Vision-Based trajectory extraction of welding holes on the sealing end of pressure vessels

Most current research on weld seam extraction based on computer vision primarily focuses on planar butt welds and fillet welds, while there is limited research on weld seams on complex surfaces. This paper proposes a method for extracting welding trajectories for nozzle welds on pressure vessel heads, specifically for holes with circular or elliptical paths. Firstly, preprocess the point clouds to enhance the quality and reliability of the point clouds data. Subsequently, extract the inner contour of the point cloudss by establishing a complex coordinate system at the centroid of the point clouds, which will serve as the original boundary. In the point cloud registration stage, the Iterative Closest Point (ICP) algorithm is employed. A nonlinear optimization problem is formulated, where the design variables include the three rotation angles (Euler angles) and the geometric parameters of features in the template point cloud (the radius for a circular hole, or the major and minor axes for an elliptical one). The objective function is defined as the registration error, which is minimized to achieve optimal alignment. Experimental results demonstrate that this algorithm can be applied to the welding of complex surfaces, realize the automatic guidance for welding robots, meet the requirements for welding accuracy and efficiency, and thus verify the feasibility of the algorithm.