| The brake disc is the core component of the automobile brake system.It realizes the deceleration and braking of the vehicle through the friction between the material attached to the surface of the brake disc and the brake lining.In industrial production,most of the friction materials are bonded through the surface coating process at present.If the gluing does not meet the technical specifications,it will bury potential safety hazards for the subsequent adhesion of friction materials.Therefore,the quality of glue directly affects the braking performance of vehicles and driving safety.It is very important to check the unqualified brake disc in time.Most of the inspection methods in factories use manual visual inspection,which has the disadvantages of high cost,secondary contact and inconsistent inspection standards.Therefore,the subject research is based on multi-eye vision to automatically detect the surface coating defects of brake discs.The paper mainly completes the following tasks:First of all,the detection algorithm of the glue defect is studied: for the position of the brake disc,the target area location based on the contour boundary tracking and Ransac method is realized;for the phenomenon of uneven illumination,a method for selecting the partition adaptive gray threshold is proposed.Combined with edge features for defect area segmentation;for multi-camera systems,the calibration based on homography transformation is realized,and a defect recognition method based on weight fusion and filling is proposed to filter out the influence of high reflection on inspection and detection..In the end,the false detection rate of the multi-eye vision detection algorithm for bubbles is0.54%,the false detection rate for uneven thin and thick areas is 0.13%,and the missed detection rate for defects is 0.17%,which meets the detection standards.Secondly,based on deep learning to study the semantic segmentation of glue defects,and create a multi-camera image data set based on polar coordinates and Cartesian coordinate transformation;at the same time,the U-Net is improved,and the grouping convolution and channel attention mechanism are integrated.Compared with the traditional U-Net based on single-camera input,the semantic segmentation performance of the improved multi-camera input network model is improved from 0.570 to 0.869 in the Io U.Finally,according to the glue acceptance standard,this study designed a set of detection system scheme based on multi-vision,which includes the selection and structure construction of hardware such as light source and camera,modular design of user operation interface and realization of each part,and finally embedded glue defect detection algorithm based on gray feature analysis.The device has been installed on the production line and put into actual brake disc coating production testing. |
