FALSE MAINTENANCE ALARM-SEQUENCE DETECTION USING FREQUENT PATTERN MINING

Maintenance management is one of the most critical
tasks in industry. It is generally carried by maintenance technicians
and engineers that monitor the health status of a production
equipment using failure detection systems. These systems merges
several senors data collected from PLCs in order to detect a possible
deviation from natural behavior of the production equipment.
If a deviation is detected, an alarm noti cation is then visualized
on the technicians’ dashboard. The technicians can then plan (or
not) maintenance actions depending on its severalty. In this paper,
we investigate the use of Frequent Pattern Mining methods application
on detecting false alarm-episodes, thus reducing the time
consumed in diagnostics phase. We propose a FPM framework that
can be generalized for most existing industrial maintenance cases.
We use the Apriori algorithm in order to mine frequent episodes
and construct association rules. These rules are then used in order
to classify alarm sequences to false/real alarm.