Research On Visual Inspection Methods For Commercial Vehicle Glass Gluing Quality

Commercial vehicle windshield gluing is an important process in the production of commercial vehicles.The overall quality and safety of commercial vehicles can be affected by the quality of gluing directly.Aiming at solving the problems of low precision and low efficiency of traditional gluing quality inspection,In this paper,a visual-based bonding quality detection method for windshield of commercial vehicles is studied which aims to provide vehicle manufactures with an efficient online gluing quality inspection method for commercial vehicle windshields,and helps these enterprises better control their gluing quality.In this paper,the background and significance of the subject are first described,and the status of the visual object detection technology at domestic and foreign is introduced,as well as the application of visual gluing detection technology in the automobile manufacturing industry.Then,the overall plan of the gluing quality inspection platform is determined according to the performance indicators and technical requirements of the platform and the key hardware equipment of the platform is selected,and a vision-based commercial vehicle glass gluing quality inspection platform is built.Next,the image processing algorithm is studied according to the characteristics of commercial vehicle windshield gluing.The ROI algorithm is used to extract the area of interest for gluing;the bilateral filtering method is used to make gluing images smooth;and the Canny edge detection algorithm is used to obtain the edge contour of glue strips.The method for measuring the diameter of the cross section of the glue strip based on the variable-step-length beetle algorithm is proposed according to the requirements for the accuracy and timeliness of the gluing detection platform.At last,in response to glue breakage,pop-up,overflow and other defects in the gluing practice,a YOLOv3-ciou model with k-means clustering method is proposed to optimize the prediction box for gluing defect detection based on the end-to-end YOLOv3 deep learning model,which solves the problem of low detection accuracy and low detection rate of the original model.The experimental results show that the H-VSBAS gluing width detection algorithm designed in this paper enjoys high accuracy and good real-time performance.The percentage of the gluing width detection error value being smaller than 0.3mm is up to 95%,meeting the requirements for 5±1mm glue width detected in the project.For the YOLOv3-CIOU gluing defect detection model proposed in this paper,93.48%,92.91% and 90.28% of the three defects,namely glue breakage,pop-up,overflow,are detected when IOU is larger than0.6.Thus,gluing defect detection can be better completed.