Classification of SUAS propellers with auditory feature extraction methods

Abstract

A measurement of one stock and three custom designed propellers was conducted with the United States Air Force Academy. The measurement consisted of a constant radius arc, and a radial array to examine acoustic attributes as a function of distance and angle. During the measurement activity the experimenters' observed that each propeller possessed different audio attributes that assisted in distinguishing the stock from any of the custom propellers. To adequately explore attributes beyond the propeller's A-weighted level as a function of thrust, a timbre and sound quality analysis were conducted. These auditory feature extraction methods were combined with a fractional octave analysis into a database for machine learning classification analysis. The new baseline propeller is distinguished by the acoustic roughness alone, but the other blade designs require additional timbre features to be segregated from the stock propeller.