A Machine Learning-Based Fast Prediction of Rotorcraft Broadband Noise

Abstract

This paper develops the data-driven fast-predicting models of rotorcraft trailing-edge broadband noise. The models are capable of predicting noise for the overall sound pressure level and the frequency-domain sound pressure level from basic rotor parameters, such as the tip Mach number, collective pitch angle, twist angle, rotor solidity, and rotor radius. A comprehensive noise data set used to train the models are generated from the rotorcraft broadband noise prediction program UCD-QuietFly, whose validations are presented against the measurements. The fast-predicting models are trained using two data-driven methods. First, the artificial neural network (ANN) is used to train the machine learning-based model. Second, the linear regression is used in which a polynomial equation along with linear combinations of the parameters is obtained. From the validations against UCD-QuietFly, it is found that the ANN model accurately captures the variations of the noise levels according to the rotor parameters and frequency. The linear regression models are also capable of predicting the general trends of noise levels with the rotor parameters.