Machine Learning-Based Modelling Of UAV Noise Metrics In Real Conditions

Abstract

Over the last decade, there has been a marked increase in the use of drones for various applications including emergency and natural disaster response, payload delivery, aerial imaging and surveillance. There are currently many private and public initiatives that aim to further increase the number of drones and diversify their tasks, offering many associated benefits such as reduction in emissions by replacing traditional and more polluting options with electric unmanned aerial vehicles (UAVs). However, several challenges have to be addressed before a broader implementation is accomplished. One of such challenges is the reported noise annoyance produced by UAVs. This study focuses on the development of data-driven models to predict the dynamics of noise metrics during real UAV operations. Extensive outdoor experimental campaigns were conducted, where array-based measurements were recorded during several manoeuvres performed by different types of drones. Beamforming techniques were applied to improve data quality and signal-tonoise ratio (SNR), and to synchronize telemetry and acoustics data streams. Using the improved experimental data, an initial machine learning model was developed to predict the backpropagated OSP L as function of telemetry-derived operational parameters for ascent, hover, and descent. The model managed to accurately predict the experimental data, and it was found that the elevation angle was the most important predictor of OSP L for the considered manoeuvres.