Research On Automatic Detection Algorithm Of Critical Component Fault In TFDS Based On Local Binary Patterns

With the continuous development of China’s railway enterprise,the traveling density and load of freight car are increasing year by year.In order to meet the high requirements about safe operation of freight car,Trouble of moving Freight car Detection System(TFDS)arises at the historic moment,this system can detect various faults of parts on freight car,and notify workers to troubleshoot timely,it ensure the safe operation of freight car.However,the current TFDS does not fully automated,because human involvement is necessary in the fault detection.The development of machine vision technology makes it possible to realize automatic fault detection on freight car.This technology can effectively improve the fault detection efficiency,and reduce the operation cost.In this paper,the bolt and bogie block key in TFDS are the research object,and using theory of image processing and deep learning.The main contents are as follows:(1)For the bolt missing fault of beam that supporting coupler,Coupler tail frame bracket and shaft end.An fault detection algorithm for bolt missing based on Local Ternary Pattern(LTP)is proposed.Firstly,three kinds of original images need to be preprocessed to improve the accuracy of positioning and simplify the subsequent algorithm.When positioning bolts,a positioning method that from rough to detailed is used.Firstly,using YOLOv3 convolutional neural network to train the positioning model to locate the region of interest,and then the bolt position is determined.At the fault recognition stage,algorithm using Adaptive Thresholds Local Ternary Pattern(ATLTP)operator to extract bolt features.The threshold of ATLTP is correlated with the standard deviation in the neighborhood of the center pixel,which can improve the anti-noise performance of the operator.After counting the bolt feature vector,using Support Vector Machine(SVM)to realize fault recognition.Theoretical analysis and experimental results show that the proposed algorithm has high accuracy in bolt positioning,better bolt fault detection rate and error detection rate.Algorithm can effectively overcome the problems of poor image quality and complex background in the TFDS,and can meet the application requirements of automatic bolt fault detection.(2)For detecting the fault of bogie block key,A fault detection algorithm for losing fault of bogie block key based on Local Directional Derivative Pattern(LDDP)is proposed.Firstly,at the stage of bogie block key location,the YOLOv3 convolutional neural network is used to train the positioning model to determine the location of the shaft end.On this basis,a mathematical model is established according to the location relationship between the bogie block key and the shaft end,the block key region image is obtained accurately.In terms of bogie block key fault recognition,in order to capture details of bogie block key and reduce the sensitivity of feature extraction operator to noise,Median Local Directional Derivative Pattern(MLDDP)operator was used to extract features of bogie block key,then detect loss fault of bogie block key in SVM.Theoretical analysis and experimental results show that the proposed algorithm is more accurate than the traditional Hough transform method,and the improved MLDDP operator is more robust than the original model.