Professor, Swiss Federal Institute of Technology, Lausanne,
Head, Swiss Federal Institute of Technology, Lausanne,
Assistant professor, Korea Advanced Institute of Science and Technology, Seoul,
Professor, Northwestern University, Evanston, IL
A large series of concrete shrinkage tests which can be used for statistical purposes is reported. The series involves two groups of 36 identical cylindrical specimens, with a diameter of 83 mm for Group 1 and 160 mm for Group 2. Statistical analysis of shrinkage strains and strain rates is presented, and the goodness of fit by normal distribution, log-normal distribution, and gamma distribution is analyzed. Correlations between the values at various times are determined.
The study reveals only the intrinsic randomness of the shrinkage process (along with measurement errors) and omits the superimposed uncertainties due to randomness of environment and to prediction uncertainties in the influence of material composition, curing, and other factors. It is found that the coefficient of variation of shrinkage values first decreases with time and then levels off at 6 to 9%. The normal, log-normal, and gamma distributions all give acceptable fits of the observed distributions. There are small systematic deviations from normal distribution: the distributions are slightly skewed to the right and slightly platykurtic. The skewness is more pronounced for the distribution of shrinkage rates, and asymmetric distributions such as gamma give slightly better fits. The correlation of single specimen shrinkage deviations from the group mean at various times is characterized by correlation coefficients. Their value is found to decrease slowly as the length of the time interval increases. For shrinkage rates, this decrease is much faster. The standard deviation of the single specimen shrinkage from its smoothed shrinkage curve is much less than (about one third of) the standard deviation of the single specimen relative to the mean for all specimens. The standard deviation of all single-specimen smoothed shrinkage curves at one time is almost the same as (about 95% of) the standard deviation of the original unsmoothed data for that time.
Paper ID: CCA10078J