Abstract:Simulation of fluid flow in porous media has many applications, from the micro-scale (cell membranes, filters, rocks) to macro-scale (groundwater, hydrocarbon reservoirs, and geothermal) and beyond. Direct simulation of flow in porous media requires significant computational resources to solve within reasonable timeframes. An integrated method combining predictions of fluid flow (fast, limited accuracy) with direct flow simulation (slow, high accuracy) is outlined. In the tortuous flow paths of porous media, Deep Learning techniques based on Convolutional Neural Networks (CNNs) are shown to give an accurate estimate of the steady state velocity fields (in all axes), and by extension, the macro-scale permeability. This estimate can be used as-is, or as initial conditions in direct simulation to reach a fully accurate result in a fraction of the compute time. A Gated U-Net Convolutional Neural Network is trained on a datasets of 2D and 3D porous media generated by correlated fields, with their steady state velocity fields calculated from direct LBM simulation. Sensitivity analysis indicates that network accuracy is dependent on (1) the tortuosity of the domain, (2) the size of convolution filters, (3) the use of distance maps as input, (4) the use of mass conservation loss functions. Permeability estimation from these predicted fields reaches over 90\% accuracy for 80\% of cases. It is further shown that these velocity fields are error prone when used for solute transport simulation. Using the predicted velocity fields as initial conditions is shown to accelerate direct flow simulation to physically true steady state conditions an order of magnitude less compute time. Using Deep Learning predictions (or potentially any other approximation method) to accelerate flow simulation to steady state in complex pore structures shows promise as a technique push the boundaries fluid flow modelling.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the huge consumption of computational resources and high time cost when conducting fluid - flow simulations in porous media. Specifically, directly simulating fluid - flow in porous media requires a large amount of computational resources to be completed within a reasonable time. Therefore, the author proposes a method that combines deep - learning prediction (fast but with limited accuracy) and direct fluid simulation (slow but with high accuracy) to significantly reduce the computation time while maintaining high accuracy. The core of the paper lies in using convolutional neural networks (CNNs), especially the gated U - Net structure, to predict the steady - state velocity fields (on all axes) in porous media. These predictions can be used as initial conditions for direct simulation, thereby accelerating the achievement of completely accurate results. In addition, the paper also explores the technical potential of using deep - learning prediction to accelerate fluid - flow simulations in complex pore structures to the steady - state, as well as the application prospects of this method in improving the boundaries of fluid - flow modeling. ### Main Objectives 1. **Improve Computational Efficiency**: Reduce the time required for direct simulation by using deep - learning techniques to predict the velocity fields of fluids. 2. **Maintain High Accuracy**: Ensure that the predicted velocity fields can be used for further simulations, such as solute - transport simulations, while maintaining high accuracy. 3. **Explore Influencing Factors**: Analyze which factors affect network accuracy, including the tortuosity of porous media, the size of convolutional filters, the use of distance maps, and the use of mass - conservation loss functions. ### Methods - **Data Generation**: Generate 2D and 3D porous - media datasets, and use the correlation - field method to generate porous media with different geometric complexities. - **Fluid Simulation**: Use the lattice Boltzmann method (LBM) for fluid simulation to obtain steady - state velocity fields. - **Network Training**: Train the gated U - Net convolutional neural network, with the binary images of porous media or Euclidean - distance - transform maps as inputs and the predicted velocity fields as outputs. - **Loss Function**: Use the mean - squared error (L2 loss) and mass - conservation loss function to optimize network performance. - **Evaluation Metrics**: Evaluate the accuracy of predictions through permeability error and fine - scale - velocity - field error. ### Results - **Permeability Prediction**: The predicted velocity fields are used for permeability estimation, achieving an accuracy of over 90%. - **Fine - Scale Velocity Field**: There are errors in the predicted velocity fields when they are used for solute - transport simulations, but as initial conditions, they can significantly accelerate direct fluid simulations to the steady - state conditions, reducing the computation time by an order of magnitude. ### Conclusions This research demonstrates the potential of using deep - learning techniques to accelerate fluid - flow simulations in porous media, especially in complex pore structures. By combining deep - learning prediction and direct simulation, computational efficiency can be significantly improved while maintaining high accuracy. This method provides new ideas and tools for fluid - flow modeling.

ML-LBM: Machine Learning Aided Flow Simulation in Porous Media

ML-LBM: Predicting and Accelerating Steady State Flow Simulation in Porous Media with Convolutional Neural Networks

Learning a General Model of Single Phase Flow in Complex 3D Porous Media

Fast direct flow simulation in porous media by coupling with pore network and Laplace models

A physics-constrained deep learning model for simulating multiphase flow in 3D heterogeneous porous media

Computationally Efficient Multiscale Neural Networks Applied to Fluid Flow in Complex 3D Porous Media

Flow prediction of heterogeneous nanoporous media based on physical information neural network

Efficient Deep Learning Techniques for Multiphase Flow Simulation in Heterogeneous Porous Media

Physics Informed Deep Learning for Flow and Transport in Porous Media

A physics-informed and hierarchically regularized data-driven model for predicting fluid flow through porous media

Improving Deep Learning Performance for Predicting Large-Scale Porous-Media Flow through Feature Coarsening

Predicting Effective Diffusivity of Porous Media from Images by Deep Learning

A Gradient-based Deep Neural Network Model for Simulating Multiphase Flow in Porous Media

Permeability Estimation for Deformable Porous Media with Convolutional Neural Network

Non-intrusive, transferable model for coupled turbulent channel-porous media flow based upon neural networks

Hybrid LBM and machine learning algorithms for permeability prediction of porous media: A comparative study

High Performance Simulation Of Complicated Fluid Flow In 3d Fractured Porous Media With Permeable Material Matrix Using Lbm

Improved Lattice Boltzmann Method to Simulate Liquid Flow in Nanoporous Media: Coupling Molecular Dynamics Simulations and Theoretical Model

Lattice Boltzmann simulations for fluid flow through porous media

Neural Network Based Pore Flow Field Prediction in Porous Media Using Super Resolution

Simulation of quasi-static drainage displacement in porous media on pore-scale: Coupling lattice Boltzmann method and pore network model