Downloads & Free Reading Options - Results

The long-range goal of this research program is to form a new system identification paradigm that fulfills all the requirements of robust control design, i.e., to produce from finite measured data a model which includes descriptors for both dynamic and disturbance uncertainty. Several directions were pursued during the reporting period: (1) least-squares related approaches (2) 'windsurfer' (learning) adaptation, and (3) uncertainty model unfalsification. The latter, together with some of the ideas in 'windsurfer' adaptation, marks a significant advance towards attaining the program objectives. The new idea is to replace identification of a single (transfer function) model with unfalsification of a family of uncertainty models and then to design a corresponding family of robust controllers to be tested on the actual system. (Both dynamic and noise uncertainty parameters are estimated along with the nominal transfer function).