Structural Parameter Identification of a Reinforced Concrete Frame Using Constrained Unscented Kalman Filter

Abstract

Field dynamic test on an as-built structure usually provides responses which are different from those generated by a corresponding finite element (FE) model. To update parameters in the FE model according to measured data, nonlinear Kalman filters, especially the unscented Kalman filter (UKF), can be applied by treating the model parameters as augmented system states. The UKF propagates the first two moments of the system states based on unscented transform, in which a set of sigma points approximating the state distribution are generated and transferred through the system equation. Although the UKF is a powerful tool for parameter identification, ignoring parameter constraints during the identification process may result in unreliable estimates. To address this challenge, this research investigates the application of constrained UKF (CUKF) on structural parameter identification. Different from the UKF, the proposed CUKF suitably generates sigma points in each iteration and make sure that all the sigma points locate within feasible region. As the weighted average of the constrained sigma points, the state estimates are guaranteed to follow the applied constraints. This paper also discusses the importance of weighting factors for sigma points for achieving second-order accuracy. Effectiveness and robustness of the proposed method are validated using experimentally measured data from a full-scale reinforced concrete frame structure. The identification results demonstrate that with properly applying parameter constraints, the proposed CUKF provides more reasonable estimation than the UKF. Finally, the updated FE model with identified model parameters is shown to achieve closer dynamic responses to experimental measurements than the initial model.