Statistical System Identification of Structures

Abstract

A general unifying approach to system identification is presented within a Bayesian statistical framework to explicitly treat the inherent uncertainties. It is shown that selecting the most probable model from a class of models for a structure based on its measured input and output leads to a rational and computationally feasible approach for response prediction. It is also asymptotically correct as the sample size is increased. The methodology is illustrated using an output-error formulation which has been successfully applied to recorded seismic motions from structures.