Pith. sign in

REVIEW 3 major objections 5 minor 68 references

A single topology-optimized specimen under simple uniaxial loading can generate enough stress–strain diversity to train a large recurrent neural network material model.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-03 15:55 UTC pith:RI4YZUUR

load-bearing objection Solid in-silico proof that topology-optimized specimens can generate diverse local strain paths for training recurrent constitutive models, but the 'single uniaxial test' framing outruns what is actually demonstrated. the 3 major comments →

arxiv 2512.14963 v1 pith:RI4YZUUR submitted 2025-12-16 physics.comp-ph cs.NAmath.NA

Design of a specimen to train path-dependent deep learning material models from a single uniaxial test: eliciting strain diversity via automatically differentiable elastoplastic topology optimization

Shunyu Yin , Bernardo P. Ferreira , Gawel Kus , Miguel A. Bessa This is my paper
classification physics.comp-ph cs.NAmath.NA MSC 74P1574C0568T07
keywords topology optimizationelastoplasticitypath-dependent constitutive modelingrecurrent neural networkstrain diversityautomatic differentiationmaterial model calibrationsingle-test characterization
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Standard test specimens like dogbones produce repetitive stress–strain paths, too uniform to train neural network constitutive models with millions of parameters. This paper shows that a specimen designed by elastoplastic topology optimization creates much more diverse local strain states under a single uniaxial cyclic test. A GRU-based network trained on data from that optimized specimen predicts unseen path-dependent plasticity with about 10% normalized error, versus 55–166% for the dogbone. The result suggests that the diverse data needed for machine-learning material models can be generated from one experiment rather than many synthetic unit-cell simulations or complex multi-axial tests.

Core claim

The central discovery is that a specimen shaped by elastoplastic topology optimization, maximizing an entropy-based measure of strain-state coverage, produces local stress–strain paths under cyclic uniaxial loading that are diverse enough to train a two-million-parameter GRU material model. Trained on these paths, the GRU reaches NRMSE of 9.39%, 11.03%, and 10.88% for the three in-plane stress components, whereas a dogbone specimen yields 87.38%, 166.36%, and 55.52%. The optimized design retains most of its advantage when the underlying material changes from von Mises to Drucker–Prager plasticity (NRMSE around 10–13%), supporting the generality of the claim.

What carries the argument

The core mechanism is an entropy-based, differentiable measure of strain diversity. Strain space is discretized into cells and each element's final strain state is softly assigned to neighboring cells with Gaussian kernels, producing a differentiable histogram; its Shannon entropy is the objective. Maximizing this entropy drives the topology optimization (using a density-based SIMP scheme, a cone filter, threshold projection, and the MMA optimizer) to sculpt a specimen whose local strain states broadly cover the strain space. The resulting geometry then serves as a data generator under cyclic loading, and the automatic differentiation–based model updating method trains a GRU material model w

Load-bearing premise

The load-bearing premise is that local stress–strain histories from every solid element—directly available in the finite element simulation—are what a physical experiment would provide; real specimens only give full-field displacements and global forces, so the single-test claim depends on a way to recover local stress–strain data without already knowing the constitutive law.

What would settle it

Fabricate the optimized specimen, apply the same two-cycle tension–compression uniaxial loading, record full-field displacement and global force, and train a GRU on those data without element-wise stress–strain labels; if the resulting model's error on unseen strain paths is comparable to the dogbone-trained model rather than the approximately 10% NRMSE obtained from simulated local data, the single-test claim as stated collapses.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If the result holds in experiments, one uniaxial test on a designed specimen could replace the many synthetic unit-cell simulations currently needed to train neural network constitutive models.
  • The framework separates specimen design from model training, so the same optimized geometry can train other material models (physics-informed, hybrid, or classical) without redesign.
  • Data redundancy drops substantially: the pruning analysis shows roughly 35% of the optimized specimen's data is redundant versus much higher redundancy for the dogbone, meaning fewer samples are needed.
  • The Drucker–Prager test suggests the design generalizes to materials that deviate from the von Mises law assumed during optimization.
  • The three-stage workflow offers a concrete route toward the 'Material Testing 2.0' idea of replacing standardized tests with a single information-rich experiment.

Where Pith is reading between the lines

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

  • The central numerical experiment uses element-wise local stress–strain data, which are not directly measurable in a physical test; a real experiment provides full-field displacements and global forces, so the single-test claim is not yet experimentally established.
  • The entropy objective is computed from final monotonic-load strain states, not from path diversity under cyclic loading; a testable extension would optimize directly for the spread of strain paths, which may further improve generalization.
  • The framework could likely be applied to other history-dependent behaviors (creep, damage, phase transformation) by changing the material model used to generate synthetic data and the diversity metric.
  • A stronger validation would be to use the optimized specimen to train a model and then test it on a different specimen geometry entirely, ensuring the learned model is material-specific rather than specimen-specific.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes an automatically differentiable elastoplastic topology optimization framework to design a specimen that, under uniaxial cyclic loading, generates a diverse set of local stress–strain paths. The optimized specimen is then used in a finite element simulation to produce a dataset of element-wise stress–strain histories, which trains a GRU-based recurrent neural network material model within the ADiMU/HookeAI framework. The authors compare the optimized design against dogbone, notched, and random geometries, report substantially lower NRMSE for the optimized design, analyze data redundancy through pruning, test robustness to a Drucker–Prager material model, and present additional 2D/3D, mesh, filter, and loading studies. The headline claim is that a single topology-optimized specimen under simple uniaxial loading can train a large path-dependent neural network, thereby reducing the experimental burden of data-driven constitutive modeling.

Significance. If the central claim holds for physical experiments, this work would be a meaningful step toward the Material Testing 2.0 vision: replacing many standardized tests or synthetic unit-cell simulations with one information-rich experiment. The in-silico evidence is coherent and relatively strong: the optimized specimen yields NRMSE values around 10%, whereas dogbone and notched specimens yield errors of 55–166%; random designs perform worse and with high variance; pruning shows lower redundancy; and the Drucker–Prager study demonstrates robustness to material-model mismatch. The novelty—integrating elastoplastic topology optimization via automatic differentiation with recurrent neural network surrogate training—is genuine and timely. However, the current manuscript establishes the claim only in a computational setting where element-wise stresses and strains are known exactly from the prescribed constitutive model; the path to physical experiments is acknowledged but not yet demonstrated.

major comments (3)
  1. [§2.2, §4] The central claim is not yet supported for physical experiments. In Stage 2, training data are element-wise local stress–strain paths extracted from FEA, where both fields are known exactly because the constitutive model is prescribed. A physical uniaxial test yields full-field displacements (e.g., DIC) and global reaction forces, not local stresses. The authors acknowledge this gap in Discussion §4 ('Unlike the present study, which relies on local stress–strain paths, such experiments will provide full-field displacement data with global force data.') but do not close it. As written, the abstract and Table 2 support an in-silico proof of specimen-design value, not the single-test claim for real materials. The manuscript should either explicitly scope the claim to simulation or provide an analysis (e.g., integrated-DIC/FEMU-style inversion) showing that local stress–strain paths can be r
  2. [§A.2, Table 5; Abstract] The adopted loading protocol is not monotonic uniaxial loading but a two-cycle tension-compression test. Table 5 shows that monotonic tension (Loading 1) gives NRMSE values of 169–180%, one-cycle tension-compression (Loading 3) gives about 29–34%, and only two-cycle tension-compression (Loading 4) reaches the reported 9–11% performance. The abstract's phrase 'simple uniaxial loading' is therefore ambiguous; the demonstrated claim is for a single cyclic tension-compression test. Because the required loading protocol materially affects the experimental effort and the scope of the claim, it should be stated explicitly in the abstract and conclusion.
  3. [§2.1.1, §2.2] The optimization objective (Eqs. 1–8) maximizes the Shannon entropy of final strain states under monotonic loading, whereas the training dataset is composed of stress–strain paths under cyclic loading. The paper treats final-state strain-space coverage as a proxy for the path diversity needed by a recurrent model, without an ablation or comparison to alternative diversity metrics. The random-design and pruning comparisons support the practical value of the optimized geometry, but the relationship between the entropy objective and recurrent-model training utility remains an assumption. A brief discussion or a simple ablation would strengthen this step.
minor comments (5)
  1. [Tables 2 and 3] The text mentions three random model initializations for some results (e.g., Fig. 7 and Fig. 9), but Tables 2 and 3 report single NRMSE values without indicating whether these are averages or a single run. Please clarify and, if averages, report standard deviations or at least state the number of seeds.
  2. [§2.1.1, item 1] The phrase 'yielding n ns b total cells' appears garbled; it should probably read 'n_b^{n_s} total cells'. Please correct the notation.
  3. [Eq. (6)] The symbol ρ_SIMP is used in the histogram construction but is not defined explicitly in the main text. Please state whether it is the filtered, projected, or raw density used in the finite element computation.
  4. [§A.4] The sentence 'the predictive performance based on the standard specimens is significantly worse than that obtained from the optimized specimen, even with fewer training paths' is ambiguous. The optimized specimen actually uses fewer paths (1280) than the dogbone (6000) or notch (2000); please rephrase to make clear which dataset has fewer paths.
  5. [General] No code/data availability statement is provided beyond the HookeAI repository link. For a computational paper of this type, a short statement on whether the topology optimization code and datasets will be released would improve reproducibility.

Circularity Check

0 steps flagged

No significant circularity: the specimen-design, dataset-generation, and GRU-generalization steps are distinct and validated independently; only a non-circular observability limitation is acknowledged.

full rationale

Walked the derivation chain. The specimen is obtained by entropy-based topology optimization (Eqs. 1–8) using von Mises FEA under monotonic loading; the optimized geometry is then independently subjected to cyclic loading to extract local element stress–strain paths (§2.2), and a two-million-parameter GRU is trained by minimizing MSE on those paths and tested on a separate randomly generated polynomial path set (§2.3.1). No parameter fitted to the test set is later reported as a prediction; the NRMSE comparisons to dogbone/notch/random designs (Tab. 2, Tabs. 4–8) are direct controlled comparisons rather than by-construction identities. The entropy objective is not defined in terms of the GRU's errors, and the test set does not reuse optimized-specimen paths. The use of the authors' ADiMU/HookeAI and the GRU architecture from [9] is tool reuse: the paper's contribution—strain-diversity maximization through elastoplastic topology optimization—is evaluated independently and would stand even if those tools were replaced. The only flagged limitation, in §4, is that real experiments supply displacement/force fields rather than local stress–strain paths; this is an observability/validity gap for the experimental claim, not a circular reduction of the derivation to its inputs. The test-set ground truth is generated by the same constitutive model used to produce training data, but that is a standard supervised setup, not circularity: the optimization never sees the test paths, and the GRU must generalize from specimen-induced paths to arbitrary strain histories. No self-definitional, fit-as-prediction, uniqueness-imported, or ansatz-by-citation step was found.

Axiom & Free-Parameter Ledger

9 free parameters · 6 axioms · 0 invented entities

The central computational demonstration rests on several hand-set objective hyperparameters (strain bounds, bin count, kernel width, continuation schedules, filter radius, volume fraction) and on the assumed von Mises ground truth. No new physical entities are introduced.

free parameters (9)
  • Strain-space bounds for entropy objective = ε11∈[−0.1,0.1], ε22∈[0,0.1], ε12∈[−0.1,0.1] (Tab. 1)
    Defines the target strain diversity region; chosen by hand and directly shapes the optimized geometry.
  • Number of histogram bins nb = 10 intervals per strain component
    Discretization of strain space in the objective; chosen, not optimized.
  • Gaussian kernel scaling factor s = 1/6
    Chosen to emulate hard cell counting while retaining differentiability (Eq. 2).
  • SIMP penalization exponent schedule p = 1→10
    Continuation strategy to push densities to 0/1; standard but hand-set.
  • Threshold projection steepness β = 1→10
    Continuation for near-binary projection (Eq. 14); hand-set.
  • Density filter radius r = 4 elements
    Controls minimum feature size; main paper uses 4, appendix studies 2–10.
  • Volume-fraction constraint and initial density = [40%,60%] and ρ=0.5
    Constraints chosen; affect feasible designs.
  • GRU architecture/hyperparameters = 2×500 GRU, LR=0.001, 200 epochs, batch 32, seq 200
    Hand-chosen; a 2M-parameter model is deliberately used as a stress test.
  • Training loading case = two-cycle tension-compression
    Selected among eight cases after observing performance (A.2); case 8 performed better but required two tests.
axioms (6)
  • domain assumption The von Mises elastoplastic model with isotropic hardening (E=110 GPa, ν=0.33, Y0=900 MPa, Et=500 MPa) is the true material behavior used for both optimization and dataset generation.
    Sec. 2.1/C.1; all synthetic data come from this model; Drucker–Prager test only partially relaxes this.
  • domain assumption A thin sheet can be modeled under plane stress in 2D, and a single-element thickness suffices for 3D.
    Sec. 2.1 and Appendix B; strain diversity measured only in-plane in 3D.
  • ad hoc to paper Maximizing Shannon entropy of a differentiable strain histogram is a valid proxy for the diversity needed to train a path-dependent neural network.
    Sec. 2.1.1; no proof ties this objective to minimal surrogate error; validated only empirically.
  • domain assumption The random polynomial strain-path test set is representative of unseen loading scenarios for the target material class.
    Sec. 2.3.1; test paths are generated from the same constitutive model used to create training data.
  • standard math The unrolled JAX finite-element implementation with reverse-mode AD yields correct sensitivities for the path-dependent plasticity residual.
    Sec. 2.1; relies on AD correctness of the implemented time-stepping; no verification against analytical adjoint provided.
  • domain assumption A 2M-parameter GRU can represent the elastoplastic constitutive map given sufficiently diverse strain paths.
    Borrowed from prior work [9]; used as a deliberate stress test.

pith-pipeline@v1.3.0-alltime-deepseek · 18666 in / 14119 out tokens · 142687 ms · 2026-08-03T15:55:08.105252+00:00 · methodology

0 comments
read the original abstract

Artificial neural networks accurately learn nonlinear, path-dependent material behavior. However, training them typically requires large, diverse datasets, often created via synthetic unit cell simulations. This hinders practical adoption because physical experiments on standardized specimens with simple geometries fail to generate sufficiently diverse stress-strain trajectories. Consequently, an unreasonably large number of experiments or complex multi-axial tests would be needed. This work shows that such networks can be trained from a single specimen subjected to simple uniaxial loading, by designing the specimen using a novel automatically differentiable elastoplastic topology optimization method. Our strategy diversifies the stress-strain states observed in a single test involving plastic deformation. We then employ the automatically differentiable model updating (ADiMU) method to train the neural network surrogates. This work demonstrates that topology-optimized specimens under simple loading can train large neural networks, thereby substantially reducing the experimental burden associated with data-driven material modeling.

Figures

Figures reproduced from arXiv: 2512.14963 by Bernardo P. Ferreira, Gawel Kus, Miguel A. Bessa, Shunyu Yin.

Figure 1
Figure 1. Figure 1: What is the optimal geometry to train (update or calibrate) a material model? The standard dogbone specimen [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Schematic workflow of the proposed framework. Step 1: Automatically-differentiable elastoplastic topology [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: One-dimensional schematic of the objective-function definition. (a) The predefined strain interval is partitioned [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Visualization of the 2D test set: (a) Accumulated plastic strain versus load step, demonstrating that all test [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Representative 2D design evolution: (a) Optimized design obtained directly from the topology optimization [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Strain paths generated under two-cycle tension–compression loading for (a) the dogbone specimen, (b) the [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: (a) Prediction performance (NRMSE) with respect to the training data set size. (b–d) The mean and standard [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Dataset redundancy analysis. (a) Average prediction NRMSE (averaged over all stress components) of [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: The mean and standard deviation (from 3 random model initializations) of predicted stress components on the [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Proposed experimental-data-driven topology optimization workflow for material model updating. The [PITH_FULL_IMAGE:figures/full_fig_p012_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Optimized 2D designs obtained using different filter radii ( [PITH_FULL_IMAGE:figures/full_fig_p014_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Randomly generated designs [PITH_FULL_IMAGE:figures/full_fig_p016_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Optimized designs for different mesh sizes. (a) In a [PITH_FULL_IMAGE:figures/full_fig_p017_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Representative 3D design evolution. (a) Optimized design obtained directly from the topology optimization [PITH_FULL_IMAGE:figures/full_fig_p018_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Optimized 3D designs obtained using different filter radii ( [PITH_FULL_IMAGE:figures/full_fig_p018_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Visualization of the 3D test set: (a) Accumulated plastic strain versus load step; (b) Probability density [PITH_FULL_IMAGE:figures/full_fig_p020_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Stress history of 3D-optimized specimen dataset. [PITH_FULL_IMAGE:figures/full_fig_p021_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Stress history of 3D testset. 21 [PITH_FULL_IMAGE:figures/full_fig_p021_18.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

68 extracted references · 21 canonical work pages · 2 internal anchors

  1. [1]

    Lemaitre, J.-L

    J. Lemaitre, J.-L. Chaboche, Mechanics of solid materials, Cambridge university press, 1994

    1994

  2. [2]

    J. C. Simo, T. J. Hughes, Computational inelasticity, Springer, 1998

    1998

  3. [3]

    L. E. Malvern, Introduction to the Mechanics of a Continuous Medium, no. Monograph, 1969

    1969

  4. [4]

    Hill, The mathematical theory of plasticity, V ol

    R. Hill, The mathematical theory of plasticity, V ol. 11, Oxford university press, 1998

    1998

  5. [5]

    Truesdell, W

    C. Truesdell, W. Noll, The non-linear field theories of mechanics, in: The non-linear field theories of mechanics, Springer, 2004, pp. 1–579

    2004

  6. [6]

    E. A. de Souza Neto, D. Peric, D. R. Owen, Computational Methods for Plasticity: Theory and Applications, John Wiley & Sons, 2008

    2008

  7. [7]

    Ghaboussi, D

    J. Ghaboussi, D. A. Pecknold, M. Zhang, R. M. Haj-Ali, Autoprogressive training of neural network constitutive models, International Journal for Numerical Methods in Engineering 42 (1) (1998) 105–126

    1998

  8. [8]

    Bessa, R

    M. Bessa, R. Bostanabad, Z. Liu, A. Hu, D. W. Apley, C. Brinson, W. Chen, W. Liu, A framework for data-driven analysis of materials under uncertainty: Countering the curse of dimensionality, Computer Methods in Applied Mechanics and Engineering 320 (2017) 633–667.doi:https://doi.org/10.1016/j.cma.2017.03.037. URLhttps://www.sciencedirect.com/science/artic...

    work page doi:10.1016/j.cma.2017.03.037 2017

  9. [9]

    Mozaffar, R

    M. Mozaffar, R. Bostanabad, W. Chen, K. Ehmann, J. Cao, M. A. Bessa, Deep learning predicts path-dependent plasticity, Proceedings of the National Academy of Sciences 116 (52) (2019) 26414–26420, publisher: Proceedings of the National Academy of Sciences.doi:10.1073/pnas.1911815116. URLhttps://www.pnas.org/doi/10.1073/pnas.1911815116

    doi:10.1073/pnas.1911815116 2019

  10. [10]

    B. P. Ferreira, M. A. Bessa, Automatically Differentiable Model Updating (ADiMU): conventional, hybrid, and neural network material model discovery including history-dependency, arXiv:2505.07801 [math] (May 2025). doi:10.48550/arXiv.2505.07801. URLhttp://arxiv.org/abs/2505.07801

    Pith/arXiv arXiv doi:10.48550/arxiv.2505.07801 2025

  11. [11]

    Zhang, Q.-J

    Y . Zhang, Q.-J. Li, T. Zhu, J. Li, Learning constitutive relations of plasticity using neural networks and full-field data, Extreme Mechanics Letters 52 (2022) 101645

    2022

  12. [12]

    Linden, D

    L. Linden, D. K. Klein, K. A. Kalina, J. Brummund, O. Weeger, M. Kästner, Neural networks meet hyperelasticity: A guide to enforcing physics, Journal of the Mechanics and Physics of Solids 179 (2023) 105363

    2023

  13. [13]

    Wang, L.-y

    C. Wang, L.-y. Xu, J.-s. Fan, A general deep learning framework for history-dependent response prediction based on UA-Seq2Seq model, Computer Methods in Applied Mechanics and Engineering 372 (2020) 113357. doi:10.1016/j.cma.2020.113357. URLhttps://www.sciencedirect.com/science/article/pii/S0045782520305429

    arXiv 2020

  14. [14]

    M. B. Gorji, M. Mozaffar, J. N. Heidenreich, J. Cao, D. Mohr, On the potential of recurrent neural networks for modeling path dependent plasticity, Journal of the Mechanics and Physics of Solids 143 (2020) 103972. doi:10.1016/j.jmps.2020.103972. URLhttps://www.sciencedirect.com/science/article/pii/S0022509620302076

    arXiv 2020

  15. [15]

    M. A. Maia, I. B. C. M. Rocha, P. Kerfriden, F. P. van der Meer, Physically recurrent neural networks for path-dependent heterogeneous materials: Embedding constitutive models in a data-driven surrogate, Computer Methods in Applied Mechanics and Engineering 407 (2023) 115934.doi:10.1016/j.cma.2023.115934. URLhttps://www.sciencedirect.com/science/article/p...

    arXiv 2023

  16. [16]

    Borkowski, C

    L. Borkowski, C. Sorini, A. Chattopadhyay, Recurrent neural network-based multiaxial plasticity model with regularization for physics-informed constraints, Computers & Structures 258 (2022) 106678. doi:10.1016/j. compstruc.2021.106678. URLhttps://www.sciencedirect.com/science/article/pii/S0045794921002005

    arXiv 2022

  17. [17]

    Eghbalian, M

    M. Eghbalian, M. Pouragha, R. Wan, A physics-informed deep neural network for surrogate modeling in classical elasto-plasticity, Computers and Geotechnics 159 (2023) 105472.doi:10.1016/j.compgeo.2023.105472. URLhttps://www.sciencedirect.com/science/article/pii/S0266352X2300229X

    arXiv 2023

  18. [18]

    T. Wang, Y . Yu, H. Luo, Z. Wang, Plastic Constitutive Training Method for Steel Based on a Recurrent Neural Network, Buildings 14 (10) (2024) 3279, number: 10 Publisher: Multidisciplinary Digital Publishing Institute. doi:10.3390/buildings14103279. URLhttps://www.mdpi.com/2075-5309/14/10/3279 22 APREPRINT- DECEMBER18, 2025

    work page doi:10.3390/buildings14103279 2024

  19. [19]

    Bigger, B

    R. Bigger, B. Blaysat, C. Boo, M. Grewer, J. Hu, A. Jones, M. Klein, P. Lava, M. Pankow, K. Raghavan, P. Reu, T. Schmidt, T. Siebert, M. Simonsen, A. Trim, D. Turner, A. Vieira, T. Weikert, E. Jones, M. Iadicola, A Good Practices Guide for Digital Image Correlation, Tech. rep., International Digital Image Correlation Society (Oct. 2018).doi:10.32720/idics...

    work page doi:10.32720/idics/gpg.ed1/print.format 2018

  20. [20]

    Leclerc, J.-N

    H. Leclerc, J.-N. Périé, S. Roux, F. Hild, Integrated digital image correlation for the identification of mechanical properties, in: International conference on computer vision/computer graphics collaboration techniques and applications, Springer, 2009, pp. 161–171

    2009

  21. [21]

    Denys, S

    K. Denys, S. Coppieters, M. Seefeldt, D. Debruyne, Multi-dic setup for the identification of a 3d anisotropic yield surface of thick high strength steel using a double perforated specimen, Mechanics of Materials 100 (2016) 96–108

    2016

  22. [22]

    Advancements in Constitutive Model Calibration: Leveraging the Power of Full-Field DIC Measurements and In-Situ Load Path Selection for Reliable Parameter Inference

    D. Ricciardi, D. T. Seidl, B. Lester, A. Jones, E. Jones, Advancements in Constitutive Model Calibration: Leveraging the Power of Full-Field DIC Measurements and In-Situ Load Path Selection for Reliable Parameter Inference, arXiv:2411.07310 [cs] (Jul. 2025).doi:10.48550/arXiv.2411.07310. URLhttp://arxiv.org/abs/2411.07310

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2411.07310 2025

  23. [23]

    Pierron, Material Testing 2.0: A brief review, Strain 59 (3) (2023) e12434, _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1111/str.12434.doi:10.1111/str.12434

    F. Pierron, Material Testing 2.0: A brief review, Strain 59 (3) (2023) e12434, _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1111/str.12434.doi:10.1111/str.12434. URLhttps://onlinelibrary.wiley.com/doi/abs/10.1111/str.12434

    work page doi:10.1111/str.12434.doi:10.1111/str.12434 2023

  24. [24]

    Guélon, E

    T. Guélon, E. Toussaint, J.-B. Le Cam, N. Promma, M. Grediac, A new characterisation method for rubber, Polymer Testing 28 (2009) 715–723, publisher: Elsevier.doi:10.1016/j.polymertesting.2009.06.001. URLhttps://hal.science/hal-01131584

    work page doi:10.1016/j.polymertesting.2009.06.001 2009

  25. [25]

    J. Fu, W. Xie, J. Zhou, L. Qi, A method for the simultaneous identification of anisotropic yield and hardening constitutive parameters for sheet metal forming, International Journal of Mechanical Sciences 181 (2020) 105756. doi:10.1016/j.ijmecsci.2020.105756. URLhttps://www.sciencedirect.com/science/article/pii/S002074031934740X

    arXiv 2020

  26. [26]

    K. T. Kavanagh, R. W. Clough, Finite element applications in the characterization of elastic solids, International Journal of Solids and Structures 7 (1) (1971) 11–23

    1971

  27. [27]

    Grédiac, Principe des travaux virtuels et identification, Comptes rendus de l’Académie des sciences

    M. Grédiac, Principe des travaux virtuels et identification, Comptes rendus de l’Académie des sciences. Série 2, Mécanique, Physique, Chimie, Sciences de l’univers, Sciences de la Terre 309 (1) (1989) 1–5

    1989

  28. [28]

    H. Shin, G. Pande, On self-learning finite element codes based on monitored response of structures, Computers and Geotechnics 27 (3) (2000) 161–178.doi:https://doi.org/10.1016/S0266-352X(00)00016-1. URLhttps://www.sciencedirect.com/science/article/pii/S0266352X00000161

    work page doi:10.1016/s0266-352x(00)00016-1 2000

  29. [29]

    Lefik, B

    M. Lefik, B. Schrefler, Artificial neural network as an incremental non-linear constitutive model for a finite element code, Computer Methods in Applied Mechanics and Engineering 192 (28) (2003) 3265–3283, multiscale Computational Mechanics for Materials and Structures. doi:https://doi.org/10.1016/S0045-7825(03) 00350-5. URLhttps://www.sciencedirect.com/s...

    work page doi:10.1016/s0045-7825(03 2003

  30. [30]

    Mathieu, H

    F. Mathieu, H. Leclerc, F. Hild, S. Roux, Estimation of Elastoplastic Parameters via Weighted FEMU and Integrated-DIC, Experimental Mechanics 55 (1) (2015) 105–119.doi:10.1007/s11340-014-9888-9. URLhttps://doi.org/10.1007/s11340-014-9888-9

    doi:10.1007/s11340-014-9888-9 2015

  31. [31]

    Gerbig, A

    D. Gerbig, A. Bower, V . Savic, L. G. Hector Jr, Coupling digital image correlation and finite element analysis to determine constitutive parameters in necking tensile specimens, International Journal of Solids and Structures 97 (2016) 496–509

    2016

  32. [32]

    Thakolkaran, A

    P. Thakolkaran, A. Joshi, Y . Zheng, M. Flaschel, L. De Lorenzis, S. Kumar, Nn-euclid: Deep-learning hyperelasticity without stress data, Journal of the Mechanics and Physics of Solids 169 (2022) 105076. doi:https://doi.org/10.1016/j.jmps.2022.105076. URLhttps://www.sciencedirect.com/science/article/pii/S0022509622002538

    arXiv 2022

  33. [33]

    X. Wu, Y . Zhang, S. Mao, Learning the physics-consistent material behavior from measurable data via pde- constrained optimization, Computer Methods in Applied Mechanics and Engineering 437 (2025) 117748. doi: https://doi.org/10.1016/j.cma.2025.117748. URLhttps://www.sciencedirect.com/science/article/pii/S0045782525000209

    arXiv 2025

  34. [34]

    Akerson, A

    A. Akerson, A. Rajan, K. Bhattacharya, Learning constitutive relations from experiments: 1. pde constrained optimization, Journal of the Mechanics and Physics of Solids 201 (2025) 106128. doi:https://doi.org/10. 1016/j.jmps.2025.106128. URLhttps://www.sciencedirect.com/science/article/pii/S0022509625001048 23 APREPRINT- DECEMBER18, 2025

    arXiv 2025

  35. [35]

    B. Chen, B. Starman, M. Haliloviˇc, L. A. Berglund, S. Coppieters, Finite Element Model Updating for Material Model Calibration: A Review and Guide to Practice, Archives of Computational Methods in Engineering 32 (4) (2025) 2035–2112.doi:10.1007/s11831-024-10200-9. URLhttps://doi.org/10.1007/s11831-024-10200-9

    doi:10.1007/s11831-024-10200-9 2025

  36. [36]

    Demmerle, J

    S. Demmerle, J. P. Boehler, Optimal design of biaxial tensile cruciform specimens, Journal of the Mechanics and Physics of Solids 41 (1) (1993) 143–181.doi:10.1016/0022-5096(93)90067-P. URLhttps://www.sciencedirect.com/science/article/pii/002250969390067P

    arXiv 1993

  37. [37]

    Creuziger, M

    A. Creuziger, M. A. Iadicola, T. Foecke, E. Rust, D. Banerjee, Insights into Cruciform Sample Design, JOM (Warrendale, Pa. : 1989) 69 (5) (2017) 902–906.doi:10.1007/s11837-017-2261-6. URLhttps://www.ncbi.nlm.nih.gov/pmc/articles/PMC5520637/

    work page doi:10.1007/s11837-017-2261-6 1989

  38. [38]

    M. B. R. Bertin, F. Hild, S. Roux, Optimization of a Cruciform Specimen Geometry for the Identification of Constitutive Parameters Based Upon Full-Field Measurements, Strain 52 (4) (2016) 307–323, _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1111/str.12178.doi:10.1111/str.12178. URLhttps://onlinelibrary.wiley.com/doi/abs/10.1111/str.12178

    work page doi:10.1111/str.12178.doi:10.1111/str.12178 2016

  39. [39]

    Chapelier, R

    M. Chapelier, R. Bouclier, J.-C. Passieux, Spline-based specimen shape optimization for robust material model calibration, Advanced Modeling and Simulation in Engineering Sciences 9 (1) (2022) 4. doi:10.1186/ s40323-022-00217-9. URLhttps://doi.org/10.1186/s40323-022-00217-9

    work page doi:10.1186/s40323-022-00217-9 2022

  40. [40]

    Souto, A

    N. Souto, A. Andrade-Campos, S. Thuillier, A numerical methodology to design heterogeneous mechanical tests, International Journal of Mechanical Sciences 107 (2016) 264–276. doi:10.1016/j.ijmecsci.2016.01.021. URLhttps://www.sciencedirect.com/science/article/pii/S0020740316000278

    work page doi:10.1016/j.ijmecsci.2016.01.021 2016

  41. [41]

    Andrade-Campos, J

    A. Andrade-Campos, J. Aquino, J. M. Martins, B. Coelho, On the design of innovative heterogeneous sheet metal tests using a shape optimization approach, Metals 9 (3) (2019) 371

    2019

  42. [42]

    R. C. Ihuaenyi, J. Luo, W. Li, J. Zhu, Seeking the most informative design of test specimens for learning constitutive models, Extreme Mechanics Letters 69 (2024) 102169.doi:10.1016/j.eml.2024.102169. URLhttps://www.sciencedirect.com/science/article/pii/S235243162400049X

    arXiv 2024

  43. [43]

    C.-H. Tung, J. Li, The anti-dogbone: Evaluating and designing optimal tensile specimens for deep learning of constitutive relations, Extreme Mechanics Letters 69 (2024) 102157.doi:10.1016/j.eml.2024.102157. URLhttps://www.sciencedirect.com/science/article/pii/S2352431624000373

    arXiv 2024

  44. [44]

    R. C. Ihuaenyi, W. Li, M. Z. Bazant, J. Zhu, Mechanics informatics: A paradigm for efficiently learning constitutive models, Journal of the Mechanics and Physics of Solids 203 (2025) 106239. doi:10.1016/j.jmps. 2025.106239. URLhttps://www.sciencedirect.com/science/article/pii/S0022509625002157

    arXiv 2025

  45. [45]

    D. E. Ricciardi, D. T. Seidl, B. T. Lester, A. R. Jones, E. M. C. Jones, Bayesian optimal experimental design for constitutive model calibration, International Journal of Mechanical Sciences 265 (2024) 108881. doi: 10.1016/j.ijmecsci.2023.108881. URLhttps://www.sciencedirect.com/science/article/pii/S002074032300783X

    arXiv 2024

  46. [46]

    M. P. Bendsøe, Optimal shape design as a material distribution problem, Structural optimization 1 (4) (1989) 193–202.doi:10.1007/BF01650949. URLhttps://doi.org/10.1007/BF01650949

    doi:10.1007/bf01650949 1989

  47. [47]

    M. P. Bendsøe, O. Sigmund, Topology Optimization, Springer, Berlin, Heidelberg, 2004. doi:10.1007/ 978-3-662-05086-6. URLhttp://link.springer.com/10.1007/978-3-662-05086-6

    doi:10.1007/978-3-662-05086-6 2004

  48. [48]

    Chamoin, C

    L. Chamoin, C. Jailin, M. Diaz, L. Quesada, Coupling between topology optimization and digital image correlation for the design of specimen dedicated to selected material parameters identification, International Journal of Solids and Structures 193-194 (2020) 270–286.doi:10.1016/j.ijsolstr.2020.02.032. URLhttps://www.sciencedirect.com/science/article/pii/...

    doi:10.1016/j.ijsolstr.2020.02.032 2020

  49. [49]

    Barroqueiro, A

    B. Barroqueiro, A. Andrade-Campos, J. Dias-de Oliveira, R. A. F. Valente, Design of mechanical heterogeneous specimens using topology optimization, International Journal of Mechanical Sciences 181 (2020) 105764. doi: 10.1016/j.ijmecsci.2020.105764. URLhttps://www.sciencedirect.com/science/article/pii/S0020740320305166

    arXiv 2020

  50. [50]

    Gonçalves, A

    M. Gonçalves, A. Andrade-Campos, S. Thuillier, On the topology design of a mechanical heterogeneous specimen using geometric and material nonlinearities, in: IOP Conference Series: Materials Science and Engineering, V ol. 1238, IOP Publishing, 2022, p. 012055. 24 APREPRINT- DECEMBER18, 2025

    2022

  51. [51]

    Gonçalves, A

    M. Gonçalves, A. Andrade-Campos, B. Barroqueiro, On the design of mechanical heterogeneous specimens using multilevel topology optimization, Advances in Engineering Software 175 (2023) 103314. doi:10.1016/j. advengsoft.2022.103314. URLhttps://www.sciencedirect.com/science/article/pii/S0965997822002150

    arXiv 2023

  52. [52]

    Ghouli, M

    S. Ghouli, M. Flaschel, S. Kumar, L. De Lorenzis, A topology optimisation framework to design test specimens for one-shot identification or discovery of material models, Journal of the Mechanics and Physics of Solids 203 (2025) 106210.doi:10.1016/j.jmps.2025.106210. URLhttps://www.sciencedirect.com/science/article/pii/S0022509625001863

    arXiv 2025

  53. [53]

    K. Yuge, N. Kikuchi, Optimization of a frame structure subjected to a plastic deformation, Structural optimization 10 (3) (1995) 197–208.doi:10.1007/BF01742592. URLhttps://doi.org/10.1007/BF01742592

    work page doi:10.1007/bf01742592 1995

  54. [54]

    Maute, S

    K. Maute, S. Schwarz, E. Ramm, Adaptive topology optimization of elastoplastic structures, Structural optimiza- tion 15 (2) (1998) 81–91.doi:10.1007/BF01278493. URLhttps://doi.org/10.1007/BF01278493

    work page doi:10.1007/bf01278493 1998

  55. [55]

    O. Amir, Stress-constrained continuum topology optimization: a new approach based on elasto-plasticity, Struc- tural and Multidisciplinary Optimization 55 (5) (2017) 1797–1818.doi:10.1007/s00158-016-1618-8. URLhttps://doi.org/10.1007/s00158-016-1618-8

    work page doi:10.1007/s00158-016-1618-8 2017

  56. [56]

    Alberdi, G

    R. Alberdi, G. Zhang, L. Li, K. Khandelwal, A unified framework for nonlinear path-dependent sensitivity analysis in topology optimization, International Journal for Numerical Methods in Engineering 115 (1) (2018) 1–56, _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/nme.5794.doi:10.1002/nme.5794. URLhttps://onlinelibrary.wiley.com/doi/abs/10.100...

    work page doi:10.1002/nme.5794.doi:10.1002/nme.5794 2018

  57. [57]

    C. C. Margossian, A review of automatic differentiation and its efficient implemen- tation, WIREs Data Mining and Knowledge Discovery 9 (4) (2019) e1305, _eprint: https://wires.onlinelibrary.wiley.com/doi/pdf/10.1002/widm.1305.doi:10.1002/widm.1305. URLhttps://onlinelibrary.wiley.com/doi/abs/10.1002/widm.1305

    work page doi:10.1002/widm.1305.doi:10.1002/widm.1305 2019

  58. [58]

    Y . Jia, W. Li, X. S. Zhang, Multimaterial topology optimization of elastoplastic composite structures, Journal of the Mechanics and Physics of Solids 196 (2025) 106018.doi:10.1016/j.jmps.2024.106018. URLhttps://www.sciencedirect.com/science/article/pii/S0022509624004848

    arXiv 2025

  59. [59]

    Y . Jia, X. S. Zhang, Multimaterial topology optimization for finite strain elastoplasticity: theory, methods, and applications, arXiv:2502.02052 [cs] (Feb. 2025).doi:10.48550/arXiv.2502.02052. URLhttp://arxiv.org/abs/2502.02052

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2502.02052 2025

  60. [60]

    Bradbury, R

    J. Bradbury, R. Frostig, P. Hawkins, M. J. Johnson, C. Leary, D. Maclaurin, G. Necula, A. Paszke, J. VanderPlas, S. Wanderman-Milne, Q. Zhang, JAX: composable transformations of Python+NumPy programs (2018). URLhttp://github.com/jax-ml/jax

    2018

  61. [61]

    Svanberg, The method of moving asymptotes—a new method for structural optimization, International journal for numerical methods in engineering 24 (2) (1987) 359–373

    K. Svanberg, The method of moving asymptotes—a new method for structural optimization, International journal for numerical methods in engineering 24 (2) (1987) 359–373

    1987

  62. [62]

    B. Bourdin, Filters in topology optimization, International Journal for Numerical Methods in Engineering 50 (9) (2001) 2143–2158, _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/nme.116.doi:10.1002/nme.116. URLhttps://onlinelibrary.wiley.com/doi/abs/10.1002/nme.116

    work page doi:10.1002/nme.116.doi:10.1002/nme.116 2001

  63. [63]

    S. Xu, Y . Cai, G. Cheng, V olume preserving nonlinear density filter based on heaviside functions, Structural and Multidisciplinary Optimization 41 (4) (2010) 495–505.doi:10.1007/s00158-009-0452-7. URLhttps://doi.org/10.1007/s00158-009-0452-7

    work page doi:10.1007/s00158-009-0452-7 2010

  64. [64]

    Sigmund, On benchmarking and good scientific practise in topology optimization, Structural and Multidisci- plinary Optimization 65 (11) (2022) 315.doi:10.1007/s00158-022-03427-2

    O. Sigmund, On benchmarking and good scientific practise in topology optimization, Structural and Multidisci- plinary Optimization 65 (11) (2022) 315.doi:10.1007/s00158-022-03427-2. URLhttps://doi.org/10.1007/s00158-022-03427-2

    doi:10.1007/s00158-022-03427-2 2022

  65. [65]

    Chung, C

    J. Chung, C. Gulcehre, K. Cho, Y . Bengio, Gated Feedback Recurrent Neural Networks, in: Proceedings of the 32nd International Conference on Machine Learning, PMLR, 2015, pp. 2067–2075, iSSN: 1938-7228. URLhttps://proceedings.mlr.press/v37/chung15.html

    2015

  66. [66]

    Y . Lou, C. Zhang, S. Zhang, J. W. Yoon, A general yield function with differential and anisotropic hardening for strength modelling under various stress states with non-associated flow rule, International Journal of Plasticity 158 (2022) 103414.doi:10.1016/j.ijplas.2022.103414. URLhttps://www.sciencedirect.com/science/article/pii/S0749641922001929 25 APR...

    arXiv 2022

  67. [67]

    K. Li, D. Persaud, K. Choudhary, B. DeCost, M. Greenwood, J. Hattrick-Simpers, Exploiting redundancy in large materials datasets for efficient machine learning with less data, Nature Communications 14 (1) (2023) 7283, publisher: Nature Publishing Group.doi:10.1038/s41467-023-42992-y. URLhttps://www.nature.com/articles/s41467-023-42992-y

    work page doi:10.1038/s41467-023-42992-y 2023

  68. [68]

    Sigmund, Morphology-based black and white filters for topology optimization, Structural and Multidisciplinary Optimization 33 (4) (2007) 401–424

    O. Sigmund, Morphology-based black and white filters for topology optimization, Structural and Multidisciplinary Optimization 33 (4) (2007) 401–424. 26

    2007