Neural computing of effective properties of random composite materials

The effective response of disordered heterogeneous materials, in general, is not amenable to the exact analysis because the phase geometry may not be completely specified. The present paper deals with the problem of effective properties such as thermal conductivity, electrical conductivity, dielectric constant, magnetic permeability, and diffusivity in the realm of disordered composites. Even though all these properties are analogous, their numerical treatment in the same unified frameworks may be difficult. In fact, depending on the physical quantity involved, there may be a large discrepancy in the order of magnitude of relevant material parameters. This paper reports a methodology for investigating the effective scalar parameter of disordered composites with the help of the same neural network, regardless of the above physical context. Particular results are obtained for effectively isotropic and macroscopically homogeneous two-phase materials.