The concept of qualitative modeling is briefly described. This paper focuse
s on the use of signed directed graphs to model fault propagation in proces
s plants. It is common to use a unit-based approach for modeling. Two kinds
of problems, which arise when constructing unit-based qualitative models,
are identified: (1) The models are difficult to construct. (2) Ambiguities
due to multiple causal paths may occur when the models are combined. A meth
od that allows models to be built simply and correctly is proposed. The met
hod incorporates features from the knowledge acquisition field, and a compu
ter-aided modeling tool, Equipment Model Builder, has been built to support
this method. A new modular approach is presented which overcomes the probl
em of ambiguous inference when qualitative models are combined. Two case st
udies were used to evaluate the method developed and the support tool.