An artificial neural network approach on crystal plasticity for material modelling in macroscopic simulations

Abstract Anisotropy plays a significant role in engineering, especially in the field of sheet metal forming. This particular characteristic stems mainly from the crystallographic structure of the metals and the influence of the rolling process, inducing preferred orientations of the grains. In this context, the crystal plasticity theory plays an important role as it accounts for the anisotropic nature of the elastic tensor and the orientation dependencies of the crystallographic deformation mechanisms. Despite the advantages and capabilities, the integration of the crystal plasticity theory in macro simulations is hindered by high computational costs. A novel approach aims to rectify this problem through the application of machine learning. Therefore, this work investigates the machine learning of crystal plasticity simulations, whereby the DAMASK simulation kit package is used both as a benchmark for quality and costs as well as for providing a data basis for the training and testing of the neural networks. A phenomenological material model for an AA5083 aluminium alloy provides the training data for a neural network study, testing different input parameters as well as network setups.