Cornpods En&wing, Vol. 4, No. 1, pp. 374 1994. 0961-9526/94 $6.00+ .

oo
Printed in Great Britain. 0 1993 Pergattton Press Ltd

COMPOSITE HOMOGENIZATION VIA THE EQUIVALENT

POLY-INCLUSION APPROACH

MAURO FERRAIU
Department of Civil Engineering, and Department of Materials Science and Mineral Engineering,
University of California, Berkeley, CA 94720, U.S.A.

(Received 9 June 1993; final version accepted 15 Juiy 1993)

Abstract-For the efficient use and development of advanced materials it is necessary to develop
simple yet accurate analytic relationships linking their average physical response to as large a set
as possible of microstructural quantifiers. With this objective in mind, the theory of the effective
elastic response of biphase composite materials, with arbitrary second-phase geometry, concentra-
tion, and orientation distribution, is examined here. A formulation is presented, which employs
an equivalent poly-inclusion methodology, in the sense that its governing variable is the average
inclusion strain concentrator for an entirely uniform body undergoing a uniform eigenstrain in
a non-dilute family of morphologically identical internal regions (inclusions). Admissibility
criteria for approximate inclusion strain concentrators are formulated. The relation of the present
approach with the traditional one, based on the inhomogenity average strain concentrator, is
investigated, and some deficiencies of current methods based on the latter approach are discussed.
A family of fully admissible poly-inclusion concentrators is proposed, which differs by a single
scalar-valued function of the volume fraction-the so-called “interaction function”. A specific
interaction function is finally identified, which satisfies all admissibility criteria, is consistent with
the known analytic solutions for biphase composites, and reproduces literature data to within
experimental error.

1. INTRODUCTION
The “effective” or “equivalent” stiffness tensor C of a composite material is defined
through the relation
6 = C&O (1)
where so is a homogeneous deformation applied at the boundary and b is the resulting
average stress.
For a biphase composite, consisting of a matrix of stiffness Cm, containing a
volumetric concentration (Y of inhomogeneities, or fibers, of stiffness Cf, the effective
stiffness may be exactly expressed as
C = Cm + a((Cf - C*)A) (2)
where the angular brackets denote orientational averaging, weighted by a fiber orientation
probability density function (Ferrari and Johnson, 1989). In eqn (2), the (average) strain
concentration tensor A was introduced. This is defined by the equation:

where so is as in (l), a superscript f refers to the fiber phase, and overbars denote spatial
averaging.
Eshelby (1957) showed the strain concentrator for the limit case of a composite
consisting of a single anisotropic ellipsoidal inhomogeneity in an infinite matrix to be
T = [I + E(C*)-‘(C’ - Cm)]-’(4)
where I is the fourth rank symmetric identity tensor, and E is defined in eqn (8) below.
At non-dilute volumetric fiber concentrations, T is no longer an appropriate con-
centrator. The determination of the adequate concentrator A is essential, for the purpose
of homogenization, in view of the fact that, once the fiber orientation profile and the
tensor A are known, the homogenization problem is reduced to a mere computational
exercise, according to (2). However, this tensor is not generally available, in the case of
arbitrary fiber spacings and concentration levels. The need thus arises for introducing
estimates for A.
37
38 M. FERRARI

One such estimate, which has received considerable attention in recent literature, is
the method of Mori-Tanaka (Mori and Tanaka, 1973), which is based on the assumption
that
A = T[(l - ar)I + ar(T>]-‘. (5)
While the appropriateness of the Mori-Tanaka method, and all other approxi-
mations on A, must be assessed experimentally, on a case-to-case basis, some physical
considerations may guide the choice of the approximate expressions for the strain
concentrator A. In particular, the following are adopted here as necessary conditions for
the general validity of any homogenization scheme:
Requirement2: The effective stiffness C must be diagonally symmetric. This ensures
reversibility of any composite deformation in the linear elastic range.
Requirement2: In the power-series expansion of C in terms of (Y, the first-order term
must be ((C’ - Cm)T). Requirement 2 ensures the recovery of Eshelby’s exact solution at
dilute concentrations.
Requirement3: At the unitary fiber concentration limit, the effective stiffness
prediction must be independent of the matrix properties.
Requirement4: The effective moduli associated with C must comply with the
variational bounds. Examples of such bounds are those obtained by Hashin and
Shtrikman (1963) for macroscopically isotropic composites with isotropic phases.
The viewpoint adopted here is that, in order to be qualified as admissible, an effective
medium theory must satisfy Requirements l-4 for any biphusecomposite,i.e. for any
material symmetry, morphology, volume fraction, and orientation distribution of the
phases. As will be apparent in the following, the most demanding of these restrictions is
associated with the orientation distribution of the embedded phase-a requirement that
has not only academic, but also a considerable technological significance in the context
of Functionally Graded Materials (FGMs), especially for their application in thermo-
chemically aggressive environments (Nadeau and Ferrari, 1993).
Elementary as they may be, the above-stated requirements suffice to prove the
general inadmissibility of the Mori-Tanaka method. In particular, this complies with
Requirements 1 and 3 if and only if the inhomogeneities are perfectly aligned, or consist
of isotropic material (Ferrari, 1991). For completeness, it is recalled that Voigt’s (1928)
assumption A = I obviously violates Requirements 2 and 4, while Eshelby’s (1957)
assumption A = T violates Requirements 3 and 4.
In an effort to simplify the search for a fully admissible effective medium theory,
in the present paper the effective medium theory is reformulated in the context of an
equivalent eigenstraining poly-inclusion problem. In this formulation, the poly-inclusion
strain concentrator & defined in (9), replaces A as the independent variable of the
homogenizing scheme. This affords a considerably simplified analytical statement of the
physical admissibility requirements.
On this basis, a family of fully admissible homogenization approaches is presented
below, that differ by a single scalar-valued function of the second-phase concentration-
the so-called interaction function, Comparison with exact analytical results and
experimental data is employed to conclude the paper with the identification of a specific
interaction function.

2. THE EQUIVALENT EIGENSTRAINING POLY-INCLUSION PROBLEM

To solve the problem of the single ellipsoidal inhomogeneity in an infinite matrix

subject to the uniform boundary strain z”, Eshelby (1957) devised to substitute the
inohomogeneity with an equal volume hz of matrix material (the “inclusion”), subject
to a uniform eigenstrain E* of unknown amount. This unknown equivalent eigenstrain
was then calculated from the condition that the inhomogeneity and its equivalent
eigenstraining inclusion be equally stressed:
C’(&O + EC)= C”(&O + EC- &*) in R.
 Composite homogenization 39

Here, sCis the strain perturbation due to 0. This procedurethen reducedthe single
inhomogeneityproblem to the previouslysolvedproblem of a largehomogeneousbody
with uniform non-zeroeigenstrainingin Szonly.
Eshelby (1957)proceededto determine that the strain in the inhomogeneityis
uniform, and givenby
gf = Ef = Tg” (7)
in duality with the fact that the strainin theequivalenteigenstraininginclusionis uniform,
and givenby
E” = E” = Eg* + go. (8)
For the generalpoly-inhomogeneitycase,relation (7) must be approximatedby (3).
Similarly, the averageinclusion strains for the caseof a nondilute concentrationof
inclusions,undergoingthe sameuniform eigenstrainE* and in the absenceof mechanical
loads, is not given by (8). In this context, the approximateexpression
g” = fig* (9)
is now introduced,in analogywith (3). By parallelingEshelby’soriginal developments,
the poly-inclusion and the poly-inhomogeneityproblemsare put in correspondence by
imposingthat the averageinhomogeneitystressand the averageinclusionstresscoincide.
Consequently,the inhomogeneityaveragestrain in the non-dilutecaseis
8’ = *go (10)
with correspondingapproximateeffectivestiffness
c = cm + a((Cf - Crn)f> (11)
wheref is obtainedfrom T by replacingE with a in (4).

3. ADMISSIBILITY CRITERIA-A FAMILY OF ADMISSIBLE CONCENTRATIONS

In the following, the poly-inclusionstrainconcentrator& ratherthanA, is employed

as the governingquantity of the homogenizationscheme.The relation betweenthese
tensorsis
E = (A-’ - I)(Cf - Cm)-‘Cm (12)
as is found by equating(2) and (11).
To analyticallyexpressthe symmetrycondition, Requirement1 above,it is first noted
that, by the definition of T,
c = cm + cY((Cf- Cm)@
= cm + a((C’ - Crn)[I + e(cm)-‘(cf - cm)]-‘>
= cm + a({[1 + 12(cm)-‘(cf - C”)](Cf - cm)-‘]-‘>
= cm + a<((C’ - cmyl + fi(cm)-l)-l>. (13)
Thus, it is concludedthat C is symmetricif and only if @(Cm)-’is.
Requirement2 is equivalentto imposing that the limit of 6 asQ approacheszerobe
Eshelby’stensorE. Requirement3 may be equivalentlyexpressedby the condition that
c, = cm + <[(Cf - Cm)-’ + R’(Cm)-y) (14)
be matrix-independentfor all composites.Here, C1and & denotethe unitary concentra-
tion limits of C and a, respectively.For a void secondphase,(14)reducesto
c, = C”(I + (fi’ - I)-‘) = Cm(ill(& - I)-‘). (15)
In order for Ci to vanish,regardlessof the matrix properties,it is thus necessarythat
<i7’(til - I)-‘) = 0. (16)
40 M. FERRARI

Since no quantity in (16) depends on Cf, the holding of (16) for all composites is
equivalent to Requirement 3. Further specialization of (16) is possible, considering the
special case of perfect second-phase alignment. In this context, the orientational
averagings may be deleted, and (16) is equivalent to
fil = 0. (17)
For texture-independent a., Requirement 3 is thus equivalent to imposing (17) for all
composites.
The family of approximate poly-inclusion strain concentrators
ti = f(a)E + g(a)1 (18)
is now introduced. Here f( *) and g( *) are arbitrary functions of the fiber concentration,
and are referred to as “interaction functions”. The effective stiffness tensor, associated
with the general form (18) via eqn (1 l), may be expressed as
C = C” + cx<[(Cf - Cm)-’ + f(a)E(Cm)-’ + g(a)(C”)-‘I-‘). (19)
All d of the form (18) are in compliance with Requirement 1, since the contraction
E(C”)-‘, also known as the “polarization tensor”, is always symmetric (Walpole, 1981).
The limit conditions, Requirements 2 and 3-eqn (17)-are satisfied if and only if
lim f(o) = 1, lim f(o) = 0, lim g(o) = 0, lim g(o) = 0. (20)
IX-0 a--r1 Or-0 a-1

The simplest concentrator that complies with Requirements l-3 is thus

ti = (1 - CY)E (21)
which is fully acceptable for biphase composites with any second-phase geometry,
anisotropy, concentration, and orientation distribution, if the associated moduli are in
compliance with the appropriate bounds.
The family (18) comprises the classical homogenization schemes. In particular, it may
be concluded from (12) that the special casef = g = 0 corresponds to Voigt’s approach,
while f = 1, g = 0 corresponds to Eshelby’s. Voigt’s method violates (20), , and thus
Requirement 2, while Eshelby’s violates (17), and thus Requirement 3.
Given (5) and (12), the assumption of Mori-Tanaka may be expressed as
d = tiMT I (1 - a)E + a((T)T-’ - I)(Cf - Cm)-‘Cm (22)
after some tensorial algebra. By enforcing eqn (17) on (22), it is concluded that the
term (T)T-’ - I must identically vanish, in order to avoid unphysical predictions at the
unitary second-phase concentration limit. Since, in general, T differs from (T), it is
concluded that the scheme of Mori-Tanaka is not acceptable for composites with
arbitrary constitution.
Under special circumstances, the Mori-Tanaka scheme does yield a matrix-
independent effective stiffness at the unitary concentration limit. In particular, this
property is satisfied for biphase composites with isotropic phases, and for isotropic-
matrix biphase composites with perfectly aligned inhomogeneities (Ferrari, 1991). For the
latter composite type, an immediate proof is obtained by noting that, in conditions of
perfect alignment, the orientation averagings may be dropped, and (22) reduces to (21),
which ensures acceptability at the unitary concentration limit.
In the form (21), the Mori-Tanaka assumption may be interpreted as the o-linear
interpolation that satisfies the dilute (fi = E) and unitary concentration (fi = 0) limits for
the eigenstraining poly-inclusion problem under fixed boundary conditions. When not
reducible to the form (21), the Mori-Tanaka theory is not applicable, as shown above.
Returning to the general case, it is now recalled that any stiffness of the form (19)
is admissible, provided (20) is satisfied and the relative moduli comply with the
appropriate variational bounds. To clarify the associated procedure, a special case of the
theory is considered next.
 Compositehomogenization 41
4. A CASE STUDY

For a void second phase, eqn (19) specializes to

C = Cm(I + (Y([~(oL)E + (g(cr) - l)I]-‘)). (23)
For isotropically distributed cavities in an isotropic matrix, the effective elastic response
is isotropic. In terms of the matrix properties and the void concentration, the optimal
bounds on the effective bulk and shear moduli k and ,u for this case are those obtained by
Hashin and Shtrikman (1963):
o ( k < 4(1 - 4 o < 4f_ < (1 - 4(8 + 9x)
- km - 4+3ax’ -pm-8+9x+6a(x+2)’ (24)

Here, x denotes the matrix bulk-to-shear modulus ratio.

The normalized effective moduli corresponding to (23) are
k (a + /2)(3x + 4) + 3fx 4f_ = 5(a + /1)(3x + 4) + 61(x + 2)
, (25)
k,= Fl Pm F2
with the definitions
h=g-1; Fl =h(3x+4)+3fx; F2 = 5/2(3x + 4) + 6f(x + 2) (26)
for the case of spherical voids. Equations (25)-(26) are conveniently obtained from (23)
by the symbolic computing methods introduced in Ferrari and Marzari (1992).
Combining (24)-(26), and simplifying sign definite terms, the bounds onf( .) and h( +)
coresponding to the upper Hashin-Shtrikman bounds may be expressed as
k I-CY & + (1 - a8 + 9x) < o
- s 0,
F+m Fl Pm F2 -’
In view of the positivity of the normalized effective moduli and of the numerators in (27),
it follows that Fl and F2 are negative quantities. With this, the bounds may be put in
the form
3x(1 - 01) I (3x + 4)g + 3fx I (1 - #X)(3x + 4) (28)
6(1 - ar)(x + 2) I 5(3x + 4)g + 6f(x + 2) I 5(1 - (r)(3x + 4). (29)
Elementary considerations show that, for
(3x - 4)f(a) s 0 (30)
the bound (28), is redundant. Otherwise, (29), is. Similarly, (28), is redundant for
(3x - 4)g(cY) s 0 (31)
while otherwise (29), is. Thus, under the assumption that both f(q) and g(a) be positive
definite, the bounds reduce to (28) for x 1 4/3, and to (29) otherwise.
Further, assuming that g(ar) be identically zero-as done in the Voigt, Eshelby and
MO&Tanaka theories-the bounds reduce to
(1 - (Y) I&) s (1 - @J(x) (32)
where
3x + 4
w -
-- 3x x 2 4/3
(33)
5(3x + 4)
U(x) = x s 4/3.
6(x + 2)
The region of admissibility corresponding to (32) is shown in Fig. 1. There, the gap
between the upper and the lower bounds is seen to be a decreasing function of OL,for a
given material, and to be maximal at the known dilute limit.
Figure 2 shows that the maximum bound gap at fixed concentration peaks for
x = 413, and then asymptotically tends to zero for increasing values of x.
42 M. FERRAIU

Fig. 1. Domain of admissibility of the functionf for g = 0. f(e) and g(e) ar defined in eqn (18).
f= 5(3x + 4)/6(x + 2) for x s 4/3, f= (3x + 4)/3x for x 2 4/3.

Fig. 2. Maximum gap between bounds as function of x, x = K,,,/p,.

From (32) it follows that Voigt’s assumption f(s) = 0 is not admissible for any
concentration levels, while the dilute approximationf(.) E 1 does not violate the bounds
if and only if
9x + 8
x I 4/3
cYs 5(3x + 4)
(34)
4
a=3x+4 x 2 4/3.

To visualize: for x = 3, representing a high-modulus ceramic, (34) shows that

Eshelby’s homogenization scheme violates the variational bounds for 01larger than 31%.
The Mori-Tanaka approach, in the context of the present case study, corresponds to
f(o) = 1 - CL Thus, the bound (32), is identically satisfied at all volume fractions and for
all materials, and the predictions actually coincide with the bound (32), .

5. ON THE INTERACTION FUNCTION

In order to progress in the search for appropriate interaction functions, recourse is
here made to comparisons with (i) known exact results, and (ii) published experimental
data. In the following, it is assumed that g(ar) identically vanish.

Comparison with exact results

First, the case of a composite with isotropic spherical inhomogeneities in an isotropic
matrix is considered, under the further assumption that the shear moduli of the two phases
coincide. For this assemblage, the effective bulk modulus is known exactly, as the relative
Hashin-Shtrikman bounds coincide. This exact modulus corresponds to what is obtained
from (13) via assumption (21)-which was to be expected, in view of the fact that for this
composite type assumption (21) reproduces the Mori-Tanaka method.
In total analogy with this first case, it is further noted that eqns (13) and (21) yields this
exact bulk modulus for a composite with isotropic randomly oriented particles reinforcing
a softer, isotropic matrix of equal shear modulus.
 Composite homogenization 43

Finally, the case of isotropic spherical inhomogeneities reinforcing a softer isotropic

matrix (different shear moduli) is considered. For such assemblies, eqns (13) and (21)
yield the lower Hashin-Shtrickman bound on the effective bulk modulus:
k = k” + a[(k’ - km)-’ + 3(1 - ar)(3krn + 4pm)-l]-‘. (35)
Explicit computation shows that the upper bound corresponds to
f(a) = (1 - cr)[(3P + 4prn)/(3k” + 4p’)] = (1 - ar)y (36)
so that the full bounding for f( ) is
1 - a! 5 f(CY) I (1 - ol)?. (37)
In analogy with the case of voids, the imposing of the shear moduli restrictions does
not further limit the range of f( ).

Comparison with experimental data

Ferrari et al. (1993) employed the measured values of the Young’s moduli of a glass
matrix with alumina (Hasselman and Fulrath, 1965a) and tungsten (Hasselman and
Fulrath, 1965b) reinforcements of approximately spherical shape as data from which to
compute via (13) the values of the interaction function at different second-phase concen-
trations. The results are shown in Figs 3 and 4, and are in excellent agreement with (21).

l.OF
\* Experimenlaipointe l
\
\
\ Lowerli-Sbound _---
0.6 - ‘\ f(a) - 1 + a
\
0 ‘. UpperH-sbound
\ f(a)=cfmetx(l-a)““”
\
0.6 - \
l ‘, \
f(a)

.A,
0.4 -
‘. \
. .. \
*... \
\
-*.. \
. ..* \
0.2 - \
--.*
.... ’ \
-**. \
‘-..\
I I I , . .y
0.0 0.2 0.4 0.6 0.6 1.0
Fiber volume fraction (a)
Fig. 3. Interaction function versus volume fraction (from Hasselman and Fulrath, 196513).

,a-

\ LowerH-Sbound ____
0.6 - ’ f(a) - 1 + a
A\
U~~erH-Sbovt-mi ,.....
0’ \ f(a)=canstx(l -a)
\
0.6 - \

f(a) 0 ‘\
\
*a ‘\
0.4 - -\\
“Y- \
-... \
\
. . .. \
-... \
0.2 - \
. .. . \
. . .. \
.... \
. . .. \
I I I
0.0 0.2 0.4 0.6 0.6 1.0
Fiber volume fracfion (a)
Fig. 4. Interaction function versus volume fraction (from Hasselman and Fulrath, 1965a).
 M. FERFARI
1.2 .

r
1.0X-
\
\ mrl-pdntl l
\
\
\ Lcwwf+-sbcund ----
0.6 - \‘\ f(a) - 1 + a
\
\
\
\ UwH-S- .....
\\ f(a)=amIx(l -aI
fta) 0.6 L \
\
\
*. \
**. *.. .
-. a. .‘\
0.4 -
OA- *-....**.
0.
*. L
*.
..
*. ‘,*
*. \
0.2- *. \
-- -. \
**._ ‘.
..
*.I,
I I I
0.0 02 0.4 0.6 0.6 1.0
Fiber volume fraction (a)
Fig. 5. Interaction function versus volume fraction (from Nishimatsa and Gurland, 1960).

6~.~
Fig. 6. C,, versus volume fraction (from Dean and Turner, 1973).

A similar procedure was employed ibidem to determine the interaction function for
a cobalt/tungsten carbide composite, with experimental data obtained from Nishimatsu
and Gurland (1960). The results of the comparison are given in Fig. 5, and again are in
agreement with (21). Finally, eqn (13) sub assumption (21) was directly employed to
compute the transverse Young’s moduli of two different unidirectional carbon-epoxy
composites. The computed values are shown in Fig. 6 to coincide within experimental
error with the measurements of Dean and Turner (1973).

6. DISCUSSION AND CONCLUSIONS

In Section 2, the theory of homogenization was reformulated in terms of the

poly-inclusion strain concentrator b. The admissibility criteria, postulated in the intro-
duction, were then re-established in this context, and the family (18) of poly-inclusion
strain concentrators was introduced. Any members of this family, for which restrictions
(20) hold, is fully admissible, once the appropriate variational bounds on the associated
moduli (19) are verified.
 Composite homogenization 45

The homogenization approach based on (18) with g( *) = 0 permits an efficient use of

the variational bounds, since the bounds on each effective modulus result in a possibly
non-redundant restriction on the single functionf(*). Expressing the effective moduli in
terms of a single admissible f( .) also simplifies the experimental determination of the full
elastic response, asf(.) may be obtained from easily realizable tensile tests, and then used
to predict all other moduli.
Progress towards the determination of the interaction function was sought in the
above by comparison with known exact results and experimental data available in the
literature. All indications from these comparisons concur in identifying the interaction
functions defined in (18) to be
f(a) = 1 - CX, gw = 0 (38)
corresponding to the homogenization method embodied in (13) under assumption (21):
C = Cm + cr([C’ - Cm)-’ + (1 - a)E(C”)-‘I-‘), (39)
which is valid for arbitrary second-phase concentration, material symmetry, morphology,
and orientation distribution.

Acknowledgements-The author thankfully acknowledges support from the National Science Foundation,
Award # ECS 9023714.

REFERENCES
Dean, G. D. and Turner, P. (1973). The elastic properties of carbon fibers and their composites. Composites,
July 1973, 174-180.
Eshelby, J. D. (1957). The determination of the elastic field of an ellipsoidal inclusion, and related problems.
Proc. R. Sot. A 241,276-303.
Ferrari, M. (1991). Asymmetry and the high concentration limit of the Mori-Tanaka effective medium theory.
Mech. Muter. 11, 251-256.
Ferrari, M., Imam. A. and Kuong, L. (1993). Recent developments in homogenization theory: on the interaction
function of the equivalent poly-inclusion formalism. In Composite Materials Technology 1993 (Edited by
D. Hui, 0. Ochoa and T. J. Kozik), Vol. PD-53, pp. 237-241. ASME, New York.
Ferrari, M. and Johnson, G. C. (1989). The effective elasticities of short-fiber composites with arbitrary
orientation distribution. Mech. Muter. 8, 67-73.
Ferrari, M. and Marzari, N. (1992). A Mori-Tanaka theory for short-fiber composites: application. J. Energy
Res. Tech. 144, 101-104.
Hashin, Z. and Shtrikman, S. (1963). A variational approach to the elastic behavior of multiphase materials. J.
Mech. Phys. Solids 11, 127-141.
Hasselman, D. P. H. and Fulrath, R. M. (1%5a). Effect of alumina dispersions on Young’s modulus of a glass.
J. Am. Ceram. Sot. 48(4), 218-219.
Hasselman, D. P. H. and Fulrath, R. M. (1965b). Effect of spherical tungsten dispersion on Young’s modulus
of a glass. J. Am. Ceram. Sot. 18(10), 548-549.
Mori, T. and Tanaka, K. (1973). Average stress in matrix and average elastic energy of materials with misfitting
inclusions. Actu Metall. 21, 571-574.
Nadeau, J. and Ferrari, M. (1993). (In preparation.)
Nishimatsu, C. and Gurland, J. (1960). Experimental survey of the deformation of the hard-ductile two-phase
alloy system WC-Co. Trans. ASM 52, 469-484.
Voigt, W. (1928). Lehrbuch der Krystaliphysik. Teubner, Leipzig.
Walpole, L. J. (1981). Elastic properties of composite materials: theoretical foundations. Adv. Appl. Mech. 21,
169-186.