Study On The Diagnosis Methods Of Composite Faults In Traction Drive Systems Of High Speed Train

High speed train has become the first choice for people to travel.As large and complex engineering systems,the traction systems of high speed train have a very important relationship with the safety of life and property of passengers.So its safety and reliability is very important.After the high speed trains having been put into use,the failures of the traction drive systems occur frequently.Based on this background,this paper uses the high-speed train traction drive systems as the research object.The diagnosis methods of three level inverter IGBT power transistor open circuit and current sensor drift composite fault,rotor broken bar and air gap eccentricity composite fault,the fault of two current sensors are studied.The specific contents are as follows:A diagnostic scheme using KPCA is designed for the composite fault of IGBT power tube open circuit and current sensor drift.Firstly,the optimized EEMD method is used to denoising the current signals.Then KPCA is used to analysis the data by detecting the Hoteling statistic and SPE statistic.And contribution values of principal elements are used to determine the fault types.The three phase current average method is finally used to locate the position of the fault IGBT power pipe.The data of different faults are collected through the fault injection benchmark(TDCS-FIB)simulation platform of traction drive control system to verify the effectiveness of the fault diagnosis method.Considering that the composite fault characteristics of rotor broken bar and air gap eccentricity are difficult to distinguish from current fundamental frequency and noise,a composite fault diagnosis method based on RBF neural network is proposed.Firstly,an adaptive filter using optimized RLS algorithm is developed to reduce noise in current signals.Wavelet packet transform and singular value decomposition are carried out to reduce fundamental frequency and extract fault features of stator current.The k-means algorithm constructs RBF neural network to distinguish rotor composite fault modes after optimized by subtractive clustering algorithm.By using the fault injection benchmark(TDCS-FIB)simulation platform of traction drive control system,the data of different fault modes are collected to verify the effectiveness of the fault diagnosis method.The fault feature is submerged under the closed-loop control of the traction system when two current sensors failed at the same time.So the fault free,single sensor fault and composite fault cannot be accurately distinguished.Considering that the samples of this kind of fault are simple and the types of faults are few,a composite fault diagnosis method based on wavelet packet and decision tree algorithm is proposed.The feature information of the signals is extracted by the wavelet packet,which is used to train the decision tree.Then the decision tree is applied to classify the fault types.Finally,using the data collected from the semi-physical simulation platform of Zhuzhou Electric Locomotive Research Institute,the composite fault diagnosis method is verified.