Mechanics of Materials 80 (2015) 351–367

Contents lists available at ScienceDirect

Mechanics of Materials
journal homepage: www.elsevier.com/locate/mechmat

Gradient damage models coupled with plasticity: Variational

formulation and main properties
Roberto Alessi a,c, Jean-Jacques Marigo c,⇑, Stefano Vidoli a,b
a
Dipartimento di Ingegneria Strutturale e Geotecnica, Sapienza Università di Roma, Via Eudossiana 18, 00184 Roma, Italy
b
Institut Jean Le Rond D’Alembert, UPMC, Paris, France
c
Laboratoire de Mécanique des Solides, Ecole Polytechnique, 91128 Palaiseau Cedex, France

a r t i c l e i n f o a b s t r a c t

Article history: A variational formulation is proposed for a family of elastic–plastic–damage models within
Received 4 November 2013 the framework of rate-independent materials. That consists first in defining the total
Received in revised form 10 December 2013 energy which contains, in particular, a gradient damage term and a term which represents
Available online 14 January 2014
the plastic dissipation but depends also on damage. Then, the evolution law is deduced
from the principles of irreversibility, stability and energy balance. Accordingly, the plastic
Keywords: dissipation term which appears both in the damage criterion and the plastic yield criterion
Gradient damage model
plays an essential role in the damage–plasticity coupling. Suitable constitutive choices on
Cohesive crack
Plasticity
how the plastic yield stress decreases with damage, allows us to obtain a rich variety of
Variational approach coupled responses. A particular attention is paid on the equations which govern the forma-
Ductile fracture tion of cohesive cracks where the displacement is discontinuous and the plasticity
localizes. In the one-dimensional traction test where the solution is obtained in a closed
form, we show that, because of damage localization, a cohesive crack really appears at
the center of the damage zone before the rupture and the associated cohesive law is
obtained in closed form in terms of the constitutive parameters. A Finite Element discrete
version of the energy functional is used to simulate a two-dimensional traction test over a
rectangular domain with mixed boundary conditions; again a localized band of plastic
strain is generated seemingly independent of the mesh size.
Ó 2014 Elsevier Ltd. All rights reserved.

1. Introduction decreasing dependency of the stiffness EðaÞ on the damage

Gradient damage models are very efficient to account
for the behavior of brittle and quasi-brittle materials. Such
models have been developed independently by different
group of authors (see for instance Peerlings et al., 1998;
Comi, 1999; Bourdin et al., 2000; Comi et al., 2006; Benallal
and Marigo, 2007; Pham and Marigo, 2010a,b; Lorentz
et al., 2011). Their main merit is that they are able to ac-
count for both the nucleation and the propagation of cracks
in a unified framework. Their basic ingredients are: (i) a
⇑ Corresponding author. Tel.: +33 169335797.
E-mail addresses: roberto.alessi@uniroma1.it (R. Alessi), marigo@
lms.polytechnique.fr (J.-J. Marigo), stefano.vidoli@uniroma1.it (S. Vidoli).
variable a; (ii) no more rigidity at the ultimate damage
state (say Eð1Þ ¼ 0); (iii) a critical stress rc ; (iv) a softening
behavior with a decrease of the stress from rc to 0 when
the damage goes to 1; (v) a gradient damage term in the
energy which necessarily contains an internal length ‘
and which limits the damage localization.
Accordingly, it can be shown in a one-dimensional set-
ting that the process of crack nucleation consists in the fol-
lowing three successive stages (see Pham et al., 2011a;
Pham and Marigo, 2013a; Sicsic et al., 2014): (i) first, dam-
age occurs in a neighborhood of a point where the stress
reaches the material critical stress; (ii) then, damage grows
inside a damage zone the width of which is related to the
material characteristic length ‘; (iii) finally, damage

http://dx.doi.org/10.1016/j.mechmat.2013.12.005
0167-6636/Ó 2014 Elsevier Ltd. All rights reserved.
352 R. Alessi et al. / Mechanics of Materials 80 (2015) 351–367
Fig. 1. In one-dimension, nucleation of a crack at the center x1 of a
damage zone of width p‘=2; ‘ being the material characteristic length. The
graphs represent the growth of the damage field with time. At final time
aðx1 Þ ¼ 1, a crack has appeared at x1 .
reaches the critical value 1 at the center of the damage zone
and a crack appears at this point, see Fig. 1. During this
crack nucleation process, some energy is dissipated inside
the damage zone and this dissipated energy involves a
quantity Gc which can be considered as the effective surface
energy of Griffith’s theory. Therefore, Gc becomes a byprod-
uct of the gradient damage model which can be expressed
in terms of the parameters of the model (specifically, Gc is
proportional to r2c ‘=Eð0Þ (Pham and Marigo, 2013a)).
In two or three dimensions, the process of crack nucle-
ation essentially follows the same stages. For example, in
Fig. 2 are illustrated in a two-dimensional setting these dif-
ferent phases of nucleation and propagation of an array of
cracks which are obtained by this type of gradient damage
model in the case of a thermal shock (see Sicsic et al., 2014)
for details. Moreover, once a crack has nucleated, it can be
proved (Sicsic and Marigo, 2013) that its propagation is
essentially governed by Griffith’s law, i.e., the law based
on the concept of critical energy release rate criterion
where the role of the critical energy release rate is played
by the dissipated energy density Gc .
However, this type of ‘‘quasi-brittle’’ models are not
able to account for residual strains and consequently can-
not be used in the case where the fracture occurs with a
significant plastic zone. Moreover there is no discontinuity
of the displacement in the damage strip before the loss of
rigidity at its center, i.e., before the nucleation of a crack.
In other words such a model cannot account for the
nucleation of cohesive cracks, i.e., the existence of surface
Fig. 2. Numerical simulation by a gradient damage model of the damage evolution during the thermal shrinking induced by cooling through the top surface
of the sample (Sicsic et al., 2014). In blue, the sound material; in red, the totally damaged material. (For interpretation of the references to colour in this
figure legend, the reader is referred to the web version of this article.)
of discontinuity of the displacement with a non vanishing
stress. The natural way to include such effects is to intro-
duce plastic strains into the model and to couple their evo-
lution with damage evolution. Of course, this idea is not
new and a great number of damage models coupled with
plasticity have been developed from the eighties in the
spirit of Lemaitre and Chaboche (1985) (see for instance
de Borst et al., 1999; Grassl and Jirásek, 2006; Belnoue
et al., 2007; Dimitrijevic and Hackl, 2011). But our purpose
is to construct such models in a softening framework with
gradient of damage terms and to see how these models can
account for the nucleation of cracks in presence of plastic-
ity. In our knowledge, the previous works are not able to go
so far. Such regularization methods by adding gradient
terms also exist for other types of behavior exhibiting
strain-softening. It is not possible to quote here all the
abundant literature devoted to this topic and it is not the
goal of the present paper to compare all or a part of these
approaches with ours. The reader interested by such a
comparison should refer to Lorentz and Andrieux (2003)
where a general overview was proposed, or to the more
recent work (Dal Corso and Willis, 2011) where a regular-
ization by gradient plastic strain is proposed for elastic–
plastic (but non damaging) materials. Let us simply remark
that none of the quoted works are able to treat in a unified
manner the delicate issue of nucleation of cohesive and
non cohesive cracks.
Here we will adopt a variational approach in the spirit
of previous works (Mielke, 2005, 2006; Bourdin et al.,
2008; Del Piero and Truskinovsky, 2009; Pham and Marigo,
2010a,b; Pham et al., 2011a; Sicsic and Marigo, 2013; Del
Piero, 2013). Such an approach is fundamental both from
the theoretical and numerical viewpoints. From the theo-
retical viewpoint, that allows us to construct the model
in a rational and systematic way. Indeed, the main steps
are the following ones: (i) one defines the total energy of
the body in terms of the state fields which include the dis-
placement field and the internal variable fields, namely the
damage, the plastic strain and the cumulated plastic strain
fields; (ii) one postulates that the evolution of the internal
variables is governed by the three general principles of
irreversibility, stability and energy balance. Besides, we have
the chance that the variational approach works and has
been already developed both in plasticity and in damage
mechanics, even though only separately up to now. So, it
‘‘suffices’’ to introduce the coupling by choosing the form
of the total energy to obtain, by virtue of our plug and play
 R. Alessi et al. / Mechanics of Materials 80 (2015) 351–367
device, a model of gradient damage coupled with plastic-
ity: the damage and plasticity criteria are deduced from
the stability principle; the normality flow rule for the plas-
tic strain has not to be postulated but is a consequence of
the stability principle and the energy balance; we derive
new jump conditions and another normality flow rule on
a surface of discontinuity (shear band or cohesive crack).
In this paper, our choice of coupling is minimalist in the
sense that it simply consists in introducing the dependence
of the yield plastic stress rP ðaÞ on the damage variable
(with the natural assumption that rP ðaÞ goes to 0 when
the damages goes to 1). In turn, by virtue of the variational
character of the model, the product r0P ðaÞp  of the deriva-
tive of the state function rP ðaÞ by the cumulated plastic
strain p enters in the damage criterion and this coupling
plays a fundamental role in the nucleation of a cohesive
crack.
From the numerical viewpoint, the variational ap-
proach allows us to use an alternate minimization algo-
rithm to solve the incremental evolution problem. This
type of algorithm is used for a long time in plasticity
and more recently in damage mechanics (Bourdin et al.,
2000). In the present context, it consists in minimizing
at a given time step the total energy of the body alterna-
tively with respect to one of the three state fields (dis-
placement, plastic strain or damage field), the other two
being fixed, and to iterate until convergence. It turns out
that such an algorithm is automatically an algorithm of
descent of the energy, what is very interesting to obtain
a final state which satisfies the stability condition. It is this
algorithm which allowed us to obtain the numerical re-
sults for the thermal shock problem plotted in Fig. 2 and
it will be used here in Section 5 to solve numerically the
uniaxial traction problem.
Specifically, the paper is organized as follows. In Sec-
tion 2 the energy and dissipation functionals of the model
are presented; the main hypotheses on their constitutive
functions are discussed and the explicit form of the balance
and consistency equations, as dictated by the stability con-
dition and by the energy balance, are explicitly stated. To
model plastic effects we choose the simple framework of
Von Mises plasticity criterion (more complex choices in
this respect could be also considered). Once stated the gov-
erning equations, the homogeneous responses are studied
in Section 3, by requiring all the fields to do not depend
on the space variable. Three different responses are
enlightened which corresponds to all basic kinds of cou-
pling between the evolution of damage and plasticity. This
classification guides also the study of non-homogeneous
solutions in Sections 4 and 5. Indeed it turns out that the
homogeneous responses are not stable if the size of the do-
main is larger than a threshold; hence we are led to study
the conditions under which localization of both the dam-
age and the plastic strain fields can occur. The interplay be-
tween damage and accumulated plastic strain is, for the
case of non-homogeneous solutions, much richer: in some
cases the model dictates the formation of a plastic hinge
where the accumulated plastic strain must localize and
the damage profile derivative, as well as the displacement
field, suffers a jump. Section 4 is devoted to the standard
one-dimensional traction test; in this case we are able to
353
compute in closed form the nucleation of cohesive cracks
and the associated, Barenblatt-type, cohesive law between
the stress and the displacement jump. The two-dimen-
sional analysis described in Section 5 requires instead a
numerical implementation of the presented model. To this
aim a Finite Element discrete version of the variational
problem is implemented within the Fenics framework
(Logg et al., 2012). Whilst for the one-dimensional traction
test the code allows to recover the discussed analytic solu-
tion, a traction test problem over a rectangular material
domain in plane-strain enlighten the formation of a 45°
shear band and the necking of the specimen. Despite the
localization phenomena occurring, these results are shown
to be mesh-independent.
Notation. Throughout the paper, the following notations
are used. N denotes the dimension of the space,
1 6 N 6 3; N ¼ 1 corresponds to the one-dimensional case
and N > 1 to the multi-dimensional case. In the multi-
dimensional case, vectors and second order tensors are
denoted by boldface letters, e.g., u for the displacement
and r for the stress. Their components are denoted by
the respective italic letters with lower indices like ui or
rij ; 1 6 i; j 6 N. The second order identity tensor is denoted
I. Fourth order tensors are denoted with typewriter letters,
like E and C for the stiffness and compliance tensors,
respectively. The summation convention is implicitly used.
The inner product between vectors or between second
order tensors is indicated by a dot. Accordingly, one reads
u  v ¼ ui v i ; e  e ¼ eij eij ; E e  e ¼ Eijkl eij ekl . The euclidean
norm of a vector or a second order tensor is denoted k  k
pffiffiffiffiffiffiffiffiffiffi
and thus kuk ¼ u  u. The trace operator for a second
order tensor is denoted Tr and thus Tr p ¼ p  I ¼ pii . The
deviatoric operator is represented by the superscript D
and thus rD denotes the deviatoric part of r, i.e.,
Tr r
rD ¼ r  I:
N
The dependence on the time parameter t is indicated by a
subscript whereas the dependence on the spatial coordi-
nate x is indicated classically by parentheses, e.g.,
x # ut ðxÞ stands for the displacement field at time t. In
general, the state functions or the material parameters
are represented by sans serif letters, like Y0 for the Young
modulus, d1 for the local dissipated energy by damage or
wðaÞ for the damage change of variable. The prime denotes
either the derivative with respect to the spatial coordinate
x in a one-dimensional setting or the derivative with re-
spect to the damage parameter, the dot stands for the time
derivative, e.g.,
dut dE
u0t ðxÞ ¼ ðxÞ; E0 ðaÞ ¼ ðaÞ;
dx da
1
u_ t ðxÞ ¼ lim ðutþh ðxÞ  ut ðxÞÞ:
h!0 h
The symbols r and rs will be respectively used to indicate
the spatial gradient and its symmetric part when this last
makes sense. Accordingly, the total strain tensor field reads
as e ¼ rs u. Table 1 summarizes the main nomenclature
used in this article.
354 R. Alessi et al. / Mechanics of Materials 80 (2015) 351–367

Table 1 denotes the gradient of damage and the gradient damage

Main nomenclature. term, which necessarily introduces an internal length
State variables and state functions ‘ > 0, is used to limit the damage localization. Note that
a Scalar damage variable
WD represents the total energy density, sum of the stored
x ¼ wðaÞ Change of damage variable used in the particular (and recoverable) elastic energy and the energy dissipated
family of damage models during the damage process. Such models are able to
ra Gradient of damage account for the nucleation of cracks in brittle and ‘‘quasi-
e Total strain tensor
brittle’’ materials. However, they are not able to account
p Plastic strain tensor

p Cumulated plastic strain for residual strains and consequently cannot be used in
EðaÞ State function giving the stiffness tensor ductile fracture. The natural way to include such effects
CðaÞ State function giving the compliance tensor is to introduce plastic strains and to couple their evolution
dðaÞ State function giving the energy density
with damage evolution.
dissipated by damage
rD ðaÞ State function giving the damage critical stress
Before to introduce the coupling between damage and
rP ðaÞ State function giving the plastic yield stress plasticity, it is important to recall that the classical plastic-
Wðe; a; ra; p; p
Þ State function giving the strain work density ity theory can also be formulated in a variational form (at
Material constants least when the plastic flow rule follows the normality rule).
Y0 ; m 0 Young modulus and Poisson ratio of the For instance, in the case of Von Mises yield criterion, the
undamaged material total energy density reads as
d1 Energy density dissipated by a fully damaged
material point 1
r D ¼ rD ð0Þ Initial damage critical stress WP ðe; p; p
Þ ¼ E ðe  pÞ  ðe  pÞ þ rP p
; ð2Þ
2
r P ¼ rP ð0Þ Initial plastic yield stress
‘ Material characteristic length where p 2 MNs denotes the plastic strain tensor and p  the
Gc Effective surface energy density cumulated plastic strain. Their definition depends on the
Space and time variables or fields dimension of the space.
N Dimension of the space In the one-dimensional case, the plastic strain is a scalar
X Reference configuration of the N-dimensional
body
and hence is simply denoted p whereas the cumulated
@D X Part of the boundary where the displacements are plastic strain is defined by
prescribed
@F X Part of the boundary where the surface forces are _ ¼j p_ j :
p
prescribed
x ¼ ðx1 ; . . . ; xN Þ Material point and its cartesian coordinates in the In the multi-dimensional case we adopt the plastic incom-
reference configuration pressibility condition and hence p is assumed to be purely
t Time variable  is a sca-
deviatoric, Tr p ¼ 0. The cumulated plastic strain p
MN s
Space of symmetric second order tensors in a N-
lar defined by
dimensional setting
u Real displacement field
rffiffiffiffiffiffiffiffiffiffiffiffiffi
N1
n ¼ ðu; a; p; p Þ Real state field of the body _ ¼
p _
kpk:
JðnÞ Surface where the state field is singular N
sut Jump discontinuity of the displacement field
Accordingly, all the cases can be synthesized by the follow-
pR ; p
R Regular parts of the plastic strain and cumulated
plastic strain fields ing formula

P Surface density of the singular part of the
cumulated plastic strain field _ ¼ kN kpk
p _ with Tr p ¼ 0 if N > 1 and
q; qR Virtual plastic strain field and its regular part (
1 if N ¼ 1;
Jðv Þ Jump set of the virtual displacement field v kN ¼ qN1 ffiffiffiffiffiffiffi ð3Þ
N
otherwise:

In (2), the fourth tensor E is the (invariable) stiffness tensor

2. The gradient damage model coupled with plasticity and the material constant rP represents the Von Mises
yield stress. Hence, the first term represents the stored
2.1. Definition of the total energy state function elastic energy and the second the energy dissipated during
the plasticity process.
For quasi-brittle materials, one uses gradient damage To couple damage with plasticity, we first define the
models which are defined by assuming that the energy state of a volume element by the quintuple
density has the following form Þ 2 MNs  ½0; 1  RN  MNs  Rþ and then de-
ðe; a; ra; p; p
fine the total energy density as the following function of
1
WD ðe; a; raÞ ¼ EðaÞe  e þ dðaÞ þ d1 ‘2 ra  ra; ð1Þ state:
2
1
where e 2 MNs is the strain tensor and a is a scalar charac- Wðe; a; ra; p; p
Þ ¼ EðaÞðe  pÞ  ðe  pÞ þ dðaÞ
2
terizing the damage level in the material; respectively þ d1 ‘2 ra  ra þ rP ðaÞ p
: ð4Þ
a ¼ 0 and a ¼ 1 mean a sound and a totally damaged
material. Here EðaÞ represents the material stiffness tensor, The definition (1) of the energy density has been altered by
supposed decreasing with the damage variable, dðaÞ introducing the plastic strain field p in the stored elastic
account for the energy dissipation due to damage, ra energy, and by adding the dissipation due to plastic defor-
 R. Alessi et al. / Mechanics of Materials 80 (2015) 351–367
mations, namely rP ðaÞ p
. Our choices in (4) are minimalist
in the sense that
(i) the damage variable is still a scalar and one does not
introduce another internal variable than the cumu-
lated plastic strain in the dissipated energy state
function. At this purpose, let us note that the most
used models in ductile fracture by the engineering
community, like Gurson-like models, contain also
only one damage variable, namely the microvoid
porosity;
(ii) for a ¼ 0 we are actually considering the standard
Von Mises model of perfect plasticity. But there is
no difficulty to change the shape of the plasticity cri-
terion and more complex choices will be considered
in the future;
(iii) the coupling between damage and plasticity simply
consists in introducing a dependence of the yield
plastic stress rP ðaÞ on the damage variable. Despite
this simplicity and by virtue of the variational char-
acter of the model, we will see that the product
r0P ðaÞp of the derivative of the state function rP ðaÞ
by the cumulated plastic strain p  enters in the dam-
age criterion and this coupling will play a funda-
mental role in the onset of damage by plastic
accumulation or in the nucleation of a cohesive
crack at the center of a damage zone.
2.2. The total energy functional for a N-dimensional body
In (4), W denotes the energy of the volume element. Let
us now consider a N-dimensional continuum body whose
reference configuration is the open bounded domain X
and which is submitted to a time-dependent loading pro-
cess characterized by time-dependent surface forces Ft
prescribed on the part @ F X of the boundary of X and by
time-dependent displacements Ut prescribed on the com-
plementary part @ D X ¼ @ X n @ F X. (To simplify the presen-
tation, we will neglect body forces.) Then, at a given time
t, the total energy of the body is the following functional
of the quadruple
n ¼ ðu; a; p; p
Þ
called the global state field, made of the displacement field
u, the damage field a, the plastic strain field p and the
cumulated plastic strain field p:
Z
E t ðu; a; p; p
Þ ¼ Wðrs uðxÞ; aðxÞ; raðxÞ;pðxÞ; p
ðxÞÞdx
X
Z
 Ft ðxÞ  uðxÞdS:
@F X
The above expression of the energy makes sense provided
that the global state field n is smooth enough. Let us briefly
discuss these questions of regularity.
1. As far as the damage field is concerned, the gradient
term requires that a 2 H1 ðXÞ so that the total energy
be finite. Therefore, the damage field is continuous
across any surface, but the gradient of damage can be
discontinuous.
355
2. As far as the displacement and the plastic strain fields
are concerned, in our perfect plasticity setting (at given
damage) the elastic strain field rs u  p only is regular
so that the elastic energy be finite (it must belong to
L2 ðXÞ). In such a setting, it is well known that plasticity
can localize in shear bands. We will see that it is actu-
ally the case once the damage has appeared, because
of the localization of damage induced by the softening
character of the model.
Accordingly, we will assume that the global state field
n ¼ ðu; a; p; p
Þ is piecewise smooth and that its ‘‘singular
part’’ is localized on a n-dependent set JðnÞ  X, called
the jump set of n, which contains a finite number of smooth
and non-intersecting surfaces in X. For a sake of simplicity,
we also assume that this set has zero intersection with the
boundary @ X, see Fig. 3. Specifically, we assume that
(i) the displacement field u is continuously differentia-
ble on X n JðnÞ and admits a jump discontinuity on
JðnÞ. Therefore, the strain field e associated with u
can be decomposed into two parts: its regular part
eR which corresponds to the usual symmetric part
of the gradient of u and its singular part which can
be seen as a Dirac measure concentrated on the
jump set JðnÞ. The singular part of e is denoted by
eS and is written with an abuse of notation as
eS ¼ suts ndJðnÞ ; ð6Þ
where n is the normal to the positive face of
JðnÞ; sut :¼ uþ  u and dJ is the Dirac surface measure
concentrated on the surface J.
(ii) in order that the elastic energy be finite, the plastic
strain field p has the same singular part as the strain
field and hence its singular part pS will also read as
pS ¼ suts ndJðnÞ ð7Þ
while its regular part (denoted x # pR ðxÞ) is assumed at
least continuous on X n JðnÞ. In the multi-dimensional case,
for p to be deviatoric its singular part must be such that
Trðsuts nÞ ¼ sut  n ¼ 0; hence in dimension higher than
one, only the tangential components of the displacement
can jump through JðnÞ.
ð5Þ
Fig. 3. The N-dimensional body with its loading and the jump set Jðnt Þ of
the current state field nt (light gray curve) inside a plastic-damaged non
singular zone (dark gray zone).
356 R. Alessi et al. / Mechanics of Materials 80 (2015) 351–367

(iii) in the same manner, the cumulated plastic strain the multi-dimensional case, Tr q ¼ 0 and if the singular
field is decomposed into regular and singular parts part of q is related to the jump of v by
R and p
p S . Its singular part reads as
qS ¼ sv ts n dJðvÞ : ð11Þ
p ¼ P dJðnÞ
S ð8Þ 1
In order that at 6 a
< 1 and that a
2 H ðXÞ, it is
 denotes its surface density.
and thus P necessary and sufficient that b  0; b 2 H1 ðXÞ and h be
small enough. A triple of fields ðv ; b; qÞ which satisfies
Finally, the total energy of the body in the global state the above conditions will be called an admissible direction
n ¼ ðu; a; p; p
Þ will read as of perturbation. We are now in a position to define the
Z condition of stability.
1
E t ðu; a;p; p
Þ ¼ EðaðxÞÞðrs uðxÞ  pR ðxÞÞ
2
XnJðnÞ Definition (Local stability). The state ðut ; at ; pt ; p
t Þ of the
 
 rs uðxÞ  pR ðxÞ dx body at a time t before the nucleation of a crack is said
Z locally stable if, for every admissible direction of pertur-

þ R ðxÞ
dðaðxÞÞ þ rP ðaðxÞÞp  > 0 such that for all h 2 ½0; h
bation ðv ; b; qÞ, there exists h 
XnJðnÞ
Z

þd1 ‘2 raðxÞ  raðxÞ dx þ 
rP ðaðxÞÞPðxÞdS E t ðut þ hv ; at þ hb; pt þ hq; p
t þ hkN kqkÞ P Eðut ; at ; pt ; p
t Þ:
JðnÞ
Z ð12Þ
 Ft ðxÞ  uðxÞdS: ð9Þ
@F X
2.3.3. Energy balance
The only singular part which appears in (9) is that of the Following the presentation of Pham and Marigo (2010b,
cumulated opening, because a and ra are not singular 2011b), the energy balance principle in our particular
whereas e and p have the same singular part. setting reads as
Z Z
d
2.3. Damage irreversibility, local stability and energy balance E t ðut ; at ; pt ; p
t Þ ¼ rt n  U_ t dS  F_ t  ut dS: ð13Þ
dt @D X @F X

At this stage of the construction, nothing was said on
the laws governing the evolution of the damage and the
plasticity. One of the main advantages of the variational
approach is that those laws are simple byproducts of the
general physical principles of irreversibility, stability and
energy balance once the total energy has been defined.
These principles are briefly recalled below and the reader
interested by more details can refer to Mielke (2005,
2006), Bourdin et al. (2008), Pham and Marigo (2010a,b)
and Francfort and Giacomini (2012). Then, this section
and Appendix A will be devoted to the deduction of the
damage and plasticity evolution laws from these three
principles and the assumed form (9) of the energy.
2.3.1. The irreversibility condition
We require that at every point the damage can only in-
crease with time, i.e.,
a_ t ðxÞ  0; 0 6 at ðxÞ 1; 8x 2 X: ð10Þ
Moreover, to simplify the presentation, we will only con-
sider the evolution before the nucleation of a full damaged
set and hence we assume that at < 1 everywhere in X.
2.3.2. Stability condition
Let nt ¼ ðut ; at ; pt ; p
t Þ be the state of the body at time t
and let n
¼ ðu
; a
; p
; p
Þ be the following virtual state first order stability conditions.
n
¼ nt þ h ðv ; b; q; kN kqkÞ;
where h is a positive constant. In order that u
be kinemat-
ically admissible, the field v must be such that v ¼ 0 on
@ D X. Moreover, the field v is assumed, like ut , piecewise
smooth and we denote by Jðv Þ the set of points where v
is discontinuous. Therefore the jump set of n
is and the Poisson ratio), dðaÞ and r  P ðaÞ which give the
Jðn
Þ ¼ Jðnt Þ [ Jðv Þ. Finally the field p
is admissible if, in dependence of the stiffness, the local damage dissipated
Therefore, (13) is a global (and single) equation which
involves the total energy of the whole body and which
must hold at every time. Note that the right hand side term
in (13) involves the rate of the data whereas the rate n_ t of
the state field will appear in the time derivative of the total
energy.
2.3.4. Consequences of the principles of irreversibility, stability
and energy balance on the evolution law
Using the above three general principles with the par-
ticular form of the total energy functional gives a set of
necessary conditions that the evolution must satisfy. The
details of this construction of the evolution problem are
given in Appendix A to which the reader is invited to refer
in order to understand the origin of each condition. All
these conditions are summarized in Table 2 in the multi-
dimensional case. Note that these conditions are necessary
but in general not sufficient in order that the ‘‘true’’ stabil-
ity condition (12) be satisfied at each time of the evolution.
One must also verify the so-called second order stability
conditions. But these additional conditions will not be used
explicitly here and the interested reader can refer to Pham
et al. (2011b), Pham and Marigo (2013b) and Sicsic et al.
(2014) where they play an essential role to select the good
evolutions in the cases where many evolutions satisfy the
2.4. Constraint on the constitutive functions
Our model (4) contains three state functions: EðaÞ
(which contains for an isotropic material two scalar state
functions, namely YðaÞ and mðaÞ giving the Young modulus
 R. Alessi et al. / Mechanics of Materials 80 (2015) 351–367 357

Table 2
In the multi-dimensional case, the conditions that the evolution must verify in order that the irreversibility condition, the first order stability conditions and the
energy balance are satisfied. Note in particular that the cumulated plastic strain is present in the damage condition and since r0P < 0 the damage critical stress
is monotonically decreasing when the plasticity evolves. Note also all the conditions which are obtained on the jump set and which are rarely mentioned in the
literature. These latter conditions are essential to obtain the cohesive law as we will see in the next sections.

Name Statement Domain

Stress–strain relation r ¼ EðaÞðrs u  pÞ X
Irreversibility a_  0 X
Equilibrium in the bulk divr ¼ 0 X n JðnÞ
Damage criterion in the bulk R  12 C0 ðaÞr  r  2d1 ‘2 Da P 0
d0 ðaÞ þ r0P ðaÞp X n JðnÞ

Damage consistency in the bulk R  12 C0 ðaÞr  r  2d1 ‘2 Da a_ ¼ 0

d0 ðaÞ þ r0P ðaÞp X n JðnÞ

Plastic yield criterion in the bulk krD k 6 kN rP ðaÞ X n JðnÞ

Plastic flow rule in the bulk D
p_ R ¼ kp_ R k k rr ðaÞ X n JðnÞ
N P

Equilibrium on the jump set srtn ¼ 0 JðnÞ

Damage criterion on the jump set r0P ðaÞP  2d1 ‘2 s@ a=@nt  0 JðnÞ

Damage consistency on the jump set r0P ðaÞP  2d1 ‘2 s@ a=@nt a_ ¼ 0 JðnÞ

Plastic yield criterion on the jump set krn  ðrn  nÞ nk 6 jN rP ðaÞ JðnÞ
Plastic flow rule on the jump set _ rnð
_ ¼ ksutk
sut rnnÞ n
j r ðaÞ
JðnÞ
N P

Dirichlet boundary condition u¼U @D X

Neumann boundary condition rn ¼ F @F X
Damage boundary condition @a @X
@n  0
Damage consistency on the boundary @a _ @X
@n a ¼ 0

energy and the plastic yield stress on the damage variable.
ðaÞ ¼ r 2 MNs j krD k 6 kN rP ðaÞ ;
The physical meaning of these constitutive functions
dictates some natural constraints to be required.
We start from the function a # dðaÞ, fixing the amount
of dissipated energy with respect to the damage state. Spe-
 
P
ð18Þ
which represents the stress domain of elasticity associated
with the plasticity criterion. All these elastic domains
depend on the damage state. Note that, since

aðaÞ ¼ E1
0 EðaÞ, the inequality defining D ðaÞ can be read as
cifically, we assume that it is smooth monotonically
increasing function with r2D ðaÞ 2 d0 ðaÞ
0 E1
0 rr 6 :¼ :
dð0Þ ¼ 0; d ðaÞ > 0; 8a 2 ½0;1Þ; dð1Þ ¼ d1 < þ1: ð14Þ Y0 ð1=aÞ0 ðaÞ
Concerning the stiffness decrement due to damage, we The definitions of D
ð0Þ and P
ð0Þ introduce two thresh-
limit here the attention to the case where olds, r  D :¼ rD ð0Þ and r  P :¼ rP ð0Þ, for the stress; whether
EðaÞ ¼ aðaÞ E0 r D < r P or r D > r P discriminates if, in a mono-axial trac-
tion test, damage or plasticity will evolve first; thus r D
is given in term of one scalar function aðaÞ and the initial and r  P could be interpreted as the initial yield limits.
sound stiffness E0 . (Of course, more general cases could The strain-hardening property and the stress-softening
be considered and that will be the subject of future works.) properties are respectively equivalent to the growth of the
In isotropic material such an assumption corresponds to set DðaÞ and to the decrease of the set D
ðaÞ with respect to
the following dependence the damage variable; these requirements translate into the
YðaÞ ¼ aðaÞ Y0 ; mðaÞ ¼ m0 ð15Þ following requests:

for the Young and Poisson moduli, respectively. Hence the Strain hardening condition:
function a # aðaÞ gives the evolution of the material Young d0 ðaÞ
modulus with its damage state. We assume that it is a a# is monotonically increasing ð19Þ
j a0 ðaÞ j
smooth monotonically decreasing function with
and
að0Þ ¼ 1; a0 ðaÞ < 0; 8a 2 ½0; 1Þ; að1Þ ¼ a0 ð1Þ ¼ 0: ð16Þ
Stress softening condition:
Additional constraints are needed on the constitutive func-
d0 ðaÞ
tion a # aðaÞ in order to obtain strain-hardening and a# is monotonically decreasing: ð20Þ
stress-softening behaviors of the material. To this aim,
ð1=aÞ0 ðaÞ
one defines the following sets: Finally, it would be natural to require that the plastic
  yield stress progressively decreases when damage grows
1
DðaÞ ¼ e 2 MNs j  E0 ðaÞe  e 6 d0 ðaÞ ; and finally vanishes when the material is fully damaged.
2
  Within this softening behavior framework, we assume
1
D
ðaÞ ¼ r 2 MNs j C0 ðaÞr  r 6 d0 ðaÞ ; ð17Þ the following properties for the plastic yield stress state
2 function a # rP ðaÞ:
which represent the domains of elasticity associated with Plastic yield stress softening:rP ð0Þ ¼ rP > 0;
the damage criterion at vanishing cumulated plasticity p, r aÞ < 0; 8a 2 ½0; 1Þ; rP ð1Þ ¼ 0; r0P ð1Þ 0:
0
Pð ð21Þ
and
358 R. Alessi et al. / Mechanics of Materials 80 (2015) 351–367

Accordingly, our model is quite different of Ambrosio et al. The solid curves in Fig. 4 illustrates the dependencies of
(2012) and Del Piero et al. (2012) even if those models have the constitutive functions on the damage variable for
also the goal for coupling fracture with plasticity by using a k ¼ 3; n ¼ 2 and h ¼ 1=2. In Section 3 we will discuss in
variational approach. some details the class of models k > 1, n P 1 and h < 1 in
which this choice falls; this class physically corresponds
2.5. A family of admissible constitutive functions to the stress–strain relation shown in Fig. 5, where a purely
plastic phase is followed by a coupled evolution of damage
A useful choice of the constitutive function respecting and plasticity.
the conditions given above is the following one which de- The conditions (14) and (16), the strain hardening con-
pends on three dimensionless parameters k > 1, h > 0, dition (19), the softening condition (20) and the plastic
n > 0 and on three parameters characterizing the material yield softening condition (21) are automatically satisfied.
stiffness, the material strength and the material length, The parameter h represents the ratio between r  P and r
 D.
namely Y0 , r D ; ‘: The limit case where h ¼ þ1 would correspond to a pure
damage model without plasticity. It corresponds to the
1  wðaÞ
aðaÞ ¼ ; type of damage models which is used in the variational
1 þ ðk  1ÞwðaÞ
approach to fracture (see Amor et al., 2009; Pham et al.,
kr
2
2011a,b; Sicsic and Marigo, 2013), and is close to those
dðaÞ ¼ D wðaÞ; rP ðaÞ ¼ ð1  wðaÞÞn hr
 D; ð22Þ
2Y0 used in Lorentz et al. (2011) for quasi-brittle materials.
where
3. Homogeneous solutions and their stability
wðaÞ ¼ 1  ð1  aÞ2 : ð23Þ
Thus the state functions depend in fact on the variable Let us study the response predicted by the damage–
x ¼ wðaÞ which grows from 0 to 1 as a does. Accordingly, plasticity coupled model when a single material point is
one gets submitted to a uniaxial traction test where r ¼ re1  e1
and where the associated strain component e ¼ e11 is con-
rD ðaÞ ¼ ð1  wðaÞÞr D ¼ ð1  aÞ2 r D ð24Þ
trolled. The axial component p11 of the plastic strain is de-
and _ ¼ jp_ j.
noted p and the cumulated plastic strain is given by p
The state of the material point can be identified with the
kr
 2D
d1 ¼ ; r P ¼ hr D : quadruple ðe; a; p; p
Þ. We assume that the material point
2Y0 is at time 0 in the unstrained, unstressed and undamaged

Fig. 4. Constitutive functions as in (22)–(24) when k ¼ 3; n ¼ 2 and h ¼ 1=2.

Fig. 5. Response of the volume element in the case h < 1. Left: for n > 1, the response corresponds to the sequence E–P–DP; Right: for n ¼ 1, the response
corresponds to the sequence E–P–D–F.
 R. Alessi et al. / Mechanics of Materials 80 (2015) 351–367 359

state, i.e., ðe0 ; a0 ; p0 ; p

0 Þ ¼ ð0; 0; 0; 0Þ, and then is submitted
to an increasing uniaxial stretching where e grows from 0
to þ1. Accordingly, we can assimilate the time parameter
with the strain, i.e., e ¼ t. The problem is to find the evolu-
tion of ða; p; p Þ with e. That evolution is assumed to be
smooth, in the sense that e # ðae ; pe ; p e Þ are at least abso-
lutely continuous, and governed by the stress–strain rela-
The plastic stage ends at the limit strain eDP ð0Þ. Indeed,
during the plastic stage, the plastic strain accumulation re-
duces the damage yield limit until the current value of
stress is reached. Hence one seeks for the value of strain
where both yield criteria are satisfied: eliminating the
plastic strain leads to the following

tion, the damage irreversibility condition, the damage eDP ðaÞ ¼ ð1  aÞ2nþ2 k  h2 ð1  aÞ2n2 ðk  2nð1 þ ðk  1Þð2  aÞaÞÞ
and plasticity yield criteria, the damage consistency equa- eD
 :
tion and the plasticity flow rule. For the reader conve- 2hn
nience, these equations, derived in the previous section
As in the plastic stage P we had a ¼ 0, we immediately get:
within a three-dimensional setting, are here reported in
scalar form: kð1  h2 Þ
eDP ð0Þ ¼ heD þ eD ;
stress—strain relation : r ¼ aðaÞY0 ðe  pÞ; ð25Þ 2nh
a value strictly larger than heD , see Fig. 5. Finally, as for
irreversibility: 0 6 a 1; a_  0; ð26Þ
e ¼ eDP ð0Þ the damage yield has been reached, the damage
evolves. It turns out that, for this class of constitutive
1
damage criterion:  a0 ðaÞY0 ðe  pÞ2 parameters, both the yield limits are identically satisfied;
2
hence we have a damage–plastic stage DP, where both
 d0 ðaÞ  r0P ðaÞp
 0 if a < 1; ð27Þ
damage and plasticity could evolve. Specifically we have

plastic yield criterion: jrj  rP ðaÞ 0; ð28Þ 8e 2 ½eDP ð0Þ; eDP ða ! 1Þ; ae ¼ e1DP ðeÞ;
e ¼ pDP ðae Þ;
pe ¼ p r ¼ rP ðae Þ;
damage consistency relation:
  where
1 0
a ðaÞY0 ðe  pÞ2 þ d0 ðaÞ þ r0P ðaÞp
 a_ ¼ 0; ð29Þ
2 keD

pDP ðaÞ :¼ ð1  aÞ2nþ2  h2 ð1  aÞ2n2 :

plastic flow rule : rP ðaÞjp_ j  rp_ ¼ 0:
Clearly when the strain fields are identified with scalar
fields, as done here and in the following section, the re-
quest on the plastic strains to be deviatoric must be re-
moved. While the study is trivial in the case of
uncoupled models, it becomes much more difficult in the
case of a coupling. In particular, the existence of the re-
sponse is not ensured in the whole range of strains without
introducing additional assumptions on the constitutive
relations. Moreover, one can obtain a great variety of re-
sponses according to the values of the material parameters
entering in the model. Thus we limit the analysis to the
class of constitutive functions (22) with k > 1; n P 1 and
h < 1. This case seems to be particularly relevant for many
applications; possible responses in terms of stress–strain
relationship are depicted in Fig. 5.
Starting from a sound, unstrained and unstressed state,
we find first an elastic stage E; comparing the two yield
functions (27) and (28) at a ¼ p ¼ p  ¼ 0 we find that the
limit strain eI for the elastic stage is
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2d0 ð0Þ r D
eI ¼ minfeD ; eP g; eD :¼ ¼ ;
ja0 ð0ÞjY0 Y0
rP ð0Þ hr
D
eP :¼ ¼ ¼ heD :
Y0 Y0
Since h < 1 then eI  heD and the plastic yield is firstly
reached. During the elastic stage E, we have
8e 2 ½0; heD ; ae ¼ 0; pe ¼ pe ¼ 0; r ¼ Y0 e:
During the plastic stage P, we have
8e 2 ½heD ; eDP ð0Þ; ae ¼ 0; pe ¼ pe ¼ e  heD ; r ¼ hr D :
2hn
ð30Þ
Note that for n > 1 the DP stage continues up to infinity as
eDP ða ! 1Þ ¼ þ1 and the plasticity actually increases
monotonically (Fig. 5, left). However, when n ¼ 1, we have
kð1 þ h2 ÞeD kð1  h2 Þ
eDP ð1Þ ¼ ¼: e1 ; e ¼
pe ¼ p eD :
2h 2h
Thus for n ¼ 1 the plastic strain remains fixed at the value
reached at the end of the plastic stage P, and we actually
have a pure damage stage D, see Fig. 5 (right). A totally
damaged material F, namely the condition a ¼ 1, is
reached for a finite strain as e1 < 1.
4. Closed form solution for the one-dimensional
traction test
4.1. General assumptions
Again with the purpose of simulating a uniaxial traction
test, we now allow the fields to be functions of one spatial
coordinate. To this aim we consider X ¼ ð0; LÞ as the refer-
ence configuration of a one-dimensional body called the
bar. Its end x ¼ 0 is fixed and the end x ¼ L is submitted
to a time dependent displacement U t with U 0 ¼ 0. The
bar is made of a material whose local behavior is given
by the plasticity–damage models described in the previous
section. The equations governing the evolution of the bar
with time t are the scalar form of the equations derived
in Section 2; their solution is given is the map
t # ðut ; at ; pt ; p
t Þ for t  0, where ut ; at ; pt and p
t denote
now, respectively, the displacement field, the damage field,
the plastic strain field and the cumulated plastic strain
field of the bar at time t. We assume that, at time t ¼ 0,
360 R. Alessi et al. / Mechanics of Materials 80 (2015) 351–367
the bar is sound and was never plasticized so that
a0 ¼ p0 ¼ p0 ¼ 0 everywhere in X.
It is easy to check that the homogeneous response, i.e.,
the response such that ut ðxÞ ¼ U t x=L; at ðxÞ ¼ at and
pt ðxÞ ¼ pt for all x 2 X, is still possible. In the case where
U t ¼ tL, i.e., for a monotonically increasing traction test,
the homogeneous response is precisely that obtained in
the previous section for the volume element. This evolu-
tion, namely the E, P, and DP stages described above, sat-
isfies the irreversibility condition, the first order stability
conditions and the energy balance because, in particular,
the gradient of damage vanishes and Jðnt Þ ¼ ;. However,
we are no more ensured that it is the unique solution.
Moreover, we are not ensured that the local stability con-
dition (12) is satisfied by the homogeneous response. If
we refer to what happens in the case of gradient damage
models with softening (without plasticity), we know that
the homogeneous response is unique and stable if and only
if the length L of the bar is sufficiently small by comparison
with the internal length ‘ of the material (Pham et al.,
2011b; Pham and Marigo, 2013b). When the length of
the bar is large enough, the homogeneous response is not
stable and it is possible to construct non homogeneous
Fig. 6. In the case h < 1 and n ¼ 1, construction of the solution in the
phase plane at a given stress r < hr
 D . The circle represents the damage
field and the dashed line corresponds to the jump of x0 at the center of
the damage zone.
Fig. 7. Case with h ¼ 2=3; n ¼ 1 and k ¼ 4: evolution of the damage after the plasticity stage. On the left, evolution in the phase plane ðx; ‘x0 =2hÞ; on the
right, evolution in the physical space. The presence of a cohesive crack from the beginning of the damage process is visible on both spaces.
responses. Accordingly, we propose here to follow the
same procedure and, assuming that L is sufficiently large
by comparison with ‘, to construct a response where the
damage, when it appears, remains localized on a time-
dependent part of the bar. To construct such an evolution,
we follow the method proposed in Pham et al. (2011b) and
Pham and Marigo (2013a).
To simplify the presentation and to prevent from con-
sidering too many cases, we construct such non homoge-
neous responses for the family of models considered at
Section 2.5 only. Therefore, a; d and rP are given by (22)
and (23) with k > 1; n ¼ 1 and h < 1. Despite the cases
n > 1 could be treated similarly, we focus the attention
on the case n ¼ 1 because the solution can be obtained in
a closed form and all the steps for constructing the solution
are easier.
4.2. Plasticity stage followed by damage localization with
nucleation and growth of a cohesive crack
The analysis starts at a time when the damage yield cri-
terion is reached somewhere in the bar. This time tc corre-
sponds to the end a P stage as h < 1. We will assume, to
simplify the presentation, that the plastic strain field and
the cumulated plastic strain field are uniform at t c . Hence,
the state of the bar at t c is ntc ¼ ðeDP ð0Þx; 0; pDP ð0Þ; pDP ð0ÞÞ,
the stress is rtc ¼ hr
 D and the damage yield criterion is
also reached at every point of the bar:
ð1  h2 ÞkeD
aðxÞ ¼ 0; pðxÞ ¼ pðxÞ ¼ pDP ð0Þ ¼ ; r ¼ hr D :
2h
When t > tc , we assume that rt is monotonically decreas-
ing from rtc to 0. We seek for non homogeneous evolutions
such that the damage zone is the interval ðx1  Dt ; x1 þ Dt Þ
where x1 is an arbitrary point of the bar sufficiently far
from its ends so that 0 < x1  Dt 6 x1 þ Dt < L. Thus, we
exclude the case where the damage zone is at the bound-
ary. The half width Dt of the damage zone, which can de-
pend on time, has to be determined. We will assume that
the center x1 of the damage zone is the unique possible sin-
gular point, i.e., Jðnt Þ ¼ ; or Jðnt Þ ¼ fx1 g.
Therefore, the plasticity can only evolve at x1 by virtue
of (43) and remains equal to pDP ð0Þ otherwise,
 R. Alessi et al. / Mechanics of Materials 80 (2015) 351–367 361

pRt ðxÞ ¼ p
Rt ðxÞ ¼ pDP ð0Þ; 8x 2 ð0; LÞ n fx1 g; 8t P t c : So Dr increases from p‘=4h to p‘=2h when r goes from hrD
to 0. For a given r, the damage profile in the damage zone
Accordingly, by virtue of (53), the damage field must
is given by
satisfy when 0 < r < hr
 D:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
aðxÞ ¼ 1  1  xðxÞ
C0 ðaÞr2 þ 2d0 ðaÞ þ 2r0P ðaÞpDP ð0Þ  4d1 ‘2 a00 ¼ 0 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 
in Ir n fx1 g ð31Þ ur hjx  x1 j
¼1 1x  r cos2 þ in Ir :
2 ‘
and
aðx1  Dr Þ ¼ a0 ðx1  Dr Þ ¼ 0; ð32Þ Since x r is increasing and ur is decreasing (when r de-
creases), the damage grows at given x and hence the irre-
where CðaÞ ¼ 1=ðaðaÞY0 Þ is the compliance state function, versibility condition is satisfied. The damage evolution is
Ir ¼ ðx1  Dr ; x1 þ Dr Þ denotes the damage zone and Dr is represented on Fig. 7 for h ¼ 2=3.
its half-width which has to be determined. The conditions As long as the cohesive law is concerned, one gets
at x1 depends on whether x1 is singular or not, but in any rffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffi!
case and since n ¼ 1 the plasticity criterion requires that hr D r
sut ¼ keD ‘  or equivalently
2 r hrD
r 6 rP ðaðx1 ÞÞ ¼ hð1  aðx1 ÞÞ r D : ð33Þ
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi !2
Multiplying (31) by 2a , one obtains a first integral with
0 sut2 sut
r ¼ hrD 1þ 2 2 2 ð35Þ
the constant given by (32). Specifically, one gets 4k eD ‘ 2keD ‘

2d1 ‘2 a02 ¼ 2dðaÞ  2ðrP ð0Þ  rP ðaÞÞpDP ð0Þ  ðCðaÞ the cohesive crack appearing as soon as the damage starts.
 Cð0ÞÞr2 in Ir n fx1 g: ð34Þ When r ¼ 0, then u0 ¼ 0; x  0 ¼ 1 and aðx1 Þ ¼ 1. A true
crack has nucleated at x1 and the damage profile is
This property holds for any plasticity–damage model. In    
the case of the models given by (22) and (23) with n ¼ 1, hjx  x1 j p‘ p‘
aðxÞ ¼ 1  sin in x 1  ; x1 þ ;
after introducing the variable x, the first integral eventu- ‘ 2h 2h
ally reads as when r ¼ 0: ð36Þ
r2 The dissipated energy inside the damage zone at the end of
‘2 x0 2 ¼ 4h2 xðx
 r  xÞ where x
r ¼ 1 2 :
h r
 2D the damage localization process, i.e., when r ¼ 0 is given
by
Hence, in the normalized phase plane ðx; ‘x0 =2hÞ, the first
Z x1 þD0

integral is a circle of center x r =2 and radius x

 r =2. The
D0 ¼ dðaðxÞÞ þ rP ðaðxÞÞpDP ð0Þ þ d1 ‘2 a0 ðxÞ2 dx;
plasticity criterion (33) requires that x1  D 0

r with a given by (36). A part, namely 2rP ð0ÞpDP ð0ÞD0 , was
xðx1 Þ 6 xr :¼ 1   :
hrD dissipated during the P stage. So, if we define Gc as the dis-
 r as soon as r < hr
But since x
r < x  D ; x0 is necessarily dis- sipated energy due to the damage process alone, then we
continuous at x1 and hence x1 is a singular point, see Fig. 6. obtain
Therefore, a cohesive crack appears as soon as r < hr  D and, Z x1 þ D 0

by virtue of (54) and (58), one must have Gc :¼ dðaðxÞÞ þ ðrP ðaðxÞÞ  rP ð0ÞÞpDP ð0Þ þ d1 ‘2 a0 ðxÞ2 dx:
x1 D0

xðx1 Þ ¼ x
r ; r0P ða
r Þsutðx1 Þ ¼ 2d1 ‘2 sa0 tðx1 Þ: After some calculations, one gets
 r cos2 u2ðxÞ gives ‘2 u02 ¼ 4h2 . Therefore,
Setting xðxÞ ¼ x
pk hr 2D
the half width Dr of the damage zone is given by Gc ¼ ‘:
2 Y0
‘ hr
D  r
Dr ¼ ð p  u r Þ with ur ¼ arccos : Note that this value of Gc involves all the parameters of the
2h hr
 þr D
model.

5. Numerical implementation

The numerical implementation of the proposed model

Fig. 8. Graph of the cohesive law obtained for n ¼ 1 which gives r in term
of sut. Note that the curve is monotonically decreasing, but tends only
asymptotically to 0.
takes advantage from the variational formulation. Indeed,
the stability condition is directly translated into a numeri-
cal strategy based on seeking local energy minimizers. The
numeric simulations will cover the most delicate aspect of
the problem, that is the ability of the algorithm to under-
stand and describe the plastic localization and the cohesive
response in both a one-dimensional and a two-dimen-
sional traction test.
For technical reasons, the constitutive functions used in
the numerical simulations are chosen to be:
362 R. Alessi et al. / Mechanics of Materials 80 (2015) 351–367

r 2P iterating alternatively on the three following subproblems

aðaÞ ¼ ð1  aÞ2 ; dðaÞ ¼ a; rP ðaÞ ¼ ð1  aÞ2 r P ;
h2 Y0 until convergence:
ð37Þ
The minimization of E with respect to u at fixed p
with 0 < h < 1. This model satisfies all the required prop- and a: this is a straightforward unconstrained opti-
erties of strain hardening (19), stress softening (20) and mization problem solved as an elastic problem with
plastic yield stress softening (21), but it does not belong prescribed boundary conditions;
to the family of models (22) because of the linear depen- The minimization of E with respect to p at fixed u
dence of d on a. However this choice allows to use dedi- and a: this is a nonlinear constrained problem. Since
cated libraries for the minimization of the damage no space derivatives involves the field p the optimal-
variable, which turns out to be a constrained quadratic ity condition is local although non linear. A common
optimization problem. The material characteristic length solution procedure involves a standard return
‘ is chosen sufficiently small by comparison to the size L mapping algorithm;
of the body so that damage localization be possible. We The minimization of E with respect to a at fixed u
still have and p: this is a box constrained quadratic optimiza-
tion problem, which can be solved by a linear bound
r P r P r D eP
r D ¼ ; eP ¼ ; eD ¼ ¼ constraint solver; in particular we use the TAO (Tool-
h Y0 Y0 h
kit for Advanced Optimization) library.1
and the damage criterion reads now
! 5.2. One-dimensional traction test

p h2
2 rr ðTr rÞ2
1  2ð1  aÞh  ð1 þ mÞ 2  m
eP ð1  aÞ 3
r P r 2P For the one-dimensional traction test, namely the same
 2‘2 Da  0: problem analytically solved in Section 4, we have

X ¼ ð0; LÞ; uð0Þ ¼ 0; uðLÞ ¼ U with U increasing from 0

5.1. Time and space discretisation and numeric algorithm and the response essentially depends on the two dimen-
sionless parameters ‘=L and h. The computations are made
As the imposed displacement boundary conditions will with
be chosen proportionally to t, the time can be interpreted pffiffiffi pffiffiffi
as a multiplier of the applied external action. We follow ‘=L ¼ 0:15 2  0:21; h ¼ 1= 2  0:71:
the minimizers of the total energy as t is increased in steps
Following the method described in the previous section, a
of uniform amplitudes until t ¼ T for T sufficiently large.
solution with a damage localization zone can be obtained
For both 1D and 2D cases, the code has been imple-
in a closed form. Let us recall its main features. The elastic
mented within the finite element framework FEniCS (Logg
stage E ends when U ¼ eP L, then a plastic stage P starts
et al., 2012). Specifically the displacement and damage
during which the plastic strain field remains uniform in
fields are projected over a piecewise affine finite element
space. The plastic stage ends when the damage criterion
space (1-Lagrange elements) using the same triangulation
is reached (everywhere) in the bar. That corresponds to
domain. Motivations for not using higher degree finite ele-  ¼ pc ; U c and pc being
the moment when U ¼ U c and p ¼ p
ments can be found for example in Bourdin et al. (2000).
given here by
Conversely, the plastic strain field p is projected over a dis-
crete discontinuous space (Quadrature elements) defined 1 þ h2
only over the Gauss integration points. These last corre-
U c ¼ eDP ð0ÞL :¼ eP L ¼ 1:5 eP L;
2h2
spond to the centers of the elements, as suggested by the 1  h2
local character of the plastic model. pc ¼ pDP ð0Þ :¼ eP ¼ 0:5 eP :
2h2
It is worthnoting that apparently there was no chance
to describe displacement jumps and plastic singularities For U > U c damage occurs and the damage field is localized
leading to a cohesive response since the chosen finite ele- in an interval I ¼ ðx1  D; x1 þ DÞ, the position of its center
ment spaces do not own the capability to describe such x1 being arbitrary. In the damage zone deprived of its cen-
solution. Nevertheless and quite surprisingly this kind of ter, i.e., in I n fx1 g, the damage field is given by the first
response is picked up from the simulations. The reason lies integral (34) which reads here
on how plasticity is implemented. Although plasticity is !
defined over discrete points the contribution of any point 2 0 2 h2 r2 1
‘ a ðxÞ ¼ aðxÞ þ 1
is spread over the entire finite element of size h. That is, 2r 2P ð1  aðxÞÞ2
the numerical approximation has the effect to regularize 1  h2

the displacement and plastic field while the mesh size h  1  ð1  aðxÞÞ2 :
2
plays the role of a convergence parameter. In the following
the responses are compared for meshes of different size. This first integral gives also the width D of the damage
As the energy functional is separately convex in each zone which depends only on r. At the center x1 ; a0 and u
variable, it seems reasonable to adopt an alternate minimi- are discontinuous, a cohesive crack appears and the cohe-
zation algorithm, respectively in the variables u; p and a.
Hence at a given time step the solution is simply found 1
<http://www.mcs.anl.gov/research/projects/tao/>.
 R. Alessi et al. / Mechanics of Materials 80 (2015) 351–367 363

sive law relating sut to r is obtained by using the plastic curs and the final cohesive phase where despite the exis-
yield criterion and the first integral above. Specifically, tence of sut still the bar sustains a non-vanishing stress
one gets (Fig. 8) r; the analytic curve is actually the plot of cohesive law
rffiffiffiffiffiffi (35). For U within the final cohesive phase, we compare
r the responses in terms of damage aðxÞ and plastic strain
1  aðx1 Þ ¼ ; sutðx1 Þ
r P pðxÞ profiles in Fig. 9a and b, respectively.
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
!ffi
u rffiffiffiffiffiffi In all these cases, the numeric solutions monotonically
2‘eD ur 1 þ h 2
r r h 2 2
r
¼ 2 t
P
þ þ  : converge towards the analytic solution as the mesh size
h r 2 r P 2r P 2r 2P
decreases. In particular, comparing Fig. 9b and c, one can
appreciate a rather surprising mesh independence of the
Finally, the stress-displacement relation r-U is obtained by
plastic localization. Indeed, for a sequence of decreasing
using the stress–strain relation, the boundary condition
mesh sizes, the numerical approximation mimics the sin-
and the first integral of the damage criterion,
gular part of the plastic response (as discussed before a
Z L
r dx Dirac delta in x ¼ x1 ) by a sequence of increasingly local-
U ¼ sutðx1 Þ þ :
Y0 0 ð1  aðxÞÞ2 ized profiles with monotonically converging subtended
areas.
Note that the graph r-U presents a snap-back at U c when
the ratio ‘=L is small enough. Therefore, if one prescribes
5.3. Two-dimensional traction test
a monotonic increasing displacement U, then the evolution
is necessarily discontinuous at U c and the stress jumps
This simulation has the purpose to highlight the effec-
from r P to a smaller value which depends on L.
tiveness of both the model and the numeric implementa-
In Fig. 9 the analytic solution is compared to the Finite
tion in spatial dimensions higher than 1. We choose:
Element approximation with different mesh sizes: 20, 50,
100 and 200 elements in X. Thick solid lines are used to X ¼ ð0; LÞ  ð0; HÞ; u1 ðx ¼ 0Þ ¼ 0;
plot the analytic response, while dotted, dotted-dashed, u1 ðx ¼ LÞ ¼ U with U increasing from 0;
dashed and thin solid lines are used to represent the
numerical responses at decreasing mesh size. In Fig. 9c where ðe1 ; e2 Þ is the natural orthonormal base and the
and d the integral of the cumulated plasticity and stress transversal contraction is left free, see Fig. 10a.
are plotted against the applied end-displacement U: it is Plane strain
pffiffiffi conditions are assumedpffiffiffi and the data are
evident the critical value where a displacement jump oc- ‘=L ¼ 0:15 2; H=L ¼ 0:2; m ¼ 0:3; h ¼ 1= 3. Accordingly,

Fig. 9. In 1D comparison of the analytic solution (solid black) with the numeric solutions for the different mesh sizes (20 elem. (dotted); 50 elem. (dot-
dashed), 100 elem. (dashed), 200 elem. (thin)): (a) and (b) represent respectively the damage field a and the additional plastic strain field p  pc at a given
time step after the occurrence of the cohesive crack; (c) and (d) represent respectively the accumulated plastic strain integrated over the domain and the
stress-displacement response.
364 R. Alessi et al. / Mechanics of Materials 80 (2015) 351–367
the deformation remains homogeneous in space during the
elastic stage and the plastic stage. The elastic stage ends
when U ¼ U e and r11 ¼ re with
Ue 1  m2 re 1 tended area in the direction normal to the discontinuity to
¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  1:024; ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  1:125:
eP L 1  m þ m2 r P 1  m þ m2
The plastic stage ends when the damage criterion is
reached everywhere in the specimen. At this moment,
the values U c ; rc11 and p
c of the displacement, the normal
stress component and the cumulated plastic strain can be
obtained in a closed form, but we give here their numerical
values only
U c  1:834 eP L; rc11  1:149 r P ; pc  0:889 eP :
All these theoretical values are well captured by the alter-
nate minimization algorithm, see Fig. 10d. After this criti-
cal time, the solution can no more be obtained in a
closed form and we will refer to the numerical results only.
The numerical computations show that, as in the 1D case,
the damage localizes in a strip whose width is of the order
of the internal length ‘ and a cohesive crack nucleates at its
center. The phenomenon is still brutal in the sense that the
evolution is discontinuous at this time and the stress
drops. Specifically, in Fig. 10d the global response of the
specimen is reported; namely the stress component r11 ,
averaged on the boundary x ¼ L, is plotted against the
end displacement U. One can see the sudden drop of the
stress value when U  1:7 eP L which is accompanied by
the formation of a localized plastic field (actually a shear
band) in the center of the specimen. For a greater value
of the end displacement U, we have plotted in Fig. 10b
and c the damage field a and the cumulated plastic strain
field p, respectively. As to be expected, the plastic localiza-
tion describes a shear band which is inclined with respect
the axis of traction (the inclination can depend on the
width H of the specimen); this zone is centered within
the support of the damage profile. However, whilst the size
Fig. 10. Numerical results for the 2D traction test: (a) the reference configuration; (b) the damage field plotted on the deformed shape when U ¼ 2eP L (blue,
a ¼ 0; red, a ¼ 1); (c) the accumulated plastic strain field plotted on the deformed shape when U ¼ 2eP L (blue, p ¼ min; red, p ¼ max); (d) the average
normal stress at x1 ¼ L versus the average prescribed stretching. (For interpretation of the references to colour in this figure legend, the reader is referred to
the web version of this article.)
of the damaged zone can be constitutively controlled,
being proportional to the internal characteristic length ‘,
the size of the plastic strain localization is inessential and
dictated by the mesh size. As in Fig. 9b, it is the actual sub-
have a true physical meaning.
6. Perspectives
Let us finish now by some perspectives.
1. We have only considered in the present paper the case
where the plasticity criterion is Von Mises criterion. A
first natural extension will consist in generalizing our
approach to arbitrary convex set.
2. We have chosen here a form of the total energy which is
the simplest one to couple damage with plasticity. More
complex coupling based on phenomenological consid-
erations or physical mechanisms should deserve to be
analyzed.
3. Our approach leads to a competition between the dam-
age criterion and the plastic yield criterion. The latter is
unbounded in the direction of hydrostatic pressure
while the former one is bounded in every stress direc-
tion. Assuming that r P < r
 D , the plastic criterion is
reached before the damage criterion in the uniaxial
traction problem and consequently the fracture occurs
along a shear band where the plasticity is concentrated
and which forces the jump discontinuity of the dis-
placement to be tangential. In more complex situations,
it could happen that the damage criterion be reached at
a point without any preliminary plasticization (because
of a triaxiality effect) and hence that a non cohesive
crack nucleates at this point as in the quasi-brittle case.
The highlighting of such phenomena will be the goal of
future works.
4. Accordingly, the numerical method will be tested on
more complex geometry and loading.
 R. Alessi et al. / Mechanics of Materials 80 (2015) 351–367 365

Appendix A. Construction of the evolution problem 2. Plasticity yield criterion in the regular part of the domain.
from the irreversibility condition, the stability principle Taking v ¼ 0; b ¼ 0 and hence qS ¼ 0, (38) gives
and the energy balance Z
ðrP ðat ÞkN kqk  rt  qÞdx P 0;
XnJðnt Þ
A.1. The first order stability conditions
8q smoothð; Tr q ¼ 0Þ: ð41Þ
Dividing (12) by h and passing to the limit when h ! 0
In the multi-dimensional case, since Tr q ¼ 0, one gets
yields the first order stability conditions:
rt  q ¼ rDt  q where rDt denotes the deviatoric part of rt

d and one deduces from (41) that
t þ hkN kqkÞ
E t ðut þ hv ; at þ hb; pt þ hq; p
dh h¼0
rDt  q 6 kN rP ðat Þ; 8q 2 MNs : Tr q ¼ 0; kqk ¼ 1:
P 0; 8ðv ; b; qÞ admissible:
ð42Þ
Using the definition (9) of the energy and the assumed
This inequality must hold everywhere in X n Jðnt Þ and
forms of the field leads to
Z Z hence the stress must satisfy:
06 rt  ðrs v  qR Þ dx þ rP ðat ÞkN kqR k dx krDt k 6 kN rP ðat Þ in X n Jðnt Þ: ð43Þ
XnðJðnt Þ[Jðv ÞÞ XnJðv Þ
Z
þ rP ðat ÞjN ksv tk dS Thus (43) is actually the standard Von Mises plastic yield
Jðv Þ criterion obtained as a stability condition associated with
Z  
1 the plastic incompressibility hypothesis and the form of
þ d0 ðat Þ  C0 ðat Þrt  rt b dx
2 the plastic dissipated energy.
ZX Z In the one-dimensional case, one simply deduces from (41)
þ 2d1 ‘2 rat  rb dx þ r0P ðat ÞpRt b dx that jrt j 6 rP ðat Þ in X n Jðnt Þ. But since rt is a constant and
XnJðnt Þ XnJðnt Þ
Z Z at is continuous in space, this inequality must hold
þ r0P ðat ÞPt b dS  Ft  v dS; ð38Þ everywhere.
Jðnt Þ @F X
3. Plasticity yield criterion on the jump set Jðnt Þ. Taking
where the spatial dependence has been dropped and b ¼ 0; qR ¼ 0; v such that Jðv Þ – ; and using (40) and
(38) gives
rt ¼ Eðat Þðrs ut  pt Þ Z
denotes the stress field at time t. The inequality (38) must 06 ðrP ðat ÞjN ksv tk  rt n  sv tÞ dS: ð44Þ
Jðv Þ
hold for all v such that v ¼ 0 on @ D X (and sv t  n ¼ 0 on
Jðv Þ), all b  0 and all q (deviatoric) which satisfies (11). In the one-dimensional case, since Jðv Þ and sv t can be cho-
Let us derive the different local conditions which are given sen arbitrarily and since jN ¼ 1, one simply re-obtains that
by (38). jrt j 6 rP ðat Þ must hold everywhere in X.
1. Equilibrium equations. Taking first b ¼ 0; q ¼ 0 and In the multidimensional case, the jump of v is restricted by
hence Jðv Þ ¼ ;, integrating by parts the gradient of v the plastic incompressibility condition, sv t  n ¼ 0 on Jðv Þ.
term over X n Jðnt Þ and considering that the boundary Therefore rt n  sv t involves the shear stress only and can
@ðX n JÞ ¼ @ X [ J þ [ J  involves both faces (with oppo- read as
site normals) of the given surface J, one gets:
Z Z rt n  svt ¼ ðrt n  ðrt n  nÞ nÞ  sv t:
rt  rs v dx ¼ rt n  v dx Considering first v such that Jðv Þ \ Jðnt Þ ¼ ; and using the
XnJðnt Þ @ðXnJðnt ÞÞ
Z fact that in such a case the direction n can be chosen arbi-
 ðr  rt Þ  v dx trarily, (44) gives the following condition which must hold
XnJðnt Þ in X n JðnÞ:
Z Z
¼ rt n  v dx  srt tn  v dx ðrt n  ðrt n  nÞ nÞ  t 6 jN rP ðat Þ;
@F X Jðnt Þ
Z
8n; t 2 RN : knk ¼ ktk ¼ 1; n  t ¼ 0:
 ðr  rt Þ  v dx: ð39Þ
XnJðnt Þ It is easy to see that this condition is equivalent to
Thus by standard localization arguments one obtains from max krt n  ðrt n  nÞ nk 6 jN rP ðat Þ in X n Jðnt Þ:
(38) and (39) the classical equilibrium equations and the n: knk¼1

natural boundary conditions: ð45Þ

r  rt ¼ 0 in X n Jðnt Þ; srt tn So, we obtain a maximal shear stress condition which dif-
¼ 0 on Jðnt Þ; rt n ¼ Ft on @ F X: ð40Þ fers from Von Mises plastic yield criterion. In fact this max-
imal shear stress condition (45) is weaker than Von Mises
Note that the vector stress rt n must be continuous on the plastic yield criterion (43) in the sense that (45) is auto-
jump set Jðnt Þ but this is not necessarily true for the other matically satisfied when (43) holds, but the converse is
components of the stress tensor. It could happen that rt is not true. (The proof of this property is left to the reader.)
discontinuous across Jðnt Þ and hence not defined on Jðnt Þ. Accordingly we can disregard (45).
366 R. Alessi et al. / Mechanics of Materials 80 (2015) 351–367

Z t
If we consider now v such that Jðv Þ  Jðnt Þ, then the normal
Rt ðxÞ ¼
p kN kp_ Rs ðxÞkds; 8x 2 X n Jðnt Þ and
vector n is fixed by nt and (44) gives the following condi- 0
tion on Jðnt Þ: Z t
 t ðxÞ ¼
P jN ksu_ s tðxÞkds; 8x 2 Jðnt Þ: ð52Þ
ðrt n  ðrt n  nÞ nÞ  t 6 jN rP ðat Þ; 8t 2 RN 0

: ktk ¼ 1; t  n ¼ 0; ð46Þ Let us now use the energy balance (13). Expanding the
time derivative of the total energy, (13) becomes
which is equivalent to Z
 
krt n  ðrt n  nÞ nk 6 jN rP ðat Þ on Jðnt Þ: ð47Þ 0¼ rt  ðrs u_ t  p_ Rt Þ þ kN rP ðat Þkp_ Rt k dx
XnJðnt Þ
Z Z
So, the plasticity yield criterion on the jump set is formu-
 rt n  U_ t dS Ft  u_ t dS
lated in terms of the norm of the shear stress vector, what @D X @F X
has a sense since the stress vector is well defined on this Z 
1
surface by virtue of (40). Of course, when all the stress ten- þ Rt a_ t
 C0 ðat Þrt  rt a_ t þ d0 ðat Þa_ t þ r0P ðat Þp
XnJðnt Þ 2
sor field is continuous and hence well defined on Jðnt Þ, then 
Von Mises yield criterion must hold and since it contains þ2d1 ‘2 rat  ra_ t dx
Z
the shear stress condition this latter condition can be dis-  0 
þ rP ðat ÞPt a_ t þ jN rP ðat Þksu_ t tk dS:
regarded. In the doubt we keep (47). Jðnt Þ
4. The damage yield criteria. Taking v ¼ 0 and q ¼ 0, inte-
Integrating by parts the terms in rs u_ t and ra_ t above,
grating by parts the term in rat  rb, we obtain by
using the equilibrium equations (40) and the boundary
classical arguments of Calculus of Variations the fol- _ t on @ D X, the energy balance leads to
conditions u_ t ¼ U
lowing damage yield criteria in the regular part, the
the following equality
singular part and the boundary @ X:
Z
 
1 0¼ kN rP ðat Þkp_ Rt k  rt  p_ Rt dx
Rt  C0 ðat Þrt  rt  2d1 ‘2 Dat P 0
d0 ðat Þ þ r0P ðat Þp XnJðnt Þ
Z
2
in X n Jðnt Þ; ð48Þ þ ðjN rP ðat Þksu_ t tk  rt n  su_ t tÞdS
Jðnt Þ
Z  
1
þ Rt  C0 ðat Þrt  rt  2d1 ‘2 Dat a_ t dx
d0 ðat Þ þ r0P ðat Þp
r0P ðat ÞPt  2d1 ‘2 s@ at =@nt  0 on Jðnt Þ; ð49Þ XnJðnt Þ 2
@ at =@n  0 on @ X::
In the case where rP does not depend on a, we recover the
damage yield criteria obtained in Pham and Marigo
(2010b) and Pham et al. (2011a,b) by the same variational
approach and in Comi (1999). But note that here, because
of the coupling term between damage and plasticity, the
localization of the plastic strain on a surface will in general
induce a discontinuity of the damage normal derivative
and vice versa.
Z   Z
@ a @a
ð50Þ
þ r0P ðat ÞPt  2d1 ‘2 s t a_ t dS þ 2d1 ‘2 t a_ t dS:
t
Jðnt Þ @n @X @n
1. The consistency equations. By virtue of the irrevers-
ibility condition (10) and the local first order stabil-
ity conditions (43) and (47)–(50), each of the five
integrands above is non negative and hence each
must vanish so that their sum be zero. Therefore,
we have obtained the three following consistency
equations for the damage evolution:
A.2. The plastic flow rules and the consistency equations as  
1
consequences of the energy balance Rt  C0 ðat Þrt  rt  2d1 ‘2 Dat a_ t
d0 ðat Þ þ r0P ðat Þp
2
t is
Assuming that the evolution t # nt is smooth, t # p ¼ 0 in X n Jðnt Þ;
obtained from t # pt by ð53Þ
_ Rt ðxÞ ¼ kN kp_ Rt ðxÞk 8x 2 X n Jðnt Þ;
p  
r0P ðat ÞPt  2d1 ‘2 s@ at =@nt a_ t ¼ 0 on Jðnt Þ; ð54Þ
P_ ðxÞ ¼ j ksu_ tðxÞk 8x 2 Jðn Þ;
t N t t
ð@ at =@nÞ a_ t ¼ 0 on @ X: ð55Þ
where
2. The plasticity flow rules. In the multi-dimensional
 case, the vanishing of the first integrand gives
1 if N ¼ 1
jN ¼ pffiffiffi : ð51Þ
kN = 2 otherwise rDt  p_ Rt ¼ kN rP ðat Þkp_ Rt k in X n Jðnt Þ:
Note that t # Jðnt Þ is not decreasing, i.e., the number of sin- But, by virtue of (42) and since Tr p_ Rt ¼ 0, the equality
gular points can only increase. Indeed, if a jump disconti- above can hold if and only if the normality flow rule for
nuity of the displacement appears at a point xs at some the plastic strain rate holds, i.e.,
 t ðxs Þ > 0 for all t P t s . Therefore those points
time t s , then P rDt
are material points and their position does not depend on p_ Rt ¼ kp_ Rt k in X n Jðnt Þ: ð56Þ
kN rP ðat Þ
time, but their number can increase because new points
can appear all along the evolution. Accordingly, using the In the same way, one deduces from the vanishing of the
initial conditions, we can set second integrand and su_ t t  n ¼ 0 that
 R. Alessi et al. / Mechanics of Materials 80 (2015) 351–367
ðrt n  ðrt n  nÞ nÞ  su_ t t ¼ jN rP ðat Þksu_ t tk on Jðnt Þ
and hence, by virtue of (46), another normality plastic flow
rule on the jump set
rt n  ðrt n  nÞ n
su_ t t ¼ ksu_ t tk on Jðnt Þ: ð57Þ
jN rP ðat Þ
Note that this latter flow rule is rarely mentioned in the lit-
erature and the interest reader can refer to Francfort and
Giacomini (2012) where this flow rule is obtained at each
interface of an heterogeneous elastic–plastic body.
In the one-dimensional case, these flow rules read as
rt 26.
p_ Rt ¼
p_ Rt
in X n Jðnt Þ; Mielke, A., 2005. Evolution of rate-independent systems. Handbook of
rP ðat Þ
rt
su_ t t ¼ jsu_ t tj on Jðnt Þ: ð58Þ
rP ðat Þ
References
Ambrosio, L., Lemenant, A., Royer-Carfagni, G., 2012. A variational model
for plastic slip and its regularization via C-convergence. J. Elast., 1–35.
Amor, H., Marigo, J.-J., Maurini, C., 2009. Variational approach to brittle
fracture with unilateral contact: numerical experiments. J. Mech.
Phys. Solids 57 (8), 1209–1229.
Belnoue, J.P., Nguyen, G.D., Korsunsky, M., 2007. A one-dimensional non
local damage–plasticity model for ductile materials. Int. J. Fract. 144,
53–60.
Benallal, A., Marigo, J.-J., 2007. Bifurcation and stability issues in gradient
theories with softening. Modell. Simul. Mater. Sci. Eng. 15 (1), S283–
S295.
Bourdin, B., Francfort, G.A., Marigo, J.-J., 2000. Numerical experiments in
revisited brittle fracture. J. Mech. Phys. Solids 48 (4), 797–826.
Bourdin, B., Francfort, G.A., Marigo, J.-J., 2008. The variational approach to
fracture. J. Elast. 91 (1–3), 5–148.
Comi, C., 1999. Computational modelling of gradient-enhanced damage in
quasi-brittle materials. Mech. Cohesive-Frictional Mater. 4 (1), 17–36.
Comi, C., Mariani, S., Negri, M., Perego, U., 2006. A one-dimensional
variational formulation for quasi-brittle fracture. J. Mech. Mater.
Struct. 1 (8), 1323–1343.
Dal Corso, F., Willis, J., 2011. Stability of strain-gradient plastic materials.
J. Mech. Phys. Solids 59 (6), 1251–1267.
de Borst, R., Pamin, J., Geers, M.G.D., 1999. On coupled gradient-
dependent plasticity and damage theories with a view to
localization. Eur. J. Mech. A 18, 939–962.
Del Piero, G., 2013. A variational approach to fracture and other inelastic
phenomena. J. Elast. 112 (1), 3–77.
Del Piero, G., Truskinovsky, L., 2009. Elastic bars with cohesive energy.
Continuum Mech. Thermodyn. 21, 141–171.
367
Del Piero, G., Lancioni, G., March, R., 2012. Diffuse cohesive energy in
plasticity and fracture. Tech. Mech. 32 (2), 174–188.
Dimitrijevic, B.J., Hackl, K., 2011. A regularization framework for damage–
plasticity models via gradient enhancement of the free energy. Int. J.
Numer. Methods Biomed. Eng. 27, 1199–1210.
Francfort, G.A., Giacomini, A., 2012. Small-strain heterogeneous
elastoplasticity revisited. Commun. Pure Appl. Math. 65, 1185–1241.
Grassl, P., Jirásek, M., 2006. Plastic model with non-local damage applied
to concrete. Int. J. Numer. Anal. Methods Geomech. 30, 71–90.
Lemaitre, J., Chaboche, J., 1985. Mécanique des Matériaux Solides. Bordas.
Logg, A., Mardal, K.-A., Wells, G.N., 2012. Automated Solution of
Differential Equations by the Finite Element Method. Springer.
Lorentz, E., Andrieux, S., 2003. Analysis of non-local models through
energetic formulations. Int. J. Solids Struct. 40, 2905–2936.
Lorentz, E., Cuvilliez, S., Kazymyrenko, K., 2011. Convergence of a gradient
damage model toward a cohesive zone model. C. R. Méc. 339 (1), 20–
Differential Equations: Evolutionary Equations, vol. 2. North-Holland,
pp. 461–559.
Mielke, A., 2006. A mathematical framework for generalized standard
materials in the rate-independent case. In: Helmig, R., Mielke, A.,
Wohlmuth, B. (Eds.), Multifield Problems in Solid and Fluid
Mechanics, Lecture Notes in Applied and Computational Mechanics,
vol. 28. Springer, Berlin, Heidelberg, pp. 399–428.
Peerlings, R., de Borst, R., Brekelmans, W., Geers, M., 1998. Gradient-
enhanced damage modelling of concrete fracture. Mech. Cohesive-
Frictional Mater. 3, 323–342.
Pham, K., Marigo, J.-J., 2010a. Approche variationnelle de
l’endommagement: I. Les concepts fondamentaux. C. R. Méc. 338
(4), 191–198.
Pham, K., Marigo, J.-J., 2010b. Approche variationnelle de
l’endommagement II. Les modèles à gradient. C. R. Méc. 338 (4),
199–206.
Pham, K., Marigo, J.-J., 2013a. From the onset of damage until the rupture:
construction of the responses with damage localization for a general
class of gradient damage models. Continuum Mech. Thermodyn. 25
(2–4), 147–171.
Pham, K., Marigo, J.-J., 2013b. Stability of homogeneous states with
gradient damage models: size effects and shape effects in the three-
dimensional setting. J. Elast. 110 (1), 63–93.
Pham, K., Amor, H., Marigo, J.-J., Maurini, C., 2011a. Gradient damage
models and their use to approximate brittle fracture. Int. J. Damage
Mech. 20 (4), 618–652.
Pham, K., Marigo, J.-J., Maurini, C., 2011b. The issues of the uniqueness
and the stability of the homogeneous response in uniaxial tests with
gradient damage models. J. Mech. Phys. Solids 59 (6), 1163–1190.
Sicsic, P., Marigo, J.-J., 2013. From gradient damage laws to Griffith’s
theory of crack propagation. J. Elast. 113 (1), 55–74.
Sicsic, P., Marigo, J.-J., Maurini, C., 2014. Initiation of a periodic array of
cracks in the thermal shock problem: a gradient damage modeling. J.
Mech. Phys. Solids 63, 256–284 http://dx.doi.org/10.1016/
j.jmps.2013.09.003.