Publication Date:
2020-01-17
Description:
This paper investigates using machine learning to rapidly develop empirical models suitable for system-level aircraft noise studies. In particular, machine learning is used to train a neural network to predict the noise spectra produced by a round jet near a surface over a range of surface lengths, surface standoff distances, jet Mach numbers, and observer angles. These spectra include two sources, jet-mixing noise and jet-surface interaction (JSI) noise, with different scale factors as well as surface shielding and reflection effects to create a multi- dimensional problem. A second model is then trained using data from three rectangular nozzles to include nozzle aspect ratio in the spectral prediction. The training and validation data are from an extensive jet-surface interaction noise database acquired at the NASA Glenn Research Center's Aero-Acoustic Propulsion Laboratory. Although the number of training and validation points is small compared a typical machine learning application, the results of this investigation show that this approach is viable if the underlying data are well behaved.
Keywords:
Aircraft Propulsion and Power; Computer Programming and Software
Type:
GRC-E-DAA-TN75937
,
AIAA SciTech 2020; Jan 06, 2020 - Jan 10, 2020; Orlando, FL; United States
Format:
application/pdf
Permalink
