Using neural networks to optimise gas turbine aero engines

A technique is presented for controlling gas turbine engines that maintains thrust while minimising fuel consumption. A neural network is used to model the engine. Fuel, which is one of the inputs to the model, is decremented and the model used to forward propagate the consequences, one of which is, typically, reduced thrust. Error derivatives to restore the thrust are backpropagated through the network and the error derivatives of the inputs to the model are then used to adjust controllable engine parameters to restore thrust. This iterative technique is continued until the optimum operating point is found. Changes to the engine as a result of temperature change and wear are tracked by updating the neural network engine model online.