| Transportation industry is one of the crucial industries of a country with respect to people's life and both economic and social development.Railway is an important part of transportation industry.It is the pillar carrier of social and economic development.EMU high-speed rail as a means of railway transport,also experienced decades of development and growth.As early as 1950s,the global economy and technology have developed rapidly,especially the railway passenger transport and highway,aviation,and EMU high-speed rail with its fast,safe and flexible features.The operation of high-speed EMU marks the fact that the country has entered the ranks of high-speed railway countries.Although high-speed railhas been demonstrated to be remarkable safe,it still involves many high-tech problems such as vehicle technology,bogie technology,traction drive technology,and fault diagnosis technology.The fault diagnosis technology of EMU is one of the key technologies to ensure the safe operation of EMU.It can make early forecast of the fault development of EMU,judge the cause of failure and put forward countermeasures and suggestions to avoid or reduce potential accident.Fault diagnosis technology is the new rapidly developing technology.It is a multidisciplinary and comprehensive technology to meet the needs of various engineering needs.Its development is closely related to the maintenance of the equipment.For quite some time,duo to the low level of technology,there have been two ways to maintain and repair the equipment:one is after maintenance,in which the analysis and maintenance of the equipment carried out after its failure.However,this way of maintenance is not only subject to damage of the equipment,but also causes great threat to people's life.There is also a periodic maintenance,but,this way has not reduced the fault,and it resulted in waste of maintenance resources in case of absence of failure.Around 1960s,the United States military put forward predictive maintenance,that is,in the process of the normal operation of the equipment it begins monitoring to detect potential faults and takes early measures to prevent sudden breakdown.This approach has quickly applied in various fields and fault diagnosis technology made quick breakthrough.At present,there are many diagnostic techniques have been applied the fault diagnosis of EMU,but the efficiency and accuracy of diagnosis are relatively low.Moreover,with the increase in the operation time and operation scale of EMU,large amount of fault data has been accumulated and the amount of data for data mining needs increased dramatically.Traditional fault diagnosis methods have been unable to meet the needs of such large-scale data analysis,and some can't even come to the conclusion of diagnosis.In order to overcome the above stated problems,it is necessary to explore more efficient and accurate methods that could be applied for the analysis of the fault data of EMU and be able to get more valuable information quickly and securely.The fault diagnosis of EMU is a nonlinear phenomenon,and the application of BP neural network in the fault diagnosis of EMU has been carried out.But,the traditional BP neural network training method is serial data processing on a single machine,and with the rapid development of information society,the amount of data needed for data mining increases rapidly,reached the level of massive data.As a result,the traditional methods will have a lot of problems in dealing with massive data sets such as long time consumption,insufficient memory to train and other issues.Inorder to solve the problem of single-line serial BP neural network training method,the parallel mode is the inevitable choice,and in the parallel mode choice,the emerging cloud computing is a very suitable technology.The EMU is a very complex large-scale electromechanical equipment.In fact,the research on fault diagnosis technology and device is a huge and complex research.In this paper,only the main converter of one of the key components of the EMU is taken as the research object.According to some faults of the main converter inverter,the failure mechanism is explored,and the BP neural network theory based on Spark is proposed.The main work of this paper is to analyze the working principle and fault mode of traction converter.By using certain technology,the fault signal processing can be realized and the fault characteristics can be obtained.On this basis,the BP neural network is applied to the identification of the fault modes.The design method of neural network learning samples and the neural network design are discussed and the BP neural network algorithm is improved in a parallel algorithm based on Spark platform,and solve the traditional BP neural network training method problems.Based on the simulation of fault diagnosis,the effectiveness and advantages of this method are verified.In short,the Spark is introduced into the EMU fault data analysis,which can be implemented by HDFS or local storage for faster read and write speed.Hardware failures can be solved by storing redundant data.Thus,we can safely,quickly,and effectively analyze the fault data of EMU,and obtain more valuable information for fault prediction,fault avoidance,troubleshooting and so on.According to the method proposed in this paper,firstly,the shortcomings of the present intelligent fault diagnosis theory applied to the traction converter of CRH5 EMU are analyzed.The internal structure and working principle of the traction converter is analyzed,and the fault characteristic parameters are determined.The traction converter used in AC drive electric locomotives and high speed EMUs is composed of four quadrant pulse rectifier and traction inverter.Traction converter for EMUs CRH5 YGNZQ213(user LY000O0001o5O)series was introduced by Alstom Technology.It consists of 8 component platforms:2 auxiliary component platforms,2 traction module assembly platform,2 user component platforms,1 cooling system platform and a resistor assembly platform.All 8 of them are connected via central line slot to form a one whole.The structure of traction unit 1 and 2 is substantially the same,mainly composed of four quadrant rectifier,inverter module,supporting capacitor and water cooling circuit interface.The traction converter is a typical large-scale analog electronic circuit.Its common faults are:overcurrent fault,overvoltage fault,under voltage fault,overheating fault,overload fault and etc.According to the fault location,the fault of traction converter can also be divided into input stage fault,internal fault and output stage fault.Input stage faults include pantograph,the main circuit breaker or the main transformer faults.Output level fault is the fault of traction motor.Internal fault is the fault of the converter itself,which can be divided into:DC link fault,inverter fault and control system fault.This paper mainly studies the inverter fault in the converter.The power semiconductor device and its control circuit are the weakest link in the inverter,and their reliability hasn't been fully resolved.According to statistics,82.5%of the failure of control system is derived from components failure,mainly IGBT or diode damage.The faults of the power supply can be divided into straight and open circuit faults.When the fault is straight,it will burn out the power device,thus,manifesting the open circuit fault.Therefore,the study of open circuit fault of power device can basically reflect the main fault phenomenon,make accurate positioning,ensure the safety of locomotive operation.After studying and analyzing the circuit structure and working principle of converter of CRH5 type EMU in detail,and by analyzing the open circuit fault of the power supply of inverter,22 cases of inverter open circuit faults are found,and by combining them all,results show that output current waveform is different under different fault conditions.Therefore,the characteristics of the fault types can be presented from these different output current waveforms.Afterwards,the neural network theory is studied thoroughly in detail.The neural network is the kind of network system which imitates human brain information processing mechanism.It is composed of a large number of connected simple neurons,which have the following characteristics:they have strong self-learning ability,they have strong fault tolerance,parallel architecture and parallel processing,and they are nonlinear.Furthermore,the structure,basic principle,and algorithm of BP neural network are studied and expounded.The main idea of BP neural network includes forward propagation of signal and back propagation of error.In the process of forward propagation,the input signal is transmitted to the output layer through hidden layer processing.If the output value is different from the expected value,and it is larger than the acceptable error range,then the back propagation process is carried out.The error is transmitted through hidden layer to input layer for error adjustment.By constantly adjusting the weights between layers,the output error reaches the acceptable range or the maximum number of learning rate.In the final stage of the thesis project,I learnt the origin and framework of Spark.On the basis of open source cloud computing platform of Spark,combined with the traditional BP neural network theory,the parallelization method of BP neural network algorithm is studied and as a result,the fault diagnosis method of traction converter based on Spark is proposed.Then,the structure of the converter fault mode and the neural network are obtained and coded.So that the neural network is relatively fixed.Combining the existing feature vectors,the design method of neural network,and the programming method of Spark,the fault diagnosis system for traction converter of EMU is designed.The non-sample data is input into the system and the fault types are obtained to verify the validity of the fault diagnosis neural network.The simulation and test of open circuit fault diagnosis of inverter is carried out,and the research shows that this method is effective.The neural network fault diagnosis method based on Spark is helpful to improve the performance of fault diagnosis,and has broad prospects for development.Finally,by using certain amount of data,the performance of traditional BP neural network fault diagnosis algorithm and the Spark based BP neural network fault diagnosis algorithm are compared.The acquired results showed that the BP neural network algorithm based on Spark that is proposed in this paper is effective in the fault diagnosis of the traction converter of EMU,and has broad prospects for development.Even though this paper only analyzes the fault of the inverter link of traction converter,the fault diagnosis method of BP neural network based on Spark can be applied to other components and systems,and can be extended to other applications.The purpose of this paper is to use BP neural network based on Spark to diagnose the fault and improve the accuracy and efficiency of fault diagnosis.Its theoretical significance and engineering application prospects are very crucial.It has certain theoretical and practical value... |
PDF Full Text Request