Kernel-based distributed-input distributed-output control for fan array wind generators

We propose a fast two-dimensional wind profile control methodology for the 3.25 × 3.25 m2 test section of the world's highest resolution fan array wind generator (FAWG) with 40 × 40 fans (Guinness World Record 2025). The advancement of unmanned aerial vehicles (UAVs) and vertical takeoff and landing (VTOL) systems has motivated the need for robust UAV control under turbulent atmospheric conditions. Unlike single-impeller traditional wind tunnels, fan array wind generators (FAWGs) have emerged as innovative facilities that generate dynamic, multi-scale gusts replicating real-world turbulence. By leveraging arrays of individually controllable fans, they achieve superior spatial and temporal resolution and fast dynamic response. Nevertheless, due to FAWG's high dimensionality and corresponding flow complexity, controlling individual fans to generate a desired flow field is challenging. This work synergizes feedback control and machine learning to advance FAWG capabilities. In particular, we establish a real-time control strategy that harnesses a Gaussian kernel model to map fan speed settings to velocity profiles, combined with its inverse model, both integrated into the internal model control (IMC) structure, improving UAV testing methods for next-generation aerial systems working in complex wind conditions. The Gaussian kernel's smoothness and differentiability facilitate gradient-based optimization, enabling computationally efficient inverse modeling for real-time velocity-to-fan-speed mapping. We experimentally tested several target velocity profiles that vary in complexity to assess the effectiveness of the proposed controller, including a uniform profile and one-dimensional and two-dimensional spatially varying profiles. The controller successfully tracked the target velocity profile. These results pave the way for accurately replicating turbulent environments critical for authorizing next-generation UAVs.