Method of Heating: Temperature rises, or drops, in a structure can result either naturally (pas-
sive) or from an external source (active). Passive temperature stimulation can occur in manufactur-
ing processes or by simple exposure to atmospheric heat. Active temperature stimulation occurs by
subjecting the system of interest to a heating source. Both types of heating have been used in SHM
projects.
Mechanisms of Heat Flow: There are three basic methods temperature (or heat) can flow: conduc-
tion, radiation, or convection. Conduction heat flow occurs when heat flows from warmer to cooler parts
of materials. Radiation occurs because warm surfaces generate MW that radiates from the warm sur-
faces, De Witt (1988). Such waves travel in near-infrared wave lengths. Infrared radiated waves behave
similar to visible light waves, except for the differences in wave lengths and frequencies. Because of
this, it can be captured by special infrared cameras. Convection occurs when heated mass is transferred.
Conduction and radiation have many applications and uses in SHM while convection has limited use.
An important metric in the field of thermography is the emissivity of the surface ε. Emissivity is
a measure of the efficiency of the surface to radiate or absorb electromagnetic radiation. A perfect
absorbing, or radiating, surface is when ε = 1.0. Thus, 0.0 ≤ ε ≤ 1.0. It is desired to have as high
emissivity as possible during the test. To improve the accuracy of an infrared imaging device in
detecting thermal radiation, it is important for the surface to be clean and free of foreign material.
Table 7.7 shows emissivity of some engineering materials.
Method of Sensing: Methods of sensing temperature flow are as varied as the mechanisms
of heat flow. Sensors can be placed directly on the object (contact). Sensing temperature can also
be done remotely (noncontact). Remote sensing thermography is of special interest in the field of
SHM because of all of the advantages that remote sensing offers (see Chapter 5). Figure 7.26 shows
remote sensing steps when utilizing thermography method.
7.4.3.3 Geometric-Based: Heating Methods

When the temperature of the system is stimulated actively, the spatial (geometric) stimulated sur-
face varies as follows:
Point Temperature Stimulation: a single point on the surface is stimulated. This is generally done
by a laser point device. This technique is similar to A-Scan in UT NDT.
Line Temperature Stimulation: a line on the surface is stimulated. This is generally done by a
linear florescent lamp. This technique is similar to B-Scan in UT NDT.
Area Temperature Stimulation: an area on the surface is stimulated. This is generally done by a
lamp, or set of lamps. This technique is similar to C-Scan in UT NDT. In SHM applications, this
method can be most efficient, since it cover more area on the surface of interest.
TABLE 7.7
Emissivity of Some Popular Materials
Material Emissivity ε
Polished steel 0.03–0.08
Dull steel 0.08–0.20
Steel mill products 0.15–0.25
Heavily oxidized and rough iron or steel 0.60–0.85
Bright brick, tile of plastics 0.80–0.95
Concrete 0.92
Wood 0.90
Sand 0.90
Source: From Botsko, R., Introduction to ultrasonic testing,

Section 13, Part 2 in Nondestructive Testing Handbook:
Vol. 10, ANST, Columbus, OH. Reprinted from ASNT
Publication.
Geometric-based:
heating the target
Sensing:
IR cameras
Time-based:
Infrared
application methods
thermography
FIGURE 7.26 Steps of remote sensing infrared thermography.
7.4.3.4 Time-Based Thermal Application Methods

The temporal wave form of the thermal stimulation is also of interest when the temperature of the system
is stimulated actively. The different possible input temporal heat wave forms are described in Table 7.8.
7.4.3.5 Sensing
7.4.3.5.1 Contact Sensing
Contact thermometry has many uses in numerous fields such as manufacturing and processing. In
SHM contact thermometry is used primarily to measure ambient temperature that can be important
to correlate with such hazards as corrosion, fatigue, or thermal behavior of the global structural sys-
tem. For these general uses, thermoelectric devices such as thermocouples would provide adequate
performance.
7.4.3.5.2 Noncontact/Remote Sensing
Noncontact (remote sensing) thermal sensors vary in their temperature range resolution as well as
distance from the test object. Hand-held thermal scanners offer the portability advantage with the
disadvantage of low imaging quality. Perhaps the most popular thermal imaging tool is the high
resolution infrared cameras, Hardy and Bolen (2005). They tend to offer high imaging resolution,
TABLE 7.8
Temporal Wave Forms of Temperature Stimulants
Method Description
Pulse The stimulus is applied with a very short time duration. The decay properties of the
temperature are measured and the needed information is deduced from this information,
Shepard et al. (2001)
Step The stimulus is applied with a finite duration. The rise of temperature is observed, and the
needed information is deduced from this observation, Maclachlan-Spicer et al. (1991)
Lock-in thermography: The thermal waves are generated and detected remotely. The waves are harmonic. The
modulation heating/ frequencies can be changed such that maximum efficiency of detection can be had,
phased pulses Maldague (2002).
Vibration/mechanical Vibrothermography induces heat at defects (by friction, for example) by externally vibrating
induced heating the system at certain driving frequencies. By observing the heat rise, the defects can be
detected. See Moore (2001) and Maldague (2002)
with wide range of temperature range. Infrared cameras also have good response time that makes
them versatile for measuring fast and slow changing temperatures.

Infrared thermographic NDT techniques can be utilized in many ways in SHM field. Table 7.9 sum-
marizes some of the potential ways of these applications. Thermography has been used in several
SHM applications. Figure 7.27 shows results of thermographic inspection of the surface of an FRP
bridge deck. FRP has also been used to retrofit bridge beams and columns. Thermography was used
to monitor status of FRP wrapped bridge beams (Figure 7.28) and columns (Figure 7.29). Table 7.10
illustrates the applicability of thermography techniques in infrastructure applications.
7.4.4 Electromagnetic Methods

7.4.4.1 Overview
Electromagnetic NDT methods utilize the interaction between the electromagnetic flux and waves and
the flaws in the test object. Numerous electromagnetic NDT methods are used. Among the more popular
methods are: Eddy current (EC), static magnetic field, magnetic particle, and microwaves (radar) tech-
niques. Some basic properties of the four methods are shown in Table 7.11. We also discuss the prop-
erties of these methods as they relate to SHM field in Table 7.12. Figure 7.30 shows different aspects
of each method that need to be considered while studying the applicability of the method for use in an
SHM project. The rest of the section will explore those methods in more details, especially as they relate
to SHM field. Note that there are other types of electromagnetic testing such as electric current perturba-
tion (ECP) method that can detect damage in different metals such as aluminum alloys, Beissner (2005).
Another modern technique is the megaabsorption technique, Rollwitz (2005a).
7.4.4.2 Eddy Current

7.4.4.2.1 General
Eddy current is a popular NDT method for detecting surface or near surface anomalies (cracks
delaminations, etc.). It is also used to detect thickness of objects as well as some material properties
such as conductivity, corrosion, and permeability. EC method is based on the following steps
1. An alternating current from an electrical circuit is induced in a circular coil in a portable
probe (Figure 7.31).
2. The current will induce a voltage field in the nearby test object.
TABLE 7.9
NDT Thermography Utilization in SHM Civil Infrastructure Field
Direct Virtual Sensing Remote Local vs. Global—
Method Sensing Paradigm (VSP) Sensing Scaling Issues Cost–Benefit Issues
Conduction Yes Yes No Mostly local Can be useful for SHM
experiments for correlation
purposes, e.g., fatigue,
corrosion, thermal expansion
problems, etc.
Convection Yes NA No NA
Radiation Yes Yes Yes Can be global and/ High potential for a higher
or local benefit to cost ratio
FIGURE 7.27 Thermographic inspection of FRP bridge deck. (Courtesy of New York State Department of
Transportation.)
FIGURE 7.28 Thermographic inspection of FRP bridge wrapping of beams. (Courtesy of New York State
Area of
delamination
25.0 °F 35.0 °F
FIGURE 7.29 Thermographic inspection of FRP bridge wrapping of columns. (Courtesy of New York State
TABLE 7.10
Applicability of Thermography Techniques to SHM Hazards
Hazard Comments
Fatigue Can be used on a local level for surface cracks
Corrosion—General Can be used for surface rust-produced cracks
Corrosion—PT/PS Not promising due to the relative embedment depth of the PT tendons
Scour NA
Earthquakes Can be valuable in detecting near surface damages on a global scale after seismic events
FRP wrapping Very useful due to the surface nature and small thickness of the wrapping materials
FRP decks Very useful for skin top (or bottom) plates
TABLE 7.11
Comparison between Different Electromagnetic Testing Methods
Static Magnetic Electromagnetic Waves
Issue Eddy Current Field Magnetic Particle (Microwaves)
Type of Surface or subsurface Surface or Surface flaws Surface, subsurface, interior,
damage flaws subsurface flaws or length of cables or solids
Size of flaws Up to 0.2 mm 0.03 mm Up to 0.5 mm 10–3 mm

Pretest None. Requires Require inducing Extensive None
preparations reference samples magnetic field in
the test object
Post-test None Require Extensive None
preparations demagnetizing the
test object
In-situ? Yes, manual. Can be Can be difficult Yes, mostly manual Can be automated
automated
Detection Computing electrical Measure magnetic Visual Measuring electromagnetic
methods impedance properties field perturbations wave reflections and scatter
can reveal damage states
TABLE 7.12
Electromagnetic NDT Technical Issues as Applies to SHM
Electromagnetic
Technical Issues Effects in SHM
Type of material Materials are limited to ferromagnetic or metals, depending on the type of test. Electromagnetic
waves (microwaves) can be used with other materials such as plastics, concrete, soil, or elastomers
Global vs. local Some methods are strictly surface-based, or near surface-based, such as magnetic particles.
Microwaves can penetrate deep into materials, as well as travel fairly long distances, e.g., steel
cables
Surface preparation Some surface preparation is needed before and after the experiment for magnetic particle testing,
which would limit its use in general real time SHM setups
Type of surface Eddy current and magnetic particle need somewhat smooth surfaces. Electromagnetic waves testing
can operate on variety of surfaces
Electromagnetic/electronic testing: Overview
Basic Detection Geometrical

Other
methods method needs
Eddy current Digital Method Safety

dependency
Magnetic field Analogue One-side Cost–benefit
Magnetic Suitability to
Visual
particle SHM
Microwave Analysis
FIGURE 7.30 Overview of electromagnetic/electronic testing.
Probe Probe
Test object
High frequency: Low frequency:

shallower signal across deeper signal across
test specimen test specimen
FIGURE 7.31 Effects of frequency on Eddy current signal depth.

3. The circular nature of the coil in the probe will cause electrons in the test object to move in
an eddy-like circular fashion. The electrons will in turn generate a circular magnetic field
in the test object.
4. The circular magnetic field can then be detected and measured in real time. The detection
can be either by the driving coils themselves or by another set of coils in the probe.
The depth of generated magnetic field is a function of the test object material properties, such as
geometry and lift off electrical properties. It is also a function of the excitation frequency. For
low frequencies, the detection depth is large and as the frequency increases, the depth decreases
(Figure 7.31). The depth of EC penetration S of an object is computed by
S = (p f m m0 s)−0.5 (7.2)
In Equation 7.2 the driving frequency is f, the test object electric conductivity is σ, the relative mag-
netic permeability is µ, and the permeability of free space is µ 0

The probes in EC testing can be either single or double probes. The single probe function is illus-
trated in Figure 7.32. As the probe moves along the surface of the test object, the measured volt-
age is recorded on display. The real part of the voltage (resistance) is represented by the horizontal
axis, while the vertical axis represents the imaginary part of the voltage (inductance reactance).
As the probe passes over a defect, going from point 2 to 3, the voltage changes as shown in the
image. When the probe moves past the defect, going from point 3 to 4, the voltage traces its foot-
print, as shown in Figure 7.32. Finally, the voltage reverts back to its baseline at point 5, away
from the defect.
A more detailed image can be produced by using a two-probe setup as shown in Figure 7.33. As
the double-probe passes over a defect, going from point 2 to 3, the voltage changes as shown in the
image. Since each of the probes is recording the voltage with slight space shift (as the first probe
moves into the defect the second probe is still in pristine material), a hysterics in the voltage space
is formed. Thus, when the probe moves past the defect, going from point 2 to 3, the voltage does not
trace its footprint, as shown in Figure 7.33. When the double probe continues its movement across
the defect, going through points 3-4-5, an antisymmetric hysteresis is formed. This type of signal is
Lissajous* signal.
The Lissajous signal serves more than just providing additional image details. It was found that
the signal characteristics can provide information as to the type of defect. Thus, by having a base
line of defect characteristic images, it is possible to discern not only the presence of damage, but
also the type of damage that is being observed. Figure 7.34 shows different signals that vary accord-
ing to the type of damage being detected.
7.4.4.2.3 Advantages–Disadvantages in NDT Applications
The advantages of EC method are as follows:
• Very simple and accurate to detect surface defects

• Most efficient in metallic materials (good electrical conductivity)
• Works well for carbon FRP materials
• Near contact between the probe and the test object is not needed
• Efficient locally
* Joules Antoine Lissajous (1857) studied the family of curves that are represented by x1 = A1 sin(a 1 x + Δ ) and
x2 = A2 sin(a 2 x ). The shape of the curves depends on the parameters, α 2, A1, A2, and D.
1—As probe moves across

the defect
Probe
1 2 4 5
Test specimen
3
Defect
imaginary part
(inductance/
reactance)
Voltage,
2—The loci in the impedance
plane moves accordingly: first
3 the approaching signal 1 to 2 to 3
2
1 Voltage, real part

(resistance)
imaginary part
(inductance/
reactance)
Voltage,
3 4 3—Then the receding

signal 3 to 4 to 5
5 Voltage, real part

(resistance)
FIGURE 7.32 Generation of impedance curves of a defect: single probe.
The disadvantages are as follows:

• Somewhat limited depth
• Sensitivity is dependent on direction of cracks. Maximum accuracy when the crack is
normal to flux lines
• Not global
• Not sensitive for nonconductive materials
7.4.4.3 Magnetic Particle (MP)

7.4.4.3.1 General
Fine ferromagnetic particles are first spread over the test object. When these particles are magne-
tized, the pattern will follow the magnetic field as shown in Figure 7.36. If there is a surface, or
near surface defect or crack that is aligned roughly normal to the magnetic flow, the pattern will
show visually; thus revealing the presence of defects. Obviously, this NDT method is capable of
detecting defects in a visual manner. The extent or depth of such defects might not be detectable by
this method. The method of application is shown in Figure 7.37. Figure 7.38 shows a plate with a
preinserted longitudinal crack; the crack is not visible. After applying the magnetic field process, the
crack becomes visible as shown in Figure 7.39.
The needed tools for this method are (a) the ferromagnetic particles themselves that are needed to
spread over the test object, and (b) the yoke that is needed to magnetize the particle. The yoke, of
course will need power supply for operations.
1 - As the double probe moves across

the defect
Double
probe
1 2 4 5
Test specimen
3
Defect
imaginary part
(inductance /
reactance)
Voltage,
2 - The loci in the impedance
plane moves accordingly: first
2 the approaching signal
1 to 2 to 3 (note that loci at
1 location 3 is same as 1)
Voltage, real part 3
(resistance)
imaginary part
(inductance /
3 - Then the receding signal

reactance)
3 to 4 to 5
Voltage,
(note that loci at location 5

is same as 1 and 3)
2
Lissajou
s sig
1 5 na
Voltage, real part 3
l
(resistance)
4
FIGURE 7.33 Impedance curves of a defect: double probe (Lissajous signal).
(a) Hole in a plate or a tube (b) Surface crack (c) Corrosion
FIGURE 7.34 Patterns of Lissajous signals as a function of type of damage (a) pattern for a hole in plate or
tube, (b) pattern for surface crack, and (c) pattern for corrosion.

The advantages of MP method are as follows:
• Very simple to detect surface defects

• No need for electrical circuits
• Ferromagnetic materials only, such as steel

• Needs surface preparation before and after the test
• Direction of defects needs to be very close to normal to the magnetic flux
• Only surface defects
Flux density
Virgin magnetization
– +
Magnetic field strength
FIGURE 7.35 Magnetic hysteresis.
Electric current
– +
Magnetic particle
Yoke surface treatment
Magnetic field
Testing object
Surface flaw
FIGURE 7.36 Components of magnetic particle testing.
FIGURE 7.37 Magnetic particle yoke and particles in use. (Courtesy of Nondestructive Evaluation Labaratory,
FHWA.)
FIGURE 7.38 Magnetic particle testing, baseline test object. (Courtesy of New York State Department of
Transportation.)
FIGURE 7.39 Magnetic particle testing, surface crack revealed. (Courtesy of New York State Department
of Transportation.)
7.4.4.4 Static Magnetic Field (SMF)

7.4.4.4.1 General
Unlike MP method, in SMF the test object itself is magnetized, see Beissner (2005). The normal
magnetic field lines are interrupted by the presence of a flaw in the test object, thus creating a flux
leakage. Figure 7.40 shows the basic concepts of static magnetic field NTD with and without flaws.
A scanning tool is then used to detect the state of magnetic field in the object. If there is a defect
near the surface, it will cause disturbance to the field and the scanner will react accordingly. During
the relationship between magnetic field strength and flux density can be displayed (Figure 7.35). The
shape of the hysteresis can give insights on location, type, and size of the defect in the object. The
method can be used to detect flaws, as well as mechanical properties (such as material hardness) in
ferromagnetic materials (see Jiles 1986). The technique was applied in several SHM applications
such as detecting cracks in steel rebars, see Klausenberger and Barton (1977), and Beissner and
McGinnis (1984).
Magnetic lines
North South
Test object without flaw
Magnetic lines
Flux Leakage
Surface
flaw
North South
Test object with flaw
FIGURE 7.40 Basics of static magnetic field method.

The advantages of SMF method are as follows:
• Can be applied to whole objects with varying sizes and shapes

• Can detect both material properties and flaws in an SHM test
The disadvantages of the method are as follows:
• Results are mostly qualitative

• Requires near contact between the scanner and the object
• In-field magnetization can be difficult
7.4.4.5 Electromagnetic Waves: Microwaves or Radio Waves

7.4.4.5.1 General
Microwaves exist in frequency range of about 300 MHz to about 325 GHz, which is an equivalent
to a wave length range of about 3 × 109 to 105 nanometers (Figure 7.8). These waves are also known
as short-pulse radar, impulse radar, or ground penetrating radar (GPR). Radar wavelengths usually
exist in the higher range of the microwave bands. Both microwaves and radar waves have been used
extensively in NDT field, specifically; they have numerous applications for both NDT and SHM in
civil infrastructure, see Alkhalifah (2005).
Microwaves can be generated in several forms. Among these forms are pulsed, constant frequen-
cies, or modulated frequency. They are aimed at the test object, and upon interaction with it, they
would reveal needed information about the state of the object. They interact with the object in one,
or more, of the following forms:
Scattering: Microwaves scatter from discontinuities or other boundary conditions can reveal
important information.
Reflection: When the microwaves reflect from surface, the reflected signals can reveal flaws, and
delaminations. This is fairly similar to the pulse-echo technique in ultrasound.
Radiation: Microwave can radiate/attenuate through solids, such as pavements, or soils.
Information regarding composition of the interior of the test object can be revealed by measuring
this radiation, or attenuation.
In addition to the above-mentioned uses of microwaves, they can also be used to measure thick-
ness of objects, discontinuities, cracks/flaws, moisture, and material anisotropy. They can also be
used in the more important stress-corrosion measurements; such an application is important for high
strength cables (suspension bridges, cable stayed bridges, and prestressed/post-tensioned bridges),
see Rollwitz (2005b).

The tools needed for a microwave testing include (a) a microwave generator, (b) antennas (transmit-
ting, and/or receiving), and (c) a phase detectors (receiver).

The advantages of SMF method are as follows:
• No need for contact with test object

• No safety concerns
• Have many applications in civil infrastructure
The disadvantages of the method are as follows
• Results can be difficult to interpret

• Equipment sizes vary, and can be unwieldy

7.4.4.6.1 General
Electromagnetic NDT techniques can be utilized in many ways in SHM field. Table 7.13 summa-
rizes some of the potential ways of these applications. Table 7.14 illustrates some potential uses of
Electromagnetic NDT in detecting the effect of hazards.
7.4.5 Penetrating Radiation

7.4.5.1 Overview
Penetrating radiation is an attractive type of testing for civil infrastructure, but with severe limi-
tations. Different issues that involve these classes of tests are shown in Figure 7.42. These tests
involve subjecting the tested object to a type of radiation waves, as shown in Figure 7.41. The
radiation waves will either pass through the object, or backscatter from the object. The passed
or scattered radiation waves can then be interrogated for information regarding the test object.
Generally speaking, there are three types of radiation waves that are used in testing: X-rays,
gamma rays, and neutron beams. X-rays are generated by subjecting a metal object to high-
energy electrons (Figure 7.41). This will generate photons that will propagate in a predetermined
direction through the tested object. Gamma rays are generated from the nuclei of the atom. They
are emitted from unstable radioactive materials, such as cobalt, cesium, or thulium. Neutron
beams are generated from reactors or isotopes. They tend to produce better resolution than X-rays
or gamma rays. They also penetrate deeper into solids than other methods. Table 7.15 shows the
TABLE 7.13
NDT Electromagnetic Utilization in SHM Civil Infrastructure Field
Local vs.
Direct Virtual Sensing Global—Scaling Cost–Benefit
Method Sensing Paradigm (VSP) Remote Sensing Issues Issues
Eddy current (EC) Yes As of the writing Even though all Only local, due to Average for local
of this volume, these methods do the surface, or detection
Magnetic particle Yes there is no body not require direct near surface, Average for local
of work that contact with the nature of the detection. Covers
relates any of test object, they method larger area than the
these methods to require close EC, but requires
damages within proximity. Thus, more effort in
the object in a their usefulness surface preparation
Static magnetic Yes VSP mode in remote sensing Due to practical Can be expensive
field (SMF) applications is pre- and for SHM
limited postpreparations applications
issues, this
method can be
very difficult to
scale from local
to global SHM
Microwaves/radar Yes Fairly suited to Given the For appropriate
remote sensing radiation cases, this method
property of can provide
microwaves, this attractive
method has a cost–benefit value
potential of
scaling local to
global detection
of state
TABLE 7.14
Applicability of Different Electromagnetic Techniques to SHM Hazards
Hazard Applicable Technique Comments
Fatigue EC, microwave, magnetic All electromagnetic methods can detect fatigue cracks. The
particle, and SMF optimal method will depend on geometry, and extent of expected
damage. Careful decision making analysis is needed to identify
the optimal method
Corrosion—General Microwave inspection Stress corrosion can be detected by microwaves, Rollwitz (2005b)
Corrosion—PT/PS
Scour Microwave (GPR) Ground penetrating radar might be suited to detect scour damage
Earthquakes EC, microwave, magnetic All methods can be used locally to detect earthquake damage.
particle, and SMF Radar waves have been used remotely to inspect postearthquake
damage, Housner and Masri (1996)
FRP wrapping EC Limited on to carbon FRP
FRP decks
Bridge security Microwaves Radar waves might be used remotely to detect post-event damages
TABLE 7.15
Differences between X-Rays and Gamma Rays (Figure 7.43)
Issue X-Rays Gamma Rays
Nature during testing Generated by X-ray machines during tests Naturally emitted
Projected power Vary Constant
Generation High-energy electrons impacting metallic Unstable radioactive isotopes
objects and generating photons
Useful life Long Limited by the half-life of the isotope
Wave length See Figures 7.8 and 7.9
basic differences between X-rays and Gamma rays from testing view points. Other types of pen-
etrating radiation testing are beyond the scope of this document, the reader should consult Greene
(2005), Emigh et al. (2007), and Bossi et al. (2007).
Penetrating radiation testing is attractive for NDT since the X-rays or the gamma rays can pen-
etrate through the object, or backscatter from the object with minimal loss of intensity leading to
accurate identification of small flaws in the object. Generally speaking, the intensity of X-rays can
be controlled such that it can be used to test thicker and denser objects. Gamma-ray intensity is
constant, so the size of the test object can be limited.
The setup of penetrating radiation testing is fairly simple (see Figure 7.41). There are three basic
components of the test: the radiation-generating devices, the testing object itself, and the colleting
devices. In addition, some testing methods rely on computational processing components.
The generating devices for X-rays vary in size and portability. The stationary machines are suited for
NDT of relatively small objects. However, they are not suited for in-site SHM projects. Portable X-ray
machines can be used in site, however, safety precautions must be taken. The energy demands of X-ray
machine also generally vary in the range of 150–450 kV. Higher energy X-ray machines can generate
more than 450 kV. As mentioned before, gamma-ray machines uses constant energy input. They house
the isotope (radiating material) inside a shielded, S-shaped housing. When needed for use, the isotope
is pushed out of the housing for the prescribed period of time, then pulled back for safe keeping.
The objects that are suitable for penetrating radiation testing, in the field of civil infrastructure,
can be steel, reinforced concrete, or FRP constructs. Both X-ray and Gamma-ray methods can
be used with all of those materials. Types of damages that can be detected include voids, cracks,
delaminations, rusting, thickness, or density.
After the radiating waves pass through the tested object, or scattered back from the tested object,
the intensity of these waves is then collected and processed using different techniques. The oldest
technique and most popular is to use radiographic film. The films are covered with thin layers of
chemicals that changes when subjected to the photons emanating from X-rays. Such changes are
proportional to the intensity of the photons. This intensity is function of the internal composition of
the test object. If there are any flows, defects, or any other changes in the composition, such changes
will affect the exposed film. Damages can be identified by observing the film. Radiographic film
collectors are suitable for static tests: no time-dependent observations can be made.
CRT (TV screens) is also used to observe the intensity of X- or Gamma rays after passing through,
or scattering back from, an object. Such a method is also called radioscopy. It has the advantage of
measuring dynamic, even real time, tests.
In addition to the above visual collecting procedures, more complex analytical methods were
developed to infer even more information about the composition of the tested object. These analyti-
cal methods process the digital information about backscatter, or passed radiation intensity, then
compute more accurate representation of the internal composition of the tested object. The two
main computational methods are computed backscatter (CB) and computed tomography (CT). Note
Source of
Schematics of penetrating
rays
(or scatter) radiation testing
Incident
rays
Backscatter
rays
Test object Collecting media

(radiation film, etc.)
Penetrating
rays
FIGURE 7.41 Elements of penetrating radiation testing.
Penetrating radiation testing: Overview
Radiation Detection Geometrical Other

type method needs
X-rays Digital 360 Degrees Safety
Gamma rays Analogue One-side Cost–benefit
Neutron Backscatter Suitability to

beams tomography SHM
Computed
tomography
FIGURE 7.42 Penetrating radiation testing overview.
also that CT is sometimes known as computer aided tomography (CAT). These methods will be
presented in more detail in the next section.

Penetrating radiation testing has been used extensively for NDT applications. As usual, NDT appli-
cations involve laboratory testing or for spot testing in site. For those situations, penetrating radiation
techniques offer unmatched accuracy for detecting small flaws, delaminations, or cracks. Changes
in density detection accuracy make it ideal for detecting rusting or corrosion advances deep within
reinforced concrete components. Using CT can result in a 3D representation of the whole object. In
addition to the accuracy advantage, there are several other basic advantages to penetrating radiation
technologies. Both X-rays and gamma rays travel in straight lines and are not sensitive to magnetic
or electrical fields; yet they can be scattered at surfaces of objects or edges of flows.
2 - ...The resulting resolution

from Gamma ray tests is
better than those resulting Gamma rays results
from X-rays tests
Hidden damage inside

a structural system
1-Because wave length of

X-rays results Gamma waves are shorter
than wave length of X-rays...
Gamma rays in
an NDT for detecting
hidden structural
damage
X-rays in an NDT for

detecting hidden
structural damage
FIGURE 7.43 Comparison between X-rays and Gamma rays in NTD tests.
Unfortunately, such advantages come with steep price. CT, radioscopy, or radiographic films
need access to at least front and back side of the test object. This imposes immense practical limi-
tation for on-site testing. Another practical limitation is the size and portability of the projectors.
Perhaps the most important limitation to this class of methods is the safety concerns. Heavy shield-
ing from the radiating waves is needed to protect workers. All of those limitations add to large
cost–benefit ratio.
7.4.5.2 Computed Tomography

Inside the laboratory or in situation when the radiation generator can rotate around the test object,
CT can be used to generate a digital illustration of the state of the interior of the object in a given
plane. By moving the plane of the radiation generator along the test object, a 3D image of the test
object can be generated. The concept is shown in Figure 7.44. The concept of CT is based on the
absorption of an incident X-ray beam with an intensity of R0 after it penetrates an object to an inten-
sity of R. The incident and penetrated intensities are related by Lambert’s Law
R = R0 e − m r (7.3)
With r as the traveled distance and µ as an attenuation coefficient of the test objet. The coefficient µ is a
material property. For cross-sections that contain different materials, such as steel rebars and concrete,
for example, the distribution of the attenuation coefficient in a cross-section is µ(x,y) that is a function
of the Cartesian coordinates of that cross-section x and y. Equation 7.3 can be generalized to
R = R0 e A (7.4)
r1
A = − ∫ m( x, y) dr (7.5)
r0

Pθ (X') = Projection along X'
X'
Parallel projection lines

Object
(2-D slice)
at angle θ2
X'
ng
alo
ion
ject Pθ (X') = Projection along X'
= Pro
Pθ (X') θ2
θ3
X'
Parall s
X' at an el project n line
jectio
gle θ ion li el pro
3 nes Parall le θ1
g
θ1 at an
X-ray tube sweeps X-ray tube sweeps

on a line normal to θ1 on a line normal to θ3
to generate a projection to generate a projection
on the other side of the on the other side of the
object object
X-ray tube sweeps After X-ray tube finish generating projection
on a line normal to θ2 at each angle θ, it moves to the next angle
to generate a projection
Basics of computed tomography (CT) on the other side of the
position, and start a new sweep, until
it finishes creating projections at every angle
object in a 360 degrees circle
FIGURE 7.44 Basics of computed tomography (CT).
The limits of the integral in Equation 7.5 range over the point of entry and the point of exit of the
beam. Operating on Equations 7.4 and 7.5 yield
r1
P = ∫ m( x, y)dr (7.6)
r0
P = ln( R0 /R ) (7.7)

The transform in Equation 7.6 of µ(x,y) is the Radon transform. The object of CT is to estimate µ(x,y)
given the test results that produce P at different stations around the test object. Several analytical
techniques have been proposed to estimate the transform of Equation 7.6. Among those techniques
are direct polar Fourier transform (FT). It is reported that such solution does not produce accurate
resolutions, Dennis (2005). More accurate resolutions for the transform can be obtained following
filtered back projection techniques, which utilizes filtering of the measured intensities by processes
of convolution; see Dennis (2005) for in-depth description of the method.
What concern us immediately is the potential uses of CT in SHM projects. Given the capabilities
of resolving interior states of solid such as steel or concrete, the process should be of great inter-
est. Since the radiation generator needs to revolve around the object, the method obviously is most
applicable to linear-type geometries such as columns, truss members, and suspension cables. Two
basic obstacles remain to be resolved:
• Physical setup: The mechanics of revolving of the radiation generator around the in-situ
test object and recording the data accurately needs to be designed.
• Safety: Penetrating radiation in the field can produce safety concern.
As of the writing of this volume, the above two issues have not been resolved in a practical sense.
However, we believe that the promise of CT method in SHM applications will prompt the resolution
of the above obstacles in a practical and acceptable manner.
7.4.5.3 Computed (Compton) Backscatter

When an X-ray, or gamma ray hit a target, photons will scatter from the target with a scattering
angle θ. This process changes the incident wave length of the incident waves λi into scattered wave
length λs. This effect was discovered by Arthur Compton in 1923. In 1927, he earned the Noble prize
in physics for his discovery. The relationship between the incident and scattered wave lengths are
h
ls − li = (1 − cos q) (7.8)
me c

Where h is Planck’s constant (h ≅ 6.626 × 10 −34 J. s); it is a measure of energy. Note also that c is the
speed of light. The wave length shift as predicted by Equation 7.8 is used to measure surface and
internal voids and anomalies in the tested object.
Again, we are mainly interested in the potential uses of CB in SHM projects. Given the capa-
bilities of resolving interior states of solids such as steel or concrete, the process should be of great
interest. The other feature of the CB approach in penetrating radiation testing is that, unlike CT, it
needs radiation exposure only from one side of the tested object, thus offering great practical advan-
tage. Because of this, the method is applicable to most infrastructure geometries. Again, two basic
obstacles remain to be resolved:
• Analytical Accuracy: The numerical resolution of backscattering problem for general,

online, and in-field setups are not popularly available as of the writing of this volume.
• Safety: Penetrating radiation in the field can produce safety concern.

Penetrating radiation NDT techniques can be utilized in many ways in SHM field. Table 7.16 sum-
marizes some of the potential ways of these applications. We only cover in-field, and near real time
SHM projects in this table. Table 7.17 illustrates some potential uses of penetrating radiation in
detecting effects of the hazards.
7.4.6 Other Methods

7.4.6.1 Visual Methods
Visual inspection is the simplest and most straightforward method of NDT. It is based on observing
the condition of the object of interest visually, and trying to detect its current state based on personal
experience and existing guidelines, if any. Visual NDT methods should be an integral part of any
NDT, or SHM project; they can save costs and time. They also can guide the progress of the project
and help in adjusting its direction midway, if any adjustments are needed.
The advantages of visual method are as follows:
• Very simple and fast

• Can scale from local to larger areas without any additional efforts
• Can apply remotely

• Qualitative
• Might need special equipment for access and other reasons
TABLE 7.16
Radiation NDT Utilization in SHM Civil Infrastructure Field
Virtual Sensing Remote Local vs. Global— Cost–Benefit
Method Direct Sensing Paradigm (VSP) Sensing Scaling Issues Issues
X-rays, Difficult, given VSP might be needed Can be used to Safety issues can be Cost–benefit
Gamma rays, safety and on a case-by-case detect state of of concern if applied issues need
and neutron physical sizes of basis if the testing is remote to more than careful studies
beams infrastructure performed on components. localized area before embarking
isolated or localized Safety on an SHM
components concerns can testing
be an issue
Computed Possible for Since the method Due to the Can be designed to Can be cost
tomography linear-type produces 3D images need to detect the state of effective if
(CT) geometries, such of the interior state revolve whole components, limitations are
as columns, of the test object, no around the such as whole resolved
truss members, need for VSP test object, column, or entire
and suspension applications. remote truss members
cables. Safety sensing in
concerns must traditional
be addressed sense is not
applicable
Computed Can have CB is a remote Can be designed to
backscatter numerous sensing detect the state of
(CB) applications in method, by entire components,
SHM. Safety default such as steel
and numerical connections, or
issues need to be bearing setup
resolved
TABLE 7.17
Applicability of Different RT Techniques to SHM Hazards
Applicable
Hazard Technique Comments
Fatigue CB Film radiography can be difficult in-situ
Corrosion—General CB/CT Depending of geometry of reinforced concrete, CT can be used, if safety can be
Corrosion—PT/PS CB/CT assured
Scour NA Due to presence of soil and water, penetrating radiation is not practical
Earthquakes CB/CT After earthquake event, CB or CT can be used to assess damaged connections
and welds
FRP wrapping CB/CT CB can be used to assess conditions of FRP wraps in most situations. CT can be
used for columns wraps, since it is possible to access the column from all sides
FRP decks CB The most effective technique is CB, since it is difficult access the deck from
all sides
7.4.6.2 Liquid Penetrant

This method is similar in concept to MP method. It is based on applying a liquid to the surface of the
object being tested. The liquid will penetrate the cracks (as small as 0.001 mm in width). After the
excess fluid is removed, a suitable powder coating is applied to the surface. If there are any cracks
in the surface, the residual liquid will interact with the powder coating and becomes visible causing
the crack to be detected.
There are several types of liquid penetrant, such as florescent and/or visible types. They can
be removed either by washing them (water), or by using solvents. The choice of the penetrant type
depends on several factors such as size of test area and the access. More details of the methods is
presented by Tracy (2007).
The advantages of liquid penetrant method are as follows:
• Very simple and quick to detect surface defects

• Applicable to most engineering materials
• Can be adapted to large as well as small surfaces
The disadvantages are as follows
• Surfaces need to be within reach

• Needs surface preparation before and after the test
• Inside of cracks needs to be clean, for example, corrosion free
• Porous materials need special procedures
• Only surface defects
• Might produce false indications of cracks for rough surfaces
7.4.6.3 Optical Methods

There are numerous NDT optical methods. We already discussed some laser-based techniques ear-
lier. Similar to the fiber optic sensors (Chapter 5), optical NDT methods are mostly based on optical
interferometry. Generally speaking, interferometry concerns itself with the interaction of two waves,
in both amplitude and phase. Interferometer is the tool that controls such an interaction. When the
resultant of the interacting two waves is known, it is possible to discern the attributes of one of the
original waves. At its simplest mode of behavior, when the original two waves are in phase, they
will reinforce each other; when the two waves are out of phase, they will cancel each other. Recall
that Fabre Grating Fiber Optic sensors operate on this principle, as explained in Chapter 5. By using
interferometer principles, state of strains (deformations) and displacements at the surfaces of the test
object can now constructed both visually and quantitatively. There are three basic NDT optical meth-
ods: Grid, holography, and shearography, see Botsco and Jones (2007). Generally speaking, there are
two major differences in the attributes of the three methods: first difference is in the way the two light
signals that are to be interrelated are generated. Second difference is in the complexity of the setup.
Moiré Method: Two light signals are split, then are beamed the object. Upon reflection, their inter-
ferences are analyzed and surface displacements and strains are computed. See Post et al. (1994)
for more details.
Holography: This type of optical NDT is based on applying a laser beam (reference beam)
to the test object. The reflected light from the object (called object beam) is then compared. with
reference beam. The two beams would show information about the object, including surface
displacements and strains. This information is revealed in real time. This general holographic
technique would yield a low-resolution images. For higher resolution, holographic interfer-
ometry is used. In such a situation, the images are generated from the interference of the two
wave fronts emanating from the deformed object, instead of the two simple laser beams. This
would result in high resolution of strains and displacements. For more details on this subject,
see ASNT (1995).
Shearography: The setup in shearography is simpler than the other optical methods. Only
one beam is directed at the test object. Upon reflection from points at different surface coordi-
nates, those reflected signals are then sheared using a birefringent crystal so as to make them
nearly collinear. The resulting interference patterns between these two collinear signals can
produce the desired surface displacement or strain images. Generally speaking, shearography
does not require special environmental requirements that are needed to generate the refer-
ence beam in holography. It is also known to produce higher resolution (sensitive) results, see
ASNT (1995).
The advantages of optical methods are as follows:
• Remote sensing
• High resolution
• Scales from local to global detection
• In general, faster than other comparable NDT techniques
• Faster in application than similar ultrasound methods
• Can be portable
• It may not be possible to detect interior defects

• Direct line of vision required; hidden, or interior surfaces can’t easily be tested
• Fairly recent and thus, body of knowledge is still being developed

Different NDT techniques discussed above can be utilized in many ways in SHM field. Table 7.18
summarizes some potential applications.
TABLE 7.18
Use of Different NDT Methods in SHM Civil Infrastructure Field
Virtual Sensing Remote Local vs. Global— Cost–Benefit
Method Direct Sensing Paradigm (VSP) Sensing Scaling Issues Issues
Visual Qualitative Due to qualitative Visual Can visually inspect Inexpensive.
methods results nature of the inspection can both local and However, given
method, VSP be done global zones/ the limited benefit,
utilization is not remotely, objects the cost–benefit
too promising either by naked rations may not be
eye, or by too effective
optical tools
Liquid Surface cracks. Remote sensing Can apply the Limited expenses.
penetrant Mostly is not penetrant both on For limited goals,
qualitative consistent with limited and large cost–benefit can
this method areas be attractive. For
general SHM
project, the
qualitative nature
of the method
limits its
cost–benefit
attractiveness
Optical Surface and Holds great Yes, by default Yes, by default. Cost–benefit can be
methods subsurface potential. For Local zones need very attractive for
stains. Can also example, STRID to be in the direct SHM projects, if
reveal using boundary line of sight, not decision making
displacements element method hidden analysis is
instead of finite performed
element method appropriately
7.5 ACOUSTIC EMISSION

7.5.1 Overview
The concept of AE is fairly simple: damage events in materials and structures generate stress waves
that propagate throughout the structure. The properties of these stress waves are dependent on the
properties of damage as well as the properties of the structure. These stress waves are called AE. By
monitoring the AE and analyzing the AE signals, information can be gathered about the damage as
well as the state of the material and the structure.
The frequency range of AE signals depends on the kind of triggering events. They can be in the
range of 1.0 Hz for example, if we consider ground seismic motions as a form of AE. However, in
the field of NDT, the frequency range of AE is similar to other UT waves (20 KHz ~ 1.0 MHz).
There are numerous differences between ultrasound and AE methods. One of the basic differences
is that AE is a passive NDT method; it relies on listening to damage signals. Ultrasound methods,
on the other hand, are an active method; it generates sound signals and monitors the interaction of
those signals with the test object.
This section will explore theoretical and practical attributes of AE in NDT field. We then will
discuss the potential applications of AE in SHM field. Several examples will be presented. Many
of the examples are from fields other than civil infrastructure; however, we believe that those appli-
cations can be utilized in civil infrastructure with minimal additional research effort. Figure 7.45
shows some basic components of AE field.
7.5.2 Theoretical Background

7.5.2.1 Basics of AE
Acoustic emission testing relies on observing AE signals. These are elastic wave forms that propagate
in the frequency range of 20 KHz to 1.0 MHz. By definition, AE waves are those waves resulting
form energy release (due to nonlinear deformations in material). As such, earthquake motions can be
considered as a form of large amplitude AE. AE also can have very small amplitude. For example,
corrosion processes in reinforced concrete can be considered as low amplitude AE, with amplitude in
the range of 30 dB*. Figure 7.46 shows some of the basic concepts of AE that will be discussed next.
Acoustic emission (AE) components
Physical process of
Damage detection
damage as an Acoustic signal Signal processing
and identification
acoustic source
Type of damage of
AE sensors Source localization Energy release
interest
Temporal Frequency - time

charecteristics of Frequency range Dispersion curves
composition
experiment
(continious,
intermittent, discrete, Noise levels Thresholds
etc.)
FIGURE 7.45 Components of acoustic emission.
* dB = 20 log (A1/A0), with A1 as the measured signal and A0 as a suitable reference baseline.
AE sensor Load: P
H
Felicity effect H
Emission count
E
G
Load (P)
F G F
B
E Source of D
Kaiser effect emission A
Emission C
C D signals (plastic behavior
A B of beam)
Load Displacement
Load history
Load-emission count Load-displacement
A-Start of plastic behavior
B-Load reduction starts
C-Reloading
D-Load reach previous max level
E-New load max level
F-Load reduction
G-Load reach previous max level
H-Load keep on increasing
FIGURE 7.46 Basics of acoustic emission.
7.5.2.2 Kaiser, Dunegan, and Felicity Behavior of AE

The load-emission count relationship is such that as the loading increases, the emission count
increases. However, there are several effects that can be observed during the loading and unloading
of a system. Three main effects govern the AE behavior, and understanding those effects can help
in performing an accurate and efficient AE experiment.
Kaiser Effects: Kaiser effects happen when materials start behaving in a nonlinear form (a form
of failure, or damage), as it starts emitting stress waves that originate from the damaged loca-
tion. The emission will stop if the loading stops, or if there is a reduction of the load level. The
emission will only start again when the load exceeds the previous loading level. For example, in
Figure 7.46, when the loading P increases, the emission count increases, as shown from pint “A”
to point “B.” If the load decreases to point “C,” the emission count stays constant. The emission
count will remain constant when the load increases again, until it reaches the previous level at
point D = B. If the load increases beyond that, the emission count starts increasing again.
Dunegan Corollary: Dunegan et al. (1970) noticed that in many situations the Kaiser effects
need to be modified. Emission can occur when the load is kept constant, which appears to be a
deviation from Kaiser effect. The discovery of this corollary led to the successful applications of
AE in many field NDT proof tests, Miller (2007).
Felicity Effects: At higher stress levels the Kaiser effect does not hold true. In such a situa-
tion, emission can occur at load levels that are lower than the previous maximum load level. For
example, Figure 7.46 shows that if at a higher load level, “E,” the load decreases to level “F,”
and the emission count will remain same. When the load starts increasing again from “E” to
“G,” and the emission count remains same. However, at level “G” that is less than the previous
maximum load level “E,” the emission count start increasing. Such an effect is called “Felicity
Effect” The Felicity ratio is defined as PR /PM with PR as the load level at which the emission
count start increasing and PM as the maximum previous load level. Obviously, if the Felicity ratio
is 1.0, we have Kaiser effects.
Felicity effects occur in time-dependent material behavior (such as FRP and concrete). Several
civil infrastructure situations can experience Felicity effects such as Hydrogen embrittlement (high
strength tendons), corrosion (reinforced concrete rebars), and FRP bridge decks. For more detail,
see Miller (2007). The effects have been incorporated in standard load tests and were devised to
generate AE signals in different structures such as for FRP pressure vessels, and bucket truck and
lift inspections. We observe the many similarities between the proof testing of those situations and
the proof testing of bridges, (see Chapter 8 by Ettouney and Alampalli 2012). On the basis of those
similarities, and the success of AE in providing satisfactory proof testing methodology, it seems that
developing a rigorous AE proof testing guide, or even standard, for bridge proof testing is a worthy
future effort for the bridge community.
7.5.2.3 Emission Initiation and Propagation

Emission is initiated when energy is released from a local source due to the state of stresses or
deformations within the material. The emission from the stressed source behaves as a pulse. This
pulse will cause elastic waves to propagate throughout the body in all directions. Depending on the
shape of the damage, some preferred propagation paths might be followed. Preferred propagation
paths might also occur in inhomogeneous materials such as FRP. The stress wave propagation in an
AE situation will follow conventional rules of wave propagation (Rose 1999).
There are two types of AE waves: burst and continuous. Burst AE, as the name implies occur
when the energy is released suddenly in a form of pulse in short time span. It occurs during proof
load tests, for example. Continuous AE signals occur when the damage in the material occur con-
tinuously, as in corrosion processes in metals. We explore examples of both types later.
7.5.2.4 Important Characteristics of AE Wave

There are several important characteristics of AE signals: amplitude, frequency content, and rate
of decay are the most important. However, the overall wave forms of AE signals have been used to
discern valuable DMID information: signal processing of AE signals is discussed later this section.
See Figure 7.47 for illustration of AE wave form characteristics. In addition to the above AE wave
forms attributes that are shared with many other wave forms (such as seismic, impact-echo, etc.),
there are specific AE wave forms attributes that are used in AE experiments such as the following:
AE Counts: The number of passages of amplitude over a given threshold during an AE signal is
called AE count (see Figure 7.47). Clearly, the AE count can be used as a measure for the strength
of an AE signal: the larger the count, the stronger the signal. The relationship between elapsed time
and AE count is used extensively in AE test. In addition to revealing the strength of AE signal, it can
also reveal the overall condition of the damage in the test object. If the rate of increase of the AE
count is high, it indicates a more severe damage state than slower rate (Figure 7.48).
AE rise time
Count of amplitudes
passing threshold is
4 in this AE signal
AE signal
Threshold
AE signal duration
Time
FIGURE 7.47 Acoustic emission signal characteristics.

Damage severe,
needs attention
Emission count
Normal behavior,
test passed
Load
Qualitative Use of Load-Emission Count

in an AE Test
FIGURE 7.48 Use of AE count.
AE Energy: AE energy is measured by RMS (root mean square) or TMS (total mean square,
which is the square of RMS) of the signal. It is a measure of the strength of the signal. Similar to
AE counts, its rate can also be used to qualitatively infer the level of damage, or the lack of it in the
object. Using AE energy as a metric is particularly useful for continuous AE Tests, which would be
valuable for SHM experiments. In general the resulting energy measures are smoother results than
the counting technique. Energy results are also less sensitive to the choice of threshold value; which
is another advantage of using AE energy in SHM tests.
AE Signal Duration: Signal duration is evaluated as in Figure 7.47. It can be used as a measure
of the strength of AE signals.
AE Transforms: Signal processing using FT, WFT, WT, or HH techniques can be used to infer
damage presence, location, and extent. In general, those techniques are limited by the sampling rate
of the signals. Thus only frequencies up to 5 MHz can be analyzed.
More details on AE signal attributes and their usefulness during NDT/SHM experiments are
discussed by Pollock (2005).
7.5.2.5 AE Data Filtering

One of the important features of AE is that the emission acoustic energy can be extremely small.
This can be a severe limitation, especially when used in an in-situ SHM experimentation where
the background noise can interfere with the AE signal. There are several practical solutions to this
limitation. Two of the popular solutions are discussed next.
Swansong Filters: The unwanted noise in AE signal can be filtered analytically by Swansong
filtering technique. This analytical filtering technique is based on a set of rules that were devel-
oped by the Association of American Railroads (1999). The rules are based on eliminating
unwanted noise that are generated by specific outside sources that are common in the railroad
community such as mechanical sliding or rubbing that can occur in rail cars. A research study
by Endo (2000) reported that those types of filters were not very effective in removing noise
from moving vehicles on bridges. The idea of developing analytical filters to remove unwanted
noise from AE signals in civil infrastructure, for example, traffic noise, should be explored
further.
Guard Sensor Technique: Guard sensors concept was developed to help the AE sensors record
only relevant signals. The concept is shown in Figure 7.49. Carlyle and Ely (1992), Carlyle and
Leaird (1992), and Carlyle (1993) successfully used guard sensors in AE bridge testing to filter
unwanted noise from AE signals. Guard sensors can be incorporated in an AE sensor network.
Traffic:
Source of noise
Damage
Unwanted noise
source
Test object
Guard sensors
Data sensors
FIGURE 7.49 Guard sensor concept.
A careful study of potential sources of noise and potential crack or damage locations can help in
optimizing the performance of an AE-based SHM experiment.
7.5.2.6 Calibration of AE Sensors

Sensors need to be calibrated prior, or during the experiment. A popular calibration technique for
AE sensors and experimental setup is the pencil lead breaking technique. It was found out that the
fracture of pencil lead produces AE signals that are similar in amplitude and frequency range to the
process of fracture in many materials, such as FRP material. Thus, the AE sensors can be calibrated
by this process, which is also known as Hsu-Nielsen pencil technique.
7.5.2.7 Source Location

One of the advantages of AE method is its capability to locate damage locations accurately. When
the AE signal has a definitive arrival time, as in the burst mode AE, for example, the damage loca-
tion can be estimated in linear, 2D, or even 3D geometries. The process is based on the well known
time of flight (TOF) technique. For example, consider the linear situation in Figure 7.50, where the
source of the damage is located on the same line between two AE sensors, S1 and S2, that are spaced
with a distance L. Assume that the arrival times of the AE signals were recorded as T1 and T2 at the
two sensors. The distance between the damage and sensor S1 can be shown to be
⎛ (T1 / T2 ) ⎞
L1 = ⎜ L (7.9)
⎝ 1 + (T1 / T2 ) ⎟⎠

The distance from the damage source to sensor S2 can be computed in a similar fashion. The lin-
ear geometry has an important application in the field of high strength wires, such as suspension
bridges, cable stayed bridges, or post-tensioned/prestressed bridge girders. A well-positioned AE
sensors along the cables can detect wire breakage (due to deterioration), and can also locate the
damage source fairly accurately. A more involved, but similar expressions can be developed for 2D
and 3D geometries. The 2D geometry is shown in Figure 7.51. Note that at least two AE sensors are
AE signal source
AE sensor, S1 AE sensor, S2
The signal will arrive after T1 The signal will arrive after T2
The location of the damage can be

estimated from different times of arrival
FIGURE 7.50 Linear setup for AE damage location.
AE signal source
AE sensor, S3
The signal will
arrive after T3
AE sensor, S1
The signal will
arrive after T1
AE sensor, S2
The signal will
arrive after T2
AE signal
wave front
The location of the damage can be

estimated from different times of arrival
Minimum of three sensors are needed to

accurately locate the source location
FIGURE 7.51 2D setup for AE damage location.
needed for linear geometries. A minimum of three AE sensors are needed for 2D geometry. For 3D
geometry, a minimum of four AE sensors are needed.
When the AE signal is continuous, as in corrosion processes, TOF method is not appropriate.
Time averaging techniques are used to estimate the damage location using the zone location method,
Baron and Ying (1987). The method is based on observing the AE sensor recording the largest
observed AE amplitude A AE_0, within a given AE sensor array. It is logical to assume that the source
of damage is closest to that sensor. Assuming a reasonable attenuation rate can place the damage on
the circumference of a circle with that sensor at its center. The source location can be narrowed by
observing the sensor with the next highest amplitude A AE_1. Using the same attenuation rate, the dam-
age source can be pinpointed using technique that is similar to the TOF technique. More involved
techniques for continuous AE signals can reveal even more information. For example, a cross-corre-
lation evaluation (Bendat and Piersol 1971) between the two continuous signals can reveal the time
difference between the two signals. When the time differences are know, the TOF method can be
used to estimate the source location. A coherence analysis between the two signals can also be made
to detect the presence of damage. For more detail see Baron and Ying (1987).
The simplicity of evaluation of damage source location, both for burst and continuous AE modes,
using AE can be utilized so scale real time damage detection in an SHM setting using an array of
AE sensors that are optimally located (both in number and location).
7.5.2.8 Properties of AE as Related to SHM
Civil infrastructure SHM projects have been gaining popularity lately. For AE to be successful in
any SHM project, several of its attributes must be utilized efficiently. Table 7.19—AE attributes in
civil infrastructure SHM—shows some of the special features of AE phenomenon, and how they
can be used in civil infrastructure.
7.5.3 AE versus Other NDT Methods

AE method and its underlying physical behavior are very different from other NDT methods, as
shown in Table 7.20. Since the UT and AE methods operate mostly within similar frequency range,
it is of interest to compare the two approaches, as shown in Table 7.21. Finally, Table 7.22 shows a
general comparison between AE method and the popular DMID vibration-based technique.
Duke and Horne (2006) provided an example of the effectiveness of AE in identifying the exis-
tence and location of the crack in metal components of bridge structures, and in detecting if the
crack is actively growing. Hence, monitoring AEs from the formation and growth of cracks in steel
commonly used for bridge structures was reviewed by the authors. The emphasis of the review
was on cracks that penetrated the thickness of the plates, commonly used in bridge girders, and
propagating in the plane of the plate. Figure 7.52 shows the relative amplitude of AE signals due to
mechanism responsible for AE during material deformation. Figure 7.52 also shows that all detect-
able AE can not be immediately attributed to the final generation of new crack surfaces.
AE wave forms were collected during the overload tests of specimens containing cracks introduced
by fatigue in A36, 1018, A588, HPS70W, and A514 steels, with each specimen overloaded numerous
TABLE 7.19
AE Attributes in Civil Infrastructure SHM
Acoustic Emission (AE)
Issues Effects in SHM
Noise Levels Traffic and other ambient noise will need special considerations in an AE testing, such as
guard sensors, or special types of numerical filters. Noise effects become important for
continuous AE signals such as AE generated by corrosion processes
Distance from When sources of damage are known, then AE sensors can be located near those sources; this is
source—attenuation an NDT-type experiment. In an SHM type experiment, where the sources/locations of
damage are not known, or when it is desired to have a continued monitoring of the structural
state; several AE sensors must be used. The distances between those sensors must be
optimized so as the AE signals will not attenuate be for they reach at least one AE sensor
Damage localization AE sensors need to be close to each other to detect damages reliably
Frequency range AE sensors need to cover as wide frequency range as possible. This is needed to accommodate
as many types and sizes of damage sources as possible
Passive vs. Active AE AE is a passive type of testing: the sensors detect the signals emitted from damage sources.
Testing Some situations pulsers might be used to send acoustic pulses throughout the test object.
Those acoustic signals will interact with defects and their reflections, or scattering from the
damage sources can then be detected by AE receivers; this is similar to pulse-echo technique
in ultrasonic testing
A summary of the difference between active and passive emissions was given by Butt and
Limaye (2002). They described both approaches as damage identification methods
TABLE 7.20
AE versus Other NDT Methods (General Comparison)
Issue AE Other NDT
Role Listener (passive) Initiator (Active)
Scope Localized, or whole structure More localized
Main limitation Background noise Complex geometries
TABLE 7.21
AE versus Ultrasonic
Issue AE UT
Physics Stress waves Stress waves
Frequency range Higher Lower
Techniques Listens to waves created Generate pulses into the media,
from anomalies then looks for responses
Sensors Use multiple sensors to Can use single or multiple
localize the damage effects sensors
Detects Location, size of damage Location, size of damage
Accuracy Less More
TABLE 7.22
AE versus Vibration-Based Testing
Issue AE Vibration-Based DMID

Physics Propagating waves Standing waves
Frequency range 20 kHZ –1.2 MHz Up to about 100 Hz
Techniques Listens to waves created from anomalies. Measures displacements or accelerations.
Emphasis on analyzing wave forms and Emphasis and generating natural
relating them to damage frequencies and mode shapes
Issue AE GV
Sensors Use multiple sensors to localize the damage Can use single or multiple sensors. Higher
effects. Higher the number of sensors, number of sensors does not necessarily
higher the possibility of damage detection mean higher number of detected mode
shapes
Detects Location, size of damage Useful for STRID, Limited DMID, value
times at different crack lengths. On the basis of these studies they concluded that the amount of AE
detected were insufficient to provide a reliable means of monitoring crack extension.
7.5.4 Applications in SHM—General

7.5.4.1 General
Constructed infrastructure includes nuclear structures, bridges, (including special types of bridges
such as suspension, cable stayed, or movable bridges), pressure vessels, dams, tunnels, pipelines,
transmission towers, and storage tanks. AE testing has been used extensively in most of those infra-
structure. In this section we look closely at issues that involve the use of AE in civil infrastructure
such as bridges, tunnels, or buildings.
90
80
Amplitude (dB ref1

70
microvolt)
60
50
40
30
20
10
0
g
n
e)
e)
n
om on
ng
e
(C leak
ur
in
ur
io
tio
io
tio
sit
sit
ki
i
ck
at
es
ut
ct
ct
a
po
po
ac
rm
rm
rm
fra
a
oh
ol
ra
e
r
cr
r
om
ev
su
c
fo
fo
sfo
de
de
ar
ge
es
sio
de
de
(C
xi
an
ul
va
ge
pr
n
fu
O
n
in
an
g
ng
o
tr
ea
ro
in
tio
gh
In
Tw
us
gr
ki
se
Cl
yd
ak
ca
i
cl
ac
te
ha
H
H
re
In
o
In
cr
cp
isl
rb
rix
D
iti
be
at
ns
Fi
M
te
ar
M
FIGURE 7.52 Mechanisms responsible for AE during material deformation and their typical AE signal
amplitudes. (Courtesy of Physical Acoustics Corporation.)
7.5.4.2 Types of AE Tests in Infrastructure

There are two modes of AE behavior that can be utilized in civil infrastructure’s DMID efforts.
They are burst and continuous AE modes as shown in Figure 7.53. The main difference between the
two modes is in the temporal behavior of the material. In burst mode AE, the energy is released in
bursts or pulses as the material undergoes plastic deformations. Burst mode AE can be with high
amplitudes, for example, when a high tension cable in a post-tensioned beam breaks suddenly, the
AE signal can be very large. Lower amplitude burst AE modes can also occur in proof tests of
bridges. Proof tests involve increasing the loading demands until a preset loading level. If the bridge
undergoes a plastic deformation, a well-placed AE sensors can detect those deformation effects.
Also, when bearing plates move against each other, thus generating friction, the friction energy
would produce a burst mode AE signals. Continuous AE signals occur when the changes in material
are continuous, such as corrosion process, or continuous scour process under bridge foundation.
Acoustic emission techniques can be utilized in many ways in SHM field. Table 7.23 summarizes
some of the potential applications. Table 7.24 and 7.25 presents specific potential uses of AE in the
field of civil infrastructure.
7.5.5 Corrosion Rate Measurements (Continuous AE)

Corrosion is one of the costliest hazards in the field of infrastructure. Detecting corrosion rates or cor-
rosion damage in reinforced concrete, or steel structures is always an issue in SHM/structural health
in civil engineering (SHCE). It has been observed that corrosion as a chemical process can generates
AE signals. The AE due to corrosion is a good example of continuous AE; since corrosion is a continu-
ous process, it is reasonable to assume that the resulting emission is also a continuous signal; thus, all
techniques and procedures of continuous AE might be used to detect corrosion.
Riahi and Khosrowzadeh (2005) tried to answer the question: can AE methods detect corrosion
rate in steel? They devised a test that induced corrosion process using different solution molari-
ties. The sampling AE frequencies during the test were in the range of 100–400 KHz. One of the
problems they tried to address during the test was the relative ambient noise level and the actual AE
amplitude level. For a successful AE detection, a good compromise has to be made between noise
at low frequencies and fast attenuation at high frequencies. The results of the test showed that the
AE sensor Load: P AE sensor Load: P

Emission
signals
Source of
Source of
Emission emission
emission
signals (Plastic behavior
(Friction of bearings
of beam)
due to rust)
Plastic behavior of beams Friction of bridge bearings
AE signal is observed while the loading AE signal is observed while the loading
is increased is increased
AE sensor AE sensor
Post-tensioned Steel rebars
tendons
Source of Source of
Emission emission Emission emission
signals (Breakage of signals (Chamical corrosion
post-tensioned process in rebars)
tendon)
Corrosion of Rebars
Breaking of post-tentioned tendons
Continuous AE monitoring:
Continuous AE monitoring: Until the breakage occurs
Corrosion rates are observed
in real time
FIGURE 7.53 Modes of AE in civil infrastructure.
TABLE 7.23
Acoustic Emission Utilization in SHM Civil Infrastructure Field
Direct Virtual Sensing Local vs. Global—
AE Method Sensing Paradigm (VSP) Remote Sensing Scaling Issues Cost–Benefit Issues
Continuous AE Yes Relationships No: Sensors Can be local or For a well-planned
between AE count, need to be in global, depending experiment,
amplitude, or direct contact on the number and benefits to cost
energy and damage with the test distribution of AE ratio may be
needs to be object sensors attractive
developed. For
example, correlation
between AE count
and corrosion rates
Burst AE Yes Relationships No: Sensors Can be local or For a well-planned
between AE count, need to be in global, depending experiment,
amplitude, or direct contact on the number and benefits to cost
energy and damage with the test distribution of AE ratio may be
needs to be object sensors attractive
developed. For
example, correlation
between AE signals,
dispersion curves
and wave velocities
TABLE 7.24
Applicability of AE Technique to Selected Problems in Infrastructure
Hazard Comments
Yielding of materials When materials reach post-elastic state, AE are generated. Thus, any scenario at which the
material reaches or operates at yielding or post-yielding state, AE method of testing offers a
reasonable choice. Such scenarios include postseismic behavior, soil scour due to flooding,
fatigue, or impact conditions
Cracking of concrete When concrete cracks, either by design or as a result of excess deformation, such cracks cause
AE. When such AE is detected, it can offer insight on the causes and nature of these cracks
Friction movements Many components in civil infrastructure are designed to move against each other, such as
movable bridge bearings. Some components are not designed to move relative to each other,
such as bolted or riveted connections or plate girders. AE resulting from relative movements,
whether by design or as a result of malfunction, can be recorded. A decision as to the nature
of such AE (within design limits, or a nonacceptable design condition) can help the system
to operate safely
Welding State of welding has been inspected by AE in many industries as well as in the civil
infrastructure field
Failure of matrix or resin State of matrix or resin has been inspected by AE in many industries as well as in the civil
in composite materials infrastructure field
TABLE 7.25
Applicability of AE Technique to SHM Hazards
Hazard Comments
Fatigue A very popular application of AE testing. The sensors can be temporarily or permanently placed.
Also the experiment can be either local of global, depending on the objective of the test
Corrosion—General Corrosion process itself in the wires might be monitored by AE (see example below)
Corrosion—PT/PS A popular application of AE is to detect breakage of PT/PS wires. Corrosion process itself in the
wires might be monitored by AE (see example below)
Scour AE was shown to detect changes and deformations in soils and rock as they occur, (Koerner et al.
1981 and Koerner and Lord 1982). AE might be useful to detect emission from soil erosion due
to rapid flow of water, however, ambient flow noise might be of concern
Earthquakes Can be useful to detect postyield deformations of systems after an earthquake. The governing
behavior in such a situation would be Dunegan Corollary
FRP wrapping Can be useful, see Karbhari et al. (2005)
FRP decks Can be useful, especially for proof testing of deck
Bridge security Can be useful to detect post yield deformations of systems after a blast event. The governing
behavior in such a situation would be Dunegan Corollary
mean AE amplitude due to corrosion process is about 40 dB, and thus they decided to use a thresh-
old of 26 dB, which is well above ambient noise level. Figure 7.54 shows the relationship between
count rates, corrosion rates, and molarity level obtained during their test. Figure 7.55 shows a cor-
relation between the AE count rate and corrosion rate. The test proves the following:
• There is a clear correlation between AE count rate and corrosion rate

• The ambient noise level is well below the threshold count rate
This test shows great promise for using AE in real time monitoring of corrosion process for infra-
structure. But, significant further research is required to make it useful for structures such as
bridges.
Corrosion rate and AE count rate versus molarity

2500 700
600
2000
Count rate *10^6 500
Corrosion rate
1500
400
1000 Count rate A 300

Count rate B
200
500 Corrosion rate A
100
Corrosion rate B
0 0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
Molarity
FIGURE 7.54 AE count, molarity, and corrosion rate. (Reprinted from ASNT Publication.)
Count rate versus corrosion rate

2500
2300
2100
cdd *10^6
1900
1700
1500
1300
1100 A
900 B
700
350 400 450 500 550 600 650
mdd
FIGURE 7.55 AE count versus corrosion rate. (Reprinted from ASNT Publication.)
7.5.6 Proof Loading (Burst AE)

A complete DMID involves identification of four damage attributes: occurrence, location, type, and
size. The experiment reported by Carlos et al. (2006) showed how AE can be used to quantify all
four DMID attributes in real time, in-situ situation. The experiment was performed on a full size
storage vessel shown in Figure 7.56. The AE sensors were placed on a regular grid around the cir-
cumference of the vessel. As the vessel is pressurized in a fashion similar to a proof loading setup,
the damages in the vessel shell started to emit signals in a burst mode similar to that in Figure 7.57.
A special processing software was used (Figure 7.58). The software used signal processing tech-
niques, such as wavelet transform (WT) analysis, and theoretical wave dispersion relations to detect
the damage sources of the signals. A built-in discrimination logic that correlated type and size of
damage sources with the detected AE signals, shown in Figure 7.59, was then used to completely
identify damages on the vessel skin. The identified damages are shown in Figure 7.60.
7.5.7 Quality Control (QC)

Another potentially important application of AE in the field of SHM/SHCE is QC and acceptance cri-
teria. It was shown that the behavior of AE count rates can reveal differences in material composition,
Pollock (2005). For example, when steel specimens with different degrees of annealment (underan-
nealed, optimally annealed, and overannealed) are subjected to a monotonically increased loading, the
AE energy rate behavior changes considerably, as shown in Figures 7.61, 7.62, and 7.63, respectively.
This suggests that a carefully designed AE experiment can reveal the adequacy of in-service structural
10850 mm
1m
1750 mm
Flow
D=4522 mm
2870 mm
Due to blistering
12
30–34 mm
Slug catcher
Scanned 470 · 0 · 11 · 8 20–29 mm
9 3
area 15–19 mm
FIGURE 7.56 Pressure vessel with damage locations. (Courtesy of Physical Acoustics Corporation.)
FIGURE 7.57 Typical AE signal. (Courtesy of Physical Acoustics Corporation.)
component. For example, the in-service behavior of gusset plates in a steel truss bridge can be exam-
ined without having to rely on numerical analysis. This becomes even more informative if the gusset
plate is suspected to have a potentially brittle behavior that numerical analysis can’t reveal.
Another potential QC application of AE would be in the emerging field of FRP bridge decks
(Chapter 6 by Ettouney and Alampalli 2012). The body of experience of AE in the behavior of con-
structed FRP systems, such as pressure vessels, is vast. Utilizing this body of knowledge for load
testing of FRP deck seems to be logical and thus further research is required in this area.
7.5.8 Incipient Failure Detection (IFD)

The real time capability of AE to detect, locate, and in sometimes size up, damage offer a unique
potential application in the SHM/SHCE field; the potential of detecting incipient failure of global
FIGURE 7.58 Processing software. (Courtesy of Physical Acoustics Corporation.)
FIGURE 7.59 Damage discrimination screen. (Courtesy of Physical Acoustics Corporation.)
systems, or system components. This capability of AE was utilized in other fields. For example,
by monitoring AE signals of pump discontinuities, signs of malfunction and ultimate failure were
evident long before similar signs of failure were observed by monitoring lower frequency behavior,
(see Figure 7.64).
The concept of IFD in civil infrastructure is an attractive one, and can save costs and improve
safety. Recall that we explored monitoring the relationship between the stability condition of a truss
compression member and its natural frequency. We showed that real time monitoring of compres-
sion members can reveal if there is an incipient loss of stability, which can lead to a catastrophic col-
lapse of the truss. The concept can be generalized into even a higher frequency range; the frequency
FIGURE 7.60 Damage locations during test. (Courtesy of Physical Acoustics Corporation.)
500
400
Energy rate, (µV)2
300
200
100
Load
0
0 10 20 30 40 50 60 70 80 90 100
Time, s
FIGURE 7.61 AE in QA of materials/construction: suboptimal composition. (Courtesy of Physical Acoustics

Corporation.)
range of AE. Thus, the concept is, at least, logically feasible. This concept of monitoring AE signals
at high frequencies in real time and trying to use potential early warning signs of incipient failure of
either structural components, or whole structural systems needs to be researched fully: the potential
rewards for such research, if successful are immense.
7.5.9 Failure of High Strength Wires

High strength wires are used extensively in civil infrastructure. They are used for prestressing/post-
tensioning of reinforced concrete components in buildings, as well as bridges. High strength wires
are also used in suspension and cable stayed bridges. Condition assessment of high strength wires
has been of great interest to officials due to higher costs of retrofit and/or replacement. AE has been
used successfully to monitor the event of breaking of wires. When a post-tensioned wire (or any
high strength wire) breaks, the energy of the event generates burst-type AE signals that travel along
500
400
Energy rate, (µV)2 300
200
100
Load
0
0 10 20 30 40 50 60 70 80 90 100
Time, s
FIGURE 7.62 AE in QA of materials/construction: optimal composition. (Courtesy of Physical Acoustics

Corporation.)
500
400
Energy rate, (µV)2
300
200
100
Load
0
0 10 20 30 40 50 60 70 80 90 100
Time, s
FIGURE 7.63 AE in QA of materials/construction: overoptimal composition. (Courtesy of Physical Acoustics

Corporation.)
the wires for long distances due to the high intensity of the wire breaking. If AE sensors are placed
along the wires, they can record the signals and through a simple TOF procedure (generally linear
along the wire) the location of the break can be accurately computed.
Cullington et al. (2001) reported on a real time SHM project that utilized AE sensors to monitor
post-tensioned wire breaking in real time. AE sensors were placed along the post-tensioned rail via-
duct of Figure 7.65 as shown in Figure 7.66. When a wire breaks, the AE recorded signal is shown
in Figure 7.67. The FT of the signal is shown in Figure 7.68. When a wire break is detected, it is easy
for the official in charge to make appropriate decisions within reasonable time, thus ensuring safety
of the system under consideration.
7.5.10 Local Area Monitoring (LAM)

Carlos et al. (2000) described the concept of LAM, and presented some of its practical applications.
The concept of LAM with AE utilizes a limited number of AE channels and takes advantage of the
structural loading resulting from normal traffic patterns. Applications suitable for LAM include
37.5 (1.5)
35.0 (1.4)
32.5 (1.3)
millimeters (inches) per second

30.0 (1.2)
27.5 (1.1)
25.0 (1.0)
22.5 (0.9)
Velocity
20.2 (0.8)
High frequency
17.5 (0.7)
acoustic signal
15.0 (0.6)
12.5 (0.5) Low frequency
10.0 (0.4) vibration signal
0.75 (0.3)
0.50 (0.2)
0.25 (0.1)
5 10 15 20 25 30 35 40 45 50 55 60 Elapsed time
(minutes)
e
vit e
ge
lur
g
o
ma
ma
ati
fai
da
da
ng
Ca
ing
ng
ari
ari
ear
Be
be
tb
us
en
rio
ipi
Se
Inc
FIGURE 7.64 Incipient failure detection by AE: discontinuities of pump behavior. (From Green, A. et al.,
(1987). Acoustic emission applications in the petroleum and chemical industries, in Nondestructive Testing
Handbook: Volume Five, Acoustic Emission Testing, ASNT, Columbus, OH. With permission. Reprinted
from ASNT Publication.)
FIGURE 7.65 Railway viaduct, Huntington UK. (Courtesy of Pure Technologies US, Inc.)
• Determining if a preexisting or detected crack is active

• Monitoring of retrofits
• Monitoring of highly stressed areas to determine structural integrity
• Monitoring a suspected area that cannot be visually inspected
At first, AE was used on 15 bridges with “known defect” areas that were monitored to obtain
more information on activity level of the defects. On the basis of this experience, an AE system for
LAM was developed and described during their work.
t
me n
Ra Ra
roa pton
ilw ilw
e
Pav
ay ay
m
Bra
d
Web - 01
HRVA-01 HRVA-05 HRVB-00
Web - 02 HRVB-11
Web - 03
HRVA-06 HRVB-01 HRVB-12
Web - 04
HRVA-12 HRVB-07
Web - 05
Web - 06 HRVA-07 HRVB-02
Junction box HRVA-13 HRVB-08

Web - 07
HRVA-03
t
HRVA-00 HRVB-04
n
HRVA-08 HRVB-14
eme
Web - 08
HRVB-09
Pav
Web - 09 HRVA-14
Web - 10 HRVA-09 HRVB-04 HRVB-15HRVB-06

Web - 11
HRVA-04 HRV-HJ-A-01
HRVA-10 HRVB-05
2m 2m 2m 2m 2m 2m 2m 2m 2m 2m 2m 2m 2m 2m 2m 2m 2m 2m 2m 2m 2m 2m 2m 2m
FIGURE 7.66 Plan and elevation of the bridge. (Courtesy of Pure Technologies US, Inc.)
A15
10
Signal (v)
–10
0 0.015 0.03 0.045 0.06 0.075 0.09 0.105 0.12 0.135 0.15
Time (s)
FIGURE 7.67 Signal of wire break. (Courtesy of Pure Technologies US, Inc.)
Local area monitoring system, shown in Figure 7.69, is rugged with shock and environmental
protection, accommodates 8 channels (option for multiples of 2 up to 8), data can also be digitized
and stored, can accommodate other than AE sensors, internal battery power if needed, and can be
connected to PC through a modem. One possible sensor configuration is shown in Figure 7.70.
Field Test 1: In an SHM project for George P. Coleman Bridge in Yorktown, Virginia, in October
1996, sensors were mounted on the south, steel pivot box girder, and the remote operating features
of the system were tested. A complete investigation of the north and south pivot box girder was
performed and reported by the Virginia Transportation Research Council.
A15
1000
Magnitude (v)
0
0 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000
Frequency (Hz)
FIGURE 7.68 Fourier transform (FT) of wire break. (Courtesy of Pure Technologies US, Inc.)
FIGURE 7.69 LAM instrument in NEMA 4 enclosure. (Courtesy of CRC Press.)
Second Field Test: In November 1998, on the Route 1 bridge south of Washington D.C., sensors
were mounted near a weld crack in a longitudinal stiffener plate. The test was designed to determine
if the crack had been arrested. The results indicated that it had been arrested.
7.5.11 Advantages and Disadvantages of AE

Among the advantages of the AE method are the following:
• Can be local or global

• Passive: no need for external stimulus
Defect area
6
7
5 1
8
2
4
3
Strain gages
FIGURE 7.70 Typical experimental setup. (Courtesy of CRC Press.)
• Has potential for proof testing, IFD, and QC

• Real time damage detection; damage is detected as it occurs
• Very amenable to long-term monitoring, that is, can span the leap from NDT to SHM easily
Disadvantages and limitations of AE are as follows
• Ambient noise can be a major issue, especially for continuous AE signals

• In many situations, AE can be mostly qualitative
• Direct contact with test object is needed
• More than one sensor is needed to locate the damage source
7.6 VIBRATION–BASED METHODS

7.6.1 General
This section addresses a popular DMID approach in civil infrastructure: the vibrations-based DMID.
The basic premise of this approach is to use measured vibration properties of the system in the pristine
and damaged states, then try to identify the damage through some analytical optimization process. In
other words, the methods study the changes between baseline and damaged states. Generally speak-
ing, these methods are based on modal, or parameter, identification and this relates them directly to the
STRID field. We explore first the basic methodologies that utilize STRID method in vibration-based
DMID. Next, we offer two examples of the use of these methods. Both examples point to the impor-
tance of being careful when choosing a vibration-based DMID technique in civil infrastructure.
7.6.2 Overview of Methods

7.6.2.1 Changes in Modal Properties
Since most methods are based on estimating effects of damage on modal properties, let us set some
rules. Consider an SDOF with stiffness, and mass as k and m, respectively. The system will have a
natural frequency of v = k / m . Rearranging we get
k =mv2 (7.10)

For a baseline (no-damage) and damaged states, the change in the frequency is
1
Δ v = v1 − v 0 =
m
( k1 − k0 ) (7.11)

If the change in stiffness due to damage is
Δ k = k1 − k0 (7.12)

we can express the dimensionless change of frequency ∆ϖ/ϖ0 as
Δv ⎛ ⎛ Δk⎞ ⎞
= ⎜ 1+ ⎜ ⎟ − 1⎟ (7.13)
v0 ⎜⎝ ⎝ k0 ⎠ ⎟⎠

Figure 7.71 shows the relationship between dimensionless change of frequency ∆ϖ/ϖ0 and the dam-
age as estimated by the dimensionless change of stiffness ∆k/k0.
To relate the frequency ranges with a simple beam situation, consider the frequency equation of
simply supported beam:
n2 rA
vn = (7.14)
2 EI

The mode number is n. The modulus of elasticity, the mass density, the cross-sectional area, and
the moment of inertia are E, ρ, A, and I. The length of the beam is ℓ. Assuming a damage state as
described in Table 7.26, the dimensionless change of frequency and dimensionless change of stiff-
ness can be computed. For a simply supported steel WF 36X194, the damage states can be related
to the changes in frequency and stiffness in Figure 7.72. For a dimensionless change in the range
of 10% or more, 25% of the web, or more, need to be damaged for accurate vibration observations.
Clearly, relying on frequency, or stiffness, changes alone might not be an efficient way of monitor-
ing damage in this situation.
7.6.2.2 Utilization of Numerical Formulations FEM/BEM

Some DMID methods rely on numerical formulation to identify damage. Most of the methods use
the changes in stiffness matrices ∆K for the DMID
Frequency-damage relationship
1.00E+00
1.00E-01
Dimensionless frequency
1.00E-02
1.00E-03
1.00E-04
1.00E-05
1.00E-06
1.00E-05 1.00E-04 1.00E-03 1.00E-02 1.00E-01 1.00E+00
Dimensionless damage
FIGURE 7.71 Frequency–damage relationship.

TABLE 7.26
Damage Levels of Steel Beam
Damage Level Description
A Loss of whole bottom half
B Loss of bottom flange
C 25% of web
D 5% of web
E 1% of web
F 0% of web
Dimensionless damege–frequency for WF 36 × 194

1.00E+00
Dimensionless result
1.00E-01
Dimensionless stiffness
Dimensionless frequency
1.00E-02
1.00E-03
A B C D E
Damaged case
FIGURE 7.72 Dimensionless damage-frequency for steel beam.
The problem can be formulated as
[ Δ K ] = [ K1 ] − [ K 0 ] (7.15)

The order of the matrices in Equation 7.15 is N, the total number of degrees of freedom. The matrix
[∆K] is an expression of damage, while [K0] and [K1] are the baseline and the damaged stiffness
matrices, respectively.
Most of the techniques of this approach rely on finding [∆K], subjected to some set of test results.
This reminds us with parameter estimation methods of Chapter 6, only executed twice: for pristine
and for damaged systems. The guidelines of using parameter estimation methods for DMID were
discussed earlier in this chapter.
Detection of damage levels and location of the damage in the system are the main goals of
most SHM projects. Several vibration-based damage detection schemes have been proposed
in recent years. The next section investigates the accuracy of different damage detection tech-
niques that utilize vibration-based methods. Two full-scale case studies are presented. Several
popular DMID methods are applied for each case. We will find that all cases produced results
with varying accuracy. When embarking on using vibrations-based DMID methods, the reader
is advised to follow guidelines similar to that presented earlier in this chapter to ensure accurate
results.
7.6.3 Case Study I: Theoretical Simulation of DMID Procedures

7.6.3.1 Introduction
Structural health monitoring is critical in several types of conventional civil engineering structures.
For example, offshore oil platforms need continuous monitoring for structural damage that might
occur below the waterline from extreme sea states and ship impact. Structural monitoring of bridges
is necessary as the effects of corrosion, ship traffic, earthquakes, and, sadly, even terrorists continu-
ously threaten the soundness of these vital structures. During the last two decades, attention has
been focused on using the vibration characteristics of a structure as an indication of structural dam-
age. See, for example, Pandey, A. K. and Biswas, M. (1994), Stubbs and Garcia (1996) and Faarar
and Jaurengui (1996). The vibration characteristics of a structure may be defined by its natural
modes of vibration. These depend only on the characteristics of the structure and not the excitation.
The vibration characteristics of a structure may be determined from taking measurements at one
or more points on the structure. Vibration signatures obtained before and after damage have been
inflicted by excessive forces may be utilized to locate and estimate the severity of damage.
A structural member develops strain as it deforms due to an applied load. If the load is excessive,
the yield strain (i.e., the strain above which the stress is no longer linearly proportional to the strain)
of the member may be exceeded. Therefore, if the change in length of a structural member divided
by its initial length is greater than the yield strain, the stiffness of a steel member will be reduced
typically by a factor of 10 to 20. For a structural steel member, the yield strain is a function of the
type of steel as well as the rate of loading. For typical commercial high strength low alloy steel with
a static yield stress of 50,000 psi, the static yield strain is 0.0017.
The physics behind the detection of damage through the measurement of vibration signatures is
based on the previously mentioned fact that if the strain in a structural member exceeds the yield
strain, the stiffness of the member is reduced. If the stiffness of the member is reduced, then the
characteristics of the vibration (natural frequency, mode shapes, and damping) of the member will
change as well as those of the complete structure.
Two challenging problems present themselves. First, the number and location of structural moni-
toring measurements ideally should be kept to a minimum. Second, correlating the results of the
measurements to a specific damage mechanism (e.g., hairline cracks, large localized strains, etc.) as
well as to the location of the damage is difficult.
Ettouney et al. (1998) investigated the applicability and accuracy of different damage detection
techniques to complex structures. A typical multi-jointed steel bridge will be considered. The dam-
age will be simulated analytically in the structural models, and the damage detection algorithms
will be applied to both the damaged and the undamaged structures.
The accuracy and efficiency of different damage detection schemes are investigated, including
the capability of detecting damage location as well as the severity and extent of the damage. It is
shown that the capability of the three algorithms to predict the location and damage levels depend
on the number of modes considered. In general, good qualitative predictions were observed from
all algorithms considered.
7.6.3.2 Vibration-Based DMID methods

7.6.3.2.1 General
Three health monitoring methods will be investigated in this study. The change in stiffness method,
the change in flexibility method, and the change in damage index (DI) method. All these methods
are based on the knowledge of the mode shapes of both the undamaged and the damaged structures
[ϕ]and [ϕ*], respectively. Also, the natural frequency diagonal matrices [Ω]and [Ω*]for both dam-
aged and undamaged structures are assumed to be known. The values of the mode shapes can be
measured and computed for the structure of interest both in the undamaged and the damaged states.
For the purpose of the current study, which is intended to investigate different analytical health
monitoring methods, the values of [Ω], [Ω*], [ϕ], and [ϕ*] will be calculated numerically. In what
follows, a summary of each of the three methods is presented. Note that the size of the mode shape
matrices is N⋅M, where M is the number of mode shapes considered in the problem, and N is the
number of degrees of freedom of the structure. In the rest of this section, an * will refer to the dam-
aged structure.
7.6.3.2.2 Change in Stiffness Method

Change of stiffness methods was investigated and discussed by several authors. Zimmerman and
Kaouk (1994), Reese (1995), and Farrar and Jauregui (1996) discussed several aspects of the method.
It is based on finding the change of the stiffness matrix of the structure [∆K], where
[ ΔK ] = [ K ] − [ K * ] (7.16)

The undamaged stiffness matrix [K] and the damaged stiffness matrix [K∗] can be calculated as
[ K ] = [f][Ω][f]T (7.17)

[ K * ] = [f* ][Ω* ][f* ]T (7.18)

Note that the matrix [∆K] contains stiffness terms, which include forces and bending moments.
Since it is difficult to mix forces and moments, it is proposed to use a normalized matrix [ �K ]
such that
⎡ ΔK ⎤ = [ ]
ΔK
⎣ ⎦ [K ] (7.19)

Note that the above division is a scalar division, such that ΔK ij = ΔK ij / K ij.
To simplify the investigation of the matrix [ �K ], the vector {DK} will be considered. The vector
{DK} contains only the diagonal terms of the matrix [ �K ]. The relative values of the components
of {DK} should indicate the location of the damaged parts of the structure. The statistical approach
used by Kim and Stubbs (1995a) to further isolate the damaged elements will be utilized here. The
differential stiffness index (DSI) will be introduced as
DK i − DK (7.20)
DSI i = ,
s DK

where DKi is the ith component of {DK}, DK is the mean of the vector {DK} and σDK is the standard
deviation of the vector {DK}. The subscript, i, has the range of i = 1, 2, … N. The index DSI can be
used to investigate the locations of the damage in the structure.
7.6.3.2.3 Change in Flexibility Method

Pandley and Biswas (1994) proposed the change of flexibility method. It is based on the computation
of the differential flexibility matrix of the damaged and the undamaged structure. The flexibility
matrix of the undamaged and the damaged structures can be computed as
[ F ] = [f][Ω]−1 [f]T (7.21)
−1 T
⎡⎣ F * ⎤⎦ = ⎡⎣f* ⎤⎦ ⎡⎣Ω* ⎤⎦ ⎡⎣f* ⎤⎦ (7.22)

From these expressions, the normalized differential flexibility matrix [ �F ], can be computed using
similar steps as in the differential stiffness matrix [ �K ]. The differential flexibility index (DFI)
can be computed using the components of [ �F ], following the procedure outlined in the previous
section.
7.6.3.2.4 Damage Index (DI) Method

Kim and Stubbs (1995a) developed DI method. Unlike the change of stiffness and change of flexibil-
ity methods, which uses the structural nodal degrees of freedom as their basic variable, it is based
on the structural elements as the basic variable. It was reported by Kim and Stubbs (1995b) that
it has the advantage of requiring small number of structural modes to obtain an accurate damage
detection results. The DI method results in a damage index DIi, where subscript i has the range of i
= 1, 2, … Ne, where Ne is the number of structural elements in the system. Excellent results using the
DI method were reported in numerous reports. Stubbs et al. (1995) and Farrar and Jauregnui (1996)
used it in an application to a highway bridge structure. Also, Garcia and Stubbs (1995) and Stubbs
and Garcia (1996) reported on the successful use of the DI methodology. A comprehensive descrip-
tion of the method can be found in Kim (1995a and 1995b) and Stubbs and Kim (1994).
7.6.3.3 Results
7.6.3.3.1 General
The three damage detection algorithms, which were described above, were applied to a complex
steel structure. Figure 7.73 shows a steel frame that is supported on five supports. The structure con-
tains 241 structural elements, and 78 structural connections. All the connections of the structure are
rigidly attached. The height of the structure is about 10 ft, and the longer dimension of the structure
is about 70 ft. The structure was modeled using finite elements so as to have accurate enough modes
up to a natural frequency of 250 Hz. The mode shapes [ϕ] and the natural frequencies [Ω], of the
base line, undamaged structure were calculated. Two damaged states of the structure were then con-
sidered. A structure with only one damaged element, as shown in Figure 7.74 and a structure with
10 damaged elements, as shown in Figure 7.75. The damaged elements were simulated numerically
by artificially specifying the modulus of elasticity of the damaged element Ed, as Ed = α Eu. Where
Eu is the modulus elasticity of the undamaged steel element. The magnitude of the factor α ≤ 1.0
represent the degree of damage in the element. Two damage levels were considered for this work, a
moderate damage case, with α = 0.5, and a severe damage case with α = 0.01. An essential factor in
FIGURE 7.73 Model of steel structure.

FIGURE 7.74 Model of steel structure, with one damaged member.
FIGURE 7.75 Model of steel structure, with ten damaged members.
any damage detection method is the number of structural modes M needed to accurately detect the
structural damage. Two cases are studied in this work, namely M = 14 and M = 140. Note that the
structural model has as many as 468 modes. It is desirable to limit the number of modes since this
translates directly into cost savings of any long-term health monitoring scheme.
7.6.3.3.2 Change in Stiffness Method

The change of stiffness method is applied to the structures of Figures 7.73 through 7.75. The DSI
were computed for the different conditions described above. For the case of one damaged element,
Figure 7.76 shows the DSI for α = 0.5 and both M = 14 and M = 140. Figure 7.77 shows similar results
for α = 0.01. Figures 7.78 and 7.79 show the same comparisons for the structure with 10 damaged
elements. In Figures 7.76 through 7.79 the zones of nodal degrees of freedom that are connected to
damage elements are highlighted. The DSI was fairly successful in detecting the location of the dam-
aged one element case. It also predicted with a lesser degree of accuracy, the locations of the damaged
10 element case. Even with only 14 modes, the method seemed to have acceptable results.
Location of zone for

Location of zone for
damaged degrees of
damaged degrees of
freedom
Differential stiffness index (DSI) freedom
Differential stiffness index (DSI)

12 12
10 10
8 8
6 6
4 4
2 2
0 0
–2 –2
100 200 300 400 100 200 300 400
Degree of freedom Degree of freedom
(a) 14 modes included (b) 140 modes included
FIGURE 7.76 Change of stiffness method, 1 damaged element; Ed/Eu = 0.50 (a) 14 modes, and (b) 140 modes.
Location of zone for Location of zone for

damaged degrees of damaged degrees of
freedom freedom
8 8
6 6
4 4
2 2
0 0
–2 –2
100 200 300 400 100 200 300 400
FIGURE 7.77 Change of stiffness method, 1 damaged element; Ed /Eu = 0.01 (a) 14 modes, and (b) 140 modes.
7.6.3.3.3 Change in Flexibility Method

Next, we apply the change of flexibility method to the structures of Figures 7.73 through 7.75. The
DFI were computed. For the case of one damaged element, Figures 7.80 and 7.81 show the DFI for
α = 0.5 and α = 0.01, each calculated for M = 14 and M = 140. Figures 7.82 and 7.83 show the same
comparisons for the structure with 10 damaged elements. The results are similar to those of the
change of stiffness method. In Figures 7.80 through 7.83, the zones of nodal degrees of freedom
which are connected to damage elements are highlighted. The DFI was fairly successful in detect-
ing the location of the damaged one element case. It also predicted with a lesser degree of accuracy
the locations of the damaged 10 element case. Increase in the number of modes from 14 to 140
Location of zones Location of zones

for damaged degrees for damaged degrees
of freedom of freedom
9 9
8 8

7 7
6 6
5 5
4 4
3 3
2 2
1 1
0 0
–1 –1
–2 –2
–3 –3
–4 –4
100 200 300 400 100 200 300 400
FIGURE 7.78 Change of stiffness method, 10 damaged elements; Ed/Eu = 0.50 (a) 14 modes, and (b) 140
modes.
Location of zones
Location of zones for damaged degrees
16 16
for damaged degrees of freedom

14 of freedom 14
12 12
10 10
8 8
6 6
4 4
2 2
0 0
100 200 300 400 100 200 300 400
FIGURE 7.79 Change of stiffness method, 10 damaged elements; Ed/Eu = 0.01 (a) 14 modes, and (b) 140
modes.
improved the results, although the 14 modes case method still produced acceptable results. The DFI
seems to predict the severity of the damage.
7.6.3.3.4 Damage Index Method

Finally, we investigate the DI method. The DI was computed for all the cases. Different results
are shown in Figures 7.84 through 7.87. Note that the DI is a function of each structural element.
Thus the DI would show the damage situation in each element, not a nodal degree of freedom. In
Figures 7.84 through 7.87 the different elements that are damaged are highlighted. In general, the
DI capabilities to predict the location and magnitude of damage are similar to those of the other two
methods described earlier.

freedom freedom
9 9
Differential flexibility index (DFI)

8 8
7 7
6 6
5 5
4 4
3 3
2 2
1 1
0 0
100 200 300 400 100 200 300 400
FIGURE 7.80 Change of flexibility method, 1 damaged element; Ed/Eu = 0.50 (a) 14 modes, and (b) 140 modes.

freedom freedom
10 10
9 9
8 8
7 7
6 6
5 5
4 4
3 3
2 2
1 1
0 0
100 200 300 400 100 200 300 400
FIGURE 7.81 Change of flexibility method, 1 damaged element; Ed/Eu = 0.01 (a) 14 modes, and (b) 140 modes.
7.6.3.4 Conclusions
Three different damage detection algorithms were investigated in this section. A realistic complex
steel structure was used for the evaluation. Different important parameters were studied. These
parameters included number of structural modes, number of damaged elements and the extent of
damage in each element. The three algorithms performed equally well in predicting both the loca-
tions of damaged elements and the extent of the damage.
Following these encouraging results, Ettouney et al. (1998) recommended to apply these algo-
rithms to more complex structures, and to increase the parametric studies so as to include more
important parameters, such as shape of damage and damages concentrated in structural joints. It is
also of interest to investigate the performance of these algorithms when the mode shapes included are
not consecutive. Experimental testing should eventually be used to verify the numerical testing.
Location of zones Location of zones

for damaged degrees for damaged degrees
of freedom of freedom
13 13

12 12
11 11
10 10
9 9
8 8
7 7
6 6
5 5
4 4
3 3
2 2
1 1
0 0
–1 –1
100 200 300 400 100 200 300 400
FIGURE 7.82 Change of flexibility method, 10 damaged elements; Ed/Eu = 0.50 (a) 14 modes, and (b) 140 modes.
Location of zones
Location of zones for damaged degrees
for damaged degrees of freedom
11 of freedom 11
10 10
9 9
8 8
7 7
6 6
5 5
4 4
3 3
2 2
1 1
0 0
100 200 300 400 100 200 300 400
FIGURE 7.83 Change of flexibility method, 10 damaged elements; Ed /Eu = 0.01 (a) 14 modes, and (b) 140
modes.
7.6.4 Case Study II: I-40 Bridge Experiment

In a field experiment by Farrar et al. (1994) and Farrar and Jauregnui (1996), different DMID algo-
rithms were applied to a bridge spanning highway I-40 in NM. The I-40 Bridge was a steel girder
bridge supporting a reinforced concrete deck. An intentional damage was introduced to the steel
girders to simulate fatigue cracks. Different vibration-based damage detection algorithms were then
employed to see how well those algorithms fare in detecting the damage. Four damage scenarios
were induced at the center of one of spans of the steel girder as shown schematically in Figure 7.88.
After each damage state is introduced, the bridge was excited using a single hydraulic shaker that
produced a 2000 pounds peak-force uniform random signal in the range of 2–12 Hz. Accelerations
at different locations were measured. The test was designed to record a coarse set of accelerations
Location of Location of
6 damaged element 6 damaged element
4 4
2 2
Damage index (DI)
Damage index (DI)

0 0
–2 –2
–4 –4
–6 –6
–8 –8
–10 –10
50 100 150 200 50 100 150 200
Element number Element number
FIGURE 7.84 Damage index method, 1 damaged element; Ed /Eu = 0.50; (a) 14 modes, and (b) 140 modes.
Location of
4 4 damaged element
2 2
Damage index (DI)
Damage index (DI)
0 0
–2 –2
–4 –4
–6 Location of –6
damaged element
–8 –8
–10 –10
50 100 150 200 50 100 150 200
FIGURE 7.85 Damage index method, 1 damaged element; Ed /Eu = 0.01; (a) 14 modes, and (b) 140 modes.
8 damaged elements damaged elements
8
6 6
4 4
Damage index (DI)
Damage index (DI)
2 2
0 0
–2 –2
–4 –4
–6 –6
50 100 150 200 50 100 150 200

FIGURE 7.86 Damage index method, 10 damaged elements; Ed/Eu = 0.50; (a) 14 modes, and (b) 140 modes.
damaged elements damaged elements
6 6
4 4
Damage index (DI)
Damage index (DI)

2 2
0 0
–2 –2
–4 –4
–6 –6
–8 –8
50 100 150 200 50 100 150 200
FIGURE 7.87 Damage index method, 10 damaged elements; Ed/Eu = 0.01; (a) 14 modes, and (b) 140 modes.
Damage scenario ‘D1’ Damage scenario ‘D2’

Partial web damage Half-height web
in the middle damage
Damage scenario ‘D3’ Damage scenario ‘D4’

Partial web and The bottom half of the
flange damage girder is damaged
Imposed damage states of the I-40 experiment
FIGURE 7.88 Damage states of I-40 Bridge.
at uniform stations along the whole bridge. Then a more refined set of accelerations were measured
near the intentional damage location. A standard modal identification technique was then employed
to identify the bridge vibration characteristics. As many as six modes were identified with their
corresponding modes shapes, natural frequencies, and damping parameters. Those modes were
identified for five bridge states: pristine state and the four intentional damaged states. The natural
frequencies ranged from 2.3 Hz for the first mode to 4.69 Hz for the sixth mode. The damping ratio
ranged from 0.62% to 1.6%. For more details on the experiment and its results see Farrar et al.
(1994), Farrar and Jauregnui (1996), and Farrar and Doebling (1999).
What concerns us in this section is the application of several vibration-based damage detection algo-
rithms to the I-40 results that were reported by Farrar and his colleagues. They employed five damage
detection algorithms in trying to identify the known damages of Figure 7.88. We describe briefly the
five algorithms, and then report on how those algorithms fared in detecting the I-40 experiment.
Damage Index Method: See description in previous section.
Mode Shape Curvature Method: This method is based on computing the analytic expression
of beam curvature, u″(x), in the form
M ( x)
u ʹʹ( x ) = (7.23)
EI
The bending moment of the girder at station x is M(x). The modulus of elasticity and the governing
moment of inertia are E and I, respectively. Equation 7.23 is used to identify the damage, as reflected
in the flexural stiffness EI from the measured beam curvature u″(x). See Pandey et al. (1991) for
more details on this method.
Change in Flexibility Method: See description in previous section.
Change in Uniform Load Surface Curvature: The method was suggested initially by Zhang and
Aktan (1995). It is also based on computing curvature of the bridge girder with the implied assumption
that it behaves as a beam. Simply stated, the method uses the test results to produce a deflected shape
that is the resultant of a uniform loading on the bridge. It achieves that by adding all the columns in
the flexibility matrix. When the experimental uniform deflected shape is computed, the curvature at
each point can be computed using central difference approach. The difference between the pristine
curvature and the damaged curvature is now computed. That curvature difference can be related to a
quantified damage since we know the analytical expressions of the bending of beams.
Change in Stiffness Method: See description in previous section.
The results of applying the five vibrations-based damage detection methods to the four damage
scenarios are shown for the coarse and refined accelerometers sets in Tables 7.27 and 7.28, respec-
tively. It is clear that the accuracy of damage detection using different methods vary greatly. For
in-depth discussions and conclusions of this informative and important experiment, see Farrar et al.
(1994), Farrar and Jauregnui (1996), and Farrar and Doebling (1999).
The I-40 experiment was an important event in the evolution of SHM in civil infrastructure. It shed
light on the different vibration-based DMID methods as stated above. Farrar and Doebling (1999)
stated some of the lesson learned from that experiment: (a) Pretest visual inspection is important; (b)
During the STRID phase, it is important to ensure that the results are both reciprocal and linear, since
such assumptions are built-in most STRID methods; (c) Sensitivity of the modal identification method
to environmental and testing procedures should be investigated; and (d) False-posivite analysis should
be performed to ensure that the procedures will not identify pristine systems as damaged.
For our current purposes, we note that all the five methods were based on STRID modal identification
techniques for both pristine and damage systems. After the STRID process, the DMID methods were
applied. Earlier in this chapter we introduced a guide for using STRID as a DMID tool. We feel that this
guide is applicable to all of the above five methods. It would be of interest to compute (quantify) the errors
involved in damage using vibration-based approach for different types and sizes of damages. It should
also be noted that the damages incorporated in the case studies are known before the testing. In many
cases, the identified damages are relatively big.
7.7 SIGNAL PROCESSING AND DMID

7.7.1 Time Signals
The use of time signals (defined as a time-varying function) in the field of SHCE is extensive.
Invariably, at some step of the process, the need to record and analyze time-dependent signals
arises. The signal can be an input signal (earthquakes, blast pressures, fluid velocity, wind pressures,
TABLE 7.27
Results of Vibration-Based DMID Methods, Coarse Sensors Set
DMID Method
Change in
Mode Uniform
Damage Shape Change in Load Surface Change in
Index Curvature Flexibility Curvature Stiffness
Damage Scenario Method Method Method Method Method
2 2 X X 2
D1:
2 X X X 2
D2:
2 2 2 X 2
D3:
1 1 1 1 1
D4:
Note: 1—Damage located; 2—damage located using only two modes; 3—damage not located.
Source: Farrar, C. et al., Dynamic characterization and damage detection in the I-40 Bridge over the Rio Grande, Los
Alamos National Laboratory Report LA-12767-MS, 1994.
etc.) or an output signal (measured responses such as accelerations, strains, displacements, etc.).
Also, many of the NDT, STRID methods are based on measuring and analyzing signals. This
shows that we need to have adequate knowledge of signal analysis for appropriate decision making
in SHCE.
One of the basic requirements for any signal processing method is that the method must be com-
patible with the signal properties. An important signal property is whether it changes its character-
istics (such as RMS) as time passes. Stationary signals do not change their characteristics with time.
Nonstationary signals change their characteristics. Most of signals that affect civil infrastructure
are nonstationary, such as earthquakes or blast pressures. Also, by definition, any nonlinear behav-
ior would produce nonstationary signal. Sometimes the nonstationary requirement is dropped for
the sake of simplicity; earthquake motions and responses are usually analyzed by FT, which is not
suited to analyze nonstationary signals.
TABLE 7.28
Results of Vibration-Based DMID Methods, Refined Sensors Set
Damage Scenario DMID Method
Change in
Mode Uniform Change
Damage Shape Change in Load Surface in
Index Curvature Flexibility Curvature Stiffness
Method Method Method Method Method
1 3 X X X
D1:
1 2 X 3 X
D2:
1 1 X 1 X
D3:
1 1 1 1 1
D4:
Note: 1—Damage located; 2—damage narrowed down to two locations; 3—damage narrowed down to three locations;
X—damage not located.
Source: Farrar, C. et al., Dynamic characterization and damage detection in the I-40 Bridge over the Rio Grande, Los
Alamos National Laboratory Report LA-12767-MS, 1994.
This section explores four popular signal processing methods: FTs, time-frequency transforms
(TFT), wavelet transforms (WT), and Hilbert–Hwang transforms (HHT). The theoretical basis of
each method is overviewed; the advantages/disadvantages and some examples of their use in STRID
and DMID are presented.
7.7.2 Fourier-Based Transforms

7.7.2.1 Conventional Fourier Transforms
The frequency content of a continuous time signal x(t) can provide valuable information during an
SHM experiment. Perhaps the most popular tool for investigating the frequency content is its FTs.
It can be expressed as
+∞
FT ( f ) = ∫ x(t ) e − i 2 p f t dt (7.24)
−∞
The frequency spectrum FT(f) is a complex illustration of the signal behavior at a given frequency
f. Figure 7.89 illustrates the mathematical principles of FT( f ). One of the reasons of the popularity
of FTs is that they exist also for discrete time signal x(n) as
∞
DFT ( f ) = ∑ x ( n) e −i n2p f (7.25)
n = −∞
assuming that
1. The discrete time signal is discretized with a time step of ∆t

2. The total time steps is N
Then following Heisenberg uncertainty principle, Hubbard (1998), DFT( f ) is discretized with a
frequency step of ∆ f with
1 (7.26)
Δf =
N Δt
One of the advantages of the FT method is that the inverse transform exist. The continuous trans-
form is expressed as
∞
x(t ) = ∫ FT ( f ) ei 2 p ft df (7.27)
−∞
Traditional fourier
For a given signal
transform concepts
Time
For a given frequency within the frequency range
Define a complex harmonic function that spans the whole signal time length
(imaginary part is dashed in the above figure)
The harmonic function is integrated at each time position with the signal
Repeat for a different frequency
The process is repeated until the frequency range of interest is covered. The resulting complex
intervals represent the desired frequency spectrum is the windowed Fourier spectrum of the signal
FIGURE 7.89 Principles of Fourier transform (FT).

The discrete transform is

∞
− i n 2 pn
x( f ) = ∑ DFT ( n) e (7.28)
n = −∞
Thus, other advantages of the FT method are as follows:
• Can be used to analyze many applications

• Well understood and used by professionals
• Inverse transforms exist, as shown above
FT method is a very efficient analytical technique for evaluating Equation 7.25, for discrete FT, and
Equation 7.28 for discrete inverse transform. The method is known as fast Fourier transform, FFT
in short.
From Equation 7.26 it is clear that the Heisenberg principle puts severe limitations on the fre-
quency or the time step of the signal. This has important implications during SHM experiments.
For example, at low frequencies, the frequency step is relatively high, thus the resolution of the
frequency spectrum is low. Another serious disadvantage of the Fourier spectrum is that it is based
on harmonic functions only. The harmonic basis can have inaccurate effects on signals with short
time duration, see Pendat and Piersol (2010). Perhaps the most limiting disadvantage is that the FT
(or FFT) spectra do not reveal any information regarding the time dependency of frequency content
of the signal. Thus, the FT method is not applicable for nonstationary signals (wind, flood, or earth-
quake signals) or nonlinear problems (such as analyzing failure potential of brittle or potentially
unstable systems).
7.7.2.2 Time and Frequency Transforms

Before we continue to explore other methods of signal processing, we need to discuss further a very
important derivative of FT, namely, the power spectral density (PSD). The PSD is defined as
1 2
G( f ) = FT (7.29)
2T
Where G( f) is the PSD. Mathematically, it can be shown that for a given frequency width, say
∆ f, the area represented by ∆f • |FT|2 is the mean square of the harmonic amplitude at such fre-
quency (see Figure 7.90 for illustration). Thus, the PSD is a representative of the energy within a
PSD
Frequency scale
FIGURE 7.90 Physical illustration of PSD.

particular spectrum width. This is an important fact that can be used to understand and analyze
time signals. Similar to the parent FT spectrum, we note a major limitation to the use of PSD
in signal processing. The time dependency of the energy occurrence is lost. We only know the
energy situation at a given frequency; we do not know when such energy occurred during the
time signal.
There are several other derivatives of FT, such as correlation, coherence, and cross-correla-
tion functions. For detailed description of those derivatives, the reader can refer to Pendat and
Piersol (2010).
7.7.2.3 Time-Fourier Transforms (TFT)

To avoid the FT limitation of losing temporal information of frequency content, the TFT was intro-
duced, see Boashash (1992). The time signal x(τ) is windowed by a window function h(τ). The win-
dowed function is then centered about time t. We are now interested in the transform of x(τ)h(τ − t)
such that
+∞
TFT ( f , t ) = ∫ x(t)h(t − t ) e − i 2 p f tt d t (7.30)
−∞
The frequency-time spectrum TFT(f, t) is now an illustration of the signal behavior at a given fre-
quency f and a given time t. Figure 7.91 illustrates the mathematical principles of FT(f). Note that a
discrete form of Equation 7.30 can be evaluated. We also note that Equation 7.26 still applies due to
Heisenberg’s principle. Similar to FT, the inverse transform of TFT exists.
The TFT resolves one of the limitations of the FT method. The TFT spectrum can now reveal
useful information regarding the time dependency of frequency content of the signal. For example,
time of excitation of particular vibration mode during a given experiment. Unfortunately, since TFT
is still dependent on a fixed-length widowed harmonic function, TFT method is still not applicable
for nonstationary signals or nonlinear problems. To summarize, the TFT method has the following
advantages:

• There are both continuous and discrete
Among the disadvantages of the TFT method are the following:
• Loses frequency resolution at low frequency (Heisenberg uncertainty principle)

• Loses time resolution at high frequency (Heisenberg uncertainty principle)
• Only harmonic functions
• Similar to the FT method, the TFT method is not applicable for nonstationary signals or
nonlinear problems
7.7.3 Wavelets
To resolve several of the FT or TFT limitations, another method of signal processing was developed
in the past few decades: the WT, see Graps (1995) and Hubbard (1998). The continuous wavelet
transform for a time signal x(t) is
WT ( s, t) = ∫ x ( t ) y ( s, t, t ) dt (7.31)

For a given signal.
Time
Choose appropriate Window time length
The frequency of the harmonic function in the window is changed

The function is complex (imaginary part is dashed in the above figure
For a given The window is moved along time and the function is
frequency of integrated at each time position with the signal
the windowed
harmonics
The position of the

translated function
is the time axis in the
windowed Fourier spectrum
The frequency
is changed
The process is repeated until the frequency range of interest is

covered. The resulting time and frequency spectrum is the windowed Fourier
spectrum of the signal
FIGURE 7.91 Windowed Fourier transform.
The basis function (also known as wavelet function, or mother wavelet) ψ(s, τ, t) include a scaling
and translation variables s and τ, respectively; formally it can be expressed as
1 ⎛ t − t⎞ (7.32)
y (s, t, t ) = y⎜ ⎟
s ⎝ s ⎠
Upon inspection, it is clear that the scaling factor controls the amplitude of the wavelet function.
The translation factor controls the location of the wavelet function on the time scale. As the scaling
factor changes, the effective width of the wavelet function changes. The translation factor control
the sweep of the wavelet function along the time scale. See Figure 7.92 for illustration of WT basic
principles. There are several forms of wavelet functions will be discussed in Section 7.7.3.2. This
is perhaps the most important distinction between WT and FT/TFT: the wavelet functions can be
designed to accommodate the particular problem on hand. The FT/TFT are based on harmonic
functions.
There is an inverse WT such that
x(t ) = ∫ ∫ WT ( s, t) y (s, t, t )ds d t (7.33)

For a given signal
Time
Choose appropriate wavelet function
In this example, the Mexican hat function is chosen
The wavelet The wavelet function is "translated" along time and

function is integrated at each time position with the signal
scaled
The scale, S, The position of the

is related to translated function
frequency is the time axis in the
wavelet spectrum
The wavelet The wavelet function is "translated" along time and

function is integrated at each time position with the signal
scaled again
The position of the

translated function
is the time axis in the
wavelet spectrum
The process is repeated until the frequency range of interest (scale range) is
covered. The resulting time and frequency spectrum is the wavelet spectrum of
the signal
FIGURE 7.92 Basic principles of wavelet transforms (WT).
The frequency-time spectrum WT(s, τ) is now an illustration of the signal behavior at a given scale
(frequency) s and a given translation (time) τ. Note that a discrete form of WT Equation 7.30, and
the inverse WT, Equation 7.33, can be evaluated.
Some of the advantages of the WT method are as follows

• There are both continuous and discrete forms
• Do not have the FT/TFT limits on frequency or time resolutions. Controlling the scal-
ing and translation factors would result in a good frequency-time resolution compromise.
This is a reasonable accommodation to the Heisenberg uncertainty principle, as discussed
later.
• Can apply to nonstationary and nonlinear problems. Can also be used for pattern recogni-
tion and data compression
7.7.3.2 Wavelet Functions

One of the many differences between TFT and WT is the flexibility of the wavelet functions h(τ).
Wavelet functions vary in shape as well as in time width. They can be chosen to better suite the
signal under consideration. Examples of wavelet functions are the following
Morlet Wavelet: They are complex in shape and resemble a decaying harmonic wave form, They
are used in many applications in SHM field such as wind and earthquake analysis (Figure 7.93).
Haar Wavelet: This wavelet is the simplest wavelet form (Figure 7.94). It is not a continuous
function, yet its simplicity makes it a favorite for illustration as well as sensitivity studies.
Mexican Hat Wavelet: This wavelet (Figure 7.95) has simple analytical expression. It is used
in several physics and engineering applications such as astronomy, ocean engineering, and seismic
waves.
Morlet wavelets
0.0003
0.0002
0.0001
Amplitude
–0.0001
–0.0002
–0.0003
–4 –3 –2 1 0 1 2 3 4
Time
Amplitude Real Imaginary
FIGURE 7.93 Morlet wavelets.
Haar wavelets
1.5
0.5
Amplitude
–0.5
–1
–1.5
–0.5 –0.3 –0.1 0.1 0.3 0.5 0.7 0.9 1.1 1.3 1.5
Time
FIGURE 7.94 Haar wavelets.

Mexican Hat wavelet
0.5
0.4
0.3
0.2
Amplitude
0.1
–0.1
–0.2
–0.3
–4 –3 –2 –1 0 1 2 3 4
Time
FIGURE 7.95 Mexican hat wavelets.
Daubechies Wavelets: One of the features of Daubechies wavelets is that they are produced
numerically by iterations; they are not analytical functions. However, these types of waves are not
finite in width, they cover the whole signal. Because of this, they do not have the truncation prob-
lems experienced by other types of wavelets.
7.7.3.3 Heisenberg Principle

In discussing the Heisenberg principle earlier, we showed that it relates time step, frequency step,
and number of time samples for FT processes. A generalized form of Heisenberg principle is
Δ t i Δf = A (7.34)
where A is a constant for the specific process under consideration; Figure 7.96 shows the implica-
tions of the Heisenberg principle for different signal processing techniques:
Time Signal: High temporal resolution, but no information about the frequency content of the
signal.
FT: Can produce high frequency resolution, but temporal information is missing.
TFT: The equal time and frequency resolutions produces coarse information at the tail ends of
both time and frequency scales.
WT: The relative time and frequency resolutions changes along the time and frequency scales.
WT produces higher time resolution at lower frequency range. It also produces higher time resolu-
tion at higher frequency range. This behavior produces more accuracy than other processes.
7.7.3.4 Basic Example of DWT

Let us consider now a simple example of DWT. The signal
f (t ) = f1 (t ) + f2 (t ) + f3 (t ) (7.35)

is sampled at time step of ∆t with sample size of N = 2 m = 2048, which means m = 11. We assume
⎛ a p (i Δ t ) ⎞
f1 (i Δ t ) = sin ⎜ 1 (7.36)
⎝ 180 ⎟⎠

Heisenberg uncertainty
principle:
Heisenberg boxes
Frequency
Frequency
Heisenberg
boxes have constant
area in the time-
frequency
Time spectrum. Time
a - Time signal: High time resolution b - Frequency spectrum: High frequency
but no information about frequency (∆t).(∆f ) = Constant resolution but no information about time
Higher time resolution

at higher frequency
range (when signal changes
rapidly in time)
Frequency
Frequency
Higher frequency resolution

at lower frequency
range (when signal changes
slowly in time)
Time Time
c - Windowed Fourier spectrum: Equal time and d - Wavelet spectrum: variable time and frequency
frequency resolution in the time-frequency spectrum resolution in the time-frequency spectrum
FIGURE 7.96 Heisenberg boxes for different signal transform methods.
f2 (i Δ t ) i < 512 = 0 ⎫
⎪⎪
⎛ a2 p ( i Δ t ) ⎞ ⎬ (7.37)
f2 (i Δ t ) i ≥ 512 = sin ⎜ ⎪
⎝ 180 ⎟⎠ ⎪⎭

f3 (i Δ t ) i < 1024 = 0 ⎫
⎪⎪
⎛ a3 p (i Δ t ) ⎞ ⎬ (7.38)
f3 (i Δ t ) i ≥ 1024 = sin ⎜ ⎪
⎝ 180 ⎟⎠ ⎪⎭

The sampled function f(i∆t) is shown in Figure 7.97a. For our purpose, there are two main features
of f(i). First, it contains three distinct frequencies (a1 π/180), (a2 π/180) and (a3 π/180). Second fea-
ture is that the three signals f1, f 2, and f 3, arrive at three distinct times t = 0, t = 512, and t = 1024,
respectively.
Performing an FFT on f(i) will produce the discrete Fourier spectrum of Figure 7.97b. The
Fourier spectrum shows, as expected, the three distinct frequencies along the frequency scale. Also,
as expected, the times of arrival of the three component signals are nowhere to be found in the
FFT. Performing WT on the signal produced the spectra in Figure 7.98. The arrivals of the three
sinusoids are clearly identified in an accurate manner, especially in the contour of Figure 7.98a. The
head-on view of Figure 7.98c (the amplitude-scale view) resembles the FT spectrum of Figure 7.97b.
It shows the three distinct frequencies, but the times of arrival of the signals are clear. The superior
performance of WT as compared to the FT is evident.
2.67
Amplitude (i)
1.33
–1.33
–2.67
–4
0 1500 3000 4500 6000 7500 9000
Time - (i)
a - Time signal
50
40
Amplitude (i)
30
20
10
0
0 20 40 60 80 100 120 140 160
Frequency
b - FT of the signal
FIGURE 7.97 Three-sinusoids signal (a) time signal and (b) FT of the signal.
7.7.3.5 Applications of Wavelets in Damage Identification

The flexibility and ease of WT in extracting information from signals and measurements resulted in
numerous applications by users and researchers. For example, Spanos and Failla (2004) developed a
wavelets-based method to estimate the evolutionary power spectral density (EPSD) of nonstationary
stochastic processes. The method can help in earthquake engineering studies by accommodating
the nonstationary property of earthquake motions. Yan et al. (2004) used WT for STRID of bridge
structures. They also considered potential uncertainties in their work while applying the WT tech-
nique. Another innovative utilization of WT was offered by Duan et al. (2004). They used WT as a
teaching mechanism for neural networks.
Hamstad et al. (2002) showed a practical example on using WT for detecting damage in plates.
The experiment was based on measuring AE signals from damages, then identifying damage infor-
mation using WT. The experimental setup is shown in Figure 7.99. Seven AE sensors are placed on
the plate. An induced damage was placed inside the plate under the middle sensor (#4). The identi-
fication process is based on identifying dispersion curves of the plate of interest. Figure 7.100 shows
typical plate dispersion curves, Rose (2004). The WT and dispersion curves are utilized during the
experiment as follows: (a) measure emission signals at different sensor locations, (b) use WT to
produce time-frequency spectrum, (c) produce theoretical dispersion curves, and (d) match appro-
priate dispersion curve with the WT spectrum. Figure 7.101 shows a typical emission signal, its WT
spectrum overlaid on a symmetric (S 0) and asymmetric (A0) dispersion lines. A very good match
between the experimental measurements and theoretical computations is shown. Since theoretical
(a) (b)
Amplitude
20
10
10
5 10 Scale
Scale
5000
0 8000
0 2000 4000 6000 8000
Time step (i) Time step (i)
WT: Contour view WT: 3D view

(c)
20
Amplitude
10
0
0 5 10 Scale
WT: Amplitude-scale view
FIGURE 7.98 WT of three-sinusoid signal (a) contour view, (b) 3D view, and (c) 2D view.
1000
500
Source below
Sensor sensor 4
numbers
500
1 2 3 4 5 6 7
1000
90 30
25.4
FIGURE 7.99 Testing specimen. All dimensions are in mm. The sensors are spaced 60 mm. (From Hamstad,
M. A. et al., J. Acoustic Emission, 20, 39-61, 2002. With permission.)
curves are associated with specific stress wave with known wave velocity, and wave number, a
simple computation can reveal the distance between the sensor location and the source of the signal,
thus identifying the location of the damage. Figure 7.102 shows the conventional FT spectrum of
the same signal. Clearly the FT does not provide any information regarding the damage location.
Figure 7.103 shows the WT spectra as captured by different sensors. The effects of distance on the
signals are evident. This shows that a triangulation scheme can pinpoint unknown damage location
6 S5 S6
S0 S4
S3
5
S1 S2
cg (mm/µs)
4
3
A0
2
A3
1 A6
A1 A2 A4 A5
0
0 2 4 6 8 10 12 14 16 18 20
f d (MHz-mm)
FIGURE 7.100 Dispersion curves (group velocities) of a plate. (From Rose, J., Ultrasonic Waves in Solid
Media, Cambridge University Press, Cambridge, UK, 2004. Reprinted with the permission of Cambridge
University Press.)
File: c\nasa_database14_2700tra Chan: 1 Set: 1 Index: 1 0 0:00:00 0.0000 Time[µs]

4
3
Amplitude [mY]
2
1
0
–1
–2
–3
–4
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140

1000
950
900 A0
850
800
750
700
650
Frequency [kHz]
600
550
500
450 S0
400
350
300
250
200
150
100
50
0
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
A0 S0 Time [µs]
FIGURE 7.101 Recorded AE signal and its wavelet spectrum. Note the correlation with first mode disper-
sion curves A0 and S0. Wavelet scales are 0–150 μs, and 0–1.0 MHz. (From Hamstad, M. A. et al., J. Acoustic
Emission, 20, 39-61, 2002. With permission.)
with accuracy. Figure 7.104 shows that the method is also applicable for detecting damage locations
that are also deep within the plate thickness. Finally Figure 7.105 shows the equivalent FT of the
measure signals. Again, the FT signals fail to provide information regarding the location of the
damage. This experiment shows that careful integration of signal proceeding technique, theoretical
analysis of wave motion (in a way, this can be considered STRID step), and NDT technique (AE)
can provide accurate and informative damage detection methodology.
160
120
Amplitude, fm
80
40
0
0.0 0.2 0.4 0.6 0.8 1.0
Frequency, MHz
FIGURE 7.102 Fourier spectrum of the AE signal of Figure 7.101. (From Hamstad, M. A. et al., J. Acoustic
Emission, 20, 39-61, 2002. With permission.)
7.7.4 Hilbert–Hwang Transform

7.7.4.1 General
Hilbert–Hwang transform is an increasingly popular method for signal analysis processes. It prom-
ises to have many successful applications in the field of SHM, especially when applied to civil infra-
structure. The method was described in detail by Hwang et al. (1998). This section summarizes the
essential theoretical basis of the method. Applications to SHM field are provided next.
7.7.4.2 Instantaneous Frequency

Hilbert transform xˆ (t ), of an arbitrary time series x(t) can be defined as
1 +∞ x ( t)
x̂ ( t ) = P∫ dt (7.39)
p −∞ t − t
Where P is the Cauchy principal value. Note that x(t) and xˆ (t ) form a complex conjugate pairs such
that an analytic series can be defined as
z ( t ) = x ( t ) + i xˆ ( t ) (7.40)
Defining
0.5
(
A ( t ) = x 2 ( t ) + xˆ 2 ( t ) )
(7.41)
⎛ x (t ) ⎞
q(t ) = tan −1 ⎜ ⎟ (7.42)
⎝ x (t ) ⎠
we end up with
z ( t ) = A ( t ) ei q(t ) (7.43)
Hwang et al. (1998) showed that an instantaneous frequency, ω can be defined as
dq( t )
w= (7.44)
dt
1000
950
900
850 60 mm
800
750
700
650
Frequency [Hz]
600
550
500
450
400
350
300
250
200
150
100
50
0
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80
Time [µs]
1000
950
900
850 120 mm
800
750
700
650
Frequency [Hz]
600
550
500
450
400
350
300
250
200
150
100
50
0
0 10 20 30 40 50 60 70 80 90 100 110 120
Time [µs]
1000
950
900 180 mm
850
800
750
700
650
Frequency [Hz]
600
550
500
450
400
350
300
250
200
150
100
50
0
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
Time [µs]
FIGURE 7.103 Effects of source distance on wavelet specturm. The time scales in the three spectra are
0–80 μs, 0–120 μs, and 0–150 μs, from top to bottom, respectively. The frequency scales are all the same
0–1.0 MHz. (From Hamstad, M. A. et al., J. Acoustic Emission, 20, 39–61, 2002. With permission.)
if the function A(t) satisfy certain conditions:

1. The number of extreme values and the number of zero crossings must be either equal, or
differ only by one at most.
2. The mean value of the envelopes that define the local maxima and the local minima at any
point is zero.
A function that satisfies the above two conditions is defined as an intrinsic mode function (IMF).
A simple method of decomposing the time series x(t) into a set of IMF components is described next.
1000 1000
950 950
900 900
850 850
800 800
750 750
700 700
650 650
Frequency [Hz]
Frequency [Hz]
600 600
550 550
500 500
450 450
400 400
350 350
300 300
@ 2.35 mm, 250 @ 1.41 mm
250 200
200 mid-plane
150 150
100
100 50
50 0
0 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
Time [µs] Time [µs]
1000 1000
950 950
900 900
850 850
800 800
750 750
700 700
650
Frequency [Hz]
650
Frequency [Hz]
600 600
550 550
500 500
450 450
400 400
350 350
300 300
250 @ 1.097 mm
250 @ 2.037 mm 200
200 150
150 100
100 50
50 0
0 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
Time [µs]
Time [µs]
1000
1000 950
950 900
900 850
850 800
800 750
750 700
700 650
Frequency [Hz]
Frequency [Hz]
650 600
600 550
550
500 500
450 450
400 400
350 350
300 300
250 @ 0.47 mm
250 @ 1.723 mm
200 200
150 150
100 100
50 50
0 0
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150
Time [µs] Time [µs]
FIGURE 7.104 Effect of depth of AE source. The measurements were all at horizontal distance of 180 mm
from AE source. (From Hamstad, M. A. et al., J. Acoustic Emission, 20, 39-61, 2002. With permission.)
7.7.4.3 Sifting Process

An ingenious process that is called empirical mode decomposition (EMD) was devised by Hwang
et al. (1998) that identify IMF components from an arbitrary signal x(t), as in Figure 7.106a. The
steps are as follows:
1. Assign i = 1, where i is the counter of the IMF component of interest

2. Assign working signal, Wij(t) = x(t)
3. Define a sifting counter, j = 1
4. Define X MAX(t) as a cubic spline function that connects all the local maxima in Wij(t). Note
that X MAX(t) now represents the upper envelope of Wij(t) (see Figure 7.106b)
5. Define X MIN(t) as a cubic spline function that connects all the local minima in Wij(t). Note
that X MIN(t) now represents the lower envelope of Wij(t) (see Figure 7.106b)
6. Define the average mij (t ) = (( X MAX (t ) − X MIN (t )) / 2) (see Figure 7.106c)
7. Define an initial estimate for the ith IMF as hij = Wij(t)−µij(t) (see Figure 7.106d)
8. The estimate hij(t) needs to be improved because: (a) riding waves exist, in the estimate,
and they need to be removed, and (b) it needs to have smoother uneven amplitudes. This
improvement can be done by increasing the sifting counter by 1 as j = j + 1.
9. Assign new working signal Wij(t) = hij(t)
10 500
@ 2.35 mm,
0 250 mid-plane
–10 0
10 500
@ 2.037 mm
0 250
–10 0
10 500
Spectral amplitude, fm
@ 1.723 mm
Displacement, pm
0 250
–10 0
10 500
@ 1.41 mm
0 250
–10 0
10 500
@ 1.097 mm
0 250
–10 0
10 500
@ 0.47 mm
250
0
–10 0
20 50 80 110 140 0 0.2 0.4 0.6 0.8 1
Time, µs Frequency, MHz
FIGURE 7.105 Measured signals and their Fourier spectra that correspond to Figure 7.104 plots. (From
Hamstad, M. A. et al., J. Acoustic Emission, 20, 39-61, 2002. With permission.)
10. Repeat steps 4 through 9 until the standard deviation SD between two consecutive sifting
signals as defined by SD = ∑ t (| Wi ( j −1) (t ) − Wij (t ) |2 / Wi 2( j − i ) (t )) is less than a preset limit.
Such a limit can be in the range of 0.2–0.3. When that limit is reached, the ith IMF signal
can be defined as ci = Wij
11. Assign i = i + 1
12. Repeat steps 3 through 10, until the new working signal Wij(t) becomes too small (less
than a preset tolerance level), or when it becomes monotonic, where no more IMF can be
extracted. In such a situation a residual, rn(t) can be defined as rn(t) = cn−1 − Wnj. Note that
n = i: the number of IMF components of the original signal x(t)
From the results of the above steps, the original signal can now be expressed as
i=n
x ( t ) = ∑ ci ( t ) + rn ( t ) (7.45)
i =1
The residue rn(t) can be a small number, or a trend. All empirical components cn(t) have the proper-
ties of IMF.
7.7.4.4 Hilbert Spectrum

The final step in evaluating the Hilbert–Hwang spectrum is to compute the Hilbert transform to each
of the IMC components and computing the instantaneous frequency according to Equation 7.44.
Thus the original signal can be expressed as
n
x (t ) = ∑ a j (t ) e y (7.46)
j =1
(a) Original signal
Upper envelope
Lower envelope
(b) Upper and lower envelopes
Averaging of envelopes Upper envelope
Lower envelope
(c) Average of envelopes
(d) Reduced signal
FIGURE 7.106 Steps for generating IMF (a) original signal, (b) upper and lower envelopes, (c) averages of
envelopes, and (d) reduced signal.
with
y = −1 ∫ w j ( t ) dt (7.47)

Hwang et al. (1998) observed that the IMF expansion of x(t) as shown in Equation 7.46 and Equation
7.47 is a generalized expression of the conventional discrete FT as expressed earlier in this section,
which we repeat for convenience as
∞
iw j t
x (t ) = ∑ a j e (7.48)
j =1
The differences between FT and IMF expansions are obvious:
• The FT produces constant amplitude for each component, while the IMF expansion pro-
duces time-dependent amplitude for each component.
• Similarly, the FT is based on constant frequency for each Fourier component, while the
IMF expansion produced a time-dependent instantaneous frequency for each component.
The just-discussed time-frequency spectrum is called Hilbert–Hwang spectrum, for consistency we

refer to it as HHT. It was shown to have the following advantages

• Applicable for nonstationary signals or nonlinear problems
• The adaptivity of the HHT makes it problem dependent, which promises to produce more
physically meaningful results than other more mathematically rigid transforms
• Adaptivity of the HHT makes it produce adequate frequency and time resolution to the
problem on hand within the time and frequency ranges of interest
7.7.4.5 Use in SHM: Earthquake Applications

The use of HHT in studying earthquake time signals was illustrated by Hwang et al. (1998).
Consider the El Centro earthquake acceleration time history of Figure 7.107. Using the procedures
of IMF evaluations, the ten IMF components of the signal are shown in Figure 7.108. Note how
the frequency content is changed for each IMF component. Figure 7.109 shows the time-frequency
spectrum of the HHT of El Centro. For comparison with WT (using Morlet wavelets), Figure 7.110
shows the WT of the same time signal. The HHT of Figure 7.109 shows too many details as com-
pared with the WT spectrum. When the HHT is smoothed as in Figure 7.111, the similarity of energy
patterns of both WT and HHT becomes evident. However, there are pronounced differences between
the two spectra that require additional study.
It is of interest to compare the FT behavior of El Centro with the marginal HHT along the fre-
quency scale. This is important since the FT of El Centro have been studied extensively by numer-
ous authors. Figure 7.112 shows the FT spectra (Figures 7.112a and 7.112b) and the HHT spectra
(Figures 7.112c and 7.112d). The HHT shows higher frequency content at low frequencies that the
200
Acceleration (cm/s)
–200
0 10 20
Time (s)
FIGURE 7.107 El Centro signal. (From Hwang, N. E. et al., The empirical mode decomposition and the
Hilbert spectrum for nonlinear and non-stationary time series analysis, Proceedings of the Royal Society,
London, 454, 903–995, 1998. Courtesy of Royal Society, London.)
100
20
C1 0 C6 0
–20
–100 –40
200 20
C2 0 C7 0
–200 –20
200 5
C3 0 C8 0
–200 –5
100 2
C4 0 C9 0
–100 –2
–4
2
50 0
C5 0 C10 –2
–50 –4
–6
0 10 20 0 10 20
Time (s) Time (s)
FIGURE 7.108 IMF Components of El Centro Signal. (From Hwang, N. E. et al., The empirical mode
decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis, Proceedings of
the Royal Society, London, 454, 903–995, 1998. Courtesy of Royal Society, London.)
20
Frequency (Hz)
10
0
0 10 20
Time (s)
FIGURE 7.109 Hilbert spectrum of El Centro time signal. (From Hwang, N. E. et al., The empirical mode
FT does not produce. This low frequency behavior can have a major effect on the seismic behavior on
many important structures such as suspension and cable stayed bridges, as well as tall buildings.
7.7.4.6 WT versus HHT in Infrastructure Applications

Both Wavelet and EMD methods are capable of detecting time of changes in the properties of a
given time signal. Vincent et al. (2000) investigated this capability in both techniques in a purely
20
Frequency (Hz)
10
0
0 10 20
Time (s)
FIGURE 7.110 Wavelet transform (Morlet) of El Centro time signal. (From Hwang, N. E. et al., The empir-
ical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis,
Proceedings of the Royal Society, London, 454, 903–995, 1998. Courtesy of Royal Society, London.)
20
Frequency (Hz)
10
0
0 10 20
Time (s)
FIGURE 7.111 Smoother Hilbert transform of El Centro. (From Hwang, N. E. et al., The empirical mode
numerical approach. They modeled a 3-DOF close coupled spring-mass system that is subjected to
a sinusoidal base motion that lasts 20 seconds. After 10 seconds, they suddenly reduced the stiff-
ness of one of the three springs in the system. The computed response of the third mass was then
analyzed by both EMD and wavelet approach. The writers found that the sudden reduction in the
spring (resulting in a softening of the whole system) was detected accurately by both methods, at 10
seconds. In addition, both methods gave some indication as to the severity of the damage, albeit in
a different manner. The changes in the frequencies that correspond to vibration modes in the EMD
approach would enable the estimation of damage level (loss of spring stiffness). The continuous
wavelet approach might also give an indication of damage severity through its fringe patterns. In all,
both methods proved to be worthy in detecting occurrence and level of the damage.
(a) (b)
0.15
0.1
0.05
(c) (d)
0.8
0.6
0.4
0.2
0
0 10 20 30 40 0 1 2 3 4
Frequency (Hz) Frequency (Hz)
FIGURE 7.112 Fourier and marginal Hilbert spectra of El Centro time signal (a) Fourier spectrum in the
range of 0-40 Hz; (b) detailed Fourier spectrum in the range of 0-5 Hz; (c) marginal HHT spectrum in the range
of 0-20 Hz; and (d) detailed marginal HHT spectrum in the range of 0-5 Hz. (From Hwang, N. E. et al., The
empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis,
Proceedings of the Royal Society, London, 454, 903–995, 1998. Courtesy of Royal Society, London.)
In a similar study that addressed only the usability of wavelet method for DMID, Hou and Noori
(2000) looked at an analytically simulated 3-DOF closed coupled system. Their investigations
addressed the effects of level of damage and the noise contamination on the efficiency of the wave-
let method in identifying the time and location of damage. They observed that the method is effec-
tive at higher damage levels (around 20% for their experiment) and/or low noise contamination.
Conversely, they observed that, at low damage levels (around 5% for their experiment), the noise
contamination might reduce the accuracy of prediction.
7.7.5 Signal Processing as an Integrator of DMID Methods

7.7.5.1 Use of Correlation and Operational Modal Analysis
As an example of an integral use of SHM different techniques in increasing safety, consider the
SHM project by Lin et al. (2005). The principals of that project wanted to (a) investigate numerical
modeling assumptions of elastomeric bearings in bridges, and (b) the temperature effects on such a
behavior. It is noted that during modeling a bridge that is supported by elastomeric bearing, those
bearings are assumed to be either hinges, or fixed boundary conditions. Such assumptions were
shown to be inaccurate (see Sanayei et al. (1999). The sensitivity of analysis and design of bridges
to such assumptions made the SHM project important. The principals of the project decided to use
dynamic modes of the bridge as the metrics in their project. However, since it is expected that tem-
perature can affect modal behavior of bridges, it was important to include sensitivity to temperature
variations in the project activities.
A simple span, skew composite steel bridge in Amherest, NY was used for the project
(Figure 7.113). There were ten steel girders that were about 10 ft apart. The girders were connected
by a set of cross bracings. The skew angle was about 40º–45º. Each end of the girder is supported
1 2 3 4
Sensor location
FIGURE 7.113 Plan of the bridge, and location of sensors. (Courtesy of Jerome O’Connor.)
by an elastomeric bearing that is 2.5-in thick, and in-plane dimensions of 30 in × 12 in. Four accel-
erometers were used. The accelerometers have a sampling range of 0–300 Hz; fairly adequate to
process the project frequency range of 0–20 Hz. The plan of the bridge and the locations of the
accelerometers are shown in Figure 7.113. As much as 168 hours of measurements were recorded.
The recordings were made at 2 hours intervals. The measurements were made at different ambient
temperatures, ranging from 80º to 32ºF.
One of the challenges of that project was the method to excite some of the bridge dynamic
modes. The bridge under consideration is a heavy bridge, and to use forced vibrations to excite the
needed modes would be costly. Because of this it was decided to use the ambient vibration mea-
surements to evaluate the modal behavior. The measured vibrations would then be analyzed using
signal processing techniques to estimate the needed information. Measured ambient vibrations
contain both the bridge natural modal behavior (forced and free vibrations) as well as unavoidable
noise. Using signal processing techniques if it was possible to eliminate, or reduce the noise, then
it would be possible to estimate accurately the bridge behavior. Two signal processing techniques
were used.
Cross-Correlation Analysis: As was mentioned earlier, cross-correlation and the cross power
density function of two time signals x1 and x2 can be described as
+∞
R12 = ∫ x1 ( t ) x2 ( t + t) dt
−∞ (7.49)
P12 = FT ( R12 )
(7.50)
Note that the effects of the noise will be reduced in P12, leaving the bridge modal effects in the
signal. To improve accuracy, the authors used an average of several cross power spectra for their
estimation. Figures 7.114 and 7.115 show the results of the cross-correlation analysis at 40º and 80ºF,
respectively. The first two natural frequencies of the bridge were insensitive to temperature at about
2 Hz and 4 Hz. The third natural frequency of the bridge was estimated as 6.3 Hz at 40º and 5.7 Hz
at 80º.
Random Decrement (Operational Modal Analysis): Random decrement y(t) of a function x(t)
of length n is defined as
1 N
y ( t) = ∑ x ( t + ti )
N i =1 (7.51)
The Fourier spectra of y(t) will also reduce noise effects and enhance modal behavior of the bridge.
It was found that the first two natural frequencies of the bridge were similar to the correlation analy-
sis at 2 Hz and 4 Hz, respectively. The third frequency was estimated at 5.8 Hz, fairly similar to the
correlation results.
Channel 1 cross power density spectrum Channel 2 cross power density spectrum
(a) 10 (b) 10
9 9
8 8
7 7
6 6
Amplitude
Amplitude
5 5
4 4
3 3
2 2
1 1
0 0
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Frequency(Hz) Frequency(Hz)
FIGURE 7.114 Cross power spectral accelerations (40º) (a) channel 1, and (b) channel 2. (Courtesy of Jerome
O’Connor.)
Channel 1–cross power density spectrum Channel 2–cross power density spectrum
(a) 10 (b) 10
9 9
8 8
7 7
Amplitude
Amplitude
6 6
5 5
4 4
3 3
2 2
1 1
0 0
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Frequency (Hz) Frequency (Hz)
FIGURE 7.115 Cross power spectral accelerations (80º) (a) channel 1, and (b) channel 2. (Courtesy of Jerome
O’Connor.)
The above estimations showed some sensitivity of the bridge behavior (third mode) to tempera-
ture effect. To further investigate the modeling effects of the elastomeric bearings, the authors used
a finite element modeling technique. They showed that by accurately modeling the bearings, it was
possible to produce natural frequencies of the bridge similar to the measurements. They found that
only vertical stiffness of the bearings affect bridge natural frequencies. They concluded that in
numerical bridge modeling, the vertical stiffness of elastomeric bearings must be assigned with care
to ensure accurate modeling results.
7.7.5.2 Filtering Multihazards Effects

Connor et al. (2002) used a specifically designed signal processing techniques that meets the goals
of the SHM experiment. They utilized high and low-pass filters to isolate signals from different
sources. The bridge was a four span continuous haunched steel deck truss. The deck was designed
to be fully composite with the main upper chord of the steel truss as well as the longitudinal string-
ers and transverse floor beams. Thus, due to this unique structural characteristic, it was decided to
instrument and monitor the structure to better understand its short- and long-term behavior. Items
of interest were: live load response, quantifying the effects of long-term creep and shrinkage of the
deck, and degree of local and global composite action.
Due to the interest in long-term response from construction to in service, vibrating wire strain
gages wee used for static type of measurements and regular strain gages for short duration dynamic
measurements (load tests, etc.). Some gages were also installed inside the deck on the rebars of inter-
est. Special attention was paid to temperature compensation effects. A wireless modem installed at
the site was used to communicate the data from the field to a remote office location. Nine 12-volt
batteries were used to power the system with a solar panel. A methodology was adopted to conserve
power based on observed activity.
Strain data was processed in three steps. First, a digital low-pass filter was used to remove the
noise spikes from the data stream caused by phenomena such as quickly changing magnetic fields
and radio transmissions.
A copy of filtered data (low pass) was passed through a second high-pass filter to remove the
strain effects from thermal effects. Thus, one file contains mechanical and thermal and the other
thermal alone. Subtracting the later from the former, gives mechanical signals alone.
The response was as expected and agreed with designers’ estimates. Truss acted compositely
with the entire deck system. Hence, live load stresses decreased and a more redundant structure
system is attained overall.
7.7.6 Rules for Using Signal Processing Tools

The properties of the four popular signal processing methods are now compared in Table 7.29.
Clearly, the methods can be helpful in SHM applications depending on their properties. Table 7.30
shows some SHM situations where these four methods can be used. We conclude this section with
some rules for applying those different signal processing methods in general SHM situations.
TABLE 7.29
Comparison of Signal Processing Methods
Issue FT TFT WT HHT
Scales Frequency Frequency and time Frequency and time Frequency and time
Discrete/Continuous Yes/yes Yes/yes Yes/yes NA
Stationary No No Yes Yes
Nonlinear systems No No Yes Yes
Adaptive No No No Yes
TABLE 7.30
Potential SHM Uses of Different Signal Processing Methods
Method Possible use in SHM Comment
FT Earthquake engineering FT has been used extensively in earthquake engineering, even though earthquake
signals are not stationary. The potential nonlinear response of the structures to
earthquakes might limit the accuracy of FT in studying earthquake responses
and damages
TFT Strain history due to Sine response of bridges to truck motions can be mostly linear, TFT might
truck motions provide accurate tool for processing the strain history. The time-varying
frequency response can be captured by TFT (WT and HHT can also be used for
this task)
WT Flood and scour Potential nonlinearities and nonstationary response of bridges to floods and
scour are reasons for considering WT (and HHT) to analysis in this situation
HHT Flood and scour Same as WT, above
• Plan Ahead: The frequency and time steps, as well as the frequency and time ranges
should be well know in advance of any SHM experiment.
• Structural Identification: Ensure that the above steps and ranges are consistent with the
structural properties of interest. For example, the frequency range should cover all struc-
tural modes of interest in a STRID situation.
• Damage Scales: Justify in advance how signal processing can aid in identifying the dam-
age of interest. Ensure that the frequency range and steps are consistent with the expected
damage type, location, and size.
• Uncertainties: Make allowance for any uncertainty or variability in the experiment. This
includes, but not limited to, setting up potential probabilistic methods that are consistent
with the signal processing method.
7.8 DAMAGE IDENTIFICATION IN SHM

7.8.1 General
The previous sections of this chapter explored several methods of DMID. It should be remembered that
the choice of a particular method for DMID should be consistent with the overall goals and objectives
of the project. They should also be consistent with the damage parameters as explained earlier in this
chapter. This section will present some important considerations that might be of help to the profes-
sional while choosing a DMID method. The presence of DMID as only one component within the
overall SHM project means that the interactions of DMID with other components must be considered.
7.8.2 Important Considerations

Quality: Quality of results of DMID method needs to be considered carefully. How capable the
method is in identifying the damage of interest? Some methods can produce qualitative damage
description, such as “corrosion is extensive in near the support” or “the connection has an unac-
ceptable deformation.” Other methods would produce a more quantitative damage estimate such
as “a 0.25 inch crack is detected at a depth of 0.12 inch below the surface.” Both qualitative and
quantitative damage estimates are useful. The professional will need to ensure that the quality of the
damage estimate is consistent with the objectives of the SHM project on hand.
Resolution: DMID methods must be capable of addressing time (frequency) and spatial resolution
of the damage of interest.
Reliability: Uncertainties need to be considered as a part of the DMID experiment. This can be
accomplished by taking sufficient sample points, repeating the experiment, or any other activity that
would accommodate uncertainties or unexpected breakdowns in parts of the experiment.
Fiscal: The cost of the DMID method must be reasonable and consistent with the potential value
of information that is expected from the activity. There are numerous cost–benefit examples in this
volume that can be used for such an evaluation.
Others: There are many other considerations for the use of a DMID method in an SHM project.
Some of those issues are environmental considerations, traffic conditions, experience of the profes-
sionals, and prevailing rules and regulations.
7.8.3 Interactions with Other SHCE Components

Damage Identification processes can only be one part of a whole SHCE process. It needs to be pre-
ceded by, or performed in conjunction with, sensing or measurement phase. DMID process is also
usually followed by a decision making process. DMID can be executed alone, or in conjunction with
STRID activity. Table 7.31 explores some of these interrelationships between DMID for different
type of damages/hazards.
7.8.4 Common Bridge Damages and SHM Applications

Damages affect bridges in many forms. They depend on the type of bridge, the bridge component
as well as the cause of the damage. Table 7.32 shows some types of damages that can occur in rein-
forced concrete bridges (Figures 7.116 and 7.117). Similarly, Tables 7.33 and 7.34 show types of dam-
ages that can occur in steel truss and steel girder bridges, respectively (Figures 7.118–7.123). Finally,
Table 7.35 shows some types of damages that might occur in bridge pedestal seats. Other damage
situations will be explored in depth by Ettouney and Alampalli (2012) (Figures 7.124–7.126).
TABLE 7.31
Interactions of DMID with Other SHCE Phases
Hazard
SHCE Phase Fatigue Corrosion Earthquakes Scour
Sensing Real time or intermittent sensing is suitable Pre-event sensing would Real time or
be different from intermittent sensing is
Post-event sensing suitable
STRID Effects of monitored Accommodating Nonlinear analysis is Soil-foundation-water
fatigue damage on corrosion effects in needed for this type interactions need to be
structural STRID efforts might of STRID. For both simulated. Direct
performance needs be difficult. Loss of pre- and afterevent modeling might be
to be investigated. section can be analysis difficult. Informal
Special modeled, however, analysis techniques
considerations accurate information such as neural
need to be taken in is needed networks might prove
the analysis phase to be convenient
due to the localized
nature of fatigue
damage
TABLE 7.32
Types of Damages and Reinforced Concrete Bridge Types
Bridge Type Damages
Cast-In-Place slabs Cracking, scaling, delamination, spalling, efflorescence, honeycombs,
Tee beams pop-outs, wear, collision damage, abrasion, overload damage,
Concrete girders reinforced steel corrosion (Figure 7.116)
Concrete channel beams
Concrete rigid frames
Concrete arches In addition to the above, stress corrosion
Precast and prestressed slabs
Prestressed double tees (Figure 7.117) Cracking, scaling, delamination, spalling, efflorescence, honeycombs,
pop-outs, wear, collision damage, abrasion, overload damage,
reinforcing steel corrosion, stress corrosion of prestressing strands
Prestressed I-beams Cracking, Delamination, Spalling, Collision damage, Overload damage,
Concrete box girders including segmental Reinforcing/prestressing steel corrosion, stress corrosion,
Efflorescence, Pop-outs
Concrete box culverts Cracking, scaling, delamination, spalling, efflorescence, honeycombs,
pop-outs, wear, abrasion, overload damage, Section loss of exposed
reinforcing bars, embankment scour at culvert inlet and outlet, roadway
settlement
FIGURE 7.116 Delamination of reinforced concrete beam. (Courtesy of New York State Department of
Transportation.)
FIGURE 7.117 Spalling of concrete and corroded rebars. (Courtesy of New York State Department of
Transportation.)
7.9 APPENDIX: LAMB WAVES

Lamb waves (also known as guided waves, or plate waves) are linear elastic waves that propagate in
solids at UT frequencies. They can travel for relatively long distances. They are guided waves that
travel in plates or shell systems, that is, systems with small thicknesses. The Lamb wave velocity is
proportional to both the plate thickness and the frequency of vibration. It has two main modes of
propagation: symmetric and asymmetric. In civil infrastructure, they have been used successfully in
detecting damages in several types of system components such as pavements and truss gusset plates.
One major limitation of utilization of Lamb waves in civil infrastructure damage detection is that
they are fairly limited to localized damage detection, as opposed to global damage detection. For
detailed discussions on Lamb waves, see Rose (1999).
7.10 APPENDIX: DISPERSION CURVES

The 3D elasticity equations in Cartesian coordinates can be expressed as
s ij , j + r fi = rui (7.52)

(
eij = 0.5 ui , j + u j ,i ) (7.53)
TABLE 7.33
Type of Damages in a Steel Truss Bridge
Truss Components Damages
Truss members Section losses due to corrosion (Figure 7.118). Corrosion between plates of built-up
members. Effects of de-icing salts
Damage due to direct collision impact resulting in cracks or large deformations.
Imperfections in welds, especially plug-welds. Damages in bolted or welded retrofits due to
drilling, flame cutting, or punching
Looseness in truss rods should be considered a type of damage, since they are not carrying
the intended tensile force
Alignment of individual Truss member misalignment should be considered as a type of damage. Such misalignment
truss members can produce force redistribution, which is inconsistent with design condition that might lead
to undesirable overstressing. DMID of alignment can be achieved by tilt meters, laser
sensing, etc.
Member overstressing Lateral or local buckling are signs of overstressing in compression members (Figure 7.119).
Necking down of cross-sections is a sign of overstressing in tension members. Special
attention should be given to less ductile steels such as high strength steel, or older cast iron
construction.
Connections Lost rivets or welds can result in a higher undesired stresses. Acoustic monitoring can help in
detecting noises from loose connections (the limit is an AE monitoring). Visual inspections
of corroded connections or supports can be of help (Figure 7.120)
Source: NYSDOT, Bridge Inspection Manual, New York State Department of Transportation, Albany, NY, 1997.
TABLE 7.34
Type of Damages in a Steel Girder Bridge
Steel Girder Bridge Components Damages
Alignment of individual girders Misalignment should be considered as a type of damage. Such misalignment can
and/or the whole girder assembly produce force redistribution, which is inconsistent with design condition that
might lead to undesirable overstressing.
Web crippling/section losses This can happen in sections with large shearing forces, such as those near supports.
Crippling indicates excessive stresses. Section losses occur due to corrosion.
DMID methods for detecting these damages are not the same and should be
considered separately (Figure 7.121)
Corrosion Corrosion would cause section losses in some hidden areas such as under the deck,
behind sidewalks, and above supports. Careful choice of DMID methods is
essential so as to ensure accuracy in detection (Figure 7.122)
Riveted or built-up sections These components are particularly susceptible to crevice corrosion (corrosion in
hidden surfaces). Careful choice of DMID methods is essential so as to ensure
accuracy in detection
s ij = leij dij + 2meij (7.54)

Eliminating stresses and strains from Equations 7.52, 7.53, and 7.54 we get
m ui , jj + ( l + m ) u j , ji + r fi = rui (7.55)

FIGURE 7.118 Corrosion loss in a steel member. (Courtesy of New York State Department of Transportation.)
FIGURE 7.119 Local buckling in a steel member. (Courtesy of New York State Department of Transportation.)
FIGURE 7.120 Heavily corroded steel bearing. (Courtesy of New York State Department of Transportation.)
FIGURE 7.121 Section loss in steel girders. (Courtesy of New York State Department of Transportation.)
FIGURE 7.122 Corrosion of flange near sidewalk due to de-icing. (Courtesy of New York State Department
of Transportation.)
FIGURE 7.123 Steel girder damage near supports. (Courtesy of New York State Department of Transportation.)
TABLE 7.35
Type of Damages in a Bridge Seats and Pedestals
Type of Bridge Pedestal Damages
Concrete Damages include cracking, scaling, spalling, delaminating, or leaching. See Figures 7.124
and 7.125. Figure 7.126 shows a heavily damaged concrete bearing seat. A temporary
bridge seat is used to support the bridge girder (shown in the foreground of Figure 7.126)
Masonry Damages include cracks and deterioration of mortar joints or stone bricks. Loss of
masonry might occur as the pedestal deteriorates
Steel Damages include loss of section, bowing, or buckling
FIGURE 7.124 Bridge pedestal with minor damages. (Courtesy of New York State Department of Transportation.)
FIGURE 7.125 Bridge pedestal with moderate damages. (Courtesy of New York State Department of
Transportation.)
FIGURE 7.126 The inspector shows a heavily damaged bridge bearing concrete seat, where the reinforce-
ment rebars are exposed.
For plate with thickness h in the x3 direction that is free from tractions at top and bottom surfaces
(x3 = −0.5h and x3 = 0.5h), Equation 7.55 can be decomposed to two equations as follows
∂2f ∂2f 1 ∂2f (7.56)

+ =
∂x12 ∂x32 cL2 ∂t 2
∂2 y ∂2 y 1 ∂2 y (7.57)
+ =
∂x12 ∂x32 cT2 ∂t 2
Equation 7.56 governs longitudinal wave motion while Equation 7.57 governs the shear wave motion.
Assuming plane strain motions in the plate, the displacements are related to ϕ and ψ by
∂f ∂y (7.58)
u1 = +
∂x1 ∂x3
u2 = 0 (7.59)

∂f ∂y (7.60)
u3 = −
∂x3 ∂x1

By separating the time and space variables, we assume the wave form solutions for Equations 7.58
and 7.60 as
f = Φ ( x3 ) ei(kx1 −w t ) (7.61)

y = Ψ ( x3 ) ei(kx1 −w t ) (7.62)

Defining the wave number as k, the driving frequency as ω, the wave phase velocity as cp and the
wave length as λ we have the following relationships:
w
k= (7.63)
cp

and
lw (7.64)
cp =
2p
We can further introduce the following variables:
w2
p2 = − k2 (7.65)
cL2

and
w2
q2 = − k2 (7.66)
cT2

Substituting Equations 7.63 through 7.66 into Equations 7.61, and 7.62, we obtain
tan ( qh ) 4 k 2 p tan ( ph )
+ 2
=0 (7.67)

q (
q2 − k 2 )
2
q tan ( qh ) +
(q 2
)
− k 2 tan ( ph )
=0 (7.68)
4k 2 p
Equation 7.67 governs the symmetric wave motions while Equation 7.68 governs antisymmetric
wave motions along x1. For detailed discussions on dispersion curves, see Rose (1999).
7.11 APPENDIX: HELMHOLTZ EQUATION

Let us define, as usual, the wave number k, as
w (7.69)
k=
c
Then Helmholtz equation can be written as

(∇ 2
)
+ k2 A = 0

(7.70)
A = A ( x1 , x2 , x3 ) (7.71)

For more details, see Rose (1999).

7.12 APPENDIX: ANGULAR SPECTRUM METHOD

By inspecting Equation 7.71, for a constant x3, we can perform the FT in the (x2, x3) plane as
+∞
Aˆ e ( 2 1 2 2 ) dxˆ1 dxˆ 2
i x xˆ + x xˆ
A = ∫∫ (7.72)
−∞
The inverse transform of this equation is
1 +∞
Aˆ = A e ( 2 1 2 2 ) dx1 dx2
− i x xˆ + x xˆ
4 p2 ∫∫ −∞
(7.73)

Note that
Aˆ = Aˆ ( xˆ1 , xˆ 2 , x3 ) (7.74)

Substituting Equation 7.74 in Equation 7.70 we get the transformed Helmholtz equation as
∂ 2 Aˆ
+ w 2 Aˆ = 0 (7.75)
∂ x32

With

w 2 = k 2 − x12 − x22 (7.76)
The solution to Equation 7.75 is
Aˆ = Aˆ1 eiw x3 + Aˆ 2 e − iw x3 (7.77)
With the separation of variables we have Aˆ1 = Aˆ1 ( xˆ1 , xˆ 2 ) and Aˆ 2 = Aˆ 2 ( xˆ1 , xˆ 2 ). We can now express
the general solution of Helmholtz equation as
+∞ +∞
Aˆ1 e ( 2 1 2 2 3 ) dxˆ1 dxˆ 2 + ∫ ∫ Aˆ 2 e ( 2 1 2 2 3 ) dxˆ1 dxˆ 2
i x xˆ + x xˆ + w x i x xˆ + x xˆ − w x
A = ∫∫ (7.78)
−∞ −∞
− iw x
The two parts of Equation 7.78 represent incoming waves and outgoing waves for e + iw x3 and e 3 ,
respectively. Each incoming or outgoing wave can either be propagating wave, or exponentially
decaying wave in the x3 direction when ω 2 ≥ 0 or ω 2 < 0, respectively. The four types of waves
can represent any wave field in a plate. Equation 7.78 have thus reduced the 3D representation of
complex wave field into a set of planar waves in the (xˆ 2 , xˆ 3 ) plane. The spectrum represented by
Aˆ1 = Aˆ1 ( xˆ1 , xˆ 2 ) and Aˆ 2 = Aˆ 2 ( xˆ1 , xˆ 2 ) is called the angular spectrum since the coordinates x̂2 and x̂3
can be related to the angle between the original (x2, x3) plane and the transformed (xˆ 2 , xˆ 3 ) plane.
REFERENCES
Alkhalifah, T. (2005). “Nondestructive Testing Resolution: Electromagnetic Waves versus Seismic Ones,” 3rd
Middle East Nondestructive Testing Conference & Exhibit, Manama, Bahrain.
ASNT. (1995). “Laser Based Nondestructive Testing Methods,” in Nondestructive Testing Handbook, 2nd
Edition, Vol. 9, Special Nondestructive Testing Methods. ASNT, Columbus, OH.
ASNT. (1999). Choosing NDT: Applications, Costs, and Benefits of Nondestructive Testing in Your Quality
Assurance Program, American Society for Nondestructive Testing, Inc., Columbus, OH.
Baron, J. and Ying, S. (1987). “Acoustic Emission Source Location,” in Nondestructive Testing Handbook: Vol.
5, Acoustic Emission Testing, ASNT, Columbus, OH.
Beissner, R. (2005). “Magnetic Field Testing,” in ASM Handbook: Non Destructive Evaluation and Quality
Control, Vol. 17, ASM, Materials Park, OH
Beissner, R. and McGinnis, C. (1984). “Laboratory Test of Magnetic Field Disturbance (MFD) System for
Detection of Flaws in Reinforcing Steel,” Report No. DTFH 61-80-C-00002, FHWA, Washington, DC.
Bendat, J. and Piersol, A. (1971). Random Data: Analysis and Measurement Procedures, Wiley-Interscience,
New York.
Birks, A., Golis, M., and Green, R. (2007). “Introduction to Ultrasonic Testing,” in Nondestructive Testing
Handbook: Vol. 10, ANST, Columbus, OH.
Boashash, B. (1992). Time-Frequency Signal Analysis, Wiley, New York.
Bossi, R., Mengers, P., and Qien C. (2007). “Radioscopy and Tomography,” in Nondestructive Testing
Handbook: Vol. 10, ASNT, Columbus, OH.
Botsco, R. and Jones, T. (2007). “Thermography and Other Special Methods,” in Nondestructive Testing
Botsko, R. (2007). “Introduction to Ultrasonic Testing,” Section 13, Part 2 in Nondestructive Testing Handbook:
Vol. 10, ASNT, Columbus, OH.
Butt, S. and Limaye, V. (2002). “Damage Detection in Concrete Bridge Deck Slabs Using Acoustic
Attenuation,” Proceedings of 1st International Workshop on Structural Health Monitoring of Innovative
Civil Engineering Structures, ISIS Canada Corporation, Manitoba, Canada.
Carlos, M. F., Miller, R. K., and Tamutus, T. A. (2000). “Acoustic Emission Local Area Monitoring System,”
Structural Materials Technology: an NDT Conference, ASNT, Atlantic City, NJ.
Carlos, M. F., Wang, D., Vahaviolos, S., Anastasopoulos, A. (2006). “Advanced Acoustic Emission for
On-stream Inspection of Petrochemical Vessels,” Proceedings of the 3rd International Conference on
Emerging Technologies in NDT, A. A. Balkema, Netherlands 2004, ISBN 90 5809 645 9 (Volume)-. 90
5809 645 7(CD), pp. 167–172.
Carlyle, J. H. (1993). “Acoustic Emission Monitoring of the I-10 Mississippi River Bridge,” Phase Report No.
R90-259, Physical Acoustics Corporation, Lawrenceville, NJ.
Carlyle, J. H. and Ely, T. M. (1992). “Acoustic Emission Monitoring of the I-95 Woodrow Wilson Bridge,”
Phase Report No. R90-259, Physical Acoustics Corporation, Lawrenceville, NJ.
Carlyle, J. H. and Leaird, J. D. (1992). “Acoustic Emission Monitoring of the I-80 Bryte Bend Bridge,” Phase
Report No. R90-259, Physical Acoustics Corporation, Lawrenceville, NJ.
Colla, C. and Wiggenhauser, H. (2000). “Developments in Impact-Echo for the investigation of Concrete
Structures,” Structural Materials Technology: an NDT Conference, ASNT, Atlantic City, NJ.
Connor, J. (1976). Analysis of Structural Member Systems The Ronald Press Company, New York.
Connor, R. J., Santousuosso, B., and Pessiki, S. P. (2002). “Long-Term Wireless Remote Monitoring of the
Lehigh River Bridge,” Proceedings, NDE Conference on Civil Engineering, ASNT, Cincinnati, OH.
Cullington, D. W., MacNeil, D., Paulson, P., and Elliot, J. (2001). “Continuous acoustic monitoring of grouted
post-tensioned concrete bridges,” NDT & E International, Vol. 34, No 2.
De Witt, D. (1988). Theory and Practice of Radiation Thermometry, John Wiley, New York.
Demma, A., Alleyne, B., and Pavlakovie, B. (2005). “Testing of Buried Pipelines using Guided Waves,”
Proceedings of the 2005 ASNT Fall Conference, Columbus, OH.
Dennis, M. (2005). “Industrial Computed Tomography,” in ASM Handbook: Non Destructive Evaluation and
Quality Control, Vol. 17, ASM, Materials Park, OH
Doebling, S.W., Farrar, C.R., Prime, M.B., and Shevitz, D. W. (1996). “Damage Identification and Health
of Structural and Mechanical Systems from Changes in their Vibrational Characteristics: A Literature
Review,” Los Alamos National Laboratory Report, LA-13070-MS.
Duan, Z., Yan, G., and Ou, J. (2004). “Structural Damage Detection in Ambient Vibration Using Wavelet Packet
Transform and Probabilistic Neural Network,” Proceedings of 2nd International Workshop on Structural
Canada.
Duke, J. C. and Horne, M. R. (2006). “Viability of Monitoring Acoustic Emission from Crack Growth in
Bridges,” NDE Conference on Civil Engineering, ASNT, St. Louis, MO.
Dunegan, H., Harris, D., and Tetelman, A. (1970). “Detection of Fatigue Crack Growth by Acoustic Emission
Techniques,” Materials Evaluation, Vol. 28, No. 10, Columbus, OH.
Emigh, C., Iddings, F., and Willenberg, J. (2007). “Radiation Principles and Sources,” in Nondestructive Testing
Endo, K. (2000). “A study on the Application of the Acoustic Emission Method for Steel Bridges,” Masters of
Science in Engineering Thesis, The University of Texas at Austin, Austin, TX.
Ettouney, M. and Alampalli, S. (2000). “Engineering Structural Health.” ASCE Structures Congress 2000,
Philadelphia, PA.
Ettouney, M. and Alampalli, S. (2012). Infrastructure Health in Civil Engineering, Applications and
Ettouney, M., Daddazio, R., Hapij, A., and Aly, A., (1998), “Health Monitoring of Complex Structures,”
Proceedings, Smart Structures and Materials 1998: Industrial and Commercial Applications of Smart
Structures Technologies, SPIE, Vol. 3326.
Farrar, C. and Doebling, S. (1999). “Damage Detection and Evaluation II,” in Modal Analysis and Testing,
Editors: Silva, J. and Maia, N. NATO Science Series, Kluwer Academic Publishers, London, UK.
Farrar, C. and Jauregui, D. (1996). “Damage Detection Algorithms Applied to Experimental and Numerical
Modal Data from the I-40 Bridge,” Los Alamos National Laboratory Report No. LA-13074-MS, Los
Alamos, NM.
Farrar, C., Baker, W., Bell, T., Cone, K., Darling, T., Duffey, T., Eklund, A., and Migliori, A. (1994). Dynamic
Characterization and Damage Detection in the I-40 Bridge over the Rio Grande, Los Alamos National
Laboratory Report LA-12767-MS.
Fisk, P. (2008). “Sonic / Ultrasonic Testing for Bridge Inspections,” NDE/NDT for Highway and Bridges:
Structural Materials Technology (SMT), ASNT, Oakland, CA.
Fu, Y. and Dewolf, J. (2001). “Monitoring and Analysis of a Bridge with Partially Restraind Bearings,” ASCE
Journal of Bridge Engineering, 6(1).
Garcia, G. and Stubbs, N. (1995). “Effect of damage size and location on the stiffness of a rectangular beam,”
Smart Structures and Materials, San Diego, CA.
Graps, A. (1995). Introduction to Wavelets, IEEE Computational Science and Engineering, IEEE, Los Alamitos, CA.
Green, A., Blackburn, P., Craig, B., Cross, N., Ferdinand, M., Fowler, T., and Robinson, D. (1987). “Acoustic
Emission Applications in the Petroleum and Chemical Industries,” in Nondestructive Testing Handbook:
Volume Five, Acoustic Emission Testing, ASNT, Columbus, OH.
Greene, A. (2005). “Radiographic Inspection,” in ASM Handbook: Non Destructive Evaluation and Quality
Control, Vol. 17, ASM, Materials Park, OH
Gucunski, N., Rascoe, C., Wang, Z., Chen, Y., and Fang, T. (2008). “Automated Interpretation of Bridge Deck
Impact Echo Data Using Three - Dimensional Visualization,” NDE/NDT for Highways and Bridges:
Structural Materials Technology (SMT), ASNT, Oakland, CA.
Hamstad, M. A., O’Gallagher, A., and Gary, J. (2002). “A Wavelet Transform Applied To Acoustic Emission
Signals: Part 1: Source Identification,” Journal of Acoustic Emission, 20, 39–61.
Hardy, G. and Bolen, J. (2005). “Thermal Inspection,” in ASM Handbook, ASM International, Materials Park, OH.
He, J. and Fu, Z. (2001). Modal Analysis, Butterworth Heinemann, Boston.
Hou, Z. and Noori, M. (2000). “Application of Wavelet Analysis for Structural Health Monitoring,” Proceedings
of 2nd International Workshop on Structural Health Monitoring, Stanford University, Stanford, CA.
Housner, G., and Masri, S. (1996). “Structural Control Issues Arising from the Northridge and Kobe
Earthquakes,” Proceedings of 11th World Conference on Earthquake Engineering, Mexico, Paper
No. 2009.
Hwang, N. E., Shen, Z., Long, S. R., Wu, M. C., Shih, H., Zheng, Q., Yen, N. C., Tung, C. C., and Liu,
H. H. (1998). “The Empirical Mode Decomposition and the Hilbert Spectrum for Nonlinear and Non-
Stationary Time Series Analysis,” Proceedings of the Royal Society, London, 454, 903–995.
Hubbard, B. (1998). The World According to WAVELETS, A. K. Peters, Natick, MA.
Jiles, D. C. (1986). “Evaluation of the Properties and Treatment of Ferromagnetic Steels Using Magnetic
Measurements,” Proceedings of the Nondestructive Evaluation of Ferromagnetic Materials, Houston, TX.
Karbhari, V., Kaiser, H., Navada, R., Ghosh, K., and Lee, L. (2005). “Methods for Detecting Defects in
Composite Rehabilitated Concrete Structures,” Submitted to Oregon Department of Transportation, and
Federal Highway Administration, Report No. SPR 336.
Kim, J. and Stubbs, N. (1995a). “Damage Detection in Offshore Jacket Structures from Limited Modal
Information,” International Journal of Offshore and Polar Engineering, 5(1), 58–66.
Kim, J. and Stubbs, N. (1995b). “Modal-Uncertainty Impact and Damage-Detection Accuracy in Plate Girders,”
Journal of Structural Engineering, ASCE Publication, 121(10).
Klausenberger, F. and Barton, J. (1977). “Detection of Flaws in Reinforcing Steel in Prestressed Concrete
Bridge Members,” Report on Contract No. DOTFH-11-8999, FHWA, Washington, DC.
Koerner, R. and Lord, A. (1982). Use of Acoustic Emissions to Predict Ground Stability Federal Highway
Administration Report No. FHWA / RD-82/052, Washington, DC.
Koerner, R., McCabe, W., and Lord, A. (1981). “Acoustic Emission Behavior and Monitoring of Soils,”
Acoustic Emissions in Geotechnical Engineering Practice. American Society for Testing and Material.
ASTM STP 750. Philadelphia, PA
Lin, L., Lee, G., Liang, Z., and O’Connor, J. (2005). “Health Monitoring of Bridge Bearings,” Proceedings of
the 2005 ASNT Fall Conference, Columbus, OH.
Maclachlan-Spicer, J., Kerns, W., Aamodt, L., and Murphy, J. (1991). “Time Resolved Infrared Radiometry of
Multilayer Organic Coatings Using Surface and Subsurface Heating,” Thermosense XIII, Editor: Baird,
G. Proceedings of Society of Photo-Opt. Instrumentation Engineering (SPIE), Vol. 1467.
Maldague, X. (2002). “Introduction to NDT by Active Infrared Thermography,” Materials Evaluation, ASNT,
Vol. 60, No. 9, Columbus, OH.
Miller, R. (2007). “Acoustic Emission Testing,” in Nondestructive Testing Handbook: Vol. 10, Nondestructive
Testing Overview, ANST, Columbus, OH.
Moore, P. (2001). NDT Handbook on Infrared Technology, Editor: Maldague, X. ASNT Handbook Series,
ASNT Press, Columbus, OH.
NYSDOT. (1997). Bridge Inspection Manual. New York State Department of Transportation, Albany, NY.
Owens, S., Rose, J., Mu, J., and Zhang, L. (2005). “A new High Frequency Long Range Ultrasonic Guided
Wave Inspection System for Pipes,” Proceedings of the 2005 ASNT Fall Conference, Columbus, OH.
Pandey, A. K. and Biswas, M. (1994). “Damage Detection in Structures using Changes in Flexibility,” Journal
of Sound and Vibration, 169(1), 3–17.
Pandey, A. K., Biswas, M., and Samman, M. (1991). “Damage Detection in from changes in curvature mode
shapes,” Journal of Sound and Vibration, 145(9), 312–332.
Pendat, J. and Piersol, A. (2010). Random Data: Analysis and Measurement Procedures, Wiley, New York.
Pollock, A. (2005). “Acoustic Emission Inspection,” in ASM Handbook: Non Destructive Evaluation and
Quality Control, Vol. 17, ASM, Materials Park, OH
Post, D., Han, B., and Ifju, P. (1994). High Sensitivity Moiré: Experimental Analysis for Mechanics and
Materials, Springer-Verlag, New York.
Reese, G. J. (1995) “The Effects of Finite Element Grid Density on Model Correlation and Damage Detection
of a Bridge,” Proceedings of the 36th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics
and Materials Conference.
Riahi, M. and Khosrowzadeh, B. (2005). “Development of Acoustic Emission AE) Method for the Diagnosis of
Corrosion in Steel Pipes,” Proceedings of the 2005 ASNT Fall Conference, Columbus, OH.
Rollwitz, W. (2005a). “Magabsorption NDE,” in ASM Handbook: Non Destructive Evaluation and Quality
Control, Vol. 17, ASM, Materials Park, OH.
Rollwitz, W. (2005b). “Microwave inspection,” in ASM Handbook: Non Destructive Evaluation and Quality
Control, Vol. 17, ASM, Materials Park, OH.
Rose, J. (2004). Ultrasonic Waves in Solid Media, Cambridge University Press, Cambridge, UK.
Sanayei, M. and Saletnik, M. J. (1996a). “Parameter Estimation of Structures From Static Strain Measurements,
Part I: Formulation,” Journal of Structural Engineering, ASCE, 122(5), 555–562.
Sanayei, M. and Saletnik, M. J. (1996b). “Parameter Estimation of Structures From Static Strain Measurements,
Part II: Error Sensitivity Analysis,” Journal of Structural Engineering, ASCE, 122(5), 563–572.
Sanayei, M., McClain, J., Wadia-Fascetti, S., and Santini, E. (1999). “Parameter Estimation Incorporating
Modal Data and Boundary Conditions,” Journal of Structural Engineering, ASCE, 125(9), 1048–1055.
Shepard, S., Lhota, J., Ahmed, T., Rubadeux, B., and Wang, D. (2001). “Qualification and Automation of Pulse
Thermographic NDE,” in Nondestructive Evaluation of Materials and Composites, Editor: Baaklini, et
al. Proceedings of Society of Photo-Opt. Instrumentation Engineering (SPIE), Vol. 4336.
Spanos, P. D. and Failla, G. (2004). “Evolutionary Spectra Estimation Using Wavelets,” Journal of Engineering
Mechanics, ASCE, Reston, VA, 130(8), 952–960.
Stubbs, N. and Garcia, G. (1996). “Application of Pattern Recognition to Damage Localization,” Microcomputers
in Engineering, Vol. 11. Computer-Aided Civil and Infrastructure Engineering, pp 395–409.
Stubbs, N. and Kim, J. T. (1994). “Field Verification of a Nondestructive Damage Localization and Severity
Estimation Algorithm,” Report prepared for New Mexico State University, Las Cruces, NM.
Stubbs, N., Kim, J. T., and Farrar, C. R. (1995). “Field Verification of a nondestructive damage localization and
severity estimation algorithm,” 13th International Model Analysis Conference, Nashville, TN.
Tracy, N. (2007). “Liquid Penetrant Testing,” in Nondestructive Testing Handbook: Vol. 10, ANST,
Columbus, OH.
Vincent, H., Hu, J. and Hou, Z. (2000). “Damage Detection Using Empirical Mode Decomposition Method and
a Comparison with Wavelet Analysis,” Proceedings of 2nd International Workshop on Structural Health
Yan, B., Goto, S., and Miyamoto, A. (2004). “Time-Frequency Analysis Based Methods for Modal Parameter
Identification of Bridge Structure Considering Uncertainty,” Proceedings of 2nd International Workshop
on Structural Health Monitoring of Innovative Civil Engineering Structures, ISIS Canada Corporation,
Manitoba, Canada.
Zhang, Z. and Aktan, E. (1995). The Damage Indices for the Constructed Facilities, Proceedings of the 13th
International Modal Analysis Conference, Vol. 2, Nashville, TN.
Zimmerman, D. C. and Kaouk, M. (1994). “Structural Damage Detection using a Minimum Rank Update
Theory,” Journal of Vibration Acoustics, (2), 222–231.
8 Decision Making in IHCE
8.1 INTRODUCTION
There should be a reason for any engineering activity. This applies well to any structural
health monitoring (SHM) project. The ultimate reasons for SHM activities should be either
reducing costs or increasing safety. The way to ascertain that the SHM activities for a particu-
lar project would ultimately achieve the cost reduction and/or increasing safety is through the
decision making tool chest. We define the decision making tool chest as a collection of quan-
titative or semiquantitative analytical methods and techniques that help the professional in
analyzing different aspects of the SHM project and making appropriate decisions that would
help to reduce costs or increase safety. This relationship is shown in Figure 8.1. Note that the
decision making tool chest can be utilized to design the SHM experiment. It can also be used
to assess the results of the SHM experiment. The utilization of the decision making tool chest
depends on the type of SHM project, as well as the ultimate goals of embarking on the SHM
project to start with.
Decision making processes involve all aspects of SHM. As defined earlier, Decision making
is the fourth step in any structural health engineering effort. We aim to generalize this concept.
Decision making is a part of every step in the first three steps, generally referred to as SHM, as
well as a separate part by itself. Figure 8.1 shows these integral interactions between decision
making (DM) and the other SHM steps. As Figure 8.1 shows, any SHM effort would have one (or
both) of the goals of (a) saving costs, and (b) improving safety. DM should always be the gateway
to any/all of the three components of SHM, it also should be present during the execution of any/
all of those components, and finally, DM should be the last step to be executed in the effort. The
final step in any effort should be an attempt to answer the question “how much additional safety,
or reduced cost, or both, have been achieved by this effort?” Lack of appropriate and timely execu-
tion of DM might result in less efficient project, which can result in costlier, and perhaps less-safe
results.
DM tools that can be utilized before, during, and after the execution of any/all of the SHM proj-
ect component(s) are immense. To streamline and formalize the discussion of the interrelationship
between the DM tools and the SHM components (and subcomponents) we first introduce the con-
cept of decision making tool box (DMTB). The DMTB will enumerate the main subcomponents of
DM tools, as shown in Table 8.1.
Decision making components, as shown in the DMTB (Figure 8.3) have been presented and stud-
ied by numerous authors in many fields. We note that one, or more, of the techniques employed by
those tools will play a role during an SHM project. As shown earlier, SHM includes measurements,
structural identifications, and damage identifications (DMIDs) components. As such, each of the
decision making tools mentioned above can be used in any of these components. Finally, since SHM
projects are usually designed to assess one or more hazards (corrosion, scour, earthquakes, impact,
normal wear and tear, etc.); it is also reasonable to conclude that the particular decision making
technique to be utilized in the project will be greatly affected by the hazard (or hazards) under
consideration.
495
Ultimate goals
Cost reduction Improve safety
SHM project
Measurments
(sensors and
instrumentations)
Structural Decision making

Identification tool chest
Damage
identification
FIGURE 8.1 Decision making and SHM effort.
TABLE 8.1
Decision Making Tool Box (DMTB)
Main components of DMTB
Risk management
Risk assessment
Decision and utility theorems (Figure 8.2)
Cost–benefit analysis
Life cycle cost analysis
Statistical analysis
Probabilistic methods and techniques
Stochastic processes, including Markov Processes
Basic economics
8.2 DECISION MAKING PROCESS AND STRUCTURAL HEALTH COMPONENTS

8.2.1 Introduction
Decision making tools should be employed at every phase of any SHM project. Many of the deci-
sions that involve the needs, goals, and methodologies of the project should be established before
embarking on an SHM project, not after the project is finished. Of these, the need for the project is
perhaps the most important. The decision maker should embark on a process that established the
need for such an effort. In performing such an effort, the decision maker will invariably use one, or
more, of the techniques of this chapter.
After the need for an SHM project is established, all of the four phases of the structural health
in civil engineering (SHCE) project as described in Chapter 2, will require the use of ensemble of
decision making tools, as shown in Figure 8.4. In the next four sections we discuss the utilizations
of the DMTB in each of the four phases of the SHCE field. We then discuss how the use of ensemble
of the DMTB techniques can produce a successful SHCE project.
Decision Making in IHCE 497
Decision making elements
Cost–benefit (value)
Utility
Management
overrule
FIGURE 8.2 Elements of decision making.
Decision making tool box
Decision and
Statistics Probabiliy Stochastic Economics Risk
utility theories
Different Different
Cost–benefit/
statistical probabilistic Marcov chains Management
value
methods methods
Signal Life cycle

Assessment
processing cost
Rate of
return,
present value,
etc.
FIGURE 8.3 Decision making tool box (DMTB).
8.2.2 Sensors and Measurements: Data Collection

Sensing and measurements are the first part of any SHM project. Different decision making tools
and processes will be needed to ensure successful execution of this phase. Among the interrelation-
ships between the sensing/measurements efforts and decision making tools are the following:
Planning of the Measurement and Sensing Phase: Optimizing costs of the SHM experiment and
the costs of measurements must be one of the goals of the SHM project. Chapter 2 offered “General
Theory of Experiments” for optimizing cost of experiment. “Special Theory of Experiments” for
optimizing cost of measurements was also presented in Chapter 2. The utilization of different
Different DM tools can be

used at different stages of
an SHM project
nts
Overall project
o ne
co mp
M
SH
DM
I D
DM
RID
ST
g
sin
Sen
Risk
Valu
e
LCC
Stat
istic
s DM
tool
ches
t
FIGURE 8.4 Decision making tool box and elements of structural health.
decision making tools, such as cost–benefit/value, utility, probability, risk, and economics theories
and techniques must be planned during the measurements/sensing phase of the SHM project.
Data Organization: Organizing data in efficient and simple databases is not an easy task, espe-
cially for large and long SHM projects. Signal processing and statistical techniques needs to be
employed in this phase.
Data Analysis: Generating histograms, probability density functions (PDFs) and cumulative
distribution functions (CDFs) is part of this step. Signal processing techniques (see Chapter 7) might
also be needed. Efficient archiving and data sorting retrieval methods and software need to be
planned and employed.
Consistency with Overall Project Goals: The lore of advanced technologies can have an
undesirable effect on achieving the goals of the SHM project. Use of advanced technology must be
consistent with the overall goals of the project. All parts of the experiment: sensors, data processing and
instrumentation must be well integrated, and in full conformity with the overall project objectives.
8.2.3 Structural Identification: Vulnerabilities and Mitigation

Decision making process, including the many methods of the process relies heavily on STRID
component of SHM. Such interaction between the two subjects should be accommodated in any
decision making process (see Figure 8.5). Some of the ways in which the two fields interact together
are as follows:
Theories of STRID: Some STRID methods are based on DM Methods. For example, artificial neu-
ral network (ANN) was presented as an STRID method in Chapter 6. The ANN method can also
be used to aid in project prioritization and cost–benefit analysis. We note that structural reliability
method that is considered a decision making technique in this document (it aids the decision maker
to decide on the safety of the structure) can also be considered a structural identification technique,
since it explore structural behavior at its very core.
Risk: Hazards, Vulnerabilities, and Consequences: The concept of risk, as described later in
Section 8.6, deals with all three aspects of structural health: hazards, vulnerabilities, and conse-
quences. All structural identification methods explore the relationships between these three aspects
from structural behavior view point. All STRID methods deal with inputs (hazards) structural
Set goals
Cost–benefit of the identification Start
Choose appropriate ID method identification
etc.
Start
Start here Decision making tool chest Structural identification

tools
Are results and goals

Structural compatible?
identification Cost–benefit implications of
process must results End of identification
start and end with What to do with results process
some DM Analysis of results (risk,
invistigations statistics, probabilities, etc.)
etc.
FIGURE 8.5 Structural identification and decision making process
properties (vulnerabilities) and consequences (structural response). The methods can vary from
simplistic to fairly complex, but the process is still the same. Because of this, the use of risk methods
as a decision making tool must utilize STRID process in one form or another.
Decision Making Pre-STRID: In many situations, the STRID process itself must be preceded
by decision-making analysis. For example, in a situation where the professional needs to evaluate
maintenance budget allocation, an STRID condition assessment investigation of some of the bridge
network might be needed to aid in prioritizing available funding.
Decision Making Post-STRID: The reverse to the previous situation is also possible in some
cases. For example, in a large network of bridges, it might be necessary to decide a priori which
bridges might require further STRID condition assessment. In such situations, a simple risk-
based assessment can help the decision maker to target fewer bridges for further STRID condition
assessment.
Capacity/Demand (Reliability): As was discussed earlier, structural reliability methods can be
considered both a DM tool and a STRID tool. Efficient and accurate reliability studies can be suc-
cessful only if this dual identity is recognized. This recognition means, among other things, that a
reliability analysis should pay equal attention to different structural analysis rules and needs and the
uncertainties of parameters rules and needs.
Probabilistic Structural Methods: Probabilistic structural methods are well known. Some of
these methods are component based, such as probabilistic behavior of beams or axial members, see,
for example, Elishakoff (1983), Neubeck (2004), or Melchers (2002). Other methods can analyze the
uncertainties in the structure on a system level, such as probabilistic finite element methods and proba-
bilistic boundary element method, Ettouney et al. (1989a and 1989b). These types of methods can help
the decision maker to accommodate structural uncertainties during any decision making analysis that
deals with uncertainties of structures. Section 8.8 of this chapter is devoted to this subject.
8.2.4 Damage Identification

Damage identification process is interrelated to decision making process in many ways. Decision-
making efforts must precede DMID efforts; they must also follow DMID effort. We describe these
relationships and others next (Figure 8.6).
Decision Making before DMID Efforts: Before embarking on any DMID, the decision maker
should decide if a simple nondestructive testing (NDT) effort (including visual inspection) can be
NDT or SHM?
Decision making cost–benefit
before Utility
damage identification Type of damage
Localized damage
NDE tools
Decision making (NDT type)? Midcourse correction?
during
damage identification When the project is/
Global experiment NDE tools should be finished?
(SHM)? DMTB tools
What to do?
Why?
Decision making
Decision making and
after Cost–benefit
damage identification
damage identification
Is there another
type of damage?
FIGURE 8.6 Decision making and damage identification process.
used instead of the more costly SHM? The advantages and limitations of both NDT and SHM are
discussed in several other parts of this document (see, for example, Chapters 2 and 7). Another
important decision to make is to evaluate the utility, or cost–benefit (value) of DMID. In particular:
the decision maker should evaluate, before embarking on a DMID process, the benefits, or utilities
that will be realized from the process.
Another less obvious decision that needs to be considered in the DMID process is the type of
damage that needs to be identified. As discussed in Chapter 7, there are numerous types of damages.
The tools to identify those different damages are different in technology, cost, and types of results.
A careful analysis is needed to ensure the most efficient DMID process.
Decision Making during DMID Efforts: Decision making tools during the DMID effort will
depend largely on whether the DMID effort is localized, NDT type effort, or a more general SHM
effort. For NDT type effort, the decision making effort is mostly technical. In such a situation, ana-
lytical NDE tools can be used. For the more general SHM projects (which cover larger areas of the
structure, and are usually done on a continuous or intermittent time spans), both NDE analytical tools
as well as decision making effort that is specific to the SHM project of interest must be performed.
For example: the quality and applicability of the data that is being gathered by the SHM project must
be evaluated often to ensure that they are consistent with the overall project goals. Also, the data
must be evaluated to see if there is a need for midcourse correction, or changes. See Section 8.4.5
on sequential decision making for more on this subject. Another decision that needs to be addressed
during the DMID effort is if/when to finish/stop the SHM project.
Decision Making after DMID Efforts: After the DMID process ends, the implications of finding/
not finding the sought after damage must be studied. A plan describing what to do, and why, if any
damage is identified must be part of the original project plan. Finally, the decision maker must be ready
for the serendipity principle, which was presented in Chapter 1. The serendipity principle in the DMID
situation is whether a damage that is different from the originally sought after damage is discovered.
8.2.5 Decision Making

Decision making process is involved with all components of SHM, as we have discussed.
However, after all information from sensing/measurements, structural identification, DMID has
been gathered, the decision making now becomes the central subject: what to do with all of
Prioritization
Network level Type of attention
Level of attention
Different needed
projects
Single system
Cost benefit
Safety life cycle issues
Simple maintenance
Cost Small repair
Specific project Large repair
Partial replacement
Do nothing
Utility of options
Project options Cost benefit
life cycle
Levels of project and decision making
FIGURE 8.7 Levels of projects and decision making issues.
this information? The answer of course depends on the needs of the decision maker. There are
many possibilities at graduating level of detail, as shown in Figure 8.7. Depending of the level
of interest, the type of decision making process differs. Actually, it is paramount that such level
and needs established before embarking on an SHM project, not after the project is finished.
Different levels and types of decision making issues for each level are discussed next. Bridge
structures will be used in this discussion; however, the discussion is applicable to any type of
structural systems.
Network Asset Level: DM tools can identify and prioritize systems (bridges) that need atten-
tion. For example, simple DM vulnerability assessment tools can help in identifying bridges that are
vulnerable to scour or seismic damage, and the degree of that vulnerability. In addition, the type and
level of attention can also result from such network-level investigation.
Single System (Bridge) Level: DM tools can prioritize between needed projects. For example:
painting, seismic retrofit, or diagnostic testing. This level can use STRID, DMID, or sensing type
projects. The varied nature of types of projects makes this particular process difficult, yet, with the
choice of appropriate DM tool, or tools, adequate decisions can be made.
A specific Project Level: After a specific project has been identified, the decision maker needs
to decide on what type and level of effort is needed: (a) simple maintenance, (b) small repair, (c)
large repair effort, (d) partial replacement effort, or (e) do nothing. This level of decision making
might require several decision making tools that range from simple cost–benefit analysis to complex
risk analysis, depending on the particular circumstances of the project.
A Given Project Option: When the professional decides upon the type of effort for a par-
ticular project, the type of action needs to be explored. For example, after deciding to retrofit and
strengthen a group of reinforced concrete columns, the decision maker needs to decide on the most
beneficial type of retrofit. Among potential retrofit option might be to use FRP (fiber reinforced
polymers) wrapping, replace steel rebars, or add steel jacketing. There are several issues that need
to be considered: useful life of retrofit, initial cost of different retrofit options, and life cycle costs.
At All Decision Making Levels: The main objective in SHCE is to ensure safety at reasonable
costs. As such, it is important to explore safety, costs, benefits, utility during all decision making
processes, if pertinent.
8.3 PROBABILITY AND STATISTICS

8.3.1 Introduction
Different concepts of statistics, probability, regression, and correlation will be discussed in this
section. Whenever possible, the use of those concepts in the SHM/SHCE field is highlighted. These
concepts are used in examples throughout this volume.
8.3.2 Statistics
8.3.2.1 Data Collection
Almost all SHM project collect information about system. This information is then grouped in data
sets or databases. One of the most important steps in an SHM project is the details of these data-
bases and how they are organized. A good data structure would have the following attributes:
• Complete/encompassing: it should cover all important aspects of the project. For example: in
a modal identification project, there must be enough sensors that are located at adequate loca-
tions for producing the minimum number of mode shapes that are required by the project.
• Simple: The simpler the data structure, the easier it would be to extract information from.
• Expandable: Can be added upon at later times without major difficulties.
• Suitable for analysis: The analysis that would be performed on the collected data must be
kept in mind while deciding upon the data and the data structure itself. Mismatch between
the data, the data structure, and the analysis type can lead to a failed project.
8.3.2.2 Frequency and Histograms

A group of data points xi that is collected in a vector X = {xi}, can be grouped in ranges, where the
number of occurrences of xi within a predefined ranges are computed into another array Y. Note that
i = 1, 2, …, Ni and Ni is the size of X. This array is the frequency array. Thus, by definition
{ }
Y = yj (8.1)
y j = Number of occurrences which xi satifies the condition R j ≤ xi <R j +1 (8.2)
Where Rj describes the range limits, with Nj describes the total number of ranges.
Identifying the frequency of a data set is used extensively in SHM projects to identify statisti-
cal properties of the data. Consider, for example, the partial data set in Table 8.2. The complete set
contains 1000 records of sampling savings (units of $1,000) of a particular project as estimated by
an official. The maximum and the minimum records of this data set are $5,402.29 and $4,518.37,
respectively. The official decided that ten range limits are adequate to study the frequency of occur-
rences of this data set. The ranges and the frequency of occurrences are shown in Table 8.3. Note
that the total number of occurrences is 1000, which is the same as the total number of records in the
data set. A graphical display of the frequency is called the histogram is shown in Figure 8.8. The
histogram is also used extensively in analyzing data during SHM projects.
8.3.2.3 Random Variables

There are numerous ways to define a random variable, see, for example, Benjamin and Cornell (1970)
or Schlaifer, R (1969). These definitions range from mathematically complex to the fairly qualita-
tive toss of coin experiment. For our purposes, we observe that almost all parameters that pertains
to SHM are random in nature, that is, they do not have a constant value. Random variables can be
defined using their statistical properties (mean, variance, etc.) In some instances, the changes in the
parameter are small enough to permit the professional to deem it constant, or deterministic at a given
time and/or space. Deterministic variables can be defined completely by a single valued number.
TABLE 8.2
Example of a Data Set
Sampling of Present Value of Savings of an SHM Project (Only first 50 records of 1000 records are shown)
Record # Value Record # Value Record # Value Record # Value Record # Value
1 5111.05 11 5048.01 21 4834.12 31 4936.97 41 4938.61
2 5066.48 12 4899.32 22 4932.45 32 4906.50 42 4948.48
3 5007.72 13 5133.57 23 5032.43 33 5156.15 43 5057.90
4 4929.55 14 4764.81 24 4897.59 34 5150.22 44 4846.75
5 4675.08 15 4944.34 25 5039.52 35 5167.41 45 4890.13
6 4906.10 16 4891.67 26 4802.79 36 4941.53 46 5084.11
7 4751.34 17 4917.50 27 5075.22 37 5044.49 47 5059.83
8 5038.29 18 5059.30 28 4954.99 38 5001.62 48 4916.91
9 4889.03 19 4914.10 29 5244.73 39 4931.59 49 4855.30
10 5010.23 20 4886.22 30 4902.96 40 5058.99 50 4813.91
Note: The maximum and minimum of the complete data set are 5402.29 and 4518.37, respectively.
TABLE 8.3
Ranges and Frequencies
of Occurrences
Range Frequency
4500 0
4600 1
4700 11
4800 55
4900 171
5000 274
5100 271
5200 153
5300 57
5400 6
5500 1
5600 0
Random variables can be either discrete or continuous. Discrete random variables as the name
implies can take only finite set of values: they can be counted. Continues random variables have
infinite set of values. In SHM, both types of random variables can be encountered. Mathematical
tools exist for both discrete and continuous versions of the random variables. We will encounter
both types throughout this document.
A discrete random variable is then collected in the form xi, with i = 1,2, …Ni and Ni is the size of
data set (sample). A continuous random variable is expressed simply as x, with −∞<x<∞.
8.3.3 Probability Distribution

8.3.3.1 Discrete versus Continuous Distributions
The CDF of a continuous random variable X is defined as
F ( x ) = P ( X ≤ x ) (8.3)
Histogram of present value data set

300
250
200
Frequency
150
100
50
0
00
00
00
00
00
00
00
00
00
00
00
00
5
55
<4
–4
–4
–4
–4
–5
–5
–5
–5
–5
–5
>
00
00
00
00
00
00
00
00
00
00
45
46
47
48
49
50
51
52
53
54
Present value ($1,000.00)
FIGURE 8.8 Histogram example: present value data set.
A PDF of a continuous random variable is

d F ( x)
f ( x) = (8.4)
dx
Similar expressions exist for discrete random variables. We note that the PDF of a discrete ran-
dom variable can easily be generated from the frequency data set that was described earlier. See
Benjamin and Cornell (1970) for more details.
8.3.3.2 Expectations and Moments of Random Variables

A general expression for the nth expectation (moment) of a random variable x is
+∞
x n f ( x ) dx (8.5)
( )
E xn = ∫
−∞
Where f(x) is the PDF of x.
For n = 1, the expectation is commonly called expected value, mean, or average of the random
variable. It is expressed as
+∞
m = E ( x) = ∫ x f ( x ) dx (8.6)
−∞
The variance V and the standard deviation σ 2 are defined as

2
V = s 2 = E ⎡⎣ x − E ( x )⎤⎦ (8.7)
We get the useful relation

( )
V = s2 = E x 2 − E [ x ]
2
(8.8)
+∞
2
V = s2 = ∫ ( x − m) f ( x ) dx (8.9)
−∞
Similar expressions exist for discrete random variables.

Higher moments n > 2 are used in statistical analysis applications.
8.3.3.3 Uniform Distribution

Uniform PDF is expressed as
1
f ( x) = for a < x < b (8.10)
b−a
f ( x ) = 0 elsewhere (8.11)
The mean and variance are

a+b
m= (8.12)
2
2
( b − a)
V= (8.13)
12
Because of its simplicity, it is used frequently in SHM projects (see Figure 8.9).
8.3.3.4 Normal and Truncated Normal Distribution

Normal distribution is expressed as
1 ⎛ ( x − m )2 ⎞
f ( x) = exp ⎜ − ⎟ (8.14)
s 2p ⎝ 2 s2 ⎠
The mean is µ and the variance is V = σ2. Equation 8.14 has infinite tails on both sides of the scale.
In many physical applications, it is required to truncate these tails to reasonable physical limits. For a
truncated normal distribution, Equation 8.14 mentioned above, can be rewritten as (Elishakoff 1983)
A ⎛ ( x − m )2 ⎞
f ( x) = exp ⎜ − ⎟ for a ≤ x ≤ b (8.15)
s 2p ⎝ 2s 2 ⎠
f ( x) = 0 for a ≤ x ≤ b (8.16)
A = 12 erf ( k ) (8.17)
1 k ⎛ y2 ⎞
erf ( k ) = ∫ exp ⎜ − ⎟ dy (8.18)
2p 0 ⎝ 2⎠
f(x)
1/(b-a)
a b
Variable - x
FIGURE 8.9 Uniform PDF.

f(x)
Truncated normal distribution curve
(Limits at x = a and x = b)
Normal distribution curve
a b
Variable - x
FIGURE 8.10 Uniform and truncated uniform PDF.
See Figure 8.10 for illustration of normal and truncated normal distributions. Uniform distribution
is used in many SHM situations. For example, it can be used to represent the random behavior of
strength of large number of concrete test samples.
8.3.3.5 Poisson’s Distribution
m x e−m
f ( x) = (8.19)
x!
The mean and the variance of Poisson distribution is µ. Poisson distribution is used in numerous
physical applications. For example, it can be used to represent the number of arriving trucks to a
highway bridge during a given interval of time.
8.3.3.6 Marginal Distributions
In many SHM applications we are confronted with more than one random variable. Consider a situ-
ation with two continuous random variables x and y. A joint CDF can be define as
F ( x, y ) = P ( X ≤ x, Y ≤ y ) (8.20)
The joint PDF is

∂ 2 F ( x, y )
f ( x, y ) = (8.21)
∂x ∂y
The marginal CDF is

x +∞
F ( x) = ∫ ∫ f ( x , x ) dx
−∞ −∞
1 2 1 dx2 (8.22)
And the marginal PDF is

+∞
f ( x) = ∫ f ( x, x ) dx
−∞
1 1 (8.23)
8.3.4 Conditional Probabilities: Bayes Theorem

For two random variables the conditional probability that relate them is
f ( x, y )
f (y x) = (8.24)
f ( x)
The conditional probability expression 8.24 is generalized to Bayes theorem which states that
( )
P Aj qi P (qi )
(
P qi Aj = ) k=N
(8.25)
(
∑ P Aj qk P (qk )
k =1
)
Bayes theorem expresses the probability of a state θi, given another state Aj in terms of
• The marginal probability of that state θi

• The conditional probabilities of Aj given all other possible states of θ
• All marginal probabilities of θ
Symbolically, Benjamin and Cornell (1970) presented Equation 8.26 as
P (Sample State ) P (State )

P (State Sample ) = (8.26)
∑ P (Sample State ) P (State )
All States
Bayes theorem has many applications in the field of SHM. There are many examples that show its
use in many parts of this volume.
8.3.5 Confidence Intervals and Chebyshev Inequality

For any random variable x the knowledge of the mean µ and standard deviation σ can be used in pro-
ducing the probability of that variable within any given range, as shown in the following equations
1
P ( x1 ≤ x ≤ x2 ) = 1 − (8.27)
a2
x1 = m − a s (8.28)
x2 = m + a s (8.29)
The above equation is known as Chebyshev inequality. It was described in full by Benjamin and
Cornell (1970). Daddazio and Ettouney (1990 and 1991) utilized the method in acoustic radiation
problems, while Ettouney et al. (1993) used it to solve soil problems, and Daddazio and Ettouney
(1987) used it to address uncertainties in boundry conditions for impulsive loads on structural sys-
tems, respectively. The expressions for one sided inequalities are
a2
P ( x1 ≤ x ) =
1+ a2 (8.30)
a2
P ( x ≤ x2 ) =
1+ a2 (8.31)
8.3.6 Functions of N Random Variables

8.3.6.1 General
It is easy to express the expectations of simple functions in an analytical form as shown in Table 8.4.
Note that Y is a function of the random variables x, x1, or x2.
8.3.6.2 Taylor Series

There are numerous situations when the known functional dependence of several random variables
is too complex. In such situations the use of the simple expressions of Table 8.4 is not possible. A
simple, yet not too widely know method of finding the expectations of such complex function is
Taylor series approach. Consider the functional relation
Y = Y ( x1 , x2 , xi ,, x N ) (8.32)
Where xi is a set of independent random variables and i = 1, 2, 3, …, N. It is assumed that the math-
ematical expectations of xi are known. Recall that the expectations of order n are related to the PDFs
f(xi) as
+∞
E xin = ( ) ∫ x f ( x ) dx n
i i (8.33)
−∞
In practice, the first two moments are used, for example, the mean (expected value) E(xi) and the
( )
mean square E xi2 .
For simplicity, we define the mean of xi as
xi = E ( xi ) (8.34)
We need to evaluate the mathematical expectations of Y. Specifically, we need to evaluate the mean
(expected value) E(Y) and the mean square E(Y2).
For complex functional relationships, it is not easy to express E(Y) and E(Y2) in simple expres-
( )
sions of E(xi) and E xi2 . A general solution that can be used is by utilizing the Taylor serious expan-
sion of Equation 8.32. The expansion around xi can be expressed as
N
1 N N
Y = Y + ∑ ei Y ,i + ∑ ∑ e e Y , + (8.35)
i 2 i j i j ij
( )
Higher order terms O e3i and higher, are assumed to be of negligible value. Note that
ei = xi − xi (8.36)
Y =Y xi = xi
(8.37)
∂Y
Y ,i = (8.38)
∂ xi xi = xi
and
∂ 2Y
Y ,ij = (8.39)
∂xi x j
xi = xi , x j = x j
TABLE 8.4
Expectations of Simple Functions
Function E(Y)
Y = cx cE(x)
Y = x1 + x2 E(x1) + E(x2)
Y = x1x2 E(x1) E(x2): assuming that
x1 and x2 are independent
Taylor series expansion Equation 8.35 is now in a form that can easily define E(Y) and E(Y2). From
Equation 8.36, by definition:
E ( ei ) = 0 (8.40)
and
( ) ( )
E ei2 = E xi2 = Vi (8.41)
Given that the random variables xi are also independent, then
( )
E ei ej = 0 (8.42)
Taking the expectations of Equation 8.35 we get
1⎡ N ⎤

2⎣ i
( )
E (Y ) = Y + ⎢ ∑ E ei2 Y ,ii ⎥ (8.43)
⎦
and
⎡N
( ) ( )( 2 ⎤
E Y 2 = Y 2 + ⎢ ∑ E ei2 Y Y ,ii + (Y ,i ) ⎥ ) (8.44)
⎣ i ⎦
If the random variables xi are dependent, that is, E(εi εj) ≠ 0, the expectations for N = 2 case are
defined as
1

2
( ) ( )
E (Y ) = Y + ⎡⎣ E e12 Y ,11 + E e22 Y ,22 ⎤⎦ + E ( e1e2 )Y12 (8.45)
and
( ) ⎢⎣ ( )( 2
) ( )( 2
E Y 2 = Y 2 + ⎡ E e12 Y Y ,11 + (Y ,1 ) + E e22 Y Y ,22 + (Y ,2 ) ⎤
⎥⎦ )
+2 E (e1e2 ) (Y Y ,12 +Y ,1 Y ,2 ) (8.46)
The expectations for N = 3 are
1
E (Y ) = Y +
2⎣
( ) ( ) ( )
⎡ E e12 Y ,11 + E e22 Y ,22 + E e32 Y ,33 ⎤
⎦
+ E (e1e2 ) Y12 + E (e1e3 ) Y13 + E (e2 e3 ) Y23 (8.47)
and
( ) ⎣⎢
( )( 2
) ( )( 2
) ( )( 2
E Y 2 = Y 2 + ⎡ E e12 Y Y ,11 + (Y ,1 ) + E e22 Y Y ,22 + (Y ,2 ) + E e32 Y Y ,33 + (Y ,3 ) ⎤
⎥⎦ )
+2 E (e1e2 ) (Y Y ,12 +Y ,1 Y ,2 ) + 2 E (e1e3 ) (Y Y ,13 +Y ,1 Y ,3 )
(8.48)
+2 E (e2 e3 ) (Y Y ,23 +Y ,2 Y ,3 )
Once the expectations of Y are computed, the variance V, and standard deviation σ are computed
using
( )
V = E Y 2 − E 2 (Y ) (8.49)
and
s= V (8.50)
respectively.
Taylor series approach is useful when the functional form of the random variables is known. For
example, it can be used in evaluating the probabilistic behavior of load rating of bridges or cost esti-
mates of failure. Also, the formal expressions of scour hazards are good applications for the Taylor
series approach.
8.3.6.3 Numerical Integration

Another method of obtaining the expectations of a function such as Y in Equation 8.35 is to utilize
numerical integration scheme. Such an approach would offer less insight to the problem than the
Taylor series approach. In such situations, the reader is advised to consider the Monte Carlo simula-
tion method as an alternative. Monte Carlo method offer more general solution, with similar level
of computational demands.
8.3.7 Regression and Correlation Analysis

Consider an SHM experiment that measure the humidity levels x at a reinforced concrete bridge
site and the rate of corrosion y in the bridge. After N readings, the professional wanted to use the
findings to estimate the relationships of corrosion rate at other locations that are similar to the
SHM testing location. Also, the professional wants to study if there is any correlation between
the humidity level and the rate of corrosion. Formally, for a given set of data pairs, xi and yi,
where i = 1, 2, …, N, find a relationship
y ( x) = a x + b (8.51)
where a and b are constants. Find also a factor r that indicates the degree of correlation between
humidity levels and rate of corrosion. We discuss both methods next.
8.3.7.1 Linear Regression

The constants a and b can be estimated using a minimum least square approach. It aims at reduc-
ing the least square of the differences between the line and the data pairs xi and yi. It can be shown
that
∑ ( x − m )(y − m )
i= N
i =1 i x i y
(8.52)
b= i= N 2
∑ (x − m )i =1 i x
and
a = m y − bm x (8.53)
Note that µx and µy are the averages of the data sets xi and yi, respectively.
Equations 8.51 through 8.53 define the linear regression method. The parameter a is the intercept
of regression line, and b is the slope of the regression line. Studying the regression line can give
insights to data trends as well as the relationships between the two sets of parameters under consid-
eration. Figure 8.11 illustrates the concept of regression lines.
Slope (trend)
Corrosion rate
Intercept
Humidity
FIGURE 8.11 Regression analysis concepts.
Note that regression analysis can be done with nonlinear equations. It also can be used to identify
moving averages for stochastic (time dependent) problems. It offers many powerful tools that can
have many applications in the field of SHM/SHCE.
One final comment on regression analysis: we note that the method of ANN (see Chapter 6)
offers a stronger tool that relates data sets with complex interrelationships. It lacks the simplicity of
regression analysis, but it offers in return a far more powerful technique that can be used in many
SHM situations.
8.3.7.2 Correlation Analysis

To obtain the correlation between xi and yi we use the equation
r=
i= N
(
(1 / N − 1)∑ i =1 ( xi − mx ) yi − my ) (8.54)
sx sy
where σx and σy are the standard deviations of xi and yi. The correlation coefficient −1≤r≤+1 indicates
how closely the two parameters correlate. For r = 0, there is no correlation between x and y, and
they are completely independent. For r = +1, there is strong positive correlation, that is, when one
variable changes, the other variable changes in same direction. For r = −1, there is strong negative
correlation, that is, when one variable changes, the other variable changes in opposite direction.
Correlation coefficient of Equation 8.54 can be generalized to form correlation matrix between
more than two parameters. In such a situation, the correlation coefficients can form a correlation
matrix that can be of benefit in the SHM field, see Pendat and Piersol (2010).
8.4 TRADITIONAL THEORIES OF DECISION MAKING

8.4.1 Elements of Traditional Decision making
8.4.1.1 Value versus Utility Methods
Perhaps the most obvious decision making metric is the monetary metric. If there are several
options that are available to a decision maker, the optimal choice would be the one that is the least
expensive, or the one that provide the most savings, and so on. These have produced the traditional
monetary cost–benefit decision making methods (see, for example, Treuman 1977). The concept of
value can be linked to the cost–benefit in a mathematical form, as a ratio or subtraction. Because of
this, value will be used in this volume interchangeably with cost–benefit.
Traditionally, it was argued that the two main disadvantages of the highly quantitative cost–
benefit approach are that
• It requires too much information that might not be available to the decision maker
• It lacks the subjective feel that many decision makers have; such subjective feel is extremely
difficult to enumerate in a true cost–benefit analysis
To avoid such limitations, Von Neumann and Morgenstern (1944) developed a new decision mak-
ing process that is based on developing utility functions. The theory of utility has been used since
then as an alternate method for decision making processes. Table 8.5 shows the main differences
between traditional cost–benefit/value and utility decision making methods.
Before we explore the two methods in more detail, we would like to note that we propose to
generalize the concept of cost–benefit/value beyond monetary or quantitative methods to include
more subjective issues. We already mixed objective and subjective parameters in Chapters 4 (when
evaluating values of experiments) and 5 (when evaluating suitability of sensors to experiments). As
such, we believe that our generalized utilization of the cost–benefit/value actually encompasses
both attributes of traditional cost–benefit and utility methods.
8.4.2 Utility Method

8.4.2.1 Theory of Utility
As was mentioned earlier, the utility method has several advantages to the traditional cost–benefit
(value) method in decision making. We present the utility theory, its mathematical basis, and its use
in the field of SHM in this section. We start by presenting a definition of utility by Trueman (1977)
as, “Utility is the subjective numerical measure of the value of an act to a decision maker when a
particular event occurs.” In other words, a utility is a subjective definition of value.
There are as many types of utility metrics as there are decision makers. Every decision maker
can establish the type of utility as it suites the project on hand. Of course, care must be taken when
deciding upon the type of utility used. Some good attributes of utility metrics are as follows:
• Accuracy: The utility used must be accurate in describing the type of project. At least it
must be consistent with the spirit of the project. For example, monetary or cost utility is
suited for projects where costs are important factor in the decision making process. Safety
as a metric might be a good utility for bridge security measure. Availability is a good met-
ric for research oriented SHM project.
• Simplicity: The utility metric should be simple to sort out. For example: monetary utility
is fairly simple to understand and apply. In some risk-related decision making processes,
utilities such as vulnerability, asset value, threat, or consequences are used. They are fairly
simple to apply in a qualitative sense.
TABLE 8.5
Comparison between Value and Utility Methods in Decision Making
Value Utility
Mostly quantitative foundations Mostly qualitative
Needs mostly objective/mathematical descriptions Can be adjusted to account for subjective situations
Analytically rigorous Less analytically rigorous
Due to its rigor it does not need as much care in formulating Needs more care in formulating the problem
the problem
Needs less care in interpreting the results Needs care in interpreting the results
• Efficiency: The utility metric should be efficient in describing as many aspects of the proj-
ect as possible. Short of this, multiple utilities might be needed to accomplish this objec-
tive. Relative risk analysis is a good example to this need. Three utility metrics are used
to describe risk levels in many situations: hazard, vulnerability, and consequences. Failure
mode and effect analysis can be used to address this issue. See Johnson and Simon (1997)
and Johnson and Dock (1998) for in-depth description of this topic.
After deciding upon the type of utility X, we need to define the utility function U(X). To accomplish
this, we need to establish few points that relates X to U(X). Some examples will be provided later.
The choice of the utility value depends on the level of experience of the analyst. Thus, in using
utility approach, it is prudent for the analyst to be as familiar with the subject as possible. SHM
projects can also help the analyst to produce accurate utility values.
8.4.2.2 Types of Utility

8.4.2.2.1 Monetary/Cost Utility
The monetary utility is a popular utility that is commonly used in financial decision. Well-known
qualitative illustrations of different monetary utility function are shown in Figure 8.12.
Cost utility describes cost attributes of different aspects of the SHM project. It is especially benefi-
cial when exact costs are not well known to the analyst. For example, if the decision maker needs
to establish a cost-utility function for cost of executing and maintaining an SHM project over the
course of 10 years. There is no available information that can help the decision maker to establish
an objective cost–benefit analysis. The utility approach is then chosen. The utility function bounds
are established first: what are the maximum and minimum expected costs for the project? The deci-
sion maker estimates that the maximum cost is $500,000.00 and the minimum cost is $0 (since this
would be the case of no SHM project to start with). The utility function is assigned values of 1.0
and 0.0 for these bounds. The midpoint U(X) = 0.5, is established next. If the decision maker is risk
neutral, then the midpoint would be X = 0.5(0+500,000) = 250,000. For a more aggressive decision
maker, the midpoint would be higher than 250,000, say X = 350,000. For a conservative decision
maker, the midpoint would be less than 250,000, say X = 150,000.
The three-point utility function is now formed, as shown in Table 8.6, and Figure 8.13. The advan-
tage of the utility approach is now clear: (a) The lack of concise information did not prevent the deci-
sion maker from establishing a utility function using previous experiences, and (b) The attitude of the
decision maker (conservative, neutral or aggressive) is clearly reflected in the utility function.
A final comment on the construction of utility functions: we used three points for the above
example. It is recommended to use five points to establish these functions for reasonable accuracy,
the five-points approach can be done with similar logic to the above.
8.4.2.2.2 Benefit Utility

We propose another possible utility in the field of SHM: the benefit utility. The benefit utility can
be established in a manner that is similar to the cost utility, above. Similar to cost utility, the benefit
1.0 1.0 1.0

Aggresive, high
Utility function
Utility function
Utility function
Neutral risk user

user
Conservative
user
0.0 0.0 0.0

Min Max Min Max Min Max
Monetary value Monetary value Monetary value
FIGURE 8.12 Monetary utility functions.

TABLE 8.6
Cost Utility Functions
Cost of SHM Project ($ Thousands)
Cost Utility Conservative Neutral Aggressive
0.00 0 0 0
0.50 150 250 350
1.00 500 500 500
Cost utility functions

1.00
0.90
0.80
0.70
0.60
Cost utility
0.50 Conservative
Neutral
0.40 Aggressive
0.30
0.20
0.10
0.00
0 50 100 150 200 250 300 350 400 450 500
Cost of SHM project over given number of years ($ thousands)
FIGURE 8.13 Cost utility functions.
TABLE 8.7
Benefit Utility Functions
Benefit of SHM Project ($ Thousands)
Benefit
Utility1 Conservative Neutral Aggressive
0.00 0 0 0
0.50 175 300 425
1.00 600 600 600
utility for both conservative and aggressive decision makers will be similar in shape. Aggressive
decision maker expects more benefit for the same cost than the conservative decision maker and
the reverse is true. For example, Table 8.7 and Figure 8.14 show the benefit utility for the above-
mentioned SHM project.
8.4.2.2.3 Utility of Experimentations

Chapter 4 presented several methods that can help the decision maker in identifying the value of the
experiments in a rigorous quantitative manner. We introduce here another method that is based on the
utility theory to evaluate experiments: utility of experimentation. The utility function can be formed
in same fashion as before. The professional would estimate the maximum and minimum values that
are to be obtained from performing the experiment. A maximum and minimum utility of 1.0 and 0.0
is assigned to the function. The midpoint is then estimated following similar procedures as before.
Benefit utility functions

1.00
0.90
0.80
0.70
Benefit utility
0.60
0.50 Conservative
Neutral
0.40 Aggressive
0.30
0.20
0.10
0.00
0 50 100 150 200 250 300 350 400 450 500 550 600
Benefit of SHM project over given number of years ($ thousands)
FIGURE 8.14 Benefit utility functions.
8.4.2.2.4 Other Types of Utility

There are many other potential utilities to be used in the SHM/SHCE field. For example, utilities
of consequences, vulnerability, damage, and so on can all be established to help the professional in
making decisions.
8.4.2.3 Multidimensional Utilities

Utilities can also be functions of other utilities. Such as the following:
U ( X1 , X 2 ) = a 1 U ( X1 ) + a 2 U ( X 2 ) (8.55)
Where U(X1 , X2) is a utility function of the addition of two utilities The factors α1 and α2 are con-
stants. Also the expression:
U ( X1 , X 2 ) = a 1 U ( X1 ) U ( X 2 ) (8.56)
represents another multidimensional utility.

A savings utility function can be established with X1 representing the SHM project costs in the
above example, and X2 representing the benefits of the SHM project. The SHM savings utility func-
tion over the next 10 years is then
U ( X1 , X 2 ) = U ( X 2 ) − U ( X1 ) (8.57)
Using Tables 8.6 and 8.7 the savings utility function can be computed as in Table 8.8 and Figure 8.15.
8.4.2.4 Expected Utilities

After forming the utility functions, they can be used to compute the expected utility. For example,
suppose that the decision maker has two options to proceed in the SHM project:
• Option A: Have a 25% probability of costing $100,000 and 75% probability of costing
$300,000.
• Option B: Have a 25% probability of costing $150,000 and 75% probability of costing
$250,000.
1.00
0.90
0.80
0.70
Savings Utility
0.60
0.50 Conservative
Neutral
0.40 Aggressive
0.30
0.20
0.10
0.00
0 25 50 75 100
Savingsof SHM project over given number of years ($ thousands)
FIGURE 8.15 Savings utility function.
TABLE 8.8
Savings Utility Function
Savings of SHM Project ($ Thousands)
Savings
Utility Conservative Neutral Aggressive
0.00 0 0 0
0.50 25 50 75
1.00 100 100 100
TABLE 8.9
Expected Cost Utility Comparison
Option P X1 ($) (1 − P) X2 ($) U(X1) U(X2) E[U]
A 0.25 100,000 0.75 300,000 0.33 0.72 0.62
B 0.25 150,000 0.75 250,000 0.41 0.65 0.59
Assuming that the decision maker is using the conservative utility function curve of Figure 8.13, the
expected utility can be estimated using
E [U ] = p U ( X1 ) + (1 − p )U ( X 2 ) (8.58)
Table 8.9 shows the details of the computations. The expected utility of option A is higher, thus it is
the better choice in this case.
The process can continue to accommodate the benefit and savings utilities. Such an exercise is
more involved computationally, but it is similar to the cost-only exercise; we will leave resolving it
to the reader.
8.4.3 Cost–Benefit/Value (Monetary) Method

The cost–benefit/value (some authors refer to it as the monetary) method in decision making is
based on constructing a payoff table similar to Table 8.10. The decision maker is faced with several
TABLE 8.10
Example of Payoff Table in Decision Making
State of
Nature/Events Probability Decisions/Actions
V1 V2 Vj VN
θ1 p1 u11 u12 u1j u1N
θ2 p2 u21 u22 u2j u2N
θi pi ui1 ui2 uij uiN
θM pM uM1 uM2 uMj uMN
TABLE 8.11
Payoff Table for Severe Flood Mitigation
Severe Flood Mitigation Costs ($1,000)
Event Probability Do Noting Method – 1 Method – 2 Method – 3
0–1 years 0.10 1000 100 150 200
1–10 years 0.30 500 75 40 20
10–50 years 0.60 100 10 5 3
Total 1.0 310.0 38.5 30.0 27.8
states of nature (events) that are placed in the rows of Table 8.10. They are designated with θi where
i = 1,2,…M, with M is the total number of possible events. The payoff table will also include a set
of decisions/actions that the decision maker would choose from. They are designated Vj with j =
1,2, …, N, with N is the total number of available choices/actions. The body of the payoff table is
filled with the cost that will be incurred if event θi occurred and the action Vj is taken. These costs
are designated uij. Finally, the knowledge of the probability of occurrence of the events θi is needed.
They are designated pi. Note that, by definition Σ1M pi = 1.0.
After the payoff table is fully formed, the cost of each decision is computed as follows:
i= M
C j = ∑ pi uij (8.59)
i =1
The preferred decision is the decision with the minimum Cj.

As an example of the payoff table, consider Table 8.11, which explores severe flood conditions and
different mitigation options. Three potential state of nature can happen: the flood might occur in 0–1
years, 1–10 years, or 10–50 years. Four decisions are available. Three mitigation options plus a do-
nothing option. The costs of different states for different actions fill the body of the payoff table. Also
the probabilities that a severe flood occur for different events are placed in the payoff table. The total
cost of each action is then computed using Equation 8.59. Obviously the do-nothing option is the least
cost-effective option. The three options are fairly close to each in cost, with mitigation option 3 costing
the least.
8.4.4 More on Traditional Decision Making Methods

The payoff and utility methods as presented above can have many extensions that make them much
more useful for decision making process, especially when used in conjunction with an SHM project.
We summarize some of those techniques next.
Posterior Analysis: Note that both utility and monetary methods utilize probability of different
events θi. If there are additional information that are available that relate the events to the different
choices in form of conditional probabilities, this would enable the decision maker to use Bayes theo-
rem in improving the probabilities used in the payoff table. Such improvements will ensure more
accurate decision making process.
Preposterior Analysis: In addition to the conditional probabilities in the posterior analysis, if
specific information regarding the problem on hand is available, it can be used to improve the deci-
sion making process even more. Such information can be in form of additional testing, or additional
data sets.
Sequential Decision Making: During the decision making process if additional information
becomes available it can change the course of the process. Such information can be obtained from
additional testing that is performed by the decision maker to enhance the accuracy of the decision.
Probabilistic versus Deterministic: Even though the methods of this section are all based on
available probabilities of different parameters, all of these methods produce single valued results.
Because of this, they sometimes are referred to as deterministic methods. Those averages will influ-
ence the decision. In many situations there is a need to also know the variances of those results. To
produce both averages and variances, probabilistic decision making techniques are needed. Some
of those probabilistic techniques are discussed in Section 8.6.
There are several examples of those three methods throughout this manuscript. All the examples
are presented in relation to the SHM/SHCE field.
We also note that the above-mentioned three forms of analysis are particularly suited to SHM
projects, since SHM projects are meant for that: to provide information that can help decision mak-
ers to improve their decisions. The only requirement is that such an interaction between SHM
results and decision making needs must be recognized before embarking on either (SHM project
and decision making process).
8.4.5 Use in SHM

Decision making method must be performed in every step of an SHM project. This is especially needed
for such an emerging new field. Some of the benefits of decision making methods can be used in prioriti-
zations of decisions, including (a) comparing costs of SHM options, (b) comparing benefits, costs, and/or,
savings, and (c) helping decision maker in arriving to a subjective decision as to the best course of action.
SHM can also be used to increase accuracy of utility methods to (a) create measurements (datasets)
that can be used to compute different probabilities of events p, (b) create data sets that can help in estab-
lishing accurate utility functions, (c) estimate payoff table entries, (d) produce conditional probabilities,
and (e) use measurements results in accurate probability estimates.
8.4.6 Decision Tree

Decision tree method is a graphical technique that can be used with any of the methods that were dis-
cussed in this section. It is particularly useful for the more complex decision making techniques such
as posterior, preposterior or sequential analysis. It has the additional advantage of showing a global
overview of the whole problem, rather than splitting the analysis into series of tables. Several SHM/
SHCE examples for the use of decision trees are presented elsewhere in this chapter, and throughout this
volume.

We explored many traditional methods for decision making in this section. Several examples were
given, and many more examples are given throughout this manuscript. The reader is encouraged to
refer to these examples. There are two more points we offer regarding traditional decision making
methods. First, those methods were largely developed before the era of modern computing, as such
they are mostly visual, and with simple computational demands. With the availability of fast and
efficient computing, the reliance on computationally demanding simulation methods paved the way
to other techniques that will be explored in other sections in this chapter.
Another attribute of traditional methods is that they seem to lump together hazards, vulner-
abilities, and consequences without a formal way to study these issues and their individual and
combined effects on decision making process. Risk and reliability techniques, as will be presented
later, seem to provide a more formal fashion to address those different issues.
This does not preclude the usefulness of the traditional methods in any manner. It simply adds to the
wealth of methods that are available to the decision maker. We recommend that decision makers utilize
all pertinent decision making methods; this will ensure accurate decisions for problems they face.
8.5 RELIABILITY
8.5.1 Introduction: Reliability and SHM
We discuss in this section the concept of structural reliability: how it can benefit from SHM prac-
tices, and how, in turn, it can aid decision making process. Structural reliability has been studied
extensively by many authors, see Rausand and Royland (2004). We will first summarize essential
concepts of structural reliability: capacity demand, reliability (safety index), and probability of fail-
ure. For the purpose of this document we will be defining structural reliability as simply the prob-
ability of failure. We next discuss general structural reliability evaluation methods, with discussions
of the utilization of these methods for and by SHM technologies. The remainder of the section will
be devoted to detailed examples of utilizing reliability techniques in the fields of SHM.
8.5.1.1 Capacity, Demand, and SHM

Conventional structural design methods included the evaluation of structural capacity C and the
demands from the loadings D. Structural capacity includes ultimate strains, yield stresses, allowable
displacements, allowable accelerations, and so on. Demands from loadings include gravity loads,
live loads, seismic motions, wind pressures, and so on. In general, structural design require that
C≥D (8.60)
Since both C and D are highly uncertain parameters, these uncertainties are accounted for by sev-
eral methods. For example, if the PDFs of the uncertain capacities and demands are defined as C(x)
and D(x), where x is a random variable, then probability of failure:
p f = P (C ≥ D ) (8.61)
Also
p f = P ((C − D ) ≥ 0 ) (8.62)
From Equation 8.61 and given C(x), and D(x), it can be shown that
+∞
pf = ∫ (C ( x) − D( x)) dx
0
(8.63)
The probability of failure pf is the area under the failure density function. The process is shown in
Figure 8.16.
We reach finally a definition of reliability R as the probability that the structure does not fail,
that is,
R = 1 − pf (8.64)
CDF of capacity FC(x)

PDF of demand D(x)
PDF scale
PDF of capacity C(x)
Random variable scale x

Failure density function
Coordinate of failure density Probability of failure pf

curve is FC(x).D(x) is the area under the
curve
Random variable scale x
FIGURE 8.16 Concepts of probability of failure.
8.5.1.2 Reliability (Safety) Index and SHM

If C(x) and D(x) are both normally distributed, then for
Y =C−D (8.65)
Then the mean and variance of Y are
mY = mC − m D (8.66)
VY = VC + VD (8.67)
respectively.
The probability of failure is then
⎛ −m ⎞
p f = P (Y ≤ 0 ) = Φ ⎜ Y ⎟ (8.68)
⎝ sY ⎠
With
sY = VY (8.69)
Note that Ω(ϕ) is the standard normal distribution function. From Equations 8.66, 8.67, and 8.68
⎛ ( −m + m ) ⎞
pf = Φ ⎜ C D
⎟ (8.70)
⎜⎝ s 2 + s 2 ⎟⎠
C D
Defining
b=
( mC − m D ) (8.71)
sC2 + s 2D
f(x)
Since E(CD) and
E(CD2) are known
the probability of failure:
p(CD < α) can easily be
determined
Probability of failure:
p(CD < α) is the area
under the tail end of
Normal distribution
curve Normal distribution
(assumed)
a b
α E(CD)
CD limit mean of CD
of failure
Random variable: capacity demand ratio (CD)
FIGURE 8.17 Probability of failure in a capacity/demand paradigm.
We end up with the well-known expression for probability of failure

p f = Φ ( −b) (8.72)

The function β is known as the reliability, or safety, index. It is the shaded area to the left of the
normal distribution curve of Figure 8.17.
Reliability index is used in SHM and decision making situations extensively. It is discussed in
Chapters 9 and 10 by Ettouney and Alampalli (2012).
8.5.2 Reliability Methods as Applied to SHM

8.5.2.1 Structural Components
Reliability of structural components have been explored by many authors, and well documented.
See, for example, Neubeck (2004) or Melchers (2002). We can relate capacity and demand of com-
ponents in as in Equation 8.65 and evaluate probability of failure as in Equation 8.72. If probabi-
listic properties of C(x) and D(x) are known pf can be computed analytically. If the expressions are
complex pf can be computed using numerical methods such as Taylor series, or general simulation
methods such as Monte Carlo method.
Monitoring techniques can be used to monitor and compute statistical properties of important
variables of Equation 8.72. This will ensure more accurate estimations of pf. An example of com-
puting pf of bridge columns that loses support due to scour is given in Chapter 1 by Ettouney and
Alampalli (2012).
8.5.2.2 Structural Systems
Even the simplest structural system is composed of several components. Because of this, simple
solutions for probability of failure of the system pfs are not possible. In some cases, however, struc-
tural systems can be categorized into systems with simple behavioral attributes. In such situations,
computing pfs can still be done in a reasonable simple fashion.
8.5.2.2.1 Parallel Systems
Parallel structural systems are shown schematically in Figure 8.18. In the literature, the parallel
systems are of interest since the pfs can be computed in a simple fashion (see Neubeck 2004).
For example, for a system with two parallel components the system probability of failure is
p fs = 1 − ( R1 + R2 − R1 R2 ) (8.73)
Force
Components
Components are linked

such that they behave
Force
in parallel
FIGURE 8.18 Parallel systems.
with
p fi = 1 − Ri (8.74)
For three parallel components
p fs = 1 − ( R1 + R2 + R3 − R1 R2 − R1 R3 − R2 R3 + R1 R2 R3 ) (8.75)
Note that pfi and Ri are the probability of failure and reliability of the ith component. Equations 8.73
and 8.75 can be generalized for systems with higher number of components.
Parallel systems are simple in their structural behavior, so, complex building or bridge super
structures are definitely not even remotely resemble parallel systems. However, on closer look, it
appears that bridge bents are good example of parallel systems. This is especially true when the
beams at the top of the bents are more rigid than the columns; this is the situation for relatively
longer column spans as compared to beam spans. This means that it is fairly simple to formulate an
algorithm to monitor the Pf of bridge bents due to most hazards, using 8.73, above, without the need
of complex structural identification system. We give an example of using SHM in monitoring pfs
of bridge bents in case of seismic hazards in Chapter 2 by Ettouney and Alampalli (2012). Similar
SHM setups are possible for other hazards.
8.5.2.2.2 Serial Structural Systems

Serial structural systems are shown schematically in Figure 8.19. Similar to parallel systems, the
serial systems are of interest since its failure probability pfs can be easily computed. The serial sys-
tem will fail if the weakest component of the system fails. Thus, the system pfs is
i= N
p fs = 1 − ∏ Ri (8.76)
i =1
Note that N is the number of serial components and Ri is the reliability of the ith component.
In-series systems can be recognized, after some reflection, as nonredundant or very-low redun-
dant structural systems. An example of SHM use in case of nonredundant steel truss system is given
later in this section.
Serial links Components
Force Force
FIGURE 8.19 Serial systems.
8.5.2.2.3 General Systems

General structural systems that do not conform to the parallel or in-series categorization are much
more difficult to evaluate their reliability. There are several general methodologies that can be used
for this purpose:
Failure Modes and Effects Analysis (FMEA): FMEA method of prioritization and assessment of
system failure have been considered a reliability method. We discuss this method in the risk sec-
tion, since we consider it a risk, not reliability method, since it includes as one of its parameters the
consequences of failure.
Failure Modes, Effects, and Criticality Analysis (FMECA): FMECA method of prioritization
and assessment of system failure is similar to FMEA except that it also adds the criticality of failure
modes in the analysis. It is also based on tabular format in its description of failure modes.
Fault Tree Analysis: Fault tree analysis is a graphical-based method that utilizes the information
that results from FMECA analysis; it identifies the critical failure modes at any level. It relies on
logical relationships such as “AND” and “OR” to relate logical events. During fault tree analysis,
primary and secondary failure modes are identified by the analyst.
Monte Carlo Simulation: This is a general purpose simulation technique that is widely used in
many reliability situations. It is used extensively in this document.
For more extensive discussions about reliability methods of structural systems, see Melchers
(2002) or Neubeck (2004).
8.5.2.2.4 SHM Techniques and Structural Reliability

Structural health monitoring (SHM) techniques can help in evaluating structural reliability in many
ways. For example:
Capacity Monitoring: This type of monitoring aims to monitor and record capacity of systems. For
example, by measuring strains, displacements, or motions, the different capacities of the structure
can be inferred. The different probabilistic properties of capacities can be used in any desired reli-
ability investigation.
Demands Monitoring: Examples of demands are magnitude of loads, frequency of load applica-
tions (such as truck traffic), hazards properties (such as earthquakes, floods, wind, etc.), or changes
in gravity loads over the years. Monitoring of these demands can produce different probabilistic
properties of the demands that can be used in any desired reliability investigation.
Separate and Integrated Capacity—Demand Monitoring: Separate capacity—demand monitoring
involves measuring metrics of capacity and demands simultaneously, then correlating them to produce
appropriate reliability estimates. Integrated capacity—demand monitoring involves using the virtual sens-
ing (VS) concept: trying to find a metric to monitor that can infer failure of a component of a system.
Examples of different types of usage of SHM to estimate reliability of structures are given next.
8.5.2.3 NDT versus SHM Reliability

The probability of failure of a whole system considers all potential failure modes of the system and
their interaction. This is in contrast with the failure of components and their probabilities. Recall that
one of the differences between NDT and SHM is that NDT presumes the knowledge of the approxi-
mate location of the damage. As such, NDT is more suited to studying the probability of failure of
structural components than the probability of failure of the whole structural system. SHM, on the
other hand looks for DMID of the whole structure, or DMID of several components assembly. Because
of this, SHM can be used for more accurate estimations of structural failure probabilities.
An example of this distinction between NDT and SHM is in the use of global vibration modes
in identifying potential loss of flexural strength as reported by Farrar and Doebling (1998). SHM
experiment helped in identifying the bridge flexural modes.
8.5.3 Capacity Monitoring: Avoiding Catastrophic Failure

The sudden and catastrophic failure of the I-35 Bridge in Minneapolis in 2007 brought into attention
the dangers of sudden bridge failures. Sudden failures of bridges, or any other type of infrastructure
should be avoided due to the tragic loss of lives, and immense social costs that will follow. One pos-
sible way to address this problem is through the utilization of SHM technologies.
A potential possibility of sudden failures has always been failure of critical compression mem-
bers in truss bridges. Note that tension members do not usually fail suddenly since they will experi-
ence noticeable elongation (ductile) behavior. Let us consider the truss of Figure 8.20. It is a typical
truss system that is used in many bridges. Note that none of compression members in the truss
is redundant. Thus: if any of them fails suddenly due to local buckling of the member, the whole
truss might fail suddenly. Before we continue further, we need to differentiate between the global
instability of the truss, and the local instability of a compression member. The global instability
of the truss involves the whole system instability, its theory and analytical methods can be found
1 - Truss bridge com-
pression members
Roadway
2 - If for any reason, the

critical load is reached
Axial force
due to reduced capacity,

or increased demand,
or both...
Deformation (lateral) 4 - The compression

member fails: if
3 - Due to the softening attri- there is no alternate
bute of axially loaded beam, if load path for the rest
the critical load is reached, if of the system...
will fail suddenly. in a brittle
manner
Roadway
5 - The whole system

becomes unstable and fails
catastrophically
FIGURE 8.20 Interrelationship between local and global loss of stability.

Axial load Axial load
Hinged-hinged beam Fixed-fixed beam

k=1 k = 0.5
If boundary conditions change,

the k factor will increase, thus
the critical load capacity will
decrease
FIGURE 8.21 Effects of boundary conditions on beam critical load capacity.
elsewhere, Timoshenko and Gere (1961). Our current interest is in the local instability (buckling) of
a truss member (see Figure 8.21). The decision maker wanted to monitor the potential local buck-
ling of a compression member in the truss. The first step in the project is to identify the reasons for
local buckling of the member. This can be done by studying the critical (buckling) axial force in the
member, which can be expressed as
p2 EI
Pcr = (8.77)
( kL )2
Thus the buckling load Pcr is dependent on the flexural stiffness of the member EI. (The modulus
of elasticity is E and the moment of inertia is I.) It is also dependent on the length of the member L.
The effective length factor k is a function of the boundary conditions of the member. Assuming that
the two connections are not experiencing a relative lateral displacement: for a pin-pin connection
k = 1.0, and for fixed-fixed member k = 0.5. If the relative lateral displacements between the two
connections become larger than zero, then k increases. In the limit, as k → ∞, the critical load Pcr →
0, and the member becomes unstable, and fail (see Figure 8.20).
Thus the potential causes of local instability become clear.
• One potential reason is the reduction of flexural stiffness EI due to erosion of material, for
example.
• A change in the factor k. Such a change can occur from changes in the boundary condi-
tions of the member. Such changes can be caused by deterioration of the conditions of one,
or both, of the member connections. Such a deterioration would cause k to increase, thus
reducing Pcr.
• An unexpected large relative displacement between the two ends of the member. Such a
deterioration would cause k to increase, thus reducing Pcr.
All of the above changes are complex changes, and they are mostly not observable by visual inspec-
tion. Moreover, due to the potential sudden nature of the instability failure visual observations
might not be of value in this situation. Another complication would become apparent by examining
Equation 8.77. Note that the critical load is independent of the applied load. Because of this, using
strain sensors is not useful for monitoring local instability of truss members. A novel monitoring
solution is needed.
Before continuing the discussion, we note that this local failure-to-global failure situation occurs
mainly because of lack of structural redundancy in the system. Such a problem is also known as the
progressive collapse problem in building community. Furthermore, we also note that this situation
can be considered as the in-series reliability problem, which was discussed earlier: if one compo-
nent failed, the rest of the system would fail, thus the probability of failure of the components is the
same as the probability of failure of the whole system.
The concept of VS can be useful to resolve this problem. The decision maker recalls that the
natural frequency of a beam that is subjected to an axial force P is
k⎛ P⎞
fP = 1− (8.78)
m ⎜⎝ Pcr ⎟⎠
The governing flexural stiffness of the beam is k and its governing mass is m. If we denote f0 as the
natural frequency of a beam without the effects of the axial force we have
k
f0 = (8.79)
m
Thus
P
f = f0 1 − (8.80)
Pcr
Or
f = 1− P (8.81)
Where f = f / f0 and P = P / Pcr are the dimensionless frequency and axial loads, respectively. Figure
8.22 shows the relationship of Equation (8.81).
This means that for a dimensionless axial load of about 10%, the dimensionless flexural
frequency is about 95%. As the dimensionless axial load increases to 50%, the dimensionless
axial frequency decreases to about 70%. In more specific terms, let us assume that a fixed-fixed
beam as in Figure 8.21 is designed for P = 0.3, which is a reasonable design range for a com-
pression member in a truss. The corresponding dimensionless flexural frequency is f ⊕0.84 .
Let us assume that the beam loses its fixity in one of its ends so that it becomes a fixed-hinged
beam, the effective length k factor can be shown to increase from k = 0.5to k = 0.699. Using
Equation 8.77 the critical load is decreased by 51%. Thus the new dimensionless axial load
is P = 0.3 /.51 = 0.59. The dimensionless flexural frequency is reduced to f ⊕0.63. This is a
25% reduction in the natural frequency, which can easily be detected by a reasonable monitor-
ing scheme. If both sides of the beam lost their fixity, the changes become more severe. The
new effective length factor becomes k = 1.0, and the new dimensionless axial load becomes
P = 0.3 / 0.25 = 1.20 . The beam becomes unstable. If the flexural frequency was monitored, it
would have observed that it is approaching zero (the instability threshold), long before the beam
becomes actually unstable.
We thus have established a simple way to monitor reduction of axial-flexural capacity of beams
by monitoring its flexural natural frequency. There is one major hurdle that remains. Beams flexural
frequencies are much higher than global system natural frequencies. To resolve such a hurdle, we
need to explore the scales of both global and local frequencies.
1.00
0.90
Dimensionless first flexural frequency

0.80
0.70
0.60
0.50
0.40
0.30
0.20
0.10
0.00
0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00
Dimensionless axial load
FIGURE 8.22 Axial load effects on beam flexural frequency.
Global Truss Frequency Scale: For the truss in Figure 8.20, we can estimate a simple expres-
sion for its first natural frequency in the following steps. First, we assume that all truss members
have equal cross-sectional area. An approximate global moment of inertia of the whole truss is
A 2
I eq ⊕ (8.82)
2
Where A is the cross-sectional area of the truss member. If ρ is the mass density of the truss mate-
rial, the total mass of the truss is
(
M = rA 2 n + ( n + 1) + n 2 ) (8.83)
Assuming that n ≫ 1 we get

M 4.14 n r (8.84)
The vertical stiffness of the truss can be approximated by
24 EI eq
k (8.85)
Given that L3
L = n (8.86)
And using Equation 8.79 we get the first natural frequency of the truss as
1 24 E I eq 1
fG = (8.87)
2p L 3
a G ( 4.14 nrA )
The equivalent mass in Equation 8.87 is α GM.

Equation 8.87 can be simplified to
0.46 E I eq
fG = 2 (8.88)
(a G L ) meq
With
meq = rA (8.89)
Substituting Equations 8.82 into 8.88 we get
0.46 1 E
fG = (8.90)
a G2 2 n
2
r
Taking α G = 0.5 (Biggs 1964) the simplified global truss first natural frequency is
1.3 E
fG = (8.91)
n2 r
Local Member Natural Frequency Scale: The diagonal member natural frequency scale can be
estimated using similar steps as before. The local stiffness kLocal is
24 EI
kLocal = (8.92)
( )
3
2
The governing mass of the diagonal is
mLocal = a G 2 rA (8.93)
The first local natural frequency of the diagonal can then be estimated as
1 24 EI 1
fLocal = (8.94)
2p aG 2 rA
( 2 )
3
Simplifying
1 24 EI 1
fLocal = (8.95)
2p aG 2 rA
( 2 )
3
Resolving Monitoring Scaling Hurdle: From Equations 8.91 and 8.95 the relative local to global
frequencies ratio is
1.3 2 r n 2
fLG = (8.96)
p
Finally
0.58 r n 2
fLG = (8.97)

To get a practical sense for the range of f LG, let us explore a steel truss with n = 20, r = 20 in, and
height l = 15 ft; the f LG ratio is about 26. This means that the first flexural frequency of the compres-
sion member is about 26 time the first global truss frequency. This large scale necessitates special
attention during an SHM project to monitor f_local. This includes the following:
• Ensure that the sampling rate during SHM measurements is high enough to measure
f_local accurately
• Utilize adequate signal processing schemes to filter away lower global frequencies
• Validate measurements using adequate structural identification methods
The final step in this project is to assign a threshold alarm level at which mitigation effort might be
needed. Such a threshold might be an upper limit to P_bar, say 0.75. Given the high uncertainty of
most of the parameters in this problem, and the high consequences of failure, it might be advisable
to treat P_bar as a random variable, instead of a deterministic variable. It should also be noted that
this may not be practical for most structures due to the nature of complexity involved, cost, and
limitations of the available monitoring systems.
8.5.4 Integrated Capacity—Demand Monitoring: Virtual Sensing

Concept of VS (see Chapter 5 of Ettouney and Alampalli 2012) can be sued to monitor directly the
probability of capacity-demand ratio, that is, probability of failure. An obvious application is the moni-
toring of strains at any location. If there is a preset failure strain level, then by estimating the probability
of situations when the measured strain will exceed such a value, we obtain the need failure probability.
Since measured strains represent demand, and preset failure strains represent capacity, what we have is
a VS technique to monitor probability of failure. We illustrate this concept here using a reported experi-
ment by Chakraborty and DeWolf (2006). We use the monitored location of the neutral axis of compos-
ite bridge girder to estimate probability of failure of the cross-section. The approach is as follows:
The SHM project by Chakraborty and DeWolf (2006) was installed on a multi-span bridge with
steel girders and reinforced concrete deck. The steel girders and concrete deck were designed to
perform in composite action. The bridge elevation, plan, and cross-section are shown in Figures
8.23, 8.24, and 8.25, respectively. The SHM system included 20 uniaxial strain gauges as shown
in Figure 8.24. The strain gauges were arranged in pairs. The gauges in each pair were placed 2 in
below the top flange and 2 in above the bottom flange at the same vertical cross-sectional plane,
respectively. Eight of the pairs were placed at the center of the eight girders at the longer span. The
other two pairs were placed at quarter of the span on two of the eight girders.
The SHM project measured strains at the 20 locations. The measured strains were then used to
study different aspects of bridge behavior as it responded to normal traffic on the bridge over extended
period of time. Among those behavioral aspects were (a) how the neutral axes of the composite cross-
section vary during a typical weekday of traffic, (b) distribution of traffic load to the eight girders,
(c) comparison of measured and AASHTO design requirements, (d) validation of analytical tech-
niques (finite elements method) and measured results, and (e) effects of slab continuity. The results
of that SHM project provided great insights to the behavior of bridges of the type they considered.
For example, the measured neutral axes during typical weekday showed that the neutral axis actually
changes its position during a typical weekday of traffic. The authors produced a histogram showing
the relationship between N and X, where X is the location of the neutral axes and N is the number of
Direction of travel
258'-8''
Span 3 Span 2 Span 1
63'-7.5'' 76'-3'' 75'-7.5''
FIGURE 8.23 Elevation view of bridge. (From Chakraborty, S. and DeWolf, J. J. Bridge Eng., ASCE, 11(6),
753–762, 2006. With permission from ASCE.)
Direction of travel
Span 3 Span 2 Span 1
63' 75' 75'
5' Midspan
12 G1
Shoulder
56 34 G2
7 @ 7'–4'' = 51'–4''
Lane 1 78 G3
G4
11 12 9 10
Lane 2 13 14 G5
15 16 G6
Lane 3 G7
17 18
19 20 G8
Diaphragm
10°37'10''
All gages strain gages
Odd gages: top
Even gages: bottom
FIGURE 8.24 Plan view of bridge. (From Chakraborty, S. and DeWolf, J., J. Bridge Eng., ASCE, 11(6),
10'' 1'–3''
25'–8''
4'–5''
C
L Roadway
1% 7.75''
FIGURE 8.25 Typical cross-section. (From Chakraborty, S. and DeWolf, J., J. Bridge Eng., ASCE, 11(6),
6000
5000
4000
Frequency
3000
2000
1000
0
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28
Position of NA w.r.t CL of girder
FIGURE 8.26 Histogram of neutral axis during a typical weekday of traffic. (From Chakraborty, S. and
DeWolf, J., J. Bridge Eng., ASCE, 11(6), 753–762, 2006. With permission from ASCE.)
occurrences of X per typical week day. Figure 8.26 shows an illustration of a typical histogram. The
authors observed that the histogram is narrowly spread, and it is centered about the design value of the
composite girder (X = 13.5 in). In addition, the measured location of the neutral axis does not extend to
the tensile (bottom) part of the concrete slab (Figure 8.25). Because of the above, they concluded that
the SHM experiment showed that the composite behavior of the subject concrete bridge is adequate.
Since composite behavior of steel and reinforced concrete deck is one of the most basic design
assumptions in modern bridges, we would like to discuss the SHM experiment, and try to formal-
ize it to gain the maximum value from such an experiment. First, we observe that the fact that the
histogram of Figure 8.26 has a spread, it would indicate a nonlinear behavior of cross-section, which
is not particularly surprising: after all, the composite steel-concrete design is a nonlinear condition
that depends on the level of loading. This nonlinearity makes the knowledge of the histograms of
the neutral axis during an SHM experiment extremely valuable. To study such a behavior further
we need to look at the SHM-produced histogram in a structural reliably sense. To formalize the
process, note that in the following discussion, we will be using hypothetical mathematical formula-
tions, which have no relationship with the above-mentioned SHM project.
Let us assume that a PDF of the location of the composite neutral axis for a particular bridge is
generated from a measured histogram during an SHM project. See Section 8.3 on PDF and histo-
grams for more details. Keeping the reasonable assumption that the composite section deform in a
linear fashion, as displayed in Figure 8.27, we can, after a bit of reflection, observe that as the load
on the composite system increases, the location of the neutral axis X decreases and as the load on
the composite section decreases, the value of X increases. There is actually an inverse nonlinear
relationship between the load level on the composite section and X.
Proceeding further, let us assume that the governing design condition of the composite section
results in x = xDESIGN. Let us associate such a design condition with a load level of L DESIGN, using
Figure 8.17, we can compute the probability that the actual load, in a typical weekday, would exceed
the design load condition P(L ≥ L DESIGN) to be
P ( L ≥ LDESIGN ) = A1 (8.98)
Where A1 is the area under the PDF to the left of x = xDESIGN.

The question now is how to relate the design loading condition L DESIGN to the failure loading con-
dition LFAILUR. Obviously, this is a case-dependent situation; so, let us relate the two values as
LFAILURE = a LDESIGN (8.99)
Locations of neutral axis

Reinforced concrete deck
– X0 – X –
X
Steel
girder + + +
For loads that Design strain For loads that

Composite section are less than the condition are higher than the
design load, Neutral axis at X0 design load,
the neutral axis the neutral axis
move downward move upward
(X > X0) (X > X0)
Thus, the condition X > X0 indicates that

the corresponding loads are higher than the
design condition of the composite section
FIGURE 8.27 Behavior of neutral axis in a composite section.

With
a ≥ 1.0 (8.100)
Upon defining an adequate α, we can define the probability of failure (reliability of the composite
system) as
Pf = P ( L ≥ LFAILURE ) = A2 (8.101)
Where A2 is the area under the PDF to the left of x = xFAILURE. x = xFAILURE can easily be computed
using the concepts of Figure 8.17 and basic composite design methods.
The above example shows how to compute reliability of composite systems using the result of
simple SHM project that is based only on strain measurements. In cases of continuous or intermit-
tent monitoring the official can detect the actual probability of failure of the system in real, or near
real, time.
We end this example by noting that it is based on assuming full composite behavior between steel
girder and reinforced concrete deck. Other SHM or NDT techniques can be utilized to validate such
an important assumption, for example, by measuring relative strains between steel and concrete. If
the full composite action assumption is not accurate, modifications of the equations of this section
are needed.
8.5.5 Separate Capacity—Demand Monitoring

8.5.5.1 General
There are several situations where the structural reliability is computed by separate considerations
of capacity and demands. This is accomplished as follows:
1. Compute statistical properties of demands (e.g., truck weights and frequency) as measured
by the SHM system
2. Compute statistics of capacity (e.g., strain levels)
3. Compute reliability index using appropriate method (e.g., Taylor series or Monte Carlo
simulation)
Two popular applications for this technique of reliability computations are proof load testing and
load and resistance factor design (LRFD)/load and resistance factor rating (LRFR) fields.
8.5.5.2 Proof Load Testing

Proof load tests essentially apply loads to the structure gradually until the load reaches a prespeci-
fied level, or the structure displays signs of distress. This type of load testing is valuable since it
gives information about the capacity of the structure to meet high loading demands. Careful analy-
sis of both capacity and demand of the structure must be performed before, during, and after the
proof load test to ensure successful testing. These analyses include both numerical and monitoring
efforts.
When designing a proof load test, the professional must consider carefully the level of loading
to be used in the test. This is demonstrated in Figure 8.28 (from Melchers 2002). Low-level proof
loads (low-level q∗/L) will entail high uncertainty (as reflected in the higher testing variances VR)
in the probability of failure that are detected by the test. On the other hand, as the proof load level
increases, the failure load probability (P(H > 0)) increases.
It should be noted that there are more limitations to proof load tests:
• There is a possibility of failure if the loads exceed the capacity of the structure.
• There might be some permanent damage due to exceedance of elastic limit states.
Probability of failure
VR = 0.18
P(H < 0)
VR = 0.18
VR = 0.13 Proof load applied

at year
VR = 0.13 25 10
10
25
Normalized proof load q */L
FIGURE 8.28 Reliability of proof test. (From Melcher, R. Structural Reliability Analysis and Prediction.
2002. Copyright Wiley-VCH Verlag GmbH & Co. KGaA. Reproduced with permission.)
• The loading pattern might not excite an important limit state, thus the test will not show
the limiting capacity of that state. For example, gravity loading of a bridge will not be
capable of exploring limit states of wind, or seismic loads.
Of course, proof load testing is still important and useful since it gives insight to behavior of struc-
tures under increasingly demanding loading conditions.
8.5.5.3 LRFD and LRFR

LRFD: LRFD method is a popular design method for structures; it is used for designing build-
ings, bridges, and all other types of structures, see Ravindra and Galambos (1978) and Barker and
Puckett (2007). The LRFD method require for design metrics (such as beams, columns, connec-
tions, etc.):
Ci ≥ Di (8.102)
Ci = fi Rin (8.103)
j=N
Di = ∑ gij Lijn (8.104)
j =1
The capacity and demands of the ith design metric is Ci and Di, respectively. The nominal resistance of
the design metric is Rin. The nominal jth load effect on the design metric is Lijn. There are N load effects
on the design metric under consideration. The resistance factor ϕi and the load factors γij are factors
that are obtained through probabilistic analysis, and extensive testing and monitoring. In important or
sensitive structures, those factors might be improved by performing case-specific SHM project.
LRFR: LRFR method is used in bridges to estimate their adequacy to resist required dead and
live loads. The LRFR method is also based on evaluating factors through probabilistic analysis, and
extensive testing and monitoring, see Moses (2001) and Murdock (2009). SHM projects can also
improve the accuracy of the generic LRFR factors. See Chapter 8 of Ettouney and Alampalli (2012)
for more on SHM use to improve LRFR predictions.
8.6 RISK
8.6.1 Overview
There are several definitions of risk as applied to civil infrastructure, see FEMA (2009), Aven
(2003), and Koller (2000). For our purpose, we define the risk as a measure of severity of the out-
come of uncertain events when applied to systems with a degree of uncertainties. This definition
accommodates all recognized important parameters in traditional risk theories:
• Uncertainties of events/hazards
• Uncertainties within the system itself (vulnerabilities)
• The outcome (consequences)
We would like to differentiate between the concept of risk and reliabilities. Reliability concerns
itself with the interrelationship between capacity and demand of systems. In other words, it con-
cerns itself with hazards and vulnerabilities. The only outcome reliability is concerned with is the
system behavior, for example, whether the system has failed, or damaged. Risk on the other hand,
concerns itself with consequences of all types: it could be system failure, or consequences beyond
system failure such as economic, social, or political consequences.
There are several quantitative ways to compute risk: linear programming, decision tree, and Monte
Carlo techniques (see Table 8.12). Also, there are several qualitative methods for computing risk.
In this section we explore risk management and its underlying different type of risk. The role of
SHM in each of those types of risk is discussed briefly. The following subsections include discussions
of relative versus absolute risk, deterministic versus probabilistic risk, and the general approaches
of evaluating different risk components. Four specific examples of the utilization of SHM in risk
management are presented. Each example is fairly different, but the four examples as a whole show
the rich and varied ways of interaction of risk techniques and SHM technologies: such interaction
would benefit decision makers in arriving to accurate, safe, and cost-effective decisions.
8.6.2 Components of Risk Management

For the purpose of this document we define risk management as the main risk module that encom-
passes several types of risk. Thus, risk manager needs to handle several types of risk (see Figure
8.29). In addition, the role and utilization of SHM vary depending on the type of risk. In what fol-
lows, we discuss briefly those different types of risk and how SHM is related to each.
8.6.2.1 Risk Acceptance
The first step in risk management is to establish the presence of a risky situation. For our purposes,
we use a generalized definition of risky situation as any situation that would affect the health and
TABLE 8.12
Techniques for Computing Risk
Method Advantage Disadvantage
Linear programming, not covered Formal Mostly Deterministic some probabilistic applications,
in this manuscript, see Trueman Elsayed and Ettouney (1994). Difficult to relate
(1977) for more details linkages and dependencies between variables
Decision tree/fault tree Visual, some linkages Can be limiting for complex cases. Probabilistic
between variables decision tree models are not as widely used as
deterministic decision tree models
Monte Carlo General and flexible. Can be computationally demanding. Difficult to “feel”
Easy to program in the physical meaning of the results
computer codes
Risk acceptance
Risk treatment
Risk
management
Risk assessment
Risk communications
Risk management and

different risk types
FIGURE 8.29 Risk management ant its components.
well-being of the infrastructure under consideration. Following such a general definition, risky situ-
ations include the smallest of effects such as painting boils and catastrophic effects as general bridge
collapse. Identifying the presence of risk and accepting it can be done qualitatively using the experi-
ence of bridge officials, or quantitatively using different components of SHM.
8.6.2.2 Risk Assessment
After the decision of accepting a particular risk, assessment of that risk follows. Risk assessment
results can be used in a decision making process in one of the three forms:
• Direct utilization
• Cost–benefit process
• Life cycle analysis (LCA) process
Risk values can be used directly in several manners, as follows:
1. Studying the components of risk (threats, vulnerabilities, and consequences) can give an
insight to causes of higher risk issues. The decision maker can then address these issues to
reduce overall risk.
2. Prioritize different projects based on risk assessment results. For example, let us assume
that there are three available projects that can reduce risk of a particular situation from
R0 to R1, R2, and R3. If R1 is the least of the three, then the first project is the preferred
project since it produces the least risk.
8.6.2.3 Risk Treatment
Soon after a type of risk assessment is performed, risk treatment phase starts. Risk treatment (some-
time also referred to as risk response) shows how the risk assessment results are handled. There are
several types of risk treatment such as
• Accept Risk: the identified risk is within the previously identified acceptable level. Levels
of acceptable risk are usually identified via cost–benefit analysis, LCA, or similar type of
quantifiable decision making techniques.
Many of the methods of assessing risk are explored later in this section.
• Apply Appropriate Risk Reduction Solutions: Risk reduction or mitigation solutions can
be accomplished by reducing vulnerabilities, minimizing threats, reducing consequences,
or a combination of some of each. For example, a painting project would mitigate risk of
corrosion hazard.
• Risk Avoidance: In some situations it is possible to avoid risks, but only if the actions that
may cause the risks can be controlled. In most situations where the risk is due to a natural
hazard, such as an earthquake, risk avoidance is not possible.
• Risk Transfer: Risk can be transferred to other organizations in some situations. The
typical risk transfer is an insurance against the risk. Since insurance involves monetary
compensation for losses, it is not always an appropriate way of risk treatment, especially if
losses may involve casualties.
8.6.2.4 Risk Communication

Risk communications establishes interchange between civil infrastructure authorities (owners,
managers, professionals) and the users of those infrastructure. Such an exchange of information
should be two ways and continuous. It also should include as many stakeholders from the two sides
as possible. The goals of risk communications are (a) Convey that no harm is going to happen. This
will ease public concerns, (b) Give some guidance on how the public should respond to the situation,
and (c) Reduce anxieties regarding the situation.
Risk communication occurs during one of three phases that relate to hazardous events (severe
floods, earthquakes, etc.) that might affect the infrastructure:
• Preevent Risk Communications: involves communicating risk assessment results, the

values of any needed mitigation solutions, training activities, and so on.
• During Event Risk Communications: On-going risks need to be communicated to the
public efficiently and in a timely fashion. Examples are prolonged hurricane events.
• After Event Risk Communications: Communicating information on the causes of events
and future mitigating solutions is essential.
Other reasons for the paramount importance of risk communications are due to the fact that it helps
the public to know and understand the relevant facts that relate to risks (before, during, and after
events). This way, the public can take appropriate action, such as orderly evacuations during a severe
wind storm. Note that risk communication involves more than just relaying information. Risk com-
munication should also accommodate personal factors and emotions such as unreasonable fears,
trust in authorities, and uncertainties. Generally speaking, there are three modes of risk communi-
cations, namely (a) written (e.g., newspaper, bulletins, flyers), (b) verbal (personal communications),
and (c) visual (TV messages).
8.6.3 Other Forms of Risk

8.6.3.1 Relative versus Absolute Risk
Risk can also be classified into two types: relative and absolute. Relative risk is generic, requires
smaller amounts of input data and forecasting and is easier to formulate. Absolute risk is more spe-
cific, requires more input data and can be more difficult to formulate. The two types of risks can be
classified, arguably, as qualitative and quantitative for relative and absolute risks, respectively.
8.6.3.2 Deterministic versus Probabilistic Risk

At first glance, the title of this section seems to be odd. The reader might ask: isn’t risk based on
uncertainties to begin with? If so, then how can it be deterministic? In a way, the question is valid.
However, we need to step back and try to observe the methods of infrastructure that are supposed to
be deterministic. Recall that all inputs in deterministic methods are based on best guesses, or aver-
ages. In additions, many parameters in deterministic methods, such as load factors in LRFD design
approaches, are based on probabilistic evaluations. On the basis of this, we identify deterministic
risk method as a method that would produce a single valued result: the expected risk value. We
identify the probabilistic risk method as the method that produces expectations (at lest two lowest
expectations) or a complete histogram of the risk as a random event.
8.6.4 Evaluation of Risk and Its Components

8.6.4.1 General Risk Expression
We use in this document a general form of risk, R as
R = ∫ HVC (8.105)
In Equation 8.105, H, V, and C are the hazard, vulnerability, and consequence. The integral in
Equation 8.105 is over the hazard space. A discrete form of Equation 8.105 is
R = ∑H VC (8.106)
Again, the sum in Equation 8.106 is over all possible hazard levels. A simple form of Equation 8.105 is
R = H VC (8.107)
This simple form presumes that the hazard is single valued. Note that both Equations 8.105
and 8.107 imply a dependence between the hazard level, or levels, and both vulnerabilities and
consequences.
Structural health monitoring can help in estimating different risk components. We explore dif-
ferent risk components-SHM interrelationships next.
8.6.4.2 Hazards
Different methods of risk evaluation process different hazards differently. The use of SHM to aid in
evaluating hazards, or the way SHM handles hazards in a risk-related activity are interrelated. Table
8.14 shows these different interrelationships.
8.6.4.3 Vulnerability
Measures of vulnerability, as used by risk professionals, vary depending on the type of risk analysis
being performed. When used in relationship with SHM, the SHM project is also affected, as shown
in Table 8.15.
8.6.4.4 Cost—Consequences
Evaluating cost or consequences is perhaps the most difficult component in evaluating risk. This is
simply because it is not well defined. For example:
• Is consequence just the direct cost of damage retrofit?

• Is it cost of damage, or failure?
• Does it include management costs?
• Does it include loss of business, or business interruption?
• Does it include other indirect costs?
• Does it include social or political costs?
A simple cost of bridge failure expression CFailure, measured in $, was developed by Stein and
Sedmera (2006) as
T D A d ⎧ ⎡ ⎛ 100 ⎞ ⎤ 1⎡ ⎛ 100 ⎞ ⎤⎫
CFailure = C1eWL + ⎨ ⎢C2 ⎜ − 1⎟ + C3 ⎥ + ⎢C4O ⎜ − 1⎟ + C5 ⎥ ⎬ + C6 X (8.108)
100 ⎩ ⎣ ⎝ T ⎠ ⎦ S⎣ ⎝ T ⎠ ⎦⎭
Where
C1 = Unit rebuilding costs (Table 8.52) unit is in $
e = Cost multiplier (Table 8.53) unit is in $
W = Bridge width, can be obtained from National Bridge Inventory (NBI), field 52 (unit is in ft.)
L = Bridge length, can be obtained from NBI, field 49 (unit is in ft.)
C2 = Cost of running cars (Table 8.54)
C3 = Cost of running trucks (Table 8.54)
D = Detour length, can be obtained from NBI, field 19 (unit is in Miles)
A = Average daily traffic on bridge (ADT), can be obtained from NBI, field 29
d = Duration of detour (Table 8.55) unit is in days
C4 = Value of time per passenger (Table 8.56)
O = Average occupancy rate per car (Table 8.57)
T = Average daily truck traffic on bridge (ADTT), can be obtained from NBI, field 109 (note
that it is a % of ADT)
C5 = Value of time per truck (Table 8.58 or Table 8.59)
S = Average detour speed, units are in mph
X = Cost per life lost
C6 = Number of deaths from failure (Table 8.60)
TABLE 8.13
Typical Values for Some
Parameters of Equation 8.108
Parameter Typical Value Units
O 1.63 Scalar
S 40 mph
X 500,000 $
TABLE 8.14
Hazard Estimation and SHM Methods
Hazard Quantitative (Absolute Risk) Qualitative (Relative Risk)
Magnitude Measure earthquake motions, flood water velocities and pressures, traffic loads, etc.
Recurrence Example: traffic patterns and frequency Use available records, e.g., flood maps,
seismic maps, etc.
Large-scale: earthquakes Measure seismic motions at different regions. Also, estimate effects of local soil conditions
Small-scale: corrosion Measure actual loss of area, or rusting volume, Measure corrosion rates, see Chapters 3
see Chapters 3 and 4 in Ettouney and and 4 in Ettouney and Alampalli (2012)
Alampalli (2012)
Small scale: fatigue Measure fatigue cracks using different NDT Estimate remaining fatigue life. See
methods (e.g., ultrasound or acoustic Chapter 5 in Ettouney and Alampalli
emission) (2012)
TABLE 8.15
Vulnerability and SHM Methods
Hazard Sensing STRID DMID
Earthquakes Use accelerometers or strain Before event: use modal After event: use different NDT
sensors for modal identification identification methods, methods to estimate earthquake
and nonlinear behavior Chapter 6 damage
Scour See Chapter 1 of Ettouney and Alampalli (2012)
Fatigue See Chapter 5 of Ettouney and Alampalli (2012)
Gravity loads Use vibration and strain sensors Use parameter or modal Use different NDT methods to
to estimate responses to gravity identification to estimate estimate deterioration of bridge
loads structural properties during components
exposure to gravity loads
Corrosion See Chapters 3 and 4 of Ettouney and Alampalli (2012)
Typical values of some of the parameters of Equation 8.108 are offered by Stein and Sedmera (2006)
as shown in Table 8.13.
Equation 8.108 contains costs of reconstruction (first term), detour (second term), and potential
fatalities, (third term). We note that the equation is hazard-independent, which makes it fairly suit-
able for most risk-based computations. We note that the equation do not include some additional
costs of bridge failures such as potential business interruptions, permanent business closures or
social/political effects. These additional effects are much more difficult to estimate; they are case
dependent. Because of this, Equation 8.108 can be considered as a lower estimate limit to the cost
of bridge failure CFailure. Ettouney and Alampalli (2012) utilized Equation 8.108 to compute the
statistical properties of bridge failure. In Chapter 10 of Ettouney and Alampalli (2012) a slightly
modified version of Equation 8.108 is used to estimate costs of hazard occurrences.
Upon closer inspection of Equation 8.108, it becomes obvious that most of its independent vari-
ables are actually estimates of averages of random variables. This indicates that CFailure itself is a
random variable, and Equation 8.108 is an estimate of an average for the cost of bridge failure. In
different decision making situations, the uncertain nature of CFailure needs to be considered. Given
the complex relationships of Equation 8.108, this can be accomplished in one of two ways: either
using a Monte Carlo simulation technique, or use the Taylor series technique. Both techniques are
explained elsewhere in this chapter.
8.6.5 Relative Risk: Prioritization

8.6.5.1 General
Relative risk methods utilize the form of Equations 8.105 through 8.107. Definitions of risk com-
ponents are mostly qualitative that are based on the experience of the professional. This is where
SHM can help to a great deal in enhancing the accuracy of any/all of the three components of risk.
There are two popular relative risk methods. They are FEMA risk method, and the FMEA Analysis
method. The FEMA approach, FEMA (2009) is discussed in Chapter 11 by Ettouney and Alampalli
(2012). The FMEA method is discussed next as it applied to bridge scour mitigation risk analysis.
8.6.5.2 Failure Mode and Effect Analysis

The FMEA method produces a qualitative measure of risk. Let us assume that for a particular haz-
ard condition N is the number of possible retrofit options for the system on hand. We observe that for
each of the possible retrofit measure there exists Mi set of possible failure modes. Let us define, for
each of the ith retrofit measure (i = 1,2, …, N) and jth failure mode (j = 1,2, …, Mi) a consequence of
failure rating RCij, a likelihood of failure occurrence R Lij and an ease (difficulty) of failure detection
R Dij. We also define the risk priority number RPNij as
RPN i j = RC i j RL i j RD i j (8.109)
0 ʺ RC i j ʺ 10 (8.110)
0 ʺ RL i j ʺ 10 (8.111
0 ʺ RD i j ʺ 10 (8.112)
There are several ways to continue on the risk ranking. For example, a simple method is to compare
qualitatively all RPNi j for a given i (retrofit method). The most preferred ith method is then chosen
qualitatively. This approach was used in scour countermeasure risk rating by Johnson and Niezgoda
(2003). A perhaps less qualitative approach can be to define a risk priority number for each retrofit
measure, RPNMi such that
j = Mi

∑ j =1
RPN i j
(8.113)
RPNMi =
Mi
A final ranking of the set RPNMi can then be easily performed analytically. The retrofit measure
with highest risk is Max(RPNMi) and the retrofit measure with lowest risk is Min(RPNMi).
The elegance of the FMEA method is that it is simple, yet it covers many parameters. It is also
expandable, since additional retrofit measures, or failure modes can be easily added. On the nega-
tive side, the method does not account for cost of retrofit measures, nor does it account for other
objective parameters such as ease of installation, time constraints, or experience of contractors. In
addition, the method is deterministic in nature, thus the uncertainties in the rankings, and their pos-
sible effect on the decision maker, are not evaluated.
Perhaps the most limiting factor in the FMEA method as presented is that it does not recognize
the interrelationship between the consequence of failure and the likelihood of failure. It is reason-
able to assume that low-level hazard is more likely to occur than high-level hazard, yet it would
produce a damage level that is fairly small. The opposite is true for high-level hazard. For example,
a magnitude 5 earthquake would produce damage levels in a given structure that are much lower
than a magnitude 6 earthquake. Yet, in a given site, the likelihood of a magnitude 5 earthquake is
much higher than magnitude 6 earthquake.
One possible way to handle this limitation is to generalize Equation 8.109 from the form of
Equation 8.107 to the form of Equation 8.106.
8.6.6 Relative Risk: Matrix Analysis

Sometimes it is desired to provide a consistent risk analysis template for bridges that have some
generic common feature. Unfortunately, those generically similar bridges (or any other form of struc-
tures) have important variability, such as the nature of topology, soil condition, and usage patterns.
If the decision maker follows a simple risk method such as the prioritization model of the previous
sect, a separate model for each of the generically similar bridges must be built. This will result in
an inconsistent modeling, and complexity on the bridge network level. Koller (2000) presented an
elegant solution for this situation: the matrix approach, we give an SHM illustration for the use of
this approach.
8.6.6.2 Problem Description

Consider the case of a typical simple span post-tensioned reinforced concrete bridge girder. Those
types of bridges are fairly common throughout the United States. Let us assume that the decision
maker decided that a consistent maintenance prioritization scheme for repairing/replacing post-
tensioned tendons is need, given the limited budget that is available for this type of bridges within
the network. The decision maker then estimates that there are four generic parameters that affect the
repair prioritization of a given bridge: (a) age of construction, (b) traffic patterns, (c) skew angle of
the bridge, and (d) repair history. The local factors that can affect each of the generic parameters are
(a) seismic zones, (b) substructure conditions, (c) visual inspection, (d) temperature, and (e) humid-
ity. The matrix of generic and local parameters is shown in Table 8.16. We should emphasize here
that the generic and local parameters of this example are hypothetical, and not based on an actual
studies: they are used here for demonstration purposes only.
TABLE 8.16
Matrix of Generic and Local Parameters
Parameters Seismic Effects Substructures Visual Inspection Temperature Humidity
Age of Age of construction Older bridges Older bridges are in Cumulative Cumulative
construction would affect would be in a more deteriorated effects of effects of
seismic behavior. more state than newer cycles of humidity
This is due to the deteriorated bridges thermal variations
changes of condition expansion and would have
construction than newer freeze and thaw more
material, ones would have effects on
construction more effects on older than
details, and the older than newer
structural system newer bridges bridges
of the bridge
Traffic Traffic patterns Designs of Visual inspection Temperature or humidity levels
patterns have minimal substructures would investigate have minimal interaction with
interaction with are directly effects of traffic traffic patterns
seismic behavior related to patterns on different
of a bridge traffic bridge components,
patterns e.g., collision and/or
deck deterioration
patterns
Skew angle Skew angle can Behavior of Components with skew Repeated cycles of temperature
have an effect on connections features, such as may be important
seismic and other connections, could
performance structural experience more signs
details are of deterioration
affected by
angle of
skew
Repair Seismic events can Repair history Visual inspection Temperature (e.g., generating
history affect repair has direct results can lead to cracks) or humidity (e.g.,
activities. effects on repair activity generating corrosion) effects can
Conversely, substructure lead to repair activities
repairing bridges behavior
can affect seismic
performance
In addition to the above matrix, a relationship between damage level and the scores of impor-
tance of generic parameters is needed. This is another situation where an SHM database for this
class of bridges can be used to generate the required relationship. Table 8.16 shows an example of
the desired relationship. The table shows the parameter score range and the corresponding damage
level. It should be noted that Table 8.17 can also be generated based on the experience of the analyst;
however, an SHM database is preferable, since it is more accurate.
The matrix in Table 8.16 can now be filled on a qualitative manner for each bridge site by a scale
from 0 to 10. An entry of 0 indicated that the local parameter do not have any impact on the generic
parameter at the same row. A score of 10 indicates a maximum effect of the local condition on the
generic parameter. Let us assume that the matrix entries are xij. Where i and j are the rows and col-
umns of Table 8.16, respectively. Thus 1 ≥ xij ≥ 0.
The values of xij can be filled by one of three methods
1. Qualitatively: Where each of the bridges of the type under considerations is evaluated by a
professional who is familiar with the generic and local conditions of the bridge.
2. Analytically: If there is an analytical relationship between the local and generic parameters,
it can be estimated. For example, the importance of the seismic effects and the skew angle of
the bridge might be possible to estimate analytically for each bridge under consideration.
3. Experimentally. If there is an SHM activity for a particular bridge, such as records of tem-
perature, humidity, or traffic patterns, they can be used to estimate appropriate xij.
Of the three methods, it is clear that the use of SHM data would generate the most accurate xij.
Next, the decision maker will build a histogram of each of the generic parameters using two
sources
1. A general histogram that relate the importance of the generic parameter Yi, to the state of
deterioration of the tendons D.
2. A relationship between Yi and xij.
Formally
D = f (Yi ) (8.114)
And
( )
Yi = g xij (8.115)
Again, the histograms of Equation 8.114 can be based on SHM data collection, or in lieu of such
data, they can be estimated by the official based on “common sense.” Similarly, the functional
TABLE 8.17
Score Range of Generic Parameters and
Damage Levels
Damage Level (%)
Parameters 00–20 20–40 40–99
Age of construction 0.00–0.40 0.40–0.80 0.80–1.00
Traffic patterns 0.00–0.70 0.70–0.95 0.95–1.00
Skew angle 0.00–0.60 0.60–0.90 0.90–1.00
Repair history 0.00–0.50 0.50–0.70 0.70–1.00
relationships g(xij) can be estimated using experience of the official. In general, in their simplest
form, they can take the form
Yi = ∑ w j xij (8.116)
j
Where wj is an adequate weighting factor.

Tables 8.18 and 8.19 show sample values of xij and wj for a particular bridge, respectively. Note
that xij is bridge specific, while wj is generic for the type of bridges under considerations.
The resulting damage score for each of the generic parameters are shown in Table 8.19.
The histograms D = f(Yi) are generated next. These histograms are the results of several SHM
projects that resulted in database for this type of bridges. Tables 8.20, 8.21, 8.22, and 8.23, respec-
tively show sample frequencies of different generic parameters as they are related to degrees of
damage of tendons. From the histograms, the cumulative frequency of tendon damage level for the
generic parameters. These results are shown in Tables 8.24, 8.25, 8.26, and 8.27, respectively. The
histograms and their cumulative distributions are also shown in Figures 8.30, 8.31, 8.32, and 8.33,
respectively. The histograms and the cumulative distributions can be used for different decisions
regarding probabilities of damage as resulted from different sources (age of construction, traffic
patterns, skew angle and repair history in the current example).
TABLE 8.18
Values of Xij for the Matrix of Generic and Local Parameters
Seismic Visual
Parameters Effects Substructures Inspection Temperature Humidity
Age of construction 1 5 9 8 8
Traffic patterns 0 4 3 0 0
Skew angle 1 7 0 9 5
Repair history 0 3 7 7 7
TABLE 8.19
Values of wj for the Matrix of Generic and Local Parameters
Seismic Visual
Parameters Effects Substructures Inspection Temperature Humidity
wj 7 4 4 8 8
TABLE 8.20
Histograms Relating Bridge Age and Damage
Levels
Damage Level
Age of Bridge
(Years) 00%–20% 20%–40% 40%–99%
00–10 0.13 0.02 0.01
10–20 0.45 0.16 0.03
20–30 0.2 0.41 0.2
30–40 0.11 0.2 0.53
40–50 0.08 0.12 0.15
50–99 0.03 0.09 0.08
Total 1 1 1
TABLE 8.21
Histograms Relating Traffic Patterns and Damage Levels
Damage Level
Traffic Pattern: Number
of Trucks per Day 00%–20% 20%–40% 40%–99%
00–50 0.003 0.001 0
50–100 0.007 0.009 0.02
100–1000 0.02 0.06 0.11
1000– 0.97 0.93 0.87
Total 1 1 1
TABLE 8.22
Histograms Relating Skew Angle and Damage
Levels
Damage Level
Skewness
(degrees) 00%–20% 20%–40% 40%–99%
00–05 0.02 0.05 0.02
05–15 0.3 0.15 0.08
15–25 0.6 0.25 0.15
25– 0.08 0.55 0.75
Total 1 1 1
TABLE 8.23
Histograms Relating Repair History and
Damage Levels
Damage Level
Repair History 00%–20% 20%–40% 40%–99%
Past 2 years 0.03 0.01 0.001
Past 10 years 0.07 0.05 0.03
Past 20 years 0.5 0.55 0.45
Never 0.4 0.39 0.519
Total 1 1 1
TABLE 8.24
Histograms Relating Bridge Age and Cumulative
Damage Levels
Cumulative Damage Level
Age of Bridge
(Years) 00%–20% 20%–40% 40%–99%
00–10 0.13 0.02 0.01
10–20 0.58 0.18 0.04
20–30 0.78 0.59 0.24
30–40 0.89 0.79 0.77
40–50 0.97 0.91 0.92
50–99 1 1 1
TABLE 8.25
Histograms Relating Traffic Patterns and Cumulative
Damage Levels
Traffic Pattern: Number
of Trucks Per Day 00%–20% 20%–40% 40%–99%
00–50 0.003 0.001 0
50–100 0.01 0.01 0.02
100–1000 0.03 0.07 0.13
1000– 1 1 1
TABLE 8.26
Histograms Relating Skew Angle and
Cumulative Damage Levels
Skew Angle
(Degrees) 00%–20% 20%–40% 40%–99%
00–05 0.02 0.05 0.02
05–15 0.32 0.2 0.1
15–25 0.92 0.45 0.25
25– 1 1 1
TABLE 8.27
Histograms Relating Repair History and
Cumulative Damage Levels
Repair History 00%–20% 20%–40% 40%–99%
Past 2 years 0.03 0.01 0.001
Past 10 years 0.1 0.06 0.031
Past 20 years 0.6 0.61 0.481
Never 1 1 1
8.6.7 Absolute Risk: Cost–Benefit

One of the most important findings of a recent workshop on engineering of structural health,
Alampalli and Ettouney (2002) was the need for cost–benefit analysis when embarking on an SHM
project, or when considering any major improvements to a particular bridge. There are several
methods to perform a cost–benefit analysis of an activity. In this section, we present a risk-based
approach that uses decision tree topology. This example will show how to apply a risk method to
the cost–benefit problem. For our immediate purpose, the example will show (a) how to utilize the
results of SHM experiments to produce accurate cost–benefit analysis, and (b) how to compare the
value (or cost–benefit) of SHM projects to the more conventional visual inspection projects in a
qualitative manner.
Effects of age of bridge on damage

1 1
0.9 0.9
0.8 0.8
Cummulative frequency
0.7 0.7
Histogram frequency
Hist. 00%–20%
0.6 0.6
Hist. 20%–40%
0.5 0.5 Hist. 40%–99%
Cumulative 20%-40%
0.4 0.4 Cumulative 40%-99%
Cumulative 00%-20%
0.3 0.3
0.2 0.2
0.1 0.1
0 0
00–10 10–20 20–30 30–40 40–50 50–99
Age (years)
FIGURE 8.30 Frequency and cumulative damage level as related to bridge age.
Effects of traffic pattern of bridge on damage

1 1
0.9 0.9
0.8 0.8
0.7 0.7
Histogram frequency
0.6 Hist. 00%–20%

0.6
Hist. 20%–40%
0.5 0.5 Hist. 40%–99%
Cumulative 20%–40%
0.4 0.4 Cumulative 40%–99%
0.3 0.3
0.2 0.2
0.1 0.1
0 0
00–50 50–100 100–1000 1000–
Number of trucks per day
FIGURE 8.31 Frequency and cumulative damage level as related to traffic patterns.

The problem that a decision maker has is that it is felt that a monitoring/inspection process for a par-
ticular bridge is needed. There are several monitoring/inspection methods that can be followed:
1. Visual/manual inspection
2. Utilize some type of SHM techniques/technologies
3. A combination of the above two methods
The decision maker estimated that after the monitoring/inspection process is finished, there can be
three outcomes:
1. There is need for immediate repair

2. There will be some repair needed within 1–2 years
3. There is no obvious need for repair within 2-year period
Effects of skewness angle of bridge on damage

1 1
0.9 0.9
0.8 0.8
0.7 0.7
Histogram frequency
Hist. 00%–20%
0.6 0.6 Hist. 20%–40%
Hist. 40%–99%
0.5 0.5 Cumulative 20%–40%
0.4 0.4 Cumulative 40%–99%
0.3 0.3
0.2 0.2
0.1 0.1
0 0
00–05 05–15 15–25 25–
Skew angle (degrees)
FIGURE 8.32 Frequency and cumulative damage level as related to skew angle.
1 1
0.9 0.9
0.8 0.8
0.7 0.7
Histogram frequency
Hist. 00%–20%
0.6 0.6 Hist. 20%–40%
0.5 Hist. 40%–99%
0.5
0.4 0.4 Cumulative 40%–99%
0.3 0.3
0.2 0.2
0.1 0.1
0 0
Past 2 years Past 10 years Past 20 years Never
Skew angle (degrees)
FIGURE 8.33 Frequency and cumulative damage level as related to repair history.
What distinguishes this problem from the other types of problems is that for each possibility there
are two basic outcomes:
1. The damage of interest can be discovered by the process

2. The damage can’t be discovered by the process
Figure 8.34 shows the generic decision tree for this problem.
On the basis of previous experiences, the decision maker proceeded to assign different probabili-
ties to different potential outcomes. On the basis of past experiences, the probabilities of damage
occurrences are estimated in Table 8.28. The potential of observing damage from different methods
are estimated in Table 8.29 and the potential monetary consequences of damage observance are
estimated in Table 8.30. The table also shows the direct cost of each monitoring method. Clearly,
Method of inspection Needed repairs Inspection results
Observed
Immediate Not observed
Observed
Visual inspection 1–2 Years
Not observed
None Observed
Not observed
Observed
Observed
SHM only 1–2 Years
Not observed
None Observed
Not observed
Observed
Observed
1–2 Years
Visual inspection Not observed
+ SHM None Observed
Not observed
FIGURE 8.34 Decision tree for the inspection methods.
TABLE 8.28
Probability of Damage Condition
Condition of Damage (s) Frequency (probability) (%)
Needs immediate repair 20
Needs repair within 1–2 years 50
Do not need repair within 2 years 30
TABLE 8.29
Probability of Occurrence of Inspection Results
Inspection Results: Probability of Occurrence
Method Condition of Damage Damage Observed (%) Damage Not Observed (%)
Visual inspection Needs immediate repair 40 60
Needs repair within 1–2 years 50 50
Do not need repair within 2 years 60 40
SHM Needs immediate repair 70 30
Both Needs immediate repair 85 15
TABLE 8.30
Costs of Different Outcomes
Cost (M. $)
Direct Cost of Inspection/Monitoring
Needed Repairs Inspection Results General Consequences Visual Inspection SHM Both
Immediate Damage observed 1.5 0.5 1.0 1.2
Damage not observed 5.5 0.5 1.0 1.2
1–2 Years Damage observed 1.0 0.5 1.0 1.2
None Damage observed 0.0 0.5 1.0 1.2
1. Cost of inspection/monitoring includes direct and indirect costs. Both estimated on a present value basis.
2. General consequences cost includes estimated rehabilitation, social, and long-term costs.
Method of inspection Needed repairs Inspection results Consequences

(Cost: $ millions)
Probabilities
Observed
Probabilities Immeidiate 0.4 2.0
Not observed
0.2 6.0
0.6
Observed
Visual inspection 0.5 1–2 years 0.5 1.5
Not observed
0.5 2.8
0.3 None 0.6 Observed
0.5
0.4 Not observed
0.5
Observed
2.5
0.7
Immidiate Not observed
6.5
0.2 0.3
Observed
SHM only 0.5 1–2 years 0.8 2.0
Not observed
0.2 3.3
0.3 None 0.6 Observed
1.0
0.4 Not observed
1.0
Observed
0.85 2.7
Immidiate Not observed
0.2 6.7
0.15
Observed
1–2 years 0.95 2.2
0.5
Visual inspection Not observed
0.05 3.5
+ SHM 0.3 None 0.9 Observed
1.2
0.1 Not observed
1.2
FIGURE 8.35 Decision tree with assigned probabilities and costs.
the direct costs of visual inspection are the least of the three methods. Application of all these
estimates to the decision tree branches is shown in Figure 8.35. The total cost of each method can
be computed by multiplying each probability along the branch by the total cost of each branch and
summing all branches within each method. The process is shown in Table 8.31. The total expected
costs for the methods are $2.1 million, $2.17 million, and $2.15 million, for visual inspection, SHM
only, and combined methods. Visual inspection method produces the lowest risk while using SHM
method would produce the highest risk, when measured in monetary terms.
TABLE 8.31
Deterministic Risk Evaluation
Needed Probability Inspection Total Method
Method Repairs (%) Results Probability Cost Cost Cost
Visual Inspection Immediate 20.00 Observed 40.00 2.000 0.160 2.10
Immediate 20.00 Not observed 60.00 6.000 0.720
1–2 Years 50.00 Observed 50.00 1.500 0.375
1–2 Years 50.00 Not observed 50.00 2.800 0.700
None 30.00 Observed 60.00 0.500 0.090
None 30.00 Not observed 40.00 0.500 0.060
SHM Only Immediate 20.00 Observed 70.00 2.500 0.350 2.17
1–2 Years 50.00 Observed 80.00 2.000 0.800
None 30.00 Observed 60.00 1.000 0.180
Both Immediate 20.00 Observed 85.00 2.700 0.459 2.15
1–2 Years 50.00 Observed 95.00 2.200 1.045
None 30.00 Observed 90.00 1.200 0.324
TABLE 8.32
Costs of Different Outcomes—High Consequences
Cost (M. $)
Direct Cost of Inspection/Monitoring
Needed
Repairs Inspection Results General Consequences Visual Inspection SHM Both
Immediate Damage observed 1.5 0.5 1.0 1.2
Damage not observed 11.0
1–2 Years Damage observed 1.0
None Damage observed 0.0
1. Cost of inspection/monitoring includes direct and indirect costs. Both estimated on a present value basis.
2. General consequences cost includes estimated rehabilitation, social, and long-term costs.
The example above shows how absolute risk can be used in prioritizing different potential deci-
sions. It is able to do that by using actual monetary values for consequences while using realistic
probabilities of different events. To appreciate the power of this approach further, let us increase
the cost of not observing the damage for this inspection situation. We can, for example, double such
a cost, while keeping all other costs and probabilities intact, as displayed in Table 8.32. The costs
of the three methods are shown in Table 8.33. Obviously, there are clear differences between the
costs. Now the visual method is having the highest cost. The lowest cost is the combined method.
The higher consequences of not observing damage necessitate that the two combined methods of
monitoring are used to produce the lowest monetary risk.
TABLE 8.33
Deterministic Risk Evaluation (High Consequences)
Needed Probability Inspection Total Method
Method Repairs (%) Results Probability Cost Cost Cost
Visual Immediate 20.00 Observed 40.00 2.000 0.160 3.34
Inspection Immediate 20.00 Not observed 60.00 11.500 1.380
1–2 Years 50.00 Observed 50.00 1.500 0.375
None 30.00 Observed 60.00 0.500 0.090
SHM Only Immediate 20.00 Observed 70.00 2.500 0.350 2.73
1–2 Years 50.00 Observed 80.00 2.000 0.800
None 30.00 Observed 60.00 1.000 0.180
Both Immediate 20.00 Observed 85.00 2.700 0.459 2.375
1–2 Years 50.00 Observed 95.00 2.200 1.045
None 30.00 Observed 90.00 1.200 0.324
TABLE 8.34
Probabilistic Risk Results
Cost ($ Million)
Method Mean ($ Million) Standard Deviation Sample Minimum Sample Maximum
Visual inspection 2.10 1.68 0.5 6.0
SHM only 2.17 1.3 1.0 6.5
Both 2.15 .97 1.2 6.7
These results are still an expression of single valued averages. The decision maker still does
not have an indication of the level of uncertainty of these numbers. A probabilistic risk method is
still needed to gain more feeling of the situation. Using the decision tree of Figures 8.34 and 8.35
it is possible to build three Monte Carlo simulation models for the three observation methods. The
number of samples of these models was chosen to be 500 samples. A uniform distribution func-
tion was assumed for all events in the tree. A COV of 15% was chosen as a realistic value for each
of those distributions. Combined with the information in Figure 8.35, the probabilistic risks can
now be evaluated. The resulting mean and standard deviation for the risks of the three observation
methods are shown in Figures 8.36 through 8.38. We note that even though the visual inspection
method still produces the lowest monetary mean, it also produces the highest standard deviation.
The lowest standard deviation is the combined methods. Such results are not entirely surprising.
When studying the probabilistic risk for the high consequences case (Table 8.34), the Monte Carlo
simulation results produces different prioritization, as in the deterministic risk case as shown in
Table 8.35. The PDF and CDF of the high consequences cost situations of the three methods are
shown in Figures 8.39 through 8.41. Both the average cost and the standard deviation of the visual
inspection method are the highest of the three methods, while the combined method produced the
PDF and CDF of cost of visual inspection

100.00%
90.00%
80.00%
70.00%
60.00%
PDF/CDF
PDF
50.00%
CDF
40.00%
30.00%
20.00%
10.00%
0.00%
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00
Cost ($ million)
FIGURE 8.36 PDF and CDF of cost of visual inspection.
PDF and CDF of cost of SHM method

100.00%
90.00%
80.00%
70.00%
60.00%
PDF/CDF
PDF
50.00%
CDF
40.00%
30.00%
20.00%
10.00%
0.00%
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00
Cost ($ million)
FIGURE 8.37 PDF and CDF of cost of SHM method.
least average cost and standard deviation. Thus, the combined method is the preferred method for
this high consequence situation: the decision of which method is preferred is fairly clear.

The method just presented utilizes a probabilistic branching of the decision tree, as opposed to
the deterministic branching approach. The probabilistic branching versus deterministic branching
methods can be compared as follows:
• The probabilistic branching method results in a CDF of the overall costs of different meth-
ods, thus making it easy to compare absolute risk of each option. The CDF gives the deci-
sion maker the option to use any nonexceedance probability that suites the project’s needs.
Using deterministic branching offers only a single valued deterministic result, this might
not be adequate to make an accurate decision.
PDF and CDF of cost of combined methods

100.00%
90.00%
80.00%
70.00%
60.00%
PDF/CDF
PDF
50.00%
CDF
40.00%
30.00%
20.00%
10.00%
0.00%
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00
Cost ($ million)
FIGURE 8.38 PDF and CDF of cost of combined (visual and SHM) methods.
TABLE 8.35
Probabilistic Risk Results
Cost ($ Million)
Method Mean ($ Million) Standard Deviation Sample Minimum Sample Maximum
Visual inspection 3.28 3.46 0.5 11.5
SHM only 2.75 2.7 1.0 12.0
Both 2.35 1.84 1.2 12.2
PDF and CDF to cost of visual inspection high consequences

100.00%
90.00%
80.00%
70.00%
60.00%
PDF/CDF
PDF
50.00%
CDF
40.00%
30.00%
20.00%
10.00%
0.00%
0.00 2.00 4.00 6.00 8.00 10.00 12.00 14.00
Cost ($ million)
FIGURE 8.39 PDF and CDF of cost of visual inspection (high consequences).
PDF and CDF of cost of SHM method high consequences

100.00%
90.00%
80.00%
70.00%
60.00%
PDF/CDF
PDF
50.00%
CDF
40.00%
30.00%
20.00%
10.00%
0.00%
0.00 2.00 4.00 6.00 8.00 10.00 12.00 14.00
Cost ($ million)
PDF and CDF of cost of combined methods

high consequences
100.00%
90.00%
80.00%
70.00%
60.00%
PDF/CDF
PDF
50.00%
CDF
40.00%
30.00%
20.00%
10.00%
0.00%
0.00 2.00 4.00 6.00 8.00 10.00 12.00 14.00
Cost ($ million)
• Advent of powerful computing power, the reliance of Monte Carlo while performing prob-
abilistic branching method is more feasible than ever.
• The probabilistic branching method can utilize the results of an SHM project. If proper
coordination and planning are in place, such results can improve the accuracy of the cost–
benefit analysis, thus ensuring an optimal financial decision regarding the project on hand.
8.6.8 Absolute Risk: Life Cycle Analysis

One implicit assumption that was made for different methods of this chapter so far has been that
there are no interdependencies between problem parameters. Witness the many branches of the
decision tree of Figure 8.34 are not interconnected, except to the nodes that immediately support
them. We also note that Monte Carlo method, which is described in Section 8.7.2, needs special
considerations to accommodate such interdependencies between variables. Of course such simpli-

fying assumption is too limiting in practical situations. There are many instances when different
problem parameters are interdependent at many levels. Such interdependency can greatly affect
the outcome of the problem and the resulting decisions that might be made. We present in this sec-
tion a fairly simple example of an absolute risk evaluation approach that includes some parameter
interdependency.

Estimating life cycle costs LCC of a particular system is a recurring dilemma for decision makers.
The LCC problem as related to SHM/SHCE is explored in Chapter 10 by Ettouney and Alampalli
(2012). For now we present a simple LCC problem with an emphasis on interdependence between
parameters. Consider a simple situation of a two parameter cost model: cost of an earthquake hazard
and cost of wear and tear (normal deterioration) of a bridge. The decision maker is tasked of esti-
mating the LCC of the bridge in this situation. For simplicity, let us ignore the effects of economic
rates (inflation, discount, etc.) over the years. The LCC becomes
LCC = ∑ (∑ ( A C ) + C )
Years
1 1 2 (8.117)
With A1 is the probability of the annual occurrence of seismic events. The cost of such an earth-
quake is C1. The annual cost of normal deterioration of the bridge is C2. To simplify the problem
even further, we assume that there is only one earthquake level that need to be considered the
Equation 8.117 becomes
LCC = ∑ ( A1C1 + C2 ) (8.118)
Years
Equations 8.117 and 8.118 illustrates the concept of risk. They are similar to the form of Equation
8.107. They account for the effects of hazards, vulnerabilities, and consequences. The risk is
expressed in monetary terms. The risk can be evaluated in a deterministic (single valued) man-
ner. Assume, for example, that the earthquake of interest has an annual probability of occurrence
of 0.2% (500-years earthquake) and the cost of such an occurrence is estimated at $2.0 million.
Furthermore, assume that the cost of annual normal deterioration is $0.5 million. The LCC for 20
years is
LCC = 20 × ( 0.002 × 2 + 0.5) = $10.8 Million

We can, of course, apply a true probabilistic risk analysis to Equation 8.118. If we assume that the
two costs are random variables with properties as shown in Table 8.36.
Noting that LCC is now a simple function of random variables, we can use the expressions of
Table 8.4 to compute the means and standard deviation or by applying a Monte Carlo simulation
analysis, with number of samples of 5000. The mean is µLCC = $10.11 million. The standard devia-
tion is σLCC = $2.01 million.
These results offer the decision-maker ability to make more educated decisions than in the
deterministic, single valued situation. For example, for conservative estimate of LCC, the deci-
sion maker may choose to use the µLCC + σLCC in the LCC estimation. Such a value would produce
TABLE 8.36
Statistical Properties of Cost
Parameter Distribution Mean $ (Million) COV $ (Million)
C1 Uniform 2 0.2
C2 Uniform 0.5 0.2
a nonexceedance probability level of 84%, if the PDF is normally distributed. Thus estimating the
LCC as
LCC = m LCC + s LCC (8.119)
This would produce LCC = $12.12 million. The difference between the single valued/deterministic,
and the more realistic/probabilistic risk is almost 20%.
Let us now turn our attention to the subject of dependency of parameters. Upon further consid-
erations of Equations 8.117 and 8.118, we can only conclude that the two costs C1 and C2 must have
some kind of dependency. The cost of normal deterioration is usually computed without accounting
for any other abnormal rehabilitation costs. In the case of a seismic event, there would be rehabilita-
tion of the bridge that would cost C1. Such rehabilitation effort would definitely affect the cost of
normal deterioration. Such a dependency has not been studied as of the writing of this manuscript.
It is clear, however, that the cost of deterioration is C2 as long as no seismic event occurred. As soon
as a seismic event occurs at a particular year, the cost of normal deterioration in the following years
will need to be different from C2. Clearly, Equations 8.117 and 8.118 can’t accommodate such a
dependency in their current simple forms.
One possible way to handle the dependency is to solve the LCC problem in a true time line fash-
ion, using a Monte Carlo simulation method. Recall that the earthquake hazard was assumed to
have an annual probability of occurrence of 0.002%. This can be restated as a uniform PDF in the
range of 0–500 years. Figure 8.42 shows both the PDF along with the CDF of this function. We can
then build a simulation model as follows:
1. Start the simulation with trial i = 1

2. Generate a random number in the range from 0 to 1.0
3. Using the CDF of Figure 8.42, generate the year that an earthquake would happen for that
random number Ni
4. If Ni ≥ N TOTC2 then there would be no seismic event in this trial
a. LCCi = N TOT
b. Go to step 6
5. If Ni < N TOT then there will be an earthquake event in this trial at the year Ni
a. LCCi = C21(N TOT − Ni) + NiC2 + C1
6. Assign I = i + 1
7. Repeat steps 1 through 6 until the desired number of Monte Carlo trials is completed
The earthquake has a

uniform probability CDF
1.0
of occurrence once
in 500 years
Probability
Uniform PDF
0 500
Year
FIGURE 8.42 PDF and CDF of a 500-years earthquake.

Earthquake Earthquake do
C1 occurs during not occur during
life span life span
C2 C2
C21
Cost
Cost
Years Years
C2 = Cost of normal deterioration per year

C1 = Cost of an earthquake event
C21 = Cost of normal deterioration per
year, after an earthquake event
FIGURE 8.43 Dependency of costs of normal deterioration and earthquake hazard.
TABLE 8.37
LCC for Different Analysis Methods
Life Cycle Costs $ (Million)
Probabilistic
Independent Dependent
Cost of Seismic Mean Mean
Life Span, Years Retrofit $(M) Deterministic Mean STD +STD Mean STD +STD
20 2.00 10.08 10.11 2.01 12.13 9.99 1.96 11.95
20.00 10.80 10.81 2.02 12.83 10.75 4.15 14.90
40.00 11.60 11.60 2.03 13.63 11.64 8.09 19.73
50 2.00 25.20 25.24 4.99 30.23 24.70 4.95 29.65
20.00 27.00 26.93 4.95 31.88 26.54 6.87 33.41
40.00 29.00 28.83 5.03 33.86 28.49 12.00 40.49
The cost C21 is the cost or normal deterioration after the earthquake rehabilitation efforts are made.
N TOT in the above steps is the required life span of the bridge. The dependency logic for this example
is shown in Figure 8.43. Upon finishing steps 1 through 7, the data series LCCi can be used to gener-
ate the CDF of LCC, as well as the mean and standard deviation of LCC.
If we assume that C21 = $0.3 million, and performing Monte Carlo simulation process, we can
compute the µLCC and σLCC for each case of interest. Table 8.37 shows the results for different life
spans, seismic retrofit costs, and deterministic, probabilistic independent, and probabilistic depen-
dent cases. The LCC varies considerably; which shows the importance of probabilistic and depen-
dent simulation modeling. The equivalent histograms for life spans of 20 years are shown in Figures
8.44 through 8.49. The CDFs are shown in Figures 8.50 through 8.52.

The importance of the dependency between parameters in risk assessment was explored in this
example. We would like to mention that the importance of this dependency is not limited to LCC.
In most practical situations there will invariably be dependencies between important parameters. It
is not an easy task to isolate then quantify those dependencies. Yet, it is essential to accommodate
Histogram of frequency of LCC Independent analysis,

life span = 20 years, cost of earthquake refrofit = $2 million
70.00%
60.00%
Frequency 50.00%
40.00%
30.00%
20.00%
10.00%
0.00%
00–10 10–20 20–30 30–40 40–50 50–60 60–70 70–
LCC range ($ million)
FIGURE 8.44 Histogram: life span = 20, cost of EQ = 20, independent analysis.
Histogram of frequency of LCC Dependent analysis,

life span = 20 years, cost of earthquake retrofit = $20 million
60.00%
50.00%
40.00%
Frequency
30.00%
20.00%
10.00%
0.00%
00–10 10–20 20–30 30–40 40–50 50–60 60–70 70–
FIGURE 8.45 Histogram: life span = 20, cost of EQ = 20, dependent analysis.
such dependencies in any risk analysis. Short of that, the decision maker might end up with solutions
of a problem that is very different from the problem on hand.
One final comment about this example: the reader might ask: what is the role of SHM in this
example? We can answer such a question by observing first that to accurately estimate the costs C1
and C2, there is a need for the following:
1. Accurate measurements and estimation of the normal deterioration of the bridge and the
different rehabilitation measures that are needed to counter such deterioration
2. Accurate structural analysis and identification to estimate the damaging effects of earth-
quake on the bridge
3. The above two needs might also include data collection (real time or intermittent)
All of the above are direct and indirect usage of SHM/SHCE tools and methodologies. Without
employing some or all of those tools and methodologies, the cost estimations of C1 and C2 will not
be more than a theoretical exercise with no great practical value.
Histogram of frequency of LCC Independent analysis,

60.00%
50.00%
Frequency 40.00%
30.00%
20.00%
10.00%
0.00%
00–10 10–20 20–30 30–40 40–50 50–60 60–70 70–
Histogram of frequency of LCC Dependent analysis,

60.00%
50.00%
40.00%
Frequency
30.00%
20.00%
10.00%
0.00%
00–10 10–20 20–30 30–40 40–50 50–60 60–70 70–
8.7 STOCHASTIC MODELS

Performance of civil infrastructure is a time-dependent issue. The time scale of performance can
be extremely short (acoustic emission from cracks, sudden or brittle collapse, etc.), short (earth-
quake seismic wave effects, hydrodynamics of flash floods, speedy truck, etc.), or long (deteriora-
tion of reinforced concrete, corrosion of steel girders, etc.). One of the commonalities between all
of these performance measures is that they all exhibit uncertain behavior in one form or another.
Such time-dependent uncertain behavior is commonly referred to as stochastic behavior, or stochas-
tic processes. The subject of stochastic processes has been studied by numerous authors; see, for
example, Lawler (1995) and Crandal and Mark (1963). Many of the stochastic processes techniques
have been used by decision makers for civil infrastructure (see Elishakoff 1983). In this chapter, we
review two of the most popular methods in stochastic processes for bridge management systems,
namely Markov processes and Monte Carlo methods. We offer some practical examples for each
technique.
Histogram of frequency of LCC Independent analysis, life

span = 20 years, cost of earmquake retrofit = $40 million
80.00%
70.00%
60.00%
50.00%
Frequency
40.00%
30.00%
20.00%
10.00%
0.00%
00–10 10–20 20–30 30–40 40–50 50–60 60–70 70–
Histogram of frequency of LCC dependent analysis,

life span = 20 Years, cost of earthquake retrofit = $40 million
60.00%
50.00%
40.00%
Frequency
30.00%
20.00%
10.00%
0.00%
00–10 10–20 20–30 30–40 40–50 50–60 60–70 70–
8.7.1 Markov Processes

Assume that the is a set of different states θi, with i = 1, 2, …, M. The states can describe states
of damage of bridge deck, or a state of deterioration of paint, and so on. Further we assume that
in a given time period, the probabilities of the ith state to change to the jth state is constant at pij.
This set of probabilities form the components of an M × M matrix [T] that is called transition
matrix. If the initial probability of the formation of each state is defined as {p 0}, then the prob-
abilities that the subject of interest is at a given state after N periods are the components of the
vector {pN} such that

{ pN } = { p0 }[T ]N (8.120)
The above process is called Markov process (Figure 8.53). It is used in several applications in
civil and structural engineering, especially in the bridge management field. It has the advantage of
Cumulative frequency of LCC

100.00%
90.00%
80.00% Independent analysis
70.00% Dependent analysis
60.00%
CDF
50.00%
40.00%
30.00%
20.00%
10.00%
0.00%
>0 >=10 >=20 >=30 >=40 >=50 >=60 >=70
FIGURE 8.50 CDF for life span = 20, cost of EQ = 20.

100.00%
90.00%
60.00%
CDF
50.00%
40.00%
30.00%
20.00%
10.00%
0.00%
>0 >=10 >=20 >=30 >=40 >=50 >=60 >=70
simplicity. However, one of the most important limitations of the methods is that probabilities in the
transition matrix are assumed to be constant.
In addition to the above assumptions, it is clear that the components of the transition matrix [T]
must be accurate, otherwise the computed probabilities, and the ensuing decisions, will be faulty.
See Benjamin and Cornell (1970) for more on theoretical applications of Markov processes. Also,
see Agrawal et al. (2009) for bridge management applications of Markov process.
The field of SHM can take advantage of Markov processes in enhancing decision making for many
situations. The above examples of bridge paint, or deck deterioration illustrates such use. Conversely,
SHM can also help in producing accurate probabilities for populating the transition matrices. We
explore both aspects of SHM/Markov process interrelationship by providing a specific example.
8.7.1.1 Example: Bridge Paint Status

As an example of Markov process use in decision making, let us consider the bridge paint
deterioration problem. In our example, the decision maker decides that there are three states

100.00%
90.00%
60.00%
CDF
50.00%
40.00%
30.00%
20.00%
10.00%
0.00%
>0 >=10 >=20 >=30 >=40 >=50 >=60 >=70
Assume three states for paint: Transition matrix

Initial paint condition: good
1-Good, 2-Fair, 3-Poor te
sta en P
od iv w rob 0.30 0.50 0.20
Probabilty Good state
tim ill b abilt

will be Fair in a given
y go in a g 0.00 0.70 0.30

bi l t d e p e Po y G
ba goo 20 erio or in ood 0.00 0.00 1.00
50%
Pro ll be eriod 0% % d
wi e p 3 a g stat
time period
ive e
tim n
Paint condition: Good Paint condition: Fair Paint condition: Poor

a g ate
ive e
n
n
ag e
a g at
will be good in a given

n
ive
Pr ill b per
Pr ill b per
Pr ill b per
d n st
d n at
ive
d n st
will be Fair in a given

Probabilty poor state
will be fair in a given
io i st
w me
ob e g io
ob e P iod
w me
io i r
ob e P iod
w me
er ood poo
Probabilty fair state
io i d
Probabilty fair state
er ood fair
ti
er ood goo
ti
ab oo d
ti
ab oo
ab oo
ilt d
ilt r
ilt r
e p e g lty
e p e g lty
e p e g lty
y p in
y f in
y p in
tim ill b abi
00%
50%
70%
tim ill b abi
tim ill b abi
%
air a g
30
20
10
oo a g
oo a g
w rob
30
00
00
w rob
time period
time period
time period
w rob
%
%
0%
r s iv
r s ive
sta ive
P
P
tat en
P
tat n
te n
e
Paint deterioration: example of Marcovian process
FIGURE 8.53 Example of Markov processes: deterioration of paint.
for paint: good, fair, and poor. Previous experiences and studies have produced a generic paint
state transition matrix (based on a 2-year period) similar to that in Table 8.38. Such a generic
matrix is used all over network of bridges that the decision maker in our hypothetical example is
managing. It is a 3 × 3 matrix. The cells in the matrix are the transition probabilities pij, which
indicates the probability that the state of paint will change from the ith state to the jth state within
2 years. Note that the sum of each of the row in the matrix is 1.0, as expected. Also, note that
the probabilities in lower triangle of the matrix are all zeros, which indicates that the state of the
paint can’t improve during the 2-years period. The transition matrix can be used to estimate the
probabilities of a given paint state at any time.
The above process is fairly simple, yet it is very sensitive to the probability values in the transition
matrices. SHM can help in producing accurate transition probability matrices, as is discussed next.
TABLE 8.38
Qualitative Estimate of
Transition Matrix of Paint Status
Paint Status Good Fair Poor
Good 0.3 0.5 0.2
Fair 0.0 0.7 0.3
Poor 0.0 0.0 1.0
TABLE 8.39
Initial Paint Status at Different SHM Observation Zones
Bridge Zone Initial Paint State
Bridge # 1. Older Bridge; thus has 1 Good
several different initial paint states 2 Fair
3 Fair
4 Poor
5 Poor
Bridge # 2. Newer Bridge, lesser number 6 Good
of different initial paint states 7 Fair
8 Fair
Bridge # 3. Newest Bridge. Less number 9 Good
of different initial paint states 10 Good
8.7.1.2 Use of SHM in Markov Processes: Paint States of a Bridge

One of the main uses of SHM activities in a Markovian processes is to accurately estimate transition
matrices. Generally speaking, transition matrices are estimated using qualitative (best guess) approach.
Such an approach is generic and does not account for specific conditions of the process or geometries
on hand. For example, the paint state of the bridge that was discussed earlier depend on many general
and local conditions. The general conditions, such as type of paint, type of bridge, and age of paint, are
applicable to most situations, and can be considered generic in nature. The local conditions, such as
temperature variation range, humidity or salt usage (snow removal solution, or sea water) are all local
conditions that can affect the deterioration of paint. These local conditions will affect transition matrices
and must be accommodated in any Markov process analysis. Let us consider how an SHM project can
help in building a transition matrix for the evaluation of the paint deterioration problem in a bridge.
The bridge official is tasked with producing a more realistic transition matrix for bridge paint
states. The paint states are kept similar to the ones mentioned earlier. Three bridges that have
similar local conditions are chosen for this project. It is decided that manual visual inspection is
suited for this project, since the estimation of paint states is mostly qualitative. Each of the three
bridges has different paint states at several zones as shown in Table 8.39, for a total of 10 zones.
The paint states at each of these zones were observed on a quarterly basis, over 8 years period, thus
producing 32 observation points at each zone. The observation results are shown in Table 8.40. The
frequencies of the different observations are then tabulated (Table 8.41). These frequencies are then
grouped in a suitable form in Table 8.42. Those probabilities are finally arranged in an SHM-based
transition matrix as shown in Table 8.43.
8.7.1.3 Value of SHM for a Markov Process: Paint States

We turn our attention now to the value of the SHM project. Recall the earlier qualitative estimate of
the paint transition matrix in Table 8.38. We had just evaluated the more accurate transition matrix
TABLE 8.40
SHM Observation Results
Bridge # 1 Bridge # 2 Bridge # 3
2-Years
Observation Increments 1 2 3 4 5 6 7 8 9 10
0 Good Fair Fair Poor Poor Good Fair Fair Good Good
1 1 Good Fair Fair Poor Poor Good Fair Fair Good Good
6 Fair Fair Fair Poor Poor Good Fair Fair Good Good
9 2 Fair Fair Fair Poor Poor Good Fair Fair Good Good
11 Fair Fair Poor Poor Poor Fair Fair Fair Good Good
14 Fair Poor Poor Poor Poor Fair Fair Fair Fair Good
17 3 Fair Poor Poor Poor Poor Fair Fair Fair Fair Good
19 Fair Poor Poor Poor Poor Fair Fair Poor Fair Good
24 Fair Poor Poor Poor Poor Poor Fair Poor Poor Good
25 4 Fair Poor Poor Poor Poor Poor Poor Poor Poor Good
26 Fair Poor Poor Poor Poor Poor Poor Poor Poor Good
27 Fair Poor Poor Poor Poor Poor Poor Poor Poor Good
28 Poor Poor Poor Poor Poor Poor Poor Poor Poor Poor
in Table 8.43. It is more accurate, obviously, since it is based on actual observations in more real-
istic conditions that are similar to the bridge conditions of interest, rather than generic conditions.
Qualitatively, it is clear that the components of Table 8.38 and Table 8.43 are considerably different.
The bridge official, however, would like to have a more objective procedure for estimating the value
of SMH procedures. The first step in estimating the value is to compute the cost of the SHM. Recall
that the measurements of the paint deterioration were done by manual observations at three bridge
sites, over the span of 8 years, at quarterly intervals, for a total of 32 observations per bridge zone
(total of ten zones). The cost of these observations, and the management, archiving, and computa-
tions is estimated by the bridge official is CSHM, while the cost of painting the bridge is CPNT.
TABLE 8.41
Frequency of Different Observations
Bridge # 1 Bridge # 2 Bridge # 3
Starting State Ending State 1 2 3 4 5 6 7 8 9 10 Total
Good Good 1 0 0 0 0 1 0 0 1 3 6
Good Fair 0 0 0 0 0 1 0 0 1 0 2
Good Poor 0 0 0 0 0 0 0 0 0 1 1
Fair Fair 2 1 1 0 0 0 3 2 1 0 10
Fair Poor 1 1 1 0 0 1 1 1 1 0 7
Poor Poor 0 2 2 4 4 1 0 1 0 0 14
Total transitions per zone 4 4 4 4 4 4 4 4 4 4
TABLE 8.42
Consolidated Frequencies of
Observations
Total End States
Total Start States
Good Fair Poor
Good 9 6 2 1
Fair 17 0 10 7
Poor 14 0 0 14
TABLE 8.43
SHM-based Transition
Matrix of the Paint Status
Paint Status Good Fair Poor
Good 0.67 0.22 0.11
Fair 0.0 0.59 0.41
Poor 0.0 0.0 1.0
The value of the SHM-based transition matrices is estimated using the potentially saved paint
costs. For example, if the initial paint state is “Good,” the initial probability vector is
{Po } = {1 0 0} (8.121)
After 2N years of operations, the paint state probability vector is
{PN } = {P0 } [T ]N (8.122)
Where [T] is a 3 × 3 probability transition matrix. The N power in Equation 8.122 is due to the fact
that the transition probabilities are computed for a 2-years time interval.
From Table 8.38 and Equations 8.121 and 8.122, the state probabilities from generic transition
matrices after 10 years are
{PN1 } = {0.00 0.21 0.79} (8.123)

From Table 8.43, the SHM-based state probabilities vector after 10 years are
{PN1 } = {0.14 0.17 0.69} (8.124)
To get a wider understanding of the deterioration process as predicted by Markov analysis, we plot
the 10 years probabilities of paint being at a given state in Figures 8.54 through 8.56 for good, fair,
and poor state, respectively. The differences between the generic and SHM-based probabilities are
self-evident. Any management decision that is based on Markov process for paint deterioration
would be fairly inaccurate if it is based on generic transition matrices. We can quantify this impor-
tant conclusion as follows.
If the management rule for that particular office is to repaint if the probabilities of the paint state
of “poor” is ≥0.75, then the generic transition matrices based estimations will cause an initiation of
repainting. The SHM-based results will not initiate repainting. Continuing the process for the SHM-
based computations reveals that 2 more years will pass before an repainting is needed. If we ignore
discount and inflation rates, for simplicity, we can deduce that over the course of the life span of the
Probability of paint deterioration to a good state
1.20
1.00
0.80
Probability
SHM based
0.60
Generic
0.40
0.20
0.00
0 1 2 3 4 5 6 7 8 9 10 11
Years
FIGURE 8.54 Probabilities of good paint state.
Probability of paint deterioration to a fair state

0.60
0.50
0.40
Probability
SHM based
0.30
Generic
0.20
0.10
0.00
0 1 2 3 4 5 6 7 8 9 10 11
Years
FIGURE 8.55 Probabilities of fair paint state.

Probability of paint deterioration to a poor state

0.90
0.80
0.70
0.60
Probability
0.50
SHM based
0.40 Generic
0.30
0.20
0.10
0.00
0 1 2 3 4 5 6 7 8 9 10 11
Years
FIGURE 8.56 Probabilities of poor paint state.
bridge the cost savings are CPNT. Further, if we assume that the number of bridges in the network
that have similar conditions as the subject bridge is NBRIDGE , the monetary value of the SHM project,
over the course of the life span of the bridge is
VSHM = N BRIDGE CPNT − CSHM (8.125)
The above example showed net monetary value that result from direct cost savings due to later
repaint needs. If the SHM-based transition matrices indicate that a sooner repaint is needed, there
is still a value for such situation. The value in this case will come from the realized performance of
repainting the bridge before its state deteriorate below the stated performance goal (≥0.75 probabil-
ity of a “poor” painting condition). The value is not monetary, but it can be computed using utility
theory techniques as mentioned in Section 8.4.2.
One final remark about the above example: it was based on a single paint state throughout the life
span of the bridge. More realistic analysis that account for different paint states can be exercised.
The computations of value are similar for those more complex examples.
8.7.2 Monte Carlo Method

Monte Carlo simulation method has been used in many fields. Its popularity is attributed to its
simplicity. Lately, because of the advances in computing power, Monte Carlo uses have grew
even further and faster. For our purposes, we describe Monte Carlo in its simplest form as
follows:
1. Input: a single, or a set, of random variables with known probability distributions

2. Using random number generator, generate a single input data set
3. Model: Create a mathematical model that uses the random values in #2 to produces a single
output data set. Note that the model itself can contain random properties
4. Repeat #2 and #3 as many times as prescribed, until an ensemble of data sets is formed
5. Form frequency analysis (see Section 8.3.2.2 of this chapter) to produce needed statistical
properties, and plot the histograms, or evaluate PDFs, of CDFs, as needed
Figure 8.57 shows a summary of the process. The Monte Carlo simulation method is utilized exten-
sively in this document. For further readings about the basis of the method, or other applications,
check Trueman (1977).
Monte Carlo simulation
Given distributions Evaluates

Given distributions
of mathematical statistics of the
of input random
model random output data
variables No
variables sample
Is total number of
desired sample
points achieved? Yes
Generate estimates of the random variable
using random generators
Generate Output is considered a

mathematical single data entry within
Generate input
model the total desired output
sample
FIGURE 8.57 Monte Carlo simulation technique.
8.8 STRUCTURAL ANALYSIS IN DECISION MAKING

8.8.1 General
Properties of infrastructure are usually assumed to be single valued, or deterministic. Such an
assumption is usually accepted during the course of conventional design procedures. In many
decision making situations, such an assumption might not be adequate. Since decision making
process involves uncertainties of almost all issues, it is reasonable to accommodate the uncertain-
ties of properties of infrastructure in the decision making process. Ignoring such uncertainties
might lead to inadequate decisions. In some situations such as when assessing probabilities and
consequences of failure, ignoring uncertainties of infrastructure properties might lead to unsafe
decisions.
The problem can be summarized as follows. Find the response of a given structure that has
some of properties as random variables. The random input can be described by its PDF, or simply
by its mean and variances. The computed output is usually in a form of a mean and variances
of output measures, such as displacements, strains, or stresses. There are many methods to this
problem. Many solutions address uncertainties in structural components or structural systems;
see Elishakoff (1983) or Melchers (2002). Whether the main interest lies in structural compo-
nents or systems, there is a need for a special probabilistic analysis program. The steps for such
an analysis are
1. Model the structure using a specialized probabilistic analysis program

2. Assign random properties as appropriate
3. Assign deterministic properties as appropriate
4. Assign the different design loads
5. Analyze for desired design measures, such as displacements, strains, or stresses. Note that
the design measures in this situation will include statistical parameters such as means and
standard deviations
6. Use the statistical results of step #5, above to evaluate confidence limits, or probabilistic
statements about the safety
7. Use the results from #6 in the remainder of the decision making process. For example,
decide if the results of step #6, above, is within acceptable safety limits of the structure on
hand
The above steps are applicable if specialized probabilistic analysis tools are available to the official.
However, such specialized tools might not be easy to obtain. In the absence of such specialized
tools, the following steps might be followed:
1. Model the structure using conventional structural analysis techniques

2. From a random number generator and utilizing the probabilistic statement concerning the
uncertain properties, find a random value(s) for those properties
3. Assign the values obtained in #2 to the appropriate locations in the conventional analytical
model of step #1
4. Assign deterministic properties to the rest of the model as appropriate
5. Assign different design loads
6. Analyze for desired design measures, such as displacements, strains, or stresses
7. Repeat steps 2 through 6
8. From the information obtained in #6, form the statistical parameters (such as means and
standard deviations) for different design measures
9. Use the statistical results of step #8, above to evaluate confidence limits, or probabilistic
statements about the safety
10. Use the results from #9 in the remainder of the decision making process. For example,
decide if the results of step #9, above are within acceptable safety limits of the structure
on hand
Structural investigations under uncertain

conditions
Specialized probabilistic Conventional structural

structural analysis tools analysis tools
Uncertain Generate Random number

properties properties generators
Probabilistic Deterministic
model model
Solve for Solve for output Iterate until

uncertain output measures desired sample
size is formed
Evaluate
probabilistic
output
Continue decision-
making process
FIGURE 8.58 Methods for probabilistic structural analysis.

The reader would recognize the above steps as a form of Monte Carlo’s method.
Both methods would enable the official to account for uncertainties in structural properties.
Figure 8.58 shows schematics of the steps of both approaches. There are advantages and limitations
of each method:
• The specialized probabilistic method is computationally efficient, but it requires a special
probabilistic analytical tools that might not be available to the decision maker
• The Monte Carlo approach is usually available, but it can be computationally inefficient
The appropriate method will then depend on the severity and complexity of the structural system
under considerations.
8.8.2 Probabilistic Finite/Boundary Element Methods

Among specialized probabilistic structural analysis methods for structural systems are probabilistic
finite element methods and probabilistic boundary element methods. We present here a brief explana-
tion of probabilistic boundary element method. For complete developments and applications in different
fields, the reader is referred to Ettouney et al. (1989a and 1989b) and Daddazio and Ettouney (1990).
Let us assume that for an elastic problem, the discretized boundary element equation can be
expressed as
M ( x1 , x2 ) p ( x1 , x2 ) = N ( x1 , x2 ) q ( x1 , x2 ) (8.126)
Where M(x1, x2) and N(x1, x2) are square matrices that include the integrals of the kernel functions
of the problem. The vectors p(x1, x2) and q(x1, x2) include the boundary displacements and traction.
We assume that x1 and x2 are two uncertain physical properties (e.g., they can represent an uncertain
shear modulus and Poisson’s ratio). Let us assume that x1 and x2 are the mean values of x1 and x2. Let
us define the perturbation values
e1 = x1 − x1 (8.127)
e2 = x2 − x2 (8.128)
We can now utilize Taylor series to expand Equation 8.126 about the mean values of x1 and x2.
Utilizing perturbation analysis techniques, Bellman (1964), Equation 8.126 will produce the fol-
lowing recursion equations
M p = Nq (8.129)
∂p ∂N ∂M (8.130)
M = q− p
∂x1 ∂x1 ∂x1
∂p ∂N ∂M
M = q− p (8.131)
∂x 2 ∂x 2 ∂x 2
∂2 p ∂M ∂p
M 2
=2 (8.132)
∂x 1 ∂x1 ∂x1
Higher order equations can be developed. We now solve the recursive equations for p, ∂p / ∂x1 , ∂p / ∂x2 ,
and ∂ 2 p / ∂ x12 and express the displacements as
∂p ∂ p 1 2 ∂2 p 1 2 ∂2 p ∂2 p
p ( x1 , x2 ) = p + e1 + e2 + e1 + e2 + e1 e2 (8.133)
∂ x1 ∂ x2 2 ∂ x12 2 ∂ x22 ∂ x1 ∂ x2
The mean and the mean square of the displacement vector (correct to the second order) can be easily
computed by
∂p ∂p
E [ p ] = p + E [ e1 ] + E [ e2 ]
∂ x1 ∂ x2
1 ∂2 p 1 ∂2 p ∂2 p
+ E ⎡⎣ e12 ⎤⎦ 2
+ E ⎡⎣ e22 ⎤⎦ 2
+ E [ e1 e2 ] (8.134)
2 ∂ x1 2 ∂ x2 ∂ x1 ∂ x2
⎛ ∂2 p ⎛ ∂ p ⎞ 2 ⎞ ⎛ ∂2 p ⎛ ∂ p ⎞ 2 ⎞
E ⎡⎣ p 2 ⎤⎦ = p 2 + E ⎡⎣ e12 ⎤⎦ ⎜ p 2
+ ⎜ ⎟ ⎟ + E ⎡ e2
⎣ 2 ⎦ ⎜ p ∂ x 2 + ⎜⎝ ∂ x ⎟⎠ ⎟
⎤ (8.135)
⎝ ∂ x1 ⎝ ∂ x1 ⎠ ⎠ ⎝ 2 2 ⎠
As soon as E[p] and E[p2] are evaluated, the required mean and variance can be established as in
Section 8.3 of this chapter.
Note that Equations 8.126 through 8.135, above can be generalized for more than two uncertain
variables. However, the equations can become too complex. In such a situation, the use of other
techniques, such as Monte Carlo technique becomes more preferable.
8.8.2.2 Example
Consider the bridge foundation problem of Figure 8.59. The foundation is a single footing that rests
directly on a soil layer that rest on a half space. The footing supports a bridge column that sup-
ports the bridge superstructure. The foundation-soil system is located in a seismically active region.
Because of this, the official desires to compute the equivalent soil springs as detailed by Arya et al.
(1979). The elastic properties of the soil layer and the half space are shown in Table 8.44. For our
purposes, the foundation has a square footprint, and is assumed to have a stiffness that is much
higher than the soil stiffness. By using conventional soil-structure analysis techniques, the official
computed the horizontal and vertical soil springs to be as shown in Table 8.45.
Upon further research, the official found out that the soil properties in Table 8.46 were obtained
by a soil consultant who provided also the coefficient of variations (COV) of those properties as
Width = 25 ft. Rigid foundation
Soil height Top soil layer

= 25 ft.
Half space
FIGURE 8.59 Soil—structure interaction problem.
TABLE 8.44
Deterministic Input
Soil Layer Modulus of Poisson
Rigidity (ksf) Ratio
Top layer 2000.00 0.35
Half space 3500.00 0.30
TABLE 8.45
Deterministic Results
Equivalent Spring Type Deterministic Results (kips/ft)
Vertical 150,366
Horizontal 123,285
TABLE 8.46
Probabilistic Input
Modulus of Poisson Ratio
Elasticity (ksf)
Soil Layer Mean COV Mean COV
Top layer 2000.00 0.21 0.35 0.05
Half space 3500.00 0.26 0.30 0.05
TABLE 8.47
Probabilistic Results
Equivalent Mean m Standard COV m+σ m-σ
Spring Type Deviation σ
Vertical 153,461 38,365 0.25 230,192 76,730
Horizontal 119,482 34,649 0.29 188,782 50,182
shown in Table 8.46. The official felt that those COV were higher than normal. It was felt that with
high COV, there is a need for a probabilistic analysis. Fortunately, a PBE computer code was avail-
able, and the official embarked on a probabilistic analysis that employed an algorithm similar to the
one described above. The results are shown in Table 8.47. Note that the means µ of the soil springs
have not changed much from the deterministic results of Table 8.45. However, the standard devia-
tions σ and COV produced a µ+2σ and µ−2σ results that are very different from the mean. Those
limits of the soil springs would represent a 97% of nonexceedance probability (if the resulting dis-
tribution is normal), which is consistent with the probability levels of the whole project. The final
decision was to use the probabilistic results. Note that two analyses would be needed for the upper
and lower bounds of the soil springs. For more on this type of problems, see Ettouney et al. (1989a,
1989b, and 1993).
8.9 FINANCIAL CONSIDERATIONS

8.9.1 Introduction
Decision making processes invariably use monetary means. This monetary means can vary from
simple present and future value considerations to more complex financial considerations of proj-
ects. We describe the basic processes of present and future values in this section. In addition, we
show the importance of uncertainty considerations in financial computations. We also introduce
the concepts of LCA and its relationship to both SHM and financial analysis concepts. Finally,
the role of peer review in the whole process is introduced and discussed. The reader may refer
to more advanced engineering economic references, such as Newnan and Lavelle (1988), Park
(1997), Riggs et al. (1996), and Ruegg and Marshall (1990) for more detailed discussions of this
all important subject.
8.9.2 Present Value

Present value concept recognizes that future costs are worth today less than in the future. This is
due to the interest-earning potential of monies at the present. If the yearly discount rate i, is con-
stant, then the present value PV of a future monies FV at n years away is
FV
PV = n
(8.136)
(1 + i )
Note that the interest rate is called discount rate in the context of present value computations.
As an extension of the present value concept, the present value PVn of future sums of monies FVi
that are realized at uniform time interval is
j=n
FV j
PVn = ∑ (8.137)
j =1 (1 + i ) j
If FVi is constant at FV, then the Equation 8.137 becomes
j=n
1
PVn = FV ∑ (8.138)
j =1 (1 + i ) j
This is a present value of an annuity.
8.9.2.1 Example I: Present Value in Decision Making

Present value is used in several situations in asset management decision making processes. For
example, if the decision maker wishes to compare savings from two projects. Project A promises
graduated cost savings, while project B promises constant cost savings. The total cost savings for
each project is identical for the first 20 years, if discount rates are ignored. By applying the present
value Equation 8.137 for each of the projects, it is possible to make a reasonable choice, as detailed
in Table 8.48. By comparing the present value of the two projects, it is clear that project B will offer
more savings than project A. If the decision maker chooses project B, the total present value savings
will be $224.69 thousands, about 4.7%.
8.9.3 Future Value

Future value FV is the inverse of present value. It is higher than the present value due to interest
effects. Thus, the future value received after n years, with a constant interest i from present monies
PV is as follows:
n
FV = PV (1 + i ) (8.139)
If there is a recurring cost, or benefit, over n years the future cumulative value is
j=n
j
FVn = ∑ Pj (1 + i ) (8.140)
j =1
With Pj is the cost, or benefit, during the jth year.
8.9.3.1 Example II: Future Value in Decision Making

For example, if a constant yearly inspection cost is $0.5 million. After 20 years, with an interest
rate of 7.0%, the compound future value of the inspection costs will be $21.93 million. For a slight
TABLE 8.48
Present Value of Two Projects
Year Discount Project A Project B
Rate (%)
Cost Savings per Present Value Cost Savings per Present Value
Year ($1,000) ($1,000) Year ($1,000) ($1,000)
1 5.00 300 285.71 400 380.95
2 5.00 300 272.11 400 362.81
3 5.00 300 259.15 400 345.54
4 5.00 300 246.81 400 329.08
5 5.00 300 235.06 400 313.41
6 5.00 400 298.49 400 298.49
7 5.00 400 284.27 400 284.27
8 5.00 400 270.74 400 270.74
9 5.00 400 257.84 400 257.84
10 5.00 400 245.57 400 245.57
11 5.00 400 233.87 400 233.87
12 5.00 400 222.73 400 222.73
13 5.00 400 212.13 400 212.13
14 5.00 400 202.03 400 202.03
15 5.00 400 192.41 400 192.41
16 5.00 500 229.06 400 183.24
17 5.00 500 218.15 400 174.52
18 5.00 500 207.76 400 166.21
19 5.00 500 197.87 400 158.29
20 5.00 500 188.44 400 150.76
Total present value 4760.19 4984.88
change of interest rate to 6.5%, the future value of the inspection cost will be $20.65 million. The
importance of interest rate is self-evident. Next section we discuss potential fluctuations (uncertain-
ties) in interest rate and their effect on present and future values.
8.9.4 Considerations of Uncertainties

For decisions concerning present of future values, we note that there are uncertainties in all the
input parameters. Discount or interest rates i, future values FVi, or Pj can all be uncertain. Because
of this, the resulting present or future value will also be uncertain. Equations 8.136 through 8.140
must be evaluated probabilistically. Monte Carlo simulation method is a common method for com-
puting those values in a probabilistic manner.
8.9.4.1 Example I, Continued: Present Value and Uncertainties

Recall that in example I assumed that the discount rate and the cost savings per year for both
potential projects A and B to be deterministic. Such an assumption is not realistic, since it is
obvious that those variables are uncertain. Because of this, the decision maker embarks on a
probabilistic comparison between the two projects. To facilitate such a comparison, the cost sav-
ings per year and the discount rate of both projects, as shown in Table 8.48 were assumed to be
the means of both parameters for the next 20 years. The decision maker made some educated
assumptions regarding realistic uncertainties for both parameters. First, it is assumed that the
uncertainties in both parameters will follow uniform distribution PDF. Second, realistic values
of the COV are estimated for that period. The resulting means and COV for both parameters are
shown in Table 8.49.
TABLE 8.49
Uncertain Parameters for the Two Projects
Year Project A Project B
Cost Savings per Cost Savings per
Discount Year ($1,000) Discount Year ($1,000)
Mean COV Mean COV Mean COV Mean COV
1 0.05 0.12 300 0.1 0.05 0.12 400 0.1
2 0.05 0.12 300 0.1 0.05 0.12 400 0.1
3 0.05 0.12 300 0.1 0.05 0.12 400 0.1
4 0.05 0.12 300 0.1 0.05 0.12 400 0.1
5 0.05 0.12 300 0.1 0.05 0.12 400 0.1
6 0.05 0.12 400 0.1 0.05 0.12 400 0.1
7 0.05 0.12 400 0.1 0.05 0.12 400 0.1
8 0.05 0.12 400 0.1 0.05 0.12 400 0.1
9 0.05 0.12 400 0.1 0.05 0.12 400 0.1
10 0.05 0.12 400 0.1 0.05 0.12 400 0.1
11 0.05 0.12 400 0.1 0.05 0.12 400 0.1
12 0.05 0.12 400 0.1 0.05 0.12 400 0.1
13 0.05 0.12 400 0.1 0.05 0.12 400 0.1
14 0.05 0.12 400 0.1 0.05 0.12 400 0.1
15 0.05 0.12 400 0.1 0.05 0.12 400 0.1
16 0.05 0.12 500 0.1 0.05 0.12 400 0.1
17 0.05 0.12 500 0.1 0.05 0.12 400 0.1
18 0.05 0.12 500 0.1 0.05 0.12 400 0.1
19 0.05 0.12 500 0.1 0.05 0.12 400 0.1
20 0.05 0.12 500 0.1 0.05 0.12 400 0.1
TABLE 8.50
Present Value of Cost Savings
Project Mean STD COV Mean – STD
($1,000) ($1,000) ($1,000)
A 4768.2 128.41 0.0269 4511.4
B 4994.9 132.4 0.0265 4730.1
After a Monte Carlo simulation process using Equation 8.137, the statistical results for the pres-
ent value of the cost savings of the two projects are shown in Table 8.50.
A conservative present value of cost savings would be the mean—two standard deviations. Thus,
project B still offer a better solution, with conservative cost savings of $218.7 thousands, or a cost
saving of about 4.8%. In this case, considerations of uncertainties provided similar results to deter-
ministic computations: the decision maker can proceed with project B confidently.
8.9.4.2 Example II, Continued: Future Value and Uncertainties

Let us consider the effects of uncertainties in example II. Consider that the yearly inspection costs
as a random variable with a mean of $0.5 million. The decision maker feels that such a yearly cost
can fluctuate; thus becoming a random variable, with a COV of 0.2. The decision maker assumes
that such variability can take a uniform probability distribution. Similarly, the interest rate has
uncertainty. The decision maker assumes that the interest rate as a uniformly distributed random
variable will have a mean of 7% and COV of 0.2. A simple Monte Carlo simulation algorithm can be
performed using Equation 8.139. The results of the simulation show that the average future value of
the inspection costs are $22.33 million. The standard deviation is $1.48 million. Thus the two-sigma
cost estimation is $25.3 million. Note that such estimation is higher than the earlier deterministic
cost of $21.93 million. The cost differential of 25.3-21.93 = $3.37 million is the cost of uncertainty.
Such uncertainty should be considered in any decision making process.
8.9.5 Life Cycle Costs/Benefits and Life Span

Life cycle analysis is the analysis of different aspects of a particular system during its expected life
span. These aspects can include different performance measures such as safety, security, or durabil-
ity. In practice, LCA refers to estimation of costs and benefits that are expected during the life span
of the system. We can state that
Optimizing life cycle aspects of a particular system, or systems, is perhaps the most important aspect
of an asset manager.
As stated above, costs/benefits are popular aspects in the field of LCA. We immediately state that
cost and/or benefits need not be of monetary units. Of course, measuring costs/benefits using mon-
etary units is used by many authors in the field.
Accurate LCA can address an important question that faces decision makers often: What is more
beneficial: to expend efforts on a short-term project, such as immediately painting a rusted girder, or a
long-term project, such as hardening an abutment for an expected 50 years earthquake? In other words,
accurate LCA can easily prioritize short-term versus long-term issues. This can be done by incorpo-
rating all aspects of the different projects accurately within the framework of the LCA. Thus, for the
painting versus earthquake hardening issue, the information in Table 8.51 needs to be ascertained.
Clearly, both issues are of importance to the decision maker. Assuming that the monetary metric
is of paramount importance to the decision maker in this particular situation, it is clear that consid-
erations of monetary costs and benefits along the life span of the bridge will aid in prioritizing the
two options. Note that the three LCA issues, namely hazards, vulnerabilities, and consequences are
similar to the main issues involved in risk evaluations, Section 8.6. What differentiate LCA from
risk is that it involves considering the passage of time up to a predetermined life span of the system
under consideration.
Estimation of LCA involves many of the decision making tools that were presented in this chap-
ter. SHM procedures can aid LCA efforts by providing information regarding performance of the
bridge/system. Specific applications of LCA are presented throughout this volume and the compan-
ion volume (see Ettouney and Alampalli 2012).
8.9.6 Peer Review

As noted in earlier sections, SHCE project can be a simple, straight forward project or compli-
cated project with significant time and financial commitment depending on project scope, structure
involved, and so on. Thus peer review or an independent peer review can be very useful in ensuring
successful SHCE project. The primary reason for peer review is to make sure that the SHCE project
will indeed produce the value that is anticipated from the project for the money/efforts planned to
TABLE 8.51
Information Needed for LCA
Issue for LCA Painting Earthquake Hardening
Hazard Immediate, high probability Lesser probability, once in every 50 years occurrence
Vulnerability Rusting, loss of section Severe damage to bridge
Consequences Continued deterioration and ultimate failure Potential bridge failure
be spent. Thus the peer review is to ensure that scientifically the proposed content makes sense, it is
practical, has realistic time frame, and, finally, will assist in the decision an owner intends to make
using the project results. In general, it is the responsibility of the owner or sponsor of the project to
initiate the peer review process.
There are several decisions that have to be made before a peer review is initiated. It is always
useful for organizations that will be involved routinely with SHCE projects to develop a set of peer
review guidelines. These should include guidelines for project selection, review process, review
team composition, compensation for reviewers, guidelines for reviewers, reviewer selection criteria,
review time, and conflict resolution (Alampalli 2001).
The project requiring peer review can be based on project cost, duration, or criticality of the
decision that will be made on the basis of the project outcome. Review can occur at beginning of the
project for small projects or at several stages for big projects covering several years. Normally the
initial review is to make sure that the project assumptions, both technical and nontechnical, makes
sense, and project results can support the reasons the project was initiated. The number of reviewers
and reviewer’s qualifications should be decided by the owner/sponsor of the project rather than the
entity conducting the project. The compensation for peer reviewers should be included in the project
budget. The scope of the review, acceptable time frame for review, compensation schedules, report-
ing requirements should be detailed well to ensure that there will be no conflicts (or minimal con-
flicts) in the future and peer review can be done on time and does not impact the project time frame.
There should be a mechanism to address the reviewers comments especially if the reviewer opinion
differs considerably than the opinions of the proposers. It should also be made clear to everyone in
this process that the peer review process is intended to enhance project quality and thus not to serve
as watchdog for errors; and review recommendations are not binding on the owner/sponsor.
8.10 APPENDIX: TABLES FOR COMPUTING COST OF BRIDGE FAILURE

Tables 8.52 through 8.60 in this appendix are used in a study by Stein and Sedmera (2006); the study
was supported by the National Cooperative Highway Research Program (NCHRP) Project 24-25.
TABLE 8.52
Cost of Bridge Construction
Bridge Superstructure Type, Demolition Total Cost ($/ft2)
Reinforced concrete flat slab; simple span 50–65a
Reinforced concrete flat slab; continuous span 60–80a
Steel deck/girder; simple span 62–75a
Steel deck/girder; continuous span 70–90a
Prestressed concrete deck/girder; simple span 50–70a
Prestressed concrete deck/girder; continuous span 65–1 10a
Post-tensioned, cast-in-place, concrete box girder cast on scaffolding; 75–110
Span length < = 240 ft
Steel Box Deck/ Girders:
Span range from 1 50 ft to 280 ft 76–120
For curvature add a 15% premium segmental concrete box girders; span range from 150 ft to 280 ft 80–100)
Movable bridges; bascule spans & piers 900–1500
Demolition of Existing Bridges:
Typical 9–15
Bascule spans & piers 63
a Increase the cost by 20% for phased construction.

Source: http://www.dot.state.fl.us/structures/Manuals/LRFDSDG2002AugChap11.pdf. Accessed May 26, 2005.
TABLE 8.53
Cost Multiplier for Early Replacement
Average Daily Traffic (ADT) Cost Multiplier for
Early Replacement
ADT < 100 1.0
100 ≤ ADT < 500 1.1
500 ≤ ADT < 1000 1.25
1000 ≤ ADT < 5000 1.5
ADT ≤ 5000 2.0
TABLE 8.54
Comparison of Total and Variable Costs
per Mile
Cost Category Automobiles Trucks
Total per mile S0.45 $1.80
Driver costs – $0.50
Total vehicle cost per mile $0.45 $1.30
Variable cost per mile $0.15 $0.43
Variable as % of total 33% 33%
Source: Minnesota Department of Transportation

(ftp://www. trrb.org/pJf/2003I9.pdf.
Accessed May 26, 2005).
TABLE 8.55
Detour Duration versus ADT
Average Daily Traffic (AUT) Detour Duration (days)
ADT < 100 1095
100 ≤ ADT < 500 730
500 ≤ Ajyr < 1000 548
1000 ≤ ADT < 5000 365
ADT ≤ 5000 183
TABLE 8.56
Value of Time
State Mean Wagea ($/hour) Value of timeb ($/hour)
Alabama 15.35 6.29
Alaska 20.27 8.31
Arizona 16.77 6.88
Arkansas 14.21 5.83
California 20.18 8.27
Colorado 19.14 7.85
Connecticut 21.35 8.75
Delaware 18.77 7.70
District of Columbia 27.87 11.43
Florida 16.23 6.65
TABLE 8.56 (continued)

Value of Time
State Mean Wagea ($/hour) Value of timeb ($/hour)
Georgia 17.23 7.06
Guam 13.20 5.41
Hawaii 17.67 7.24
Idaho 15.76 6.46
Illinois 18.55 7.61
Indiana 16.26 6.67
Iowa 15.38 6.31
Kansas 16.24 6.66
Kentucky 15.47 6.34
Louisiana 15.02 6.16
Maine 16.09 6.60
Maryland 19.89 8.15
Massachusetts 21.78 8.93
Michigan 19.03 7.80
Minnesota 19.15 7.85
Mississippi 13.77 5.65
Missouri 16.57 6.79
Montana 14.37 5.89
Nebraska 15.89 6.51
Nevada 16.49 6.76
New Hampshire 18.01 7.38
New Jersey 20.69 8.48
New Mexico 15.87 6.51
New York 20.96 8.59
North Carolina 16.40 6.72
North Dakota 14.72 6.04
Ohio 17.26 7.08
Oklahoma 14.97 6.14
Oregon 17.78 7.29
Pennsylvania 17.29 7.09
Puerto Rico 10.61 4.35
Rhode Island 18.38 7.54
South Carolina 15.35 6.29
South Dakota 13.98 5.73
Tennessee 15.74 6.45
Texas 16.98 6.96
Utah 16.40 6.72
Vermont 16.66 6.83
Virgin Islands 13.62 5.58
Virginia 18.81 7.71
Washington 19.65 8.06
West Virginia 14.65 6.01
Wisconsin 16.94 6.95
Wyoming 15.63 6.41
Source: http://www.bls.gov/oes/current/oessrest.htm. Accessed January 12, 2006.

The value of time is assumed to be 41% of the mean wage as suggested by José
A. Gómez-Ibáñez, William B. Tye, Clifford Winston, Essays in Transportation
Economics and Policy: A Handbook in Honor of John R. Meyer, 1999.
TABLE 8.57
Occupancy per Vehicle Mile by Daily Trip Purpose
Trip Purpose Mean Standard Error
All personal vehicle trips 1.63 0.012
Work 1.14 0.007
Work related 1.22 0.020
Family/personal 1.81 0.016
Church/school 1.76 0.084
Social/recreational 2.05 0.028
Other 2.02 0.130
Source: NHTS, Highlights of the 2001 National Household Travel

Survey, Bureau of Transportation Statistics, U.S. Department of
Transportation, Washington, DC, 2001.
TABLE 8.58
Value of Time Used in the Derivation of Road User Costs
Value of Time from Value of Time Adjusted
Vehicle Type MBC (1990 Dollars) (1998 Dollars using CPI)
Small passenger car $9.75 $12.16
Medium/large passenger car $9.75 $12.16
Pickup/van $9.75 $12.16
Bus $10.64 $13.27
2-axle single unit truck $13.64 $17.01
3-axle single unit truck $16.28 $20.30
2-S2 semi truck $20.30 $25.32
3-S2 semi truck $22.53 $28.10
2-S1-2 semi truck $22.53 $28.10
3-S2-2 semi truck $22.53 $28.10
3-S2-4 semi truck $22.53 $28.10
Source: http://tti.tamu.edu/documents/407730.pdf. Accessed on May 26, 2005.
TABLE 8.59
Estimates of the Values of Travel Time
Automobiles Trucks
Travel Purpose Small Medium 4-Tire 6-Tire
Business Travel
Value per persona $21.20 $21.20 $21.20 $18.10
Average vehicle occupancy 1.43 1.43 1.43 1.05
Total business $31.55 $31.96 $32.47 $22.01
Personal Travel
Value per persona $10.60 $10.60 $10.60
Average vehicle occupancy 1.67 1.67 1.67
Total personal $17.70 $17.70 $17.70
a 2000 Dollars.
Source: FHWA website (http://isdde.dot.gov/olpfiles/fhwa/010617.pdf. Accessed
May 26, 2005.)
TABLE 8.60
Assumed Number of Lives Lost in Bridge Failure
Number of Lives
Average Daily Traffic (ADT) Lost
ADT < 100 0
100 < ADT < GOO 1
500 < ADT < 1000 2
1000 < ADT < 5000 1
ADT > 5000 (Not an interstate or arterial) 5
ADT > 5000 (interstate or arterial) 10
REFERENCES
Agrawal, A., Kawaguchi, A., and Qian, G., (2009) “Bridge Element Deterioration Rates: Phase I Reprot,” New
York State Department of Transportation, Project # C-01-51, Albany, NY.
Alampalli, S., (2001) “Peer Review to Enhance State Transportation Research Program,” Leadership and man-
agement in Engineering, ASCE, Vol 1, No. 2, pp. 27–28.
Alampalli, S. and Ettouney, M., (2002) Proceedings of the Workshop on Engineering Structural Health,
Sponsored by NYS-DOT, Albany, NY.
Arya, S., O’Neill, M., and Pincus, G., (1979), Design of Structures and Foundations for Vibrating Machines,
Gulf Publishing Company, Houston, TX.
Aven, T., (2003) Foundations of Risk Analysis, John Wiley and Sons, West Sussex, England.
Barker, R. and Puckett, J., (2007) Design of Highway Bridges: An LRFD Approach, John Wiley Sons,
Hoboken, NJ.
Bellman, R., (1964) Perturbation Techniques in Mathematics, Physics and Engineering, Rinehart and Winston,
New York, NY.
Benjamin, J., and Cornell, A., (1970) Probability, Statistics, and Decision for Civil Engineers, McGraw-Hill,
New York, NY.
Biggs, J., (1964) Intorduction to Structural Dynamics, McGraw Hill, New York, NY.
Chakraborty, S. and DeWolf, J., (2006) “Development and Implementation of a Continuous Strain Monitoring
on Multi-Girder Composite Steel Bridge,” Journal of Bridge Engineering, ASCE, Vol 11, No. 6,
pp. 753–762.
Crandal, S. and Mark, W., (1963) Random Vibrations, Academic Press, New York.
Daddazio, R. and Ettouney, M., (1987) “The Effect of Uncertain Boundary Conditions on the Large Displacement
Elasto-Plastic Response of Impulsively Loaded Structures,” Proceedings, ASCE Engineering Mechanics
6th Specialty Conference, Buffalo, NY.
Daddazio, R. and Ettouney, M., (1990) “Boundary Element Methods in Probabilistic Acoustic Radiation
Problems,” Journal of Vibration and Acoustic, ASME, Vol 112, pp. 556–560.
Daddazio, R. and Ettouney, M., (1991) “Probabilistic Acoustic Analysis,” in Boundary Element Methods
in Acoustics, Editors: Ciskowski, R. D. and Brebbia, C. A. Computational Mechanics Publications,
Southampton, UK, pp. 95–108.
Elishakoff, A., (1983) Probabilistic Methods to the Theory of Structures, John Wiley & Sons.
Elsayed, A. E. and Ettouney, M. M., (1994) “Perturbation Analysis of Linear Programming Problems with
Random Parameters,” Computers and Operation Research, Vol 21, No 2, pp. 211–224.
Ettouney, M., Benaroya, H., and Wright, J., (1989a) “Probabilistic Boundary Element Methods (PBEM),” in
Computational Mechanics of Probabilistic and Reliability Analysis, Editors: Liu, W. K. and Belytschko,
T. Elmpress International, Lausanne, Switzerland, pp. 141–166.
Ettouney, M., Benaroya, H., and Wright, J., (1989b) “Boundary Element Methods in Probabilistic Structural
Analysis (PBEM),” Applied Mathematics Modeling, Vol 13, No. 7, pp. 432–441.
Ettouney, M., Daddazio, R. and Abboud, N. (1993) “Stochastic Analysis of Soil Dynamics,” in Computational
Stochastic Mechanics, Editors: Cheng, A. H-D. and Yang, C. Y. Computational Mechanics Publications,
Southampton, UK.
FEMA, (2009). “Handbook for Rapid Visual Screening of Buildings to Evaluate Terrorism Risk,” FEMA 455,
Federal Emergency Management Agency, FEMA Publication, Washington, DC.
Farrar, C. R. and Doebling, S.W., (1998) Damage Detection - Field Application to a Large Structure. Proceedings
of the Modal Analysis & Testing, NATO-Advanced Study Institute, Sesimbra, Portugal.
Johnson, P. A., and Dock, D. A., (1998) “Probabilistic bridge scour estimates,” Journal of Hydraulic
Engineering, ASCE, 124(7), 750–755.
Johnson, P. A., and Niezgoda, S. L., (2003) “Risk-based method for selecting scour countermeasures,” Journal
of Hydraulic Engineering, ASCE, 130(2), 121–128.
Johnson, P. A., and Simon, A., (1997) “Reliability of bridge foundations in modified channels,” Journal of
Hydraulic Engineering, ASCE, 123(7), 648–651.
Koller, G., (2000) Risk Modeling for Determining Value and Decision Making, Chapman & Hall/CRC Press,
Boca Raton, FL.
Lawler, G. F., (1995) Introduction to Stochastic Processes, Chapman & Hall/CRC, New York.
Melcher, R., (2002) Structural Reliability Analysis and Prediction, John Wiley & Sons, New York, NY.
Moses, F., (2001) “Calibration of Load Factors for LRFR Bridge Evaluation,” National Cooperative Highway
Research Program, NCHRP Project 454, Washington, DC.
Murdock, M. B., (2009) “Comparative Load Rating Study Under LRFR And LFR Methodologies For Alabama
Highway Bridges,” Auburn University, Auburn, AL.
Neubeck, K., (2004) Practical Reliability Analysis, Prentice-Hall, NJ.
Newnan, D. G. and Lavelle, J. P., (1988) Engineering Economic Analysis, 7th Edition, Engineering Press,
Austin, TX
NHTS, (2001), “Highlights of the 2001 National Household Travel Survey,” Bureau of Transportation Statistics,
U.S. Department of Transportation, Washington, DC.
Park, C. S., (1997) Contemporary Engineering Economics, 2nd Edition, Addison- Wesley, Menlo Park, CA.
Pendat, J., and Piersol, A., (2010) Random Data: Analysis and Measurement Procedures, Wiley, New York.
Riggs, J. L., Bedworth, D. D., and Randhawa, S. U., (1996) Engineering Economics, 4th Edition, McGraw-Hill,
New York.
Ruegg, R. T. and Marshall, H. E., (1990) Building Economics: Theory and Practice, Van Nostrand Reinhold,
New York.
Rausand, M. and Hoyland, A., (2004) System Reliability Theory: Models, Statistical Methods, and Applications,
Wiley, NJ.
Ravindra, M. and Galambos, T., (1978) “Load and Resistance Factor Design for Steel,” Journal of Structural
Division, ASCE, Vol 104, no 9, pp. 1337–1353.
Schlaifer, R. S., (1969) Analysis of Decision under Uncertainty, McGraw-Hill, New York, NY.
Stein, S. and Sedmera, K., (2006) “Risk-Based Management Guidelines for Scour at Bridges with Unknown
Foundations,” National Cooperative Highway Research Program, NCHRP Project 24-25, Washington, DC.
Timoshenko, S. and Gere, J., (1961) Theory of Elastic Stability, McGraw-Hill, New York.
Trueman, R. (1977) An Introduction to Quantitative Methods for Decision Making, Holt, Rinehart, and Winston,
New York, NY.
Von Neumann, J. and Morgenstern, O., (1944) Theory of Games and Economics, Princeton University Press,
Princeton, NJ.
Appendix Unit Conversion
Table A.1 contains the units of major engineering metrics used in this volume and the relationships
between the units in both SI and US Customary units. The conversion factors are rounded off to a
reasonable decimal point.
TABLE A.1
Conversion Table between SI and US Customary Units
To Convert From . . . to . . .
Engineering Metric SI US Customary Divide by
Linear acceleration m/s2 in/s2 0.0254
m/s2 ft/s2 0.3048
Angle radian radian 1
Angular acceleration radian/s2 radian/s2 1
Angular velocity radian/s radian/s 1
Area m2 in2 0.000645
m2 ft2 0.0929
mm2 in2 645.2
Energy J ft.lb 1.356
Force N or kg.m/s2 lbf 4.448
kN kipf 4.448
Frequency Hz Hz 1
Impulse N.s or kg.m/s lb . s 4.448
Distance/length m in 0.0254
mm in 25.40
m ft 0.3048
km mi 1.609
Mass kg lb (mass) 0.4536
G oz 28.35
kg slug 14.59
kg ton 907.2
Moment of a force N.m lb.in 0.112867
Power W or J/s ft.lbf/s 1.356
W or J/s hp 745.7
Pressure or stress Pa psi or lb/in2 6895
Pa lb/ft2 47.88
kPa psi or lb/in2 6.895
Work J lb.ft 1.356
Velocity m/s in/s 0.0254
m/s mi/h 0.447
km/h mi/h 1.609
Volume: solid m3 in3 1.64E-05
m3 ft3 0.02832
cm3 in3 16.39
continued
583
584 Appendix
TABLE A.1 (continued)

Conversion Table between SI and US Customary Units
To Convert From . . . to . . .
Engineering Metric SI US Customary Divide by
Volume: liquid l gal (US) 3.785
l qt 0.946
Time s s 1
Notes: m = meter, s = seconds, kg = kilogram, J = joule, kN = kilo newton, qt = quart, N = newton,

Hz = hertz, W = watt, Pa = pascal, l = liter, kipf = kilo pounds (force), in = inches,
lb = pounds (force), gal = gallon, ft = feet, mm = millimeter, oz = ounce, mi = mile,
km = kilo meter, g = gram.
Civil Engineering
Mohammed M. Ettouney • Sreenivas Alampalli
Continually increasing demands on infrastructures mean that maintenance and

renewal require timely, appropriate action that maximizes benefits while minimizing
cost. To be as well informed as possible, decision makers must have an optimal
understanding of an infrastructure’s condition—what it is now, and what it is
expected to be in the future. Written by two highly respected engineers, the first
volume, Infrastructure Health in Civil Engineering: Theory and Components,
integrates the decision making concept into theoretical and practical issues.
It includes:
• An overview of the infrastructure health in civil engineering (IHCE) and

associated theories
• In-depth description of the four components of IHCE: measurements,
structural identification, damage identification, and decision making
• Discussion of how IHCE and asset management are applied
• Exploration of analogies between structural and human health
Built to correspond to the ideas presented in its companion volume, Applications

and Management, this is an invaluable guide to optimized, cost-saving methods
that will help readers meet safety specifications for new projects, as well as the
aging infrastructure at great risk of failure.
2040

Health in Civil Engineering

Uploaded by

Copyright:

Available Formats

Health in Civil Engineering

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Health in Civil Engineering

Uploaded by

Copyright:

Available Formats

Mohammed M.

Ettouney • Sreenivas Alampalli

© 2012 by Taylor & Francis Group, LLC

No claim to original U.S. Government works

International Standard Book Number-13: 978-1-4200-0375-8 (eBook - PDF)

and the CRC Press Web site at

Chapter 2 Elements of Structural Health in Civil Engineering (SHCE)..................................... 41

Chapter 3 Lives and Deaths of Infrastructure............................................................................. 81

3.2.1.2 Design Phase........................................................................ 83

Chapter 4 Theories and Principles of IHCE.............................................................................. 133

4.2.4.1 General.............................................................................. 141

Chapter 5 Sensors and Infrastructure Health............................................................................ 175

5.1.2 This Chapter...................................................................................... 175

5.5.5 Distributed (Long) FOS: Brillouin Optical Time Domain

5.9.1 Overview........................................................................................... 252

Chapter 6 Structural Identification (STRID)............................................................................. 281

6.3.1.4 Frequency Response Function (FRF)................................ 294

6.6.2 Energy-Based Methods..................................................................... 337

Chapter 7 Damage Identification (DMID)................................................................................. 373

7.3.2.2 Objectives of DMID.......................................................... 378

7.5.2.4 Important Characteristics of AE Wave.............................. 425

7.8 Damage Identification in SHM....................................................................... 481

Chapter 8 Decision Making in IHCE........................................................................................ 495

8.4.4 More on Traditional Decision Making Methods............................... 517

8.7.1.2 Use of SHM in Markov Processes: Paint States of a

PART I: OVERVIEW AND THEORIES OF IHCE

Infrastructure Health in Civil Engineering (IHCE)

PART I PART II PART III Part IV

3—Lives and 10—Bridge life

PART II: COMPONENTS OF IHCE

Aspects of sensors in IHCE

Categories Case studies

FIGURE 0.2 Sensor topics.

1.2 GENERAL CONCEPTS OF INFRASTRUCTURES HEALTH

1.2.2 What Is Structural Health?

1.2.3 Infrastructure Health in Civil Engineering

(b) Appropriate input

FIGURE 1.10 (a) Passive (b) active SHM experiments.

Sensor Acoustic emission Sensor

Data acquisition and

(b) Lamb waves from

Sensor Lamb waves from Actuator

Data acquisition and

Data acquisition and

1.2.3.2 Objectives of IHCE

1.2.4 The 95% Solution!

1.3 HISTORY OF STRUCTURAL HEALTH EFFORTS

1.3.2 Fulfilling Defined Objectives

1.3.3 Operations and Management

1.4 NEED FOR INFRASTRUCTURE HEALTH EFFORTS

Bridge health monitoring

Service/operation Maintenance Repair-retrofit

Intrinsic force Damage

Diagnostic Diagnostic Diagnostic

Damage indices: Displacement (flexibility) influence coefficients, strain

Views of the principal structural system

822′ 1644′ 822′

innovative practices/materials, better understand performance, and effects of factors influencing

1.4.2 Specific Needs

Infrastructure life span

1.9.2 Treatment of SHCE Multidisciplinary Needs in