Sunteți pe pagina 1din 38

Crack propagation in a three

dimensional FRC unit cell


Three months research training preriod at Brunel University
Vincent VISSEQ
Supervisor : Giulio ALFANO
August 11 2008
Contents
1 Overview 6
1.1 Background of this study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 A brief introduction on composites . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Multi-scale approaches and homogenization . . . . . . . . . . . . . . . . . . . . 9
1.4 Nonlinear modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2 Cohesive elements in Abaqus 15
2.1 Cohesive elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Initial behavior and damage initiation criteria . . . . . . . . . . . . . . . . . . . 17
2.3 Softening onset prediction (Mixed-mode) . . . . . . . . . . . . . . . . . . . . . . 18
2.4 Delamination propagation prediction and damage evolution law . . . . . . . . . 20
3 Representative volume element characterization 21
3.1 Size of the RVE, number of bres and location of the crack . . . . . . . . . . . 21
3.2 Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Materials and interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4 Specicities of the modeling under Abaqus 24
4.1 Contact status and mesh of interface elements . . . . . . . . . . . . . . . . . . . 24
4.2 Partition of the matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5 Numerical simulation 26
5.1 Stress concentration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.2 Damage evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.3 Global response of the unit cell - comparison of Abaqus/Standard and Abaqus/Explicit 28
5.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3
6 Damage modeling 30
6.1 Some damage modelings and their deciencies . . . . . . . . . . . . . . . . . . . 30
6.2 Gradient-based damage modelings . . . . . . . . . . . . . . . . . . . . . . . . . 30
6.3 Interpretation and tools used . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
7 Numerical simulation with a matrix damage model 33
7.1 Parameters of the Abaqus damage model . . . . . . . . . . . . . . . . . . . . . 33
7.2 Simulations for the Small Unit Cell . . . . . . . . . . . . . . . . . . . . . . . . . 34
7.3 Simulations for the Bigger Unit Cell . . . . . . . . . . . . . . . . . . . . . . . . 35
7.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
Introduction
Composite materials are now widely used in several industrial elds. The most spectacular
one is probably aviation were companies have built airliners with 50 per cent of composites.
This explains why the international research community has produced many studies on it, and
why others are undergoing. Indeed, a good knowledge upon composite materials allow us both
to predict their behavior and to design other ones which hold out properties needed.
Many dierent types of composite materials exist and they are increasing every day, which
is well illustrated by new reinforcements made of nanotubes. More precisely, if we focus on
reinforced composites, one can see that reinforcements are mainly bers (long or short) or
particles (spheroid or akes). In this study we will be particularly interested in (long) bre
reinforced composites (FRC).
Even if a combination of both numerical simulations and experiments would have been wise,
we concentrate ourselves on an accurate computational modeling. Nowadays, even if numerical
tools allow us to perform complex simulations of mechanical problems, it is still very dicult
to accurately simulate a structural response taking into account micromechanical phenomena.
That is the reason why multi-scale approaches and homogenization techniques have been de-
veloped to link phenomena which occur at micro and macro scales.
If a multiscale approach is needed to accurately resolve mechanical problems for a whole struc-
ture, it is then obviously the case for composite ones. Indeed, composites, and (as a particular
case) unidirectional bre reinforced composites are anisotropic materials. Furthermore, both
debonding at the interfaces between matrix and bres and crack propagation might be taken
into account in the model to understand why, how and when micro cracks leads to a macro
crack.
In this study we will model a Representative Volume Element (RVE) of unidirectional FRC,
with a preexisting cracked bre, up to failure. The aim is to understand the failure history,
and particularly how the stress previously carried by the broken bre is redistributed to the
unbroken bres and to the matrix.
To this end, preliminary investigation in theoretical tools used has been made [Chapter 1
and 2].
The dierent steps of the study are summarized as follows:
1. The design of a representative volume element (RVE) which might be appropriate for
the work considered: ability to take into account the redistribution of stress and strain and
representativeness of a whole structure [Chapter 3].
2. A nite element modeling using the software ABAQUS [Chapter 4].
3. The simulations of bres/matrix debonding under longitudinal load [Chapter 5].
4. The simulations of fracture types taking into account the damage of the matrix [Chapter
6 and 7].
5
Chapter 1
Overview
1.1 Background of this study
In this rst part we are going to give an idea of the background of this study. Several concepts
and tools are needed which are related to the general theory of continuum mechanics an the
study of bre reinforced composites.
We have decided to give explanations especially on the following points:
1. Fibre reinforced composites: products, processes and materials
2. Interface and interphase between materials
3. Homogenization techniques
4. Nonlinear nite element analysis
1.2 A brief introduction on composites
As we have already said, good knowledge on composite materials allow us both to predict their
behavior and to design other ones with properties searched. As an example, it is very useful to
know precisely what parameters are involved in energy dissipation during a crash. With those
informations, one could theoretically determine what the matrix/bres brace which maximizes
the energy dissipated is.
Nevertheless, to have an extended idea of the properties of a composite material, it is useful
to consider the triptych : product-process-material.
1. Product: Composites materials are often called designed materials which means that
composites are designed by engineers and searchers to obtain properties which do not
exist in classical materials. For example one can have the idea to conjugate the high
ductility and strong stiness of steel to the high strength of ceramique.
This example highlight the possibility of combining dierent materials. Nevertheless, the
way in which those materials are combined is determinant. According to the shape of the
structure and its main mechanical solicitations, one might adapt the kind of combination
between materials (e.g. bonded lamina, multi or uni-directional reinforcement).
2. Process: Many specic processes have been developed in the last century to obtain
reliable composites. They are mainly classied in terms of rates of production and global
6
Crack propagation in a three dimensional FRC unit cell year 2008
cost. One can cite, for FRC polymers, bag molding, compression molding, poltrusion or
lament winding.
A good knowledge on those processes is needed to accurately model interfaces between
matrix and reinforcement as well as both materials. Depending on the degree, the pres-
sure and the length of cure, the process can leads to modify the behavior at mesoscopic
and structural level of the composite (e.g. prestress due to shrinkage). Furthermore,
manufacturing causes defects as voids inside the matrix or buckling of bres.
The frequency and spacial distribution of those preexisting voids, distortions and pre-
stress are key informations to predict initiation of failure.
Finally, the process is also a constraint in the design steps.
3. Material: material properties have several facets. If the global characteristics of several
materials are well known (density, fusion temperature,...) one could also need to know
some details about smaller scales. Like the length of a polystyrene macromolecule, its
shape, the fraction of polymer cristallised, etc...
As a further example, some damage modelings involve to determine a characteristic
length at mesoscopic scale (e.g. gradient based strategies to avoid mesh dependency)[1].
1.2.1 Some studies on bre reinforced composites
A number of authors have addressed micromechanical analysis of a representative unit cell. A
brief description of some of these recent studies is reported below.
1. Xia and al. [2] designed a three dimensional unit cell. They have studied how the load
carried by an broken bre is then redistributed among the matrix and the unbroken
bres. This paper is very important for us since our study is very close to it. Indeed,
the main objective is to handle the evolution of damage at the larger scale of the bers
by results at the micro scale. In this purpose, its necessary to establish the connection
between the detailed deformation at micro scale for a multibre damage problem and
large-scale component performance. In their paper, the stress transfer is modeled by
Coulomb friction end they are using contact (gap) elements to model interfaces.
2. Bonora and Ruggiero [3] tried to take into account the stress/strain histories due to
manufacturing. Thus they incorporate material dependency as well as damage processes
and follow the associated progressive degradation of the overall properties in advanced
special algorithms. They shown that matrix/bre interface properties result of manu-
facturing process. An example of that is the prestress due to the nal cooling phase in
forming process. They developed a 2D unit cell model for SiC ber reinforced metal
matrix composite (MMC) laminate.
3. Gonzalez and Llorca [4] formulated the project to have a precise knowledge of the lamina
behavior under transverse loading until failure to develop a robust failure criteria. The
paper shows a two dimensional representative volume element (RVE) where circular
elastic bres are randomly distributed. They also deal with both the size of RVE (results
for 30 bres and 70 bres are shown) and properties of the interface (one case for a
strong interface and another with a weak interface). Those interfaces were modeled
by cohesive elements under ABAQUS and the damage of the matrix was evaluated in
term of accumulated plastic strain.
report.pdf 7 August 11 2008
Crack propagation in a three dimensional FRC unit cell year 2008
Figure 1.1: Fibres reinforced composite under transverse loading
1.2.2 Interface and interphase between materials
As in the rst part of this study we are going to model debonding between matrix and bres,
it seems to be essential to dene precisely the concept of interface.
An interface is widely considered as a mathematical concept, the common boundary of
reinforcing bre and matrix which maintain the bond in between for the transfer of loads.
This denition leads to consider the thickness of this interface as equals to zero. Furthermore,
its physical and mechanical properties are unique from the bre and the matrix.
In opposition to this concept, some authors are using the word of interphase, which is
dened as a region of nite volume extending from the bulk bre through the actual interface
into the matrix. Then it embraces all the volume altered during the fabrication process of
the composite. As a consequence, the chemical, physical and mechanical properties of the
interphase vary continuously or in a stepwise manner along the nite thickness.
This last concept allows us to describe the adhesion mechanisms (absorption and wetting,
electronic attraction, chemical bonding, Vander Waals forces,...) which lead to the bond
between the two materials. However, as these adhesion mechanisms are reduced to mechanical
properties, the distinction of those two concepts is often ignored.
Indeed, the interface properties are essentially reduced to:
- the shear and tension strengths
- the critical strain energy release rate for mode I, II and III (G
IC
, G
IIC
, G
IIIC
).
These quantities are measured by means of two kind of tests:
- In the rst category fall the tests in which bres are embedded in specially constructed
blocks of matrix (single bre compression test, bre fragmentation test, bre pull/push-out
test, slice compression test) which are used to determine the bond shear and tension strengths.
- The second type is represented by the interlaminar/intralaminar tests where bulk lami-
nate composites are used (Double Cantilever Beam (DCB) test to estimate G
IC
, End Notched
Flexure or End Loaded Slit test to estimate G
IIC
).
The main problem concerning these tests is the large data scattered from one laboratory
to another. J-K Kim and Y-W Mai [5] proposed an explanation of that weird fact, based on
the idea that dierent assumptions have been made by searchers concerning the type of failure
which had taken place without conrmation.
report.pdf 8 August 11 2008
Crack propagation in a three dimensional FRC unit cell year 2008
1.3 Multi-scale approaches and homogenization
Multiscale approaches are necessary in mechanical modelings to have simplied methods of
resolution. Those methods derive from continuum mechanics, but are widely linked with
material(s) properties and the geometry of the structure studied. The main idea is to discretise
the bigger scale without taking into account the smaller one.
1.3.1 Example of multi-scale approach
One of the most famous two-scale method is certainly the beam theory. Because of the slen-
dernessed ratio of the structure considered, one of the three dimensions of the body is favored,
which allows us to consider its sections as rigid. This fact is taken into account by the expres-
sion of the displacement eld as follows:
u(M) = u(m) +(m) X
Where m is the geometrical gravity center of the section S
D
(s), s the curvilinear abscissa
and X the vector mM. (See gure 1.2)
Figure 1.2: Modelisation of a beam
Remark : The classic theory based on the expression above of the displacement assumes
that an elastic deformation energy exists and the external loads derive from a potential. Under
these hypotheses one can prove that the general expression of the deformation energy can be
written as follows:

1
2
_

: =
1
2
_
S
D
[T(s) (s) +M(s) (s)]
One can see that this theory is very similar, in the general form, to the plate theory.
Indeed, this last theory is based on the same geometrical type of assumption, with two favored
directions.
1.3.2 Homogenization
Multiscale approaches concerning composites especially focus on the fact that the material is
heterogeneous at the microscale but can be considered as anisotropic and homogeneous at the
macro scale.
report.pdf 9 August 11 2008
Crack propagation in a three dimensional FRC unit cell year 2008
Homogenization is a class of theories which allows us to obtain the explicit dependence of
the bigger scale on the smaller scale.
The following two main assumptions are made:
- there exists a pattern (smaller scale) which allows us to reconstitute the whole structure
- the characteristic dimension of the pattern is negligible compared with the structures one.
Several homogenization methods have been developed based on rst or second order gra-
dient theories [6]. In the rst case one can advert asymptotic homogenization [7] of periodic
material and auto-coherent method [8] for randomly distributed materials.
1.3.3 Kinematic homogenization
In this section we present some details of a special type of homogenization technique called
Kinematic homogenization [9].
As other homogenization methods, the aim is to obtain an homogeneous equivalent material
which gives an approximazised response on the RVE. To this end, a series of homogeneous loads
are applied to determine the local response. Then the homogeneous operators are build using
the equivalence in deformation energy.
Kinematic homogenization dene the homogeneous behavior which gives the real RVE
deformation energy for boundary conditions written as

om = u
d
Let us consider the mechanical problem written upon the RVE as:
Find u(m), (m) and (m) such
u

om
div = 0 (1.1)
= K(m) (u(m))
where K(m) is the heterogeneous constitutive tensor of the RVE, (m) the local strain, and
(m) the local stress.
Remark : One can see that u(m) =

(m) om is the unique solution of (1.1).


Then the idea is to write the equivalence in deformation energy as
e
real
d
=
1
2

_

0
Tr
_
(m) (m)
_
d
0
=
1
2
vol()Tr
_

_
,

(1.2)
where

K denes the homogeneous behavior, and
1
2
vol()Tr
_

_
the deformation
energy of the homogeneous problem.
The next step is to calculate (m)
1
for

given.
The local deformation depends linearly from

as (m) = L(m)

.
1
Engineering notations are used further for simplicity
report.pdf 10 August 11 2008
Crack propagation in a three dimensional FRC unit cell year 2008

=
i

i
is associated with

, where, for example


1
= [1, 0, 0, 0, 0, 0] and
4
=
[0, 0, 0,

2, 0, 0].
More generally, we have :

= [
1
,
2
,
3
,

2
4
,

2
5
,

2
5
]
Then one solves six typical problems as:

=
1
u
d
=
_
_
_
1 0 0
0 0 0
0 0 0
_
_
_

_
_
_
x
y
z
_
_
_
=
_
_
_
x
0
0
_
_
_
where
1
(m) is related to
1
.
Finally we have
1
(m) = L(m)
1
as follows:

1
(m) =
_

1
(m);
2
(m);
3
(m);
4
(m);
5
(m);
6
(m)
_

_
_
_
_
_
_
_
_
_
1
0
0
0
0
0
_
_
_
_
_
_
_
_
_
Then, with the knowledge of L(m) (matrix 6 6), one can determine the homogeneous
behavior of the RVE. For any

the local deformation solution upon the RVE equals to:


(m) = L(m)

(1.3)
That implies, assuming a linear constitutive law (m) = D (m) :
(m) = H(m)

(1.4)
With : H(m) = D(m) L(m)
Then one can write the local deformation energy as:
e
local
d
=
1
2
(m)
t
(m)
And energy equivalence as:
1
2

_

t
L(m)
t
D L(m)

_
d =
1
2
vol ()

(1.5)
Finally, the homogeneous constitutive matrix is written as follows

D =
_
L(m)
t
D L(m)
_
(1.6)
Where =
1
vol()
_

d has been used.


report.pdf 11 August 11 2008
Crack propagation in a three dimensional FRC unit cell year 2008
1.4 Nonlinear modeling
Consider the mechanical problem schematized in gure 1.3:
Figure 1.3: General simple mechanical problem
and dened by the following conditions of admissibility:
Stress admissibility
is a symmetric matrix
div +f
d
= 0 in all end n = F
d
upon
2

Displacement admissibility
u = U
d
in
1

1.4.1 Linear case


Let us consider the linear constitutive law (engineering notations)
= D
and the linear strain-displacement relationship (Finite Element Method (FEM))
= B u
where u is the vector of nodal displacements.
Then one can write the internal forces as follows
q
int
=
_

B
t
d
i.e.
q
int
=
_

B
t
D B ud
As u is independent from x,y,z, one can write
q
int
=
_

B
t
D Bd u
Where
_

B
t
D Bd is the stiness matrix of our problem.
Finally we have the quasi static equilibrium equation
K u = q
ext
Where K is the tangent stiness matrix end q
ext
the external forces.
report.pdf 12 August 11 2008
Crack propagation in a three dimensional FRC unit cell year 2008
1.4.2 Nonlinear case
Let us consider the nonlinear constitutive law:
= ()
In addition we assume that the strain-displacement relationship remain the same as before,
i.e.
= (B u)
Accordingly, the internal forces have the following expression:
q
int
=
_

B
t
(B u)d
Then we introduce a residual function g = q
int
-q
ext
. This function is useful in order to use
an iterative method (Newton-Raphson, arclength, etc...). Indeed it allows us to estimate the
error committed and to check the convergence of the problem.
Now, let us consider the derivative of g with respect to u
dg
du
=
dq
int
du
=
_

B
t
d
d
Bd
The matrix
_

B
t d
d
Bd = K
t
is the tangent stiness matrix of the nonlinear problem.
This kind of tangent stiness matrix is then computed at each iteration (or increment).
1.4.3 Newton-Raphson method
Considering a one-dimensional problem, for a generic increment the force will change from an
initial value q
exti
to the the value at the end of the increment, q = q
exti
+q. Let us denote
u
0
the value of the displacement at the end of the increment. The equilibrium equation is
generally not satised and a residual g is computed, g = f(u) q
ext
= 0. Then the Taylor
series gives:
y = g(u
0
) +
dg
du
(u
0
)(u u
0
)
and y=0 leads to:
q = u
0

dg
du
(u
0
)
1
g(u
0
) if
dg
du
(u
0
) = 0
Then, if we set
u
0
=
dg
du
(u
0
)
1
g(u
0
) = K
1
0
g(u
0
)
the next solution u
1
will be written as
u
1
= u
0
+u
0
This procedure is then iterated until u
i
is suciently small.
The error committed with this approximation is of the order of u
2
i
(Taylor series expanded
up to the rst order).
This Newton-Raphson method is called full because we calculate at each iteration a
new tangent stiness matrix. Some simplied method are frequently used where the tangent
stiness matrix is computed for the rst iteration only.
report.pdf 13 August 11 2008
Crack propagation in a three dimensional FRC unit cell year 2008
1.4.4 Dynamic explicit method
The dynamic explicit method proposed in Abaqus is based on a central-dierence operator. Let
us consider an increment at the time t. First, the dynamic equilibrium equations are satised
at the beginning of the increment. Then the accelerations at time t are used to advance the
velocity to time t + t/2 and, nally, the displacement to time t + t.
Practically, the accelerations at the beginning are computed as follow:
u
N
i
=
_
M
NJ
_
1

_
P
J
i
I
J
i
_
With M
NJ
the diagonal (lumped) mass matrix, P
J
i
is the applied load vector and I
J
i
the
internal force vector at the increment i.
Remark: I
J
is assembled from the contributions of each elements such that a global stiness
matrix need to be formed. However, the explicit procedure requires no iterations and no
tangent stiness matrix which leads to a high eciency of the procedure.
The velocities and displacements are then computed as:
u
N
i+
1
2
= u
N
i
1
2
+
t
i+1
+ t
i
2
u
N
i
u
N
i+1
= u
N
i
+ t
i+1
u
N
i+
1
2
The main disadvantage of an explicit analysis is that the global equilibrium is never checked,
so that an error is inevitably introduced. As a consequence, in order to keep this error within
an acceptable range, very small time increments are necessary. In particular, a maximum time
increment is evaluated, which is equal to, or an estimate of, the time needed by a dilatational
wave to cover the smallest element of the mesh.
report.pdf 14 August 11 2008
Chapter 2
Cohesive elements in Abaqus
The nite-element code Abaqus, which has been used in this project includes special cohesive
elements to model adhesive joints or bonded interfaces taking into account degradation of
material properties. This model is widely based on the work of PP. Camanho and C.G D avila
[12] and [13].
It is possible to dene three types of responses of these elements: continuum, traction/
separation and gasket. In our particular case, as we are dealing with interfaces between bres
and matrix, the constitutive thickness of the elements is basically very close to zero. This
directly implies the use of cohesive element behavior dened in terms of traction separation
law which allows us to:
- Model delamination at interfaces
- Specify materials data (as fracture energy) as a function of the ratio of normal to shear
strength at the interface
- Characterize failure by progressive degradation of the material stiness, which is driven by
damage process
- Take into account multiple damage mechanisms.
Moreover, cohesive elements are fully nonlinear (nite strain and rotation) and can have
mass so that they can be used in dynamic analysis (i.e. they are available in Abaqus/Standard
and Abaqus/ Explicit).
Failure mechanisms can generally be modeled in two steps:
- First : A damage initiation criterion
- Second : A damage evolution law.
In Abaqus, a third step is used with the possibility to remove elements upon reaching a
complete damage state.
2.1 Cohesive elements
In this section, a description of the three dimensional cohesive elements available in Abaqus
[11] is given.
The functions N
k
used to interpolate displacements from the nodal ones are standard
lagrangian shape functions. We can write (in the [, , ] coordinate system) those functions
for the 8-nodes cohesive element as :
N
1
= ( 1)( 1)(1); N
2
= ( 1)(1); N
3
= (1); N
4
= ( 1)(1);
15
Crack propagation in a three dimensional FRC unit cell year 2008
Figure 2.1: 3D cohesive elements
N
5
= ( 1)( 1); N
6
= ( 1); N
7
= ; N
8
= ( 1)
The main dierences between cohesive elements (dened in terms of traction separation
law) and classic nite elements is that they are supposed to symbolize interfaces. Quantities,
therefore, which represent the internal state of stress are no longer stress a tensor but a vector
of normal t
n
, tangential t
t
and shear t
s
components (gure 2.2).
Figure 2.2: Denition of stress quantities along the interface
The denition of these quantities implies the use of a particular direction for each element,
called direction of through-thickness behavior, along which t
n
is aligned. Obviously, this
direction is perpendicular to the interface. Practically, one needs to construct a bottom and
a top face of the element. To do that, a particular meshing method called sweep mesh -
advancing front is used. The idea is to dene a rst surface which will be meshed (two
dimensional mesh). Then this mesh will be extruded to the top face.
This procedure allows us to obtain an element dened as shown in gure 2.2.
These particularities of cohesive elements in Abaqus imply, in the visualisation step, to
reconsider the signication of the stress values [
33
;
32
;
31
] and the corresponding strains
[
33
;
32
;
31
]. Indeed we have no longer stress and strain tensor but vectors of stresses and
relative displacements at the interface. That leads to identify [
33
;
32
;
31
] with [t
n
; t
s
; t
s
]
and also [
33
;
32
;
31
] with [
n
;
s
;
t
], where
i
, i = n, s, t is the relative displacement at the
interface (or separation displacement).
report.pdf 16 August 11 2008
Crack propagation in a three dimensional FRC unit cell year 2008
2.2 Initial behavior and damage initiation criteria
2.2.1 Initial behavior
Initially, the interface behavior is dened by a linear elastic traction separation law. This law
is written in terms of an elastic constitutive matrix K that relates the nominal stress t to the
nominal strain across the interface:
t =
_
_
_
E
nn
E
ns
E
nt
E
ns
E
ss
E
st
E
nt
E
st
E
tt
_
_
_

_
_
_

t
_
_
_
= E = K =
_
_
_
K
nn
K
ns
K
nt
K
ns
K
ss
K
st
K
nt
K
st
K
tt
_
_
_

_
_
_

t
_
_
_
Where the corresponding strains
n
,
s
and
t
are derived from displacements on the inter-
face and from its thickness as follows :

n
=

n
T
0
;
s
=

s
T
0
;
t
=

t
T
0
Remark 1: The constitutive behavior presented above is a coupled law. To obtain an un-
coupled one it is sucient to dene the o-diagonal terms of K as equal to 0.
Remark 2: T
0
is, in Abaqus, an articial constitutive thickness of the cohesive element
which is, by default, dened equal to 1. That allows us to assume that, initially, the nominal
strains equal the separation displacements (i.e. = ).
Remark 3: Nevertheless, the real constitutive thickness T
c
is used to dene the stiness, as
proposed by L. Daudeville, O. Allix and P. Ladev`ese [14] :
K
nn
=
E
33
T
c
, K
ss
=
2G
13
T
c
and K
tt
=
2G
23
T
c
.
K
ij
i, j = n, s, t are users dened values.
As the real thickness is close to zero, K is associated to a penalty stiness. One needs to
dene carefully this value because a very large penalty stiness is detrimental to the stable
time increment and may result in ill-conditioning of element operation.
Remark 4: To have a positive dened matrix one might consider the condition that :
K
ii

_

i=j
K
2
ij
2.2.2 Damage initiation criteria
Damage initiation criteria are based on the schematic traction-separation response (shown
in gure 2.3) for the uncoupled delamination mode [11]. (The rst linear part is explicated
subsection above.)
One enters the second linear part once the damage criterion used is met.
The criteria proposed in Abaqus are :
- Two criteria based on stress ratio, i.e.
t
i
t
0
i
called Maxs and Quads damage criterion.
- Two criteria based on strain ratio, i.e.

i

i
0
with i = [n, s, t] called Maxe and Quade damage
criterion.
They are written as follow:
report.pdf 17 August 11 2008
Crack propagation in a three dimensional FRC unit cell year 2008
Figure 2.3: Typical traction-separation response
max
_
t
n

t
0
n
;
t
s
t
0
s
;
t
t
t
0
t
_
= 1
. .
MaxsDamageCriterion
;
_
t
n

t
0
n
_
2
+
_
t
s
t
0
s
_
2
+
_
t
t
t
0
t
_
2
= 1
. .
QuadsDamageCriterion
(2.1)
And
max
_

0
n
;

s

0
s
;

t

0
t
_
= 1
. .
MaxeDamageCriterion
;
_

0
n
_
2
+
_

0
s
_
2
+
_

0
t
_
2
= 1
. .
QuadeDamageCriterion
(2.2)
Remark: . symbolize the fact that only positive value are taken, i.e. pure compressive
stress or deformation does not initiate damage.
2.3 Softening onset prediction (Mixed-mode)
In the case of a mixed mode delamination [12] and [13] propose a formula to obtain the value
of the total displacement corresponding to
0
n
(delamination in mode I only). A demonstration
of this formula is proposed below.
In mixed mode one might dene the total displacement at the interface as:

m
=
_

2
+
2
s
+
2
t
(2.3)
Let us consider the quadratic damage initiation criteria
_
t
n

t
0
n
_
2
+
_
t
s
t
0
s
_
2
+
_
t
t
t
0
t
_
2
= 1 (2.4)
and assume that K
nn
= K
ss
= K
tt
= K. Immediately we have the fact that t
i
= K
i
and
t
0
i
= K
0
i
, i = n, s, t.
Then it leads to:
_

0
n
_
2
+
_

0
s
_
2
+
_

0
t
_
2
= 1 (2.5)
report.pdf 18 August 11 2008
Crack propagation in a three dimensional FRC unit cell year 2008
Case 1:
n
= 0:
In the paper, the authors proposed to consider that
0
s
=
0
t
and to introduce the mode
mixity ratio as:
(
n
= 0)
_
=
_

2
s
+
2
t

n
_
(2.6)
so that we can rewrite (2.5) as:

2
n

_
1
(
0
n
)
2
+
_

0
s
_
2
_
= 1 (2.7)
With (2.7) one can express the value of
2
n
which corresponds to the mixed mode initiation:

2
n
=
_

0
n
_
2

0
s
_
2

1
(
0
s
)
2
+ (
0
n
)
2

2
(2.8)
Indicating by

n
the value of
n
corresponding to damage initiation in a mixed mode
decohesion:

n
=
0
n

0
s

1
(
0
s
)
2
+ (
0
n
)
2

2
(2.9)
we can see that generally
0
n
=

n
, but in the case where
2
s
+
2
t
= 0, and then = 0 we
are able to retrieve that
0
n
=

n
, which is the onset softening in mode I only.
With this value of

n
, the mixed mode relative displacement corresponding to the onset
softening is:

0
m
=
_

2
n
+

2
s
+

2
t
(2.10)
According to the denition of :
=
_

2
s
+
2
t

n
=
_

2
s
+

2
t

n
(2.11)
Then:

0
m
=

_
1 +
2
(2.12)
Finally the result is:

0
m
=
0
n

0
s

1 +
2
(
0
s
)
2
+ (
0
n
)
2

2
(2.13)
Case 2:
n
= 0.
By (2.5) and the assumption
0
s
=
0
t
one obtains immediately that:
_

s
2
+
t
2
=
0
s
=
0
m
(2.14)
Hence we retrieve the formula given in [12].
report.pdf 19 August 11 2008
Crack propagation in a three dimensional FRC unit cell year 2008
2.4 Delamination propagation prediction and damage evolu-
tion law
The criteria used to predict delamination propagation under mixed-mode loading conditions
are usually established in terms of the energy release rates (G
I
; G
II
; G
III
) and fracture
toughness (G
IC
; G
IIC
; G
IIIC
).
In an uncoupled law the nal displacements (or displacement at failure) are obtained as

f
n
= 2G
IC
/t
0
n
,
f
s
= 2G
IIC
/t
0
s
,
f
t
= 2G
IIIC
/t
0
t
.
However, in mixed-mode, an interaction between the energy release rates appears. Hence,
two predictive criteria are dened as:
_
G
I
G
IC
_

+
_
G
II
G
IIC
_

= 1
. .
PowerLawCriterion
; G
IC
+ (G
IIC
G
IC
)
_
G
shear
G
T
_

= G
C
. .
BKCriterion
(2.15)
With G
shear
= G
II
+G
III
and G
T
= G
I
+G
shear
; equals 1 or 2; [0.5, 2].
Then, the energy release rates corresponding to the total decohesion are obtained from:
G
I
=
_

nf
m
0
t
n
d
n
, G
II
=
_

sf
m
0
t
s
d
s
, G
III
=
_

tf
m
0
t
t
d
t
(2.16)
Using the irreversible, bi-linear, softening constitutive behavior (which summarize the dif-
ferent steps exposed in this chapter) :
t
i
=
_

_
K
i

max
i

0
i
(1 d
i
)K
i

0
i
<
max
i

f
i
0
max
i

f
i
(2.17)
and the linear softening law:
d =

f
m
_

max
m

0
m
_

max
m
(
f
m

0
m
)
, d [0, 1] (2.18)
also than (2.3) and (2.6) in equations (2.16 a., b. and c.) and substituting in (2.15.a) or in
(2.15.b), the criterion for total decohesion can be established in terms of
m
and . Solving
the equation for
m
, the mixed-mode displacements corresponding to a total decohesion,
f
m
,
are obtained for the power law criterion as:

f
m
=
_
_
_
2(1+
2
)
K
0
m
__
1
G
IC
_

+
_

2
G
IIC
_

_
1/

n
> 0
_
(
f
s
)
2
+ (
f
t
)
2

n
0
(2.19)
and for the B-K criterion as:

f
m
=
_
_
_
2
K
0
m
_
G
IC
+ (G
IIC
G
IC
)
_

2
1+
2
_

n
> 0
_
(
f
s
)
2
+ (
f
t
)
2

n
0
(2.20)
Remark: An exponential damage evolution law is proposed in Abaqus which leads to a
similar denition of
f
m
.
report.pdf 20 August 11 2008
Chapter 3
Representative volume element
characterization
The peculiarity of our study is that we consider a ber reinforced composite with one cracked
bre. That is why one might neither use a RVE with just one ber (gure 3.1) nor a two
dimensional one. Indeed, the aim is to model the propagation of cracks and the redistribution
of the stress previously carried by the unbroken ber to the matrix and the other bers of the
composite. Furthermore, the stress concentration which exists around the crack is subject to
move along the three axes when debonding phenomenon or damage appears.
Figure 3.1: Classical RVE for bre reinforced composite
As a rst step of modeling this type of behavior, we concentrate ourselves on the aim to
proof that such studies are possible and accurate (or not) with a software as Abaqus. That is
why we are beginning with a relatively simple RVE which is subject to change regarding to
the matrix modeling, the number of bres and their location, some size criterion or even the
crack shape and location.
3.1 Size of the RVE, number of bres and location of the crack
Many studies have been made to establish a critical size of RVE with criteria like Hill condition,
Eective properties or Coecient of correlation [15]. Nevertheless the main information which
is of interest for us is the global magnitude of the bers radius, according to the aim expressed
above. (For example, the radius of AS4 carbon bers is approximately 2.5m.)
From this information, one of the simplest case which takes into account the redistribution
of stress in the matrix and some bers can be thought as a cube with one central broken bre
and four quadrant cylinders [gure 3.2].
21
Crack propagation in a three dimensional FRC unit cell year 2008
Figure 3.2: Representative Volume Element
Concerning the location and the shape of the crack we will assume that:
- the central bre is totally broken
- the broken bre can be modeled as two perfect cylinders (crack shape dened as z = C
te
)
- Those two cylinders have the same length (the crack appears at the very middle of the
central bre)
The last assumption derives straightly from the hypothesis of a RVE, because of the in-
nitesimal signication of it. Concerning the rst two assumptions, it is clear that alternative
modeling could have been chosen. For example, the shape of the crack can be randomly chosen
as well as the average of the broken part of the central ber.
Remark: One can consider that in a longitudinal load, we could have used the symmetries
of the problem to simplify the model. But, even if in this step we are applying this type of
load, the idea is to build, in the long term, a RVE able to support every kind of loads. Fur-
thermore, the assumption made of representativeness of the whole structure implies to keep a
cubic (or a cobble) shape.
3.2 Boundary conditions
The boundary conditions that we are applying to the unit cell are very similar to those used
in kinematic homogenization, i.e. : u
d
=

OM.
Then for the particular case of longitudinal load we have, with respect to the coordinate
system used in our Abaqus model:

=
_
_
_
0 0 0
0 0 0
0 0 1
_
_
_
Then:
u
d
= [0, 0, ] on
top
, and u
d
= [0, 0, ] on
bottom
()
Furthermore, kinematic homogenization boundaries leads to dene:
u
d
= [0, 0, z] on
Slat
1


Slat
2

report.pdf 22 August 11 2008


Crack propagation in a three dimensional FRC unit cell year 2008
The other possibility is to impose, for every lateral surfaces (apart from the top and bottom
ones, where () remains valid), that the displacement following e
1
and e
2
of every nodes are
the same, i.e.:
(q
n
, q
m
)
Slat
i

, u(q
n
) e
i
= u(q
m
) e
i
, i = 1, 2
3.3 Materials and interfaces
In the very rst step of the modeling we have used an isotropic linear elastic behavior for the
matrix and bres. The materials and their properties are:
Fibre Carbon (AS4) : E = 210000MPa and = 0.3
Matrix Epoxy (914) : E = 10000MPa and = 0.2
Moreover the unit cell count seven interfaces (six between the matrix and the bres and
one between the two parts of the broken central bre). The rst six are modeled by cohesive
elements and the seventh by frictionless contact.
Remark 1: One can see that from the rst increment of the calculus (in the case of a lon-
gitudinal pull load), there is no longer contact between the two parts of the central bre. In
this sense, the seventh interface is not taken into account for the calculus.
Remark 2: Concerning the cohesive element properties, as observed earlier (Chapter 2),
one has to specify the characteristics used to dene the fracture initiation criteria and the
damage evolution law, i.e. the constitutive matrix K, the vector of relative displacement at
softening oset
0
and the interlaminar fracture toughnesses G
I
, G
II
, G
III
In the case of a longitudinal pull load, cohesive elements are stressed in both mode I and
mode II (opening and sliding). Indeed, sliding is given by both crack and dierences of Youngs
modulus, and opening by dierences of Poissons coecients (the presence of the crack changes
the values of mode I relative displacements compared with the uncracked unit cell).
Even if mode III is theoretically inexistent, we can see that Abaqus gives us a (very low)
value of relative displacement corresponding to this mode.
Remark 3: For utility, the length unit used is the m, which leads to dene the Youngs
modulus in TPa. Accordingly, the stress values results will have to be reinterpreted.
report.pdf 23 August 11 2008
Chapter 4
Specicities of the modeling under
Abaqus
The construction of the three dimensional numerical model leads to some diculties we are
going to expose as well as the solutions we have proposed.
Indeed, the specicities of cohesive elements concerning geometrical, status and meshing
methods have to be treated with a particular care.
4.1 Contact status and mesh of interface elements
The constitutive thickness of interface elements is basically very thin as regards to the whole
model dimensions. In our case, we dened a thickness equals to 0.1m which leads to a ner
mesh than that of the matrix or the ber. This fact implies to give a slave status of the
cohesive elements in the tie (completely xed) interactions.
Furthermore, compatible meshes between cohesive and surrounding elements have to be
carefully constructed.
Moreover, the specic meshing method used to solve those problems for central interface
elements (where the most important issues appear) are as follows.
First, one needs to partition the cylinder in two parts and to dene a new meshing path,
as the automatic one leads to ill denition of bottom and top faces of the interface.
Figure 4.1: Partition and path denition
24
Crack propagation in a three dimensional FRC unit cell year 2008
Moreover, without partition no user dened path is allowed for cylindric bodies. Then the
proposed model leads to well dened elements as shown gure 4.2:
Figure 4.2: Bottom and top denition. 4.2.b: Local coordinates.
where the bottom face is colored in purple and the top in brown.
Remark: The local coordinate system linked with every cohesive elements is here : e
3
the
through-thickness direction, e
2
parallel for every element with the z global axis and e
1
= e
3
e
2
(where symbolize the vectorial product).
The gure 4.2.b shows the local coordinate systems for two elements C
n
and C
n+m
:
4.2 Partition of the matrix
Because of the relatively complex geometry of the matrix, the automatically computed mesh
is not satisfying. Then the partition of the matrix seems to be recommended.
To improve the quality of the mesh we performed the partition as shown gure 3.2, which
partition can be performed by the partition tool in the mesh module or, more simply, by
constructing the part in two times (the central cylinder and the remaining volume) with the
option of keeping internal boundaries.
report.pdf 25 August 11 2008
Chapter 5
Numerical simulation
The results shown in this chapter are mainly useful to prove the accuracy of a three dimen-
sional model for a mixed mode delamination. As we do not have the possibility to compare
them with experiments, we will focus on qualitatively regain the classical phenomena (stress
concentration, redistribution of stress from the bre to the matrix) which appear in cracked
reinforced composites under longitudinal pull load.
The materials properties used are exposed in chapter 3. Nevertheless, numerical values of
interface properties and boundary conditions are detailed below:
Properties of the interface:
Relative displacement values at onset softening
0
n
=
0
t
=
0
s
= 0.1m, which leads
(formula (2.13)) to
0
m
= 0.1m.
K is diagonal and K
nn
= K
ss
= K
tt
= 1GPa.
The interlaminar fracture energy G
I
= G
II
= G
III
= 5.10
4
kJ/m
2
, which correspond
to a displacement at failure equals to 1m.
Displacement control:
The boundary conditions are: U
d
= [0, 0, 2m] on
top
, U
d
= [0, 0, 2m] on
bottom
and U
d
= [0, 0, 0] on every lateral faces.
Where:
=
_
_
_
1 0 0
0 1 0
0 0 0
_
_
_
5.1 Stress concentration
Firstly, the numerical simulation allows us to follow the displacement of the stress concentration
from the outskirts of the crack and then along the central bre outline.
Figure 5.1 gives an idea of the phenomenon. As recalled in chapter 2, the stress quantity

23
represent the longitudinal shear stress upon the interface (linked with Mode II).
Remark: In the case of stronginterface between bre and matrix, the stress concentration
can not move, as it happens here, from the outskirts of the crack to the bottom and top face
along the interface between the crack bre and the matrix. Indeed this phenomenon supposes
the degradation, up to failure, of the cohesive elements where the stress concentration was
26
Crack propagation in a three dimensional FRC unit cell year 2008
Figure 5.1: Longitudinal shear load / S23 - Step 1, 10 et 20
previously located. To simulate the displacement of stress concentration in this next case, one
need to model the matrix as a damageable material.
5.2 Damage evolution
In the corresponding in gure 5.2 of damage values in cohesive elements, the location of the
stress concentration widely depends on the interface softening.
The SDEG value is the value of D in the relation
i
= (1 D) K
ii
t
i
. Then SDEG = 1
(red in the color classication) signify the fact that the materials are no longer bonded.
We can see that mode II and mode I occurred in the simulation. Then one need to be
careful in the interpretation of the stress/displacement curve for one cohesive element.
Figure 5.2: Damage variable at Step 13. 5.2.b: Strain/Stress curve in mode II
For one element located near the crack , the gure 5.2.b show the strain/stress curve in
mode II.
The strain/stress curve corresponding to mode II is decomposed, as expected, in:
0
s
< 0.098m the linear initial behavior
0.098
s
< 0.98m the linear softening part
0.98 <
s
the fully damaged behavior
Remark: If we mark, as proposed chapter 2,

s
= 0.098, i.e. the value of
s
at onset
softening in mixed mode delamination, we obtain that
_

n
2
+

s
2
+

t
2
= 1.
report.pdf 27 August 11 2008
Crack propagation in a three dimensional FRC unit cell year 2008
5.3 Global response of the unit cell - comparison of Abaqus/Standard
and Abaqus/Explicit
The mean stress (of
33
) on the central ber (top face) versus the displacement is presented
in gure 5.3 for several types of resolution. The idea is to compare the results given by
Abaqus/Standard and Abaqus/Explicit for the same problem. In chapter 1, section 4, we have
explained the main dierences between implicit and explicit resolution methods. It is now the
place to show how both methods are available for our study.
Indeed, the use of an explicit method take into account the density of materials (to build
the mass matrix). However, the fact that the length unity used is the micro meter leads to
dene the density in 10
18
Kg/m
3
. Then, the time increment used for the explicit resolution
tends to zero as :
t =
L
min
c
d
= L
min

+ 2
In our case, L
min
1 for the coarse mesh, 1.10
18
Kg/m
3
, E = 0.21TPa and = 0.3.
Thus t 1.10
9
, which implies to compute one million iterations to solve the simulation.
This trouble leads to use a pseudo mass, which is an arbitrary value of the density. Because
of the quasi-static problem we are solving, this procedure is frequently used. However, it is
important to be sure that no dynamic eects are appearing during the process.
Figure 5.3: Stress-displacement curve of the central bre (top face) for implicit and explicit
method
The rst curve (serie 1) is the result of the Abaqus/Standard resolution. Series 2, 3 , 4
and 5 are respectively the resolution of the same mechanical problem under Abaqus/Explicit
with = 1.10
10
, = 1.10
8
, = 1.10
7
and = 1.10
6
. Moreover, their time increments
are respectively t = 2.10
5
, 2.10
4
, 7.10
4
and 2.10
3
.
Dynamic eects inter into account in the last two series, where oscillations appears. This
gives us the condition to use a value of 1.10
8
for a coarse mesh (L
min
1). However,
even with a rened mesh we are going to use this condition.
Remark 1: One identify time and displacement in our case as we applied displacement of
the top and the bottom face linearly in function of the time.
report.pdf 28 August 11 2008
Crack propagation in a three dimensional FRC unit cell year 2008
Remark 2: The load carried by the central ber is clearly dependent from the level and the
number of damaged elements of interface. Obviously, the load carried when all the cohesive
elements are fully damaged tends to zero.
Remark 3: This phenomenon leads to a redistribution of the stress in the matrix. Hence,
the stress/displacement curve for an element of the matrix close to the central bre and to the
top face is shown (gure 5.5) .
Figure 5.4: Stress/displacement in an element of the matrix
5.4 Discussion
The crack propagation in bre reinforced composites is theoretically allowed to expand in the
three space dimensions. Several types of failures are commonly listed as: opening of the matrix,
debonding at the interfaces, failure of bres which are surrounding the crack. Obviously, a
real failure of the material can be a combination of several failure types.
In this numerical simulation, we tried to qualitatively represent sliding between bre and
matrix. The stress concentration which is theoretically predicted has been shown. Further-
more, behavior of interface elements accurately correspond to the results expected.
In that sense, this simulation validates the use of cohesive elements in a three dimensional
problem involving a bre reinforced composite unite cell.
However, it could be wise to:
Dene a bigger RVE which involves more bres to reduce the eects of the boundary
conditions on the propagation of the crack.
Have a damage matrix model to allow the crack to expand in three dimensions. Then
it could be possible to consider the competition between debonding along the interface
and opening of the matrix.
In the next chapters, we are going to expose the main tools and the work related to those ends.
report.pdf 29 August 11 2008
Chapter 6
Damage modeling
6.1 Some damage modelings and their deciencies
Damage appears in materials as a consequence of the presence of voids. Their growth and
coalescence under loading is characterized by localized cracks and softening behavior.
In function of the type of material, the plastic strain at onset softening and the rate of loss
of load-carrying capacity is more or less important. That is the reason why several kinds of
damage modeling have been developed.
But even for a quasi-brittle material (for example), computational models can roughly be
separated into three categories :
- Micromechanical models where materials are modeled by lattice structures or braces of
constituents and interfaces. However, because of the high level of detail considered, these
methods lead to large computational costs.
- Damage modeled by dominant macroscopic crack, using the concept of cohesive zones,
is also proposed. Its main disadvantage consists in an extensive remeshing of the structure to
follow the propagation of the crack.
- The third damage model is based on continuum mechanics and the use of a damage quan-
tity dened at each points of the structure. This method leads to a pathological dependence
on the neness of the spatial discretisation as well as on its orientation. Mathematically this
deciency is related to a loss of ellipticity of the equilibrium equation. Owing to the local tran-
sition of these partial dierential equations (elliptic to parabolic and nally hyperbolic) the
tangential material stiness change of sign. Then the mechanical problem becomes ill-posed.
Practically, this deciency leads to have a localized damage, with the consequence that the
nner the mesh is, the smaller the energy needed to damage the material becomes.
In the next section, we are going to briey expose some modied (isotropic) damage models
which were designed to avoid this deciency.
6.2 Gradient-based damage modelings
6.2.1 Method based on the principle of virtual work
One of the proposed methods to avoid this problem is based on the rewriting of the principle
of virtual work taking into account the damage quantity and its gradient [16].
Then the power of internal forces and the power of external forces are written as:
30
Crack propagation in a three dimensional FRC unit cell year 2008
P
i
_
u,
d
dt
_
=
_

: D(u) d
_

_
B
d
dt
+H grad
_
d
dt
__
d
P
e
_
u,
d
dt
_
=
_

f u d +
_

F u d +
_

A
d
dt
d +
_

b
d
dt
d
With:
(X, t): the macroscopic damage quantity
B: the internal work of damage
H: the ux vector of internal work of damage
A, b: the volumetric and surfacic external source of damage work
Then for a quasi-static problem one obtains that:
, v, , P
i
(v, ) +P
e
(v, ) = 0
6.2.2 Gradient-enhanced damage
An isotropic damage model (which is implemented in CAST3M) has been developed by R.H.J.
Peerlings et al. in [17-19]. The idea is to link the classical constitutive law for damage (6.1)
to a set of equations ( and ) dening the new mechanical problem.
= (1 D) H (6.1)
div +f = 0 ()

eq
c
eq
=
eq
()
With:

eq
: the local equivalent strain dened at each point of the body

eq
: the corresponding non-local equivalent strain
c: the constant gradient parameter with the dimension of length square
The link between D and
eq
is performed by the fact that D = D(), where is an historic
parameter which appears in the Kuhn Tucker relations as below:
0 ,
eq
0 , (
eq
) = 0
After the establishing of the weak forms of () and () (for the boundary conditions
n = F
D
on
2
and
eq
n = 0 on ), the iterative method is initiated by the dierentiation
of (6.1):
= (1 D) H D H (6.2)
The Kuhn Tucker relation gives us that, for the increment i,
i
=
eq,i
if =
eq
or

i
= 0 otherwise. Furthermore, the linearization of D() gives us that D = (D/) .
Then the weak formulation of () can be written, for the increment i, with
eq,i
and u
i
as
only unknown.
Finally, the new nite element problem to solve is written as:
report.pdf 31 August 11 2008
Crack propagation in a three dimensional FRC unit cell year 2008
_
_
K
uu
i1
K
u
i1
K
u
i1
K

_
_

_
u
i

eq,i
_
=
_
f
u
ext
f

i1
_

_
f
u
int,i1
K


eq,i
_
Remark 1: As K
u
i1
= K
u
i1
, the generalized stiness matrix is non-symmetric.
Remark 2: The iterative method proposed by the authors is based on the idea that for a
quantity , its value at the increment i is :
i
=
i1
+
i
.
6.3 Interpretation and tools used
One can see that an advanced damage model (even for an homogeneous isotropic material)
leads to deeply change the mechanical problem to solve. Moreover, the underlying idea of the
models above is that non-local eects are acting at each point of the material. Then those
models are questioning the fact that stress and strain are locally dened.
The damage modeling (proposed in Abaqus/Explicit) we are going to use, simply em-
ploy the microscale characteristic length of the material to reduce the mesh dependency phe-
nomenon. Thus, the concept of equivalent plastic displacement is dene as:

u
pl
= L


pl
(6.3)
For solid elements, L is taken as the cube root of the integration point volume.
One can see that damage modeling is very close to the one applied for cohesive elements
degradation. Indeed, damage occurs after that a specied criterion has been reached and the
very same softening laws (linear or exponential) are available.
However, this model is called plastic-damage because of the fact that the damage occurs
after yielding of the material. This type of modeling is drawn gure 6.1 [11].
Figure 6.1: Plastic-damage model under Abaqus
Remark: The fact that the damage models are available for Abaqus/Explicit only is the
reason why most of the simulations with this resolution method also than the developments
given concerning the theory of explicit procedures (chapter 1) and its validity using a pseudo-
mass (chapter 5 and 7).
report.pdf 32 August 11 2008
Chapter 7
Numerical simulation with a matrix
damage model
In this chapter, we are mainly using the boundary conditions and material properties dened
previously, apart from the parameters of the Plastic-Damage model of the matrix.
7.1 Parameters of the Abaqus damage model
The previous chapter gives us an idea of the dierent types of damage modeling. Practically,
the Plastic-Damage model under Abaqus leads to dene some specic parameters which are
the Yield Stress, the Fracture Strain, the Deformation at Fracture (or Fracture Energy) and
the Fracture ratio.
Figures 7.1 and 7.1.b show the inuence of the Fracture Strain and Deformation at fracture
for a simple example (one 3-dimensional element xed on one side and pulled on the opposite
side) for a Yield Stress given. The Fracture Ratio is not taken into account because the results
are insensitive to its changes.
Figure 7.1: Variation of damage model parameters
Those two parameters can be considered as means to adapt the damage model for the type
of material used. A small fracture Strain gives a damage model well tted for a quasi-brittle
type of material and a large one to a ductile material.
33
Crack propagation in a three dimensional FRC unit cell year 2008
7.2 Simulations for the Small Unit Cell
Mesh of the small unit-cell:
Figure 7.2: Mesh of the small Unit Cell
Parameters of the damage model:
Fracture Strain = 1 ; Yield Stress = 100MPa ; Fracture Energy = 1.10
5
kJ/m
2
With those parameters we are obtaining the case where the crack is propagating inside the
matrix to lead to its opening.
Figure 7.3: Values of the damage initiation criterion - opening of the matrix
For this simulation, the dynamics eects linked with the use of a pseudo-mass are controlled
using two values of (gure 7.4).
Figure 7.4: Dynamic eects
Hence we can see the dierences of results between these two simulations for some particular
quantities (evolution of the normal stress in a cohesive element and evolution of the average
report.pdf 34 August 11 2008
Crack propagation in a three dimensional FRC unit cell year 2008
of normal stress upon the central bre top face). Figure 7.4, serie1 correspond to = 7.10
8
and serie2 to = 7.10
10
.
Remark : The other case (sliding at the interfaces without failure of the matrix) is easily
obtained in changing the parameters of the Plastic-Damage model or the properties of the
interface.
7.3 Simulations for the Bigger Unit Cell
Mesh of the small unit-cell:
To have a more accurate modeling of the failure, a bigger unit-cell has been build. Thus,
thirteen bres are involved and the dimensions of the RVE become 40 40 40m. Figure
7.5 and 7.5.b show this unit-cell and the mesh used for the simulations.
Figure 7.5: Bigger unit-cell and its mesh
With this representative volume element, several simulations have been made for dierent
values of Fracture Strain which allows us to obtain the case of debonding of the interface and
the opening of the matrix. These two type of failure are shown gures 7.6 and 7.6.b.
Figure 7.6: Values of the damage initiation criterion - a. opening of the matrix - b. debonding
at the interface
report.pdf 35 August 11 2008
Crack propagation in a three dimensional FRC unit cell year 2008
7.4 Discussion
The prediction of failure and its propagation in a composite material is frequently based
on Linear Elastic Fracture Mechanics (LEFM). In this simulation we have proposed a nite
element approach of this phenomenon, at the micro scale. This allowed us to represent two
types of fracture depending to the properties of the materials and interfaces.
However, one of the main questions of this project can be expressed as : is Abaqus accurate
to model this unit cell and the phenomena of crack propagation? This question can now be
answered in three principal view angles:
The possibility to use special types of interface elements, called cohesive elements (Abaqus/
Standard and Abaqus/Explicit), to represent debonding at the interface between bre
and matrix, accurately model the physical phenomenon. This model is also very close
to other recents works as Alfanos ones [20, 21].
However, some work has been done to implement the model presented in [21], where the
friction at the interface is added to the damage model in the cohesive zone. Regrettably,
we did not have the time to solve all the troubles linked with the implementation of the
model as a user subroutine Abaqus/Explicit. Although some existing user-subroutines
implemented for Abaqus/Standard (implicit) were available, the implementation of the
code for Abaqus/Explicit resulted in some error at run-time, which was not possible to
eliminate in the little time available. That is why no results with this Friction-Damage
Model are presented in this report.
The fact that we are modeling the unit-cell at the micro scale leads to a stunning diculty.
Indeed, the use of a damage model implies to run the computation under Abaqus/Explicit
and then to dene the density of the materials. Thereby, the time increment (which
represent that in one increment a mechanical wave cover the smallest element of the
mesh) tends dramatically to zero.
As proposed chapter 6, it is possible to use a pseudo mass in order to avoid this diculties.
Nevertheless, a particular study is necessary to establish the validity of the results.
Finally, we can estimate that the damage models proposed under Abaqus are quite
insucient. Indeed, the mesh dependency troubles are solved by the use of characteristic
length of the material, without taking into account the gradient of the damage quantity.
It seems that the use of a more advanced model, as the Gradient-Enhanced Damage
Model could improve the accuracy of the simulations.
report.pdf 36 August 11 2008
Conclusion
In this project we have simulated the propagation of fracture in a Representative Volume Ele-
ment, using an industrial nite element software. The simulation of dierent type of fracture,
opening of the matrix, delamination at the interfaces and a combination of both types, have
been shown.
To this end, two types of RVE have been built (ve bres and sixteen bres involved re-
spectively). Special interface elements and a damage model proposed in Abaqus were used.
Furthermore, we began the implementation of a Friction-Damage Model to improve the be-
havior of the Cohesive Elements.
Along this report, several special studies have exposed the background of the project (Fibre
Reinforced Composites and some recents works related, Multi-Scale approaches, Non-linear
modeling - chapter 1 - and, nally, some current damage models - chapter 6).
This work leads to several research lines. One could summarize them as:
- Complete the implementation of the Friction-Damage Model for the interface elements
in Abaqus/Explicit,
- Implement other better damage models or discrete-crack models (X-FEM) also for the
implicit analysis,
- Include a brittle damage model for the bre,
- Incorporate stochastic approaches concerning the position or/and the properties of the
bres,
- And, last but not least, investigate on the link between micro and macro scales, to develop
a multi-scale strategy.
Acknowledgments
I really thank Mr Giulio Alfano, lecturer and researcher at Brunel University, who guided me
during these three months research training period. I am also grateful to Mr Ali Bahtui and
Mr Fiorenzo Decicco who gave me their support and precious advice.
37
References
[1] Mallick. 1993. Fibers Reinforced Composites. Marcel Dekker, inc..
[2] Z. Xia, W. A. Curtin and P.W. Peters. 2001. Multiscale Modeling of Failure in Metal
Matrix Composites. Acta Materialia 273-287.
[3] N. Bonora, Andrew Ruggerio. 2006. Micromechanical Modeling of Composites Inter-
face Part 1 and 2. Composites Science and Technology 214-322.
[4] C. Gonz alez, J. LLorca. 2007. Mechanical Behavior of Unidirectional Fiber Reinforced
Polymers Under Transverse Compression Composites Science and Technology 2795-2806.
[5] J.-K. Kim and Y.-W. Mai. 1998, Engineered Interfaces in Fibre Reinforced Compos-
ites. Elvesier.
[6] M.G.D. Geers et al.. 2002. Perspectives in multi-scale modeling of composites and
heterogeneous materials upon localization. WCCM V July 7-12, Vienna, Austria.
[7] Sanchez Palancia. 1974.
[8] Kroner. 1965.
[9] O. Allix. Courses given at ENS Cachan. 2008.
[10] D. Violeau. Numerical analisys of delamination fatigue in composite via cohesive zone
model. 2001.
[11] Abaqus users manual
[12] PP. Camanho and C.G D avila. 2001. Mixed-Mode Decohesion Finite Elements for
the Simulation of Delamination in Composites Materials. NASA.
[13] C.G D avila et al.. 2001. Mixed-Mode Decohesion Elements for Analyses of Progressive
Delamination. Structural Dynamics and Materials Conference.
[14] Daudeville, L., O. Allix and P. Ladev`ese. 1995. Delamination Analysis by Damage
Mechanics : Some Applications. Composites Engineering 5(1):17-24.
[15] D. Trias et al.. 2006. Critical Size for a Statistical Representative Volume Element.
Acta Materialia 54 3471-3484.
[16] M. Fremond and B. Nedjar. 1996. Damage, Gradient of Damage and Principle of
Virtual Power. Int. J. Solids Structures Vol.33, No8. pp. 1083-1103.
[17] R. H. J. Peerlings et al.. 1996. Gradient Enhanced Damage for Quasi-Brittle Mate-
rials. Num. Methods in Eng., vol. 39, 3391-3403.
[18] M.G.D. Geers et al.. 1998. Strain-based transient-Gradient Damage Model for Failure
Analyses. Comput. Methods Appl. Engrg. 160, 133-153.
[19] R. H. J. Peerlings et al.. 1998. Gradient-enhanced Damage Modelling of Concrete
Fracture. Mech. Cohes.-Fric. Mater. 3, 323-342.
[20] G. Alfano and M.A. Criseld. 2000. Finite element interface model for the delamina-
tion analisys of laminated composites. Int. J. Numer. Meth. Engng; 50: 1701-1736.
[21] G. Alfano and Elio Sacco. 2006. Combining interface damage and friction in a
cohesive-zone model. Int. J. Numer. Meth. Engng; 68: 542-582.
38

S-ar putea să vă placă și