An Integrated Deep Learning and Physics-Constrained Upscaling Workflow for Robust Permeability Prediction in Digital Rock Physics

Abstract

Research on rock permeability plays a crucial role in understanding fluid flow within geological formations, which contributes significantly to addressing challenges in sustainable geo-energy production and CO2 sequestration. Currently, numerous advanced techniques are employed for predicting permeability in porous media, including experimental methods, numerical simulations, and deep learning approaches. However, existing methods often face difficulties in accurately predicting properties across multiple scales. In this study, we propose an integrated workflow that combines deep learning and physical constraints to achieve accurate and multi-scale permeability prediction. The workflow begins with a 3D pore segmentation, where a novel architecture of 3D Inception U-Net is developed to achieve a segmentation accuracy exceeding 0.99. We then introduce a progressive transfer learning approach to directly predict permeability at varying scales. This method achieves R2 scores of 0.94, 0.83, and 0.84 for sub-volumes of 1503, 3003, and 6003 voxels (2.25 µm/voxel), respectively. To address reduced accuracy for larger sub-volumes, we further develop a physics-constrained upscaling method. This approach enhances predictive performance, achieving R2 scores of 0.98 for transitions from 1503 to 3003 and 0.99 for transitions from 3003 to 6003 sub-volumes. This research underscores the potential of integrating advanced deep learning with physics-based constraints, providing a robust framework for accurate and scalable permeability prediction in digital rock physics and paving the way for core-scale applications and future studies.

Keywords: Digital rock, deep learning, pore segmentation, permeability prediction, multi-scale permeability analysis, physics constraint