ECF15

A COHESIVE ZONE MODEL WITH BARENBLATT SURFACE

ENERGY : THEORETICAL CONSIDERATIONS AND
NUMERICAL SIMULATIONS
Jrme Laverne* **, Eric Lorentz* & Jean-Jacques Marigo**
*LaMSID UMR2832 CNRS-EDF R&D, 1 av. du Gnral de Gaulle, 92141 Clamart, France
**LPMTM UPR9001 CNRS Univ. Paris 13, 99 av. J.B. Clment, 93430 Villetaneuse, France
[email protected]

Abstract
By using a principle of least energy and a Barenblatt-type surface energy assumed as a
smooth function of the jump of the displacements, we prove that it is possible to predict the
onset of cracking by searching relative minima of the total energy of a body, the yield
criterion depending on the choice of the energies. These types of brittle fracture models are
then tested numerically by using cohesive elements.

1. Introduction
It is known, see Francfort and Marigo [1], that the linear fracture Mechanics based on
Griffith's hypotheses relative to the surface energy (the surface energy associated to a surface
of discontinuity is proportional to the area of this surface) and on the criterion formulated in
terms of the critical energy release rate is (generally) unable to predict crack initiation in an
initially sound elastic structure.
One of the remedies proposed by Francfort and Marigo was at first, while keeping
Griffith's hypothesis on the surface energy, to replace the criterion of propagation by a
principle of least energy (Griffith revisited). This principle, along with numeric adapted
methods (see Bourdin et al. [2]), allows to widen the frame of application of Griffith's theory
and is able to predict initiation and propagation of cracks following a priori non restricted
space-time paths : multi-cracking, brutal propagation, branching, . In counterpart the seek
of global minima leads to non-admissible defects like spurious size effects or the inability of
the material to sustain body forces. Moreover, considering only local minima is (almost)
equivalent to Griffiths criterion and its incapacity to predict the initiation.
As a remedy, we propose to change the form of the surface energy by adopting
Barenblatts idea. In the spirit of what happens at an atomic scale when atomics bonds break
we will assume that the surface energy depends on the value of the displacement jump
through the crack. This approach corresponds to the cohesive zone models (CZM) frequently
developed in the literature. With this change of the form of the energy and the seek of local
minima we remedy to the main defects of Griffith's theories (original or revisited). It is at
least what suggest analyses in dimension one made by Del Piero and Truskinovsky [4] or by
Charlotte et al. [3] where it is shown that one of the necessary conditions of local minimum is
that, in each sound point of the structure, the stress is lower than a critical stress
corresponding to the derivative (with regard to the displacement jump) of the surface energy
at zero. In other words the principle of local minimum applied in the case of a surface energy
of Barenblatts type leads to a yield criterion formulated in terms of the stress tensor.

ECF15

We suggest here to generalize this result for three-dimensional structures. This is the
object of the following section. We will then present the numerical implantation of this type
of model by using cohesive elements. Finally this brittle fracture model will be tested
numerically on a few configurations of fracture.

2. Surface energy and yield criterion

We consider three-dimensional homogenous isotropic brittle bodies. The energy Es
necessary to create a surface of discontinuity of normal n and through which the
displacement undergoes a discontinuity is, by following the idea of Barenblatt :
Es =

( , n)d

where is a concave function, vanishing if = 0 and growing towards the tenacity Gc

when the jump displacement tends to infinity. Moreover, for an isotropic material, the
function surface energy has to satisfy the condition (Q , Qn) = ( , n) for any
orthogonal matrix Q . By decomposing the jump displacement in a normal and a tangential
part : = n n + t , t n = 0 , the surface energy can read as :

( , n) = ( n, t

To prevent the interpenetration of the lips of the crack we impose n 0 . Finally we

suppose that ( 0, 0 ) = 0 and that the derivative of at ( 0, 0 ) is defined by :
lim

h 0 +

1
(h , h ) = ( , ) with positively homogeneous of degree 1.
h

Let us consider such a structure submitted to a given load. The state of this structure is
characterized by the displacement field u and the stress field (cracks are identified with the
surfaces of discontinuity of u and thus known once u is known). A necessary and sufficient
condition so that the state of the structure is a locally stable equilibrium state is that it is a
local minimum of the energy. In other words, there exists a neighbourhood (according to the
chosen norm) of u such that the energy of the structure in this state is less than the energy of
the structure in any other admissible field v in this neighbourhood.
The energy of the structure is constituted here by three terms : the elastic energy (defined
on the sound zones of v), the surface energy (defined on the surfaces S v of discontinuity of v)
and the potential of the dead loads f ( v ) :
(v ) =

(v n, v ) d f (v)

A (v) (v) dx +

Sv

with v : jump of v

Sv

From a mathematical point of view, the most convenient space of displacement fields is
the space of bounded variation which gives the norm used in the condition of local minimum.
However, in this paper, we will limit our attention to piecewise smooth displacement fields.
Thus the condition of local minimum for u reads as :

r > 0 v : v u r

E (u ) < E (v)

ECF15

The necessary condition so that u is a local minimum, called optimality condition of first
order (OC1), formally reads as E ' (u)(v) 0 for any admissible direction v. From classic
arguments of calculus of variations, we deduce that the OC1 are equivalent to the following
local conditions :
(a) Equilibrium equations :

div + f = 0 in \ Su

(b) Boundary condition :

n = F

(c) Jump condition :

n = ,1 ( n, t ) n + ,2 ( n, t

on

(d) Yield criterion : n ( n , n n

on Su

n, : n = = 1

in \ Su

with = A (u ) and = u = u + u .
Depending on the properties of the derivative of at ( 0, 0 ) we will find different types of
yield criteria.
- If is differentiable at ( 0, 0 ) : is linear and we have ( , ) = c + c , c and

c being characteristic stresses of the material. The yield criterion (d) becomes :

n c n + c n n ,

n, : n = = 1

in \ Su

which can be written in term of principal stresses { i }i =1,3 :

supi c and

1i3

sup i j c .

1i, j3

Its a criterion on both maximal traction and maximal shear :

- If is not differentiable at ( 0, 0 ) : When admits only directional derivatives at

( 0, 0 ) ,

is no more linear. By using Legendre transform, it appears that the criterion

(d) is of intrinsic curve type, the maximal shear stress decreasing monotonically with
the normal stress. It depends only on the extreme principal stresses 1 and 3 and can
be written :
1 + 3
0
2

1 3 + 2

with

( s ) = sup

[ 0, 2]

{s cos ( cos ,sin )}

The difference in comparison to Griffith's theory lives in the conditions (c) and (d). For
Griffiths theory, the first one would read n = 0 , which implies the absence of interaction
between the lips of cracks, while the second one would be : G Gc the famous Griffith
criterion which is global and no more local. We see that with a Barenblatts surface energy
type, the lips of the cracks interact, which correspond to the notion of cohesive forces.
Remarks: The OC1 are only necessary conditions for u to be a local minimum. It is
necessary to add to them second order conditions which are generally global ones (see [3] or
[4]). For more details on this part see Laverne and Marigo [5].

ECF15

3. Numerical Model
One of the main difficulties of numerical implantation of such models lives in the necessity to
suppose a priori any surface of discontinuity in the structure. This difficulty had been bypassed in the revisited Griffith theory by making an "elliptic regularization" of the energy
which allowed to work with regular fields (see [2]), the discontinuities being replaced by
strong gradients. Such a process of regularization is not available yet for Barenblatts surface
energies. We have therefore opted for an implantation (in two dimensions) by cohesive
elements with internal discontinuities. In other words we suppose the crack path is known a
priori.
3.1 Choice of the surface energy
We choose a density of surface energy depending only on the norm of the jump displacement
and, on this fact, independent of the orientation n of the discontinuity line :

Gc

(n, ) = Gc 1 exp

We note that its a concave function of the norm of the jump displacement, starting from zero
and progressively growing to the tenacity Gc (see [3]).

3.2 Choice of the cohesive element

The element is a quadrangle with an internal discontinuity noted [AB] which is located at the
centre of the sides [1-4] and [2-3] (see Fig. 1). It allows to define a local mark in the element :
n and t are respectively a normal and a tangent unit vector to the discontinuity line. The main
idea consists in considering the jump of displacement in the element, denoted = ( n , t ) , as
a constant internal variable. Moreover we consider that the normal jump n cannot be
strictly negative to take into account the condition of non-interpenetration of the crack lips.
3

t
n
Y

4
A
1

FIGURE 1. Geometry of the cohesive element.

ECF15

3.3 Numerical solving

By using a principle of least energy the solution of the problem consists in finding the
displacement field as well as the jumps in each cohesive element.
Minimisation problem : find (U , ) = arg min ( E (U , ) )
U ,

Numerically, the seek for a local minimum of the total energy at a given level of loading
will be made in two steps : successive minimization with respect to the jump then with
respect to the displacement.
Remark : in this part we will detail only the computation of the jump in a cohesive
element for the onset of cracking. The calculation of the jump for propagation is similar. We
take into account the irreversibility of cracking thanks to a threshold variable.
First step : Determination of for a fixed U . Thanks to the choice of constant by
element, a local analysis is sufficient to determine in every element according to the
displacement. Indeed, for given U , will be a minimizer of the element total energy.
Supposing that the element is free of body forces, its energy reads as (with matrix notations) :
E (U , ) =

1
T
BU D ) A ( BU D ) dV + LGc 1 exp c
(

2 V
Gc

the first term corresponds to the elastic energy (the strain read as = BU D and the
stress = A ) and the second one to the surface energy (L is the length of the discontinuity
line AB on Fig. 1). E is a strictly convex function of as long as the size of the element is
small enough, which fixes the maximal mesh size. There is then a unique minimum. We
+
obtain (by noting x = Max {x, 0} ) :
If

n T D T AB U

) + ( t D AB U )

+ 2

L 2 c 2 , then = 0 . It corresponds to the onset

criterion (d) in part 2. Otherwise the jump reads as :

t T D T AB U
= a

t T DT AD t a + L c exp c

Gc

t +

n D AB U

nT DT AD n a + L c exp c
Gc

with a > 0 solution of the following non linear scalar equation :

2

T
T
+
T
T
n D AB U
t D AB U

1 =
+

t T DT AD t a + L c exp c a
nT DT AD n a + L c exp c a

Gc
Gc

Second step : Determination of U. Once is determined in every cohesive element

according to the displacement, we solve the discrete equilibrium equation (equivalent to (a)
and (b) in part 2) KU = F where F , the vector forces, must be updated to take into account
stresses associated to in cohesive elements.
Iterations : For a given load level, the two steps are iterated : from an initial vector U 0
we build, until convergence, a sequence ( i , U i ) with a global Newton-Raphson algorithm.

ECF15

4. Simulations
4.1 Validation test
The aim of this test case is to validate the numerical model by confronting it with an
analytical solution. Let us consider a crack along which the jump displacement is not
constant. The idea is then, in addition to a good prediction of the jump, to notice a decrease of
the numerical error when we refine the mesh.
We consider a two-dimensional rectangular elastic plate and one of its sides. The
Airy stress function ( x, y ) , governed by the equation : = 0 on , when body forces
vanish, generates stresses that satisfy the equations of equilibrium and compatibility (see
Fung [8]). In this case xx , yy and xy are derived from ( x, y ) according to the
following equations :

xx =

2
2
,

=
yy
y 2
x 2

and xy =

2
xy

Lets chose a biharmonic function ( x, y ) such that the stress tensor reads as :

xx = x + y +

yy = 0

xy = y +

with , , and arbitrary constants.

For a Young modulus : E = 1 and a Poissons ratio : = 0 we deduce the displacement

field in the plate (and in particular the boundary conditions on ) :
x2

u ( x, y ) = + y 2 + x ( y + )
2

2
x
+ 2 x
v ( x, y ) =
2

(1)

Furthermore, knowing the vector .n on , we can deduce the jump S on ,

corresponding to such a force, by inverting Barenblatts law. The new prescribed
displacement on , generating such a jump, thus reads as : U = U 0 S (with U 0 the
prescribed displacement on given by (1)). So we build an analytical solution of the elastic
plate containing a crack along which the jump displacement is not constant.
Finally, for the numerical simulation, we put cohesive elements along , we apply a
displacement U on and a displacement given by (1) on . Let us note S num the
numerical solution for the jump. We notice that the numerical error S S num
when we refine the mesh (see Fig. 2).

L2 ( )

decreases

ECF15

FIGURE 2. Error vs. number of cohesive elements (logarithmic scale).

4.2 The pull-out problem

This study concerns a cylindrical beam composed of two parts : a matrix with cross circular
section reinforced by a fibre immersed on its centre. The matrix is assumed to be linear
homogeneous isotropic material with an elastic behaviour while the fibre is taken rigid. The
beam is clamped on its lateral surface, while a longitudinal displacement U i is prescribed at
the top of the fibre. Cohesive elements are located everywhere in the matrix so that we
authorize only longitudinal debonding (see Fig. 3).
Moreover a technique of path-following of the load was developed to take into account a
possible brutal opening of the cracks see Badel and Lorentz [7]. It allows to follow the
unstable branches of the global response of the structure.
The numerical simulation leads to a debonding at the interface Matrix / Fibre on all the
height of the beam. The results are validated by comparing the numeric curve of the global
response with the analytical solution (see Fig. 4).

Ui

y
x
Fibre
Matrix
Cohesive elements
Fixed boundary

FIGURE 4. Global response of the beam.

Force vs. prescribed displacement.

Interface Matrix / Fibre

ECF15

FIGURE 3. Geometry of the beam.

4.3 Crack propagation through a perforated plate

This study concerns the simulation of crack propagation through a perforated plate which is
submitted to a prescribed force Fi applied on both ends, see Fig. 5. By postulating a priori the
potential crack paths, we put cohesive elements along the symmetry line A-A right and left
from the hole. Global response and displacement field are given in Figs. 6-7.
Fi

650

350

R = 500

500

Fi

FIGURE 5. Geometry of the plate

and load conditions (dimension in mm).

FIGURE 6. Global response of the plate.

Prescribed force vs vertical displacement
at the top of the hole.

FIGURE 7. Evolution of the displacement field.

References
1. Francfort, G.A. and Marigo, J. -J., J. Mech. Phys. Solids, vol. 46(2), 1319-1342, 1998.
2. Bourdin, B., Francfort, G.A. and Marigo, J. -J., J. Mech. Phys. Solids, vol. 48, 797-826,
2000.
3. Charlotte, M., Francfort G.A. , Marigo, J. -J. and Truskinovsky L., The data science
library, Elsevier, edited by A. Benallal, Paris , 7-18, 2000.
4. Del Piero, G. and Truskinovsky, L., Int. J. Solids Structures, vol. 38, 1135-1148, 2000.
5. Laverne, J. and Marigo, J. -J., Acadmie des Sciences C.R. Mcanique 332 (2004).
6. Laverne, J., Lorentz, E. and Marigo, J. J., In Proceedings of the 16th French Congress of
Mechanics Nice 2003.
7. Badel, P.-B. and Lorentz, E., In Proceedings of the 16th French Congress of Mechanics,
Nice, 2003.

ECF15

8. Fung, Y.C., Foundations of Solid Mechanics, Prentice-Hall, 1965.