Eur. J. Mech.

A/Solids 19 (2000) 187–208

 2000 Éditions scientifiques et médicales Elsevier SAS. All rights reserved
S0997-7538(00)00161-3/FLA

Anisotropic damage law of evolution

Jean Lemaitre a, *, Rodrigue Desmorat b , Maxime Sauzay a,c

a Laboratoire de mécanique et technologie, 61, av. du président Wilson, 94235 Cachan cedex, France
b Laboratoire de modélisation et mécanique des structures, 8, rue du Capitaine Scott, 75015 Paris, France
c Laboratoire d’elaboration et de développement des produits plats, USINOR, 17, av. des tilleuls, 57190 Florange, France

(Received 19 August 1999; revised and accepted 2 November 1999)

Abstract – A formulation for anisotropic damage is established in the framework of the principle of strain equivalence. The damage variable is still
related to the surface density of microcracks and microvoids and, as its evolution is governed by the plastic strain, it is represented by a second order
tensor and is orthotropic. The coupling of damage with elasticity is written through a tensor on the deviatoric part of the energy and through a scalar
taken as its trace on the hydrostatic part. The kinetic law of damage evolution is an extension of the isotropic case. Here, the principal components of
the damage rate tensor are proportional to the absolute value of principal components of the plastic strain rate tensor and are a nonlinear function of the
effective elastic strain energy. The proposed damage evolution law does not introduce any other material parameter. Several series of experiments on
metals give a good validation of this theory. The coupling of damage with plasticity and the quasi-unilateral conditions of partial closure of microcracks
naturally derive from the concept of effective stress. Finally, a study of strain localization makes it possible to determine the critical value of the damage
at mesocrack initiation.  2000 Éditions scientifiques et médicales Elsevier SAS
damage / anisotropy thermodinamique / effective stress / constitutive equations

1. Introduction

The formulation of anisotropic damage is already a long story with several not fully satisfactory theories.
The present one also has a price to pay in the introduction of an extra-parameter η although it disappears in the
final constitutive equation.
The extension of isotropic damage theory to anisotropy is not a straight forward task in the coupling between
elasticity and damage. In the case of isotropic damage represented by the scalar variable D, the effective stress
concept associated to the principle of strain equivalence, σ̃ij = σij /(1 − D) = Eij kl εkl
e
for elasticity solves the
problem (Lemaitre, 1992). In the case of general anisotropy, the damage variable is represented by a fourth
order tensor (Kracinovic, 1985; Leckie and Onat, 1981; Lemaitre and Chaboche, 1985). Here we consider a
second order tensor D (Murakami and Ohno, 1978; Murakami, 1988) which corresponds to orthotropy justified
by the mechanism of damage essentially related to plasticity represented by the second order plastic strain
tensor ε p . In these conditions the direct generalization of the concept of effective stress and the principle of
strain equivalence generally does not yield existence of an elastic potential (Cordebois and Sidoroff, 1982).
Using the concept of effective stress only on the principal stress components (Hayhurst and Leckie, 1973)
does not solve the problem. Replacing the strain equivalence principle by an energy equivalence (Cordebois
and Sidoroff, 1982; Cordebois, 1983) is a possible solution which has been followed by several authors (Chow
and Wang, 1987; Ju, 1989). It yields existence of the elastic potential but the physical definition of the damage
variable is lost. The damage is no longer directly related to the surface density of defects but is a mathematical

* Correspondence and reprints. E-mail: Jean.Lemaitre@lmt.ens-cachan.fr

188 J. Lemaitre et al.

variable defined by its coupling with elasticity. Then, no more rules exist to describe the coupling with plasticity
or the conditions of partial closure of microcracks (quasi-unilateral effect) (Horii and Nemat-Nasser, 1983;
Ladevèze, 1983; Ladevèze and Lemaitre, 1984; Liénard, 1989; Mazars et al., 1990; Chaboche, 1993; Dragon
and Halm, 1995) nor to define the critical value of the damage at mesocrack initiation. Micromechanics has
also been used (Kachanov, 1993; Dragon et al., 1994; Fichant, 1996) but restricted to brittle failure it does not
yield a closed form constitutive equation.

2. Elastic potential

This problem of incompatibility between physics and thermodynamics is avoided if two damage variables
corresponding to two mechanisms are introduced as in composites for delamination and matrix cracking
(Ladevèze, 1983). The Gibbs elastic potential is split into its deviatoric part affected by a tensorial damage
variable D and its hydrostatic part affected by another (scalar) damage variable dH .
For monolithic materials subjected to one damage mechanism, only one damage variable is considered, the
second order tensor D. The same decomposition of the Gibbs energy may be kept if the coupling of the damage
with the hydrostatic part of the energy is written through a function of the mean damage DH , the trace of D

1
dH = g DH , DH = Dkk , (1)
3

E and ν being the Young’s modulus and the Poisson’s ratio of the virgin material, σ D being the stress deviator
and σH the hydrostatic stress
1
σijD = σij − σH δij , σH = σkk (2)
3
the Gibbs energy for linear and initially isotropic elasticity is

1+ν 3(1 − 2ν) σH2

ρψ ∗ = Hij σjDk Hkl σliD + , (3)
2E 2E 1 − dH

where
H = (1 − D)−1/2 . (4)

A procedure to identify the function g is given in the next section and a good approximation consists in
taking
dH = ηDH , (5)
η is a necessary parameter for a correct representation of experiments concerning the variations in the Poisson’s
ratio with damage. It depends upon materials and characterizes their sensibility to hydrostatic stress. It is one
more parameter to identify. The particular value η = 1 corresponds to isotropic damage, but often for metals
η ≈ 3. Furthermore, it does not appear in the damage law of evolution proposed in Section 4.
The law of elasticity derives from the Gibbs potential

∂ψ ∗ 1 + ν ν
εije = ρ = σ̃ij − σ̃kk δij . (6)
∂σij E E
 Anisotropic damage law of evolution 189

It naturally introduces a symmetric effective stress tensor which does not depend upon the elasticity parameters
and which can be used for the coupling with plasticity (Section 6)

D σH
σ̃ij = Hik σklD Hlj + δij . (7)
1 − ηDH

The associated variable to D is the strain energy density release rate tensor Y which also derives from the
Gibbs energy, Yij = ρ∂ψ ∗ /∂Dij . The final derivative needs some care owing to the possible variation of the
principal direction of the damage. With H dH + dH H = H 2 dD H 2 written
1
Aij kl Ḣkl = Hip2 Ḋpq Hqj2 , Aij kl = (Hik δj l + Hj l δik + Hil δj k + Hj k δil ) (8)
2
1+ν D η(1 − 2ν) σH2
Yij = σkp Hpq σqlD A−1
klmn Hmi Hj n +
2 2
δij . (9)
E 2E (1 − ηDH )2
By chance the law of damage evolution will not be a function of Y . Nevertheless it is possible to verify that the
dissipation Yij Ḋij is positive or zero at least for all practical cases.

3. Damage measurements

As for the isotropic case (Lemaitre and Dufailly, 1987) the values of the components of the damage may be
obtained from the changes in elasticity characteristics.

3.1. Uniaxial tension (TU)

x1 , xE2 , xE3 ) of figure 1

Consider a representative volume element (RVE) damaged in the orthotropic frame (E
 
D1 0 0
 
D=
 0 D2 0
. (10)
0 0 D3

The elastic strains in this frame are

   
ε1e 0 0 1 0 0
  1 − 2ν σ1  
 0 0 0 1 0
 ε2e = 3E 1 − ηDH  
0 0 ε3e 0 0 1
1   1 D
√ 0 0 2σ1 √ 0 0
 1 − D1  0 0 
  1 − D1 
  3  
1+ν 
 1 
  0 −σ1 0  1 
 . (11)
+ 0 √ 0  0 √ 0
E 

 1 − D2 
 3 

1 − D2 

 1  −σ1 1 
0 0 √ 0 0 0 0 √
1 − D3 3 1 − D3
Defining the damaged elasticity modulus in direction 1 and the associated contraction ratios by
σ1 ε2e ε3e
Ee1 = , ν̃12 = − , ν̃13 = − . (12)
ε1e ε1e ε1e
190 J. Lemaitre et al.

Figure 1. Damaged elasticity in the orthotropic frame.

Then
 
E 1+ν 4 1 1 1 − 2ν
= + + + , (13)
Ee1 9 1 − D1 1 − D2 1 − D3 3(1 − ηDH )
 
E 1+ν 2 2 1 1 − 2ν
ν̃12 = + − − , (14)
Ee1 9 1 − D1 1 − D2 1 − D3 3(1 − ηDH )
 
E 1+ν 2 1 2 1 − 2ν
ν̃13 = − + − . (15)
Ee1 9 1 − D1 1 − D2 1 − D3 3(1 − ηDH )
The same operation for directions 2 and 3 gives nine equations to determine the three components of the damage
D1 , D2 , D3 and the coefficient η.
Practically, most of the time, it is not possible to obtain three specimens in three orthogonal directions but
four measures are sufficient if the orthotropic directions are known. In the case of a damaged plane sheet, the
only possible directions for measures are xE1 and xE2 to give Ee1 , Ee2 and ν̃12 then
 −1
Ee1 Ee1
D1 = 1 − (1 + ν) 2 + ν̃12 − , (16)
E Ee2
 −1
Ee2 Ee2
D2 = 1 − (1 + ν) 2 − (1 − ν̃12 ) , (17)
E Ee1
Ee1 1 − 2ν
ηDH = 1 − . (18)
E 1 − 2ν̃12
If tension is applied in direction 1, D1 and D2 are determined by equations (16) and (17), D3 = D2 for a
material initially isotropic and DH = (D1 + 2D2 )/3. Then, η is obtained from equation (18). Examples are
given in Section 5.

3.2. Plane tension (TP)

In the case of plane sheets, complementary tests may be performed on large specimens loaded in tension
ensuring a plane strain state in the middle of each sheet. This induces damage under a different stress triaxiality.
 Anisotropic damage law of evolution 191

If xE1 is the direction of tension, xE2 the direction in the width and xE3 the direction in the thickness, the strain ε22
vanishes owing to the plane strain condition. Furthermore, if during loading elastic strains are neglected then
p
ε22 ≈ 0 and
 p 
ε11 0 0
 
εp = 
 0 0 0 
. (19)
p
0 0 −ε11
During loading and unloading the states of stress are, respectively,
   
σ 0 0 σ1 0 0
 σ   

σ = 0 0
 and σ = 
 0 σ2 0  . (20)
 2 
0 0 0 0 0 0

The plane strain condition is then ε̇22 = ε̇22

e
= 0. The elastic strain is calculated from the law of elasticity (6)
with a diagonal damage matrix D = diag[D1 , D2 , D3 ]. Avoiding the calculations, let us just recall that
   
σ1 4 1 1 σ2 4 2 1
D
σ̃11 = + + − − + , (21)
9 1 − D1 1 − D2 1 − D3 9 1 − D1 1 − D2 1 − D3
   
σ1 2 4 1 σ2 1 4 1
D
σ̃22 =− − + + + + + , (22)
9 1 − D1 1 − D2 1 − D3 9 1 − D1 1 − D2 1 − D3
   
σ1 1 2 2 σ2 1 2 2
D
σ̃33 =− − + + − − + + , (23)
9 1 − D1 1 − D2 1 − D3 9 1 − D1 1 − D2 1 − D3
σ1 + σ2
σ̃H = . (24)
3(1 − dH )
For the unloading, the relation between σ2 and σ1 as well as the determination of the damaged modulus are still
obtained by consideration of the plane strain condition.
In order to measure the damage tensor, small uniaxial tension specimens are then cut in the homogeneous
deformation zones of the large sheets to obtain Ee1 , Ee2 , ν̃12 from which D1 , D2 , η (and D3 ) are derived from
equations (16)–(18).

4. Kinetic law of damage evolution

It is possible to formulate a law which generalizes the isotropic damage law of evolution (Lemaitre, 1992) in
the framework of generalized standard materials (Halphen and Nguyen, 1975; Germain et al., 1983)
 s
Y
Ḋ = ṗ, (25)
S

where Y = 12 Eij kl εkl

e e
εij is the strain energy release rate density. For the anisotropic case Y is also the trace of Y
p p
(equation (9)) when Dij = 0 and η = 1. ṗ = ( 23 ε̇ij ε̇ij )1/2 is the accumulated plastic strain rate, S and s are two
parameters characteristic of each material. This law considers the damage governed by the elastic energy and
the plastic strain as shown by many observations (Lemaitre et al., 1998).
192 J. Lemaitre et al.

4.1. Damage evolution

In order to obtain the above isotropic law as a particular case of the anisotropic law the simplest choice of
the dissipative potential is
 s  p
Y (ε e )  dε 
F = f + FD = f + Yij   , (26)
S dr 
ij

where |·| applied to a tensor means the absolute value of the principal values, r is the internal variable associated
to the isotropic hardening R, X is the kinematic hardening back stress variable

∂ψ ∗ (σ , D, R, X)
r = −ρ . (27)
∂R
R
Y (ε e ) is also the effective strain energy σ̃ij dεije which can be written as a function of the effective stress

2
1 1 σ̃eq Rν
Y = Eij kl εkl
e e
εij = σ̃ij εije = . (28)
2 2 2E
This introduces the triaxiality function as
 2
2 σ̃H
Rν = (1 + ν) + 3(1 − 2ν) (29)
3 σ̃eq

with
σH
σ̃eq = (H σ D H )eq , σ̃H = . (30)
1 − ηDH
f is the von Mises plastic loading function in which σy is the yield stress

f = (σ̃ − X)eq − R − σy . (31)

The law of plasticity coupled to damage derives from the potential F or f if linear kinematic hardening is
assumed (if not refer to Section 6)

p ∂f 3 [Hik (σ̃klD − Xkl )Hlj ]D

ε̇ij = λ̇ = λ̇ , (32)
∂σij 2 (σ̃ − X)eq
∂f (σ̃ − X)eq
ṙ = −λ̇ = λ̇ = ṗ . (33)
∂R [H (σ̃ D − X)H ]eq
The damage law also derives from the damage potential F or FD
 s  p 
∂F dr Y  dε 
Ḋij = λ̇ =  
∂Yij dt S  dr  . (34)
ij

And finally
 s
Y  p
Ḋij = ε̇  . (35)
ij
S
 Anisotropic damage law of evolution 193

The principal directions of the damage rate coincide with those of the plastic strain rate. Furthermore, the
damage exists only above a threshold written in terms of the accumulated plastic strain

Ḋij = 0 if p 6 pD . (36)

4.2. Stored energy based damage threshold

Several attempts have been made to determine the damage threshold as a function of the loading. The
simplest is to consider pD having for any loading a constant value εpD which is the measured plastic strain
threshold in pure tension. This is a rough approximation valid only for monotonic loadings. In fact this threshold
is much (up to 1 000 times!) larger in cyclic loadings inducing fatigue because it is related to the stored energy
(Lemaitre, 1992) related to the partial irreversibility of the crystalline plastic slips (Tanaka and Mura, 1981). In
order to fit this last statement, let us consider that damage occurs at a given amount φD (material dependent) of
energy stored in the material φs , i.e.

Ḋij = 0 if φs 6 φD , (37)
Z t

φs = (Xij α̇ij + R ṙ) dt, R = R∞ 1 − e−γ0 r . (38)
0

Equation (38) is the standard expression of the stored energy with an exponetial isotropic hardening R. The
parameter R∞ = σu − σy is the saturated value of the hardening and γ0 the nonlinearity exponent. Please note
that as long as there is no damage, the equality r = p stands.
In the further developments we choose to omit the kinematic hardening contribution which is small in fatigue
and which does not change the fact that the classical standard material framework greatly overestimates the
value of the stored energy. Usual standard models lead to an increasing stored energy with the accumulated
plastic strain, when very elegant experiments made on several materials (Chrysochoos, 1987) have shown that
the rate φ̇s becomes negligible at large p. This means that the correct expression of the stored energy should be
Z p
φs = R(p)z(p) dp with z(∞) = 0. (39)
0

Chrysochoos
R (1987) proposes a new set of thermodynamical variables (Q0 , q0 ) from which
φs (p) = 0t Q0 q̇0 dt is proportional to the exponential hardening R
 −b/γ0
−(b+γ0 )p

σy R
q0 = q∞ 1 − e , Q0 = 1− . (40)
q∞ (b + γ0 ) R∞

The expression (but not the value) of the yield criterion is changed, f = σeq − R(Q) − σy which modifies
the plastic constitutive equations. It is possible to avoid this problem by the introduction of new plasticity
variables Q and q such as

Q(q) = R(p), dq = z(p) dp. (41)

As the previous (Q0 , q0 ) they generally differ from the standard couple (R, p) but lead to the more correct
expression of the stored energy (39). The law Q(q) derives from the Gibbs energy corrected in consequence
(q = −ρ∂ψ ∗ /∂Q).
194 J. Lemaitre et al.

Several choices of z(p) are possible (Sermage, 1998) according to several series of tests. We consider the
power law
A (1−m)/m
z(p) =p , with m > 1, (42)
m
where A and m are material dependent parameters. The expression of the stored energy (39) calculated in the
general case and in the uniaxial tension case allows for the determination of the corresponding plastic strain
threshold pD as a function of the threshold in monotonic tension εpD
Z pD
1 1/m
φD = A σeq − σy p (1−m)/m dp = A(σu − σy )εpD , (43)
0 m
where h·i denotes the positive part.
If monotonic or symmetric periodic loading are considered together with perfectly plastic behaviour for the
maximum stress σeq = const = σeq Max
then we obtain the formula
 m
σu − σy
pD = εpD Max − σ
(44)
σeq y

to be used in equation (36). At least a few specific tests are needed to identify the parameter m for each material.

4.3. Crack initiation criterion

Last, a mesocrack is initiated when the damage reaches a critical value Dc which corresponds to the
instability under loading in the plane (of normal n) in which the density of microcracks is maximum. According
to the principle of strain equivalence it is defined by the norm of the damage vector Dij nj or by the larger
principal value of the damage
sup DI = Dc → mesocrack initiation. (45)
I

5. Some verifications

5.1. Uniaxial measurements

Five sets of experiments were used to check that the measured damage obeys the theory and that the η
coefficient may be taken as a constant. They are experiments of ductile damage in uniaxial tension for which
the elasticity modulus Ẽ1 and the contraction coefficient ν̃12 are measured. The materials under consideration
are
– an aluminium alloy 2024 T4 (Cordebois and Sidoroff, 1982);
– a pure copper CUA1 (Cordebois and Sidoroff, 1982);
– a steel XC38 (Nouailhas, 1980);
– an other aluminum alloy 2024 T3 (Chow and Wang, 1987);
– a dispersoid steel SOLDUR 355 studied at LEDEPP (SOLLAC) and LMT Cachan.
In principle two measurements Ee1 and ν̃12 are not sufficient to identify the principal components of the damage
D1 , D2 , D3 and coefficient η but the symmetry with respect to the direction 1 allows us to consider D3 = D2
 Anisotropic damage law of evolution 195

Figure 2. Elasticity modulus and contraction ratio changes for 2024 T4.

Figure 3. Damage evolutions (2024 T4).

and to make the approximation

Ee1
D1 ≈ 1 − . (46)
E
For the extremal values 12 6 ν̃ν 6 1 and 34 6 EEe 6 1, this corresponds to a maximal relative error of 15 % on D1
with respect to equation (46). With this approximation,

Ee1 1+ν
D2 ≈ 1 − . (47)
E 1 + 3ν̃12 − 2ν

The parameter η is determined by the slope of the best straight line fitting the experimental points in the
graph dH fonction of DH (equation (5)). As an example figure 3 shows the evolution of D1 , D2 and the value
of η from experimental data of figure 2 for the aluminium alloy 2024 T4.
For all the cases the following results are obtained:
– 2024 T4: η = 2.6;
196 J. Lemaitre et al.

Figure 4. Checking D1 = 2D2 .

– CUA1: η = 3.5;
– XC 38: η = 2.6;
– 2024 T3: η = 2.1;
– SOLDUR 355: η = 2.8.
Figure 3 and similar results for other materials (figure 4) show that D1 = 2D2 is verified with an accuracy
of the order of magnitude of the scatter which is always large in this kind of experiments (Dufailly, 1995).
D1 = 2D2 is precisely the result given by the damage evolution law as in uniaxial tension ε̇ p = ṗ[1, −1/2,
−1/2] from which |ε̇ p |11 = 2|ε̇ p |22 and from equation (35), D1 = 2D2 .

5.2. Multiaxial measurements

We have performed tension experiments of ductile damage on large plates made of SOLDUR 355 steel (TU
and TP as explained in Section 3.2). These experiments are used to induce the two different plastic strain fields
p p
ε p = p[1, −1/2, −1/2] (TU) and ε p = [ε11 , 0, ε11 ] (TP).
For each tension TU and TP, three small specimens have then been cut in three directions (0, 90 and
45 degrees with respect to the axis of the previous tension, figure 5) of three sheets previously damaged
at three levels of plastic strains: ε11 = 7 10−2 , ε11 = 15 10−2 and ε11 = 33 10−2 , for uniaxial tension,
p p p
p −2 p −2 p −2
ε11 = 5 10 , ε11 = 6.7 10 and ε11 = 11 10 for plane tension. They have then been tested in uniaxial
tension and the elasticity modulus as well as the contraction coefficient have been measured for each of them.
The Young’s moduli of the small specimens coincide of course with the plates moduli Ee1 , Ee2 and E(45).
e
Only specimens cut in the 0 degree direction of the large TU and TP sheets have been used to identify the
constitutive equation of the damage (equation (35)). The damage law thus identified was then applied to all
other tests in order to calculate the components Dij and then Ee2 , E(45),
e ν̃12 to compare with experiments.
The measurements of strains have been made by strain gauges glued axialy and transversaly. The scatter is
important.
 Anisotropic damage law of evolution 197

Figure 5. Small specimens in large specimens.

5.2.1. Identification
The material coefficients to be identified are S, s, Dc for the evolution law (35) and εpD , σu , σy , m for
the threshold (44) (with in uniaxial tension pD = εpD ). We also need the elasticity coefficients E and ν of the
virgin material. One of the main values of the law (35) stands in its identification procedure identical to the
one for isotropic law: the non-isotropic character is given by the plastic strain tensor alone. Furthermore, the
additional parameter η may be ignored as it only becomes influential for large damages.
In order to determine the damage evolution D1 the plastic criterion function f = 0 must be considered
(equation (31)). If the damage occurs upon saturated strain hardening a good approximation of the function
criterion is σ̃eq − σu = 0 where σu is the ultimate stress. Then, from equations (28)–(31) and (35) the evolution
of D is obtained as an implicit function of the plastic strain and the triaxiality
Z D1 Z p σu2
s
Rν−s
p
dD1 = dε11 . (48)
0 pD 2ES

For uniaxial tension the Rν triaxiality function is a function of D1 alone because D2 = D3 = D1 /2 and
DH = 2D1 /3
  −2
2 2η 2 1
RνT U = (1 + ν) + 3(1 − 2ν) 1 − D1 + . (49)
3 3 1 − D1 1 − D21
p
As p = ε11 the relation betweeen D1 and the plastic strain is
 s Z D1
p p 2ES −s
ε11 = ε11 (D1 ) = εpD + RνT U dD1 . (50)
σu2 0

p √
For plane tension p = 2ε11 / 3. The damage law (35) gives D3 = D1 , D2 = 0 and
 2
2 1 − D1
RνT P = (1 + ν) + (1 − 2ν) , (51)
3 1 − 2η
3
D1
198 J. Lemaitre et al.

Figure 6. Damage evolutions (SOLDUR 355).

Figure 7. Determination of parameters S and s.

√  s Z
3 2ES D1
p −s
ε11 = εpD + RνT P dD1 . (52)
2 σu2 0

The results are reported in figure 6 where the damage is calculated by equation (46) and the triaxiality
functions by equations (49)–(51).
Knowing the ultimate stress from the uniaxial tension test, the S and s parameters are deduced from the
dD1
graph Log dε p function of Log(Rν )
11

 s
dD1 σu2
Log p = Log + sLogRν . (53)
dε11 2ES

The corresponding points are reported in figure 7. Due to the scatter and the small variation of Rν , the accuracy
is poor.
Nevertheless identification of the steel SOLDUR 355 is as follows: S = 0.57 MPa, s = 4,
εpD = 2.5 10−2 , Dc = 0.3 with E = 230 GPa, ν = 0.3, σu = 474 MPa, η = 2.8.
 Anisotropic damage law of evolution 199

Figure 8. Comparison theory–experiments for uniaxial tension.

5.2.2. Comparison between theory and experiments

Once the damage law is identified it is possible to calculate the components of the damage for any history
p
of loading. D1 is obtained as a function of ε11 by equations (49)–(52). Then, D2 = D3 = D1 /2 for uniaxial
tension and D2 = 0, D3 = D1 for plane tension. Ee1 (ε11 ), Ee2 (ε11 ), ν̃12 (ε11 ) are calculated to be compared with
p p p

experimental data in figures 8 and 9. Note that the curves Ee1 (ε11 ) for TU and TP only verify the identification.
p

The other curves verify the validity of the model. In figures 8 and 9 the result concerning the specimens cut at
45 degrees are also reported. For this direction
 −1
e 1 1 1 ν̃12
E(45) =4 + + −2 , (54)
Ee1 Ee2 e 12
G Ee1
 
Ee45 ν̃12 1 1 1
ν̃(45) = 2 + − − . (55)
4 e e e
E1 G12 E1 E2 e

The shear modulus may also be obtained as

p
e 12 = G
G e 12 ε = G (1 − D1 )(1 − D2 ).
p
(56)
11

Due to experimental discrepancy it is not possible to conclude that the proposed theory exactly fits the
experimental results. Nevertheless no contradiction exists and, at least qualitatively, the agreement is good. The
200 J. Lemaitre et al.

Figure 9. Comparison theory–experiments for plane tension.

differences between experiments and numerical simulations partially come from a small initial anisotropy of
the material not taken into account here.

6. Coupling between damage and plasticity

The greatest value of the effective stress being associated with the principle of strain equivalence is to
naturally allow any mechanical coupling. For coupled plasticity, the only change in comparison with natural
plasticity is to replace the stress σ by the effective stress σ̃ (equation (7)) in the criterion function. The non-
linear kinematic hardening or back stress is introduced by considering the non-associated potential f + FX

3γ
f = (σ̃ − X)eq − R − σy , FX = X : X, (57)
4C
C and γ are the non-linear kinematic hardening material parameters. The associated variable to X is
α = 3X/2C whose evolution law is given by

∂(f + FX )
α̇ij = −λ̇ . (58)
∂Xij

The normality rule gives the following set of equations

 Anisotropic damage law of evolution 201
p
εij = εije + εij ,
1+ν ν
εije = σ̃ij − σ̃kk δij ,
E E


R = R∞ 1 − e−γ0 r ,
p 3 [Hik (σ̃klD − Xkl )Hlj ]D
ε̇ij = ṙ, (59)
2 (σ̃ − X)eq
2 3 (σ̃ijD − Xij )
Ẋij = C β̇ij − γ Xij ṙ, β̇ij = ṙ,
3 2 (σ̃ − X)eq
 s
Y  p 
Ḋij = ε̇ ij if r > pD
S
in which the plastic multiplier λ̇ is given by the consistancy condition f = 0, f˙ = 0 and ṙ = λ̇, the effective
stress tensor is defined by equation (7), the effective elastic energy density is
 
E εijeD εijeD e 2
εkk 2
σ̃eq Rν
Y= + = (60)
2 1+ν 3(1 − 2ν) 2E
and the damage threshold is given by equation (43). To write the previous set of equations as driven by the
variable r instead of the accumulated plastic strain p makes the equations simpler. Let us finally note that

[H (σ̃ D − X)H ]eq

ṗ = ṙ. (61)
(σ̃ − X)eq

7. Quasi-unilateral conditions

The partial closure of microcracks loaded in compression often induces a damage rate much smaller in
compression than in tension (Horii and Nemat-Nasser, 1983; Ladevèze, 1983; Ladevèze and Lemaitre, 1984;
Chaboche, 1993; Dragon and Halm, 1995). The difficulty is to recognize what is compression and what is
tension in a three dimensional state of stress and to write a Gibbs energy able to be differentiated. In the
isotropic case the problem has been solved by the introduction of a parameter h (0 6 h 6 1, often h = 0.2)
which operates by (1 − hD)−1 on the negative part of the principal stresses, the positive part being the classical
effective stresses σI /(1 − D) (Ladevèze and Lemaitre, 1984; Liénard, 1989).
Using the same concept in the present anisotropic damage theory consists in writing
H + = (1 − D)1/2 , H − = (1 − hD)1/2 , (62)
dH+ = ηDH , dH− = hηDH , (63)
 
1+ν − D − D
3(1 − 2ν) hσH i2+ hσH i2−
ρψ ∗ = Tr H + σ D + D
+H σ + + H σ −H σ − + + , (64)
2E 2E 1 − dH+ 1 − dH−
h·i+ and h·i− are the positive and the negative parts, σ D D
+ (resp. σ − ) is built with the positive (resp. negative)
eigenvalues and the corresponding eigenvectors of (H σ ) (resp. (H − σ D )) (Ladevèze, 1983, 1995). We
+ D

finally obtain the effective stress taking into account the quasi-unilateral effect
 

D
D hσH i+ hσH i−
σ̃ = H + σ D
+H
+
+ H −σ D
−H
−
+ + + 1. (65)
1 − dH 1 − dH−
202 J. Lemaitre et al.

The energy function Y to be introduce in the damage law is:

 
1 + ν  + D +
2 −
2  3(1 − 2ν) hσH i2+ hσH i2−
Y= H σ + H eq + h H − σ D
− H + + h . (66)
3E eq
2E (1 − dH+ )2 (1 − dH− )2

8. Strain localization

The classical conditions of localization (Rice and Rudnicki, 1980; Borre and Mayer, 1989; Benallal et al.,
1992)


det nLn = 0 (67)
are studied here in the framework of elasto-plasticity coupled to anisotropic damage. In this section, we prefer
intrinsic notations (second order tensors are underlined one time when fourth order tensors are underlined
twice), L is the tangent operator defined by
σ̇ = L : ε̇, (68)
n is the vector normal to the localization plane. Condition (67) makes it possible to determine the critical value
of the damage Dc by considering that a mesocrack initiation is the final stage of the strain localization.
In order to calculate L we write the law of elasticity as


σ̃˙ = E : ε̇ − ε̇ p , (69)

where the plastic strain is given by equations (32)–(33). The effective stress σ̃ (equation (7)) defines the fourth
order tensor Mij kl = Hik Hlj − 13 [Hkl2 δij + Hij2 δkl ] + 19 Hpp
2
δij δkl such that

σ̃ = M : σ . (70)

Then
∂ σ̃
σ̃˙ = σ̇˜ + : Ḋ, σ̇˜ = M : σ̇ . (71)
∂D
The calculation of ∂ σ̃ /∂D is similar to the calculation of Y (equation (9)) as

∂ σ̃ij ∂ ∂ 2 ρψ ∗ ∂Yrs
= e
Eij kl εkl = Eij kl = Eij kl . (72)
∂Drs ∂Drs ∂Drs ∂σkl ∂σkl

For commodity no kinematic hardening is considered (X = 0). The laws governing the evolution of the plastic
strain, of the isotropic strain hardening and of the damage are written as
3 (H σ̃ D H )D
ε̇ p = a ṗ with a = , (73)
2 (H σ̃ D H )eq
σ̃eq
ṙ = 1ṗ with 1 = , (74)
(H σ̃ D H )eq
 s
Y
Ḋ = d ṗ with d = |a| (75)
S
 Anisotropic damage law of evolution 203

and the accumulated plastic strain rate is determined by the consistancy condition f˙ = 0

b : σ̃˙ 3 σ̃ D
ṗ = with k = 1R 0 and b = . (76)
k 2 σ̃eq

With the notations just introduced

b : σ̃˙
σ̃˙ = E : ε̇ − 2Ga, (77)
k
b : σ̃˙ 2G
= b : ε̇ (78)
k k + 2Ga : b
and with equation (71)
 
2G ∂ σ̃
σ̇˜ = E : ε̇ − 2Ga + : d b : ε̇. (79)
k + 2Ga : b ∂D
Finally the expression for the tangent operator is for any loading
   
2G
∂ σ̃
L = Ẽ − Ẽ : a ⊗ b + M −1 : :d ⊗ b , (80)
k + 2Ga : b ∂D

where Ee = M −1 : E is the elastic effective tensor.

We derive here the conditions of localization in tension for isotropic as well as anisotropic damages. When
dealing with anisotropy, we consider an hydrostatic sensibility parameter η larger than 1 (usually 2 < η < 3).
If a tension loading σ = σ xE1 ⊗ xE1 is applied parallel to the xE1 direction, then a = b = diag[1, −1/2, −1/2].
The vector normal to the strain localization plane is
 
c
 
n= 
 s , c = cos θ, s = sin θ. (81)
0

For isotropic damage we introduce Lamé coefficients λ and µ such that

 
µ + (λ + µ)c2 (λ + µ)sc 0
 
e = (1 − D) 
nEn (λ + µ)sc µ + (λ + µ)s 2 0
 . (82)
0 0 µ

The strain localization conditions (67)–(80) are

       
 µ 2 5µ 1 
 (1 − D) µ + λ − c − Zµc2 (1 − D) λ + sc + Zµsc 
 3 3 2  
    =0 (83)
 5µ 2µ 2 
 (1 − D) λ + (1 − D) µ + λ +
 sc s 
3 3
204 J. Lemaitre et al.

in which
 s
2 σu2 σu
Z= . (84)
3 2ES µ
Further derivations give the closed form expression for the localization conditions as a critical damage Dc
depending on the normal n,
 s
σu2 σu 2 − 4ν + (5 − 4ν)s 2 c2
Dc (n) = 1 − g(θ) , g(θ) = . (85)
2ES µ 3 − 6s 2 c2 + (1 − 2ν)(2s 2 − c2 )

It can be verified that g is of the order of magnitude of unity. Furthermore, the ratio σu /µ is very small for
any material. We obtain the classical result of Benallal et al. (1992) for which the condition of localization in
tension is close to Dc ≈ 1.
For anisotropic damage, the same result (a critical damage almost equal to 1) is obtained but in terms of
hydrosatic damage: the condition of strain localization is close to

3
ηDH c ≈ 1 or DH c ≈ 1/η or D1c = 2D2c = 2D3c ≈ . (86)
2η

With η = 3, we obtain D1c ≈ 0.5. This value is in accordance with the critical damages often observed in
experiments with a poor accuracy owing to the difficulty in defining in practice what is the initiation of a crack!
The use of a strain localization criterion to model a mesocrack initiation is made possible owing to the
consideration of anisotropic damage. Such a criterion depends on the value of the parameter η which could
finally be identified by strain localization or mesocrack initiation measurements.

9. Conclusion

As a conclusion let us develop the topic ‘How to use anisotropic damage’ to predict crack initiation in real
structures. In our opinion the main difficulty is the identification of the constitutive equations for each material
(at each temperature considered!).

9.1. Identification

To identify the damage law which can be used either in monotonic loadings for ductile fracture or in cyclic
loadings for low cycle or high cycle loadings, it is best to consider experiments in the largest possible domain
of stresses, strains, time and number of cycles.
The parameters to be identified are defined by the basic equations
 s
Y  p 
Ḋij = ε̇ ij if p > pD ,
S
sup DI = Dc mesocrack initiation
I

with
 
E εijeD εijeD e 2
εkk 2
σ̃eq Rν
Y= + = ,
2 1+ν 3(1 − 2ν) 2E
 Anisotropic damage law of evolution 205

Figure 10. Tension test for identification.

 m
σu − σy
pD = εpD Max − σ
(87)
σeq y

they are
– S, s, εpD , m, Dc for damage;
– E, ν for elasticity;
– σu , σy for low cycle fatigue and fracture in tension. Add the fatigue limit σf for high cycle fatigue (two
scale damage model of Section 9.3).

9.1.1. Uniaxial tension with unloadings

One (at least) experiment in uniaxial tension with unloadings (figure 11) defines E, ν, σu , εpD , the isotropic
hardening law R(r) and the damage by the simplified equation (46)

Ee1
D11 ≈ 1 − . (88)
E
For the two scale model of Section 9.3 where only linear kinematic hardening is assumed, the tension
experiments allow then for the determination of the plastic sequent modulus C.

9.1.2. Low cycle fatigue tests

About ten very low cycle fatigue tests (at numbers of cycles to failure NR = 5, 10, 50, 100, 1 000 cycles)
allow us to calculate the damage evolution by equation (46), to draw the graph D11 (p) from which Dc is
determined from all tests, and εpD from monotonic tension/compression tests, pD from fatigue tests (figure 11).
The damage parameters S and s are obtained from the graph dD/dp function of the maximum stress σMax
as for fatigue test σ̃eq ≈ σMax for p > pD and Rν ≈ 1 as long as damage remains small
 s
dD11 σMax
= . (89)
dp 2ES

If needed, h comes from the measurements of the elasticity modulus in compression (figure 11).

9.1.3. High cycle fatigue tests

High cycle fatigue tests define the asymptotic value of the Woehler curve from which the fatigue limit is
appreciated as σf = σMax for a number of cycles to failure NR → ∞ with a zero mean stress.
206 J. Lemaitre et al.

Figure 11. Low cycle tests for identification.

The two scale model of Section 9.3 introduces plasticity and damage at a microscopic scale at which the yield
stress is taken equal to the fatigue limit σf . The exponent m is then obtained from the graph of the damage
threshold at microscale as a function of the macroscopic maximum applied stress σMax
pD σu − σf
ln = m ln . (90)
εpD σM − σf

Some two level high cycle fatigue tests help to define the graph for large pD as for p < pD , Dij = 0 (still at
microscale).

9.2. Post processing by uncoupled method

An elasto-plastic structure calculation by a finite element analysis or a Neuber plastic correction of an elastic
p
calculation gives the histories of the elastic strains εije (t) and of the plastic strain rates ε̇ij (t). The damage
evolution is obtained by a simple time integration of the laws (28)–(35) with an eventual initial damage
(Dij0 = D0 δij )
Z t s
Y  p
Dij = Dij0 + ε̇  dt if p > pD . (91)
ij
tO S
Ductile damage and low cycle fatigue may be analysed by this method.

9.3. Post processing by locally coupled method

When the damage is very localized, an elastic structure calculation gives the stresses and strains at the critical
point. Here a two-scale model solves the problem of a weak inclusion embedded in an elastic (or elasto-plastic)
representative volume element (Lemaitre and Doghri, 1994; Lemaitre et al., 1999). The law of localization
(Kroener, 1961; Berveiller and Zaoui, 1980) is used to calculate the microscopic stresses as a function of the
mesoscopic stresses applied on the RVE (µ stands for the microscopic values)
µp p

σijµ = σij − aE εij − εij , a ≈ 0.4. (92)

The constitutive equations at microscale are those of elasto-plasticity coupled to damage with the fatigue
limit taken as the yield stress and the strain hardening considered only as linear kinematic. The code
 Anisotropic damage law of evolution 207

DAMAGE97 (Sermage, 1998) solves the equations in the case of isotropic damage and ‘soon’ in the case
of anisotropy.
This method applies with success for high cycle fatigue.

9.4. Fully coupled method

When the damage is not localized and if a high accuracy is needed, the full set of the coupled constitutive
equations (59) needs to be solved as field variables. The routines ABAQUS (UMAT) PLASTENDO and
VISCOENDO (Benallal et al., 1988; Florez, 1992) solve these equations in case of isotropic damage and
‘soon’ in case of anisotropy but the price to pay is a very important time of calculation.

References
Benallal A., Billardon R., Doghri I., 1988. An integration algorithm and the corresponding consistent tangent operator for fully coupled elastoplastic
and damage equations. Comm. Appl. Numer. Meth. 4, 731–740.
Benallal A., Comi C., Lemaitre J., 1992. In: Ju J.M. et al. (Eds.), Damage Mechanics and Localisation, Vol. 13, AMD-vol. 142.
Berveiller M., Zaoui A., 1980. Self-consistent schemes for heterogeneous solids mechanics. In: Comportement Rhéologique et Structure des
matériaux, CR 15th coll. GFR, Paris.
Borre G., Mayer G., 1989. On linear versus nonlinear flow rules in strain localization analysis. Meccanica 24 (1), 36.
Chaboche J.L., 1993. Development of continuum Damage Mechanics for elastic solids sustaining anisotropic and unilateral damage. Int. J. Damage
Mech. 2, 311–329.
Chrysochoos A., 1987. Thèse d’état de l’université Paris 6.
Cordebois J. P., 1983. Critère d’instabilité plastique et endommagement ductile en grandes déformations. Doctorat d’état de l’université Paris 6.
Cordebois J.P., Sidoroff F., 1982. Endommagement anisotrope en élasticité et plasticité. J.M.T.A., Numéro spécial, 45–60.
Chow C.L., Wang J., 1987. An anisotropic theory for continuum damage mechanics. Int. J. Fract. 33, 3–16.
Dragon A., Halm D., 1995. A model of anisotropic damage by microcrack growth, unilateral effect. Proc. Workshop Mechanical Behavior of
Damaged Solids, Fontainebleau, France, November.
Dragon A., Cormery F., Desoyer T., Halm D., 1994. Localized failure analysis using damage models. In: Vardoulakis (Ed.), Localisation and
Bifurcation Theory for Soils and Rocks, Chambon, Desrue, A.A. Balkema, pp. 127–140.
Dufailly J., 1995. Identification des lois de comportement à variables internes. Habilitation à diriger des recherches, Université Paris 6.
Fichant S., 1996. Endommagement et anisotropie induite du béton de structures. Modélisations approchées. Thèse de doctorat de l’université Paris 6.
Florez, 1992. Viscoendo user’s guide. Notice 14 LMT-Cachan.
Germain P., Nguyen Q.S., Suquet P., 1983. Continuum thermodynamics. J. Appl. Mech. 50, 1010–1020.
Halphen B., Nguyen Q.S., 1975. Sur les matériaux standards généralisés. J. Méc. 14 (1).
Hayhurst D.R., Leckie F.A., 1973. The effect of creep constitutive and damage relationships upon the rupture time of a solid circular torsion bar. J.
Mech. Phys. Solids 21, 431–446.
Horii H., Nemat-Nasser S., 1983. Overall moduli of solids with microcracks: Load induced anisotropy. J. Mech. Phys. Solids 31, 155–171.
Ju J., 1989. On energy-based coupled elasto-plastic damage theories: Constitutive modeling and computational aspects. Int. J. Solids Struct. 25 (7),
803–833.
Kachanov M., 1993. Elastic solids with many cracks and related problems. In: Hutchinson J., Wu T. (Eds.), Advances in Appl. Mech., Vol. 1, Acad.
Press Pub., pp. 260–445.
Kracinovic D., 1985. Continuous damage mechanics revisited: Basic concepts and definitions. J. Appl. Mech. 52, 829–834.
Kroener E., 1961. On the plastic deformation of polycrystals. Acta Metall. 9, 155–161.
Ladevèze P., 1983. On an anisotropic damage theory. In: Boehler J.P. (Ed.), Proc. CNRS Int. Coll. 351 Villars-de-Lans, Failure criteria of structured
media, pp. 355–363.
Ladevèze P., 1995. Modeling and simulation of the mechanical behavior of CMCs. In: Evans A.G., Naslain R. (Eds.), High Temperature Ceramic-
Matrix Composite, Ceramic Transaction 57, 53–64.
Ladevèze P., Lemaitre J., 1984. Damage effective stress in quasi unilateral conditions. 16th International Congress of Theoretical and Applied
Mechanics, Lyngby, Denmark.
Leckie F.A., Onat E.T., 1981. Tensorial nature of damage measuring internal variables. In: Hult J., Lemaitre J. (Eds.), Physical Non-Linearities in
Structural Analysis, Springer, Berlin, pp. 140–155.
Lemaitre J., 1992. A Course on Damage Mechanics, Springer-Verlag.
Lemaitre J., Chaboche J.L., 1985. Mécanique des matériaux solides. Dunod, Mechanics of Solid Materials, Springer-Verlag, 1987 (English
translation).
Lemaitre J., Doghri I., 1994. Damage 90: A post-processor for crack initiation. Comput. Meth. Appl. Mech. Engrg 115, 197–232.
208 J. Lemaitre et al.

Lemaitre J., Dufailly J., 1987. Damage measurements. Engrg. Fract. Mech. 28, 643–661.
Lemaitre J., Desmorat R., Sauzay M., 1998. Some considerations about anisotropic damage. International Symposium on Transient Loading and
Response of Structures, NTNU Trondheim, Norway.
Lemaitre J., Sermage J.P., Desmorat R., 1999. A two scale damage model. Int. J. Fract., in press.
Liénard C., 1989. Plasticité couplée à l’endommagement en conditions quasi-unilatérales pour la prévision de l’amorçage de fissures. Thèse de
doctorat de l’université Paris 6.
Mazars J., Berthaud Y., Ramtani S., 1990. Engrg. Fract. Mech. 35 (4/5), 629.
Murakami S., 1988. Mechanical modeling of material damage. J. Appl. Mech. 55, 280–286.
Murakami S., Ohno N., 1978. A constitutive equation of creep damage in pollicristalline metals. IUTAM Colloquium Euromech 111, Marienbad.
Nouailhas D., 1980. Etude expérimentale de l’endommagement de plasticité ductile anisotrope. Thèse de doctorat de l’Université Paris 6.
Rice J.R., Rudnicki J.W., 1980. Int. J. Solids Struct. 16, 597.
Sermage J.P., 1998. Fatigue multiaxiale à température variable. Thèse de doctorat de l’université Paris 6.
Tanaka K., Mura T., 1981. A dislocation model for fatigue crack initiation. J. Appl. Mech. 48, 97–103.