A Multiphase Geological Segmentation of the Rotondo Granite

Accurate analysis of rock pore structures is pivotal for predicting their performance in geological applications, such as CO₂ storage, geothermal energy extraction, and radioactive waste disposal. Advanced imaging techniques, such as Digital Rock Physics (DRP), have transformed the characterization of porous media by providing detailed insights into microstructural properties.

DRP employs high-resolution X-ray computed tomography (CT) to scan rock samples, generating 3D images that reveal pore structures, mineral phases, and fracture networks. This enables the computation of key rock properties, including porosity, permeability, thermal conductivity, and elastic moduli. These properties are critical for applications in sustainable energy production and seismic hazard monitoring.

The classical five-step workflow of DRP begins with the preparation of a high-resolution X-ray computed tomography (CT) image. This is followed by tomographic reconstruction of the image using simple back-propagation techniques. Next, preprocessing operations are conducted to assess and manage artifacts before proceeding to the segmentation of individual phases. Based on the segmentation results, key rock properties such as thermal conductivity, permeability, and elastic properties are computed by solving the corresponding physical equations.

The accurate segmentation of digital rock models into different phases and features holds a significant importance and remains a formidable challenge, as it directly impacts the precise characterization of subsequent physical properties. However, most studies using classical segmentation techniques, such as grayscale histogram processing or watershed algorithms usually relied on a binary segmentation process. This means that rock samples are divided into just two phases: pore and solid. This oversimplifies the complexity of multiphase rocks, neglecting the heterogeneity of, i.e., granitic rocks and compromising the accuracy of reservoir characterizations.

This study presents a multiphase geological segmentation approach for the Rotondo Granite, integrating geological ground truth for potential machine learning applications. By accounting for every distinct mineral phase within a granitic reservoir rock, this method moves beyond traditional binary segmentation, enabling a more detailed and accurate representation of rock microstructures. The improved segmentation accuracy enhances the precision of DRP analyses, contributing to more reliable underground reservoir assessments and supporting the transition toward sustainable energy technologies.