Research On Defect Detection Technology Of Inner Ring Outer Surface Of Train Tapered Roller Bearing Based On Image Recognition

Wheelset tapered roller bearing is an important transmission part in train operation and its condition is related to the safety of train operation.At present,the surface defects of inner ring of railway train bearing are mainly detected by manual inspection.In view of the problems such as visual fatigue,long detection time,insufficient stability,low detection efficiency and difficult information integration,etc.,when the inner ring of railway bearing is detected by manual inspection,it is easy to produce visual fatigue.Machine vision and image processing technology are used to study the detection and recognition of defects on the outer surface of bearing inner ring.The specific work is as follows:(1)Scheme design for detecting the defects on the outer surface of inner ring of semidismantled train roller bearing by machine vision.Through the analysis of the structural characteristics of the roller bearing of the semi-disassembled train,the selection and scheme design of the industrial camera,light source and lighting environment are carried out.In the lighting environment with direct lighting as the main mode and dark field lighting as the auxiliary mode,industrial endoscope was used to complete the image acquisition of the outer surface of the bearing inner ring.(2)image pretreatment.The industrial endoscope image can be obtained using the weighted average method for gray level transformation,through the calculation of the standard deviation value to determine whether the sampled bearing image containing defects,in terms of image filtering,through the optimized median filter for filtering operation,finally filtering gamma transform for the image enhancement,improve the overall effect of the image,make its have greater contrast.(3)Segmentation of defect images.Image segmentation realizes an important process from image processing to image analysis.The edge of defect image is the most basic feature description of defect.In this thesis,an edge detection algorithm based on entropy fusion and multi-direction morphology is proposed to extract the edges from four kinds of common defect images(pitting,corrosion,indentation and scratch)on bearing surface.The results show that the proposed algorithm can effectively suppress the noise interference and retain the edge information of the defect,which has good feasibility and robustness.(4)Feature extraction and classification recognition.A feature pool containing 40 features was established through statistics and sorting.Ten features with the most distinguishing degree and representativeness were selected in terms of regional shape feature,gray feature and texture feature.Feature vectors of four common defects on bearing surface were calculated and normalized,and SVM classification prediction model was designed.The grid search method based on cross validation is used to optimize the parameters of the classification model.Finally,the test sample proves that the recognition rate of the classification model reaches 94.81%,which indicates that the model has a good recognition effect under the basic conditions of feasibility,and also verifies the validity of the features selected in the thesis.Through machine vision and image processing technology,automatic detection and recognition of defects on the outer surface of inner ring of train tapered roller bearing are studied.The design scheme basically meets the expected effect.At the same time,the research work provides some reference value for forming a set of stable and efficient automatic visual online detection and recognition system.