Neural Network Modeling of Measured Tiltrotor Acoustics for Designing Low-Noise Approach Profiles

A neural network was designed to model the acoustic results from a flight test of the XV-15 tiltrotor as a function of the aircraft's performance parameters. A judicious choice of the inputs to the neural network, and a thorough study of the flight test results, yielded a model capable of predicting the A-weighted sound pressure levels on a hemisphere surrounding the aircraft within 1.2 dBA during typical approaching flights. The neural network accurately predicted the level and directivity of highly nonlinear phenomena such as blade-vortex interaction (BVI) noise and high-speed impulsive (HSI) noise during steady-state descending flights at various combinations of airspeed, nacelle tilt, and flight path angle. The neural network model was combined with the quasi-static acoustic modeling (Q-SAM) method to predict noise hemispheres during slow maneuvering flights. The resulting performance/acoustic model was used to systematically design approach profiles that reduce BVI noise. A reduction of 8 dBA in the peak hemisphere A-weighted sound pressure level is predicted as compared to a baseline approach.