An event-based plant alarm system is proposed as an alternative to the current variable-based alarm system used in the process industry. While a traditional variablebased process alarm indicates an unsafe deviation of a process variable, an event alarm will be a direct indicator of an undesired event. A conceptual framework for the design of such an alarm processor is outlined in this article. A methodology for allocation of alarms based on the information content of variables is highlighted. The annunciation philosophy for event alarms is detailed, and alarm prioritization method and operator guidance are discussed. The method uses Bayesian network and Bayesian inference to estimate probability of an event and to identify its root causes.

The concept of an event-based alarm system is introduced and a Bayesian method to annunciate event alarms is presented in this article. Unlike the current variable-based system, event alarms are allocated to abnormal events and are annunciated based on the risk of such events. The event-based alarm system design involves identification of abnormal events, identification of the sets of variables associated with each event and development of a model to predict the risk of each event using process measurements. Risk estimation involves the estimation of probabilities of individual events and their consequences. This article presents a method for estimating the probability of occurrence of abnormal events for the purpose of annunciation of an alarm. The methodology is demonstrated using the example of a simulated continuous stirred tank reactor.