Scholarlink

<p>This study proposes a physics-guided, vibration-based structural health monitoring (SHM) framework for steel truss bridges by integrating experimental modal analysis, finite element simulation, and hybrid machine learning. Progressive material degradation was physically induced in SS400 steel specimens through controlled annealing and corrosion treatments to represent systematic stiffness loss. Experimental modal testing reveals that a stiffness reduction of approximately 50% constitutes a critical transition point, marking a shift from stiffness-dominated to degradation-dominated dynamic behavior. This transition is characterized by pronounced reductions in natural frequencies, accompanied by increased damping ratios and accelerated vibration decay rates. These experimentally validated modal parameters were subsequently employed to construct a two-stage machine learning pipeline. Unsupervised K-Means clustering was first used to identify latent stiffness degradation states without prior labeling, followed by supervised Random Forest classification to establish robust and physically interpretable decision boundaries. The proposed framework achieved near-perfect classification accuracy across four degradation levels, with misclassification confined to adjacent transitional states. Feature importance analysis confirms that dominant frequency, damping ratio, and decay time are the most influential diagnostic indicators, reinforcing the physical consistency of the model. Finally, the transfer of specimen-scale degradation indicators to a Warren truss bridge case study demonstrates the scalability of the proposed approach to system-level applications. The results highlight the feasibility of a non-destructive, explainable, and deployment-ready SHM framework capable of supporting real-time condition assessment and maintenance prioritization for aging steel truss bridge infrastructure.</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

Integration of finite element–based modal analysis and hybrid machine learning for early-stage stiffness degradation diagnosis in steel truss bridges

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月27日

查看全文
  • Bambang Sugiantoro,
  • Utis Sutisna,
  • Susilo Adi Widyanto,
  • Achmad Widodo,
  • Sukamta Sukamta

摘要

This study proposes a physics-guided, vibration-based structural health monitoring (SHM) framework for steel truss bridges by integrating experimental modal analysis, finite element simulation, and hybrid machine learning. Progressive material degradation was physically induced in SS400 steel specimens through controlled annealing and corrosion treatments to represent systematic stiffness loss. Experimental modal testing reveals that a stiffness reduction of approximately 50% constitutes a critical transition point, marking a shift from stiffness-dominated to degradation-dominated dynamic behavior. This transition is characterized by pronounced reductions in natural frequencies, accompanied by increased damping ratios and accelerated vibration decay rates. These experimentally validated modal parameters were subsequently employed to construct a two-stage machine learning pipeline. Unsupervised K-Means clustering was first used to identify latent stiffness degradation states without prior labeling, followed by supervised Random Forest classification to establish robust and physically interpretable decision boundaries. The proposed framework achieved near-perfect classification accuracy across four degradation levels, with misclassification confined to adjacent transitional states. Feature importance analysis confirms that dominant frequency, damping ratio, and decay time are the most influential diagnostic indicators, reinforcing the physical consistency of the model. Finally, the transfer of specimen-scale degradation indicators to a Warren truss bridge case study demonstrates the scalability of the proposed approach to system-level applications. The results highlight the feasibility of a non-destructive, explainable, and deployment-ready SHM framework capable of supporting real-time condition assessment and maintenance prioritization for aging steel truss bridge infrastructure.