ARTICLE IN PRESS

Physica B 368 (2005) 302–307

[Link]/locate/physb

Model for thermal conductivity of carbon

nanotube-based composites
Q.Z. Xue
College of Physics Science and Technology, China University of Petroleum, Dongying, Shandong 257061, P.R. China
Received 29 September 2004; received in revised form 28 July 2005; accepted 28 July 2005

Abstract

Considering the carbon nanotubes’ (CNTs) orientation distribution a new model of effective thermal conductivity of
CNTs-based composites is presented. Based on Maxwell theory, the formulae of calculating effective thermal
conductivity of CNTs-based composites are given. The theoretical results on the effective thermal conductivity of
CNTs/oil and CNTs/decene suspensions are in good agreement with the experimental data. The model is valid for the
transport properties of the CNTs-based composites.
r 2005 Elsevier B.V. All rights reserved.

PACS: 44.30.+v; 44.10.+I; [Link]

Keywords: Model; Thermal conductivity; Nanofluid

1. Introduction conductivity. It can be expected that the CNTs-

based suspensions can enhance thermal conduc-
Carbon nanotubes (CNTs), due to their unique tivity and the improved thermal performance
structure and remarkable physical properties, have would be applied to energy systems [6–8].
attracted much attention in the past few years Recently, Chio et al. reported that the enhance-
[1–4]. Recently, the rapid advance in the bulk ment in thermal conductivity of suspension for
synthesis of CNTs makes it possible to produce a 1.0 vol% CNTs in oil is 160% [8]. The anomalous
host of CNTs-based composites [5]. Recent studies thermal conductivity enhancement is theoretically
reveal that CNTs have anomalously high thermal interesting because the measured thermal conduc-
tivities have been considered to be much larger
than the prediction from the existing models.
Tel.: +86 546 8392836; fax: +86 546 8392123. The expressions of the conventional models of
E-mail address: xueqingzhong@[Link]. the effective thermal conductivity of two-phase

0921-4526/$ - see front matter r 2005 Elsevier B.V. All rights reserved.
doi:10.1016/[Link].2005.07.024
 ARTICLE IN PRESS

Q.Z. Xue / Physica B 368 (2005) 302–307 303

composites are briefly summarized as follows [8]: theory, two formulae of calculating the effective
ke 3ða  1Þf thermal conductivity of CNTs-based composites
¼1þ ; Maxwell ½9, are given. The model shows that the axial ratio
km ða þ 2Þ  ða  1Þf
and the space distribution of the CNTs can largely
ke a þ ðn  1Þ  ðn  1Þð1  aÞf affect the effective thermal conductivity of CNTs-
¼ , based composites so that the dispersion of quite a
km a þ ðn  1Þ þ ð1  aÞf
small amount of CNTs can result in a remarkable
Hamilton and Crosser ½10, enhancement in the effective thermal conductivity

 of the composites that the existing models cannot
ke 2 3b2 9b3 a þ 2 understand.
¼ 1 þ 3bf þ 3b þ þ þ    f 2,
km 4 16 2a þ 3
Jeffery ½11,
2. Model for thermal conductivity of carbon
ke 3ða  1Þ nanotube-based composites
¼1þ ½f þ f ðaÞf 2 þ 0ðf 3 Þ,
km ða þ 2Þ  ða  1Þf
Davis ½12, We assume that ke is the effective thermal
conductivity of the CNTs-based composites, kc
ke and km are the thermal conductivities of the CNTs
¼ 1 þ af þ bf 2 ; Lu and Lin ½13, and the main media, respectively.
km ~ and heat flux ~
Let the temperature field E q be,
where ke is the effective thermal conductivity of respectively, defined as
two-phase composites, km and kc are the thermal
conductivities of the main phase and scattered ~ ¼ rf,
E (1)
phase, respectively, n is the shape factor of a
particle given by n ¼ 3=j, where j is defined as the ~ ~
q ¼ kE, (2)
ratio of the surface area of a sphere (with a volume where f and k are the temperature distribution
equal to that of the particle) to the surface area of function and the thermal conductivity.
the particle, f is the particle volume fraction of the The field factor is defined as the ratio between
scattered phase, and a ¼ kc =km , b ¼ ða  1Þ= the temperature field E ~c in the CNT to the mean
ða þ 2Þ, a and b are pure fitting parameters [9–11]. ~
temperature field E 0 in the composite. We assume
All the existing models are only valid for the that the applied field is along the j-axis (j ¼ x; y; z),
composites containing the spherical or rotational according to Maxwell theory the field factor
elliptical particles with small axial ratio M component along j-axis can be expressed as
(M ¼ a=b), where a, b, c ( ¼ b) are the semi-radii
along j-axis (j ¼ x; y; z). However, CNTs, in fact, E c;j 1
cj ¼ ¼  , (3)
can be regarded as rotational elliptical nanoparticles E 0;j 1 þ Bj kc =km  1
with a very large axial ratio M (M ¼ ða=bÞb1) so where Bj is the depolarization factor component of
that the existing models cannot work on the CNTs- the CNT along j-axis, which is a parameter used to
based composites. Besides, the existing models characterize the shape of the particle [14–17]. The
cannot describe the effect of the space distribution CNTs can be considered as the same rotational
of the CNTs on the thermal conductivity. Due to elliptical nano-particles with very large axial ratio
the defects, the existing models fail to describe the M (M ¼ a=bb1). So, the depolarization factor
experimental data, and there is really no way to component is expressed as follows [15,16]:
make them work on the CNTs-based composites.
1 M pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Therefore, considering the very large axial ratio
Bx ¼ þ q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi lnðM þ M 2  1Þ,
and the space distribution of the CNTs we built a 1  M2 ð1  M Þ 2 3
new model for the effective thermal conductivity
of CNTs-based composites. Based on Maxwell M41, ð4Þ
 ARTICLE IN PRESS

304 Q.Z. Xue / Physica B 368 (2005) 302–307

1  Bx The effective thermal conductivity component

By ¼ Bz ¼ . (5)
2 ke;j can be obtained as
According to Eq. (4), when the length direction hqj i fkc hE c;j i þ ð1  f Þkm hE 0;j i
of CNTs is parallel (or vertical) to the x-axis ke;j ¼ ¼
direction, the value of Bx is equal to 0 (or 1/2). As hE j i f hE c;j i þ ð1  f ÞhE 0;j i
we know, the CNTs are randomly dispersed in the fkc c̄j þ ð1  f Þkm
¼ , ð9Þ
main media. In other words, their space orienta- f c̄j þ ð1  f Þ
tion is stochastic. According to Eq. (4) and the
above discussion, when the applied field is along a where f is the total volume fraction of the CNTs.
stable direction, say x-axis, the value of Bx Due to the assumption that the thermal conduc-
will range from 0 to 1/2, due to the angle, between tivity of the CNTs-based composites is isotropy,
the applied field and the length direction of the the effective thermal conductivity components is
CNT, ranges from 0 to p=2. In other words, the isotropy.
value of the depolarization factor of CNTs, Substituting Eq. (8) into (9), we have the expres-
determined by their orientation distribution, sion of the effective thermal conductivity of CNTs-
ranges from 0 to 1/2. Therefore, in order to based composites.
describe this orientation distribution, we can pffiffiffiffiffiffiffiffiffiffiffiffiffi  pffiffiffiffiffiffiffiffiffiffiffiffiffi
introduce a distribution function PðBj Þ of the 1  f þ ð4f =pÞ kc =km arctg p=4 kc =km
ke ¼ km pffiffiffiffiffiffiffiffiffiffiffiffiffi  pffiffiffiffiffiffiffiffiffiffiffiffiffi .
depolarization factor. Therefore, the equivalent 1  f þ ð4f =pÞ km =kc arctg p=4 kc =km
field factor will be expressed as (10)
Z 1=2 As discussed above, the CNTs’ distribution can
hqj i 1
c̄j ¼ ¼ PðBj Þ dBj , affect effective thermal conductivity of CNTs-
hE j i 0 1 þ Bj ðkc =km  1Þ based composites. For comparison, we assume
(6) that the form of distribution function PðBj Þ can
selected as
where hqj i and hE j i are the spatial average of the
heat flux component and the temperature field PðBj Þ ¼ 2. (11)
component along the j-axis, respectively. In this case, the equivalent field factor will be
The distribution function PðBj Þ must satisfy the expressed as
R 1=2
two conditions: 0obj o1=2 and 0 PðBj Þ dBj ¼ 1. Z 1=2
1
Usually, the CNTs are randomly distributed in the jj ¼ 2   dBj
0 1 þ B j k c =km  1
nanofluid, a normal-like distribution function
2km  
PðBj Þ can be adopted [14]: ¼ ln kc þ km =2km . ð12Þ
kc  km
2 The corresponding expression of the effective
PðBj Þ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi . (7)
p Bj ð1  Bj Þ thermal conductivity of CNTs-based composites
can be deduced as
For the nanofluid with randomly distributed
kc
CNTs, the equivalent field factor will be expressed 1  f þ 2f kc km ln kc2kþkmm
as ke ¼ km . (13)
1  f þ 2f km
kc km ln kc2k
þkm
m

Z 1=2
2 1
c̄j ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi   dBj
0 p B j ð1  B j Þ 1 þ B j k c =km  1 3. Numerical calculations and discussion
sffiffiffiffiffiffi sffiffiffiffiffiffi!
4 km p kc
¼ arctg . ð8Þ Using Eqs. (10) and (13) we calculated the
p kc 4 km
thermal conductivity of the CNTs–oil and
 ARTICLE IN PRESS

Q.Z. Xue / Physica B 368 (2005) 302–307 305

CNTs–decene suspensions, respectively. In the those theoretical results given by the existing
calculation, the thermal conductivity of CNTs models are much smaller than the experimental
is taken as 600–3000 W/mK, and the thermal data, even if we select the thermal conductivity of
conductivities of the oil and decene are taken CNTs 3000 W/mK (Fig. 1b). For example, the
as 0.1448 W/mK [8] and 0.14 W/mK [7], res- measured enhancement in thermal conductivity
pectively. for 1 vol% CNTs in oil is 160%, whereas the
As shown in Fig. 1a, when the thermal enhancement predicted by Hamilton–Crosser
conductivity of CNTs is taken as 600 W/mK the model is not more than 10% (Fig. 1b) and the
theoretical results given by Eq. (10) is in good enhancements predicted by the other four models
agreement with the experimental data, whereas are almost identical, about 3%.

5 1.2
Kc = 3000 W/mK

Thermal conductivity ratio (Ke/Km)

Thermal conductivity ratio (Ke/Km)

kc=600 W/mK Kc = 2000 W/mK

kc=900 W/mK Kc = 1500 W/mK
4 Kc =1000 W/mK
kc=1500 W/mK
Experimental data
kc=2000 W/mK
Experimental data 1.1
3

1.0
1 0.0 0.2 0.4 0.6 0.8 1.0
0.0 0.2 0.4 0.6 0.8 1.0
(a) CNTs volume fraction %
(a) CNTs volume fraction (%)

1.08 1.2
Maxwell
Thermal conductivity ratio (Ke/Km)
Thermal conductivity ratio (Ke/Km)

Maxwell Davis
Davis Jeffery
Jeffery Lu-Lin
1.06
Lu-Lin Hamiton-Crosser
Experimental data
Hamiton-Crosser

1.04 1.1

1.02

1.00 1.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
(b) CNTs volume fraction (%) (b) CNTs volume fraction %

Fig. 1. (a) Comparison between experimental data of CNTs–oil Fig. 2. (a) Comparison between experimental data of
suspensions and the calculated values using Eq. (10). (b) CNTs–decene suspensions and the calculated values using
Calculated thermal conductivity of CNTs–oil suspensions by Eq. (13). (b) Calculated thermal conductivity of CNTs–decene
the conventional models. In the calculation, n ¼ 3 for the suspensions by the conventional models. In the calculation, n ¼
Hamiton–Crosser model [10]; f ðaÞ ¼ 0:5 for Davis model [12]; 3 for the Hamiton–Crosser model [10]; f ðaÞ ¼ 0:5 for Davis
a ¼ 3 and b ¼ 4:51 for Lu–Lin model [13]. model [12]; a ¼ 3 and b ¼ 4:51 for Lu–Lin model [13].
 ARTICLE IN PRESS

306 Q.Z. Xue / Physica B 368 (2005) 302–307

As shown in Fig. 2a, when the thermal conductivity of the CNTs–oil suspensions again.
conductivity of CNTs is taken as 3000 W/mK, We find that the theoretical results are still smaller
the theoretical results given by Eq. (13) is in good than the experimental data, even though the
agreement with the experimental data, whereas the theoretical results are much larger that that of
theoretical results given by the existing models are conventional models.
much smaller than the experimental data, even if From the discussion above, the distribution
we select the thermal conductivity of CNTs state of the CNTs has an important effect on the
3000 W/mK (Fig. 2b). For example, the measured thermal conductivity of the nanofluid. Because we
enhancement in thermal conductivity for 1 vol% have considered the effect of the distribution of the
CNTs in decene is 19.6%, whereas the enhance- CNTs in our model, the thermal conductivity
ment predicted by Hamilton–Crosser model is not enhancement in the composites with dilute CNTs
more than 10% (Fig. 2b) and the enhancements is not anomalously beyond our model predictions.
predicted by the other four models are almost
identical, about 3%.
As we noted, Choi et al. reported that the
4. Conclusion
thermal conductivity enhancement for suspension
with 1.0 vol% CNTs in oil (km ¼ 0.1448 W/mK)
In summary, considering the very large axial
was 160%. However, Xie et al. reported that the
ratio and the space distribution of the CNTs, a
thermal conductivity enhancement for nanofluid
simple model of effective thermal conductivity of
with 1.0 vol% CNTs in decene (km ¼ 0.14 W/mK)
CNTs-based composites is presented. Based on
was only 19.6%. We believe that this discrepancy
Maxwell theory, two formulae of calculating the
would be attributed to the difference of the
effective thermal conductivity of CNTs-based
distribution state of the CNTs in the nanofluid.
composites are given. The model shows that
As shown in our model, the distribution state of
dispersion of quite a small amount of CNTs can
the CNTs in the nanofluid has an important effect
result in a remarkable enhancement in the effective
on the thermal conductivity of the nanofluid.
thermal conductivity of the composites that the
Therefore, in this model we introduce two forms of
existing models cannot understand. By comparing
the distribution function PðBj Þ of the CNTs. In
with the experiment data of the CNTs-contained
Fig. 3, using Eq. (13) we calculated the thermal
suspensions, we can find that the observed
anomalous thermal conductivity enhancement lies
within reasonable range covered by our model.
2.5
Kc = 3000 W/mK
Thermal conductivity ratio (Ke/Km)

Kc = 2000 W/mK
Kc =1500 W/mK References
Experimental data
2.0 [1] S. Iijima, Nature (London) 354 (1991) 56.
[2] P.M. Ajayan, O. Stephen, C. Cplliex, D. Trauth, Science
265 (1994) 1212.
[3] W.A. de Heer, W.S. Bacsa, A. Chatelain, R.H. Baker,
L. Forro, D. Ugarte, Science 268 (1995) 845.
1.5 [4] H. Dai, E.W. Wong, C.M. Lieber, Science 272 (1996) 523.
[5] R. Ramasubramaniam, J. Chen, H.Y. Liu, Appl. Phys.
Lett. 83 (2003) 2928.
[6] M.J. Biercuk, M.C. Liaguno, M. Radosavljevic, J.K.
Hyun, A.T. Johnson, J.E. Fischer, Appl. Phys. Lett. 80
1.0
0.0 0.2 0.4 0.6 0.8 1.0 (2002) 2767.
CNTs volume fraction % [7] H.Q. Xie, Hohyun Lee, Wonjin Youn, M. Choi, J. Appl.
Phys. 94 (2003) 4967.
Fig. 3. Comparison between experimental data of CNTs–oil [8] S.U.S. Choi, Z.G. Zhang, W. Yu, F.E. Lockwood, E.A.
suspensions and the calculated values using Eq. (13). Grulke, Appl. Phys. Lett. 79 (2001) 2252.
 ARTICLE IN PRESS

Q.Z. Xue / Physica B 368 (2005) 302–307 307

[9] J.C. Maxwell, A Treatise on Electricity and Magnetism, [12] R.H. Davis, Int. J. Thermopyhs. 7 (1986) 609.
second ed., Oxford University Press, Cambridge, UK, [13] S. Lu, H. Lin, J. Appl. Phys. 79 (1996) 6761.
1904, p. 435. [14] L. Gao, J.Z. Gu, J. Phys. D: Appl. Phys. 35 (2002)
[10] R.L. Hamilton, O.K. Crosser, IEC Fundamentals 2 (1962) 267.
187. [15] Q.Z. Xue, Eur. Polym. J. 40 (2004) 323.
[11] D.J. Jeffrey, Proc. Roy. Soc. London, Ser. A 335 (1973) [16] W.T. Doyle, I.S. Jacobs, J. Appl. Phys. 71 (1992) 3926.
355. [17] Q.Z. Xue, Phys. Lett. A 307 (2003) 313.