You are on page 1of 9

International Journal of Heat and Mass Transfer 49 (2006) 30753083

www.elsevier.com/locate/ijhmt

A new approach to modelling the eective thermal conductivity


of heterogeneous materials
a,*
Jianfeng Wang , James K. Carson b, Mike F. North b, Donald J. Cleland a

a
Institute of Technology and Engineering, Massey University, Private Bag 11-222, Palmerston North, New Zealand
b
AgResearch Ltd., Private Bag 3123, Hamilton, New Zealand

Received 26 October 2005; received in revised form 3 February 2006


Available online 5 April 2006

Abstract

A unifying equation for ve fundamental eective thermal conductivity structural models (Series, Parallel, two forms of Maxwell
Eucken, Eective Medium Theory) was derived. A procedure for modelling complex materials as composites of these ve basic structural
models using simple combinatory rules based on structure volume fractions was proposed. The combined models have advantages over
other generic models such as the semi-empirical Krischer model, in that each has a distinct physical basis, and that they are not depen-
dent on any empirical parameter. As a by-product, a physical description has been identied for Levys model, which was previously used
with reservation by some researchers because it was derived solely by mathematical reasoning without any explicit physical basis.
2006 Elsevier Ltd. All rights reserved.

Keywords: Eective thermal conductivity; Models; Heterogeneous material; Anisotropic material

1. Introduction empirical parameter may be inserted to account for dier-


ences in structure [2,7]. Another common way of estimat-
Modelling the eective thermal conductivity of hetero- ing eective thermal conductivity for composite materials
geneous or composite materials is of interest in many heat with known microstructures is to make rigorous numerical
transfer applications. Progelhof et al. [1] and Carson et al. simulations using the nite dierence or nite element
[2] provide reviews of relevant modelling approaches. A methods [810]. However, analytical models are preferred
substantial number of eective thermal conductivity mod- over numerical models in many applications due to their
els have been proposed, some of which have been intended physical basis, rapid and low cost of calculation, and rea-
for highly specic applications, while others have wider sonable accuracy even when microstructure is uncertain.
applicability. A heterogeneous materials eective thermal Many (if not most) eective thermal conductivity mod-
conductivity is strongly aected by its composition and els found in the literature are based on one or more of ve
structure, and, as yet, there does not appear to be any sin- basic structural models; specically, the Series, Parallel,
gle model equation that is applicable to all types of struc- MaxwellEucken (two forms) [11,12] and Eective Med-
ture. Instead, a common approach has been to develop a ium Theory (EMT) models [13,14]. The physical structures
set of equations based on a conceptual parent model, that assumed in the derivations of the Series and Parallel mod-
is modied to varying extents to account for variations els are of layers of the components aligned either perpen-
in composition and structure [36]. Alternatively, an dicular or parallel to the heat ow, as their names
indicate. The MaxwellEucken model assumes a dispersion
*
Corresponding author. Tel.: +64 6 356 9099x7284; fax: +64 6 350
of small spheres within a continuous matrix of a dierent
5604. component, with the spheres being far enough apart such
E-mail address: J.F.Wang@massey.ac.nz (J. Wang). that the local distortions to the temperature distributions

0017-9310/$ - see front matter 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijheatmasstransfer.2006.02.007
3076 J. Wang et al. / International Journal of Heat and Mass Transfer 49 (2006) 30753083

Nomenclature

A, B, C, D, F, G intermediate variables dened within Greek symbols


the text e structure volume fraction
d shape factor / structure composition factor
f empirical weighting factor j eective thermal conductivity of a structure
k thermal conductivity of a component (W m1 K1)
(W m1 K1)
K eective thermal conductivity of a material Subscripts
(W m1 K1) i ith component
m number of components j jth structure
n number of structures
v volume fraction of component

Table 1
Five fundamental eective thermal conductivity structural models for two-component materials (assuming the heat ow is in the vertical direction)
Model Structure schematic Eective thermal Reference Eq. (1) parameter
conductivity equation values
Parallel model K = v1k1 + v2k2 di ! 1 or ~k k i

3k 1
k 1 v1 k 2 v2
2k 1 k 2
MaxwellEucken 1 (ME1) (k1 = continuous phase, K [8,9] di = 3 and ~k k 1
3k 1
k2 = dispersed phase) v1 v2
2k 1 k 2

k1  K k2  K
EMT model v1 v2 0 [10,11] di = 3 and ~k K
k 1 2K k 2 2K

3k 2
k 2 v2 k 1 v1
2k 2 k 1
MaxwellEucken 2 (ME2) (k1 = dispersed phase, K [8,9] di = 3 and ~k k 2
3k 2
k2 = continuous phase) v2 v1
2k 2 k 1

1
Series model K di = 1 or ~k ! 0
v1 =k 1 v2 =k 2

around each of the spheres do not interfere with their in particular with regards to dening thermal conductivity
neighbours temperature distributions. For a two-compo- bounds for certain classes of physical structure [1518]. In
nent material, two forms of the MaxwellEucken model this paper we present a procedure for modelling complex
arise depending on which of the components forms the physical structures as composites of these basic elementary
continuous phase. The EMT model assumes a completely structural models using simple combinatory rules.
random distribution of all the components. Table 1 lists
the two-component forms of the equations for each of 2. Model development
these models along with a schematic of its assumed physi-
cal structure. 2.1. A unifying equation for the basic structural models
Several eective thermal conductivity studies have recog-
nised the importance of these basic structural models in the- Following on from work by Brailsford and Major [16]
oretical analyses and for developing more complex models, each of the model equations shown for two components
J. Wang et al. / International Journal of Heat and Mass Transfer 49 (2006) 30753083 3077
0P 1
in Table 1 may be derived for a multi-component material m di~ k
i1 k i vi d i 1~
K f@
kk i A
from Eq. (1) by suitable choice of the parameters di and Pm d i ~k
~k: i1 vi d i 1~kk i
structure-1
0Pm0 1
Pm d i ~k
0 0 d 0i ~k 0
i1 k i vi d i 1~kk i i0 1 k i vi d 0 1~k 0 k 0
K Pm 1 1  f @ Pm0 0 d 0i k~0
i iA
3
d i ~k
i1 vi d 1~kk i0 1 vi d 0 1~k 0 k 0
i i i i structure-2

As indicated in Table 1, the Series model is obtained when However, while such weighted mean combination models
di = 1, or ~k ! 0; the Parallel model is obtained when may be useful in some situations, the fact that the value
di ! 1 or ~k k i ; the MaxwellEucken equation is of the weighting parameters cannot be determined mecha-
obtained when di = 3, ~k k cont ; and the EMT equation is nistically from information about the physical structure is a
obtained when di = 3, ~k K. signicant shortcoming. Therefore, we do not advocate
The di parameter can have a physical interpretation. such an approach unless a more analytical and mechanistic
Kirkpatrick [19] related a similar parameter to the number approach cannot be found.
of Euclidean dimensions of the system involved, while
Fricke [20] and Hamilton and Crosser [21] related it to 2.3. Structure volume fractions and structure composition
the sphericity of the dispersed phase. However, it may be factors
possible to dene a parameter that combines both aspects
of component shape and number of Euclidean dimensions. Krischers approach assumed that a complex structure
This is the topic of ongoing investigations by the authors, could be approximated by a mixture of simpler structures,
but falls outside the scope of this paper. The most common where the relative amounts of each of the simpler structures
approach is to take di = 3 (spherical dispersed phase). was determined empirically. In this work we dene struc-
ture volume fractions (as distinct from component volume
2.2. Empirically weighted mixtures of basic physical fractions) for the jth type of structure by Eq. (4):
structures X
ej vi /ij 4
i
A common approach has been to combine structural
models using empirical weighting. The Series and Parallel where ej represents the volume fraction of a material that is
models dene the upper and lower bounds (sometimes made up of structure j. Since the ej are fractions of the total
referred to as the Wiener bounds [22]) for the eective ther- volume:
mal conductivity of any heterogeneous material for which X
the components volume fractions and thermal conductivi- ej 1 5
j
ties are known accurately, provided conduction is the only
mechanism of heat transfer involved. Krischer [7] reasoned The structure composition factors, /ij, are a measure of
that since the thermal conductivity of any two-component the fraction of component material i that is part of struc-
material must lie between the Wiener bounds, its structure ture j, and therefore since the total amount of component
could be modelled as a mixture of Series and Parallel struc- i must be distributed between the structures:
tures. He proposed that the eective thermal conductivity X
of the combined structure should be the weighted harmonic /ij 1 6
j
mean of the Series and Parallel conductivities:
For m components, the thermal conductivity of structure j
1 is a function of the v0i s; /0ij s and k 0i s:
K 2
f =k series 1  f =k parallel
jj jj v1 ; v2 ; . . . ; vm ; /1j ; /2j ; . . . ; /mj ; k 1 ; k 2 ; . . . ; k m 7
Chaudhary and Bhandari [23] and Renaud et al. [24] used If the structure is modelled by one of the ve fundamental
similar approaches, based on weighted geometric and models listed in Table 1, Eq. (7) may be written:
arithmetic means, respectively. Pm d i ~k
Clearly this reasoning may be extended using narrower i1 k i vi /ij d i 1~kk i
bounds, such as those proposed by Hashin and Shtrikman jj Pm di~
k
8
i1 vi /ij d 1~ kk
[15] which were equivalent to the two forms of the Max- i i

wellEucken model, or Carson et al. [18], which were based For n structures, the overall material eective thermal con-
on the MaxwellEucken and EMT models. Making use ductivity, K, is a function of the j0j s:
of Krischers approach and Eq. (1), a generic weighted
K Kj1 ; j2 ; . . . ; jn 9
model can be dened that would allow a combination of
any of the equations listed in Table 1, and therefore would In order to solve Eq. (9), based only on ki and vi and not on
be suitable for structures that t between any of these any empirical parameters, it is necessary to determine
bounds: expressions for /ij as functions of ki and vi. This may
3078 J. Wang et al. / International Journal of Heat and Mass Transfer 49 (2006) 30753083

be achieved by making assumptions about the relative 3. Selected binary-structure models


amounts of each structure, ej, in the total volume and the
relative contributions of the dierent jj to K. There are The schematics in Fig. 1ad represent four of the 10 the-
an innite number of relationships between the e0j s and jj oretical two-component models with an equal mixture of
that could be chosen. It may be possible to relate the e0j s two of the ve basic structures. Fig. 1a shows half of the
to structural characteristics. However, such detailed infor- volume has the MaxwellEucken structure with component
mation is often unavailable. Therefore the simplest possible 1 as the continuous phase (structure 1 = ME1), while the
relationships were adopted. They are to assume that each other half of the volume has the MaxwellEucken structure
structure comprises an equal fraction of the total volume: with component 2 as the continuous phase (structure
2 = ME2); similarly, in Fig. 1b structure 1 = Parallel and
1
ej 10 structure 2 = ME2; in Fig. 1c structure 1 = ME1 and struc-
n ture 2 = EMT; and in Fig. 1d structure 1 = ME2 and struc-
and that the thermal conductivity of each structure is equal ture 2 = EMT.
to the eective thermal conductivity:
3.1. ME1 + ME2 model
K j1 j2    jn 11
Other more complex approaches may be justied in some Starting with Eq. (8), the appropriate values of di and ~k
circumstances. For example, for a material made up of are chosen such that structure 1 is the ME1 structure and
three structures the distribution of structure volumes does structure 2 is the ME2 (Table 1). Hence from Eqs. (8)
not have to be uniform and could arbitrarily be e1 = 0.2, and (11):
e2 = 0.5, e3 = 0.3 or similar. However, the overall method-
k 1 v1 /11 k 2 v2 /21 2k3k 1
1 k 2
ology is applicable irrespective of what relative values of K j1
e0j s and j0j s are chosen. v1 /11 v2 /21 2k3k 1
1 k 2
The model is now completely dened for two-compo- k 2 v2 /22 k 1 v1 /12 2k3k 2
2 k 1
nent materials. The eective thermal conductivity, K, can j2 12
be calculated by solving Eqs. (4), (8), (10) and (11) simulta- v2 /22 v1 /12 2k3k 2
2 k 1

neously subject to Eqs. (5) and (6) for a material with


known components and specied structures. Worked
examples of the detailed model derivation process are
shown below for two-component materials. If alternative
relationships are used instead of Eqs. (10) and (11), the
mathematics of solving for K will become slightly more
complex.
For materials with more than two components, extra
assumptions must be made in order for the model to be
completely and uniquely specied. Models for multi-com-
ponent materials are not considered further in this paper;
instead it is assumed that they may be dealt with by sequen-
tial application of the two-component models. For exam-
ple, for a material with three components x, y and z, the
overall eective thermal conductivity is calculated in two
stages. First, the eective thermal conductivity Kxy for
the mixture of two components, x and y, is predicted using
a two-component structural models with the following
adjusted volume fractions: vxx = vx/(vx + vy) and vyy =
vy/(vx + vy). Second, the overall eective thermal conduc-
tivity, K, is predicted for the mixture of xy as one compo-
nent and z as the other component. The volume fraction of
component xy is vxy = vx + vy, while the volume fraction of
z is vz, and the conductivities are Kxy and kz, respectively.
The two-component structural model used for this second
stage may, or may not, be the same as that used in the rst
stage. Clearly, there are two other possible sequences for
the calculations based on the choice of components to
include in the rst mixture (i.e. xz or yz instead of xy). Fig. 1. Schematic representations of some two-component materials as
The three alternative sequential calculations will usually uniform mixtures of two fundamental structural models: (a) ME1 + ME2,
not give identical results. (b) Parallel + EMT, (c) ME1 + EMT and (d) ME2 + EMT.
J. Wang et al. / International Journal of Heat and Mass Transfer 49 (2006) 30753083 3079

Rearranging Eq. (12) and cancelling terms yields: 1 Parallel


Series
/11 /22 A /12 /21 B 0 13 ME1
0.8
ME2
where EMT
0.6 ME1+ME2
A v1 v2 k 1  k 2 14

K/k1
Parallel+ME2
ME1+EMT
and 0.4
ME2+EMT
3k 1 3k 2 9A
B A 15 0.2
2k 1 k 2 2k 2 k 1 5 2k 1 =k 2 2k 2 =k 1

Based on Eqs. (4) and (10): 0


0 0.2 0.4 0.6 0.8 1
v2
v1 /11 v2 /21 1=2 16
Fig. 2. Plots of the ve fundamental eective thermal conductivity
v1 /12 v2 /22 1=2 17
structural models listed in Table 1, along with plots of the binary-
From Eq. (6): structure models shown schematically in Fig. 1ad for a two component
material with k1/k2 = 20.
/11 /12 1 18
/21 /22 1 19
3.2. Parallel + ME2 model
Eqs. (12) or (13) plus (16)(19) must be solved simulta-
neously to give values of /11, /12, /21, /22 and K given that For the Parallel + ME2 model, Eqs. (8) and (11) give
values of k1, k2, v1 and v2 are known. For this structure
3k 2
combination an analytical solution is possible. /12, /21 k 1 v1 /11 k 2 v2 /21 k 2 v2 /22 k 1 v1 /12 2k2 k 1
and /22 may all be written in terms of /11: K 26
v1 /11 v2 /21 v2 /22 v1 /12 2k3kk
2
2 1

/12 1  /11 20 Again, an analytical solution is possible. Eq. (26) is very


 
1  2v1 /11 similar to Eq. (12), which suggests that an expression for
/21 21 /11 will be similar to Eq. (24). In fact, it can be shown that
2v2
  Eqs. (24) and (25) may be used to calculate K for the
2v2 2v1 /11  1 Parallel + ME2 model provided B is calculated from
/22 22
2v2 Eq. (27) rather than Eq. (15):

Substituting for /12, /21 and /22 in Eq. (13): 3k 2 A


B 27
    2k 2 k 1
2v2 2v1 /11  1 1  2v1 /11
/11 A 1  /11 B 0 A plot of K/k1 vs. v2 for the Parallel + ME2 model is
2v2 2v2
shown in Fig. 2.
23
3.3. ME1 + EMT model
Rearranging Eq. (23) to be explicit in terms of /11:
q
B2v1 1  A2v2  1  A2v2  1  B2v1 12  8Bv1 A B For the ME1 + EMT model, Eqs. (8) and (11) give
/11
4v1 A B
k 1 v1 /11 k 2 v2 /21 2k3k 1
1 k 2
24 K
v1 /11 v2 /21 2k3k 1
1 k 2
(Note: if k1 > k2 then the positive square root in Eq. (24)
3K 3K
should be used, and vice versa). K can be calculated from k 1 v1 /12 2Kk 1
k 2 v2 /22 2Kk 2
28
the left-hand side version of Eq. (12) by substituting /21 3K
v1 /12 2Kk 3K
v2 /22 2Kk
1 2
using Eq. (21) and using the /11 value calculated from
Eq. (24): The right-hand side of Eq. (28) may be rearranged to be ex-
   plicit in terms of K:
k 1 v1 /11 k 2 v2 12v 1 /11
2v2
3k 1
2k 1 k 2
q
2
2k 1  k 2 v1 /12 2k 2  k 1 v2 /22 2k 1  k 2 v1 /12 2k 2  k 1 v2 /22  2k 1 k 2
K    25 K
12v1 /11 3k 1 2
v1 /11 v2 2v2 2k 1 k 2
29
A plot of K/k1 vs. v2 for the ME1 + ME2 model is shown in Expressing /11, /12 and /22 in terms of /21, Eq. (28)
Fig. 2 for a material with k1/k2 = 20. becomes:
3080 J. Wang et al. / International Journal of Heat and Mass Transfer 49 (2006) 30753083
 
1  2v2 /21 3k 1 to be expressed explicitly. Eq. (32) has been tted to /21
k 1 v1 k 2 v2 /21
2v1 2k 1 k 2 curves for k1/k2 values between 1 and 100:
K  
1  2v2 /21 3k 1 /21 f0:1353 lnk 1 =k 2  0:1193gv31
v1 v2 /21
2v1 2k 1 k 2
p f0:2551 lnk 1 =k 2  0:1711gv21
2
C C 2k 1 k 2 f0:1203 lnk 1 =k 2  0:0523gv1 0:5 32
30
2
Fig. 4 shows that Eq. (32) provides good approximations
where of /21 without the inconvenience of using an iterative
 
2v1 2v2 /21  1 numerical solution of the simultaneous equations. At
C 2k 1  k 2 v1 2k 2  k 1 v2 1  /21
2v1 k1/k2 values lower than 100, the agreement is closer than
31 shown in Fig. 4. A plot of K/k1 vs. v2 for the ME1 + EMT
model is shown in Fig. 2.
Due to the non-integral powers of /21 on its right-hand
side, Eq. (30) cannot be simply rearranged to be explicit 3.4. ME2 + EMT model
in terms of /21. Therefore alternative solution procedures
must be adopted. One possibility is to solve Eqs. (30), For the ME2 + EMT model, Eqs. (8) and (11) give
(31) and (16)(19) by numerical iteration. Fig. 3a shows
3k 2
the plots of /11, /12, /21 and /22 calculated by numerical k 2 v2 /21 k 1 v1 /11
iteration using the Solver function in Microsoft ExcelTM 2k 2 k 1
K
for a two-component material with k1/k2 = 20.0. The shape 3k 2
v2 /21 v1 /11
of the curve for /21 suggests that it could be accurately t- 2k 2 k 1
ted by a polynomial function, which would then allow /21 3K 3K
k 1 v1 /12 k 2 v2 /22
2K k 1 2K k 2
33
3K 3K
1 v1 /12 v2 /22
11 2K k 1 2K k 2
12
0.8
Following the same approach as for the ME1 + EMT
21
model, Eq. (33) becomes
22  
1  2v1 /11 3k 2
0.6 k 2 v2 k 1 v1 /11
2v2 2k 2 k 1
K  

1  2v1 /11 3k 2
0.4 v2 v1 /11
2v2 2k 2 k 1
p
D D2 2k 1 k 2
0.2 34
2
where
0  
0 0.2 0.4
v1
0.6 0.8 1 2v2 2v1 /11  1
(a) D 2k 1  k 2 v1 1  /11 2k 2  k 1 v2
2v2
1 11
35
12
0.8 21
22 1

0.6 0.8
21 (polynomial)

0.6
0.4

0.4
0.2
0.2

0 0
0 0.2 0.4 0.6 0.8 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
v1 21 (iteration)
(b)

Fig. 3. Plots of structure volume factors /11, /12, /21 and /22 for a two- Fig. 4. Comparison of /21 for the ME1 + EMT model calculated by
component material with k1/k2 = 20: (a) ME1 + EMT model and (b) numerical iteration with /21 calculated using Eq. (32) for a two
ME2 + EMT model. component material with k1/k2 = 100.
J. Wang et al. / International Journal of Heat and Mass Transfer 49 (2006) 30753083 3081

Again, a simple analytical solution is not possible. Fig. 3b v1 /13 v2 /23 1=3 41
shows the plots of /11, /12, /21 and /22 calculated by /11 /12 /13 1 42
numerical iteration for the ME2 + EMT model for a two
component material with k1/k2 = 20. As with the ME1 + /21 /22 /23 1 43
EMT model, a polynomial expression for /11 can be de- Due to the complexity of the algebra involved in solving
rived to avoid solution of Eq. (34) by numerical iteration. Eqs. (37)(43) simultaneously, solution by iteration with
For k1/k2 between 1 and 100: the aid of a tool such as ExcelTM Solver becomes a more
/11 f0:0526 lnk 1 =k 2 2 0:2125 lnk 1 =k 2  0:1689gv31
practical option than solving algebraically. Fig. 5 shows a
plot of the EMT + ME2 + Series model calculated using
f0:0886 lnk 1 =k 2 2  0:5758 lnk 1 =k 2 0:4879gv21 ExcelTM, along with plots of the ve basic structural models.
f0:0407 lnk 1 =k 2 2 0:3881 lnk 1 =k 2 0:3478gv1 0:5 Quaternary models (an example of which is plotted in
36 Fig. 5) and the ve-structure model (plotted in Fig. 5)
may also be derived by further extension of this method.
A plot of K/k1 vs. v2 for the ME2 + EMT model is shown By using every possible combination of the ve basic
in Fig. 2. structural models, 10 binary models, 10 ternary models,
ve quaternary models and one ve-structure model may
4. Multi-structure models be derived, giving a total of 26 new eective thermal con-
ductivity models, each with a distinct physical basis.
Extension of the procedure to a ternary-structure model
(EMT + ME2 + Series) gives: 5. Practical application
v1 /11 v2 /21
K
v1 /11 =k 1 v2 /21 =k 2 Clearly these combined models may not be applicable
3K 3K for materials whose structures could be mathematically
k 1 v1 /12 k 2 v2 /22 dened. In such cases, more specic models are often justi-
2K k 1 2K k 2
37 ed by greater prediction accuracy. But for naturally
3K 3K
v1 /12 v2 /22 occurring materials which are characterised by high degrees
2K k 1 2K k 2
of variability (such as soils and biological materials), sim-
3k 2 pler, more generic models, are more convenient.
k 2 v2 /23 k 1 v1 /13
2k 2 k 1 The ve basic structural models listed in Table 1 have
K
3k 2 been widely used, but without modication they cannot
v2 /23 v1 /13
2k 2 k 1 account for wide ranges of structure. Models such as
3K 3K Krischers use an empirical approach to account for dier-
k 1 v1 /12 k 2 v2 /22 ences in structure; however, unless there are data in the lit-
2K k 1 2K k 2
38 erature for f, the value of this parameter must be
3K 3K
v1 /12 v2 /22 determined by experimentation. This often defeats the pur-
2K k 1 2K k 2
pose of thermal conductivity prediction because it is very
v1 /11 v2 /21 1=3 39
dicult to perform an intuitive estimate of the value of f
v1 /12 v2 /22 1=3 40

1 f = 0.04
Parallel
1
Series f =0
ME1
0.8
ME2
0.8
EMT f = 0.1
EMT+ME2+Series
Parallel+ME1+EMT+ME2 0.6
0.6 f = 0.2
K/k1

Parallel+ME1+EMT+ME2+Series
K/k1

0.4 0.4 f = 0.35

0.2 0.2 f = 0.6

f =1
0
0 0.2 0.4 0.6 0.8 1 0
v2 0 0.2 0.4 0.6 0.8 1
v2
Fig. 5. Plots of ve fundamental eective thermal conductivity structural
models along with plots of selected ternary, quaternary and ve-structure Fig. 6. Plots of Krischers model (Eq. (2)) for dierent values of f for a
models for a two-component material with k1/k2 = 20. two-component material with k1/k2 = 20.
3082 J. Wang et al. / International Journal of Heat and Mass Transfer 49 (2006) 30753083

due to the highly non-linear dependence of K on f for a ling the thermal conductivity of frozen meat and oal prod-
given v2 as shown in Fig. 6. ucts, stated that: The best predictions by far are obtained
Models derived by the procedure outlined above have with Levys model, which yields correct values. . .over the
two major advantages over models such as Krischers that whole range of compositions. . .and temperatures. . .consid-
use empirical, structure-related parameters. Firstly, regard- ered. The drawback of Levys equation is a certain lack of
less of the value of f, Krischers model assumes a highly physical justication, since it was based on mathematical
anisotropic physical structure. Carson [25] has shown that rather than physical arguments.
over a range of compositions, thermal conductivity models A comparison of the eective thermal conductivities pro-
that assume anisotropic structures will not t experimental duced by Levys model and the ME1 + ME2 model above
data for isotropic structures as well as conductivity models shows that they are identical, and hence the two models
that are based on isotropic physical models, and, it is are equivalent. Therefore the physical model of Levys
reasonable to assume, vice versa. Models derived from equation, which must be the same as the ME1 + ME2
Eq. (1) may have either isotropic or anisotropic struc- model, is a homogeneous mixture (on the macroscopic
tures. scale) of equal volumes of the two Maxwell structures, in
Secondly, the models described in this work have dis- which the conductivities of both component structures are
tinct individual physical bases, which allow for an intuitive equal, as represented schematically in Fig. 1a.
approach to the selection of the appropriate model. For
example, Carson et al. [18] proposed thermal conductivity 7. Conclusions
bounds for two classes of porous materials. Internal
porosity materials were dened as materials in which A unifying equation has been developed for ve funda-
the gaseous phase was dispersed in a continuous high con- mental eective thermal conductivity structural models
ductivity phase the upper bound of thermal conductivity (Series, Parallel, two forms of MaxwellEucken, Eective
of internal porosity materials was given by the Maxwell Medium theory), which form the basis of many of the
Eucken model with the gaseous phase dispersed (ME1 in more complex models available in the literature. Structure
this paper), while the lower bound was dened by the volume fractions and structure composition factors were
EMT model. The ME1 + EMT model provides a logical dened and used in a new procedure for modelling complex
intermediate structure between these two extremes that materials as composites of these ve basic structures, using
may be more accurate for an internal porosity material that simple combinatory rules (equal structure volumes and
has neither a true Maxwell structure nor a true EMT struc- equal structure eective thermal conductivities). Worked
ture. With a model such as Krischers, the choice of f (or examples for deriving models using the proposed procedure
the equivalent parameter) might be little more than a guess. were presented. Each new model derived using this proce-
dure is dependent only on the component materials vol-
6. A physical basis for Levys model ume fractions and thermal conductivities, and not on any
empirical parameter. In addition, each model has a distinct
Levy [26] produced a model based on the MaxwellEuc- physical basis. The model combining the two forms of the
ken model that avoided the perceived problem of decid- MaxwellEucken model was shown to be equivalent to
ing which of the two MaxwellEucken equations to use for Levys equation, thereby providing Levys model with a
a given material, since they produced dierent results (in physical basis.
actual fact this was not a problem, since the two forms
were not supposed to produce the same result). The model Acknowledgement
was
This work was funded in part by the Foundation for Re-
2k 1 k 2  2k 1  k 2 F search, Science and Technology (New Zealand) as part of
K k1 44
2k 1 k 2 k 1  k 2 F Objective 2 of Contract C10X0201.
where
q References
2=G  1 2v2  2=G  1 2v2 2  8v2 =G
F 45 [1] R.C. Progelhof, J.L. Throne, R.R. Reutsch, Methods for predicting
2 the thermal conductivity of composite systems: a review, Polym. Eng.
and Sci. 16 (1976) 615625.
[2] J.K. Carson, S.J. Lovatt, D.J. Tanner, A.C. Cleland, Predicting the
2
k 1  k 2 eective thermal conductivity of unfrozen, porous foods, J. Food
G 2
46 Eng. 75 (3) (2006) 297307.
k 1 k 2 k 1 k 2 =2 [3] D.A.G. Bruggeman, Berechnung verschiedener physikalischer Konst-
However, since this model was derived solely by algebraic anten von heterogenen Substanzen. I. Dielektrizitatskonstanten und
Leitfahigkeiten der Mischkorper aus isotropen Substanzen (Calcula-
manipulation, with no stated physical basis, some subse-
tion of dierent physical constants of heterogeneous substances. I.
quent researchers have been hesitant to recommend its Dielectric constant and conductivity of media of isotropic sub-
use. For example, Pham and Willix [27], who were model- stances), Annalen der Physik 24 (1935) 636664.
J. Wang et al. / International Journal of Heat and Mass Transfer 49 (2006) 30753083 3083

[4] E. Behrens, Thermal conductivities of composite materials, J. [16] A.D. Brailsford, K.G. Major, The thermal conductivity of aggregates
Compos. Mater. 2 (1968) 217. of several phases including porous materials, Brit. J. Appl. Phys. 15
[5] T.H. Bauer, A general analytical approach toward the thermal (1964) 313319.
conductivity of porous media, Int. J. Heat Mass Transfer 36 (1993) [17] G.C.J. Bart, Thermal conduction in non homogeneous and phase
41814191. change media, Ph.D. thesis, Delft University of Technology, The
[6] E.E. Gonzo, Estimating correlations for the eective thermal Netherlands, 1994.
conductivity of granular materials, Chem. Eng. J. 90 (2002) 299302. [18] J.K. Carson, S.J. Lovatt, D.J. Tanner, A.C. Cleland, Thermal
[7] O. Krischer, Die wissenschaftlichen Grundlagen der Trocknungs- conductivity bounds for isotropic, porous materials, Int. J. Heat
technik (The Scientic Fundamentals of Drying Technology), Mass Transfer 48 (11) (2005) 21502158.
Springer-Verlag, Berlin, 1963. Cited in English in: R.B. Keey, Drying [19] S. Kirkpatrick, Percolation and conduction, Rev. Modern Phys. 45
of Loose and Particulate Materials, Hemisphere Publishing Corpo- (1973) 574588.
ration, New York, 1992 (Chapter 7). [20] H. Fricke, The electrical conductivity of two-phase materials, Phys.
[8] R.P.A. Rocha, M.E. Cruz, Computation of the eective conductivity Rev. 24 (1924) 575587.
of unidirectional brous composites with an interfacial thermal [21] R.L. Hamilton, O.K. Crosser, Thermal conductivity of heterogeneous
resistance, Numer. Heat Transfer, Part A: Applicat. 39 (2) (2001) two-component systems, Indust. Eng. Chem., Fund. 1 (1962) 187
179203. 191.
[9] G. Buonanno, A. Carotenuto, The eective thermal conductivity of [22] O. Wiener, Die theorie des Mischkorpers fur das Feld der Statonaren
packed beds of spheres for a nite area, Numer. Heat Transfer, Part Stromung. I. Die Mittelwertsatze fur Kraft, Polarisation und Energie,
A: Applicat. 37 (4) (2000) 343357. Der Abhandlungen der Mathematisch-Physischen Klasse der Konigl,
[10] M. Christon, P.J. Burns, R.A. Sommerfeld, Quasi-steady temperature Sachsischen Gesellschaft der Wissenschaften 32 (1912) 509604.
gradient metamorphism in idealized dry snow, Numer. Heat Transfer, [23] D.R. Chaudhary, R.C. Bhandari, Heat transfer through a three-phase
Part A: Applicat. 25 (3) (1994) 259278. porous medium, J. Phys. D (Brit. J. Appl. Phys.) 1 (1968) 815817.
[11] J.C. Maxwell, A Treatise on Electricity and Magnetism, third ed., [24] T. Renaud, P. Briery, J. Andrieu, M. Laurent, Thermal properties of
Dover Publications Inc., New York, reprinted 1954 (Chapter 9). model foods in the frozen state, J. Food Eng. 15 (2) (1992) 8397.
[12] A. Eucken, Allgemeine Gesetzmaigkeiten fur das Warmeleitvermo- [25] J.K. Carson, Prediction of the thermal conductivity of porous foods,
gen verschiedener Stoarten und Aggregatzustande, Forschung Ph.D. thesis, Massey University, Palmerston North, New Zealand,
Gabiete Ingenieur 11 (1) (1940) 620. 2002.
[13] C.J.F. Bottcher, Theory of Electric Polarization, Elsevier, Houston, [26] F.L. Levy, A modied MaxwellEucken equation for calculating the
1952, pp. 415420. thermal conductivity of two-component solutions or mixtures, Int. J.
[14] R. Landauer, The electrical resistance of binary metallic mixtures, J. Refrigeration 4 (4) (1981) 223225.
Appl. Phys. 23 (1952) 779784. [27] Q.T. Pham, J. Willix, Thermal conductivity of fresh lamb meat, oal
[15] Z. Hashin, S. Shtrikman, A variational approach to the theory of the and fat in the range 40 to +30 C: measurements and correlations, J.
eective magnetic permeability of multiphase materials, J. Appl. Phys. Food Sci. 54 (3) (1989) 508515.
33 (1962) 31253131.