Application of Extended, Unscented, Iterated Extended and Iterated Unscented Kalman Filter for Real-Time Structural Identification
Mohsen Askari,Jianchun Li,Bijan Samali
2012-01-01
Abstract:System identification refers to any systematic way of deriving or improving models of dynamical systems through the use of experimental and field testing input-output data. In the field of civil engineering, identification of the state of a structure during service condition under dynamic loading, such as earthquake, in order to detect any damage as it occurs, has posed a great challenge to the research community. Therefore, online and real-time structural identification has attracted a great deal of attentions in the structural engineering research over the past decades, especially when input-output measurement data are contaminated by high-level noise. Among real-time identification methods, one of the most successful and widely used methods for estimation of states and parameters is the Kalman filter and its various nonlinear extensions like Extended Kalman Filter (EKF) and iterated Extended Kalman Filter (IEKF). However, these methods are not effective in the case of highly nonlinear problems. To overcome the problem, two filtering techniques, namely unscented Kalman filter (UKF) and iterated unscented Kalman filter (IUKF), have been recently developed to handle any functional nonlinearity. In this paper, an investigation has been carried out on the aforementioned methods for their effectiveness and efficiencies through a highly nonlinear SDOF structure as well as a two-storey linear structure. Results show that, although IEKF is an improved version of EKF, the IUKF, in most of cases, produces better results on state estimation and parameter identification than UKF and IEKF. IUKF is also more robust to measurement noise levels compared to the other approaches.