Towards Generality In Modelling Complex Physical Systems

Qualitative physics has achieved considerable success in small and moderate scale problems but methods for managing the scope and depth of knowledge used by human problem solvers remain elusive. We argue that generality in a modelling approach is essential for extending qualitative physics to complex systems. A novel new model architecture is proposed which combines qualitative and quantitative knowledge in an ontology which draws from existing component and process based approaches. The significance of causal order in knowledge representation and inference is discussed and we introduce a new technique called dynamic binding which assigns causal order at run-time. We show how the context of a component is used to construct local perspectives of a complicated system and how this locality provides a unique demarcation in both knowledge representation and inference in the system architecture. Our mixed-model approach provides for physical modelling of engineering problems as a precursor to mathematical modelling or for interactive high-level querying by a non-expert user.