pith. sign in

arxiv: 2510.21804 · v2 · submitted 2025-10-21 · 💻 cs.LG · physics.flu-dyn

XRePIT: A deep learning-computational fluid dynamics hybrid framework implemented in OpenFOAM for fast, robust, and scalable unsteady simulations

Shilaj Baral , Youngkyu Lee , Sangam Khanal , Joongoo Jeon This is my paper

Pith reviewed 2026-05-18 05:35 UTC · model grok-4.3

classification 💻 cs.LG physics.flu-dyn
keywords hybrid simulationneural surrogatecomputational fluid dynamicsresidual monitoringunsteady flowOpenFOAM frameworkerror controllong-term stability
0
0 comments X

The pith

Residual-guided coupling lets neural surrogates run stable long-term 3D fluid simulations by switching to OpenFOAM when drift appears.

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

The paper develops an automated hybrid framework that runs a neural surrogate for most steps of an unsteady flow simulation but hands control to the full OpenFOAM solver whenever a residual threshold detects growing errors. Standalone neural models accumulate mistakes and eventually produce non-physical results, so the hybrid switch prevents that drift while retaining most of the speed gain. Tested on three-dimensional buoyancy-driven flow, the method reaches nearly three times the wall-clock speed of a pure solver run and holds relative errors near one part in a thousand. The same residual monitor works with more than one kind of neural network, showing the guardrail does not depend on a specific architecture. This supplies a practical, open-source route to longer and more reliable accelerated simulations inside an existing CFD code.

Core claim

The central claim is that a monitored residual threshold can manage state transitions between a neural surrogate and the OpenFOAM finite-volume solver. When the surrogate prediction exceeds the threshold, the framework transfers the current state to the physics solver for corrective steps and then returns to the surrogate. This residual-guided loop keeps the simulation physical over time intervals that exceed the stability limit of the surrogate alone. On buoyancy-driven flow the hybrid achieves up to 2.91 times wall-clock acceleration while relative L2 errors stay within order 1E-03. The stabilizing action remains effective when the underlying neural model is replaced by a finite-volume FNO

What carries the argument

Residual threshold monitor that detects non-physical drift in the neural surrogate and triggers corrective steps from the OpenFOAM solver.

If this is right

  • Unsteady 3D flow simulations can extend to longer time horizons without the instability that limits pure neural rollouts.
  • Wall-clock time is reduced by up to a factor of 2.91 while accuracy stays comparable to full-order numerical solutions.
  • The residual coupling logic works with multiple neural architectures without requiring changes to the switching mechanism.
  • The entire workflow is automated inside OpenFOAM, removing the need for custom manual hybrid code.

Where Pith is reading between the lines

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

  • The same residual guard could be applied in other physics domains where surrogate models drift, such as structural or thermal problems.
  • Making the threshold adaptive or learned might reduce the number of solver interventions and raise the overall speedup.
  • The pattern of using a physics solver as a safety net suggests a general approach for hybrid modeling across scientific computing.
  • Testing on flows with higher Reynolds numbers would show whether the current threshold choice generalizes or needs regime-specific adjustment.

Load-bearing premise

A single residual threshold can reliably detect non-physical drift in the neural surrogate and trigger corrective solver steps without introducing coupling instabilities or excessive overhead across different flow regimes and neural architectures.

What would settle it

Run the hybrid simulation for a duration several times longer than the point at which a standalone surrogate produces visible non-physical artifacts or diverges, then verify whether L2 errors remain below 0.001 and whether flow features stay physical.

read the original abstract

Autoregressive neural surrogates offer computational acceleration for fluid dynamics but inherently suffer from error accumulation and non-physical drift during long-term rollouts. Although hybrid strategies combining surrogate models and physics-based solvers have been proposed, they are limited to manual implementations for low-dimensional benchmarks. In this study, we propose an OpenFOAM-based hybrid framework, XRePIT (eXtensible Residual-based Physics-nformed Transfer learning), characterized by its fastness, robustness, and scalability. Unlike prior manual implementations (e.g., RePIT), XRePIT integrates a fully automated open-source workflow that manages the state transition between a neural surrogate and a traditional numerical solver (OpenFOAM) based on a monitored residual threshold. Using 3D buoyancy-driven flow as a testbed, we demonstrate that this residual-guided coupling enables stable long-term simulation-ell beyond the stability horizon of standalone surrogates. Our results indicate that the hybrid loop achieves up to 2.91x wall-clock acceleration while maintaining relative L2 errors within O(1E-03) Furthermore, we benchmark the framework's extensibility by introducing a finite-volume-based Fourier neural operator (FVFNO), confirming that the stabilizing effect of the residual guardrail is agnostic to the underlying neural architecture. This study provides a deployable methodology for fast, robust, and automated hybrid simulation in 3D unsteady flow.

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 / 1 minor

Summary. The manuscript introduces XRePIT, an OpenFOAM-based hybrid framework that automatically switches between neural surrogate predictions and full physics-based solves using a monitored residual threshold. On a 3D buoyancy-driven flow test case it reports stable long-term rollouts beyond standalone surrogate limits, up to 2.91x wall-clock acceleration, and relative L2 errors of O(1E-03); the same residual guardrail is shown to work with both an original surrogate and a finite-volume Fourier neural operator (FVFNO), supporting the claim of architecture-agnostic behavior.

Significance. An automated, open-source hybrid workflow inside OpenFOAM that mitigates surrogate drift while delivering measurable wall-clock gains is a practical contribution to ML-accelerated CFD. The explicit demonstration that the residual coupling remains effective when the surrogate is replaced by FVFNO is a clear strength. Significance is nevertheless constrained by the narrow validation scope: a single flow regime and parameter set.

major comments (2)
  1. [Abstract and 3D testbed results] Abstract and the 3D buoyancy-driven flow testbed results: the claim that a single residual threshold reliably detects non-physical drift and hands control back to OpenFOAM without introducing instabilities or excessive overhead is load-bearing for the central contribution, yet the manuscript provides no ablation on threshold value, no variation of Rayleigh number or domain aspect ratio, and no reporting of intervention frequency versus error growth. The demonstration is limited to one test case and two architectures.
  2. [Abstract and results section] Abstract and results section: the reported 2.91x acceleration and O(1E-03) L2 errors are presented without details on training data, threshold selection procedure, baseline comparisons (pure OpenFOAM, standalone surrogate), or statistical validation across multiple runs, limiting assessment of whether the data support the robustness and speedup claims.
minor comments (1)
  1. [Abstract] Abstract contains the typographical error 'simulation-ell beyond' that should read 'simulation beyond'.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback and for recognizing the practical value of the automated residual-guided hybrid workflow in XRePIT. We address each major comment below with clarifications and indicate the revisions we will make to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract and 3D testbed results] Abstract and the 3D buoyancy-driven flow testbed results: the claim that a single residual threshold reliably detects non-physical drift and hands control back to OpenFOAM without introducing instabilities or excessive overhead is load-bearing for the central contribution, yet the manuscript provides no ablation on threshold value, no variation of Rayleigh number or domain aspect ratio, and no reporting of intervention frequency versus error growth. The demonstration is limited to one test case and two architectures.

    Authors: We agree that additional analysis of the threshold would strengthen the central claim. In the revised manuscript we will add an ablation on threshold values, including plots of intervention frequency versus error growth for the 3D buoyancy-driven flow case. We will also expand the results section to report the number and timing of interventions during the long-term rollouts. While the study is centered on a single representative 3D test case, we will add a brief discussion of expected generalization and note that the residual guardrail remains effective across the two distinct architectures (original surrogate and FVFNO). A full parametric sweep over Rayleigh number and aspect ratio would require substantial new simulations beyond the current scope; we will therefore treat this as a limitation to be addressed in future work rather than claiming broad coverage. revision: partial

  2. Referee: [Abstract and results section] Abstract and results section: the reported 2.91x acceleration and O(1E-03) L2 errors are presented without details on training data, threshold selection procedure, baseline comparisons (pure OpenFOAM, standalone surrogate), or statistical validation across multiple runs, limiting assessment of whether the data support the robustness and speedup claims.

    Authors: We will revise the results section and supplementary material to include: (i) a clear description of the training dataset and surrogate training procedure, (ii) the rationale and preliminary tests used to select the residual threshold, and (iii) direct baseline comparisons of wall-clock time and L2 error against both pure OpenFOAM and the standalone surrogate on the same hardware. Because the underlying CFD solver and surrogate inference are deterministic for fixed initial conditions, we will clarify this point and, where possible, report consistency across a small set of runs with minor initial-condition perturbations to address statistical robustness. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical framework demonstration without self-referential derivation

full rationale

The paper presents an engineering framework (XRePIT) for hybrid surrogate-solver coupling in OpenFOAM, with performance claims (2.91x acceleration, O(1E-03) L2 errors) obtained from direct numerical experiments on a 3D buoyancy-driven flow testbed. No mathematical derivation chain exists that reduces predictions or results by construction to fitted parameters, self-definitions, or self-citation load-bearing premises. The residual-threshold switching mechanism is implemented and validated empirically rather than derived; prior work (RePIT) is referenced only for context and does not substitute for independent verification of the reported metrics. The contribution is self-contained through external benchmark comparisons in the simulations themselves.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The framework rests on standard CFD governing equations and neural network approximation capabilities plus the practical assumption that residual monitoring suffices for stable coupling; no new physical entities are introduced.

free parameters (1)
  • residual threshold
    The value that triggers solver intervention is a tunable parameter whose selection affects the speed-accuracy tradeoff and is not derived from first principles.
axioms (1)
  • domain assumption Neural surrogates produce short-term predictions sufficiently close to the true solution that residual-based correction can restore physical consistency.
    Implicit premise required for the hybrid loop to remain stable rather than oscillate or diverge.

pith-pipeline@v0.9.0 · 5805 in / 1220 out tokens · 57780 ms · 2026-05-18T05:35:50.091443+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.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. A Numerical Method for Coupling Parameterized Physics-Informed Neural Networks and FDM for Advanced Thermal-Hydraulic System Simulation

    cs.LG 2026-04 conditional novelty 6.0

    A single trained parameterized NA-PINN coupled to FDM delivers low-error solutions for gravity-driven draining across multiple time steps and initial conditions without retraining or simulation data.

Reference graph

Works this paper leans on

49 extracted references · 49 canonical work pages · cited by 1 Pith paper · 3 internal anchors

  1. [1]

    Progress in Nuclear Energy 73, 75–85 (2014) 19

    Locatelli, G., Bingham, C., Mancini, M.: Small modular reactors: A comprehen- sive overview of their economics and strategic aspects. Progress in Nuclear Energy 73, 75–85 (2014) 19

    work page 2014

  2. [2]

    arXiv preprint arXiv:2407.17822 (2024)

    Jeon, J., Rabault, J., Vasanth, J., Alc´ antara- ´Avila, F., Baral, S., Vinuesa, R.: Advanced deep-reinforcement-learning methods for flow control: group-invariant and positional-encoding networks improve learning speed and quality. arXiv preprint arXiv:2407.17822 (2024)

    work page arXiv 2024

  3. [3]

    Control Engineering Practice13(6), 789–803 (2005)

    Kim, C.-S., Hong, K.-S., Kim, M.-K.: Nonlinear robust control of a hydraulic elevator: experiment-based modeling and two-stage lyapunov redesign. Control Engineering Practice13(6), 789–803 (2005)

    work page 2005

  4. [4]

    Nuclear Engineering and Technology51(8), 1939–1950 (2019)

    Jeon, J., Kim, Y.S., Choi, W., Kim, S.J.: Identification of hydrogen flammability in steam generator compartment of opr1000 using melcor and cfx codes. Nuclear Engineering and Technology51(8), 1939–1950 (2019)

    work page 1939

  5. [5]

    Journal of loss prevention in the process industries52, 125–139 (2018)

    Tolias, I., Stewart, J., Newton, A., Keenan, J., Makarov, D., Hoyes, J., Molkov, V., Venetsanos, A.: Numerical simulations of vented hydrogen deflagration in a medium-scale enclosure. Journal of loss prevention in the process industries52, 125–139 (2018)

    work page 2018

  6. [6]

    Matematiˇ ceskij sbornik47(3), 271–306 (1959)

    Godunov, S.K., Bohachevsky, I.: Finite difference method for numerical computa- tion of discontinuous solutions of the equations of fluid dynamics. Matematiˇ ceskij sbornik47(3), 271–306 (1959)

    work page 1959

  7. [7]

    Handbook of numerical analysis7, 713–1018 (2000)

    Eymard, R., Gallou¨ et, T., Herbin, R.: Finite volume methods. Handbook of numerical analysis7, 713–1018 (2000)

    work page 2000

  8. [8]

    John Wiley & Sons, ??? (2012)

    Dhatt, G., Lefran¸ cois, E., Touzot, G.: Finite Element Method. John Wiley & Sons, ??? (2012)

    work page 2012

  9. [9]

    Neural networks2(5), 359–366 (1989)

    Hornik, K., Stinchcombe, M., White, H.: Multilayer feedforward networks are universal approximators. Neural networks2(5), 359–366 (1989)

    work page 1989

  10. [10]

    Nature communications11(1), 233 (2020)

    Vinuesa, R., Azizpour, H., Leite, I., Balaam, M., Dignum, V., Domisch, S., Fell¨ ander, A., Langhans, S.D., Tegmark, M., Fuso Nerini, F.: The role of artificial intelligence in achieving the sustainable development goals. Nature communications11(1), 233 (2020)

    work page 2020

  11. [11]

    Annual review of fluid mechanics52(1), 477–508 (2020)

    Brunton, S.L., Noack, B.R., Koumoutsakos, P.: Machine learning for fluid mechanics. Annual review of fluid mechanics52(1), 477–508 (2020)

    work page 2020

  12. [12]

    Proceedings of the National Academy of Sciences118(21), 2101784118 (2021)

    Kochkov, D., Smith, J.A., Alieva, A., Wang, Q., Brenner, M.P., Hoyer, S.: Machine learning–accelerated computational fluid dynamics. Proceedings of the National Academy of Sciences118(21), 2101784118 (2021)

    work page 2021

  13. [13]

    Journal of Ocean Engineering and Technology36(3), 194–210 (2022) 20

    Kim, T., Lee, W.-D.: Review on applications of machine learning in coastal and ocean engineering. Journal of Ocean Engineering and Technology36(3), 194–210 (2022) 20

    work page 2022

  14. [14]

    Nature Computational Science2(6), 358–366 (2022)

    Vinuesa, R., Brunton, S.L.: Enhancing computational fluid dynamics with machine learning. Nature Computational Science2(6), 358–366 (2022)

    work page 2022

  15. [15]

    Engineering Applications of Artificial Intelligence160, 111879 (2025) https:// doi.org/10.1016/j.engappai.2025.111879

    Sinha, S., Bharill, N., Patel, O.P., Jetta, M.: Active learning with gaussian process regression for solving non-linear time-dependent partial differential equations. Engineering Applications of Artificial Intelligence160, 111879 (2025) https:// doi.org/10.1016/j.engappai.2025.111879

    work page doi:10.1016/j.engappai.2025.111879 2025

  16. [16]

    Engineering Applications of Artificial Intelligence153, 110898 (2025) https://doi.org/10

    Raut, R., Ball, A.K., Basak, A.: Scalable and transferable graph neural networks for predicting temperature evolution in laser powder bed fusion. Engineering Applications of Artificial Intelligence153, 110898 (2025) https://doi.org/10. 1016/j.engappai.2025.110898

    work page arXiv 2025

  17. [17]

    Computer Methods in Applied Mechanics and Engineering440, 117912 (2025) https://doi.org/10.1016/j.cma.2025.117912

    Nguyen, H.V., Chen, J.-U., Bui-Thanh, T.: A model-constrained discontinuous galerkin network (dgnet) for compressible euler equations with out-of-distribution generalization. Computer Methods in Applied Mechanics and Engineering440, 117912 (2025) https://doi.org/10.1016/j.cma.2025.117912

    work page doi:10.1016/j.cma.2025.117912 2025

  18. [18]

    In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp

    Guo, X., Li, W., Iorio, F.: Convolutional neural networks for steady flow approx- imation. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 481–490 (2016)

    work page 2016

  19. [19]

    Advances in neural information processing systems 25(2012)

    Krizhevsky, A., Sutskever, I., Hinton, G.E.: Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems 25(2012)

    work page 2012

  20. [20]

    A Deep Learning based Approach to Reduced Order Modeling for Turbulent Flow Control using LSTM Neural Networks

    Mohan, A.T., Gaitonde, D.V.: A deep learning based approach to reduced order modeling for turbulent flow control using lstm neural networks. arXiv preprint arXiv:1804.09269 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  21. [21]

    Journal of Computational physics378, 686–707 (2019)

    Raissi, M., Perdikaris, P., Karniadakis, G.E.: Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computational physics378, 686–707 (2019)

    work page 2019

  22. [22]

    Journal of Computational Physics428, 110079 (2021)

    Gao, H., Sun, L., Wang, J.-X.: Phygeonet: Physics-informed geometry-adaptive convolutional neural networks for solving parameterized steady-state pdes on irregular domain. Journal of Computational Physics428, 110079 (2021)

    work page 2021

  23. [23]

    arXiv preprint arXiv:2504.16447 (2025)

    Shin, J., Kim, C., Yang, S., Lee, M., Kim, S.J., Jeon, J.: Node assigned physics-informed neural networks for thermal-hydraulic system simulation: Cvh/fl module. arXiv preprint arXiv:2504.16447 (2025)

    work page arXiv 2025

  24. [24]

    Engineering Applications of Artificial Intelligence127, 107324 (2024) https://doi.org/10.1016/j.engappai.2023.107324 21

    Sahli Costabal, F., Pezzuto, S., Perdikaris, P.: Delta-pinns: Physics-informed neural networks on complex geometries. Engineering Applications of Artificial Intelligence127, 107324 (2024) https://doi.org/10.1016/j.engappai.2023.107324 21

    work page doi:10.1016/j.engappai.2023.107324 2024

  25. [25]

    Fourier Neural Operator for Parametric Partial Differential Equations

    Li, Z., Kovachki, N., Azizzadenesheli, K., Liu, B., Bhattacharya, K., Stuart, A., Anandkumar, A.: Fourier neural operator for parametric partial differential equations. arXiv preprint arXiv:2010.08895 (2020)

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  26. [26]

    DeepONet: Learning nonlinear operators for identifying differential equations based on the universal approximation theorem of operators

    Lu, L., Jin, P., Karniadakis, G.E.: Deeponet: Learning nonlinear operators for identifying differential equations based on the universal approximation theorem of operators. arXiv preprint arXiv:1910.03193 (2019)

    work page internal anchor Pith review Pith/arXiv arXiv 1910

  27. [27]

    Computer Methods in Applied Mechanics and Engineering 427, 117035 (2024) https://doi.org/10.1016/j.cma.2024.117035

    Ye, X., Li, H., Huang, J., Qin, G.: On the locality of local neural operator in learn- ing fluid dynamics. Computer Methods in Applied Mechanics and Engineering 427, 117035 (2024) https://doi.org/10.1016/j.cma.2024.117035

    work page doi:10.1016/j.cma.2024.117035 2024

  28. [28]

    Computer Meth- ods in Applied Mechanics and Engineering437, 117750 (2025) https://doi.org/ 10.1016/j.cma.2025.117750

    Wang, X., Li, P., Lu, D.: Phase-field hydraulic fracturing operator network based on en-deeponet with integrated physics-informed mechanisms. Computer Meth- ods in Applied Mechanics and Engineering437, 117750 (2025) https://doi.org/ 10.1016/j.cma.2025.117750

    work page doi:10.1016/j.cma.2025.117750 2025

  29. [29]

    Advances in neural information processing systems34, 26548–26560 (2021)

    Krishnapriyan, A., Gholami, A., Zhe, S., Kirby, R., Mahoney, M.W.: Charac- terizing possible failure modes in physics-informed neural networks. Advances in neural information processing systems34, 26548–26560 (2021)

    work page 2021

  30. [30]

    International Journal of Heat and Mass Transfer220, 124900 (2024)

    Jeon, J., Lee, J., Vinuesa, R., Kim, S.J.: Residual-based physics-informed transfer learning: A hybrid method for accelerating long-term cfd simulations via deep learning. International Journal of Heat and Mass Transfer220, 124900 (2024)

    work page 2024

  31. [31]

    arXiv preprint arXiv:2502.06837 (2025)

    Khanal, S., Baral, S., Jeon, J.: Comparison of cnn-based deep learning architectures for unsteady cfd acceleration on small datasets. arXiv preprint arXiv:2502.06837 (2025)

    work page arXiv 2025

  32. [32]

    Nature Machine Intelligence, 1–11 (2024)

    Zhang, E., Kahana, A., Kopaniˇ c´ akov´ a, A., Turkel, E., Ranade, R., Pathak, J., Karniadakis, G.E.: Blending neural operators and relaxation methods in pde numerical solvers. Nature Machine Intelligence, 1–11 (2024)

    work page 2024

  33. [33]

    https://arxiv.org/abs/ 2405.12380

    Lee, Y., Liu, S., Zou, Z., Kahana, A., Turkel, E., Ranade, R., Pathak, J., Kar- niadakis, G.E.: Fast meta-solvers for 3D complex-shape scatterers using neural operators trained on a non-scattering problem (2025). https://arxiv.org/abs/ 2405.12380

    work page arXiv 2025

  34. [34]

    Computer Methods in Applied Mechanics and Engineering429, 117133 (2024)

    Sousa, P., Rodrigues, C.V., Afonso, A.: Enhancing cfd solver with machine learn- ing techniques. Computer Methods in Applied Mechanics and Engineering429, 117133 (2024)

    work page 2024

  35. [35]

    npj Computational Materials10(1), 145 (2024) 22

    Oommen, V., Shukla, K., Desai, S., Dingreville, R., Karniadakis, G.E.: Rethink- ing materials simulations: Blending direct numerical simulations with neural operators. npj Computational Materials10(1), 145 (2024) 22

    work page 2024

  36. [36]

    In: International Workshop on Coupled Methods in Numerical Dynamics, vol

    Jasak, H., Jemcov, A., Tukovic, Z.,et al.: Openfoam: A c++ library for com- plex physics simulations. In: International Workshop on Coupled Methods in Numerical Dynamics, vol. 1000, pp. 1–20 (2007). Dubrovnik, Croatia)

    work page 2007

  37. [37]

    International Journal of Energy Research46(8), 10770–10795 (2022)

    Jeon, J., Lee, J., Kim, S.J.: Finite volume method network for the accelera- tion of unsteady computational fluid dynamics: Non-reacting and reacting flows. International Journal of Energy Research46(8), 10770–10795 (2022)

    work page 2022

  38. [38]

    Network Theory Ltd

    Van Rossum, G., Drake, F.L.: An Introduction to Python. Network Theory Ltd. Bristol, ??? (2003)

    work page 2003

  39. [39]

    (English translation of Ph

    Nielsen, P.V.: Flow in air conditioned rooms. (English translation of Ph. D. thesis from the Technical University of Denmark (1976)

    work page 1976

  40. [40]

    Journal of Advanced Research in Fluid Mechanics and Thermal Sciences107, 1–18 (2023)

    Kit, L.W., Mohamed, H., Luon, N.Y., Chan, L.: Numerical simulation of ventila- tion in a confined space. Journal of Advanced Research in Fluid Mechanics and Thermal Sciences107, 1–18 (2023)

    work page 2023

  41. [41]

    Ferziger, J.H., Peri´ c, M.: Computational Methods for Fluid Dynamics vol. 586. Springer, ??? (2002)

    work page 2002

  42. [42]

    https://github.com/xu- xianghua/ofpp (2017)

    Xianghua, X.: OpenFOAM Python Parser (Ofpp). https://github.com/xu- xianghua/ofpp (2017)

    work page 2017

  43. [43]

    Ascher, D., Dubois, P.F., Hinsen, K., Hugunin, J., Oliphant, T., et al.: Numerical python (2001)

    work page 2001

  44. [44]

    Packt Publishing Ltd, ??? (2014)

    L´ opez, F., Romero, V.: Mastering Python Regular Expressions. Packt Publishing Ltd, ??? (2014)

    work page 2014

  45. [45]

    Journal of Computational Physics230(3), 628–643 (2011)

    Hajibeygi, H., Jenny, P.: Adaptive iterative multiscale finite volume method. Journal of Computational Physics230(3), 628–643 (2011)

    work page 2011

  46. [46]

    Greenshields, H.G.W.: Notes on Computational Fluid Dynamics: General Principles

    C.J. Greenshields, H.G.W.: Notes on Computational Fluid Dynamics: General Principles. https://doc.cfd.direct/notes/cfd-general-principles/residual (2022)

    work page 2022

  47. [47]

    In: Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pp

    Akiba, T., Sano, S., Yanase, T., Ohta, T., Koyama, M.: Optuna: A next- generation hyperparameter optimization framework. In: Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pp. 2623–2631 (2019)

    work page 2019

  48. [48]

    In: International Conference on Machine Learning, pp

    Ioffe, S., Szegedy, C.: Batch normalization: Accelerating deep network training by reducing internal covariate shift. In: International Conference on Machine Learning, pp. 448–456 (2015). pmlr

    work page 2015

  49. [49]

    Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., Salakhutdinov, R.: Dropout: a simple way to prevent neural networks from overfitting. The journal of machine learning research15(1), 1929–1958 (2014) 23 7 Author contributions Shilaj Baral:Writing – original draft, Validation, Software, Methodology, Investi- gation.Youngkyu Lee:Writing – review/ed...

    work page 1929