Industrial alarms, Discrete Event Systems, alarm filtering, pattern mining, Petri nets

Abstract :

Alarm systems play an important role for the safe and efficient operation of modern industrial alarms. However, in most of industrial alarm systems,
alarm flows cannot always be correctly managed by dependencies contained in the alarm log are identified. To ease this analysis, the discovered
the operators as they often turn into alarm floods, sequences of numerous alarms occurring in a short period of time.
To reduce the alarm flows, this report focuses on detection of redundant alarms that could be removed. This objective is met by, first, looking for frequent adjacency in the alarm log. To identify them, the frequent pattern mining algorithm AprioriAll is adapted. Another way to find potentially redundant alarms is to look for systematic predecessors. To discover them, dominations and mutual dependencies contained in the alarm log are
identified. To ease this analysis, the discovered relations are depicted in the form of Petri nets.
Both those methods are then tested against an industrial alarm log made available by General Electric. The results show that both methods allow a
reduction of the alarm flow, with the biggest reduction being during alarm floods.