Permeability and porosity upscaling method using Machine Learning and Digital Rock Physics

In this paper, we introduce a novel upscaling method in Digital Rock Physics for porosity and permeability properties. The upscaling method is based on a machine learning method characterizing quantitatively image textures from 3D X-ray Micro-Computed Tomography (MCT) images. The procedure starts by imaging a core plug sample at a coarse scale, corresponding to a resolution of around 20 ìm, to visualize texture heterogeneity. The machine learning method identifies representative texture spatial locations by classification. We extract physically subsets representing each texture and image the subset at a resolution of 1 ìm. Then, we run pore scale simulations to obtain porosity and permeability properties for each representative texture. Finally, we upscale rock properties using classification result to populate coarse scale model from fine scale results. We illustrate our workflow results for a carbonate rock sample from an oilfield reservoir in Abu Dhabi.