AeroCloud Validation Study for Sports Aerodynamics
AeroCloud vs a full-scale wind tunnel across ±15° yaw: 6.9% average CdA difference, no calibration
AeroCloud is validated for sports aerodynamics applications in a direct comparison against full-scale wind tunnel measurements. Wind tunnel tests performed on a cyclist mannequin on a time-trial bicycle have been replicated in the AeroCloud environment to evaluate the absolute accuracy of the automated CFD workflow. Across the yaw range tested, the average absolute difference in drag area between AeroCloud and the wind tunnel is 6.9%, falling to 1.3% at 0° yaw. No calibration or correction factors were applied to the simulations.
What is being measured
The comparison metric is drag area (CdA) the product of drag coefficient and frontal area, and the single figure that determines the aerodynamic power required to hold a given speed. Results are reported across yaw angles from -15° to +15°; yaw is the angle at which the free-stream wind meets the rider, where 0° is a pure headwind and positive and negative values represent crosswinds from either side.
Wind tunnel set-up
Wind tunnel measurements were performed on a full-scale mannequin model in a static time-trial position on a time-trial bicycle. The mannequin was fixed at the handlebars and pedals to ensure a stable position and was tested without any garments. The wind tunnel is an open-jet construction the flow passes through an open test section rather than a closed duct with a fan diameter of 2.5 metres. The bicycle is attached to a 6-component force balance (a sensor measuring forces and moments in all six directions) through 4 support struts. The complete assembly is placed on a turntable, allowing the model to be rotated inside the test section to each yaw angle. Horizontal forces were measured, and the drag area is calculated based on the free-stream wind speed for yaw angles ranging from -15° to +15°. The free-stream wind speed is constant at 50 km/h throughout the yaw sweep.
CFD set-up in AeroCloud
The setup from the wind tunnel experiment was replicated in AeroCloud with the original CAD models of the mannequin, bicycle and helmet. The bicycle support struts were not included in the simulation model. The default simulation domain dimensions were used, meaning that any blockage or expansion effects present in the wind tunnel environment the acceleration and redirection of flow caused by a finite test section are not considered. CFD simulations were performed on AeroCloud using the “PRO” quality setting. The ground boundary condition was set to slip, so there is no boundary layer build-up on the ground as there would be in the wind tunnel; the boundary layer is the thin region of retarded flow that forms against any surface. The boundary layer treatment on the model is set to “Resolved boundary layer” in order to better capture the flow separation on the mannequin, where the flow detaches from the body and generates drag. The simulation mesh, the division of the flow volume into discrete cells, was automatically generated by AeroCloud, containing 32.7 M cells. AeroCloud uses steady-state RANS simulations with the k-ω SST turbulence model. The simulations were run until convergence for all force components, meaning the computed forces no longer change between iterations; this was achieved after approximately 3000 iterations for all yaw angles. The flow field is subsequently averaged to smooth out oscillations in the solution resulting from the unsteady nature of the wake behind the mannequin.
Results
On average across all yaw angles, the absolute difference in CdA between AeroCloud and experiments is 6.9%. This is considered a good agreement for uncalibrated CFD simulations, considering also that the wind tunnel test section environment was not fully replicated in the CFD simulations. The drag forces from the CFD simulations are consistently lower than from the wind tunnel measurements. At 0° the drag is underestimated by just 1.3% by CFD, which is within the expected uncertainty of the wind tunnel experiments. The maximum deviation is -13%, found at the largest negative yaw angle. At large absolute yaw angles the influence of the support struts and surface roughness is also expected to be larger than at 0° yaw.