Realistic wind modeling is essential to inform Advanced Air Mobility vehicle design, to test the limits of the operational envelope, and to ensure that trajectory planning, and control design can maintain high levels of safety. To represent urban winds, we performed Large Eddy Simulations to generate a wind field around the Boone Pickens stadium, located on the Oklahoma State University campus. We selected a realistic inflow condition based on an average wind magnitude in north central Oklahoma and generated the wind field over a several minute period. The Large Eddy Simulation is computationally expensive. To support numerous eventual Advanced Urban Mobility operations, it is not feasible to run such simulations to cover every Urban environment for every conceivable wind condition. Rather, we expect early demonstrated success developing complex wind fields using machine learning approaches will be leveraged to represent complex wind environments much more rapidly in and around cityscapes. Developing machine learning-based wind models requires some assessments to be made regarding the information content retained in the model. Thus, after performing the simulations of the wind field, we generated several reduced order model versions to assess the fidelity required.
To test the reduced order wind field models relative to the full-order model, we compare the dynamic response of a representative advanced air mobility platform operating in wing-borne flight as it traverses across our computational domain just above the stadium. The aircraft outer-mold line, mass and inertia characteristics, and flight speed come from a NASA tilt-wing design concept popularized in recent literature. The aerodynamic forces are from an unsteady vortex lattice method, which allows us to fully capture the time and spatially varying gust components. Spatial variation is captured based on location relative to the stadium, and by applying appropriate gusts to each vortex-ring panel on all the lifting surfaces by interpolating the high-resolution wind model. Early dynamic simulation results indicated the need to have some basic feedback control operating on the aircraft to keep the flight within the computational domain.
To assess wind model fidelity, we develop aircraft motion trajectories at various start points. Within each start point, start times are varied to fully sample the time varying nature of the LES derived wind. Statistical comparisons of trajectories show that wind-field fidelity is a minor effect when at and above the 80% information content threshold, particularly when the wind and flight direction are aligned. This is likely due to the high wing-loading of the concept vehicle. We do see significant variation in the control surface activity as information content is reduced from the full-order model. Overall, we suggest that the 80% information content to reconstruct the wind is a good threshold for this specific configuration and flight profile. While 50% ROM also captures the unsteady phenomenon, its effects on the AAM vehicle are less accurate when compared with wind from full order model, especially for the more active starting positions. Nevertheless, 50% ROM model could be comfortably employed for algorithms like path planning, especially for offline planning which requires a good approximation of the unsteady wind field. However as observed, 50% ROM could lead to grossly underestimating the energy requirements for control or control surface deflections. The mean wind model is clearly not appropriate for any application considering control activity or energy usage requirements.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner