Coherent flow structures such as aircraft wake vortices may pose a hazard to follower aircraft potentially subjecting them to severe rolling moments. Particularly in their final approach towards the runway such disturbances may prove consequential. The characterization of wake vortices generated by leading aircraft therefore constitutes a decisive element for ensuring safe and efficient airport runway operation. Light detection and ranging (lidar) instruments are most frequently employed for undertaking measurements of wake vortices. The one-dimensionality of the measured velocity field necessitates processing algorithms with large-scale assumptions to determine the position and strength of wake vortices within. The accuracy of available algorithms, such as the radial velocity (RV) method, is so far not known, in particular for turbulent atmospheric cases. The lack of a ground truth in field measurements motivates this thesis to simulate lidars within high-fidelity Reynolds-Averaged Navier-Stokes and Large Eddy Simulation (LES) aircraft landing runs. Wake vortices are simulated from their generation behind a detailed aircraft geometry to their decay, with known position and strength. Lidar simulations within the LES, LES Lidar Simulators (LLS) including lidar parameters and characteristics, allow wake vortex processing algorithms to be assessed. Realistic modeling of both the complex wake vortex flow and the lidar instrument has not been done previously. This thesis determines the RV method to overestimate the strength of wake vortices by 4 %, while dislocating them by 6 %. The RV method is particularly inaccurate in ground proximity, as it lacks the modeling of image vortices. Efficiently processing extensive data amounts or characterizing wake vortices in fast-time, for dynamic aircraft separations at airports, is not accommodated with available lidar processing algorithms. Herein, a novel Machine Learning (ML) pipeline is developed for individual wake vortex characterizations, enabling fast-time, reliable, and accurate characterizations (at least as accurate as the RV method) with no bias or undesirable wall effects. For the first time, LES data are employed for training ML models and evaluating field measurements. This enables the processing of field measurement datasets that lack labeled field measurements for training. The necessity for lidar simulations to include atmospheric noise and measurement noise, abbreviated as LLSₙ, is determined. LLSₙ-trained ML pipelines detect over 90 % of hazardous wake vortices in representative field measurements. This is on a par with the reliability of field-measurement-trained ML pipelines, suggesting that trustworthy wake vortex processing is achievable using simulated lidars when the expected measurement setup, aircraft categories, and atmospheric conditions are considered.
Niklas Louis Wartha (Thu,) studied this question.