Adaptive Automotive Chassis Welding Joint Inspection Using a Cobot and a Multi-modal Vision Sensor: Adaptive welding joint inspection robotic vision system

摘要

Visual inspection of welding joints has the advantages of being non-contact, low-cost and highly efficient. Automotive chassis welding joints have complex forms, coupled with the increasing demand for flexible manufacturing in factories, their inspection requires more flexibility and affordability. This report proposes an adaptive robotic system for welding joint localization and inspection by placing a baseline-adjustable multi-modal (2D + 3D) vision sensor on a 6-axis cobot mounted under a movable track. The system visually recognizes the model and pose of the chassis on the production belt at a distance, and automatically generates the necessary sensor positions and poses, and trajectories for image acquisition of the welding joints. For each welding joint, the system further manipulates the sensor positions and poses to image it at a close distance from a proper direction. A coarse-to-fine registration of the obtained 3D point clouds is employed to extracts the region of interest in the images and align them to known good welding joint templates using a modified Iterative Closest Point (ICP) algorithm with a conditional similarity metric. Finally, a KD-Tree based method is employed for defect detection using defective point ratio in the point cloud of the identified welding joint region. Experimental results show that the proposed system and methods can efficiently and effectively locate welding joints of different automobile chassis and detects defects on a variety of welding joints, achieving defect detection rate of 99.05%.