A data-driven controller design method with a minimum-variance objective and constraints on robust D stability, has been developed. The controller depends on data in three ways: via (i) an estimated model, (ii) an estimated model uncertainty (ellipsoid), and (iii) an estimated disturbance rejection. A simple procedure is used to place all closed-loop poles for the ellipsoidal set of models inside a specified D region. This gives robust performance. The shape of the D region is chosen by the user and defines the desired response. The approach has been applied to a magnetic suspension system that is unstable, nonlinear and has noisy sensors. High performance can be achieved, which extends the operational range of the suspension system. A tailor-made parameterization can improve both the estimation and the controller performance. It is also shown that more improvement in performance can be achieved by estimating the disturbance rejection (internal model principle) than by iterating between model estimation and subsequent controller design.
Holmberg U., S. Valentinotti and D. Bonvin. An Identification-for-control Procedure with Robust Performance. Control Engineering Practice, 8, 1107-1117
Valentinotti S., U. Holmberg, C. Cannizzaro and D. Bonvin. Modeling for Control of Fed-Batch Fermenters. ADCHEM 2000, Pisa, Italy (June, 2000), 491-496
Holmberg U., P. Myszkorowski and D. Bonvin. A Pole-Placement Approach that Matches Identification and Control Objectives. European Control Conference, Brussels, Belgium, Session TH-A-F 6 (July 1-4, 1997).