pith. sign in

arxiv: 2604.02976 · v1 · submitted 2026-04-03 · 💻 cs.CE

Extending deep learning U-Net architecture for predicting unsteady fluid flows in textured microchannels

Ganesh Sahadeo Meshram , Partha Pratim Chakrabarti , Suman Chakraborty This is my paper

Pith reviewed 2026-05-13 18:32 UTC · model grok-4.3

classification 💻 cs.CE
keywords U-Netdeep learningfluid flow predictionmicrochannelslattice Boltzmann methodattention mechanismvelocity field regression
0
0 comments X

The pith

A U-Net model with attention mechanism predicts fluid velocities in textured microchannels at 5.18 percent average error

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

The paper applies the U-Net deep learning architecture, originally for image segmentation, to regress unsteady fluid velocity fields in textured microchannels. Training data comes from lattice Boltzmann simulations that are preprocessed and normalized before being fed to the model. Adding an attention mechanism improves results over the basic U-Net across multiple error metrics, with the enhanced version reaching an average error of 5.18 percent that optimization could bring down to 2.1 percent. The approach shows that spatial-hierarchy capture in U-Net can substitute for repeated full simulations in microscale flow problems.

Core claim

The U-Net equipped with an attention mechanism predicts the velocity magnitude and components for textured microchannels with an average error of 5.18%, which upon optimization may subsequently lower to 2.1%. The U-Net model including an attention mechanism (U-Net AM) regularly surpasses the conventional U-Net model in all measures, evidencing enhanced accuracy and generalization.

What carries the argument

U-Net architecture augmented with attention mechanism, trained on preprocessed lattice Boltzmann simulation data to regress velocity magnitude and vector components

If this is right

  • Velocity predictions depend on the solid-fluid interaction parameter and surface wettability.
  • The attention-enhanced model consistently records lower MSE, RMSE, MAE and higher R-squared scores than the standard U-Net.
  • The method offers a faster alternative to repeated lattice Boltzmann runs for similar microchannel geometries.
  • Further parameter tuning can reduce average error from 5.18 percent toward 2.1 percent.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The trained model could support rapid iterative design of microfluidic devices by generating velocity fields in seconds rather than hours of simulation.
  • Extending the same architecture to other unsteady flow problems or different surface textures would require only retraining on new simulation batches.
  • Coupling the predictor with real-time sensor data could enable closed-loop control in lab-on-chip systems.

Load-bearing premise

Lattice Boltzmann simulations supply accurate ground truth that matches real fluid behavior in textured microchannels without biases from data preprocessing or normalization.

What would settle it

Direct comparison of model outputs against velocity measurements from physical experiments on fabricated textured microchannel devices would confirm or refute the reported error rates.

read the original abstract

In this study, we have explored an application of deep learning architecture of the U-Net model, originally designed for biomedical image segmentation, in a regression analysis aimed at predicting fluid flows through textured microchannels. The data for this analysis is generated using the lattice Boltzmann method through extensive simulations, capturing the intricate behaviors of fluid dynamics in a microscale environment. The raw simulation data was meticulously preprocessed to prepare it for training the U-Net model, ensuring that the input features and labels were appropriately formatted and normalized to optimize the learning process of the model. The U-Net model, with its inherent capability of capturing spatial hierarchies and producing better predictions, proved effective in this novel application. We have evaluated the performance of the model using metrics including MSE, RMSE, MAE, and $R^2$ scores. These metrics were crucial in assessing the accuracy and reliability of the model predictions. The results demonstrate that the U-Net model can predict fluid flows with high accuracy and less error, indicating its potential for broader applications in fluid dynamics and other fields requiring precise regression modeling. A parametric analysis of the U-Net with attention mechanism showed that the velocity field prediction is contingent upon the solid-fluid interaction parameter and surface wettability. The U-Net equipped with an attention mechanism predicts the velocity magnitude and components for textured microchannels with an average error of 5.18%, which upon optimization may subsequently lower to 2.1%. The U-Net model including an attention mechanism (U-Net AM) regularly surpasses the conventional U-Net model in all measures, evidencing enhanced accuracy and generalization.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper applies a U-Net architecture augmented with an attention mechanism to regress velocity magnitude and components for unsteady flows in textured microchannels. Training data are generated exclusively via lattice Boltzmann method (LBM) simulations that incorporate a solid-fluid interaction parameter and surface wettability; the data are preprocessed and normalized before training. Performance is quantified with MSE, RMSE, MAE, and R², yielding an average error of 5.18 % (claimed to be reducible to 2.1 % after optimization) and consistent superiority of the attention-augmented U-Net over the baseline U-Net. A parametric study links prediction quality to the interaction parameter and wettability.

Significance. If the LBM-generated fields are shown to be physically faithful, the work would supply a fast surrogate model for microchannel flow prediction, potentially useful for design iteration in microfluidics. The attention mechanism's reported improvement over plain U-Net is a modest but concrete technical increment for regression tasks in fluid dynamics. However, the complete absence of experimental or analytical validation of the ground-truth data substantially reduces the immediate significance of the numerical results.

major comments (2)
  1. [Abstract / Data Generation] Abstract and data-generation description: all reported errors (5.18 % average, 2.1 % optimized) are measured exclusively against LBM velocity fields. No comparison is supplied to experimental PIV data, to a Navier-Stokes solver on identical geometries, or to analytical limits such as Poiseuille flow in smooth channels; this leaves the central claim of “predicting unsteady fluid flows” dependent on an unverified label.
  2. [Results] Results section: the manuscript states performance metrics (MSE, RMSE, MAE, R²) but supplies no information on train/test splits, cross-validation strategy, number of independent runs, or error bars. Without these, the generalization claim for unsteady flows and the superiority of U-Net AM cannot be rigorously assessed.
minor comments (2)
  1. [Abstract] The abstract asserts that the flow is unsteady yet reports only time-independent velocity magnitude and components; clarification is needed on whether the model predicts instantaneous fields at multiple time steps or time-averaged quantities.
  2. [Methods] Notation for the solid-fluid interaction parameter and wettability is introduced without an explicit equation or table of values; a short methods table would improve reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help improve the rigor and clarity of the manuscript. We address each major point below and will revise the paper to incorporate additional details and comparisons where feasible.

read point-by-point responses

  1. Referee: [Abstract / Data Generation] Abstract and data-generation description: all reported errors (5.18 % average, 2.1 % optimized) are measured exclusively against LBM velocity fields. No comparison is supplied to experimental PIV data, to a Navier-Stokes solver on identical geometries, or to analytical limits such as Poiseuille flow in smooth channels; this leaves the central claim of “predicting unsteady fluid flows” dependent on an unverified label.

    Authors: We agree that experimental PIV validation would strengthen the work but lies outside the current computational scope focused on LBM surrogate modeling. In revision we will add an analytical comparison to Poiseuille flow in smooth (non-textured) channels as a baseline sanity check, expand the methods section to discuss LBM's established physical fidelity for wettability-driven microflows (citing standard references), and clarify that the model predicts LBM-generated fields as a fast surrogate rather than claiming direct experimental equivalence. revision: partial

  2. Referee: [Results] Results section: the manuscript states performance metrics (MSE, RMSE, MAE, R²) but supplies no information on train/test splits, cross-validation strategy, number of independent runs, or error bars. Without these, the generalization claim for unsteady flows and the superiority of U-Net AM cannot be rigorously assessed.

    Authors: We acknowledge this omission reduces reproducibility. The revised manuscript will explicitly state the train/test split ratio (80/20), describe the k-fold cross-validation procedure used, report results averaged over five independent training runs with random seeds, and include error bars (mean ± standard deviation) for all metrics to support the generalization and superiority claims. revision: yes

Circularity Check

0 steps flagged

No circularity: standard supervised learning on held-out LBM data

full rationale

The paper generates velocity-field labels via lattice Boltzmann simulations, preprocesses them into normalized inputs, trains a U-Net (optionally with attention) to regress velocity magnitude and components, and evaluates on held-out test splits using MSE/RMSE/MAE/R². No equation defines a prediction as a function of a fitted parameter by construction, no self-citation supplies a uniqueness theorem or ansatz that the central claim depends on, and the reported 5.18 % error is an empirical test-set statistic rather than a renaming or tautological reduction of the training procedure itself. The pipeline is therefore self-contained against external benchmarks and receives the default non-circularity finding.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the assumption that simulation data serves as reliable ground truth and that standard neural network training will yield generalizable predictors for this spatial regression task. No free parameters are explicitly fitted beyond typical model weights, and no new entities are postulated.

axioms (2)
  • domain assumption Lattice Boltzmann method simulations accurately capture the fluid dynamics in textured microchannels for use as training labels.
    Invoked implicitly when using LBM data as ground truth for the regression task.
  • domain assumption Standard neural network optimization converges to a predictor that generalizes to unseen microchannel configurations.
    Required for the reported accuracy and superiority of U-Net AM to hold.

pith-pipeline@v0.9.0 · 5592 in / 1432 out tokens · 50506 ms · 2026-05-13T18:32:09.485443+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

30 extracted references · 30 canonical work pages

  1. [1]

    Physical mechanisms relevant to flow resistance in textured microchannels,

    S. E. Game, M. Hodes, E. E. Keaveny, and D. T. Papageorgiou, “Physical mechanisms relevant to flow resistance in textured microchannels,” Phys. Rev. Fluids, vol. 2, p. 094102, Sep 2017. TFEC-2025-55517 13

    work page 2017

  2. [2]

    Physics of fluid flow in an hourglass ( converging-diverging) microchannel,

    S. Goli, S. K. Saha, and A. Agrawal, “Physics of fluid flow in an hourglass ( converging-diverging) microchannel,” Physics of Fluids, vol. 34, no. 5, 2022

    work page 2022

  3. [3]

    Single Droplet on Micro Square-Post Patterned Surfaces-Theoretical Model and Numerical Simulation,

    Y. Q. Zu and Y. Y. Yan, “Single Droplet on Micro Square-Post Patterned Surfaces-Theoretical Model and Numerical Simulation,” Scientific Reports, vol. 6, no. February 2015, pp. 1–12, 2016

    work page 2015

  4. [4]

    An Experimental Investigation if Transition of a Plane Poiseuille Flu,

    M.A.Karnitz, M. C. Potter, and M. C. Smith, “An Experimental Investigation if Transition of a Plane Poiseuille Flu,” tec h. rep. 1974

    work page 1974

  5. [5]

    REVIEW PDMS microfluidics: A mini review,

    R. M. Kiran and S. Chakraborty, “REVIEW PDMS microfluidics: A mini review,” 2020

    work page 2020

  6. [6]

    Wetting failure condition on rough surfaces,

    F. C. Yang and X. P. Chen, “Wetting failure condition on rough surfaces,” Chinese Physics B, vol. 28, no. 4, 2019

    work page 2019

  7. [7]

    Direct measurement of slip flows in superhydrophobic microchannels with transverse grooves,

    D. Byun, J. Kim, H. S. Ko, and H. C. Park, “Direct measurement of slip flows in superhydrophobic microchannels with transverse grooves,” Physics of Fluids, vol. 20, no. 11, 2008

    work page 2008

  8. [8]

    Wetting transitions on rough surfaces,

    C.Ishino, K. Okumura, and D. Qu´ er´ e, “Wetting transitions on rough surfaces,” Europhysics Letters, vol. 68, no. 3, pp. 419–425, 2004

    work page 2004

  9. [9]

    Prediction of heat transfer on a flat plate subjected to a transversely pulsating j et using artificial neural networks,

    U. Akdag, M. A. Komur, and S. Akcay, “Prediction of heat transfer on a flat plate subjected to a transversely pulsating j et using artificial neural networks,” Applied Thermal Engineering, vol. 100, pp. 412–420, 2016. Publisher: Elsevier Ltd

    work page 2016

  10. [10]

    The stability of Poiseuille flow in a pipe of circular cross -section,

    H. A. R. O L DSalwen and C. E. Grosch, “The stability of Poiseuille flow in a pipe of circular cross -section,” tech. rep., 1972. Publication Title: J. Fluid Mech Volume: 54 Issue: 1

    work page 1972

  11. [11]

    Slippage of water over hydrophobic surfaces,

    O.I. Vinogradova, “Slippage of water over hydrophobic surfaces,” International Journal of Mineral Processing, vol. 56, pp. 31 60, Apr. 1999. Publisher: Elsevier Science Publishers B.V

    work page 1999

  12. [12]

    Machine learning for fluid mechanics,

    S. L. Brunton, B. R. Noack, and P. Koumoutsakos, “Machine learning for fluid mechanics,” Annua l review of fluid mechanics, vol. 52, pp. 477–508, 2020. Publisher: Annual Reviews

    work page 2020

  13. [13]

    DeepM&Mnet: Inferring the electroconvection Multiphysics fields based on operator approximation by neural networks,

    S. Cai, Z. Wang, L. Lu, T. A. Zaki, and G. E. Karniadakis, “DeepM&Mnet: Inferring the electroconvection Multiphysics fields based on operator approximation by neural networks,” Journal of Computational Physics, vol. 436, July 2021. arXiv: 2009.12935 Publisher: Academic Press Inc

    work page arXiv 2021

  14. [14]

    A composite neural network that learns from multi -fidelity data: Application to function approximation and inverse PDE problems,

    X. Meng and G. E. Karniadakis, “A composite neural network that learns from multi -fidelity data: Application to function approximation and inverse PDE problems,” Feb. 2019. arXiv: 1903.00104

    work page arXiv 2019

  15. [15]

    Multi -step wind speed prediction based on turbulence intensity and hybrid deep neural networks,

    F. Li, G. Ren, and J. Lee, “Multi -step wind speed prediction based on turbulence intensity and hybrid deep neural networks,” Energy Conversion and Management, vol. 186, no. January, pp. 306–322, 2019. Publisher: Elsevier

    work page 2019

  16. [16]

    Review and Examination of Input Feature Prepar ation Methods and Machine Learning Models for Turbulence Modeling,

    S.Luo,J.Cui, M.Vellakal, J. Liu, E. Jiang, S. Koric, and V. Kindratenko, “Review and Examination of Input Feature Prepar ation Methods and Machine Learning Models for Turbulence Modeling,” arXiv preprint arXiv:2001.05485, pp. 1–26, 2020. arXiv: 2001.05485

    work page arXiv 2001

  17. [17]

    Flows Over Periodic Hills of Parameterized Geometries: A Dataset for Data -Driven Turbulence Modeling From Direct Simulations,

    H. Xiao, J.-L. Wu, S. Laizet, and L. Duan, “Flows Over Periodic Hills of Parameterized Geometries: A Dataset for Data -Driven Turbulence Modeling From Direct Simulations,” Computers & Fluids, p. 104431, 2020. arXiv: 1910. 01264 Publisher: Elsevier Ltd

    work page 2020

  18. [18]

    Deep unsupervised learning of turbulence for inflow generation at various Reynolds numbers,

    J. Kim and C. Lee, “Deep unsupervised learning of turbulence for inflow generation at various Reynolds numbers,” Journal of Computational Physics, vol. 406, p. 109216, 2020. arXiv: 1908.10515 Publisher: Elsevier Inc

    work page arXiv 2020

  19. [19]

    Unsteady flow -field forecasting leveraging a hybrid deep -learning architecture,

    C. Guo, Y. Wang, Y. Han, M. Ji, and Y. Wu, “Unsteady flow -field forecasting leveraging a hybrid deep -learning architecture,” Physics of Fluids, vol. 36, no. 6, 2024

    work page 2024

  20. [20]

    Deepxde: A deep learning library for solving differential equations

    L. Lu, X. Meng, Z. Mao, and G. E. Karniadakis, “DeepXDE: A deep learning library fo r solving differential equations,” SIAM Review, vol. 63, no. 1, pp. 208 –228, 2021. arXiv: 1907.04502 Publisher: Society for Industrial and Applied Mathematics Publications

    work page arXiv 2021

  21. [21]

    Linking Gaussia n process regression with data -driven manifold embeddings for nonlinear data fusion,

    S. Lee, F. Dietrich, G. E. Karniadakis, and I. G. Kevrekidis, “Linking Gaussia n process regression with data -driven manifold embeddings for nonlinear data fusion,” Interface Focus, vol. 9, no. 3, 2019. arXiv: 1812.06467 Publisher: Royal Society Publishing

    work page arXiv 2019

  22. [22]

    Automated pulmonary nodule detection in CT images using deep convolutional neural networks,

    H.Xie, D.Yang, N.Sun, Z.Chen, and Y.Zhang, “Automated pulmonary nodule detection in CT images using deep convolutional neural networks,” Pattern Recognition, vol. 85, pp. 109–119, Jan. 2019. Publisher: Elsevier Ltd

    work page 2019

  23. [23]

    Discovering a universal variable -order fractional mo del for turbulent Couette flow using a physics-informed neural network,

    P. P. Mehta, G. Pang, F. Song, and G. E. Karniadakis, “Discovering a universal variable -order fractional mo del for turbulent Couette flow using a physics-informed neural network,” Fractional Calculus and Applied Analysis, vol. 22, pp. 1675–1688, Dec

    work page

  24. [24]

    Publisher: De Gruyter

    work page

  25. [25]

    Physics -informed neural netw orks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations,

    M. Raissi, P. Perdikaris, and G. E. Karniadakis, “Physics -informed neural netw orks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations,” Journal of Computational Physics, vol. 378, pp. 686–707, 2019. Publisher: Elsevier Inc

    work page 2019

  26. [26]

    NSFnets (Navier -Stokes Flow nets): Physics -informed neural networks for the incompressible Navier-Stokes equations,

    X. Jin, S. Cai, H. Li, and G. E. Karn iadakis, “NSFnets (Navier -Stokes Flow nets): Physics -informed neural networks for the incompressible Navier-Stokes equations,” Mar. 2020. arXiv: 2003.06496

    work page arXiv 2020

  27. [27]

    Classification of machine learning frameworks for data-driven thermal fluid models,

    C. W. Chang and N. T. Dinh, “Classification of machine learning frameworks for data-driven thermal fluid models,” International Journal of Thermal Sciences, vol. 135, no. August 2018, pp. 559–579, 2019. arXiv: 1801.06621 Publisher: Elsevier

    work page arXiv 2018

  28. [28]

    Data -driven modeling and learning in science and engineering,

    F. J. Mont´ans, F. Chinesta, R. Gomez-Bombarelli, and J. N. Kutz , “Data -driven modeling and learning in science and engineering,” Comptes Rendus-Mecanique, vol. 347, no. 11, pp. 845–855, 2019. Publisher: Elsevier Masson SAS

    work page 2019

  29. [29]

    Kr¨ uger, H

    T. Kr¨ uger, H. Kusumaatmaja, A. Kuzmin, O. Shardt, G. Silva, and E. M. Viggen, The lattic e Boltzmann method, vol. 10. Springer, 2017. Publication Title: Springer International Publishing Issue: 978-3

    work page 2017

  30. [30]

    Numerical Investigation of Wettability and its Effects on Flow through Textured Micro-channels using Lattice Bo ltzmann Method,

    G.Meshram and S.Kondaraju, “Numerical Investigation of Wettability and its Effects on Flow through Textured Micro-channels using Lattice Bo ltzmann Method,” in Proceedings of the 26thNational and 4th International ISHMT -ASTFE Heat and Mass Transfer Conference December 17-20, 2021, IIT Madras, Chennai-600036, Tamil Nadu, India, 2021

    work page 2021