Articles and whitepapers

We introduce a novel method capable of generating large-scale 3D images of material microstructures from a single 2D image. Our approach circumvents the need for 3D image data while offering a cost-effective, high-resolution alternative to existing imaging techniques. This technique facilitates our method’s ability to gradually generate 3D images that accurately capture the geometric properties and statistical characteristics of the original 2D input.

We performed unsteady-state quasi-static oil- and waterflooding experiments to study multiphase flow in two sister heterogeneous sandstones with variable wettability conditions.

The pore-scale fluid distributions during this process were imaged by micro- CT. In the mixed-wet case, we observed pore filling events where the fluid interface appeared to be at quasi-equilibrium at every position along the pore body (13% by volume), in contrast to capillary instabilities typically associated with slow drainage or imbibition.

We investigated the sensitivity of the multi-phase flow properties to the slow filling events using a state-of-the-art multi-scale pore network model. Our study sheds new light on poorly understood multiphase fluid dynamics in complex rocks, of interest to e.g. groundwater remediation and subsurface CO2 storage.

We present SIT-SR 3D, a novel self-supervised method for 3D single image super-resolution (SISR).  The proposed approach learns to super-resolve 3D low-resolution images in a self-supervised manner.

SIT SR-3D is especially useful when 3D ground-truth is not available, but 2D ground-truth is available. In the case of digital rocks, it is often the case that 3D high-resolution ground truth is not available.

We demonstrated our results on an open-source digital rock CT dataset.

Image registration is a process used to align or register multiple images or volumes to facilitate comparison or combination of the data. In the context of 3D wet and dry images of rock samples, it is essential to accurately align these images to analyze and utilize the data in various experiments.

We propose a new optimization-based algorithm for image registration of large 3D wet and dry images of rock samples. We have demonstrated that our algorithm can provide a solution in under a minute for samples of size 1000³ cube, compared to the several hours of expert time needed by the current industrial practice.

We provide quantitative and qualitative results and compare our algorithm to the solution time of a human expert.

In this technical paper we describe a robust methodology for performing advanced analysis on smaller sample volumes from unconventional reservoirs. The described methodology incorporates a multi-scale and multi-modal imaging workflow where scanning electron microscopy and spatial mineralogy are used for rock characterization, followed by acquisition of 3D volumes with nanometer resolution for pore scale analysis and generation of digital 3D models.

Elastic properties were calculated in addition to basic petrophysical properties, and comparisons with laboratory data as well as literature data have been conducted to validate the predicted properties.

In this work, we propose a size-invariant multi-step 3D generation workflow from a single 2D image. The proposed workflow tackles several major challenges in the generation of 3D images since it is designed to not only satisfy the large size constraint (>1000³ voxels cube) but also to generate statistically representative pore structures.

To validate our methodology, we have generated several large-size 3D rock images and compare them to real 3D images in terms of physical properties (porosity, permeability, and Euler characteristic).

In this work, we show how to super-resolve a single image using an implicit neural network to produce sharp and photo-realistic images. We improved the perceptual quality of the implicit neural network based single image super-resolution. The main hindrance in utilizing adversarial losses for continuous image representation models was the random co-ordinate-based sampling procedure adopted by previous works. We proposed to use a patch-based sampling method. Then we trained the implicit neural network with additional objectives based on adversarial and perceptual losses. We demonstrated that the resulting network produces sharp and photo-realistic images while maintaining the desirable properties of the implicit neural networks.

Recent multi-scale models incorporate sub-voxel porosity maps, created by differential micro-CT imaging of a contrast fluid in the pores. These maps delineate different microporous flow zones in the model, which must be assigned petrophysical properties as input. The uncertainty on the pore scale physics in these models is therefore heightened by uncertainties on the representation of unresolved pores, also called sub-rock typing.

We address this by validating a multi-scale pore network model using a drainage experiment imaged with differential micro-CT on an Estaillades limestone sample. We found that porosity map-based sub-rock typing was unable to match the micrometer-scale experimental fluid distributions. To investigate why, we introduce a novel baseline sub-rock typing method, based on a 3D map of the experimental capillary pressure function.

We have developed a tool that aims to fully automate the segmentation process in one step, without the need for any extra image processing steps such as noise filtering or artifact removal.

The results show a robust and accurate segmentation performance for the most common types of noises present in real micro-CT images. We also compared the segmentation of our method and five expert users, using commercial and open software packages on real rock images. We found that most of the current tools fail to reduce the impact of local and global noises and artifacts.

We introduce a generalized multiscale imaging and pore-scale modelling workflow to derive transport properties of complex rocks having broad pore size distributions.

The multiscale digital rock workflow is applied to two heterogeneous rock samples: a mixed wet thinly laminated reservoir sandstone and an oil wet reservoir carbonate. Experimentally measured mercury–air primary drainage and oil–water imbibition capillary pressure curves are used to anchor the multiscale pore network model. Waterflood relative permeability is calculated in a blind test and compared with high-quality experimental data. A very encouraging agreement between computed and measured properties is found.

Comparison of the fluid arrangement evolution measured in fast synchrotron micro-CT experiments on two rock types to quasi-static simulations which implement capillary-dominated pore filling and snap-off, including a sophisticated model for cooperative pore filling.

The results indicate that such pore network models can, in principle, predict fluid distributions accurately enough to estimate upscaled flow properties of strongly wetted rocks at low capillary numbers.

Three-dimensional X-ray computed tomography was used to characterize the pore structure of the reservoir core. We obtained in-situ pore-scale images of the distribution of CO2: brine analogue fluid pairs within reservoir samples during low capillary number drainage and imbibition flooding experiments. The micro- CT images are used directly as input to a pore-scale simulation model. The validity is investigated by comparing on a pore-by-pore basis the simulated and imaged fluid distributions. The pore filling states are in good agreement both for drainage and imbibition displacements and the computed capillary trapping curve agrees with experimental data.

A pore-network model study of capillary trapping in water-wet porous media was presented. The amount and distribution of trapped non-wetting phase is determined by the competition between two trapping mechanisms - snap-off and cooperative pore-body filling. A new model to describe the pore- body filling mechanism in geologically realistic pore-networks was developed. The model accounts for the geometrical characteristics of the pore, the spatial location of the connecting throats and the local fluid topology at the time of the displacement, which was validated by comparing computed capillary trapping curves with published data for four different water-wet rocks. Computations were performed on pore-networks extracted from micro-CT images and process-based reconstructions of the actual rocks used in the experiments. Compared with commonly used stochastic models, the new model describes more accurately the experimental measurements, especially for well connected porous systems where trapping is controlled by subtleties of the pore structure. The new model successfully predicts relative permeabilities and residual saturation for Bentheimer sandstone using in-situ measured contact angles as input to the simulations. The simulated trapped cluster size distributions are compared with predictions from percolation theory.