Research And Design Of Visual Inspection System For Bearing Roller Surface Defects Based On Machine Learning

With the development of computer technology,industrial production has higher requirements for automation.Especially driven by the Smart Manufacturing 2025 program,industrial production is developing towards a higher degree of intelligence.With the emergence and popularity of machine vision technology,product quality inspection in production has become more automated,more informational and more intelligent.This paper studied the visual inspection method for defects on bearing roller surface based on machine learning.This method designed a defect detection system based on different defects using machine vision technology and image processing technology.The main research contents are as follows:First,the design of the visual inspection system was carried out and an optical imaging system was built.The optical imaging system mainly consisted of two plane-array CCDs,two line-array CCDs,four lenses,four ring light sources and two strip light sources.Three workplaces were designed to capture the cylindrical surface and end surfaces of the bearing roller.The system can acquire stable and high-quality images through reasonable hardware setup and software development,thus realizing real-time defect detection.Then,the defect detection methods based on SVM and CNN were studied respectively.Firstly,the SVM-based defect detection process was introduced in detail,including image filtering,segmentation,feature extraction and classification.Then we focused on the defect detection method based on CNN.This method applied the object detection technology to the surface defect detection.An object detection model called SSD with high detection accuracy and high detection efficiency was used,and a better performance basic network was replaced for the SSD model to obtain better detection results.Finally,the hyper parameters were selected reasonably through experiments,which made the method achieve a good result in the classification task.Finally,the experimental results showed that the bearing roller defect detection method based on CNN has good robustness.In this paper,the CNN-based and SVM-based defect detection methods were compared.The CNN-based method was superior to the SVM-based method in defect classification accuracy,which was over 98%.This paper also applied the CNN-based method to the defect detection equipment and compared it with the traditional method that used in production line.The experimental results showed that the accuracy of the detection of qualified rollers of CNN-based method,which was over 95%,was much higher than the traditional method.The experiments proved that the bearing roller defect detection system can replace the manual detection,improve the detection efficiency and accuracy,and has engineering application value.