Personal tools
You are here: Home Science & Projects Deisa Extreme Computing Initiative Projects 2010 - 2011 Distributed particle transport simulation in a Grid-like HPC CFD environment

Distributed particle transport simulation in a Grid-like HPC CFD environment

Project DiParTS
Research Area Engineering
Principal Investigator(s) Prof.Dr. Hans-Joachim Bungartz
Dr. Tobias Weinzierl
  • Technische Universität München, Institut für Informatik, Garching bei München, Germany
  • Technical University of Cluj-Napoca, Cluj-Napoca, Romania
  • King Abdullah University of Science and Technology, Thuwal, Kingdom of Saudi Arabia


The DiParTS project (Distributed Particle Transport Simulation in a Grid-like HPC CFD Environment) numerically studies particles dispersed in non-stationary fluids within tube-like geometries on the micro-scale, where the fluid and, as a consequence, the particles are stimulated by an oscillating pressure. The particles’ long-time behaviour due to the pressure oscillations, i.e. their averaged movement on the long-term time-scale, allows us to draw conclusions, for example, on the causes of particle sedimentary deposition and centrifugal particle separation in several applications, as the particles exhibit a drift along the stimulation amplitude. Here, classical fluid-structure interaction phenomena interplay with Brownian motion and particle-wall interaction. In a preceding DEISA project, we already studied simplified experimental setups on the short-time time-scale. Despite some promising and interesting insights from a fluid-dynamics point of view, the full simulation of the situation described above however proved to be far from solvable with today’s computing power. Due to this proposal, we nevertheless will broaden the horizon of computability, as we switch from a fully coupled system to an approach where the fluid simulation without particles on an extremely fine spatial and temporal resolution is cut into small time intervals, these chunks of computational challenges are deployed to supercomputers, and the fluid fields are coarsened spatially before the supercomputer streams the data back to the scientist’s local workstation where it is post-processed, i.e. the Brownian motion and the particles’ effect are remotely added to the flow field after the fluid dynamics time step has terminated. The extreme computing power spent on this waterfall process – in particular on the fine-scale fluid dynamics simulation – will yield new insights on the long-time behaviour of the overall simulation setup, while the approach is validated simultaneously by a comparison of a fully-coupled fluid-interaction setting with the decoupled simulation for several small time steps.

Document Actions