Seam extraction strategy and path generation based on point cloud for robotic welding system

Abstract Due to the complexity of welding processes and the generation of welding defects, traditional methods such as manual teaching and offline programming have become inefficient and less adaptable to diverse workpieces. In this paper, weld seam extraction and path planning methods based on structured light 3D scanners are proposed to enhance the accuracy and stability of weld seam trajectories and provide standardized and reliable input for intelligent welding systems. The 3D modelling of welds is realized through the robot’s hand-eye system, addressing the inability of offline programming systems to handle inaccurately modelled or partially missing weld data. A novel weld seam extraction algorithm combining robust Mahalanobis distance and principal component analysis is proposed. Compared with robust principal component analysis (RPCA) and random sampling consensus (RANSAC), the proposed method improves the accuracy of identifying outliers by 1.15% and 51.14%, respectively, when the initial point cloud outlier ratio is 20%. The running time is reduced by 1.23 s and 2.36 s, respectively when 10000 points are processed. Polynomial curve fitting methods are presented to optimize the generation of weld seam trajectories. Experimental results demonstrate that the proposed method can achieve autonomous welding with maximum weld point deviation of less than 0.6mm. And it is suitable for promotion and application in practical welding robots.