Interpretable material descriptors for critical pitting temperature in austenitic stainless steel via machine learning

Austenitic stainless steel is renowned for its exceptional corrosion and mechanical properties, yet it remains susceptible to localized corrosion, such as pitting in the presence of aggressive ions or/and extreme environmental conditions. Critical pitting temperature (CPT) serves as a key metric for evaluating the susceptibility to pitting corrosion, and its accurate prediction is essential for engineering pitting-resistant alloys. In this work, through optimized feature selection processes, three critical features − standard reduction potential ( $${E}_\max $$ ), valence electron ( $${{VEC}}_{{sum}}$$ ), and geometric parameters ( $$\lambda$$ ) − are identified as crucial descriptors for CPT. Utilizing interpretable machine learning techniques, a predictive model for CPT is developed and confirmed via cross-validation, demonstrating superior predictive accuracy. This work not only deepens our understanding of the underlying factors affecting pitting but also facilitates the development of austenitic stainless steel with improved resistance to pitting corrosion.