Machine learning-mediated composition-process optimization and high strength-toughness performance achievement of ductile iron

To overcome the high cost and inefficiency associated with traditional trial-and-error methods in optimizing the strength and toughness of ductile cast iron, this study proposes an intelligent design framework that combines physical metallurgy features with machine learning. By combining composition-process data with atomic-scale elemental attributes, a high-dimensional feature space was constructed. Multi-level feature screening was employed to extract key interactive features influencing ultimate tensile strength (UTS) and elongation (EL). High-precision performance prediction models were developed based on a multilayer perceptron, and inverse design of three composition-process schemes for ductile cast iron with balanced high strength and high toughness was achieved by integrating the NSGA-II multi-objective optimization algorithm. Experimental validation confirmed their excellent mechanical properties, the UTS reaches 940 MPa, and the EL reaches 6.3%. SHAP analysis further elucidated the regulatory mechanisms of the key features. As a preliminary exploratory extension, the proposed feature engineering method was preliminarily applied to a 7xxx series aluminum alloy in the T6 temper. The model achieved promising predictive performance, with an R 2 of 0.90 for UTS and 0.87 for EL. These results preliminarily suggest that this feature construction strategy may have the potential to be transferred to other precipitation-hardened alloy systems.