Metamodels of a Gas Turbine Powered Marine Propulsion System for Simulation and Diagnostic Purposes

The paper presents the application of artificial neural network for simulation and diagnostic purposes applied to a gas turbine powered marine propulsion plant. A simulation code for the propulsion system, developed by the authors, has been extended to take into account components degradation or malfunctioning with the addition of performance reduction coefficients. The above coefficients become input variables to the analysis method and define the system status at a given operating point. The simulator is used to generate databases needed to perform a variable selection analysis and to tune response surfaces for both direct (simulation) and inverse (diagnostic) purposes: two ANN are trained to reply simulator behavior in steady state conditions with considerably reduced calculation time (direct ANN), and to invert simulator inputs and outputs in order to obtain information on the system health state starting from measured variables (diagnostic ANN). The application of the methodology to the propulsion system of an existing frigate version demonstrate the potential of the approach for simulation and diagnostics: the simulation model behavior is replied with acceptable errors, while the health state of the propulsion system is successfully identified from the selected state variables by properly trained artificial neural networks.