Incorporating engineering formalisms into automated model builders

Abstract

We present a new knowledge representation and reasoning framework for modeling nonlinear dynamic systems. The goals of this framework are to smoothly incorporate varying levels of domain knowledge and to tailor the search space and the reasoning methods accordingly. In particular, we introduce a new structure for automated model building known as a meta-domain which, when instantiated with domain-specific components, tailors the space of candidate models to the system at hand. We combine this abstract modeling paradigm with ideas from generalized physical networks, a meta-level representation of idealized two-terminal elements, and a hierarchy of qualitative and quantitative analysis tools, to produce dynamic modeling domains whose complexity naturally adapts to the amount of available information about the target system. Since the domain and meta-domain representation use the same type of techniques and formalisms as practicing engineers, the models produced from these frameworks are naturally communicable to their target audience. © Springer-Verlag Berlin Heidelberg 2007.