You are on page 1of 15

Composite Structures 131 (2015) 692706

Contents lists available at ScienceDirect

Composite Structures
journal homepage: www.elsevier.com/locate/compstruct

Effective anisotropic stiffness of inclusions with debonded interface


for Eshelby-based models
Atul Jain a,b,, Yasmine Abdin b, Wim Van Paepegem c, Ignaas Verpoest b, Stepan V. Lomov b
a

Siemens Industry Software NV, Interleuvenlaan 68, B-3001 Leuven, Belgium


Department of Materials Engineering, KU Leuven, Belgium
c
Department of Material Science and Engineering, Ghent University, Belgium
b

a r t i c l e

i n f o

Article history:
Available online 17 June 2015
Keywords:
A Short-ber composites
B Fiber matrix debonding
C Finite element analysis (FEA)
C Micro-mechanics

a b s t r a c t
Inclusions in short ber reinforced composites (SFRC) suffer from debonding and cannot be directly modeled using Eshelby based mean eld methods. This paper proposes a method of treatment of inclusions
with debonded interface by replacing them with a ctitious equivalent bonded inclusion (EqBI) whose
properties are calculated based on the reduced load bearing capacity of the inclusion due to the debonded
interface. Approximate expressions are derived for stress redistribution in an inclusion due to the
presence of debonded interface for the six elementary loading cases and the corresponding terms in
the stiffness tensor are estimated as a function of the reduced average stress in the inclusion.
Mechanical equivalence of the EqBI is conrmed by comparison with nite element models having inclusions with debonded interface and the overall stress strain response of a SFRC composite is validated
against experimental data.
2015 Elsevier Ltd. All rights reserved.

1. Introduction
The properties of a short ber reinforced composites (SFRC)
depend on the properties of its constituents and on the capacity
of stress transfer in the interface [1]. Mean eld homogenization
schemes are computationally cheap, easy to implement and reasonably accurate making them a good candidate for predicting
local effective stiffness properties of SFRC. Different mean-eld
homogenization schemes described in literature depend on the
solution of Eshelby [2] and thus are based on the assumption of
a perfect interface between the inclusions and the matrix. The
MoriTanaka (MT) formulation [3] is probably the most commonly
used mean eld homogenization scheme. It is known to predict
with reasonable accuracy not only the effective response of the
composite but also the stresses in individual inclusions [4] in spite
of formal mathematical inconsistencies appearing in some specic
cases [5].
SFRC, when subjected to loading, suffer from bermatrix
debonding which leads to a debonded interface and consequently
the formulation of Eshelby and MT cannot be used. In this paper
we propose a method to deal with these inclusions with a
Corresponding author at: Department of Materials Engineering, KU Leuven,
Belgium.
E-mail addresses: atul.jain@siemens.com, atul.jain@mtm.kuleuven.be (A. Jain).
http://dx.doi.org/10.1016/j.compstruct.2015.06.007
0263-8223/ 2015 Elsevier Ltd. All rights reserved.

debonded interface by replacing them by an equivalent bonded


inclusion (EqBI). The properties of the EqBI are estimated as a function of the reduced stress transfer in the inclusion due to the
debonded interface. Depending upon the direction of applied load
with respect to the ber axis, the debonded interface surfaces
could lie (A) at the tip or (B) elsewhere on the surface of the
inclusion; we name these two congurations Type A and Type B,
respectively (Fig. 1).
Within the framework of mean-eld homogenization there are
several ways to estimate the onset of debonding in an inclusion in
a composite based on the average stresses in the ber and some
failure criteria; a few of them were proposed in [69]. There is also
a family of methods which rely on Weibull statistics to estimate
the extent of debonding in SFRC [10,11]. Such methods require a
number of additional experimentally determined parameters as
input. Once the state of debonding is calculated the next step is
to model the properties of debonded inclusion.
However, there is limited literature describing the methods to
deal with inclusions having a debonded interface in the framework
of mean-eld homogenization. Zhao and Weng [12,13], Fitoussi
et al. [14] and Mihai [15] proposed a scheme to estimate the
effective properties of a composite containing perfectly bonded
inclusions as well as inclusions with a debonded interface by
replacing the inclusions with a debonded interface by a perfectly
bonded equivalent inclusion, so that the Eshelby [2] solution

A. Jain et al. / Composite Structures 131 (2015) 692706

693

Fig. 1. Schematic representation of the concept of the equivalent bonded inclusion (EqBI) used to calculate the effective properties of SFRC containing inclusions with
debonded interface. Two different congurations of the debonded interface (Types A and B) are shown as EqBI of different color.

and consequently the mean eld homogenization schemes could


be used.
Koyama et al. [16] proposed a method to deal with inclusions
with debonded interface by introducing a virtual matrix; this
approach is deemed to be simplistic and is known to lead to a
lower bound of the elastic properties. Zairi et al. [11] and Nhung
[17] modeled ber matrix debonding by replacing part of the
debonded ber with an equivalent volume of matrix material.
The volume of the additional matrix was determined as a function
of the fraction of debonding.
Caroll and Dharani [18] derived an analytic expression for the
loss of axial elastic modulus in unidirectional SFRC due to debonding initiating at the tip of the bers. They noted that the effects of
debonding in terms of loss of axial stiffness of bers could be
different based on the exact location and extent of debonding.
More work in this eld was done [1923].
All the models described above suffer from certain shortcomings. Zhao and Weng [12,13] and Mihai [15] dealt exclusively with
spherical isotropic inclusions. The loss of modulus in the direction
tangential to the debonded interface is neglected and the exact
area and position of the debonded interface is not taken into
account in the model. These ideas are not directly adaptable for
ellipsoidal inclusions.
If the debonded inclusions are replaced partially by the voids as
was proposed by Koyama et al. [16], Zairi et al. [11] and Nhung
[17], the exact region of debonding cannot be taken into account.
Therefore that approach cannot differentiate between Type A and
Type B debonding. The MT formulation cannot take into account
the exact location of the void, this scheme therefore cannot
conrm that the stress redistribution due to debonded interface
and the average stresses in the debonded inclusion are correctly
estimated.
Also the friction between the ber and the matrix cannot be
accounted for by either of the two sets of methods.
There have been limited attempts to validate predictions of the
average stress in an inclusion with a debonded interface. Average
stresses in individual inclusions are important if further damage
is to be modeled.
The model for treating imperfectly bonded inclusions,
presented in this paper, is based on the original idea of Zhao and
Weng [13], Fitoussi et al. [14] and Mihai [15], who replaced

inclusions with a debonded interface by an equivalent inclusion


with a perfect interface.
Fitoussi et al. derived the properties of an equivalent anisotropic inhomogeneity (EAUI) by estimating the average stresses in a
perfectly bonded ber and a completely debonded ber. The stiffness of the EAUI was estimated as a debonded interface length
weighted average of these two scenarios. The proposed model of
Fitoussi et al. was developed for sheet molding compound (SMC)
composites which have long bers. For long bers, tip debonding
was deemed to be not very important and therefore they only considered debonding at the equator. Their basic hypothesis of interpolation does not hold true for inclusions with Type A debonded
interface as complicated stress redistribution takes place in such
cases. Zhao and Weng [13] and Mihai [15] proposed that the modulus of EqBI can be estimated by assuming that the load bearing
capacity of the inclusion reduces to zero in the direction perpendicular to the debonded interface surface. Such an assumption is
too simplistic and valid only if the inclusions are spherical.
Unlike spherical inclusions, ellipsoidal inclusion can still bear some
load even if the tip is debonded. In injection molded composites,
the aspect ratio of inclusions is much smaller than in SMCs (but
the inclusions are not spherical either) and consequently the stress
build up in the ber is often not high enough for inducing ber
breakage without initiation of ber matrix debonding rst.
Hence, debonded interfaces at the tip of the inclusion must be
accounted for.
Building on the ideas of Fitoussi et al. [14], a model to deal with
both Types A and B debonded interface is developed. This proposed
model relates the stress redistribution due to debonded interface
to the stiffness of EqBI. The proposed methods in this paper can
account for both the exact extent and region of debonding, can take
into account the friction between the ber and the matrix and also
conrm that the average stresses in the debonded inclusion are
correctly estimated.
There is a signicant amount of literature trying to model the
stress redistribution of debonded inclusions, one of the rst being
a paper by Cox [24] which can take into account friction between
ber and matrix, different volume fraction and lengths of the ber.
There are several other papers proposing advanced theories for
stress redistribution of stresses in inclusions due to loading in
the axial direction [2527]. These methods typically are

694

A. Jain et al. / Composite Structures 131 (2015) 692706

computationally expensive and need more input parameters as


compared to the Cox model. In this paper we rely on the expressions given by Cox [24] for the axial loading case and develop
simple relationships for the other loading cases. The Cox model
gives reasonable predictions for the average stress in the inclusion
but unlike recent theories, it cannot predict the stress concentrations at the tip of the debond surface. The proposed methods of this
paper depend upon the average stresses in the inclusions with
imperfect interface and therefore the stress concentrations at the
tip of debond surface is not important.
At this point it is important to note that the shape of the
inclusion is assumed to be ellipsoidal in all the Eshelby based formulations including the MT formulation, while the shape of the
inclusion considered in the Cox model is cylindrical. The real shape
of a glass ber in a SFRC might be closer to a cylinder (chopped or
broken long bers). Modeling bers using ellipsoidal shape is considered an acceptable abstraction and both analytic [28] and FE
based [29,30] studies have conrmed that the average stress/strain
in a cylindrical ber and an equivalent ellipsoidal inclusion do not
differ by more than few percentage. In this paper, the expressions
for the EqBI are derived assuming the shape of the inclusions as
cylindrical while also the shape of the inclusion in the FE-VE model
is considered to be a cylinder. On the other hand, for the MT
scheme the shape of the inclusion is ellipsoidal.
Non-linearity in SFRC is caused mainly by ber matrix debonding and matrix cracking; these damage events have been conrmed by a number of experimental studies [3133]. In this
paper the overall stress strain response prediction will be validated
by a new model to treat ber matrix debonding, while the matrix
non-linearity is modeled by the well known secant modulus
approach.
The main contributions of this paper are the derivation of
expressions for the stiffness tensor of the EqBI for both Types A
and B debonded interfaces. The second contribution is the nite
element (FE) based validation of the mechanical equivalence of
EqBI for each of the six different elementary load cases viz. three
uniaxial tensile and three shear loads. We validate not only the
effective response of the EqBI composite but also the average stresses in the EqBI. Additionally the overall stressstrain behavior of
SFRC is validated using experimental data.
In Section 2, we present the derivations of expressions for the
properties of the EqBI. In Section 3 we present the methodology
for creating the FE models, performing the mean eld homogenization and the experimental procedure. The results of the validation
are presented in Section 4 and conclusions are presented in
Section 5.
2. Mathematical formulation
It is proposed in this paper to replace an inclusion with a
debonded interface by an EqBI. To estimate the stiffness tensor of
an EqBI, a VE consisting of a single inclusion is assumed and
approximate expressions for the stress (re)distribution in the
inclusion due to a debonded interface are derived. The average
stresses in the inclusions are then calculated. The diagonal terms
(C 0ii ) of the stiffness tensor of the EqBI are determined as the
product of the corresponding diagonal terms (Cii) of the stiffness
tensor of the original inclusion and the ratio of the average stress
in the inclusion with debonded interface to the average stress that
would exist in the inclusion, if it were perfectly bonded:

C 0ii

hr
C ii
hrii i
0
ii i

where Cii is the diagonal component of the stiffness tensor which


relates the average stress rii in a composite to applied strain eii ,

r0ii and rii are stresses in the inclusion with debonded and perfect
interface respectively; h_i indicates averaging. The non-diagonal
terms of the stiffness tensor of the EqBI are also calculated assuming
that the Poissons ratio of the EqBI is the same as the original
inclusion. Ideally the non-diagonal terms of the EqBI, Cij must be
calculated based on the reduced average of the stress component
rjj when the applied strain is ejj. However formulations to calculate
the stress in transverse direction of an inclusion (both bonded and
debonded) when subjected to strain loading in the axial direction
are not available. Therefore the non-diagonal terms of the stiffness
of the EqBI are based on the diagonal terms of the stiffness of the
EqBI and the unchanged Poissons ratio.
In the absence of an analytical solution, it is assumed that the
diagonal terms of the EqBI tensor which are calculated on the basis
of the stress redistribution in the axial direction can also somehow
account for the lowered stress in the transverse direction.
For the rest of the paper, z coordinate dene the axis parallel to
the axial direction of the ber, while x and y are the two transverse
directions.
For uniaxial load in the axial direction, Cox [24] proposed a formula for the stress prole of the inclusion with Type A debonded
interface. The prole of the axial stresses in the interface is given
by the following expression

r0f Ef  e1  Ef  e1  2  si  s  m 

 
cosh nzr
coshn  s1  m

si lrres  m1  Em  e1
where r0f are the stresses in the axial direction of the inclusion with
debonded interface. Ef ; m1 are the Youngs modulus and Poissons
ratio of the inclusion, e1 is the applied strain and si is the shear
stress in the interface. rres is the residual stress in the interface, m
is fraction of the inclusion which has debonded interface and z is
the coordinate of the inclusion in the axial direction with the centroid as origin. n is dened by the following expression:

n2

2  Gm
Ef  ln Rr

where Gm is the shear modulus of the matrix and the term R/r is the
ratio of the radius of the VE and ber, it is estimated as a function of
the volume fraction of the inclusion and s is the aspect ratio of the
inclusion. The stress r0f is integrated over the length of the inclusion
to get the average value of stress in the inclusion.
For the Type B debonded interface, we assume that there is no
signicant loss of stiffness in the axial direction as the stress transfer has been completed when moving from the tip to the center of
the ber.
When subjected to loading in the transverse direction (y or x),
the stress transfer in the inclusion is limited solely to the regions
with perfect interface while there is negligible stress transfer in
the regions where there is a debonded interface. Thus we assume
that the average stress in the region with debonded interface is
zero whereas the average stress in the regions with perfect interface is assumed to remain unchanged. If the applied load is in
the y-direction, the stress in the inclusion in the y direction is

r0incl rf for perfect interface


r0incl 0 for debonded interface

where r0incl is the stress in the inclusion in the applied load direction
(y), rf is the average stress in the inclusion in the applied load direction that would be present if there was perfect interface in the
inclusion. The average stress in the inclusion can thus be derived
as r0f v proj  rincl =v incl , where v proj is the projected volume of the
inclusion segment with debonded interface in the normal direction

695

A. Jain et al. / Composite Structures 131 (2015) 692706

of load and v incl is the volume of the inclusion. A similar derivation


can be done for loading in the second transverse direction, leading
to an expression of stiffness tensor component Cxx of the EqBI.
When an inclusion is subjected to shear loading, then there is
stress transfer in the debonded interface due to sliding friction.
Stress transfer to the inclusion due to sliding friction takes place
in the interface where the interface stresses in the outward normal
direction of the ber are negative. In other parts of the debonded
interface where there is positive normal stress in the interface,
the matrix simply separates from the inclusion and there is zero
stress transfer. The stress transfer in those parts of inclusions
which is perfectly bonded is assumed to remain unchanged.
These assumptions are exactly the same as used by Fitoussi et al.
[14], but the implementation is slightly different. Mathematically,

r0incl rf if interface is perfect


r0incl lrrr if interface is debonded and rrr is negative
r0incl 0 if interface is debonded and rrr is positive

where r0incl is the stress in the inclusion in the applied load direction,
rf is the average stress in the inclusion in the applied load direction
that would be present if there was a perfect interface in the inclusion. rrr is the stress component in the outward normal direction
of the inclusion if there was perfect interface in the inclusion and
l is the coefcient of friction. The average stress in the inclusion
can then be calculated as a surface area weighted average of the
stresses in the inclusion. A summary of the expressions is presented
in Table 1.
The stiffness tensor of the EqBI thus derived is expressed in the
local co-ordinate system of the inclusion and must be transformed
to the global coordinate system for further application of the MT
formulation. The effect of the orientation of the debonded bers
is accounted for in the MT formulation by means of the image
strain during the implementation of the MT formulation [5].
3. Methodology of numerical experiments and experimental
validation
The material considered for the validation is a Polybutylene
terephthalate (PBT) matrix reinforced with glass ber, with a
Youngs modulus of 2.9 and 72 GPa, respectively and Poissons
ratio of 0.37 and 0.22, respectively.

3.1. Creation of the Finite Element Model


A number of FE VE was created to validate the mechanical
equivalence of the two composite systems, viz. the composite with
imperfect interface, and the composite containing an EqBI. The
mechanical equivalence here implies that both the effective
response of composite systems and the average stresses in the
inclusions must be similar for every possible type of loading.
According to the interpretation of the MT formulation by
Beneveniste [5], each inclusion is assumed to be surrounded by
matrix and the inuence of the other inclusions are felt indirectly
through the matrix by means of the image strain. Therefore, for the
validation of the mechanical equivalence of EqBI, it is enough to
model a single inclusion surrounded by matrix and then validate
the mechanical equivalence for six elementary load cases (three
uniaxial tensile and three shear loads).
We created a FE-VE with a single inclusion inside a matrix
placed in such a way that the centroids of the matrix cuboid and
the cylindrical inclusion coincide (Fig. 2). The aspect ratio of the
inclusion is taken to be 15 and the volume fraction is taken to be
0.1. This aspect ratio is chosen as it is a typical value of aspect ratio
of inclusions in injection molded SFRC component. The cylindrical
shape of the inclusion in the nite element VE is chosen to ensure
acceptable mesh quality of the inclusion.
Periodic boundary conditions are applied to the cubic cell by
applying the following equations:

ux; y; 0  uz ux; y; L
ux; 0; z  uy ux; L; z

u0; y; z  ux uaL; y; z
where, u is the displacement vector on the different faces of the
cube and ux, uy, uz depends on the particular loading applied to
the cubic cell. Uniaxial strain ex is applied by ux = (exaL, 0, 0),
uy = (0, uy, 0) and uz = (0, 0, uz). uy and uz is then computed from
R
R
the conditions. X T y dX 0 on y = L and X T z dX 0 on z = L.
Where Ty and Tz are the normal tractions acting on the prism faces
contained in the transverse planes y = L and z = L. Similar boundary
conditions can be applied for different loading directions as well.
Commercial FE solver ABAQUS version 6.13-3 [34] was used for
the calculations.

Table 1
A summary of the expressions of the stresses in inclusions with debonded interface, r0incl .The stiffness of the EqBI is calculated as a function of the reduced stress in the inclusion
with the debonded interface.
Stiffness
component
Czz

Cyy

Type A
 
cosh nz
r
coshn  s1  m
2

G
m
where, si lrres  m1  Em  e1 and n2
Ef :ln Rr
Average stress is calculated by integration over length of inclusion, z;
other terms have usual meanings

r0incl Ef  e1  Ef  e1  2  si  s  m 

r0incl rf if interface is perfect


r0incl 0 if interface is debonded

Type B
No change in average
stress

Same as type A

Average stress is calculated by volume weighted averaging

rf is the average stress in the inclusion that would be present if there was
perfect interface in the inclusion
Cxx

Same as Cyy

Same as Type A

Cxy, Cxz, Cyz

r0incl rf if interface is perfect


r0incl lrrr if interface is debonded and rrr is negative
r0incl 0 if interface is debonded and rrr is positive

Same as Type A

Average stress is calculated by surface area weighted averaging of r0incl


rrr is the stress component in the outward normal direction of the inclusion if there was perfect interface in the
inclusion

696

A. Jain et al. / Composite Structures 131 (2015) 692706

Fig. 2. Finite element modeling of the single ber FE VE geometry and contact surfaces to model debonded interfaces. The blue body is ber. The red surface indicates contact
surface used to model debonded interface, (a) Geometry of the FE VE showing inclusion and the matrix and the datum lines used to section the surface of the inclusions
surface. (b) Type A debonded interface with fraction of debonded interface is 4/15. (c) Type B debonded interface with fraction of debonded interface is 4/15. (For
interpretation of the references to colour in this gure legend, the reader is referred to the web version of this article.)

For the meshing structured hexahedron elements C3D8R were


used. The mesh size was chosen by performing a convergence analysis. We choose the seed size for mesh generation to be one-tenth
of the diameter of the cylindrical inclusion, since on further reduction of the seedsize no signicant change was seen in the value of
the average stresses in the inclusions.
Both Types A and B debonded interfaces are modeled by partitioning the inclusion and the matrix and using appropriate contact
surface properties. The fraction of debonded interface length was
chosen in such a way that the increase in the debonded length fraction was progressively higher. Higher number of debonding length
fractions can be studied easily but ve debonded length fractions
were considered enough to compare the average stresses in the
inclusions and to study the trend in loss of load bearing stiffness
due to debonding. Five FE models with debonded interface length
fractions 0, 2/15, 4/15, 8/15 and 14/15 were created for modeling
the Type A debonded interface.
For the Type B debonded interface, the aspect ratio of the rectangle with debonded interface is chosen to be the same as the
aspect ratio of the inclusion. This particular choice of the debonded
surface conguration is chosen so that the debonded interface

surface can now be characterized by only value (the fraction of


debonded interface length). The outward normal of the Type B
interface is across the x-axis with the value of fraction of debonded
interface length same as Type A.
The friction formulation in the debonded interface is chosen as
static-kinetic exponential decay. This formulation allows modeling the friction between the surface of the inclusion and the
corresponding matrix surface with a smooth transition between
the two. The coefcient of static friction is taken to be 0.28; kinetic
friction is 0.1, the values of friction are as reported by Chua and
Piggott [35] and are used also for the calculations of the properties
of the EqBI. The normal behavior of the debonded interface is
modeled as a hard contact, where the two bodies are allowed to
separate, but not penetrate into each other. The homogenization
of the composite containing EqBI is performed by the MT
algorithm. For the Type A debonded interface, the geometry and
location of the debonded interface is transversely isotropic, therefore we performed calculations for four load cases viz., tensile
loads in axial and transverse load directions and shear in
in-plane and out of plane directions. For the Type B debonded
interface, all six possible elementary load cases (three uniaxial

697

A. Jain et al. / Composite Structures 131 (2015) 692706

tensile and shear) were studied. The average stresses in both the
VE (ber and matrix) and inclusion are calculated by volume averaging the stresses in different regions.
3.2. Experimental tests
For the experimental validations, a BASF Ultradur B4040 G10
compound was injection molded to plates 170  170  2 mm.
This material consists of PBT reinforced with 50% weight fraction
of glass ber (equivalent volume fraction is 35%). The ber volume
fraction is chosen to be as high as possible since it is expected that
the higher content of ber will lead to higher ber related damage,
especially ber matrix debonding. Higher content of ber leads to
higher stresses in the ber and also there is a larger interface area
which can debonded.
Coupons were machined from the plates in three directions,
inclined at angles U = 0, 45 and 90 with respect to the prevailing

ow direction (Fig. 3). Dynamic extensometers were mounted on


the coupons and load controlled tensile tests were performed.
The manufacturing simulation was performed using the software SIGMASOFT [36] to estimate the orientation distribution of
the inclusions. Particular attention was paid to the region comprising the gauge length of the coupon. The simulation results showed
that there was very little variation of the 2nd order orientation
tensor along the gauge length of the coupon. This conrmed that
the orientation distribution of the inclusions is similar in the gauge
length of the coupon and therefore we could treat the entire coupon as a single RVE during further simulation.
The orientation tensor of the 0-degree coupon was found to be

0:81

6
aij 4 0:018

0:018 0:137
0:11

0:137 0:004

7
0:004 5

0:079

The orientation tensor of the 45 and 90-degree coupons were calculated by rotating of the 0 orientation tensor.
3.3. Mean eld homogenization scheme for the composite
A realization of 500 inclusions is generated from the 2nd order
orientation tensor described in previous section using the formulation described by Onat and Leckie [37]. Small load increments are
applied per time step and the onset of debonding is checked at
each time step. The onset of debonding is calculated using the
well-known Modied Coulomb criteria which account for both
normal and shear stress. This criterion has been extensively used
for checking the onset of debonding in ellipsoidal inclusions,
including SFRC [3840]. The yield strength of the PBT matrix is
assumed to be 56 MPa [41], the value of the shear contribution
coefcient is taken to be 0.5 as advised by Huysmans et al. [39].
The modied Coulomb criterion is applied at a number of
equally spaced locations along the three orthogonal cross-section
surfaces of each ellipsoid inclusion (Fig. 4). For the calculations in
this paper, the number of points is set to 100, since it was seen that
there was no difference in the simulated stress strain curve if the
number of points for debonding check is increased.
At the end of each time step, the debonded inclusion is replaced
by EqBI and the MT calculations as well as further damage checks
are performed on a modied RVE which contains both undamaged
inclusions as well as EqBIs. The stiffness of the EqBI is predicted in
the ber co-ordinate system and is transformed to the global RVE
direction before a full MT formulation can be used. Matrix
non-linearity is implemented using the well-known secant
approach originally proposed by Tandon and Weng [42] and implemented for SFRC [40]. The input for modeling the secant model is
the stressstrain curve of the matrix, we have used the stress strain
data of pure PBT matrix (BASF Ultradur B4520) [41].
Two sets of stress strain simulations are performed for each
coupon. In the rst simulation only matrix non-linearity is treated.
While for the second simulation both ber matrix debonding and
matrix non-linearity are considered.
4. Results
In this section we present the validations of the proposed
models for both Types A and B. For both the cases rst we look

Fig. 3. Specimens preparation (a) plates used for injection molding, the point of
injection is the center region in between the two plates. (b) The geometry of the dog
bone specimen with dimensions in mm (c) orientation designation of coupon.

Fig. 4. The three orthogonal cross-sections where the onset of debonding is


checked. Equally spaced points are considered for each of the three cross sections.

698

A. Jain et al. / Composite Structures 131 (2015) 692706

at the FE contours to examine the assumptions made during the


formulations (Section 2) and next present the results for the properties of EqBI, average stresses in the VE and the EqBI.
4.1. Type A debonded interface
4.1.1. Stress redistribution
The contour stress plots calculated by the FE calculations are
shown below in Fig. 5(a) and (b) for axial and transverse loading
cases respectively for Type A debonded interface.
A qualitative analysis of the plots conrms the assumptions
made during derivations of expressions (Eqs. (1)(4)). The
FE-calculated stress redistribution in the inclusion with a
debonded interface, which is subjected to axial loading, follows
the trend as predicted by the formulation of Cox [24]. There is negligible stress transfer in the part of the inclusion which has a

debonded interface; this is mainly due to sliding friction between


the ber and matrix surfaces, which is also accounted for by the
Cox model. A comparison of the stress proles of rzz on the surface
of an inclusion with Type A debonded interface (fraction of the
debonded interface surface is 4/15) predicted by the FE calculations and the Cox model is presented in Fig. 6(a). There is some
noise and scatter in predicted stress values at the tip of interface,
which the Cox model cannot take into account, also it is seen that
the peak stress predicted by the Cox model is slightly higher than
the FE predictions. Also there is some difference in the non-linear
stress prole predicted by the Cox model and the FE calculations.
Barring this there is good match between the FE and Cox model
predictions.
The part of the inclusion which has a perfect interface is
stressed higher but the distribution of stress is not uniform with
maximum stress at the center. The magnitude of stresses in the

Fig. 5. Stress distribution in inclusion with debonded interface type A, fraction of debonded surface is 0, 0.133, 0.266, 0.533 and 0.93, respectively (top to bottom). (a):
Applied strain is 1% in the z-direction. (b) Applied strain is 1% in the x direction.

A. Jain et al. / Composite Structures 131 (2015) 692706

699

Fig. 6. Comparison of stress distribution by Cox model and FE. (a) Stress distribution in inclusion surface with debonded interface type A, fraction of debonded surface is
0.266. by FE and the Cox model (b) Peak stress in inclusions with debonded interface by the Cox model and FE calculations.

Fig. 7. Stiffness predictions of the EqBI for different fractions of Type A debonded interface.

700

A. Jain et al. / Composite Structures 131 (2015) 692706

Fig. 8. Average stresses in the inclusions for different loads and different fractions of
Type A debonded interface at the tip of inclusion by both the MT formulation of VE
with EqBI and FE VE consisting of a single inclusion. Aspect ratio of inclusion is 15,
vf = 0.1 (a) applied load, ezz is 1% uniaxial strain. (b) applied load, exx is 1% uniaxial
strain. (c) applied load, exy is 1% shear strain. (d) applied load, eyz is 1% shear strain.

Fig. 9. Average stresses in the VE for different loads and different fractions of Type
A debonded interface at the tip of inclusion by both the MT formulation of VE with
EqBI and FE VE consisting of a single inclusion. Aspect ratio of inclusion is
15, vf = 0.1 (a) applied load, ezz is 1% uniaxial strain. (b) applied load, exx is 1%
uniaxial strain (c) applied load, exy is 1% shear strain. (d) applied load, eyz is 1% shear
strain.

A. Jain et al. / Composite Structures 131 (2015) 692706

701

Fig. 10. Stress redistribution in inclusion with debonded interface Type B, fraction of debonded surface is 0.133, 0.266, 0.533 and 0.93, respectively (top to bottom). Applied
strain is 1%. (a) loading z-direction. (b) loading y direction. (c) loading x direction.

702

A. Jain et al. / Composite Structures 131 (2015) 692706

center of inclusion reduces progressively as the region with


debonded interface increases. There is good match between the
Cox model and FE results for the 5 cases (Fig. 6(b)).
When the inclusion is subjected to the transverse loading there
is negligible stress in the part of the inclusion with debonded interface. The part of the inclusion with perfect interface is uniformly
stressed and the magnitude of stress is independent of the extent
of the debonded interface between the inclusion and matrix.

Fig. 11. Stiffness predictions of the EqBI for different extents of Type B debonded
interface.

4.1.2. EqBI properties and mechanical equivalence


The properties of the EqBI are presented in Fig. 7 as function of
the fraction of the debonded surface.
A comparison of the real average stresses, calculated by FE,
and the estimated average stresses using the MT-inclusion
model, after replacing the real debonded inclusions by their EqBI
is presented. Comparison of the predictions of the average stresses

Fig. 12. Average stresses in the inclusions for different loads and different fractions of Type B debonded interface by both the MT formulation of VE with EqBI and FE VE
consisting of single inclusion. Aspect ratio of inclusion is 15, vf = 0.1 (a) applied load, ezz is 1% uniaxial strain. (b) applied load, exx is 1% uniaxial strain. (c) applied load, eyy is 1%
uniaxial strain. (d) applied load, exy is 1% shear strain. (e) applied load, exz is 1% shear strain. (f) applied load, eyz is 1% shear strain.

A. Jain et al. / Composite Structures 131 (2015) 692706

703

Fig. 13. Average stresses in the VE for different loads and different fractions of Type B debonded interface by both the MT formulation of VE with EqBI and FE VE consisting of
single inclusion. Aspect ratio of inclusion is 15, vf = 0.1 (a) applied load, ezz is 1% uniaxial strain. (b) applied load, exx is 1% uniaxial strain. (c) applied load, eyy is 1% uniaxial
strain. (d) applied load, exy is 1% shear strain. (e) applied load, exz is 1% shear strain. (f) applied load, eyz is 1% shear strain.

in the inclusions (Fig. 8(a)(d)) and VE (Fig. 9(a)(d)) by the Full FE


calculations and the MT formulation showed that there was a good
match for different types of loading. It is seen that there is a better
match when the average stresses in the VE is compared to the
stresses in the inclusions.
For the loading in the axial direction (Figs. 8(a) and 9(a)) the
average stresses in the EqBI and EqBI composite are slightly higher
than the FE predicted values, this can be ascribed to the fact that
the Cox formulation which was used to predict the stiffness of
the EqBI over predicted the peak stress value. For the other load
cases, the correlation between the EqBI based MT formulation
and FE model is reasonably good.
The biggest errors are noticed when the EqBI based MT formulation and FE predicted average stresses are compared for an
applied load of ryz. However, for a perfectly bonded inclusion
(debonding fraction = 0), the difference in the average stresses in
the inclusions predicted by the MT formulation and FE was also

seen to be the highest when the shear load ryz is considered. The
error in the proposed scheme could be due to the fact that the
inclusion is modeled as a cylinder in the FE model, while the MT
formulation treats the inclusion as an ellipsoid.
It is also seen that the average stresses reduces in a different
manner for different uniaxial loading. This also explains the different trends predicted for the stiffness of the EqBI and conrms
that different components of the stiffness tensor must be treated
in a different manner during the calculations of the stiffness of
the EqBI.
4.2. Type B debonded interface
4.2.1. Stress redistribution
Fig. 10(a)(c) show FE stress contours for axial and the two
transverse loading for VE containing inclusions with Type B
debonded interface.

704

A. Jain et al. / Composite Structures 131 (2015) 692706

Fig. 14. Experimentally derived and simulated stress strain behavior of injection molded SFRC (Cauchy strain is plotted); (a) 0-degree coupon (b) 45-degree coupon (c) 90degree coupon.

A. Jain et al. / Composite Structures 131 (2015) 692706

No change in the stress distribution is observed for inclusion


with Type B debonded interface when the VE is subjected to a load
in the axial direction (z-direction). When subjected to the loading
in the transverse direction normal to the debonded interface surface, there is negligible stress in the part of the inclusion with
debonded interface. The part of the inclusion with perfect interface
is uniformly stressed and the magnitude of stress is independent of
the extent of the debonded interface. This conrms the assumptions during the formulation.
4.2.2. EqBI properties and mechanical equivalence
For each of the 5 extents of debonded interface, the normalized
stiffness values of the EqBI were calculated and seen to vary differently (Fig. 11).
Comparison of the predictions of the average stresses in the
inclusion (Fig. 12(a)(f)) and the EqBI composite (Fig. 13(a)(f))
by the full FE calculations and the MT formulation showed that
there is a good match for different types of loading. It is seen that
in general the loss of stress bearing capacity of the composite is
lower than observed for Type A debonding.
It is seen that the match for the average stress (both for EqBI and
EqBI composite) matches well with the FE results for uni-axial
loads. Both the FE and EqBI based MT formulation predict no loss
in load bearing capacity of the composite when subjected to uniaxial load. When the loading is in the transverse direction, the drop in
average stress in the inclusion is gradual for low values of the fraction of debonded interface. There is a sudden drop in the average
stress when the debonded interface fraction is above 0.5. The
EqBI based MT formulation is able to predict exactly the same trend
as the FE calculations.
However, for the shear loads the difference in the average stress
value in the EqBI composite and the inclusion in the FE is slightly
higher. It is unclear whether the difference in the average stress
is due to the different shape considered or it is due to the approximations in the calculation of the EqBI properties. This is because
the difference in the average stresses (particularly in the inclusion)
was highest when the fraction of debonded length considered was
0 (fully bonded inclusion).
From the results it is also clear that the loss of load bearing
capacity is different for each loading type and is estimated by the
proposed method. Subsequently each component of stiffness
matrix of EqBI therefore varies in a different manner and is taken
into account by the proposed model.
4.3. Experimental validation
A good match was also observed in the overall stress strain
behavior prediction for the SFRC coupon when both ber matrix
debonding and material non-linearity are taken into account.
(Fig. 14). It is concluded that modeling ber matrix debonding
together with the matrix non-linearity leads to an improvement
in the predictions of the overall stress strain response. This is particularly true for the 0 and 45 degree coupons but for the
90-degree coupon the major cause of non-linearity is due to matrix
non-linearity.
Tensile failure of the coupon is probably due to complex damage events which is not yet accounted for by the proposed methods
described in this paper.

5. Conclusions
A method for treating debonded interface inclusions under the
MT formulation by the introduction of an EqBI was proposed and
validated in this paper. The proposed method takes into account
the stress distribution and corresponding reduced load bearing

705

capacity of a ber due to the presence of debonded interface. The


stress distribution due to the presence of a debonded interface is
different for different applied loads and subsequently the components of the stiffness tensor of the EqBI vary differently.
Validation of the mechanical equivalence of the EqBI was performed by modeling a VE with debonded interface in FE and comparing the predictions of the average stresses in the VE as well as in
the inclusions. A good match was observed for the average stresses
both for the full VE and in the inclusions for all possible types of
elementary loading. Good predictions were also obtained for overall stress strain behavior of SFRC. The derived algorithms for creation of EqBI are ready to use in MT modeling of SFRC.
Acknowledgments
IWT Vlaanderen is thanked for funding this research as a part of
the project Fatigue life prediction of random ber composites
using hybrid multi-scale modelling methods COMPFAT
Baekeland mandate number 100689. PART Engineering GmbH
opened suitable interfaces in their software Converse to facilitate
the analysis presented in this paper. Stefan Straesser, Dr. Michael
Hack and Christophe Liefooghe from Siemens Industry Software
are thanked for useful discussions and constructive feedback. I.
Verpoest is holder of the Toray Chair in Composites at KULeuven.
References
[1] Thomason JL. Interfacial strength in thermoplastic composites at last an
industry friendly measurement method? Compos Part A Appl Sci Manuf
2002;33(10):12838.
[2] Eshelby JD. The determination of the elastic eld of an ellipsoidal inclusion,
and related problems. Proc R Soc London Ser A Math Phys Sci
1957;241(1226):37696.
[3] Mori T, Tanaka K. Average stress in matrix and average elastic energy of
materials with mistting inclusions. Acta Metall 1973;21(5):5714.
[4] Jain A, Lomov SV, Abdin Y, Verpoest I, Paepegem WV. Pseudo-grain
discretization and full Mori Tanaka formulation for random heterogeneous
media: predictive abilities for stresses in individual inclusions and the matrix.
Compos Sci Technol 2013;87:8693.
[5] Benveniste Y. A new approach to the application of MoriTanaka theory in
composite-materials. Mech Mater 1987;6(2):14757.
[6] Lawrence P. Some theoretical considerations of ber pull-out from an elastic
matrix. J Mater Sci 1972;7(1):16.
[7] Leung CKY, Li VC. Applications of a 2-way debonding theory to short ber
composites. Composites 1990;21(4):30517.
[8] Takaku A, Arridge RGC. The effect of interfacial radial and shear stress on bre
pull-out in composite materials. J Phys D: Appl Phys 1973;6(17):2038.
[9] Meraghni F, Blakeman CJ, Benzeggagh ML. Effect of interfacial decohesion on
stiffness reduction in a random discontinuous-bre composite containing
matrix microcracks. Compos Sci Technol 1996;56(5):54155.
[10] Desrumaux F, Meraghni F, Benzeggagh ML. Generalised MoriTanaka scheme
to model anisotropic damage using numerical Eshelby tensor. J Compos Mater
2001;35(7):60324.
[11] Zairi F, Nait-Abdelaziz M, Gloaguen JM, Bouaziz A, Lefebvre J-M.
Micromechanical modelling and simulation of chopped random ber
reinforced polymer composites with progressive debonding damage. Int J
Solids Struct 2008;45(20):522036.
[12] Zhao YH, Weng GJ. Plasticity of a two-phase composite with partially
debonded inclusions. Int J Plast 1996;12(6):781804.
[13] Zhao YH, Weng GJ. Transversely isotropic moduli of two partially debonded
composites. Int J Solids Struct 1997;34(4):493507.
[14] Fitoussi J, Bourgeois N, Guo G, Baptiste D. Prediction of the anisotropic
damaged behavior of composite materials: introduction of multilocal failure
criteria in a micromacro relationship. Comput Mater Sci 1996;5(1
3):87100.
[15] Mihai I. Micromechanical constitutive models for cementitious composite
materials. School of Engineering [Ph.D.]. Cardiff: Cardiff University; 2012.
[16] Koyama S, Katano S, Saiki I, Iwakuma T. A modication of the MoriTanaka
estimate of average elastoplastic behavior of composites and polycrystals with
interfacial debonding. Mech. Mater. 2011;43(10):53855.
[17] Nhung LTT. Prevision de la loi de comportement des composites bmc.
Mcanique et Matriaux [Ph.D.]. Paris; 2011.
[18] Carroll DR, Dharani LR. Elastic properties of imperfectly bonded short ber
composites. Compos Struct 1996;35(2):195206.
[19] Benveniste Y. The effective mechanical behaviour of composite materials with
imperfect contact between the constituents. Mech Mater 1985;4(2):197208.
[20] Gorbatikh L, Lomov S, Verpoest I. Elastic compliance of a partially debonded
circular inhomogeneity. Int J Fract 2005;131(3):21129.

706

A. Jain et al. / Composite Structures 131 (2015) 692706

[21] Ju JW, Lee HK. A micromechanical damage model for effective elastoplastic
behavior of ductile matrix composites considering evolutionary complete
particle debonding. Comput Methods Appl Mech Eng 2000;183(34):20122.
[22] Lee HK, Pyo SH. Micromechanics-based elastic damage modeling of particulate
composites with weakened interfaces. Int J Solids Struct 2007;44(25
26):8390406.
[23] Yanase K, Ju JW. Effective elastic moduli of spherical particle reinforced composites
containing imperfect interfaces. Int J Damage Mech 2012;21(1):97127.
[24] Cox HL. The elasticity and strength of paper and other brous materials. Br J
Appl Phys 1952;3(72):729.
[25] Amirbayat J, Hearle JWS. Properties of unit composites as determined by the
properties of the interface. Part I: Mechanism of matrix-bre load transfer.
Fibre Sci Technol 1969;2(2):12341.
[26] Nairn JA. A variational mechanics analysis of the stresses around breaks in
embedded bers. Mech Mater 1992;13(2):13154.
[27] Wu W, Verpoest I, Varna J. A novel axisymmetric variational analysis of stress
transfer into bres through a partially debonded interface. Compos Sci Technol
1998;58(12):186377.
[28] Goh KL, Aspden RM, Mathias KJ, Hukins DWL. Finite-element analysis of the
effect of material properties and bre shape on stresses in an elastic bre
embedded in an elastic matrix in a bre-composite material. Proc R Soc A
Math Phys Eng Sci 2004;460(2048):233952.
[29] Buryachenko V. Micromechanics of heterogeneous materials. Springer; 2007.
[30] Giordano S. Analytical procedure for determining the linear and nonlinear
effective properties of the elastic composite cylinder. Int J Solids Struct
2013;50(24):405569.
[31] Arif MF, Meraghni F, Chemisky Y, Despringre N, Robert G. In situ damage
mechanisms investigation of PA66/GF30 composite: effect of relative
humidity. Compos Part B Eng 2014;58:48795.

[32] Mouhmid B, Imad A, Benseddiq N, Benmedakhene S, Maazouz A. A study of the


mechanical behaviour of a glass bre reinforced polyamide 6,6: experimental
investigation. Polym Test 2006;25(4):54452.
[33] Sato N, Kurauchi T, Sato S, Kamigaito O. Microfailure behavior of randomly
dispersed short ber reinforced thermoplastic composites obtained by direct
sem observation. J Mater Sci 1991;26(14):38918.
[34] ABAQUS. A general-purpose nite element software, ABAQUS Inc, Pawtucket
RI, USA; 2005.
[35] Chua PS, Piggott MR. The glass brepolymer interface: pressure and
coefcient of friction. Compos Sci Technol 1985;22(3):18596.
[36] Sigmasoft. SIGMA Engineering GmbH, Aachen. In: http://www.sigmasoft.de/,
editor.; 2014.
[37] Onat ET, Leckie FA. Representation of mechanical behavior in the
presence of change internal structure. Trans ASME J Appl Mech
1988;55(1):110.
[38] Furuhashi R, Huang JH, Mura T. Sliding inclusions and inhomogeneities with
frictional interfaces. J Appl Mech 1992;59(4):7838.
[39] Huysmans G, Verpoest I, Van Houtte P. A damage model for knitted fabric
composites. Compos Part A Appl Sci Manuf 2001;32(10):146575.
[40] Jao Jules SL, Ignace Verpoest, P. Naughton, A. W. Beekman, Van Daele R.
Prediction of non-linear behaviour of discontinuous long glass bres
polypropylene composites. In: 15th ICCM Durban; 2005.
[41] PLASTICS C. CAMPUS a material information system for the plastics
industry. <http://www.campusplastics.com/>; 2014.
[42] Tandon GP, Weng GJ. A theory of particle-reinforced plasticity. J Appl Mech
Trans ASME 1988;55(1):12635.

You might also like