pith. sign in

arxiv: 2510.22517 · v2 · submitted 2025-10-26 · 💻 cs.CE · cs.LG· cs.SY· eess.SY

Data-driven Sensor Placement for Predictive Applications: A Correlation-Assisted Attribution Framework (CAAF)

Sze Chai Leung , Di Zhou , H. Jane Bae This is my paper

Pith reviewed 2026-05-18 04:51 UTC · model grok-4.3

classification 💻 cs.CE cs.LGcs.SYeess.SY
keywords optimal sensor placementfeature attributionclusteringpredictive modelingdynamical systemsstructural health monitoringturbulent flowmachine learning
0
0 comments X

The pith

Clustering candidate sensor locations before feature attribution identifies optimal placements despite input correlations.

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

The paper develops a machine-learning approach for choosing the best spots to install sensors so that predictions about a physical system's behavior remain accurate with fewer devices. Standard feature attribution methods run into trouble when sensor readings are highly correlated, a common situation in real monitoring setups. The proposed Correlation-Assisted Attribution Framework first groups similar candidate locations through clustering, then applies attribution to cut redundancy and improve how well the selections transfer to new conditions. Tests on structural health monitoring, airfoil lift, and turbulent channel flow show clearer advantages than methods that skip this step when nonlinear dynamics and multi-scale effects appear. A reader would care because the result points to cheaper, more practical sensor networks for control and inference tasks.

Core claim

The central claim is that introducing a clustering step on candidate sensor locations before feature attribution reduces redundancy from correlated inputs and thereby enables feature attribution to identify optimal sensor placements effectively for target predictions in dynamical systems that contain nonlinear dynamics, chaotic behavior, and multi-scale interactions.

What carries the argument

The Correlation-Assisted Attribution Framework (CAAF), which clusters candidate sensor locations before performing feature attribution to reduce redundancy and enhance generalizability.

If this is right

  • The framework supports effective use of feature attribution for optimal sensor placement in real-world environments that feature nonlinear dynamics and multi-scale interactions.
  • It outperforms alternative approaches that struggle with chaotic behavior and correlated measurements in applications such as structural health monitoring and fluid-flow estimation.
  • It improves the generalizability of sensor selections when measurement inputs exhibit high correlations.
  • It enables more efficient monitoring, control, and inference with reduced numbers of sensors in complex physical systems.

Where Pith is reading between the lines

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

  • Pre-clustering may help other attribution-based selection methods when working with redundant high-dimensional inputs from physical sensors.
  • The same idea could be tested for adaptive placement where sensor importance changes over time or under varying operating conditions.
  • Combining the clustering step with cost or power constraints on individual sensors would be a natural next check on practicality.

Load-bearing premise

That performing a clustering step on candidate sensor locations before feature attribution will reliably reduce redundancy and enhance generalizability when inputs are highly correlated.

What would settle it

A controlled test on a system with known strong correlations and nonlinear dynamics in which the clustered version produces sensor sets whose predictions are no better than those from direct feature attribution without clustering would falsify the central claim.

Figures

Figures reproduced from arXiv: 2510.22517 by Di Zhou, H. Jane Bae, Sze Chai Leung.

Figure 1
Figure 1. Figure 1: Clustering and naive FA applied to image classification. Clustering and naive FA outcome for 2 banana images (a,e,i,b,f,j) and 2 starfish images (c,g,k,d,h,l). The top row (a–d) shows the original ImageNet samples. The middle row (e–h) displays the naive FA results, where cyan crosses × indicate the 50 pixels with the highest FA scores. The bottom row (i–l) presents the clustering results, with each cluste… view at source ↗
Figure 2
Figure 2. Figure 2: Variation of X3’s contribution fraction with correlation. As the correlation between X3 and the target variable Y increases, the fractions of the total FA scores contributed by X3 from the naive FA ( ) and clustered FA ( ) approaches are shown. that the authentic mode shapes and target frequencies can be accurately recovered31. Additionally, SHM strategies rely on OSP techniques for various applications, i… view at source ↗
Figure 3
Figure 3. Figure 3: Locations and performance of SHM sensors. (a) 15 SHM sensors on the cantilever beam identified by the CAAF, EI, and KE methods, with 19 node clusters (colored bars) and their centers (thick colored bars). Performance variation measured using (b) RMS, (c) CN , and (d) DET versus the number of sensors (nsensor) for configurations determined by CAAF ( ), EI ( ), and KE ( ). RMS/CN values or low DET values ind… view at source ↗
Figure 4
Figure 4. Figure 4: Clustering of airfoil surface pressure sensor candidates. Clustering results showing 27 identified clusters (colored dots) and their centers (◦). Bayesian experimental design, and uniform sensor distributions methods. Following our previous work43, gusty inflow conditions are generated by placing a cylinder directly upstream of the airfoil. The Reynolds number Rec based on chord length C and free-stream ve… view at source ↗
Figure 5
Figure 5. Figure 5: Optimal airfoil surface pressure sensor locations. The optimal 10-sensor configurations identified by (a) CAAF, (b) naive FA, (c) POD-based QR pivoting, (d) Bayesian experimental design, and (e) uniform distribution. Red crosses × mark the locations of the sensors. Numerical labels indicate the sensors’ importance ranking, reflecting their selection order, except for the POD-based QR pivoting sensors, whic… view at source ↗
Figure 6
Figure 6. Figure 6: Prediction performance comparison between sensor configurations identified using different OSP methods. Lift prediction error for different number of desired sensors (nsensor) using sensors identified by CAAF ( ), naive FA ( ), POD with QR pivoting ( ), Bayesian experimental design ( ), and uniform distribution ( ). The error bars represent the standard deviation computed over a minimum of ten independent … view at source ↗
Figure 7
Figure 7. Figure 7: Instantaneous data visualization from DNS of channel flow. a Wall-normal velocity fluctuation v ′+ at y + = 10. b Wall pressure fluctuation p ′+ w . where (·) ′ denotes fluctuations, h·i is the time-averaging operator, and σ represents the standard deviation. As shown in [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Clustering and optimal sensor configuration for wall-normal velocity estimation. a Initial candidate sensor locations (361 probes) and AP clustering results (reduced to 58 representative locations). Cluster members are shown as colored dots with cluster centers indicated by black open circles (◦). b Normalized cross-correlation coefficient between pw and v10 time series. The ten most optimal sensor locatio… view at source ↗
Figure 9
Figure 9. Figure 9: Wall-normal velocity (v + 10) time-series predictions. CAAF-identified sensors ( ) versus uniform benchmark sensor array ( ) plotted against reference data ( ). results agree with physical intuition, demonstrating their interpretability through domain-specific logic. Therefore, the choice of clustering and FA algorithms must also be carefully considered to ensure robust and interpretable results. Furthermo… view at source ↗
Figure 10
Figure 10. Figure 10: Schematics of the 5 simulated cases listed in [PITH_FULL_IMAGE:figures/full_fig_p014_10.png] view at source ↗
read the original abstract

Optimal sensor placement (OSP) is critical for efficient, accurate monitoring, control, and inference in complex physical systems. We propose a machine-learning-based feature attribution (FA) framework to identify OSP for target predictions. FA quantifies input contributions to a model output; however, it struggles with highly correlated input data often encountered in practical applications for OSP. To address this, we propose a Correlation-Assisted Attribution Framework (CAAF), which introduces a clustering step on the candidate sensor locations before performing FA to reduce redundancy and enhance generalizability. We first illustrate the core principles of the proposed framework through a series of validation cases, then demonstrate its effectiveness in realistic dynamical systems such as structural health monitoring, airfoil lift prediction, and wall-normal velocity estimation for turbulent channel flow. The results show that the CAAF outperforms alternative approaches that typically struggle due to the presence of nonlinear dynamics, chaotic behavior, and multi-scale interactions, and enables the effective application of FA for identifying OSP in real-world environments.

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 proposes the Correlation-Assisted Attribution Framework (CAAF) for data-driven optimal sensor placement (OSP). It augments standard feature attribution (FA) methods, which struggle with correlated inputs, by inserting a clustering step on candidate sensor locations prior to attribution. The approach is first validated on synthetic cases and then demonstrated on three applications: structural health monitoring, airfoil lift prediction, and wall-normal velocity estimation in turbulent channel flow. The central claim is that CAAF enables effective FA-based OSP in systems with nonlinear dynamics, chaos, and multi-scale interactions, outperforming alternatives that typically fail in such regimes.

Significance. If the clustering step can be shown to stabilize attribution scores and produce generalizable sensor selections independent of the downstream model, the framework would supply a practical, ML-based tool for OSP in engineering systems where input correlations and nonlinearity are pervasive. The choice of turbulent channel flow and airfoil test cases is appropriate for stressing multi-scale behavior, and successful results here would strengthen the case for broader adoption in predictive monitoring and control.

major comments (2)
  1. [§4 (Framework description) and §5.3 (turbulent channel flow results)] The manuscript introduces the clustering step to mitigate input correlations before FA, yet provides no ablation that isolates its contribution (e.g., attribution-score stability, selected-sensor overlap, or predictive error with vs. without clustering) in the turbulent channel flow or airfoil examples. Without this, performance gains cannot be confidently attributed to the proposed CAAF component rather than the choice of downstream regressor.
  2. [§5.2–5.3 (application results)] In the airfoil and channel-flow demonstrations, the paper reports improved OSP relative to baselines, but does not quantify how clustering affects redundancy reduction under the nonlinear, multi-scale regimes that are central to the claim. A direct metric such as pairwise correlation of selected sensors or variance of attribution ranks across cross-validation folds would be required to substantiate the generalizability argument.
minor comments (2)
  1. [Abstract] The abstract asserts outperformance across three domains but supplies no numerical metrics, error bars, or baseline names; adding one representative quantitative result would improve readability.
  2. [§3 (methods)] Notation for the correlation matrix and attribution scores should be defined once in a dedicated subsection and used consistently thereafter to avoid ambiguity when readers compare the clustering and FA stages.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment below and indicate where revisions will be made to strengthen the evidence for the clustering component.

read point-by-point responses

  1. Referee: [§4 (Framework description) and §5.3 (turbulent channel flow results)] The manuscript introduces the clustering step to mitigate input correlations before FA, yet provides no ablation that isolates its contribution (e.g., attribution-score stability, selected-sensor overlap, or predictive error with vs. without clustering) in the turbulent channel flow or airfoil examples. Without this, performance gains cannot be confidently attributed to the proposed CAAF component rather than the choice of downstream regressor.

    Authors: We agree that an explicit ablation isolating the clustering step would strengthen attribution of gains to CAAF. The validation cases in §4 illustrate the role of clustering in reducing redundancy, and the application results show overall improvements over baselines. To directly address the concern, we will add ablation studies for the airfoil and turbulent channel flow cases, reporting attribution-score stability, selected-sensor overlap, and predictive error with versus without the clustering step. revision: yes

  2. Referee: [§5.2–5.3 (application results)] In the airfoil and channel-flow demonstrations, the paper reports improved OSP relative to baselines, but does not quantify how clustering affects redundancy reduction under the nonlinear, multi-scale regimes that are central to the claim. A direct metric such as pairwise correlation of selected sensors or variance of attribution ranks across cross-validation folds would be required to substantiate the generalizability argument.

    Authors: We acknowledge that additional quantitative metrics would better substantiate redundancy reduction and generalizability under nonlinear, multi-scale conditions. The current results demonstrate improved OSP, but to strengthen this, we will incorporate direct metrics including pairwise correlations of selected sensors and variance of attribution ranks across cross-validation folds in the revised airfoil and channel-flow sections. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation chain remains self-contained

full rationale

The paper proposes CAAF by adding a pre-FA clustering step on candidate locations to mitigate input correlations before attribution. This is presented as a methodological addition with validation on structural health monitoring, airfoil, and turbulent flow cases. No load-bearing step reduces by construction to a fitted parameter renamed as prediction, a self-citation chain, or an ansatz smuggled via prior work. The central claim of improved OSP identification rests on empirical demonstrations rather than definitional equivalence or forced uniqueness. The clustering contribution is asserted to reduce redundancy, but the paper does not equate it tautologically to the final attribution scores; external validation cases supply independent content. This is the common honest non-finding for a framework paper whose improvements are shown end-to-end without internal re-derivation of inputs.

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 framework implicitly assumes clustering will mitigate correlation effects without detailing how clusters are formed or validated.

pith-pipeline@v0.9.0 · 5714 in / 978 out tokens · 23171 ms · 2026-05-18T04:51:32.284992+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

67 extracted references · 67 canonical work pages · 2 internal anchors

  1. [1]

    & Meng, X

    Y u, L., Wang, N. & Meng, X. Real-time forest fire detection wit h wireless sensor networks. In Proc. 2005 Int. Conf. Wirel. Commun. Netw. Mobile Comput., vol. 2, 1214–1217 (2005)

    work page 2005

  2. [2]

    A., Gomez V argas, R

    Tibaduiza Burgos, D. A., Gomez V argas, R. C., Pedraza, C., Ag is, D. & Pozo, F. Damage identification in structural health monitoring: A brief review from its implementation t o the use of data-driven applications. Sensors 20, 733 (2020)

    work page 2020

  3. [3]

    Andersson, T. R. et al. Environmental sensor placement with convolutional gaussi an neural processes. Environ. Data Sci. 2, e32 (2023)

    work page 2023

  4. [4]

    C., Fan, J., Davari, A

    Castello, C. C., Fan, J., Davari, A. & Chen, R.-X. Optimal sen sor placement strategy for environmental monitoring using wireless sensor networks. In 2010 42nd Southeastern Symp. Syst. Theory (SSST) , 275–279 (IEEE, 2010)

    work page 2010

  5. [5]

    J., V elasco-Hernandez, G., Barry, J

    Y eong, D. J., V elasco-Hernandez, G., Barry, J. & Walsh, J. Sensor and sensor fusion technology in autonomous vehicles: A review. Sensors 21, 2140 (2021)

    work page 2021

  6. [6]

    Kammer, D. C. Sensor placement for on-orbit modal identifica tion and correlation of large space structures. J. Guid. Control. Dyn. 14, 251–259 (1991)

    work page 1991

  7. [7]

    & Soares, C

    Liu, K., Y an, R.-J. & Soares, C. G. Optimal sensor placement a nd assessment for modal identification. Ocean. Eng. 165, 209–220 (2018)

    work page 2018

  8. [8]

    W ., Kutz, J

    Manohar, K., Brunton, B. W ., Kutz, J. N. & Brunton, S. L. Data- driven sparse sensor placement for reconstruction: Demonstrating the benefits of exploiting known patterns. IEEE Control. Syst. Mag. 38, 63–86 (2018)

    work page 2018

  9. [9]

    & Rasheed, A

    Tabib, M., Skare, K., Bruaset, E. & Rasheed, A. Data-driven s patio-temporal modelling and optimal sensor placement for a digital twin set-up. Eng. Proc. 39, 98 (2023). 16/19

    work page 2023

  10. [10]

    Singh, A. K. & Hahn, J. Determining optimal sensor locations for state and parameter estimation for stable nonlinear systems. Ind. & Eng. Chem. Res. 44, 5645–5659 (2005)

    work page 2005

  11. [11]

    & K oumoutsakos, P

    V erma, S., Papadimitriou, C., Lüthen, N., Arampatzis, G. & K oumoutsakos, P . Optimal sensor placement for artificial swimmers. J. Fluid Mech. 884, A24 (2020)

    work page 2020

  12. [12]

    & Marzouk, Y

    Huan, X. & Marzouk, Y . M. Simulation-based optimal bayesian experimental design for nonlinear systems. J. Comput. Phys. 232, 288–317 (2013)

    work page 2013

  13. [13]

    & Guestrin, C

    Krause, A., Singh, A. & Guestrin, C. Near-optimal sensor pla cements in gaussian processes: Theory, efficient algorithm s and empirical studies. J. Mach. Learn. Res. 9 (2008)

    work page 2008

  14. [14]

    & Arranz, G

    Lozano-Durán, A. & Arranz, G. Information-theoretic formu lation of dynamical systems: causality, modeling, and control. Phys. Rev. Res. 4, 023195 (2022)

    work page 2022

  15. [15]

    Otto, S. E. & Rowley, C. W . Inadequacy of linear methods for minimal sensor placement and feature selection in nonlinear systems: A new approach using secants. J. Nonlinear Sci. 32, 69 (2022)

    work page 2022

  16. [16]

    & Chen, C

    Wang, Z., Li, H.-X. & Chen, C. Reinforcement learning-based optimal sensor placement for spatiotemporal modeling. IEEE Trans. Cybern. 50, 2861–2871 (2019)

    work page 2019

  17. [17]

    & Taira, K

    Zhong, Y ., Fukami, K., An, B. & Taira, K. Sparse sensor recons truction of vortex-impinged airfoil wake with machine learning. Theor. Comput. Fluid Dyn. 1–19 (2023)

    work page 2023

  18. [18]

    & Wang, S

    Shi, S., Du, D., Mercan, O., Kalkan, E. & Wang, S. A novel data- driven sensor placement optimization method for un- supervised damage detection using noise-assisted neural n etworks with attention mechanism. Mech. Syst. Signal Process. 209, 111075 (2024)

    work page 2024

  19. [19]

    Shapley, L. S. A value for n-person games. Annals Math. Stud. 307–317 (1953)

    work page 1953

  20. [20]

    C., Lundberg, S

    Chen, H., Covert, I. C., Lundberg, S. M. & Lee, S.-I. Algorith ms to estimate shapley value feature attributions. Nat. Mach. Intell. 5, 590–601 (2023)

    work page 2023

  21. [21]

    Towards better understanding of gradient-based attribution methods for Deep Neural Networks

    Ancona, M., Ceolini, E., Öztireli, C. & Gross, M. Towards bet ter understanding of gradient-based attribution methods f or deep neural networks. arXiv preprint arXiv:1711.06104 (2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  22. [22]

    Lundberg, S. M. & Lee, S.-I. A unified approach to interpretin g model predictions. Adv. Neural Inf. Process. Syst. 30 (2017)

    work page 2017

  23. [23]

    & Løland, A

    Aas, K., Jullum, M. & Løland, A. Explaining individual predi ctions when features are dependent: More accurate approx- imations to shapley values. Artif. Intell. 298, 103502 (2021)

    work page 2021

  24. [24]

    & Y an, Q

    Sundararajan, M., Taly, A. & Y an, Q. Axiomatic attribution f or deep networks. In Proc. Mach. Learn. Res. , 3319–3328 (PMLR, 2017)

    work page 2017

  25. [25]

    & Heese, K

    Sikdar, S., Bhattacharya, P . & Heese, K. Integrated directi onal gradients: Feature interaction attribution for neura l nlp models. In Proc. 59th Annu. Meet. Assoc. Comput. Linguist. 11th Int. Jo int Conf. Nat. Lang. Process. (V ol. 1: Long Pap.), 865–878 (2021)

    work page 2021

  26. [26]

    & Kotsiantis, S

    Linardatos, P ., Papastefanopoulos, V . & Kotsiantis, S. Exp lainable ai: A review of machine learning interpretability methods. Entropy 23, 18 (2020)

    work page 2020

  27. [27]

    & Fredrikson, M

    Wang, Z., Mardziel, P ., Datta, A. & Fredrikson, M. Interpreting interpretations: Organizing attribution methods by criteria. In Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. W orkshops, 10–11 (2020)

    work page 2020

  28. [28]

    Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps

    Simonyan, K., V edaldi, A. & Zisserman, A. Deep inside convolutional networks: Visualising image classification models and saliency maps. arXiv preprint arXiv:1312.6034 (2013)

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  29. [29]

    & Kundaje, A

    Shrikumar, A., Greenside, P . & Kundaje, A. Learning importa nt features through propagating activation differences. I n Proc. Int. Conf. Mach. Learn. , 3145–3153 (PMLR, 2017)

    work page 2017

  30. [30]

    Deng, J. et al. Imagenet: A large-scale hierarchical image database. In IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 248–255 (IEEE, 2009)

    work page 2009

  31. [31]

    & Garcia, E

    Papadopoulos, M. & Garcia, E. Sensor placement methodologi es for dynamic testing. AIAA J. 36, 256–263 (1998)

    work page 1998

  32. [32]

    Friswell, M. I. & Castro-Triguero, R. Clustering of sensor l ocations using the effective independence method. AIAA J. 53, 1388–1391 (2015)

    work page 2015

  33. [33]

    & Penny, J

    Garvey, S., Friswell, M. & Penny, J. Evaluation of a method fo r automatic selection of measurement locations based on subspace-matching. In Proc. 14th Int. Modal Anal. Conf. , vol. 2768, 1546 (1996). 17/19

    work page 1996

  34. [34]

    & Zumpano, G

    Meo, M. & Zumpano, G. On the optimal sensor placement techniq ues for a bridge structure. Eng. Struct. 27, 1488–1497 (2005)

    work page 2005

  35. [35]

    & Fritzen, C

    Li, D., Li, H. & Fritzen, C. The connection between effective independence and modal kinetic energy methods for sensor placement. J. Sound Vib. 305, 945–955 (2007)

    work page 2007

  36. [36]

    & Peinke, J

    Boettcher, F., Renner, C., Waldl, H.-P . & Peinke, J. On the st atistics of wind gusts. Boundary-Layer Meteorol. 108, 163–173 (2003)

    work page 2003

  37. [37]

    & Williams, D

    Granlund, K., Monnier, B., Ol, M. & Williams, D. Airfoil long itudinal gust response in separated vs. attached flows. Phys. Fluids 26 (2014)

    work page 2014

  38. [38]

    & Fusina, G

    Galway, D., Etele, J. & Fusina, G. Modeling of the urban gust e nvironment with application to autonomous flight. In Proc. AIAA Atmos. Flight Mech. Conf. Exhibit , 6565 (2008)

    work page 2008

  39. [39]

    Renn, P . I. & Gharib, M. Machine learning for flow-informed ae rodynamic control in turbulent wind conditions. Commun. Eng. 1, 45 (2022)

    work page 2022

  40. [40]

    & Eldredge, J

    Beckers, D. & Eldredge, J. D. Deep reinforcement learning of airfoil pitch control in a highly disturbed environment using partial observations. Phys. Rev. Fluids 9, 093902 (2024)

    work page 2024

  41. [41]

    Nair, N. J. & Goza, A. Bio-inspired variable-stiffness flaps for hybrid flow control, tuned via reinforcement learning. J. Fluid Mech. 956, R4 (2023)

    work page 2023

  42. [42]

    & Dandois, J

    Paris, R., Beneddine, S. & Dandois, J. Robust flow control and optimal sensor placement using deep reinforcement learning. J. Fluid Mech. 913, A25 (2021)

    work page 2021

  43. [43]

    C., Zhou, D

    Leung, S. C., Zhou, D. & Bae, H. J. Integrated gradients for op timal surface pressure sensor placement for lift predictio n of an airfoil subject to gust. In Proc. AIAA Aviat. F orum ASCEND 2024, 4148 (2024)

    work page 2024

  44. [44]

    & Kim, J

    Choi, H., Moin, P . & Kim, J. Active turbulence control for dra g reduction in wall-bounded flows. J. Fluid Mech. 262, 75–110 (1994)

    work page 1994

  45. [45]

    & Choi, H

    Park, J. & Choi, H. Machine-learning-based feedback contro l for drag reduction in a turbulent channel flow. J. Fluid Mech. 904, A24 (2020)

    work page 2020

  46. [46]

    Guastoni, L. et al. Convolutional-network models to predict wall-bounded turbulence from wall quantities. J. Fluid Mech. 928, A27 (2021)

    work page 2021

  47. [47]

    & Ianiro, A

    Güemes, A., Discetti, S. & Ianiro, A. Sensing the turbulent l arge-scale motions with their wall signature. Phys. Fluids 31 (2019)

    work page 2019

  48. [48]

    & Koumoutsakos, P

    Milano, M. & Koumoutsakos, P . Neural network modeling for ne ar wall turbulent flow. J. Comput. Phys. 182, 1–26 (2002)

    work page 2002

  49. [49]

    & Lee, J

    Y un, J. & Lee, J. Prediction of near wall velocity in turbulen t channel flow using wall pressure based on artificial neural network. In Proc. Korean Soc. Mech. Eng. Conf., 457–460 (KSME, 2017)

    work page 2017

  50. [50]

    & Moin, P

    Choi, H. & Moin, P . On the space-time characteristics of wall -pressure fluctuations. Phys. Fluids 2, 1450–1460 (1990)

    work page 1990

  51. [51]

    & Wang, M

    Liu, Y ., Zhou, D. & Wang, M. Wall-pressure fluctuations induc ed by a forward step in a high-reynolds-number turbulent boundary layer. In Proc. AIAA/CEAS Aeroacoust. Conf. 2024 , 3392 (2024)

    work page 2024

  52. [52]

    Frey, B. J. & Dueck, D. Clustering by passing messages betwee n data points. Science 315, 972–976 (2007)

    work page 2007

  53. [53]

    Tan, M. & Le, Q. Efficientnetv2: Smaller models and faster tra ining. In International Conference on Machine Learning , 10096–10106 (PMLR, 2021)

    work page 2021

  54. [54]

    & Moin, P

    Y ou, D., Ham, F. & Moin, P . Discrete conservation principles in large-eddy simulation with application to separation control over an airfoil. Phys. Fluids 20, 101515 (2008)

    work page 2008

  55. [55]

    Ruge, J. W . & Stüben, K. Algebraic multigrid (amg). In McCorm ick, S. F. (ed.) Multigrid Methods, vol. 3 of Frontiers Appl. Math., 73–130 (SIAM, Philadelphia, P A, 1987)

    work page 1987

  56. [56]

    & Cabot, W

    Germano, M., Piomelli, U., Moin, P . & Cabot, W . H. A dynamic su bgrid-scale eddy viscosity model. Phys. Fluids 3, 1760–1765 (1991)

    work page 1991

  57. [57]

    Lilly, D. K. A proposed modification of the germano subgrid-s cale closure method. Phys. Fluids 4, 633–635 (1992)

    work page 1992

  58. [58]

    R., Jarman, L

    Lefebvre, J., Jones, A. R., Jarman, L. & Smith, M. J. Experime ntal and numerical investigation of airfoil performance in cylinder wake. In Proc. 2018 Fluid Dyn. Conf. , 3232 (2018). 18/19

    work page 2018

  59. [59]

    & Gursul, I

    Zhang, Z., Wang, Z. & Gursul, I. Aerodynamics of a wing in turb ulent bluff body wakes. J. Fluid Mech. 937, A37 (2022)

    work page 2022

  60. [60]

    & Liu, H.-Y

    Jiang, Y ., Mao, M.-L., Deng, X.-G. & Liu, H.-Y . Numerical investigation on body-wake flow interaction over rod–airfoil configuration. J. Fluid Mech. 779, 1–35 (2015)

    work page 2015

  61. [61]

    & Samtaney, R

    Gao, W ., Zhang, W ., Cheng, W . & Samtaney, R. Wall-modelled large-eddy simulation of turbulent flow past airfoils. J. Fluid Mech. 873, 174–210 (2019)

    work page 2019

  62. [62]

    & Grassberger, P

    Kraskov, A., Stögbauer, H. & Grassberger, P . Estimating mutual information. Phys. Rev. E 69, 066138 (2004)

    work page 2004

  63. [63]

    55 (Springer, 2000)

    Orlandi, P .Fluid flow phenomena: a numerical toolkit , vol. 55 (Springer, 2000)

    work page 2000

  64. [64]

    Wray, A. A. Minimal storage time advancement schemes for spe ctral methods. NASA T ech. Rep.MS 202 (1990)

    work page 1990

  65. [65]

    Chorin, A. J. Numerical solution of the navier-stokes equat ions. Math. Comput. 22, 745–762 (1968)

    work page 1968

  66. [66]

    J., Lozano-Duran, A., Bose, S

    Bae, H. J., Lozano-Duran, A., Bose, S. & Moin, P . Turbulence intensities in large-eddy simulation of wall-bounded flows. Phys. Rev. Fluids 3, 014610 (2018)

    work page 2018

  67. [67]

    J., Lozano-Durán, A., Bose, S

    Bae, H. J., Lozano-Durán, A., Bose, S. T. & Moin, P . Dynamic sl ip wall model for large-eddy simulation. J. Fluid Mech. 859, 400–432 (2019). Acknowledgments We gratefully acknowledge the funding provided for this res earch by the SciAI Center, supported by the Office of Naval Research (ONR) under grant number N00014-23-1-2729, the Ca rver Mead New Adventu...

    work page 2019