Typical Fault Diagnosis Of Highly Reliable Switching Power Supplies For Trains Based On Thermal Imaging Characteristics Of High-frequency Transformers

In the railway train,the high reliable switching power supply as the power supply of the train instrument is the basis to ensure the normal operation of the train control system.Its fault diagnosis is of great significance to ensure the normal operation of the train measurement and control system.In this paper,taking the thermal imaging characteristics of the high-frequency transformer of the train switching power supply as the research object,the typical fault diagnosis model of the highly reliable switching power supply of the train is established based on the improved BP neural network based on the extraction and analysis of the thermal imaging fault characteristics.(1)Aiming at the problem that the heat edge of the key devices is low and the detail characteristics are not obvious in the original infrared image,the mean filter,median filter and adaptive median filtering method are used for smoothing,and the evaluation indicators of PSNR and SSIM show that the adaptive median filter method has the best effect.At the same time,through image sharpening,the detail characteristics of the elements in the image are enhanced.The infrared image processed by image preprocessing operation can provide a good image basis for the segmentation of the heating area and the extraction and analysis of the thermal imaging characteristics of key devices.(2)In order to improve the accuracy of the heat generation area segmentation of key devices(Devices that are prone to failure),according to the characteristics of the infrared image of the train switching power supply,an image segmentation algorithm combining regional growth and block Otsu threshold segmentation method is proposed,which can better realize the segmentation of the heating area of high-frequency transformers.The image shape characteristics of high-frequency transformers(Area,Perimeter,Compactness,Mean gray value,Standard deviation of gray value and Regional centroid)and feature D,which characterize the influence of environmental factors,are extracted as typical fault diagnosis characteristics of highly reliable switching power supplies of trains.(3)Using the K-L transform to reduce the dimensionality of the thermal fault feature vector,a new feature matrix composed of the original features is linearly combined,which can improve the classification accuracy and use it as a sample set for the fault diagnosis model.Aiming at the shortcomings of BP neural network,a fault diagnosis model is constructed based on the BP neural network based on the additional momentum-adaptive learning rate BP neural network and the additional momentum-adaptive learning rate BABP neural network,and the optimal parameters of the model are determined through experiments.The two fault diagnosis models optimized by the improved algorithm can realize the fault diagnosis of the highly reliable switching power supply of the train through training and learning,and the accuracy rate is 72.776% and 78.776%.(4)Based on the development tools Open CV and MFC,the infrared thermal image detection system for train high reliability switching power supply faults is designed in the VS environment,and a series of steps of fault diagnosis can be completed at the system interface using the fault diagnosis model.The results of the operation show that the system can realize the diagnosis of typical faults of highly reliable switching power supply of trains,and improve the efficiency of fault diagnosis.There are 37 pictures,14 tables and 71 references in the body.