Compositional Modeling of Physical Systems

Automating analysis of physical systems requires techniques for managing complexity and finding an appropriate model for an analysis . Compositional modeling is a strategy for organizing multi-grain, multi-perspective models of physical phenomena which addresses these problems . In this paper, we identify several limitations with the approach presented in [7] and describe important extensions that address these limitations . First, we separate ontological commitments (e .g ., energy flows, contained stuffs, etc) from granularity commitments . We present a new framework for explicitly reasoning about the appropriate ontology and the structural nesting of systems . Second, we provide a generalized modeling language which enables the compositional modeling approach to be applied to either quantitative or qualitative models . We show that model composition cannot a priori determine that a quantitative model will be sufficiently accurate . However, we show that by using explicit modeling assumptions, inaccurate models may be detected and these inconsistencies can be used to focus a new round of model composition, leading to using more accurate, but more costly, models only when they are needed . We describe an implemented algorithm for model composition, which can automatically ascertain the appropriate modeling assumptions to answer simple questions about physical and functional characteristics of its operation .