pith. sign in

arxiv: 2605.29911 · v1 · pith:Q6JBK55Znew · submitted 2026-05-28 · 💻 cs.LG · cs.CV

Reducing Experimental Testing in Space Propulsion Film Cooling Analyses by Pixelwise Generative Image Interpolation

Adam T. M\"uller , Philipp J. Teuffel , Konstantin Manassis , Nicolaj C. Stache This is my paper

Pith reviewed 2026-06-29 08:46 UTC · model grok-4.3

classification 💻 cs.LG cs.CV
keywords machine learningimage regressionfilm coolingspace propulsionneural networksparse dataexperimental optimizationgenerative interpolation
0
0 comments X

The pith

A lightweight neural network generates accurate film cooling images from 30% fewer experimental measurements.

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

The authors present a machine learning approach for regressing full images from sparse measurements in film cooling studies for space propulsion. A feed-forward neural network with positional encoding produces images conditioned on input parameters, validated to have RMSE under 8% and SSIM over 93% on real and synthetic data. This supports reducing measurements by 30% while keeping data quality high enough for injector optimization. A knowledge-informed extension is added for better local adaptability.

Core claim

A lightweight feed-forward neural network with positional encoding generates pixelwise images from sparse experimental measurements in film cooling analyses, conditioned by input parameters, achieving high similarity to ground truth while enabling a 30% reduction in the number of required tests.

What carries the argument

Lightweight feed-forward neural network with positional encoding for conditioned image generation from sparse data.

Load-bearing premise

The neural network trained on sparse measurements will generalize to new injector configurations and flow conditions without missing key physical effects in the film cooling process.

What would settle it

A new set of experiments on an unseen injector configuration where the generated images deviate substantially from measured data in terms of cooling effectiveness or temperature distribution.

Figures

Figures reproduced from arXiv: 2605.29911 by Adam T. M\"uller, Konstantin Manassis, Nicolaj C. Stache, Philipp J. Teuffel.

Figure 1
Figure 1. Figure 1: Neural network architecture, where (a) shows the full architecture from input parameters to output and (b) [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Schematic diagram of the proposed neural network, illustrating (left) the training phase, where the network [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Experimental setup for conducting cold flow tests of a film cooling setup with a single coolant injector on [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Different image processing steps for an example image of a coolant film with our surrogate ink and dem￾ineralized water solution. (a) shows the image captured by the imaging system, (b) the binarization result, (c) the downsampled binary image and (d) a grayscale image, obtained by using (c) to mask (a). High-speed imaging was conducted using a Photron FASTCAM SA1.1 camera equipped with a Nikon ED AF MICRO… view at source ↗
Figure 5
Figure 5. Figure 5: PixCOIN regression results on synthetic images for datasets of (1.) a binary function plot, (2.) a represen [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: PixCOIN interpolation results on real images. This one-dimensional depiction shows varying the injection [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: PixCOIN regression results on real images showing the variation of the injection angle [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Real images that were cropped ((a) and (d)), generated by PixCOIN trained on binary data ((b) and (e)) [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Exemplary comparison of real (1.) and generated (2.) image data for two operating points. Column (3.) [PITH_FULL_IMAGE:figures/full_fig_p010_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Histogram of pixel values in bins of intensity values of multiples of 10 and their contribution to the RMSE [PITH_FULL_IMAGE:figures/full_fig_p011_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Schematic flowchart of the extended training workflow to enable an expert informed adaptation. [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: GradCAM activation (b) on an inverted image (a) from the classifier to detect zones contributing to a [PITH_FULL_IMAGE:figures/full_fig_p012_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: PixCOIN image generation results for two operating points without ((a), (c)) and with ((b), (d)) the IML [PITH_FULL_IMAGE:figures/full_fig_p012_13.png] view at source ↗
read the original abstract

We propose a machine learning approach for image regression from sparse experimental measurements. We show the application of the proposed method on film cooling studies in propulsion system development, aiming to reduce the need for extensive physical testing. Our method employs a lightweight feed-forward neural network with positional encoding to generate images conditioned by input parameters. Validated on real and synthetic data, it achieves high image similarity (RMSE < 8 %, SSIM > 93 %) while maintaining accuracy with a 30 \% reduction of measurements. We further propose a knowledge-informed extension for local adaptability of the generated images. This approach significantly reduces required tests while preserving high-quality data, enabling efficient optimization of coolant injector configurations with applications beyond aerospace.

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

1 major / 1 minor

Summary. The paper proposes a lightweight feed-forward neural network with positional encoding for pixelwise image regression from sparse experimental measurements in film cooling analyses for space propulsion systems. Conditioned on input parameters, the model generates full images; a knowledge-informed extension is also introduced for local adaptability. Validated on real and synthetic data, it reports RMSE < 8 % and SSIM > 93 % while claiming a 30 % reduction in required physical measurements, with the goal of enabling efficient optimization of new coolant injector configurations.

Significance. If the reported accuracy holds under proper out-of-distribution testing, the approach could meaningfully lower the experimental burden in propulsion development by substituting interpolation for additional physical tests. The method is lightweight and directly targets a practical bottleneck in aerospace testing.

major comments (1)
  1. [Abstract] Abstract: the central claim that the method enables a 30 % reduction in measurements for previously unseen injector configurations and flow conditions requires demonstration that held-out test cases are out-of-distribution with respect to the training parameter space. The reported RMSE and SSIM figures on 'real and synthetic data' do not establish this, as no information is given on data partitioning, parameter ranges in the test set, or whether new geometries were included.
minor comments (1)
  1. [Abstract] The abstract supplies performance numbers but omits any description of training procedures, data partitioning strategy, error analysis, or the precise experimental design that produced the 30 % reduction figure.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful review and for emphasizing the importance of clarifying whether the held-out cases support claims of applicability to previously unseen injector configurations. We address the single major comment below and will revise the manuscript to supply the requested details on data partitioning.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that the method enables a 30 % reduction in measurements for previously unseen injector configurations and flow conditions requires demonstration that held-out test cases are out-of-distribution with respect to the training parameter space. The reported RMSE and SSIM figures on 'real and synthetic data' do not establish this, as no information is given on data partitioning, parameter ranges in the test set, or whether new geometries were included.

    Authors: We agree that the manuscript does not currently supply the information needed to evaluate whether the reported 30 % reduction applies to out-of-distribution cases. In the revised version we will add a dedicated paragraph in the 'Dataset and Experimental Setup' section that (i) describes the exact splitting procedure, (ii) lists the numerical ranges of all conditioning parameters (mass-flow ratio, blowing ratio, etc.) used for training versus testing, and (iii) states whether any test images correspond to injector geometries absent from the training set. If the test cases prove to lie inside the convex hull of the training distribution, we will also revise the abstract and conclusion to qualify the claim as a reduction achieved via interpolation within the sampled parameter space, while noting the potential for further OOD validation. revision: yes

Circularity Check

0 steps flagged

No significant circularity; supervised regression validated on held-out data

full rationale

The paper describes a standard supervised image regression task: a feed-forward NN with positional encoding is trained on sparse experimental measurements (real and synthetic) to generate full images conditioned on input parameters. Performance is reported via RMSE and SSIM against held-out test cases, with no equations or claims that reduce a prediction to a fitted parameter by construction. No self-citations are load-bearing for the core method, no uniqueness theorems are invoked, and no ansatz is smuggled via prior work. The derivation chain is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no explicit free parameters, axioms, or invented entities; the method implicitly relies on standard neural network assumptions such as sufficient training data coverage and that pixel-wise regression captures the relevant physics.

pith-pipeline@v0.9.1-grok · 5658 in / 974 out tokens · 24604 ms · 2026-06-29T08:46:03.461958+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

34 extracted references · 20 canonical work pages · 1 internal anchor

  1. [1]

    Review on film cooling of liquid rocket engines,

    S. R. Shine and S. S. Nidhi, “Review on film cooling of liquid rocket engines,”Propulsion and Power Research, vol. 7, no. 1, pp. 1–18, Mar. 2018. doi: 10.1016/j.jppr.2018.01.004

    work page doi:10.1016/j.jppr.2018.01.004 2018

  2. [2]

    Optimal Liquid Engine Architecture by Performance- Cooling TradeoffAnalysis,

    S. Tauchi, C. Inoue, Z. Wang, Y . Daimon, and G. Fujii, “Optimal Liquid Engine Architecture by Performance- Cooling TradeoffAnalysis,”Journal of Propulsion and Power, vol. 40, no. 4, pp. 631–641, 2024. doi: 10.2514/1.B39409

    work page doi:10.2514/1.b39409 2024

  3. [3]

    Data-Driven Aerospace Engineering: Reframing the Industry with Machine Learning,

    S. L. Brunton, J. N. Kutz, K. Manohar, A. Y . Aravkin, K. Morgansen, J. Klemisch, N. Goebel, J. Buttrick, J. Poskin, A. W. Blom-Schieber, T. Hogan, and D. McDonald, “Data-Driven Aerospace Engineering: Reframing the Industry with Machine Learning,”AIAA Journal, vol. 59, no. 8, pp. 2820–2847, 2021. doi: 10.2514/1.J060131

    work page doi:10.2514/1.j060131 2021

  4. [4]

    Green Propellants for Satellite Propulsion: Status of Research and Developments at DLR Lampoldshausen,

    C. U. Kirchberger, T. Hörger, J. Scholl, F. Merz, P. Teuffel, M. Kurilov, F. Lauck, L. Werling, R. Scholl, S. Ricker, and D. Freudenmann, “Green Propellants for Satellite Propulsion: Status of Research and Developments at DLR Lampoldshausen,” inAIAA SCITECH 2024 F orum, American Institute of Aeronautics and Astronautics. doi: 10.2514/6.2024-1789

    work page doi:10.2514/6.2024-1789 2024

  5. [5]

    Test results of a film cooled 200 N hypergolic green propellant thruster using hydrogen peroxide as coolant,

    P. Teuffel, F. Lauck, F. Merz, D. Turner, L. Werling, and C. Kirchberger, “Test results of a film cooled 200 N hypergolic green propellant thruster using hydrogen peroxide as coolant,”Space Propulsion Conference, 2024

    2024

  6. [6]

    Saito, Z

    R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, and D. Batra, “Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization,” in2017 IEEE International Conference on Computer Vision (ICCV), Oct. 2017, pp. 618–626. doi: 10.1109/ICCV .2017.74

    work page doi:10.1109/iccv 2017

  7. [7]

    Thermal performance of MMH/NTO rocket thrust chamber based on pintle injector by using liquid film cooling,

    G. Zhang, G. Li, L. Li, and G. Tang, “Thermal performance of MMH/NTO rocket thrust chamber based on pintle injector by using liquid film cooling,”Appl. Therm. Eng., vol. 223, p. 120035, Mar. 2023. 13 REDUCING EXPERIMENTAL TESTING VIA GENERATIVE IMAGE INTERPOLATION

    2023

  8. [8]

    Evaporation of Three-Dimensional Wavy Liquid Film Entrained by Turbulent Gas Flow,

    T. Inoue, C. Inoue, G. Fujii, and Y . Daimon, “Evaporation of Three-Dimensional Wavy Liquid Film Entrained by Turbulent Gas Flow,”AIAA J., vol. 60, no. 6, pp. 3805–3812, Jun. 2022

    2022

  9. [9]

    Experimental analysis of hydrogen peroxide film-cooling method for nontoxic hypergolic thruster,

    H. Kang, H. Kim, S. Heo, S. Jung, and S. Kwon, “Experimental analysis of hydrogen peroxide film-cooling method for nontoxic hypergolic thruster,”Aerospace Sci. Technol., vol. 71, pp. 751–762, Dec. 2017

    2017

  10. [10]

    Flow distribution in the liquid film created by a coherent circular water jet impinging obliquely on a plane wall,

    M. W. L. Chee, P. Chauhan, J. Georgiou, and D. I. Wilson, “Flow distribution in the liquid film created by a coherent circular water jet impinging obliquely on a plane wall,”Experimental Thermal and Fluid Science, vol. 140, p. 110748, Jan. 2023. doi: 10.1016/j.expthermflusci.2022.110748

    work page doi:10.1016/j.expthermflusci.2022.110748 2023

  11. [11]

    Data Driven Models for the De- sign of Rocket Injector Elements,

    J. F. Zapata Usandivaras, A. Urbano, M. Bauerheim, and B. Cuenot, “Data Driven Models for the De- sign of Rocket Injector Elements,”Aerospace, vol. 9, no. 10, p. 594, Oct. 2022. [Online]. Available: https://www.mdpi.com/2226-4310/9/10/594

    2022

  12. [12]

    Surrogate Neural Network for Rapid Flight Performance Evaluation of Hybrid Rocket Engines,

    A. Zavoli, P. M. Zolla, L. Federici, M. T. Migliorino, and D. Bianchi, “Surrogate Neural Network for Rapid Flight Performance Evaluation of Hybrid Rocket Engines,”J. Spacecr . Rockets, vol. 59, no. 6, pp. 2003–2016, 2022. [Online]. Available: https://doi.org/10.2514/1.A35369

    work page doi:10.2514/1.a35369 2003

  13. [13]

    Experimental and Simulative Evaluation of a Reinforcement Learning Based Cold Gas Thrust Chamber Pressure Controller,

    T. Hörger, L. Werling, K. Dresia, G. Waxenegger-Wilfing, and S. Schlechtriem, “Experimental and Simulative Evaluation of a Reinforcement Learning Based Cold Gas Thrust Chamber Pressure Controller,”Acta Astronau- tica, vol. 219, pp. 128–137, Jun. 2024

    2024

  14. [14]

    Machine Learning Methods for the Design and Operation of Liquid Rocket Engines – Research Activities at the DLR Institute of Space Propulsion,

    G. Waxenegger-Wilfing, K. Dresia, J. Deeken, and M. Oschwald, “Machine Learning Methods for the Design and Operation of Liquid Rocket Engines – Research Activities at the DLR Institute of Space Propulsion,”arXiv, no. arXiv:2102.07109, Feb. 2021. [Online]. Available: http://arxiv.org/abs/2102.07109

    work page arXiv 2021

  15. [15]

    Conditional image synthe- sis with diffusion models: A survey.arXiv preprint arXiv:2409.19365,

    Z. Zhan, D. Chen, J.-P. Mei, Z. Zhao, J. Chen, C. Chen, S. Lyu, and C. Wang, “Conditional Image Syn- thesis with Diffusion Models: A Survey,”arXiv preprint arXiv:2409.19365, Jun. 2025. [Online]. Available: http://arxiv.org/abs/2409.19365

    work page arXiv 2025

  16. [16]

    Pixel Recurrent Neural Networks,

    A. van den Oord, N. Kalchbrenner, and K. Kavukcuoglu, “Pixel Recurrent Neural Networks,” inProc. 33rd Int. Conf. Mach. Learn. (ICML), 2016, pp. 1747–1756. [Online]. Available: https://proceedings.mlr.press/v48/oord16.html

    2016

  17. [17]

    FILM: Frame Interpolation for Large Motion,

    F. Reda, J. Kontkanen, E. Tabellion, D. Sun, C. Pantofaru, and B. Curless, “FILM: Frame Interpolation for Large Motion,” inComputer Vision - ECCV 2022, Cham: Springer Nature Switzerland, 2022, pp. 250-266. doi: 10.1007/978-3-031-20071-7_15

    work page doi:10.1007/978-3-031-20071-7_15 2022

  18. [18]

    Machine-learning enabled prediction of 3D spray under engine combustion network spray G conditions,

    J. Hwang, P. Lee, S. Mun, I. K. Karathanassis, P. Koukouvinis, L. M. Pickett, and M. Gavaises, “Machine-learning enabled prediction of 3D spray under engine combustion network spray G conditions,”Fuel, vol. 293, p. 120444, Jun. 2021. doi: 10.1016/j.fuel.2021.120444

    work page doi:10.1016/j.fuel.2021.120444 2021

  19. [19]

    Video Frame Interpolation: A Comprehensive Survey,

    J. Dong, K. Ota, and M. Dong, “Video Frame Interpolation: A Comprehensive Survey,”ACM Trans. Multimedia Comput. Commun. Appl., vol. 19, no. 2, pp. 78:1–78:31, 2023. doi: 10.1145/3556544

    work page doi:10.1145/3556544 2023

  20. [20]

    DiffMorpher: Unleashing the Capability of Diffusion Models for Image Morphing,

    K. Zhang, Y . Zhou, X. Xu, B. Dai, and X. Pan, “DiffMorpher: Unleashing the Capability of Diffusion Models for Image Morphing,” inProc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2024, pp. 7912–7921

    2024

  21. [21]

    Conditional Image Generation with PixelCNN Decoders,

    A. van den Oord, N. Kalchbrenner, L. Espeholt, K. Kavukcuoglu, O. Vinyals, and A. Graves, “Conditional Image Generation with PixelCNN Decoders,” inAdvances in Neural In- formation Processing Systems, vol. 29, Curran Associates, Inc., 2016. [Online]. Available: https://proceedings.neurips.cc/paper/2016/hash/b1301141feffabac455e1f90a7de2054-Abstract.html

    2016

  22. [22]

    Gonzalez and R

    R. Gonzalez and R. Woods,Digital Image Processing. Pearson Deutschland, 2017, pp. 699–810. ISBN: 9781292223070

    2017

  23. [23]

    PyTorch: An Imperative Style, High-Performance Deep Learning Library

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. Köpf, E. Yang, Z. DeVito, M. Raison, A. Tejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, and S. Chintala, “PyTorch: An Imperative Style, High-Performance Deep Learning Library,” arXiv preprint arXiv:1912.01703, Dec. 2019. do...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1912.01703 1912

  24. [24]

    Hydrogen peroxide and its analogues: I. Density, refractive index, viscosity, and surface tension of deuterium peroxide - deuterium oxide solutions,

    M. K. Phibbs and P. A. Giguère, “Hydrogen peroxide and its analogues: I. Density, refractive index, viscosity, and surface tension of deuterium peroxide - deuterium oxide solutions,”Canadian Journal of Chemistry, vol. 29, no. 2, pp. 173–181, Feb. 1951. doi: 10.1139/v51-022. 14 REDUCING EXPERIMENTAL TESTING VIA GENERATIVE IMAGE INTERPOLATION

    work page doi:10.1139/v51-022 1951

  25. [25]

    unwrap_labels,

    A. Zankevich, “unwrap_labels,” GitHub repository, https://github.com/Nepherhotep/unwrap_labels, Accessed: Jun. 5, 2025

    2025

  26. [26]

    The OpenCV Library,

    G. Bradski, “The OpenCV Library,”Dr . Dobb’s Journal of Software Tools, 2000

    2000

  27. [27]

    Image quality metrics: PSNR vs. SSIM,

    A. Horé and D. Ziou, “Image quality metrics: PSNR vs. SSIM,” inProc. 2010 20th Int. Conf. on Pattern Recog- nition (ICPR), Aug. 2010, pp. 2366–2369. doi: 10.1109/ICPR.2010.579

    work page doi:10.1109/icpr.2010.579 2010

  28. [28]

    Image quality assessment through FSIM, SSIM, MSE and PSNR– A comparative study,

    U. Sara, M. Akter, and M. S. Uddin, “Image quality assessment through FSIM, SSIM, MSE and PSNR– A comparative study,”Journal of Computer and Communications, vol. 7, no. 3, pp. 8–18, Mar. 2019. doi: 10.4236/jcc.2019.73002

    work page doi:10.4236/jcc.2019.73002 2019

  29. [29]

    Video quality assessment through PSNR estimation for different compression standards,

    R. G. Deshpande, L. L. Ragha, and S. K. Sharma, “Video quality assessment through PSNR estimation for different compression standards,”Indonesian Journal of Electrical Engineering and Computer Science, vol. 11, no. 3, 2018

    2018

  30. [30]

    Cosine similarity to determine similarity measure: Study case in online essay assessment,

    A. R. Lahitani, A. E. Permanasari, and N. A. Setiawan, “Cosine similarity to determine similarity measure: Study case in online essay assessment,” inProc. 2016 4th Int. Conf. on Cyber and IT Service Management (CITSM), Apr. 2016, pp. 1–6. doi: 10.1109/CITSM.2016.7577578

    work page doi:10.1109/citsm.2016.7577578 2016

  31. [31]

    An overview of perceptual hashing,

    H. Farid, “An overview of perceptual hashing,”Journal of Online Trust and Safety, vol. 1, no. 1, Oct. 2021. doi: 10.54501/jots.v1i1.24

    work page doi:10.54501/jots.v1i1.24 2021

  32. [32]

    scikit-image: Image processing in Python,

    S. van der Walt, J. L. Schönberger, J. Nunez-Iglesias, F. Boulogne, J. D. Warner, N. Yager, E. Gouillart, T. Yu, and the scikit-image contributors, “scikit-image: Image processing in Python,”PeerJ, vol. 2, p. e453, Jun. 2014. doi: 10.7717/peerj.453

    work page doi:10.7717/peerj.453 2014

  33. [33]

    cosine_similarity.py,

    O. Okonji, “cosine_similarity.py,” GitHub Gist, https://gist.github.com/onyekaokonji/7224787b3f127003b14f6ab 36570622c #file-cosine _similarity-py, Accessed: Jun. 5, 2025

    work page arXiv 2025

  34. [34]

    ImageHash,

    J. Buchner, “ImageHash,” GitHub repository, https://github.com/JohannesBuchner/imagehash, Accessed: Jun. 5, 2025. 15

    2025