XGBoost model for the quantitative assessment of stress corrosion cracking

Abstract

This study employs a data-driven methodology to assess the susceptibility of Fe-Cr-Ni alloys to stress corrosion cracking (SCC) in chloride-containing environments. Historical data from constant-load SCC testing in boiling magnesium chloride were used to train an XGBoost regression model. This model overcomes limitations related to multicollinearity and insufficient sample sizes seen in previous studies. The XGBoost model captures complex interactions between alloy compositions and stresses, explaining 94.9% (R² = 0.949) of SCC susceptibility of the specimens. Shapley additive explanations (SHAP) were employed to interpret the model, offering new metallurgical insights, such as the critical role of nickel content. The SHAP analysis identified an optimal nickel range between 14.5 and 45 wt%, which markedly enhances SCC resistance. The XGBoost-SHAP framework in this work comprehensively isolates the contributions of chemical constituents and stress, offering a path toward more systematic alloy design—departing from the traditional reliance on trial and error or serendipity.