KFU. Card publication. Physics-Guided Machine Learning for Predicting Gas Permeability of Standard Carbonate Core Plugs from Low-Resolution Microtomography Image Stacks

- RUS
- 中文
- ES

PHYSICS-GUIDED MACHINE LEARNING FOR PREDICTING GAS PERMEABILITY OF STANDARD CARBONATE CORE PLUGS FROM LOW-RESOLUTION MICROTOMOGRAPHY IMAGE STACKS

Form of presentation

Articles in Russian journals and collections

Year of publication

2025

��

��

Authors, employees of KFU

Kadyrov Rail Ilgizarovich, author

Nguen Tkhan Khyng , author

Stacenko Evgeniy Olegovich, author

Bibliographic description in the original language

Kadyrov, R. I. Physics-Guided Machine Learning for Predicting Gas Permeability of Standard Carbonate Core Plugs from Low-Resolution Microtomography Image Stacks / R. I. Kadyrov, T. H. Nguyen, E. O. Statsenko // Scientific Visualization. – 2025. – Vol. 17, No. 3. – P. 35–48. – DOI: 10.26583/sv.17.3.04.

Annotation

This study presents a physics-guided workflow for predicting the gas permeability of car-bonate reservoirs directly from low-resolution microtomography (μCT) imagery. Standardcore plugs were scanned at 34.6–36 μm/voxel, and a total of 52,327 grayscale d aggregationagainst experimental plug-scale measurements. The grayscale images and log-transformedpermeability labels were used to train a Swin Transformer model, pre-trained on ImageNet.Two models were developed independently: one using harmonic-mean aggregation and theother using the bottleneck approach. Both models demonstrate stable convergence despitethe highly skewed data distribution. The harmonic-mean model achieved R² = 0.904 on thevalidation set, while the bottleneck model yielded R² = 0.879. Although the higher R² reflectsa closer fit to the overall trend, the bottleneck model, in blind testing on ten independentsamples (0.4–2300 μm² × 10⁻³), reduced the MAE from 165 to 104 μm² × 10⁻³ (−37 %) andthe RMSE from 255 to 140 μm² × 10⁻³ (−45 %) relative to the harmonic-mean model. Themethod provides a fast and interpretable permeability prediction based solely on raw μCTslices, without requiring image segmentation or 3D reconstruction. The proposed approachdemonstrates robust performance across a wide range of standard carbonate plugs and effec-tively captures permeability trends even in the presence of structural heterogeneity. Whilesamples with extremely large fractures or vugs can introduce local inconsistencies in labellingdue to the limitations of slice-based estimation, these cases are rare and can be systematicallyaddressed in future work. Overall, the results highlight the strong potential of physics-guidedmachine learning to accelerate digital core analysis and provide reliable, image-driven per-meability predictions for complex carbonate reservoirs.

Keywords

permeability, carbonates, μCT, digital core, porous structure, standard coreplug, physics-guided machine learning, 2D image analysis

The name of the journal

Scientific Visualization

Please use this ID to quote from or refer to the card

https://repository.kpfu.ru/eng/?p_id=316697&p_lang=2

Full metadata record

Field DC	Value	Language
dc.contributor.author	Kadyrov Rail Ilgizarovich	ru_RU
dc.contributor.author	Nguen Tkhan Khyng	ru_RU
dc.contributor.author	Stacenko Evgeniy Olegovich	ru_RU
dc.date.accessioned	2025-01-01T00:00:00Z	ru_RU
dc.date.available	2025-01-01T00:00:00Z	ru_RU
dc.date.issued	2025	ru_RU
dc.identifier.citation	Kadyrov, R. I. Physics-Guided Machine Learning for Predicting Gas Permeability of Standard Carbonate Core Plugs from Low-Resolution Microtomography Image Stacks / R. I. Kadyrov, T. H. Nguyen, E. O. Statsenko // Scientific Visualization. – 2025. – Vol. 17, No. 3. – P. 35–48. – DOI: 10.26583/sv.17.3.04.	ru_RU
dc.identifier.uri	https://repository.kpfu.ru/eng/?p_id=316697&p_lang=2	ru_RU
dc.description.abstract	Scientific Visualization	ru_RU
dc.description.abstract	This study presents a physics-guided workflow for predicting the gas permeability of car-bonate reservoirs directly from low-resolution microtomography (μCT) imagery. Standardcore plugs were scanned at 34.6–36 μm/voxel, and a total of 52,327 grayscale d aggregationagainst experimental plug-scale measurements. The grayscale images and log-transformedpermeability labels were used to train a Swin Transformer model, pre-trained on ImageNet.Two models were developed independently: one using harmonic-mean aggregation and theother using the bottleneck approach. Both models demonstrate stable convergence despitethe highly skewed data distribution. The harmonic-mean model achieved R² = 0.904 on thevalidation set, while the bottleneck model yielded R² = 0.879. Although the higher R² reflectsa closer fit to the overall trend, the bottleneck model, in blind testing on ten independentsamples (0.4–2300 μm² × 10⁻³), reduced the MAE from 165 to 104 μm² × 10⁻³ (−37 %) andthe RMSE from 255 to 140 μm² × 10⁻³ (−45 %) relative to the harmonic-mean model. Themethod provides a fast and interpretable permeability prediction based solely on raw μCTslices, without requiring image segmentation or 3D reconstruction. The proposed approachdemonstrates robust performance across a wide range of standard carbonate plugs and effec-tively captures permeability trends even in the presence of structural heterogeneity. Whilesamples with extremely large fractures or vugs can introduce local inconsistencies in labellingdue to the limitations of slice-based estimation, these cases are rare and can be systematicallyaddressed in future work. Overall, the results highlight the strong potential of physics-guidedmachine learning to accelerate digital core analysis and provide reliable, image-driven per-meability predictions for complex carbonate reservoirs.	ru_RU
dc.language.iso	ru	ru_RU
dc.subject	permeability	ru_RU
dc.subject	carbonates	ru_RU
dc.subject	μCT	ru_RU
dc.subject	digital core	ru_RU
dc.subject	porous structure	ru_RU
dc.subject	standard coreplug	ru_RU
dc.subject	physics-guided machine learning	ru_RU
dc.subject	2D image analysis	ru_RU
dc.title	Physics-Guided Machine Learning for Predicting Gas Permeability of Standard Carbonate Core Plugs from Low-Resolution Microtomography Image Stacks	ru_RU
dc.type	Articles in Russian journals and collections	ru_RU