Webpdf

Nonlinear Dynamics of
Structures Under Extreme

Transient Loads
The effect of combined extreme transient loadings on a structure is not well understood – whether the
source is man-made, such as an explosion and fire, or natural, as an earthquake or extreme wind loading.
A critical assessment of current knowledge is timely (with Fukushima-like disasters or terrorist threats).
The central issue in all these problems is structural integrity, along with their transient nature, their
unexpectedness, and often the uncertainty behind their cause. No single traditional scientific discipline
provides full answer, but a number of tools need to be brought together: nonlinear dynamics, probability
theory, some understanding of the physical nature of the problem, as well as modeling and computational
techniques for representing inelastic behavior mechanisms.
The book covers model building for different engineering structures and provides detailed presentations
of extreme loading conditions. A number of illustrations are given: quantifying a plane crash or explo-
sion induced impact loading, quantifying the effects of strong earthquake motion, quantifying the impact
and long-duration effects of strong stormy winds - along with a relevant framework for using modern
computational tools. The book considers the levels of reserve in existing structures, and ways of reducing
the negative impact of high-risk situations by employing sounder design procedures.
Nonlinear Dynamics of
Structures Under Extreme
Transient Loads
By
Adnan Ibrahimbegovic and Naida Ademovic
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
© 2019 by Taylor & Francis Group, LLC

CRC Press is an imprint of Taylor & Francis Group, an Informa business
No claim to original U.S. Government works
Printed on acid-free paper
International Standard Book Number-13: 978-1-138-03541-6 (Hardback)
This book contains information obtained from authentic and highly regarded sources. Reasonable efforts have been
made to publish reliable data and information, but the author and publisher cannot assume responsibility for the valid-
ity of all materials or the consequences of their use. The authors and publishers have attempted to trace the copyright
holders of all material reproduced in this publication and apologize to copyright holders if permission to publish in this
form has not been obtained. If any copyright material has not been acknowledged please write and let us know so we
may rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or
utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including pho-
tocopying, microfilming, and recording, or in any information storage or retrieval system, without written permission
from the publishers.
For permission to photocopy or use material electronically from this work, please access www.copyright.com (http://
www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA
01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and registration for a variety of
users. For organizations that have been granted a photocopy license by the CCC, a separate system of payment has been
arranged.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for
identification and explanation without intent to infringe.
Library of Congress Cataloging‑in‑Publication Data
Names: Ibrahimbegovic, Adnan, author. | Ademovic, Naida, author.

Title: Nonlinear dynamics of structures under extreme transient loads / Adnan
Ibrahimbegovic and Naida Ademovic.
Description: First edition. | Boca Raton, FL : CRC Press/Taylor & Francis
Group, [2019] | Includes bibliographical references and index. |
Identifiers: LCCN 2019001405 (print) | LCCN 2019002310 (ebook) | ISBN
9781351052498 (Adobe PDF) | ISBN 9781351052474 (Mobipocket) | ISBN
9781351052481 (ePub) | ISBN 9781138035416 (hardback : acid-free paper) |
ISBN 9781351052504 (ebook)
Subjects: LCSH: Live loads. | Load factor design. | Structural dynamics. |
Nonlinear mechanics.
Classification: LCC TA654.3 (ebook) | LCC TA654.3 .I27 2019 (print) | DDC
624.1/72--dc23
LC record available at https://lccn.loc.gov/2019001405
Visit the Taylor & Francis Web site at

http://www.taylorandfrancis.com
and the CRC Press Web site at

http://www.crcpress.com
Contents
Authors ix
1 Initial boundary value problem 1

1.1 Newton’s Second Law—Strong Form of Elastodynamics 1
1.2 D’Alembert Principle—Weak Form of Equations of Motion 3
1.3 Hamilton’s Principle—Energy Conservation 5
1.4 Modal Superposition Method 6
1.4.1 Rayleigh Damping 8
1.4.2 Nonproportional Damping and Fast Computation Methods 9
1.5 Time-Integration Schemes 11
1.5.1 Central Problem of Dynamics (Time Evolution) 12
1.5.2 Central Difference (Explicit) Scheme 12
1.5.3 Trapezoidal Rule or Average Acceleration (Implicit) Scheme 15
1.6 Motivation for Developments Presented in Part I and Part II 19
1.6.1 Damping Model Characterization 19
1.6.2 Extreme Loading Representation 20
References 20
Part I
2 Steel structures 25
2.1 Essential Ingredients of Plasticity Model 25
2.1.1 1D Plasticity with Hardening and Softening 25
2.2 Localized Failure Plasticity Model 31
2.3 Structural Plasticity Model 34
2.3.1 Simple Form 34
2.3.1.1 Numerical Examples of Small Experiments 36
2.3.2 Multi-Scale Model Parameters Identification 37
2.3.2.1 Beam Element With Embedded Discontinuity 40
2.4 Computation of Beam Plasticity Material Parameters 48
2.4.1 Numerical Examples of Frames and Frame Elements 51
2.4.1.1 Example 1 — Computation of Beam
Plasticity Material Parameters 51
2.4.1.2 Example 2 — Push-Over of an Asymmetric Frame 55
v
vi Contents
2.4.1.3 Example 3 — The Darvall–Mendis Frame 57

2.4.1.4 Example 4 — Push-Over Analysis of a Symmetric Frame 58
2.4.1.5 Example 5 — Cyclic Loading of a Steel Structure 60
2.5 Conclusion 62
References 62
3 Reinforced concrete structures 65

3.1 Concrete Modeling with a Damage Model 66
3.1.1 A Numerical Example of Small Experiments 71
3.2 Bond-Slip Model 71
3.3 Steel Model with Classical Elastoplasticity 74
3.4 Reinforced Concrete Model 75
3.4.1 Kinematics of Bond-Slip Along Reinforcement
Bar and its X-FEM Representation 76
3.4.2 Solution Procedure for Relative
Displacements of Bond-Slip Element 78
3.4.3 Numerical Examples 81
3.4.3.1 Traction Test and Proposed RC Model Validation 81
3.4.3.2 Four-Point Bending Test and Crack-
Spacing Computations 84
3.5 Reinforced Concrete Model for Cyclic Loading 88
3.5.1 Constitutive Concrete Model 89
3.5.1.1 Thermodynamics with Internal Variables (TIV) 89
3.5.1.2 Governing Equations of the Constitutive Model 91
3.5.1.3 Euler–Lagrange Equations of a Concrete
1D Structure in Dynamic Loading 95
3.5.2 Numerical Examples 98
3.5.2.1 Identification of a Concrete Law 98
3.5.3 Numerical Examples Seismic Application 102
References 108
4 Masonry structures 111

4.1 Computational Modeling of Masonry Structures 112
4.2 Micro-Modeling of Masonry 113
4.3 Strain Localization Phenomena 114
4.3.1 Modeling of In-Plane and Out-Of-Plane
Loading on a Brick Wall 115
4.3.1.1 Modeling of Failure Mechanisms in Bricks 115
4.3.1.2 Modeling of Mortar Joints in a Brick Wall 117
4.3.2 Examples 122
4.3.2.1 Example 1 122
4.3.2.2 Example 2 124
4.4 Further Research 125
References 126
Contents vii
Part II
5 Dynamics extreme loads in earthquake engineering 131

5.1 The Free-Field vs. Added Motion Approach in
Structure-Foundation Interaction 132
5.2 Localized Nonlinearities in Structure-Foundation
Interface 136
5.3 Reduction of Model: Numerical Techniques 138
5.4 Case Studies of Structure-Foundation Interaction Problems 140
Bibliography 143
6 The dynamics of extreme impact loads in an airplane crash 145

6.1 Central Difference Scheme Computation of Impact Problems 147
6.2 Local Model Extensions 151
6.3 Field Transfer Strategy for Complex Structure Impact Loading 152
6.4 Field Transfer in Space and Projection to Coarse Mesh 154
6.5 Field Transfer in Time and Minimization Problem
for Optimal Transfer Procedure 157
6.6 Numerical Examples 159
6.6.1 Example 1 — Element Level 159
6.6.2 Example 2 — Structure Level 161
References 163
7 Fire-induced extreme loads 165

7.1 Transient Heat Transfer Computations 166
7.2 Damage Mechanisms Representation Under Increased Temperature 167
7.3 Coupled Thermomechanical Problem Computations 170
7.4 Operator Split Solution Procedure with Variable Time Steps 171
7.5.1 Example 1 — Circular Ring Heating 172
7.5.2 Example 2 — Cellular Structures 174
7.5.3 Example 3 — Hollow Brick Wall 175
7.6 Detailed Theoretical Formulation of Localized
Thermomechanical Coupling Problem 179
7.6.1 Continuum Thermo-Plastic Model and its Balance Equation 180
7.6.2 Thermodynamics Model for Localized Failure
and Modified Balance Equation 183
7.6.2.1 Thermodynamics Model 183
7.6.2.2 Mechanical Balance Equation 186
7.6.2.3 Local Balance of Energy at the Localized Failure Point 186
7.7 Embedded Discontinuity Finite Element Method (ED-FEM)
Implementation 187
7.7.1 Domain Definition 187
7.7.2 “Adiabatic” Operator Split Solution Procedure 187
viii Contents
7.7.3 ED-FEM Implementation for the Mechanical Part 188

7.7.4 ED-FEM Implementation for the Thermal Part 192
7.8.1 Example 1 — Simple Tension Imposed
Temperature Example with a Fixed Mesh 195
7.8.1.1 Material Properties Independent on Temperature 195
7.8.1.2 Material Properties are Linearly
Dependent on Temperature 197
7.8.1.3 Material Properties Non-Linearly Dependent
on Temperature (Eurocode 1993-1-2) 200
7.8.2 Model Generalization to a 2D/3D Case: Formulation,
Implementation and Numerical Results 206
7.8.2.1 Theoretical Formulation 206
References 212
8 Fluid-induced extreme loads 215

8.1 Structure and Fluid Formulations 215
8.1.1 Structure Equation of Motion 215
8.1.2 Free-Surface Flow 218
8.2 Fluid–Structure Interaction Problem Computations 221
8.2.1 Theoretical Formulation Of Fluid–Structure Interaction Problem 221
8.3.1 Example 1a — Flexible Appendix in a Flow 224
8.3.2 Example 1b — Flexible Appendix in a Flow—3d 228
8.3.3 Example 2 — Three-Dimensional Sloshing
Wave Impacting a Flexible Structure 229
References 234
Index 239
Authors
Adnan Ibrahimbegovic obtained his PhD in Structural Engineering from the University of
California at Berkeley and Habilitation in Mechanics from University Pierre Marie Curie
in Paris. He is currently a Full Professor in Mechanical Engineering at the Université de
Technologie Compiègne in France. He holds the Chair for Computational Mechanics for
interdisciplinary research within the Sorbonne Universités group, and positions of Senior
Member of the Institut Universitaire de France and Head of Joint Mechanics-Mathematics
Research Platform at UTC. He was formerly Head of teaching and research in Civil
Engineering at the Ecole Normale Supérieure (ENS-Cachan), one of elite engineering
schools of French system of higher education. He is a Fellow of the International Association
for Computational Mechanics, recipient of Humboldt Research Award in Technischen
Mechanik from Germany, Claude Levy Strauss Award from Brazil, Asgard Award from
Norway, Research Agency Award from Slovenia, and was invited professor and short course
instructor at KAIST (South Korea), IPN (Mexico), USP (Brazil), Universities of Innsbruck
(Austria), TU Tampere (Finland), Luxembourg, TU Braunschweig (Germany), Pavia (Italy),
Ljubljana (Slovenia). He has written over 500 publications, including 8 books, 29 book
chapters, 24 special issues/proceedings and close to 200 scientific papers in refereed journals.
Naida Ademović is an Associate Professor in Civil (Structural) Engineering at the Faculty

of Civil Engineering at the University of Sarajevo in Bosnia and Herzegovina, which she
joined in 2001. She is also an Associate Professor at the Center of Interdisciplinary Studies,
University of Sarajevo, at the new Masters program “Natural Disasters Risk Management
in Western Balkan Countries.”
Naida received her 5-year Diploma (Bachelor of Science) in Structural Engineering from
the University of Sarajevo. She received a Master of Science degree in “Computational
Engineering” from Rühr University in Bochum, Germany, in 2004, and a second
Masters degree (Advanced Masters in Structural Analysis of Monuments and Historical
Constructions - SAHC) from the University of Minho, Portugal and University of Padova,
Italy, in 2011. The SAHC program she was involved in won an EU Prize for Cultural
Heritage/Europa Nostra Award in 2017. Naida received her PhD in 2012 in the field of
earthquake engineering and masonry structures at the University of Sarajevo.
She has co-authored two books and published over 70 peer-reviewed scientific papers in
refereed journals and international conferences.
ix
Chapter 1
Initial boundary value problem
1.1 NEWTON’S SECOND LAW—STRONG

FORM OF ELASTODYNAMICS
The first general equation of motion was developed by Newton and it is known as Newton’s
Second Law of Motion. This law enforces for the motion momentum either conservation, in
the absence of external force, or change, due to an applied external force. Newton’s Second
Law applies in the same manner to point-like particles and to any point in a rigid body,
apart from their constrained motion. It is also applicable to each point in a mass continua,
like deformable solids or fluids, for either small or large motion, but the latter requires the
motion to be accounted for by material derivative. The essential restriction concerns the
mass conservation hypothesis; hence, if the mass is not constant, it is necessary to make
some modifications to Newton’s Second Law, which are consistent with the conservation of
momentum.
The model problem of one-dimensional (1D) elasticity in large displacements is selected
with the aim to demonstrate the finite element method applied to nonlinear dynamics. It is
important to note that the main solution steps for the chosen model problem will remain the
same for a number of other problems in nonlinear dynamics.
The reference frame in which the initial configuration of the 1D solid corresponds to a
closed interval Ω = [0,l ] is selected. Each particle of this configuration is identified by its
position vector:
x ∈ Ω ≡ [0, l ] (1.1)
The dynamic loads are applied in terms of time-dependent distributed loading b(x,t), pro-
ducing the motion of the body that can be described by the trajectory of each particle from
its initial position x to the current position at time t, denoted as φ(x,t). Alternatively, the
motion can also be described by the displacement field specifying the displacement value
along x for each particle:
x ϕ = x + d(x, t) =: ϕ(x, t) ≡ ϕt (x) (1.2)
The deformed configuration is defined as the assembly of all particles in their current posi-
tion (see Figure 1.1):
x ϕ ∈ Ωtϕ = 0, ltϕ  (1.3)
Velocity and acceleration of motion can be written in either a spatial or material descrip-
tion, by using xφ or x as the independent variable, respectively. The point transformation
1
2 Nonlinear Dynamics of Structures Under Extreme Transient Loads
Figure 1.1 Mapping from the initial to current configuration.
in (1.2) allows one to easily switch from one description to other. This also being applicable
for inertia effects which are important for dynamic problems.
The velocity in the material description can simply be computed as the partial derivative
of motion in (1.2) with respect to time while keeping the particle material position fixed;
this results with:
∂ϕ(x, t)
υ(x, t) = (1.4)
∂t x
It is important to understand that the velocity vector in (1.4) acts in the current configura-
tion, but that it is parameterized in the material description with respect to the particle posi-
tion in the initial configuration through the point transformation x ϕ = ϕt (x). By assuming
that the motion is sufficiently regular to define the inverse of this transformation, ϕt−1 , the
velocity vector can also be expressed by the spatial description:
( )
υˆ ϕt−1(x ϕ ), t =: υϕ (x ϕ , t) ≡
(
∂ϕ ϕt−1(x ϕ ), t ) (1.5)
∂t ϕ
x
Further, in the same manner, one can continue on with the relationship between spatial and
material descriptions of the particle acceleration. Namely, in a material description, the
latter can be expressed simply as the second partial derivative of the motion with respect
to time:
∂υ(x, t) ∂ 2φ(x, t)
a(x, t) = ≡ (1.6)
∂t x ∂t 2 x
On the other hand, the spatial description of the acceleration vector will result in a so-called
convective term:
(
aϕ xϕ , t = ) (
∂υ ϕ x ϕ , t ) + ∂υ ( x , t ) υ ( x , t ) (1.7)
ϕ ϕ
ϕ ϕ
ϕ
∂t ∂x
The mass conservation principle postulates that in any deformed configuration of a deform-
able solid body, Ωφ, the total mass of the body will remain constant. Therefore, the mass
density ρφ(xφ,t) will change in each particular deformed configuration Ωφ, but the total mass
will not:
( ) ∫ ρ ( x , t ) dx (1.8)
m Ωϕ = ϕ ϕ ϕ
Ωϕ
Initial boundary value problem 3
Denoting the mass density as ρ(x), in the initial configuration, the principle of mass conser-
vation allows one to write the following:

∫ ρ ( x , t ) dx = ∫ ρ(x)dx (1.9)
Ωϕ
ϕ ϕ ϕ
By considering that the mass conservation principle also applies to an arbitrary sub-domain,
the local form of this mass conservation principle can be acquired as:
 ϕ  ρ(x) dx ϕ
ρϕ  x  , t = ; λ = (1.10)
 ρt (x)  λ ( x, t ) dx
The last result derived for the 1D case can easily be generalized to the three-dimensional
(3D) case taking into account a proper change of infinitesimal volume through the determi-
nant of the deformation gradient:
ρ
ρϕdV ϕ = ρdV ⇒ ρϕ = ; J = det [F] (1.11)
J
Taking into account these results, it is possible to deliver the spatial and material descrip-
tions of the local form of equations of motion. The former, which applies to the current
position of the particle “xφ” in a given deformed configuration, reads:
( ) (
ρϕ x ϕ , t aϕ x ϕ , t =) (
∂σ ϕ x ϕ , t ) + b ( x ) ; in dx
ϕ ϕ ϕ
(1.12)
ϕ
∂x
The driving force of motion in the spatial description is defined by the expression equivalent
to the static equilibrium equations written in term of the Cauchy or true stress.
As is well known, static equilibrium equations can also be written in the material descrip-
tion by using the first Piola–Kirchhoff stress P(x,t) and the distributed loading b(x,t) per
unit volume of the initial configuration. This kind of transformation can be applied to the
problem in dynamics by adding the inertia term in the material description and the principle
of mass conservation defined in (1.6) and (1.10), respectively. The final result is the material
description of the equations of motion, which can be written:
∂P(x, t)
ρ(x)a(x, t) = + b(x, t); in dx (1.13)
∂x
It is important to note that the material description results in the linear inertia term, which
follows from using the fixed reference frame where a(x, t) = (∂ 2φ(x, t)) / (∂t 2 ) . The linear form
of the inertia terms in the material description of the equations of motion also applies to the
3D case, which is written by using the direct tensor notation:
ρ(x)a(x, t) = div P(x, t) + b(x, t); ∀x ∈ Ω & ∀t ∈ [0, T ] (1.14)
1.2 D’ALEMBERT PRINCIPLE—WEAK FORM

OF EQUATIONS OF MOTION
The D’Alembert principle is the dynamic analog to the principle of virtual work for applied
forces in a static system. It is more general than Hamilton’s principle, avoiding the restric-
tion to holonomic systems (Morris 1997). In this respect, the D’Alembert principle is cho-
sen as the starting point in the development of the weak form of the equations of motion.
The snap-shot of motion taken at time “t” can be described formally with the equilib-
rium equations. However, unlike the statics problem, these equilibrium equations should
also include an extra external load in terms of the inertia force; the latter, denoted as
( ) ( )
finertia = −ρ x ϕ , t a x ϕ , t , is proportional to the mass and directed opposite to acceleration.
Thus, it can be written:
( ) (
0 = −ρϕ x ϕ , t aϕ x ϕ , t + ) (
∂σ ϕ x ϕ , t ) + b ( x , t ) (1.15)
ϕ ϕ
ϕ
∂x
This point of view of D’Alembert provides an important conceptual advantage in allowing
one to reduce a new problem of describing the motion in dynamics to a familiar problem
of equilibrium in statics. As stated above, the later can then be treated by the principle of
virtual work, stating that the work of internal forces on chosen virtual displacement should
be equal to the virtual work of external forces, including the inertia force among them. In
this approach, the time is kept fixed at the chosen value t corresponding to a particular
deformed configuration; hence, the virtual displacement field is independent of time and can
be denoted: w ϕ = wˆ ϕ ( ϕ(x, t )), see Figure 1.2.
Equations of motion can be multiplied by the chosen virtual displacement (or weighting func-
tion) and integrated over the whole domain. We then use the integration by parts to remove deriv-
atives on the stress field and to provide the final weak form of the 1D problem in elastodynamics:
(
0 = Gdyn ϕt ; w ϕ := ) ∫

( ) (
w ϕ ρϕ x ϕ , t aϕ x ϕ , t −
∂σ ϕ x ϕ , t
) (
− bϕ x ϕ , t
) (

) dx ϕ
Ωϕ  ∂x ϕ

(1.16)
 ϕ ϕ ϕ ϕ

=
∫ (ϕ ϕ
) (
w ρ x , t a x , t +
dw ϕ ϕ
dx ϕ ) ( ) (
σ x , t − w ϕbϕ x ϕ , t dx ϕ − w ϕ (lt )t ϕ (t) )

Ωϕ 
Taking into account the result obtained in (1.10), together with the material parameteriza-
tion of the virtual displacement field w(x) = w ϕ (ϕ(x, t )), the weak form of the equations of
motion can be recast in the material description:
0 = Gdyn ( ϕt (x); w(x)) :=

∫ [w(x)ρ(x)a(x, t)dx
Ω
(1.17)
dw(x) 
+ P(x, t) − w(x)b(x, t) dx − w(l)t (t)
dx 

Gstat
where Gstat(·) is the result of the 1D elastostatics. The same kind of result for the 3D case can
be achieved, and by using the direct tensor notation it reads:
0 = Gdyn ( ϕt (x); w(x)) :=

∫ w(x) ⋅ ρ(x)a(x, t)dV + G
Ω
stat (ϕt (x); w(x)) (1.18)
Figure 1.2 Application of the D’Alembert principle.

1.3 HAMILTON’S PRINCIPLE—ENERGY CONSERVATION
The weak form of equations of motion can be expressed in terms of variational formulation,
which applies to hyperelastic materials. Here, the strain energy density ψ(φ) and external
loads together define the potential energy functional Π(φt(x)), which can be written:
Π(ϕ) =
∫ ψ(ϕ)dx − Π
Ω
ext (ϕ) (1.19)
The first variation of such a functional is equivalent to the weak form of the equilibrium
equations in statics, and, moreover, a positive value of the second variation indicates that
the equilibrium state remains stable:
d
0 = Gstat ( ϕt (x); w(x)) :=  Π ( ϕt , s (x))
ds  s = 0 (1.20)
d2
0<  Π ( ϕt , s (x))
ds2  s =0
It is possible to generalize this result to dynamics, by appealing to the Hamilton variational prin-
ciple employing the total energy functional, which consists of both potential and kinetic energy:
H ( ϕ(x, t), υ(x, t)) := T ( υ(x, t)) + Π ( ϕ(x, t)) (1.21)
In the material description, the kinetic energy can be written as a quadratic form in velocities:
1
T ( υ(x, t)) =
∫ 2ρ(x)[υ(x, t)] dx (1.22)
2

Ω
One can appeal to the principle of least action, to postulate that in the dynamic motion of
a hyperelastic body over a time interval [t1, t2], the zero value of the first variation of total
energy (or yet referred to Hamiltonian) will produce the equations of motion, whereas a
positive value of the second variation will confirm the motion stability:
 t2 
d 
0 = Gdyn :=
ds 
 t1
∫
H ( ϕt , s (x)) dt  ;

 s = 0 (1.23)

 t2 
d2 
0<
ds2 
 t1
H
∫( ϕ t,s (x)) d t 

 s =0
It is easy to confirm the last result by using the directional derivative computation. In fact,
the only new term with respect to statics concerns the directional derivative of the kinetic
energy in (1.22), which can be written:
 t2  ∂ ϕ (x)  
( t,s )  dt = ∂T ( υ ) ∂ 2 ϕ t , s
t2
d 
ds ∫
 t1
T
 ∂t 

 s =0
∫
t1
∂υ ∂s∂t
dt =
t2 t2
 ∂T (υ) ∂ϕt , s  ∂  ∂T (υ)  (1.24)
=
 ∂υ ∂s
 −
s = 0  t1
∫t1
∂t  ∂υ 
wdt
(
 ∂T υ ( t 2 ) ) ∂T υ (t1 ) (  t2 )
=
 ∂υ
w (t 2 ) −

0
∂υ
w (t1 ) −

0  t1 Ω
∫∫
wρυ ( x, t )dxdt
The last result is obtained by imposing a supplementary admissibility condition on zero value
of the variations at both limits of the time interval, with w(t1) = 0 and w(t2) = 0, which allows for
the recovering of the inertia term of equations of motion from the first variation of the kinetic
energy. The last result along with the variation of the potential energy producing will confirm
that the first variation of the Hamiltonian corresponds to the equations of motion (1.17):
 t2 
d 
0=
ds ∫
 t1
H ( ϕt , s (x)) dt 

 s =0
≡ Gdyn (1.25)
In the absence of external load with Π ext ( ϕ(x, t)) = 0, the Hamilton principle implies that the
total energy of the motion remains conserved:
 t2 
d 
0=
ds ∫
 t1
H ( ϕt , s (x)) dt 

 s =0
⇒ H(t) = cst. (1.26)
This is an important observation that has to be taken into account in the development of
time-integration schemes.
1.4 MODAL SUPERPOSITION METHOD
Generally, superposition is applicable when there is a linear relationship between external

forces and corresponding structural displacements. For the dynamic analysis of linearly
elastic systems, modal superposition is a powerful idea of obtaining solutions.
The modal superposition procedure is also applicable to dynamic linearly elastic systems
having continuously distributed properties, which have an infinite number of degrees of
freedom requiring that their equations of motion are written in the form of partial differ-
ential equations. This is all in the sense that for obtaining practical solutions only a limited
number of the lower modes of vibration can be considered.
It is applicable to both free-vibration and forced-vibration problems. This method has
often been used in the analysis of linear response, and it has seldom been applied in nonlin-
ear problems (Morris 1997).
The modal superposition procedure can be used to obtain an independent SDF equation
for each mode of vibration of the undamped structure. Thus, the use of the normal coordi-
nates serves to transform the equations of motion from a set of N simultaneous differential
equations, which are coupled by the off-diagonal terms in the mass and stiffness matrices, to
a set of N independent equations in normal coordinates. The dynamic response, therefore,
can be obtained by solving separately for the response of each normal (modal) coordinate
and then superposing these to obtain the response in the original geometric coordinates
(Clough and Penzien 2003).
The equations of motion for a free undamped system of vibration can be written in the form:
+ Kv = 0 (1.27)
Mv
In order to determine in which case the conditions set in (1.27) will be satisfied, we have to
solve the problem of free-vibrations. It can be assumed that the free-vibration motion is a
simple harmonic function, expressed as:
v(t) = φ sin ( ωt + θ ) (1.28)
where ϕ is the shape of the system (amplitude) and θ is a phase angle.

Taking the second derivative of equation (1.28) and inserting it and the equation (1.28)
into (1.27) one obtains:
−ω 2 Mφ sin ( ωt + θ ) + Kφ sin ( ωt + θ ) = 0 (1.29)
As the sine term is arbitrary it can be omitted, so the equation now reads:
 K − ω 2 M  φ = 0 (1.30)
which is a well-known characteristic equation of the system, where ϕi and ωi are natural
modes of free-vibrations and natural frequencies.
The free-vibration mode shapes ϕi have certain special properties which are very useful
in structural dynamics analyses. The orthogonality relationships state that the vibrating
shapes are orthogonal with respect to the stiffness matrix as well as with respect to the mass
matrix.
The equation of motion for the system (1.29) can be written as:
Kφ = ω 2 Mφ
(1.31)
fs = −fI
where:
fs is the elastic-resisting force factor and
−f I is the applied-inertia-load vector.
Using Betti’s law on the two vibration modes, one obtains, as indicated in Figure 1.3:
T
−fIm φn = −fInT φm (1.32)
Combining equations (1.31) and (1.32), one obtains:
ω 2mφTm Mφn = ω n2φTn Mφm (1.33)
As the mass matrix is symmetric and a scalar is obtained as the result of the matrix opera-
tions, it is evident that equation (1.33) can be presented in the form:
(ω 2
m )
− ω 2n φTm Mφn = 0 (1.34)
Figure 1.3 Vibration mode shapes and resulting inertial forces (Clough and Penzien 2003).
As the two frequencies are not the same, one obtains the first orthogonality conditions,
which reads:
φTm Mφn = 0 ω m ≠ ω n (1.35)
The second orthogonality condition states that
φTm Kφn = 0 ω m ≠ ω n (1.36)
For a free-vibration problem in linear dynamics, with d0 and v0 as the initial displacement
and velocity, the solution is obtained with mode superposition
∑ φ  φ Md cos ω t + ( 1ω ) φ Mv sin ω t (1.37)

n
T T
dfree (t) = i i 0 i i 0 i
i
i =1
The mode superposition method can also be employed to construct the solution to a forced-
vibration problem in linear dynamics, where the forcing term is supplied by Duhamel’s
integral with:
m t 
dforced (t) = dfree (t) + ∑ ∫(
i =1

0
)
φi  φTi Mf (s) / ω i sin ω i (t − s)ds (1.38)

1.4.1 Rayleigh Damping
In order to apply the modal analysis of undamped systems to damped systems, it is common
to assume the proportional damping, a special type of viscous damping. The proportional
damping model can be represented as a linear combination of the mass and stiffness matri-
ces, leading to well known Rayleigh damping, or classical damping model.
C = αM + βK (1.39)
where α and β are real scalars, known as proportionality constants and they have units 1/
sec and sec units respectively. Modes of classically damped systems preserve the simplicity
of the real normal modes as in the undamped case. The properties of mode orthogonality
as stated in (i) and (j) are used here. These are called mass proportional and stiffness pro-
portional damping, and the damping behavior associated with them may be recognized by
evaluating the generalized modal damping value for each.
When introducing the damping matrix into the equation of motion, equation (1.27) now
reads:
+ Cv + Kv = 0 (1.40)
Mv
where:
C is the damping matrix and
v is the first derivative of the displacement, meaning velocity.
Using the orthogonality characteristics of the modes generalized modal damping value
can be written for each as:
Cn = φTn Cφn = αφTn Mφn (1.41a)

or:
Cn = φTn Cφn = βφTn Kφn (1.41b)
Combining (1.41a) and (1.41b) with (1.42):
2ω n M nξ n = αM n or K n = ω n2 M n (1.42)
one obtains:
α βω n
ξn = or ξ n = (1.43)
2ω n 2
From (1.43), it is obvious that for mass proportional damping, damping is inversely propor-
tional to the frequency on one hand, and on the other for stiffness proportional damping
it is directly in proportion with the frequency. Combining these two gives the above men-
tioned Rayleigh’s damping (see Figure 1.4). The two damping factors α and β can be evalu-
ated using the solution of a pair of simultaneous equations if the damping ratios ξm and ξn
connected with the two specific modes of frequencies ωm and ωn are known.
As damping is a rather uncertain parameter and its variation with frequency is usually
unknown, in most practical applications it is feasible to assume the same damping ratio for
the tow control frequencies. The proportional factors α and β then are given as:
α  2ξ ω m ω n 
 =   (1.44)
β  ωm + ωn  1 
1.4.2 Nonproportional Damping and Fast Computation Methods

A nonproportional damping matrix is formed as an assembly of proportional matrices for
each part of the structure, with characteristic materials and specified damping ratios for
each segment of the structure.
In this situation, the assembly of the stiffness, mass and damping property matrices will
be formulated as
Steel Steel Steel
Ks Ms Cs
X X X
K= M= C=
Concrete Concrete Concrete
Kc Mc Cc
where X DoF are common to both parts of the structure. The same procedure as explained
above is used for solving this problem.
The vast majority of engineering structures in practical applications are characterized
by nonproportional damping. Many practical examples with nonproportional damping
matrices can be given, starting from the frictional support systems, over mechanical struc-
tures with energy-absorbing or vibration-isolation devices and coupled systems, to soil-
structure and fluid–structure interaction problems. Dynamic response of the systems with
nonproportional damping can be obtained by direct integration or by the calculation of
the complex eigenvectors of the complete system, which are then used to transform the
system into an uncoupled set of complex modal equations (Veletsos and Ventura 1986;
Chen 1987).
Figure 1.4 Relationship between damping ratio and frequency (for Rayleigh damping) (Clough and Penzien 2003).
For a nonproportionally damped system, the equations of motion in the modal coor-
dinates are coupled through the off-diagonal terms of the modal damping matrix and
consequently the system possesses complex modes instead of real normal modes. This has
been investigated by many authors (Ibrahimbegović et al. 1991; Phani 2004; Adhikari
and Woodhouse 2001; Udwadia 2009). Researchers have proposed different methods
to overcome this issue and to take into account nonlinearities (for further details see
Bernal 1994; Hall 2006; Charney 2005; Medina and Krawinkler 2004; Zareian 2006).
However, the proposed solutions by all the researchers above have real limitations; for
example, Bernal’s (Bernal 1994) approach does not preclude spurious damping forces
appearing when masses are assigned to rotational degrees of freedom. Charney’s (Charney
2005) proposal is questionable once the overall tangent stiffness of the system becomes
negative due to second-order, P-Δ effects and/or significant material strength and stiffness
deterioration.
A general method for handling nonproportional damping was proposed by Ibrahimbegović
and Wilson (1989), which is characterized by high accuracy and low computational cost.
The main idea of this solution method is to make an approximation in loading time varia-
tion—making it linear in each increment (this fits well with the earthquake loading, where
one measurement is available at each increment; hence, all the measurement can be con-
nected linearly). After, there is no approximation, but the exact computation of the modal
response. Namely, if time variation of modal load fi within a small time interval ∆t = t1 − t0
can be approximated by a linear function (fi = a + b ⋅ t ), then the total solution can be writ-
ten as a series of consecutive piecewise exact solutions of the form:
yi (ti ) = A0 + A1∆t + A2 exp(−ξi ω i ∆t)cos ω Di ∆t + A3 exp(−ξi ω i ∆t)sin ω Di ∆t (1.45)
where the coefficients in equation (1.45) above are:

a 2ξ b
A0 = − i3 (1.46)
ω i2 ωi
b
A1 = (1.47)
ω i2
A2 = yi (t0 ) − A0 (1.48)
1
A3 =  y i (t0 ) + ξi ω i A2 − A1  (1.49)
ω Di 
The pseudo-force concept can be applied in a very similar way to account for modal cou-
pling effects in the analysis of systems with nonproportional damping. In this case, the cru-
cial step is to separate the damping matrix into a proportional component that is uncoupled
by the modal coordinate transformation, plus a nonproportional component that is trans-
ferred to the right-hand side of the equations of motion, where its effects are represented as
pseudo-forces. In this context, we will assume that the adequate nonproportional viscous
damping matrix is denoted as C and that it has been formulated to represent the system’s
actual damping mechanism. Now modal coordinate transformation will be applied, and
the damping matric will be transformed into diagonal elements which are representing the
proportional damping contribution and the nonproportional damping effects are presented
by the off-diagonal elements. Thus, we introduce the additive split of the modal damping
matrix:
Cm = Cd + Cˆ = diag(2ξi ω i ) + Cˆ (1.50)
The damping coupling between the modes is expressed by these off-diagonal coefficients
and are considered as pseudo-forces applied on the right side of the equation of motion.
i + 2ξi ω i y i + ω i2 yi = fi − Cˆ ij y j , i = 1, 2,...m. (1.51)

y
The contribution of the off-diagonal modal damping coefficients to these pseudo-forces can
be defined in an iterative manner:
mm
i(k) + 2ξi ω i y i(k) + ω i2 yi(k) = fi −

y ∑ Cˆ y
j =1
ij
(k −1)
j , i = 1,2,… m. (1.52)
This method is characterized by high accuracy and low computational cost thanks to its
unconditionally stable iterative procedure with strong convergence properties. The impor-
tant aspect is that the procedure uses an exact solution for piecewise linear loading excita-
tion; hence, its accuracy depends on the proper representation of real loading variation in
time. In addition, the algorithm can easily be extended to systems subjected to periodic
loading (Ibrahimbegović and Wilson 1989).
1.5 TIME-INTEGRATION SCHEMES
In this case, we make an approximation not only in loading time variation, but also response
function approximation (e.g. sin or cos functions replaced by polynomials), and we arrive at
the time-integration schemes.
The final product of semi-discretization procedure carried out by the finite element
method in nonlinear dynamics is a system of nonlinear ordinary differential equations in
time. Any such nonlinear dynamics problem is quite different from the nonlinear equilib-
rium problem in statics, which is governed by a set of nonlinear algebraic equations param-
eterized by pseudo-time. Despite this difference, in constructing the solution to a nonlinear
dynamics problem by using the time-integration schemes, all the main ingredients of the
incremental/iterative solution procedure for quasi-static problems are present. The tradi-
tional methods for solving linear dynamics problems, such as modal superposition method
(e.g. Clough and Penzien 2003; Géradin and Rixen 1992) are of small interest for nonlinear
dynamics problems since the free-vibration modes would change with each incremental/
iterative modification of the tangent stiffness matrix. A few words will be given in the fol-
lowing subsection regarding modal superposition.
In order to solve the nonlinear dynamics problems by using the time-integration schemes,
a few steps have to be followed. In this respect, the time interval of interest [0, T] is subdi-
vided into the chosen number of time steps, which will specify the time instants where the
selected time-integration scheme should deliver the solution:
0 < t1 < t2 < ... < tn < tn +1 < ... < T (1.53)
This corresponds to the incremental analysis for quasi-statics problems; however, contrary
to statics, the complete solution requires not only the nodal values of displacements but also
of velocities and accelerations. For example, the solution at time tn is defined with:
d n = d (tn ) ; v n = d (tn ) ; a n = d
(t ) (1.54)
n
1.5.1 Central Problem of Dynamics (Time Evolution)

In order to simplify the computer code architecture, in general, one-step time-integration
schemes are used that construct the solution over a single step at the time. For a typical
time step starting at time tn, the time-integration scheme should deliver the nodal values of
displacements, velocities and acceleration at time tn+1, which verify the equations of motion;
this can formally be stated in terms of:
Central problem in nonlinear dynamics for [tn, tn + 1]
Given: d n , v n , a n
Find: d n +1, v n +1, a n +1
such that: Ma n +1 + fˆ int (d n +1 ) = fnext
+1
One possibility for solving the central problem in dynamics is by exploiting the one-step
time-integration schemes. In fact, it is always possible to recast a set of second-order dif-
ferential equations that characterize the nonlinear dynamics in terms of the set of the first-
order equations containing twice as many equations. In this way, we switch to the so-called
state space form of the system, where both displacements and velocities are considered as
independent variables:
d   v 
z(t) = h ( z(t)) ; z =   ; h(z)=  −1 ext ˆ int
( )
 (1.55)
v  M f − f (d) 
 
Thus one can directly apply the time-integration schemes for integrating the first-order sys-
tems (heat transfer equation or internal variable evolution equations). However, this would
not lead to the most efficient implementation. The computational efficiency can be improved
significantly with the time-integration schemes applicable directly to the original form of
the second-order differential equations. Several time-integration schemes of this kind, both
explicit and implicit, are presented subsequently.
1.5.2 Central Difference (Explicit) Scheme

The central difference scheme is based upon the second-order approximation of the dif-
ferential equations of motion, with a truncated series development employing only the first
derivative of displacements and velocities in (1.47). This kind of approximation can be
constructed in a systematic manner by using the central difference operator (e.g. Dahlquist
and Björck 1974, p. 353); the corresponding approximation of the first differential equation
in (1.29) is obtained, which can be written:
d − dn −1
d (t) = v(t) ⇒ n +1 = vn (1.56)
2h
The last result confirms the explicit form of the central difference approximation since in
computing the displacement dn + 1 only the displacement and the velocity in two previous
steps is required. However, for simplicity of computer code architecture, it is preferred that
the numerical implementation of the central difference scheme, where the only values needed
for computing the displacement at time tn + 1, are those from the previous time tn. In order to
ensure such a format of the central difference scheme, the corresponding approximations of
the velocity at the mid-point of each time step is used, which allow one to write:
(dn +1 − dn −1 ) = vn +1/2  1

1
h 1

1
⇒
2 h
(dn +1 − dn −1 ) = 2 (vn +1/2 + vn −1/2 ) (1.57)
(dn − dn −1 ) = vn −1/2 
h 
Combining equation (1.56) and (1.57), the corresponding approximation of velocity at time
tn can be written as:
1
vn =
2
(vn +1/2 + vn −1/2 ) (1.58)
The same kind of central difference approximation based upon the mid-point values can
also be constructed for the acceleration vector:
vn + 1 / 2 − vn −1 / 2
a(t) = v (t) ⇒ an = (1.59)
h
The corresponding approximation for the velocity vector in the middle of the interval of
interest is obtained by adding up equations (1.58) and (1.59), and it can be written:
h
vn + 1 / 2 = vn + a n (t) (1.60)
2
By inserting (1.60) into (1.57), the final form of the central difference approximation for the
displacement vector at time tn + 1 is obtained, and it reads:
(1.61)
dn+ 1 = dn + hvn + h2 an
It is evident that this kind of approximation for displacement is fully explicit in that it only
employs the known values of displacements, velocities and accelerations at time tn.
The final form of the central difference approximation of velocity at time tn + 1 can be
obtained as a linear combination of the results in (1.59) and (1.61) above, as well as the
equivalent result for the subsequent time step,1 which results in:
h
vn + 1 = vn +
2
( an + an +1 ) (1.62)
1
1 vn + 1 =
2
(vn + 3 / 2 + vn +1/ 2 ) and vn + 3 / 2 = vn +1/ 2 + 2h an +1
Table 1.1 Central Difference Scheme in Nonlinear Dynamics
(
Initialize: d 0 , υ0 → a0 = M−1 f 0ext − fˆ int (d 0 ) )
Compute for each step: n = 0,1, 2... ( given : d n , υn , an and h)
d n+1 = d n + hv n + h2 an
Man +1 + fˆ int (d n +1) = f next

+1 ⇒ an +1 = M
−1
( ˆ int (d )
+1 − f
f next )
n +1
h
υn +1 = υn +
2
(an + an +1)
Next: n ← n + 1
In the last expression, the acceleration vectors an and an + 1 are computed directly from the
equations of motion at times tn and tn + 1, respectively. The most efficient implementation of
the central difference scheme is presented in Table 1.1.
The most costly phase in the proposed implementation of the central difference scheme
clearly pertains to the solution of a set of algebraic equations for computing the acceleration
vector an + 1. However, this cost can be reduced considerably by using a diagonal form of the
mass matrix, which allows for the solution to be obtained with the number of operations
equal only to the number of equations “n.” With this kind of computational efficiency, we
can easily accept a very small time step that is often required to meet the conditional stabil-
ity of the explicit scheme. For example, for the 1D hyperelastic bar (with free energy density
ψ(λ)) and 2-node finite element approximations (with a typical element length le), the condi-
tional stability of the central difference scheme requires (e.g. Oden and Fost 1973) the time
step “h” no larger than:
1/ 2
le  C (λ)  d 2 ψ(λ)
h≤ ; cmax = max   ; C (λ) = (1.63)
6cmax ∀λ > 0
 ρ  dλ2
Interestingly enough, by using a diagonal form of the mass matrix, the largest acceptable
time step which guarantees the stability of the central difference scheme increases to:
le
h≤ (1.64)
2cmax
However, in a number of cases of practical interest, the restriction placed by the conditional
stability of the explicit central difference scheme can be too severe, forcing us to take a time
step which is too small with respect to the required result accuracy. For that reason, the
central difference scheme is mostly used in applications with dynamics phenomena of short
duration, such as impact problems, explosions or wave propagation with very short dura-
tion; in any such problem, the computed response would have a significant contribution of
high-frequency modes and would require very small time steps, which can easily be handled
with the central difference scheme.
Remarks:
1. For nonlinear inelastic behavior in dynamics, the central difference scheme provides
a very interesting alternative to fully implicit schemes. Namely, we can solve the
equations of motion by the explicit central difference scheme, combined with an
implicit scheme solution of a small set of the evolutions equations for internal vari-
ables at each numerical integration point. This kind of explicit–implicit approach to
nonlinear dynamics problems with inelastic behavior can be implemented within the
framework of the operator split procedure. We, thus, obtain the highest computa-
tional efficiency in solving a large set of equations of motion that provides the best
value of displacements dn(i+1 )
as well as the corresponding value of the total deforma-
(i )
tion field ε n +1, along with results reliability for the local computations of internal
variables and admissible value of stress that is carried out by an implicit scheme (e.g.
the backward Euler scheme). The latter should be implemented in a manner which
guarantees the convergence of the local computation that enforces the stress admis-
sibility constraint in agreement with chosen plasticity or damage criteria (e.g. the
stress value cannot be larger than given plasticity or damage threshold). It is impor-
tant to note that only stress values are needed for this kind of strategy, with no need
to compute the tangent moduli.
2. We have implemented this kind of strategy for a coupled damage-plasticity constitu-
tive model suitable for representing the behavior of concrete under impact (see Hervé,
Gatuingt and Ibrahimbegović 2005). This model employs a judicious combination
from the damage model of Mazars (see Mazars 1986) for representing the concrete
cracking with the threshold related to the principal elastic strains, and the Gurson-like
(see Gurson 1977 or Needleman and Tvergaard 1984) viscoplasticity model, capable
of representing the concrete hardening behavior under compressive stress. A graphic
illustration of the elastic domain defined by the proposed criterion is represented in
Figure 1.5. We note that there is a considerable complexity of the constitutive model
of this kind with a fairly long list of internal variables (such as the damage variable
dn + 1, viscoplastic strains ε nυp+1, the porosity f*, as well as the hardening variables; e.g.
Hervé, Gatuingt and Ibrahimbegović 2005), which would represent quite a significant
challenge for guaranteeing the convergence of the fully implicit schemes. However, the
explicit computations of equations of motion, accompanied by only implicit computa-
tions of internal variables does not have any such difficulty. A couple of illustrative
results which concern the impact computation on a concrete slab, both for the cases
with and without perforation, are presented in Figures 1.6 and 1.7.
1.5.3 Trapezoidal Rule or Average Acceleration (Implicit) Scheme

For a large number of problems in low-velocity dynamics, such as the problems typical
of earthquake engineering with significant damping, the dynamic response is to a large
extent governed by fairly few, low-frequency modes. One finds that the contribution of
Figure 1.5 Steel and concrete frame structure (Clough and Penzien 1993).
Figure 1.6 Elastic domain for a coupled damage-plasticity model of concrete under impact (see Hervé,
Fabrice Gatuingt and Ibrahimbegović 2005).
Figure 1.7 Impact on a reinforced concrete slab: contours of damage variable for the cases with and without
perforation (see Hervé, Fabrice Gatuingt and Ibrahimbegović 2005).
high-frequency modes for such problems is not very pronounced and that they are quickly
damped out. Integrating the equations of motion for one such problem in dynamics by using
an explicit scheme, with a very small time step that must be selected to guarantee the sta-
bility, would imply a significant, yet unnecessary increase in computational cost. We, thus,
have to turn to implicit schemes in order to allow for larger time steps.
For linear problems in dynamics where the response is dominated by low frequencies, one
can use the modal superposition method with mode truncation keeping a few modes only,
in order to reduce the computational cost without sacrificing the accuracy. For a nonlinear
dynamics problem, where the mode superposition method is not applicable, the computa-
tional cost of implicit time-integration schemes is not easy to reduce. However, this does
not mean we will discard implicit schemes; for them, we can provide a more reliable solu-
tion satisfying the equations of motion for all chosen instants of the incremental sequence.
Any implicit time-integration scheme for nonlinear dynamics will lead to a set of coupled
nonlinear algebraic equations, which requires an iterative solution scheme with a signifi-
cant increase in computational cost at each step. The only manner of compensating for this
increase of computational cost is by using (much) larger time steps with respect to those
typical of explicit schemes.
The trapezoidal rule, sometimes called the average acceleration method, is the first implicit
scheme we will discuss subsequently. By applying the trapezoidal rule (e.g. Dahlquist and
Björck 1974, p. 347) to the equations of motion, we obtain the second-order approximation
to evolution equations for displacement and velocity, which can be written:
h
d (t) = v(t) ⇒ dn +1 − dn = ( vn + vn +1 )
2 (1.65)
h
v(t) = a(t) ⇒ vn +1 − vn = ( an + an +1 )
2
Rewriting the result in (1.59)1 we can obtain the corresponding approximation for the veloc-
ity vector in terms of displacement increment:
2
vn + 1 = − vn +
h
(dn +1 − dn ) (1.66)
With this result in hand, we can then obtain from (1.59)2 the same kind of approximation
of the acceleration vector at time tn + 1 which reads:
4 4
an +1 = − an − vn + 2 ( dn +1 − dn ) (1.67)
h h
Note that both of these approximations are implicit in the sense that they depend upon the
displacement value at time tn + 1. Combining these velocity and acceleration approximations
with the equations of motion written at time tn + 1, we can obtain the most suitable implemen-
tation of the trapezoidal rule as presented in Table 1.2.
In terms of computational cost, the most demanding phase in the proposed implementa-
tion of the trapezoidal rule pertains to iterative solution of the set of nonlinear algebraic
equations with displacements at time tn + 1 as unknowns. If Newton’s method is employed for
Table 1.2 Trapezoidal Rule or Average Acceleration

Scheme for Nonlinear Dynamics
(
Initialize: d 0 , υ0 → M−1 f 0ext − f int (d 0 ) )
At each step: n = 0,1,2…(given: dn, vn, an and h)
Find dn + 1, υn + 1, an + 1
Iterate (i) = 1,2…
( ) h4 (d − d ) − 4h υ − a
an( i+)1 = a d n( i+)1 := 2
(i)
n +1 n n n
Maˆ ( d ) + fˆ ( d ) = f → d
(i)
n +1
int (i)
n +1
ext
n +1
( i +1)
n +1
υ = (d − d ) − a
(i) 2 (i)
n +1 n +1 n n
h
IF f − Ma − f ( d ) THEN (i ) → (i + 1) → d
ext
n +1
(i)
n +1
int (i)
n +1 n +1
IF
ELSE n → n + 1
this computation, we have to solve, at each iteration, the consistently linearized form of the
system, which can be written:

4 (i )  (i ) ext int (i )
( (i )
)
 h2 M + Kn +1  un +1 = fn +1 − f dn +1 − Ma dn +1 (1.68) ( )
The matrix Kn +1 = (∂f int (dn +1)) / (∂dn +1) above is the tangent stiffness already defined for non-
linear problems of static equilibrium. The tangent operator in dynamics is yet referred to
as the effective tangent stiffness, which includes both the true tangent stiffness for statics
and the corresponding contribution from the mass matrix. The presence of the mass matrix
(with full rank) in the effective tangent stiffness has the regularizing effect, which ensures
the correct rank. We can thus always obtain the unique solution for incremental displace-
ment and carry out the displacement update according to:
dn(i++11) = dn(i+) 1 + un(i+) 1 (1.69)
This iterative procedure continues until the chosen convergence tolerance is reached, provid-
ing the final displacement value dn+1 for that particular step. For linear dynamics problems,
( )
where the internal force can be written as fˆ int dn(i+) 1 = Kdn +1, the trapezoidal rule guaran-
tees the unconditional stability of computations regardless of the time step size. Moreover,
the trapezoidal rule ensures energy conservation in linear dynamics problems, for the free-
vibration phase of motion in the absence of external loading. Namely, with the time deriva-
tive of total energy identical to the equations of motion, it is easy to conclude that:

d
dt
( )
[E(t)] = d(t) × Md(t) + Kd(t) = E(t) = cst. (1.70)

Gnnn (ϕ , v ) = 0
The discrete approximation of the free-vibration problem in linear dynamics provided by

the trapezoidal rule will also ensure the conservation of the total energy with:
dn + 1 − dn a + an dn +1 − dn d + dn + 1
0= ⋅ M n +1 + ⋅K n =
h 2 h 2
1 1
=
2
( v n + 1 + v n ) ⋅ M ( v n + 1 − v n ) + 2 ( dn + 1 − dn ) ⋅ K ( dn + 1 − dn ) =
(1.71)
1 1  1 1 
=  vn +1 ⋅ Mvn +1 + dn +1Kdn +1  −  vn ⋅ Mvn + dn ⋅ Kdn  =
2 2   2 2 
= En +1 − En
The result of this kind is an additional confirmation of the unconditional stability of the
trapezoidal rule method for linear dynamics problems. The unconditional stability of the
same scheme in nonlinear dynamics is guaranteed only (see Belytschko and Schoeberle
1975) in the case where the discrete approximation of the internal energy verifies the fol-
lowing condition:1
( )
ψ n +1: = ψ n + un +1 ⋅ fˆ int ( dn +1 ) + fˆ int ( dn ) / 2 ≥ 0 (1.72)
1 We note in passing that the inertia term in the Lagrangian formulation of nonlinear dynamics still remains
linear; hence, the stability condition for nonlinear dynamics only concerns the internal energy.
Such a condition is violated (e.g. see Hughes 1977) for nonlinear dynamics problems with
softening behavior.1 Moreover, the trapezoidal rule can no longer ensure the energy conser-
vation for nonlinear dynamics problems, which is one of the main motivations for the study
of other implicit time-integration schemes, such as mid-point rule.
1.6 MOTIVATION FOR DEVELOPMENTS

PRESENTED IN PART I AND PART II
1.6.1 Damping Model Characterization

For structural systems where a reasonable degree of homogeneity in the energy loss mech-
anism exists, modeling damping as proportional or classical is suitable and convenient.
However, for structures made up of more than a single type of material, where the different
materials provide drastically differing energy loss mechanisms in various parts of the struc-
ture, the distribution of damping forces will not be similar to the distribution of the inertial
and elastic forces; in other words, the resulting damping will be nonproportional (Clough
and Penzien 1993).
An easier remedy to this problem would be to use the material damage as the source of
damping. Here, one needs to define nonlinear inelastic models, which pertains to any par-
ticular material. In Part I of this book, we present detailed studies for the most frequently
used materials in engineering applications, such as steel, concrete and masonry.
During vibration motion (earthquake, etc.) of each and every structure, there would be
some energy loss. The latter pertains to damping phenomena in vibrating structures, which
in many expositions on the subject have been defined with great uncertainty with no clear
recommendations on how to define the damping forces. Damping capacity is defined as the
ratio of the energy dissipated in one cycle of oscillation to the maximum amount of energy
accumulated in the structure. Material damping and interface damping are the most fre-
quently present damping sources in the structure among many others. The material damp-
ing contribution comes from different damage mechanisms and their interaction within the
material from which the structure is built. Thus, the damping is dependent upon the type of
material, as well as on manufacturing methods (Kareem and Gurley 1996). Taking all this
into account and the fact that the material characteristics vary due to material heterogene-
ities, providing the reliable estimation of material damage on structural damping is perhaps
the most reliable path to constructing predictive models. This is illustrated in Part I, for
most frequently used structure materials.
The interface damping mechanism is Coulomb friction between members and connec-
tions of a structural system and nonstructural components like partitions, facades and other
frictional mechanisms still being unknown and not precisely identified. It is interesting to
note that many experiments have shown that most of the energy dissipation mechanisms
through the structure are dependent on displacement amplitude rather than the frequency of
the structure. The mathematical formulation of the Coulomb friction is written as:
u
fd = µ = µ sgn ( u )
u
where:
μ is the friction coefficient of contacting surfaces and
u is the first derivative of displacement.
1 The softening behavior implies the stress decrease with increasing strain.
It is clear that a nonlinear analysis is required here which most probably will consume
much computational time and as such is not commonly used in the engineering practice.
This is simplified into a linear viscous damping model represented in the well-known
formulation:
fd = cu
In contrast to the real structure’s behavior (see above), viscous damping is frequency depen-
dent and not displacement dependent. It is clear that nonlinearity is not covered by this
model.
1.6.2 Extreme Loading Representation

Once we develop a more detailed model of damping phenomena, we can significantly
increase the model’s predictive capabilities. Further increase can also be achieved through
the more reliable representation of extreme loading conditions. In Part II, we provide the
corresponding developments for some of the extreme loads applied to engineering struc-
tures. We include among them the most frequently encountered types of loading such as
earthquakes and fires, as well as the impact loads applied to structures, either from another
structure (e.g. airplane) or fluid (e.g. dam break).
Any of these representations for extreme loading applications require a more general
framework, which pertains both to improving the structure formulations and to additional
field computations (e.g. temperature, fluid flow or ground or other structure motion). The
large diversity of such applications leads to multi-physics and interaction problems that need
to be solved in order to sufficiently obtain a predictive model for engineering structures
under extreme loading conditions. Several illustrative examples presented in Part II, deal-
ing with earthquakes, airplane impact, fire and dam break, are most likely to be useful for
engineering practice.
REFERENCES
Adhikari, S. and J. Woodhouse. 2001. “Identification of Damping: Part 1. Viscous Damping”. Journal
of Sound and Vibration 243 (1): 43–61. doi:10.1006/jsvi.2000.3391.
Belytschko, T. and D. F. Schoeberle. 1975. “On the Unconditional Stability of an Implicit
Algorithm for Nonlinear Structural Dynamics”. Journal of Applied Mechanics 42 (4): 865.
doi:10.1115/1.3423721.
Bernal, Dionisio. 1994. “Viscous Damping in Inelastic Structural Response”. Journal of Structural
Engineering 120 (4): 1240–54. doi:10.1061/(asce)0733-9445(1994)120:4(1240).
Charney, F. A. 2005. “Consequences of Using Rayleigh Damping in Inelastic Response History
Analysis”. Congreso Chileno de Sismologia e Ingenieria Antisismica. Conception. Paper No.
A10–17.
Chen, H. C. 1987. Solution Methods for Damped Linear Systems. UCB/SEMM Rep. 87/08, Berkeley:
University of California.
Clough, Ray W. and Joseph Penzien. 1993. Dynamics of Structures. First edition.New York: McGraw-
Hill Education (ISE Editions).
Clough, Ray W. and Joseph Penzien. 2003. Dynamics of Structures. Second edition.New York:
McGraw-Hill Education (ISE Editions).
Dahlquist, G. and A. Björck. 1974. Numerical Methods. Englewood Cliffs, NJ: Prentice-Hall.
Géradin, M. and D. Rixen. 1992. Théorie des vibrations: Applications à la dynamique des struc-
tures. Paris: Masson.
Gurson, A. L. 1977. “Continuum Theory of Ductile Rupture by Void Nucleation and Growth: Part
I —Yield Criteria and Flow Rules for Porous Ductile Media”. Journal of Engineering Materials
and Technology 99. doi:10.1115/1.3443401.
Hall, John F. 2006. “Problems Encountered from the Use (or Misuse) of Rayleigh Damping”.
Earthquake Engineering & Structural Dynamics 35 (5): 525–45. doi:10.1002/eqe.541.
Hervé, Guillaume, F. Gatuingt and Adnan Ibrahimbegović. 2005. “On Numerical Implementation of
a Coupled Rate-Dependent Damage‐plasticity Constitutive Model for Concrete in Application
to High‐rate Dynamics”. Engineering Computations 22 (5/6).
Hughes, Thomas J. R. 1977. “A Note on the Stability of Newmark’s Algorithm in Nonlinear
Structural Dynamics”. International Journal for Numerical Methods in Engineering 11 (2):
383–86. doi:10.1002/nme.1620110212.
Ibrahimbegović, A., H. C. Chen, E. L. Wilson and R. L. Taylor. 1991. “Ritz Method for Dynamic
Analysis of Large Discrete Linear System with Non-Proportional Damping”. Earthquake
Engineering and Structural Dynamics 19: 887–99.
Ibrahimbegović, A. and E. L. Wilson. 1989. “Simple Numerical Algorithms for the Mode Superposition
Analysis of Linear Structural Systems with Non-Proportional Damping”. Computers &
Structures 33 (2): 523–31. doi:10.1016/0045-7949(89)90026-6.
Ibrahimbegović, Adnan and Edward L. Wilson. 1989. “Simple Numerical Algorithm for the Mode
Superposition Analysis of Linear Structural Systems with Non-Proportional Damping”.
Computer and Science 33: 523–33.
Kareem, A. and K. Gurley. 1996. “Damping in Structures: Its Evaluation and Treatment of
Uncertainty”. Journal of Wind Engineering and Industrial Aerodynamics 59 (2–3): 131–57.
doi:10.1016/0167-6105(96)00004-9.
Mazars, Jacky. 1986. “A Description of Micro- and Macroscale Damage of Concrete Structures”.
Engineering Fracture Mechanics 25 (5–6): 729–37. doi:10.1016/0013-7944(86)90036-6.
Medina, R. A. and H. Krawinkler. 2004. Seismic Demands for Non-Deteriorating Frame Structures
and their Dependence on Ground Motions Berkeley. PEER Report, Berkeley, CA: Pacific
Earthquake Engineering Research Cente.
Morris, Nicholas F. 1997. “The Use of Modal Superposition in Nonlinear Dynamics”. Computers &
Structures 7 (1): 65–72. doi:10.1016/0045-7949(77)90061-x.
Needleman, A. and V. Tvergaard. 1984. “An Analysis of Ductile Rupture in Notched Bars”. Journal
of the Mechanics and Physics of Solids 32 (6): 461–90. doi:10.1016/0022-5096(84)90031-0.
Oden, J. T. and R. B. Fost. 1973. “Convergence, Accuracy and Stability of Finite Element
Approximations of a Class of Non-Linear Hyperbolic Equations”. International Journal for
Numerical Methods in Engineering 6 (3): 357–65. doi:10.1002/nme.162006030.
Phani, A. S. 2004. Damping Identification In Linear Vibrations. PhD Thesis, Cambridge: Cambridge
University Engineering Department.
Udwadia, Firdaus E. 2009. “A Note on Nonproportional Damping”. Journal of Engineering
Mechanics 135 (11): 1248–56. doi:10.1061/(asce)0733-9399(2009)135:11(1248).
Veletsos, Anestis S. and Carlos E. Ventura. 1986. “Modal Analysis of Non-Classically Damped
Linear Systems”. Earthquake Engineering & Structural Dynamics 14 (2): 217–43. doi:10.1002/
eqe.4290140205.
Zareian, F. 2006. Simplified Performance-Based Earthquake Engineering. PhD Dissertation,
Stanford, CA: Department of Civil & Environmental Engineering, Stanford University.
Part 1
Chapter 2
Steel structures
In the following sections the governing equations of a boundary value problem are developed
for the case where the computed strain field can be highly non-homogeneous, with one or
more sub-domains where the peak resistance is passed leading to a strain softening regime
with very large strains, while the rest of the structure still remains in either elastic or elas-
toplastic strain hardening regime with fairly small strains. The model developed herein for
representing this kind of problem is capable of describing both strain hardening and strain
softening regimes, as well as identifying the stress state that corresponds to the passage from
hardening to softening. In particular, the latter is considered to coincide with the satisfac-
tion of the localization condition (e.g. Hill 1962; Rice 1976) modified for the presence of
hardening.
2.1 ESSENTIAL INGREDIENTS OF PLASTICITY MODEL
As it is well known, plasticity theory began with Tresca in 1864, when he undertook an
experimental program into the extrusion of metals and published his famous yield criterion.
Further advances with yield criteria and plastic flow rules were made by different research-
ers such as Saint-Venant, Levy, von Mises, Hencky and Prandtl. It was in the 1980s and
1990s that a more rigorous foundation based on thermodynamic principles was developed
opening the door to many numerical and computational aspects of the plasticity problem.
It is still the most well understood constitutive model with internal variables capable of
describing several observed phenomena of inelastic behavior. In the following section, the
plasticity model with hardening and introduced softening will be elaborated in detail. This
is used as a basis for implementation in structures and further in push-over analysis. At the
end of the chapter, a refined model is presented that represents a novelty to the current prac-
tice. The chapter closes with several examples.
2.1.1 1D Plasticity with Hardening and Softening

The plasticity model with hardening is just a minor modification of the perfect plasticity
model, in the sense that an increase of the plasticity threshold is enabled beyond the limit of
elasticity σy as a function of accumulated plastic deformations. On the other hand, softening
can be seen as a decrease of the stress. In this sense, it can be stated that the sign of the yield
stress changes; an increase would lead to hardening and a decrease (or negative) hardening
would be equivalent to softening. For simplicity, a 1D model for a simple shear test is con-
sidered. This model is represented by a bar of length l (of a unit section), fixed at one end
and submitted to an imposed shear force at the other end. (Figure 2.1).
25
Figure 2.1 Simple tension test.
As displacements are considered to be infinitesimal, the small displacement gradient the-

ory is valid, so the application of standard kinematics and equilibrium equations are valid:
∂u
ε= (2.1)
∂x
∂σ
+ b = 0 (2.2)
∂x
where u ,ε and σ, are, respectively, transverse displacement, infinitesimal deformation and

the Cauchy stress field, b is the external distributed loading. All the fields with a superposed
bar indicated in (2.1) are considered as smooth, which tacitly implies hardening plasticity.
Setting the external loading b represented in (2.2) to zero, it is obvious that the stress σ will
also be smooth (constant) along the bar.
Taking the essential ingredients, one can construct the constitutive model of plasticity.
The total strain can be decomposed into elastic ε e and plastic components ε p:
ε = ε e + ε p (2.3)
The thermodynamic for plasticity with hardening is represented by the potential strain
energy function depending upon the elastic strain and hardening variable ξ:
(
ψ εe, ξ =) 1 e e
2
()
ε Eε + Ξ ξ (2.4)
And finally, for the developments to follow, the yield criterion specifying the admissible
values of stress and stress-like hardening variable q is represented in terms of variables σ, q :
φ ( σ, q ) = σ − ( σ y − q ) ≤ 0 (2.5)
where σy denotes the yield stress.

All the remaining ingredients of the plasticity model can be obtained from standard ther-
modynamic considerations and the principle of the maximum plastic dissipation (Lubliner
1990). Namely, the second law of thermodynamics (Truesdell and Noll 1965; Lubliner 1990;
Steel structures 27
Maugin 1992) states that the dissipation always remains non-negative, which can be written
by making use of the results from (2.3) and (2.4) as:
d
0 ≤  = σε − ψ ε e , ξ
dt
( )
(2.6)
 ∂ψ  ∂ψ ∂Ξ
=  σ − e  ε e + e ε p − ξ
 ∂ε  ∂ε ∂ξ
It is clear from (2.6) that for any stress state in the elastic domain, the values of internal

( )
variables will not change, ε p = 0, ξ = 0 and dissipation remains equal to zero. Then the
stress-like variables can be defined through the following constitutive equations:

φ < 0, ε p = 0, ξ = 0
(2.7)
∂ψ ∂Ξ
 =0⇒σ= ;q = −
∂ε e ∂ξ
By assuming the same constitutive relation in (2.7) remains valid for plastically admissible
stress on the yield surface, the corresponding (positive) value of plastic dissipation from the
second principle of thermodynamics can be obtained according to:
∂ε p ∂ξ
φ = 0, 0 <  p = σ σ+q (2.8)
∂t ∂t
The principle of maximum plastic dissipation can be used in order to obtain the evolution
equations of internal variables for such a process. One looks into all the plastically admis-
( )
sible variables of stress satisfying (2.5) (for which φ σ*, q* = 0), for the true solution ( σ, q )
that will maximize plastic dissipation, or otherwise:
 
(σ, q ) = arg  min (
 −  p σ*, q* ) (2.9)
φ (σ* , q* ) 
 
The Lagrange multiplier method can be used to handle constraint and define the corre-
sponding unconstrained minimization problem:
( )
Lp σ, q, γ = max
*
∀γ > 0
(
min Lp σ*, q*, γ ;
∀(σ* , q* )
)
(2.10)
∂γ
p
( )
L σ, q, γ = − D p ( σ, q ) +
∂t
φ ( σ, q )
The Kuhn–Tucker optimality conditions (e.g. Luenberger 1984; Strang 1986) of this prob-
lem will provide the evolution equation of the plastic strain and hardening variable.
∂p ∂ ε p ∂φ
=0⇒ =γ = σsign ( σ )
∂σ* σ
∂t ∂σ
(2.11)
∂p ∂ ξ ∂γ ∂ φ
=0⇒ = =γ
∂q q
∂t ∂t ∂q
The constraint equation is obtained as well, which can be combined with the corresponding
value in (2.8) in an elastic case in order to provide the final form of the loading/unloading
conditions:
γ ≥ 0, φ ≤ 0, γφ
= 0 (2.12)
If one recalls the model of perfect plasticity, it is clearly observed that the presented model
gives the same evolution equation for plastic deformations and the same stress constitutive
equation for plasticity deformation and the same stress constitutive relation.
The appropriate value of the plastic multiplier can be obtained from the consistency con-
ditions to guarantee the admissibility of stress for subsequent states, meaning the only case
of plastic loading that gives a non-zero value of γ :
 
 ∂ φ 
σ = E  ε − γ
e

∂ σ
 
ε p 
d 2Ξ
q = − 2 ξ
d ξ ∂φ
γ
K ∂q
d ∂ φ ∂σ ∂ φ ∂q
0= φ= +
dt ∂σ ∂t ∂q ∂t
∂ φ ∂ ε ∂γ  ∂φ ∂φ ∂φ ∂φ 
= E −  ∂σ E ∂σ + ∂q K ∂q  (2.13)
∂σ ∂t ∂t
∂φ ∂ ε
E
γ = ∂γ = ∂σ ∂t
∂t ∂φ ∂φ ∂φ ∂φ
E + K
∂σ ∂ σ ∂q ∂q
Further, the stress rate form of the stress–strain constitutive equations can be expressed
according to:
 Eε

  ∂φ 
σ =   E E
∂σ ∂ φ  ε , γ > 0 (2.14)
  E − 
∂ φ ∂ φ ∂ φ ∂ φ ∂σ 
 E + K

  ∂ σ ∂σ ∂ q ∂ q 

Figure 2.2 clearly indicates linear isotropic vs. saturated hardening. In particular, for the
yield criterion set in (2.5), leading to ∂ φ / ∂σ = sign(σ) and ∂ φ / ∂q = 1, a simplified form of
the stress rate equation for plastic loading can be defined as:
EK
σ = ε γ > 0 (2.15)
E+K
Steel structures 29
Figure 2.2 (a) Linear isotropic hardening (b) Saturated hardening.
This clearly reveals that a further increase of stress is allowed beyond elastic limit σy with
increasing values of strains only as long as one is kept in a strain hardening regime with
( )
K = d 2Ξ / d ξ 2 > 0. The peak resistance in the stress–strain diagram is identified at:
EK d 2Ξ
 ep = =0⇒K = = 0 (2.16)
E+K dξ2
( ) ( )
Assuming that K ξu = 0, defining the ultimate value of stress as σ u = σ y − q ξu and if then
the softening phase is entered, with K < 0, according to (2.15) the increase in strain will lead
to a decrease in stress. Namely, in the latter case, the 1D problem under consideration with
zero external load and constant stress state will lead to the gradual spreading of the plastic-
ity throughout the domain, starting from the first plastified section, since the stress will keep
increasing. In the same constant stress state and strain softening case, the first section that
passes the peak (typically a slightly weakened section that is supposed to transform a bifur-
cation into a limit load problem) will reduce the stress level leading to unloading in all other
sections where peak is not passed; all subsequent inelastic deformation will be accumulated in
the particular section where the shear band is created. For this specific problem, taking into
account that the shear band thickness tends to zero, the displacement field can be written as:
u(x, t) = u(x, t) + η(t)Mx (x) (2.17)
with
Mx (x) = H x (x) − N a (x);
1, x>x
H x (x) =  ; (2.18)
0, x<x
1, x=l
N a (x) =  ;
0, x=0
In (2.17), u(x, t) is the smooth displacement field, and η(t) is the displacement discontinuity
which appears at the shear band location x.
By differentiating (2.18), the corresponding strain field is obtained:
ε(x, t) =
ε(x
, t) +η x) + η(t)δ x (x) (2.19)
(t)G(
ε
It consists of regular (smooth) ε and singular components (the third term in (2.19)), where
δ x denotes the Dirac-delta function at point x.
In line with this new form of the strain field, the strain energy consists of a regular, as set
in (2.4), and a singular part (the second term) as:
( ) ( ) ()
ψ ε, ξ, ξ = ψ ε e , ξ + ψ ξ δ x (x) (2.20)
The internal variable ξ describes the softening phenomena on discontinuity. The results
obtained in (2.20) can be used to write the dissipation inequality as:
∫ σε − dt ψ (ε, ξ, ξ )dx

 d 
0 ≤ Ωloc =
Ω
(2.21)
Ω
∫


d
= σε − ψ ε e , ξ
dt
( )   d 
dx + t η − dt ψ ξ  ()
x
The definition of the Dirac-delta function in (2.19) is used in order to obtain the traction at
discontinuity as:
t = σ x (2.22)
Comparing the expression in (2.21) with the one in (2.8), it can be concluded that the dis-
sipation in the post-localization phase can be reduced to the sum of the regular and singular
components which can be written as:
∂ψ
Ωloc =
∫ σε
Ω
+ qξ dx + q ξ 

p

Ω
  x
;q = −
∂ξ
(2.23)
with the orthogonality condition set in (2.24), which should be imposed and solved simulta-
neously with the global equilibrium equations, the following condition applies:

∫ G(x)σdx + t = 0 (2.24)
Ω
The contribution of the discontinuity to the total dissipation is defined in (2.21) and repre-
sents the second term in the equation. In order to further develop the exact nature of such a
contribution, the yield criterion at discontinuity is postulated, which defines the correspond-
ing admissible traction values according to:
( ) (
φ t, q := t − σ u − q = 0 (2.25) )
The principle of maximum plastic dissipation applied to the case where only discontinuity
remains active can be written by introducing the Lagrange multiplier in the form:
γ = γδ x (2.26)
which further restricts the corresponding Lagrangian to the discontinuity according to:
   
( )
p q, γ =  −q ξ  + =
 x ∫

γδ φdx =  −q ξ + γφ  (2.27)

x

Ω γ x
Steel structures 31
This reduces the Kuhn–Tucker optimality conditions to:

ξ = γ
(2.28)
γ ≥ 0, φ ≤ 0, γφ
=0
The time derivative of this yield criterion provides the consistency condition for computing
the corresponding value of the plastic multiplier with:
∂q 1
0 = sign(t)t + ξ ⇒ γ = t sign(t) (2.29)
∂ξ

K
−K
where the result in (2.28) is exploited. The rate of change of the traction at discontinuity can
be obtained from the time derivative of the orthogonality condition in (2.23), leading to:
1
γ =
KΩ ∫
Gσ dx (2.30)
The presence of discontinuity does not at all affect the elastic response as the rigid-plastic
traction displacement law is employed. The corresponding coefficients of the traction dis-
placement law to be chosen are the traction yield value t y = σ u , which is chosen in accor-
dance with the ultimate values of stress in the hardening regime, and the softening modulus
which depends on a particular choice of softening law.
The total external work (Gf ) expended in the process of driving the effective flow traction
t yeff = t y − q is set to zero. If a linear softening law is chosen, the corresponding value of the
softening modulus K can be obtained from the following condition:
ty
0 = t y − K ξu ⇒ ξu =
K
(2.31)
1 1 t y2 1 t y2
Gf = t y ξu = ⇒ K =−
2 2 K 2 Gf
Any other softening regime, instead of the linear softening law chosen in (2.31), can be
selected, and in this respect adequate modification should be made. For example, with an
exponential softening response it leads to:
∞
( ) ( ) ()  ty 
∞
ty 
Gf =
∫
0
t y exp −αξ d ξ = −
α 

exp −αξ  ⇒ t y = −q ξ = t y exp  −

 0  Gf
ξ  (2.32)

This can be further extended to the 2D/3D case, with the tensor notation as the only nov-
elty. For further details see Ibrahimbegović (2009).
2.2 LOCALIZED FAILURE PLASTICITY MODEL
In the spirit of the classical works on localization problems (Hill 1962; Mandel 1996; Rice
1976), it is assumed that the bifurcation phenomena in an elastoplastic response can be
interpolated as the difference between two smooth stress fields, defining the corresponding
jump in the stress rate as:
σ = Cep ( m ⊗ nη ) (2.33)

Additionally, according to Newton’s Third Law which imposes the continuity of traction
across the displacement discontinuity line Γs and the Cauchy principle which relates the trac-
tion vector to stress tensor, it can be written:
0 = t = σ n + [σ ] n
(2.34)
=0
The second term in (2.34) is equal to zero as it is assumed that the direction of the discon-
tinuity remains fixed in time. By combining (2.33) and (2.34), the following is obtained:

(
0 = Cep m ⊗ nη n ) (2.35)

= = ηA ep m
where Aep is the acoustic tensor. The acoustic tensor in this case remains compatible with
the von Mises yield criterion, where the plastic deformation in the softening phase remains
incompressible and can be written as:
2µ
A ep = µ1 − sn ⊗ sn (2.36)
 K 2
 1 + 3µ  s
where μ and K are, respectively, the shear and hardening moduli.

The characteristic equations are obtained from the principal values of the acoustic tensor,
which can be written in the reference system of the principal values of stress:
 
 
  λ
( ) 2
( )
2
0 = det  A ep − λ1 = λ* − 1  × n12 s12 + n22 s22 + λ* − 1 ; λ* = (2.37)
 1+ K  2  µ
s
  3µ  
By imposing the localization condition in (2.35) with λ*1 = 0 from (2.37) it leads to:
 2 τ 
2
 1
K = 3µ  sin 2θ − 1 ; τ = ( σ1 − σ 2 ) (2.38)
 s   2
 
Additionally, assuming that the localization condition occurs at peak resistance where
( )
K ξu = 0 gives an angle θ under which the slip line is formed:
1 −1  s 
θ= sin  (2.39)
2  2 τ 
By computing the value of s in terms of the principal stress value, it can further lead to:
Steel structures 33
2 2
 2τ 
1= 
 p + 3Bε 33
p 
(1 − 2ν)2 ;
 +  3
 s 2  s 2 
(2.40)
E tr [σ ]
B= ; p=
3 (1 − 2 ν) 3
p
When the pressure p is related to 3Bε 33 , or for the case of (quasi-)incompressible material,
the slip line will occur at 45° with respect to the maximum principal stress as is clear from
(2.39) and (2.40).
Having established that the localization condition coincides with the initiation of the
displacement discontinuity, it is necessary to turn to the final phase of the analysis as the
discontinuity remains active in this location. The strain field, in this case, is written as:
h + η ( m ⊗ n ) δ Γ s (2.41)
ε = ∇s u = ∇s u + G
s

ε
ε
such that:
∫ G dΩ + ∫ (m ⊗ n) dΓ
s
s = 0 (2.42)
Ω Γs
The well-known result in distribution theory is used (e.g. Stakgold 1998) so that:

∫ φ∇H
Ω
Γs
dΩ =
∫ φndΓ (2.43)
Γs
s
In accordance with this form of the strain field, the strain energy can be split into two parts.
A smooth part and the discontinuity contribution:
( ) ( ) ()
ψ ε, ξ, ξ = ψ ε e , ξ + ψ ξ δ Γs (2.44)
The dissipation inequality can thus be written as:
∫ σ ⋅ ε − dt ψ (ε, ξ, ξ )dΩ

 d 
0 ≤ Ωloc =
Ω
(2.45)

∫
d
= σ ⋅ ε − ψ ε e , ξ
Ω
 dt
( ) 
d Ω +
Γ
∫ 

t ⋅ η
 ds − ∫
Γs
d
() 
 dt ψ ξ  ds
s
If, further, it is assumed that the stress orthogonality condition must be satisfied:

∫ σ ⋅ G h dΩ + ∫ t ⋅ h ds = 0
Ω Γs
(2.46)
From (2.45), the stress constitutive equation is recovered and the total dissipation reduced to
the sum of two parts, in the volume and slip line:

Ωloc =
∫ ∫
σ ⋅ ε p + q ξd Ω + q ⋅ ξds (2.47)

Ω

Γ

s

p p
Ω Γ
s
Once the localized solution develops, the stress orthogonality condition in (2.46) should be
joined and solved along with the set of global equilibrium equations. The yield condition
controlling the inelastic deformation at discontinuity is set directly in terms of the traction
vector component tm = t ⋅ m:
( )
φ t, q = tm − t y − q ≤ 0 (2.48) ( )
Γ , the principle of
Assuming, further, that the plastic multiplier takes the form γ = γ + γδ s
maximum plastic dissipation can be used with:
max

∀γ > 0 ∀(t* , q* )
(
min Lp σ*, q*, γ ; Lp (⋅) )
(2.49)
= − DΩloc (⋅) + γφ
dΩ +
∫
Ω
∫
Γs
γ φ (⋅) ds
The corresponding Kuhn–Tucker optimality conditions then reads:
 ∂φ 
0=
∫  − ε
Ω
p
+ γ
∂σ 
dΩ
(2.50)
 ∂φ 
0=
∫
Ω
 − ξ + γ ∂q  d Ω
together with:
 ∂φ 
∫  −ξ + γ ∂q  ds ⇒ ∫ ξds = ∫ γds (2.51)
0=
Γs Γs Γs
It is evident that the combination of two kinds of dissipation is essentially handled by the
corresponding value of the plastic multiplier. Namely, for the plastic loading case, it is either
γ or γ , or both which remain positive.
The corresponding value of the latter is obtained from the time derivative of (2.48) by
exploiting the results in (2.46) and (2.50) to get:
1
γ =
KΩ ∫
Gσ d Ω (2.52)
If the localization condition in (2.35) is reached within a given time increment, one can then
separate the step in one part where γ evolves and the remaining part, where only γ , will be
modified computing the corresponding dissipation as indicated in (2.52). The corresponding
“ultimate” value of hardening variable ξu occurring in such a time step can as well be computed.
2.3 STRUCTURAL PLASTICITY MODEL
2.3.1 Simple Form
Going further into the structural level, it is of great advantage to transform the weak form
of the differential equilibrium equation into a set of algebraic equations with the nodal
values of displacements as unknowns using the finite element based discrete approximation
(Hughes 1987; Bathe 1996; Zienkiewicz and Taylor 2000).
Steel structures 35
In this part, the basic idea will be presented with two examples. Further, the application
of such a model will be illustrated and implemented in a practical frame structure composed
of columns and beams. Why such structures? From experimental tests and seismic actions,
it has been observed from failure modes, that these structures fail due to localized failures
in critical zones, termed as plastic (inelastic) hinges. So, it has been observed that this model
is convenient for such representation.
Nel
A f int, e (t) − f ext,e (t) = 0
e =1
(2.53)
f int, e (t) =
∫
Ωe
BT σ(t)d W e , t ∈ [0, T ]
where (⋅) and (⋅) denotes, respectively, the element contributions toward the external and
ext,e int,e
internal force vector, whereas W e is the element domain. Parameter t in (2.53) is the pseudo-
time which is used to describe the particular loading program, which is typically described
by an incremental sequence.
It is only possible, if the evolution of the plastic strains ε p (t) and hardening variable ξ(t) is
known, to obtain, the complete response and the corresponding value of stress σ(t) in (2.53)
in the strain hardening phase.
The corresponding value of stress σ(t) can be obtained one incremental step at a time by
integrating the rate equations such that the loading/unloading conditions in (2.12) remain
satisfied. If the backward Euler is used for this kind of integration (Bathe 1996; Simo and
Hughes 2000), it is possible to compute the evolution of the plastic strains and hardening
variable over any time interval ∆t = tn +1 − tn with:
∂φ
εnp+1 = εnp + γ n +1
∂s n +1 (2.54)
ξn + 1 = ξn = + γ n + 1
which further implies that the corresponding stress evolution reads:
 ∂φ 
s n +1 = s n + C  ∆ε n +1 − γ n +1 (2.55)
 ∂s n +1 
The plastic multiplier γ n+1 in (2.54) and (2.55) is computed from plastic admissibility condition:
γ n +1 > 0; φ ( γ n +1 ) = 0 (2.56)
The computations described in (2.54) to (2.56) are carried out independently from one qua-
drative point to another for the chosen numerical integration rule (e.g. 2×2 Gauss quadratic)
(e.g. Hughes 1987; Bathe 1996) which is employed to compute the internal force integral in
(2.53). The main result of such a computation which is needed at the global level for solv-
ing the set of equilibrium equations in (2.53) is the plastically admissible value of stress in
(2.55), as well as the elastoplastic tangent modulus of the discrete problem:
Cep
∂s n + 1 2µ ( 2µ ) γ n + 1
2
n +1 = =C− s⊗s−
∂ εn +1  K 2 sn + 2µ∆en +1
 1 + 3µ  s
(2.57)
 1 1 
I − 2 sn +1 ⊗ sn +1 − 1 ⊗ 1
 sn +1 3 
The starting point for equivalent computations is the state where the localized conditions in
(2.35) are verified for one of the quadrative points where we check the stress values. Using
the same computations as in the hardening regime, the following is obtained:
( ) ( )
φ γ n +1 = 0; tm,n +1 = t y − q ξn +1 (2.58)
dq
Kn + 1 = (2.59)
d ξn + 1
The choice of the discrete approximation of the displacement field or the choice of element
type is detrimental in finite element implementation. Two problems ought to be addressed
in this respect: first, the locking phenomena in the plastic hardening behavior of the quasi-
incompressible type, which occurs for a given choice of the von Mises yield criterion ren-
dering the plastic deformation incompressible (Nagtegaal, Parks and Rice 1974). Second,
the displacement discontinuity representation in the finite element setting. In this respect, a
mixed-enhanced strain field is introduced, which, on one side, introduces the independent
volume-change deformation measure θ(x) and, on the other side, the displacement disconti-
nuity mode which produces a purely deviatoric strain field.
The set of equilibrium equations can be written as:
Nel  
 Ωe
∫
r = A  B ⋅ σd Ω − f ext  = 0
e =1  

(2.60)
he =
∫ G ⋅ σdΩ + ∫ hm ⋅ tdΓ = 0, e ∈[1, Nel]
Ωe Γs
where the supplementary equation expresses a weak form of equilibrium over a particular
element Ωe with respect to the variation of the enhanced strains.
In this way, the problem is reduced to its original size without the incompatible modes by
appealing to the static condensation procedure (e.g. Ibrahimbegović, Fadi and Lotfi 1998).
2.3.1.1 Numerical Examples of Small Experiments

The finite element models all employ the 4-node quadrilateral element with B strain interpo-
lations and embedded discontinuity. Two examples are provided: one is the pure shear test
and the other the traction test.
The former is presented in Figure 2.3 and the material properties are given in Table 2.1.
The problem is solved using the arclength method. In order to change the bifurcation
problem into a localization one, a band of elements (in gray in Figure 2.3) has been weak-
ened by slightly reducing the limit stress. The sensitivity of the proposed model to mesh
refinement and mesh distortion is checked by considering the different meshes as presented
in Figure 2.3.
The results are given in Figure 2.4 in terms of the horizontal shear load vs. the horizontal
displacement of point A. The numerical results are fully independent of mesh refinement
and distortion.
In the latter, a double notched beam submitted to a traction test under a plain strain
condition (see Figure 2.5) is considered. The tested sample is 20 cm long and 10 cm high,
and the material parameters are kept the same as in the former example with exceptions of
σ y = 0.55 GPa, σ ∞ = 0.75 GPa, K = −0.10 GPa/mm. The test is carried out under displacement
Steel structures 37
Figure 2.3 Pure shear test.
Table 2.1 Material Properties for Pure Shear Test

Continuum Plasticity Model
Young’s Modulus 210 GPa
Poisson Ratio 0.4999
σy 0.35 GPa
σ∞ 0.60 GPa
β 200
Softening Rule Parameter
−0.05 GPa/mm
K
control. Figure 2.6 shows the corresponding force–displacement diagram computed for this
simple shear test.
It is between the two notches of the beam where the shear band develops. Accumulated
plastic deformation and discontinuity lines activated in localized elements at the end of the test
are illustrated in Figures 2.7 and 2.8 respectively. It is evident from these two illustrations that
localized elements correspond to those where plastic deformation has significantly developed.
Figures 2.7 and 2.8 show the contributions of two different dissipative mechanisms. One
corresponds to bulk dissipation mechanism that is characterized by the development of plastic
deformation in fracture process zone and the other to a surface dissipation mechanism that
is characterized by the creation of a plastic slips resulting with displacement discontinuity.
2.3.2 Multi-Scale Model Parameters Identification

The application of the combined hardening and softening constitutive model of plasticity
can be applied to structures. One of the possible applications is in metallic frames with thin-
walled cross-sections, composed of columns and beams. When studying the full collapse of
Figure 2.4 Load/displacement results for a pure shear test.
Figure 2.5 Traction test.
a frame, a softening response is observed after reaching its limit capacity; the load reduces
with additional frame deformation. This structural softening response can be modeled using
an elastoplastic constitutive model with a softening relation between the generalized strain
measures and the corresponding stress resultants. As mentioned in the previous subsec-
tion, the key point here is the introduction of localized energy dissipation. This is achieved
by introducing a strong discontinuity in the kinematic fields. In this case, it would be in
axial displacement and beam axis rotation. Second, the local dissipative mechanism will
be defined at the discontinuity in terms of a softening cohesive law (e.g. a softening law
between the bending moment and the rotation jump). For beams, the introduced discontinu-
ity can be naturally regarded as a softening plastic hinge.
In respect of the previous example, a planar straight stress-resultant beam finite element
is developed with the following features: (i) Euler–Bernoulli kinematics, (ii) an elastoplastic
Steel structures 39
Figure 2.6 Load/displacement results for a simple shear test.
Figure 2.7 Accumulated plastic deformation.
stress-resultant constitutive model with isotropic hardening, (iii) a localized softening plas-
tic hinge related to the strong discontinuity in generalized displacements, and (iv) an approx-
imation of the geometrically nonlinear effects by using the von Karman strains for the
virtual axial deformations.
The finite element that will be developed in the following sections can be successfully
used for the limit load analysis, the push-over analysis and the complete failure analysis of
planar frames made of steel. More complex material models are used. In order to describe
the beam material behavior, stress-resultant elastoplasticity with hardening is used on one
hand and on the other stress-resultant rigid-plastic softening is employed to describe mate-
rial behavior at the discontinuity.
Figure 2.8 Localized elements and discontinuity lines.
The material parameters are obtained via numerical simulations on a representative part
of the beam by using a refined model, which is superior to the beam model as it has the abil-
ity to describe the beam response in more detail. For these kind of frames, the refined model
can be chosen as the nonlinear shell model (e.g. Brank, Perić and Damjan 1970; Brank
2005). The shell model is superior to the beam model in providing a proper local descrip-
tion of the strain/stress fields and the overall spread of the plasticity. This model goes even
further in the domain of the representation of the local buckling of the flanges and the web,
which is, in bending dominated conditions, very often the reason for localized beam failure.
Considering the above, the shell model can be seen as the mesoscale model and the beam
model as the macroscale model.
2.3.2.1 Beam Element With Embedded Discontinuity

As stated, a planar Euler–Bernoulli beam finite element is considered which can represent
elastoplastic bending, including the localized softening effects associated with the strong
discontinuity in the rotation. Further geometrical nonlinearity is taken into account approx-
imately by virtual axial strains of the von Karman type, which allows this element to cap-
ture the global buckling modes.
2.3.2.1.1 Kinematics
A straight planar frame member is taken into consideration where the middle axis occupies
domain Ω ∈ . A typical 2-node finite element is presented in Figure 2.9. The following
notation is used: ui are nodal axial displacements, wi are nodal transverse displacements, w ′
Figure 2.9 Beam finite element with embedded discontinuity.

Steel structures 41
are nodal values of the beam axis rotation and i = 1; 2 is the node number. A strong discon-
tinuity in axial displacement is assumed αu and beam axis rotation αθ at xd ∈ L(e) .
Further, it is presupposed that the domain of the discontinuity influence corresponds to a
single element. In respect to all these assumptions, the axial displacement is then defined as:
uh ( x, xd ) = N u (x)u + M u ( x, xd ) α u (2.61)
where N u (x) = {1 − x / L(e) , x / L(e) }, u = {u1, u2 }T , and M u (x, xd ) is a function with zero value at the
nodes with a unit jump of xd, i.e. M u (0, xd ) = M u (L(e) , xd ) = 0 and M u (xd+ , xd ) = M u (xd− , xd ) + 1.
In the same way, the transverse displacement reads:
w h (x, xd ) = N w (x)w + N w ′ (x)w ′ + Mθ ( x, xd ) α θ (2.62)
where:
  x  3  x 
2
 x 
3
 x  
2
N w (x) = 2  (e)  − 3  (e)  + 1, −2  (e)  + 3  (e)  
  L  L  L  L  
 (2.63)
w = {w1, w2 }
T
 x  3  x 
2
x  x 
3
 x  
2
N w ′ (x) = L(e)  (e)  − 2  (e)  + (e) ,  (e)  −  (e)  
 L  L  L L  L  
 (2.64)
w ′ = {w1′ , w2′ }
T
and Mθ (x, xd ) is a function with zero value at the nodes and a unit jump of its first derivative
( ) ( )
at xd i.e. Mθ (0, xd ) = Mθ (L(e) , xd ) = 0 and Mθ ′ xd+ , xd = Mθ ′ xd− , xd + 1.
Making the derivative of equation (2.61) in respect to x, one obtains the beam axial strain:
∂u h
ε ( x, xd ) = = Bu (x)u + Gu ( x, xd ) α u + δ xd α u (2.65)
∂x
ε ε
{ }
where Bu (x) = −1 / L(e) , 1 / L(e) , Gu ( x, xd ) = ∂M u / ∂x, and δ xd Dirac-delta, which appears
due to the discontinuous nature of axial displacement at xd. As in equation (2.19), strain is
composed of a regular part and a singular part. The beam curvature is computed as:
∂ 2w h
κ ( x, xd ) = = Bw (x)w + Bw ′ (x)w ′ + Gθ ( x, xd ) α θ + δ xd α θ (2.66)
∂x 2
κ κ
where:
 6  2x  6  2x  
Bw (x) =  − (e)2  1 − (e)  , (e)2  1 − (e)   (2.67)
 L  L  L  L 
 2  3x  2  3x  
Bw ′ (x) =  − (e)  2 − (e)  , − (e)  1 − (e)   (2.68)
 L  L  L  L 
and Gθ (x, xd ) = ∂ 2 Mθ / ∂x 2 . The curvature κ is as well divided into a regular part κ and a
singular part κ . It is clear that ε can be interpreted as a localized plastic axial strain on one
hand and in the other κ as a localized plastic curvature. In matrix notation, beam strains
can be rewritten as:
ε = ε + ε (2.69)
ε = Bd
+ Ga
a, ε = δ αd a (2.70)
ε
{ } { }
T
Where ε = {ε, κ } , ε = {ε , κ } , ε = ε , κ
T T T
, ε = ε, κ and
 Bu 0 0 
{ }
T
B= T T
 , d = u , w , w′
T
(2.71)
0 Bw Bw ′ 
G = DIAG Gu , Gθ ,{ } α = {α u , α θ } (2.72)
T
Once derivation of G operator is done, the kinematic description of the element is finalized. It
may be derived indirectly through the requirement that an element has to be able to describe
a strain-free mode at some non-zero values of α̂ u and α̂ θ , (e.g. Armero and Ehrlich 2006).
According to Figure 2.10, the generalized nodal displacements dˆ hinge = {uˆ 1, uˆ 2 , wˆ 1, wˆ 2 , wˆ 1′ , wˆ 2′ }
T
of such a strain-free mode are composed of:

T
 0 1 0 0 0 0 
dˆ hinge = dˆ rigid + Dhingeaˆ , Dhinge = (2.73)
(e )
0 0 0 L − xd 0 1
{ }
T
where dˆ rigid = uˆ 1, uˆ 1, wˆ 1, wˆ 1 + wˆ 1′L(e) , wˆ 1′ , wˆ 1′ are generalized nodal displacements due to the
rigid body motion of a complete beam, and Dhingeâ a are generalized nodal displacements
due to the rigid body motion of one part of the beam due to imposed strong discontinuity
aˆ = {αˆ u , αˆ θ } . If we now set strains defined in (2.69) to be equal to zero ford̂ hinge , one obtains:
T
0 = Bdˆ hinge + Gαˆ = Bdˆ hinge + (G + BDhinge ) aˆ (2.74)

=0
Since the above equation should hold for any â

a, the G operator needs to read:
G = − BDhinge (2.75)
and this leads to:

1
Gu ( x, xd ) = − (2.76)
L(e)
Figure 2.10 Strain-free mode of the element.

Steel structures 43
 2x   2x 
1 + 3  1 − (ed)   1 − (e) 
 L   L 
Gθ ( x, xd ) = − (e )
(2.77)
L
Equation (2.77) concludes the kinematic description of the geometrically linear element.
Interpolation functions Mu and Mθ can be developed by using (2.61) and (2.62)
with the aim of describing the strain-free mode of Figure 2.10. Defining the val-
ues in (2.61) as u1 = uˆ 1 = 0, u2 = uˆ 2 = αˆ u , uh = 0 for x < xd, and uh = α̂ u for x ≥ xd,
it follows that M u = H ( x − xd ) − N u {0, 1}. Here, H(x − xd) is a unit-step function,
which is 0 for x < xd and 1 for x ≥ xd. In the same sense for bending in (2.62) one gets:
{ }
Mθ = H ( x − xd ) ( x − xd ) − N w ⋅ 0, L(e) − xd − N w ′ ⋅ {0, 1} .
The geometrically nonlinear effects, and related global buckling, are defined by using the
von Karman axial strain when computing the virtual axial strain. The real axial strain, used
for computing the internal forces, will still be assumed as linear, as given in (2.65). The von
Karman axial strain is defined as: εVK = ∂uh / ∂x + (1 / 2)(∂w h / ∂x)2 . So, the corresponding
virtual axial strain reads:
∂δu h ∂w h ∂δw h
δεVK = + (2.78)
∂x ∂x ∂x
By interpolating δuh , w h and δw h in (2.78) as δuh = N u (x)δu + M u (x, xd )δα u ,
w h = N w (x)w + N w ′ (x)w ′ , and δw h = N w (x)δw + N w ′ (x)δw ′ , where δu = {δu1, δu2 } is a
T
vector of virtual nodal axial displacements, δw = {δw1, δw2 } and δw ′ = {δw1′ , δw2′ } are
T T
vectors of virtual nodal transverse displacements and rotations, and δα u is a virtual discon-
tinuity in axial displacement at xd, the chosen interpolations direct to:
δεVK = Bu (x)δu + Bu,w (x)δw + Bu,w ′ (x)δw ′ + Gu ( x, xd ) δα u + δ

x δα
u (2.79)
d
δεVK δε
where:
dN w u , w ′ dN w ′  dN w dN w ′ 
Bu,w (x) = C , B (x) = C ,C =  ⋅w + ⋅ w ′ (2.80)
dx dx  dx dx 
When computing virtual strains δε = {δε , δκ } , the linear matrix operator B from (2.71)
T
should be thus replaced with the nonlinear matrix operator BVK , meaning:
δε = δε VK   Bu Bu, w Bu, w ′ 

 =   δd + Gδa
a (2.81)
 δκ  0 B

w
Bw ′ 

BVK
{ }
T
The generalized virtual nodal displacements in (2.81) are denoted as δd = δuT , δwT , δw ′T
whereas virtual jumps at xd are denoted δa = {δα u , δα θ } .
T
The tangent stiffness matrix of the beam finite element with von Karman virtual axial
strain has a symmetric geometric part and non-symmetric material part. Here a non-sym-
metric tangent stiffness matrix is used. It is important to emphasize that if von Karman
definition of axial strains is used for both real and virtual strains (e.g. Reddy 2004), the
element exhibits severe locking.
2.3.2.1.2 Equilibrium Equations
The principle of virtual work, as the weak form of the equilibrium equations for an element
e of a chosen finite element mesh with Nel finite elements, can be written as:
δΠ int,(e) − δΠ ext,(e) = 0 (2.82)
A single element contribution to the virtual work of internal forces can be defined and writ-
{ }
T
ten, if one denotes the virtual strains as δε = δεVK , δκ and the virtual curvatures to be of
the same form as the real curvatures as stated in (2.66):
L(e )
∫ (δε) sdx
int,(e) T
δΠ =
0
L(e ) L(e )
(2.83)
=
∫ δd (B ) sdx + ∫ δa (G s + δ s)dx
T VK T T T
xd

0

0

standard additional
The vector of the internal beam forces that contains axial force N and bending moment M
are defined as:
s = {N , M} (2.84)
T

The vector of the internal nodal forces is acquired from the first intergral in equation (2.83)
and reads:
L(e )
∫ (B ) sdx (2.85)
VK T
int,( e )
f =
0
From the virtual work of external forces δΠ ext,(e) , the vector of the element external nodal
ext,( e )
forces f will be obtained representing the external load applied to the element. The finite
int,( e ) ext,( e )
element assembly of vectors f and f , for all elements of the chosen mesh, leads to a
set of global (i.e. mesh related) equations:
AeN=el1 f ( int,( e )
−f
ext ,( e )
) = 0 (2.86)
where A is the assembling operator.
The contribution of the second integral in equation (2.83) will be treated locally element
by element. Then, in view of (2.82), the following two equations are obtained for each ele-
ment of the chosen mesh:
L(e )
h
(e )
= h ,h { (e )
N
(e )
M } = ∫ (G s + δ s)dx =T
xd
0
(2.87)
L(e ) L(e )
=
∫ G sdx + s = ∫ G sdx + t =0, ∀e ∈[1, N ]
0
T
xd
=t 0
T
el
Steel structures 45
In (2.87) vector t = s xd = {t N , t M } is defined with components tN and tM, which represent

T
respectively the axial traction and moment (bending) traction at the discontinuity. The com-
ponent form of (2.87) can be derived by using (2.77) and (2.84), and it reads:
L(e )
hN(e) =
∫ G Ndx + t
0
u
N =0
(2.88)
L(e )
(e )
hM =
∫ G Mdx + t
0
θ
M = 0, ∀e ∈ [1, N el ]
It is evident here that there is a set of global equations (2.86) and a set of equations on an
element level (2.87) that need to be solved and interrelated. This will be addressed later in
the chapter.
2.3.2.1.3 Constitutive Relations
It is assumed that the axial response of the beam material always remains elastic. For the
bending behavior of the beam material, the following constitutive models are chosen: (i)
a stress-resultant elastoplastic constitutive model with linear isotropic hardening, (ii) a
stress-resultant rigid-plasticity model with linear softening at the softening plastic hinge.
The basic ingredients of the chosen constitutive relations are built on classical plasticity
(Ibrahimbegović, Fadi and Lotfi 1998) and are summarized in the following paragraphs.
In the same sense as in (2.70), the regular strains ε can be split into an elastic and plastic
part. As the axial strain of the beam (2.65) always remains elastic, it is clear that:
ε = ε = ε e , ε = 0 ⇔ ε p = 0, α u = 0 (2.89)
The free energy of the beam material (before localized softening is activated) becomes:
( ) ( ) ()
ψ ε e , ξ := W ε e + Ξ ξ =
1 eT e 1
2
ε C ε + Kh ξ 2 (2.90)
2
where C = DIAG {EA, EI }, E is the elastic modulus, A and I are the area and moment of
inertia of the cross-section, ξ ≥ 0 is the strain-like bending hardening variable and Kh ≥ 0 is
the linear bending hardening modulus.
The yield criterion for the beam material is defined in terms of the bending moment. The
admissible values of the bending moment and the stress-like bending hardening variable
()
q ξ are governed by the function:
φ ( M, q ) = M − ( My − q ) ≤ 0 (2.91)
where My > 0 denotes the positive yield moment of the cross-section. In order to take into
account the influence of the axial force N on the cross-section yielding, it is necessary to
define My and q as functions of N. This is done via the localization (failure) criterion that
activates softening in the discontinuity at xd and is defined in terms of the bending traction
()
tM and the stress-like softening bending variable q ξ (the latter is defined in terms of the
bending strain-like softening variable ξ)
( ) (
φ t M , q = t M − Mu − q ≤ 0 (2.92))
where Mu > My>0 denotes the positive ultimate (failure) moment of the cross-section.
Influence of the axial force N on the cross-section failure is taken into account by defining
Mu and q as functions of N, as shown below.
The bending traction tM at the discontinuity xd is related to the rotation jump as shown
in Figure 2.11.
t M = t M ( α θ ) (2.93)
Further thermodynamics of associative plasticity and the principle of maximum plastic dis-
sipation are utilized for determination of the remaining ingredients of the elastoplasticity
with hardening (e.g. Ibrahimbegović 2009; Lubliner 1990; Simo and Hughes 1998). In the
present beam model, the elastoplasticity with hardening happens for α=0, which leads to
ε = ε . By using the fact that regular strain can be decomposed into an elastic and plastic
part on one hand and the free energy of the beam, as stated in (2.90), on the other hand the
mechanical dissipation can be written as:
T
( ) ∂ψ  ∂ψ
def
0 ≤  = σT ε − ψ ε e , ξ =  σ − e  ε e + σT ε p − ξ (2.94)
 ∂ε  ∂ξ
Once again, it is assumed that the elastic process is non-dissipative (i.e. = 0), and that the
plastic state variables do not change, from (2.94) one obtains:
∂ψ
s=
∂ε e
(
= C ε e ⇒ N = EAε , M = EI κ − κ p (2.95) )
Taking into account the free energy of the beam material and mechanical dissipation, the
hardening variable q is expressed as:
∂ψ ∂Ξ
q=− =− = −Kh ξ (2.96)
∂ξ ∂ξ
With the above constitutive equations, the plastic dissipation thus reads:
p p

 = sT ε p + q ξ ⇒  = Mκ p + q ξ (2.97)
Applying the principle of maximum plastic dissipation for the variables ( M, q ) that satisfy

the yield criteria φ ( M, q ) ≤ 0 at frozen rates κ p and ξ, the constrained optimization problem
can be written:
n max
M,q γ  ( )
Lp M, q , γ = − D p ( M, q ) + γφ
( M, q ) (2.98)

Figure 2.11 Rigid-plastic cohesive law at discontinuity.

Steel structures 47
where γ ≥ 0 plays the role of the Lagrange multiplier. The evolution equations for internal
variables is formulated by using (2.92) and (2.97):
∂p ∂φ
= −κ p + γ = 0 ⇒ κ p = sign ( M ) γ
∂M ∂M
(2.99)
∂p ∂φ
= − ξ + γ = 0 ⇒ ξ = γ
∂q ∂q
along with the Kuhn–Tucker loading/unloading conditions and the consistency condition:
γ ≥ 0, φ ≤ 0, γφ = 0 (2.100)
= 0, γφ
It is necessary isolate the softening plastic hinge in order to find the remaining ingredients of
the rigid-plastic response describing softening at the discontinuity xd. The bending softening
potential at the discontinuity is expressed as Ξ = ψ = (1 / 2)Ks2 ξ , where ψ is the strain energy
function due to softening. The softening potential depends on the strain-like (bending) soft-
ening variable ξ ≥ 0 and the linear (bending) softening modulus Ks ≤ 0. The dissipation at
xd can be then written as:
∂ψ
()
def
0 ≤  = t Mα θ − ψ ξ = t Mα θ − ξ (2.101)
∂ξ
By defining:
∂ψ ∂Ξ
q=− =− = −Ks ξ = Ks ξ (2.102)
∂ξ ∂ξ
the result in (2.101) can be rewritten as:

 =  p = t Mα θ + q ξ (2.103)
The principle of maximum plastic dissipation at the rigid-plastic discontinuity can then be
defined as:
min max
tM ,q
γ  M ( ) M (
Lp t , q , γ = − D p t , q + γφ
)
t ,q
M ( ) (2.104)
and here γ ≥ 0 is the Lagrange multiplier. Taking into account the localization failure crite-
rion and plastic dissipation, the following evolution equations are obtained:
∂p ∂φ
= −α θ + γ = 0 ⇒ α θ = sign (t M ) γ
∂t M ∂t M
(2.105)
∂p ∂φ
= − ξ + γ = 0 ⇒ ξ = γ
∂q ∂q
By noting that sign(t M ) = sign(α θ ) (see (2.93) and Figure 2.11), it follows from (2.105) that:

sign(α θ )α θ = ξ ⇒ α θ = ξ (2.106)
Here, once again, the Kuhn–Tucker loading/unloading conditions and the consistency con-
dition apply:

γ ≥ 0, φ ≤ 0, γ φ = 0, γ φ = 0 (2.107)
Finally, the total dissipation of the beam finite element, when the element is in the softening
regime, can be written as:
∫ (s ε − ψ ( ε , ξ )) dx + (t ( ))
L( e )
Ltot(e) = T e
M α θ − ψ ξ
0
L(e)  
 sT ε + sT Ga − sT ε e + q ξ  dx +  t Mα θ + q ξ  (2.108)
=
∫
0



∼ 

 
εe
∫ (Mκ +qξ ) dx + ∫

L(e)
 L(e)

θ

= p
G Mdx + t  α + q
ξ M θ
0 
0 Ks ξ
p =0
It can be seen from (2.108) that enforcing vector equations (2.88) will decouple the dissipa-
tion in the softening plastic hinge from the dissipation in the rest of the beam. In order to
calculate tM, (2.88) is employed.
The plastic work of the beam cross-section in the hardening regime is defined as:

(
W p = Mκ p = M ξ = My + Kh ξ ξ (2.109) )
and plastic work for the beam finite element in the softening regime as:

(
W p = t Mα θ = t M ξ = Mu + Ks ξ ξ (2.110))
with this, all constitutive relations are defined.
2.4 COMPUTATION OF BEAM PLASTICITY

MATERIAL PARAMETERS
The material parameters that need to be known for the chosen material models are: (i) My
and Kh for the plasticity with hardening, and (ii) Mu and K s for the softening plastic hinge.
The yield moment My can be determined by considering the uniaxial yield stress of the
material σy, the bending resistance modulus of cross-section W, the cross-section area A and
the level of axial force N. One can associate the yield moment of the cross-section with the
yielding of the most-stressed material fiber to get:
 N 
My ( N ) = W σ y  1 − (2.111)
 Aσ y 
If one takes material hardening into consideration, as well as the possibility of local buck-
ling, a better estimate for Mu is obtained in contrast to the standard assumption of the
elastic–perfectly-plastic response of material fibers. In this respect, a refined finite element
model based on a geometrically and materially nonlinear shell element, which is able to
Steel structures 49
capture local buckling and gradual spreading of plasticity over the cross-section, is formu-
lated. The ultimate bending resistance Mu can be obtained by using results of such a shell
model computation, as can be moduli Kh and K s.
To obtain the desired results, a part of the frame member with a reference length Lref < Ltot
(Ltot is the total length of the frame member under consideration) is: (i) modeled with shell finite
elements, (ii) subjected to an external axial force N̂ in the first loading step, and (iii) subjected to
a varying external bending moment at the end cross-sections in the second loading step, while
keeping N̂ fixed (see Figure 2.12b). It is assumed that such a loading pattern will produce an
approximately constant internal axial force N = Nˆ during the analysis. The computation with
the shell model takes into account the geometrical and material nonlinearity which includes
plasticity with hardening and strain softening, strain softening regularization and local buckling
effects. The results of the shell analysis are illustrated in the diagrams presented in Figure 2.12d
and g. One can associate the ultimate bending moment Mu with the peak point in the diagram
in Figure 2.12d, where the applied end moment is plotted vs. the end rotation, i.e.
Mu (N) = Muref (N) (2.112)
This point can be used as well as a border-point between the hardening regime and the soft-
ening regime, where the softening can be due to material and/or geometric effects. To deter-
mine the values of the beam model hardening and softening parameters, it is assumed that
the plastic work at failure should be equal for both the beam and the shell model. Meaning
that the internal forces of the beam model have to produce the same amount of the plastic
work as the internal forces of the shell model when considering the full complete failure of
the frame element with length of Lref.
Since the plastic work is done in two regimes (hardening and softening), it is necessary to
assure that the amount of plastic work in each regime matches for both models, i.e.
p p , ref p p , ref
EW (N) = EW (N), EW (N) = EW (N) (2.113)
p,ref p,ref
The plastic work in both regimes EW (hardening) and EW (softening) are obtained
from shell model analysis, Figure 2.12g.
p
The plastic work of the beam model in the hardening regime, EW , can be determined by
observing that each cross-section in the frame member of length Lref is approximately under the
same force–moment state during the hardening regime. Integrating equation (2.109), one obtains:
t at Mu
Lref
 1 
E Wp
=
∫
0 ∫
0
W pd τdl =Lref  My ξ + Kh ξ 2  (2.114)
 2 

In (2.114) above, ξ is the value of hardening variable that corresponds to the bending
moment Mu, Figure 2.12e. As linear hardening in the beam model is assumed (2.96), the
hardening variable reads:
Mu − My
ξ= (2.115)
Kh
Hardening modulus is easily obtained by employing the three last equations:
K (N) =
h
(M (N) − M (N)) L 2
u
2
y
ref
(2.116)
p , ref
2EW (N)
p
On the other hand, the plastic work of the beam model in the softening regime, EW , can be
determined by assuming that the softening part of the M − ϕ curve in Figure 2.12d, obtained
from the shell model analysis, is produced by a very localized phenomenon (in a single cross-
section) related to the local buckling and/or to the localized strain softening. Integrating
(2.110), one can compute the plastic work in the softening regime for the beam model as:
t ξ at t M = 0
E Wp
∫ p
= W dτ =
0
∫0
( M + K ξ )d ξ = K ξ (2.117)
u s
1
2

s
2

Here ξ is the value of the softening variable that corresponds to the total cross-section fail-
ure, Figure 2.12f. As linear softening is assumed in the beam model, one obtains:
M
ξ = u (2.118)
Ks
Figure 2.12 Evaluation of beam material parameters by using results of a refined analysis (a) beam model,
(b) refined model, (c) moment-curvature/rotation (beam model), (d) moment-rotation (refined
model), (e) moment-hardening variable (beam mode), (f) moment-softening variable (beam
model) and (g) plastic work (refined model).
Steel structures 51
Taking into account the last two expressions, as well as the fact that the amount of plastic
work needs to be the same in both models, the expression for softening modulus reads:
Mu2 (N)
Ks (N) = p , ref , Ks ≤ 0 (2.119)
2EW (N)
It is realized that the influence of the following values Lref, Kh and K s is very small; however,
one should keep in mind that if the chosen length of Lref enables large deflections then a large
displacement correction of Mu has to be done.
A procedure for solving the set of global (mesh related) and the set of local (element
related) nonlinear equations, generated by using the stress-resultant plasticity beam finite
element with embedded discontinuity, can be found in Ibrahimbegović (2009).
2.4.1 Numerical Examples of Frames and Frame Elements

The beam model computer code was generated by using the symbolic manipulation code
AceGen, and the examples were computed by using the finite element program AceFem (e.g.
Korelc n.d.).
2.4.1.1 Example 1 — Computation of Beam Plasticity Material Parameters

As seen in the theoretical background, first of all, it is necessary to compute the beam plas-
ticity material parameters Mu, Kh and K s. A frame member is considered with an I-cross-
section with flange width bf = 30 cm, flange thickness tf = 1.5 cm, web height bw = 40 cm and
web thickness tw = 0.8 cm. The cross-section area is A = 122 cm 2 and the bending resistance
modulus is I = 43034.2 cm4. A part of the frame member of length Lref = 2L = 300 cm is mod-
eled and it should be sufficient to capture the local softening effects due to local buckling
and/or strain softening. Material properties of the frame member are presented in Table 2.2,
and the uniaxial response is plotted in Figure 2.13.
The example has been computed with the finite element code ABAQUS (Hibbit, Karlsson
and Sorensen n.d.) by using the shell finite element S4R with five integration points
throughout the thickness. As only one half of the considered geometry was discretized (see
Figures 2.14 and 2.15), the symmetry conditions uz = ϕ x = ϕ y = 0 were employed. The mesh
consists of equal squared elements. The free-end cross-section of the model was made rigid
by coupling the degrees of freedom of that particular cross-section.
According to Ibrahimbegović (2009) the linear softening modulus is computed as:
leσ 2u
Ksle = − (2.120)
2 gs
in order to minimize the mesh-dependency effect.
Table 2.2 Material Properties

of the Frame Member
E 210 GPa
σy 0.24 GPa
σu 0.36 GPa
εy = σy / E 200
εu 0.1
εf 0.12778
Figure 2.13 Uniaxial stress–strain curve.
Figure 2.14 Boundary conditions for the shell model analysis.
Figure 2.15 Failure mode of the representative part of the frame member as computed by the shell model.
Steel structures 53
where le = 5 cm is a characteristic dimension of the element and gs = 0.5kN/cm 2 is the plastic
work density (plastic work per unit volume) in the softening regime of the shell material
that corresponds to the gray-colored area in Figure 2.13. The strain at failure, adjusted to
the mesh, is then:
σu
ε les = ε u + = 0.10727 (2.121)
EKsle
E + Ksle
The load was applied in two steps. For the first step, a desired level of axial force N (N=−0.3Ny
to N = 0.3Ny, Ny = Aσy) at the mid-point MP of the rigid cross-section was applied and then
the application of the bending moment M at the point MP took place. Nonlinear analysis
with the path-following method was performed. The results of the analyses are presented in
Figure 2.15 to Figure 2.17. The final deformed configuration of the shell model and distri-
bution of the equivalent plastic strain are presented in Figure 2.15 for a pure bending case.
It is noted that the considered part of the frame member failed by localized buckling of the
bottom flange and strong localized yielding of the flange is observed which is concentrated
in the neighborhood of the web. All other cases had the same failure mode. The level of axial
force has a significant influence on the peak bending resistance and on the overall response
(see Figure 2.16).
The value of the axial force does not influence the shape of the curve to a great extent
(see Figure 2.17). The relation between the rotation and plastic work is almost linear. In
Figures 2.16 and 2.17, it is assumed that points with the maximum bending moment repre-
sent a transition point from a hardening to softening regime.
The results obtained using the shell model are now used for evaluation of the beam model
material parameters. Summarized results of the shell model are given in Table 2.3.
By using Table 2.3, a bi-linear approximation function for Muref is formulated as:
uref (N) = Mu

M
(
 ref ,0 1.03 + 0.85 N
Ny ) if N < − 0.035N y
(2.122)
 Muref ,0 if N ≥ −0.035N y
uref (N) can be used to evaluate the ultimate

where Muref ,0 = Muref (N = 0). It is assumed that M
bending moment of the beam model Mu, i.e.
uref (N) (2.123)

Mu (N) = M
see Figure 2.18. The values for the ultimate resistance obtained with the shell analyses are
marked with dots in Figure 2.18.
The beam model hardening modulus Kh can be evaluated pointwise by using (2.116),
(2.111), (2.123) and the third column of Table 2.3. Average value of Kh(N)=1.06×107kN/cm 2
assuming that the axial force has no influence on the hardening modulus.
In the same fashion, the beam model softening modulus Ks can be evaluated pointwise
by using (2.119) and (2.123) and the last column of Table 2.3. The gray-colored fields in
Table 2.3 present unreliable results since for those cases the shell analysis did not converge.
Again, assuming that the axial force has no influence on softening modulus the average
value K s = −3.28×105kN/cm 2 can be adopted.
Table 2.4 gives a comparison between the shell analysis results and the corresponding
beam model results. It is noted that the error in the ultimate bending moment is small, while
the error in dissipated plastic work can be quite large.
Figure 2.16 Bending moment vs. rotation curves for the end cross-section.
Figure 2.17 Plastic work vs. end cross-section rotation curves.
Table 2.3 Summary of Results of the Shell

Model Analyses
Muref (kNm)
p,ref p,ref
N/Ny EW (kJ) EW (kJ)
0 550 86 53
−0.1 521 73 45
−0.2 480 46 29
−0.3 427 43 27
0.1 560 72 32
0.2 579 149 0
0.3 571 142 0
Steel structures 55
Figure 2.18 Approximation of the ultimate bending moment of the cross-section.
Table 2.4 Comparison Between Approximations and Shell Analyses Results

p p ,ref p p ,ref
Mu − Muref EW − EW EW − EW
(%) E W (kJ) (%) E W (kJ) (%)
p p
N/Ny Mu(kNm)
Muref
p ,ref p ,ref
EW EW
0 550 0.00 50 42 46 12
−0.1 519 0.26 59 20 41 8
−0.2 473 1.54 54 17 34 19
−0.3 426 0.22 48 12 28 ?
0.1 550 1.94 81 14 46 ?
0.2 550 5.07 109 27 16 ?
0.3 550 3.72 134 6 46 ?
2.4.1.2 Example 2 — Push-Over of an Asymmetric Frame

An asymmetric frame is presented in Figure 2.19, where LB1 = 600 cm, LB2 = 500 cm, LB3 = 400 cm,
HC = 250 cm. All the other geometrical materials are the same as in the previous example.
The vertical load is constant and equals qυ = 0.05kN/cm. The lateral loading is presented in
Figure 2.19, where F0 = 1kN, is a concentrated force and λ is the load multiplier. The mesh
consists of eight beam finite elements per each frame member.
The results are presented in Figures 2.20 and 2.21. The results of the geometrically lin-
ear and geometrically nonlinear analyses are nearly the same before ultimate resistance is
reached at 1581.8 kN for the geometrically linear case and at 1578.4 kN for the geometri-
cally nonlinear case. After that point, the difference between the two analyses is bigger.
The final computed equilibrium configuration for the geometrically nonlinear case is at
utop = 26.52 cm. In the next load step, one additional softening plastic hinge is activated,
which results in the global failure mechanism. Since the path-following algorithm used is
only governed by the increase of utop, it is not possible to capture the remaining part of the
load–displacement curve.
In the right part of Figure 2.20, we present the dissipated energy vs. utop curves. Namely,
first, there is the elastic non-dissipative phase, followed by the pure hardening dissipation
phase, followed by the combined hardening and softening dissipation phase and finally
the pure softening dissipation phase. In the geometrically nonlinear case, there is no final
pure softening dissipation phase due to the activation of the global failure mechanism.
Figure 2.21 illustrates the deformed configuration of the frame at utop = 26.52 cm. Locations,
Figure 2.19 Asymmetric frame: geometry and loading.
Figure 2.20 Load vs. displacement and dissipated energy vs. displacement curves.
Figure 2.21 Deformed shape and locations of softening plastic hinges.

Steel structures 57
where softening plastic hinges were activated at utop ≈ 26.52 cm, are presented in the middle
part of Figure 2.21 for the geometrically linear case and in the right part of the same figure
for the geometrically nonlinear case.
2.4.1.3 Example 3 — The Darvall–Mendis Frame

The clamped portal frame under vertical loading first studied by Darvall and Priyantha
(1985) and later examined by Armero and Ehrlich (2006) and Wackerfuß (2008) was elabo-
rated. The geometry of the frame is presented in Figure 2.22 (Table 2.5).
The inelastic response is defined by the ultimate bending moment M u,C = 158.18kNm for
the columns, the ultimate bending moment M u,B = 169.48 kNm for the beam and the soft-
ening modulus K s = aEI/10L, where the values a = 0, −0.04, −0.06, −0.0718 are considered.
Note, in this example, the inelastic response does not include any material hardening. It
is assumed that the axial force has no influence on the ultimate bending resistance of the
frame members. The mesh consists of eight geometrically linear beam finite elements (see
Figure 2.22). The frame is loaded with a vertical load λF0(F0 = 1kN) applied at node 5 (see
Figure 2.22). In the numerical simulations, the load multiplier λ and the vertical displace-
ment uυ is controlled at node 5 using the path-following method.
The vertical load vs. vertical deflection curves are presented in Figure 2.23, where the
points on the curves mark configurations where the softening plastic hinge was activated in
one of the elements of the mesh. In all cases, hinges form at the same locations.
Figure 2.22 Geometry and loading of the portal frame.
Table 2.5 Geometric and Material

Properties of the Frame
Lc 3.048 m
A 0.103 m2
I 0.001 m4
E 2.068×107 kN/m2
Figure 2.23 Vertical load vs. vertical deflection curves.
The first hinge forms in elements 4 and 5 at node 5, the second hinge forms in element 7
at node 7 and the third hinge forms in element 2 at node 3. This example shows the signifi-
cant influence of the softening modulus on the ultimate load of the structure. The ultimate
load in the case of a perfectly plastic response of the hinges đa a = 0 is 434 kN when the
third hinge forms. In cases a = −0.04 and a = −0.06, the structure fails when the second hinge
forms when the vertical load reaches 383 kN and 350 kN, respectively. Finally, in the case of
a = −0.0718, the structure fails when the first hinge forms at 336 kN. In Table 2.6, results are
compared with the results obtained by Darvall and Priyantha (1985), Armero and Ehrlich
(2006) and Wackerfuß (2008).
2.4.1.4 Example 4 — Push-Over Analysis of a Symmetric Frame

Additionally, the symmetric steel frame was exposed to a push-over. The geometrical and
material properties of the frame are presented in Table 2.7, while the geometry and the load-
ing of the frame is presented in Figure 2.24.
It should be noted that in order to have strong columns and a weak beak system the
cross-section properties of the columns are 10% stronger than the cross-section properties
of the beams. The elements which connect the beams to the columns are 10% weaker than
the cross-section properties of beams; these elements are chosen to simulate the behavior of
connections in the global analysis of the steel frame structure. All beams are loaded with the
Table 2.6 Summary of Results of the Shell Model Analyses
Darvall and Mendis Armero and Ehrlich Wackerfu Present
a Hinge uυ (cm) λF0 (kN) uυ (cm) λF0 (kN) uυ (cm) λF0 (kN) uυ (cm) λF0 (kN)
0 1 0.50 336 0.50 337 0.53 342 0.50 336
2 1.14 427 1.14 428 1.13 435 1.13 427
3 1.34 433 1.34 434 1.33 440 1.34 434
−0.04 1 0.50 336 0.50 337 0.53 349 0.50 336
2 1.14 387 1.18 388 1.16 401 1.19 383
−0.06 1 0.50 336 0.50 337 0.52 348 0.50 336
2 1.19 336 1.29 337 1.23 349 1.23 350
−0.0718 1 0.50 336 0.50 337 0.50 336
Steel structures 59
Table 2.7 Geometric and Material

Properties of the Frame
Geometric and Material Properties of the Frame
A 28.50 cm2
I 1940 cm4
E 2.10×104 kN/m2
K 0.5E
Yielding and Ultimate Values

My 3100 kNcm
Mu 3100 kNcm
Vy 355 kN
Vu 400 kN
Fracture Energies
Gf,M 550 kN/cm
Gf,V 450 kN/cm
Figure 2.24 Geometrical characteristics of the frame and loading scheme.
same vertical constant uniformly distributed load simulating the dead load, while the lateral
loading is applied in terms of an imposed incremental displacement (utop) at the upper corner
(point A, see Figure 2.24) (Imamovic, Ibrahimbegović and Mesic 2017).
The final position of the formed plastic hinges is presented in Figure 2.25 together with
the obtained softening response presented in the well-known form (force–displacement dia-
gram). The force which is taken as the adequate measure, in this case, is the reaction in the
corner where the displacement is imposed (point A).
Ultimate limit analysis of any frame planar steel structure can be conducted with the
application of this geometrically exact beam model with the inclusion of the second-order
effects as well as different failure mechanisms. The geometrically exact beam model can
perform the corresponding nonlinear analysis that allows the ultimate limit load compu-
tations with large displacement without any need for correction of the equivalent beam
Figure 2.25 Locations of softening plastic hinges and load vs. displacement.
property. (Dujc, Bostjan and Ibrahimbegović 2010). This advantage is very important in a
steel frame structure because of the large ductility of steel.
2.4.1.5 Example 5 — Cyclic Loading of a Steel Structure

A steel structure presented in Figure 2.26 was exposed to cyclic loading. During the experi-
ment, it was observed that the large deformations of the connection components under cyclic
loading cause a less stiff response of the experimental structure. This can be explained by
the formation of large deformations on the welded plate located in the tension zone, which
can cause a partial loss of contact between the plate and the horizontal beam. (Figure 2.27a).
As the direction of the applied load changes, the compression and tension zones are changes.
In the compression zone the full contact between elements is lost, and as a consequence there
is a reduction of the initial stiffness of the connection. The reduction of stiffness in the con-
nection is caused by the partially lost contact in the compression zone. (Figure 2.27b). Once
full contact is provided the stiffness reaches its full value.
In order to model this phenomenon, the geometrically exact beam model (Imamovic,
Ibrahimbegović and Mesic 2017) with hardening/softening was used with certain modi-
fications with respect to the response under a cyclic load. The plasticity part governs the
Figure 2.26 Experimental setup and tests.

Steel structures 61
Figure 2.27 Experimental setup and tests.
Figure 2.28 Constitutive model.
Figure 2.29 Computed vs. measured response of the experimental structure.
hardening and unloading phases, whereas the damaged part provides the reduced stiffness
of the connection after the change in the sign of the bending moment: from positive to nega-
tive or vice versa (Figure 2.28). The damage model governs the connection response until
full contact between the plate and horizontal beam is reached. After the full contact has
been reached, the plasticity model is again activated. The gap δ corresponds to the plastic
deformation in the bolts (Imamovic, Ibrahimbegović i Mesic 2018). The modification of the
beam model requires the splitting of the internal variables into two groups, depending on
the sign of bending moment.
A comparison of the response of the structure exposed to cyclic loading obtained by
the experiment and calculated are given in Figure 2.29. One can notice that there is a dif-
ference between the responses; however, the proposed beam model significantly improved
the response prediction compared to the model of the plasticity or the damage (Imamovic,
Ibrahimbegović and Mesic 2018).
2.5 CONCLUSION
The presented multi-scale model for computing the load limit of planar metal frames under
push-over and full collapse analysis combines the best of two worlds: on one side, the effec-
tiveness and robustness of the macroscale beam model for the entire structure, and, on
another side, a refined representation of the localized instability effects (both geometric and
material) through mesoscale effects based upon the geometrically nonlinear elastoplastic
shell formulation. The latter is captured and stored within the macroscale beam model in
the manner which is compatible with enhanced beam kinematics with embedded disconti-
nuity. The most appropriate choice of the mesoscale shell model can be further guided by
the error-controlled adaptive finite element method for shell structures (by using a model
error estimation, e.g. Bohinc, Ibrahimbegović and Brank (2009)), which could automati-
cally find the most appropriate model for representing the particular local phenomenon
under consideration.
The multi-scale procedure proposed in this chapter belongs to the class of weak cou-
pling methods, where we carry out sequential computations. The results of the shell model
computations, accounting for material and geometric localized instability, are stored to be
used within the beam model softening response. As presented by numerical simulations, the
performance of the proposed multi-scale computational approach is very satisfying. One
of its main features is that detection and development of the softening plastic hinges in the
frame is fully automatic, and spreads gradually in accordance with stress redistribution in
the course of the nonlinear analysis. This is in contrast with many standard computational
approaches to load limit, under the push-over and the full collapse analysis of frames, which
rely on predefined locations of plastic hinges and the corresponding inelastic deformations.
REFERENCES
Armero, F. and D. Ehrlich. 2006. “Numerical Modeling of Softening Hinges in Thin Euler–Bernoulli
Beams”. Computers & Structures 84 (10–11): 641–56. doi:10.1016/j.compstruc.2005.11.010.
Bathe, K. J. 1996. Finite Element Procedures. Upper Saddle River, NJ: Prentice Hall.
Bohinc, U., A. Ibrahimbegović and B. Brank. 2009. “Model Adaptivity for Finite Element Analysis
of Thin and Thick Plates Based on Equilibrated Boundary Stress Resultants”. Engineering
Computations 26 (1/2): 69–99.
Brank, Boštjan. 2005. “Nonlinear Shell Models with Seven Kinematic Parameters”. Computer
Methods in Applied Mechanics and Engineering 194 (21–24): 2336–62. doi:10.1016/j.
cma.2004.07.036.
Brank, Boštjan, Djordje Perić and Frano B. B. Damjan. 1970. “On Large Deformations of Thin
Elasto‐Plastic Shells: Implementation of a Finite Rotation Model for Quadrilateral Shell
Element”. International Journal for Numerical Methods in Engineering 40 (4): 689–726.
doi:10.1002/(sici)1097-0207(19970228)40:4<689::aid-nme85>3.3.co;2-z.
Steel structures 63
Darvall, Peter LePoer and Anuruddha Mend Priyantha. 1985. “Elastic‐Plastic‐Softening

Analysis of Plane Frames”. Journal of Structural Engineering 111 (4): 871–88. doi:10.1061/
(asce)0733-9445(1985)111:4(871).
Dujc, J., B. Bostjan and A. Ibrahimbegovic. 2010. “A Multi-Scale Computational Model for Failure
Analysis of Metal Frames That Includes Softening and Local Buckling”. Computational
Methods in Applied Mechanics and Engineering 1371–1385.
Hibbit, Karlsson and Sorensen. n.d. “Abaqus Manuals”.
Hill, R. 1962. “Acceleration Waves in Solids”. Journal of the Mechanics and Physics of Solids 10 (1).
doi:10.1016/0022-5096(62)90024-8.
Hughes, T. J. R. 1987. The Finite Element Methods. Englewood-Cliffs, NJ: Prentice-Hall.
Ibrahimbegović, Adnan. 2009. Nonlinear Solid Mechanics: Theoretical Formulations and Finite
Element Solution Methods. Springer.
Ibrahimbegović, Adnan, Gharzeddine Fadi and Chorfi Lotfi. 1998. “Classical Plasticity and
Viscoplasticity Models Reformulated: Theoretical Basis and Numerical Implementation”.
International Journal for Numerical Methods in Engineering 42 (8): 1499–535. doi:10.1002/
(sici)1097-0207(19980830)42:8<1499::aid-nme443>3.3.co;2-o.
Imamovic, Ismar, Adnan Ibrahimbegović and Esad Mesic. 2018. “Coupled Testing-Modeling
Approach to Ultimate State Computation of Steel Structure with Connections for Statics and
Dynamics”. Couples System Mechanics 7 (5): 555–81.
Imamovic, Ismar, Adnan Ibrahimbegović and Esad Mesic. 2017. “Nonlinear Kinematics Reissner’s
Beam with Combined”. Computer and Structures 12–20.
Korelc, J. n.d. “Fakulteta za gradbeništvo in geodezijo, AceGen, AceFem”. http://www.fgg.uni-lj.si/
Symech.
Lubliner, Jacob. 1990. Plasticity Theory. New York: Macmillan.
Luenberger, David G. 1984. Linear and Non-Linear Programming. 2nd ed. Reading, MA:
Addison-Wesley.
Mandel, J. 1996. “Conditions de Stabilite´ et postulat de drucker”. Rheology and Soil Mechanics. In:
Kravtchenko J. and Sirieys P. M. (eds). Grenoble 1964: IUTAM Symposium.
Maugin, Gerard A. 1992. The Thermomechanics of Plasticity and Fracture. Cambridge: Cambridge
university press.
Nagtegaal, J. C., D. M. Parks and J. R. Rice. 1974. “On Numerically Accurate Finite Element
Solutions in the Fully Plastic Range”. Computer Methods in Applied Mechanic and Engineering
4: 153–78.
Reddy, J. N. 2004. An Introduction to Nonlinear Finite Element Analysis. Oxford: Oxford University
Press.
Rice, J. R. 1976. “The Localization of Plastic Deformation. In: Theory and Applied Mechanics”. 14th
IUTAM Congress. Delft, North-Holland, Amsterdam, 207–20.
Simo, J. C. and T. J. R. Hughes. 1998. Computational Inelasticity. Springer.
Stakgold, Ivar. 1998. Green’s Functions and Boundary Value Problems. New York: John Wiley &
Sons.
Strang, G. 1986. Introduction to Applied Mathematics. Cambridge: Wellesley- Cambridge Press.
Truesdell, C. and W. Noll. 1965. The Non-Linear Field Theories. Berlin: Springer.
Wackerfuß, Jens. 2008. “Efficient Finite Element Formulation for the Analysis of Localized Failure in
Beam Structures”. International Journal for Numerical Methods in Engineering 73 (9): 1217–50.
doi:10.1002/nme.2116.
Zienkiewicz, O. C. and R. L. Taylor. 2000. The Finite Element Method. 5th ed. Oxford:
Butterworth-Heinemann.
Chapter 3
Reinforced concrete structures
The modeling of reinforced concrete (RC) structures in order to ensure the optional design
with respect to structural integrity has a long tradition, which has been summarized in sev-
eral reference works (e.g. Chen 1982; Hofstetter and Mang 1995). The finite element models
are used extensively in this respect (e.g. Stein and Tihomirov 1999; Bažant 1986) where
the equivalent properties of the RC composite are obtained by smearing the reinforcement
within the concrete layers to provide a corresponding match of the total resistance. This
can be acceptable with respect to the main design goal pertaining to ensuring the structural
integrity, but not taking into consideration the accuracy of the behavior of each component
of the RC composite. It is well known that the most efficient approach of this kind can be
developed for RC structures in terms of stress resultants, i.e. moments, shear and membrane
forces (e.g. Ibrahimbegović and Frey 1993a,b; Armero 1997). It is clear that any such stress-
resultant RC model cannot provide reliable information on the detailed sequence of events
pertaining to different failure mechanism creations, but only the corresponding bounds
with respect to the structure or structural component resistance and integrity.
However, the current engineering design practice of RC structures no longer only requires
proof of structural integrity, but also structural durability, either under long term, exploi-
tation loading (such as aging problems) or under short term, extreme loading (such as fire
exposure or aggressive agent exposure). For this, detailed information on concrete cracking,
with the results on crack-spacing and opening or the bond-slip, is needed and cannot be
provided by the smeared RC models. In this respect, a more detailed modeling of each of the
RC composite ingredients, capable of describing the cracking of concrete, the bond-slip as
well as the plastic yielding of steel is required.
One such model is proposed herein, combining the concrete finite element model capable
of representing concrete cracking, both in terms of micro-cracks in the fracture process zone
(FPZ) and macro-cracks (e.g. Brancherie and Ibrahimbegović 2009), with the elastoplas-
tic model for the yielding and localized failure of reinforcement (e.g. Ibrahimbegović and
Brancherie 2003; Ibrahimbegović 2009) coupled together with a simple finite element model
for bond-slip based upon Drucker–Prager plasticity.
The goal is to compute crack-spacing and opening in a case when the external load is
applied upon the concrete, and where only the cracking of the concrete will lead toward
bond-slip activation. It is the crack-spacing and opening that pertains to the structural
durability.
For bond-slip, a simple Drucker–Prager plasticity model is proposed as it has a reduced
number of parameters with a clear physical meaning in respect to the essential features
of bond-slip within the context of the RC composite. The simple bond-slip model, simi-
lar to the non-associative perfect plasticity, sensitive to confining pressure, is shown to be
quite sufficient for the problems of interest where the bond-slip is not excessive. The failure
65
mechanism with pull-out of the reinforcement is not considered herein, which allows for
limitation of the analysis to the linear kinematics framework.
The new approach proposed herein combines all the ingredients of the RC composite
into a single finite element model, which has practically the same outside appearance as the
standard finite element model in terms of the macro finite element, but provides an inside
detailed description of the strain field for each ingredient, concrete, steel or bond-slip. The
latter is achieved with the corresponding enhancements of the macro element strain field,
where the incompatible mode method (e.g. Ibrahimbegović and Wilson 1991) is used to han-
dle the macro-crack appearance in the concrete, and the X-FEM method (e.g. Belytschko
et al. 2001; Belytschko et al. 2002; Belytschko et al. 2002; Belytschko and Black 1999) is
used for representing the bond-slip along a particular reinforcement bar.
The suitable form of the operator split procedure is then constructed in order to separate
the main update procedure for the concrete displacement, which also implies the corre-
sponding steel displacement with the frozen value of bond-slip, from the new slip computa-
tion. The latter is not equivalent to the local (Gauss quadrature point) plastic deformation,
since it is computed from the corresponding slip equilibrium equations assembled from all
the bond-slip contributions along the particular reinforcement bar. The key ingredient that
allows for this kind of operator split solution procedure, separating the bond-slip compu-
tation from the RC displacement (with the value of slip kept fixed), pertains to describing
the bond-slip with relative degrees of freedom. This is in sharp contrast with the stan-
dard parametrization of interface discontinuities (e.g. Belytschko et al. 2001; Jirásek 2002),
which (as noted by Ibrahimbegović and Melnyk (2007)) would not allow for the use of the
same operator split procedure.
3.1 CONCRETE MODELING WITH A DAMAGE MODEL
The concrete model has been developed in view of failure models for massive structures
(e.g. Ibrahimbegović and Brancherie 2003) where the elastic behavior is followed by the
creation of the FPZ with a large number of micro-cracks and subsequent final failure mode
in terms of the macro-cracks. The FPZ is represented by the isotropic continuum damage
model since the number of micro-cracks is considered sufficiently large and their orientation
random. The macro-crack is represented by the displacement discontinuity across the FPZ,
as done for a number of damage models for localized failure (e.g. Armero 1997; Jirásek and
Zimmermann 2001; Oliver 2000; Oliver, Huespe and Sánchez 2006). The original feature
of this concrete cracking model, with respect to those proposed previously, is its ability to
combine two failure modes: FPZ and macro-cracks. Therefore, the proposed model not only
accounts for damage-induced anisotropy, but also provides the most appropriate representa-
tion of failure mechanisms for massive structures, and allows for the reproduction of the
proper sequence of failure modes creation and the induced stress redistribution leading to
a better prediction of macro-cracks orientation (e.g. Ibrahimbegović and Brancherie 2003;
Kucerova et al. 2009). The concrete failure model of this kind is built from three main ingre-
dients: a non-homogeneous representation of the strain field, the strain energy in a quadratic
form and, finally, a corresponding damage criterion indicating the creation of the FPZ and
the creation of the failure-inducing macro-cracks.
(i) A non-homogeneous representation of the strain field is proposed in order to capture

the strain localization at the onset of macro-cracks creation, leading to the softening
phenomena. The total strain in the concrete ε c (x, t) can be written as the sum of the
Reinforced concrete structures 67
ac + δ Γ ac corresponding to the displace-

smooth part ε c (x, t) and the singular part Ga
ment discontinuity contribution of the macro-cracks described by the surface Γ.
c
ε
c c
ε (x, t) =
ε (x, t) + G ( )
(x)ac (t) + δ Γ (x) ac (t) ⊗ n s (3.1)
∇s u
where Γ denotes the surface of discontinuity of the displacement field, δΓ is the Dirac-
c (x) corresponds to
Delta function on Γ and a denotes the displacement jump on Γ. G
the influence function and is chosen in order to limit the macro-cracks influence on
the FPZ; in the finite element setting, the representation in terms of the enhanced con-
stant strain triangle (CST) element is chosen (e.g. Zienkiewics and Taylor 2005) with
embedded discontinuity (see Figure 3.1).
Such a choice for the embedded discontinuity permits the formulation of the cor-
responding approximation for G (x) by using:
=−
G ∑ ∇ N (3.2)
i ∈Ωe +
s
i
The virtual strain field γ c (x, t) is typically constructed in a slightly modified form
(Ibrahimbegović and Brancherie 2003; Ibrahimbegović and Wilson 1991), which
guarantees that the patch test condition for convergence (e.g. Zienkiewics and Taylor
2005; Ibrahimbegović 2009) is satisfied:
γ c (x,t) = γ c (x,t) + G (Γ )
ˆ ( x)βc (t) + δ e (x) βc (t) ⊗ n s (3.3)
where G ˆ (x) is the modified version of the function G

(x) and βc denotes the virtual
displacement discontinuity.
(ii) The strain energy density of the concrete model ψ c (x, t) is constructed by taking into
account both the smooth field and the discontinuity contributions:
1 1 1
ψ c (x, t) = s ⋅ ε c − s ⋅ Ds + K ζ 2 + K ζ 2 (3.4)
2 2 2
where s is the Cauchy stress tensor, D is the damage compliance tensor, K and K
are, respectively, the hardening and softening moduli for the FPZ and macro-crack-
induced discontinuity, whereas ζ and ζ are the corresponding internal variables gov-
erning the evolution of these two mechanisms.
Figure 3.1 Macro-cracks with the fracture process zone and its finite element representation.
(iii) The damage criteria indicating the creation of the FPZ and the macro-crack are writ-
ten, respectively, as:
1
(σ f − q ) ≤ 0
c
φ c ( s, q ) = s ⋅ D es −
E
( ) ( )
φn c s, q = t Γ ⋅ n − σ f − q ≤ 0 (3.5)
 σ 
( )
φs c s, q = t Γ ⋅ m −  σ s − s
 σf
q ≤ 0

where:
D e = C −1 is the elastic damage compliance tensor (equal to the inverse of the
elasticity tensor),
E is Young’s modulus,
t Γ = sn is the driving traction at the discontinuity,
σ f is the elasticity limit indicating the creation of the FPZ,
σ f and σ s are the threshold values indicating the beginning of the localized fail-
ure in traction (mode I) and in shear (mode II); whereas, q and q are
the stress-like variables governing the evolution of the stress threshold.
The remaining governing equations can be obtained using the standard thermodynam-
ics arguments based on the second principle defining dissipation in concrete  c (e.g.
Ibrahimbegović 2009) according to:

0 ≤  c = σ ⋅ ε c −
d c
dt
( ) 1 σ + q ζ + q ζδ
ψ = σ ⋅ Dσ − ε c + σ ⋅ D
2
Γ (3.6)
where q = −K ζ and q = −K ζ. The last equation applies to the linear softening law, which
is often replaced by exponential softening (Ibrahimbegović and Brancherie 2003) in order
to ensure the robustness of the computations. However, either case of softening law can
be characterized by physically sound parameters, such as the total fracture energy Gf
(Ibrahimbegović 2009). For the elastic process, as seen in (3.6), where internal variables
remain frozen with D = 0, ζ = 0, ζ = 0 :
ε c = Dσ (3.7)
From (3.1), it can be worked out that:
D = D + Dδ Γ
α c = Ds (3.8)
ε c := ε c + G
α c = Dt Γ
For the inelastic process it is assumed that (3.8) is still valid, which means that the charac-
teristic evolution time for internal variables is much shorter than the one for other state vari-
ables (e.g. Ibrahimbegović 2009). The total damage dissipation for concrete can be defined
from (3.6):
1 1 
0 ≤ c = σ ⋅ Dσ + q ζ +  t Γ ⋅ D t Γ + q ζ  δ Γ (3.9)
2
 2 
c

 c
Applying the principle of maximum dissipation, the corresponding evolution equations

for damage variables for concrete cracking can be obtained according to:
 ∂φc
 Ds s = γ
 ∂s
 ∂φc
max c ⇒  ζ = γ (3.10)
c
φ =0
 ∂q
 γ ≥ 0, φ c ≤ 0, γφ
c =0


and
 ∂φc
Dt Γ =
 a = m, n
∑γ a a
∂t Γ
maxc c ⇒  (3.11)
 ζ = γ ∂ φa
c
∑
c
φn = 0, φm = 0
 a
∂q
 a = m, n
In the finite element implementation based on the enhanced CST element (Figure 3.1), the
internal variables for the FPZ are computed with a single quadrature point. A single value
per element is also used for each discontinuity inducing macro-crack in both mode I and
mode II. For a typical time step ∆t = tn +1 − tn . The computation is started by assuming the
trial elastic step:
γ nc,+trial trial trial

1 = 0 ⇒ Dn +1 = Dn , ζn +1 = ζn
−1 c ,(i )
s ntrial n +1, qntrial
+1 = Dn ε +1 = − K ζn = qn (3.12)
1
φnc,+trial
1 = s ntrial e trial
+1 ⋅ D s n +1 −
E
(σ f − qn )
If γ nc,trial
+1 ≤ 0 , the step is indeed elastic and no change of the FPZ will result. However, if
γ nc,trial c
+1 ≥ 0 , a correct (positive) value has to be obtained γ n +1 as:
φnc,+trial
γ nc +1 = 1
2 (3.13)
∂ φnc,+trial ∂ φ c, trial  ∂ φ c,trial 
1
⋅ Dn−+11 n +1 + K  n +1 
∂s ∂s  ∂q 
and the corresponding change of smooth damage variables according to:
∂ φn +1
Dn +1s n +1 = Dns n +1 + γ nc +1
∂s n + 1
(3.14)
c ∂ φn +1
ζn +1 = ζn + γ n +1
∂qn +1
The final stress from the obtained results can be then obtained as:
scn +1 = Dn−+11ε cn,(+i1) (3.15)
It is from (3.15) that one can test for the eventual evolution of the macro-cracks. In this
sense, one needs to first compute the trial value of traction.
t ntrial
+1 = − G
∫
ˆ T scn +1dV (3.16)
Ωe
and the corresponding trial values of the discontinuity damage criteria by assuming that no
change in the macro-crack would occur:
( )
φn c, trial s, q = t ntrial (
+1 ⋅ n − σ f − qn (3.17a) )
 σs 
( )
φn c, trial s, q = tntrial
+1 ⋅ m −  σ s −

qn (3.17b)
σ f 

If these trial values remain negative, no change will occur for any of the fracture modes
(mode I or mode II). At the opposite end, one has to compute the corresponding values of the
damage multipliers γ nc +1 and γ nc +1 giving the final contribution of each mode in the current
step. The final result of such a computation concerns the potential evolution of the disconti-
nuity parameters, acn +1, due to the macro-crack evolution, which in return will also modify
ac . Now, the concrete part contribution to the RC
the total deformation field ε c (x, t) = ε c + G
residual rnc+1 can be computed according to:
rnc+1 =
∫N b
Ω e
T
dV −
n +1
∫B s
Ω e
T c
dV (3.18)
n +1
c
where s is the final admissible value of stress from the standpoint of the all damage cri-
n +1
teria φ , φ , φmc . The concrete part also provides a corresponding contribution to the tangent
c c
n
moduli of the RC composite, which can be obtained upon the static condensation of the
macro-crack-induced displacement discontinuity parameters according to:
cn +1 = K cn +1 − FnT+1H n−1+1Fn +1 (3.19)

K
with:
K cn +1 =
∫B C
Ωe
T d
n +1 BdV , FnT+1 =
∫B C
Ωe
T d dV
G
n +1
(3.20)
H n +1 =
∫
Ωe
dV + D n−1+1, Fn +1 =
GT Cn +d1G
∫
Ωe
GT Cn +d1BdV
d
where Cn +1 denotes the consistent tangent modulus of the bulk material (outside the discon-
tinuity). As K cn+1 , the tangent stiffness of the concrete part, is non-symmetric it can provide
detailed information on different concrete fracture modes and increased computational time
is here of secondary importance.
3.1.1 A Numerical Example of Small Experiments

A simple example of concrete specimen failure in a simple tension test illustrates the main
advantages of such a concrete model for a RC structure in providing crack-spacing infor-
mation and a decrease in resistance with an increase in crack-opening is shown. A square
specimen of length and width equal to 200 mm, with a unit thickness, is chosen having the
mechanical properties as stated in Table 3.1.
Owing to symmetry, only half of the specimen is considered in the computations (see
Figure 3.2). The computations are performed by incrementing the imposed displacement
at the free-end of the specimen in order to simulate a displacement-controlled tension test.
For the loading program that includes both loading and unloading, the force–displacement
response computed for coarse and irregular vs. fine and regular mesh remains practically
the same (see Figure 3.3).
The only difference concerns the computed total crack lengths for both a coarse and fine
finite element mesh (see Figure 3.2). This crack length difference appears as the consequence
of the adopted strategy to allow for crack creation only at the element centers. The presented
concrete model (with additional inelastic mechanism representing the FPZ) remains capable
of providing an invariant computed response (see Figure 3.3).
3.2 BOND-SLIP MODEL
In the spirit of the bond-slip role for the RC composite, several features have to be adopted.
First, the bond-slip element should be reduced to the concrete–steel interface and should be
thus without thickness. The bond-slip should not be activated as long as the dilatations in
the adjacent steel and concrete remain the same. The bond-slip is represented through the
shear strain, and its resistance can be increased by the normal pressure-induced confinement,
as well as decreased through transverse tension. This kind of interface model is developed
by combining the finite element interpolation for contact problems (Ibrahimbegović and
Table 3.1 Material Properties for Simple Tension Test

Continuum model Localized model
Young’s modulus 38 GPa σf 2.55 MPa
Poisson ratio 0.18 2.5 MPa (weakened element)
σf 2 MPa σ s /σ f 0.3
K 1000 MPa Fracture energy Gf 1.275 N/mm
Figure 3.2 Finite element model and computed macro-crack representation for coarse irregular mesh and
fine regular mesh.
Figure 3.3 Load–displacement diagram for regular and irregular meshes.
Wilson 1992) with the thermodynamics-based formulation of the inelastic slip mechanism.
The latter and essential features are here represented by the Drucker–Prager non-associative
plasticity model (Ibrahimbegović 2009). The equation for the bond-slip model then reads:
3 1
φ bs (s) = dev [s ] − κ(p) ≤ 0, κ ( p) = σc − pf , p = tr [s ] (3.21)
2 3
where dev [s ] and tr [s ] denote, respectively, the deviatoric and spherical parts of the stress
tensor, and σc is the sliding resistance at zero pressure, with no confinement. It is evident that
with the increase of pressure, the sliding resistance with a factor proportional to the model
parameter f will also increase. On the contrary, the tension stress across the interface can
actually reduce the lateral resistance (see Figure 3.5). Once the slip criterion is satisfied, the
slip is activated. This kind of slip is considered as the internal variable of the model, and its
evolution is governed by the von Mises plasticity potential with:
∂ φ vM vM 3
ε p = γ , φ (s) = dev [s ] − σ y (3.22)
∂s 2
Slip accumulation is therefore governed by a non-associative plasticity model leading to the

non-symmetric tangent stiffness matrix as in the concrete part. The bond-slip should not get
activated when the adjacent concrete and steel share the same strain field so in this respect
the bond-slip dilatation and the corresponding stress are not taken into account. In order to
implement such a model, a degenerate quadrilateral element with zero thickness in the FEM
is selected (see Figure 3.4).
By using a penalty-like formulation for the zero-thickness condition (e.g. Dominguez et
al. 2005), defined through small parameter hpen, the kinematics of the degenerated element
can be constructed by following the standard procedure for isoparametric quadrilateral
elements (e.g. Zienkiewics and Taylor 2005; Ibrahimbegović 2009)). The shape function
derivatives which are required for strain computations, can be formulated according to:
Figure 3.4 Degenerated quadrilateral bond-slip interface element.
∂N i   ∂N i   ∂y   ∂N i   ∂y  
=    −    / j
∂x   ∂ζ   ∂η   ∂η   ∂ζ  
∂N i   ∂N i   ∂x   ∂ N i   ∂ x  
=    −     / j (3.23)
∂y   ∂ζ   ∂η   ∂η   ∂ζ  
 ∂x   ∂y   ∂x   ∂y 
j =    −  
 ∂y   ∂η   ∂η   ∂ζ 
where we obtain the derivatives with respect to the natural coordinates:
∂y 
( ) (
= − y1 + y2 + y3 + hpen − y4 + hpen  / 4 + y0 η
∂ζ 
)
∂y 
( ) ( )
= − y1 − y2 + y3 + hpen + y4 + hpen  / 4 + y0ζ (3.24)
∂η  
( ) (
y0 =  − y1 − y2 + y3 + hpen − y4 + hpen  / 4 )
As hpen allows for one to carry out this kind of computation even for zero thickness of
the bond-slip element (nodal coordinates coincide for the corresponding nodes pairs),
from the given current iterative displacements of the four nodes of the bond-slip element
d(ii,)n +1, i = 1, 2, 3, 4, the corresponding best iterative guess of the total strain field for this ele-
ment can be computed as:
 ∂N i 
ε 
bs 4  0 ∂y  (i)

yy bs,(i )
 γ  =: ε n +1 =
 
bs
xy
∑i =1
Bi di ,n +1, Bi = 
bs (i ) bs
 ∂N i ∂N i 
{ }
 , di ,n +1 : d1c , d1s , d2c , d2s (3.25)
 ∂y ∂x 

where Bbs
i is the appropriate form of the strain–displacement matrix for the chosen ordering
of concrete and steel nodal displacements. The trial value of the bond-slip stress field is then
computed by assuming that the slip remains frozen with:
s ntrial
+1 = C
bs
(
ε bs ,(i ) bs, p
n +1 − ε n )
+1 = Ktr 
⇔ pntrial  ε n +1  ,
bs,(i )
dev s ntrial
+1  ( )
 = µ dev  ε n +1  − ε n (3.26)
bs,(i ) bs, p
φbs, trial =
3
2
dev s ntrial
+1  (
trial
 − κ pn +1 )
If the bond-slip criterion is φbs,trial < 0, the trial step is indeed the right guess and the slip
remains the same as in the previous step. For φbs,trial ≥ 0, the total bond-slip in the current
step and the current stress have to be corrected with:
∂ φnvM
+1 , dev [s n +1 ] =
pn +1 = pntrial
2
3
( )
κ pntrial
+1
+1
trial
∂s n +1 (3.27)

n +1 = pn +11 + dev [s n +1 ]
sbs
The consistent tangent can then be computed for the active slip case according to:
K bs
n +1 =
∫B
Γe
bs,T
C nep+1BbsdA,
(3.28)
 1 ∂ φ vM+1 ∂ φ vM+1  2 ∂ φnvM+1
C nep+1 = K1 ⊗ 1 + 2G  I − 1 ⊗ 1 − ntrial ⊗ ntrial  + K κ ′pn + 1 trial
⊗1
 3 ∂ s n + 1 ∂ s n + 1  3 ∂ s n +1
3.3 STEEL MODEL WITH CLASSICAL ELASTOPLASTICITY
The standard elastoplastic model with isotropic hardening, where the stress is defined from
the total and plastic strain, is chosen for steel making sure that the plastic admissibility of
stress is guaranteed (Figure 3.5).
(
σ ns +1 = E ε ns +1 − ε ns,+p1 )
(3.29)
φ s (σ) = σ − ( σ y + Kζ ) ≤ 0, γ n +1 ≥ 0, γ n +1φ s = 0
The value of this total strain is computed from the simplest two-node displacement field
representation for the bar with:
d2s ,n +1 − d1s,n +1
ε ns +1 = (3.30)
le
Figure 3.5 Elastoplastic bond-slip law accounting for transverse pressure confinement.

The plastic strain is computed from the standard procedure (e.g. Ibrahimbegović 2009) for
the model of this kind:
σ ns,+trial (s s, p
1 = E ε n +1 − ε n )
1 = σ n +1 − ( σ y + Kζ n )
φns,+trial s , trial
(3.31)
φns,+trial
γ n +1 = 1
≥0
K+E
( )
ε ns,+p1 = ε ns, p + γ n +1sign σ s
The steel bar stiffness matrix pertaining to steel displacement dis can then be written as:
Cnep+1A  I −I 
Kns +1 = (3.32)
l  −I I 
where l and A are, respectively, the reinforcement bar length and cross-section and Cep
is the corresponding value of the elastoplastic tangent modulus for elastic or plastic step,
respectively:
 E if γ n +1 = 0

Cnep+1 =  EK (3.33)
 E + K if γ n +1 > 0
We can further study the failure mode related to reinforcement bar failure if we consider
the post-peak softening response for the steel bar; the details of such a model enhancement
are very much similar to the corresponding modification of the concrete softening model,
and are thus omitted (e.g. Ibrahimbegović and Brancherie (2003) for a detailed discussion).
3.4 REINFORCED CONCRETE MODEL
It is important to make a correct assembly using the corresponding contributions from each
of the presented ingredients of the RC model. Such an assembly should be able to provide the
correct representation for a number of possible damage modes along with the right order of
activation between the concrete cracking, bond-slip and steel yielding. It is anticipated that
it will capture the complete failure mode for the RC, which is caused either by the failure
of the concrete with the creation of a macro-crack or a failure of the steel with the rupture
of the reinforcement. A potential failure mode of RC related to the eventual pull-out of the
reinforcement bar is not considered, assuming either that special provisions are made to
ensure steel reinforcement anchorage or that the concrete pressure confinement at the end
of each steel bar is sufficient to prevent this kind of extreme bond-slip. Such a hypothesis
implies that the slips remain equal to zero at both ends of the bar and sufficiently small in-
between so that the analysis of the failure mode can be carried out within the framework
of the small displacement gradient theory or linearized kinematics. It is important to note
that even such a small amount of slip along the inner portion of steel reinforcement bar can
lead to significant stress redistribution between the concrete and the steel, and, more impor-
tantly, provide the reliable information on final crack-spacing and opening, which will be
very valuable for durability studies.
3.4.1 Kinematics of Bond-Slip Along Reinforcement

Bar and its X-FEM Representation
A typical finite element for the composite RC model (see Figure 3.6) is considered, repre-
sented by structured finite element mesh.
The black nodes of the finite element mesh provide the corresponding description of the
concrete displacement field, whereas the white nodes are used to describe the displacement
field of the steel bar. It is considered that the bond-slip element possesses two black nodes
attached to concrete and two white nodes attached to steel. It is important to note that the
bond-slip element has a zero thickness, and therefore the black and white nodes share the
same sets of coordinates. The corresponding nodal displacements, however, are different
due to a bond-slip between the concrete and the steel. Such a displacement field can conve-
niently be represented by using X-FEM (see Belytschko et al. 2001; Belytschko et al. 2002;
Belytschko and Black 1999) where the global bond-slip along a particular reinforcement bar
is represented by the extended finite element interpolation; for example, with the horizontal
steel bar placed parallel to x-axis at the distance y , the X-FEM representation of the dis-
placement covering one entire composite element for RC can be written as:
4
1, y < y
u(x) Ωe = ∑ N (x)(d
i =1
i
c
i )
+ H y (y)a(x) , H y (y) = 
0, y ≥ y
(3.34)
where Ni(x) represents the standard isoparametric shape functions of the corresponding
element used for concrete, dci are the nodal values of the concrete displacement, H y (y) is the
Heaviside function and a(x) is the distribution of slip along the part of the reinforcement
bar that belongs to a particulate element. The Heaviside function H y (y) in (3.34) introduces
a displacement discontinuity in y direction due to bond-slip, whereas a(x) defines the corre-
sponding slip variation in x direction. It is important to emphasize that a(x) is a global field
describing the complete variation of the slip along a particular steel bar. The expression in
(3.34) represents the part of that field that belongs within a chosen element Ωe, which is con-
structed by appealing to the partition of unity properties (e.g. Melenk and Babuška 1996)
of the isoparametric shape functions Ni(x). Therefore, the appropriate local representation
of the global slip field can be obtained, and thus the standard finite element method for the
computations related to the slip field is exploited.
Such a discontinuity of the displacement field to account for bond-slip will impose the
corresponding strain discontinuity, which can be written as follows:
∂u(x, y) ∂α(x)
ε xx (x, y) Ωe = + H y (y) (3.35)
∂x ∂x
∂v (x, y)
ε yy (x, y) Ωe = + δ y (y)β(x) (3.36)
∂y
Figure 3.6 Finite element model of reinforced concrete constructed by a structured mesh, with nodes that
belong to concrete and nodes that belong to steel.
∂u(x, y) ∂v (x, y) ∂β(x, y)

γ xy (x, y) Ωe = + + δ y (y)α(x, y) + H y (y) (3.37)
∂y ∂x ∂x
where δ y (y) is the Dirac function obtained as the corresponding derivative of the Heaviside
function. The main advantage of the X-FEM representation of the bond-slip field is the abil-
ity to provide the appropriate coupling between all the bond-slip elements along a particular
steel bar. This is shown for the simplest case (chosen for clarity, but without any essential
restriction), where the bond-slip coincides with an element edge with y = 0. In this particu-
lar case, the slip field function over a single element can be represented by a linear polyno-
mial expression, which is uniquely defined by two sets of nodal values a j , j = 1, 2:
2
a(x, y) Γ e =
bs
∑ M (x)a , M (x) = N (x, 0) (3.38)
j =1
j j j j
where Γ bse is the corresponding element edge where the bond-slip element is placed. The finite
element interpolation in (3.31) can be written as:
4  2 
u(x) Ωe = ∑
i =1
Ni (x)  dci +
 ∑ j =1
Mj (x)a j  (3.39)

In such a case, a j can be defined as the nodal values of bond-slip displacements relative to
the concrete nodal displacement dci . The same displacement field can also be represented by
separating the description of the concrete vs. steel displacements according to:
uc (x)
Ωe
= ∑ N (x)d
i =1
i
c
i
(3.40)
2
u s (x) e
Γ bs
= ∑ M d (x), d
j =1
j
s
j
s
j = dcj + a j
where d sj are now the absolute nodal displacements of steel bar expressed in the fixed refer-
ence frame. While these two descriptions of the RC composite indeed represent the same
displacement field for the concrete and the steel components, the crucial difference between
them concerns the description of the bond-slip. In the first case in (3.39), the bond-slip is
measured in a relative sense in respect to the concrete displacement, which further implies
that one part of the steel bar displacement will accompany a given concrete displacement
as defined before by the current value of slip. This includes the starting value of zero slip,
in which case both the concrete and the steel undergo the same displacement. The inelastic
value of slip is what remains as the permanent differential displacement between the con-
crete and the steel, and plays the role of the inelastic deformation localized at the interface
between the concrete and the steel, with a given non-local distribution of bond-slip along
the steel bar. In the second case in (3.40), the bond-slip is measured in the absolute sense,
produced either by a given displacement of concrete (while keeping the steel fixed) or by a
given displacement of steel (while keeping the concrete fixed). The two displacement field
descriptions in (3.39) and (3.40) lead to very different solution procedures.
Here only the solution procedure for the relative displacement of the bond-slip element
(3.40) will be presented as the procedure for the absolute displacement of the bond-slip ele-
ment (3.39) is not robust enough to make this approach plausible. The main reason for that
lack of robustness is in the very unfavorable sequence of bond-slip computations, where the
motion of concrete nodes, keeping the steel nodes fixed, does not really correspond to the
real behavior of the RC assembly. An alternative manner to achieve the same goal of simpli-
fying the system while keeping the robustness is presented in the next section.
3.4.2 Solution Procedure for Relative

Displacements of Bond-Slip Element
The bond-slip phenomenon is considered in terms of relative motion in respect to the con-
crete. More precisely, the total motion of the RC composite is split between the motion of RC
with a fixed slip (including the zero slip value initially) imposing the mutually constrained
displacement of concrete and steel nodes, followed by the residual bond-slip that occurs due
to the redistribution of the internal force upon the concrete cracking. This kind of split is
more closely related to the more physical phenomena, which take place during the cracking
of concrete and the related stress redistribution within the RC composite. Additionally, this
kind procedure provides more robust computations, where the operator split procedure can
be used to separate the global slip of RC motions computation (with a fixed slip) from the
local computation of slip along a particular bar. It is obvious that the latter plays the same
role as the local damage deformation in concrete or local plastic deformation in steel, but
it is related to the complete steel bar. The key ingredient that allows one to integrate this
kind of non-local field within standard finite element computational procedure is related to
the special representation of the bond-slip field in terms of the X-FEM approximations (e.g.
Belytschko et al. 2001; Belytschko et al. 2002; Belytschko and Black 1999). To that end the
virtual displacement field will be constructed in the same manner as the real displacement
field defined in (3.39), but with nodal value of real displacement and relative bond-slip dci
and a j replaced by the corresponding virtual ones, denoted as w ci and h j , respectively; this
gives the corresponding representation of the virtual displacement field over a typical ele-
ment Ωe of RC composite with the bond-slip, which can be written as:
4  2 
w(x) Ωe = ∑
i =1
Ni (x)  w ci +
 ∑
j =1
Mj (x)h j  (3.41)

By assuming that such a virtual displacement field is applied at the deformed configuration
at pseudo-time tn+1 of a particular loading program, the weak form of the equilibrium equa-
tions can be obtained. The latter contains the first global term where the contribution of all
concrete elements is accounted for along with the corresponding contribution of the steel
bar, which is constrained to move along with concrete as defined by the given value of bond-
slip. The corresponding residual then reads:
bs
nel
( ) ( )
nel
r cs :=  f c,int,e dcn +1 +  f s,int, e dcn +1, a n − fnext
+1 = 0
e =1 e =1
( ) ∫ (∇ N ) s (d ) dV
T
fic, int,e dcn +1 = s
i
c
n +1
c
n +1
Ωe (3.42)
T
 dMj  s
(
fjc,int,e dcn +1, a n = ) ∫ c
(
 dx  σ n +1 d n +1, a n dA )
e
Γ bs
fiext
, n +1 =
∫N
Ωe
i
T
bdV
This global equation is accompanied by the local equilibrium equation, which concerns
redistribution of the slip with:
( )
nel
r bss :=  f s,int,e dcn +1, a n +1 + f bs,int, e ( a n +1 ) = 0
e =1
(
fjs,int,e dcn +1, a n +1 = ) ∫  dM 
dx 
j
( )
σ ns +1 dcn +1, a n dA (3.43)
e
Γ bs
fjbs,int,e ( a n +1 ) =
∫Be
Γ bs
bs ,T
s nsb+1 ( a n +1 ) dA
The computation of the stress values for each component of the RC is again carried out as
described previously. From equations (3.42) and (3.43) it is evident that only the steel bar
contributes to both sets of equations, whereas the bond-slip computation is limited only to
the last equation. In fact, such an equation is very much like the one used for computing
the internal variables for representing concrete cracking and steel yielding; only it concerns
more than one element at the time since connecting all the bond-slip elements along a par-
ticular bar.
For that reason, the operator split methods, where the sets of in (3.42) and (3.43) are
solved sequentially, provide the most appropriate approach. The benefit of this approach is
dual; firstly, one is not to deal simultaneously with two sets of equations of different sizes,
as well as the guarantee of the computational robustness where the corresponding slip evolu-
tion within a particular time step is always computed by starting from the converged value
obtained in the previous step. The latter allows avoiding eventual spurious loading–unload-
ing slip cycle, which can occur when advancing the slip and the concrete displacement com-
putations simultaneously.
For a typical time step between tn and tn+1 we can thus write:
• For n=0,1,2…
• Given: dcn , a n dc n, an, internal variables at tn
• Find: dcn +1, a n +1, internal variables at tn+1
• Iterate: (i)=1,2…
• Iterate: (j)=1,2…
compute internal variable evolution for given dcn,(+i1) , a(nj+) 1.
∂r sbs,(i)

∂a n + 1
( ) ( )
a n +1( j +1) − a n( j +1) = − r sbs dcn,(+i1) , a n +1( j) (3.44)
Test convergence locally:
( )
IF r sbs dcn,(+i1) , a n +1( j) > tol NEXT (j)
( )
ELSE r sbs dcn,(+i1) , a n +1( j) < tol ⇒ a n +1(i +1) = a n +1( j +1) NEXT (i)
 ∂r cs ∂r cs  ∂r sbs  ∂r sbs  c,(i +1)

−1
 c,(i) − (i )  (i )  (i )
 d n +1 a n +1  a n +1  a n +1 
( )
 d n +1 − dcn,(+i1) = − rncs+,(1i) (3.45)
Test convergence globally:
( )
IF r cs dcn,(+i1+1) , a(ni ++11) > tol NEXT (i)
ELSE r cs
(d c ,(i +1)
n +1 )
, α(ni ++11) < tol ⇒ a n ← a(ni ++11) , d n ← dcn,(+i1+1) NEXT (n)
From (3.44) and (3.45) it is evident that the computation of the bond-slip parameters is
carried out for the given nodal values of concrete displacements. Hence, a smaller set of
equations governing bond-slip is forced to converge for a given iterative value of concrete
displacement. This could be interpreted as imposing the admissibility of the bond-slip forces
in the given slip criterion, which is fully equivalent to imposing the admissibility of the stress
field for each component of the RC composite in the sense of the given damage criterion for
concrete and the plasticity criterion for steel. It is important to note that such a computa-
tion is also carried out for each iterative sweep (either (i) or (j)) in order to provide the cor-
responding values of the internal variables governing the cracking of concrete or the plastic
yielding of steel. With these values in hand, the corresponding form of the tangent moduli
for concrete can also be obtained, steel and bond-slip, which allows one to compute accord-
ingly the tangent operators needed in (3.45). Hence, explicitly it can be written as:
∂risbs
Hijsbs,( j) =
∂a(jj,)n +1 e
=
∫ dx
( )
dMi ep, s c,(i) c,(i) dMj
Cn + 1 d n + 1 , a n + 1
dx
dA
Γ bs e
Γ bs
(3.46)
+
∫B bs,T
i C ep, bs
n +1 (a ) B(i )
n +1
bs
j dA
e
Γ bs
where Cnep,+1s and C n+1

ep,bs
are given in (3.33) and (3.28), respectively. In (3.46) above, Bbs is the
strain–displacement matrix describing the kinematics of the bond-slip element with relative
degrees of freedom. Namely, by accounting for the relation between the absolute and the
relative displacement given in (3.40) it can be stated that:
 I 0
Bbs = Bbs T, T =  (3.47)
I I 
The explicit form of the tangent stiffness for concrete will not change with respect to the
result in (3.19) and if Newton’s iterative method is used, the tangent operators can be com-
puted as:
∂r cbs ∂r cbs
K ncbsc
+1
,(i )
= , K cbss,(i )
n + 1 =
∂dcn,(+i1) ∂d ns +1
(3.48)
sbsc ,(i ) ∂r sbs ∂r sbs
K n +1 = c,(i) , K sbss
n +1
,(i )
=
∂d n + 1 ∂d ns +1
Finally, the coupling term in the global tangent stiffness matrix is provided by the steel com-
ponent and can be written as:
∂ r sbs,(i),e dNi ep, s dN j

Fijsbs , e ,(i )
, n +1 :=
∂dcn +1 ∫=
e
Γ bs
dx
Cn + 1
dx
dA
(3.49)
∂r cs,(i),e dMa ep, s dNb
F bss, e ,(i )
ij , n +1 := i
∂a n + 1 ∫e
Γ bs
=
dx
Cn + 1
dx
dA
where Cep,s is given in (i), for the corresponding case of internal variables’ computations.
3.4.3 Numerical Examples
In this section, the results of several numerical simulations are validated, which are given in
order to illustrate the corresponding advantages of the proposed approach. All the compu-
tations are carried out by a research version of the computer program FEAP developed by
Prof. R.L. Taylor at UC Berkeley (e.g. Zienkiewics and Taylor 2005).
3.4.3.1 Traction Test and Proposed RC Model Validation

The first example considers the traction test on an RC beam, which was carried out by
Mivelaz (1996). The experimental results presented in Mivelaz (1996) pertain both to a
global response, such as the resulting force–displacement diagram, and to a local response,
such as the observed cracks’ distribution and measured cracks’ width. For that reason, this
particular test was used to provide complete validation of the proposed RC model. In the
first phase formation of a macro-crack (visible to the naked eye) is observed once the ulti-
mate concrete resistance is reached. In the second phase, once the bond-slip mechanism
is activated redistribution of the internal force between the steel and the cracked concrete
entering the softening response phase is observed and this is tested as well. It is the latter
that decides, to a large extent, not only the global response but also the details of the local
response with resulting crack-spacing and opening.
The selected specimen geometry and loading conditions used in the test are presented in
Figure 3.7. The test specimen is in a large RC panel with the dimensions: 3000×420×1000
mm, reinforced with three layers of steel bars. Each layer contains six steel bars whose
diameter is equal to 16 mm. The material properties of concrete, steel and the bond-slip are
given in Table 3.2.
The finite element mesh is constructed with CST elements (Brancherie and Ibrahimbegović
2009), as shown in (Figure 3.8), by using 200×28 elements. Three layers of enhanced RC
finite elements presented herein, which are used to provide the proper representation of the
Figure 3.7 Finite element model of reinforced concrete constructed by a structured mesh, with nodes that
belong to concrete and nodes that belong to steel.
Table 3.2 The Material Properties of Concrete, Steel and the Bond-Slip
Material properties
Concrete Steel Bond-slip
E c = 45700 MPa E s = 210000 MPa Kb = 25400 MPa
νc = 0.20 νs = 0.30 G b = 19000 MPa
Gf = 193 N/m τ y = 6 MPa
σ f = 2.77 MPa
Figure 3.8 RC specimen used in simple tension validation test: Chosen finite element mesh.
bond-slip field along the steel reinforcement, are indicated in a light gray color and posi-
tioned all along the three steel reinforcement layers. The bond-slip is prevented at both ends
of each steel reinforcement layer, which is used in order to model the anchors and ensure the
corresponding zero value of bond-slip.
The computations are carried out under imposed end displacement. The computed global
response, in terms of an imposed strain–force diagram is presented in Figure 3.9 (top curve).
The very good performance of the proposed model and very satisfying results of the numeri-
cal simulations with respect to those measured experimentally (see top graph in Figure 3.10),
in regard to the prediction of the global force–displacement curve, are observed. It is obvi-
ous that the model is capable of capturing the initial elastic part and FPZ creation, followed
by the first fully developed crack that leads to a sudden drop in resistance. The model is
also able to express further stress redistribution and the resulting recovery of the specimen,
before the next macro-crack develops, as well as the same scenario repeating for the final
crack appearance in this specimen. It is important to note that it is not only the force values
but also the imposed strain values, computed at each new crack, that match the experimen-
tal results very well.
Validation of the local results pertains to the cracks’ width. Comparing the bottom curves
in Figure 3.9 against those in Figure 3.10, it is clear that the computed cracks’ width is
capable of matching quite well with the experimental results. All the salient features of
local response are captured. Namely, a first sudden jump is seen when opening the first
macro-crack due to the brittle localized failure of the concrete, at an imposed strain value
Figure 3.9 Computed force–displacement response and crack width.

Figure 3.10 Experimentally obtained force–displacement response and crack width.
of 0.75×10 −4. This crack-opening keeps growing (almost linearly) until an imposed strain
value of 1.75×10 −4 where the second crack appears. The sudden appearance of the second
crack will lead to a partial closing of the first crack. From then on, both existing cracks will
keep growing, until the sudden appearance of the third crack at imposed strain value of
2.75×10 −4, when they both repeat the same scenario of partial closing again while the new
crack takes off. The final picture for crack openings is thus reached, with the final values of
roughly 0.2 to 0.25 mm.
The computational results presented in Figure 3.11 illustrate quite well the fundamental
role of the bond-slip in internal force transfer and resulting crack-spacing. Figure 3.11 pres-
ents the bond-stresses along the particular reinforcement layer, recorded at different stages
of the simulation for one, two and three activated macro-cracks, in two Gauss numerical
integration points (GNP) in each bond-slip element. It is seen that the concrete and the steel
share roughly the same strain value before the first crack appears and that the bond-slip has
pretty much no effect. The bond-slip gets activated once the first crack appears. The bond-
stress keeps growing up to the yield value quickly spreading in the crack neighborhood (in
accordance with the perfect plasticity model representation), leading to plastic reinforce-
ment sliding relative to concrete. The direction of plastic sliding is not the same on either
side of the crack, on one side it is positive, and on the other side negative sliding is observed.
For example, on the left side of the crack, the concrete moves sharply to the left as the
activation of the element embedded discontinuity, whereas steel remains in place. Therefore,
the positive relative sliding is observed resulting in the positive value of bond-stress. The
opposite signs and the same scenario apply to the right side, resulting in the negative value
of bond-stress.
The bond-stress subsequent to activation of each of three cracks is illustrated in Figure 3.11.
The first one opens near GNP 140, leading to different directions of relating sliding on both
sides of the cracks as described above. The second crack appears near GNP 20, clearly
leading to unloading at the first crack. And finally, the last crack appearing near GNP
100 unloads each of the first two cracks. Figure 3.12 shows the relative displacement field
together with strong discontinuity field indicating the macro-cracks’ locations.
Figure 3.11 Bond-stress distribution along reinforcement layer at different Gauss numerical integration
points (GNP).
Figure 3.12 Relative displacement field contours and final spacing of macro-cracks (indicated by solid lines).
The efficiency in the iterative procedure is shown in Figure 3.13 (relative vs. absolute
bond-slip displacement approach). Namely, in Figure 3.13, we give the total number of itera-
tions needed to carry out the computations by using the relative bond-slip displacements
(solid-line) vs. the absolute bond-slip displacements (dashed-line). We recall that the former
always start with the frozen value of the bond-slip when predicting the RC displacement in
the subsequent step, as opposed to the latter when the initial displacements of concrete and
reinforcement are advanced independently.
Once the first crack appears and the bond-slip gets activated, the relative bond-slip dis-
placement approach remains fairly stable with a nearly constant number of iterations in
each time step. Whereas the absolute bond-slip displacement approach faces more and more
pronounced difficulties in converging with each new crack that is activated, ending with an
ever increasing number of iterations because initial displacements of concrete and reinforce-
ment are advanced independently.
3.4.3.2 Four-Point Bending Test and Crack-Spacing Computations

Here, a for four-point bending test (very frequently used) for quasi-brittle materials is elabo-
rated. Here a qualitative illustration pertaining to crack patterns for different reinforced
cases, with either heavy or weak reinforcement case is provided.
a) Crack-Spacing Computations for Heavy Reinforcement

The selected specimen is presented in Figure 3.14. For this RC specimen, only one steel layer
is modeled (2×16 mm located at 30 mm from the bottom line).
Figure 3.13 Total number of iterations in each time step for relative (solid-line) vs. absolute (dashed-line)
bond-slip displacement approach.
Figure 3.14 Four-point bending test on RC specimen: geometry, loading and finite element mesh.
In the same way, as in the previous example, we can observe different mechanisms at dif-
ferent stages of the loading process. Figure 3.15 clearly illustrates the appearance of the first
macro-cracks in-between the section where the loading is applied.
The crack pattern keeps growing along with the bond-slip increase, with some cracks
developing up to the core of the structure. Figure 3.16 illustrates the crack pattern at an
advanced post critical scenario.
Finally, the global response (Figure 3.17) clearly shows a loss of rigidity due to the dam-
aged concrete structure. After an elastic part, subsequent loading/unloading sessions are
observed when cracks appear and keep growing. As the structure is heavily reinforced, large
discontinuity in the response is not observed. Namely, the steel plays for this case a domi-
nant role in ensuring the structural resistance.
b) Crack-Spacing Computations for Weak Reinforcement

Here the steel reinforcement is only half of the one from the previous example, while all
other properties still remain the same.
The crack pattern represented in Figure 3.18 shows the cracks that are developed more
regularly than in the previous case with high reinforcement percentage. This phenomenon is
due to the weaker role that steel plays in the structural process. Figure 3.19 shows the role
played by the bond-slip. On both parts of a crack, positive and negative sliding is observed.
On the left side, concrete moves sharply to the left due to the discontinuity, whereas steel
remains at the same place. As a consequence, positive relative sliding is observed. The oppo-
site signs and the same scenario apply to the right side.
Figure 3.15 Bond-slip along the bar and crack pattern for the first macro-cracks.
Figure 3.16 Crack pattern at an advanced loading stage.
Figure 3.17 Global response.

Figure 3.18 FPZ and cracks’ opening.
Figure 3.19 Crack pattern and bond-slip sliding.
Strong discontinuities in the loading/displacement curve for weak reinforcement is

seen (see Figure 3.20) due to the fact that concrete is predominant in the structural
response. Once concrete is totally damaged, an elastic part corresponding to the steel
behavior is evident.
A novel approach to finite element modeling of the RC structures is presented. The pro-
posed model is capable of providing detailed information about each of the ingredients of
RC; the resulting information on crack-spacing and opening, which is of great importance
for durability problems.
The problem of interest here concerns the case where the bond-slip is not excessive (no
steel reinforcement pull-out), where even the simple Drucker–Prager perfect plasticity model
can be used for representing the bond-slip constitutive behavior. Even a simple model for
bond-slip is shown to ensure the proper description of internal force redistribution upon
concrete cracking and determines to a large extent the crack-spacing.
Figure 3.20 Global response for weak reinforcement.
The main idea presented here, which has to be emphasized, is pertaining to a judicious
combination of the ED-FEM-based local representation of concrete cracking with the
X-FEM-based global representation of connected bond-slips along a particular steel rein-
forcement in order to provide the corresponding strain field representation in an RC struc-
ture with cracks.
The choice of the relative displacement for connected bond-slip parametrization based
upon X-FEM is shown to reduce the total number of iterations in respect to absolute
displacement.
Finally, we should also indicate the potential of the proposed methodology for RC to deal
with other composite materials with long fibers.
3.5 REINFORCED CONCRETE MODEL FOR CYCLIC LOADING
Modeling the behavior of concrete until the collapse in dynamic loading applications is a
rather complex problem. A robust elastoplastic-damage model and its implementation in
finite element procedures is already shown in Ibrahimbegović, Jehel and Daven (2007). In
addition, here, in the physical formulation: (i) viscosity is added by appealing to the so-called
Kelvin–Voigt viscoelastic model; (ii) the modeling of the softening part, which is introduced
in the formulation by a discrete plastic or damage model that requires the definition of an
enhanced strain field, is presented as well and (iii) a clear distinction between tensile and
compressive constitutive models is made. On the numerical side, the conditions under which
the model can enter the generalized standard materials class (Halphen and Nguyen 1975),
whose elements benefit from good robustness properties, are identified.
Here, it is interesting to note the motivation of viscosity introduction into the material model.
Viscosity in concrete can be associated to its loading-rate-dependent macroscopic behavior that
results from several nano and microscopic phenomena such as the complex interaction between
moisture and the micro-structural solid skeleton, micro-cracking process, Stefan effect, micro-
inertia of the material surrounding the crack tip (Pedersen, Simone and Sluys 2008). Second,
viscosity can also be associated to creep phenomena in highly dissipative materials such as
asphalt concrete (Panoskaltsis et al. 2008). Thirdly, its introduction leads to the explicit appear-
ance of a time step in the governing equations of the discretized problem. The possibility of
determining a critical time step then provides a tool to regularize numerical problems that
can be ill-posed in the presence of materials that exhibit strain softening behaviors (Oliver
and Huespe 2004). And finally, viscosity can also be used to introduce a source of energy dis-
sipation at the material level: it can lump the energy dissipation that physically comes from
unknown or not (accurately) modeled nonlinear physical phenomena in the materials.
3.5.1 Constitutive Concrete Model

3.5.1.1 Thermodynamics with Internal Variables (TIV)
To represent the salient phenomena observed in the experimental cyclic response of a con-
crete representative elementary volume (REV) (Figure 3.21), a set α of internal variables is
chosen. The phenomena that are modeled: (i) loading-rate-dependent behavior, (ii) strain
hardening, (iii) strain softening, (iv) appearance of residual deformation, (v) stiffness and
strength degradations, (vi) hysteresis loops. It is considered that the phenomena (ii), (iv) and
(vi) occur only in compression, that strain softening in compression is due to the localization
of permanent deformation whereas strain softening in tension is due to the localization of
deformation that completely disappears after unloading. Internal variables are associated
with each of these phenomenologically identified mechanisms.
The set of internal variables is defined in Table 3.3. Continuum quantities are denoted as
• whereas discrete quantities being localized, are marked with •.
A noteworthy advantage of this material model is that all its parameters have a clear
meaning and can, therefore, be easily identified according to experimental curves. The vis-
cous parameter η can be seen as a material property and identified from free-vibration tests;
the tensile softening curve coefficient a a is related to the fracture energy Gf, a fracture
mechanics concept (Hillerborg, Modéer and Petersson 1976; Bažant, Yu and Zi 2002) that
quantifies the total amount of energy that has to be furnished in tension to a concrete sec-
tion between the time when displacement begins to localize and the time when the section
is completely broken; all the other parameters can be identified from stress–strain curves of
concrete specimen in quasi-static cyclic loading.
Figure 3.21 (a) Experimental (adapted from Ramtani 1990) and (b) numerical (by using the proposed consti-
tutive concrete model) cyclic behavior of concrete in quasi-static loading.
Table 3.3 Set of Internal Variables α, Corresponding Affinities  and Phenomenological Coefficients PCs (Parameters of the Model)
t/c Compression Tension
α ευ εp ξp λp D ξp up ξp D ξd
(viscous (plastic (isotropic (kinematic (stiffness (isotropic (plastic (plastic (stiffness (damage
deformations) deformations) plastic plastic degradation) damage deformations) strain degradation) strain
strain strain strain softening) softening)
hardening) hardening) hardening)
 συ σ qp κp σ2 qd t qp t2 qd
2 2
PC E, η E Kp Hp E −1 Kd σ∞ , a
E Kp E −1
3.5.1.2 Governing Equations of the Constitutive Model

3.5.1.2.1 Basic Ingredients
The three basic ingredients for developing a constitutive model in the framework of the TIV
are defined as:
1. The split of the total strain into a viscoelastic, a plastic, a damage and another dis-
crete part due to the likely localization of the displacement. -space is divided into
two parts: the terms involved in tension σ ≥ 0 are mentioned by the sign •+ and those
involved in compression σ < 0 by the sign •−
u(x, t) = u(x, t) + u(t)x (x)

∂u(x, t) (3.50)
ε(x, t) =
∂x
= ε v + ε p + u◊ + u p δx ( )
∂u(x, t)
where ε(x, t) =
∂ x
= ε v + ε p is the continuum part of the strain and uδ x = u ◊ + u p δ x ( )
− with u ◊ = u d in tension and u ◊ = u e in compression—its discrete part introduced by
the Dirac function δ x that is the derivative of the Heaviside’s function x that intro-
duces the displacement jump u in the description of the displacement. ε v corresponds
to the deformation in a Kelvin–Voigt rheological model: a spring−elastic in tension
(ε v = ε p ) and damaging in compression (ε v = ε d ) − and a dashpot in parallel. For sim-
plicity, it is assumed that the displacement jumps only take place in one section; exten-
sion to several localization sections is straightforward.
2. Helmholtz free energy functional

( ) ( ( ) ( ))
ψ + (u, α) = ψ e u + + ψ d u + , D + Ξd ξ d δ x
(3.51)
−
ψ (u, α) = ψ d
( u , ε , D) + Ξ ( ξ ) + Λ ( λ ) + Ξ ( ξ ) + ( ψ ( u , u ) + Ξ
− p p p p p d d e − p p
( ξ )) δ
p
x
3. The nonlinear plasticity and damage mechanisms are activated for a positive value of
the corresponding plasticity and damage criteria
• For σ ≥ 0 (tension), the dissipative mechanisms that can be involved in the evolu-
tion process are activated according to the following criteria:
φ v, + (σ) = σ v = σ − σ e ≤ 0
(3.52)
( ) (
φ d , + t, q d = t − σtu − q d ≤ 0 )
where t is the stress at the discontinuity and σtu is the ultimate stress in tension
• For σ < 0 (compression), the criteria functions are:
( )
φ v, − ( σ ) = −σ v = − σ − σ d ≤ 0
φ p, − (σ, q , κ ) = σ + κ − (σ − q ) ≤ 0 (3.53)
p p p
y
p
φd,− (σ, q ) = −σ − (σ − q ) ≤ 0
d
f
d
φ p, − ( t , q ) = −t − ( σ − q ) ≤ 0
p c
u
p
where σy, σf, and σcu are yield, fracture and ultimate stresses in compression.
State Equations of the System

The expression of the energy dissipated by this viscoelastic-plastic-damage model with dif-
ferent hardenings/softenings within a time unit can now be written. This can be computed
as the difference between the total amount of energy imparted to the system and the amount
of energy stored by the system during this time unit.
In tension:
( ) (
 + = σε + + tu + − ψ + u + , u + , D, ξ d = )

∂ψ e  ∂ψ d ∂ψ d ∂Ξ d d 
( e v
)
σ + σ ε + + tu + δ x − + ε + −  + u + +
∂ε  ∂u ∂D
D+
∂ξ d
∂ξ  δ x

It is clear that ε + = ε e = ε v and u + = u d because u p cannot evolve in tension according to

equations (3.52) and (3.53). Thus:
 ∂ψ e   ∂ψ d  ∂ψ d ∂Ξ d d
 + =  σ e − e  ε e + σ v, + ε v, + +  t − d  u d δ x − Dδ x − ξ δ x (3.54)
 ∂ε   ∂u  ∂D ∂ξ d
In compression:
( ) (
 − = σε − + tu − − ψ − u − , u − , ε p , ξ p , λ d D, ξ d , u p , ξ p =)
 ∂ψ d ∂ψ d ∂Ξ p p ∂Λ p p ∂Ξ d d 
(σ ) )( )
d
+ σ v ε − + tu − δ x −  ε −
− ε p + D + p ∂ξ + λ + d ∂ξ 
(
 ∂ ε − − ε p ∂D ∂ξ ∂λ p ∂ξ 
 
−
∂ψ e
(
 ∂ u − u p
)(
u − − u p +
∂ξ p
)
∂Ξ p p 
∂ξ δ x

 
It is clear that ε − = ε v + ε p with ε v = ε d and u − = u e + u p . Thus:
 ∂ψ d  ∂ψ d ∂ Ξ p p ∂ Λ p p
 − =  σ d −  ε + σ ε + σε −
d v v, − p
D − p ∂ξ − λ
 ∂ ε− − εp( )  ∂D ∂ξ ∂λ p
(3.55)
∂Ξ d  ∂ψ e  ∂Ξ p p
− d ∂ξ d +  t − e p
 u δ x + tu δ x − p ∂ ξ δ x
∂ξ  ∂ u− − up( )  ∂ξ
Note that ε e , u d , ε d , u e are not internal variables. Moreover, when there is no evolution of
any internal variable, the dissipation is null. Therefore, according to equations (3.54) and
(3.55), the following state equations are valid:
∂ψ e d ∂ψ d + ∂ψ d − ∂ψ e
σe = ,σ = ,t = ,t = (3.56)
∂ε e ∂ε d ∂u d ∂u e
For the thermodynamic forces, expressions that will then lead to an accurate reproduc-
tion of the experimentally observed behavior of a concrete REV, the different terms in the
expression of the Helmholtz free energy are defined as follows:
ψe =
1 e e
2
ε Eε ψd =
1 −
2
( ) (
ε − ε p D−1 ε − − ε p )
1 1
ψ d = u d D−1u d Ξ p = ξ pK p ξ p
2 2
( 
) 1 d
Ξd = − σtu − σ ∞  ξ d + e − aξ 
 a 
1
Λ p = λ pH pλ p
2
(3.57)
1
Ξd = ξ d K d ξ d
2
( ) (
ψe = u− − up E u− − up )
1 p p p
Ξp = ξ K ξ
2
The state equations of the internal variables are defined from the Helmholtz free energy
functional that plays the role of a thermodynamic potential:
∂ψ
i = − (3.58)
∂α i
Finally, the state equations of the system, that is the equations that characterize the state of
a concrete material point at a given instant in -space, are given in Table 3.4.
3.5.1.2.2 Equations of Evolution of the Internal Variables

Here, the equations that govern the evolution of the internal variables of this concrete phe-
nomenological constitutive law are determined. The same as in the models for steel, it is
assumed that the evolution is driven by the principle of maximum dissipation. Table 3.4
presents the state equations.
Taking this into account the dissipation equations (3.54) and (3.55) can be rewritten as:
p p 1
 = σ

v
ε+ σεp
+ q ξ + κ pλ p + σDσ + q d ξ d

2

Dv
Dp

Dd
(3.59)
  1 
+  tu p + q p ξ p  δ x +  tDt + q d ξ d  δ x
  2 

Dp Dd
To characterize the evolution of the system, except the viscous one, the principle of maximum
dissipation is to this model and thus the flow of the internal variables is computed by assum-
ing that they maximize the dissipation. A problem of constrained maximization that involves
inequalities has to be solved, recalling that the criteria functions equations (3.52) and (3.53)
( )
must be satisfied. Denoting f = σ, t, q p , κ p , qd , q p , q d , the following problem has to be solved:
max f  with constraints φ p (f ) ≤ 0, φd (f ) ≤ 0, φ p (f ) ≤ 0, φ d (f ) ≤ 0 (3.60)
This problem is solved by utilizing the Lagrange multipliers method and the Lagrangian is
defined as:
φ p (f ) ≤ 0, φd (f ) ≤ 0, φ p (f ) ≤ 0, φ d (f ) ≤ 0 (3.61)
( )
where γ = γ p , γ d , γ p , γ d is the set of Lagrange’s multipliers associated to each constraint.
According to the Kuhn–Tucker defined in Chapter 2 equation (1.11), and here repeated due
{ }
to consistency. If the set f * = fi* is a solution of the problem, then there exists a unique set
of Lagrange multipliers such that the following relations are verified for all i and for all j
where the last three conditions are referred to as the loading/unloading conditions.

(
∂ f *, γ * ) = 0, ∂ ( f , γ ) ≥ 0, γ
* *
*
j ( )
≥ 0, γ *j φ j f * = 0 (3.62)
∂ fi γ j
Finally, the equations of evolution of the internal variables are given in Table 3.5.
For the viscous internal variable ε v , the evolution is expressed, in an associative form to
(Perzyna 1966):
φ v ∂φ v
ε v = (3.63)
η ∂σ
Table 3.4 State Equations of the System

Tension Compression
σe = E ε e σd = D ε d
t = D −1u d t = E ue
)( )
Ξd Ξp
qd = −
ξd d
(
= σut − σ ∞ 1 − e − aξ
d
qp = −
ξp
= −K p ξ p
1 2 ∂ψ Λp
t =− κ p = − p = −H p λ p
2 ∂D λ
Ξd
q = − d = −K d ξ d
d
ξ
1 2 Ξp
σ =− = −K p ξ p
2 ξp
Table 3.5 Equations of the Evolution of the Internal Variables

Continuous plasticity Continuous damage
∂φ p ∂φ d
ε p = γ p D σ = γ d
∂σ ∂σ
ξ p = γ p ∂φ ξ d = γ d ∂φ
p d
∂q p
∂q d
λ p = γ p ∂φ
p
∂κ p
γ p ≥ 0, φ p ≤ 0, γ pφ p = 0 γ d ≥ 0, φd ≤ 0, γ d φd = 0
Discrete plasticity Discrete damage
∂φ
p
∂φ
d
up = γp Dt = γ d
∂t ∂t
p p ∂φ p d d ∂φ d
ξ =γ ξ =γ
∂q p ∂q d
p d
γ ≥ 0, φ p ≤ 0, γ p φ p = 0 γ ≥ 0, φ d ≤ 0, γ d φ d = 0
where η is the viscous parameter assumed here to be rate-insensitive although this might
not be the case (Pedersen, Simone and Sluys 2008). It is interesting to see that (3.63) is in
accordance with the rheological relation in a dashpot σ v = ε v η.
Note also that concrete is in general not an associative material. However, in the 1D con-
text of this paper, we nevertheless assume that it is the case. All the equations of evolution
of the internal variables are thus expressed in an associative form. The normality of the flow
rule to the loading surface is in particular implied by the appeal to the maximum dissipation
principle.
3.5.1.3 Euler–Lagrange Equations of a Concrete

1D Structure in Dynamic Loading
The local constitutive model is used as the basis for the formulation of the structural,
mechanical problem. This latter, illustrated here in the 1D case, has to be adapted to the
local constitutive model and requires the definition of an enhanced displacement field.
3.5.1.3.1 Enhanced Displacement Field Kinematics

Figure 3.22 illustrates that the displacement field is written as the sum of a smooth lin-
ear part u (x, t) enhanced by an additional part u (x, t) to take into account the possible
appearance of a displacement jump u(x, t) in the 1D structure (Garikipati and Hughes 1998;
Ibrahimbegović and Brancherie 2003)
u(x, t) = u (x, t) + ( x (x) − ϕ(x)) u(t) (3.64)

u ( x ,t )
which can also be written, in the form adapted for identification with (3.50), as:
u(x, t) = u (x, t) − ϕ(x)u(t) + u(t)x (x) (3.65)

u(x,t )
Then with ε(x, t) = ε(x, t) + u(t)δ x
∂u(x, t) ∂u (x, t) d ϕ(x)

ε(x, t) = = − u(t) (3.66)
∂x ∂x dx
Figure 3.22 Enhanced displacement field in a 1D structure. The solid, dashed and dotted lines represent the
enriched displacement field, the smooth displacement field and the continuum displacement field.
In this 1D context, only failure mode I is considered; we refer to Dujc et al. (2010) for a
presentation of the kinematics with an embedded crack of quadrilateral 2D finite elements
where several failure modes are considered.
3.5.1.3.2 Complementary Lagrangian Variational Formulation

The kinetic energy, assumed to depend only on the smooth part of the displacement, is:
1  ∂u (x, t) 
T (u , t) =  ∫
2  ∂t 
Ω
 d Ω (3.67)
where the volume mass ρ in the volume Ω is considered as constant.

The total potential energy can be written:
U ( u , u, α, t ) =
∫ ψ (u, u, α, t ) dΩ − U
Ω
ext
( u , t ) (3.68)
To introduce the stress fields σ and t as other independent variables of the problem and thus
give a more general setting that can be helpful to derive the governing equations of the prob-
lem (Markovic and Ibrahimbegović 2006; Simo, Kennedy and Taylor 1989), it is necessary
to appeal to the partial Legendre transformation of the Helmholtz free energy:
1 e e  d 1 
ψ + = σe ε e − σ Eσ +  tu − tDt + Ξd  δ x
2  2 
(3.69)
1  1 
ψ = σ ε − σ d Dσ d + Ξ p + Λ p + Ξd +  tu e − tE−1t + Ξ p  δ x
− d d
2  2 
Recalling that ε e = ε v in tension, ε d = ε v in compression, σ = σ v + σ e / d and according to equa-

tions (3.50) and (3.66), these equations can be rewritten with respect to u (x, t), u(t), σ(x, t), t(t)
and α:
 ∂u d ϕ
ψ+ = σ  −
 ∂x dx
 1
(
u − ε p  − σ v ε v − σ − σ v E−1 σ − σ v +
 2
) ( )


( 1
) 
+  t u − u p − tDt + Ξd  δ x
2 
(3.70)
 ∂u d ϕ
ψ− = σ  −
 ∂x dx
 1
u − ε p  − σv ε v − σ − σv D σ − σv
 2
( ) ( )


(
1
) 
+ + Ξ p + Λ p + Ξd +  t u − u p − tE−1t + Ξ p  δ x
2 
L is denoted as the Lagrangian of the mechanical system. It is defined as L = T − U and
Lagrange’s variational principle can be written as:
t2

∫ δL (u, u, σ, t, t ) dt = 0
t1
∀t1, ∀t2 ξ (3.71)
with δu (x, t), δu(t), δσ(x, t), δt(t) = 0 when t = t1 and t = t2 .
The following set of governing equations available for tension and compression are
obtained after some mathematical operations:
∂ 2 u ∂ψ
∫
Ω
ρ
∂t 2
Ω+
δud
∫ ∂u δud Ω − δU
Ω
ext
=0
∂ψ
∫ ∂u dΩδu = 0
Ω
(3.72)
∂ψ
∫
Ω
∂σ
δσd Ω = 0
∂ψ
∫ ∂t dΩδt = 0
Ω
which leads, with Γ denoting the section in which the displacement is likely to occur and
by remembering that neither u nor t depends on the position x (they are only defined at the
position x Γ of the section Γ), to the set of Euler–Lagrange equations of a concrete 1D struc-
ture written in Table 3.6.
The first equation in Table 3.6 enforces a satisfying level of global force equilibrium
(either for tension or compression); the second equation gives the condition of compatibility
between the continuum σ and discrete t stresses; the third and fourth equations correspond
to the weak form of the local continuum and discrete constitutive models.
One can take advantage of the dependency between the continuum and the discrete
 dϕ 
stresses that appear in Table 3.6: 
the Lagrangian in the form:  dx
Ω
 ∫
σ − tδ x  d Ω = 0 ⇒ t = t(σ) and give expressions of
  ∂u   1  
∫ ψ dΩ = ∫  σ  ∂x − ε
Ω
+
Ω
v
− ε p  +  −tu p − tDt + Ξd  δ x  d Ω
  2  
(3.73)
  ∂u p  1 −1 p 
∫ ψ dΩ = ∫
Ω
−
Ω
v p p d p
 σ  ∂x − ε − ε  + Ξ + Λ + Ξ +  −tu − 2 tE t + Ξ  δ x  d Ω
Table 3.6 Euler–Lagrange Equations of a Concrete 1D Structure

Tension Compression
 ∂2u ∂δu   ∂2u ∂δu 

∫  ρ ∂t
Ω
2
δu + σ d Ω − δU ext = 0
∂x  ∫  ρ ∂t
Ω
2
∂x 
 dϕ   dϕ 

 dx
Ω
∫σ − tδ x  d Ωδ u = 0


∫
 dx
Ω
σ − tδ x  d Ωδ u = 0

 ∂u dϕ   ∂u dϕ 
∫ δσ  ∂x − dx u − ε
Ω
v
− ε p dΩ = 0
 ∫ δσ  ∂x − dx u − ε
Ω
v
− ε p dΩ = 0

∫ (u − u
Γ
p
)
− Dt d Γδt = 0
∫ (u − u
Γ
p
)
− E −1t d Γδt = 0
Moreover, the response of the localization section is supposed to be infinitely rigid before
the stress reaches an ultimate value and therefore E −1 → 0. Which, finally, denotes A Γ as the
area of the section Γ where displacement localization can take place, the simplified form of
these equations is written in Table 3.7. Only two unknown fields remain: the smooth dis-
placement u (x, t) and the continuum stress σ(x, t).
3.5.2 Numerical Examples
As in the previous examples, this numerical implementation was performed using the gen-
eral purpose finite element computer program FEAP. Due to the fact any external distrib-
uted loading is involved (self-weight is, in particular, neglected) in these applications, the
concrete structure is always modeled with constant stress bar elements. Figure 3.23 shows
the 1D structure that was tested.
Neglecting the concrete mass m = ρAL as compared to the added mass M, the stiffness k,
the fundamental pulsation w and period T of this structure is computed as:
ES k 2π
k= = 1.4 GPa; ω = = 44.7 s −1; T = = 0.14 s (3.74)
L m ω
3.5.2.1 Identification of a Concrete Law

All the parameters of the 1D cyclic concrete model presented here have a clear physical
meaning. It will be shown that (i) the viscous parameter η is related to the material critical
damping ratio ξ, (ii) the tensile softening law parameter a is related to the fracture energy
Gf and (iii) all the other parameters can be directly identified from the experimental stress–
strain response of a concrete specimen in quasi-static cyclic loading.
Table 3.7 Final Euler–Lagrange Equations of a Concrete 1D Structure

Tension Compression
 ∂2u ∂δu   ∂ 2u ∂δu 

∫  ρ ∂t
Ω
2
∂x  ∫  ρ ∂t 2 δu + ∂x σ dΩ − δU = 0
ext
∫
Ω
 ∂u
δσ 
 ∂x (
− ε v − ε p − u p + Dt
dt 
)
δ x dΩ = 0
dσ 
Ω
 ∂u
∫ δσ  ∂x − ε v
− εp − up
dt 
δ x dΩ = 0
dσ 
1 dϕ 1 dϕ
with t = t( σ ) =
A ∫ dx σdΩ
Ω
with t = t( σ ) =
A ∫ dx σdΩ
Ω
Figure 3.23 1D concrete structure model used for the numerical applications. The cross-section area
S = 0.04 m2 , the length L = 1 m, the added mass M = 7.0 E5 kg, the volume mass ρ = 2400 kg/m3
and the concrete elastic modulus E = 35 GPa.
a) Identification of the Viscous Parameter η

The viscous parameter η can be interpreted as a material property that can be identified
according to, among others, the experimental results of free low-amplitude—so that the
structure remains elastic—vibration tests (Figure 3.24):
1. First, denoting ξ as the critical damping ratio and c as the structural viscous param-
eter, if one assumes that ξ << 1, one can write (Chopra 2001):
c = 2ξmω (3.75)
2. Then, the material viscous parameter η can be connected to structural parameter c
starting from the 1D local form of the equilibrium, and the Kelvin–Voigt model con-
stitutive equations:
∂σ ∂u d  ∂u 
ρu
+ = 0; σ = E + η   (3.76)
∂x ∂x dt  ∂x 
For a tested structure with constant cross-section area S and strain field ( ∂u / ∂x = u / L ),
the introduction of (3.76)1 into (3.76)2 and then the integration of the resulting local equa-
tion over the whole structure gives:
S ES
ρ

+ η u +
SL u u = 0 (3.77)
m
L
L

c k
3. The Kelvin–Voigt viscoelastic model exhibits a hysteretic—energy dissipative—behav-

ior in cyclic loading as illustrated in Figure 3.25 for excitation of the form with, except
for other indicated values, A = 0.3 mm, Ω= 157.08 s−1 and τ = 0.001 s. The amount of
energy dissipated per cycle Dcyc—the area of the elliptical loop—can be analytically
written as (Wang 2009):
Dcyc = πcΩA2 (3.78)

The main drawback of the Kelvin–Voigt model is that Dcyc is dependent on the
loading frequency Ω. It is indeed not realistic in particular for seismic excitations
Figure 3.24 Free-vibrations viscoelastic response of the tested structure for various critical damping ratios.
Figure 3.25 Hysteretic response of the viscoelastic model in quasi-static sine loading for several (a) viscous
parameters, (b) loading amplitudes and (c) forcing pulsations.
(Wang 2009). This dependency is only negligible in the cases where the response is
primarily represented by the resonant pulsation, that is Ω ω .
4. If it is supposed that a concrete structure is excited by an impulsion that is weak
enough not to active the nonlinear phenomena that are represented by the concrete
model presented in the previous sections, and that, assuming that the response is pri-
marily represented by the fundamental pulsation ω, a critical damping ratio ξ = 1% is
measured (for instance thanks to the logarithmic decrement), then η can be seen as a
material property and identify it according to the following relation:
2ξmωC
η= (3.79)
S
In the case of a tested structure identical to that in Figure 3.23, the following is
computed:
η
η = 15.7 MPa; τ = ≅ 4.5 e −4s (3.80)
E
b) Identification of the Tensile Softening Parameter a

The tensile softening parameter a can be computed according to the fracture energy Gf. Gf
is one of the parameters that characterizes the softening stress-separation (crack-opening)
curve of the cohesive crack model (Hillerborg, Modéer and Petersson 1976; Bažant, Yu
and Zi 2002) and represents the total amount of energy that has to be furnished in tension
to a section between the time when displacement begins to localize and the time when the
local softening problem becomes (numerically) ill-posed, because K d < 0; physically, tcri has
to coincide with the time when the section is completely broken: when all the energy Gf is
consumed. In the case of this concrete model, one can demonstrate that tcri is always defined
and that:
tcri
t t
= σ u ⇒ a = σ u (3.81)
Gf =
∫
tloc
tudt
2a 2Gf
In addition, there is an empirical relation (Bažant and Becq-Giraudon 2002) that allows for
computing Gf according to material characteristics:
0.46 0.22 −0.30
 σcu   da   w
Gf = 2.5α 0   1 +    (3.82)
 0.051  11.27  c
with σcu in MPa and where α 0 = 1 for rounded aggregates, α 0 = 1.44 for crushed or angu-
lar aggregates, d a is the maximum aggregate size in mm and w/c is the water-cement ratio.
Finally, for concrete with: α 0 = 1.44, da = 25 mm, w / c = 0.5 , computed fracture energy is
Gf = 136 N/m that leads, from (3.80) with σtu = 3MPa, to:
a = 22060 m −1 (3.83)
c) Identification of the Other Parameters

The remaining parameters of the model can be identified from the stress–strain experimen-
tal response in Figure 3.21 (a).
1. Identify Young’s modulus for tensile and compressive parts. It is considered that they
have the same value: E = E = 35 GPa.
2. Identify the set of stresses so as to characterize the changes in the slope of the backbone
curve. It is chosen that: σ y = 3 MPa, σ f = 37 MPa, σcu = 49 MPa, σtu = 0.6 ⋅ σcu = 3 MPa
σy = 3 MPa, σf = 37 MPa, = 49 MPa and = 0.6 · = 3 MPa.
3. Simultaneously identify K p and H p to describe both the strain hardening phase of the
backbone curve in the range σ y < σ < σ f and the shape of the local hysteresis loops.
4. Identify K d so as to describe the remaining strain hardening phase of the backbone
curve in the range of σ f < σ < σcu .
5. Identify K p so as to describe the softening part of the backbone curve.
Figure 3.21 (b) is plotted with the parameters indicated above. Note that although all
the parameters have a clear physical meaning, they are also all connected and the results
of the first identification process often thus need to be refined. Indeed, the slope of the
hardening part of the backbone curve in the range of σ y < σ < σ f is C1 = (EK p ) / (E + K p ),
in the range of σ f < σ < σcu is C2 = (C1K d ) / (C1 + K d )and the slope of the softening part of
the backbone curve is :C3 = (C2K p ) / (C2 + K p ). Note also that although we observed that
activating plasticity before damage leads to better identification, it is possible to invert the
roles of σy and σf .
d) Mesh Objectivity
In the formulation of this model, we introduced a strong discontinuity—a displacement
jump—and thus concentrated the localization in a zero-length zone (Ibrahimbegović and
Brancherie 2003; Simo, Oliver and Armero 1993). This method leads to a formulation that
does not need any characteristic length and thus satisfies the mesh objectivity requirement,
that is the unicity of the solution.
e) Loading-Rate-Sensitive Response
The four main effects of the loading strain rate on the response of concrete are: the increases of
(i) the compressive strength (Bischoff and Perry 1995; Lu and Xu 2004), (ii) the tensile strength
(Weerheijm and Van Doormaal 2007), (iii) Young’s modulus (Weerheijm and Van Doormaal
2007) and (iv) the brittle behavior (Dilger, Koch and Kowalczyk 1984). For seismic excitations,
the strain rates are comprised between 10−5 1/s and 10−2 1/s, which respectively corresponds
to an increase of (i) the compression strength between 0 and 30% (Bischoff and Perry 1995;
Lu and Xu 2004), (ii) the tensile strength between 0 and 60% (Weerheijm and Van Doormaal
2007) and (iii) Young’s modulus between 0 and 10% (Weerheijm and Van Doormaal 2007).
The numerical applications presented in Figure 3.26(a) show that Young’s modulus com-
puted for a loading strain rate of 10 −2 1/s is about 5.7% larger than the one computed for
a loading strain rate of 10 −5 1/s. The increase of Young’s modulus is thus represented by
the model but a little underestimated here. However, this increase is related to the viscous
parameter τ computed in these numerical applications from the damping ratio ξ=1% whose
value has been arbitrarily chosen: one can thus hope that, in reality, the right value for the
damping ratio would lead to correct modeling of the increase of elastic stiffness. The com-
pressive response is presented in Figure 3.26(b). The proposed concrete model has not yet
been developed to represent the rate-dependent tensile and compressive strengths by itself.
This can be done in a simplified manner with the commonly accepted practice in engineer-
ing where the concrete strength parameters are majorated for seismic analysis.
3.5.3 Numerical Examples Seismic Application

a) Global Seismic Response
It is only the compressive part of the concrete law that is in focus. Thus, a static compres-
sive normal force in imposed to the bar before the seismic excitation. In the absence of
Figure 3.26 Tensile (a) and (b) compressive concrete response for several loading rates.
steel, once the strength degradation process has begun the structure is no longer capable of
resisting the seismic force demand and global equilibrium can no longer be satisfied; there-
fore, it is necessary to couple the concrete bar to an elastic steel bar with cross-section area
Ss = 0.02 ⋅ Sc = 8e −4m2 and Young’s modulus Es = 210 GPa. The behavior law for concrete is
that presented in Figure 3.20(b). The loading pattern is shown in Figure 3.27 and the global
structural response in Figure 3.28.
b) Local Response and Intrinsic Dissipated Energies

TIV provides a useful framework for quantifying the material intrinsic dissipation. Indeed,
the amount of continuum viscous, plastic and damage energies dissipated throughout the
structure and the discrete plastic and damage energies dissipated in the section where dis-
placement localizes can be computed according to (3.59)
τ τ
Ev =
∫∫
0 Ω
 v d Ωdt =
∫ ∫ σ ε dΩdt
0 Ω
v v
∫ ∫ (σε )
τ τ

∫∫

p
E =  pd Ωdt = p
+ q p ξ p + κ pλ p d Ωdt
0 Ω 0 Ω
τ τ
1 p
∫∫ ∫ ∫  2 σDσ + q ξ  dΩdt

d
E =  d d Ωdt = p
(3.84)
0 Ω 0 Ω
τ τ
 p p p
Ep =
∫∫
0 Ω
 pd Ωdt =
∫∫
0 Γ
 tu + q ξ  d Γdt
τ τ
1 d
d
E =
∫∫
0 Ω
 d d Ωdt =
∫ ∫  2 tDt + q ξ  dΓdt
0 Γ
d
Note that the total intrinsic dissipation is composed of a volume and a surface part in the
section Γ where displacement will localize.
Figure 3.27 Loading pattern: static loading + seismic loading time-histories.

Figure 3.28 (a) Right-node displacement and (b) reaction-displacement time-histories.
Figure 3.29 shows the local response of material points located outside and inside Γ: the
localization of the displacement in Γ leads to a larger amount of dissipated energy in this
section. The amount and source of the intrinsic energy dissipation within the seismic exci-
tation are detailed in Figure 3.29. Around the time t = 11 s, a big amount of seismic energy
is imparted to the structure, so as this latter does not collapse, it has to dissipate this input
energy. The input energy is converted both into stored energy in the material and dissipated
energy (irreversible mechanisms). Figure 3.30 illustrates the amount and sources of intrinsic
energy dissipation within the seismic excitation.
Taking into account the material identification test for concrete (Ramtani 1990, see
Figure 3.12, UPMC thesis on experimental identification of dynamic response on concrete),
allowed us to suggest the most appropriate model combining viscosity, plasticity, damage
and softening. As seen the model, it can capture both the change of elastic response (due
to damage) and residual deformation (due to plasticity) in both the hardening and soften-
ing phase. More importantly, the model can capture the three salient phases of dynamic
response in an earthquake, with small-amplitude vibrations for minor earthquakes, strong
Figure 3.29 Local response of a material point (a) located and (b) not located in section Γ where displace-
ment localizes.
Figure 3.30 Continuum plastic and damage, and discrete plastic intrinsic dissipated energies time-histories.
shaking for a major earthquake, followed by free-vibrations of a damaged structure in a

post-earthquake regime (Figure 3.31).
It is important that the proposed inelastic structural model and additional damping model
is able to represent the salient phenomena corresponding to each of these three phases.
Phase 1 represents an elastic phase with no or only a few incursions passing in the inelas-
tic domain. This means that the energy dissipation comes from the friction in the cracks
that appeared when applying a dead load and from other mechanisms always present in
mechanical systems. The intensified ground motion is seen in phase 2 where a large amount
of seismic energy is imparted to the structure and some parts of the structure then exhibit
inelastic behavior. The part of the energy that is not accounted for in the inelastic structural
model is introduced with the additional damping model. In the third and final phase, the
structure suffered irreversible degradations that modified its dynamic properties. Due to
the formation of cracks and friction in the cracks that were formed during the second phase
the energy dissipative mechanisms are different.
Our strategy consists of providing a comprehensive coupled plasticity-damage model
with both hardening and softening, where the evolution of all the different mechanisms is
defined through the corresponding failure (plasticity or damage) criterion. For simplicity,
Figure 3.31 Three-phase response (Jeleh, Ibrahimbegović and Leger 2012).

the hardening/softening moduli are kept constant, and the identification problem seeks to
address the same amount of dissipation in a typical hysteresis loop (Figure 3.32).
Another judicious choice concerns the hybrid-stress formulation of the proposed model,
which allows for the direct construction of the elasto-visco-plastic-damage compliance,
with the main advantage in its additive structure (format), which can easily be inverted to
obtain the corresponding value of the stiffness (e.g. 3.85).
 ct M e G e  d(ni)+1  R(di,)n +1 
 e,T  = ⇒
G − Hˆ  βn +1   0  (3.85)
K r , tan(i) = ct M e + G e Hˆ −1G e,T

Such computational strategy leads to very robust computations, with local computations of
internal variables governing the evolutions of different failure modes would need only one
iteration for each of them. Clearly, such computational strategy leads to very robust local
computations.
The proposed model was implemented in a multi-fiber beam for reinforced concrete struc-
tures (see Figure 3.33), leading to typical results. The fiber element and the constitutive
behavior law were implemented in the finite element computer program FEAP (2002). Fibers
were defined as a regular mesh of rectangles.
It is interesting to note that the proposed coupled damage-plasticity model can also be
brought in correspondence with a dynamic response at the structural level typical of viscous
Figure 3.32 Dissipation of energy (Jeleh, Ibrahimbegović and Leger 2012).
Figure 3.33 Multi-fiber beam for reinforced concrete structures.

damping (with an exponential-type of decay of the vibration amplitude), if we account for

heterogeneities common to concrete structures (Figure 3.34), which is defined as a lognor-
mal random variable leading to spatial variability in the material behavior law. Although the
yield stress is considered heterogeneous, the loading and unloading slopes remain constant
throughout (Jehel and Cottereau 2012).
Figure 3.35 shows the displacement time-history of the free-end of the cantilever beam
for three simulations of the same beam model but with a different variance parameter of the
lognormal random field describing the spatial variability of σ y ( x ) and σu ( x ) . It can be seen
from Figure 3.35 that almost all the imparted energy is dissipated after 14 s when τ = 1.06.
This model is able to well represent the material energy dissipated during seismic loading.
This kind of analysis leads to very robust results and a short computational time.
Figure 3.34 Local elastoplastic behavior law with linear kinematic hardening. The yield stress varies accord-
ing to the position in the beam.
Figure 3.35 Dynamic response in free-vibration: free-end displacement time-history (Jehel and Cottereau 2012).
REFERENCES
Armero, F. 1997. “Localized Anisotropic Damage of Brittle Materials”. In: Owen D. R. J., Hinton
E. (eds). Onate E. COMPLAS — Computational Plasticity: Fundamentals and Applications.
Proceedings of the 5th International Conference, Spain Barcelona. 635–640.
Bažant, Z. P., Q. Yu and G. Zi. 2002. “Choice of Standard Fracture Test for Concrete and Its
Statistical Evaluation”. International Journal of Fracture 118: 303–337.
Bažant, Zdeněk P. 1986. “Mechanics of Distributed Cracking”. Applied Mechanics Reviews 30 (5):
675–705. doi:10.1115/1.3143724.
Bažant, Zdeněk P and Emilie Becq-Giraudon. 2002. “Statistical Prediction of Fracture Parameters of
Concrete and Implications for Choice of Testing Standard”. Cement and Concrete Research 32
(4): 529–56. doi:10.1016/s0008-8846(01)00723-2.
Belytschko, T. and T. Black. 1999. “Elastic Crack Growth in Finite Elements with Minimal
Remeshing”. International Journal for Numerical Methods in Engineering 45 (5): 601–20.
doi:10.1002/(sici)1097-0207(19990620)45:5<601::aid-nme598>3.0.co;2.
Belytschko, Ted, Chandu Perimi, Nicolas Moës and Shuji Usui. 2001. “Arbitrary Discontinuities in
Finite Elements”. International Journal for Numerical Methods in Engineering 50 (4): 993–1013.
doi:10.1002/1097-0207(20010210)50:4<993::aid-nme164>3.0.co;2-m.
Belytschko, Ted, Chandu Perimi, Nicolas Moës, S. Sukumar and Shuji Usui. 2002. “Structured
Extended Finite Element Methods for Solids Defined by Implicit Surfaces”. International
Journal for Numerical Methods in Engineering 56 (4): 609–3.
Bischoff, Peter H. and Simon H. Perry. 1995. “Impact Behavior of Plain Concrete Loaded in
Uniaxial Compression”. Journal of Engineering Mechanics 121 (6): 685–93. doi:10.1061/
(asce)0733-9399(1995)121:6(685).
Brancherie, D. and Adnan Ibrahimbegović. 2009. “Novel Anisotropic Continuum‐discrete Damage
Model Capable of Representing Localized Failure of Massive Structures”. Edited by Adnan
Ibrahimbegović. Engineering Computations 26 (1/2): 100–127. doi:0.1108/02644400910924.
Chen, W. F. 1982. Plasticity in Reinforced Concrete. New York: McGraw-Hill.
Chopra, Anil K. 2001. Dynamics of Structures: Theory and Applications to Earthquake Engineering.
Upper Saddle River: Prentice-Hall.
Dilger, W. H., R. Koch and R. Kowalczyk. 1984. “Ductility of Plain and Confined Concrete Under
Different Strain Rates”. ACI Journal Proceedings 81 (1). doi:10.14359/10649.
Dominguez, Norberto, Delphine Brancherie, Luc Davenne and Adnan Ibrahimbegović. 2005.
“Prediction of Crack Pattern Distribution in Reinforced Concrete by Coupling a Strong
Discontinuity Model of Concrete Cracking and a Bond‐slip of Reinforcement Model”.
Engineering Computations 22 (5/6): 558–82. doi:10.1108/02644400510603014.
Dujc, Jaka, Bostjan Brank, Adnan Ibrahimbegović and Delphine Brancherie. 2010. “An Embedded
Crack Model for Failure Analysis of Concrete Solids”. Computers & concrete 7 (4): 331–46.
doi:10.12989/cac.2010.7.4.331.ž.
Garikipati, Krishna and Thomas J. R. Hughes. 1998. “A Study of Strain Localization in a Multiple
Scale Framework — The One-Dimensional Problem”. Computer Methods in Applied Mechanics
and Engineering 159 (3–4): 193–222. doi:10.1016/s0045-7825(97)002.
Halphen, B. and Q. S. Nguyen. 1975. “On generalized standard materials (in French)”. J. de
Mecanique 14: 39–63.
Hillerborg, A., M. Modéer and P. E. Petersson. 1976. “Analysis of Crack Formation and Crack
Growth in Concrete by Means of Fracture Mechanics and Finite Elements”. Cement and
Concrete Research 6 (6): 773–81. doi:10.1016/0008-8846(76)90007-7.
Hofstetter, G. and H. A. Mang. 1995. Computational Mechanics of Reinforced Concrete Structures.
Braunschweig: Vieweg.
Ibrahimbegović, A. and D. Brancherie. 2003. “Combined Hardening and Softening Constitutive
Model of Plasticity: Precursor to Shear Slip Line Failure”. Computational Mechanics 31 (1–2):
88–100. doi:10.1007/s00466-002-0396-x.
Element Solution Methods. Berlin: Springer.
Ibrahimbegović, Adnan and Edward L. Wilson. 1992. “Unified Computational Model for Static
and Dynamic Frictional Contact Analysis”. International Journal for Numerical Methods in
Engineering 34 (1): 233–47. doi:10.1002/nme.1620340115.
Ibrahimbegović, Adnan and E. L. Wilson. 1991. “A Modified Method of Incompatible Modes”.
Communications in Applied Numerical Methods 3 (7): 187–94. doi:10.1002/cnm.1630070303.
Ibrahimbegović, Adnan and François Frey. 1993a. “Stress Resultant Finite Element Analysis of
Reinforced Concrete Plates”. Engineering Computations 10 (1): 15–30. doi:10.1108/eb023892.
Ibrahimbegović, Adnan and François Frey. 1993b. “An Efficient Implementation of Stress Resultant
Plasticity in Analysis of Reissner–Mindlin Plates”. International Journal for Numerical
Methods in Engineering 36 (2): 303–20. doi:10.1002/nme.
Ibrahimbegović, Adnan and Sergiy Melnyk. 2007. “Embedded Discontinuity Finite Element Method
for Modeling of Localized Failure in Heterogeneous Materials with Structured Mesh: An
Alternative to Extended Finite Element Method”. Computational Mechanics 40 (1): 149–55.
doi:10.1007/s00466-006-0091-4.
Ibrahimbegović, Adnan, Pierre Jehel and Luc Daven. 2007. “Coupled Damage-Plasticity Constitutive
Model and Direct Stress Interpolation”. Computational Mechanics 42 (1): 1–11. doi:10.1007/
s00466-007-0230-6.
Jehel, P. and R. Cottereau. 2012. “On damping Created by the Heterogeneity of the Mechanical Properties
in RC Frame Seismic Analysis”. 15th World Conference on Earthquake Engineering. Lisbon.
Jeleh, P, A Ibrahimbegović and P Leger. 2012. “Efficient Inelastic Fiber Frame Element and Consistent
Damping Model for Seismic Time-History Analyses”. 15th World Conference on Earthquake
Engineering. Lisbon. 1–9.
Jirásek M. and T. Belytschko 2002. “Computational Resolution of Strong Discontinuities”. WCCM —
World Congress on Computational Mechanics. Vienna, Austria. 1–7.
Jirásek, Milan and Thomas Zimmermann. 2001. “Embedded Crack Model. Part II: Combination
with Smeared Cracks”. International Journal for Numerical Methods in Engineering 50 (6):
1291–305. doi:10.1002/1097-0207(20010228)50:6<1291::aid-nme1>.
Kucerova, A., D. Brancherie, J. Zeman and Z. Bitt. 2009. “Novel Anisotropic Continuum‐Discrete
Damage Model Capable of Representing Localized Failure of Massive Structures”.
Edited by Adnan Ibrahimbegović. Engineering Computations 26 (1/2): 128–44.
doi:10.1108/02644400910924834.
Lu, Yong, and Kai Xu. 2004. “Modelling of Dynamic Behaviour of Concrete Materials Under
Blast Loading”. International Journal of Solids and Structures 41 (1): 131–43. doi:10.1016/j.
ijsolstr.2003.09.019.
Markovic, Damijan and Adnan Ibrahimbegović. 2006. “Complementary Energy Based FE Modelling
of Coupled Elasto-Plastic and Damage Behavior for Continuum Microstructure Computations”.
Computer Methods in Applied Mechanics and Engineering 195 (37–50): 5077–93. doi:10.1016/j.
cma.2005.05.058.
Melenk, J. M. and I. Babuška. 1996. “The Partition of Unity Finite Element Method: Basic Theory and
Applications”. Computer Methods in Applied Mechanics and Engineering 139 (1–7): 289–314.
doi:10.1016/s0045-7825(96)01087-0.
Mivelaz, P. 1996. Waterproofness of Reinforced Concrete Structures: Leaking Through the Cracks.
Ph.D. Thesis (in French). Lausanne, Switzerland: EPFL.
Oliver, J. 2000. “On the Discrete Constitutive Models Induced by Strong Discontinuity Kinematics
and Continuum Constitutive Equations”. International Journal of Solids and Structures 37
(48–50): 7207–29. doi:10.1016/s0020-7683(00)00196-7.
Oliver, J., A. E. Huespe and P. J. Sánchez. 2006. “A Comparative Study on Finite Elements for
Capturing Strong Discontinuities: E-FEM vs. X-FEM”. Computer Methods in Applied
Mechanics and Engineering 195 (37–40): 4732–52. doi:10.1016/j.cma.2005.0.
Oliver, J. and A. E. Huespe. 2004. “Theoretical and Computational Issues in Modelling Material
Failure in Strong Discontinuity Scenarios”. Computer Methods in Applied Mechanics and
Engineering 193 (27–29): 2987–3014. doi:10.1016/j.cma.2003.08.007.
Panoskaltsis, V. P., K. D. Papoulia, S. Bahuguna and I. Korovajchuk. 2008. “The Generalized Kuhn
Model of Linear Viscoelasticity”. Mechanics of Time-Dependent Materials 11 (3–4): 217–30.
doi:10.1007/s11043-007-9044-3.
Pedersen, R. R., A. Simone and L. J. Sluys. 2008. “An Analysis of Dynamic Fracture in Concrete with
a Continuum Visco-Elastic Visco-Plastic Damage Model”. Engineering Fracture Mechanics 75
(13): 3782–805. doi:10.1016/j.engfracmech.2008.02.
Perzyna, P. 1966. “Fundamental Problems in Viscoplasticity”. Advances in Applied Mechanics
(Academic Press) 9: 243–377.
Ramtani, S. 1990. Contribution to the modeling of the multi-axial behavior of damaged concrete
with description of the unilateral characteristics. Ph.D. thesis, Paris 6: University.
Simo, J. C., J. G. Kennedy and R. L. Taylor. 1989. “Complementary Mixed Finite Element
Formulations for Elastoplasticity”. Computer Methods in Applied Mechanics and Engineering
74 (2): 177–206. doi:10.1016/0045-7825(89)90102-3.
Simo, J. C., J. Oliver and F. Armero. 1993. “An Analysis of Strong Discontinuities Induced by Strain-
Softening in Rate-Independent Inelastic Solids”. Computational Mechanics 12 (5): 277–96.
doi:10.1007/bf00372173.
Stein, E. and D. Tihomirov. 1999. “Anisotropic damage-plasticity modelling of reinforced concrete”.
ECCM’99 — European Conference on Computational Mechanics. Munchen, Germany. 1–19.
Wang, Judith. 2009. “Intrinsic Damping: Modeling Techniques for Engineering Systems”. Journal of
Structural Engineering 135 (3): 282–91. doi:10.1061/(asce)0733-9445(2009)135:3(282).
Weerheijm, J. and J. C. A. M. Van Doormaal. 2007. “Tensile Failure of Concrete at High Loading
Rates: New Test Data on Strength and Fracture Energy from Instrumented Spalling Tests”.
International Journal of Impact Engineering 34 (3): 609–26.
Zienkiewics, O. C. and R. L. Taylor. 2005. The Finite Element Method. 6th ed. New York: Elsevier/
Chapter 4
Masonry structures
Masonry structures are defined as structures made from individual units of different shapes,
sizes and layouts (bricks, stones, marble, granite, travertine, limestone, cast stone, concrete
block etc.), with or without mortar as a bonding material. The masonry structures are
well known for being easy to construct and less costly compared to steel and reinforced
concrete structures, as well as their benefits in respect to thermal insulation, fire protec-
tion, low maintenance, widespread geographic availability and durability. This method of
construction is the oldest (Abrams 2001) and simplest; however, one which is still the least
understood in terms of strength and deformation characteristics even under static loading.
The development of design rules for masonry structures have not kept pace with the
advance of other civil engineering materials, such as reinforced concrete and steel as they
became more competitive, and masonry has been put aside. There were several reasons for
this. First, new buildings in developed countries are mainly made of reinforced concrete
and steel; second, there is a lack of insight into and models for the complex behavior of
masonry units, mortar, joints and masonry as a complex material. However, the situation is
completely different in developing countries where structural masonry is still widely used.
Additionally, in the last decades, the need for preservation of old historic buildings has
increased in the sense that old buildings have intrinsic value; they are reminders of a city’s
culture and complexity and attract people and business. This all has had an influence on the
research on masonry structures. It is clear that research has to be two-sided, experimental
and numerical which is the basis for validation and further improvement of existing models.
Besides the checking of the ULS (ultimate limit state) and SLS (serviceability limit state) in
the new masonry structures assessment, strengthening of existing masonry structures, espe-
cially historical monuments, have opened a new phase in the research of masonry structures
and their modeling.
The complexity of these structures is even more emphasized in conditions of seismic load-
ing. Characteristics of masonry structures depend on their individual components (units
and mortars) as well as the interface between the units and mortar joints. Additionally, the
behavior of head and bed joints are different. This all results in modeling difficulties and
still remains an open challenge for engineers due to the nature of masonry structures, which
are characterized by complex and particularly nonlinear behavior. The nonlinear behavior
of masonry walls is highly complicated. Nonlinearity is mainly dependent on the cracking
and crushing of masonry, which is rather heterogeneous and anisotropic. Softening and
dilatancy play a curial rule in the nonlinear processes (Lourenço 1996). For one to define
the nonlinear behavior of a masonry wall, it is necessary to first define the failure modes. It
is well known that there are four primary in-plane failure modes: rocking, bed joint sliding,
diagonal tension failure along masonry units or along the head and bed joints in a stair-
stepped fashion and toe-crushing. These four primary in-plane failure modes are not suf-
ficient to define the inelastic behavior of masonry wall, and additional coupled failure modes
111
are needed, as defined in FEMA 306. Through experimental studies, it has been seen that
different failure modes exhibit different hysteretic energy dissipation, displacement capacity
and strength degradation. From the experimental results, it was observed that the aspect
ratio (height/length) and vertical stress are the most important factors that have a direct
impact on the type of the failure mode of the masonry walls.
4.1 COMPUTATIONAL MODELING OF MASONRY STRUCTURES
It is only in recent years that numerical modeling techniques based on the finite element
method have emerged for the analysis of masonry structures. The complexity lies in the
strong nonlinear behavior of masonry; modeling needs to be done by two- (2D) or three-
dimensional (3D) elements which makes it more complex in regard to steel and concrete
structures. The main problem is the development of robust and reliable models capable of
predicting the nonlinear behavior of such structures.
Masonry represents a composite material made of units and mortar joints. Modeling,
depending on the required accuracy and desired simplicity, can have different modeling
strategies (Lourenço, Computer Strategies for Masonry Structures 1996):
• Detailed micro-modeling—units and mortar in the joints are represented by continuum

elements whereas the unit–mortar interface is represented by discontinuous elements;
• Simplified micro-modeling—expanded units are represented by continuum elements
whereas the behavior of the mortar joints and unit–mortar interface is lumped in dis-
continuous elements;
• Macro-modeling—units, mortar and unit–mortar interface are smeared out in the
continuum.
More details regarding macro-modeling can be found in Rots (1991) and for micro-model-
ing, either detailed or simplified, see Lourenço (1996). Preferring one model over the other is
not possible as they are to be used on different scales. Micro-modeling studies are necessary
for a better understanding of the local behavior of masonry structures, mainly if one is to be
interested in the strong heterogeneous states of stress and strain. On the other hand, macro-
models are applicable when the structure is composed of solid walls with sufficiently large
dimensions, so that the stresses across or along a macro-length will be essentially uniform.
This type of modeling is most valuable when a compromise between accuracy and efficiency
is needed (Lourenço 1996).
Accuracy is proportional to model complexity and inversely proportional to computer
memory and time. In this respect, detailed micro-modeling is formulated in such detail that
its applicability can be seen only for small samples. In order to make this type of model-
ing more usable, a simplified model Figure 4.1(c)) was developed where the mortar joints
and interface are lumped into discontinuous elements, whereas the units remain as con-
tinuum elements as in Figure 4.1(b) but are further expanded. In this respect, masonry is
viewed as composed of elastic blocks that are bonded by potential fracture/slip lines at the
joints. Anthoine (1992), in his work, gave a very detailed review of different micro-modeling
attempts by different authors.
It should be mentioned that some authors have studied only the joint (Rosson and Suelter
2001; Nwofor 2011), and others investigated and modeled the blocks (Gambarotta and
Lagomarsino 1997; Guinea et al. 2000).
Finally, global structural behavior modeling at such a level of detail is impractical but
above all the interface modeling can be disregarded as it has been shown that its effect are
Masonry structures 113
Figure 4.1 Modeling strategies for masonry structures: (a) masonry sample; (b) detailed micro-modeling; (c)
simplified micro-modeling; (d) macro-modeling. (Lourenço 1996).
negligible. In this respect, many authors have turned to macro-modeling. Here, material is
referred to as an anisotropic composite, and in forming constitutive relations one deals with
average stresses and strains (Lourenço, Rots and Blaauwendraad 1998; Papa and Nappi
1997; Yan, Thambiratnam and Corderoy 1998). In order to describe the material behavior
to a full degree, orthotropic material needs to be implemented, which is very complex due
to the fact that only a few material models of this kind exist (Lourenço 1996; Anthoine
1992). If the structures are made in such a way that there is a rather regular pattern of units
and joints, then homogeneous models based either on plasticity (Lopez, Oller and Lubliner
1999; Ma, Hao and Lu 2001) or damage (Raimondo and Sacco 1997; Pegon and Anthoine
1997; Marfia and Secco 2001) have been seen as applicable.
In order to make an accurate model regardless of its scale material parameters are
required. It is quite obvious that the diversity of these characteristics is large; from material
properties of individual components, the arrangement of bed and head joints, their existence
or non-existence, interface between mortar and units, the anisotropy of units, quality of
workmanship, the degree of curing, atmospheric effects, age etc.
Due to all these reasons, the analysis and modeling of masonry structures currently rep-
resents one of the most important fields in civil engineering.
4.2 MICRO-MODELING OF MASONRY
Here, a micro-model for masonry will be presented. The starting point is the identifica-
tion of various local failure modes, as clearly performed by Lourenço and Rots (1997). As
illustrated in Figure 4.2a and b, which represent joint mechanisms, formerly associated with
joint tension cracking and later with sliding along the bed joint due to low values of normal
stress. Brick mechanisms are illustrated in Figure 4.2c represented by unit direct tension
cracking. Figures 4.2d and 4.2e represent the combined brick and joint mechanics. The for-
mer is associated with the diagonal tension cracking of masonry units at values of normal
stress sufficient to cause friction in the joints, and the latter is connected to masonry being
crushed as a result of mortar dilatancy at high values of normal stress.
Figure 4.2 Local modes of failure: (a) joint tension cracking, (b) joint slip, (c) unit direct tension crack, (d)
unit diagonal tension crack, and (e) masonry crack. (Lourenço and Rots 1997).
Based on the work of Alfano, Rosati and Valoroso (2000), it seems that these patterns are
of great importance in the search for a predictive model. However, the softening behavior
of materials led to some difficulties, such as the lack of digital objectivity. In Lourenço and
Rots (1997), softening in compression is located in the mortar joints and in the brick ele-
ments; pure tension cracks in the units is placed vertically in the middle of each unit and
using interface elements to model the joints this has given good results in comparison to the
tests that were carried out. The approach that will be presented here models the two mate-
rials using continuous elements, while reintegrating the incompatible modes in the blocks;
this is done using strong displacement discontinuities and thus solving the problem of digital
objectivity. In this respect, the model is explained below.
4.3 STRAIN LOCALIZATION PHENOMENA
Softening and localization have a clear physical origin in microstructure evolution even
though they are observed in a macroscopic material response. Materials that exhibit soften-
ing (concrete, masonry, soli-cohesive materials) are usually described with a stress–strain
relation in numerical analysis. Until now, it has been well known (de Borst 1999) that local-
ized deformations, closely related to material softening, is one of the primary difficulties
for classical continuum models when representing inelastic deformation and assuring mesh
independency. Here it leads to an ill-posed boundary value problem, since equilibrium or
motion equations do not remain elliptic or hyperbolic once the descending branch is entered.
The only way to define the localization is to set the measure to zero. This leads to mesh-
dependency. There are two ways to resolve this issue. One possibility is to concentrate the
deformation on interfaces or discrete cracks. Here, constitutive relations are a function of the
tractions on relative displacements of two interface sides. This is done using interface finite
elements (Rots 1988). Here, it is necessary to predict the location of the crack. The meshless
paradigm has provided new insights into the finite element method (Babuška and Melenk
1996; Duarte and Oden 1996; Fries and Belytschko 2010) by using so-called extended finite
elements, placed within the partition of unity framework. This partition of unity framework
(Babuška and Melenk 1996) is a powerful technique for modeling discontinuities and singu-
larities. This is done through element domains owing to an enrichment of the approximation
basis with Heaviside-type functions.
Here, a model which takes into account both micro and macrocracking is presented, as
this is physically closer to reality. Additionally, it interprets, in a correct manner, the nature
of the associated inelastic energy dissipation.
4.3.1 Modeling of In-Plane and Out-Of-Plane

Loading on a Brick Wall
4.3.1.1 Modeling of Failure Mechanisms in Bricks
The novelty of this model is that the nonlinear softening behavior of these elements is mod-
eled by introducing strong displacement discontinuities via jump functions (Oliver et al.
2002; Simo, Oliver and Armero 1993; Colliat, Davenne and Ibrahimbegovic, 2002). This
means that potential cracks in the units must always be included in micro-modeling in order
for the correct behavior of the masonry structures to be captured. In this respect, the dis-
placement field is written in such a way that it is allowed to have a discontinuity (crack or
opening), and it can be written as:
u ( x, t ) = u ( x, t ) + u ( t ) H x ( x ) (4.1)
where u ( x, t ) is the smooth part, H x ( x ) is the Heaviside function and u is the displace-
ment jump corresponding to the crack-opening.
The strain field obtained from (4.1) and the well-known relation from the kinematics
equation:
∂u ( x, t )
ε ( x, t ) = (4.2)
∂x
is therefore expressed by:
∂u ( x, t ) ∂u ( x, t )
ε ( x, t ) = = + u ( t ) δ x ( x ) (4.3)
∂x ∂x
where, δ x ( x ) is the Dirac-delta function.
The elastoplastic softening behavior is modeled using the theory of plasticity on the dis-
sipation surface; this defines an elastic field in the well-known form of:
( )
φ σ, q = t −( σy −q ) (4.4)
where t is the normal stress at the displacement discontinuity, q is the hardening variable and
σy is the yield stress, separating the elastic and plastic domains. If (4.4) is less or equal to zero,
this assumes that there is an elastic domain where no change of the inelastic deformation
will occur. Additive decomposition of the total deformations into elastic and plastic parts
is acceptable for small deformations in plasticity models (Lubliner 1990). It follows that the
stress depends on the elastic deformation part, while the plastic flow can be written as:
σ
ε ( x, t ) =
E
()
+ γ ( x, t ) sign σ (4.5)
where:
E is Young’s modulus and
γ is the plastic multiplier.
Once the results obtained in (4.3) are introduced in (4.5), and taking into account the
smoothness of the stress field, it is obvious that the following must be true:
γ ( x, t ) = γ ( x, t ) + γ ( x, t ) δ x ( x ) (4.6)
Going back to (4.6) and introducing (4.3) it follows that:
 ∂u ( x, t ) 
σ ( x, t ) = E  ()
−γ ( x, t ) sign σ  (4.7)
 ∂x 
Leading to:
γ ( t ) = u (4.8)
The consistency conditions follow from (4.4) and read:
∂u ( x, t )
( ) () ()
2
0 = φ σ, q = sign σ E −γ ( x, t ) E sign σ 
∂x  
(4.9)
(

()

x

)
+ q + E u sign σ −γ δ x ( x )
0
It is assumed that the micro and macro-cracks will appear subsequently; the first phase can
be defined as:
∂u ( x, t )
()
sign σ E
∂x
γ ( x, t ) = (4.10)
()
2
E sign σ  + K
 
From this it implies that the non-local form of the consistency condition leads to:
()
q = −sign σ σ (4.11)
indicating that the hardening stress-line variable has to be bonded. Taking into account
that q = −Kξ , ξ = γ , where K is the hardening modulus and ξ is the hardening variable.
Considering the equality set in (4.8) and the above mentioned qualities, it follows that the
second part in (4.6) becomes:
γ ( x, t ) δ x ( x ) = u δ x ( x ) = −K −1q (4.12)
This additional implies that:
K −1 = K −1δ x ( x ) (4.13)
This plasticity model is incorporated in the following manner: four node elements are con-
sidered and one discontinuity, meaning the potential crack is placed in the center of the ele-
ment and the interpolation is constructed as shown in Figure 4.3. In addition to the usual
interpolation of the displacement field, a form of the function called the “Incompatible
mode” M(x) is superimposed.
The potential crack is placed in the center of the real strain field and is constructed as
follows:
4
ε ( x, t ) = ∑ ∇ N ( x) d + G ( x) α (4.14)
a= 1
s
a a
where α is the parameter for incompatible mode, and G is produced from the chosen dis-
placement interpolation by applying the same differential operator to the smooth and dis-
continuous part.
The interpolation of the virtual strain is then constructed with:
4
δε ( x, t ) = ∑ ∇ N ( x) w + G ( x) β (4.15)
a= 1
s
a a
where Ĝ is a modified incompatible mode strain–displacement matrix with a zero-mean,

which should guarantee satisfaction of patch test conditions (Ibrahimbegović and Wilson
1991). As the same shape for the virtual strain field is chosen, the discrete formulation of
the problem is of the standard form of the equations of the incompatible mode method
(Ibrahimbegović and Wilson 1991), where the set of global equilibrium equations is replaced
by element based local equilibrium equations. The advantage of this approach is that that
the fracture energy Gf of the material corresponds to the given mode; meaning that the total
amount of dissipated energy is independent of the foreseen element sizes, and in this way the
problem of numerical objectivity is answered.
4.3.1.2 Modeling of Mortar Joints in a Brick Wall

The modeling of the interfaces between the unit elements (in this specific case bricks) in the
structural assemblies is of utmost importance. Its contribution is crucial to the strength of
Figure 4.3 Form functions for elements at the discontinuity.

masonry structures. Two issues have to be taken into account here: the size of the interface
elements in respect to the size of the structure and the weak mechanical properties of mortar
in existing structures. The complexities of modeling masonry structures are numerous: from
the irregularity of the structure, the constitutive equations for different materials (stone for
example is a rigid block), the presence or absence of mortar as well as the thickness of the
mortar etc. Until now either phenomenological modeling or deductive modeling was used.
In the former, the thickness of the interface is equated as zero (Lourenço 1996). This means
that the mortar and the adhesion surface between the brick and the mortar are modeled as
a unique interface. Whereas, in the later, interphase elements are modeled at the microme-
chanical level, mortar is specifically modeled as a continuum material and the joint (adhe-
sion between the mortar and the unit) by a specific interface (Sacco and Toti 2010).
In order to model the mortar between the two blocks, it is clear that the nonlinear behav-
ior of this zone is associated with two phenomena: one linked to the opening in tension,
and the other to shear. So, the important feature of the composite yield surface is the cou-
pling of tension and shear softening due to the fact that both phenomena are related to the
bond or adhesion between the brick units and the mortar. For the latter, one should also
note the importance of the dilatancy of the mortar joints as they have a low dilatancy so it
is necessary to formulate the model with non-associated plasticity. The dilatancy angle is
considered as a function of the plastic-relative shear displacement and the normal confining
pressure. Under the increasing values of these quantities, the dilatancy angle tends to zero.
This is physically realistic, given the microstructure of the unit–mortar interface and it being
confirmed by experiments. Also, softening behavior should be included for all the modes of
the composite yield surface.
The theory of plasticity is used as the basis for the development of this model (Colliat, Davenne
and Ibrahimbegovic 2002); the elastic range is defined by a multisurface (see Figure 4.4).
The same basis was used by Giuseppe and Di Gati (1997) where a simple, cohesive interface
Figure 4.4 The plasticity criterion for the joints.

model adopted a yield surface using a classical bi-linear Coulomb condition with a tension
cut-off and with a non-associative flow law.
A perfect plasticity model for joints has been chosen, and the constant dilatancy angle
is part of this approach. This treatment has been developed through the use of the strong
discontinuities of displacement under the incompatible modes method.
For the shear mode a Drucker–Prager criterion is implemented in the form:
φ1 = s + βI1 −cψ (4.16)
While a tensile limit for the opening of the joints is defined as:
φ2 = I1 −T ψ (4.17)
where I1 and s are the hydrostatic and deviatoric part of the stress tensor and β is the mate-
rial parameter which can characterize the internal friction.
All elements for the multisurface criteria are described in Non-Smooth Multisurface
Plasticity and Viscoplasticity. Loading/Unloading Conditions and Numerical Algorithms
(Simo, Kennedy and Govindjee 1988). The equal degradation of strength can be assumed as
both phenomena occur due to the breakage of the same bridge that exists at the micro-level
between the unit and mortar.
The law of plastic flow has the usual form:
m
∂φ*α
ε p = ∑
α= 1
γ α
∂σ
(4.18)
α
where γ are the different plasticity multiplications.
Indeed, it is shown in Simo, Kennedy and Govindjee (1988) that φαtrial > 0 does not imply
α
γ > 0 or which may have φαtrial > 0, but φα,n+1 > 0.
 ∂φ* 
In fact, γ α can be seen as the contravariant components of C : ∆ε np+1 with the base C, α  .
The condition γ α > 0° for α = 1, 2 defines a mode in the corner.  ∂σ 
a) Drucker Mode
One has:
∂φ1 β ∂φ*1 ψ s
= n + 1 and = n + 1 with n = (4.19)
∂σ 3 ∂σ 3 s
The consistency condition gives the value of the plastic multiplier in this mode:
φ1trial
γ1 = (4.20)
2G + βKψ
which now gives the first part of the new state of constraints:
I1,n+1 = I1trial −γ 1Kψ (4.21)
sn+1 = sntrial
+ 1 −2Gγ 1n (4.22)
and, the other part, the elastoplastic part being consistent (Simo and Taylor 1985) with:
2G sn+1 α
α= and α = K − (4.23)
sn+1 + 2Gγ 1 3
1
C ep = αI + α1 ⊗ 1 − α2n ⊗ n + αβKn ⊗ 1 + αψ K 1 ⊗ n + βK 2 ψ 1 ⊗ 1  (4.24)
α + βKψ 
b) Mixed Mode (Corresponding to Corner)

In addition to the above equations, the following expression is added:
∂φ 2 1
= 1 (4.25)
∂σ 3
The consistency condition can express both plastic multipliers:
s trial −c + βT I1trial −T
γ1 = and γ 2 = −ψγ 1 (4.26)
2G K
from which the new state of stress is obtained:
I1,n+1 = I1trial −Kψγ 1 −Kγ 2 (4.27)
sn+1 = sntrial
+ 1 −2Gγ 1n (4.28)
and the elastoplastic module reads:

α
C ep = α I − α 1 ⊗ 1 − α n ⊗ n (4.29)
3
So, the plastic flow rule used is:
φ*1 = s + ψ I1 (4.30)
and
∂φ*1 1
ε p = = n + ψ 1 (4.31)
∂σ 3
Most of the research was focused on the in-plane failure of masonry structures (Figures 4.5
and 4.6) due to earthquake actions.
It is necessary for the wall to have a sufficient capacity to withstand out-of-plane failure
as it has been seen in past earthquakes that it can trigger partial or complete collapse of
the structure. It has been observed in past earthquakes from Banja Luka, the Bosnia and
Herzegovina earthquake in 1969 (Figure 4.7) to the Christchurch earthquake (Figure 4.8)
in 2011 (Ingham and Griffith 2011). The lack of sufficient connections between wall to wall
and wall to floor results in out-of-plane damage and the failure of masonry members. This
is something that is characteristic of the old masonry structures being more vulnerable than
the new structures with rigid diaphragms. This means that even in cases where connections
are shown to be effective, this type of failure can be triggered due to the excessive vertical
Figure 4.5 In-plane failure—Christchurch (New Zealand) (Ingham and Griffith 2011).
Figure 4.6 In-plane failure—Structural damage to masonry building resulting from the 1994 Northridge,
California, earthquake (NAS 2003).
Figure 4.7 Out-of-plane failure—Banja Luka earthquake (Bosnia and Herzegovina).

Figure 4.8 Out-of-plane failure—Christchurch (New Zealand) (Ingham and Griffith 2011).
and/or horizontal slenderness of the wall. In this respect, the out-of-plane failure of masonry
is mainly related to the geometric stability rather than to the strength of the materials. In
these cases, the out-of-plane failure mechanisms are activated by lower values of ground
motion than the in-plane ones. An additional element that needs to be taken into account is
the question of the type of floors, with flexible or rigid diaphragms. Diaphragm flexibility
significantly increases the out-of-plane displacement response. Considerable experimental
work is also reported for the strength and response of brick walls under out-of-plane loads
(Griffith et al. 2007; Derakhshan, Ingham and Griffith 2009; Meisle, Elwood and Ventura
2007; Maheri, Najafgholipour and Rajabi 2011).
Even though there have been many attempts and recommendations in modern codes
for avoiding out-of-plane failures, it is still an existing problem and an important field of
research. Masonry is still an open area for further research and especially out-of-plane
failure which even now is not fully understood. As Paulay and Priestley (1992) state, out-
of-plane failure is “one of the most complex and ill-understood areas of seismic analysis”.
4.3.2 Examples
Traditionally, experiments on shear walls have been adopted by the masonry structures
research community as full scale in-plane tests. Results are compared with the ones obtained
in Lourenço and Rots (1997) and CUR (1994).
4.3.2.1 Example 1
The shear walls tested have a width/height ratio of 1 with dimensions of 990×1000 [mm 2],
and are built up with 18 courses, having 16 courses active while 2 courses are clamped in
steel beams. The walls are made of wire-cut solid clay blocks with dimensions of 210×52×
100 [mm 2] and the thickness of the mortar is 10 [mm].
First of all vertical pre-compression is applied and then the horizontal load is monolithi-
cally increased under top displacement as indicated in Figure 4.9. The boundary conditions
are formulated in such a way that both the bottom and top boundaries are horizontal, pre-
venting vertical movement and rotations.
The material properties are summarized in Table 4.1.
It should be noted that the choice of zero tensile strength is due to the use of a perfect plas-
ticity model; open joints will thus transmit the force. The dilatancy angle ψ is a measure of
the uplift of one unit over the other upon shearing. The average value of tan ψ falls into the
Figure 4.9 Shear walls tested (a) vertical loading—phase 1, (b) horizontal loading under displacement con-
trol-phase 2 (Lourenço 1996).

Bricks Joints
E(MPa) = 16700 E(MPa) = 20000
υ-Poisson ratio = 0.15 υ-Poisson ratio = 0.20
Compression (vertical) c = 0.35(MPa ) = tan ϕ = 0.75
s y = 8(MPa )
G c = 220( J/m2 )
Tension (horizontal) tan ψ =0.35 T = 0(MPa )
s y = 1(MPa )
Gf = 80( J/m2 )
range of 0.2 to 0.7 for low confining pressure, depending on the roughness of the unit sur-
face. With increasing slip, tan ψalso decreases to zero due to the smoothening of the sheared
surface (Lourenço and Rots 1997). As for the dilatancy angle, sensitivity tests have shown
that regarding normal stresses it has little influence, which agrees with Lourenço and Rots
(1997), where it has been seen that after the horizontal displacement of d = 7.5 mm the solu-
tion with tan ψ = 0 and tan ψ = 0.4 are similar because the residual compressive strength
controls the response.
The comparison between the experimental results and the model is given in Figure 4.10
shows the total horizontal total vs. displacement of the top of the wall. The proposed model
fits well with the experimental tests. It may be noted that the initial stiffness of the wall is
essentially correct without any need for the modification of the elastic parameters; this is an
advantage in view of a predictive model. Regarding the peak value of the force this is satis-
factory; however a greater displacement gives the wall a more ductile behavior by modeling
than it is observed in the experiment.
The failure is caused by crushing. In the reinforced concrete shear walls investigated
in Feenstra and De Borst (1996), it was seen that higher initial vertical loads lead to an
increase of strength but decrease of ductility, meaning an increase of brittleness. In shear
walls, the increase of confinement does not lead to an increase of ductility in the case where
the softening regime is generally governed by the failure of the compressed toes of the wall.
Figure 4.10 Comparison based on the CUR program testing; pre-compression of vertical loading 2 MPa.
As the material can always fall into the out-of-plane failure this can happen. However, in
masonry, shear wall experiments showed that higher initial loads lead to the increasing
strength of the shear walls, but higher failure loads did not correspond to increasing brittle-
ness. This can be explained by the internal force distribution at failure (for more informa-
tion see Lourenço (1996)).
4.3.2.2 Example 2
This test program was conducted by the Centre Scientifique et Technique du Bâtiment; the
masonry was made out of different types of burned brick elements and was unreinforced.
Here, only the case where the walls were loaded horizontally in their plane will be discussed.
The considered wall was made of brick elements having dimensions of 500×200×
200 [mm3], the mechanical characteristics of vertical compression are summarized in
Table 4.2. It was composed of 12 horizontal rows with seven bricks in each row. Details on
the tests can be found in Capra et al. (2001).
The calculation was carried out by imposing horizontal displacement on the upper part
of the wall and without any possible rotation thereof. The results are plotted in Figure 4.11.
The obtained numerical results were reasonably consistent with the measurements obtained
from experimental tests.

Bricks
E = 2300(MPa )
υ = 0.12 -Poisson ratio
s c = 2.70(MPa )
Figure 4.11 Comparison based on the CSTB program testing and numerical model.
The first pseudo-linear phase coincided with the nonlinear behavior of the joints (open-
ing and shear); it was observed that the initial stiffness of the wall was correct, and had an
absence of timing parameters (elastic and inelastic for the brick joints).
The failure of the wall occurred with the crushing of the bricks, mainly in the two corners,
which was seen in the experiments. The experimental behavior was satisfactorily reproduced
and the collapse load could be estimated within a 10% range of the experimental values,
which was very good. From the diagram, it is evident that the displacement of the top of the
wall was quite consistent with the values obtained in the model. The good match between
the proposed model and measurements is explained by the choice of the brick pattern.
The importance of local failure modes and particularly those related to the compression
resistant blocks was strongly emphasized.
The implementation of this model in the code of the finite element FEAP allowed one to
obtain very satisfactory results in the case of hollow brick walls. This is consistent with the
assumption made previously about the importance of compression failure modes. It is quite
difficult to solve larger problems using micro-modeling and in this respect it is necessary to
move to macro-modeling. It should be noted that the micro-modeling has shown applicabil-
ity for small walls (Rekik and Lebon 2010; Rekik and Lebon 2012).
4.4 FURTHER RESEARCH
Most of the available models for masonry structures provide analyses in either the in-plane
or out-of-plane direction. However, in reality, most loading occurs as a combination of in-
plane and out-of-plane motion. Very few studies were carried out on the numerical response
under simultaneous in-plane and out-of-plane loading. Therefore, a comprehensive model-
ing strategy is needed to simulate the behavior of masonry structures for 3D cyclic loading
taking into account the in-plane and out-of-plane interaction (Dolatshahi and Aref 2011;
Shaprio et al. 1994; Flanagan and Bennett 1999; Milani 2008; Maheri and Najafgholipour
2012; Najafgholipour, Maheri and Lourenço 2013; Kong, Zhai and Wang 2016).
REFERENCES
Abrams, Daniel P. 2001. “Performance-Based Engineering Concepts for Unreinforced Masonry

Building Structures”. Progress in Structural Engineering and Materials 3 (1) (January): 48–56.
Alfano, G., L. Rosati and N. Valoroso. 2000. “A Numerical Strategy for Finite Element Analysis of
No-Tension Materials”. International Journal for Numerical Methods in Engineering 48 (3):
317–50.
Anthoine, A. In-Plane Behaviour Of Masonry: A Literature Review. EUR 13840 EN, Ispra, Italy:
Commission of the European Communities, JRC-Institute for Safety Technology, 1992.
Babuška, I. and J. M. Melenk. 1996. “The Partition of Unity Finite Element Method: Basic Theory
and Applications”. Computer Methods in Applied Mechanics and Engineering 139 (1–7)
(December): 289–314.
Capra, B., J. Cruz-Diaz, P. Delmotte, A. Mebarki, P. Rivillon and A. Sellier. 2001. Murs de contrev-
entement en maconneries de terre cuite. Report, Cahiers du CSTB.
Colliat J.B., L. Davenne and A. Ibrahimbegovic. 2002. “Modelisation jusqu’a rupture de murs en
maconnerie charges dans leur plan”. Revue francaise de genie civil 6 (4): 395–408.
CUR. 1994. Structrural Masonry: An Experimental/Numerical Basis for Practical Design Rules.
Report 171, Goudam, The Netherlands: CUR.
de Borst, R. 1999. “Computational Methods for Localisation and Failure”. European Conference on
Computational Mechanics. Munchen, 150.
Derakhshan, H., J. M. Ingham and M. C. Griffith. 2009. “Tri-Linear Force-Displacement Models
Representative of Out-Of-Plane Unreinforced Masonry Wall Behaviour”. 11th Canadian
Masonry Symposium. Toronto.
Dolatshahi, Kiarash M. and Amjad J. Aref. 2011. “Three Dimensional Modeling of Masonry
Structures and Interaction of In-Plane and Out-of-Plane Deformation of Masonry Walls”.
Engineering Mechanics Institute Conference 1–6.
Duarte, Armando C. and Tinsley J. Oden. 1996. “An H-P Adaptive Method Using Clouds”. Computer
Methods in Applied Mechanics and Engineering 139 (1–4) (December): 237–62.
Feenstra, Peter H. and René De Borst. 1996. “A Composite Plasticity Model for Concrete”.
International Journal of Solids and Structures 33 (5) (February): 707–30.
Flanagan, R. D. and R. M. Bennett. 1999. “Bidirectional Behaviour of Structural Clay Tile Infilled
Frames”. Journal of Structural Engineering, ASCE, 125 (3): 236–44.
Fries, Thomas-Peter and Ted Belytschko. 2010. “The Extended/generalized Finite Element Method:
An Overview of the Method and Its Applications”. International Journal for Numerical
Methods in Engineering 84 (3) (August): 253–304.
Gambarotta, L. and S. Lagomarsino. 1997. “Damage Models for the Seismic Response of Brick
Masonry Shear Walls. Part I: The Mortar Joint Model and Its Applications”. Earthquake
Engineering & Structural Dynamics E26 (4) (April): 423–39.
Giuseppe, Giambanco and Leonardo Di Gati. 1997. “A Cohesive Interface Model for the Structural
Mechanics of Block Masonry”. Mechanics Research Communications 24 (5) (September):
503–12.
Griffith, M. C., J. Vaculik, N. T. K. Lam, J. Wilson and E. Lumantarna. 2007. “Cyclic Testing of
Unreinforced Masonry Walls in Two-Way Bending”. Earthquake Engineering and Structural
Dynamics 36: 801–21.
Guinea, G. V., G. Hussein, M. Elices and J. Planas. 2000. “Micromechanical Modeling of Brick-
Masonry Fracture”. Cement and Concrete Research 30 (5) (May): 731–37.
Communications in Applied Numerical Methods 7 (3) (April): 187–94.
Ingham, Jason M. and Michael C. Griffith. 2011. The Performance of Unreinforced Masonry
Buildings in the 2010/2011 Canterbury Earthquake Swarm. Technical report, Canterbury:
Canterbury Earthquake Royal Commission of Inquiry.
Kong, Jingchang, Changhai Zhai and Xiaomin Wang. 2016. “In-Plane Behavior of Masonry Infill
Wall Considering Out-of-Plane Loading”. Periodica Polytechnica Civil Engineering 60 (2):
217–21.
Lopez, L., S. Oller and J. Lubliner. 1999. “A Homogeneous Consitutive Model for Masonry”.
International Journal for Numerical Methods in Engineering 46 (10).
Lourenço, Paulo B. and Jan G. Rots. 1997. “Multisurface Interface Model for Analysis of Masonry
Structures”. Journal of Engineering Mechanics 123 (7) (July): 660–68.
Lourenço, Paulo B., Jan B. Rots and Johan Blaauwendraad. 1998. “Continuum Model for Masonry:
Parameter Estimation and Validation”. Journal of Structural Engineering 124 (6) (June): 642–52.
Lourenço, Paulo Jose Brandao Barbosa. 1996. Computational strategies for Masonry Structures.
PhD Thesis, Delft: Delft University Press, Stevinweg 1, 2628 CN Delft, The Netherlands.
Lubliner, Jacob. 2005. Plasticity Theory. Berkeley: University of California at Berkeley, 1990, revised 2005.
Ma, Guowei, Hong Hao and Yong Lu. 2001. “Homogenization of Masonry Using Numerical
Simulations”. Journal of Engineering Mechanics 127 (5) (May).
Maheri, M. R., M. A. Najafgholipour and A. R. Rajabi. 2011. “The Influence of Mortar Head Joints
on the In-Plane and Out of Plane Seismic Strength of Brick Masonry Walls”. Iranian Journal of
Science and Technology 35: 63–79.
Maheri, Mahmoud R. and M. A. Najafgholipour. 2012. “In-Plane Shear and Out-Of-Plane Bending
Capacity Interaction In Brick Masonry Wall”. 15WCEE. Lisbon, 1–10.
Marfia, A. and E. Secco. 2001. “Modeling of Reinforced Masonry Elements”. International Journal
of Solids and Structures 38 (24–25): 4177–98.
Meisle, C. S., K. J. Elwood and C. E. Ventura. 2007. “Shake Table Tests on the Out-Of-Plane Response
of Unreinforced Masonry Walls”. Canadian Journal of Civil Engineering 34: 1381–92.
Milani, G. 2008. “3D Upper Bound Limit Analysis of Multi-Leaf Masonry Walls”. International
Journal of Mechanical Sciences 50 (4): 817–36.
Najafgholipour, M. A., Mahmoud R. Maheri and P. B. Lourenço. 2013. “Capacity Interaction in
Brick Masonry Under Simultaneous In-Plane and Out-of-Plane Loads”. Construction and
Building Materials 38: 619–26.
NAS. 2003. Preventing Earthquake Disasters. National Academy of Sciences.
Nwofor, T. C. 2011. “Finite Element Stress Analysis of Brick-Mortar Masonry Under Compression”.
Journal of Applied Science and Technology 16 (1–2) (March): 48–67.
Oliver, J., A. E. Huespe, M. D. G. Pulido and E. Chaves. 2002. “From Continuum Mechanics to
Fracture Mechanics: The Strong Discontinuity Approach”. Engineering Fracture Mechanics 69
(2) (January).
Papa, Enrico and Alfonso Nappi. 1997. “Numerical Modelling of Masonry: A Material Model
Accounting for Damage Effects and Plastic Strains”. Applied Mathematical Modelling 21 (6)
(June): 319–35.
Paulay, T. and M. J. N. Priestley. 1992. Seismic Design of Reinforced Concrete and Masonry
Buildings. New York: Wiley, John & Sons.
Pegon, P. and A. Anthoine. 1997. “Numerical Strategies for Solving Continuum Damage Problems
with Softening: Application to the Homogenization of Masonry”. Computers & Structures 64
(1–4) (July): 623–42.
Raimondo, Luciano and Elio Sacco. 1997. “Homogenization Technique and Damage Model for Old
Masonry Material”. International Journal of Solids and Structures 34, (24) (August): 3191–208.
Rekik, A. and F. Lebon. 2012. “Homogenization Methods for Interface Modeling in Damaged
Masonry”. Advances in Engineering Software 46 (1) (April): 35–42.
Rekik, Amna and Frédéric Lebon. 2010. “Identification of the Representative Crack Length Evolution
in a Multi-Level Interface Model for Quasi-Brittle Masonry”. International Journal of Solids
and Structures 47 (22–23) (November): 3011–21.
Rosson, Barry T. and Jason L. Suelter. 2001. “Closed-Form Equations for Hardening of Sand-Lime
Mortar Joints”. Journal of Engineering Mechanics 127 (6) (June): 574–81.
Rots, J. G. 1988. Computational modeling of concrete fracture. PhD Thesis, Delft: Delft University
of Technology.
Rots, J. G. 1991. Numerical Simulation of Cracking in Structural Masonry. Heron.
Sacco, Elio and Jessica Toti. 2010. “Interface Elements for the Analysis of Masonry Structures”.
International Journal for Computational Methods in Engineering Science and Mechanics 11
(6) (November): 354–73.
Shaprio, D., J. Uzarski, M. Webster, R. Angel and D. Abrams. 1994. Estimating out of plane strength
of cracked masonry infills. Structural Research Series No. 588, Illinois: University of Illinois at
Urbana-Champaign, Civil Engineering Studies.
Simo, J. C., J. G. Kennedy and S. Govindjee. 1988. “Non-Smooth Multisurface Plasticity and
Viscoplasticity. Loading/unloading Conditions and Numerical Algorithms”. International
Journal for Numerical Methods in Engineering 26 (10) (October): 2161–85.
Simo, J. C., J. Oliver and F. Armero. 1993. “An Analysis of Strong Discontinuities Induced by Strain-
Softening in Rate-Independent Inelastic Solids”. Computational Mechanics 12 (5): 277–96.
Simo, J. and R. Taylor. 1985. “Consistent Tangent Operators for Rate-Independent Elastoplasticity”.
Computer Methods in Applied Mechanics and Engineering 48 (3): 101–18.
Yan, Zhuge, David Thambiratnam and John Corderoy Corderoy. 1998. “Nonlinear Dynamic Analysis
of Unreinforced Masonry”. Journal of Structural Engineering 124 (3) (March): 270–77.
Part 1I
Chapter 5
Dynamics extreme loads in

earthquake engineering
Considerable research has been conducted on the dynamic analysis of structure-foundation

systems. However, due to the use of frequency domain analysis (inherently implied superpo-
sition), the application is restricted to linear analysis. This approach is used to facilitate the
modeling of the radiation condition in a semi-infinite foundation domain. In order to con-
duct nonlinear analysis of structure-foundation systems, time domain analysis is required.
Time domain analysis, however, implies a local approximation to the radiation condition,
i.e. the transmitting boundaries are used to prevent the reflection of outgoing waves in the
finite-size foundation model. For nonlinear problems, it is reasonable to expect that the dis-
sipation of energy due to inelastic material behavior will compensate and diminish the effect
of approximating the radiation condition. The paraxial boundary (Cohen and Jennings
1983; Kausel 1988; Wolf 1986) emerged as systematic local approximation of the global
boundary conditions and the generalization of the ideas of Engquist and Majda (1977) from
the scalar wave equation to the elasticity equations of motion. In this sense, paraxial bound-
ary conditions represent local boundary conditions based on paraxial approximations of
one-way wave equations to finite element analysis.
The models used for the dynamic analysis of structure-foundation systems are usually
very complex, and their detailed nonlinear analysis represents a prohibitive computational
expense. In light of the uncertainties associated with earthquake ground motion character-
istics, one is prone to choose an approach that requires a moderate computational effort so
that the analysis can be placed in an appropriate framework of random vibrations theory
(Lin 1976).
In light of a sound earthquake-resistant design philosophy, nonlinearities should be
localized in a small predetermined part of the structure-foundation system (Clough and
Wilson 1979). There are different base isolation systems that can be utilized here. It is
important to indicate that the locations where nonlinear displacements may occur have
to be known in advance. This means that it is convenient to eliminate the purely linear
response degrees of freedom by static condensation before performing the dynamic analy-
sis; in order to implement this reduction scheme, the structure has to be idealized as an
assemblage of substructures. One, which will be presented in the further sections, is the
dynamic contact, a model problem for local nonlinearity, i.e. the structure is permitted
to uplift. This all indicates that the specific feature of this model is that on one hand the
foundation interaction effect by utilizing the dynamic substructuring concept (Wilson and
Bayo 1986) is included and on the other hand the model is being reduced the dynamic sub-
structuring is used to reduce the size of the linear part to an easily manageable form prior
to nonlinear analysis. This is in sharp contrast to the vast majority of ad-hoc simplified
models used for the same purpose.
131
The formulation is based on the added motion approach of Clough and Penzien (1993).
The added motion approach is used for the analysis of the inertial interaction. By utilizing
the added motion approach, the kinematic interaction phase is not avoided, except in a spe-
cial case of a constant free-field motion across structure-foundation interface.
5.1 THE FREE-FIELD VS. ADDED MOTION APPROACH

IN STRUCTURE-FOUNDATION INTERACTION
A general methodology for dynamic analysis of linear structure-foundation systems is pre-

sented in this section with localized nonlinearities occurring in the structure-foundation
interface only. Earthquake ground motion is usually recorded at a single point, the so-called
control point (Roesset and Kim 1987). In order to use this information, the added motion
approach is suggested by Clough and Penzien (1993). The added motion approach for the
dynamic analysis of structure-foundation systems is based on an additive split of the total
motion field into two motion fields: the free-field motion (motion of foundation with no
structure on it) and the added motion field (due to addition of a structure). The evaluation
of the structural component behavior within the added motion approach is straightforward,
since for the structural component of a structure-foundation system the added motion is the
same as the total motion. On the other hand, the evaluation of the foundation component
behavior becomes more complicated, since two motion fields have to be obtained. However,
the model, which includes the “source” of the earthquake excitation is now chosen of suf-
ficient size since transmitting boundaries are supplied to model the radiation condition.
The basic concepts on which the added motion approach of Clough and Penzien (1993)
is based is presented in the following paragraphs, in order to make it clear before implying
the nonlinear analysis. The system presented in Figure 5.1 is utilized for the derivation of
the added motion approach.
The free-field motion is determined considering only the foundation part. The semi-dis-
crete equations of motion read:
v(t) + C f v (t) + K f v(t) = fff (t) (5.1)

M f
where:
Mf, Cf , Kf represent mass, damping and stiffness matrices of the semi-discrete foun-
dation model;
v(t), v (t),
v(t) represent free-field displacement, velocity and acceleration vectors, and
fff(t) is the far-field boundary loading (where the earthquake ground motion
originated).
Adding the structure component to the foundation one, a complete structure-foundation

system is formed. Now, the equations of motion for the complete system (structure and
foundation) exposed to the same far-field boundary loading fff(t) can be expressed as:
 M s + M f  ( u(t)) + C s + Cf  ( v (t) + u (t)) +  K s + K f  ( v(t) + u(t)) = fff (t) (5.2a)
v(t) +
[M ](v(t) + u(t)) + [C]( v (t) + u (t)) + [K ]( v(t) + u(t)) = fff (t) (5.2b)
where:
u(t), u (t), u
(t) are added motion displacement, velocity and acceleration vectors, and
M s, C s, K s are structural mass, damping and stiffness matrices.
Dynamics extreme loads in earthquake engineering 133
Figure 5.1 Structure-foundation system and the free-field.
Mass, damping and the stiffness of the complete system is denoted as M, C, K. A parti-
tioned form for some of the matrices and vectors of (5.2) are given for clarity, with a remark
that the structure degrees of freedom (dofs) carry subscript s, foundations dofs subscript f,
while interface dofs (dofs common to both structure and foundation) carry subscript i.
Free-field motion v(t)added motion u(t)
 0   u s (t)
v(t) =  v i (t) ; u(t) =  u (t) (5.3)
   i 
 v f (t)  u f (t)
structure mass matrix M s and foundation mass matrix Mf
 M ss M si 0 0 0 0 
 
M s =  M is M iis 0 ; M f = 0 M iif Mif  (5.4)
 0 0 0 0 M fi M ff 
similar partitioned forms apply for damping and stiffness matrices.
The equations of motion can be written in the form of (5.5) if (5.1) and (5.2b) are com-
bined and by taking advantage of sparsity:
 M si  C si   K si 
Mu(t) + Cu(t) + Ku(t) = −  M ii  v i (t) −  Cii  v i −  Kiis  v i (t) (5.5)
 s 
 s 

 0   0   0 
It is interesting to see that only the added motion field will be obtained using the solution of
the equations (5.5).
For the structure, added the motion field represents the total motion field (see partitioned
equations (5.3)); hence, the added motion formulation in the form of equations (5.5) is ame-
nable to the dynamic analysis of a structure-foundation system with nonlinear constitutive
equations for the structural component. Specifically, if the inelastic constitutive equations
for the structure and linear viscoelastic equations for the foundation are assumed, then the
set of semi-discrete equations of motion (5.2b) can be restated as:
 M si  C si   K si 
(t) + C f u (t) + K f u(t) + rs ( u(t), u (t)) = − M ii vi (t) − Cii v i −  Kiis  v i (t) (5.6)
Mu  s   s 
     
 0   0   0 
where the only new notation is introduced for the internal force vector for the structure
dofs, rs ( u(t), u (t)) .
If equations (5.5) or (5.6) are to be used for the structure-foundation dynamic analy-
sis, the free-field velocities and displacements have to be obtained by integrating the given
acceleration record which is not convenient. This difficulty can be avoided by separating the
added motion field into its dynamic and pseudo-static component, and this can be done only
for the linear system:
u(t) = u d (t) + u ps (t) (5.7)
The pseudo-static motion represents the motion of the complete system produced by the free-
field displacements imposed at the interface dofs, when the dynamic effects are neglected.
Going back to (5.5), it is clear that pseudo-static motion can be obtained as:
 K si   K si 
Ku ps (t) = −  K iis  v i (t); u ps (t) = Rv i (t); R = − K −1  Kisi  (5.8)
 0   0 
where R represents to as the influence-coefficient matrix (Clough and Penzien 1993).
Inserting equations (5.7) and (5.8) into the added motion formulation (5.5) (and neglect the
damping proportional forcing function), the new formulation of the motion equations reads:
  M si  
 

d (t) + Cu d (t) + Ku d (t) = Mv
Mu i (t); M − MR +  M iis   (5.9)

 
  0  

The equations of motion (5.9) are further simplified for the embedded structures and the
structures supported at the ground surface as the assumption of free-field motion of the
same value at each interface nodal point can be employed. This means that the pseudo-
static displacements of the structure due to unit displacement at the interface dofs, (5.8), are
actually rigid body displacements. So, the equations of motion for the rigid base earthquake
ground motion are now furnished:
d (t) + Cu d (t) + Ku d (t) = −MsR rb

Mu vi (t) (5.10)
where Rrb is the rigid body transformation matrix (a set of rigid body displacements for unit
displacement at the structure base).
This means that with the added motion approach, leading to equations of motion in
(5.10), splits the total motion field into the dynamic and pseudo-static added motion for the
structure, and into the dynamic, pseudo-static and free-field motion components for the
foundation. However, the assumption on the constant free-field motion overall interface
dofs simplifies the stress recovery in the structure, i.e. the solution of (5.10) is sufficient for
that purpose.
Equation (5.6), for a linear system, can be modified in another way as well. In this assump-
tion, the structure and the foundation are governed by linear viscoelastic constitutive equa-
tions, while only the interface constitutive equations are considered as inelastic. Figure 5.2
is given as a clear illustration of the second proposal.
In this case, the total motion field for the interface dofs on the foundation side
(Figure 5.2), according to formulation (5.6), is split into the free-field motion and the
added motion field. With the solution of (5.6), only the added motion field in the foun-
dation is obtained. However, for the inelastic constitutive equations of the interface,
the total motion field is normally required.
Further additive split of the structure added motion (which is equal to the total motion
of the structure) is proposed between the reference motion introduced by the structure
interface motion identical to the foundation interface free-field motion and the remain-
ing part (see Figure 5.2). If the assumption of constant free-field motion over all interface
dofs is introduced (equivalent to a rigid base assumption), then the reference motion
caused by the free-field motion represents rigid body motion, and the corresponding
stress field is zero for both the interface and the structure. Quite the opposite, if the
free-field motion varies along the structure-foundation interface, the stress field will be
Figure 5.2 Structure-foundation system with inelastic interface-additive split of total motion field.
non-zero in both the structure and the interface. However, from the specified variation
of the free-field motion, this stress field can be easily obtained (state determination), even
for the inelastic interface constitutive equations, and further introduced in (5.6) as the
initial stress in the interface.
This indicates that the split of the structure added motion is equivalent to the split of the
total motion into the pseudo-static and the dynamic components for the linear structure-
foundation system.
This requires the equality of the foundation free-field v if (t) for the interface dofs on the
foundation side (upper side) and the structure reference motion v is (t) for the interface dofs on
the structure side (lower side):
v if (t) ≡ v is (t) (5.11)
Total structure motion is composed of two additive parts:
ut (t) = u(t) + u vi (t) (5.12)
where, by definition:
u vi (t) = − K ss−1 K si v is (t) = Rv if (t) (5.13)
and additionally insertion of (5.11) into (5.13) is done.

Now the equations of motion (5.5) can be rewritten as:
(t) + Cu (t) + Ku(t) + r ( u(t), u (t)) = Mv

Mu is (t) (5.14)
where interface dofs for the structure and the foundation have been separated. For clarity,
expanded form of global dofs and interface internal force r ( u(t), u (t))are given below:
 u s (t)  0 
 uis (t)  ris ( u(t), u (t))
u(t) =  f  ; r ( u(t), u (t)) =  f  (5.15)
 ui (t)  ri ( u(t), u (t))
 u (t)  0 
 f 
and partitioned form of the mass matrix M now becomes a 4x4 matrix:
 M ss M si 0 0 
M M iis 0 0 
M=
is
(5.16)
 0 0 M iif M if 
 
 0 0 M fi M ff 
The damping matrix C and the stiffness matrix K are partitioned in the same way as M
is of the form given in (5.9).
above, while M
5.2 LOCALIZED NONLINEARITIES IN STRUCTURE-FOUNDATION

INTERFACE
Natural base isolation is achieved if uplifting of the structure is allowed (Huckelbridge and
Clough 1977). The generalization of such an isolation concept is the dynamic frictional
contact which occurs at the structure-foundation interface, i.e. a structure can both uplift
and slide. Here we first consider the case when only uplift of the structure is allowed.
The formulation for dynamic contact relies on the approach of Wilson (E. L. Wilson
1975), who introduced relative dof in order to avoid numerical sensitivity in contact prob-
lems. The relative dofs possibly should be defined in a rotated coordinate system (with axes
normal and tangent to the contact boundary) so that the ill-conditioning of tangent opera-
tors does not occur. It should be emphasized that relative dofs fit naturally within the pro-
posed split of the structure total motion field for an inelastic interface only (see (5.12)), since
structure reference motion causes no contact stress (this follows from (5.11)).
The normal interface compliance is assumed in the form of a power law. This choice for
the normal interface compliance plays a crucial role in modeling a more general case fric-
tional contact phenomena (Ibrahimbegović and Wilson 1992). Normal interface compliance
is obtained as the limit for the summation of elastic compliances of the individual asperities;
different distributions of asperity heights result in different powers in the constitutive law for
the normal interface. To account for energy dissipation that occurs in impact, a simple non-
linear viscoelastic model is suggested as a supplement to the elastic power law for the normal
penetration interface. It can be considered as a generalization of the dissipative model of Hunt
and Crossley (1975), which uses the restitution coefficient for different materials. The normal
interface stress σn is given as:
mn ln
σ n = −Cn un + Bn un u n (5.17)
where ⋅ denotes a Macaulay bracket (i.e. un = un if un > 0 and un = 0 if un ≤ 0), and un

is the normal interface deformation (approach). Normal interface deformation un is deter-
mined as the difference between the relative normal displacement Δ and the initial normal
gap g (see Figure 5.3). In (5.17), Cn , Bn , mn , ln are the constitutive coefficients.
Figure 5.3 Initial gap, normal displacement and penetrating approach.

A family of isoparametric finite elements is developed to incorporate the model for

dynamic contact presented herein. Contact elements which can be used for the modeling of
contact boundary Γc in the form of a point, a line or a surface can be developed in a unified
manner. The interpolation of contact boundary geometry x and displacement field u in each
element is:
np np
x= ∑ N x ; u = ∑ N u (5.18)
I =1
I I
I =1
I I
where NI is the standard isoparametric shape function, while xI are nodal coordinates and
uI are relative nodal displacements.
T
If the unit’s normal vector is denoted on the contact boundary Gc as n = n1, n2 , n3 and
defines vector N,
(n1N1 ) ; (n2N2 ) ; (n3N3 ) ;...; (n1Nnp ) ; (n2Nnp ) ; (n3Nnp )

T
N= (5.19)
the contact element (e) contribution to the interface residual ri (u(t), u (t)) in (5.15) is:
ri(e) ( u(t ), u (t )) =
∫ Nσ dΓ (5.20)
Γ(ce )
n
Utilizing the consistent linearization procedure (Hughes and Pister 1978), for the contact
interface residual (5.20), the tangent stiffness matrix can be written:
∫ ∫
mn −1 ln −1
K(iie) = Cnmn un dΓ + Bn ln un u nσ n NN T d Γ (5.21)
Γ(ce ) Γ(ce )
and the tangent damping matrix resulting from the linearization of the normal interface
residual with respect to velocities:
γ
∫B
ln
C(iie) = n un NN T d Γ (5.22)
β∆t
Γ(ce )
where β and γ are Newmark parameters. This model can readily be extended to the fric-
tional contact interface by following approach proposed in (Ibrahimbegovic, Wilson 1992).
5.3 REDUCTION OF MODEL: NUMERICAL TECHNIQUES
To enhance computational efficiency, the complete set of semi-discrete equations of motion

(5.14) is projected onto the subspace spanned by the selected sets of Ritz vectors (for both
the structure and the foundation) as well as the interface global finite element coordinates.
Modal transformation are used in order to reduce the number of dofs for the structure:
 u s (t) Ys R s   y s (t)

 u s (t) =  0 I i   uis (t)
(5.23)
i 
and the foundation:
 u f (t) Yf R f   y f (t)

 u f (t) =  0 I i   uif (t)
(5.24)
i 
where:
y s and y f are the sets of Ritz vectors (subspace basis) for the structure and the foun-
dation, respectively;
R s and Rf are the corresponding pseudo-static transformations (following from
(5.23) and (5.24));
Ii is the identity matrix of dimension equal to the number of interface dofs; and
ys(t) and yf(t)
are sets of generalized Ritz coordinates for the structure and the
foundation.
The sufficient number of Ritz vectors for representing the structure and the foundation can
be determined by monitoring truncation criteria for the Ritz vector basis (Ibrahimbegović
and Wilson 1990).
In the case of the paraxial boundary, the discretized damping matrix C has a nonpropor-
tional form. If lumping techniques (Chow 1985) for the evaluation of paraxial boundary
damping matrices are used, the projection of the nonproportional damping matrix on the
Ritz vector subspace can be easily obtained as a summation of rank-one matrices which
affects only far-field boundary dofs. Material damping can also be dealt with only at the
level of the subspace either by giving to it a Rayleigh damping form or directly specifying
the damping ratios. An algorithm that accommodates nonproportional damping within the
real Ritz vector subspace is selected, since it was demonstrated (Ibrahimbegović, Chen et al.
1990) to be more efficient than the adequate one that employs the complex Ritz vector
subspace.
Introducing transformations (5.23) and (5.24) into the weak form of the equations of
motion (5.14) and utilizing the additive split of the modal nonproportional damping matrix
for the foundation substructure C f = diag(2ξω f ) + Cˆ f , the new form of the equations of
motion is obtained:
*(t) + C*u *(t) + K*u*(t) + r ( u(t), u (t)) = T T Mv

M*u is (t) − C f y f (t) (5.25)
The form of the internal force vector r(u(t), u (t)) is preserved under modal transformation.
The displacement vector in (5.25) has the form:
 y s (t)
 y f (t)
u*(t) =  s  (5.26)
 ui (t)
 u f (t)
i
which is obtained by the reduction of the form u*(t) = Tu(t). Now, the mass matrix of the
complete system in an expanded form is given as:
 Is 0 ˆ si
M 0 
 
 0 If 0 ˆ fi 
M
M=  (5.27)
M ˆ Tsi 0 ˆ iis
M 0 
 ˆ ˆ f 
 0 MTfi 0 M ii 
where:
ˆ si = YsT M ss R s + YsT M si ; M
M ˆ iis = RTs M ss R s + M si R s + RTs M is + M iis (5.28)
Similar partitioned forms can be written for the damping matrix C* and stiffness matrix
K*, except that the upper left blocks of those matrices have the diagonal form with elements
2ξi ω i and ω i2 instead of the identity matrices Is and If (ξi are modal damping ratios and ωi mi
are approximate modal frequencies).
The semi-discrete equations of motion (5.25) can be further solved by utilizing step-by-
step integration methods, i.e. fully discretizing them. For this purpose the implicit Newmark
algorithm family’ can be used. This requires the tangent operators (5.21) and (5.22) for all
contact elements and the solution of the nonlinear set of equations at each time step. The
task of solving a set of nonlinear equations is greatly simplified by first reducing the size of
the linear part of the complete system to a small set of generalized Ritz coordinates, and
second by reducing the linear part to its Schur complement (Duff, Erisman and Reid 1989),
i.e. prior to the solution of the nonlinear equations set the static condensation is performed
on the effective stiffness matrix K̂:
 Kˆ 11 Kˆ 12 
  →  Kˆ 22 − Kˆ 21Kˆ 11−1Kˆ 12  (5.29)
 Kˆ 21 Kˆ 22   
where the static condensation is made trivial by the diagonal form of K̂11 for all the matrices
in (5.25), and consequently for the effective stiffness matrix. For the step-by-step algorithm
with variable step size (a requirement of efficient solution for dynamic frictional contact),
the linear part of the effective stiffness matrix has to be reformed and refactorized quite
often. Computation of the Schur complement (5.29) of the linear part is much more efficient
for the equations of the form (5.25) than for the complete set (5.14) where the linear part is
represented in the finite element coordinates (Bathe and Gracewski 1981). Hence, the initial
cost of formulating the Ritz vector subspace for both the structure and the foundation is
likely to be compensated for, especially for loading of long duration (e.g. earthquake load-
ing). In addition, the number of operations used to form the effective load vector is also
significantly reduced, since dim(y s ) dim(u s ) and dim(y f ) dim(u f ) .
5.4 CASE STUDIES OF STRUCTURE-FOUNDATION

INTERACTION PROBLEMS
The analysis of the dam-foundation model presented in Figure 5.4 is performed when the
uplifting of the dam occurs. The final selection for the foundation model is done after exten-
sive parametric studies of radiation effects (Ibrahimbegović and Wilson 1989a). The model
of the dam is very similar to the Pine Flat Dam analyzed by Chakrabarti and Chopra (1973)
for hydrodynamic effects. Only the uplifting effect is considered; the tangential sliding is
assumed to be restricted either by embedment or otherwise. Material properties for the dam
and the foundations are given in the Table 5.1.
Material damping is given the form of Rayleigh damping with control frequencies equal
to 13.27 and 21.39 rad/sec and damping ratio ξ = 5%, while the nonproportional damp-
ing arises from paraxial boundaries used to represent the radiation condition. To model a
maximum credible earthquake, the 1952 Taft earthquake (record S96E with peak ground
acceleration 0.17949 at 3.74 sec) is arbitrarily scaled by 3. Since a nonlinear problem is
elaborated, hydrostatic water pressure and dead load are considered simultaneously with
the earthquake excitation. Hydrodynamic water pressure is completely disregarded. The
initial conditions are represented by the deformed configuration under the influence of
static loads.
Figure 5.4 Finite element models of dam-foundation system.
Table 5.1 Material Properties for the Dam and the

Foundations, and Contact Boundary
Dam and Foundations Contact Boundary
Young modulus 22150 MN/m2 Cn 106 MN/m2
Mass density 2500 kg/m3 mn 2
Poisson’s ratio 0.25
Figure 5.5 Horizontal displacement at dam tip.

Figure 5.6 Horizontal displacement at dam bottom.
Figure 5.7 Difference in horizontal displacement—dam top vs. bottom.
Both the dam and the foundations are represented by sets of 10 Ritz vectors which have,
by the previous analysis, proved to carry most of the information. The nonproportional
damping, which arises from the paraxial approximation to the radiation condition, is han-
dled by the additive split of the modal damping matrix. The equilibrium position of the
dam-foundation system under the static load only is determined from the analysis of the
complete system represented in finite element coordinates (I 154 dofs). At the time of con-
ducting this example, a research version of the computer program SAP26 was used to per-
form the computations.
Figures 5.5 and 5.6 illustrates the displacements at the dam top and the bottom, respec-
tively, for both cases where the uplifting of the dam is permitted and prevented. Over a
Figure 5.8 Vertical stress at dam bottom.
certain range of frequencies response amplification occurs due to the uplifting, but in the
part of the strong shaking the beneficial isolation effects of the uplifting are evident. To
assess the stress field in the dam, the difference of the horizontal displacement at the dam
top vs. the bottom, as well as the vertical stress time-history at the dam bottom, are plotted
in Figures 5.7 and 5.8, respectively. As it is assumed that the dam is made of low quality con-
crete, say C15/20 (which is consistent with the choice of Young’s modulus value for the dam
material), then the allowable compressive stress is 8 MPa while the allowable tensile stress
is 0.8 MPa. Hence, if the uplifting of the dam is allowed for, no cracking of the dam body
would occur. Figure 5.8 shows that if the uplifting is prevented this will not be the case.
BIBLIOGRAPHY
Bathe, Klaus-Jürgen and Sheryl Gracewski. 1981. “On Nonlinear Dynamic Analysis Using
Substructuring and Mode Superposition”. Computers & Structures 13 (5–6): 699–707.
doi:10.1016/0045-7949(81)90032-8.
Chakrabarti, P. and Anil K. Chopra. 1973. “Earthquake Analysis of Gravity Dams Including
Hydrodynamic Interaction”. Earthquake Engineering & Structural Dynamics 2 (2): 143–60.
doi:10.1002/eqe.4290020205.
Chow, Y. K. 1985. “Accuracy of Consistent and Lumped Viscous Dampers in Wave Propagation
Problems”. International Journal for Numerical Methods in Engineering 21 (4): 723–32.
doi:10.1002/nme.1620210411.
Clough, Ray W. and Edward L. Wilson. 1979. “Dynamic Analysis of Large Structural Systems with
Local Nonlinearities”. Computer Methods in Applied Mechanics and Engineering (17–18):
107–29. doi:10.1016/0045-7825(79)90084-7.
Clough, Ray W. and Joseph Penzien. 1993. Dynamics of Structures. Edited by 2nd. New York:
McGraw-Hill Education (ISE Editions).
Cohen, M. and P. C. Jennings. 1983. Silent Boundary Methods for Transient Analysis. Edited by T.
Belytschko and T. J. R. Hughes. New York: Elsevier Science Publications.
Duff, I. S., A. M. Erisman and J. K. Reid. 1989. Direct Methods for Sparse Matrices. Oxford: Oxford
University Press.
Engquist, Bjorn and Andrew Majda. 1977. “Absorbing Boundary Conditions for the Numerical
Simulation of Waves”. Mathematics of Computation 31 (139): 629–51. doi:10.2307/2005997.
Huckelbridge, A. and R. W. Clough. 1977. “Seismic Response of Uplifting Building Frame”. Journal
of Structural Engineering (ASCE) 1222–1229.
Hughes, Thomas J. R. and Karl S. Pister. 1978. “Consistent Linearization in Mechanics of Solids and
Structures”. Computers & Structures 8 (3–4): 391–97. doi:10.1016/0045-7949(78)90183-9.
Hunt, K. H. and F. R. E. Crossley. 1975. “Coefficient of Restitution Interpreted as Damping in
Vibroimpact”. Journal of Applied Mechanics 42 (2): 440–45. doi:10.1115/1.3423596.
Ibrahimbegović, A. and E. L. Wilson. 1989a. Dynamic Analysis of Large Linear Structure-
Foundation Systems with Local Nonlinearities. UCB/SEMM 89/14, Berkley, CA: University of
California Berkley.
Ibrahimbegović, A. and E. L. Wilson. 1989b. “Simple Numerical Algorithms for the Mode
Superposition Analysis of Linear Structural Systems with Non-Proportional Damping”.
Computers & Structures 33 (2): 523–31. doi:10.1016/0045-7949(89)90026-6.
Ibrahimbegović, Adna, and Edward L. Wilson. 1990. “Automated Truncation of Ritz Vector Basis
in Modal Transformation”. Journal of Engineering Mechanics 116 (11): 2506–20. doi:10.1061/
(asce)0733-9399(1990)116:11(2506).
Ibrahimbegović, Adnan, Harc C. Chen, Edward L. Wilson and Robert L. Taylor. 1990. “Ritz Method
for Dynamic Analysis of Large Discrete Linear Systems with Non-Proportional Damping”.
Earthquake Engineering & Structural Dynamics 19 (6): 877–89. doi:10.1002/eqe.4290190608.
Kausel, Eduardo. 1988. “Local Transmitting Boundaries”. Journal of Engineering Mechanics 114
(6): 1011–27. doi:10.1061/(asce)0733-9399(1988)114:6(1011).
Lin, Y. K. 1976. Probabilistic Theory of Structural Dynamics. Huntington, NY: R. E. Krieger Pub.
Co.
Roesset, J. M. and J. S. Kim. 1987. “Specification of Control Point for Embedded Foundations”. 5th
Canadian Conference on Earthquake Engineering. 63–86.
Wilson, E. L. 1975. “Finite Elements for Foundations, Joints and Fluids”. Conf. on Numerical
Methods Soil Roch Mechanics, September 15–19. Karlsruhe, Germany.
Wilson, Edward L. and Eduardo P. Bayo. 1986. “Use of Special Ritz Vectors in Dynamic
Substructure Analysis”. Journal of Structural Engineering 112 (8): 1944–54. doi:10.1061/
(asce)0733-9445(1986)112:8(1944).
Wolf, John P. 1986. “A Comparison of Time-Domain Transmitting Boundaries”. Earthquake
Engineering & Structural Dynamics 14 (4): 655–73. doi:10.1002/eqe.4290140412.
Chapter 6
The dynamics of extreme impact

loads in an airplane crash
Present day challenges in fighting man-made and natural hazards have brought about new
issues in dealing with extreme transient conditions for engineering structures. As already indi-
cated, the main advantage of the multi-scale approach is that it has by far the greatest capabili-
ties for providing a basis for constructing a sufficiently predictive model. Namely, this kind
of approach provides refined prediction capabilities of the mechanics model with respect to
the smeared models of plasticity or damage (Lemaitre and Chaboche 1988; Lubliner 1990),
as well as more detailed information regarding inelastic mechanisms rather than just the total
inelastic dissipation. It was only at the beginning of the 21st century that models for predict-
ing detailed crack-spacing and opening (fracture process zones developed in more detail) were
developed. Application to metals was performed by Ibrahimbegović and Brancherie (2003)
and to concrete by Brancherie et al. (2005) and Ibrahimbegović (2006) or to reinforced con-
crete by Dominguez et al. (2005).These models are used as the basis of models for massive
structures exposed to dynamic extreme airplane actions, as a man-made hazard for which
most engineering structures have not been designed. From the standpoint of nonlinear analy-
sis, the main difficulty in such a problem pertains to the significant high-frequency content
typical of impact phenomena, as well as the likely presence of extensive structural damage. In
this respect, first of all, it is necessary to develop models that will be able to predict inelastic
behavior and damage for both the airplane and the massive structure. Second, it is interesting
to examine different impact scenarios in trying to quantify any potential reserve that might
exist in the given design of engineering structures for taking on a higher level of risk. The
complementary goal, which also of great importance, pertains to providing the best-reduced
basis for carrying out the parametric studies that are needed for a sound design procedure.
The idea for such analysis was born after the September 11, 2001, attacks on the World
Trade Center (Figure 6.1).
In order to solve such a complex problem, it is necessary to incorporate the significant
advances in nonlinear analysis within the design procedure of a complex engineering struc-
ture under an equally complex dynamic loading. One possible example of this would be the
impact by an airplane on a massive structure, such as a nuclear power plant (Figure 6.2).
It is important to note that traditional design procedures are not applicable to these types
of complex structures, and even less so to a complex, nonproportional loading program
such as the one produced by the frictional contact of the airplane with the structure. At the
moment, some empirical expressions are used, which are based on rather simplified interpre-
tations of experimental results of projectile tests perforating different plate-like structures,
and this is very difficult to apply on complex structures.
This all shows that there was and still is the need for a novel approach to the nonlinear
analysis of these kinds of problems.
One of the elements that has a major influence on structure damage, as well as the struc-
ture–projectile interaction, is the true nature of the impact. In this respect, it is necessary to
145
Figure 6.1 World Trade Center attack, 9/11/2001 (www.telegraph.co.uk).
Figure 6.2 Nuclear power plant: massive structure built to sustain airplane impact.
The dynamics of extreme impact loads in an airplane crash 147
define the type of impact action, either hard or soft impact. In the case of hard impact, the
projectile is much stiffer than the target; whereas in the case of the soft impact, the target is
at least as stiff as the projectile (Figure 6.3). Additionally, a great diversity in damage modes
creates another problem that has to be taken into account and elaborated in detail, including
perforation, cauterization or spalling (Figure 6.3).
In order to solve this kind of complex problem, it is necessary to split the analysis into
two phases, local and global. The local phase deals with the analysis of a single structural
component (e.g. an impacted area of a nuclear power plant or a concrete slab in a facility,
see Figure 6.4); whereas the global analysis deals with the complete (complex) structure. The
result of the local analysis will be incorporated into the final stage of the global analysis in
the sufficiently representative manner. This kind of formulation will provide a much higher
efficiency of the computations in the global phase. In this way, all the parametric studies,
needed in the design phase, can be performed in a very efficient manner, which is the main
advantage of this kind of approach (Figure 6.4).
6.1 CENTRAL DIFFERENCE SCHEME

COMPUTATION OF IMPACT PROBLEMS
The local phase of the analysis is carried out first using a detailed local model of a single
component corresponding to the impacted zone, which is exposed to an impact action of
short duration. Analysis of this kind considers the finite element model of the projectile (for
example an airplane), which can account for the large plastic deformations developing at the
impact and the frictional contact with potentially large frictional sliding. The explicit, cen-
tral difference scheme (Bathe and Wilson 1976; Owen and Hinton 1980) is most appropri-
ate for such an analysis. The state variable values over a typical increment can be computed
according to:
d n +1 = d n + ∆tv n + ∆t 2a n ,
(
Ma n +1 = rcontact, n +1 − fˆ int d n +1  ε vp , ,…
GNP ), (6.1)
∆t
v n +1 = v n
2
(a n + a n +1 )
Figure 6.3 Hard vs. soft impact and damage modes.

Figure 6.4 An analysis-design procedure based on the split between the local phase and the global phase.
where d, v and a are, respectively, displacements, velocities and accelerations, and M is the
mass matrix in a diagonal form for computational efficiency. The main sources of nonlin-
earity belong to the impact and frictional contact of the airplane with contact forces denoted
as rcontact, as well as to the plastic and damage deformations of the impacted component of
the structure, which are all stored in internal force term fˆ int .
A reliable representation of damage in the impacted structural component made of rein-
forced concrete has to be given, which can then be used for the structural design. Several
damage mechanisms should be represented: the first is concrete damage in tension leading
to cracking; the second is damage in compression with the original feature that the concrete
initially hardens due to compaction with an increase of stress until the ultimate value where
the concrete will break; the last is the dependence of the compressive and tensile strength of
the concrete with respect to the rate of deformation (Figure 6.5).
The fracture of concrete in tension is described by the first criterion that depends directly
upon the principal values of the principle tensile elastic strain. Same criterion can also be used
to detect concrete cracks in compression (with cracks parallel to compressive loading direc-
tion), when modified to account for the corresponding increase of fracture energy. The latter
criterion for the concrete compaction and damage in compression is chosen (Hervé, Gatuingt
and Ibrahimbegović 2005) as a Gurson-like plasticity criterion, defined with:
( )
φ σij , σ m , f * =
3 J2
σ 2m
 I 
( (
+ 2q1f * cosh  q2 1  − 1 + q3f *
 2σ m 
) ) ≤ 0 (6.2)
2
where:
σij are the components of the nominal stress tensor
σm is the reference value of the stress in the matrix
f* is the concrete porosity and
q1, q2 , q3 are the chosen coefficients.
The graphic illustration of this criterion for both compressive and tensile stress is shown
in Figure 6.5.
The constitutive model also features porosity-dependent hardening that evolves as a func-
tion of the equivalent compressive plastic deformation. The same parameter defines the
threshold for element erosion, defining the stage where the element is damaged to such an
extent that it ought to be completely removed from the mesh.
The evolution equations for plastic deformation in compression and damage deformation
in tension are chosen as rate-dependent in order to account for the effects of rate of defor-
mation, which are quite noticeable for this class of problem (Figure 6.5). The numerical
implementation of the proposed model is incorporated within the proposed computational
framework in (6.1). However, contrary to the explicit scheme computations for the global
momentum balance equations, the integration of the evolution equations for internal vari-
ables, such as plastic strains, damage compliance and hardening/softening variables, is car-
ried out by an implicit scheme. The latter provides at each step the admissible values of stress
with respect to the chosen criteria and leads to a robust numerical implementation.
The tests for simulating impact were done in Sandia Laboratory, and the reliability of
the proposed constitutive model was checked. This was done using different sized missiles,
which simulated aircraft engines, and their impact on the reinforced concrete slabs. In the
Figure 6.5 A constitutive model of concrete for dynamic analysis of the dependence of compressive and
tensile strength on the rate of deformation, yield damage criterion and hardening introduced
by compaction.
simulation process, it was seen as best to simulate both rigid and soft missiles. In this respect,
simulations for large size equivalent and deformable missiles, medium size equivalent and
deformable missiles and small size equivalent rigid missiles (SER) were done (see Figure 6.6
for details). The results of two of the conducted experiments are given in Table 6.1.
The missiles where modeled using the shell finite element models, except for the SER
missiles where 3D solid elements were used. The concrete slabs were modeled using under-
integrated 3D solid elements with the proposed constitutive model, and the reinforcement
was modeled with truss-bar elastoplastic elements. In order to define the different types of
damage, it was necessary to distinguish between penetration, perforation and scabbing.
Penetration denotes that the missile penetrated the slab without having gone through it;
Figure 6.6 (a) Experimental results of slab perforation obtained at Sandia US National Lab, (b) and the results
of numerical simulations.
Table 6.1 Simulated Impact Test—Missile and Slab Characteristics and Experimental
Results—Sandia Laboratory
Missile Velocity Slab Reinforcement Spalling/
No Type (m/s) Thickness (M) Ration Slab Type Perforation Scabbing Penetration
S10 SER 141 0.15 0.4 Small no.1 No Yes Yes
L5 LED 214 1.60 0.4 Large no.3 No Some Yes
perforation denotes that the missile went through; and scabbing denotes that the impact
generated a scab on the rear face of the slab.
Figure 6.6b shows the numerical simulations for the selected examples. It is clear that
quite similar results were obtained. Once the velocity of the missile stopped decreasing,
which implied the missile was finally stopped by the concrete, the computations where
stopped. Numerical simulation of example S10 (thin slab) showed damage of the slab, spall-
ing occurred on the rear face, but there was no penetration of the missile. The second exam-
ple was a thick slab (L5), where the damage remained localized in the impact area with a
large undamaged volume separating damage on the front face and damage on the rear face.
The latter implies the presence of only some small cracks. A very good correlation between
the model and the tests were obtained.
6.2 LOCAL MODEL EXTENSIONS
One phenomena that needs to be elaborated further is spalling, where a damaged piece from
the main structure can detach and fly away. This represents an additional issue that has to
be modeled. Damage of the interior equipment or storage material depends on the size of
the detached piece, more precisely its mass, and its velocity. If the velocity is too high and/
or if the piece of detached material is too big, it can cause significant damage to the equip-
ment or storage material without causing larger damage to the structure. However, this can
cause significant damage, explosion or fire, and lead to structural damage and even collapse.
The best way to represent the spalling phenomena is through a three-point bending test
in dynamics. This test is performed in the manner that a weight is dropped from a certain
height. In this way, an impact is caused on the fallen structure producing a compressive wave
that travels from the upper surface in a downward direction toward the bottom surface
of the specimen. When this wave reaches the lower surface of the specimen, it will reflect,
double in size and turn into a tensile wave. This is the moment when the damage can be
introduced in tension sensitive material, with pieces that can detach and freely move away
from the main structure.
The results of the performed numerical simulation are illustrated in Figure 6.7.
This damage-plasticity continuum model is capable of predicting the extent of the dam-
age zone associated with a large value of the damage variables or even eliminate all the
elements which are extensively damaged by activating the erosion criterion. The continuum
model enables one to eliminate the pieces of the damaged structure that have detached dur-
ing the impact; however, it is not possible to follow the motion of these detached pieces.
In order to grasp these phenomena thoroughly, it is necessary to use a discrete model. In
this case, a model based on the Vornoi cell representation of the specimen was developed
by Ibrahimbegović and Delaplace (2003), where the cohesive forces between the adjacent
cells are represented by a geometrically exact Reissner beam model (Ibrahimbegović and
Taylor 2002). This particular feature of the beam model with its capability for representing
the overall large motion (accompanied by small strains) is crucial for the present applica-
tion, where the detached piece is represented by several Vornoi cells with preserved cohesive
forces (Figure 6.7). There is a very delicate question here concerning the implementation
of appropriate time-integration schemes (Delaplace and Ibrahimbegović 2006) capable of
controlling the high-frequency content of motion (Ibrahimbegović and Mamouri 2002) and
thus minimizing the risk of the spurious stress oscillation introduced by brittle fracture.
An example of an anisotropic damage model is given; which can be generalized to even
a complex model. Namely, a coupled damage-plasticity model can be obtained, which is
capable of representing the inelastic behavior of porous metals (Ibrahimbegović, Markovic
Figure 6.7 The numerical simulation of three-point bending testing: A continuum model with element ero-
sion can completely remove damage elements, and the discrete model can also represent spalling
as well as the motion of any detached piece of the specimen.
and Gatuingt 2003) or concrete under compaction (Hervé, Gatuingt and Ibrahimbegović
2005). The latter is used in the application of a commercial airplane impact problem on a
massive structure. Finally, the same kind of development can be carried out for an even more
general framework of the reinforced concrete model, which is assembled from the aniso-
tropic damage model for concrete, the elastoplastic model for steel and a coupled damage-
plasticity model for bond-slip (Dominguez et al. 2005). The latter is implemented within
the zero-thickness bond element with normal and tangential degrees of freedom, which is
constructed in the same manner as the contact element described by Ibrahimbegović and
Wilson (1992). An increase in slip resistance due to confinement pressure can be easily rep-
resented. Figure 6.8 illustrates the predictive capabilities with respect to crack-spacing and
opening for this kind of model, where the important role played by a bond-slip element in
redistributing the stress transfer mechanism along the steel bar more evenly was demon-
strated, as well as an excellent correlation between the experimental and numerical results,
even in the case of the dispersion of slip resistance (Dominguez et al. 2005).
6.3 FIELD TRANSFER STRATEGY FOR COMPLEX

STRUCTURE IMPACT LOADING
Once the local analysis phase of the airplane impact on a single structural component (or
impacted zone) has been completed, the global phase can start. The main goal of this phase
is to check the integrity of the whole structural assembly and perform any eventual para-
metric studies where different design possibilities are proposed. Efficiency is ensured by the
chosen coarse mesh analysis at the structural scale, which incorporates the results from the
structural component which were considered in the local phase. Global analysis can directly
integrate the results obtained in the local analysis phase, providing a couple of possible sim-
plifications. First of all, the fine mesh used in the local analysis phase is replaced by a coarse
Figure 6.8 Pull-out test for a reinforced concrete beam: Experimental results, numerical results (with no
bond-slip, with constant bond-slip and with variable bond-slip resistance using the standard
variation of 0.05, 0.10 and 0.20 and the resulting force–displacement diagram).
mesh which is chosen from the same grading as the mesh used for the rest of the structural
assembly. In this way, any potential risk that the waves propagating across the interface
between the impacted zone and the rest of the structure remain trapped within the refined
mesh zone is eliminated. The second simplification consists of replacing the true loading on
the impacted structural component, which stems from time-consuming computations of
frictional contact and the impact of the airplane, with the equivalent nodal loading applied
at the nodes of the coarse mesh of the structural component, which is used in the global
analysis phase.
This represents a typical field transfer problem. The biggest issue here is how, in the best
possible way, to perform the field transfer task between the fine and the coarse mesh. This is
easily explained in Figure 6.9 where a slab represented by a fine mesh of 3D solid elements,
is impacted by a small plate.
This nonlinear impact problem is solved using the central difference scheme which typi-
cally employs very small time steps as required by the Courant-Friedrichs-Levy stability
condition (Bathe and Wilson 1976; Ibrahimbegović 2009) and the small finite element sizes
in the fine mesh. Further, it is necessary to recover these results as close as possible from the
nonlinear impact analysis on the coarse mesh, where again the central difference scheme
will be used with larger time steps due to larger element sizes. The coarse mesh analysis is no
longer driven by contact, but with an equivalent loading applied on the coarse mesh, which
can further and dramatically improve the computational efficiency.
6.4 FIELD TRANSFER IN SPACE AND

PROJECTION TO COARSE MESH
The first transfer problem concerns the space coordinates. Namely, the transfer between the
fine and the coarse mesh at any chosen time, say tn or tn+1. One such field transfer should be
done for any point in space on the coarse mesh, denoted as x, where the best possible coarse
mesh representation of the results obtained on the fine mesh is required. The coarse mesh
representation is constructed in accordance with the least moving square approximation of
the fine mesh transfer. More precisely, for the displacement component on the coarse mesh
dicoarse (x) at a chosen location x, it can be assumed that the projection, which is based on the
nodal values of the corresponding displacement component difine (x) obtained from solution
to (6.1) in the neighborhood of x on the fine mesh can be used:
dicoarse (x) = ∏ (d (x )); ∀x

x
fine
i
fine
j j ∈  (x) (6.3)
where Π denotes the corresponding projection operator. The subscript x indicates that such
x
a projection operator remains dependent on the chosen location, through space dependence
of its coefficients. More precisely, projection of this kind should be in accordance with the
moving least square approximation which is written as:
 b0 ( x )
 
b1 ( x ) 
T 
dicoarse (x) = [1 x1 x2 x
3  ] b 2 ( x ) 
 = p(x)T b(x) (6.4)
 b3 ( x )
where b(x) are the approximation parameters to be determined from the least square fit at
each desired location x . This location is typically selected in accordance with the coarse mesh,
either as the nodal point of the numerical integration point, with a known coordinates values
x . Projection to the coarse mesh is constructed using the complete linear polynomials p(x).
Figure 6.9 Coarse and fine mesh representation of a plate-like structural component and space–time field
transfer between two meshes based on diffuse approximation.
The choice of linear approximation is conditioned by coarse mesh interpolation, where the
linear approximation is used for all fields of either displacement, velocity or acceleration.
The minimization problem for computing the moving least square approximation coef-
ficients can thus formally be written for each motion component according to:
2
 
 
∑
1
W ( x, x j )  p ( x j ) b ( x )− difine ( x j ) (6.5)
T
min J(b); min J(b) :=
b b 2 j ∈ (x)  dicoarse ( xj ) 
 
where W ( x, x j ) are the bell-shaped weighting functions (Figure 6.9), which are chosen in order
to limit the influence of the points placed farther away from the point x, as well as to provide
a continuous approximation when moving across the domain of influence of the neighboring
nodes on the coarse mesh (Villon, Borouchaki and Khemais 2002). The latter is true for the
present case only when taking more than a minimum of the four closest points in the neighbor-
hood x, x j ∈  (x), j = 1,...m ≥ 4; or otherwise the weighting functions will play no role. The
cost function in the minimization problem in (6.5) can also be stated in matrix notation as:
1 T 1
J(b) = b PWP T b − bT PWd fine + d fine,T Wd fine (6.6)
2 2
where the weighting factors, interpolation polynomial and fine mesh displacement values
are stored as:
W ( x, x1 ) .. 0   p ( x )T   d fine ( x ) 
   1
  i 1

..
W=  ; PT =   ;
d fine
=   (6.7)
 ..    fine 
T

 0 ...

W ( x, x m )  p ( x m )  di ( x m )
The Kuhn–Tucker optimality condition of this minimization problem leads to the optimal
value of approximation parameters:
∂J(b) −1
0= ⇒ b = PWP T  PWd fine (6.8)
∂b
which defines completely the chosen approximation in (6.3). In order to improve the system
conditioning and efficiency, to ensure the “partition of unity” property for the weighting
functions, the following simple transformation is used:
W ( x, x i )
W ( x, x i ) ← (6.9)
∑
j ∈ (x )
W ( x, x j )
By this, the field transfer in space between the fine and the coarse mesh is clarified. The next
step is the transfer of the evolution problem in time. It is evident that usually, and in the
most cases, the time steps used in the fine mesh are not kept the same for the coarse mesh
but are much bigger. It is important to make a correct field transfer regarding the different
time steps with respect to the two different mesh scales.
A single time step in the coarse mesh computation is considered which corresponds to a
number of small steps on the fine mesh (Figure 6.9).
There are two possible ways for this field transfer. The first possible way is by first carry-
ing out the computations on the fine mesh using the central difference scheme in (6.1) and
then use the projection in (6.3) of computed result on the coarse mesh. This kind of the field
transfer is carried out for each component of the displacement, velocity and acceleration
vectors, which can formally be written as:
d ncoarse
+1 (x) := ∏ (d (x )); ∀x
x
fine
n +1
fine
j j ∈  (x)
v ncoarse
+1 (x) := ∏ (v (x )); ∀x ∈  (x)
x
fine
n +1
fine
j j
(6.10)
a coarse
n +1 (x) := ∏ (a (x )); ∀x ∈  (x)
x
fine
n +1
fine
j j
fncoarse
+1 (x) := ∏ (r
x
fine
contact, n +1 (x )); ∀x
fine
j j ∈  (x)
coarse coarse coarse coarse

where d n+1 , v n+1 ,a n+1 and fn+1 are, respectively, the computed vectors of displacements,
fine
velocities and acceleration at time tn+1, while rcontact , n+1 are the nodal forces computed from
the contact problem on the fine mesh. The last expression states that any time we need the
corresponding values on the coarse mesh (typically, either at nodes or at numerical integra-
tion points), they can be obtained by projection, which completely eliminates the need for
central difference scheme computations on the coarse mesh.
6.5 FIELD TRANSFER IN TIME AND MINIMIZATION

PROBLEM FOR OPTIMAL TRANSFER PROCEDURE
The second way that transfer is possible is where the computations are kept on the
coarse mesh. In this procedure, the results obtained on the fine mesh are transferred,
and then computations are done on the coarse mesh. The latter can formally be written
according to:
d ncoarse
+1 (x) = ∏ (d (x )) + ∆t∏ (v (x )) + ∆t ∏ (a (x ))
x
fine
n
fine
j
x
fine
n
fine
j
2
x
fine
n
fine
j
Ma coarse
n +1 (x) = g n +1(x) − f n +1 (
ˆ int d coarse (x),… v coarse (x) =
n +1 ) (6.11)
∆t  
= v coarse
n (x) +
2 
 ∏ ( a ( x )) + a
x
fine
n
fine
j
coarse
n +1 (x)

When referring to computational efficiency, there is one crucial difference between the com-
putational procedure on fine and on coarse mesh. The computation on the fine mesh is
motivated by impact and frictional contact while the latter is driven by equivalent nodal
loading gn+1. This equivalent nodal loading is obtained as the solution of the minimization
problem seeking to render the results of two field transfer procedures as close as possible to
each other. Indicating that:
min F ( gn +1 ) (6.12)
gn + 1
having the explicit form of the cost function defined by:

T
 
∏ (d )
1
F ( g n +1 ) =
2 ∫
Ωcoarse 
coarse
 d n +1 (x) −
x
fine
n +1 (x)  ×

 
×  d ncoarse

+1 (x) − ∏ (d
x
fine
n +1 )
(x) 

(6.13)
T
   
+  v ncoarse

+1 (x) − ∏ (v
x
fine
n +1 )
(x)  ×  v ncoarse
 
+1 (x) − ∏ (v x
fine
n +1 )
(x)  +

T
    
+  a coarse

n +1 (x) − ∏ (a
x
fine
n +1 )
(x)  ×  a ncoarse
 
+1 (x) − ∏ (ax
fine
n +1 )
(x)   dV
 
The dependence of such a cost function on the equivalent load vector acting in the coarse
mesh gn+1 is defined through the corresponding central difference equation (6.11). A differ-
ent weighting factor for enforcing the equivalence of the displacement, velocity or accel-
eration fields between the coarse and the fine meshes can be selected, and in this way the
particular results could be enhanced.
Further improvement is possible if the work of the contact forces on the fine mesh are
matched as close as possible over any time step of the coarse mesh computations through
the work of equivalent loads on the coarse mesh. The latter can be obtained using the trap-
ezoidal rule for computing the work of equivalent loads on the coarse mesh, which is in
accordance with the accuracy of the central difference scheme. It can be stated that:
 tn +1 − tn T coarse
C ( g n +1 ) =
∫ 
 2
( )
fine 
gn +1v n +1 + gTn vTn − Wcontact dV (6.14)

Ωcoarse
fine
where Wcontact is the work of the contact forces which is computed on the fine mesh during
the same time step. The optimization problem in (6.12) can be customized by adding the
last condition as the constraint C ( gn +1 ) = 0. This kind of problem can be solved using the
well-known method of Lagrange multipliers (Strang 1986; Luenberger 1984) to define the
corresponding Lagrangian:
min max L ( gn +1, λ n +1 ) ; L ( gn +1, λ n +1 ) = F ( gn +1 ) + λ n +1C ( gn +1 ) (6.15)

gn + 1 λn +1
The Kuhn–Tucker equations for this most general case of constrained minimization is writ-
ten in the form:
∂L ∂F ∂C
0= := + (6.16)
∂g n + 1 ∂g n + 1 ∂g n + 1
∂L
0= := C ( g n +1 ) (6.17)
∂λ n + 1
Four different implementations of the proposed field transfer method were developed and
tested, considering different bases of the minimization problems presented herein.
The first considered the direct transfer of the nodal values computed on the fine mesh. The
second took the same cost function and added the work conservation constraint on top. The
third method seeks to improve upon the computation of the cost function by using patch-
like computations. Finally, the fourth method was using the direct transfer of the values at
the Gauss numerical points of the coarse mesh, leading in general to the highest precision
of results.
6.6 NUMERICAL EXAMPLES
6.6.1 Example 1 — Element Level

Computations with these four methods were performed for the simply supported thick
plate component impacted with a square projectile (see Figure 6.9). The results representing
Figure 6.10 (a) Computed displacement and energy with fine and coarse mesh and (b) four different project
methods—elastic case.
Figure 6.10 (Continued)
the upper and lower bonds on the result accuracy are obtained by carrying out the com-
putations of the problem on the fine and on the coarse mesh, respectively. The fine mesh
has 7,538 degrees of freedom, which are reduced to 225 in the coarse mesh. The first set
of computations, performed for the case of the elastic plate (Figure 6.10), have shown
that all the methods can give fairly good results, since a reasonably good representation
of the fundamental vibration modes is the only condition which ought to be fulfilled. The
second computation using these four methods is carried out for the case of inelastic plate
(Figure 6.11). In this case, the fourth method clearly shows far superior results, which is
the consequence of the highest level of consistency that we are able to impose for this kind
of field transfer.
Figure 6.11 (a) Computed displacement and energy with fine and coarse mesh and (b) four different project
methods—inelastic case.
6.6.2 Example 2 — Structure Level

The second example concerns the application of this strategy to a more realistic case of
application. Due to the confidential nature of the study, detailed data are not provided
(Figure 6.12), but the order of magnitude of the obtained results correspond to a quite real-
istic case. The roof slab was placed on the walls with simple supports, in a manner which
Figure 6.11 (Continued)
Figure 6.12 A structure under the impact of a large airplane (a) 3D model; (b) plane view.
Figure 6.13 Computed energy under the impact of a large airplane (a) after 350 ms; (b) after 1,000 ms.
allowed the slab to uplift and reduce negative effects in the case of an internal explosion. The
local problem of the impact on the roof slab was solved with a fine mesh including roughly
75,000 degrees of freedom. In the global computational phase, the roof slab finite element
representation was reduced in order of magnitude to close to 2,500 degrees of freedom, with
as many degrees of freedom used for the remaining part of the structure. The corresponding
strain energy plots computed at 350 ms and 1,000 ms of motion are shown in Figure 6.13.
REFERENCES
Bathe, K.-J. and E. Wilson. 1976. Numerical Methods in Finite Element Analysis. Englewood Cliffs,
NJ: Prentice-Hall.
Brancherie, D., P. Villon, A. Ibrahimbegović, A. Rassineux and P. Breitkopf. 2005. “Transfer
Operator Based on Diffuse Interpolation and Energy Conservation for Damage Materials”.
Edited by Bathe K.-J. International Journal of Heat and Mass Transfer. Amsterdam: Elsevier.
doi:10.1016/j.ijheatmasstransfer.2004.11.006.
Delaplace, Arnaud and Adnan Ibrahimbegović. 2006. “Performance of Time-Stepping Schemes for
Discrete Models in Fracture Dynamic Analysis”. International Journal for Numerical Methods
in Engineering 65 (9): 1527–44. doi:10.1002/nme.1509.
Dominguez, Norberto, Delphine Brancherie, Luc Davenne and Adnan Ibrahimbegović. 2005.
“Prediction of Crack Pattern Distribution in Reinforced Concrete by Coupling a Strong
Discontinuity Model of Concrete Cracking and a Bond‐Slip of Reinforcement Model”.
Engineering Computations 22 (5/6): 558–82. doi:10.1108/02644400510603014.
Hervé, Guillaume, F. Gatuingt and Adnan Ibrahimbegović. 2005. “On Numerical Implementation of
a Coupled Rate Dependent Damage‐Plasticity Constitutive Model for Concrete in Application
to High‐rate Dynamics”. Engineering Computations 22 (5/6).
Ibrahimbegović, Adnan. 2009. Nonlinear Mechanics of Deformable Solids: Theoretical Formulation
and Finite Element Implementation (in French). Paris: Hermes Science – Lavoisier.
88–100. doi:10.1007/s00466-002-0396-x.
Ibrahimbegović, A. and A. Delaplace. 2003. “Microscale and Mesoscale Discrete Models for Dynamics
Fracture of Structures Built of Brittle Material”. Computers & Structures 81: 1255–65.
Ibrahimbegović, A. and S. Mamouri. 2002. “Energy Conserving/Decaying Implicit Time-Stepping
Scheme for Nonlinear Dynamics of Three-Dimensional Beams Undergoing Finite Rotation”.
Computer Methods in Applied Mechanics and Engineering 191: 4241–5.
Ibrahimbegović, A., D. Markovic and F. Gatuingt. 2003. “Constitutive Model of Coupled Damage-
Plasticity and Its Finite Element Implementation”. European Journal of Finite Elements 12:
381–405.
Ibrahimbegović, A. and R. Taylor. 2002. “On the Role of Frame-Invariance in Structural Mechanics
Models at Finite Rotations”. Computer Methods in Applied Mechanics and Engineering 191:
5159–79.
Lemaitre, J. and J.-L. Chaboche. 1988. Mecanique des matriaux solides. Paris: Dunod.
Lubliner, J. 1990. Plasticity Theory. New York: Macmillan.
Luenberger, David. 1984. Linear and Nonlinear Programming. Reading, MA: Addison-Wesley.
Owen, D. and E. Hinton. 1980. Finite Elements in Plasticity: Theory and Practice. Swansea: Pineridge
Press.
Strang, G. 1986. Introduction to Applied Mathematics. Cambridge: Wellesley-Cambridge Press.
Villon, Pierre, Houman Borouchaki and Saanouni Khemais. 2002. “Transfert de Champs
Plastiquement Admissibles”. Comptes Rendus Mécanique 330 (5): 313–18. doi:10.1016/
s1631-0721(02)01457-2.
Chapter 7
Fire-induced extreme loads
How to determine the inelastic behavior of a structure subjected to mechanical and ther-
mal loads jointly applied is an important task in civil and nuclear engineering, especially
in the case of accidental loading scenarios and/or fire resistance. A vast number of cellular
structures are jointly subjected to mechanical and thermal loads, especially in the case of
accidental loading scenarios and/or fire resistance, and in this respect the development of
predictive models capable of describing inelastic behavior is needed. Most of them are built
either of folded plates and/or non-smooth shells. Heat transfer problems are dealt with
and solved in a satisfactory way for solid bodies (e.g. Armero and Simo 1992; Coleman
and Gurtin 1967; Ibrahimbegović, Lotfi Chorf and Gharzeddine 2001; Lewis et al. 1996;
Simo and Miehe 1992; Simmonds 2001; Stabler and Baker 2000), including the pertinent
aspects of thermomechanical coupling; however, for cellular structures built of clay or con-
crete hollow blocks, several novel issues arise (e.g. Bischoff and Armero 2001; Simo and
Kennedy 1992; Simmonds 2001) both in terms of describing pertinent heat transfer phe-
nomena and of accounting for thermomechanical coupling. It is envisaged that this model
should initially supplement and then eventually replace the standard testing procedure for
evaluating the fire resistance of the clay or concrete hollow blocks cellular structures. The
model should be able to account for a number of complex phenomena of heat conduc-
tion and radiation, as well as the inelastic behavior of material with thermomechanical
coupling.
Due to the complexity of the problem here only cellular structures will be treated. The
classical shell model (e.g. Naghdi 1972; Libai and Simmonds 2005; Simo, Fox and Rifai
1989) is not applicable for the modeling of folded plates or non-smooth shells. The main
difficulty in this sense is the lack of compatibility between the displacement degrees-of-free-
dom-only, which one uses to describe the membrane deformation field, and the displacement
combined with rotations, which are necessary to describe the bending deformations. This is
solved using the application of shell model with so-called drilling rotations (e.g. Hughes and
Brezzi 1989; Ibrahimbegović and Frey 1994; Gruttmann, Wagner and Wriggers 1992). From
the standpoint of the finite element implementation, the most convenient format of the shell
theory with drilling rotations is for shallow shells (e.g. Ibrahimbegović and Frey 1994). The
latter combines the kinematics hypothesis appropriate for a particular shell model, such as
the Kirchhoff hypothesis where the shear deformation is neglected or the Reissner–Mindlin
hypothesis including constant shear deformation, with the hypothesis of Marguerre on shal-
low shell geometry, which allows one to work with the projected form of the shallow shell.
165
7.1 TRANSIENT HEAT TRANSFER COMPUTATIONS
The 3D classical form of the energy balance equation when the mechanical part is ignored
is described as:
∂θ ∂q ∂θ ∂θ
c + = r ; q = −k ⇒c = −qα ,α + r (7.1)
∂t ∂α ∂α ∂t
where c is the heat capacity coefficient, θ is temperature, r is heart source, qα,α stands for
the heat flux and k is the thermal conductivity coefficient. The non-stationary heat transfer
is described by a partial differential equation featuring the temperature field θ which is a
function of space and time, as well as its partial derivatives with respect to space and with
time. Besides the boundary conditions, in order to acquire unique solutions, it is necessary
to obtain the initial conditions as well. In this respect, the initial conditions, specifying the
temperature field at the time where the heat transfer starts:
θ(x, 0) = θ0 (x) in Ω (7.2)
and the boundary conditions, specifying the value of temperature or its space derivative on
the domain boundary as:
θ ∂Ωθ = θ or qα nα ∂Ωq = qn (7.3)
In order to develop the shell-like formulation, representation of the shell 3D domain has to
be split into mid-surface A and thickness direction, and further assume (in analogy with the
mechanics part) a linear variation of the weighting temperature field:
θ* ( xα , ζ ) = ϑ* ( xα ) + ζϕ* ( xα ) (7.4)
where ϑ*is the mid-surface temperature and ϕ* is the through-the-thickness gradient. The
weak form (the advantage is the reduction of the solutions regularity) of the energy balance
equation (7.1) in a shell-like domain can be written as:
 t3 * 
0 = Gθ ( ϑ, ϕ ) :=
∫
A
 ct ϑ *
ϑ + c
12
ϕ ϕ  dA −
 ∫ (ϑ
A
*
,α α )
p + ϕ*,α rα + ϕ*p3 dA
(7.5)
 t   t  
∫
+   ϑ* + ϕ*  qn+ +  ϑ* − ϕ*  qn−  dA

A
2   2  
and by qn+ and qn− the heat fluxes on the upper and lower surface, respectively.
The resultant heat fluxes can be related to temperature and temperature gradient through
a set of constitutive relations. For example, the choice made to assure compatibility with the
resultant stress variations leads to the following form:
pα ( xα ) = −kt ϑ ,α ; p3 = ϕe3 , (7.6)
kt 3
rα ( xα ) = − ϕ,α (7.7)
12
The discrete finite element approximation is based upon the weak form. Implementation of
the thermal part of the problem is carried out by the application of the same isoparametric
Fire-induced extreme loads 167
interpolations. First of all, it is necessary to choose the finite element interpolation for the
weighting functions ϑ and φ. For the chosen four node shell element the finite element inter-
polation is:
4 4
ϑ → ϑh
Ae
= ∑
a =1
N a ( ξ, η) ϑ a , ϕ → ϕ h e =
A
∑ N (ξ, η) ϕ (7.8)
a =1
a a
ϑ a and ϕ a are the nodal values of temperature and temperature gradients, and Na are the

chosen shape functions for this type of element. By choosing the same kind of interpolations
for weighting temperature, the weak form for the thermal part in (7.5) can be reduced to a
set of algebraic equations (7.9). In order to integrate the temperature evolution over a given
time step ∆t = tn +1 − tn , a backward Euler method was selected.
f θ = 0, f θ =  faθ  ,

 ϑ − ϑ a, n t3 ϕ − ϕ a, n  (7.9)
∫
faθ =  ctN a a,n +1
A
 ∆t
+c
12
N a a, n +1
∆t
− N a,α ( pα + rα + p3 ) dA

7.2 DAMAGE MECHANISMS REPRESENTATION

UNDER INCREASED TEMPERATURE
The choice of a material model is of crucial importance for determining the most appropriate
constitutive equations for the mechanics part of the shell problem. A constitutive model able
to describe the behavior of brittle materials in terms of a generalized plasticity model based
on a Saint-Venant or Rankine-like yield criterion has been chosen. It is well known that the
von Mises plasticity criterion is used for metals, but this will not be further elaborated, as
many authors have been and are working on this matter. This model is based on limiting
the elastic domain using maximum mechanical strain values, which is in accordance with
the experimentally observed behavior of brittle materials. Failure or brittle materials are
primarily driven by extensions (positive strains) leading to cracking in the direction per-
pendicular to the principal tension stress or parallel to the principal compressive stress (e.g.
Colliat, Ibrahimbegović and Davenne 2005b). Accordingly, the elastic domain is defined in
terms of principal values by using a multisurface plasticity criterion of mechanical strain
(e.g. Colliat, Ibrahimbegović and Davenne 2005 for details). This criterion can be recast in a
stress-resultant space, leading to a multisurface yield criterion which consists of four surfaces
intersecting in a non-smooth fashion of the form (Hughes and Brezzi 1989; Ibrahimbegović,
Taylor and Wilson 1990; Ibrahimbegović and Wilson 1991; Ibrahimbegović and Frey 1993):
4µ 2µ
K+ K−
Φ1 = 3 nˆ + m
αβ ˆ αβ − αβ ˆ αβ − σ y (θ ) − q (θ ) (7.10)
3 nˆ + m
2µ I 2µ II
4µ 2µ
K+ K−
Φ2 = 3 nˆ + m
3 nˆ + m
2µ II 2µ I
4µ 2µ
K+ K−
Φ3 = 3 nˆ - m
3 nˆ + m
2µ I 2µ II
4µ 2µ
K+ K−
Φ4 = 3 nˆ − m
3 nˆ − m
2µ II 2µ I
with the stress-resultant and couple-normalized values defined with:
nαβ m
nˆ αβ = ˆ αβ = 2αβ (7.14)
,m
t t
and ⋅ I / II denoting the principal values of the symmetric tensor, and K and μ denoting,
respectively, the bulk and the shear moduli. It is for both the elastic limit σy(θ) and the vari-
able which controls the evolution of the elastic domain q(θ) that temperature dependence is
assumed. The latter is typically related to a stress softening branch which eventually drives
the stress to zero (e.g. Ibrahimbegović and Brancherie 2003). A special provision is taken
(e.g. Colliat, Ibrahimbegović and Davenne 2005) to incorporate different behaviors and cor-
responding fracture energies in compression and in tension. A plasticity model of this kind
also features the standard additive split of generalized strain measures for both membrane
and bending components with:
p p
ε αβ = ε eαβ + ε αβ , χαβ = χeαβ + χαβ (7.15)
The free energy can then be written in terms of the elastic strain components according to:
1 e 3

( e
Ψ ε αβ e
, χαβ ,ξ = ) e e t
 ε αβtCαβγδ ε γδ + χαβ
2 12
( )
C αβγδ χeγδ  +  ξ, ε eαβ (7.16)

where:
C αβγδ is the fourth-order elasticity tensor modified for plane stress case and
ξ is the internal variable which controls hardening/softening response.
Further developments of this plasticity model’s ingredients follow in the footsteps of the
work of Simo, Kennedy and Govindjee (1998). In particular, the evolution equations for
plastic components of the generalized strain measures are obtained by the Koiter rule; the
latter can be written as:
∂Φ j p ∂Φ j
ε αβ
p
= ∑ γ
j ∈Jact
j
∂nαβ
, χ αβ = ∑ γ
j ∈Jact
j
∂mαβ
(7.17)
with j ∈ J act denoting the active yield surfaces and γ j denoting the corresponding plastic
multipliers. In the determination of Jact, special care has to be provided to the case in which:
tC 0 
 ∂Φ 2 ∂Φ 2   αβγδ   ε γδ  ≤ 0 (7.18)
 ⋅ 
3
 ∂nˆ ∂m t
αβ ˆ αβ 0 Cαβγδ   χ γδ 
 12 
and γ > 0 at the same time (the same problem exists for i = 4) (e.g. Colliat, Ibrahimbegović
and Davenne 2005). The plastic multipliers are computed from the consistency condition of
the given plastic state and consistent linearization provides the elastoplastic tangent moduli.
The time-integration of the constitutive response is carried out using the backward Euler
scheme and the return mapping algorithm. The internal variables marked as e p = ε αβ p
( p
)
, n , χ αβ, n
and ξn which are known at time step tn need to be determined at the next time step tn+1.
Before moving to the plastic step, the elastic trial step has to be assumed. At this moment all
the values of internal variables at time tn remain frozen, trial values for stress resultants and
couples as well as the corresponding yield criterion values are being calculated employing
the following:
trial ∂Ψ trial ∂Ψ ∂Ψ
nαβ , n +1 = e , trial
, mαβ , n +1 = e , trial
, qntrial
+1 = e , trial
(7.19)
∂ε αβ , n +1 ∂ χ αβ, n +1 ∂ ξ αβ, n +1
giving:
(
trial trial
)
, n +1 = Φ i nαβ, n +1 , mαβ, n +1 , qn +1 , i ∈ [1...4] (7.20)
Φ itrial trial
The elastic trial step is admissible only if:
, n +1 ≤ 0, ∀i ∈ [1...4] (7.21)
Φ itrial
leading to:
p p p p
ε αβ , n +1 = ε αβ, n , χ αβ, n +1 = χ αβ, n , ξ n +1 = ξ n (7.22)
If any of the yield surfaces is active, as indicated by a positive trial value, the elastic trial step
values have to be corrected:
∃i, Φ itrial p p p p
, n +1 > 0 ⇒ ε αβ, n +1 ≠ ε αβ, n , χ αβ, n +1 ≠ χ αβ, n , ξ n +1 ≠ ξ n (7.23)
It is extremely important to emphasize that if only a single yield surface is active, this will
not necessarily imply that only a single Lagrange multiplier is non-zero, i.e.
Φ1trial trial
, n +1 > 0 and Φ 2, n +1 ≤⇔ γ 1, n +1 > 0 and γ 2, n +1 = 0 (7.24)
or:
Φ 3trial trial
, n +1 > 0 and Φ 4, n +1 ≤⇔ γ 3, n +1 > 0 and γ 4, n +1 = 0 (7.25)
By taking into account these and other possibilities to obtain the number of unknown
p p
Lagrange multipliers, a complete set of nonlinear equations is formed with ε αβ,n +1 , χ αβ,n +1 ,
ξ n+1 and λ i ,n +1 = ∆t γ i , n +1 as the unknowns.
Rn+1 (⋅) = 0 (7.26)
The elastoplastic tangent modulus needed for the incremental, iterative solution procedure
via Newton’s method can be acquired by application of the consistent linearization of the last
expression and the systematic application of the static condensation (e.g. Ibrahimbegović,
Gharzeddine and Chorfi 1998).
Lin Rni++11  = Rni +1 + DRni +1 ⇒ C nep+1 (7.27)

Finally, the consistent tangent for the mechanics part of the problem can be written as:
K uu =
∫ B C
A
T ep
BdA (7.28)
n +1

where B is the strain–displacement matrix which contains all the corresponding sub-matri-
ces for membrane and bending strain fields.
7.3 COUPLED THERMOMECHANICAL PROBLEM COMPUTATIONS
In this case, the thermomechanical coupling is taken into account for the shell model and the
constitutive model should change. The starting point in defining the constitutive response
can be selected in terms of the Helmholtz free energy which is now defined as:
( 1
)
Ψ ε eαβ , χeαβ , ξ, ϑ, ϕ α =  ε eαβtC αβγδ ε eγδ + χeαβ
2
t3
12

(
Cαβγδ χeγδ  +  ξ, ε eαβ , ϑ, ϕ α

)
(7.29)
ρc  2 t3 2   e t3 
− t ϑ + ϕ  −  ε αβt αC αβγδ δ γδ ϑ + χeαβ αCαβγδ δ γδ ϕ 
2θref  12   12 
where α a is the thermal expansion coefficient. In the last expression, the free energy is
written by adding upon the mechanics part in the first term a more general temperature
dependent form of the hardening potential in the second term; followed by the thermal
potential in the third term and the thermomechanical coupling effect in the fourth term.
With this choice of free energy, a modified form of the stress-resultant constitutive equation
is acquired which contains the temperature dependent term according to:
∂Ψ
nαβ = meca
= nαβ + t αC αβγδ δ γδ ϑ (7.30)
∂ε eαβ
∂Ψ meca t3
mαβ = = mαβ + αCαβγδ δ γδ ϕ (7.31)
∂χeαβ 12
For the case of thermomechanical coupling, the stress-resultant in (7.30) is replaced by the
weak form of the equilibrium equations in the mechanical part.
On the other hand, the weak form of the thermal balance equation will not change if the
t3
structural heating terms are neglected ( ε eαβt αC αβγδ δ γδ 0 and χ eαβ αCαβγδ δ γδ 0) and the
12
plastic dissipation. The former is done because the time variation of elastic deformations is
very slow and the latter is done because we deal herein with brittle materials which have
very small inelastic dissipation.
Moreover, with the free energy (7.30), the constitutive equations defining entropy with
both average and gradient component are gained with:
∂Ψ ∂Ψ
η= ,ς = (7.32)
∂ϑ ∂ϕ
Thermomechanical coupling still remains through boundary conditions. Namely, the classi-
cal format of radiative heat exchange is exploited herein, implying that within each cell the
resultant flux over each surface should be equal to zero, which can be written as:
nsurf
q ±
n,i ←q ±
n,i −σ ∑ ε A F (θ ) (7.33)
j =1
j j ij
± 4
j
±
where q represents the outgoing flux, and the last term represents the contribution of
n,i
incoming fluxes dependent upon the Stefan–Boltzmann constant σ, the surface emittance εj,
the area of the surface Aj, the relative shape factor of each set of surfaces Fij and the surface
temperature to the power four.
It is evident that each such equation is nonlinear in temperature terms (or in terms of
the mid-surface and temperature gradient); thus, it requires several iterations at the global
level to converge. In addition, the modification of such an equation in (7.33) has to be done
each time the fracture of the barriers changes the cell configuration. Please refer to Colliat,
Ibrahimbegović and Davenne(2005) for a more detailed discussion of this issue. A detailed
theoretical formulation is given in Section 7.6.
7.4 OPERATOR SPLIT SOLUTION PROCEDURE

WITH VARIABLE TIME STEPS
The set of equations governing the semi-discretized problem of the thermomechanical cou-
pling of shells consists of the nonlinear algebraic equations expressing mechanics equilib-
rium equations, along with differential equations describing the heat flow. These equations
are accompanied by the evolution equations of internal variables (plastic strain and hard-
ening variables), which are defined and solved at the local level, at each Gauss numerical
integration point. The problem can thus formally be written as:
r := 
( )
 r u d u , d θ , ε p (θ ) , χ p (θ ) , ξ (θ ) 
 = 0 (7.34)

(
 Mθθd θ − r θ d θ , ε p (θ ) , χ p (θ ) , ξ (θ ) ) 

 ∂Φ i 
 ε −

p
∑ γi
i
∂n 

 ∂Φ i 
 χ p −

∑ i
γ i  = 0, Φ i ≤ 0 ∀GNP (7.35)
∂m 
 
∂Φ i 
 ξ −
 ∑ i
γ i
∂q 
For the rate-independent constitutive response, the dependence of real-time is therefore pres-
ent only in the heat transfer problem. However, since all the equations are, in general, tightly
coupled, the same time parameter can be employed throughout. The system is thus first
recast in a form where only algebraic equations will appear by integrating the evolution
equations for internal variables as well as the heat transfer equation using the backward
Euler scheme. The problem thus reduces to:
(
given: d u , d θ , ε np , χnp , ξ np
θ
)∀GNP
( )
GNP
u p p p
find: dn +1, dn +1, ε n +1, χn +1, ξ n +1
GNP
such that:


(
r u dnu+1, dnθ+1, ε np+1, χnp+1, ξ n +1

) 

rn +1 :=  θ
− θ  = 0 (7.36)
θθ
M

d n + 1
∆t
d n

θ

θ
( p p
− rn +1 dn +1, ε n +1, χn +1, ξ n +1 ) 

 p ∂Φ i 

p
 ε n +1 − ε n − ∑λ i
i
∂n 

 ∂Φ i 
 χnp+1 − χnp −

∑ i
λi  = 0, Φ i ≤ 0 ∀GNP (7.37)
∂m 
 
∂Φ i 
 ξ n +1 − ξ n −
 ∑ i
λi
∂q 
It is interesting to note that the coupling effect in the discretized heat transfer equation is
present between temperatures on one side and internal variables on the other side, not only
through the plastic heating effect but also through the plastic deformation which induces
modification of the cell assembly and determines the heat radiation conditions. Therefore,
even when the plastic heating is neglected, which appears to be justified for brittle models
of this kind where no large ductile deformations occur, the coupling still persists through a
special remeshing procedure which one has to provide.
By reducing the coupling effect to a minimum, the linearized form of the system of alge-
braic equations to be solved can be written as:
K uu K uθ 
 ⋅  ∆d  = −r (k) (7.38)
u

 0 M θθ
 θ
+ K θθ   ∆d 
n +1

 ∆t  
These equations are solved for the converged values of the internal variables, which would
ensure the plastic admissibility of the stress state Φ i ≤ 0.
7.5.1 Example 1 — Circular Ring Heating

For this example, an analytical solution exists to verify the model. A circular ring with
average radius R and initial temperature θ0 is considered. With respect to the chosen refer-
ence frame and cylindrical coordinates (r, ϕ, z), temperature θ1 is applied to all points which
satisfy ϕ = 0/= 0 a h1 (see Figure 7.1).
For convenience, we recast this problem using dimensionless expressions and write:
ϕ θ − θ1 at
ϕ= ,θ = , Fo = 2 (7.39)
2π θ0 − θ1 R
where:
a is the thermal diffusivity [m 2 /s] and
Fo the Fourier number.
The analytic solution of the heat equation for the middle fiber at r = R can be written as:
∞  2 n +1 
∑
4
( )
− Fo 
θ ( ϕ, Fo ) = sin ( 2n + 1) πϕ e  2 
(7.40)
n =0
(2n + 1) π
In particular, at the point ϕ = 1 / 2 , the dimensionless temperature evolves according to:
Figure 7.1 Circular ring transient problem.
∞  2 n +1 
(−1)n  −
∑
Fo 

θ(Fo ) = e 2
(7.41)
n =0
2n + 1
Coupled thermomechanical analysis was performed limited to the elastic regime by using
the finite element model constructed with a presented shell element. Only half of the cylin-
der has been modeled, due to the symmetry, using a six element mesh (Figure 7.2).
Evolution of the bending moment at the opposite point ϕ = 1 / 2 is shown in Figure 7.3.
The peak value is reached during the transient heat transfer phase and followed by a decrease
to zero limit in the subsequent steady-state phase. Comparison between the exact solution
and the numerical results is illustrated in Figure 7.4 which pertains to the evolution of the
dimensionless temperature for the same point. An excellent match is observed. However,
this is not the case for other fibers placed at r ≠ R. The reasoning for such is that in the tran-
sient phase the temperature distribution along the r coordinates could be described using
Bessel’s functions of the first kind. In the steady-state, the limit distribution is a logarithmic
one. Thus, it is only as the thickness of the element decreases that such a limit expression
Figure 7.2 Circular ring mesh.

Figure 7.3 Bending moment.
Figure 7.4 Temperature evolution.
becomes closer to a linear distribution as assumed for shell element. It follows that the strat-
egy we developed is the most suitable for thermal analysis of thin shells.
7.5.2 Example 2 — Cellular Structures

Cellular structures as a more complex problem are of major practical engineering interest
and will be elaborated in the following section. Considering cellular structures, like hollow
bricks, the evaluation of their fire resistance leads to a coupled thermomechanical problem.
Moreover, the geometry of such a structure leads naturally to the choice of flat shell ele-
ments for constructing the finite element model. Concerning heat transfer, the large varia-
tion of temperature requires that one should take into account the radiative heat exchanges.
Like clay bricks, the flue block consists of a hollow structure made of clay, which is verti-
cally assembled in order to form a vertical stack, whose role is to evacuate smoke. The stack
is submitted to an important temperature increase on the inner face, which may lead to the
appearance of cracks and a loss of efficiency in the smoke evacuation. In order to study the
cracking of the flue block, a thermomechanical analysis employing nonlinear mechanical
model is to be employed. Only a quarter of the model is analyzed due to the symmetry of
the structure, and mesh of 16 shell elements and five solid radiative elements is used (see
Figure 7.5).
On the inner face, boundary conditions of the third kind are applied with a driven con-
vection temperature which increases from 0° to 1000° in 10 min. Figure 7.6 shows the com-
parison of the corresponding average temperature evolution at four different positions, with
decreasing values from inside to outside. It is noted that the temperature gradient between
the inner face and the outer face is still very large even after 10 min. This gradient is at the
origin of thermally induced stresses which can lead to flue cracking and eventually failure.
Since one of the interests is to eliminate the risk of cracking, it is important to have a look
into the thermal strain field and induced stresses. The stress evolution in a horizontal direc-
tion for the outer face is presented in Figure 7.7. Since the inner face is heated, this part is
generally in tension. Moreover, the supplementary global bending effect of the block is pro-
duced by the 2D nature of the problem and a non-uniform distribution of the temperature
on the inner face. Both the peak stress level and the delay effect before this peak is reached,
which are very clearly seen in Figure 7.6. The latter is a major factor in trying to estimate
the block quality.
7.5.3 Example 3 — Hollow Brick Wall

A thermomechanical coupling is considered in the cellular units placed within a brick wall.
It is assumed that the geometry and the loading allow for exploitation of the periodicity
conditions. This implies that the analysis can be carried out on a single cellular unit, isolated
Figure 7.5 Flue block-mesh.

Figure 7.6 Flue block-temperature evolution from inside to outside.
Figure 7.7 Flue block-reaction evolution.

from the whole structure at the level of interface with neighboring units, by applying the
corresponding boundary conditions which assure periodicity. More precisely, for the typical
unit assembly in a brick wall (see Figure 7.8) with only partial overlapping of successive lay-
ers, the same periodicity conditions are enforced only over half of the brick.
Therefore, the domain which is retained in the analysis corresponds to one typical unit of
the size 570×200×200 mm3. This domain also includes half of the vertical and horizontal
joints with a thickness equal to 10 mm.
The chosen finite element mesh (see Figure 7.9) consists of three vertical layers of flat shell
elements which brings the total number of this element to 384 for the entire brick. Another
subtlety of the model is the choice which is made for representing the interface joints. This
one is modeled using solid elastic elements covering the cells of the brick placed only at the
top. However, the latter does not introduce any non-symmetry into the problem, consider-
ing the periodicity in the boundary conditions.
Both mechanical and thermal loading is applied in this case. The mechanical loading is
supposed to represent the dead load on the brick chosen as a compressive loading of 1.3
MPa, which is introduced directly at the level of each element as the initial compressive load-
ing in the bricks and remaining constant afterward. The thermal loading is then applied, in
terms of the uniform temperature field applied only at the brick facet exposed to fire. The
time evolution of this temperature field is given as:
θ(t) = θ0 + 345 log(8t + 1) (7.42)
where:
θ0 is the initial temperature and
t the time in minutes.
The mechanical and thermal properties of the brick material and interface are given in
Table 7.1.
Figure 7.10 shows the location of the cells.
Figure 7.8 Unit assembly in a brick wall and periodic BC.

Figure 7.9 FE mesh for a cellular unit.
Table 7.1 Mechanical and Thermal Properties of the Brick and Interface
Mechanical and Thermal Properties Brick Interface
Density 1.870 2.100
Heat capacity 836 J kg−1 K−1 950 J kg−1 K−1
Conductivity (parallel to flakes) 0.55 Wm−1 K−1 1.55 Wm−1 K−1
Conductivity (perpendicular to flakes) 0.35 Wm−1 K−1
Thermal expansion coefficient at θref 7.10−6 K−1 1.10−6 K−1
Young’s modulus 12 GPa 15 GPa
Poisson’s ratio 0.2 0.25
σy at θref 14.5 MPa
Fracture energy 80 J m−2
Figure 7.10 Brick-cells facets locating.
First, the results are presented in terms of the temperature field. Figure 7.11 shows the
evolution of the temperature in three different cells marked in Figure 7.10. The experimental
results are provided by thermocouples inside the cells. Therefore, these values are compared
with the temperatures of the two surfaces on both sides of each cell obtained by the finite
element analysis. The comparison shows that the model is able to capture the temperature
Figure 7.11 Brick-temperature evolution and profile after 48 mm.
evolutions even far away from the exposed face of the wall. This result is confirmed by
Figure 7.11b which shows a temperature profile 48 min after the beginning of heating. The
key point here is that in order to obtain such a good result the introduction of radiative
exchanges into the heat transfer model is necessary.
From a mechanical point of view, Figure 7.12a shows the evolution of the sum of vertical
reactions at selected nodes. Each curve corresponds to a line of nodes parallel to the exposed
face of the wall and positive values are for compression (with a prestressed initial value due
to constant mechanical loading). Figure 7.12b shows the comparison of the horizontal dis-
placement of the wall built with ten rows of bricks. This bending is due to the temperature
gradient through the wall. It is shown that the stiffness provided by the analysis is quite
correct even if the displacement is slightly over-estimated by the absence of mechanical
boundary conditions.
7.6 DETAILED THEORETICAL FORMULATION OF LOCALIZED

THERMOMECHANICAL COUPLING PROBLEM
A detailed theoretical formulation for coupled thermomechanical failure problems that

take into account both the fracture process zone and softening behavior at a localized
failure zone is presented herein. A model is developed that describes the localized thermo-
mechanical failure; introducing the displacement and deformation discontinuity for the
Figure 7.12 Brick-total vertical reaction for the first lines and horizontal displacement.
mechanical part along with the discontinuity in temperature gradient for the thermal part.
A very careful choice of finite element approximation is selected in the presence of thermo-
mechanical coupling and localized failure which allows the usage of the structured mesh.
Here, enhancement of the strain field to accompany displacement discontinuity is selected,
which is needed to accommodate the temperature dependent material properties in the
fracture process zone in the presence of a non-homogeneous temperature field induced by
localized failure.
The efficiency of the numerical implementation is ensured by using the structured finite
element mesh, which is constructed by employing the finite element method with embed-
ded discontinuities (ED-FEM). As shown in Ibrahimbegović and Melnyk (2007), the pro-
posed ED-FEM has proven to be a very successful alternative to the extended finite element
method or X-FEM (Belytschko and Black 1999), providing higher computational robustness
with the discontinuities in displacement and in heat flux defined at the element level.
7.6.1 Continuum Thermo-Plastic Model and its Balance Equation

The free energy of the continuum thermo-plastic consists of three components: mechanical
energy, thermal energy and thermomechanical energy:
  θ 
( ) 1
( )
E ε − ε p − qξ + ρc (θ − θ0 ) − θ ln   
2
Ψ ε, ε p , ξ, θ =
2   θ0 

Ψm

Ψt (7.43)
− β ( θ − θ0 ) ε − ε p

( )
Ψtm
where:
E is Young’s modulus,
ε is the total strain,
εp is the plastic strain,
q is the stress-like variable associated with hardening,
ξ is the hardening variable,
ρ is the mass density,
θ is the temperature,
θ 0 is the reference temperature,
c is the density heat capacity and
β is the thermal stress (for unit temperature change in the material).
It is assumed that mechanical properties are temperature dependent.

The state equations are given as:
dΨ
σ :=
dε
( )
= E(θ) ε − ε p − β(θ) (θ − θ0 ) (7.44)
dΨ  θ
ηe := −
dθ
( )
= β(θ) ε − ε p + ρ(θ)c(θ)n   (7.45)
 θ0 
where σ is the stress and ηe is the reversible part of the entropy or “elastic” entropy (e.g.
Ibrahimbegović 2009).
The thermal stress can also be expressed in terms of the thermal expansion coefficient α:
β(θ) = E(θ)α(θ) (7.46)

By taking the last result into account, (7.44) can be rewritten in an alternative form:
 
dΨ
σ :=
dε 
(
 )
= E(θ)  ε − ε p − α(θ) (θ − θ0 ) = σ m + σth (7.47)
 ε th 
where εth denotes the thermal deformation, while σm denotes the mechanical part and σth the
thermal part of the stress. Denoting with ηp the irreversible or “plastic” part of the “total”
entropy η (with the additive split of entropy, η = ηe + ηp), the local form of internal dissipa-
tion rate can be expressed as follows:
+ σε −
0 ≤ int := θη
de + σε −
= θη
(
d Ψ + ηeθ )
(7.48)
dt dt
where e = Ψ + ηeθ is the internal energy. The additive split of the dissipation rate into a
mechanical and thermal part can be obtained with:
p + σε − d Ψ θ − d Ψ ε − d Ψp ε p − d Ψ ξ − θη
e + η
0 ≤ int = θ η ( )
e − ηeθ (7.49)
d
θ d
ε d
ε d
ξ
e
−η σ −σ −q
p
0 ≤ int = θη p + qξ (7.50)
+ σε

ther mech
The temperature dependent yield criterion for the material in the fracture process zone is
defined as:
φ ( σ, q, θ ) := σ − ( σ y (θ) − q(θ)) ≤ 0 (7.51)
where:
σy(θ) is the initial yield stress of the material at temperature θ and
q(θ) is the stress-like hardening variable controlling the evolution of the yield threshold.
The temperature dependence of these two variables is expressed in the following equations:
σ y (θ) = σ y 1 − ω (θ − θ0 ) (7.52)
q = −K(θ)ξ; K(θ) = K 1 − ω (θ − θ0 ) (7.53)
where σy and K are the values at the reference temperature θ0.

The evolution laws of the state variables are established by the second law of thermo-
dynamics, in which the internal dissipation reaches maximum value. In particular, the
Kuhn–Tucker condition is used to find the maximum of internal dissipation int among the
admissible stress values with φ ( σ, q, θ ) ≤ 0. This can be defined as the corresponding con-
strained minimization:
max int ( σ, q, θ ) ⇔ min max L ( σ, q, θ, γ ) ;

φ (σ , q, θ)≤0 (7.54)
L ( σ, q, θ, γ ) = − int ( σ, q, θ ) + γφ
( σ, q, θ )
The corresponding optimality conditions can be written as follows:

∂L ∂φ
0= → ε p = γ = γsign ( σ ) (7.55)
∂σ ∂σ
∂L ∂φ
0= → ξ = γ = γ (7.56)
∂q ∂q
∂L p = γ ∂φ = γ ( σ y + Kξ ) ω (7.57)
0= →η
∂θ ∂θ
where γ is the Lagrange multiplier.
The balance equations for this problem are obtained using the force equilibrium equation
and the first law of thermodynamics. The force equilibrium equation can be written as:
d 2u dσ
−ρ + + b = 0 (7.58)
dt 2 dx
where:
ρ is the mass density,
u is the displacement,
σ is the stress and
b is the distributed load.
The energy balance is then established by using the first principle:
d  1  du  
2
du d  du  dQ
 e + ρ   =b + σ  + R − (7.59)
dt  2  dt   dt dx  dt  dx
where:
e is the internal energy density,
R is distributed heat supply and
Q is the heat flux.
The last equation can be rewritten explicitly as:
de du  d 2 u  du d σ du ∂2u dQ
+ρ   =b + +σ +R− (7.60)
dt dt  dt  dt dx dt ∂x∂t dx
By combining this result with the force equilibrium equation, the reduced form of the first
principle is obtained:
de dε dQ
=σ +R− (7.61)
dt dt dx
By exploiting the Legrendre transformation, e = Ψ + ηeθ, the free energy potential ψ can be
further introduced:
de d Ψ e e (7.62)
= + θη + θη
dt dt
de d Ψ d Ψ dΨ dΨ e e (7.63)
= θ+ ε + p ε p + ξ + θη + θη
dt dθ dε
dε
d
ξ
− ηe σ −σ −q
Replacing this expression with (7.62), the final form of the balance equations is obtained:
e = − dQ + σε p + qξ + R (7.64)

θη
dx
mech
= − dQ + int + R (7.65)
θη
dx
It is noted that the definition of thermal dissipation in (7.48) allows for the final result in
(7.62). By further considering only quasi-static loading applications, (7.56) and (7.62) can be
recast as the final form of the balance equations:
dσ
0= +b
dx
(7.66)
= − dQ + int + R
θη
dx
7.6.2 Thermodynamics Model for Localized Failure
and Modified Balance Equation
7.6.2.1 Thermodynamics Model
When the localized failure happens, the free energy is decomposed into a regular part in the
fracture process zone and an irregular part of the free energy at the localized failure point:
( ) ( ) ( )
Ψ ε, ε p , ξ, θ = Ψ ε , ε p , θ, ξ + δ X Ψ ξ, θ1 (7.67)
where:
* denotes the regular part and
* represents the irregular part of the potential
θ denotes the temperature in any position and
θ1 denotes the temperature at the localized failure point x.
In (7.67) above, the irregular part of the energy is limited to the localized failure point by
using δ x , the Dirac-delta function:
 ∞; x = x
δ x (x) =  (7.68)
0; otherwise
The regular part of the free energy pertains to the fracture process zone, and it keeps the
same form as written in (7.43). The localized free energy is assumed to be equal to:
( )
Ψ ξ, θ1 =
1
2
K (θ1 ) ξ 2 (7.69)
where ξ is the internal variable quantifying the softening behavior due to the localized
failure. The chosen quadratic form of the softening potential in (7.69) further allows us to
obtain the corresponding stress-like internal variable:
( )
q θ, ξ :=
dΨ ξ ( ) = −K (θ ) ξ (7.70)
1
dξ
This variable drives the current ultimate stress value to zero, when the failure process is
activated, as confirmed by the corresponding yield criterion:
( ) ( ( ))
φ t x , q := t x − σ u (θ1 ) − q θ1, ξ ≤ 0 (7.71)
where:
t x is the traction at the localized failure point x
σu(θ1) is the initial value of the ultimate stress.
The mechanical properties at the localized failure are assumed to have the same depen-
dence on temperature as the bulk part; hence, it can be written in the form:
σ u (θ ) = σ u 1 − ω (θ1 − θ0 ) (7.72)
K (θ ) = K 1 − ω (θ1 − θ0 ) (7.73)
where σu and K are, respectively, the ultimate stress and softening modulus at the reference
temperature.
Once the localized failure occurs, the crack-opening (further denoted as u (t ), see
Figure 7.13) contributes to a “jump” or irregular part in the displacement field.
The total displacement field is thus the sum of the regular (smooth) part and irregular
part:
{ }
u ( x, t ) = u ( x, t ) + u (t ) H x ( x ) − ϕ ( x ) (7.74)
where H x is the Heaviside function introducing the displacement jump:
0, x ≤ x
Hx =  (7.75)
1, x > x
In (7.74) above, φ(x) is a (smooth) function, introduced to limit the influence of the dis-
placement jump within the “failure” domain. The usual choice for φ(x) in the finite element
implementation pertains to the shape functions of the selected interpolation. For example,
for a 1D truss-bar with two nodes and element length le, it can be selected as:
x
ϕ ( x ) = N2 ( x ) = (7.76)
le
The corresponding illustrations for H x (x) and φ(x) in a two-node truss-bar element are
given in Figure 7.14.
Figure 7.13 Displacement discontinuity at localized failure for the mechanical load.

Figure 7.14 Displacement discontinuity for a two-node truss-bar element: Heaviside function H x ( x ) and
φ(x).
Denoting with uˆ ( x, t ) the continuous part of the displacement field, and with u(t) the
“jump” in displacement, an additive decomposition of displacement field can be written as:
u ( x, t ) = u ( x, t ) + u (t ) H x ( x ) (7.77)
The corresponding strain field can then be obtained by exploiting the kinematic equation:
du ∂
ε ( x, t ) : = =  uˆ ( x, t ) + u (t ) H x ( x ) = εˆ + δ x ( x ) u (t ) (7.78)
dx ∂x 
The rate of internal dissipation can then be written as:
+ σε −
0 ≤ int = θη
d
dt
e ε e , ξ, ξ, θ = ( )

dt  (
+ σε − d  Ψ ε e , ξ, ξ, θ + ηeθ  =
= θη
 ) (7.79)
p + ε e σ − ∂ Ψ ε e , ξ, θ
= θη ( ) + σε p
−
∂Ψ
ξ − δx
( )
∂Ψ ξ, θ1
ξ+
 ∂ε ∂ξ ∂ξ
dϕ ( x )
+ δ x σ x u (t ) + u (t )
dx
For the elastic loading case where the rate of internal variables and the internal dissipation
are equal to zero, the constitutive stress equation reads:
σ :=
d
dε
( ) ( )
Ψ ε e , ξ, θ = E (θ ) ε − ε p − β (θ ) (θ − θ0 ) (7.80)
For the bulk material, this equation remains the same as presented in (7.44). Taking all
this into account, the final expression for the internal dissipation in a plastic loading case
can be obtained, where the correct interpretation is be given in terms of distribution (e.g.
Ibrahimbegović and Brancherie 2003):
∫ (θη )
 
int =
∫
Ωe
int dx =
Ωe
p
+ σε p + q ξ dx +  q ξ  (7.81)
 x
The evolution laws for the localized variables are established in the same way as for the clas-
sical continuum model. In particular, the evolution equation for the internal variables which
control softening can be written as:
dLΩ ∂ φ = γδ
(7.82)
0= → ξδ x = γδ x x
dq ∂q
where γ is the plastic multiplier at the point of localized failure.
7.6.2.2 Mechanical Balance Equation

The set of force equilibrium equations consists of two equations:
1. the local force equilibrium (established for the entire bulk domain)
dσ
0= + b (7.83)
dx
2. the stress orthogonality condition for defining the traction at the localized failure
point
d ϕ(x)
0 = tx +
∫
Ωe
dx
σ x dx (7.84)
7.6.2.3 Local Balance of Energy at the Localized Failure Point

For the regular part, the local energy balance is still described by a thermodynamic con-
tinuum model:
= − dQ + int + R (7.85)
θη
dx
The corresponding state equation reads:
dψ  θ
ηe = −
dθ
( )
= β ε − ε p + ρc ln   → θη
 θ0 
( )
e = θβ ε − ε p + ρcθ (7.86)
p, the local energy balance

By considering that η = ηe + ηp, int = mech + ther and ther = θη
can finally be rewritten in the format equivalent to the heat transfer equation:
ρcθ = −
dQ
dx
( )
+ mech − θβ ε − ε p + R (7.87)
where mech acts as an additional heat source. This equation holds at any point of the mate-
rial in the regular part.
It is further considered that at the localized failure point, the material has no more ability
to store heat, which implies setting the heat capacity to zero (ρc = 0). It is also taken into
account that at the localized failure point there is no heat source (R = 0) nor thermal stress
(b = 0). Therefore, the mechanical dissipation at the localized failure can be balanced only
against the change of heat flux. Moreover, the local energy balance equation at the localized
failure point ought to be interpreted in the distribution sense, resulting in the corresponding
jump in the heat flux:
 dQ   
0 = −
 dx 
x Ω ∫
+ δ x mech → Q =  q ξ  (7.88)
 x
where the mechanical dissipation mech acts as the heat source at the failure point. As indi-
cated in (7.87) above, this results in the corresponding “jump” of the heat flux through the
localized failure section. The jump in the heat flux leads to a change of the temperature gra-
dient at the localized point. In the finite element implementation, one needs additional shape
functions for the describing not only displacement but also temperature field.
7.7 EMBEDDED DISCONTINUITY FINITE ELEMENT

METHOD (ED-FEM) IMPLEMENTATION
7.7.1 Domain Definition
A 1D heterogeneous truss-bar subjected simultaneously to mechanical loading (including
distributed load b(x) and prescribed displacements at both ends) and heat transfer along the
bar (Figure 7.15) is considered. The material heterogeneity is the direct result of temperature
dependent material parameters under a heterogeneous temperature field. In particular, the
bar is built of an elastoplastic material, occupying two different sub-domains separated by
a localized failure point at x:
Ωe = Ω1e ∩ Ω2e ; Ωe = [0, le ]; Ω1e = [0, x ]; Ω2e = [ x, le ] (7.89)
The localized mechanical failure is assumed to happen at the interface x (see Figure 7.16).
In the following, the indices “1” is used for all the thermodynamics variables related to
sub-domain Ω1e , and the indices “2” to the second sub-domain Ω2e .
7.7.2 “Adiabatic” Operator Split Solution Procedure

Given positive experience (e.g. Kassiotis, Colliat and Matthies 2009), the operator split
method is selected based upon adiabatic split to solve this problem. In the most general
Figure 7.15 Heterogeneous two-phase material for a truss-bar, with phase-interface placed at x .
Figure 7.16 Two sub-domain Ω1e and Ω2e separated by a localized point at x .

case with active localized failure, the coupled thermomechanical problem is described by
a set of mechanical balance equations defined in (7.80) and (7.81), accompanied by the
energy balance equations in (7.84) and (7.85). Solving all of these equations simultane-
ously is certainly not the most efficient option. In order to increase the solution efficiency,
one can choose between two possible operator split implementations: isothermal and
adiabatic (e.g. Ibrahimbegović 2009). The isothermal operator split is not capable of
providing the stability of the computation (e.g. Kassiotis, Colliat and Matthies 2009).
Therefore, one focuses only upon the adiabatic operator split method. In this method,
the problem is divided into two phases, with each one contributing to the change of
temperature:
Phase 1—the mechanical part with Phase 2—the

“adiabatic” condition thermal part
 dσ dQ
0= +b ρcθ = − +R
 dx dx

η
 (
= 0 → ρcθ = mech − θβ ε − ε p )  
Q =  q ξ
(at the localized point): x
σ1 x = σ2 x = t x
The computations of the mechanical and thermal states remain coupled through the adia-
batic condition.
7.7.3 ED-FEM Implementation for the Mechanical Part

The basis of the numerical implementation is the weak form of the balance equations. For
the mechanical part, one can write (e.g. Ibrahimbegović 2009):
dw

∫ wbdx − ∫ dx σ dx + wb (l ) − wb (0) = 0 (7.90)
Ωe Ωe
e
where w is the virtual displacement field. In the numerical implementation, the simplest
two-node truss-bar element with linear shape functions is chosen:
x
N1 ( x ) = 1 − (7.91)
le
x
N2 ( x ) = (7.92)
le
where le is the element length. When the localized failure occurs, a displacement discontinu-
ity at the failure point is introduced, with parameter α1m (t) representing the crack-opening
displacement. The latter is multiplied using the shape function M1(x) (see Figure 7.17), in
order to limit the influence of crack-opening to that particular element. Due to the tempera-
ture dependence of the material properties, one might have potentially different values of
Young’s modulus in two parts of the element.
Considering that the stress remains continuous inside the element, as shown in
Ibrahimbegović and Melnyk (2007), one must introduce the corresponding strain discon-
tinuity at the localized failure point. This is carried out by using the shape function M 2(x)
shown in Figure 7.18 with the corresponding parameter α m 2 (t). It is noted that both M1(x)
Figure 7.17 Displacement discontinuity shape function two sub-domain M1(x).
Figure 7.18 Strain discontinuity shape function M2(x) and its derivatives.
and M 2(x) are chosen with respect to the localized failure that occurs in the middle of the
element, so that x = le / 2. Thus, the displacement field interpolation can be written as:
u ( x, t ) = N a ( x ) da (t ) + M1 ( x ) α1m (t ) + M2 ( x ) α 2m (t ) (7.93)
with
 x  le 
 − l if x ∈ 0, 2 
 e  
M1(x) = H x − N 2 (x) =  (7.94)
1 − if x ∈  e , l 
x l
 le  2 e 
 x  le 
 − l if x ∈ 0, 2 
 e  
M2 (x) =  (7.95)
 − 1 if x ∈ e , l 
x  l
 le  2 e 
The corresponding strain interpolation can then be written as:

du(x, t)
ε(x, t) = = Ba (x)da (t) + G1(x)α1m (t) + G2 (x)α m
2 (t) (7.96)
dx
 1  l  l 
 − if x ∈ 0, e  ∪  e , le 
dM1(x)  le  2 2  1 = − 1 (7.97)
G1(x) = = = G1 + δ x le ; G
dx 1
 − + δ x if x = x = e l x=
2 le
 le 2
 1  le 
 − l if x ∈ 0, 2 
dM2 (x)  e  
G2 (x) = = (7.98)
dx  1 if x ∈  le , l 
 le  2 e 
The corresponding discrete approximation of the virtual displacement and strain can be
written in an equivalent form:
w(x, t) = N a (x)wa + M1(x)β1m + M2 (x)β2m (7.99)
dw(x, t) 1(x)β1m + G2 (x)βm

= Ba (x)wa + G 2 (7.100)
dx
where β1m and βm m m
2 are the variations corresponding to α1 and α 2 , respectively. With these
interpolations on hand, the weak form of the equilibrium equation can be recast in an
incompatible mode format (Ibrahimbegović and Wilson 1991) as the set of equations:

(
 e =1 ) ∫ ( a i )
 Anel f int, e − f ext,e = 0; f int,e = Beσ d , α m , α m , θ dx
a 1 2 i
 Ωie



e e
 h1 = 0; h1 = ∑∫ G ( ) ( )
1eσi da , α1m , α 2m , θi dx + t x α1m , α 2m (7.101)
Ωie



h2e = 0; h2e = ∑∫ ( )
G2e σi da , α1m , α 2m , θi dx
 Ωie
Given highly the nonlinear material behavior, this set of equations ought to be solved using
an iterative scheme. If Newton’s method is used, systematic use of the consistent lineariza-
tion (e.g. Ibrahimbegović 2009) is used, where the corresponding incremental stress–strain
relation has to be obtained. It is noted that the chosen isoparametric elements provide con-
tinuum consistent interpolation, and furthermore that the continuum and discrete tangent
moduli remain the same in a 1D setting (e.g. Ibrahimbegović 2009). Thus, one starts with
the consistent linearization of the continuum problem to obtain the stress rate constitutive
equation, one in each sub-domain “i”:
( )
σ i = Ei ε i − ε ip − βiθ i (7.102)
The time derivative of temperature can be computed by imposing the adiabatic step:
ie = θiβi ε i − ε ip + ρiciθ i = 0 → θ i = − βiθi ε i − ε ip (7.103)

θi η ( ) ( )
ρici
Combining the last two results, finally it is obtained with:
β 2θ
( )
σ i = Cie,ad ε i − ε ip ; Cie,ad = Ei + i i (7.104)
ρici
where Cie,ad denotes the adiabatic tangent modulus. For sub-domain i, undergoing elastic
loading, with ε jp = 0, the constitutive equation can simplified as:
σ i = Cie,ad ε i (7.105)
On the other hand, if sub-domain i undergoes plastic loading, the consistency condition
requires:
d φi dφ dφ
φ i ( σi , qi , θi ) = σ i + i ξi + i θ i = 0 (7.106)
dσi dξi dθi
With the expression φi chosen herein, (7.101) can further be simplified to:
( )
sign ( σi ) σ i − Ki ξi + σiy + Ki ξi ω iθ i φ i = 0 (7.107)
By using (7.98), the constitutive equation in rate form is obtained:
(
σ i = sign ( σi ) σiy + Ki ξi ω i ) βi θi
ρici
εi
(7.108)
 βθ 
( )
+ Ki − sign ( σi ) σiy + Ki ξi ω i i i  ε ip
 ρici 
From (7.102), one has:
σ i
ε ip = ε i − e,ad (7.109)
Ci
Combining (7.103) and (7.104), the constitutive equation for a plastic domain “i” can be
formulated:
(
σ i = sign ( σi ) σiy + Ki ξi ω i ) βi θi
ρici
εi
 β θ  σ 


( ρ c
i i  
)
+ Ki − sign ( σi ) σiy + Ki ξi ω i i i   ε i − e,iad  (7.110)
C i 
KiCie,ad
σ i =
( βθ
Cie,ad + Ki − sign ( σi ) σiy + Ki ξi ω i i i
ρc
)
ii
p , ad
Ci
In conclusion, the following constitutive equation can be employed:
 C e,ad ; if φi < 0 elastic case

σ i = Ciad ε i ; Ciad =  pi ,ad (7.111)
Ci ; if φi = 0 plastic case
where Cie,ad and Cip,ad are defined in (7.99) and (7.105), respectively. To solve the problem, a
two operator split is employed (e.g. Ibrahimbegović 2009) with “local” and “global” phases
of computation. The former provides the internal variables, while the latter gives the nodal
values of displacement.
7.7.4 ED-FEM Implementation for the Thermal Part

The local form of the energy balance equation can be written as:

ie = − ∂Q + σi ε ip + qi + δ x  q ξ  + Ri (7.112)
θη
∂x  
In each of the two sub-domains, the heat transfer obeys the Fourier heat conduction law:
∂θ i
Qi = −ki (7.113)
∂x
By using the evolution equation for “elastic” entropy, the local energy balance can be rewrit-
ten in an equivalent form to the heat equation:
d 2θ i
ρcθ i = ki 2
+ R0i + mechδ x ( x ) ; mech = q ξ (7.114)
dx
where:

( )
R0i = σi ε ip + qi ξi − θiβi ε i − ε ip + Ri (7.115)
The strong form (7.114) is further transferred into the weak form by introducing an arbi-
trary temperature field, denoted as ϑ , and by applying the virtual work laws:
l
 d 2θ i 

∫0
ϑ  ρcθ i − ki
 dx 2
− R0i − mechδ x  dx = 0 (7.116)

After integration by part, we can finally obtain the following weak form:
l l l
∫ ϑ (R + δ  )dx (7.117)
d ϑ dθi

∫0
ϑρiciθ i dx +
∫
0
ki
dx dx
dx =
0
i
0 x mech
A two-node truss-bar element is considered. The nodal values of temperature and the weight-
ing temperature at node i are denoted as dθi and dϑ i , respectively. Thus, dθ and d ϑ denote the
real and the arbitrary nodal temperature vector, respectively. For a two-node element, one has:
 dθ1   dϑ1 
dθ =   ; dϑ =  
 dθ 2  dϑ 2 
The real and weighting temperature fields along the element are constructed with interpola-
tion shape functions. Furthermore, the jump of temperature gradient at the localized failure
point, is represented by an additional shape function, the same one for the real and weight-
ing temperature fields:
θi (x) = N a (x)dθa (t) + M2 (x)α t2 (t) (7.118)
ϑ(x) = N a (x)dϑa + M2 (x)βt2 (7.119)
where Na(x) and M 2(x) are defined in (7.91) and illustrated in Figure 7.18 for a two-node
truss-bar element, whereas at α t2 (t) is the variable controlling the “jump” in temperature
( )
gradient. It is noted that: θ1(t) = 1 / 2 dθ1(t) + dθ2 (t) + α 2 (t), where θ1 is the temperature at the
interface (at the middle of the element).
By considering these interpolations, the weak form (7.112) is finally reduced to:
∫ N (x)ρ c N (x)d

a i i a
a
θ + N a (x)ρici M2 (x)α 2 (t) + Ba (x)ki Ba (x)dθa
Ωie (7.120)
+ Ba (x)kiG2 (x)α 2 (t)] dx = Qθa + N a ( x ) mech
∫ M (x)ρ c N (x)d

2 i i a
a
θ + M2 (x)ρici M2 (x)α 2 (t) + G2 (x)ki Ba (x)dθa
Ωie (7.121)
+G2 (x)kiG2 (x)α 2 (t)] dx = M2 ( x ) mech

Finally, the finite element equations to be solved for the “thermal” phase are given with:
 M edeq + P eα 2e + K edeq + Feα 2e = Q1q

 eT e (7.122)
eT e
P d q + L α 2 + F d q + H α 2 = Q q
e e e e 2
where:
M e2 x 2 = ∑ ∫ N (x)ρ c N (x)dx
a i i a
Ωie
(7.123)
l  7ρ1c1 + ρ2c2 2 (ρ1c1 + ρ2c2 )
M e2 x 2 = e  
24 2 (ρ1c1 + ρ2c2 ) ρ1c1 + 7ρ2c2 
P1ex 2 = ∑ ∫ N (x)ρ c M (x)dx

a i i 2
Ωie
(7.124)
le 2ρ1c1 + ρ2c2 
P1ex 2 = −  
24 ρ1c1 + 2ρ2c2 
K e2 x 2 = ∑ ∫ B (x)k B (x)dx
a i a
Ωie
(7.125)
K e2 x 2 =
(k1 + k2 )  1 −1
2le  −1 1 

F1ex 2 = ∑ ∫ B (x)k G (x)dx

a i 2
Ωie
(7.126)
k1 − k2 1
F1ex 2 =  
2le  −1
Le1x1 = ∑ ∫ M (x)ρ c M (x)dx

2 i i 2
Ωie
(7.127)
le
Le1x1 =
24
(ρ1c1 + ρ2c2 )
H1e x1 = ∑ ∫ G (x)k G (x)dx

2 i 2
Ωie (7.128)
H1e x1 =
(k1 + k2 )
2le
Q1q1x 2 = Qθa + N a ( x ) mech ;
1 (7.129)
le 3R01 + R02  1
Q1q1x 2 =  1 2
+ mech  
8 R0 + 3R0  2 1
Q q21x1 = M2 ( x ) mech = −0.5mech (7.130)
There are many methods capable of solving the time-dependent equation (7.117) (e.g.
Ibrahimbegović 2009). Here the Newmark integration scheme is chosen. Assuming that the
heat transfer problem lasts for a duration [0,T], This duration can be divided into n incre-
ments: [t0 = 0, t1,..., ti ,..., tn -1, tn = T ] with the time step h = ti +1 − ti .
Given the state variables at time i:
di = dθ ( ti ) , di = υi = d θ ( ti )

αi2 = αi2 ( ti ) , α 2 ( ti )
i2 = i2 = α
Find the state variables at the time i + 1:
di + 1 = dθ ( t i + 1 )
υi +1 = di +1 = dθ (ti +1 )
α i2+1 = α 2 (ti +1 )
α i2+1 = i2+1 = α i2 (ti +1 )
γ γ
By considering the equation of Newmark: ∆υi = ∆d and ∆ = ∆α 2 and by lineariza-
tion, (7.75) becomes: β∆ t β∆t
 γ   γ e 
 M e + K e  ∆d +  P + Fe  ∆α 2 = R1i
  β∆t   β∆t 
 (7.131)
 γ P eT + FeT  ∆d +  γ Le + H e  ∆α = Ri
 β∆t   β∆t  2 2

where the residuals are computed using the following equation
 R1i = Qθ1 − M evi − P ei2 − K edi − F eα i2

 i (7.132)
2 eT e i eT e i
R2 = Qθ − P vi − L 2 − F di − H α 2
From (7.84)2 , the expression of Δα2 from Δd is established:
−1
 γ e 
∆α 2 =  L + He R2i
 β∆t 
(7.133)
−1
 γ e   γ eT eT 
− L + He  β∆t P + F  ∆d
 β∆t 
Combining (7.133) and (7.131)1, Δd is obtained from the current nodal temperature and the
current value of heat flux “jump” as the solution of:
 γ  γ e
−1

e e e  γ e  γ eT T 
 M +K − P +F  e
L +H   P + F e   ∆d =
 β∆t  β∆t   β∆t   β∆t 

(7.134)
−1
 γ e  γ e 
R1i −  P + Fe   L + He R2i
 β∆t   β∆t 
or:
ˆ
Κ∆d ˆ
= F (7.135)
Once Δd is known, Δα2 can be computed. The nodal temperature and the “jump” in heat
flow through the localized mechanical failure at the next time step can be updated using the
formula:
di +1 = di + ∆d , α i2+1 = α i2 + ∆α 2
7.8.1 Example 1 — Simple Tension Imposed

Temperature Example with a Fixed Mesh
In this example, a steel bar 5 mm long is elaborated. The bar is built-in at the left end and
subjected to an imposed displacement at the right end. The imposed displacement increases
1.6 × 10 −4 mm in each step. Simultaneously, the right end of the bar is heated, and its temper-
ature is raised from 0°C to 1000°C, with a 10°C increase in each step. The temperature at
the left end is kept equal to 0°C. The loading increases until the localized failure of the bar.
The problem geometric data and loading program are described in Figures 7.19 and 7.20,
respectively.
The problem is subsequently considered for three different variations of the material prop-
erties: (i) the material properties are independent of temperature, (ii) the material proper-
ties are linearly dependent on temperature and (iii) the material properties are non-linearly
dependent on temperature (following a suggestion given by regulation Eurocode (CEN-
Eurocode1 2003)).
7.8.1.1 Material Properties Independent on Temperature

In this case, the material properties of the bar are assumed to be constant regardless of the
temperature change. The chosen values for the material parameters are given in Table 7.2.
Figure 7.19 Bar subjected to imposed displacement and temperature applied simultaneously.

Figure 7.20 Time variation of imposed displacement and temperature.
Table 7.2 Material Properties of Steel Bar

Material Properties Value Dimension
Young’s modulus (E) 205 000 MPa
Initial yield stress (σy) 250 MPa
Ultimate stress (σu) 300 MPa
Plastic hardening modulus (Kp) 20 000 MPa
Localized softening modulus K

Mass density (ρ)
() −30 000
7 865 × 10−9
MPa m−1
Ns−2 mm−4
Thermal conductivity (k) 45 Ns−1 K−1
Heat specific (c) 0.46 × 109 mm2 s−2 K−1
Thermal elongation (α) 0.00001
The computed results for stress–strain curves in two sub-domains are presented in
Figure 7.21, while the force–displacement curve of the bar is given in Figure 7.22.
In Table 7.3 and Figure 7.23, the resulting time evolution of temperature and its distri-
bution along the bar is presented. For the case with material properties independent on
temperature, it can be concluded that there is no difference in the strain values between the
two sub-domains. The “jump” in temperature gradient (αth), which appears at the local-
ized failure point, also remains very small. The computed dissipation due to plasticity in the
Figure 7.21 Stress–strain curves in two sub-domains (line for 1st sub-domain, squares for 2nd).
Figure 7.22 Force–displacement curve of the bar.
Table 7.3 Time Evolution of Temperature along the Bar

Time θ at x = 0 θ at x = 0.25le θ at x = 0.50le θ at x = 0.75le θ at x = le Δθ
0.0 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
0.1 0.0000 25.0000 50.0000 75.0000 100.0000 0.0000
0.2 0.0000 50.0000 100.0000 150.0000 200.0000 0.0000
0.3 0.0000 75.0000 150.0000 225.0000 300.0000 0.0000
0.4 0.0000 100.0000 200.0000 300.0000 400.0000 0.0000
0.5 0.0000 125.0000 250.0000 375.0000 500.0000 0.0000
0.6 0.0000 150.0000 300.0000 450.0000 600.0000 0.0000
0.7 0.0000 175.0005 350.0000 525.0005 700.0000 0.0000
0.8 0.0000 200.0007 400.0014 600.0007 800.0000 0.0014
0.9 0.0000 225.0008 450.0015 675.0008 900.0000 0.0015
1.0 0.0000 250.0008 500.0016 750.0008 1000.0000 0.0016
Figure 7.23 Distribution of temperature (°C) along the bar at chosen values of time.
fracture process zone is 36.63 Nmm, while the dissipation due to localized failure is 29.44
Nmm. In summary, the total mechanical dissipation in the bar is equal to 66.07 Nmm.
7.8.1.2 Material Properties are Linearly Dependent on Temperature

In this example, the same mechanical material properties of the steel bar are chosen as in
the first example (see Table 7.2), and they hold only at a reference temperature (equal to 0°C)
(see Figure 7.24).
Figure 7.24 Evolution of Δθ vs. time (in K).
For other temperature values, they vary linearly according to the expression given in Table 7.4.
The thermal material properties are independent on temperature and equal to those in the
first example. The resulting stress–strain curves in the two sub-domains and the resulting
force–displacement diagram are presented in Figures 7.25 and 7.26, respectively.
The distribution of temperature along the bar is shown in Table 7.5 and Figure 7.27.
In this example, the total plastic dissipation and the total localized dissipation are 14.08
Nmm and 13.82 Nmm, respectively. Thus, the total mechanical dissipation is equal to
27.90 Nmm.
Table 7.4 Material Properties of Steel Bar Linearly Dependent

Material Properties Expression Dimension
Young’s modulus E(θ) E (θ) = 2.05 × 105 (1 − 0.0008θ) MPa
Initial yield stress σy(θ) σ y (θ) = 250 (1 − 0.001θ) MPa
Ultimate stress σu(θ) σu (θ) = 300 (1 − 0.0015θ) MPa
Plastic hardening modulus Kp(θ) K p (θ) = 2 × 10 (1 − 0.0008θ)
4
MPa
Localized softening modulus K (θ) K (θ) = −3 × 10 4 (1 − 0.0008θ) MPa m−1
Figure 7.25 Stress–strain curves in two sub-domains (bold line for 1st sub-domain, squared line for 2nd).
Figure 7.26 Force–displacement curve.
Table 7.5 Time Evolution of Temperature along the Bar

0.000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
0.1000 0.0000 25.0000 50.0000 75.0000 100.0000 0.0000
0.2000 0.0000 50.0000 100.0000 150.0000 200.0000 0.0000
0.3000 0.0000 75.0000 150.0000 225.0000 300.0000 0.0000
0.3500 0.0000 87.5000 175.0000 262.5000 350.0000 0.0000
0.4000 0.0000 100.0000 199.9999 300.0000 400.0000 −0.0001
0.4500 0.0000 112.5002 225.0005 337.5002 450.0000 0.0005
0.5000 0.0000 125.0004 250.0007 375.0004 500.0000 0.0007
0.5500 0.0000 137.5004 275.0008 412.5004 550.0000 0.0008
0.6000 0.0000 150.0004 300.0008 450.0004 600.0000 0.0008
0.6300 0.0000 157.5004 315.0008 472.0004 630.0000 0.0008
Figure 7.27 Evolution of temperature (°C) along the bar in time.
From the results presented in the figures above, it can be clearly concluded that the tem-
perature variations deeply influence the behavior of the bar. In particular, the displace-
ment at the end of the bar when failure occurs reduces from 0.016 mm to 0.011 mm, the
initial yield stress falls down to approximately 225 MPa from 250 MPa and so the ultimate
strength reduces from 300 MPa to about 220 MPa. The total dissipation in this example is
also reduced, from 66.07 Nmm to 27.90 Nmm. Figure 7.25 indicates that the variation of
the temperature field leads to a significant difference in the material behavior and computed
stress–strain curves in two parts of the bar. As seen from Figure 7.28, the “jump” in tem-
perature gradient accompanying localized failure remains relatively small.
Figure 7.28 Evolution of Δθ versus time (in K).
7.8.1.3 Material Properties Non-Linearly Dependent

on Temperature (Eurocode 1993-1-2)
In Eurocodes (CEN-Eurocode1 2003), the material properties of the steel bar subjected to
thermal loading are not constant but dependent on temperature as multi-linear functions.
Based on those regulations, the evolution of mechanical properties as functions of tempera-
ture can be established as shown in Table 7.6.
where ω* are the temperature dependent coefficients. The values of the temperature
dependent coefficients for yield stress, ultimate strength and Young’s modulus are taken
from Eurocode 1993-1-2. The corresponding values of the coefficients for the plastic
hardening modulus and localized softening modulus are taken as the same as the one for
Young’s modulus. All the values used for these coefficients are presented in Table 7.7 and
Figure 7.29.
The evolution of thermal properties is also taken from Eurocodes (CEN-Eurocode1 2003).
Thermal Elongation
1.2 × 10−5 θa + 0.4 × 10−8 θa2 − 2.416 × 10−4 , if 20°C ≤ θa < 750°C

α = 1.1 × 10−2 , if 750°C ≤ θa < 860°C
2 × 10−5 θ − 6.2 × 10−3 , if 860°C ≤ θa < 1200°C
 a
Specific Heat
425 + 7.73 × 10−1 θ a − 1.69 × 10−3 θ2a

 if 20°C ≤ θ a < 600°C
+2.22 × 10−6 θ3a J / (kgK),

 13002
c = 666 + 738 − θ J / (kgK), if 600°C ≤ θ a < 735°C
a

0
545 + 1782 J / (kgK), if 735°C ≤ θ a < 900°C
 θ a − 731

650 J / (kgK), if θ a ≥ 900°C
Thermal Conductivity
54 − 3.33 × 10−2 θ a W/(mK), if 20°C ≤ θ a < 800°C

k=
27.3 W/(mK), if 800°C ≤ θ a ≤ 1200°C
Table 7.6 Material Properties of a Linearly Dependent Steel Bar

Material Properties Expression Dimension
Young’s modulus E(θ) E (θ) = 2.05 × 105 (1 − ω E (θ − 20)) MPa
Initial yield stress σy(θ) (
σ y (θ) = 250 1 − ω σ y (θ − 20) ) MPa
Ultimate stress σu(θ) σu (θ) = 300 (1 − ω σu (θ − 20)) MPa
Plastic hardening modulus Kp(θ) (
K p (θ) = 2 × 10 4 1 − ω Kp (θ − 20) ) MPa
Localized softening modulus K (θ) (

K (θ) = −3 × 10 4 1 − ω K (θ − 20) ) MPa m−1
Table 7.7 Temperature Dependent Coefficients
θ(°C) ω σu ωσy ωE ω Kp ωK
0.000 0.00000 0.00000 0.00000 0.00000 0.00000
20 0.00000 0.00000 0.00000 0.00000 0.00000
100 0.00000 0.00000 0.00000 0.00000 0.00000
200 0.00000 0.00107 0.00056 0.00056 0.00056
300 0.00000 0.00138 0.00071 0.00071 0.00071
400 0.00000 0.00153 0.00079 0.00079 0.00079
500 0.00046 0.00133 0.00083 0.00083 0.00083
600 0.00091 0.00141 0.00119 0.00119 0.00119
700 0.00113 0.00136 0.00128 0.00128 0.00128
800 0.00114 0.00122 0.00117 0.00117 0.00117
900 0.00107 0.00109 0.00106 0.00106 0.00106
1000 0.00098 0.00099 0.00097 0.00097 0.00097
1100 0.00091 0.00091 0.00091 0.00091 0.00091
1200 0.00085 0.00085 0.00085 0.00085 0.00085
Figure 7.29 Temperature dependent coefficients (according to Eurocode 1993-1-2).
The main results obtained considering those evolutions are described subsequently in terms
of the stress–strain curves, force–displacement diagram and corresponding temperature
variations.
Figure 7.30 clearly shows the large difference in strain between the two sub-domains,
both before and after the initiation of a localized failure. Mathematically, this difference is
due to different values of α1m and α m
2 (see (7.93)). Before the initiation of localized failure,
the difference in temperature will lead to a difference in tangent modulus between the two
sub-domains, which results in the appearance of α m 2 which represents the difference in strain
Figure 7.30 Stress–strain curves in two sub-domains (squared line for subdomain 1, bold line for subdomain 2).
between the two sub-domains. After a localized failure occurs, α1m increases and contributes
to the different behaviors in two parts of the bar (see Figure 7.31).
From Table 7.8 and Figure 7.32, it can be seen that the temperature distribution is nonlin-
ear. Its gradient changes in the middle of the bar. This change can be computed through α2
(see (7.71)). It is noted that the magnitude of α2 increases and then decreases with time (see
Table 7.8 and Figure 7.33). However, Figure 7.21 also shows that the change in temperature
gradient is relatively small in comparison to the temperature at the localized failure point
(the maximum ratio of ∆θ / θl (θl is the temperature of the localized point) is approximately
0.0136%), and therefore does not significantly contribute to the final results.
In this example, once again, a reduction in the strength of the bar is observed (see
Figure 7.31): the maximum displacement that can be applied to the bar now reduces to
roughly 0.006 mm from 0.010 mm and 0.016 mm in the second and the first example,
respectively.
Figure 7.31 Force–displacement diagram for the bar.

Table 7.8 Distribution of Temperature Along the Bar

0.000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
0.0500 0.0000 12.5000 25.0000 37.5000 50.0000 0.0000
0.1000 0.0000 25.0000 50.0000 75.0000 100.0000 0.0000
0.1500 0.0000 37.5000 75.0000 112.5000 150.0000 0.0000
0.2000 0.0000 49.9998 99.9996 149.9998 200.0000 −0.0004
0.2500 0.0000 62.5020 125.0041 187.5020 250.0000 0.0041
0.3000 0.0000 75.0053 150.0106 225.0053 300.0000 0.0106
0.3500 0.0000 87.5094 175.0188 262.5094 350.0000 0.0188
0.3900 0.0000 97.5133 195.0265 292.5133 390.0000 0.0265
Figure 7.32 Distribution of temperature (°C) along the bar due to time.
Figure 7.33 Evolution of Δθ vs. time (in K) where we computed ∆θ = θ x = 0.5le − 0.5(θ x = 0 + θ x = le ).

The total mechanical dissipation along the bar is significantly smaller than the second and
the first example (15.01 Nmm in comparison to 27.90 Nmm and 66.07 Nmm). The majority
contribution comes from the localized dissipation: 10.55 Nm in comparison with the total
plastic dissipation: 4.47 Nm.
7.8.1.3.1 Mesh Refinement, Convergence and Mesh Objectivity

The influence of the chosen number of elements upon the computed final results is presented
further. The geometry description is given in Figure 7.34.
A steel bar is considered built-in at the left end and subjected to an imposed displacement
at the right end (increasing linearly to 2 mm). Simultaneously, the right end of the bar is
heated, and its temperature is raised from 0°C to 100°C. The temperature of the left end is
kept constant and equal to 0°C. The material properties of the bar are considered as tem-
perature independent and are the same as in Table 7.2, except localized softening failure
which amounts to (K) = 45 MPa m −1.
Figure 7.35 shows the load–displacement diagram of the bar computed by using 3, 5, 7
and 9 elements. From this figure it is clearly noted that the computed curve after localized
failure is not dependent on the chosen mesh). This result proves the convergence of the
numerical solution with respect to mesh refinement (Ibrahimbegović 2009).
7.8.1.3.2 Heating Effect of Mechanical Loading

In this example, the heating effect produced by mechanical dissipation in a bar when a
localized failure occurs is illustrated. A steel bar of 10 mm long, fixed at the left end and
Figure 7.34 Bar subjected to imposed loading and imposed temperature.
Figure 7.35 Load–displacement diagram with a different number of elements.

subjected to an increasing displacement (0.045 mm/s) at the right end until collapse is
elaborated. The initial temperature is constant along the bar and equal to 0°C. Material
properties of the bar are given in Table 7.2. Due to a problem in manufacturing, the ulti-
mate stress at the middle point reduces to 299 MPa instead of 300 MPa in other parts (see
Figure 7.36).
This problem is solved with two different meshes: five elements and nine elements. In
these two meshes, the middle element represents the zone with a smaller ultimate stress (σu =
299 MPa). The localized failure will, therefore, occur in this element. The computed load–
displacement diagram of the bar is given in Figure 7.37, while the evolution of temperature
in the bar is shown in Figures 7.38 and 7.39.
The computed results clearly demonstrate the heating effect produced by the mechanical
dissipation. Namely, the plastic dissipation equals the heat supply leading to the temperature
increase. Initially, the dissipation in FPZ is equally distributed along the bar, so that the
temperature at every part of the bar remains the same. However, with the start of localized
failure, additional dissipation at the failure point acts as a concentrated heat supply. This
further leads to a heat transfer process in the bar and results in the evolution of temperature,
as shown in Figures 7.38 and 7.39.
Figure 7.36 Description of the example and its mesh.
Figure 7.37 Load–displacement curve.

Figure 7.38 Temperature evolution along the bar before and after the localized failure occurs (computed
with five elements mesh).
Figure 7.39 Temperature evolution along the bar before and after the localized failure occurs (computed
with nine elements mesh).
7.8.2 Model Generalization to a 2D/3D Case: Formulation,

Implementation and Numerical Results
7.8.2.1 Theoretical Formulation
The balance equations in a 2D/3D case are written the same as the corresponding general-
ization of a 1D case employing the tensors; this applies both to the strong form in (7.63) or
the weak form in (7.85).
The first ingredient of the theoretical formulation in a 2D/3D case is the free energy,
which can be written as an additive decomposition of the regular and irregular part:
( ) ()
ψ = ψ ε, ε p , ξ, θ + δ Γ s ( x ) ψ ξ (7.136)
where Γs denotes the localized failure surface. The regular part of the free energy is the
appropriate generalization of (7.43), with the strain fields represented by the second-order
tensors (e.g. Ibrahimbegović 2009), which can be written as:
  θ 
Ψ=
1
( ) ( )
ε − ε p : C ε − ε p + Ξ (θ, ξ ) + ρc (θ − θ0 ) − θ ln   
2  θ0 

 (7.137)
Ψm
Ψt

(
× − ( θ − θ0 ) β : ε − ε
p

)
Ψth
The constitutive equations for the stress tensor are the tensor generalization of the 1D result
in (7.44):
dΨ
σ :=
dε
( )
= C ε − ε p − (θ − θ0 ) β (7.138)
where the second-order stress tensor σ is defined in terms of the fourth-order elasticity
tensor C and the second-order strain tensors ε, εp, along with the second-order tensor of
thermal stress β. It is noted that in a 2D/3D case, the temperature is still a scalar field, and
so is the elastic entropy:
dΨ  θ
ηe := −
dθ
( )
= β : ε − ε p + ρc ln  
 θ0 
(7.139)
The second main ingredient of the model is the yield criterion, indicating the creation of the
FPZ, which can be written in terms of the chosen norm of the stress tensor and hardening
variable q:
∂Ξ
φ ( σ, q, θ ) = σ − ( σ y − q) ≤ 0; q := − = −Κξ (7.140)
∂ξ
The stress orthogonality condition in (7.81) can be generalized to 2D/3D to define the stress
traction vector:
tΓs =
∫ G σ
Γs
Γs d Γ s (7.141)
The latter is then used to define the corresponding yield criteria indicating the localized
failure:
( )
φ1 ( σ, q, θ ) = t Γ s ⋅ n − σ f − q ≤ 0 (7.142)
where σ f is the ultimate stress at the start of localized failure, and q is the softening variable
controlling the evolution of plasticity threshold; the latter can be computed by assuming the
simplest quadratic form of the softening potential:
( )
ψ ξ, θ1 =
1
K (θ1 ) ξ 2 → q (θ ) := −
dψ ξ ()
= K ξ (7.143)
2 dξ
The second law of thermodynamics can be used to obtain the total dissipation written as
the additive decomposition of the regular part and irregular part in terms of distribution:

p + σ : ε p + qξ +  q ξ  δ (7.144)
int = θη
  Γ s
int
int
7.8.2.1.1 Finite Element Approximation

Once the localized failure occurs over surface Γs, the displacement field can be split into a
regular part and an irregular part, which controls the strong discontinuity at the localized
failure surface:
( )
u ( x, t ) = u ( x, t ) + u ( x, t ) Γ s ( x ) − ϕ ( x ) (7.145)
The complete domain Ω is thus decomposed into two sub-domains Ω+ and Ω−, and the cor-
responding choice of φ(x) is made to limit the influence of the particular discontinuity.
The discrete approximation is constructed by using the structured finite element mesh
so that one again needs the embedded discontinuities to represent the localized failure sur-
face. In term of the finite element approximation, the “jump” in the displacement field due
to the localized failure is presented using a method of incompatible modes (e.g. Baker and
Borst 2005; Bathe 1996). In particular, for a 2D case, and the simplest choice of discrete
approximation based upon the three node CST element (e.g. see Bathe 1996), one can write
the corresponding FE interpolation of displacement field:
3
u ( x, y ) = ∑ N ( x, y) d
a =1
a a + M1 ( x, y ) u + M2 ( x, y ) α 2a (7.146)
where:
Na are standard linear polynomial shape functions;
M1(x) is the approximation of displacement jump, and
M 2(x) is the jump in the derivative.
It is noted that the choice of these incompatible modes is set with respect to the localized
failure surface (see Figure 7.40).
A similar enhancement is chosen for constructing the discrete approximation of the tem-
perature field within the element that is crossed by the localized surface failure. More pre-
cisely, it can be written:
3
θ ( x, y ) = ∑ N ( x, y)d
a =1
a
θ
a + M2 ( x, y ) α 2a = N a daq + M2 ( x, y ) α 2a (7.147)
The final product of the finite element discretization is a set of nonlinear algebraic equa-
tions, with extra unknowns for each element crossed by the localized failure surface. The
system is fully equivalent to the system in (7.96). Given this, one can again exploit the adia-
batic operator split solution procedure by closely following the steps of development already
presented in detail for a 1D case.
7.8.2.1.2 Numerical Results
Evaluating the fire resistance of RC structures in current practice relies upon the standard
fire test, specified by ISO 834. However, the standard fire resistance tests should not be the
only ones relied upon in order to determine the survival time of concrete elements under
realistic fire scenarios, since they can often define less severe heating environments than
Figure 7.40 Additional shape functions for representing displacement and temperature fields.
real fires. One example from the nuclear industry with a more demanding fire scenario, as
illustrated in Figure 7.41, considers more rapid time-evolutions of the temperature and heat
flux than the standard tests.
The first example deals with plain concrete. More precisely, a concrete slab is considered
(with no reinforcement) with dimensions of 2350 mm×2350mm×300 m. The behavior of
the slab is modeled using the thermoplasticity model proposed in Ibrahimbegović, Colliat
and Davenne (2005), which considers that the yield criterion is set in terms of the principal
values of stress. The aim is to check if the proposed thermoplasticity model for localized fail-
ure can reproduce the corresponding fire-induced damage conditions in a slab submitted to
a temperature gradient through the thickness. For that reason, the slab is heated on one side
using the critical fire scenario for nuclear installation shown in Figure 7.41. This produces
the bending effect in the slab and the typical damage pattern of tensile failure, as shown in
Figure 7.42. It can be seen that the localized failure surface (presented by a corresponding
red lines approximation in each element) is typical of slab failure in bending.
One can provide a more thorough analysis of the computed result. Namely, the chosen sce-
nario characterizes fires where very high temperatures are reached in a few seconds. Due to this
quick thermal load increase, leaving no time for diffusion, a high thermal gradient is located in
a thin layer at the bottom of the slab. This kind of loading generates compressive stresses at the
Figure 7.41 Fire scenario for nuclear installation in terms of temperature and flux variations.
Figure 7.42 Localized failure surface (cracks) in a plain concrete slab under a temperature gradient produced
by the heating of one side (bold lines indicate approximations to localized failure surface).
bottom, accompanied by traction stresses in the surface layer producing cracks in the horizon-
tal direction. Thus, it is indeed possible to reproduce the mechanism of spalling, by connecting
these cracks where one would finally obtain a part of the surface concrete layer to detach from
the rest of the structure. This result corroborates the mechanical theory of spalling induced
by steep temperature variation. The further temperature increase leads to a dominant vertical
direction of crack propagation due to slab bending. Computed crack-opening, in the range
of 1–4 mm, can be considered critical in nuclear installations in that it is sufficient to allow
nuclear waste to dissipate. The crack-spacing corresponding to pure bending remains uniform,
which is the consequence of a homogeneous temperature field imposed on the bottom of the
slab. It is noted that this result on crack-spacing can further change for a particular fire scenario
with a concentrated fire source producing a heterogeneous temperature field.
In the second example, the fire-induced cracking in a reinforced concrete slab is con-
sidered, which illustrates the positive influence of reinforcement in preventing or limiting
Figure 7.43 Localized failure surface (cracks) in a reinforced concrete slab under a temperature gradient
produced by the heating of one side (bold lines indicate approximations to localized failure surface).
the spalling phenomena. The reinforced concrete slab has the same dimensions as in the
previous example. The main difference concerns the presence of steel reinforcement with
bars of 16 mm diameter at each 10 cm, with a concrete cover of 30 mm. The reinforcement
is modeled with the 1D model described at the beginning, and directly connected to con-
crete elements (practically assuming the perfect bond). This reinforced concrete slab is then
loaded with critical thermal loading as shown in Figure 7.41, and the computed results are
presented in Figure 7.43.
In comparison with the first example, with a plain concrete slab, the crack pattern for a
reinforced concrete slab is much less pronounced, which confirms that the reinforcement
plays an important role. First, the vertical cracks are fewer and of smaller width. Second,
there are no horizontal cracks which connect, and thus no mechanical spalling would occur
in a case when reinforcement is used to strengthen the concrete slab.
REFERENCES
Armero, F. and J. C. Simo. 1992. “A New Unconditionally Stable Fractional Step Method for Non-
Linear Coupled Thermomechanical Problems”. International Journal for Numerical Methods
in Engineering 35 (4): 737–66. doi:10.1002/nme.1620350408.
Baker, G. and R. D. Borst. 2005. “An Anisotropic Thermomechanical Damage Model for Concrete
at Transient Elevated Temperatures”. Philosophical Transactions of the Royal Society
A: Mathematical, Physical and Engineering Sciences 363 (1836): 2603–28. doi:10.1098/
rsta.2005.1589.
Bathe, K. J. 1996. Finite Element Procedures. Upper Saddle River, NJ: Prentice-Hall.
Belytschko, T. and T. Black. 1999. “Elastic Crack Growth in Finite Elements with Minimal
Remeshing”. International Journal for Numerical Methods in Engineering 45 (5): 601–20.
doi:10.1002/(sici)1097-0207(19990620)45:5<601::aid-nme598>3.0.co;2.
Bischoff, M. and F. Armero. 2001. Stabilised finite element methods for the analysis of heat transfer
in shells. ECCM-2001. Crackow, Poland.
Coleman, Bernard D. and Morton E. Gurtin. 1967. “Thermodynamics with Internal State Variables”.
The Journal of Chemical Physics 47 (2). doi:10.1063/1.1711937.
Colliat, J. B., A. Ibrahimbegović and L. Davenne. 2005a. “Heat Conduction and Radiative Heat
Exchange in Cellular Structures Using Flat Shell Elements”. Communications in Numerical
Methods in Engineering 22 (3): 167–80. doi:10.1002/cnm.784.
Colliat, J.-B., A. Ibrahimbegović and L. Davenne. 2005b. “Saint‐Venant Multi‐surface Plasticity
Model in Strain Space and in Stress Resultants”. Engineering Computations 22 (5/9): 536–57.
doi:10.1108/02644400510603005.
EN 1991-1-5 (2003). 2003. “(English): Eurocode 1: Action on structures. Part 1–5: General actions –
thermal actions”. Committee European de Normalization: Bruxelles, Belgium.
EN 1993-1-2 (2005). 2004. “(English): Eurocode 3: Design of steel structures. Part 1–2: General
rules–Structure fire design”. Committee European de Normalization: Bruxelles, Belgium.
Gruttmann, F., W. Wagner and P. Wriggers. 1992. “A Nonlinear Quadrilateral Shell Element with
Drilling Degrees of Freedom”. Archive of Applied Mechanics 62 (7): 474–86. doi:10.1007/
bf00810238.
Hughes, T. J. R. and F. Brezzi. 1989. “On Drilling Degrees of Freedom”. Computer Methods in
Applied Mechanics and Engineering 72 (1): 105–121.
Ibrahimbegović, A., J. B. Colliat and L. Davenne. 2005. “Thermomechanical Coupling in Folded
Plates and Non-Smooth Shells”. Computer Methods in Applied Mechanics and Engineering
194 (21–24): 2686–707. doi:10.1016/j.cma.2004.07.052.
88–100. doi:10.1007/s00466-002-0396-x.
Element Solution Methods. Berlin: Springer.
Communications in Applied Numerical Methods 7 (3): 187–94. doi:10.1002/cnm.1630070303.
Ibrahimbegović, Adnan and Sergiy Melnyk. 2007. “Embedded Discontinuity Finite Element Method
for Modeling of Localized Failure in Heterogeneous Materials with Structured Mesh: An
Alternative to Extended Finite Element Method”. Computational Mechanics 40 (1): 149–55.
doi:10.1007/s00466-006-0091-4.
Ibrahimbegović, Adnan, Chorfi Lotfi Chorf and Fadi Gharzeddine. 2001. “Thermomechanical
Coupling at Finite Elastic Strain: Covariant Formulation and Numerical Implementation”.
Communications in Numerical Methods in Engineering 17 (4): 275–89. doi:10.1002/cnm.405.
Ibrahimbegović, Adnan, Robert L. Taylor and Edward L. Wilson. 1990. “A Robust Quadrilateral
Membrane Finite Element with Drilling Degrees of Freedom”. International Journal for
Numerical Methods in Engineering 30 (3): 445–57. doi:10.1002/nme.1620300305.
Ibrahimbegović, Adnan and François Frey. 1993. “An Efficient Implementation of Stress Resultant
Plasticity in Analysis of Reissner–Mindlin Plates”. International Journal for Numerical
Methods in Engineering 36 (2): 303–20. doi:10.1002/nme.1620360209.
Ibrahimbegović, Adnan and François Frey. 1994. “Stress Resultant Geometrically Nonlinear Shell
Theory with Drilling Rotations—Part II. Computational Aspects”. Computer Methods in
Applied Mechanics and Engineering 118 (3–4): 285–308. doi:10.1016/0045-7825(94)90004-3.
Ibrahimbegović, Adnan, Fadi Gharzeddine and Lotfi Chorfi. 1998. “Classical Plasticity and
Viscoplasticity Models Reformulated: Theoretical Basis and Numerical Implementation”.
International Journal for Numerical Methods in Engineering 42 (8): 1499–535. doi:10.1002/
(sici)1097-0207(19980830)42:8<1499::aid-nme443>3.3.co;2-o.
Kassiotis, C., J.-B. Colliat, A. Ibrahimbegovic and H. G. Matthies. 2009. “Multiscale in Time
and Stability Analysis of Operator Split Solution Procedures Applied to Thermomechanical
Problems”. Edited by Adnan Ibrahimbegović. Engineering Computations 26 (1/2): 205–23.
doi:10.1108/02644400910924870.
Lewis, R. W., K. Morgan, R. H. Thomas and K. N. Seetharamu. 1996. The Finite Element Method
in Heat Transfer Analysis. New York: John Wiley.
Libai, Avinoam and James G. Simmonds. 2005. The Nonlinear Theory of Elastic Shells. 2. Cambridge:
Cambridge University Press.
Naghdi, P. M. 1972. Theory of Shells, Handbuch der Physik. Vol. VIa/2, New York: SpringerVerlag.
Ngo, Van-Minh, Adnan Ibrahimbegović and Delphine Brancherie. 2013. “Model for Localized
Failure with Thermo-Plastic Coupling: Theoretical Formulation and ED-FEM Implementation”.
Computers & Structures 127: 2–18. doi:10.1016/j.compstruc.2012.12.0.
Simmonds, J. G. 2001. “A Simplified, Nonlinear Thermodynamic Theory of Beamshells”. Quarterly
of Applied Mathematics 59 (3): 401–12. doi:10.1090/qam/1848525.
Simo, J. C. and C. Miehe. 1992. “Associative Coupled Thermoplasticity at Finite Strains: Formulation,
Numerical Analysis and Implementation”. Computer Methods in Applied Mechanics and
Engineering 98 (1): 41–104. doi:10.1016/0045-7825(92)90170-o.
Simo, J. C. and J. G. Kennedy. 1992. “On a Stress Resultant Geometrically Exact Shell Model. Part V.
Nonlinear Plasticity: Formulation and Integration Algorithms”. Computer Methods in Applied
Mechanics and Engineering 96 (2): 133–71. doi:10.1016/0045-7825(92)90129-8.
Simo, J. C., D. D. Fox and M. S. Rifai. 1989. “On a Stress Resultant Geometrically Exact Shell
Model. Part II: The Linear Theory; Computational Aspects”. Computer Methods in Applied
Mechanics and Engineering 73 (1): 53–59. doi:10.1016/0045-7825.
Simo, J. C., J. G. Kennedy and S. Govindjee. 1998. “Non-Smooth Multisurface Plasticity and
Viscoplasticity. Loading/unloading Conditions and Numerical Algorithms”. International
Journal for Numerical Methods in Engineering 26 (10): 2161–85. doi:10.1002/nme.1620261003.
Stabler, J. and G. Baker. 2000. “On the Form of Free Energy and Specific Heat in Coupled Thermo-
Elasticity with Isotropic Damage”. International Journal of Solids and Structures 37 (34):
4691–713. doi:10.1016/s0020-7683(99)00292-9.
Chapter 8
Fluid-induced extreme loads
Among the multi-physics problems that are currently entering the mainstream of scien-
tific research in computational mechanics (e.g. see Ibrahimbegović and Brank 2005; Oden
et al. 2003), perhaps the most frequently studied are the problems of fluid–structure interac-
tion, some dealing with problems in medicine, (e.g. Barcelos, Bavestrello and Maute 2006;
Bathe and Zhang 2009; Bazilevs et al. 2008; Bazilevs, Calo and Zhang et al. 2006; Bazilevs
et al. 2009a,b,c; Causin, Gerbeau and Nobile 2005; Degroote, Bathe and Vieren 2009;
Deparis et al. 2006; Dettmer and Perić 2007), among others. The fluid–structure interaction
is already an interesting problem in its own right with a vast number of important applica-
tions. The goal is to conduct the direct coupling of different codes developed for a particular
sub-problem (i.e. either solid or fluid mechanics) into a single code.
A novel partitioned approach for a nonlinear fluid–structure interaction will be presented
here (Kassiotis, Ibrahimbegovic and Matthies 2010; Kassiotis et al. 2011a,b). A combination
of the finite volume method for fluid and finite element method for the structure will be uti-
lized. This partitioned approach has been preferred in the earlier attempts to solve this prob-
lem (Belytschko, Yen and Mullen 1979; Felippa and Park 2004; Hughes and Liu 1978). The
instabilities in the previous procedures are eliminated by the application of implicit schemes
for each sub-problem, but also for the partitioned coupling. The resulting algorithm provides
the computational robustness, which is of great interest not only for fluid–structure interac-
tion but also for any other multi-physics problems of current interest, where one would like
to re-use the available codes for any particular sub-problem in a more general framework.
The main advantage of the code-coupling approach for fluid–structure interaction con-
cerns the fact that the coupling is limited only to the fluid–structure interface. Therefore, the
main difficulty is reduced to enforcing the interface matching in respect of the two different
discretization schemes, the finite element versus the finite volume, as well as the two differ-
ent time-integration schemes and different time steps and their matching at the interface.
The strategy of re-using the stand-alone software for particular sub-problems is likely to
become the most efficient way for the development of software products for multi-physics
applications.
8.1 STRUCTURE AND FLUID FORMULATIONS
8.1.1 Structure Equation of Motion

Structure motion is based on the Lagrangian description. Namely structure domain Ωs is
considered with imposed displacements u on the Dirichlet boundary ∂Ω s,D and moving
under the loading of traction forces t on the Neumann boundary ∂Ω s, N and body force
215
b applied in the whole domain Ωs. This dynamic motion has to be computed in the time
interval [0, T]. The Cauchy governing equation for the structure describes the momentum
conservation. The strong form of this equation can be written with respect to the deformed
configuration as follows; given u on ∂Ω s,D × [0, T ], t and b × [0,T ] in, find: u ∈ Ω s × [0, T ]
u ∈ _s × [0, T] so that:
( )
∇ ⋅ J sF −T + ρs b − ∂t2 u = 0 in Ω s × [0, T ] (8.1)

P
where ρs denotes the material density of the solid domain, u its displacement field and ∂t2 u
the accelerations. The Cauchy stress tensor σ in the deformed configuration can be linked to
the first Piola–Kirchhoff stress tensor P, formulated in the initial configuration through the
gradient F of the deformation and its Jacobian J (e.g. Ibrahimbegović 2009).
To close this partial differential equations system, the displacements (or rather its deriva-
tives) are linked with the stresses through constitutive law. For instance, an elastic material
model based on the Saint-Venant–Kirchhoff constitutive equation is assumed that links the
Cauchy stress tensor σ and the Green-Lagrange strain tensor E through:
F −1 J sF −T = C : E, E =
1 T
2
( )
F F − I and F = I + ∇u (8.2)
where C denotes the constitutive fourth-order elasticity tensor. Thus, due to geometric non-
linearities it is a priori impossible to directly find an exact solution to the problem defined
above. The goal is to find the best approximation of the solution in a finite-dimensional
space where the solution can be found numerically. The FEM approximation (Belytschko,
Liu and Moran 2000; Ibrahimbegović 2009; Zienkiewics and Taylor 2005) derives from
weak forms of the equilibrium in (8.1) and can be written for this problem as given t on
∂Ω s, N × [0, T ] and b in Ω s × [0, T ], finds u ∈ U such that, for allδu ∈ U0 :
Gs ( u; δu) :=
∫ ρ ∂ u ⋅ δu + ∫ s : ∇ δu − ∫ b ⋅ δ u − ∫ t ⋅ δ u = 0
Ωs
2
s t
Ωs Ωs ∂Ω s
where U and U0 are functional spaces for the solution and its variation.
The solid domain Ωs is then discretized in a finite number of sub-domains or elements
I h = (κ e )e =1,...,nel so that the whole space is covered with the finite elements that do not inter-
sect. The solution space associated with this approximation solution is restrained to the
space of continuous element-wise polynomial functions, which is denoted:
{
U h = U ∩ u ∈ C 0 (Ω s ) u κ ∈ P p (κ), ∀κ ∈ I h (8.3)}
where P p (κ) is the space of polynomials of order p on κ. The same restriction U0h holds on
the associated vector space. The semi-discretized FE problem is defined as: given t on ∂Ω s, N
and b in Ωs, find u ∈ U h such that, for all δu ∈ U0h :
Gs ( u; δu ) = 0 (8.4)
This semi-discrete problem can also be written in matrix notation by using the real valued
vectors u ∈ nd-o-f :
Rs ( u s ; λ ) := M s
u s + fsint ( u s ) − fsext (λ) = 0 (8.5)
Fluid-induced extreme loads 217
where:
M s is the mass matrix
fsint is the internal force which is highly nonlinear if large deformation or complex mate-
rial behavior is used and
ext
fs is the external force vector.
Here the λ represents the boundary forces computed from the fluid flow problem and
imposed on the fluid–structure interface. Each matrix and vector of this semi-discrete equa-
tion is properly defined by assembling locally computed array in each element with the
polynomial basis Ne of P (κ e ):
M s, e =
∫ρ N N
κe
s
T
e e
,e ( ue ) =
fsint
∫ ∇N : P (u N ) (8.6)
κe
e e
fsext
,e =
∫N b + ∫ N t
κe
T
e e
∂κ e
T
e e
In order to complete the discretization process, the time-integration of the structural prob-
lem can be carried out by using standard time-stepping schemes (Belytschko 1983; Hughes,
Pister and Taylor 1979). In particular, the generalized HHT-α method (Chung and Hulbert
1994) is used herein. The time interval [0,T] is discretized into a finite number of time steps
tN such as t0 =0 and t Nmax = T . In a typical time step ∆t = t N +1 − t N , the time derivatives of
nodal displacement are approximated with:
 1  
u N +1 = u N + ∆tu N + ∆t 2  − β u N + β
u N +1 
  2  
u N +1 = u N + ∆t (1 − γ ) u N +1 
N + γu
(8.7)
u N + α f = (1 − α f ) u N + α f u N +1
u N + αm = (1 − α m ) u
N + α m
u N +1
In the semi-discrete form of the solid equation of motion in (8.5), the acceleration u and
the displacement u are evaluated at t N +α f and t N +αm . For the elastic linear case, it is shown
(Chung and Hulbert 1994) that there are optimum values for the parameters β, γ , α f , α m for
a given spectral radius ρ∞ ∈ [0, 1]ρ∞ ∈ [0, 1].
(1 + α − αf )
2
m 1
β= ,γ = + αm − α f
4 2
(8.8)
1 2 − ρ∞
αf = , αm =
1 + ρ∞ 1 + ρ∞
The spectral radius controls the numerical damping of the time-integration scheme. The
damping decreases with smaller values of ρ∞ which is maximum for ρ∞= 0. For ρ∞= 1 the
method is the classic trapezoidal rule. Other time-integration schemes can be easily derived
from this general formulation (Hilber, Hughes and Taylor 1977).
8.1.2 Free-Surface Flow
An Arbitrary Lagrangian–Eulerian description of a two-phase flow (taking into account both
water and the surrounding air) is utilized when discretized by finite volume (e.g. Ferziger and
Perić 2002). In this volume of fluid (V.O.F.) method a volume fraction is used to represent
the interface; the main remaining issue is how to convert the interface without diffusing, dis-
persing or wrinkling it. V.O.F. methods use precise convection schemes that reconstruct the
interface from the volume fraction distribution before advection.
For complex flows (with jets, cavitation and aeration in the sloshing wave), it is natural to
consider the Navier– Stokes equations for two immiscible and incompressible flows (water
and air, for instance) occupying transient domains Ωi(t) so that the whole fluid domain
is considered as Ω f (t) = Ω1(t) ∪ Ω2 (t) . The interface between both domains Ω1 and Ω2 is
denoted as Γ. In space–time domain Ω f (t) × [0, T ], the Navier–Stokes equations formulated
in an ALE framework apply. For an arbitrary motion of the total fluid domain Ωf described
by a displacement field um, it can be written as:
ρ∂t u + ρ ( u − u m ) ∇ ⋅ u − ∇ ⋅ 2µD(u) =
= −∇p + fΓ + ρg in Ω f (t) × [0, T ] (8.9)
∇⋅u = 0 in Ω f (t) × [0, T ]
where p denotes the pressure field and υ the velocity. Symbol g is introduced which depicts
the gravity field, with fΓ showing the surface tension forces. It can be expressed as fΓ = σκδ Γ ,
where σ is the surface tension, κ the curvature of the free-surface (i.e. interface between Ω1
and Ω2) and δΓ the mass distribution concentrated at the surface (equivalent to a Dirac dis-
tribution). Fluid material properties are the dynamic viscosity μ and the density ρ. To write a
unique formulation in the whole domain Ωf(t), we express the local material property values
as the function of i:
ρ = ιρ1 + (1 − ι ) ρ2 and µ = ιµ1 + (1 − ι ) µ 2 in Ω f (t) × [0, T ] (8.10)
where the characteristic function or fluid volume fraction ι is defined as:
1, for x ∈ Ω1(t)

ι(x, t) =  (8.11)
0, for x ∈ Ω2 (t)
It is observed that when the whole domain is filled with one fluid ( ι = 1 ∈ Ω f ), the classical
Navier–Stokes equations in the ALE framework appear. The fluid volume fraction ι and the
mass distribution δΓ are linked by the following relation:
∇ι = δ Γ n (8.12)
To close the set of equations the conservation of ι has to be written. When no reaction
between phases occurs, the fluid volume fraction evolves only by advection:
∂t ι + ( u − u m ) ∇ι = 0 (8.13)
The conservation equation system on the whole domain Ωf in d dimensions can be written
as a function of the 2d + 2 unknowns: um , u, p and ι are the solutions for equations (8.9) and
(8.10) are verified. Using a fixed (Eulerian) grid creates difficulties in the in fluid–structure
interaction problems because of moving domains at the fluid interface which follows the
deformations of the structure. Application of Arbitrary Lagrangian–Eulerian (ALE) method
where the whole grid is moved inside the fluid domain leads to the difficulty of pertaining
the quality and the validity of the fluid inner mesh for different new shapes of the boundary.
This is solved by building a suitable map for the domain motion given its interface displace-
ment. The fluid displacement um is arbitrary inside domain Ωf, but on the boundary it has
to fulfill the condition:
um = um on ∂Ω f (t) × [0, T ] (8.14)
Inside the fluid domain Ωf , the fluid displacement is an arbitrary extension of um ∂Ωf :
um = Ext um ∂Ωf ( ) (8.15)

The latter is possible to construct by solving the Laplace smoothing equation:
∇ ⋅ ( γ∇um ) = 0 in Ω f (8.16)
This kind of Laplace smoothing equation is known to have some limitations when the defor-
mation of the fluid domain is governed by large rotations. In this case, it shows sufficient
quality performance when the diffusion coefficient is made spatially dependent upon the
distance to the fluid–structure interface.
The second major difference from solids concerns the most popular discretization tech-
nique for fluid in terms of the finite volume method. The latter will transform the weak form
of the continuous equations in (8.9) to (8.16) into a set of algebraic equations that can be
solved numerically.
The FV formulation can be written directly using an integrated form of the conserva-
tion equation written in (8.9). Another possibility, which is utilized here, is to consider the
restriction of the solution space of weak form problems. The weak form of the Navier–Stokes
equation can be written as defined in (Glowinski 2003): find ( um , u, p and ι ) ∈ U × I × V × P
such that, for all ( δum , δu, δp and δι ) ∈ U0 × I 0 × V 0 × P0 :∈ U × I × V × P :
Gf :=
∫ ∇ ⋅ ( γ∇u m )δum + ∫ (∂t ι + (u − um ) ∇ι) δι × ∫ ρ∂t u ⋅ δu +
Ωf Ωf Ωf
+
∫ ρ∇ (u − u
Ωf
m ) ⊗ u ⋅ δu − ∫ ∇ ⋅ µ f D (u) ⋅ δu + ∫ p∇ ⋅ δu +
Ωf Ωf
(8.17)
+
∫ ∇ ⋅ uδp + [B.C. in a weakk form] = 0
Ωf
where U , I , V , and P are suitable functional spaces for the solution fields.
For this method, the whole of volume Ωf is divided into a set of discrete elements, here
∪
nel
called discrete volumes (κ f ,e )e =1,nel covering the whole domain (Ω f = κ f ,e ) without over-
e =1
∩
nel
lapping κ f ,e = 0 ). For finite element discretization, the solution spaces are restricted to
e =1
suitable spaces of piecewise polynomial functions over the set of discrete elements. For the
finite volume discretization used herein, the test functions were chosen in the space of the
characteristic discrete volume functions. For instance, the velocity can be approximated as:
V h
{
= V ∩ u u κ ∈ span ( ικ ) , ∀κI h
(Ω )} (8.18)
f
where ικ is the characteristic function of the element defined as:
ικ : Ω f →
1 if x ∈ κ (8.19)
x→
0 if x ∈ Ω / κ
The same kind of restriction holds for the solution spaces of (8.17). Therefore, the
functions are piecewise constant by elements, and with the restriction of the weak
formulation it gives: (u h
m , ιh , v h , p h ) ∈ U × I × V × V , such that, for all: (δu , δι , δv , δp
h h h h h
m
h h h
( δu h
m )
, διh , δv h , δp h ∈ U0h × I 0h × V 0 h × V 0 h f h
m : G ((u , ι , v , p ) ; δu , δι , δv , δp ) = 0. The diver-
h h h h
m
h h h
gence terms in (8.17) can be written in terms of flux at the boundary of the volume controls
using the Gauss theorem.
Hence, the weak formulation can be written as:
 
0= ∑ ∫  d Γ ⋅ ( γ∇um ) − [B.C.]
κ  ∂κ 
 
0= ∑∫  ∂t ι + ∫ d Γ ⋅ (v − u m ) ι − [B.C.]
κ  κ ∂κ 

 
0= ∑∫
κ
 ρ∂tv +
 κ
∫
∂κ ∂κ
∫ ∂κ
∫
ρd Γ ⋅ (v − u m ) ⊗ v − d Γ ⋅ µ f D (v ) + pd Γ − [B.C.]

 
0= ∑ ∫ dΓ ⋅ v  − [B.C.]
κ  ∂κ 
where dΓ is the elementary surface vector. It is interesting to see that there is no continu-
ity requirement for the solution (contrary to classical FE), and therefore the approximate
solutions need not be defined at the interface. Their flux can be computed without being
imposed by the restriction of the solution space to the FV space. The difficulty that arises is
to build an accurate representation of the fluxes at the boundaries from a piecewise constant
field. On each control volume, three levels of numerical approximations are applied to build
the boundary fluxes: interpolation to express variable values at the control volume surface
in terms of nodal values (depending on where the variable is stored); differentiation to build
convective and diffusive fluxes the value of the gradient of the quantity of interest—or at
least its approximation—is required; integration to approximate surface and volume inte-
gral using quadrature formulas.
The semi-discrete form of the discretized fluid problem can be written in a matrix form
as follows. The fluid mesh motion considers that um is imposed by the motion of interface u:
Rm ( um , u ) := Km um − Dm u = 0 (8.20)
where Dm is a projection/restriction operator and Km governs the extension of the boundary

displacement (see (8.16)). The volume fraction ιι, the d components of velocity v and pres-
sure p are coupled through a set of nonlinear equations. Written in a matrix form, it gives
the following semi-discrete problem:
 M ι ι + N ι ( v f − u m ) ι 
 
Rf ( ι, v f , p f , um ) := M f ( ι ) v f + N f ( ι, v f − u m ) v + K f ( ι ) v f + B f pf − f f ( ι ) = 0 (8.21)
 BTf v f 
 
where Mι and Nι are the matrices associated with the advection problem of the fluid volume
fraction, Mf is a positive definite mass matrix, Nf is an unsymmetrical advection matrix, Kf
is the conduction matrix describing the diffusion terms and Bf is for the gradient matrix,
whereas f f is the discretized nodal loads on the flow. This matrix form also takes into
account the boundary conditions; special care has to be taken concerning the discretization
of boundary conditions—and especially normal flux—when using the finite volume method
(Ghidaglia and Pascal 2005).
For a given motion of the fluid domain, the coupling between the mesh deformation and
the Navier–Stokes equation is weak in the sense that no variable, such as velocity v or pres-
sure p, influences the fluid domain deformation under imposed boundary displacements.
The coupling between the mesh motion problem and the fluid momentum equation can,
therefore, be ensured explicitly.
The next that needs to be solved is the choice of velocity variation in the time step
Dt = tN +1 − tN . Due to the fact that the mesh motion is arbitrary and does not rely on any
physical phenomenon, it is a priori possible to take any velocity evolution on the window
[TN ,TN +1 ] enabling the initial mesh deformation to be equal to um,N , and the final mesh
deformation is equal to um,N +1 .
The geometric conservation law demands a numerical scheme to reproduce exactly and
independently from the mesh motion a constant solution. This condition can be found in the
literature for ALE formulation discretized either by the finite volume (Demirdžić and Perić
1988) or the stabilized finite element methods (Förster, Wall and Ramm 2006). It is proved
(Farhat, Geuzaine and Grandmont 2001) that the velocity of the dynamic mesh needs to
be computed for all first- and second-order time-integration schemes (like implicit Euler or
Crank–Nicholson):
um,N +1 − um,N
u m = (8.22)
Dt
The volume fraction function is supposed to be sharp at the interface between water and
gas, and, therefore, standard FV discretization that can be strongly diffusive cannot be
applied, as they would smear the interface. One way to guarantee a sharp and bounded solu-
tion is by using a numerical scheme designed for the multidimensional advection equation
(LeVeque 1996). In this case, explicit treatment referred to as MULES (multidimensional
universal limiter with explicit solution) is used (e.g. Behzadi, Issa and Rusche 2004; Ubbink
and Issa 1999).
8.2 FLUID–STRUCTURE INTERACTION PROBLEM COMPUTATIONS
8.2.1 Theoretical Formulation Of Fluid–Structure

Interaction Problem
The interaction problem of motion of the fluid (denoted further with f) and structure
(denoted with s) is elaborated. For the sake of generality, it is supposed that these prob-
lems are nonlinear and time-dependent. This interaction is handled by the classical direct
force-motion transfer (DFMT, e.g. Felippa and Park 2004; Ross et al. 2009) that can
formally be expressed in terms of corresponding Steklov–Poincaré operator (e.g. Deparis

et al. 2006):

Si : H 1/ 2
(Γ ) → H 1/ 2 (Γ ) (8.23)
ui → li
The Steklov–Poincaré operator gives the evolution of dual field li for an imposed primal
field ui on the interface space and time domain Γ × [0,T ]. In fact, this operator requires the
computation of the fluid and structure problem on the complete space–time domain, gov-
erned by Navier–Stokes equations for incompressible flow and nonlinear dynamics equa-
tions for structural motion. For the problems of this kind, the Steklov–Poincaré operator
is not available analytically, but rather as the results of numerical approximations. These
results not only depend on the chosen model, material properties and boundary conditions,
but also on discretization techniques, time-integration algorithms, discrete equations solv-
ers etc.
The interface matching considered in the following is based on two classical mechanics
principles:
Continuity of primal quantities or the perfect matching condition over the interface:
us = u f = u on Γ × [0,T ] (8.24)
where u denotes the value of the primal variable at the interface. In the continuum setting,
the time derivatives of this condition give the equivalent equations for velocity v = u and
acceleration a = u :
u s = u f = u; as = af = a on Γ × [0, T ] (8.25)
However, as the result of time discretization, these conditions are no longer equivalent.
Equilibrium of dual quantities or action–reaction principle, which implies:
l f + l s = 0 on Γ × [0, T ] (8.26)
The action–reaction principle in (8.26) can be reformulated using the Steklov–Poincaré

operator defined in (8.23), resulting with the so-called Steklov–Poincaré formulation (e.g.
Deparis et al. 2006):
Find: u on Γ × [0, T ], so that: Sf (u) + Ss (u) = 0 (8.27)
Using the inverse of the first Steklov–Poincaré operators allows to rewrite the equilibrium
of dual quantities as the following fixed-point equation that concerns only the unknown at
the interface:
Find: u on Γ × [0, T ], so that : u = Ss−1 ( − Sf (u)) (8.28)
The fixed-point equation can be reformulated in order to get a root equation:
Find: u on Γ × [0, T ], so that : Ss−1 ( − Sf (u)) − u = 0 (8.29)
where all the requested quantities are defined at the interface.

It is recalled that these Steklov–Poincaré operators are associated with the semi-dis-
crete form of the continuum equations. Namely, the fluid problem is defined by an ALE
formulation (e.g. Demirdžić and Perić 1988; Hughes, Liu and Zimmermann 1981) of the
Navier–Stokes equations. The latter considers the fluid mesh motion that um is imposed by
the motion of the interface u and can be written as:
Rm ( u m , u ) := K m u m − Dm u = 0 (8.30)
where Dm is a projection/restriction operator and Km governs the extension of the boundary

displacement either by a diffusion process or a pseudo-solid equation (e.g. Ferziger and Perić
2002). The fluid flow in this moving domain is described by the semi-discrete Navier–Stokes
equations:
 M f v f + N f ( v f − u m ) v + K f v f + Bf p f − ff 
Rf ( v f , pf ; u f ) :=   = 0 (8.31)
 BTf v f 
where v f , v f and pf are fluid velocity, its derivative and fluid pressure Mf is a (positive defi-
nite) mass matrix, Nf is a (non-symmetric) advection matrix, Kf is the matrix with diffusion
terms, Bf stands for the gradient matrix, and f f is the driving force on the flow. The discreti-
zation process leading to fluid equations of motion and incompressibility constraint in (8.31)
is carried out by the finite volume method (Ferziger and Perić 2002).
The finite element method (Zienkiewics and Taylor 2005) is used for the structure sub-
problem, resulting with structural semi-discrete equations of motion as set in (8.27)
With this notation on hand, the explicit form of the Steklov–Poincaré operator for fluid
and for structure, which can be expressed in terms of algorithms 1 and 2, is delivered
Tables 8.1 and 8.2.
Table 8.1 Algorithm 1—Steklov–Poincaré Operator for Fluid Sf

Require: Fluid state variable at the time TN
1: Impose displacement of mesh at fluid–structure interface:
um = u on Γ × [TN , TN +1]
2: Solve mesh intern nodes displacement:
Rm (um , u) = 0 on Ωf × [TN , TN +1]
3: Solve fluid problem in ALE formulation:
Rf ( u f , pf ; uf ) = 0 on Ωf × [TN , TN +1]
4: Get boundary traction force at the interface:
λ = −σ f n on Γ × [TN , TN +1]
Table 8.2 Algorithm 2—Poincaré–Steklov Operator for Fluid Ss−1

Require: Solid state variable at the time TN
1: Impose boundary traction force at fluid–structure interface:
σ s n = λ on Γ × [TN , TN +1]
2: Solve structure problem:
Rs (us , λ ) = 0 on Ω s × [TN , TN +1]
3: Get boundary displacement at the interface:
u = us on Γ × [TN , TN +1]
8.3.1 Example 1a — Flexible Appendix in a Flow

An example which is elaborated in this part was first proposed in Wall and Ramm (1988),
and subsequently also studied in Dettmer and Perić (2007), Hübner, Walhorn and Dinkler
(2004), Matthies, Niekamp and Steindorf (2006); and has been used as a benchmark for
fluid–structure interaction.
A fixed square bluff body, with a flexible appendix attached to it, immersed within an
incompressible flow (see Figure 8.1) filling the whole domain is considered. At a sufficiently
long distance from this body, the flow is uniform with an imposed velocity u; the corre-
sponding value of the Reynolds number with respect to the characteristic size of the obstacle
is Re = 330. For this Reynolds number, the flow exhibits a transient behavior with vortices
separating from the corner of the square. These vortices induce an alternative drop and
increase in the pressure field behind the rigid bluff body at a frequency that depends on
the Reynolds number and the shape of the bluff body (Roshko 1952). The vortex shedding
induces oscillations of the flexible appendix.
The thickness of the appendix as well as the material properties is chosen so that its first
eigenfrequency is close to the frequency of the vortex shedding (Table 8.3).
The first eigenfrequency of the problem f = 3.03 1/s obtained considering a linear material
is closed to the natural frequency of vortex shedding behind a square bluff body at Reynolds
Re = 330.
The chosen fluid discretization contains 5,080 FV cells (i.e. around 20 × 103 dof), which
is quite sufficient to get an accurate representation of the flow, and also of the fluid loading
on the structure. The model accuracy is comparable to the one used in Dettmer and Perić
(2007), built with 4,300 finite elements. The PISO algorithm and the Euler implicit scheme
with a time step ∆t = 0.004s are used. The PCG solver is used for pressure correction step
and mesh motion equations and PBiCG for the momentum predictor.
The flexible appendix is discretized with 20 nine-node elements, with quadratic polynomes
description of large structural displacement. Neo-Hookean and Saint-Venant–Kirschoff
Figure 8.1 The benchmark used for fluid–structure interaction problems.

Table 8.3 Material Properties f Steel Bar

FLUID
Density ρf 1.18 × 10−3 kg/m3
νf 1.54 × 10−1 m2/s air 20°C
Imposed velocity at the left-hand side is u (51.3 m/s, 0)
SOLID
Density ρs 0.1 kg/m3
Es 2.50 × 106 N/m2/s
νs 0.35
materials are used. The time discretization is carried out by a generalized HHT-α scheme
with the following parameters: ρ∞ = 1/2; β = 4/9; γ = 5/3; and α = 2/3. At each iteration, the
linear system is solved by the direct solver for asymmetric matrices.
The interface matching computation is carried out by DFMT-BGS solver, that can
easily converge since the channel is open and the flow is not mainly driven by incom-
pressibility. The Aitken relaxation technique is used with an initial value of 0.5, that
rapidly increases to 1. No more than four iterations are required to reach the required
tolerance on the interface displacement residual that is set to 1 × 10 −7. The results for
the pressure and velocity field at several instants are given in Figure 8.2. The deformed
shapes of the appendix also given in Figure 8.2, reveal the oscillations dominated by
the first mode.
The displacement at the free-end of the appendix is plotted in Figure 8.3 for both Saint-
Venant–Kirschoff and Neo- Hookean solid materials. As it can be seen the two results are
very close, since the appendix deformations remain small, despite its large displacement and
rotations.
The long term response (see Figure 8.4) indicates an almost harmonic response domi-
nated by the first eigenfrequency of the structure. A comparison with the results from the
literature in term of the maximum amplitude of motion is also presented in Figure 8.4.
Despite the well-known sensitivity of the computed result with respect to the initial condi-
tion (Hübner, Walhorn and Dinkler 2004), the answers obtained are very close to the previ-
ous results from the literature based upon a FE discretization for both fluid and solid parts
and obtained either by a monolithic (Dettmer and Perić 2007) or a partitioned approach
(Matthies and Steindorf 2003; Wall, Mok and Ramm 1999).
Contrary to the lid-driven cavity with a flexible bottom, the small number of iterations
required to solve the fluid– structure interaction problem of the oscillating appendix sug-
gests that an explicit coupling can also be used for solving this problem. The results from
explicit DFMT algorithm presented in Figure 8.5 for the free-end displacement compare to
a reference solution obtained with an implicit computation.
Using better predictors is supposed to reduce the errors made in term of residual and
energy. In Figure 8.6, the energy error time-history is represented for the zero, first and
second-order predictor. All the results confirm the expected trends, with a decrease in errors
when the predictor order increases.
The final study is then considered with respect to the size of time steps. In Figure 8.7, the
maximum residual error on the time interval t ∈ [0, 15 s] is presented as a function of the
time step size. The error is observed to decrease with a decreasing time step size. However,
when the time steps become too small, the added mass effect triggers the divergence of the
computation. Thus, only the less sensitive schemes with a zero order predictor are able to
solve the coupled problem with the smallest time step.
Figure 8.2 Problems oscillating appendix in flow: velocity and pressure field snapshots.
Figure 8.3 Implicit coupling: displacement of the appendix of the appendix extremity for two nonlinear
materials (Saint-Venant and Neo-Hookean).
Figure 8.4 Implicit coupling: long term response with implicit coupling. Maximum amplitude comparison
with (Dettmer and Perić 2007) (dotted-line), (Matthies and Steindorf 2003) (solid-line), (Wall,
Mok and Ramm 1999) (dashed-line).
Figure 8.5 A comparison between explicit and implicit coupling algorithm-based computations for the dis-
placement of the appendix.
Figure 8.6 Energy error at the interface for different explicit coupling schemes (a) energy error for different
time step size and (b) energy error for different predictors.
Maximum residual error for explicit coupling schemes with different time step sizes and
Figure 8.7
predictors.
8.3.2 Example 1b — Flexible Appendix in a Flow—3d

The material characteristics are chosen as in Table 8.4.
The chosen boundary conditions, specified in Figure 8.8 are as follows: for the lateral
walls of the fluid domain, the velocity boundary condition allows for slipping; at the inflow,
a constant velocity is imposed with u = (100 cm/s, 0, 0); zero gradient pressure is specified
at the outflow.
The fluid problem is discretized with FV method, and then split into six sub-domains by
METIS software tool in order to perform parallel computations (see Figure 8.9).
Fluid only computation from t = −2 s to t = 0 s with the inflow velocity increasing in a smooth
1    t + 1  
way the velocity according to  sin  π    + 1 . In this way, the first steps of the fluid–
2    2  
structure interaction computations are smoothened. The fluid–structure interaction will start
at t = 0 s. All the points of the fluid mesh will move in the ALE strategy, with their motion
governed by a smoothing process based on the Laplacian operator and a diffusivity coefficient
whose value depends on the distance to the appendix. The fluid discretization techniques and
solvers are equivalent to the one used in the above shown 2D example. Three-dimensional ele-
ments with quadratic shape functions are used herein, with each element therefore containing
27 nodes. Two mesh grading are used: the coarse mesh with 663 nodes, and the fine mesh with
2475 nodes. The same integration scheme is used as in the 2D problem.
It is rather difficult to select the most pertinent result for 3D flow problems, and even
more for fluid–structure interaction problems. Hence, in order to obtain a qualitative pic-
ture of computed results, in Figure 8.10 represent the stream-tubes going through the two
lines (x = 6.0, y, z = 3.0) and (x = 6.0, y, z = 7.0) along with the corresponding deformed shape
of the flag.

FLUID
Density ρf 1.18 × 10−3 kg/m3
νf 1.54 × 10−1 m2/s air 20°C
Imposed velocity at the left-hand side is u (51.3 m/s, 0)
SOLID
Density ρs 2.0 kg/m3
Es 2.00 × 106 N/m2/s
νs 0.35
Figure 8.8 Boundary conditions.
Figure 8.9 Flag in the wind: Decomposed fluid domain and zoom on the structure (a) fluid domain decom-
posed and (b) zoom on the deformed mesh.
The displacement of the points at the free-end represented in Figure 8.11 show that the
motion of the flag corresponds to the first flexural mode. It is hard to predict the exact solu-
tion for such a complex three-dimensions flow in with a relatively high Reynolds number.
These results are more in agreement with the ones provided for the 2D version of the same
problem (e.g. Dettmer and Perić 2007; Matthies and Steindorf 2003; Wall, Mok and Ramm
1999) with a flexible appendix.
8.3.3 Example 2 — Three-Dimensional Sloshing

Wave Impacting a Flexible Structure
This example is a simplified representation of a dam-breaking event that brings about the
sloshing wave impact on a flexible structure standing in its way, as presented in Figure 8.12.
At initial time t = 0 s, a three-dimensional water column starts falling down under the gravity
Figure 8.10 Flag in the wind: motion of the structure and stream-tube snapshots for some time steps.
Figure 8.11 Displacement of structure extremity—Oscillating 3D Appendix: extremity displacements for

points A =(10.0, 5.5, 3.0), B=(10.0, 5.5, 5.0) and C=(10.0, 5.5, 7.0).
Figure 8.12 Three-dimensional wave impacting an obstacle: geometry and boundary conditions.
loading. Spreading further at a later time, it hits the obstacle which is a slender plate-like
solid body made of elastic material that can undergo large deformations. The dimension of
the problem and the imposed boundary conditions are given in Figure 8.12.
In order to prevent that the water will bounce-back and again hit the structure after
breaking on the walls, only the left and bottom planes of the fluid domain are defined as
non-slipping walls, while the others are defined with atmospheric boundary condition for
the pressure. Material properties are given in Table 8.5.
The mesh motion problem is solved using a Laplacian smoothing material where the dif-
fusion coefficient is a quadratic inverse function of the distance to the interface between
structure and fluid.
The fluid domain is discretized with finite volume cells always covering the complete
domain either by one phase or by the other. The computations are performed for two differ-
ent meshes with the chosen discretization and the number of cells given in Table 8.6.
An explicit–implicit algorithm is used to compute the two-phase flow evolution from the
corresponding Navier–Stokes equations. In the V.O.F. method used herein (e.g. Ghidaglia,
Kumbaro and Le Coq 2001), an indicator function (volume fraction, level set or phase-field)
is used to represent the phases; leaving only the remaining issue on how to convert the inter-
face without diffusing, dispersing or wrinkling it. This is particularly troublesome when
the volume fraction is chosen as an indicator function because the convection scheme has
to guarantee that the volume fraction stays bounded, with the values that remain within its
physical bounds of 0 and 1. The idea proposed in Behzadi, Issa and Rusche (2004) is used:
FLUID
High-density fluid (water) ρf,1 1.00 × 103 kg/m3
Kinetic viscosity νf,1 1.00 × 105 m/s air 20°C
FLUID-water
Low-density fluid (the air in the remaining part of the domain) ρf,2 1.00 kg/m3
Kinetic viscosity νf,2 1.00 × 106 m/s air 20°C
SOLID
Es 1 × 106 Pa
νs 0
ρs 2500 kg/m3
Table 8.6 Number of Dof for Coarse and Fine Discretization of the
Three-Dimensional Dam-Breaking Problem
Fluid Solid
Number of
Discretization Cells Dof Nodes Dof Time Steps
Coarse 13 × 103 63 × 103 363 1.1 × 103 1 × 105
Fine 104 × 103 520 × 103 2205 6.6 × 103 1 × 105
a V.O.F. method using convection schemes that reconstructs the interface from the volume
fraction distribution before adverting it. The equation associated with the characteristic
function is solved with an explicit time-integration scheme whereas the remaining terms are
solved with implicit time-integration schemes (e.g. Behzadi, Issa and Rusche 2004; Ubbink
and Issa 1999). The fluid is handled by a second-order space discretization with a VanLeer
limiter used for the advection terms, and the implicit Euler time-integration scheme. The
small time steps are required for the explicit solution of the phase function indicator equa-
tion, as well as the half-implicit nature of the coupling between the momentum predictor
and the pressure corrector. At this scale of modeling, it is not required to consider surface
tension between the two fluids. The structure model is constructed by using three-dimen-
sional elements with quadratic shape functions, where each element has 27 nodes. The time-
integration is carried out by a generalized-α scheme as in the previous examples.
Figure 8.13 Number of iterations in order to make the DFMT-BGS algorithm converge for the three-
dimensional dam-breaking problem.
The computation of the coupled problem is carried out by an iterative scheme. The results
of fluid and structure computations are matched for a time step of 1 × 10 −4 for the coarse and
2 × 10 −5 for the fine discretization. The coupling scheme is DFMT-BGS with Aitken’s relax-
ation with the initial parameter ω = 0.25 and the predictor of order 1. The absolute tolerance
for coupled computation is equal to:
rNk ≤ 1 × 10−6 (8.32)
Figure 8.14 3D dam break problem. Evolution of the free-surface and motion of the structure.
Figure 8.15 3D dam break example: obstacle displacement measured at the center of the top face.
The number of iterations required to reach the convergence criteria is given in Figure 8.13.
There is no coupling iteration before the water hits the structure (the effect of air flow can
almost be deemed negligible with respect to the structure). During the water–structure con-
tact, the number of iteration depends on the chosen the discretization density. In reaching
the opposite wall, the water does not rebound on the wall but simply flows away, which
again does not require any iteration.
The water or high-density fluid domain is represented, as well as some part of the fluid
mesh and the structure displacement is shown in Figure 8.14. In the first 0.1 s of the simula-
tion, the water column falls under the gravity loading. There is no effect whatsoever on the
structure until the high-density flow reaches its bottom. The maximum amplitude of the
motion is obtained at t = 0.25 s before the solid comes back to free-vibration phase.
In Figure 8.15, the motion of the free-end of the obstacle is plotted. The motion of the
solid part remains rather well described even with the coarsest grid.
A solution approach to fluid–structure interaction problems that allow coupling of differ-
ent space discretization methods, such as FE for structures and FV for fluids is presented.
The proposed strategy is applicable to demanding problems of this kind that deal with com-
plex free-surface flows interacting with geometrically nonlinear structures. The proposed
strategy can also employ different time-integration schemes, and interface matching condi-
tions between fluid and structure in the spirit of implicit analysis and chosen interface field
representation.
The key idea here pertains to producing the final software tools for complex FSI problems
by coupling the existing software products that were developed (and tested) previously for
either structure or fluid. The code-coupling strategy is capable of dealing with full 3D mod-
els, where the computational efficiency is of paramount interest. The latter is ensured by the
nested parallelization, where the parallelization is carried out not only for coupling of fluid
and structure through interface matching condition but also for flow computations that are
the most expensive task. Both code-coupling and nested parallelization are handled by the
CTL, with the latter as its new feature.
REFERENCES
Barcelos, Manuel, Henri Bavestrello and Kurt Maute. 2006. “A Schur–Newton–Krylov Solver for
Steady-State Aeroelastic Analysis and Design Sensitivity Analysis”. Computer Methods in
Applied Mechanics and Engineering 195 (17–18): 2050–69. doi:10.1016/j.cma.2004.09.013.
Bathe, Klaus-Jürgen and Hou Zhang. 2009. “A Mesh Adaptivity Procedure for CFD and Fluid-
Structure Interactions”. Computers & Structures 87 (11–12): 604–17. doi:10.1016/j.
compstruc.2009.01.017.
Bazilevs, Y., V. M. Calo, T. J. R. Hughes and Y. Zhang. 2008. “Isogeometric Fluid-Structure
Interaction: Theory, Algorithms, and Computations”. Computational Mechanics 43 (1): 3–37.
doi:10.1007/s00466-008-0315-x.
Bazilevs, Y., V. M. Calo, Y. Zhang and T. J. R. Hughes. 2006. “Isogeometric Fluid–structure
Interaction Analysis with Applications to Arterial Blood Flow”. Computational Mechanics 38
(4–5): 310–22. doi:10.1007/s00466-006-0084-3.
Bazilevs, Y., J. R. Gohean, T. J. R. Hughes, R. D. Moser and Y. Zhang. 2009a. “Patient-Specific
Isogeometric Fluid–structure Interaction Analysis of Thoracic Aortic Blood Flow Due to
Implantation of the Jarvik 2000 Left Ventricular Assist Device”. Computer Methods in Applied
Mechanics and Engineering 198 (45–46): 3534–50. doi:10.1016/j.cma.2009.04.015.
Bazilevs, Yuri, M. C. Hsu, D. J. Benson, S. Sankaran and A. L. Marsedn. 2009b. “Computational
Fluid–structure Interaction: Methods and Application to a Total Cavopulmonary Connection”.
Computational Mechanics 45 (1): 77–89. doi:10.1007/s00466-009-0419-y.
Bazilevs, Y., M. -C. Hsu, Y. Zhang, W. Wang, X. Liang, T. Kvamsdal, R. Brekken and J. G. Isaksen.
2009c. “A Fully-Coupled Fluid-Structure Interaction Simulation of Cerebral Aneurysms”.
Computational Mechanics 46 (1): 3–16. doi:10.1007/s004.
Behzadi, A., R. I. Issa and H. Rusche. 2004. “Modelling of Dispersed Bubble and Droplet Flow at High
Phase Fractions”. Chemical Engineering Science 59 (4): 759–70. doi:10.1016/j.ces.2003.11.018.
Belytschko, T. 1983. “An Overview of Semidiscretization and Time Integration Procedures.
Computational Methods for Transient Analysis”. Edited by Hughes T. J. R., Belytschko T.
Journal of Applied Mechanics (North-Holland, Amsterdam), 1–65.
Belytschko, T., Kam W. Liu and B. Moran. 2000. Nonlinear Finite Elements for Continua and
Structures. New York: Wiley.
Belytschko, T., H. J. Yen and R. Mullen. 1979. “Mixed Methods for Time Integration”. Computer Methods
in Applied Mechanics and Engineering 17–18: 259–75. doi:10.1016/0045-7825(79)90022-7.
Causin, P., J. F. Gerbeau and F. Nobile. 2005. “Added-Mass Effect in the Design of Partitioned
Algorithms for Fluid–structure Problems”. Computer Methods in Applied Mechanics and
Engineering 194 (42–44): 4506–27. doi:10.1016/j.cma.2004.12.005.
Chung, Jintai and Gregory M. Hulbert. 1994. “A Family of Single-Step Houbolt Time Integration
Algorithms for Structural Dynamics”. Computer Methods in Applied Mechanics and
Engineering 118 (1–2): 1–11. doi:10.1016/0045-7825(94)90103-1.
Degroote, Joris, Klaus-Jürgen Bathe and Jan Vieren. 2009. “Performance of a New Partitioned
Procedure Versus a Monolithic Procedure in Fluid–structure Interaction”. Computers &
Structures 87 (11–12): 793–801. doi:10.1016/j.compstruc.2008.11.01.
Demirdžić, I. and M. Perić. 1988. “Space Conservation Law in Finite Volume Calculations of Fluid
Flow”. International Journal for Numerical Methods in Fluids 8 (9): 1037–50. doi:10.1002/
fld.1650080906.
Deparis, Simone, Marco Discacciati, Gilles Fourest and Alfio Quarteroni. 2006. “Fluid–structure
Algorithms Based on Steklov–Poincaré Operators”. Computer Methods in Applied Mechanics
and Engineering 195 (41–43): 5797–812. doi:10.1016/j.cma.
Dettmer, W. G. and D. Perić. 2007. “A Fully Implicit Computational Strategy for Strongly Coupled
Fluid–Solid Interaction”. Archives of Computational Methods in Engineering 14 (3): 205–47.
doi:10.1007/s11831-007-9006-6.
Farhat, Charbel, Philippe Geuzaine and Céline Grandmont. 2001. “The Discrete Geometric
Conservation Law and the Nonlinear Stability of ALE Schemes for the Solution of Flow
Problems on Moving Grids”. Journal of Computational Physics 174 (2): 669–94. doi:10.1006/
jcph.2001.6932.
Felippa, C. A. and K. C. Park. 2004. “Synthesis Tools for Structural Dynamics and Partitioned
Analysis of Coupled Systems”. Edited by Ibrahimbegovic A. and Brank B. NATO Advanced
Research Workshop (IOS Press, The Netherlands), 50–111.
Ferziger, J. H. and M. Peric. 2002. Computational Methods for Fluid Dynamics. Berlin:
Springler-Verlag.
Förster, C. H., W. A. Wall and E. Ramm. 2006. “On the Geometric Conservation Law in Transient
Flow Calculations on Deforming Domains”. International Journal for Numerical Methods in
Fluids 50 (12): 1369–79. doi:10.1002/fld.1093.
Ghidaglia, Jean-Michel, Anela Kumbaro and Gérard Le Coq. 2001. “On the Numerical Solution to
Two Fluid Models via a Cell Centered Finite Volume Method”. European Journal of Mechanics -
B/Fluids 20 (6): 841–67. doi:10.1016/s0997-7546(01)01150.
Ghidaglia, Jean-Michel and Frédéric Pascal. 2005. “The Normal Flux Method at the Boundary for
Multidimensional Finite Volume Approximations in CFD”. European Journal of Mechanics -
B/Fluids 24 (1): 1–17. doi:10.1016/j.euromechflu.2004.05.003.
Glowinski, R. 2003. Numerical Methods for Fluids (Part III). Handbook of Numerical Analysis.
Edited by Lions J. L., Ciarlet P. G. Vol. 9. North-Holland: Elsevier.
Hilber, Hans M., Thomas J. R. Hughes and Robert L. Taylor. 1977. “Improved Numerical
Dissipation for Time Integration Algorithms in Structural Dynamics”. Earthquake Engineering
& Structural Dynamics 5 (3): 283–92. doi:10.1002/eqe.4290050306.
Hübner, Björn, Elmar Walhorn and Dieter Dinkler. 2004. “A Monolithic Approach to Fluid–struc-
ture Interaction Using Space–Time Finite Elements”. Computer Methods in Applied Mechanics
and Engineering 193 (23–26): 2087–2104. doi:10.1016/j.cma.2004.
Hughes, Thomas J. R., Wing Kam Liu and Thomas K. Zimmermann. 1981. “Lagrangian-Eulerian
Finite Element Formulation for Incompressible Viscous Flows”. Computer Methods in Applied
Mechanics and Engineering 29 (3): 329–49. doi:10.1016/0045–7825.
Hughes, T. J. R. and W. K. Liu. 1978. “Implicit-Explicit Finite Elements in Transient Analysis:
Stability Theory”. Journal of Applied Mechanics 45 (2): 371. doi:10.1115/1.3424304.
Hughes, Thomas J. R., Karl S. Pister and Robert L. Taylor. 1979. “Implicit-Explicit Finite Elements
in Nonlinear Transient Analysis”. Computer Methods in Applied Mechanics and Engineering
17–18: 159–82. doi:10.1016/0045-7825(79)90086-0.
Ibrahimbegovic, Adnan. 2009. Nonlinear Solid Mechanics: Theoretical Formulations and Finite
Element Solution Methods. New York: Springer.
Ibrahimbegovic, A. and B. Brank. 2005. Engineering Structures Under Extreme Conditions: Multi-
Physics and Multi-Scale Computer Models in Non-Linear Analysis and Optimal Design.
Amsterdam: IOS Press.
Kassiotis C., A. Ibrahimbegovic, H. Matthies. 2010. “Partitioned solution to fluid-structure inter-
action problem in application to free-surface flow”. European Journal of Mechanics, Part B:
Fluids 29: 510–21.
Kassiotis C., A. Ibrahimbegovic, R. Niekamp, H. Matthies. 2011a. “Partitioned solution to nonlinear
fluid-structure interaction problems. Part I: implicit coupling algorithms and stability proof”.
Computational Mechanics 47: 305–23.
Kassiotis C., A. Ibrahimbegovic, R. Niekamp, H. Matthies. 2011b. “Partitioned solution to nonlinear
fluid-structure interaction problems. Part II: CTL based software implementation with nested
parallelization”. Computational Mechanics 47: 335–57.
LeVeque, Randall J. 1996. “High-Resolution Conservative Algorithms for Advection in Incompressible
Flow”. SIAM Journal on Numerical Analysis 33 (2): 627–65. doi:10.1137/0733033.
Matthies, Hermann G., Rainer Niekamp and Jan Steindorf. 2006. “Algorithms for Strong Coupling
Procedures”. Computer Methods in Applied Mechanics and Engineering 195 (17–18): 2028–
49. doi:10.1016/j.cma.2004.11.032.
Matthies, Hermann G. and Jan Steindorf. 2003. “Partitioned Strong Coupling Algorithms for
Fluid–structure Interaction”. Computers & Structures 81 (8–11): 805–12. doi:10.1016/
s0045-7949(02)00409-1.
Oden, J. Tinsley, Ted Belytschko, Ivo Babuska and T. J. R. Hughes. 2003. “Research Directions in
Computational Mechanics”. Computer Methods in Applied Mechanics and Engineering 192
(7–8): 913–22. doi:10.1016/s0045-7825(02)00616-3.
Roshko, A. 1952. Of the development of turbulent wakes from vortex streets. Ph.D. thesis, Pasadena,
CA: California Institute of Technology.
Ross, Michael R., Michael A. Sprague, Carlos A. Felippa and K. C. Park. 2009. “Treatment of Acoustic
Fluid–structure Interaction by Localized Lagrange Multipliers and Comparison to Alternative
Interface-Coupling Methods”. Computer Methods in Applied Mechanics and Engineering 198
(9–12): 986–1005. doi:10.1016/j.cma.2008.11.006.
Ubbink, O. and R. I. Issa. 1999. “A Method for Capturing Sharp Fluid Interfaces on Arbitrary
Meshes”. Journal of Computational Physics 153 (1): 26–50. doi:10.1006/jcph.1999.6276.
Wall, W. A., D. P. Mok and E. Ramm. 1999. “Partitioned Analysis Approach of the Transient Coupled
Response of Viscous Fluids and Flexible Structures”. In: Solids, Structures and Coupled Problems
in Engineering, Proceedings of the European Conference on ComputationalMechanics
ECCM’99, Munich, August/September 1999.
Wall, W. A. and E. Ramm. 1988. Fluid-Structure Interaction Based Upon a Stabilized (ALE) Finite
Element Method. Sonderforschungsbereich 404, Institut für Baustatik und Baudynamik,
German.
Zienkiewics, O. C. and R. L. Taylor. 2005. The Finite Element Method. 6th. New York: Elsevier/
Index
acceleration, 2, 4, 12, 14, 17, 63, 132, 134, 140, constitutive law, 93, 137, 216
148, 156–158, 216, 222 contact problem, 71, 137, 157
additive decomposition, 185, 206 continuity, 32, 222
adiabatic, 187, 188, 190, 191 Coulomb friction, 19, 119
adiabatic condition, 188 coupled plasticity-damage model, 105
adiabatic operator split, 188, 208 crack opening, 71, 83, 100, 115, 184, 188, 210
adiabatic split, 187 critical damping ratio, 98, 99, 100
anisotropic, 108, 109, 110, 111, 113, 151, cyclic loading, 60, 62, 88, 89, 98, 99, 125
152, 212
anisotropic damage model, 151, 152 damage dissipation, 68
average temperature evolution, 175 damage model, 15, 61, 66, 88, 92, 105, 108,
average acceleration (implicit) scheme, 15 109, 110, 126, 127, 151, 152, 212
damping matrix, 8, 9, 10, 11, 136, 138, 139,
backward Euler scheme, 15, 168, 171 140, 142
best approximation, 216 damping ratio, 9, 10, 98, 99, 100, 102, 139, 140
bond-slip, 65, 66, 71–88, 152, 163 damping model, 8, 19, 20, 105, 109, 110
boundary conditions, 52, 122, 131, 144, 166, deductive modeling, 118
170, 175, 177, 179, 221, 222, 228, deformation discontinuity, 179
229, 231 deformation gradient, 3
boundary displacement, 220, 221, 223 deformed configuration, 1, 2, 3, 4, 53, 55, 78,
boundary forces, 217 140, 216
brick, 111, 113, 114, 115, 117, 118, 122, determinant, 3
123–127, 174, 175, 177, 178, 179 deviatoric part, 119
differentiation, 220
Cauchy, 3, 26, 67, 216 dilatancy angle, 118, 119, 122, 123
Cauchy principle, 32 Dirac-delta function, 30, 67, 115, 183
central difference (explicit) scheme, 12 d’Alembert principle, 3, 4
cohesive interface model, 118, 126 diagonal tension cracking, 113
compressive stress, 15, 143, 167, 209 Dirichlet boundary, 215
concrete cracking, 15, 65, 66, 69, 75, 78, 79, 87, displacement discontinuity, 29, 32, 33, 36, 37,
88, 108, 163 66, 67, 70, 76, 116, 180, 184, 185,
concrete cracking model, 66 188, 189
concrete displacement, 66, 76, 77, 79, 80 displacement gradient, 26, 75
concrete failure model, 66 displacement vector, 13, 139
concrete hardening behavior, 15 dissipation, 19, 26, 27, 30, 33, 34, 37, 38, 46,
concrete finite element model, 65 47, 48, 55, 68, 89, 92, 93, 95, 103,
concrete model, 66, 67, 71, 75, 88, 89, 98, 100, 104, 105, 106, 112, 115, 131, 137, 145,
101, 102, 152 170, 181, 183, 185, 186, 196, 197, 198,
concrete–steel interface, 71 199, 204, 205, 207, 236
configuration, 1, 2, 3, 4, 53, 55, 57, 78, 140, Drucker mode, 119
171, 216 Drucker–Prager criterion, 119
consistency condition, 28, 31, 47, 48, 116, 119, Drucker–Prager plasticity, 65
120, 168, 191 dynamic frictional contact, 109, 140, 144, 163
239
240 Index
earthquake, 10, 15, 19, 20, 21, 104, 105, 108, hardening plasticity 26
109, 120, 121, 126, 127, 131, 132, hardening regime, 25, 31, 36, 48, 49
133, 134, 137, 139, 140, 143, 144, 236 hardening variable, 15, 26, 27, 34, 35, 45, 46,
eigenfrequency, 224, 225 49, 50, 116, 149, 171, 180, 181, 207
elastic-vibration, 99 heat capacity, 166, 178, 180, 186
elastoplastic-damage model, 88 heat capacity coefficient, 166
energy conservation, 18, 19, 163 Hamilton’s principle, 3, 5
energy error, 225, 227 Hamilton variational principle, 5
entropy, 170, 180, 192, 207 Heaviside function, 76, 77, 115, 184, 185
equation of motion, 1, 7, 8, 11, 215, 217 Helmholtz free energy, 91, 92, 93, 96, 170
essential ingredients, 25, 26 homogeneous models, 113
Eulerian, 218 homogeneity, 19
evolution equations, 12, 17, 27, 47, 69, 149, 168, hydrostatic, 119
171, 218 hydrostatic water pressure, 140
explosion, 14, 151, 163
impact, 14, 15, 16, 20, 108, 110, 112, 137,
finite element, 1, 11, 14, 21, 34, 36, 38, 39, 40, 146–151, 154, 157, 162, 163, 229
43, 44, 48, 51, 55, 57, 62, 63, 65, 66, impact loads, 20, 145, 152
67, 69, 71, 76, 77, 78, 81, 82, 85, 87, impact phenomena, 145
88, 96, 98, 106, 108, 109, 110, 112, incompressible flows, 218
115, 125, 126, 127, 131, 138, 140, 141, interface concrete-steel, 71, 77, 112, 113
142, 144, 147, 150, 154, 163–167, 167, interface discontinuity, 66, 115
173, 174, 177, 178, 180, 184, 187, 193, interface fluid-structure, 215, 217, 219, 223
207, 208, 212, 213, 215, 216, 219, interface mortar-brick, 111, 112, 118
221, 223–237 interface structure-foundation, 132, 135, 136
fire-induced damage conditions, 209 integration, 4, 6, 9, 11, 12, 14, 16, 19, 35, 51,
fire resistance, 165, 174, 208 83, 84, 99, 140, 149, 151, 154, 157,
fire test, 208 168, 171, 192, 194, 213, 215, 217, 220,
first Piola–Kirchhoff stress tensor, 216 221, 222, 228, 232, 234, 235, 236
first principle, 182 interpolation, 36, 43, 71, 76, 77, 109, 117, 138,
forced-vibration problems, 6 156, 163, 167, 184, 189, 190, 192, 193,
free energy potential, 182 208, 220
free undamped system of vibration, 6 incremental analysis, 12
free-vibration, 6, 7, 8, 12, 18, 89, 99, 105, 107 incremental displacement, 18, 59
free-vibration phase, 234 indicator function, 218, 231
free-vibration problem, 8, 18 infinitesimal deformation, 26
frictional contact, 109, 137, 140, 144, 145, 147, infinitesimal volume, 3
148, 154, 157, 163 initial condition, 140, 166, 225
frictional mechanisms, 19 initial configuration, 1, 2, 3, 216
frictional sliding, 147 initial mesh deformation, 221
frictional support systems, 9 in-plane failure modes, 111
functional spaces, 216, 219 internal energy, 18, 181, 182
fundamental vibration modes, 160 internal variables, 14, 15, 25, 27, 47, 62, 67, 68,
69, 79, 80, 81, 89, 90, 92, 93, 94, 95,
Gauss numerical integration points, 83, 84 106, 149, 168, 171, 172, 185, 186, 191
Gauss quadrature point, 66 isoparametric, 72, 76, 138, 166, 190
Gauss theorem, 220 isothermal, 188
generalized standard materials, 88, 108 isothermal operator, 188
geometrical nonlinearity, 40 isotropic damage, 90, 213
Green-Lagrange strain tensor, 216
Jacobian, 216
hardening, 28, 29, 37, 39, 45, 46, 60, 61, 74, 92, joints, 111, 112, 113, 114, 117, 118, 119, 122,
105, 107, 108, 127, 149, 163, 168, 170, 123, 125, 127, 144, 177
180, 212 jump functions, 115
hardening behavior, 15, 36
hardening dissipation, 55 Kelvin–Voigt viscoelastic model, 88, 99
hardening moduli, 32, 53, 67, 106, 116, 196, kinematic equation, 185
198, 200, 201 kinematic hardening, 107
hardening phase, 35, 101, 104 kinematics hypothesis, 165
Index 241
kinetic energy, 5, 6, 96 nonproportional damping, 9, 10, 11, 21, 139,

kinetic viscosity, 232 140, 142
Kirchhoff hypothesis, 165 nonproportional loading, 145
Kuhn–Tucker optimality conditions, 27, 31, 34 numerical integration, 14, 35, 83, 84, 154,
157, 171
Lagrangian, 18, 30, 93, 96, 97, 158, 215
Lagrangian–Eulerian description, 218, 219, 236 operator split method, 79, 187, 188
Legendre transformation, 96 orthogonality condition, 8, 30, 31, 33, 34,
linear dynamics, 1, 2, 4, 6, 8, 10, 11, 18 186, 207
linear viscoelastic constitutive equations, 135 out-of-plane failure, 120, 121, 122, 124
linear viscous damping model, 20
loading/unloading conditions, 28, 35, 47, 48, patch test, 67, 117
94, 119, 128, 213 perfect plasticity, 25, 28, 65, 83, 87, 119, 122
localization condition, 25, 32, 33 phenomenological modeling, 118
localization failure (criterion), 45, 47 PISO algorithm, 224
localization problems, 31 plastic deformations, 25, 28, 90, 147
locking, 36, 43 plastic flow, 25, 116, 119, 120
logarithmic decrement, 100 plastic reinforcement sliding, 83
plastic sliding, 83
macro-cracks, 65, 66, 67, 70, 83, 84, 85, 86, 116 plasticity, 15, 16, 21, 25, 28, 29, 31, 34, 37, 39,
mass conservation principle, 2, 3 40, 45, 46, 48, 49, 51, 60, 61, 63, 65,
mass density, 2, 3, 141, 180, 182, 196 72, 74, 80, 83, 87, 91, 94, 101, 104,
mass matrix, 7, 14, 18, 133, 136, 139, 148, 217, 105, 106, 108, 109, 110, 113, 115, 117,
221, 223 118, 119, 122, 126, 127, 128, 145, 148,
material damping, 19, 139, 140 151, 152, 163, 164, 167, 168, 196, 207,
material derivative, 1 209, 212, 213
material description, 1–5 plastic yielding, 65, 80
material hardening, 48, 57 porous metals, 151
masonry, 19, 112, 114, 115, 117–125, 128 post-localization phase, 30
masonry units, 111, 113 position vector, 1
masonry walls, 111, 112, 126, 127 principle of the maximum plastic dissipation, 26
maximum principal stress, 33 principle of virtual work, 3, 4, 44
method of incompatible modes, 109, 126, proportional damping, 8, 9, 11
208, 212 pseudo-time, 11, 78
mid-point rule, 19
mixed mode, 120 quasi-static problems, 11
modal nonproportional damping matrix, 139 quasi-brittle materials, 84
mode of vibration, 6 quasi-static cyclic loading, 89, 98
modeling of the interfaces, 117 quasi-static sine loading, 100
moment-curvature/rotation, 50
mortar, 111, 112, 113, 114, 117, 118, 119, 122, Rankine-like yield criterion, 167
126, 127 Rayleigh damping, 8, 10, 20, 21, 139, 140
mortar dilatancy, 113 realistic fire scenarios, 208
multisurface, 118, 119, 127, 128, 167, 213 real-time, 171
reinforced concrete, 16, 65, 76, 81, 106, 108,
natural frequencies, 7 109, 110, 111, 123, 127, 145, 148, 149,
Neo-Hookean, 224, 225, 226 152, 153, 163, 210, 211
Neumann boundary, 215 reinforcement bar 66, 75, 76
Newton, 1 relaxation technique, 225
Newton’s method, 17 residual, 70, 78, 89, 104, 123, 138, 194,
Newton’s iterative method, 80 225, 228
Newton’s method, 169, 190 Ritz coordinates, 139, 140
Newton’s Second Law of Motion, 1 Ritz vectors, 138, 139, 142, 144
Newton’s Third Law, 32
non-associative perfect plasticity, 65 second law of thermodynamics, 26, 181, 207
nonlinear behavior of masonry, 111, 112 seismic actions, 35
nonlinear dynamics, 1, 11, 12, 14, 16, 17, 18, Saint-Venant–Kirschoff, 224, 225
19, 21, 163, 222 Saint-Venant–Kirchhoff constitutive equation, 216
nonlinear plasticity, 91, 213 semi-discretization, 11
242 Index
small-amplitude vibrations, 104 thermal strain field, 175

softening, 25, 29, 30, 37, 38, 39, 40, 45, 47, thermal stress, 180, 186, 207
49, 50, 51, 53, 59, 62, 66, 68, 69, 81, thermodynamics, 26, 27, 46, 68, 72, 89, 181,
88, 100, 101, 104, 105, 108, 110, 111, 182, 183, 187, 207, 212
114, 118, 127, 128, 163, 168, 183, 186, thermodynamic principles, 25
207, 212 thermoplasticity model, 209
softening behavior, 19, 89, 114, 115, 118, total damage dissipation for concrete, 68
179, 183 trapezoidal rule, 17, 18, 19, 158, 217
softening dissipation, 55 true stress or Cauchy stress, 3, 26, 67, 216
softening failure, 204 truncation criteria, 139
softening law, 31, 38, 68, 98 truss-bar 150, 184, 185, 187, 188, 192
softening modulus, 31, 47, 51, 53, 57, 58, 67,
184, 196, 198, 200, 201 undamped, 6, 8
softening phase, 29, 32, 104
softening plastic hinge, 38, 39, 45, 47, 48, 55, validation, 81, 82, 111, 127
56, 57, 60, 62 VanLeer limiter, 232
softening regime, 25, 31, 48, 49, 50, 53, 123 Variation, 5, 6, 9, 10, 11, 36, 76, 96, 136, 153,
softening stress-separation, 100 166, 170, 174, 190, 195, 196, 199, 201,
softening variable, 47, 50, 207 209, 210, 216, 221
spatial description, 2, 3 variational formulation, 5, 96
standard materials, 88, 108 velocity, 1, 2, 8, 13, 15, 17, 132, 150, 151, 156,
state variable, 46, 68, 147, 181, 194, 212, 223 157, 158, 218–226, 228
static condensation, 36, 70, 131, 140, 169 vibration-isolation devices, 9
steel bar, 75–79, 81, 103, 152, 195–198, 200, vibration modes, 7, 12, 160
201, 204, 225 vibration motion, 6, 19
steel bar stiffness matrix, 75 virtual axial deformations, 39
steel displacement, 66, 75, 77 virtual axial strains, 40
steel reinforcement anchorage, 75 virtual discontinuity, 43
steel yielding, 75, 79 virtual jump, 43
stiffness matrix, 7, 75, 136, 140 virtual strain, 43, 44, 67, 117
strain energy, 5, 26, 30, 33, 47, 66, 163 virtual nodal transverse displacements, 43
strain energy density 67 viscoelastic equations, 134
strain hardening, 25, 29, 89, 90 viscoelastic model, 88, 99, 100, 137
strain localization, 66 viscoelastic-plastic-damage model, 92
strain softening, 49, 50, 51, 89, 90 viscoelastic response, 99
strong form, 192, 206, 216 viscoplasticity model, 15, 63, 213
viscous damping, 8, 11, 20
tangent modulus, 35, 70, 75, 169, 191, 201 viscosity, 88, 89, 104, 218, 232
tangent stiffness matrix, 12, 43, 72, 80, 138 viscous parameter, 89, 95, 98, 99, 100, 102
temperature gradient, 166, 167, 171, 175, 179, von Karman axial strain, 43
180, 187, 192, 196, 199, 202, 209, von Mises yield criterion, 32, 36
210, 211
thermal balance equation, 170 weak form, 3, 4, 5, 34, 36, 44, 78, 97, 139,
thermal conductivity coefficient, 166 166, 167, 170, 188, 190, 192, 193,
thermal deformation, 181 206, 216, 219, 220
thermal diffusivity, 172 weighting functions, 156, 167
thermal dissipation, 183
thermal energy, 180 yield condition, 34
thermal expansion coefficient, 170, 178, 180 yield criterion, 25, 26, 28, 30, 31, 32, 36, 45,
thermal loading, 177, 200, 211 167, 169, 181, 184, 207, 209
thermal part, 166, 180, 181, 188, 192 Young’s modulus, 37, 68, 71, 101, 102, 103,
thermal potential, 170 116, 143, 178, 180, 188, 196, 198,
thermal properties, 177, 178, 200 200, 201

Webpdf

Uploaded by

Copyright:

Available Formats

Webpdf

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

Webpdf

Uploaded by

Copyright:

Available Formats

Nonlinear Dynamics of

Structures Under Extreme

© 2019 by Taylor & Francis Group, LLC

No claim to original U.S. Government works

Printed on acid-free paper

International Standard Book Number-13: 978-1-138-03541-6 (Hardback)

Library of Congress Cataloging‑in‑Publication Data

Names: Ibrahimbegovic, Adnan, author. | Ademovic, Naida, author.

Visit the Taylor & Francis Web site at

and the CRC Press Web site at

1 Initial boundary value problem 1

2.4.1.3 Example 3 — The Darvall–Mendis Frame 57

3 Reinforced concrete structures 65

4 Masonry structures 111

5 Dynamics extreme loads in earthquake engineering 131

6 The dynamics of extreme impact loads in an airplane crash 145

7 Fire-induced extreme loads 165

7.7.3 ED-FEM Implementation for the Mechanical Part 188

8 Fluid-induced extreme loads 215

Naida Ademović is an Associate Professor in Civil (Structural) Engineering at the Faculty

Initial boundary value problem

1.1 NEWTON’S SECOND LAW—STRONG

x ϕ = x + d(x, t) =: ϕ(x, t) ≡ ϕt (x) (1.2)

x ϕ ∈ Ωtϕ = 0, ltϕ  (1.3)

Figure 1.1 Mapping from the initial to current configuration.

ρ(x)a(x, t) = div P(x, t) + b(x, t); ∀x ∈ Ω & ∀t ∈ [0, T ] (1.14)

1.2 D’ALEMBERT PRINCIPLE—WEAK FORM

0 = Gdyn ( ϕt (x); w(x)) :=

0 = Gdyn ( ϕt (x); w(x)) :=

Figure 1.2 Application of the D’Alembert principle.

1.3 HAMILTON’S PRINCIPLE—ENERGY CONSERVATION

H ( ϕ(x, t), υ(x, t)) := T ( υ(x, t)) + Π ( ϕ(x, t)) (1.21)

1.4 MODAL SUPERPOSITION METHOD

Generally, superposition is applicable when there is a linear relationship between external

v(t) = φ sin ( ωt + θ ) (1.28)

where ϕ is the shape of the system (amplitude) and θ is a phase angle.

−ω 2 Mφ sin ( ωt + θ ) + Kφ sin ( ωt + θ ) = 0 (1.29)

Combining equations (1.31) and (1.32), one obtains:

ω 2mφTm Mφn = ω n2φTn Mφm (1.33)

φTm Mφn = 0 ω m ≠ ω n (1.35)

The second orthogonality condition states that

φTm Kφn = 0 ω m ≠ ω n (1.36)

∑ φ  φ Md cos ω t + ( 1ω ) φ Mv sin ω t (1.37)

Cn = φTn Cφn = αφTn Mφn (1.41a)

Cn = φTn Cφn = βφTn Kφn (1.41b)

Combining (1.41a) and (1.41b) with (1.42):

1.4.2 Nonproportional Damping and Fast Computation Methods

yi (ti ) = A0 + A1∆t + A2 exp(−ξi ω i ∆t)cos ω Di ∆t + A3 exp(−ξi ω i ∆t)sin ω Di ∆t (1.45)

where the coefficients in equation (1.45) above are:

i + 2ξi ω i y i + ω i2 yi = fi − Cˆ ij y j , i = 1, 2,...m. (1.51)

i(k) + 2ξi ω i y i(k) + ω i2 yi(k) = fi −

1.5.1 Central Problem of Dynamics (Time Evolution)

1.5.2 Central Difference (Explicit) Scheme

(dn +1 − dn −1 ) = vn +1/2  1

Table 1.1 Central Difference Scheme in Nonlinear Dynamics

Man +1 + fˆ int (d n +1) = f next

1.5.3 Trapezoidal Rule or Average Acceleration (Implicit) Scheme

Table 1.2 Trapezoidal Rule or Average Acceleration

dn(i++11) = dn(i+) 1 + un(i+) 1 (1.69)

1.1 NEWTON’S SECOND LAW—STRONG

Figure 1.1 Mapping from the initial to current configuration.

1.2 D’ALEMBERT PRINCIPLE—WEAK FORM

Figure 1.2 Application of the D’Alembert principle.

1.3 HAMILTON’S PRINCIPLE—ENERGY CONSERVATION

1.4 MODAL SUPERPOSITION METHOD

1.4.2 Nonproportional Damping and Fast Computation Methods

i + 2ξi ω i y i + ω i2 yi = fi − Cˆ ij y j , i = 1, 2,...m. (1.51)

i(k) + 2ξi ω i y i(k) + ω i2 yi(k) = fi −

1.5.1 Central Problem of Dynamics (Time Evolution)

1.5.2 Central Difference (Explicit) Scheme

Table 1.1 Central Difference Scheme in Nonlinear Dynamics

1.5.3 Trapezoidal Rule or Average Acceleration (Implicit) Scheme

Table 1.2 Trapezoidal Rule or Average Acceleration

1.6 MOTIVATION FOR DEVELOPMENTS

1.6.1 Damping Model Characterization

1.6.2 Extreme Loading Representation

2.1 ESSENTIAL INGREDIENTS OF PLASTICITY MODEL

2.1.1 1D Plasticity with Hardening and Softening

Figure 2.1 Simple tension test.

Figure 2.2 (a) Linear isotropic hardening (b) Saturated hardening.

∫ σε − dt ψ (ε, ξ, ξ )dx

2.2 LOCALIZED FAILURE PLASTICITY MODEL

σ = Cep ( m ⊗ nη ) (2.33)

∫ σ ⋅ ε − dt ψ (ε, ξ, ξ )dΩ

2.3 STRUCTURAL PLASTICITY MODEL

2.3.1.1 Numerical Examples of Small Experiments

Figure 2.3 Pure shear test.

Table 2.1 Material Properties for Pure Shear Test

2.3.2 Multi-Scale Model Parameters Identification

Figure 2.4 Load/displacement results for a pure shear test.

Figure 2.5 Traction test.

Figure 2.6 Load/displacement results for a simple shear test.

Figure 2.7 Accumulated plastic deformation.

Figure 2.8 Localized elements and discontinuity lines.

2.3.2.1 Beam Element With Embedded Discontinuity

Figure 2.9 Beam finite element with embedded discontinuity.

Figure 2.10 Strain-free mode of the element.

δεVK = Bu (x)δu + Bu,w (x)δw + Bu,w ′ (x)δw ′ + Gu ( x, xd ) δα u + δ

δε = δε VK   Bu Bu, w Bu, w ′ 

Figure 2.11 Rigid-plastic cohesive law at discontinuity.

∫ (Mκ +qξ ) dx + ∫

2.4 COMPUTATION OF BEAM PLASTICITY

2.4.1 Numerical Examples of Frames and Frame Elements

2.4.1.1 Example 1 — Computation of Beam Plasticity Material Parameters

Table 2.2 Material Properties

Figure 2.13 Uniaxial stress–strain curve.

Figure 2.14 Boundary conditions for the shell model analysis.

uref (N) = Mu

uref (N) can be used to evaluate the ultimate

uref (N) (2.123)

Figure 2.17 Plastic work vs. end cross-section rotation curves.

Table 2.3 Summary of Results of the Shell

Figure 2.18 Approximation of the ultimate bending moment of the cross-section.

Table 2.4 Comparison Between Approximations and Shell Analyses Results

2.4.1.2 Example 2 — Push-Over of an Asymmetric Frame

Figure 2.19 Asymmetric frame: geometry and loading.

Figure 2.21 Deformed shape and locations of softening plastic hinges.

2.4.1.3 Example 3 — The Darvall–Mendis Frame

Figure 2.22 Geometry and loading of the portal frame.

Table 2.5 Geometric and Material

Figure 2.23 Vertical load vs. vertical deflection curves.

2.4.1.4 Example 4 — Push-Over Analysis of a Symmetric Frame

Table 2.7 Geometric and Material

Figure 2.24 Geometrical characteristics of the frame and loading scheme.