Parallel mesh loading and partitioning for large-scale simulation


Prof. Dr. Florian Frank
Lehrstuhl für Modelierung und Numerik
Department of Mathematic
Friedrich-Alexander-Universität Erlangen-Nürnberg

Project Overview

In the context of our large-scale CFD simulator, the purpose of this project is to accomplish scalability of the mesh loading and distribution procedure. Digital Rock describes the digitalization and prediction of petrophysical experiments in the oil and gas industries. It aims to replace or supplement expensive and time-consuming laboratory experiments , in particular, in the context of chemical enhanced oil recovery. The starting points are high-resolution 3D images of rock samples obtained by microcomputed-tomography(μ-CT) scanning. Subsequently, the pore space of the scanned rock is either abstracted by a pore-network model or used as a mesh for direct numerical simulation (DNS). In the simple case of a single-phase fluid, e. g., the absolute permeability (resistance due to connectivity and anisotropy of the porous medium) is an effective property of interest and can be computed by upscaling the pore-scale simulation results. With our code Simulicious, we follow the DNS approach. From a computational perspective, an ultimate goal is to handle image sizes of 1000³ to 2000³ voxels with voxel sizes of 1–10μm, thus extending the capability to cm-scale volumes. This requires both the design of an efficient solution strategy for large sparse linear systems and the parallel scalability of all required time-consuming operations such as linear system assembly, solving/preconditioning, and mesh generation.