Brake Disc Defect Detection And Sorting System

Remanufacturing is an effective way to develop circular economy and realize resource conservation and environmental friendliness.However,due to the diversity,randomness and complexity of the actual waste parts damage,the accuracy and reliability of the detection and evaluation of the remanufacturing blank have become one of the key problems restricting the remanufacturing production process.With the waste brake discs with different damage characteristics,the following work focuses on the key problems of real-time detection and classification of surface defects based on machine vision algorithm:First of all,in view of the traditional manual inspection speed is slow,subjectivity is strong,repeatability is poor,and can not provide continuous consistency inspection,a machine vision-based remanufacturing blank defect inspection platform is proposed,and the system’s mechanical structure,electrical Control system and defect image acquisition scheme to improve the speed and reliability of brake disc defect detection.Secondly,machine vision is applied to extract and identify brake disc defects.In order to improve the detection rate,the original image of the brake disc is positioned,cropped,and extracted,thereby reducing the amount of data.Then the median filter is used to denoise the image,and the histogram equalization method is used to achieve image enhancement.In order to extract the characteristic regions of defects,the OTSU method and iterative method are used to perform threshold segmentation on the image,and the faster OTSU method is selected after data analysis.Then,the eight-neighbor labeling method is used to obtain the defect connected domain,and the number of defective pixels in the defect connected domain is calculated to achieve grade classification.The experimental results show that the class II defect types can be effectively identified.Thirdly,the Faster-RCNN target detection algorithm is used for classification based on the characteristics of class II defect types in this paper.In order to strengthen the model training effect of class II defects,the strip steel surface defect data of Northeastern University（NEU）was used as a data set,and preprocessing was carried out through data enhancement and equalization.After comparing the characteristics of various target detection algorithms based on deep learning,the Faster-RCNN target detection algorithm with high accuracy and fast speed is finally selected.Compare the VGG and Inception series model networks,and select the Res Net50 model with residual structure and strong model performance as the backbone feature extraction network.The trained model predicts three types of defects including scratches,grooves,and pits among grade II defects.The statistical data analysis shows that the Faster-RCNN target detection algorithm can further classify grade II defects.Finally,an experimental platform is built and a software system for brake disc defect identification and classification is developed.After running the experimental platform and software system,the detection speed reaches 2/s,and the accuracy rate reaches 82.1%.Practice has proved the practicability of the experimental platform and software system and the effectiveness of the image processing module.