A Failure Detection Method for Gas Turbine Sensors Based on Arx, Neural Network and Fuzzy Logic Models

In a free energy market the reduction of operating costs becomes a primary goal and, in this context, energy producers require the availability of assistance systems, with a particular attention to maintenance periods and fault prevention. In order to accomplish this task, utilities need diagnostic systems able to describe plant components health state, analysing the measured data signals in real time. The present paper deals with the FDI (Fault Detection and Isolation) environment, focusing on the problem of sensor diagnostics. This work describes the development of a diagnosis system able to individuate and isolate faults of different entity occurring in gas turbine sensors, provided that the fault affects only one sensor at a time. The applied methodology is based on the gas path analysis using the technique of analytical redundancy, by means of three different kinds of black box simulators: ARX, Artificial Neural Network and Fuzzy Logic. The sensors control has been effected disposing in parallel the three above mentioned models, in order to reliably identify the monitored sensors faults. The diagnosis system has been implemented in the Matlab-Simulink environment and tested on a data series, coming from a single shaft gas turbine power plant supplied by Ansaldo Energia.