Analyzing the Robustness of Hybrid, Output-Only, Kalman Filtering–Based System Identification Method

Abstract

This paper investigates, in detail, the robustness of a previously introduced approach to output-only structural system identification using the random decrement method and unscented Kalman filter (RD-UKF) []. Unscented Kalman filters have been widely used for structural system identification and damage detection purposes. These filter’s divergence in estimating the desired states of a structural system with unknown excitations is a well-known weakness, considerably limiting their application. To overcome this difficulty, the current study initially employs the random decrement method to extract a system’s free decaying response from its measured responses. Subsequently, it applies an unscented Kalman filter to the extracted free response in order to estimate the system’s dynamic properties. Our previous study demonstrated this method’s proficiency. The present study conducts further sensitivity analysis to show the RD-UKF method’s robustness vis-à-vis different uncertainties in the process of identification. First, we estimate the stiffness and damping matrices of a three-degrees-of-freedom (DoF) system with three different kinds of excitations. Next, we examine the RD-UKF method’s robustness in 100 experiments (Monte Carlo simulation). Besides, it will be shown that the method is robust in addressing uncertainties related to mass distribution and missing data (sensor malfunction or a loss of communication connectivity) during the modelling and measurement process. The results of the study show that the RD-UKF method is sufficiently robust for all the uncertainties of the system identification process.