Absolute permeability estimation from microtomography rock images through deep learning super-resolution and...

Area of Science:

• Digital rock physics
• Artificial intelligence in geosciences
• Carbon capture and storage (CCS)

Background:

• Accurate subsurface reservoir characterization is crucial for effective carbon capture and storage (CCS) projects.
• X-ray microtomography (μ-CT) is vital in digital rock physics for analyzing rock properties like porosity and permeability.
• High-resolution μ-CT imaging is costly and time-consuming, limiting its widespread application in reservoir simulations.

Purpose of the Study:

• To develop a cost-effective and time-efficient method for petrophysical property estimation in reservoir rocks.
• To leverage deep learning for enhancing low-resolution μ-CT images and substituting complex numerical simulations.

Main Methods:

• Implementation of a deep learning-based super-resolution model to upscale low-resolution μ-CT images.
• Development of a surrogate model to predict petrophysical properties, bypassing traditional numerical simulations.
• Integration of a generative adversarial network (GAN)-inspired correction process to refine image quality and property estimations.

Main Results:

• The super-resolution model successfully enhanced the quality of low-resolution μ-CT images.
• The surrogate model, guided by the GAN-inspired correction, provided accurate estimations of petrophysical properties.
• The combined approach demonstrated significant improvements in permeability estimation accuracy on the DeePore dataset.

Conclusions:

• Deep learning offers a powerful solution to overcome the limitations of high-resolution imaging in digital rock physics.
• The proposed method enhances the efficiency and accuracy of petrophysical characterization for CCS applications.
• This approach facilitates better reservoir simulations and supports the development of more reliable carbon storage strategies.