Multi-objective optimization of a wing fence on an unmanned aerial vehicle using surrogate-derived gradients

In this paper, the multi-objective, multifidelity optimization of a wing fence on an unmanned aerial vehicle (UAV) near stall is presented. The UAV under consideration is characterized by a blended wing body (BWB), which increases its efficiency, and a tailless design, which leads to a swept wing to ensure longitudinal static stability. The consequence is a possible appearance of a nose-up moment, loss of lift initiating at the tips, and reduced controllability during landing, commonly referred to as tip stall. A possible solution to counter this phenomenon is wing fences: planes placed on top of the wing aligned with the flow and developed from the idea of stopping the transverse component of the boundary layer flow. These are optimized to obtain the design that would fence off the appearance of a pitch-up moment at high angles of attack, without a significant loss of lift and controllability. This brings forth a constrained multi-objective optimization problem. The evaluations are performed through unsteady Reynolds-Averaged Navier–Stokes (URANS) simulations. However, since controllability cannot be directly assessed through computational fluid dynamics (CFD), surrogate-derived gradients are used. An efficient global optimization framework is developed employing surrogate modeling, namely regressive co-Kriging, updated using a multi-objective formulation of the expected improvement. The result is a wing fence design that extends the flight envelope of the aircraft, obtained with a feasible computational budget.