| Qualitative simulation is a promising technique for analyzing dynamic systems with incomplete knowledge. The QSIM algorithm for qualitative simulation provides a framework for constructing qualitative versions of process models normally represented by ordinary differential equations. Previous work with the QSIM algorithm modeled systems with relatively simple dynamics. In this work, qualitative models were developed for a variety of dynamic chemical processes including first-, second-, and third-order linear processes, nonlinear processes such as chemical reactors, multivariable systems, and a process under PI control. In addition, extensions to the QSIM algorithm were developed for reasoning about the higher-order derivatives that help lead to tractable simulations of many of these systems.Much of this work focused on identifying the necessary structure of the qualitative model that would minimize the number of spurious predictions. For some systems, the qualitative model required redundant representations of the process equations in order to capture quantitative knowledge lost in formulating the qualitative representation. Furthermore, many models needed additional reasoning techniques to cope with ambiguity inherent in qualitative mathematics. The most effective extensions placed additional constraints on higher-order derivatives to eliminate large classes of spurious behaviors. For every system studied, the qualitative model predicted all of the behaviors expected from a traditional quantitative analysis but without precise knowledge of the process parameters. When quantitative information was included in a qualitative model, the number of predictions generated was reduced and the resulting behaviors were more descriptive. |
