Advanced AI and ML Solution to Expedite Geological Rocktyping from Petrographic Thinsections to Improve Reservoir Characterization

Abstract Heterogenous carbonate reservoirs exhibit diverse flow characteristics. Petrographic study on thinsection images is essential for rock typing studies. Optical point counting is a conventional method which is semi-quantitative, and thinsection descriptions are in text format. Performing optical point counting is a cumbersome and time-consuming process requiring specialist skills and domain expertise. Optical point counting provides data for modal analysis to assess reservoir quality for rocktyping. Image processing of thinsection is a swift and robust alternative to optical point counting method. Proposed AI and ML solution automates modal analysis for digital reservoir description and augments available optical count data to large number of thinsection images in quick turnaround to enhance geological rocktyping. An innovative AI and ML solution was developed by leveraging image processing algorithms and computing power to improve reservoir characterization. Computational algorithms mainly, multilevel thresholding and pixel intensity clustering algorithms were programmed to segment images and labelled by comparing with original image. The labeled elements were interpreted for geological elements such as matrix, pores, cement, and other granular content. Algorithms were customized for thinsection image processing by honoring complex pore types in carbonate reservoirs. The interpreted geological elements were measured for their physical properties like area, equivalent diameter, perimeter, solidity, eccentricity, and entropy. Geological rocktyping was carried out using K-Means classification based on digital proxy reservoir quality (DPRQ) parameters and compared with rock types made from conventional methods. The whole process was automated in a batch process for a specific reservoir type and computational cost was analyzed for optimization. Automated modal analysis workflow produced digital thinsection data comparable to optical count data under microscopic analysis. Digital proxy rock quality (DPRQ) parameters were introduced and proved to be quick and reliable reference to expedite geological rocktyping instead of conventional workflows. Integration of laboratory nuclear magnetic resonance data enabled the bridge required for the integration of 2-D geometric properties with three-dimensional nature of conventional core analysis data from core plugs. This paper demonstrated a unique and innovative solution to extract digital data from thinsections for modal analysis. Rocktyping workflow is enhanced over conventional workflow by defining DPRQ parameters. This paper presents a novel means to use thinsection images directly in digital format in geoscience applications. As conventional workflows are limited by number of samples, time, cost and expert skill, this workflow can be scalable to large number of thinsections to expedite rocktyping in contemporary reservoir characterization workflows.