Scholarlink

<p>Accurate detection and identification of weld flaws in the X-ray image have been realized as important in ensuring structural integrity and preventing possible failures of welded components within different industries. Accurate analysis of X-ray images plays a key role in quality control by reducing failure risks and ensuring structural reliability. This study focuses on the detection and localization of welding defects in such images in the welds in X-ray images. To handle the challenges in defect identification, a sliding window convolutional neural network (SW-CNN) model has been implemented. The CNN model is applied for image classification, while the sliding window technique refines the defect localization after classification. Our model reached a mean average precision of 84%, which showed better weld defect detection and localization accuracy. This methodology works out the critical issues that usually come up in the analysis of X-ray images. Furthermore, the serialization of the CNN model for easy deployment and sharing makes this proposed methodology practical and versatile for applications in weld-quality assessment.</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

Dynamic Deep Learning for Weld Defects Detection and Localization

  • 期刊论文

  • 发表日期：

    2025年11月13日

查看全文
  • Dalila Say,
  • Mounira Tarhouni,
  • Salah Zidi

摘要

Accurate detection and identification of weld flaws in the X-ray image have been realized as important in ensuring structural integrity and preventing possible failures of welded components within different industries. Accurate analysis of X-ray images plays a key role in quality control by reducing failure risks and ensuring structural reliability. This study focuses on the detection and localization of welding defects in such images in the welds in X-ray images. To handle the challenges in defect identification, a sliding window convolutional neural network (SW-CNN) model has been implemented. The CNN model is applied for image classification, while the sliding window technique refines the defect localization after classification. Our model reached a mean average precision of 84%, which showed better weld defect detection and localization accuracy. This methodology works out the critical issues that usually come up in the analysis of X-ray images. Furthermore, the serialization of the CNN model for easy deployment and sharing makes this proposed methodology practical and versatile for applications in weld-quality assessment.