A flexible seam detection technique for robotic laser welding

This thesis deals with the problem of seam detection for robotic laser welding applications. Within the Mechanical Automation laboratory of the Engineering Technology faculty, a compact, lightweight and multi-purpose welding head is developed. The integrated welding head is attached to a robot and should be able to detect seams, learn trajectories, do the laser welding and inspect the welds. A video camera is attached to the laser focussing optics to obtain a realtime stream of images of the work piece when the robot moves the head along the seam. This report focusses on compensating the camera images for radial distortions introduced by the welding optics, detection of the seam in the camera images and the translation of image coordinates (in pixels) to real world coordinates (in millimetres). For estimating the distortion of the images a calibration procedure based on Zhang’s algorithm [1] has been implemented in Matlab. Also an algorithm for automated extraction of reference points from calibration images and an undistortion function have been implemented. Extensive testing has shown that the effect of distortions can be well compensated for after calibration of the optical system. The welding optics are not optimised for imaging applications. Together with the wish for a large field of view, this results in images with large distortions and aberrations. Image processing algorithms have been developed to cope with these problems in a robust way. The seam detection algorithm makes use of a structured light projection onto the work piece. Using the triangulation principle the seam position can be estimated from the camera images. Tests have been done with a single straight line projection and with a projection of two crossing lines over an overlap joint. In these tests both methods proved to be capable of tracking several test seams. For the world coordinate calibration an automated procedure has been developed which detects a marker in the camera images. The marker is positioned at the tool centre point, the focussing spot of the welding laser (or at a known offset). Position changes of the marker in the images are then related to prescribed robot movements, in order to determine the position and orientation of the welding head with respect to the marker.