# improved rigorous bounds on the effective elastic...

Post on 16-Oct-2020

1 views

Embed Size (px)

TRANSCRIPT

.J. Mech. Phys. Solids Vol. 32, No. 1, PP. 41-62, 1984. Printed in Great Britain.

0022-5096/84 $3.00 + 0.00 0 1984 Pergamon Press Ltd.

IMPROVED RIGOROUS BOUNDS ON THE EFFECTIVE ELASTIC MODULI OF A COMPOSITE MATERIAL

Y. KANTOR and D. J. BERGMAN

Dept. of Physics and Astronomy, Tel Aviv University, 69978 Tel Aviv, Israel

(Received 13 April 1983)

ABSTRACI

A NEW METHOD for deriving rigorous bounds on the effective elastic constants of a composite material is presented and used to derive a number of known as well as some new bounds. The new approach is based on a presentation of those constants as a sum of simple poles. The locations and strengths of the poles are treated as variational parameters, while different kinds of available information are translated into constraints on these parameters. Our new results include an extension of the range of validity of the Hashin-Shtrikman bounds to the case of composites made of isotropic materials but with an arbitrary microgeometry. We also use information on the effective elastic constants of one composite in order to obtain improved bounds on the effective elastic constants of another composite with the same or a similar microgeometry.

1. INTRODUCTION

THEORIES of the effective elastic properties of composite materials (i.e. macroscopically inhomogeneous systems made of regions or grains of different homogeneous sub- stances) have many practical applications in technology and geophysics. Various approaches to the problem were reviewed by HASHIN (1970), WATT et al. (1976) (this review contains an exhaustive list of references), HALE (1976), and WILLIS (1982).

In principle, exact values for the effective elastic constants can be obtained only when the microgeometry of the composite is known precisely. In many cases, however, the precise microgeometry is unknown, e.g. when the composite has a certain randomness in its microstructure. In that case, exact theories are limited to the derivation of rigorous bounds on the elastic constants. These range from the simplest bounds of Voigt and Reuss (HILL, 1952), for which only the volume fractions of the components need to be known, through the more complicated Hashin-Shtrikman bounds and their extensions (HASHIN and SHTRIKMAN, 1963 ; HILL, 1963 ; WALPOLE, 1966a,b) for isotropic mixtures, and up to some very complicated bounds which require knowledge of the two- and three-point correlation functions (see, e.g. BERAN and MOLYNEUX, 1966; MILLER, 1969; MCCOY, 1970). Recently some of these bounds (the Hashin-Shtrikman and some of the higher order bounds) were modified and improved by MILTON and PHAN-THIEN (1982). The derivation of bounds is usually based on variational principles (e.g. HILL, 1952; HASHIN and SHTRIKMAN, 1961, 1962; BERAN and MOLYNEUX, 1966). Sometimes these principles are used in the context of a scattering-theory-like approach (e.g. DEDERICHS and ZELLER, 1973 ; WILLIS, 1982).

41

42 Y. KANTOK and D. J. BERGMAN

Recently we introduced a new approach to the problem of evaluation of the effective elastic constants of composite materials (KANTOR and BERGMAN, (1982a)-this will be referred to as I). We showed that any effective elastic constant of a composite can be written as a sum of simple poles, and reduced the problem to that of a systematic evaluation of the locations and weights of these poles. The usefulness of this approach was demonstrated on several well defined microgeometries (see I, and also KANTOR and BERGMAN (1982b)). In this article we will apply this pole representation to the case when only partial information is available on the microgeometry of the system. We will treat the locations and the weights of the poles as unknown parameters, and we will derive rigorous upper and lower bounds on the effective elastic constants by varying them subject to certain constraints imposed by the known information. A similar method was developed and applied by BERGMAN (1978a, b, 1982), for the derivation of bounds on the effective dielectric constants of composites.

In Section 2, we rederive the general theory of I for the elastic properties of composites in a simplified form, and thereby also introduce the main concepts to be used later. In Section 3 we use our formalism to obtain some of the simple known bounds, and we also extend the range of validity of the Hashin-Shtrikman bounds to include composites with an arbitrary (i.e. not necessarily isotropic or cubic) micro- geometry and without any information about the microgeometric correlation func- tions. A more general and also more flexible formalism is introduced in Section 4. This is applied in Section 5 to derive improved bounds that require information of a new type about the composite. This information is in the form of known values of the effective elastic constants for a composite with the same microgeometry but different constituents. Such information can be obtained either by measurement or by another calculation. It is clearly information of a “physical” nature about the composite, as opposed to the purely geometrical information that is contained in the correlation functions which are also sometimes used to obtain improved bounds.

2. THE GENERAL THEORY

The position dependent local elastic stiffness tensor C(r) of a two-component composite made of homogeneous materials with stiffness tensors C(” and C”’ can be written with the help of a step function 0, :

C(r) = P’ + d,(r) (P- IF) 3 cC2) + 8,6C, (2.1)

where

O,(r) = i

1, r inside C(i) material,

0, r outside C(l) material. (2.2)

In (2.1) and often also in subsequent discussions, we have suppressed the tensorial indices.

The effective elastic stiffness tensor C@) is usually defined by means of the volume averages of the stress tensor g and the strain tensor E in the inhomogeneous sample, so that

(0) a” = C’“‘(c) a”) (2.3)

Effective elastic moduli of a composite material 43

where ( ),, denotes a volume average. Alternatively, C@) may be defined by requiring that the elastic energy density that would exist in a homogeneous material with stiffness tensor C@) be equal to the volume averaged energy density in the actual inhomo- geneous sample when it is subjected to the same boundary conditions on the displacement vector u. We will use the boundary conditions (see HASHIN, 1970, pp. 44

47)

ui = ,$xj, for r = (x1,x2,x& on the boundary, (2.4)

where .st is some constant symmetric tensor (i.e. a0 has the same value over the entire boundary). Here and subsequently we use the Einstein summation convention on repeated tensorial indices. These boundary conditions would cause the strain E to be a uniform constant E = so in the entire volume of a homogeneous material, while in the case of an inhomogeneous material only the volume average of the position dependent z(r) would be equal to so. Thus the alternative definition of C”’ under the boundary conditions of (2.4) is

&°C(e)ao = (&C(T)&),, (2.5)

for any constant symmetric tensor E ‘. For these boundary conditions, it can be shown that the two definitions of C’“‘, namely, (2.3) and (2.5) coincide.

We now introduce a somewhat generalized form of (2.1), allowing C to depend on a continuous parameter s :

C(r;s) = Cc2’+ A131(r)X. s

(2.6)

By allowing s to take arbitrary values, we are actually replacing the C(i) material by a different material C(l)‘, where

C”” = C(Z)+ 16C = I@‘+ S-l _CV) (2.7) S S s .

This replacement also makes C@) a function of s. We note that when s lies in certain ranges, the tensor C (I)’ becomes unphysical, i.e. it ceases to be positive definite. From (2.5) and (2.6) we can obtain the expression for C(‘) as (cf. I)

s°C%o -E~C’~‘E~ = -& s

8,s”6Cs dV = F(s). (2.8)

In order to simplify the notation, we now introduce two definitions : For any tensor E we define a complementary tensor E;

E” = (&X)*, (2.9)

where the asterisk denotes complex conjugation, and we also define a scalar product between two arbitrary tensors

(E I E’) = s

Ol(r)$@)&(r) dV. (2.10)

44 Y. KANT~K and D. J. BERGMAN

We can now rewrite (2.8) in the form

F(s) = $ (a0 1 e). (2.11) The use of non-real (i.e. complex) tensors E, b is mandatory only ifs or C are complex. Otherwise we can always restrict ourselves to real F,E. However, even then it is sometimes convenient to allow complex 8, E:

The strain tensor c(r) in a composite material, the boundaries of which undergo the displacement (2.4), is the solution of the linear integral equation (see, e.g. Wu and MCCULLOUGH (1977))

Eij(r) = C: + f s

0 1 (r’)Gijk.(r, r’ ; C(2))‘6CklmnEmn(r’) d V’.

Here G is the tensor Green’s function of the problem, which depends on C(” as well as on the shape of the sample, and has the symmetries

Gijkl(r, r’) = Gijlk(r, r’) = Gjikl(r, r’) = Gklij(r’, r). (2.13)

The integral equation (2.12) can be written in a more concise bra- and -ket notation

Ic) = ,cO)+ fc,,:). (2.14)

We can formally solve this equation, and substitute that solution in (2.11) to obtain

(2.15)

In order to make further progress, we introduce the eigenstates of the operator (? :

Q&(Q)) = S,(&(“)). (2.16)

Since G is a non-hermitian operator, its right eigenstates differ from its left eigenstates. However, from (2.13) and the definition of G we can easily see, that