Study On Poly-reference Time-frequency Domain Parametric Estimation Of Time-varying Structures From Output Observations

Engineering structures have been widely used in various mechanical engineering fields especially in aerospace engineering nowadays.Obtaining the dynamic characteristics of engineering structures has important application value and significance.As an important means to obtain the dynamic characteristics of engineering structures,modal parameter identification methods have been widely used in the engineering.With the large-scale,complex,and high-speed structures,the time-varying characteristics cannot be neglected.Study of modal parameter identification methods for time-varying structures becomes one of the research hotspots in structural dynamics.For the large-scale and complex time-varying structures under operating conditions,the excitations are complex and usually unknown,thus developing the modal parameter identification methods of time-varying structures based on response observations is of great significance.Modal parameter identification methods of time-varying structures are classified into time-domain methods and time-frequency domain methods,while the parameterized modal parameter identification method of time-varying structures in time-frequency domain has potential advantages in estimating high-damping structures and modal shapes.This paper studies on the parameterized modal parameter identification method of time-varying structures in time-frequency domain.In order to solve some problems existing in this kind of method,the following works are carried out:(1)To improve the identification ability of parameterized time-frequency domain methods for dense modes,a modal parameter identification method of time-varying structures in time-frequency domain based on right matrix fraction description is proposed.A right matrix fraction model is found,with which a single-reference form can be extended to a poly-reference form.On this basis,a poly-reference least square estimator of time-varying structures in time-frequency domain is developed.(2)For the problem that the least square method cannot make use of the uncertainty information and has poor ability in identifying the low signal-noise ratio(SNR)data,a poly-reference maximum likelihood estimator of time-varying structures in time-frequency domain is developed.To obtain better robustness,a logarithmic poly-reference maximum likelihood estimator of time-varying structures in time-frequency domain is developed.Since the covariance matrix are not available with single measurement,a novel approach of time-frequency-domain average method to estimate the covariance matrix is proposed and it is applied to the maximum likelihood estimation method.(3)To improve the non-parametric time-dependent power spectrum,a time-dependent power spectra estimation method based on vector time-varying autoregressive model is established.To improve the ability of time-frequency domain method to identify low-frequency weak modes,numerical integration is used to obtain the fake velocity instead of the acceleration as the original data to estimating the power spectra.(4)The proposed method in this paper is validated by a time-varying structural laboratory experiment,and compared with a classical time-domain method.The methods of specifying the bandwidth and changing the power spectra estimation method are used to improve the recognition effect of the proposed method.The following innovative results are achieved in the research.The thesis presents a parametric identification method in time-frequency domain of time-varying structures based on the right matrix fractional model,which improves the recognition effect of the parametric modal parameter identification method in time-frequency domain for the dense modes.A novel approach of covariance matrix estimation method based on the time-frequency domain averaging method is proposed,in case to solve the problem that the covariance matrix are not available for the single observation in the maximum likelihood estimator.The pseudo-velocity signal is obtained by numerical integration using the acceleration signal,and it can be used to improve the recognition ability of the estimation methods for the low-frequency weak modes.