Machine-Learning Groundwork and Cluster Analysis for Vortex Detection

Author(s):

Prabhat Kumar, Melvin Felton, David Ligon, Sandra Collier, and Deryck James

Abstract:

During flight, an aircraft produces trailing wingtip wake vortices as a result of the relative motion of the craft and surrounding air. To detect these vortices, Doppler LIDAR is used (both on the ground and on the craft) to measure (relative) radial wind speeds using various scan geometries. These measurements can be processed to produce information about strength, circulation, and overall dynamics of the wake vortices. In this vortex detection project, we lay a machine-learning groundwork for the goal of characterizing and tracking coherent atmospheric structures produced by the wingtips of a C-17 aircraft. This study will aid research that aims to provide essential information used to dictate flight pattern protocol for safe and efficient aircraft maneuvers and group orientations during airdrops. Two outcomes of this project are (1) the LIDAR Cluster Analysis Program (LCAP), a preliminary software using k-means and DBSCAN clustering algorithms to partition a LIDAR scan into regions of interest, and (2) a methodology to isolate the vortex regions (with little noise) from the remainder of the LIDAR scan.