|Name:||Large-Scale Simulation of Respiratory Airflows|
|Time:||Thursday, June 26, 2014
11:40 am - 12:00 pm
CCL - Congress Center Leipzig
|Speaker:||Nicola Varini, iVEC|
|Abstract:||In this talk we will explain how we have used state-of-the-art supercomputing to create what is believed to be, at the time of writing, the most realistic simulation of the human respiratory system.
Almost all existing models for respiratory airflow and aerosol deposition utilise rigid analogues of airways, which do not account for the effects of lung motion on airflow and particle deposition. Mead-Hunter et al.  compared traditional computational fluid dynamics approaches with a novel moving mesh method. A distinct difference was found to exist between the results obtained using a stationary geometry (with either constant or oscillating flow) and a moving-mesh geometry. It was found that a hybrid moving-mesh and oscillating-flow method was required to produce results in agreement with in vivo experimental data for local and global deposition in rat airways. A prescribed lung-motion model was developed by the authors, for OpenFOAM, to account for expansion of the lungs and to drive the flow. To achieve this, the boundary of the mesh was expanded relative to a fixed reference point, with the ability to adjust the expansion along each of the three axes independently. The mesh was generated from an X-ray, computed tomography scan dataset, using approximately thirteen million unstructured tetrahedral mesh cells.
It was found that, for previous studies, the application of laminar or various turbulent flow models was made to the simulation for relatively arbitrary reasons. In order to provide a definitive guide for the most appropriate airflow model for such simulations, the authors’ work examines the largest and most realistic lung simulation ever conducted (to the knowledge of the authors), using the spectrum of respiration rates and both laminar and Large Eddy Simulation (LES) turbulent solvers. In order to simulate realistic respiratory cycles, approximately 20 seconds of real-time was simulated, which required around 50,000 CPU hours on Magnus. Magnus is a 70-Teraflops Cray XC30 system, based around the Intel Xeon ‘Sandy Bridge’ processor. The Cray Aries interconnection network is particularly beneficial for this kind of computation since the computation is communication intensive.