Pith. sign in

REVIEW 5 major objections 5 minor 22 references

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

T0 review · 5 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A graph neural network trained on micro-CT images can predict how gas and liquid move through rock pores.

desk verdict Promising proof-of-concept for learning pore-scale multiphase flow from micro-CT data, but the generalization claim is untested and the evaluation lacks baselines. read the letter →

arxiv 2411.14192 v1 pith:L63OQPX4 submitted 2024-11-21 physics.flu-dyn cs.LG

Yuxuan Gu , Catherine Spurin , Gege Wen This is my paper · ORCID
classification physics.flu-dyncs.LG
keywords graphneuralnetworkpore-scalemultiphaseflowmicro-CTimagingporousmediaautoregressivepredictionintermittentmeshnetworksCO2storage
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper sets out to show that a graph neural network can learn the pore-scale physics of multiphase flow directly from real micro-CT experimental images, without solving the Navier-Stokes equations. Its proposed model, LSE-MGN, represents the pore space as a graph and autoregressively predicts whether each pore voxel holds gas or liquid at the next time step. The authors report that the model captures both bubble emergence and minor oscillations in a carbonate rock sample, with mean surface-area error around 8.8 to 9.9 percent on held-out frames. If this holds, it would offer a fast, accurate route to modelling flows relevant to CO2 storage, hydrogen storage, and fuel cells.

What carries the argument

The load-bearing mechanism is a graph representation of the segmented pore space in which each node is a pore voxel with features for gas/liquid state, interface status, coordinates, and local gas fraction, and edges connect nodes within a radius of $\sqrt{3}$ voxel units. The LSE-MGN architecture uses an encoder-processor-decoder layout, but replaces single-scale message passing with a long-short-edge scheme: all edges propagate in fine processors, then only 'long' edges (length $\sqrt{3}$) propagate in coarse processors, letting pivotal nodes exchange information across distant pores, then all edges propagate again. This expands each node's receptive field while keeping the coarse graph under 30 percent of the original edges. The training objective is the soft Dice BCE loss, which balances per-voxel accuracy against the class imbalance between gas and liquid.

What would settle it

Take the trained LSE-MGN and run autoregressive rollout on a different subvolume of the same core, or on a different rock sample with a different pore structure; if the mean surface-area error rises well above the reported 8-10 percent range, the claim that the model learned general pore-scale physics from this experimental dataset would be contradicted.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a graph network trained on only 40 time steps of experimental micro-CT data—20 of an emerging bubble and 20 of minor oscillation—can autoregressively generate future gas/liquid distributions that match the measured pore-scale flow. The model tracks the constantly rearranging flow pathways characteristic of intermittent flow, something the authors say current numerical models cannot capture. The best configuration, F6C6F6 with six fine, six coarse, and six fine message-passing layers, reaches a mean surface-area error of 8.81 percent on a 9-step validation set, and the soft Dice BCE loss achieves 9.88 percent on the 16-step test set. The authors take this as evidence that the model successfully captures the physics from high-resolution experimental data and generalises across the boundary conditions present in the experiment.

Load-bearing premise

The paper assumes that a single hand-picked 50 by 80 by 50 voxel subvolume of one carbonate core, with only 40 training time steps, contains enough variety of pore-scale flow behaviour to teach a model that generalises to other geometries, flow rates, and boundary conditions.

Editorial extensions

If this is right

  • If the method holds, pore-scale flow prediction becomes possible at micro-CT resolution without solving the Navier-Stokes equations, enabling fast screening of injection strategies for CO2 storage.
  • The graph-based representation avoids the grid limitations of convolutional networks and can in principle scale to larger rock volumes through section-based training.
  • Autoregressive rollout from a single initial state means long flow sequences can be generated cheaply, potentially replacing expensive pore-network model calibrations.
  • The same architecture could be retrained on other fluid pairs or rock types once experimental imaging data are available.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper's evidence for 'generalisation across varying boundary conditions' is confined to two flow regimes inside one subvolume; a stronger test would train on several subvolumes or cores and measure transfer performance, which the authors do not report.
  • The surface-area error metric may not detect errors in gas-phase connectivity or trapping; adding topological metrics such as the Euler characteristic would make evaluation stricter and might change the ranking of loss functions.
  • A natural follow-up experiment is to train on multiple intermittent-flow subvolumes and test on a steady-flow subvolume, which would clarify whether the model learns intermittent dynamics specifically or a more general pore-scale flow rule.
  • If the model were instead trained on simulated data and fine-tuned on these 40 experimental frames, the comparison would reveal how much of the learned behaviour comes from the experimental data itself.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper proposes LSE-MGN, a graph neural network that learns pore-scale two-phase (gas/liquid) flow directly from segmented micro-CT experimental images of a carbonate rock. The model represents pore voxels as nodes, uses a long-short-edge multi-scale message-passing architecture, and is trained with a soft Dice-BCE loss to autoregressively predict future gas/liquid distributions. The authors report a mean surface-area error of about 8.8-9.9% on held-out time steps from the same experimental sequence and claim that the model captures pore-scale physics and generalizes across boundary conditions.

Significance. Learning pore-scale multiphase flow from real experimental data rather than from idealized simulations is a worthwhile goal with clear relevance to CO2 storage, hydrogen storage, and fuel-cell applications. The paper introduces a sensible graph representation of the pore space and a multi-scale message-passing scheme that appears to enlarge receptive fields while reducing computational cost. The fact that the model is trained and evaluated on synchrotron micro-CT data, not synthetic data, is a strength. However, the quantitative evidence is currently limited: there is no comparison with any baseline (persistence, prior ML models, or numerical simulation), no error bars, and the generalization claim is not tested outside the single 50x80x50 subvolume and single flow condition. As a result, the paper demonstrates a promising architecture but does not yet substantiate the central claims of learning generalizable pore-scale physics.

major comments (5)
  1. [Section 4 and Appendix C] The dataset split is reported inconsistently. Section 4 states the training set comprises 20 emerging-bubble and 20 oscillation time steps, and the test set includes 8 emerging-bubble and 8 oscillation steps. Table 1 and Appendix C instead state a 20-time-step training set and a 9-time-step test set. Table 2 reports testing error 0.0988 while Table 1 reports 0.0881 for the same configuration; these numbers cannot both correspond to the same split. This inconsistency directly affects the reproducibility of the headline results and must be resolved.
  2. [Section 4, Eq. (7)] No baseline comparison is provided for the mean surface-area error. Because the flow is intermittent but locally quasi-static and the temporal resolution is 2 s, a persistence forecast (predicting the previous frame) may achieve a surface-area error of similar magnitude. Without such a baseline, the reported 8.8-9.9% error does not demonstrate that the model has learned pore-scale physics; it could be largely a near-identity map on the gas/liquid field. The paper should report the surface-area error of a persistence predictor and ideally also a simple voxel-wise majority-class baseline on the same test frames.
  3. [Abstract and Section 1] The claim that the model 'can effectively capture complex fluid behaviors and generalise well across varying boundary conditions' is not tested. All training and test data come from the same 50 x 80 x 50 subvolume of one carbonate core, one fluid pair, and one flow condition; the test set consists of held-out time steps from the same experimental run. To support the generalization claim, the authors should either provide out-of-distribution experiments (different subvolumes, flow rates, or fluid pairs) or substantially temper the claim to 'generalizes to unseen time steps within the same experiment.'
  4. [Table 1 and Section 3] No error bars or repeated-seed experiments are reported. The differences between configurations in Table 1 (e.g., 8.81% vs 9.05% vs 10.10%) are likely within run-to-run variability for a stochastic optimizer, so the model selection F6C6F6 is not statistically justified. The paper should report the mean and standard deviation over at least a few training seeds, and ideally show that the chosen configuration's advantage over F7C4F7 persists.
  5. [Section 4, Eq. (6)-(7)] The surface-area error is a global quantity that can be insensitive to spatial mislocalization of fluid clusters. Two predictions with identical surface area can place gas in completely different pores. The paper should supplement this metric with a voxel-wise metric such as intersection-over-union or accuracy, and ideally report errors separately for the bubble-emergence and oscillation test sets, since these are distinct scenarios.
minor comments (5)
  1. [Section 3, Eq. (4)] The coarse-message-passing equation uses the indicator I(|e'_{ij}| = sqrt(3)), but e'_{ij} is a latent edge embedding, not a distance. The filtering should apply to the original edge length (which is one of the input edge attributes), not to the updated embedding. Please clarify the notation.
  2. [Section 3, Eq. (5)] The symbols N and n in the soft Dice BCE loss are not defined. Presumably N is the total number of nodes and n is the number of gas voxels in the ground truth, but this should be stated explicitly.
  3. [Section 2] The text says micro-CT imaging provides 'nanometer-scale spatial resolution,' but the stated resolution is 2.75 µm (micrometres), not nanometers. This should be corrected.
  4. [Figure 2 caption] The caption refers to 'the testing set with emerging bubbles' and shows time steps t=221 to 229, which is 9 frames, while Section 4 states the test set contains 8 emerging-bubble steps. Please clarify whether the displayed interval is a subset or whether the test set description is inaccurate.
  5. [Algorithm 1] The surface-area metric computed via Marching Cubes depends on the voxel discretization. This is acceptable since the same algorithm is applied to predictions and ground truth, but the units should be stated (e.g., voxel surface area, normalized by the domain size).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: empirical supervised learning with external experimental data; the evaluation metric is not the optimized loss by construction.

full rationale

This is an empirical supervised-learning paper, not a derivation, so most circularity patterns do not apply. The model is trained on micro-CT experimental frames from Spurin et al. [8] and evaluated on held-out frames from the same experiment; the labels are produced by a published segmentation workflow [17]. The training objective is soft Dice BCE (Eq. 5), while the reported metric is surface-area error (Eqs. 6-7); the metric is not the loss function or a direct algebraic transform of it, so the headline result is not forced by construction. The cited GNN architectures [14, 15, 18] are used as architectural inspiration, not as a justification that the empirical result is true. The self-citations to [8] and [17] refer to external experimental data and a segmentation tool, which are inputs to the study rather than conclusions derived from this model. The paper's claim of generalization across boundary conditions is not actually tested—all data come from a single 50x80x50 subvolume of one carbonate core under one flow condition—and the train/test split is described inconsistently, but these are validation and reporting weaknesses, not circularity. No fitted parameter is renamed as a prediction, and no uniqueness theorem or ansatz is imported via self-citation. Score 0.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central result rests on the representativeness of one 50x80x50 subvolume and 40 training frames, the adequacy of voxel-graph local features to encode flow physics, and the validity of surface-area error as a physical fidelity metric. The graph radius, processor counts, and latent dimension are hand-chosen or validation-selected. No new physical entities are introduced.

free parameters (4)
  • Graph edge radius R = sqrt(3) voxels (about 1.732)
    Hand-chosen so each node connects to at most 27 neighbors; determines the long/short edge split that defines LSE-MGN. No sensitivity analysis is reported.
  • Fine/coarse processor counts (n, m) = n=6, m=6 (F6C6F6)
    Selected by lowest validation mean surface-area error among four configurations reported in Table 1.
  • Latent embedding dimension = 128
    Hand-chosen for node and edge encoders; no ablation is reported.
  • SGD learning rate = 0.001
    Standard choice with momentum; no sensitivity analysis is reported.
assumptions (5)
  • domain assumption The segmented micro-CT images from Spurin et al. correctly identify gas, liquid, and rock at each voxel.
    The entire input and ground truth are binary occupancy labels from the segmentation pipeline of reference [17]; any segmentation error propagates directly into training and evaluation.
  • domain assumption The 50x80x50 subvolume and 40 training frames are representative of pore-scale multiphase flow dynamics beyond this specific experiment.
    The paper chooses one intermittent-flow subvolume and does not test on other rocks, flow rates, or fluid systems, yet claims generalizable physics learning.
  • domain assumption A graph with voxel nodes and edges within radius sqrt(3) captures the local physics needed for next-frame prediction.
    Message passing is limited to 27 immediate neighbors; the paper provides no comparison with larger radii or physics-informed features.
  • domain assumption Marching-Cubes surface area is a sufficient proxy for physical fidelity.
    The paper reports only surface-area error and visual checks; it does not report mass conservation, connectivity, or pressure and velocity fields.
  • domain assumption Held-out frames from the same experiment provide a valid generalization test.
    Train and test frames are temporally adjacent in one sequence; this tests interpolation in time, not generalization to new boundary conditions or rock samples.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning Pore-scale Multi-phase Flow from Experimental Data with Graph Neural Network." pith.science (2026). https://pith.science/paper/L63OQPX4

@misc{pith2026241114192,
  author       = {Pith},
  title        = {Pith review of: Learning Pore-scale Multi-phase Flow from Experimental Data with Graph Neural Network},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/L63OQPX4}},
  note         = {Machine review of arXiv:2411.14192}
}
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.

Figures

Figures reproduced from arXiv: 2411.14192 by the authors.

Figure 1
Figure 1. (a) A section in the rock is used to train a GNN model fθ. 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… view at source ↗
Figure 2
Figure 2. Visualisation (see video) of ground truth and autoregressive rollout from [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
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 √ 3 units apart. B More visualisation of the prediction [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Visualisation (see video) of ground truth and autoregressive rollout from [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Visualisation (see video) of ground truth and autoregressive rollout from [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Visualisation (see video) of ground truth and autoregressive rollout from [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 2
Figure 2. Figure 2 [PITH_FULL_IMAGE:figures/full_fig_p010_2.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

22 extracted references · 16 canonical work pages

  1. [1]

    Masson-Delmotte, P

    V . Masson-Delmotte, P. Zhai, H.-O. Pörtner, D. Roberts, J. Skea, P. R. Shukla, A. Pirani, W. Moufouma-Okia, C. Péan, R. Pidcock, S. Connors, J.B.R. Matthews, Y . Chen, X. Zhou, M.I. Gomis, E. Lonnoy, T. Maycock, M. Tignor, and T. Waterfield. Summary for policymakers. In V . Masson-Delmotte, P. Zhai, H.-O. Pörtner, D. Roberts, J. Skea, P. R. Shukla, A. Pi...

    2018

  2. [2]

    Progress in carbon capture technologies

    Tabbi Wilberforce, AG Olabi, Enas Taha Sayed, Khaled Elsaid, and Mohammad Ali Abdelka- reem. Progress in carbon capture technologies. Science of The Total Environment, 761:143203, 2021

    work page 2021

  3. [3]

    Carbon capture and storage: history and the road ahead

    Jinfeng Ma, Lin Li, Haofan Wang, Yi Du, Junjie Ma, Xiaoli Zhang, and Zhenliang Wang. Carbon capture and storage: history and the road ahead. Engineering, 14:33–43, 2022. 5

    work page 2022

  4. [4]

    The role of carbon capture and storage to achieve net-zero energy systems: Trade-offs between economics and the environment

    David Yang Shu, Sarah Deutz, Benedikt Alexander Winter, Nils Baumgärtner, Ludger Leenders, and André Bardow. The role of carbon capture and storage to achieve net-zero energy systems: Trade-offs between economics and the environment. Renewable and Sustainable Energy Reviews, 178:113246, 2023

    work page 2023

  5. [5]

    A review on underground hydrogen storage: Insight into geological sites, influencing factors and future outlook

    Nasiru Salahu Muhammed, Bashirul Haq, Dhafer Al Shehri, Amir Al-Ahmed, Mohammed Miza- nur Rahman, and Ehsan Zaman. A review on underground hydrogen storage: Insight into geological sites, influencing factors and future outlook. Energy Reports, 8:461–499, 2022

    work page 2022

  6. [6]

    A review of cell-scale multiphase flow modeling, including water management, in polymer electrolyte fuel cells

    Martin Andersson, SB Beale, M Espinoza, Z Wu, and W Lehnert. A review of cell-scale multiphase flow modeling, including water management, in polymer electrolyte fuel cells. Applied Energy, 180:757–778, 2016

    work page 2016

  7. [7]

    Pore-scale imaging and modelling

    Martin J Blunt, Branko Bijeljic, Hu Dong, Oussama Gharbi, Stefan Iglauer, Peyman Mostaghimi, Adriana Paluszny, and Christopher Pentland. Pore-scale imaging and modelling. Advances in Water resources, 51:197–216, 2013

    work page 2013

  8. [8]

    Real-time imaging reveals distinct pore-scale dynamics during transient and equilibrium subsurface multiphase flow

    Catherine Spurin, Tom Bultreys, Maja Rücker, Gaetano Garfi, Christian M Schlepütz, Vladimir Novak, Steffen Berg, Martin J Blunt, and Samuel Krevor. Real-time imaging reveals distinct pore-scale dynamics during transient and equilibrium subsurface multiphase flow. Water Resources Research, 56(12):e2020WR028287, 2020

    work page 2020

Show all 22 references
  1. [9]

    Compre- hensive comparison of pore-scale models for multiphase flow in porous media

    Benzhong Zhao, Christopher W MacMinn, Bauyrzhan K Primkulov, Yu Chen, Albert J Valocchi, Jianlin Zhao, Qinjun Kang, Kelsey Bruning, James E McClure, Cass T Miller, et al. Compre- hensive comparison of pore-scale models for multiphase flow in porous media. Proceedings of the Na...

    2019

  2. [10]

    An intercomparison of the pore network to the navier–stokes modeling approach applied for saturated conductivity estimation from x-ray ct images

    Bartłomiej Gackiewicz, Krzysztof Lamorski, Cezary Sławi ´nski, Shao-Yiu Hsu, and Liang- Cheng Chang. An intercomparison of the pore network to the navier–stokes modeling approach applied for saturated conductivity estimation from x-ray ct images. Scientific reports, 11(1): 5859, 2021

    2021

  3. [11]

    Review of pore network modelling of porous media: Experimental characterisations, network constructions and applications to reactive transport

    Qingrong Xiong, Todor G Baychev, and Andrey P Jivkov. Review of pore network modelling of porous media: Experimental characterisations, network constructions and applications to reactive transport. Journal of contaminant hydrology, 192:101–117, 2016

    2016

  4. [12]

    Poreflow-net: A 3d convolutional neural network to predict fluid flow through porous media

    Javier E Santos, Duo Xu, Honggeun Jo, Christopher J Landry, Maša Prodanovi´c, and Michael J Pyrcz. Poreflow-net: A 3d convolutional neural network to predict fluid flow through porous media. Advances in Water Resources, 138:103539, 2020

    2020

  5. [13]

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

    Xu-Hui Zhou, James E McClure, Cheng Chen, and Heng Xiao. Neural network–based pore flow field prediction in porous media using super resolution. Physical Review Fluids, 7(7): 074302, 2022

    2022

  6. [14]

    Learning mesh-based simulation with graph networks

    Tobias Pfaff, Meire Fortunato, Alvaro Sanchez-Gonzalez, and Peter W Battaglia. Learning mesh-based simulation with graph networks. arXiv preprint arXiv:2010.03409, 2020

    2010 arXiv

  7. [15]

    Learning to simulate complex physics with graph networks

    Alvaro Sanchez-Gonzalez, Jonathan Godwin, Tobias Pfaff, Rex Ying, Jure Leskovec, and Peter Battaglia. Learning to simulate complex physics with graph networks. In International conference on machine learning, pages 8459–8468. PMLR, 2020

    2020

  8. [16]

    Learning large-scale subsurface simulations with a hybrid graph network simulator

    Tailin Wu, Qinchen Wang, Yinan Zhang, Rex Ying, Kaidi Cao, Rok Sosic, Ridwan Jalali, Hassan Hamam, Marko Maucec, and Jure Leskovec. Learning large-scale subsurface simulations with a hybrid graph network simulator. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge ...

    2022

  9. [17]

    Python workflow for segmenting multiphase flow in porous rocks

    Catherine Spurin, Sharon Ellman, Dane Sherburn, Tom Bultreys, and Hamdi Tchelepi. Python workflow for segmenting multiphase flow in porous rocks. EarthArXiv, 2023

    2023

  10. [18]

    Multiscale meshgraphnets

    Meire Fortunato, Tobias Pfaff, Peter Wirnsberger, Alexander Pritzel, and Peter Battaglia. Multiscale meshgraphnets. arXiv preprint arXiv:2210.00612, 2022. 6

    2022 arXiv

  11. [19]

    On the optimal combi- nation of cross-entropy and soft dice losses for lesion segmentation with out-of-distribution robustness

    Adrian Galdran, Gustavo Carneiro, and Miguel A González Ballester. On the optimal combi- nation of cross-entropy and soft dice losses for lesion segmentation with out-of-distribution robustness. In Diabetic Foot Ulcers Grand Challenge, pages 40–51. Springer, 2022

    2022

  12. [20]

    Distance map loss penalty term for semantic segmentation

    Francesco Caliva, Claudia Iriondo, Alejandro Morales Martinez, Sharmila Majumdar, and Valentina Pedoia. Distance map loss penalty term for semantic segmentation. arXiv preprint arXiv:1908.03679, 2019

    1908 arXiv

  13. [21]

    Focal loss for dense object detection

    Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár. Focal loss for dense object detection. In Proceedings of the IEEE international conference on computer vision , pages 2980–2988, 2017

    2017

  14. [22]

    Tversky loss function for image segmentation using 3d fully convolutional deep networks

    Seyed Sadegh Mohseni Salehi, Deniz Erdogmus, and Ali Gholipour. Tversky loss function for image segmentation using 3d fully convolutional deep networks. In International workshop on machine learning in medical imaging, pages 379–387. Springer, 2017. 7 A Pore structure As illus...

    2017

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.