A Machine Learning-based Constitutive Model for Nonlinear Analysis via Finite Element Method
Álefe F. Figueiredo, Saulo S. Castro, Roque L. S. Pitangueira, Samir S. Saliba
CILAMCE 2020 - XLI Ibero-Latin-American Congress on Computational Methods in Engineering , Foz do Iguaçu , 2020
Resumo (em inglês)
This paper addresses a machine learning technique in the context of constitutive modelling. Since it has been proven that multilayer perceptrons with the backprogapation algorithm are capable of approximating any class of functions, studies have been developed with the objective of using it as approximation functions for the nonlinear behaviour of complex material media. This is only possible because neural networks have a powerful adaptability, capability of learning and generalizability. In this context, a multilayer perceptron is trained with stress-strain results from a nonlinear analysis via finite element method with Mazars material in order to develop a neural network-based constitutive model. This implementation is carried out with the help of a recognized machine learning package in order to obtain more accurate results. To validate the proposed constitutive model, the results obtained through the multilayer perceptron are compared with the ones of the finite element numerical analysis.