A Comparison of Nondestructive Testing Backcalculation Techniques For Rigid and Flexible Pavements

University of Arkansas, Fayetteville
ScholarWorks@UARK
Civil Engineering Undergraduate Honors Theses Civil Engineering
5-2008
A Comparison of Nondestructive Testing

Backcalculation Techniques for Rigid and Flexible
Pavements
Trenton Ellis
University of Arkansas, Fayetteville
Follow this and additional works at: http://scholarworks.uark.edu/cveguht

Part of the Civil Engineering Commons, and the Structural Engineering Commons
Recommended Citation
Ellis, Trenton, "A Comparison of Nondestructive Testing Backcalculation Techniques for Rigid and Flexible Pavements" (2008). Civil
Engineering Undergraduate Honors Theses. 1.
http://scholarworks.uark.edu/cveguht/1
This Thesis is brought to you for free and open access by the Civil Engineering at ScholarWorks@UARK. It has been accepted for inclusion in Civil
Engineering Undergraduate Honors Theses by an authorized administrator of ScholarWorks@UARK. For more information, please contact
scholar@uark.edu.
1
A COMPARISON OF NONDESTRUCTIVE TESTING BACKCALCULATION
TECHNIQUES FOR RIGID AND FLEXIBLE PAVEMENTS
2
A COMPARISON OF NONDESTRUCTIVE TESTING BACKCALCULATION
TECHNIQUES FOR RIGID AND FLEXIBLE PAVEMENTS
A thesis submitted in partial fulfillment

of the requirements for Honors Studies in Civil Engineering
By
Trenton Blake Ellis
May 2008
University of Arkansas
3
A COMPARISON OF NONDESTRUCTIVE TESTING TECHNIQUES FOR RIGID
AND FLEXIBLE PAVEMENTS
Trenton B. Ellis
ABSTRACT
In designing new pavements, engineers rely on many different measures to characterize

the average traffic, climate and soil conditions of the region. Of those, soil conditions are
the most elusive, but are also the most crucial for designing a pavement of proper
thickness and stiffness. This problem is compounded when the design is for an overlay
instead of a new pavement. For overlay designs, engineers require a quantitative
characterization of the strength of the existing pavement as well as the underlying soil.
Especially for new Mechanistic-Empirical design procedures, direct measures and/or
estimates of pavement stiffness are essential inputs. A variety of methods exist for
measuring the appropriate strength properties for soil, concrete and asphalt. The most
traditional method – regardless of the material – is to retrieve a sample of the material
from the field and use laboratory tests to determine the strength of the material, which is
then considered representative for other materials near the location from which the
sample was taken. The problem with this approach is that, in the field, construction
materials experience a confining pressure from the other materials that surround them;
even when carefully removed, the strength properties measured in the laboratory are not
truly representative of the strength of the materials in the field. Nondestructive testing
(NDT) refers to a collection of methods that are used to estimate material properties
without removing or otherwise damaging the material. Generally, NDT is faster, cheaper
and less intrusive to the traveling public. Two NDT methods, the falling weight
deflectometer (FWD) and the spectral analysis of surface waves (SASW), have both
become popular for their ability to estimate in-situ (in-place) stiffness properties of
pavement materials. Each method measures a different local response (i.e. deflection or
wave propagation) caused by a specific load. The measurement of the local response is
then used to “backcalculate” the strength property that is desired. The appropriate
backcalculation technique is different for each test, and for each test the appropriate
technique varies according to the type of material on which the test is being performed
and for which stiffness is sought. This study proposes to assess the validity of several
backcalculation procedures associated with these two NDT methods and to compare their
relative appropriateness in backcalculating pavement stiffness.
4
This thesis is approved for
recommendation to the
Undergraduate Council
Thesis Director:
_______________________________
Kevin D. Hall, PhD.
Thesis Committee:
_______________________________
Norman D. Dennis, PhD.
_______________________________
Robert P. Elliott, PhD.
5
THESIS DUPLICATION RELEASE
I hereby authorize the University of Arkansas Libraries to duplicate this thesis when
needed for research and/or scholarship.
Agreed __________________________
Refused _________________________
6
ACKNOWLEDGEMENTS
I would like to thank my thesis director Dr. Kevin D. Hall for his support and
guidance throughout this project. Thanks are also extended to the remaining members of
my thesis committee, Dr. Robert P. Elliott and Dr. Norman D. Dennis. Their reviews and
advice throughout the writing process have been extremely helpful. In addition, I could
not have completed this project without the help that I have received from Dr. Brady
Cox, Dr. Nam Tran, Meagan Berlau, Sarah Tamayo, and Sutapa Hazra. Finally, I would
like to thank Dr. Anastasios Ioannides for making backcalculation software available to
this project.
7
TABLE OF CONTENTS
Section Page
1. INTRODUCTION ...................................................................................................... 1
1.1. Background .......................................................................................................... 1
1.2. Problem Statement ............................................................................................... 2
1.3. Objectives ............................................................................................................. 3
1.4. Experimental Plan ................................................................................................ 3
1.4.1. Task 1: Literature Review ............................................................................. 3
1.4.2. Task 2: Data Analysis ................................................................................... 4
1.4.2.1. Software Acquisition ............................................................................. 4
1.4.2.2. Collection of Field Data ........................................................................ 5
1.4.2.3. Perform Backcalculations ...................................................................... 5
1.4.2.4. Document Results from Backcalculations ............................................. 6
1.4.3. Task 3: Report Findings ................................................................................ 6
2. LITERATURE REVIEW ........................................................................................... 7
2.1. Nondestructive Testing Background .................................................................... 7
2.2. Falling Weight Deflectometer .............................................................................. 9
2.2.1. Theoretical Background ................................................................................ 9
2.2.2. Field Procedures.......................................................................................... 12
2.2.3. Backcalculation Methods ............................................................................ 17
2.2.3.1. 1993 AASHTO Method for Estimating Mr from Deflection Tests ..... 19
2.2.3.2. ILLI-BACK ......................................................................................... 23
2.2.3.3. NUS-BACK for Rigid Pavements ....................................................... 25
2.2.3.4. AREA Method for Rigid Pavements ................................................... 33
vi
2.2.3.5. Best-Fit Procedure ............................................................................... 38
2.2.3.6. LTPP Forward Calculation .................................................................. 40
2.2.3.7. MODULUS ......................................................................................... 45
2.2.3.8. ELMOD 5 ............................................................................................ 50
2.2.3.9. BAKFAA ............................................................................................. 52
2.2.3.10. DIPLO-DEF ........................................................................................ 52
2.3. Spectral Analysis of Surface Waves .................................................................. 56
2.3.1. Theoretical Background .............................................................................. 56
2.3.2. Field Procedures.......................................................................................... 59
2.3.3. Backcalculation Procedure.......................................................................... 68
2.4. Case Studies: Comparisons of Backcalculation Procedures .............................. 76
2.4.1. Oregon: FWD vs. Laboratory for HMA and Base Course Materials ......... 77
2.4.2. LTPP Database: Comparison of FWD Backcalculation Methods .............. 81
3. DATA ANALYSIS................................................................................................... 86
4. RESULTS ................................................................................................................. 92
4.1. SASW Results .................................................................................................... 92
4.2. NUS-BACK Results ........................................................................................... 92
4.3. AREA72 Results.................................................................................................. 95
4.4. LTPP Forward Calculation Results .................................................................... 96
4.5. BAKFAA Results............................................................................................... 96
4.6. ELMOD Results ................................................................................................. 98
4.7. Comparisons ....................................................................................................... 98
5. DISCUSSION ......................................................................................................... 109
vii
5.1. SASW Results .................................................................................................. 109
5.2. NUS-BACK ..................................................................................................... 109
5.3. AREA72 ............................................................................................................ 110
5.4. LTPP Forward Calculation............................................................................... 111
5.5. BAKFAA ......................................................................................................... 111
5.6. ELMOD ............................................................................................................ 112
5.7. NUS-BACK Variations .................................................................................... 113
5.8. NUS-BACK vs. AREA72 ................................................................................. 114
5.9. ELMOD Variations .......................................................................................... 115
5.10. Other Comparisons for Estimating Subgrade Stiffness ................................ 116
5.11. Other Comparisons for Estimating PCC Stiffness ....................................... 117
6. CLOSING ............................................................................................................... 118
REFERENCES ............................................................................................................... 121
APPENDIX A Raw Data from SASW Testing ......................................................... 125
APPENDIX B Raw Data from FWD Testing............................................................ 134
APPENDIX C Example SASW Inversion ................................................................. 139
LIST OF FIGURES
FIGURE 1 Dynatest FWD (11) ..................................................................................... 13
FIGURE 2 KUAB FWD (11)........................................................................................ 14
FIGURE 3 Effect of Existing Distresses On FWD Testing (12) .................................. 16
FIGURE 4 Sensor Location Effect On Sampling Depth (13) ....................................... 16
FIGURE 5 Graphical NUS-BACK Dense Liquid Solution for ℓ.................................. 31
FIGURE 6 Graphical NUS-BACK Elastic Solid Solution for ℓ ................................... 31
viii
FIGURE 7 Graphical NUS-BACK Solutions for Fk and FE ......................................... 32
FIGURE 8 AREA36 Deflection Sensor Layout ............................................................. 34
FIGURE 9 AREAS60 Deflection Sensor Layout ........................................................... 34
FIGURE 10 AREA72 Deflection Sensor Layout ......................................................... 35
FIGURE 11 Curves Relating AREAX to General AREA Parameter (12)................... 36
FIGURE 12 AREA12 Deflection Sensor Layout ......................................................... 43
FIGURE 13 WESDEF Relationship Between Deflection and Modulus (19) ............. 54
FIGURE 14 Idealized Layered Half Space ................................................................. 58
FIGURE 15 SASW Configuration .............................................................................. 60
FIGURE 16 HP 35670A Digital Signal Analyzer (37) ............................................... 61
FIGURE 17 Common Receivers Midpoint Geometry for SASW (38) ....................... 63
FIGURE 18 Common Source Geometry for SASW (38) ........................................... 64
FIGURE 19 Typical Coherence Function (35) ............................................................ 66
FIGURE 20 Typical Wrapped Cross-Power Spectrum (35) ....................................... 67
FIGURE 21 Strains Under Repeated Loads (19) ........................................................ 71
FIGURE 22 Normalized Shear Modulus Reduction Curve (19) ................................. 73
FIGURE 23 HMA Moduli for Rufus Project (47)....................................................... 79
FIGURE 24 HMA Moduli for Centennial Project (47) ............................................... 79
FIGURE 25 Base Course Moduli for Rufus Project (47) ............................................ 80
FIGURE 26 Base Course Moduli for Centennial Project (47) .................................... 80
FIGURE 27 Comparison of Backcalculated and Measured k Values (22) ................. 82
FIGURE 28 Comparison of Backcalculated and Measured EPCC Values for JPC (22)83
ix
FIGURE 29 Comparison of Backcalculated and Measured EPCC Values for CRCP (22)
83
FIGURE 30 Errors of Backcalculated k Values (22)................................................... 84
FIGURE 31 Errors of Backcalculated EPCC Values for JPC (22) ................................ 84
FIGURE 32 Errors of Backcalculated EPCC Values for CRCP (22) ............................ 85
FIGURE 33 Spatial Distribution of Soil Types in Nashville, AR (48) ....................... 87
FIGURE 34 Backcalulated EPCC from NUS-BACK ................................................... 93
FIGURE 35 Backcalculated Soil Stiffness from NUS-BACK .................................... 94
FIGURE 36 Backcalculated EPCC from AREA72 ...................................................... 95
FIGURE 37 Backcalculated k from AREA72 .............................................................. 95
FIGURE 38 Mr and EComp from LTTP Forward Calculation ...................................... 96
FIGURE 39 ES and EPCC from BAKFAA ................................................................... 97
FIGURE 40 ES and EPCC from ELMOD ...................................................................... 98
FIGURE 41 NUS-BACK: D3-D1 vs. D7-D4 for EPCC .................................................. 99
FIGURE 42 NUS-BACK: D3-D1 vs. D7-D4 for k, Es .................................................. 99
FIGURE 43 NUS-BACK: Dense Liquid vs. Elastic Solid for EPCC.......................... 100
FIGURE 44 NUS-BACK: Dense Liquid vs. Elastic Solid for k & ES ...................... 100
FIGURE 45 NUS-BACK (Elastic Solid) vs. AREA72 for EPCC ................................ 101
FIGURE 46 NUS-BACK Average (Elastic Solid) vs. AREA72 for EPCC.................. 101
FIGURE 47 NUS-BACK D7-D4 (Dense Liquid) vs. AREA72 for k .......................... 102
FIGURE 48 ELMOD: Radius of Curvature vs. Basin-Matching .............................. 102
FIGURE 49 NUS-BACK D7-D4 (DL) k vs. BAKFAA ES ........................................ 103
FIGURE 50 NUS-BACK D7-D4 (DL) k vs. LTPP Mr............................................... 103
x
FIGURE 51 NUS-BACK D7-D4 (DL) k vs. ELMOD ES .......................................... 103
FIGURE 52 AREA72 k vs. BAKFAA ES................................................................... 104
FIGURE 53 AREA72 k vs. LTPP Mr ......................................................................... 104
FIGURE 54 AREA72 k vs. ELMOD ES ..................................................................... 104
FIGURE 55 BAKFAA ES vs. LTPP Mr .................................................................... 105
FIGURE 56 BAKFAA ES vs. ELMOD ES ................................................................ 105
FIGURE 57 LTPP Mr vs. ELMOD ES ...................................................................... 106
FIGURE 58 AREA72 EPCC vs. BAKFAA EPCC ......................................................... 106
FIGURE 59 AREA72 EPCC vs. LTPP Ecomp................................................................ 107
FIGURE 60 AREA72 EPCC vs. ELMOD EPCC............................................................ 107
FIGURE 61 BAKFAA EPCC vs. LTPP EComp ............................................................ 107
FIGURE 62 BAKFAA EPCC vs. ELMOD EPCC ......................................................... 108
FIGURE 63 LTTP EComp vs. ELMOD EPCC .............................................................. 108
xi
1. INTRODUCTION
1.1. Background
Maintaining structurally and functionally sound roads is an expensive endeavor in the
United States and around the world. Much of the current transportation infrastructure in
the United States was built in the 1960s and 1970s, and is reaching the end of its design
life. There is a need for roadway rehabilitation on a large scale, and maintenance
measures are most effective when pavement distresses are detected early on and treated
properly before more costly damage occurs.
Laboratory tests can be used for detecting structural deficiencies in subgrade soil
and pavement layers. For these tests, a sample of the material in question is excavated
from the roadway and taken to the laboratory for examination. Care must be taken when
collecting samples in the field because the stresses applied to the specimen during
excavation can alter the material properties under investigation. Indeed, removing the
specimen from the field will inevitably result in measured material properties that stray
from those that exist in the field.
A more favorable method is to conduct tests in the field without removing
materials from the roadway. Methods collectively known as nondestructive tests (NDT)
have become increasingly popular in recent years because of their speed, low cost, and
reliability (1). By testing the materials in-situ (in-place), field conditions can be
maintained and represented in the results. The falling weight deflectometer (FWD) and
the spectral analysis of surface waves (SASW) are two types of NDT commonly used for
pavement analysis.
1
1.2. Problem Statement
Nondestructive testing techniques such as FWD and SASW have gained popularity
because of their ease of use as well as the potential accuracy of their results. As these
tools have increased in use, a number of different techniques for reducing their field data
have been proposed. The process of reducing field data from an NDT into more pertinent
information about the roadway is commonly referred to as backcalculation.
A growing number of backcalculation methods exist to accommodate variables
such as the type of test being used, the underlying theory used to model the pavement
structure, and the amount of computing power that is available. These accommodations,
while appropriate, have resulted in a large collection of backcalculation methods that do
not lend themselves to a simple comparison. One straight-forward manner of comparison
is to perform field tests on a single section of pavement and use several different
techniques to reduce the common data set. Then, the ability and/or accuracy of each
method to backcalculate the engineering properties of the pavement can be gauged by
comparing the backcalculation results to known parameters of the pavement structure on
which the field tests were conducted.
If the number and variety of backcalculation procedures that are compared is large
enough, then the results of the comparison could be used to gauge the effect of factors
such as type of test being used, underlying theory used to model the pavement structure,
and/or available computing power upon the accuracy of the final results. This study is
limited to backcalculation procedures for FWD and SASW tests, but methods for
reducing data from both rigid and flexible pavements will be considered for each type of
test. Backcalculation procedures to be considered include closed form solutions via
2
spreadsheet, database software, and iterative software. The set of data reduction methods
that is used for this study will be chosen so that it is large enough for a worthwhile
comparison between the types of tests being used (FWD and SASW) and between the
different types of backcalculation software inherent to each method.
1.3. Objectives
The objective of this research is to compare various backcalculation techniques for both
FWD and SASW testing of rigid and flexible pavements. This has been accomplished by
completing the following related objectives:
• Identify commonly used backcalculation procedures for data from FWD
and SASW field tests.
• Assess the ability and/or accuracy of several different procedures for
evaluating structural properties of pavement layers (both individually and
as a single effective layer) and underlying subgrade by using several
different methods to reduce the data from a common set of field tests.
• Compare the results from the backcalculations to known properties of the
pavement on which the field tests were run. Report the relative ease and
accuracy of each of the techniques considered at reducing field test data.
1.4. Experimental Plan
The objectives of this study will be achieved by completing the following tasks:
1.4.1. Task 1: Literature Review
3
A comprehensive literature review of FWD and SASW tests will be undertaken to gain
an understanding of the backing theories, field procedures, and common backcalculation
techniques for each test. It is not the goal of this research to provide a detailed
explanation for the backing theory of each test. Instead, this report will describe the
theories in sufficient detail so that the reader may gain a working knowledge of how
and/or why the corresponding backcalculation procedures work.
Similarly, this report will describe the manner in which field tests are performed
for FWD and SASW testing. This section of the literature review is included so that
readers who are unfamiliar with the equipment and/or procedures involved with each test
may become sufficiently aware of each so that references to these things later on the
report will not be a source of confusion.
Finally, for FWD and SASW, a number of backcalculation procedures will be
described in detail. The two goals of this portion of the literature review are to introduce
the reader to the most common types of backcalculation procedures that are currently in
use for FWD and SASW tests and to provide an explanation of how these procedures
work.
1.4.2. Task 2: Data Analysis
1.4.2.1. Software Acquisition
Based upon results from the literature review, a number of different types of software
will be sought for use in the comparison of this study. Software for the study will be
limited by what is available for use at either no cost or a modest price. In addition, some
procedures employing database software may not lend themselves to this study for the
4
reason that the software is available, but the database that the software uses is not.
Computer processing speed has been an issue to consider with some procedures in the
past, but initial thoughts are that modern computers possess such high processing speeds
that this will not be a limiting issue for the present study.
It is anticipated that some closed-form solution backcalculation techniques will be
most efficiently utilized via spreadsheet applications such as Microsoft Excel. For this
reason, a number of spreadsheets will likely be programmed with applicable closed-form
solutions for the comparison of this study. These spreadsheets, along with the software
that is obtained according to the preceding paragraph, will comprise the backcalculation
techniques studied during the comparison.
1.4.2.2. Collection of Field Data
Field data from actual FWD and SASW tests will be obtained in order to compare the
backcalculation methods of the study. Raw data will be collected from past and ongoing
research at the University of Arkansas as well as records from the Arkansas State
Highway and Transportation Department (AHTD).
1.4.2.3. Perform Backcalculations
After a sufficient number of spreadsheets and other software have been collected for
backcalculation, the raw data from the FWD and SASW tests will be used to test the
ability and/or accuracy of each applicable backcalculation method to determine various
properties of the pavement structure. This process will be repeated as many times as is
necessary to ensure that the backcalculation techniques are being used correctly.
5
1.4.2.4. Document Results from Backcalculations
Once the backcalculations have been performed using all of the techniques under
investigation, the calculated pavement properties will be organized in a manner
conducive to comparison. The results obtained using each technique will be
accompanied by notes that outline the general instructions for using the method as well as
potential difficulties specific to that method. The end result of this step will be a table of
raw output data and user notes for each of the backcalculation techniques investigated in
this study.
1.4.3. Task 3: Report Findings
The final phase involves the actual comparison of the results from the backcalculations.
This comparison may range from broad observations such as user friendliness and
outlying results to more advanced observations from statistical analysis. After all
comparisons have been drawn, a concise summary of the findings will be presented that
identifies the most pertinent findings from this study as well as unanswered questions that
may be relevant for future studies.
6
2. LITERATURE REVIEW
2.1. Nondestructive Testing Background
Nondestructive testing (NDT) refers to a collection of field testing methods that facilitate
the in-situ (in-place) testing of materials instead of conventional laboratory testing. NDT
is attractive to highway engineers for several reasons. Testing subgrade and pavement
materials in-situ preserves the confining stresses that are present in the field, whereas
laboratory testing requires the removal of a sample and consequently introduces a change
in confining stress. In addition, NDT requires significantly less time and money to
estimate the same material properties that would otherwise be measured in the laboratory.
Thus, NDT may provide a relatively fast, inexpensive and accurate alternative to
laboratory testing.
A great deal of the current highway system in the United States was constructed
in the 1960s and the 1970s, and much of that system is beginning to reach the end of its
design life. In order to maintain a functional transportation infrastructure, maintenance is
and will be required on a large scale. The cost of this maintenance effort can be
minimized if highway engineers are able to obtain a more detailed knowledge of the
structural condition of the aging highway system. Laboratory tests are the conventional
method for measuring the structural properties and stiffness of pavements, but the time
and expense associated with laboratory testing allows only a minimal amount of locations
to be assessed. NDT, on the other hand, can estimate the same material properties as
laboratory tests, but at a much larger number of locations and at a fraction of the cost.
Integrating NDT into standard highway evaluation techniques has and will continue to
7
improve the extent to which the structural capacity of existing highways can be
measured.
NDT has been practiced in several different forms for a number of decades, but
two types of NDT have emerged as the front-runners in recent years: the falling weight
deflectometer (FWD) and the spectral analysis of surface waves (SASW). Currently, the
FWD is the type of NDT that is preferred by most agencies. The FWD involves dropping
a weight on the ground and “backcalculating” the stiffness of the pavement and/or
subgrade from the measured pavement surface deflections. If the pavement system
consists of multiple layers, a rough idea of the thickness of each layer must be known
prior to backcalculation so that iterative bounds may be set.
SASW is also emerging as a valuable tool for highway engineers. This method is
also used to determine the stiffness of pavement and/or subgrade from an impulse load.
The main differences are that SASW can also determine the thickness of multiple layers,
and that SASW estimates these properties from measurements of wave propagation. The
techniques for measuring the necessary characteristics of wave propagation and for
converting those measurements to a usable output (inversion) have been advanced greatly
in the past two decades. Advancements in technique and computing power have and will
continue to make SASW a considerable alternative and/or compliment to the FWD.
Both FWD and SASW testing are promising candidates for providing estimates of
pavement and subgrade stiffnesses, which are the most important input parameters used
by engineers to plan maintenance of existing roadways. These tests will become
increasingly valuable with time as mechanistic-empirical (M-E) design becomes more
popular in the United States. M-E design software requires extensive inputs such as
8
subgrade stiffness that will need to be closely approximated for each location that uses
the software. NDT will facilitate the development of much less expensive and more
representative catalogs of local subgrade properties for M-E design of new pavements.
2.2. Falling Weight Deflectometer
The falling weight deflectometer (FWD) is an “impulse load” variety of NDT, and it is
currently the most popular type of NDT that is practiced by highway officials in the
United States. The FWD operates much in the way that its name implies. Some amount
of mass is dropped from a specific height in order to impart a particular magnitude of
impulse load onto a strategic location on the pavement surface. That stress causes a
strain through the pavement and subgrade, and several sensors are positioned on the
pavement surface at precise horizontal distances from the location of the impulse load to
measure the vertical deflection at each point. The vertical deflections measured by each
sensor define a single deflection basin that is caused by the falling mass. The stiffness of
the pavement and/or subgrade is then backcalculated from the magnitude of the imparted
load and the size of the measured deflection basin.
2.2.1. Theoretical Background
The type of backcalculation that is used to estimate pavement and/or subgrade stiffness
depends upon the assumptions that are made and also upon the type of theoretical model
that is used to idealize the pavement structure.
The most significant assumption that is commonly made for FWD
backcalculation is that despite the dynamic nature of the load, the pavement materials
9
react as though the load is purely static. That is, the loading rate is assumed to be low
enough so that pavement materials experience no dynamic effects such as damping. This
assumption is often made in order to simplify the mechanics that are utilized for
backcalculation. If, however, a significant time lapse occurs between the peak stress and
the peak deflection, the assumption of static loading is not appropriate. Generally, static
behavior is a more realistic assumption for pavements with stiff concrete or thick asphalt
layers than for pavements comprised of a thin asphalt layer (2).
The mechanical behavior of soil is affected by a variety of factors, including (i)
the shape, size and mechanical properties of the individual soil particles, (ii) the
configuration of the soil structure, (iii) the intergranular stresses and stress history, and
(iv) the presence of soil moisture, the degree of saturation and the soil permeability.
Considering the myriad factors that influence the mechanical behavior of soil, it is no
surprise that naturally occurring soils exhibit non-linear, irreversible and time-dependent
stress-strain behavior. Furthermore, naturally occurring soil masses almost always
exhibit anisotropic and non-homogeneous material properties. Any analysis that
accounts for all of these factors - if possible - would be far too rigorous and time-
consuming to be practical. Therefore, several simplifying assumptions other than static
behavior are also commonplace for FWD backcalculation (3).
As the science of theoretical soil mechanics has advanced in recent decades, a
number of prudent assumptions for idealizing in-situ soil behavior have emerged. In
general, a prudent assumption is one that simplifies analytical computations while
introducing a minimal amount of error in the corresponding results. While the models
and assumptions used for different backcalculation methods may vary slightly, the most
10
common and important assumptions for any idealization of naturally occurring soils are
as follows (4):
1. The soil is assumed to be an elastic, homogeneous, isotropic medium.
2. The soil is assumed to act as an ideal half-space, bound by a horizontal ground
surface.
3. The material obeys Hooke’s law of proportionality between stress and strain.
4. The deformations of the soil medium are so small that they can be neglected in
setting up equilibrium conditions.
The previous assumptions idealize the soil structure on a micro-level, but there is
also a need to model the soil structure on a macro-level. This can be accomplished by
employing one of two common models. The simpler of the two methods was introduced
by Winkler, and it is commonly known as the dense liquid model. For this model, the
subgrade is assumed to act as a field of springs that do not interact with each other. That
is, each spring’s movement is only affected by the load directly above it, and springs do
not share any of their load with neighboring springs. The spring constant, k, is called the
coefficient of subgrade reaction. Winkler’s method is the most commonly used model
because of its computational simplicity. More information on the development of this
model is provided elsewhere (5).
Most backcalculation procedures are derived not from Winkler’s original model,
but instead from amendments to that model that were proposed by Westergaard (6).
Westergaard accepted the soil behavior that was proposed by Winkler, but advanced the
model by including a rigid plate at the top of the half-space to represent a rigid pavement.
Westergaard’s studies provide a simplified means to find the maximum tensile stress in a
11
concrete slab (plate) under point loads at mid-slab, slab edge, and slab corner. The
methods used by Westergaard are not well documented, but a detailed review of his
results and their applicability is available (7).
The dense liquid idealization proposed by Winkler does have its shortcomings.
For instance, soils with any degree of cohesion do not obey the assumptions of a Winkler
foundation. That is, the soil directly underneath a point load is not the only soil that
experiences a vertical deflection. Instead, a somewhat continuous deflection basin forms
around the point load. This type of deflection basin was originally described by
Boussinesq (8), and it leads to the second type of model for macro-level behavior of soils:
the elastic continuum model. This model, in contrast to the Winkler model, assumes
continuous behavior throughout the soil half-space. Both two- and three-dimensional
analyses may be performed using the Boussinesq distribution and plate theory to model
pavement-soil systems. The elastic continuum (also called elastic solid or ES) model
provides a more accurate representation of in-situ behavior, but the accompanying
analysis involves mathematics that are often too complex to be practical (3).
Backcalculation procedures that do employ the ES model are mostly based off the works
of Panc (9) and Losberg (10), which model an infinitely long slab on an elastic
foundation.
2.2.2. Field Procedures
FWD apparatus are usually trailer-mounted and carried by a vehicle with onboard
computer processors. Two commonly used commercial FWDs, the Dynatest and KUAB
FWDs, are shown in Figures 1 and 2, respectively (11). During field testing, the FWD is
12
driven to the testing location and positioned so that the load plate is above the location to
be tested. For the purposes of determining pavement layer moduli, rigid (or asphalt-
overlaid rigid) pavements are tested at slab interiors and flexible pavements are tested in
the outer wheel path of the outer lane. When flexible (or asphalt-overlaid) pavements are
tested, it is important to record the temperature of the asphalt pavement at least 3 times
during a full day of testing. The temperature measurements can be used to calibrate the
measurements for a more accurate backcalculation. In every case, tests are conducted
every 100 to 500 ft. In general, longer intervals are more appropriate for newer
pavements and vice versa (11).
FIGURE 1 Dynatest FWD (11)
13
FIGURE 2 KUAB FWD (11)
Once the FWD is in place, the loading begins. After the load plate and deflection
sensors are hydraulically lowered to the pavement, a seating load is applied to ensure
proper placement of the load plate. The magnitude of the pre-load varies for different
equipment, but is usually between 8 to 18% of the maximum impulse load. Once the pre-
load is performed, a series of impulse loads are performed that mimic the design load of
the roadway being tested. In other words, a larger peak load is applied for roads with
predominantly truck traffic than for roads with relatively light traffic loads. The
magnitude of the load that is delivered by the FWD is altered by changing the amount of
mass that is dropped and by changing the height from which it is dropped. A load cell at
the load plate reports the delivered load. Peak loads and deflections are stored in the
14
onboard computer for backcalculation, the equipment is hydraulically lifted back to the
trailer, and the entire apparatus is moved to the next testing location (12).
At least three different loadings are usually performed at each testing location.
This manner of testing allows highway officials to identify any non-linear response of the
foundation, and it also helps to ensure that at least one of the deflection basins is
satisfactory for backcalculation. Typical target loads for FWD testing on highway
pavements are 6000, 9000, and 12000 pounds (11).
At least two drops should be performed at each target load to check for any
permanent deformation under the loading plate. If the difference in the deflections
caused by two loads of the same magnitude at the same location is greater than 3% for
any sensor, the apparatus should be moved and the applied force should be reduced (1).
It is worth noting that existing pavement distresses can and do affect the
deflections that are measured during FWD testing. Figure 3 shows an example in which
alligator cracking causes higher deflection readings. An effort should be made to avoid
FWD testing in the vicinity of existing distresses insofar as possible.
It is well known that deflections measured closer to the point load are more
indicative of pavement stiffness, whereas deflections measured further horizontally
outward from the point load are more indicative of subgrade stiffness. Consider Figure 4
(13). The dotted line represents the boundary of the zone of stress, and beyond a radial
distance from the load of a3e all deflections at the surface are due to stress that occurs
only in the subgrade layer. This simplistic explanation assumes Winkler behavior, but it
is sufficient to explain why the outer sensors are indicative of subgrade behavior.
15
FIGURE 3 Effect of Existing Distresses On FWD Testing (12)
FIGURE 4 Sensor Location Effect On Sampling Depth (13)
16
It is desirable to place at least one sensor far enough away from the load plate so
that its measured deflection can be used to estimate subgrade stiffness. Seven deflection
sensors are commonly used, including one sensor at the middle of the load plate. The
most common configuring for the seven-sensor system on highway pavements is 0, 8, 12,
18, 24, 36, and 60 inches from the center of the load plate. Whatever configuration is
used, it is important to place one sensor further than 36 inches from the center of the load
plate to account for subgrade stiffness (11).
2.2.3. Backcalculation Methods
Backcalculation methods vary according to a number of factors such as the theory that
was used to develop the method and the type of pavement system that is under
investigation. Rigid pavements are typically the simplest types of pavement structures to
analyze with NDT. Portland cement concrete (PCC) layer(s) are typically assumed to act
as rigid plates, which can allow for a closed-form solution.
A more common pavement structure is a 3-layered system consisting of a PCC
and a base layer that are underlain by subgrade soil. The addition of the base layer adds
complexity to the backcalculation procedure, but closed-form solutions are still possible.
In fact, two classes of closed-form solutions are available for 3-layer rigid pavement
systems, depending on the type of interface that exists between the PCC and base layers
(14). If the two layers are completely bonded, the parallel axis theorem is used to
combine layer thicknesses and stiffnesses for backcalculation. Otherwise, if the two
layers are not bonded together, the composite layer of PCC and base is simply defined by
the PCC thickness and the sum of the layer stiffnesses.
17
Most backcalculation procedures are too sensitive to adhere to general rules or
procedures, but research has shown a general trend for 3-layer rigid pavement systems
(15). First, neglecting the presence of a base layer tends to overestimate the stiffness of
the overlying PCC layer. Second, assuming a full bond between the PCC and base layers
usually underestimates the stiffness of the base material. Slab curling and moisture
warping can cause a rigid slab to maintain a full bond at mid-slab, where deflection
testing occurs, while exhibiting no bond at the edges. From these observations, a
decision could be made to model the PCC-base interface as unbonded unless field
conditions clearly indicate a full bond.
Flexible pavements involve a higher degree of complexity for backcalculations
than rigid pavements. This is because flexible pavements are viscoelastic materials,
meaning that they act both as an elastic solid and as a viscous liquid. The viscous
behavior causes a variation in HMA stiffness with changes in temperature. Thus, the
effects of temperature gradients cause more computational complications with flexible
pavements (13). In addition, flexible pavement deflections occur not only because of
movement of the underlying subgrade, but also because of deflections within the hot-mix
asphalt (HMA) layer(s).
Typical backcalculation procedures for flexible pavements are not closed-form
and require some form of computer software to carry out the iterative, finite element,
and/or database-search type of computations. Some of the more common types of
software will be discussed in terms of their computational approach, appropriateness for
various pavement types, and general availability.
18
The ultimate goal of deflection testing is to provide material characterizations that
can be used as inputs for the design of new pavements or for overlays of existing
pavements. The primary material characterization parameters have not changed
drastically from 1993 AASHTO Pavement Design Guide to the Mechanistic-Empirical
Pavement Design Guide (MEPDG). Subgrade soils for flexible pavements are
characterized by Resilient Modulus (Mr), and those for rigid pavements by the
Coefficient of Subgrade Reaction (k). Both of these quantities can be estimated from
FWD testing (16).
The characterization of pavement stiffness has changed in the new mechanistic-
empirical procedures. The structural number (SN) was used to indicate material stiffness
in the 1993 guide, but it has been eliminated in the MEPDG software. If pavement
designers choose to use the 1993 guidelines to obtain baseline designs, FWD results can
be used to estimate the effective structural number (SNeff) or effective depth (Deff). SNeff
is used to characterize the structural capacity of a flexible pavement as a whole, and Deff
does the same for rigid pavements. MEPDG software, on the other hand, uses the
material thicknesses and stiffnesses that are obtained from FWD backcalculation directly.
2.2.3.1. 1993 AASHTO Method for Estimating Mr from Deflection Tests
The 1993 AASHTO specifications for pavement design offers the most straight-forward
method for estimating the resilient modulus (Mr) of subgrade soils (16). To use this
method, however, it is necessary to iterate the subgrade resilient modulus and total
pavement thickness. The benefit is that Mr can be estimated using a single deflection
measurement.
19
To begin, it is necessary to identify the radial distance from the load plate at
which deflections will be outside the influence of the pavement layers. Equation 1 is
used to estimate this value, which is termed a3e in Figure 4. Note that Equation 1 requires
an estimate of the subgrade resilient modulus (Mr) as well as the effective modulus of all
pavement layers (EP) above the subgrade. Table 1 provides a list of reasonable guesses
and ranges that can be used in this equation or any other application that requires an
estimate of material moduli (13). Equation 2 is used to find a value for EP to be used in
Equation 1. Finally, after ae has been calculated, Mr is found via Equation 3. If the
pavement is comprised of a single PCC layer that rests directly on the subgrade, Equation
4 can also be used to calculate a k-value from the previously estimated MR value. Note,
however, that Equation 4 has been shown to consistently underestimate k values (17, 18).
The obtained value for Mr must also be multiplied by a reduction factor not less
than 0.33 for fine-grained, stress-dependent soils. While FWD loads do emulate the
loads that are considered for AASHTO design standards, the amount of soil that exists
between the load impact and the deflection sensor tends to absorb some of the stress
before it reaches the deflection sensor. For fine-grained, stress-dependent soils, Mr
decreases as the magnitude of strain increases. Thus, the reduction factor is in place to
correct the falsely inflated estimate of Mr that occurs because of some amount of stress
that is absorbed by soil that lies between the load plate and the deflection sensor (16).
2
⎛ E ⎞
a e = a + ⎜ D ⋅ 3 P ⎟ …….…………..…………..Eq. 1
2
⎜ M r ⎟⎠
⎝
20
⎧ ⎡ ⎤⎫
⎪ ⎢ ⎥⎪
⎪ ⎢ 1 ⎥⎪
⎪ ⎢1 − 2 ⎥⎪
⎪ ⎢ ⎛D⎞ ⎥⎪
⎪ ⎢ 1 + ⎜ ⎟ ⎥⎪
⎪ 1 ⎣ ⎝a⎠ ⎦⎪
d 0 = 1.5 pa ⋅ ⎨ + ⎬ ……………….Eq. 2
2 EP
⎪ ⎛D E ⎞ ⎪
⎪ M R 1 + ⎜⎜ ⋅ 3 P ⎟⎟ ⎪
⎪ ⎝a Mr ⎠ ⎪
⎪ ⎪
⎪ ⎪
⎪⎩ ⎪⎭
0.24 P
Mr = …………………..…….……….Eq. 3
dX ⋅r
Mr
k= ………………..….……..………..Eq. 4
19.4
Where: ae = Radius of stress bulb at the subgrade-pavement interface (in.)
a = FWD load plate radius (in.)
D = Total thickness of pavement layers above the subgrade (in.)
EP = Effective modulus of all pavement layers (psi)
d0 = Deflection measured at the center of the load plate (in.)
p = FWD load plate pressure (psi)
a = FWD load plate radius (in.)
MR = Subgrade resilient modulus (psi)
P = Applied load (lb)
dX = Deflection at a distance X from the center of the load (in.)
r = Distance from center of load (in.)
k = Composite modulus of subgrade reaction (psi/in.)
21
Table 1 Reasonable Estimates and Ranges for Material Moduli (13)
Estimate of Moduli Range of Moduli

Material Type
(ksi) (ksi)
Crushed Stone, Gravel or Slag

Bases 50.0 10.0 to 150.0
Subbases 30.0 10.0 to 100.0
Gravel or Soil-Agg. Mix, Coarse

Bases 30.0 10.0 to 100.0
Subbases 20.0 5.0 to 80.0
Sand
Bases 20.0 5.0 to 80.0
Gravel or Soil-Agg. Mix, Fine

Bases 20.0 5.0 to 80.0
Lime-Treated Base/Subbase
fc’ < 250 psi 30.0 5.0 to 100.0
250 - 500 psi 50.0 10.0 to 150.0
> 500 psi 70.0 15.0 to 200.0
Asphalt-Treated Base/Subbase
fc’ < 300 psi 100.0 10.0 to 300.0
300 - 800 psi 150.0 25.0 to 800.0
> 800 psi 200.0 50.0 to 1500.0
Cement-Treated Base/Subbase
fc’ < 750 psi 400.0 50.0 to 1500.0
750 - 1250 psi 1000.0 100.0 to 3000.0
> 1250 psi 1500.0 150.0 to 4000.0
Fractured PCC 500.0 100.0 to 3000.0
PCC 3500 3000.0 to 8000.0
HMA 400 250 to 500
22
2.2.3.2. ILLI-BACK
ILLI-BACK was the first closed-form backcalculation method to be developed for rigid
pavements (19). Very similar to the AREA method (see Section 2.2.3.3), ILLI-BACK
relies on the unique relationship between radius of relative stiffness to a ratio of
deflection measurements. To better explain this relationship, consider Westergaard’s
equations for normalized deflections both under a load (d0) and at some distance from the
load (di), presented below as Equations 5 and 6, respectively. If the radius of the load
plate (a) is known and the deflection (w) is measured, then an important observation can
be drawn. The normalized deflection, whether it be directly underneath the load or at
some distance from the load, is a function of only one unknown: the radius of relative
stiffness (ℓ).
w0 kl 2 ⎛a⎞
d0 = = f ⎜ ⎟ ………………………………..Eq. 5
P ⎝l⎠
wi kl 2
di = = f i (l ) …………………………..……Eq. 6
P
ILLI-BACK works by the relationship that is found by taking the ratio of the two
above equations. The result, given below as Equation 7, indicates that the ratio of any
two measured deflections may be defined as a function of ℓ. This means that a
theoretically sound relationship can be developed to calculate ℓ that is representative of
how much one location deflects relative to another from the same load. Thus, for an
FWD setup, the relative deflections between the different sensors are representative of a
unique value of ℓ for the pavement system. Furthermore, instead of limiting the
relationship to the ratio of two deflections, the AREA parameter can be used to provide
an even more reliable development of ℓ.
23
d 1 w1 f
= = 1 = f (l ) …………………….……..Eq. 7
d 2 w2 f0
Though this derivation was presented in terms of a dense liquid foundation, the
relationship between relative deflections and radius of relative stiffness also exists for
elastic solid foundation models. ILLI-BACK can use either of the two models to analyze
a pavement structure. After ℓ is determined, and depending on the subgrade model that is
specified, the k value or Es is calculated using Equation 8 or 9, respectively. Finally, the
stiffness of the PCC layer is calculated using Equation 10 or 11.
Pd i
k= ………………………….……..Eq. 8
l 2 wi
Es =
(
2 1 − v s Pd i
2
)
……………………………Eq. 9
lwi
E PCC =
(
12 1 − v PCC kl 4
2
) …………….…………Eq. 10
h3
E PCC =
(
6 1 − v PCC E s l 3
2
) ………………….……Eq. 11
(1 − v s
2
)h 3
Where: k = Coefficient of subgrade reaction
P = Impact load
d0, di = Normalized loading under the center of loading and at
some distance from the load, respectively
ℓ = Radius of relative stiffness
w0, wi = Measured vertical deflections under the center of loading
and at some distance from the load, respectively
Es = Young’s modulus of the subgrade
24
νs = Poisson’s ratio of the subgrade
νPCC = Poisson’s ratio of the PCC
ILLI-BACK is available as a DOS based computer program, and it has been
developed for both 4-sensor arrangements (AREA36) and 7-sensor arrangements (both
AREA72 and AREAS60) (20). If necessary, Table 2 can be used to find reasonable
estimates of Poisson’s Ratio to use during manual backcalculation (13).
Table 2 Typical Poisson’s Ratio Values for Pavement Materials (13)
Material Type Poisson’s Ratio

HMA
E > 500 ksi 0.30
E < 500 ksi 0.35
PCC 0.15
Stabilized Base/Subbase
Lime 0.20
Cement 0.20
Asphalt 0.35
Other (stabilized subgrade) 0.35
Other (fractured PCC) 0.30
Granular Base/Subbase 0.35
Cohesive Subgrade 0.45
Cohesionless Subgrade 0.35
2.2.3.3. NUS-BACK for Rigid Pavements
The simplest type of pavement system for backcalculation is a PCC layer resting on a
homogeneous subgrade. In this case, a closed-form solution may be developed using
either the dense liquid or elastic solid subgrade models (21). This simple method
requires only two measured deflections to estimate the stiffness of the rigid pavement and
25
the subgrade. Since FWD tests provide more than two deflections, NUS-BACK can be
used for certain types of quality control in addition to just stiffness estimation. Since the
deflections from any two sensors may be used as input data, faulty sensors can be easily
spotted and their data discarded. The variance in estimated stiffnesses from different
sensor combinations can also be used to test the robustness of the backcalculation theory.
At least one study has shown that using the fourth and seventh sensors is most accurate
for estimating subgrade stiffness, and using the first and third sensors is most accurate for
estimating pavement stiffness (22).
Simplifying assumptions for NUS-BACK include the ignoring of effects from
load transfer across joints and dynamic loading as well as slab curling. These
assumptions allow deflections in the two-layer system (PCC and subgrade) to be defined
solely in terms of the elastic modulus of the slab (EPCC) and a measure of stiffness for the
subgrade. For the dense liquid model this would be the coefficient of subgrade reaction
(k) and for the elastic solid model this would be the elastic modulus of the subgrade (ES).
The situation is described in terms of two equations that may be simultaneously
solved to provide unique values for radius of relative stiffness (ℓ) and either k or ES for
two measured values of deflection (Dm1 and Dm2). Equations 12 and 13 describe this
setup for a dense liquid model, and Equations 15 and 16 do the same for an elastic solid
model.
After unique solutions have been found for radius of relative stiffness and
subgrade stiffness, the stiffness of the PCC pavement is solved directly from the
definition of the radius of relative stiffness. Equation 14 or 17 is used to calculate EPCC
for the dense liquid or elastic solid model, respectively.
26
For dense liquid-modeled subgrades:
FK (l K , r1 ) ……………………………..Eq. 12
P
Dm1 =
kπa 2
FK (l K , r2 ) ……………………………..Eq. 13
P
Dm 2 =
kπa 2
1
⎛ EPCC hPCC ⎞ 4
l K = ⎜⎜ ⎟ ………………….………….Eq. 14
⎝ 12 (
1 − μ PCC
2
K ⎟
⎠ )
For elastic solid-modeled subgrades:
Dm1 =
(
P 1 − μS
2
)
FE (l E , r1 ) …………………..……….Eq. 15
ES l E
Dm 2 =
(
P 1 − μS
2
)
FE (l E , r2 ) ………………………….Eq. 16
ES l E
( )⎞⎟
1
⎛ E h 3 1− μ 2 3
l E = ⎜⎜ PCC PCC 2 S
(
⎝ 6 1 − μ PCC ES ) ⎟ ……………..…………….Eq. 17
⎠
Where: Dm1,2 = Measured deflections

P = Applied load
K = Modulus of subgrade reaction
a = Radius of loaded area
ES = Subgrade elastic modulus
μS = Poisson’s ratio of the subgrade
EPCC = Elastic modulus of the pavement slab
hPCC = Slab thickness
r1,2 = Horizontal distances of points 1 and 2 from the center of
loaded area
ℓ1,2 = Radii of relative stiffness
27
NUS-BACK exists in the form of a computer program, but it is also possible to go
through the same process graphically. For the graphical procedure the two deflections
are labeled Dmi and Dmj so that Dmi is greater than or equal to Dmj. The first step is to
compute the ratio of the two deflections, Dmj/Dmi. Depending on the preferred analysis
(dense liquid or elastic solid) the appropriate radius of relative stiffness (ℓk or ℓE) is taken
from the chart in Figure 5 or 6. Then, instead of solving for Fk or FE analytically, it is
taken from a chart (Figure 7) using the previously determined value of ℓk or ℓE. This is
done for ri and rj, and the rest of the procedure is performed in the previously described
manner using Equations 12 – 17. Regardless of whether the software- or graphical-based
method is used, stiffness values can be estimated using many different combinations
deflection readings to check the accuracy of the sensors and the model itself.
It is common practice to include a granular base layer between PCC slabs and the
subgrade to facilitate drainage. Accordingly, there are also two NUS-BACK procedures
for analyzing a two-layer rigid pavement. A computer program called NUS-BACK3 may
be used, or the two upper pavement layers may be combined into a single equivalent
layer so that the three-layer pavement can be analyzed as a two-layer pavement. If the
interface between PCC and base material is considered bonded, Equations 18 and 19 are
used to define an equivalent pavement layer; otherwise, if the interface is completely
unbonded, Equation 20 is used. For these analyses, there are three unknowns: modulus
of the concrete slab (EPCC), modulus of the base layer (Eb), and stiffness of the subgrade
(ES or k). In order to find unique solutions for all three parameters, an initial input must
be provided. The most convenient additional parameter is the modular ratio (β) of EB to
EPCC. Typical values are provided in Table 3 (13).
28
E eq =
4 E PCC
heq
3
[h 0
3
− (h0 − hPCC ) +
3
] 4Eb
heq
3
[(h PCC
3 3
]
+ hb − h0 ) − (h0 − h PCC ) ………..Eq. 18
2 2
E h + 2 Eb hb hPCC + Eb hb
h0 = PCC PCC …………………..…..Eq. 19
2(E PCC hPCC + Eb hb )
3 3
⎛h ⎞ ⎛ ⎞
Eeq = E PCC ⎜ PCC ⎟ + Eb ⎜ hb ⎟ ………………………..Eq. 20
⎜ h ⎟ ⎜h ⎟
⎝ eq ⎠ ⎝ eq ⎠
Where:
Eeq, EPCC, Eb = Young’s modulus of equivalent pavement layer, top
PCC layer, and base layer, respectively
heq, hPCC, hb = Thickness of equivalent pavement layer, top PCC
layer, and base layer, respectively
h-0 = Depth to neutral axis of equivalent pavement layer
The method of solution for NUS-BACK3 is similar to the two-layer solution that
was previously discussed, except that it only uses the elastic solid subgrade model and it
utilizes the work of Burmister (23) as well as Panc (9). The three-layer system is
analyzed by simultaneously solving Equations 21, 22, and 23 for unique values of ℓ, c
and Eb. It is noted that the 3-layer system requires three deflection measurements instead
of just two in order to find a unique solution.
29
Table 3 Typical Modular Ratios (β) for EBase / EPCC (13)
Base Type β Base Type β
Hot-mixed, hot-laid asphalt concrete (AC), dense

10 Sand asphalt 50
graded
Hot-mixed, hot-laid AC, open graded 15 PCC, prestressed 1
Jointed plain concrete pavement 1 PCC, fiber reinforced 1
Jointed reinforced concrete pavement 1 Recycled JPCP 100
Continuously reinforced concrete pavement 1 Recycled JRCP 100
Plant mix (emulsified asphalt) material, cold-laid 20 Recycled CRCP 100
Plant mix (cutback asphalt) material, cold-laid 20 Crushed Rock 150
Recycled AC, hot-laid, central plant mix 10 Gravel, uncrushed 200
Recycled AC, cold-laid, central plant mix 15 Crushed stone 150
Recycled AC, cold-laid, mixed-in-place 15 Crushed gravel 175
Recycled AC, heater scarification/recompaction 15 Crushed slag 175
Fine-grained soils: lime-treated soil 100 Sand 250
Fine-grained soils: cement-treated soil 50 Hot-mixed AC 15
Bituminous treated subgrade soil 100 Asphalt-treated mix 50
Soil-aggregate mixture (predominantly fine-

400 Econocrete 4
grained)
Soil-aggregate mixture (predominantly coarse-

250 Cement-treated soil 50
grained)
Dense-graded, hot-laid, central plant mix AC 10 Lean concrete 2
Dense-graded, cold-laid, central plant mix AC 15 Cracked/seated PCC 25
Dense-graded, cold-laid, mixed-in-place AC 15 Treatment: lime 100
Open-graded, hot-laid, central plant mix AC 15 Treatment: lime, fly ash 150
Open-graded, cold-laid, central plant mix AC 15 Treatment: bitumen 100
Pozzolanic-aggregate
Open-graded, cold-laid, mixed-in-place AC 15 100
mixture
30
FIGURE 5 Graphical NUS-BACK Dense Liquid Solution for ℓ
FIGURE 6 Graphical NUS-BACK Elastic Solid Solution for ℓ
31
FIGURE 7 Graphical NUS-BACK Solutions for Fk and FE
d m1 =
(
2 1 − μb P
2
)
F (l, c, r1 ) …………………………..Eq. 21
πaEb
dm2 =
(
2 1 − μb P
2
)
F (l, c, r2 ) ………………….……….Eq. 22
πaEb
d m3 =
(
2 1 − μb P
2
)
F (l, c, r3 ) ……………….…………Eq. 23
πaEb
c=
(1 + μb )ES ………………………………….Eq. 24
(1 + μS )Eb
( )⎞⎟
1
⎛ EC hC 3 1 − μb 2 3
l = ⎜⎜
(
⎝ 6 1 − μc Eb
2
) ⎟ …………………….……….Eq. 25
⎠
32
In the above equations, μb is the Poisson ratio of the base; c and ℓ are given by
Equations 24 and 25, respectively. The definition of all other variables is consistent with
the definitions given for two-layer NUS-BACK equations. Unfortunately, a graphical
solution for NUS-BACK3 is not available.
2.2.3.4. AREA Method for Rigid Pavements
Another graphical backcalculation process has been developed to account for preferential
FWD sensor spacings in which the radius of the load plate is 5.9 inches (12). The first
step for this graphical procedure is to calculate the AREA of the deflection basin. The
AREA parameter is the area of the deflection basin normalized to some height to negate
the effect of load magnitude. Figures 8, 9, and 10 illustrate three common layouts for
arranging sensors during FWD testing, AREA36, AREAS60, and AREA72. Equations 26,
27, and 28 are used to calculate the AREA parameter for the AREA36, AREAS60, and
AREA72 layouts, respectively. In each of the figures below, the term “dx” refers to the
vertical deflection that is measured at a horizontal distance “x” from the center of the
load plate.
The AREAS60 layout is generally considered to provide the best measure of
curvature in the basin, but the following graphical procedure may be used for AREA36,
AREAS60, or AREA72. After the AREAX parameter is calculated using the appropriate
equation, the next step is to calculate the radius of relative stiffness. This may be done
empirically using Equation 29 and the coefficients from Table 4.
⎛ d d d ⎞
AREA36 = 6 ⋅ ⎜⎜1 + 2 ⋅ 12 + 2 ⋅ 24 + 36 ⎟⎟ …………………Eq. 26
⎝ d0 d0 d0 ⎠
33
12 in. 12 in. 12 in.
Load Plate
d36
d24
d12
d0
FIGURE 8 AREA36 Deflection Sensor Layout
d8 d d d d d
AREAS 60 = 4 + 6 ⋅ + 5 ⋅ 12 + 6 ⋅ 18 + 9 ⋅ 24 + 18 ⋅ 48 + 12 ⋅ 60 ……...Eq. 27
d0 d0 d0 d0 d0 d0
8 in. 4 in. 6 in. 6 in. 24 in. 12 in.

Load Plate
d60
d48
d18 d24
d8 d12
d0
FIGURE 9 AREAS60 Deflection Sensor Layout
⎛ d d d d d d ⎞
AREA72 = 6 ⋅ ⎜⎜1 + 2 ⋅ 12 + 2 ⋅ 24 + 2 ⋅ 36 + 2 ⋅ 48 + 2 ⋅ 60 + 72 ⎟⎟ ……..Eq. 28
⎝ d0 d0 d0 d0 d0 d0 ⎠
34
12 in. 12 in. 12 in. 12 in. 12 in. 12 in.
Load Plate
d72
d60
d48
d36
d24
d12
d0
1
⎡ ⎛ k1 − AREA ⎞ ⎤ k4
⎢ ln⎜⎜ ⎟⎟ ⎥
l=⎢ ⎝ k 2 ⎠⎥ ………………..……………….Eq. 29
⎢ − k3 ⎥
⎢ ⎥
⎢⎣ ⎥⎦
Table 4 Empirical Coefficients for Calculating Radius of Relative Stiffness (12)
k1 k2 k3 1/k4
AREA36 36 1812.597 2.559 4.387
AREAS60 60 289.708 0.698 2.566
AREA72 72 242.385 0.442 2.205
The radius of relative stiffness is then used to match the AREAX parameter with
the general AREA parameter using an appropriate curve from Figure 11 that corresponds
to the appropriate sensor layout.
35
One last empirical quantity, a nondimensional deflection coefficient (dr*), has to
be calculated before the pavement and subgrade stiffness can be estimated. The
deflection coefficient is calculated using Equation 30 and the empirical coefficients from
Table 5. After dr* is found, the stiffness of the subgrade and pavement are estimated for
each individual deflection measurement using Equations 31 and 32, respectively. The
arithmetic average of those values is taken to be representative of the entire basin.
FIGURE 11 Curves Relating AREAX to General AREA Parameter (12)
d r* = a ⋅ e − b⋅e⋅
− c⋅l
……………………………….Eq. 30
Where: dr* = Nondimensional deflection coefficient for the deflection at a
radial distance r from the load plate
a, b, c = Coefficients from Table 2
ℓ = Radius of relative stiffness
36
Table 5 Empirical Coefficients for Calculating Deflection Coefficient (12)
dX* a b c
d0* 0.12450 0.14707 0.07565
d8* 0.12323 0.46911 0.07209
d12* 0.12188 0.79432 0.07074
d18* 0.11933 1.38363 0.06909
d24* 0.11634 2.06115 0.06775
d36* 0.10960 3.62187 0.06568
d48* 0.10241 5.41549 0.06402
d60* 0.09521 7.41241 0.06255
d72* 0.08822 9.59399 0.06118
The k value obtained from the AREA method is representative of dynamic
loading conditions (kd), but the 1993 AASHTO Pavement Design Guide uses the static k
value (ks). Therefore, the obtained value must be converted to a static k value. Robust
methods for converting between the two measurements are available, but AASHTO
recommends that Equation 33 be used for a simplified conversion (16, 24).
P ⋅ d X*
k= ……………………………….Eq. 31
d X ⋅ l2
E PCC =
( )
12 ⋅ l 4 ⋅ 1 − μ 2 ⋅ k
…….…………………Eq. 32
h3
ks = kd ÷ 2 ……………….……………..Eq. 33
Where: k = Modulus of subgrade reaction (psi/in.)
P = Applied load (lb)
37
dX* = Nondimensional deflection coefficient for deflection at
radial distance r from load
dX = Measured deflection at radial distance r from the load (in.)
ℓ = Radius of relative stiffness (in.)
EPCC = PCC elastic modulus (psi)
μ = Poisson’s ratio of PCC
h = Slab thickness (in.)
2.2.3.5. Best-Fit Procedure
The preceding graphical AREA approach is one of two common AREA
backcalculation methods that utilize Westergaard’s solution for the interior loading of a
horizontally infinite plate on a dense liquid foundation. The other main method is called
the Best Fit procedure (15). The graphical AREA method matched a theoretical
deflection basin to the measured deflection basin. That is, layer properties were found
that provided the best match between the deflection basins, as a whole. The Best-Fit
Procedure, however, matches deflections point-by-point. The point-by-point criteria is
less stringent than fitting an entire deflection basin, and the Best-Fit Method actually
provides a better match between measured and calculated deflection basins.
Consider Equation 34, which describes the distribution of deflections (w) at a
distance (r) from the center of a vertical load evenly distributed across a circular area
with radius a. The Best Fit method finds the combination of k and ℓk that causes
Equation 34 to predict a deflection basin closest to the actual measured basin. Then, an
algorithm uses those two values to find the combination of k and EPCC that again causes
38
the best match between calculated and measured deflection basins. The mechanics of the
algorithm involve minimizing an error function, the mathematics of which are described
elsewhere (15).
w(r ) = ⋅ f (r , l k ) ………………………………Eq. 34
p
k
Where: w = Vertical deflection
r = Radial distance from center of load
p = Applied pressure
k = Coefficient of subgrade reaction
ℓk = radius of relative stiffness
The Best-Fit Method can also be used to analyze two-layered rigid pavements
such as a PCC slab that is underlain by a granular base (21, 25). The analyzing of such a
pavement is facilitated by the same method that was described for NUS-BACK in
Section 2.2.3.3.
The main advantage to using the Best Fit method is that the theoretically derived
deflection basin is made to be as close as possible to the measured deflection basin. The
biggest problem with this approach is that there exist too many combinations of
parameters (ℓk, k, and EPCC) that will make the Best Fit procedure appear to work even if
the results are incorrect and/or unreasonable. Best Fit procedures are performed by
software, and most of the time the computer program will request “seed” values and
acceptable ranges of necessary parameters to ensure that the values chosen for the “best
fit” are at least reasonable. There is no guarantee, however, that the values will be
correct.
39
2.2.3.6. LTPP Forward Calculation
The Long-Term Pavement Performance (LTPP) program has documented a closed-form
procedure termed “forward calculation” for estimating the stiffness of subgrade soils as
well as rigid and flexible pavements on a closed-form basis (14). Forward calculation
spreadsheets are available in both English and Metric format, free of charge, from LTPP.
Note, however, that forward calculation is only intended to be used for generating
approximate estimations of pavement and subgrade stiffnesses. Those estimates can then
be used to check the reasonableness of other estimates from more complex
backcalculations.
Forward Calculation uses two deflection values to approximate the upper
subgrade stiffness. The deflection under the center of the FWD load plate (d0) and the
deflection at an offset where the deflection is approximately one-half of d0 are used as
inputs into a model based on the Hogg model, which idealizes the pavement structure as a
thin plate on an elastic foundation (26, 27). The model used for Forward Calculation can
be altered to reflect 3 distinct cases: (1) infinite elastic foundation and μ = 0.5; (2) and (3)
finite elastic foundation with thickness assumed equal to 10 times the characteristic
length of the deflection basin and μ = 0.4 and 0.5, respectively. Case 2 is strongly
recommended.
The upper subgrade stiffness (E0) is approximated from the sequence of Equations
35 - 38 in conjunction with Table 6.
E0 = I ⋅
(1 + μ0 )(3 − 4μ0 ) ⎛ S0 ⎞⎛⎜ p ⎞⎟ …..............…………..Eq. 35
⎜ ⎟⎜
2(1 − μ0 ) ⎝ S ⎠⎝ Δ 0l ⎟⎠
40
r50 = r
(1α ) 1
β
−B
…………………..………..Eq. 36
1/ β
⎡ 1 ⎛ Δ 0 ⎞⎤
⎢ ⎜⎜ − 1⎟⎟⎥ −B
⎢⎣ α ⎝ Δ r ⎠⎥⎦
l = y0
r50
2
[
+ ( y0 r50 ) − 4mar50
2
]
1
2
………………………..Eq. 37
< 0.2 , then l = ( y0 − 0.2m )r50

a
if
l
⎛ S0 ⎞ ⎛a ⎞
⎜ ⎟ = 1 − m⎜ − 0.2 ⎟ ………….…....…….……..Eq. 38
⎝S ⎠ ⎝l ⎠
a S0
if < 0.2 , then = 1.0
l S
Where: E0 = Upper subgrade modulus
μ0 = Poisson’s ratio for subgrade
S0 = Theoretical point load stiffness
S = Pavement stiffness = p/Δ0 (area loading)
p = Applied load
Δ0 = Deflection at center of load plate
Δr = Deflection at offset distance r
r = Distance from center of load plate
r50 = Offset distance where Δr/Δ0 = 0.5
l = Characteristic length
h = Thickness of subgrade
I = Influence factor
α, β, B = Curve fitting coefficients
y0, m = Characteristic length coefficients
m(bar) = Stiffness ratio coefficient
41
Table 6 Forward Calculation Subgrade Stiffness Coefficients (14)
Case 1 2 3
I 0.1925 0.1689 0.1614
Range Δr/Δ0 All Values > 0.43 > 0.70
A 0.3210 0.3804 0.4065
R50 = ƒ(Δr/Δ0) β 1.7117 1.8246 1.6890
B 0 0 0
Range Δr/Δ0 < 0.43 < 0.70
A 4.3795E-4 2.6947E-3
R50 = ƒ(Δr/Δ0)
β 4.9903 4.5663
B 3 2
y0 0.527 0.603 0.642

l= ƒ(r50,a)
m 0.098 0.108 0.125
(S/S0) = ƒ(a/l) m(bar) 0.185 0.208 0.219
The LTTP Forward Calculation procedure applies the previously described 4-
sensor AREA36 method for rigid pavements as well as a revised AREA12 method for
flexible pavements to estimate the stiffness of bound surface course. AREA12 is defined
by Equation 39 and the setup shown in Figure 12.
⎡ ⎛ d ⎞ ⎛ d ⎞⎤
A12 = 2 ⋅ ⎢2 + 3⎜⎜ 8 ⎟⎟ + ⎜⎜ 12 ⎟⎟⎥ …………………………Eq. 39
⎣⎢ ⎝ d 0 ⎠ ⎝ d 0 ⎠⎦⎥
42
8 in. 4 in.
Load Plate
d12
d8
d0
The Forward Calculation procedure is based on the AREA of a given pavement
system relative to that of an idealized system whose layers all have identical values for
stiffness and Poisson’s ratio. If continuity is assumed between layers of the ideal
pavement system, AREA36 and AREA12 will reach minimum values of 11.04 and 6.85,
respectively. This holds for any value of stiffness and/or Poisson’s ratio so long as it is
held constant through the different layers. A more realistic pavement structure consists
of a surface course that is stiffer than the underlying pavement and/or subgrade layers.
The larger the difference between the stiffness of the wearing course and underlying
layers, the larger the AREA will be. Equations 40 and 41 are used to calculate the AREA
factors for rigid and flexible pavements, respectively. These area factors essentially give
a measure of how much a given pavement increases the AREA from its theoretical
minimum value. Finally, Equations 42 and 43 are used to estimate the stiffness of the
upper PCC or HMA layers, respectively.
43
1.79
⎡ ⎤
⎢ k2 − 1 ⎥
AFPCC =⎢ ⎥ ………….…………..Eq. 40
⎢ k 2 − ⎛⎜ AREA36 ⎞⎟ ⎥
⎣⎢ ⎝ k1 ⎠ ⎦⎥
Where: AFPCC = AREA factor for PCC
k1 = 11.04 (theoretical minimum AREA36)
k2 = 3.262 (theoretical maximum AREA36 = 36/11.04)
1.35
⎡ ⎤
⎢ k2 − 1 ⎥
AFHMA =⎢ ⎥ ………………..……..Eq. 41
⎢ k 2 − ⎛⎜ AREA12 ⎞⎟ ⎥
⎣⎢ ⎝ k1 ⎠ ⎦⎥
Where: AFHMA = AREA factor for HMA
k1 = 6.85 (theoretical minimum AREA12)
k2 = 1.752 (theoretical maximum AREA36 = 12/6.85)
⎛ 1 ⎞
E ⋅ AFPCC ⋅ k3⎜⎝ AFPCC ⎟⎠
EPCC = 0 2.38
…………………………..Eq. 42
k3
⎛ 1 ⎞
E 0 ⋅ AFHMA ⋅ k 3 ⎜⎝ AFHMA ⎟⎠
E HMA = 2
…………………………..Eq. 43
k3
Where: EPCC = Elastic modulus of upper PCC layer(s)
EHMA = Elastic modulus of upper HMA layer(s)
AFPCC, AFHMA = AREA factors
k3 = Thickness ratio of upper layer thickness to load plate diameter
a = Load plate radius
E0 = Composite modulus of the entire pavement system beneath
the load plate
44
1 .5 ⋅ a σ 0
And, E0 =
d0
σ0 = Peak pressure of FWD impact load under the load
plate
2.2.3.7. MODULUS
MODULUS is a program that utilizes the database-search technique to backcalculate the
moduli of pavement layers (19). Such a program is developed by using a linear elastic
computer program to simulate the deflection basins that would result from an FWD-type
load on thousands of combinations of hypothetical layer thicknesses and stiffnesses. All
of the fictitious deflection basins are then stored in a permanent database. The program
works by searching and interpolating between these previously calculated deflection
basins to find the combination of layer thicknesses and stiffnesses that would
theoretically exhibit a deflection basin closest to what was measured in the field. Until
recently, MODULUS has been a DOS based program. The newest version, MODULUS
6.0, has been built on a Windows platform and appears to be much more user friendly
than previous versions of the program.
The manner in which MODULUS calculates a deflection is depicted by Equation
44. Ultimately, an error function in the form of Equation 45 is minimized to find the best
solution from the database. In order for this process to work, the user must input known
layer thicknesses, plus seed values and acceptable ranges for the moduli of those layers.
Equation 44 shows that MODULUS works in terms of modular ratios instead of
independent moduli. For this reason, an estimate of the subgrade modulus (En) must be
provided so that the program can report actual moduli instead of ratios.
45
qa ⎛ E1 E 2 E ⎞
w ci = f i ⎜⎜ , ,..., n −1 ⎟⎟ …………………………..Eq. 44
En ⎝ En En En ⎠
2 2
⎛ wim − wic
s
⎞ s
⎛ wc ⎞
ε = ∑ ⎜⎜
2
⎟ = ∑ ⎜1 − mi
⎟ ⎜ w
⎟ …………….………..Eq. 45
⎟
i =1 ⎝ wim ⎠ i =1 ⎝ i ⎠
Where: wc = Computed deflection
wm = Measured deflection
i = sensor number from 1 to S
Ex = Elastic modulus for layer number x
q = Contact pressure
a = Contact radius
ε2 = Squared error
fi = Value from layer system program and used as a data base
The Strategic Highway Research Program (SHRP) found in 1993 that
MODULUS showed superior performance to other programs (MODCOMP3 and
WESDEF) for analyzing the roughly 800 test sections that comprised the SHRP
deflections database at the time (13). SHRP published a set of recommendations for
performing backcalculations with MODULUS that focused on three major areas:
definition of layer moduli ranges, modeling of the pavement structure, and evaluation of
the analysis results. A summary of those recommendations is now presented.
When deciding a seed value for HMA layers, Equation 46 should be used
whenever the necessary inputs are available. Equation 47 should be used in lieu of
Equation 46 for cases in which little is known about the HMA mixture. Table 7 can be
used to supply some of the variables for Equation 47 if they are unknown.
46
log10[E*] = 2.2250053 – 0.091756Vbe – 0.027949Va – 0.096881P200 + ……….Eq. 46
0.250094Pabs – 0.006447tp + 0.060612f – 0.00007404tp2 +
0.00191539Vbe2 + 0.0082813P2002 – 0.0010225P3/42 +
0.0001909P3/82 – 0.0801155Pabs2 + 0.0148592η70,10^62 – 0.0024159f2
+ 0.00094015P3/8Vbe + 0.00084534P3/4Vbe + 0.0004965P3/4P4 –
0.00034328P3/8Pabs
log10[E*] = 0.553833 + 0.028829P200f-0.17033 – 0.03476Va + 0.070377η70,10^6 + ..Eq. 47

0.000005[tp(1.3 + 0.49825log(f))PHMA0.5] – 0.00189[tp(1.3 + 0.49825log(f))PHMA0.5f-1.1]
Where: E* = HMA modulus, in 105 psi

Vbe = Effective binder content, by volume percentage
Vbe = [(Pac − Pabs − (Pac Pabs 100)) × Gmb ] ÷ Gb
Gmb = Maximum specific gravity of the mix

Gb = Specific gravity of bitumen (assume 1.010 if unknown)
Va = Percent air voids in mix
P200 = Percentage aggregate weight passing the No. 200 sieve
Pabs = Percent asphalt absorption, by weight of aggregate
G se − G sb
Pabs = × Gb ⋅ 100%
G sb − G se
PHMA = Percent asphalt content by weight of mix

f = Test frequency of load wave, assume 16 Hz in all cases
tp = Mid-depth HMA temperature, degrees Fahrenheit
P4, P3/8, P3/4 = Percent aggregate weight retained in No. 4, 3/8”,
and 3/4” sieves, respectively
۫ in 106 Poises
η70,10^6 = Asphalt viscosity at 70◌F,
47
Table 7 Default Values for Determining HMA stiffness (19)
Va PHMA P200
Surface Courses 4% 6% 6%
Binder Courses 5% 5% 5%
Base Courses 7% 4% 4%
Sand Asphalt Mixtures - 8% 6%
All HMA layers that have the same construction age are combined into a single
composite asphalt layer with thickness hcomp and stiffness Ecomp as defined by Equations
48 and 49, respectively. The range of moduli that is reported to MODULUS should be
determined according to Equation 50.
n
hcomp = ∑ hi …………………………..……Eq. 48
i =1
3
⎡⎛ h ⎞ ⎤
E comp = ∑ ⎢⎜ i ⎟ ⋅ 3 Ei ⎥ …………………...…….Eq. 49
⎢⎣⎜⎝ hcomp ⎟
⎠ ⎥⎦
Range = 0.25*E(initial or composite) to [3.00*E(initial or composite) < 3,000,000 psi]……Eq. 50
MODULUS also requires seed and range values for PCC layers in the pavement
structure. A seed value for EPCC is determined according to the methods in Table 8,
which are listed in order of highest to lowest accuracy. The most accurate method for
which input parameters are known should be used. The range of moduli that is reported
to MODULUS is determined using Equation 51.
MODULUS is capable of modeling 4 layers, and if a given pavement system
contains more than 4 layers, SHRP recommends 7 prioritized steps (available elsewhere)
48
for combining the layers so that all composite layers will maintain an accurate
representation of field conditions.
Table 8 Methods for Estimating Seed Value for EPCC (19)
Required Inputs Methodology

Static modulus (E) test results Report E directly
Compressive strength (fc’) in psi E = 57,000*(fc’)0.5
fc’ = [12.53*Splitting tensile strength] – 1275
Splitting tensile strength in psi
E = 57,000*(fc’)0.5
No data Assume E = 4,000,000 psi
Range = 0.25*Einitial to [3.00*Einitial < 9,000,000 psi]…………Eq. 51
Unbound granular base and subbase materials may be estimated using values
Table 1 unless the reported lower modulus bounds are lower than the seed value for
subgrade modulus. In such a case, the seed value for subgrade modulus should be used
as the lower bound for the unbound granular base or subbase material.
The seed value for subgrade modulus is calculated in two steps. First, Equations
52 and 53 are used to calculate the composite modulus of the pavement structure for each
deflection that is measured at a radius beyond that of the load plate. The seed value is
then taken as the minimum of the calculated composite moduli values.
E comp =
2
( )
Pc ⋅ a c ⋅ 1 − μ 2 ⋅ C
…………………..……Eq. 52
dr ⋅ r
⎛ r ⎞
C = 1.1 log⎜⎜ ⎟⎟ + 1.15 ………………..….……..Eq. 53
⎝ ac ⎠
49
Where: Ecomp = Composite modulus of the pavement
Pc = Contact pressure applied by FWD
ac = Load plate radius, in.
μ = Poisson’s ratio (assume 0.4)
dr = Measured deflection at radial distance r
r = Radial distance from deflection
C = deflection constant
When all of the preceding recommendations are followed, the backcalculated
moduli have a good chance of being reasonable, but there is no guarantee. To ensure
some degree of accuracy, SHRP also includes recommendations for evaluating the results
of a MODULUS backcalculation. First, a limit of 2% average error is set for the
matching of individual deflections. In addition, if the reported modulus for any layer is
equal to either the upper or lower bound of acceptable moduli for that layer, the reported
modulus is considered invalid. Finally, SHRP recommends that all backcalculation
results be closely reviewed by a qualified engineer to check for reasonableness. If the
results are considered unreasonable for any reason, the backcalculation process should be
screened for obvious errors and/or the results should be discarded.
2.2.3.8. ELMOD 5
ELMOD is a Windows based backcalculation program that is marketed by Dynatest® and
usually sold with the Dynatest® Falling Weight Deflectometer (28). The name, ELMOD,
is an acronym which stands for Evaluation of Layer Moduli and Overlay Design. LTPP
50
has noted that ELMOD is well suited for processing large batches of FWD readings
without constant attention from the user (14).
ELMOD uses an iteratation-based backcalculation procedure that can forward
model the pavement response using a variety of techniques. ELMOD is also known for
its ability to model the temperature-dependent behavior of HMA and the nonlinear
response of subgrade and aggregate materials (29). Older versions of the software were
limited to two different types of forward calculation. The first option was a common
deflection basin matching routine. The other option was based on matching the radius of
curvature of the deflection basin instead of trying to match the basin as a whole. The
literature suggests that the radius of curvature option is more accurate, but still
acknowledges the merit of the more typical deflection basin matching technique (14).
The most current version, ELMOD 5, is equipped with a finite element module (FEM)
can treat every layer as non-linear elastic.
ELMOD is used by the Arkansas State Highway and Transportation Department
(AHTD), and is generally a good candidate for governmental agencies for several
reasons. First of all, the software is available with the Dynatest® FWD, which is likely to
be purchased by state DOTs even without the inclusion of software. Second, as noted by
LTPP, ELMOD is capable of running large batches of FWD field data. This is a valuable
feature when long stretches of highway or other roadways are being analyzed. In
addition, ELMOD contains optional automated tools for determining the remaining life of
a pavement and also for overlay design (28). The accuracy of these tools is beyond the
scope of the current study, but their potential convenience is noted.
51
2.2.3.9. BAKFAA
BAKFAA, as the name implies, was developed by the Federal Aviation Administration
(FAA) specifically for the backcalculation of FWD data on airfield pavements (30).
BAKFAA used to be called FAABACKCAL, but the difference in names does not
signify any significant difference in the software (31). BAKFAA is similar to ELMOD
in that it is an iteration-based backcalculation that utilizes layered elastic theory (32).
BAKFAA is unique because it can model up to 10 pavement layers, and also
because its code is written with a special emphasis on computational efficiency. It can be
used for airfield or highway pavements for which a deflection basin (such as one created
during FWD loading) is measured. A special feature of BAKFAA is its ability to
calculate strains due to aircraft gear loads, which is accommodated by the FAA’s linear
elastic analyzing software, LEAF. The program is Windows based, and has a graphical
user interface that is simple to use.
For the purposes of analyzing highway pavements, BAKFAA works by
minimizing the error between LEAF-generated deflections and field-measured
deflections. LEAF iteratively alters the moduli of user-defined layers and layer
thicknesses until the generated and measured deflection measurements match within
some tolerance.
2.2.3.10. DIPLO-DEF
DIPLODEF is the newest and most versatile backcalculation software that is currently
available (33). This program falls into the “iteration” class of backcalculation programs.
The user must give DIPLO-DEF a seed and range moduli and thickness for each layer,
52
and then the program begins a process of forward-calculating theoretical deflections.
DIPLO-DEF calculates theoretical deflections over and over again for different moduli
until the theoretical, forward-calculated deflections match the measured deflections
within some tolerance.
Even though DIPLO-DEF is theoretically superior to most existing
backcalculation programs, it is relatively difficult to use. DIPLO-DEF is DOS based
software that requires various sets of strings of alphanumeric characters to describe the
deflection basin and the preferred method for backcalculation.
To better illustrate the features that set DIPLO-DEF apart from other programs, it
is necessary to analyze the program in terms of each of its two main components: (1) the
iterative search subprogram and (2) the methods by which forward calculations are
conducted. DIPLO-DEF was created by combining two existing components. The
computer optimization routine from a program called WESDEF is used as the iterative
search subprogram, and a separate program called DIPLOMAT is used to perform
forward calculations.
WESDEF works in the same manner as a stiffness matrix, and therefore will only
work if it can generate enough equations to solve for the number of unknowns that are
present. For backcalculation of pavement layer moduli, the number of layers for which
moduli are to be determined represents the number of unknowns, and the number of
deflection measurements represents the number of equations (19). For every deflection
that is measured (j), a relationship in the form of Equation 54 is developed for each
pavement and subgrade layer for which modulus is sought (i). Figure 13 illustrates the
relationship that is developed between deflection and modulus. The system of equations
53
is solved to provide slopes (Sji) and intercepts (Aji) that can be used to predict a modulus
to match the measured deflection. A reasonable solution is usually obtained within three
iterations.
log (deflectionj) = Aji + Sji (log Ei) ……….…………Eq. 54
FIGURE 13 WESDEF Relationship Between Deflection and Modulus (19)
The second component of DIPLO-DEF, DIPLOMAT, is the part of the program
that calculates new theoretical deflections after WESDEF performs an iteration.
DIPLOMAT is different from other forward calculation sub-programs because it can
perform forward calculations using any combination of layered elastic theory and plate
theory for pavement layers and either dense liquid or elastic solid modeling for the
54
subgrade. In addition, DIPLOMAT can handle any number of layers and model the layer
interfaces as either fully bonded or unbonded.
In the grand scheme of pavement analysis, DIPLOMAT is most valuable because
of its ability to analyze pavements with both HMA and PCC layers. The principles
behind the software are not new, but DIPLOMAT is the first successful attempt to unify
the various models that have been generally considered mutually exclusive since the
onset of backcalculation in nondestructive testing. The development of DIPLOMAT is
documented elsewhere (34).
The developers of DIPLO-DEF have recommended several guidelines for its use.
First of all, care must be taken when choosing seed and range values for layer moduli.
As with all iterative backcalculation procedures, the seed values do have an effect on the
final backcalculated values. The user-input value for Poison’s Ratio of the subgrade
should also be carefully considered. The developers found that using values above 0.45
seemed to cause DIPLO-DEF to abort early. Therefore, it is recommended that Poisson’s
Ratio values for the subgrade not be assumed above 0.45.
As a general note, the developers of DIPLO-DEF also stress the importance of
patience when performing backcalculations. DIPLO-DEF offers many different options
for modeling any given pavement structure, and the user should be prepared to exhaust
all practical options and combinations in order to obtain reasonable results. In this effort,
it is important to exercise engineering judgment and also to keep a careful record of all
trials.
55
2.3. Spectral Analysis of Surface Waves
A measure of seismic disturbances in pavement and subgrade layers can be used to
estimate the stiffness and thickness of the different layers. More specifically, the elastic
waves that are generated by a transient impulse load may be monitored by a recording
device and then converted into a dispersion curve. The dispersion curve is then
iteratively inverted to obtain a vertical profile of the shear wave velocity. Finally,
discrepancies in the shear wave velocity profile are used to identify pavement and/or
subgrade layer interfaces, and the average shear wave velocity within each layer is used
to estimate the elastic modulus of the layer. This method of NDT is called the spectral
analysis of surface waves (SASW) (35).
2.3.1. Theoretical Background
SASW is based on the theory of stress waves propagating in elastic media. For analysis,
the pavement and subgrade layers are considered to comprise a layered elastic half space.
This means that all of the various layers of pavement and subgrade are elastic,
homogeneous, and isotropic, and that the entire elastic space ends at the ground surface.
In contrast, an elastic whole space extends endlessly in all three dimensions. Two types
of waves propagate radially outward from a disturbance in an elastic whole space:
primary waves (also called P-waves or compression waves) and secondary waves (also
called S-waves or shear waves). Primary waves cause particles in the elastic medium to
deflect in the direction of wave propagation, and secondary waves cause particles in the
elastic medium to deflect in the direction perpendicular to wave propagation. Primary
waves always propagate at a higher velocity than secondary waves (35).
56
Half spaces accommodate two additional types of elastic waves: Rayleigh (R-
waves) and Love (L-waves). Rayleigh waves are analogous to S-waves because they
propagate perpendicular to the direction of wave propagation. The difference is that R-
waves only exist near the boundary of a half space; the amplitude of an R-wave decays
rapidly beyond a depth of 1.5 times the wavelength (λ) of the R-wave. L-waves are
similarly analogous to P-waves. The velocity at which these various types of waves
travel is dependent upon the Poisson’s ratio (υ) of the elastic medium. As υ increases to
its upper limit of 0.5, the ratio of P-wave velocity to S-wave velocity approaches infinity.
This is due to the fact that a material with υ = 0.5 is theoretically incompressible,
incapable of allowing the manifestation of S-waves. In contrast, the ratio of R-wave
velocity to S-wave velocity remains within 0.86-0.95 for all values of υ (35). This leads
to a common practice of approximating shear wave velocity as 1.1 times the measured R-
wave velocity.
A vertical disturbance on the surface of an idealized half space such as the one
shown in Figure 14 results in the propagation of many different stress waves of myriad
frequencies. Considering each different frequency of wave as a unique disturbance, it is
convenient to define phase velocity (Vph) as the velocity at which a seismic disturbance
travels. Low-amplitude, high-frequency waves will penetrate only shallow depths.
Therefore, the phase velocity of such a wave is influenced mostly by the thin upper layers
of the pavement system. Conversely, a high-amplitude, low-frequency wave penetrates
through greater depths; its phase velocity is indicative of the properties of every layer
through which it passes. Thus, the upper pavement layers do influence high-amplitude
57
waves, but with increasing amplitude, semi-infinite subgrade layers play a greater role in
regulating the velocity of wave propagation.
For the purposes of SASW it is necessary to think of the phase velocity as a
function of wave frequency instead of as a function of time. In the frequency domain, it
is clear to see that a wave of particular frequency corresponds to a certain phase velocity,
as a function of several elastic layer properties. Phase velocity depends upon the
frequency of the wave (f) and also the mass density (ρ), Poisson’s Ratio (υ), and
H1 GS1, ρ1, υ1 Layer 1
H2 GS2, ρ2, υ2 Layer 2
H(n-1) GS(n-1), ρn-1, υn-1 Layer (n-1)
GSn, ρn, υn Layer n

Semi-Infinite
FIGURE 14 Idealized Layered Half Space
especially the Shear Modulus (GS) of the material in that layer. The height (H) of each
layer defines how much weight the elastic properties of that layer receive when averaging
their impact with that from other layers through which a wave travels. Thus, by inputting
known or assumed values for elastic layer properties, phase velocity can be solved
directly in terms of an independent variable, wave frequency.
58
The relationship between wave frequency (or wavelength) and phase velocity
over some frequency domain is called a dispersion function, the graph of which is called
a dispersion curve. Plotting wavelength versus phase velocity is the more tactical choice
because wavelength is used to estimate the sampling depth of the phase velocity. In
practice the sampling depth is usually assumed to be between (½)λ and (1/3)λ; a common
assumption is to let the sampling depth equal (0.4) λ.
Dispersion curves that are created from hypothetical (not measured) elastic layer
properties are referred to as theoretical dispersion curves. The ultimate goal of SASW
testing is to identify pavement layer thicknesses and stiffnesses. This is done by creating
a dispersion curve from measured R-wave phase velocities and then re-creating that
shape with a theoretical dispersion curve that is defined by distinct layer thicknesses and
properties. Though there is no unique solution of layer thicknesses and properties that
will re-create a given dispersion curve, a solution can be found that adequately describes
the layered pavement system under investigation through trial-and-error simulations (36).
2.3.2. Field Procedures
The basic idea behind SASW testing is to impart a transient vertical impact to the ground
surface and measure the resulting particle motion from R-wave propagation at two or
more known distances from the disturbance. The setup shown in Figure 15 will be used
to illustrate the process of field data collection.
The transient impact that is used to generate seismic waves may be imparted by a
hammer, falling weight, sinusoidal noise source, or a vehicle depending on the desired
sampling depth. As was discussed in the previous section, high-frequency, low-
59
amplitude waves are used to sample shallow depths. Hand-held hammers are usually
sufficient for creating such disturbances. Large vehicles and/or sinusoidal noise sources
are more appropriate for generating the high-amplitude, low-frequency waves that are
required for larger sampling depths.
Waveform Analyzer or
Transient Impact
Dynamic Signal Analyzer
Source
Receiver 1 Receiver 2
Various
Frequencies of
R-waves
d/2
d1 d
d2
φi
FIGURE 15 SASW Configuration
Sensor spacing (d) is another setup issue that is chosen based upon desired
sampling depth. For shallow sampling depths, small sensor spacings are both appropriate
and convenient. As the sampling depth increases, however, so must the sensor spacing.
The reason for this accommodation will be explained with the discussion of data
reduction.
First, though, it is necessary to discuss how the raw field data is transformed into
a format that can be used to create a dispersion curve. The first issue to consider is the
switch from the time-domain to the frequency-domain. Fourier Transform is the
mechanism by which the switch is made, and it is performed by an electronic machine
60
called a Wave Form Analyzer or a Digital Signal Analyzer. An example of one of these
devices is shown in Figure 16 (37).
Depending on the type of analyzer that is being used, there are at least two - and
sometimes many more - inputs for seismic disturbance readings. The device that is used
to measure the particle motion from R-waves is chosen based on the depth and frequency
of sampling. For large amplitudes and low frequencies, motion detectors (transducers) or
velocity transducers (geophones) are sufficient to measure the particle motion for
calculations of phase velocity. When performing SASW tests on a layered pavement
system, however, it is more appropriate to use accelerometers. Accelerometers have a
high frequency response that is better suited to the high-frequency waves that are used to
analyze stiff pavement layers (38). The time domain can be used to convert the
acceleration measurements to the desired value of vertical particle motion.
FIGURE 16 HP 35670A Digital Signal Analyzer (37)
61
One problem with using the Fourier Transform for transient impacts is the noise
that is present from surrounding activities and from the reflection of waves between
elastic layers. Noise can be thought of as vibrations in the elastic half space that are
caused by a phenomenon other than direct propagation of individual R-waves. Examples
of such phenomena could be nearby traffic, multi-modal R-waves, or even the reflection
of just one R-wave.
SASW configurations in which only two measurements of vertical particle motion
are taken are most susceptible to noise. The two most common and accepted ways to
arrange a set of two receivers - common receivers midpoint (CRMP) geometry and
common source (CS) geometry - are shown in Figures 17 and 18, respectively. It has
been argued that CRMP geometry reduces scatter by maintaining equidistant spacing
between receivers and impact, but others have not found the reduction in scatter to be
significant enough to justify abandoning the convenience of CS geometry (38).
When using CRMP geometry the two receivers are positioned equal distances
from an imaginary centerline and the transient impact is applied on either side of the
receivers at a distance equal to the distance between the receivers. Next, the receivers
and transient load are moved further away from the stationary imaginary centerline and
the test is repeated, again from both sides. This process is usually repeated 5 or 6 times
so that different ranges of R-wave wavelength may be measured. From Figure 17 it can
be seen that for each repetition of the test both receivers and the transient impact must be
moved. CS geometry is more convenient because the source of the transient impact can
be left in one place for many different receiver spacings. The transient impact is still
positioned at a distance equal to the distance between the two receivers, but instead of
62
rearranging the setup about a stationary imaginary centerline, the receivers are simply
offset a further distance from the transient impact source (39).
Taking readings of seismic reactions for different sensor spacings provides a larger
number of wavelength samples that can be averaged to reduce the error from noise.
Time-domain signals from all of the different spacings are averaged before being Fourier
transformed (a process called stacking) to improve the quality of signal. It is particularly
important to measure the seismic effect that the transient impact has from each side of the
sensors. This serves to eliminate the error from phase shifts that occur within the
FIGURE 17 Common Receivers Midpoint Geometry for SASW (38)
63
FIGURE 18 Common Source Geometry for SASW (38)
receivers and also to measure the homogeneity of the sublayers in the direction of the
receivers (38, 39).
Another method of eliminating error is to increase the number of sensors
(accelerometers) that are used during testing. When more than two receivers are used the
process is sometimes referred to as the multi mode analysis of surface waves (MASW).
Some wave analyzers may be limited to two input channels, but it is common for newer
models to have inputs for several dozen sensors (38). A setup consisting of more than
one receiver is called an array. The advantage of using an array is time efficiency. For
just one transient impact, wavelengths can be sampled using any combination of two of
the receivers from the array, whereas a setup of only two receivers can only measure a
relatively limited amount of time-amplitude data. Thus, the more sensors that are
included in the array, the more measurements can be obtained for averaging. Even for
64
MASW it is important to perform the test from both directions to eliminate any internal
phase shifts from the receivers. The literature suggests, however, that testing at higher
frequencies is more beneficial for pavement analysis than testing with more than two
receivers (38, 40, 41).
For either SASW or MASW, the purpose of field testing is to measure particle
motion that corresponds to the amplitude of R-wave motion in the time domain. After
this data is collected for a given site, the measurements are averaged in an attempt to
eliminate noise in the readings that will hinder contingent calculations. After the field
measurements are averaged, the Waveform Analyzer or Dynamic Signal Analyzer is used
to perform a Fourier transform so that the data can be analyzed in the frequency domain.
The output from the Fourier transform includes two important pieces of information: the
coherence function and the cross-power-spectrum.
The coherence function is essentially a measure of the signal-to-noise ratio that
provides a quick way to evaluate the quality of recorded data. Coherence values range
from zero to one, with values of one corresponding to signals absent of noise and values
of zero corresponding to signals so noisy that no relationships can be found. Typically,
data is considered to be clean enough to use for analysis when the coherence value is
between 0.9 and 1.0. Figure 19 shows the plot of a typical coherence function (35).
Along with the coherence function, the Waveform Analyzer or Digital Signal Analyzer
gives the cross-power-spectrum. Consider the idealized R-wave motion represented in
Figure 15. Each of the various R-waves travels a sinusoidal path, traversing a phase
angle (φi) in the time domain that is dependant on the wave’s frequency (fi). The cross-
power-spectrum uses the Fourier transformed data to represent the particle motion in
65
terms of phase angle and frequency. Since, at any given point, a higher-frequency wave
will have traversed a greater total phase angle than a lower-frequency wave, it is rational
to expect a constant decreasing relationship between phase angle and frequency. As can
be seen in Figure 20, the output of a cross-power-spectrum is not a constantly decreasing
relationship. Instead, the relationship is asymptotic at regular frequency intervals. This
is the result of plotting phase angle in terms of relative displacement (i.e. in terms of 0 <
|φi| < 180o) instead of total displacement (i.e. 0 < |φi| < ∞ ). When the phase angle is
represented in terms of relative displacement it is called the wrapped phase. The phase
can be unwrapped by adding 360◌۫ to the magnitude of φi every time the wrapped phase
crosses the dotted line corresponding to φi = 0o (35).
FIGURE 19 Typical Coherence Function (35)
66
FIGURE 20 Typical Wrapped Cross-Power Spectrum (35)
The ultimate goal of the field test is to come up with a dispersion curve relating
wavelength to phase velocity, and these quantities can be calculated using the cross-
power-spectrum. The first step is to pick a point on the cross-power spectrum and
identify its frequency and unwrapped phase angle. The time (t) that it took the wave to
travel to the unwrapped phase angle is calculated using Equation 55. Referring again to
Figure 15, the phase velocity (Vph) is calculated by dividing the space between the 2
receivers (d) by the previously calculated travel time (t), Equation 56. The next step is to
calculate the wavelength, which can be done using Equation 57. Finally, the process is
repeated for a sufficient number of points to clearly define the experimental dispersion
curve (35).
ϕ
t= .......................................................Eq. 55
360 ⋅ f
67
d
V ph = ..........................................................Eq. 56
t
V ph
λ ph = ........................................................Eq. 57
f
2.3.3. Backcalculation Procedure
Backcalculation for the SASW method consists of two major steps. First, the field
dispersion curve is used to create a “true” profile of Shear wave velocity versus depth.
The second and final step is forward modeling of pavement and subgrade properties to
recreate the “true” shear wave velocity profile. The forward modeling process for SASW
is called inversion. Inversion can be as simple as “eyeballing” dominant ranges of a
dispersion curve for representative values, or considerably more exhaustive methods may
also be undertaken.
Inversion can be a relatively complex task that requires a large amount of
engineering judgment. Research is underway across the United States to develop
automated inversion procedures so that SASW tests may be conducted with difficulty not
significantly greater than other field tests. The University of Arkansas currently uses a
computer program called SignalCapture2.0 that was written in LabVIEW to collect and
store field data from SASW testing. SignalCapture collects data in the time domain,
performs a Fourier analysis, and provides the necessary coherence function and cross-
power spectrum. The program was first developed in 2003 by Kyle Bennett and later
modified by Andrew Wilson and Sutapa Hazra. The cross-power spectra from
SignalCapture are imported into a macro-enabled Excel spreadsheet to produce a field
dispersion curve. The spreadsheet employs the approximations that are noted in
68
Equations 58 and 59 below. In addition, the user is allowed to mask any frequency
ranges that appear to be inaccurate or scattered.
V S = 1.1 ⋅ V R ………………………..………..Eq. 58
d s = 0.4 ⋅ λ …………………………………Eq. 59
Where: VS = Shear wave velocity
VR = Rayleigh wave velocity
ds = Sampling depth
λ = Wavelength
The inversion process is most robustly accomplished using the Haskell-Thompson
transfer matrix, as discussed by Kausel and Roësset (42). The transfer matrix is used to
describe the measured phenomenon in terms of all of the parameters that significantly
affect the seismic behavior of the pavement system. The transfer matrix utilizes
approximately 60 points from the field dispersion curve to estimate the shear wave
velocity in roughly 15 discrete layers. Statistical methods are then used to minimize the
error of the transfer matrix. The end result of this inversion is a profile of shear wave
velocity versus depth, mechanistically based on measured field data (39). While this
method is considerably more involving than the simplified procedure of “eyeballing” the
field dispersion curve, it is the more robust method that will eventually allow the
automation of SASW.
In either case, inversion involves a type of forward calculation to determine the
actual thickness and stiffness of pavement and subgrade layers. First, either visual or
69
analytical inspection of the shear wave velocity profile is used to identify obvious
discontinuities where layer interfaces are likely to exist. Then, some reasonable values
for layer thickness, Poisson’s ratio, density and shear modulus are assumed for each
layer. For the simplified method, VS may be directly read from the field dispersion curve
and converted to the shear modulus using Equation 60.
G = VS ⋅ ρ ………………………………..Eq. 60
2
Where: G = Seismic shear modulus (lb/ft2)
VS = Shear wave velocity (ft/s)
ρ = Mass density (lb*s2/ft4)
For the transfer matrix method, these assumed values are used to create a
theoretical dispersion curve which is then compared to the field dispersion curve. If the
two curves do not agree within some tolerance, the assumed pavement parameters are
changed and the theoretical dispersion curve is recreated. This process is continued by
trial-and-error until the theoretical and field dispersion curves are sufficiently similar.
The assumed pavement properties that result in the best-fitting theoretical dispersion
curve are taken to be representative of field conditions (39).
Once the shear modulus is obtained for the pavement layers (either by the
simplified method or the transfer matrix method), Equation 61 is used to convert the
shear moduli to Young’s moduli (E).
E = 2G (1 + ν ) ………..………………………Eq. 61
70
Where: E = Young’s modulus
G = Shear modulus
ν = Poisson’s ratio
The 1993 AASHTO Pavement Design Guide and the new Mechanistic-Empirical
Pavement Design Guide (MEPDG) both use Resilient Modulus (Mr) as the primary
subgrade characterization criteria. Mr is defined as axial stress divided by recoverable
strain, whereas Young’s Modulus is the same axial stress divided by total strain. On a
theoretical basis, Mr and E represent two different values. On a more realistic basis,
however, the two quantities can be nearly equal.
As shown in Figure 21, as the number of load cycles increases, the recoverable
strain begins to equal the total strain (19). This is especially true for loads well shy of a
material’s capacity and after 100 to 200 repetitions. Since seismic disturbances impart
stresses that are well below the strength of all pavement materials, the characteristic E
and Mr values should be nearly equal (43). Thus, a common assumption in SASW
FIGURE 21 Strains Under Repeated Loads (19)
71
testing is to take the Resilient Modulus to be equal to Young’s Modulus.
The last step of backcalculation is to adjust the seismically estimated modulus to a
more appropriate strain level. Moduli that are measured at seismic strain levels (< 10-3
%) are essentially equal to the maximum modulus for the material (43). With increasing
loads and corresponding strains, soils exhibit a non-linear decrease in modulus. This
trend is shown graphically in terms of Shear Modulus (G) in Figure 22 (19).
Since traffic loads induce strains that are significantly greater than what is
experienced during seismic testing, the modulus of a given soil is expected to be
significantly less under traffic loading than for seismic loading. Accordingly, the moduli
that are obtained from theoretical dispersion curves must be reduced by some amount to
avoid reporting Resilient Modulus values far in excess of what is true or reasonable. An
appropriate percent reduction can be estimated using a normalized modulus reduction
curve, which relates G/Gmax to shear strain. Three methods for modeling G/Gmax and/or
E/Emax against corresponding strain are now presented.
The simplest method for adjusting seismic moduli is to use a two-parameter
hyperbolic model (44). Equation 62 shows Darendeli’s simplified hyperbolic model
where the reference shear strain for normalization (γr) represents 0.5Gmax. . Equations 63
and 64 give the reference strain and curvature coefficient (a) for coarse and fine grained
soils, respectively.
G 1
= a
……………………….…….Eq. 62
Gmax ⎛γ ⎞
1 + ⎜⎜ ⎟⎟
⎝γr ⎠
72
FIGURE 22 Normalized Shear Modulus Reduction Curve (19)
0.5×Cu −0.15
⎛σ ' ⎞ ⎛σ '⎞
γ r = 0.12 × C u −0.6 × ⎜⎜ 0 ⎟⎟ ; a = 0.86 + 0.1 × log⎜⎜ 0 ⎟⎟ …………Eq. 63
⎝ Pa ⎠ ⎝ Pa ⎠
γ r = (0.0352 + 0.0010 × PI × OCR 0.3246 )⋅ σ 0'

0.3483
; a = 0.9190 ………….Eq. 64
Where: G = Shear modulus at some arbitrary load/strain level
Gmax = Shear modulus at seismic strain level
γ = Shear strain corresponding to G
γr = Reference shear strain corresponding to G = 0.5Gmax
a = Curvature coefficient
Cu = Uniformity coefficient
σ0’ = Effective confining pressure (atm)
Pa = Atmospheric pressure (atm)
73
PI = Plasticity index
OCR = Overconsolidation ratio
Ke and Nazarian have developed another model that uses different parameters
than the hyperbolic model (45). Their model is defined by Equation 65, which
normalizes Young’s modulus in terms of relative confining and deviatoric pressures. At
very small or large deviatoric stresses, Equation 65 may report unrealistic values. As
such, Nazarian recommends upper and lower limits of Eseis and 0.05Eseis, respectively.
The statically determined coefficients k2 and k3 for some pavement materials are listed in
Table 9.
k2 k3
E ⎛σ ⎞ ⎛ σ d −ult ⎞
= ⎜⎜ c −ult ⎟⎟ ⎜⎜ ⎟⎟ ………………………Eq. 65
E seis ⎝ σ c −init ⎠ ⎝ σ d −init ⎠
Where: E = Young’s modulus at some arbitrary load/strain level

Eseis Young’s modulus at seismic strain level
σc-ult, σc-init = Ultimate and initial confining pressures, respectively
σd-ult, σd-init = Ultimate and initial deviatoric stresses, respectively
k2, k3 = Statically determined coefficients, Table 9
Table 9 Statically Determined Coefficients k2 and k3 for Equation 65 (45)
Material Characterization k2 k3
High Quality 0.4 -0.2
Base Average 0.2 -0.3
Poor Quality 0.1 -0.4
Sandy 0.4 -0.2
Subgrade Low Plasticity Clay 0.1 -0.3
High Plasticity Clay 0.0 -0.4
74
Two final constitutive models are now presented that provide separate, but similar
models for sands and clays (46). The primary difference between the two models is that
the model for clays includes an input for plasticity of the soil. Equations 66 - 68 are for
the sand model, and Equations 69 - 72 are for modeling clays or other fines with
plasticity.
= log K (γ ) + {m(γ ) − m0 }log σ ' 0 ………….……….Eq. 66

G
log
Gmax
⎡ ⎧⎪ ⎛ 0.000102 ⎞ 0.492 ⎫⎪⎤

k (γ ) = 0.5⎢1 + tanh ⎨ln⎜⎜ ⎟⎟ ⎬⎥ ……..…………….Eq. 67
⎢⎣ ⎪⎩ ⎝ γ ⎠ ⎪⎭⎥⎦
⎡ ⎧⎪ ⎛ 0.000556 ⎞ 0.4 ⎫⎪⎤

m(γ ) − m0 = 0.272 ⎢1 − tanh ⎨ln⎜⎜ ⎟⎟ ⎬⎥ ……………….Eq. 68
⎢⎣ ⎪⎩ ⎝ γ ⎠ ⎪⎭⎥⎦
= K (γ , PI )(σ ' 0 )
G m (γ , PI )− m0
………………….…..Eq. 69
G seis
⎧⎪ ⎡ ⎛ 0.000102 + n(PI ) ⎞ 0.492 ⎤ ⎫⎪

K (γ , PI ) = 0.5⎨1 + tanh ⎢ln⎜⎜ ⎟⎟ ⎥ ⎬ ……..……..Eq. 70
⎪⎩ ⎢⎣ ⎝ γ ⎠ ⎥⎦ ⎪⎭
⎧⎪ ⎡ ⎛ 0.000556 ⎞ 0.4 ⎤ ⎫⎪ (−0.0145 PI 1.3 )

m(γ , PI ) − m0 = 0.272⎨1 − tanh ⎢ln⎜⎜ ⎟⎟ ⎥ ⎬e ……Eq. 71
⎪⎩ ⎢⎣ ⎝ γ ⎠ ⎥⎦ ⎪⎭
⎧ 0.0 PI = 0
⎪3.37 × 10 −6 PI 1.404 0 < PI ≤ 15
⎪
n(PI ) = ⎨ −7
……………..…..Eq. 72
⎪ 7.0 × 10 PI 1.976
15 < PI ≤ 70
⎪⎩ 2.7 × 10 −5 PI 1.115 PI > 70
75
Where: G = Truck traffic load level modulus
Gseis = Seismic modulus
k2, k3 = Statically determined coefficients
σc-init and σc-ult = Initial and ultimate confining pressures
σd-init and σd-ult = Initial and ultimate deviatoric pressures
PI = Plasticity index of the base or subgrade material
γ = Truck traffic load shear strain
σ’0 = Mean effective confining pressure
This paper does not argue the relative accuracy of the three aforementioned
methods, but instead suggests that is important to consider several different models
and/or methods when attempting to estimate quantities that are not well understood or
defined. For the case of adjusting seismic moduli, it is necessary to estimate properties
such as effective confining stresses and/or shearing strains before the ultimate estimation
of corrected modulus can ever be considered. Thus, in each step of the process,
engineering judgment should be used and documented so that unreasonable results may
be compared with their corresponding methods and assumptions.
2.4. Case Studies: Comparisons of Backcalculation Procedures
The literature includes a number of comparisons between pavement and subgrade moduli
determined from different NDT and laboratory methods. Because of the relative
popularity of FWD compared to SASW, most of the available comparisons either
consider different FWD backcalculation procedures or a certain FWD procedure to
76
laboratory testing. For that reason, this section will only address two comparisons, and
those will be limited to FWD and laboratory-determined stiffnesses. Sections 3 and 4 of
this paper include a direct comparison of SASW and FWD-backcalculated moduli for a
particular section of roadway.
2.4.1. Oregon: FWD vs. Laboratory for HMA and Base Course Materials
A KUAB FWD was used to estimate pavement layer moduli for two typical and similar
flexible pavement sections in Oregon (47). The first section, the Rufus Project, consisted
of a 173-mm (6.8-in.) thick HMA layer underlain by a 457-mm (18-in.) thick base layer.
The HMA layer was moderately rutted and extensively cracked; the subgrade was non-
plastic sandy gravel. The second section, the Centennial Project, consisted of a 102-mm
(4-in.) thick HMA layer underlain by a 406-mm (16-in.) thick base layer. The HMA
layer was moderately cracked; the subgrade was a well-graded sandy gravel.
A software called BOUSDEF was used to backcalculate layer moduli from FWD
data. BOUSDEF is an iterative software that estimates moduli from user-input layer
thicknesses and Poisson’s ratios as well as measured deflection basin data. In addition,
BOUSDEF determines a non-linear regression to model the stress-dependent nature of
base course modulus if varying load levels are used during testing. ASTM D-4123 was
used for laboratory resilient modulus testing.
Figures 23 and 24 show the backcalculated and laboratory-measured resilient
moduli for the HMA layers of the Rufus and Centennial projects, respectively. It is clear
from the figure that laboratory values are consistently higher than backcalculated values.
This could be the result of removing cores at locations that are in relatively good
77
condition. If that is the case, the laboratory test would have no way to simulate the
distresses that were encountered in the pavement that was tested via FWD.
Figures 25 and 26 show plots of the backcalculated regression for base course
modulus versus laboratory-measured values for east-bound and west-bound lanes of the
Rufus and Centennial projects, respectively. These values have exceptional agreement
for the Rufus project and very close agreement for the Centennial project.
The Oregon report includes backcalculated values for subgrade resilient modulus,
but states nothing of laboratory or otherwise determined values for comparison. This
short comparison provides two pieces of information. First, iterative backcalculation
software appears to be able to estimate the non-linear resilient modulus of base materials
with a good deal of accuracy. While there is no clear reason why the backcalculated
HMA moduli did not match up, it is likely that the field samples were taken at locations
that were not representative of field conditions. This has dual consequence. First, it
emphasizes the need to characterize pavements at many different locations. Also, it
shows the importance of using engineering judgment throughout sampling and testing. If
relatively few samples are taken from a material in the field, those samples need to err on
the conservative side of the average. If an overlay were designed based on the laboratory
testing of an unrepresentatively strong sample, the overlay may be too thin to serve its
purpose.
78
FIGURE 23 HMA Moduli for Rufus Project (47)
FIGURE 24 HMA Moduli for Centennial Project (47)
79
FIGURE 25 Base Course Moduli for Rufus Project (47)
FIGURE 26 Base Course Moduli for Centennial Project (47)
80
2.4.2. LTPP Database: Comparison of FWD Backcalculation Methods
The Long Term Pavement Performance (LTPP) Program has compiled a database of
pavement sections for which FWD and/or laboratory testing has been performed, and one
study used that database compare the effectiveness of different FWD backcalculation
methods to estimate the measured stiffness of rigid pavements and subgrades (22).
Twenty-six jointed concrete pavement (JCP) sections were obtained that had deflection
data and k values measured from plate-load tests, 50 JPC sections were chosen that had
deflection data and measured EPCC values, and 76 continuously reinforced concrete
pavement (CRCP) sections were chosen that had deflection data and measured EPCC
values.
The deflection data was used as an input in ILLIBACK4, ILLIBACK7,
NUSBACK, and the LTPP Best-Fit methods to determine the relative ability of each to
estimate k and/or EPCC for rigid pavements. Figures 27, 28, and 29 compare measured to
backcalculated k values, EPCC for JPC, and EPCC for CRCP, respectively, (a) based on the
average value for each pavement section and (b) based on each deflection basin.
Similarly, Figures 30, 31, and 32 illustrate how many of the estimates contain varying
levels of error.
The authors of the LTPP study offer some interpretation of the results. First, the
relative inaccuracy of backcalculated k values is attributed to the likelihood that the plate
load tests would have been conducted at different times and temperature than the FWD
tests. The same is true for EPCC measurements, but k is more dependent on environmental
factors.
81
It is also noted that ILLIBACK4 and ILLIBACK7 have the most stringent
deflection basin conformance requirement, and that this may be the cause of their
relatively poor performance. The LTTP Best-Fit method attempts to match individual
deflections instead of the entire deflection basin, and that is probably at least part of the
reason why the forward calculation was more accurate. NUSBACK was clearly the most
accurate backcalculation method, and the authors guess that this is because of
NUSBACK’s computational flexibility. NUSBACK can use any two deflection
measurements, and this study used different strategic pairs for calculating k (d4 and d7)
and EPCC (d1 and d3). As a final note, the inclusion of tests on JCP seemed to prove that
jointed pavements are sufficiently “infinite” to use layered elastic theory as long as the
FWD test is conducted at an interior point on the slab.
FIGURE 27 Comparison of Backcalculated and Measured k Values (22)
82
FIGURE 28 Comparison of Backcalculated and Measured EPCC Values for JPC(22)
FIGURE 29 Comparison of Backcalculated and Measured EPCC Values for CRCP
(22)
83
FIGURE 30 Errors of Backcalculated k Values (22)
FIGURE 31 Errors of Backcalculated EPCC Values for JPC (22)
84
FIGURE 32 Errors of Backcalculated EPCC Values for CRCP (22)
85
3. DATA ANALYSIS
Falling Weight Deflectometer (FWD) and (Spectral Analysis of Surface Wave) SASW
tests were conducted for a FHWA project in Nashville, AR during the summer of 2007
by the Arkansas Highway and Transportation Department (AHTD) and the University of
Arkansas, respectively. The current study is completely separate from the research that
was being undertaken in Nashville, but the raw field data from those tests has been
provided for this project at no expense. The data from Nashville were all taken from a
single stretch of road comprised of approximately 76 adjacent 30-ft slabs, for a total
length of just under half a mile. The road, AR-4/US-278/Main St, consisted of a roughly
6-inch thick, JRCP layer that is underlain by native soils. At the time of testing the road
had never undergone major rehabilitation, and the primary distress was poor joint load
transfer between slabs.
The Natural Resources Conservation’s Web Soil Survey was used to classify
Nashville’s native soil (48). Figure 33 shows the spatial distribution of the three primary
soil types that are found in the immediate vicinity of AR-4/US-278/Main St., which is
drawn in yellow. Table 10 gives the engineering properties of the three types of soils that
were available from the Web Soil Survey. The information from Table 10 was used to
make educated assumptions about the remaining soil properties that were required for
analysis. Table 11 documents the assumptions that were made for backcalculating
moduli at the Nashville site.
SASW tests were also conducted at the Nashville site, but on a less frequent basis.
For each SASW test, a plot of the field dispersion curve is presented in Appendix A. The
estimated subgrade and pavement moduli are reported in Table 12 of Section 4. SASW
86
FIGURE 33 Spatial Distribution of Soil Types in Nashville, AR (48)
Table 10 Description of Soil Types at Nashville Site (48)
Symbol Name LL PI %Sand %Clay %Silt AASHTO USCS

Ruston Fine Sandy
RuB 31 12 46.0 22.7 31.3 A-6 CL
Loam
Sacul Fine Sandy
SaC 45 21 25.2 32.1 42.7 A-7-6 CL
Loam
Savannah Fine
SfB Sandy Loam
30 11 41.7 23.8 34.5 A-6 CL
87
Table 11 General Assumptions for Materials at Nashville Site
Property Assumed Quantity

Poisson’s Ratio of Subgrade Soil 0.40
Poisson’s Ratio of JRCP 0.15
Unit Weight of Subgrade Soil 115 pcf
Unit Weight of JRCP 150 pcf
Thickness of JRCP 6 inches
Truck-load-level Strain for Subgrade 0.05%
tests were documented according to the slab at which they were conducted, but
unfortunately there is not a similar record for the FWD tests. It is believed, however, that
all of the nondestructive tests were performed in a close enough proximity to one another
to facilitate a worthwhile comparison of backcalculated moduli. FWD tests were
performed at mid-slab at two load levels for 65 adjacent slabs, both northbound and
southbound, for a total of 260 measured deflection basins. The raw data from these
measurements is presented in Appendix B.
In an effort to compare the relative performance of the various FWD
backcalculation procedures, the raw data from Appendix B has been used to
backcalculate moduli using all of the reviewed methods that were available and/or
feasible for use. A spreadsheet was programmed to perform NUS-BACK and AREA72
backcalculations for all 260 deflection basins. The LTPP spreadsheet for forward
calculation of a rigid pavement was used to estimate the overall pavement stiffness and
subgrade resilient modulus for 30 deflection basins. Twenty of the basins were taken
straight from the raw data, and the other 10 were created by averaging the two sets of
load and deflections from 10 different locations. All 30 basins were chosen from basins
that produced reasonable results in both NUS-BACK and AREA72. BAKFAA software,
88
which is available to the public at no charge, was used to backcalculate the same 30
deflection basins. Lastly, ELMOD was used to evaluate the pavement structure using the
data file that was output from the FWD. Tables 34 - 40 in Section 4 document the results
of the various backcalculations that were performed.
The literature suggests that DIPLO-DEF and MODULUS are both valuable tools
for estimating the stiffness of pavement layers, but they were not used for this study.
Both programs were obtained, but the DOS environment proved too tedious and foreign
to be practical. The largest single factor that hindered the feasibility of these programs
was the process of identifying input and output files for computation. ILLI-BACK was
also excluded from the analysis portion of this project for the same reason. A newer,
Windows-based version of Modulus has been published, but it was not available for this
study. The omission of these programs is justified by considering that any increase in
accuracy from DIPLO-DEF or MODULUS would likely be offset by human error that
could arise from attempting to implement such tedious programs for widespread use in
state highway agencies.
SASW inversion was also conducted, but a significant additional assumption was
required. Either the frequencies that were used or the sensors that were used to measure
the surface waves generated field dispersion curves that are inconsistent and
unreasonable for the PCC layer. The shear wave velocities were expected to be highest
for the PCC layer, but the results on their own are not consistent enough to facilitate such
a conclusion.
The readings from sensor spacings greater than 3 inches consistently report shear
wave velocities similar to what is shown for the subgrade soil, and only the 3-inch
89
spacing gave results that showed the shear wave velocity to be highest in the PCC layer.
There are two likely reasons for this problem. The simplified equation for estimating
sampling depth could be inaccurate, causing the larger sensor spacings to report accurate
readings at inaccurate depths. The other likely scenario is that the larger sensor spacings
tend to mask the high-frequency signal that is required to analyze a stiff layer like PCC
with lower-frequency background noise. For the purposes of this study, the modulus of
the PCC layer was estimated by considering the most common shear wave velocity that
was estimated from the 3-inch sensor spacing and ignoring all other readings.
Another issue with SASW is the correction of seismic subgrade moduli to account
for the nonlinear stress-dependant variation. Since the soil at the Nashville site is a low
plasticity silty soil, Darendeli’s simplified hyperbolic relationship for fine grained soils
was used. The process of modulus adjustment is discussed in further detail in Appendix
C. A simplified example of the inversion process is also included in Appendix C.
One last technique of backcalcuation is now discussed before the results are
presented. The layered elastic models provide estimates for Young’s modulus of the
pavement and subgrade values, but rigid pavement design procedures necessitate a value
for k, the coefficient of subgrade reaction. The two values are related, and Equation 73 is
used to convert between the two throughout Sections 4 and 5. This is a modification of a
relationship that is given in the ASTM specification for measuring soil stiffness by the
electromechanical method, which gives stiffness in terms of force per depth penetration.
In order to change that value into a pressure per depth penetration, the ASTM equation
was simply divided by the area of the loading plate (49). For the purposes of this study,
the load plate had a radius of 5.91 inches.
90
1.77 RE 0.563E
k= ÷ πR 2 = ………………..………..Eq. 73
(1 −υ 2
) (
R 1−υ 2)
Where: k = Coefficient of subgrade reaction (psi/in.)
R = Radius of load plate (in.)
E = Young’s modulus (psi)
υ = Poisson’s ratio
The only laboratory measurement that is available for comparison with the
nondestructive estimations is a single measurement of concrete compressive strength
(f’c). This value was determined according to the specifications in AASHTO T 22. The
single test reported a compressive strength of 9,441 psi. Equation 74 was used to
estimate a value of 5,538,000 psi for EPCC.
E PCC = 57,000 f c' ………………………………Eq. 74
Where: EPCC = Young’s Modulus of PCC (psi)
f’c = Concrete compressive strength (psi)
91
4. RESULTS
4.1. SASW Results
The dispersion curves for the Nashville SASW tests are presented in Appendix A. The
reduced data from those curves is presented below in Table 12. A detailed explanation
of the reduction procedure is presented in Appendix C.
Table 12 Profile of Young’s Modulus with Depth from SASW
Depth Young’s modulus (psi) @ Slab (#)

(in.) 3 17 25 34 42 52 66 76
0–6 - 2,478,000 6,999,000 11,624,000 3,142,000 189,000 214,000 297,000
6 – 12 68,000 321,000 564,000 - - 645,000 191,000 222,000
12 – 18 134,000 321,000 564,000 618,000 - 645,000 191,000 333,000
18 – 24 134,000 321,000 564,000 618,000 395,000 645,000 191,000 333,000
24 – 30 134,000 321,000 564,000 - 395,000 161,000 191,000 333,000
30 – 36 134,000 321,000 - - 175,000 161,000 - 333,000
36 - 42 134,000 - - - 175,000 161,000 - 333,000
4.2. NUS-BACK Results
Figure 34 provides a graphical illustration of the NUS-BACK estimates for EPCC using
different combinations of sensor spacing and subgrade modeling. Figure 35 provides the
same illustration for estimates of k and ES.
92
FIGURE 34 Backcalulated EPCC from NUS-BACK
93
FIGURE 35 Backcalculated Soil Stiffness from NUS-BACK
94
4.3. AREA72 Results
The AREA72 backcalculation procedure was also used to estimate EPCC and k for the
Nashville pavement and subgrade. Figure 36 shows the estimates for EPCC, and Figure 37
shows those for k.
FIGURE 36 Backcalculated EPCC from AREA72
FIGURE 37 Backcalculated k from AREA72
95
4.4. LTPP Forward Calculation Results
The LTPP forward calculation spreadsheet is capable of estimating the resilient modulus
and composite pavement modulus for rigid pavements. Figure 38 shows the LTPP
estimations for both.
FIGURE 38 Mr and EComp from LTTP Forward Calculation
4.5. BAKFAA Results
Pavement and subgrade moduli were also estimated using BAKFAA, which requires seed
moduli for the subgrade and the pavement. Two seed values were used for each input,
and the results are given in Figure 39. Seed values for subgrade modulus were 15,000 psi
and 20,000 psi. Those for pavement modulus were 4 million psi and 5 million psi.
96
FIGURE 39 ES and EPCC from BAKFAA
97
4.6. ELMOD Results
ELMOD 5 backcalculation results are presented in Figure 40 using both the radius of
curvature and the basin matching methods.
FIGURE 40 ES and EPCC from ELMOD
4.7. Comparisons
Figure 41 compares the effectiveness of different sensor spacings for estimating EPCC in
NUS-BACK for both dense liquid and elastic solid foundations.
98
FIGURE 41 NUS-BACK: D3-D1 vs. D7-D4 for EPCC
Figure 42 compares the different sensor spacings for estimating k (for the dense
liquid model) and ES (for the elastic solid model).
FIGURE 42 NUS-BACK: D3-D1 vs. D7-D4 for k, Es
Figure 43 illustrates the variability in the dense liquid and elastic solid models for
estimating EPCC.
99
FIGURE 43 NUS-BACK: Dense Liquid vs. Elastic Solid for EPCC
Similarly, Figure 44 compares the ability of the two subgrade models to estimate
subgrade stiffness. The line of equality is not 1:1 because the relationship between k and
ES is not 1:1. The reader is referred to section 5.6 for a more detailed explanation of this
topic.
FIGURE 44 NUS-BACK: Dense Liquid vs. Elastic Solid for k & ES
The NUS-BACK elastic solid model for both sensor spacings is compared to the
AREA72 model for estimations of EPCC in Figure 45.
100
FIGURE 45 NUS-BACK (Elastic Solid) vs. AREA72 for EPCC
For completeness, the average of the two NUS-BACK elastic solid models (both
D3/D1 and D7/D4) is compared to the AREA72 model for estimating EPCC in Figure 46.
FIGURE 46 NUS-BACK Average (Elastic Solid) vs. AREA72 for EPCC
Figure 47 compares the estimations of k from the NUS-BACK D7/D4 dense liquid
setup with those from the AREA72 method.
101
FIGURE 47 NUS-BACK D7-D4 (Dense Liquid) vs. AREA72 for k
Figure 48 compares the estimations of subgrade and pavement stiffness from
ELMOD 5 according to which optimization procedure is used, basin-matching or radius
of curvature.
FIGURE 48 ELMOD: Radius of Curvature vs. Basin-Matching
The NUS-BACK D7/D4 dense liquid setup is compared to BAKFAA, LTPP, and
ELMOD 5 for estimating subgrade stiffness in Figures 49, 50, and 51, respectively.
Similarly, AREA72 is compared to BAKFAA, LTPP, and ELMOD 5 in Figures 52, 53
and 54.
102
FIGURE 49 NUS-BACK D7-D4 (DL) k vs. BAKFAA ES
FIGURE 50 NUS-BACK D7-D4 (DL) k vs. LTPP Mr
FIGURE 51 NUS-BACK D7-D4 (DL) k vs. ELMOD ES
103
FIGURE 52 AREA72 k vs. BAKFAA ES
FIGURE 53 AREA72 k vs. LTPP Mr
FIGURE 54 AREA72 k vs. ELMOD ES
104
The subgrade stiffness estimates from BAKFAA and LTPP are presented in
Figure 55. Figure 56 shows the same for BAKFAA and ELMOD 5, and Figure 57 makes
the comparison for LTPP and ELMOD 5.
FIGURE 55 BAKFAA ES vs. LTPP Mr
FIGURE 56 BAKFAA ES vs. ELMOD ES
105
FIGURE 57 LTPP Mr vs. ELMOD ES
Estimations of the different procedures for EPCC are presented in figures 58 – 63.
The estimates of AREA72 are compared to those from BAKFAA, LTPP, and ELMOD 5
in Figures 58, 59, and 60, respectively. The BAKFAA estimates of EPCC are plotted
against those from LTPP and ELMOD 5 in Figures 61 and 62, respectively. Finally, a
comparison between LTPP and ELMOD 5 for estimating EPCC is given in Figure 63.
FIGURE 58 AREA72 EPCC vs. BAKFAA EPCC
106
FIGURE 59 AREA72 EPCC vs. LTPP Ecomp
FIGURE 60 AREA72 EPCC vs. ELMOD EPCC
FIGURE 61 BAKFAA EPCC vs. LTPP EComp
107
FIGURE 62 BAKFAA EPCC vs. ELMOD EPCC
FIGURE 63 LTTP EComp vs. ELMOD EPCC
108
5. DISCUSSION
5.1. SASW Results
SASW results are presented in Table 12 of Section 4.1. The subgrade stiffness is
consistently overestimated, and less than half of the estimates of EPCC are reasonable. In
fact, only one dispersion curve (Figure A2, Slab 17) is actually considered reasonable. It
is not clear exactly why SASW performed so poorly at the Nashville site, but it is evident
that SASW did not provide any valuable estimates for the Nashville site. This could have
been due to human error, but the consistently unreasonable estimates suggest that SASW
was not appropriate for the Nashville pavement. This could have been because of voids
underneath the pavement that interfered with wave propagation, or simply because
SASW is not an effective tool for analyzing rigid pavements.
5.2. NUS-BACK
NUS-BACK is capable of modeling a foundation as dense liquid or elastic solid, and it
can calculate layer moduli based on any two of seven measured deflections. The
literature suggests that using the deflection measured directly under the load plate along
with the third measured deflection (D1 and D3, respectively) is best for estimating the
pavement stiffness since those two deflections are due almost entirely to pavement
response. Similarly, D7 and D4 are considered to work best for estimating subgrade
moduli. Thus, NUS-BACK, on its own, provides many possible ways to backcalculate
rigid pavement and subgrade stiffness.
Figures 34 and 35 in Section 4 show a graphical representation of the
backcalculated values for pavement and subgrade stiffness, respectively, for different
109
combinations of the previously discussed methods. The EPCC values were unreasonably
large for the dense liquid model, and only moderately reasonable (between 2 and 3
million psi) for the D3/D1 layout of the elastic solid model. The elastic solid D3/D1 model
consistently predicted values of EPCC roughly 2 million psi lower than the laboratory
estimated moduli.
The ES values were more consistent throughout the different subgrade models and
sensor spacings. The most consistent results, however, were observed when the dense
liquid model was used. The dense liquid model gave k values mostly between 100 and
200 psi/in for both D3/D1 spacing and D7/D4 spacing. Unexpectedly, the D3/D1 spacing
estimated the most reasonable values for ES with the elastic solid model, having a range
of 5,000 to 10,000 psi. Further comparisons within the NUS-BACK results are discussed
in section 5.6.
5.3. AREA72
The AREA72 method provides an estimation of the coefficient of subgrade reaction (k)
and the subgrade stiffness (EPCC). The estimate is made by taking the average of seven
individual layer moduli estimations, one for each deflection measurement. Figures 36
and 37 in Section 4 show a graphical representation of the backcalculated moduli for the
pavement and subgrade, respectively. The coefficients of subgrade reaction vary
between about 75 and 125 psi/in., and the PCC moduli mostly fall between 2.5 and 4.5
million psi. The estimates of EPCC were consistently 1 million psi below the laboratory
estimate of EPCC. Both of these ranges are within reason for the Nashville pavement
system.
110
5.4. LTPP Forward Calculation
The LTPP forward calculation spreadsheet was used to estimate moduli for 30 selected
deflection basins from the Nashville project. The NUS-BACK and AREA72 results were
used to screen the database of Nashville data for basins that were most likely to yield
reasonable estimates of layer moduli. Basins were chosen that corresponded to NUS-
BACK and AREA72 estimates between 115 and 250 psi/in. for k and between 3 million
and 6 million psi for EPCC. The subgrade resilient modulus (Mr) and composite pavement
stiffness (Ecomp) that were estimated for those selected basins are illustrated graphically in
Figure 38. The Mr estimates range between about 18,000 and 25,000 psi with good
consistency. The composite pavement stiffness should be a close estimation of the PCC
modulus since the Nashville pavement was only comprised of that single layer. The
range of EComp from forward calculation was about 400,000 to 600,000 psi with a few
exceptions around 1 million psi. These estimates are unreasonably low for PCC stiffness.
5.5. BAKFAA
BAKFAA was used to analyze the same 30 deflection basins that were loaded into the
LTPP forward calculation spreadsheets. BAKFAA is interation-based software that
relies on user-input values for seed moduli. Thus, unlike the previous closed-form
methods, BAKFAA allows the user to alter the final estimate of moduli by using different
seed values. Figure 39 shows the estimates that BAKFAA provides for various seed
moduli. The EPCC ES seeds were chosen based on the results of NUS-BACK and
AREA72 backcalculations. Accordingly, ES seeds of 15,000 and 20,000 psi were used,
111
and EPCC seeds of 4 million and 5 million psi were used. The corresponding
combinations and results are shown graphically in Figure 39.
At least for the variation in the seed values that were chosen, BAKFAA did not
appear to be very sensitive to differences in seed values. For all 8 trial combinations the
ES estimate ranged between 19,000 and 24,000 psi fairly consistently. Similarly, the EPCC
estimates ranged between roughly 3 and 4 million psi. The EPCC estimates are between 1
million and 2 million psi lower than the laboratory estimated moduli. The effect of
averaging the basins from different load levels did not have any significant effect. The
backcalculated moduli from the averaged basins are not noticeably different than those
from the true basins.
5.6. ELMOD
ELMOD 5.0 was used to backcalculate the layer moduli straight from the FWD output
files. In practice, this feature makes ELMOD significantly more convenient than the
other methods, because deflection basins are not entered manually by individual
deflections and sensor spacings. The data is included in the .F25 file, and ELMOD
provides the pertinent output as a Microsoft Access database file. As was mentioned in
the literature review, ELMOD is capable of backcalculating moduli by matching either
the radius of curvature of the basin or by matching the basin itself. Figure 40 summarizes
the results from both techniques for ES and EPCC.
While there was significant scatter for the different basins, the overall results did
not vary much based solely on the backcalculation technique. Furthermore, the scatter is
likely due to inconsistent pavement distresses rather than errors in the backcalculation
112
program. The estimates for subgrade modulus were mostly between 5,000 and 10,000
psi. Those for PCC modulus were mostly between 2 million and 5 million psi. These
ranges are both reasonable for the Nashville pavement and subgrade. However, the EPCC
estimates are consistently lower than the laboratory estimates.
5.7. NUS-BACK Variations
Considering the different sensor arrangements and pavement modeling schemes that are
possible with NUS-BACK, there are many different potential ways for NUS-BACK to
backcalculate FWD data. If the NUS-BACK software were being used, it would be
feasible to estimate moduli from 21 different 2-sensor arrangements, and each of those
arrangements could be modeled as elastic solid or dense liquid. Since the graphical
procedure was used for this study, the 2-sensor arrangements had to be limited for
practicality. The two chosen sensor arrangements were D3/D1 and D7/D4. The literature
suggests that these two spacings give the best idea of pavement and subgrade stiffness,
respectively. All 260 deflection basins from the Nashville project were analyzed in this
manner.
Figure 41 in Section 4.7 compares the values that NUS-BACK estimated for EPCC
using D3/D1 and D7/D4. Points lying on the dashed diagonal line represent deflection
basins for which D3/D1 and D7/D4 both predict the same pavement stiffness. From the
figure, it is clear that the D3/D1 configuration estimates much higher PCC moduli than the
D7/D4 configuration for both dense liquid and elastic solid models. Figure 42 shows a
similar comparison for subgrade stiffness instead of pavement stiffness. In this case, the
sensor arrangement makes little difference for the dense liquid model, but for the elastic
113
solid model the D7/D4 model is observed to estimate higher moduli than the D3/D1
configuration.
Another important comparison to make for NUS-BACK procedures is the relative
strength of the dense liquid and elastic solid models. Figure 43 compares the two
models’ estimates of pavement stiffness. The figure suggests that the dense liquid model
predicts significantly larger EPCC values than the elastic solid model. In addition, it can
be seen that the values predicted by the dense liquid are unreasonably large.
For the comparison of k and ES, two equality lines were initially drawn to
represent Equations 4 and 73. It was quickly seen, however, that the difference in the
two relationships was insignificant compared to the scatter of data and the two lines laid
so close together that they could not be differentiated. Thus, for the remainder of this
study, a single equality line was drawn for all comparisons between k and ES that
corresponds to ES = 9.55k. Figure 44 shows the comparison between backcalculated ES
and k for the dense liquid and elastic solid models. Interestingly, the dense liquid model
predicted higher values for the D3/D1 spacing while the elastic solid model predicted
higher values for the D7/D4 spacing.
5.8. NUS-BACK vs. AREA72
Comparisons were made between AREA72 and several variations of NUS-BACK for
estimating pavement and subgrade moduli. The comparison of Section 5.6 indicates that
the dense liquid model is not appropriate for estimating the moduli of pavement layers in
NUS-BACK. Therefore, only the elastic solid model was used within NUS-BACK for
EPCC backcalulation.
114
Figure 45 compares the both NUS-BACK arrangements to the AREA72 model for
estimating pavement stiffness. Figure 46 does the same, but takes the average of the
D3/D1 and D7/D4 arrangements for NUS-BACK. The results show that the D7/D4
arrangement causes NUS-BACK and AREA72 to show the greatest agreement for EPCC
backcalculation. The average NUS-BACK estimates also show fairly close agreement
with AREA72 estimates for PCC modulus. It is noted that the two comparisons including
D3/D1 show the least scatter and indicate an underpriction by AREA72, while the D7/D4
NUS-BACK estimates seem to show an overprediction by AREA72. These results are not
significant enough to indicate which method is best, but they do indicate that NUS-
BACK and AREA72 report EPCC values fairly close to one another.
In addition, NUS-BACK was compared to AREA72 for the ability to estimate the
subgrade stiffness. Figure 47 shows the dense liquid D7/D4 NUS-BACK estimates
plotted against the AREA72 estimates. This relationship shows a high degree of
continuity, with NUS-BACK consistently and slightly predicting higher soil moduli than
AREA72. Again, this does not indicate whether one method is more accurate than the
other, but it shows that the two methods estimate very similar ranges of moduli for the
same deflection basins.
5.9. ELMOD Variations
It was noted in Section 5.5 that the two backcalculation options with ELMOD do not
have a significant impact on the final results. The two options are to simulate the radius
of curvature of the deflection basin or to replicate the basin based on the individual
deflections. Figure 48 shows the effect of these options for estimating both subgrade and
115
PCC stiffness. The figure shows a significant amount of scatter for both cases, but in
both cases the scatter is centered on the line of equality.
5.10. Other Comparisons for Estimating Subgrade Stiffness
NUS-BACK D7/D4 (DL) and AREA72 were found to give consistent and reasonable
results for the coefficient of subgrade reaction. The other backcalculation programs do
not give k as a direct output, but the soil modulus (ES) was compared to k from NUS-
BACK and AREA72 by the equation discussed in Section 5.6.
Figure 49 compares NUS-BACK with BAKFAA for estimating k and ES,
respectively. The figure shows that, according to the line of equality, BAKFAA predicts
significantly higher soil moduli than NUS-BACK. Figure 50 compares NUS-BACK to
the LTPP forward calculation spreadsheet for estimating k and Mr, respectively. In this
case, the graph again suggests that NUS-BACK is estimating lower values than the
compared technique. Figure 51 compares NUS-BACK to ELMOD for estimating soil
stiffness. This case continues the trend of underestimation by NUS-BACK, but the
agreement is much better than for the BAKFAA or LTPP methods. As expected, Figures
52 – 54 show very similar results for comparisons in which the NUS-BACK data was
replaced by AREA72 data.
In order to further assess the various techniques for estimating subgrade stiffness,
several additional comparisons were drawn. Figure 55 shows a direct comparison
between the BAKFAA and LTPP methods for estimating ES and Mr, respectively. The
two methods show a consistent agreement with the LTPP method predicting slightly
lower values than BAKFAA. Figure 56 then compares BAKFAA to ELMOD. In this
116
case, BAKFAA again predicts the higher values, but the agreement is less consistent and
more scattered. Finally, LTPP and ELMOD are compared in Figure 57. The figure
shows a mild agreement between the two methods, with ELMOD consistently predicting
lower values than the LTPP spreadsheet. This shows that ELMOD predicted lower
values than BAKFAA and the LTPP spreadsheet for soil stiffness.
5.11. Other Comparisons for Estimating PCC Stiffness
AREA72 was compared to the other studied methods for estimating PCC modulus.
Figure 58 shows such a comparison between AREA72 and BAKFAA with excellent
agreement between the two data sets. Figure 59 compares AREA72 to LTPP, but with
less agreement in the data. The LTPP spreadsheet is observed to predict a lower value
for EPCC than AREA72 in almost every case. Figure 60 shows close agreement between
the estimated EPCC values from AREA72 and ELMOD. As with the other ELMOD plots,
there is a good deal of scatter, but the scatter is centered on the line of equality.
To further investigate these techniques, BAKFAA was compared to the LTPP
spreadsheet and ELMOD in Figures 61 and 62, respectively. Figure 61 shows little
agreement between the LTPP and BAKFAA data, and it suggests that the LTPP
spreadsheet estimated lower values than BAKFAA. Figure 62 shows a wide range of
scatter for the BAKFAA and ELMOD estimates of EPCC, but the data is centered on the
line of equality. Finally, the LTPP and ELMOD estimates were compared in Figure 63.
The figure shows relatively little agreement, and shows again that the LTPP spreadsheet
tends to predict lower values than other similar procedures.
117
6. CLOSING
The purpose of this project was to analyze and compare backcalculation techniques for
rigid and flexible pavements. Most of the techniques discussed have been in terms of the
falling weight deflectometer (FWD), but the spectral analysis of surface waves (SASW)
is also a viable approach. The literature review suggests that DIPLO-DEF is the most
flexible program for analyzing FWD field data, and that automated procedures are being
developed to ease the process of SASW inversion.
Field data for this project was limited to raw FWD and SASW data from a rigid
pavement in Nashville, AR. The raw FWD data was used to perform backcalculations
with five different procedures. DIPLO-DEF and MODULUS both seemed promising
from the literature review, but the input/output process in a DOS-based environment
proved too difficult to be practical for this project. The SASW field dispersion curves
appeared to be flawed by some aspect of the field conditions. It is likely that air pockets
existed between the concrete pavement and the subgrade which inhibited the regular
propagation of Rayleigh waves. Nevertheless, the field dispersion curves were inverted
by a simplified method.
Closed form solutions for rigid pavements were found to be very useful during
this study because they facilitate the development of a user-friendly spreadsheet for
backcalculation. ELMOD proved even more convenient by allowing the user to input a
raw FWD file instead of entering individual basin dimensions. NUS-BACK and AREA72
were analyzed via spreadsheet; LTPP forward calculations were performed using a
publically available spreadsheet; BAKFAA was run by entering basin dimensions
118
through an easy-to-use graphical user interface; ELMOD was run by inputting raw FWD
files.
The SASW results were not ideal, and a larger library of field data would likely
give more reasonable and valuable results for comparison with FWD backcalculations.
The simplified procedure that was used for this study is valuable for comparison with
other testing methods, but an analysis using more advanced inversion procedures would
also be practical to gauge the effect of automation on the final SASW-estimated layer
moduli.
The discussion in Section 5 can be summarized by the following findings:
• The simplified SASW method was an ineffective and likely inappropriate tool for
analyzing the rigid pavement in Nashville.
• The backcalculation procedures consistently estimated EPCC values between 1
million and 2 million psi lower than the laboratory estimation. This could be due
to flawed laboratory data.
• NUS-BACK D7/D4 with dense liquid model and AREA72 are equally effective at
estimating reasonable values for the coefficient of subgrade reaction, k, for rigid
pavements. However, they both appear to predict much lower values than the
other methods that were studied.
• The LTPP forward calculation spreadsheet for rigid pavements consistently
underpredicts values of Mr and EComp
• BAKFAA is an effective tool for estimating PCC moduli, but it seems to
overpredict soil stiffness. This could be remedied by changing the values used for
seed moduli.
119
• ELMOD reflects the scatter that is inherent in the field, but the overall quality of
data is more consistent and reasonable than any of the other methods.
It is stressed that there is no “right” way to estimate in-situ pavement and soil
properties. Every method is “wrong” to some extent, but the art of backcalculation
requires the engineer to tweak whatever method(s) he or she is using to get the most
reasonable results without resorting to destructive and expensive laboratory testing.
The engineer’s knowledge of typical values, reasonable assumptions, and common
sense is the most important element of backcalculation. That being said, FWD has
shown a greater applicability to the type of pavement that was tested in this study,
jointed reinforced concrete pavement (JRCP).
Since laboratory modulus tests were not available for the materials on which the
FWD and SASW tests were run, the current study cannot assess the relative accuracy
of the backcalculation procedures, but only their relative performance. Future studies
would optimize their value by performing laboratory resilient modulus tests for a
more valid comparison. In addition, the LTPP forward calculation, BAKFAA,
ELMOD, and SASW could have been more adequately compared if raw data had
been obtained for flexible or overlaid pavements. Nevertheless, the current study
provides a detailed comparison of a simplified SASW inversion and several FWD
backcalculation techniques for assessing the stiffness of rigid pavements and
subgrade soils.
120
REFERENCES
1. ASTM Designation: D 4694 – 96 (Reapproved 2003). Standard Test Method for

Deflections with a Falling-Weight-Type Impulse Load Device.
2. Lytton, R.L, F.P. Germann, Y.J. Chou, and S.M. Stoffels. “Determining Asphaltic
Concrete Pavement Structural Properties by Nondestructive Testing”, NCHRP
Report 327, Transportation Research Board, Washington, D.C., 1990.
3. Selvadurai, A.P.S. “Elastic Analysis of Soil-Foundation Interaction”, Developments

In Geotechnical Engineering, vol 17, Elsevier Scientific Publishing Company, 1979.
4. Jumikis, A.R. “Theoretical Soil Mechanics”, American Book Company, 1969.
5. Winkler, E. Theory of Elasticity and Strength. Prague: H. Dominicus, 1867 (in

German).
6. Westergaard, H.M. “Stresses in Concrete Pavements Computed by Theoretical

Analysis,” Public Roads, 7, No. 2, 23-35 (1926).
7. Scott, R.F. “Foundation Analysis”, Prentice-Hall, Inc., Englewood Cliffs, NJ, 1981.
8. Boussinesq, J. Application des Potentiels a l’Etude de l’ Equilibre et du Mouvement

des Solides Elastique. Gauthier-Villars, Paris, 1885.
9. Panc, V., Theories of Elastic Plates, Noordhoff International Publishing, Leyden,

Czechoslovakia, 1975.
10. Losberg, A. “Structurally Reinforced Concrete Pavement”, Doktorsavhandlingar Vid

Chalmers Tekniska Hogskola, Goteborg, Sweden, 1960.
11. Hall, K.T., C.E. Correa, S.H. Carpenter, and R.P. Elliott. Rehabilitation Strategies for
Highway Pavements. National Cooperative Highway Research Program, Web
Document 35 (Project C1-38), May 2001.
12. National Highway Institute. Portland Cement Concrete Pavement Evaluation and
Rehabilitation. FHWA-NHI-131062. CD-ROM. NHI, Virginia, 2001.
13. Strategic Highway Research Program: SHRP’s Layer Moduli Backcalculation

Procedure, Strategic Highway Research Program, December 1993.
14. Stubstad, R.N., Jiang, Y.J., and E.O. Lukanen. “Guidelines for Review and
Evaluation of Backcalculation Results”, Long-Term Pavement Performance Program,
FHWA-HRT-05-152, 2006.
121
15. Khazanovich, L., McPeak, T.J., and S.D. Tayabji. “LTPP Rigid Pavement FWD
Deflection Analysis and Backcalculation Procedure”, Nondestructive Testing of
Pavements and Backcalculation of Moduli: Third Volume, ASTM STP 1375, S.D.
Tayabji and E.O. Lukanen, Eds., American Society for Testing and Materials, West
Conshohocken, PA, 2000.
16. “AASHTO Guide for Design of Pavement Structures”, American Association of State
Highway and Transportation Officials, Washington, D.C., 1993.
17. Hall, K.T., Darter, M.I., and C.M. Kuo. “Improved Methods for Selection of k Value
for Concrete Pavement Design”, Transportation Research Record 1505, pp. 128-136,
Washington, D.C., 1995.
18. Netemeyer, R. and S. Munsell. “Layer By Layer Evaluation of a PCC Pavement As It

Was Constructed. Comparison of Backcalculated Material Values to Lab Material
Values”, RDT-99-002,Missouri Department of Transportation, Material and Research
Division, 1995.
19. Huang, Y.H. Pavement Analysis and Design, Prentice Hall, pp. 316-317, 1993.
20. Ioannides, A.M. ILLI-BACK: Experimental Version 3.3 User’s Manual, 1990.
21. Fwa, T.F., Tan, K.H., and S. Li. “Closed-Form and Semi-Closed-Form Algorithms
for Backcalculation of Concrete Pavement Parameters” Nondestructive Testing of
Pavements and Backcalculation of Moduli: Third Volume, ASTM STP 1375, S.D.
Tayabji and E.O. Lukanen, Eds., American Society for Testing and Materials, West
Conshohocken, PA, 2000.
22. Fwa, T.F. and B.H. Setiadji. “Evaluation of Backcalculation Methods for
Nondestructive Determination of Concrete Pavement Properties”, Transportation
Research Record: Journal of the Transportation Research Board, No. 1949,
Transportation Research Board of the national Academies, Washington, D.C., 2006,
pp. 83-97.
23. Burmister, D.M. “The General Theory of Stresses and Displacements in Layered
System”, Journal of Applied Physics, Vol. 16, pp. 89-94 and 126-127, 1945.
24. Khazanovich, L. “Dynamic Analysis of FWD Test Results for Rigid Pavements”,
Nondestructive Testing of Pavements and Backcalculation of Moduli: Third Volume,
ASTM STP 1375, S.D. Tayabji and E.O. Lukanen, Eds., American Society for
Testing and Materials, West Conshohocken, PA, 2000.
25. McPeak, T.J and L. Khazanovich. “Backcalculation of Asphalt-Overlaid Concrete

Pavements”, Proceedings, Aircraft/Pavement Technology: In the Midst of Change,
Seattle, Washington, American Society of Civil Engineers, 1997.
122
26. Hogg, A.H.A. “Equilibrium of a Thin Plate on an Elastic Foundation of Finite
Depth”, Philosophical Magazine, Vol. 35 (243), pp. 265-276, 1944.
27. Wiseman, G. and J. Greenstein. “Comparison of Methods of Determining Pavement

Parameters from Deflection Bowl Measurements”, Proceedings of the 7th Asian
Regional Conference on Soil Mechanics and Foundation Engineering, pp. 158-165,
1983.
28. ELMOD® 5. Dynatest International.

http://www.dynatest.com/downloads/2007jan/Elmod5%2025102006.pdf. Accessed
March 19, 2008.
29. Appeah, A.K. and L. Al-Qadi. “Assessment of Falling Weight Deflectometer Data for
Stabilized Flexible Pavements”, Transportation Research Record 1709, pp. 19-25,
Washington, D.C., 2000.
30. Federal Aviation Administration. “R&D Review”, TC07-0045, Issue 2, pp. 13-14,
2007.
31. Gopalakrishnan, K. and M.R. Thompson. “Comparative effect of B777 and B747
Trafficking On Elastic Layer Moduli of NAPTF Flexible Pavements”, 2004 FAA
Worldwide Airport Technology Transfer Conference, Atlantic City, New Jersey,
April 2004.
32. Hayhoe, G.F. “LEAF – A New Layered Elastic Computational Program for FAA
Pavement Design and Evaluation Procedures”, 2002 Federal Aviation Administration
Technology Transfer Conference, 2002.
33. Kothandram, S. and A.M. Ioannides. “DIPLODEF: A Unified Backcalculation

System for Asphalt and Concrete Pavements”, Transportation Research Record 1764,
pp. 20-29, Washington, D.C., 2001.
34. Khazanovich, L. and A.M. Ioannides. “DIPLOMAT: Analysis Program for

Bituminous and Concrete Pavements”, Transportation Research Record 1482, pp. 52-
60, Washington, D.C., 1995.
35. Nazarian, S. and K. H. Stokoe II. “Use of Surface Waves in Pavement Evaluation” In
Transportation Research Record 1070, pp. 132-144, Washington, D.C., 1986.
36. Rosenblad, B., E. M. Rathje, and I. I. Stokoe. “Shear Wave Velocity Profiling by the
SASW Method at Selected Strong-Motion Stations in Turkey”, 99?
37. HP 35670A Product Overview.

http://www.mae.wmich.edu/Labs/Noise_and_Vibration/HP_analyzer.pdf. Accessed
February 19, 2008.
123
38. Gucunksi, N., and R. D. Woods. “Instrumentation for SASW Testing”, Geotechnical
Special Publication, n 29, Recent Advances In Instrumentation, Data Acquisition and
Testing In Soil Dynamics, October 21, 1991, pp. 1-16.
39. Rix, G. J. and E. A. Leipski. “Accuracy and Resolution of Surface Wave Inversion”,
Geotechnical Special Publication, n 29, Recent Advances In Instrumentation, Data
Acquisition and Testing In Soil Dynamics, October 21, 1991, pp. 17-32.
40. Goel, A. and A. Das. “A Brief Review on Different Surface Wave Methods and Their
Applicability for Non-Destructive Evaluation of Pavements”, Nondestructive Testing
and Evaluation, January 2008, pp. 337-350.
41. Ryden, N., Ulriksen, P., Park, C.B., Miller, R.D., Xia, J., and J. Ivanov. “High
Frequency MASW for Non-Destructive Testing of Pavements – Accelerometer
Approach”, Proceedings of the SAGEEP, Denver, Colorado, RBA-5, 2001.
42. Kausel, E. and J.M. Roesset. “Stiffness Matrices for Layered Soils”, Bulletin of the
Seismological Society of America, Vol. 71, No. 6, p. 1743-1761, 1981.
43. Barfield, O.E. “Mechanistic Procedure for Prediction of the In-situ Resilient Modulus
of Unbound Aggregate Base from Seismic Testing”, Thesis for Masters of Science in
Engineering, The University of Texas at Austin, 2006.
44. Darendeli, M.B., Stokoe, K.H., II, and R. Gilbert. “Development of a New Family of
Normalized Modulus Reduction and Material Damping Curves”, Ph.D., University of
Texas at Austin, 2001.
45. Meshkani, A., Abdallah, I., and S. Nazarian. “Determination of Nonlinear Parameters
of Flexible Pavement Layers from Nondestructive Testing”, Conducted for TXDOT,
Research Report TX 1780-3, 2004.
46. Ishibashi, I. and X. Zhang. “Unified Dynamic Shear Moduli and Damping Ratios of
Sand and Clay”, Soils and Foundations, Vol. 33, No. 1, pp. 182-191, 1993.
47. Zhou, H. “Comparison of Backcalculated and Laboratory Measured Moduli on AC

and Granular Base Layer Materials”, Nondestructive Testing of Pavements and
Backcalculation of Moduli: Third Volume, ASTM STP 1375, S.D. Tayabji and E.O.
Lukanen, Eds., American Society for Testing and Materials, West Conshohocken,
PA, 2000.
48. Web Soil Survey. Natural Resources Conservation. United States Department of
Agriculture. http://websoilsurvey.nrcs.usda.gov/app/. Accessed March 22, 2008.
49. ASTM Designation: D 6758-02. Test Method for Measuring Stiffness and Apparent
Modulus of Soil and Soil-Aggregate In-Place By an Electro-Mechanical Method.
124
APPENDIX A RAW DATA FROM SASW TESTING
125
FIGURE A1 Nashville, AR – AR4/US278/Main St – Slab 3
TABLE A1 Analysis of Slab 3
Depth Measured VS Assumed Assumed γ Gseis G

E (psi)
(in.) (ft/s) υ (pcf) (psi) (psi)
0–6 - - - - - -
6 – 11 1,500 0.40 115 55,000 24,200 67,760
11 – 40 2,100 0.40 115 109,000 47,960 134,288
126
Depth Measured VS Assumed Assumed γ

Gseis (psi) G (psi) E (psi)
(in.) (ft/s) υ (pcf)
0–6 8,700 0.15 150 2,448,000 1,077,000 2,478,000
6 – 36 3,600 0.40 115 321,000 141,240 395,472
127

0–6 9,700 0.15 150 3,043,000 3,043,000 6,999,000
6 – 28 4,300 0.40 115 458,000 201,520 564,256
128
Depth Measured Assumed Assumed γ

(in.) VS (ft/s) υ (pcf)
0–6 12,500 0.15 150 5,054,000 5,054,000 11,624,000
6 – 10 - - - - - -
10 – 23 4,500 0.40 115 502,000 220,880 618,464
129

0–6 6,500 0.15 150 1,366,000 1,366,000 3,142,000
6 – 15 - - - - - -
15 – 30 3,600 0.40 115 321,000 141,240 395,472
30 – 40 2,400 0.40 115 142,000 62,480 174,944
130
Depth Measured VS Assumed Assumed γ Gseis

G (psi) E (psi)
(in.) (ft/s) υ (pcf) (psi)
0–6 1,600 0.15 150 82,000 82,000 189,000
6 – 23 4,600 0.40 115 524,000 230,560 645,568
23 – 40 2,300 0.40 115 131,000 57,640 161,392
131
Depth Measured VS Assumed Assumed γ Gseis G

E (psi)
(in.) (ft/s) υ (pcf) (psi) (psi)
0–6 1,700 0.15 150 93,000 93,000 214,000
6 – 28 2,500 0.40 115 155,000 68,200 190,960
132
Depth Measured VS Assumed Assumed γ Gseis

G (psi) E (psi)
(in.) (ft/s) υ (pcf) (psi)
0–6 2,000 0.15 150 129,000 129,000 297,000
6 – 13 2,700 0.40 115 180,000 79,200 221,760
13 – 40 3,300 0.40 115 270,000 118,800 332,640
133
APPENDIX B RAW DATA FROM FWD TESTING
134
Table B1 Raw FWD Data from Nashville, Northbound Lane
135
Table B2 Raw FWD Data from Nashville, Northbound Lane (Continued)
136
Table B3 Raw FWD Data from Nashville, Southbound Lane
137
Table B4 Raw FWD Data from Nashville, Southbound Lane (Continued)
138
APPENDIX C EXAMPLE SASW INVERSION
139
The purpose of this appendix is to provide a transparent documentation of the
method by which moduli were obtained from SASW tests. As was described in the
literature review, SASW field tests involve measuring the phase velocity of Rayleigh
waves for various seismic disturbances at several different sensor spacings. Wider sensor
spacings (i.e. 24-in. instead of 6-in.) are used to measure relatively low-frequency waves
that correspond to deeper sampling depths. Figure C1 shows an idealized field dispersion
curve for a pavement structure.
Note that the data points from smaller sensor spacings correspond to shallower
depths than the data points from larger sensor spacings. Also notice that the field
dispersion curve exhibits a decrease in shear wave velocity with increasing depth. This is
because, for a pavement system, the stiffest layers are at the top. If a similar test were
run on bare soil for the design of a new roadway, the field dispersion curve would have a
different shape. There would not be a stiff layer on top, and the soil stiffness would
likely increase with depth due to increased confining pressure. The range of shear wave
velocities for this scenario would be lower than what is measured for pavement layers,
and the general shape would be close to a lateral reflection of the curve in Figure C1.
The process now described may be thought of as a simplified inversion process,
as mentioned in the literature review. In this process, the first step is to identify depths at
which the dominant shear wave velocity exhibits an abrupt change. Such changes are
indicative of a change in the material through which the shear wave is traveling. If the
depth of the pavement layer(s) is already known, this information should be used to aid in
the interpretation of the field dispersion curve. Especially if the shear wave velocity
changes uniformly, minimally, haphazardly, or in any fashion that hinders inversion,
140
Figure C1 Idealized Field Dispersion Curve
the known layer thicknesses should be used to aid in interpretation.
After the field dispersion curve has been discretized into sections of semi-uniform
shear wave velocity a representative shear wave velocity is estimated for each discretized
layer. More complex inversion methods utilize stiffness matrices and the analytic
minimization of corresponding error functions for this step. Such methods are
convenient for automation, but unless they enhance the resolution of the field dispersion
curve, they will not ultimately enhance the accuracy of the estimated layer moduli. In the
present study layer moduli were estimated by “eyeballing” the field dispersion curve and
141
drawing in sections of vertical lines (representing constant shear wave velocity through a
depth) that best and most realistically represent the field data.
Table C1 summarizes the data that can be taken from the hypothetical field
dispersion curve in Figure C2. The next step for data reduction is to either measure,
estimate or assume the quantities that are needed to relate shear wave velocity to Young’s
Modulus. These quantities include unit weight and Poisson’s ratio. If the unit weight of
the material in any given layer is unknown, the probable constituent of that layer should
be guessed and its unit weight assumed. Poisson’s Ratio is not usually measured, and it
may be estimated using sources such as Table 2 in the literature review or estimated
using engineering judgment.
VS ≈ 17,000 ft/s
VS ≈ 7,000 ft/s
VS ≈ 5,500 ft/s
VS ≈ 4,000 ft/s
Figure C2 Discretized Field Dispersion Curve
142
After measuring, estimating or assuming the necessary material characteristics,
the shear wave velocity is converted into a value for seismic shear modulus (G) using
Equation 60 from the literature review, shown below. The mass density may be
calculated by dividing the material’s unit weight by gravity. Then, Young’s modulus is
estimated from the assumed value for Poisson’s ratio using Equation 61 from the
literature review, also shown below. As discussed in the literature review, seismic values
for Young’s modulus may be taken as equal to resilient modulus (Mr).
G = VS ⋅ ρ ………………………….……Eq. 60
2
E = 2G (1 + ν ) ………………………….…Eq. 61
Lastly, Mr is adjusted to account for the stress-sensitive behavior of soil moduli.
For this study, Mr was adjusted using Darendeli’s model for fine-grained soils (44). This
relationship is defined in the literature review by Equations 62 and 64, and is repeated
below. This model (as would similar models) is used to generate a curve for G/Gmax
versus strain.
G 1
= a
……………………..…….Eq. 62
Gmax ⎛γ ⎞
1 + ⎜⎜ ⎟⎟
⎝γr ⎠
γ r = (0.0352 + 0.0010 × PI × OCR 0.3246 )⋅ σ 0'

0.3483
; a = 0.9190 ……..Eq. 64
This model requires several additional assumptions. First, the overconsolidation
ratio is assumed equal to one. This is reasonable since the soils in Nashville, AR have
not supported glaciers or other tremendous weights in their geologic past. Secondly, the
effective confining stress was taken equal to 0.5 atmospheres. The literature suggests
143
that values in this region are common and appropriate (44). These two assumptions
facilitate the development of the graph in Figure C3, which was used to reduce all of the
moduli estimates from the Nashville site.
Figure C3 Darendeli’s Modulus Reduction Curve for Fine-Grained Soils
Next, the estimated layer modulus for each of the discretized layers is adjusted to
a FWD/Truck-loading strain using Figure C3. This may be accomplished on an
analytical basis. The first step in this process is to identify a shear strain that is
representative of FWD/Truck loading. The FWD can be used to identify such a strain.
Using a representative deflection basin, first calculate the horizontal distance from the
load plate at which deflections are due entirely to subgrade deformation. This distance is
called ae and is found using Equations 1-4. The normal strain at this point (εFWD) is
calculated as the vertical deflection divided by the subgrade sampling depth. The
144
subgrade sampling depth is taken as 1.5 times the radius of the FWD load plate minus the
depth of pavement layer(s).
εFWD may then be converted to a shearing strain (γFWD) using the principles of
Equation 61 and Mohr’s Circle. The first step is to replace the Young’s Modulus and
Shear Modulus terms in Equation 61 with their component quantities, as shown below in
Equation 61.B. Rearranging this relationship to the form of Equation 61.C gives a
solvable expression for shear strain (γ). The normal stress (σ) and strain (ε) are available
from FWD data, Poisson’s Ratio (υ) is assumed, and the shear stress (τ) may be found
using Mohr’s circle.
σ τ
= 2 (1 + ν ) ……………………..………Eq. 61.B
ε γ
2τ (1 + ν )
γ = ………………………………Eq. 61.C
E
The current analysis used a simplified procedure for determining γFWD. Instead of
an analytical estimation, a general rule of thumb was employed. That is, the FWD load
was assumed to impart a shear strain of 0.05%. This undoubtedly introduces some error
into the problem, but the assumption is that this error will not significantly affect the final
estimation.
Using γFWD as an input, Figure C3 is used to find a single reduction factor for all
of the seismically determined modulus. Figure C4 illustrates this procedure. Finally,
each of the seismically measured moduli is multiplied by the reduction percentage.
These final values are representative of the soil moduli at an appropriate load level for
highway design.
145
G/Gmax = 0.44
γFWD = 0.05%
Figure C4 Simplified Determination of γFWD
146
147

A Comparison of Nondestructive Testing Backcalculation Techniques For Rigid and Flexible Pavements

Uploaded by

Copyright:

Available Formats

A Comparison of Nondestructive Testing Backcalculation Techniques For Rigid and Flexible Pavements

Uploaded by

Document Information

Original Title

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

A Comparison of Nondestructive Testing Backcalculation Techniques For Rigid and Flexible Pavements

Uploaded by

Copyright:

Available Formats

University of Arkansas, Fayetteville

A Comparison of Nondestructive Testing

Follow this and additional works at: http://scholarworks.uark.edu/cveguht

A thesis submitted in partial fulfillment

Trenton Blake Ellis

AND FLEXIBLE PAVEMENTS

In designing new pavements, engineers rely on many different measures to characterize

Kevin D. Hall, PhD.

Norman D. Dennis, PhD.

Robert P. Elliott, PhD.

1.1. Background .......................................................................................................... 1

1.2. Problem Statement ............................................................................................... 2

1.3. Objectives ............................................................................................................. 3

1.4. Experimental Plan ................................................................................................ 3

1.4.1. Task 1: Literature Review ............................................................................. 3

1.4.2. Task 2: Data Analysis ................................................................................... 4

1.4.2.1. Software Acquisition ............................................................................. 4

1.4.2.2. Collection of Field Data ........................................................................ 5

1.4.2.3. Perform Backcalculations ...................................................................... 5

1.4.2.4. Document Results from Backcalculations ............................................. 6

1.4.3. Task 3: Report Findings ................................................................................ 6

2. LITERATURE REVIEW ........................................................................................... 7

2.1. Nondestructive Testing Background .................................................................... 7

2.2. Falling Weight Deflectometer .............................................................................. 9

2.2.1. Theoretical Background ................................................................................ 9

2.2.2. Field Procedures.......................................................................................... 12

2.2.3. Backcalculation Methods ............................................................................ 17

2.2.3.2. ILLI-BACK ......................................................................................... 23

2.2.3.3. NUS-BACK for Rigid Pavements ....................................................... 25

2.2.3.4. AREA Method for Rigid Pavements ................................................... 33

2.2.3.6. LTPP Forward Calculation .................................................................. 40

2.2.3.7. MODULUS ......................................................................................... 45

2.2.3.8. ELMOD 5 ............................................................................................ 50

2.2.3.9. BAKFAA ............................................................................................. 52

2.2.3.10. DIPLO-DEF ........................................................................................ 52

2.3. Spectral Analysis of Surface Waves .................................................................. 56

2.3.1. Theoretical Background .............................................................................. 56

2.3.2. Field Procedures.......................................................................................... 59

2.3.3. Backcalculation Procedure.......................................................................... 68

2.4. Case Studies: Comparisons of Backcalculation Procedures .............................. 76

2.4.2. LTPP Database: Comparison of FWD Backcalculation Methods .............. 81

3. DATA ANALYSIS................................................................................................... 86

4. RESULTS ................................................................................................................. 92

4.1. SASW Results .................................................................................................... 92

4.2. NUS-BACK Results ........................................................................................... 92

4.3. AREA72 Results.................................................................................................. 95

4.4. LTPP Forward Calculation Results .................................................................... 96

4.5. BAKFAA Results............................................................................................... 96

4.6. ELMOD Results ................................................................................................. 98

4.7. Comparisons ....................................................................................................... 98

5. DISCUSSION ......................................................................................................... 109

5.2. NUS-BACK ..................................................................................................... 109

5.3. AREA72 ............................................................................................................ 110

5.4. LTPP Forward Calculation............................................................................... 111

5.5. BAKFAA ......................................................................................................... 111

5.6. ELMOD ............................................................................................................ 112

5.7. NUS-BACK Variations .................................................................................... 113

5.8. NUS-BACK vs. AREA72 ................................................................................. 114

5.9. ELMOD Variations .......................................................................................... 115