You are on page 1of 29

International Journal of Solids and Structures 39 (2002) 28272855

www.elsevier.com/locate/ijsolstr

Some issues in the application of cohesive zone models


for metalceramic interfaces
N. Chandra
a

a,*

, H. Li a, C. Shet a, H. Ghonem

Department of Mechanical Engineering, College of Engineering, Florida State University, Tallahassee, FL 32310, USA
b
Department of Mechanical Engineering, University of Rhode Island, Kingston, Rhode Island, USA
Received 29 November 2000

Abstract
Cohesive zone models (CZMs) are being increasingly used to simulate discrete fracture processes in a number of
homogeneous and inhomogeneous material systems. The models are typically expressed as a function of normal and
tangential tractions in terms of separation distances. The forms of the functions and parameters vary from model to
model. In this work, two dierent forms of CZMs (exponential and bilinear) are used to evaluate the response of interfaces in titanium matrix composites reinforced by silicon carbide (SCS-6) bers. The computational results are then
compared to thin slice push-out experimental data. It is observed that the bilinear CZM reproduces the macroscopic
mechanical response and the failure process while the exponential form fails to do so. From the numerical simulations,
the parameters that describe the bilinear CZM are determined. The sensitivity of the various cohesive zone parameters
in predicting the overall interfacial mechanical response (as observed in the thin-slice push out test) is carefully examined. Many researchers have suggested that two independent parameters (the cohesive energy, and either of the
cohesive strength or the separation displacement) are sucient to model cohesive zones implying that the form (shape)
of the tractionseparation equations is unimportant. However, it is shown in this work that in addition to the two
independent parameters, the form of the tractionseparation equations for CZMs plays a very critical role in determining the macroscopic mechanical response of the composite system.  2002 Elsevier Science Ltd. All rights reserved.
Keywords: Cohesive zone models; Metalceramic interface; Push-out test; Thermal residual stress; Interface failure

1. Introduction
Metallic and intermetallic matrix composites (MMCs and IMCs) are being considered as potential
material systems of choice for advanced propulsion systems of future air and space craft because of their
high specic stiness and strength. A critical issue in the successful application of these composites is the
mechanical characteristics of bermatrix interfaces. Interfaces are narrow regions separating well-dened
domains and are primarily responsible for a range of key properties including stiness, strength, and
fracture behavior (Jayaraman et al., 1993). The role of interfaces is very vital to the stress transfer between

Corresponding author. Fax: +1-850-410-6420.


E-mail address: chandra@eng.fsu.edu (N. Chandra).

0020-7683/02/$ - see front matter  2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 2 0 - 7 6 8 3 ( 0 2 ) 0 0 1 4 9 - X

2828

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

the ber and matrix and interfaces inuence the mechanical performance and fracture behavior of composites under various loading conditions (Kim and Mai, 1991a,b).
Interfaces have been modeled in a number of ways, e.g., a narrow region of continuum with graded
properties, an innitely thin surface separated by springs, and cohesive zones with specic tractionseparation relations. One of the key research issues is to determine the best way to characterize interfaces within
the framework of continuum mechanics rather than using ad hoc methods just to facilitate numerical
implementations, e.g. springs in nite element methods. In the past few years, Chandra and co-workers
(Ananth and Chandra, 1995a,b; Chandra and Ananth, 1995; Mukherjee et al., 1997, 1998; Chandra and
Ghonem, 2001) simulated the interfacial mechanical behavior of composites in thin-slice push-out tests
incorporating the spring layer model and fracture mechanics approach. They analyzed both the initiation
and the propagation of interfacial failure to explain many of the experimental observations of ber pushout behavior at room and elevated temperatures. In the spring layer model (Ananth and Chandra, 1995b), a
stress based criterion for debonding and a frictional resistance based criterion for interfacial sliding have
been used to capture debonding and sliding. Debonding is postulated to occur under the combined action
of normal tensile stress (mode I) and shear stress (mode II) at the interface. In the fracture mechanics
approach strain energy release rate at the tip of the interface cracks is computed using the equivalent
domain integral method (Mukherjee et al., 1997). In this approach a critical strain energy release rate is
used as the criterion to determine the propagation of the crack.
Recently the cohesive zone approach is being increasingly used in describing fracture and failure behavior in a number of material systems. CZM has been used in the past to study crack tip plasticity and
creep under static and fatigue loading conditions, crazing in polymers, adhesively bonded joints, interface
cracks in bimaterials, and crack bridging due to bers and ductile particles in composites. Interface modeling using CZM has the distinct advantage compared to other global approaches (e.g. shear lag model), in
that it is based on a micro-mechanical approach. CZM was originally proposed by Barenblatt (1959, 1962)
as a possible alternative to the concept of fracture mechanics in perfectly brittle materials. Later, Dugdale
(1960) extended this concept to perfectly plastic materials by postulating the existence of a process zone at
the crack tip. Based on the Barenblatt and Dugdale models, Hillerborg and co-authors (van Mier, 1996)
proposed the ctitious crack model (FCM) for analyzing crack growth in cementitious composites. FCM
has made rapid strides in the analysis of concrete structures in the eld of civil engineering.
CZM has spawned a plethora of models in fracture of metals, ceramics, and polymers and their composites (Needleman, 1990a,b; Rice and Wang, 1989; Wappling et al., 1998). It is not the purpose of this
paper to review all of those models and applications, but to outline some of the key works relevant to the
present study. Needleman was one of the rst to use polynomial and exponential types of tractionseparation equations to simulate the particle debonding in metal matrices (Needleman, 1987, 1990a, 1990b,
1997). Xu and Needleman (1993, 1994, 1995) further used the above models to study the void nucleation at
the interface between particle and matrix, fast crack growth in brittle materials under dynamic loading, and
dynamic crack growth at the interface of bimaterials. Tvergaard (1990) used a quadratic tractiondisplacement jump form to analyze interfaces. Tvergaard and Hutchinson (1992) used a trapezoidal shape of
the tractionseparation model to calculate the crack growth resistance. Camacho and Ortiz (1996) employed a linear tractionseparation equation with an additional fracture criterion to propagate multiple
cracks along arbitrary paths during impact damage in brittle materials. Geubelle and Baylor (1998) have
utilized a bilinear CZM to simulate the spontaneous initiation and propagation of transverse matrix cracks
and delamination fronts in thin composite plates subjected to low-velocity impact. In all the CZMs (except
Dugdales model and Camacho et al.s model; see Table 1), the tractionseparation relations for the interfaces are such that with increasing interfacial separation, the traction across the interface reaches a
maximum, then decreases and eventually vanishes permitting a complete decohesion. The main dierence
lies in the shape and the constants that describe that shape. In this context, the term shape is used in a
loose sense to describe the normal (or tangential) traction vs. the normal (or tangential) separation response

Table 1
Various cohesive zone models and their parameters
Author
(year)
Barenblatt
(1959, 1962)

Model parameters
R dd 0 G1 t dt
p
0
t
q
pET
1m2 (ductile)
q
pET0
1m
2 (brittle)

T T0 T1
T0 is work of separation
for brittle materials
T1 is work of plastic deformation
 
s
2 sin2 p4 rTy
a
For small value of T =ry
 2
s
1:23 rTy
1

Problem
solved

Model constants

Comments

Perfectly
brittle
materials

The rst to
propose the
cohesive zone
concept

Yielding of
thin ideal
elasticplastic
steel sheets
containing slits

Plastic zone
ranges from
0.042 to 0.448
(in.)

Cohesive stress
equated to yield
stress of material

d 109 to
108 m cohesive
energy 110 J/m2
rmax 1000
1400 MPa
ry 350450 MPa

Phenomenological
model; predicts
normal separation

Needleman
(1987)

/sep is work of separation


d are normalizing parameters
rmax is cohesive strength

Particle-matrix
decohesion

Rice and
Wang
(1989)

E0 is initial Youngs modulus


h is normalizing parameter
rmax is maximum
stress


a is constant Ea02h 2c

Solute
segregation

Needleman
(1990a)

/sep is work of separation


d are normalizing parameters
rmax is cohesive strength

Particle-matrix
decohesion

Ascending part is
equated to E0
considers normal
separation and
ignores shear
separation

d 109 to
108 m

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

Dugdale
(1960)

Proposed model

Predicts normal
separation

2829

(continued on next page)

Author
(year)

Proposed model

Problem
solved

Model constants

Comments

Needleman
(1990b)

/n , /t are work of normal


and shear separation
dn , dt are critical displacements
rmax is cohesive strength

Decohesion of
interface under
hydrostatic
tension

dn dt
2  1010 to
2  109 m
JIC =/n
0:572:59
rmax =r0 2, 3

Periodic shear
traction to model
Pieriels shear
stress due to slip

Tvergaard
(1990)

dn , dt are critical displacements


rmax is cohesive strength

Interfaces of
whisker reinforced metal
matrix composites

dn dt
1  109 m
E 60 GPa
Youngs mod
ry =E 0:005
rmax =ry 59

Quadratic model

Tvergaard
and Hutchinson (1992)

C0 is work of separation
dc is critical displacement
rmax is peak normal traction
=interface strength d1 ,
d2 are factors governing shape

Crack growth
in elasto-plastic material,
peeling of adhesive joints

Css =C0 010


Css Pl wk:,
dcn =dct 1
rmax =ry 014 d1 ,
d2 0:15, 0.5
ry =E 1=300

Claims shape of
separation law are
relatively unimportant

Xu and
Needleman
(1993)

/n is work of normal separation


/t is work of shear separation
dn , dt are critical displacements
rmax is cohesive strength

Particle-matrix
decohesion

dn dt
2  1010 to
2  109 m

Predicts shear and


normal separation

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

Model parameters

2830

Table 1 (continued)

r0 , s0 are normal and shear stress


at fracture initiation
drcr ; dscr are critical normal and
shear opening displacement
GC fracture energy

Impact

Alumina:
r0 400 MPa,
drcr 1:7 
107 m;
steel:
r0 1500 MPa,
drcr 5:43  106 m

Uses additional
fracture criterion;
predicts failure by
both shear and
normal separation
in tension and by
shear separation in
compression

Geubelle
and Baylor
(1998)

/n , /t are work of normal and


tangential separation
Dn , Dt are normal and tangential
displacement jump
dn , dt are normal and tangential
interface characteristic length

Delamination
by low-velocity
impact

rmax E=100 to
E=10 critical
normal displacement jump
Dcn 105 106 m

Bilinear model;
ascending curve
can be matched to
initial stiness of
the material

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

Camacho
and Ortiz
(1996)

2831

2832

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

as typically used in the literature. A more precise concept would be to examine the functional form of the
relation between the traction vector and displacement vector that exists between the separating surfaces
during the fracture process. However, for the sake of simplicity the term shape will be used in this work.
The magnitude of the parameters in CZMs vary widely, ranging from MPa to GPa for traction, J to kJ
for energy, and nanometers to micrometers for separation distance. Table 1 describes the details of some
popular CZMs mentioned above with specic attention focused on the shape and the values of the model
parameters.
The basic question raised in this paper is whether the detailed form, i.e., the shape of tractionseparation
equations, aects the macroscopic mechanical response of a material system. We attempt to answer this
question by comparing the predictions of two dierent models to the experimental results of thin slice push
out tests. In this research work, the metalceramic interfaces in Titanium Beta 21S/SCS composites have
been studied by using exponential and bilinear forms of CZMs. This paper is arranged as follows: in Section
2, formulation of the interface problem is presented; in Section 3, critical issues of various CZMs are
discussed; in Section 4, two CZMs used in the present work are outlined; following Section 4, the nite
element approach is given in Section 5; nally, the detailed results are discussed in Section 6, which includes
comparison between experimental observations and computational results and comparison between simulation results from two dierent models. The eects of various parameters that describe CZMs are investigated systematically to examine the sensitivity of the interfacial parameters of CZM. Currently, many
researchers (e.g. Rahulkumar et al. (2000), Mohammed and Liechti (2000), Hutchinson and Evans (2000))
have assumed that two independent parameters (cohesive energy, and either of the cohesive strength or the
separation displacement) are sucient to model interfaces using CZM. Series of parametric studies indicate
that this assumption leads to meaningless responses (see Figs. 10 and 12). The responses clearly show (by
the theory of contradiction) that those two parameters are not sucient to represent the physics of the
problem of the interface separation process. In order to accurately simulate the interface and reproduce the
macroscopic mechanical behavior of composites, CZM should include, apart from those two parameters,
the shape of tractionseparation law.

2. Interface problem
In general, continuum models of materials are stated as boundary value problems in which we seek
quantities such as displacements, velocities, stresses or temperatures at each point in a given domain. The
governing equations of the continuum are identied with the balance laws of classical physics such as
conservation of energy, mass and momentum. The geometries of deformation are prescribed in term of a
kinematic relationship between strain and displacements or deformation gradients. The specic choice of
the relationship depends on whether the body is undergoing nite deformation or not. To complete the
description of the boundary value problem a constitutive relationship between the kinematic quantity
(strain, deformation gradient) and the kinetic quantity (stress, stress rate) is postulated. This constitutive
equation is specic to the body under consideration within the range of temperature, rate of loading, type of
loading and environment and is usually obtained phenomenologically.
If the boundary value problem consists of domains where one set of equations cannot be applied, then
dierent sets are used to represent the individual domains. A case in point is the modeling of heterogeneous
composites in which dierent forms of constitutive relations are used. For such bodies, there is a distinct
separation between the various regions where the usual theories of continuous media are valid. Such a
separation leads to discontinuity in at least a few of the eld quantities, e.g. strains. In the sense of continuum mechanics, this separation denes the interface that needs to be independently characterized. In
some cases, interfaces are not modeled explicitly and the two regions are rigidly bonded leading to continuous displacement elds with discontinuous strain and stress elds across the interface.

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

2833

Thus, the thermo-mechanical characterization of interfaces should be explicitly included for determining
the response of bodies represented by regions of dierent materials. This prescription is even more critical if
the interfaces were to physically separate (open up or slide) during the loading or unloading process.
2.1. Generic formulation
Consider the two solid bodies X1 and X2 separated by a common boundary S as shown in Fig. 1(a),
where S can be considered as the same surface S1 2 X1 and S2 2 X2 , in the initial conguration, i.e. S1
S2 S. Mathematically, we would like to dene S as an innitesimally thin 3-D domain with surfaces S1
and S2 being the part of X1 and X2 respectively, before separation occurs. For all practical purposes the
surface S1 or S2 can be identied as a single surface as a part of either of the domains. A material particle
initially located (within either of the domains X1 and X2 ) at some position X , moves to a new location x,
with a one to one correspondence between x and X given by the equation of motion x vX ; t or
xi vXj ; t.
In a generic sense, S denes the interface between any two domains. If X1 is a metal and X2 a ceramic,
then S represents a metalceramic interface; if X1 and X2 belong to the same material depicting grains of
dierent orientations then S is a grain boundary, and if X1 and X2 represent the same domain X1 [ X2 X,
then S is an internal surface not yet separated.
In any one of those cases, if S separates (fractures) to
and
as shown in Fig. 1(b), then the process
creates a new internal/external surface violating the fundamental laws of continuity. Obviously the newly
formed region cannot be uniquely mapped from the undeformed conguration. The equation of motion of

Fig. 1. Conceptual frame work of cohesive zone model for interface.

2834

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

the body xi vXj ; t cannot identify the new region. This is the fundamental problem in modeling fracture
(creation of internal/external surface) in the framework of the mechanics of continuous media.
The surface S represented by the unit normal N N 1 2 S1 , and N 2 2 S2 and N N 1 N 2 ) acting along the
boundary separating the domain prior to deformation is as shown in Fig. 1(a). In the deformed conguration as shown in Fig. 1(c) n^1 and n^2 represent the unit normal of the surfaces (separated or otherwise).
2.2. Domains X1 and X2
Consider two neighboring particles located at X and X dX in the initial conguration. Let dX deform
to dx such that
dx Fe dX ;

where Fe is the deformation gradient evaluated with respect to the material coordinates. When constitutive
equations of materials are formulated in the rate form (e.g. elastoplasticity, viscoplasticity) then it is
convenient to dene velocity gradient e
L such that
o
v
e
L ;
2
ox
e is decomposed into
where the gradient of velocity vector v is evaluated with respect to spatial coordinates. L
e and antisymmetric part W
e such that
a symmetric part D
e
e W
e;
LD
where
e 1 e
D
Le
LT
2

3
and

e 1 e
W
Le
L T ;
2

e is the rate deformation tensor and W


e is the spin tensor.
where D
In order to formulate an appropriate constitutive relation to describe the thermo-mechanical behavior of
material, balance laws (conservation of mass, linear and angular momentum) and thermodynamic laws
(rst and second law) should be satised. Additionally, the equations should remain invariant under a
change of frame of reference. This condition can be satised by selecting quantities that are objective (frame
invariant) in nature.
If the material is governed by Cauchy elasticity, which states that the current stress depends only on the
e , such that
current deformation, then the Cauchys stress can be written in terms of right stretch tensor U
e G U
e R
eT ;
r~ R

eU
e and R
e is the orthogonal
where G is the material response function and deformation gradient Fe R
rotational tensor. However, if the material is hyperelastic (possesses a strain energy density function W,
that depends on the current conguration), then
oW
e and e
W W E
S
;
6
e
oE
e is second PiolaKircho stress tensor, and E
e is the GreenLagrangian strain tensor. If the strain
where S
energy density is W 1=2Eij Cijkl Ekl , then
e
eE
e;
SC

e is the fourth order elasticity tensor. In the nite deformation inelastic models, the constitutive
where C
equations are formulated in the rate form based on hypoelasticity. In this case, it is assumed that the
material has no memory and that the Cauchy stress rate D~
r=Dt only depends on the current state r~ and
velocity gradient e
L . Thus

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

D~
r
g~
r; e
L :
Dt

2835

In order to fulll the objectivity principle (principle of the material-frame indierence), the above equation
can be written as

e ;
r h~
r; D

where r is the objective Jaumann rate of Cauchy stress tensor, given by


D~
r e
e T:
 W r~  r~W
Dt

10

It is noted that D~
r=Dt is the material derivative of the Cauchy stress. If r is a linear function of the rate of
deformation, then

e ~
e:
rC
r D

11

The above hypo-elastic formulation can be extended to inelastic material by adopting an additive decomposition of the rate of deformation tensor
e D
e El D
e In ;
D

12

e El and D
e In are the elastic and inelastic parts of the rate of deformation tensor. Although the validity
where D
of the above equation has been questioned, for the present analysis the formulation is acceptable. Additionally, the equation has been implemented in most of the commercial codes including the one used in the
present work. The constitutive equation can then be written as

e D
eD
e In :
rC

13

e is assumed to be isotropic. Since the rate of deformation tensor D


e is objective, the
The elasticity tensor C
above equation satises the principle of material frame indierence.
2.3. Interface S
If S continues to be a part of X1 and X2 (having points/particles common to both), then the motion of S
b would have rotated
can be uniquely dened by the motion of either X2 or X1 . Though the surface normal N
b K Fe 1 n^, (K ds=dS stretch ratio, S and s are the
and deformed to n^, in the sense of the kinematics N
lengths of a small segment in the original and deformed congurations, respectively). Thus, we have a oneto-one relationship between the deformed and undeformed congurations.
However, if S is to be separated as shown in Fig. 1(b), then we have created a pair of new surfaces in the
traditional sense of the term. Consider the region bounded by
and
belonging to a new domain X .

Assume that X is a 3-D domain made of extremely soft glue, which can be shrunk to an innitesimally thin
surface but can be expanded to a 3-D domain. The constitutive relation of X is expressed quite dierently
from that of a typical 3-D solid (e.g. X1 or X2 ). The two surfaces that are initially part of X1 and X2 (S1 and
b 1 and N
b 2 in the undeformed conguration; N
b 1 and N
b 2 are equal and opposite. During
S2 ) have normal N
deformation the surfaces rotate to new normal n^1 and n^2 . As the surfaces separate we have two surfaces
(
2 X1 \ X ) and
(
2 X2 \ X ). The constitutive equation is written in terms of the normal and
shear tractions and their corresponding separation displacements. For a narrow region (crack tip region)
the directions of n^1 and n^2 are approximated to be same. A typical constitutive relation of X is given by T d
relations (see Fig. 1(d)), such that
if jdj < jdsep j;

r~n^ T :

14

2836

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

Beyond a separation distance of jdj > jdsep j, the traction is identically zero within X ,
jdj P jdsep j;

r~n^ T 0:

15

It can also be construed that when jdj > jdsep j in the domain X , the stiness Cijkl  0. In order to implement
the vectorial inequalities given in Eqs. (14) and (15) typically two separate identities are postulated for the
normal and tangential components with limits set for each of them. Other methods of implementation are
equally possible.
In the numerical schemes, such as in nite element method, interfaces have been modeled as contact
surfaces or surfaces separated by spring elements. Such approaches usually have ad hoc criteria for separation and sliding and usually do not conserve energy. For example, when a nodal spring constant is
reduced to zero upon satisfying a failure criterion, stored energy is not appropriately dissipated.
In the cohesive zone approach, an independent specication of a local constitutive relation is established.
This relationship reects the coupling of stresses (or tractions) at the interfaces to the displacements they
suer. One class of interfacial constitutive model is based on the existence of a potential /D that measures
the energy cost to displace the adjacent planes across an interface by a relative displacement D (Needleman,
1992). The resulting tractions are given by
T

o/D
:
oD

16

3. Critical issues in the application of cohesive zone models


Since CZMs provide a convenient way to model fracture (or for that matter any separation) process, it is
instructive to critically review the available models. It is our opinion that at this stage of their development,
CZMs should be truly called models rather than sanctifying them as constitutive equations or as cohesive
laws. We examine the most popular CZMs in terms of their forms, physical signicance and applications.
We examine not only the functional form of tractionseparation distance relation but also the various
parameters that describe such relations, e.g. the rising part of the curve, the peak values (both the maximum
traction, rmax , and corresponding dmax ), and the value of dsep when complete separation occurs. Usually the
T d relation is expressed in the form of Tn vs. Dn =dc . The normalizing process Dn =dc is necessitated by the
requirement that when traction is obtained as a function of cohesive strength, the multiplier has to be a
non-dimensional quantity. An important issue is whether dc is just a scaling parameter or if it has any
other physical signicance. It should be noted that dc varies anywhere from 1010 to 105 m and rmax ranges
from MPa to GPa. A nal issue to be addressed is the physical signicance of the area under the curve.
Does it represent just the surface energy or all the dissipative work associated with the separation process?
Barenblatt (1959, 1962) was the rst to propose the concept of CZM while analyzing brittle fracture. He
postulated that the molecular force of cohesion exists near the edge of a crack. The intensity of the molecular force of cohesion Tn is found to vary as shown in row 1 of Table 1. Very interestingly, in this work
Barenblatt did not suggest any specic form of T d relationship. He postulated the existence of the cohesive
stress (thus reducing the singular stress). Crack propagation occurs when the integral eect of the cohesive
force acting ahead of the crack tip (along the crack plane) for the length equal to the cohesive zone size d
reaches a set value for the modulus of the cohesion, K. While investigating the yielding of an elasticplastic
steel sheet, Dugdale (1960) identied the existence of cohesive force (termed as internal stresses) acting at
the tip of the crack for a length equal to plastic zone size.
Needleman (1987) was one of the earliest to apply the tractionseparation cohesive relationship to model
void nucleation and debonding of inclusions in metallic materials. A number of potentials have been
suggested in order to obtain dierent forms of response between T and u. For example, a polynomial

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

2837

function in 1987s model (row 3, Table 1), an exponential type in 1990as model (row 5, Table 1), an exponential function for the normal traction and periodic function for the tangential traction in 1990bs
model (row 6, Table 1) have all been proposed by Needleman and co-workers. In 1993, Needleman again
used exponential forms for both normal and shear tractions (row 9, Table 1). While the polynomial form
was chosen for analytical convenience, the exponential form was motivated by atomistic considerations
following the work of Rose et al. (1981). The periodic response for shear was postulated to model the
Pieriels shear stress associated with slips.
The area under the normal traction vs. separation distance is assumed to be the work of separation /sep
and is in the range of 1 to 10 J/m2 . In most of their (Needleman and co-investigators) work, the maximum
stress rmax is typically assumed to be about three times that of yield stress (ry ) of the matrix material.
Typical values of rmax 10001400 MPa, and dmax 109 to 108 m have been used in their analysis. In the
case of the exponential model, separation dsep theoretically occurs as d ! 1 . However, for practical reason
it can be assumed that complete separation occurs when dsep is about 67 times dmax . For the polynomial
model dmax dsep =3.
Rice and Wang (1989) analyzed embrittlement (or ductilization) eects of atomic scale separation of
solutes to grain boundaries. The normal traction vs. separation distance proposed in their work is shown in
row 4 of Table 1. The area under the curve was assumed to be
Z 1
G
rd dd 2cint A;
17
0

where G is the strain energy release rate in an elastic-brittle material. Rice raises a very important question
regarding the area (A) under the curve. It can be interpreted as the adhesion energy or work of fracture, i.e.,
the energy consumed in opening up two new surfaces from a single one. As a rst approximation, for
perfectly brittle materials the area under the curve can be assumed to be G. However, for ductile materials
we should recognize that if J-integral represents the strain energy release rate J oU =oA, then
JIC Wp 2cint Wp A;

18

where Wp is the plastic (or viscoplastic) work ahead of the crack tip. While in brittle inclusions (carbides in
steel), Wp is about 1.32.5 times that of A, in the failure of polycrystalline ductile materials Wp is about 500
1000 times that of A.
Tvergaard (1990) proposed a simplied interface model, where the traction shows a quadratic variation
with respect to the traction separation, having a steep ascending curve and gradual softening curve as
shown in row 7 of Table 1. Tvergaard and Hutchinson (1992) proposed a CZM as shown in row 8 of Table
1, for simulating crack growth processes in elasticplastic solids. The T d response is made up of linear
segments with separate slopes for rising and falling parts with a horizontal segment at the traction value of
rmax . For mode I cracks, in elasticplastic solids, Tvergaard and Hutchinson (1992) have noted that the
shape parameters d1 =dc ; d2 =dc are relatively unimportant and only rmax and the area under the curve C0
play a key role.
Camacho and Ortiz (1996) used CZM concepts in their nite element implementation of ballistic impact
problems. They departed from the earlier works and dened a separate multi-axial fracture criterion based
on local normal and tangential stresses. Once this criterion is satised a cohesive relationship, as shown in
row 10 of Table 1, is specied. The area under the curve is assumed to be the fracture energy, discounting
the energy consumed prior to the stress at the tip of the crack reaching rmax . The parameters used are
dsep 5:43 lm, rmax 1500 MPa, G 4074 J/m2 for steel (impactor), and dsep 17 lm, rmax 400 MPa,
G 34 J/m2 for alumina. This work deviates once again from the others discussed in that it attempts to
solve a macroscopic boundary value problem using CZM only to model the separation and not the initiation of cracks. Here G inherently includes the surface energy and all other inelastic energies associated with
the elasticplastic processes.

2838

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

Geubelle and Baylor (1998) have simulated the delamination process in a composite plate subjected to
low velocity impact using CZMs. They proposed bilinear models both for normal and tangential separation
processes and we have followed this model in the present work. The area under the normal traction
separation curve is considered as GIC and that due to shear tractionseparation is GIIC . In applying this
model to graphite/epoxy delamination the following values have been adapted: for PMMA, rmax 324
MPa, GIC 352:3 J/m2 , dsep 2:17 lm; and for epoxy, rmax 91 MPa, GIC 88 J/m2 , GIIC 315 J/m2
2
(interface), Gply
IC 147 J/m .
A number of other investigators have followed the above models in solving a variety of boundary value
problems. Rahulkumar et al. (2000) simulated large deformation peel tests of a fully relaxed polymer using
the model shown in row 8 of Table 1. They use /sep 510 J/m2 and dsep 2550 lm and have a corresponding rmax 0:147 MPa. Espinosa et al. (2000) simulated dynamic impact of woven ber composites
using models shown in row 7, 8 and 11 of Table 1. They use a value of rmax 50 MPa, /sep 2002000 J/
m2 , dsep 48 lm for the PMMA/epoxy interface. Foulk et al. (2000) modeled SCS-6/Timetal 21s metal
matrix composite interface using the model 8 in Table 1 with values of rmax 5131500 MPa and
dsep 0:10.2 lm. From the above discussion it is abundantly clear that there are many forms of cohesive
zone models and the parameters in each of them vary by orders of magnitude.
4. The cohesive zone models
In this work, two CZMs, exponential (Xu and Needleman, 1993) and bilinear (Geubelle and Baylor,
1998) forms are used to simulate the interfacial mechanical response by considering push-out test process as
the example.
For the exponential CZM, the interfacial potential is given by
/Dn ; Dt /n /n expDn =dn f1  r Dn =dn 1  q=r  1
 q r  q=r  1Dn =dn  expD2n =d2t g;

19

with q /t =/n , r D n =dn , where /n and /t are the work of normal and shear separation, respectively; Dn
and Dt are the normal and tangential displacement jumps, respectively; dn and dt are the normal and
tangential interface characteristic-lengths; and D n is the value of Dn after complete shear separation takes
place under the condition of normal tension being zero, Tn 0.
Using Eq. (19) in Eq. (16) the interfacial tractions are obtained as
Tn /n =dn expDn =dn fDn =dn expD2t =d2t 1  q=r  11  expD2t =d2t r  Dn =dn g;
20
Tt /n =dn 2dn =dt Dt =dt fq r  q=r  1Dn =dn g expDn =dn expD2t =d2t :
The works of normal and shear separations are related to rmax and smax , respectively,
p
/t e=2smax dt ;
/n rmax edn ;

21

22

where e exp1.
With Dt 0, the normal tractionseparation relation obtained from Eq. (20) is shown in Fig. 2(a), and
the variation of Tt with Dt given by Eq. (21) for Dn 0 is shown in Fig. 2(b).
The interfacial constitutive relations for the bilinear CZM are given below:
For dn > 0,

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

2839

Fig. 2. (a) Variation of normal traction, Tn , with Dn for Dt 0. (b) Variation of shear traction, Tt , with Dt for Dn 0.

rmax
Tn

d ;
dmax n
rmax 1d
d 1dmax

dn ;

(r
Tt

c
max Dn
d;
dmax Dct t
rmax 1d Dcn
d 1dmax Dct

d 6 dmax
dt ;

For dn 0,
(r
Tt

c
max Dn
d;
dmax Dct t
rmax 1d Dcn
d 1dmax Dct

d 6 dmax
d > dmax ;

d > dmax :

d 6 dmax
dt ;

d > dmax ;

23

24

25

where rmax and smax are interface normal and tangential strength, respectively; dmax is interface characteristic length parameter; Dcn and Dct are the critical normal and tangential separations at which complete
separation is assumed; and dn , dt and d denote the non-dimensional normal, tangential and total displacement jumps respectively, dened by
q
Dt
Dn
dt c ; dn c ; d d2t d2n :
26
Dt
Dn
For pure opening Dt 0 and pure shear separation Dn 0, the variation of normal and tangential
tractions with respect to Dn and Dt are shown in Fig. 3(a) and (b), respectively. The normal Cn and
tangential Ct works of separation per unit area of interface are given by

Fig. 3. (a) Normal traction Tn as a function of the normal separation dn for dt 0, (b) shear traction Tt vs. shear separation dt for
dn 0.

2840

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

Cn rmax Dcn =2;

Ct smax Dct =2:

27

5. Finite element approach


To evaluate the interfacial properties of composites, the single ber push-out test has emerged as the defacto experimental tool. The objective of any test for characterizing the interface is to extract qualitative as
well as quantitative information about the properties. But many factors make it clear that extracting
quantitative data from the single push-out test is not trivial.
The thin-slice push-out test is modeled as an axisymmetric problem (Kallas et al., 1992), since the
loading (both thermal and mechanical) produces axially symmetric displacement elds. The discretized
mesh has 4800 axisymmetric four-node elements to model the ber and matrix. Duplicate nodes are created
at the interface on the ber and matrix sides. 240 axisymmetric cohesive elements with each having four
nodes and zero thickness in the direction normal to the interface are used to model the interface behavior.
The general-purpose commercial code ABAQUS (1998) is employed to carry out the analysis due to its
capability of handling non-linear problems and also, because of its exibility in allowing user-dened
subroutines to be linked to the main program. The cohesive element model is input as a user-dened element subroutine UEL into ABAQUS. The mesh used in the simulation is shown in Fig. 4.
To simulate the single ber push-out test, the elastic constitutive behavior is assumed for the ber, and
the matrix is assumed to be a rate independent elasticplastic material. The temperature dependency of the
elastic and inelastic properties (Nimmer et al., 1991; Kroupa and Neu, 1994) of the constituent phases is
included in a piece-wise linear manner. The material properties used in the analysis are given in Fig. 5. The
analysis is done in the following three steps:
1. In step 1, the cooling process after the composite consolidation at high temperature is modeled. A reference temperature Tref is assumed above which the composite is stress free, which is taken as 815 C.
Thermal residual stress (TRS) is induced due to the coecient of thermal expansion mismatch when
the sample is cooled down from Tref to ambient temperature. The boundary conditions for this step
are shown in Fig. 6(a). The nodes on the top surface are tied to have equal displacement to create a generalized plane strain condition.
2. When a thin slice is cut out of the bulk composite plate, the residual stresses tend to relax. This is simulated by removing the existing tying constraints and boundary conditions from the top and bottom, and

Fig. 4. The model used for push-out test simulation and details of interface element.

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

2841

Fig. 5. Variation of material properties with temperature.

allowing the stresses in the specimen to relax and reach equilibrium. The symmetry boundary conditions
used for this step are shown in Fig. 6(b). This process results in shear stresses at the interface due to the
dierential axial residual strains between the ber and matrix, as shown in the Fig. 7.
3. Axial displacement is applied to the punch shown in Fig. 6(c) until the ber slides out of the matrix after
the entire length of the ber debonds.
6. Results and discussion
6.1. Distribution of thermal residual stresses
The role of TRS is often ignored in analytical and experimental considerations of interfacial eects in
metallic based composite materials. This oversight is unfortunate because the resultant interpretation of
properties and behaviors is usually misleading. TRS are inherent characteristics of composite materials
arising from the diering coecients of thermal expansion of the component materials. Since composites

2842

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

Fig. 6. Boundary conditions used in the nite element analysis.

Fig. 7. Distribution of TRS along cross-section after cooling at ambient for sample thickness of 0.63 mm.

are invariably used at a dierent temperature than that at which they are fabricated, the diering thermal
expansions or contractions of the ber and matrix set up the thermally induced residual stresses around the
interface on cool down from consolidation temperature. Typical MMCs and IMCs are fabricated at
temperatures which are quite high relative to ambient (above 815 C for Ti-MMCs (Yang et al., 1991; Sohi
et al., 1991)), and thus hold the possibility of producing very high stress levels that inuence the fracture
and failure process of the material system.
The distribution of TRS induced during the cooling process (step 1) is shown in Fig. 7. It can be seen that
the ber and matrix are under uniform axial compression and tension respectively to satisfy the force
equilibrium. Also, the ber has similar distribution for radial and tangential stresses and the matrix is
subjected to compressive radial and tensile tangential stresses.

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

2843

Fig. 8. Shear stress distribution along the interface. (a) After slicing, (b) beginning of loading, (c) debonding occur and (d) complete
debonding. Note: the displacement is applied at the left side of ber (see arrow).

The evolution of shear stress distribution along the interface from step 2 to 3 (see Fig. 6) of the push-out
test is shown in Fig. 8.
6.2. Comparison between experimental results and numerical predictions
The non-aged Timetal-21S/SCS-6 composite system was fabricated and push-out tested as a part of this
work. The reinforcing bers are made of SCS-6 materials with 142 lm diameter, with the volume fraction of
ber being 35%. The following three cases selected from a set of experimental data of Osborne et al. (2000),
are used in the current study:
test temperature: 25 C; sample thickness: 0.63 mm,
test temperature: 500 C; sample thickness: 1.07 mm,
test temperature: 650 C; sample thickness: 1.10 mm.
Forcedisplacement data obtained from the experiments are shown in Fig. 9, and are the average of at
least two data sets in each case.
Finite element simulation was carried out on the model shown in Fig. 4 as per the three steps discussed
earlier. The bilinear CZM (see Eqs. (23)(27)) for the interface is characterized by four parameters: dmax ,
smax , Dct and Dcn . Since shear failure is by far the dominant failure mechanism in current study, it is assumed
that Dcn is equal to Dct . Therefore, only three parameters need to be determined in the simulation. To determine smax and Dct , the work of shear separation /t is rst estimated based on the area under the force
displacement curves obtained from the experiment (Fig. 9). The values of smax for various temperatures are
chosen so that the experimental response is closely matched. And then Dct is determined by using Eq. (27).
The values used in the simulation are listed in Table 2.

2844

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

Fig. 9. Experimental measurement and numerical predictions, at dierent temperatures. Note: displacement for each temperature is
relative to the starting position.

Table 2
The values of CZM parameters used for simulation
Temperature

dmax

smax (MPa)

Dct (m)

Dcn (m)

/t (J m2 )

25 C
500 C
650 C

1.0
1.0
1.0

200
100
90

3.5E5
1.3E5
1.42E5

3.5E5
1.3E5
1.42E5

3000
650
640

The experimental and computational results for various temperatures (i.e. 25, 500 and 650 C) are shown
in Fig. 9. As noted above, the samples for 500 and 650 C are approximately of the same thickness and can
be compared directly. However, due to maximum force limitations, the sample for room temperature is
thinner. First, it can be seen from both experiments and simulations that the peak load decreases with the
increase of temperature due to the relaxation of compressive residual stresses which act on the ber as a
clamping force at elevated temperatures (Davidson, 1992). The simulation response for various temperatures matches the experimental results very well. Also, it can be seen that it is linear for room temperature
until the maximum load is reached. This is followed by catastrophic and complete debonding. For elevated
temperatures, each curve has an initial linear portion, and as the load continues to increase, a reduction in
slope can be observed (especially for the plot at 650 C). It was reported that the change in slope from
linearity for elevated temperatures corresponds to the initiation of the interfacial debond crack (Kerans and
Parthasarathy, 1991). However, from the current simulation results, it is found that the debonding does not
take place until the peak load is reached. It was reported (Davidson, 1992) that the matrix material behaves
elastically at lower temperature. At higher temperature (>400 C), the ability of the matrix phase to ow
under the applied stress is enhanced. It is shown by simulation that the plastic deformation in the matrix
indeed takes place along the entire interface during the loading process at elevated temperatures. Even at
ambient temperature, the plastic deformation occurs near the bottom along the interface. However, it is just
not enough to cause the slope change in macroscopic forcedisplacement curve. Hence, it should be reasonable to conclude that the change of slope in the curves during loading at elevated temperatures is due
not to the initiation of interfacial debonding, but to plastic deformation of the matrix material at the in-

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

2845

terface. It is clear that the choice of smax is very critical to reproduce the macroscopic mechanical response
of the material system.
6.3. Comparison between two cohesive zone models
At the outset it appears that the CZM for the interface and conventional eld equations with appropriate constitutive equations for the bulk (matrix and ber) will reproduce the test results. Consequently,
the agreement leads to a set of cohesive zone parameters (see Table 2) that can then be used for other
geometric and loading congurations. Before such a positive assertion can be made, it is worthwhile to
analyze the uniqueness of the shape of the CZM and the sensitivity of the chosen parameters on matching
the experimental data.
For the purpose of examining whether the form of the CZM is important if the energy (area under force
displacement curve) is maintained at the same level, two forms (exponential and bilinear) are chosen. An
energy value of /n /t 3000 J/m2 was used for the two forms. The simulation results for bilinear and
exponential forms are shown in Fig. 10(a) and (b), respectively. The value of smax was selected as 200 MPa
(as in Table 2 for ambient temperature case), and then altered to 300 and 400 MPa, which corresponds to
curves 13 (Fig. 10(a) and (b)), respectively. It should be noted that a bilinear CZM with smax 200 MPa
matches the experiment very well.
Fig. 10(a) and (b) bring home some very important conclusions regarding a number of issues. Fig. 10(a)
illustrates that even though the cohesive energy is the same, increasing the interfacial shear strength changes
the F DU response of the push-out test. This result is in contrast to the common belief that the separation
process is governed mainly by the energy of separation. Again, focusing on smax 300 and 400 MPa in Fig.
10(a), there is a non-linear response when the force is above 50 N. On closer examination, it is observed that
matrix yields under those conditions, leading to the slope change in the macroscopic response. Arrows in
the Fig. 10(a) indicate the loading at which the initial plastic deformation occurs. The initial debonding for
all three cases, however, takes place at the peak load. Hence, the average shear stress required for failure
savg Ppeak =2p dL is dierent in the three cases. The present discussions again clearly reveal that selection of
smax is critical in determining the macroscopic mechanical response.

Fig. 10. Loaddisplacement plots simulated using exponential and bilinear CZMs. (a) Bilinear model (Geubelle et al., 1998), (b) exponential model (Xu and Needleman, 1993). Note: Pdplastic deformation; Dbdebonding.

2846

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

Fig. 10(b) shows the overall response of the push-out test for the exponential model keeping the cohesive
energy and interfacial strength identical to those of Fig. 10(a). It is immediately clear that the predicted
response is in no way similar to the experimental observations. Arrows in the Fig. 14(b) indicate the loading
at which the initial debonding occurs. The important question is why there is such a drastically dierent
response. In fact, no amount of parametric variations on the exponential CZM can make it match with
experimental results making it unsuitable for modeling the interface in this problem. Focusing on Fig.
10(b), for smax 200 MPa, the peak load (about 53 N) does not correspond to either the initial debonding
or the matrix yield stress. As the push-out load is increased (truly the push-out displacement, see Fig. 8(a)),
the interface responds to the external force according to its constitutive equation. When the peak stress in
each element is reached (Fig. 10(b)), the slope is reversed and the interface becomes more and more
compliant. Thus the overall stiness of all the interfaces in tandem is continually altered with dierent
cohesive elements changing their stiness as the push out displacement increases. Since the separation in
this model takes place with a signicantly large displacement, the rst debonding does not occur until very
late in the process.
It is interesting to note that in both Fig. 10(a) and (b), the shape of the overall forcedisplacement response (F DU , along the longitudinal ber axis) is similar to that of input tractionseparation curve of
CZM (T d, normal to the interface in the radial direction, see Figs. 2 and 3). This enigmatic coincidence can
be explained on the basis of the geometry of the push-out test and the loading process. A simplied model is
presented in the next section to explain why the two responses show a close resemblance.
6.4. A simplied model for push-out test based on cohesive zone model
In this section, we develop a simplied model of the push-out test problem. The circular ceramic ber is
connected to the metallic matrix by means of interfaces. While the ceramic domain X1 is modeled as purely
elastic, the metallic matrix X2 is modeled as temperature-dependent elastic-linearly strain hardening plastic
material governed by von Mises yield criterion. X is modeled by cohesive zone elements discretized as N
four-node axisymmetric elements governed by tractionseparation laws, through T d relations in the nite
element analysis. When the push-out load F is applied, the end of the ber moves by DU . We would like to
establish a relationship between the global F DU and the cohesive zone T d relations.
As shown in Fig. 11, we have simplied cohesive zone nite elements as nonlinear springs. We have pairs
of nodes connected by springs at f1; 10 g, f2; 20 g; . . . ; fn; n0 g co-located in the ber and the matrix before the
application of load. Let Ti be the traction at a spring i after some global load F is applied. Without loss of
generality, let fi be the shear force given by fi pdTi li , where d is the ber diameter, Ti is the average shear
traction in the region of spring i  i  1=2 to i i 1=2 and li the axial length of the region. The
shear displacement of spring i is governed by T d relation.
From the equilibrium of force, we have
F

N
X

28

fi :

i1

6.4.1. Case (1): bilinear model d < dmax


Assume that the matrix has not yielded; for ith spring, Ti ki di and fi pdi li ki d, so that
F pdki

N
X

d i li :

29

i1

Since the eective length of the spring


P remains constant, the total length can be expressed as a linear
combination of each of them. Also N1 di DU , leading to

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

2847

Fig. 11. (a) The schematic free body diagram of all cohesive zone elements. (b) The bilinear and exponential tractionseparation
relations of CZM.

F K 0 DU :

30
0

Thus the global F DU curve will be linear with a dierent slope, K .


6.4.2. Case (2): exponential model d < dmax
As shown in Fig. 11, since k kd, we have the following equation,
F pd

N
X

kdi di li ;

i1

31

F KD DU :
It is thus seen that we have a monotonically increasing curve, with the slope changing with DU , i.e. KD.
6.4.3. Case (3): d P dmax (for some springs)
Since only a few springs will have d P dmax , the traction of others will still be below the peak value. Thus
in both cases (bilinear and exponential) the slope of F DU will change when any of the spring exceeds
d > dmax .
6.4.4. Case (4): matrix yielding
Whenever the matrix yields, the stressstrain response of the matrix becomes non-linear. Consequently
the slope of the F DU curve also changes.
6.4.5. Case (5): spring failure, d P dmax
Since the number of loaded springs will be less than n, the slope of F DU will change.
Thus, the simplied analysis shows that a change in the global response of F DU may be caused by (1)
change in slope (change in sign) of tractionseparation (T d) law of CZM, (2) matrix yielding, or (3) cohesive zone failure.

2848

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

6.5. Parametric study


In Section 6.2, we evaluated a set of parameters (Table 2) within the framework of the bilinear CZM that
matched the experimental results. However, numerical simulation clearly shows that there are many other
parameters that signicantly alter the progression of debonding and hence the loaddisplacement behavior.
In this section, the eects of various parameters, i.e., work of shear separation /t , interfacial shear
strength smax , test temperature and interface characteristic length dmax are investigated systematically by
using bilinear CZM in terms of loaddisplacement response of the push-out test. The main purpose is not
only to simulate a specic material system, but also to provide some insights on the eects of the chosen
parameters on the deformation and failure behavior of the materials in general.
6.5.1. Eect of interface characteristic length, dmax
In this section, the eect of interface characteristic length, dmax , on the forcedisplacement curve is
studied and the results are shown in Fig. 12. The values of dmax are selected to be 0.1, 0.4, 0.7 and 1.0; the
corresponding F DU curves are I, II, III and IV, respectively. It is interesting to nd that the shape of the
forcedisplacement curve is dependent on the value of interface characteristic length, which determines the
shape of shear traction vs. shear separation. The arrows in the gure indicate the loading at which debonding occurs at the interface. Comparing the numerical results with the experimental data, it can be
concluded that for the current material system, the suitable value for dmax is 1.0. Therefore, it is reasonable
to conclude that the shape of the tractionseparation curve for the CZM is an important factor in determining the macroscopic mechanical behavior of the interface in addition to the other interfacial parameters
such as the cohesive energy. Thus, it is obvious that setting dmax 1:0 xes the shape of the CZM; in other
words, any other values of dmax corresponding to dierent shapes will yield unacceptable results for the
present problem of metalceramic interfaces. This is an important point to be considered while selecting
specic shapes of CZM for a given problem.

Fig. 12. Eect of interface characteristic length dmax on loaddisplacement plots (I, II, III and IV represent corresponding force
displacement curves to dierent values of characteristic lengths 0.1, 0.4, 0.7 and 1.0, respectively).

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

2849

Having established that the shape is critical, we proceed to conduct parametric study of other parameters
(/t , smax or dsep ) that describe the bilinear model.
6.5.2. Eect of work of shear separation, /t
The eect of the work of shear separation, /t , on the forcedisplacement curve is discussed in the following section and the results are shown in Fig. 13. The work of shear separation is varied from 100 to 3000
J/m2 and the curves 14 correspond to the work of shear separation of 100, 500, 1000 and 3000 J/m2 ,
respectively. The interfacial shear strength smax (200 MPa) is kept constant. The total area under the force
displacement curve represents the overall work required to separate the ber from the matrix (complete
debonding). It is found from the simulation that the debonding begins to occur at the right end (Fig. 8) just
when the peak load is reached. And the subsequent debonding process along the interface from right to the
left takes place at a very fast pace (corresponding to the unloading portion of curves in Fig. 13). The slope
of the F DU curve is steeper for the case with lower /t and this slope reduces with an increase in /t . As
happens in the push-out test the entire ber is debonded soon after the rst cohesive element (at the left
end) fails. The actual critical separation distances Dct at which the cohesive element debonds completely are
Dct 1  106 , 5  106 , 1  105 , 3  105 m, corresponding to /t 100, 500, 1000, 3000 J/m2 respectively. Thus, suciently large tangential displacement occurs at the left element before the entire ber is
being pushed out.
6.5.3. Eect of shear strength, smax
In this section, simulations were carried out by keeping /t constant and varying smax (and correspondingly critical tangential displacement Dct through the Eq. (27) for the bilinear CZM). The value of dmax
is set equal to 1 as explained earlier.
Fig. 14 shows the macroscopic response of forcedisplacement curves with dierent interfacial shear
strengths, smax . It is seen that when smax 6 200 MPa, there is a linear relationship between force and displacement. For cases when smax 300 or 400 MPa, the slope of the ascending part of the curve is found to

Fig. 13. Eect of work of shear separation on forcedisplacement curves.

2850

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

Fig. 14. Eect of interfacial strength on forcedisplacement curves.

change when the load reaches about 55 N. It is observed from the results that yielding in the matrix occurs
at this load and is responsible for the change in slope. It is worthwhile to note that at critical points where
debonding initiates, the matrix yields when the stress due to applied loading combined with the TRS
reaches the yield stress of 1043 MPa for a given external load (55 N in this case). It can also be noted that
the forcedisplacement curve response for smax 200 MPa is one closer to the experimental result. It is
generally assumed that given the cohesive energy /t and the cohesive strength smax , the behavior of the
CZM can be uniquely described. But it is clear from the present work, apart from /t and smax , the shape
dmax 1 is required to uniquely and completely describe the CZM. On the other hand when dmax 6 1, or
other CZMs are used, the response will be quite dierent from that of experimental results (see Figs. 10(b)
and 12).
6.5.4. Eect of temperature
For sample with thickness of 0.63 mm, the simulation is performed to study the eect of temperature on
the forcedisplacement curve and the results are shown in Fig. 15(a) and (b) for cohesive energies of 3000
and 1000 J/m2 , respectively. As expected, the peak value of load decreases as the temperature increases. It
can also be seen that magnitude of peak force reduces with the decrease in the cohesive energy (Fig. 15(b)
compared to Fig. 15(a)). For the simulation at room temperature, the load keeps increasing linearly with
displacement until the load reaches the peak value, and then debonding initiates from the left end. On the
other hand, at higher temperatures the load increases linearly initially; then the slope changes with the onset
of plastic deformation in the matrix. Also, it should be noted that at higher temperatures, the temperaturedependent yield stress value is considerably lower than that at room temperature which results in plastic
deformation in the matrix at a much lower external load (<55 N).
6.6. The evolution of cohesive traction and energy
As observed in the experiments and conrmed by the simulations, the fracture initiation takes place at
the opposite end of the specimen (left) compared to where the load is applied (right). It is also clearly seen

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

2851

Fig. 15. The eect of temperature on forcedisplacement curve (a) /t 3:0  103 and (b) /t 1:0  103 .

that there is a signicant amount of residual shear stress even before the application of any load. So it is
interesting to examine the absorption of cohesive energy by various elements along the length, especially at
both ends. The growth of interfacial shear traction and work of shear separation in the cohesive elements
during mechanical loading (step 3), are shown in Figs. 16 and 17 for two cases (case 1: dmax 0:1; case 2:
dmax 1:0). The evolution of shear traction Tt and interfacial cohesive energy /t in Figs. 16 and 17 is
plotted along the interface at various stages of loading as indicated by 15 in Fig. 12.
Figs. 16(a) and 17(a) show that the distribution of shear traction at early stages of mechanical loading is
anti-symmetric with respect to the sample center; the right side (top) is in positive shear and the left side
(bottom) is in negative shear, indicated by the curve 1. With slight increase in loading, the shear traction of

Fig. 16. (a) The evolution of shear traction and (b) work of shear separation for case 1 dmax 0:1, /t 3000, smax 200 (15 indicate
various loading positions shown on curve I in Fig. 12).

2852

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

Fig. 17. (a) The evolution of shear traction and (b) work of shear separation for case 1dmax 1:0, /t 3000, smax 200). (14 indicate
various loading positions shown on curve IV in Fig. 12.)

all elements along the interface turns to negative shear traction and also reaches the maximum value smax as
indicated by 2 and 3 in Figs. 16(a) and 17(a). These stages indicated by 13 lie along the ascending part of
the T d curve (Fig. 3). After reaching the peak value smax the shear traction begins to reduce to zero and the
interface debonds as, indicated by curves 4 and 5 in Fig. 16(a) and curve 4 in Fig. 17(a). The stages indicated by 4 and 5 lie along the descending part of T d curve. It is interesting to note that mechanical
loading is applied at the right end (right side in the plot) and the shear traction rst reaches zero in a few
elements on the bottom (left side on the plot), indicating that the fracture has initiated from the left end,
which is consistent with the experimental results (Ghosn et al., 1994).
Figs. 16(b) and 17(b) show the evolution of the work of shear separation (cohesive energy) in the cohesive elements along the interface. In both gures the cohesive energy along the interface is zero at the
beginning of loading and gradually increases with the increase in the mechanical loading, as indicated by
curves 15 in Fig. 16(b) and curves 14 in Fig. 17(b). As observed earlier, with the increase of mechanical
loading, the work of shear separation reaches its critical value /t rst for a few elements near the bottom
end (left side in the plot) rather than that at right side (top) indicating that the debonding always begins to
occur at the bottom for the current material system. The cause of bottom fracture initiation during thinslice push-out tests of MMCs has been found to be predominantly due to TRSs (Ananth et al., 1995;
Chandra et al., 1995).

7. Summary and conclusions


This work can be summarized as follows:
The concept of continuum mechanics is extended to include a zone of discontinuity modeled by cohesive
zones. A comprehensive analysis of some of the popular CZMs has been presented with respect to the
form of the equations and the magnitude of the parameters. It is clearly demonstrated that energy values
ranging from J to kJ, traction values ranging from MPa to GPa and separation distances ranging from
nm to lm are employed to describe interfaces. We illustrate some of the outstanding issues, e.g., physical

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

2853

interpretation of the area under the curve, form of the equations (shape of the traction curves), cohesive
strength and scales in CZMs.
Two CZMs (exponential and bilinear) have been employed to model the interface failure in metal matrix
composites. The bilinear CZM is suitable to simulate the push-out test process and the calculated results
match the experiment data very well. The macroscopic mechanical behavior (F DU curve shape) of
interfaces in the material system has a striking resemblance to that of CZM (shape of interfacial tractionseparation law) provided no plastic deformation in the matrix takes place. This implies that the
macroscopic mechanical response of the interface of the material system is very sensitive to the shape
of the interfacial constitutive relation of the CZM in addition to the other interfacial parameters.
Debonding of cohesive elements takes place when the accumulated cohesive energy reaches the critical
value, as prescribed in the model. It has been observed during the push-out tests, when the load/displacement is applied at the top end, a few cohesive elements near the bottom end rst attain the critical energy, leading to initiation of fracture at the bottom end.
CZMs can be used to describe a broad range of fracture processes in a wide variety of material systems.
However, in order to have truly predictive capabilities, this phenomenological failure model must rely on
carefully designed and conducted experiments to select a proper cohesive zone constitutive relation. Such
a relation should include the shape and appropriate material parameters such as interfacial strength,
work of separation and critical displacement discontinuities, and response during loading and unloading.
It is our view that the CZM represents the physics of the interface separation process and hence the
shape of the CZM should in some sense depend on the inelastic processes occurring at the micromechanical level. Hence when using CZMs to model separation in a given material system, an appropriate shape
(form), depending on the type of material system and the inelastic micromechanical processes, should be
used. Otherwise, the CZM based modeling and simulation will not yield meaningful results.

Acknowledgements
The authors wish to acknowledge the Air Force Oce of Scientic Research (F49620-99-10275) for
providing partial nancial assistance in support of this research.

References
ABAQUS, 1998. Manual, version 5.8, Habbit, Karlsson & Sorensen, Inc., USA.
Ananth, C.R., Chandra, N., 1995a. Evaluation of interfacial shear properties of metal matrix composites from ber push-out tests.
Mechanics of Composite Materials and Structures 2, 309328.
Ananth, C.R., Chandra, N., 1995b. Numerical modeling of ber push-out test in metallic and intermetallic matrix composites
Mechanics of the failure processes. Journal of Composite Materials 29 (11), 14881514.
Barenblatt, G.I., 1959. The formation of equilibrium cracks during brittle fracture. General ideas and hypothesis. Axially-symmetric
cracks. Prikl. Matem. I mekham 23, 434444.
Barenblatt, G.I., 1962. Mathematical theory of equilibrium cracks in brittle fracture. In: Dryden, H.L., von Karman, T. (Eds.),
Advances in Applied Mechanics, vol. VII. Academic Press, New York, pp. 55125.
Camacho, G.T., Ortiz, M., 1996. Computational modeling of impact damage in brittle materials. International Journal of Solids and
Structures 33, 28992938.
Chandra, N., Ananth, C.R., 1995. Analysis of interfacial behavior in MMCS and IMCS using thin-slice push-out tests. Composite
Science and Technology 54 (1), 87100.
Chandra, N., Ghonem, H., 2001. Interfacial mechanics of push-out tests: theory and experiments. Composites Part AApplied
Science and Manufacturing 32 (34), 575584.
Davidson, D.L., 1992. The micromechanics of fatigue crack growth at 25 C in Ti-6Al-4V reinforced with SCS-6 bers. Metallurgical
Transactions 23A, 865879.

2854

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

Dugdale, D.S., 1960. Yielding of steel sheets containing slits. Journal of the Mechanics and Physics of Solids 8, 100104.
Espinosa, H.D., Dwivedi, S., Lu, H.C., 2000. Modeling impact induced delamination of woven ber reinforced composites with
contact/cohesive laws. Computational Methods in Applied Mechanics and Engineering 183 (34), 259290.
Foulk, J.W., Allen, D.H., Helms, K.L.E., 2000. Formulation of a three-dimensional cohesive zone model for application to a nite
element algorithm. Computational Methods in Applied Mechanics and Engineering 183, 5166.
Geubelle, P.H., Baylor, J., 1998. Impact-induced delamination of laminated composites: a 2D simulation. Composites Part B
Engineering 29 (5), 589602.
Ghosn, L.J., Eldridge, J.I., Kantzos, P., 1994. Analytical modeling of the interfacial stress state during push-out testing of SCS-6/
Ti-based composites. Acta Metallurgica et Materialia 42, 38953908.
Hutchinson, J.W., Evans, A.G., 2000. Mechanics of materials: top-down approaches to fracture. Acta Materialia 48 (1), 125135.
Jan, G.M. van Mier, 1996. Fracture Processes of Concrete, Assessment of Material Parameters for Fracture Models. CRC press, Boca
Raton, Florida, p. 291.
Jayaraman, K., Reifsnider, K.L., Swain, R.E., 1993. Elastic and thermal eects in the interphase: part II. Comments on modeling
studies. Journal of Composites Technology & Research 15 (1), 1422.
Kallas, M.N., Koss, D.A., Hahn, H.T., Hellman, J.R., 1992. Interfacial stress state present in a thin-slice bre push-out test. Journal of
Material Science 27 (14), 38213826.
Kerans, R.J., Parthasarathy, T.A., 1991. Theoretical analysis of the ber pullout and push out tests. Journal of the American Ceramic
Society 74 (7), 15851596.
Kim, J.K., Mai, Y.W., 1991a. High strength high fracture toughness ber composites with interface controlA review. Composites
Science Technology 41 (4), 333378.
Kim, J.K., Mai, Y.W., 1991b. The eect of interfacial coating and temperature on the fracture behaviors of unidirectional KFRP and
CFRP. Journal of Materials Science 26 (17), 47024720.
Kroupa, J.L., Neu, R.W., 1994. The non-isothermal viscoplastic behavior of a titanium-matrix composite. Composites Engineering 4
(9), 965977.
Mohammed, I., Liechti, K.M., 2000. Cohesive zone modeling of crack nucleation at bimaterial corners. Journal of the Mechanics and
Physics of Solids 48, 735764.
Mukherjee, S., Ananth, C.R., Chandra, N., 1997. Eect of residual stresses on the interfacial fracture behavior of metal-matrix
composites. Composite Science and Technology 57 (11), 15011512.
Mukherjee, S., Ananth, C.R., Chandra, N., 1998. Eect of interface chemistry on the fracture properties of titanium matrix
composites. Composites Part AApplied Science and Manufacturing 29 (910), 12131219.
Needleman, A., 1987. A continuum model for void nucleation by inclusion debonding. Journal of Applied Mechanics 54,
525531.
Needleman, A., 1990a. An analysis of tensile decohesion along an interface. Journal of the Mechanics and Physics of Solids 38, 289
324.
Needleman, A., 1990b. An analysis of decohesion along an imperfect interface. International Journal of Fracture 42, 2140.
Needleman, A., 1992. Micromechanical modeling of interfacial decohesion. Ultramicroscopy 40 (3), 203214.
Needleman, A., 1997. Numerical modeling of crack growth under dynamic loading conditions. Computational Mechanics 19 (6), 463
469.
Nimmer, R.P., Bankert, R.J., Russel, E.S., Smith, G.A., Wright, P.K., 1991. Micromechanical modeling of ber-matrix
interface eects in transversely loaded SiC/Ti-6-4 metal matrix composites. Journal of Composites Technology & Research 13
(1), 313.
Osborne, D., Ghonem, H., Chandra, N., 2000. Interface behavior of Ti matrix composites at elevated temperature. Composites Part
AApplied Science and Manufacturing 32 (34), 545553.
Rahulkumar, P., Jagota, A., Bennison, S.J., Saigal, S., 2000. Cohesive element modeling of viscoelastic fracture: application to peel
testing of polymers. International Journal of Solids and Structures 37, 18731897.
Rice, J.R., Wang, J.-S., 1989. Embrittlement of interfaces by solute segregation. Material Science and Engineering A 107, 2340.
Rose, J.H., Ferrante, J., Smith, J.R., 1981. Universal binding energy curves for metals and bimetallic interfaces. Physical Review
Letters 47 (9), 675678.
Sohi, M., Adams, J., Maahapatra, R., 1991. Transverse constitutive response of titanium aluminum metal matrix composites. In:
Desai, C.D. (Ed.), Constitutive Laws for Engineering Materials. ASME Press, New York, pp. 617626.
Tvergaard, V., 1990. Eect of bre debonding in a whisker-reinforced metal. Material Science and Engineering A 125 (2), 203213.
Tvergaard, V., Hutchinson, J.W., 1992. The relation between crack growth resistance and fracture process parameters in elasticplastic
solids. Journal of the Mechanics and Physics of Solids 40 (6), 13771397.
Wappling, D., Gunnars, J., Stahle, P., 1998. Crack growth across a strength mismatched bimaterial interface. International Journal of
Fracture 89, 223243.
Xu, X.P., Needleman, A., 1993. Void nucleation by inclusion debonding in a crystal matrix. Modelling and Simulation in Materials
Science and Engineering 1 (2), 111132.

N. Chandra et al. / International Journal of Solids and Structures 39 (2002) 28272855

2855

Xu, X.P., Needleman, A., 1994. Numerical simulation of fast crack growth in brittle solids. Journal of the Mechanics and Physics of
Solids 42, 13971434.
Xu, X.P., Needleman, A., 1995. Analysis of ductile crack growth by means of a cohesive damage model. International Journal of
Fracture 81, 99112.
Yang, J.M., Jeng, S.M., Yang, C.J., 1991. Failure mechanisms of ber reinforced titanium alloy matrix composites, part I: interface
behavior. Material Science and Engineering A 138, 155167.

You might also like