Automated Carbonate Reservoir Pore and Fracture Classification by Multiscale Imaging and Deep Learning

Abstract

Summary Carbonate rocks are heterogeneous at microscopic and macroscopic scales, hence, their characterization is challenging, expensive and time-consuming. Petrophysical analyses cannot provide information on the full geometry and pore space connectivity. Moreover, these analyses are time consuming and require evaluation by experts. Current imaging techniques do not cover a representative range of scales, have difficulty to image microporosity, and are not well integrated with the expert knowledge available. An automated tool for identifying and classifying pores and fractures does not yet exist. This contribution presents a novel multi-scale workflow dedicated to carbonate rocks that integrates innovative methods with state-of-the-art imaging technologies for automated classification of connected pores from nano- to centimeter scale: 1) Virtual Petrograph (ViP), an automated high resolution petrographic microscope to acquire and visualize high-resolution cross-polarized image-maps of ultra-thin sections; 2) Broad Ion Beam – Scanning Electron Microscopy (BIB-SEM), a 2D preparation and imaging technique that preserves the most delicate microstructures and images microporosity in detail over representative areas; 3) Liquid Metal Injection (LMI) followed by BIB-SEM, a porosimetry technique to distinguish connected and unconnected pore space. Validated pore maps will be the input for statistical analysis and used to train deep learning algorithms for pore segmentation and classification.