Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning Pore-scale Multi-phase Flow from Experimental Data with Graph Neural Network

Yuxuan Gu, Catherine Spurin, Gege Wen

Abstract

Understanding the process of multiphase fluid flow through porous media is crucial for many climate change mitigation technologies, including CO$_2$ geological storage, hydrogen storage, and fuel cells. However, current numerical models are often incapable of accurately capturing the complex pore-scale physics observed in experiments. In this study, we address this challenge using a graph neural network-based approach and directly learn pore-scale fluid flow using micro-CT experimental data. We propose a Long-Short-Edge MeshGraphNet (LSE-MGN) that predicts the state of each node in the pore space at each time step. During inference, given an initial state, the model can autoregressively predict the evolution of the multiphase flow process over time. This approach successfully captures the physics from the high-resolution experimental data while maintaining computational efficiency, providing a promising direction for accurate and efficient pore-scale modeling of complex multiphase fluid flow dynamics.

Learning Pore-scale Multi-phase Flow from Experimental Data with Graph Neural Network

Abstract

Understanding the process of multiphase fluid flow through porous media is crucial for many climate change mitigation technologies, including CO geological storage, hydrogen storage, and fuel cells. However, current numerical models are often incapable of accurately capturing the complex pore-scale physics observed in experiments. In this study, we address this challenge using a graph neural network-based approach and directly learn pore-scale fluid flow using micro-CT experimental data. We propose a Long-Short-Edge MeshGraphNet (LSE-MGN) that predicts the state of each node in the pore space at each time step. During inference, given an initial state, the model can autoregressively predict the evolution of the multiphase flow process over time. This approach successfully captures the physics from the high-resolution experimental data while maintaining computational efficiency, providing a promising direction for accurate and efficient pore-scale modeling of complex multiphase fluid flow dynamics.

Paper Structure

This paper contains 11 sections, 14 equations, 6 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Data Description
  3. Methodology
  4. Graph construction.
  5. Model architecture.
  6. Results
  7. Conclusion
  8. Pore structure
  9. More visualisation of the prediction
  10. Loss functions comparison
  11. Links of visualisation videos

Figures (6)

  • Figure 1: (a) A section in the rock is used to train a GNN model $f_{\theta}$. It learns to predict the next state based on its current state. (b) Five steps in the prediction pipeline: step 1. encoder that embeds node and edge features into latent space; step 2. fine message passing with all edges; step 3. coarse message passing only with long edges; step 4. fine message passing with all edges; step 5. a decoder that outputs the final node states. (c) Node encoder. (d) Edge encoder. (e) Decoder.
  • Figure 2: Visualisation (https://www.youtube.com/watch?v=CyOpiv1anCY) of ground truth and autoregressive rollout from $t = 221$ to 229 on the testing set with emerging bubbles. Red parts represent gas. Blue parts represent liquid.
  • Figure 3: (a) An example of a cross section of a porous rock showing gas and liquid particles separated by solid rocks. (b) Edge formation: connect nodes within $\sqrt{3}$ units apart.
  • Figure 4: Visualisation (https://www.youtube.com/watch?v=hbbWUJxL3ng) of ground truth and autoregressive rollout from $t = 288$ to 296 on the testing set with minor oscillations. Red parts represent gas. Blue parts represent liquid.
  • Figure 5: Visualisation (https://www.youtube.com/watch?v=nm4u_G7pOcs) of ground truth and autoregressive rollout from $t = 61$ to 69 on the training set with emerging bubbles. Red parts represent gas. Blue parts represent liquid.
  • ...and 1 more figures