Data Mining Methods to Analyze Alarm Logs in IoT Process Control Systems

Abstract

Process industries use complex control systems to control manufacturing operations. Control systems collect a large variety and volume of sensor data that measure processes and equipment functions. Alarms constitute an integral component of data collected by control systems. These alarms are generated when there is a deviation from normal operating conditions in equipment and processes. With large number of alarms potentially occurring in a plant, it is imperative that operators and plant managers focus on the most important alarms and dismiss un-important alarms. This paper discusses a novel approach on how to reduce unimportant alarms in a control system and how to show operators the most important alarms using Sequence Data Mining and Market Basket Analysis concepts. These approaches help reduce the number of unimportant alarms and highlight alarms that can lead to expensive breakdowns or potential accidents.