Monitoring behavioral evolution for on-line fault detection

The evolution graph monitor is defined and illustrated for on-line monitoring of computer-controlled processes with both discrete and continuous dynamics. Using a behavioral model specification of the process dynamics, the evolution graph monitor builds and maintain an efficient representation of the set of admissible system trajectories which are consistent with the on-line observations as they are received. In contrast to simulation-based methods, the evolution graph approach handles delayed, uncertain, and infrequent observations for systems with widely varying operating characteristics. A behavioral model specification and evolution graph monitor have been developed to perform on-line fault detection for an electric arc furnace in a continuous-caster steel mill. This application is used throughout the paper to illustrate the details of the approach.