Predicting Elastic Constants of Refractory Complex Concentrated Alloys Using Machine Learning Approach.

Area of Science:

• Materials Science
• Computational Materials Science

Background:

• Refractory complex concentrated alloys (RCCAs) exhibit desirable mechanical and thermal properties, including creep resistance, ductility, and oxidation resistance.
• Elastic constants are critical for understanding RCCA properties but are challenging and costly to determine experimentally.
• Accurate prediction of elastic constants is essential for the efficient design and application of new RCCAs.

Purpose of the Study:

• To develop and validate a machine learning (ML) approach for predicting the elastic constants of RCCAs.
• To compare the performance of different ML regression models, including Random Forest, Gradient Boosting Regressor (GBR), and XGBoost, for this prediction task.

Main Methods:

• Utilized density functional theory (DFT) simulation data combined with ML algorithms.
• Trained and evaluated Random Forest Regressor, Gradient Boosting Regressor (GBR), and XGBoost regression models.
• Assessed model performance using R-squared, mean average error, and root mean square error metrics.
• Validated the most promising model (GBR) on an independent dataset of refractory high-entropy alloys (RHEAs).

Main Results:

• The Gradient Boosting Regressor (GBR) model demonstrated superior performance in predicting RCCA elastic constants compared to Random Forest and XGBoost.
• GBR model predictions showed reasonable agreement with experimental and computational results when validated on unseen RHEA datasets.
• The developed ML approach significantly reduces the need for expensive experimental and computational efforts.

Conclusions:

• The Gradient Boosting Regressor (GBR) model is a highly effective tool for accurately predicting the elastic constants of new refractory complex concentrated alloys (RCCAs).
• This ML-driven approach accelerates the discovery and optimization of RCCAs with tailored properties.
• The findings pave the way for more efficient materials design and development in the field of advanced alloys.