Deep Learning-Driven Optimization of Dynamic Morphing in VR-12 Airfoil During Pitching Motion

This study presents a surrogate-assisted multi-objective optimization framework for enhancing the aerodynamic performance of the Boeing VR-12 airfoil during pitching motion incorporating leading and trailing edge morphing at a Reynolds number of 250,000. Both the pitching and morphing motions are sinusoidal, with the pitching motion held constant throughout the optimization process. Morphing motion limits and frequencies are varied in an attempt to optimize the combination of pitching and morphing parameters for optimum aerodynamic performance. A deep neural network (DNN) surrogate model is developed and trained using a dataset generated from computational fluid dynamics (CFD) simulations, with sampling points selected via Latin Hypercube Sampling (LHS). The trained surrogate model enables rapid and accurate prediction of aerodynamic responses, significantly reducing computational costs compared to direct CFD evaluations. The NSGA-II algorithm is then employed to explore the design space and simultaneously optimize leading morphing parameters, specifically focusing on leading edge morphing frequency and phase lag. The results indicate that synchronized morphing—where the morphing frequency matches the pitching frequency and no phase lag is present—consistently yields the best aerodynamic performance. Additionally, the presence of another optimal solution at twice the pitching frequency highlights the potential for alternative morphing strategies to further enhance airfoil efficiency under specific operating conditions.