Sei sulla pagina 1di 10

Computational Materials Science 25 (2002) 329–338

www.elsevier.com/locate/commatsci

Apparent thermal conductivity of periodic


two-dimensional composites
M. Jiang a, I. Jasiuk b,*
, M. Ostoja-Starzewski c

a
GE Global Research Center, 1 Research Circle, Building KW, Rm. C1610J, Niskayuna, NY 12309, USA
b
The George W. Woodruff School of Mechanical Engineering, Georgia Institute of Technology, Atlanta, GA 30332-0405, USA
c
Department of Mechanical Engineering, McGill University, 817 Sherbrooke Street West, Montr
eal, Que., Canada H3A 2K6
Received 1 July 2001; received in revised form 10 February 2002; accepted 1 March 2002

Abstract

Effects of scale and boundary conditions, as well as mismatch between component phases, on apparent thermal
conductivity of periodic, square-array, two-dimensional composites are investigated in this paper. The apparent con-
ductivities are defined as those under either essential, natural, mixed or periodic boundary conditions applied to finite
size material domains. It is shown that apparent conductivities obtained under mixed and periodic conditions are the
same within numerical accuracy, and they are bounded by those obtained under essential and natural boundary
conditions. Bounds are very sensitive to the mismatch of phase conductivities: the higher the mismatch, the wider the
bounds. Bounds tighten as the window size increases. It is also observed that the windowÕs location strongly affects the
boundsÕ sharpness.
Ó 2002 Elsevier Science B.V. All rights reserved.

1. Introduction tions, phase microgeometries (distribution infor-


mation) and scale sizes can only be determined by
Scale and boundary conditions effects have been experiment or computational micromechanics.
investigated during the last decade for random in- Few investigations on these issues have been
homogeneous materials—for both linear [1–3], and done for periodic composites, even though they
nonlinear materials [4,5]. The basic idea is to use are easier to carry out compared with those of
variational principles to obtain bounds, and then random composites. Hollister and Kikuchi [6],
their order structures and hierarchy structures. The Pecullan et al. [7] and Jiang et al. [8] investigated
details of these structures, critically depending on scale and boundary conditions effects in elastic
the mismatch of phase properties, volume frac- or elasto-plastic behavior of periodic two-dimen-
sional (2D) composites. The motivations for such
investigations include:
*
Corresponding author. Tel.: +1-404-894-6597; fax: +1-404-
894-0186. (a) The understanding of periodic composites can
E-mail address: iwona.jasiuk@me.gatech.edu (I. Jasiuk). give better insights into random composites.

0927-0256/02/$ - see front matter Ó 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 9 2 7 - 0 2 5 6 ( 0 2 ) 0 0 2 3 4 - 3
330 M. Jiang et al. / Computational Materials Science 25 (2002) 329–338

(b) The properties of periodic structures consist-


ing of a finite number of periodic cells may
be very different from the effective ones.

In this paper, we investigate thermal conduc-


tivity of 2D composites with locally isotropic
phases; this is equivalent to anti-plane elasticity
problem, electrical conductivity problem, etc., (see
the Appendix A for a list of equivalent problems).
It is governed by the 2D LaplaceÕs equation, which
is easier to solve than the governing equation of
2D (in-plane) and three-dimensional (3D) elastic-
ity problems. In addition, because only one ma-
terial constant, i.e., thermal conductivity, is needed Fig. 1. Periodic square-array composite.
to describe each phase, the parametric investiga-
tion of such a problem is much simpler.
The influence of several factors on apparent
conductivity is investigated, such as the window (a
piece of composite with the same volume fraction
as the original composite, in ensemble average
sense for the random composite) size effect, the
mismatch in material properties effect, and the
boundary conditions effect. The boundary condi-
tions include essential, natural, mixed and peri-
odic. The property obtained under a certain
boundary condition applied to a window is re-
ferred to as an apparent property, following Huet
[2]. The apparent property of a periodic composite Fig. 2. Unit cells.
under periodic boundary conditions (PBCs) is just
an effective property.
We show that the apparent properties under we should be able to construct the whole com-
mixed and periodic conditions are the same within posite by repeating of unit cell in x1 and x2 direc-
numerical accuracy, and are bounded by those tions. Second, the effective conductivity obtained
obtained under essential and natural boundary from the unit cell should be isotropic. If we define
conditions (NBCs). Bounds are very sensitive to the scale of a square unit cell of Fig. 2 to be d ¼ d0
mismatch of phase conductivities: the higher the (d0 ¼ L0 =d, where L0 is the edge length of a unit
mismatch, the wider the bounds. We also investi- cell and d is the diameter of inclusion), by applying
gate the effect of a windowÕs location on boundsÕ the second requirement on any scale of window of
sharpness. edge length L, we choose the windows of sizes
d ¼ 2d0 , d ¼ 4d0 and d ¼ 8d0 , respectively. nd0
means that the windowÕs edge is n times larger than
2. Composite structure, unit cells and boundary the unit cell in both x1 and x2 directions.
conditions For each phase the system is described by
 2 
o T o2 T
We study the composite with a square packing C þ ¼0 ð1Þ
ox21 ox22
of inclusions as shown in Fig. 1. The two choices
of unit cells are sketched in Fig. 2. We choose these with Cð~ xÞ ¼ v1 ð~
xÞC1 þ v2 ð~
xÞC2 . ~
x is a position
unit cells on the basis of two considerations. First, vector, T is temperature, Ci ¼ ki I (i ¼ 1 for matrix,
M. Jiang et al. / Computational Materials Science 25 (2002) 329–338 331

or 2 for inclusions) is the conductivity tensor for assumptions of uniform temperature gradient and
isotropic phase with a constant conductivity ki , I is uniform heat flux, respectively, and S is the resis-
the identity tensor of second order, and vi ð~
xÞ are tivity tensor. Also following [9,1], one can prove
the indicator functions of the regions occupied by that, for an arbitrary window of size d
phase i. Locally, we also have
ðSdN Þ1 6 CdM 6 CdE ð7Þ
qð~
~ xÞ ¼ Cð~
xÞ  rT ð~
xÞ ð2Þ
The relationships in (6) and (7) are to be under-
where ~ qð~
xÞ is the local heat flux vector and rT ð~
xÞ stood as follows: for two second-rank tensors
is the local temperature gradient. A and B, the order relation B 6 A means that
The apparent properties of a given size window ~
t  B ~
t 6~ t  A ~
t for any vector ~ t 6¼ ~0. For the
are obtained under four kinds of boundary con- composite studied in this paper, the microstructure
ditions, which are: is isotropic in the sense of thermal conductivity,
Periodic boundary condition (PBC) i.e., C eff ¼ k eff I, where constant k eff depends on
phasesÕ conductivities and volume fraction of in-
x þ~
T ð~ xÞ þ rT  ~
LÞ ¼ T ð~ L; clusions. From Eqs. (6) and (7) we know that ap-
qð~
~ x þ~
LÞ ¼ ~qð~
xÞ; 8~x 2 oBd ð3Þ parent properties depend on boundary conditions
and window size, and the influence of these factors
which yields the effective conductivity tensor C eff . disappears as the window size goes to infinity.
oBd is the boundary of window domain Bd . rT is
the spatial average temperature gradient, ~ L ¼ L~ e
(L is the length of periodicity or, equivalently, d; ~
e 3. Numerical experiments
is the unit vector) and ~q is the heat flux.
Essential boundary condition (EBC) To find apparent properties as defined above,
one must solve the corresponding boundary value
T ¼ rT ~
x; 8~
x 2 oB ð4Þ problems by a numerical method. In this paper, we
which yields a tensor CdE (E stands for an essential employ the finite element software ANSYS 5.4.
condition). In particular, the volume fraction of inclusions is
chosen to be 0.35. The thermal conductivity of
Natural boundary condition (NBC) matrix is assumed (without loss of generality) to
q ~
~ n ¼~
q ~
n ð5Þ be 1, and the mismatch between the thermal con-
1 ductivities of inclusion and matrix varies from
which yields a tensor ¼ CdN ðSdN Þ
(N stands for a 0.001 to 1000 (in this paper, we refer to the mis-
natural condition), where ~q is the spatial average match less than unity as poorly conducting inclu-
heat flux, and ~ n is the outer normal to the win- sion case, while mismatch greater than unity to
dowÕs boundary. be a highly conducting inclusion case). After the
Mixed boundary condition (MBC): We apply es- temperature field is obtained, we can use two
sential condition on one pair of edges and natural methods to find the apparent properties, i.e., by
condition on the other pair of edges. This condi-
tion yields a tensor CdM (M stands for a mixed q ¼ C app  rT
~ ð8Þ
condition). or
Following Huet [2], it can be proved that
U ¼ 12Vd  rT  C app  rT ð9Þ
R R 1 1 1 N 1
C
ðS Þ 6 ðSdN0 Þ 6 ðSnN0 d0 Þ 6 ðSnd0
Þ 6C eff

where U is the entropy rate, C app is an apparent


E 0
6 Cnd0
6 CnE0 d0 6 CdE0 6C ; V
81 < n < n ð6Þ conductivity tensor under a certain boundary
condition, Vd is the area of a given window.
where C V and C R denote the Voigt and Reuss It has been shown that, under EBCs, apparent
bounds of conductivity tensor, corresponding to properties obtained from Eqs. (8) and (9) are
332 M. Jiang et al. / Computational Materials Science 25 (2002) 329–338

equivalent. The same holds separately for natural So, in the following, we take the form of C app to be
and properly constructed mixed conditions. This  app 
k 0
results from the Hill condition [1]. In our numer- C app ¼ ð10Þ
0 k app
ical approach, we use Eq. (9) to obtain the ap-
parent properties. and we only report k app for different window sizes,
During our numerical calculation of mismatches and boundary conditions. From (6)
 app app  and (9), we can obtain the following hierarchy
C11 C12
C app ¼ app app structure
C21 C22
a given rT is used in PBCs and EBCs, and a given k R 6 kdN0 6 knN0 d0 6 knd
N
0
6 k eff 6 knd
E
0
6 knE0 d0
q is used in NBCs. Specifically, for PBCs and
~
T app 6 kdE0 6 k V ; 81 < n0 < n ð11Þ
EBCs, we apply rT ¼ ð1; 0Þ to calculate C11 ¼
app T
C22 , then we apply rT ¼ ð1; 1Þ to calculate
app app T and we only report k app for different window sizes,
C12 ¼ C21 . For NBCs, we apply  q ¼ ð1; 0Þ to
N N T mismatches and boundary conditions. The nu-
calculate S11 ¼ S22 , then we apply ~ q ¼ ð1; 1Þ to
N N merical results are given in Tables 1 and 2, and
calculate S12 ¼ S21 . When boundary condition is
shown in Fig. 3.
mixed, on the pair of edges in the x1 -direction, we
fix the temperature of nodes according to their x1
coordinates, and we specify the heat flux to be zero
on the pair of edges in x2 -direction. First we il- Table 1
lustrate some results under EBCs and NBCs. For Apparent property as a function of mismatch, boundary con-
the unit cell (1) of Fig. 2, for example, when d ¼ d0 dition and window size; windows are generated using the unit
cell (1)
and k1 ¼ 1, k2 ¼ 10, we have
NBC PBC or EBC Window
  MBC size
E
3:1128 0:0019
C ¼ and
0:0019 3:1128 k2 =k1 ¼ 0:001 0.480e)2 0.4821 0.5264 d ¼ d0
  0.954e)2 0.5052 d ¼ 2d0
0:5919 3:3 105 0.0187 0.4939 d ¼ 4d0
SN ¼
3:3 105 0:5919 0.0732 0.4855 d ¼ 8d0
0.01 0.0458 0.4890 0.5327 d ¼ d0
and when k1 ¼ 1, k2 ¼ 1000, we have 0.0841 0.5118 d ¼ 2d0
0.1435 0.5007 d ¼ 4d0
 
208:84 0:3350 0.3278 0.4980 d ¼ 8d0
CE ¼ and
0:3350 208:84 0.1 0.3217 0.5546 0.5927 d ¼ d0
  0.4078 0.5747 d ¼ 2d0
0:5251 3:9 105
SN ¼ 0.4703 0.5651 d ¼ 4d0
3:9 105 0:5251 0.5544 0.5601 d ¼ 8d0

app app app app app 10 1.689 1.803 3.113 d ¼ d0


We can see that C11 ¼ C22 , C12  C11 , S11 ¼ 1.744 2.459 d ¼ 2d0
app app app eff
S22 and S12  S11 , which is trivial for C and 1.772 2.131 d ¼ 4d0
S eff , but not for C E and S N . When window size is 1.773 1.872 d ¼ 8d0
larger, C app becomes more isotropic. For example, 100 1.879 2.044 21.86 d ¼ d0
for d ¼ 4d0 , when k1 ¼ 1, k2 ¼ 10, we have 1.960 11.97 d ¼ 2d0
2.002 7.009 d ¼ 4d0
  2.011 3.281 d ¼ 8d0
E 2:1308 1:1 108
C ¼ and 1000 1.902 2.074 208.8 d ¼ d0
1:1 108 2:1308 1.986 105.6 d ¼ 2d0
 
1:7725 2:4 105 2.030 53.85 d ¼ 4d0
SN ¼ 2.040 15.10 d ¼ 8d0
2:4 105 1:7725
M. Jiang et al. / Computational Materials Science 25 (2002) 329–338 333

Table 2 From Table 1 it is found that, for the same


Apparent property as a function of mismatch, boundary con- volume fraction, under PBCs and MBCs, the ap-
dition and window size; windows are generated using the unit
cell (2)
parent properties are the same for both unit cells
(1) and (2), within numerical accuracy. This tells us
NBC PBC or EBC Window
MBC size
that one can use either of the unit cells to estimate
the apparent––which happens to be effective––
k2 =k1 ¼ 0:001 0.4703 0.4821 0.4955 d ¼ d0
property under either MBC or PBC. It should be
0.4768 0.4894 d ¼ 2d0
0.4796 0.4859 d ¼ 4d0 pointed out that the MBC is a common setup in
0.4817 0.4832 d ¼ 8d0 experiments. The above observation gives us rea-
sons to use the mixed condition in practice to
0.01 0.4775 0.4890 0.5021 d ¼ d0
0.4838 0.4961 d ¼ 2d0 obtain the effective conductivity for 2D isotropic
0.4866 0.4928 d ¼ 4d0 periodic composite. Because of the equivalence of
0.4881 0.4912 d ¼ 8d0 the apparent properties obtained under MBCs and
0.1 0.5462 0.5546 0.5648 d ¼ d0 PBCs, there are no scale effects for those proper-
0.5511 0.5604 d ¼ 2d0 ties, i.e., such apparent properties are the same for
0.5533 0.5580 d ¼ 4d0 other window sizes. The contour plots of temper-
0.5540 0.5553 d ¼ 8d0 ature are shown in Fig. 4. It should be noted that
10 1.771 1.803 1.831 d ¼ d0 the contour plots of temperature under MBCs and
1.786 1.816 d ¼ 2d0 PBCs are not exactly the same. The effective con-
1.794 1.809 d ¼ 4d0 ductivity and apparent properties under the mixed
1.799 1.804 d ¼ 8d0
condition are bounded from above and below by
100 1.993 2.044 2.095 d ¼ d0 the apparent conductivities obtained under EBCs
2.018 2.070 d ¼ 2d0 and NBCs, respectively, as is shown earlier in Eqs.
2.032 2.058 d ¼ 4d0
(6) and (7).
2.041 2.049 d ¼ 8d0
For the unit cell (2), shown in Fig. 2, the ap-
1000 2.019 2.074 2.127 d ¼ d0 parent properties under different boundary con-
2.046 2.100 d ¼ 2d0
ditions are close. The temperature contour plots
2.061 2.088 d ¼ 4d0
2.071 2.079 d ¼ 8d0 for d ¼ d0 are given in Fig 5. We can see that the
two plots in Fig. 5 are similar. However, for the

Fig. 3. Apparent conductivities as functions of mismatch and window size: (a) k2 =k1 ¼ 0:1; (b) k2 =k1 ¼ 10.
334 M. Jiang et al. / Computational Materials Science 25 (2002) 329–338

Fig. 4. Temperature contour plots under different boundary conditions, unit cell (1), d ¼ d0 , k2 =k1 ¼ 10: (a) PBC; (b) MBC; (c) EBC;
(d) NBC.

Fig. 5. Temperature contour plots under EBC and NBC, unit cell (2), d ¼ d0 , k2 =k1 ¼ 10: (a) EBC; (b) NBC.
M. Jiang et al. / Computational Materials Science 25 (2002) 329–338 335

Fig. 6. Temperature contour plots under EBC and NBC, unit cell (1), d ¼ 4d0 , k2 =k1 ¼ 10: (a) EBC; (b) NBC.

unit cell (1), the apparent conductivities obtained natural conditions is pretty uniformly distributed.
under EBCs and NBCs are much different from From Fig. 8, we see that, for a poorly conducting
the effective conductivity and they give very inclusion case, the concentration of U is on the
wide bounds. Their temperature contour plots pair of edges where uniform heat flux is ap-
given in Fig. 4(c) and (d) are quite different. We plied, while U is quite uniformly distributed under
further observe that for a poorly conducting EBCs, which is similar to those under MBCs and
inclusion case, jk E  k eff j < jk N  k eff j, while for PBCs.
a highly conducting inclusion case, jk N  k eff j < We have observed that choice of unit cells has
jk E  k eff j. a big influence on bounds. From Fig. 3, we see
From the Tables 1 and 2 and Fig. 3, we see that, the big difference between bounds obtained for
as window size increases, the difference between unit cells (1) and (2). It can be seen that for unit
apparent properties under different conditions be- cell (2), when d ¼ d0 , the bounds are already very
comes smaller and bounds become tighter. This close. For unit cell (1), on the other hand, even
convergence is a function of mismatch in con- when the window size is pretty big, the bounds are
ductivity constants: the larger the mismatch, the still very wide. The convergence of bounds for
slower the boundsÕ convergence. It is further ob- this case may require a very large window, which
served that for a highly conducting inclusion case, will exceed the capacity of computer simulation,
the effective conductivity is closer to the lower as illustrated in Fig. 3 for k2 =k1 ¼ 0:1 and k2 =k1 ¼
bound, while for a poorly conducting inclusion 10.
case, the effective conductivity is closer to the Concentration of U caused by location of in-
upper bound. The details of this phenomenon can clusions on boundary, when the conditions are
be seen in Figs. 6–8. Actually, the temperature essential and natural, gives us some explanation on
contour plot patterns under EBC and NBC are not the wide bounds obtained for random composite
very distinct as shown in Fig. 6. However, from the [10]. For random composite, when we cut a win-
contour plots of entropy rate, we can see that the dow from the whole material, the probability that
patterns are totally different under essential and there is at least one inclusion falling on the window
natural conditions. From Fig. 7, we observe that, edges is almost 1. Thus, the entropy (energy in
for a highly conducting inclusion case, there is U elasticity problem) concentration on edges, which
concentration on the pair of edges where linear is the reason for wide bounds, is inevitable. This
temperature constraint is applied, while U under also indicates that for random composite, the only
336 M. Jiang et al. / Computational Materials Science 25 (2002) 329–338

Fig. 8. Contour plot of U under MBC, EBC and NBC, unit cell
(1), d ¼ 4d0 , k2 =k1 ¼ 0:1. (The plots under MBC and PBC are
Fig. 7. Contour plot of U under MBC, EBC and NBC, unit cell almost identical, so only that of PBC is shown.) (a) PBC;
(1), d ¼ 4d0 , k2 =k1 ¼ 10. (The plots under MBC and PBC are (b) EBC; (c) NBC.
almost identical, so only that of PBC is shown.) (a) PBC;
(b) EBC; (c) NBC.

prehensive study on the apparent out-of-plane


way to obtain close bounds is to choose a large elasticity of random composite under different
window, whose size depends on mismatches and boundary conditions was done in [11] by using a
volume fraction of inclusions. Recently, a com- spring network method.
M. Jiang et al. / Computational Materials Science 25 (2002) 329–338 337

Table 3
Analogous quantities of problems controlled by Laplace equation
Physical subject T rT C q
Thermal conductivity Temperature Gradient Thermal conductivity Heat flux
Out-of-plane elasticity Displacement Strain Elastic moduli Stress
Torsion Stress function Strain Shear moduli Stress
Electrical conduction Potential Intensity Electrical conductivity Current density
Electrostatics Potential Intensity Permittivity Electric induction
Magnetostatics Potential Intensity Magnetic permeability Magnetic induction
Diffusion Concentration Gradient Diffusivity

4. Conclusions gence rate of bounds under essential and


natural boundary conditions could be different
(1) Windows of square packing composite yield quantitatively from that of 2D case, with the
isotropic apparent conductivity tensor in the same propertiesÕ mismatch between fibers and
form of k app I, approximately, where I is a matrix. This is an interesting topic for future
2 2 identity matrix. investigation. For 3D composites, the com-
(2) When mixed and periodic boundary condi- puter requirements to do the FEA simulation
tions are applied, unit cells (1) and (2) yield would be much higher.
the same apparent conductivity, within numer-
ical accuracy, which, in this case, is the effec-
tive conductivity. This important observation
provides us the way in numerical simulation Acknowledgements
or real experiment to obtain effective conduc-
tivity of periodic composite by applying the Support by the NSF under grants CMS-
special mixed boundary conditions introduced 9713764 and CMS-9753075, and by the Canada
in present paper. Research Chairs program, is gratefully acknowl-
(3) Effective conductivity is bounded from above edged.
and below by apparent conductivities obtained
under essential and natural conditions.
(4) The apparent conductivity under either essen-
Appendix A
tial or natural condition is very sensitive to
the unit cell choice.
The thermal conductivity problem is equivalent
(5) Bounds become tighter as window size in-
to a series of problems governed by the Laplace
creases and mismatch decreases.
equation. A list of analogous quantities is given in
(6) For highly conducting inclusion case, lower
Table 3 (e.g., [12]).
bound approaches to effective conductivity
faster; for poorly conducting inclusion case,
the opposite is true.
(7) Inclusions on edges are the major reason for References
wide bounds, which implies the need for very
big windows for the investigation of random [1] S. Hazanov, C. Huet, Order relationships for boundary
composites. conditions effect in heterogeneous bodies smaller than
(8) The investigation in the present paper can also representative volume, J. Mech. Phys. Solids 42 (1994)
1995–2011.
be generalized to 3D periodic composites, and [2] C. Huet, Application of variational concepts to size effects
we believe similar conclusions would be ob- in elastic heterogeneous bodies, J. Mech. Phys. Solids 38
tained. However, for the 3D case, the conver- (1990) 813–841.
338 M. Jiang et al. / Computational Materials Science 25 (2002) 329–338

[3] M. Ostoja-Starzewski, Random field models of heteroge- [8] M. Jiang, I. Jasiuk, M. Ostoja-Starzewski, Apparent elastic
neous materials, Int. J. Solids Struct. 35 (1998) 2429–2455. and elastoplastic behavior of periodic composites, Int. J.
[4] S. Hazanov, On apparent properties of nonlinear hetero- Solids Struct. 39 (2002) 199–212.
geneous bodies smaller than the representative volume, [9] S. Hazanov, M. Amieur, On overall properties of elastic
Acta Mech. 134 (1999) 123–134. heterogeneous bodies smaller than the representative
[5] M. Jiang, M. Ostoja-Starzewski, I. Jasiuk, Scale-dependent volume, Int. J. Engng. Sci. 23 (1995) 1289–1301.
bounds on effective elastoplastic response of random [10] M. Ostoja-Starzewski, J. Schulte, Bounding of effective
composites, J. Mech. Phys. Solids 49 (2001) 655–673. thermal conductivities of multiscale materials by essential
[6] S.J. Hollister, N. Kikuchi, A comparison of homogeniza- and natural boundary conditions, Phys. Rev. B 54 (1996)
tion and standard mechanics analyses for periodic porous 278–285.
composites, Comput. Mech. 10 (1992) 73–95. [11] M. Jiang, K. Alzebdeh, I. Jasiuk, M. Ostoja-Starzewski,
[7] S. Pecullan, L.V. Gibiansky, S. Torquato, Scale effects on Scale and boundary conditions effects in elastic properties
the elastic behavior of periodic and hierarchical two- of random composites, Acta Mech. 148 (2001) 63–78.
dimensional composites, J. Mech. Phys. Solids 47 (1999) [12] Z. Hashin, Analysis of composite materials––A survey,
1509–1542. J. Appl. Mech. 50 (1983) 481–505.

Potrebbero piacerti anche