arxiv: 2605.00937 · v1 · submitted 2026-05-01 · ⚛️ physics.flu-dyn · cs.LG

Recognition: unknown

An ALE-Consistent Graph Neural Operator-Transformer Framework for Fluid-Structure Interaction

Haokui Jiang, Mart\'in Saravia, Shihang Zhao, Shunxiang Cao, Zhiyang Xue

Authors on Pith no claims yet

Pith reviewed 2026-05-09 19:11 UTC · model grok-4.3

classification ⚛️ physics.flu-dyn cs.LG

keywords fluid-structure interactionarbitrary Lagrangian-Euleriangraph neural operatorvision transformermachine learning surrogatelong-term predictionpartitioned couplingmoving interface

0 comments

The pith

An ALE-consistent coupling of graph neural operators and transformers predicts stable long-term fluid-structure interactions on deforming meshes.

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

The paper develops a machine learning surrogate for fluid-structure interaction problems that must handle moving interfaces between fluid and solid domains. It uses a graph neural operator combined with a vision transformer to forecast fluid flow on changing meshes and a recurrent network to track structural motion, linking them through a partitioned solver. A key addition is a correction step based on arbitrary Lagrangian-Eulerian formulation that aligns velocities at the shared boundary using the latest structural predictions. This setup, refined through staged training to curb error buildup, produces phase-accurate forecasts over many time steps and holds up when inlet flows change, offering a path to faster simulations than full physics solvers for engineering applications.

Core claim

The framework models fluid dynamics with a graph neural operator and vision transformer surrogate on deforming unstructured meshes while an LSTM predicts structural kinematics, coupled via partitioned procedure with an ALE-consistent boundary-correction that updates fluid interface velocity from predicted structural velocity to enforce compatibility, yielding accurate phase-consistent long-term predictions and generalization under inlet variations on the flexible beam in cylinder wake benchmark.

What carries the argument

The ALE-consistent boundary-correction step that enforces kinematic compatibility by updating the fluid-side interface velocity with the predicted structural velocity at each coupling update.

If this is right

Phase-consistent predictions are maintained over long autoregressive rollouts in fluid-structure interaction simulations.
Robust performance holds under variations in inlet profiles for both interpolation and extrapolation regimes.
The vision transformer module, ALE boundary correction, and long-term autoregressive fine-tuning each contribute to overall accuracy and stability as shown in ablation studies.

Where Pith is reading between the lines

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

The approach may extend to other moving-boundary multiphysics problems where interface conditions are critical for stability.
Reducing computational cost for repeated FSI queries could enable optimization loops or uncertainty quantification in design.
Further validation on three-dimensional or turbulent cases would test the scalability of the mesh-handling and correction strategy.

Load-bearing premise

Kinematic compatibility between fluid and structure at the deforming interface holds when fluid velocities are corrected using the structural velocity predictions from the LSTM.

What would settle it

Observation of accumulating phase errors or interface mismatches in long-term rollouts without the boundary correction step, or failure to generalize when inlet conditions differ substantially from training data.

Figures

Figures reproduced from arXiv: 2605.00937 by Haokui Jiang, Mart\'in Saravia, Shihang Zhao, Shunxiang Cao, Zhiyang Xue.

**Figure 1.** Figure 1: Schematic of the flexible beam (gray shaded region) clamped to a rigid cylinder (white circle) and view at source ↗

**Figure 2.** Figure 2: ALE-consistent GNO–Transformer framework for long-horizon FSI prediction. (a) Overview of the view at source ↗

**Figure 3.** Figure 3: Gmsh generate fluid mesh based on the structure boundary. view at source ↗

**Figure 4.** Figure 4: Prediction and truth for the time histories of normalized transverse displacement at monitoring point A in view at source ↗

**Figure 5.** Figure 5: Comparison of prediction and truth of flow field view at source ↗

**Figure 6.** Figure 6: Prediction and truth for the time histories of normalized structural displacements at monitoring point A in view at source ↗

**Figure 7.** Figure 7: Comparison of prediction and truth of flow field at time view at source ↗

**Figure 8.** Figure 8: Truth and prediction results of flow field by different methods at step 1 and step 15 for view at source ↗

**Figure 9.** Figure 9: Time evolution of the correlation coefficient view at source ↗

**Figure 10.** Figure 10: Comparison of boundary pressure predictions by GNO-ViT, GNO and GNO-ViT-NoBM for view at source ↗

**Figure 11.** Figure 11: Time-evolution of predicted 𝑝 𝑓 errors by different method for 𝑐2 = −2: (a) GNO-ViT, (b) GNO, (c) GNO-ViT-noBM view at source ↗

**Figure 12.** Figure 12: Structural boundary node indexing and representative displacements of the flexible beam in the view at source ↗

**Figure 13.** Figure 13: Prediction and distribution of structural boundary velocity for GNO-ViT and GNO-ViT-NoBM: predicted view at source ↗

**Figure 14.** Figure 14: Prediction and truth for the time histories of normalized structural displacements at monitoring point A view at source ↗

**Figure 15.** Figure 15: Time evolution of the flow-field correlation coefficient view at source ↗

read the original abstract

We propose an arbitrary Lagrangian-Eulerian (ALE)-consistent machine learning framework for long-term fluid-structure interaction (FSI) prediction on deforming unstructured meshes. Specifically, the fluid dynamics are modeled by a surrogate that combines a graph neural operator (GNO) with a vision Transformer (ViT) for spatiotemporal prediction, while a lightweight long short-term memory (LSTM) network predicts structural kinematics at the interface. The two surrogates are coupled through a standard partitioned procedure. Most importantly, kinematic compatibility at the moving interface is enforced via an ALE-consistent boundary-correction step that updates the fluid-side interface velocity with the predicted structural velocity at each coupling update, thereby improving near-interface accuracy and long-term rollout stability. To mitigate autoregressive error accumulation, a two-stage training strategy is adopted, consisting of single-step supervised pretraining followed by long-term autoregressive fine-tuning. The proposed framework is validated on the benchmark problem of a flexible beam vibration in the wake of a cylinder. Results demonstrate accurate phase-consistent predictions over long rollouts and robust generalization under inlet-profile variations in both interpolation and extrapolation settings. Systematic ablation studies further assess the respective contributions of the ViT module, ALE-consistent boundary correction, and long-term training to predictive accuracy and rollout robustness.

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.

Desk Editor's Note private letter to a colleague

The paper's main addition is an ALE-consistent boundary velocity correction that couples a GNO-ViT fluid surrogate to an LSTM structure model and appears to help long-term rollout stability on the cylinder-beam benchmark.

read the letter

The work combines a graph neural operator with a vision transformer to handle fluid prediction on deforming unstructured meshes, uses a simple LSTM for the structural response at the interface, and couples them through a partitioned scheme. The distinctive step is the ALE-consistent correction that overwrites the fluid-side interface velocity with the structure prediction at each step to enforce kinematic compatibility. They also use two-stage training: first single-step supervised, then autoregressive fine-tuning to limit error growth over long horizons. Ablations check the separate roles of the ViT, the boundary correction, and the long-term training phase. Tests cover the standard flexible beam in cylinder wake case and check generalization across inlet profiles in both interpolation and extrapolation regimes. The abstract claims phase-consistent predictions and improved stability from the correction, which is plausible given the design. The approach is technically coherent and the validation protocol follows common practice in the field. The main limitation is that the provided description supplies no concrete error values, mesh resolutions, or rollout lengths, so the size of the gains from each component remains unclear. If the full paper contains solid quantitative tables and figures without obvious overfitting, the contribution is real but incremental rather than transformative. This is useful reading for groups already working on operator learning or reduced-order models for fluid-structure problems who want a concrete coupling recipe. It is not yet something I would cite in my own work, but the method is clear enough and the benchmark standard enough that a serious editor should send it out for review rather than desk-reject.

Referee Report

2 major / 3 minor

Summary. The manuscript proposes an ALE-consistent machine learning framework for long-term fluid-structure interaction (FSI) on deforming unstructured meshes. Fluid dynamics are modeled by a graph neural operator (GNO) combined with a vision Transformer (ViT) for spatiotemporal prediction, while structural kinematics at the interface are predicted by a lightweight LSTM network. The surrogates are coupled via a partitioned procedure, with kinematic compatibility enforced by an ALE-consistent boundary-correction step that updates the fluid-side interface velocity using the predicted structural velocity. A two-stage training strategy (single-step supervised pretraining followed by autoregressive fine-tuning) is used to mitigate error accumulation. The framework is validated on the cylinder-flexible beam benchmark, claiming accurate phase-consistent long rollouts and robust generalization to inlet-profile variations (interpolation and extrapolation), supported by systematic ablations on the ViT module, boundary correction, and long-term training.

Significance. If the empirical results hold with the claimed quantitative support, this work would represent a meaningful advance in surrogate modeling for FSI problems involving moving interfaces and long-term stability. The explicit handling of ALE consistency and the two-stage training address known challenges in autoregressive rollout for coupled physics, and the ablation studies provide a clear assessment of component contributions. Such frameworks could reduce computational cost in engineering applications like aeroelasticity or biomedical flows, provided they demonstrate reliable generalization beyond the specific benchmark.

major comments (2)

[Abstract and Results/Validation sections] The central empirical claim of 'accurate phase-consistent predictions over long rollouts' and 'robust generalization' is load-bearing for the paper's contribution, yet the abstract (and by extension the validation description) supplies no quantitative metrics, error norms, error bars, baseline comparisons, or data details (e.g., mesh resolution, rollout length in time steps, or specific L2 errors). This gap prevents confirmation that the data support the stated performance; the results section must include these to substantiate the claims.
[Method section (ALE boundary correction)] § on the ALE-consistent boundary-correction step: the description states that updating the fluid-side interface velocity with the predicted structural velocity improves near-interface accuracy and long-term stability, but no derivation or proof of ALE consistency (e.g., satisfaction of the geometric conservation law or interface velocity matching) is provided. This is central to the framework's novelty and requires explicit verification or reference to the underlying ALE formulation.

minor comments (3)

[Method] Notation for the GNO-ViT-LSTM architecture and coupling variables should be defined consistently in a single table or early section to avoid ambiguity when reading the partitioned procedure.
[Figures and Results] Figure captions for the benchmark results and ablation studies should explicitly state the rollout horizon, inlet conditions tested, and any error metrics plotted.
[Training strategy] The two-stage training procedure would benefit from a clearer description of the loss functions used in each stage and the number of fine-tuning steps relative to pretraining.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. The comments highlight important areas where the empirical support and theoretical grounding can be strengthened. We address each major comment below and commit to revisions that will improve clarity and substantiation without altering the core contributions.

read point-by-point responses

Referee: [Abstract and Results/Validation sections] The central empirical claim of 'accurate phase-consistent predictions over long rollouts' and 'robust generalization' is load-bearing for the paper's contribution, yet the abstract (and by extension the validation description) supplies no quantitative metrics, error norms, error bars, baseline comparisons, or data details (e.g., mesh resolution, rollout length in time steps, or specific L2 errors). This gap prevents confirmation that the data support the stated performance; the results section must include these to substantiate the claims.

Authors: We agree that the absence of explicit quantitative metrics in the abstract and validation description weakens the ability to evaluate the performance claims. The current manuscript does not report specific L2 errors, rollout lengths, mesh details, error bars, or baseline comparisons in those sections. In the revised manuscript we will update the abstract to include key quantitative indicators (e.g., average L2 norms for fluid and structural fields over long rollouts, mesh resolution, number of time steps, and comparisons to baselines) and expand the validation section with the same metrics, error bars, and data details drawn from our existing experiments. These additions will directly substantiate the claims of phase-consistent long-term predictions and generalization. revision: yes
Referee: [Method section (ALE boundary correction)] § on the ALE-consistent boundary-correction step: the description states that updating the fluid-side interface velocity with the predicted structural velocity improves near-interface accuracy and long-term stability, but no derivation or proof of ALE consistency (e.g., satisfaction of the geometric conservation law or interface velocity matching) is provided. This is central to the framework's novelty and requires explicit verification or reference to the underlying ALE formulation.

Authors: The referee is correct that the manuscript provides no explicit derivation or proof of ALE consistency for the boundary-correction step. Although the correction is motivated by the standard kinematic matching requirement in ALE formulations, we did not derive its consistency with the geometric conservation law or supply numerical verification. In the revision we will insert a dedicated subsection in the Methods section that (i) recalls the relevant ALE geometric conservation law, (ii) derives how the velocity-update step enforces interface matching, and (iii) presents numerical checks confirming conservation properties on the cylinder-beam benchmark. This will make the novelty claim fully rigorous. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper presents an empirical ML architecture for FSI (GNO+ViT fluid surrogate coupled to LSTM structural predictor via partitioned scheme plus ALE boundary-velocity correction and two-stage training). All load-bearing claims are validated by benchmark rollouts, inlet-profile generalization tests, and systematic ablations that isolate the contribution of the boundary-correction step, ViT module, and autoregressive fine-tuning. No equation or modeling choice reduces by construction to its own inputs; the ALE correction is an explicit algorithmic design whose effect is measured externally rather than assumed. No self-citation chain is invoked to justify uniqueness or to substitute for independent evidence. The derivation chain is therefore self-contained against the reported external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The framework rests on standard domain assumptions from FSI and ML time-series modeling; the main empirical addition is the boundary correction whose benefit is shown via ablation.

free parameters (1)

GNO-ViT-LSTM network weights and hyperparameters
Fitted during single-step pretraining and autoregressive fine-tuning to achieve reported accuracy on the benchmark.

axioms (2)

domain assumption Partitioned coupling procedure remains valid when augmented with ALE-consistent velocity correction.
Invoked in the coupling description as the standard way to link fluid and structure surrogates.
domain assumption Two-stage training mitigates autoregressive error accumulation in spatiotemporal rollouts.
Stated as the strategy adopted to improve long-term stability.

pith-pipeline@v0.9.0 · 5537 in / 1463 out tokens · 73838 ms · 2026-05-09T19:11:21.782281+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

55 extracted references · 4 canonical work pages · 1 internal anchor

[1]

Vortex-induced vibration of flexible cylinders in cross-flow

Francisco Huera-Huarte. Vortex-induced vibration of flexible cylinders in cross-flow. Annual Review of Fluid Mechanics, 57, 2025

2025
[2]

Numerical investigation of wave run-up and load on heaving cylinder subjected to regular waves.Ocean Engineering, 268:113415, 2023

Xinlong Wang, Dongsheng Qiao, Linxiao Jin, Jun Yan, Bin Wang, Binbin Li, and Jinping Ou. Numerical investigation of wave run-up and load on heaving cylinder subjected to regular waves.Ocean Engineering, 268:113415, 2023

2023
[3]

Data-driven modeling for unsteady aerodynamics and aeroelasticity.Progress in Aerospace Sciences, 125:100725, 2021

Jiaqing Kou and Weiwei Zhang. Data-driven modeling for unsteady aerodynamics and aeroelasticity.Progress in Aerospace Sciences, 125:100725, 2021

2021
[4]

Yujie Sun, Jiayi Huang, Qingshuang Lu, Xinhai Yue, Xuanming Huang, Wei He, Yun Shi, and Ju Liu. Modeling fibrous tissue in vascular fluid–structure interaction: A morphology- based pipeline and biomechanical significance.International Journal for Numerical Methods in Biomedical Engineering, 41(1):e3892, 2025

2025
[5]

Shunxiang Cao and Daniel Zhengyu Huang. Bayesian calibration for large-scale fluid struc- ture interaction problems under embedded/immersed boundary framework.International Journal for Numerical Methods in Engineering, 123(8):1791–1812, 2022

2022
[6]

Uncertainty quantification in aeroelasticity.Annual review of fluid mechanics, 49(1):361–386, 2017

Philip Beran, Bret Stanford, and Christopher Schrock. Uncertainty quantification in aeroelasticity.Annual review of fluid mechanics, 49(1):361–386, 2017

2017
[7]

Bio-inspiredvariable-stiffnessflapsforhybridflowcontrol, tuned via reinforcement learning.Journal of Fluid Mechanics, 956:R4, 2023

NirmalJNairandAndresGoza. Bio-inspiredvariable-stiffnessflapsforhybridflowcontrol, tuned via reinforcement learning.Journal of Fluid Mechanics, 956:R4, 2023

2023
[8]

A novel framework for predicting active flow control by combining deep reinforcement learning and masked deep neural network.Physics of Fluids, 36(3), 2024

Yangwei Liu, Feitong Wang, Shihang Zhao, and Yumeng Tang. A novel framework for predicting active flow control by combining deep reinforcement learning and masked deep neural network.Physics of Fluids, 36(3), 2024

2024
[9]

Comparative study of rein- forcement learning and reduced-order model-based control for mitigating vortex-induced vibration.Physics of Fluids, 37(4), 2025

Yujia Zhao, Haokui Jiang, Jichao Li, and Shunxiang Cao. Comparative study of rein- forcement learning and reduced-order model-based control for mitigating vortex-induced vibration.Physics of Fluids, 37(4), 2025

2025
[10]

Model predictive control of fluid–structure interaction via koopman-based reduced-order model.Journal of Fluid Mechanics, 1026:A32, 2026

Haokui Jiang and Shunxiang Cao. Model predictive control of fluid–structure interaction via koopman-based reduced-order model.Journal of Fluid Mechanics, 1026:A32, 2026

2026
[11]

Co-flow jet control for strong turbulence over an airfoil based on deep reinforcement learning and attention mechanism.Physics of Fluids, 38(1), 2026

Dongxu Yin, Yumeng Tang, Feitong Wang, Shihang Zhao, and Yangwei Liu. Co-flow jet control for strong turbulence over an airfoil based on deep reinforcement learning and attention mechanism.Physics of Fluids, 38(1), 2026

2026
[12]

Anoverviewofcomputationalfluidstructureinteraction: Methods and applications, 2020

SumantRMorabandAtulSharma. Anoverviewofcomputationalfluidstructureinteraction: Methods and applications, 2020. URLhttps://arxiv.org/abs/2006.04068

work page arXiv 2020
[13]

Added-mass effect in the design of partitioned algorithms for fluid–structure problems.Computer methods in applied mechanics and engineering, 194(42-44):4506–4527, 2005

Paola Causin, Jean-Frédéric Gerbeau, and Fabio Nobile. Added-mass effect in the design of partitioned algorithms for fluid–structure problems.Computer methods in applied mechanics and engineering, 194(42-44):4506–4527, 2005

2005
[14]

Robin-neumann transmission conditions for fluid-structure coupling: embedded boundary implementation and parameter analysis

Shunxiang Cao, Alex Main, and Kevin G Wang. Robin-neumann transmission conditions for fluid-structure coupling: embedded boundary implementation and parameter analysis. International Journal for Numerical Methods in Engineering, 115(5):578–603, 2018. 25

2018
[15]

John Wiley & Sons, 2013

M’hamed Souli and David J Benson.Arbitrary Lagrangian Eulerian and fluid-structure interaction: numerical simulation. John Wiley & Sons, 2013

2013
[16]

Fluid–structure interaction in engineering based on its applications: A plenary and up-to-date review.Physics of Fluids, 36(10), 2024

Muhammad Idrees Afridi, Mohsen Izadi, Mohammad Eizadi, Hussein A Mohammed, Reza Karimi, Faris Alqurashi, Mohamed H Mohamed, Ahmad Hajjar, Mohamed Bechir Ben Hamida, Ibrahim Mahariq, et al. Fluid–structure interaction in engineering based on its applications: A plenary and up-to-date review.Physics of Fluids, 36(10), 2024

2024
[17]

Machine learning for fluid mechanics.Annual review of fluid mechanics, 52(1):477–508, 2020

Steven L Brunton, Bernd R Noack, and Petros Koumoutsakos. Machine learning for fluid mechanics.Annual review of fluid mechanics, 52(1):477–508, 2020

2020
[18]

Deep learning in computational mechanics: a review.Computational Mechanics, 74(2):281–331, 2024

Leon Herrmann and Stefan Kollmannsberger. Deep learning in computational mechanics: a review.Computational Mechanics, 74(2):281–331, 2024

2024
[19]

Modal analysis of fluid flows: An overview.AIAA journal, 55(12): 4013–4041, 2017

Kunihiko Taira, Steven L Brunton, Scott TM Dawson, Clarence W Rowley, Tim Colonius, Beverley J McKeon, Oliver T Schmidt, Stanislav Gordeyev, Vassilios Theofilis, and Lawrence S Ukeiley. Modal analysis of fluid flows: An overview.AIAA journal, 55(12): 4013–4041, 2017

2017
[20]

Harsh Sharma, David A Najera-Flores, Michael D Todd, and Boris Kramer. Lagrangian operator inference enhanced with structure-preserving machine learning for nonintrusive model reduction of mechanical systems.Computer Methods in Applied Mechanics and Engineering, 423:116865, 2024

2024
[21]

Aviral Prakash and Yongjie Jessica Zhang. Projection-based reduced order modeling and data-driven artificial viscosity closures for incompressible fluid flows.Computer Methods in Applied Mechanics and Engineering, 425:116930, 2024

2024
[22]

Galerkin-free technique for the reduced-order modeling of fluid-structure interaction via machine learning

Matthew J Whisenant and Kivanc Ekici. Galerkin-free technique for the reduced-order modeling of fluid-structure interaction via machine learning. InAIAA Scitech 2020 Forum, page 1637, 2020

2020
[23]

Balanced proper-orthogonal-decomposition-based feedback control of vortex-induced vibration.Physical Review Fluids, 9(7):073901, 2024

Haokui Jiang and Shunxiang Cao. Balanced proper-orthogonal-decomposition-based feedback control of vortex-induced vibration.Physical Review Fluids, 9(7):073901, 2024

2024
[24]

Koopman reduced-order modeling and analysis of flag flapping in the wake of a cylinder.Physical Review E, 111(4):045101, 2025

Haokui Jiang, Jean-Lou Pfister, Daniel Zhengyu Huang, and Shunxiang Cao. Koopman reduced-order modeling and analysis of flag flapping in the wake of a cylinder.Physical Review E, 111(4):045101, 2025

2025
[25]

A novel method for predicting fluid–structure interaction with large deformation based on masked deep neural network.Physics of Fluids, 36(2), 2024

Yangwei Liu, Shihang Zhao, Feitong Wang, and Yumeng Tang. A novel method for predicting fluid–structure interaction with large deformation based on masked deep neural network.Physics of Fluids, 36(2), 2024

2024
[26]

Differentiable hybrid neural modeling for fluid-structure interaction.Journal of Computational Physics, 496:112584, 2024

Xiantao Fan and Jian-Xun Wang. Differentiable hybrid neural modeling for fluid-structure interaction.Journal of Computational Physics, 496:112584, 2024

2024
[27]

Pretraining codomain attention neural operators for solving multiphysics pdes.Advances in Neural Information Processing Systems, 37:104035–104064, 2024

Md Ashiqur Rahman, Robert Joseph George, Mogab Elleithy, Daniel Leibovici, Zongyi Li, Boris Bonev, Colin White, Julius Berner, Raymond A Yeh, Jean Kossaifi, et al. Pretraining codomain attention neural operators for solving multiphysics pdes.Advances in Neural Information Processing Systems, 37:104035–104064, 2024. 26

2024
[28]

Mesh-based super-resolution of fluidflowswithmultiscalegraphneuralnetworks.ComputerMethodsinAppliedMechanics and Engineering, 443:118072, 2025

Shivam Barwey, Pinaki Pal, Saumil Patel, Riccardo Balin, Bethany Lusch, Venkatram Vishwanath, Romit Maulik, and Ramesh Balakrishnan. Mesh-based super-resolution of fluidflowswithmultiscalegraphneuralnetworks.ComputerMethodsinAppliedMechanics and Engineering, 443:118072, 2025

2025
[29]

Enhancing cfd solver with machine learning techniques.Computer Methods in Applied Mechanics and Engineering, 429:117133, 2024

Paulo Sousa, Carlos Veiga Rodrigues, and Alexandre Afonso. Enhancing cfd solver with machine learning techniques.Computer Methods in Applied Mechanics and Engineering, 429:117133, 2024

2024
[30]

Machine learning–accelerated computational fluid dynamics.Proceedings of the National Academy of Sciences, 118(21):e2101784118, 2021

DmitriiKochkov,JamieASmith,AyyaAlieva,QingWang,MichaelPBrenner,andStephan Hoyer. Machine learning–accelerated computational fluid dynamics.Proceedings of the National Academy of Sciences, 118(21):e2101784118, 2021

2021
[31]

Synergizing machine learning with fluid–structure interaction research: An overview of trends and challenges

Muk Chen Ong and Guang Yin. Synergizing machine learning with fluid–structure interaction research: An overview of trends and challenges. InOcean, volume 1. Tsinghua Press, 2025

2025
[32]

Machine-learning enhanced predictors for accelerated convergence of partitioned fluid-structure interaction simulations.Computer Physics Communications, 310:109522, 2025

Azzeddine Tiba, Thibault Dairay, Florian De Vuyst, Iraj Mortazavi, and Juan Pedro Berro Ramirez. Machine-learning enhanced predictors for accelerated convergence of partitioned fluid-structure interaction simulations.Computer Physics Communications, 310:109522, 2025

2025
[33]

A novel spatial-temporal prediction method for unsteady wake flows based on hybrid deep neural network.Physics of Fluids, 31(12), 2019

Renkun Han, Yixing Wang, Yang Zhang, and Gang Chen. A novel spatial-temporal prediction method for unsteady wake flows based on hybrid deep neural network.Physics of Fluids, 31(12), 2019

2019
[34]

Machine learning for modelling unstructured grid data in computational physics: a review.Information Fusion, page 103255, 2025

Sibo Cheng, Marc Bocquet, Weiping Ding, Tobias Sebastian Finn, Rui Fu, Jinlong Fu, Yike Guo, Eleda Johnson, Siyi Li, Che Liu, et al. Machine learning for modelling unstructured grid data in computational physics: a review.Information Fusion, page 103255, 2025

2025
[35]

Deep learning of vortex-induced vibrations.Journal of fluid mechanics, 861:119–137, 2019

MaziarRaissi,ZhichengWang,MichaelSTriantafyllou,andGeorgeEmKarniadakis. Deep learning of vortex-induced vibrations.Journal of fluid mechanics, 861:119–137, 2019

2019
[36]

Predicting fluid-structure interaction with graph neural networks.Physics of Fluids, 36(1), 2024

Rui Gao and Rajeev K Jaiman. Predicting fluid-structure interaction with graph neural networks.Physics of Fluids, 36(1), 2024

2024
[37]

Data-driven nonlinear reduced-order modeling of unsteady fluid–structure interactions.Physics of Fluids, 34(5), 2022

XinshuaiZhang,TingweiJi,FangfangXie,ChangdongZheng,andYaoZheng. Data-driven nonlinear reduced-order modeling of unsteady fluid–structure interactions.Physics of Fluids, 34(5), 2022

2022
[38]

A transfer learning-physics informed neural network (tl-pinn) for vortex-induced vibration.Ocean Engineering, 266: 113101, 2022

Hesheng Tang, Yangyang Liao, Hu Yang, and Liyu Xie. A transfer learning-physics informed neural network (tl-pinn) for vortex-induced vibration.Ocean Engineering, 266: 113101, 2022

2022
[39]

Physics-informed neural networks for prediction of a flow-induced vibration cylinder.Journal of Offshore Mechanics and Arctic Engineering, 146(6):061203, 2024

Guang Yin, Marek Jan Janocha, and Muk Chen Ong. Physics-informed neural networks for prediction of a flow-induced vibration cylinder.Journal of Offshore Mechanics and Arctic Engineering, 146(6):061203, 2024

2024
[40]

A projection-based time- segmented reduced order model for fluid-structure interactions.Journal of Computational Physics, 520:113481, 2025

Qijia Zhai, Shiquan Zhang, Pengtao Sun, and Xiaoping Xie. A projection-based time- segmented reduced order model for fluid-structure interactions.Journal of Computational Physics, 520:113481, 2025. 27

2025
[41]

Conditional neural field latent diffusion model for generating spatiotemporal turbulence.Nature Communications, 15(1):10416, 2024

PanDu,MeetHemantParikh,XiantaoFan,Xin-YangLiu,andJian-XunWang. Conditional neural field latent diffusion model for generating spatiotemporal turbulence.Nature Communications, 15(1):10416, 2024

2024
[42]

Text2pde: Latent diffusion models for accessible physics simulation.arXiv preprint arXiv:2410.01153, 2024

Anthony Zhou, Zijie Li, Michael Schneier, John R Buchanan Jr, and Amir Barati Farimani. Text2pde: Latent diffusion models for accessible physics simulation.arXiv preprint arXiv:2410.01153, 2024

work page arXiv 2024
[43]

Fourier neural operator based fluid–structure interaction for predicting the vesicle dynamics.Physica D: Nonlinear Phenomena, 463:134145, 2024

Wang Xiao, Ting Gao, Kai Liu, Jinqiao Duan, and Meng Zhao. Fourier neural operator based fluid–structure interaction for predicting the vesicle dynamics.Physica D: Nonlinear Phenomena, 463:134145, 2024

2024
[44]

A hybrid partitioned deep learning methodology for moving interface and fluid–structure interaction.Computers & Fluids, 233:105239, 2022

Rachit Gupta and Rajeev Jaiman. A hybrid partitioned deep learning methodology for moving interface and fluid–structure interaction.Computers & Fluids, 233:105239, 2022

2022
[45]

Deep neural network based reduced-order model for fluid–structure interaction system

Renkun Han, Yixing Wang, Weiqi Qian, Wenzheng Wang, Miao Zhang, and Gang Chen. Deep neural network based reduced-order model for fluid–structure interaction system. Physics of Fluids, 34(7), 2022

2022
[46]

Physics-embedded neural networks: Graph neural pde solvers with mixed boundary conditions.Advances in Neural Information Processing Systems, 35:23218–23229, 2022

Masanobu Horie and Naoto Mitsume. Physics-embedded neural networks: Graph neural pde solvers with mixed boundary conditions.Advances in Neural Information Processing Systems, 35:23218–23229, 2022

2022
[47]

Learning mesh-based simulation with graph networks

Tobias Pfaff, Meire Fortunato, Alvaro Sanchez-Gonzalez, and Peter Battaglia. Learning mesh-based simulation with graph networks. InInternational conference on learning representations, 2020

2020
[48]

Afiniteelement-inspiredhypergraphneural network: Application to fluid dynamics simulations.Journal of Computational Physics, 504:112866, 2024

RuiGao,InduKantDeo,andRajeevKJaiman. Afiniteelement-inspiredhypergraphneural network: Application to fluid dynamics simulations.Journal of Computational Physics, 504:112866, 2024

2024
[49]

Neural operator: Graph kernel network for partial differential equations

Anima Anandkumar, Kamyar Azizzadenesheli, Kaushik Bhattacharya, Nikola Kovachki, Zongyi Li, Burigede Liu, and Andrew Stuart. Neural operator: Graph kernel network for partial differential equations. InICLR 2020 workshop on integration of deep neural models and differential equations, 2020

2020
[50]

Geometry-informedneuraloperatorforlarge-scale3dpdes.AdvancesinNeuralInformation Processing Systems, 36:35836–35854, 2023

Zongyi Li, Nikola Kovachki, Chris Choy, Boyi Li, Jean Kossaifi, Shourya Otta, Moham- mad Amin Nabian, Maximilian Stadler, Christian Hundt, Kamyar Azizzadenesheli, et al. Geometry-informedneuraloperatorforlarge-scale3dpdes.AdvancesinNeuralInformation Processing Systems, 36:35836–35854, 2023

2023
[51]

Springer, 2006

Stefan Turek and Jaroslav Hron.Proposal for numerical benchmarking of fluid-structure interaction between an elastic object and laminar incompressible flow. Springer, 2006

2006
[52]

An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

A. Dosovitskiy and et al. An image is worth 16x16 words: Transformers for image recognition at scale.arXiv:2010.11929, 2020

work page internal anchor Pith review Pith/arXiv arXiv 2010
[53]

Long short-term memory.Neural Computation, 9(8):1735–1780, 1997

Sepp Hochreiter and Jürgen Schmidhuber. Long short-term memory.Neural Computation, 9(8):1735–1780, 1997. 28

1997
[54]

Data-driven nonlinear parametric model order reduction framework using deep hierarchical variational autoencoder.Engineering with Computers, 40(4):2385–2400, 2024

Sihun Lee, Sangmin Lee, Kijoo Jang, Haeseong Cho, and Sangjoon Shin. Data-driven nonlinear parametric model order reduction framework using deep hierarchical variational autoencoder.Engineering with Computers, 40(4):2385–2400, 2024

2024
[55]

Supplementary videos s1–s2 for: An ALE-consistent graph neural operator- Transformer framework for fluid-structure interaction

Shihang Zhao. Supplementary videos s1–s2 for: An ALE-consistent graph neural operator- Transformer framework for fluid-structure interaction. Zenodo, 2026. URL https: //doi.org/10.5281/zenodo.18458033. 29

work page doi:10.5281/zenodo.18458033 2026