Intelligent alarm filtering — A dynamic approach

Abstract Alarm filtering and fault diagnosis are becoming more and more difficult for human operators due to the complexity of process schematics, the interaction between various upstream and downstream variables, and the existence of time-delay. Time-delay and phantom alarms management is the most challenging task out of all. There are many existing fault diagnosis methodologies but only a few of them address the practical problem of phantom alarms. In this paper, a probabilistic knowledge-based alarm filtering expert system is described. The system employed a rule-based technique to handle cyclic loops, which may occur in the active causal network during diagnosis. For phantom alarm management, the system dynamically updated its causal network to include appropriate phantom alarms as exogenous causes. The system was implemented on a pilot distillation column and the results were satisfactory.