Diagnosis based on explicit means-end models

Abstract

This article describes three diagnostic methods for use with industrial processes. They are measurement validation, i.e., consistency checking of sensor and measurement values using any redundancy of instrumentation; alarm analysis, i.e., analysis of multiple alarm situations to find which alarms are directly connected to primary faults and which alarms are consequential effects of the primary ones; and fault diagnosis, i.e., a search for the causes of and remedies for faults. The three methods use multilevel flow models (MFM), to describe the target process. They have been implemented in the real-time expert system tool G2, in C, and in Common Lisp, and successfully tested on simulations of several processes.The knowledge representation ontology used is based on the notion of flows, of mass, energy, and information, which are used to describe physical systems. The relationships between structure and function of a system is described by teleological relations, which connect the flow structures into a graph, built at model construction time. This allows the diagnostic reasoning to be implemented as searches in a static graph structure, and it can thus be performed extremely rapidly. As with other model-based approaches, general algorithms are used over a representation with generative capacities. The representation gains strength from being functional with a very abstract physical level, more abstract than most qualitative physics models. It works well with systems that can be described using flows, while it currently lacks the capability of capturing important aspects of other types of systems, for example, electronic circuits.