UKF estimation of SP-TAR models for the identification of time-varying structures

Abstract

An important class of methods for the effective identification of time-varying structures based on random vibration response data records is that of stochastic parameter evolution methods. Methods of this class rely on parametric time-varying models with a stochastic structure imposed on the time evolution of their parameters. The latter are considered as random variables allowed to vary in time, with their evolution being subject to stochastic smoothness constraints (smoothness priors constraints). In the present study, Smoothness Priors Timedependent (SP-TAR) models characterized by stochastic smoothness constraint equations with unknown a-priory coefficients are considered. The SP coefficients of the model along with the time-varying AR coefficients have to be estimated based on the measured response of the structure. This is achieved by expressing the generalized SP-TAR model in a nonlinear state-space form and employing the Unscented Kalman Filter (UKF) algorithm. The introduced method is validated through its application for the identification of a simulated gantry crane system.