Phase-parameterised gaussian process for predicting UAV aerodynamic loads in operational turbine wakes

Operational inspection of wind turbines using uncrewed aerial vehicles (UAVs) is an attractive method for reducing offshore operation and maintenance costs, but robust flight in the turbine near-wake remains challenging due to the presence of strong unsteady disturbances that can degrade stability and quality of the inspection. This paper presents a control-oriented modelling framework for predicting UAV aerodynamic loads in operational turbine wakes with faster-than-real-time capability. A free-vortex-wake representation is used to capture the dominant near-wake physics, and a phase-parameterised Gaussian Process Regression (GPR) surrogate model is employed in a wake-aligned coordinate system to exploit the periodic structure of the induced velocity field. The surrogate model is coupled to a nonlinear, physics-informed rotorcraft flight dynamics model to predict wake-induced force perturbations on a quadrotor UAV. The approach is validated against data obtained from an experimental campaign with a scaled three-bladed turbine and a 250 mm-class quadrotor mounted on a load cell. The relevant data and experimental settings are reported as a reference for future research. Across lateral and vertical traverses in the near wake, the predicted streamwise and vertical force components agree with measurements in both magnitude and trend. A frequency-domain error attribution analysis shows the turbine–UAV interaction component, and the results indicate the measured spectral peaks at the turbine rotational and blade-passing harmonics are consistent with the simulations. The present formulation is primarily intended for wake conditions with coherent low-order periodic content, and its prediction fidelity may degrade under strongly turbulent or weakly phase-coherent inflow. Within this scope, the results demonstrate that phase-parameterised GPR surrogates can provide accurate, data-efficient, uncertainty-aware wake disturbance predictions suitable for real-time simulation platforms supporting UAV autonomy and control development for operational turbine inspection.