Scholarlink

<p>Gas metal arc welding (GMAW) of AISI 304 stainless steel is extensively employed in structural and industrial applications due to its corrosion resistance and weldability, yet it remains prone to porosity, lack of fusion, spatter, and pen- etration defects that compromise joint integrity. Weld quality has traditionally been evaluated through visual inspection (VI), a method influenced by inspec- tor skill, fatigue, and environmental conditions. To address these limitations, machine vision (MV) was comparatively evaluated under AWS D1.6 (2017). Six specimens were prepared, one defect-free and five with induced discontinuities. VI was conducted by certified inspectors, while MV used digital cameras, CMY preprocessing, supervised labeling, and contour detection. Validation employed. Confusion matrices and performance metrics. Results showed VI detected major discontinuities but lacked reproducibility. MV consistently identified all induced defects, achieving 0.92 precision, 1.00 recall, and 0.96 F1-score, confirming its robustness as an alternative for weld quality control.</p> Graphical abstract <p></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

Comparative evaluation of visual inspection and two-dimensional machine vision for the detection of discontinuities in GMAW welds of AISI 304 stainless steel

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月25日

查看全文
  • Melitsa Torres-Torres,
  • Roosvel Soto-Diaz,
  • Jose Escorcia-Gutierrez,
  • Aida Valls

摘要

Gas metal arc welding (GMAW) of AISI 304 stainless steel is extensively employed in structural and industrial applications due to its corrosion resistance and weldability, yet it remains prone to porosity, lack of fusion, spatter, and pen- etration defects that compromise joint integrity. Weld quality has traditionally been evaluated through visual inspection (VI), a method influenced by inspec- tor skill, fatigue, and environmental conditions. To address these limitations, machine vision (MV) was comparatively evaluated under AWS D1.6 (2017). Six specimens were prepared, one defect-free and five with induced discontinuities. VI was conducted by certified inspectors, while MV used digital cameras, CMY preprocessing, supervised labeling, and contour detection. Validation employed. Confusion matrices and performance metrics. Results showed VI detected major discontinuities but lacked reproducibility. MV consistently identified all induced defects, achieving 0.92 precision, 1.00 recall, and 0.96 F1-score, confirming its robustness as an alternative for weld quality control.

Graphical abstract