pith. machine review for the scientific record. sign in

arxiv: 2604.25690 · v1 · submitted 2026-04-28 · 💻 cs.CE

Recognition: unknown

Data Driven Calibration of Analytical Concrete Creep Models Considering Preloading Effects Using Gaussian Processes

Christopher Taube, Gledson Rodrigo Tondo, Guido Morgenthal, Leonie Heller

Authors on Pith no claims yet

Pith reviewed 2026-05-07 14:13 UTC · model grok-4.3

classification 💻 cs.CE
keywords concrete creepGaussian process regressionpreloading effectsmodel calibrationuncertainty quantificationdata-driven modelinganalytical modelsstructural concrete
0
0 comments X

The pith

Gaussian Process Regression improves analytical concrete creep models by incorporating preloading effects from experiments

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

The paper sets out to show that Gaussian Process Regression can adjust conventional analytical models of concrete creep so they better match data from experiments that include preloading. Preloading consists of a short-term higher load applied before the long-term sustained load, and it tends to lower both the size and the spread of creep strains. The calibration step also produces estimates of uncertainty and can indicate which future tests would add the most value. Readers would care because improved creep forecasts allow concrete structures to be designed with less material while remaining safe over decades.

Core claim

The authors demonstrate that Gaussian Process Regression can be trained on experimental creep data that records preloading intensity, timing, and concrete age in order to adjust the parameters of analytical creep models, thereby raising prediction accuracy, supplying uncertainty bounds, and clarifying how preloading reduces creep magnitude and variability.

What carries the argument

Gaussian Process Regression used to calibrate analytical creep models while taking preloading intensity, timing, and concrete age as inputs.

If this is right

  • Creep strain predictions become more accurate once preloading history is included in the model.
  • Uncertainties in the forecasts are quantified so that engineers can make informed safety decisions.
  • Future experimental programs can be planned to deliver the largest improvement in model performance.
  • The role of preloading in altering concrete microstructure is clarified, supporting more sustainable material use.

Where Pith is reading between the lines

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

  • The same calibration approach could be tested on other viscoelastic materials or on concrete under varying environmental conditions.
  • Embedding the calibrated models into structural analysis software would allow direct use in design calculations.
  • Collecting longer-duration data could test whether the observed benefits of preloading persist over service lives of decades.

Load-bearing premise

The available experimental data sets on preloading effects are representative enough that the Gaussian Process model will generalize reliably to other concrete mixes, load histories, and time spans.

What would settle it

A new series of creep tests on preloaded concrete specimens whose conditions lie outside the training data range, in which the Gaussian Process calibrated predictions show no gain in accuracy or produce uncertainty intervals that fail to contain the measured strains.

Figures

Figures reproduced from arXiv: 2604.25690 by Christopher Taube, Gledson Rodrigo Tondo, Guido Morgenthal, Leonie Heller.

Figure 1
Figure 1. Figure 1: First-order and total-order Sobol indices of the EC view at source ↗
Figure 2
Figure 2. Figure 2: Predicted creep coefficient (left) and distribution of optimised model parameters (right) based on the test data view at source ↗
Figure 3
Figure 3. Figure 3: Predicted creep coefficient (left) and distribution of optimised model parameters (right) based on the test data view at source ↗
Figure 4
Figure 4. Figure 4: Parameter correlations of the creep model and the hyperparameters of the covariance kernel. view at source ↗
Figure 5
Figure 5. Figure 5: Predicted creep coefficient with one parameter view at source ↗
Figure 6
Figure 6. Figure 6: Predicted creep function for equidistant (left) and logarithmic (right) training data and 0 % (green), 30 % view at source ↗
Figure 7
Figure 7. Figure 7: Predicted creep function for equidistant (left) and logarithmic (right) training data and 0 % (green), 30 % view at source ↗
Figure 8
Figure 8. Figure 8: Influence of preload intensity on calibrated model parameters and prediction of the final creep coefficient for view at source ↗
read the original abstract

The time-dependent deformation of concrete, particularly creep, remains a key challenge for reliable and material-efficient design. Experimental results show that tailored preloading, short-term loads exceeding the subsequent sustained load, can reduce both the magnitude and variability of creep strains which may be associated with beneficial microstructural changes. Building on these insights, this article employs Gaussian Process Regression (GPR) to calibrate analytical creep models, incorporating the effects of preloading intensity, timing, and concrete age into conventional predictions. The study pursues three main objectives: (i) calibrating a creep model using GPR based on experimental data, (ii) evaluating the impact of training data selection and preparation, and (iii) analysing model performance depending on the available experimental duration. The results demonstrate that GPR can improve model accuracy, quantify uncertainties, and support optimal test planning, while also enhancing understanding of preloading effects and contributing to more reliable and sustainable concrete creep predictions.

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 / 2 minor

Summary. The manuscript proposes calibrating analytical concrete creep models with Gaussian Process Regression (GPR) to incorporate preloading intensity, timing, and concrete age effects from experimental data. It pursues three objectives: GPR-based calibration, evaluation of training data selection/preparation impacts, and analysis of model performance as a function of available experimental duration. The central claim is that this yields improved accuracy, uncertainty quantification, support for optimal test planning, and enhanced insight into preloading benefits for more reliable creep predictions.

Significance. If the GPR calibration demonstrably improves predictive accuracy and provides well-calibrated uncertainties that generalize, the work could support more material-efficient concrete design by exploiting preloading to reduce creep magnitude and variability, with direct implications for sustainability in structural engineering.

major comments (2)
  1. [Results section] Results section (performance analysis by experimental duration): No quantitative hold-out validation is described (e.g., training on short-duration data and testing on longer durations or unseen preloading intensities), which is required to substantiate generalization claims for GPR given the sparsity and high cost of creep experiments; without this, the assertion that the approach supports optimal test planning and reliable predictions beyond training conditions cannot be evaluated.
  2. [Abstract and results] Abstract and results: The claim that GPR 'can improve model accuracy' is asserted without reported quantitative metrics (RMSE, R², or coverage of uncertainty intervals) or direct comparisons to the uncalibrated analytical model or alternative calibration methods, making the magnitude and robustness of the improvement impossible to assess from the presented evidence.
minor comments (2)
  1. [Abstract] The specific analytical creep model (e.g., B3, ACI 209, or Eurocode) being calibrated should be named explicitly in the abstract and methods to allow readers to understand the baseline functional form before GPR augmentation.
  2. [Methods] Notation for GPR inputs (preloading intensity, timing, age) and kernel choice should be defined consistently when first introduced to improve readability for readers outside machine learning.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments, which highlight important aspects for strengthening the validation and presentation of our results. We address each major comment below and have revised the manuscript accordingly to provide the requested quantitative evidence.

read point-by-point responses
  1. Referee: [Results section] Results section (performance analysis by experimental duration): No quantitative hold-out validation is described (e.g., training on short-duration data and testing on longer durations or unseen preloading intensities), which is required to substantiate generalization claims for GPR given the sparsity and high cost of creep experiments; without this, the assertion that the approach supports optimal test planning and reliable predictions beyond training conditions cannot be evaluated.

    Authors: We agree that explicit quantitative hold-out validation is necessary to support claims of generalization, especially given the limited availability of long-duration creep data. In the revised manuscript we have added a new subsection to the Results section that performs and reports hold-out experiments: GPR models are trained exclusively on short-duration subsets of the experimental data (e.g., up to 28 or 90 days) and then evaluated on both longer-duration observations from the same experiments and on preloading intensities withheld from training. We now report RMSE, R², and the empirical coverage rate of the 95 % credible intervals on these hold-out sets, together with a comparison against the uncalibrated analytical model. These additions directly substantiate the statements concerning support for optimal test planning and reliable extrapolation beyond the training conditions. revision: yes

  2. Referee: [Abstract and results] Abstract and results: The claim that GPR 'can improve model accuracy' is asserted without reported quantitative metrics (RMSE, R², or coverage of uncertainty intervals) or direct comparisons to the uncalibrated analytical model or alternative calibration methods, making the magnitude and robustness of the improvement impossible to assess from the presented evidence.

    Authors: We accept that the original manuscript did not supply the quantitative metrics needed to evaluate the claimed accuracy improvement. The revised version updates both the Abstract and the Results section with direct numerical comparisons. Tables now present RMSE and R² values for the GPR-calibrated model, the original analytical model, and at least one alternative calibration approach across the range of preloading conditions. We also report the observed coverage of the GPR predictive intervals on both in-sample and hold-out data. These metrics allow readers to assess the magnitude and robustness of any improvement. revision: yes

Circularity Check

0 steps flagged

No significant circularity; calibration uses independent external data

full rationale

The paper's core chain consists of fitting a Gaussian Process Regression model to independent experimental creep datasets that incorporate preloading intensity, timing, and age as inputs, then using the resulting surrogate to predict creep strains, quantify uncertainty, and analyze performance as a function of observation duration. No equation or step reduces a claimed prediction to a fitted parameter by construction, no self-citation supplies a load-bearing uniqueness theorem, and no ansatz is smuggled in; the GPR is a standard nonparametric regressor whose outputs are determined by the external measurements rather than by internal redefinition of the target quantities. The objectives explicitly include evaluating training-data selection and duration dependence, confirming an external-data-driven workflow rather than a self-referential loop.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Only the abstract is available, so the ledger is necessarily incomplete; standard GPR kernel hyperparameters are expected to be fitted to data, and domain assumptions about preloading effects are invoked without further detail.

free parameters (1)
  • GPR kernel hyperparameters
    Typical in Gaussian Process models; fitted to experimental creep data to control smoothness and uncertainty estimates.
axioms (1)
  • domain assumption Preloading produces consistent, repeatable reductions in creep strain magnitude and variability that can be captured by a statistical model
    Invoked in the abstract as the basis for including preloading variables in the calibration.

pith-pipeline@v0.9.0 · 5464 in / 1240 out tokens · 48119 ms · 2026-05-07T14:13:44.778802+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

6 extracted references

  1. [1]

    Taube, A

    C. Taube, A. Flohr, and G. Morgenthal. Experimental investigations on the influence of the short-term load history on the creep behaviour of normal concrete.Construction and Building Materials, 473:140916, 2025

  2. [2]

    Taube, A

    C. Taube, A. Flohr, G. Morgenthal, and A. Osburg. Experimental investigations on the influence of tailored preloading on the creep behaviour of concrete and its variability. In4th fib International Conference on Concrete Sustainability (ICCS2024), pages 71–79, 2025

  3. [3]

    Heller, C

    L. Heller, C. Taube, and G. Morgenthal. Analytical modelling for the prediction of the improved creep behaviour of tailored preloaded concrete. In4th fib International Conference on Concrete Sustainability (ICCS2024), pages 195–202, 2025

  4. [4]

    Eurocode 2: Design of concrete structures - Part 1-1: General rules and rules for buildings; German version EN1992-1-1:2004 + AC:2010

  5. [5]

    E. J. Cross, T. J. Rogers, D. J. Pitchforth, S. J. Gibson, S. Zhang, and M. R. Jones. A spectrum of physics-informed gaussian processes for regression in engineering.Data-Centric Engineering, 5, 2024

  6. [6]

    Saltelli, M

    A. Saltelli, M. Ratto, and T. Andres.Global sensitivity analysis: The Primer. Wiley-Interscience, Chichester, England, 2008. 10