A Novel Weld Adaptive Path Generation Method Based on 3-D Point Cloud for Robotic Multilayer Multipass Welding

In the field of medium-thick plate welding, multilayer multipass (MLMP) welding technology with high precision and a wide range of adaptability has attracted much attention. The traditional welding robot’s “teaching and playback” mode cannot be efficiently and flexibly applied to the robot MLMP welding. Currently, weld feature extraction relies on 2-D planar information, which cannot fully utilize the geometric information of the weld seam. Moreover, the existing MLMP welding path generation methods are only for V-beveled weld seams, which cannot flexibly cope with different shapes of welded workpieces. In this article, a novel weld adaptive path generation method based on a 3-D point cloud for robotic MLMP welding is proposed. A time-of-flight (TOF) sensor is used to extract the 3-D point cloud of the weld seam as input, and the weld seam is segmented by random sample consistent weld segmentation algorithm based on density clustering (DC-RANSAC) to extract the feature points of the weld seam. The feature points are interpolated and smoothed to generate an MLMP weld path of the weld seam based on the geometric features of the weld workpiece point cloud, and finally, the end pose of the weld torch is calculated by the normal estimation method of weighted principal components analysis (PCA). Experiments show that this method has higher accuracy as well as stability than the traditional MLMP method, and it has efficiently experimented on different shapes of welded workpieces such as filet, lap, and butt joints.