Study Of Autonomous Guidance Of Initial Welding Position Based On Vision Sensing

It is not only one of important steps of automatic welding process, but also of key technologies of realizing intelligent welding robot to recognize initial welding position and guide based on vision sensing with welding robots. It has practical significance to achieve the welding tasks in such environments as space, deep-sea and nuclear. In order to accomplish the actual project, two CCD cameras were fixed on the end-effecter of robot for the purpose of imitating welder to observe the welding process. The method of one CCD camera and moving two different positions which formed binocular stereo vision was successfully used to capture and recognize the initial welding position so as to achieve the guidance, which may be a foundation of realizing the automatic welding for welding robots.The precision of recognition and guidance of initial welding position based on vision sensing is directly determined by the structure of the sensor and calibration results. A new vision sensor which can change and adjust its function based on modular thoughts was developed. Meanwhile, improved calibration pattern marked with dots and blocks was adopted which makes camera calibration and hand-eye calibration simultaneous and much easier.According to the project demands and workpiece features, a set of algorithm suitable for recognition of initial welding position of flat butt joint, flat flange and space curve seam with marks was developed. Started with global image process, including smoothing, adaptive threshold segmentation, edge detection and other steps, then located the object region which contains the initial welding position so as to extract the feature points, i.e. initial welding positions, by using improved Harris algorithm or grey scale scanning method which is called local image processing. The algorithm offers multi-path and multi-method so that it can reduce the feature information annihilation caused by global information, also improve the speed and accuracy of image recognition.The feature points in the pair of images captured by binocular stereo vision configured by the two positions movement of one camera matching follows epipolar constraints. This dissertation also substitutes the midpoint of the both vertical line of two radials for the intersection point in case of the disjoint conjugate points. Thus, three dimensional coordinate read directly by robot can be calculated by two dimensional coordinate through stereo vision principle and coordinate transformation.A set of software system for guidance of initial welding position was developed to meet the project demands. The precision, that is mean square error, about guidance of flat butt joint in X, Y, Z direction is 0.4491, 0.8187, 1.4794mm and 0.5398,0.4260,1.2596mm for flat flange with improved marks. On the analysis of influencing factors about the recognition and guidance of initial welding position, precision of image processing, hand-eye calibration results and movement disparity are the most important factors. At last, guidance experiments on simulating parts were presented whose success rate is up to 98% and precision meets the project demands.