Research On Key Technologies Of Fault Prediction And Health Management For High-speed EMU Based On Big Data

While providing convenience for people to travel,EMU sets higher requirements for operation safety.Prognostics and Health Management(PHM)is a technology that evaluates the health status of equipment by collecting all kinds of data and equipment characteristics,and makes prediction of equipment failure,so that it can effectively take timely measures to maintain the equipment before the failure occurs.It is the development direction and effective means of EMU operation.In this thesis,big data technology and PHM technology are combined to study and explore the PHM system architecture,data reception and processing,fault prediction and health assessment algorithms of high-speed EMU.In addition,combined with digital twin technology and multi-task deep learning algorithm,a digital twin-based PHM technology framework of high-speed EMU is proposed.Real-time data receiving and processing and offline data analysis and processing methods are designed,fault diagnosis and fault prediction methods are constructed respectively,and a high-speed EMU PHM system is developed and implemented.The main research contents of this thesis are as follows:1.PHM system architecture of high speed EMU based on digital twin.By constructing the digital twin model of high-speed EMU physical entity,integrating the real-time data and related twin data of physical and virtual equipment,the PHM system architecture of high-speed EMU based on digital twin is proposed,which provides theoretical guidance for PHM system implementation.On this basis,the EMU data receiving and processing method based on big data distributed computing technology is studied.The real-time data processing adopts the message processing mechanism to receive and send messages in real time,and then load the messages to spark platform for data flow calculation;Offline data processing is used for model training and statistical analysis.Hive is used as data warehouse,and the construction method of five layer data hierarchical model in data warehouse is described in detail.2.Protocol parsing method for high speed EMU data transmission with configuration and distributed computing characteristics.Aiming at the problems of various types of communication protocols,huge workload of parsing and parallel parsing delay of various types of high-speed EMUs,a new standardized protocol parsing method is proposed,which has the characteristics of configurable,extensible and parallel computing.It includes the standardized coding rules of protocol variables,the configuration method of parsing rule base based on parsing function library,and the protocol parsing method based on spark engine dynamic loading parsing rule base to ensure the rapid expansion of new models and the rapid response of old model protocol changes.3.Fault diagnosis and maintenance prediction method based on multi-task deep neural networks self-organizing map(MTDNN-SOM).Aiming at the problem of traction converter fault caused by the abnormal cooling filter state of traction converter of high-speed train,a fault diagnosis method based on multi-task deep learning was proposed by comprehensively analyzing the correlation between the fault classification of traction converter and the clogged degree of filter.Firstly,a multi-task deep neural network was constructed,which included the main task of filter blockage fault diagnosis for traction converter and the sub-task of filter clogging degree.Then,the fault features extracted by multi task deep neural network were combined with the self-organizing mapping method to construct MTDNN-SOM.The experimental results show that this method is superior to single task or traditional fault diagnosis methods in accuracy and efficiency.4.Multi-dimensional data fusion method for EMU condition prediction and health management.In view of the complexity of high-speed train structure and operation environment,in order to effectively evaluate the health status of the train during operation and predict the possible faults in advance,a bearing condition prediction and health management method for high-speed train based on condition identification and multi task deep learning is constructed.Taking the axle box,gearbox and traction motor as the research object,this thesis firstly analyzes and identifies the operating condition parameters of high-speed train,then constructs the bearing temperature prediction model based on the condition identification and multi-task deep learning.Finally,according to the theory of statistical quality control,the abnormal state of high-speed train bearing is evaluated and the accuracy and availability of the model are verified by an example.5.A prototype system of fault prediction and health management for high-speed EMUs is designed and developed.Based on the industrial big data and the idea of hierarchical system design,the PHM technology architecture of high-speed EMU is designed.Based on this architecture,the core functions and knowledge base construction technology of EMU PHM system are elaborated.Finally,through the system function construction,the EMU health management function is realized.