Scholarlink

<p>Fatigue failure is a critical concern for Mg alloy welded joints used in lightweight structural applications. Accurate assessment of fatigue life is essential for ensuring structural reliability and service safety. In this study, machine-learning-assisted models are developed to predict the high-cycle fatigue life of Mg alloy welded joints by incorporating stress state, welding method, and material type as input features. Additional physically motivated features derived from empirical fatigue relations are introduced to enhance model robustness. The predictive performance of five representative machine learning models is systematically compared, and the results show that XGBoost achieves the best overall predictive performance on the present dataset. Furthermore, Shapley Additive exPlanations are employed to quantify the influence of key stress parameters, welding methods, and material characteristics on fatigue life. The results indicate that mean stress and stress range dominate fatigue behavior, while welding process and material type also play important roles. This study provides an effective data-driven framework for fatigue performance assessment of Mg alloy welded joints with practical engineering relevance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

High-Cycle Fatigue Performance Assessment of Mg Alloy Welded Joints Using Machine-Learning-Assisted Modeling

  • 期刊论文

  • 发表日期：

    2026年6月24日

查看全文
  • Haili Sun,
  • Qi Dong,
  • Jiaqi Zhang,
  • Yuedong Wang

摘要

Fatigue failure is a critical concern for Mg alloy welded joints used in lightweight structural applications. Accurate assessment of fatigue life is essential for ensuring structural reliability and service safety. In this study, machine-learning-assisted models are developed to predict the high-cycle fatigue life of Mg alloy welded joints by incorporating stress state, welding method, and material type as input features. Additional physically motivated features derived from empirical fatigue relations are introduced to enhance model robustness. The predictive performance of five representative machine learning models is systematically compared, and the results show that XGBoost achieves the best overall predictive performance on the present dataset. Furthermore, Shapley Additive exPlanations are employed to quantify the influence of key stress parameters, welding methods, and material characteristics on fatigue life. The results indicate that mean stress and stress range dominate fatigue behavior, while welding process and material type also play important roles. This study provides an effective data-driven framework for fatigue performance assessment of Mg alloy welded joints with practical engineering relevance.