Data structure improvements for 3D polyhedral grids with application to unstructured discrete fracture models.

Thomas Viard and Claude Cavelius and Bradley Mallison and Charles H. Sword. ( 2012 )
in: Proc. 32nd Gocad Meeting, Nancy

Abstract

Multi-million cell grids are a common practice in the realm of structured grids and allow finescale 3D rock properties to be effectively captured in the model. However, structured grids usually involve approximations on the subsurface geometry such as stair-stepped faults, or wasteful fitting strategies such as dead cells in the vicinity of fault T-junctions. Unstructured grids allow a better fit to complex geological structures. Unfortunately, their larger footprint in memory has a negative impact on performance. Consequently, unstructured grids are frequently built at a coarser scale than structured grids, which can result in a loss of resolution on rock properties. Designing a compact, memory-efficient data structure is the key to removing this limitation. In this paper, we present a fully 3D general-purpose unstructured framework, whose main benefits are: the amount of memory required per polyhedron is much lower, allowing larger unstructured grids to be used in practice, it is not restricted to a specific type of polyhedron (e.g., only tetrahedra), nor to the typical limitations of 2.5D unstructured grids and locally unstructured grids, and all adjacency queries can be performed in constant or bounded time, making algorithms scalable with the total number of polyhedra. We demonstrate the efficiency of our framework for building unstructured grids for discrete fracture-matrix models. The grid generation algorithm is flexible enough to closely fit a large amount of geological features (e.g., wells, faults, fractures, horizons) and refine the mesh in the vicinity of these features, so that it largely takes advantage of our unstructured grid capabilities. Our data structures are an improvement upon the TopoKernel plug-in, which is directly available from the Gocad Research Group.

Download / Links

BibTeX Reference

@INPROCEEDINGS{ViardGM2012,
    author = { Viard, Thomas and Cavelius, Claude and Mallison, Bradley and Sword, Charles H. },
     title = { Data structure improvements for 3D polyhedral grids with application to unstructured discrete fracture models. },
 booktitle = { Proc. 32nd Gocad Meeting, Nancy },
      year = { 2012 },
  abstract = { Multi-million cell grids are a common practice in the realm of structured grids and allow finescale 3D rock properties to be effectively captured in the model. However, structured grids usually involve approximations on the subsurface geometry such as stair-stepped faults, or wasteful fitting strategies such as dead cells in the vicinity of fault T-junctions. Unstructured grids allow a better fit to complex geological structures. Unfortunately, their larger footprint in memory has a negative impact on performance. Consequently, unstructured grids are frequently built at a coarser scale than structured grids, which can result in a loss of resolution on rock properties.
Designing a compact, memory-efficient data structure is the key to removing this limitation. In this paper, we present a fully 3D general-purpose unstructured framework, whose main benefits are:
the amount of memory required per polyhedron is much lower, allowing larger unstructured grids to be used in practice,
it is not restricted to a specific type of polyhedron (e.g., only tetrahedra), nor to the typical limitations of 2.5D unstructured grids and locally unstructured grids, and
all adjacency queries can be performed in constant or bounded time, making algorithms scalable with the total number of polyhedra.
We demonstrate the efficiency of our framework for building unstructured grids for discrete fracture-matrix models. The grid generation algorithm is flexible enough to closely fit a large amount of geological features (e.g., wells, faults, fractures, horizons) and refine the mesh in the vicinity of these features, so that it largely takes advantage of our unstructured grid capabilities.
Our data structures are an improvement upon the TopoKernel plug-in, which is directly available from the Gocad Research Group. }
}