Simultaneous Estimation of System and Input Parameters from Output Measurements

Abstract

System identification of very large structures is of necessity accomplished by analyzing output measurements, as in the case of ambient vibration surveys. Conventional techniques typically identify system parameters by assuming (arguably) that the input is locally Gaussian white, and in so doing, effectively reduce the number of degrees of freedom of the estimation problem to a more tractable number. This paper describes a new approach that has several novel attributes, among them, elimination of the need for the Gaussian white input assumption. The approach involves a filter applied to an identification problem formulated in the frequency domain. The filter simultaneously estimates both system parameters and input excitation characteristics. The estimates we obtain are not guaranteed to be unique (as is true in all other approaches: simultaneous estimation of both system and input possesses too many degrees of freedom to guarantee uniqueness); but we do, nonetheless, identify system parameters and input excitation characteristics that are physically plausible and intuitively reasonable, without making input excitation assumptions. Simulated and laboratory experimental data are used to verify the algorithm and demonstrate its advantages over conventional approaches.