Machine Learning Aided Low-Order Predictions of Fan Stage Broadband Interaction Noise

Abstract

A fast method for predicting turbofan fan-stage broadband interaction noise is being developed. The downstream propagating acoustic power in the bypass duct due to the response of the fan exit guide vane (FEGV) to fan wake turbulence is computed based on two-dimensional flat-plate cascade analysis and Green’s method. This study focused on using machine learning to define the fan wake parameters used as inputs to the FEGV response and noise calculation. Machine-learning algorithms are being trained using computational fluid dynamics results. This paper describes the accuracy of machine learning given the available rotor wake data. Further, the effect of errors in the learned input data on the acoustic prediction was studied. Based on this study, the method shows great promise.