Complexity and Experiment Design in Identification for Control

University Of California-San Diego, La Jolla CA

Investigators

Abstract

Relationships and interactions between the two disciplines of system identification and model-based control design are the prime milieu of the proposed research. The unanswered question, What is a good model for control design?, will be examined from a constructive viewpoint by considering a range of design choices within the methods of iterative identification and control, which deals with the recently developed techniques for successively generating controllers and models based on closed-loop data. In this context of iteration, each choice of design variable is informed (or prejudiced) by the previous model and controller plus experimental data. The design parameters which we study are connected with modeling in a framework where undermodeling and approximation are the norm, as opposed to the unrealistic aim of seeking exact descriptions. Both theoretical and practical areas will be studied, since each is an important guide and adjunct to the other in this field, where successful applications appear to be running ahead of the prevailing available theoretical support. The research has high scientific and practical merit, as it deals with fundamental problems of understanding the nature of modeling for a purpose and with issues of applied industrial control. Outcomes should be expected in both arenas. High performance control systems are regarded as a major enabling technology for many industrial applications. Since the provision of a dynamic model is the starting point for most high-performance control design, an understanding of the methodology for deriving this model from closed-loop operating data is critical to realizing this potential. The research areas addressed here are core issues whose resolution will provide the theoretical comprehension and guidance to yield useful practical techniques.