Research On Key Technique Of Automatic Welding Of Structure Based On Binocular Vision

With the continuous development of science and technology,welding methods have gradually transitioned from traditional manual welding to robot welding.At present,there are two methods in the automatic welding,which are manual teaching and off-line programming.The manual teaching that is a simple technique lacks intelligence,which is difficult to guarantee the welding quality and efficiency.The off-line programming comes over the shortcomings of manual teaching,but it is mainly applicable to the welding of small components.Because the off-line programming can't communicate with the outside world in real time,there will be problems that are mis-welding or lack-welding for the component which often occurs to deviate the position what we want.Therefore,it is necessary to develop an automatic welding system that can perceive and correct deviations without programming teaching.Hence,we designed an automatic welding solution for components,and studied on its key technologies.Firstly,in this thesis,we use the calibration toolbox to experiment,which obtains the camera's parameters and relative position matrix.We use SGM semiglobal stereo matching algorithm to find the disparity map.On time,we got the 3D point cloud data of the component.Then,In the thesis,a new Point Cloud Matching technology will be applied to make comparison between draft model and physical model.In this way,we can identify the position of artifacts in real time and then realize the pre-welding location and guidance for robots.in the process of Point Cloud Matching,we innovatively put forward a new matching method to better traditional matching method in which low accuracy and efficiency were always existed.And based on feature points,we also compare the new method with traditional PCA and SAC-IA through experiments,which turns out that the new method works better both in accuracy and efficiency.Finally,the ACA will be introduced to plan for 3D route for welding and re-define the output code of draft model,so that we short the robotic welding length and improve the robotic welding efficiency greatly.In this thesis,we used goose-neck-groove and goose-back-beam as experimental subjects to design and abstract the component's draft model and physical model to verify and analyse.The result shows that the point clouds in draft model and physical model are exactly overlapped,so that the coordinates can be unified and transformation matrix can be solved.