Nonlinear interaction in composites using physics informed neural networks

Abstract

Modelling of composites requires the consideration of various components that work together and interact in a linear and nonlinear way. Linear and nonlinear modelling in view of demanding needs, like representative volume element calculations within numerical homogenization and the advent of new tools, like physics informed neural networks, are reviewed in this article. In particular, a concept is proposed towards the implementation of a unilateral contact mechanics law within physics-informed neural networks. The theoretical framework and related applications are presented. Results indicate that the proposed deep learning approach can further be applied towards solving contact mechanics problems, considering the mechanical interactions between the constituents of composites.