Machine learning-based prediction of hydrogen-assisted fatigue crack growth rate in Cr–Mo steel

Abstract

In this study, linear and tree-based algorithms were utilized to train the hydrogen-assisted fatigue crack growth test data. Through accuracy verification, it was found that the Gradient Boosting Regression (GB) belonging to tree-based algorithms had the best predictive performance. For the model trained by the GB algorithm, SHapley Additive exPlanations (SHAP) values were employed to evaluate the impact of stress intensity factor range, hydrogen pressure, ultimate tensile strength, stress ratio, frequency and chemical compositions on the hydrogen-assisted fatigue crack growth. The results show that material properties, experimental conditions and environment all have an effect on the crack growth rate. Besides, the feature influence patterns derived from machine learning models are consistent with the literature, demonstrating that the model has the potential to accurately predict the hydrogen-assisted fatigue crack growth rate of Cr–Mo steel.