A Crystal Plasticity Finite Element—Machine Learning Combined Approach for Phase Transformation Prediction in High Entropy Alloy

Abstract

The mechanical properties of an alloy depend on its microstructure. The strength-ductility trade-off is a paradigm that existed for a long time. Advanced alloys, such as high entropy alloys (HEAs), utilize a dual-phase strengthening mechanism, which originates from the microstructural phenomena consisting of twinning and phase transformation, to significantly improve their mechanical properties. To understand the impact of phase transformation mechanism on stress–strain response, developments of crystal plasticity finite element models (CPFEM) and machine learning (ML) together with experimental methods have potential to capture the relationships between descriptive features and targeted phenomena. Here, ML models on local crystallography, local stresses, and energy-based driving forces are leveraged for phase transformation prediction in a HEA. The ML model (XGBoost classification model) uses a hybrid training data combining electron backscatter diffraction (EBSD) experimental data and CPFEM simulation results. This approach enhances prediction performance at optimum data sizes. This predictive model is implemented in multiple experimental measurements to validate our models and evaluates importance of different physical quantities on phase transformation phenomenon. The prediction accuracy reached over 95% compared to experimental data. The CPFEM-ML framework used in this study is expected to be applicable to other HEA systems to facilitate the understanding and prediction of the phase transformation.