DYANA: Best-first hierarchical diagnosis based on differential equation models

Abstract

In this paper we evaluate and extend existing methods for an intended application for monitoring and diagnosis of a combined heat-power system. Only a few methods in model based diagnosis take advantage of the available design model by using (semi-)qualitative abstractions of ordinary differential equations, e.g. by Dvorak and Kuipers (1989), Ng (1990) and Lackinger and Nejdl (1991). Diagnosis of complex devices benefits from hierarchical models, both from a modelling perspective and from a computational perspective. This leads us to prefer the hierarchical diagnosis methoddiamon (Lackinger and Nejdl, 1991). We propose an extension of diamon, calleddyana, that is suitable for on-line diagnosis and uses an alternative method for dynamic model zooming which is based on heuristics regarding parsimony of diagnoses and consistency of fault models.dyana also uses numerical simulation, based on differential equations, instead of qualitative simulation.