(162a) Alarm Management Using Data-Driven Methods

Alarm systems are a crucial part of modern control systems. They serve as a layer of protection and provide a prior indication to the operator about an abnormal process event. Alarms are designed and configured in the control systems and logged into historian for future references and reviews. With the progress in data mining and analysis methods, the alarm and event logs can be mined to generate diagnostic information about the healthiness and configuration of the alarm system. To demonstrate this, we propose a method to mine information from alarm & event log. This method is developed on an open-source programming platform. The input to the tool is an alarm and event log file from an asset management system or control system. This tool aids in (1) identifying and extracting alarm flood sequences, (2) providing graphs for visual analysis, (3) mining frequent alarm patterns and (4) bench marking the system against the Key Performance Indicators (KPI). The outputs can be used for offline analysis, rationalization of the alarm system, definition of suppression logics, and reduction in the alarm floods. The effectiveness of the proposed method and tool is established by application to a real industrial data set.