Abstract: Currently, the non-destructive testing for weld formation quality of welded structural components still predominantly employs visual evaluation with low efficiency and low accuracy. To improve the situation, an online inspection technology for weld formation quality based on structured-light vision is proposed. The established 3D coordinate recognition model endows this technology with the capability to perceive spatial 3D information; the image processing algorithm integrating YOLOv5 and the spatial distance judgement method autonomously identifies weld types from structured-light images and automatically measures geometrical dimensions such as weld width, reinforcement, and leg sizes; Particularly, this algorithm is not limited to monitoring weld profiles, but can also accurately identify, classify, and locate various geometrical defects including excess weld metal, misalignment, incompletely filled groove, undercut, overlap, excessive convexity, and excessive asymmetry fillet weld, while autonomously evaluating defect quality grades online according to the ISO 5817:1992 standard. Experimental results demonstrate that this technology achieves measurement accuracy at the 10⁻² mm level for weld geometrical dimensions, attains 100% grading accuracy for geometrical defect quality, and fulfills the requirements for online quality inspection of weld formation in automated welding production lines.