GeoScale - Direct Reservoir Simulation on Geocellular Models

Open-source software
SINTEF is a research institute without teaching activities by itself. Instead, we supervise master and PhD students. This not only allows us to explore new research ideas but is also central in our efforts to make open-source reservoir simulation tools.

The Matlab Reservoir Simulation Toolbox is our open-source test-bench for new methodologies and contains a comprehensive set of routines and data structures for reading, representing, processing, and visualizing unstructured grids, with particular emphasis on the corner-point formate used within the petroleum industry. The toolbox also has various flow and transport solvers for single-phase, two-phase, and black-oil flow. We have several research challenges that are suitable for students:

  • Efficient numerical (non)linear algebra - MRST currently uses Matlab's built-in linear solver (i.e., A\b). We have also experimented a little with AGMG, an algebraic multigrid toolbox for Matlab. However, there are a large number of other possibilities for linear algebra. The project is simple to formulate: invstigate various other linear solvers and try to make MRST more efficient!
  • Multiscale finite-volume methods in MRST - A competing technology to our MsMFE method is the multiscale finite volume (MsFV) method. The MsFV method is formulated using a priamal and a dual coarse grid and is therefore more difficult to implement on non-Cartesian grids. Recently, however, the MsFV method has been formulated on an abstract operator form that opens the possibility for compact implementations on unstructured grids. In this project, you will use matrix formulations in Matlab and try to develop an implementation of the MsFV method in MRST that works on non-Cartesian and unstructured grids.
  • Unstructured grids - MRST offers quite extensive support for general unstructured grids and is an ideal test-bench for studying new ways of representing complex reservoirs. In this project, you will study various methods for generating Voronoi grids, constrained grids, anisotropic grids, advancing front methods, etc, and compare them with the standard method based on corner-point grids.
  • Efficient visualization - MRST relies heavily on the 'patch' function in Matlab for 3D visualization, which allows us to use quite fancy presentations of surfaces. Still, the visualization could definitely be improved, e.g., by adding possibilities for volume visualization, GPU acceleration, etc. Moreover, it would be highly desireable to develop some solution that allows a full port of MRST's current visualization capabilities to Octave.
  • GPU acceleration - use CUDA or OpenCL on programmable GPUs to accelerate MRST.

 We offer supervision at NTNU and at the universities of Bergen and Oslo. For all the projects listed above, you must work in close collaboration with the GeoScale team and we expect you to do quite a bit of programming. 

Contact: Knut-Andreas Lie 

Published November 18, 2009

A portfolio of strategic research projects funded by the Research Council of Norway