On the Performance of Different Architectures in Modelling Elasto-Plasticity with Neural Network

Abstract

Constitutive models have been utilized to study the mechanical behaviors of solid material. The formulation of constitutive relations is difficult and could be associated with limiting hypothesis. This work proposes neural network-based approaches to reproduce the complex nonlinear constitutive relations of solid materials including elastic behavior and plastic deformation. It is shown that the proposed history-based and internal variable-based strategies can represent exactly the von Mises elasto-plastic material model in uni-axial stress state. Furthermore, close investigation suggests that the internal variable-based approach is most suitable.

Full Paper