Research On Laser Vision Multi-layer Multi-pass Welding Seam Tracking Based On Deep Learning

Multi-layer and multi-channel welding method is widely used in aerospace,shipbuilding and other industrial fields for joining medium and thick plate,which is important joining technology.And due to its great advantages of high precision,strong anti-interference ability and low price,the robotic seam welding is becoming the most popular method based on laser vision.However,the seam pictures obtained by visual sensor will inevitably be polluted by strong reflection,spatter and arc noise when performing the actual seam tracking,which make it difficult to guarantee the accuracy and stability of welding.At the same time,seam planning in multilayer and multi-pass welding is uncertain.Therefore,it is significant to study the tracking of multilayer and multichannel welds based on laser vision.The main research contents are listed as follows:Firstly,a multi-layer and multi-channel welding seam tracking platform for the welding robot was built for experimental research.In order to avoid arc light spatter in welding process which has great influence on image quality,the related components and installation methods of laser vision system are selected and designed.The calibration principles of camera,hand eye and laser plane in the laser vision system are analyzed,and the parameters needed to realize the transformation from 2D coordinates to 3D coordinates are solved by the proposed calibration method.Secondly,an autonomous identification method of multi-layer and multi-pass welds based on laser fringe edge guidance network is proposed.In this method,the global features of weld images are extracted by the improved VGG main network,then the guidance module in the network is used to fuse the laser fringe information obtained by the multi-scale extraction module and the laser fringe edge extraction module.thus,the laser fringe image of weld seam with clearer and more accurate boundary can be obtained in the multi-feature fusion output module for subsequent processing.Then,the automatic extraction of initial feature points and the continuous tracking algorithm of feature points of welds are studied.According to the different characteristics of V-groove single weld and multi-layer multi-weld,the random sampling consistent and non-uniform rational B-spline fitting methods were proposed respectively to extract the initial feature points of the weld.At the same time,a continuous weld tracking algorithm based on deep regression network is proposed to continuously detect the position of weld feature points.Finally,the multi-layer and multi-channel weld tracking experiment is studied.In order to realize multi-layer and multi-pass automatic welding of V-groove workpiece of medium thick plate,the cross section of weld pass and welding torch attitude were planned.The communication system of welding seam tracking is designed to ensure the smooth progress of automatic welding seam tracking.Finally,a multi-layer and multi-channel weld tracking experiment was carried out on the plane V-groove workpiece,and the error analysis was carried out on the collected data,thus proving the accuracy and stability of the algorithm and the weld tracking system proposed in this paper.