An agent-based implementation of hidden Markov models for gas turbine condition monitoring

This paper considers the use of a multi-agent system (MAS) incorporating hidden Markov models (HMMs) for the condition monitoring of gas turbine (GT) engines. Hidden Markov models utilizing a Gaussian probability distribution are proposed as an anomaly detection tool for gas turbines components. The use of this technique is shown to allow the modeling of the dynamics of GTs despite a lack of high frequency data. This allows the early detection of developing faults and avoids costly outages due to asset failure. These models are implemented as part of a MAS, using a proposed extension of an established power system ontology, for fault detection of gas turbines. The multi-agent system is shown to be applicable through a case study and comparison to an existing system utilizing historic data from a combined-cycle gas turbine plant provided by an industrial partner.