pith. sign in

arxiv: 2512.03280 · v2 · pith:CGZXCCDZnew · submitted 2025-12-02 · 💻 cs.LG · cs.AI

BlendedNet++: A dataset and benchmark for field-resolved aerodynamics and inverse design of blended wing body aircraft

Nicholas Sung , Steven Spreizer , Mohamed Elrefaie , Matthew C. Jones , Faez Ahmed This is my paper

Pith reviewed 2026-05-21 17:14 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords blended wing bodyaerodynamic datasetinverse designdiffusion modelsgeometric deep learningRANS simulationslift-to-drag ratiosurrogate modeling
0
0 comments X

The pith

BlendedNet++ dataset of 12,492 BWB shapes trains models to predict fields and generate designs meeting exact lift-to-drag targets.

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

The paper introduces a large collection of blended wing body geometries together with their full Reynolds-averaged Navier-Stokes flow solutions. It demonstrates that geometric deep learning can deliver real-time surface pressure and friction predictions while conditional diffusion models, refined by gradients, can produce multiple valid new shapes for any chosen lift-to-drag ratio. A sympathetic reader sees this as a concrete step that replaces slow iterative CFD loops with direct generation of feasible early-stage concepts. The work therefore supplies both the data and the benchmark pipelines needed to move conceptual BWB design from analysis to synthesis.

Core claim

BlendedNet++ supplies 12,492 distinct BWB geometries, each accompanied by steady RANS data that includes integrated forces and dense surface fields of pressure and skin friction. Five surrogate architectures are benchmarked for field prediction, with Transolver emerging as the most accurate. A generative inverse-design pipeline that combines conditional diffusion models with gradient-based refinement then produces multiple feasible designs whose lift-to-drag ratios match prescribed targets at R-squared greater than 0.99, and these designs are independently confirmed by fresh CFD runs.

What carries the argument

Conditional diffusion model plus gradient-based refinement that enforces lift-to-drag targets on the BlendedNet++ data.

If this is right

  • Surface aerodynamic fields can be evaluated in real time rather than through repeated full-order simulations.
  • Designers can request and receive multiple feasible BWB shapes that meet a chosen lift-to-drag target in a single forward pass.
  • The same pipeline supplies a quantitative benchmark for any future surrogate or generative model aimed at BWB aerodynamics.
  • Early-stage aircraft studies shift from iterative analysis to direct synthesis of candidate geometries.

Where Pith is reading between the lines

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

  • The same data and modeling approach could be extended to include structural or noise constraints without changing the generative backbone.
  • If the sampling density proves sufficient, analogous datasets for other non-conventional aircraft families would enable comparable inverse-design workflows.
  • The reported CFD verification loop offers a practical test that future papers can adopt as a standard acceptance criterion for generated shapes.

Load-bearing premise

The 12,492 geometries and their RANS simulations sample the high-dimensional BWB design space densely enough for trained models to produce physically valid shapes for lift-to-drag targets lying outside the training set.

What would settle it

Generate a design for a lift-to-drag target never seen during training, run independent high-fidelity CFD on that geometry, and check whether the achieved lift-to-drag ratio falls within the reported accuracy band.

Figures

Figures reproduced from arXiv: 2512.03280 by Faez Ahmed, Matthew C. Jones, Mohamed Elrefaie, Nicholas Sung, Steven Spreizer.

Figure 1
Figure 1. Figure 1: BlendedNet++ overview: (left) geometry parameterization and meshing, (middle) surrogate benchmarks [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Parameterization of the BWB planform [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Representative renderings from BlendedNet++, illustrating the shape diversity across the dataset. The [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Example from BlendedNet++, showing the 3D shape alongside surface flow quantities. From left to right: [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Example visualization of an automatically generated computational fluid dynamics mesh. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Scatter plots of aerodynamic relationships: (Left) Lift coefficient ( [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: CFD-validated inverse designs: FUN3D L/D versus surrogate-predicted L/D for one CDM→Opt sample per condition. Overall correlation is R2 = 0.9974. operator-learning studies on full volumetric flow fields in the current release. (3) CFD and meshing fidelity. Automated meshing at scale can introduce variability in near-wall resolution and element quality, and steady RANS with a single turbulence model may dev… view at source ↗
Figure 8
Figure 8. Figure 8: Distributions (counts) of normalized planform parameters for BlendedNet++. Each panel shows the marginal [PITH_FULL_IMAGE:figures/full_fig_p019_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Flight-condition distributions for BlendedNet++: altitude, Mach number, Reynolds number (log-scaled on [PITH_FULL_IMAGE:figures/full_fig_p019_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: t-SNE of PointNet autoencoder latent space for BlendedNet and BlendedNet++. Each point is one geome [PITH_FULL_IMAGE:figures/full_fig_p020_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Distribution of flight conditions using Altitude, Reynolds Number, Mach Number and Angle of Attack. [PITH_FULL_IMAGE:figures/full_fig_p020_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Distributions of CD, CL, CMy, and efficiency CL/CD for BlendedNet vs. BlendedNet++. A.3 Forward Surrogate Training Details A.3.1 Common preprocessing We follow the default settings in the NeuralSolver library [54]. The 3 condition inputs log10 ReL, M∞, α are standardized to have zero mean and unit variance. Each output channel Cp, Cfx , Cfz  is z scored independently. During training we draw a uniform s… view at source ↗
read the original abstract

The conceptual design of Blended Wing Body (BWB) aircraft is often constrained by the high computational cost of resolving complex aerodynamics over a high-dimensional design space. While deep learning offers a pathway to rapid aerodynamic prediction and inverse design, its adoption in aerospace engineering is limited by a lack of large-scale, field-resolved training data. This work addresses this gap by introducing BlendedNet++, a comprehensive aerodynamic dataset comprising 12,492 unique BWB geometries, each evaluated using steady Reynolds-Averaged Navier--Stokes (RANS) simulations to provide integrated forces and dense surface fields (Cp, Cf). Leveraging this data, we establish a robust framework for two critical engineering tasks: (1) real-time prediction of surface aerodynamic fields using geometric deep learning models, and (2) generative inverse design. We benchmark five surrogate architectures, identifying Transolver as the most accurate for field predictions. Furthermore, we demonstrate a generative inverse design pipeline using conditional diffusion models combined with gradient-based refinement. This hybrid approach is shown to generate multiple feasible designs that satisfy specific lift-to-drag targets with high accuracy (R^2 > 0.99), as confirmed by computational fluid dynamics (CFD) simulation. These resources enable a shift from iterative analysis to direct generation in early-stage BWB design.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces BlendedNet++, a dataset of 12,492 unique BWB geometries each evaluated with steady RANS simulations to supply integrated forces and dense surface fields (Cp, Cf). It benchmarks five geometric deep learning surrogate architectures for real-time prediction of aerodynamic surface fields, identifying Transolver as the most accurate, and presents a generative inverse design pipeline that combines conditional diffusion models with gradient-based refinement to produce designs meeting specified lift-to-drag targets, reporting R² > 0.99 on subsequent CFD verification.

Significance. If the inverse-design results prove robust, the work supplies a valuable large-scale, field-resolved resource that could accelerate early-stage BWB conceptual design by moving from iterative CFD analysis to direct generation of performance-targeted geometries. The dataset itself and the reported CFD-confirmed accuracy on the generative task constitute concrete contributions to data-driven aerodynamics.

major comments (2)
  1. [Abstract] Abstract: The central claim that the conditional diffusion + gradient-refinement pipeline generates multiple feasible designs satisfying arbitrary lift-to-drag targets with R² > 0.99 rests on the assumption that the 12,492 RANS-evaluated geometries adequately sample the high-dimensional BWB design manifold (typically 8–15 continuous parameters). No coverage diagnostics (parameter-range histograms, convex-hull volume, or discrepancy measures) are supplied, and no separate metrics are reported for in-distribution versus out-of-range targets, leaving generalization risk unaddressed.
  2. [Inverse-design pipeline and verification sections] Inverse-design pipeline and verification sections: The manuscript provides no information on the data splits used to train the diffusion model, the procedure for selecting or tuning its hyperparameters, or explicit post-generation checks for physical consistency (force balance, moment equilibrium). These omissions make it impossible to determine whether the reported R² > 0.99 reflects genuine extrapolation or merely interpolation within the training distribution.
minor comments (2)
  1. [Dataset description] Dataset description: The symbols Cp and Cf are introduced without explicit definitions or reference to standard aerodynamic conventions; a brief parenthetical clarification would improve accessibility.
  2. [Figure captions] Figure captions: Captions for the generated-geometry figures should list the prescribed lift-to-drag target together with the CFD-verified value achieved by each design.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback. We address each major comment below and have revised the manuscript to incorporate additional methodological details and diagnostics where the original submission was incomplete.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that the conditional diffusion + gradient-refinement pipeline generates multiple feasible designs satisfying arbitrary lift-to-drag targets with R² > 0.99 rests on the assumption that the 12,492 RANS-evaluated geometries adequately sample the high-dimensional BWB design manifold (typically 8–15 continuous parameters). No coverage diagnostics (parameter-range histograms, convex-hull volume, or discrepancy measures) are supplied, and no separate metrics are reported for in-distribution versus out-of-range targets, leaving generalization risk unaddressed.

    Authors: We agree that explicit coverage diagnostics and in-distribution versus out-of-range metrics were not provided in the original manuscript. The 12,492 geometries were generated by varying 12 continuous design parameters over ranges informed by existing BWB literature, but without accompanying visualizations or split metrics this coverage was not transparent. In the revised manuscript we have added parameter-range histograms, a description of the Latin-hypercube sampling procedure, and separate R² values computed on targets inside versus near the boundary of the training convex hull. These additions directly address the generalization concern while remaining within the scope of the existing dataset. revision: yes

  2. Referee: [Inverse-design pipeline and verification sections] Inverse-design pipeline and verification sections: The manuscript provides no information on the data splits used to train the diffusion model, the procedure for selecting or tuning its hyperparameters, or explicit post-generation checks for physical consistency (force balance, moment equilibrium). These omissions make it impossible to determine whether the reported R² > 0.99 reflects genuine extrapolation or merely interpolation within the training distribution.

    Authors: We acknowledge that the original submission omitted these implementation details. The revised manuscript now contains an expanded 'Inverse Design Pipeline' subsection that specifies the 80/10/10 train/validation/test split used for the diffusion model, the Bayesian optimization procedure employed for hyperparameter selection on the validation set, and the explicit physical-consistency checks performed during gradient-based refinement (enforcing lift, drag, and moment equilibrium within CFD-verified tolerances). These clarifications demonstrate that the reported accuracy is obtained under standard training protocols and post-generation physical constraints. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper's core contribution is the release of a new dataset of 12,492 RANS-evaluated BWB geometries, followed by training of surrogate models (e.g., Transolver) on that data for field prediction and a conditional diffusion + gradient-refinement pipeline for inverse design. The headline performance claim (R^2 > 0.99 on L/D targets) is established by running fresh CFD simulations on the generated geometries; these verification runs are independent of the training set and are not obtained by re-fitting or re-using the original data points. No equations, parameters, or self-citations are shown to reduce the generative result to the input data by construction. The derivation chain therefore remains externally falsifiable and self-contained.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Central claims rest on RANS simulations serving as accurate ground truth for both training and verification, plus the assumption that geometric deep learning and diffusion models can capture the relevant physics from the sampled geometries without explicit physical constraints.

free parameters (1)
  • diffusion model and refinement hyperparameters
    Chosen during training of the generative pipeline; not enumerated in abstract.
axioms (1)
  • domain assumption Steady RANS equations with chosen turbulence closure accurately represent the relevant aerodynamics for the BWB geometries considered.
    Invoked to generate all ground-truth surface fields and forces used for training and validation.

pith-pipeline@v0.9.0 · 5779 in / 1374 out tokens · 94852 ms · 2026-05-21T17:14:25.403964+00:00 · methodology

discussion (0)

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

Forward citations

Cited by 2 Pith papers

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

  1. CarCrashNet: A Large-Scale Dataset and Hierarchical Neural Solver for Data-Driven Structural Crash Simulation

    cs.LG 2026-05 unverdicted novelty 6.0

    CarCrashNet releases a large-scale open benchmark dataset of structural crash simulations and a hierarchical neural solver for data-driven full-vehicle crash prediction.

  2. CarCrashNet: A Large-Scale Dataset and Hierarchical Neural Solver for Data-Driven Structural Crash Simulation

    cs.LG 2026-05 accept novelty 6.0

    CarCrashNet supplies a large multi-modal crash simulation benchmark and CrashSolver neural model for data-driven full-vehicle crash prediction, validated against experiments and commercial solvers.

Reference graph

Works this paper leans on

55 extracted references · 55 canonical work pages · cited by 1 Pith paper

  1. [1]

    R. H. Liebeck. Design of the blended wing body subsonic transport.Journal of Aircraft, 41(1):10–25, 2004. doi: 10.2514/1. 9244. 15 BlendedNet++: A Large-Scale Blended Wing Body Aerodynamics Dataset and Benchmark

    work page doi:10.2514/1 2004

  2. [2]

    Blended-wing-body transonic aerodynamics: Summary of ground tests and sample results

    Melissa Carter, Dan Vicroy, and Dharmendra Patel. Blended-wing-body transonic aerodynamics: Summary of ground tests and sample results. In47th AIAA Aerospace Sciences Meeting including the New Horizons Forum and Aerospace Exposition, Orlando, FL, Jan 2009. doi: 10.2514/6.2009-935. AIAA Paper 2009-935

    work page doi:10.2514/6.2009-935 2009

  3. [3]

    Assessment on critical technologies for conceptual design of blended-wing-body civil aircraft.Chinese Journal of Aeronautics, 32(8): 1797–1827, Aug 2019

    Zhenli Chen, Minghui Zhang, Yingchun Chen, Weimin Sang, Zhaoguang Tan, Dong Li, and Binqian Zhang. Assessment on critical technologies for conceptual design of blended-wing-body civil aircraft.Chinese Journal of Aeronautics, 32(8): 1797–1827, Aug 2019. doi: 10.1016/j.cja.2019.03.008

    work page doi:10.1016/j.cja.2019.03.008 2019

  4. [4]

    Phd dissertation, Technische Universit¨at Braunschweig, 2023

    Achyuth Attravanam.High-fidelity CFD-based Shape Optimization of a Blended-Wing-Body Aircraft for Improved Aerody- namic Performance, Considering Engine Integration Effects. Phd dissertation, Technische Universit¨at Braunschweig, 2023

    work page 2023

  5. [5]

    Chan, and Howard Smith

    Shang Lyu, Yicheng Sun, Joseph L. Chan, and Howard Smith. Blended wing body aircraft conceptual design optimisation with nonlinear multi-fidelity aerodynamic surrogate model. InAIAA Aviation Forum and ASCEND, 2024. doi: 10.2514/6. 2024-3979. AIAA Paper 2024-3979

    work page doi:10.2514/6 2024

  6. [6]

    Deep learning in aircraft design, dynamics, and control: Review and prospects.IEEE Transactions on Aerospace and Electronic Systems, 57(4):2346–2368, 2021

    Yiqun Dong, Jun Tao, Youmin Zhang, Wei Lin, and Jianliang Ai. Deep learning in aircraft design, dynamics, and control: Review and prospects.IEEE Transactions on Aerospace and Electronic Systems, 57(4):2346–2368, 2021. doi: 10.1109/ TAES.2021.3056082

    work page arXiv 2021

  7. [7]

    Fast predictions of aircraft aerodynamics using deep-learning techniques.AIAA Journal, 60(9):5249–5261, September 2022

    Christian Sabater, Philipp St ¨urmer, and Philipp Bekemeyer. Fast predictions of aircraft aerodynamics using deep-learning techniques.AIAA Journal, 60(9):5249–5261, September 2022. doi: 10.2514/1.J061145

    work page doi:10.2514/1.j061145 2022

  8. [8]

    Mohamed Elrefaie, Florin Morar, Angela Dai, and Faez Ahmed. Drivaernet++: A large-scale multimodal car dataset with computational fluid dynamics simulations and deep learning benchmarks.Advances in Neural Information Processing Sys- tems, 37:499–536, 2024

    work page 2024

  9. [9]

    Drivaernet: A parametric car dataset for data-driven aerodynamic design and prediction.Journal of Mechanical Design, 147(4):041712, 2025

    Mohamed Elrefaie, Angela Dai, and Faez Ahmed. Drivaernet: A parametric car dataset for data-driven aerodynamic design and prediction.Journal of Mechanical Design, 147(4):041712, 2025

    work page 2025

  10. [10]

    Drivaernet: A parametric car dataset for data-driven aerodynamic design and graph-based drag prediction

    Mohamed Elrefaie, Angela Dai, and Faez Ahmed. Drivaernet: A parametric car dataset for data-driven aerodynamic design and graph-based drag prediction. InInternational Design Engineering Technical Conferences and Computers and Information in Engineering Conference, volume 88360, page V03AT03A019. American Society of Mechanical Engineers, 2024

    work page 2024

  11. [11]

    Jones, and Faez Ahmed

    Nicholas Sung, Steven Spreizer, Mohamed Elrefaie, Kaira Samuel, Matthew C. Jones, and Faez Ahmed. Blendednet: A blended wing body aircraft dataset and surrogate model for aerodynamic predictions. InProceedings of the ASME 2025 Inter- national Design Engineering Technical Conferences and Computers and Information in Engineering Conference (IDETC/CIE 2025),...

    work page 2025

  12. [12]

    M. H. Zhang, Z. L. Chen, and B. Q. Zhang. A conceptual design platform for blended wing-body transports. InProceedings of the 30th Congress of the International Council of the Aeronautical Sciences (ICAS), pages 25–30, Daejeon, South Korea, Sep 2016

    work page 2016

  13. [13]

    Multifidelity data fusion: Application to blended-wing- body multidisciplinary analysis under uncertainty.AIAA Journal, 58(2):889–906, 2020

    Alex Feldstein, David Lazzara, Norman Princen, and Karen Willcox. Multifidelity data fusion: Application to blended-wing- body multidisciplinary analysis under uncertainty.AIAA Journal, 58(2):889–906, 2020

    work page 2020

  14. [14]

    Lynch, Shaunak D

    Soumalya Sarkar, Sudeepta Mondal, Michael Joly, Matthew E. Lynch, Shaunak D. Bopardikar, Ranadip Acharya, and Paris Perdikaris. Multifidelity and multiscale bayesian framework for high-dimensional engineering design and calibration. volume 141, page 121001, Dec 2019. doi: 10.1115/1.4044543

    work page doi:10.1115/1.4044543 2019

  15. [15]

    Multi-fidelity design framework integrating compositional kernels to facilitate early-stage design exploration of complex systems.Journal of Mechanical Design, 147(1):011701, 2025

    Nikoleta Dimitra Charisi, Hans Hopman, and Austin A Kana. Multi-fidelity design framework integrating compositional kernels to facilitate early-stage design exploration of complex systems.Journal of Mechanical Design, 147(1):011701, 2025

    work page 2025

  16. [16]

    Loeser and Andreas Schuette

    Thomas D. Loeser and Andreas Schuette. Saccon forced oscillation tests at dnw-nwb and nasa langley 14x22-foot tunnel. In 28th AIAA Applied Aerodynamics Conference, Chicago, IL, Jun 2010. doi: 10.2514/6.2010-4394. AIAA Paper 2010-4394

    work page doi:10.2514/6.2010-4394 2010

  17. [17]

    Melissa B. Rivers. Nasa common research model: A history and future plans. InAIAA Aviation 2019 Forum, Dallas, TX, Jun

    work page 2019

  18. [18]

    AIAA Paper 2019-3725

    doi: 10.2514/6.2019-3725. AIAA Paper 2019-3725

    work page doi:10.2514/6.2019-3725 2019

  19. [19]

    Dpw-5 analysis of the crm in a wing-body configuration using structured and unstructured meshes

    Anthony Sclafani, John Vassberg, Mortaza Mani, Chad Winkler, Andrew Dorgan, Michael Olsen, and James Coder. Dpw-5 analysis of the crm in a wing-body configuration using structured and unstructured meshes. In51st AIAA Aerospace Sciences Meeting Including the New Horizons Forum and Aerospace Exposition, Grapevine, TX, Jan 2013. doi: 10.2514/6.2013-0048. AIA...

    work page doi:10.2514/6.2013-0048 2013

  20. [20]

    Going for experimental and numerical unsteady wake analyses combined with wall interference assessment by using the nasa crm-model in etw

    Thorsten Lutz. Going for experimental and numerical unsteady wake analyses combined with wall interference assessment by using the nasa crm-model in etw. In51st AIAA Aerospace Sciences Meeting Including the New Horizons Forum and Aerospace Exposition, Grapevine, TX, Jan 2013. doi: 10.2514/6.2013-0871. AIAA Paper 2013-0871

    work page doi:10.2514/6.2013-0871 2013

  21. [21]

    Aircraftverse: A large-scale multimodal dataset of aerial vehicle designs.Advances in Neural Information Processing Systems, 36:44524–44543, Dec 2023

    Adam Cobb, Anirban Roy, Daniel Elenius, Frederick Heim, Brian Swenson, Sydney Whittington, James Walker, Theodore Bapty, Joseph Hite, Karthik Ramani, and Craig McComb. Aircraftverse: A large-scale multimodal dataset of aerial vehicle designs.Advances in Neural Information Processing Systems, 36:44524–44543, Dec 2023

    work page 2023

  22. [22]

    Edwards, Vaishnavi L

    Kristen M. Edwards, Vaishnavi L. Addala, and Faez Ahmed. Design form and function prediction from a single image. In Proceedings of the ASME 2021 International Design Engineering Technical Conferences and Computers and Information in Engineering Conference (IDETC-CIE), volume 85376, page V002T02A032, Virtual Conference, Aug 2021. American Society of Mecha...

    work page doi:10.1115/detc2021-68566 2021

  23. [23]

    Mart ´ın, Andr´es Mateo-Gab´ın, Thomas Wagenaar, Gonzalo Rubio, and Sven A

    Marta A. Mart ´ın, Andr´es Mateo-Gab´ın, Thomas Wagenaar, Gonzalo Rubio, and Sven A. Lanzan Ferran. Ai-based generative algorithms applied to the design of blended wing body aircraft. InAIAA Aviation Forum and ASCEND 2025, Las Vegas, NV , United States, July 2025. doi: 10.2514/6.2025-3292

    work page doi:10.2514/6.2025-3292 2025

  24. [24]

    Joaquim R. R. A. Martins and Andrew B. Lambe. Multidisciplinary design optimization: A survey of architectures.AIAA Journal, 51(9):2049–2075, 2013. doi: 10.2514/1.J051895

    work page doi:10.2514/1.j051895 2049

  25. [25]

    Zhoujie Lyu and Joaquim R. R. A. Martins. Aerodynamic design optimization studies of a blended-wing-body aircraft. Journal of Aircraft, 51(5):1604–1617, 2014. doi: 10.2514/1.C032491

    work page doi:10.2514/1.c032491 2014

  26. [26]

    Laban, K

    Ning Qin, Armando Vavalle, Alan Le Moigne, M. Laban, K. Hackett, and P. Weinerfelt. Aerodynamic considerations of blended wing body aircraft.Progress in Aerospace Sciences, 40(6):321–343, Aug 2004. doi: 10.1016/j.paerosci.2004.07.001

    work page doi:10.1016/j.paerosci.2004.07.001 2004

  27. [27]

    Wayne Mastin, Robert E

    C. Wayne Mastin, Robert E. Smith, Ideen Sadrehaghighi, and Michael R. Wiese. Geometric model for a parametric study of the blended-wing-body airplane. In14th Applied Aerodynamics Conference, pages AIAA–96–2416. AIAA, 1996

    work page 1996

  28. [28]

    Challenges in realizing 3rd generation multidisciplinary design optimization.AIMS Applied Com- puting and Simulation Engineering, 2025

    Alexandru Antonau et al. Challenges in realizing 3rd generation multidisciplinary design optimization.AIMS Applied Com- puting and Simulation Engineering, 2025. doi: 10.3934/acse.2025001

    work page doi:10.3934/acse.2025001 2025

  29. [29]

    Framework of airfoil max lift-to-drag ratio prediction using hybrid feature mining and gaussian process regression.Energy Conversion and Management, 243:114339, Sep 2021

    Yaoran Chen, Zhikun Dong, Jie Su, Yan Wang, Zhaolong Han, Dai Zhou, Yongsheng Zhao, and Yan Bao. Framework of airfoil max lift-to-drag ratio prediction using hybrid feature mining and gaussian process regression.Energy Conversion and Management, 243:114339, Sep 2021. doi: 10.1016/j.enconman.2021.114339

    work page doi:10.1016/j.enconman.2021.114339 2021

  30. [30]

    X. Liu, Q. Zhu, and H. Lu. Modeling multiresponse surfaces for airfoil design with multiple-output-gaussian-process regres- sion.Journal of Aircraft, 51(3):740–747, May 2014. doi: 10.2514/1.C032493

    work page doi:10.2514/1.c032493 2014

  31. [31]

    Artificial neural networks to predict aerodynamic coefficients of transport airplanes.Aircraft Engineering and Aerospace Technology, 89(2):211–230, Mar 2017

    Ney Rafael Secco and Bento Silva de Mattos. Artificial neural networks to predict aerodynamic coefficients of transport airplanes.Aircraft Engineering and Aerospace Technology, 89(2):211–230, Mar 2017. doi: 10.1108/AEAT-06-2015-0151

    work page doi:10.1108/aeat-06-2015-0151 2017

  32. [32]

    A machine learning enabled multi-fidelity platform for the integrated design of aircraft systems.Journal of Mechanical Design, 141(12):121405, 2019

    Ana Garcia Garriga, Laura Mainini, and Sangeeth Saagar Ponnusamy. A machine learning enabled multi-fidelity platform for the integrated design of aircraft systems.Journal of Mechanical Design, 141(12):121405, 2019

    work page 2019

  33. [33]

    Tripnet: Learning large-scale high-fidelity 3d car aerodynamics with triplane networks.arXiv preprint arXiv:2503.17400, 2025

    Qian Chen, Mohamed Elrefaie, Angela Dai, and Faez Ahmed. Tripnet: Learning large-scale high-fidelity 3d car aerodynamics with triplane networks.arXiv preprint arXiv:2503.17400, 2025

    work page arXiv 2025

  34. [34]

    Qi, Hao Su, Kaichun Mo, and Leonidas J

    Charles R. Qi, Hao Su, Kaichun Mo, and Leonidas J. Guibas. Pointnet: Deep learning on point sets for 3d classification and segmentation. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 652–660. IEEE, 2017. doi: 10.1109/CVPR.2017.16

    work page doi:10.1109/cvpr.2017.16 2017

  35. [35]

    Hamilton, Rex Ying, and Jure Leskovec

    William L. Hamilton, Rex Ying, and Jure Leskovec. Inductive representation learning on large graphs. InAdvances in Neural Information Processing Systems (NeurIPS), pages 1025–1035, 2017

    work page 2017

  36. [36]

    Graph u-nets

    Hongyang Gao and Shuiwang Ji. Graph u-nets. InProceedings of the 36th International Conference on Machine Learning (ICML), 2019

    work page 2019

  37. [37]

    Gnot: A general neural operator transformer for operator learning, 2023

    Zhongkai Hao, Zhengyi Wang, Hang Su, Chengyang Ying, Yinpeng Dong, Songming Liu, Ze Cheng, Jian Song, and Jun Zhu. Gnot: A general neural operator transformer for operator learning, 2023

    work page 2023

  38. [38]

    Transolver: A fast transformer solver for pdes on general geometries

    Haixu Wu, Huakun Luo, Haowen Wang, Jianmin Wang, and Mingsheng Long. Transolver: A fast transformer solver for pdes on general geometries. InInternational Conference on Machine Learning, 2024

    work page 2024

  39. [39]

    FiLM: Visual reasoning with a general conditioning layer

    Ethan Perez, Florian Strub, Harm de Vries, Vincent Dumoulin, and Aaron Courville. FiLM: Visual reasoning with a general conditioning layer. InProceedings of the AAAI Conference on Artificial Intelligence, volume 32, pages 3942–3951, Apr 2018

    work page 2018

  40. [40]

    Sharma and Serhat Hosder

    Rohan S. Sharma and Serhat Hosder. Mission-driven inverse design of blended wing body aircraft with machine learning. Aerospace, 11(2):137, Feb 2024. doi: 10.3390/aerospace11020137

    work page doi:10.3390/aerospace11020137 2024

  41. [41]

    Neural fields for rapid aircraft aerodynamics simulations.Scientific Reports, 14(1):25496, Oct 2024

    Giovanni Catalani, Siddhant Agarwal, Xavier Bertrand, Fr ´ed´eric Tost, Michael Bauerheim, and Joseph Morlier. Neural fields for rapid aircraft aerodynamics simulations.Scientific Reports, 14(1):25496, Oct 2024. doi: 10.1038/s41598-024-32213-z

    work page doi:10.1038/s41598-024-32213-z 2024

  42. [42]

    Diffairfoil: An efficient novel airfoil sampler based on latent space diffusion model for aerodynamic shape optimization

    Zhen Wei, Edouard Dufour, Colin Pelletier, Micha ¨el Bauerheim, and Pascal Fua. Diffairfoil: An efficient novel airfoil sampler based on latent space diffusion model for aerodynamic shape optimization. InAIAA AVIATION Forum, 2024. doi: 10.2514/6.2024-3755

    work page doi:10.2514/6.2024-3755 2024

  43. [43]

    Airfoil diffusion: Denoising diffusion model for conditional airfoil generation, 2024

    Reid Graves and Amir Barati Farimani. Airfoil diffusion: Denoising diffusion model for conditional airfoil generation, 2024

    work page 2024

  44. [44]

    Multi-point aerodynamic inverse design of flying wing using the conditional diffusion model.Physics of Fluids, 37(7):077116, 2025

    Jiahao Lin, Shusheng Chen, Shiyi Jin, Quanfeng Jiang, Muliang Jia, and Dong Li. Multi-point aerodynamic inverse design of flying wing using the conditional diffusion model.Physics of Fluids, 37(7):077116, 2025. doi: 10.1063/5.0271723

    work page doi:10.1063/5.0271723 2025

  45. [45]

    Cooling-guided diffusion model for battery cell ar- rangement

    Nicholas Sung, Zhiyao Liu, Peishan Wang, and Faez Ahmed. Cooling-guided diffusion model for battery cell ar- rangement. InProceedings of the ASME 2024 International Design Engineering Technical Conferences and Computers and Information in Engineering Conference, page V03AT03A009, Washington, DC, USA, August 2024. ASME. doi: 10.1115/DETC2024-143373. V olum...

    work page doi:10.1115/detc2024-143373 2024

  46. [46]

    Conceptual design of a blended wing body airliner

    Jeffrey Trac-Pho. Conceptual design of a blended wing body airliner. Master’s thesis, San Jos´e State University, San Jose, Cal- ifornia, December 2022. Available athttps://www.sjsu.edu/ae/docs/project-thesis/Jeffrey.Trac-Pho-F22. pdf. 17 BlendedNet++: A Large-Scale Blended Wing Body Aerodynamics Dataset and Benchmark

    work page 2022

  47. [47]

    McDonald and James R

    Robert A. McDonald and James R. Gloudemans. Open vehicle sketch pad: An open source parametric geometry and analysis tool for conceptual aircraft design. InAIAA SciTech 2022 Forum, San Diego, CA, Jan 2022. doi: 10.2514/6.2022-0004. AIAA Paper 2022-0004

    work page doi:10.2514/6.2022-0004 2022

  48. [48]

    Fidelity pointwise (version 2024.1)

    Cadence Design Systems, Inc. Fidelity pointwise (version 2024.1). https://www.cfd-technologies.co.uk/fidelity-pointwise, March 2025. Accessed: 2025-03-12

    work page 2024

  49. [49]

    Anderson, Robert T

    William K. Anderson, Robert T. Biedron, Jan-Rene ´e Carlson, Joseph M. Derlaga, Boris Diskin, Cameron T. Druyor Jr, Peter A. Gnoffo, Dana P. Hammond, Kevin E. Jacobson, William T. Jones, and William L. Kleb.FUN3D Manual: 14.1. NASA Langley Research Center, Hampton, V A, 2024. NASA Technical Manual

    work page 2024

  50. [50]

    Anderson.Fundamentals of aerodynamics

    John D. Anderson.Fundamentals of aerodynamics. McGraw-Hill series in aeronautical and aerospace engineering. Mc- Graw Hill LLC, New York, NY , seventh edition. edition, 2024. URLhttps://ebookcentral.proquest.com/lib/mit/ detail.action?docID=7192797

    work page 2024

  51. [51]

    Kitware, 2006

    Will Schroeder, Ken Martin, and Bill Lorensen.The Visualization Toolkit (4th ed.). Kitware, 2006. ISBN 978-1-930934-19-1

    work page 2006

  52. [52]

    Interactive supercomputing on 40,000 cores for machine learning and data analysis

    Albert Reuther, Jeremy Kepner, Chansup Byun, Siddharth Samsi, William Arcand, David Bestor, Bill Bergeron, Vijay Gade- pally, Michael Houle, Matthew Hubbell, Michael Jones, Anna Klein, Lauren Milechin, Julia Mullen, Andrew Prout, Antonio Rosa, Charles Yee, and Peter Michaleas. Interactive supercomputing on 40,000 cores for machine learning and data analys...

    work page doi:10.1109/hpec.2018.8547629 2018

  53. [53]

    Airfrans: High fidelity computational fluid dynamics dataset for approximating reynolds-averaged navier–stokes solutions

    Florent Bonnet, Jocelyn Ahmed Mazari, Paola Cinnella, and Patrick Gallinari. Airfrans: High fidelity computational fluid dynamics dataset for approximating reynolds-averaged navier–stokes solutions. InNeurIPS 2022 Datasets and Benchmarks Track, 2022. URLhttps://openreview.net/forum?id=HEJ2K8QLHwJ

    work page 2022

  54. [54]

    Special issue: Design by data: Cultivating datasets for engineering design.Journal of Mechanical Design, 147(4):040301, 02 2025

    Faez Ahmed, Cyril Picard, Wei Chen, Christopher McComb, Pingfeng Wang, Ikjin Lee, Tino Stankovic, Douglas Allaire, and Stefan Menzel. Special issue: Design by data: Cultivating datasets for engineering design.Journal of Mechanical Design, 147(4):040301, 02 2025. ISSN 1050-0472. doi: 10.1115/1.4067871. URLhttps://doi.org/10.1115/1.4067871

    work page doi:10.1115/1.4067871 2025

  55. [55]

    Neural-solver-library: A library for advanced neural pde solvers

    Tsinghua University Machine Learning Group (THUML). Neural-solver-library: A library for advanced neural pde solvers. https://github.com/thuml/Neural-Solver-Library, 2025. Last accessed: September 29, 2025. 18 BlendedNet++: A Large-Scale Blended Wing Body Aerodynamics Dataset and Benchmark A Appendix A.1 Supplementary Dataset Distributions Figures 8 and 9...

    work page 2025