Poisson's ratio of ber-reinforced composites

Theoretical Physics, Royal Institute of Technology, S-100 44 Stockholm, Sweden Department of Mathematical Sciences, Rensselaer Polytechnic Institute, Troy, New York 12180-3590

Henrik Christiansson

Johan Helsing

Abstract
Poisson's ratio ow diagrams, that is, the Poisson's ratio versus the ber fraction, are obtained numerically for hexagonal arrays of elastic circular bers in an elastic matrix. High numerical accuracy is achieved through the use of an interface integral equation method. Questions concerning xed point theorems and the validity of existing asymptotic relations are investigated and partially resolved. Our ndings for the transverse e ective Poisson's ratio, together with earlier results for random systems by other authors, make it possible to formulate a general statement for Poisson's ratio ow diagrams: For composites with circular bers and where the phase Poisson's ratios are equal to 1/3, the system with the lowest sti ness ratio has the highest Poisson's ratio. For other choices of the elastic moduli for the phases, no simple statement can be made.

I. INTRODUCTION
Several authors have recently devoted their attention to Poisson's ratio ow diagrams, that is, the e ective Poisson's ratio as a function of the inclusion volume fraction, for composite materials. 1{14]. Some authors consider explicitly discrete materials such as discrete fractals 1] and lattice networks of springs 2{5], but most work has been for continua. The Poisson's ratio for two-dimensional elastic composites has been studied analytically in the framework of e ective medium approximations 6,16], and numerically with methods such as nite element methods 13] and boundary element methods 11], and with discretized-spring schemes applied to digital images of materials 8{10,12,13,15]. The studies undertaken with the digital image based method of the Poisson's ratio ow diagrams have been for isotropic, random or periodic structures of circular voids 9] and circular rigid inclusions 13]. Computations have also been done for composites with random structure, where the phases have nite elastic moduli 10,14]. The error in the above computations typically is of the order of three per cent. For systems with strong inhomogeneities and smaller inter-inclusion distances, the error is larger. 1

Two-dimensional e ective medium approximations for random circular inclusions predict so-called xed points for the e ective Poisson's ratio of two-phase composites in di erent ways. One xed point relates to composites where both phases have the same Poisson's ratio. If the phase Poisson's ratio is equal to 1/3, then the e ective Poisson's ratio should also be 1/3 for any matrix Young's modulus and any volume fraction of the inclusions 6,10] and this xed point will be a straight line in the Poisson's ratio ow diagram. Numerical calculations with the digital image based method on random ber-reinforced composites 14] suggest that this prediction may be a good approximation for a wide class of composites. Another xed point for the e ective Poisson's ratio relates to composites where one phase consists of either circular voids or circular rigid inclusions. For such materials, when the inclusions form a connected path (percolation), the e ective medium approximation predicts that the e ective Poisson's ratio is independent of both the matrix Young's modulus and its Poisson's ratio. The existence of this type of Poisson's ratio xed point for percolating voids has been proved analytically 8,17] and demonstrated numerically 8,9], for any suspension of voids. The precise value of the xed point, however, depends on the geometrical con guration 8]. For example, the xed point for a hexagonal array of circular voids is 1. Recent numerical calculations 13] suggest that systems with rigid inclusions do not exhibit a xed point of this type. Asymptotic relations for the e ective Poisson's ratio have been derived for periodic systems of circular voids 9] or circular rigid bers 13]. Those asymptotic relations are in good agreement with numerical results 9,13]. Recently, one of us 24] presented a numerical algorithm for solving the elastostatic equation of a two-dimensional, locally isotropic, two-component composite. For a hexagonal array of circular bers the implementation of this algorithm is particularly simple. Machine accurate solutions can be computed in a few seconds for both high degree of inhomogeneity and small inclusion separation distances. The purpose of this paper is to test the validity of xed point theorems and asymptotic expressions for the e ective Poisson's ratio when the bers are near touching. This is accomplished by using the above mentioned algorithm to calculate detailed Poisson's ratio ow diagrams from near homogeneity to the regime where asymptotic expressions become valid, for hexagonal arrays of elastic circular bers. The paper is organized as follows. Section II summarizes some results from linear elasticity and gives relations between di erent moduli. Section III describes our numerical algorithm, which we refer to as the interface integral equation method. Section IV presents Poisson's ratio ow diagrams and compares those to earlier results and predictions 9,10,13,14].

II. LINEAR ELASTICITY

The stress-strain relations for a linear elastic and isotropic material can be written as = 1 (1 + (d)) ? (d) ]; (1)
ij ij kk ij

E (d)

ij

=(

(d)

2 ?d

(d)

) kk ij + 2 2

(d)

ij :

(2)

Here ij and ij are the strain and stress tensors and d is the dimension, which can be two or three. Eq. (1) de nes the Young's modulus E (d) and Poisson's ratio (d). Eq. (2) de nes the bulk modulus (d) and the shear modulus (d). Two-dimensional elasticity can be considered as an approximation to either of two important special cases in three-dimensional elasticity: plane-stress and plane-strain elasticity. When a material is in the form of a thin sheet and subjected to planar loads, stress across its thickness can often be neglected. The resulting, two-dimensional, elastostatic problem is called plane-stress. When a very thick material that is translationally invariant in the direction of its thickness is subjected to planar loads, the strain along the thickness can often be neglected. The resulting, twodimensional, elastostatic problem is called plane-strain. Plane-strain elasticity is physically relevant when considering ber-reinforced materials subjected to loads across the bers. The mappings from three-dimensional elastic moduli to the relevant two-dimensional elastic moduli can be retrieved from Eqs. (1,2), using the assumptions in the two special cases. The results are summarized in Table 1 12]. In the following we will consider two-dimensional elasticity for ber-reinforced composites. The e ective elastic two-dimensional moduli will be denoted by e , e , e and Ee .

III. THE INTERFACE INTEGRAL EQUATION METHOD

Integral equation methods are by now standard for solving simple boundary value problems in locally isotropic elasticity. See Chen and Zhou 18] for a theoretical discussions and a literature survey. Reasons why integral equation methods are often preferred over other numerical schemes are related to storage requirements, accuracy, and discretization. A numerical example of particular interest is the paper of Greenbaum, Greengard, and Mayo 19]. The authors solve the elastostatic equation rapidly and to high accuracy using iterative techniques and a biharmonic extension of the Fast Multipole Method 20{22]. A recent numerical example involving inclusions is the work on Ostwald ripening by Jou, Leo, and Lowengrub 23]. For some reason, integral equation methods have not been used much for solving homogenization problems in locally isotropic elasticity. We know of only two papers in the literature. The rst papers is by Eischen and Torquato 11]. The authors use a standard boundary element method for nite body problems, match boundary conditions at interfaces, and compute three-digit accurate e ective moduli for some hexagonal arrays of bers. The ratio of ber separation distance to ber radius is larger than 0.06. The other paper is by one of us 24]. Here an integral equation for a periodic composite is explicitly written down and is solved with Fourier methods. A simple formula for extracting e ective properties from the solution is derived. This leads to a fast and accurate algorithm which we refer to as the interface integral equation method. For moderate inhomogeneity we can compute e ective moduli for arbitrarily dense arrays in a few seconds on a simple workstation. The two main reasons that the interface integral equation method is orders of magnitudes better than the boundary element method in this case are 1) We only discretize the interfaces, not the boundary of the unit cell. The periodic boundary conditions are treated with lattice sums 25]. 2) We use a quadrature rule with superalgebraic convergence, not low-order accurate quadrature. 3

IV. RESULTS AND DISCUSSION

We have computed the e ective Poisson's ratio e for hexagonal arrays of bers in a matrix. Most calcualtions were for systems where the Poisson's ratios of the matrix 1 and of the bers 2 were equal and the Young's moduli of the matrix E1 and bers E2 were di erent. It follows immediately from the integral equation 24], that the e ective Poisson's ratio depends only on the sti ness ratio E2=E1 and the Poisson's ratios of the phases. The resulting Poisson's ratio ow diagrams are shown in Figures 1-4 with E2=E1 and 1 = 2 as variable parameters. We have considered non-overlapping bers in hexagonal arrays which means that the calculations of Poisson's ratio do not extend beyond the percolation threshold p p f2 = =2 3. First we calculated the e ective Poisson's ratio for systems with 1 = 2 = 0:5 and sti ness ratios (E2=E1) = 2000 and (E2=E1) = 0:0005 which is approximately equivalent to rigid inclusions and voids, respectively. Those structures have been studied before with other methods 9,13]. E ective Poisson's ratio ow diagrams for those systems are shown together with ow diagrams for other sti ness ratios in Figure 1. Our results are in agreement with existing graphical data for voids 9] (open circles) and rigid inclusions 13] ( lled circles). It has been suggested 14] that for random suspensions of inclusions and 1 = 2 = 1=3, the e ective Poisson's ratio is e = 1=3 for all ber fractions of inclusions f2 and all sti ness ratios E2=E1. Figure 2 illustrates that this does not hold for hexagonal arrays of bers. Figure 3 shows Poisson's ratio ow diagrams for 1 = 2 = 0. Previous calculations on periodic structures with the digital image based method 9,13] have only been possible with 1=3. The graphs with (E2=E1) = 2000 and (E2=E1) = 500 are almost on top of 1 = 2 each other. From gure 3 it is obvious that even when the Poisson's ratio is zero, a large range of the e ective Poisson's ratios can be obtained. Figure 4 shows the Poisson's ratio ow diagrams for sti ness ratio (E2=E1) = 100 and with 1 = 2 varying from -0.5 to 0.5 with the step 0.1. This could be compared to the result from e ective medium approximations 10] and numerical calculations 10,14] on random suspension of bers. The conclusion in those studies was stated as follows: when 1 = 2 > 1=3, the e ective Poisson's ratio always decreases and is bounded below by 1/3 when two phases are mixed. If 1 = 2 < 1=3 the value of e always increases and is bounded above by 1/3 when two phases are mixed. Figure 4 shows that for hexagonal arrays of bers this statement is valid at low ber fractions of bers when the sti ness ratio is larger than one, but invalid for moderate and high ber fractions of bers. For instance the graph with 1 = 2 = 0:4 is not bounded by e = 1=3. A similar Poisson's ratio ow diagram for systems with sti ness ratios less than one would violate this statement even more. Our conclusion is that the sti ness ratio has a large in uence on the e ective Poisson's ratio for periodic structures. The parameter choice 1 = 2 = 1=3, suggested 10,14] as a xed point for random systems of bers, plays a certain role even for this periodic composite structure. We have investigated Poisson's ratio ow diagrams with various parameter choices and reached the following conclusions: For every f2, there is a lowest number up(f2) and a highest number low (f2) such that e < 1 = 2 whenever 1 = 2 > up(f2 ) and such that e > 1 = 2 whenever 1 = 2 < low (f2). In general up(f2), and low (f2) are quite di erent. Only in the dilute limit, f2 < 0:2, do they coincide to up(0) = low (0) = 1=3. This conclusion is a 4

generalization of a suggestion by Garboczi and Day 14] that the e ective Poisson's ratio e satis es 1 = 2 < e < 1=3 and 1 = 2 > e > 1=3 for any random geometry involving overlapping and nonoverlapping bers. For 1 = 2 = 1=3 the system with lowest sti ness ratio has the highest e ective Poisson's ratio as can be seen in Fig. 2. Numerical calculations using the digital image based method on random systems of circular bers 10] and 1 = 2 = 1=3, showed that the system with lowest sti ness ratio had the highest e ective Poisson's ratio although the deviation from 1/3 was small for all considered sti ness ratios and not as large as in our calculations. A general statement for circular bers, supported by our calculations, would be that for 1 = 2 = 1=3 and given a structure, periodic or random, the system with the lowest sti ness ratio has the highest e ective Poisson's ratio. For other values of 1 = 2, no general statement for the e ective Poisson's ratio can be made. Chen, Thorpe and Davis 13] derived an asymptotic formula, near percolation (f2p = p =2 3) for the e ective Poisson's ratio of a hexagonal array of rigid bers; 1+ 1 p (3) e = 7? : 1 We calculated the e ective Poisson's ratio with the integral equation method for two systems with ber fractions near percolation and high sti ness ratio, (E2=E1) = 2000. The rst system had the phase Poisson's ratios 1 = 2 = 0:5 and the second had the phase Poisson's ratios 1 = 0:5, 2 = 0:2. Fiber fraction f2 = 0:90689 gave e = 0:245 for the rst system and e = 0:181 for the second system. Fiber fraction f2 = 0:906899 gave e = 0:21 for the rst system and e = 0:14 for the second system. Equation (3) gives ep = 0:23 with 1 = 0:5. The di erence between our results and the asymptotic result has to do with the order in which we approach in nitely rigid bers and touching bers. In the derivation of Eq. (3) the bers are rst assumed to be in nitely rigid, then touching is approached. Our calculations show that large but nite values of the ber Young's modulus could give results for the e ective Poisson's ratio near percolation that di er signi cantly from the asymptotic relation Eq. (3). To our knowledge, no asymptotic expressions of uniform validity exist that include more material parameters. See McPhedran, Poladian and Milton 26] for an uniformly valid asymptotic formula in the context of electrostatics. Our results show that no xed point is reached for rigid inclusions when percolation is approached. This conclusion has also been reached by Chen, Thorpe and Davis 13] using the digital image based method on a system of rigid circular bers. This paper demonstrates the possibility of accurate calculations of the e ective Poisson's ratio for two-phase ber reinforced composites over the whole parameter space (fi; i; Ei) (i = 1; 2), using the interface integral equation method. Our calculations show that a hexagonal array of circular bers does not exhibit Poisson's ratio xed points in any of the ways discussed in the introduction. The only exception is of course voids in any two-dimensional structure, which have been shown analytically 8,17] to exhibit Poisson's ratio xed points. We also conclude that the sti ness ratio of the phases has a larger in uence on the e ective Poisson's ratio for periodic systems, than can be seen in the existing results 10,14] for random systems. A general statement about the e ective Poisson's ratio for circular bers in random or periodic structures is made. The validity of existing asymptotic relations for the e ective Poisson's ratio near percolation is investigated. 5

ACKNOWLEDGMENTS
We would like to acknowledge useful discussions with L. Gibiansky, G. W. Milton, A. B. Movchan and M. F. Thorpe. HC gratefully acknowledges support from the Swedish research councils NFR and NUTEK and the Goran Gustafsson foundation.

REFERENCES
1] D. J. Bergman and Y. Kantor, Phys. Rev. Lett. 53, 511 (1984). 2] D. J. Bergman, Phys. Rev. B 31, 1696 (1985). 3] S. Feng and M. F. Thorpe, Phys. Rev. B 31, 276 (1985). 4] E. J. Garboczi and M. F. Thorpe, Phys. Rev. B 31, 7276 (1985). 5] L. M Schwartz, S Feng, M. F. Thorpe and P. N. Sen, Phys. Rev. B 32, 4607 (1985). 6] M. F. Thorpe and P. N. Sen, J. Acoust. Soc. Am. 77, 1674 (1985). 7] I. Jasiuk, J. Chen and M. F. Thorpe, J. Mech. Phys. Solids 40, 373 (1992). 8] M. F. Thorpe and I. Jasiuk, Proc. R. Soc Lond. A 438, 531 (1992). 9] A. R. Day, K. A. Snyder, E. J. Garboczi and M. F. Thorpe, J. Mech. Phys. Solids 40, 1031 (1992). 10] K. A. Snyder, E. Garboczi and A. R. Day, J. Appl. Phys. 72, 5948 (1992). 11] J. W. Eischen and S. Torquato, J. Appl. Phys. 74, 159 (1993). 12] I. Jasiuk, J. Chen and M. F. Thorpe, Appl. Mech. Rev. 47, 18 (1994). 13] J. Chen, M. F. Thorpe and L. C. Davis, J. Appl. Phys. 77, 4349 (1995). 14] E. J. Garboczi and A. R. Day, J. Mech. Phys. Solids 43, 1349 (1995). 15] L. C. Davis, K. C. Hass, J. Chen and M. F. Thorpe, Appl. Mech. Rev. 47, 5 (1994). 16] J. Berryman, J. Acoust. Soc. Am. 68, 1820 (1980). 17] A. V. Cherkaev, K. A. Lurie and G. W. Milton, R. Proc. Soc. A 438, 519 (1992). 18] G. Chen and J. Zhou, Boundary Element Methods (Academic Press: London 1992). 19] A. Greenbaum, L. Greengard and A. Mayo, Physica D 60, 216 (1992). 20] V. Rokhlin, J. Comput. Phys. 60, 187 (1985). 21] L. Greengard and V. Rokhlin, J. Comput. Phys. 73, 325 (1987). 22] J. Carrier, L. Greengard and V. Rokhlin, SIAM J. Sci. and Stat. Comput. 9, 669 (1988). 23] H. J. Jou, P. H. Leo and J. S. Lowengrub, submitted to J. Comp. Phys. (1995). 24] J Helsing, J. Mech. Phys. Solids 43, 815 (1995). 25] L. Berman and L. Greengard, J. Math. Phys. 35, 6036 (1994). 26] R. C. McPhedran, L. Poladian and G. W. Milton, Proc. R. Soc. Lond. A 415, 185 (1988).

1.0
holes rigid incl. E2/E1=0.0005 E2/E1=0.002 E2/E1=0.01 E2/E1=0.05

eff

0.5
E2/E1=20 E2/E1=100 E2/E1=500 E2/E1=2000

0.0 0.0

0.5

1.0

f2

FIG. 1 E ective Poisson's ratio e versus the ber fraction f2 for a hexagonal array of circular bers. The phase Poisson's ratios are 1 = 2 = 0:5. The dashed lines have sti ness ratios (E2=E1) < 1. The exact sti ness ratios are listed to the right and the top value in the column is associated with the dashed curve with largest maximum value. The solid lines have sti ness ratios (E2=E1 ) > 1. The corresponding sti ness ratios are listed to the right and the top value in the column is associated with the solid curve with smallest minimum value. Earlier results for holes 9] (open circles) and rigid inclusions 13] ( lled circles) are included.

1.0
E2/E1=0.0005 E2/E1=0.002 E2/E1=0.01 E2/E1=0.05

eff

0.5
E2/E1=20 E2/E1=100 E2/E1=500 E2/E1=2000

0.0 0.0

0.5

1.0

f2

FIG. 2 E ective Poisson's ratio e versus the ber fraction f2 for a hexagonal array of circular bers. The phase Poisson's ratios are 1 = 2 = 1=3. The dashed lines have sti ness ratios (E2=E1) < 1. The solid lines have sti ness ratios (E2=E1) > 1. The corresponding sti ness ratios are listed to the right in the same way as in Fig. 1.

1.0
E2/E1=0.0005 E2/E1=0.002 E2/E1=0.01 E2/E1=0.05

eff

0.5

E2/E1=2000 E2/E1=500 E2/E1=100 E2/E1=20

0.0 0.0

0.5

1.0

f2

FIG. 3 E ective Poisson's ratio e versus the ber fraction f2 for a hexagonal array of circular bers. The phase Poisson's ratios are 1 = 2 = 0. The dashed lines have sti ness ratios (E2=E1) < 1. The solid lines have sti ness ratios (E2=E1) > 1. The corresponding sti ness ratios are listed to the right in the same way as in Fig. 1.

10

0.5

eff

0.0

-0.5 0.0

0.5

1.0

f2

FIG. 4 E ective Poisson's ratio e versus the ber fraction f2 for a hexagonal array of circular bers. The phase Poisson's ratios are 1 = 2 and take the values -0.5 to 0.5 with step 0.1. The sti ness ratio is (E2=E1) = 100.

11

V. TABLES
TABLE I. The connections between the two-dimensional elastic moduli and the three-dimensional elastic moduli in the plane-strain condition and the plane-stress condition. plane-strain elasticity
(2)

plane-stress elasticity
(2)

= 1?

(3) (3) (3) (3) 2

(3)

E (2)

= 1 ?E (

E (2)

= E (3)
(3) (3)

(2)

= (1 ?
(2)

E (3) (3))(1

? 2 (3))
(3)

(2)

= 2(1E ? = 2(1E +

) )

= 2(1E +

(3)

(2)

(3) (3)

12