Fault template extraction to assist operators during industrial alarm floods

In industrial systems, a fault occurring on a process can create an alarm flood, a succession of alarms raised at a rate per minute so high it overwhelms the process operator in charge of the monitoring of the process. In this paper, a method to extract fault templates from a set of alarm lists raised on the occurrence of several faults is proposed. Alarm lists generated by the same fault are condensed into a weighted sequential fault template formed of the sequence of alarms the most frequently produced on the occurrence of the fault. Each alarm is weighted according to its relevance to diagnose the fault. It is further shown how the fault templates can be used to extract relevant information on the alarm system and be used by operators as guidelines for fault diagnosis. Moreover, an on line fault isolation method using a weighted sequential similarity measure is proposed. The results obtained by the method on a data set formed of alarm lists raised by the control system of the CERN LHC connected to a simulator of one of the LHC processes are presented and discussed.