Modelling drying and autogenous shrinkage of high strength concrete with and without mineral and chemical admixtures

Abstract

Parameters of some well known concrete shrinkage prediction models have not been updated to account for modern high performance concrete data. Consequently, their predictions are not accurate for high strength concrete with chemical admixtures and high mineral admixtures content. This study considered modifying three well known shrinkage models, the RILEM B4, MC 2010 and WITS models, to predict drying and autogenous (RILEM B4 and MC 2010 only) shrinkage for high strength concrete. Experimental data for concrete shrinkage specimens that met the criteria of rapid hardening or rapid development of early age strength, a water-to-cementitious material ratio ≤ 0.42 or 28th day compressive strength ≥ 60 MPa was extracted from the 2018 version NU database (Northwestern University, 2018), a technical report (Al-Manaseer and Fayyaz, 2011) and the Concrete Institute of South Africa database. This gave reliable data for 220 drying and 342 autogenous shrinkage experiments. These data were used to (i) assess accuracies of the original versions of the selected models in predicting shrinkage of high strength concrete (using only data within the covariate ranges on which each model was developed), (ii) update model parameters to improve the accuracy of high strength concrete shrinkage predictions using data subsets (from the 562 experiments) for comparable experiments and (iii) propose composite models constructed as logistic dose curves (combining two or more individual functions) to fit high strength concrete drying shrinkage data that had an early age peak before reaching the final shrinkage value. Excel Solver® was used to update model parameters. Shrinkage residuals of both original and modified models were used to rank the models for the complete HSC datasets, the data subsets and for individual shrinkage time periods (0 to 99, 100 to 199, 200 to 499 and ≥ 500 days). Ranking was done using the statistical indicators Root Mean Square Error, adjusted Coefficient of Determination, Akaike’s Information Criterion and overall coefficient of variation. High strength concrete drying shrinkage predictions of the original models were best overall for the WITS, then the RILEM B4 and MC 2010 models. After parameter modification they were best for overall for the WITS, then the MC 2010 and RILEM B4 models. For high strength concrete autogenous shrinkage prediction, the RILEM B4 model performed better than the MC 2010 overall (original and modified versions). The proposed composite models outranked the existing models in overall performance and per shrinkage term for the high strength concrete data subsets with an early age peak. Prediction errors for the original models were high for drying shrinkage experiments, of the order –235% to +100% for short-term shrinkage (0 to 99 days) and –257% to +74% for medium- and long-term shrinkage (≥ 100 days). For the modified models, residuals were generally much smaller for the medium- and long-term shrinkage, with errors ranging from –57% to +48%. For some data subsets the model parameters could not be improved, due to the large variations in the actual shrinkage data. For autogenous shrinkage experiments, original model prediction errors ranged from -2943% to +81% for short-term shrinkage and -321% to 35% for medium- and long-term shrinkage. The modified model prediction errors ranged from -381% to 99% for short-term and -98% and +29% for medium- and long-term shrinkage. Comparisons were also made across the different geographical regions from which the experiments originated, because of their different test specifications and cement classifications. The original RILEM B4 and MC 2010 models predicted worse for North American concretes than for European concretes, but the RILEM B4 model was the more accurate for concretes from both these regions and from East Asia. Surprisingly the MC 2010 achieved the lowest overall coefficient of variation (<40%) for Southern African concretes.