Synergistic process optimization for strength and ductility in an Mg-Gd-Y-Zn-Zr alloy using an interpretable ensemble learning framework with data augmentation

Achieving high performance has long been a primary objective in alloy design. With the advancement of materials informatics, the integration of data‑driven approaches with domain knowledge has provided a new design paradigm for the synergistic optimization of multiple properties in alloys. This work proposes an Ensemble Learning with Noise Augmentation for SHAP‑guided Multi‑Objective Optimization (ELNASMO) framework to enable the synergistic optimization of strength and ductility in a Mg‑13.9Gd‑2.1Y‑0.77Zn‑0.51Zr alloy. ELNASMO first employed Gaussian noise-based data augmentation to expand the original small-sample dataset. Building on this augmented dataset, multiple ensemble learning models based on Bagging, Boosting, and Stacking strategies were constructed and evaluated. The results indicated that the Stacking model was the optimal predictor, achieving coefficients of determination (R 2 ) of 0.98 for both ultimate tensile strength (UTS) and elongation (EL) predictions. By integrating the Stacking model with the SHapley Additive exPlanations (SHAP) method, the influence mechanisms of processing parameters on UTS and EL were analyzed. Based on these insights and physical metallurgy criteria, the search space for the Non-dominated Sorting Genetic Algorithm II (NSGA-II) was refined. Subsequently, by coupling the Stacking model with the NSGA-II algorithm, a Pareto front for UTS and EL was constructed to maximize strength while maintaining ductility no lower than the as-cast baseline. The optimal processing parameters were identified as: homogenization at 530°C for 10 h, extrusion at 440°C, extrusion ratio of 16, and extrusion speed of 0.5 mm/s. Under these optimized conditions, the Stacking model predicted UTS and EL values of 456.6 MPa and 7.4%, respectively. Subsequent experimental validation yielded actual values of 451.1 MPa and 6.3%. Consequently, the ELNASMO framework effectively achieves high-strength design in rare-earth Mg alloys while satisfying ductility engineering constraints. Furthermore, it provides a versatile and efficient methodological strategy for the synergistic optimization of multiple properties in other complex alloy systems.