Machine Learning Modeling Of Advanced Air Mobility Rotor Inflow From A Mid-Fidelity Aerodynamic Solver

Abstract

In this study, two different rotor configurations were analyzed using DUST, a mid-fidelity aerodynamic solver, to investigate accurate rotor inflow prediction for advanced air mobility applications. A feedforward machine learning model was trained on the computed inflow fields and then integrated into a blade element momentum theory (BEMT) solver to evaluate its predictive capability. The trained model achieved an R2 of about 0.95 and a range-normalized root mean square error (RN-RMSE) of around 0.03 when predicting the airflow field under given test conditions. Furthermore, when these airflows were fed into the BEMT solver, the resulting rotor thrust and torque predictions reached an R2 of approximately 0.99 and an NRMSE of about 0.04, closely matching the reference analysis data. These results suggest that machine learning-based surrogate inflow models can significantly enhance the accuracy of real-time or near-real-time aerodynamic load predictions, and potentially aid certification processes.