Nonlinear Model-Data Fusion for Post-Earthquake Assessment of Structures

This paper proposes a real-time methodology estimate earthquake induced structural damage in instrumented buildings. The methodology relies on model-based state estimation to perform the assessment throughout the complete structure. In a similar fashion to existing data assimilation methods for nonlinear systems, the proposed observer uses sparse noise contaminated measurements of structural response and a model in order to perform the estimation. In contrast with existing methods, such as the extended and ensemble Kalman filters, the proposed observer does not require model linearization or Monte Carlo simulations. It is shown by means of stochastic simulation in a ten-story nonlinear building structure that the proposed model-based observer outperforms the extended Kalman filter and provides similar results to the ensemble Kalman but with the advantage of requiring only a small fraction of the computational time.