Health Condition Recognition Of Lateral Damper For High Speed Train

It is highly concerned that a health status identification maintenance system of certain key components is needed due to the rapid development and actual needs of highspeed railway in modern China.In this paper,lateral damper is used as a model of this maintenance system because of it plays an important role in high-speed train bogies.Related literatures indicated that vibration statement of car body and passengers' comfort are directly affected by the changes in damping parameters which cannot be detected during train running.Thus,the relations between health states of train component and signal data is needed to be built which leading to the revolution from regular maintena nce to state maintenance.However,it is still a missing part that define the early degradation states.In this research,a feature exploration and selection method has been proposed in a more effectively way.The paper proposes feature extraction and analysis methods of wavelet packet crosscorrelation analysis and composite feature selection model on the basis of the state identification of the conventional characteristic parameters of the degradation signal,aiming at the feature extraction of the degradation signal,Multi-feature selection and state recognition issues,the following research work was carried out:(1)The paper established a finite element model of the EMU,and carried out a series of simulation calculations after the flexib le treatment of it.The influence of the degradation on the dynamic performance of the train and other suspension components was analyzed.(2)Taking the vibration signals of different health states as the research object,the paper extracted the characteristic parameters of the vibration signal,and at differe nt operating speeds,the sensitivity of the degradation state was compared.(3)The paper introduced the wavelet packet cross-correlation coefficient on the basis of signal time-frequency domain analysis to establish the correlation between the vibration signals of each degradation state and the vibration signal in the normal state,and determined the degradation state of the component by quantitatively analyzing this correlation.The correct identification of the degradation state of verified the feasibilit y and effectiveness of the wavelet packet correlation coefficient index in signal feature extraction.(4)In order to avoid the defect that the single feature cannot comprehensive ly evaluate the quality of the components,the multi-dimensional feature vector was used to identify the health state.However,many features have different contributions to classification recognition.In order to select the most favorable feature subset for classification recognition,to better remove redundant features and bad features,and improve the efficiency of the operation,the paper constructed a Relief algorithm and Mahalanobis distance composite feature extraction model,the effective selection and dimensionality reduction of the feature information of the degradation signal.While reducing the number of dimensions,it can also significantly improve the state recognit io n rate and noise resistance.