pith. sign in

arxiv: 1907.02531 · v1 · pith:XYLGHOUAnew · submitted 2019-07-04 · 📊 stat.ML · cs.LG

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

Somdatta Goswami , Cosmin Anitescu , Souvik Chakraborty , Timon Rabczuk This is my paper

Pith reviewed 2026-05-25 08:56 UTC · model grok-4.3

classification 📊 stat.ML cs.LG
keywords phase-field fracturephysics informed neural networksvariational energytransfer learningbrittle fracturespline CAD geometryGauss quadrature
0
0 comments X

The pith

A neural network trained by minimizing variational energy with exact boundary conditions models brittle fracture more accurately than residual-based PINNs.

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

The paper introduces a physics-informed neural network for phase-field modeling of fracture that minimizes the variational energy of the system instead of the residual of the governing equations. Boundary conditions are satisfied exactly by modifying the network output, and the variational energy is computed efficiently using Gauss quadrature on spline-based CAD geometry. Transfer learning is applied to obtain crack paths efficiently. In four example problems the approach produces results matching literature values and yields higher accuracy than conventional residual-based PINNs on the first two cases.

Core claim

The central claim is that minimizing the variational energy functional with a neural network whose output is adjusted to enforce boundary conditions exactly, combined with Gauss quadrature integration on spline CAD models and transfer learning, produces accurate crack paths and load-displacement responses in phase-field fracture problems while requiring only lower-order derivatives and achieving better accuracy than residual-minimizing networks.

What carries the argument

Neural network output modified to satisfy boundary conditions exactly, trained by minimizing variational energy computed via Gauss quadrature on spline CAD geometry, with transfer learning for crack-path prediction.

If this is right

  • Boundary conditions are enforced exactly without penalty terms or soft constraints.
  • Only lower-order derivatives appear in the loss, reducing training cost.
  • Results on standard fracture problems match literature values.
  • Transfer learning accelerates computation of successive crack configurations.

Where Pith is reading between the lines

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

  • The same variational-energy approach could apply directly to other mechanics problems whose governing equations derive from energy minimization.
  • Native use of spline CAD input may allow seamless coupling with existing engineering design tools.
  • Efficiency improvements could extend the method to three-dimensional or dynamic fracture scenarios without prohibitive cost.

Load-bearing premise

Minimizing the variational energy with the modified network output and Gauss quadrature on spline geometry produces the correct crack path and load-displacement response without additional regularization.

What would settle it

A clear mismatch in predicted crack path or load-displacement curve compared with established finite-element or experimental results on a single-edge-notched tension specimen.

Figures

Figures reproduced from arXiv: 1907.02531 by Cosmin Anitescu, Somdatta Goswami, Souvik Chakraborty, Timon Rabczuk.

Figure 1
Figure 1. Figure 1: (a) Schematic representation of the proposed physics informed neural network. For computing the deriva [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: (a) This is the original geometry to be modeled. (b) Modeling the geometry using the piecewise polynomial [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The initial tensor product mesh at level 0 is refined via cross-insertion technique using the quadtree decom [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The parameters for each layer is represented as [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Geometrical setup of one-dimensional elastic bar with crack. [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Results for one-dimensional elastic bar with crack using variation energy based PINN (proposed approach). [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: (b) presents the modeled mesh used for the generation of Gauss points for training the deep neural network. The objective is to compute the crack path and the failure load of the system. (a) Geometrical setup and boundary conditions. (b) Mesh for generating the Gauss points [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Predicted crack pattern for prescribed displacement of (a) [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Scatter plots of the deformed configuration for prescribed displacement of (a) [PITH_FULL_IMAGE:figures/full_fig_p016_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Plots of convergence of the loss function. The top row shows the plots of convergence without the reuse [PITH_FULL_IMAGE:figures/full_fig_p017_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Perforated and notched asymmetric bending example. The diameter of the holes are 0.5. [PITH_FULL_IMAGE:figures/full_fig_p017_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Plots showing the predicted phase-field for prescribed displacement of (a) initialization of crack, (b) [PITH_FULL_IMAGE:figures/full_fig_p018_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Plots showing the predicted y-displacement for prescribed displacement of (a) initial crack (without any [PITH_FULL_IMAGE:figures/full_fig_p018_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Geometrical setup and boundary conditions of three dimensional mode-I tension test. [PITH_FULL_IMAGE:figures/full_fig_p019_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Plots of cube showing the predicted phase-field for prescribed displacement of (a) initial crack (without [PITH_FULL_IMAGE:figures/full_fig_p019_15.png] view at source ↗
read the original abstract

We present a new physics informed neural network (PINN) algorithm for solving brittle fracture problems. While most of the PINN algorithms available in the literature minimize the residual of the governing partial differential equation, the proposed approach takes a different path by minimizing the variational energy of the system. Additionally, we modify the neural network output such that the boundary conditions associated with the problem are exactly satisfied. Compared to conventional residual based PINN, the proposed approach has two major advantages. First, the imposition of boundary conditions is relatively simpler and more robust. Second, the order of derivatives present in the functional form of the variational energy is of lower order than in the residual form. Hence, training the network is faster. To compute the total variational energy of the system, an efficient scheme that takes as input a geometry described by spline based CAD model and employs Gauss quadrature rules for numerical integration has been proposed. Moreover, we utilize the concept of transfer learning to obtain the crack path in an efficient manner. The proposed approach is used to solve four fracture mechanics problems. For all the examples, results obtained using the proposed approach match closely with the results available in the literature. For the first two examples, we compare the results obtained using the proposed approach with the conventional residual based neural network results. For both the problems, the proposed approach is found to yield better accuracy compared to conventional residual based PINN algorithms.

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 / 1 minor

Summary. The paper proposes a physics-informed neural network (PINN) for phase-field modeling of brittle fracture that minimizes the variational energy functional rather than the PDE residual. The network output is modified to enforce boundary conditions exactly, Gauss quadrature on spline CAD geometry is used for integration, and transfer learning is applied to obtain crack paths. The approach is demonstrated on four fracture problems, claiming close agreement with literature results and better accuracy than residual-based PINNs on the first two examples.

Significance. If validated with quantitative metrics, the energy-based formulation could offer advantages in BC enforcement simplicity and reduced derivative order for training efficiency in fracture simulations. The CAD integration and transfer learning elements provide practical implementation strengths for engineering use cases.

major comments (2)
  1. [Abstract] Abstract: the claim that the proposed approach yields better accuracy than conventional residual-based PINN algorithms on the first two examples supplies no quantitative error metrics, error bars, mesh convergence data, or description of how accuracy was measured, rendering the central comparative claim unverifiable from the reported information.
  2. [Method description (variational energy minimization)] The method relies on direct minimization of the non-convex phase-field variational energy to recover physical crack paths and load-displacement responses, yet provides no discussion of initialization sensitivity, multiple training runs, or stabilization techniques to guard against spurious local minima; this assumption is load-bearing for the accuracy claims.
minor comments (1)
  1. [Abstract] The abstract states results 'match closely with the literature' for four problems but does not specify which quantitative or qualitative measures (e.g., crack path deviation, energy error) were used for this assessment.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We appreciate the referee's insightful comments on our manuscript. Below, we provide point-by-point responses to the major comments and indicate the revisions we plan to make.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that the proposed approach yields better accuracy than conventional residual-based PINN algorithms on the first two examples supplies no quantitative error metrics, error bars, mesh convergence data, or description of how accuracy was measured, rendering the central comparative claim unverifiable from the reported information.

    Authors: We agree that the abstract's claim of better accuracy would benefit from quantitative support to be fully verifiable. In the revised manuscript, we will update the abstract to include specific quantitative error metrics (e.g., relative L2 errors in displacement and phase-field variables) and a brief description of how accuracy was measured against reference solutions from the literature. We will also ensure that the main text provides the corresponding data, error bars from multiple runs if applicable, and any mesh or discretization convergence information. revision: yes

  2. Referee: [Method description (variational energy minimization)] The method relies on direct minimization of the non-convex phase-field variational energy to recover physical crack paths and load-displacement responses, yet provides no discussion of initialization sensitivity, multiple training runs, or stabilization techniques to guard against spurious local minima; this assumption is load-bearing for the accuracy claims.

    Authors: The referee correctly identifies a potential vulnerability in relying on energy minimization without addressing optimization challenges. We will revise the method section to include a discussion of the initialization procedure (e.g., starting from a uniform phase-field or pre-trained states via transfer learning), results from multiple independent training runs with varied random initializations to assess sensitivity, and any stabilization approaches such as incremental loading or the use of transfer learning to guide the optimization toward physically relevant minima. This will strengthen the reliability of our accuracy claims. revision: yes

Circularity Check

0 steps flagged

No circularity: algorithmic proposal and empirical comparisons are independent of inputs

full rationale

The paper introduces a variational-energy minimization PINN with exact BC enforcement via output modification, spline CAD quadrature, and transfer learning. These are presented as distinct algorithmic choices compared to residual-based PINNs. No equations, parameters, or claims reduce by construction to fitted values or self-citations; accuracy comparisons on four examples are external numerical results, not tautological. The derivation chain remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The approach rests on the standard variational formulation of phase-field fracture and on the assumption that neural-network training can reliably locate the global energy minimum for these problems.

axioms (2)
  • domain assumption The total variational energy of the phase-field fracture system can be computed accurately via Gauss quadrature on spline-based CAD geometry.
    Invoked to justify the integration scheme used to evaluate the loss.
  • domain assumption Transfer learning from a prior similar fracture problem yields a useful initialization for the crack-path solution.
    Central to the efficiency claim but not derived in the abstract.

pith-pipeline@v0.9.0 · 5792 in / 1290 out tokens · 37701 ms · 2026-05-25T08:56:06.863644+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

29 extracted references · 29 canonical work pages · 2 internal anchors

  1. [1]

    Physics Informed Deep Learning (Part I): Data-driven Solutions of Nonlinear Partial Differential Equations

    Maziar Raissi, Paris Perdikaris, and George E. Karniadakis. Physics informed deep learning (part I): data-driven solutions of nonlinear partial differential equations. CoRR, abs/1711.10561, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  2. [2]

    Deep hidden physics models: Deep learning of nonlinear partial differential equations

    Maziar Raissi. Deep hidden physics models: Deep learning of nonlinear partial differential equations. Journal of Machine Learning Research, 19:25:1–25:24, 2018

    work page 2018

  3. [3]

    Hidden Fluid Mechanics: A Navier-Stokes Informed Deep Learning Framework for Assimilating Flow Visualization Data

    Maziar Raissi, Alireza Yazdani, and George Em Karniadakis. Hidden fluid mechanics: A Navier-Stokes informed deep learning framework for assimilating flow visualization data. CoRR, abs/1808.04327, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  4. [4]

    Physics-constrained deep learning for high-dimensional surrogate modeling and uncertainty quantification without labeled data

    Yinhao Zhu, Nicholas Zabaras, Phaedon-Stelios Koutsourelakis, and Paris Perdikaris. Physics-constrained deep learning for high-dimensional surrogate modeling and uncertainty quantification without labeled data. Journal of Computational Physics, 394:56–81, oct 2019

    work page 2019

  5. [5]

    Modeling the Dynamics of PDE Systems with Physics-Constrained Deep Auto-Regressive Networks

    Nicholas Geneva and Nicholas Zabaras. Modeling the Dynamics of PDE Systems with Physics-Constrained Deep Auto-Regressive Networks. Arxiv prep, jun 2019

    work page 2019

  6. [6]

    Simulator-free Solution of High- Dimensional Stochastic Elliptic Partial Differential Equations using Deep Neural Networks

    Sharmila Karumuri, Rohit Tripathy, Ilias Bilionis, and Jitesh Panchal. Simulator-free Solution of High- Dimensional Stochastic Elliptic Partial Differential Equations using Deep Neural Networks. Arxiv prep, feb 2019

    work page 2019

  7. [7]

    Bourdin, G.A

    B. Bourdin, G.A. Francfort, and J.-J. Marigo. Numerical experiments in revisited brittle fracture. Journal of the Mechanics and Physics of Solids, 48(4):797–826, apr 2000. 20 A PREPRINT - J ULY 8, 2019

    work page 2000

  8. [8]

    Davidson

    Xuekun Sun and Barry D. Davidson. A Direct Energy Balance Approach for Determining Energy Release Rates in Three and Four Point Bend End Notched Flexure Tests. International Journal of Fracture, 135(1-4):51–72, sep 2005

    work page 2005

  9. [9]

    Weinan and Yu Bing

    E. Weinan and Yu Bing. The deep ritz method: A deep learning-based numerical algorithm for solving variational problems. Communications in Mathematics and Statistics, 6(1):1–12, Mar 2018

    work page 2018

  10. [10]

    Hybrid discrete element/finite element method for fracture analysis

    N Monteiro Azevedo and JV Lemos. Hybrid discrete element/finite element method for fracture analysis. Com- puter Methods in Applied Mechanics and Engineering, 195(33-36):4579–4593, 2006

    work page 2006

  11. [11]

    The finite element method , volume 3

    Olgierd Cecil Zienkiewicz, Robert Leroy Taylor, Perumal Nithiarasu, and JZ Zhu. The finite element method , volume 3. McGraw-Hill London, 1977

    work page 1977

  12. [12]

    Isogeometric analysis: toward integration of CAD and FEA

    J Austin Cottrell, Thomas JR Hughes, and Yuri Bazilevs. Isogeometric analysis: toward integration of CAD and FEA. John Wiley & Sons, 2009

    work page 2009

  13. [13]

    Miehe, F

    C. Miehe, F. Welschinger, and M. Hofacker. Thermodynamically consistent phase-field models of fracture: Variational principles and multi-field FE implementations. International Journal for Numerical Methods in Engineering, 83(10):1273–1311, sep 2010

    work page 2010

  14. [14]

    A phase field model for rate-independent crack propagation: Robust algorithmic implementation based on operator splits

    Christian Miehe, Martina Hofacker, and Fabian Welschinger. A phase field model for rate-independent crack propagation: Robust algorithmic implementation based on operator splits. Computer Methods in Applied Me- chanics and Engineering, 199(45-48):2765–2778, nov 2010

    work page 2010

  15. [15]

    Borden, Clemens V

    Michael J. Borden, Clemens V . Verhoosel, Michael A. Scott, Thomas J.R. Hughes, and Chad M. Landis. A phase-field description of dynamic brittle fracture. Computer Methods in Applied Mechanics and Engineering , 217-220:77–95, apr 2012

    work page 2012

  16. [16]

    A. Griffith. The Phenomena of Rupture and Flow in Solids. Philisophical Transactions of the Royal Society of London, 221(Series A):163–198, 1921

    work page 1921

  17. [17]

    Borden, Thomas J.R

    Michael J. Borden, Thomas J.R. Hughes, Chad M. Landis, and Clemens V . Verhoosel. A higher-order phase- field model for brittle fracture: Formulation and analysis within the isogeometric analysis framework.Computer Methods in Applied Mechanics and Engineering, 273:100–118, may 2014

    work page 2014

  18. [18]

    Timo Heister, Mary F Wheeler, and Thomas Wick. A primal-dual active set method and predictor-corrector mesh adaptivity for computing fracture propagation using a phase-field approach.Computer Methods in Applied Mechanics and Engineering, 290:466–495, 2015

    work page 2015

  19. [19]

    Verhoosel

    Julien Vignollet, Stefan May, René de Borst, and Clemens V . Verhoosel. Phase-field models for brittle and cohesive fracture. Meccanica, 49(11):2587–2601, nov 2014

    work page 2014

  20. [20]

    R. Rojas. Neural network: A systematic introduction. Springer, 1996

    work page 1996

  21. [21]

    The NURBS book

    Les Piegl and Wayne Tiller. The NURBS book. Springer Science & Business Media, 2012

    work page 2012

  22. [22]

    Polynomial splines over hierarchical T-meshes

    Jiansong Deng, Falai Chen, Xin Li, Changqi Hu, Weihua Tong, Zhouwang Yang, and Yuyu Feng. Polynomial splines over hierarchical T-meshes. Graphical Models, 70(4):76–86, 2008

    work page 2008

  23. [23]

    Understanding the difficulty of training deep feedforward neural networks

    Xavier Glorot and Yoshua Bengio. Understanding the difficulty of training deep feedforward neural networks. In Proceedings of the thirteenth international conference on artificial intelligence and statistics, pages 249–256, 2010

    work page 2010

  24. [24]

    Tensorflow: Large-scale machine learning on heterogeneous systems,

    Martın Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S Corrado, Andy Davis, Jeffrey Dean, Matthieu Devin, et al. Tensorflow: Large-scale machine learning on heterogeneous systems,

    work page

  25. [25]

    org, 1(2), 2015

    Software available from tensorflow. org, 1(2), 2015

    work page 2015

  26. [26]

    Borden, and Henryk K

    Dominik Schillinger, Michael J. Borden, and Henryk K. Stolarski. Isogeometric collocation for phase-field fracture models. Computer Methods in Applied Mechanics and Engineering, 284:583–610, feb 2015

    work page 2015

  27. [27]

    A fenics implementation of the phase field method for quasi-static brittle fracture

    Sundararajan Natarajan, Ratna Kumar Annabattula, et al. A fenics implementation of the phase field method for quasi-static brittle fracture. Frontiers of Structural and Civil Engineering, pages 1–17, 2019

    work page 2019

  28. [28]

    Quasi-automatic simulation of crack propagation for 2d lefm problems

    TN Bittencourt, PA Wawrzynek, AR Ingraffea, and JL Sousa. Quasi-automatic simulation of crack propagation for 2d lefm problems. Engineering Fracture Mechanics, 55(2):321–334, 1996

    work page 1996

  29. [29]

    2d and 3d abaqus implementation of a robust staggered phase-field solution for modeling brittle fracture

    Gergely Molnár and Anthony Gravouil. 2d and 3d abaqus implementation of a robust staggered phase-field solution for modeling brittle fracture. Finite Elements in Analysis and Design, 130:27–38, 2017. 21

    work page 2017