The Application In Safety Estimation Of High-Speed Train Based On MEMD

With the continuous improvement of the running speed of high-speed train,the guarantee of train's safe operation is facing a higher challenge.In the course of train running,the sensors installed on the train monitor large amounts of vibration signal data.Analyze these data by signal feature analysis method and obtain the running state of the train in service by the classifier.So as to evaluate the safety state of the train quickly and accurately.There is a strong correlation between the sensors arranged on the high-speed train.The existing feature analysis methods can only process single channel in the process of feature extraction of high-speed train monitoring data,and then carry on data fusion on feature level,which is not conducive to the analysis of the vibration signals generated by the same physical system.Aiming at the multi-channel problem of high-speed train monitoring data,this thesis analyzes the Multivariate Empirical Mode Decomposition(MEMD)algorithm and researches the application of MEMD algorithm in high-speed train signal processing.The main work of the thesis are as follows.1)This paper studies the MEMD algorithm.To reduce the aliasing phenomenon further,the Gauss white noise is added to the MEMD decomposition process,called the noise assisted MEMD.The performance of the noise assisted MEMD method is evaluated by using the data of the full disassembly of bogie.The results indicate the noise assisted MEMD is more advantageous than the standard MEMD,and the best performance is achieved when the number of noise channels is 2.2)A feature analysis method based on MEMD and information entropy is proposed.The MEMD algorithm is used for the pretreatment of two kinds of fault signals,including the full disassembly of bogie and different number of lateral damper failure.The correlation coefficient is used to filter the decomposition results.Then three kinds of information entropy,including sample entropy,permutation entropy and multivariate multi-scale entropy are extracted from the selected intrinsic mode function(IMF).Finally,the extracted features are classified and recognized.The MEMD method is more effective than other feature extraction methods verified by the single variable method.3)The MEMD decomposition experiment of two kinds of fault data of the bogie is carried out by combining different number of channels,and the optimal number of channels that can describe the complete motion of the high-speed train is obtained in the corresponding fault state.The results indicate that for the full disassembly of bogie,the direction of transverse and vertical of a single measure point can give relatively complete description,so it only need to combine 3 channels for decomposition can achieve the best results.For different number of lateral damper failure,the channel including the direction of transverse and vertical and portrait can give complete motion description of high-speed train,so the identification effect of combining 6 channels for decomposition is the best.4)To research the state estimation in the process of performance parameters degeneration of damper,this paper presents evaluation method of performance parameter degeneration degree based on MEMD and compressed sensing.Decompose multi-channel data using MEMD.Extract information entropy feature of IMF,obtaining the original high dimensional feature set.Then the compressed sensing is used to reduce the dimension(The optimal dimension of feature is determined by Fisher ratio),obtaining low dimensional features with redundancy removed.The low dimensional features are classified and identified by the classifier.The results show that the low dimensional feature after compression and reduction by compressed sensing algorithm can effectively recognize phase states of damper during its performance parameter degeneration.