LAUSR

Abstract Recent rapid progress in metal forming has brought difficult problems when it comes to accurate prediction in plastic deformation in which high-strength advanced metals are used. If a metal that exhibits strong anisotropy is applied, the use of some advanced material model is recommended. Usually, advanced material models require large number of material parameters to be determined by experiments. To avoid this situation, construction of numerical material testing could be beneficial if such methodology can be an alternative for some difficult-to-conduct experiments. Therefore, we have developed a numerical material testing using finite element polycrystalline model based on successive integration method. The proposed method consists of crystal plasticity-based model and a deep neural network to capture the microstructural behavior of polycrystalline metals. In this study, a description of the proposed method that is based on the concept of material learning, and some verification with experimental data are presented. In the learning phase, experimental data obtained from in-plane tensile tests are provided as teaching data, and after the multiscale material learning, the virtual material will acquire generality to non-learned out-of-plane mechanical characteristics. In this work, prediction for equi-biaxial stress-strain relation resulted in acceptable agreement with experimental data from a literature.