Building a pore network model from a pore space 3D image to precisely predict permeability tensor

Pore network models (PNM) are widely used for more than 50 years to study pore scale
transports. It is commonly admitted that, in order to give realistic property predictions, the PNM
must summarize all the relevant topological and geometrical aspects of the considered porous
medium. Now that 3D imaging techniques give us very precise representations of this geometry,
it is difficult to state that most of the PNM, which are based on regular cubic lattices, are
representative of the pore space. But, even if 3D images are easy to obtain, the construction of
a representative PNM only using information directly extracted from the 3D images is still
difficult.
In this work we present a methodology for such a construction. Starting from a 3D binary image
of a porous sample, all the crucial steps of the method are analysed in depth. Permeability
values obtained by our PMN are compared to values obtained by direct numerical modeling
where Stokes equations are solved using the voxelized 3D image as computation grid.
An extension of the model allowing computing the full permeability tensor is also presented.