A distributed fault diagnosis method based on digraph models : steady-state analysis

Abstract We develop a highly modular fault diagnosis methodology using digraph models of process behavior. Digraph models are appealing because they are easy to understand, they can be developed from either empirical or fundamental knowledge, and they can be analyzed using the host of results available from graph theory. A modular approach is desirable for implementation in modern distributed control systems. All the available digraph methods rely on a global view of the process data, so they are not easily implemented in a highly distributed setting. The present method is developed carefully from graph theory, and uses off-line analysis of digraph structure to reduce the on-line computation wherever possible. The on-line work is specifically designed for a distributed implementation in which each sensor and controller currently in an abnormal state examines the states for a small number of adjacent measurements to suggest possible faults, including both measured and unmeasured variables. Attention is paid to the issues of diagnostic completeness, diagnostic resolution and computational effort.