摘要： The quality of robotic laser welded parts is related to the joint location, the trajectory of the laser focal point and the process parameters. By performing in-process monitoring, it is possible to acquire sufficient process knowledge for post-inspection to evaluate the geometrical weld quality. The existing solutions for such systems operate along linear welds. This paper contributes with a quality inspection system for robot laser welding, that can handle double-curved geometries. The data acquisition system includes a CMOS camera, which is mounted such that it looks through the laser optics, external LED illumination and matching optical filters. During the process, the area around the moving laser focal point is captured, resulting in a sequence of images. The trajectory of the focal point is determined by estimating the 2D displacement field between each image using template matching and subsequently filtering the data through a Kalman filter to improve the accuracy and robustness of the system. The joint location is determined by applying a Canny edge detector and a standard Hough transform within a specified region of interest. As this paper deals with double-curved geometries, the region of interest is moved in relation to the laser trajectory, such that it always contains the visible part of the joint, that is closest to the focal point. The developed post-inspection system evaluates the quality of the weld by comparing the estimated trajectory relative to the determined location of the joint. The performance of the proposed quality inspection system was validated empirically on 18 samples. The tests showed promising results, as the system was able to accurately detect changes in the welding trajectory relative to the location of the joint with an accuracy of ± 0.2 mm.