Distributed data fusion for condition monitoring of graphite nuclear reactor cores

• Main Author: Wallace, Christopher John
• Published: University of Strathclyde 2013
• Subjects: 621.3
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.581920

Nuclear power stations worldwide are exceeding their originally specified design lives and with only limited construction of new generation underway, there is a desire to continue the operation of existing stations to ensure electricity supply. Continued operation of nuclear power stations with degrading and life-limiting components necessitates increased monitoring and inspection, particularly of the reactor cores, to ensure they are safe to operate. The monitoring of a large number of components and their related data sources is a distributed and time consuming process for the engineer given the lack of infrastructure available for collecting, managing and analysing monitoring data. This thesis describes the issues associated with nuclear Condition Monitoring (CM) and investigates the suitability of a distributed framework utilising intelligent software agents to collect, manage and analyse data autonomously. The application of data fusion techniques is examined to estimate unre corded parameters, provide contextualisation for anomalies in order to quickly identify true faults from explainable anomalies and to extract more detail from existing CM data. A generalised framework is described for nuclear CM of any type of reactor, specifying the required components and capabilites based on the design of a suitable Multi Agent System, including the interaction of the framework with existing CM systems and human users. A high level ontology for nuclear CM is proposed and is emphasised as a crucial aspect of the data management and extendability of the framework to incorporate further data sources and analyses. A prototype system, based on the generalised framework is developed for the case of the Advanced Gas-cooled Reactor, with new and existing CM analyses formalised within intelligent agents. Using real station data and simulated fault data, the prototype system was shown to be capable of performing the existing monitoring tasks considerably faster than a human user while retaining all data and analyses for justification and traceability of decisions based on the analyses.