Tracking particle clusters in turbulent channel flows via density-based clustering algorithms

The dynamics of clusters of inertial particles in a fully developed turbulent flow are analyzed using a new methodology for cluster identification and temporal tracking. It builds on density-based clustering algorithms DbScan and Optics to identify particle clusters from 3D snapshots of dispersed particles, which are then defined in terms of skeleton markers and boundary particles. Tracking and connectivity of cluster markers in snapshots from sequential time steps provide a computationally efficient tool to study the cluster dynamics. The computational cost of the new methodology is compared to that of conventional Voronoi tessellations. We apply these tools to study the evolution of clusters in a particle-laden turbulent square-duct flow, using data from DNS simulations with point-particle Lagrangian models. By tracking in time information of individual clusters such as their volume and location of constituent markers we analyze the different formation and disintegration processes, i.e. break up and merging of clusters, coagulation or dispersal of particles not assigned to clusters in previous or later times. Spatial information encoded in the cluster labeled markers facilitates the analysis of spatial and temporal correlations between clusters and flow dynamics.