Research On Weld Recognition And Location Based On Vision Sensing

With the development of equipment manufacturing industry towards intelligent manufacturing in China,welding robot has become the main technical means of welding process.The welding robot mainly works in the way of teaching and reproduction,which has the disadvantages of fixed process and unable to correct the clamping trajectory deviation,processing error and thermal deformation in time.So the welding quality and efficiency was affected.Industrial robotics was integrated with image processing technology and deep learning technology,which makes the automation and intelligence of welding technology become the main development direction in the future.Therefore,the research on welding seam identification and location based on the combination of visual sensing and robot is of great significance.The robot vision system,the development of the PC software,vision system calibration integration,image processing,data calculation,data interactive visualization,robot motion control and so on were carried out in this paper.The line structured light and the industrial camera were formed into a vision system.The transformation matrix of the feature points from the pixel coordinates to the robot base coordinates was completed by calibration.The YOLO algorithm was used to locate the welding seam in the collected pictures.The pixel coordinates of feature points in the image were obtained by the method of weld feature point image recognition.the method of constructing a vector was used to determine the position of the welding torch during welding.The experiment was designed and completed,and the results of weld identification and location were verified and analyzed through the experiment.Firstly,the transformation relationship between the coordinate systems of each part of the welding robot was established,and the internal and external parameters of the camera were calibrated.According to the external parameter and image processing method,a structured light plane calibration method was proposed to obtain the light plane coefficient.The two-step method was used to solve the hand-eye calibration part,and its non-linear optimization was carried out to complete the conversion from pixel coordinates to base coordinates,and experimental verification is carried out.Secondly,a variety of image preprocessing methods were compared to preprocess the collected weld image,remove the noise to make the image clearer,highlight the key image features of the weld at the laser stripe.The YOLO algorithm was used to identify the area near the feature points of the welding seam,reduce the range of image processing,and improve the processing speed.The method of automatic threshold selection was used to extract the laser stripe image,and the sub-pixel algorithm was used to extract the center line of the laser stripe,the intersection point of center line was extracted as the feature point of the welding seam to complete the image processing of the welding feature point identification.Finally,for the V-shaped thin plate welding seam,the spatial vector was established to determine the robot pose information of feature points according to the spatial position relationship of laser fringes.The actual welding conditions were simulated to calculate the rotation angle of the welding position from the tool to the welding feature points.The off-line welding method was used as the base to design the overall experiment,and the error of the fitted welding trajectory was analyzed.The experiment shown that the welding seam identification and location method based on deep learning can meet the requirements of welding.It has a certain effect on the realization of welding automation process.