Development of an intelligent system for detection of exhaust gas temperature anomalies in gas turbines

An unplanned outage can be costly for a utility, and gas turbines are expensive pieces of equipment to repair or replace. It is therefore vital that anomalous behaviour is flagged before damage can occur that may cause a prolonged outage. An anomaly detection system is proposed for gas turbines to monitor the related parameters and raise alarms when anomalies are identified. The proposed system incorporates machine learning algorithms based on artificial neural networks (ANN). By using ANNs trained on normal plant behaviour, it is possible to identify anomalous behaviour by the high residuals between actual and predicted outputs. Within this paper, the data mining methodology is described and the process followed before arriving at the successful approach is documented. Results from testing the approach on an industrial case study are presented and, based on these results, areas for further development are identified. It is intended to deploy the system along with several other algorithms as part of a multi-agent system for plant-wide condition monitoring. This paper will focus on the design and testing of the developed anomaly detection system.