Physics-aware wind surrogate model for UAV aerodynamic response assessment

Abstract

Purpose – This study aims to use dimensionality reduction techniques applied to a detailed wind flow computational fluid dynamics (CFD)-generated database to develop a fast numerical tool that predicts, using the available weather forecast data, the airflow around any urban environment. The tool is aimed for its use in path planning design and optimization of unmanned aerial vehicles (UAVs) in urban mobility. Design/methodology/approach – A complex urban site is selected as an example of vertiport. Geospatial data and land models are used to automate the CFD computational domain, mesh generation and terrain classification. To enhance efficiency, some mesh cells, corresponding to dense vegetation and remote buildings, are solved as porous media. After validation, a CFD database is created using a Reynolds-averaged Navier−Stokes model by sweeping different wind flow boundary conditions. The database is processed with high order singular value decomposition techniques, and interpolation methods enable real-time wind flow predictions, producing detailed maps with resolution under 1 m in approximately 1 s. Findings – The surrogate model accelerates predictions by a factor of 7200 compared to direct CFD simulations while maintaining acceptable accuracy: mean relative deviations in velocity predictions near the buildings of interest are of the order of 2%. Examples of UAV trajectories and their dynamic responses are obtained using the developed tool. Originality/value – The computational domain is automated using geospatial data, facilitating mesh classification and improving simulation efficiency. The surrogate model, which uses wind forecasts from the meteorological as inputs, provides real-time wind-flow predictions and improves UAV flight path design by identifying high-risk areas before take-off.