Interpretable machine learning design for concurrent and significant enhancement of the mechanical properties and corrosion resistance of low-density Mg-Li alloys

Abstract

• An interpretable machine learning strategy was developed to guide the design of low-density Mg-Li alloys. • Clarify the key intrinsic factors affecting the ultimate tensile strength, elongation, and corrosion rate of Mg-Li alloys. • The newly designed alloys achieve 34%∼44% higher strength and 35%∼79% lower corrosion rates than LAZ103, while maintaining high ductility and low density. Designing low-density, high-strength Mg-Li alloys is a major challenge in achieving extreme lightweighting of high-end equipment. This study proposes an interpretable machine learning strategy to simultaneously enhance the mechanical properties and corrosion resistance of Mg-Li alloy. Key alloy factors (KAFs) influencing ultimate tensile strength (UTS), elongation (EL), and corrosion rate (CR) were identified through alloy factor construction and screening. Using KAFs and processing parameters as inputs, gradient boosting regression models for UTS, EL, and CR were established, achieving the coefficients of determination of test-set above 0.85. Then, SHapley Additive exPlanations (SHAP) analysis quantified the impact of KAFs, and an element evaluation method was established to identify Al, Si, Ca, and Zn as candidates for alloy design. Finally, three new alloys were designed via multi-objective optimization. In the hot-extruded state, they exhibited UTS of 253∼273 MPa, EL of 18.4%∼27.9%, CR of 0.55∼1.61 mg/(cm 2 ·day), and ρ of 1.49∼1.54 g/cm 3 . Compared to LAZ103, the new alloys show 34%∼44% higher UTS, 35%∼79% lower CR, and comparable ρ. Microstructural analysis revealed increased α-Mg, decreased β-Li, reduced coarse secondary phases, and fine Ca-/Si-rich precipitates which are conducive to grain refinement and dislocation density increasing, synergistically enhancing comprehensive property.

Topics & Concepts