It is not completely understood fatigue strength at this time due to its complex formation mechanism. Therefore, in order to address this issue, machine learning has been used to examine the important factors involved in predicting fatigue strength. In this study, a hybrid model was proposed based on the modified bagging method by combining XGBoost and LightGBM, in which the hyperparameters of the models were optimized by a grey wolf algorithm. Moreover, an interpretable method, referred to as Shapley additive explanations (SHAP), was introduced to explain the fatigue strength predictions made by ML models. The SHAP values were calculated, and feature importance of fatigue strength by XGBoost, LightGBM and the hybrid model was discussed. The final results demonstrated that the SHAP method had major potential for interpreting fatigue strength predictions, which would provide constructive guidance for the development of antifatigue steel material in the future.

中文翻译：

---

基于机器学习的钢疲劳强度预测及机理分析

疲劳强度由于其复杂的形成机制，目前还没有完全了解。因此，为了解决这个问题，机器学习已被用于检查预测疲劳强度所涉及的重要因素。在本研究中，结合XGBoost和LightGBM提出了一种基于改进bagging方法的混合模型，其中模型的超参数通过灰狼算法进行优化。此外，还引入了一种称为 Shapley 加性解释 (SHAP) 的可解释方法来解释 ML 模型所做的疲劳强度预测。计算SHAP值，讨论了XGBoost、LightGBM和混合模型疲劳强度特征的重要性。