pith. sign in

arxiv: 2307.10161 · v1 · submitted 2023-07-19 · ❄️ cond-mat.other

Variability estimation in a non-linear crack growth simulation model with controlled parameters using Designed Experiments testing

Seungju Yeoa , Paul Funkenbuscha , Hesam Askari This is my paper

Pith reviewed 2026-05-24 07:43 UTC · model grok-4.3

classification ❄️ cond-mat.other
keywords variability estimationtolerance designdesigned experimentscrack growth simulationpropagation of errorsmonte carlo methodnonlinear finite element model
0
0 comments X

The pith

Tolerance Design estimates variability in nonlinear crack growth models accurately with moderate numbers of designed experiments.

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

The paper compares three methods for estimating how uncertainty in multiple input parameters propagates to overall system response. Propagation of Errors becomes inaccurate once input coefficient of variation exceeds five percent, while Monte Carlo requires very large numbers of trials for complex models. Tolerance Design, implemented through designed experiments, is tested on both a linear three-point bending beam and a nonlinear extended-finite-element crack-growth model. When benchmarked against a 10,000-trial Monte Carlo reference, Tolerance Design produces reliable variability estimates from far fewer runs in both cases. The work shows how these techniques behave inside deterministic numerical simulations and supplies a practical performance guideline for planning physical experiments.

Core claim

The Tolerance Design method works very well with moderately sized trials of designed experiment for estimating variability in both the linear elastic 3-point bending beam model and the nonlinear extended finite elements crack growth model, while the Propagation of Errors method works suboptimally for coefficients of variance above 5 percent in the input variables.

What carries the argument

Tolerance Design method, which applies designed experiments to quantify how input-parameter variability affects system response.

If this is right

  • Tolerance Design supplies usable variability estimates for both linear and nonlinear simulation models using only moderate numbers of trials.
  • Propagation of Errors loses accuracy once input coefficients of variation rise above five percent.
  • The performance comparison supplies a concrete guideline for choosing variability-estimation methods when planning physical tests from simulation results.

Where Pith is reading between the lines

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

  • Tolerance Design could be inserted into automated design-optimization loops to enforce robustness without prohibitive run counts.
  • The same designed-experiment approach might scale to models containing dozens of uncertain parameters where full Monte Carlo becomes prohibitive.
  • Direct comparison of Tolerance Design against analytic sensitivity methods on the identical crack-growth model would clarify when each is preferable.

Load-bearing premise

A Monte Carlo run of 10,000 trials supplies an accurate reference value for the true output variability.

What would settle it

A Monte Carlo simulation run with 100,000 trials that produces variability statistics differing substantially from those obtained by Tolerance Design on the same crack-growth model.

Figures

Figures reproduced from arXiv: 2307.10161 by Hesam Askari, Paul Funkenbuscha, Seungju Yeoa.

Figure 1
Figure 1. Figure 1: The model setting illustrates a 3-point bending setup of a linear elastic Timoshenko beam, fixed on one end, and simply supported on the other. The beam has a square cross-section with significant dimensions and properties shown. The maximum vertical displacement, 𝑣, is chosen to be the system response. The vertical displacement is a function of six parameters according to the analytical solution provided … view at source ↗
Figure 3
Figure 3. Figure 3: (a) X-Y plane view of Finite Element Model of aluminum plate (b) Deformed model with a scale factor of 100 in y direction. (c) Perspective view of the plate. (d) Model view of the plate with the boundary conditions. The fixed constraint is created at the initial step at the bottom edge of the plate. Displacement of 0.1167 mm to positive y direction is applied at the top left half of the edge, created at th… view at source ↗
read the original abstract

Variability in multiple independent input parameters makes it difficult to estimate the resultant variability in the system's overall response. The Propagation of Errors and Monte-Carlo techniques are two major methods to predict the variability of a system. However, in the former method, the formalism can lead to an inaccurate estimate for systems that have parameters varying over a wide range. For the latter, the results give a direct estimate of the variance of the response, but for complex systems with many parameters, the number of trials necessary to yield an accurate estimate can be very large to the point the technique becomes impractical. In this study, the effectiveness of the Tolerance Design method to estimate variability in complex systems is studied. We use a linear elastic 3 point bending beam model and a nonlinear extended finite elements crack growth model to test and compare the PE and MC methods with the TD method. Results from an MC estimate, using 10,000 trials, serve as a reference to validate the result in both cases. We find that the PE method works suboptimal for a coefficient of variance above 5% in the input variables. In addition, we find that the TD method works very well with moderately sized trials of designed experiment for both models. Our results demonstrate how the variability estimation methods perform in the deterministic domain of numerical simulations and can assist in designing physical tests by providing a guideline performance measure.

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 / 0 minor

Summary. The manuscript compares Propagation of Errors (PE), Monte Carlo (MC with 10,000 trials as benchmark), and Tolerance Design (TD) methods for estimating output variability arising from multiple input parameters. It applies the methods to a linear elastic 3-point bending beam and a nonlinear XFEM crack-growth model, concluding that PE becomes suboptimal for input coefficients of variation above 5% while TD performs well with moderately sized designed-experiment trials for both models.

Significance. If the central validation holds, the work supplies a concrete, simulation-based demonstration that TD can serve as a computationally lighter alternative to MC for variability quantification in deterministic mechanics models, offering guidelines that could inform the design of physical experiments in fracture mechanics and related fields.

major comments (2)
  1. [Abstract] Abstract: the claim that the 10,000-trial MC estimate constitutes an accurate reference benchmark for both models is load-bearing for all performance comparisons, yet the text provides no convergence diagnostics, bootstrap error estimates, or subsample-stability tests; this is especially problematic for the nonlinear XFEM case where crack-length jumps can produce heavy-tailed or thresholded response distributions whose sample variance may retain appreciable Monte-Carlo error at 10k draws.
  2. [Abstract] Abstract: the reported superiority of TD over PE rests on concrete performance differences, but the manuscript supplies neither the exact design matrices employed in the TD trials, the explicit error-propagation formulas used for PE, nor the statistical tests used to quantify agreement with the MC reference, preventing independent assessment of the claimed accuracy.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful review and constructive comments. We address each major comment below and will revise the manuscript accordingly to strengthen the validation and reproducibility of the results.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that the 10,000-trial MC estimate constitutes an accurate reference benchmark for both models is load-bearing for all performance comparisons, yet the text provides no convergence diagnostics, bootstrap error estimates, or subsample-stability tests; this is especially problematic for the nonlinear XFEM case where crack-length jumps can produce heavy-tailed or thresholded response distributions whose sample variance may retain appreciable Monte-Carlo error at 10k draws.

    Authors: We agree that convergence diagnostics are important to support the MC benchmark, particularly for the nonlinear XFEM model. In the revised manuscript we will add bootstrap standard-error estimates on the sample variance from the 10,000 trials and subsample-stability plots (e.g., variance versus number of draws) for both the linear beam and XFEM cases. These additions will directly address the concern about possible heavy-tailed behavior in crack-length outputs. revision: yes

  2. Referee: [Abstract] Abstract: the reported superiority of TD over PE rests on concrete performance differences, but the manuscript supplies neither the exact design matrices employed in the TD trials, the explicit error-propagation formulas used for PE, nor the statistical tests used to quantify agreement with the MC reference, preventing independent assessment of the claimed accuracy.

    Authors: We acknowledge that the current manuscript does not provide the full design matrices, the explicit PE formulas, or the precise statistical comparison metrics. In the revision we will include (i) the exact orthogonal-array design matrices and factor levels used for the TD experiments, (ii) the full first-order Propagation-of-Errors expressions applied to each model, and (iii) the quantitative agreement measures (relative error and mean-absolute-percentage deviation) employed against the MC reference. These details will be placed in a new supplementary section or expanded methods subsection. revision: yes

Circularity Check

0 steps flagged

No circularity; MC benchmark is independent external reference

full rationale

The paper is an empirical comparison study of three variability estimation methods (Propagation of Errors, Monte Carlo, Tolerance Design) on two simulation models. It explicitly treats the separate 10,000-trial MC run as an external benchmark reference against which PE and TD results are validated, rather than deriving the benchmark from the other methods or fitting parameters that are then renamed as predictions. No self-definitional equations, fitted-input predictions, self-citation load-bearing steps, or ansatz smuggling appear in the abstract or described chain. The central claim (TD performs well on moderately sized designed experiments) rests on direct numerical comparison to an independent simulation benchmark and is therefore self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim depends on two standard domain assumptions: that the chosen simulation models faithfully represent the physics and that input parameters can be treated as varying independently. No free parameters or invented entities are introduced.

axioms (2)
  • domain assumption The linear elastic 3-point bending beam and nonlinear XFEM crack-growth models accurately capture the deterministic response of the physical systems under study.
    Invoked when MC results are used as reference truth for both models.
  • domain assumption Input parameters vary independently over the tested ranges.
    Required for the Propagation of Errors formalism and for the design of the Tolerance Design experiments.

pith-pipeline@v0.9.0 · 5781 in / 1465 out tokens · 28939 ms · 2026-05-24T07:43:43.153273+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

19 extracted references · 19 canonical work pages

  1. [1]

    The opened l ength of the crack along the long edge is a function of many unknown parameters. Among those, seven parameters are assumed to have a significant level of uncertainty in their value that affects the final crack opening displacement: the length of the alumin um plate (L), initial depth of the crack (d), offset location of the crack (m), Poisson...

    work page 2021

  2. [2]

    Ch. 23. Tolerancing — Approaches and related issues in industry,

    T. S. Arthanari, “Ch. 23. Tolerancing — Approaches and related issues in industry,” in Handbook of Statistics, vol. 22, Elsevier, 2003, pp. 823–865. doi: 10.1016/S0169-7161(03)22025-3

    work page doi:10.1016/s0169-7161(03)22025-3 2003

  3. [3]

    Kleiber, The stochastic finite element method: basic perturbation technique and computer implementation / Michael Kleiber, Tran Duong Hien

    M. Kleiber, The stochastic finite element method: basic perturbation technique and computer implementation / Michael Kleiber, Tran Duong Hien. Chichester [England] ; Wiley, 1992

    work page 1992

  4. [4]

    Worst case optimization: a method to account for uncertainties in the optimization of intensity modulated proton therapy,

    D. Pflugfelder, J. J. Wilkens, and U. Oelfke, “Worst case optimization: a method to account for uncertainties in the optimization of intensity modulated proton therapy,” Phys. Med. Biol., vol. 53, no. 6, p. 1689, Feb. 2008, doi: 10.1088/0031-9155/53/6/013

    work page doi:10.1088/0031-9155/53/6/013 2008

  5. [5]

    ASME Measurement Uncertainty,

    R. B. Abernethy, R. P. Benedict, and R. B. Dowdell, “ASME Measurement Uncertainty,” Journal of Fluids Engineering, vol. 107, no. 2, pp. 161–164, Jun. 1985, doi: 10.1115/1.3242450

    work page doi:10.1115/1.3242450 1985

  6. [6]

    S. Huet, A. Bouvier, M.-A. Poursat, and E. Jolivet, Statistical Tools for Nonlinear Regression. New York: Springer-Verlag, 2004. doi: 10.1007/b97288

    work page doi:10.1007/b97288 2004

  7. [7]

    H. H. Ku, Notes on the use of propagation of error formulas. National Bureau of Standards, 1966. Accessed: Sep. 06, 2022. [Online]. Available: http://archive.org/details/jresv70Cn4p263

    work page 1966

  8. [8]

    Estimation and propagation of error in measurements of river channel movement from aerial imagery,

    N. Mount and J. Louis, “Estimation and propagation of error in measurements of river channel movement from aerial imagery,” Wiley InterScience, vol. 30, pp. 635–643, Apr. 2005, doi: 10.1002/esp.1172

    work page doi:10.1002/esp.1172 2005

  9. [9]

    Error propagation of DEM-based surface derivatives,

    J. Oksanen and T. Sarjakoski, “Error propagation of DEM-based surface derivatives,” Computers & Geosciences, vol. 31, no. 8, pp. 1015–1027, Oct. 2005, doi: 10.1016/j.cageo.2005.02.014

    work page doi:10.1016/j.cageo.2005.02.014 2005

  10. [10]

    Review of statistical approaches to tolerance analysis,

    S. D. Nigam and J. U. Turner, “Review of statistical approaches to tolerance analysis,” Computer- Aided Design, vol. 27, no. 1, pp. 6–15, Jan. 1995, doi: 10.1016/0010-4485(95)90748-5

    work page doi:10.1016/0010-4485(95)90748-5 1995

  11. [11]

    Timoshenko versus Euler beam theory: Pitfalls of a deterministic approach,

    A. T. Beck and C. R. A. da Silva, “Timoshenko versus Euler beam theory: Pitfalls of a deterministic approach,” Structural Safety, vol. 33, no. 1, pp. 19–25, Jan. 2011, doi: 10.1016/j.strusafe.2010.04.006. 15

    work page doi:10.1016/j.strusafe.2010.04.006 2011

  12. [12]

    M. S. Phadke, Quality engineering using rubust design / Madhav S. Phadke. Englewood Cliffs, N.J. : Prentice Hall

    work page

  13. [13]

    Statistical Tolerancing Using a Modification of Taguchi’s Method,

    J. R. D’Errico and N. A. Zaino, “Statistical Tolerancing Using a Modification of Taguchi’s Method,” Technometrics, vol. 30, no. 4, pp. 397–405, Nov. 1988, doi: 10.1080/00401706.1988.10488434

    work page doi:10.1080/00401706.1988.10488434 1988

  14. [14]

    Tolerance Design,

    G. Taguchi, S. Chowdhury, and Y. Wu, “Tolerance Design,” in Taguchi’s Quality Engineering Handbook, John Wiley & Sons, Ltd, 2004, pp. 340–351. doi: 10.1002/9780470258354.ch16

    work page doi:10.1002/9780470258354.ch16 2004

  15. [15]

    Probabilistic finite element prediction of knee wear simulator mechanics,

    P. J. Laz, S. Pal, J. P. Halloran, A. J. Petrella, and P. J. Rullkoetter, “Probabilistic finite element prediction of knee wear simulator mechanics,” J Biomech, vol. 39, no. 12, pp. 2303–2310, 2006, doi: 10.1016/j.jbiomech.2005.07.029

    work page doi:10.1016/j.jbiomech.2005.07.029 2006

  16. [16]

    Designing Experiments for Tolerancing Assembled Products,

    S. Bisgaard, “Designing Experiments for Tolerancing Assembled Products,” Technometrics, vol. 39, no. 2, pp. 142–152, 1997, doi: 10.2307/1270902

    work page doi:10.2307/1270902 1997

  17. [17]

    Standard errors: A review and evaluation of standard error estimators using Monte Carlo simulations,

    B. Harding, C. Tremblay, and D. Cousineau, “Standard errors: A review and evaluation of standard error estimators using Monte Carlo simulations,” The Quantitative Methods for Psychology, vol. 10, no. 2, pp. 107–123, 2014, doi: 10.20982/tqmp.10.2.p107

    work page doi:10.20982/tqmp.10.2.p107 2014

  18. [18]

    P. D. Funkenbusch, Practical Guide To Designed Experiments: A Unified Modular Approach. Boca Raton: CRC Press, 2004. doi: 10.4324/9780203997314

    work page doi:10.4324/9780203997314 2004

  19. [19]

    J. M. Gere and B. J. Goodno, Mechanics of Materials. Cengage Learning, Stamford, Conn, 2009

    work page 2009