Optimizing effective perceived noise in distributed electric propulsion with neural networks and differential propeller rotation

Aeroacoustic studies are critical for developing quieter aircraft, especially in urban areas. Propeller noise has two components: tonal and broadband. While broadband noise relates to small, low-Reynolds propellers, the tonal component, with its prominent frequencies, dominates sound pressure, disturbing humans and impacting metrics like EPNdB. Although it is difficult to change the noise signature of single or twin rotor aircraft, distributed electric propulsion offers alternatives. By operating propellers differentially, the noise signature can be significantly shifted while maintaining overall performance. This study aims to reduce the effective perceived noise from a distributed propeller system by optimizing individual propeller RPMs using a machine learning-based model. The method focuses on offsetting the tonal frequencies of each propeller to disperse acoustic energy over a broader spectrum, therefore reducing acoustic annoyance. Experimental data from wind-tunnel tests at Delft University of Technology will train a neural network surrogate model, which will predict noise levels during propeller speed optimization to lower EPNdB while ensuring required thrust. By achieving an optimal balance between noise reduction and aerodynamic performance, this approach seeks to advance aeroacoustic technologies and reduce the noise impact of next-generation aircraft.