Data-based quantification of process nonlinearity

Abstract

The field of quantification of nonlinearity has drawn much attention from researchers in recent times. However, the scope of research on quantification of nonlinearity, especially for process plants, remains a vast area to explore. Nonlinearity in process plants may arise from two factors, namely, instruments and process characteristics. Instruments can induce nonlinearity due to their nonlinear dynamics or faults such as stiction, hysteresis and saturation. Process may themselves be nonlinear in nature or may show nonlinear behavior due to violation of some physical limits or constraints. Nonlinearity quantifications are useful for many purposes such as checking the adequacy of a linear controller for a nonlinear process and finding the root cause of a fault that arose due to increase in nonlinearity of the instruments. Model based quantification of nonlinearity is difficult, expensive and time consuming. On the other hand, data-based nonlinearity quantification methods are easy to use and becoming popular due to readily available data from Distributed Control System (DCS) or data historian of the plants. Four data-based nonlinearity measures, namely, Bicoherence based, Surrogate Data based, Correlation Dimension and Maximal Lyapunov Exponent are studied and their performances are compared in this thesis. Both simulations and experimental investigations have been undertaken to compare and find the suitability of the data-based nonlinearity measures for quantification of nonlinearity for chemical processes. Bicoherencebased measure was found to be the most successful among them and the Surrogate data-based measure was next to it.