Speaker
Description
Accurately characterizing rock properties at the core scale is fundamental for reliable reservoir-scale modeling. This task is particularly challenging in carbonate rocks due to their pronounced heterogeneities across multiple spatial scales. Although conventional core analysis methods yield precise laboratory measurements, they often fail to capture pore-scale variability within core plug samples.
Over the past decades, Digital Rock Physics (DRP) has emerged as a powerful framework to bridge this gap by combining X-ray computed tomography (CT), micro-CT imaging, and numerical simulations to analyze rock microstructures and derive physical properties. DRP has been extensively used to estimate porosity, permeability, and elastic moduli in both carbonate and siliciclastic rocks. However, despite these advances, a universally accepted workflow for the numerical characterization of carbonate rocks remains elusive.
This study introduces three innovative applications that leverage computer vision and machine learning for enhanced rock characterization. The first application centers on texture classification of X-ray CT images to identify and categorize rock fabrics. By modeling CT data, extracting representative textural descriptors, and applying the Kohonen self-organizing map (SOM)—an unsupervised learning method—distinct lithological textures within core samples are effectively classified.
The second application focuses on interpolating laboratory-measured rock properties, such as porosity and density, along entire core samples. This is achieved through a Convolutional Neural Network (CNN) trained on 3D X-ray CT data, which exploits the spatial continuity of these properties to generate high-resolution interpolations.
Finally, the third application presents a multiscale simulation framework for permeability and porosity prediction in heterogeneous carbonate samples using 3D X-ray CT images. This approach integrates machine learning–based texture classification results—enhanced by scattering and fractional scattering descriptors—into numerical upscaling workflows to quantitatively model heterogeneity across scales. To enrich the textural analysis, scattering and fractional scattering transforms are employed as advanced feature extraction techniques. These approaches capture multiscale spatial correlations and fine-grained structural details, offering a more robust and physics-inspired representation of rock textures compared to traditional statistical descriptors.
The proposed methodologies are demonstrated on two carbonate samples from a Middle Eastern carbonate oilfield, highlighting the potential of combining DRP, scattering-based texture analysis, and deep learning for comprehensive rock characterization and improved reservoir modeling.
| Country | United Arab Emirates |
|---|---|
| Acceptance of the Terms & Conditions | Click here to agree |
