Qualitative-model simplification with an application to multistage separation processes

Abstract

Deep-model-based qualitative reasoning provides a systematic framework for describing qualitative physical phenomena. However, as physical systems become more and more complex, the computational load and complexity of qualitative simulation increase dramatically. Multistage separation processes are among the most important unit operations in chemical engineering. An industrial-scale multistage separation process is normally described by some tens or hundreds of quantitative equations. Qualitative reasoning for this type of process is difficult, if not impossible. Since multistage separation processes are highly structured (e.g. cascaded stages), the concept of model simplification (or model reduction), often seen in the practice of process control, is applied to qualitative modeling. The objective of this work is to simplify the qualitative process model according to the structural properties of the multistage separation processes. The simplified qualitative model can, on the one hand, reduce the computational load in qualitative simulation and, on the other hand, retain the qualitative description of the system. In this work, the simplest qualitative model, the signed directed graph (SDG), is used and the most common multistage separation process, using distillation columns, is studied. Qualitative simulation results show that qualitative-model simplification offers an attractive approach for reasoning about important, practical problems.