Pith. sign in

REVIEW 4 major objections 5 minor 58 references

Physics-informed machine learning surrogate for scalable simulation of thermal histories during wire-arc directed energy deposition

T0 review · 4 major / 5 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 A physics-informed neural network trained solely on the heat equation and boundary conditions reproduces FEM thermal histories for wire-arc DED with a relative L2 error of 7.267e-2 and up to 98.6% reduction in reported compute time.

desk verdict A useful data-free PINN benchmark for wire-arc DED, but the 98.6% speed-up is an artifact of comparing against the finest FEM mesh; the paper needs a mesh-convergence study and matched-baseline timing before the efficiency claim can be taken at face value. read the letter →

arxiv 2507.09591 v1 pith:ATAZMIBD submitted 2025-07-13 cs.CE

Michael Ryan , Mohammad Hassan Baqershahi , Hessamoddin Moshayedi , Elyas Ghafoori This is my paper
classification cs.CE
keywords physics-informedneuralnetworkswire-arcdirectedenergydepositionthermalhistorysimulationSobol'sequencescollocationpointsamplingdata-freesurrogateGoldakheatsourcefiniteelementmethodbenchmark
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

The paper tries to show that a physics-informed neural network (PINN), trained with no experimental or FEM data, can serve as a fast surrogate for thermal-history simulation in wire-arc directed energy deposition. On a single-layer 40 mm × 6 mm × 4 mm block, the network reproduces the temperature field of a fine finite-element benchmark with relative L2 error $7.267\times10^{-2}$ while cutting the reported compute time by up to 98.6%. The key enabler is a sampling strategy based on Sobol' sequences, which covers the space-time domain with far fewer collocation points than uniform grids. Because the trained network can be evaluated at any point in space and time, it offers \u201csuper-resolution\u201d that FEM would need a finer mesh to match. If the result holds at larger scales, repeated thermal simulations for process planning could move from days to under an hour.

What carries the argument

The carrying mechanism is a feed-forward MLP that outputs temperature $\hat{u}(x,y,z,t)$ and is trained by minimizing normalized residuals of the heat-conduction PDE (with the Goldak double-ellipsoid moving heat source $Q_{\text{goldak}}$), the Robin convection/radiation boundary condition, and the Dirichlet bottom condition, all computed by automatic differentiation. The Goldak source is a Gaussian-shaped, double-ellipsoidal volumetric heat flux that is standard for welding and DED. Three design choices carry the scalability argument: hard enforcement of the initial condition, which removes one loss term; a self-adaptive weighting scheme that balances boundary and PDE residuals during training; and Sobol' low-discrepancy quasi-random collocation points, which cover the four-dimensional space-time domain with far fewer samples than uniform grids and change from time step to time step. The Sobol' sequence, a low-discrepancy point set designed to cover a unit hypercube uniformly, is the central scalability component: it keeps the collocation-point count and per-epoch cost low as domain size and, later, parameter dimensionality grow.

What would settle it

Run a mesh- and time-step-convergence study on the same 40 mm × 6 mm × 4 mm block, comparing peak temperatures at the five probe points across FEM-CC, FEM-FC, and FEM-FF; if FEM-CC agrees with FEM-FF within engineering tolerance, then the 98.6% speed-up over FEM-FF is not the relevant comparison and the PINN is slower than FEM-CC. Alternatively, compare PINN predictions to thermocouple measurements on a deposited wall: a relative error much larger than 7.3% would show the PDE-only training misses physics that the FEM benchmark already approximates.

Watch

Extended reading notes

Core claim

The central claim is that a fully connected feed-forward neural network, constrained only by the transient heat-conduction equation with a Goldak double-ellipsoid volumetric heat source and by convection/radiation (Robin) plus Dirichlet boundary conditions, can learn the thermal history of a wire-arc DED deposition without any external training data. With hard enforcement of the uniform ambient initial condition, self-adaptive loss weighting, Sobol'-sampled collocation points, and an Adam-then-L-BFGS schedule, the trained network matches the finest FEM benchmark (0.1 mm mesh, 5 ms steps) to a relative L2 error of $7.267\times10^{-2}$ and reproduces peak temperatures along the heat-source path to an average difference of about 85 \u00b0C (6.5%). The authors conclude that a strictly PDE-driven PINN is a justifiable substitute for FEM when fine-resolution thermal histories are needed, and that folding even one FEM run into training would erase the computational advantage for a single use case.

Load-bearing premise

The load-bearing premise is that the finest FEM model (0.1 mm mesh, 5 ms time steps, about two days on a laptop CPU) is the required fidelity baseline, so the 98.6% speed-up is the right comparison; the paper also concedes in Section 3.4 that a strictly PDE-driven approach has inherent accuracy limits, and if the coarser 0.5 mm / 20 ms FEM (about 1.5 minutes) is already accurate enough, the PINN's 45-minute training is slower than FEM, a possibility no mesh-convergence study rules out.

Editorial extensions

If this is right

  • For applications that genuinely require FEM-FF-level resolution (0.1 mm mesh, 5 ms steps), a single PINN training run replaces a multi-day FEM run, and additional queries at new points or times cost only a forward pass.
  • Because the PINN is trained from the governing equations alone, it sidesteps the shortage of large experimental datasets for wire-arc DED; the same workflow could be applied to a new geometry or material by redefining the residual and initial/boundary conditions.
  • Repeated what-if simulations for heat management, process-parameter choice, or printing strategy become feasible: the network can be evaluated thousands of times at negligible marginal cost.
  • Sobol' collocation reduces per-epoch cost and sample count relative to uniform multi-grid sampling, and the advantage grows with domain size and parameterized inputs, pointing toward parametric PINNs trained over ranges of process parameters.

Reading between the lines

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

  • If a mesh-convergence study shows the coarse FEM already captures the thermal history, the honest comparison is PINN versus FEM-CC; then PINN wins only when the same network is reused for many different queries or process parameters, not for a single thermal history.
  • The 'super-resolution' property could be exploited the other way: instead of matching a fine FEM mesh, use the PINN to interpolate sparse thermocouple data onto any spatial-temporal grid, effectively turning the surrogate into a measurement-enhancement tool.
  • The single-layer, constant-geometry test leaves open how the method behaves with layer-by-layer deposition, where the heat source moves over newly added material and boundary conditions change; a natural next check is a two- or three-layer wall with the same Sobol' sampling and a comparison to FEM or experiment.
  • The 45-minute training cost is paid once; amortized over many simulations, even a slower-than-coarse-FEM training time can be economically justified, an argument the paper makes only implicitly through its parametric-PINN outlook.
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

4 major / 5 minor

Summary. The paper presents a physics-informed neural network (PINN) surrogate for predicting thermal histories in wire-arc directed energy deposition (DED). The PINN is trained using only the heat equation, boundary/initial conditions, and a Goldak heat source, without external FEM or experimental data. The authors introduce Sobol' sequence-based collocation point sampling, a power-law z-warp, hard initial conditions, and self-adaptive loss weighting. They compare the trained PINN against three FEM simulations (FEM-CC, FEM-FC, FEM-FF) with increasing spatial and temporal resolution, reporting a relative L2 error of 7.267e-2 and computational time reductions of up to 98.6% relative to the finest FEM model. The paper also discusses scalability and future directions for large-scale DED simulation.

Significance. If the reported efficiency gains hold, the work would be a useful step toward making PINN surrogates practical for repeated thermal simulation of large-scale DED components, where FEM runs are expensive. The claimed novelty lies in data-free training combined with efficient quasi-Monte Carlo sampling. The manuscript gives a clear description of the network architecture, loss weighting, sampling strategy, and training schedule, which facilitates reproducibility. The main limitations are that the central efficiency claim is not yet adequately supported because the comparison baseline and hardware settings are not justified, and the single-run error estimate does not demonstrate robustness.

major comments (4)
  1. [Section 3.3, Table 4] The headline efficiency claim of up to 98.6% reduction is computed against FEM-FF (2d 6h), while the coarsest FEM-CC baseline runs in 1m33s and is approximately 29 times faster than the PINN training time (45m40s). The paper's only justification for preferring FEM-FF as the comparison baseline is that it 'most closely resembles the discretization on which the PINN was trained' (Section 3.3). This is a resolution-matching argument, not an accuracy argument. The authors need to provide a mesh-convergence study, reporting L2 errors and thermal-history differences between FEM-CC, FEM-FC, and FEM-FF, and then justify which FEM resolution is sufficient for engineering accuracy. Without this, the efficiency conclusion reverses if FEM-CC is adequate.
  2. [Section 3.3, Table 4] The timing comparison mixes hardware and workflows: PINN training was performed on an NVIDIA RTX 3070Ti GPU, while FEM runs used four CPU cores of a laptop. The PINN time is a one-time training cost, and the paper does not quantify inference time for new queries, storage cost, or the break-even number of repeated simulations. For a single thermal-history prediction, the FEM-CC run is already faster than PINN training. To make the efficiency claim meaningful, the authors should specify the total cost model (training plus inference) and state for how many parameter variations or repeated simulations the PINN becomes cheaper.
  3. [Section 3.3, L2 error] The reported relative L2 error of 7.267e-2 is a single value with no information about variation across random seeds, network initializations, or Sobol' sequence draws, and no spatial or temporal decomposition of the error. Given the stochastic Adam optimizer and the sensitivity of PINN training, one training run does not establish the robustness of the error. The manuscript should report the mean and standard deviation over at least several independent training runs, together with the precise definition of the error norm over the spatiotemporal domain.
  4. [Section 3.3, super-resolution] The 'super-resolution' property is invoked as an advantage, but it is not demonstrated. The PINN was trained with time discretized at 5 ms and the FEM-FF baseline also uses 5 ms temporal and 0.1 mm spatial resolution, so the comparison does not show prediction beyond the discretization of the reference. To substantiate the super-resolution claim, the authors should evaluate the trained PINN at spatial and temporal resolutions finer than any FEM run (for example, 2.5 ms or 0.05 mm) and compare against an appropriate refined reference or interpolation, or clearly define what 'super-resolution' means in this context.
minor comments (5)
  1. [Table 1] The unit for power P is given as 'g mm/s2', which is not dimensionally consistent with the Goldak source term in Eq. (2). Please verify the unit and use standard SI or consistent CGS units throughout.
  2. [Section 2.2, Eq. (10)] The summation index i and the sample count N are not defined for the gradient-based weight update; please specify that N corresponds to the number of collocation points in the respective loss term.
  3. [Figure 7 caption] The phrase 'collapsed time dimension' is unclear; please add a sentence explaining that the points are projected onto the spatial domain.
  4. [Section 3.2.1] The statement that Sobol' sampling gives 'approximately half of the necessary computation time per epoch' lacks quantitative support; report the measured epoch times for uniform and Sobol' sampling.
  5. [Section 4] In the Conclusion, the reduction for FEM-FC is stated as 62.5%, whereas Table 4 implies 62.4%; please make the numbers consistent.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the PINN is trained purely on PDE and boundary residuals without FEM data in the loss, and the reported speed-up is a baseline-selection concern rather than a definitional reduction.

full rationale

The PINN is trained entirely on the governing-equation residuals and boundary-condition residuals, with no FEM output used in the loss function; the relative L2 error of 7.267e-2 is a post-hoc comparison against the FEM benchmark. Since the PINN and FEM solve the same heat-conduction equation with the same Goldak heat source, material properties, and boundary conditions, agreement is expected when both methods converge, making the comparison a consistency check rather than a fitted prediction. The headline computational saving is reported only against the finest FEM-FF baseline (0.1 mm spatial, 5 ms temporal), and the paper does not provide a mesh-convergence study demonstrating that the faster FEM-CC model is inadequate; this is a weakness in the fairness and generality of the efficiency claim, but it is not circular because the PINN does not inherit its result from the FEM or from a fitted parameter. The one self-citation, referencing the authors' prior experimental validation of the FEM model [10], supports the physical plausibility of the FEM benchmark but is externally falsifiable and does not enter the PINN training; it is therefore not load-bearing circularity. No step in the derivation reduces, by construction or by self-citation, to the paper's own inputs.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The central claim rests on a physical model (heat equation plus Goldak source) taken as given, on hyperparameters selected by tuning, and on the assumption that the FEM-FF result is the right accuracy baseline. No new physical entities are introduced.

free parameters (5)
  • collocation point counts = 185,669 boundary; 112,635 domain; 3,509 initial
    Number of Sobol points chosen roughly as a fraction of the estimated points required by a uniform grid (Section 3.1), not derived from accuracy targets.
  • power-law z-warp exponent = not specified
    A power-law z-warp was applied to shift domain points toward the top surface (Section 3.1); the exponent is a hand-chosen tuning parameter that affects where collocation points concentrate.
  • network hyperparameters = depth 4, width 64, Adam lr 0.001, L-BFGS iterations 50, epochs 14850
    Selected through substantial parameter analysis (Section 3.2.2); these control training cost and accuracy.
  • GELU Kaiming gain factor = 1.48
    Modified gain factor for GELU activation (Section 2.3), an empirical initialization choice.
  • loss balancing alpha = 0.9, update every 1000 epochs
    Self-adaptive weight update follows Wang et al. with alpha chosen per their recommendations (Section 2.2); affects convergence.
assumptions (6)
  • domain assumption Transient heat conduction PDE (Eq. 1) with constant material properties describes the thermal process.
    Assumed without justification for the temperature range involved; the FEM benchmark uses the same model.
  • domain assumption Goldak double-ellipsoid heat source (Eq. 2) with literature parameters represents the arc heat input.
    Parameters adopted from Liang et al. [42]; no calibration for this specific wire-arc setup.
  • domain assumption Convection, radiation, and Dirichlet boundary conditions (Eqs. 3-6) with constant coefficients represent the surface heat loss.
    h and epsilon fixed; emissivity 0.2 is low for metal surfaces, and temperature-dependent effects are ignored.
  • domain assumption FEM solutions (Abaqus) with DC3D8 elements are a valid reference for the PINN comparison.
    The finest FEM (0.1 mm mesh) is treated as ground truth without an independent convergence or experimental check in this paper.
  • standard math Sobol sequences provide better sampling coverage than uniform grids for PINN training.
    Low-discrepancy/QMC theory (Niederreiter [49]); this is a standard result, though its benefit for this specific PINN loss landscape is empirical.
  • standard math GELU activation is C2-smooth, enabling second-order automatic differentiation for the PDE residual.
    Standard property of GELU used for the Laplacian; the requirement is satisfied by the chosen activation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Physics-informed machine learning surrogate for scalable simulation of thermal histories during wire-arc directed energy deposition." pith.science (2026). https://pith.science/paper/ATAZMIBD

@misc{pith2026250709591,
  author       = {Pith},
  title        = {Pith review of: Physics-informed machine learning surrogate for scalable simulation of thermal histories during wire-arc directed energy deposition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ATAZMIBD}},
  note         = {Machine review of arXiv:2507.09591}
}
read the original abstract

Wire-arc directed energy deposition (DED) has emerged as a promising additive manufacturing (AM) technology for large-scale structural engineering applications. However, the complex thermal dynamics inherent to the process present challenges in ensuring structural integrity and mechanical properties of fabricated thick walls and plates. While finite element method (FEM) simulations have been conventionally employed to predict thermal history during deposition, their computational demand remains prohibitively high for actual large-scale applications. Given the necessity of multiple repetitive simulations for heat management and the determination of an optimal printing strategy, FEM simulation quickly becomes entirely infeasible. Instead, advancements have been made in using trained neural networks as surrogate models for rapid prediction. However, traditional data-driven approaches necessitate large amounts of relevant and verifiable external data, during the training and validation of the neural network. Regarding large-scale wire-arc DED, none of these data sources are readily available in quantities sufficient for an accurate surrogate. The introduction of physics-informed neural networks (PINNs) has opened up an alternative simulation strategy by leveraging the existing physical knowledge of the phenomena with advanced machine learning methods. Despite their theoretical advantages, PINNs have seen limited application in the context of large-scale wire-arc DED for structural engineering. This study investigates the scalability of PINNs, focusing on efficient collocation points sampling, a critical factor controlling both the training time and model performance. Results show PINNs can reduce computational time and effort by up to 98.6%, while maintaining the desired accuracy and offering "super-resolution". Future directions for enhancing PINN performance in metal AM are discussed.

Figures

Figures reproduced from arXiv: 2507.09591 by the authors.

Figure 1
Figure 1. Schematic representation of wire-arc DED Wire-arc DED has already been used for strengthening of conventional profiles [8–11] and for manufacturing of structural components such as columns [12], beams [13], and nodes [14, 15], as well as entire structures [16–18]. What distinguishes the application of wire-arc DED in structural engineering from other fields is the scale of components that can be up to meters, as sho… view at source ↗
Figure 2
Figure 2. MX3D: Stainless Steel Additive-Manufactured Bridge - made through wire-arc DED - Amsterdam, NL While opening up space for unprecedented design freedom, wire-arc DED also raises several open questions re￾garding mechanical properties, the structural reliability, and certification, among others [19]. An intrinsic complexity of the wire-arc DED technique lies in its unique temperature history, characterized by high the… view at source ↗
Figure 3
Figure 3. To simulate movement, the equation is modified to include explicit time dependence. For our purposes, the [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Network Topology- 4 Dense Layers with 64 Neurons 2.4. Finite Element Modelling The FEM numerical simulation was carried out using Abaqus software 2023 [40], purely as a benchmark analysis independent of the training process. An implicit time integration method with New…
Figure 5
Figure 5. Figure 5: FEM model and mesh configuration with a mesh size of 0.5 mm A transient heat transfer simulation was conducted using DC3D8 elements to model the temperature evolution during deposition. Two models were prepared and analyzed, featuring mesh sizes of 0.5 mm ( [PITH_FULL…
Figure 6
Figure 6. Figure 6: Schematic representation of the simplified test scenario 3. Results and Discussion 3.1. Problem Description To evaluate the described methodology, a simplified test scenario was developed. The object domain is a bare plate with dimensions 40 mm × 6 mm × 4 mm. A Goldak …
Figure 7
Figure 7. Figure 7: Example collocation point sampling comparison with collapsed time dimension more than double that of Figure 7b. According to preliminary testing on current hardware, the expected computation time is linearly related to the sample size, resulting in approximately half o…
Figure 8
Figure 8. Figure 8: Sobol’ sequence sampled collocation points with collapsed time dimension amplified. The maximum temperature has been capped at 1400◦C to highlight the melt pool. The relative L 2 error, calculated across the entire domain, including inter-melt pool regions, was 7.267 ×…
Figure 9
Figure 9. Figure 9: Comparison of the temperature field between FEM-FC and trained PINN model on the top surface (z=4 mm) 0 10 20 30 40 x [mm] 0 4 z [mm] 0 10 20 30 40 x [mm] 0 10 20 30 40 x [mm] 25 254 483 712 942 1171 1400 PINN - Temperature [ ±C] 0 10 20 30 40 x [mm] 0 4 z [mm] 0 10 20…
Figure 10
Figure 10. Figure 10: Comparison of the temperature field between FEM and trained PINN model on the y-midsection surface (y=3 mm) 11 [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]
Figure 11
Figure 11. Figure 11: Thermal history comparison of representative points on the top-surface (z=4 mm) 0 5 10 15 20 25 30 x [mm] 0 500 1000 1500 2000 T e m p e r a t u r e [ ±C] t=0.60s (PINN) t=0.60s (FEM-FF) t=0.60s (FEM-FC) t=1.60s (PINN) t=1.60s (FEM-FF) t=1.60s (FEM-FC) t=2.60s (PINN) …
Figure 12
Figure 12. Figure 12: Thermal comparison between FEM-FC, FEM-FF, and PINN at t = 0.6 s 1.6 s and 2.6 s 12 [PITH_FULL_IMAGE:figures/full_fig_p012_12.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

58 extracted references · 27 canonical work pages

  1. [10]

    Dahaghin, M

    H. Dahaghin, M. Motavalli, H. Moshayedi, S. Zahrai, E. Ghafoori, Wire and arc additive manufacturing for strengthening of metallic components, Thin-Walled Structures 203 (2024) 112074.doi:10.1016/j.tws. 2024.112074

    work page arXiv 2024

  2. [1]

    Buchanan, L

    C. Buchanan, L. Gardner, Metal 3d printing in construction: A review of methods, research, applications, op- portunities and challenges, Engineering Structures 180 (2019) 332–348.doi:10.1016/j.engstruct.2018. 11.045

    work page doi:10.1016/j.engstruct.2018 2019

  3. [2]

    S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal, P. Colegrove, Wire+arc additive manufacturing, Materials science and technology 32 (7) (2016) 641–647.doi:10.1179/1743284715Y.0000000073

    doi:10.1179/1743284715y.0000000073 2016

  4. [3]

    M. H. Baqershahi, C. Ayas, E. Ghafoori, Design optimisation for hybrid metal additive manufacturing for sustainable construction, Engineering Structures 301 (2024) 117355.doi:10.1016/j.engstruct.2023. 117355

    work page doi:10.1016/j.engstruct.2023 2024

  5. [4]

    M. H. Baqershahi, C. Ayas, E. Ghafoori, Topology optimisation for large-scale wire-arc directed energy de- position considering environmental impact and cost, Automation in Construction 177 (2025) 106313.doi: 10.1016/j.autcon.2025.106313

    arXiv 2025

  6. [5]

    Jafarabadi, I

    A. Jafarabadi, I. Ferretto, M. Mohri, C. Leinenbach, E. Ghafoori, 4d printing of recoverable buckling-induced ar- chitected iron-based shape memory alloys, Materials & Design 233 (2023) 112216.doi:10.1016/j.matdes. 2023.112216

    arXiv 2023

  7. [6]

    I. O. Felice, J. Shen, A. F. Barragan, I. A. Moura, B. Li, B. Wang, H. Khodaverdi, M. Mohri, N. Schell, E. Ghafoori, et al., Wire and arc additive manufacturing of fe-based shape memory alloys: microstructure, mechanical and functional behavior, Materials & Design 231 (2023) 112004.doi:10.1016/j.matdes.2023. 112004

    work page doi:10.1016/j.matdes.2023 2023

  8. [7]

    Lopes, D

    J. Lopes, D. Martins, K. Zhang, B. Li, B. Wang, X. Wang, N. Schell, E. Ghafoori, A. Baptista, J. Oliveira, Unveiling the microstructure evolution and mechanical properties in a gas tungsten arc-welded fe–mn–si–cr–ni shape memory alloy, Journal of Materials Science 59 (17) (2024) 7387–7408.doi:10.1007/s10853-024- 09606-4

    work page doi:10.1007/s10853-024- 2024

Show all 58 references
  1. [8]

    J. Yang, M. A. Wadee, L. Gardner, Strengthening of steel i-section beams by wire arc additive manufactur- ing—concept and experiments, Engineering Structures 322 (2025) 119113.doi:10.1016/j.engstruct. 2024.119113

    2025

  2. [9]

    X. Meng, L. Gardner, Hybrid construction featuring wire arc additive manufacturing: review, concepts, chal- lenges and opportunities, Engineering Structures 326 (2025) 119337.doi:https://doi.org/10.1016/j. engstruct.2024.119337

    2025

  3. [11]

    Ghafoori, H

    E. Ghafoori, H. Dahaghin, C. Diao, N. Pichler, L. Li, M. Mohri, J. Ding, S. Ganguly, S. Williams, Fatigue strengthening of damaged steel members using wire arc additive manufacturing, Engineering Structures 284 (2023) 115911.doi:10.1016/j.engstruct.2023.115911

    2023

  4. [12]

    Laghi, M

    V . Laghi, M. Palermo, G. Gasparini, T. Trombetti, Computational design and manufacturing of a half-scaled 3d-printed stainless steel diagrid column, Additive Manufacturing 36 (2020) 101505.doi:10.1016/j.addma. 2020.101505

    2020

  5. [13]

    J. Ye, P. Kyvelou, F. Gilardi, H. Lu, M. Gilbert, L. Gardner, An end-to-end framework for the additive manu- facture of optimized tubular structures, IEEE Access 9 (2021) 165476–165489.doi:10.1109/ACCESS.2021. 3132797. 14

    2021 doi

  6. [14]

    H. Wang, W. Du, Y . Zhao, Y . Wang, R. Hao, M. Yang, Joints for treelike column structures based on generative design and additive manufacturing, Journal of Constructional Steel Research 184 (2021) 106794.doi:10. 1016/j.jcsr.2021.106794

    2021

  7. [15]

    Snijder, L

    A. Snijder, L. van der Linden, C. Goulas, C. Louter, R. Nijsse, The glass swing: a vector active structure made of glass struts and 3d-printed steel nodes, Glass Structures & Engineering 5 (1) (2020) 99–116.doi: 10.1007/s40940-019-00110-9

    2020 doi

  8. [16]

    Gardner, P

    L. Gardner, P. Kyvelou, G. Herbert, C. Buchanan, Testing and initial verification of the world’s first metal 3d printed bridge, Journal of constructional steel research 172 (2020) 106233.doi:10.1016/j.jcsr.2020. 106233

    2020 doi

  9. [17]

    Huang, L

    C. Huang, L. Li, N. Pichler, E. Ghafoori, L. Susmel, L. Gardner, Fatigue testing and analysis of steel plates manufactured by wire-arc directed energy deposition, Additive Manufacturing 73 (2023) 103696.doi:https: //doi.org/10.1016/j.addma.2023.103696. URLhttps://www.sciencedi...

    2023

  10. [18]

    Huang, Y

    C. Huang, Y . Zheng, T. Chen, E. Ghafoori, L. Gardner, Fatigue crack growth behaviour of wire arc additively manufactured steels, International Journal of Fatigue 173 (2023) 107705.doi:https://doi.org/10.1016/ j.ijfatigue.2023.107705

    2023

  11. [19]

    Gardner, Metal additive manufacturing in structural engineering–review, advances, opportunities and outlook, Structures 47 (2023) 2178–2193.doi:10.1016/j.istruc.2022.12.039

    L. Gardner, Metal additive manufacturing in structural engineering–review, advances, opportunities and outlook, Structures 47 (2023) 2178–2193.doi:10.1016/j.istruc.2022.12.039

    2023 doi

  12. [20]

    T. A. Rodrigues, V . Duarte, J. A. Avila, T. G. Santos, R. Miranda, J. Oliveira, Wire and arc additive manufacturing of hsla steel: Effect of thermal cycles on microstructure and mechanical properties, Additive Manufacturing 27 (2019) 440–450.doi:10.1016/j.addma.2019.03.029

    2019 doi

  13. [21]

    W. J. Sames, F. List, S. Pannala, R. R. Dehoff, S. S. Babu, The metallurgy and processing science of metal additive manufacturing, International materials reviews 61 (5) (2016) 315–360.doi:10.1080/09506608. 2015.1116649

    2016

  14. [22]

    Mukherjee, H

    T. Mukherjee, H. Wei, A. De, T. DebRoy, Heat and fluid flow in additive manufacturing—part i: Modeling of powder bed fusion, Computational Materials Science 150 (2018) 304–313.doi:10.1016/j.commatsci. 2018.04.022

    2018 doi

  15. [23]

    S. Liao, T. Xue, J. Jeong, S. Webster, K. Ehmann, J. Cao, Hybrid thermal modeling of additive manufactur- ing processes using physics-informed neural networks for temperature prediction and parameter identification, Computational Mechanics 72 (3) (2023) 499–512.doi:10.1007/s00...

    2023 doi

  16. [24]

    X. Lu, X. Lin, M. Chiumenti, M. Cervera, Y . Hu, X. Ji, L. Ma, H. Yang, W. Huang, Residual stress and dis- tortion of rectangular and s-shaped ti-6al-4v parts by directed energy deposition: Modelling and experimental calibration, Additive Manufacturing 26 (2019) 166–179.doi:10...

    2019 doi

  17. [25]

    L. Yang, T. Özel, Physics-based simulation models for digital twin development in laser powder bed fusion, International Journal of Mechatronics and Manufacturing Systems 14 (2) (2021) 143–163.doi:10.1504/ IJMMS.2021.119155

    2021

  18. [26]

    Hosseini, P

    E. Hosseini, P. Scheel, O. Müller, 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:10.1016/j.cma....

    2023

  19. [27]

    G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, L. Yang, Physics-informed machine learning, Nature Reviews Physics 3 (6) (2021) 422–440.doi:10.1038/s42254-021-00314-5. 15

    2021 doi

  20. [28]

    Q. Zhu, Z. Liu, J. Yan, Machine learning for metal additive manufacturing: predicting temperature and melt pool fluid dynamics using physics-informed neural networks, Computational Mechanics 67 (2021) 619–635. doi:10.1007/s00466-020-01952-9

    2021 doi

  21. [29]

    J. Xie, Z. Chai, L. Xu, X. Ren, S. Liu, X. Chen, 3d temperature field prediction in direct energy deposition of metals using physics informed neural network, The International Journal of Advanced Manufacturing Technol- ogy 119 (5) (2022) 3449–3468.doi:10.1007/s00170-021-08542-w

    2022 doi

  22. [30]

    Uhrich, N

    B. Uhrich, N. Pfeifer, M. Schäfer, O. Theile, E. Rahm, Physics-informed deep learning to quantify anomalies for real-time fault mitigation in 3d printing, Applied Intelligence 54 (6) (2024) 4736–4755.doi:10.1007/s10489- 024-05402-4

    2024 doi

  23. [31]

    Sampaio, J

    R. Sampaio, J. Pragana, I. Bragança, C. Silva, C. Nielsen, P. Martins, Modelling of wire-arc additive manufac- turing – a review, Advances in Industrial and Manufacturing Engineering 6 (2023) 100121.doi:10.1016/j. aime.2023.100121

    2023

  24. [32]

    Goldak, A

    J. Goldak, A. Chakravarti, M. Bibby, A new finite element model for welding heat sources, Metallurgical Trans- actions B 15 (2) (1984) 299–305.doi:10.1007/BF02667333

    1984 doi

  25. [33]

    S. Wang, S. Sankaran, H. Wang, P. Perdikaris, An Expert’s Guide to Training Physics-informed Neural Net- works, preprint (2023).arXiv:2308.08468

    2023 arXiv

  26. [34]

    Hendrycks, K

    D. Hendrycks, K. Gimpel, Gaussian Error Linear Units (GELUs), preprint (Jun. 2023).arXiv:1606.08415

    2023 arXiv

  27. [35]

    K. He, X. Zhang, S. Ren, J. Sun, Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification, in: 2015 IEEE International Conference on Computer Vision (ICCV), IEEE, Santiago, Chile, 2015, pp. 1026–1034.doi:10.1109/ICCV.2015.123

    2015 doi

  28. [36]

    A. G. Baydin, B. A. Pearlmutter, Automatic Differentiation of Algorithms for Machine Learning, preprint (Apr. 2014).arXiv:1404.7456

    2014 arXiv

  29. [37]

    D. P. Kingma, J. Ba, Adam: A Method for Stochastic Optimization, preprint (Jan. 2017).arXiv:1412.6980

    2017 arXiv

  30. [38]

    D. C. Liu, J. Nocedal, On the limited memory method for large scale optimization, Mathematical Programming, Series B 45 (3) (1989) 503–528.doi:10.1007/BF01589116

    1989 doi

  31. [39]

    P. Roy, S. Castonguay, Exact Enforcement of Temporal Continuity in Sequential Physics-Informed Neural Net- works, preprint (2024).arXiv:2403.03223

    2024 arXiv

  32. [40]

    Smith, ABAQUS/Standard User’s Manual, Version 6.9, Dassault Systèmes Simulia Corp, United States, 2009

    M. Smith, ABAQUS/Standard User’s Manual, Version 6.9, Dassault Systèmes Simulia Corp, United States, 2009

    2009

  33. [41]

    Y . Ueda, K. Fukuda, K. Nakacho, S. Endo, A new measuring method of residual stresses with the aid of finite el- ement method and reliability of estimated values, Journal of the Society of Naval Architects of Japan 1975 (138) (1975) 499–507.doi:10.2534/jjasnaoe1968.1975.138_499

    1975

  34. [42]

    Liang, Q

    X. Liang, Q. Chen, L. Cheng, D. Hayduke, A. C. To, Modified inherent strain method for efficient prediction of residual deformation in direct metal laser sintered components, Computational Mechanics 64 (6) (2019) 1719– 1733.doi:10.1007/s00466-019-01748-6

    2019 doi

  35. [43]

    J. Ding, P. Colegrove, J. Mehnen, S. Ganguly, P. Sequeira Almeida, F. Wang, S. Williams, Thermo-mechanical analysis of wire and arc additive layer manufacturing process on large multi-layer parts, Computational Materials Science 50 (12) (2011) 3315–3322.doi:10.1016/j.commatsci...

    2011 doi

  36. [44]

    Paszke, S

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. Köpf, E. Yang, Z. DeVito, M. Raison, A. Tejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, S. Chintala, PyTorch: An Imperative Style, High-Perform...

    2019 arXiv

  37. [45]

    S. Li, G. Wang, Y . Di, L. Wang, H. Wang, Q. Zhou, A physics-informed neural network framework to predict 3D temperature field without labeled data in process of laser metal deposition, Engineering Applications of Artificial Intelligence 120 (2023) 105908.doi:10.1016/j.engappa...

    2023

  38. [46]

    I. Sobol’, On the distribution of points in a cube and the approximate evaluation of integrals, USSR Computa- tional Mathematics and Mathematical Physics 7 (4) (1967) 86–112.doi:10.1016/0041-5553(67)90144-9

    1967 doi

  39. [47]

    W. Cho, M. Jo, H. Lim, K. Lee, D. Lee, S. Hong, N. Park, Parameterized Physics-informed Neural Networks for Parameterized PDEs, preprint (Aug. 2024).arXiv:2408.09446

    2024 arXiv

  40. [48]

    K. Liu, K. Luo, Y . Cheng, A. Liu, H. Li, J. Fan, S. Balachandar, Parameterized physics-informed neural net- works (p-pinns) solution of uniform flow over an arbitrarily spinning spherical particle, International Journal of Multiphase Flow 180 (2024) 104937.doi:10.1016/j.ijmul...

    2024

  41. [49]

    Niederreiter, Low-discrepancy and low-dispersion sequences, Journal of Number Theory 30 (1) (1988) 51– 70.doi:10.1016/0022-314X(88)90025-X

    H. Niederreiter, Low-discrepancy and low-dispersion sequences, Journal of Number Theory 30 (1) (1988) 51– 70.doi:10.1016/0022-314X(88)90025-X

    1988 doi

  42. [50]

    M. D. McKay, R. J. Beckman, W. J. Conover, A comparison of three methods for selecting values of input variables in the analysis of output from a computer code, Technometrics 21 (2) (1979) 239–245.doi:10.1080/ 00401706.1979.10489755

    1979

  43. [51]

    J. H. Halton, On the efficiency of certain quasi-random sequences of points in evaluating multi-dimensional integrals, Numerische Mathematik 2 (1) (1960) 84–90.doi:10.1007/BF01386213

    1960 doi

  44. [52]

    S. Mo, D. Lu, X. Shi, G. Zhang, M. Ye, J. Wu, J. Wu, A Taylor Expansion-Based Adaptive Design Strategy for Global Surrogate Modeling With Applications in Groundwater Modeling, Water Resources Research 53 (12) (2017) 10802–10823.doi:10.1002/2017WR021622

    2017 doi

  45. [53]

    Crombecq, E

    K. Crombecq, E. Laermans, T. Dhaene, Efficient space-filling and non-collapsing sequential design strategies for simulation-based modeling, European Journal of Operational Research 214 (3) (2011) 683–696.doi:10. 1016/j.ejor.2011.05.032

    2011

  46. [54]

    B. Peng, A. Panesar, Multi-layer thermal simulation using physics-informed neural network, Additive Manufac- turing 95 (2024) 104498.doi:10.1016/j.addma.2024.104498

    2024

  47. [55]

    S. Wang, H. Wang, J. H. Seidman, P. Perdikaris, Random weight factorization improves the training of continu- ous neural representations, preprint (2022).arXiv:2210.01274

    2022 arXiv

  48. [56]

    Tancik, P

    M. Tancik, P. P. Srinivasan, B. Mildenhall, S. Fridovich-Keil, N. Raghavan, U. Singhal, R. Ramamoorthi, J. T. Barron, R. Ng, Fourier features let networks learn high frequency functions in low dimensional domains, preprint (2020).arXiv:2006.10739

    2020 arXiv

  49. [57]

    Loshchilov, F

    I. Loshchilov, F. Hutter, Sgdr: Stochastic gradient descent with warm restarts, preprint (2017).arXiv:1608. 03983

    2017

  50. [58]

    L. N. Smith, N. Topin, Super-convergence: very fast training of neural networks using large learning rates, in: T. Pham (Ed.), Artificial Intelligence and Machine Learning for Multi-Domain Operations Applications, V ol. 11006, International Society for Optics and Photonics, SP...

    2019 doi

Pith tools

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