Expert Systems Methodology for Real-time Process Control

Abstract The methodology of real-time expert systems is discussed, within the constraints of achievable paradigms and computational capabilities. Serious applications will require the representation of hundreds of objects, thousands of knowledge frames and often thousands of rapidly changing data inputs. Static expert systems based on snap-shot views of the process and pattern-matching paradigms cannot cope with such large, complex problems within current or nearterm expected computational capabilities. An expert system design for real-time problems requires the consideration of representation of temporal knowledge, efficient inference paradigm design, truth maintenance and other issues. A design appropriate for process control includes consideration of large knowledge bases, knowledge maintenance, knowledge validation, representation of deep knowledge and other issues. The authors present a survey of prior work, and a framework of specific criteria and design suggestions for real-time expert systems for process control.