Deep learning based phase transformation model for the prediction of microstructure and mechanical properties of hot-stamped parts

Abstract

Hot stamping of boron steel is an effective way for weight reduction of automobiles parts, but suffers from the loss of mechanical properties due to incomplete martensitic transformation. Different from general heat treatment, phase transformation in hot stamping relates to both thermal loading path and deformation history. Unfortunately, the existing models have difficulty in capturing the history-dependent phase transformation behavior of boron steel under hot stamping conditions. In the present work, the gated recurrent unit (GRU) are combined with fully connected neural network (FCNN) to capture the coupling effect of both nonlinear strain history and thermal loading path on phase transformation. A unified deep-learning based phase transformation (DLPT) model is developed to integrate both diffusive and diffusionless transformation to cover all feasible thermo-mechanical loading path in hot stamping process. A hybrid driven thermo-mechanical-metallurgical (TMM) framework is developed by integrating the DLPT model into general TMM framework, which providing a novel approach for two-scale simulation of hot stamping process. A comprehensive database of phase transformation is constructed for model training based on well-designed thermo-mechanical loading experiments to cover hot stamping conditions, and the optimal topology of neural network in DLPT model is obtained by training on this database. The accuracy and reliability of DLPT model is systematically demonstrated by dieless hot V-bending and hot stamping of T-shaped parts besides CCT and DCCT tests. The net effect of local plastic strain on subsequent phase transformation is clarified with accuracy by dieless V-bending with the removal of disturbing factors. Non-uniform distribution in hardness of the final hot stamped parts can be predicted with the DLPT model successfully, which is better in accuracy than in case of working with the conventional model.