Regression-based modelling of a fleet of gas turbine engines for performance trending

Module performance analysis is a well-established framework
to assess changes in the health condition of the components
of the engine gas-path. The primary material of the technique is
the so-called vector of residuals, which are built as the difference
between actual measurement taken in the gas-path and values
predicted by means of an engine model. Obviously, the quality of
the assessment of the engine condition depends strongly on the
accuracy of the engine model.
The present paper proposes a new approach for data-driven
modelling of a fleet of engines of a given type. Such black-box
models can be designed by operators such as airlines and thirdparty
companies. The fleet-wide modelling process is formulated
as a regression problem that provides a dedicated model for each
engine in the fleet, while recognising that all engines are of the
same type. The methodology is applied to a virtual fleet of engines
generated within the ProDiMES environment. The set of
models is assessed quantitatively through the coefficient of determination
and is further used to perform anomaly detection.