Design and Development of Real-Time Vision AI Welding Analyzer

Abstract

Welding training needs simple, reliable tools that give quick feedback on bead quality to help learners improve skills and meet industry standards. The main purpose of the study was to develop a real-time vision AI welding analyzer, specifically describing its technical features, its accuracy, sensitivity, response time, and acceptability in terms of technical features, composition, operating performance, and safety features. The research used a developmental method, and the evaluation was conducted using a researcher-made questionnaire, validated by the members of the panel. Thirty (30) evaluators, comprising professors, welders, Technology and Livelihood Education (TLE) instructors, and end-users, assessed the developed device, providing valuable insights for its proper utilization in welding training. The device provided real-time feedback on weave and stringer beads via a web dashboard with pass/fail metrics and safety features. The findings showed that the device delivered reliable weld bead evaluation with advanced technical features, strong operating performance, and comprehensive safety measures. It achieved 90% accuracy in classifying acceptable stringer and weave beads, 100% precision with no false positives, and instant response times within the 10 second threshold. It was very acceptable, confirming its effectiveness as a supportive tool for welding education and training.