Robust approximation and control

An algorithmic procedure has been implemented. Its input data are the frequency samples of a MIMO high order (possibly infinite dimensional) model and its output is a controller for that system. The procedure considers the approximation error as additive uncertainty and designs for robust performance of a mixed sensitivity problem by using H/sup /spl infin// optimal control. For cases where not only modeling uncertainty but also other types of uncertainties are considered, μ-synthesis can be used to design for robust performance From a practical point of view, the same procedure can be used for robust identification. This is the case when the values of H(j/spl omega//sub k/) are corrupted by additive bounded noise and the uniform bound on the noise is small enough. This is an important practical extension of this methodology, because a controller can be obtained directly from experimental data. Although there are assumptions which should be made concerning the identified plant, these are less restrictive than the knowledge of a mathematical model of the (possibly) infinite dimensional system as is the case in the approximation procedure.