Towards Data-Driven Aeroacoustic Modeling to Predict Propeller Installation Noise: A First-Principle based Approach

Abstract

Prediction of propeller installation noise is a critical aspect of developing low-noise aircraft and advanced air mobility systems. A novel data-driven approach, grounded in first-principles aerodynamics and acoustic theory, has been developed to accurately predict tonal noise generated by installed propeller configurations. This research combines mid-fidelity aerodynamic simulations using the DLR's Unsteady Full-Span Free-Wake Panel Method (UPM) with acoustic predictions based on the Ffowcs Williams-Hawkings (FW-H) equation. Two vortex methods, namely, the Vortex Filament Method (VFM) and the Vortex Particle Method (VPM), have been evaluated for modeling propeller tonal noise in both isolated and installed configuration. The present study shows that VPM is more suitable for the installed propeller configuration due to its superior ability to capture wake-wing interactions. As a result, VPM-based UPM simulations were utilized to generate a comprehensive dataset of far-field noise emissions. Sensitivity studies were conducted to identify key geometric and operational parameters influencing tonal noise emissions, and a reduced-order model was constructed using these parameters. A fully connected neural network (FCNN) was trained on the generated dataset to learn the functional relationship between input parameters and sound pressure levels (SPL) for the first five harmonics, as observed at an array of virtual microphones on an acoustic hemisphere. The model's performance was validated using flight conditions from the DLR DO-228 fly-over measurement campaign. The results demonstrate that the ANN model accurately captures tonal noise features, including installation-induced higher harmonics up to 1 kHz, with good agreement to measured data. These findings suggest that the proposed approach has significant potential as a fast and reduced-order tool for predicting installation noise in propeller-driven aircraft. Future work will focus on incorporating multi-fidelity simulations and measured datasets to extend the model's capability to include broadband noise prediction, thereby enhancing its utility in the development of low-noise aircraft and advanced air mobility systems.