You are on page 1of 28

The current issue and full text archive of this journal is available at

www.emeraldinsight.com/0264-4401.htm

EC
26,1/2 Novel anisotropic
continuum-discrete damage
model capable of representing
100
localized failure of massive
structures
Part I: theoretical formulation and numerical
implementation
D. Brancherie
Laboratoire Roberval, Université de Technologie de Compiègne, Compiègne,
France, and
A. Ibrahimbegovic
Laboratoire de Mécanique et Technologie, École Normale Supérieure de Cachan,
Cachan, France

Abstract
Purpose – The purpose of this paper is to present a finite element model capable of describing both
the diffuse damage mechanism which develops first during the loading of massive brittle structures
and the failure process, essentially due to the propagation of a macro-crack responsible for the
softening behaviour of the structure. The theoretical developments for such a model are presented,
considering an isotropic damage model for the continuum and a Coulomb-type criterion for the
localized part.
Design/methodology/approach – This is achieved by activating subsequently diffuse and
localized damage mechanisms. Localized phenomena are taken into account by means of the
introduction of a displacement discontinuity at the element level.
Findings – It was found that, with such an approach, the final crack direction is predicted quite well,
in fact much better than the prediction made by the fracture mechanics type of models considering
combination of only elastic response and softening.
Originality/value – The presented model has the potential to describe complex damage phenomena
in a cyclic and/or non-proportional loading program, such as crack closing and re-opening, cohesive
resistance deterioration due to tangential sliding, by using only a few parameters compared to the
traditional models for cyclic loading.
Keywords Modelling, Structural engineering, Structural design, Failure (mechanical)
Paper type Research paper

Engineering Computations: 1. Introduction


International Journal for
Computer-Aided Engineering and Ever increasing demand to achieve a more economical structural design requires that a
Software better understanding be obtained for the non-linear behavior of a particular structural
Vol. 26 No. 1/2, 2009
pp. 100-127
q Emerald Group Publishing Limited
0264-4401
This work was supported by the French Ministry of Research and by the project “ACI jeunes
DOI 10.1108/02644400910924825 chercheurs”. This support is gratefully acknowledged.
system and reliable estimate of its limit load be furnished. However, the limit state of a Continuum-
complex system often implies that particular components – whose peak resistance has discrete damage
already been defeated – be able to function in the post-peak regime (also called strain
softening). A reliable evaluation of the limit load of the considered system is thus model: Part I
highly dependent upon an efficient and reliable description of the softening branch of
the material behavior.
It is well known by now (Needleman, 1988; de Borst et al., 1992; Bažant et al., 1984; 101
Pijaudier-Cabot and Bažant, 1987) that the strain softening phenomena cannot be
reliably represented by using classical continuum mechanics models without severe
mesh dependencies of the computed response.
This has led to the development of a number of remedies aiming at representing
objectively, with respect to the chosen mesh discretization, the post-peak response of
the material and/or structure. Most of those techniques, usually referred to as
“localization limiters”, often lead to significant increase of complexity in formulating
the FE problem and solving it. Among them, we can mention: non-local continuum
(Pijaudier-Cabot and Bažant, 1987), higher order gradient damage (Muhlhaus and
Aı̈fantis, 1991), viscoplastic regularization (Needleman, 1988), crack-band models
(Hillerborg et al., 1976), etc.
A new family of methods, introduced more recently, is based on a modification of
classical continuum models allowing discontinuities either in the strain field (weak
discontinuities) (Ortiz et al., 1987) or in the displacement field (strong discontinuities)
(Simo et al., 1993; Armero and Garikipati, 1995; Wells and Sluys, 2001). The main
advantage of those modified continuum models is to provide an adequate measure of
the total dissipation with respect to the chosen finite element discretization. However,
the vast majority of the previous works on the subject only consider problems where
dissipation in the structure can be attributed only to the strain localization zone. Such
models only consider the combination of the discontinuity with an elastic behavior of
the continuum part of the material. Thus, the inelastic response which can develop
outside the localization is completely ignored.
The latter appears insufficient for dealing with massive structures. Indeed, massive
structures often present at the crack tip a “process zone” where damage develops,
induced by the creation of micro-cracks, until the macro-crack forms due to the
coalescence of some micro-cracks. The dissipation produced in the process-zone is far
from being negligable with respect to the total dissipation and thus it has to be taken
into account by exploiting a continuum damage model.
The main contributions in this work can be stated as follows:
.
We present a model capable of taking into account both the contribution of
diffuse dissipative mechanisms (cf. micro-cracks) accompanied by strain
hardening as well as localized damage modes with strain softening effect. This is
achieved by introducing displacement discontinuities coupled with a continuum
damage mechanics model. The combination of the two types of dissipative
mechanisms is taken into account by building adequate continuum model to
describe diffuse dissipation and an adequate localized or discrete model to deal
with the dissipation taking place in localized zones.
.
The combined result of two dissipative mechanisms can also be interpreted as an
alternative approach to constructing an anisotropic damage model to a number
of only partially successfully works using only the classical continuum
EC mechanics framework (Rots et al., 1985; Govinjee et al., 1995; La Borderie, 1991),
26,1/2 where the macro-crack creation is guided by the micro-cracking phase. One can
thus obtain not only a more robust implementation, but also a more reliable
estimate of the final orientation of the macro-crack which follows the
corresponding stress redistribution in the micro-cracking phase.
.
Finally, we present a multi-surface damage model taking into account the
102 contribution of the localized dissipative mechanism, which accounts both for
normal interface and tangential interface damage modes.
The outline of the paper is as follows. In Section 2, we present the theoretical formulation of
both the bulk damage model and the discontinuity damage model as well as the
modifications induced by the introduction of a displacement discontinuity. In Section 3,
the key points of the finite element implementation are presented, with particular
developments dedicated to the necessary modifications of the solution strategy due to the
introduction of a displacement discontinuity. In Section 4, results of several numerical
simulations are presented. Finally, some closing remarks are given in Section 5.

2. Theoretical formulation
In this section, we first present the ingredients for the continuum models needed to
describe the diffuse damage mechanism. Second, we will present the ingredients for
constructing the response of the surface of discontinuity. In order to distinguish
variables associated to the bulk material and variables associated to the surface of
discontinuity, we will note them, respectively, with one or two bars superposed.

2.1 Continuum damage model


The model presented here is an isotropic damage model with isotropic “hardening”. Such
a model is assumed to represent the progressive nucleation of micro-cracks in the bulk
material distributed of more or less random orientation which can be considered as
inducing the damage in roughly isotropic way. The internal variables of the model are the
(forth order) compliance tensor denoted as: D  and the hardening variable denoted as j.
The Helmholtz free energy of such a model can be written as:

cð1;  jÞ ¼ 1 1 : D
 D;  21 : 1 þ Xð
 jÞ ð1Þ
2|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}

c^ð1; DÞ

where Xð jÞ is the function describing hardening. The damage function defining the
“elastic” domain is written in a generalized way as:

fðs; qÞ ¼ f^ðsÞ 2 ðsf 2 q Þ # 0 ð2Þ


where f^ is a homogeneous function of degree one, which implies that:

›f^
s ¼ f^ðsÞ ð3Þ
›s
where sf is the initial damage threshold and q is the stress-like variable associated
with hardening. The damage-induced dissipation can then be written by appealing to
the second law of thermodynamics as:
 ¼ s : 1_ 2 d   
cð1;
Continuum-
0#D  D; jÞ ð4Þ
dt discrete damage
By exploiting further the Legendre transformation of c^ in order to define the model: Part I
complementary energy: x :

 ¼ s : 1 2 c^ ð1;  1  :s
xðs; DÞ  DÞ 5 s:D
2
ð5Þ 103
The dissipation inequality can further be reduced to:
d  2 s_ : 1 2 s : 1_ 2 d Xð
 ¼ s : 1_ þ
0#D xðs; DÞ  jÞj_
dt d j
›x ›x _ d
¼ 2s_ : 1 þ :sþ  :D 2  Xð  jÞj_ ð6Þ
›s ›D dj
¼ 2s_ : 1 þ D : s : s_ þ 1 s : D
_ : s 2 d Xð  jÞj_
2 d j
In order to determine further the state equation and evolution law, two cases have to be
treated:
(1) the system is supposed to remain elastic with no evolution of internal variables:
_ ¼ 0;
D j_ ¼ 0 ð7Þ
with zero inelastic dissipation which can be written as:
D  : s
 ¼ 0 ¼ s_ : ½21 þ D ð8Þ
Since the last result must hold for any s_ we can thus obtain the first state
equation:
›cð1;  jÞ
 D;
 21 : 1 ¼
s¼D ð9Þ
›1
A similar state equation is postulated for the stress-like hardening variable q :
d
q ¼ 2  Xð  jÞ ð10Þ
dj

(2) By assuming further that those results remain valid when defining the
 for an inelastic process we can obtain that the latter reduces to:
dissipation D
D ¼ 1s : D
_ : s þ q j_: ð11Þ
2

In the inelastic damage process of this kind, the internal variables are modified:
_ – 0;
D j_ – 0 ð12Þ
and the dissipation in the material is positive. In order to provide the evolution of
internal variables, we consider here the principle of maximum damage dissipation
resulting with an associated damage model. In other words, among all the values of the
EC couple of dual variables which must remain admissible with respect to given
damage criterion, we select those which are the solution of the following maximisation
26,1/2 problem:
 s; q Þ
max Dð ð13Þ
fðs;qÞ

104 By appealing to the Lagrange multiplier method, the previous problem can be replaced
by a maximisation problem without constraints: the couple of dual variables ðs; q Þ is
solution of:
max min ½2Dð  s; q Þ þ gfðs; q Þ ð14Þ
g$0 ðs;qÞ |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
 s;qÞ

where g is the Lagrange multiplier. The Kuhn-Tucker optimality conditions of this


problem allow us to write the evolution laws of internal variables as:
 s; q Þ
›Lð  s; q Þ
›Dð ›fðs; q Þ _ : s ¼ g ›fðs; q Þ
¼0¼2 þ g )D
›s ›s ›s ›s
ð15Þ
 s; q Þ
›Lð  s; q Þ
›Dð ›fðs; q Þ › 
f
¼0¼2 þ g ) j_ ¼ g
›q ›q ›q ›q
The result in equation (15) is obtained by exploiting the fact that f^ is a homogeneous
function of degree one which allows us to write the evolution of damage model
compliance as:
 
s: D _ 2 g ›f ^ ›f 1 :s¼0 ) D _ 5 g ›f ^ ›f 1 ð16Þ
›s ›s f^ðsÞ ›s ›s f^ðsÞ
The Kuhn-Tucker conditions of the maximisation problem under consideration also
include the loading/unloading conditions:
g $ 0; fðs; q Þ # 0; gfðs; q Þ ¼ 0 ð17Þ
Finally, the model description is completed by the consistency condition which takes
the following form:

gf_ ¼ 0 ð18Þ
The latter equation gives the only non-zero value of g for the case of damage
loading:
›f ›f _
f_ ðs; q Þ ¼ 0 ¼ : s_ þ q ð19Þ
›s ›q
From the state equations (9) and (10), we can further write:
8
< s_ ¼ D 21 : 1_ þ D _ 21 : 1
2  ð20Þ
: q_ ¼ 2 › Xð2jÞ j_
›j
and noting that: Continuum-
 :D
D  21 ¼ I ) _ 21 ¼ 2D
D _ : D
 :D  ð21Þ
discrete damage
model: Part I
Equation (19) can be rewritten as:
" #
_ ›f _  jÞ _
›f ›2 Xð

f¼  21 _   
: D : 1 2 D : D : |ffl{zffl}
D : 1 2 j ð22Þ 105
›s ›q ›j2

s

By exploiting the evolution equation (15) and denoting K ¼ ›2 X=


 ›j2 , we finally
can deduce the value of the Lagrange multiplier g for the damage step
according to:
 21 : 1_
ð›f=›sÞ : D
g ¼ ð23Þ

ð›f=›sÞ : D : ð›f=›sÞ þ Kð
21  ›f=›q Þ2

Combining those results we can write the stress-rate constitutive equations giving
the evolution of s with respect to the evolution of 1:
8
> 21
< s_ ¼ D : 1_ g ¼ 0
h i
21
21 21
ðD :ð›f=›sÞÞ^ðD :ð›f=›sÞÞ ð24Þ
: s_ ¼ D 2 ð›f=›sÞ:D 21 :ð›f=›sÞþKð›f=›qÞ2 : 1_ g . 0
>

The damage model developed herein is very general in the sense that it applies to
any damage criterion. A particular choice made in this work is further discussed
in Section 2.3.

2.2 Discrete damage model


The damage model presented in the previous section is capable of describing the
diffuse mechanisms corresponding to the microcracking in the process zone of the bulk
material. We present here the development to construct a model for a macrocrack by
using the displacement discontinuity surface. The model of this kind is thus able to
represent all the salient phenomena which appear in the damage process of a massive
structure. In particular, one can consider that the displacement discontinuity will take
into account the dissipation due to all the mechanisms taking place at the scale of
the localization zones without going into details of such a damage process. In that
sense, the main input parameter is the fracture energy density.
Then, the displacement field can be written as:
uðx; tÞ ¼ uðx;
^  ðx; tÞðHGs ðxÞ 2 wðxÞÞ
tÞ þ u ð25Þ
where Gs is the displacement surface of discontinuity splitting the domain V into two
sub-domains Vþ and V2 as shown in Figure 1, whereas wðxÞ is a at least C 0 function,
permitting the displacement boundary conditions on ›V not to be perturbed by the
presence of the displacement discontinuity:
(
0 x [ ›V 2
wðxÞ ¼ ð26Þ
1 x [ ›V þ
EC where:
26,1/2 ›V2 ¼ ›V > V2 ›Vþ ¼ ›V > Vþ
Equation (25) can then be rewritten as:
uðx; tÞ ¼ uðx;
  ðx; tÞHGs
tÞ þ u ð27Þ
106  ðx; tÞwðxÞ is the continuous part of the displacement field.
 tÞ ¼ uðx;
where uðx; ^ tÞ 2 u
This displacement field leads to the following expression of the strain field:
1ðx; tÞ ¼ 7s uðx; tÞ ¼ 7s uðx;
 tÞ þ ðu  ^nÞs dGs ðxÞ ð28Þ
|fflfflfflfflffl{zfflfflfflfflffl}
1ðx;tÞ


By assuming that the result on constitutive equation in equation (9) still applies, we have:
1ðx; tÞ ¼ 1ðx;  ^nÞs dGs ðxÞ ¼ D : s
 tÞ þ ðu ð29Þ
By taking into account that the stress field must remain bounded, the last result implies
that:
D¼D d
 þD ð30Þ
G s

from where, we can identify:


(  :s
1 ¼ D on V\Gs
 :s ð31Þ
 ^nÞs ¼ D
ðu on Gs

 ^nÞs dGs , the


As the strain field is decomposed into a regular part 1 and a singular part ðu
hardening variable j should also be split into a regular part j and a singular part jd  :
Gs
Considering these results, we can also introduce an additive split of the Helmoltz free
energy into a regular and a singular part:

cð1; D; jÞ ¼ cð1;  jÞ þ c ðu


 D;  jÞd ðxÞ
 ; Q; ð32Þ
Gs

with:
8
< cð1;
 D; jÞ ¼ 1 1 : D
 21 : 1 þ Xð
 jÞ
2
 jÞ ¼ 1 u  21 · u  ðjÞ ð33Þ
: c ðu
 ; Q;
2
 ·Q  þX

m
n

Ω+
s
Figure 1. Ω– Γs
Slip line Gs separating the
domain into Vþ and V2
where we introduced a new internal variable for describing the damage response at the
 ¼ ðn · D
 21 · nÞ21 :
Continuum-
discontinuity: Q
The dissipation inequality can then be written showing that the dissipation appears
discrete damage
decomposed into two parts as: model: Part I
0#D¼D  þDd
Gs
   
d    _

d   

ð34Þ 107
_
¼ s : 1 2 cð1;  D; jÞ þ t Gs · u 2 cðu; Q; jÞ dGs
dt dt
where the first term is the bulk dissipation, due to diffuse damage mechanisms,
and the second is the localized dissipation, due to the development of localization
zones.
Considering further the case where only D is non-zero, we have:
 _   ðjÞ _
d 2X
0#D Gs
 21 · u
 ¼ ðt 2 Q _ þ 1 u
Þ·u  21 · u
· Q  2 j;
2 dj 2
 ðjÞ
_  21  d 2 X
 21 · u
¼ ðt Gs 2 Q _ þ 1 u
Þ·u  21 · Q
 ·Q ·Q ·u 2 ; ð35Þ
2 d j 2
 ðjÞ
d 2X
 21 · u
¼ ðt Gs 2 Q _ þ 1 t G · Q
Þ·u _
· t Gs 2
dj 2
s
2
We also assume that the damage function for the surface of discontinuity can be
written as:
^
f ðt Gs ; qÞ ¼ f ðt Gs Þ 2 ðs f 2 qÞ # 0 ð36Þ
^
where f ðt Gs Þ is a homogeneous function of degree one, s f is the initial damage
threshold and q is the traction-like variable associated to softening phenomena at the
discontinuity. As for continuum damage models, we have then to distinguish between
two different cases: the first with f , 0 where the discontinuity is assumed to evolve
elastically. In this case, no dissipation and no evolution of internal variables will occur
with:
_ _
 ¼ 0;
D Q ¼ 0; j ¼ 0 ð37Þ
so that equation (35) reduces to equality leading to the corresponding form of
constitutive equation:

 21 · u
 ¼ 0 ¼ ðt 2 Q  21 · u
_ ) t G ¼ Q ›c
D Gs Þ·u  ¼ ð38Þ
s

›u
By introducing a similar constitutive equation for the traction-like variable q :

dX
q ¼ 2 ð39Þ
dj
and by assuming that these constitutive equations apply in damage process, we can
EC obtain from equation (35) the total inelastic dissipation at discontinuity as:
26,1/2 D ¼ 1t ·Q
_ · t þ qj_ ð40Þ
G Gs
2 s
 and j; in order to
It only remains to determine the evolution of the internal variables Q
master the damage evolution of the discontinuity. Then, as for continuum damage
models, the principle of maximum damage dissipation on the discontinuity can
108 be invoked resulting with the associated damage evolution equations. More precisely,
the couple ðt Gs ; q Þ has to verify:

max minLðt    Þ þ g f ðt Gs ; q Þ
Gs ; qÞ L ¼ 2Dðt Gs ; q ð41Þ
g.0 ðt Gs ;qÞ

where g is the Lagrange multiplier introduced for the discontinuity.


The stationarity of the functional L leads to the following Kuhn-Tucker equations:
8
> _ 1 ›f ›f
< Q ¼ g f^ ›t Gs ^ ›t Gs
>
ð42Þ
> _ ›f
: j ¼ g ›q
>

In order to obtain the last result we have exploited the expression of the dissipation in
^
equation (40) and have taken into account that f is a homogeneous function of
degree one.
The Kuhn-Tucker conditions above are completed by the loading/unloading
conditions as:
f # 0; g $ 0; g f ¼ 0 ð43Þ
As for continuum damage models, the only non-zero value of the Lagrange multiplier
occurs for the damage loading at the discontinuity. In that case, we can write the
consistency condition which applies at the discontinuity:

_ _ ›f _ ›f _
g . 0 and g f ¼ 0 ) f ¼ 0 ¼ · t Gs þ q ð44Þ
›t G s ›q
The last result and a couple of auxiliary results simple to obtain allow us to write:
8 _  21 
>  21 · u
t_ Gs ¼ Q  21 · Q
_ 2 Q ·Q ·u
>
>
>
>
< ›2 X _
q_ ¼ 2  ð45Þ
> › 2 j
j
>
> |{z}
>
>
:  
K

and to obtain the Lagrange multiplier g :


 21 · u
ð›f =›t Gs Þ · Q _
g ¼   ›f =›q Þ2
ð46Þ
ð›f =›t Gs Þ · Q 21 · ð›f =›t Gs Þ þ Kð
with these results in hand, we can finally write the stress-rate constitutive equations as:
8
>
>
 21 · u
Q _ g ¼ 0 Continuum-
< 
t_ Gs ¼  21 · ð›f =›t ÞÞ^ðQ  21 · ð›f =›t ÞÞ ð47Þ discrete damage
 21
> Q 2  ðQ Gs Gs  g ¼ 0
>
:  21 · ð›f =›t ÞþKð
ð›f=›t Gs Þ · Q  ›f =›qÞ2 u model: Part I
Gs

The latter development applies for very general damage model and any particular form
of the discontinuity surface. 109
2.3 Choice of damage criteria
All the foregoing developments where presented assuming no particular form of damage
functions neither for the continuum part nor for the localized part of the model. In the
following, we discuss a couple of particular forms which are used in our computations.
2.3.1 Continuum damage model. By assuming that the continuum damage describes
the initial damage phase with no preferred direction, we can choose an isotropic
damage model. With respect to the developments in Section 2.1, this implies a
particular choice for the damage function f, defining the elastic domain:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1
fðs; q Þ ¼ s : D e : s 2 pffiffiffiffi ðsf 2 q Þ ð48Þ
E
where D e is the undamaged elastic compliance of the bulk material and E is the Young
modulus. With such a choice, we can easily compute that:
›f De : s ›f 1
¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi and ¼ pffiffiffiffi ð49Þ
›s s : De : s ›q E
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
where s : D e : s defines a norm in the stress space, which is further denoted as
k · kD e .
The evolution equations of internal variables in equation (15) will thus simplify to:
8_ ›f ›f 1
 De
>
< D ¼ g kskD e ¼ g ›s ^ ›s f^
 ð50Þ
>
: j ¼ g ››fq ¼ g p1ffiffiffi E

By introducing generalized Lagrange multiplier m_ ¼ ðg=kskD e Þ, expression (50) can


also be written as:
D_ ¼ m_ D e ð51Þ
By further integrating this equation in the interval ½0; t we further obtain:
Z t Z t
_Dð
 tÞd t ¼ DðtÞ
 2 Dð0Þ
 ¼ m_ D e dt ¼ ½mðtÞ 2 mð0ÞD e ð52Þ
0 0

By taking into account that Dð0Þ ¼ D e is the damage-free elastic compliance and
mð0Þ ¼ 0, we can finally obtain:
 ¼ ½1 þ mðtÞD e
DðtÞ with m [ ½0; þ1 ð53Þ
which confirms the isotropic character of the chosen damage model. We note in
EC passing that the chosen damage function allows to reformulate, the model in terms of
the effective stresses (Lemaitre, 1992). Namely, by introducing damage variable d as:
26,1/2
m
d¼ ; d [ ½0; 1 ð54Þ
1 þ m

110 we can rewrite the damage evolution according to:

C ¼ D 21 ¼ ð1 2 d ÞC e ð55Þ

which, in accordance with the result in equation (9) further gives:

s ¼ ð1 2 d ÞC e : 1 ð56Þ
Finally, the stress rate constitutive equation can now be written:
8 1 e
< s_ ¼ 1þm C : 1_
> g ¼ 0
h 2
i ð57Þ
1 e ðs^sÞ=ð1þmÞ
>
: s_ ¼ 1þm C 2 ð1=1þmÞkskD e þð1=EÞK : 1_ g . 0

In our case, we have chosen Kð ›2 Xð


 jÞ=›j2 Þ as a constant so that the stress-like variable
q is obtained from a linear hardening law:

q ¼ 2K j ð58Þ

2.3.2 Localized damage model. In this part, we will develop the particular damage
model chosen and implemented to describe the behavior of the cohesive discontinuity.
As we are dealing with quasi-brittle materials, the preponderant rupture mode that
have to be reproduced is normal interface crack, i.e. mode I (Edwards and Wanhill,
1986). We consider that the displacement discontinuity is introduced as soon as the
maximal principal stress reaches the given fracture stress s f .
Two possible ways can be followed to determine the criterion for the inception of the
displacement discontinuity and its orientation via the normal vector n (Figure 1). The
first way pertains to the classical localization criterion in the continuum models, which
is thoroughly studied by Oliver (2000) and coworkers. In that case, the behavior of the
discontinuity surface is deduced from the behavior of the bulk material by assuming
that equations written in the classical continuum models are still valid when dealing
with the strong discontinuities and the behavior of the surface of discontinuity is
constructed by exploiting the fact that stresses and tractions must remain bounded on
the discontinuity.
An alternative way which is chosen herein to determine the orientation of the
discontinuity and the moment at which it occurs is to postulate them independently
from the continuum model and thus to identify the behavior that has to be reproduced
by the discontinuity and then to deal with the surface of discontinuity simply as a
cohesive dissipative interface. The discontinuity initiation is based on the principal
stress values analysis. Denoting by n the direction of this principal stress and the
normal unit vector to the discontinuity at point x, we construct the damage criterion
related to the discontinuity according to:
f 1 ðs; qÞ ¼ t Gs · n 2 ðs f 2 q Þ # 0 ð59Þ Continuum-
|fflffl{zfflffl}
^
discrete damage
f 1 ðt Gs Þ
model: Part I
where s f is the given fracture stress and q is the stress-like variable describing strain
softening. The anisotropy of the model is accounted for in equation (59). Namely, the
discontinuity damage is only damaging in traction. If the normal component traction 111
on the discontinuity became negative, no evolution of localized damage is considered.
Further anisotropy is introduced in order to be able to deal with shear effects on the
discontinuity. The chosen model for the discontinuity is thus turned into a
multi-surface type of model, by introducing an added damage function concerning the
tangential component of the traction defined by:
 
 s s

f2 ðs; qÞ ¼ jt Gs · mj 2 s s 2 q # 0 ð60Þ
^ s f
f2 ðt Gs Þ

where s s is the limit value of the shear on the discontinuity. This multi-surface
treatment of the discontinuity shear is compatible with the classical shear retention
factor treatment (Armero, 1997; Rots et al., 1985). A multi-surface damage requires that
the developments presented in the previous section be slightly modified to take into
account the added damage function f 2 . This was achieved by adapting the method
developed by Govinjee et al. (1995) and Simo et al. (1988) to treat multi-surface
continuum damage model to the present case of an interface behavior at the surface of
discontinuity. To that end, the Helmoltz free energy for the discontinuity can be
written:

c ðu  jÞ ¼ 1 u
 ; Q; Q 21 u  ðjÞ
 þX ð61Þ
2
which further leads the damage dissipation:

 ¼t u d   jÞ ¼ 1 t Qt
 _
0#D Gs  2  ; Q;
cð u G Gs þ qj ð62Þ
dt 2 s
The principal of maximum dissipation has to be enforced under the two constraints:

f 1 # 0 and f 2 # 0 ð63Þ

By introducing Lagrange multipliers g 1 and g 2 , we can write the corresponding


Lagrange functional:
X
max  ) max min Lðt
D 
Gs
; q 
 Þ; Lðt G s
; q  þ
 Þ ¼ 2D g k f k ðs; qÞ ð64Þ
  
f1 ðt G ;qÞ#0 g1 $0 ðt Gs ;qÞ

s
g2 $0 k
f2 ðt G ;qÞ#0
s

where k ¼ 1; 2 in our case. Exploiting the stationarity of the Lagrange functional with
respect to t Gs and q leads to the evolution laws of internal variables as:
EC ›L _ X ›f k _ X ›f k _ X  1 ›f k ›f k
¼ 2Q·t Gs þ g ¼ 0 ) Qt Gs ¼ g ) Q ¼ gk ^ ð65Þ
26,1/2 ›t Gs k
t Gs k
›q k
^
f ›t Gs ›t Gs k

and:
 ›f k
_ X  ›fk X
›L
112 _
¼ 2j þ g ¼0 ) j ¼ g k ð66Þ

›q ›q ›q
k k

By taking into account the expressions of f 1 and f 2 in equations (59) and (60), we can
write explicit form of these evolution equations:

 ¼ g 1 1
Q 1 n^n þ g 2 m^m ð67Þ
t Gs · n jt Gs · mj
and:
_ s s
j ¼ g 1 þ g 2 ð68Þ
s f
The loading-unloading conditions can also be deduced from the minimisation problem
(64) as:

g 1 $ 0; f 1 ðt Gs ; q Þ # 0; g 1 f 1 ðt Gs ; q Þ ¼ 0
ð69Þ
g 2 $ 0; f ðt ; q Þ # 0;
2 Gs g f ðt ; q Þ ¼ 0
2 2 Gs

The consistency conditions allowing to compute the corresponding values of Lagrange


multipliers are then written as:
_ _
g 1 f 1 ðt Gs ; qÞ ¼ 0 g 2 f 2 ðt Gs ; q Þ ¼ 0 ð70Þ
In the case where the damage process is activated at the surface of discontinuity, by
assuming the most general case involving both damage surfaces we can determine the
value of the Lagrange multipliers g 1 and g 2 from:
_  
f 1 ðt Gs ; qÞ ¼ 0 ¼ ››tfG1 · t_ Gs þ ››fq1 q_ ¼ 0
s
ð71Þ
_ ›f 2 
f 2 ðt Gs ; qÞ ¼ 0 ¼ ›t Gs · t_ Gs þ ››fq2 q_ ¼ 0

By exploiting the time derivative of the state equations (38) and (39):
8
>  _  21 
 21 · Q
_ 2 Q
< t_ Gs ¼ Q 21 · u ·Q ·u
  _   ðjÞ
›2 X ð72Þ
: q_ ¼ 2KðjÞj; K ¼ ›j 2
>

we can further derive the explicit form of consistency equations for computing the
Lagrange multipliers:
2 3 2 3
›f j 4  21 _ X ›f k ›  X ›f
f
Continuum-
_  21 g
 2Q 
52K j 4 g k k 5
fj ¼
›t G s
Q u k
›t Gs ›q ›q
ð73Þ discrete damage
k k
model: Part I
where we denoted with J act ¼ {ijf i ¼ 0}; j [ J act and k [ J act the active set of damage
constraints.
This can also be written in a matrix form as: 113
" #
›f k  21 _ ›f i  21 ›f j ›f i  ›f j
½G · ½g k k[J act ¼ Q u  ; G ij ¼ ·Q þ K ð74Þ
›t G s ›t G s ›t G s ›q ›q
k[J act

where ½g k k[J act is the vector of Lagrange multipliers associated to active damage
functions f k ¼ 0 and [G ] is a matrix with components G ij ; ði; jÞ [ J act £ J act .
By solving equation (74), we can obtain the vector ½g k k[J act of Lagrange multipliers
which gives:
X
 21 ›f j _
g j ¼ ½G21
ij Q

u ð75Þ
j[J act
›t G s

By introducing this result in the expression for the traction rate in equation (72) we can
rewrite the constitutive equations as:
8
> _
Q 21 u ;g k ¼ 0 k ¼ 1; 2
>
>2
>
> 3
>
>
>
>
<6
>
6
7
h  i h i7
t_ Gs ¼ 66  21
P 21  21 j
› f  21 i 7
› 
f 7 _  ð76Þ
> 6Q 2 ½G ij Q ›t Gs ^ Q ›t Gs 7u ’ kjgk . 0
>
>
>
> 6 i [ J act 7
>
> 4 5
>
>
>
: j [ J act

We have chosen an exponential softening law, which allows to write for K  that:
" !# !
b   ›q  b 
q ¼ s f 1 2 exp 2 j ) K ¼ 2  ¼ 2bexp 2 j ð77Þ
s f ›j s f

where b is a parameter which allows to adjust the total fracture energy dissipated at
the discontinuity by driving the traction to zero.

3. Finite element implementation


In this section, we present the introduced modifications in the FE interpolation and
solution procedure pertaining to the introduction of a displacement discontinuity
surface in the element. More precisely, we first present the modifications in the global
FE strategy in constructing interpolations functions. The second part will be dedicated
to the resolution of the complementary local equilibrium equation that each element
presenting a surface of discontinuity has to satisfy.
EC 3.1 Finite element interpolations
26,1/2 Before the introduction of the discontinuity, the interpolations of displacement and
strains remain the classical ones:
NX
nodes NX
nodes

dðxÞ ¼ N a ðxÞd a ) 1ðxÞ ¼ LN a ðxÞ d a ð78Þ


|fflfflfflffl{zfflfflfflffl}
a¼1 a¼1
114 Ba

where da is the displacement of node a, Na(x) is the shape function associated to node a
and L denotes the matrix associated to the operator 7s .
With those interpolations for the displacement and strain fields, the discretisation of
the equilibrium equation leads to the following system:
Nel
Z
A ½f int;e ðtÞ 2 f ext;e ðtÞ ¼ 0; f int;e ðtÞ ¼ B T sðtÞdV e ; t [ ½0; T ð79Þ
e¼1 Ve
ext,e int,e
where, ( · ) and ( · ) refer to the external and internal forces, respectively.
In equation (79) above A denotes the finite element assembly operator (Zienkiewicz
and Taylor, 2000) and Ve a particular finite element domain (an element). The values of
s(t) are obtained by integrating the rate constitutive equations of the continuum
models.
When localisation is detected in one element, the basis of the finite element
interpolations functions of the displacement field has to be completed by a function
presenting a jump in order to take into account the displacement jump on the surface of
discontinuity. One such a function M(x) is presented here on Figure 2 for a three-node
triangular element.
With this function added to the finite element interpolation function basis, we obtain
for the interpolation of the real and virtual strain field:
;
1ðx; tÞ ¼ Bd þ G r u d1ðx; tÞ ¼ Bw þ G v b ð80Þ
where G r ðxÞ ¼ LM: w and b are, respectively, the virtual displacement and virtual
displacement jump fields. Gv is an incompatible mode function modified in order to
guarantee the verification of the patch-test.
It has to be noticed that the form of the function M implies that Gr must be
decomposed into a continuous part and a discontinuous (or singular) part and that the
same holds for Gv:

Gr ¼ G  d
rþG and Gv ¼ G  d
vþG ð81Þ
r Gs v Gs

n
Figure 2.
Shape function M added
for a CST element
M(x)
With those interpolations, the discretisation of the equilibrium equation, leads to the Continuum-
system which is typically obtained in the framework of the incompatible mode method
(Wilson and Ibrahimbegović, 1990; Ibrahimbegovic and Wilson, 1991):
discrete damage
8 Nel model: Part I
< Ae¼1 ½f int;e ðtÞ 2 f ext;e ðtÞ ¼ 0
R ð82Þ
: h e ¼ Ve GTv sdVe ¼ 0; ;e [ ½1; Nel
115
3.1.1 Incompatible mode method: solution of the equilibrium equation. The first
equation of system (82) is no more than the classical global equilibrium equation. The
second one which is written for each finite element crossed by a surface of
discontinuity is a local element-wise equilibrium equation. For the choice of finite
element interpolations in equation (81) the latter is no more than the weak form of the
traction continuity along the surface of discontinuity Gs.
The linearization of system (82) leads to the set of equilibrium equations
(considering time step n þ 1 and iteration i ):
8 n o n o
>
> Nelem eði Þ ði Þ eði Þ  ði Þ Nelem
fe;ext e;intði Þ
< Ae¼1 Knþ1 Ddnþ1 þ Fr nþ1 Du nþ1 ¼ Ae¼1 nþ1 2 fnþ1
h eði Þ i h i ð83Þ
>
> T ði Þ ði Þ eði Þ ði Þ  ðinþ1
Þ
¼ 2heði Þ
: Fv nþ1 þ Kdnþ1 Ddnþ1 þ Hnþ1 þ Kanþ1 Du nþ1

where:
Þ R ði Þ
Þ
R ði Þ
Keði
nþ1 ¼ Ve B T Ced e
nþ1 BdV ; Feði
r nþ1 ¼ Ve B T Ced 
nþ1 G r dV
e

R R
Feði Þ
B T Ced
ði Þ
 e
Heði Þ  T Ced ði Þ G
 e
v nþ1 ¼ Ve nþ1 G v dV ; nþ1 ¼ Ve G v nþ1 r dV ð84Þ
R  T › t Gs R  T › t Gs
Kðidnþ1
Þ
¼ Gs Gv ›d dGs ; Kðianþ1
Þ
¼ G s Gv › u  dGs

At that stage, there exists a couple of possibilities to solve the above set of equilibrium
equations.
The first possibility consists in solving simultaneously the two equilibrium
Þ eði Þ ði Þ ði Þ eði Þ
equations at the global level. For this, matrices Feði v nþ1 , Hnþ1 , Kdnþ1 , Kanþ1 and hnþ1
need to be stored for one iteration to the next which impose an important secondary
storage.
The second possibility which is exploited herein consists in using the fact that the
second equilibrium equation is written on each localized element independently and
use an “operator split” solution strategy (Ibrahimbegovic and Kožar, 1995). For a given
displacement increment at global iteration i: Ddðinþ1 Þ
, the local elementary equilibrium
equation is solved by an iterative strategy.
The equation to be solved at the element level is then written as:
Z Z
eði Þ
hnþ1 ¼  T ði Þ  ði Þ
Gv sðdnþ1 ; unþ1 ÞdV þe
G T t ðdði Þ ; u  ðinþ1
Þ
ÞdGs ¼ 0 ð85Þ
v Gs nþ1
Ve Gs

The resolution is carried out at the element level for a given value of the nodal
displacement dinþ1 which remains fixed during the process of solving equation (85)
EC  ði;jÞ
(Ibrahimbegovic and Kožar, 1995). Considering iteration j and assuming u nþ1 is given,
 ði;jÞ  ði;jþ1Þ  ði;jÞ
computation of Dunþ1 ¼ unþ1 2 unþ1 is made by considering the linearized form of
26,1/2 equation (85) with respect to u :
h i
eði;jÞ
hnþ1 þ Hnþ1eði;jÞ
þ Kði;jÞ  ði;jÞ
anþ1 Dunþ1 ¼ 0 ð86Þ

116 We finally solve for the increment of displacement jump as:


h i21
 ði;jÞ
Du eði;jÞ ði;jÞ
nþ1 ¼ 2 Hnþ1 þ Kanþ1
eði;jÞ
hnþ1 ð87Þ

 ði;jÞ
The value of u nþ1 is then incremented according to:

ði;jþ1Þ
 nþ1  ði;jÞ  ði;jÞ
u ¼u nþ1 þ Dunþ1 ð88Þ

and the procedure is repeated at the element level until the local equilibrium equation is
satisfied within a prescribed tolerance.
At the end of this iterative procedure, the vector of displacement jump obtained
satisfies:
 
Þ ði Þ  ði Þ
heði
nþ1 dnþ1 ; unþ1 ¼ 0 ð89Þ

The linearized global system to be solved is then written:


8 n o n o
>
> A Nelem
K eði Þ
Ddði Þ
þ F eði Þ  ði Þ
D u ¼ A Nelem
fe;ext
2 f e;intði Þ
< e¼1 nþ1 nþ1 r nþ1 nþ1 e¼1 nþ1 nþ1
h eði Þ i h i ð90Þ
>
> T ði Þ ði Þ eði Þ ði Þ  ðinþ1
Þ
: Fv nþ1 þ Kdnþ1 Ddnþ1 þ Hnþ1 þ Kanþ1 Du ¼0

Or, in a matrix form:


2 32 3 "
Keði Þ
Feði Þ #
nþ1 r nþ1 Ddðinþ1
Þ
fe;ext 2 fe;intði Þ
6 74 5¼ nþ1 nþ1
4 T eði Þ 5 ð91Þ
Fv nþ1 þ Kðidnþ1
Þ
Heði Þ ði Þ  ðinþ1
Þ
nþ1 þ Kanþ1 Du 0

The second equation is condensed at the element level so that the resolution of the
previous system is reduced to the classical form:
Nelem n o
A ^ eði Þ Ddði Þ ¼ fe;ext 2 fe;intði Þ
K ð92Þ
nþ1 nþ1 nþ1 nþ1
e¼1

with:
h i21 h eði Þ i
^ eði Þ ¼ Keði Þ 2 Feði Þ Heði Þ þ Kði Þ
K F T
þ K ði Þ
ð93Þ
nþ1 nþ1 r nþ1 nþ1 anþ1 v nþ1 dnþ1

taking into account the behavior of the surface of discontinuity.


4. Numerical examples Continuum-
In this section, we present the results of several numerical simulations, which provide an discrete damage
illustration of the particular features of the proposed anisotropic damage model. All the
examples presented in this section are performed considering plane strain hypothesis. model: Part I
The finite element model which is employed in all the examples we considered concerns
a three-node triangular element with constant strain and stress element fields, so-called
CST element in the terminology of Zienkiewicz and Taylor (2000), with the strain field 117
modified accordingly by the incompatible mode which is activated upon the
discontinuity (or “crack”) initiation. The element computations are performed by one
point quadrature to account for the continuum damage dissipative phenomena spread
throughout the element and by one point quadrature accounting for displacement
discontinuity contribution. All the computations are carried out by a research version of
the computer program FEAP, developed by Professor R.L. Taylor at the University of
California at Berkeley. Meshes are designed with the 2D mesh generator developed by
Alain Rassineux at the University of Technology of Compiègne.

4.1 Simple tension test: mesh-independence of solution


In this example, we seek to clearly illustrate the ability of the proposed model to
furnish a mesh-independent solution for all phases of the computed response. To that
end, we consider the simple tension test on a rectangular strip of length equal to
200 mm, width equal to 100 mm and a unit thickness. Owing to symmetry, only a half
of the specimen is considered in computations, with the corresponding choice of the
symmetry enforcing boundary conditions (Figure 3). The chosen mechanical properties
are given in Table I.
The computations are performed by incrementing the imposed displacement at the
free-end of the specimen in order to simulate a displacement-controlled tension test.
In the first phase of the response concerning the elastic regime, one obtains the
homogeneous stress state which can be exactly represented by either one of the chosen

U U

(a) Coarse unstructured mesh (b) Fine structured mesh

Figure 3.
Finite element model and
boundary conditions
(c) Fine unstructured mesh
EC finite element meshes. In the subsequent phase with damage, homogeneous stress state
26,1/2 would typically lead to a bifurcation problem which implies that any of particular
element can be the first to reach the ultimate stress limit and enter the subsequent
softening phase. The latter would in general be decided by the round-off errors in
numerical computations. Therefore, in order to avoid dealing with such a bifurcation
problem, we introduce a perturbation by slightly reducing the value of the ultimate
118 stress for a single element in the mesh which ensures that the softening phase should
first start in that element. This kind of perturbation is used for each finite element
models employed in computations (see Figure 3, where the weakened element is
indicated by the shaded area).
The computed responses by the three finite element models are shown in Figure 4 in
terms of the force-displacement diagrams. The chosen loading program considers both
load increase and unloading phase. The unloading is first attempted in the damage
hardening phase, where only the continuum damage model is active, and then in the
later stage where damage softening model dictates the computed result. The difference
in computed response by the three different meshes is a way too small to display in
the given scale concerning the force-displacement diagram (Figure 4).

Continuum model
Young modulus 38 GPa
Poisson ratio 0.18
sf 2 MPa
K 1,000 MPa
Localised model
s f 2.55 MPa
Table I. 2.5 MPa (weakened element)
Mechanical properties of s s =s f 0.3
the specimen in simple b f 2.55 MPa/mm
tension test Fracture energy Gf 1.275 N/mm

300

A
250

200 Coarse mesh


Fine unstructured mesh
F (N/mm)

Fine structured mesh


150

100

50
Figure 4.
Load/displacement curve
for structured and 0
0 20 40 60 80 100 120
unstructured meshes
U ×100 (mm)
The only difference concerns the computed positions of discontinuity lines for one Continuum-
finite element mesh and another (Figure 5). For the coarse mesh, the discontinuity lines discrete damage
approximately add to 157 mm, whereas for the fine mesh the total length of
discontinuity is 133 mm. This kind of difference is the consequence of the adopted model: Part I
strategy to allow for displacement discontinuity creation only in the element centers. It
is very important to note, however, that the different discontinuity lines do not imply
any notable difference in the computed response. The latter is the consequence of the 119
stress orthogonality condition in equation (85), which is capable of establishing the
same kind of correspondence between the two types of dissipative mechanisms and
thus guarantee the invariance of the computed response regardless to the length of
discontinuity lines. This is very important finding which further indicates that,
contrary to the case of simplified models combining only elastic response with
post-peak softening, there is no need to construct a continuous representation of the
displacement discontinuity line between the neighboring elements. In other words, the
displacement discontinuity line can be interpreted more like a dominant crack direction
at the given loading level, rather than the continuous crack appearing as the only
source of dissipation in the fracture mechanics kind of models.

4.2 Simple tension test: model performance in cyclic loading


In this example, we consider the same specimen as in the previous example, but now
submitted to a cyclic loading. The first goal of this example is to show that the
proposed anisotropic damage model can capture a number of salient features of cracks
closing and re-opening during a loading cycle.
The computations are performed by using the fine structured mesh with two
elements weakened (Figure 3) in order to better control the response. The response is
driven by the free-end displacement incremental sequence. The displacement is first
increased to the level of activating the continuum damage phase, followed by a short
unloading and reloading process bringing the stress state to the same point where the
unloading has started (Figure 6). The loading is then continued further passing the

U U

(a) Unstructured mesh (b) Fine mesh

Figure 5.
Discontinuity lines at the
end of the test
(c) Fine unstructured mesh
EC 300
26,1/2
200

100

120
F (N/mm)
0

– 100

– 200

– 300
Figure 6.
Load/displacement curve – 400
for structured mesh – 0.6 – 0.4 – 0.2 0 0.2 0.4 0.6 0.8 1 1.2
U (mm)

ultimate load level and entering the softening response phase. The latter is, as shown
on Figure 7 characterized by the progressive opening of the discontinuities as the
prescribed displacement increases.
The unloading is carried out again on the softening phase, but this time all the way
to reversing the sign of stress. The complete unloading of stress to zero is accompanied
by the closing of the previously opened discontinuities. This closing is characterized by
evolution neither of the continuum damage nor of the discrete damage which remain
equal to their maximum value (Figure 8). When reversing the sign of stress and
beginning loading in compression, we note an increase of the apparent stiffness of the
structure. This is due to the unilateral behavior of the discontinuities leading to
stiffness recovery: the first apparent compliance in compression corresponds to its
maximum value reached at the ultimate stress point. If the loading in compression is
carried on till the damage yield is reached, the continuum compliance increases again.

1.4 300

1.2 200

1 100 F

u=n
F (N/mm)
un(mm)

0.8 0

0.6 – 100
=

0.4 – 200

Figure 7. 0.2 – 300


Evolution of displacement
jump and measured load 0 – 400
0 0.5 1 1.5 2 2.5 3 3.5 4
according to pseudo-time
t (pseudo-time)
50 300 Continuum-
m– and 50 × Qnn (MPa–1.m) 45
200 discrete damage
40
F
model: Part I
35 100
=

F (N/mm)
30 0 50 × Qnn
25 121
=


– 100 µ
20
15 – 200
10 Figure 8.
– 300 Evolution of damage
5 variables and measured
0 – 400 load according to
0 0.5 1 1.5 2 2.5 3 3.5 4
pseudo-time
t (pseudo-time)

Unloading subsequently and going back to traction loading, we can reach the same
level of stress before the softening in tension starts again. However, due to change in
continuum damage compliance, we do not recover the same strain (Figures 6 and 8).
Yet another presentation of the complex phenomena taking place in this example can
be given by focussing on the evolution of the displacement discontinuity during this
loading program (Figure 7). During the initial loading phase, the displacement
discontinuity gets activated only upon the stress reaching the ultimate value and then
keeps evolving by further increase of the imposed displacement. Once the loading
(imposed displacement) is reversed, we start by closing the “crack” and the
displacement discontinuity is reduced to zero (the crack is fully closed) when complete
unloading is performed.
The further evolution of the displacement discontinuity will occur only when the
crack opening reaches the value at which the unloading had started.
Therefore, this example shows that a very complex phenomena of crack closing and
reopening can be fully captured by the present model, even if the latter has only a few
parameters. This should be compared against a considerable complexity of the
previous damage models for cyclic loading (La Borderie, 1991), which typically have a
very large number of parameters.

4.3 Non-homogeneous stress field problem: three-point bending test


In this example, we seek to illustrate the capabilities of the proposed model for
representing the failure under a non homogeneous stress field. More precisely, we
consider a three-point bending test on a notched specimen. The presence of the notch
introduces a considerable heterogeneity of the stress field, and represents the kind of
tests which remains out of reach for traditional models of failure based upon non-local
formulation (Pijaudier-Cabot and Bažant, 1987). Dimensions of the specimen along
with boundary conditions are shown on Figure 9. The set of material parameters is
given in Table II. Two different unstructured meshes have been considered (Figure 10)
for the computation which is carried out under displacement control.
In Figure 11, we present the results obtained with the two different FE meshes for
the load/displacement diagram, where vertical load is measured at the points where
vertical displacements are imposed. We can note that the results given by the two
EC different meshes are quite similar both for the limit load as well as the total dissipated
26,1/2 energy predicted by each computation.
Figure 12 shows the evolution of applied load in terms of crack mouse opening
displacement (CMOD). We find again that the results obtained with computation

u
122
Figure 9. 500
Notched specimen:
geometric characteristics

200
(in mm) and boundary
conditions 890 20 890

Continuous model
Young modulus 38 GPa
Poisson’s coefficient 0.1
sf 2.2 MPa
K 1,000 MPa
Discrete model
Table II. s f 2.35 MPa
Mechanical properties of b f 23.5 MPa/mm
the anisotropic damage Fracture energy Gf 0.1175 N/mm
model s s 0.235 MPa

500

400

300

200

100

0
0 200 400 600 800 1,000 1,200 1,400 1,600 1,800
(a) Coarse mesh

500

400

300

200

100
Figure 10.
Two different meshes 0
0 200 400 600 800 1,000 1,200 1,400 1,600 1,800
used for the computation
(b) Fine mesh
140
Coarse mesh
Continuum-
Fine mesh discrete damage
120
model: Part I
100
Load (N/mm)

80 123
60

40

20
Figure 11.
Load displacement
0 response for the three
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
point bending test
Prescribed displacement (mm)

140
Coarse mesh
Fine mesh
120

100
Load (N/mm)

80

60

40

20
Figure 12.
Measured load in terms of
0 crack mouse opening
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
displacement
CMOD (mm)

employing two different FE meshes are finally quite similar. In short, the results in
Figures 11 and 12 show displacement evolution both for absolute vertical and relative
horizontal component as a function of applied loading.
Similar tendencies are noticed for discrete model internal variable which can be
interpreted as the prolongation of CMOD. Namely, Figure 13 shows the color map
representation of crack opening (u  n ) and crack sliding (u
 m ) for these two different
meshes at the final deformed configuration (scaled by 500). We can observe that the
local results in terms of crack opening and crack sliding remain quite similar for the
two different discretisations.
EC STRESS 5 STRESS 6
26,1/2 –1.40E-02
2.46E-05
1.41E-02
–3.24E-02
–2.83E-02
–2.42E-02
2.81E-02 –2.00E-02
4.21E-02 –1.59E-02
5.61E-02 –1.18E-02
7.02E-02 –7.66E-03
8.42E-02 –3.54E-03
9.82E-02 5.87E-04
1.12E-01 4.71E-03
1.26E-01 8.83E-03
124 1.40E-01
1.54E-01
1.30E-02
1.71E-02

Time = 3.10E+01 Time = 3.10E+01


=
(a) Coarse mesh: u=n (b) Coarse mesh: um

STRESS 5 STRESS 6
–1.35E-02 –2.80E-02
2.36E-05 –2.42E-02
1.35E-02 –2.04E-02
2.70E-02 –1.65E-02
4.05E-02 –1.27E-02
5.40E-02 –8.89E-03
6.76E-02 –5.07E-03
8.11E-02 –1.25E-03
9.46E-02 2.57E-03
Figure 13. 1.08E-01
1.22E-01
6.39E-03
1.02E-02
1.35E-01 1.40E-02
Crack opening and sliding 1.49E-01 1.79E-02
at the end of the
Time = 3.10E+01 Time = 3.10E+01
computation for the two
different discretisations =
(c) Fine mesh: u=n (d) Fine mesh: um

4.4 Non-proportional loading and mixed mode failure


In this example, we test the Nooru-Mohamed (1992) specimen under non-proportional
loading. The geometry, boundary conditions and mesh discretisation are shown on
Figure 14. The material parameters used for the computation are the same as in the
previous test (Table II).

Un

Fs
200 mm

Figure 14.
Double notched specimen:
problem description and 200 mm
mesh discretisation
25mm
The test is carried out by first imposing increasing horizontal load Fs to its maximum Continuum-
value 5 kN, in the second part of the loading path, this value is kept constant while the discrete damage
vertical displacement Un is increased.
Figure 15(a) shows the crack path obtained at a normal displacement U n ¼ 0:024 model: Part I
mm. Figure 15(b) shows the deformed mesh scaled by 1,000 at the same point. We can
note that the crack path is close to the one observed experimentally. This simulation
shows again the efficiency of the proposed model capable, with little parameters, to 125
describe mixed mode failure under non proportional loading.

5. Conclusions
In this work, we have presented a new way for constructing an anisotropic damage
model, which is not carried out locally at a stress point but rather at the level of a single
element. The latter is accomplished by a judicious combination of the isotropic
continuum damage model with the discrete damage model introduced directionally
along the displacement discontinuity line. Additional anisotropy pertains to discrete
damage mechanisms described by two-surface damage criterion, coupling damage in
mode I and in mode II. As the result we obtained a novel continuum-discrete
anisotropic damage, which is very representative of failure mechanisms of massive
structures characterized by a significant contribution not only from “macro” cracks but
also from fracture process zone.
The presented model possesses a number of desirable features. First of all, it
combines successfully the volume dissipation in the element with the surface
dissipation along the discontinuity line. The introduced stress orthogonality condition
allows to couple two kinds of dissipations and to provide that the final computed
response remains independent on the chosen mesh or the displacement discontinuity
line. For the same reason, one need not insist that the displacement discontinuity line
should pass continuously from one element onto its neighbor. In fact, in the present
model, the displacement discontinuity can be interpreted as the direction of the
dominant crack, which is formed subsequently to the creation of a number of

Crack path

200

150

100

50

Figure 15.
Crack path and deformed
0
– 20 0 20 40 60 80 100 120 140 160 180 200 mesh ( £ 1,000) at U n ¼
0:024 mm
(a) Crack path (b) Deformed mesh
EC micro-cracks and the corresponding stress phase distribution phase. With such an
26,1/2 approach the final crack direction is predicted quite well, in fact much better than the
prediction made by the fracture mechanics type of models considering combination of
only elastic response and softening.
Finally, it is also shown that the presented model has a big potential in describing
complex damage phenomena in a cyclic and/or non-proportional loading program,
126 such as crack closing and re-opening, cohesive resistance deterioration due to
tangential sliding, by using only a few parameters compared to the traditional models
for cyclic loading.

References
Armero, F. (1997), “Localized anisotropic damage of brittle materials”, in Oñate, E., Owen, D.R.J.
and Hinton, E. (Eds), Computational Plasticity. Fundamentals and Applications, CIMNE,
Barcelona, pp. 635-40.
Armero, F. and Garikipati, K. (1995), “Recent advances in the analysis and numerical simulation
of strain localization in inelastic solids”, in Owen, E., Oñate, D.R.J. and Hinton, E. (Eds),
Proceedings of Computational Plasticity IV, CIMNE, Barcelona, pp. 547-61.
Bažant, Z.P., Belytschko, T. and Chang, T.P. (1984), “Continuum theory for strain softening”,
Journal of Engineering Mechanics, Vol. 110 No. 12, pp. 1666-91.
de Borst, R., Muhlhaus, H.B., Pamin, J. and Sluys, L.J. (1992), “Computational modelling
of localisation of deformation”, Proceedings of Computational Plasticity III, pp. 483-502.
Edwards, H.L. and Wanhill, R.J.H. (1986), Fracture Mechanics, Edward Arnold, London.
Govinjee, S., Kay, G.J. and Simo, J.C. (1995), “Anisotropic modelling and numerical simulation
of brittle damage in concrete”, International Journal for Numerical Methods in
Engineering, Vol. 38, pp. 3611-33.
Hillerborg, A., Modeer, M. and Petersson, P.E. (1976), “Analysis of crack formation and crack
growth in concrete by means of fracture mechanics and finite elements”, Cement and
Concrete Research, Vol. 6, pp. 773-82.
Ibrahimbegovic, A. and Kožar, I. (1995), “Non-linear Wilson’s brick element for finite elastic
deformations of three-dimensional solids”, Communications in Numerical Methods in
Engineering, Vol. 11, pp. 655-64.
Ibrahimbegovic, A. and Wilson, E.L. (1991), “A modified method of incompatible modes”,
Communications in Applied Numerical Methods, Vol. 7, pp. 187-94.
La Borderie, C. (1991), “Phénomènes unilatéraux dans un matériau endommageable:
modélisation et application à l’analyse de structures en béton”, PhD thesis, Université
Paris VI, Paris.
Lemaitre, J. (1992), A Course on Damage Mechanics, Springer, New York, NY.
Muhlhaus, H.B. and Aı̈fantis, E.C. (1991), “A variational principle for gradient plasticity”,
International Journal of Solids and Structures, Vol. 28, pp. 845-57.
Needleman, A. (1988), “Material rate dependence and mesh sensitivity in localization problems”,
Computer Methods in Applied Mechanics and Engineering, Vol. 63, pp. 69-85.
Nooru-Mohamed, M.B. (1992), “Mixed mode fracture of concrete: an experimental approach”,
PhD dissertation, Delft University of Technology, Delft.
Oliver, J. (2000), “On the discrete constitutive models induced by strong discontinuity kinematics
and continuum constitutive equations”, International Journal of Solids and Structures,
Vol. 37, pp. 7207-29.
Ortiz, M., Leroy, Y. and Needleman, A. (1987), “A finite element method for localized failure Continuum-
analysis”, Computer Methods in Applied Mechanics and Engineering, Vol. 61, pp. 189-214.
Pijaudier-Cabot, G. and Bažant, Z.P. (1987), “Nonlocal damage theory”, ASCE Journal of
discrete damage
Engineering Mechanics, Vol. 113, pp. 1512-33. model: Part I
Rots, J.G., Nauta, P., Kusters, G. and Blaauwendraa, T. (1985), “Smeared crack approach and
fracture localization in concrete”, Heron, Vol. 30 No. 1, pp. 1-48.
Simo, J.C., Kennedy, J.G. and Govindjee, S. (1988), “Non-smooth multisurface plasticity and 127
viscoplasticity. Loading/unloading conditions and numerical algorithms”, International
Journal for Numerical Methods in Engineering, Vol. 26, pp. 2161-85.
Simo, J.C., Oliver, J. and Armero, F. (1993), “An analysis of strong discontinuity induced by strain
softening solutions in rate-independent solids”, Journal of Computational Mechanics,
Vol. 12, pp. 277-96.
Wells, G.N. and Sluys, L.J. (2001), “Analysis of slip planes in three-dimensional solids”, Computer
Methods in Applied Mechanics and Engineering, Vol. 190, pp. 3591-606.
Wilson, E.L. and Ibrahimbegović, A. (1990), “Use of incompatible displacement modes for the
calculation of element stiffnesses or stresses”, Finite Elements in Analysis and Design,
Vol. 7, pp. 229-41.
Zienkiewicz, O.C. and Taylor, R.L. (2000), The Finite Element Method, 5th ed.,
Butterworth-Heinemann, Oxford.

Further reading
Ibrahimbegovic, A. (2006), Mécanique non linéaire des solides déformables: formulation théorique
et résolution numérique par éléments finis, Lavoisier, Paris.

Corresponding author
A. Ibrahimbegovic can be contacted at: ai@lmt.ens-cachan.fr

To purchase reprints of this article please e-mail: reprints@emeraldinsight.com


Or visit our web site for further details: www.emeraldinsight.com/reprints

You might also like