Computed Tomography 3D Super-Resolution with Generative Adversarial Neural Networks: Implications on Unsaturated and...

Area of Science:

• Geosciences
• Artificial Intelligence
• Petroleum Engineering

Background:

• Accurate fluid flow characterization is vital for reservoir performance assessment.
• Laboratory methods are insufficient for determining these characteristics, necessitating numerical approaches.
• Computed tomography (CT) scans, while useful, face resolution limitations impacting accuracy.

Purpose of the Study:

• To investigate the impact of CT scan resolution on pore network properties and fluid flow.
• To develop and apply super-resolution generative adversarial networks (GANs) to enhance CT image resolution.
• To evaluate the effectiveness of GANs in improving predictions of fluid flow characteristics.

Main Methods:

• Analysis of how varying resolution affects pore network characteristics and fluid flow (single-phase, unsaturated, two-phase).
• Development of super-resolution GANs to increase CT image resolution from 12 µm to 4 µm.
• Direct application of GANs to raw CT images of geological materials.

Main Results:

• Lower resolution leads to larger average pore/throat sizes and decreased surface area, overestimating permeability and affecting moisture uptake and residual oil.
• Super-resolution GANs successfully enhanced image resolution, yielding better predictions of pore network and fluid flow properties.
• Improved resolution better captured small pores and surfaces, leading to more accurate estimates of unsaturated and two-phase flow.

Conclusions:

• Super-resolution GANs offer a viable method to overcome CT resolution limitations in geological material analysis.
• Enhanced resolution significantly improves the prediction accuracy of critical fluid flow properties.
• This study is the first to apply GANs directly to raw CT images for super-resolution in geology and assess its impact on fluid flow predictions.