Gas Turbine Modelling

It is becoming increasingly common for the controls available to the pilot of modern aircraft to be only indirectly connected to the control surfaces, engines, etc.. Instead the pilot's controls are connected to one or more computers whose job it is to map the demands of the pilot to the actual control surfaces or engine controls. The term used to describe this is fly-by-wire.

One advantage of fly-by-wire is that the controls can be calibrated in terms of a parameter of direct relevance to the pilot rather in terms of the degree of openness of the throttle, for instance. To the pilot a throttle control calibrated in units of thrust is more useful.

In order to understand the mapping between the various controls of an engine and the consequent thrust, an accurate model is needed. For research purposes thermodynamic models that take account of the complex combustion processes are constructed. These are extremely large programs. Despite their size such programs are sometimes unable to model some of the more subtle aspects of a gas turbine engine such as the effects of changing clearances as the engine heats up (heat soak) or as it wears.

The mapping between the positions of the various controls of an engine and its consequent thrust is highly non-linear and therefore a neural network is an appropriate choice of model. Furthermore, since the neural network requires only a corpus of training examples, it can easily be trained to model the effects of heat soak or wear.

The research carried out by Neural Solutions in this area yielded a gas turbine model that outperformed the conventional thermodynamic model and furthermore was able to include the effects of heat soak and wear. This was then used successfully to optimise the gas turbine, resulting in significant benefits in performance, economy and longevity.

next ...