Diagnostic models for engineering process management: A critical review of objectives, constraints and applicable tools

Abstract

Abstract Automated and computer-based systems for the dynamic management of complex engineering processes have been the subject of active research in recent years. Many of these systems have adopted methodology and technology from the field of artificial intelligence and expert system research. In real-time process diagnosis and process management applications, where model based reasoning may be highly beneficial in terms of speed and functionality, a larger confluence of methods and approaches has started to form. The objective of these applications often includes, besides performing automated fault analysis and diagnosis, also establishing levels of confidence for the results that are obtained. Thus, both deterministic and uncertainty-based models and reasoning frameworks are often needed. Of interest among the deterministic modeling techniques that can be used in a PMS (Process Management System) are binary trees, rule networks, and graph networks, which in turn include influence diagrams, logic flowgraph methodology and signed directed graph. PMS uncertainty management can be based on formal probabilistic methods, such as Bayesian estimation and updating, or less traditional methods such as certainty factors, Dempster—Schafer theory and fuzzy-set theory. The main features, advantages and disadvantages of the approaches and methods that are suitable for use in a PMS are critically examined and discussed.