On Pore Space Partitioning in Relation to Network Model Building for Fluid Flow Computation in Porous Media

We present an improved method for generating a representative network model
(NM) of a porous medium from the structure of its pore space. We differentiate the problem of
positioning pore bodies (PB) and delimiting them via pore throats (PT). From a binary 3D
microtomographic reconstruction of a porous material, we use the common watershed
approach on the Euclidean distance map, examining several aspects in detail: the skeleton
and the necessary pre-processing to remove digitization artefacts, the conversion to a graph
that will support the NM, along with the post-processing needed to account for geometric
features, delimitation of the PB by a watershed method and comparison to direct PT
construction, and finally evaluation of the robustness of the NM generation by application on
synthesized images of varying resolutions.