Integrating Fourier Neural Operators into High-Fidelity Helicopter Flight Simulation for Real-Time Urban Wind Prediction

Abstract

High-fidelity helicopter flight simulators are essential for preparing pilots for complex and hazardous environments, yet realistic urban wind dynamics are difficult to reproduce in real time when relying on precomputed computational fluid dynamics (CFD) data. We present the first integration of a Fourier Neural Operator (FNO) into a Level D full flight simulator for real-time, physics-based urban wind field generation. Trained on high-resolution urban flow simulations, the FNO predicts one-minute-averaged 3D wind fields that dynamically adapt to flight state and location, replacing static wind inputs in the simulator pipeline. Turbulence levels are computed from the predictions and injected directly into the simulation loop. Professional pilots evaluated the system in an urban scenario and reported that it reproduced wind effects they would expect, such as turbulence and directional changes when landing behind buildings. They highlighted its value for less experienced pilots to develop wind awareness and for realistic training in critical operations, including offshore platform landings.