Statistical Extrapolation of Shrinkage Data-Part I : Regression

A large series of dala on carefu/ly COnlrolled shrinkage tests of concrete involving groups of large numbers of identical specimens is reporttd. The data are used to compare existing shrinkage formulas in ACI, CEB-FIP, and BP models. By far the besl agreement is obtained for the BP model. Assuming that only the measured data for a certain initial period are known, predictions are made for long times and are compared with the subsequently observed shrinkage strains. In this manner, various possible statisllca/ regression modeis are examined and compared. Best predictions are obtained when the shrinkage formula is fitted to lest data using nonlinear optimi;,ation, then linear regression in transformed variables is used to obtain the confidence limits for long-time predictions. It is concluded that good long-time predictions of shrinkage can be obtained on the basis of shrinkage lest results on 80 mm diameter cylinders for a 3-week duration. These predictions involve only the intrinsic uncertainty of the material, on which the uncertainly due to random enVironment, curing history, and differences in concrete composition must be superimposed for practical application.