Heterogeneities at the pore scale influence flow and reactive transport processes in porous media such as natural (clay) and engineered (cement-based) barriers, both highly relevant to radioactive waste disposal. Flow properties of porous media are controlled by the geometry of their pore space (two-phase system, i.e. solid or pore), while reactive transport properties additionally depend on the spatial distribution of different minerals within the solid matrix (multiphase system) and their specific characteristics. Flow and reactive transport modelling in porous media therefore requires detailed 3D information about the pore’s structure. Yet 3D imaging techniques are cumbersome and expensive while for some porous media such as clay, the required resolution can only be achieved by 2D imaging. That is why 2D-to-3D reconstruction of porous media, that is, the process of stochastically generating 3D image realizations that are statistically consistent with a given set of 2D images or slices, has been an intensive field of research for more than a decade. Relatively recently, deep learning (DL) has been shown to be a promising tool for 2D-to-2D and 3D-to-3D reconstruction of porous media from training images while the booming development of DL for texture synthesis in general makes it a good candidate for handling challenging 2D-to-3D reconstructions. Yet so far DL has only been used for 2D-to-3D reconstruction of binary subsurface rocks and geologic layers and its potential to deal with more complex heterogeneous porous media remains unexplored. This PhD focusses on developing a DL-based 2D-to-3D reconstruction approach suitable for clay and other complex porous materials relevant to radioactive waste disposal.
The minimum diploma level of the candidate needs to be
- Master of sciences in engineering
- Master of sciences
The candidate needs to have a background in