Phase Formation in Vacuum Arc Melted High Entropy Alloys: Insights From Machine Learning and Experimental Validation

In recent years, High Entropy Alloys (HEAs) have gained significant interest due to their unique properties such as high strength, wear resistance, and high temperature stability. However, accurately predicting the phases of HEAs remains a challenge due to multiple elements within the alloy. This study proposes a Machine Learning (ML) framework for phase prediction of HEAs fabricated using the Vacuum Arc Melting (VAM) process. For this, a dataset containing 316 unique experimentally validated HEA combinations were used to train and test seven different classifiers including Support Vector Machine (SVC), K-Nearest Neighbours (KNN), XGBoost (XGBoost), Random Forest (RF), Decision Tree (DT), Artificial Neural Network (ANN), and Gradient Boost (GB). Each alloy’s elemental composition and five different thermodynamic parameters were used as input features. Grid search Cross Validation (CV) method was used for hyperparameter tuning, and all the models were evaluated using various parameters such as 5-fold CV, F1, and Area Under the Curve (AUC) scores. KNN and ANN models achieved the best overall performance with F1 score of 0.83 and 0.81 respectively and AUC score of 0.92 each. XGBoost also performed well with F1 and AUC scores of 0.81 and 0.96 respectively. KNN and ANN have the best overall phase prediction accuracy of 82.81% each. In addition, KNN, ANN, and XGBoost predicted 9, 9, and 8 phases correctly. The model validation uses 10 novel HEA combinations not utilized during training and testing. These results illustrate the potential of ML techniques in design and phase prediction of arc melted HEAs.