A Fast Point Cloud Reconstruction Algorithm for Saddle-Shaped Weld Seams in Boiler Header Joints

In boiler industries, the automated welding of header and tube-seat joints is a critical topic. Traditional offline programming method suffers from deviations between models and real workpieces. Vision aided welding provides a feasible way to deal with the problem, but the point cloud acquisition and reconstruction are yet to be developed. In this paper, a fast point cloud reconstruction method for saddle-shape welding seams in boiler header and tube-seat joints has been proposed. The point clouds of header and tube-seat joints in boilers are acquired by a partition scanning strategy, and then reconstructed using a two-step registration method. The partial point clouds are first coarsely spliced using position relationship, and then fine registered to improve accuracy. Besides, a simple but effective benchmark, the registration mark, has been proposed. The performance of splicing in each stage has been evaluated with the registration mark. The registration mark after fine registration in each iteration is lower than 1 mm. Finally, the partial clouds are spliced to form a complete cloud.