An Agent-Based Anomaly Detection Architecture for Condition Monitoring

Online diagnostics and online condition monitoring are important functions within the operation and maintenance of a power plant. When there is knowledge of the relationships between the raw data and the underlying phenomena within the plant item, typical intelligent system-based interpretation algorithms can be implemented. Increasingly, health data is captured without any underlying knowledge concerning the link between the data and their relationship to physical and electrical phenomena within the plant item. This leads to the requirement for dynamic and learning condition monitoring systems that are able to determine the expected and normal plant behavior over time. This paper describes how multi-agent system technology can be used as the underpinning platform for such condition monitoring systems. This is demonstrated through a prototype multi-agent anomaly detection system applied to a 2.5-MW diesel engine driven alternator system.