Prediction of alloying element effects on the mechanical behavior of high-pressure die-cast Mg-based alloys

Abstract

Achieving optimal mechanical performance in high-pressure die-cast (HPDC) Mg-based alloys through experimental methods is both costly and time-intensive due to significant variations in composition. This study leverages machine learning (ML) techniques to accelerate the development of high-performance Mg-based alloys. Data on alloy composition and mechanical properties were collected from literature sources, focusing on HPDC Mg-based alloys. Six ML models—extra trees, CatBoost, k-nearest neighbors, random forest, gradient boosting, and decision tree—were trained to predict mechanical behavior. CatBoost yielded the highest prediction accuracy with R² scores of 0.95 for ultimate tensile strength (UTS) and 0.92 for yield strength (YS). Further validation using published datasets reaffirmed its reliability, demonstrating R² values of 0.956 (UTS) and 0.936 (YS), MAE of 1% and 2.8%, and RMSE of 1% and 3.5%, respectively. Among these, the CatBoost model demonstrated the highest predictive accuracy, outperforming other ML techniques across multiple optimization metrics.