Model-based identification of primary causes

Abstract

In this paper we describe a model-based diagnosis system for dynamic processes. Modelling relies on a causal graph whose nodes represent significant process variables and whose arcs represent causal relations. Each arc has an associated propagation function that represents the way in which a change in a variable is propagated to another. Different modes of behaviour are defined through the association of different propagation functions to arcs. Propagation functions are defined in a qualitative arithmetic allowing the use of numerical values and the use of some approximate values.<>