Fatigue Life Prediction of 25CrMo4 Alloy Steel Based on Interpretable Methods

The fatigue failure of railway axles is directly associated with the operational safety of trains. As 25CrMo4 steel is commonly employed for high-speed train axles, precise evaluation of its fatigue life is essential for transportation reliability. This study compared six machine learning models following hyperparameter optimization via a differential evolution algorithm. The DE-optimized Gaussian process regression (DE-GPR) model exhibited superior predictive performance, achieving a coefficient of determination (R²) of 0.8020 and a root mean square error (RMSE) of 0.1250 on the most significant outer test fold. Furthermore, an interpretable analysis of the model utilized a combination of SHapley Additive exPlanations (SHAP) and partial dependence plots (PDP) to elucidate feature importance. The results indicate that the applied stress level is the predominant feature affecting fatigue life predictions and that it slightly interacts with surface residual stress and full width at half maximum to influence the predicted fatigue life. This study can provide valuable insights into the fatigue life assessment and process optimization of 25CrMo4 steel components.