Speaker
Description
Predictive modelling of relative permeabilities in representative carbonate samples remains a challenging problem in the Digital Rock Physics (DRP) community. Traditional DRP workflow, comprising of image acquisition and pore network model (PNM) extraction and simulation [1] is verified against homogeneous rock samples, fails to capture sub-micron porosity, prominent in carbonates. Recent research efforts focus on multiscale PNMs [2,3]. This approach is necessary to capture the complexity of bimodal samples but is affected by uncertainties in differential image processing and physics definition in Darcy regions of the model.
The following research presents an application of the existing DRP workflow to monomodal carbonates, investigating the practical limits of current image super-resolution and network extraction tools and aiming to extend the domain of application of an already verified single-scale physics solver [4]. With sample pore sizes ranging from 100 µm to 100nm, generation of representative volume models posed unique computational challenges, tackled in this work.
This large range of pore sizes opened the question of the optimal resolution to perform the proposed study. Using a sub-volume of the full sample, the impact of the resolution from 100nm to 1µm on porosity, permeability and capillary pressure was evaluated. While porosity decreased with coarser resolution due to closure of smaller pores, the impact on both permeability and capillary pressure remained minimal up to 650 nm resolution, suggesting that closed pores have limited influence on flow behaviour. A target resolution of 500 nm was selected.
To achieve the required level of detail, a CycleGAN-inspired unpaired image transfer and super-resolution neural network was employed [5]. This approach enhanced a 5 μm μCT image with high-resolution, low-noise 2D SEM images to reach the target 500 nm resolution. The resulting 13,000×13,000×12,000 voxel image reproduced the resolved porosity of the μCT acquisition while providing details consistent with the SEM image in its previously unresolved regions.
Processing such large images exceeds the capabilities of most current tools. Even PNM "simplification" approaches typically require 20-60 times the image size in memory for network extraction. A tiled processing strategy was developed: splitting the full image into overlapping tiles, extracting networks from each tile independently, and merging these into continuous PNM representing the complete pore space. Each tile was extracted using an in-house tool inspired by both GNExtract [6] and PoreSpy [7]. Critical to this approach was determining the overlap size necessary to guarantee network continuity across tile boundaries, a proper study on the impact of this overlap size was performed to set the tile size before extraction.
Applied to a carbonate sample, this workflow produced an explicit 85-million-pores network compatible with a single scale Stokes PNM solver and suitable for comparison with experimental relative permeability measurements. This study demonstrates the feasibility of extending DRP workflows to sub-micron resolutions at representative volumes, while identifying computational bottlenecks that must be addressed for routine applications.
| References | [1] Regaieg, M., Varloteaux, C., Farhana Faisal, T., & ElAbid, Z. (2023). Towards Large-Scale DRP Simulations: Generation of Large Super-Resolution images and Extraction of Large Pore Network Models. Transport in Porous Media, 147(2), 375-399. [2] Foroughi, S., Bijeljic, B., Gao, Y., Blunt, M. J. (2024). Incorporation of Sub-Resolution Porosity Into Two-Phase Flow Models With a Multiscale Pore Network for Complex Microporous Rocks. WaterResources Research, 60, e2023WR036393 [3] Wang, S., Mascini, A., Ruspini, L.C., Øren, P.-E., & Bultreys, T. (2023). Imaging and modeling the impactof multi-scale pore connectivityon two-phase flow in mixed-wetrock. Water Resources Research,59, e2022WR034308 [4] Regaieg, M., Nono, F., Faisal, T. F., & Rivenq, R. (2023). Large-Pore network simulations coupled with innovative wettability anchoring experiment to predict relative permeability of a mixed-wet rock. Transport in Porous Media, 147(2), 495-517. [6] Raeini, A. Q., Bijeljic, B., & Blunt, M. J. (2017). Generalized network modeling: Network extraction as a coarse-scale discretization of the void space of porous media. Physical Review E, 96(1), 013312. [7] Khan, Z. A., Elkamel, A., & Gostick, J. T. (2020). Efficient extraction of pore networks from massive tomograms via geometric domain decomposition. Advances in Water Resources, 145, 103734. |
|---|---|
| Country | France |
| Acceptance of the Terms & Conditions | Click here to agree |
