Nonlinear System Identification with Application to Combustion Instability Control

University Of California-San Diego, La Jolla CA

Investigators

Abstract

Abstract-0200449 Robert R. Bitmead, U. of Cal-San Diego Nonlinear System Identification with Application to Combustion Instabillty Control Robert R Bitmead, Department of Mechanical & Aerospace Engineering, U of California-San Diego. System Identification is the science of fitting mathematical dynamical models using measured data from experiments together with candidate model structures from physical reasoning. Identification is critical in developing models of complex and imprecisely known systems prior to their analysis. This project is concerned with developing techniques and underpinning theories for nonlinear system identification. At present there is a dearth of tools and guidance of how to proceed in this case. While the focus is scientific and broad, the study will proceed by considering in tandem with theory a major driving application of combustion instability control, of interest in jet engines and gas power turbines. The project has phases in which general concepts are developed and then are explored in the context of the combustion instabilities. Developing suitable measures of fit and then tests of confidence of models will be early goals. These should capture the important features in the data records from the nonlinear oscillating system while ignoring the unimportant detail. Here tests can be made with models already fitted to the data. Specifically nonlinear tools such as bifurcation analysis then can be applied for the assessment of model properties which are preserved from experiment to experiment and which might form the basis of control design. Chief among the theoretical tasks is to address the concept of model validation for such systems as oscillators. That is, how might one develop an experiment, which is capable of revealing the weaknesses of a particular candidate model fitted using some criterion. This is intimately connected with confidence in the identified model and with uncertainty in its suitability for a use. Should generalized concepts and meaningful tests be developed, there is a ready market for the application of these tools. Indeed, nonlinear control design has moved forward greatly recently in theory and practice, but it needs good models of known reliability to go much further.