Scholarlink

<p>Bolted connections are integral components in steel structures and often are vulnerable to loosening due to cyclic loading and fatigue. Detecting bolt loosening in early stages is critical to prevent sudden catastrophic failures. Recent advancements in image capturing and processing, combined with machine learning techniques have significantly improved the accuracy of detecting bolt loosening. Deep learning techniques, in specific those based on image analysis, have proven to be effective tool for damage detection in structural members. In this study, initially captured dataset of 120 images from a laboratory model is augmented to 576 images using the Roboflow platform. Investigations in this work are tied to one image data set and the metrics are analysed for how well a model performs. Usually, it depends on each use case as to which model would be best suited. In the present work, dataset featured two states of bolt loosening: tight and loose. Convolution Neural Network architectures AlexNet and ResNet are applied individually to the augmented dataset to predict the bolt loosening states as binary classification problem. The results demonstrated an accuracy of 87.931%, indicating that the AlexNet model effectively distinguished between loose and tight conditions. Whereas the ResNet algorithm exhibited relatively better accuracy of 90.8%. Further, an integrated hybrid model (CNN model with LSTM) is also applied on the same image data to further enhance the efficiency of the deep learning frame work. The performance of the trained CNN algorithms was further validated with image responses from various real time scenarios of various lighting conditions and found to be fairly good. This study highlights the potential of using deep learning techniques, specifically AlexNet, ResNet and CNN with LSTM for accurately detecting bolt loosening, thereby contributing to the safety and reliability of steel structures.</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

Efficacy of image based deep learning CNN models for bolt loosening detection in steel structures

  • 期刊论文

  • 发表日期：

    2025年3月29日

查看全文
  • Mallika Alapati,
  • G. Ramesh Chandra,
  • Shivani Abboju

摘要

Bolted connections are integral components in steel structures and often are vulnerable to loosening due to cyclic loading and fatigue. Detecting bolt loosening in early stages is critical to prevent sudden catastrophic failures. Recent advancements in image capturing and processing, combined with machine learning techniques have significantly improved the accuracy of detecting bolt loosening. Deep learning techniques, in specific those based on image analysis, have proven to be effective tool for damage detection in structural members. In this study, initially captured dataset of 120 images from a laboratory model is augmented to 576 images using the Roboflow platform. Investigations in this work are tied to one image data set and the metrics are analysed for how well a model performs. Usually, it depends on each use case as to which model would be best suited. In the present work, dataset featured two states of bolt loosening: tight and loose. Convolution Neural Network architectures AlexNet and ResNet are applied individually to the augmented dataset to predict the bolt loosening states as binary classification problem. The results demonstrated an accuracy of 87.931%, indicating that the AlexNet model effectively distinguished between loose and tight conditions. Whereas the ResNet algorithm exhibited relatively better accuracy of 90.8%. Further, an integrated hybrid model (CNN model with LSTM) is also applied on the same image data to further enhance the efficiency of the deep learning frame work. The performance of the trained CNN algorithms was further validated with image responses from various real time scenarios of various lighting conditions and found to be fairly good. This study highlights the potential of using deep learning techniques, specifically AlexNet, ResNet and CNN with LSTM for accurately detecting bolt loosening, thereby contributing to the safety and reliability of steel structures.