Physics-guided self-supervised upscaling of three-dimensional digital rock models from multi-resolution micro-CT data
Abstract
Digital Rock Physics relies on multiscale upscaling workflows that bridge pore-scale imaging and Darcy-scale flow simulation in heterogeneous, low-permeability reservoirs. Existing approaches use convolutional neural networks to transform low-resolution micro-computed tomography images into multiclass Darcy-scale models, but depend on supervised training with carefully curated labeled minicubes, which is costly and difficult to extend to new lithologies or scarce-label regimes. This work introduces a physics-guided self-supervised pretraining framework for three-dimensional digital rock models that combines volumetric contrastive learning with a permeability-aware regularization term. The encoder first learns volumetric representations by predicting contextual positions of three-dimensional image crops and then enforces consistency between embedding similarities and proxy permeabilities derived from percolation-based analysis of segmented pore space. After fine-tuning for rock typing, the encoder is integrated into an upscaling pipeline that maps low-resolution scans to Darcy-scale multiclass models used for single-phase flow simulations. The self-supervised model was compared with a purely supervised baseline in terms of rock-typing performance, visual fidelity of the upscaled models, and preservation of key petrophysical properties relative to laboratory and high-resolution numerical benchmarks. The results indicate that physics-guided self-supervised pretraining improves rock-typing accuracy, yields Darcy-scale models with more consistent connectivity of high-permeability channels and barriers, and reduces discrepancies in effective permeability, especially in low-label regimes. These findings suggest that self-supervised, physics-informed representation learning can enhance both classification robustness and the reliability of digital rock upscaling workflows for heterogeneous carbonate rocks.
Document Type: Original article
Cited as: Gainitdinov, B., Prochii, R., Orlov, D., Sharaev, M., Ren, Y., Koroteev, D. Physics‑guided self‑supervised upscaling of three‑dimensional digital rock models from multi‑resolution micro‑CT data. Advances in Geo‑Energy Research, 2026, 21(1): 13‑28. https://doi.org/10.46690/ager.2026.07.04
DOI:
https://doi.org/10.46690/ager.2026.07.04Keywords:
Self-supervised learning, upscaling, digital rock physics, reservoir characterization, darcy-scale rock modelsReferences
Azizi, S., Mustafa, B., Ryan, F., et al. Big self‑supervised models advance medical image classification. Paper Presented at IEEE/CVF International Conference on Computer Vision, Montreal, QC, Canada, 10‑17 October, 2021.
Babu, P. P. S., Brindha, T. Deep learning fusion for intracranial hemorrhage classification in brain CT imaging. International Journal of Advanced Computer Science & Applications, 2024, 15(8): 884‑894.
Banterle, F., Corsini, M., Cignoni, P., et al. A low‑memory, straightforward and fast bilateral filter through subsampling in spatial domain. Computer Graphics Forum, 2012, 31(1): 19‑32.
Behbahani, H., Azin, R., Osfouri, S. A comprehensive review of digital rock physics: From tomographic images to pore network modeling. Chemical Process Design, 2023, 2(1): 6‑30.
Blunt, M. J., Bijeljic, B., Dong, H., et al. Pore‑scale imaging and modelling. Advances in Water Resources, 2013, 51: 197‑216.
Blunt, M. J., Sun, S., Boone, M. A., et al. Digital rock physics and fluid flow in the context of the energy transition. Advances in Geo‑Energy Research, 2025, 18(3): 299‑302.
Bultreys, T., De Boever, W., Van Hoorebeke, L., et al. A multiscale, image‑based pore network modeling approach to simulate two‑phase flow in heterogeneous rocks. Paper SCA2015‑027 Presented at International Symposium of the Society of Core Analysts, St. John's Newfoundland and Labrador, Canada, 16‑21 August, 2015.
Chen, T., Kornblith, S., Norouzi, M., et al. A simple framework for contrastive learning of visual representations. Paper Presented at International Conference on Machine Learning, Vienna, Austria, 12‑18 July, 2020.
Ebadi, M., Orlov, D., Alekseev, V., et al. Lift the veil of secrecy in sub‑resolved pores by xe‑enhanced computed tomography. Fuel, 2022, 328: 125274.
Faisal, T. F., Islam, A., Jouini, M. S., et al. Numerical prediction of carbonate elastic properties based on multi‑scale imaging. Geomechanics for Energy and the Environment, 2019, 20: 100125.
Fourer, D., Sidibe, D., Lecomte, J.-F., et al. A comparative evaluation of self‑supervised methods applied to rock images classification. Paper Presented at 19th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications, Rome, Italy, 27‑29 February, 2024.
Grady, L. Random walks for image segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2006, 28(11): 1768‑1783.
Hasan, N., Bao, Y., Shawon, A., et al. Densenet convolutional neural networks application for predicting covid-19 using ct image. SN Computer Science, 2021, 2(5): 389.
He, K., Fan, H., Wu, Y., et al. Momentum contrast for unsupervised visual representation learning. Paper Presented at IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, WA, USA, 13-19 June, 2020.
Jackson, S. J., Niu, Y., Manoorkar, S., et al. Deep learning of multiresolution X‑ray micro‑computed‑tomography images for multiscale modeling. Physical Review Applied, 2022, 17(5): 054046.
Jing, L., Tian, Y. Self‑supervised visual feature learning with deep neural networks: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 43(11): 4037-4058.
Law, B. E., Spencer, C. W., Howell, D. G. Gas in tight reservoirs - an emerging major source of energy. The Future of Energy Gases: US Geological Survey Professional Paper, 1993, 1570: 233-252.
Molinski, N. S., Kenda, M., Leithner, C., et al. Deep learning‑enabled detection of hypoxic‑ischemic encephalopathy after cardiac arrest in ct scans: A comparative study of 2D and 3D approaches. Frontiers in Neuroscience, 2024, 18: 1245791.
Mukhametdinova, A., Kazak, A., Karamov, T., et al. Reservoir properties of low‑permeable carbonate rocks: Experimental features. Energies, 2020, 13(9): 2233.
Niu, Y., Wang, Y., Mostaghimi, P., et al. An innovative application of generative adversarial networks for physically accurate rock images with an unprecedented field of view. Geophysical Research Letters, 2020, 47(23): e2020GL089029.
Orlov, D., Ebadi, M., Muravleva, E., et al. Different methods of permeability calculation in digital twins of tight sandstones. Journal of Natural Gas Science and Engineering, 2021, 87: 103750.
Orlov, D., Gainitdinov, B., Koroteev, D. Darcy‑scale digital core models for rock properties upscaling and computational domain reduction. Journal of Computational Science, 2025, 92: 102715.
Orlov, D. M., Alekseev, V. V., Pimanov, V. O., et al. Multiscale digital core model for complex carbonate reservoirs. Gas Science Bulletin, 2022, 3(52): 78-89. (in Russian).
Portal, D. P. M. Estallades carbonate micro‑CT dataset (3.6 μm voxel size), 2020a.
Portal, D. P. M. Portal, D. P. M. Estallades carbonate micro‑CT dataset (3.1 μm voxel size), 2020b.
Ruspini, L. C., Oren, P. E., Berg, S., et al. Multiscale digital rock analysis for complex rocks. Transport in Porous Media, 2021, 139(2): 301-325.
Spurin, C., Callas, C., Darraj, N., et al. The importance and challenges associated with multi‑scale heterogeneity for geological storage. InterPore Journal, 2025, 2(1): IPJ260225-2.
Vidal, A. D., Neta, A. P. M., de Castro Vargas Fernandes, J., et al. Multi‑resolution X‑ray micro‑computed tomography images of carbonate rocks from brazilian pre‑salt. Scientific Data, 2024, 11(1): 1361.
Wu, L., Zhuang, J., Chen, H. Voco: A simple‑yet‑effective volume contrastive learning framework for 3D medical image analysis. Paper Presented at IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, WA, USA, 16-22 June, 2024.
Zheng, J., Zheng, L., Liu, H., et al. Relationships between permeability, porosity and effective stress for low‑permeability sedimentary rock. International Journal of Rock Mechanics and Mining Sciences, 2015, 78: 304-318.
Zhou, Y., Nie, X. Identification and parameter characterization of pores and fractures in shales based on multi‑scale digital core data. Paper Presented at Asia Petroleum Geoscience Conference and Exhibition 2024, Kuala Lumpur, Malaysia, 20-21 November, 2024.
Zhou, Z., Sodha, V., Rahman Siddiquee, M. M., et al. Models genesis: Generic autodidactic models for 3D medical image analysis. Paper Presented at International Conference on Medical Image Computing and Computer‑Assisted Intervention, Shenzhen, China, 13-19 October, 2019.
Zou, C., Zhu, R., Liu, K., et al. Tight gas sandstone reservoirs in china: Characteristics and recognition criteria. Journal of Petroleum Science and Engineering, 2012, 88: 82-91.
Downloads
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 Batyrkhan Gainitdinov, Rinat Prochii, Denis Orlov, Maxim Sharaev, Yili Ren, Dmitry Koroteev
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.