abstract = "Dynamic aeroengine models have an important role in
the design of real-time control systems. Modelling of
aeroengines using dynamic performance simulations is a
key step in the design process in order to reduce costs
and the development period. A dynamic model can provide
a numerical counterpart for the development of control
systems and for the study of the engine behaviour in
both steady and unsteady scenarios. The latter
situation is particularly felt in the military field.
The Viper 632-43 engine analysed in this work is a
military turbojet, so it was necessary to develop a
model that would replicate its behaviour as
realistically as possible. The model was built using
the Gas turbine Simulation Program (GSP) software and
validated both in steady and transient conditions. Once
the engine model was validated, different machine
learning techniques were used to estimate (data mining)
and predict an engine parameter; the Exhaust Gas
Temperature (EGT) has been chosen as the key parameter.
A MultiGene Genetic Programming (MGGP) technique has
been used to derive simple mathematical relationships
between different input parameters and the EGT. These,
then, can be used to calculate the EGT value of a real
Viper 632-43 engine knowing a priori the input
parameters and in any operating condition.
Finally, the EGT estimated by this algorithm has been
added to the dataset used for the one-step-ahead EGT
prediction by Artificial Neural Network (ANN). A
time-series ANN was used for the EGT prediction, i.e.
the Nonlinear AutoRegressive with eXogenous inputs
(NARX) neural network. This network recognizes the
input data as a real time series and is therefore able
to predict the output in the next time step. It was
chosen to use, as forecasting method, the
one-step-ahead technique which allows to predict the
EGT in the immediately next time step",