Research On The Theory And Method Of Intelligent Maintenance For Key Components Of High-speed Emus Based On PHM Technology

With the increase in the operating speed of China's high-speed EMUs and the growth of the scale of use,the operational safety and security technical challenges of EMUs have become increasingly prominent.Prognostics and Health Management(PHM),as a kind of equipment health management technology,can realize equipment status monitoring,abnormal prediction,fault diagnosis,maintenance prediction and maintenance decision-making functions.In order to improve the safety protection capability of high-speed EMUs,reduce operation and maintenance costs,and improve maintenance efficiency,this thesis deeply integrates the maintenance business of high-speed EMUs and PHM technology,and explores the theory and method of intelligent maintenance of key components of high-speed EMUs based on PHM technology.This theis concentrates on the research of component state prediction and fault diagnosis technology based on risk prediction and prevention,and key component maintenance prediction and maintenance decision-making technology based on fixed maintenance plan.Provide theoretical basis and technical support for the transformation of the EMU maintenance mode from "planned maintenance" to "planned maintenance +predictive maintenance".The main research content includes the following four aspects:(1)In order to solve the problem of abnormal state prediction under the background of complex service environment and diverse failure modes of key components of the high-speed EMUs,a method combining online temperature monitoring and multiple hidden layer neural network predictions is proposed.This method carries out online monitoring of the bearing status of the key components of high-speed EMUs,and predicts the health status of the components through a multi-hidden-layer neural network prediction method.Collected sample data of approximately 30,000 kilometers in 8 key locations including EMU gearbox bearings,traction motor drive end bearings and so on.Analyze the change characteristics of bearing temperature in different temperature conditions,different operating states and intervals,study the correlation of bearing temperature with time and driving characteristics,and predict the trend of bearing temperature data.Experimental results show that,compared with common methods,the accuracy of bearing temperature prediction is significantly improved,with MAPE within 3% and RMSE within 1.(2)Aiming at the accuracy problem caused by the lack of fault sample data of key components of high-speed EMUs,an optimized sequential extreme learning machine fault diagnosis method for unbalanced data(MS-ABC-OSELM)is proposed.This method uses the K-Means SMOTE method and the under-sampling method based on Euclidean distance to reconstruct the sample data set,uses the non-balanced data classification evaluation function as the fitness function,globally optimize the parameters of the fault diagnosis model and construct the fault diagnosis model.At the same time,the accurately classified data is used as a sequential sample to continuously update the diagnostic model.The actual operation data of the EMU axle box bearing is used as a sample for verification.The results show that compared with the existing method,the value is increased by more than 6.9%,and the F1-measure value is increased by more than 9%.(3)In view of the problem of the unpredictable operating mileage in the high-level maintenance plan of high-speed EMUs,a mileage prediction algorithm based on empirical mode decomposition and optimized deep learning is proposed.This method uses the empirical mode decomposition method to decompose the mileage time series into high and low frequency time series,and calculates the mileage prediction results by constructing an optimized deep confidence network prediction model,to realize the prediction of the high-level maintenance time window.Take the three-year operating data of the EMU as a sample,the impact of the sample period on the prediction results is analyzed,and the sample period of the mileage prediction model is determined.The experimental results show that compared with the traditional forecasting method,the proposed method reduces MSE by more than 23.9%,MAE by more than 22%,and RMSE by more than 12.7%,which can be used as an effective basis for the preparation of high-level maintenance plans for EMUs.(4)In order to solve the problems of vacant and running maintenance resources and low utilization rate of EMUs caused by unbalanced maintenance demand,a high-level maintenance plan optimization method based on particle swarm optimization is proposed.This method comprehensively analyzes the influencing factors of the advanced repair plan of the EMU,and establishes the evaluation index of the advanced repair plan based on the lost mileage,the maintenance capacity and the number of maintenance days on holidays.The evaluation index is used as the fitness function of the particle swarm algorithm to construct the optimization model of the high-level maintenance plan.The global optimization plan for the high-level maintenance plan is obtained.The experimental results show that compared with the traditional advanced repair plan preparation method,the proposed method reduces the evaluation index by35.9% and reduces the time consumption significantly.This thesis conducts research from two aspects of key component safety assurance and capability maintenance,builds an intelligent maintenance model,and verifies the effectiveness of the proposed method through the actual operation data and real business scenarios of the EMU.There are 70 figures,35 tables,and 153 references.