Ceg Simone 20031013

Continuous-Discontinuous
Modelling of Failure
Continuous-Discontinuous
Modelling of Failure
Proefschrift
ter verkrijging van de graad van doctor
aan de Technische Universiteit Delft,
op gezag van de Rector Magnicus prof. dr. ir. J. T. Fokkema,
voorzitter van het College van Promoties,
in het openbaar te verdedigen op maandag 13 oktober 2003 om 10.30 uur
door Angelo SIMONE
ingegnere civile laureato al Politecnico di Milano
geboren te Taranto, Itali e
Dit proefschrift is goedgekeurd door de promotor:
Prof. dr. ir. J. Blaauwendraad
Toegevoegd promotor:
Dr. ir. L. J. Sluys
Samenstelling promotiecommissie:
Rector Magnicus Voorzitter
Prof. dr. ir. J. Blaauwendraad Technische Universiteit Delft, promotor
Dr. ir. L. J. Sluys Technische Universiteit Delft, toegevoegd promotor
Prof. dr. ir. M. G. D. Geers Technische Universiteit Eindhoven
Prof. dr. ir. E. van der Giessen Rijksuniversiteit Groningen
Dr. M. Jir asek

Ecole Polytechnique F ed erale de Lausanne, Zwitserland
Prof. dr. R. Larsson Chalmers Tekniska H ogskola, Zweden
Prof. dr. ir. F. Molenkamp Technische Universiteit Delft, reservelid
Prof. dr. ir. J. G. Rots Technische Universiteit Delft
Published and distributed by DUP Science
DUP Science is an imprint of
Delft University Press
P.O. Box 98
2600 MG Delft
the Netherlands
Telephone: +31 15 27 85 678
Telefax: + 31 15 27 85 706
E-mail: info@library.tudelft.nl
ISBN 90-407-2434-2
Keywords: continuous-discontinuous failure, nite-element method, damage
Copyright c 2003 by A. Simone
All rights reserved. No part of the material protected by this copyright notice may be repro-
duced or utilised in any form or by any means, electronic or mechanical, including photocopy-
ing, recording or by any information storage and retrieval system, without written permission
from the publisher: Delft University Press
This document was set in Palatino and Helvetica using L
A
T
E
X 2
together with the KOMA-

Script bundle and the indexing package AUTHIDX
Printed in the Netherlands
Contents
Preface vii
List of symbols and abbreviations ix
1 From continuous to continuous-discontinuous failure representation 1
1.1 The need for discontinuous failure descriptions . . . . . . . . . . . . . . . . . . 1
1.2 Failure characterisation and numerical strategies . . . . . . . . . . . . . . . . . 4
1.3 Requirements for distributed failure models . . . . . . . . . . . . . . . . . . . . 6
1.4 Models for inelastic bulk deformation . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 A model for separation across cohesive surfaces . . . . . . . . . . . . . . . . . 12
2 Continuous-discontinuous failure in standard media 13
2.1 Problem elds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Governing equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Variational formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4 Discretisation and linearisation . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.5 Element technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.6 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.7 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3 Continuous-discontinuous failure in gradient-enhanced media 41
3.1 Problem elds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.2 Governing equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.3 Variational formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.4 Discretisation and linearisation . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.6 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.7 Stress-strain relationships . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.8 Damage initiation and discontinuities . . . . . . . . . . . . . . . . . . . . . . . 65
4 Continuous-discontinuous failure in rate-dependent media 71
4.1 Rate-dependent elastoplastic-damage models . . . . . . . . . . . . . . . . . . . 72
4.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.4 Traction-free discontinuities in rate-dependent and non-local media . . . . . . 96
5 Conclusions 101
vi Contents
A Conventional interface and PU-based discontinuous elements 105
A.1 Conventional continuous interface element . . . . . . . . . . . . . . . . . . . . 105
A.2 Partition of unity-based discontinuous elements . . . . . . . . . . . . . . . . . 108
A.3 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
B Some essentials of generalised functions 115
C Constitutive models for softening materials 117
C.1 Considerations on numerical modelling of concrete . . . . . . . . . . . . . . . 119
C.2 Non-local versus viscous regularisation . . . . . . . . . . . . . . . . . . . . . . 121
D Interpolation requirements for a class of gradient-enhanced media 125
D.1 Governing equations and spatial discretisation . . . . . . . . . . . . . . . . . . 125
D.2 Element performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
D.3 About terminology: mixed method versus coupled problem . . . . . . . . . . 130
D.4 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
E Incorrect failure characterisation in non-local media 135
E.1 Some basic notions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
E.2 Damage characterisation in mode I problems . . . . . . . . . . . . . . . . . . . 137
E.3 Damage characterisation in shear band problems . . . . . . . . . . . . . . . . . 145
E.4 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
References 153
Author/editor index 167
Subject index 171
Summary 173
Samenvatting 175
Curriculum vitae 177
Preface
Failure of most engineering materials is a phenomenon which is charac-
terised by the development of a process zone in which microcracks arise,
deformations localise and the accumulation of damage eventually leads to
traction-free macrocracks with eventual total loss of load-carrying capacity
at the nal stage. This thesis explores applications of a method combining
novel techniques of continuous and discontinuous numerical failure anal-
ysis. Continuous and discontinuous approaches to failure have been ex-
tensively studied in the past and have now reached a reasonable degree
of maturity. Various strategies are known to cope with the problems origi-
nating from the use of continuous softening stress-strain relationships, and
techniques are available which allow displacement discontinuities to cross
through solid elements when the discontinuous stress-crack opening ap-
proach is considered. It is desirable that a numerical approach to failure
description is performed in a unied fashion, in all its stages.
The basic idea behind this study is the combination of continuous and
discontinuous failure descriptions to achieve a better characterisation of the
whole failure process. Aim of this thesis is to illustrate that a discontinuous
interpolation of the problem elds
- enables a more realistic description of failure,
- adds exibility to continuous modelling,
- solves some problems inherent to some regularised models, and
- allows a different interpretation of model parameters.
Outline. The motivations behind a continuous-discontinuous approach to
failure are put forward in Chapter 1. The chapter is completed by some basic
notions regarding models for degradation in the bulk volume and across the
discontinuity surface used in the remainder of this thesis.
The basic tool of the continuous-discontinuous approach to failure, dis-
placement discontinuities based on the partition of unity property of nite-
element shape functions, is described in Chapter 2. A similarity is drawn
between conventional interface elements and partition of unity-based dis-
continuous elements; some applications to elastic and strain-hardening me-
dia conclude the chapter.
viii Preface
When strain-softening media are considered, regularisation techniques
must be employed to avoid a dependence of the solution on the discretisa-
tion. The applicability of discontinuities in regularised media is discussed in
Chapters 3 and 4 where discontinuities are considered in gradient-enhanced
and rate-dependent media respectively, and where it is shown that the un-
derlying regularised continuum determines the quality of the discontinuous
enhancement.
Notation. In most of this thesis tensor notation is used. Details on tensorial
algebra can be found in References [77, 84]. Matrix notation (referred to as
engineering notation in the remainder of the thesis) is considered in the lin-
earisation and discretisation of the governing equations in Chapters 2 and
3 and in Appendix A. First-order tensors (representing vectors) are denoted
by bold lower-case Latin letters, second-order tensors (vectors in engineer-
ing notation) by bold lower-case Latin or Greek letters, and fourth-order
tensors (matrices in engineering notation) by bold capital Latin letters. In
a = b c = d, (1)
(1)
2
indicates the second relation (c = d), and in
a = b = c, (2)
(2)
2
relates to a = c. A list of symbols and abbreviations is given on page ix.
Acknowledgements. The research presented in this thesis was carried out
at the Faculty of Civil Engineering and Geosciences at Delft University of
Technology under the supervision of L.J. Sluys. Support for this research
was provided by the BEO programme (special fund from Delft University
of Technology for excellent research).
I am greatly indebted to J. Alfaiate, H. Askes, J. Blaauwendraad,
G.L. Chiusa, K. De Proft, F.P.X. Everdij, M.G.D. Geers, M. Jir asek, P. Ka-
bele, E. Kuhl, P. Lura, A. Meda, A.V. Metrikine, J. Pamin, T. Pannachet,
R.H.J. Peerlings, J.J.C. Remmers, L.J. Sluys and G.N. Wells for their advice
and suggestions.
A.S.
Delft, the Netherlands
June 2003
List of symbols and abbreviations
Latin symbols
a softening parameter in softening evolution law
a vector of regular nodal displacement degrees of freedom
A cross-sectional area
b softening parameter in softening evolution law
interface width (Appendix A)
b vector of extra nodal displacement degrees of freedom
B
e
matrix containing derivatives of N
e
B
u
matrix containing derivatives of N
u
c gradient parameter
C interface constitutive matrix
C
e
fourth-order compliance tensor
d spring stiffness (Chapter 2)
notch size (Chapter 3)
d
i
vector in the direction of the Gauss point i
D tangent matrix for the bulk material
D
e
fourth-order linear-elastic constitutive tensor
linear-elastic constitutive tangent matrix
D
p
fourth-order elastoplastic consistent tangent tensor
D
pd
fourth-order elastoplastic-damage consistent tangent tensor
e non-local equivalent strain
e component of e when

is crossed by
d
e
+/
positive/negative part of e
e component of e when

is crossed by
d
e
l
local equivalent strain
E Youngs modulus
f yield function
f
cd
stress level corresponding to
cd
f
t
tensile strength
f
t,
stress level corresponding to a specic damage value
f
f /
f
f /
f
f /
f
/
f
ext/int,i
external/internal force vector for dof i
G
f
fracture energy
H hardening parameter
H
1
0
, H
1
Sobolev spaces
H
d
Heaviside function centred at
d
x List of symbols and abbreviations
I
1
rst invariant of the strain tensor
I complementary energy
I fourth-order identity tensor
J
2
second invariant of the deviatoric strain tensor
J
2
second invariant of the deviatoric stress tensor
J total potential energy
k ratio of the compressive and tensile strength for concrete
K
I
mode I stress intensity factor
K global stiffness matrix
K
i j
partition of K related to i j degrees of freedom
K partition of K in interface and PU-based discontinuous elements

l length scale
l
f
bre length
L length
L
2
Sobolev space
L Lagrangian functional
L differential operator
m shape parameter for cohesive law at the discontinuity (Chapter 2)
nodes in interface and PU-based discontinuous elements (Appendix A)
m inward unit normal to
+
n outward unit normal to
N parameter in the overstress function
N displacement shape function matrix (Appendix A)
N
e
non-local equivalent strain shape function matrix
N
u
displacement shape function matrix
p vector of regular nodal non-local equivalent strain degrees of freedom
P applied load
q vector of extra nodal non-local equivalent strain degrees of freedom
r interaction radius
r special second-order tensor (Chapter 4)
D
e
:
s
u (Appendix D)
R distance from the crack tip along the crack line
R special fourth-order tensor (Chapter 4)
rotation matrix (Appendix A)
s special second-order tensor (Chapter 4)
test function (Appendix D)
t time
t tractions at the discontinuity
t prescribed tractions
t
i
traction in the tangential s or normal n direction
t
max
tensile strength
T tangent matrix for the discontinuity
u uniaxial displacement jump
u
y
vertical displacement
u displacement eld
u displacement jump
u prescribed displacements
List of symbols and abbreviations xi
u component of u when

is crossed by
d
u component of u when

is crossed by
d
U
u
space of trial displacements
v deection
v generic unit vector (Chapter 3 and Appendix B)
test function (Appendix D)
V
f
bre volume fraction
V
i
volume related to Gauss point i
x Cartesian line coordinate
x, y Cartesian spatial coordinates
w crack opening displacement
w
e
weight function for the non-local equivalent strain e
w
e
component of w
e
when

is crossed by
d
w
e
component of w
e
when

is crossed by
d
w
i
weight associated to the Gauss point i
w
max
crack opening displacement at zero load
w
u
weight function for the displacement u
w
u
component of w
u
when

is crossed by
d
w
u
component of w
u
when

is crossed by
d
W
u,0
space of admissible displacement variations
Greek symbols
softening parameter in damage evolution law
softening parameter in damage evolution law
boundary surface of

d
discontinuity surface
+/
d
positive/negative part of
d
t
boundary surface where prescribed tractions are applied
+/
t
t
u
boundary surface where prescribed displacement are applied
+/
u
u
d
Dirac-delta function
t time increment
uniaxial strain
i
principal strain in the i direction
cd
strain level for continuous to discontinuous transition
strain tensor
strain array in engineering notation
e
elastic component of the strain tensor
p
plastic component of the strain tensor
vp
viscoplastic component of the strain tensor
angle
equivalent plastic strain (plasticity)
deformation history parameter (damage)
xii List of symbols and abbreviations
0
threshold of damage initiation
c
ultimate equivalent strain at which = 1
equivalent plastic strain for rate-dependent effective stress space
plastic multiplier
Poissons ratio
distance between points x and y
uniaxial stress
uniaxial yield stress
0
yield stress for perfect plasticity

0
initial yield stress (cohesion) for hardening/softening plasticity
e
effective stress dened by the yield criterion in plasticity
i
principal stress in the i direction
stress tensor
stress array in engineering notation
d
deviatoric part of the stress tensor
relaxation time
overstress function (Chapter 4)
regular function (Appendix B)
weight function in non-local averaging
scalar damage variable
crit
critical scalar damage value
critical for the introduction of a cohesive discontinuity
body volume or surface or length (boundary excluded)
+/
body volume or surface or length (boundary included)
+/

Meaning of indices
Subscripts
a related to a dofs
b related to b dofs
d related to discontinuity surface
e related to non-local equivalent strain e
e related to the elastic zone
h discretised quantity
n related to the normal direction to
d
p related to p dofs
p related to the process zone
q related to q dofs
s related to the tangential direction to
d
u related to displacement u
List of symbols and abbreviations xiii
Superscripts
admissible variation
e elastic quantity
p plastic quantity
s symmetric part
T transpose
vp viscoplastic quantity
Abbreviations
C continuous modelling
cf confer = Latin for compare
cmod crack mouth opening displacement
cmsd crack mouth sliding displacement
COD crack opening displacement
D discontinuous modelling
dof(s) degree(s) of freedom
ECC engineered cementitious composite
et al. et alia = Latin for and others
FEM nite-element method
FRC bre-reinforced cement/concrete
HPFRCC high performance bre-reinforced cement composite
i.e. id est = Latin for that is
LVDT linear variable differential transformer
PU partition of unity
RHS right-hand side
SFR steel-bre reinforced
SIFCON slurry inltrated bre concrete
Chapter 1
From continuous to continuous-discontinuous
failure representation
A numerical strategy for a realistic characterisation of failure should con-
sider the representation of nucleation of defects into microcracks, the evolu-
tion of microcracks into macrocracks and the correct macrocrack-microcrack
interaction [87]. This is an extremely complicated task which can be pursued
in a phenomenological approach to failure, although the value of such a pro-
cedure is debatable (see Appendix C).
When modelling failure phenomena, the use of displacement disconti-
nuities is advocated to achieve a better representation of the entire fail-
ure process. Compared to a continuous failure analysis, a continuous-
discontinuous failure analysis, in which discontinuities in the problemelds
arise as a natural consequence of strain localisation, may lead to a more
realistic description of the entire failure process, from diffuse microcrack-
ing to macroscopic traction-free cracks as depicted in Figure 1.1. The use
of a continuous-discontinuous approach in which displacement discontinu-
ities are considered as the natural evolution of strain localisation has been
considered e.g. by Grootenboer [75], Rots [152], Ren and Bi cani c [145], Ka-
bele and Horii [89], Jir asek and Zimmermann [86], Oliver et al. [123, 124]
and Wells et al. [194]. An indication of other continuous-discontinuous ap-
proaches can be found in References [24, 2628, 57, 81, 90, 106, 111, 112] and
references herein.
In the rest of this chapter, the motivations for a continuous-discontinuous
approach to failure are briey recalled and a parallel is drawn between
failure characterisations and numerical strategies for failure description. A
summary of the models for failure description used in the remainder of this
thesis is given in the last part of this chapter.
1.1 The need for discontinuous failure descriptions
In a smeared approach to failure in quasi-brittle materials [59, 60, 151, 153,
154, 180, 181], the cracked material is treated as a continuum and displace-
2 Chapter 1 From continuous to continuous-discontinuous failure representation
(a) (b) (c)
Figure 1.1 Three different stages of failure in a compact-tension specimen: (a) failure ini-
tiation, (b) failure propagation and (c) close to complete failure (the shaded part indicates
microcracking while the thick white line represents a macrocrack).
ment continuity is assumed across the cracked region. To describe the loss
of load-carrying capacity, specic material properties [19, 138] and strain-
softening constitutive relationships with a residual stress are usually em-
ployed (see Figure 1.2a).
Reasons of practical and of theoretical nature are behind the use of strain-
softening constitutive relationships with residual stress. From the practical
side, this setting is, obviously, a very convenient one since it allows numer-
ical analyses to be performed in a continuous framework. Using e.g. con-
tinuum damage softening constitutive relationships with full stress relax-
ation at signicant strain values, such as the one depicted in Figure 1.2b
(solid line), poses the problem of dealing with damage values equal to
unity, i.e. with a singular stiffness matrix. Conversely, considerations of the-
oretical nature lead to the conclusion that the asymptote of such constitu-
tive relationships might be useful in reproducing the long tail observed in
load-displacement diagrams of concrete specimens which can be related to
crack bridging [78, 108]. Unfortunately, due to the inability of describing a
kinematic discontinuity in the primal eld in a continuous setting, the as-
sumption of a smeared reproduction of displacement discontinuity across a
cracked region usually results in a too stiff mechanical response [48, 152] and
in spurious damage growth [64]. Smeared degradation constitutive models,
such as continuum damage or plasticity models, are best suited for mod-
elling diffuse microcracking, in strain-softening or -hardening materials, be-
fore macrocracks become dominant. A better approximation of failure pro-
cesses can be achieved by using numerical techniques in which a disconti-
nuity is naturally endowed in the model itself and is activated at some stage
during localisation (by using a cohesive discontinuity) or after the locali-
1.1 The need for discontinuous failure descriptions 3
strain strain
(a) (b)
s
t
r
e
s
s
s
t
r
e
s
s
Figure 1.2 Softening stress-strain curves employed in a continuous failure description:
(a) with residual stress and (b) with full stress relaxation (solid line) or negligible residual
stress (dashed line).
sation process has been completed (by using a traction-free discontinuity).
Failure can then be realistically described as progressive material degrada-
tion which develops into a discrete crack, for which a discontinuity in the
displacement eld is a suitable representation [110, 193].
To illustrate one of the incongruities due to an erroneous use of strain-
softening relationships in a continuous setting, the global response in terms
of load-deection curve for a beam loaded in four-point bending [78, 130]
is reported in Figure 1.3. The analysis has been performed in the frame-
work of a regularised damage model with a strain-softening constitutive
relationship of the type depicted in Figure 1.2a (see Reference [130] and
Section 3.6.1). In this regularised model (implicit gradient-enhanced dam-
age continuum model [131]), damage evolution is governed by a non-local
measure (non-local equivalent strain e) of the strain tensor. The compari-
son with the experimental curve reported in Figure 1.3a can be considered
satisfactory but it must be realised that it is the result of an unrealistic repre-
sentation of the stress-strain situation in the points surrounding the notch.
From the rst principal stressnon-local equivalent strain curve depicted in
Figure 1.3b, it is evident that part of the load-deection curve in Figure 1.3a
is to be attributed to the contribution of the integration points surrounding
the notch along the crack line. In a more realistic failure representation of
quasi-brittle materials, the contribution of the points surrounding the notch
should vanish close to local failure. The relevance of strain-softening laws
with residual stress (of the type shown in Figure 1.2a) is questionable since
there is no physical rationale behind their denition and their use alters the
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
b
a
v [mm]
experiment
simulation
(a)
0
0.5
1
1.5
2
2.5
3
P

[
k
N
]
0 0.01 0.02 0.03 0.04 0.05
b
e []
a
(b)
0
1
2
3
4
1
[
M
P
a
]
Figure 1.3 Notched beam in four-point bending (30 mm notch depth [78, 130]): (a) load-
deection curve and (b) rst principal stress (
1
) non-local equivalent strain (e) curve for
a point ahead the notch (the points a and b in (b) indicate the integration point contribution
to the global response depicted in (a) for an integration point ahead the notch; also, a and
b indicate the value of the deection v in (a) and of the corresponding strain level e for the
integration point ahead the notch in (b) at different stages of the loading processnote that
the contribution b in (b) is pathological since it is related to a value of
1
which remains
constant, and different than zero, for increasing deformation values e).
understanding of the whole failure process. The use of constitutive laws
with full stress relaxation at signicant strain, in conjuction with a numeri-
cal technique in which the problem elds are allowed to develop a disconti-
nuity, may enable a more realistic characterisation of failure processes.
1.2 Failure characterisation and numerical strategies
Considerations of different nature from the ones in the previous section sug-
gest that the use of a discontinuous interpolation of problem eld could
be benecial, in terms of added exibility with respect to continuous mod-
elling, to the description of the failure behaviour of cementitious compos-
ites.
Failure in cementitious composite materials ranges from quasi-brittle
failure of plain concrete or conventional bre-reinforced cement or con-
crete (FRC) to ductile failure observed in some high performance bre-
reinforced cement composite (HPFRCC) [3, 158]. In plain concrete and
conventional FRC, an initial linear response is followed by a softening
branch which is characterised by the widening of a single crack. Most of
the deformation relates to crack splitting and can be coherently described,
1.2 Failure characterisation and numerical strategies 5
u u
(b) (a)
s
t
r
e
s
s
C
b
D
a
s
t
r
e
s
s
C
b
D
a
Figure 1.4 Strategies for failure analysis of cementitious composites: idealised tensile be-
haviour for (a) quasi-brittle composites and (b) ductile composites (C = continuous mod-
elling, D = discontinuous modelling, = strain, u = displacement jump across the discon-
tinuity surface).
in a computational framework, by softening cohesive surfaces where all
the non-linearity is concentratedthese formulations make use of stress-
displacement relations across the discontinuity surface. An alternative re-
sides in the use of regularised continuous approaches to failure in which
softening stress-strain relations are employed. In some HPFRCC, an initial
linear response is followed by an extended hardening branch characterised
by a uniform deformation stage due to the multiple cracking process. The
multiple cracking phase is followed by a softening branch which stems from
the widening of a single crack. This characterisation is typical of various
composites such as SIFCON [117] (randomly distributed short steel bres
inltrated with cement slurry with bre volume fraction V
f
= 5 20%) or
ECC (cementitious matrix with e.g. V
f
= 1 2% of short random synthetic
bres [100]). The behaviour prior to softening can be described by a stress-
strain relationship which can be deduced after direct translation of uniax-
ial tensile tests due to the relatively homogeneous distribution of cracks in
the hardening regime. The softening part of the global response can be de-
scribed by means of cohesive surfaces.
The numerical strategy just described is purely phenomenological and is
based on the direct translation of the above failure characterisations into
a computational framework. A schematic representation of the strategies
for failure analyses of cementitious composites is reported in Figure 1.4,
where C and D indicates the continuous and the discontinuous range re-
spectively. In the representation of quasi-brittle failure (see Figure 1.4a),
analyses can be conducted in a continuous framework or in a combined
continuous-discontinuous setting. Traditionally, this is done, in the latter ap-
proach, by plugging a discontinuity at the end of the elastic regime, just be-
fore the softening phase (at point a), thus avoiding the use of regularised me-
dia. However, the introduction of a discontinuity close to failure (at point b)
could be benecial to a class of regularised models, improving failure rep-
resentation as will be discussed in Chapter 3. A discontinuity can also be
introduced between points a and b. This may be useful when only some
inelasticity is allowed in the continuum. This approach requires the use of
regularised media and could be considered representative of the behaviour
of some steel bre-reinforced composites. When failure in ductile compos-
ite is considered (see Figure 1.4b), it is necessary to allow some degree of
inelasticity to the bulk to represent diffuse degradation. A discontinuity is
then introduced at the onset of localisation (at point a).
1.3 Requirements for distributed failure models
When introducing discontinuities in the problems elds, some requirements
on the underlying continuum description must be satised. Obvious re-
quirements are related to the nature of the continuum model which should
be regularisednumerical results must be independent of mesh type, size and
orientationand should produce a failure mode which is physically reasonable
(the location of failure initiation and the evolution of failure should be cor-
rectly predicted). Less obvious requirements, but equally important, reside
in the ability of the model to allow the formation of a localised strain prole
with full stress relaxation at signicant deformation and without spurious damage
growth close to failure for a correct description of a stress-free crackin other
words, the problem elds of the continuous model should converge to a re-
alistic discontinuous state. The above requirements should also be taken into
account when a conventional continuous failure analysis is considered.
1.4 Models for inelastic bulk deformation
Inelastic bulk behaviour can be effectively described by continuum damage
or plasticity theories which can be opportunely tailored to account for the
representation of the behaviour of various materials. A summary of quasi-
static damage and plasticity models used in this thesis is given next. The
summary is limited to the geometrically linear case. A more detailed de-
scription can be found in Criseld [46, 47] and Jir asek and Ba zant [84].
1.4 Models for inelastic bulk deformation 7
Continuum damage theories
Continuum damage theories are usually considered to describe material
degradation as a consequence of the growth of microstructural defects. The
basic assumption characterising isotropic elasticity-based damage models
resides in the progressive reduction of the elastic stiffness via a scalar dam-
age parameter , ranging from 0 (virgin material) to 1 (completely damaged
material). Consequently, the basic relation between total stress and total
strain reads
= (1 ) D
e
:, (1.1)
where D
e
is the fourth-order linear-elastic constitutive tensor. It is also as-
sumed that the damage history of the material is considered a function of
the monotonically increasing deformation history parameter , whose evo-
lution is governed by the Kuhn-Tucker relations
0, e
l
0, (e
l
) = 0, (1.2)
which are related to the denition of a scalar measure e
l
of the strain tensor.
Failure characterisation is dened next through a damage evolution lawand
an equivalent strain denition.
Equivalent strain denition. Different weight can be given to the compo-
nents of the strain tensor in the construction of an equivalent strain measure.
When only tensile strains are relevant, use can be made of the expression
e
l
=
_
3
i=1
i
)
2
, (1.3)
due to Mazars [104, 105], with
i
) = (
i
+[
i
[)/2 and
i
the principal strains,
based on the positive principal strain components
i
). Conversely, when all
the components are considered, the modied von Mises denition [63, 188]
e
l
=
k 1
2k (1 2)
I
1
+
1
2k
_
(k 1)
2
(1 2)
2
I
2
1
+
6k
(1 +)
2
J
2
, (1.4)
with
I
1
= tr () , J
2
= tr ( )
1
3
tr
2
() , (1.5)
k the ratio of the compressive and tensile strength and the Poissons ratio,
can be used.
Damage evolution laws. Some characteristics of global behaviour of a
specimen can be directly related to the shape of the damage evolution law
as briey discussed in Section 1.1. There are two classes of damage evolu-
tion laws which are diversied by the strain value at which a fully damaged
state is reached. In the rst category there are laws for which a critical dam-
age value is reached at a signicant strain level such as the linear softening
damage evolution law
=
_
_
_
0 if <
0
1
0
if
0

c
1 if >
c
,
(1.6)
which is characterised by a linear decrease of the stress until a zero-stress
level is reached at ultimate strain
c
with
0
the threshold of damage initia-
tion, or the modied power softening law [64]
=
_
_
0 if <
0
1
_
0
_
if
0

c
1 if >
c
,
(1.7)
which is a modied version of the linear softening law (1.6) in which the
model parameters and inuence the slope and the shape of the softening
curve, respectively.
In the second category of damage evolution laws, critical damage is never
reachedit is possible to reach unit damage only asymptotically, at innite
strain values. One such law is the exponential softening law [134]
=
_
0 if <
0
1
(1 +exp ((
0
))) if
0
,
(1.8)
with and model parameters inuencing the residual stress level and the
slope of the softening curve, respectively.
Regularisation in presence of softening constitutive behaviour. When
strain-softening constitutive equations are exploited to describe the progres-
sive loss of load-carrying capacity in a continuous setting, regularisation
techniques must be considered to preserve well-posedness of the govern-
ing equations. A widely used technique relies on the inclusion of non-local
terms in the constitutive equations. In the non-local damage model pro-
posed by Pijaudier-Cabot and Ba zant [139], non-locality enters the consti-
tutive equations through the denition of a non-local scalar measure e of
the strain tensor dened in the body as
e(x) =
(y; x) e
l
(y) d(y)
(y; x) d(y)
, (1.9)
where is a homogeneous and isotropic weight function which determines
the inuence of point y on x through a decaying weight which is a function
of a length scale l. The normalised Gaussian function
() =
1
2l
2
exp
_
2
2l
2
_
in R
2
, (1.10)
where is dened as the distance between the points y and x, is usually
taken as the weight function in integral non-local models. The non-local
state variable e replaces its local counterpart e
l
in the Kuhn-Tucker rela-
tions (1.2) which now read
0, e 0, (e ) = 0. (1.11)
An approximate differential format of (1.9)-(1.10) has been derived by Peer-
lings et al. [134] and reads
e c
2
e = e
l
in , (1.12)
where c =
1
2
l
2
and
2
is the Laplacian operator. Equation (1.12) is a modi-
ed Helmholtz equation and is combined with the boundary condition
e n = 0 on , (1.13)
where n is the outward unit normal at the boundary of and e is the
gradient of e. In a nite-element context, e represents an additional degree of
freedom to the standard ones. The equivalence of the two formats has been
discussed by Peerlings et al. [135].
Continuum plasticity theories
The ow theory of plasticity is usually considered to describe elastoplastic
material behaviour. In the geometrically linear case, the basic assumption is
the decomposition of the strain tensor into an elastic and a plastic compo-
nent:
=
e
+
p
. (1.14)
Consequently, the relation between total stress and elastic strain yields
= D
e
:
e
= D
e
: (
p
) . (1.15)
The plastic strain rate
p
is postulated, in the case of an associated ow rule
to which this study is restricted, as

p
=

f
, (1.16)
where the rate

of the plastic strain multiplier determines the magnitude
of the plastic ow while the gradient f
= f / to the yield surface f is

a tensor indicating the direction of the plastic strain-rate. Similar to damage
theories, loading-unloading conditions are expressed by the Kuhn-Tucker
relations
0, f 0,

f = 0. (1.17)
In ideal plasticity, the yield function is expressed as
f () =
e
()
0
, (1.18)
which is a function of the stress tensor only and where
e
is the effective
stress, as dened by the yield criterion, and
0
is the yield stress. Soften-
ing/hardening constitutive behaviour is introduced by making the yield
stress a function of a scalar measure of the plastic strain tensor. The equiv-
alent plastic strain is dened through the strain-hardening hypothesis
which denes the relation of proportionality between the equivalent plas-
tic strain-rate and the plastic strain-rate multiplier

. The characterisation
of elastoplastic material behaviour is based on the denition of the yield
function, limited here to isotropic hardening, through the effective and yield
stress denitions.
Yield function. Two yield functions have been considered. The rst is the
classical von Mises yield function
f (, ) =
_
3J
2
() , (1.19)
Figure 1.5 Smoothed Rankine yield function [128].

where () is the uniaxial yield stress whose expression will be specied
later and J
2
=
1
2
d
:
d
is the second invariant of the deviatoric stress ten-
sor
d
. The second yield function is the Rankine yield function which is
widely used to describe cracking in quasi-brittle materials. It is a principal
stress criterion and is characterised by a vertex in the principal stress space.
Its smoothed version, proposed by Pamin [128] and depicted in Figure 1.5,
reads
f (, ) =
_
_
_
1
() if
2
0
_
2
1
+
2
2
() if
2
> 0,
(1.20)
where
i
are the principal stresses (
1
>
2
), and has been considered in
some of the analyses reported in Chapter 4 and Appendices C and E. For
both yield criteria, under the assumption of strain hardening, =

. Plane
stress and plane strain plasticity have been considered (plane stress plas-
ticity has been formulated in the framework proposed by Simo and Tay-
lor [167]).
In describing softening constitutive behaviour, the rule governing ()
has been given an exponential form [69] according to
() =
0
((1 + a) exp (b) a exp (2b)) , (1.21)
with a and b model parameters and
0
the initial cohesion (or yield stress).
Conversely, in case of hardening material behaviour, a linear hardening
model of the type
() =
0
+ H, (1.22)
where H is the hardening parameter, is considered.
Regularisation in presence of softening constitutive behaviour. The
introduction of rate terms in the constitutive equations has been consid-
ered to preserve the regularity of the governing equations in case of strain-
softening constitutive equations. Rate-dependent regularisation is consid-
ered in a viscoplasticity framework specied by the models of Perzyna [137]
and Duvaut-Lions [54, 165]. Similar to rate-independent plasticity, in small
strain viscoplasticity the strain tensor is decomposed into an elastic and a
viscoplastic component according to
=
e
+
vp
(1.23)
which is combined with the elastic stress-strain law (1.15)
1
. The two vis-
coplasticity formats will be described in Chapter 4 in the context of their
coupling to damage.
1.5 A model for separation across cohesive surfaces
As an approximation of the behaviour of quasi-brittle materials, material
degradation can be considered as lumped at cohesive surfaces [185, 197].
Depending on the modelling characterisation of the failure process, degra-
dation can be initiated on cohesive surfaces or can be considered as the nat-
ural evolution of diffuse bulk degradation into a localised crack with some
residual stress. In general terms, the traction at a discontinuity surface
d
reads
t = f (u, history) on
d
. (1.24)
A simple law can be formulated in terms of tractions t
n
and displacement
jumps (Crack Opening Displacement, COD) w normal to the discontinuity
surface:
t
n
(w) =
_
t
max
_
1
w
w
max
_
m
if 0 w w
max
0 if w > w
max
,
(1.25)
where t
max
is the tensile strength, w
max
is the COD at zero stress and m > 1
is a shape parameter. In the analysis of bre composites, w
max
relates to the
bre length and m depends on the bre type [82].
Chapter 2
Continuous-discontinuous failure in standard media
Displacement discontinuities are incorporated in standard nite elements

for the numerical treatment of geometrical discontinuities, e.g. rock
joints [72], or for the description of problems with evolving boundaries, e.g.
cracks in quasi-brittle materials [159] or delamination in composite mate-
rials [6]. The incorporation of a discontinuity in the displacement eld is
conventionally pursued by using interface elements or embedded disconti-
nuity elements. With interface elements, a discontinuous displacement eld
is described by the relative displacement of a double set of nodes at inter-
element boundaries. This requires special mesh generators and, in the case
of evolving boundaries, remeshing procedures [25]. Alternatively, the effect
of a discontinuity in the displacement eld can be described by the use of
embedded discontinuity elements in which the displacement discontinuity,
described by using the Heaviside function, is incorporated into the nite-
element formulation as an incompatible strain mode and can pass through
solids elements [7, 122, 166].
A different approach employs the partition of unity property of nite-
element shape functions (the sum of the shape functions must equal unity
at each spatial point) [52, 107]. Within this approach, the standard approxi-
mation basis is enriched locally with special functions. This enrichment re-
sults in extra degrees of freedom for the nodes in the domain subjected to
the enrichment, without modication of the mesh topology. The effect of
a displacement discontinuity is described by enriching the standard nite-
element polynomial basis with the Heaviside function [110]. The outcome of
this approach is a class of elements, in the following referred to as partition
of unity-based discontinuous elements (in short PU-based discontinuous
elements), which is kinematically equivalent to the class of conventional
interface elements, the key difference being the possibility of arbitrarily lo-
cating the discontinuity within the domain of an element. When PU-based
discontinuous elements are used, the interface response is described by an
additional set of global degrees of freedom and by a constitutive law at the
Based on References [168, 171, 174]

14 Chapter 2 Continuous-discontinuous failure in standard media
t
+
d
d
u
n
m
t
Figure 2.1 Body

crossed by a discontinuity
d
.
discontinuity.
Next, the characterisation of the problem elds for a body crossed by a
discontinuity is recalled. The variational formulation and its linearised dis-
crete format are derived following standard procedures. Some issues related
to the element technology are discussed and some applications to stationary
and propagating discontinuities in elastic and inelastic bulk materials con-
clude the chapter.
2.1 Problem elds
When a body is crossed by a discontinuity, it is necessary to characterise the
problem elds in a proper way. Here, the displacement jump is described by
the Heaviside function operating on a smooth and continuous function. Dif-
ferent from Wells and Sluys [193], the strain eld is dened everywhere in
the body except at the discontinuity surface. This format leads to a straight-
forward derivation of a variational statement in local and non-local (e.g.
gradient-enhanced) continua as will be illustrated later in this chapter and
in Chapter 3, respectively.
The body

, depicted in Figure 2.1, is bounded by and is crossed by
a discontinuity surface
d
which divides the body into two sub-domains,
+
and
( =
+

). The boundary surface of the body

consists
of three mutually disjoint boundary surfaces,
u
,
t
and
d
( =
u

t
).
Displacements u are prescribed on
u
, while tractions

t are prescribed on
t
.
2.1 Problem elds 15
Figure 2.2 Discretised body

crossed by a dis-
continuity
d
(the white circles indicates the nodes
with extra degrees of freedom b).
In the body

, the displacement eld can be decomposed as
u(x, t) = u(x, t) +H
d
(x) u(x, t), (2.1)
where H
d
(x) is the Heaviside function centred at the discontinuity surface
d
(H
d
= 1 if x

+
, H
d
= 0 if x

) and u and u are continuous func-

tions on

. In the geometrically linear case, the strain eld in is computed
as the symmetric part of the gradient of the displacement eld and reads
(x, t) =
s
u (x, t) +H
d
(x)
s
u (x, t) if x /
d
, (2.2)
where ()
s
refers to the symmetric part of ().
2.1.1 Problem eld interpolation
The discrete representation of a discontinuity can be rigorously achieved
through a discontinuous interpolation of the problem elds [110, 193]. To
begin with, it is useful to consider the discretised version of the body

depicted in Figure 2.2. In standard nite elements, each node of the discre-
tised medium is given a standard set a of degrees of freedom representing
the displacements in the Cartesian directions. With PU-based discontinu-
ous elements, a node is given an extra set b of degrees of freedom when
a discontinuity crosses the support of that node. Following standard pro-
cedures [110, 193], the discretised format of (2.1) reads, for nodes whose
support is crossed by a discontinuity,
u
h
= N
u
a +H
d
N
u
b, (2.3)
where N
u
is an array containing standard nite-element shape functions
and the global nodal degrees of freedom a and b represent, in the arrange-
ment of (2.3), the total displacement eld. The displacement jump across the
discontinuity
d
is given by
u
h
= N
u
b
d
. (2.4)
The strain eld in (2.2) can be discretised in a similar fashion and reads,
using engineering notation for the stress tensor,
h
= B
u
a +H
d
B
u
b if x /
d
, (2.5)
where B
u
= LN
u
and L is the differential operator
L =
_
x
0 0

y
0

z
0

y
0

x
z
0
0 0

z
0

y
x
_
_
T
. (2.6)
For nodes whose support is not crossed by a discontinuity, the Heaviside
function is a constant function over their supports and therefore it is not con-
sidered. Consequently, since there is no enhancement, the standard nite-
element interpolation u
h
= N
u
a is retrieved. Note that the inclusion of in-
ternal discontinuity surfaces in a nite element is equivalent to the applica-
tion of natural boundary conditions, without modications of the original
nite-element mesh, at the discontinuity surface.
The interpolation in (2.3) can be understood as an enrichment of the stan-
dard polynomial nite-element spaces by a special function (the Heaviside
function H
d
) which reects known information (the presence of a displace-
ment jump) about the boundary value problem. The special function H
d
is
multiplied with the shape functions N
u
(which are also partitions of unity)
and then pasted to the existing nite-element basis to construct a conform-
ing approximation [52, 107, 121].
The PU-based discontinuous element approach can be termed as a
smeared discontinuity approach since the displacement jump is described
at the nodes whose support is crossed by a discontinuity by the b degrees of
freedom. Indeed, there are no modications of the existing mesh topology,
i.e. no extra nodes are added where a discontinuity intersects an element
(see e.g. the application in Section 2.6.1). However, contrary to embedded
discontinuity elements [7, 122, 166], in PU-based discontinuous elements the
discontinuity contribution to the element stiffness matrix is properly taken
into account (see Section 2.5 under Numerical integration). A formulation of
embedded discontinuity elements with extra nodes where a discontinuity
intersects an element has been derived by Alfaiate et al. [4, 5].
2.2 Governing equations 17
2.2 Governing equations
The equilibrium equations and boundary conditions for the body

without
body forces can be summarised by
= 0 in , n =

t on
t
, m = t on
d
, (2.7)
where is the Cauchy stress tensor, n is the outward unit normal to the
body and m is the inward unit normal to
+
on
d
(see Figure 2.1). Equa-
tion (2.7)
3
represents traction continuity at the discontinuity surface
d
. The
strong form is completed by the boundary conditions
u = u on
u
, u = 0 on
u
, (2.8)
where u is a prescribed displacement. Equation (2.8)
1
is a standard essential
boundary condition while (2.8)
2
has been imposed to simplify the nite-
element implementation [193]. The constitutive relationships for bulk and
discontinuity degradation will be specied in Section 2.6.
2.3 Variational formulation
Following standard procedures, the governing equations (2.7)
1
to (2.7)
3
will
be cast in a weak form. The space of trial displacements is dened by the
function u = u +H
d
u with u and u U
u
where
U
u
=
_
u
i
and u
i
[ u
i
and u
i
H
1
() and u
i
[
u
= u
i
, u
i
[
u
= 0
_
(2.9)
with H
1
a Sobolev space; u
i
and u
i
indicate the i
th
component of u and u,
respectively. The space of admissible displacement variations is dened by
the weight function w
u
= w
u
+H
d
w
u
with w
u
and w
u
W
u,0
where
W
u,0
=
_
w
u,i
and w
u,i
[ w
u,i
and w
u,i
H
1
()
and w
u,i
[
u
= w
u,i
[
u
= 0
_
, (2.10)
where w
u,i
and w
u,i
indicate the i
th
component of w
u
and w
u
, respectively.
The equilibrium equations (2.7)
1
are multiplied by the weight function w
u

W
u,0
, which is decomposed into w
u
and w
u
consistent with the displacement
decomposition in (2.1), and integrated over the domain to obtain the weak
equilibrium statement
( w
u
+H
d
w
u
) () d = 0. (2.11)
The term related to the continuous part of the displacement eld ( w
u
) can be
expanded using integration by parts, Gauss theorem and the relationship
n =

t on
t
to yield
w
u
() d =
( w
u
) d
s
w
u
: d
=
t
w
u
t d
s
w
u
: d. (2.12)
Similarly, the term related to the discontinuous part of the displacement
eld (H
d
w
u
) is expanded using integration by parts:
d
w
u
() d =
+
w
u
() d
=
+
( w
u
) d
s
w
u
: d. (2.13)
Using Gauss theorem and the relationships m = t and n =

t, the rst
term of the RHS of (2.13) reads
+
( w
u
) d =
+
t
w
u
(n) d
+
d
w
u
(m) d
=
+
t
w
u
t d
+
d
w
u
t d , (2.14)
where
+
t
and
+
d
are parts of the boundary
+
. The weak form therefore
reads
s
w
u
: d+
s
w
u
: d+
d
w
u
t d
=
t
( w
u
+H
d
w)
t d , (2.15)
in which the terms related to
t
and
+
t
have been collected under the same
integral using the Heaviside function. Since the function w
u
is continuous
across the discontinuity and since the notation
+
d
has been introduced to
indicate which part of the discontinuity is under analysis, the domain
+
d
of
the integral of the traction t has been changed into
d
.
From the decomposition of the displacement eld it follows that any ad-
missible variation w of u can be regarded as admissible variations w
u
and
w
u
, thus leading to two variational statements. Taking rst variation w
u
2.4 Discretisation and linearisation 19
( w
u
= 0) and then w
u
( w
u
= 0) yields the nal form of the variational
statements:
s
w
u
: d =
t
w
u
t d w
u
W
u,0
(2.16a)
s
w
u
: d+
d
w
u
t d =
+
t
w
u
t d w
u
W
u,0
. (2.16b)
The second variational statement ensures that traction continuity is satised
in a weak sense across the discontinuity
d
. The two variational statements
in (2.16) resemble a coupled problem in which the elds u and u are coupled
in the continuum through the expression of the stress eld and the effect of
the discontinuity is taken into account by the integral over
d
. It is also worth
noting that the variational statements in (2.16), compared to the variational
statement related to a body under a continuous kinematic eld,
s
w
u
: d =
t
w
u
t d w
u
H
1
0
(), (2.17)
could have also been derived writing the principle of virtual work applied
to the body

under the continuum displacement eld u and applied to
the body

+
under the displacement eld H
d
u related to the displacement
jump. Note also that the principle of virtual work for a body crossed by a
discontinuity is recovered by summing (2.16a) and (2.16b) [4, 103].
2.4 Discretisation and linearisation
Next, the linearised format of the equilibrium equation is derived. In the
following, the subscript h is dropped from discretised quantities and engi-
neering notation for and is used.
2.4.1 Problem eld description
Following a Bubnov-Galerkin approach, (2.1) and (2.2) can be discretised in
each element with extra degrees of freedom b using a format similar to (2.3)
and (2.5):
u = N
u
a u = N
u
b
s
u = B
u
a
s
u = B
u
b (2.18)
w
u
= N
u
a
w
u
= N
u
b

s
w
u
= B
u
a

s
w
u
= B
u
b
. (2.19)
In the above relations, the primes refer to admissible variations. For ele-
ments with standard degrees of freedom a only, the problem elds can be
discretised in a standard fashion.
2.4.2 Discretised and linearised weak governing equations
The discrete format of the problem elds leads to the two discrete weak
governing equations
B
T
u
d =
t
N
T
u
t d (2.20a)
+
B
T
u
d+
d
N
T
u
t d =
+
t
N
T
u
t d , (2.20b)
from which the equivalent nodal force vector related to admissible varia-
tions of a and b result in
f
int,a
=
B
T
u
d f
ext,a
=
t
N
T
u
t d (2.21a)
f
int,b
=
+
B
T
u
d+
d
N
T
u
t d f
ext,b
=
+
t
N
T
u
t d . (2.21b)
The linearised form of the discretised weak governing equation is obtained
by substituting the stress rate in the bulk
= D = D
_
B
u
a +H
d
B
u

b
_
, (2.22)
with D the tangent matrix for the bulk material, and the traction rate at a
discontinuity
t = T u = T
_
N
u

b
_
d
, (2.23)
where T relates traction rate

t and displacement jump rate u, in (2.20).
After standard manipulations, the linearised weak form of the governing
equations at iteration i within a time step n reads
_
K
n,i1
aa
K
n,i1
ab
K
n,i1
ba
K
n,i1
bb
_ _
a
n,i
b
n,i
_
=
_
f
n
ext,a
f
n
ext,b
_
_
f
n,i1
int,a
f
n,i1
int,b
_
, (2.24)
where
K
aa
=
B
T
u
DB
u
d (2.25a)
K
ab
=
+
B
T
u
DB
u
d (2.25b)
2.5 Element technology 21
K
ba
= K
T
ab
=
+
B
T
u
DB
u
d (2.25c)
K
bb
=
+
B
T
u
DB
u
d+
d
N
T
u
TN
u
d (2.25d)
f
int,a
=
B
T
u
d (2.25e)
f
int,b
=
+
B
T
u
d+
d
N
T
u
t d (2.25f)
f
ext,a
=
t
N
T
u
t d (2.25g)
f
ext,b
=
+
t
N
T
u
t d . (2.25h)
Symmetry of the stiffness matrix is assured if the material tangent matrices
D and T are symmetric. The derivation of the variational formulation pre-
sented here differs from the one reported by Wells and Sluys [193] but yields
the same weak and linearised weak form of the governing equations for a
body crossed by a discontinuity.
Note that the arrays N
u
multiplying a and b in (2.3) are normally not the
same since only part of the extra degrees of freedom in the array b might
be activated. Consequently, the matrices B
u
multiplying a and b in (2.5) are
also normally not the same. However, since the system of equations in (2.24)
is assembled only for the active degrees of freedom, it is still possible to use
the standard N
u
and B
u
matrices.
2.5 Element technology
In the following, some issues pertinent to the implementation of PU-based
discontinuous elements are discussed. Other issues, such as the choice
and the activation of the extra degrees of freedom, are discussed in Ref-
erences [51, 110, 193].
Introducing a discontinuity. As soon as a critical situation is detected in
the element ahead of a discontinuity tip the discontinuity is extended. In
the examples reported in Sections 2.6.3 and 2.6.4, a cohesive discontinuity
is extended from a pre-existing traction-free discontinuity; this corresponds
to the introduction of a discontinuity at point a as depicted in Figure 1.4. A
principal stress criterion is used. This criterion is applied by sampling the
discontinuity tip
Figure 2.3 Extension of a discontinuity within
the body

(the black circles indicates nodes with
standard degrees of freedom a while the white
circles indicates nodes with extra degrees of free-
dom b for the nodes whose support is crossed by
a discontinuity).
maximum principal tensile stress at all the integration points in the element
ahead the discontinuity tip at the end of a load increment. Mesh renement
studies suggest that the total energy dissipated during crack propagation in
an elastic medium is a constant material parameter [193]. This gives an in-
dication of the validity of the above criterion. From a mathematical point of
view this is not correct since, for very high mesh densities, the elastic solu-
tion is recovered, and criteria based on stress or strain quantities are mean-
ingless. In that perspective, the use of criteria based on energy considera-
tions should be considered [109]. However, since the nite-element solution
is known to converge very slowly to the elastic solution, the above maxi-
mum stress criterion can be accepted with some condence. For the mesh
densities considered in this chapter, this criterion can be considered equiva-
lent to the sampling of the maximum principal stress at some distance from
the crack tip as suggested by Williams and Ewing [195]. In inelastic local
media, this issue is of no concern since the stress at the discontinuity tip is
bounded in the inelastic regime.
To preserve the robustness of the Newton-Raphson solution procedure, a
discontinuity is introduced as a straight segment at the end of a load incre-
ment in the element ahead of a discontinuity tip if the initiation criterion is
fullled. This procedure is repeated in the elements ahead of this element
with the extended discontinuity until the initiation criterion is no longer
satised.
To reproduce a crack tip, the displacement jump at the discontinuity tip
is set to zero. This is achieved by considering only standard a dofs for the
nodes on an element boundary touched by a discontinuity (see Figure 2.3).
When a discontinuity is extended in a neighbouring element, the nodes be-
hind the discontinuity tip are enhanced.
Figure 2.4 Composite integration scheme for a quadrilateral
element crossed by a discontinuity (the crossed circles are in-
tegration points on the discontinuity and the black dots are
integration points for the continuum).
discontinuity
Orienting a discontinuity. For the example in Section 2.6.4, the direction
of discontinuity extension is computed using the principal directions of a
non-local stress tensor which is calculated as a weighted average of stresses
using a Gaussian weight function [86, 193]. The weight w
i
associated with
integration point i are computed from
w
i
=
1
(2)
3/2
r
3
exp
_
|d
i
|
2
2r
2
_
, (2.26)
where d
i
is the vector in the direction of the integration point i and the in-
teraction radius r is equal to three times the average element size ahead the
discontinuity tip. The discontinuity propagates in the direction normal to
the direction of the maximum non-local principal stress. When the disconti-
nuity is close to a boundary, the discontinuity extension direction is aligned
with the previous discontinuity segment to avoid an incorrect direction de-
termination due to the bias produced by an unsymmetric domain in (2.26).
Numerical integration. When dealing with integration of the element ma-
trices only on a part of an element domain, it is necessary to consider al-
ternative integration rules [110]. When an element is intersected by a dis-
continuity, the two resulting domains are triangulated and each triangular
sub-domain is mapped to a parent unit triangle over which a three-point
symmetric quadrature rule, with interior points within the triangular sub-
domain, is considered (see Figure 2.4). The discontinuity contribution to the
stiffness matrix and the internal force vector are integrated on the discon-
tinuity through a 2point Gauss integration scheme unless otherwise indi-
cated. For elements not crossed by a discontinuity, a standard 2 2 Gauss
integration scheme is used for numerical convenience.
It is worth noting that the resulting composite integration scheme for the
continuum reported in Figure 2.4, which consists of 24 integration points in
eight triangular sub-domains, integrates correctly a quadratic function over
spring
L L
x
P
Figure 2.5 Geometry and boundary conditions for the tension test.
a parent quadrilateral element. In the numerical integration of the element
matrices, this composite integration scheme results, for the terms containing
the strain energy density, in a full integration rule for bilinear quadrilateral
elements and in a reduced integration rule for a quadratic quadrilateral ele-
ments. The actual construction of the composite integration rule is based on
an exact algebraic inverse iso-parametric mapping for bilinear quadrilateral
elements [201]. This technique is based on the use of quadrilateral elements
with straight sides and evenly spaced side nodes.
Transfer of history data. The transfer of history data is performed within
an element only for those elements crossed by a discontinuity in which the
bulk experiences inelastic deformations [194]. The maximum value of the
equivalent plastic strain within the element not yet crossed by a disconti-
nuity is considered as representative of the state of the entire element. Once
this value is identied, the transfer of the history data to it related is per-
formed by copying this history array to the integration points related to the
composite integration scheme as dened in the above paragraph. After the
transfer is performed, the increment which led to the discontinuity exten-
sion is recomputed to allow a consistent dissipation in the elements around
the discontinuity tip.
2.6 Applications
Some applications of PU-based discontinuous elements are presented. To
begin with, a simple one-dimensional example is given to get acquainted
with this class of elements. Application to quasi-brittle and to ductile fail-
ure of cementitious composites follows a study of the performance of the
method with respect to a problem in which conventional interface elements
do not perform well.
2.6 Applications 25
2 1 3 4
L L
1
L L/2
2 3 4
L/2
(a) (b)
discontinuity discontinuity
Figure 2.6 Discretisation for the tension test with (a) conventional interface elements and
(b) PU-based discontinuous elements (the black circles indicates the standard degrees of
freedom a while the white circles indicates the extra degrees of freedom b; the discontinuity
is indicated by the vertical dotted line).
2.6.1 Setting the scene: a simple one-dimensional example
A simple one-dimensional example is given next to illustrate some of the
issues involved in the element technology for PU-based discontinuous ele-
ments. The example deals with the elastic solution of a one-dimensional bar
in tension with a spring in the cross section as depicted in Figure 2.5. The so-
lution will be given for standard and for PU-based discontinuous elements.
The bar is modelled by means of the two discretisations depicted in Fig-
ure 2.6. The spring is rst represented by a conventional interface element
(see Figure 2.6a) while the discontinuous interpolation of (2.3) is exploited
for the discretisation depicted in Figure 2.6b. The domain
+
is given by
0 < x < L in Figure 2.5. In the following, E is the Youngs modulus, A is the
cross-sectional area, and d is the spring stiffness. The element types needed
for the nite-element solution are depicted in Figure 2.7.
Stiffness matrix computation. For completeness, the stiffness matrices
of truss and conventional interface elements are reported. For the one-
dimensional truss element of length L depicted in Figure 2.8a the stiffness
matrix reads
K
truss
=
EA
L
_
1 1
1 1
_
. (2.27)
A one-dimensional conventional interface element can be conceived as a
translational spring element with stiffness d = (EA) /L in (2.27).
The sub-matrices in (2.25) are expanded for the truss element with a dis-
continuity in the middle section depicted in Figure 2.8b. Note that the pos-
3
2 3
(c)
1 4 2 3
(a) (b)
1 2 4
Figure 2.7 Element assembly for the discretisation with PU-based discontinuous elements
(note the effect of the choice of the domain
+
on the activation of the extra degrees of
freedom b).
itive part
+
of the domain goes from x = 0 to x = L/2 (H
d
= 1 for
x < L/2; see Figure 2.8b) and the presence of extra degrees of freedom b
for both nodes. Sub-matrix K
aa
is the same as K
truss
. The remaining sub-
matrices are expanded as
K
ab
= K
ba
=
+
B
T
u
DB
u
d =
L/2
0
B
T
u
DB
u
d x =
EA
2L
_
1 1
1 1
_
. (2.28)
For K
bb
, the volume integral equals K
ab
while the surface integral is evalu-
ated on x = L/2 and yields
d
N
T
u
TN
u
d =
_
N
T
u
TN
u
_
x=L/2
=
d
4
_
1 1
1 1
_
. (2.29)
Assembly of the sub-matrices into the element stiffness matrix gives
K =
_
K
aa
K
ab
K
ba
K
bb
_
=
_
_
EA
L

EA
L
EA
2L

EA
2L
EA
L
EA
L

EA
2L
EA
2L
EA
2L

EA
2L
EA
2L
+
d
4

EA
2L
+
d
4
EA
2L
EA
2L

EA
2L
+
d
4
EA
2L
+
d
4
_
_
. (2.30)
If the discontinuity does not cross the element but one of the element nodes
has got extra degrees of freedomb (see e.g. Figure 2.7a) the element stiffness
matrix can be derived following the procedure described above and reads
K =
2EA
L
_
_
1 1 1
1 1 1
1 1 1
_
_
, (2.31)
2.6 Applications 27
b
i
b
j
j
L
i
a a
x
i j
L
x
a a
i j i j
(a) (b)
Figure 2.8 One-dimensional (a) truss element and (b) PU-based discontinuous truss ele-
ment.
where the degrees of freedom have been ordered in the sequence
_
a
1
a
2
b
2
(note that the truss length is L/2).

Assembly and solution. For the bar with the spring modelled by a con-
ventional interface element (see Figure 2.6a), assembly of local stiffness ma-
trices into the global stiffness matrix for the active degrees of freedomresults
in the system of equations
_
_
EA
L
+ d d 0
d
EA
L
+ d
EA
L
0
EA
L
EA
L
_
_
_
_
a
2
a
3
a
4
_
_
=
_
_
0
0
P
_
_
, (2.32)
which yields
_
a
2
a
3
a
4
=
_
PL
EA
PL
EA
+
P
d
2PL
EA
+
P
d
. (2.33)
For the discretisation with the PU-based discontinuous element (see Fig-
ure 2.6b), the global system of equations reads
_
_
3EA
L

EA
L
5
2
EA
L

1
2
EA
L
0
EA
L
3EA
L

1
2
EA
L
1
2
EA
L

2EA
L
5
2
EA
L

1
2
EA
L
5
2
EA
L
+
d
4

1
2
EA
L
+
d
4
0
1
2
EA
L
1
2
EA
L

1
2
EA
L
+
d
4
1
2
EA
L
+
d
4
0
0
2EA
L
0 0
2EA
L
_
_
_
_
a
2
a
3
b
2
b
3
a
4
_
_
=
_
_
0
0
0
0
P
_
_
(2.34)
1 2 4 3
+
=
H
(a)
(b)
(c)
u
u
~
u
^
Figure 2.9 Displacement eld for the one-
dimensional tension test: (a) total displace-
ment eld given by the sum of (b) the u eld
and (c) the H
d
u eld (see (2.35) for the nu-
merical values).
and its solution yields
_
a
2
a
3
b
2
b
3
a
4
=
_
1
2
PL
EA
+
P
d
3
2
PL
EA
+
P
d

P
d

P
d
2PL
EA
+
P
d
. (2.35)
For interface elements, the displacement jump across the discontinuity
d
is expressed as the difference u
h
= a
3
a
2
= P/d of the displacement of
the doubled nodes. In PU-based discontinuous elements, the displacement
jump is expressed through (2.4) as u
h
= N
u
b[
d
= 0.5 (b
2
+ b
3
) = P/d
(see Figure 2.9). Note that the minus sign in the displacement jump value
indicates that the local frame of the discontinuity is in the opposite direction
with respect to the global frame xin other words, the negative sign is due
to the choice of the positive part of the domain.
2.6.2 Linear-elastic analysis of a notched beam: the problem of
oscillations in the traction prole
As partition of unity-based discontinuous elements become more and more
used [110, 143, 193], it is important to assess their limitations. The perfor-
mance of PU-based discontinuous elements in a more realistic setting is ex-
amined through the analysis of the two-dimensional notched beamdepicted
in Figure 2.10 in which the discontinuity is allowed only along the central
line of the beam, later referred to as the crack line. This linear elastic test
was used by Rots [151] to test conventional interface elements. Four-node
and eight-node quadrilateral elements are used under plane stress condi-
tions. A Youngs modulus equal to 2 10
4
N/mm
2
and a Poissons ratio
2.6 Applications 29
20
450
100
P
Figure 2.10 Notched beam (depth=100 mm; all dimensions in millimetres).
equal to 0.2 have been used for the continuum. The traction-separation rela-
tion in (1.24) is formulated in a local s,n coordinate system. A simple elastic
law of the type
_
t
s
t
n
_
= T
sn
u
sn
=
_
d
s
0
0 d
n
_ _
u
s
u
n
_
(2.36)
is used, where d
s
and d
n
are constant, u
s
and u
n
are the displacement jumps
in the local (discontinuity) reference system and t
s
and t
n
are the tangen-
tial and normal interface tractions, respectively. To reproduce pure mode I
opening, only displacement jumps in the horizontal direction are activated.
The notch is described as a traction-free discontinuity (d
s
= d
n
= 0 N/mm
3
)
and to simulate perfect contact prior to the loss of material coherence along
the crack line, a high value of the interface stiffness is there considered. This
approach is usually called the dummy stiffness approach and it is often
used in combination with conventional interface elements.
The analyses performed with conventional interface elements reported
by Rots [151] show that the normal traction prole along the crack line is
highly dependent on the stiffness of the interface and on the chosen numer-
ical integration scheme. In particular, it was shown that high values of the
normal stiffness in combination with exact Gauss integration (2point rule
for linear element and 3point rule for quadratic element) lead to signicant
oscillations in the normal tractions. The results of the analyses performed
with PU-based discontinuous elements are reported in Figures 2.12 and 2.13,
where the normal tractions have been sampled at the integration points on
the discontinuity (similar results have been reported by Audi et al. [11] and
by Remmers et al. [144]). The stiffness d
n
at the discontinuity ranges from
2 10
3
to 2 10
5
N/mm
3
. Results depicted in the upper part of Figures 2.12
Figure 2.11 Unstructured mesh for the notched beam(943 elements;
the discontinuity is indicated by the heavy line).
and 2.13 have been obtained using structured meshes with the discontinu-
ity lying within elements (element size = 3.33 mm)the discontinuity inter-
sected undistorted quadrilateral elements through opposite midsides. The
mesh in Figure 2.11 has been used for the results depicted in the lower part
of Figures 2.12 and 2.13. In structured meshes, PU-based discontinuous ele-
ments show the same spurious traction oscillations of conventional interface
elements when an exact Gauss quadrature scheme is used for the integration
of the traction forces at the discontinuity (upper part of Figures 2.12 and
2.13). Only a nodal integration scheme gives a smooth traction prole for all
the values of the dummy stiffness (in the computations, the trapezoidal rule,
also called 2point closed Newton-Cotes formula, for linear element and the
Simpsons rule, also called 3point Newton-Cotes formula, for quadratic el-
ement were used). These results are similar to those reported by Rots [151]
and by Schellekens and de Borst [160] for conventional interface elements.
However, in unstructured meshes with a high value of the dummy stiffness,
the oscillations in the traction prole are always present (lower part of Fig-
ures 2.12 and 2.13). Results for the limit case of a discontinuity lying on an
element side for a structured mesh are reported in Figure 2.14. It is stressed
that the oscillations are not due to a poorly conditioned system of equations
but stem from the numerical integration of the contribution of the disconti-
nuity to the stiffness matrix. Indeed, the oscillations are present with Gauss
integration schemes (see Figures 2.12a and 2.12b) and disappear with nodal
integration schemes (see Figures 2.13a and 2.13b) when a structured mesh is
considered.
Similar to conventional interface elements, the reasons behind the oscil-
lations in the traction prole for PU-based discontinuous elements are not
well understood. Aheuristic remedy is the use of nodal integration schemes.
However, this approach is successful only for special mesh congurations.
2.6 Applications 31
2 2
2 2
1 1
1 1
0 0
0 0
1 1
1 1
2 2
2 2
3 3
3 3
4 4
4 4
5
4
3
5
4
3
5
4
3
5
4
3
normal traction [MPa] normal traction [MPa]
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
20 20
20 20
40 40
40 40
60 60
60 60
80 80
80 80
100 100
100 100
p
o
s
i
t
i
o
n

[
m
m
]
p
o
s
i
t
i
o
n

[
m
m
]
p
o
s
i
t
i
o
n

[
m
m
]
p
o
s
i
t
i
o
n

[
m
m
]
(a) (b)
(c) (d)
Figure 2.12 Traction proles along the crack line with Gauss integration scheme for
two mesh structures: structured mesh with (a) linear and (b) quadratic elements and
unstructured mesh with (c) linear and (d) quadratic elements (the position is measured
from the bottom of the beam).
In Appendix A, an attempt to link the behaviour of standard interface ele-
ments and PU-based discontinuous elements is presented along the lines of
the derivations pursued by Schellekens and de Borst [160]. Indeed, an anal-
ysis of the stiffness matrices of the elements showed that the two method-
ologies share the structure of the terms governing the interface response.
The problem of the oscillations in the traction prole can be circumvented
by activating the degrees of freedom responsible for the displacement jump
when they are required [110, 193], thus avoiding the initial elastic branch,
as in the following examples. However, this problem may still be present if
favourable conditions exist, e.g. high stress gradient and high discontinuity
stiffness in elastic regime.
2 2
2 2
1 1
1 1
0 0
0 0
1 1
1 1
2 2
2 2
3 3
3 3
4 4
4 4
5
4
3
5
4
3
5
4
3
5
4
3
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
20 20
20 20
40 40
40 40
60 60
60 60
80 80
80 80
100 100
100 100
p
o
s
i
t
i
o
n

[
m
m
]
p
o
s
i
t
i
o
n

[
m
m
]
p
o
s
i
t
i
o
n

[
m
m
]
p
o
s
i
t
i
o
n

[
m
m
]
(a) (b)
(d) (c)
Figure 2.13 Traction proles along the crack line with nodal integration scheme
(trapezoidal rule for linear element and Simpsons rule for quadratic element) for two
mesh structures: structured mesh with (a) linear and (b) quadratic elements and un-
structured mesh with (c) linear and (d) quadratic elements (the position is measured
from the bottom of the beam).
2.6.3 Quasi-brittle failure in a FRC beam in bending
Similar to quasi-brittle materials such as plain concrete, some bre-
reinforced materials fail as a consequence of the development of a major
crack. Their behaviour up to failure can then be described by the same tech-
niques used for plain concrete by employing discrete softening constitutive
laws [193].
The exural behaviour of a series of mortar beams reinforced with
aramide bres depicted in Figure 2.15 has been analysed by Ward and
Li [191]. The beam was tested with different bre types and volume frac-
tions. Aramide (kevlar) bres with l
f
= 6.4 mm and bre volume fraction
V
f
= 0.5, 1.0, 1.5% were used in the beams considered for the numerical
2.6 Applications 33
2 2
2 2
1 1
1 1
0 0
0 0
1 1
1 1
2 2
2 2
3 3
3 3
4 4
4 4
5
4
3
5
4
3
5
4
3
5
4
3
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
d=2 x 10 MPa/mm
20 20
20 20
40 40
40 40
60 60
60 60
80 80
80 80
100 100
100 100
p
o
s
i
t
i
o
n

[
m
m
]
p
o
s
i
t
i
o
n

[
m
m
]
p
o
s
i
t
i
o
n

[
m
m
]
p
o
s
i
t
i
o
n

[
m
m
]
(a) (b)
(d) (c)
Figure 2.14 Traction proles along the crack line with (a, b) Gauss and (c, d) nodal in-
tegration scheme for structured mesh with discontinuity along element side: (a, c) lin-
ear and (b, d) quadratic elements (the position is measured from the bottom of the
beam).
analyses. In the experiment, the load was applied via displacement control
and the midspan deection was measured by a linear variable differential
transformer (LVDT) placed at the centre of the beam. The beam has been
analysed using a structured mesh with 8node quadrilateral elements in
plane stress with element size in the central part of the beam equal to 3 mm;
a Youngs modulus equal to 30 GPa and a Poissons ratio equal to 0.2 were
used. Inelasticity was introduced through the cohesive law in (1.25) with
m = 2 and composite tensile strength t
max
= 1.96, 2.60, 3.25 MPa for a bre
content equal to 0.5, 1.0, 1.5%, respectivelythese model parameters have
been derived by Kullaa [94] from micromechanical properties using a mi-
cromechanical statistical tensile model. This analysis falls under the case
depicted in Figure 1.4a, with the discontinuity activated at point a (a co-
d
d
d d
thickness=0.55d
d = 114 mm
Figure 2.15 Geometry and boundary con-
ditions for the bending test.
0
simulation
experiment
increasing V
f
0.2 0.4 0.6 0.8 1 1.2
Deflection [mm]
0
5
10
15
L
o
a
d

[
k
N
]
Figure 2.16 Load-deection curves for -
bre volume fractions V
f
= 0.5, 1.0, 1.5%.
hesive discontinuity is extended along the central line of the beam from a
pre-existing vertical traction-free discontinuity placed at the bottom of the
beam).
The numerical and experimental load-deection curves are reported in
Figure 2.16 in which the applied load is plotted against the midspan de-
ection. A good agreement was found only for the beam with the highest
bre volume fraction (V
f
= 1.5%); for the other bre volume fractions the
peak load of the beams was predicted lower and the post-peak response
was more ductilethese results are very similar to those reported by Kul-
laa [94] who used conventional interface elements. As suggested by Wells
and Sluys [193], a possible remedy to modify the post peak regime could
be the introduction of a more complicated dissipation mechanism through
the introduction of displacement contributions parallel to the crack (slid-
ing). Such an approach may lead to a better t between numerical and ex-
perimental curves but is rather difcult to justify on physical grounds (see
Appendix C for a brief discussion on this issue).
2.6.4 Ductile failure in a composite compact-tension specimen
In a class of engineered cementitious composites (ECC) [100], inelasticity
develops as a consequence of multiple cracking which is initiated on planes
normal to the maximum principal stress. In tensile specimens, cracking is
diffuse over a large part of the specimen and the portion of total displace-
ment associated to it is not recoverable when the specimen is unloaded to
2.6 Applications 35
63.5
127
63.5
25.4
360
310
117
thickness=61
300
Figure 2.17 Compact-tension specimen (all dimensions in mm).
zero stress state. The composite can then be idealised as homogeneous and
continuous in the multiple cracking stage and the Rankine yield criterion
can be used to characterise inelastic behaviour in the hardening phase be-
fore localisation [89]. It is noted that the use of anisotropic plasticity is better
suited to describe diffuse cracking with a preferential directional pattern.
However, since in this study the failure mode is controlled by uniaxial ex-
tension, the use of anisotropic plasticity is not required [61]. Localisation of
deformation is accounted for by means of the cohesive law in (1.25).
A compact-tension specimen, tested by Li and Hashida [98], was anal-
ysed using the procedure depicted in Figure 1.4b with a cohesive discon-
tinuity activated at point a. The material was an ECC with 2% by vol-
ume of 12 mm long polyethylene bres. The geometry of the specimen
is depicted in Figure 2.17. In the experiment, the load was applied at the
loading pins via deformation control using a LVDT placed near the notch
tip. The following model parameters have been used for the analysis [89]:
Youngs modulus 22 GPa, Poissons ratio 0.2, matrix cracking strength
f
t
=
0
= 2.2 MPa (see (1.22)), and continuous-discontinuous transition
strain value
cd
= 5.78% with corresponding stress level f
cd
= 4.32 MPa
as depicted in Figure 2.18a. The above model parameters could have been
determined on the basis of the composite microstructure following the pro-
cedure described in Kanda et al. [92]. The cohesive discontinuity is extended

(b)
cd
f
t
f
cd
f
cd
w
max u
(a)
Figure 2.18 Stress-deformation relations for ECC: (a) contin-
uum and (b) discrete constitutive laws (dashed lines indicate
unloading behaviour).
0 10 20 30 40 50
Loadline displacement [mm]
0
experiment
simulation
2
4
6
8
10
L
o
a
d

[
k
N
]
Figure 2.19 Load-displacement curves for ECC compact-
tension specimen.
from a pre-existing horizontal traction-free discontinuity which represents
part of the notch and its direction is computed according to the procedure
described in Section 2.5 under Orienting a discontinuity.
Inelasticity in the continuum bulk was introduced by Rankine plasticity
(see (1.20) and Figure 2.18a), while the cohesive lawin (1.25) with t
max
= f
cd
,
w
max
= 0.5l
f
, l
f
= 12 mm and m = 1 was used to describe the inelas-
tic behaviour across the macrocrack (see Figure 2.18b). In the nite-element
simulations, an unstructured mesh of 2964 4node quadrilateral elements in
plane stress with an average element size in the central part of the specimen
equal to 4 mm was used.
The simulated behaviour compares reasonably well with the experimen-
tal data as depicted in Figure 2.19 and the softening effect due to the activa-
2.6 Applications 37
5 mm
50 mm 20 mm
10 mm
0.3 % 6 %
Figure 2.20 Multiple and localised cracking at different load-line displacements (the thick
line represents the discontinuity; traction-free and cohesive segments are represented in white
and black, respectively).
Figure 2.21 Close-up of the failure zone close to the notch.
(a) (b)
Figure 2.22 Experimental results for the failure of the ECC compact tension specimen:
(a) close-up of the cracked zone near the notch for load-line displacement equal to
21.12 mm and (b) detail of a crack bridged by bres (from Li and Wu [100] and Kabele
and Horii [88]).
tion of the discrete crack is fully captured by the model. Note that the high
value of the total deformation is due to the high density of multiple cracks
(visual inspection of the specimen indicated the presence of several cracks
per square centimetre [98]). The extent of multiple cracking can be estimated
from the equivalent plastic strain contour plots depicted in Figure 2.20 along
with cohesive and traction-free discontinuity segments. The reproduced lo-
cal behaviour is similar to the experimental one, reported in Figures 2.22a
and 2.23 [88, 100], with a large portion of the specimen undergoing multiple
cracking. The discontinuity propagates through solids elements as depicted
in Figure 2.21, where a close-up of the failure zone close to the notch is rep-
resented.
Figures 2.22a and 2.23 clearly motivate a continuous-discontinuous ap-
proach to failure according to the strategies described in Section 1.2 as a
proper approach to the description of failure processes in which a localised
crack stems from diffuse degradation.
2.7 Concluding remarks
PU-based discontinuous elements represent a promising alternative to con-
ventional interface elements in problems with evolving boundaries. Com-
pared to conventional interface elements, the major advantage is that mesh
adjustment along the propagating boundary is avoided since a discontinu-
2.7 Concluding remarks 39
Figure 2.23 Experimental results for the failure of the ECC compact tension speci-
men: multiple cracking and localised crack at failure (from Kabele and Horii [88]).
ity can pass arbitrarily through solid elements.
It is worth noting that the performance of PU-based discontinuous ele-
ments deteriorates when this class of elements is used with a dummy stiff-
ness approach, a key ingredient in many cohesive-zone models [185, 197]
(see also Appendix A where PU-based discontinuous elements are con-
trasted with conventional interface elements).
Chapter 3
Continuous-discontinuous failure in
gradient-enhanced media
The performance of some non-local models, in either integral [139] or dif-

ferential format [134], deteriorates in the nal stage of failure [64]. At com-
plete failure of a material point, which is understood as a discontinuity i.e. a
physical crack, numerical interaction between the two physically separated
parts of the body persists and causes an exchange of information between
the two sides of the discontinuity. This information, in the form of locally ex-
tremely high strain values, stimulates damage growth away from the pro-
cess zone, i.e. a spurious extension of the continuum damaged zone. This
phenomenon, most evident in problems with large crack opening, such as
e.g. in the failure of composite specimens [64], manifests itself also in prob-
lems with small crack opening [130].
In this chapter, the introduction of traction-free discontinuities in a differ-
ential version of a non-local damage model, the so-called implicit gradient-
enhanced continuum damage model [134], is presented. Discontinuities are
introduced when the material reaches a fully damaged state. The inclusion
of internal discontinuity surfaces, where boundary conditions are applied
without modications of the original nite-element mesh, avoids non-local
interactions across the crack. The unbounded strain on the surface has no
inuence on the non-local eld and the unrealistic damage growth typi-
cal of this class of regularised continuum models is avoided. After the in-
troduction of a discontinuity in the problem elds, coupling between fully
damaged locations and the surrounding material ceases and damage away
from the discontinuity tip is frozen; additional degradation is described by
the damaging material in the process zone, around the discontinuity tip.
As a discontinuity propagates, the non-local equivalent strain increases at
the discontinuity tip and ceases behind it since the non-local interaction be-
tween opposite sides of the discontinuity is avoided. This allows a realistic
description of macroscopic cracks evolving from a material with microc-
Based on References [175, 176]

42 Chapter 3 Continuous-discontinuous failure in gradient-enhanced media
racks. Numerical simulations illustrate the performance of the continuous-
discontinuous gradient-enhanced damage model in preventing the spurious
response normally observed prior to complete failure with the continuous
model alone.
3.1 Problem elds
In the gradient-enhanced damage continuum model proposed by Peerlings
et al. [134], the problem is characterised by the displacement eld u and by
the scalar non-local equivalent strain eld e (e is an extra degree of freedom
which is added to the standard displacement degrees of freedom). The de-
composition of the displacement eld u and of the strain eld into a con-
tinuous and a discontinuous part has been described in Section 2.1 (see (2.1)
and (2.2)). Similar to the strain eld in (2.2), the non-local equivalent strain
eld for the body

crossed by a discontinuity (depicted in Figure 2.1) can
be decomposed as
e(x, t) = e (x, t) +H
d
(x) e (x, t) if x /
d
, (3.1)
where e and e are continuous functions on

. In the domain of an element
where extra degrees of freedom are active, an approximation of the non-
local equivalent strain eld in (3.1) is given by
e
h
= N
e
p +H
d
N
e
q if x /
d
, (3.2)
where N
e
is a matrix containing usual nite-element shape functions and
the global nodal degrees of freedom p and q represent, in the arrangement
of (3.2), the total non-local equivalent strain eld. When the support of a
node is not crossed by a discontinuity, the standard nite-element interpola-
tion is retrieved for that node. Note that (3.1) can be derived by substitution
of the local strain eld from (2.2) into the non-local averaging in (1.9). The
decomposition in (3.1) is a direct consequence of the denition of the strain
eld (2.2).
3.2 Governing equations
The equilibrium equations and boundary conditions for the elastic body

,
recalled in (2.7), can be used in a damage continuum model by adopting the
constitutive relation
= (1 ) D
e
: in (3.3)
3.2 Governing equations 43
for an elasticity-based isotropic description of continuum damage. As
briey recalled in Section 1.4, in presence of softening constitutive be-
haviour, an approximate gradient-enhanced version of the above dam-
age continuum model can be conveniently employed to preserve well-
posedness of the governing equations. The approximate differential format
of the non-local averaging of the local equivalent strain e
l
in (1.9)-(1.10) re-
sults in the modied Helmholtz equation [134]
e c
2
e = e
l
in , (3.4)
for e , where c =
1
2
l
2
is the gradient parameter with l the length scale. Equa-
tion (3.4) is an extra equation which is added to the equilibrium equation.
The homogeneous natural boundary conditions
e n = 0 on (3.5)
complete the coupled system of equations.
From the decomposition of the non-local equivalent strain e in (3.1) and
using (3.5), the boundary conditions at the discontinuity surface
d
(see Fig-
ure 2.1) can be written as
e m = 0 on
d
(3.6)
( e + e) m = 0 on
+
d
. (3.7)
Since the function e is a continuous function, e
+
= e
, where e
+/
indicates
the value of e on
+/
d
. Therefore,
( e m)
+
d
= ( e m)
d
= 0. (3.8)
From (3.7) and (3.8), it follows that e m = 0 on
+
d
. In summary, the
boundary conditions for the non-local equivalent strain e at the discontinu-
ity surface are
e m = 0 on
+/
d
(3.9)
e m = 0 on
+
d
. (3.10)
3.2.1 Propagating discontinuities and boundary conditions
Together with the boundary conditions for the displacement elds, the
above boundary conditions allow the development of the process zone into
d
: = 1
: < < 0 1
0 : =
Figure 3.1 Damage distribution in a continuum (adapted from Peerlings [131]).
a macroscopic crack. Indeed, (3.4) is only valid on the domain
e

p
of
the body where 0 < 1 (see Figure 3.1). Since the process zone
p
and
the fully damaged domain
d
will increase during a computation, the nu-
merical procedure must consider the evolution of the internal boundary
d
.
The evolution of the fully damaged zone can be dealt with by modifying
the domain of the problem (using remeshing and adaptive techniques [8, 39]
or removing fully damaged elements as done by Peerlings [131, 133]). To
avoid singularities in the system due to = 1, exponential-like soften-
ing laws, such as e.g. (1.8), can be used. In this way, the damaging body
will never experience complete damage. This may sound appealing from
the numerical point of view in the sense that it is then possible to perform
computations in a continuous framework and consider as fully damaged
zones those parts of the body where damage is close to unity. However,
since in those parts of the body the local equivalent strain reaches extremely
high values, due to the non-local averaging either integral or differential
damage models transfer information from physically fully damaged zones
to partly damaged zones, resulting in spurious damage growth away from
the discontinuity tip [64]. With the continuous-discontinuous description of
failure, the incorporation of the standard boundary conditions for the prob-
lem elds where damage has reached its critical value is automatic and al-
lows the development of boundary surfaces. The application of the standard
boundary conditions to u and e on the internal boundary surfaces alters the
averaging procedure and implies that the non-local averaging for point A in
Figure 3.2b is limited to the shaded part only. Classical non-local models are
3.3 Variational formulation 45
Figure 3.2 Non-local averaging
for point A close to a numerical
macrocrack (highly damaged lo-
cations) in the darker area: (a) a
standard non-local model aver-
ages on and across the numeri-
cal macrocrack while (b) the non-
local averaging is limited to the
shaded part when a true discon-
tinuity is present (adapted from
Wells [192]).
(a) (b)
discontinuity
highly damaged zone
A A
unable to detect a numerical macrocrack (highly damaged locations) and
non-local averaging for point A is performed across and on the numerical
macrocrack (see Figure 3.2a) with consequent spurious damage growth.
3.3 Variational formulation
In this section, following standard procedures, the modied Helmholtz
equation will be cast in a weak form. The weak form of the equilibrium
equations has been derived in Section 2.3. The weak statements of the gov-
erning equations of the continuous-discontinuous gradient-enhanced dam-
age model will be coupled through the denition of the stress eld in Sec-
tion 3.4. To begin with, the space of trial non-local equivalent strains is de-
ned by the function e(x, t) = e(x, t) + H
d
e(x, t) with e and e H
1
()
while the space of admissible non-local equivalent strain variations is de-
ned by the weight function w
e
(x) = w
e
(x) + H
d
w
e
(x) with w
e
and w
e
H
1
(). Equation (3.4) can be cast in a variational form by multiplication
with a scalar weight function w
e
H
1
() (split into w
e
and w
e
) and by
integration over the domain . The weak form of the modied Helmholtz
equation reads
( w
e
+H
d
w
e
) ( e +H
d
e) dc
( w
e
+H
d
w
e
)
2
( e +H
d
e) d
=
( w
e
+H
d
w
e
) e
l
d w
e
H
1
(). (3.11)
From the decomposition of the problem elds it follows that any admissible
variation w
e
of e can be regarded as admissible variations w
e
and w
e
, thus
leading to two variational statements. Taking rst variations w
e
( w
e
= 0),
and then w
e
( w
e
= 0), leads to:
w
e
( e +H
d
e) dc
w
e
2
( e +H
d
e) d
=
w
e
e
l
d w
e
H
1
() (3.12a)
d
w
e
( e +H
d
e) dc
d
w
e
2
( e +H
d
e) d
=
d
w
e
e
l
d w
e
H
1
(). (3.12b)
Using the product rule for the Laplacian of a discontinuous scalar eld (see
Appendix B), the term
2
(H
d
) in the previous equations is expressed as
2
(H
d
) = H
2
+
d
m+ 2
d
m, (3.13)
where
d
is the Dirac-delta function centred at the discontinuity surface
d
. Substitution of the above relation into (3.12) leads to terms of the type
d
m d which can be expanded using the directional derivative of
a function in the direction of a generic unit vector v (D
v
= v):
d
v) d =
D
v
d
d =
d
D
v
d
=
d
v d . (3.14)
Note that (3.14) has been derived using the relation
d
) d =
d
d (3.15)
for the Dirac-delta function
d
(see Appendix B). Using Greens theorem
and after the application of the boundary conditions (3.5), (3.9) and (3.10),
the variational statements in (3.12) can be written as
w
e
e d+
+
w
e
e d+ c
w
e
e d+ c
+
w
e
e d
=
w
e
e
l
d w
e
H
1
() (3.16a)
+
w
e
e d+
+
w
e
e d+ c
+
w
e
e d+ c
+
w
e
e d
=
+
w
e
e
l
d w
e
H
1
(). (3.16b)
It is interesting to recall the weak statement of the continuous model
alone, as derived by Peerlings et al. [134]:
w
e
e d+ c
w
e
e d =
w
e
e
l
d w
e
H
1
(). (3.17)
Unlike the variational statement of the equilibrium counterpart of the cou-
pled problem (see Section 2.3), the coupling is already present in the varia-
tional statements in (3.16a) and (3.16b). However, collecting the continuous
and discontinuous part of the interpolation under the expression in (3.1)
allows for a comparison with the standard variational statement along
the lines of the previous comparison for the equilibrium equations in Sec-
tion 2.3.
3.4 Discretisation and linearisation
In this section the linearised form of the governing equations for a body
crossed by a discontinuity is developed. In what follows, the subscript h
will be dropped from discretised quantities and engineering notation for
and will be used.
3.4.1 Problem eld description
Using a Bubnov-Galerkin approach, (3.1) can be discretised in each element
affected by the enhancement using
e = N
e
p e = N
e
q e = B
e
p e = B
e
q (3.18)
w
e
= N
e
p
w
e
= N
e
q
w
e
= B
e
p
w
e
= B
e
q
(3.19)
for the non-local equivalent strain eld where p and q are standard and
extra dofs, respectively. A similar discretisation has been considered for the
displacement and strain elds in Section 2.4.
3.4.2 Discretised and linearised weak governing equations
Substitution of the discretised non-local equivalent strain into (3.16), leads
to two discrete weak governing equations:
N
T
e
N
e
p d+
+
N
T
e
N
e
q d+
B
T
e
cB
e
p d
+
+
B
T
e
cB
e
q d =
N
T
e
e
l
d (3.20a)
+
N
T
e
N
e
p d+
+
N
T
e
N
e
q d+
+
B
T
e
cB
e
p d
+
+
B
T
e
cB
e
q d =
+
N
T
e
e
l
d, (3.20b)
from which the equivalent nodal force vectors related to admissible varia-
tions of p and q result in
f
int, p
=
_
N
T
e
N
e
p +B
T
e
cB
e
p N
T
e
e
l
_
d
+
+
_
N
T
e
N
e
+B
T
e
cB
e
_
q d (3.21a)
f
int,q
=
+
_
N
T
e
N
e
p +B
T
e
cB
e
p N
T
e
e
l
_
d
+
+
_
N
T
e
N
e
+B
T
e
cB
e
_
q d. (3.21b)
It is interesting to note that the equivalent nodal force vectors take care of the
diffusive nature of the modied Helmholtz equation since they are not self-
equilibrated vectors within a nite element. The external part of the RHS is
empty due to the applied boundary conditions (f
ext, p
= f
ext,q
= 0). The equi-
librium counterpart of the above equations has been derived in Section 2.4.
To develop a consistent incremental-iterative full Newton-Raphson proce-
dure, the governing equations are linearised following standard procedures
(see e.g. References [64, 84]). At iteration i within a time step n, the discre-
tised coupled boundary value problem can be written in matrix format as:
_
_
K
n,i1
aa
K
n,i1
ab
K
n,i1
ap
K
n,i1
aq
K
n,i1
ba
K
n,i1
bb
K
n,i1
bp
K
n,i1
bq
K
n,i1
pa
K
n,i1
pb
K
n,i1
pp
K
n,i1
pq
K
n,i1
qa
K
n,i1
qb
K
n,i1
qp
K
n,i1
qq
_
_
_
_
a
n,i
b
n,i
p
n,i
q
n,i
_
_
=
_
_
f
n
ext,a
f
n
ext,b
0
0
_
_
f
n,i1
int,a
f
n,i1
int,b
f
n,i1
vint, p
f
n,i1
int,q
_
_
, (3.22)
with the symmetries K
ba
= K
ab
, K
bp
= K
bq
= K
aq
, K
qa
= K
qb
= K
pb
,
K
qp
= K
qq
= K
pq
, and
K
aa
=
B
T
u
(1 ) D
e
B
u
d (3.23a)
K
ab
=
+
B
T
u
(1 ) D
e
B
u
d (3.23b)
K
ap
=
B
T
u
_
_ _
e
_
D
e
N
e
d (3.23c)
K
aq
=
+
B
T
u
_
_ _
e
_
D
e
N
e
d (3.23d)
K
bb
=
+
B
T
u
(1 ) D
e
B
u
d+
d
N
T
u
TN
u
d (3.23e)
K
pa
=
N
T
e
_
e
l
_
T
B
u
d (3.23f)
K
pb
=
+
N
T
e
_
e
l
_
T
B
u
d (3.23g)
K
pp
=
_
N
T
e
N
e
+B
T
e
cB
e
_
d (3.23h)
K
pq
=
+
_
N
T
e
N
e
+B
T
e
cB
e
_
d, (3.23i)
u
discontinuous continuous
Figure 3.3 From continuous to discontinuous displace-

ment/strain proles as a consequence of strain localisation
(adapted from Jir asek [83]).
where D
e
is the linear-elastic constitutive matrix. As in the standard
gradient-enhanced damage formulation, the stiffness matrix is not symmet-
ric. The terms in the RHS of the discretised boundary value problem are
dened in (2.21) and (3.21).
The nite-element implementation for the continuum response mainly
follows the one proposed by Peerlings et al. [134] for the gradient-
enhanced model. The proposed model has been implemented with linear
and quadratic quadrilateral elements. A discussion of the performance of
linear and quadratic elements in the gradient-enhanced damage model can
be found in Appendix D. Next, some issues pertinent to the element tech-
nology for the continuous-discontinuous gradient-enhanced model will be
discussed. The integration scheme has been described in Section 2.5.
Introducing a discontinuity. In a damaging continuum a critical situation
can be dened as one which corresponds to damage values close, or equal,
to unity in the element ahead of a discontinuity tip. The introduction of a
discontinuity at (almost) total loss of load-carrying capacity is in line with
the narrowing of the strain prole as a consequence of strain localisation
d
i
2 /
r
i
discontinuity tip
Figure 3.4 Determination of the propagation direction.
(see Figure 3.3). Introducing a discontinuity before complete localisation,
e.g. when damage is around 90%, poses the problem of dening the cor-
rect energy dissipation for the regularised continuous-discontinuous model
which is problematic, even in a one-dimensional setting. In the applications
in Sections 3.6.1 and 3.6.2, a traction-free discontinuity is inserted into an
element when the damage at all its integration points is larger than a crit-
ical value
crit
. The insertion of a discontinuity with damage larger than
crit
at one integration point is considered in Section 3.6.3 and in Chapter 4.
Simulations performed with both criteria ( in one integration point versus
in all integration points larger than
crit
) on the same test gave identical
results upon mesh renement.
Orienting a discontinuity. In a regularised continuum, the direction of a
discontinuity cannot be analytically derived from bifurcation analyses and
phenomenologically based criteria must be used. In this context, since cri-
teria based on the stress tensor cannot be reliably used [193], the direction
of maximum accumulation of the non-local equivalent strain in a V-shaped
window ahead of a discontinuity tip is used (the V-shaped window spans a
circular sector with = 90
, see Figure 3.4). This criterion can be justied by

the observation of numerically computed non-local equivalent strain pro-
les and by the experimental strain elds shown by Geers et al. [68]. The
vector in the direction of the discontinuity propagation is computed as
d
d
=
iS
e
i
V
i
w
i
d
i
|d
i
|
, (3.24)
where S is the set of integration points i in the V-shaped window ahead of a
discontinuity tip, e
i
is the non-local equivalent strain at integration point i,
V
i
and w
i
are the volume and the weight associated with integration point
i, and d
i
is the vector in the direction of the integration point i [194]. The
weights w
i
are computed using the Gaussian weight function in (2.26) with
interaction radius r equal to four times the length scale of the gradient-
enhanced damage model. When the discontinuity is close to a boundary, the
discontinuity extension direction is aligned with the previous discontinuity
segment.
Transfer of history data. The transfer of history data within an element
is performed considering the maximum value of within the element as
representative of the state of the entire element (see also Section 2.5 under
Transfer of history data). Smoothing techniques, as e.g. the one employed by
Wells et al. [194], could produce smaller values for the maximum value of
the non-local equivalent strain e driving damage evolution, violating thus
the Kuhn-Tucker conditions in (1.11). The error introduced diminishes upon
mesh renement. The increment which led to the critical damage value is
recomputed after the transfer is performed.
3.6 Applications
In this section, the continuous-discontinuous approach to failure is ap-
plied to quasi-brittle failure in concrete and composite specimens. A full
Newton-Raphson procedure has been used to trace the response in the non-
linear regime. Although the algorithm was consistently derived to obtain
quadratic convergence, the performance of the model close to failure and
before the extension of a discontinuity in a critically damaged area deterio-
rated (as 1). However, after the extension of a discontinuity, quadratic
convergence was retrieved. All the applications are analysed following the
procedure described in Figure 1.4a on page 5 with a traction-free disconti-
nuity (i.e. with t = 0 in (1.24) and (2.25f) and T = 0 in (3.23e)) activated
at point bthe use of cohesive discontinuities is not feasible because of the
problems discussed in Section 3.8.
The rst application (concrete beam in four-point bending) illustrates that
the problem of spurious damage growth close to failure can be avoided
by using the proposed continuous-discontinuous strategy. This application
also illustrates that softening laws with high residual stress induce an in-
correct energy dissipation. In the second application (composite compact-
tension test), the continuous and the continuous-discontinuous approaches
3.6 Applications 53
150 150
500
150
d
100
1/2 P 1/2 P
Figure 3.5 Four-point bending test (thickness = 50 mm; all dimensions in mm).
are properly compared and the results highlight the effect of a constant
length scale (see also Section 3.8). The last application (single-edge notched
beam in anti-symmetric four-point-shear loading) illustrates the quality of
the criterion for the determination of the direction of discontinuity propa-
gation.
3.6.1 Concrete beam in four-point bending
A four-point bending test of a concrete beam with different notch sizes d
is analysed (see Figure 3.5). To enable comparison with the experiments re-
ported in Reference [78], the vertical displacement of a point placed at the
bottom of the beam and with an offset of 7.5 mm from the centreline of the
beam is used for the measurement of the deection v. The following model
parameters, tted for the continuous problem, are adopted for the simu-
lation [130]: Youngs modulus E = 40000 MPa; Poissons ratio = 0.2;
exponential damage evolution law (1.8) with
0
= 0.000075, = 0.92
and = 300; modied von Mises denition of the local equivalent strain
(1.4) with k = 10 and gradient parameter c = 4 mm
2
. The simulation is
performed under plane stress conditions. The load is applied via an im-
posed displacement and quadrilateral elements with quadratic interpola-
tion for the displacement and bilinear interpolation for the non-local equiv-
alent strain have been used (see Appendix D). The notch is simulated as a
traction-free discontinuity (t = 0 in (1.24)). The traction-free discontinuity
is extended, starting from the tip at the notch, when the damage at all in-
0 0.1 0.2 0.3 0.4 0 0.5 0.1 0.6 0.2
50
30
10
0.7 0.3 0.4 0.5 0.6 0.7
0
1
2
3
4
5
0
1
2
3
4
5
P

[
k
N
]
P

[
k
N
]
continuousdiscontinuous
continuous experiment
continuous
v [mm]
(a)
v [mm]
(b)
h=10, 5, 2.5 mm
Figure 3.6 Load-deection curves for simulations with (a) different meshes for a 10 mm
deep notch (h is the element size in the central part of the beam) and (b) different notch size
d for the medium element size mesh (h = 5 mm) and experimental results [78].
notch
continuous failure
continuousdiscontinuous failure
Figure 3.7 Four-point bending test: non-local equivalent strain evolution (plotted in the
same scale for the 10 mm deep notch beam with medium element size mesh; close-up of the
central part of the beam).
tegration points in the element ahead of the discontinuity tip is larger than
a critical value, set to
crit
= 0.999; the discontinuity is prescribed to be
vertical (consequently, (3.24) is not used here).
3.6 Applications 55
Results of the analyses for the 10 mmdeep notch beamare reported in Fig-
ure 3.6a for different meshes. In the central part of the mesh, the coarse,
medium and ne mesh element size h is 10 mm, 5 mm and 2.5 mm, respec-
tively. Due to the small difference in the response between the medium and
the ne discretisation (see Figure 3.6a), the former, with a 10 mmdeep notch,
has been used for the results depicted in Figures 3.7 and 3.8. From the anal-
ysis, it is evident that the introduction of a discontinuity during the com-
putation inuences the global (load-displacement curve) and local (dam-
age and non-local equivalent strain proles) behaviour. In particular, Fig-
ure 3.7 shows that the activity of the non-local equivalent strain is mobilised
only around the discontinuity tip for the continuum-discontinuous model.
This translates into the more realistic damage proles depicted in Figure 3.8
where, in contrast to the continuous model alone, the width of the damage
zone at the bottom of the specimen does not increase with the continuous-
discontinuous failure model. However, the use of an exponential softening
relationship with a high residual stress at the moment of the enhancement
causes the marked drops reported in Figure 3.6a and makes the comparison
1
0.1
continuousdiscontinuous failure
continuous failure
Figure 3.8 Four-point bending test: damage evolution (the discontinuity is represented by
the white thick line; 10 mm deep notch beam with medium element size mesh; close-up of
the central part of the beam).
thickness 3.8 mm
W/50
0.25 W
0.275 W
0.325 W
0.25 W
1.20 W
1.25 W
d
Figure 3.9 Compact-tension specimen (W = 50 mm, d = 10 mm).
with the experiment, reported in Figure 3.6b, difcult (this issue is discussed
in Section 3.7). Responses of computations with higher values of
crit
show
a better agreement at the global level with experimental results but this is
due to the delayed or, in some cases, precluded extension of a discontinu-
ity. The use of different damage law parameters to achieve lower residual
stress values in the softening relationship produced an unsatisfactory com-
parison in the post-peak response (note that this consideration relates to the
continuum model alone since the discontinuity is extended at a later stage
as depicted in Figure 3.6). Introduction of displacement discontinuities re-
quires a re-assessment of the continuum model parameters governing the
post-peak response, which has not been done.
3.6.2 Composite compact-tension specimen
The compact-tension specimen depicted in Figure 3.9 [10], experimentally
tested by Geers et al. [68] and numerically analysed by Geers et al. [65],
Peerlings [131] and de Borst et al. [34], has been numerically investigated
with the continuous-discontinuous approach. The specimen is placed on
two loading pins whose action has been modelled by applying two verti-
3.6 Applications 57
Figure 3.10 Load-cmod diagrams for the
compact-tension specimen for different mesh
resolutions (h indicates the average element
size in the central part of the specimen).
0 1 2 3 4 5
cmod [mm]
0
500
1000
1500
2000
F

[
N
]
experiment
h=2, 1, 0.5 mm
cal forces at the uppermost and lowermost node of the pinholes via defor-
mation control. In the simulations, indirect displacement control has been
used, with the displacement (crack mouth opening displacement, cmod)
measured between two markers placed 25 mmfromthe left edge and 14 mm
from the symmetry axis of the specimen [34]. The vertical displacement of
the mid-side point on the right central part of the specimen has been re-
strained, as well as the horizontal displacement of the right lowermost and
uppermost corners. A small part of the notch is simulated as a traction-free
discontinuity. The traction-free discontinuity is extended when the dam-
age at all integration points in the element ahead of the discontinuity tip
is larger than
crit
= 0.99995, and its direction is computed according to
the direction of maximum non-local equivalent strain as described in Sec-
tion 3.5 under Orienting a discontinuity. The simulations are performed using
unstructured meshes of bilinear quadrilateral elements under plane stress
conditions with average element sizes h in the central part of the specimen
equal to 2, 1 and 0.5 mm with 443, 1209 and 3955 elements, respectively.
The model parameters have been adopted from Geers et al. [65]: damage
growth is expressed via the power law in (1.7) with
0
= 0.011,
c
= 0.5,
= 5 and = 0.75; the equivalent strain denition is based on the positive
principal strain components (see (1.3)); Youngs modulus E = 3200 MPa;
Poissons ratio = 0.28 and gradient parameter c = 2 mm
2
. To avoid dam-
age growth in the elements around the pinholes, a higher value of
0
has
been given to elements in these areas. Furthermore, an almost horizontal
branch was added to the softening law at = 0.385
c
, which corresponds,
in an ideal one-dimensional uniform tension test, to an almost nil residual
stress (0.175%
0
E; see Section 3.8).
The load-cmod response is shown in Figure 3.10. The agreement between
the experimental response and the results of the simulations is excellent.
cmod = 2.00 mm
cmod = 2.75 mm
cmod = 5.00 mm
(a) (b)
e
0.1 1 0.1 0.5
Figure 3.11 Evolution of the failure process: (a) damage and (b) non-local equivalent
strain e contour plots.
3.6 Applications 59
cmod = 5 mm
(a) (b)
e
0.1 1 0.1 0.5
Figure 3.12 Final failure pattern with the continuous model (no propagating discontinuity):
(a) damage and (b) non-local equivalent strain e contour plots.
Figure 3.13 Load-cmod diagrams
for the compact-tension specimen
(h = 0.5 mm): comparison between
the continuous and the continuous-
discontinuous model.
0 1 2 3 4 5
cmod [mm]
0
500
1000
1500
2000
F

[
N
]
continuous
However, the response computed with the ner mesh (h = 0.5 mm) shows
bumps after the extension of the discontinuity (this issue is discussed in
detail in Section 3.8). The evolution of the fracture process is shown in Fig-
ure 3.11 for the simulation related to the nest of the meshes used (h =
0.5 mm). For a comparison, the local and global response of the standard
continuous model (no propagating discontinuity) and of the continuous-
discontinuous (with a propagating discontinuity) model are shown in Fig-
ure 3.12 and Figure 3.13, respectively. Although the differences in the global
response are not signicant (see Figure 3.13), the continuous-discontinuous
1
0
0
2
0
5
20 20
20 20
10/11P 1/11P
440
Figure 3.14 Single-edge notched beam [161] (depth = 100 mm; all dimensions in mm).
model provides a better representation of the failure process (compare Fig-
ure 3.11 to Figure 3.12).
3.6.3 Single-edge notched beam in anti-symmetric four-point-shear
loading
A single-edge notched beam, depicted in Figure 3.14, is subjected to an anti-
symmetric four-point-shear loading [161]. Figure. 3.14 shows the applied
boundary conditions which result in a curved crack path, from the lower
right part of the notch towards a point to the right of the lower-right sup-
port as depicted in Figure 3.15. The boundary conditions are specied by
constraining the displacement at the upper-right support in both directions
and by constraining the vertical displacement at the upper-left support. The
supports have widths of 20 mm with the centre located 20 mm out of the
midspan of the beam.
The beamis analysed in a plane stress situation with bilinear quadrilateral
elements; the simulations are performed using unstructured meshes with
average element size h in the central part of beam equal to 4, 2 and 1 mm
with 674, 2148 and 7308 elements, respectively. The model parameters are
Figure 3.15 Experimental crack patterns from three single-edge
notched beams (adapted from Schlangen [161]).
3.6 Applications 61
0 0.05 0.1 0.15 0.2
cmsd [mm]
0
10
20
30
40
50
P

[
k
N
]
continuous
h=4, 2, 1 mm
(a)
0 0.05 0.1 0.15 0.2
cmsd [mm]
0
10
20
30
40
50
P

[
k
N
]
continuous
experiment
(b)
Figure 3.16 Applied load against crack mouth sliding displacement (cmsd): (a) convergence
studies and (b) comparison with the experimental response [161] for the h = 1 mm mesh (h
indicates the average element size in the central part of the beam).
adopted unaltered from Peerlings et al. [132] who performed analyses in
a purely continuous setting: Youngs modulus E = 35000 MPa, Poissons
ratio = 0.2, gradient parameter c = 1 mm
2
, modied von Mises equiv-
alent strain denition (1.4) with k = 10, exponential softening law (1.8)
with
0
= 0.00006, = 0.96 and = 100. The loading platens have a
Youngs modulus one order of magnitude larger than the concrete. The load
is applied by means of an indirect displacement control procedure with the
crack mouth sliding displacement (cmsd), dened as the relative vertical
displacement of the opposite faces of the notch, taken as control parame-
ter [131]. The traction-free discontinuity is extended when damage is larger
that
crit
= 0.99 at one integration point in the element ahead of the discon-
tinuity tip. It starts from an initial horizontal traction-free discontinuity (the
horizontal traction-free discontinuity goes from (222.5; 80.1) to (223; 80.1),
with the origin of the coordinate system placed at the lowermost left corner
of the beam). As in the previous example, the direction of the discontinu-
ity is aligned with the direction of the maximum accumulation of non-local
equivalent strain.
Similar to the case of the four-point bending test reported in Section 3.6.1,
the global response of the continuous-discontinuous model, reported in Fig-
ure 3.16 in terms of load-crack mouth sliding displacement, is more brit-
tle than that of the continuous model. This makes the comparison with
the experiment difcult (see Figure 3.16b). This issue is discussed in the
(a)
(b)
(c)
0.1 1
Figure 3.17 Final damage distribution and traction-free discontinuity for the (a) h = 4 mm,
(b) h = 2 mm and (c) h = 1 mm meshes (close-up of the central part of the beam).
3.6 Applications 63
Figure 3.18 Close-up of the nal damage distribution and traction-
free discontinuity for the h = 1 mm mesh near the notch.
(a)
(b)
0.1 1
Figure 3.19 Final failure state for (a) the continuous-discontinuous

model and (b) the continuous model (no propagating discontinuity)
for the h = 1 mm mesh (close-up of the central part of the beam).
next section. However, it is worth noting that the convergence of the iter-
ative scheme for the continuous-discontinuous approach is faster (see Fig-
ure 3.16a). The nal failure pattern is reported in Figures 3.17 to 3.19 to-
gether with a comparison of failure patterns between the continuous and
the continuous-discontinuous model. In particular, Figure 3.17 shows that
the evolution of damage is in good agreement with the observed crack pat-
tern reported in Figure 3.15 and that the direction of the discontinuity is
correctly determined even with the coarsest of the discretisations (see Fig-
ure 3.17a). A close-up of the nal damage distribution and traction-free
discontinuity for the h = 1 mm mesh is shown in Figure 3.18, and the
continuous-discontinuous model and the continuous model are contrasted
in Figure 3.19 where it is clear that the introduction of the discontinuity
avoids the unrealistic damage growth close to the notch (see Figure 3.19a).
3.7 Stress-strain relationships
Enhancing an element with a discontinuous interpolation of the problem
elds when the damage at all its integration points is larger than
crit
=
0.999 or larger than
crit
= 0.9999 may result in signicant differences
in terms of the global response. To understand the differences originat-
ing from the choice of stress-strain relationships, the normalised softening
stress-strain paths for a one-dimensional uniform loading state eld are
contrasted. The model parameters of the four-point bending test and the
compact-tension test previously analysed are adopted. The results of this
analysis are shown in Figure 3.20. For the exponential law (1.8) (see Fig-
ure 3.20a), an increment of less than 1% in damage requires an increment
of approximately 512% in deformation which corresponds to a drop in the
normalised stress of about 39%; for the power law (1.7) (see Figure 3.20b),
only an increase of 23% in the deformation is necessary to drop the stress of
about 88% for the same increment of damage.
The drops in the global response reported in Figures 3.6 and 3.16 are due
to an incorrect energy dissipation related to the high residual stress at the
moment of the enhancement of the displacement elddespite high dam-
age ( 1), high stresses can still develop. Although the asymptote of the
exponential softening curve in Figure 3.20a might be useful in reproducing
the long tail observed in load-displacement diagrams of concrete specimens
which is related to crack bridging [78, 108], the physical relevance of these
laws is questionable since an analogous and more pronounced phenomenon
3.8 Damage initiation and discontinuities 65
0 0.1 0.2 0.3 0.4 0 0.5 0.01
deformation
0.02 0.03 0.04 0.05 0.06
deformation
(a) (b)
0
0.2
0.4
0.6
0.8
1
n
o
r
m
a
l
i
s
e
d

s
t
r
e
s
s
0
0.2
0.4
0.6
0.8
1
n
o
r
m
a
l
i
s
e
d

s
t
r
e
s
s
=0.9999
=0.08
=0.06
=0.999
=0.02551
=0.2985
=0.9999
=0.0031
=0.36875
=0.999
=0.12974
=0.0098
Figure 3.20 Normalised stress-strain softening curves for (a) exponential and (b) power law
of damage growth.
occurs in bre-reinforced composite polymers where softening laws with
full stress relaxation at nite strainor softening laws with a small residual
stresssufce.
3.8 Damage initiation and discontinuities
A different phenomenon can be observed in the global response of the com-
pact tension test depicted in Figure 3.10. In the simulations of the compact-
tension test with the nest of the meshes, the load-displacement curve
showed a bump after the extension of a discontinuity. To better understand
the cause of this problem, it is important to realise that, as a result of non-
local averaging, the non-local equivalent strain differs quantitatively and
qualitatively from its local counterpart. With a strongly concave boundary,
as in the case of a sharp notch, non-local regularisation results in a nite
value of the damage driving quantity at the crack tip (see Appendix E),
unlike in the case of the standard continuum where the equivalent strain
mimics the singularity of the strain eld. However, in contrast to a standard
continuum, the maximum of the damage driving quantity is not located at
the tip, but shifted away from it. Simple analytical considerations reported
in Appendix E show that, for a planar crack, the position of the maximum
of the non-local equivalent strain is a function of the length scale (i.e. the
maximum of the non-local equivalent strain is at the crack tip for a length
scale equal to zero). As a consequence, damage reaches its maximum inside
the specimen, away from the tip. Note that although the condition for the
x
y
(a) (b)
60 mm
4 mm
12 mm
60 mm
Figure 3.21 Compact-tension test: (a) geometry (thickness = 1 mm) and (b) discreti-
sation (traction-free discontinuity from x = 12 mm to x = 16 mm at y = 30 mm).
propagation of a discontinuity might be satised in the specimen, it is not
necessarily at the crack tip. Eventually, due to increasing loading, the por-
tion of the specimen where damage has its maximum enlarges and reaches
the crack tip, thus allowing for the propagation of the discontinuity. Further,
for computational reasons, since damage can reach values larger than
crit
away from the discontinuity tip, the use of exponential-like softening laws
(or any other law with full stress relaxation at innite strain) is necessary to
avoid singularity in the global stiffness matrix.
The discontinuity propagates through more that one element at a time
and, as a consequence, is accompanied by strong variations of the local
equivalent strain which in turn affect its non-local counterpart. It is shown
later that this inuence is proportional to the length scale. This affects the
loading condition at points ahead of the new discontinuity tip which now
experience unloading and, due to an increased loading level, generates a
structural pseudo-elastic loading, resulting in an incorrect global response.
Numerical considerations. To gain more insight into the nature of the
problem, a simpler compact-tension test geometry, depicted in Figure 3.21a,
has been considered. To reproduce the effect of the discretisation on non-
local averaging, different values of c have been used, keeping the mesh un-
changed (structured mesh with 4 mm4 mm linear quadrilateral elements;
element size e
size
= 4
2 mm; see Figure 3.21b). The load is applied at the

uppermost and lowermost left corners (x = 0 mm, y = 0 mm / y = 60 mm)
via an applied displacement. The uppermost and lowermost right corners
3.8 Damage initiation and discontinuities 67
Figure 3.22 Load-displacement curves for in-
creasing non-locality parameter l (the dots rep-
resent the extension of the discontinuity).
0 5 10
displacement [mm]
0
100
200
300
400
500
r
e
a
c
t
i
o
n

[
N
]
0 5 10
0
100
200
300
400
500
l=4, 6, 8 mm
are restrained in the horizontal direction while the nodes at x = 60 mm,
y = 2 mm are restrained in the vertical direction. To avoid damage where
the load is applied, a higher value of
0
(see (1.7)) has been given to the
material in the area around the uppermost and lowermost left corners. The
model parameters used in the previous simulation of the compact-tension
test (see Section 3.6.2) with = 0 and = 1 have been considered. A hor-
izontal traction-free discontinuity is placed in the shaded element depicted
in Figure 3.21b. The traction-free discontinuity is extended horizontally in
the neighbouring element(s) if at all its(their) integration points damage is
larger than
crit
= 0.99999. The inuence of the non-local averaging is il-
lustrated in Figure 3.22, where load-displacement curves related to the ver-
tical uppermost restrained node for different values of the length scale have
been plotted. The values used for c are 8 mm
2
, 18 mm
2
and 32 mm
2
which
correspond to a length scale l equal to 4 mm, 6 mm and 8 mm. Compared
to the element size e
size
, only the last one permits sufcient non-local in-
teraction. However, no estimate of the necessary/sufcient number of el-
ements in the process zone is available for differential non-local models
and the ratio e
size
/l is often taken very close to unity and sometimes even
larger [34, 66, 93, 130, 132]. From the curves in Figure 3.22 it is evident that
an increasing non-local interaction leads to bumps in the load-displacement
response after the extension of the discontinuity (the extension is marked by
the dots). The bumps in Figure 3.22 originate fromthe loading situation after
the extension of the discontinuity in the elements ahead of the discontinuity
tip. This phenomenon can be understood as the combination of two effects:
(i) the sudden extension of the discontinuity through several elements due
to the shift of the maximum of the non-local equivalent strain away from the
crack tip and (ii) the inuence of non-local interaction on the unloading of
points adjacent to the discontinuity, with the inuence proportional to the
0
tip new tip tip new tip
0 10 10 20 20 30 30 40 40 50 50 60 60
e
(b)
x
y=+32
(a)
x
y=+32
e
Figure 3.23 Non-local equivalent strain prole before (solid line) and after (dotted line) the
extension of the discontinuity for (a) l = 4 mm and (b) l = 8 mm (not to scale).
degree of non-local interaction.
The non-local equivalent strain proles depicted in Figure 3.23, for differ-
ent degrees of non-local interaction before and after the rst extension of the
discontinuity, illustrate the problem with the loading condition discussed
above. Due to the denition of the loading function (see (1.11)), when non-
local interaction is strong (see Figure 3.23b) all the integration points ahead
of the discontinuity tip will temporarily unload after its extension. Despite
the high damage value in these integration points, they contribute to the
RHS through residual stresses resulting in spurious loading at the global
level. If the non-local interaction is weak (see Figure 3.23a), the difference in
the proles of the non-local equivalent strain ahead of the discontinuity tip
is small. The integration points between the new discontinuity tip and the
point where the non-local equivalent strain exhibits its maximum value will
unload and, since the unloading is limited, there is no evidence at the global
level. The inuence of the length scale on the evolution of the solution ar-
ray (u and e) is depicted in Figure 3.24. The extension of the discontinuity,
marked by the dots, is accompanied by strong variations of the non-local
equivalent strain eld (see section 4.4 for similar analyses performed with a
rate-dependent model). Note that this phenomenon is related to the resolu-
tion of the mesh with respect to the length scale as depicted in Figure 3.24:
if a coarse mesh is used it is likely that bumps in the load-displacement
response will not appear (see Figure 3.6). It is also stressed that the use of
cohesive discontinuities do not solve nor alleviate the problem of the bumps
which is only related to the non-local nature of the underlying continuum.
Finally, the use of a model with a variable length scale, for which the length
pseudotime
(a) (b)
0
0.2
0.4
0.6
0.8
u
y
u
y
|u |
x
|u |
x
1
e
e
s
c
a
l
e
d

s
o
l
u
t
i
o
n

a
r
r
a
y
pseudotime
0
0.2
0.4
0.6
0.8
1
s
c
a
l
e
d

s
o
l
u
t
i
o
n

a
r
r
a
y
Figure 3.24 Evolution of u and e for a point located at x = 32 mm, y = 32 mm for (a) l =
4 mm and (b) l = 8 mm (the dots represent the extension of the discontinuity and the dotted
line indicates the rst moment at which the point is behind the discontinuity tip).
scale goes to zero close to failure (see e.g. Reference [64]), might solve the
problem of the unloading after the extension of a discontinuity, thus reduc-
ing or eliminating the bumps. However, the continuous model will still be
unable to predict damage initiation correctly as illustrated in Appendix E.
When discontinuities are allowed in an implicit gradient-enhanced contin-
uum damage model, spurious growth of damage can be prevented. The
model successfully eliminates the interaction between fully and partially
damaged material locations. A continuous model with the same goal was
introduced by Geers et al. [64] but, unlike that approach, in which a variable
length scale was used for the gradual transition from a regularised to a stan-
dard continuum, in the continuous-discontinuous model considered here
only minimal assumptions (i.e. (2.1) and (3.1)) are necessary to introduce a
true discontinuity. Although in a different context, the transition from a reg-
ularised continuum model to discrete cracking was analysed by Jir asek and
Zimmermann [86] using embedded discontinuities.
The model presented in this chapter eliminates the problem of spurious
damage growth but some properties of the underlying non-local continuum
model, illustrated in Appendix E, makes the transition from continuous to
continuous-discontinuous failure problematic. In the next chapter, a differ-
ent regularisation technique, based on rate-dependence of the governing
equations, will be employed.
Chapter 4
Continuous-discontinuous failure in rate-dependent
media
A displacement discontinuity can be considered as originating either from

complete material failure in quasi-brittle materials, representing genuine
material separation, or from failure of the cementitious matrix in some
steel bre-reinforced concrete (SFR-concrete), representing crack bridging.
In both cases, prior to the development of a displacement discontinuity, the
material undergoes a phase of degradation which can be described by a con-
tinuous approach. In this chapter, degradation in the continuumis described
by means of a rate-dependent elastoplastic-damage model in which rate-
dependency is considered in the framework of Perzyna viscoplasticity. The
coupling to damage is crucial to the subsequent introduction of a disconti-
nuity as it (i) allows the necessary narrowing of the width of the localisation
zone (converging to a discrete surface) which preludes the discontinuity in
the displacement eld and (ii) allows full stress relaxation at the integration
point level response (see Section 1.3). This is an important aspect of the cou-
pling to damage since in standard viscoplastic models [54, 137] full stress
relaxation is difcult to obtain due to the viscous contribution to the stress
eld [179]. In describing the behaviour of SFR-concrete, the discontinuity
is given cohesive forces to describe the crack bridging effect of the bres.
Further, the use of a coupled elastoplastic-damage model ensures a more
realistic representation of the behaviour of quasi-brittle materials since the
assumption of unloading to the origin in a damage model is simplistic [17]
and does not reect experimental evidence [108].
An approach similar to the one presented in this chapter has been con-
sidered by Wells et al. [194] who used traction-free discontinuities in a stan-
dard Perzyna viscoplastic model. However, as already mentioned and as
will be illustrated next, the use of such a model does not represent a real-
istic physical situation since a traction-free discontinuity is considered at a
stage in which the material (i.e. the integration point in a nite-element con-
Based on References [172, 173, 177]

72 Chapter 4 Continuous-discontinuous failure in rate-dependent media
text) still has some residual stress. Note that in standard viscoplasticity it is
problematic to control the residual stress in the strain-softening constitutive
relationship when the localisation process is completed. If the simulation is
controlled in displacements, a constant residual stress is usually obtained
with constant stepping; conversely, in load control situations, it is even pos-
sible to obtain hardening effects. With the coupling to damage, as proposed
next, it is always possible to control the residual stress level.
After the detailed description of the continuum constitutive model, some
aspects of the nite-element technology will be discussed (the kinematics
of the continuous-discontinuous model used in this chapter has been de-
scribed in Chapter 2). Three applications in failure mechanics problems il-
lustrate the performance of the approach. The case of inserting discontinu-
ities in a differential version of a non-local continuum description [134] (as
described in Chapter 3) is qualitatively compared to the present approach.
In Section 4.4 it is illustrated that the nature of the regularisation technique
plays a major role in a continuous-discontinuous approach.
4.1 Rate-dependent elastoplastic-damage models
In small strain viscoplasticity, the strain tensor is decomposed into an elas-
tic
e
and a viscoplastic
vp
component according to (1.23). Consequently,
the elastic stress-strain relationship (1.15)
1
yields
= D
e
: (
vp
) . (4.1)
The viscoplastic ow rule will be specied later for Perzyna and Duvaut-
Lions viscoplasticity modelsa general overview of the two viscoplasticity
formats can be found in Jir asek and Ba zant [84] while a detailed comparison
can be found in Sluys [179] and Runesson et al. [157]. The coupling of iso-
tropic damage and plasticity is introduced by adopting the effective stress
concept and the strain equivalence principle [96]. In such a framework, a
simple and quite general algorithmic formulation, based on the operator
splitting technique [42, 163, 164], can be derived. An alternative algorithmic
formulation is based on the direct linearisation of all the relevant quantities
as will be specied next. The two formulations are equivalent [42] and the
latter approach will be used for a Duvaut-Lionstype model while the for-
mer for a Perzyna model. In the following derivations, the quantity refers
to a rate-dependent homogenised quantity (like the quantity with the super-
script
vp
),

to a rate-dependent effective quantity,

to a quasi-static (rate-
4.1 Rate-dependent elastoplastic-damage models 73
independent) effective quantity and

to a quasi-static (rate-independent)
homogenised quantity.
The algorithmic procedure for the coupled model hinges on the knowl-
edge of the stress tensor, the algorithmic tangent moduli and the equivalent
plastic strain, all dened in the effective space. Key point for the following
derivations is the expression of the rate-dependent effective stress tensor
as a function of (i) the rate dependent homogenised stress tensor , (ii) of the
elastic strain tensor
e
, or (iii) of the total strain tensor and the viscoplastic
strain tensor
vp
:
=

1
= D
e
:
e
= D
e
: (
vp
) , (4.2)
where is a scalar valued damage variable (0 1). To preserve well-
posedness of the governing equations when softening constitutive relation-
ships are used, damage evolution must be postulated as some function of
a regularised monotonically increasing deformation history invariant . In
this chapter, damage evolution is postulated as
( ) =
_
0 if
0
(1 exp ( )) if >
0
,
(4.3)
with and model parameters and
0
the threshold of damage initiation
(note that damage is plastically induced).
4.1.1 Duvaut-Lionstype model
In the viscoplastic model proposed by Simo et al. [165]this model is
a modied version of the model originally proposed by Duvaut and Li-
ons [54]the viscoplastic ow rule is written in the form

vp
=
1
C
e
: ( ) , (4.4)
where is the relaxation time, C
e
= (D
e
)
1
is the fourth-order elastic com-
pliance tensor and indicates the projection of on the static yield surface,
i.e. the rate-independent stress. The rate of the hardening parameter is
dened in a similar fashion as
=
1
( ) . (4.5)
A backward Euler time integration scheme gives
n+1
=

+t

n
+
t
+t

n+1
. (4.6)
For the next derivations, it is useful to recall the expressions [116, 198],
for the rate-independent case, of the differential of the incremental plastic
multiplier
d (
) =

f
R:d ()
R:

f
, (4.7)
with d (
) = d
, and of the fourth-order consistent tangent tensor
D
p
=

R
R:

f
R:

f
(4.8)
where
R =
_
I +
D
e

f
_
1
D
e
, (4.9)
=

f / ,

f
=

f
/ ,

f
=

f / ,
= /
and I is the fourth-

order identity tensor.
Stress update and algorithmic tangent
Since the following operations are performed in a rate-independent effec-
tive stress space while the previous relations ((4.4) to (4.9)) were dened in
a rate-independent homogenised stress space, it is necessary to replace the
quantities with

and

with

.
The stress update is obtained by combining relations (4.2)
1
and (4.2)
3
to
express the stress-strain relationship as [69]
= (1 ) D
e
: (
vp
) . (4.10)
For the following derivations, the viscoplastic strain rate

vp
=
1
C
e
: (

) , (4.11)
formulated in the rate-independent effective stress space, can be expressed
more conveniently as a function of and

:

vp
=
1
C
e
:
_

1

_
. (4.12)
Time differentiation of (4.10), with substitution of
vp
from the above ex-
pression, gives the stress rate
+
_
1
+

1
_
=
1
+ (1 ) D
e
: (4.13)
which can be integrated, replacing time derivatives by their discrete coun-
terparts, using a backward Euler algorithm:
t
+
_
1
+

t (1
n+1
)
_
n+1
=
1
n+1

n+1
+ (1
n+1
) D
e
:
t
. (4.14)
Rearranging terms yields the incremental stress update relation
n+1
=
(
n
+ (1
n+1
) D
e
:) +
t
(1
n+1
)

n+1
1 +
t
+

n+1
n
1
n+1
(4.15)
as a function of quantities from the previous step and from the rate-
independent effective stress space. After the update of the effective rate-
independent stress

n+1
with a standard return mapping scheme in the ef-
fective stress space, and the update of
n+1
in (4.6) and of
n+1
in (4.3), the
update of
n+1
follows as mere function evaluations.
The algorithmic tangent operator D
pd
is obtained from the linearised for-
mat of the constitutive relationships at the end of the time stepnote that
d (
n
) = 0 d (
n+1
) = d (). To this end, (4.14) is expressed in a
format which is more convenient for the next derivations as
+
t

n+1
+

1
n+1
n+1
(1
n+1
)

n+1
(1
n+1
) D
e
: = 0. (4.16)
Differentiation of the above stress update relation requires the evaluation of
d () which can be expressed, using (4.6), as
d () =

n+1

n+1
d (
n+1
) =

n+1

n+1
t
+t
d (

n+1
) . (4.17)
From the relation between the plastic multiplier and the equivalent plastic
strain (d
= d
for the specic choice of the yield function, see Section 1.4),
after using the discrete form of (4.7) and employing the symmetry of

R,
the variation of the damage increment can be related to the variation of the
strain tensor:
d () =

n+1

n+1
t
+t
R
n+1
:

f
,n+1
f
,n+1
:
R
n+1
:

f
,n+1
f
,n+1

,n+1
:d ()
= r
n+1
:d () , (4.18)
where d () has been expressed as a function of quantities in the effective
rate-independent space. After noting that
d
_

1
n+1
n+1
_
=
1
n
1
n+1
D
e
:
e
n+1
d ()
+

n+1
n
1
n+1
d () , (4.19)
where use has been made of
n+1
/(1
n+1
) = D
e
:
e
n+1
, and that
d ((1
n+1
) D
e
:) = (1
n+1
) D
e
:d () (D
e
:) d () , (4.20)
the algorithmic tangent can be written as
D
pd
n+1
=
t
(1
n+1
)

D
p
n+1
+ (1
n+1
) D
e
s
n+1
r
n+1
1 +
t
+

n+1
n
1
n+1
(4.21)
with
s
n+1
=
t

n+1
+
1
n
1
n+1
D
e
:
e
n+1
+D
e
:. (4.22)
Note that the algorithmic tangent is not symmetric. The integration proce-
dure requires only the evaluation of D
pd
n+1
after the computation of

D
p
n+1
in
the effective stress space for the rate independent problem [46, 162].
4.1.2 Perzyna model
The rate-dependent isotropic elastoplastic-damage model described next is
derived from the class of models proposed by Ju [87], where the notion of
operator split [42] allows a very simple algorithmic treatment [87, 163, 164].
Stress update and algorithmic tangent
The stress update relation at the end of the time step follows from rela-
tions (4.2)
1
:
n+1
= (1
n+1
)
n+1
(4.23)
where the damage value is updated through
n+1
= (1 exp (
n+1
)) , (4.24)
with the equivalent plastic strain in the effective space for the rate-
dependent elastoplastic problem.
The fourth-order algorithmic tangent stiffness tensor D
pd
n+1
is dened by
d () = D
pd
n+1
:d () for variations d () of the current strain increment
. To derive the consistent tangent operator, (4.23) is differentiated at t
n+1
to obtain (dropping the subscript n + 1):
d () = (1 ) d ( ) d () . (4.25)
The variation d () of the damage increment can be related to d () by
d () =

d =

d =

r:d () , (4.26)
where r is a second-order tensor which depends on the plasticity model
in the effective stress space and which will be specied later. Substituting
d () from the above expression in (4.25), and recalling that d ( ) =
D
p
:d (), yields the consistent tangent operator for the elastoplastic-
damage model:
D
pd
= (1 )

D
p

r. (4.27)
The rate-dependent response of the plasticity model in the effective stress
space is governed by the Perzyna viscoplastic model [137]. In presence of
plastic ow(

f 0, where

f is the yield function in the effective stress space),
the viscoplastic strain rate for the Perzyna model is expressed in the asso-
ciative form according to

vp
=
1
, (4.28)
where the overstress function is given by the following power-law form
_

f
_
=
_

f

0
_
N
, (4.29)
with
0
the initial yield stress and N (N 1) a real number. After stan-
dard manipulations [47, 116], the algorithmic treatment of the constitutive
equations for Perzyna viscoplasticity in the effective stress space yields the
relation for d () as
d () =

f
R:d ()
R:

f
+/
_
t

f
_
, (4.30)
with d () = d, the increment of the plastic multiplier, and the consistent
tangent
D
p
=

R
R:

f
R:

f
+/
_
t

f
_
(4.31)
with

R dened in (4.9). After employing the symmetry of

R, the second-
order tensor r (see (4.26)), required for the evaluation of the consistent tan-
gent operator (4.27) for the elastoplastic-damage model, reads
r =

R:

f
R:

f
+/
_
t

f
_
. (4.32)
The consistent tangent operator for the elastoplastic-damage model is read-
ily available by direct substitution of the above expressions into (4.27). As for
the rate-dependent elastoplastic-damage model of the Duvaut-Lionstype,
the consistent tangent operator in (4.27) is not symmetric. The step-by-step
integration procedure is very similar to that of standard plasticity, the dif-
ference being the presence of the damage update which requires only the
evaluation of (4.23) and (4.27).
4.1.3 Inuence of model parameters
The inuence of the model parameters on the two rate-dependent
elastoplastic-damage models is studied by considering the integration point
level response for the von Mises yield function by means of a one-element
0 0.025 0.05 0.075 0.1
displacement [mm]
0
0.5
1
1.5
2
s
t
r
e
s
s

[
M
P
a
]
0 0.025 0.05 0.075 0.1
displacement [mm]
0
0.5
1
1.5
2
2.5
3
s
t
r
e
s
s

[
M
P
a
]
0 0.025 0.05 0.075 0.1
displacement [mm]
0
0.5
1
1.5
2
2.5
3
s
t
r
e
s
s

[
M
P
a
]
0 0.025 0.05 0.075 0.1
displacement [mm]
0
0.5
1
1.5
2
2.5
3
s
t
r
e
s
s

[
M
P
a
]
0 0.025 0.05 0.075 0.1
displacement [mm]
0
0.5
1
1.5
2
2.5
3
s
t
r
e
s
s

[
M
P
a
]
0 0.025 0.05 0.075 0.1
displacement [mm]
0
0.5
1
1.5
2
s
t
r
e
s
s

[
M
P
a
]
=1
=0
=0
=100
=1 s
=0.0001 s
=1 s
=0.0001 s
a=2
a=1
(e)
b=100
b=1
(f)
influence of softening law parameters
(a) (b)
influence of damage law parameters
(c) (d)
influence of relaxation time
Figure 4.1 Duvaut-Lionstype model, inuence of the model parameters on the constitutive
response: effect of the damage law parameters (a) with = 0, 0.5, 1 and = 100 and (b)
with =0, 1, 10, 100 and = 1 on the rate-dependent elastoplastic-damage model (a = 1,
b = 1, = 0.01 s); effect of the relaxation time with = 0.0001, 0.01, 1 s on the rate-
dependent (c) elastoplastic model with a = = = 0, b = 1 and (d) elastoplastic-damage
model with a = 0, b = 1, = 1, = 100; effect of softening law parameters (e) a with
a = 1, 0, 1, 2 and b = 10 and (f) b with b = 1, 10, 100 and a = 0 on the rate-independent
( = 0 s) elastoplastic model.
0 0.025 0.05 0.075 0.1
displacement [mm]
0
0.5
1
1.5
2

s
t
r
e
s
s

[
M
P
a
]

0 0.025 0.05 0.075 0.1
displacement [mm]
0
0.5
1
1.5
2

s
t
r
e
s
s

[
M
P
a
]

0 0.025 0.05 0.075 0.1
displacement [mm]
0
0.5
1
1.5
2

s
t
r
e
s
s

[
M
P
a
]

0 0.025 0.05 0.075 0.1
displacement [mm]
0
0.5
1
1.5
2

s
t
r
e
s
s

[
M
P
a
]

0 0.025 0.05 0.075 0.1
displacement [mm]
0
0.5
1
1.5
2

s
t
r
e
s
s

[
M
P
a
]

=0
=1 =100
=0
=0.001 s
=1000 s
=0.001 s
=1000 s
0 0.025 0.05 0.075 0.1
displacement [mm]
0
0.5
1
1.5
2

s
t
r
e
s
s

[
M
P
a
]

influence of softening law parameters
a=2
a=1
(e) (f)
b=100
b=1
(a) (b)
influence of damage law parameters
influence of relaxation time
(c) (d)
Figure 4.2 Perzyna model, inuence of the model parameters on the constitutive response:
effect of the damage law parameters (a) with = 0, 0.5, 1 and = 100 and (b) with
=0, 1, 10, 100 and = 1 on the rate-dependent elastoplastic-damage model (a = 1, b = 1,
= 1 s); effect of the relaxation time with = 0.001, 1, 1000 s on the rate-dependent
(c) elastoplastic model with a = = = 0, b = 1 and (d) elastoplastic-damage model with
a = 0, b = 1, = 1, = 100; effect of softening law parameters (e) a with a = 1, 0, 1, 2 and
b = 10 and (f) b with b = 1, 10, 100 and a = 0 on the rate-independent ( = 0 s) elastoplastic
model.
0
0
0.025
0.025
0.05
0.05
0.075
0.075
0.1
0.1
displacement [mm]
displacement [mm]
0
0
0.5
0.5
1
1
1.5
1.5
2
2
s
t
r
e
s
s

[
M
P
a
]
s
t
r
e
s
s

[
M
P
a
]
(a) (b)
Figure 4.3 Response of the model to a series of loading-unloading-reloading cycles com-
pared to a monotonic load response for (a) Duvaut-Lionstype model with = 0.01 s and
(b) Perzyna model with N = 1 and = 0.5 s.
test in displacement control on a 8-node quadrilateral element (element size
1 mm 1 mm). The element is subjected to monotonic linearly increas-
ing uniaxial loading at constant strain rate (t = 0.0001 s until the nal
displacement of 0.1 mm is reached in 1000 steps). The model parameters
adopted are: Youngs modulus E = 100 MPa and Poissons ratio = 0. The
softening rule governing the cohesion capacity of the material is given an
exponential form according to
() =
0
((1 + a) exp (b) a exp (2b)) , (4.33)
where a and b are model parameters, with the initial cohesion
0
= 1 MPa.
Note that the softening behaviour is represented both by damage and plas-
ticity effects through (4.3) and (4.33), respectively. The results of the analy-
ses are shown in Figures 4.1 and 4.2. The effective reduction of the residual
stress due to damage as in Figures 4.1a and 4.1b and the effect of the relax-
ation time reported in Figures 4.1c and 4.1d are worth noting. The effect of
the softening rule parameters is depicted in Figures 4.1e and 4.1f. Analogous
considerations hold for Figure 4.2. The constitutive response to a series of
loading-unloading-reloading cycles is reported in Figure 4.3 along with the
response of the model to monotonic loading (model parameters are: a = 0,
b = 1, = 1, = 100; relaxation time = 0.01 s and = 0.5 s with N = 1
for the Duvaut-Lionstype model and the Perzyna model, respectively).
4.1.4 Regularisation properties
The regularisation properties of the models are demonstrated considering
a bar of length L = 100 mm, thickness t = 1 mm and width increas-
ing from 8 mm at the restrained end to 10 mm at the free end (see Fig-
ure 4.4). This geometrical conguration was chosen to avoid dependence of
the width of the localisation zone on the nite-element discretisation [189].
In the nite-element discretisations, the boundary conditions are prescribed
restraining both bottom node directions and vertical top node direction; the
mesh consists of one row of equally spaced elements. The bar is subjected to
monotonic tensile loading with constant average strain rate obtained by in-
creasing the displacement at the free end linearly in time in 1000 steps with
t = 0.0001 s until the nal displacement of 0.1 mm. Von Mises yield func-
tion with yield stress
0
= 2 MPa is adopted. Other model parameters are:
Youngs modulus E = 24000 MPa, Poissons ratio = 0, a = 0 and b = 100
in the exponential softening law (4.33) and = 1 and = 300 for the expo-
nential damage evolution law (4.3). Relaxation time is set to = 0.005 s for
the Duvaut-Lionstype model and to = 3 s with N = 1 for the Perzyna
model. These model parameters have been chosen for numerical conve-
nience. Linear quadrilateral elements have been used. The results of the
simulations for different discretisations (20, 40, 80 and 160 equally spaced el-
ements) have been reported in Figure 4.5 for the Duvaut-Lionstype model
and in Figure 4.6 for the Perzyna model together with the results for the
rate-dependent elastoplastic model ( = = 0 in the exponential dam-
age evolution law (4.3)). Close to failure, computations with the Duvaut-
Lionstype model did not converge in the local Newton-Raphson iteration
scheme for the computation of the plastic multiplier (a maximum of 50 local
iteration was allowed) and the analyses were stopped (see Figure 4.5b). An
analysis of the problem showed that the evaluation of the derivatives of the
yield function was problematic due to an almost zero value of the yield func-
tion for some integration points in the localisation zone (this situation was
found to be related to the trial stress being of the same order of magnitude
as the stress increment D
e
:). Similar problems were encountered for stan-
u
_
100
8 10
Figure 4.4 Geometry and boundary
conditions for the tensile test (all dimen-
sions in mm).
0 0.01 0.02 0.03
elongation [mm]
0
5
10
15
20
r
e
a
c
t
i
o
n

f
o
r
c
e

[
N
]
20 el.
160 el.
20 160 el.
0.02 0.0225 0.025 0.0275 0.03
elongation [mm]
0
1
2
3
4
r
e
a
c
t
i
o
n

f
o
r
c
e

[
N
]
160 el.
20 el.
(a) (b)
Figure 4.5 Duvaut-Lionstype model: (a, b) load-displacement curves for 20, 40, 80 and
160 element discretisations for the rate-dependent elastoplastic (dashed line) and the rate-
dependent elastoplastic-damage (solid line) model; (b) close-up.
Figure 4.6 Perzyna model: load-
displacement curves for 20, 40, 80 and
160 element discretisations for the rate-
dependent elastoplastic (dashed line) and
the rate-dependent elastoplastic-damage
(solid line) model.
0 0.025 0.05 0.075 0.1
elongation [mm]
0
5
10
15
20
r
e
a
c
t
i
o
n

f
o
r
c
e

[
N
]
20 el.
160 el.
160 el.
20 el.
dard Duvaut-Lions viscoplasticity, i.e. without the coupling to damage. In
the following, only the Perzyna model will be considered. The curves for the
Perzyna model reported in Figure 4.6 show convergence to a unique solu-
tion and the curves obtained with the rate-dependent elastoplastic-damage
model clearly show mesh-dependence close to failure due to strain localisa-
tion in one element (of course, this does not mean that there is a patholog-
ical mesh dependence since the model is regularised, but indicates that the
model is able to reproduce a highly localised strain eld). This is clearer from
the stroboscopic evolution plot of the equivalent plastic strain reported in
Figure 4.7 where strain localisation due to damage is evident. The effect of
the viscous regularisation is evident from Figures 4.84.9 where a higher
relaxation time corresponds to a higher energy dissipation and a wider lo-
calisation zone. In particular, it is worth noting the inuence of the coupling
0 25 50 75 100
distance from clamped end [mm]
0
0.5
1.0
1.5
0 25 50 75 100
0
0.5
1.0
1.5
0 25 50 75 100
0
0.5
0.1
1.5
0 25 50 75 100
0
0.5
1.0
1.5
]
x

1
0
0
]
x

1
0
0
]
x

1
0
0
]
x

1
0
0
(a)
20 el.
(c)
20 el.
elastoplastic model
elastoplasticdamage model
(b)
160 el.
(d)
160 el.
Figure 4.7 Perzyna model: stroboscopic evolution of the equivalent plastic strain for the
rate-dependent (a, b) elastoplastic and (c, d) elastoplastic-damage model for (a, c) 20 and (b,
d) 160 element discretisations.
between damage and plasticity on the equivalent plastic strain reported in
Figure 4.9: it is evident that the coupling allows a more localised equivalent
strain prole compared to the plasticity model alone while leaving almost
unchanged the width of the localisation zone. The load-displacement curves
and the damage proles depicted in Figure 4.8 indicate that this model can
properly represent strain localisation at the global and local level through
full stress relaxation and localised damage prole, respectively (see also Sec-
tion 1.3 and Figure 3.3).
In this section some aspects of the nite-element implementation are
discusseddetails on the integration scheme and on the transfer of history
data can be found in Section 2.5.
4.3 Applications 85
0 0.05 0.1 0.15 0.2 0 20
elongation [mm]
40 60 80 100
0
5
10
15
20
r
e
a
c
t
i
o
n

f
o
r
c
e

[
N
]
0
0.2
0.4
0.6
0.8
1
d
a
m
a
g
e

[
]
=3 s
=9 s
=3 s
=9 s
(a) (b)
Figure 4.8 Perzyna model: effect of the relaxation time on (a) the global response and on
(b) damage prole at the end of the computation.
Introducing a discontinuity. The inclusion of displacement discontinu-
ities represents, in this context, either genuine separation of material (i.e. a
stress-free discontinuity) or the formation of cohesive surfaces. In the former
case, a discontinuity is extended when damage at one or more integration
points in the element ahead of the discontinuity tip is above a critical value;
in the latter case this condition has to be fullled at all the integration points
in the element ahead of the discontinuity tip. The critical damage value,
close to unity for quasi-brittle materials, will be specied in the application
to SFR-concrete.
Orienting a discontinuity. The discontinuity direction is aligned with the
direction of maximum dissipation. This is achieved performing a non-local
averaging of the equivalent plastic strain, the quantity driving damage evo-
lution, ahead of the discontinuity tip. The vector in the direction of the dis-
continuity propagation is computed as in Section 3.5, replacing e
i
with
i
in (3.24). The interaction radius for the Gaussian weight function in (2.26)
is taken equal to three times the average element size ahead of the discon-
tinuity tip. When the discontinuity is close to a boundary, the discontinuity
extension direction is aligned with the previous discontinuity segment.
0 20 0 40 60 20 80 40 100 60 80
100
0
0.005
0
0.01
0.005
]
0.01
]
=3 s
=9 s
=9 s
=3 s
(b) (a)
Figure 4.9 Perzyna model: effect of the relaxation time on the equivalent plastic strain for
the rate-dependent (a) elastoplastic and (b) elastoplastic-damage model.
0
500
1000
1500
2000
0 0.5 1 1.5 2 2.5 3
a
p
p
l
i
e
d

l
o
a
d

[
N
]
u [mm]
Figure 4.10 Inclusion of discontinuities in
a rate-dependent elastoplastic model: load-
displacement response for a biaxial speci-
men under tensile loading (adapted from
Wells [192]).
4.3 Applications
The model for bulk degradation described in Section 4.1.2 is endowed with
some properties which make it a suitable tool for failure analyses. In contrast
to standard rate-dependent elastoplastic models, characterised by a constant
width of the localisation zone (see Figure 4.7b) and by a high residual stress
due to the viscous contribution, this rate-dependent elastoplastic-damage
model allows the progressive narrowing of the localisation zone (see Fig-
ure 4.7d) and full stress relaxation, which can be interpreted as a stress-free
crack in a continuous setting. If the regularised model is unable to describe
the narrowing of the degraded zone, the inclusion of a traction-free discon-
tinuity is problematic and its use should be avoided. An example of such
a model is the gradient elastoplasticity model proposed by M uhlhaus and
Aifantis [114] and Pamin [128] which is not able to properly model com-
4.3 Applications 87
Figure 4.11 Load-cmod diagrams for the
compact-tension specimen for different mesh
resolutions (h indicates the average element
size in the central part of the specimen).
0 1 2 3 4 5
cmod [mm]
0
500
1000
1500
2000
F

[
N
]
experiment
h=2, 1, 0.5 mm
plete failure since gradient contributions make it impossible to reach zero
stress values (quoted from Engelen et al. [55]). A similar problem is to be
found in classical viscoplasticity formulations (like e.g in Duvaut-Lions or
Perzyna viscoplasticity) in which the stress-strain relationship presents a
horizontal plateau or even an increasing stress due to the viscous stress
contribution [179]. When traction-free discontinuities are considered in a
rate-dependent viscoplastic model, load-displacement curves exhibit a saw-
toothlike shape caused by high residual stresses which have not been dis-
sipated [192, 194] (see Figure 4.10 and also Figures 3.6 and 3.16). In the fol-
lowing, some applications to problems involving strain localisation will be
analysed.
4.3.1 Composite compact-tension specimen
The compact-tension specimen studied in Section 3.6.2 is now analysed con-
sidering the model for bulk degradation described in Section 4.1.2. The
simulations are performed using the same unstructured meshes of bilinear
quadrilateral elements used in Section 3.6.2. The total cmod (5 mm) has been
applied in 200 steps with cmod = 0.025 mm. Plastic ow is described by
a smoothed Rankine yield criterion [128] with
0
= 35.2 MPa under the
assumption of a plane stress situation. Softening parameters are: a = 1.5
and b = 13 in (4.33) and = 1 and = 20 in (4.3). Other model parame-
ters have been chosen as follows: N = 1 and relaxation time = 20 s for
the viscous regularisation of (4.28), Youngs modulus E = 3200 MPa and
Poissons ratio = 0.28. Of the above model parameters, E, and
0
have
been adopted from Geers et al. [65]. The remaining parameters have been
chosen to ensure a qualitative t at the global and local levels in terms of
load-cmod curve and damage prole, respectively, with the data reported
(a) (b)
0.1 1
Figure 4.12 Failure state at cmod = 5 mm: damage contour plot for (a) coarse (h =
2 mm) and (b) ne (h = 0.5 mm) mesh (the thick white line indicates the traction-free
discontinuity).
by Geers et al. [65]. To avoid damage growth in the elements around the
pinholes, a higher value of
0
has been given to elements in these areas. The
critical damage value for discontinuity extension (traction-free) has been set
to 0.99. An initial horizontal traction-free discontinuity has been placed in
the notch zone as starting point for the discontinuity extension. This has
no effect on the global/local response since the traction-free discontinuity
mimics the real notch.
The load-cmod response is shown in Figure 4.11. The marked drop in the
load-cmod curve for the coarsest of the discretisations (h = 2 mm) relates to
the high residual stress in the element at the moment of the rst extension
of the discontinuityonly one integration points had damage values larger
than 0.99 while in the remaining integration points damage was around 0.7.
The following discontinuity extensions are characterised by a smooth global
response as a consequence of comparable damage values at all integration
points of one element. The failure pattern at cmod = 5 mm is shown in Fig-
ures 4.12 and 4.13. The continuous-discontinuous failure strategy provides a
clear indication of the extent of the discontinuity and of the zone of continu-
ous degradation. However, it is worth noting that no signicant differences
have been observed with respect to computations performed in a continu-
ous framework with the nest of the discretisations in terms of local and
4.3 Applications 89
(a) (b)
Figure 4.13 Failure state at cmod = 5 mm: discontinuity extension for (a) coarse
(h = 2 mm) and (b) ne (h = 0.5 mm) mesh.
global responses.
A comparison of results obtained with the continuous-discontinuous im-
plicit gradient-enhanced continuum damage model described in Chapter 3
and the continuous-discontinuous rate-dependent elastoplastic-damage
model is reported in Figures 4.14 and 4.15. Both continuous-discontinuous
regularised models are able to reproduce the experimental results with rea-
sonable accuracy, as depicted in Figure 4.14. However, the comparison in
terms of failure evolution in Figure 4.15 clearly indicates that the gradient-
enhancement produces a delayed discontinuity extension caused by the
shift of the maximum of the quantity driving damage evolution (see Sec-
tions 3.8 and 4.4 and Appendix E for more details on this issue).
4.3.2 Strip footing near a slope
A more interesting failure pattern can be observed from the numerical anal-
ysis of the rigid and rough strip footing resting on the crest of a slope de-
picted in Figure 4.16. A 0.6 m deep initial traction-free discontinuity is lo-
cated on the right side of the footing. The domain has been discretised by
using 8-node quadrilateral elements with reduced integration scheme930
elements for the coarse mesh and 3720 elements for the ne mesh. Plane
strain J2 softening plasticity with cohesive capacity according to (4.33) with
initial cohesion
0
= 0.00005 MPa, a = 1 and b = 100 and with damage
growth according to (4.3), with = 0.9999 and = 200, has been used
0 1 2 3 4 5
cmod [mm]
0 1 2 3 4 5
cmod [mm]
0
500
1000
1500
2000
F

[
N
]
0
500
1000
1500
2000
F

[
N
]
experiment
h=2, 1, 0.5 mm
experiment
h=2, 1, 0.5 mm
(b) (a)
Figure 4.14 Comparison of load-cmod diagrams for the compact-tension specimen for dif-
ferent mesh resolutions h for (a) implicit gradient-enhanced continuum damage model (see
Chapter 3) and (b) rate-dependent elastoplastic-damage model.
to describe the material behaviour. The remaining model parameters have
been chosen as follows: N = 1 and relaxation time = 1000 s, Youngs
modulus E = 0.1 MPa (the strip is assumed to be elastic and one hundred
times stiffer) and Poissons ratio = 0.2. The above model parameters have
been chosen for numerical convenience but may be thought of as represent-
ing the undrained behaviour of a heavily overconsolidated soil. The high
value of the relaxation time allows a meaningful width of the localisation
band. The rigid footing has been loaded in displacement control and the
total vertical displacement u = 100 mm has been applied in 50 steps with
u = 2 mm. The critical damage value for discontinuity extension (traction-
free) has been set to 0.99.
The failure of the slope develops in the typical curved shape depicted
in Figures 4.17b and 4.17d and is accompanied by localised deformations
which are well described by the continuous model and by the following
discontinuity. The direction of the discontinuity is properly described even
with the coarse mesh due to the signicant extension of the equivalent plas-
tic strain beyond the discontinuity tip. Note however that, due to the un-
symmetric geometry of the slope stability problem, the alignment of the
discontinuity with the previous discontinuity segments close to the left
boundary is necessary in order to obtain a realistic failure pattern. Similar to
the simulations of the composite compact-tension specimen, no signicant
differences have been observed between a continuous and a continuous-
discontinuous analysis in terms of local and global responses. The use of
4.3 Applications 91
cmod = 2.00 mm
cmod = 2.75 mm
cmod = 5.00 mm
(a) (b)
0.1 1
Figure 4.15 Evolution of the failure process for (a) implicit gradient-enhanced continuum
damage model (see Chapter 3) and (b) rate-dependent elastoplastic-damage model.
u =0
y
u

=
0
x
5 5
12
1
0
u
0
.
6
Figure 4.16 Geometry and boundary conditions for
the slope (all dimensions in m).
traction-free discontinuities does not improve failure representation when
bulk degradation obeys the rate-dependent elastoplastic-damage model de-
scribed in Section 4.1.2.
4.3.3 Steel bre-reinforced concrete beam
Discontinuities can be better employed for the description of SFR-concrete.
Numerical modelling of SFR-concrete is based on the use of stress-crack
opening relationships [76]. Within the stress-crack opening approach, a rea-
sonable approximation of the tensile behaviour of SFR-concrete can be ob-
tained by using a bilinear relationship [99] of the type depicted in Fig-
ure 4.18a. According to Stang and Olesen [184], the rst part of the bilin-
ear relationship reects a combination of the concrete contribution and the
initial bre bridging action, while the second part reects the bre bridging
action only. For normal strength concrete and low bre content, the rst
branch of the bilinear relationship can be taken equal to that of plain con-
crete, so that the effect of the bres is reected only by the second branch.
The numerical strategy proposed for the modelling of SFR-concrete (see Fig-
ure 1.4a) consists in the use of the model for bulk degradation, presented in
4.3 Applications 93
(a)
(c) (d)
(b)
Figure 4.17 Slope failure: (a, c) damage contour plot and discontinuity at failure and (b,
d) deformed state (magnied 100 times) for different mesh resolutions (the thick white line
indicates the discontinuity).
Section 4.1.2, to describe, in a continuous fashion, the initial degradation of
the cementitious matrix, represented by the rst part of the curve in Fig-
ure 4.18a. A cohesive discontinuity is introduced at a specied level of
damage of the cementitious matrix, corresponding to a residual stress f
t
(see Figures 4.18b and 4.19), and represents the second part of the curve in
Figure 4.18a. The constitutive model for the cohesive discontinuity is de-
ned by a linear decreasing relationship between the stress and the crack
u
w
max
f
(a)
(b)
f
t
t
1
0.7
0.8
0.9
=
=
=
Figure 4.18 Constitutive models: (a) bilinear constitutive relationship for SFR-
concrete and (b) qualitative stress-equivalent plastic strain plot with critical dam-
age values for the activation of the cohesive discontinuity.
u
w
max
f
t

f
t
cohesive discontinuity softening continuum
Figure 4.19 From continuous to discontinuous failure description in SRF-concrete

(the dashed line indicates that the continuum close to the discontinuity experiences
increasing dissipation after the introduction of a cohesive discontinuity).
opening w with = f
t
at w = 0 and = 0 for w w
max
where w
max
is
related to the bre length.
Anotched beamin four point bending with a 150 mm150 mmcross sec-
tion, a 550 mm span and a 25 mm notch is analysed. The load is applied via
displacement control at a distance of 75 mm from the midspan and the total
deection v (0.4 mm) has been applied in 160 steps with v = 0.0025 mm.
The beam was discretised with bilinear quadrilateral elements with element
size, in the central part of the beam, of 2.73 mm 3.125 mm. Bulk degra-
dation is described by plane stress smoothed Rankine plasticity [128] with
f
t
=
0
= 2 MPa and with softening parameters a = 0, b = 200, = 0.99
and = 400. The other model parameters have been chosen as follows:
N = 1 and relaxation time = 20 s, Youngs modulus E = 30 GPa and
Poissons ratio = 0.2. The notch has been described by a traction-free
discontinuity. SFR-concrete is characterised by the extension of a cohesive
discontinuity related to the bre action at three different damage values,
= 0.7, 0.8 and 0.9, from the traction-free notchnote that the cohesive
4.3 Applications 95
0 0.1 0.2 0.3 0.4
deflection [mm]
plain concrete
SFRconcrete
0
5
10
15
20
l
o
a
d

[
K
N
]
0.7
0.8
0.9 =
Figure 4.20 Load-deection curves for

plain and SFR-concrete.
(a) (b)
0.1 1
Figure 4.21 Damage contour plot for
(a) plain and (b) SFR-concrete ( = 0.8) at
deection = 0.4 mm.
discontinuity is extended if all integration points reach the critical damage
value . For plain concrete, a traction-free discontinuity was extended at
crit
= = 0.989 from the traction-free notch.
The values of f
t
at which damage reaches the above values are depicted
in Figure 4.18b in the qualitative plot of the rst principal stress
1
versus
the equivalent plastic strain for an integration point in the element ahead
of the notch. The value of w
max
was chosen equal to 25 mm for all the sim-
ulations. Figures 4.20 and 4.21 clearly indicate the effect of the cohesive dis-
continuity on the global and local response, respectively. Figure 4.21 indi-
cates that the use of a cohesive discontinuity promotes dissipation across
the discontinuity surface as well as in the continuum. A more ductile re-
sponse, which corresponds to a wider zone of high damage concentration,
is obtained with a cohesive discontinuity (see Figure 4.21b) compared to
the response obtained with a traction-free discontinuity (see Figure 4.21a),
where high damage values are concentrated only close to the discontinuity.
The damage values for the activation of the bre and the shape of the co-
hesive relationship have been chosen for numerical conveniencethey can
be related e.g. to the strength of the cementitious matrix. The aim of this aca-
demic example was to illustrate how changes at integration point level (un-
derstood as changes at the material level) inuence the structural response
in a regularised continuous-discontinuous model. Note that this approach
is not feasible in a non-local continuum because of the severe oscillations in
the quantity governing damage growth (see Section 3.8).
4.4 Traction-free discontinuities in rate-dependent and
non-local media
It is interesting to further compare the performance of this rate-dependent
continuous-discontinuous model to the one obtained when discontinuities
are considered in a non-local medium (see Chapter 3). To this end, numeri-
cal analyses similar to those performed with the gradient-enhanced damage
model in Section 3.8 have been performed with the rate-dependent model.
Although some of the following issues have already been touched upon
in Section 3.8, they are recalled here to draw a parallel between the use
of traction-free discontinuities in gradient-enhanced and in rate-dependent
media. In the examples, the model parameters for the viscous regularisation
are similar to those used in Section 4.3.1. A ner discretisation, compared to
the one used in Section 3.8, has been employed in the numerical simulations.
An important conclusion can be drawn from the results of the simulations
of the compact-tension test, shown in Figures 4.22 and 4.23, and it is related
to the actual meaning of the inclusion of discontinuities (of any nature, co-
hesive or traction-free), which are local in space, in problem elds which
are, by denition, non-local. As already discussed in Section 3.8 and as de-
picted in Figure 4.22a, for the analyses performed with the implicit gradient-
enhanced continuum damage model [134], of all the basic problem elds
(vertical displacement eld u
y
and non-local equivalent strain eld e), only
the non-local strain eld suffered from severe oscillations upon a traction-
free discontinuity extension. Analogous analyses depicted in Figure 4.22b
and performed with the rate-dependent model did not show oscillations in
the equivalent plastic strain . This can be considered as a consequence of
the nature of the regularisation employed which is non-local in space for the
implicit gradient-enhanced continuum damage model and local in space for
this rate-dependent elastoplastic-damage.
The examples reported in Section 3.8 showed that the use of discontinu-
ities in a gradient-enhanced medium can be problematic due to the signi-
cant changes in failure initiation and characterisation (see also Appendix E).
When a sharp crack is considered in a non-local medium, damage initia-
tion is predicted ahead of the crack tip and not at the crack tip itself. This
shift in the location of damage initiation is proportional to the length scale
parameter. Numerical evidence showed that the shift of the maximum of
the non-local equivalent strain away from the crack tip is even more pro-
nounced in the non-linear regime than it is in the elastic regime. The con-
4.4 Traction-free discontinuities in rate-dependent and non-local media 97
pseudotime
0
0.2
0.4
0.6
0.8
u
y
u
y
a
1
e
s
c
a
l
e
d

s
o
l
u
t
i
o
n

a
r
r
a
y
pseudotime
0
0.2
0.4
0.6
0.8
1
s
c
a
l
e
d

s
o
l
u
t
i
o
n

a
r
r
a
y
(a) (b)
Figure 4.22 Pseudo-time evolution of (a) the vertical displacement eld and the
dissipation-driving quantity e for implicit gradient-enhanced continuum damage model
(see Chapter 3) and of (b) the vertical displacement eld and for the rate-dependent
elastoplastic-damage model in a compact tension test for point a (the discontinuity prop-
agates from the notch along the dotted line; the dots represent the extension of the dis-
continuity and the dotted line indicates the rst moment at which the point is behind the
discontinuity tip).
0 1 2 3 4
displacement [mm]
0
100
200
300
r
e
a
c
t
i
o
n

f
o
r
c
e

[
N
]
0 1 2 3 4
displacement [mm]
0
100
200
300
r
e
a
c
t
i
o
n

f
o
r
c
e

[
N
]
(a) (b)
Figure 4.23 Load-displacement curves for (a) implicit gradient-enhanced continuum
damage model (see Chapter 3) and (b) rate-dependent elastoplastic-damage model with a
propagating discontinuity (the dots represent the extension of the discontinuity; note that
the extension is performed through several elements at a time for the gradient-enhanced
model and through one element at a time for the rate dependent model).
sequence is that, for a reasonably ne discretisation, several elements are
crossed at a time and the discontinuity extension can be delayed as reported
in Figure 4.15a. The oscillations of the dissipation driving quantity are more
pronounced by the release of more than one element at a time. This causes a
0 0 10 20 10 30 20 40 30 50 40 60 50 60
position [mm] position [mm]
(b) (a)
0
0.05
0.1
0.15
0
]
0.5
1
1.5
2
]
crack tip
initial initial
crack tip
propagating disc.
fixed disc.
propagating disc.
Figure 4.24 Prole of the equivalent plastic strain for the rate-dependent elastoplastic-
damage model: (a) comparison of proles with xed and propagating discontinuity and
(b) close-up of the prole in case of propagating discontinuity (proles plotted close to
failure).
situation of articial shock-wise crack propagation and of spurious unload-
ing in the points ahead of the newly extended discontinuity tip resulting
in the bumps in the load-displacement curve as depicted in Figure 4.23a.
For the rate-dependent elastoplastic-damage model (plane stress Rankine),
some numerical evidence suggests that the dissipation-driving eld quan-
tity is maximum, at the beginning of the dissipation process, at the crack tip.
Mesh renement studies also indicated that the energy dissipated during a
load process converges to a non-zero value. From here it can be concluded
that the dissipation-driving eld quantity converges to a nite value at the
crack tip (these conclusions are based on numerical considerations and, as
such, must be validated by analytical studies). When the rate-dependent
model is considered, usually one element at a time is crossed by the dis-
continuity and the load-displacement curves do not show bumps (see Fig-
ure 4.23b).
A discontinuous problem eld interpolation is intended to give a better fail-
ure representation. The justication for the use of displacement discontinu-
ities in the rate-dependent elastoplastic-damage model stems from a better
interpretation of some eld quantities which gain a clearer physical mean-
ing. With reference to the example reported in Section 4.4, when consider-
ing the evolution of the equivalent plastic strain ahead of the disconti-
nuity tip as depicted in Figure 4.24, the propagating discontinuity avoids
the articial growth of the equivalent plastic strain which, in a continuous
setting, is the response of the model to strain localisation. When a traction-
free discontinuity is considered, this is the only observable difference when
the responses of the continuous and the continuous-discontinuous models
are compared (load-displacement curves and damage proles are identical).
Conversely, when a cohesive discontinuity is considered, as was presented
in Section 4.3.3, the continuous-discontinuous strategy proved to be effec-
tive in the qualitative description of SFR-concrete.
Unlike standard regularised models in which failure description can be
problematic close to complete failure (e.g. difculties in reaching full stress
relaxation in rate- [179] and gradient-dependent [55] media and spurious
damage growth in gradient-dependent media [64]), the continuum model
alone is a valuable tool for the description of strain localisation phenomena.
Further, its combination to a technique in which displacement discontinu-
ities are considered can enhance failure representations in a wide range of
applications. It is stressed that the continuous-discontinuous strategy de-
scribed in this chapter is intended, at this stage, as a tool to the description of
phenomena in which a displacement discontinuity arises as a consequence
of strain localisation. More insight into the determination of model param-
eters, using the technique described e.g. in References [44, 91], is needed to
transform this approach from a descriptive tool into a predictive tool.
Chapter 5
Conclusions
The design of safe structures is intimately linked to the understand-
ing of failure processes of engineering materials. Only in exceptional
circumstancese.g. in building demolitionfailure is a desirable event; in
all other cases, failure must be delayed or, if possible, prevented. The role
of the analyst is that of providing practitioners with reliable tools for fail-
ure analysis so that failure processes can be prevented or controlled. In re-
cent times, a host of constitutive models and computational tools have been
developed to assist practitioners in the analysis of (quasi-)brittle and duc-
tile fracture. Amongst all these models and tools, damage and plasticity
theories and continuous and discontinuous analyses have emerged, respec-
tively. Damage and plasticity theories as well as continuous and discontinu-
ous analyses indicate relative situations; whether a particular failure process
is better represented within a damage or a plasticity framework through a
continuous or a discontinuous analysis depends on the situation. In most
cases, a realistic failure representation should consider a coupled damage-
plasticity model in a combined continuous-discontinuous failure represen-
tation, rather than a specic constitutive model in a continuous or discon-
tinuous framework.
The development of computational strategies to combine techniques of
continuous and discontinuous numerical failure analysis was the objective
of this thesis. A continuous-discontinuous approach to failure is advocated
as a way to improve classical continuous failure representation by (i) allow-
ing a different interpretation of model parameters (see Chapters 1 and 3),
(ii) enabling a more realistic failure representation by solving some prob-
lems inherent to some regularised models (see Chapter 3), and (iii) adding
exibility to continuous modelling alone (see Chapters 2 and 4). Besides
the applications presented in this thesis, the superiority of a continuous-
discontinuous approach to failure, compared to a conventional continuous
approach, is evident. The particular choice of partition of unity-based dis-
continuous elements as the tool for the discontinuous enhancement in the
continuous-discontinuous failure approach was instrumental in (i) provid-
102 Chapter 5 Conclusions
ing a mechanically sound approach when compared to embedded disconti-
nuity element techniques, and (ii) in providing a clear framework in which
the continuous-discontinuous analysis is cast when compared to the use of
conventional interface elements coupled to remeshing and adaptive tech-
niques. The only assumption behind the application of this method is the
decomposition of the displacement eld into the sum of a continuous and a
discontinuous eld.
However, the simplicity of the partition of unity-based discontinuous el-
ement concept is counterbalanced by the difculties related to its imple-
mentation which requires heavy modications to element routines and ac-
cess to the whole code (which is not an option in commercial nite-element
method packages). Partition of unity-based discontinuous elements are not
the panacea for all the problems involving discrete cracking. Indeed, parti-
tion of unity-based discontinuous elements are best suited to describe major
cracks. When the focus is on the description of distributed failure in a dis-
continuous setting, conventional interface elements or embedded discon-
tinuity elements, within a cohesive-zone approach, should be used. When
partition of unity-discontinuous elements are used, there should be an ex-
tra set of degrees of freedom dened for each crack; although possible in
principle, the use of many extra set of degrees of freedoms is problematic to
handle, expecially in the case of intersecting discontinuities, and should be
avoided.
The main conclusions of this study can be summarised as follows:
1. model parameters should be reassessed when continuous(-discontin-
uous) failure descriptions are considered (see Sections 1.1 and 3.6.1
and Appendix C);
2. partition of unity-based discontinuous elements represent no general
remedy against the shortcomings of conventional interface elements
(only in terms of stress oscillations) and their use should not be consid-
ered in combination to a dummy stiffness approach (see Section 2.6.2
and Appendix A);
3. although the use of discontinuities helps in avoiding spurious damage
growth in gradient-enhanced media (see Chapter 3), discontinuities
and non-locality do not get along well (see Sections 3.8 and 4.4);
4. Duvaut-Lions viscoplasticity is poorly suited for the description of
softening media close to failure (see Section 4.1.4);
Conclusions 103
5. compared to the gradient-enhanced damage model used described in
Chapter 3, the rate-dependent elastoplastic-damage model described
in Chapter 4 can be used with coarser meshes to adequately represent
strain localisation (see Section 4.3);
6. the continuous-discontinuous rate-dependent elastoplastic-damage
model described in Chapter 4 results in the most realistic and com-
putationally effective description of failure;
7. the underlying regularised continuum determines the quality of the
discontinuous enhancement (see Section 4.3.1);
8. gradient models derived from non-local formulations can be classied
as coupled problems, not requiring mixed nite-element formulations
in the classical sense (see Appendix D), and
9. constitutive models based on a non-local dissipation-driving variable
may lead to incorrect failure initiation and propagation in arbitrary
loading scenarios due to a fundamental aw in damage characterisa-
tion (see Appendix E).
Appendix A
Conventional interface and PU-based discontinuous
elements
The performance of PU-based discontinuous elements deteriorates when

these elements are used with a dummy stiffness approach because of the
incorrect representation of the traction prole along the crack line (see Sec-
tion 2.6.2). When oscillations in the traction prole are present, the non-
linear interface response cannot be activated correctly.
An examination of the interface contribution to the stiffness matrix of PU-
based discontinuous elements through the conventional interface element
perspective [160] may provide clarity on their limitation and cast some light
towards possible improvements. In what follows, it is argued that the cause
of the problem could be traced back to the contribution of the displacement
discontinuity to the element stiffness matrix. A comparison between plane
conventional and PU-based discontinuous elements is elaborated, and it is
shown that the two methodologies share the same structure of the part of
the stiffness matrix responsible for interface behaviour. The effect of differ-
ent integration schemes on conventional continuous interface elements and
on PU-based discontinuous elements is investigated and their inuence on
the structure of the stiffness matrix of PU-based discontinuous elements is
illustrated.
A.1 Conventional continuous interface element
The stiffness matrix of a conventional isoparametric m-node line interface
element is expressed as
K = b
=+1
=1
B
T
CB
x
d, (A.1)
Based on References [168, 171]

106 Appendix A Conventional interface and PU-based discontinuous elements
where b is the width of the interface and B is the 2 2m matrix
B =
_
N N 0 0
0 0 N N
_
(A.2)
containing isoparametric shape functions N
i
with
N =
_
N
1
, ..., N
m
. (A.3)
The interface constitutive matrix C is obtained by the constitutive matrix
in the local s,n system after the transformation C = RT
sn
R
T
, with R the
rotation matrix from the the local frame into the global frame of reference,
R =
_
cos
xs
sin
xs
sin
xs
cos
xs
_
, (A.4)
and T
sn
a traction-separation relation of the type
T
sn
=
_
d
s
0
0 d
n
_
, (A.5)
where d
s
and d
n
are constant. For the element depicted in Figure A.1, the lo-
cal and global frames of reference coincides, therefore C = T
sn
. The degrees
of freedom have been ordered in the sequence
_
u
1
n
, ..., u
m
n
u
1
s
, ..., u
m
s
according to the node numbering of Figure A.1. For a conventional linear

line interface element the N matrix is equal to
N =
_
1
2
(1 )
1
2
(1 +)

(A.6)
while for a conventional quadratic line interface elements it reads
N =
_
1
2
_
+
2
_
1
2
_
+
2
_ _
1
2
_
. (A.7)
Expansion of the term B
T
CB results in the block diagonal matrix
B
T
CB =
_
K
n
0
0 K
s
_
, (A.8)
where K
i
is given by
K
i
= d
i
_

K
K

K
_
(A.9)
A.1 Conventional continuous interface element 107
x
y
5
2 3
6
1
4
s
n
Figure A.1 Conventional 6-node line interface element.
and d
i
is the stiffness at the interface in the direction i. The structure of the
sub-matrix

K depends on the interpolation along the conventional interface
element. For line elements it results in
K =
_
N
2
1
N
1
N
2
N
1
N
2
N
2
2
_
. (A.10)
Analytical integration (and a 2point Gauss integration scheme) of the terms
N
i
N
j
in (A.10) results in a full matrix while nodal integration (2point closed
Newton-Cotes formula i.e. trapezoidal rule) removes the coupling terms
N
i
N
j
with i ,= j. For a quadratic line interface element the submatrix

K
reads
K =
_
_
N
2
1
N
1
N
2
N
1
N
3
N
2
N
1
N
2
2
N
2
N
3
N
3
N
1
N
3
N
2
N
2
3
_
_
, (A.11)
and the use of analytical integration (and a 3point Gauss integration
scheme) results in a full matrix while the coupling terms N
i
N
j
are deleted
for i ,= j using again nodal integration (3point Newton-Cotes formula i.e.
Simpsons rule). In other words, the terms responsible for the coupling be-
tween the degrees of freedom are deleted if a nodal integration scheme is
used.
(a) (b) (c) (d) (e)
3
4
1
4
1
2
3
3
2
4
1
1
4
3
2
1
4
3
2
2
Figure A.2 Possible congurations for a horizontal discontinuity crossing a PU-based
discontinuous linear quadrilateral element.
A.2 Partition of unity-based discontinuous elements
For the two-dimensional m-node PU-based discontinuous element depicted
in Figure A.2, where m = 4, the contribution of K
bb
(see (2.24)) on
d
reads
K
bb,
d
=
d
N
T
RT
sn
R
T
N d = b
=+1
=1
N
T
T
sn
N
x
d, (A.12)
where N is the 2 2m matrix
N =
_

N 0
0

N
_
(A.13)
containing the isoparametric shape functions of the element with
N =
_
N
1
, ..., N
m
. (A.14)
The sequence
_
u
1
x
, ..., u
m
x
u
1
y
, ..., u
m
y
_
, with the corner nodes in the rst
m positions and the interior nodes in the remaining positions, has been used
for the ordering of the degrees of freedom in (A.13). Expansion of the inte-
grand in (A.12) results in the block diagonal matrix
N
T
T
sn
N =
_

K
n
0
0

K
s
_
, (A.15)
where

K
i
= d
i

K and

K is given by
K =
_
_
N
2
1
N
1
N
2
. . . N
1
N
m
N
2
N
1
N
2
2
. . . N
2
N
m
.
.
.
.
.
.
.
.
.
.
.
.
N
m
N
1
N
m
N
2
. . . N
2
m
_
_
. (A.16)
A.2 Partition of unity-based discontinuous elements 109
0.5000 0.5000
0.1667 0.1667
(a)
0.5000 0.5000
0.5000 0.5000
(b)
Figure A.3 Eigenmodes and corresponding eigenvalues for a linear PU-based discontin-
uous element with discontinuity placed along an element side for 2point (a) Gauss and
(b) trapezoidal integration rule.
It is worth to stress the similarity between matrix

K in (A.16) and matrix

K
in (A.9) dened for conventional line interface elements.
To illustrate the signicance of the coupling between degrees of freedom
in a PU-based discontinuous element, a linear quadrilateral element (m = 4)
is analysed next. In the case of Figure A.2a and with reference to matrix

K
in (A.16), the use of a trapezoidal rule will activate the interaction between
nodes 1 and 4 and between nodes 2 and 3 only (natural coupling), pre-
venting, for instance, the coupling between node 1 and node 2 (pathologi-
cal coupling); more specically, the terms N
1
N
2
, N
1
N
3
, N
2
N
1
, N
2
N
4
, N
3
N
1
,
N
3
N
4
, N
4
N
2
and N
4
N
3
cancel. When a discontinuity is placed similar to the
one depicted in Figure A.2b and a 2-point Gauss integration scheme is con-
sidered, the coupling is understood to be between node 2 and the remaining
nodes. By using a trapezoidal rule, the pathological coupling is only par-
tially reduced (see the lower part of Figure 2.13 where the oscillations are
less pronounced than in the lower part of Figure 2.12) but it is still present
since the terms N
1
N
2
, N
2
N
1
, N
2
N
3
and N
3
N
2
do not cancel. The cases of
Figures A.2c and A.2d fall under the the same category of Figure A.2a (i.e.
pathological coupling disappears with a trapezoidal rule). In the limit case
of a discontinuity placed along an element side (see Figure A.2e), matrix N
in (A.12) reduces to a 2 4 matrix containing the isoparametric shape func-
tions of the element related to the side on which the discontinuity lies (i.e.
N
1
and N
2
for side 12 in Figure A.2e) and matrix

K in (A.16) is identical to
matrix

K of a conventional line interface element as dened in (A.9). This
identity links the two approaches since conventional interface elements and
PU-based discontinuous elements retain the same block diagonal structure
for the terms related to the discontinuity.
The previous observations on the coupling of degrees of freedom are con-
rmed by an eigenvalue analysis performed on the part

d
N
T
TN d of the
sub-matrix K
bb
(see (2.24)) which directly contributes to the stiffness of the
discontinuity. The interpretation of the eigenvalue analysis is straightfor-
ward and clearly indicates that the reason of the oscillating behaviour could
be indeed rooted in the pathological coupling between degrees of freedom.
Unit values for the length, the surface area and the dummy stiffnesses d
s
and
d
n
have been assumed. The results of the eigenvalue analyses are shown
in Figures A.3 and A.4 for a PU-based discontinuous element with a dis-
continuity placed along the right vertical side. The pathological coupling
of the nodal displacements is evident when a Gauss integration scheme is
used; on the contrary, the pathological coupling disappears with nodal in-
tegration scheme. These results are similar to those reported by Rots and
Schellekens [155]. Consequently, with two-dimensional linear and quadratic
PU-based discontinuous elements and in the particular situation of a dis-
continuity placed along an element side, only the use of nodal integration
scheme for the integration of the terms related to the discontinuity
d
guar-
antees a smooth traction prole for all the values of the dummy stiffness.
Analogous analyses have been performed on a wide range of element and
discontinuity congurations and the results are reported in Table A.1. The
analysis for the 8node quadrilateral element (marked with an asterisk in
table A.1) shows pathological coupling but the results reported in the up-
per part of Figure 2.13 do not show stress oscillations. This can be explained
by further decomposition of the matrix

K (see (A.16)) into a matrix which
contains the contribution to the stiffness matrix of the integration points lo-
cated along the element sides and a matrix which contains the contribution
to the stiffness matrix of the integration point located in the centre of the
discontinuity. Along the line of reasoning pursued by Schellekens and de
Borst [160], it can be shown that the pathological coupling is introduced by
the integration point located in the centre of the discontinuity and that, only
A.3 Concluding remarks 111
Table A.1 Appearance of pathological coupling between degrees of freedom for pos-
sible congurations between a discontinuity and a plane element (YES indicates the
appearance of coupling; the asterisk indicates a particular situation which is explained
on page 110).
Gauss (3 pts)
Simpson
Gauss (2 pts)
Trapezoidal
Yes Yes Yes Yes
Yes Yes No Yes *
Yes Yes Yes Yes
Yes No No No
Gauss (2 pts)
Trapezoidal
Gauss (3 pts)
Simpson
Yes Yes No
Yes Yes Yes
Yes Yes Yes
Yes Yes No
for this special conguration, the element behaves like if no pathological
coupling exists. To summarise, the use of nodal integration is effective in
deleting the pathological coupling between degrees of freedom and in pre-
venting the appearance of oscillations in the stress prole only in particular
situations.
(b)
1.6667
1.6667
1.6667 1.6667
6.6667
(a)
0.8037
1.6667
0.8037
5.5296
1.6667
5.5296
6.6667
Figure A.4 Eigenmodes and corresponding eigenvalues for a quadratic PU-
based discontinuous element with discontinuity placed along an element side
for 3point (a) Gauss and (b) Simpsons integration rule.
A.3 Concluding remarks 113
A.3 Concluding remarks
When partition of unity-based discontinuous elements are used with a
dummy stiffness approach, similar to conventional interface elements,
their performance is severely affected by the position of the discontinuity.
Spurious oscillations in the traction prole that appear when a dummy
stiffness approach is considered (see Chapter 2), are linked to the appear-
ance of pathological coupling between degrees of freedom in the part of the
stiffness matrix responsible for the interface behaviour. It was illustrated
that the pathological coupling between degrees of freedom can be elimi-
nated only in few circumstances. Therefore, partition of unity-based discon-
tinuous elements represent no general remedy against the shortcomings of
conventional interface elements and their use should not be considered in
combination to a dummy stiffness approach.
Appendix B
Some essentials of generalised functions
Some relations involving the Dirac-delta function
d
(x) and the Heaviside
function H
d
(x) are recalled [73, 74, 183] and derived. The body

, depicted
in Figure B.1, is crossed by a discontinuity surface
d
which divides the body
into two sub-domains,
+
and
( =
+

). The Dirac-delta func-

tion is centred at the discontinuity surface
d
and the Heaviside function
is discontinuous across the discontinuity surface
d
(H
d
= 1 if x

+
,
H
d
= 0 if x

). A regular function (x) is dened in the domain and

m is the inward unit normal to
+
. The following relations hold between
the continuous function and the Dirac-delta function
d
:
d
d =
d
d , (B.1)
d
) d =
d
d . (B.2)
The directional derivative of the function in the direction of a generic
unit vector v is dened as D
v
= v. All the properties of standard
m
Figure B.1 Body

crossed by a discontinuity
d
: denition of
+
and
.
116 Appendix B Some essentials of generalised functions
derivation hold and
d
v) d =
D
v
d
d =
d
D
v
d
=
d
v d . (B.3)
The last term in (B.3) is non zero only if the vector v is not perpendicular to
. The gradient of the Heaviside function, with reference to Figure B.1, is
dened as [73]
H
d
=
d
m (B.4)
while the Laplacian is derived as
2
H
d
= (H
d
) = (
d
m) =
d
m+
d
m =
d
m. (B.5)
The Laplacian of the discontinuous scalar function H
d
then yields
2
(H
d
) = H
2
+
d
m+ 2
d
m. (B.6)
Note that the function must be continuously differentiable in (B.2) and
twice continuously differentiable in (B.6)
Appendix C
Constitutive models for softening materials
The phenomenological behaviour of many engineering materials is charac-
terised by a phenomenon known as softening: when a specimen is loaded
beyond its elastic limit, an increase of load-carrying capacity is still possi-
ble up to a certain level of increasing deformation; upon further loading
the load-carrying capacity is progressively lost until failure (this last phe-
nomenon is referred to as softening [108]; see Figure C.1a).
In a continuous setting, softening behaviour can be reproduced by con-
sidering strain-softening constitutive relationships such as the one depicted
in Figure C.1b. However, when a standard inviscid continuum model
is equipped with strain-softening constitutive relationships its numeri-
cal characterisation is not unique as numerical solutions of the boundary
value problem suffer from a severe dependence on the spatial discretisa-
tion [14, 16, 139]. Mesh-dependent results are caused by a change in type of
the governing boundary value problem in the process zonethe problem
becomes ill-posed: elliptic equations for quasi-static problems become hy-
perbolic [36] and hyperbolic equations in wave propagation problems with
classical continuum models become elliptic [95]. Nevertheless, in the frame-
work of the existing phenomenological description of material behaviour,
it is possible to objectively characterise localisation phenomena through the
inclusion, implicitly or explicitly, of a length scale into the eld equations
or the material description (models making use of a length scale are usu-
ally known as regularised models or as localisation limiters). Common
approaches are indicated below.
Fracture energy-based approaches: the softening portion of the stress-
strain curve is adjusted according to the element size to ensure proper
internal energy dissipation during the failure process [19, 138, 140, 151,
153, 154].
Rate-dependent models: these models are based on the inclusion of rate-
dependence in the constitutive equations [53, 87, 101, 118, 163, 164,
179, 180, 189, 190]it could be articial viscosity or material rate-
118 Appendix C Constitutive models for softening materials
strain
s
t
r
e
s
s
(b)
0 0.05 0.1 0.15 0.2
cmsd [mm]
0
10
20
30
40
50
P

[
k
N
]
experiment
(a)
Figure C.1 Modelling structural softening: (a) envelop of load-displacement curves
(adapted from Schlangen [161]) and (b) softening stress-strain constitutive relationship.
dependence such as e.g. Perzyna [136, 137] or Duvaut-Lions [54] vis-
coplasticity.
Higher-order continuum models:
- micro-polar continuum models: additional degrees of freedom,
micro-rotations [45], are added to the continuum description al-
lowing the strains to localise in a band of nite size, dictated by
an internal length scale [31, 35, 50, 113, 115, 179];
- gradient methods: the stresses depend on strain gradients [1, 2, 33,
95, 114, 128, 179, 182];
- non-local continuum models: the stresses at a point depend on the
strains in a non-vanishing region around the point itself (math-
ematically this is similar to gradient methods above) [15, 17, 18,
21, 134, 139, 141].
Although there is no commonly accepted way of treating strain localisa-
tion problems, their correct characterisation is of prime interest since strain
localisation can be usefully interpreted as a failed region. It is therefore im-
portant to understand the limitations of current approaches for the descrip-
tion of failure. To this end, some considerations on numerical modelling of
concrete structures are reported next while a non-local and a rate-dependent
model are contrasted in Section C.2.
C.1 Considerations on numerical modelling of concrete 119
Figure C.2 Schematic representation of material
knowledge when dening a micromechanical consti-
tutive model for bre-reinforced concrete (note that
geometrical and mechanical properties can be dened
for each of the material constituents).
f
i
b
r
e
f
i
b
r
e
a
g
g
r
e
g
a
t
e
a
g
g
r
e
g
a
t
e
c
e
m
e
n
t

p
a
s
t
e
c
e
m
e
n
t

p
a
s
t
e
x
x
x
?
?
C.1 Considerations on numerical modelling of concrete
There are different opinions regarding the strategy to choose when a con-
stitutive model is used. Basically one could consider micromechanical or
macromechanical approaches. Both procedures require a certain degree of
homogenisation which is considered at the end of the analysis in the for-
mer case and at the beginning in the latter, assuming a priori the knowl-
edge of the mutual action of each single material constituents. In both cases,
the level of knowledge regarding material constituents and their mutual in-
teraction is imperfect. Considering e.g. micromechanical modelling, mate-
rial knowledge can be represented in a matrix form (see Figure C.2 for the
micromechanical modelling of bre-reinforced concrete) in which only the
diagonal terms are more or less known. The knowledge regarding some
of the off-diagonal terms is incomplete or neglected because of the dif-
culties in determining the mechanical responses of the single constituents
and their coupling in independent experiments (see e.g. References [92, 97]
where bre-bre interaction and crack-closure effect in bre-reinforced con-
crete is not considered). Similar considerations hold for macromechanical
approaches.
When dening a constitutive model, missing information could be re-
trieved through inverse methods for parameter determination. These meth-
ods make use of specic information derived e.g. from the local behaviour
in terms of stress, strain or displacement elds [49, 63, 67, 68, 80, 102] or by
using the size effect method [40]. Considering only global information in
terms of a load-displacement curve [150] gives rise to an ill-posed problem
and parameters cannot be uniquely identied (see References [40, 63])as
a side remark, it is worth noting that an inverse method which makes use
of the global behaviour is suggested in the FRC design guidelines in Refer-
ences [146, 147] to determine the stress-crack opening relationship for FRC
materials from load-deection test data [184]. Further, when inverse meth-
ods are considered, it is fundamental that the parameters identied by the
inverse procedure are the most basic parameters and do not represent a col-
lection of other factors. Also, it has to be made sure that all the mechanisms
involved in the failure process are clearly identied and that their interac-
tion is well dened. Equally important is the use of error-estimation-driven
adaptive procedure in connection with one of the above inverse methods for
parameter determination to avoid the inuence of nite-element discreti-
sation errors [8, 149]. Unfortunately, inverse methods are prone to fail (i.e.
convergence is difcult to achieve) when many parameters are considered
in the parameter identication procedure. This circumstance poses serious
limitations to their application.
To conclude, there are strong arguments against the use of many of the
models and procedures currently considered in computational mechanics
of quasi-brittle materials (see e.g. References [20, 23, 63, 70, 71, 85, 126, 127,
178, 184, 196, 200] and Sections 1.1 and 3.6.1 and Appendix E). As an exam-
ple, the fracture energy G
f
, a key ingredient in many cohesive-zone models,
seems to be size-dependent [187, 196, 200]. Further, in Reference [126] it is
demonstrated that the length scale l of the non-local damage model pro-
posed by Pijaudier-Cabot and Ba zant [139] is not a material property de-
pending on the maximum aggregate size [22]; rather, it is a material func-
tion depending on the strain and stress eld in the neighbourhood of a
point, especially for points in the fracture process zone [126] (an analogous
conclusion can be found in the review paper by Ba zant et al. [20]). It is also
rather obvious that a model for material characterisation should be valid un-
der complex stress states and loading histories. Although the abundance of
formulations to describe failure in concrete, and their widespread use, none
of the constitutive models derived so far seems to be general enough to deal
Figure C.3 Experimental crack patterns from three single-edge
notched beams (adapted from Schlangen [161]).
C.2 Non-local versus viscous regularisation 121
Figure C.4 Applied load against crack
mouth sliding displacement (cmsd) for the
rate-dependent elastoplastic-damage model
(with Rankine plasticity) and for the im-
plicit gradient-enhanced continuum damage
model (with modied von Mises and Mazars
equivalent strain denitions).
0 0.05 0.1 0.15 0.2
cmsd [mm]
0
10
20
30
40
50
P

[
k
N
]
Rankine
modified von Mises
Mazars (curved crack)
experiment
Mazars (straight curve)
with real-life situations. Even the obvious requirement of obtaining similar
predictions with different constitutive models for the same test is not war-
ranted [71]. This suggests that the mechanics governing failure in concrete
is far from clearly understood.
C.2 Non-local versus viscous regularisation
The implicit gradient-enhanced continuum damage model [134] used in
Chapter 3 and the rate-dependent elastoplastic-damage model described in
Chapter 4 are contrasted through the analysis of a single-edge notched beam
in anti-symmetric four-point-shear loading [161]. This specimen has already
been analysed in Chapter 3 with the continuous-discontinuous approach.
The single-edge notched beam has been often considered in the literature to
test constitutive models with respect to their ability to reproduce the experi-
mental failure pattern [9, 32, 59, 60, 109, 128, 132, 149, 151, 153, 154, 161, 193].
Here this test is considered to highlight some of the characteristics of the
regularised models used in Chapter 3 and 4. In particular, attention will be
drawn to the ability of the models to reproduce the experimental crack pat-
tern reported in Figure C.3.
The simulations are performed considering the nest of the discretisation
used in Section 3.6.3. For the analyses with the gradient-enhanced damage
model, the parameters reported on page 61 are used in combination with
the Mazars (1.3) and modied von Mises (1.4) denitions of the equivalent
strain. For the rate-dependent elastoplastic-damage model, plastic ow is
described by a smoothed Rankine yield criterion [128] with a yield stress
equal to 1.4 MPa under the assumption of a plane stress situation. Softening
parameters are: a = 1 and b = 300 in (4.33) and = 0.9999 and = 170
in (4.3). Other model parameters are taken as follows: N = 1 and relaxation
(a) (b)
(c) (d)
0.1 1
Figure C.5 Final failure pattern with (a) the rate-dependent elastoplastic damage model
(Rankine plasticity) and the gradient-enhanced damage model with (b) modied von Mises
and (c, d) Mazars equivalent strain (close-up of the central part of the beam; the failure prole
in (d) relates to softening parameters different from the ones used in (b) and (c)).
time = 10 s for the viscous regularisation of (4.28), Youngs modulus E =
35000 MPa and Poissons ratio = 0.2. The total cmsd (0.2 mm) has been
applied in 200 steps with cmsd = 0.001 mm. The above model parameters
have been chosen to ensure a qualitative t with the experimental data at
the global level in terms of the initial slope of the load-cmsd curve and the
peak load.
Results of the simulations are reported in Figure C.4 and C.5. The re-
sults reported with the dotted line in Figure C.4 and in Figure C.5d have
been obtained with a modied set of softening parameters as suggested
by Rodrguez-Ferran and Huerta [149] (the softening parameter in (1.8)
is taken equal to 7000 instead of 100; this results in a steeper stress-strain
relationships). Although the global response in terms of applied load ver-
sus cmsd, reported in Figure C.4 for the gradient-enhanced model with the
C.2 Non-local versus viscous regularisation 123
modied von Mises equivalent strain and the rate-dependent elastoplastic-
damage model with Rankine plasticity, is satisfactory, the local response, in
terms of damage proles reported in Figures C.5a-b, indicates that a dif-
ferent failure mode has taken place. In particular, Figures C.5a indicate the
presence of two curved cracks of which the one on the right is the domi-
nant one. All the cracks in Figure C.5a develop simultaneously while of all
the cracks reported in Figure C.5b, only the major curved crack is active un-
til failure; the lateral cracks are active only in the early stages of the failure
process. This latter fact is in agreement with the experimental observation
by Schlangen [161]. However, the damage level of those secondary cracks is
not conrmed by the experimental crack pattern in Figure C.3.
Figure C.5d indicates that the correct failure pattern, reported in Fig-
ure C.3, can be reproduced with a mode I failure criterion (i.e. by using
the Mazars equivalent strain with the gradient-enhanced damage model).
However, softening parameters must be properly chosen to obtain an ade-
quate t of the global response (see Figure C.4 and Rodrguez-Ferran and
Huerta [149] for a similar discussion).
Appendix D
Interpolation requirements for a class of
gradient-enhanced media
Some properties of the gradient-enhanced continuum damage model pro-

posed by Peerlings et al. [134] are discussed by means of a numerical and
theoretical study. In particular, it is shown that gradient-enhanced damage
models [32, 134] derived from integral non-local models [21, 139] do not re-
quire the use of mixed nite-element formulation, as suggested by de Borst
et al. [32], but rather belong to the class of coupled problems. Hence, the
Babu ska-Brezzi condition does not apply to this type of models. Conse-
quently, the use of linear interpolation functions for displacements as well
for non-local equivalent strains yields excellent performance in terms of rate
of convergence and of convergence of the numerical procedure. Stress oscil-
lations may arise for this particular interpolation, but these do not originate
from the variational defects of the discretisation technique. On the contrary,
they can be handled by simple post-processing techniques.
D.1 Governing equations and spatial discretisation
Following standard procedures [134], the governing equations of the im-
plicit gradient-enhanced continuum damage model (see Chapter 3) are cast
in a weighted residual weak form. For completeness, the equilibrium equa-
tions (2.7) (without body forces) and the modied Helmholtz equation (3.4)
are recalled:
= 0 in , e c
2
e = e
l
in . (D.1)
The weak format of the previous governing equations results in
s
w
u
: d =
t
w
u
t d (D.2)
(cw
e
e + w
e
(e e
l
)) d = 0, (D.3)
Based on Reference [169]

126 Appendix D Interpolation requirements for a class of gradient-enhanced media
p, u
u
p
x
x x
x x
e
Figure D.1 Origin of the stress oscillations with linear interpolation functions
for N
u
and N
e
in a one-dimensional tension test.
where w
u
and w
e
are test functions for the displacement and the non-local
equivalent strain eld. The weak governing equations are discretised, at el-
ement level, using a Bubnov-Galerkin approach. The displacement and the
non-local equivalent strain elds are discretised in each element by
u
h
= N
u
a, e
h
= N
e
p, (D.4)
where the interpolation functions N
u
and N
e
link the continuous-valued dis-
placement and non-local equivalent strain elds to their discretised nodal
components in the vectors a and p. Details on the numerical formulation
can be found in Reference [134].
A question now arises with respect to the choice of the order of the inter-
polation functions N
u
and N
e
. Since the non-local equivalent strain eld is
a scalar function of the strain tensor, an obvious choice would be to take the
order of the interpolation for the displacement one order higher than for the
non-local equivalent strain eld (i.e. N
u
quadratic and N
e
linear or N
u
cubic
and N
e
quadratic, etc.). Using the same order for the interpolation functions
(linear for instance) could lead to some misunderstandings in the interpre-
tation of the results of the numerical simulation. The issue was illustrated
with reference to a one-dimensional tension test with piece-wise linear in-
D.2 Element performance 127
Figure D.2 Linear-linear (left) and quadratic-linear
(right) bar element (white and black circles indi-
cate displacement and non-local equivalent strain
degrees of freedom, respectively).
terpolation of u and e by Peerlings [131]. Due to the linearity of the displace-
ment eld within each element, the local strain eld (equal to the equivalent
strain eld in this case) is piecewise constant within each element. However,
the damage, depending on the non-local equivalent strain, is close to linear
within the element. The combination of the linearly varying damage and of
the constant strain within each element through = (1 ) yields the os-
cillatory behaviour in the stress depicted in Figure D.1. However, it must be
realised that the stress eld depicted in Figure D.1 is a valid solution of the
weak equilibrium problem and it is accompanied by correct displacement,
non-local equivalent strain and damage proles and by quadratic conver-
gence of the numerical algorithm.
When a strain-based stress update algorithm is used [134], the oscillations
in the stress prole do not constitute a problem and this apparent anomaly
can be easily solved at the post-processor level, e.g. sampling the stress at
a point located in the middle of the element in case of a one-dimensional
setting. A heuristic way to avoid oscillations has been suggested by Peer-
lings [131], in which the use of linear interpolation functions for N
u
and N
e
and taking the damage variable uniform in each element ensures a smooth
stress prole. Note that this technique can also be used for stress-based up-
date algorithms.
D.2 Element performance
The effect of the interpolation function on the results is analysed here by
using a one-dimensional bar which is clamped at one end and with a pre-
scribed displacement incrementally applied at the free end. The bar has a
length of 100 mm, a cross sectional area of 10 mm
2
and a geometrical imper-
fection in the middle section (10% reduction of the cross sectional area for
10% of the length of the bar). Youngs modulus is equal to 20000 MPa. The
linear softening damage evolution law(1.6) with
0
= 0.001 and
c
= 0.0125
has been used. The gradient parameter is set to c = 1 mm
2
. In the following,
the use of linear interpolation functions for N
u
and N
e
is indicated with the
wording linear-linear while the use of quadratic interpolation functions for
0 0.01 0.02 0.03 0.04 0.05
displacement [mm]
0
5
10
15
20
l
o
a
d

[
N
]
linearlinear
quadraticlinear
962 dofs
3842 dofs
Figure D.3 Load-displacement curve for different interpolation schemes.
10
10
10
10
10
6
5
4
3
2
10
2
10
3
10
4
linearlinear
quadraticlinear
p p
| | a a
| |
ref
ref
number of degrees of freedom
e
r
r
o
r

i
n

s
o
l
u
t
i
o
n

n
o
r
m
nonlocal equivalent strain norm
displacement norm
Figure D.4 Rate of convergence for different interpolation schemes.
D.2 Element performance 129
0
20
40
60
80
100
0
0.005
0.010
0.015
0.020
0.025
0.030
0
0.5
1
1.5
2
2.5
s
t
r
e
s
s

[
M
P
a
]
end displacement [mm]
x [mm]
Figure D.5 Evolution of the stress prole for the linear-linear discretisation.
0
20
40
60
80
100
0
0.005
0.010
0.015
0.020
0.025
0.030
0
0.5
1
1.5
2
2.5
x [mm]
s
t
r
e
s
s

[
M
P
a
]
end displacement [mm]
Figure D.6 Evolution of the stress prole for the quadratic-linear discretisation.
N
u
and of linear interpolation functions for N
e
is indicated with quadratic-
linear. The comparison has been done with the same number of degrees
of freedom and for two nite-element discretisations. In the rst discretisa-
tion, the total number of dofs is 962 (480 linear-linear elements versus 320
quadratic-linear elements) and in the second it is 3842 (1920 linear-linear
elements versus 1280 quadratic-linear elements). As a measure of the accu-
racy of the approximation (see Figure D.3) the difference in solution norms
[a
re f
a[ and [p
re f
p[ between a reference solution (the solution obtained
with a quadratic-linear bar element with 7682 dofs) and the solutions ob-
tained in various computations have been compared. Figure D.4 shows that
the use of a higher-order interpolation for the displacement eld with re-
spect to the non-local equivalent strain leads to a smaller computational ef-
fort (at equivalent error level). However, the rate of convergence is the same
for the two interpolations.
The curves in Figure D.4 have been obtained imposing a nal displace-
ment of 0.425 mm at the free end of the bar in displacement control and
evaluating the difference in solution norm in 81 equally distributed sample
points. The stress prole for different interpolation orders is shown in Fig-
ures D.5 and D.6. Figure D.5 clearly shows the stress oscillations with linear-
linear interpolation functions while the use of a quadratic-linear interpola-
tion function restores the correct stress prole (see Figure D.6). The compar-
ison has been done with 120 linear-linear bar elements and 80 quadratic-
linear bar elements for a given displacement of 0.03 mm at the free end of
the bar. Note that the proles were constructed by connecting the values
sampled at Gauss points. The real proles resemble the one in Figure D.1. It
is worthwhile noting that oscillations remain local (they only occur where
the damage varies quickly) and that the amplitude of oscillations is and
remains limited. It is therefore questionable how serious these oscillations
should be taken. In the next section some theoretical arguments will be
given to support the idea of using interpolation functions of the same or-
der for N
u
and N
e
.
D.3 About terminology: mixed method versus coupled problem
In order to justify the use of different orders of interpolation related to the
discretisation of the displacement and the non-local equivalent strain elds,
it has been argued by de Borst et al. [32] that gradient-enhanced damage
models [32, 134] derived from integral non-local models [21, 139] requires
D.3 About terminology: mixed method versus coupled problem 131
the use of mixed nite-element formulations. Broadly speaking, a mixed
method for a second-order elliptic boundary value problem is a method
in which one or more auxiliary variables are introduced, the second-order
problem is rewritten as a rst-order system of equations which is cast into
a variational form and then discretised within a Bubnov-Galerkin method,
thus obtaining direct approximations to both the original and the auxiliary
variables.
For instance, considering the eld equations of the equilibrium problem
with a homogeneous Dirichlet boundary condition [142]
_
+f = (D
e
:) +f = (D
e
:
s
u) +f = 0 in
u = 0 on ,
(D.5)
where f are the body forces and D
e
is the fourth-order linear-elastic consti-
tutive tensor, its solution u is given by the minimisation, with respect to a
suitable test function v belonging to the Sobolev space H
1
0
(), of the total
potential energy (v):
(v) =
1
2
:D
e
: d
fv d
=
1
2
s
v:D
e
:
s
v d
fv d. (D.6)
By introduction of the quantity r = D
e
:
s
u, where u is the solution to the
equilibrium problem stated above, the original problem can be restated with
respect to the new couple of unknowns u and r. The problem has been re-
stated into the rst-order system
_
_
_
r = D
e
:
s
u in
r +f = 0 in
u = 0 on
(D.7)
and its solution r is given by the minimisation of the complementary energy
1(s) =
1
2
s: (D
e
)
1
:s d, (D.8)
where s is a suitable test function which belongs to the Sobolev space
L
2
() and satises s + f = 0 in and whose divergence belongs to
L
2
(). The solution r is related to the solution u of (D.5) satisfying (D.7)
3
through (D.7)
1
[142].
The restated second-order equilibrium problem in (D.7) is more appeal-
ing than the original one with respect to those situations in which an ap-
proximation of r has to be found which is at least as accurate as that of the
approximation v. Starting from the minimisation of the total complemen-
tary energy, it is possible to build a discretised nite-element solution to the
problem. However, this is a difcult task because of the fullment of the con-
straint r +f = 0. These problems can be solved by making use of a mixed
method. These methods are based on the introduction of a Lagrangian mul-
tiplier to relax the constraint r +f = 0. The Lagrangian functional can be
formulated as
L(s, v) = 1(s) +
(s +f) v d. (D.9)
The solution to the problem is then to minimise over s and maximise over
v, that is to search for the couple s and v which is the saddle-point of
L(, ). The existence and uniqueness of a solution to the discretised form
of the saddle-point problem of L(, ) is assured by a compatibility condi-
tion, which links the order of interpolation for the unknowns (this condi-
tion is known in literature as inf-sup condition or Babu ska-Brezzi condi-
tion [12, 37, 79]), and by the ellipticity condition [38, 142]. Moreover, con-
vergence of the numerical implementation cannot be assured if the Babu ska-
Brezzi condition is not satised [13].
From the above denition, it is clear that the problem dened by the
two equations (D.1)
1
and (D.1)
2
of the gradient-enhanced description of the
continuum is not of the mixed type, since the diffusion equation can not
be retrieved from the equilibrium equation: (D.1)
1
and (D.1)
2
form a cou-
pled problem in which neither domain or set of dependent variables can be
solved if separated from the other [202, vol 1, p 542]. Therefore, the compati-
bility condition is not necessary and, as a consequence, the order of interpo-
lation of the enhanced strain eld does not have to be related to the one of
the displacement eld. Stress oscillations that appear in a linear-linear inter-
polated element can be solved by means of post-processing techniques. For
the class of gradient-enhanced damage models derived from integral non-
local models [21, 139], it is always possible to formulate the weak form in
such a way that it resembles a coupled formulation. The use of mixed for-
mulations is also possible but leads to unnecessary complications (see e.g.
Comi [43] where, although in a different gradient-enhanced damage for-
mulation, a mixed formulation is used). Further, according to Zienkiewicz
and Taylor [202, vol 1, p 542], in coupled systems the solution of any single
D.4 Concluding remarks 133
system is a well-posed problem and is possible when the variables corre-
sponding to the other system are prescribed. This is not always the case in
mixed formulations: due to the well-posed nature of each of the two sys-
tems in the gradient-enhanced damage models proposed by Peerlings et
al. [134], namely a set of equilibrium equations (see (D.1)
1
) and a modied
Helmholtz equation (see (D.1)
2
), well-posedness for gradient-enhanced con-
tinuum damage formulations is guaranteed.
D.4 Concluding remarks
Gradient models derived from non-local formulations can be classied as
coupled problems, not requiring mixed nite-element formulations in the
classical sense. Consequently, the inf-sup or Babu ska-Brezzi condition is not
relevant for this class of models. In practice, this means that linear-linear in-
terpolation functions can be used, instead of quadratic-linear interpolation
functions. This can be advantageous when quadratic nite-element meshes
are not available or difcult to construct. The use of linear-linear elements
gives the same convergence rate as the use of quadratic-linear elements.
Stress oscillations that appear in a linear-linear interpolated element can be
solved by means of post-processing techniques. It is emphasised that the use
of higher order functions is not related to the well-posedness of the problem
but rather to the occurrence of stress oscillations: stress oscillations are not
related to the Babu ska-Brezzi condition.
Appendix E
Incorrect failure characterisation in non-local media
Correct failure characterisation, in terms of damage initiation and propa-

gation, is a fundamental property of any sound model. Considering failure
initiation, a quantitative discrepancy between the response of a regularised
and a standard continuum is accepted but a qualitative resemblance must
be maintained. It is obvious that a wrong prediction of the correct location
or moment of initiation may lead to a misrepresentation of the failure mode
and therefore of the failure load. Failure propagation is as important as fail-
ure initiation and, in a continuous failure representation, gives an indication
of the failure mechanism. A realistic failure propagation is of paramount
importance when the assessment of a constitutive model is concerned. In
the class of non-local models analysed here [55, 129, 134, 139], only the
dissipation-driving variable is given a non-local character.
In this appendix it is shown that the choice of a non-local quantity as
damage-driving quantity produces non-physical damage initiation away
from the crack tip in mode I problems and a wrong failure pattern in shear
band problems. Damage initiation in a class of non-local damage models,
within an integral [139] as well as a differential [134] formulation, is anal-
ysed. The ndings of this study are not limited to damage mechanics but
extend easily to other dissipation mechanisms, e.g. plasticity [55, 129], if a
similar form of regularisation is employed.
E.1 Some basic notions
In the remainder of this appendix, use is made of some notions from Chap-
ter 1 which are briey recalled for convenience. In particular, the non-local
equivalent strain e is dened as [139]
e(x) =
(y; x) e
l
(y) d(y)
(y; x) d(y)
, (E.1)
Based on Reference [170]

136 Appendix E Incorrect failure characterisation in non-local media
u
u
4h
h
b a
p
2h
y
x
o
b a
e
e
(a) (b)
l
Figure E.1 Compact tension specimen: (a) geometry and bound-
ary conditions and (b) local (e
l
) and non-local (e) equivalent strain
eld along line ab.
with the homogeneous and isotropic weight function
() =
1
2l
2
exp
_
2
2l
2
_
in R
2
(E.2)
which depends on the length scale l and on the distance between the
source point y and the receiving point x. With these denitions, the de-
nominator in (E.1) sums to unity for an innite and regular domain. In an
approximate differential version of the non-local model (implicit gradient-
enhanced damage model [134]), (E.1) is expanded in a Taylors series around
x which yields, after some manipulations and with the use of (E.2), the mod-
E.2 Damage characterisation in mode I problems 137
ied Helmholtz equation
e
1
2
l
2
2
e = e
l
in . (E.3)
Equation (E.3) is complemented by the homogeneous natural boundary con-
ditions
e n = 0 on , (E.4)
where n is the outward unit normal at the boundary of . The equivalence
of (E.1)-(E.2) and (E.3)-(E.4) has been discussed by Peerlings et al. [135].
E.2 Damage characterisation in mode I problems
Proper failure characterisation relies on correct failure initiation. In quasi-
brittle failure analyses of notched specimens, experimental evidence shows
that cracks propagate from the notch [108]. Proper modelling of quasi-brittle
material behaviour must reproduce this phenomenon.
Damage initiation in mode I is analysed by means of the compact ten-
sion specimen with a pre-existing crack of length h depicted in Figure E.1a.
Numerical analyses showed that the elastic contour plots of the non-local
damage-driving quantity e is maximum at some distance from the crack
tip. More specically, the maximum of the non-local equivalent strain e was
found along the line ab, as qualitatively depicted in Figure E.1b (analytical
result) and in Figure E.2b (numerical result), and not at the crack tip. Thus,
damage initiation is predicted inside the specimen, rather than at the crack
tip. In Figure E.1b, the proles of the local equivalent strain e
l
and of the
non-local equivalent strain ethe latter obtained through (E.1)are plotted
along the line ab. First, consider the prole of e
l
. The local equivalent strain,
as predicted by Grifths theory, equals 0 from point a to the crack tip. At
the crack tip it is innite, after which it decays monotonically to its nite
value at point b. The prole of the non-local counterpart e differs from the
prole of e
l
in that (i) no singularity is present (as already noted by Peer-
lings et al. [135]), and (ii) the maximum occurs not at the crack tip but at
some point between the crack tip and point b along the crack line. Note that
the crack is discretised as a set of zero measure in this example and, as such,
along line ab, it does not inuence the integral in the denominator in (E.1)
the denominator in (E.1) is the normalising factor in the non-local averaging
(b) (a)
min max
Figure E.2 Qualitative contour plot of (a) the local elastic equiv-
alent strain e
l
and (b) the non-local elastic equivalent strain e: due
to the non-local averaging the maximum of the non-local equiv-
alent strain shifts away from the tip.
near free boundaries. In other words, for all points along line ab that are rea-
sonably far from the edges of the specimen, the denominator of (E.1) yields
the same value, therefore the shift of maximum from e
l
to e is not the result
of a varying averaging volume. The shift of the maximum of the non-local
equivalent strain e from the crack tip is a phenomenon which is independent
of the stress situation (plane stress/plane strain) and of the choice of the lo-
cal equivalent strain denition as will be illustrated in Section E.2.2. Indeed,
this can be explained by considering that non-local averaging of the unsym-
metrical local strain eld e
l
is performed through a symmetric function .
The shift of the maximum away from the crack tip will be proven analyt-
ically and illustrated numerically in Sections E.2.1 and E.2.2, respectively.
E.2.1 Analytical considerations
In the ideal situation of a planar crack in an innite plate loaded in mode I,
such as the one depicted in Figure E.3, the linear elastic stress eld is singu-
lar at the crack tip [58]. Analytical manipulations under the assumption of a
x
2
r
o R, x
1
Figure E.3 Linear elastic crack problem in an innite domain.
plane stress situation yield the Cartesian strains at a distance r and an angle
from the crack tip o:
11
(r, ) =
K
I
E
2r
cos
1
2
_
1 (1 +) sin
1
2
sin
3
2
_
(E.5)
22
(r, ) =
K
I
E
2r
cos
1
2
_
1 + (1 +) sin
1
2
sin
3
2
_
(E.6)
12
(r, ) =
K
I
E
2r
(1 +) cos
1
2
sin
1
2
cos
3
2
(E.7)
33
(r, ) =

1
(
11
+
22
) =
2K
I
E
2r
cos
1
2
, (E.8)
fromwhich the local equivalent strain according to the von Mises equivalent
strain
e
l
=
1
1 +
_
3J
2
(E.9)
reads
e
l
(r, ) =
2K
I
4E
r
_
(1 + cos) (5 3 cos). (E.10)
In the above relations, E is the Youngs modulus, the Poissons ratio, J
2
is
the second invariant of the deviatoric strain tensor and K
I
the mode I stress
intensity factor.
Next, the non-local equivalent strain e is investigated. First, its value at the
crack tip is considered. Second, it will be shown that the crack tip value e
o
of e is not maximum since larger values of e occur at locations away from
the crack tip. To demonstrate that damage initiation is incorrectly predicted
with the class of non-local models analysed here, it is necessary and suf-
cient to demonstrate that the absolute maximum of the damage-driving
quantity does not occur at the crack tip.
Crack tip value of non-local equivalent strain. Following Peerlings et
al. [135], it can be demonstrated that the non-local equivalent strain e has a
nite value e
o
at the crack tip. To this end, the two-dimensional normalised
isotropic Gaussian weight function in (E.2) along with the local equivalent
strain expression in (E.10) is substituted in the non-local averaging of (E.1)
to give
e
o
=
K
I
8
3/2
El
2
0
1
r
exp
_
r
2
2l
2
_
r dr
+
_
2 (1 + cos) (5 3 cos) d. (E.11)
The rst integral in (E.11) can be rewritten as
r exp
_
r
2
2l
2
_
dr =
l
3/2
_
3
4
_
4
2
, (E.12)
where () indicates the gamma function; the second integral in (E.11)
yields:
+
_
2 (1 + cos) (5 3 cos) d = 8
2
2
3
6 ln
_
7 4
3
_
. (E.13)
Thus,
e
o
=
K
I
E
l
4
18
_
3
4
_
12
3/2
_
4
3 ln
_
7 4
3
__
0.3612
K
I
E
l
, (E.14)
which proves that the non-local equivalent strain is not singular at the crack
tip (for l ,= 0). An analogous result was presented in References [131, 135].
One of the conclusions reported by Peerlings [131] is that the absolute max-
imum of the non-local equivalent strain is at the crack tip. This issue is dis-
cussed next.
Non-local equivalent strain along the crack line. The search for larger
values of the non-local equivalent strain is restricted to the points along the
crack line ab (see Figure E.1b). A point p is considered along line ab whereby
the distance from the crack tip to p is denoted by R. The weight function in
(E.2) is written for a point p as
p
(r, , R) =
1
2l
2
exp
_
(rcos R)
2
+ (rsin)
2
2l
2
_
, (E.15)
and with this expression the non-local equivalent strain at a distance R from
the crack tip along the crack line reads
e
R
=
0
+
p
(r, , R) e
l
(r, ) r d dr (E.16)
for which a closed form solution could not be obtained. Numerical evalua-
tion of the integral in (E.16), for a given R, indicates that the maximum of
the non-local equivalent strain is not positioned at the crack tip as depicted
in Figure E.4a. The linear dependence of the position of the maximum of the
non-local equivalent strain on the length scale l is depicted in Figure E.4b
which shows that the non-local equivalent strain is maximum at the crack
tip only for l = 0 mm, i.e. only for a local damage model. These results ex-
tend to a nite specimen width if the effect of a nite geometry is reected
in the stress intensity factor K
I
.
Note that, in general, the use of non-local averaging of eld quantities
with isotropic weight functions results in a modication of failure charac-
terisation. In the class of non-local elasticity models proposed by Eringen
et al. [56], the stress eld value at the crack tip is nite but, similar to the
non-local damage model considered here, its maximum occurs at some dis-
tance from the crack tip along the crack line. The modication of the weight
function in order to preserve the qualitative character of the eld to be
averaged (the local eld) is problematic and leads to tailor-made solutions
which are not recommended when a differential version of the non-local
damage model is considered.
E.2.2 Numerical analysis
The compact tension specimen depicted in Figure E.1a has been numerically
analysed by using an integral and a differential non-local damage model
with the nite-element method. In the numerical simulations only the upper
part of the specimen has been discretised due to symmetry and the load has
0 5 10
R [mm]
0 1 2 3 4
l [mm]
[mm]
0
0.1
0.2
0.3
0.4
0
0.5
e
R
1
2
3
4
R
max
R
max
(a) (b)
Figure E.4 Prole of the non-local equivalent strain at distance R from the tip for
l = 1 mm (a) and (b) linear dependence of the position of the maximum of the non-
local equivalent strain R
max
on the length scale l for unit values of E and K
I
.
been applied via an imposed displacement. The following model parame-
ters have been adopted for the simulation: Youngs modulus E = 1000 MPa;
Poissons ratio = 0; exponential damage evolution law
=
_
0 if
0
1
_
1 +e
(
0
)
_
if >
0
,
(E.17)
with threshold of damage initiation
0
= 0.0003; softening parameters =
0.99 and = 1000; length scale l = 0.2 mm. The equivalent strain denition
e
l
=
_
3
i=1
i
)
2
, (E.18)
based on the positive principal strain components
i
) = (
i
+[
i
[)/2 where
i
are the principal strains, has been used. The height of the specimen has
been taken as 4h = 2 mm. The mesh used for the simulations has been
chosen such that a sufcient resolution of the non-local eld is obtained.
Contour plots of the non-local equivalent strain at the onset of damage
initiation are reported in Figure E.5. Clearly, the maximum of the non-local
equivalent strain has shifted, both for the integral model and the differen-
tial model. Due to the shifting, damage is expected to initiate, wrongly, away
from the crack tip. However, as depicted in Figure E.6, damage proles close
to global failure of the specimen give no indication of the incorrect damage
(a) (b)
x
0 0.25 0.5 0.75 1
0
0.25
0.5
0.75
1
x
0 0.25 0.5 0.75 1
0
0.25
0.5
0.75
1
y y
0.0003
2E05
e
Figure E.5 Contour plot of the non-local equivalent strain for (a) the integral non-local
damage and (b) the differential non-local damage model at the onset of damage initiation,
i. e. in the elastic stage (measures in mm; crack tip at 0.5 mm).
(a) (b)
x
y
0 0.25 0.5 0.75 1
0
0.25
0.5
0.75
1
y
0 0.25 0.5 0.75 1
0
0.25
0.5
0.75
1
x
1
0
Figure E.6 Contour plot of the damage eld for (a) the integral non-local damage and
(b) the differential non-local damage model close to failure of the specimen (measures in
mm; crack tip at 0.5 mm).
P, v
(a)
P, v
2h
h
(b)
P, v
Figure E.7 Geometry and boundary conditions for the specimen in biaxial
compression: (a) full specimen and (b) half specimen (h = 60 mm; the shaded
part indicates the imperfection; imperfection size in (a) is h/10 h/10).
initiation. Experience with the differential version of the non-local damage
model indicates that this is a common situation in mode I problems and
that failure characterisation close to failure is quite similar to the ones ob-
tained with other constitutive models. Consequently, the shift of the maxi-
mum of the non-local equivalent strain away from the crack tip can be con-
sidered harmless as long as the nal failure characterisation is concerned.
However, it must be realised that the use of a non-local dissipation-driving
variable leads to a non-physical damage initiation. The shift of the maxi-
mum of the non-local equivalent strain away from the crack tip is present,
although less evident, also in case of cracks or notches modelled as strongly
non-convex entities with non-zero volume, i.e. when there are no strain sin-
gularities.
E.3 Damage characterisation in shear band problems 145
(a) (b) (c)
p p p
Figure E.8 Schematic formation of a shear band: (a) initiation and (b) expansion of the
plastic zone and (c) further localisation/expansion within/of the plastic zone (strain
localisation is triggered by a weak spot in the left bottom corner).
E.3 Damage characterisation in shear band problems
The correct determination of shear bands is of prime interest and it is di-
rectly linked to the occurrence of possible collapse mechanisms in many
engineering problems. Specimens under compressive loading are usually
characterised by the formation of shear bands whose inclination can be de-
termined analytically. Results obtained within the ow theory of plastic-
ity [156] have been extended to scalar damage models by Rizzi et al. [148]
and Carol and Willam [41] and apply to an innite geometry for a standard
(i.e. not regularised) medium. Their results have been derived for a specic
choice of the equivalent strain denition and their generalisation to other
equivalent strain denitions, although possible in principle, is not within
the scope of this study and therefore is not considered here. In what fol-
lows, it is illustrated that non-local regularisation techniques signicantly
alter failure propagation during strain localisation.
To illustrate the problem, shear band simulations under a plane stress and
a plane strain condition have been performed with the gradient-enhanced
continuum damage model for the two-dimensional specimen depicted in
Figure E.7. Both geometries were considered to assess the validity of the
boundary condition (E.4) which was indeed veried: both geometries gave
identical results and in the following only the geometry in Figure E.7b
is considered. A detailed analysis regarding the treatment of the non-
0 0.02 0.04 0.06 0.08
v [mm]
0
50
100
150
200
P

[
N
]
Figure E.9 Load-displacement curve for
the shear band problem (relevant elds
corresponding to the white circles are de-
picted in Figures E.10 and E.11).
local equivalent strain at the boundaries for a mode I problem was given
by Peerlings et al. [135]. In numerical simulations of quasi-static shear
band formation under compressive loading, shear bands are usually trig-
gered by an imperfection (placed on the left bottom corner for the spec-
imen in Figure E.7b). After the shear band has been initiated, expansion
of the plastic zone and further localisation within the plastic zone is ob-
served [199] (for a schematic initiation and evolution of the shear band see
Figure E.8). Shear bands are characterised by their stationary nature in the
sense that their position is determined after their formation (see e.g. Ref-
erence [120] and references herein for experimental shear bands and Ref-
erences [29, 30, 62, 119, 125, 179, 182, 186, 194, 199] for some numerical re-
sults). The inclination angle that the shear band forms with the horizon-
tal axis is determined mainly by assumptions related to the constitutive
model, to the Poissons ratio and to the plane stress or plane strain con-
dition [41, 148, 156, 179, 182] while the width of the shear band, in a contin-
uous description of failure, is dictated by the length scale (i.e. the larger the
length scale, the wider the band width).
In the numerical analyses, the material has been characterised by a
Youngs modulus E = 20000, a Poissons ration = 0.2, the exponen-
tial softening law (E.17) with
0
= 0.0001, = 0.99 and = 300 and
the von Mises equivalent strain (E.9). The load has been applied via dis-
placement control. The imperfection has been given a reduced value of
0
(
0
= 0.00005) and the mesh density has been chosen to ensure a sufcient
resolution of the non-local eld. To begin with, the evolution of the shear
band, in terms of non-local equivalent strain and damage elds has been
analysed for the specimen in Figure E.7b with a length scale l = 2 mm un-
der a plane strain condition. Results depicted in Figures E.10 and E.11 relate
v=0.009 mm v=0.013 mm v=0.016 mm
v=0.021 mm v=0.030 mm v=0.038 mm v=0.080 mm
v=0.006 mm
0.03 0.0003
e
Figure E.10 Shear band evolution: contour plots of the non-local equivalent strain eld (see
Figure E.9).
v=0.006 mm v=0.009 mm v=0.013 mm v=0.016 mm
v=0.021 mm v=0.030 mm v=0.038 mm v=0.080 mm
1.0 0.5
Figure E.11 Shear band evolution: contour plots of the damage eld (see Figure E.9).
1.0 0.5
0.03 0.0003
e
Figure E.12 Shear band close to failure: contour plots of the non-local equivalent strain eld
(top) and of the damage eld (bottom) for l = 1 mm (left) and l = 2 mm (right) for plane
stress (odd columns) and plane strain (even columns).
to the load-displacement diagram in Figure E.9, where the applied load P
is plotted against the absolute value v of the vertical displacement. In the
contour plots in Figures E.10 and E.11, only values larger than the threshold
in the respective legends have been plotted. It is clear that the shear band
migrates from the weak spot, where it was initiated, to the opposite side of
the specimen along the horizontal boundary as depicted in Figures E.10 and
E.11. Similar results have been reported by Engelen et al. [55] and Pamin et
al. [129]. It is stressed that the migrating shear band is not the product of
an improper treatment of the boundary conditions since, as already stated
before, the geometries reported in Figure E.7 give identical results. This par-
ticular appearance of the shear band is simply due to a wrong prediction of
the positioning of localised shearing and has the same nature of the shift of
the maximum of the non-local equivalent strain in mode I problems.
The effect of a larger length scale is reported in Figure E.12 together with a
comparison between plane stress and plane strain conditions close to failure.
Similar to shear bands in a plasticity context [156] and as reported by Carol
and Willam [41], the only noticeable difference between plane stress and
plane strain resides in a different inclination of the shear band with respect
to the horizonal axis which does not correspond to the numerical results in
Figure E.12. Further, with an increasing non-local effect a wider shear band
width is expected, while it is also noted that a more pronounced shift of the
shear band takes place. In one case (plane stress situation with l = 2 mm)
the non-local equivalent strain eld mimics a mode I situation and almost
half of the specimen is damaged, which is not realistic.
0 0.02 0.04 0.06 0.08
v [mm]
0
50
100
150
200
P

[
N
]
(a)
0 0.05 0.1 0.15 0.2
v [mm]
0
50
100
150
200
P

[
N
]
(b)
Figure E.13 Load-displacement curve for the shear band problem with the rate-dependent
elastoplastic-damage model (see Chapter 4): relaxation time (a) = 4 s and (b) = 8 s
(relevant elds corresponding to the white circles are depicted in Figures E.14 and E.15 for
= 4 s and = 8 s, respectively).
The shear band problem was also analysed with the viscous model de-
scribed in Chapter 4. The simulations were performed with a von Mises
plane stress rate-dependent damage-elastoplasticity for two values of the
relaxation time in (4.28). Other model parameters are: Youngs modulus
E = 20000, Poissons ratio = 0.2, N = 1 in (4.29), softening parameters
a = 1 and b = 200 in (1.21) and = 0.99 and = 300 in (4.3) and
yield stress equal to 2 MPa with a reduction of 10% in the weak spot. The
total displacement was applied at constant strain rate (for = 4 s the -
nal displacement (0.08 mm) was applied in 160 steps while for = 8 s the
nal displacement (0.2 mm) was applied in 400 steps). Results reported in
Figures E.14 and E.15 relate to the load-displacement diagram reported in
Figure E.13. The shear band evolves correctly, is stationary and its width is
set by the relaxation time [179]. The slight shift of the shear band is due to
the rectangular geometry of the imperfection and is not inuenced by the
relaxation time.
1.0 0.5
v=0.028 mm v=0.043 mm v=0.062 mm v=0.080 mm
Figure E.14 Shear band evolution: contour plots of the damage eld with the rate-
dependent elastoplastic-damage model (see Chapter 4) for = 4 s (see Figure E.13a).
1.0 0.5
v=0.050 mm v=0.075 mm v=0.099 mm v=0.200 mm
Figure E.15 Shear band evolution: contour plots of the damage eld with the rate-
dependent elastoplastic-damage model (see Chapter 4) for = 8 s (see Figure E.13b).
E.4 Concluding remarks
When non-local averaging is considered as a regularisation technique, the
use of a non-local variable as degradation-driving variable induces incorrect
failure initiation and propagationthis issue has been investigated through
analytical and numerical analyses for mode I and shear band problems. The
particular choice of the examples in mode I failure was aimed at empha-
sising that the standard formulation of non-local damage model, either in
integral [139] or in differential [134] format, is unsuited to correctly describe
damage initiation when degradation stems from a strong inhomogeneity
of the strain eld. It is stressed that this phenomenon is not caused, in the
present case, by boundary effects on the non-local averaging. Analogous
results can be found in cases with strong gradients of the quantity to be
averaged due to e.g. concave boundaries, concentrated loads and material
E.4 Concluding remarks 151
inhomogeneities. Although the shift of the maximum of the dissipation-
driving variable may not alter the nal failure representation in mode I dom-
inated problems, it does affect the transition from continuous to continuous-
discontinuous failure in a gradient-enhanced damage model as discussed in
Chapters 3 and 4. The numerical study of a shear band problem illustrated
that the non-local averaging is responsible for the non-stationary shear band
which results in an unrealistic failure pattern. This has been studied for var-
ious length scale values under plane stress and plane strain conditions. The
shear band problem was also analysed with a rate-dependent elastoplastic-
damage model which gave correct results.
To summarise, constitutive models based on a non-local dissipation-
driving variable may lead to incorrect failure initiation and propagation in
arbitrary loading scenarios due to a fundamental aw in damage character-
isation. Their use is acceptable with some reserve only in mode I problems.
References
The number printed in italics indicates the page where the publication is
cited.
[1] E. C. Aifantis. On the microstructural origin of certain inelastic models. Journal of Engineering
Materials Technology, 106:326330, 1984. {118}
[2] E. C. Aifantis. The physics of plastic deformation. International Journal of Plasticity, 3:211247,
1987. {118}
[3] P. C. Atcin. High Performance Concrete. E & FN Spon, London, 1998. {4}
[4] J. Alfaiate, A. Simone, and L. J. Sluys. Non-homogeneous displacement jumps in strong embed-
ded discontinuities. International Journal of Solids and Structures, 2002. In press. {16, 19}
[5] J. Alfaiate, A. Simone, and L. J. Sluys. A new approach to strong embedded discontinuities. In
N. Bi cani c, R. de Borst, H. A. Mang, and G. Meschke, editors, Computational Modelling of Concrete
Structures EURO-C 2003, pages 3341. Swets & Zeitlinger, Lisse, 17-20 March 2003. {16}
[6] G. Alfano and M. A. Criseld. Finite element interface models for the delamination analysis of
laminated composites: mechanical and computational issues. International Journal for Numerical
Methods in Engineering, 50(7):17011736, 2001. {13}
[7] F. Armero and K. Garikipati. An analysis of strong discontinuities in multiplicative nite strain
plasticity and their relation with the numerical simulation of strain localization. International
Journal of Solids and Structures, 33(2022):28632885, 1996. {13, 16}
[8] H. Askes. Advanced Spatial Discretisation Strategies for Localised Failure. PhDthesis, Delft University
of Technology, 2000. {44, 120}
[9] H. Askes and L. J. Sluys. Remeshing strategies for adaptive ALE analysis of strain localisation.
European Journal of Mechanics A/Solids, 19:447467, 1999. {121}
[10] ASTM. Metals Test Methods and Analytical Procedures, Designation: E 39990; E 64799. In
Annual Book of ASTM Standards, volume 03.01. American Society for Testing and Materials, West
Conshohocken, Pennsylvania, 2000. {56}
[11] M. Audi, J. Kr cek, J. Zeman, and M.

Sejnoha. Constant strain triangular element with embedded
discontinuity based on partition of unity. Building Research Journal, 2003. Submitted. {29}
[12] I. Babu ska. The nite element method with lagrangian multipliers. Numerische Mathematik,
20:179192, 1973. {132}
[13] K. J. Bathe. Finite Element Procedures. Prentice-Hall, New Jersey, 1996. {132}
[14] Z. P. Ba zant. Instability, ductility and size-effect in strain-softening concrete. Journal of Engineering
Mechanics, 102:331344, 1976. {117}
154 References
[15] Z. P. Ba zant. Mechanics of distributed cracking. Applied Mechanics Reviews, 39:675705, 1986.
{118}
[16] Z. P. Ba zant and L. Cedolin. Stability of Structures: Elastic, Inelastic, Fracture and Damage Theories.
Oxford University Press, New York, 1991. {117}
[17] Z. P. Ba zant and M Jir asek. Nonlocal integral formulations of plasticity and damage: survey of
progress. Journal of Engineering Mechanics, 128(11):11191149, 2002. {71, 118}
[18] Z. P. Ba zant and F. B. Lin. Non-local yield limit degradation. International Journal for Numerical
Methods in Engineering, 26:18051823, 1985. {118}
[19] Z. P. Ba zant and B. Oh. Crack band theory for fracture of concrete. Materials and Struc-
tures/Mat eriaux et Constructions, 16:155177, 1983. {2, 117}
[20] Z. P. Ba zant, J. O zbolt, and R. Eligehausen. Fracture size effect: review of evidence for concrete
structures. Journal of Structural Engineering, 120(8):23772398, 1994. {120}
[21] Z. P. Ba zant and G. Pijaudier-Cabot. Nonlocal continuum damage, localization instability and
convergence. Journal of Applied Mechanics, 55:287293, 1988. {118, 125, 130, 132}
[22] Z. P. Ba zant and G. Pijaudier-Cabot. Measurement of characteristic length of nonlocal continuum.
Journal of Engineering Mechanics, 115:755767, 1989. {120}
[23] T. Belytschko and K. Mish. Computability in non-linear solid mechanics. International Journal for
Numerical Methods in Engineering, 52(12):321, 2001. {120}
[24] N. Bi cani c. From continua to discontinuamodelling of progressive discontinuities. In R. de
Borst, N. Bi cani c, H. A. Mang, and G. Meschke, editors, Computational Modelling of Concrete Struc-
tures, pages 921930, Rotterdam, 1998. Balkema. {1}
[25] T. N. Bittencourt, P. A. Wawrzynek, A. R. Ingraffea, and J. L. Souza. Quasi-automatic simulation
of crack propagation for 2D LEFM problems,. Engineering Fracture Mechanics, 52(2):321334, 1996.
{13}
[26] L. Bod e, J. L. Tailhan, G. Pijaudier-Cabot, C. La Borderie, and J. L. Cl ement. Failure analysis of
initially cracked concrete structures. Journal of Engineering Mechanics, 123:11531160, 1997. {1}
[27] G. Bolzon. Formulation of a triangular element with an embedded interface via isoparametric
mapping. Computational Mechanics, 27:463473, 2001. {1}
[28] G. Bolzon and A. Corigliano. Finite elements with embedded displacement discontinuity. a gen-
eralised variable formulation. International Journal for Numerical Methods in Engineering, 49:1227
1266, 2000. {1}
[29] R. I. Borja. A nite element model for strain localization analysis of strongly discontinuous elds
based on standard Galerkin approximation. Computer Methods in Applied Mechanics and Engineer-
ing, 190:15291549, 2000. {146}
[30] R. de Borst. Simulation of strain localisation: a reappraisal of the Cosserat continuum. Engineering
Computations, 8:317332, 1991. {146}
[31] R. de Borst. A generalisation of J
2
-ow theory for polar continua. Computer Methods in Applied
Mechanics and Engineering, 103:347362, 1993. {118}
[32] R. de Borst, A. Benallal, and O. M. Heeres. A gradient-enhanced damage approach to fracture.
Journal de Physique IV, 6:491502, 1996. {121, 125, 130}
References 155
[33] R. de Borst and H.-B. M uhlhaus. Gradient-dependent plasticity - formulation and algorithmic
aspects. International Journal for Numerical Methods in Engineering, 35(3):521539, 1992. {118}
[34] R. de Borst, J. Pamin, and M. G. D. Geers. On coupled gradient-dependent plasticity and damage
theories with a view to localization analysis. European Journal of Mechanics A/Solids, 18(6):939962,
1999. {56, 57, 67}
[35] R. de Borst and L. J. Sluys. Localisation in a Cosserat continuumunder static and dynamic loading
conditions. Computer Methods in Applied Mechanics and Engineering, 90:805827, 1991. {118}
[36] R. de Borst, L. J. Sluys, H.-B. M uhlhaus, and J. Pamin. Fundamental issues in nite element
analyses of localization of deformation. Engineering Computations, 10:99121, 1993. {117}
[37] F. Brezzi. On the existence, uniqueness and approximation of saddle-point problems arising from
lagrangian multipliers. Revue Fran caise dAutomatique, dInformatique et de Recherche Operationnelle.
Analyse Num` erique, 8:129151, 1974. {132}
[38] F. Brezzi and M. Fortin. Mixed and Hybrid Finite Element Methods. Springer-Verlag, New York,
1991. {132}
[39] D. Brokken. Numerical Modelling of Ductile Fracture in Blanking. PhD thesis, Eindhoven University
of Technology, 1999. {44}
[40] J. Carmeliet. Optimal estimation of gradient damage parameters from localization phenomena in
quasi-brittle materials. Mechanics of Cohesive-Frictional Materials, 4(1):116, 1999. {119}
[41] I. Carol and K. Willam. Application of analytical solutions in elasto-plasticity to localization anal-
ysis of damage models. In D. R. J. Owen, E. O nate, and E. Hinton, editors, Computational Plasticity,
Fundamentals and Applications, pages 714719, Barcelona, Spain, 1997. CIMNE. {145, 146, 148}
[42] A. Chorin, T. J. R. Hughes, M. F. McCracken, and J. E. Marsden. Product formulas and numerical
algorithms. Communications in Pure and Applied Mathematics, 31:205256, 1978. {72, 76}
[43] C. Comi. Computational modelling of gradient-enhanced damage in quasi-brittle materials. Me-
chanics of Cohesive-Frictional Materials, 4:1736, 1999. {132}
[44] A. Corigliano and S. Mariani. Parameter identication of a time-dependent elastic-damage in-
terface model for the simulation of debonding in composites. Composites Science and Technology,
61:191203, 2000. {99}
[45] E. Cosserat and F. Cosserat. Th eorie des Corps D eformables. Librairie Scientique A. Herman et Fils,
Paris, 1909. {118}
[46] M. A. Criseld. Non-Linear Finite Element Analysis of Solids and Structures. John Wiley & Sons Ltd,
Chichester, England, 1996. {6, 76}
[47] M. A. Criseld. Non-Linear Finite Element Analysis of Solids and Structures: Advanced Topics. John
Wiley & Sons Ltd, Chichester, England, 1997. {6, 78}
[48] M. A. Criseld and J. Wills. Numerical comparison involving different concrete-models. In
Computational Mechanics of Concrete Structures Advances and Applications, IABSE Colloquium
Delft 1987, Report, volume 54, pages 177187. Delft University Press, Delft, the Netherlands, 1987.
{2}
156 References
[49] K. De Proft, G. N. Wells, L. J. Sluys, and W. De Wilde. A combined experimental-numerical study
to cyclic behaviour of limestone. In D. R. J. Owen, E. O nate, and B. Su arez, editors, Proceedings
of the 7th International Conference on Computational Plasticity Complas 2003, page 120. CIMNE,
Barcelona, Spain, 7-10 April 2003. {119}
[50] A. Dietsche, P. Steinmann, and K. Willam. Micropolar elasto-plasticity and its role in localisation
analysis. International Journal of Plasticity, 9:813831, 1992. {118}
[51] J. Dolbow, N. Mo es, and T. Belytschko. Discontinuous enrichment in nite elements with a parti-
tion of unity method. Finite Elements in Analysis and Design, 36:235260, 2000. {21}
[52] C. A. M. Duarte and J. T. Oden. H-p clouds an h-p meshless method. Numerical Methods for
Partial Differential Equations, 12(6):673705, 1996. {13, 16}
[53] J. F. Dub e, G. Pijaudier-Cabot, and C. La Borderie. Rate dependent damage model for concrete in
dynamics. Journal of Engineering Mechanics, 122(10):939947, 1996. {117}
[54] G. Duvaut and J. L. Lions. Les In equations en M echanique et en Physique. Dunod, Paris, France, 1972.
{12, 71, 73, 118}
[55] R. A. B. Engelen, M. G. D. Geers, and F. P. T. Baaijens. Nonlocal implicit gradient-enhanced elasto-
plasticity for the modelling of softening behaviour. International Journal of Plasticity, 19(4):403433,
2003. {87, 99, 135, 148}
[56] A. C. Eringen, C. G. Speziale, and B. S. Kim. Crack-tip problem in non-local elasticity. Journal of
the Mechanics and Physics of Solids, 25:339255, 1977. {141}
[57] H. Espinosa, P. D. Zavattieri, and S. K. Dwivedi. A nite deformation continuum/discrete model
for the description of fragmentation and damage in brittle materials. Journal of the Mechanics and
Physics of Solids, 46(10):19091942, 1998. {1}
[58] H. L. Ewalds and R. J. H. Wanhill. Fracture Mechanics. Edward Arnold, London, England, 1984.
{138}
[59] P. H. Feenstra. Computational Aspects of Biaxial Stress in Plain and Reinforced Concrete. PhD thesis,
Delft University of Technology, 1993. {1, 121}
[60] P. H. Feenstra and R. de Borst. Aspects of robust computational modeling for plain and reinforced
concrete. HERON, 38(4):376, 1993. {1, 121}
[61] S. Fichant, C. La Borderie, and G. Pijaudier-Cabot. Isotropic and anisotropic descriptions of dam-
age in concrete structures. Mechanics of Cohesive-Frictional Materials, 4:339359, 1999. {35}
[62] F. X. Garaizar and J. Trangenstein. Adaptive mesh renement and front-tracking for shear bands
in an antiplane shear model. SIAM Journal on Scientic Computing, 20(2):750779, 1998. {146}
[63] M. G. D. Geers. Experimental Analysis and Computational Modelling of Damage and Fracture. PhD
thesis, Eindhoven University of Technology, 1997. {7, 119, 120}
[64] M. G. D. Geers, R. de Borst, W. A. M. Brekelmans, and R. H. J. Peerlings. Strain-based transient-
gradient damage model for failure analyses. Computer Methods in Applied Mechanics and Engineer-
ing, 160:133153, 1998. {2, 8, 41, 44, 48, 69, 99}
[65] M. G. D. Geers, R. de Borst, W. A. M. Brekelmans, and R. H. J. Peerlings. Validation and in-
ternal length scale determination for a gradient damage model: application to short glass-bre-
reinforced polypropylene. International Journal of Solids and Structures, 36:25572583, 1999. {56, 57,
87, 88}
References 157
[66] M. G. D. Geers, R. de Borst, and R. H. J. Peerlings. Damage and crack modeling in single-edge
and double-edge notched concrete beams. Engineering Fracture Mechanics, 65:247261, 2000. {67}
[67] M. G. D. Geers, R. de Borst, and A. A. J. M. Peijs. Mixed numerical-experimental identication of
nonlocal characteristics of random bre reinforced composites. Composite Science and Technology,
59:15691578, 1999. {119}
[68] M. G. D. Geers, T. Peijs, W. A. M. Brekelmans, and R. de Borst. Experimental monitoring of
strain localization and failure behaviour of composite materials. Composite Science and Technology,
56:12831290, 1996. {51, 56, 119}
[69] J. F. Georgin, W. Nechnech, L. J. Sluys, and J. F. Reynouard. A coupled damage-viscoplasticity
model for localization problems. In H. A. Mang, F. G. Rammerstorfer, and J. Eberhardsteiner,
editors, Proceedings of the Fifth World Congress on Computational Mechanics WCCM V, volume I,
page 428. Vienna University of Technology, Austria, July 7-12 2002. {11, 74}
[70] W. H. Gerstle and A. R. Ingraffea. Does bond-slip exist? In S. P. Shah, S. E. Swartz, and M. L.
Wang, editors, Proceedings of Conference on Micromechanics of Quasi-brittle Materials, pages 407416.
Elsevier Science Publishing LTD, June 6-8 1990. {120}
[71] S. Ghavamian and I. Carol. Benchmarking of concrete cracking constitutive laws: MECA project.
In N. Bi cani c, R. de Borst, H. A. Mang, and G. Meschke, editors, Computational Modelling of Concrete
Structures EURO-C 2003, pages 179187. Swets & Zeitlinger, Lisse, 17-20 March 2003. {120, 121}
[72] R. E. Goodman, R. L. Taylor, and T. L. Brekke. A model for the mechanics of jointed rocks. Journal
of the Soil Mechanics and Foundations Division, 94:637659, 1968. {13}
[73] W. G. Gray and P. C. Y. Lee. On the theorems for local volume averaging of multiphase systems.
International Journal of Multiphase Flow, 3:333340, 1977. {115, 116}
[74] W. G. Gray, A. Leijnse, R. L. Kolar, and C. A. Blain. Mathematical Tools for Changing Scales in the
Analysis of Physical Systems. CRC, Boca Raton, Florida, 1993. {115}
[75] H. J. Grootenboer. Finite Element Analysis of Two-Dimensional Reinforced Concrete Structures, Tak-
ing Account of Non-Linear Physical Behaviour and the Development of Discrete Cracks. PhD thesis,
Technische Hogeschool Delft, 1979. {1}
[76] A. Hillerborg. Analysis of fracture by means of the ctitious crack model, particularly for bre
reinforced concrete. The International Journal of Cement Composites, 2(4):177184, 1980. {92}
[77] G. A. Holzapfel. Nonlinear Solid Mechanics. John Wiley & Sons, New York, 2000. {viii}
[78] D. A. Hordijk. Local Approach to Fatigue of Concrete. PhD thesis, Delft University of Technology,
1991. Also published as Tensile and tensile fatigue behaviour of concrete: experiments, modelling
and analyses. HERON, 37(1):379, 1992. {24, 53, 54, 64}
[79] T. J. R. Hughes. The Finite Element Method - Linear Static and Dynamic Finite Element Analysis.
Prentice-Hall, London, England, 1987. {132}
[80] C. Iacono, L. J. Sluys, and J. G. M. van Mier. Development of an inverse procedure for parameter
estimates of numerical models. In N. Bi cani c, R. de Borst, H. A. Mang, and G. Meschke, editors,
Computational Modelling of Concrete Structures EURO-C 2003, pages 259268. Swets & Zeitlinger,
Lisse, 17-20 March 2003. {119}
[81] J. Janson and J. Hult. Fracture mechanics and damage mechanics - a combined approach. Journal
de M echanique Appliqu ee, 1(1):6983, 1977. {1}
158 References
[82] Y. S. Jenq and S. P. Shah. Crack propagation in ber-reinforced concrete. Journal of Structural
Engineering, 112(1):1934, 1986. {12}
[83] M. Jir asek. Modelling of localized damage and fracture in quasibrittle materials. In P. A. Vermeer,
S. Diebels, W. Ehlers, H. J. Herrmann, S. Luding, and E. Ramm, editors, Continuous and Discon-
tinuous Modelling of Cohesive-Frictional Materials, volume 568, pages 1719. Springer, Berlin, 2001.
{50}
[84] M. Jir asek and Z. P. Ba zant. Inelastic Analysis of Structures. John Wiley & Sons, New York, 2002.
{viii, 6, 48, 72}
[85] M. Jir asek and B. Patz ak. Models for quasibrittle failure: theoretical and computational aspects.
In Z. Waszczyszyn and J. Pamin, editors, Solids, Structures and Coupled Problems in Engineering,
Proceedings of the Second European Conference on Computational Mechanics, volume 1, pages 7071,
Cracow, Poland, 2001. Fundacja Zdrowia Publicznego, Vesalius Publisher. On CD-ROM. {120}
[86] M. Jir asek and T. Zimmermann. Embedded crack model. Part II: Combination with smeared
cracks. International Journal for Numerical Methods in Engineering, 50:12911305, 2001. {1, 23, 69}
[87] J. W. Ju. On energy-based coupled elastoplastic damage theories: constitutive modeling and com-
putational aspects. International Journal of Solids and Structures, 25(7):803833, 1989. {1, 76, 117}
[88] P. Kabele. Assessment of structural performance of engineered cementitious composites by com-
puter simulation. Habilitation thesis, Czech Technical University, Prague, November 2000. {38,
39}
[89] P. Kabele and H. Horii. Analytical model for fracture behaviour of pseudo strain-hardening ce-
mentitious composites. Concrete Library International (proc. of JSCE), 29:105120, 1997. {1, 35}
[90] P. Kabele, E. Yamaguchi, and H. Horii. FEM-BEM superposition method for analysis of quasi-
brittle structures. International Journal of Fracture, 100:249274, 1999. {1}
[91] R. E. Kalman. A new approach to linear ltering and prediction problems. Journal of Basic Engi-
neering, 82:3545, 1960. {99}
[92] T. Kanda, Z. Lin, and V. C. Li. Modeling of tensile stress-strain relation of pseudo strain-hardening
cementitious composites. Journal of Materials in Civil Engineering, 12(2):147156, 2000. {35, 119}
[93] E. Kuhl, E. Ramm, and R. de Borst. An anisotropic gradient damage model for quasi-brittle ma-
terials. Computer Methods in Applied Mechanics and Engineering, 183:87103, 2000. {67}
[94] J. Kullaa. Finite element modelling of bre-reinforced brittle materials. HERON, 42(2):7595, 1997.
{33, 34}
[95] D. Lasry and T. Belytschko. Localization limiters in transient problems. International Journal of
Solids and Structures, 24:581597, 1988. {117, 118}
[96] J. Lemaitre. A Course on Damage Mechanics. Springer-Verlag, Berlin, second edition, 1996. {72}
[97] F. Li and Z. Li. Continuum damage mechanics based modeling of ber reinforced concrete in
tension. International Journal of Solids and Structures, 38:777793, 2001. {119}
[98] V. C. Li and T. Hashida. Ductile fracture in cementitious materials? In Z. P. Ba zant, editor, Fracture
Mechanics of Concrete Structures. Proceedings of the First International Conference on Fracture Mechanics
of Concrete and Concrete Structures FramCoS 1, pages 526535, London, 1992. Elsevier. {35, 38}
References 159
[99] V. C. Li, H. Stang, and H. Krenchel. Micromechanics of crack bridging in ber reinforced concrete.
Materials and Structures/Mat eriaux et Constructions, 26:486494, 1993. {92}
[100] V. C. Li and H. C. Wu. Conditions for pseudo strain-hardening in ber reinforced brittle matrix
composites. Applied Mechanics Reviews, 45(8):390398, 1992. {5, 34, 38}
[101] B. Loret and J. H. Prevost. Dynamic strain localization in elasto-(visco)plastic solids. Part I. Gen-
eral formulation and one-dimensional example. Computer Methods in Applied Mechanics and Engi-
neering, 83:247273, 1990. {117}
[102] R. Mahnken and E. Kuhl. Parameter identication of gradient enhanced damage models with the
nite element method. European Journal of Mechanics A/Solids, 18:819835, 1999. {119}
[103] L. E. Malvern. Introduction to the Mechanics of a Continuous Medium. Prentice-Hall International,
Englewood Cliffs, New Jersey, 1969. {19}
[104] J. Mazars. Application de la M ecanique de lEndommagement au Comportement Non-Lin eaire et ` a la
Rupture du B eton de Structure. PhD thesis, Universit e Paris VI, France, 1984. {7}
[105] J. Mazars and G. Pijaudier-Cabot. Continuum damage theory - application to concrete. Journal of
Engineering Mechanics, 115:345365, 1989. {7}
[106] J. Mazars and G. Pijaudier-Cabot. From damage to fracture mechanics and conversely: a com-
bined approach. International Journal of Solids and Structures, 33(2022):33273342, 1996. {1}
[107] J. M. Melenk and I. Babu ska. The partition of unity nite element method: basic theory and
applications. Computer Methods in Applied Mechanics and Engineering, 139(14):289314, 1996. {13,
16}
[108] J. G. M. van Mier. Fracture Processes of Concrete. CRC Press, Inc., Boca Raton, Florida, 1997. {2, 64,
71, 117, 137}
[109] N. Mo es and T. Belytschko. Extended nite element method for cohesive crack growth. Engineer-
ing Fracture Mechanics, 69:813833, 2002. {22, 121}
[110] N. Mo es, J. Dolbow, and T. Belytschko. Anite element method for crack growth without remesh-
ing. International Journal for Numerical Methods in Engineering, 46(1):131150, 1999. {3, 13, 15, 21,
23, 28, 31}
[111] S. Mohammadi, D. R. J. Owen, and D. Peric. A combined nite/discrete element algorithm for
delamination analysis of composites. Finite Elements in Analysis and Design, 28:321336, 1998. {1}
[112] S. Mohammadi, D. R. J. Owen, and D. Peric. 3D progressive damage analysis of composites by -
nite/discrete element approach. In E. O nate, G. Bugeda, and B. Su arez, editors, European Congress
on Computational Methods in Applied Sciences and Engineering, Barcelona, Spain, 2000. CIMNE. On
CD-ROM. {1}
[113] H.-B. M uhlhaus. Application of Cosserat theory in numerical solutions of limit load problems.
Archive of Applied Mechanics (Ingenieur Archiv), 59:124137, 1989. {118}
[114] H.-B. M uhlhaus and E. C. Aifantis. A variational principle for gradient plasticity. International
Journal of Solids and Structures, 28:845857, 1991. {86, 118}
[115] H.-B. M uhlhaus and I. Vardoulakis. The thickness of shear bands in granular materials.
G eotechnique, 37:271283, 1987. {118}
160 References
[116] T. M unz, K. Willam, and K. Runesson. Viscoplastic algorithmic operators and their localization
properties. In R. de Borst, N. Bi cani c, H. A. Mang, and G. Meschke, editors, Computational Mod-
elling of Concrete Structures, pages 219228, Rotterdam, 1998. Balkema. {74, 78}
[117] A. E. Naaman. SIFCON: tailored properties for structural performance. In A. E. Naaman and
H. W. Reinhardt, editors, High Performance Fibre Reinforced Cement Composites HPFRCC-91, pages
1838, London, 1991. E & FN Spon. {5}
[118] A. Needleman. Material rate dependence and mesh sensitivity in localization problems. Computer
Methods in Applied Mechanics and Engineering, 67:6985, 1988. {117}
[119] A. Needleman and V. Tvergaard. Finite element analysis of localization in plasticity. In J. T.
Oden and G. F. Carey, editors, Finite Elements: Special Problems in Solid Mechanics, volume V, pages
94157. Prentice-Hall, Inc., Englewood Cliffs, New Jersey, 1984. {146}
[120] S. Nemat-Nasser and N. Okada. Radiographic and microscopic observation of shear bands in
granular materials. G eotechnique, 51(9):753765, 2001. {146}
[121] J. T. Oden, C. A. M. Duarte, and O. C. Zienkiewicz. A new cloud-based hp nite element method.
Computer Methods in Applied Mechanics and Engineering, 153(12):117126, 1998. {16}
[122] J. Oliver. Modelling strong discontinuities in solid mechanics via strain softening constitutive
equations. Part 2: numerical simulation. International Journal for Numerical Methods in Engineering,
39(21):36013623, 1996. {13, 16}
[123] J. Oliver, M. Cervera, and O. Manzoli. Strong discontinuities and continuum plasticity models:
the strong discontinuity approach. International Journal of Plasticity, 15:319351, 1999. {1}
[124] J. Oliver, A. E. Huespe, M. D. G. Pulido, and E. Chaves. From continuum mechanics to fracture
mechanics: the strong discontinuity approach. Engineering Fracture Mechanics, 69:113136, 2002.
{1}
[125] M. Ortiz, Y. Leroy, and A. Needleman. A nite element method for localized failure analysis.
Computer Methods in Applied Mechanics and Engineering, 61:189214, 1987. {146}
[126] J. O zbolt. General microplane model for concrete. In F. H. Wittmann, editor, Numerical Models in
Fracture Mechanics of Concrete, pages 173187, Rotterdam, Brookeld, 1993. A.A. Balkema. {120}
[127] J. O zbolt and Z. P. Ba zant. Numerical smeared fracture analysis: nonlocal microcrack interaction
approach. International Journal for Numerical Methods in Engineering, 39:635661, 1996. {120}
[128] J. Pamin. Gradient-Dependent Plasticity in Numerical Simulation of Localisations Phenomena. PhD
thesis, Delft University of Technology, 1994. {11, 86, 87, 94, 118, 121}
[129] J. Pamin, H. Askes, and R. de Borst. Two gradient plasticity theories discretized with the Element-
free Galerkin Method. Computer Methods in Applied Mechanics and Engineering, 192:23772403,
2003. {135, 148}
[130] J. Pamin and R. de Borst. Gradient-enhanced damage and plasticity models for plain and rein-
forced concrete. In W. Wunderlich, editor, Proceedings of the European Conference on Computational
Mechanics ECCM99, pages 482483, paper no. 636, Munich, 1999. Technical University of Mu-
nich. {3, 4, 41, 53, 67}
[131] R. H. J. Peerlings. Enhanced Damage Modelling for Fracture and Fatigue. PhD thesis, Eindhoven
University of Technology, 1999. {3, 44, 56, 61, 127, 140}
References 161
[132] R. H. J. Peerlings, R. de Borst, W. A. M. Brekelmans, and M. G. D. Geers. Gradient-enhanced
damage modelling of concrete fracture. Mechanics of Cohesive-Frictional Materials, 3:323342, 1998.
{61, 67, 121}
[133] R. H. J. Peerlings, R. de Borst, W. A. M. Brekelmans, and M. G. D. Geers. Gradient-enhanced
damage modelling of high-cycle fatigue. International Journal for Numerical Methods in Engineering,
49:15471569, 2000. {44}
[134] R. H. J. Peerlings, R. de Borst, W. A. M. Brekelmans, and J. H. P. de Vree. Gradient-enhanced dam-
age for quasi-brittle materials. International Journal for Numerical Methods in Engineering, 39:3391
3403, 1996. {8, 9, 4143, 47, 50, 72, 96, 118, 121, 125127, 130, 133, 135, 136, 150}
[135] R. H. J. Peerlings, M. G. D. Geers, R. de Borst, and W. A. M. Brekelmans. A critical comparison of
nonlocal and gradient-enhanced softening continua. International Journal of Solids and Structures,
38:77237746, 2001. {9, 137, 140, 146}
[136] P. Perzyna. The constitutive equations for rate sensitive plastic materials. Quarterly of Applied
Mathematics, 20(4):321332, 1962. {118}
[137] P. Perzyna. Fundamental problems in viscoplasticity. In Advances in Applied Mechanics, volume 9,
pages 243377. Academic Press, New York, 1966. {12, 71, 77, 118}
[138] S. Pietruszczak and Z. Mr oz. Finite element analysis of deformation of strain softening materials.
International Journal for Numerical Methods in Engineering, 17:327334, 1981. {2, 117}
[139] G. Pijaudier-Cabot and Z. P. Ba zant. Nonlocal damage theory. Journal of Engineering Mechanics,
113:15121533, 1987. {9, 41, 117, 118, 120, 125, 130, 132, 135, 150}
[140] G. Pijaudier-Cabot, Z. P. Ba zant, and M. Tabbara. Comparison of various models for strain-
softening. Engineering Computations, 5:141150, 1988. {117}
[141] G. Pijaudier-Cabot and A. Benallal. Strain localization and bifurcation in a nonlocal continuum.
International Journal of Solids and Structures, 30:17611775, 1993. {118}
[142] A. Quarteroni and A. Valli. Numerical Approximation of Partial Differential Equations. Springer-
Verlag, Berlin, 1994. {131, 132}
[143] J. J. C. Remmers, R. de Borst, and A. Needleman. A cohesive segments method for the simulation
of crack growth. Computational Mechanics, 31(12):6977, 2003. {28}
[144] J. J. C. Remmers, G. N. Wells, and R. de Borst. Analysis of delamination growth with discon-
tinuous nite elements. In Z. Waszczyszyn and J. Pamin, editors, Solids, Structures and Coupled
Problems in Engineering, Proceedings of the Second European Conference on Computational Mechanics,
volume 2, pages 906907, Cracow, Poland, 2001. Fundacja Zdrowia Publicznego, Vesalius Pub-
lisher. On CD-ROM. {29}
[145] Z. Ren and N. Bi cani c. Simulation of progressive fracturing under dynamic loading conditions.
Communications in Numerical Methods in Engineering, 13:127138, 1997. {1}
[146] RILEM. TC 162-TDF: Test and design methods for steel bre reinforced concrete. Materials and
Structures/Mat eriaux et Constructions, 33(225):35, 2000. {120}
[147] RILEM. TC 162-TDF: Test and design methods for steel bre reinforced concrete. Materials and
Structures/Mat eriaux et Constructions, 33(226):7581, 2000. {120}
[148] E. Rizzi, I. Carol, and K. Willam. Localization analysis of elastic degradation with application to
scalar damage. Journal of Engineering Mechanics, 121(4):541554, 1995. {145, 146}
162 References
[149] A. Rodrguez-Ferran and A. Huerta. Error estimation and adaptivity for nonlocal damage models.
International Journal of Solids and Structures, 37:75017528, 2000. {120123}
[150] P. E. Roelfstra and F. H. Wittmann. Numerical method to link strain softening with failure of
concrete. In F. H. Wittmann, editor, Fracture Toughness and Fracture Energy, pages 163175, Ams-
terdam, 1986. Elsevier. {119}
[151] J. G. Rots. Computational Modeling of Concrete Fracture. PhD thesis, Delft University of Technology,
1988. {1, 2830, 117, 121}
[152] J. G. Rots. Removal of nite elements in strain-softening analysis of tensile fracture. In Z. P. Ba zant,
editor, Fracture Mechanics of Concrete Structures. Proceedings of the First International Conference on
Fracture Mechanics of Concrete and Concrete Structures FramCoS 1, pages 330338, London, 1992.
Elsevier. {1, 2}
[153] J. G. Rots and J. Blaauwendraad. Crack models for concrete: Discrete or smeared? Fixed, multi-
directional or rotating? HERON, 34(1):359, 1989. {1, 117, 121}
[154] J. G. Rots, P. Nauta, G. M. A. Kusters, and J. Blaauwendraad. Smeared crack approach and fracture
localization in concrete. HERON, 30(1):348, 1985. {1, 117, 121}
[155] J. G. Rots and J. C. J. Schellekens. Interface elements in concrete mechanics. In N. Bi cani c and H. A.
Mang, editors, Computer Aided Analysis and Design of Concrete Structures, pages 909918, Swansea,
U.K., 1990. Pinerigde Press. {110}
[156] K. Runesson, N. S. Ottosen, and D. Peric. Discontinuous bifurcation of elasto-plastic solutions in
plane stress and plane strain. International Journal of Plasticity, 7:99121, 1991. {145, 146, 148}
[157] K. Runesson, M. Ristinmaa, and L. M ahler. A comparison of viscoplasticity formats and algo-
rithms. Mechanics of Cohesive-Frictional Materials, 4(1):7598, 1999. {72}
[158] H. G. Russel. ACI denes High Performance Concrete. Concrete International, 21(2):5657, 1999.
{4}
[159] V. E. Saouma, E. Br uhwiler, and H. L. Boggs. A review of fracture mechanics applied to concrete
dams. International Journal of Dam Engineering, 1(1):4157, 1990. {13}
[160] J. C. J. Schellekens and R. de Borst. On the numerical integration of interface elements. Interna-
tional Journal for Numerical Methods in Engineering, 36:4366, 1993. {30, 31, 105, 110}
[161] E. Schlangen. Experimental and Numerical Analysis of Fracture Processes in Concrete. PhD thesis,
Delft University of Technology, 1993. Also published in HERON, 38(2):3117, 1993. {60, 61, 118,
120, 121, 123}
[162] J. C. Simo and T. J. R. Hughes. Computational Inelasticity. Springer, London, 1998. {76}
[163] J. C. Simo and J. W. Ju. Strain- and stress-based continuum damage modelsI. Formulation. Inter-
national Journal of Solids and Structures, 23(7):821840, 1987. {72, 76, 117}
[164] J. C. Simo and J. W. Ju. Strain- and stress-based continuum damage modelsII. Computational
aspects. International Journal of Solids and Structures, 23(7):841869, 1987. {72, 76, 117}
[165] J. C. Simo, J. G. Kennedy, and S. Govindjee. Non-smooth multisurface plasticity and viscoplas-
ticity. loading/unloading conditions and numerical algorithms. International Journal for Numerical
Methods in Engineering, 26:21612185, 1988. {12, 73}
References 163
[166] J. C. Simo, J. Oliver, and F. Armero. An analysis of strong discontinuities induced by strain-
softening in rate-independent inelastic solids. Computational Mechanics, 12:277296, 1993. {13, 16}
[167] J. C. Simo and R. L. Taylor. A return mapping algorithm for plane stress elastoplasticity. Interna-
tional Journal for Numerical Methods in Engineering, 22:649670, 1986. {11}
[168] A. Simone. Partition of unity-based discontinuous elements for interface phenomena: computa-
tional issues. 2003. Submitted. {13, 105}
[169] A. Simone, H. Askes, R. H. J. Peerlings, and L. J. Sluys. Interpolation requirements for implicit
gradient-enhanced continuum damage models. Communications in Numerical Methods in Engineer-
ing, 19(7):563572, 2003. See also the corrigenda which will appear in Communications in Numerical
Methods in Engineering. {125}
[170] A. Simone, H. Askes, and L. J. Sluys. Incorrect initiation and propagation of failure in non-local
and gradient-enhanced media. International Journal of Solids and Structures, 2003. In press. {135}
[171] A. Simone, J. J. C. Remmers, and G. N. Wells. An interface element based on the partition of unity.
Technical Report CM2001-007, Delft University of Technology, Delft, the Netherlands, November
2001. {13, 105}
[172] A. Simone and L. J. Sluys. Continuous-discontinuous failure analysis in a rate-dependent elasto-
plastic damage model. In F. Bontempi, editor, Proceedings of the 2nd International Structural Engi-
neering and Construction Conference ISEC-02, 23-26 September 2003. {71}
[173] A. Simone and L. J. Sluys. The use of displacement discontinuities in a rate-dependent medium.
Computer Methods in Applied Mechanics and Engineering, 2003. In press. {71}
[174] A. Simone, L. J. Sluys, and P. Kabele. Combined continuous/discontinuous failure of cementi-
tious composites. In N. Bi cani c, R. de Borst, H. A. Mang, and G. Meschke, editors, Computational
Modelling of Concrete Structures EURO-C 2003, pages 133137. Swets & Zeitlinger, Lisse, 17-20
March 2003. {13}
[175] A. Simone, G. N. Wells, and L. J. Sluys. Discontinuous modelling of crack propagation in a
gradient-enhanced continuum. In H. A. Mang, F. G. Rammerstorfer, and J. Eberhardsteiner, edi-
tors, Proceedings of the Fifth World Congress on Computational Mechanics WCCM V, volume I, page
130. Vienna University of Technology, Austria, 712 July 2002. {41}
[176] A. Simone, G. N. Wells, and L. J. Sluys. From continuous to discontinuous failure in a gradient-
enhanced continuum damage model. Computer Methods in Applied Mechanics and Engineering,
2002. In press. {41}
[177] A. Simone, G. N. Wells, and L. J. Sluys. Discontinuities in regularised media. In D. R. J. Owen,
E. O nate, and B. Su arez, editors, Proceedings of the 7th International Conference on Computational
Plasticity Complas 2003, page 90. CIMNE, Barcelona, Spain, 7-10 April 2003. On CD-ROM. {71}
[178] D. Simons. Grid convergence in several simple problems. In G. H. Hulbert, editor, Sixth U.S. Na-
tional Congress on Computational Mechanics, page 279. Mechanical Engineering Department, Uni-
versity of Michigan, 13 August 2001. {120}
[179] L. J. Sluys. Wave Propagation, Localisation and Dispersion in Softening Solids. PhD thesis, Delft
University of Technology, 1992. {71, 72, 87, 99, 117, 118, 146, 149}
[180] L. J. Sluys and R. de Borst. Wave propagation and localization in a rate-dependent cracked
mediummodel formulation and one-dimensional examples. International Journal of Solids and
Structures, 29:29452958, 1992. {1, 117}
164 References
[181] L. J. Sluys and R. de Borst. Failure in plain and reinforced concretean analysis of crack width
and crack spacing. International Journal of Solids and Structures, 33(2022):32573276, 1996. {1}
[182] L. J. Sluys, R. de Borst, and H.-B. M uhlhaus. Wave propagation, localization and dispersion in a
gradient-dependent medium. International Journal of Solids and Structures, 30(9):11531171, 1993.
{118, 146}
[183] I. Stakgold. Greens Functions and Boundary Value Problems. John Wiley & Sons, New York, 1979.
{115}
[184] H. Stang and J. F. Olesen. On the interpretation of bending tests on frc-materials. In H. Mihashi
and K. Rokugo, editors, Fracture Mechanics of Concrete Structures. Proceedings of the Third Interna-
tional Conference on Fracture Mechanics of Concrete and Concrete Structures FramCoS 3, volume 1,
pages 511520, Freiburg, Germany, 1998. Aedicatio Publishers. {92, 120}
[185] M. G. A. Tijssens, L. J. Sluys, and E. van der Giessen. Numerical simulation of quasi-brittle fracture
using damaging cohesive surfaces. European Journal of Mechanics A/Solids, 19(5):761779, 2000. {12,
39}
[186] V. Tvergaard, A. Needleman, and K. K. Lo. Flowlocalization in the plane strain tensile test. Journal
of the Mechanics and Physics of Solids, 29(2):115142, 1981. {146}
[187] M. R. A. van Vliet. Size Effect in Tensile Fracture of Concrete and Rock. PhD thesis, Delft University
of Technology, 2000. {120}
[188] J. H. P. de Vree, W. A. M. Brekelmans, and M. A. J. van Gils. Comparison of nonlocal approaches
in continuum damage mechanics. Computers and Structures, 55:581588, 1995. {7}
[189] W. M. Wang. Stationary and Propagative Instabilities in Metals A Computational Point of View. PhD
thesis, Delft University of Technology, 1997. {82, 117}
[190] W. M. Wang, L. J. Sluys, and R. de Borst. Viscoplasticity for instabilities due to strain softening
and strain-rate softening. International Journal for Numerical Methods in Engineering, 40:38393864,
1997. {117}
[191] R. J. Ward and V. C. Li. Dependence of exural behavior of ber reinforced mortar on material
fracture resistance and beam size. ACI Materials Journal, 87(6):627637, 1990. {32}
[192] G. N. Wells. Discontinuous Modelling of Strain Localisation and Failure. PhD thesis, Delft University
of Technology, 2001. {45, 86, 87}
[193] G. N. Wells and L. J. Sluys. A new method for modelling cohesive cracks using nite elements.
International Journal for Numerical Methods in Engineering, 50(12):26672682, 2001. {3, 14, 15, 17,
2123, 28, 31, 32, 34, 51, 121}
[194] G. N. Wells, L. J. Sluys, and R. de Borst. Simulating the propagation of displacement disconti-
nuities in a regularised strain-softening medium. International Journal for Numerical Methods in
Engineering, 53(5):12351256, 2002. {1, 24, 52, 71, 87, 146}
[195] J. G. Williams and P. D. Ewing. Fracture under complex stress the angled crack problem. Inter-
national Journal of Fracture Mechanics, 8(4):441446, 1972. {22}
[196] F. H. Wittmann, H. Mihashi, and N. Nomura. Size effect on fracture energy of concrete. Engineer-
ing Fracture Mechanics, 35:107115, 1990. {120}
[197] X.-P. Xu and A. Needleman. Numerical simulations of fast crack growth in brittle solids. Journal
of the Mechanics and Physics of Solids, 42:13971434, 1994. {12, 39}
References 165
[198] L. F. Zeng, G. Horrigmoe, and R. Andersen. Numerical implementation of constitutive integration
for rate-independent elastoplasticity. Computational Mechanics, 18:387396, 1996. {74}
[199] A. Zervos, P. Papanastasiou, and I. Vardoulakis. A nite element displacement formulation for
gradient elastoplasticity. International Journal for Numerical Methods in Engineering, 50:13691388,
2001. {146}
[200] D. Zhang and K. Wu. Fracture process zone of notched three-point-bending concrete beams.
Cement and Concrete Research, 29:18871892, 1999. {120}
[201] C. Zhao, B. E. Hobbs, H.-B. M uhlhaus, and A. Ord. A consistent point-searching algorithm for so-
lution interpolation in unstructured meshes consisting of 4-node bilinear quadrilateral elements.
International Journal for Numerical Methods in Engineering, 45(10):15091526, 1999. {24}
[202] O. C. Zienkiewicz and R. L. Taylor. The Finite Element Method. Butterworth-Heinemann, Oxford,
UK, fth edition, 2000. {132}
Author/editor index
Aifantis, E. C., 118
Aifantis, E. C., see M uhlhaus, H.-B., 86, 118
Atcin, P. C., 4
Alfaiate, J., 16, 19
Alfano, G., 13
Andersen, R., see Zeng, L. F., 74
Armero, F., 13, 16
Armero, F., see Simo, J. C., 13, 16
Askes, H., 44, 120, 121
Askes, H., see Pamin, J., see Simone, A., 125,
135, 148
ASTM, 56
Audi, M., 29
Baaijens, F. P. T., see Engelen, R. A. B., 87, 99,
135, 148
Babu ska, I., 132
Babu ska, I., see Melenk, J. M., 13, 16
Bathe, K. J., 132
Ba zant, Z. P., 1, 2, 35, 38, 71, 117, 118, 120, 125,
130, 132
Ba zant, Z. P., viii, see Jir asek, M., see O zbolt, J.,
see Pijaudier-Cabot, G., 6, 9, 41, 48,
72, 117, 118, 120, 125, 130, 132, 135,
150
Belytschko, T., 120
Belytschko, T., see Dolbow, J., see Lasry, D., see
Mo es, N., 3, 13, 15, 2123, 28, 31,
117, 118, 121
Benallal, A., see Borst, R. de, see
Pijaudier-Cabot, G., 118, 121, 125,
130
Bi cani c, N., 1, 13, 16, 110, 119121
Bi cani c, N., see Borst, R. de, see Ren, Z., 1, 74, 78
Bittencourt, T. N., 13
Blaauwendraad, J., see Rots, J. G., 1, 117, 121
Blain, C. A., see Gray, W. G., 115
Bod e, L., 1
Boggs, H. L., see Saouma, V. E., 13
Bolzon, G., 1
Bontempi, F., 71
Borja, R. I., 146
Borst, R. de, 1, 56, 57, 67, 74, 78, 117, 118, 121,
125, 130, 146
Borst, R. de, see Bi cani c, N., see Feenstra, P. H.,
see Geers, M. G. D., see Kuhl, E., see
Pamin, J., see Peerlings, R. H. J., see
Remmers, J. J. C., see Schellekens, J.
C. J., see Sluys, L. J., see Wang, W.
M., see Wells, G. N., 14, 8, 9, 13, 16,
24, 2831, 4144, 47, 48, 5053, 56,
57, 61, 67, 69, 71, 72, 87, 88, 96, 99,
105, 110, 117121, 125127, 130,
133, 135137, 140, 146, 148, 150
Brekelmans, W. A. M., see Geers, M. G. D., see
Peerlings, R. H. J., see Vree, J. H. P.
de, 2, 79, 4144, 47, 48, 50, 51, 56,
57, 61, 67, 69, 72, 87, 88, 96, 99, 118,
119, 121, 125127, 130, 133,
135137, 140, 146, 150
Brekke, T. L., see Goodman, R. E., 13
Brezzi, F., 132
Brokken, D., 44
Br uhwiler, E., see Saouma, V. E., 13
Bugeda, G., see O nate, E., 1
Carey, G. F., see Oden, J. T., 146
Carmeliet, J., 119
Carol, I., 145, 146, 148
Carol, I., see Ghavamian, S., see Rizzi, E., 120,
121, 145, 146
Cedolin, L., see Ba zant, Z. P., 117
Cervera, M., see Oliver, J., 1
Chaves, E., see Oliver, J., 1
Chorin, A., 72, 76
Cl ement, J. L., see Bod e, L., 1
Comi, C., 132
Corigliano, A., 99
Corigliano, A., see Bolzon, G., 1
Cosserat, E., 118
Cosserat, F., see Cosserat, E., 118
Criseld, M. A., 2, 6, 76, 78
Criseld, M. A., see Alfano, G., 13
De Proft, K., 119
De Wilde, W., see De Proft, K., 119
Diebels, S., see Vermeer, P. A., 50
Dietsche, A., 118
Dolbow, J., 21
Dolbow, J., see Mo es, N., 3, 13, 15, 21, 23, 28, 31
Duarte, C. A. M., 13, 16
Duarte, C. A. M., see Oden, J. T., 16
Dub e, J. F., 117
Duvaut, G., 12, 71, 73, 118
Dwivedi, S. K., see Espinosa, H., 1
Eberhardsteiner, J., see Mang, H. A., 11, 41, 74
Ehlers, W., see Vermeer, P. A., 50
168 Author/editor index
Eligehausen, R., see Ba zant, Z. P., 120
Engelen, R. A. B., 87, 99, 135, 148
Eringen, A. C., 141
Espinosa, H., 1
Ewalds, H. L., 138
Ewing, P. D., see Williams, J. G., 22
Feenstra, P. H., 1, 121
Fichant, S., 35
Fortin, M., see Brezzi, F., 132
Garaizar, F. X., 146
Garikipati, K., see Armero, F., 13, 16
Geers, M. G. D., 2, 7, 8, 41, 44, 48, 51, 56, 57, 67,
69, 87, 88, 99, 119, 120
Geers, M. G. D., see Borst, R. de, see Engelen, R.
A. B., see Peerlings, R. H. J., 9, 44,
56, 57, 61, 67, 87, 99, 121, 135, 137,
140, 146, 148
Georgin, J. F., 11, 74
Gerstle, W. H., 120
Ghavamian, S., 120, 121
Giessen, E. van der, see Tijssens, M. G. A., 12,
39
Gils, M. A. J. van, see Vree, J. H. P. de, 7
Goodman, R. E., 13
Govindjee, S., see Simo, J. C., 12, 73
Gray, W. G., 115, 116
Grootenboer, H. J., 1
Hashida, T., see Li, V. C., 35, 38
Heeres, O. M., see Borst, R. de, 121, 125, 130
Herrmann, H. J., see Vermeer, P. A., 50
Hillerborg, A., 92
Hinton, E., see Owen, D. R. J., 145, 146, 148
Hobbs, B. E., see Zhao, C., 24
Holzapfel, G. A., viii
Hordijk, D. A., 24, 53, 54, 64
Horii, H., see Kabele, P., 1, 35
Horrigmoe, G., see Zeng, L. F., 74
Huerta, A., see Rodrguez-Ferran, A., 120123
Huespe, A. E., see Oliver, J., 1
Hughes, T. J. R., 132
Hughes, T. J. R., see Chorin, A., see Simo, J. C.,
72, 76
Hulbert, G. H., 120
Hult, J., see Janson, J., 1
Iacono, C., 119
Ingraffea, A. R., see Bittencourt, T. N., see
Gerstle, W. H., 13, 120
Janson, J., 1
Jenq, Y. S., 12
Jir asek, M, see Ba zant, Z. P., 71, 118
Jir asek, M., viii, 1, 6, 23, 48, 50, 69, 72, 120
Ju, J. W., 1, 76, 117
Ju, J. W., see Simo, J. C., 72, 76, 117
Kabele, P., 1, 35, 38, 39
Kabele, P., see Simone, A., 13
Kalman, R. E., 99
Kanda, T., 35, 119
Kennedy, J. G., see Simo, J. C., 12, 73
Kim, B. S., see Eringen, A. C., 141
Kolar, R. L., see Gray, W. G., 115
Kr cek, J., see Audi, M., 29
Krenchel, H., see Li, V. C., 92
Kuhl, E., 67
Kuhl, E., see Mahnken, R., 119
Kullaa, J., 33, 34
Kusters, G. M. A., see Rots, J. G., 1, 117, 121
La Borderie, C., see Bod e, L., see Dub e, J. F., see
Fichant, S., 1, 35, 117
Lasry, D., 117, 118
Lee, P. C. Y., see Gray, W. G., 115, 116
Leijnse, A., see Gray, W. G., 115
Lemaitre, J., 72
Leroy, Y., see Ortiz, M., 146
Li, F., 119
Li, V. C., 5, 34, 35, 38, 92
Li, V. C., see Kanda, T., see Ward, R. J., 32, 35,
119
Li, Z., see Li, F., 119
Lin, F. B., see Ba zant, Z. P., 118
Lin, Z., see Kanda, T., 35, 119
Lions, J. L., see Duvaut, G., 12, 71, 73, 118
Lo, K. K., see Tvergaard, V., 146
Loret, B., 117
Luding, S., see Vermeer, P. A., 50
M ahler, L., see Runesson, K., 72
Mahnken, R., 119
Malvern, L. E., 19
Mang, H. A., 11, 41, 74
Mang, H. A., see Bi cani c, N., see Borst, R. de, 1,
13, 16, 74, 78, 110, 119121
Manzoli, O., see Oliver, J., 1
Mariani, S., see Corigliano, A., 99
Marsden, J. E., see Chorin, A., 72, 76
Mazars, J., 1, 7
McCracken, M. F., see Chorin, A., 72, 76
Melenk, J. M., 13, 16
Meschke, G., see Bi cani c, N., see Borst, R. de, 1,
13, 16, 74, 78, 119121
Mier, J. G. M. van, 2, 64, 71, 117, 137
Mier, J. G. M. van, see Iacono, C., 119
Mihashi, H., 92, 120
Mihashi, H., see Wittmann, F. H., 120
Author/editor index 169
Mish, K., see Belytschko, T., 120
Mo es, N., 3, 13, 15, 2123, 28, 31, 121
Mo es, N., see Dolbow, J., 21
Mohammadi, S., 1
Mr oz, Z., see Pietruszczak, S., 2, 117
M uhlhaus, H.-B., 86, 118
M uhlhaus, H.-B., see Borst, R. de, see Sluys, L.
J., see Zhao, C., 24, 117, 118, 146
M unz, T., 74, 78
Naaman, A. E., 5
Nasser, S. Nemat-, see Nemat-Nasser, S.
Nauta, P., see Rots, J. G., 1, 117, 121
Nechnech, W., see Georgin, J. F., 11, 74
Needleman, A., 117, 146
Needleman, A., see Ortiz, M., see Remmers, J. J.
C., see Tvergaard, V., see Xu, X.-P.,
12, 28, 39, 146
Nemat-Nasser, S., 146
Nomura, N., see Wittmann, F. H., 120
Oden, J. T., 16, 146
Oden, J. T., see Duarte, C. A. M., 13, 16
Oh, B., see Ba zant, Z. P., 2, 117
Okada, N., see Nemat-Nasser, S., 146
Olesen, J. F., see Stang, H., 92, 120
Oliver, J., 1, 13, 16
Oliver, J., see Simo, J. C., 13, 16
O nate, E., 1
O nate, E., see Owen, D. R. J., 71, 119, 145, 146,
148
Ord, A., see Zhao, C., 24
Ortiz, M., 146
Ottosen, N. S., see Runesson, K., 145, 146, 148
Owen, D. R. J., 71, 119, 145, 146, 148
Owen, D. R. J., see Mohammadi, S., 1
O zbolt, J., 120
O zbolt, J., see Ba zant, Z. P., 120
Pamin, J., 3, 4, 11, 41, 53, 67, 86, 87, 94, 118, 121,
135, 148
Pamin, J., see Borst, R. de, see Waszczyszyn, Z.,
29, 56, 57, 67, 117, 120
Papanastasiou, P., see Zervos, A., 146
Patz ak, B., see Jir asek, M., 120
Peerlings, R. H. J., 3, 8, 9, 4144, 47, 50, 56, 61,
67, 72, 96, 118, 121, 125127, 130,
133, 135137, 140, 146, 150
Peerlings, R. H. J., see Geers, M. G. D., see
Simone, A., 2, 8, 41, 44, 48, 56, 57,
67, 69, 87, 88, 99, 125
Peijs, A. A. J. M., see Geers, M. G. D., 119
Peijs, T., see Geers, M. G. D., 51, 56, 119
Peric, D., see Mohammadi, S., see Runesson, K.,
1, 145, 146, 148
Perzyna, P., 12, 71, 77, 118
Pietruszczak, S., 2, 117
Pijaudier-Cabot, G., 9, 41, 117, 118, 120, 125,
130, 132, 135, 150
Pijaudier-Cabot, G., see Ba zant, Z. P., see Bod e,
L., see Dub e, J. F., see Fichant, S., see
Mazars, J., 1, 7, 35, 117, 118, 120,
125, 130, 132
Prevost, J. H., see Loret, B., 117
Pulido, M. D. G., see Oliver, J., 1
Quarteroni, A., 131, 132
Ramm, E., see Kuhl, E., see Vermeer, P. A., 50,
67
Rammerstorfer, F. G., see Mang, H. A., 11, 41,
74
Reinhardt, H. W., see Naaman, A. E., 5
Remmers, J. J. C., 28, 29
Remmers, J. J. C., see Simone, A., 13, 105
Ren, Z., 1
Reynouard, J. F., see Georgin, J. F., 11, 74
RILEM, 120
Ristinmaa, M., see Runesson, K., 72
Rizzi, E., 145, 146
Rodrguez-Ferran, A., 120123
Roelfstra, P. E., 119
Rokugo, K., see Mihashi, H., 92, 120
Rots, J. G., 1, 2, 2830, 110, 117, 121
Runesson, K., 72, 145, 146, 148
Runesson, K., see M unz, T., 74, 78
Russel, H. G., 4
Saouma, V. E., 13
Schellekens, J. C. J., 30, 31, 105, 110
Schellekens, J. C. J., see Rots, J. G., 110
Schlangen, E., 60, 61, 118, 120, 121, 123
Sejnoha, M., see Audi, M., 29

Shah, S. P., 120
Shah, S. P., see Jenq, Y. S., 12
Simo, J. C., 1113, 16, 72, 73, 76, 117
Simone, A., 13, 41, 71, 105, 125, 135
Simone, A., see Alfaiate, J., 16, 19
Simons, D., 120
Sluys, L. J., 1, 71, 72, 87, 99, 117, 118, 146, 149
Sluys, L. J., see Alfaiate, J., see Askes, H., see
Borst, R. de, see Georgin, J. F., see
Iacono, C., see Simone, A., see
Tijssens, M. G. A., see Wang, W. M.,
see Wells, G. N., see De Proft, K., 1,
3, 1117, 19, 2124, 28, 31, 32, 34,
39, 41, 51, 52, 71, 74, 87, 117119,
121, 125, 135, 146
Souza, J. L., see Bittencourt, T. N., 13
Speziale, C. G., see Eringen, A. C., 141
170 Author/editor index
Stakgold, I., 115
Stang, H., 92, 120
Stang, H., see Li, V. C., 92
Steinmann, P., see Dietsche, A., 118
Su arez, B., see Owen, D. R. J., see O nate, E., 1,
71, 119
Swartz, S. E., see Shah, S. P., 120
Tabbara, M., see Pijaudier-Cabot, G., 117
Tailhan, J. L., see Bod e, L., 1
Taylor, R. L., see Goodman, R. E., see Simo, J.
C., see Zienkiewicz, O. C., 11, 13,
132
Tijssens, M. G. A., 12, 39
Trangenstein, J., see Garaizar, F. X., 146
Tvergaard, V., 146
Tvergaard, V., see Needleman, A., 146
Valli, A., see Quarteroni, A., 131, 132
Vardoulakis, I., see M uhlhaus, H.-B., see
Zervos, A., 118, 146
Vermeer, P. A., 50
Vliet, M. R. A. van, 120
Vree, J. H. P. de, 7
Vree, J. H. P. de, see Peerlings, R. H. J., 8, 9,
4143, 47, 50, 72, 96, 118, 121,
125127, 130, 133, 135, 136, 150
Wang, M. L., see Shah, S. P., 120
Wang, W. M., 82, 117
Wanhill, R. J. H., see Ewalds, H. L., 138
Ward, R. J., 32
Waszczyszyn, Z., 29, 120
Wawrzynek, P. A., see Bittencourt, T. N., 13
Wells, G. N., 1, 3, 14, 15, 17, 2124, 28, 31, 32,
34, 45, 51, 52, 71, 86, 87, 121, 146
Wells, G. N., see Remmers, J. J. C., see Simone,
A., see De Proft, K., 13, 29, 41, 71,
105, 119
Willam, K., see Carol, I., see Dietsche, A., see
Rizzi, E., see M unz, T., 74, 78, 118,
145, 146, 148
Williams, J. G., 22
Wills, J., see Criseld, M. A., 2
Wittmann, F. H., 119, 120
Wittmann, F. H., see Roelfstra, P. E., 119
Wu, H. C., see Li, V. C., 5, 34, 38
Wu, K., see Zhang, D., 120
Wunderlich, W., 3, 4, 41, 53, 67
Xu, X.-P., 12, 39
Yamaguchi, E., see Kabele, P., 1
Zavattieri, P. D., see Espinosa, H., 1
Zeman, J., see Audi, M., 29
Zeng, L. F., 74
Zervos, A., 146
Zhang, D., 120
Zhao, C., 24
Zienkiewicz, O. C., 132
Zienkiewicz, O. C., see Oden, J. T., 16
Zimmermann, T., see Jir asek, M., 1, 23, 69
Subject index
applications
composite compact-tension specimen,
5660, 8789
concrete beam in four-point bending,
5356
damage characterisation in mode I
problems, 137
damage characterisation in shear band
problems, 145
discontinuities
dummy stiffness, 2831
cohesive, 3238, 9295
traction-free, 5369, 8798
linear-elastic analysis of a notched beam,
2831
single-edge notched beam, 6064,
121123
steel bre-reinforced concrete, 9295
strip footing near a slope, 8992
Babu ska-Brezzi condition, 132
bumps, see discontinuity and bumps
comparisons
Duvaut-Lions vs Perzyna
rate-dependent
damage-elastoplasticity, 78, 82
interface vs PU-based elements, 105111
Mazars vs modied von Mises
equivalent strain, 121123
rate-dependent model vs
gradient-enhanced model, 8991,
9698, 121123, 145150
continuum models
considerations on numerical modelling
of concrete, 119121
damage, 79
for softening materials, 89, 12, 117
gradient-enhanced damage, 42
limitations, 14, 5356, 103, 119121,
135151
plasticity, 912
rate-dependent damage-elastoplasticity
Duvaut-Lions, 73, 102
general formulation, 72
Perzyna, 76
rate-dependent elastoplasticity, 86
reassessment of model parameters, 56,
61, 102
coupled problem, 130133
crack propagation
shock-wise, 6569, 98
smooth, 98
damage evolution laws
exponential softening, 8, 44, 53, 64
linear softening, 8, 127
modied power softening, 8, 57, 64, 67
discontinuity
and boundary conditions, 16, 17, 41, 43
and bumps, 59, 67, 68, 98
and drops, 55, 61, 64, 87, 88
and non-locality, 6569, 9698, 102, 137
and rate-dependence, 9698
cohesive, 94
direction, 22, 51, 85, 90
in problem elds, 14, 42
insertion, 21, 50, 85
motivations, 14
quality of the discontinuous
enhancement, 103, see
discontinuity and bumps, see
discontinuity and drops
requirements on the underlying
continuum model, 6
distributed failure in discontinuous setting,
102
drops, see discontinuity and drops
element size vs length scale, 67
energy
complementary energy, 131
total potential energy, 131
equivalent strain denition
von Mises, 139, 146
modied, 7, 53, 61, 121
positive principal strains, 7, 57, 121, 142
fracture energy, 120
Helmholtz equation, 9, 43, 45, 125, 137
homogenisation, 119
172 Subject index
inf-sup condition, see Babu ska-Brezzi
condition
interpolation
linear-linear, 127
quadratic-linear, 130
requirements for gradient-enhanced
media, 125133
inverse methods, 119, 120
knowledge, imperfect, 119
Lagrangian functional, 132
Lagrangian multiplier, 132
length scale
constant, 53
variable, 68, 69
localisation limiters, see continuum models,for
softening materials
mixed method, 130133
modelling
macromechanical approach, 119
micromechanical approach, 119
numerical integration, 23, 2831, 105113
oscillations
in stress, 125, 127
in non-local equivalent strain, 6669,
9698
in traction prole, 28
panacea for discrete cracking, 102
realistic physical situation?, 14, 6469, 71,
135151
regularised models, see continuum models,for
softening materials
softening, 117
strategies for failure analysis, 46
traction-separation law, 12
transfer of history data, 24, 52
yield function
von Mises, 10, 78, 82, 89, 149
Rankine, 11, 87, 94, 98, 121
Summary
Continuous-Discontinuous Modelling of Failure
by A. Simone
The foundations of a safe structural design lie on the understanding of fail-
ure processes of engineering materials and in their correct representation.
In a numerical context, failure representation in engineering materials can
be pursued either in a continuous or in a discontinuous setting. Both ap-
proaches can model certain failure modes, but in some cases do not reect
the physical processes behind failure properly. To some extent, continuous
failure representation can be improved by enriching the standard kinemat-
ics with displacement discontinuities, which can be thought of as a natural
consequence of material failure processes.
In this work, a continuous-discontinuous approach is applied to elas-
tic, strain-hardening and regularised strain-softening media. It is shown
that the success of a continuous-discontinuous analysis depends largely
on the underlying continuous model. Analyses performed with elastic and
strain-hardening media gave satisfactory results; conversely, when strain-
softening media were considered, the performance of the approach was re-
lated to the nature of the regularisation employed. It is shown that, in a
continuous-discontinuous setting, softening models in which the underly-
ing continuum description is enriched through a temporal regularisation
formalism (rate-dependence) perform better than models obeying a spatial
regularisation concept (non-locality). This issue is examined with respect to
a differential version of a non-local model (implicit gradient-enhanced dam-
age continuum model) and a rate-dependent elastoplastic-damage model.
More specically, it is shown that a class of regularised models based on
a non-local dissipation driving variable is not adequate for failure descrip-
tion in arbitrary loading scenarios. Several applications illustrate the im-
proved exibility of a continuous-discontinuous approach to failure when
compared to a continuous approach alone.
Samenvatting
Continue-Discontinue Modellering van Bezwijkgedrag
door A. Simone
Een veilig constructief ontwerp is gebaseerd op een goed begrip van
bezwijkprocessen van technische materialen en op hun correcte represen-
tatie. Het bezwijkgedrag van technische materialen kan in een numerieke
context worden beschreven in een continu of in een discontinu kader. Beide
aanpakken zijn ideaal voor de modellering van bepaalde bezwijkmechanis-
men, maar weerspiegelen in sommige gevallen de fysische processen niet
correct. Continue representaties van bezwijken kunnen tot op zekere hoogte
worden verbeterd door de standaard kinematica te verrijken met verplaat-
singsdiscontinuteiten, hetgeen kan worden beschouwd als een natuurlijke
consequentie van bezwijkprocessen in materialen.
In deze studie is een continue-discontinue aanpak toegepast van elasti-
sche, strain-hardening en strain-softening media. Aangetoond is dat het suc-
ces van een continue-discontinue analyse grotendeels afhangt van het on-
derliggende continue model. Analyses met elastische en strain-hardening
media vertoonden bevredigende resultaten; wanneer echter strain-softening
media worden beschouwd hangt de prestatie van de aanpak af van het
type regularisatie dat is gebruikt. Voor een continu-discontinu raamwerk
is aangetoond dat softening modellen, waarvan het onderliggende con-
tinu um is verrijkt via een tijdsregularisatietechniek (snelheidsafhankelijk-
heid), beter presteren dan modellen met een ruimtelijk regularisatieconcept
(niet-localiteit). Dit aspect is onderzocht voor een differentiaalversie van een
niet-locaal model (impliciet gradi entverrijkte schade continu ummodel) en
een snelheidsafhankelijk elastoplastisch schademodel. In het bijzonder is
getoond dat een bepaalde klasse van geregulariseerde modellen, gebaseerd
op een niet-locale dissipatiesturende variabele, niet geschikt zijn voor het
beschrijven van bezwijkgedrag in willekeurige belastingsscenarios. Di-
verse toepassingen illustreren de verbeterde exibiliteit van een continue-
discontinue benadering van bezwijkgedrag vergeleken met een continue be-
nadering als zodanig.
Curriculum vitae
Dec. 19
th
, 1973 born in Taranto, Italy, as Angelo Simone
June 1992 maturit` a scientica
Oct. 1998 laurea in ingegneria civile, Politecnico di Milano
June 1999 Sep. 2003 research assistant, Faculty of Civil Engineering and
Geosciences, Delft University of Technology

Ceg Simone 20031013

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

Ceg Simone 20031013

Uploaded by

Copyright:

Available Formats

Continuous-Discontinuous

together with the KOMA-

K partition of K in interface and PU-based discontinuous elements

body volume or surface or length (boundary included)

= f / to the yield surface f is

Figure 1.5 Smoothed Rankine yield function [128].

Displacement discontinuities are incorporated in standard nite elements

Based on References [168, 171, 174]

Figure 2.1 Body

). The boundary surface of the body

) and u and u are continuous func-

(note that the truss length is L/2).

The performance of some non-local models, in either integral [139] or dif-

Based on References [175, 176]

Figure 3.3 From continuous to discontinuous displace-

, see Figure 3.4). This criterion can be justied by

Figure 3.19 Final failure state for (a) the continuous-discontinuous

2 mm; see Figure 3.21b). The load is applied at the

A displacement discontinuity can be considered as originating either from

Based on References [172, 173, 177]

, and of the fourth-order consistent tangent tensor

and I is the fourth-

cohesive discontinuity softening continuum

Figure 4.19 From continuous to discontinuous failure description in SRF-concrete

Figure 4.20 Load-deection curves for

The performance of PU-based discontinuous elements deteriorates when

Based on References [168, 171]

according to the node numbering of Figure A.1. For a conventional linear

). The Dirac-delta func-

). A regular function (x) is dened in the domain and

Some properties of the gradient-enhanced continuum damage model pro-

Based on Reference [169]

Correct failure characterisation, in terms of damage initiation and propa-

Based on Reference [170]

Sejnoha, M., see Audi, M., 29

You might also like