A STRUCTURED LIGHT-BASED VISUAL SENSING SYSTEM FOR DETECTING MULTI-LAYER AND MULTI-TRACK WELDING

摘要

In practice, there is an urgent need for on-line monitoring and control to improve the welding performance, especially for an automatic welding system. In this work, a vision servo system based on a structured light vision sensor was established for detecting multi-layer and multi-track welding seams, which can acquire the region of interest (ROI) for the welding seam by a laser projection method. First, the image noise was removed using a median filter and threshold segmentation according to the welding seam width information. Second, an improved Hessian centreline extraction algorithm was established to extract the centreline of the welding seam with an enhancement of speed over 10 times by employing a graphics processing unit (GPU). The Hessian eigenvector was employed to locate the positions of inflection points, which provide the key information for recognizing the welding seam features. Following that, three-dimensional coordinates were obtained to reconstruct a 3D model of the welding seam through calibrations, including camera calibration, structured-light plane calibration and hand-eye calibration. Finally, using the least square and greedy triangle methods, the virtual model of the welding seam can be reconstructed, and the features of width and height can be calculated to evaluate the welding process, thereby providing an effective on-line method for the welding inspection, evaluation and control.