Accelerated design of high entropy alloys by integrating high throughput calculation and machine learning

Abstract

Designing high entropy alloys (HEAs) from a large compositional space is challenging. The design process is accelerated by integrating computational techniques. The current study developed a framework for the accelerated identification of phases in HEAs by integrating high throughput calculations and machine learning (ML). The 19968 composition information was generated using high throughput CALculation of PHAse Diagrams (CALPHAD) calculations and used as a dataset for the machine learning algorithms. The Synthetic Minority Over-sampling Technique (SMOTE) makes unbiased data, and multiple machine learning algorithms are trained and compared. The Support Vector Machine (SVM) and Artificial Neural Network (NN) show test accuracy of ∼99%. The trained algorithm was used to predict phases in ternary, quaternary and quinary systems. The experimental comparison confirms that the algorithms successfully predicted the phase transitions in HEAs with respect to varying compositions. The developed framework can be adopted for designing new materials.