Research On Automatic Detection And Positioning Technology Of Weld Groove Based On Vision

Automated and intelligent welding robotics has become an essential part of the modern industrial manufacturing process.Some current welding robots mainly track the weld seam by means of a line laser active vision sensor in conjunction with traditional image algorithms,which requires manual guidance of the robot to the vicinity of the workpiece before tracking,reducing the degree of automation of robotic welding.To this end,this dissertation proposes a vision-based robotic system for automatic detection and positioning of weld grooves,introducing a depth camera with a deep learning-based weld groove identification and positioning algorithm that identifies and positions the weld without human intervention,thus improving the level of robotic welding automation and providing technical support for flexible processing robotic welding.The specific research work is as follows:(1)Extensive research has been conducted on the coordinate systems of the system.Initially,the definitions of various coordinate systems were established,accompanied by the completion of fundamental theories and formula derivations for each coordinate system.Subsequently,the calibration of camera,hand-eye,and TCP(Tool Center Point)was accomplished,ensuring the calibration of coordinate relationships.This transformation unified the coordinates within the robot’s base coordinate system,providing a standardized representation of coordinates.Such groundwork serves as a solid foundation for the subsequent design of weld seam detection and positioning methods.(2)A weld groove detection algorithm is studied and designed.The YOLOv5 algorithm is investigated and improved by introducing a simple and lightweight Coordinate attention mechanism in its network with little extra computation to improve the extraction of weld grooves in complex environments.The results show that the m AP is 82.3% before the improvement and 90.8% after the improvement,which is a significant improvement compared with that before the improvement.The real-time frame rate of weld groove detection by the improved YOLOv5 algorithm reaches 20 FPS,which meets the requirements of weld groove detection in actual production.(3)The research designs a set of "coarse" and then "fine" weld groove positioning algorithm.The detection algorithm is fused with the depth camera ranging algorithm to calculate the coordinates of the weld groove in the robot coordinate system,and guide the robot to approach the weld seam to complete the coarse positioning of the weld groove.Fine positioning extracts the 3D point cloud on the weld groove surface by laser displacement sensor,and performs ROI extraction by direct-pass filtering,outlier removal by Gaussian filtering method,RANSAC linear segmentation,least-squares linear fitting,linear intersection point calculation and other steps to extract the weld feature points,and all feature points form the weld trajectory to complete the fine positioning of the weld.(4)Based on the system scheme and the laboratory conditions to complete the experimental hardware selection,according to the system content to carry out the design of the upper computer software system,and complete the development of the upper computer interface.The feasibility of the method,algorithm accuracy and system performance were experimentally studied by the vision-based robot system for automatic identification and positioning of weld grooves,using butt weld grooves and fillet weld grooves as objects.The results show that the positioning error of the butt weld groove is within 0.3 mm in the X and Y directions and 0.5 mm in the Z direction,while the positioning error of the fillet weld groove is within 0.3 mm in the X,Y and Z directions,which can meet the working requirements of the welding robot for weld detection and positioning.