Machine learning assisted modelling and design of solid solution hardened high entropy alloys

Abstract

High entropy alloys (HEAs) are considered as a way to unlock the unlimited potentials of materials during material design, where solid solution hardening (SSH) is one of the major contributors to their excellent mechanical properties. In this work, machine learning (ML) is applied for modelling SSH in HEAs, and a ML system is established for designing solid solution hardened HEAs. The ML-SSH model is built by considering critical factors in SSH theories and parameters associated with the atomic environment and interactions in HEAs as input features, and is demonstrated to be superior to physical SSH models in terms of hardness prediction. The effects of charge transfer and short range order (SRO) and local composition fluctuations on SSH in HEAs are confirmed using feature engineering approaches. Furthermore, two physical models are modified by introducing charge transfer to enhance their accuracy. Finally, an alloy design system is built by combining the ML-SSH model and ML models for single solid solution phase prediction, achieving good agreement with the experimental results of FeNiCuCo and CrMoNbTi families. The non-equiatomic counterparts with 28.3% and 8.8% hardness values higher than their equiatomic counterparts of FeNiCuCo and CrMoNbTi families respectively are discovered.

Keywords

• Solid solution hardening
• High entropy alloys
• Machining learning