Controlling Multivariable Systems with Significant Uncertainty

Abstract The problem of designing a robust controller for a multivariable system with significant uncertainty is considered. In particular, the class of uncertainty is represented by a set of finite demensional linear time-invariant models of the system. A physically realizable, finite demensional linear time-invariant controller in unit feedback configuration is designed such that the resulting feedback system simultaneously satisfies performance objectives for all models in the set. The performance objectives considered include not only stability but also regulation, tracking, weak interaction and desirable transcient behaviour. A new approach to a robust controller design which reaches the above objectives is developed in a frequency domain setting, and its effectiveness of this approach is verified by the design and simulation of a papermachine control system. The result exhibits quite satisfactory performances and strong robustness.