Prediction of autogenous shrinkage in concrete from material composition or strength calibrated by a large database, as update to model B4

In modern concretes, the autogenous shrinkage, i.e., the shrinkage of sealed specimens, is much more important than it is in traditional concretes. It dominates the shrinkage of thick enough structural members even if exposed to drying. A database of 417 autogenous shrinkage tests, recently assembled at Northwestern University, is exploited to develop empirical predictive equations, which improve significantly those embedded in RILEM Model B4. The data scatter is high and the power law (time)^0.2 is found to be optimal for times ranging from hours to several decades of years, as the test data give no hint of upper bound. Statistics of data fitting yields the approximate dependence of the power law parameters on the water-cement and aggregate-cement ratios, cement type, additives such as the blast furnace slag and silica fume, and curing type and duration. Alternatively, the power law parameters can be reasonably well predicted from the compression strength alone. Since some database entries do not report all these composition parameters and others do not report the compressive strength, and since the concrete strength is often the only material property specified in design, two types of models are formulated—composition based, and strength based. Both are verified by statistical comparisons with individual tests, and optimized by nonlinear statistical regression of the entire database, so as to minimize the coefficient of variation of deviations from the data points normalized by the overall data mean. The regression is weighted so as to compensate for the bias due to crowding of data in the short-time range. Statistical comparisons with the prediction models in the JSCE code, Eurocode and CEB MC90-99 code (identical to fib Model Code 2010) show the present model to give significantly better data fits. Finally it is emphasized that, in presence of external drying and creep, accurate predictions will require treating the autogenous shrinkage as a consequence of pore humidity drop caused jointly by self-desiccation due to hydration and by moisture diffusion, and solving the time evolution of humidity profiles. The present model is proposed as an update for the autogenous shrinkage formula in model B4, although recalibration of the whole B4 would be needed.