The Digital Rock workflow is an emerging framework utilizing advances in imaging technologies and state-of-the-art image processing algorithms to construct digital models of reservoir rocks. These digital models become input to a physics-based simulations to compute several petrophysical properties of interest. A typical digital rock image analysis pipeline consists of acquiring projections using X-ray micro-computed tomography (CT) scans of mini core plugs drilled out of reservoir cores, from which a three-dimensional (3D) model of the rock sample will be reconstructed by applying various reconstruction algorithms (FDK, MLEM, etc.). These reconstructed models are further segmented, resulting in geometric models of their solid and pore spaces to perform numerical simulations of different physical phenomena like fluid flow, wave propagation, electric flow, and other physical phenomena. Several petrophysical properties (Porosity, Permeability, etc.) can be estimated by these simulations, which would otherwise be difficult to measure in standard experimental settings as there the rock samples may get physically damaged.
The accuracy of the digital rock pipeline relies crucially on the quality of acquired images, the resolution being limited by the hardware of micro-CT scanning technology. This thesis develops two novel and fully data driven methodologies to improve the digital rock workflow performance with these methods being computationally efficient.
Enhancing Quality of Acquired Images:
The quality of the 3D segmented rock model subjected to various physics simulations may not be optimal due to the noise present in micro-CT scans because of shortcomings in the acquisition process, such as limited acquisition time. Therefore, a pre-processing step that involves denoising of the micro-CT scans is mandatory. For rock image analysis, preserving the size of the narrowest corners and throats in the pore space is of primary importance. This thesis proposes a novel denoising pipeline consisting of morphology-based noise reduction using multi-resolution version of structurally varying bitonic filtering, followed by a non-local means smoothing. We show the efficacy of the proposed approach in terms of a non-reference metric and also use the Equivalent Number of Looks (ENL) to incorporate geologists marking while computing signal-to-noise ratio (SNR) of different important regions in the filtered images.
Enhancing Resolution of Acquired Images: The accuracy of the petro-physical properties in digital rock is crucially dependent on the resolution of the digital images of the 3D rock samples, which is currently limited by the resolution of the micro-CT scanning technology. Furthermore, for heterogeneous rock samples with a large Representative Elementary Volume (REV), capturing a large field of view with a high-resolution acquisition can be very difficult and time-consuming. Therefore, in this part of the thesis, there is a major emphasis on developing computational techniques to enhance the resolution of the rock volumes digitally. We have proposed a novel deep learning-based super-resolution model called Siamese-SR to improve the resolution of Digital Rock images whilst retaining the texture and providing optimal denoising. The Siamese-SR model consists of a generator that is adversarially trained with a relativistic and a Siamese Discriminator utilizing Materials In Context (MINC) loss estimator.
Proposing Physics-based Quantification Metric: Another key highlight of the thesis is, for the evaluation of the super-resolution performance, we propose to move away from image-based metrics such as Structural Similarity (SSIM) and Peak Signal to Noise Ratio (PSNR) as they do not correlate well with expert geological and petrophysical evaluations. Instead, we propose to subject the super-resolved images to the next step in the Digital Rock workflow to calculate porosity, which is a petrophysical property of interest, and use it as a metric for evaluation of our proposed Siamese-SR or denoising methods. Furthermore, we have also demonstrated that the improvement due to super-resolution with Siamese-SR model is not just restricted to improved image-based porosity, but also more accurate estimation of petrophysical properties of interest derived using Mercury Injection Capillary Pressure (MICP) simulations.