pith. sign in

arxiv: 2606.23940 · v1 · pith:3VAAFXDDnew · submitted 2026-06-22 · ⚛️ physics.flu-dyn · cs.LG

Prediction of Viscoelastic Droplet Impact Dynamics Using a Vision Transformer-Based Approach

Diego A. de Aguiar , Cassio M. Oishi This is my paper

Pith reviewed 2026-06-26 06:33 UTC · model grok-4.3

classification ⚛️ physics.flu-dyn cs.LG
keywords viscoelastic droplet impactVideo Vision TransformerVolume of Fluidspreading regimebouncing regimetemporal predictioncomputational acceleration
0
0 comments X

The pith

A Video Vision Transformer predicts the full impact evolution of viscoelastic droplets from only the first 10-20% of a volume-of-fluid simulation.

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

The paper establishes that a Video Vision Transformer trained on early volume fraction fields can generate the later stages of droplet impact dynamics for viscoelastic fluids. It handles variations in Reynolds number, Weber number, solvent viscosity ratio, and Weissenberg number, and reproduces both spreading and bouncing outcomes. The approach cuts the required simulation length by 80-90% while preserving geometric features and physical consistency across parameter sets. Readers would care because repeated full simulations become prohibitive when exploring many combinations of fluid properties in manufacturing processes. The work treats the initial trajectory segment as sufficient to determine the remainder without explicit additional physics equations.

Core claim

The ViViT architecture takes initial volume fraction fields from VOF simulations of viscoelastic droplet impacts and outputs the subsequent time evolution. It produces predictions that remain consistent with the governing physics for different values of Re, We, beta, and Wi, correctly distinguishing spreading from bouncing regimes and maintaining structural similarity to full simulations.

What carries the argument

Video Vision Transformer (ViViT) that ingests sequences of early volume fraction fields and autoregressively predicts later fields of the droplet-surface interaction.

If this is right

  • The method reduces total computational cost by 80-90% while still capturing both spreading and bouncing regimes.
  • Geometric features such as droplet shape and contact area remain structurally similar to full simulations.
  • Predictions stay physically consistent when parameters change within the training distribution.
  • Volume fraction fields from experiments could be substituted for simulation data during training.
  • The same early-to-late mapping applies across different prediction horizons.

Where Pith is reading between the lines

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

  • The same early-trajectory shortcut might apply to other free-surface flows where inertia and elasticity set the long-term outcome after a short initial transient.
  • Training on mixed simulation and experimental video data could reduce reliance on any single modeling assumption about the fluid rheology.
  • Extending the horizon further could test whether the transformer learns an implicit constitutive relation rather than simple pattern continuation.

Load-bearing premise

The initial 10-20% of a VOF simulation trajectory contains sufficient information to determine the full subsequent dynamics for unseen parameter combinations without loss of physical fidelity.

What would settle it

Full VOF simulations run to completion for a held-out combination of Re, We, beta, and Wi; direct pixel-wise or shape comparison of the ViViT-generated volume fraction fields at late times against the ground-truth fields reveals whether regime, contact line position, or droplet height deviate systematically.

Figures

Figures reproduced from arXiv: 2606.23940 by Cassio M. Oishi, Diego A. de Aguiar.

Figure 1
Figure 1. Figure 1: General pipeline of the proposed framework. (1) Initial temporal context obtained from [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of the droplet impact dynamics obtained with different mesh resolutions for [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Distribution of bouncing and spreading cases. [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: General architecture of the implemented MLP. [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Basic transformer encoder architecture. Input embedding: to generate vector representations that capture relevant information for the at￾tention layers, the input data first passes through dense layers. Since the generated output does not carry any positional information, they are then passed through a time embedding process. Time embedding: incorporates information about the position of each element in th… view at source ↗
Figure 6
Figure 6. Figure 6: The image is divided into non-overlapping patches, which are then rearranged into an ordered [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Process of converting an image patch into an embedding vector. [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: General architecture of the implemented ViViT. [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Comparison between expected and predicted values using MLP and ViViT (50 [PITH_FULL_IMAGE:figures/full_fig_p015_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Comparison between expected and predicted droplet dynamics using MLP and ViViT (50 [PITH_FULL_IMAGE:figures/full_fig_p016_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Comparison between expected and predicted values using MLP and ViViT (50 [PITH_FULL_IMAGE:figures/full_fig_p017_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Comparison between expected and predicted droplet dynamics using MLP and ViViT (50 [PITH_FULL_IMAGE:figures/full_fig_p018_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Comparison between expected and predicted values using MLP and ViViT (100 [PITH_FULL_IMAGE:figures/full_fig_p019_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Comparison between expected and predicted droplet dynamics using MLP and ViViT (100 [PITH_FULL_IMAGE:figures/full_fig_p020_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Comparison of expected and predicted volume fraction fields obtained with MLP and ViViT [PITH_FULL_IMAGE:figures/full_fig_p021_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Comparison of expected and predicted volume fraction fields obtained with MLP and ViViT [PITH_FULL_IMAGE:figures/full_fig_p022_16.png] view at source ↗
read the original abstract

Droplet impact on solid surfaces is a complex fluid dynamics problem with applications in spray cooling, inkjet printing, and pharmaceutical processing. Although numerical simulations are widely used to investigate these dynamics, their computational cost becomes significant when multiple parametric variations are considered. In this work, we investigate the use of a Video Vision Transformer (ViViT) architecture to predict the temporal evolution of viscoelastic droplets impacting solid surfaces using volume fraction fields obtained from the Volume of Fluid (VOF) method. In Newtonian fluids, impact dynamics are mainly characterized by the Reynolds number $Re$, representing the ratio of inertial to viscous forces, and the Weber number $We$, representing the ratio of inertial to surface tension forces. For viscoelastic fluids, additional parameters are required to account for elastic effects, namely the solvent viscosity ratio $\beta$ and the Weissenberg number $Wi$, increasing simulation complexity and cost. Instead of simulating the entire droplet dynamics, the proposed approach uses only the initial 10% to 20% of the simulation to predict the remaining evolution. Depending on the prediction configuration, this strategy reduces computational cost by approximately 80% to 90% compared to full numerical simulations. The ViViT produces physically consistent predictions across different parameters and prediction horizons, successfully capturing both spreading and bouncing regimes while preserving geometric features and structural similarity. Since volume fraction fields can also be extracted from experimental videos, the proposed framework could be extended to incorporate experimental data during training, potentially improving the physical fidelity of the predicted dynamics.

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

3 major / 2 minor

Summary. The paper proposes a Video Vision Transformer (ViViT) surrogate model that takes the initial 10-20% of a VOF simulation trajectory for viscoelastic droplet impact (parameterized by Re, We, β, Wi) and predicts the remaining temporal evolution of the volume-fraction field, claiming 80-90% computational savings while producing physically consistent results that capture both spreading and bouncing regimes for unseen parameter combinations.

Significance. If the quantitative accuracy and extrapolation fidelity hold, the method would enable substantially faster parametric sweeps in viscoelastic droplet dynamics relevant to spray cooling and inkjet applications; the potential to incorporate experimental video data is a further strength.

major comments (3)
  1. [Abstract, Results] Abstract and Results section: the central claim that the ViViT produces 'physically consistent predictions' and 'preserves geometric features' for unseen (β, Wi) rests on qualitative visual inspection and structural similarity only; no L2 norms, interface error metrics, or quantitative comparison against held-out full VOF trajectories are reported, leaving the extrapolation error for delayed elastic recoil unquantified.
  2. [Methods, Results] Methods and Results: the assumption that the first 10-20% of the trajectory encodes the full viscoelastic memory (via Wi) is load-bearing for the cost-saving claim, yet no ablation on prediction horizon length versus parameter distance from the training distribution is presented; viscoelastic stress history can produce recoil whose onset lies outside the supplied window.
  3. [Results] Results: the reported 'across different parameters and prediction horizons' performance lacks any table or figure quantifying error growth with increasing Wi or decreasing β, which directly tests whether the initial-window sufficiency holds for the viscoelastic regime.
minor comments (2)
  1. [Introduction] Notation for the solvent viscosity ratio β and Weissenberg number Wi should be introduced with explicit definitions in the Introduction rather than assumed from the Newtonian case.
  2. [Methods] The manuscript should state the precise training/test split ratios and whether any parameter combinations in the test set lie outside the convex hull of the training set.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their insightful comments, which have helped us identify areas where the manuscript can be strengthened with additional quantitative validation. We address each major comment below.

read point-by-point responses

  1. Referee: [Abstract, Results] Abstract and Results section: the central claim that the ViViT produces 'physically consistent predictions' and 'preserves geometric features' for unseen (β, Wi) rests on qualitative visual inspection and structural similarity only; no L2 norms, interface error metrics, or quantitative comparison against held-out full VOF trajectories are reported, leaving the extrapolation error for delayed elastic recoil unquantified.

    Authors: We agree that the current presentation relies on qualitative assessments and the structural similarity index. In the revised manuscript, we will include quantitative error metrics, specifically L2 norms between predicted and ground-truth volume fraction fields, as well as interface-specific errors such as the average distance between predicted and actual interfaces. These will be computed for held-out trajectories, including cases with delayed elastic recoil, and presented in the Results section to better quantify the extrapolation performance. revision: yes

  2. Referee: [Methods, Results] Methods and Results: the assumption that the first 10-20% of the trajectory encodes the full viscoelastic memory (via Wi) is load-bearing for the cost-saving claim, yet no ablation on prediction horizon length versus parameter distance from the training distribution is presented; viscoelastic stress history can produce recoil whose onset lies outside the supplied window.

    Authors: This is a valid point regarding the robustness of the input window choice. Although our model was trained with varying input lengths in the 10-20% range, a dedicated ablation study correlating input horizon with parameter deviation (e.g., distance in Wi-β space) was not included. We will add such an analysis in the revised Methods and Results sections, showing how prediction accuracy degrades with shorter input windows or larger parameter distances, to substantiate the sufficiency of the initial trajectory segment for capturing viscoelastic memory effects. revision: yes

  3. Referee: [Results] Results: the reported 'across different parameters and prediction horizons' performance lacks any table or figure quantifying error growth with increasing Wi or decreasing β, which directly tests whether the initial-window sufficiency holds for the viscoelastic regime.

    Authors: We acknowledge the absence of explicit error quantification with respect to the viscoelastic parameters. The revised manuscript will feature a new figure and accompanying table that report the error metrics as functions of Wi and β across different prediction horizons. This will directly address whether the initial 10-20% window remains adequate as elastic effects (higher Wi, lower β) become more pronounced. revision: yes

Circularity Check

0 steps flagged

No circularity: data-driven surrogate model with no self-referential derivations

full rationale

The paper trains a ViViT model on VOF volume-fraction fields to map initial 10-20% trajectories to later frames for viscoelastic droplet impact. No analytic derivation chain exists; the central claim is an empirical performance statement about learned predictions on held-out parameter combinations (β, Wi). No equations reduce a claimed result to a fitted parameter or self-defined quantity by construction. No self-citation load-bearing steps, uniqueness theorems, or ansatzes appear in the provided text. The information-sufficiency assumption is tested via model outputs rather than presupposed by definition. This is a standard supervised learning surrogate and remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that early simulation data suffices for full trajectory prediction and on standard machine-learning training assumptions; no new physical entities are introduced.

free parameters (1)
  • ViViT model weights and hyperparameters
    Trained on simulation data; specific values not reported in abstract.
axioms (1)
  • domain assumption Initial 10-20% of VOF trajectory determines full dynamics
    Invoked to justify the partial-input prediction strategy and cost savings.

pith-pipeline@v0.9.1-grok · 5807 in / 1256 out tokens · 44833 ms · 2026-06-26T06:33:47.777521+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

71 extracted references · 24 canonical work pages

  1. [1]

    2023 , publisher =

    Jingzu Yee and Daichi Igarashi and Shun Miyatake and Yoshiyuki Tagawa , title =. 2023 , publisher =. doi:10.1088/2632-2153/acc727 , url =

    work page doi:10.1088/2632-2153/acc727 2023

  2. [2]

    2023 , publisher =

    Hossein Hassanzadeh and Saptarshi Joshi and Seyed Mohammad Taghavi , title =. 2023 , publisher =. doi:doi:10.1515/cppm-2023-0026 , url =

    work page doi:10.1515/cppm-2023-0026 2023

  3. [3]

    2023 , publisher =

    Demo, Nicola and Tezzele, Marco and Rozza, Gianluigi , title =. 2023 , publisher =. doi:10.1186/s40323-023-00249-9 , url =

    work page doi:10.1186/s40323-023-00249-9 2023

  4. [4]

    2022 , publisher =

    Csala, Hunor and Dawson, Scott and Arzani, Amirhossein , title =. 2022 , publisher =. doi:10.1063/5.0127284 , url =

    work page doi:10.1063/5.0127284 2022

  5. [5]

    2024 , publisher =

    Xiao, Qianhao and Zeng, Dongping and Yu, Zheqin and Zou, Shuyun and Liu, Zhong , title =. 2024 , publisher =. doi:10.1016/j.ijhydene.2024.08.465 , url =

    work page doi:10.1016/j.ijhydene.2024.08.465 2024

  6. [6]

    2024 , eprint=

    Sanderse, Benjamin and Stinis, Panos and Maulik, Romit and Ahmed, Shady E , title =. 2024 , eprint=

    2024

  7. [7]

    , volume=

    , author=. , volume=. , publisher=

  8. [8]

    2024 , publisher =

    Zhou, Hao and Cheng, Sibo and Arcucci, Rossella , title =. 2024 , publisher =. doi:10.1016/j.cma.2024.116758 , url =

    work page doi:10.1016/j.cma.2024.116758 2024

  9. [9]

    ModelFLOWs-app: data-driven post-processing and reduced order modelling tools , journal =

    Hetherington, Ashton and Corrochano, Adri. ModelFLOWs-app: data-driven post-processing and reduced order modelling tools , journal =. 2024 , publisher =. doi:10.1016/j.cpc.2024.109217 , url =

    work page doi:10.1016/j.cpc.2024.109217 2024

  10. [10]

    2024 , publisher =

    Oishi, Cassio M and Kaptanoglu, Alan A and Kutz, J Nathan and Brunton, Steven L , title =. 2024 , publisher =. doi:10.1098/rsos.240995 , url =

    work page doi:10.1098/rsos.240995 2024

  11. [11]

    2024 , publisher =

    Wang, Zihao and Zhang, Guiyong and Xing, Xiuqing and Xu, Xiangguo and Sun, Tiezhi , title =. 2024 , publisher =. doi:10.1016/j.oceaneng.2023.116421 , url =

    work page doi:10.1016/j.oceaneng.2023.116421 2024

  12. [12]

    2024 , publisher =

    Chen, Wenqian and Stinis, Panos , title =. 2024 , publisher =. doi:10.1016/j.jcp.2023.112683 , url =

    work page doi:10.1016/j.jcp.2023.112683 2024

  13. [13]

    2019 , publisher =

    Omata, Noriyasu and Shirayama, Susumu , title =. 2019 , publisher =. doi:10.1063/1.5067313 , url =

    work page doi:10.1063/1.5067313 2019

  14. [14]

    2004 , publisher =

    Daniel Lörstad and Marianne Francois and Wei Shyy and Laszlo Fuchs , title =. 2004 , publisher =. doi:https://doi.org/10.1002/fld.746 , url =

    work page doi:10.1002/fld.746 2004

  15. [15]

    2022 , publisher =

    Hai-Zhuan Yuan and Qing Liu and Gang Zeng , title =. 2022 , publisher =. doi:https://doi.org/10.1002/fld.5061 , url =

    work page doi:10.1002/fld.5061 2022

  16. [16]

    Markel and Robert N K Browning , title =

    Hua Tan and Erik Torniainen and David P. Markel and Robert N K Browning , title =. 2015 , publisher =. doi:https://doi.org/10.1002/fld.3997 , url =

    work page doi:10.1002/fld.3997 2015

  17. [17]

    Al-Dirawi and Andrew E

    Karrar H. Al-Dirawi and Andrew E. Bayly , title =. 2019 , publisher =. doi:10.1063/1.5085762 , url =

    work page doi:10.1063/1.5085762 2019

  18. [18]

    Proceedings of the Royal Society A , volume=

    Multi-fidelity reduced-order surrogate modelling , author=. Proceedings of the Royal Society A , volume=. 2024 , publisher=

    2024

  19. [19]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Multi-task learning using uncertainty to weigh losses for scene geometry and semantics , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  20. [20]

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

    Sepp Hochreiter and Jürgen Schmidhuber , title =. 1997 , publisher =. doi:10.1162/neco.1997.9.8.1735 , url =

    work page doi:10.1162/neco.1997.9.8.1735 1997

  21. [21]

    2021 , publisher =

    Seol-Hyun Noh , title =. 2021 , publisher =. doi:10.3390/info12110442 , url =

    work page doi:10.3390/info12110442 2021

  22. [22]

    University of Toronto Computer Science , year =

    Grosse, Roger , title =. University of Toronto Computer Science , year =

  23. [23]

    1988 , publisher=

    Spectral Methods inFfluid Dynamics , author=. 1988 , publisher=

    1988

  24. [24]

    Kelshaw, Daniel and Lestang, Thibault and Elise , title =

  25. [25]

    2021 , author =

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale , eprint=. 2021 , author =

    2021

  26. [26]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Vivit: A video vision transformer , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  27. [27]

    2017 , publisher =

    Vaswani, Ashish and Shazeer, Noam and Parmar, Niki and Uszkoreit, Jakob and Jones, Llion and Gomez, Aidan N and Kaiser, Lukasz and Polosukhin, Illia , booktitle =. 2017 , publisher =

    2017

  28. [28]

    Journal of computational physics , volume=

    A continuum method for modeling surface tension , author=. Journal of computational physics , volume=. 1992 , publisher=

    1992

  29. [29]

    2013--2025 , howpublished =

    Stéphane Popinet and collaborators , title =. 2013--2025 , howpublished =

    2013

  30. [30]

    Journal of Non-Newtonian Fluid Mechanics , volume=

    Time-dependent simulation of viscoelastic flows at high Weissenberg number using the log-conformation representation , author=. Journal of Non-Newtonian Fluid Mechanics , volume=. 2005 , publisher=

    2005

  31. [31]

    Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining , pages=

    Optuna: A next-generation hyperparameter optimization framework , author=. Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining , pages=

  32. [32]

    2022 , publisher =

    Detlef Lohse , title =. 2022 , publisher =. doi:10.1146/annurev-fluid-022321-114001 , url =

    work page doi:10.1146/annurev-fluid-022321-114001 2022

  33. [33]

    Advanced drug delivery reviews , volume=

    3D bioprinting of tissues and organs for regenerative medicine , author=. Advanced drug delivery reviews , volume=. 2018 , publisher=

    2018

  34. [34]

    Physical Review Fluids , volume=

    Influence of the impact energy on the pattern of blood drip stains , author=. Physical Review Fluids , volume=. 2018 , publisher=

    2018

  35. [35]

    Advances in colloid and interface science , volume=

    Impact of a heterogeneous liquid droplet on a dry surface: Application to the pharmaceutical industry , author=. Advances in colloid and interface science , volume=. 2010 , publisher=

    2010

  36. [36]

    Heat Transfer Research , volume=

    A review on thermo-fluidic study of droplet impact in spray cooling , author=. Heat Transfer Research , volume=. 2025 , publisher=

    2025

  37. [37]

    Food research international , volume=

    Drop impact behavior on food using spray coating: Fundamentals and applications , author=. Food research international , volume=. 2013 , publisher=

    2013

  38. [38]

    Fluid dynamics research , volume=

    Phenomena of liquid drop impact on solid and liquid surfaces , author=. Fluid dynamics research , volume=. 1993 , publisher=

    1993

  39. [39]

    Journal of Fluid Mechanics , volume=

    Maximal deformation of an impacting drop , author=. Journal of Fluid Mechanics , volume=. 2004 , publisher=

    2004

  40. [40]

    Journal of Fluid Mechanics , volume=

    Drop impact of yield-stress fluids , author=. Journal of Fluid Mechanics , volume=. 2009 , publisher=

    2009

  41. [41]

    Journal of Fluid Mechanics , volume=

    Normal and oblique drop impact of yield stress fluids with thixotropic effects , author=. Journal of Fluid Mechanics , volume=. 2019 , publisher=

    2019

  42. [42]

    Journal of Fluid Mechanics , volume=

    The role of viscoplastic drop shape in impact , author=. Journal of Fluid Mechanics , volume=. 2024 , publisher=

    2024

  43. [43]

    Lunkad and Vivek V

    Siddhartha F. Lunkad and Vivek V. Buwa and K.D.P. Nigam , title =. 2007 , publisher =. doi:https://doi.org/10.1016/j.ces.2007.07.036 , url =

    work page doi:10.1016/j.ces.2007.07.036 2007

  44. [44]

    Roisman , title =

    Ilia V. Roisman , title =. 2009 , publisher =. doi:10.1063/1.3129283 , url =

    work page doi:10.1063/1.3129283 2009

  45. [45]

    de Bruin and Denis Bartolo and Christophe Josserand and Daniel Bonn , title =

    Nick Laan and Karla G. de Bruin and Denis Bartolo and Christophe Josserand and Daniel Bonn , title =. 2014 , publisher =. doi:10.1103/PhysRevApplied.2.044018 , url =

    work page doi:10.1103/physrevapplied.2.044018 2014

  46. [46]

    2016 , publisher =

    Jae Bong Lee and Dominique Derome and Robert Guyer and Jan Carmeliet , title =. 2016 , publisher =. doi:10.1021/acs.langmuir.5b04557 , url =

    work page doi:10.1021/acs.langmuir.5b04557 2016

  47. [47]

    2018 , publisher =

    Shiji Lin and Binyu Zhao and Song Zou and Jianwei Guo and Zheng Wei and Longquan Chen , title =. 2018 , publisher =. doi:https://doi.org/10.1016/j.jcis.2017.12.086 , url =

    work page doi:10.1016/j.jcis.2017.12.086 2018

  48. [48]

    2023 , publisher =

    Jiguo Tang and Shengzhi Yu and Xiaofan Hou and Tianhui Wu and Hongtao Liu , title =. 2023 , publisher =. doi:10.1021/acs.iecr.3c01560 , url =

    work page doi:10.1021/acs.iecr.3c01560 2023

  49. [49]

    Chemical Engineering Science , volume=

    A machine learning-based approach to predict the outcome of binary droplet collision , author=. Chemical Engineering Science , volume=. 2026 , publisher=

    2026

  50. [50]

    IEEE Access , year=

    Machine Learning-Based Classification of Suspension Droplet-Solid Wall Impacts for Control of Droplet Fragmentation , author=. IEEE Access , year=

  51. [51]

    IEEE Transactions on Instrumentation and Measurement , volume=

    Robust fluid motion estimator based on attentional transformer , author=. IEEE Transactions on Instrumentation and Measurement , volume=. 2023 , publisher=

    2023

  52. [52]

    Nature Communications , volume=

    -variational autoencoders and transformers for reduced-order modelling of fluid flows , author=. Nature Communications , volume=. 2024 , publisher=

    2024

  53. [53]

    Physics of Fluids , volume=

    Self-supervised learning based on transformer for flow reconstruction and prediction , author=. Physics of Fluids , volume=. 2024 , publisher=

    2024

  54. [54]

    Physics of Fluids , volume=

    Transformer model learning driven analysis of Casson fluid flow influenced by Hall currents and Darcy--Forchheimer effects , author=. Physics of Fluids , volume=. 2025 , publisher=

    2025

  55. [55]

    Physics of Fluids , volume=

    Reconstruction of time-resolved three-dimensional flow fields based on a multi-domain fusion transformer , author=. Physics of Fluids , volume=. 2026 , publisher=

    2026

  56. [56]

    Processes , volume=

    Reservoir Surrogate Modeling Using U-Net with Vision Transformer and Time Embedding , author=. Processes , volume=. 2025 , publisher=

    2025

  57. [57]

    Mathematical and Computational Applications , volume=

    Permeability prediction using vision transformers , author=. Mathematical and Computational Applications , volume=. 2025 , publisher=

    2025

  58. [58]

    AIP Advances , volume=

    A new fluid flow approximation method using a vision transformer and a U-shaped convolutional neural network , author=. AIP Advances , volume=. 2023 , publisher=

    2023

  59. [59]

    arXiv preprint arXiv:2604.02483 , year=

    A Multimodal Vision Transformer-based Modeling Framework for Prediction of Fluid Flows in Energy Systems , author=. arXiv preprint arXiv:2604.02483 , year=

    Pith/arXiv arXiv

  60. [60]

    Proceedings of the European conference on computer vision (ECCV) , pages=

    Folded recurrent neural networks for future video prediction , author=. Proceedings of the European conference on computer vision (ECCV) , pages=

  61. [61]

    Proceedings of the IEEE international conference on computer vision , pages=

    Learning spatiotemporal features with 3d convolutional networks , author=. Proceedings of the IEEE international conference on computer vision , pages=

  62. [62]

    Advances in neural information processing systems , volume=

    Convolutional LSTM network: A machine learning approach for precipitation nowcasting , author=. Advances in neural information processing systems , volume=

  63. [63]

    Physics of Fluids , volume=

    Impact of capillary drops of complex fluids on a solid surface , author=. Physics of Fluids , volume=. 2019 , publisher=

    2019

  64. [64]

    Journal of Non-Newtonian Fluid Mechanics , volume=

    Numerical simulation of drop impact and jet buckling problems using the eXtended Pom--Pom model , author=. Journal of Non-Newtonian Fluid Mechanics , volume=. 2012 , publisher=

    2012

  65. [65]

    arXiv preprint arXiv:2411.15949 , year=

    PINNs4Drops: Video-conditioned physics-informed neural networks for two-phase flow reconstruction , author=. arXiv preprint arXiv:2411.15949 , year=

    arXiv

  66. [66]

    International journal of heat and mass transfer , volume=

    An experimental study of bouncing Leidenfrost drops: comparison between Newtonian and viscoelastic liquids , author=. International journal of heat and mass transfer , volume=. 2009 , publisher=

    2009

  67. [67]

    Journal of Non-Newtonian Fluid Mechanics , volume=

    Impact of viscoelastic droplets , author=. Journal of Non-Newtonian Fluid Mechanics , volume=. 2017 , publisher=

    2017

  68. [68]

    Physical Review Fluids , volume=

    Effects of viscoelasticity on drop impact and spreading on a solid surface , author=. Physical Review Fluids , volume=. 2016 , publisher=

    2016

  69. [69]

    Microfluidics and nanofluidics , volume=

    The role of viscoelasticity in drop impact and spreading for inkjet printing of polymer solution on a wettable surface , author=. Microfluidics and nanofluidics , volume=. 2013 , publisher=

    2013

  70. [70]

    Journal of Rheology , volume=

    Influence of fluid elasticity on drops impacting on dry surfaces , author=. Journal of Rheology , volume=. 2000 , publisher=

    2000

  71. [71]

    Journal of Fluid Mechanics , volume=

    Spreading of a viscoelastic drop on a solid substrate , author=. Journal of Fluid Mechanics , volume=. 2024 , publisher=

    2024