ABSTRACT. Last year, the Deepwater Horizon’s disaster in the Gulf of Mexico caused human casualties, serious injuries and environmental catastrophes. According to the Abnormal Situation Management Consortium, petrochemical plants on average suffer this kind of major accidents every three years. Accidents like these demonstrate what may happen when the alarm system and operator response fail as a layer of protection in a hazardous process. In the process industries, people often struggle with reducing the number of nuisance alarms and yet be able to capture meaningful information from alarm flood during abnormal events. There is a big gap between reality and the industrial standards on alarms (EEMUA 191 – 2007 and ISA 18.2 - 2009). Although the demands from industry are huge, the effective approaches are still deficient, which needs support from academia to obtain knowledge from the abnormality and to improve the design of alarm systems.

The causality between process variables provides useful information to grasp the mechanism as to how a fault can propagate and track the root cause of the abnormality that requires special attention; this is important in incipient fault detection, process hazard assessment, smart alarm monitoring, and online operator guidance. Unfortunately, to capture this causality, analysis of binary alarm data is not enough; numerical process data and process knowledge, particularly the connectivity and topology information also has to be employed. In this talk, graphical models will be introduced for process topology description. Several emerging techniques that may help capture causality by fusion of process knowledge and process data will be discussed with case studies, such as the information-based causality capture, multivariate data visualization techniques, and the ontological modeling and reasoning.

 

BIOBRAPHY: Dr. Fan Yang obtained his B.Eng. and Ph.D. in 2002 and 2008 respectively, both from the Department of Automation at Tsinghua University. Since February 2009, he has been working at the University of Alberta in Canada. Currently he is a Postdoctoral Research Fellow with Prof. Sirish L. Shah in the Department of Chemical and Materials Engineering and Prof. Tongwen Chen in the Department of Electrical and Computer Engineering. His research interests include smart alarm monitoring, process safety analysis, qualitative fault detection, and modeling of large-scale processes.