A new welding path planning method based on point cloud and deep learning

Abstract

Scan the weldment by 3D vision sensors to obtain the 3D point cloud data that can accurately describe the spatial position information of the weldment to enable automatic welding. Accurate point cloud segmentation and optimized path generation are the two pivotal issues that need to be solved in robot welding path planning. To cope with the highly redundant and uneven dense point clouds, this paper proposed a deep learning based welding path planning method with a real missile air rudders welding case. According to the structural characteristics of air rudder products, a 3D point cloud segmentation method for air rudders is designed. A convolutional neural network segmentation model for air rudder 3D point clouds is proposed to accurately segment weldment. Then, an extraction method for the characteristic points of the air rudder welding seam and the curve fitting method for the irregular welding path are designed to generate an accurate air rudder welding path. Finally, the effectiveness of the proposed method is verified by eight types of air rudder products. The results show that the proposed method achieves competitive performance than other comparative methods.