Research On Fault Detection Of High-speed Train Traction System Based On Autoencoder

High-speed train is a complex large electromechanical system,which contains many subsystems of different forms.The subsystems cooperate with each other to achieve train operation and various functions.As the core system of high-speed trains,the traction system provides traction force for the whole train,and its reliability is very important for the safe operation of high-speed trains.This thesis takes the high-speed train traction system as the research background,and through the data-driven fault detection method,it can effectively detect the internal faults of the train traction system in actual working conditions.The data-driven combination of autoencoder and variational autoencoder methods in deep learning enables accurate judgment and analysis of problems in the actual operation of trains.This thesis mainly has the following research work:(1)The operation principle and structure of the traction system of high-speed trains are introduced in detail.Through understanding the various faults that are easy to occur,the specific causes of the faults are analyzed,the sensor faults of modules such as traction inverters in the traction system are described,and then the data model is built(2)The current main methods of fault detection are analyzed,and the more novel methods of neural network detection are studied.Based on the analysis of autoencoder structure in deep learning method,the fault detection of the high-speed train traction system is realized by combining with data-driven piecewise autoencoder method.The sensor fault signals collected by the system are analyzed,and the fault points are accurately checked out,which proves the validity and feasibility of this method in the actual system,and compared with the traditional methods to verify its accuracy.(3)Based on the known piecewise autoencoder,a piecewise variational autoencoder is designed to achieve a better fault detection effect for the traction system.It can handle the same sensor fault accurately and efficiently.Its validity is verified by experiments,and a comparison test is carried out to achieve a better detection effect.