Scholarlink

<p>Crimping defects in strain clamps represent a significant threat to the structural integrity and reliability of overhead power transmission lines, yet current diagnostic methods are often reactive and lack a systematic framework for automated defect identification. This paper presents an integrated methodology that synergizes validated finite element modeling with a deep learning framework to address this challenge. A finite element model of the NY-500/45 strain clamp was first developed to simulate ten distinct defect types, generating a comprehensive dataset of their unique mechanical failure signatures, with simulation fidelity confirmed against experimental data (failure force error &lt; 10%). The experimental dataset was then used to train and validate a novel hybrid neural network architecture, combining YOLOv5 for region proposal and Faster R-CNN for precise classification. The resulting system demonstrates high efficacy, accurately identifying defect locations from X-ray imagery with a mean positional error below 3.5 pixels and achieving a higher mean Average Precision (mAP) than standalone models. Ultimately, this work establishes a robust foundation for developing real-time, intelligent diagnostic tools, offering a critical advancement towards proactive quality assurance and enhanced power grid resilience.</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

Simulation analysis and intelligent recognition of defects for strain clamps

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月27日

查看全文
  • Jiahui Chen,
  • Qian Peng,
  • Chunqing Yang,
  • Ruibin Chen,
  • Yi Xie,
  • Qianchuan Xiang,
  • Yue Ma,
  • Jigang Huang

摘要

Crimping defects in strain clamps represent a significant threat to the structural integrity and reliability of overhead power transmission lines, yet current diagnostic methods are often reactive and lack a systematic framework for automated defect identification. This paper presents an integrated methodology that synergizes validated finite element modeling with a deep learning framework to address this challenge. A finite element model of the NY-500/45 strain clamp was first developed to simulate ten distinct defect types, generating a comprehensive dataset of their unique mechanical failure signatures, with simulation fidelity confirmed against experimental data (failure force error < 10%). The experimental dataset was then used to train and validate a novel hybrid neural network architecture, combining YOLOv5 for region proposal and Faster R-CNN for precise classification. The resulting system demonstrates high efficacy, accurately identifying defect locations from X-ray imagery with a mean positional error below 3.5 pixels and achieving a higher mean Average Precision (mAP) than standalone models. Ultimately, this work establishes a robust foundation for developing real-time, intelligent diagnostic tools, offering a critical advancement towards proactive quality assurance and enhanced power grid resilience.