Scholarlink

<p>Surface quality inspection is essential in modern manufacturing, where local surface defects can strongly influence component performance and reliability. This study presents a multiscale curvature-based method for extracting surface features, such as grooves, scratches, and dimples, with greater accuracy than conventional segmentation methods. Unlike watershed segmentation or particle analysis, which often suffer from over-segmentation or loss of orientation information, the proposed approach leverages principal curvature magnitude and direction to provide physically meaningful descriptors of local geometry. Validation was carried out on three morphologically different surfaces: a polymeric human skin replica, machined 304 stainless steel, and a calibration artifact. Results show that the method can (i) isolate large convex ridges on the skin replica and separate them by orientation, (ii) distinguish machining grooves created at two distinct angles on steel surfaces and further extract individual grooves via clustering, and (iii) discriminate between concave dimples and convex flanges on calibration artifacts, including separation of valleys aligned with x- and y-directions, capabilities not achieved by watershed or particle-based approaches. The method’s rotation-invariance and scale-adaptability make it particularly suited for in-line quality inspection, defect detection, and process validation in manufacturing contexts.</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

Curvature-based multiscale feature extraction for surface quality inspection in manufacturing

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月18日

查看全文
  • François Berkmans,
  • Tomasz Bartkowiak,
  • Karol Grochalski,
  • Michal Wieczorowski,
  • Maxence Bigerelle

摘要

Surface quality inspection is essential in modern manufacturing, where local surface defects can strongly influence component performance and reliability. This study presents a multiscale curvature-based method for extracting surface features, such as grooves, scratches, and dimples, with greater accuracy than conventional segmentation methods. Unlike watershed segmentation or particle analysis, which often suffer from over-segmentation or loss of orientation information, the proposed approach leverages principal curvature magnitude and direction to provide physically meaningful descriptors of local geometry. Validation was carried out on three morphologically different surfaces: a polymeric human skin replica, machined 304 stainless steel, and a calibration artifact. Results show that the method can (i) isolate large convex ridges on the skin replica and separate them by orientation, (ii) distinguish machining grooves created at two distinct angles on steel surfaces and further extract individual grooves via clustering, and (iii) discriminate between concave dimples and convex flanges on calibration artifacts, including separation of valleys aligned with x- and y-directions, capabilities not achieved by watershed or particle-based approaches. The method’s rotation-invariance and scale-adaptability make it particularly suited for in-line quality inspection, defect detection, and process validation in manufacturing contexts.