Statistical filtering of useful concrete creep data from imperfect laboratory tests

Mohammad Rasoolinejad, Saeed Rahimi-Aghdam, Zdeněk P. Bažant*

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

6 Scopus citations


The reporting and evaluation of creep tests of concrete is complicated by the fact that creep is significant even for the shortest observable load durations. Compared to the strain after 0.1 s load duration, the strain at 2 h duration is typically 53% greater. Most experimenters have for decades been unaware of this fact. Consequently, the reported creep curves require correction by a time shift, which ranges from 0 to 2 h. This further implies a vertical shift of entire creep curve, important for all times up to structure lifetime. To filter out the errors, it is argued that, within an initial period during which the advance of hydration is negligible, which is normally about 1 day, the initial basic creep must follow a power law of the time. Creep test data from the literature are used to prove it. Corrections by time and deformation shifts are determined by minimization of the sum of squared deviations of the power law from the creep test data. For a fixed exponent n and time shift s, the optimization is reduced to linear regressions of two kinds, depending on whether the data are given in terms of either the compliance function or the creep coefficient. For both, the linear regression parameters depend nonlinearly on the chosen values of n and s. To avoid nonlinear optimization, which need not converge to the correct result, a set of many discrete values of n and s within their realistic ranges is selected and the (n, s) combination minimizing the objective function is obtained by a search. Enforcing a power law form of the initial creep curve is found to lead to better data fits. The optimum exponent n for the entire database is around 0.3, applicable to the time period cca (10 s, 1 day). After that, the exponent transits to about 0.1, and prior to that it is about 0.08. After filtering out the errors, the corrected database will allow better calibration of the general creep prediction model such as B3 or B4.

Original languageEnglish (US)
Article number153
JournalMaterials and Structures/Materiaux et Constructions
Issue number6
StatePublished - Dec 1 2018


  • Creep testing
  • Decontamination of big databases
  • Error filtering
  • Instantaneous elastic strain
  • Optimization and Statistics
  • Testing errors
  • Viscoelasticity

ASJC Scopus subject areas

  • Civil and Structural Engineering
  • Building and Construction
  • Materials Science(all)
  • Mechanics of Materials


Dive into the research topics of 'Statistical filtering of useful concrete creep data from imperfect laboratory tests'. Together they form a unique fingerprint.

Cite this