Aerodynamic optimization of a Martian drone integrating propulsive effects and neural-network viscous corrections

Abstract

This work presents an optimization procedure for the conceptual design of a fixed-wing Martian exploration drone, combining low-fidelity aerodynamic computation with viscous corrections assisted by neural-network. The aircraft geometry is described through a dual aligned parameterization of airfoil sections and wing planform, enabling synchronized updates of airfoil shape and wing planform within the optimization loop. The methodology is assessed for a $1\mathrm{km}$ low-altitude mission in the Martian atmosphere and an estimated free-stream Mach number of $M_{\infty }=0.2$. The optimization maximizes the lift-to-drag ratio under mass and geometric constraints, while enforcing a preliminary longitudinal static-stability requirement. Aerodynamic performance is evaluated through a fast hybrid framework in which a Vortex Lattice method predicts the three-dimensional aerodynamic loads, while a Physics-Informed Neural Network provides viscous and profile-drag corrections guiding the objective-function. The optimized configuration achieves a lift-to-drag ratio of approximately ${(L/D)}_{\max }=15$, with a total mass of $M_{\mathrm{TOM}}=26\text{kg}$ at the design operating point, showing a 15% efficiency improvement over the baseline configuration, while satisfying the longitudinal static-stability constraint with $C_{\mathrm{My}\alpha }=-0.35{\mathrm{rad}}^{-1}$. A preliminary propulsive assessment of the hovering phase, indicates a power demand of approximately 6 kW. The optimized architecture preserves the baseline mission constraints while achieving an efficiency-oriented configuration, offering promising potential for future Vertical Take-Off and Landing applications.