| With the rapid development and wide application of laser welding technology,more and more attention has been paid to the research on welding quality inspection methods.In order to solve the problems related to the detection efficiency,detection accuracy and detection rate of laser weld defects in stainless steel sheet,this paper studies the visual detection technology of linear structured light,and carries out the following research work to realize the automatic extraction and recognition of laser weld defects:(1)Based on the principle of line structured light visual inspection and the characteristics of laser welding seam of stainless steel sheet,a set of line structured light visual inspection system was designed,including machine vision system and software system.The machine vision system is composed of a pair of 3D laser cameras and a pair of linear robots.The 3D laser camera is arranged on the translating platform of the linear robot to realize the high-precision acquisition of welding seams on the front and back sides of stainless steel plates.The software system mainly integrates the function modules of mechanical motion and control,image acquisition,automatic defect extraction,automatic defect recognition,detection result preservation and so on.(2)The problem of initial laser weld depth image is analyzed and the necessity of image correlation pretreatment is clarified.Firstly,Canny edge detection method was used to extract the upper and lower edges of the plate,and the area over the initial image was eliminated to achieve the segmentation of stainless steel plate area.CLAHE image enhancement method is used to enhance the contrast of the plate image,which is convenient for inspectors to observe the overall morphology and quality of laser welding seam.Finally,the improved algorithm of fast mean filtering based on integral graph and the linear fitting algorithm based on Hough change were used to extract the weld zone.(3)The automatic segmentation method of laser weld surface defects and the calculation and description method of defect related features are studied and designed.Firstly,an improved algorithm of fast mean filtering based on integral graph is used to construct the ideal weld.Then subtracting with the original image,extracting the suspected defect regions exceeding the gray difference threshold;Then,the effects of filter size and gray threshold on defect segmentation were analyzed,and the optimal detection parameters were determined.Finally,the 12 features of defects are extracted and analyzed.(4)An automatic recognition algorithm for laser weld surface defects is designed and developed.Firstly,qualitative analysis was made on six kinds of defects,such as welding penetration,welding bump,staggered edge,incomplete welding,broken welding and edge bite.Then,a decision binary tree-logistic regression recognition model is designed,and 353 defect samples are used to train the recognition model.Finally,143 samples out of 155 defect prediction samples were correctly predicted,so the defect identification accuracy of the model reached 92.3%.(5)The high efficiency,high precision and automatic line structured light visual inspection system is built and integrated.The accuracy of the system in X,Y and Z directions is 0.0182 mm,0.0234 mm and 0.0028 mm,respectively.By comparing with the defect size measurement of DR digital radiographic detection technology,it is found that the detection results of the two technologies are relatively consistent.Moreover,the visual detection technology of line structured light has more advantages in the detection of defects such as incomplete welding,edge biting,broken welding and staggered edge,and the measurement errors of defect size are less than 0.1mm in both X and Y directions.Finally,the overall detection capability of the system is summarized.The detection speed of the system is 150mm/s,and the total time of image preprocessing,automatic extraction and recognition of defects is less than 10 s. |
PDF Full Text Request