Machine learning to predict effective conductivity of composite materials for shielding applications

The design of shielding systems such as enclosures for embedded electronics is often based on numerical modelling. When composite materials are used in these shields, it’s necessary to determine their effective properties of equivalent homogeneous materials to simplify the modelling work. When the properties of the internal structure are well known and simple, analytical homogenisation methods are suitable for determining the effective properties of the equivalent homogeneous material. In the case of composite materials with complex or unknown structure, analytical homogenisation approaches often become invalid. To deal with these cases, we propose an approach based on machine learning techniques to evaluate the effective conductivity of composite materials from near-field shielding effectiveness measurements. This approach consists of solving an inverse problem with shielding effectiveness as input and conductivity as output. Results from two composite samples are presented. The first is a composite reinforced by woven aluminium grid, for which an analytical mixing formula can be applied, and the second is a composite reinforced by randomly distributed carbon fibres, for which the internal properties are unknown. Our results are validated by comparing the measured shielding effectiveness with those calculated using the effective conductivities obtained from the machine learning approach and the analytical homogenised formula.