Online Detection Technology For Windshield Frit Band Defects

Automobile glass is an indispensable part of automobile body,and various problems will inevitably appear in the production process.For example,the pattern of frit band around the windshield will appear adhesion,incomplete and hiatus defects in the process of screen printing and sintering.The existence of these defects not only affects the service life of the glass,affect the driver’s perception.At present,the detection of frit band quality still relies heavily on manual work.This is mainly because the pattern is formed by a large number of tiny round or square black dots.The pattern is arranged in different ways for different models of cars,which leads to not only various patterns but also small defects.Therefore,it is difficult to realize online automatic detection and classification of frit band defects,which has become an urgent problem for enterprises to solve.In view of the above problems,this subject carries out research on the automatic detection method of three types of frit band defects common in automobile windshield,and the main work completed is as follows:（1）Design of automobile windshield frit band image acquisition system.Considering the large area of the automobile windshield,and the pattern of frit band is only distributed around the edge of the windshield,in order to avoid the high resolution industrial cameras to increase the hardware cost of the image acquisition system,this subject uses two 5 million pixel industrial cameras to collect the pattern of frit band in sections.Then,the contour curve is extracted from the collected pattern of frit band,and the contour curve is used to plan the moving position of the camera,so that the field of view of each camera can cover the area of black edge frit band to be photographed.The image acquisition system not only avoids the overlap between the pattern of the frit band and the image of the conveyor belt roller,but also ensures that the pattern of the frit band falls in the middle part of the image,which reduces the difficulty of subsequent image processing and improves the real-time detection of the flaw of the frit band.（2）The development of online detection algorithm for the defects of frit band.Aiming at the problems of small defects,various patterns and difficult detection,an improved CSi-YOLOv5s pattern defect detection algorithm is proposed in this paper.This defect detection algorithm is based on YOLOv5s model and integrates CBAM attention mechanism to obtain the attention weight in channel and spatial dimension of frit band feature map,so as to improve the relationship between various features.At the same time,by improving the activation function,the nonlinear combination performance is introduced into the algorithm to optimize the training effect of the deep network.The m AP value and t_ave value of the defect detection accuracy of the algorithm reached 98.9%and 0.134s for the three trained frit band patterns,and the average m AP value of the untrained several common frit band patterns reached 89.6%.The experimental data show that the improved CSi-YOLOv5s pattern defect detection algorithm has better detection accuracy and real-time performance.（3）The development of online monitoring software for defects.Aiming at the problems of low efficiency and chaotic data management in the manual detection of automobile glass frit band defects,the frit band defect monitoring software developed in this subject can realize the display of current glass detection information,query of historical glass detection information,fault alarm and diagnosis and other functions.The monitoring software not only realizes the fine management of detection data but also the visual control of production process data,which greatly improves the work efficiency.The research results of this subject can realize the online detection of the three types of defects of the frit band pattern of automobile glass,and the performance index is in line with the demand of the actual production line,and has a certain promotion significance for the development of automobile glass defect detection technology.