High-throughput calculation integrated with stacking ensemble machine learning for predicting elastic properties of refractory multi-principal element alloys

The traditional trial-and-error method for designing refractory multi-principal element alloys (RMPEAs) is inefficient due to a vast compositional design space and high experimental costs. To surmount this challenge, the data-driven material design based on machine learning (ML) has emerged as a critical tool for accelerating materials design. However, the absence of robust datasets impedes the exploitation of machine learning in designing novel RMPEAs. High-throughput (HTP) calculations have enabled the creation of such datasets. This study addresses these challenges by developing a data-driven framework for predicting the elastic properties of RMPEAs, integrating HTP calculations with ML. A big dataset of RMPEAs including 4536 compositions was constructed using the new proposed HTP method. A novel stacking ensemble regression algorithm combining multilayer perceptron (MLP) and gradient boosting decision tree (GBDT) was developed, which achieved 92.9% accuracy in predicting the elastic properties of Ti-V-Nb-Ta alloys. Verification experiments confirmed the ML model's accuracy and robustness. This integration of HTP calculations and ML provides a cost-effective, efficient, and precise alloy design strategy, advancing RMPEAs development.