Research On Welding Robot Visual Servo Control

With the rapid development of society and technology, high-precision, high-quality,high-efficiency products become inevitable, but the key technology of connecting thevarious parts, welding, is also gradually gaining importance. As a result of the limit ofonce forming in capacity technology and cost, welding has become a significantmanufacturing part since it was produced. Traditional human welding has been unable tomeet the increasing needs of accuracy and efficiency, welding automation as well as otherindustries has become an inevitable trend of the industry. Along with the ever changing inrobots, computers, sensors and image processing, people began to focus on researchingrobot with visual function, and modern welding techniques are increasingly automatedand intelligent. How to quickly obtain high-definition images which can reflects weldfeatures, obtain the welding seam information more rapid and reliable and obtain thedeviation between torch and seam center more precise are the urgent need in the field ofwelding automation project in the automatic welding process. In this paper, a specialstructure binocular stereo vision, combined with structural light and image processingtechnology is used to study the application of robotic welding. The main research work issummarized as follows:1. DH parameter model is established according to the laboratory existingMOTOMAN UP6robot movement mechanism and linkage parameters, and calculate thetransformation matrix between the various joint coordinate system, and verified usesimulation with robotics toolbox for MATLAB written by Queensland University ofTechnology Professor Peter Corke.2. Study binocular stereo vision based on arc welding robot welds image processing,image processing section for structure light based robot seam tracking system designedsimple fast image processing algorithms.3. Aiming at the main problem such as interference arc and spatter which can makeband broke in image processing and cannot determine the position, propose a localexposure method only the structure light can normal exposure and can accurately extract the structure light position by using the idea of exposure in taking photos, according tothe laser wavelength and intensity of illumination. And then extract the feature points ofweld using Hough transform process the image.4. Establish the model of the binocular vision and derive Jacobin matrix, using theprinciples of image-based visual servo mode and error exponentially converge on thedesign of the controller so that the torch can quickly and smoothly welds starting point,and the use of neural networks combine the genetic algorithms method optimize thecontroller parameters and use MATLAB simulation proved the rationality and feasibilityof optimization.