(373g) Ontology Based Process Supervision System for Large Scale Chemical Plants

Process supervision is a set of activities for determining the state or condition of the process and controlling it to ensure foremost that it remains in a safe regime. The first step in process supervision, determining the condition of the process, involves interpreting real-time information from the process primarily from online measurements. This interpretation requires a model of the process that describes the expected behavior of the process, for a particular state. The development of this model requires various forms of information such as fundamental knowledge (chemical engineering principles), process design related information (flowsheet, instrumentation, etc) or data and information from operating experience (operating procedures, historical data, etc). Implicit in the development of most process supervision methods is the assumption that the underlying information, or descriptors, is static. Changes in descriptors, for instance a change in the flowsheet configuration during maintenance activities and retrofits is not uncommon and has been widely recognized as affecting the safety profile of the plant. The root cause(s) of many major accidents such as the Flixborough, Piper Alpha and PetroBras have originated from significant changes in the process flowsheet, equipment specification, or other such primary descriptors. Alarm management systems too highlight the importance of reflecting changes in the process descriptors to process supervision systems; omissions would lead to spurious alarms and alarm floods that obfuscate abnormal situation management. In other words, while process supervision methods are usually developed for a specific process configuration and modes of operation, modern day plants change significantly throughout their lifecycle.

Process supervision methods therefore need to be cognizant of changes in the underlying process and adapt to them. In cases when the process or instrumentation structure changes during operation, the results from process supervision methods, developed assuming a static nature, would be unreliable. Such adaptation requires that each process supervision method is (1) explicitly aware of the information that has been used for developing its model(s), and (2) able to respond to the changes as necessary or become inactive so as not to generate misleading results. Specifically, since supervision models are dependent on disparate information drawn from distributed sources, an explicit linkage to each is essential. Shared semantics based on a common ontology offers a way to develop these linkages. Hence, an ontology, specifically suited for process supervision is developed and presented here. This ontology forms the basis for the development of a multi agent based process supervision system called ENCORE. The performance of this system under several scenarios is reported for the Kongsberg Generic Oil and Gas Production Simulator.