Determining magnetic properties of high entropy alloys by molar volume difference predicted by machine learning

Abstract

Magnetic high entropy alloys (HEAs) have attracted intensive attention for applications in functional devices, ascribed to the vast composition space for designing properties. However, a large number of experiments are needed for designing magnetic HEAs with identified properties by connecting the complex relationship between composition and properties. Herein, we proposed the importance of molar volume difference in determining magnetic properties of five-element HEAs by using machine learning (ML). The database is established first, and feature parameters connected with compositions and experimental factors are chosen as the input values of ML. ML is performed by using extreme gradient boosting and random forest algorithms, which all present acceptable training results for predicting the saturation magnetization and coercivity. It is found that the molar volume difference plays a key role in determining both saturation magnetization and coercivity after calculating the feature importance. Our results could give some tips for the experimental design of magnetic HEAs, and the method can be extended to predict the comprehensive performance of magnetic HEAs by improving the dataset.