Experimental Training and Validation of a System for Aircraft Acoustic Signature Identification

This paper deals with the experimental validation of an innovative system for the aircraft acoustic signature identification which has been developed by the vibration and acoustics laboratory of the Italian Aerospace Research Center. The system is composed of an algorithm for the acoustic signature identification and a dedicated neural network classifier, trained with a set of experimental aircraft noise data. The algorithm test and validation has been performed for different airplanes during takeoff and landing maneuvers. The experimental activity of ground noise measurements has been carried out at the Naples airport of Capodichino. More than 200 aircraft noise events of five aircraft types (Airbus A320, Boeing B737, McDonnel Douglas MD80, Fokker F100, Aerospatiale/Alenia ATR72), during both takeoff and landing maneuver, have been measured. This paper demonstrates the feasibility of developing a suitable artificial neural network to establish if a time signal, elaborated through a wavelet process, is or is not similar to others, having been recognized as originated from a defined type of aircraft. The artificial neural network was trained by the use of a subset of experimental data and then validated through a comparison with another subset of data from the same experimental campaign. The developed software demonstrated to give more than satisfactory results for each of the acquired spectra, with the maximum error always being under (10)%.