Pith. sign in

REVIEW 4 major objections 4 minor 49 references

Generative Spatio-temporal GraphNet for Transonic Wing Pressure Distribution Forecasting

T0 review · 4 major / 4 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-convolutional autoencoder with a temporal layer can forecast unsteady transonic wing pressures with accuracy comparable to CFD — about one percent error — at a fraction of the cost.

desk verdict A competent integration of known components for unsteady wing pressure prediction, but the evidence skips its own test set and the 'generative' label overclaims. read the letter →

arxiv 2411.11592 v1 pith:ZL2C6QO4 submitted 2024-11-18 cs.LG cs.CE

Gabriele Immordino , Andrea Vaiuso , Andrea Da Ronch , Marcello Righi This is my paper · ORCID
classification cs.LGcs.CE
keywords graphneuralnetworksautoencoderunsteadytransonicaerodynamicspressurecoefficientforecastingreduced-ordermodelbenchmarksupercriticalwingspatio-temporal
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

This paper tries to establish that a machine-learning surrogate can replace computational fluid dynamics (CFD) for a hard, specific task: forecasting the unsteady pressure distribution on a transonic wing in which shock motion and boundary-layer separation dominate. The proposed GST GraphNet couples a pre-trained autoencoder, built from graph convolutional layers that pool and unpool the 86,840-node surface mesh, with a temporal layer that evolves the compressed pressure field forward in time. On the Benchmark Super Critical Wing (BSCW), the surrogate predicts two held-out validation signals with mean absolute percentage error around 0.75 to 1.0 percent and $R^2$ close to 0.99, while a single prediction takes about two minutes on one GPU against roughly 6,000 CPU-hours per CFD run. If the claim holds, unsteady transonic analysis for flutter studies and design loops becomes cheap enough to run interactively, and the dominant cost shifts from prediction to one-time dataset generation.

What carries the argument

The load-bearing object is the GST GraphNet, a two-stage surrogate whose first stage is a pre-trained autoencoder made of graph convolutional layers with gradient-aware pooling and unpooling, and whose second stage is a graph-based temporal layer (GRU, LSTM, attention, or STGCN) acting on the compressed latent space. Pooling keeps nodes with large pressure gradients, where shocks and separation live, and drops low-gradient nodes; the reduced graph is reconnected with the Mahalanobis distance and interpolated by moving weighted least squares, shrinking the mesh from 86,840 to 9,600 nodes before forecasting. The temporal layers are graph-convolutional versions of standard recurrent cells, and the STGCN variant convolves along the time axis with a gated linear unit, $\Gamma *_T Y = P \odot \sigma(Q)$, and this layer together with the LSTM variant yields the lowest forecast errors. Training uses the loss $\mathrm{Loss} = \mathrm{MAE} + \lambda \cdot C_{My}$ with $\lambda = 0.01$, adding a pitching-moment penalty, and backpropagation through time over mini-sequences of length three.

What would settle it

Retrain the surrogate with training data kept at the full $2 \times 10^{-4}$ s CFD timestep instead of the downsampled $2 \times 10^{-3}$ s and compare validation MAPE and $R^2$; if coarse sampling aliases shock dynamics, the reported errors should change measurably. A second check: run one validation signal with frequency content above the Nyquist limit of the downsampled training data and compare the predicted pressure field near the shock against the CFD reference — if the surrogate still tracks the shock, the low-cost claim survives; if not, “accuracy comparable to CFD” holds only inside the training band.

Watch

Extended reading notes

Core claim

The central claim, stated in the conclusions, is that the integrated autoencoder–graph–temporal approach “can achieve an accuracy comparable to traditional CFD methods, while significantly reducing computational costs.” Concretely, the model compresses a surface pressure field defined on 86,840 mesh nodes into a 9,600-node latent space by keeping high-pressure-gradient regions — shocks and separation — at higher resolution, then steps the latent field forward three timesteps with a graph-based temporal layer and decodes back to the full mesh. Across the temporal layers compared, LSTM and STGCN give the lowest errors, with the feedforward architecture (which never feeds back its own predictions) consistently more stable than the ARMAX architecture, which accumulates error once ground-truth pressures are withdrawn from its input. The reported validation results — MAPE around 0.75 to 1.0 percent with $R^2$ close to 0.99, and a per-prediction cost of about two minutes on an NVIDIA RTX A4000 versus about 6,000 CPU-hours per CFD run — are the quantitative basis for the claim.

Load-bearing premise

The claim rests on the assumption that four damped Schroeder-phased training maneuvers, spanning a narrow range of pitch and plunge amplitudes and frequencies and sampled at $2 \times 10^{-3}$ s (ten times coarser than the CFD timestep), capture the wing's shock and separation dynamics well enough that the two validation signals are genuine generalization tests rather than near-duplicates of the training data.

Editorial extensions

If this is right

  • Unsteady transonic pressure fields on the BSCW can be produced at about 0.03 GPU-hours per prediction instead of roughly 6,000 CPU-hours per CFD run, with validation errors around one percent MAPE and $R^2$ near 0.99, which is fast enough to survey many maneuvers or run inside a design loop.
  • The feedforward architecture is the safer default for time-marching forecasts, because the ARMAX variant's feedback of its own predictions compounds error once ground truth is removed; ARMAX with ground-truth inputs can still beat feedforward, so it remains useful when reliable inputs are available.
  • The choice of temporal layer matters most under fast oscillations: on the single-harmonic validation signal, LSTM and STGCN keep MAPE near one percent while GRU and attention climb to roughly 1.5 to 1.8 percent in the feedforward model.
  • Dataset generation, not inference, becomes the dominant cost — about 75,000 CPU-hours for the 12 CFD runs here versus 0.03 GPU-hours per prediction — so further savings must come from reducing the amount of high-fidelity training data, as the paper itself concludes.

Reading between the lines

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

  • My inference: although validation signal 2 (single harmonic, 3-degree pitch amplitude) lies outside the training amplitude range, the training envelope is still narrow — four damped Schroeder signals — so the strongest defensible reading is “comparable to CFD within a limited motion-parameter regime”; a cross-regime test such as a different Mach number or a much larger amplitude would show how far
  • My inference: because training data were downsampled from $2 \times 10^{-4}$ s to $2 \times 10^{-3}$ s, forecast skill near the coarse Nyquist frequency is untested; validating against full-rate CFD outputs, or on signals with spectral content close to that limit, would reveal whether the reported one-percent error hides aliased shock dynamics.
  • My inference: the same two-stage recipe — gradient-aware pooling onto a small latent graph, temporal forecasting, unpooling — should transfer to other integrated loads such as lift or hinge moments by replacing the pitching-moment penalty with the corresponding target quantity.
  • My inference: the cost comparison implies that adopting such surrogates at scale is bottlenecked by the creation of high-fidelity training data, making multi-fidelity sampling or active learning a natural next test rather than a distant wish.
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

4 major / 4 minor

Summary. The paper proposes Generative Spatio-temporal GraphNet (GST GraphNet), a hybrid autoencoder/graph-convolutional architecture with interchangeable temporal layers (GRU, LSTM, attention, STGCN) for forecasting unsteady pressure-coefficient distributions on the Benchmark Super Critical Wing. The pipeline pre-trains an AE on four damped Schroeder-phased training signals, compresses the surface mesh into a latent graph, applies a temporal layer to sequences of length three, and decodes the result to predict the next pressure field. Two model families are compared: a feedforward model and an ARMAX model that feeds back its own predictions. Validation is conducted on two signals (one damped Schroeder, one single-harmonic), with MAPE, R2, and RMSE reported for CP, plus comparisons of CL and CM against CFD. The paper reports about two minutes of GPU time per prediction versus roughly 6,000 CPU hours per CFD run and concludes that the framework achieves accuracy comparable to CFD.

Significance. If the reported accuracy holds, the framework would be a practically valuable fast surrogate for unsteady transonic pressure prediction on a relevant aeroelastic test case. The paper's strengths include the detailed architecture description in Appendix B, the systematic comparison of temporal layers, the explicit accounting of CFD training-data cost in Table 6, and the use of a recognized benchmark configuration. It also adopts several sensible design choices, including pre-training of the autoencoder, pressure-gradient-guided pooling, and BPTT for the autoregressive variant. However, the evidence base is narrow: the central claim rests on metrics from two validation signals only, with no test-set results, no confidence intervals or repeated-seed statistics, and no external surrogate baseline. The name 'generative' is not justified by any probabilistic or sampling component. Reproducibility is further limited by the absence of released code, data, or trained models.

major comments (4)
  1. [Section 3, Table 3; Section 4, Tables 4 and 5] Section 3 (Table 3) defines a six-signal test set (Test 1–6) and states that it is used to 'accurately assess the model accuracy and sensitivity,' but Section 4 reports quantitative results only for Validation 1 and Validation 2. No MAPE, R2, or RMSE values, and no figures, are provided for Test 1–6, which include signal types the model did not see in training (US, pitch-only, plunge-only). Without those results, the paper's own evaluation protocol is incomplete, and the central claim of 'accuracy comparable to CFD' rests entirely on two validation signals. The authors should report the Test 1–6 results or explain why they were excluded.
  2. [Section 4.3 and Table 4] Section 4.3 states that the STGCN temporal layer was selected 'because it consistently yielded the most accurate results across both validation signals,' but Table 4 shows that in the feedforward model LSTM outperforms STGCN on both signals in MAPE (0.7471 vs 0.8524 for DS; 0.9695 vs 0.9975 for SH) and R2 (0.9937 vs 0.9918 for DS; 0.9909 vs 0.9897 for SH). This inconsistency needs to be resolved, and if validation signals were used to select the temporal layer, the reported metrics are selection-biased and should be supplemented by test-set or nested-validation results.
  3. [Section 4.3, Tables 4 and 5] Section 4.3 describes an ARMAX evaluation in which ground-truth CP values are used for the first half of the signal before switching to self-predicted values. It is unclear whether the MAPE, R2, and RMSE values in Table 5 were computed under this hybrid protocol or under a fully closed-loop rollout. The two protocols can give very different numbers, and the hybrid protocol depends on the arbitrary switching point. The authors should define the evaluation protocol for Table 5 and report closed-loop and teacher-forced errors separately.
  4. [Tables 4 and 5] Tables 4 and 5 report single point estimates for each temporal layer and architecture, with no confidence intervals, standard deviations across training seeds, or multiple initializations. The differences between the top performing temporal layers are sometimes small (for example, LSTM versus STGCN in Table 4), so the ranking and the 'comparable to CFD' conclusion are not statistically grounded as presented. Repeated-seed experiments or an equivalent uncertainty quantification should be reported.
minor comments (4)
  1. [Title, Abstract, Section 2.1] The model is called 'generative,' but the paper describes only deterministic regression through an autoencoder and temporal layers; no likelihood, sampling, or generative objective is defined. Please either remove the term or add the corresponding component.
  2. [Section 2.3, Eq. (4)] The probability function p(i) reuses the symbol p used for pressure, and the parameters p1 and pn are not defined. Please introduce distinct notation and specify how the values are chosen.
  3. [Section 2.1] The loss is written as MAE + λ·CMy with λ=0.01 'for dimensional consistency,' but both MAE on CP and CMy are dimensionless; the rationale for this term should be clarified or rephrased.
  4. [Table 6] The comparison mixes CPU hours (CFD) with GPU hours (model) without stating the hardware configuration for the GPU run or whether the two-minute prediction includes the full decoding and I/O pipeline; a footnote would clarify the comparison.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular reductions: forecasting accuracy is measured against held-out CFD signals, and the cited prior work supplies components, not the central claim.

full rationale

The central claim is an empirical benchmark: the model maps past motion/coordinate (and, in the ARMAX variant, past predicted pressure) inputs to a future CP distribution and is scored by MAPE/R2/RMSE on Validation 1 (DS) and Validation 2 (SH), which are not part of the training set defined in Table 3. No equation in the paper defines the target CP_t as an alias of an input or of a fitted parameter; the architecture (Eqs. 1-7, Tables B.7-B.8) is a standard AE+GCN+temporal encoder-decoder. The CMy term in the loss is a physical regularization, not a quantity that identifies the prediction with the input. Self-citations [24] and [5] provide the pooling/unpooling scheme and Bayesian-optimization recipe, but the paper's own held-out CFD comparisons carry the load; there is no uniqueness theorem or ansatz smuggled in via citation that forces the reported numbers. The manuscript does have evaluation weaknesses: the test set defined in Section 3 (Table 3, Test 1-6) is never reported, validation signals were also used to compare temporal layers and select hyperparameters, and the conclusion that STGCN is 'most accurate' is not fully supported by Table 4. These are correctness and generalization concerns, not circular reductions, so they do not increase the circularity score.

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

The framework's performance depends on several hand-chosen hyperparameters and on domain assumptions about the CFD data, the downsampled timestep, and the gradient-based pooling strategy. These are empirically selected rather than derived from first principles. No new physical entities are introduced.

free parameters (5)
  • loss penalty weight lambda = 0.01
    Added to MAE loss as lambda times CMy, chosen by hand for dimensional consistency in Section 2.1.
  • temporal sequence length n = 3
    Input includes n=3 previous timesteps, chosen based on performance in Section 2.1.
  • data augmentation noise level and fraction = Gaussian noise, 10% standard deviation, 30% augmentation
    Used in AE pre-training to improve robustness, chosen by hand in Section 2.4.
  • AE architecture hyperparameters and pooling node counts = not reported
    Selected by Bayesian optimization following [24]; pooling node counts 28,600 and 9,600 appear in Appendix B without selection criteria in this paper.
  • training hyperparameters (learning rate, batch size, epochs) = 0.001, 1, 50
    Set manually as described in Appendix B.
assumptions (5)
  • domain assumption URANS with the Spalart-Allmaras turbulence model in SU2 provides accurate ground truth for unsteady transonic flow.
    All training and validation labels come from these CFD simulations (Section 3); if the CFD is inaccurate, 'comparable to CFD' is meaningless.
  • domain assumption Downsampling the CFD timestep from 2e-4 s to 2e-3 s preserves the shock and separation dynamics.
    Section 3 states the timestep was reduced for training, but no temporal resolution study is reported.
  • domain assumption The four damped Schroeder-phased training signals are representative of the unsteady flow regime.
    Table 3 lists only four training signals; generalization to the two validation signals is claimed in Section 4 without a coverage analysis.
  • ad hoc to paper Pressure-gradient based pooling and MWLS interpolation preserve the information needed for unsteady forecasting.
    Section 2.3 describes pooling based on pressure gradients, but for moving shocks the gradient field changes over time and the paper does not analyze how pooling affects latent-space temporal consistency.
  • domain assumption The pre-trained autoencoder reconstructs CP accurately enough that decoder errors do not dominate forecasting errors.
    Section 2.4 describes pre-training, but no AE reconstruction error is reported in this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Generative Spatio-temporal GraphNet for Transonic Wing Pressure Distribution Forecasting." pith.science (2026). https://pith.science/paper/ZL2C6QO4

@misc{pith2026241111592,
  author       = {Pith},
  title        = {Pith review of: Generative Spatio-temporal GraphNet for Transonic Wing Pressure Distribution Forecasting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZL2C6QO4}},
  note         = {Machine review of arXiv:2411.11592}
}
read the original abstract

This study presents a framework for predicting unsteady transonic wing pressure distributions, integrating an autoencoder architecture with graph convolutional networks and graph-based temporal layers to model time dependencies. The framework compresses high-dimensional pressure distribution data into a lower-dimensional latent space using an autoencoder, ensuring efficient data representation while preserving essential features. Within this latent space, graph-based temporal layers are employed to predict future wing pressures based on past data, effectively capturing temporal dependencies and improving predictive accuracy. This combined approach leverages the strengths of autoencoders for dimensionality reduction, graph convolutional networks for handling unstructured grid data, and temporal layers for modeling time-based sequences. The effectiveness of the proposed framework is validated through its application to the Benchmark Super Critical Wing test case, achieving accuracy comparable to computational fluid dynamics, while significantly reducing prediction time. This framework offers a scalable, computationally efficient solution for the aerodynamic analysis of unsteady phenomena.

Figures

Figures reproduced from arXiv: 2411.11592 by the authors.

Figure 1
Figure 1. Overview of the GST GraphNet architecture for predicting wing pressure distributions. [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Diagram of the pooling and unpooling modules used in the AE for dimensionality reduc [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Schematic of the pre-trained AE architecture for compressing and reconstructing the [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figures from the paper (16 more)
Figure 4
Figure 4. Figure 4: Impression of the BSCW CFD grid. ditions where these variables exhibit both positive and negative values ( [PITH_FULL_IMAGE:figures/full_fig_p015_4.png]
Figure 5
Figure 5. Figure 5: Training signal 1: DS with κθ = 0.114, aθ = 0.80 [deg], κξ = 0.152, and aξ = −0.098 [m]. 4. Results In this section, we present the results of the reconstructed validation signals for two different types of architectures: feedforward model and ARMAX model. By comparing…
Figure 6
Figure 6. Figure 6: Validation signal 1 - DS type: Effect of temporal layer selection on [PITH_FULL_IMAGE:figures/full_fig_p018_6.png]
Figure 7
Figure 7. Figure 7: Validation signal 1 - DS type: Effect of temporal layer selection on [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]
Figure 8
Figure 8. Figure 8: Validation signal 2 - SH type: Effect of temporal layer selection on [PITH_FULL_IMAGE:figures/full_fig_p020_8.png]
Figure 9
Figure 9. Figure 9: Validation signal 2 - SH type: Effect of temporal layer selection on [PITH_FULL_IMAGE:figures/full_fig_p021_9.png]
Figure 10
Figure 10. Figure 10: Validation signal 1 - DS type: Impact of temporal layer selection on [PITH_FULL_IMAGE:figures/full_fig_p022_10.png]
Figure 11
Figure 11. Figure 11: Validation signal 1 - DS type: Impact of temporal layer selection on [PITH_FULL_IMAGE:figures/full_fig_p024_11.png]
Figure 12
Figure 12. Figure 12: Validation signal 2 - SH type: Impact of temporal layer selection on [PITH_FULL_IMAGE:figures/full_fig_p025_12.png]
Figure 13
Figure 13. Figure 13: Validation signal 2 - SH type: Impact of temporal layer selection on [PITH_FULL_IMAGE:figures/full_fig_p027_13.png]
Figure 14
Figure 14. Figure 14: Validation signal 1 - DS type: ARMAX vs Feedforward model using STGCN temporal [PITH_FULL_IMAGE:figures/full_fig_p029_14.png]
Figure 15
Figure 15. Figure 15: Validation signal 1 - DS type: Evolution of MAPE of [PITH_FULL_IMAGE:figures/full_fig_p029_15.png]
Figure 16
Figure 16. Figure 16: Validation signal 1 - DS type: ARMAX vs Feedforward model using STGCN temporal [PITH_FULL_IMAGE:figures/full_fig_p030_16.png]
Figure 17
Figure 17. Figure 17: Validation signal 2 - SH type: ARMAX vs Feedforward model using STGCN temporal [PITH_FULL_IMAGE:figures/full_fig_p031_17.png]
Figure 18
Figure 18. Figure 18: Validation signal 2 - SH type: Evolution of MAPE of [PITH_FULL_IMAGE:figures/full_fig_p032_18.png]
Figure 19
Figure 19. Figure 19: Validation signal 2 - SH type: ARMAX vs Feedforward model using STGCN temporal [PITH_FULL_IMAGE:figures/full_fig_p033_19.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 16 canonical work pages

  1. [1]

    J. D. Anderson, J. Wendt, Computational fluid dynamics, volume 206, Springer, 1995

    work page 1995

  2. [2]

    Blazek, Computational fluid dynamics: principles and applications, Butterworth-Heinemann, 2015

    J. Blazek, Computational fluid dynamics: principles and applications, Butterworth-Heinemann, 2015. doi:10.1016/C2013-0-19038-1

    work page doi:10.1016/c2013-0-19038-1 2015

  3. [3]

    Sabater, P

    C. Sabater, P. St¨ urmer, P. Bekemeyer, Fast predictions of aircraft aerodynamics using deep-learning techniques, AIAA Journal 60 (2022) 5249–5261. doi: 10. 2514/1.J061234. 36 Encoding A Layer Type Output Size Input m × 3 × 86840 × 8 GCN m × 3 × 86840 × 256 GCN m × 3 × 86840 × 224 GCN m × 3 × 86840 × 96 Pooling 1 m × 3 × 28600 × 96 GCN m × 3 × 28600 × 64 ...

    work page 2022

  4. [4]

    Castellanos, J

    R. Castellanos, J. B. Varela, A. Gorgues, E. Andr´ es, An assessment of reduced- order and machine learning models for steady transonic flow prediction on wings, in: ICAS 2022, 2022

    work page 2022

  5. [5]

    Immordino, A

    G. Immordino, A. Da Ronch, M. Righi, Steady-state transonic flowfield predic- tion via deep-learning framework, AIAA Journal (2024) 1–17. doi: 10.2514/1. J063545

    doi:10.2514/1 2024

  6. [6]

    Tompson, K

    J. Tompson, K. Schlachter, P. Sprechmann, K. Perlin, Accelerating eulerian fluid simulation with convolutional networks, in: International conference on machine learning, PMLR, 2017, pp. 3424–3433. doi: 10.48550/arXiv.1607.03597

    arXiv doi:10.48550/arxiv.1607.03597 2017

  7. [7]

    M. M. Bronstein, J. Bruna, Y. LeCun, A. Szlam, P. Vandergheynst, Geometric deep learning: going beyond euclidean data, IEEE Signal Processing Magazine 34 (2017) 18–42. doi: 10.1109/MSP.2017.2693418

    arXiv 2017

  8. [8]

    M. Gori, G. Monfardini, F. Scarselli, A new model for learning in graph domains, 37 in: Proceedings. 2005 IEEE International Joint Conference on Neural Networks, 2005., volume 2, IEEE, 2005, pp. 729–734. doi: 10.1109/IJCNN.2005.1555942

    arXiv 2005

Show all 49 references
  1. [9]

    Scarselli, M

    F. Scarselli, M. Gori, A. C. Tsoi, M. Hagenbuchner, G. Monfardini, The graph neural network model, IEEE transactions on neural networks 20 (2008) 61–80. doi:10.1109/TNN.2008.2005605

    2008

  2. [10]

    Z. Wu, S. Pan, F. Chen, G. Long, C. Zhang, S. Y. Philip, A comprehensive survey on graph neural networks, IEEE transactions on neural networks and learning systems 32 (2020) 4–24. doi: 10.1109/TNNLS.2020.2978386

    2020

  3. [11]

    Zhang, P

    Z. Zhang, P. Cui, W. Zhu, Deep learning on graphs: A survey, IEEE Trans- actions on Knowledge and Data Engineering 34 (2020) 249–270. doi: 10.1109/ TKDE.2020.2981333

    2020

  4. [12]

    J. Zhou, G. Cui, S. Hu, Z. Zhang, C. Yang, Z. Liu, L. Wang, C. Li, M. Sun, Graph neural networks: A review of methods and applications, AI open 1 (2020) 57–81. doi:10.1016/j.aiopen.2021.01.001

    2020 doi

  5. [13]

    T. N. Kipf, M. Welling, Semi-supervised classification with graph convolutional networks, arXiv preprint arXiv:1609.02907 (2016). doi: 10.48550/arXiv.1609. 02907

    2016 doi arXiv

  6. [14]

    X. Jin, P. Cheng, W.-L. Chen, H. Li, Prediction model of velocity field around circular cylinder over various reynolds numbers by fusion convolutional neural networks based on pressure on the cylinder, Physics of Fluids 30 (2018) 047105. doi:10.1063/1.5024595

    2018 doi

  7. [15]

    Fukami, K

    K. Fukami, K. Fukagata, K. Taira, Super-resolution reconstruction of turbulent flows with machine learning, Journal of Fluid Mechanics 870 (2019) 106–120. doi:10.1017/jfm.2019.238

    2019 doi

  8. [16]

    Omata, S

    N. Omata, S. Shirayama, A novel method of low-dimensional representation for temporal behavior of flow fields using deep autoencoder, Aip Advances 9 (2019) 015006. doi:10.1063/1.5067313

    2019 doi

  9. [17]

    J.-Z. Peng, S. Chen, N. Aubry, Z.-H. Chen, W.-T. Wu, Time-variant prediction of flow over an airfoil using deep neural network, Physics of Fluids 32 (2020) 123602. doi:10.1063/5.0022222. 38

    2020 doi

  10. [18]

    Rozov, C

    V. Rozov, C. Breitsamter, Data-driven prediction of unsteady pressure distri- butions based on deep learning, Journal of Fluids and Structures 104 (2021) 103316. doi:10.1016/j.jfluidstructs.2021.103316

    2021

  11. [19]

    R. Han, Y. Wang, Y. Zhang, G. Chen, A novel spatial-temporal prediction method for unsteady wake flows based on hybrid deep neural network, Physics of Fluids 31 (2019). doi: 10.1063/1.5127247

    2019 doi

  12. [20]

    Saetta, R

    E. Saetta, R. Tognaccini, G. Iaccarino, Machine learning to predict aerodynamic stall, International Journal of Computational Fluid Dynamics 36 (2022) 641–

    2022

  13. [21]

    Massegur Sampietro, A

    D. Massegur Sampietro, A. Da Ronch, Graph convolutional multi-mesh autoen- coder for steady transonic aircraft aerodynamics, Machine Learning: Science and Technology (2023). doi:10.1088/2632-2153/ad36ad

    2023 doi

  14. [22]

    G. E. Hinton, R. R. Salakhutdinov, Reducing the dimensionality of data with neural networks, science 313 (2006) 504–507. doi: 10.1126/science.1127647

    2006 doi

  15. [23]

    Vincent, H

    P. Vincent, H. Larochelle, Y. Bengio, P.-A. Manzagol, Extracting and com- posing robust features with denoising autoencoders, in: Proceedings of the 25th international conference on Machine learning, 2008, pp. 1096–1103. doi:10.1145/1390156.1390294

    2008

  16. [24]

    Immordino, A

    G. Immordino, A. Vaiuso, A. Da Ronch, M. Righi, Predicting transonic flow- fields in non-homogeneous unstructured grids using autoencoder graph convolu- tional networks, arXiv preprint arXiv:2405.04396 (2024). doi: 10.48550/arXiv. 2405.04396

    2024 doi arXiv

  17. [25]

    Hochreiter, J

    S. Hochreiter, J. Schmidhuber, Long short-term memory, Neural computation 9 (1997) 1735–1780. doi: 10.1162/neco.1997.9.8.1735

    1997 doi

  18. [26]

    K. Cho, B. Van Merri¨ enboer, D. Bahdanau, Y. Bengio, On the properties of neural machine translation: Encoder-decoder approaches, arXiv preprint arXiv:1409.1259 (2014). doi: 10.3115/v1/W14-4012

    2014 arXiv

  19. [27]

    Salehinejad, S

    H. Salehinejad, S. Sankar, J. Barfett, E. Colak, S. Valaee, Recent advances in recurrent neural networks, arXiv preprint arXiv:1801.01078 (2017). doi: 10. 48550/arXiv.1801.01078. 39

    2017 doi arXiv

  20. [28]

    K. Li, J. Kou, W. Zhang, Deep neural network for unsteady aerodynamic and aeroelastic modeling across multiple mach numbers, Nonlinear Dynamics 96 (2019) 2157–2177. doi: 10.1007/s11071-019-04915-9

    2019 doi

  21. [29]

    S. L. Brunton, B. R. Noack, P. Koumoutsakos, Machine learning for fluid me- chanics, Annual review of fluid mechanics 52 (2020) 477–508. doi: 10.1146/ annurev-fluid-010719-060214

    2020

  22. [30]

    Q. Wang, C. E. Cesnik, K. Fidkowski, Multivariate recurrent neural network models for scalar and distribution predictions in unsteady aerodynamics, in: AIAA Scitech 2020 Forum, 2020, p. 1533. doi: 10.2514/6.2020-1533

    2020 doi

  23. [31]

    Mannarino, P

    A. Mannarino, P. Mantegazza, Nonlinear aeroelastic reduced order modeling by recurrent neural networks, Journal of Fluids and Structures 48 (2014) 103–121. doi:10.1016/j.jfluidstructs.2014.02.016

    2014 doi

  24. [32]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, I. Polosukhin, Attention is all you need, Advances in neural in- formation processing systems 30 (2017). doi: 10.48550/arXiv.1706.03762

    2017 doi arXiv

  25. [33]

    Cheng, L

    J. Cheng, L. Dong, M. Lapata, Long short-term memory-networks for machine reading, arXiv preprint arXiv:1601.06733 (2016). doi: 10.18653/v1/D16-1053

    2016 arXiv

  26. [34]

    Cheng, F

    X. Cheng, F. Shi, M. Zhao, G. Li, H. Zhang, S. Chen, Temporal attention convolutional neural network for estimation of icing probability on wind tur- bine blades, IEEE Transactions on Industrial Electronics 69 (2021) 6371–6380. doi:10.1109/TIE.2021.3090702

    2021

  27. [35]

    X. Han, H. Gao, T. Pfaff, J.-X. Wang, L.-P. Liu, Predicting physics in mesh- reduced space with temporal attention, arXiv preprint arXiv:2201.09113 (2022). doi:10.48550/arXiv.2201.09113

    2022 doi arXiv

  28. [36]

    J. Du, X. Li, S. Dong, Z. Liu, G. Chen, A novel attention enhanced deep neural network for hypersonic spatiotemporal turbulence prediction, Physics of Fluids 36 (2024). doi: 10.1063/5.0210966

    2024 doi

  29. [37]

    J. Bai, J. Zhu, Y. Song, L. Zhao, Z. Hou, R. Du, H. Li, A3t-gcn: Attention tem- poral graph convolutional network for traffic forecasting, ISPRS International Journal of Geo-Information 10 (2021) 485. doi: 10.48550/arXiv.2006.11583. 40

    2021 doi arXiv

  30. [38]

    D. K. Duvenaud, D. Maclaurin, J. Iparraguirre, R. Bombarell, T. Hirzel, A. Aspuru-Guzik, R. P. Adams, Convolutional networks on graphs for learning molecular fingerprints, Advances in neural information processing systems 28 (2015). doi:10.48550/arXiv.1509.09292

    2015 doi arXiv

  31. [39]

    M. Fey, J. E. Lenssen, Fast graph representation learning with pytorch geomet- ric, 2019. doi: 10.48550/arXiv.1903.02428. arXiv:1903.02428

    2019 doi arXiv

  32. [40]

    D. K. Hammond, P. Vandergheynst, R. Gribonval, Wavelets on graphs via spectral graph theory, Applied and Computational Harmonic Analysis 30 (2011) 129–150. doi:10.1016/j.acha.2010.04.005

    2011 doi

  33. [41]

    X. SHI, Z. Chen, H. Wang, D.-Y. Yeung, W.-k. Wong, W.-c. WOO, Convolu- tional lstm network: A machine learning approach for precipitation nowcasting, in: C. Cortes, N. Lawrence, D. Lee, M. Sugiyama, R. Garnett (Eds.), Advances in Neural Information Processing Systems, volume 2...

  34. [42]

    Y. Seo, M. Defferrard, P. Vandergheynst, X. Bresson, Structured sequence modeling with graph convolutional recurrent networks, in: Neural Information Processing: 25th International Conference, ICONIP 2018, Siem Reap, Cambo- dia, December 13-16, 2018, Proceedings, Part I 25, Sp...

    2018 doi arXiv

  35. [43]

    B. Yu, H. Yin, Z. Zhu, Spatio-temporal graph convolutional networks: A deep learning framework for traffic forecasting, arXiv preprint arXiv:1709.04875 (2017). doi:10.24963/ijcai.2018/505

    2017 arXiv

  36. [44]

    De Maesschalck, D

    R. De Maesschalck, D. Jouan-Rimbaud, D. L. Massart, The mahalanobis distance, Chemometrics and intelligent laboratory systems 50 (2000) 1–18. doi:10.1016/j.patcog.2008.05.018

    2000 doi

  37. [45]

    Heeg, Overview of the aeroelastic prediction workshop, in: 51st AIAA aerospace sciences meeting including the new horizons forum and aerospace exposition, 2013, p

    J. Heeg, Overview of the aeroelastic prediction workshop, in: 51st AIAA aerospace sciences meeting including the new horizons forum and aerospace exposition, 2013, p. 783. doi: 10.2514/6.2013-783

    2013 doi

  38. [46]

    T. D. Economon, F. Palacios, S. R. Copeland, T. W. Lukaczyk, J. J. Alonso, Su2: An open-source suite for multiphysics simulation and design, Aiaa Journal 54 (2016) 828–846. doi: 10.2514/1.J053813

    2016 doi

  39. [47]

    D. P. Kingma, J. Ba, Adam: A method for stochastic optimization, arXiv preprint arXiv:1412.6980 (2014). doi: 10.48550/arXiv.1412.6980. 41

    2014 doi arXiv

  40. [654]

    doi:10.1080/10618562.2023.2171021

    2023

  41. [2015]

    doi:10.48550/arXiv.1506.04214

    doi arXiv

Pith tools

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