Hybrid Model-Based Framework for Alarm Anticipation

Abstract

Modern chemical plants consist of a number of integrated and interlinked process units. When an abnormal situation occurs, the automation system alerts the operators through alarms. In this work, we introduce a new type of alarms, known as anticipatory alarms, aimed to enable operators to orient holistically to the abnormal situation. These anticipatory alarms are developed based on an alarm anticipation algorithm that utilizes dynamic process models to offer an accurate short-term prediction of the process state. In particular, these models predict the rate-of-change of process variables, which are then translated into predictions of time horizons for occurrence of various critical alarms. Anticipatory alarms seek to improve the sensemaking facilities offered to the operator through advance warning of impending alarms. As a result, operators can adopt a more proactive approach in managing abnormal situations. The benefits of anticipatory alarms have been demonstrated through six fault scenarios in a depropanizer unit case study. All alarms are successfully predicted, providing a diagnosis time benefit of around 35 s to the operators.