Comparison of belief networks and rule-based expert systems for fault diagnosis of chemical processes

Abstract

Chemical process diagnosis requires a framework to represent and process uncertain and probabilistic knowledge. This paper deals with the use of probability theory as a framework to represent the uncertain elements of the diagnosis problem, and the use of distributed network (parallel) computations to determine the most probable diagnostic hypotheses. Knowledge is represented by belief networks, graph encoding and quantifying causal relations and conditional independence among variables. This representation is contrasted with a rule-based expert system to show how the probabilistic approach can overcome several limitations of expert systems. While constructing a belief network, possible knowledge sources, and their use, validity and integration, are discussed.