Predicting Permeability and Capillary Pressure in Low-Resolution Micro-CT Images of Heterogeneous Laminated Sandstones

Abstract

Subsurface oil and gas reservoirs and fresh water aquifer systems are defined by fundamental geological characteristics such as mineral assemblage, grain and pore texture (size and shape), and porosity, and a range of petrophysical properties such as permeability, tortuosity, and capillary pressure, all of which contribute to fluid flow behaviour during extraction, injection, and storage. Computer-based models of reservoir and aquifer systems use these fundamental rock characteristics and petrophysical properties for large-scale fluid flow simulations. Designing and testing accurate static models is essential for reliable flow predictions. A wide range of analytical techniques has been developed over many years to expand the range and quality of formation modelling data. The most commonly used techniques include down-hole logging systems and laboratory-based core analysis. Down-hole logging tools measure the geophysical properties of formations, for example: gamma radiation and electrical resistivity, and typically collect data at the scale of tens of centimetres to metres, though image logs from micro-resistivity tools can collect millimetre to centimetre scale data. Commonly used laboratory-based analytical techniques involve the use of drill core, core plugs, and drill cuttings, for routine and special core/cuttings analysis to determine reservoir and seal rock properties. Modern X-ray micro-Computed Tomography (µCT) core imaging, in combination with petrophysical simulation software, often referred to as Digital Rock Physics, is fast becoming a standard tool for augmenting formation characterisation and modelling. Due to the nature of high-resolution µCT imaging and the associated analytical equipment, sample size is limited and governs the attainable resolution. It follows that metre-scale whole core samples cannot be imaged at the same high resolution as centimetre- and millimetre-scale core plugs. High-resolution images are critical to achieve reliable results from simulations of transport properties such as permeability and threshold injection pressure, which relies on all significant pathways in the pore space being correctly represented in the image. With current technology a µCT image of a 25mm diameter x 100mm tall sample, imaged using a detector with 2000 pixels per row, will have a minimum voxel size of ~13 µm, which implies that rock bands with grain and pore textures smaller than ~50 µm (i.e. 4 voxels across) cannot be represented with enough detail to reliably simulate petrophysical properties. The main research objective is to investigate the relationships between geological characteristics and petrophysical properties of heterogeneous laminated sandstone with the aim of estimating fluid flow properties for low-resolution images of larger rock volumes where fluid flow cannot be computed directly because of insufficient image resolution. This thesis presents an imaging and computation workflow for predicting absolute permeability, threshold pressure, lambda (a parameter in the Brooks-Corey equation describing the shape of drainage capillary pressure curves), and residual non-wetting phase saturation for sample volumes that are too large to allow direct computation of these properties or where traditional correlation methods fail. The workflow involves computing the above-mentioned petrophysical properties from high-resolution µCT images, along with a series of rock characteristics from spatially registered low-resolution images. Multiple linear regression models correlating the petrophysical properties to rock characteristics provide a means of predicting and mapping those property variations in larger scale low-resolution images.

Two core samples of 25 mm diameter 80 mm tall of heterogeneous sandstone, for which 5 µm/voxel resolution is required to compute permeability and capillary pressure directly, were investigated in this study. Results show good agreement between statistical predictions of petrophysical properties made from intermediate-resolution images at 16 µm/voxel and low-resolution images at 64 and 61 µm/voxel for samples 1 and 2 respectively. The statistical models to predict permeability from low-resolution images at 64 and 61 µm/voxel (similar to typical whole core image resolutions) include open pore fraction and formation factor as predictor characteristics. Although binarized images at this resolution do not completely capture the pore system, I infer that these characteristics implicitly contain information about the critical fluid flow pathways, which control permeability. Capillary pressure simulations were performed using both pore-morphology and network model-based methods. A prediction model of threshold pressure containing open pore fraction, formation factor, and, in this case, clay fraction is similar to the model of permeability from the low-resolution image of sample 1. My conclusion, which is similar to that of the permeability model results, is that formation factor and clay fraction, because their computation takes into account the image gray scale values, inherently capture information about the pore system length scale that controls threshold pressure. A surprising yet important result is that of sample 2, where the set of predictor characteristics are unable to accurately predict threshold pressure. I conclude that this is because of image processing difficulties arising from a low signal to noise ratio in the high-resolution image, which complicates the segmentation of pore space from grain volume. The result suggests that image quality is critically important, which potentially eliminates the use of data collected using imaging techniques like ‘region of interest’ scans. Statistical models of lambda using characteristics from pore morphology-based simulations describe 62% of the parameter variance. The predictor characteristics included in the model using low-resolution characteristics are open pore fraction, surface area, and mean curvature. Correlations between lambda computed from network model-simulations and low-resolution predictors are more encouraging with formation factor and clay fraction describing 93% of the variance in lambda. Predicting residual non-wetting phase saturation poses a significant challenge and was not successfully addressed in this project. Neither the morphology-based nor the network model simulations produced data that correlate well with predictor characteristics. In the case of the network model-derived data it is possible that a larger dataset may improve residual non-wetting phase predictions.