Application Research On Machine Vision For Weld Seam Extraction

The development of advanced manufacturing has made new requirements on the intelligence of welding technology continuously.The conventional welding robot in the way of teaching and playback supersedes labour in high strength,high precision welding field.Machine vision technology is used to overcome the limitations of robot which using teach and playback method that can’t correct errors online.The actual weld image has a variety of sources of noise,especially the interference of pre-welding points.The diversity of the weld shape also increases the difficulty of the weld extraction algorithm.The difficulty of the application of machine vision technology in weld seam extraction lies mainly in the real-time,reliability and accuracy of the algorithm.To solve the problems above,this paper builds a machine vision system to acquire the weld image and performs its camera calibration.In the weld image preprocessing stage,this paper proposes a weld area segmentation method that based on the gray scale distribution and local fractal feature of the weld area to extract the weld area and reduce the processing time of the subsequent real-time algorithm.In the real-time processing stage,this paper proposes a weld area enhancement algorithm based on energy adjustment to preprocess weld image.In this algorithm,the low frequency enhancement uses the guided filter to suppress the texture noise in non-weld areas while preserving the edge of weld seam.The high frequency enhancement enhances the edges of the weld seam by enhancing the vertical high frequency coefficient by the piecewise linear threshold method proposed in this paper according to the wavelet decomposition feature of the weld image.The energy adjustment is achieved by the adaptive Gamma correction to prevent the image gray value from changing too large resulting in the shift of the weld seam.The experimental results smooth the non-weld area of the image excellently and enhance the edge of the weld seam,laying the foundation for better weld extraction.Aiming at the problems of conventional Canny edge detection algorithm in fuzzy edge and gradient threshold selection.This paper proposes a method to improve the conventional Canny edge detection algorithm by combining the preprocessing algorithm with the high and low threshold selection method based on the slope of the gradient mean histogram.The improved algorithm has better weld edge integrity in the edge detection experiment.In this paper,the weld seam extraction algorithm based on neighborhood is proposed to eliminate the missing detection,false detection and double edge.This algorithm uses cubic spline interpolation to eliminate the problem of missing detection.The real-time processing stage of this paper takes about 21.7ms,which satisfies the real-time requirement.The experimental results show that the algorithm has a reliable weld extraction effect for S-type and non-continuous perforated,straight type weld seam,and has high positioning accuracy.