Semantic analysis of alarm response entries for plant health monitoring

In most health monitoring systems, there is always a final step in the process that involves the human response, and in many cases the engineer may decide that it is a false alarm. While everything is done to eliminate false alarms, understanding why the human has decided to ignore an alarm is imperative, and what that means for improving the performance of the alarming sensors. Here we deconstruct the manual reporting log of alarm responses for an in-situ health monitoring system for a series of identical plant equipment in a power station, by applying regular expressions to categorise the response. We can show how often certain types of event occur, but also compare the data at times when a user changed the alarm limits manually to see what was happening in the sensor array.