The Modeling and Identification of Dynamically Uncertain Systems: Theory and Experiment

点击翻译按钮获取中文摘要

9634498 Smith The focus of this work is the development of methodologies for the modeling and identification of dynamically uncertain systems in a closed-loop control configuration. The work has a strong experimental component, in addition to the theoretical work required to develop the framework. The experiments serve as testbeds for the theoretical and algorithmic work, and addition, contribute to the development of feedback control technology within each of their application areas. A chemical vapor deposition system for semiconductor (GaAs/InP) growth is used as a closed-loop control and modeling testbed. The issues raised by this experiment include: uncertain flow dynamics; process nonlinearities; highly uncertain reaction dynamics; and model based sensing using optical measurements. A magnetic bearing experiment is also studied. A rotor shaft is levitated at each end by magnetic bearings. This is a particularly difficult identification and control problem and the relevant issues include: closed-loop identification of a multiple-input, multiple-output, unstable system; uncertain resonant modes in the rotor shaft; and rotor rotation dependent coupling between control loops. The third experiment is a radiant thermal processing system, consisting of an array of lamps for the heating of a plate. The issues involved here are: modeling of an infinite dimensional system; variable interaction between lamps; and significant uncertainty in the plate characteristics (emissivity, thermal coefficients). The theoretical side of the project is based on a robust control modeling framework. Uncertainty in the system dynamics is modeled by bounded perturbations to a nominal model. Prior work in the area of model validation has developed computationally tractable techniques for assessing the minimum size of the perturbations required to adequately describe an experimental observation. We propose extending the approach to develop methodologies for the experimental identification of model parameters for perturbed models. ***