Rotorcraft Acoustic Noise Estimation and Outlier Detection

This paper focuses on the application of artificial neural networks for rotorcraft acoustic data modeling, prediction, and outlier detection. The original data is recorded by microphones mounted inside a wind tunnel at NASA Ames Research Center, Moffett Field, CA. The experimental data is first acquired in the time-domain as a time history measurement; then the sound pressure level (SPL) that represents the acoustic noise in frequency domain is derived from the time history dataset. In this study, neural networks based models are developed in both time domain and frequency domain. Outlier detection is then performed using modified Z-scores for SPL data to find test points that are statistically inconsistent with the neural network model. Satisfactory computer simulation results are obtained.