Modeling, control, fault detection and isolation of chemical processes using a bond graph framework
A multitude of different approaches have been proposed to cope with external disturbances and unpredictable faults that are associated with chemical processes. A large number of these approaches rely on a model to deduce and predict the process performance. In many cases, the model neither represents the distinct physical domains (e.g., as reaction, hydraulic, mechanic, electric, etc.) nor their inter-dependencies and therefore is unable to predict critical information such as root cause of a fault. In this work, a unified modeling concept, the bond graph is used to model multiple domains. The basic variable of a bond graph is power that unifies the distinct domains. Additionally, the bond graph network reflects the physical structure in which power exchanges are tracked and the dynamics associated with power conversions can be captured quantitatively. Thus, a complete model of the process can be developed. Bond graph theory embodies causality - the cause and effect relationship between variables. This feature along with power conservation will be emphasized in this work to facilitate causal control design and fault detection and isolation. The methodology begins with extending bond graph theory into the realm of biochemical reactions, so that a unified modeling platform is obtained for biochemical processes involving biochemical reactions, hydraulics, and mechanics. Then a biochemical process for the purpose of wastewater treatment is used as the testbed to validate the extension. And with the completed model of the process, concepts on control design and fault detection and isolation under the unified bond graph framework are proposed and demonstrated using several examples.