Research On Welding Seam Tracking In Fusion Welding And Additive Process Based On Feature Segmentation

With the rapid development of manufacturing industry and science and technology,in the field of fusion welding and additive manufacturing,traditional manual welding has been unable to meet the needs of the industry for mass production,so an automated welding scheme using robots instead of manual welding is needed.In intelligent welding,with the complex and changeable welding mode and welding environment,the weld tracking scheme using traditional image processing algorithm has been difficult to achieve the ideal effect.In this paper,welding seam tracking technology in additive manufacturing is studied,and image segmentation algorithm based on deep learning is applied to intelligent industrial welding field to realize accurate extraction of welding seam characteristics.The main research contents of this paper are as follows:(1)Referring to active vision technology,the pusher and sweep weld tracking system based on line laser is established.Through the laser line calibration of the vision system and the handeye calibration to determine the relative position relationship between the vision system and the robot,the vision system can accurately obtain the required weld features in the welding process,providing reliable data support for the realization of real-time weld tracking.In this paper,the advance distance of the vision system relative to the welding robot gun is determined through experiments,which provides an important reference for the on-line correction of welding path.(2)Aiming at groove filling in additive manufacturing,ERFNet is applied to weld feature extraction algorithm in this paper.Due to the large amount of background noise in the actual welding process will drown the laser fringe,as well as the complex laser fringe contour in additive welding,traditional image processing algorithms cannot obtain accurate weld feature information from the images collected by the visual system.Therefore,the image segmentation based on deep learning is introduced into the weld feature extraction algorithm in this paper to realize the complex laser fringe center line extraction and feature point detection in additive welding.Different from the traditional scheme in which the centerline is extracted from the centerline and then the feature points are solved,this algorithm can obtain the two types of features needed for weld tracking at the same time,and the reliability of the algorithm is verified by the weld feature extraction experiment.Finally,the weld tracking scheme proposed in this paper is proved to have good applicability in groove filling task through additive welding experiment.(3)Aiming at the task of rapid prototyping solid parts in additive manufacturing,this paper optimizes and improves ERFNet and applies it to weld feature extraction algorithm.Compared with the groove filling task,the number of welding layers and welding paths in the task of rapid prototyping solid parts are more,and the laser fringe contour is more complex.Therefore,the application of ERFNet directly in the feature extraction algorithm cannot achieve the ideal segmentation accuracy.In order to meet the high precision requirements of additive manufacturing,the structure and loss function of ERFNet were optimized and improved in this paper,and the multi-scale feature fusion strategy and Focal Loss were introduced into ERFNet.The experiment of weld feature extraction verifies that the improved scheme proposed in this paper can improve the accuracy of the algorithm and ensure that the efficiency of the algorithm is not affected.Finally,the improved welding seam tracking scheme is proved to have high reliability in the task of rapid prototyping solid parts through the plate additive welding experiment.