Machine learning-driven insights into phase prediction for high entropy alloys

Abstract

The unique properties of high-entropy alloys (HEAs) have attracted considerable attention, largely due to their dependence on the choice among three distinct phases: solid solution (SS), intermetallic compound (IM), or a blend of both (SS + IM). For this reason, precise phase prediction is key to identifying the optimal element combinations needed to develop HEAs with the required characteristics. Due to large compositional domain of HEAs is opportune to design new HEAs with desired output. A machine learning tool is exploited to discover and characterize high entropy alloys with satisfying targets. Herein, a method of designing substitutional high entropy alloys with optimization of input features and predict their phase formation, using different ML algorithms are proposed. The ML models such as multi layer precreptron MLP, Decision Tree (DT), Random Forest (RF), Gradient Boosting (GB), KNN, XGB nad SVM Classifier algorithm were used for the identifying the phase of HEAs. After assessing the accuracy and tuning of each model, an random forest classifier (accuracy = 0.914. precision = 0.916, ROC-AUC score = 0.97) model showed the best predictive capabilities for phase prediction. The new HEA was designed based on prediction and successfully validated with thermodynamic simulation. Data AvailabilityData will be made available on request