Exploring North Sea Fractured Sandstone Properties: Artificial Intelligence, Multiscale Imaging, Pore-Fracture Network Analysis and Experimental Results

Abstract

Abstract This paper presents a novel integrated workflow that enhances the understanding of matrix pore-fracture flow in lower-margin reservoir engineering. The workflow, which is applied to a North Sea reservoir core sample, combines experiments, imaging, deep-learning segmentation, and pore-scale simulation techniques. Advanced Artificial Intelligence (AI) models are used to analyse images from fractured and unfractured micro-computed tomography (micro-CT) scans. This enables a comprehensive multi-scale analysis crucial for optimising production in challenging reservoirs. The study commences with an examination of a relatively clean sandstone sample from a depleted North Sea sandstone hydrocarbon reservoir. A specially developed geomechanical-flow experimental cell induces and monitors fractures, offering critical insights. Post-fracture, in-situ imaging accurately captures fracture geometry. The analysis is further enhanced by AI-powered segmentation of image pairs, followed by a multiscale pore-network analysis, which experimentally validates the fracturing-flow processes. This study's findings have significant implications for reservoir development. By demonstrating how multi-scale, image-derived data can enhance understanding of porous features, the study provides a valuable tool for more efficient resource extraction in marginal fields. The workflow, which includes two-dimensional (2D) and three-dimensional (3D) deep convolutional neural networks (CNNs) with tailored objective functions and a novel algorithm for large-scale domain decomposition and pore network extraction, improves core-scale fracture-pore network modelling (fracture-PNM). The fluid simulation reveals intricate flow behaviours in matrix, fracture, and combined systems, offering crucial insights for advancing subsurface geo-energy processes like hydraulic fracturing, carbon and hydrogen storage, and deep geothermal energy systems. We introduce cutting-edge segmentation models using 2D and 3D CNNs tailored for multi-scale analysis of fractured systems. A novel 3D large-image PNM extraction and domain decomposition algorithm is proposed, enhancing the fidelity of core-scale PNM modelling. The study offers new perspectives on matrix-fracture flow mechanisms through experimentally validated modelling, enriching the current understanding of fluid dynamics in complex subsurface environments.