Estimating Model Uncertainty using Confidence Interval Networks: Applications to Robust Control

Robust control techniques require a dynamic model of the plant and bounds on model uncertainty to formulate control laws with guaranteed stability. Although techniques for modeling dynamic systems and estimating model parameters are well established, very few procedures exist for estimating uncertainty bounds. In the case of $H∞$ control synthesis, a conservative weighting function for model uncertainty is usually chosen to ensure closed-loop stability over the entire operating space. The primary drawback of this conservative, “hard computing” approach is reduced performance. This paper demonstrates a novel “soft computing” approach to estimate bounds of model uncertainty resulting from parameter variations, unmodeled dynamics, and nondeterministic processes in dynamic plants. This approach uses confidence interval networks (CINs), radial basis function networks trained using asymmetric bilinear error cost functions, to estimate confidence intervals associated with nominal models for robust control synthesis. This research couples the “hard computing” features of $H∞$ control with the “soft computing” characteristics of intelligent system identification, and realizes the combined advantages of both. Simulations and experimental demonstrations conducted on an active magnetic bearing test rig confirm these capabilities.