Optical monitoring for distinct defect detection in laser micro-welding via signal signatures and machine learning

This study examines defect recognition in high-speed continuous-wave laser micro-welding of thin stainless-steel sheets. Using a three-channel photodiode system to monitor laser back-reflection, infrared, and visible emissions, optical signals were correlated with defect formation under dynamic conditions. The effects of laser angle, power, focus, and inter-sheet gap on signal behavior and defect sensitivity were analyzed. Distinct signal patterns were linked to specific defect types, and machine learning enabled high-confidence classification. The results demonstrate that integrated optical sensing coupled with machine learning can support real-time quality monitoring and defect classification in laser micro-welding.