Neural Nonlinear Autoregressive Model with Exogenous Input (NARX) for Turboshaft Aeroengine Fuel Control Unit Model

Abstract

One of the most important parts of a turboshaft engine, which has a direct impact on the performance of the engine and, as a result, on the performance of the propulsion system, is the engine fuel control system. The traditional engine control system is a sensor-based control method, which uses measurable parameters to control engine performance. In this context, engine component degradation leads to a change in the relationship between the measurable parameters and the engine performance parameters, and thus an increase of control errors. In this work, a nonlinear model predictive control method for turboshaft direct fuel control is implemented to improve engine response ability also in presence of degraded conditions. The control objective of the proposed model is the prediction of the specific fuel consumption directly instead of the measurable parameters. In this way is possible decentralize controller functions and realize an intelligent engine with the development of a distributed control system. Artificial Neural Networks (ANN) are widely used as data-driven models for modelling of complex systems such as aeroengine performance. In this paper, two Nonlinear Autoregressive Neural Networks have been trained to predict the specific fuel consumption for several transient flight maneuvers. The data used for the ANN predictions have been estimated through the Gas Turbine Simulation Program. In particular the first ANN predicts the state variables based on flight conditions and the second one predicts the performance parameter based on the previous predicted variables. The results show a good approximation of the studied variables also in degraded conditions.