Target-driven design of high strength yet corrosion resistant medium Mn steel via interpretable machine-learning

Corrosion-resistant steels with high strength and large ductility are desirable for industrial applications. In this study, an interpretable machine learning (ML) framework including data collection, data augmentation, model selection, and experimental validation was developed for predicting and optimizing the mechanical yet corrosion properties of medium Mn steels. Furthermore, a model interpretation based on Shapley additive explanation (SHAP) approach was utilized to analyze the feature importance. A significant finding in SHAP analysis is that the austenitization temperature serves as the key factor in tailoring mechanical yet corrosion properties. Microstructural analysis further reveals that the performance-optimized medium Mn steel shows a heterogeneous microstructure with both lamellar and block morphology. The block austenite with lower Mn content decreases its potential difference and corrosion sites compared with the lamellar microstructure for micro-galvanic corrosion. The heterogeneous austenite yields a sustainable transformation-induced plasticity (TRIP) effect during deformation and improved strength-ductility synergy. Thus, the mechanical and corrosion properties of medium Mn steels are synchronously enhanced. The interpretable ML strategy for identifying key variables affecting mechanical yet corrosion performance, along with heterogeneous microstructure design to improve the comprehensive properties in medium Mn steel, can also be applied to other materials systems.