Investigation Of Fault Diagnosis Based On Singular Value Feature Extraction For High Speed Train Axle Box Bearing

With the rapid development of China's high-speed railway,the operating mileage and speed are constantly rising,the stability and safety of high-speed train operation have become the essential issue of research and attention in the railway field.As a key rotating component in a high-speed train,axle box bearing suffered not only the static and dynamic radial loads caused by train weight,but also the radial impact loads caused by track irregularity,wheel polygon effect,and the axial loads caused by over-curve or serpentine motion.Therefore,the stable and reliable operation of axle box bearing directly affects the stability of the whole train system health state.Once the bearing fails,the complex service environment will lead to the performance degradation of the bearing and its adjacent parts,even results in the occurrence of malignant accidents.Therefore,the research on the fault diagnosis method of axle box bearings is of great practical significance and application value for improving the reliability,safety and maintainability of high-speed trains in China.An extension singular value decomposition packet(ESVDP)method is proposed to extract the characteristic resonance frequency band of bearing fault signals adaptively.Based on the decomposition structure of singular value decomposition packet(SVDP),the proposed method improves the performance of extracting multiple resonance frequency band components by designing decomposition accuracy,also it improves the modal aliasing problem in the original SVDP method by reconstructing characteristic signals with paired singular values.Moreover,taking the energy of component signal as the screening index of effective component signals and combining with the decomposition structure of ESVDP,a fast algorithm of ESVDP method is proposed,which greatly improves the engineering application value of the method.The validity and stability of the method were verified by the simulation signal and the data of axle box bearing wheelset in a test rig.Through the researches of the cyclostationary and the pseudo-cyclostationary characteristics of fault bearing,a feature enhancement method based on periodic segment matrix for noise elimination in fault frequency band is proposed to separate the different types of fault signals in a resonance frequency band.Still,the rank-one characteristic of periodic segment matrix constructed by the cyclostationary signal has been studied and the spectrum norm of periodic segment matrix is used to reconstruct the periodic mono-component signal.The relationship between the signal-to-noise ratio and the matrix rank of the periodic component signal separated from the periodic mono-component signal is derived,which provides a reference for the design of the sequence length in the analysed signal,and provides a basis for setting the embedding dimension.Since the singular value ratio spectrum is not suitable for extracting the embedding dimension of the pseudo-cyclostationary signal,the research of the singular value distribution of cyclostationary signal is conducted.It is found that the periodic segment matrix constructed by the pseudo-cyclostationary signal is not rankone matrix anymore.Hence,the determining method of the embedding dimension and the singular value effective rank order using the energy differences singular value ratio spectrum and energy difference spectrum is proposed.The proposed determining method improves the accuracy of the periodic segment matrix and enhances the separation ability of the periodic component signal.The effectiveness of the method is verified by the simulation and test rig data.Comparing with many non-stationary signal decomposition methods,the significant advantage of the proposed method is separating different fault signals in a resonance frequency band,which greatly improves the accuracy and reliability of the following envelope spectrum analysis.In order to realize the adaptive interval recognition of the multi-fault rolling element,a method based on convolution averaging for recognizing the interval in the multi-fault rolling element is proposed.The time interval relationship is established according to the maximum pulses of adjacent pulse groups lied in self-rotation period for the multi-fault rolling element vibration signal,and the limitation of time-domain method in identifying the fault rolling element interval is analysed.The proposed multi-fault rolling element interval recognition method has high efficiency and anti-interference ability.