Research On Welding Seam Detection And Tracking Method Based On Machine Vision

Aiming at the problems existing in the welding process of the visual laser welding robot,this thesis studies the visual image acquisition method,image processing method,weld fringe extraction method,weld trajectory fitting method,and weld adaptive tracking control method,and does the simulation analysis.The main work contents are as follows:1.The composition structure of the laser vision seam tracking system platform is studied,Including camera model,lens structure,laser emitter and camera position relationship,laser triangulation principle.According to Zhang's calibration principle,the internal parameters of the camera were calibrated by MATLAB,use the direct calibration method to calibrate the laser plane,and use the linear solution method to calibrate the Eye-in-Hand system.2.Aiming at the characteristics of the collected laser weld images,a series of image processing methods are studied,including image filtering,image enhancement,binarization,and image morphological changes.A method for extracting weld seam ROI based on the image gray level projection method is proposed,which improves the image processing speed.The open operation is performed on the binarized image of the tower weld,which reduces the edge burrs of the laser stripes,and the closed operation is used to process the V-shaped binary image to reduce the cracks of the laser stripes.3.The welding seam detection process includes two steps: extraction of the center line of the laser weld fringe and extraction of the weld characteristics.The method of extracting the center line of laser welding seam stripes is studied.Sobel and other operators are used to detect the edge of image.The experiment proves that the operator can effectively detect the edge of laser fringe.The feature extraction of tower joint welds,Shi.Tomasi corner detection able to completely detect the weld position.4.Aiming at weld trajectory planning,the least square algorithm based on polynomial weld trajectory fitting and the Multi-Innovation least square algorithm are proposed.By expanding single newness into Multi-Innovation,the fitting accuracy is improved.B spline fitting methods including uniform B spline fitting curve,quasi-uniform B spline fitting and piecewise Bezier B spline fitting were proposed.The simulation results show that:(1)With the increase of weld position data points,the fitting data tends to the true value.(2)The effect of uniform B-spline fitting curve and quasi-uniform B-spline fitting curve is better than that of segment Bezier B-spline fitting curve5.The thesis put forward BP neural network PID control strategy,and MATLAB simulation is carried out according to the simplified model of the system.We can,from the simulation,conclude that the controller has good ability.