Prediction of phases in high entropy alloys using machine learning

Abstract

Predicting the phases of high entropy alloys (HEAs) is a complex task as it depends on a large number of thermal, physical and geometrical parameters. The phases consist of intermetallic (IM) and amorphous (AM) forms. As, the trial-and-error approach is exhaustive and time-consuming process for both experiments and numerical simulations. Standard statistical models such as XGboost, random forest, adaboost, decision tree, logistic regression, support vector machine (SVM) and k-nearest neighbors (KNN) were adopted. The XGboost algorithm yields the better accuracy (90 %). Machine learning methods well predicted the phases and also identified the main significant variables such as mean atomic radius, atomic size difference, mixing enthalpy, ideal mixing entropy and valence electron concentration, which are in consistence with experimental results. As a whole, it is recommended that the widespread use of machine learning approaches could accelerate HEA process with tailored configurations. The use of multiple models in ensemble learning reduces the risk of overfitting, increases the accuracy of predictions, and enhances the generalizability of models.