Research On Vision-based Seam Tracking System

Since the beginning of the twentieth century,the automation level of China's manufacturing industry has been continuously improved,but the localization of its high-end welding automation products is under increasingly unsound development.At present,most welding automation equipment still uses manual teaching to plan the robot's motion trajectory,which meets high requirements for the accuracy of workpiece installation and positioning.The welding accuracy depends entirely on the accuracy of manual teaching.Lacking detection and feedback functions for the weld makes the welding gun unadjustable according to the thermal deformation of the welding workpiece,and makes it difficult to guarantee the quality of the weld.Therefore,it is urgent to study the weld tracking system for real-time detection and correction of the welding position.In this paper,we design a vision-based weld tracking sensor based on the problems of current weld tracking,and conduct in-depth analysis and research on its key technical problems.In view of different welds which use different identification methods,a back propagation neural network is proposed to identify weld types.In detail,the images of flat lap weld,V-groove weld and right-angle weld are used as training sets to train the neural network model.And the neural network model reads the image of weld tracking vision sensor,distinguishes the current weld type,and supports weld tracking and positioning.Aiming at the fact that the weld feature image is affected by arc light and spatter,the characteristics of arc wavelength in the welding process are analyzed.Specifically,the laser and filter corresponding to the wavelength band with insignificant wavelength characteristics are selected,reducing the influence of the welding arc on the image of weld feature.And aiming at the deviation of the weld feature from the preset ROI area due to the offset of the welding position during the tracking process,a method which extracts the feature image of the weld from the floating ROI area is proposed to provide a stable feature image of the weld for the weld tracking.Aiming at the fact that the data fluctuation of welding position during the weld tracking process causes frequent adjustment of the welding torch position and affects the weld quality,a method to reduce the position data fluctuation by data filtering is proposed.By analyzing the data filtering effects of median filtering and Kalman filtering,a combined filtering method by combining median filtering and Kalman filtering is proposed in order to obtain better data filtering effects.Based on the research on the related technology of weld visual tracking,a weld tracking test platform is built in the laboratory,and simulated weld tracking tests are conducted for the weld visual tracking system.The test results show that the tracking accuracy of the vision-based weld tracking system for flat lap welds,V-groove welds and right-angle welds is within the allowable range of welding errors,and on-site welding tests have been completed accordingly.