Data-based Uncertainty Modeling by Convex Optimization Techniques

A procedure based on convex optimization techniques for deriving normbounded uncertainty models for MIMO systems is presented. The procedure is developed for unstructured additive uncertainty models, but in principle this is no limitation since any uncertainty model of LFT type can be transformed into such a model. The models are determined by matching to process data available in the form of frequency responses of a set of individual models or sets of input-output data. Conditions for the existence of solutions to the data-matching problems are defined by LMIs. Uncertainty models that tightly match the data are obtained by minimizing an ellipsoidal uncertainty region. An application to distillation is included. Copyright © 2006 IFAC