Research On The Fault State Identification Method Of Train Traction Seat Crack

The locomotive traction seat is an important part connecting the car body and bogie,which bears and transmits the longitudinal force of the locomotive,so the status of the traction seat affects the running safety of the locomotive.Taking the locomotive traction seat as the research object,based on the feature extraction and pattern recognition,the recognition method of the fault state of the locomotive traction seat is studied.Based on the combination of Ensemble Empirical Mode Decomposition(EEMD)and K-NearestNeighbor(KNN),and the combination of wavelet packet decomposition and Hidden Markov Model(HMM),the effectiveness of each method is verified and compared through experiments and analysis;on this basis,combined with the modal interval theory,the wavelet packet-HMM method is improved to the recognition method of wavelet packet-Generalized Hidden Markov Model(GHMM)and this method is verified.First of all,the structural characteristics,fault causes,fault classification and other aspects of the locomotive traction seat are analyzed,the specific parameters of the traction seat model is determined,the excitation frequency of the model is determined by modal analysis,the experimental platform of crack failure is built and the relevant experimental data are obtained.Then a recognition method based on EEMD-KNN algorithm is proposed.After feature extraction,three different states of traction seat are identified.The feasibility and effectiveness of this method are verified by the recognition results.Secondly,in view of the long time-consuming of EEMD feature extraction,which leads to the long process of EEMD-KNN recognition and takes up a large amount of memory,a recognition method based on wavelet packet decomposition and HMM model is proposed.The same data is decomposed by wavelet packet.Combining the time domain features,the proportion of wavelet packet energy is extracted as the sensitive feature and imported into the HMM model to establish and optimize the model base.The maximum log likelihood value is obtained by matching the samples to be tested with the model base,and the corresponding model is the recognition result.It is found that the feature extraction method of wavelet packetHMM has the advantages of small computation and short time,which improves the recognition efficiency and accuracy.Finally,in view of the uncertainty of the experimental process,the wavelet packet-HMM method is insufficient to analyze it,so an algorithm based on modal interval and HMM model is proposed.The GHMM model is formed by combining the mode interval theory with the HMM model.After the processing of the generalized data,combining the interval time domain features,the wavelet packet decomposition is also carried out to obtain the sensitive feature interval of the generalized characteristics,and GHMM model is established and optimized.GHMM model is used for state identification.The result shows that the method has better recognition accuracy,and effectively solves the problem of data uncertainty in the experimental process,and improves the reliability and reliability of the recognition results.