(187c) Data Mining and Analysis of the Alarm Management

With the advancement in automation technology and connected devices, modern industrial systems routinely have thousands of sensors deployed in the field that are connected to control system units. In addition, most of these sensors are configured as alarms in the control system. This has resulted in a higher number of alarms, poor system performance, additional workload on operators, and in some cases has led to abnormal situations. These situations can further escalate to catastrophic incidents, if not managed properly. Although there are several solutions available to address these challenges related to the alarms and their management, most of them are complex, costly and time-consuming.

To overcome these challenges, we propose an alarm management framework which is relatively simple and efficient, and involves with four distinct levels - design, rationalized, advanced, and intelligent. This paper provides the details of a generic method developed with the use of data mining and analytics including pre-processing, similarity analysis, and correlation mapping. The method benchmarks the system for the defined KPI’s, identifies the bad actors, provides insights about the alarm system to manage the alarm flooding, and hence improves the overall efficiency of the system. An example historical alarm and event log dataset is used to demonstrate the proposed generic method, assess, benchmark the alarm system, provide details about the bad actors and assist in the overall decision support. From the example, it is clear that the new method results in reliable decision making, contributing to better alarm management and safer operations.