Damage detection of a large structure based on strong motion record. Theory of adoptive forward-backward Kalman filter

The report presents a new system identification procedure for a time-varing system to estimate natural frequency transition of a damaged building from a strong seismic motion record, named as 'Adaptive Forward-Backward Kalman Filter (AFB-KF)'. The AFB-KF is compared to the conventional Kalman filter in the below three points: (1) Forgetting Factor for corvatiance functions to track time-varying structural parameters rapidly, (2) Time-backward estimation scheme and global iteration scheme of the forward processes to estimate unknown initial value of structural parameters, (3) The time series renewal algorithm of statistical properties by reflecting the previous analysis information to improve the identification accuracy. It is useful to accurate identify from natural frequency transition of a building during earthquake for structural health monitoring which evaluates structural integrity. (author)