Pith. sign in

REVIEW 3 major objections 4 minor 61 references

Physics-guided denoiser network for enhanced additive manufacturing data quality

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Physics-guided denoiser beats plain networks on noisy 3D-printing data.

desk verdict The synthetic benchmarks are honest and show the denoiser idea works, but the LPBF real-data claim is circular and the Fisher-score novelty is never specified. read the letter →

arxiv 2508.02712 v1 pith:3I3KFUBO submitted 2025-07-31 eess.SP cs.LGcs.SYeess.SY

Pallock Halder , Satyajit Mojumder This is my paper
classification eess.SPcs.LGcs.SYeess.SY
keywords physics-informedneuralnetworkdenoisingenergy-basedmodelFisherscorelaserpowderbedfusionthermalemissiondatameltpoolmonitoringadditivemanufacturing
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Modern metal 3D-printing processes like laser powder bed fusion (LPBF) rely on noisy, low-cost sensors because high-fidelity monitoring tools are too expensive for routine use. This paper claims that a denoiser trained with physics guidance, rather than data alone, can remove that noise while keeping the output physically consistent. The authors build a physics-informed neural network (PINN) surrogate of the LPBF heat-conduction problem and use its outputs as the clean target for a feedforward denoiser, regularizing training with either an energy-based model (EBM) or a Fisher-score loss. They demonstrate the approach on three benchmark PDEs with up to 25% Gaussian noise and then on real photodiode thermal emission (TEP) data from AlSi10Mg single tracks, reporting that the physics-guided denoisers beat a vanilla neural network. If correct, the method makes low-cost thermal sensors usable for real-time process control and defect mitigation in additive manufacturing.

What carries the argument

The load-bearing mechanism is a two-network training loop: a pretrained PINN-AM surrogate solves the laser heat-conduction equation $\rho_0 \tilde{C}_p \dot{T} = \nabla \cdot (\tilde{k} \nabla T) + q_{vol}$ and outputs a physics-consistent temperature field; a separate feedforward denoiser takes spatial coordinates, time, process parameters, and the noisy measurement as inputs and is trained so that its output matches the surrogate through a data loss. An energy-based model (low energy for plausible outputs, negative log-likelihood loss) or a Fisher-score loss (gradient of log-likelihood with respect to parameters) adds the statistical regularizer. The surrogate target is what pulls the denoised reading toward a plausible thermal peak instead of toward the noise.

What would settle it

Take the trained denoiser to TEP tracks from the same photodiode monitoring system for which an independent high-speed measurement, such as synchrotron X-ray imaging or a co-axial radiometric reference, is available, and compare the denoised temperature-like profiles against that independent ground truth; if the denoised output inherits the surrogate's deviations rather than tracking the independent measurement, the surrogate target, not the noise, is the dominant error.

Watch

Extended reading notes

Core claim

The central claim is that adding a physics-derived target and a statistical regularizer to a denoiser network yields cleaner and more physically plausible temperature reconstructions than training on noisy data alone. The paper validates the idea first on the harmonic oscillator, Burgers' equation, and a 2D Laplace problem, where both Denoiser-EBM and Denoiser-Fisher keep RMSE and R-squared nearly constant as Gaussian noise rises from 5% to 25%, while a vanilla network degrades sharply. For LPBF, a PINN-AM surrogate of the transient heat equation is trained and validated against a simulation benchmark, then used as the clean reference for denoising synthetic and real photodiode TEP signals. On real single-track AlSi10Mg data the paper reports Denoiser-Fisher with RMSE 462.25 and SNR 13.09 dB versus RMSE 532.54 and SNR 12.56 dB for the vanilla baseline, and concludes that the physics-guided denoiser performs robustly across a range of LPBF conditions.

Load-bearing premise

The load-bearing premise is that the PINN-AM surrogate, validated only against a Hastelloy X simulation benchmark, produces correct clean temperature targets for AlSi10Mg experimental TEP signals, because the denoiser is trained to match that target and the real-data quality scores are measured against the same target.

Editorial extensions

If this is right

  • If the central claim holds, low-cost photodiode TEP data can be interpreted in near real time, removing a key obstacle to inline LPBF process control.
  • The same trained denoiser should stay stable under varying laser power and scan speed, since the PINN-AM surrogate was trained across a range of those parameters.
  • The reported metrics imply that Fisher-score regularization gives smoother, slightly more accurate denoised profiles, while EBM gives the lowest RMSE on some benchmark cases, so the regularizer can be chosen per application.
  • Because the denoiser is a feedforward network, inference is cheap enough to sit inside a sensor-data pipeline, which is the prerequisite for predicting next-layer printing parameters.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the surrogate target is trustworthy for AlSi10Mg, the same design should transfer to other low-cost sensors such as acoustic emission or broadband photodiode channels simply by swapping the physics engine, though the paper does not test that.
  • The consistent edge of Denoiser-Fisher over Denoiser-EBM across the tested cases suggests Fisher-score regularization may be the simpler default for real-time deployment, a conclusion the paper leaves implicit.
  • A strong test the paper does not run is applying the trained denoiser to a different alloy or a multi-track build without retraining; passing that would turn the robustness claim from 'across the tested LPBF conditions' into genuine process generalization.
  • Because the real-data metrics compare denoised outputs to the surrogate rather than to an independent measurement, the practical value of the framework depends on how faithfully PINN-AM reproduces true melt-pool physics for the experimental alloy.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes a physics-guided denoising framework that combines a physics-informed neural network (PINN) surrogate with either an energy-based model (EBM) or a Fisher-information-based regularizer, plus a data-consistency loss against the physics model output. The method is first tested on three synthetic benchmark PDEs with added Gaussian noise, then applied to laser powder bed fusion (LPBF) experimental photodiode signals. The authors report that the denoisers outperform a vanilla neural network and a standard PINN on the synthetic benchmarks, and they claim robust denoising performance across a range of LPBF conditions.

Significance. If the central claim were properly supported, the framework would be practically valuable: it would enable low-cost photodiode-based thermal monitoring in LPBF to be denoised in real time for process control and defect mitigation. The synthetic benchmark results are a useful contribution: they show that the proposed denoisers are more robust than a vanilla network across noise levels on the harmonic oscillator, Burgers' equation, and Laplace's equation. However, the real-data LPBF evaluation is circular because the denoiser is trained against the PINN-AM surrogate and then scored against the same surrogate, and the surrogate is not validated for the experimental material. The Fisher loss is also never specified. The main additive-manufacturing claim therefore rests on unverified ground truth, and the paper in its current form does not establish the headline result.

major comments (3)
  1. [Section 6.2, Table 5] The real-data evaluation is circular. The training data loss is the mean squared error between the denoiser output and the PINN-AM "physics model output," and Table 5 reports RMSE, MAE, and SNR for the same experimental tracks with this same PINN-AM output as the reference. Since the text itself notes the absence of ground truth in the experimental data, the metrics in Table 5 measure how closely the denoised output matches the surrogate, not how much noise is actually removed or how accurate the physical temperature is. Independent validation, such as calibrated thermometry or a surrogate validated against AlSi10Mg measurements, is required to support any real-data denoising claim.
  2. [Section 4.2, Section 4.3, Section 5] The PINN-AM surrogate is not validated for the experimental conditions to which it is applied. Section 4.3 validates the surrogate only against the simulation-based benchmark of reference [58], and the density is fixed at the Hastelloy X value rho_0 = 8352 kg/m^3, while the experimental data in Section 5 are from AlSi10Mg. No AlSi10Mg validation or recalibration is provided. In addition, the TEP values in Section 5 are photodiode signals filtered to the range 1000-2500 with no demonstrated calibration to temperature in Kelvin, whereas the PINN-AM outputs temperatures in Kelvin. The data loss in Section 6.2 may therefore compare physically incompatible quantities. If the surrogate is biased for AlSi10Mg or if TEP is not proportional to temperature, the denoised "clean" temperatures are wrong by construction.
  3. [Section 2.4, Section 6.2] The Fisher loss term is never specified. Section 2.4 defines the Fisher score and Fisher information in Eqs. (5)-(7), but the actual formula for L_Fisher used in training the Denoiser-Fisher models (Tables 1-5) is not given anywhere. Without an explicit definition of this regularization loss, the Denoiser-Fisher results are not reproducible, and it is unclear what quantity is being optimized or why it should improve denoising.
minor comments (4)
  1. [Section 6.1, Table 4] The text states that the Vanilla model's RMSE rises from 103.37 to 498.30 with increasing noise, but Table 4 reports values from 129.55 to 629.51; the table also contains malformed entries such as "12.962512.975421" and "13.026868 8.950125 17.0053313".
  2. [Section 2.4] The paper conflates the Fisher score (the gradient of the log-likelihood) with the Fisher score feature-selection criterion (a between-class to within-class variance ratio). These are different concepts, and the text should clarify which one is actually used in the proposed regularization.
  3. [Section 6.2] The denoiser architecture is described only as a fully connected feedforward network "with before mentioned details" followed by Tanh activation and dropout; the number of layers and hidden units for the denoiser itself is not stated, although the EBM and Fisher networks are specified. Full architecture details are needed for reproducibility.
  4. [Section 3.4] The discussion states that the PINN model's performance "still degrades with increasing noise," but Tables 1-3 show that the PINN RMSE and R^2 are constant across noise levels because the PINN does not take noisy data as input. The discussion should be reconciled with the tables.

Circularity Check

2 steps flagged · score 8.0 of 10

Real-data LPBF validation is circular: in Section 6.2 the denoiser is trained by a data loss that matches its output to the PINN-AM surrogate, and Table 5's RMSE/MAE/SNR are scored against that same surrogate output, so the reported noise reduction measures self-consistency with the training target rather than removal of true sensor noise.

  1. fitted input called prediction [Section 6.2, Case 2 (experimental TEP data), Table 5; framework defined in Section 2]
    "The total loss function comprised three components: (1) the data loss, computed as the mean squared error between the denoised prediction and the physics model output; (2) the physics loss; and (3) an additional regularization loss from either an Energy-Based Model (EBM) or a Fisher Score model. ... Table 5 presents the performance comparison among the three models using RMSE, SNR, and MAE."

    In Case 2 the experimental ground truth is unknown, so the only reference available for the RMSE/MAE/SNR of Table 5 is the PINN-AM surrogate output. That same output is the target of the training data loss ('the mean squared error between the denoised prediction and the physics model output'). The denoiser is therefore scored against its own training target: Table 5's Denoiser-Fisher RMSE 462.25 vs Vanilla 532.54 largely reflects that the denoiser was trained to match the surrogate while the Vanilla net was trained to reproduce the noisy data. The conclusion that the physics-guided denoiser 'reduc[es] noise' on real TEP data is forced by the evaluation protocol, not by an external measure of true noise removal.

  2. self definitional [Section 4.3 (test-case framing) and Section 6.1, Case 1]
    "This pretrained PINN-AM model is used with the denoiser network for two test cases: i) synthetic noise on single track temperature measurement where ground truth is known as the validated PINN-AM solution, ii) photodiode-based real LPBF noisy measurement (where the ground truth and noise level is unknown)"

    The LPBF 'ground truth' is defined to be the output of the very PINN-AM surrogate that supplies the denoiser's training targets, so both LPBF test cases measure only how faithfully the denoiser reproduces the surrogate, never agreement with external temperature measurements. The definitional chain is: PINN-AM output = training data-loss target = 'ground truth' = evaluation reference. Since the surrogate is validated only against the Hastelloy X benchmark of [58] (MAE 45.26, RMSE 63.53) and experimental tracks are AlSi10Mg with density fixed at the Hastelloy X value rho0 = 8352 kg/m^3, nothing in the LPBF pipeline ties the surrogate-agreement metric to true AlSi10Mg melt-pool temperatures.

full rationale

The three benchmark demonstrations (Sections 3.1-3.3) are self-contained and not circular: the denoisers are trained on analytically known solutions corrupted by known Gaussian noise and are evaluated against those same analytical ground truths, so those results stand independently. The PINN-AM surrogate is also constructed independently, from PDE residuals (Eqs. 16-17) with no fitting to the experimental TEP signals, so its training is not circular; the load-bearing external citation [58] is not by the present authors, and the authors' own prior works cited (e.g., [43]-[45]) are background rather than load-bearing. The circularity is confined to the LPBF validation chain: for the real TEP data the training data loss is the MSE between the denoiser output and the PINN-AM output (Section 6.2), and the reported real-data RMSE, MAE, and SNR (Table 5) are computed against that same surrogate output, since ground truth is acknowledged to be unknown. The improvement over the Vanilla baseline is therefore expected by construction: the Vanilla network was trained to reproduce the noisy measurements, while the denoisers were trained to match the surrogate, and both are scored against the surrogate. The synthetic case (Section 6.1) has the same definitional structure, with the surrogate output declared to be the 'ground truth.' The circularity is aggravated by validity gaps that remove any external anchor: the surrogate is validated only against a Hastelloy X benchmark (MAE 45.26, RMSE 63.53) at density 8352 kg/m^3, while the experimental tracks are AlSi10Mg, and uncalibrated TEP values in the 1000-2500 range are compared with the surrogate's temperatures in kelvin. The paper's own conclusion concedes that the model was 'not tested under wide experimental range,' undercutting the 'wide range of LPBF conditions' claim. Because the paper's headline real-data claim reduces by construction to surrogate self-consistency, the score is 8.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central real-data claim rests on unvalidated transfer of a physics surrogate across materials, an unstated equivalence between TEP signals and temperature, hand-set filtering thresholds, and unspecified loss weights. The Fisher-score regularizer is invoked but never defined, which makes the method's novel component an assumption rather than a derivation.

free parameters (2)
  • relative loss weights for L_Data, L_Physics, L_EBM/L_Fisher = not reported
    The total loss combines data, physics, and regularization terms, but no weighting coefficients or balancing schedule are given; these weights materially change the denoising behavior.
  • TEP valid range filter [1000, 2500] = 1000 to 2500
    Hand-selected thresholds discard values outside this range (Section 5); the choice affects which measurements are denoised and removes possibly meaningful spikes.
assumptions (5)
  • domain assumption The PINN-AM surrogate trained on Hastelloy X data and validated against a simulation benchmark provides accurate temperature targets for AlSi10Mg experimental TEP signals.
    Section 4.3 validates only against [58] benchmark profiles, not against experimental thermal measurement for the actual alloy used; Section 6.2 relies on this surrogate as the denoiser's supervision target.
  • domain assumption TEP photodiode values can be treated as comparable to temperature values for the purpose of denoising and RMSE evaluation.
    TEP is a raw spectral emission indicator, not a calibrated temperature; Section 5 and Figure 13 treat noisy TEP and denoised temperature on the same scale without a calibration step.
  • domain assumption The single-track heat conduction model with volumetric Gaussian source and neglected convection/radiation is sufficient to define physically consistent denoised outputs.
    Equations (16)-(17) and Section 4.1 omit melt-pool fluid dynamics, convection, and radiation; the physics loss and PINN targets inherit this simplification.
  • standard math The EBM partition function can be approximated by a discrete sum over sampled pairs.
    Equation (2)-(3) uses a standard approximation, though no sampling details are given.
  • ad hoc to paper Fisher information as defined in Eq. (7) can act as a regularizer that improves denoising.
    No derivation links Eq. (7) to a loss term; the paper's own references treat Fisher score as a feature-ranking method, not a denoising regularizer.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Physics-guided denoiser network for enhanced additive manufacturing data quality." pith.science (2026). https://pith.science/paper/3I3KFUBO

@misc{pith2026250802712,
  author       = {Pith},
  title        = {Pith review of: Physics-guided denoiser network for enhanced additive manufacturing data quality},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3I3KFUBO}},
  note         = {Machine review of arXiv:2508.02712}
}
read the original abstract

Modern engineering systems are increasingly equipped with sensors for real-time monitoring and decision-making. However, the data collected by these sensors is often noisy and difficult to interpret, limiting its utility for control and diagnostics. In this work, we propose a physics-informed denoising framework that integrates energy-based model and Fisher score regularization to jointly reduce data noise and enforce physical consistency with a physics-based model. The approach is first validated on benchmark problems, including the simple harmonic oscillator, Burgers' equation, and Laplace's equation, across varying noise levels. We then apply the denoising framework to real thermal emission data from laser powder bed fusion (LPBF) additive manufacturing experiments, using a trained Physics-Informed Neural Network (PINN) surrogate model of the LPBF process to guide denoising. Results show that the proposed method outperforms baseline neural network denoisers, effectively reducing noise under a range of LPBF processing conditions. This physics-guided denoising strategy enables robust, real-time interpretation of low-cost sensor data, facilitating predictive control and improved defect mitigation in additive manufacturing.

Figures

Figures reproduced from arXiv: 2508.02712 by the authors.

Figure 1
Figure 1. Overview of the proposed physics-informed denoiser framework combining a PINN, data consis [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Comparison of SHO predictions from (a) Denoiser-EBM and (b) Denoiser-Fisher models. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Comparison of the reference and denoised solutions for the Burgers’ equation. (a) Analytical refer [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: Denoised temporal profiles for the Burgers’ equation at selected time instances. (a), (b) Solutions [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Comparison of analytical, noisy, and denoised solutions for the 2D Laplace equation with a centrally [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: Cross-sectional slices of the denoised Laplace solution at [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]
Figure 7
Figure 7. Figure 7: Domain geometry used for LPBF single-track simulation, adapted from [58]. [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]
Figure 8
Figure 8. Figure 8: PINN-AM architecture for LPBF thermal analysis with 8 inputs, 6 hidden layers, and physics [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]
Figure 10
Figure 10. Figure 10: The system is based on the PrintRite3D® platform (Sigma Additive Solutions, USA), which captures high-speed spectral emissions from the melt pool during laser pow￾der bed fusion (LPBF). Emissions are collected using three photodiodes that separately 16 [PITH_FULL_IMA…
Figure 9
Figure 9. Figure 9: Validation of the proposed model against benchmark data from [58]. The predicted temperature [PITH_FULL_IMAGE:figures/full_fig_p017_9.png]
Figure 10
Figure 10. Figure 10: Schematic of the experimental setup for in-situ melt pool monitoring system using photodiodes. [PITH_FULL_IMAGE:figures/full_fig_p018_10.png]
Figure 11
Figure 11. Figure 11: Noisy TEP data across four single laser tracks collected using the photodiodes. These noisy data [PITH_FULL_IMAGE:figures/full_fig_p019_11.png]
Figure 12
Figure 12. Figure 12: Comparison of denoised temperature profiles against noisy and reference solutions. (a) Denoiser [PITH_FULL_IMAGE:figures/full_fig_p020_12.png]
Figure 13
Figure 13. Figure 13: Comparison of denoiser models trained with different uncertainty regularization strategies on [PITH_FULL_IMAGE:figures/full_fig_p020_13.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

61 extracted references · 43 canonical work pages

  1. [58]

    Hosseini, P

    E. Hosseini, P. Scheel, O. M¨ uller, R. Molinaro, S. Mishra, Single-track thermal analysis of laser powder bed fusion process: Parametric solution through physics-informed neural networks, Computer Methods in Applied Mechanics and Engineering 410 (2023) 116019. doi:https://doi.org/10.1016/j.cma.2023.116019. URL https://www.sciencedirect.com/science/articl...

    work page arXiv 2023

  2. [40]

    Pilar, N

    P. Pilar, N. Wahlstr¨ om, Physics-informed neural networks with unknown measurement noise, in: A. Abate, M. Cannon, K. Margellos, A. Papachristodoulou (Eds.), Proceed- ings of the 6th Annual Learning for Dynamics and amp Control Conference, Vol. 242 of Proceedings of Machine Learning Research, PMLR, 2024, pp. 235–247. URL https://proceedings.mlr.press/v24...

    work page 2024

  3. [1]

    Real-Time 2D Temperature Field Prediction in Metal Additive Manufacturing Using Physics-Informed Neural Networks

    P. Sajadi, M. R. Dehaghani, Y. Tang, G. G. Wang, Real-time 2d temperature field pre- diction in metal additive manufacturing using physics-informed neural networks (2024). arXiv:2401.02403. URL https://arxiv.org/abs/2401.02403

    work page Pith review arXiv 2024

  4. [2]

    V. V. Bhandarkar, B. Das, P. Tandon, Real-time remote monitoring and defect detection in smart additive manufacturing for reduced material wastage, Measurement 252 (2025) 117362. doi:10.1016/j.measurement.2025.117362. URL https://www.sciencedirect.com/science/article/pii/S0263224125007213

    arXiv 2025

  5. [3]

    Richter, T

    A. Richter, T. Gehling, K. Treutler, V. Wesling, C. Rembe, Real-time measurement of temperature and volume of the weld pool in wire-arc additive manufacturing, Measure- ment: Sensors 17 (2021) 100060. doi:https://doi.org/10.1016/j.measen.2021.100060. URL https://www.sciencedirect.com/science/article/pii/S2665917421000234

    work page arXiv 2021

  6. [4]

    Z. Ren, L. Gao, S. J. Clark, K. Fezzaa, P. Shevchenko, A. Choi, W. Everhart, A. D. Rollett, L. Chen, T. Sun, Machine learning-aided real-time detection of key- hole pore generation in laser powder bed fusion, Science 379 (6627) (2023) 89–94. doi:10.1126/science.add4667. URL https://www.science.org/doi/10.1126/science.add4667

    work page doi:10.1126/science.add4667 2023

  7. [5]

    Z. Ren, J. Shao, H. Liu, S. Clark, L. Gao, L. Balderson, K. Mumm, K. Fezzaa, A. Rollett, L. Kara, T. Sun, Sub-millisecond keyhole pore detection in laser powder bed fusion using sound and light sensors and machine learning, Materials Futures 3 (11 2024). doi:10.1088/2752-5724/ad89e2

    work page doi:10.1088/2752-5724/ad89e2 2024

  8. [6]

    Mutswatiwa, L

    L. Mutswatiwa, L. Katch, N. J. Kizer, J. A. Todd, T. Sun, S. J. Clark, K. Fezzaa, J. S. Lum, D. M. Stobbe, G. Jones, K. C. Meinert, A. P. Arg¨ uelles, C. M. Kube, High-speed synchrotron x-ray imaging of melt pool dynamics during ultrasonic melt processing of al6061, Communications Materials 5 (1), publisher Copyright: © The Author(s) 2024 (Dec. 2024). doi...

    work page doi:10.1038/s43246-024-00584-3 2024

Show all 61 references
  1. [7]

    Real-Time Print Tracking in Metal Additive Manufacturing Using Acoustic Emission Sensors and Vision Transformer Algorithms, ASME,

  2. [8]

    W. Li, A. Amanov, K. M N, B. Li, B. Ravichander, R. Zhang, H. Lu, D. Qian, G. Kumar, Y. Pyun, Processing aluminum alloy with hybrid wire arc additive man- ufacturing and ultrasonic nanocrystalline surface modification to improve porosity, sur- face finish, and hardness, Journa...

    2023 doi

  3. [9]

    Carter III, D

    F. Carter III, D. Kozjek, C. Porter, S. Clark, K. Fezzaa, M. Fujishima, N. Irino, J. Cao, Melt pool instability detection using coaxial photodiode system validated by in-situ x-ray imaging, CIRP Annals 72 (04 2023). doi:10.1016/j.cirp.2023.03.031

    2023 doi

  4. [10]

    Jeong, S

    J. Jeong, S. Webster, S. Liao, J.-E. Mogonye, K. Ehmann, J. Cao, Cool- ing rate measurement in directed energy deposition using photodiode-based planck thermometry (pdpt), Additive Manufacturing Letters 3 (2022) 100101. doi:https://doi.org/10.1016/j.addlet.2022.100101. URL htt...

    2022

  5. [11]

    L. Chen, X. Yao, N. Ng, S. Moon, In-situ melt pool monitoring of laser aided additive manufacturing using infrared thermal imaging, in: 2022 IEEE International Conference on Industrial Engineering and Engineering Management (IEEM), 2022, pp. 1478–1482. doi:10.1109/IEEM55944.20...

    2022

  6. [12]

    Kozjek, F

    D. Kozjek, F. Carter, C. Porter, J. Mogonye, K. Ehmann, J. Cao, Data-driven prediction of next-layer melt pool temperatures in laser powder bed fusion based on co-axial high- resolution planck thermometry measurements, Journal of Manufacturing Processes 79 (2022) 81–90, publis...

    2022 doi

  7. [13]

    Kozjek, C

    D. Kozjek, C. Porter, F. Carter, J. Mogonye, J. Cao, Data-driven predic- tion of inter-layer process condition variations in laser powder bed fusion, Addi- tive Manufacturing 88, publisher Copyright: © 2024 Elsevier B.V. (May 2024). doi:10.1016/j.addma.2024.104230

    2024

  8. [14]

    Mahmoodi, B

    S. Mahmoodi, B. Sharif, Noise reduction, smoothing and time interval segmentation of noisy signals using an energy optimisation method, Vision, Image and Signal Processing, IEE Proceedings - 153 (2006) 101 – 108. doi:10.1049/ip-vis:20045205

    2006 doi

  9. [15]

    Shtayat, S

    A. Shtayat, S. Moridpour, B. Best, H. Daoud, Application of noise-cancelling and smoothing techniques in road pavement vibration monitoring data, Inter- national Journal of Transportation Science and Technology 14 (2024) 110–119. doi:https://doi.org/10.1016/j.ijtst.2023.04.002...

    2024 doi

  10. [16]

    Aggarwal, V

    V. Aggarwal, V. Gupta, P. Singh, K. Sharma, S. Neetu, Detection of spatial outlier by using improved z-score test, 2019, pp. 788–790. doi:10.1109/ICOEI.2019.8862582

    2019

  11. [17]

    Thompson, Llnl-tr-659435 an empirical evaluation of denoising techniques for stream- ing data (08 2014)

    J. Thompson, Llnl-tr-659435 an empirical evaluation of denoising techniques for stream- ing data (08 2014). doi:10.2172/1165751

    2014 doi

  12. [18]

    Spiegelberg, J

    J. Spiegelberg, J. Rusz, Can we use pca to detect small signals in noisy data?, Ultrami- croscopy 172 (2017) 40–46. doi:https://doi.org/10.1016/j.ultramic.2016.10.008. URL https://www.sciencedirect.com/science/article/pii/S0304399116302480

    2017 doi

  13. [19]

    Samann, T

    F. Samann, T. Schanze, Denoising by spectral selections of svd representations of han- kel matricificated data with application to ppg signals, IF AC-PapersOnLine 58 (24) (2024) 175–180, 12th IF AC Symposium on Biological and Medical Systems BMS 2024. doi:https://doi.org/10.10...

    2024 doi

  14. [20]

    G. Ji, C. Wang, A denoising method for seismic data based on svd and deep learning, Applied Sciences 12 (24) (2022). doi:10.3390/app122412840. URL https://www.mdpi.com/2076-3417/12/24/12840

    2022 doi

  15. [21]

    R. E. Kalman, A new approach to linear filtering and prediction problems, Journal of Basic Engineering 82 (1) (1960) 35–45. doi:10.1115/1.3662552. URL https://doi.org/10.1115/1.3662552

    1960 doi

  16. [22]

    Park, M.-S

    S. Park, M.-S. Gil, H. Im, Y.-S. Moon, Measurement noise recommendation for ef- ficient kalman filtering over a large amount of sensor data, Sensors 19 (5) (2019). doi:10.3390/s19051168. URL https://www.mdpi.com/1424-8220/19/5/1168

    2019 doi

  17. [23]

    Camerlingo, I

    N. Camerlingo, I. Siviero, M. Vettoretti, G. Sparacino, S. Del Favero, A. Facchinetti, Bayesian denoising algorithm dealing with colored, non-stationary noise in continuous glucose monitoring timeseries, Frontiers in Bioengineering and Biotechnology 11 (2023) 1280233, publishe...

    2023

  18. [24]

    B. Deng, H. Yang, N. Y. Kim, A denoising autoencoder based on u-net and bidirectional long short-term memory for multi-level random telegraph sig- nal analysis, Engineering Applications of Artificial Intelligence 135 (2024) 108685. doi:https://doi.org/10.1016/j.engappai.2024.1...

    2024

  19. [25]

    Gondara, Medical image denoising using convolutional denoising autoencoders, in: 2016 IEEE 16th International Conference on Data Mining Workshops (ICDMW), 2016, pp

    L. Gondara, Medical image denoising using convolutional denoising autoencoders, in: 2016 IEEE 16th International Conference on Data Mining Workshops (ICDMW), 2016, pp. 241–246. doi:10.1109/ICDMW.2016.0041

    2016

  20. [26]

    Bengio, L

    Y. Bengio, L. Yao, G. Alain, P. Vincent, Generalized denoising auto-encoders as gener- ative models (2013). arXiv:1305.6663. URL https://arxiv.org/abs/1305.6663 24

    2013 arXiv

  21. [27]

    Z. Zhao, Z. Wu, Y. Zheng, P. Ma, Recurrent neural networks for at- mospheric noise removal from insar time series with missing values, IS- PRS Journal of Photogrammetry and Remote Sensing 180 (2021) 227–237. doi:https://doi.org/10.1016/j.isprsjprs.2021.08.009. URL https://www....

    2021 doi

  22. [28]

    D. Zhu, L. Fu, V. Kazei, W. Li, Diffusion model for das-vsp data denoising, Sensors 23 (20) (2023). doi:10.3390/s23208619. URL https://www.mdpi.com/1424-8220/23/20/8619

    2023 doi

  23. [29]

    Lagaris, A

    I. Lagaris, A. Likas, D. Fotiadis, Artificial neural networks for solving ordinary and partial differential equations, IEEE Transactions on Neural Networks 9 (5) (1998) 987–

    1998

  24. [30]

    Raissi, P

    M. Raissi, P. Perdikaris, G. Karniadakis, Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear par- tial differential equations, Journal of Computational Physics 378 (2019) 686–707. doi:https://doi.org/10.1016...

    2019 doi

  25. [31]

    C. Park, Y. Lu, S. Saha, T. Xue, J. Guo, S. Mojumder, D. W. Apley, G. J. Wag- ner, W. K. Liu, Convolution hierarchical deep-learning neural network (c-hidenn) with graphics processing unit (gpu) acceleration, Comput. Mech. 72 (2) (2023) 383–409. doi:10.1007/s00466-023-02329-4....

    2023 doi

  26. [32]

    H. Hu, L. Qi, X. Chao, Physics-informed neural networks (pinn) for computational solid mechanics: Numerical frameworks and applications, Thin-Walled Structures (12 2024). doi:10.1016/j.tws.2024.112495

    2024

  27. [33]

    Haghighat, M

    E. Haghighat, M. Raissi, A. Moure, H. Gomez, R. Juanes, A physics-informed deep learning framework for inversion and surrogate modeling in solid mechan- ics, Computer Methods in Applied Mechanics and Engineering 379 (2021) 113741. doi:10.1016/j.cma.2021.113741

    2021

  28. [34]

    S. Cai, Z. Mao, Z. Wang, G. E. Karniadakis, Physics-informed neural networks (pinns) for fluid mechanics: a review, Acta Mechanica Sinica 37 (2021) 1727–1738, published online: 23 Jan 2022. doi:10.1007/s10409-021-01148-1. URL https://doi.org/10.1007/s10409-021-01148-1

    2021 doi

  29. [35]

    D. Kats, Z. Wang, Z. Gan, W. K. Liu, G. J. Wagner, Y. Lian, A physics-informed machine learning method for predicting grain structure characteristics in directed energy deposition, Computational Materials Science 202 (2022) 110958, © 2021 Elsevier B.V. doi:10.1016/j.commatsci....

    2022

  30. [36]

    Zamiela, R

    C. Zamiela, R. Stokes, W. Tian, H. Doude, M. W. Priddy, L. Bian, Physics-informed approximation of internal thermal history for surface deformation predictions in wire arc directed energy deposition, Journal of Manufacturing Science and Engineering 146 (8) 25 (2024) 081007. do...

    2024 doi

  31. [37]

    Cervantes, M

    ´Alvaro Moreno Soto, A. Cervantes, M. Soler, Physics-informed neural networks for high- resolution weather reconstruction from sparse weather stations, Open Research Europe 4 (2024) 99. doi:10.12688/openreseurope.17388.1. URL https://doi.org/10.12688/openreseurope.17388.1

    2024 doi

  32. [38]

    LeCun, S

    Y. LeCun, S. Chopra, R. Hadsell, M. A. Ranzato, F. J. Huang, A tutorial on energy- based learning, in: G. Bakir, T. Hofman, B. Sch¨ olkopf, A. Smola, B. Taskar (Eds.), Predicting Structured Data, MIT Press, 2006

    2006

  33. [39]

    M. R. Osborne, Fisher’s method of scoring, International Statistical Review / Revue Internationale de Statistique 60 (1) (1992) 99–117. URL http://www.jstor.org/stable/1403504

    1992

  34. [41]

    F. K. Gustafsson, M. Danelljan, G. Bhat, T. B. Sch¨ on, Energy-based models for deep probabilistic regression, in: A. Vedaldi, H. Bischof, T. Brox, J.-M. Frahm (Eds.), Com- puter Vision – ECCV 2020, Vol. 12365 of Lecture Notes in Computer Science, Springer, Cham, 2020, pp. 344...

    2020 doi

  35. [42]

    Q. Gu, Z. Li, J. Han, Generalized fisher score for feature selection, in: Proceedings of the Twenty-Seventh Conference on Uncertainty in Artificial Intelligence, UAI’11, AUAI Press, Arlington, Virginia, USA, 2011, p. 266–273

    2011

  36. [43]

    W. Liu, S. Mojumder, W. K. Liu, W. Chen, D. W. Apley, Simulation-free determi- nation of microstructure representative volume element size via fisher scores, APL Machine Learning 2 (2) (2024) 026101. arXiv:https://pubs.aip.org/aip/aml/article- pdf/doi/10.1063/5.0195232/1986004...

    2024 doi

  37. [44]

    Y. Li, S. Mojumder, Y. Lu, A. A. Amin, J. Guo, X. Xie, W. Chen, G. J. Wagner, J. Cao, W. K. Liu, Statistical parameterized physics-based machine learning digital shadow models for laser powder bed fusion process, Additive Manufacturing 87 (2024) 104214. doi:10.1016/j.addma.2024.104214

    2024

  38. [45]

    Mojumder, Z

    S. Mojumder, Z. Gan, Y. Li, A. Amin, W. Liu, Linking process parameters with lack- of-fusion porosity for laser powder bed fusion metal additive manufacturing, Additive Manufacturing 68 (2023) 103500. doi:10.1016/j.addma.2023.103500. 26

    2023

  39. [46]

    Z. Gan, Y. Lian, S. Lin, K. Jones, W. Liu, G. Wagner, Benchmark study of ther- mal behavior, surface topography, and dendritic microstructure in selective laser melt- ing of inconel 625, Integrating Materials and Manufacturing Innovation 8 (04 2019). doi:10.1007/s40192-019-00130-x

    2019 doi

  40. [47]

    J. P. Leonor, G. J. Wagner, Go-melt: Gpu-optimized multilevel execution of lpbf ther- mal simulations, Computer Methods in Applied Mechanics and Engineering 426 (2024) 116977. doi:10.1016/j.cma.2024.116977

    2024

  41. [48]

    Sarker, S

    A. Sarker, S. Mojumder, Data-driven insights into melt pool dynamics and energy absorption in the laser powder bed fusion process, SSRN, 34 Pages, Posted: 30 Jan 2025 (2025). doi:10.2139/ssrn.5117777. URL https://ssrn.com/abstract=5117777

    2025 doi

  42. [49]

    J. Guo, G. Domel, C. Park, H. Zhang, O. C. Gumus, Y. Lu, G. J. Wagner, D. Qian, J. Cao, T. J. R. Hughes, W. K. Liu, Tensor-decomposition-based a priori surrogate (taps) modeling for ultra large-scale simulations (2025). arXiv:2503.13933. URL https://arxiv.org/abs/2503.13933

    2025

  43. [50]

    Y. Lu, H. Li, L. Zhang, C. Park, S. Mojumder, S. Knapik, Z. Sang, S. Tang, D. W. Apley, G. J. Wagner, et al., Convolution hierarchical deep-learning neural networks (c-hidenn): finite elements, isogeometric analysis, tensor decomposition, and beyond, Computational Mechanics 72...

    2023 doi

  44. [51]

    Y. Lu, S. Mojumder, J. Guo, Y. Li, W. K. Liu, Extended tensor decomposition model reduction methods: training, prediction, and design under uncertainty (2023). arXiv:2307.15873. URL https://arxiv.org/abs/2307.15873

    2023 arXiv

  45. [52]

    J. Tang, S. Kumar, L. De Lorenzis, E. Hosseini, Neural cellular automata for solidifi- cation microstructure modelling, Computer Methods in Applied Mechanics and Engi- neering 414 (2023) 116197. doi:https://doi.org/10.1016/j.cma.2023.116197. URL https://www.sciencedirect.com/s...

    2023

  46. [53]

    J. Tang, P. Scheel, M. S. Mohebbi, C. Leinenbach, L. De Lorenzis, E. Hosseini, On the calibration of thermo-microstructural simulation models for laser powder bed fusion process: Integrating physics-informed neural networks with cellular automata, Additive Manufacturing 96 (20...

    2024

  47. [54]

    Tae, Understanding the fisher information matrix, accessed: 2025-07-19 (2021)

    J. Tae, Understanding the fisher information matrix, accessed: 2025-07-19 (2021). URL https://jaketae.github.io/study/fisher/

    2021

  48. [55]

    Burgers, A mathematical model illustrating the theory of turbulence, Vol

    J. Burgers, A mathematical model illustrating the theory of turbulence, Vol. 1 of Advances in Applied Mechanics, Elsevier, 1948, pp. 171–199. doi:https://doi.org/10.1016/S0065-2156(08)70100-5. URL https://www.sciencedirect.com/science/article/pii/S0065215608701005 27

    1948 doi

  49. [56]

    L. Lu, X. Meng, Z. Mao, G. E. Karniadakis, Deepxde: A deep learning library for solv- ing differential equations, SIAM Review 63 (1) (2021) 208–228, dataset retrieved from https://github.com/lululxvi/deepxde/blob/master/examples/dataset/Burgers.npz. doi:10.1137/19M1274067. URL...

    2021 doi

  50. [57]

    Zhang, Y

    L. Zhang, Y. Lu, S. Tang, W. K. Liu, Hidenn-td: Reduced-order hierarchical deep learning neural networks, Computer Methods in Applied Mechanics and Engineering 389 (2022) 114414. doi:https://doi.org/10.1016/j.cma.2021.114414. URL https://www.sciencedirect.com/science/article/p...

    2022

  51. [59]

    Porter, F

    C. Porter, F. Carter, D. Kozjek, A. Cardona, J.-E. Mogonye, K. Ehmann, J. Cao, Generation and analysis of alsi10mg l-pbf single track data set enabling deeper process insights, in: Proceedings of the 2022 International Solid Freeform Fabrication Sympo- sium, University of Texa...

    2022 doi

  52. [1000]

    doi:10.1109/72.712178

    doi

  53. [2024]

    URL https://doi.org/10.1115/MSEC2024-125391

    arXiv:https://asmedigitalcollection.asme.org/MSEC/proceedings- pdf/MSEC2024/88117/V002T07A009/7366787/v002t07a009-msec2024-125391.pdf, doi:10.1115/MSEC2024-125391. URL https://doi.org/10.1115/MSEC2024-125391

    doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.