Research On Initial Weld Position Guidance And 3D Reconstruction Of Weld Feature Based On Single Vision

With the development of intelligent manufacturing technology,it shows an inevitable trend that robotic welding will be the future technology development in the industry.Visual sensor technology plays an indispensable role in robotic welding,especially in the aspects of robot initial weld position guidance,weld tracking,molten pool monitoring and so on.At present,most welding robots are in teach-play mode which needs to be accurately taught and spends a lot of time when there is welding task to be completed.To some extent,it has low efficiency and poor adaptability for complex weld programming.It is the key step of the intelligent robotic welding to guide the robot to the initial welding position before welding and control the movement of the robot to finish the welding seam identification and path planning.In this paper,a set of intelligent welding robot system with vision sensing and laser measurement is set up based on the modularization idea including hardware system and software system.The hardware system mainly includes welding robot module that mainly based on FANUC sixaxis welding robot and self-developed laser monocular vision sensing Module.The software system consists of three parts including the calibration of laser structured light parameters,the initial welding position identification and three-dimensional reconstruction and the welding seam identification.As is known,the accuracy of initial welding position guidance and weld identification based on visual sensing mainly depends upon the factors such as system calibration,feature point extraction,stereo vision matching and so on.In this paper,a custom dot-shaped array calibration target with the precision of 4um,the spot diameter of 2mm and the distance between centers of 4mm was used to obtain the internal parameters of the camera first and then the hand-eye relation matrix based on MATLAB calibration toolbox.At the same time,a rectangular target was used to calibrate the normal direction of line-structured light plane first and then the distance information from the optical center to the light plane based on the principle of cross-ratio invariance and P3 P three-point perspective model.The calibration software was developed by C++ language.Taking the flat butt welds as an example,the image coordinates of the initial welding position were obtained based on passive vision sensing.A set of two-step method and dynamic variable region template matching method are proposed to identify the weld centerline and a set of image processing algorithms are designed for the local area of the weldment to obtain the image coordinates of the initial welding position feature point which can achieve sub-pixel level.The passive stereoscopic vision model was established by the format of single vision and two positions and the feature points are matched by epipolar line constraints and image feature constraints.Then the 3D coordinate of the target point in the tool coordinate system is calculated by combining the result of hand-eye calibration and the robot pose status.Finally,send dimensional coordinates to the FANUC robot to achieve the autonomous guidance of the robot work.In this paper,a total of 10 sets of initial welding position guidance experiments were carried out at different heights of the camera.In these experiments,the distance from the tip of the wire to the surface of the workpiece was 15 mm to 60 mm.Correspondingly,the distance from the optical center of the camera to the surface of the workpiece was 160 mm ~ 180 mm.The experimental results show that the overall guidance accuracy is within ± 1mm,where e(X)?0.3mm,e(Y)?0.5mm and e(Z)?1mm,which meets the requirements of actual welding.In addition,a stereo vision model based on laser monocular vision is established according to the calibration results of line-structured light parameters.For the polyline and S-shaped welds,the laser stripe images sequence was acquired by laser scanning.For the fringe images,the laser stripe centerline is extracted by the projection method,and the edge feature points of the weld are located.Then,the 3D reconstruction of the weld information is completed by combining the results of line structure optical calibration,image picking period and robot moving speed.The experimental results show that the welding seam recognition is accurate,and it can be used to detect and track the weld before welding which can replace the trajectory teaching of the robot before welding and thus realizing the autonomous welding of the robot.