Joint Time Frequency Analysis And Their Applications On Nonstationary Vibration Processing

Vibration signals in mechanical systems always contain paramater characteristics and fault information of dynamical systems.In the area of modal identification,fault detection and excitation identification,advanced vibration signal processing is required to better understand system behavior.The traditional Foriuer analytical mathod has succesfully transformed signals into frequency domain.However,it is only valid for stationary signal since Foriuer transform only represents the global frequency components.Non-stationary vibration signals can be found in varisous engineering systems and always contain time-varying and/or nonlinear characteristics of the dynamical system.Thus,it is significent to obtain local informations of the vibration signals.Time-frequency analysis has become the unique and promising approach to deal with non-stationary signals in the last two decades.It enables us to analyze signals in time and frequency domain simultaneously.Several time-frequency analytical methods have been widely studyed during the last 20 years,e.g empirical mode decomposition,wavelet transform,Wigner-Ville Distribution and chirplet transform.These methods are all capable of describing time-varying frequency components.In this dissertation,time-frequency analysis theories are systematically addressed,and both advantages and limitation of each method are discussed.Seavral modified methods are presented to deal with those limitations.The presented methods are verified using both simulated and practical signals.The main reaserch work in this dissertation are as follows:Firstly,the dissertation reviewed some time-frequency analysis methods and discussed the advantages and disadvantages of each method.Secondly,analytical mode decomposition method is intruduced as an alternative of empirical mode decomposition in order to address the modal aliasing issue.Disrete wavelet transform is utilized to reduced the noise of the original vibration signal.The method is also capable of identifying the existence of nonlinearities in dynamical systems.The accuracy of the proposed method is verified by simulation results.Third,a complete time-varying modal parameter identification procedure is developed.Time-varying stiffness and damping are identified by using the ridge and skeleton of Morlet wavelet scalogram.The proposed method is verified by both single degree-of-freedom and two degree-of-freedom systems with abruped and smooth varying parameters.Simulation results show that the time-varying parameters are well tracked.Fourth,this dissertation point out the disadvantage of adaptive chirplet transformation in analysing nonlinear frequency modulated signals.A modified chirplet transform method is presented to address this problem.By intruducing a restriction in time domain support of the chirplet basis,the error brought by piecewise approximation of the original chirplet transform is reduced significently,especially in signals with fast mudulated frequency.Fifth,cross-terms in the Wigner-Ville distribution of multi-component signals are seriously considered.In this dissertation,a novel method is presented to obtain high resolution time-frequency distribution without cross-terms.A time-frequency filter constructed by wavelet scalogram is utilized to eliminate cross terms in the original Wigner-Ville distribution.The proposed method also has high computational efficiency and is verified by both simulated and practical vibration signals.At last,the proposed time-frequency analysis methods are verified using several practical vibration signals,including train wheel,vehicle-bridge system and echolocation signal.