Implicit material modelling using artificial intelligence techniques

Abstract

Simulation of forming operations, particularly using the finite element method, is clearly dependent on the accuracy of the constitutive models. In the last years, several methodologies were developed to improve the accuracy of constitutive models through parameter identification and calibration methodologies. However, independently of the efficacy of the calibration methods, the accuracy of a constitutive model is always constrained to its predefined mathematical formulation. Today, artificial intelligence (AI), such as Machine-learning (ML) techniques, can be used to overpass these limitations. However, their use in the reproduction of material behaviour was not fully explored. This work proposes to model the behaviour of a metal material using ML techniques and use them in forming simulations. In this preliminary work, the ML model is defined by an artificial neural network and trained using a virtual material, whose behaviour is reproduced by a classical Chaboche-type elastoviscoplasticity model. This procedure allows evaluating the ML competence at least to replace classical models. Different ML topologies and optimization techniques are used to train the model. Then, the ML model is introduced into a finite element analysis (FEA) code, as a user subroutine, and its attainment in more complex strain states is evaluated. The replacement of classical formulations by AI techniques for the material behavior definition is analysed and discussed.