Mechanics-informed machine learning for predicting mechanical properties of heat-treated multi-metal alloys

This study investigates the mechanical behavior and damage evolution of heat-treated multi-metal alloys through an integrated framework combining machine learning and mechanics-based analysis. A comprehensive dataset comprising approximately 1500 alloy samples was analyzed to predict ultimate tensile strength (Su) using various regression models. Among the evaluated algorithms, XG Boost demonstrated superior predictive performance, achieving a coefficient of determination of R2 = 0.98 and a root mean square error (RMSE) of 44.26 MPa. To ensure model interpretability and physical consistency, SHAP (SHapley Additive ex-Planations) analysis was employed, revealing that shear modulus (G), Young’s modulus (E), elongation at fracture (A5), Brinell hardness (BHN), and yield strength (Sy) are the most influential parameters governing tensile strength. The results indicate that higher elastic and shear moduli significantly enhance Su, in agreement with fundamental mechanical principles. In addition, key mechanical responses including fatigue crack propagation, energy absorption capacity (up to 2420 J), fracture toughness (KIC ranging from 50 to 120 MPa m1/2), and surface hardness gradients were analyzed to provide a comprehensive understanding of deformation and failure mechanisms. The proposed approach offers a robust, physically interpretable, and data-driven tool for predicting and optimizing the mechanical performance of heat-treated metallic alloys.