Rotor-Airframe Interaction Noise: Predicting and Mitigating Noise with Artificial Neural Networks

Abstract

For small unmanned aerial vehicles, a rotor is typically mounted to the vehicle with a support rod. With the rotor operating in close proximity to the rod, an aeroacoustic installation effect known as the rotor-airframe interaction noise creates an unsteady deterministic loading noise that excites the harmonics of the blade passing frequency by as much as 20 dB, which meant that the installation effect could produce more noise than the rotor itself. As a result, it was imperative to parameterize and model the rotor-airframe interaction noise. Methods to predict the interaction noise were estimated using analytical and artificial neural network models. The artificial neural network required training data to correlate parameters with the acoustic noise, which was acquired from an experimental test campaign conducted within an anechoic chamber. Measurements were taken for numerous combinations of rotor radii, rotational speeds, rotor-rod proximities, and rod diameters to examine the parameter space expected of a rotor-rod configuration typically found on a small unmanned aerial system. Microphones were placed in a semi-hemisphere cluster to capture the directivity for the observer in-plane and below the plane of rotation. The analytical and artificial neural network models predicted the time-domain acoustic pressure emitted by the interaction, allowing for a more insightful inspection of the acoustic emission and providing a psychoacoustic analysis tool. In addition to acquiring an extensive database for training the artificial neural network, this database was also used to evaluate the prediction performance of the analytical model, which relies on a potential-flow model to represent the pressure fluctuations exhibited on the surfaces of the rotor and rod during the interaction. Results showed that the analytical and artificial neural network models had good prediction performance for lower harmonics. As the harmonic number increased above 11×BPF, the artificial neural network outperformed the analytical model due to the assumption built into the analytical model being invalid at higher harmonics. Efforts taken in the experimental test campaign also found that by curving the rod or sweeping the rotor blade, the rotor-airframe interaction noise was significantly reduced to the baseline case where the rotor operated without a rod within the flow. An analytical optimization method was developed to find the optimal rod shape to diminish the acoustic noise based on constraints, where the optimization algorithm can be quickly executed. In addition, a second study was conducted in an outdoor setting on the DJI S-1000 drone to demonstrate that the noise emitted by a drone could be reduced when a curved rod was installed instead of the conventional straight rod.