Distributed Percolation Analysis for Turbulent Flows

Research focus area – In Situ visualization and computational steering

Application – Large-scale turbulent flow simulations

Turbulent flow analysis plays a crucial role in many domains e.g. design of fuel-efficient cars and is an active research area primarily addressed through direct numerical simulations (DNS) of the Navier-Stokes equations. Percolation analysis studies the connectivity of sublevel-sets in derived scalar fields and has been proven useful to understand intense Reynolds stress or vorticity events. A major limitation of current approaches, however, is their slow runtime. To overcome this limitation, we at first devised an algorithm exploiting the union-find data structure, which is much faster. But since it is a global algorithm it is still challenging to parallelize. However, the size of turbulent-flow data requires analysis algorithms to run in parallel on distributed memory. This problem is addressed in our second algorithm working. It also builds upon the union-find data structure, however, keeping inter-process communication to a minimum by exploiting that many operations can be executed out-of-order and synchronization is only necessary a small number of times.

Image: Result of a Percolation analysis for a duct flow data set, simulated by SerC Researchers at KTH Mechanics. On the right, connected components for several thresholds are shown, with the respective largest component highlighted in red. The percolation analysis on the left allows identifying the point at which one large component spanning the entire simulation domain appears. This point ($H_c$) is relevant in further analysis of the turbulent flow.


A. Friederici et al. Distributed Percolation Analysis for Turbulent Flows. IEEE Symposium on Large Data Analysis and Visualization (LDAV), 2019 (Best paper Honorable mention)
W. Köpp et al. Notes on Percolation Analysis of Sampled Scalar Fields. In Topology- Based Methods in Visualization (TopoInVis), 2019