Characterizing wake vortex pairs using nonsupervised machine learning

Abstract

We present sequential Doppler lidar observations of wake vortex pairs generated by a Boeing C-17 Globemaster III aircraft. We calculated weighted gradients for the Doppler data products, then applied two cluster analysis algorithms in serial to detect, track, and characterize the vortex pairs. Density-based spatial clustering of applications with noise (DBSCAN) was used to isolate vortex pairs from the background. Then k-means clustering was applied to partition the pair into three clusters that correspond well to the prototypical regions of a vortex pair. This study demonstrates the power of cluster analysis on sequential Doppler lidar data for the detection and characterization of aircraft wake vortices, offering what we believe to be a novel and promising approach for advancing automated detection capabilities.