Research On Key Technologies Of Automatic Teaching Of Welding Robot Based On Vision Feedback

Robotic welding technologies have been widely used in the domain of manufactures, which boosts the great advance of the automation of welding industries. Path planning is one of key phases of robotic welding, it can be classified into: manual teaching, off-line programming and automatic teaching. Off-line programming is suitable for structured welding environment, manual teaching still dominates robotic welding industries. But manual teaching is both time-consuming and difficult to satisfy the welding accuracy requirement for 3-dimensional (3-D) seam with complex trajectory, the automatic teaching of vision-guided welding robot under the control of computer has become a developing trend instead of manual teaching. However some key technology problems such as automatic welding target recognition, starting point location of welding robot, stereovision-based measurement of seam and its surroundings and so on have become choke points, in order to facilitate the automatic teaching of welding robot, these key technologies have been studied in the dissertation, and the related researches and results are as follows:(1) A sequence similarity detection algorithm (SSDA) guided by wavelet singularity and a rapid correlation algorithm for gray-level template matching are presented to detect and locate welding targets with different sizes respectively. A normalized singular value decomposition of image is proposed to extract the features of welding target to eliminate the effect of terget orientation, and the stable features are used for automatic recognition of welding targets. A combined BP neural network classifier is constructed and the network outputs are fused together with Dempster-Shafer theory so that the accuracy of welding target recognition can be improved.(2) A fundamental matrix expression and epipolar constraint equation that fit into welding robot vision system with"eye-in-hand"configuration are established, a constraint equation with parallel epipolar lines is derived from epipolar rectification principle, and using it the stereovision-based detection and reconstruction of both welding targets and their surroundings are fulfilled according to edge-sparse-point or region-dense-point matching criteria, the depth distribution in front of camera is given from the dense disparity graph. A stereo-matching technique that combines epipolar constraint with laser stripe indication is applied to reduce the search space from a line with one dimension to 1 or 2 points, and both matching efficiency and matching accuracy are boosted because of eliminating uncertainty. Saddle-shaped seams and their surroundings are accurately measured using the proposed matching technique, and their 3-dimensional surfaces are reconstructed using Delaunay triangulation, so that the structure information of welding environment is generated.(3) A estimation modelling of image Jacobian based on support vector regression(SVR) machine is presented for non-linear mapping between target image features and robot joint angles in robotic visual servoing, and an expression of SVR-Jacobian estimator is derived from SVR equation with Gaussian kernel, and the experiments of robotic visual servoing with both eye-in-hand and eye-to-hand camera configuration are conducted using the SVR-Jacobian estimator, two kinds of experimental results have shown that the robotic visual servoing converges at the desired goal with high target-tracking accuracy, the robotic visual servoing quality based on the SVR estimator is better than conventional Broyden-estimator, and it can be used for robot guidance to the starting point of seam welding.(4) A posture computation modelling of a 6-degree of freedom(DOF) robot is completed using Denavit-Harbenterg representation. A controlling strategy for the automatic teaching of 6-DOF welding robot is presented based on ant colony optimization(ACO) algorithm, in which the angle increment of robot joint is discretized as the nodes of ACO graph and a corresponding pheromone updating strategy based on robot posture error and heuristic value computation method allowing for obstacle effect are given. The automatic teaching of the robot is completed for both spacial saddle-shaped seams and planar curvilinear seams using the proposed controlling strategy and taking the seam posture generated by stereovision measurement as a controlling goal, and experimental results show that the teaching trajectories generated by ACO-based controlling strategy are better than ones by conventional controlling strategy. The robotic arc welding experiments of saddle-shaped seams are conducted using these teaching trajectories as welding paths, and the good quality of welded seam is achieved.