Some Features of Quality Improvement of the Neural Network Identifying an Aeroengine Low-emission Combustion Chamber

Abstract

The paper aims to the problem of an automatic control of low-emission combustion chambers of gas-turbine aeroengine (DLN GTE) reliability by using mathematic models. These models are built into the engine’s automatic control system and serve as the virtual sensors, estimating unmeasured parameters. This study proposes a solution of this problem based on the neural network (NN). The structure of NN is determined by the available composition of input and output variables, the training sample. The task of an optimal neural network developing is reduced to the choice of its main parameters: the number of hidden layers and neurons in hidden layers. The sensor of pressure pulsation in the flame tubes of the combustion chamber and the emission of nitrogen oxides NOx and carbon CO based on neural network was built for DLN GTE. The main neural network design principles were discussed and tasted. The study of the influence of the number of neurons in the hidden layers on the quality of the neural network model was carried out.