A physics-informed and hierarchically regularized data-driven model for predicting fluid flow through porous media
Abstract
This paper presents a new deep learning data-driven model for predicting structure dependent pore-fluid velocity fields in rock. The model is based on a Convolutional Auto-Encoder (CAE) artificial neural network capable of learning from image data generated by direct numerical simulations of fluid flow through pore-structures, such as by Lattice Boltzmann or molecular dynamics methods. The main novelty of the model in comparison to previous CAE-based data-driven approaches consists of three parts. The first is a methodology for decomposing the full-domain of the porous media into sub-regions, or “sub-domains”, in order to reduce the overall size of the CAE, batch process the sub-domains in parallel, and enable the CAE to learn local and generalizable nonlinear mappings of pore-fluid velocities. The second consists of embedding the finite difference solutions of the incompressible Navier-Stokes and continuity equations into convolutional layers prior to the CAE in order to provide the CAE with knowledge of fluid dynamics physics (PhyFlow). The third main novelty is that the training of the CAE is regularized with a hierarchical loss function that encourages the learning of fluid flow patterns (in a way similar to ranked modes in principal component analysis), ranking from most to least important. This ismore »
- Authors:
-
- Los Alamos National Lab. (LANL), Los Alamos, NM (United States)
- Publication Date:
- Research Org.:
- Los Alamos National Lab. (LANL), Los Alamos, NM (United States)
- Sponsoring Org.:
- USDOE National Nuclear Security Administration (NNSA)
- OSTI Identifier:
- 1804349
- Alternate Identifier(s):
- OSTI ID: 1815191
- Report Number(s):
- LA-UR-20-27444
Journal ID: ISSN 0021-9991; TRN: US2212271
- Grant/Contract Number:
- 89233218CNA000001; 20190005DR
- Resource Type:
- Accepted Manuscript
- Journal Name:
- Journal of Computational Physics
- Additional Journal Information:
- Journal Volume: 443; Journal ID: ISSN 0021-9991
- Publisher:
- Elsevier
- Country of Publication:
- United States
- Language:
- English
- Subject:
- 71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSICS; Porous Media; Deep Learning; Permeability; Machine learning; CAE; Porous flow; Shale; Lattice Boltzmann; Molecular dynamics
Citation Formats
Wang, Kun, Chen, Yu, Mehana, Mohamed Zakaria Saad, Lubbers, Nicholas Edward, Bennett, Kane, Kang, Qinjun, Viswanathan, Hari S., and Germann, Timothy Clark. A physics-informed and hierarchically regularized data-driven model for predicting fluid flow through porous media. United States: N. p., 2021.
Web. doi:10.1016/j.jcp.2021.110526.
Wang, Kun, Chen, Yu, Mehana, Mohamed Zakaria Saad, Lubbers, Nicholas Edward, Bennett, Kane, Kang, Qinjun, Viswanathan, Hari S., & Germann, Timothy Clark. A physics-informed and hierarchically regularized data-driven model for predicting fluid flow through porous media. United States. https://doi.org/10.1016/j.jcp.2021.110526
Wang, Kun, Chen, Yu, Mehana, Mohamed Zakaria Saad, Lubbers, Nicholas Edward, Bennett, Kane, Kang, Qinjun, Viswanathan, Hari S., and Germann, Timothy Clark. Fri .
"A physics-informed and hierarchically regularized data-driven model for predicting fluid flow through porous media". United States. https://doi.org/10.1016/j.jcp.2021.110526. https://www.osti.gov/servlets/purl/1804349.
@article{osti_1804349,
title = {A physics-informed and hierarchically regularized data-driven model for predicting fluid flow through porous media},
author = {Wang, Kun and Chen, Yu and Mehana, Mohamed Zakaria Saad and Lubbers, Nicholas Edward and Bennett, Kane and Kang, Qinjun and Viswanathan, Hari S. and Germann, Timothy Clark},
abstractNote = {This paper presents a new deep learning data-driven model for predicting structure dependent pore-fluid velocity fields in rock. The model is based on a Convolutional Auto-Encoder (CAE) artificial neural network capable of learning from image data generated by direct numerical simulations of fluid flow through pore-structures, such as by Lattice Boltzmann or molecular dynamics methods. The main novelty of the model in comparison to previous CAE-based data-driven approaches consists of three parts. The first is a methodology for decomposing the full-domain of the porous media into sub-regions, or “sub-domains”, in order to reduce the overall size of the CAE, batch process the sub-domains in parallel, and enable the CAE to learn local and generalizable nonlinear mappings of pore-fluid velocities. The second consists of embedding the finite difference solutions of the incompressible Navier-Stokes and continuity equations into convolutional layers prior to the CAE in order to provide the CAE with knowledge of fluid dynamics physics (PhyFlow). The third main novelty is that the training of the CAE is regularized with a hierarchical loss function that encourages the learning of fluid flow patterns (in a way similar to ranked modes in principal component analysis), ranking from most to least important. This is shown to increase the stability in learning, reduce over-fitting, and promote interpretability of the CAE neural network layers (HierCAE). The comprehensive new data-driven model, which we call the PhyFlow-HierCAE model, is shown to exhibit improved accuracy and generalizability of flow field predictions over conventional CAE models, attributable to the embedded physical knowledge and the hierarchical regularization, as well as realize orders of magnitude speed-ups in computation times as a surrogate for the direct numerical simulations. Examples of training and forward predictions on unseen pore-structures are provided and evaluated for data from Lattice Boltzmann and molecular dynamics simulations of pore-fluid flow. The model is shown to be a fast and accurate emulator (or “surrogate”) for predicting effective permeability of unseen pore-structures based on learning from relatively small direct numerical simulation datasets.},
doi = {10.1016/j.jcp.2021.110526},
journal = {Journal of Computational Physics},
number = ,
volume = 443,
place = {United States},
year = {Fri Oct 15 00:00:00 EDT 2021},
month = {Fri Oct 15 00:00:00 EDT 2021}
}
Works referenced in this record:
On the convergence of discrete approximations to the Navier-Stokes equations
journal, May 1969
- Chorin, Alexandre Joel
- Mathematics of Computation, Vol. 23, Issue 106
Modelling two-phase flow in porous media at the pore scale using the volume-of-fluid method
journal, July 2012
- Raeini, Ali Q.; Blunt, Martin J.; Bijeljic, Branko
- Journal of Computational Physics, Vol. 231, Issue 17
Molecular dynamics of flow in micropores
journal, August 1987
- Bitsanis, I.; Magda, J. J.; Tirrell, M.
- The Journal of Chemical Physics, Vol. 87, Issue 3
Finite-Difference Approximation for Fluid-Flow Simulation and Calculation of Permeability in Porous Media
journal, June 2012
- Shabro, Vahid; Torres-Verdín, Carlos; Javadpour, Farzam
- Transport in Porous Media, Vol. 94, Issue 3
Transport of Multicomponent Hydrocarbon Mixtures in Shale Organic Matter by Molecular Simulations
journal, September 2015
- Collell, Julien; Galliero, Guillaume; Vermorel, Romain
- The Journal of Physical Chemistry C, Vol. 119, Issue 39
Computations of Absolute Permeability on Micro-CT Images
journal, December 2012
- Mostaghimi, Peyman; Blunt, Martin J.; Bijeljic, Branko
- Mathematical Geosciences, Vol. 45, Issue 1
Deep learning
journal, May 2015
- LeCun, Yann; Bengio, Yoshua; Hinton, Geoffrey
- Nature, Vol. 521, Issue 7553
Stochastic generation of explicit pore structures by thresholding Gaussian random fields
journal, November 2014
- Hyman, Jeffrey D.; Winter, C. Larrabee
- Journal of Computational Physics, Vol. 277
Enhancing images of shale formations by a hybrid stochastic and deep learning algorithm
journal, October 2019
- Kamrava, Serveh; Tahmasebi, Pejman; Sahimi, Muhammad
- Neural Networks, Vol. 118
The neural particle method – An updated Lagrangian physics informed neural network for computational fluid dynamics
journal, August 2020
- Wessels, Henning; Weißenfels, Christian; Wriggers, Peter
- Computer Methods in Applied Mechanics and Engineering, Vol. 368
Direct pore-level modeling of incompressible fluid flow in porous media
journal, September 2010
- Ovaysi, Saeed; Piri, Mohammad
- Journal of Computational Physics, Vol. 229, Issue 19
Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations
journal, February 2019
- Raissi, M.; Perdikaris, P.; Karniadakis, G. E.
- Journal of Computational Physics, Vol. 378
Deconvolution and Checkerboard Artifacts
journal, October 2016
- Odena, Augustus; Dumoulin, Vincent; Olah, Chris
- Distill, Vol. 1, Issue 10
A quasi-continuum hydrodynamic model for slit shaped nanochannel flow
journal, August 2013
- Bhadauria, Ravi; Aluru, N. R.
- The Journal of Chemical Physics, Vol. 139, Issue 7
Machine learning materials physics: Integrable deep neural networks enable scale bridging by learning free energy functions
journal, August 2019
- Teichert, G. H.; Natarajan, A. R.; Van der Ven, A.
- Computer Methods in Applied Mechanics and Engineering, Vol. 353
Machine learning framework for analysis of transport through complex networks in porous, granular media: A focus on permeability
journal, August 2016
- van der Linden, Joost H.; Narsilio, Guillermo A.; Tordesillas, Antoinette
- Physical Review E, Vol. 94, Issue 2
Hybrid pore-network and lattice-Boltzmann permeability modelling accelerated by machine learning
journal, April 2019
- Rabbani, Arash; Babaei, Masoud
- Advances in Water Resources, Vol. 126
Fast Parallel Algorithms for Short-Range Molecular Dynamics
journal, March 1995
- Plimpton, Steve
- Journal of Computational Physics, Vol. 117, Issue 1
Nanoscale Two-Phase Flow of Methane and Water in Shale Inorganic Matrix
journal, October 2018
- Liu, Bing; Qi, Chao; Zhao, Xiangbin
- The Journal of Physical Chemistry C, Vol. 122, Issue 46
Nonequilibrium molecular dynamics simulations of nanoconfined fluids at solid-liquid interfaces
journal, June 2017
- Morciano, M.; Fasano, M.; Nold, A.
- The Journal of Chemical Physics, Vol. 146, Issue 24
Numerical Calculation of Time-Dependent Viscous Incompressible Flow of Fluid with Free Surface
journal, January 1965
- Harlow, Francis H.; Welch, J. Eddie
- Physics of Fluids, Vol. 8, Issue 12
The Nose–Hoover thermostat
journal, October 1985
- Evans, D. J.; Holian, B. L.
- The Journal of Chemical Physics, Vol. 83, Issue 8
Comprehensive comparison of pore-scale models for multiphase flow in porous media
journal, June 2019
- Zhao, Benzhong; MacMinn, Christopher W.; Primkulov, Bauyrzhan K.
- Proceedings of the National Academy of Sciences, Vol. 116, Issue 28
A reality check on the shale revolution
journal, February 2013
- Hughes, J. David
- Nature, Vol. 494, Issue 7437
Methods of estimating shale gas resources – Comparison, evaluation and implications
journal, September 2013
- McGlade, Christophe; Speirs, Jamie; Sorrell, Steve
- Energy, Vol. 59
Transferable Potentials for Phase Equilibria. 1. United-Atom Description of n-Alkanes
March 1998
- Martin, Marcus G.; Siepmann, J. Ilja
Impact of Nanoporosity on Hydrocarbon Transport in Shales’ Organic Matter
journal, January 2018
- Obliger, Amaël; Ulm, Franz-Josef; Pellenq, Roland
- Nano Letters, Vol. 18, Issue 2
Lattice Boltzmann simulations of liquid CO2 displacing water in a 2D heterogeneous micromodel at reservoir pressure conditions
journal, May 2018
- Chen, Yu; Li, Yaofa; Valocchi, Albert J.
- Journal of Contaminant Hydrology, Vol. 212
Neural network studies. 1. Comparison of overfitting and overtraining
journal, September 1995
- Tetko, Igor V.; Livingstone, David J.; Luik, Alexander I.
- Journal of Chemical Information and Modeling, Vol. 35, Issue 5
Reconstruction of three-dimensional porous media using generative adversarial neural networks
journal, October 2017
- Mosser, Lukas; Dubrule, Olivier; Blunt, Martin J.
- Physical Review E, Vol. 96, Issue 4
Hidden fluid mechanics: Learning velocity and pressure fields from flow visualizations
journal, January 2020
- Raissi, Maziar; Yazdani, Alireza; Karniadakis, George Em
- Science, Vol. 367, Issue 6481
Gradient-based learning applied to document recognition
journal, January 1998
- Lecun, Y.; Bottou, L.; Bengio, Y.
- Proceedings of the IEEE, Vol. 86, Issue 11
Multiphase lattice Boltzmann simulations for porous media applications: A review
journal, December 2015
- Liu, Haihu; Kang, Qinjun; Leonardi, Christopher R.
- Computational Geosciences, Vol. 20, Issue 4
Deep learning for universal linear embeddings of nonlinear dynamics
journal, November 2018
- Lusch, Bethany; Kutz, J. Nathan; Brunton, Steven L.
- Nature Communications, Vol. 9, Issue 1
Modeling and scale-bridging using machine learning: nanoconfinement effects in porous media
journal, August 2020
- Lubbers, Nicholas; Agarwal, Animesh; Chen, Yu
- Scientific Reports, Vol. 10, Issue 1
Computational engineering and science methodologies for modeling and simulation of subsurface applications
journal, August 2002
- Wheeler, Mary F.; Peszyńska, Małgorzata
- Advances in Water Resources, Vol. 25, Issue 8-12
A simplified MAC technique for incompressible fluid flow calculations
journal, October 1970
- Amsden, Anthony A.; Harlow, Francis H.
- Journal of Computational Physics, Vol. 6, Issue 2
Lattice Boltzmann Method for Fluid Flows
journal, January 1998
- Chen, Shiyi; Doolen, Gary D.
- Annual Review of Fluid Mechanics, Vol. 30, Issue 1
PoreFlow-Net: A 3D convolutional neural network to predict fluid flow through porous media
journal, April 2020
- Santos, Javier E.; Xu, Duo; Jo, Honggeun
- Advances in Water Resources, Vol. 138
Reducing the Dimensionality of Data with Neural Networks
journal, July 2006
- Hinton, G. E.
- Science, Vol. 313, Issue 5786
Development of a Hybrid Process and System Model for the Assessment of Wellbore Leakage at a Geologic CO 2 Sequestration Site
journal, October 2008
- Viswanathan, Hari S.; Pawar, Rajesh J.; Stauffer, Philip H.
- Environmental Science & Technology, Vol. 42, Issue 19
Numerical calculation of multiphase fluid flow
journal, January 1975
- Harlow, Francis H.; Amsden, Anthony A.
- Journal of Computational Physics, Vol. 17, Issue 1
Training with Noise is Equivalent to Tikhonov Regularization
journal, January 1995
- Bishop, Chris M.
- Neural Computation, Vol. 7, Issue 1
CFD Python: the 12 steps to Navier-Stokes equations
journal, November 2018
- Barba, Lorena; Forsyth, Gilbert
- Journal of Open Source Education, Vol. 1, Issue 9
Data-driven computational mechanics
journal, June 2016
- Kirchdoerfer, T.; Ortiz, M.
- Computer Methods in Applied Mechanics and Engineering, Vol. 304
Inertial Effects During the Process of Supercritical CO 2 Displacing Brine in a Sandstone: Lattice Boltzmann Simulations Based on the Continuum‐Surface‐Force and Geometrical Wetting Models
journal, December 2019
- Chen, Yu; Valocchi, Albert J.; Kang, Qinjun
- Water Resources Research, Vol. 55, Issue 12
The Effects of Adding Noise During Backpropagation Training on a Generalization Performance
journal, April 1996
- An, Guozhong
- Neural Computation, Vol. 8, Issue 3