Research On The Defect Detection Technology Of The Graphite Ball Surface Based On Machine Vision

With the rapid development of the nuclear industry,the fourth generation nuclear industry core system aimed at achieving high safety and high combustion efficiency has been developed.Although its combustion performance has achieved a qualitative leap,the capacity to adapt to the relevant workpieces of the fourth generation nuclear industry system is still low.Among them,graphite balls,the fuel component of the fourth generation nuclear industry system,have defects such as bumps,risers and tough marks in the production process.In view of the safety requirements of the nuclear industry,all graphite balls need to be detected.Because the high radiation of graphite ball will damage workers detecting,and workers detecting graphite ball has the problems of low efficiency,high false detection rate and high cost,at the present stage,graphite ball detection is difficult to meet the capacity demand of the fourth generation nuclear industry system,so it is urgent to study new detecting technology to improve detecting efficiency.Aiming at the low detection efficiency of graphite ballzhe at the present stage,a set of graphite ball surface defect detection system is designed.The main research work of the paper is as follows:(1)According to the characteristic analysis and detection requirements of graphite ball workpiece,the graphite ball imaging scheme is designed,the imaging hardware is selected,and the imaging platform for graphite ball surface defect detection is built.Firstly,based on the traditional image processing algorithm,the research on graphite ball defect detection is carried out.Gaussian filter is used for image preprocessing,OTSU algorithm is used for image segmentation,morphology and Canny algorithm are used to extract the feature points of graphite ball,and a boundary circle fitting algorithm based on least square method is proposed to realize the rough positioning of graphite ball.On this basis,the surface defects of graphite balls are recognized through the shape and position information of the contour,and then the surface defects of graphite balls are classified based on the gray value characteristics of the defect centroid and contour.The experimental results show that the traditional image processing algorithm has a certain detection effect for the surface defect detection of graphite ball,but its accuracy needs to be further improved,and its classification effect is poor.(2)In view of the general effect of traditional image processing algorithms on graphite ball surface defect recognition and classification,the research on graphite ball surface defect recognition and classification based on deep learning object detection is carried out.Firstly,the rough positioning of graphite ball is realized by the boundary circle fitting method based on the least square method,and then the recognition and classification training of graphite ball defects are carried out by using deep learning object detection such as Faster R-CNN,SSD,Retina Net and YOLOv5 series with better performance at this stage,so as to finally realize the high-precision and high-efficiency detection of graphite ball surface defects.The experimental results show that the m AP value of graphite ball surface defect detection based on YOLOv5s is as high as 97.18%.(3)Aiming at the process oriented and frameless programming mode of visual software system development at the present stage,a graphite ball defect detection software framework is proposed.The main software framework adopts the consumer producer mode to realize high encapsulation,high concurrency and high robustness,and then further improve the flexibility of the software through user-defined process and flexible configuration.the flexible configuration includes flexible library based on the adapter mode to realize the low coupling of the moving program and encapsulation into a library to realize the high reuse of the program.The program framework can be extended and oriented to machine vision software.