Nonlinear Response Estimation of Structures at Unobserved Stories Using an Augmented Kalman Filter

Abstract

We conducted a study on non-linear response estimation in unobserved stories using Augmented Kalman Filter (AKF) in the former building of the Civil Engineering and Architectural Department in Tohoku University. Firstly, we selected small earthquake records that could be considered as linear responses using the Subspace method and estimated the layer stiffness and damping ratio of each story, which is required for AKF. Then, we applied AKF to earthquake records that could be considered as non-linear responses and examined the estimation accuracy of absolute acceleration and relative displacement in unobserved stories. The results showed that AKF is effective for actual building that exhibit nonlinearity, and that the estimation accuracy is improved especially when both acceleration and displacement are used as outputs, and that L-curve method can be used to set optimal values for the process noise of the nonlinear terms that greatly affects the estimation accuracy of the nonlinear response.