A multimodel-based approach to robust and self-scheduled control design

This paper presents a design technique that aims at “flattening” the real-μ-curve corresponding to a given initial feedback by combining μ-analysis and multimodel design. μ-analysis is used sequential to detect worst cases. Multimodel design is used for treating simultaneously the set of worst cases. The resulting controller can be nonscheduled or scheduled with respect to measured (slowly) varying parameters. This paper is mostly devoted to multimodel design. The multimodel design problem is written in terms of eigenstructure assignment. Two classes of resolution techniques are presented: first, the dynamic feedback is written in a transfer matrix form, second, it is written as an observer structure