EVALUATION OF PARTICLE CLUSTERING ALGORITHMS IN THE PREDICTION OF BROWNOUT DUST CLOUDS

Abstract

A study of three Lagrangian particle clustering methods has been conducted with

application to the problem of predicting brownout dust clouds that develop when rotor-

craft land over surfaces covered with loose sediment. A significant impediment in per-

forming such particle modeling simulations is the extremely large number of particles

needed to obtain dust clouds of acceptable fidelity. Computing the motion of each and

every individual sediment particle in a dust cloud (which can reach into tens of billions per

cubic meter) is computationally prohibitive. The reported work involved the development

of computationally efficient clustering algorithms that can be applied to the simulation of

dilute gas-particle suspensions at low Reynolds numbers of the relative particle motion.

The Gaussian distribution, k-means and Osiptsov's clustering methods were studied in

detail to highlight the nuances of each method for a prototypical flow field that mimics

the highly unsteady, two-phase vortical particle flow obtained when rotorcraft encounter

brownout conditions. It is shown that although clustering algorithms can be problem

dependent and have bounds of applicability, they offer the potential to significantly re-

duce computational costs while retaining the overall accuracy of a brownout dust cloud

solution.