Studying the micromechanical behaviors of a polycrystalline metal by artificial neural networks

Abstract Though physics-based crystal plasticity models are able to accurately predict the material response under complex loading conditions, the high computational cost limits their engineering applications. Machine learning techniques, widely used to understand and predict data trends, can provide advantageous computational efficiency over conventional numerical techniques. In the current work, an artificial neural network (ANN) model is proposed and trained based on the datasets generated by the physics-based viscoplastic self-consistent (VPSC) model on the material behavior of OFHC copper. The ANN model successfully captures the loading path-dependent micromechanical behavior of the copper polycrystals with arbitrary texture, even beyond the bounds of the generated dataset. The proposed ANN model, without any major loss in accuracy, can greatly improve the computational efficiency. It can be envisioned that the engineering application of the physics-based model becomes promising through alternatively using the ANN model, which treats the underlying complex physics as a blackbox.