Search

This paper presents a hybrid deep learning framework that combines graph neural networks with convolutional neural networks to predict porous media properties. This approach capitalizes on the capabilities of pre-trained convolutional neural networks to extract ndimensional feature vectors from processed three dimensional micro computed tomography porous media images obtained from seven different sandstone rock samples. Subsequently, two strategies for embedding the computed feature vectors into graphs were explored: extracting a single feature vector per sample (image) and treating each sample as a node in the training graph, and representing each sample as a graph by extracting a fixed number of feature vectors, which form the nodes of each training graph. Various types of graph convolutional layers were examined to evaluate the capabilities and limitations of spectral and spatial approaches. The dataset was divided into 70/20/10 for training, validation, and testing. The models were trained to predict the absolute permeability of porous media. Notably, the proposed architectures further reduce the selected objective loss function to values below 35 mD, with improvements in the coefficient of determination reaching 9%. Moreover, the generalizability of the networks was evaluated by testing their performance on unseen sandstone and carbonate rock samples that were not encountered during training. Finally, a sensitivity analysis is conducted to investigate the influence of various hyperparameters on the performance of the models. The findings highlight the potential of graph neural networks as promising deep learning-based alternatives for characterizing porous media properties. The proposed architectures efficiently predict the permeability, which is more than 500 times faster than that of numerical solvers. [ABSTRACT FROM AUTHOR]

This paper presents a novel technique for upscaling multiphase fluid flow in complex porous materials that combines asymptotic homogenization approach with hydrodynamicand thermodynamic bounds. Computational asymptotic homogenization has been widely utilised in solid mechanics as a method for analysing multiscale expansion and convergence coefficients in heterogeneous systems. Computations are performed over several volumes by increasing the size until convergence of the material parameters under different load scenarios is achieved. It works by simplifying the problem with a homogenization method and is ideally suited for estimating the representative elementary volume of microporous material by expanding algorithms. The validity of the method to include complex multiphase hydrodynamic processes and their interaction with the matrix structure of porous media lacks a sound theoretical foundation. To overcome this problem, a variational thermodynamic approach is used. Upper and lower bounds of entropy production are proposed to provide effective material properties with uncertainties. This allows multiple possibilities to address dynamics via thermodynamically linked processes. This work utilizes volume of fluid approach to model multiphase porous media flow in models based on micro-computerized tomography x-ray data of Bentheimer sandstone and Savonnieres carbonate. It is found that the representative elementary volume sizes obtained by the conventional asymptotic homogenization methods do not satisfy thermodynamic bounds which consistently require larger representative elementary volume sizes. For the Savonnieres carbonate the entropic bounds have not converged fully questioning the reliability of the effective properties obtained from the classical method. [ABSTRACT FROM AUTHOR]

A tomographic or image representation of three-dimensional (3D) porous media is a primary tool for the consistent correlation and prediction of multiple physical properties. For multiscale media a single imaging scale may not be sufficient, and a compromise between field of view and resolution is required. The resulting lower resolution compared to relevant length scales in part of the tomogram impacts on the quality of petrophysical cross-correlations. This situation is overcome in geostatistics by carrying out stochastic simulations with lower resolution constraints and propagating uncertainty from smaller scales through upscaling procedures capturing the behavior of the system at smaller scales. Applying this technique to full tomograms requires a fast high-resolution reconstruction technique. We propose a local-similarity statistic reconstruction (LSSR) method to reconstruct 3D high-resolution porous structure by combining a set of increasing resolution micro-computed-tomography (micro-CT) images with decreasing field of view (FOV) to overcome this limitation. The reconstruction technique is based on two assumptions, universally existent local similarity at a given scale and fixed image degradation when downscaling by the same factor between corresponding cubes of high- and low-resolution images. Utilizing the flexibility of micro-CT images in terms of resolution and FOV, a sample was scanned first with low resolution, and then a subset was taken from this sample and scanned with high resolution. A significant number of small cube pairs are extracted from corresponding parts of the low-resolution and high-resolution images, where both high- and low-resolution images are accessible. These cube pairs contain abundant information about features of local porous structure under different resolutions allowing reconstruction of a higher-resolution micro-CT image. Instead of the "Search-Statistic" strategy popularly used in current reconstruction algorithm a "Decomposition-Reconstruction" strategy is applied to accelerate computations and improve the reconstruction accuracy. Local porosity theory and the Minkowski measures are used to estimate the performance of reconstruction algorithms. Compared to multipoint statistics we improve computational efficiency by employing a sparse representation and principal component analysis for compression, while also improving on reconstruction accuracy. [ABSTRACT FROM AUTHOR]

Carbon dioxide (CO 2 ) injection in unconventional gas-bearing shale reservoirs is a promising method for enhancing methane recovery efficiency and mitigating greenhouse gas emissions. The majority of methane is adsorbed within the micropores and nanopores (≤50 nm) of shale, which possess extensive surface areas and abundant adsorption sites for the sequestration system. To comprehensively discover the underlying mechanism of enhanced gas recovery (EGR) through CO 2 injection, molecular dynamics (MD) provides a promising way for establishing the shale models to address the multiphase, multicomponent fluid flow behaviors in shale nanopores. This study proposes an innovative method for building a more practical shale matrix model that approaches natural underground environments. The grand canonical Monte Carlo (GCMC) method elucidates gas adsorption and sequestration processes in shale gas reservoirs under various subsurface conditions. The findings reveal that previously overlooked pore slits have a significant impact on both gas adsorption and recovery efficiency. Based on the simulation comparisons of absolute and excess uptakes inside the kerogen matrix and the shale slits, it demonstrates that nanopores within the kerogen matrix dominate the gas adsorption while slits dominate the gas storage. Regarding multiphase, multicomponent fluid flow in shale nanopores, moisture negatively influences gas adsorption and carbon storage while promoting methane recovery efficiency by CO 2 injection. Additionally, saline solution and ethane further impede gas adsorption while facilitating displacement. Overall, this work elucidates the substantial effect of CO 2 injection on fluid transport in shale formations and advances the comprehensive understanding of microscopic gas flow and recovery mechanisms with atomic precision for low-carbon energy development.

Carbon dioxide (CO 2 ) injection in shale and coal seam gas reservoirs has become one of the most popular ways to promote methane (CH 4 ) production. However, geological factors affecting the CO 2 enhanced gas recovery (CO 2 -EGR) projects have not been studied in great depth, including underground moisture, subsurface water salinity, and other gases accompanying CH 4 . Thus, a hybrid methodology of molecular dynamics (MD) and grand canonical Monte Carlo (GCMC) simulation is employed to reveal the gas adsorption and displacement mechanisms at a fundamental molecular level. This study generates a type II-D kerogen matrix as the adsorbent. The simulation environment includes 0-5 wt % moisture content, 0-6 mol/L NaCl saline, and 0-5 wt % C 2 H 6 for up to 30 MPa at 308, 338, and 368 K. The impressions of moisture, C 2 H 6 , and salinity on gas adsorption and competitive adsorption characteristics are analyzed and discussed. On the basis of the simulation results, the preloaded H 2 O molecules negatively influence CH 4 adsorption, leading to a 44.9% reduction at 5 wt % moisture content. Additionally, 6 mol/L NaCl within 5 wt % moisture content exhibits a further 9.8% reduction on the basis of the moisture effect. C 2 H 6 presents a more noticeable negative impact, of which 5 wt % results in a 73.2% reduction in CH 4 adsorption. Moreover, the competitive process indicator, preferential selectivity S CO 2 /CH 4 , is analyzed and discussed in the presence of the mentioned factors. Moisture positively influences S CO 2 /CH 4 , salinity promotes S CO 2 /CH 4 , and C 2 H 6 develops S CO 2 /CH 4 . These factors would encourage the displacement processes of CH 4 by CO 2 injection. This study provides essential information for better gas resource estimation and gas recovery improvement in unconventional systems.