Mechanics of Materials: Kamran A. Khan, Romina Barello, Anastasia H. Muliana, Martin Lévesque

Mechanics of Materials 43 (2011) 608625
Contents lists available at ScienceDirect
Mechanics of Materials
journal homepage: www.elsevier.com/locate/mechmat
Coupled heat conduction and thermal stress analyses

in particulate composites
Kamran A. Khan a, Romina Barello b, Anastasia H. Muliana a,, Martin Lvesque b
a
b
Department of Mechanical Engineering, Texas A&M University, USA

CREPEC, Department of Mechanical Engineering, Ecole Polytechnique de Montreal, Canada
a r t i c l e
i n f o
Article history:
Received 2 May 2010
Received in revised form 15 April 2011
Available online 13 July 2011
Keywords:
Heat conduction
Thermal stresses
Particulate composites
Finite element
Micromechanical model
a b s t r a c t
This study introduces two micromechanical modeling approaches to analyze spatial variations of temperatures, stresses and displacements in particulate composites during transient heat conduction. In the rst approach, a simple micromechanical model based on a
rst order homogenization scheme is adopted to obtain effective mechanical and thermal
properties, i.e., coefcient of linear thermal expansion, thermal conductivity, and elastic
constants, of a particulate composite. These effective properties are evaluated at each
material (integration) point in three dimensional (3D) nite element (FE) models that represent homogenized composite media. The second approach treats a heterogeneous composite explicitly. Heterogeneous composites that consist of solid spherical particles
randomly distributed in homogeneous matrix are generated using 3D continuum elements
in an FE framework. For each volume fraction (VF) of particles, the FE models of heterogeneous composites with different particle sizes and arrangements are generated such that
these models represent realistic volume elements cut out from a particulate composite.
An extended denition of a RVE for heterogeneous composite is introduced, i.e., the number of heterogeneities in a xed volume that yield the same expected effective response for
the quantity of interest when subjected to similar loading and boundary conditions. Thermal and mechanical properties of both particle and matrix constituents are temperature
dependent. The effects of particle distributions and sizes on the variations of temperature,
stress and displacement elds are examined. The predictions of eld variables from the
homogenized micromechanical model are compared with those of the heterogeneous
composites. Both displacement and temperature elds are found to be in good agreement.
The micromechanical model that provides homogenized responses gives average values of
the eld variables. Thus, it cannot capture the discontinuities of the thermal stresses at the
particlematrix interface regions and local variations of the eld variables within particle
and matrix regions.
2011 Elsevier Ltd. All rights reserved.
1. Introduction
The existence of thermal stresses has always been a
subject of discussion when a body is subjected to coupled
heat conduction and mechanical loadings. In composites,
signicant thermal stresses can arise due to the mismatch
in the coefcient of thermal expansions (CTEs) of the con Corresponding author.
E-mail address: amuliana@neo.tamu.edu (A.H. Muliana).
0167-6636/$ - see front matter 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mechmat.2011.06.013
stituents, affecting their overall performance. The use of

composite materials in structural components requires
understanding the variations in the eld variables, such
as stress, strain, temperature and displacement, both at
micro- and macro scales. At the macro scale, composite
structures are often analyzed as homogeneous structures
through the use of effective properties, which allow performing large-scale structural analyses. Composites are
heterogeneous materials that can exhibit large variations
and even discontinuities in the eld variables. These
K.A. Khan et al. / Mechanics of Materials 43 (2011) 608625
609
Fig. 1. Schematic diagram of integration of micro-macro scale approach for particulate composites.
Fig. 2. Homogenization of the sub-regions of a composite component.
variations and discontinuities at the micro-scale cannot be

captured if one treats composites as homogenous (homogenized) materials.
Different types of micromechanical models with simplied microstructures have been developed to obtain effective thermal and mechanical properties. The composite
spheres model (Hashin, 1962), the self consistent approach
(Budiansky, 1965 and Hill, 1965); the generalized self
consistent scheme (Christensen and Lo, 1979, 1986), the

Mori-Tanaka model (Mori and Tanaka, 1973), the probabilistic approach of Chen and Acrivos (1978), the differential
method (McLaughlin, 1977 and Norris, 1985) are some
examples. Detailed discussion of various micromechanical
models and bounds on the effective mechanical properties
can be found in Aboudi (1991), Mura (1987), Nemat-Nasser and Hori (1999). Khan and Muliana (2010) presented
610
Fig. 3. 3D FE models of (a) homogenized and (b) heterogeneous composites. (c) Schematic of thermal and mechanical loading directions and proles along
which eld variables are evaluated, i.e., AB, CD, EF and GH.
Table 1
Temperature dependent mechanical and physical properties of materials of Ti6Al4V and ZrO2 used in 3D FE analyses.
Property
Ti6Al4V
Zirconia (Zr02)
Young modulus (E) (Pa)

Poisson ratio (t)
Coefcient of thermal expansion (a)106,
1/K
Thermal conductivity, (k), W/m/K
Specic heat (c), J/kg K
Density (q), kg/m3
1.23 101156.46 106T

0.3
7.58 106 + 4.93 109
T + 2.39 1012 T2
1.21 + 0.0169 T
625.297 0.264 T + 4.49 104 T2
4429
2.44 1011334.28 106T + 295.24 103T2 89.79 T3

0.3
1.28 105 19.07 109 T + 1.28 1011
T2 8.67 1017 T3
1.7 + 2.17 104 T + 1.13 105 T2
487.343 + 0.149 T 2.94 105 T2
5700
T is temperature in Kelvin (K).
a short review of the capabilities and shortcoming of few

analytical and numerical models for effective CTE and
effective thermal conductivity (ETC) of composites. Analytical expressions for ETC for two-phase composites made of
randomly distributed and dilute concentrations of spheres
in a homogeneous medium are given, for example, by Maxwell (1954), Verma et al. (1991) and Hasselman and Johnson (1987). Expressions for non-dilute concentrations of
spheres are provided by the models of Jeffrey (1973), Davis
(1986) and Sangani and Yao (1988). These models lead to
accurate ETC predictions, when compared to experimental
results, when the volume fraction (VF) is relatively small or
when the conductivities of the constituents are comparable (Matt and Cruz, 2002). Bounds on the effective thermal
conductivity have been derived using variational principle
(Hashin and Shtrikman, 1962; Beran, 1965) and three point
probability function for the distribution of microstructures.
Several numerical modeling approaches have been
developed to estimate effective thermal and mechanical
behavior of a composite. To obtain effective heat conduction response and ETC of a composite the composites are
often described with random microstructures (Ostoja-Starzewski and Schulte, 1996; Nogales and Bohm, 2008) as
well as with periodic microstructures (Auriault, 1983; Verma et al., 1991; Jiang et al., 2002). Real composite microstructures are generally non periodic; however, the use
of periodic microstructures can give approximate effective
properties for a smaller computational cost when com-
pared to that of random microstructures. To obtain the

effective thermo-mechanical response, among the numerical methods, the Finite Element (FE) technique has been
used for determining the micro- and macro-structural performance of composites by meshing their detailed microstructures; see for example the works of Llorca and
Segurado (2002, 2003 and 2006), Lvesque et al. (2004,
2008) and Zohdi and Wriggers (2001). The composite
behavior was obtained from simulations of a 3D cubic Representative Volume Element (RVE) containing several randomly distributed non-overlapping identical spheres.
Periodic boundary conditions were imposed on the meshes
in order to obtain the effective properties. Linearly elastic,
elasticplastic and linearly (Lvesque and Barello, 2009)
and nonlinearly viscoelastic (Lvesque et al. 2004) responses were analyzed. Kwon and Kim (1998) developed
a 3D micromechanical model to compute thermal stresses
of particulate composites. Using a similar type of micromechanical model, Meijer et al. (2000) studied the inuence of
inclusion geometry and thermal residual stresses- and
strains on the mechanical behavior of an aluminum alloy
reinforced with aluminum oxide particles. A unit-cell consisting of eight subcells was considered. One cell represents the particle and the rest of the cells surrounding
the particle were considered as matrix. Both constituents
were assumed to be linearly elastic and the thermal properties were considered to be temperature independent. The
effects of the size of the subcells on the stresses and strains
611
700
700
Detail Microstructural Model (Volume fraction = 20%, 20 particles)

Model-1 (CPU Time = 12785s) Model-4 (CPU Time = 12519s)
Micromechanical Model(CPU Time = 277s)
Steady State Time = 150 seconds
t =26s
500
400
t =2s
t =12s
Micromechanical Model
600
Temperature (K)
Temperature (K)
600
300
500
t =26s
400
t =12s
300
0
10
Distance (mm)
700
Temperature (K)
500
t =26s
400
t =12s
600
500
t =26s
400
10
700
t =12s
t =2s
t =2s
300
Model-1
Model-3
Model-2
Model-4
600
Distance (mm)
Temperature (K)
t =2s
10
Distance (mm)
300
10
Distance (mm)
Fig. 4. Comparison of temperature proles for FE models with the unit cell (micromechanical model) at each integration point (solid line) and the FE
models with 3D microstructural detail (symbols) for volume fraction of 20%. (a) and (b) are actual values of temperature at top (corner) edge {(X1, 10, 10);
0 6 X1 6 10}, (c) and (d), mean value of temperatures of different FE models measured at extreme top and bottom (corner) edges of the cubes along the
temperature gradient direction..
were analyzed. It was found that the sharp corner and

edges of the cube resulted in localized stresses and strains.
Recently, Aboudi (2008) presented a micromodel for fully
coupled thermo-mechanical analysis for multiphase composites where heat generation from the interconvertibility
of the mechanical and thermal energy was accounted for.
This study focuses on understanding the effects of constituents properties and microstructural details on the
variation of stress, displacement and temperature in composites. The results are used to justify the capability of
micromechanical models to analyze the overall response
of composites subjected to simultaneous mechanical and
thermal stimuli, within a certain degree of accuracy. A
simplied micromechanical model for predicting the effective thermo-mechanical behavior of particulate composites
subjected to both time and space varying temperature eld
is introduced. The micromechanical model is called at each
integration point in a FE mesh to calculate the local effective
properties of the composite. The responses thus obtained
are then compared to FE simulations of composites using
the meshes of Barello and Lvesque (2008). A sequentially
thermo-mechanical coupled problem, i.e., the temperature
eld inuences the deformation eld, is considered. The
effects of particle volume contents and temperature dependent constituent properties on the overall thermo-elastic
behavior of the composites are examined.
The paper is organized as follows. A brief outline of the
simplied micromechanical model is presented in Section 2.
The effective thermoelastic stress-strain relations, effective
heat ux equation and uncoupled energy equation for an
isotropic homogenized composite are also discussed.
Section 3 presents a micro-mechanical framework used
for computing the response of a real-size composite. FE
modeling of composites with microstructural details is
given in Section 4. Coupled heat conduction and thermal
stress analysis of particulate composites are discussed in
Section 5. A summary of the research ndings is given in
Section 6.
2. A simplied micromechanical model for particle

reinforced composites
Muliana and Kim (2007), Muliana (2008) and Khan and
Muliana (2010) developed a micromechanical model for
determining the effective viscoelastic responses and
612
are arranged uniformly in a homogeneous matrix. The RVE

is dened as a single particle embedded in a cubic matrix.
Periodic boundary conditions are imposed to the RVE. Due
to the three-plane symmetry, a unit-cell model that consists
of four particle and matrix subcells is considered. Fig. 1
thermal properties (CTE, and thermal conductivity) of a particle reinforced polymer composite. The model is modied
for simulating sequentially coupled heat conduction and
deformation in particulate composites. The model idealizes
particles in the microstructure as cubes. The cubic particles
700
700

Model-1
Model-2
Model-3
Model-4
t =26s
500
t =12s
400
600
Temperature (K)
Temperature (K)
600
t =26s
500
t =12s
400
t = 2s
t =2s
300
300
0
10
Distance (mm)
10
Distance (mm)
700
700

Model-1
Model-2
600
Temperature (K)
Temperature (K)
600
Model-3
Model-4
t =26s
500
t =12s
400
t =26s
500
t =12s
400
t =2s
t =2s
300
300
0
10
Distance (mm)
10
Distance (mm)
700
700

Model-1
Model-2
600
Temperature (K)
Temperature (K)
600
Detail Microstructural Model(Volume fraction = 30%, 45 particles)

Model-3
Model-4
t =26s
500
t =12s
400
t =26s
500
t =12s
400
t =2s
t =2s
300
300
0
Distance (mm)
10
10
Distance (mm)
Fig. 5. Comparison of temperature proles for FE models with the unit cell (micromechanical model) at each integration point (solid line) and the FE
models with 3D microstructural detail (symbols) for volume fraction of 30%. (a), (b) and (c) are actual values of temperature at top (corner) edge{(X1, 10,
10); 0 6 X1 6 10}, (d), (e) and (f), mean value of temperatures of different FE models measured at extreme top and bottom(corner) edges of the cubes along
the temperature gradient direction.
613
r ij C ijkl ekl a kl T T 0
constituents in the composite body. The effective heat

capacity is obtained using a volume average method.
The linearized thermo-elastic constitutive equations are
expressed in an incremental form. A macroscopic strain
and temperature gradient are known and used as input
variables to the micromechanical formulation. The current
microscopic strain and temperature gradient are expressed
Dt
Dt
as etij et
detij and uti ut
duti , respectively. For
ij
i
simplicity, the superscript t indicating current time, will
be dropped. The macroscopic incremental strain (dekl ) is
linked to the average incremental strain of each sub-cell
a
a
(deij ) using the strain concentration tensor (Bijkl ), written
as
a
dui
M ij duj
700
Detail Microstructural Model (Volume fraction = 20%)
600
20 particles
40 particles
500
t =26s
t =12s
t =2s
where qcxk is the effective heat capacity that depends on

the composition, density and specic heat of the two
300
0
10
Distance (mm)
700
15 particles
30 particles
45 particles
600
Temperature (K)
@ T
@xj
i and u
j are the components of the average heat
where q
ux and temperature gradient vectors, respectively. In order to obtain the temperature proles during heat conduction in the composite, the energy equation needs to be
solved. For the thermo-elastic case, in the absence of internal heat generation and thermo-mechanical coupling effect, the energy equation can be written as:
qcxk T_ qi;i i; k 1; 2; 3
t =8s
t =18s
ijkl and
where C
Sijkl are the components of the effective elas kl
tic stiffness and compliance tensors, respectively. The a
are the components of the effective CTE tensor. The parameters T and T 0 are the effective current and reference temperatures, respectively.
The average heat ux equation for a homogeneous composite medium is expressed by Fouriers law of heat conduction as:
j
where u
where superscript (a) denotes the subcell number, i.e.,

a
a = 1, 2, 3, 4. The strain concentration tensor (Bijkl
) is
400
eij Sijkl r
ij T T 0
kl a
Similarly, the average temperature gradient in each

a
subcell dui ) is related through to overall temperature
j ) by the concentration tensor (M ija ), written
gradient (du
as
or
ij u
i K
j;
q
deij Bijkl dekl
Temperature (K)
shows the RVE idealization, unit-cell and its integration

with the FE framework. The choice of the four-cell model
was primarily done to reduce computational costs; however, this turned out to cause extra effort in incorporating
material symmetry. It is noted that the chosen unit-cell
model, which is only one-eighth of the full model, was
due to the three planes of symmetry that allows interchanging the principal axes in formulating the stress, strain, temperature gradient and heat ux quantities. As a result, the
unit-cell response is independent of the loading direction.
When both particle and matrix are isotropic, the outcome
of the homogenized micromechanical model does not
always fulll the isotropic condition with regard to the
mechanical response. The micromechanical formulation
has been discussed in detail elsewhere (Muliana and Kim,
2007). Perfect bonds are assumed along the subcells interfaces. Thermo-elastic constitutive models with temperature dependent material parameters are used for both
isotropic constituents. Linearized micromechanical relations are formulated in terms of incremental average eld
variables, i.e., stress, strain, heat ux and temperature
gradient, of the subcells. The effective CTE is derived by satisfying total displacement compatibility and traction continuity at the interfaces during thermo-elastic deformations.
This formulation leads to an effective temperature-dependent CTE. The effective thermal conductivity is formulated
by imposing heat ux and temperature continuities at the
subcells interfaces. This micromechanical model is
compatible with general displacement based FE software,
which can be used to perform thermo-mechanical analyses
of composites structures.
For linearly thermo-elastic problems, the effective
ij ) and strain (eij ) are related through:
stress (r
t =26s
500
t =12s
t =18s
t =8s
400
t =2s
300
10
Distance (mm)
Fig. 6. Mean temperature proles for FE models with the unit cell
(micromechanicalmodel) at each integration point (solid line) and the FE
models with 3D microstructural detail (symbols) for volume fraction of
(a) 20% and (b) 30%.
614
determined by imposing the constitutive relation of each

subcell and micromechanical relations such that the displacement compatibility and traction continuity conditions are satised. Two sets of equations are formed. The
rst set of the equations are determined from the strain
compatibility equations which are given as:
8 1 9
e >
>
>
>
>
< 2 >
=
fRe g AM
1
121
1224
>
e3 >
>
>
>
: 4 >
;

DM
1 feg
126 61
where {Rr} is the stress residual vector. The matrix O is the

M
M
zero matrix and the components of matrix AM
1 ; A2 and D1
can be found elsewhere (Muliana and Kim, 2007). For linearized elasticity problems, the components of the residual
vectors are zero and thus the micromechanical relations
are exactly satised. When any of the constituents exhibit
nonlinear response, imposing linearized micromechanical
relations leads to non-zero residual vectors {Re} and {Rr}.
The Newton Raphson iterative method is used to minimize
these residual vectors. Upon minimizing the residual, the
strain concentration matrix B(a) is obtained from:
"
241
where {Re} is the strain residual vector. The second set of

the equations satises the traction continuity relations
within subcells:
8 1 9
e >
>
>
> 2 >
>
<
=
fRr g AM
2
121
1224
>
e3 >
>
>
>
: 4 >
;
a;t
246
AM
1
#1 "
AM
2
2424
DM
1
#
9
O
246
a
O feg
126 61
To formulate the concentration tensor, M ij ; the micromechanical relations and the constitutive equations for
heat ux are imposed. This requires forming twelve (12)
equations based on the temperature and heat ux continuities at the interface of each subcell as:
241
0.015
Model-1
Model-2
Model-3
Model-4
Model-5
Model-6
t =26s
Displacement (mm)
0.01
0.005
0.015
0.01
Displacement (mm)
t =12s
0
t =26s
0.005
t =12s
0
t =2s
t =2s
-0.005
-0.005
10
Distance (mm)

Model-1
Model-2
Model-3
Model-4
0.01
Displacement (mm)
10
d 0.015
0.015

0.01
Displacement (mm)
Distance (mm)
t =26s
0.005
t =12s
t =26s
0.005
t =12s
t =2s
t =2s
-0.005
Distance (mm)
10
-0.005
10
Distance (mm)
Fig. 7. Comparison of axial displacements for FE models with the unit cell (micromechanical model) at each integration point (solid line) and the FE models
with 3D microstructural detail (symbols) for volume fraction of 20%.(a) and (b) are actual values of displacements at top (corner) edge {(X1, 10, 10);
0 6 X1 6 10}, (c) and (d), mean value of displacements of different FE models measured at extreme top and bottom (corner) edges of the cubes along the
temperature gradient direction.
615
8 1 9
du >
>
>
>
>
< du2 >
=
T
g
fRu g A1
DT1 fdu
3
> du >
> 93
31
912 >
91
>
: 4 >
;
du
8 1 9
du >
>
>
>
>
< du2 >
=
T
g
O fdu
fRq g A2
3
> du >
> 33 31
312 >
31
>
: 4 >
;
du
10
121
0.02
Model-1
Model-2
Model-3
Model-4
0.015
Displacement (mm)
121
0.01
0.02
0.015
Displacement (mm)
11
t =26s
0.005
t =12s
0
0.01
t =26s
0.005
t =12s
0
t =2s
t =2s
-0.005
-0.005
0
10

Model-1
Model-2
Model-3
Model-4
0.02
t =26s
0.005
t =12s
0
0.01
t =26s
0.005
t =12s
t =2s
t =2s
-0.005
0.02
Model-1
Model-2
Model-3
Model-4
t =12s
0.02
0.01
t =26s
0.005
t =12s
0
t =2s
t =2s
-0.005
10
Distance (mm)
t =26s
0.005
0.015
0.01
-0.005
10
Distance (mm)
0.015
Displacement (mm)
Displacement (mm)
10
Distance (mm)
0.015
0.01
0.02
0.015
Displacement (mm)
Distance (mm)
Displacement (mm)
Distance (mm)
10
-0.005
10
Distance (mm)
Fig. 8. Comparison of axial displacements for FE models with the unit cell (micromechanical model) at each integration point (solid line) and the FE models
with 3D microstructural detail (symbols) for volume fraction of 30%. (a), (b) and (c) are actual values of displacements at top (corner) edge {(X1, 10, 10);
0 6 X1 6 10}, (d), (e) and (f), mean value of displacements of different FE models measured at extreme top and bottom (corner) edges of the cubes along the
temperature gradient direction.
where {R/} and {Rq} are the temperature gradient and heat
ux residual vectors, respectively. By substituting Eq. (6)
into Eqs. (10) and (11), the M(a) matrix can be determined
as:
M
121
AT1
#1 "
AT2
1212
DT1
12
121
where AT1 , AT2 and DT1 can be found elsewhere (Khan and
Muliana, 2009). The effective (homogenized) stress and
effective tangent stiffness matrix can be obtained from
the following equations:
r ij
ijkl
C
4
1X
a a
a
ijkl ekl a

kl DT
C
V a C ijkl Bklrs ers akl DT
V a1
4
1X
a a
V a C ijkl Bklrs
V a1
13
4. Three dimensional FE models of particulate
composites
14
(a)
where V and V are the total volume of the unit-cell model and sub-cell volume, respectively. From Eq. (13), the
ij ) can be obtained and written as:
effective CTE, (a
a ij
4
1 X
C
ijkl
a
V a C klmn amn
The multi-scale model of Section 3 can be considered as

an approximation to a complicated thermo-mechanical
15
a1
Using the heat conduction equation for each sub-cell

and the effective heat ux relation (Eq. (3)), the tangent
effective thermal conductivity matrix of the composite
can be expressed as:
4
X
a a
ik 1
K
V a K ij Mjk
V a1
4.1. General methodology for evaluating the performance of

the multi-scale framework
16
3. A multi-scale model for computing the response of a

homogenized composite
The micro-mechanical model of Section 2 has been integrated into a multi-scale FE framework in order to compute the eld variables of a real-size composite.
Boundary conditions are imposed on the real-size composite model and initial values of the eld variables are assumed. Local effective properties (thermal, mechanical)
are computed at each integration point using the micromechanical model. Computations of the effective properties are based on the assumption that the each integration
point is associated with a much smaller volume than that
of the whole composite. As a result, we assumed that each
volume associated with each integration point was at a
uniform temperature. Therefore, at the micro-scale, both
the matrix and reinforcement were assumed to be at the
same temperature. This allows determining temperaturedependent thermal conductivities for the particle and matrix constituents and calculating the effective thermal conductivities of the composite at that instant of time in each
material point. Therefore, in this context, the periodic
boundary conditions are fully justied for obtaining the
effective properties. At macroscopic scale during the transient heat conduction, the temperature-dependent properties of the constituents lead to spatially dependent ETC.
0.015
Displacement (mm)
"
a
To simulate the effective thermo-elastic response of the

particulate composite, the micromechanical model is integrated with the ABAQUS/standard FE code. At each integration point in the FE mesh, the user subroutine UMATH is
rst called to evaluate the effective thermal conductivity,
heat uxes, and temperature gradient for solving the equation that governs the conduction of heat in the composite
body. The temperature distributions obtained from heat
transfer analyses at various instants of time in the composites are used as input transient thermal load to determine
the thermo-elastic deformation in the composite body.
UMAT and UEXPAN subroutines were used to evaluate
the effective mechanical response and CTE, respectively.
Fig. 1 illustrates the whole multi-scale framework.
20 particles
40 particles
0.01
t =26s
0.005
t =12s
0
t =2s
-0.005
10
Distance (mm)
0.015
Displacement (mm)
616
15 particles
30 particles
45 particles
0.01
t =26s
0.005
t =12s
0
t =2s
-0.005
10
Distance (mm)
Fig. 9. Mean displacement proles for FE models with the unit cell
(micromechanical model) at each integration point (solid line) and the FE
models with 3D microstructural detail (symbols) for volume fraction of
(a) 20% and (b) 30%.
617
volume of the same material. The size of the RVE is measured in terms of inhomogeneities it contains (e.g. the
number of particles meshed, like 5, 10, 15, etc., particles).
One of the techniques used for obtaining the RVE size is
to use numerical homogenization based on FE. The method
consists in generating many FE meshes of the composite
microstructure with a xed number of reinforcements
(10, 15, 50, etc.). Since the particles are distributed according to a statistical distribution, each mesh, or realization,
will be different. Therefore, each realization should lead
problem: the problem of computing the temperature and

stresses inside a heterogeneous material subjected to both
thermal and mechanical loads. In order to evaluate the
reliability of the micro-mechanical model, it is therefore
required to compare its predictions against the numerically exact solution. The objective of this section is to
generate this so called numerically exact solution.
When dealing with effective properties, one of the key
issues is the denition of the RVE. A RVE can be seen as a
volume of material having the same behavior as any larger
400
Model-1
Model-4
Model-5
Model-2
Model-6
Model-3
200
Axial Stresses (11) MPa
t =2s
-200
-400
400
Model-1
Model-2
Model-3
Model-4
200
0
t =2s
-200
-400
10
400
Model-1
Model-4
Model-5
Model-2
Model-6
Model-3
200
t =12s
-200
-400
200
t =12s
-200
t =26s
-200
-400
Distance (mm)
10
10
400
Model-1
Model-2
Model-3
Model-4
200
t =26s
-200
-400
Distance (mm)
f
Model-1
Model-4
Model-5
Model-2
Model-6
Model-3
10
10
400
200

Model-1
Model-2
Model-3
Model-4
Distance (mm)
400
-400
Distance (mm)
Distance (mm)
10
Distance (mm)
Fig. 10. Axial thermal stresses for FE models with the unit cell (micromechanical model) at each integration point (solid line) and the FE models with 3D
microstructural detail (symbols) for volume fraction of 20% at different times.
618
ments. The relative precision (for example the Youngs

modulus is estimated to be x y%) can be adjusted by varying the number of realizations. If many realizations are
performed, the condence interval can be adjusted to the
desired width. Kanit et al. (2003) mentioned that for
microstructures containing numerous reinforcements,
smaller numbers of realizations are required to estimate
to different (within certain accuracy) effective properties.

For the same number of reinforcements and load history,
the effective responses are computed for each realization
and then averaged. Computing a condence interval (for
example a two-tail 95% condence interval) on this data
could give an estimation of the composites effective properties and its precision for a given number of reinforce-
400
300
400
300

200
100
0
t =2s
-100
-200
-300
200
100
0
-100
t =2s
-200
-300
-400
0
-400
10
Distance (mm)
400
300
Axial Stresses (11 ) MPa
200
100
0
t =12s
-100
-200
100
0
-100
t =12s
-200
-300
-400
10
10
400
300

200
100
0
t =26s
-100
Distance (mm)
400
300
10
200
Distance (mm)
-200
200
100
0
-100
t =26s
-200
-300
-300
-400

300
-300
400
-400
Distance (mm)
Distance (mm)
10
-400
10
Distance (mm)
microstructural detail (symbols) for volume fraction of 20% at different times with C.I. of 95%. (a), (b) and (c) are mean value of stresses of different FE
models with 20 particles and (d), (e) and (f) with 40 particles, measured at extreme top and bottom (corner) edges of the cubes along the temperature
gradient direction.
619
In this work, emphasis was put on eld variables distributions rather than on effective properties. For illustration
purposes, consider a macroscopic component made of a
nonlinear composite constituted of many reinforcements,
as shown in Fig. 2(a). Consider that the component is divided into many sub-regions of length l (Fig. 2(b)). For each
the wanted overall property within desired precision. For

relatively small numbers of particles, the homogenized
properties vary statistically until a certain number of particles are meshed. The number of particles after which
the effective response does not change anymore is called
the representative volume element.
b
Detail Microstructural Model (Volume fraction = 30%,, 15 particles)
Model-1
Model-2
Model-3
Model-4
200
a 400
t =2s
-200
-400
400
200
-200
-400
10

Model-1
Model-2
Model-3
Model-4
t =12s
Distance (mm)
d
Model-1
Model-2
Model-3
Model-4
c 400
200
t =2s
-200
-400
10
400
Model-1
Model-2
Model-3
Model-4
200
t =12s
-200
10
Distance (mm)
f
Axial Stresses ( 11) MPa

Model-1
Model-2
Model-3
Model-4
200
t =2s
-200
Distance (mm)
10
Distance (mm)
e 400
-400
0
-400
Distance (mm)
10
400
Model-1
Model-2
Model-3
Model-4
200
t =12s
-200
-400
10
Distance (mm)
microstructural detail (symbols) for volume fraction of 30% at different times.
620
sub-region, assume that the RVE size has been reached so

that effective properties could be calculated (Fig. 2(c)).
Therefore, the composite component can be divided into a
number of homogeneous sub-regions, as was done in Section 3. If the component is subjected to an external loading
(thermal, mechanical, etc.), the effective properties of each
sub-region can be iteratively computed in order to estimate
the eld variable distribution into the homogenized component. It is recalled that each effective property is known
within a given accuracy and different realization of the
same composite component should lead to different eld
variables distributions. As l decreases and the number of
sub-regions increases, the difference in the eld variables
distributions between each realization should decrease until an acceptable scatter is reached. Since each sub-region is
constituted of a nite number of reinforcements, there is a
nite number of reinforcements inside the composite component for which the homogenized response from one realization to the other stays within a given tolerance. In this
work, the meshed composites are considered as the macroscopic components mentioned above. The simulations
performed in Section 5 aim at determining the number of
reinforcements required for obtaining eld variables distributions corresponding to the homogenized component, as
well as the distribution themselves.
4.2. Generation of the FE meshes used in this study
The FE meshes used in this study are those of Barello
and Lvesque (2008). Their generation is recalled here.
The detailed composite consists of randomly distributed
identical spherical particles reinforced matrix. The microstructures were generated using the Random Sequential
Adsorption Algorithm (Segurado and LLorca, 2002). The
algorithm consists in generating the position of a rst
spherical particle center into a cubic volume using a uniform random number generator. Then, the center position
of a second sphere is generated. If the distance between the
centers of the rst two particles and the distance from the
particle center from the cubes faces is smaller than a preset value, then the second particle is rejected and a new
Tx1 ; x2 ; x3 ; 0 300 K;
edges. The realizations thus obtained were therefore periodic and always had an integer number of complete
spheres. The minimum distance between two particles
centers was set to 2.07r, where r is the particle radius while
the minimum distance from a particle center to a cubes
face was set to 0.1r. These distance criteria were obtained
through trial and error with the meshing software until
elements of acceptable aspect ratios were obtained.
A Matlab program was used for generating the particle
centers. This program wrote an ANSYS command le for
generating the FE mesh of the microstructure. Finally, a
Matlab program was used for converting the ANSYS
model to ABAQUS. The mesh consisted of 10-noded
tetrahedra.
5. Simulation execution and results
5.1. Simulations performed in this study
For both the multi-scale framework and the detailed
models of Section 4, cubic models of dimensions
10 10 10 mm were used. Fig. 3 shows these models as
well as the axes used for dening the boundary conditions
below. The studied composite is a ZrO2 matrix reinforced by
randomly distributed Ti-6Al-4V spherical particles. The
heterogeneous composites directly incorporate nonlinear
thermo-elastic behaviors for the particle and matrix regions. The thermal as well as the mechanical properties
used in the simulations can be found in Khan and Muliana
(2010) and are given in Table 1. Two volume fractions of
reinforcements were studied, namely 20% and 30%. For
the detailed FE meshes, cubes containing 15, 20, 30, 40
and 45 spheres were generated. The transient thermal analysis consisted in a problem where a composite was initially
at 300 K, except for one face that was at 600 K. This transient heat transfer problem was solved until a steady state
was reached. A uniform stress of 10 MPa was applied on the
face that was at 600 K in order to simulate effective transient thermal stresses. The models were subjected to the
following mixed uniform boundary conditions:
0 6 x1 6 10; 0 6 x2 6 10; 0 6 x3 6 10
T10; x2 ; x3 ; t 600 K;
0 6 x2 6 10; 0 6 x3 6 10; t P 0
1 ; 10; x3 ; t
@Tx1 ; 0; x3 ; t @ Tx
0:0; 0 6 x1 6 10; 0 6 x3 6 10; t P 0

@x2
@x2

@Tx1 ; x2 ; 0; t @ Tx1 ; x2 ; 10; t
0:0; 0 6 x1 6 10; 0 6 x2 6 10; t P 0

@x3
@x3
center position is generated until the minimum distance

criterion mentioned above is met. The other particles are
sequentially added, following the same process where
the distance criteria are checked with all the existing particles. The particles are added until the desired volume
fraction is reached.
The particles were allowed to cut the edges and the
faces of the cube. When this happened, the particles were
completed periodically on the corresponding faces and
u1 0; x2 ; x3 ; t 0:0;
17
0 6 x2 6 10; 0 6 x3 6 10; t P 0
u2 x1 ; 0; x3 ; t 0:0;
0 6 x1 6 10; 0 6 x3 6 10; t P 0
u3 x1 ; x2 ; 0; t 0:0;
0 6 x1 6 10; 0 6 x2 6 10; t P 0
p1 10; x2 ; x3 ; t 10:0 MPa; 0 6 x2 6 10; 0 6 x3 6 10; t P 0

p2 x1 ; 10; x3 ; t 0:0 MPa;
0 6 x1 6 10; 0 6 x3 6 10; t P 0
p3 x1 ; x2 ; 10; t 0:0 MPa;
0 6 x1 6 10; 0 6 x2 6 10; t P 0
18
621
where ui and pi (i = 1, 2, 3) are the components of the displacements and the surface tractions, respectively. It is recalled that the set of boundary conditions affects the size of
the representative volume element, and hence, that of the
representative component. It is therefore expected that the

same component subjected to different boundary conditions requires a different number of heterogeneities in order to lead to converged eld variables distributions.
400
400
300
300
200
100
0
-100
t =2s
-200
-300
-400
200
100
0
-100
-200
t =12s
-300
-400
10
Distance (mm)
400
200
100
0
t =2s
-200
200
100
0
-100
t =12s
-200
-400
10
Distance (mm)
400
300
10
400
300
200
100
0
-100
Distance (mm)
10
-300
-300
-400
300
-100
400
300
Distance (mm)
t =2s
-200
-300
200
100
0
-100
t =12s
-200
-300
-400
-400
Distance (mm)
10
10
Distance (mm)
microstructural detail (symbols) for volume fraction of 30% at different times with C.I. of 95%. (a), (b) and (c) are mean value of stresses of different FE
models with 15 particles and (d), (e) and (f) with 30 particles, (g), (h) and (i) with 45 particles, measured at extreme top and bottom (corner) edges of the
cubes along the temperature gradient direction.
Fig. 4(a) and (b) show the temperature distributions obtained from the homogenized model as well as from the
heterogeneous composite reinforced by 20% of Ti6Al4V
particles for model sizes of 20 and 40 particles, respectively, for different times. Fig. 4(c) and (d) show the mean
responses of the various realizations, along with 95% condence intervals for models of 20 and 40 particles, respectively. For the 20 particle model, the largest width of the
condence interval is 1.74% of the mean value while it is
of 3.42% for the 40 particles model. The width of the condence interval decreases as the time increases. Fig. 5(a)(f)
show temperature proles from similar type of analyses
but for a composite reinforced by 30% of Ti6Al4V particles and for models containing 15, 30 and 45 particles.
The largest widths of the condence intervals are of
3.25%, 1.88% and 3.7% for the 15, 30 and 45 particle models,
respectively. It is interesting to note that the condence
interval widths did not decrease as the number of particles
increased, as would have been expected. The reasons for
this phenomenon are unclear at this time. Conducting
more realizations and/or simulating larger number of particles might lead to the expected tendency and this phenomenon might be of statistical nature.
Fig. 6(a) shows the mean temperature curves for a composite with 20% of Ti6Al4V particle volume content and
for models containing 20 and 40 particles. Considering
their relatively narrow condence intervals, it can be seen
that the RVE has been reached for these microstructures
since both the 20 and 40 particle models lead to very similar results. Fig. 6(b) shows the mean temperature curves
for the 15, 30 and 45 particle models for composites with
30% particle volume content. It can also be concluded that
the RVE size has been reached and overcome. Figs. 4 and 5
show that the micromechanical model predicts fairly well
the temperature proles for the range of material properties simulated.
400
300
5.2. Temperature distribution
respectively. For the 20 particles model, the largest width

of the condence interval is 96% of the mean value while
it is of 52% for the 40 particles model.
Fig. 8(a)(f) show displacements from similar analyses
but for a composite reinforced by 30% of Ti6Al4V particles
for models containing 15, 30 and 45 particles. The largest
20 particles
40 particles
200
100
0
t =2s
-100
-200
-300
-400
10
10
10
Distance (mm)
400
300
Due to the boundary conditions, the eld variables

distribution on the four cube segments oriented along x1
(see Fig. 3) should be identical for a large number of
spheres. For the detailed models, the eld variables were
extracted at identical x1 coordinates and then averaged.
At each x1 coordinate, 95% condence intervals on the
mean value were computed. Finally, the averaged distributions of the detailed models were compared to the
distributions of the multi-scale model.
In the following subsections, distributions of the eld
variables predicted from the multi-scale framework are
compared to those of the detailed FE meshes of Section 4.
20 particles
40 particles
200
100
0
t =12s
-100
-200
-300
-400
Distance (mm)
400
300
622
20 particles
40 particles
200
100
0
t =26s
-100
-200
5.3. Displacement distribution

-300
Fig. 7(a) and (b) show the displacement distributions

obtained from the homogeneous and heterogeneous models for a composite containing 20% of Ti6Al4V particles
for models having 20 and 40 particles, respectively.
Fig. 7(c) and (d) show the average response of the various
realizations, along with 95% condence intervals on the
mean value for models containing 20 and 40 particles,
-400
Distance (mm)
Fig. 14. Mean stress proles for FE models with the unit cell (micromechanical model) at each integration point (solid line) and the FE models
with 3D microstructural detail (symbols) for volume fraction of (a) 20% at
different times.
623
widths of the condence intervals were of 30%, 65% and

143% for the 15, 30 and 45 particle models, respectively.
Fig. 9(a) shows the superimposed curves for 20 and 40
particles model for a sphere volume fraction of 20%. Considering the width of the condence intervals of Fig. 7, it can be
seen that the average responses are reasonably close and
hence that the RVE has been reached. These observations
allow to conclude that the micro-mechanical model predicts relatively well the macroscopic response of the
composite for this specic microstructure. Fig. 9(b) shows
the superimposed curves for 15, 30 and 45 particles model
for a sphere volume fraction of 30%. For times t = 12 s and
t = 26 s (seconds), the huge widths of the condence intervals (see Fig. 8(d)(f)) do not allow to conclude whether the
size of the RVE has been reached or not within a reasonable
precision and hence render these RVE analyses meaningless. However, for t = 2 s, the condence intervals are relatively narrow and it is possible to conclude from Fig. 9(b)
that for this time, the RVE size has been reached. For
t = 2 s, it seems that the micro-mechanical model predicts
relatively well the homogenized displacement distribution,
although it is less accurate than the microstructure having

20% of reinforcements. Moreover, it seems that performing
simulations with more than 45 reinforcements might lead
to narrower condence intervals for a better assessment
of the RVE size. Finally, it can be observed that the micromechanical model predicts with more accuracy the temperature distribution than the displacement eld, for the
cases studied here.
5.4. Thermal stresses distributions
The contrast in the CTEs values of the constituents and
high temperature gradient are the main cause for the generation of high thermal stresses. Figs. 1013 show the variation of thermal stresses for spheres volume fractions of
20% and 30% at different times for the homogenized and
heterogeneous composites, respectively. For all the gures,
except for t = 2 s over a certain distance, the width of the
condence intervals cannot be used to determine if the
RVE size has been reached with a high degree of condence. For t = 2 s, it seems that the micro-mechanical
TOP ELEMENTS
TOP ELEMENTS
BOTTOM ELEMENTS
BOTTOM ELEMENTS
400
400
Detail Microstructural Model
Detail Microstructural Model

Model-1
Model-2
200
0
t =26s
-200
-400
200
t =26s
-200
-400
0
Distance (mm)
10
10
Distance (mm)
Fig. 15. Axial thermal stresses for FE models with 3D microstructural detail for volume fraction of 20% at t = 26 s. (a) and (b) are actual values of stresses at
top (corner) edge {(X1, 10, 10); 0 6 X1 6 10} for model-2 and model-1, respectively.
624
For example, consider model-2 shown in Fig. 15(a) for

which the high compressive stresses are found in the matrix region that restraints the free expansion of two particle regions. Similar behavior is found for the elements
neighboring the particle region approximately at 2.5 mm
and 8.3 mm, respectively. For the same temperature difference the particle expands more than the matrix but the
surrounding restraints provided by the matrix elements
are the main cause for the generation of such high values
of compressive stresses. The same description is applicable
to other models where such micro-geometrical features
are found; for example, see below Fig. 15(b) of model-1.
model can predict reasonably well the thermal stresses

distribution. However, for all the other cases, the results
suggest that the micromechanical model is not capable of
capturing the thermal stresses with good accuracy. To corroborate the above-mentioned hypothesis, the mean values of axial thermal stress at different times are shown
in Fig. 14(a)(c) for the microstructures having 20% of reinforcements. More realizations, and possibly with models
having more reinforcements, are required for conrming
this hypothesis with more condence.
The localized stresses are found in some models which
are generally due to the specic micro-geometrical features and the high uctuation about the mean stress prole is due to the presence or absence of the particle
along the prole where the stresses are computed. These
high compressive stresses are found in those matrix elements which surround the particle region. In this study
the thermal expansion of the particle is higher than the
surrounding matrix at all temperatures. Therefore, during
transient heat conduction the free expansion of the particle
is constrained by the surrounding matrix elements. The
larger CTE mismatch of particle/matrix elements results
in such high values of compressive stresses in the neighboring elements of particle.
The effective displacement, (

d1 ), is dened as
d1 e11 L,
where e11 is the volume average of the strains in x1 direction and L is the length of the cube. For both the multi-scale
and the detailed models, (
d1 ) was computed at the face of
loading (BDHF in Fig. 3(c)) for composites having a sphere
volume fraction of 20% and 30%, respectively. The
d1 as a
function of time is plotted in Fig. 16(a) and (b). The mean
values of effective displacements (along with 95%
condence intervals) for heterogeneous composite models
0.02
Model-1
Model-4
Model-5
Model-2
Model-6
Model-3
0.015
0.02
Displacements (mm)
Displacements (mm)
5.5. Effective displacement
0.01
600K
Center line
0.005
300K
0.015
0.01
0.005
10
15
20
25
30
10
Time (seconds)
0.02
Model-1
Model-4
Model-2
Model-3
0.015
0.01
0.005
10
15
Time (seconds)
20
25
30
20
25
0.02
Displacements (mm)
Displacements (mm)
15
Time (seconds)
30
0.015
0.01
0.005
10
15
20
25
30
Time (seconds)
Fig. 16. Effective Axial displacements for FE models with the unit cell (micromechanical model) at each integration point (solid line) and the FE models
with 3D microstructural detail (symbols) for volume fraction of (a) 20% and (b) 30%. Mean values of effective displacements for (c) 20% and (d) 30% with C.I
of 95%.
having 20% and 30% reinforcement particles are shown in

Fig. 16(c) and (d). Agreement of these results corroborate
that the present micromechanical formulation is suitable
for the prediction of effective responses of composites
through the incorporation of a nonlinear thermo-elastic
constitutive material model.
6. Summary
The transient responses of the homogenized and heterogeneous composites due to coupled heat conduction and
mechanical loading have been studied. For the temperature response, the RVE size was reached for both models
having 20% and 30% reinforcements. It was found that
the temperature distribution is relatively well predicted
with the multi-scale model. The width of the condence
intervals for the displacements were larger than those for
the temperature but allowed nevertheless to conclude that
the multi-scale framework can also predict with a reasonable accuracy the displacement eld inside the composite.
The RVE size was not reached for the thermal stresses and
it is not possible to conclude that the multi-scale framework is suitable for representing accurately these stresses.
Larger RVEs or many more simulations for the same RVE
sizes would be required in order to narrow the condence
intervals. However, the mean results obtained are encouraging and running more simulations might reveal that the
multi-scale framework is also suitable for evaluating the
thermal stresses. Finally, the multi-scale model reasonably
predicts the effective displacement. Therefore, the main
contribution of this work was the development and the
partial validation of a multi-scale framework that allows
predicting the eld variables of a temperature dependent
thermo-mechanical problem.
Acknowledgements
This research is sponsored by the Air Force Ofce of Scientic Research (AFOSR) under Grant No. FA 9550-10-10002. We also thank the Texas A&M Supercomputing Facility (http://sc.tamu.edu/) for providing computing resources
useful in conducting the research reported in this paper.
References
Aboudi, J., 1991. Mechanics of Composite Materials. A Unied
Micromechanical Approach. Elsevier, New York.
Aboudi, J., 2008. Thermomechanically coupled micromechanical analysis
of multiphase composites. J. Eng. Math. 61 (24), 111132.
Auriault, J.L., 1983. Effective macroscopic description for heat conduction
in periodic composites. Int. J. Heat Mass Transfer 26, 861869.
Barello, R.B., Lvesque, M., 2008. Comparison between relaxation spectra
obtained from homogenization model and FE simulation. Int. J. Solids
Struct. 45, 850867.
Beran, M., 1965. Nuovo Cimento 38, 771.
Budiansky, B., 1965. On the elastic moduli of some heterogeneous
materials. J. Mech. Phys. Solids 13, 223227.
Chen, H.S., Acrivos, A., 1978. The effective elastic moduli of composite
materials
containing
spherical
inclusions
at
non-dilute
concentrations. Int. J. Solids Struct. 14 (5), 349364.
Christensen, R.M., Lo, K.H., 1979. Solutions for effective properties of
composite materials. J. Mech. Phys. Solids 27, 315.
Christensen, R.M., Lo, K.H., 1986. J. Mech. Phys. Solids 34, 639.
Davis, R.H., 1986. The effective thermal conductivity of a composite
material with spherical inclusions. Int. J. Thermophys. 7 (3), 609620.
625
Hashin, Z., 1962. The elastic moduli of heterogeneous materials. In: Trans.
ASME 84 (E) Ser. J. Appl. Mech., vol. 29. pp. 143150.
Hashin, Z., Shtrikman, S., 1962. J. Mech. Phys. Solids 10, 343352.
Hasselman, D.P.H., Johnson, L.F., 1987. Effective thermal conductivity of
composites with interfacial thermal barrier resistance. J. Compos.
Mater. 21, 508515.
Hill, R., 1965. A self consistent mechanics of composite materials. J. Mech.
Phys. Solids 13, 213222.
Jeffrey, D.J., 1973. Conduction through a random suspension of spheres.
Proceedings of the Royal Society of London Series A. Mathematical
and Physical Sciences 335 (1602), 355367.
Jiang, M., Jasiuk, I., Ostoja-Starzewski, M., 2002. Apparent thermal
conductivity of periodic two-dimensional composites. Comput.
Mater. Sci. 25, 329338.
Kanit, T., Forest, S., Galliet, I., Mounoury, V., Jeulin, D., 2003.
Determination of the size of the representative volume element for
random composites: statistical and numerical approach. Int. J. Solids
Struct. 40, 36473679.
Khan, K.A., Muliana, A.H., 2010. Effective thermal properties of
viscoelastic composites having eld-dependent constituent
properties. Acta Mech. 209 (12), 153178.
Khan, K.A., Muliana, A.H., 2009. A multi-scale model for coupled heat
conduction and deformations of viscoelastic functionally graded
materials. Compos Part B: Eng 40 (6), 511521.
Kwon, Y.W., Kim, C., 1998. Micromechanical model for thermal analysis of
particulate and brous composites. J. Therm. Stress. 21, 2139.
Lvesque, M., Barello, R.B., 2009. On the homogenization of nonlinearly
viscoelastic composite materials. In: ASME International Mechanical
Engineering Congress and Exposition, Proceedings, vol. 12. pp. 303304.
Lvesque, M., Derrien, K., Mishnaevski, L., Baptiste, D., Gilchrist, M.D.,
2004. A micromechanical model for nonlinear viscoelastic particle
reinforced polymeric composite materials: undamaged state.
Composites A 35, 905913.
Lvesque, M., Derrien, K., Baptiste, D., Gilchrist, M.D., 2008. On the
development and parameter identication of Schapery-type
constitutive theories. Mech. Time-Depend. Mat. 12 (2), 95127.
Maxwell, J.C., 1954. A Treatise on Elec. and Magnetism, third ed. Dover,
New York.
Matt, C.F., Cruz, M.A.E., 2002. Application of a multiscale nite-element
approach to calculate the effective thermal conductivity of particulate
media. Comput. Appl. Math. 21, 429460.
McLaughlin, R., 1977. A study of the differential scheme for composite
materials. Int. J. Eng. Sci. 15, 237.
Meijer, G., Ellyin, F., Xia, Z., 2000. Aspects of residual thermal stress/strain in
particle reinforced metal matrix composites. Composites, Part B 31, 2937.
Mori, T., Tanaka, K., 1973. Average stress in matrix and average elastic energy
of materials with mistting inclusions. Acta Metall. 231, 571574.
Muliana, A.H., Kim, J.S., 2007. A concurrent micromechanical model for
predicting nonlinear viscoelastic responses of composites reinforced
with solid spherical particles. Int. J. Solids Struct. 44 (21), 68916913.
Muliana, A.H., 2008. Multi-scale framework for the thermo-viscoelastic
analyses of polymer composites. Mech. Res. Commun. 35 (12), 8995.
Mura, T., 1987. Micromechanics of Defects in Solids. Martinus Nijhoff, The
Hague.
Nemat-Nasser, S., Hori, M., 1999. Micromechanics: Overall Properties of
Heterogeneous Materials, second ed. Elsevier, Amsterdam.
Nogales, S., Bohm, H.J., 2008. Modeling of the thermal conductivity and
thermomechanical behavior of diamond reinforced composites. Int. J.
Eng. Sci. 46 (6), 606619.
Norris, A.N., 1985. A differential scheme for the effective moduli of
composites. Mech. Mater. 4, 116.
OstojaStarzewski, M., Schulte, J., 1996. Bounding of effective thermal
conductivities of multiscale materials by essential and natural
boundary conditions. Phys. Rev. B 54 (1), 278285.
Sangani, A.S., Yao, C., 1988. Bulk thermal conductivity of composites with
spherical inclusions. J. Appl. Phys. 63, 1334.
Segurado, J., Gonzalez, C., Llorca, J., 2003. A numerical investigation of the
effect of particle clustering on the mechanical properties of
composites. Acta Mater. 51, 23552369.
Segurado, J., Llorca, 2002. A numerical approximation to the elastic properties
of sphere-reinforced composites. J. Mech. Phys. Solids 50, 21072121.
Segurado, J., Llorca, 2006. Computational micromechanics of composites:
The effect of particle spatial distribution. Mech. Mater. 38, 873883.
Verma, L.S., Shrotriya, A.K., Singh, R., Chaudhary, D.R., 1991. Thermal
conduction in two phase materials with spherical and non spherical
inclusions. J. Phys. D 24, 17291737.
Zohdi, T.I., Wriggers, P., 2001. Aspects of the computational testing of the
mechanical properties of microheterogeneous material samples. Int. J.
Numer. Methods Eng. 50, 25732599.

Mechanics of Materials: Kamran A. Khan, Romina Barello, Anastasia H. Muliana, Martin Lévesque

Transféré par

Informations du document

Description originale:

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Mechanics of Materials: Kamran A. Khan, Romina Barello, Anastasia H. Muliana, Martin Lévesque

Transféré par

Droits d'auteur :

Formats disponibles

Mechanics of Materials 43 (2011) 608625

Contents lists available at ScienceDirect

Coupled heat conduction and thermal stress analyses

Department of Mechanical Engineering, Texas A&M University, USA

stituents, affecting their overall performance. The use of

K.A. Khan et al. / Mechanics of Materials 43 (2011) 608625

Fig. 2. Homogenization of the sub-regions of a composite component.

variations and discontinuities at the micro-scale cannot be

consistent scheme (Christensen and Lo, 1979, 1986), the

K.A. Khan et al. / Mechanics of Materials 43 (2011) 608625

Young modulus (E) (Pa)

1.23 101156.46 106T

2.44 1011334.28 106T + 295.24 103T2 89.79 T3

T is temperature in Kelvin (K).

a short review of the capabilities and shortcoming of few

pared to that of random microstructures. To obtain the

K.A. Khan et al. / Mechanics of Materials 43 (2011) 608625

Detail Microstructural Model (Volume fraction = 20%, 20 particles)

Detail Microstructural Model (Volume fraction = 20%, 20 particles)

Steady State Time = 150 seconds

Detail Microstructural Model (Volume fraction = 20%, 40 particles)

Detail Microstructural Model (Volume fraction = 20%, 40 particles)

were analyzed. It was found that the sharp corner and

2. A simplied micromechanical model for particle

K.A. Khan et al. / Mechanics of Materials 43 (2011) 608625

are arranged uniformly in a homogeneous matrix. The RVE

Detail Microstructural Model (Volume fraction = 30%, 15 particles)

Detail Microstructural Model (Volume fraction = 30%, 15 particles)

Detail Microstructural Model (Volume fraction = 30%, 30 particles)

Detail Microstructural Model (Volume fraction = 30%, 30 particles)

Detail Microstructural Model (Volume fraction = 30%, 45 particles)

Detail Microstructural Model(Volume fraction = 30%, 45 particles)

K.A. Khan et al. / Mechanics of Materials 43 (2011) 608625

constituents in the composite body. The effective heat

where qcxk is the effective heat capacity that depends on

where superscript (a) denotes the subcell number, i.e.,

Similarly, the average temperature gradient in each

deij Bijkl dekl

shows the RVE idealization, unit-cell and its integration

K.A. Khan et al. / Mechanics of Materials 43 (2011) 608625

determined by imposing the constitutive relation of each

where {Rr} is the stress residual vector. The matrix O is the

where {Re} is the strain residual vector. The second set of

Detail Microstructural Model (Volume fraction = 20%, 40 particles)

Detail Microstructural Model (Volume fraction = 20%, 40 particles)

K.A. Khan et al. / Mechanics of Materials 43 (2011) 608625

Detail Microstructural Model (Volume fraction = 30%, 30 particles)

Detail Microstructural Model (Volume fraction = 30%, 45 particles)

K.A. Khan et al. / Mechanics of Materials 43 (2011) 608625

The multi-scale model of Section 3 can be considered as

Using the heat conduction equation for each sub-cell

4.1. General methodology for evaluating the performance of

3. A multi-scale model for computing the response of a

To simulate the effective thermo-elastic response of the

K.A. Khan et al. / Mechanics of Materials 43 (2011) 608625

problem: the problem of computing the temperature and

Axial Stresses (11) MPa

Axial Stresses (11) MPa

Axial Stresses (11) MPa

Axial Stresses (11) MPa

Axial Stresses (11) MPa

Axial Stresses (11) MPa