Accurate Detection Of Random Small Car Body Welding Slags And Autonomous Decision-making Of Robotic Grinding Path

The spot welding process of the car body is easy to cause random spatter and small-sized welding slags,which not only affects the overall aesthetics of the car body,but also affects the later spraying process.At present,domestic automobile manufacturers mainly remove the welding slags by manual visual inspection and manual grinding.Using machine vision technology to detect welding slags on the car body surface and automatic grinding can greatly improve the efficiency and automation level of welding slag grinding,but the premise is that the welding slags must be accurately detected and the grinding path of the industrial robot must be independently determined.In this thesis,the car door frame is taken as the research object.Based on the target detection neural network and the path decision algorithm,the research on the accurate detection of random tiny welding slags on the door frame and the autonomous decision-making of the robotic grinding path is investigated.The main research contents of this thesis are as follows:Firstly,an improved car body welding slags detection algorithm based on the YOLO v3 algorithm is proposed to improve the model ability to detect small targets.By comparing the YOLO v3 algorithm before and after improvement with other target detection algorithms,the improved YOLO v3 algorithm can better adapt to the detection of welding slags,improve the recall rate and accuracy of detection.This step provides support for the accurate positioning of subsequent welding slags areas.Secondly,the K-Means algorithm is used to divide the welding slags area,and the convergent binocular camera is used to accurately locate the welding slags area.The positional relationship of the calibrated binocular camera relative to the industrial robot is determined by the hand-eye calibration method;Also,the detected welding slags are divided into regions by the clustering algorithm,and the position of the welding slags region relative to the robot is determined according to the calibration results,which lays the foundation for the follow-up grinding path decision-making.Thirdly,according to the detection and positioning results,the Dijkstra algorithm is used to make autonomous decisions on the set grinding trajectory of the car body welding slags,and the autonomous decision-making software for the grinding processing path of the car door frame robot is developed.The global grinding paths are planned through offline programming of the robot,and the entry and exit points and the transition points required for decision-making are set to avoid the problem of axis configuration when the robot is grinding.Autonomous decision-making to get the shortest grinding path.Based on the software and hardware conditions and processing requirements,the autonomous decision-making software for the grinding processing path of the car door frame robot is developed to realize the local selfadaptation of the grinding robot for the car body welding slags.Finally,the door frame welding slags grinding experiment is carried out.Based on the improved YOLO v3 algorithm and visual calibration,the welding slags positioning is completed,and the feasibility of the detection and positioning method is verified by positioning error analysis.Based on the Dijkstra algorithm and offline trajectory planning,the autonomous decision-making of the grinding path of the robot is realized.The experimental results verify the feasibility of the autonomous decisionmaking algorithm for the grinding path of the robot proposed.The research work in this thesis combines deep learning and path decisionmaking algorithms,which provides a new solution for randomly distributed tiny welding slags detection and local adaptive grinding and polishing of welding slags,which has a high application prospect.