Representing Procedural Knowledge in Expert Systems: An Application to Process Control

Abstract

The paper presents a novel expert system architecture which supports explicit representation and effective use of both declarative and procedural knowledge . These two types of expert knowledge are represented by means of production rules and event-graphs respectively, and they are processed by a unified inference engine . Communication between the rule level and the event-graph level is based on a full visibility of each level on the internal state of the other, and it is structured in such a way as to allow each level to exert control on the other . This structure offers several advantages over more traditional architectures . Knowledge representation is more natural and transparent ; knowledge acquisition turns out to be easier as pieces of knowledge can be immediately represented without the need of complex transformation and restructuring; inference is more effective due to reduced non- determinism resulting from explicit representation of fragments of procedural knowledge in eventgraphs; finally, explanations are more natural and understandable . The proposed architecture has been adopted for the design of PROP, an expert system for on-line monitoring of the cycle water pollution in a thermal power plant . PROP is running on a SUN-2 workstation and has been tested on a sample of real cases