Vision-based Weld Recognition And Path Tracking For Underwater Welding Robot In Nuclear Power Plant

With the rapid development of nuclear power technology,more and more nuclear power plants have been built and put into use.Nuclear power safety issues have gradually attracted people's attention.One of the major nuclear accidents is the leakage of spent fuel pools in nuclear power plants.When the nuclear power pool leaks due to strong external forces such as natural disaster or falling heavy objects,it will seriously threaten human life safety.Manual repair of leaks is not only inefficient and costly,but also radiation damage to rescue workers.Therefore,it is necessary to use underwater robots for welding identification and welding repair.In this paper,a welding detection method and a welding path tracking method for automatic welding of underwater robots in a nuclear water pool environment are designed,and the key issues therein are analyzed and studied in depth.Aiming at the problem of imaging distortion caused by the watertight sealing glass in front of the lens of the underwater camera,the causes of the distortion are analyzed,and the true underwater image information is restored through distortion correction.The underwater weld image was preprocessed to remove noise interference,and the effects of different edge detection algorithms on weld detection were compared.The Canny algorithm has the best detection effect.By analyzing the deficiencies of the Canny algorithm,an adaptive Canny algorithm based on the K-means clustering algorithm is proposed,which effectively improves the adaptability of the algorithm.The Zhang-Suen algorithm was used to extract the skeleton of the weld image,and the center line of the weld was obtained to provide data for welding path tracking.The influences of three major disturbance factors,such as bubbles,arcs,and water ripples,on the visual recognition of welds were analyzed.Based on this,an image bubble noise removal method based on optical flow method,an arc light noise removal method based on Haar wavelet decomposition,and a water ripple noise removal method based on illuminance analysis are proposed.These methods can effectively remove the identification of underwater weld and improve the accuracy of weld position recognition.By analyzing the underwater vision welding system,the transformation relationship between the image coordinate system and the world coordinate system is established.Trajectory planning of welding robot arm is carried out in Cartesian space.According to the difference between the adjacent trajectory segments,different transfer transition strategies are adopted to ensure the stable welding.By analyzing the influence of underwater robot body disturbance on welding error,the anti-disturbance path following control method based on fuzzy PID is designed.The simulation model was built in Simulink and the error compensation control simulation experiment was completed.According to the study of underwater weld recognition and path tracking,an underwater visual welding experimental platform was set up to test the visual identification error and the tracking error,and the error compensation experiment of the robot arm was carried out.The experimental results show that the integrated error of the compensated system is within the welding accuracy,and the vision-based underwater welding experiment was finally completed.