Research On On-Line Defect Feature Detection Technology Of Aluminum Body Based On Machine Vision

With the development of automobile industry,aluminum alloy and other materials with low quality and low cost are widely used in body production.The traditional spot welding,MIG welding and other thermal connection technologies are not suitable for all aluminum body,so riveting technology is more and more widely used in the automotive industry.However,the current detection of riveting quality of automobile body is mostly manual detection,which has the problems of low detection efficiency and high misjudgment rate.Therefore,this paper mainly uses machine vision technology to detect the riveted joint of aluminum body.This paper designs the attitude defect detection scheme of aluminum body riveted point based on improved template matching algorithm and improved edge detection algorithm,and the surface defect detection scheme of aluminum body riveted point based on deep learning.Firstly,this paper introduces the riveting process and riveting defect characteristics of aluminum body and gives the basis for selecting visual inspection.Then the hardware system of vision on-line detection is designed,and the specific model and performance parameters are given for the necessary hardware structure.The radial single point detection strategy is selected to complete the image acquisition,and the detection algorithm scheme is designed according to the characteristics of two kinds of defects.The detection results of the two algorithms are integrated to determine the riveting quality of aluminum body.For the attitude defect of aluminum body riveting point,the detection content is completed from four aspects: system calibration,template matching,edge detection and contour fitting.In the system calibration,Zhang Zhengyou calibration method is used to complete the camera calibration and distortion correction;In template matching,an NCC algorithm based on Gaussian pyramid search is proposed,which can locate the detection area quickly and accurately;In edge detection,an adaptive Canny algorithm is designed.Adaptive median filter is used instead of Gaussian filter,and Otsu algorithm is used to adaptively divide high and low thresholds;In contour fitting,the least square ellipse fitting algorithm is used to fit the edge of riveting point,and the long and short axis values and attitude values are output.Finally,combined with the quantitative parameters of riveting attitude,the detection of riveting point attitude characteristics in riveting quality judgment is completed.Faster R-CNN algorithm is used to detect the surface defects of riveting points on aluminum body.Aiming at the problem of insufficient capacity of rivet point surface defect data set,the data samples are expanded by using translation,rotation and scaling,and the defect samples are marked by MATLAB software.The map and detection time are selected as the final detection standard of body rivet point surface defects.Finally,the three rivet point surface defects of crack,pit and notch are trained and learned.The final map value can reach 0.843,and the detection time of single image is 0.77 s,which meets the requirements of online detection.