Design On Automatic Detection System For Rolling Bearing Inner Ring Groove Surface Defect Of Railway Freight Car

The main purpose of the design for detection system in this paper is tosolve the problems,which exists in the artificial visual for railway wagonrolling bearing inner ring groove surface at present,and realize theautomation of detection for railway wagon rolling bearing inner ringgroove surface. The test system mainly consists of three parts:(1)The design for structure of image acquisition system, which mainlyrealizes the automation of images acquisition for railroad freight rollingbearing inner ring groove surface.In the image acquisition process, this paper put forward an idea thatthe collection for image by use of the CCD camera to replace the naked eye.While at work, the position of light source and CCD camera is fixed, themotor through the reducer slowing turntable to turn at a certain speed, andat the same time, the CCD camera collecting image for bearing inner ringgroove surface even the way clap, and sequence keep.(2)The design for program of defect detection, which completes thedetection that internal circle groove face whether have defect through theanalysis and identification for the collected images. Useing the meature that image processing combines with artificialneural network to judges whether bearing inner ring groove surface havedefects.Two image preprocessing method,such as median filtering andbinary, removes the noise of the image and makes the distinction betweenthe defects and the background is more obvious,Based on the three texturecharacteristics that on the basis of gray symbiotic matrix,we get a newcharacteristic features(two ratio)through the statistical analysis,andcombines its into one feature vector as artificial neural network’s input tocarry on the training and test for the good designed network and determinewhether bearing inner ring groove face have defects.(3)The design for program of defect classification,which completesits thinning and classification to the images that inner circle groove surfacehave defects.In the defect detection program basis,the image processing method thatis produced in the way that Convolution filtering combines with open,close operation removes interferences which come from the edge pointsaround with the defects effectively.The confirm for minimumcircumscribed rectangular of defect avoids calculation characteristics whichcome from not related area.Increasing the feature quantities, such as degree of compression, line length, distance extremes ratio, NMI feature andinvariant moments, based on the traditional features enhances basis whichbe used to divide defects. The improvement on the input matrix of BPneural network and normalized method improve neural network memorycapacity and recognition speed.The detection for the recognition result ofdefect classification system determines this system, reliability.