Research On Intersecting Weld Laser Tracking Based On Deep Learnin

Pipe weldments are widely used in modern industrial manufacturing and nation-al pillar industries.The intersecting pipeline,which is an important part of pip-eline weldment,has also been widely used.However,it is difficult to locate the welding seam and plan the welding gun posture during the welding of space intersecting pipeline.It still relies on manual welding,traditional robot teaching reproduction and off-line programming.The research on automatic and intelligent welding of intersecting welds has great scientific research and engineering application value.Moreover,in the actual weld tracking,the weld image obtained by structured light stereo vision sensor will inevitably be polluted by strong reflection,spatter and arc noise,resulting in the inability to ensure the stability and accuracy of welding.Traditional algorithms can not adapt to the problems caused by complex welding environment.Therefore,in order to improve the adaptability and anti-interference ability of the intersecting weld tracking system,this paper proposes to combine the depth learning with the intersecting weld tracking to study the intersecting weld tracking system.The main contents of this paper are as follows:Firstly,aiming at the interference of spatter,arc and other noise in the process of circumferential welding of intersecting welds,a intersecting weld recognition algorithm based on deep learning semantic segmentation is proposed.The algorithm extracts the features of the laser stripe image of the intersecting weld through the full convolution neural network as the backbone network,and then uses the proposed joint up sampling module to carry out multi-scale fusion up sampling on the extracted features,and finally obtains the laser stripe image of the intersecting weld without noise interference with the same size as the original image.Secondly,aiming at the problem that the spatial position of the weld changes at any time in the process of intersecting weld tracking,a intersecting weld tracking algorithm based on convolution neural network feature region recognition is proposed.The algorithm consists of two parts: query guidance region recommendation network module and query guidance region convolution neural network module.Firstly,the weld feature point position of the first image in the laser stripe image of the intersecting weld collected in the whole welding process is manually marked as the query image,and the other images are used as the search image after being processed by the intersecting weld recognition network,and then input into the network model for prediction,so as to obtain the weld feature point position of the search image.Finally,the intersecting weld tracking system is verified by experiments.Aiming at the problem of locating the feature points of intersecting welds,the structured light stereo vision sensor system is calibrated,and the transformation relationship between two-dimensional image coordinates and three-dimensional actual coordinates is obtained.Aiming at the difficulty of tracking the welding process caused by the intersecting weld is a spatial special-shaped curved surface weld,the posture control and path planning scheme of the welding gun are proposed to ensure the stability of the welding process.The welding experimental results are analyzed and verified,and it is proved that the intersecting weld tracking welding system has stability and accuracy.