(504e) An Ontology Based Cross-Hierarchical Multi Agent Collaboration for Monitoring Offshore Oil and Gas Production Processes

Offshore oil and gas production platforms are uniquely hazardous in that operating personnel have to work in a perilous environment surrounded by extremely flammable hydrocarbons. A failure in an equipment could quickly propagate to others resulting in leaks, fires and explosions. Such accidents could be prevented by deploying intelligent monitoring tools which continuously supervise the process and the health of equipments to provide context?specific decision support to operators during safety-critical situations.

Offshore production is a complex activity and no single FDI approach is suited to deal with all the possible failure scenarios. For instance, offshore platforms are characterized by several instrumentation and equipment level failures. Monitoring the overall process for detecting faults is not adequate as deviations in these are lagging indicators and the fault may be detected too late by which time platform safety may be compromised. However, different deviations are not independent and results from various instruments and equipment-level monitoring algorithms have to be combined considering the process connectivity and flows. Hence multiple levels of monitoring algorithms, ranging from the incipient detection of equipment related faults to process level faults for detecting human errors are necessary. We therefore propose an agent-based approach to collaboratively monitor offshore oil and gas production processes.

We have previously developed a Message Passing Interface (MPI) based architecture called CAMEO. In this work, its extension called ENCORE is reported. ENCORE uses an ontology to explicitly capture the hierarchy of the offshore platform, comprising the entire process at the highest level to the individual instruments and equipment at the lowest. Similar to CAMEO, each FDI method in ENCORE is modeled as an Agent. Each FDI agent can specialize in monitoring a different aspect ? at varying levels of granularity (tag level to unit level), scope, and monitoring methodology.

A key issue in monitoring any complex system using multiple independent methods in parallel is that the individual agents may not always concur. In this work, a consolidator agent is hence developed. The consolidator agent performs the matching between the results from the various FDI agents by mapping their evidences to the process ontology and seeking coherence among the various evidences using voting, Bayesian probability and Dempster-Shafer fusion strategies.

Finally, offshore platforms operate in a dynamic environment with numerous state changes (due to maintenance activities such as pig-launching, corrosion inhibitor injection, kinetic hydrate inhibitor injection etc) and disturbances. To effectively monitor the process, the FDI agents and the consolidator agents have to learn and adapt to the current conditions. Agents in ENCORE have this ability to continuously adapt to process and system changes as well as improve their innate data-based models by observing the outputs and maps of other agents.

ENCORE agents are compliant with the FIPA (Foundation of Intelligent Physical Agents) specifications; hence they can co-exist and cooperate with other agents developed in other implementation packages as well. The current implementation is also inherently multi-threaded; hence it can seamlessly scale-out to additional processors as computational load of monitoring increases during abnormal situations.

The multi agent framework as implemented using JADE (Java Agent DEvelopment Framework), in which the results of several FDI agents, each specialized in monitoring certain aspects of the overall process, are meaningfully fused by a consolidator agent based on the process ontology is demonstrated in this paper. Comparisons of the monitoring performance against other monolithic approaches are also shown.

References

Ng Y. S., and Srinivasan R., (2007). Multi-agent Framework for Fault Detection and Diagnosis in Transient Operations, Presented at the European Symposium on Computer Aided Process Engineering, Bucharest, Romania, Paper T1-474.