A Data-Driven Approach for Enhancement of Propeller Performance Prediction

Abstract

Accurate and efficient prediction of propeller aerodynamic and acoustic performance is of paramount importance to the analysis and design of various urban air mobility concepts emerging over the last decade. This paper presents the first exploratory effort in developing a data-driven framework to improve the prediction of propeller performance. In this framework, a multi-fidelity machine learning (ML) model, based on transfer learning (TL) and active learning (AL), is first trained using a large number of computationally inexpensive low-fidelity simulations and enhanced iteratively by a small number of high-fidelity aeroacoustic wind tunnel measurements using TL. Additional propellers which were manufactured and experimentally measured were intelligently selected based on an AL algorithm designed to minimize the predictive error of the ML model at two farfield observer locations. In the deployment stage, this ML model enables rapid prediction of the tonal noise level of isolated 2-bladed propellers at two farfield observer locations given the airfoil profile, pitch-to-diameter ratio as well as forward speed and rotational speed of the propeller. Results evaluated based on experimental data from two extra sets of propeller blades, which were not used in ML training process indicate that the multi-fidelity ML model trained with both low- and high-fidelity data based on the TL approach delivers far more accurate predictions than the ML models trained solely with low- or high-fidelity data. Furthermore, the inclusion of additional high-fidelity dataset selected by an AL algorithm was shown to further improve the predictive accuracy of the ML model.