Material informatics for uranium-bearing equiatomic disordered solid solution alloys

Near-equiatomic, multi-component alloys with disordered solid solution phase (DSSP) are associated with outstanding performance in phase stability, mechanical properties and irradiation resistance, and may provide a feasible solution for developing novel uranium-based alloys with better fuel capacity. In this work, we build a machine learning (ML) model of disordered solid solution alloys (DSSAs) based on about 6000 known multi-component alloys and several materials descriptors to efficiently predict the DSSAs formation ability. To fully optimize the ML model, we develop a multi-algorithm cross-verification approach in combination with the SHapley Additive exPlanations value (SHAP value). We find that the ΔSC, Λ, Φs, γ and 1∕Ω, corresponding to the former two Hume − Rothery (H − R) rules, are the most important materials descriptors affecting DSSAs formation ability. When the ML model is applied to the 375 uranium-bearing DSSAs, 190 of them are predicted to be the DSSAs never known before. 20 of these alloys were randomly synthesized and characterized. Our predictions are in-line with experiments with 3 inconsistent cases, suggesting that our strategy offers a fast and accurate way to predict novel multi-component alloys with high DSSAs formation ability. These findings shed considerable light on the mapping between the material descriptors and DSSAs formation ability.