pith. machine review for the scientific record. sign in

arxiv: 2512.13746 · v5 · submitted 2025-12-15 · 💻 cs.CE · cond-mat.mtrl-sci· cs.LG

Recognition: no theorem link

Probabilistic Predictions of Process-Induced Deformation in Carbon/Epoxy Composites Using a Deep Operator Network

Elham Kiyani , Amit Makarand Deshpande , Madhura Limaye , Zhiwei Gao , Zongren Zou , Sai Aditya Pradeep , Srikanth Pilla , Gang Li
show 2 more authors
Zhen Li George Em Karniadakis
Authors on Pith no claims yet

Pith reviewed 2026-05-16 22:39 UTC · model grok-4.3

classification 💻 cs.CE cond-mat.mtrl-scics.LG
keywords process-induced deformationDeepONetcarbon/epoxy compositestransfer learningcure cycle optimizationuncertainty quantificationcomposite manufacturingFiLM network
0
0 comments X

The pith

A FiLM DeepONet pretrained on simulations and fine-tuned via transfer learning on final deformation measurements predicts time histories of cure state, viscosity, and process-induced deformation in carbon-epoxy composites while quantifying,

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

The paper establishes that a physics-informed surrogate model can forecast process-induced deformation across many cure cycles by first generating training data from a two-mechanism simulation that tracks thermal expansion and cure shrinkage. A Feature-wise Linear Modulation DeepONet is trained on these simulated time histories, after which only its final layer is updated with sparse experimental measurements of final shape. This hybrid setup yields predictions of degree of cure, viscosity, and deformation at any time during a new non-isothermal cycle. Ensemble Kalman Inversion is then applied to the same framework to produce uncertainty bounds that reflect experimental variability. A sympathetic reader would care because such predictions can guide the design of cure schedules that reduce residual stresses and warping without requiring exhaustive physical trials for every new temperature profile.

Core claim

The central claim is that a FiLM DeepONet whose trunk and branch networks are trained on high-fidelity simulations of thermal and cure shrinkage can be adapted, by updating only the final layer with measured final deformation, to deliver accurate time-dependent predictions of degree of cure, viscosity, and deformation for arbitrary non-isothermal cure cycles; the same architecture combined with Ensemble Kalman Inversion further supplies probabilistic uncertainty estimates that support optimization of manufacturing schedules to minimize process-induced deformation.

What carries the argument

The FiLM DeepONet with transfer learning, in which external parameters modulate branch features and only the final layer is retrained on limited experimental final-deformation data.

If this is right

  • Time-history predictions allow virtual screening of cure cycles to identify schedules that keep final deformation below acceptable thresholds.
  • Uncertainty estimates from Ensemble Kalman Inversion enable robust optimization that accounts for variability in material response.
  • The transfer-learning step reduces the number of physical experiments needed to calibrate the model for new initial conditions or resin batches.
  • Probabilistic outputs support statistical process control and risk assessment during composite part manufacturing.

Where Pith is reading between the lines

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

  • The same architecture could be retrained on data from other fiber-matrix systems to test whether the two-mechanism model remains sufficient across different resin chemistries.
  • Coupling the surrogate with in-situ sensor readings could enable real-time adjustment of cure temperature to steer deformation toward a target value.
  • The approach suggests a general template for hybrid modeling in materials processing where full experimental time series are expensive but endpoint measurements are cheap.

Load-bearing premise

The two-mechanism physics model generates simulation data whose statistical distribution is close enough to real manufacturing conditions that transfer learning from those simulations to sparse final-deformation measurements will generalize to new cure cycles.

What would settle it

Direct comparison of the model's predicted final deformations and uncertainty intervals against measurements collected from a fresh set of cure cycles withheld from the transfer-learning step would falsify the claim if the observed errors systematically exceed the reported uncertainty bounds.

Figures

Figures reproduced from arXiv: 2512.13746 by Amit Makarand Deshpande, Elham Kiyani, Gang Li, George Em Karniadakis, Madhura Limaye, Sai Aditya Pradeep, Srikanth Pilla, Zhen Li, Zhiwei Gao, Zongren Zou.

Figure 1
Figure 1. Figure 1: Dynamic DSC run at heating rate of 20 °C/min indicating heat flow versus time for the AS4 Cf/3501-6 epoxy prepreg, with the area under the peaks indicating the heat of reaction resulting from the curing of the resin due to heat input during the DSC run 2.3. Manufacturing Trials The experimental validation runs were performed on a lab-scale setup for compression molding. The setup utilized a 10 kN load capa… view at source ↗
Figure 2
Figure 2. Figure 2: Experimental setup to consolidate and heat unbalanced [0/90] lay-up of prepregs to replicate the cure cycle to [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Specimen preparation methodology indicating layup of a 150 mm (6") x 150 mm (6") unbalance laminate [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Temperature–time profiles for the three cure cycles: an isothermal baseline cycle and two non-isothermal cycles [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Measurement of process induced deformation in unbalanced ply lay-up. The chord center point was identified [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Simulation data for a laminate subjected to a cure cycle. The top panel combines a parametrized cure [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Schematic of the DeepONet architecture for predicting DoC, viscosity, and deformation of a composite material [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Comparison of FiLM-DeepONet predictions with true simulation results at three temperature coordinates, [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Mean and standard deviation of DeepONet ensemble predictions for deformation. (Left) Deformation histo [PITH_FULL_IMAGE:figures/full_fig_p012_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Prediction uncertainty and ensemble behavior for DoC, viscosity, and deformation using EKI. (a) Uncertainty [PITH_FULL_IMAGE:figures/full_fig_p013_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Ensemble samples generated using EKI for DoC, viscosity, and deformation based on simulation data. Solid [PITH_FULL_IMAGE:figures/full_fig_p014_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Ensemble samples generated using EKI for the deformation in experimental optimal cases R11 and R21, [PITH_FULL_IMAGE:figures/full_fig_p015_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Result of the cure schedule optimization. The left panel shows the optimized temperature profile with the [PITH_FULL_IMAGE:figures/full_fig_p016_13.png] view at source ↗
read the original abstract

Fiber reinforcement and polymer matrix respond differently to manufacturing conditions due to mismatch in coefficient of thermal expansion and matrix shrinkage during curing of thermosets. These heterogeneities generate residual stresses over multiple length scales, whose partial release leads to process-induced deformation (PID), requiring accurate prediction and mitigation via optimized non-isothermal cure cycles. This study considers a unidirectional AS4 carbon fiber/amine bi-functional epoxy prepreg and models PID using a two-mechanism framework that accounts for thermal expansion/shrinkage and cure shrinkage. The model is validated against manufacturing trials to identify initial and boundary conditions, then used to generate PID responses for a diverse set of non-isothermal cure cycles (time-temperature profiles). Building on this physics-based foundation, we develop a data-driven surrogate based on Deep Operator Networks (DeepONets). A DeepONet is trained on a dataset combining high-fidelity simulations with targeted experimental measurements of PID. We extend this to a Feature-wise Linear Modulation (FiLM) DeepONet, where branch-network features are modulated by external parameters, including the initial degree of cure, enabling prediction of time histories of degree of cure, viscosity, and deformation. Because experimental data are available only at limited time instances (for example, final deformation), we use transfer learning: simulation-trained trunk and branch networks are fixed and only the final layer is updated using measured final deformation. Finally, we augment the framework with Ensemble Kalman Inversion (EKI) to quantify uncertainty under experimental conditions and to support optimization of cure schedules for reduced PID in composites.

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 manuscript develops a FiLM DeepONet surrogate trained on high-fidelity simulations from a two-mechanism physics model (thermal expansion/shrinkage plus cure shrinkage) of process-induced deformation in AS4 carbon/epoxy prepreg. Simulation-trained trunk and branch networks are fixed while only the final layer is updated via transfer learning on limited experimental final-deformation measurements; Ensemble Kalman Inversion then provides uncertainty quantification to support cure-schedule optimization for reduced PID.

Significance. If the transfer-learning step demonstrably preserves temporal fidelity, the approach would offer a practical route to fast, uncertainty-aware surrogates for composite manufacturing, combining physics-generated training data with sparse experimental calibration. The combination of operator networks, FiLM modulation by initial degree of cure, and EKI is a coherent methodological contribution to process modeling.

major comments (2)
  1. [Transfer learning and surrogate architecture] Transfer-learning procedure (abstract and § on surrogate construction): fixing the simulation-trained trunk and branch while retraining only the final layer on final-deformation scalars assumes that the learned features remain aligned with experimental dynamics. The two-mechanism model omits viscoelastic relaxation, resin flow, and fiber-bed compaction; any mismatch in these omitted mechanisms will propagate unchanged into the predicted degree-of-cure and viscosity trajectories, undermining the subsequent EKI uncertainty bands and cure optimization. Quantitative checks of intermediate time histories against any available experimental data or physics-based baselines are required to substantiate the claim.
  2. [Validation and results] Validation and results sections: the abstract states that the model is validated against manufacturing trials and that transfer learning is used, yet no error metrics (RMSE, MAE on time histories), data-split protocol, or comparison against simpler baselines (e.g., physics-only or standard DeepONet) are reported. Without these, the central assertion that the surrogate “accurately predicts time histories” cannot be assessed and remains the load-bearing gap identified in the stress-test.
minor comments (2)
  1. [Experimental data description] Clarify the precise experimental data points used for the final-layer update (number of specimens, cure cycles, measurement uncertainty) so readers can judge the information content available to the transfer step.
  2. [Figures] Figure captions should explicitly state whether plotted curves are simulation, surrogate, or post-transfer predictions to avoid ambiguity when comparing time histories.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive feedback. We address the major comments below and have updated the manuscript with additional metrics, comparisons, and discussion to strengthen the validation of the surrogate model.

read point-by-point responses

  1. Referee: [Transfer learning and surrogate architecture] Transfer-learning procedure (abstract and § on surrogate construction): fixing the simulation-trained trunk and branch while retraining only the final layer on final-deformation scalars assumes that the learned features remain aligned with experimental dynamics. The two-mechanism model omits viscoelastic relaxation, resin flow, and fiber-bed compaction; any mismatch in these omitted mechanisms will propagate unchanged into the predicted degree-of-cure and viscosity trajectories, undermining the subsequent EKI uncertainty bands and cure optimization. Quantitative checks of intermediate time histories against any available experimental data or physics-based baselines are required to substantiate the claim.

    Authors: We agree that the two-mechanism model has limitations by omitting viscoelastic relaxation, resin flow, and fiber-bed compaction, which is a common simplification in such process models. The high-fidelity simulations were validated against manufacturing trials for final deformation, and the transfer learning calibrates the output to match experimental final values. To substantiate the temporal fidelity, we have added quantitative comparisons of the predicted degree-of-cure and viscosity time histories against the original physics-based simulations (used as baselines) in the revised validation section. Since intermediate experimental time histories are not available, we cannot provide direct experimental checks for those; however, the EKI uncertainty quantification accounts for discrepancies by incorporating the experimental final deformation. We have also added a discussion on these model assumptions and their potential impact. revision: partial

  2. Referee: [Validation and results] Validation and results sections: the abstract states that the model is validated against manufacturing trials and that transfer learning is used, yet no error metrics (RMSE, MAE on time histories), data-split protocol, or comparison against simpler baselines (e.g., physics-only or standard DeepONet) are reported. Without these, the central assertion that the surrogate “accurately predicts time histories” cannot be assessed and remains the load-bearing gap identified in the stress-test.

    Authors: We have revised the results section to include RMSE and MAE metrics for the predicted time histories of degree of cure, viscosity, and deformation, both for the simulation test set and the transfer-learned experimental cases. The data-split protocol is now detailed: 80% of the simulation dataset for training the DeepONet, 20% for testing, with the experimental final deformation used solely for the transfer learning step on the final layer. Additionally, we have included comparisons to a physics-only model and a standard DeepONet without FiLM conditioning, demonstrating the improvements from our approach. These additions allow assessment of the surrogate's accuracy. revision: yes

standing simulated objections not resolved
  • Direct quantitative validation of intermediate time histories against experimental data, as only final deformation measurements are available from the manufacturing trials.

Circularity Check

0 steps flagged

Minor self-citation to DeepONet method; central predictions remain independent of fitted inputs

full rationale

The derivation trains a FiLM DeepONet on high-fidelity simulations from the two-mechanism physics model (thermal + cure shrinkage), fixes trunk/branch layers, and updates only the final layer via transfer learning on measured final-deformation scalars. Time-history outputs for degree of cure, viscosity, and deformation are generated by the fixed feature extractors rather than being algebraically identical to the scalar endpoint data. No equation reduces the predicted trajectories to the transfer-learning inputs by construction. Self-citation to the original DeepONet architecture (Karniadakis et al.) is present but not load-bearing; the paper supplies independent simulation data, experimental validation of the physics model, and EKI uncertainty quantification outside the fitted layer. This yields a low circularity score consistent with normal methodological reuse.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the accuracy of the two-mechanism physics model for generating training data and on the assumption that transfer learning from simulation to sparse experimental final values will produce reliable time histories for unseen cure cycles.

free parameters (1)
  • initial degree of cure
    Treated as an external conditioning parameter in the FiLM layer; its value is taken from experimental measurements but must be known for each new prediction.
axioms (1)
  • domain assumption The two-mechanism framework (thermal expansion/shrinkage plus cure shrinkage) sufficiently captures process-induced deformation for the AS4/amine epoxy system.
    Invoked to justify generating the simulation dataset used to train the DeepONet.

pith-pipeline@v0.9.0 · 5628 in / 1448 out tokens · 59749 ms · 2026-05-16T22:39:14.028819+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. Multi-Head Residual-Gated DeepONet for Coherent Nonlinear Wave Dynamics

    cs.LG 2026-04 unverdicted novelty 6.0

    MH-RG DeepONet adds parallel residual-gated conditioning pathways to DeepONet to achieve lower error and better phase coherence on nonlinear wave benchmarks by modulating state predictions with physical descriptors.

  2. Multi-fidelity surrogates for mechanics of composites: from co-kriging to multi-fidelity neural networks

    physics.comp-ph 2026-05 unverdicted novelty 2.0

    A review of multi-fidelity surrogates from co-kriging to neural networks for composite mechanics, with applications in prediction, optimization, and workflow integration.

Reference graph

Works this paper leans on

67 extracted references · 67 canonical work pages · cited by 2 Pith papers · 2 internal anchors

  1. [1]

    A. U. Sudhin, M. Remanan, G. Ajeesh, K. Jayanarayanan, Comparison of properties of carbon fiber reinforced thermoplastic and thermosetting composites for aerospace applications, Materials Today: Proceedings 24 (2020) 453–462.doi:10.1016/J.MATPR.2020.04.297

    work page doi:10.1016/j.matpr.2020.04.297 2020

  2. [2]

    R. E. Murray, R. Beach, D. Barnes, D. Snowberg, D. Berry, S. Rooney, M. Jenks, B. Gage, T. Boro, S. Wallen, et al., Structural validation of a thermoplastic composite wind turbine blade with comparison to a thermoset composite blade, Renewable Energy 164 (2021) 1100–1107

    work page 2021

  3. [3]

    A. M. Deshpande, R. R. Sadiwala, N. Brown, S. A. Pradeep, L. M. Headings, N. Zhao, B. Losey, R. Hahnlen, M. J. Dapino, G. Li, et al., Design and development of a multi-material, cost- competitive, lightweight mid-size sports utility vehicle’s body-in-white, CAMX 2022 (2022) 1–18

    work page 2022

  4. [4]

    A. M. Deshpande, R. Sadiwala, N. Brown, P.-Y. Lavertu, S. A. Pradeep, L. M. Headings, N. Zhao, B. Losey, R. Hahnlen, M. J. Dapino, G. Li, S. Pilla, Design optimization of a multi-material, fiber-reinforced composite-intensive body-in-white of a mid-size suv, in: CAMX 2023 Conference Proceedings, 2023, pp. 1–19.doi:10.33599/NASAMPE/C.23.0169

    work page doi:10.33599/nasampe/c.23.0169 2023

  5. [5]

    S.A.Pradeep, A.M.Deshpande, M.Limaye, R.K.Iyer, H.Kazan, G.Li, S.Pilla, Aperspectiveon the evolution of plastics and composites in the automotive industry, Applied Plastics Engineering Handbook (2024).doi:10.1016/B978-0-323-88667-3.00016-3

    work page doi:10.1016/b978-0-323-88667-3.00016-3 2024

  6. [6]

    S. R. White, H. T. Hahn, Cure cycle optimization for the reduction of processing-induced residual stresses in composite materials, http://dx.doi.org/10.1177/002199839302701402 27 (1993) 1352– 1378.doi:10.1177/002199839302701402

    work page doi:10.1177/002199839302701402 1993

  7. [7]

    C. Li, N. Zobeiry, K. Keil, S. Chatterjee, A. Poursartip, Advances in the characterization of residual stress in composite structures, International SAMPE Technical Conference (01 2014)

    work page 2014

  8. [8]

    L. G. Zhao, N. A. Warrior, A. C. Long, A micromechanical study of residual stress and its effect on transverse failure in polymer–matrix composites, International Journal of Solids and Structures 43 (2006) 5449–5467.doi:10.1016/J.IJSOLSTR.2005.08.012

    work page doi:10.1016/j.ijsolstr.2005.08.012 2006

  9. [9]

    S. L. Agius, M. Joosten, B. Trippit, C. H. Wang, T. Hilditch, Rapidly cured epoxy/anhydride composites: Effect of residual stress on laminate shear strength, Composites Part A: Applied Science and Manufacturing 90 (2016) 125–136.doi:10.1016/J.COMPOSITESA.2016.06.013

    work page doi:10.1016/j.compositesa.2016.06.013 2016

  10. [10]

    H. T. Hahn, Residual stresses in polymer matrix composite laminates, Journal of Composite Materials 10 (1976) 266–278.doi:10.1177/002199837601000401

    work page doi:10.1177/002199837601000401 1976

  11. [11]

    S. R. White, C. Kim, A simultaneous lay-up and in situ cure process for thick com- posites, http://dx.doi.org/10.1177/073168449301200503 12 (1993) 520–535.doi:10.1177/ 073168449301200503

    work page doi:10.1177/073168449301200503 1993

  12. [12]

    D. Li, X. Li, J. Dai, S. Xi, A comparison of curing process-induced residual stresses and cure shrinkage in micro-scale composite structures with different constitutive laws, Applied Composite Materials 25 (2018) 67–84.doi:10.1007/S10443-017-9608-6. 17

    work page doi:10.1007/s10443-017-9608-6 2018

  13. [13]

    A. Ding, S. Li, J. Wang, A. Ni, L. Sun, L. Chang, Prediction of process-induced distortions in l-shaped composite profiles using path-dependent constitutive law, Applied Composite Materials 23 (2016) 1027–1045.doi:10.1007/S10443-016-9501-8

    work page doi:10.1007/s10443-016-9501-8 2016

  14. [14]

    X. Liu, Z. Guan, X. Wang, T. Jiang, K. Geng, Z. Li, Study on cure-induced residual stresses and spring-in deformation of l-shaped composite laminates using a simplified constitutive model con- sidering stress relaxation, Composite Structures 272 (2021) 114203.doi:10.1016/J.COMPSTRUCT. 2021.114203

    work page doi:10.1016/j.compstruct 2021

  15. [15]

    Zobeiry, R

    N. Zobeiry, R. Vaziri, A. Poursartip, Computationally efficient pseudo-viscoelastic models for evaluation of residual stresses in thermoset polymer composites during cure, Composites Part A: Applied Science and Manufacturing 41 (2010) 247–256.doi:10.1016/J.COMPOSITESA.2009.10. 009

    work page doi:10.1016/j.compositesa.2009.10 2010

  16. [16]

    X. Li, X. Han, S. Duan, G. R. Liu, A two-stage genetic algorithm for molding parameters opti- mization for minimized residual stresses in composite laminates during curing, Applied Composite Materials 28 (2021) 1315–1334.doi:10.1007/S10443-021-09912-Z/TABLES/10

    work page doi:10.1007/s10443-021-09912-z/tables/10 2021

  17. [17]

    Q. Wang, X. Yang, X. Zhang, X. Chang, M. Yap, Z. Chen, M. Ren, Effect of cure cycles on residual stressesinthickcompositesusingmulti-physicscoupledanalysiswithmultipleconstitutivemodels, Materials Today Communications 32 (2022) 104094.doi:10.1016/J.MTCOMM.2022.104094

    work page doi:10.1016/j.mtcomm.2022.104094 2022

  18. [18]

    X. Hui, Y. Xu, W. Zhang, W. Zhang, Multiscale collaborative optimization for the thermochem- ical and thermomechanical cure process during composite manufacture, Composites Science and Technology 224 (2022) 109455.doi:10.1016/J.COMPSCITECH.2022.109455

    work page doi:10.1016/j.compscitech.2022.109455 2022

  19. [19]

    Behbahani, H

    R. Behbahani, H. Y. Sarvestani, E. Fatehi, E. Kiyani, B. Ashrafi, M. Karttunen, M. Rahmat, Machine learning-driven process of alumina ceramics laser machining, Physica Scripta 98 (1) (2022) 015834

    work page 2022

  20. [20]

    T. T. Ademujimi, M. P. Brundage, V. V. Prabhu, A review of current machine learning techniques used in manufacturing diagnosis, in: IFIP International Conference on Advances in Production Management Systems, Springer, 2017, pp. 407–415

    work page 2017

  21. [21]

    S. Y. Park, D. Kang, S. Y. On, S. S. Kim, Machine learning-driven optimization of the cure cycles of self-polymerizing epoxy molding compounds for semiconductor packaging applications, Composite Structures (2025) 119795

    work page 2025

  22. [22]

    Kiyani, H

    E. Kiyani, H. Y. Sarvestani, H. Ravanbakhsh, R. Behbahani, B. Ashrafi, M. Rahmat, M. Kart- tunen, Designing architectured ceramics for transient thermal applications using finite element and deep learning, Modelling and Simulation in Materials Science and Engineering 32 (1) (2023) 015001

    work page 2023

  23. [23]

    Y. Lin, Z. Guan, The use of machine learning for the prediction of the uniformity of the degree of cure of a composite in an autoclave, Aerospace 8 (5) (2021) 130

    work page 2021

  24. [24]

    S. Fan, J. Zhang, B. Wang, J. Chen, W. Yang, W. Liu, Y. Li, A deep learning method for fast predicting curing process-induced deformation of aeronautical composite structures, Composites Science and Technology 232 (2023) 109844

    work page 2023

  25. [25]

    Schoenholz, N

    C. Schoenholz, N. Zobeiry, An accelerated process optimization method to minimize deformations in composites using theory-guided probabilistic machine learning, Composites Part A: Applied Science and Manufacturing 176 (2024) 107842

    work page 2024

  26. [26]

    Lavaggi, M

    T. Lavaggi, M. Samizadeh, N. Niknafs Kermani, M. M. Khalili, S. G. Advani, Theory-guided ma- chine learning for optimal autoclave co-curing of sandwich composite structures, Polymer Com- posites 43 (8) (2022) 5319–5331.doi:https://doi.org/10.1002/pc.26829

    work page doi:10.1002/pc.26829 2022

  27. [27]

    Limaye, D

    M. Limaye, D. Morgan, A. M. Deshpande, S. A. Pradeep, S. Pilla, G. Li, Use of non-isothermal cure cycles for reducing cure induced deformation in laminated composites, Manufacturing Letters (4 2024).doi:10.1016/J.MFGLET.2024.03.016. 18

    work page doi:10.1016/j.mfglet.2024.03.016 2024

  28. [28]

    Q. Liu, X. Wang, Z. Guan, Z. Li, Rapid prediction and parameter evaluation of process-induced deformation in l-shape structures based on feature selection and artificial neural networks, Journal of Composites Science 8 (11) (2024) 455

    work page 2024

  29. [29]

    Zhang, H

    Y. Zhang, H. Li, L. Ge, L. Zheng, Z. Tang, F. Zhao, A deep learning model optimized by bayesian optimization with hyperband for fast prediction of the elastic properties of 3d tubular braided composites at different temperatures, Polymer Composites 46 (4) (2025) 3296–3307

    work page 2025

  30. [30]

    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) 218–229

    work page 2021

  31. [31]

    M. Lu, C. Lin, M. Maxey, G. E. Karniadakis, Z. Li, Bridging scales in multiscale bubble growth dynamics with correlated fluctuations using neural operator learning, International Journal of Multiphase Flow 180 (2024) 104959

    work page 2024

  32. [32]

    B. Shih, A. Peyvan, Z. Zhang, G. E. Karniadakis, Transformers as neural operators for solu- tions of differential equations with finite regularity, Computer Methods in Applied Mechanics and Engineering 434 (2025) 117560

    work page 2025

  33. [33]

    Kovachki, Z

    N. Kovachki, Z. Li, B. Liu, K. Azizzadenesheli, K. Bhattacharya, A. Stuart, A. Anandkumar, Neural operator: Learning maps between function spaces with applications to pdes, Journal of Machine Learning Research 24 (89) (2023) 1–97

    work page 2023

  34. [34]

    L. Liu, K. Nath, W. Cai, A causality-deeponet for causal responses of linear dynamical systems, Communications in Computational Physics (2024) 1194–1228

    work page 2024

  35. [35]

    R. Wan, E. Kharazmi, M. S. Triantafyllou, G. E. Karniadakis, Deepvivonet: Using deep neu- ral operators to optimize sensor locations with application to vortex-induced vibrations, arXiv preprint arXiv:2501.04105 (2025)

    work page arXiv 2025

  36. [36]

    A. Bora, S. Zhang, K. Shukla, B. Harrop, G. E. Karniadakis, L. R. Leung, Retrofitting earth system models with cadence-limited neural operator updates, arXiv preprint arXiv:2512.03309 (2025)

    work page arXiv 2025

  37. [37]

    Z. Li, N. Kovachki, K. Azizzadenesheli, B. Liu, K. Bhattacharya, A. Stuart, A. Anand- kumar, Fourier neural operator for parametric partial differential equations, arXiv preprint arXiv:2010.08895 (2020)

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  38. [38]

    Ronneberger, P

    O. Ronneberger, P. Fischer, T. Brox, U-net: Convolutional networks for biomedical image seg- mentation, in: Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part III 18, Springer, 2015, pp. 234–241

    work page 2015

  39. [39]

    S. Wang, H. Wang, P. Perdikaris, Learning the solution operator of parametric partial differential equations with physics-informed deeponets, Science Advances (2021) eabi8605

    work page 2021

  40. [40]

    Kiyani, M

    E. Kiyani, M. Manav, N. Kadivar, L. De Lorenzis, G. E. Karniadakis, Predicting crack nucleation andpropagationinbrittlematerialsusingdeepoperatornetworkswithdiversetrunkarchitectures, Computer Methods in Applied Mechanics and Engineering 441 (2025) 117984

    work page 2025

  41. [41]

    S. Lee, Y. Shin, On the training and generalization of deep operator networks, SIAM Journal on Scientific Computing 46 (4) (2024) C273–C296

    work page 2024

  42. [42]

    Peyvan, V

    A. Peyvan, V. Oommen, A. D. Jagtap, G. E. Karniadakis, Riemannonets: Interpretable neural operators for riemann problems, Computer Methods in Applied Mechanics and Engineering (2024) 116996

    work page 2024

  43. [43]

    Perez, F

    E. Perez, F. Strub, H. De Vries, V. Dumoulin, A. Courville, Film: Visual reasoning with a general conditioning layer, in: Proceedings of the AAAI conference on artificial intelligence, Vol. 32, 2018, pp. 1–10. 19

    work page 2018

  44. [44]

    Brockschmidt, Gnn-film: Graph neural networks with feature-wise linear modulation, in: International Conference on Machine Learning, PMLR, 2020, pp

    M. Brockschmidt, Gnn-film: Graph neural networks with feature-wise linear modulation, in: International Conference on Machine Learning, PMLR, 2020, pp. 1144–1152

    work page 2020

  45. [45]

    S. J. Pan, Transfer learning, Learning 21 (2020) 1–2

    work page 2020

  46. [46]

    R. M. Neal, Bayesian learning for neural networks, Vol. 118, Springer Science & Business Media, 2012

    work page 2012

  47. [47]

    D. J. MacKay, A practical bayesian framework for backpropagation networks, Neural computation 4 (3) (1992) 448–472

    work page 1992

  48. [48]

    Lakshminarayanan, A

    B. Lakshminarayanan, A. Pritzel, C. Blundell, Simple and scalable predictive uncertainty estima- tion using deep ensembles, Advances in neural information processing systems 30 (2017)

    work page 2017

  49. [49]

    A. G. Wilson, P. Izmailov, M. D. Hoffman, Y. Gal, Y. Li, M. F. Pradier, S. Vikram, A. Foong, S. Lotfi, S. Farquhar, Evaluating approximate inference in bayesian deep learning, in: NeurIPS 2021 Competitions and Demonstrations Track, PMLR, 2022, pp. 113–124

    work page 2021

  50. [50]

    Reich, A nonparametric ensemble transform method for bayesian inference, SIAM Journal on Scientific Computing 35 (4) (2013) A2013–A2024

    S. Reich, A nonparametric ensemble transform method for bayesian inference, SIAM Journal on Scientific Computing 35 (4) (2013) A2013–A2024

    work page 2013

  51. [51]

    Pearce, M

    T. Pearce, M. Zaki, A. Brintrup, N. Anastassacos, A. Neely, Uncertainty in neural networks: Bayesian ensembling, stat 1050 (2018) 12

    work page 2018

  52. [52]

    Pearce, F

    T. Pearce, F. Leibfried, A. Brintrup, Uncertainty in neural networks: Approximately bayesian ensembling, in: International conference on artificial intelligence and statistics, PMLR, 2020, pp. 234–244

    work page 2020

  53. [53]

    Z. Gao, G. E. Karniadakis, Scalable bayesian physics-informed kolmogorov-arnold networks, SIAM/ASA Journal on Uncertainty Quantification 13 (3) (2025) 1543–1577

    work page 2025

  54. [54]

    Pensoneault, X

    A. Pensoneault, X. Zhu, Uncertainty quantification for deeponets with ensemble kalman inversion, Journal of computational physics 523 (2025) 113670

    work page 2025

  55. [55]

    Z. Liu, Y. Wang, S. Vaidya, F. Ruehle, J. Halverson, M. Soljačić, T. Y. Hou, M. Tegmark, Kan: Kolmogorov-Arnold Networks, arXiv preprint arXiv:2404.19756 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  56. [56]

    Shukla, J

    K. Shukla, J. D. Toscano, Z. Wang, Z. Zou, G. E. Karniadakis, A comprehensive and fair compar- ison between mlp and kan representations for differential equations and operator networks, arXiv preprint arXiv:2406.02917 (2024)

    work page arXiv 2024

  57. [57]

    S. A. Faroughi, F. Mostajeran, A. H. Mashhadzadeh, S. Faroughi, Scientific machine learning with kolmogorov-arnold networks, arXiv preprint arXiv:2507.22959 (2025)

    work page arXiv 2025

  58. [58]

    D. A. Sprecher, S. Draghici, Space-filling curves and Kolmogorov superposition-based neural net- works, Neural Networks (2002) 57–67

    work page 2002

  59. [59]

    Köppen, On the training of a kolmogorov network, in: Artificial Neural Networks—ICANN 2002: International Conference, Madrid, Spain, August 28–30, 2002, 2022, pp

    M. Köppen, On the training of a kolmogorov network, in: Artificial Neural Networks—ICANN 2002: International Conference, Madrid, Spain, August 28–30, 2002, 2022, pp. 474–479

    work page 2002

  60. [60]

    L. Sun, S. S. Pang, A. M. Sterling, I. I. Negulescu, M. A. Stubblefield, Thermal analysis of curing process of epoxy prepreg, Journal of Applied Polymer Science 83 (2002) 1074–1083.doi: 10.1002/APP.10053

    work page doi:10.1002/app.10053 2002

  61. [61]

    Hargis, B

    M. Hargis, B. P. Grady, L. Aktas, K. R. Bomireddy, S. Howsman, M. C. Altan, T. Rose, H. Rose, Calorimetric and rheological measurements of three commercial thermosetting prepreg epoxies, Journal of Composite Materials 40 (2006) 873–897.doi:10.1177/0021998305056380

    work page doi:10.1177/0021998305056380 2006

  62. [62]

    J. Kim, T. J. Moon, J. R. Howell, Cure kinetic model, heat of reaction, and glass transition tem- peratureofas4/3501-6graphite–epoxyprepregs, http://dx.doi.org/10.1177/0021998302036021712 36 (2002) 2479–2498.doi:10.1177/0021998302036021712. 20

    work page doi:10.1177/0021998302036021712 2002

  63. [63]

    W. I. Lee, A. C. Loos, G. S. Springer, Heat of reaction, degree of cure, and viscosity of hercules 3501-6 resin, http://dx.doi.org/10.1177/002199838201600605 16 (1982) 510–520.doi:10.1177/ 002199838201600605

    work page doi:10.1177/002199838201600605 1982

  64. [64]

    B. C. Chern, T. J. Moon, J. R. Howell, W. Tan, New experimental data for enthalpy of reaction and temperature- and degree-of-cure-dependent specific heat and thermal conductivity of the hercules 3501-6 epoxy system, http://dx.doi.org/10.1177/0021998302036017238 36 (2002) 2061– 2072.doi:10.1177/0021998302036017238

    work page doi:10.1177/0021998302036017238 2002

  65. [65]

    T. H. Hou, J. M. Bai, A semi-empirical approach for the chemoviscosity modeling of reactive resin system, SAMPE Journal 24 (1988)

    work page 1988

  66. [66]

    M. A. Iglesias, K. J. Law, A. M. Stuart, Ensemble kalman methods for inverse problems, Inverse Problems 29 (4) (2013) 045001

    work page 2013

  67. [67]

    G. Evensen, Sequential data assimilation with a nonlinear quasi-geostrophic model using monte carlo methods to forecast error statistics, Journal of Geophysical Research: Oceans 99 (C5) (1994) 10143–10162. 21

    work page 1994