Real Time Alarm Processing for Predictive Failure Diagnosis in Petrochemical Plants

In this paper it is proposed a methodology for real time processing of alarms as a way to point out possible abnormal operating condition predictively. The proposed methodology consists of a knowledge base regarding critical scenarios. The critical scenarios are modeled by rules with expected alarm sequences for each deviation process. Each monitored scenario is assigned an occurrence index, which measures how close these scenarios are from happening. Finally, the system ranks the most probable critical scenario based on the similarity index with the expected scenario. In order to validate the proposal, a study was carried out using a simulated case in typical oil refining industrial plant.