A data-driven controller design procedure is proposed in this paper. The co
ntroller is based on both an estimated plant model and its estimated uncert
ainty described by an ellipsoid in parameter space. Desired performance is
specified by the speed and the damping of the modeled response. The unmodel
ed response is rejected by requiring robust performance with respect to a g
eneralized stability region. Moreover, estimation of a disturbance model en
ables further rejection of the unmodeled response. The methodology is appli
ed to a nonlinear and unstable magnetic suspension system. High performance
is achieved for various specifications over a large operational range. (C)
2000 Elsevier Science Ltd. All rights reserved.