Benchmarking Neural Networks Architecture for Airfoil Database

This paper presents the contribution of machine learning (ML) models developed at Alfred GessowRotorcraftCenter(AGRC)attheUniversityofMarylandCollegePark(UMD)tothe Applied Aerodynamics Surrogate Modeling Benchmark Cases. Two cases are performed: Case 1 that uses OVERFLOW Machine Learning Airfoil Performance (PALMO) database to predict lift coefficient, drag coefficient and pitching moment coefficient for NACA 4-series airfoils for a range of Mach and Reynolds number. Case 3 to predict coefficient of pressure (Cp ) around airfoils using computational fluid dynamics (CFD) solver DLR TAU which is a Reynolds-Averaged Navier-Stokes (RANS) solver that uses Spalart-Allmaras turbulence model. Comparisons are made with the data presented in published literature for both cases. The ML framework developed at AGRC-UMD achieves higher accuracy in predicting the quantities of interest compared to conventional surrogate approaches, surpassing the standard benchmark results reported in the literature.