Machine Learning Approach to Design High Entropy Alloys with Heterogeneous Grain Structures

Abstract

Heterogeneous nanocrystalline high-entropy alloys (HEAs) have excellent mechanical properties. However, it is still difficult to obtain the optimized grain size in the heterogeneous-grained HEAs, which achieves their outstanding mechanical properties. Here, using a novel integration method of machine learning, a physical model and atomic simulation, the optimal grain size is designed for achieving high yield strength of heterogeneous-grained CrCoFeNi HEAs. Atomic simulations give the stress–strain curve, yielding strength and microstructure with the increase of small grain size. The physical-based strength model expands the data from the atomic simulations and obtains the transition region from the Hall–Petch to inverse Hall–Petch relationship. The results show that the strength of CrCoFeNi HEAs derives mainly from the contribution of the grain boundary compared to lattice friction stress. The machine learning model shows that the obvious transition point from the Hall–Petch to inverse Hall–Petch relationship occurs at the grain size of 38.4 nm for the heterogeneous-grained CrCoFeNi HEAs with the large grain size of 165 nm. This result agrees with the prediction from the subsequent atomic simulation. This integrated model makes significant contributions to understanding deformation and designing the microstructure of heterogeneous-grained HEAs. Importantly, the developed model including simulation, a theoretical model, experiment and machine learning can be widely applied to explore the advanced material with the desired performance.