From Computational Fluid Dynamics to Structure Interpretation via Neural Networks: An Application to Flow and Transport in Porous Media

Industrial & Engineering Chemistry Research, Journal Year: 2022, Volume and Issue: 61(24), P. 8530 - 8541

Published: March 25, 2022

The modeling of flow and transport in porous media is the utmost importance many chemical engineering applications, including catalytic reactors, batteries, CO2 storage. aim this study to test use fully connected (FCNN) convolutional neural networks (CNN) for prediction crucial properties systems: permeability filtration rate. data-driven models are trained on a dataset computational fluid dynamics (CFD) simulations. To end, geometries created silico by discrete element method, rigorous setup CFD simulations presented. have as input both geometrical operating conditions features so that they could find application multiscale modeling, optimization problems, in-line control. average error lower than 2.5%, rate 5% all models. These results achieved with at least ∼100

Language: Английский

---

Data-driven methods for flow and transport in porous media: A review

International Journal of Heat and Mass Transfer, Journal Year: 2024, Volume and Issue: 235, P. 126149 - 126149

Published: Sept. 7, 2024

Language: Английский

---

Materials processing model-driven discovery framework for porous materials using machine learning and genetic algorithm: A focus on optimization of permeability and filtration efficiency

Chemical Engineering Journal, Journal Year: 2022, Volume and Issue: 453, P. 139540 - 139540

Published: Oct. 3, 2022

Language: Английский

---

Estimating permeability of 3D micro-CT images by physics-informed CNNs based on DNS

Computational Geosciences, Journal Year: 2023, Volume and Issue: 27(2), P. 245 - 262

Published: Jan. 31, 2023

Abstract In recent years, convolutional neural networks (CNNs) have experienced an increasing interest in their ability to perform a fast approximation of effective hydrodynamic parameters porous media research and applications. This paper presents novel methodology for permeability prediction from micro-CT scans geological rock samples. The training data set CNNs dedicated consists labels that are typically generated by classical lattice Boltzmann methods (LBM) simulate the flow through pore space segmented image data. We instead direct numerical simulation (DNS) solving stationary Stokes equation efficient distributed-parallel manner. As such, we circumvent convergence issues LBM frequently observed on complex geometries, therefore, improve generality accuracy our set. Using DNS-computed permeabilities, physics-informed CNN (PhyCNN) is trained additionally providing tailored characteristic quantity space. More precisely, exploiting connection problems graph representation space, additional information about confined structures provided network terms maximum value, which key innovative component workflow. robustness this approach reflected very high accuracy, variety sandstone samples archetypal formations.

Language: Английский

---

A computationally efficient modeling of flow in complex porous media by coupling multiscale digital rock physics and deep learning: Improving the tradeoff between resolution and field-of-view

Advances in Water Resources, Journal Year: 2024, Volume and Issue: 188, P. 104695 - 104695

Published: April 9, 2024

Digital rock physics is at the forefront of characterizing porous media, leveraging advanced tomographic imaging and numerical simulations to extract key properties like permeability. However, fully capturing heterogeneity natural rocks necessitates increasingly larger sample volumes, presenting a significant challenge. Direct these scales become either prohibitively expensive or computationally unfeasible due limitations in resolution field view (FOV). This issue particularly pronounced carbonate rocks, known for their complex, multiscale pore structures, which exacerbate resolution-FOV tradeoff. To address this, we introduce machine learning strategy that merges data from various resolutions with 3D convolutional neural network (CNN) model. approach innovative its ability identify cross-scale correlations, thereby enabling estimation transport volumes—properties are difficult simulate directly—using trainable proxies. The integration deep allows accurate permeability predictions beyond those feasible traditional direct simulation methods. By employing transfer across different during training phase, our model achieves robust performance, an R² exceeding 0.96 when evaluated on diverse lower-resolution domains FOVs. Notably, this method significantly enhances computational efficiency, reducing time by orders magnitude. Originally developed intricate structures shows promise application wide range offering viable solution longstanding tradeoff between FOV digital physics.

Language: Английский

---

A universal structure of neural network for predicting heat, flow and mass transport in various three-dimensional porous media

International Journal of Heat and Mass Transfer, Journal Year: 2025, Volume and Issue: 241, P. 126688 - 126688

Published: Jan. 13, 2025

Language: Английский

---

Intensification of catalytic reactors: A synergic effort of Multiscale Modeling, Machine Learning and Additive Manufacturing

Chemical Engineering and Processing - Process Intensification, Journal Year: 2022, Volume and Issue: 181, P. 109148 - 109148

Published: Sept. 21, 2022

Language: Английский

---

Neural network–based pore flow field prediction in porous media using super resolution

Physical Review Fluids, Journal Year: 2022, Volume and Issue: 7(7)

Published: July 12, 2022

Predicting the pore flow velocity directly from sub-sampled structure is an ill-conditioned problem. Inspired by multi-grid methods for solving systems of linear equations, we use fields simulated on coarse meshes to remedy such ill-conditioning. This leads a super-resolution-assisted geometry-to-velocity mapping porous media.

Language: Английский

---

Identifying the dominant transport mechanism in single nanoscale pores and 3D nanoporous media

Fundamental Research, Journal Year: 2022, Volume and Issue: 3(3), P. 409 - 421

Published: Jan. 5, 2022

Gas transport mechanisms can be categorized into viscous flow and mass diffusion, both of which may coexist in a porous media with multiscale pore sizes. To determine the dominant mechanism its contribution to gas capacity, diffusion processes are analyzed single nanoscale pores via theoretical method, simulated 3D nanoporous pore-scale lattice Boltzmann methods. The apparent permeability from diffusivity estimated. A dimensionless parameter, i.e., diffusion-flow ratio, is proposed evaluate mechanism, function permeability, diffusivity, bulk dynamic viscosity, working pressure. results show that increases by approximately two orders magnitude when average Knudsen number (Knavg) or (Kn) 0.1 10. Under same conditions, increment only one order magnitude. When Kn < 0.01, has lower bound (i.e., absolute permeability). > 10, an upper diffusivity). for 0.01 100, where maximum ratio less than one. In media, relies heavily on Knavg structural parameters. For throat diameter 3 nm, = 0.2 critical point, above dominant; otherwise, dominant. As 3.4, overwhelming, reaching ∼4.

Language: Английский

---

A 3D reconstruction method of porous media based on improved WGAN-GP

Computers & Geosciences, Journal Year: 2022, Volume and Issue: 165, P. 105151 - 105151

Published: May 22, 2022

Language: Английский

---