(662d) Identification of Hybrid Systems with Application to Fault Detection of a Reverse-Flow Reactor System

Many physical processes have dynamics that are well described using a hybrid modeling framework. This framework includes a finite number of continuous dynamical systems each representing a discrete state. Traditional approaches to model identification and fault detection may fail to distinguish between modeling errors and changes in the model dynamics due to a switch between discrete states. Recently, an algebraic approach to generating linear-hybrid models has been proposed. This approach allows models to be fitted to data without a-priori knowledge of the discrete dynamics. In this work, the suitability of such methods to a real-world noisy system is studied. Specifically, an algebraic polynomial approach is used to model the dynamics of a reverse-flow reactor system that exhibits several discrete modes of operation. Using this reactor data, the algebraic approach is benchmarked against both spectral (or frequency based) and traditional PCA-based identification and fault detection techniques.