Machine Vision Based Trouble Of Moving Emu Detection System For Railway

As a high speed and large capacity railway vehicle,its safety is undoubtedly the top priority.There are many aspects of the work safety guarantee of EMU.It is an effective measure to improve the efficiency of the repair work of the EMU,to ensure the quality of the overhaul of the EMU and to monitor the running state of the EMU.Trouble of Moving EMU Detection System(TEDS)applies computer,network communication,automatic control and machine vision technology and introduces scientific management methods and systematic development methods.It provides dynamic collection,storage,transmission and automatic identification of early warning service for the fault detection of railway EMU,and improves the quality and efficiency of the repair work of EMU.And the level of vehicle safety precautions,and strengthening the collection and management of fault basic information in EMU operation is an important part of 5T system.The early TEDS uses the array camera to match the LED light source,the system is weak in anti sun interference,the image quality is poor,and the image can not be stitching seamlessly,which also brings limited conditions for the later automatic recognition.In view of the image quality problems existing in the early TEDS,line scanning technology and laser light source are introduced to TEDS.According to the characteristics of line scanning imaging,a set of line scanning image acquisition and installation is developed successfully through the feasibility analysis,calculation,selection,design and debugging in the laboratory stage.Seamless splicing,better solve the problem of sunshine interference,and completely avoid the phenomenon of lost map.The early TEDS automatically identifies the difference between the current vehicle and the last train on the same train.Although there is a real fault alarm in the application process,the fault false alarm rate and the number of faults are high,and the classification of the fault level can not be carried out.By introducing the method of machine vision deep learning,the malfunction images of the EMU are classified according to the maintenance procedures,and the computer is used to train and train different positive and negative samples of the fault feature vectors,and is applied to the extraction learning of the image features of important parts based on the vgg16 depth network based on the keras framework.After the recognition algorithm iscompleted and deployed,the fault recognition error rate can be reduced to less than50% by the test and verification of the fault recognition test of the key parts of the key components of the key parts of the EMU,and the fault recognition error rate can be reduced to less than that.After the algorithm is verified,it will gradually realize the automatic identification and alarm of the key parts of the fault,and realize the real dynamic fault detection of EMU.