pith. sign in

arxiv: 2606.23719 · v1 · pith:PYSQVNC5new · submitted 2026-06-17 · 🪐 quant-ph · cs.LG

A Hybrid Quantum-Classical Approach for Melt Pool Prediction in Laser Powder Bed Fusion

Matthew M. Sato , Kincho H. Law This is my paper

Pith reviewed 2026-06-26 20:54 UTC · model grok-4.3

classification 🪐 quant-ph cs.LG
keywords hybrid quantum-classical modelquantum feature encoderlaser powder bed fusionmelt pool predictionadditive manufacturingNISQ devicesclustering for quantum computationshot noise analysis
0
0 comments X

The pith

A hybrid quantum-classical model improves melt pool predictions in laser powder bed fusion beyond what classical networks achieve alone.

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

The paper develops a hybrid model that first clusters process parameters to limit the number of quantum computations, then applies a quantum feature encoder to extract information using entanglement and superposition, and finally feeds those features into a classical neural network to predict melt pool morphology. The authors test this pipeline on a quantum simulator, examine the impact of measurement shot noise, and confirm results on actual quantum hardware. They report better predictive performance than purely classical networks and extract insights from the most influential quantum features to guide circuit design. A reader would care because accurate melt pool prediction helps assess part quality before manufacturing in an additive process known for quality variability.

Core claim

The hybrid quantum-classical model, which employs a quantum feature encoder after clustering and before classical neural network processing, produces more accurate melt pool morphology predictions from process parameters than purely classical networks, with the improvement holding under shot noise on simulators and verified on quantum hardware.

What carries the argument

Quantum feature encoder that processes clustered LPBF process parameters to produce features for input to a classical neural network.

If this is right

  • Clustering reduces the number of quantum circuit evaluations enough to make the approach feasible for realistic dataset sizes.
  • Measurement shot noise affects but does not eliminate the hybrid model's advantage over classical baselines.
  • Analysis of which quantum features matter most supplies guidance for designing improved quantum encoding circuits.
  • The overall pipeline constitutes a working engineering demonstration of NISQ hardware in an additive manufacturing prediction task.

Where Pith is reading between the lines

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

  • The same clustering-plus-quantum-encoder pattern could be tested on other manufacturing processes whose physics resist classical feature extraction.
  • If hardware noise decreases, the performance gap between hybrid and classical models may widen without changes to the encoding circuit.
  • The feature-importance analysis opens a route to co-designing quantum circuits specifically for process-parameter data rather than generic encodings.

Load-bearing premise

The quantum feature encoder must extract predictive information that classical networks cannot obtain on their own, and the clustering step must retain all variations in process parameters that matter for melt pool outcomes.

What would settle it

A classical neural network trained on the same unclustered or identically clustered dataset achieving equal or higher prediction accuracy than the hybrid model on the same test cases would falsify the performance improvement claim.

read the original abstract

Laser powder bed fusion (LPBF) is a promising additive manufacturing technique that suffers from quality assurance concerns. Predicting melt pools from process parameters is crucial for assessing quality prior to manufacturing but remains a difficult problem because of the complex physical processes underlying LPBF. Quantum computers present a new computing paradigm, providing a new approach to information processing using quantum entanglement and superposition. This paper presents a practical demonstration of a hybrid quantum-classical model that leverages quantum computing to improve process parameter feature extraction with a quantum feature encoder. To make the quantum approach computationally feasible for large datasets, we first employ a clustering algorithm to reduce the number of expensive quantum computations. These quantum features are then processed by a classical neural network to predict the melt pool morphology, allowing for more accurate predictions of melt pools. We demonstrate the method using a quantum simulator, analyze the effect of measurement shot noise on the predictive performance of the network, and verify the results using quantum hardware. Finally, by examining which quantum features are most important, we provide insights that can inform the future design of more effective quantum encoding circuits. Ultimately, the performance improvement over purely classical networks validates the hybrid approach, demonstrating an engineering application of quantum computing using noisy and intermediate scale quantum (NISQ) devices.

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 proposes a hybrid quantum-classical model for predicting melt-pool morphology in laser powder bed fusion (LPBF). Process parameters are clustered to reduce the number of quantum circuit evaluations; a quantum feature encoder extracts features from one representative per cluster; these features are fed to a classical neural network for prediction. The work includes simulator experiments, shot-noise analysis, hardware verification on NISQ devices, and an analysis of which quantum features are most important.

Significance. If the claimed performance lift over classical baselines is reproducible and attributable to the quantum encoder rather than to dataset reduction, the result would constitute a concrete engineering demonstration of NISQ hardware for an industrially relevant regression task and would supply empirical guidance on quantum encoding design.

major comments (2)
  1. [Clustering subsection] Clustering subsection (method description): the manuscript supplies no quantitative check (within-cluster variance of melt-pool depth/width, silhouette score correlated with the target variable, or ablation of cluster count versus prediction error) that would confirm the single quantum feature vector per cluster preserves the mapping from process parameters to morphology. This check is load-bearing for the central claim that any observed improvement is due to the quantum encoder rather than an artifact of the reduced dataset.
  2. [Results section] Results section (performance comparison): the abstract asserts a performance improvement over purely classical networks, yet the manuscript provides neither numerical values for the lift, baseline architectures, error bars, nor dataset size/split details. Without these, the empirical validation of the hybrid approach cannot be assessed.
minor comments (2)
  1. [Hardware verification] The hardware verification paragraph would benefit from explicit statement of the number of shots used and the device noise model employed.
  2. [Quantum feature encoder] Notation for the quantum feature map (e.g., the precise form of the encoding circuit) should be given in an equation rather than only in prose.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback on our manuscript. We address each major comment below and commit to revisions that strengthen the empirical validation of our hybrid approach.

read point-by-point responses

  1. Referee: [Clustering subsection] Clustering subsection (method description): the manuscript supplies no quantitative check (within-cluster variance of melt-pool depth/width, silhouette score correlated with the target variable, or ablation of cluster count versus prediction error) that would confirm the single quantum feature vector per cluster preserves the mapping from process parameters to morphology. This check is load-bearing for the central claim that any observed improvement is due to the quantum encoder rather than an artifact of the reduced dataset.

    Authors: We agree that a quantitative validation of the clustering step is essential to rule out artifacts from dataset reduction. In the revised manuscript we will add silhouette scores for the chosen clustering, within-cluster variance statistics on the target melt-pool variables, and an ablation study of cluster count versus final prediction error. These additions will directly address whether the single representative per cluster preserves the parameter-to-morphology mapping. revision: yes

  2. Referee: [Results section] Results section (performance comparison): the abstract asserts a performance improvement over purely classical networks, yet the manuscript provides neither numerical values for the lift, baseline architectures, error bars, nor dataset size/split details. Without these, the empirical validation of the hybrid approach cannot be assessed.

    Authors: We acknowledge the omission of explicit numerical details. The revised manuscript will report the precise performance lifts (including the specific error metrics and their magnitudes), fully describe the classical baseline network architectures, include error bars obtained from repeated training runs with different random seeds, and state the total dataset size together with the train/validation/test split ratios used. revision: yes

Circularity Check

0 steps flagged

No circularity; empirical model comparison is independent of inputs

full rationale

The paper describes an empirical hybrid quantum-classical pipeline for melt-pool prediction: clustering reduces quantum evaluations, a quantum feature encoder produces inputs to a classical neural network, and performance is assessed by direct comparison against purely classical baselines on held-out data. No derivation chain, uniqueness theorem, or fitted-parameter-as-prediction step is present; the central claim rests on measured accuracy lift rather than any equation that reduces to its own inputs by construction. Self-citations, if present, are not load-bearing for the validation. The method is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Ledger is necessarily incomplete because only the abstract is available; it captures high-level assumptions from the described approach. No explicit free parameters, axioms, or invented entities are detailed in the provided text.

axioms (2)
  • domain assumption Quantum feature encoding via superposition and entanglement can extract useful features for this regression task that classical methods miss.
    Central to the hybrid model design stated in the abstract.
  • domain assumption Clustering reduces quantum computation cost without discarding information needed for accurate melt pool prediction.
    Invoked to make the method feasible for large datasets.

pith-pipeline@v0.9.1-grok · 5750 in / 1326 out tokens · 28555 ms · 2026-06-26T20:54:10.378309+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

32 extracted references · 28 canonical work pages · 1 internal anchor

  1. [1]

    This manuscript is the author's accepted version

    1 [Accepted for presentation at the ASME IDETC/CIE 2026 Conference. This manuscript is the author's accepted version. Copyright © ASME] A Hybrid Quantum-Classical Approach for Melt Pool Prediction in Laser Powder Bed Fusion Matthew M. Sato and Kincho H. Law Engineering Informatics Group Department of Civil and Environmental Engineering Stanford University...

    2026

  2. [2]

    Process parameter optimization of metal additive manufacturing: a review and outlook

    Chia, H. Y ., Wu, J., Wang, X., and Yan, W. “Process parameter optimization of metal additive manufacturing: a review and outlook.” Journal of Materials Informatics Vo l . 2 No. 4 (2022): pp

    2022

  3. [3]

    DOI 10.20517/jmi.2022.18

    work page doi:10.20517/jmi.2022.18 2022

  4. [4]

    Effects of process parameters on porosity in laser powder bed fusion revealed by X-ray tomography

    du Plessis, A. “Effects of process parameters on porosity in laser powder bed fusion revealed by X-ray tomography.” Additive Manufacturing Vo l . 3 0 ( 2 0 1 9 ) : p p . 1 0 0 8 7 1 . D O I 10.1016/j.addma.2019.100871

    work page doi:10.1016/j.addma.2019.100871 2019

  5. [5]

    In-situ sensing, process monitoring and machine control in Laser Powder Bed Fusion: A review

    McCann, R., Obeidi, M. A., Hughes, C., McCarthy, É., Egan, D. S., et al. “In-situ sensing, process monitoring and machine control in Laser Powder Bed Fusion: A review.” Additive Manufacturing Vo l . 4 5 ( 2 0 2 1 ) : p p . 1 0 2 0 5 8 . D O I 10.1016/j.addma.2021.102058

    work page doi:10.1016/j.addma.2021.102058 2021

  6. [6]

    On the application of in-situ monitoring systems and machine learning algorithms for developing quality assurance platforms in laser powder bed fusion: A review

    Taherkhani, K., Ero, O., Liravi, F., Toorandaz, S., and Toyserkani, E. “On the application of in-situ monitoring systems and machine learning algorithms for developing quality assurance platforms in laser powder bed fusion: A review.” Journal of Manufacturing Processes Vo l . 9 9 (2023): pp. 848–897. DOI 10.1016/j.jmapro.2023.05.048

    work page doi:10.1016/j.jmapro.2023.05.048 2023

  7. [7]

    and Google Quantum AI and Collaborators , title =

    Acharya, R., Abanin, D. A., Aghababaie-Beni, L., Aleiner, I., Andersen, T. I., et al. “Quantum error correction below the surface code threshold.” Nature Vo l . 6 3 8 N o . 8 0 5 2 (2025): pp. 920–926. DOI 10.1038/s41586-024-08449-y

    work page doi:10.1038/s41586-024-08449-y 2025

  8. [8]

    Evidence for the utility of quantum computing before fault tolerance,

    Kim, Y ., Eddins, A., Anand, S., Wei, K. X., van den Berg, E., et al. “Evidence for the utility of quantum computing before fault tolerance.” Nature Vo l . 6 1 8 N o . 7 9 6 5 ( 2 0 2 3 ) : pp. 500–505. DOI 10.1038/s41586-023-06096-3

    work page doi:10.1038/s41586-023-06096-3

  9. [9]

    Nielsen, M. A. and Chuang, I. L. Quantum Computation and Quantum Information: 10th Anniversary Edition. Cambridge University Press (2010)

    2010

  10. [10]

    , 𝜎&!, 𝜎'(), and 𝜎() observables. 10 fusion additive manufacturing

    Liu, J., Ye, J., Silva Izquierdo, D., Vinel, A., Shamsaei, N., et al. “A review of machine learning techniques for process and performance optimization in laser beam powder bed Figure 7: The 20 highest importance observables and the histograms for the 𝜎%", 𝜎&!, 𝜎'(), and 𝜎() observables. 10 fusion additive manufacturing.” Journal of Intelligent Manufactur...

    work page doi:10.1007/s10845-022-02012-0

  11. [11]

    Quantum Computing in the NISQ era and beyond

    Preskill, J. “Quantum Computing in the NISQ era and beyond.” Quantum Vo l . 2 ( 2 0 1 8 ) : p p . 7 9 . D O I 1 0 . 2 2 3 3 1 / q-2018-08-06-79

    2018

  12. [12]

    Selection of Optimal Process Parameters to Obtain Defect-Free Builds in IN718 Made by Laser Powder Bed Fusion

    Reddy, K. P. K., Rao, B. N., Nazeemudheen, M. N., Manwatkar, S. K., and Murty, S. V . S. N. “Selection of Optimal Process Parameters to Obtain Defect-Free Builds in IN718 Made by Laser Powder Bed Fusion.” Journal of Materials Engineering and Performance Vo l . 3 3 N o . 1 9 (2024): pp. 10228–10241. DOI 10.1007/s11665-023-08677-9

    work page doi:10.1007/s11665-023-08677-9 2024

  13. [13]

    Physics-based modeling of metal additive manufacturing processes: a review

    Xu, S., Younes Araghi, M., and Su, Y . “Physics-based modeling of metal additive manufacturing processes: a review.” The International Journal of Advanced Manufacturing Technology Vo l . 1 3 4 N o . 1 ( 2 0 2 4 ) : p p . 1–13. DOI 10.1007/s00170-024-14156-9

    work page doi:10.1007/s00170-024-14156-9

  14. [14]

    Identification and Interpretation of Melt Pool Shapes in Laser Powder Bed Fusion with Machine Learning

    Sato, M. M., Wong, V . W. H., Yeung, H., Witherell, P., and Law, K. H. “Identification and Interpretation of Melt Pool Shapes in Laser Powder Bed Fusion with Machine Learning.” Smart and Sustainable Manufacturing Systems Vo l . 8 N o . 1 ( 2 0 2 4 ) : p p . 1–23. DOI 10.1520/SSMS20230035

    work page doi:10.1520/ssms20230035

  15. [15]

    Deep learning based porosity prediction for additively manufactured laser powder-bed fusion parts

    Mohammed, A. S., Almutahhar, M., Sattar, K., Alhajeri, A., Nazir, A., et al. “Deep learning based porosity prediction for additively manufactured laser powder-bed fusion parts.” Journal of Materials Research and Technology Vo l . 27 (2023): pp. 7330–7335. DOI 10.1016/j.jmrt.2023.11.130

    work page doi:10.1016/j.jmrt.2023.11.130 2023

  16. [16]

    Machine learning assisted prediction and optimization of mechanical properties for laser powder bed fusion of Ti6Al4V alloy

    Cao, Y ., Chen, C., Xu, S., Zhao, R., Guo, K., et al. “Machine learning assisted prediction and optimization of mechanical properties for laser powder bed fusion of Ti6Al4V alloy.” Additive Manufacturing Vo l . 9 1 ( 2 0 2 4 ) : pp. 104341. DOI 10.1016/j.addma.2024.104341

    work page doi:10.1016/j.addma.2024.104341 2024

  17. [17]

    Nature , author =

    Biamonte, J., Wittek, P., Pancotti, N., Rebentrost, P., Wiebe, N., et al. “Quantum machine learning.” Nature Vo l . 549 No. 7671 (2017): pp. 195–202. DOI 10.1038/nature23474

    work page doi:10.1038/nature23474 2017

  18. [18]

    Quantum Machine Learning in Feature Hilbert Spaces

    Schuld, M. and Killoran, N. “Quantum Machine Learning in Feature Hilbert Spaces.” Physical Review Letters Vo l . 122 No. 4 (2019): pp. 040504. DOI 10.1103/PhysRevLett.122.040504

    work page doi:10.1103/physrevlett.122.040504 2019

  19. [19]

    Nature Reviews Physics , author =

    Cerezo, M., Arrasmith, A., Babbush, R., Benjamin, S. C., Endo, S., et al. “Variational quantum algorithms.” Nature Reviews Physics Vo l . 3 N o . 9 ( 2 0 2 1 ) : p p . 6 2 5–644. DOI 10.1038/s42254-021-00348-9

    work page doi:10.1038/s42254-021-00348-9

  20. [20]

    Variational Quantum Algorithms for Computational Fluid Dynamics

    Jaksch, D., Givi, P., Daley, A. J., and Rung, T. “Variational Quantum Algorithms for Computational Fluid Dynamics.” AIAA Journal Vo l . 6 1 N o . 5 ( 2 0 2 3 ) : p p . 1 8 8 5–1894. DOI 10.2514/1.J062426

    work page doi:10.2514/1.j062426

  21. [21]

    Variational quantum reinforcement learning via evolutionary optimization

    Chen, S. Y .-C., Huang, C.-M., Hsing, C.-W., Goan, H.-S., and Kao, Y .-J. “Variational quantum reinforcement learning via evolutionary optimization.” Machine Learning: Science and Technology Vo l . 3 N o . 1 ( 2 0 2 2 ) : p p . 015025. DOI 10.1088/2632-2153/ac4559

    work page doi:10.1088/2632-2153/ac4559

  22. [22]

    Barren plateaus in variational quantum computing,

    Larocca, M., Thanasilp, S., Wang, S., Sharma, K., Biamonte, J., et al. “Barren plateaus in variational quantum computing.” Nature Reviews Physics Vo l . 7 N o . 4 ( 2 0 2 5 ) : pp. 174–189. DOI 10.1038/s42254-025-00813-9

    work page doi:10.1038/s42254-025-00813-9

  23. [23]

    Enhanced fill probability estimates in institutional algorithmic bond trading using statistical learning algorithms with quantum computers

    Ciceri, A., Cottrell, A., Freeland, J., Fry, D., Hirai, H., et al. “Enhanced fill probability estimates in institutional algorithmic bond trading using statistical learning algorithms with quantum computers.” (2025): DOI 10.48550/arXiv.2509.17715

    work page doi:10.48550/arxiv.2509.17715 2025

  24. [24]

    Quantum machine learning for additive manufacturing process monitoring

    Choi, E., Sul, J., Kim, J. E., Hong, S., Gonzalez, B. I., et al. “Quantum machine learning for additive manufacturing process monitoring.” Manufacturing Letters Vo l . 4 1 (2024): pp. 1415–1422. DOI 10.1016/j.mfglet.2024.09.168

    work page doi:10.1016/j.mfglet.2024.09.168 2024

  25. [25]

    Zur Theorie des Ferromagnetismus

    Heisenberg, W. “Zur Theorie des Ferromagnetismus.” Zeitschrift für Physik Vo l . 4 9 N o . 9 ( 1 9 2 8 ) : p p . 6 1 9–636. DOI 10.1007/BF01328601

    work page doi:10.1007/bf01328601

  26. [26]

    Generalized trotter’s formula and systematic approximants of exponential operators and inner derivations with applications to many-body problems,

    Suzuki, M. “Generalized Trotter’s formula and systematic approximants of exponential operators and inner derivations with applications to many-body problems.” Communications in Mathematical Physics Vo l . 5 1 N o . 2 (1976): pp. 183–190. DOI 10.1007/BF01609348

    work page doi:10.1007/bf01609348 1976

  27. [27]

    Web-Scale K-Means Clustering

    Sculley, D. “Web-Scale K-Means Clustering.” Proceedings of the 19th international conference on World wide web (2010): pp. 1177–1178. DOI 10.1145/1772690.1772862

    work page doi:10.1145/1772690.1772862 2010

  28. [28]

    Process Monitoring Dataset from the Additive Manufacturing Metrology Testbed (AMMT): ‘Three-Dimensional Scan Strategies.’

    Lane, B. M. and Yeung, H. “Process Monitoring Dataset from the Additive Manufacturing Metrology Testbed (AMMT): ‘Three-Dimensional Scan Strategies.’” NIST Journal of Research (2019): DOI 10.6028/jres.124.033

    work page doi:10.6028/jres.124.033 2019

  29. [29]

    Quantum computing with Qiskit

    Javadi-Abhari, A., Treinish, M., Krsulich, K., Wood, C. J., Lishman, J., et al. “Quantum computing with Qiskit.” arXiv (2024): DOI 10.48550/arXiv.2405.08810

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2405.08810 2024

  30. [30]

    Efficient Classical Simulation of Slightly Entangled Quantum Computations

    Vidal, G. “Efficient Classical Simulation of Slightly Entangled Quantum Computations.” Physical Review Letters Vo l . 9 1 N o . 1 4 ( 2 0 0 3 ) : p p . 1 4 7 9 0 2 . D O I 10.1103/PhysRevLett.91.147902

    work page doi:10.1103/physrevlett.91.147902

  31. [31]

    Noise tailoring for scalable quantum computation via randomized compiling,

    Wallman, J. J. and Emerson, J. “Noise tailoring for scalable quantum computation via randomized compiling.” Physical Review A Vo l . 9 4 N o . 5 ( 2 0 1 6 ) : p p . 0 5 2 3 2 5 . D O I 10.1103/PhysRevA.94.052325

    work page doi:10.1103/physreva.94.052325

  32. [32]

    Machine Learning45(1), 5–32 (2001)

    Breiman, L. “Random Forests.” Machine Learning Vo l . 4 5 No. 1 (2001): pp. 5–32. DOI 10.1023/A:1010933404324

    work page doi:10.1023/a:1010933404324 2001