Modeling process diagnostic knowledge through causal networks

Diagnosis in process industry is a complex task which can be effectively supported by knowledge based system technology. The majority of applications in this field have focused on rule-based implementations of the heuristic classification approach. In this paper we argue for the exploitation of a causal model based approach to knowledge representation and diagnostic reasoning which can overcome some of the problems manifested by first generation expert systems. The proposed approach can be more viable than other model-based approaches requiring a complete qualitative or quantitative modeling of correct behavior. The paper presents the causal representation formalism, a definition through a logical framework of the process diagnosis problem considered and the reasoning process which has been defined to solve it. Examples concerning both representation and reasoning issues are taken from DIOGENE, a system devoted to support on-line diagnosis of steam generation process in thermal power plants.