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

Area of Science:

• Materials Science
• Mechanical Engineering
• Computational Science

Background:

• Fatigue failure in railway axles compromises train operational safety.
• 25CrMo4 steel is a critical material for high-speed train axles, necessitating accurate fatigue life assessment.
• Reliable transportation depends on understanding and predicting material fatigue.

Purpose of the Study:

• To compare the predictive performance of six machine learning models for 25CrMo4 steel fatigue life.
• To optimize model hyperparameters using a differential evolution algorithm.
• To identify key features influencing fatigue life predictions in 25CrMo4 steel.

Main Methods:

• Six machine learning models were evaluated.
• Hyperparameter optimization was performed using a differential evolution (DE) algorithm.
• SHapley Additive exPlanations (SHAP) and partial dependence plots (PDP) were used for interpretable analysis.

Main Results:

• The DE-optimized Gaussian process regression (DE-GPR) model demonstrated superior predictive accuracy (R²=0.8020, RMSE=0.1250).
• Applied stress level was identified as the most significant feature impacting fatigue life predictions.
• Interactions between applied stress, surface residual stress, and FWHM were found to influence predicted fatigue life.

Conclusions:

• The DE-GPR model offers a robust approach for accurate fatigue life assessment of 25CrMo4 steel.
• Understanding feature importance enhances the reliability of fatigue life predictions.
• This research provides valuable insights for optimizing manufacturing processes and ensuring the safety of railway components.