Research On K-TIG Welding Seam Tracking System Based On Mask-RCNN Model

With the rapid development of modern manufacturing industry,welding as an important processing technology in manufacturing industry plays an increasingly important role.K-TIG welding is an efficient deep penetration welding method,which can weld medium and deeply plate without groove.It has high welding efficiency.Therefore,it is very important to realize the automation of K-TIG welding process.Aiming at the characteristics of strong arc and small gap in k-tig welding process,this paper studies the k-tig seam tracking process,which can lay a solid foundation for the automation of k-tig welding application,and has important research significance.In this paper,a K-TIG welding seam tracking platform is built,in which the vision system uses HDR camera with high dynamic visual range to obtain high brightness arc area and low brightness welding seam area at the same time.The traditional calibration method is used to calibrate the HDR camera.The conversion relationship between the actual distance and pixel distance in HDR camera is fitted by using the least square method through the calibration experiment.It lays a foundation for the acquisition of welding deviation of K-TIG.In order to extract the keyhole entrance from the K-TIG welding image,this paper uses mask RCNN model to identify the keyhole entrance accurately.Based on this,image processing algorithms such as HSV color space conversion,bilateral filtering,image binarization,keyhole removal and keyhole entrance center extraction are used to extract the keyhole entrance center of k-tig welding Wave,image binarization,Laplace edge detection,Hough line detection and other image processing methods are used to extract the weld centerline,and finally the distance between the entrance center of k-tig keyhole and the weld centerline of k-tig welding is the detected welding deviation.Finally,welding experiments show that the accuracy of the algorithm is ±0.226 mm.In order to make the welding deviation detection algorithm based on Mask-RCNN meet the real-time requirements of weld seam tracking,this paper deploys the deviation detection algorithm to the second generation neural network calculation bar through Open VINO for reasoning acceleration.After acceleration,the average reasoning speed of the algorithm reaches 15.65 fps,which basically meets the real-time requirements of K-TIG weld seam tracking.In order to verify the tracking accuracy of the welding seam tracking system based on Mask-RCNN proposed in this paper,two kinds of welding seam tracking experiments under different working conditions are designed.The tracking accuracy of the system is ±0.218)8)8)8)in the plane straight line welding seam tracking experiment,and ±1.398)8)8)8)in the plane curve welding seam tracking experiment,which basically meets the requirements of K-TIG welding seam tracking accuracy.