Estimation of stationary structural system parameters from non-stationary random vibration data: A locally stationary model method

Abstract

Stationary structural system (building) natural frequency and damping parameters are estimated from a long non-stationary ambient vibration time series record. The non-stationary vibration record is decomposed into continuous, not necessarily equal duration, stationary time series segments by a locally stationary autoregressive (AR) model method. Estimates of the stationary structural system parameters are extracted from each of the locally stationary AR models. The final estimate of the structural system parameters is a time duration weighted average of the estimates in each stationary segment. Under reasonable conditions the theoretical variance of the structural system parameter estimates decreases with the duration of the long non-stationary record. The results of the analysis of a long duration non-stationary record are compared with the results of an earlier approach in which the long duration non-stationary record was decomposed into relatively short fixed duration vibration records and analyzed as stationary time series [1]. The variances of the damping parameter estimates achieved by the current method are one-half to one-quarter of those obtained by the earlier method. The sample variance of the structural system parameter estimates decreases with the reciprocal of the duration of the overall non-stationary record.