Scholarlink

<p>In modern sheet metal manufacturing, steel coils serve as critical intermediate products for downstream processes, such as cutting, stamping, and forming. However, coil loosening can cause issues such as misalignment and jamming, creating safety risks during material handling. Timely and accurate detection of such loosening is essential for ensuring production stability, improving yield, and preventing mechanical failures. Traditionally, coil loosening is either overlooked, which potentially compromises product quality and safety, or assessed through manual inspection, which is often inconvenient and time-consuming. Herein, we propose an automated coil loosening detection method based on deep learning techniques. A pre-trained (VGG) model is utilized to extract scale-adaptive convolutional features, and coil images are captured from video streams when these features match the labeled reference. The coil region is then segmented using U²-Net to remove background noise. To identify coil loosening, edge detection and radial analysis techniques are applied. Furthermore, the roundness of the coil is quantified via ellipse fitting and Delaunay triangulation to assess the severity of the loosening. The proposed algorithm was validated using videos captured in a real-world manufacturing environment. The experimental results demonstrate detection accuracies of 97.5% and 85% for outer and inner coil loosening, respectively, which confirms the robustness and reliability of the proposed method for real-world applications.</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

Deep Learning–based automated detection of steel coil loosening in sheet metal manufacturing

  • 期刊论文

  • 发表日期：

    2025年12月8日

查看全文
  • Lijing Chen,
  • Yinan Miao,
  • Yeseul Kong,
  • Seunghwan Lee,
  • Penghua Zhang,
  • Gyuhae Park

摘要

In modern sheet metal manufacturing, steel coils serve as critical intermediate products for downstream processes, such as cutting, stamping, and forming. However, coil loosening can cause issues such as misalignment and jamming, creating safety risks during material handling. Timely and accurate detection of such loosening is essential for ensuring production stability, improving yield, and preventing mechanical failures. Traditionally, coil loosening is either overlooked, which potentially compromises product quality and safety, or assessed through manual inspection, which is often inconvenient and time-consuming. Herein, we propose an automated coil loosening detection method based on deep learning techniques. A pre-trained (VGG) model is utilized to extract scale-adaptive convolutional features, and coil images are captured from video streams when these features match the labeled reference. The coil region is then segmented using U²-Net to remove background noise. To identify coil loosening, edge detection and radial analysis techniques are applied. Furthermore, the roundness of the coil is quantified via ellipse fitting and Delaunay triangulation to assess the severity of the loosening. The proposed algorithm was validated using videos captured in a real-world manufacturing environment. The experimental results demonstrate detection accuracies of 97.5% and 85% for outer and inner coil loosening, respectively, which confirms the robustness and reliability of the proposed method for real-world applications.