Alarm Management for Improved Process Safety, Reliability, and Product Quality

Most chemical plants have hundreds of variables that monitor the dynamics of their processes.  When the variables move out of their normal operating ranges, alarms are triggered to notify the human operators.  The frequency and selectivity of alarms influence the performance of operators, which in turn, impacts the plant performance and reliability.  This work introduces techniques to perform alarm management that improves the reliability of plants.  To ensure the relevance and specificity of alarms, particularly during abnormal situations, the work focused on providing operators with actionable alarm information.  Bad Actors (BAs), which are alarms that deteriorate the performance of alarm systems, are identified and resolved.  Typically, BAs are poorly-configured alarms that have the highest alarm counts, often contributing to periods of high alarm frequency (referred to as alarm floods).  In this work, they include alarms that experience the most-critical abnormal events (when variables cross their Emergency Shutdown thresholds) and/or lengthy departures from their normal operating ranges – jeopardizing plant safety, reliability, and product quality.  The latter are important for many plants operated unmanned over nights/weekends.

Several months of alarm data were analyzed for an industrial air-separation unit operated by Air Liquide.  The BAs were identified and assigned different grade levels.  Several Key Performance Indicators (KPIs) were developed to evaluate the performance of alarm systems.  The methodologies developed in this work improve upon the existing industry-standard techniques, which focus only on the alarm counts and not on the criticality of abnormal situations.  Results show that the average frequency of alarms was well under manageable limits, with the alarm system in the flood condition just 1% of the total time.  However, 45% of the total alarms during the study period occurred during alarm floods.  These alarm floods are referred to as near-misses, which often distract operators, but don't necessarily result in trips or accidents.