06.2 - Flight Dynamics and Control (UAV related)
VANE-LESS 3D WIND ESTIMATION FOR FIXED-WING UAVS: A PHYSICS-AUGMENTED TEMPORAL RESIDUAL NETWORK APPROACH
Z. Zhang¹, S. Li¹, X.P. Xu¹, Z. Zhou¹; ¹Northwestern Polytechnical University, China
This paper proposes DeepWindNet, a physics-augmented framework for vane-less wind estimation in fixed-wing UAVs. By integrating kinematic equations as structural anchors to learn nonlinear residuals, the method utilizes a Squeeze-and-Excitation (SE) attention mechanism to recalibrate sensor weights based on physical flight contexts. Experimental results demonstrate that while achieving a horizontal RMSE of less than 0.3 m/s, this approach effectively eliminates the high-frequency oscillations and systematic biases inherent in traditional dynamics inversion.
DOI: 10.71945/icas2026_0859