IP Library Granted Patent US 12,657,358
Granted Patent B2
US 12,657,358 · App. 17/708,319 · Granted Jun 16, 2026

Neural network predictions of fluid flow in porous media

Inventors: Rodrigo Neumann Barros Ferreira (Rio de Janeiro, BR); Jaione Tirapu Azpiroz (Rio de Janeiro, BR); Ronaldo Giro (São Paulo, BR); Mathias B Steiner (Rio de Janeiro, BR)
Assignee: International Business Machines Corporation
G06F30/28G06F30/27G06N3/08G06F2113/08
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,657,358
App. No.
17/708,319
Granted
Jun 16, 2026
Kind
B2
Abstract

Systems and methods for predicting fluid flow of porous media are provided. In implementations, a method includes: accessing, by a computing device, a capillary network representation of a porous medium sample; generating, by the computing device, a set of simplified network representations from the capillary network representation; determining, by the computing device, simulated fluid flow properties of each of the simplified network representations using a simulator to perform fluid flow simulations; and training, by the computing device, a neural network (NN) model utilizing the set of simplified network representations as inputs and the simulated fluid flow properties as model targets, thereby generating a trained NN model for predicting fluid flow properties of the porous medium.

Claims (49)

1 . A method, comprising:

accessing, by a computing device, a capillary network representation of a porous medium sample;

generating, by the computing device, a set of simplified network representations from the capillary network representation;

determining, by the computing device, simulated fluid flow properties of each of the simplified network representations using a simulator to perform fluid flow simulations; and

training, by the computing device, a convolutional neural network (NN) model utilizing the set of simplified network representations as inputs and the simulated fluid flow properties as model targets, thereby generating a trained convolutional NN model for predicting fluid flow properties of the porous medium sample, wherein the trained convolutional NN model comprises a model that performs convolutions with link vectors to generate convolutional layers followed by a pooling operation to generate pooling layers.

2 . The method of claim 1 , wherein the capillary network representation is a three-dimensional (3D) network of interconnected capillaries representing interconnecting pore structures of a 3D micro-CT scanner model of a porous medium sample, and further comprising obtaining, by the computing device, the 3D micro-CT scanner model from a remote micro-CT scanner via a network connection.

3 . The method of claim 1 , wherein the capillary network representation is a three-dimensional (3D) network of interconnected capillaries representing interconnecting pore structures of a 3D micro-CT scanner model of a porous medium sample, and further comprising generating, by the computing device, the 3D micro-CT scanner model by imaging the porous medium sample with micro-CT.

4 . The method of claim 1 , wherein the capillary network representation is a three-dimensional (3D) network of interconnected capillaries representing interconnecting pore structures of a 3D micro-CT scanner model of a porous medium sample, and further comprising generating, by the computing device, the capillary network representation from the 3D micro-CT scanner model.

5 . The method of claim 1 , further comprising:

determining, by the computing device, predicted fluid flow properties of the porous medium sample by inputting the capillary network representation into the trained convolutional NN model;

aggregating, by the computing device, the simulated fluid flow properties of each of the simplified network representations to obtain simulated fluid flow property results for the porous medium sample; and

validating, by the computing device, the trained convolutional NN model by comparing the simulated fluid flow property results for the porous medium sample to the predicted fluid flow properties of the porous medium sample, wherein variation of the simulated flow properties from the predicted flow properties exceeding a predetermined threshold value indicates an error in the trained convolutional NN model.

6 . The method of claim 1 , further comprising: inputting, by the computing device, a new capillary network representation of a new porous medium sample into the trained convolutional NN model, thereby generating predicted fluid flow properties of the new porous medium sample, wherein the new capillary network representation comprises a single capillary network representation of the complete new porous medium sample, or a set of simplified capillary network representations of the new porous medium sample.

7 . The method of claim 1 , wherein the simulated fluid flow properties and the predicted fluid flow properties comprise one or more fluid flow properties selected from the group consisting of: porosity, permeability, flow rates, and residual fluid saturation.

8 . The method of claim 1 , wherein the convolutional neural network comprises a graph convolutional neural network (GCNN) and the computing device includes software provided as a service in a cloud environment.

9 . A computer program product comprising one or more computer non-transitory readable storage media having program instructions collectively stored on the one or more computer readable storage media, the program instructions executable to:

access a capillary network representation, wherein the capillary network representation is a three-dimensional (3D) network of interconnected capillaries representing interconnecting pore structures of a 3D micro-CT scanner model of a porous medium sample;

generate a set of simplified network representations from the capillary network representation, wherein each simplified network representation in the set of simplified network representations comprises a two dimensional (2D) or 3D network of interconnected capillary structures that represents a subset of the interconnected capillary structures of the capillary network representation;

perform fluid flow simulations, using a physics-based simulator, to provide simulated fluid flow properties of each of the simplified network representations; and

train a convolutional neural network (NN) model utilizing the set of simplified network representations as inputs and the simulated fluid flow properties as model targets, thereby generating a convolutional trained NN model for predicting fluid flow properties, wherein the trained convolutional NN model comprises a convolutional neural network model that performs convolutions with link vectors to generate convolutional layers followed by a pooling operation to generate pooling layers.

10 . The computer program product of claim 9 , wherein the program instructions are further executable to obtain the 3D micro-CT scanner model from a remote micro computed tomography (micro-CT) scanner via a network connection.

11 . The computer program product of claim 9 , wherein the program instructions are further executable to generate the 3D micro-CT scanner model by imaging the porous medium sample with micro-CT.

12 . The computer program product of claim 9 , wherein the program instructions are further executable to generate the capillary network representation from the 3D micro-CT scanner model.

13 . The computer program product of claim 9 , wherein the program instructions are further executable to:

determine predicted fluid flow properties of the porous medium sample by inputting the capillary network representation into the trained convolutional NN model;

aggregate the simulated fluid flow properties of each of the simplified network

representations to obtain simulated fluid flow property results for the porous medium sample; and

validate the trained convolutional NN model by comparing the simulated fluid flow property results for the porous medium sample to the predicted fluid flow properties of the porous medium sample.

14 . The computer program product of claim 9 , wherein the program instructions are further executable to input a new capillary network representation of a new porous medium sample into the trained convolutional NN model, thereby generating predicted fluid flow properties of the new porous medium sample, wherein the new capillary network representation comprising a single capillary network representation of the complete new porous medium sample, or a set of simplified capillary network representations of the new porous medium sample.

15 . The computer program product of claim 9 , wherein the set of simplified network representations have a number of interconnected capillary structures that is smaller than a number of interconnected capillary structures of the capillary network representation by orders of magnitude.

16 . A system comprising:

a processor, a computer readable memory, one or more computer readable storage media, and program instructions collectively stored on the one or more computer readable storage media, the program instructions executable to:

access a capillary network representation, wherein the capillary network representation is a three-dimensional (3D) network of interconnected capillaries representing interconnecting pore structures of a 3D micro-CT scanner model of a porous medium sample;

generate a set of simplified network representations from the capillary network

representation, wherein each simplified network representation in the set of simplified network representations comprises a two dimensional (2D) or 3D network of interconnected capillary structures that represents a subset of the interconnected capillary structures of the capillary network representation;

determine simulated fluid flow properties of each of the simplified network representations using a physics-based simulator to perform fluid flow simulations;

generate one or more convolutional layers by performing one or more convolutions on vectors characterizing links of the set of simplified network representations;

train a convolutional neural network (NN) model utilizing the set of simplified network representations as inputs and the simulated fluid flow properties as model targets, thereby generating a trained convolutional NN model for predicting fluid flow properties, wherein the trained convolutional NN model performs convolutions with link vectors to generate convolutional layers followed by a pooling operation to generate pooling layers; and

input a new capillary network representation of a new porous medium sample into the trained convolutional NN model, thereby generating predicted fluid flow properties of the new porous medium sample.

17 . The system of claim 16 , wherein the program instructions are further executable to display, by the computing device, the predicted fluid flow properties to a user via a user interface.

18 . The system of claim 16 , wherein the program instructions are further executable to:

generate the 3D micro-CT scanner model by imaging the porous medium sample with micro-CT; and

generate the capillary network representation from the 3D micro-CT scanner model.

19 . The system of claim 16 , wherein the program instructions are further executable to:

determine predicted fluid flow properties of the porous medium sample by inputting the capillary network representation into the trained convolutional NN model;

aggregate the simulated fluid flow properties of each of the simplified network

representations to obtain simulated fluid flow property results for the porous medium sample; and

validate the trained convolutional NN model by comparing the simulated fluid flow property results for the porous medium sample to the predicted fluid flow properties of the porous medium sample.

20 . The system of claim 16 , wherein the set of simplified network representations have a number of interconnected capillary structures that is smaller than a number of interconnected capillary structures of the capillary network representation by orders of magnitude.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Mar 30, 2022
From: NEUMANN BARROS FERREIRA, RODRIGO; TIRAPU AZPIROZ, JAIONE; GIRO, RONALDO; STEINER, MATHIAS B
To: INTERNATIONAL BUSINESS MACHINES CORPORATION
Reel/Frame 059441/0707 →
Continuity (1)
Related Publication 20230315956A1 · Oct 5, 2023
References Cited (51)
US 10648292B2 · Bryant et al. · 2020 [cited by applicant]
US 10830713B2 · Zhang · 2020 [cited by applicant]
US 11704457B2 · Riasi · 2023 [cited by examiner]
US 20050256643A1 · Boitnott · 2005 [cited by examiner]
US 20130096900A1 · Usadi et al. · 2013 [cited by applicant]
US 20140019053A1 · de Prisco · 2014 [cited by applicant]
US 20180252076A1 · Bryant et al. · 2018 [cited by applicant]
US 20180253514A1 · Bryant · 2018 [cited by examiner]
US 20200320386A1 · Filippov et al. · 2020 [cited by applicant]
CN 103699664 · 2014 [cited by applicant]
CN 105160058 · 2015 [cited by applicant]
CN 103822865 · 2016 [cited by applicant]
CN 107018011 · 2017 [cited by applicant]
CN 108173903 · 2018 [cited by applicant]
CN 108988346 · 2018 [cited by applicant]
CN 109120452 · 2019 [cited by applicant]
CN 109902846 · 2019 [cited by applicant]
CN 110365519 · 2019 [cited by applicant]
CN 110851655 · 2020 [cited by applicant]
CN 111460596 · 2020 [cited by applicant]
CN 111950192 · 2020 [cited by applicant]
CN 111950193 · 2020 [cited by applicant]
CN 112187658 · 2021 [cited by applicant]
CN 112329358 · 2021 [cited by applicant]
CN 112801499 · 2021 [cited by applicant]
CN 112967238 · 2021 [cited by applicant]
CN 113139989 · 2021 [cited by applicant]
CN 113947026 · 2022 [cited by applicant]
CN 119013670A · 2024 [cited by applicant]
WO 2013009512 · 2013 [cited by applicant]
WO 2021019361 · 2021 [cited by applicant]
WO 2023185858A1 · 2023 [cited by applicant]
https://www.merriam-webster.com/dictionary/convolution, 1 page, Aug. 15, 2024 (Year: 2024). [cited by examiner]
https://www.arm.com/glossary/convolutional-neural-network, 1 page, What Is a Convolutional Neural Network (Year: 2025). [cited by examiner]
https://developer.nvidia.com/discover/convolutional-neural-network, 5 pages (Year: 2025). [cited by examiner]
International Search Report and Written Opinion for PCT/CN2023/084412 dated Jun. 23, 2023; 8 Pages. [cited by applicant]
Mell et al., “The NIST Definition of Cloud Computing”, NIST, Special Publication 800-145, Sep. 2011, 7 pages. [cited by applicant]
Garttner et al., “Estimating permeability of 3D micro-CT images by physics-informed CNNs based on DNS,” https://arxiv.org/abs/2109.01818, Sep. 4, 2021, 23 pages. [cited by applicant]
Neumann et al., “High accuracy capillary network representation in digital rock reveals permeability scaling functions.” Scientific Reports, 2021, 8 pages. [cited by applicant]
Miao et al., “A new way to parameterize hydraulic conductances of pore elements: A step towards creating pore-networks without pore shape simplifications,” Advances in Water Resources, Nov. 28, 2016, 29 pages. [cited by applicant]
Santos et al., “PoreFlow-Net: A 3D convolutional neural network to predict fluid flow through porous media,” Advances in Water Resources, Nov. 20, 2019, 12 pages. [cited by applicant]
Rabbani et al., “Estimation of carbonates permeability using pore network parameters extracted from thin section images and comparison with experimental data,” J. Natural Gas Science & Engineering, Oct. 25, 2016, 30 pag… [cited by applicant]
Jia et al., “A Pore-Network-Modeling Approach to Predict Petrophysical Properties of Diatomaceous Reservoir Rock,” SPE Res Eval & Eng 10 (06), Dec. 20, 2007, 12 pages. [cited by applicant]
Van der Linden, “Machine learning framework for analysis of transport through complex networks in porous, granular media: A focus on permeability”, American Physical Society, 2016, 16 pages. [cited by applicant]
Wu, “Predicting Effective Diffusivity of Porous Media from Images by Deep Learning”, Scientific Reports, 2019, 12 pages. [cited by applicant]
Sudakov, “Driving Digital Rock towards Machine Learning: predicting permeability with Gradient Boosting and Deep Neural Networks”, Mar. 14, 2018, 22 pages. [cited by applicant]
Liu, “Using machine learning to predict permeability in porous materials”, 2019, 13 pages. [cited by applicant]
Zhang et al., “Graph convolutional networks: a comprehensive review”, Computational Social Networks, 2019, 23 pages. [cited by applicant]
Jepsen, “How to do Deep Learning on Graphs with Graph Convolutional Networks”, https://towardsdatascience.com/how-to-do-deep-learning-on-graphs-with-graph-convolutional-networks-7d2250723780, Sep. 18, 2018, 13 pages. [cited by applicant]
Rabbani et al., “Hybrid pore-network and Lattice-Boltzmann permeability modelling accelerated by machine learning”, Advances in Water Resources, Dec. 30, 2018, 41 pages. [cited by applicant]
Xie, et al., “Crystal Graph Convolutional Neural Networks for an Accurate and Interpretable Prediction of Material Properties”, Phys. Rev. Lett., Apr. 6, 2018, 7 pages. [cited by applicant]