International Journal of Architecture, Engineering and Construction

Vol.1, No. 1, March 2012, pp. 14-29

Evaluation of Effective Thermal Conductivity of

Fiber-Reinforced Composites
Changyong Cao1 , Aibing Yu2 , Qing-Hua Qin1,
1
2

Research School of Engineering, Australian National University, Canberra, ACT 0200, Australia

School of Materials Science and Engineering, University of New South Wales, Sydney, NSW 2052, Australia

Abstract: In this paper, effective thermal conductivity of fiber-reinforced composites are estimated by the
newly developed hybrid finite element method (FEM). In the hybrid FEM, foundational solutions are employed
to approximate the intra-element displacement field in any given element, while the polynomial shape functions
used in traditional FEM are utilized to interpolate the frame field. The homogenization procedures using the
representative volume element are integrated with the hybrid fundamental solution based finite element method
(HFS-FEM) to estimate the effective thermal conductivity of the composites and to investigate the effect of fiber
volume fraction and fiber arrangement pattern on the effective thermal conductivity. A special element with an
inclusion is constructed by means of related special fundamental solutions. Due to the fact that the proposed
special element exactly satisfy its boundary conditions along the fibre-matrix interface, only element boundary
integrals are involved and significant mesh reduction can be achieved. Mesh regeneration may be avoided as
well when the fiber volume fraction is slightly changed. The accuracy of the numerical results obtained by
the proposed method is verified against with that obtained from commercial software package ABAQUS. The
results indicate that the proposed method is efficient and accurate in analyzing the micromechanical thermal
behavior of fiber-composites and has the potential to be scaled up to macro-scale modeling of practical problems
of interest.
Keywords: Thermal conductivity, fiber-reinforced composites, representative volume element, hybrid finite
element method, fundamental solutions, special element

1 INTRODUCTION

Cheng 1967; Farooqi and Sheikh 2006; Kachanov and

Sevostianov 2005; Landis et al. 2000; Li et al. 2011;
Fiber-reinforced composites are structural materials Tsukrov and Novak 2002). It is expected to save much
that consist of fiber reinforcing phase and matrix phase effort, time and expense if the properties of the new
in which the fiber is embedded at a macroscopic level to reinforced composites could be predicted accurately or
possibly experience a range of mechanical, thermal and designed from micro-structural properties of its conchemical environment during their service life (Chung stituents.
1994). Due to the superiority of their physical properIn literatures, the averaged or homogenized method
ties over the single matrix, such as high thermal and using representative volume element (RVE) is usually
electrical conductivity, high stiffness and strength etc., employed in micro mechanical modeling of composfiber-reinforced composites have been widely used in ites (Miehe 2003; Zohdi and Wriggers 2008). The
engineering applications. The determination of effec- boundary value problems defined on the RVEs can be
tive properties of composite materials is of paramount analyzed by proper numerical methods. In the past
importance in engineering design and application of two decades, considerable attention has been given to
composite materials. The effective thermal conductiv- the determination of the thermal or mechanical propity and other thermo-physical properties of composites erties of composites by using finite element method
have attracted considerable interest in theoretical, nu- (FEM) (Islam and Pramila 1999). The reason for usmerical and experimental researchers during the last ing FEM is that by employing it we can easily simseveral decades (Brennan and Walrath 2009; Chen and ulate the effect of various possible defects to the me*Corresponding author. Email: qinghua.qin@anu.edu.au
14

Cao et al./International Journal of Architecture, Engineering and Construction 1 (2012) 14-29

chanical properties, i.e., the effect of improper bond

between fiber and matrix and the effect of cracks having different orientations. However, the drawback of
this method is that refined meshes near the defects
(cracks, holes or inclusions) are usually required to
achieve the desired accuracy. This is not practical for
analysis of composites whose fiber distribution might
change repeatedly. Unlike FEM, the boundary element
method (BEM) simply requires division on boundaries
of the domain under consideration, reducing the dimensionality of the problem by one. This approach has
been successfully applied to steady-state and transient
heat conduction (Ma et al. 2008), interface performance (Chen and Papathanasiou 2004), and thermoelastic behavior (Henry et al. 2007) of fiber-reinforced
composites. Singular or hyper-singular integrals, however, are unavoidable in BEM. Moreover, the BEM solution process becomes extremely complex for imposing
continuity conditions across the interface between fiber
and matrix when solving multi-material problems like
fiber-reinforced composites (Gao and Davies 2002).
To overcome these difficulties, hybrid Trefftz
FEM(HT-FEM) was developed based on two independent fields: an intra-element field and an auxiliary
frame field (Qin 2000; Qin and Wang 2008). Interelement continuity is enforced by using a modified variational principle, which is used to construct the stiffness equation and to establish relationship of the frame
field and the internal field of the element. This approach involves the element boundary integrals only,
which inherits the advantages of both conventional
FEM and BEM, and has been successfully applied to
various engineering problems (Qin 1994; Qin 1995; Qin
1996; Qin 2003; Qin and Wang 2008). The property
of non-singular element boundary integral appearing in
HT-FEM enables us to construct arbitrary shaped elements conveniently; however, the terms of truncated Tcomplete functions must be carefully selected to achieving the desired results, and it is difficult to generate Tcomplete functions for certain complex or new physical
problems. Further, in the HT-FEM a coordinate transformation is required to keep the system matrix stable,
and the necessary variational functional is somewhat
complex for practical use.
As an alternative to HT-FEM, a novel hybrid finite
formulation based on the fundamental solutions, called
HFS-FEM, was developed for solving two-dimensional
linear heat conduction problems (Wang and Qin 2009;
Wang and Qin 2011a), and isotropic elastic (Wang and
Qin 2011a), functionally graded elastic (Wang and Qin
2012) and piezoelectric problems (Cao et al. 2012). In
the approach, a linear combination of the fundamental solution at different source points located outside
of the elements is used to approximate the unknown
field within the element instead of the truncated Tcomplete functions used in HT-FEM. The proposed
HFS-FEM inherits all the advantages of HT-FEM and
obviates the difficulties that occur in HT-FEM. If only

the general hybrid element is used, however, the presented HFS-FEM has no significant improvement in
efficiency for handling heterogeneous composites.
In the present work, the formulations of HFS-FEM
for the heat conduction problem is presented to model
heterogeneous fiber-reinforced composites and to investigate the applicability of the new method in predicting
the effective thermal conductivity of ideal fiber reinforced composites. Both of general element and special
element for circular fiber inclusions are proposed based
on the relevant fundamental solutions. The special element is based on a special fundamental solution which
analytically satisfies the continuity of temperature and
heat flux on the interface between fiber and matrix
and is constructed to reduce the mesh refine effort in
modeling heterogeneous composites. Then, independent intra-element and frame fields as well as a modified variational functional are constructed to derive
final stiffness equations and determine the unknowns.
The RVE is utilized for estimating the effective thermal property of the composites. Two examples are
presented in order to get insight into the influence of
fiber volume fraction and fiber arrangement pattern to
the effective thermal conductivity. The accuracy of the
numerical results obtained by the proposed method is
verified against by those calculated by commercial software package ABAQUS. The results indicate that the
proposed method is efficient and accurate in analyzing
the thermal behavior of fiber-composites and has the
potential to be scaled up to macro-scale modeling of
large-scale practical problems of considerable interest.
The paper is organized as follows: the governing
equations of the heat conduction problem and the basic
concepts of RVE and homogenization procedures are
first introduced in Section 2. Then, the fundamental
solutions for interpolation of temperature field within
element domain are given out in Section 3. Further,
the detailed solution procedures for the derivation of
the HFS-FEM are presented in Section 4. Finally, two
numerical examples for RVEs with simple and complex
fiber patterns are considered in Section 5, and some
concluding remarks are presented in Section 6.

2 GOVERNING EQUATIONS AND HOMOGENIZATION

In this section, a brief review of the basic equations
and concepts used in micromechanical analysis of the
heat conduction problem is presented to introduce the
notation and provide a common source for reference in
late sections.

2.1 Governing Equations

The thermal equilibrium governing equation for the
temperature field can be expressed as (Temizer and
Wriggers 2010):

15

q = 0

(1)

Cao et al./International Journal of Architecture, Engineering and Construction 1 (2012) 14-29

Figure 1. Periodic fiber-reinforce composites and RVE

with the boundary conditions:

= on ,
h=qn=h

on k

(2)

where = [/x1 , /x2 ] is the divergence operator;

= T T0 is unknown temperature change, where T
is the current absolute temperature and T0 is a reference absolute temperature; g = is the temperature
gradient; h is the normal heat flux; n is the outward
normal vector to the boundary = k ; and
are specified values on the related boundaries,
and h
respectively. The space derivatives are denoted by a
comma, i.e., ,i = /xi , and the subscript i takes
values 1 and 2 in our analysis.
The constitutive law for flux vector q = [q1 , q2 ]T follows the Fouriers law as (Temizer and Wriggers 2010):
q = k g

(3)

where k is the thermal conductivity tensor, which is

symmetry and positive-definite and depends on the local materials in the heterogeneous composites. The
matrix and fiber are herein assumed to be locally
isotropic and homogeneous, so k11 = k22 = km in the
matrix and equal to kf in the fiber area, respectively.

2.2 Representative
(RVE)

Volume

Elements

The microstructure of the composite, such as the

shape, size distribution, spatial distribution, and orientation distribution of the reinforcing inclusions in the
matrix, has significant influence of the overall (effective) properties of the heterogeneous materials (NematNasser and Hori 1999). Although most composites possess inclusions of random distributions, great insight of
the effect of microstructure on the effective properties
can be gained from investigation of composites with periodic structures. For a composite with periodic fibres,
it is usually sufficient to draw conclusions for the whole
structure by considering only a unit cell, so called RVE,
which is shown in Figure 1 (Zohdi and Wriggers 2008).

For simplification, following assumptions for an ideal

fiber-reinforced composites will be applied (Islam and
Pramila 1999): (1) the composites are macroscopically
homogeneous, (2) locally both the matrix and the fiber
are homogeneous and isotropic, (3) the thermal contact
resistance between the fibre and the matrix is negligible, (4) the composite is free of voids, (5) the problem is
two dimensional, and (6) reinforced fibers are arranged
in a square periodic array, i.e., they are uniformly distributed in the matrix. This last assumption implies
that fibers are equal and uniform in shape and size and
are symmetrical about the x1 - and x2 - directions.
The macroscopic overall properties of the heterogeneous materials are primarily extracted from the RVEs
in an averaged or homogenized method. In the present
analysis, a RVE consisting of matrix material and fiber
phase, as shown in Figure 1, is chosen to be statistically representative of the two-phase composite. It is
noted that the characteristic size of the heterogeneities
is supposed to be much smaller than the dimension of
RVE, which in turn is supposed to be small compared
to the characteristic length of the macroscopic structure. As a result, a RVE is regarded just as a point with
a homogenized constitutive law on the macro-level.

2.3 Homogenization for RVE

The macroscopic constitutive formulation must be determined by materials testing on RVE, from which the
heterogeneous conduction problem is solved with specified boundary conditions. The macroscopic flux and
the macroscopic gradient fields are identified as the volume averages of the microscopic counterparts and they
are related to each other by the macroscopic constitutive formulations.
The thermal constitutive law that governs each material or phase in a RVE is given by the standard Fouriers
law as shown in Eq. (3). To evaluate the effective
thermal conductivity of the microscopically heteroge
neous fiber-reinforce composites, the effective flux q

16

Cao et al./International Journal of Architecture, Engineering and Construction 1 (2012) 14-29

are defined as the

and effective temperature gradient g
volume average values of the respective fields in the
RVE (Zohdi and Wriggers 2008).
Z
Z
1
1
= hqi =
qd =
hXd
(4)
q

and random materials (Hazanov and Huet 1994; Huet

1990). Consequently, PR-BC will be employed for the
RVE models in the present research.
From the above homogenization procedures, we can
see that the flux and temperature on the boundary of
the RVE are sufficient to calculate the effective therZ
Z
mal conductivity of the composites. This feature will
1
1
= hgi =
g
d =
nd
(5) enable us to easily employ the special element intro

duced in Section 5 to evaluate the effective parameters
where is the volume of RVE. It can be seen from of RVE.
Eqs. (4) and (5) that the volume average gradient and
flux are related only to the flux on the boundary of
RVE.
For the isotropic case, the thermal tensor k can be 3 FUNDAMENTAL SOLUTIONS
expressed as k = k I, where k is the conductivity coefficient and I is the identity tensor. Thus, according For the proposed hybrid FEM, it is essential to find
to Fouriers law, the effective thermal conductivity can the fundamental solutions of the plane heat conducbe calculated by:
tion problems to interpolate the intra-element approx
imation fields. The Greens function of the twok
kq
k
q
k
k = =
(6) dimensional heat transfer problem in an infinite dokg k
k
gk
main can be defined by:
It should be noticed that the material test for RVE
k2 G(x, y) + (x, y) = 0
(11)
must be conducted under steady-state conditions and
there should be no external heat supply in order to where (x, y) is the Dirac delta function, x = (x , x )
1
2
eliminate any external influents or rate effects (Tem- denotes the field point where response is calculated
izer and Wriggers 2010).
and y = (y , y ) denotes the source point where unit
1

2.4 Boundary Conditions for RVE

Using the RVE model described above, three types
of boundary condition are usually employed to evaluate the overall/effective thermal properties of heterogeneous materials: (1) Uniform flux boundary condition, (2) Linear temperature gradient boundary condition and (3) periodic boundary condition (Temizer and
Wriggers 2010; Zohdi and Wriggers 2008). The three
boundary conditions are listed as follows:

concentrated heat applied. For plane heat conduction

problems of fiber-reinforced composites two kinds of
fundamental solutions will be used: one is a general
fundamental solution for homogeneous materials without any inclusions, and the other is a special fundamental solution satisfying the interfacial conditions between the circular inclusion and the matrix.

3.1 General Fundamental Solution

Consider a unit heat source applied at the source point

z0 = x10 + ix20 in the infinite homogeneous domain
h=Qn
(7) m , the temperature response Gm at any field point z
is given in the form (Chao and Shen 1997):
so that hqi = Q.
1
Gm (z, z0 ) =
Re {ln(z z0 )}
(12)
2. Linear temperature boundary condition (LT2km
BC):
where Re denotes the real part
of the bracketed ex =GX
(8) pression, z = x1 + ix2 and i = 1 is imaginary number (Wang and Qin 2011b).
so that hgi = G.
1. Uniform flux boundary condition (UF-BC):

3. Periodic boundary condition (PR-BC):

+ = G (X+ X )

(9)

3.2 Special Fundamental Solution

and

For the case that a central circular inclusionis embedded in an infinite domain m , if a unit heat source is
applied
at the source point z0 in m , as shown in Figso that hqi = Q, where Q and G are controlled
2,
the
temperature responses Gm and Gf at any
ure
constant vectors.
field point z in matrix or fiber regions can be obtained
Previous investigations have found that the PR- by using the complex potential theory by considering
BC is much more accurate in micromechanical anal- the interface (| z | = R) continuity conditions between
ysis of composite materials for both periodic materials inclusion and matrix as (Chao and Shen 1997):
h+ = h

(10)

17

Cao et al./International Journal of Architecture, Engineering and Construction 1 (2012) 14-29

where ns is the number of source points outside the element domain, which is equal to the number of nodes of
an element in the present research based on the generation approach of the source points; ce is an unknown
coefficient vector (not nodal temperatures); and Ne is
the fundamental solution matrix which can be written
as:

Ne = G1 (x, ys1 ) G2 (x, ys2 ) . . . Gns (x, ysns )

(15)
ce = [c1 c2 . . . cns ]T

Figure 2. Fundamental solutions for plane heat conduction problems in fiber-reinforced composites

(16)

where Gi (x, ysj ) represents the corresponding fundamental solution.

Gm (x, ysj ) x m
Gi (x, ysj ) =
(17)
Gf (x, ysj ) x f

It is noted that since the fundamental solutions already include the presence of interface between the

fillers and matrix, it is not necessary to model the tem

perature and heat flux continuity conditions on the

Gm =
Re ln(z z0 ) +

2km

interface and then the analysis will become simpler.

z m

km kf

One of the advantages in the presented HFS-FEM is

Re ln(z z0 )
(13) to reduce computation effort by using special purpose
km + kf

elements. In addition, due to use of two groups of inde

Gf =

(k
+
k
)

pendent interpolation functions in the HFS-FEM, we

z

m
f
f

can construct arbitrarily shaped elements to be used

Re[ln(z z0 )]
for analysis, as shown in Figure 3.
Similarly, the induced temperature Gm in the matrix
For a particular element as shown in Figure 3, we
shows a proper singular behavior at the source point
can use the nodes of the element to generate related
z0 , while Gf in the fiber is regular because the source
source points for simplicity, so that the singular sources
point z0 is outside the fiber.
locate on the pseudo boundary to achieve certain numerical stability. The source point ysj (j = 1, 2, . . . , ns )
4 FORMULATIONS OF THE HFS-FEM can be practically generated by means of the following
method (Wang and Qin 2010):
The main idea of the proposed hybrid FEM is origiys = x0 + (x0 xc )
(18)
nated from the HT-FEM, which utilizes two independent approximate fields: intra-element field and an independent frame field along element boundary. But
unlike the HT-FEM, the intra-element fields in the
HFS-FEM are constructed based on the fundamental
solutions, rather than a truncated T-complete function
set. In this section, the solution procedures of the hybrid finite element model with the fundamental solutions as interior approximation functions are described
to solving the linear heat transfer problems of composite materials.

4.1 Intra-Element Field

For a particular element e, which occupies sub-domain
e , we assume that the temperature field defined in
the element domain is approximated by a linear combination of foundational solutions at different source
points located outside of the element domain (see Figure 3, Wang and Qin 2010).
e (x) =

ns
X

N (x, ysj )cj = Ne (x)ce

j=1

x e , ysj 6 e

(14)

Figure 3. Intra-element field, frame field in a particular element in HFS-FEM for a particular element

18

Cao et al./International Journal of Architecture, Engineering and Construction 1 (2012) 14-29

where is a dimensionless coefficient and x0 is the

point on the element boundary (the nodal point in this
work) and xc is the geometrical centroid of the element.

4.3 Hybrid Variational Functional

For the boundary value problem defined in Eqs. (1)

and (2), since the stationary conditions of the traThe corresponding outward normal derivative of ue
ditional potential or complementary variational funcon e is:
tional can not guarantee the required inter-element
e
he = k
= Qe ce
(19) continuity condition in the proposed hybrid finite eln
ement model, a modified potential functional is develwhere
oped as follows:
Y XY
Ne
=
(24)
Qe = k
= knTe
(20)
n
m
e me
with
with
Z
Z
Ne Ne T
Y
1
Te =
(21)
d+

h
,i
,i
x1 x2
2

e
qe
me
(25)
Z

h( )d

4.2 Auxiliary Frame Field

in which the governing equation, i.e., Eq. (1), is assumed to be satisfied a priori. The boundary e of a
particular element consists of the following parts:

To enforce conformity on the field variable , for instance, e = f on e f of any two neighboring elements e and f , an auxiliary inter-element frame field
e = e he Ie
(26)
is used and expressed in terms of nodal displacement
where Ie represents the inter-element boundary of the
vector, d, as used in conventional FEM. In this case,
element e.
is confined to the whole element boundary, that is
Applying the divergence theorem, we can eliminate
e (x)de x e
(22) the domain integral and obtain the final form of the
e (x) = N
the HFS-FE model.
which is independently assumed along the element functional for
Z
Z
Z
Y
1
boundary in terms of nodal degree of freedom (DOF)
d
+

hd

h
hd
(27)
=

e represents the conventional finite element

de , where N
2

e
qe
e
e
interpolating functions. For example, a simple interpoFinally, substituting Eqs. (14), (19) and (22) into the
lation of the frame field on the side with three nodes of
functional
Eq. (27) produces:
a particular element as shown in Figure 4 can be given
Y
1
in the form:
= cTe He ce dTe ge + cTe Ge de
(28)
2

= N1 1 + N2 2 + N3 3
(23)
e
R
R
e d and
i (i = 1, 2, 3) stands for shape functions in where He =
QTe Ne d, Ge = e QTe N
where N

e
R

T hd.
terms of natural coordinate defined in Figure 4.
ge = qe N
e
To enforce inter-element continuity on the common
element boundary, the unknown vector ce should be
expressed in terms
Q of nodal DOF de . Minimization of
the functional e with respect to ce and de , respectively, yields:
Q
e

cT = He ce + Ge de = 0
Qe
(29)

=Ge ce ge = 0
dTe
from which the optional relationship between ce and
de , and the stiffness equation can be produced as:
ce = H1
e Ge de

(30)

and
Ke de = ge

(31)

GTe H1
e Ge

Figure 4.
frame field

where Ke =
stands for the symmetric element stiffness matrix. It is worth pointing out that the
Typical quadratic interpolation for the evaluation of the vector in Eq. (31) is the same as that
in the conventional FEM, which is obviously convenient
19

Cao et al./International Journal of Architecture, Engineering and Construction 1 (2012) 14-29

for the implementation of HFS-FEM into the existing

FEM program. The matrices He , Ge and vector ge
can be calculated by the commonly used Gaussian numerical integration.

and the problem on how to generate realistic RVE

micro-level geometry for numerical analysis is not discussed in the present work.

5.1 Example 1: RVE with One Reinforced

Fiber

4.4 Recovery of Rigid-Body Motion

Considering the physical definition of the fundamental
solution, it is necessary to recover the missing rigidbody motion modes from the above results. Following
the method presented in (Qin 2000), the missing rigidbody motion can be recovered by writing the internal
potential field of a particular element e as:

In the first example, a square RVE with only one reinforced fiber, as shown in Figure 5, is investigated by the
proposed method. The thermal conductivities of the
fiber and matrix are respectively assumed as k2 = 20,
and k1 = 1. The mismatch ratio between fiber and matrix m = k2 /k1 = 20 will be maintained in this analysis
unless otherwise specified. For verification and come (x) = Ne (x)ce + c0
(32)
parison purpose, the numerical results from traditional
where the undetermined rigid-body motion parameter FEM (here we employ ABAQUS for the analysis) are
c0 can be calculated using the least square matching of also given in the analysis as well. It should be pointed
e and e at element nodes.
out that a constant temperature gradient G = [1, 0]T
Z n

2
or G = [0, 1]T is applied in the analysis and the PRe (x) e (x) = min
(33) BC illustrated in Section 2 are employed for both the
i=1
HFS-FEM and ABAQUS modeling.
which finally gives:
Z
1 n
c0 =
M ei
(34)
n i=1
in which M ei = (e Ne ce )|node i and n is the number
of element nodes.Once the nodal field is determined by
solving the final stiffness equation, the coefficient vector ce can be evaluated from Eq. (30), and then c0 is
evaluated from Eq. (34). Finally, the temperature field
at any internal point in an element can be determined
by means of Eq. (32).

5 NUMERICAL EXAMPLE AND DISCUSSION

In this section, numerical experiments are conducted to
demonstrate the performance and efficiency of the hybrid algorithm with special element and also to study
the micro-thermal behavior of composites. To evaluate the effective thermal conductivity of the composites
implies a homogenization procedure that requires the
application of several independent loading conditions
on the RVE of the heterogeneous materials (Zohdi and
Wriggers 2008). In the present work, temperature gradient controlled method is employed to conduct the
necessary tests. Resolving the heat flux field in heterogeneous materials through a steady-state heat transfer
analysis allows for the calculation of the average flux
field components q. Since the average temperature gradient g is imposed, the effective thermal conductivity
coefficient k for the equivalent homogenous material
can be directly calculated from Eq. (6).
In the following two examples, effective properties
of two-phase composite materials are investigated by
means of the HFS-FEM. It is assumed that both fiber
and matrix are linear thermal conduction and that they
are perfectly bonded at their interface. However, complex inclusion geometry is not included in the RVEs

Figure 5. Schematic of RVE containing one central

fiber

Convergence Verification
Before we employ the proposed method for heat transfer analysis of composite materials, the convergence
study is first carried out by checking against the concerned variables. Three different meshes: Mesh 1 (128node elements and one 16-node special elements with a
total of 60 nodes), Mesh 2 (328-node elements and one
16-node special elements with a total of 128 nodes),
and Mesh 3 (618-node elements and one 16-node special element with a total of 220 nodes) are employed.
The results calculated by ABAQUS are employed as
a reference benchmark for comparison, which are obtained using a very fine mesh (as shown in Figure 6,
1940 3D8R element with 5941 nodes).
Figure 7 presents the effective thermal conductivities
k versus the number of element meshes or nodes. It
is shown that the results obtained from HFS-FEM are

20

Cao et al./International Journal of Architecture, Engineering and Construction 1 (2012) 14-29

(a) HFS-FEM with special element

(b) ABAQUS

Figure 6. Mesh configurations for RVE by (a) for HFS-FEM and by (b) for ABAQUS
converging to the benchmark values with the increase
of number of DOF. It is also noted that HFS-FEM is
able to give similar accuracy, when using 128 nodes
only, to that from ABAQUS using thousands of nodes.

Effect of Fiber Volume Fraction on the Effective

Thermal Properties
Fiber volume fraction (FVF)has significant influence to
the overall thermal conductivity of the heterogeneous
composites. In this section, the effect of FVF on the
macroscopic effective thermal conductivities k of is investigated. The FVF is changed from 3.14% to 63.62%
by varying the radius of the fiber. The predicted effective thermal conductivities from HFS-FEM are graphically shown in Figure 8. It can be observed, as was
expected, that effective thermal conductivities k rises

along an increase in the FVF. However, the increasing rate becomes larger along with an increase in the
FVF. When the FVF increases to 63.62%, the effective
thermal conductivity k of the composite reaches up to
more than four times of that of the matrix. It is also
observed from Figure 8 that, for all the cases investigated here, better accuracy can be achieved by HFSFEM with special elements than that from ABAQUS.
Compared to the results calculated from ABAQUS,
which are achieved by refined meshes around and
within the fiber, the specially purposed inclusion elements can achieve very good accuracy. This conclusion can also be confirmed by Figures 9 and 10,
which presents the variations of the effective thermal
conductivities k varies with the materials mismatch
ratio m = k2 /k1 or 1/m = k1 /k2 .

Figure 7. RVE with one central fiber: Convergence of the effective thermal conductivity
21

Cao et al./International Journal of Architecture, Engineering and Construction 1 (2012) 14-29

Figure 8. Effective thermal conductivities k of composite for different fiber volume factions

Effect of Material Mismatch Ratio on the Effective the effect will become weaker after the thermal conducThermal Properties
tivity of fiber larger than 20 times of that of the matrix. For comparison, the predicted results obtained
Figure 9 shows the effective thermal conductivity k from ABAQUS are also presented in Figure 9. The
of the heterogeneous composite for different material same conclusion can be obtained from the results and
mismatch ratios when the thermal conductivity k2 of both methods provide equivalent accuracy.
the fiber is larger than the thermal conductivity k1
Figure 10 presents the variation of the effective therof the matrix, i.e., a conducting fiber embedded into mal conductivity k for the different material mismatch
an insulated matrix. The FVFis constant at 19.63%, ratios when the thermal conductivity k2 of the fiber
i.e., R = 2.5. It can be seen from Figure 9 that for is smaller than the thermal conductivity k1 of matrix,
a given matrix the effective thermal conductivity in- i.e., an insulated fiber embedded into a conducting macreases with the increasing conductivity of the fiber trix. The FVF also keeps constant at 19.63%. Comand the heat conduction performance can be dramat- pared with Figure 9, it is obvious that the nearly linear
ically improved when adding the conducting fiber but relationship between effective thermal conductivity k

Figure 9. Effective thermal conductivity k of the composite for different mismatch ratios when k2 > k1
22

Cao et al./International Journal of Architecture, Engineering and Construction 1 (2012) 14-29

Figure 10. Effective thermal conductivity k of the composite for different mismatch ratios when k1 > k2

Figure 11. Contour plots of the temperature and heat flux distribution in the RVE (m = 20, FVF=19.63%)

Figure 12. Contour plots of the temperature and heat flux distribution in the RVE (1/m = 20, FVF=19.63%))

23

Cao et al./International Journal of Architecture, Engineering and Construction 1 (2012) 14-29

Figure 17 that the effective thermal conductivities k

of pattern 1 is slightly better than those of pattern 2,
in other words the conductivity reducing effect of pattern 2 is slight better than that of pattern 1, as the
increasing effect for the case of k2 > k1 .
Figure 18 presents the contour plots of the temperature distribution and heat flux components across the
RVE when m = 20 and the FVF is 15.71%. It is obvious that the fiber inclusions has a significant influence
for the heat flux distribution in the RVE, and that the
heat fluxes are concentrated through the fibers when
there is fiber appearing at a cross section. It is indicated that most of the heat are carried out through the
fibers in the composites when the thermal conductivity
5.2 Example 2: RVE with Multi-Fibers
of the fiber is higher than that of the matrix. OppoTwo different fiber patterns as shown in Figure 13 are sitely, it can be found from Figure 19 that the insulated
investigated to reveal the effect of fiber configuration fibers expel the heat flux to the matrix and most flux go
on effective thermal properties of heterogeneous com- through the matrix, not the fiber. Comparing the two
posites. The geometry and dimensions of the RVEs patterns, it can be seen that this phenomenon is much
are given in Figure 13, in which L = 10 and R = 1. In more significant for pattern 2. Numerical results obthe two cases, the FVF is kept as a constant at 15.71%. tained above also show that the results from HFS-FEM
Mesh configuration of the RVEs for HFS-FEM is shown have a good agreement with those from ABAQUS alin Figure 14, in which the 16-node special element (i.e., though much less meshes are employed by HFS-FEM.
8-edges) for fibers are employed. Mesh configurations It can be concluded that the proposed method is accurate and efficient in analyzing micro heat transfer
of the RVEs for ABAQUS are shown in Figure 15.

The predicted effective thermal conductivities k of problems.

the two-phase composites by the HFS-FEM for the two
different patterns are given in Figures 16 and 17, re- 6 CONCLUSIONS
spectively. For the case of conducted fibres embedded in an insulated matrix, i.e., k2 > k1 , it can be A new hybrid FEM based on the fundamental soseen from Figure 16 that for both patterns the effec- lutions is proposed and successfully applied to heat
tive thermal conductivity k are increasing with the transfer problems of heterogeneous composites. The
increasing of fiber thermal conductivity k2 and the in- two independent intra-element and frame fields facilicreasing rates are dramatic first between 1 and 20, then tate the construction of arbitrary-shaped elements and
tends to be slightly smooth. The relationship between the modified variational functional for stiffness maeffective thermal conductivities k and mismatch ratio trix derivation involves element boundary integral only.
m are nonlinear. When the thermal conductivities k2 Based on the special fundamental solution, a type of
of the fiber is 20 times of that of matrix, the effective special element with inclusion(fiber) involved is prothermal conductivities k of the composites, compared posed for mesh reduction in analyzing heterogeneous
with the pure matrix materials, can increase by 33.3% composites.
for pattern 1 and 35.2% for pattern 2, respectively. It
The effective thermal conductivity of composites is
is obvious from Figure 16 that the effective thermal evaluated through the RVEs with single or multiconductivities k of pattern 2 is slightly better than ple fibers using the homogenization technique. The
those of pattern 1, and the beneficial influence will be- method is used to investigate the effect of the FVF
come clearer when the mismatch ratio becomes larger and the mismatch ratio between fiber and matrix
(m > 20).
in the composites. Numerical results obtained from
For the case of insulated fiber embedded in a conduc- ABAQUS are also presented for comparison and veritor matrix, i.e., k2 < k1 , it can be seen from Figure 17 fication. It is shown that the numerical solutions obthat the effective thermal conductivity k for both pat- tained by HFS-FEM coincided with the results calcuterns are linearly increasing with the matrix thermal lated by ABAQUS with fine element mesh. It is found
conductivity k1 if fixed k2 as a constant, which is com- that the effective thermal conductivity rises with the
pletely different from the former case of nonlinearity. increasing FVF. The employment of special inclusion
For such case, it is found that the thermal property element can significantly reduce model meshing effort
of the matrix has a significant influence on the overall and computing cost, and simultaneously avoid mesh
material prosperity. The matrix conductivity reduces regeneration when the FVF is slightly changed.
It may be concluded that the proposed micromechanby 26.80% for pattern 1 and 28.02% for pattern 2, respectively, when the thermal conductivity k1 of matrix ical models based on HFS-FEM have the potential to
is 100 times of that of fiber. It can be observed from model fiber-reinforced composites and to be further deand the matrix thermal conductivity k1 can be found,
which is different from the nonlinearity shown in Figure 9.
Contour plots of the temperature and heat flux distributions in the RVE under constant temperature gradient and periodic boundary conditions are shown in
Figures 11 and 12, in which the FVF is kept as 19.63%.
It can be seen from Figure 11 that for higher conduction fiber the heat flux are mainly pass through fiber
while for the insulated fiber with lower conductivity
than that of matrix, most flux will carry out by the
matrix (see Figure 12).

24

Cao et al./International Journal of Architecture, Engineering and Construction 1 (2012) 14-29

(a) Pattern 1

(b) Pattern 2

Figure 13. RVEs of the composite with two different fiber configurations

(a) Pattern 1

(b) Pattern 12

Figure 14. Mesh configurations of the RVEs for HFS-FEM

(a) Pattern 1

(b) Pattern 2

Figure 15. Mesh configurations of the RVEs for ABAQUS

25

Cao et al./International Journal of Architecture, Engineering and Construction 1 (2012) 14-29

Figure 16. Effective thermal conductivity k of the composite for different mismatch ratios when k2 > k1

Figure 17. Effective thermal conductivity k of the composite for different mismatch ratios when k1 > k2

26

Cao et al./International Journal of Architecture, Engineering and Construction 1 (2012) 14-29

Figure 18. Contour plots of the temperature and heat flux distribution in the RVE (m = 20, FVF=15.71%)

Figure 19. Contour plots of the temperature and heat flux distribution in the RVE (1/m = 20, FVF=15.71%)

27

Cao et al./International Journal of Architecture, Engineering and Construction 1 (2012) 14-29

veloped for considering the defects such as the cracks

and pore voids in microstructures and then for multiscale simulation in the future work.

REFERENCES
Brennan, K. P., and Walrath, D. E. (2009). A selfcontained module for predicting micro-scale material
properties during fiber-reinforced polymer processing. Journal of Thermoplastic Composite Materials,
22, 727751.
Cao, C., Qin, Q. H., and Yu, A. (2012). Hybrid
fundamental-solution-based FEM for piezoelectric
materials. Computational Mechanics (in press).
Chao, C., and Shen, M. (1997). On bonded circular
inclusions in plane thermoelasticity. Journal of applied mechanics, 64, 10001004.
Chen, C., and Cheng, S. (1967). Mechanical properties
of fiber reinforced composites. Journal of Composite
Materials, 1, 3041.
Chen, X., and Papathanasiou, T. (2004). Interface
stress distributions in transversely loaded continuous
fiber composites: parallel computation in multi-fiber
RVEs using the boundary element method. Composites science and technology, 64, 11011114.
Chung, D. D. L. (1994). Carbon Fiber Composites. Butterworth-Heinemann, Newton, MA, United
States.
Farooqi, J. K., and Sheikh, M. A. (2006). Finite element modelling of thermal transport in ceramic matrix composites. Computational Materials Science,
37, 361373.
Gao, X. W., and Davies, T. G. (2002). Boundary Element Programming in Mechanics. Cambridge University Press, Cambridge, United Kingdom.
Hazanov, S., and Huet, C. (1994). Order relationships
for boundary-conditions effect in heterogeneous bodies smaller than the representative volume. Journal
of the Mechanics and Physics of Solids, 42, 1995
2011.
Henry, D., Ma, F., Chatterjee, J., and Banerjee, P.
(2007). Steady state thermoelastic analysis of 3d
solids with fiber inclusions by boundary element
method. Computer Methods in Applied Mechanics
and Engineering, 197, 294307.
Huet, C. (1990). Application of variational concepts to
size effects in elastic heterogeneous bodies. Journal
of the Mechanics and Physics of Solids, 38, 813841.
Islam, M. R., and Pramila, A. (1999). Thermal conductivity of fiber reinforced composites by the FEM.
Journal of Composite Materials, 33, 16991715.
Kachanov, M., and Sevostianov, I. (2005). On quantitative characterization of microstructures and effective properties. International Journal of Solids and
Structures, 42, 309336.
Landis, C. M., Beyerlein, I. J., and McMeeking, R. M.
(2000). Micromechanical simulation of the failure of
fiber reinforced composites. Journal of the Mechan-

ics and Physics of Solids, 48, 621648.

Li, H., Li, S., and Wang, Y. (2011). Prediction of effective thermal conductivities of woven fabric composites using unit cells at multiple length scales. Journal of Materials Research, 26, 384394.
Ma, F., Chatterjee, J., Henry, D., and Banerjee, P.
(2008). Transient heat conduction analysis of 3d
solids with fiber inclusions using the boundary element method. International Journal for Numerical
Methods in Engineering, 73, 11131136.
Miehe, C. (2003). Computational micro-to-macro
transitions for discretized micro-structures of heterogeneous materials at finite strains based on the minimization of averaged incremental energy. Computer
Methods in Applied Mechanics and Engineering, 192,
559591.
Nemat-Nasser, S., and Hori, H. (1999). Micromechanics: Overall Properties of Heterogeneous Materials.
Elsevier Science, Amsterdam, Netherlands.
Qin, Q. H. (1994). Hybrid trefftz finite-element approach for plate-bending on an elastic-foundation.
Applied Mathematical Modelling, 18, 334339.
Qin, Q. H. (1995). Hybrid-trefftz finite-element
method for reissner plates on an elastic-foundation.
Computer Methods in Applied Mechanics and Engineering, 122, 379392.
Qin, Q. H. (1996). Transient plate bending analysis by
hybrid trefftz element approach. Communications
in Numerical Methods in Engineering, 12, 609616.
Qin, Q. H. (2000). The Trefftz Finite and Boundary
Element Method. WIT Press, Southampton, United
kingdom.
Qin, Q. H. (2003). Solving anti-plane problems of
piezoelectric materials by the trefftz finite element
approach. Computational Mechanics, 31, 461468.
Qin, Q. H., and Wang, H. (2008). Matlab and C Programming for Trefftz Finite Element Methods. CRC
Press, New York, United States.
Temizer, I., and Wriggers, P. (2010). A micromechanically motivated higher-order continuum formulation
of linear thermal conduction. Journal of Applied
Mathematics and Mechanics/ZAMM, 90, 768782.
Tsukrov, I., and Novak, J. (2002). Effective elastic
properties of solids with defects of irregular shapes.
International Journal of Solids and Structures, 39,
15391555.
Wang, H., and Qin, Q. H. (2009). Hybrid fem with
fundamental solutions as trial functions for heat conduction simulation. Acta Mechanica Solida Sinica,
22, 487498.
Wang, H., and Qin, Q. H. (2010). Fundamentalsolution-based finite element model for plane orthotropic elastic bodies. European Journal of Mechanics - A/Solids, 29, 801809.
Wang, H., and Qin, Q. H. (2011a). Fundamentalsolution-based hybrid FEM for plane elasticity with
special elements. Computational Mechanics, 48,
515528.

28

Cao et al./International Journal of Architecture, Engineering and Construction 1 (2012) 14-29

Wang, H., and Qin, Q. H. (2011b). Special fiber elfunctionally graded plates. European Journal of Meements for thermal analysis of fiber-reinforced comchanics - A/Solids, 33, 1233.
posites. Engineering Computations, 28, 10791097. Zohdi, T. I., and Wriggers, P. (2008). An Introduction
Wang, H., and Qin, Q. H. (2012). Boundary inteto Computational Micromechanics. Springer, Heidelgral based graded element for elastic analysis of 2d
berg, Germany.

29