Unlocking strength-ductility synergy in tungsten heavy alloys through interpretable machine learning

Tungsten heavy alloys (WHAs), critical for defense and kinetic energy systems, face persistent challenges in achieving simultaneously high strength and ductility within their dual phase, multi-element structures, where conventional design approaches often yield unpredictable outcomes. Here, an interpretable machine learning (ML) framework is developed to accelerate WHA design by integrating independent predictive models for tensile strength (TS) and elongation (EL) through a dual-objective Pareto optimization strategy. Three ML-guided alloys were experimentally validated. The best alloy achieved 1244 MPa in TS and 30 % EL, exceeding the reported mechanical performance of conventional WHAs while maintaining excellent prediction accuracy (3.7 % for TS and 8.4 % for EL). SHapley Additive exPlanations (SHAP) and Generalized Additive Model (GAM) analyses reveal the compositional origins of the strength-ductility synergy and identify key composition-process interactions. The superior mechanical performance is attributed to synergistic improvements of both phases, accompanied by W particle refinement and nanoscale in-situ Ni 3 Ta precipitates reinforcing the γ phase. This interpretable ML framework provides a quantitative, data-driven route to accelerate the design of high-performance WHAs with simultaneously enhanced strength and ductility.