Speaker
Description
Pore-scale modeling of heterogeneous materials is challenging for a variety of reasons, many of them centered around maintaining accuracy over multiple scales. For direct computational methods (i.e., those that operate on underlying grids or meshes, rather than pore-network modeling for instance), the challenge includes efficient distribution of grid nodes – ensuring sufficient resolution in tight pore dimensions, but without wasting computational resources in large pores, which can consume nodes rapidly because of the cubic scaling of volume with respect to characteristic length. The obvious solution to this problem is grid refinement. However, its use has been limited for pore-scale modeling, largely because traditional techniques used for applications such as shock fronts are not as effective within confined pore structures.
In this work, we test modern mesh-generation algorithms that can be used to create tetrahedral meshes of heterogeneous and multiscale structures. These meshes offer improved geometric conformity as well as steep spatial gradients in refinement. Single-phase flow is simulated using a second-order (velocity) finite element method implemented in the open-source computing platform FEniCS.
The work addresses two central questions. First, how coarse is too coarse (characterized by a total breakdown in fidelity of the velocity field)? This question seeks to provide guidelines for using coarse meshes to efficiently model regions of multiscale systems where high accuracy may not be required. Second, what aspects of pore-structure are most critical to capture when refining from this coarse limit toward more accurate solutions? Both questions factor into strategies for multiscale problems.
In this presentation we contrast velocity fields across a dramatic range of mesh resolutions, quantifying how structural features lost at coarse resolution impact accuracy. We test targeted refinement strategies that can be used to balance computational demand against accuracy. The results reveal potentially surprising results: overall accuracy of direct methods (even if refined) is worse than what may be commonly assumed; capturing local pore structure is not always the most important aspect of reproducing velocity fields; local mesh/grid resolution, when used alone, is a poor predictor of overall quality of solution. Based on these findings, we assess the viability of using coarse meshing with direct simulation as a compromise between highly efficient but approximate methods (such as PNM) and highly accurate but computationally demanding methods that employ brute-force refinement.
| Country | USA |
|---|---|
| Acceptance of the Terms & Conditions | Click here to agree |
