Wheelset Tread Detection Technology Based On Machine Visio

With the change of time,China’s rail transit industry has developed vigorously.At the same time,the safe operation of rail transit has also attracted more and more attention.As an important tool to support train weight and ensure train operation,train wheel sets are in relatively poor working conditions and environment frequently,which are prone to peeling,abrasion,wear and other damage.Therefore,train wheel sets must be tested regularly to ensure the safe and stable operation of the train and prevent the occurrence of major train events.At first,the detection of train wheel set tread is manual measurement.The disadvantage of this measurement method is low efficiency and inconsistent manual detection accuracy.With the continuous development of machine vision,image processing technology is gradually applied to the field of tread wear detection.On this basis,a detection method for tread profile is proposed in this paper.The specific contents are as follows:First of all,a laser light source and a CCD camera are placed on both sides at low-speed train wheels.When the laser is illuminated to the wheel pair,the outline of the wheel to the surface will form a chart of the wheel pair.When the wheels are running,use a CCD camera to collect enough contours.Then,process the parameter information of the collected pedal contour.The tread profile is preprocessed by graying,filtering and edge detection.The internal parameters,external parameters and distortion coefficients were obtained by camera calibration method,and then the tread parameters were obtained.Next,the wear area in the tread profile is identified.The edge fitting of preprocessed images was carried out to remove irrelevant background information.According to the loss function and MIo U value of the model,it can be concluded that FCN-8S model has the best detection effect.Compared with FCN-16 S and FCN-32 S,FCN-8S model has more accurate identification of point wear area.Finally,this thesis designs a wheel-set tread detection system,and applies the image processing method used in the experiment and the model to identify the tread wear area to the system.Pyqt5 was used to write the real-time detection interface,and the tread pretreatment and tread recognition were written by Python software.This paper selects the most suitable image preprocessing method for this system by comparing the effect pictures of various graphics processing methods.Since the contour graph is a curve,semantic segmentation model is adopted in this paper to achieve pixel-level recognition of the image.Among the three models,f CN-8S model with the best recognition effect is selected to ensure the accuracy of the system.