Wake Vortex Characterisation of Landing Aircraft using Artificial Neural Networks and LiDAR Measurements

Kurzfassung

An often proposed method for increasing airport capacities to tackle the ever-growing flight demand, is the re-categorisation of currently used aircraft separations. There are good reasons for the existence of these separations: wake vortices generated by an aircraft can be a hazard to those following. The danger is greatest for landing aircraft, as most follow the same glide path before hitting the runway tarmac. The goal is to have dynamic aircraft separations, allowing individual judgement for each aircraft pair, airport and atmospheric condition. The fast-time strength and location characterisation of wake vortices with LiDAR scans is suggested for monitoring predictive WVAS. Current, partly manual, algorithms cannot accommodate this for all LiDAR types. This work proposes the use of ANNs to automate the characterisation of wake vortices. Artificially generated LiDAR proxy data and MLPs are used to develop suitable network architectures and evaluate different feature engineering. Findings are thereafter applied to measured LiDAR scans, made up of radial velocity measurements from Vienna International Airport, and state-of-the-art perceptual ANNs - CNNs. Feasible feature engineering includes the removal of crosswind using LiDAR scans before overflights, the use of ideal and global measurement grids, and the disregarding of faulty scans as well as velocity measurements at low altitudes. CNNs prove suitable for monitoring WVAS and could be of value for processing large data sets of future wake vortex LiDAR campaigns and routine measurements. ANN characterisation predictions, with circulation errors as low as 26 m^2/s and localisation errors as low as 13 m, can be obtained in the fraction of a second, enabling fast-time wake vortex characterisation. The reliability of these results is up to 91%, suggesting the enormous capabilities of the proposed approach and coming one step closer to dynamic aircraft separations.