Machine learning-based constitutive model for J2- plasticity

This research aims to propose a machine learning (ML)-based constitutive model to predict elastoplastic behavior for J2-plasticity. An artificial neural network (ANN) was constructed to replace the nonlinear stress-integration scheme conducted in the conventional theoretical constitutive model under isotropic hardening and associated flow rule. The training dataset required for the ANN Model was numerically generated based on the conventional return mapping scheme in the principal stress space. The training has been effectively carried out with one element simulation along all the possible plastic loading paths for problem independent training. A conventional theoretical method is used for the unloading procedure. Therefore, ANN is selectively utilized only for nonlinear plastic loading while keeping linear elastic loading and the unloading with a physics-based model. After one element training, the ML-based constitutive model was implemented in Abaqus User MATerial (UMAT) and its performance was verified. For this purpose, one element and tensile test simulations were applied to examine the accuracy of the ANN-based model. Also, for fully nonlinear strain-paths, a circular cup drawing simulation was applied to predict the cup profiles which was compared with that to the conventional J2 plasticity. It was concluded that the simulation results predicted from the ANN-based model show good agreement with those from the conventional J2-based constitutive model. Also, according to simulation time, the ANN-based model shows an advantage in computational efficiency.