Research And Application Of TEDS System Image Aided Recognition Technology Based On Deep Learning For CRH

The railway as the main way of transporting passengers and goods in China,the composition of the system is extraordinary complex,and any tiny defects of small parts may have a great impact on the safe operation of the railway.The existing railway fault detection system adopts man-machine combination to check the potential safety hazards.On the one hand,the huge data brings great workload to technicians,and visual fatigue directly leads to misjudgment.On the other hand,the delay of manual inspection cannot ensure the timeliness of data processing.With the development of artificial intelligence technology,the deep learning will make full use of the data to realize intelligent and efficient fault detection of train image and ensure the safe operation of high-speed railway.EMU has obvious change in resolution of train and component,and the existing defect sample is few,various and unpredictable,which brings more difficult challenges to train component detection and analysis.Based on the preprocessing of spatial overlap,the component location is realized by YOLO V3,and IRGAN is proposed which is a new unsupervised learning and diagnosis method for component defects.In the end,Qt is used to encapsulate the call of discrete subroutine,and the intelligent detection of components is realized in EMU detection software.The main research contents and achievements are as follows:(1)Based on the target detection algorithm of YOLO V3,the project solved the problem that the precision parts could not be accurately identified due to the big change of dimension between the images of trains with the precision parts,and the serious loss of the resolution of the underlying network.(2)Based on the unsupervised algorithm of image reconstruction,an unsupervised learning method of fault component image detection is proposed.The unsupervised learning of defect anomaly diagnosis network is used to create the template image which can identify the existing or possible abnormal parts,so as to ensure the fault components are not missed.(3)Because the process of the image training needs to call many discrete subroutines,which requires professional developers to spend a great deal of manpower,and the efficiency is low,this paper develops a set of integrated image training software based on QT to realize the automation of the railway data training process and improve the efficiency of image training.(4)Based on the learning of the data from EMU,trouble of moving EMU detection system is developed to realize the efficient detection and processing of train image which guaranteed the safe operation of high-speed railway train.