Research On The Acquisition Method Of Welding Path For Arc Welding Robot Based On Visual Servoing

At present, most welding robots used in practical production are teach-and-playback robots. They must be taught the welding job before welding workpiece. Teach operation will occupy many production time particularly in small batch or low volume manufacturing. Therefore, it is valuable that the robot can autonomously acquire welding path via visual sensor. It can also promote intelligentized level of the robot.In the previous studies on welding path acquisition, most vision systems work in open-loop mode, which usually leads to low precision and less reliability. In this dissertation, visual servoing approach was introduced into the research on welding path acquisition. The relevant technologies were deeply studied.A servo visual sensing device was designed when setting up the test system. This device allows camera to rotate about the torch by a desired angle with a servomotor, which can enhance the flexibility of controlling the capturing position of the camera and can avoid the problem of cable twist when tracking large angle seam. Additionally, it can automatically load or unload light filter to make the camera be suitable for the two different capturing cases: welding and before welding.When visual servoing approach is used to acquire seam coordinates, the anti-interference capacity of seam image processing algorithms is very important since image information is used to control the robot's motion in closed-loop mode. Therefore, this dissertation developed two set of noiseproof seam image processing algorithms: the region-based algorithms and the edge-based algorithms.The procedures of region-based seam image processing algorithms include median filtering, quasi-constant pixel number thresholding, thinning and size filtering. According to the feature that the number of seam pixels in each image can be thought of as a constant when tracking a seam, quasi-constant pixel number thresholding can avoid threshold too more or too less seam pixels. Size filtering further increases the anti-interference capacity of this set of algorithms. The region-based seam image processing algorithms can rapidly obtain the central line of the welding seam due to its low computation load.SUSAN algorithm is used as the edge detection algorithm in the edge-based seam image processing algorithms due to its advantages over Canny algorithm and other algorithms. According to the feature that the gray level values of the scratch pixels are obviously bigger than other pixels in a seam image, the scratch marks on the workpiece in the image are removed before edge detecting. The edge-based seam image processing algorithms can obtain the edges of the welding seam or the workpiece. So the seam's width can be achieved using this set of algorithms. In the thinned image, the cases often appear that there are branches on the useful lines and that noise line consist of many branches. It is difficult to extract accurate information form the useful line with branches and to remove the noise line comprising many braches via size filter. Aiming at solving these problems, this dissertation took the strategy that was firstly to cut the braches and then to remove them by using size filter, and proposed a cutting branch algorithm. The function of the algorithm is to break the lateral branch from the two main branches.The basic principle of the cutting branches algorithm can be interpreted as follow: Three included angles between two of the three straight lines along the braches are calculated in the 16-neighbors of the crotch pixel, which consist of a straight line included angles group. Then, the smallest element in the included angles group is found out and the order number of this element is used to determine the position of the branch which will be break from the other branches. In order to design the cutting branch algorithm better, the distribution structures of the branches were classified based on the definition of the straight line included angles group, and the solutions to special cases were discussed. Based on the prior knowledge such as the probability that the seam edge was horizontal line was great, the customized cutting branch algorithm was developed which was suitable for this application.In order to control the robot to move torch along the seam with the image information, the visual sensing system was calibrated firstly. A BP neural network was used to describe the mapping between the torch position in the image frame and the angle of the servo-motor, which solved the problem that it was difficult to detect the position of the torch in the seam image. Then, based on the defined seam position error and seam angle error, the seam tracking controller was designed. The controller combined feedforward control and a PD feedback controller. Aiming at solving the existed problem when acquiring the coordinates of the seam with curvature discontinuity, a feasible solution was put forward. During seam tracking, this dissertation corrected the current torch coordinates in the base frame and the angle of the servomotor in the visual sensing device with the seam errors to acquire more accurate seam coordinates and capturing position data of the camera.As a part of acquisition of welding path for lap joint, the method was developed that the torch and camera's orientations were automatically planned and adjusted based on the acquired seam coordinates and the required angles of torch. According to the features of welding technology, the geometrical contains of the torch were defined which composed of three parameters: weld direction angle, torch work angle and weld travel angle. The X-Y-Z fixed angles of tool orientation were calculated using these parameters. The orientation adjusting data from current tool orientation to target tool orientation were calculated used differential motion vector equation. After interpolation, the rotation angle increments were used to adjust tool orientation for the robot. Aiming at solving the problem that the errors increased with the larger orientation adjusting data, a feasible solution was put forward.The test results of acquiring welding path show that the developed visual servo system can acquire nearly all shapes of planar seam with butt joint or lap joint. The acquired seam coordinates is better than±0 .5mm, which can be directly used for high quality welding process. At the same time, the accuracy of the acquired 7th axis angle is better than±5°, which is sufficient for the image processing program to easily detect the seam and provides the convenience for welding quality control in the future. For the seam with lap joint, the torch orientation at the different seam positions can be automatically calculated and adjusted. Meanwhile, the optimal capturing position of the camera can be insured.