Abstract Replacing physical laboratory conventional and special core analysis experiments (CCA/SCAL) with their digital alternatives has, so far, failed to become the main stream. This is especially the case for carbonate reservoir rocks that are known for their complexity in pore types, pore size distribution and architecture. The lack of inertia towards these potentially revolutionizing concepts is due to the difficulty in reaching a micron-scale resolution. Therefore, imaging micro-porosity seems a valid justification in a delayed adoption of digital SCAL for carbonate reservoir rocks. As the industry tackles ever more challenging and lower permeability reservoirs, a wider adoption of digital SCAL becomes a more desirable objective, with significant time and cost savings compared to conventional physical experiments. This paper leverages an extensive review of the technical literature as recent research contributed to bridging the gaps between pore space modeling and physical properties such as grain size distribution, pore size distribution, connate water saturation and plug permeability. The workflow is subdivided into two steps: (1) generating high-resolution rock images at the plug scale, and (2) estimating permeability values from these images. Resolving microporosity at the core plug scale was attempted using (1) multi-fractal behavior of the grain and pore size distribution, and (2) machine learning and pixel-based multi-point geo-statistics. For permeability estimation, multiple methods were tested including (1) graph neural network and (2) ball and stick modelling and simulation. Physical plug permeability measurements, mercury injection capillary pressure results and nuclear magnetic resonance (NMR) data (calibrated to well log NMR to account for in-situ conditions) were used for validation. This paper shows that it is possible to embed micropores into a thresholded 3D micro-CT low-resolution volume/image at core plug scale using multi-point statistics and a high resolution 2D image from confocal microscopy as a training image. Multiple realizations were generated to quantify the residual uncertainty. However, it is not computationally feasible to run traditional pore-scale flow simulation on these high-resolution (∼200 nm resolution) plug scale (inch-scale diameter) images. As an alternative for flow simulation, a graph neural network was trained to estimate permeability. The network was trained on twenty-two rock samples and produced a coefficient of determination of 0.78 and 0.87 for carbonates and sandstones, respectively. The assessment was performed on a test dataset of rock subsamples obtained from the same formations used for training. This demonstrates the viability of using such a methodology to estimate permeability for an ensemble of realizations in an efficient manner. Overall, with advances in imaging technology, computational resources, and algorithms (including machine learning), it is feasible to perform digital SCAL at a much larger scale.
Ibrahim et al. (Tue,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: