Assessment of the applicability of creep and shrinkage models (SP 63, EN 1992, fib MC2010, ACI 209, B4) using international databases

Abstract

The divergence of long-term concrete deformation predictions among code models reaches magnitudes comparable to the deformations themselves; in the design of long-span and prestressed structures this translates into underestimated deflections and prestress losses. Five creep and shrinkage prediction models – SP 63.13330.2018, EN 1992-1-1, fib Model Code 2010, ACI 209.2R and B4 (RILEM TC-242-MDC) – rest on different theoretical foundations and are calibrated against non-overlapping test sets, so the question of their applicability admits no single answer and splits into particular ones: which model is more accurate for a given deformation component and over a given time horizon. The aim is to quantitatively compare the five models with one another and with the aggregated statistical characteristics of the international NU database. An author's model calculation of deformations was carried out for concrete of class C30/37 at 65% relative humidity and a notional section size of 150 mm over the interval from 100 to 10 000 days; the results were compared with published coefficients of variation of model error relative to the NU database. The spread of shrinkage predictions by 10 000 days amounted to about 33% of the mean, and that of the creep coefficient to about 28%; the ranking of models by accuracy changes on passing from shrinkage to creep and from short to multi-decade terms. Model B4 retains the lowest coefficient of variation for autogenous shrinkage and basic creep, whereas for total shrinkage EN 1992-1-1 proves somewhat more accurate; ACI 209.2R exhibits the largest scatter at long durations. The applicability of a model is determined not by its general status but by the combination of the deformation component, the time range and the completeness of the calibration set on which the model is tuned.