Data-driven design of high bulk modulus high entropy alloys using machine learning

Abstract

In the current research, machine learning (ML) models were used as a tool for predicting the bulk modulus of High Entropy Alloys (HEAs). ML was employed to optimize HEA compositions for superior bulk modulus values. The study assessed five regression models: Random Forest (RF), K-Nearest Neighbors (KNN), XGBoost (XGB), Support Vector Regression (SVR), and Lasso regression. The XGB regression model delivered the best results, with an R-squared (R2) value of 95.2 % and an RMSE of 2.6 % on the validation dataset. The XGB model's performance was further validated by experimental work, showing an R2 value of 94.8 % and an RMSE of 3.6 %. The R-squared, RMSE, and MAE values during training, testing, and validation for the XGB model ranged from 93.2 % to 99.62 %, 0.97 to 3.64, and 0.12 to 1, respectively. Furthermore, we used the top three trained models to predict the bulk modulus of six new HEAs that were not part of the training, testing, or validation datasets. These predictions achieved R² values of 94.8 %, 93.4 %, and 92.4 %, RMSE values of 3.6 %, 4.1 %, and 4.4 %, along with MAE values of 3.4 %, 3.8 %, and 4.1 %, for the XGB, Lasso, and SVR models, respectively. This work advances the field by bridging the gap in HEA discovery and property evaluation, offering novel methods for designing HEAs with desirable bulk modulus values, and unlocking new possibilities for HEA applications.

Keywords

• High entropy alloys
• Mechanical properties
• Machine learning
• Bulk modulus
• HEAs applications