State Characterization And Fault Diagnosis Research On Wheel Bearing Of High-Speed Train

With the rapid development of high-speed railway in our country and the rapid expansion of the road network, the running safety of high-speed train has become a key issue and top priority for train construction and operation. The wheel bearing is the key part of the train running system, which must support the train weight and the static and dynamic direction load, and the the additional unsteady force. So it has visible effect on the train running safety that the reliability and stability of the wheel bearing. With the improvement of train running speed, the dynamic load of wheel/rail is increasing, and the vibration of wheel bearing is more complicated, and the wear and abrasion of wheel bearing is increased. Thus, the state detection and fault diagnosis research has become very necessary and urgent for high-speed wheel bearing.Empirical Mode Decomposition (EMD), including Ensemble Empirical Mode Decomposition, is a new method of the data processing for non-stationary and nonlinear signal proposed by Huang E. Academician. It is considered as a major breakthrough in the linear and steady state spectrum analysis based on Fourier transform since 2000. EMD is an adaptive data decomposition method, which is based on the local time scale of signal itself, without the prior knowledge of the signal to be analyzed. Any non-stationary signal is decomposed into a number of Mode Function Intrinsic (IMF), each of which has a definite physical meaning. The instantaneous frequency and instantaneous amplitude of each IMF can be obtained, and the time-frequency distribution of the original signal can be clearly described. There are many parameters of the intrinsic properties in EMD decomposition method, which reflect some of the key inherent characteristics of the original signal. Therefore, based on the characteristics of EMD, this paper aims to study the state characterization and fault diagnosis method for the wheel bearing of high-speed train using the vibration signal of wheel bearings.In the process of combing the EMD time frequency analysis method, a new method for EEMD noise identification based on singular value curvature spectrum is proposed, which can effectively identify the noise signals generated by EEMD decomposition process and the noise signals contained in the original signal.Three quantitative characterization methods are proposed based the intrinsic property of EMD for the vibration state of high speed train wheel bearing. The first characterization method is based on the shape function of the vibration signal. EMD shape function is a new generalization for IMFs. Based on this. the standard form shape function of the vibration signal of the wheel bearing in health status is constructed in this paper. The measurement criterion of the shape function for failure wheel bearing is presented, which can effectively quantify the vibration state of the wheel bearing, and also can be used to identify the fault type of the wheel bearing. Stability test is the most important index to measure the performance of mechanical system. In order to avoid the prior condition of the shape function representation method, this paper puts forward the concept of EMD stable deviator. It can be used to quantitatively characterize the stability of the system and the degree of stability. The stability deviator index based on amplitude and instantaneous frequency are proposed, and the mearuing criterions for different status of wheel bearings are measured. The results showed that the fault types of the wheel bearing can be well identified by the stable deviator. In order to improve the problem of measurement interval mixed of the stable deviator measurement method, and based on the wide application of signal to noise ratio in signal processing and the simplicity of the method, the paper studies the signal to noise ratio calculation methods using EMD decomposition, and the strong mapping relationship between the signal to noise ratio and the state of the wheel bearing is established. It proposes three indexes of signal to noise ratio which are respectively based on the significance test, the normalized energy of dominant IMF and the singular value. This method has high identification ability and high adaptability, and is suitable for on-line intelligent detection and monitoring of high speed train wheel bearing.Based on the quantitative characterization of wheel bearing state, a new method of fault diagnosis for high speed train wheel bearing is presented, which is based on IMF automatic selection. The IMF value quantization theory is established to quantify the IMF value using self correlation, skewness, kurtosis and impact properties the IMF’s own property parameter. The high value IMFs can be selected to reconstruct the new signal, then the new signal is analysised through the Hilbert marginal spectrum, and the feature frequency will be highlighted. This method can effectively extract the fault characteristic frequency of the wheel bearing, and even can able to extract the fault feature of many kinds of faults in compound fault vibration signal simultaneously.In order to realize the full automation of whole process of fault diagnosis for high speed train wheel bearing, based on the theory of entropy weight, the stability of signal is described, and a method of fault diagnosis for wheel bearing based on EMD target entropy weight and SVM method is proposed. SVM is used as the basic tool to determine the target entropy weight of the wheel bearing, that is, the combination coefficient of IMFs relative time frequency entropy weight and IMFs singular value, based on which, the matching model is established. The method has strong modeling capability for small sample, nonlinear characteristic of fault data, and has higher classification accuracy, and can effectively identify the wheel bearing multiple fault states.In a word, this paper studied the status characterization and faults diagnosis for the wheel bearing of high-speed train comprehensively based on the characteristics of EMD. The research output can extract the information of the early stage or early fault for the wheel bearing health deterioration using the wheel bearing vibration signal, and grasp of the health status of the wheel bearing deviation with a high accuracy, and then distinguish the fault types of wheel bearing effectively. It would provide the theory and technical support for the on-line detection of wheel bearing, and provide state recognition basis and decision guidance for the realization of "state repair" for wheel bearing.