Modelling and prediction of hardness in multi-component alloys: A combined machine learning, first principles and experimental study

Abstract

Abstract In present study, the relations between hardness and elemental descriptors in the multi-component alloys (MACs) are particularly uncovered via machine learning (ML) and first-principles calculations. The RBF neural network is utilized to efficiently train a large database which allows an acceptable accuracy for identifying the overall role of elemental features for target properties. Detailed information from ML predictions indicates the critical element of Al and its significant advantages for the hardness in a model Al–Cr–Fe–Ni system. Investigations on elastic properties using first-principles calculations provide relations between physical quantities and match well with ML model. Furthermore, the Al 1.2 CrFeNi alloy was selected as an example, experimentally synthesized and properties-identified, to provide a strong validation in this case. Combined with some other reported alloys in this system, the prediction of the developed model can achieve above 90%. Further experimental analysis found that a dual phase microstructure presents in the Al 1.2 CrFeNi alloy, leading to a superior work hardening ability compared with the reported high performance alloys. This study offers reliable composition-properties features and encourages more researches using this integrated approach to guide for tuning novel MACs.