Interpretable machine learning-assisted inverse design of high-strength and ductile cast magnesium alloys

Le résumé fourni par la source

Cast rare-earth (RE) magnesium (Mg) alloys are governed by strong coupling among multicomponent compositions, elemental interactions, and heat treatment parameters, making on-demand design difficult using trial-and-error methods or conventional ML models with limited interpretability. Here, an interpretable ML-assisted composition-processing inverse design strategy was developed from a dataset of 1367 cast RE Mg alloy records. Model screening identified LightGBM for tensile yield strength (TYS) prediction and XGBoost for ultimate tensile strength (UTS) and elongation (EL). After hyperparameter optimization and stacking integration, the final models achieved test-set R 2 values of 0.92, 0.92, and 0.84 for TYS, UTS, and EL, respectively. SHAP and PDP analyses revealed the effects of key compositions, elemental interaction features, and heat treatment parameters, which were coupled with NSGA-II-based multi-objective inverse optimization. Pareto-front screening yielded 139 candidates, and two representative alloys were experimentally validated. After T6 treatment, strength-oriented Alloy1 reached 294 MPa TYS, 336 MPa UTS, and 11.2% EL, while ductility- oriented Alloy2 achieved 207 MPa, 301 MPa, and 21.4%, respectively, with all prediction errors below 10.5%. These results demonstrate interpretable ML-enabled on-demand design of cast RE Mg alloys with targeted strength or ductility.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.