Detection Of Surface Defects Of Train Roller Bearings Based On Machine Vision

Train roller bearings are a key component of railway train transportation safety.It plays a very important role in the safe driving of trains.Therefore,relevant departments attach great importance to the inspection and maintenance of bearings.Traditional testing methods mainly depend on the experience and responsibility of employees.Repeated operations for a long time will cause visual fatigue,which will reduce the accuracy and efficiency of test results.Moreover,such a maintenance method does not store,upload,and statistics the detected data,and does not have a complete test and data report,which makes it difficult to analyze based on the existing test data when such problems occur in the future.In view of this situation,this paper improves the existing research and proposes a method for detecting surface defects of train roller bearings based on machine vision.The main research works are:(1)Image acquisition and initial classification.In order to realize the automatic processing of bearing defect detection,this paper improves a bearing image acquisition device.It proposes to use industrial endoscopes instead of human eyes to acquire images.The captured images are stored in a database system to realize automatic acquisition of bearing defects.In addition,the image is judged to be defective according to the standard deviation of the gray value of the image,and the bearings are classified for the first time.(2)Binarization and morphological filtering.This paper proposes a method to determine the threshold value based on the gray-scale mean range of the defective image,and binarizes the defective image.Compared with the traditional method,this method is more accurate and effective.Due to the uneven light in the image acquisition process,there will be noise in the collected image.The morphological filtering of the binary image is performed in this paper,which can eliminate the noise and retain the main features of the defective image.(3)Edge detection.The edge detection of defective images is a key step in image processing,which has a very important impact on the classification and recognition ofdefective images.After comparison,this paper chooses Canny algorithm for edge detection,so the defect edge features obtained are more accurate.(4)Image tagging and feature extraction.In this paper,an 8-connected region labeling method based on morphological expansion operation is used to obtain labeled images,and some features that can be used to distinguish different defects,such as area,perimeter,aspect ratio,and number of connected regions,are extracted from the labeled image.(5)Classification and identification.After comparing several commonly used recognition algorithms,the paper uses the method of classification and recognition decision tree to classify the defect image and determine the threshold.After experimental verification,the recognition rate reaches 96.7%.Compared with other classification algorithms,this method has no sample training process,so it is more simple and practical.This paper has carried out beneficial exploration and provided specific solutions for the automatic detection of surface defects of train roller bearings.This method also has important reference value for the research of other mechanical parts surface detection technology.