Linear and Nonlinear Indentification of Low Complexity Uncertain Models

University Of California-Riverside, Riverside CA

Investigators

Abstract

9912533 Chen This proposal initiates a nobel research program into the area of control-oriented identication, investigating linear and nonlinear methods for identification of low-complexity uncertain models. The principal goal is to develop new, theoretically sound and practically useful identification theories and techniques for modeling physical systems in a way compatible to control design theory and practice. The program is motivated and guided by problems of fundamental interest and problems of significant relevance to practicing engineers, with specific application areas focused on active noise suppression and combustion instability problems. The research plan addresses outstanding issues and calls for attention to new, emerging problem areas. It recognizes specifically computational and model complexity as the key obstacle, and nonlinear identification and model validation as the new thrust area, both of which are believed to have the potential to develop into important research areas in systems and control research. The main technical objectives to be accomplished are: To investigate a new paradigm of mixed deterministic/probabilistic identification problems. The task attempts to provide techniques and algorithms for identification and validation of deterministic uncertain models under probabilistic/stochastic noise assumptions. To develop nonlinear identification and model validation methods for special classes of nonlinear systems. The main focus will be on nonlinear systems with typical memoryless nonlinearities of engineering relevance. A central thrust will be the identification and validation of nonlinear dynamical systems with limit cycles. To test and gu ide the theoretical development via simulation and experimental work. Noise control and combustion instability problems are identified as application and testing benchmarks. A detailed research plan has been formulated to address these issues, which consists of concrete, realistic solution strategies that are believed to be both theoretically signficant and practically feasible. The plan seeks to merge theory, simulation, and experimentation, and is supported by well-established concepts and tools found in classical interpolation theory, convex optimization, stochastic processes, system identification, robust control, and nonlinear dynamics. It is felt that the successful completion of this project would significantly impact the contemporary as well as future theoretical and applied research in the areas of system identification and control. In the short term, it would advance signficantly the current state-of-the-art identification techniques, in development of computationally efficient identification algorithms. In the long term, it would lend insight and thrust to key issues and problems found in modeling and identification of nonlinear dynamical systems. Additionally, the application studies link the program directly to, and hence have the the potential to be of direct impact, on a number of ongoing research projects currently under pursuit in industrial R&D laboratories. The program is projected to span a course of three years from July 1, 2000 to June 30, 2003. Support is requested for the PI, for two summer months full time, each year, and for one graduate student, throughout the entire award period. The completed project will yield as main deliverables new, theoretically significant and practically feasible identification approaches, consisting of: a fully developed paradigm and algorithms for identification and model validation based upon a mixed deterministic/probabilistic setting, nonlinear identification and model validation techniques, worked-out benchmark examples and problems for the mixed deterministic/probabilistic paradigm and the nonlinear model validation methds, Tested noise control data and combustion models. ***