Scholarlink

<p>Despite its impressive structural features, the fatigue behavior of orthotropic steel deck (OSD) is a major concern under the effect of repeated traffic loading. This study aims to propose machine learning (ML) based predictive models for fatigue life prediction of rib-to-floorbeam welded connection (RF-WC) in OSD strengthened by ultra-high reinforced concrete (UHPC). The fatigue crack propagation of RF-WC was first studied using ABAQUS/FRANC3D crack simulation technology to create the dataset for the fatigue life prediction. Three ML-based predictive models were established, including one conventional ML model (SVR model) and two deep learning (DL) models (DNN and LSTM models) for fatigue life prediction of RF-WC. Numerical analysis results indicated that the crack growth rate increased significantly as the stress range and the size of the initial crack increased. On the contrary, a thicker UHPC layer remarkably slows down the crack growth rate. Results showed a strong agreement between the predicted fatigue life using the ML-based predictive models and the numerical analysis results. Additionally, the predicted fatigue life closely aligns with the test results, which further emphasizes the efficiency and high accuracy of the proposed models for fatigue life prediction of RF-WC. It is noticed that, among the proposed ML-based models, the LSTM model demonstrated the highest accuracy for the fatigue life prediction of RF-WC.</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

Machine Learning-Based Predictive Model for Fatigue Life of RF Welded Connection in OSD Strengthened by UHPC

  • 期刊论文

  • 发表日期：

    2026年5月11日

查看全文
  • Hesham Abdelbaset,
  • Zhiwen Zhu

摘要

Despite its impressive structural features, the fatigue behavior of orthotropic steel deck (OSD) is a major concern under the effect of repeated traffic loading. This study aims to propose machine learning (ML) based predictive models for fatigue life prediction of rib-to-floorbeam welded connection (RF-WC) in OSD strengthened by ultra-high reinforced concrete (UHPC). The fatigue crack propagation of RF-WC was first studied using ABAQUS/FRANC3D crack simulation technology to create the dataset for the fatigue life prediction. Three ML-based predictive models were established, including one conventional ML model (SVR model) and two deep learning (DL) models (DNN and LSTM models) for fatigue life prediction of RF-WC. Numerical analysis results indicated that the crack growth rate increased significantly as the stress range and the size of the initial crack increased. On the contrary, a thicker UHPC layer remarkably slows down the crack growth rate. Results showed a strong agreement between the predicted fatigue life using the ML-based predictive models and the numerical analysis results. Additionally, the predicted fatigue life closely aligns with the test results, which further emphasizes the efficiency and high accuracy of the proposed models for fatigue life prediction of RF-WC. It is noticed that, among the proposed ML-based models, the LSTM model demonstrated the highest accuracy for the fatigue life prediction of RF-WC.