Ring-shaped Automotive Steel Parts Defect Detection System

At present,most of the mechanical processing manufacturers in my country still use manual methods to inspect the quality of annular automobile steel parts,which have poor inspection efficiency,low precision,and large interference from human factors.In order to efficiently complete the defect detection task of annular automobile steel parts,this paper uses machine vision technology to detect and study the dimensional defects and surface defects of annular automobile steel parts,and designs and realizes the defect detection of annular automobile steel parts.The system has practical engineering application value.The key contents of this article are as follows:(1)Using visual measurement technology to detect the dimensional defects of annular automobile steel parts.On the basis of image preprocessing,the sub-pixel edge detection method of coarse and fine localization cascade is adopted comprehensively.Among them,improvements are made to the shortcomings of the Zernike moment method for sub-pixel edge detection: 1)By adding sub-pixel edge point judgment conditions to solve the problem of the accuracy error of Zernike moment odd-sized templates on the x-axis;2)Combined with Otsu method to achieve Zernike moment grayscale The optimal threshold of the step is automatically selected,which improves the detection efficiency.On the basis of edge detection,the four dimensional parameters of the inner and outer circle radius and the inner and outer circle roundness of the part are measured,and the dimensional defect of the part is determined according to the measurement error.Using the idea of comparative research,this paper compares various existing algorithms and improved parts size measurement methods in many aspects.The experimental results show that the improved method in this paper has smaller measurement errors and higher accuracy,which can meet the requirements of industrial parts size defect detection.Require.(2)Using deep learning technology to detect surface defects of annular automobile steel parts.According to the surface defect characteristics of annular automobile steel parts,and by improving YOLOV4,a surface defect detection algorithm for annular automobile steel parts based on SE-R-YOLOV4 is proposed.The improvement points are as follows: 1)Cluster the surface defect data set of steel parts by the weighted Kmeans algorithm,and obtain the anchor pre-selection box that matches the sample better,so as to solve the problem of unbalanced defect samples.2)The SE module is introduced into the residual unit in the YOLOV4 backbone network,and more attention is paid to the channel containing a lot of information to improve the accuracy.3)Connect the RFB-s module after the feature map with a size of 76×76 output by the network to increase the receptive field,expand the range of information extraction,improve the resolution,and extract more detailed information.In the comparative experiments with multiple target detection algorithms,the improved SE-R-YOLOV4 algorithm achieves m AP50 of 90.5% on the surface defect data set of circular automobile steel parts,FPS is 53.1 frames per second,and the overall performance is better.It can meet the real-time requirements of industrial production.Through the transfer experiments on the MS COCO dataset,it is shown that the improved algorithm can handle different datasets and perform multi-classification tasks,which verifies its transferability and generality.(3)A defect detection system for circular auto steel parts is designed and implemented,which can complete the dimensional defect detection and surface defect detection tasks of circular auto steel parts with one click,and is initially applied to industrial detection.