Study On Detection Technology For Wheel Defects Of Trains Based On Machine Vision

Wheel is one of the most important parts in rolling stock.Not only it has to withstand the vertical and horizontal forces,but also withstand strong friction from brake shoes and rails especially when it comes to train braking.Its main function is to carry the frame and the car body,and to deliver the traction and braking force of the train.Due to the bad working conditions,it often leads to defects such as abrasion and peeling of the wheel.The pros and cons of wheel's running performance have a great impact on driving safety.This paper is based on the image processing module in machine vision technology and studies the detection method of surface defect of wheels,including rim crack detection,location of tread abrasion and pit defect detection,and the rim crack detection method of Multi-feature screening,tread abrasion location method of gradient angle screening and pit defect detection method are raised.According to the image characteristics of train wheel crack on flange and rim,the various detection algorithms of crack are studied,and a new wheel crack recognition method is proposed based on crack image characteristics and Fisher criterion.First of all,among many methods of image denoising and enhancement,the fuzzy enhancement method is chosen after comparison.Algorithm uses the image segmentation method based on local statistics variable threshold after testing multiple segmention methods,with a small area threshold filtering to remove speckle noises,combining with mathematical morphology operation,and calculating the respective four features in every image after image preprocessing.Then,the extremums of four features in the continuous filtering image are calculated,and the Fisher method is used to recognize the crack image.Meanwhile,the image crack line coordinates are extracted,and the whole crack line is extracted by polynomial least square curve fitting method.Aiming at the problem of train wheel tread defect detection via images,the segmentation and location method of tread abrasion based on screening between gradient and gradient angle is proposed.Firstly,according to the direction feature of the tread and abrasion,setting the gradient angle of the specified direction to filter the specified direction gradient,then the gradient image is segmented by the Otsu threshold method,and obtaining the tread edge line by morphological method and polynomial curve fitting method.Finally,the tread is segmented and determined according to the width of the standard wheel tread.For tread areas after segmentation,the morphological transformation method is used to weak the background.Using the Otsu method to segment the abrasion area of the tread,drawing the abrasion location frame and locating the area on the original image combining with morphological operation and area threshold screening according to the abrasion area.Aiming at the detection of pits defects,according to the jumping characteristics of the gray distribution on the defect area,the standard gray distribution is established by the fitting method.The difference gray distribution and the difference gradient distribution are calculated.Defections are defected by observing the difference gray distribution and the difference gradient distribution according to the jumping effect on overall gray distribution trend.