Pith. sign in

REVIEW 4 major objections 5 minor 49 references

Numerical simulation of transient heat conduction with moving heat source using Physics Informed Neural Networks

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Window-by-window transfer learning lets a single physics-informed neural network reproduce a moving-source temperature field, matching a fine finite-element simulation at about half the cost.

desk verdict Workmanlike PINN application with an incremental training twist; the results are plausible but the paper's central accuracy claim rests on visuals and a single timing comparison, and Eq. (5) under-specifies the moving source. read the letter →

arxiv 2506.17726 v1 pith:IYCDRRNP submitted 2025-06-21 math.NA cs.LGcs.NA

Anirudh Kalyan , Sundararajan Natarajan This is my paper · ORCID
classification math.NAcs.LGcs.NA MSC 65M6068T0780A19
keywords physics-informedneuralnetworkstransientheatconductionmovingsourceGaussiantransferlearningtimesteppingmeshlessmethod
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 sets out to show that a physics-informed neural network (PINN) can compute the transient temperature field produced by a moving Gaussian heat source, a situation that normally requires very fine finite-element meshes near the source. To avoid scaling the network with simulation time, it splits the interval into short windows and trains the same network sequentially, using the final temperature field of one window as the initial condition of the next. Temperature fields from this scheme are compared with a finite-element solution on a mesh of size $h=0.05$ mm with time step $\Delta t=0.1$ s, and the paper reports good agreement along the heat-source path while taking about 3100 s of training time versus 6900 s for the finite-element run. If that comparison holds, the result is a meshless route to moving-source problems where adaptive refinement is difficult.

What carries the argument

The load-bearing mechanism is continuous time-stepping through transfer learning: the simulation interval is split into windows of length $\Delta t$, and one network is trained recursively, with the solution at the end of window $n$ becoming the initial condition for window $n+1$, so network size does not grow with simulation length. Because each window still represents time continuously, the scheme differs from Euler-style time discretization; querying the network at an arbitrary time only requires loading the weights for that window. The loss is a weighted sum of heat-equation residual, initial-condition, and boundary-condition terms, with the moving source entering through the Gaussian source term $f(\mathbf{x},t)=Q_0 e^{-r^2/r_0^2}$.

What would settle it

Run the same PINN against a converged finite-element reference on a finer mesh (for example $h=0.025$ mm) with a smaller time step, and evaluate the difference over the entire domain at the end of every time window; if the error grows with the number of windows, or if the peak-temperature error at late times exceeds the reported visual agreement, the claim of stable long-horizon transfer learning is falsified.

Watch

Extended reading notes

Core claim

The central claim is that the PINN framework solves transient heat conduction with a moving Gaussian source under mixed Dirichlet–Neumann boundary conditions, and that a sliding-window time-stepping scheme based on transfer learning makes this feasible without growing the network. The total time interval is divided into windows of length $\Delta t$; a single feed-forward network is trained on the first window, then warm-started on the next with the previous window's final temperature as the initial condition, so the same architecture covers the whole horizon. The paper compares the resulting temperature field along the source path with a finite-element solution using $h=0.05$ mm and $\Delta t=0.1$ s and finds good agreement at $t=2,4,6,8$ s, with the PINN training taking about 3100 s versus 6900 s for the reference FEM. No data beyond initial conditions, boundary conditions, and the governing equation are supplied.

Load-bearing premise

The load-bearing premise is that the finite-element reference at mesh size $h=0.05$ mm and time step $\Delta t=0.1$ s is converged, and that visual agreement along the heat-source path at the displayed times is enough to certify the PINN; the paper provides no study of error accumulation across the sequential warm-started windows.

Editorial extensions

If this is right

  • One fixed network can cover arbitrarily long simulation intervals because the parameter count does not grow with the number of windows; training time scales roughly linearly with the number of windows.
  • The moving-source test case shows agreement with the finite-element reference along the heat-source path while using about 3100 s versus 6900 s, so the meshless route can be competitive in cost at comparable resolution.
  • Only initial conditions, boundary conditions, and the governing PDE are supplied; no experimental data or precomputed snapshots are needed during training.
  • The velocity study shows that a faster-moving source lowers the peak temperature at the same spatial location, consistent with shorter exposure time of the material.

Reading between the lines

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

  • Chaining many more windows than the eight seconds shown would test whether warm-start drift accumulates; if end-of-window errors grow with window count, the long-horizon claim would need a correction term.
  • Because agreement is demonstrated along one line and at selected times, the error over the whole 20 mm by 10 mm domain remains unknown; a full-domain comparison would be the natural next check.
  • The same sliding-window schedule should carry over to temperature-dependent properties and coupled problems, but the paper's own observation that geometric rather than temporal complexity drives network size warns that 3-D extensions may need new scaling analysis.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 proposes a Physics Informed Neural Network (PINN) approach for two-dimensional transient heat conduction with a moving Gaussian heat source. To handle long time intervals, the authors introduce a sequential training strategy that divides the time domain into windows and reuses the same network across windows through transfer learning, taking the final state of one window as the initial condition for the next. The method is tested on a 20 mm × 10 mm domain with mixed Dirichlet–Neumann boundary conditions and compared visually against a finite element solution. The authors report good visual agreement, a reduction in wall-clock time (about 3100 s versus 6900 s for FEM), and expected behavior when the source velocity is varied. The manuscript claims that the proposed training regime enables the computation of large temporal intervals without increasing the network's complexity.

Significance. If fully supported, the paper would offer a practical meshless alternative for a class of manufacturing-related heat conduction problems and a clear way to extend PINNs to long time horizons without growing the network. The sequential transfer-learning idea is clearly described, the governing setup is a standard PINN formulation, and the authors provide a public GitHub repository with the implementation. These are strengths. However, the central numerical claim rests on visual comparison only, and the autoregressive time-stepping is not accompanied by an error-accumulation analysis. The paper's significance is therefore conditional on a quantitative validation that is currently missing.

major comments (4)
  1. [Section 3.1, Eq. (5)] The source term f(x,t) is printed as Q0 exp(-r^2/r0^2) with no definition of r in terms of a time-dependent source location. Since the text later states that the heat source travels along the path E-F at 2 mm/s, the actual source term used in the PINN residual of Eq. (6) must be something like Q0 exp(-||x - x_c(t)||^2 / r0^2), where x_c(t) follows the prescribed trajectory. As printed, Eq. (5) is time-independent and does not encode a moving source, which under-specifies the governing equation and the residual loss actually minimized.
  2. [Section 4, Figures 5 and 6] The agreement between the PINN and FEM is claimed on the basis of visual inspection only. No error norm (e.g., relative L2 error in temperature or maximum pointwise deviation) is reported, and no convergence study of the FEM reference (h = 0.05 mm, Δt = 0.1 s) is provided. Since FEM is treated as the benchmark and the PINN solution is autoregressive, the manuscript should include a refinement check for the FEM and a per-time-window error measure for the PINN. Without these, the central claims of 'good agreement' and 'similar accuracy at about half the time' are not quantitatively established.
  3. [Section 4, artifacting at t = 2 s] The paper concedes 'some artifacting at t = 2s' but does not quantify its magnitude or location, nor state whether it coincides with a phase boundary of the sequential training. Because each window re-initializes from the previous window's final state, an artifact at a window boundary can be carried forward and amplified. The authors should identify the artifact's magnitude, its spatial location, whether it occurs at a window transition, and whether it propagates into subsequent windows; this is directly relevant to the claimed long-time capability of the method.
  4. [Section 3.2 and Section 4] The sequential time increment Δt and the number of training phases are never specified for the numerical example. The text defines Δt as a hyperparameter but the actual value used, and the criterion for choosing it, are not given. This information is necessary for reproducibility and for interpreting the claim that training time scales linearly with the number of phases, since the number of phases is determined by Δt and the total time interval.
minor comments (5)
  1. [Figure captions and text] The text says Figure 5 shows temperature distribution at t = 2 s and 8 s, but the caption of Figure 5 says 'temperature evolution along the path E-F'; likewise, the text says Figure 6 shows two time instances, while the caption lists four (t = 2, 4, 6, 8 s). Please align captions with the figure content and the text.
  2. [Section 4, velocity study] The paragraph discussing the influence of heat source velocity refers to 'Figure 1', but Figure 1 in the manuscript shows the temperature output with and without time stepping at t = 8 s. The figure containing the velocity comparison appears to be misnumbered or missing.
  3. [Eq. (7)] The boundary condition for edge B-C is labeled Γ_AB, duplicating the label for edge A-B; this should be Γ_BC.
  4. [Section 3.1] Eq. (4) is written for a heterogeneous conducting body, while the numerical study and Table 1 use homogeneous material properties; please reconcile the wording.
  5. [References] Some references are incomplete or inconsistently formatted (e.g., [16] and [26] lack full bibliographic information), and the GitHub link appears only in the appendix rather than in the main text where the code is mentioned.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the PINN is trained on the PDE residual, initial and boundary conditions, with FEM used only as an external post-hoc benchmark.

full rationale

The paper's central claim is that a PINN with continuous time-stepping via transfer learning solves transient heat conduction with a moving Gaussian source, and that it agrees with an FEM reference. The network is trained against the strong-form PDE residual, the initial condition, and the boundary conditions through the loss in Eqs. (2), (3) and (6); no FEM output, measured data, or target solution is fed into the loss. The FEM solution is used only after training, as a comparison benchmark in Figures 5 and 6, so the agreement claim is an external check rather than a consequence of the training inputs. The sequential training procedure, in which the network output at one time window becomes the initial condition for the next window, is an algorithmic construction and does not presuppose the final solution. The only apparent self-citation, reference [4] listing one of the authors, appears in a background remark about partition-of-unity methods and is not load-bearing for the derivation. The paper's own caveat about 'some artifacting at t = 2s' and the absence of quantified error norms concern accuracy and robustness, not circularity. Likewise, the choices of network size, optimizer, and FEM mesh are calibration and verification choices rather than circular reductions. No step in the claimed derivation chain reduces by definition to its own inputs, so the appropriate circularity score is 0.

Assumptions & free parameters 8 free parameters · 4 assumptions · 0 invented entities

The central claim rests on manually chosen loss weights, unstated collocation counts, and an unstated time increment. The FEM reference and the no-drift assumption are the two least-supported pillars. No new physical entities are introduced.

free parameters (8)
  • Loss weight for initial condition lambda_ic = 250
    Chosen by hand in Section 4; no sensitivity study. Affects balance between PDE residual and initial condition terms.
  • Loss weight for boundary condition lambda_bc = 250
    Chosen by hand in Section 4; no sensitivity study. Affects how strongly Neumann and Dirichlet conditions are enforced.
  • Loss weight for PDE residual lambda_r = 1000
    Chosen by hand in Section 4; no sensitivity study. Dominates the total loss and therefore the trained solution.
  • Time increment dt for sequential phases = not stated
    Hyper-parameter controlling the temporal window length in the proposed time-stepping; value affects the number of phases and error accumulation.
  • Number and distribution of collocation points = not stated
    Critical for the PDE residual accuracy; the paper does not report counts or sampling strategy.
  • Learning rate for Adam optimizer = not stated
    Not reported in Section 4; directly affects convergence and the final approximation.
  • Network architecture depth and width = 9 layers, 128 neurons per layer
    Manual architectural choice in Section 4; no architecture search or justification.
  • Number of training epochs per phase = 20,000
    Fixed stopping criterion; not based on validation error or residual tolerance.
assumptions (4)
  • standard math Universal approximation theorem guarantees that a neural network can represent the solution.
    Invoked in Section 2 to justify NN representation, but it says nothing about whether gradient descent finds the right network or whether the residual loss reaches zero.
  • domain assumption The FEM solution with h=0.05 mm and dt=0.1 s is sufficiently converged to serve as reference truth.
    Section 4 states that the mesh size is 'found to be adequate' after progressive refinement, but no convergence table or error measures are shown.
  • ad hoc to paper Sequential warm-start training preserves the previous solution and does not accumulate error across many time windows.
    This is the core assumption behind the new training method; the paper does not analyze drift, catastrophic forgetting, or long-time error accumulation.
  • domain assumption The strong form of the heat equation with a Gaussian source is the correct physical model for this process.
    Section 3.1 adopts the standard heat conduction model; this is accepted background, though the moving-center term is not written explicitly.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Numerical simulation of transient heat conduction with moving heat source using Physics Informed Neural Networks." pith.science (2026). https://pith.science/paper/IYCDRRNP

@misc{pith2026250617726,
  author       = {Pith},
  title        = {Pith review of: Numerical simulation of transient heat conduction with moving heat source using Physics Informed Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IYCDRRNP}},
  note         = {Machine review of arXiv:2506.17726}
}
read the original abstract

In this paper, the physics informed neural networks (PINNs) is employed for the numerical simulation of heat transfer involving a moving source. To reduce the computational effort, a new training method is proposed that uses a continuous time-stepping through transfer learning. Within this, the time interval is divided into smaller intervals and a single network is initialized. On this single network each time interval is trained with the initial condition for (n+1)th as the solution obtained at nth time increment. Thus, this framework enables the computation of large temporal intervals without increasing the complexity of the network itself. The proposed framework is used to estimate the temperature distribution in a homogeneous medium with a moving heat source. The results from the proposed framework is compared with traditional finite element method and a good agreement is seen.

Figures

Figures reproduced from arXiv: 2506.17726 by the authors.

Figure 1
Figure 1. Temperature distribution output of PINN at [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Schematic representation of four layer neural network: [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Sequential Training Procedure LSTMs [16], the neural network is treated as a sliding window function. The network is trained for a range of time steps, and the output of the final time step of the network is taken as the initial condition for the next training phase. This autoregressive method allows for moderate function complexity while enabling a modular structure for solving large time frames. As a result, the m… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Schematic of geometry and boundary conditions [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Comparison of temperature evolution along the path [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Comparison of temperature evolution along the path [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 42 canonical work pages

  1. [3]

    Burgers' pinns with implicit euler transfer learning

    Vit´ oria Biesek and Pedro Henrique de Almeida Konzen. “Burgers’ pinns with implicit euler transfer learning”. In: arXiv preprint arXiv:2310.15343 (2023)

    work page Pith review arXiv 2023

  2. [1]

    TensorFlow: A system for large-scale machine learning

    Martin Abadi et al. “TensorFlow: A system for large-scale machine learning”. In: 12th USENIX Symposium on Operating Systems Design and Implementation (OSDI 16) . VMware, NetApp, Google Inc., Microsoft, Facebook. USENIX Association, 2016, pp. 265–283

    work page 2016

  3. [2]

    Automatic Differentiation in Machine Learning: A Survey

    Atılım G¨ unes Baydin et al. “Automatic Differentiation in Machine Learning: A Survey”. In: J. Mach. Learn. Res. 18.1 (Jan. 2017), 5595–5637. issn: 1532-4435. 10

    work page 2017

  4. [4]

    Partition of Unity Methods

    St´ ephane PA Bordas, Alexander Menk, and Sundararajan Natarajan. Partition of Unity Methods. Wiley, 2024

    work page 2024

  5. [5]

    Jax: Autograd and xla

    James Bradbury et al. “Jax: Autograd and xla”. In: Astrophysics Source Code Library (2021), ascl–2111

    2021

  6. [6]

    Physics-informed neural networks for heat transfer problems

    Shengze Cai et al. “Physics-informed neural networks for heat transfer problems”. In: Journal of Heat Transfer 143.6 (2021), p. 060801

    work page 2021

  7. [7]

    Physics-informed neural networks (PINNs) for fluid mechanics: A review

    Shengze Cai et al. “Physics-informed neural networks (PINNs) for fluid mechanics: A review”. In: Acta Mechanica Sinica 37.12 (2021), pp. 1727–1738

    work page 2021

  8. [8]

    Distributed learning machines for solving forward and inverse problems in partial differential equations

    Vikas Dwivedi, Nishant Parashar, and Balaji Srinivasan. “Distributed learning machines for solving forward and inverse problems in partial differential equations”. In: Neurocomputing 420 (2021), pp. 299–316

    work page 2021

Show all 49 references
  1. [9]

    Physics Informed Extreme Learning Machine (PIELM) – A rapid method for the numerical solution of partial differential equations

    Vikas Dwivedi and Balaji Srinivasan. “Physics Informed Extreme Learning Machine (PIELM) – A rapid method for the numerical solution of partial differential equations”. In: Neurocom- puting 391 (2020), pp. 96–118

    2020

  2. [10]

    Physics-informed kernel function neural networks for solving partial differential equations

    Zhuojia Fu, Wenzhi Xu, and Shuainan Liu. “Physics-informed kernel function neural networks for solving partial differential equations”. In: Neural Networks 172 (2024), p. 106098

    2024

  3. [11]

    Physical informed neural network for thermo-hydral analysis of fire-loaded concrete

    Zhiran Gao et al. “Physical informed neural network for thermo-hydral analysis of fire-loaded concrete”. In: Engineering Analysis with Boundary Elements 158 (2024), pp. 252–261

    2024

  4. [12]

    An isogeometric boundary element method for heat transfer problems of multiscale structures in electronic packaging with arbitrary heat sources

    Yanpeng Gong et al. “An isogeometric boundary element method for heat transfer problems of multiscale structures in electronic packaging with arbitrary heat sources”. In: Applied Mathematical Modelling 109 (2022), pp. 161–185

    2022

  5. [13]

    Transfer learning enhanced physics informed neural network for phase-field modeling of fracture

    Somdatta Goswami et al. “Transfer learning enhanced physics informed neural network for phase-field modeling of fracture”. In: Theoretical and Applied Fracture Mechanics 106 (2020), p. 102447

    2020

  6. [14]

    A deep learning framework for solution and discovery in solid me- chanics

    Ehsan Haghighat et al. “A deep learning framework for solution and discovery in solid me- chanics”. In: arXiv preprint arXiv:2003.02751 (2020)

    2020 arXiv

  7. [15]

    Deep learning in computational mechanics: a review

    Leon Herrmann and Stefan Kollmannsberger. “Deep learning in computational mechanics: a review”. In: Computational Mechanics 74 (2024), pp. 281–331

    2024

  8. [16]

    Long Short-term Memory

    S Hochreiter. “Long Short-term Memory”. In: Neural Computation MIT-Press (1997)

    1997

  9. [17]

    Multilayer feedforward networks are universal approximators

    Kurt Hornik, Maxwell Stinchcombe, and Halbert White. “Multilayer feedforward networks are universal approximators”. In: Neural Networks 2.5 (1989), pp. 359–366. issn: 0893-6080. doi: https://doi.org/10.1016/0893-6080(89)90020-8 . url: https://www.sciencedirect. com/science/art...

    1989

  10. [18]

    Solving heat conduction problems with a moving heat source in arc welding processes via an overlapping nodes scheme based on the improved element- free Galerkin method

    Juan C. ´Alvarez Hostos et al. “Solving heat conduction problems with a moving heat source in arc welding processes via an overlapping nodes scheme based on the improved element- free Galerkin method”. In: International Journal of Heat and Mass Transfer 192 (2022), p. 122940

    2022

  11. [19]

    hp-VPINNs: Variational physics-informed neural networks with domain decomposition

    Ehsan Kharazmi, Zhongqiang Zhang, and George Em Karniadakis. “hp-VPINNs: Variational physics-informed neural networks with domain decomposition”. In: Computer Methods in Applied Mechanics and Engineering 374 (2021), p. 113547. 11

    2021

  12. [20]

    Variational physics- informed neural networks for solving partial differential equations

    Ehsan Kharazmi, Zhongqiang Zhang, and George Em Karniadakis. “Variational physics- informed neural networks for solving partial differential equations”. In:arXiv preprint arXiv:1912.00873 (2019)

    2019 arXiv

  13. [21]

    Adaptive finite element solution for the heat conduction with a moving heat source

    Chi-Kyung Kim. “Adaptive finite element solution for the heat conduction with a moving heat source”. In: Journal of Mechanical Science and Technology 26 (2012), pp. 967–972

    2012

  14. [22]

    Temperature Field Due to a Moving Heat Source: A Moving Mesh Finite Element Analysis

    Z.-B Kuang and SN Atluri. “Temperature Field Due to a Moving Heat Source: A Moving Mesh Finite Element Analysis”. In: Journal of Applied Mechanics 52 (1985), pp. 274–280

    1985

  15. [23]

    Artificial neural networks for solving ordinary and partial differential equations

    I.E. Lagaris, A. Likas, and D.I. Fotiadis. “Artificial neural networks for solving ordinary and partial differential equations”. In: IEEE Transactions on Neural Networks 9.5 (1998), pp. 987–1000

    1998

  16. [24]

    Neural-network methods for boundary value problems with irregular boundaries

    I.E. Lagaris, A.C. Likas, and D.G. Papageorgiou. “Neural-network methods for boundary value problems with irregular boundaries”. In: IEEE Transactions on Neural Networks 11.5 (2000), pp. 1041–1049

    2000

  17. [25]

    Reynolds averaged turbulence mod- elling using deep neural networks with embedded invariance

    Julia Ling, Andrew Kurzawski, and Jeremy Templeton. “Reynolds averaged turbulence mod- elling using deep neural networks with embedded invariance”. In: Journal of Fluid Mechanics 807 (2016), 155–166

    2016

  18. [26]

    A review of automatic differentiation and its efficient implementa- tion

    Charles C Margossian. “A review of automatic differentiation and its efficient implementa- tion”. In: Wiley interdisciplinary reviews: data mining and knowledge discovery 9.4 (2019), e1305

    2019

  19. [27]

    An efficient boundary-type meshfree method for analysis of two-dimensional laser heating problems

    M. Mohammadi, M.R. Hematiyan, and Y.C. Shiah. “An efficient boundary-type meshfree method for analysis of two-dimensional laser heating problems”. In: Engineering Analysis with Boundary Elements 132 (2021), pp. 460–468

    2021

  20. [28]

    Numerical prediction of fluid flow and heat transfer in welding with a moving heat source

    K Mundra, T DebRoy, and K M Kelkar. “Numerical prediction of fluid flow and heat transfer in welding with a moving heat source”. In: Numerical Heat Transfer, Part A: Applications 29.2 (1996), pp. 115–129. doi: 10.1080/10407789608913782

    1996 doi

  21. [29]

    The partition-of-unity method for linear diffusion and convection problems: Accuracy, stabilization and multiscale interpretation

    EA Munts, SJ Hulshoff, and R De Borst. “The partition-of-unity method for linear diffusion and convection problems: Accuracy, stabilization and multiscale interpretation”. In: Int J Numer Meth Fluids 43 (2003), pp. 199–213

    2003

  22. [30]

    Parametric deep energy approach for elasticity accounting for strain gradient effects

    Vien Minh Nguyen-Thanh et al. “Parametric deep energy approach for elasticity accounting for strain gradient effects”. In: Computer Methods in Applied Mechanics and Engineering 386 (2021), p. 114096. issn: 0045-7825. doi: https://doi.org/10.1016/j.cma.2021.114096 . url: https:...

    2021

  23. [31]

    Generalized finite element analysis of three-dimensional heat transfer problems exhibiting sharp thermal gradients

    P O’Hara, C Duarte, and T Eason. “Generalized finite element analysis of three-dimensional heat transfer problems exhibiting sharp thermal gradients”. In: Comput Methods Appl Mech Eng 198 (2009), pp. 1857–1871

    2009

  24. [32]

    A paradigm for data-driven predictive modeling using field inversion and machine learning

    Eric J. Parish and Karthik Duraisamy. “A paradigm for data-driven predictive modeling using field inversion and machine learning”. In: Journal of Computational Physics 305 (2016), pp. 758–774

    2016

  25. [33]

    Automatic differentiation in pytorch

    Adam Paszke et al. “Automatic differentiation in pytorch”. In: (2017)

    2017

  26. [34]

    Adaptive physics-informed neural networks for dynamic coupled thermo- mechanical problems in large-size-ratio functionally graded materials

    Lin Qiu et al. “Adaptive physics-informed neural networks for dynamic coupled thermo- mechanical problems in large-size-ratio functionally graded materials”. In: Applied Mathe- matical Modelling 140 (2025), p. 115906. 12

    2025

  27. [35]

    Spectral integrated neural networks (SINNs) for solving forward and inverse dynamic problems

    Lin Qiu et al. “Spectral integrated neural networks (SINNs) for solving forward and inverse dynamic problems”. In: Neural Networks 180 (2024), p. 106756

    2024

  28. [36]

    Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial dif- ferential equations

    M. Raissi, P. Perdikaris, and G.E. Karniadakis. “Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial dif- ferential equations”. In: Journal of Computational Physics 378 (2019), pp. 686–707

    2019

  29. [37]

    SPINN: Sparse, Physics-based, and partially Interpretable Neural Networks for PDEs

    Amuthan A. Ramabathiran and Prabhu Ramachandran. “SPINN: Sparse, Physics-based, and partially Interpretable Neural Networks for PDEs”. In:Journal of Computational Physics 445 (2021), p. 110600. issn: 0021-9991. doi: https://doi.org/10.1016/j.jcp.2021.110600 . url: https://www....

    2021

  30. [38]

    Two-dimensional numerical simu- lation of heat transfer with moving heat source in welding using the Finite Pointset Method

    Edgar O. Res´ endiz-Flores and F´ elix R. Saucedo-Zendejo. “Two-dimensional numerical simu- lation of heat transfer with moving heat source in welding using the Finite Pointset Method”. In: International Journal of Heat and Mass Transfer 90 (2015), pp. 239–245. issn: 0017-

    2015

  31. [39]

    Mathematical theory of heat distribution during welding and cutting, aw s

    D Rosenthal. “Mathematical theory of heat distribution during welding and cutting, aw s”. In: Jourual, May (1941)

    1941

  32. [40]

    An energy approach to the solution of partial differential equations in computational mechanics via machine learning: Concepts, implementation and applications

    E. Samaniego et al. “An energy approach to the solution of partial differential equations in computational mechanics via machine learning: Concepts, implementation and applications”. In: Computer Methods in Applied Mechanics and Engineering 362 (2020), p. 112790. issn: 0045-78...

    2020

  33. [41]

    Advances in computational mod- eling for laser powder bed fusion additive manufacturing: A comprehensive review of finite element techniques and strategies

    Devashish Sarkar, Angshuman Kapil, and Abhay Sharma. “Advances in computational mod- eling for laser powder bed fusion additive manufacturing: A comprehensive review of finite element techniques and strategies”. In: Additive Manufacturing 85 (2024), p. 104157

    2024

  34. [42]

    On the performance of a Chimera-FEM implementation to treat moving heat sources and moving boundaries in time- dependent problems

    Bruno A Storti, Alejandro E Albanesi, and Ignacio Peralta. “On the performance of a Chimera-FEM implementation to treat moving heat sources and moving boundaries in time- dependent problems”. In: Finite Elements in Analysis and Design 208 (2022), p. 103789

    2022

  35. [43]

    An Extrinsic Approach Based on Physics- Informed Neural Networks for PDEs on Surfaces

    Zhuochao Tang, Zhuojia Fu, and Sergiy Reutskiy. “An Extrinsic Approach Based on Physics- Informed Neural Networks for PDEs on Surfaces”. In: Mathematics 22 (2022), p. 2861

    2022

  36. [44]

    Artificial intelligence for partial differential equations in computational me- chanics: A review

    Y Wang et al. “Artificial intelligence for partial differential equations in computational me- chanics: A review”. In: arXiv (2024). doi: https://doi.org/10.48550/arXiv.2410.19843

    2024 doi

  37. [45]

    Isogeometric boundary element method for steady-state heat transfer with concentrated/surface heat sources

    Quansheng Zang et al. “Isogeometric boundary element method for steady-state heat transfer with concentrated/surface heat sources”. In: Engineering Analysis with Boundary Elements 122 (2021), pp. 202–213

    2021

  38. [46]

    Multi-domain physics-informed neural networks for solving transient heat conduction problems in multilayer materials

    Benrong Zhang, Fajie Wang, and Lin Qiu. “Multi-domain physics-informed neural networks for solving transient heat conduction problems in multilayer materials”. In:Journal of Applied Physics 133 (2023), p. 245103

    2023

  39. [47]

    A Comprehensive Survey on Transfer Learning

    Fuzhen Zhuang et al. “A Comprehensive Survey on Transfer Learning”. In: Proceedings of the IEEE 109.1 (2021), pp. 43–76. doi: 10.1109/JPROC.2020.3004555

    2021

  40. [48]

    Deep autoencoder based energy method for the bending, vibration, and buckling analysis of Kirchhoff plates with transfer learning

    Xiaoying Zhuang et al. “Deep autoencoder based energy method for the bending, vibration, and buckling analysis of Kirchhoff plates with transfer learning”. In: European Journal of Mechanics - A/Solids 87 (2021), p. 104225. issn: 0997-7538. doi: https://doi.org/10. 1016/j.eurom...

    2021

  41. [9310]

    org / 10

    doi: https : / / doi . org / 10 . 1016 / j . ijheatmasstransfer . 2015 . 06 . 023. url: https://www.sciencedirect.com/science/article/pii/S0017931015006456

    2015

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.