Search

This paper presents an observer designed under the assumption that differences between predicted and measured outputs arise from discrepancies between the real structural system and the nominal model used to represent it. The observer gain is independent of the assumed model error parametrization and proves to be the transpose of the state to output matrix of a state space formulation. The estimated state with the proposed observer is shown to be identical to that obtained by exciting the nominal model with the known input while adjusting the measured portion of the state to match the measurements at the start of every step. Numerical experiments suggest that the proposed observer can provide state estimates that are substantially more accurate than results predicted by projecting the measurements in a truncated modal space. [ABSTRACT FROM AUTHOR]