Pith. sign in

REVIEW 4 major objections 7 minor 20 references

Model-based reinforcement corrosion prediction: Continuous calibration with Bayesian optimization and corrosion wire sensor data

T0 review · 4 major / 7 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Progressive calibration of a physics-based chloride diffusion model with Bayesian optimization, using wire-sensor break times, yields increasingly accurate predictions of chloride penetration depth and effective diffusion coefficients…

desk verdict A competent calibration study whose 'increasingly accurate predictions' claim is really about fitting the same four data points, not predicting new ones. read the letter →

arxiv 2411.16447 v1 pith:PWWCIAVZ submitted 2024-11-25 cs.CE

A. Potnis , M. Macier , T. Leusmann , D. Anton , H. Wessels , D. Lowke This is my paper · ORCID
classification cs.CE
keywords chloride-inducedcorrosioneffectivediffusioncoefficientBayesianoptimizationwiresensormodelcalibrationconcretedurabilitychloridemigrationneuralnetworksurrogate
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 asks whether a physics-based chloride diffusion model can be kept accurate over time by recalibrating it every time a cheap wire sensor in the concrete breaks. It shows that, as more wire-break events are added to a Bayesian optimization routine, the model's predicted chloride penetration depths converge to the observed ones, and the fitted effective diffusion coefficients land in the same range as values obtained from independent drilling-dust chloride profiles. In contrast, a neural network trained on the same four events overfits and cannot generalize, although it reproduces the calibration points. The practical payoff would be a monitoring system where corrosion risk is updated continuously from simple resistance measurements, reducing the need for destructive sampling until visible damage appears.

What carries the argument

The central object is the analytical chloride diffusion model of reference [1], which gives the chloride concentration profile $C(x,t)$ as an error-function expression with an effective diffusion coefficient that depends on time and temperature. By inverting the error function, the model becomes a deterministic map from the parameters (aging exponent $a$, temperature sensitivity $b_e$, and migration-coefficient factor $D_t$) to the depth at which the critical chloride content is reached at a given sensor time. Bayesian optimization—a Gaussian-process surrogate with expected-improvement acquisition—finds the parameter triple that minimizes the squared error between calculated and observed depths. The wire sensor supplies the observed depths and times: each resistance jump is converted into a calibration point by assuming the wire began corroding 24 days earlier at a chloride concentration of 0.6 M% by cement mass, equivalent to 1.62 kg/m$^3$.

What would settle it

Run the same optimization while varying the assumed corrosion-start lag (for example, ±7 and ±30 days) and the critical chloride concentration (0.2 to 2 M%) and observe whether the fitted diffusion coefficients remain within the drilling-dust values; alternatively, place additional wires at the same depths on a new exposure and compare predicted versus measured break times without re-fitting.

Watch

Extended reading notes

Core claim

The paper demonstrates that progressively calibrating a physics-based chloride diffusion model with wire-sensor break times via Bayesian optimization yields increasingly accurate predictions of chloride penetration depth and effective diffusion coefficients. Starting from the analytical solution of Fick's second law for chloride ingress, the authors isolate the penetration depth expression via the inverse error function and treat the measured (depth, time) wire-break pairs as data for fitting the parameters $a$, $b_e$, and $D_t$ in the effective diffusion coefficient. They use Bayesian optimization with a Gaussian-process surrogate and expected-improvement acquisition to find the parameter triple that minimizes the mean squared error in predicted depth. Calibrating progressively with one, two, three, and four wire-break events makes the predicted time-versus-depth curve align ever more closely with the observed events, and the effective diffusion coefficients extrapolated to the sampling dates agree with the drilling-dust values ($0.48 \times 10^{-12}$ m$^2$/s at 509 days, $0.12 \times 10^{-12}$ m$^2$/s at 6692 days) when the critical chloride content is taken at the lower bound of its uncertainty range. The authors conclude that the progressive calibration of the model, paired with Bayesian optimization, leads to increasingly accurate predictions.

Load-bearing premise

The whole calibration rests on the assumption that a resistance jump implies the wire began to corrode exactly 24 days earlier, when the chloride concentration at the wire depth was 0.6 M% by cement mass; if the corrosion-start lag or that critical concentration is wrong, every fitted parameter and every subsequent prediction shifts.

Editorial extensions

If this is right

  • If the calibration is correct, a concrete structure's corrosion risk can be updated continuously from resistance measurements alone, without destructive dust sampling.
  • The progressive calibration procedure makes model-predicted penetration depths trustworthy only after the first few sensor events; earlier predictions are less reliable.
  • Uncertainty in critical chloride content can be propagated through the model to give a band of predicted corrosion times, improving maintenance planning.
  • The comparison shows that the physics-based model needs less data than the flexible neural network to generalize, and the two approaches could be combined to exploit both interpretability and flexibility.

Reading between the lines

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

  • The 24-day lag and the 0.6 M% critical concentration are the main assumptions that convert resistance jumps into calibration pairs; replacing them with direct electrochemical depassivation measurements would make the method transferable to other structures without re-estimating these offsets.
  • Because the neural network overfits with only four data points, adding more wire sensors at intermediate depths would allow a meaningful test of whether the data-driven approach catches up to the physics-based model as data grow.
  • The model assumes one-dimensional, homogeneous diffusion; extending the calibration to cracked or layered concrete with a numerical solver would check whether the fitted parameters remain physically stable.
  • A direct cross-validation on a second exposure cycle would test whether the calibrated diffusion coefficients predict the timing of new sensor breaks before they occur.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. The paper calibrates two models for chloride ingress into concrete using wire-sensor breakage times from the 'Concerto' test bridge: a physics-based Gehlen diffusion model with three parameters (a, Dt, be) estimated by Bayesian optimization, and a feed-forward neural network that maps time and temperature to an effective diffusion coefficient. The authors claim that progressive calibration with additional wire-sensor data yields increasingly accurate predictions of chloride penetration depth and that the calibrated diffusion coefficients agree with independent drilling-dust measurements. The paper also includes a Sobol sensitivity analysis, an optimizer sanity check, and a comparison against a rapid chloride migration (RCM) test.

Significance. If the progressive-calibration claim were established, the work would be practically valuable for non-destructive corrosion monitoring of concrete infrastructure. The manuscript has genuine strengths: it uses multi-year real sensor data from a near-real-scale bridge, presents a transparent physics-based model, checks the Bayesian optimizer on synthetic data (Appendix C.1), and honestly acknowledges that the neural network is overfitted with four training points (Section 5.2). The commitment to publish data and code on Zenodo is also a positive feature. However, the central evidence for 'increasingly accurate predictions' is currently in-sample calibration, and the only independent comparison is sparse and not quantitatively evaluated. These load-bearing issues prevent the paper from supporting its main claim in its present form.

major comments (4)
  1. [Section 5.1, Eq. (10), Table 4, Figures 6-7] The demonstration of 'increasingly accurate predictions' is an in-sample calibration result, not a prediction on held-out data. In Table 4, the listed x_calc values equal the x_exp values by construction, because the optimization objective in Eq. (10) is the mean squared error over those exact depth-time pairs. Figures 6 and 7 plot the fitted model against the same points used for fitting. This does not support generalization to new wire depths or later exposure times. To substantiate the progressive-calibration claim, the authors should perform a genuine out-of-sample evaluation, such as leave-one-out calibration over the four points, or predict the deepest wire (0.03 m) using only the first three calibration points, and report quantitative predictive errors.
  2. [Section 5.4, Figure 13] The only non-circular validation is the comparison with drilling-dust-derived effective diffusion coefficients, but the text merely asserts 'good agreement' without a numerical tolerance or uncertainty band. Using the four-point calibrated parameters from Table 4 (a = 0.16, Dt = 1.29e-12 m2/s, be = 1010.7 K) in Eq. (4) gives DEff,C values on the order of 0.9e-12 and 0.6e-12 m2/s at t = 509 and 6692 days, respectively, whereas the chloride-profile regressions in Section 5.4 report 0.48e-12 and 0.12e-12 m2/s. The later value is therefore overestimated by roughly a factor of five. The paper should either quantify the discrepancy, provide confidence intervals from the parameter uncertainties, or explicitly state whether the discrepancy is expected within the CCrit uncertainty range shown in Figure 8.
  3. [Section 5.1, Table 4] The calibrated parameters are strongly unstable across the progressive-calibration steps: a varies from 0.10 to 0.78 and Dt from 1.18e-12 to 8.69e-12 m2/s. With only four data points and three free parameters, this indicates weak identifiability. This instability undermines the physical interpretation of the calibrated Dt and weakens the reliability claim made in Section 5.1. The authors should report parameter uncertainties or posterior ranges (for example, using the GP surrogate in the Bayesian optimizer), and they could implement the dimensionality reduction they propose in Section 3.1 by fixing be, then check whether the remaining parameters and predictions are stable.
  4. [Section 3, Assumption 2] The conversion of each resistance jump into a calibration point is load-bearing: the paper assumes that corrosion started exactly 24 days before the observed wire break and that at that earlier time the chloride concentration at the wire depth equaled 0.6 M% (1.62 kg/m3). This assumption turns every resistance jump into a depth-time observation used in Eq. (10). No sensitivity analysis or validation of the 24-day offset is provided, and the offset is not treated as an uncertain quantity. The authors should justify the offset with a quantitative reference or include it as an additional calibration parameter (or at least perturb it over a plausible range and show the effect on the fitted parameters and on the Section 5.4 comparison).
minor comments (7)
  1. [Section 3, Eq. (6)] The cosine argument in Eq. (6) is ambiguous because of the commas in the large numbers and the missing parentheses: '2π · t + 2, 542, 453.44 / 32, 407, 303.30' could be read as (2πt + 2,542,453.44) / 32,407,303.30 or as 2πt + (2,542,453.44 / 32,407,303.30). Please rewrite with clear parentheses and without thousands separators inside the mathematical expression.
  2. [Appendix C.1, Table 6] The text states that the true value of Dt is '2e-11', but Table 6 lists the true value as '2 × 10−12'. These are inconsistent and should be reconciled.
  3. [Section 1.2, Section 3] The wire is described as having a 'cross section area' of 0.065 mm, but mm is a unit of length, not area. Please specify either a diameter of 0.065 mm or give the cross-sectional area in mm2 (e.g., π·(0.0325 mm)2).
  4. [General formatting] Several table titles contain the typo 'T able' instead of 'Table' (e.g., Tables 1-3), and the heading 'Calibration using one Data pair' in Table 4 should read 'one data point'.
  5. [Section 5.2, Algorithm 2] The neural network training details are incomplete: the number of epochs, learning rate, and final loss value are not reported. Since the paper acknowledges overfitting, these details would help the reader judge the convergence claim.
  6. [Section 4] The description of fitting DEff,C from the chloride profiles says 'adjusting the calculated chloride concentration CS,Δx to the assumed chloride profile', which is unclear; it should clarify which parameters are adjusted in the nonlinear regression and how the uppermost layers are excluded.
  7. [References] Reference [11] is cited as 'DIN EN 12390-11:2015-11' but the standard number is 'DIN EN 12390-11'; also reference [6] is listed as a BAW data sheet without a full title or year, making it difficult to locate.

Circularity Check

1 steps flagged · score 6.0 of 10

The central 'increasingly accurate predictions' claim is demonstrated on the calibration set: Table 4 and Figures 6-7 compare fitted depths to the same measured depths that Eq. (10) minimizes over, so the progressive-calibration evidence is in-sample fit, not held-out prediction; the drilling-dust comparison is external but not quantified enough to rescue the claim.

  1. fitted input called prediction [Section 3.1, Eq. (10); Section 5.1, Table 4 and Figures 6-7]
    "The aim of this optimization problem is to find the optimal parameters {a, be, Dt}∗ such that the the mean squared error (MSE) between the depths x according to 9 and the depths ˆxi of the measurement points listed in Table 2 is minimized. ... As seen in Figure 6, the calibration results show that the predicted depth aligns increasingly well with the observed data as more calibration points are added."

    Equation (10) defines the fitted parameters as the minimizers of the discrepancy between computed and measured depths at exactly the four Table 2 points. Table 4 then reports x_calc equal to x_exp for those same points (0.015/0.015, 0.02/0.02, 0.025/0.025, 0.03/0.03), and Figures 6-7 plot the calibrated curve through those same measured points. Therefore the 'increasingly accurate predictions' shown are the in-sample residuals of the fitting objective, not out-of-sample predictions at new wire depths or times; the improvement is not evidence of generalization. The central progressive-calibration claim thus reduces to the fitting construction.

full rationale

The paper contains a genuine external element: the drilling-dust chloride profiles and RCM test in Section 5.4 are independent of the wire-sensor calibration, and the Gehlen model, the 24-day corrosion-start assumption, and the critical-chloride ranges are taken from external sources [1,5], not from the present authors' own prior work. No load-bearing self-citation chain is present. However, the main claim that progressive calibration with Bayesian optimization leads to increasingly accurate predictions is supported by Table 4 and Figures 6-7, which compare the fitted model to the same four points that Eq. (10) minimizes over. Table 4 lists x_calc = x_exp at each calibration stage by construction, and Figure 7, despite the word 'validation', is a calibration plot through those points. Thus the load-bearing demonstration of the central claim is in-sample fit. The only out-of-sample content is the extrapolated prediction curve in Figure 8 and the qualitative comparison with two drilling-dust diffusion coefficients in Section 5.4, but the latter is not quantified against the calibrated model's predictions with uncertainty bands. The score is 6 because one or more 'predictions' reduce by construction to the calibration objective, while the presence of independent external data prevents full circularity.

Assumptions & free parameters 5 free parameters · 7 assumptions · 0 invented entities

The calibration pipeline rests on domain-specific diffusion models and on a number of assumptions that convert wire break times into chloride concentration events. The key numerical free parameters are the three Gehlen parameters plus the temperature fit coefficients and the neural network weights. No new physical entities are introduced.

free parameters (5)
  • a (aging exponent in Gehlen model) = ≈0.16 (four-data-point calibration); ranges from ≈0.1 to ≈0.78 across calibrations
    Fitted by Bayesian optimization to minimize depth error in Eq. 10.
  • Dt (transfer factor times migration coefficient, kt * DRCM,0) = ≈1.29e-12 m2/s (four-point calibration); ranges from ≈1.18e-12 to ≈8.69e-12
    Fitted by Bayesian optimization; represents kt*DRCM,0 in Eq. 4.
  • be (temperature regression coefficient) = ≈1010.70 K (four-point calibration); ranges from ≈1000 to ≈2054
    Fitted by Bayesian optimization; Sobol analysis shows it has low sensitivity.
  • Temperature model coefficients (offset, amplitude, phase, period) = Offset=284.39 K, amplitude=11.10 K, phase=2542453.44 s, period=32407303.30 s in Eq. 6
    Obtained by regression on temperature sensor data; used as input to both models.
  • Neural network weights and biases (all θ) = not reported, architecture 2-10-10-1 with ReLU, log-scaled output
    Trained with Adam; effectively many free parameters; acknowledged to overfit with four data points.
assumptions (7)
  • domain assumption Fick's second law with a time-dependent effective diffusion coefficient describes chloride ingress in concrete (Eq. 2 and Eq. 3).
    Invoked in Section 2 as the governing equation; standard in chloride transport modeling.
  • domain assumption The analytical error-function solution of Gehlen (Eq. 3) is valid for this concrete and exposure.
    Adopted from [1] and used for all depth calculations.
  • ad hoc to paper A wire break indicates the wire fully corroded, and corrosion initiation occurred exactly 24 days before the break, at which time the chloride content at the wire depth equaled 0.6 M% (1.62 kg/m3).
    Stated as assumptions 1 and 2 in Section 3; this converts each resistance jump into a (depth, time) calibration point. If the 24-day offset or the 0.6 M% value is wrong, all fitted parameters shift.
  • domain assumption The chloride surface concentration CS,Δx is 18.19 kg/m3 for the sensor data, derived from a 3 wt% NaCl solution.
    Computed in Appendix A; assumes the solution concentration at the surface equals the bulk sodium chloride solution concentration.
  • ad hoc to paper The internal concrete temperature follows the cosine regression in Eq. 6 with fixed coefficients.
    Used as input to both models; fitted to temperature sensor data.
  • domain assumption For the dust sample regression, Fick's law with constant D applies, and the uppermost 5 mm (2007) and 20 mm (2024) are excluded as non-Fickian zones.
    Section 4: used to derive the reference DEff,C values.
  • domain assumption The critical chloride content CCrit is uncertain within 0.54 to 5.4 kg/m3 (0.2 to 2 M%), and all values in that range are treated as equally plausible.
    Section 5.1 and Table 3; used for uncertainty bands.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Model-based reinforcement corrosion prediction: Continuous calibration with Bayesian optimization and corrosion wire sensor data." pith.science (2026). https://pith.science/paper/PWWCIAVZ

@misc{pith2026241116447,
  author       = {Pith},
  title        = {Pith review of: Model-based reinforcement corrosion prediction: Continuous calibration with Bayesian optimization and corrosion wire sensor data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PWWCIAVZ}},
  note         = {Machine review of arXiv:2411.16447}
}
read the original abstract

Chloride-induced corrosion significantly contributes to the degradation of reinforced concrete structures, making accurate predictions of chloride migration and its effects on material durability critical. This paper explores two modeling approaches to estimate the effective diffusion coefficient for chloride transport. The first approach follows Gehlen's interpretable diffusion model, which is based on established physical principles and incorporates time and temperature dependencies in predicting chloride migration. The second approach is a neural network-based method, where the neural network approximates the effective diffusion coefficient. In a subsequent step, the calibrated models are used to predict the penetration depth of the critical chloride content, taking into account the uncertainty in the critical chloride content. The models are calibrated using experimental data measured by a wire sensor installed in a concrete test bridge. The calibration results are compared to effective diffusion coefficients derived from drilling dust samples. A comparison of both approaches reveals the advantages of the physics-based model in terms of transparency and interpretability, while the neural network model demonstrates flexibility and adaptability in data-driven predictions. This study emphasizes the importance of combining traditional and machine learning-based methods to improve the accuracy of chloride migration predictions in reinforced concrete.

Figures

Figures reproduced from arXiv: 2411.16447 by the authors.

Figure 1
Figure 1. The ’Concerto’ test bridge in 2005 1.2 Investigations at ’Concerto’ and laboratory experiments In addition to the investigation of monitoring techniques for locating prestress steel fractures and the subsequent installation of RFID wire sensors [4] a major focus of the investigations carried out at ‘Concerto’ was on monitoring corrosion activity using the wire sensor developed at iBMB [5]. Localised corrosive damage… view at source ↗
Figure 2
Figure 2. Resistance jumps over time period temperature profile over time and determine the unknown parameters by regressing the approach on the data. The regression results in the following analytical equation: TIST(t) = 11.10 · cos  2π · t + 2, 542, 453.44 32, 407, 303.30  + 284.39 (6) [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Temperature vs. time with best-fit cosine curve This equation, shown in [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (14 more)
Figure 4
Figure 4. Figure 4: Measured chloride content from the drilling dust analysis or from the regression analysis as a function of depth and bottom sides of the cubes. A drill core with a diameter of 100 mm was taken from each of the two cubes, which was then divided into three samples with a…
Figure 5
Figure 5. Figure 5: Termination Criteria (a) Calibration with one data pair (b) Calibration with two data pairs (c) Calibration with three data pairs (d) Calibration with four data pairs [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: Calibration plots: 0.54 < CCrit < 5.4 (approach A) [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]
Figure 7
Figure 7. Figure 7: Calibration plot: Time vs. Depth validation (approach A) [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]
Figure 8
Figure 8. Figure 8: Calibration plot: Time vs. Depth prediction (approach A) [PITH_FULL_IMAGE:figures/full_fig_p014_8.png]
Figure 9
Figure 9. Figure 9: Calibration plots: 0.54 < CCrit < 5.4 (approach B) 3. Ease of Implementation: The process of setting up Approach A involves tuning parameters such as a, Dt, and be using Bayesian optimization. This process requires detailed knowledge of the system and careful calibrati…
Figure 10
Figure 10. Figure 10: Calibration plot: Time vs. Depth (approach B) [PITH_FULL_IMAGE:figures/full_fig_p017_10.png]
Figure 11
Figure 11. Figure 11: Calibration plot: Time vs. Depth prediction (approach B) methodology. Both Approach A and Approach B provide valuable insights into chloride migration prediction, utilizing different strategies. Approach A, with its basis in physics, emphasizes structure and interpret…
Figure 12
Figure 12. Figure 12: Comparison of chloride concentration predictions using physics-based and data-driven ap￾proaches for CCrit = 1.62 (xi, ti) True CCrit Approach A Prediction Approach B Prediction (0.015, 55194877) 1.62 1.9881 1.61981744 (0.02, 102827181) 1.62 1.4169 1.62008777 (0.025, …
Figure 13
Figure 13. Figure 13: Comparison of DEff,C from the RCM test, the wire sensor, and drilling dust samples over time 6 Conclusion This study addresses the need for accurate predictions of chloride migration in reinforced concrete struc￾tures to mitigate the risks of chloride-induced corrosio…
Figure 14
Figure 14. Figure 14: Sobol indices [PITH_FULL_IMAGE:figures/full_fig_p024_14.png]
Figure 15
Figure 15. Figure 15: Sanity Check:True vs. Predicted results Parameter True Value Optimized Value a 0.2 0.20010373848673135 Dt 2 × 10−12 2.0015664373556944 × 10−12 be 2050 2054.4906785267476 [PITH_FULL_IMAGE:figures/full_fig_p025_15.png]
Figure 16
Figure 16. Figure 16: Termination Criteria During Sanity Check C.2 Sensitivity Analysis Before conducting the Sobol sensitivity analysis on Gehlen’s model, a sanity check was performed using the dummy model to validate the methodology and ensure proper implementation. f(X) = 2X2 1 + X2 + 3…
Figure 17
Figure 17. Figure 17: Sobol indices sanity check [PITH_FULL_IMAGE:figures/full_fig_p027_17.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

20 extracted references · 19 canonical work pages

  1. [1]

    Gehlen, Probabilistische Lebensdauerbemessung von Stahlbetonbauwerken – Zu- verl¨ assigkeitsbetrachtungen zur wirksamen Vermeidung von Bewehrungskorrosion

    C. Gehlen, Probabilistische Lebensdauerbemessung von Stahlbetonbauwerken – Zu- verl¨ assigkeitsbetrachtungen zur wirksamen Vermeidung von Bewehrungskorrosion. DIN Media GmbH, 06 2000

    work page 2000

  2. [2]

    Holst, Korrosionsmonitoring und Bruchortung vorgespannter Zugglieder in Bauwerken

    A. Holst, Korrosionsmonitoring und Bruchortung vorgespannter Zugglieder in Bauwerken. PhD thesis, Institut f¨ ur Baustoffe, Massivbau und Brandschutz (IBMB), 2007

    work page 2007

  3. [3]

    Monitoring von Betonbauwerken: Sensoren, Methoden, Erfahrungen,

    Harald Budelmann, “Monitoring von Betonbauwerken: Sensoren, Methoden, Erfahrungen,” in Sicherheitsgewinn durch Monitoring?, Darmstadt: Freunde des Inst. f¨ ur Massivbau der TU, 2006

    work page 2006

  4. [4]

    Korrosionsmonitoring von stahlbetonbauwerken mit einem funkbasierten drahtsensor,

    I. Dressler, H.-J. Wichmann, and H. Budelmann, “Korrosionsmonitoring von stahlbetonbauwerken mit einem funkbasierten drahtsensor,” Bautechnik, vol. 92, no. 10, pp. 683–687, 2015

    work page 2015

  5. [5]

    Korrosionsmonitoring von stahlbetonbauwerken als element des lebensdauermanagements,

    A. Holst, H. Budelmann, and H.-J. Wichmann, “Korrosionsmonitoring von stahlbetonbauwerken als element des lebensdauermanagements,” Beton- und Stahlbetonbau, vol. 105, no. 12, pp. 536–549, 2010

    work page 2010

  6. [6]

    BA W - Merkblatt: Dauerhaftigkeitsbemessung bei Carbon- atisierung und Chlorideinwirkung (MDCC)

    Bundesanstalt f¨ ur Wasserbau - BA W, “BA W - Merkblatt: Dauerhaftigkeitsbemessung bei Carbon- atisierung und Chlorideinwirkung (MDCC).”

    work page

  7. [7]

    Chloride transport in concrete: theory and application,

    R. F. M. Bakker, “Chloride transport in concrete: theory and application,” HERON, vol. 39, no. 1, pp. 45–83, 1994

    work page 1994

  8. [8]

    Lausanne, Switzerland: CEB Bulletin No

    Comit´ e Euro-International du B´ eton (CEB) Task Group V,Durability of Concrete Structures: State- of-the-Art Report. Lausanne, Switzerland: CEB Bulletin No. 238, 1997

    work page 1997

Show all 20 references
  1. [9]

    Bayesian Optimization: Open source constrained global optimization tool for Python,

    F. Nogueira, “Bayesian Optimization: Open source constrained global optimization tool for Python,” 2014–

    2014

  2. [10]

    Goodfellow, Y

    I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning. online: MIT Press, 2016. https: //www.deeplearningbook.org

    2016

  3. [11]

    DIN EN 12390-11:2015-11, Pr¨ ufung von Festbeton - Teil 11: Bestimmung des Chloridwiderstandes von Beton - Einseitig gerichtete Diffusion; Deutsche Fassung EN 12390-11:2015

    “DIN EN 12390-11:2015-11, Pr¨ ufung von Festbeton - Teil 11: Bestimmung des Chloridwiderstandes von Beton - Einseitig gerichtete Diffusion; Deutsche Fassung EN 12390-11:2015.”

    2015

  4. [12]

    Monte carlo and quasi-monte carlo sampling

    S. U. Hai Wang Group, “Monte carlo and quasi-monte carlo sampling.” https://web.stanford. edu/group/haiwanglab/FFCM2/docs/ResponseSurface/MonteCarlo/, 2024. Accessed: 2024-11- 21. 21 A Calculations for CS,∆X The concentration of NaCl in the solution was determined based on a 3...

    2024

  5. [13]

    Define the Problem : Specify the input parameters (i.e., be, Dt, and a) and their respective distribution (refer Subsection B.1). 23

  6. [14]

    The Sobol sampling algorithm ensures that the points are distributed uniformly, covering the range for each input parameter effectively

    Generate Sobol Samples : • Sobol samples are quasi-random, low-discrepancy sequences that ensure good coverage of the input space with fewer samples than purely random sampling.[12] • For each input parameter, generate N samples and collect them in X. The Sobol sampling algori...

  7. [15]

    Evaluate the Model for Each Combination of x, t, and TIST: For each combination of x, t, and TIST, evaluate the model f (X) for all Sobol samples to obtain the output Y : Y = f (X). (20)

  8. [16]

    • Total-order Sobol indices S(j) T for be, Dt, and a, representing the total contribution of each parameter, including all interactions, for the j-th combination of data points

    Perform Sobol Analysis : For each combination j, where j corresponds to a specific combination of depth x, time t, and temperature TIST as listed in Table 2, calculate: • First-order Sobol indices S(j) 1 for be, Dt, and a, representing the independent effect of each parameter ...

  9. [17]

    • Xi represents the i-th input parameter ( be, Dt, or a)

    V alidate V ariance Decomposition: For each combination j, check if the sum of first-order Sobol indices S1 approximates the total variance V (Y ) by verifying: Variance Sum(j) = kX i=1 S(j) 1 (Xi) × V (Y (j)) ≈ V (Y (j)), (21) where: • j refers to a specific combination of x,...

  10. [18]

    • m is the total number of combinations of x, t, and TIST as defined in Table 2

    Average Sobol Indices Across All Runs : Compute the average first-order S1 and total-order ST indices across all combinations of x, t, and TIST (indexed by j) for each input parameter ( be, Dt, and a)(indexed by i): Average S1(Xi) = 1 m mX j=1 S(j) 1 (Xi), Average ST (Xi) = 1 ...

  11. [19]

    Plot Results: Plot the averaged first-order Sobol indices S1 and total-order Sobol indices ST as bar charts

  12. [20]

    Return Results: Return a dictionary containing S1 and ST averaged over all runs. The Sobol sensitivity analysis revealed that the first-order as well as the total-order Sobol index for a was the highest, followed by Dt, with the least contribution by be, confirming its dominan...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.