| In the engineering practice,most of the mechanical equipment are running under non-stationary conditions,especially the phenomenon of speed fluctuation is obvious.For instance,wind turbine is composed of complex structure and many transmission components.Especially,the load and rotation speed has typical time-varying features.There are still some problems in the traditional time-frequency analysis method under variable operating conditions.For instance,the time-frequency representations is easy to be interfered by noise and ridges extraction,multi-channel signal features are not easy to be recognized,and the feature of impulse and strong modulation is hard to identified.To solve these four typical problems,this paper firstly studies the methods of signal denoising and ridge extraction.Then,based on Synchrosqueezing Transform,the improved time-frequency analysis methods is researched to accomplish high-resolution representations and accurate extraction for different time-varying features of complex vibration signals.Finally,the proposed methods have been applied to mechanical fault diagnosis.The main research work of this dissertation includes the following aspects:(1)Based on convex optimization,a novel vibration signal denoising and robust ridge extraction method is proposed.Different from the classical spectral analysis method based on inner product operation,this paper firstly considers the signal denoising as a problem of mathematical optimization.The theoretical framework of convex optimization for signal denoising is established and the dependence on parameter selection has been reduced.The sparse representations and preprocessing of one-dimensional vibration signal can be accomplished by using the non-convex penalty function.On this basis,the optimal time-frequency coefficient matrix in the time-frequency domain is taken as the objective function of the convex optimization problem.Then,the optimum time-frequency coefficient of multi-component signals and the robust ridge extraction under viable working conditions are realized by using the generalized minimum and maximum concave penalty function,which provide a basis for the ridge extraction and signal reconstruction in the subsequent Synchrosqueezing Transform.(2)A multi-channel signal denoising algorithm based on Multivariate Matching Synchrosqueezing Wavelet Transform is proposed.Different from the traditional single-channel signal processing method,this paper proposes a Multivariate Synchrosqueezing Transform on the basis of Matching Synchrosqueezing Wavelet Transform.Multi-sensor testing of key components is increasingly common in practical engineering applications.The co-existence of multi-channel signals can be used to accurately reflect the change of equipment operating state.Therefore,the multi-channel signal is firstly processed by Matching Synchrosqueezing Wavelet Transform to obtain the time-frequency representation.Then,based on the multi-modulation oscillation model,the shared characteristics of multi-channel signals are obtained,that is,the joint analysis spectrum is determined.Meanwhile,according to the bandwidth difference of different components,the adaptive time-frequency domain segmentation technique is proposed to realize the multi-channel vibration signal denoising for rolling bearings and gears.(3)Multiple Horizontal Synchrosqueezing Transform is presented for impulse characteristics identification.Different from the classical Synchrosqueezing Transform,Horizontal Synchrosqueezing Transform builds a new theoretical framework around the reassignment of time-frequency coefficients in the time direction.It has realized the accurate time location and time-frequency coefficient rearrangement along time scale by second-order group delay estimation,which is more suitable for extracting the strong impulse feature of multi-component signals.To further enhance the performance of time-frequency representation and increase the resolution in the strong noise environment,this paper studies Multiple Horizontal Synchrosqueezing Transform.The purpose is to continuously improve the energy concentration of the time-frequency plane through iterative reassignment procedure.Therefore,it can generate the sharp representation of time-frequency ridges to rolling bearing impulse signal.(4)Improved Synchrosqueezing Transform based on frequency rearrangement is presented to identify the strong time-varying feature.Different from the feature extraction method under constant working condition,this paper proposes the Multiple Second-order Synchrosqueezing Transform around the reassignment of time-frequency coefficients in the frequency direction.It fully combines the advantages of the Second-order Synchrosqueezing Transform based on Gaussian modulated linear chirp and the multiple iterative reassignment procedure.Since the instantaneous frequency of the strong modulated signal can be estimated more accurately by this method,the fault characteristic extraction and order spectrum analysis of the variable speed gear and bearing faults has been performed.Simultaneously,considering that Rearrangement Method realizes synchrosqueezing rearrangement both on two scales of time and frequency,it has better time-frequency representations ability,but it does not support signal reconstruction.Taking wind turbine as the research object,the paper defines compensation frequency distance and enhances the performance of Synchrosqueezing Transform by means of Rearrangement Method.It proposes to Reassigned Second-Order Synchrosqueezing Transform,which performed the order analysis of bearing vibration signal of wind generator without tachometer. |
