Pith. sign in

REVIEW 3 major objections 7 minor 64 references

Uncertainty-triggered multi-fidelity surrogates cut RANS use to under 15% while lifting airfoil cruise efficiency 41% and take-off lift 21% on a two-point design.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-13 23:21 UTC pith:M7EPNERS

load-bearing objection Solid engineering integration of LF-informed GPs with uncertainty-triggered RANS and synchronized elitism; the cost claim is real but only vs an all-RANS counterfactual of the same trajectory, not equal-budget baselines. the 3 major comments →

arxiv 2603.17057 v2 pith:M7EPNERS submitted 2026-03-17 physics.flu-dyn cs.LGcs.NEmath.OC

Optimization-Embedded Active Multi-Fidelity Surrogate Learning for Multi-Condition Airfoil Shape Optimization

Isaac Robledo , Alberto Vilari\~no , Arnau Mir\'o , Oriol Lehmkuhl , Carlos Sanmiguel Vila , Rodrigo Castellanos This is my paper
classification physics.flu-dyn cs.LGcs.NEmath.OC
keywords multi-fidelity surrogateactive learningairfoil shape optimizationGaussian process regressionCST parametrizationRANSXFOILhybrid genetic algorithm
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Airfoil shape optimization usually needs many expensive high-fidelity CFD runs. This paper claims you can keep RANS-level answers while spending far fewer of those runs by embedding an active multi-fidelity surrogate inside a hybrid genetic optimizer. Cheap panel-method (XFOIL) evaluations supply features; a Gaussian-process transfer map turns them into high-fidelity predictions and also reports when those predictions are untrustworthy. Only then—or for elite individuals that must stay consistent—is a RANS simulation run, and the population is re-scored so evolution never ranks designs on an outdated surrogate. On a two-point problem (cruise L/D at 2° and take-off lift at 10°, Re=6e6, 12-parameter CST shapes), the method improves the best first-generation design by 41% in cruise efficiency and 21% in take-off lift while using RANS for only about 15% and 9.5% of the condition-wise candidate evaluations. The practical payoff is multi-condition aerodynamic design that stays accurate without an all-RANS budget.

Core claim

An optimization-embedded active multi-fidelity strategy—low-fidelity-informed Gaussian-process transfer maps, uncertainty-triggered RANS calls, condition-wise decoupled surrogates, and synchronized elitism with population re-evaluation—delivers RANS-consistent multi-point airfoil performance while requiring high-fidelity evaluations for only 14.78% (cruise) and 9.5% (take-off) of condition-specific candidates, with measured gains of 41.05% in cruise efficiency and 20.75% in take-off lift relative to the best first-generation individual.

What carries the argument

Uncertainty-triggered, LF-informed GPR transfer: the high-fidelity metric is learned as a nonparametric function of the CST parameters plus the XFOIL output; a coefficient-of-variation threshold decides when to escalate to RANS, after which elites are forced to high fidelity and the whole population is re-evaluated so selection never rests on a stale surrogate.

Load-bearing premise

That gains versus the best first-generation design plus RANS-usage percentages versus an all-RANS counterfactual of the same candidates are enough to prove the framework works, without a same-budget pure high-fidelity evolutionary run or a direct contest against other multi-fidelity methods on this problem.

What would settle it

Re-run the identical two-point CST optimization with a pure high-fidelity hybrid genetic algorithm under the same total wall-clock or core-hour budget; if the final cruise L/D and take-off CL are no better (or worse) than the multi-fidelity result, or if a fixed-schedule multi-fidelity baseline matches the gains at equal or lower RANS count, the central efficiency claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 7 minor

Summary. The manuscript proposes an optimization-embedded active multi-fidelity framework for multi-condition airfoil shape optimization. Low-fidelity XFOIL outputs and CST parameters are mapped to RANS-consistent metrics via independent Gaussian-process transfer models per flight condition; high-fidelity RANS is invoked when a coefficient-of-variation uncertainty metric exceeds a calibrated threshold, with mandatory HF validation of elites and population re-evaluation after surrogate updates (synchronized elitism) inside the hybrid genetic algorithm HyGO. The method is demonstrated on a two-point 12-parameter CST problem at Re=6e6 (cruise α=2°, maximize E=L/D; take-off α=10°, maximize CL). Relative to the best first-generation individual, the optimized design improves cruise efficiency by 41.05% and take-off lift by 20.75%, while RANS is used for only 14.78% (cruise) and 9.5% (take-off) of condition-specific candidate evaluations versus an all-RANS evaluation of the same campaign trajectory.

Significance. If the efficiency and consistency claims hold under fair comparison, the work is a useful engineering contribution to multi-point aerodynamic shape optimization: uncertainty-triggered fidelity escalation at the candidate level, condition-wise decoupled surrogates, and synchronized elitism address practical failure modes of static multi-fidelity models inside evolutionary search. Strengths already present include a mesh-convergence study, explicit HF admissibility checks, pre-update (not in-sample) surrogate RMSRE reporting (E: 0.64→0.09; CL: 0.06→0.01), generation-wise HF accounting, and physically interpretable Cp/flow-field analysis. Planned open-source release of the framework would further increase impact. The main significance risk is that headline cost and improvement numbers are currently framed against weak baselines (all-RANS of the same trajectory; Gen-1 best after dual-fidelity LHS+DSM), so the comparative value of the active multi-fidelity mechanism is not yet fully established.

major comments (3)
  1. Abstract and §3 (cost discussion around Fig. 10): the central cost claim—that RANS is required for only 14.78% (cruise) and 9.5% (take-off) of condition-specific candidate evaluations—is defined relative to evaluating all 1042 campaign individuals at RANS for both conditions (an all-HF counterfactual of the identical search trajectory). That does not establish that the framework finds comparable designs under a fixed HF budget against (i) pure-HF HyGO or (ii) standard multi-fidelity alternatives (e.g., co-Kriging, fixed-schedule or EI-driven infill) on the same 12-parameter CST two-point problem. Without equal-budget controls or a carefully restated claim, the load-bearing efficiency result remains under-supported even though the internal campaign accounting is coherent.
  2. §3, Table 6 and Fig. 5: the reported 41.05% (E) and 20.75% (CL) gains are relative to the best first-generation individual, which already benefits from 83 dual-fidelity LHS designs plus DSM local-search exploitation. This baseline does not isolate the contribution of uncertainty-triggered multi-fidelity refinement and synchronized elitism from ordinary evolutionary progress on a well-initialized population. A same-budget pure-HF run, or at least an ablation that freezes the surrogate after initialization, is needed to support the claim that the active multi-fidelity machinery drives the multi-point gains.
  3. §2.3–2.5 and Algorithm 2: death-penalty treatment of (a) XFOIL failures and (b) individuals that violate CV≥κ after the post-elite surrogate re-evaluation bounds the HF budget but can systematically remove designs near separation or in regions where the LF–HF map is poorly calibrated. For the intended attached-flow mission this may be acceptable, but the manuscript should quantify how often death penalties are applied after re-evaluation, whether elite-adjacent designs are discarded, and discuss the resulting search bias—especially at cruise where drag-sensitive E drives more HF calls and a looser κ.
minor comments (7)
  1. Abstract vs body: the abstract states RANS for “only 14.78% and 9.5% of evaluated individuals,” while the body correctly frames these as fractions of condition-specific candidate evaluations including the initial dual-fidelity set; align wording so the denominator is unambiguous.
  2. Eq. (3): the damping constant ε=10 is large relative to O(1) normalized objectives; a short sensitivity check or justification that ranking is preserved would help readers assess selection pressure.
  3. Table 1 / §2.1: several CST bounds reach ±1 after the 15% expansion; a brief note on whether bound saturation occurred for any optimized or Pareto designs would clarify whether the envelope was active.
  4. Figures 5–7, 12–13: generation markers and color conventions are dense; ensure legends remain readable in grayscale and that the Pareto front construction (how non-dominated points are selected under the scalarized J) is stated in the caption or text.
  5. §2.2 mesh study uses NACA 0012 at α=10°; a short remark that the same automated mesh settings remain adequate for highly cambered CST optima (Fig. 6) would strengthen the HF reference claim.
  6. Typographical/consistency: “high-hidelity” (§2.4), “Cummulative” (Fig. 10 caption), and mixed “multi-fidelity” hyphenation; also reconcile abstract “RANS-consistent” with body “RANS-level accuracy.”
  7. Related work (§1) surveys co-Kriging and multi-fidelity BO but the results do not return to those methods; even a qualitative discussion of why LF-informed GPR transfer was preferred over hierarchical co-Kriging for this XFOIL–RANS pair would help place the contribution.

Circularity Check

0 steps flagged

No circular derivation: empirical multi-fidelity campaign with pre-update HF validation and accounting metrics, not identity-by-construction claims.

full rationale

This is an empirical methods paper. The LF-informed GPR maps (θ, φ_LF) → φ_HF and is assessed on newly acquired HF points before assimilation (pre-update RMSRE), so surrogate quality is not scored in-sample. Uncertainty thresholds κ are calibrated once on a held-out LF sample distribution and then used as fixed escalation rules; they are not fitted to the reported performance gains. The 41.05%/20.75% improvements are measured against the Gen-1 best using RANS-validated elites and triggered HF evaluations, not quantities forced by the cost-function definition. HF-usage percentages (14.78%, 9.5%) are accounting ratios of RANS calls to candidate evaluations versus an all-RANS counterfactual of the same trajectory—definitional bookkeeping, not a prediction that reduces to a fitted identity. Self-citation of HyGO supplies the base evolutionary optimizer; the active multi-fidelity, condition-wise decoupling, and synchronized elitism mechanisms are developed and demonstrated in this work and do not rest on a uniqueness theorem or ansatz imported from prior author papers. Weaknesses in experimental design (lack of equal-budget pure-HF or co-Kriging baselines) are methodological, not circularity. No step reduces Eq. X to Eq. Y by construction.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 2 invented entities

Load-bearing content is engineering assumptions and hand-chosen control knobs, not new physics entities. The claim rests on RANS/XFOIL adequacy in the stated regime, CST feasibility constraints, fixed equal-weight scalarization, and calibrated uncertainty thresholds. Free parameters dominate the ledger; invented entities are algorithmic constructs with no independent physical status.

free parameters (5)
  • Uncertainty thresholds κ (cruise/take-off)
    Set to 90th percentile of CV on 1000 random LF samples after init: κα=2°=0.05, κα=10°=0.02; directly control HF escalation rate and thus the central cost claim.
  • Cost damping ε and equal weights 0.5/0.5
    J = 1/(0.5Ê + 0.5ĈL + ε) with ε=10 and normalization by initial-population means; shapes selection pressure and multi-point trade-off.
  • HyGO GA/DSM hyperparameters
    Ng, population sizes, Pc/Pm/pm, Ne, de, Nexploit, Ns, Nf_b (Table 5) chosen pragmatically; govern search path and when elites force HF.
  • GPR kernel hyperparameters (σf, l, σn)
    Optimized by log-marginal likelihood on accumulating LF/HF pairs; define predictive mean/variance and thus CV triggers.
  • RANS mesh/y+ and admissibility cutoffs
    Finest mesh (~1.2M cells), y+~40, std(CL,CD)/mean < 0.05 over last 2000 steps, positive coefficients; define what counts as valid HF truth.
axioms (5)
  • domain assumption 2D incompressible RANS with k-ω SST and wall functions is an adequate high-fidelity reference for the multi-point metrics at Re=6e6.
    HF truth for training and elite validation is this RANS setup (§2.2); transition/separation physics beyond RANS are out of scope.
  • domain assumption XFOIL LF outputs plus CST parameters are informative features for a nonparametric map to RANS coefficients.
    Core of Eq. (1) and the multi-fidelity environment (§2.3); justified by trend capture in attached regimes.
  • ad hoc to paper CV = σ/φ̂HF with fixed κ is a valid fidelity-escalation policy (not an optimization acquisition function).
    Authors choose CV and 90th-percentile κ rather than EI/UCB multi-fidelity acquisition; policy is design choice (§2.3).
  • ad hoc to paper Death-penalty treatment of LF failures and of re-eval CV≥κ after surrogate update does not fatally bias the search for the intended mission envelope.
    Algorithms 1–2 and §2.5 discard rather than HF-evaluate some uncertain/failed designs to bound budget.
  • domain assumption 12-parameter CST bounds (+15% expansion) and geometric constraints define a design space representative enough for the claimed multi-condition gains.
    §2.1 Tables 1–2; invalid shapes death-penalized.
invented entities (2)
  • Synchronized elitism with mandatory elite HF and population re-evaluation under updated GPR no independent evidence
    purpose: Prevent evolutionary selection on stale surrogate fitness after online model updates.
    Algorithmic mechanism introduced in §2.5 / Algorithm 2; no independent physical evidence required—it is a control rule whose value is empirical.
  • Condition-wise decoupled LF-informed multi-fidelity GPR transfer models no independent evidence
    purpose: Allow independent refinement and κ per flight condition/coefficient.
    Architectural choice (§2.3); standard multi-output alternative exists; evidence is the campaign’s asymmetric HF counts.

pith-pipeline@v1.1.0-grok45 · 28566 in / 3874 out tokens · 37735 ms · 2026-07-13T23:21:57.849905+00:00 · methodology

0 comments
read the original abstract

Active multi-fidelity surrogate modeling is developed for multi-condition airfoil shape optimization to reduce high-fidelity CFD cost while retaining RANS-consistent aerodynamic metrics. The framework couples a low-fidelity-informed Gaussian process regression transfer model with uncertainty-triggered sampling and a synchronized elitism rule embedded in a hybrid genetic algorithm. Low-fidelity XFOIL evaluations provide inexpensive features, while sparse RANS simulations are adaptively allocated when predictive uncertainty exceeds a threshold; elite candidates are mandatorily validated at high fidelity, and the population is re-evaluated to prevent evolutionary selection based on outdated fitness values produced by earlier surrogate states. The method is demonstrated for a two-point problem at $Re=6\times10^6$ with cruise at $\alpha=2^\circ$ (maximize $E=L/D$) and take-off at $\alpha=10^\circ$ (maximize $C_L$) using a 12-parameter CST representation. Independent multi-fidelity surrogates per flight condition enable decoupled refinement. The optimized design improves cruise efficiency by 41.05% and take-off lift by 20.75% relative to the best first-generation individual. Over the full campaign, RANS evaluations were required for only 14.78% and 9.5% of the condition-specific candidate evaluations at cruise and take-off, respectively. These percentages quantify the reduction in high-fidelity usage relative to the fixed automated RANS workflow adopted as the high-fidelity reference in this study.

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

64 extracted references · 13 canonical work pages

  1. [1]

    Jameson, Aerodynamic design via control theory, Journal of scientific computing (1988)

    A. Jameson, Aerodynamic design via control theory, Journal of scientific computing (1988)

    1988

  2. [2]

    Reuther, A

    J. Reuther, A. Jameson, J. Farmer, L. Martinelli, D. Saunders, Aerodynamic shape optimization of complex aircraft configurations via an adjoint for- mulation, in: 34th aerospace sciences meeting and exhibit, 1996, p. 94

    1996

  3. [3]

    J. P. Slotnick, A. Khodadoust, J. Alonso, D. Darmo- fal, W. Gropp, E. Lurie, D. J. Mavriplis, CFD vision 2030 study: a path to revolutionary computational aerosciences, Technical Report No. NF1676L-18332, NASA, 2014

    2030

  4. [4]

    Yondo, E

    R. Yondo, E. Andrés, E. Valero, A review on de- sign of experiments and surrogate models in air- craft real-time and many-query aerodynamic anal- yses, Progress in aerospace sciences (2018)

    2018

  5. [5]

    Z.-H. Han, S. Görtz, Hierarchical kriging model for variable-fidelity surrogate modeling, AIAA journal 50 (2012) 1885–1896

    2012

  6. [6]

    Nagawkar, L

    J. Nagawkar, L. T. Leifsson, X. Du, Applications of polynomial chaos-based cokriging to aerodynamic design optimization benchmark problems, in: AIAA Scitech 2020 Forum, 2020, p. 0542

    2020

  7. [7]

    Laurenceau, P

    J. Laurenceau, P. Sagaut, Building efficient response surfaces of aerodynamic functions with kriging and cokriging, AIAA journal (2008)

    2008

  8. [8]

    Z.-H. Han, Y . Zhang, C.-X. Song, K.-S. Zhang, Weighted gradient-enhanced kriging for high- dimensional surrogate modeling and design optimiza- tion, Aiaa Journal (2017)

    2017

  9. [9]

    M. A. Bouhlel, J. R. Martins, Gradient-enhanced kriging for high-dimensional problems, Engineering with Computers (2019)

    2019

  10. [10]

    M. A. Bouhlel, N. Bartoli, A. Otsmane, J. Morlier, Improving kriging surrogates of high-dimensional de- sign models by partial least squares dimension reduc- tion, Structural and Multidisciplinary Optimization (2016)

    2016

  11. [11]

    Wu, X.-Y

    M.-Y . Wu, X.-Y . Yuan, Z.-H. Chen, W.-T. Wu, Y . Hua, N. Aubry, Airfoil shape optimization using genetic algorithm coupled deep neural networks, Physics of Fluids 35 (2023)

    2023

  12. [12]

    Esfahanian, M

    V . Esfahanian, M. J. Izadi, H. Bashi, M. Ansari, A. Tavakoli, M. Kordi, Aerodynamic shape op- timization of gas turbines: a deep learning sur- rogate model approach, Structural and Multidis- ciplinary Optimization 67 (2023). doi: 10.1007/ s00158-023-03703-9

    2023

  13. [13]

    Raissi, P

    M. Raissi, P. Perdikaris, G. E. Karniadakis, Physics- informed neural networks: A deep learning frame- work for solving forward and inverse problems in- volving nonlinear partial differential equations, Jour- nal of Computational physics (2019)

    2019

  14. [14]

    Shukla, V

    K. Shukla, V . Oommen, A. Peyvan, M. Penwarden, L. Bravo, A. Ghoshal, R. M. Kirby, G. E. Karni- adakis, Deep neural operators can serve as accurate surrogates for shape optimization: a case study for airfoils, arXiv (2023). 18 Optimization-Embedded Active Multi-Fidelity Surrogate Learning for Multi-Condition Airfoil Shape Optimization

    2023

  15. [15]

    L. Lu, P. Jin, G. Pang, Z. Zhang, G. E. Karniadakis, Learning nonlinear operators via deeponet based on the universal approximation theorem of operators, Nature machine intelligence (2021)

    2021

  16. [16]

    Pereira, F

    D. Pereira, F. Afonso, F. Lau, End-to-end deep- learning-based surrogate modeling for supersonic airfoil shape optimization, Aerospace 12 (2025). doi:10.3390/aerospace12050389

    work page doi:10.3390/aerospace12050389 2025

  17. [17]

    Y . Y . Liu, J. X. Shen, P. P. Yang, X. W. Yang, A cnn-pinn-drl driven method for shape optimization of airfoils, Engineering Applications of Computational Fluid Mechanics (2025)

    2025

  18. [18]

    Rajnarayan, A

    D. Rajnarayan, A. Ning, J. A. Mehr, Universal airfoil parametrization using b-splines, in: 2018 Applied Aerodynamics Conference, 2018, p. 3949

    2018

  19. [19]

    Della Vecchia, E

    P. Della Vecchia, E. Daniele, E. D’Amato, An airfoil shape optimization technique coupling parsec param- eterization and evolutionary algorithm, Aerospace Science and Technology (2014)

    2014

  20. [20]

    W. He, X. Liu, Improved aerofoil parameterisation based on class/shape function transformation, The Aeronautical Journal (2019)

    2019

  21. [21]

    Kulfan, A Universal Parametric Geometry Rep- resentation Method - "CST", Aerospace Sciences Meetings, American Institute of Aeronautics and As- tronautics, 2007

    B. Kulfan, A Universal Parametric Geometry Rep- resentation Method - "CST", Aerospace Sciences Meetings, American Institute of Aeronautics and As- tronautics, 2007. doi:10.2514/6.2007-62, 0

    work page doi:10.2514/6.2007-62 2007

  22. [22]

    N. Vu, J. Lee, Aerodynamic design optimization of helicopter rotor blades including airfoil shape for forward flight, Aerospace Science and Technology 42 (2015) 106–117

    2015

  23. [23]

    Zhang, X

    Y . Zhang, X. Fang, H. Chen, S. Fu, Z. Duan, Y . Zhang, Supercritical natural laminar flow air- foil optimization for regional aircraft wing design, Aerospace Science and Technology (2015)

    2015

  24. [24]

    D. J. Toal, N. W. Bressloff, A. J. Keane, C. M. Holden, Geometric filtration using proper orthogonal decomposition for aerodynamic design optimization, AIAA journal (2010)

    2010

  25. [25]

    X. Wu, W. Zhang, X. Peng, Z. Wang, Benchmark aerodynamic shape optimization with the pod-based cst airfoil parametric method, Aerospace Science and Technology (2019)

    2019

  26. [26]

    J. Li, M. A. Bouhlel, J. R. Martins, Data-based approach for fast airfoil analysis and optimization, AIAA Journal (2019)

    2019

  27. [27]

    Yonekura, K

    K. Yonekura, K. Suzuki, Data-driven design explo- ration method using conditional variational autoen- coder for airfoil design, Structural and Multidisci- plinary Optimization (2021)

    2021

  28. [28]

    Achour, W

    G. Achour, W. J. Sung, O. J. Pinon-Fischer, D. N. Mavris, Development of a Conditional Generative Adversarial Network for Airfoil Shape Optimiza- tion, AIAA SciTech Forum, American Institute of Aeronautics and Astronautics, 2020. doi:10.2514/ 6.2020-2261, 0

    2020

  29. [29]

    W. Chen, K. Chiu, M. D. Fuge, Airfoil design param- eterization and optimization using bézier generative adversarial networks, AIAA journal (2020)

    2020

  30. [30]

    Sekar, Q

    V . Sekar, Q. Jiang, C. Shu, B. C. Khoo, Fast flow field prediction over airfoils using deep learning approach, Physics of Fluids (2019)

    2019

  31. [31]

    Giselle Fernández-Godino, Review of multi- fidelity models, Advances in Computational Science and Engineering 1 (2023) 351–400

    M. Giselle Fernández-Godino, Review of multi- fidelity models, Advances in Computational Science and Engineering 1 (2023) 351–400. doi: 10.3934/ acse.2023015

    2023

  32. [32]

    Schouler, A

    M. Schouler, A. Belme, P. Cinnella, Comparison of multi-fidelity surrogate models for multi-objective aerodynamic optimization in turbomachinery under extreme cost imbalance, Advanced Modeling and Simulation in Engineering Sciences 12 (2025) 35. doi:10.1186/s40323-025-00316-3

    work page doi:10.1186/s40323-025-00316-3 2025

  33. [33]

    N. M. Alexandrov, R. M. Lewis, C. R. Gumbert, L. L. Green, P. A. Newman, Approximation and model management in aerodynamic optimization with variable-fidelity models, Journal of Aircraft 38 (2001) 1093–1101. doi:10.2514/2.2877

    work page doi:10.2514/2.2877 2001

  34. [34]

    V . O. Balabanov, A. A. Giunta, O. Golovidov, B. Grossman, W. H. Mason, L. T. Watson, R. T. Haftka, Reasonable design space approach to re- sponse surface approximation, Journal of Aircraft 36 (1999) 308–315. doi:10.2514/2.2438

    work page doi:10.2514/2.2438 1999

  35. [35]

    Zheng, X

    J. Zheng, X. Shao, L. Gao, P. Jiang, Z. Li, A hybrid variable-fidelity global approximation mod- elling method combining tuned radial basis function base and kriging correction, Journal of Engineering Design 24 (2013). doi: 10.1080/09544828.2013. 788135

    work page doi:10.1080/09544828.2013 2013

  36. [36]

    C. C. Fischer, R. V . Grandhi, P. S. Beran, Bayesian low-fidelity correction approach to multi-fidelity aerospace design, in: 58th AIAA/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Con- ference, 2017, p. 0133

    2017

  37. [37]

    Schouler, A

    M. Schouler, A. Belme, P. Cinnella, Bayesian and non-bayesian multi-fidelity surrogate models for multi-objective aerodynamic optimization under ex- treme cost imbalance, 2025. URL: https://arxiv. org/abs/2505.17279.arXiv:2505.17279

    Pith/arXiv arXiv 2025

  38. [38]

    P. Z. G. Qian, C. F. J. Wu, Bayesian hierarchical mod- eling for integrating low-accuracy and high-accuracy experiments, Technometrics 50 (2008) 192–204. doi:10.1198/004017008000000082

    work page doi:10.1198/004017008000000082 2008

  39. [39]

    A. I. Forrester, A. J. Keane, Recent advances in surrogate-based optimization, Progress in Aerospace Sciences 45 (2009) 50–79. doi:https://doi.org/ 10.1016/j.paerosci.2008.11.001

    doi:10.1016/j.paerosci.2008.11.001 2009

  40. [40]

    L. L. Gratiet, J. Garnier, Recursive co-kriging model for design of computer experiments with multiple lev- els of fidelity, International Journal for Uncertainty Quantification 4 (2014) 365–386. doi:10.1615/Int. J.UncertaintyQuantification.2014006914. 19 Optimization-Embedded Active Multi-Fidelity Surrogate Learning for Multi-Condition Airfoil Shape Optimization

    work page doi:10.1615/int 2014

  41. [41]

    Brevault, M

    L. Brevault, M. Balesdent, A. Hebbal, Overview of gaussian process based multi-fidelity techniques with variable relationship between fidelities, application to aerospace systems, Aerospace Science and Tech- nology 107 (2020) 106339. doi:https://doi.org/ 10.1016/j.ast.2020.106339

    doi:10.1016/j.ast.2020.106339 2020

  42. [42]

    X. Meng, G. E. Karniadakis, A composite neural network that learns from multi-fidelity data: Appli- cation to function approximation and inverse pde problems, Journal of Computational Physics 401 (2020) 109020. doi: https://doi.org/10.1016/ j.jcp.2019.109020

    arXiv 2020

  43. [43]

    X. Meng, H. Babaee, G. E. Karniadakis, Multi-fidelity bayesian neural networks: Al- gorithms and applications, Journal of Com- putational Physics 438 (2021) 110361. URL: https://www.sciencedirect.com/science/ article/pii/S0021999121002564. doi: https: //doi.org/10.1016/j.jcp.2021.110361

    work page doi:10.1016/j.jcp.2021.110361 2021

  44. [44]

    Cutajar, M

    K. Cutajar, M. Pullin, A. Damianou, N. Lawrence, J. González, Deep gaussian processes for multi- fidelity modeling, 2019. URL: https://arxiv. org/abs/1903.07320.arXiv:1903.07320

    Pith/arXiv arXiv 2019

  45. [45]

    A. A. Howard, M. Perego, G. E. Karniadakis, P. Stinis, Multifidelity deep operator networks for data-driven and physics-informed problems, Journal of Computational Physics 493 (2023) 112462. doi: https://doi.org/10.1016/j.jcp. 2023.112462

    doi:10.1016/j.jcp 2023

  46. [46]

    Bhola, S

    S. Bhola, S. Pawar, P. Balaprakash, R. Maulik, Multi- fidelity reinforcement learning framework for shape optimization, Journal of Computational Physics 482 (2023) 112018. doi: https://doi.org/10.1016/ j.jcp.2023.112018

    arXiv 2023

  47. [47]

    K. Li, F. Li, Multi-fidelity methods for optimization: A survey, 2024. URL: https://arxiv.org/abs/ 2402.09638.arXiv:2402.09638

    Pith/arXiv arXiv 2024

  48. [49]

    MOHAMMAD ZADEH, M

    P. MOHAMMAD ZADEH, M. SAY ADI, An efficient aerodynamic shape optimization of blended wing body uav using multi-fidelity models, Chinese Jour- nal of Aeronautics 31 (2018) 1165–1180. doi:https: //doi.org/10.1016/j.cja.2018.04.004

    work page doi:10.1016/j.cja.2018.04.004 2018

  49. [50]

    C. M. Aye, K. Wansaseub, S. Kumar, G. G. Tejani, S. Bureerat, A. R. Yildiz, N. Pholdee, Airfoil shape optimisation using a multi-fidelity surrogate-assisted metaheuristic with a new multi- objective infill sampling technique, CMES - Computer Modeling in Engineering and Sciences 137 (2023) 2111–2128. doi:https://doi.org/10. 32604/cmes.2023.028632

    arXiv 2023

  50. [51]

    Zhang, F

    X. Zhang, F. Xie, T. Ji, Z. Zhu, Y . Zheng, Multi- fidelity deep neural network surrogate model for aerodynamic shape optimization, Computer Meth- ods in Applied Mechanics and Engineering 373 (2021) 113485. doi: https://doi.org/10.1016/ j.cma.2020.113485

    arXiv 2021

  51. [52]

    Charayron, T

    R. Charayron, T. Lefebvre, N. Bartoli, J. Morlier, Towards a multi-fidelity & multi-objective bayesian optimization efficient algorithm, Aerospace Science and Technology 142 (2023) 108673. doi:https:// doi.org/10.1016/j.ast.2023.108673

    work page doi:10.1016/j.ast.2023.108673 2023

  52. [53]

    X. Wu, Z. Zuo, L. Ma, W. Zhang, Multi- fidelity neural network-based aerodynamic optimiza- tion framework for propeller design in electric air- craft, Aerospace Science and Technology 146 (2024) 108963. doi: https://doi.org/10.1016/ j.ast.2024.108963

    arXiv 2024

  53. [54]

    Mourousias, B

    N. Mourousias, B. G. Marinus, M. C. Runacres, A novel multi-fidelity optimization framework for high- altitude propellers, Aerospace Science and Technol- ogy 153 (2024) 109407. doi:https://doi.org/10. 1016/j.ast.2024.109407

    arXiv 2024

  54. [55]

    Robledo, Y

    I. Robledo, Y . Li, G. Y . Cornejo Maceda, R. Castel- lanos, Fast and robust parametric and functional learning with hybrid genetic optimisation (hygo), Ad- vances in Engineering Software 216 (2026) 104139. doi:10.1016/j.advengsoft.2026.104139

    work page doi:10.1016/j.advengsoft.2026.104139 2026

  55. [56]

    Drela, Xfoil: An analysis and design system for low reynolds number airfoils, in: T

    M. Drela, Xfoil: An analysis and design system for low reynolds number airfoils, in: T. J. Mueller (Ed.), Low Reynolds Number Aerodynamics, Springer Berlin Heidelberg, Berlin, Heidelberg, 1989, pp. 1– 12

    1989

  56. [57]

    Morgado, R

    J. Morgado, R. Vizinho, M. Silvestre, J. Páscoa, Xfoil vs cfd performance predictions for high lift low reynolds number airfoils, Aerospace Science and Technology (2016). doi:https://doi.org/10. 1016/j.ast.2016.02.031

    2016

  57. [58]

    Zhang, H

    S. Zhang, H. Li, A. A. Abbasi, Design method- ology using characteristic parameters control for low reynolds number airfoils, Aerospace Science and Technology (2019). doi:https://doi.org/10. 1016/j.ast.2019.01.003

    2019

  58. [59]

    Carreño Ruiz, D

    M. Carreño Ruiz, D. D’Ambrosio, Experimental validation of virtual wind tunnel testing for ultra- low reynolds numbers flows, Aerotecnica Missili & Spazio (2024)

    2024

  59. [60]

    Jasak, Openfoam: Open source cfd in research and industry, International Journal of Naval Architec- ture and Ocean Engineering 1 (2009) 89–94

    H. Jasak, Openfoam: Open source cfd in research and industry, International Journal of Naval Architec- ture and Ocean Engineering 1 (2009) 89–94. URL: https://www.sciencedirect.com/science/ article/pii/S2092678216303879. doi: https: //doi.org/10.2478/IJNAOE-2013-0011

    doi:10.2478/ijnaoe-2013-0011 2009

  60. [61]

    F. R. Menter, Two-equation eddy-viscosity turbu- lence models for engineering applications, AIAA journal (1994). 20 Optimization-Embedded Active Multi-Fidelity Surrogate Learning for Multi-Condition Airfoil Shape Optimization

    1994

  61. [62]

    D. C. Jespersen, T. H. Pulliam, M. L. Childs, Over- flow turbulence modeling resource validation results, Technical Report ARC-E-DAA-TN35216, NASA Ames Research Center, 2016. URL: https://ntrs. nasa.gov/citations/20190000252

    arXiv 2016

  62. [63]

    F. R. Menter, Improved Two-Equation k-omega Tur- bulence Models for Aerodynamic Flows, Techni- cal Report NASA-TM-103975, NASA Ames Re- search Center, 1992. URL: https://ntrs.nasa. gov/citations/19930013620, NASA Technical Memorandum

    arXiv 1992

  63. [64]

    J. A. Nelder, R. Mead, A simplex method for function minimization, The Computer Journal 7 (1965) 308– 313

    1965

  64. [65]

    C. A. Coello Coello, Theoretical and numerical constraint-handling techniques used with evolution- ary algorithms: a survey of the state of the art, Com- puter Methods in Applied Mechanics and Engineer- ing 191 (2002) 1245–1287. doi:https://doi.org/ 10.1016/S0045-7825(01)00323-1. 21

    work page doi:10.1016/s0045-7825(01)00323-1 2002