Neural networks for modelling aircraft noise radiation

Abstract

This paper deals with a method for estimating the spectra of aircraft flyover noise radiation by means of neural network (NN) techniques. The method employs geometrical flight path data obtained by radar and one-third octave spectra of aircraft flyover noise, recorded simultaneously on the ground to train an NN-spectra model of flyover noise. The procedure uses a standard neural network backpropagation algorithm applied to minimise the error between measured and network-simulated octave or one-third octave spectra of the noise. The outputs of the NN-noise radiation model are octave or one-third octave spectra of aircraft noise (normalised to standard atmospheric/weather conditions) at some reference distance around an aircraft as a function of frequency, polar directivity angle (emission angle), azimuthal directivity angle (elevation angle), flight speed and thrust settings. Knowledge of these spectra opens up a wide range of new possibilities in the domain of aircraft noise prediction. In particular, phenomena such as radiation dirctivity of aircraft noise, atmospheric attenuation and weather conditions during the propagation, local ground effects, reflection, and conversion of received spectra to time dependent A-level or Dlevel flyover noise history can be incorporated correctly in the prediction methodology, without the need of restoring to empirical approximations like lateral attenuation correction. The proposed methodology is illustrated for the case of NN-modelling of flyover octave spectra, measured for a small body commercial aircraft.