Core rock samples provide direct evidence of pore systems of subsurface reservoirs which significantly impact fluid flow, storage, and transport dynamics within the reservoirs. Images including thin sections, micro-CT and SEM are the most readily available methods for visualizing rock pore systems and have been extensively used for reservoir characterization. However, conventional imaging techniques are subject to a variety of resolution, sample size and two-dimensionality limitations. This Ph.D. thesis integrates diverse rock sample images from reservoir equivalent outcrop analogues of the Upper Jurassic formations in Saudi Arabia, and presents three image-based machine-learning workflows designed to achieve efficient, precise and non-subjective characterization of reservoir rocks. This study begins with using deep-learning methods to predict the Dunham texture and depositional facies of thin sections. The high predictive accuracy rates of over 86% indicate the reliability of this technique. To the best of our knowledge, this is the first instance of utilizing integrated deep-learning architectures to predict depositional facies. As a next step a workflow has been developed for estimating reservoir properties from thin sections which consists of image segmentation, 2D to 3D image reconstruction and pore network modeling.  The low-resolution and two-dimensionality of thin section images prevent accurate characterization of the complex inner structures of rock samples. Thus, we developed a novel workflow that considers resolution, sample size and computational demand. It commences with using Super Resolution to segment the µ-CT images into solids, macropores and microporous medium with over 83% pixel-wise accuracy. Next, we generate an individual pore network model (PNM) for each pore structure with the macro-PNM directly extracted from the µ-CT images, and the micro-PNM stochastically generated from SEM images. These PNMs are then stitched together to create a multi-scale PNM that represents the intricate pore structures of the rock sample. Comparative analysis of flow simulation using the multi-scale PNM against laboratory measurements highlights the accuracy, efficiency, and robustness of the workflow. The technology developed by this study, provides both academia and industry with validated workflows to comprehensively analyze reservoirs from the core- to pore-scale and subsequently facilitate accurate predictions of reservoir performance and optimization of recovery strategies.

Zoom Link: https://kaust.zoom.us/j/95536260668, Meeting ID: 955 3626 0668