Real time monitoring and the diagnostics of fault critical power plants, such as small scale gas turbine based ones, is proved to increase their availability and rentability. However, the diagnostic tool requires high speed calculators and fast response models to produce time efficient action, therefore encouraging model free application such as Artificial Neural Networks (ANN) techniques.
To this aim, the present paper investigates the performances of an ANN based diagnostic system realized for a small scale commercial gas turbine.
After a data harvesting campaign on two existing installations, a diagnostic tool (DT), provided by the gas turbine manufacturer, was operated to produce a sufficient comprehensive diagnostic data base to be used to training and testing the ANN system. The DT system evaluates performance deterioration causes (i.e. compressor fouling) modifying diagnostic parameters, such as efficiencies and combustion parameters, in order to meet data provided by thermodynamic simulation with data gathered on field therefore requiring a relatively high number of input variables (namely 9).
The results obtained show good agreement with the ones provided by the DT diagnostic tool with a percentage error not exceeding 4%. Moreover, the ANN model utilized in the diagnosis was implemented considering only 3 input parameters therefore resulting in a less complicated system, not considering the dramatic increase in execution time due to the lack of any iterative calculation.
The next step is to extend the monitoring to the overall gas turbine’s components and to integrate the monitoring system with an artificial intelligence based supervising system for diagnostic purposes which derives from ANN outputs damage presence, fault source and evaluation.