A novel welding path generation method for robotic multi-layer multi-pass welding based on weld seam feature point

Traditional “teach and playback” mode limits the efficiency and adaptability of robotic multi-layer multi-pass (MLMP) welding. The tracking solely based on the seam points may result in an unstable welding process with bad filling quality. In this paper, a novel welding path generation method for MLMP weld based on seam feature points is proposed. The 3D weld reconstruction is realized during the welding torch round-trip movement in the MLMP welding process. The FPLDN network is proposed to detect the seam feature points for each welding pass. To achieve accurate key direction vector estimation, an adaptive weighted PCA-based normal estimation method and an improved RANSAC method are used for weld segmentation and fitting. Then, the welding torch position and posture can be estimated in the nearest neighbor of seam feature points with further smoothing and interpolating. In the experiment, this method showed better performance in precision and stability than traditional methods with the root mean square error (RMSE) less than 0.771 mm.