SHRINKAGE OF SELF-COMPACTING CONCRETE. A COMPARATIVE ANALYSIS

Self-compacting concrete (SCC) is a concrete type that does not require vibration for placing and compacting. SCC possesses special technical features and properties that recommend its application in many jobs. Nevertheless, in some situations, it has been observed an inadequate behavior of the material at early ages due to shrinkage. The existing shrinkage prediction models were developed for standard concrete. In this paper three SCC mixtures, with different compressive strength, are studied in terms of autogenous and total shrinkage. The results are compared with the Eurocode 2 model. For the studied mixtures it was found that this model underestimates the autogenous shrinkage, while the total shrinkage is generally overestimated.


INTRODUCTION
The estimation of time-dependent behaviour is still one of the most difficult aspects in designing a concrete structure. Over the last few years, the use of self-compacting concrete (SCC) has increased [1][2][3]. SCC is a technically advanced material, which has shown to have a high potential in the areas of productivity, working conditions and even in matters arising from their inherent characteristics. This material possesses special properties which makes it more suitable for repair jobs. Nevertheless, in some real cases, unsuitable behaviour of the repair material was observed in the early stages of hydration, due to shrinkage [4][5][6]. The structural concrete codes which deal with time-dependent behaviour provide general rules for standard concrete, but the validation of some established stress-strain-relations have to be confirmed via laboratory testing when special mixtures are used [7,8]. The most important changes in mix design between conventionally vibrated concrete (VC) and SCC are the higher paste volume, the large use of mineral additions, and the high dosage of superplasticiser, as well as the optional resource to a viscosity-modifying agent. The variations in paste volume and binder composition lead to a significant influence on the viscoelastic properties of this type of concrete [9][10][11][12][13]. According to the ACI Terminology [14], shrinkage is the decrease in either length or volume of a material subsequent from changes in moisture content or chemical changes. This decrease occurs in the absence of stress attributable to actions external to the concrete [15][16][17]. When no moisture transference with the surrounding environment is allowed, and temperature is kept constant, this volume change is called autogenous shrinkage and is attributed to self-desiccation due to binder hydration [16,[18][19][20][21][22][23][24][25][26][27][28][29][30][31][32]. It is accepted that autogenous shrinkage is a consequence of RH changes in the pores [33][34][35]. The volume of internal liquid water decreases due to hydration. Depending on the pore structure and water available, different mechanisms are triggered. Changes in the surface tension of the solid gel particles, disjoining pressure and tension in capillary water are parameters that have been discussed [16,23,36,37]. In addition to these main mechanisms, other phenomena may be involved in the early volume changes: swelling phase related to sulfate-to-alkali ratio of the clinker and the amount of free lime [19], influence of the type of hydration products [21,24,27,32,34,38], creep in the C-S-H phases [30]. Autogenous shrinkage does not usually appear significantly in normal VC, but in highperformance concrete types such as high-strength and SCC with a low water-cement ratio (w/c), autogenous shrinkage it is not an unimportant role [6,22,26,28,29,39,40,41]. In those cases, the low water/powder ratio leads to refined pores, and the SCC is more sensible to cracking at early shrinkage than VC, even when good practical curing is applied. As the use of SCC becomes more widespread, some innovative techniques to combat this singularity have been developed. The means and procedures for mitigating shrinkage include cement modification, chemical admixtures, mineral additives, control of curing conditions, fibers and advanced methods of internal curing [42][43][44]. Lately, innovative shrinkage control methods using the combined effect of expansive and shrinkage reducing admixtures have been presented [45][46][47][48]. In order to optimize the shrinkage reducing effect, an appropriate curing is usually suggested, since curing conditions affect both shrinkage and cracking processes [49,50]. In this study, autogenous and total shrinkage in three different SCC mixtures with different compressive strength are studied. The curing effect on the total shrinkage is also assessed. The objective of the experimental work carried out is to compare the results with the Eurocode 2 model (EC 2) [7]. A better understanding of early and long-term shrinkage will promote good performance of the concrete structure during its service-life.

MATERIALS AND METHODS
In this study, three different concrete mixtures were studied. The preparation of specimens (40 mm x 40 mm x 160 mm) was carried out according to EN 196-1, in a room with a temperature of 20 ± 2 ºC and relative humidity of 555%. Nevertheless, the aggregates and mixture proportions used in the study are different from those established in EN 196-1 (see Tables 4 and 7). Furthermore, due to the low viscosity of the mixture (SCC), the test specimens were not compacted mechanically, and the SCC was just poured into molds. The removal of molds took place about 8, 18 and 26 hours after mixing, according to the strength class. This length of time was defined as the minimum necessary to ensure concrete strength between 2 and 5 MPa, to avoid specimen's damage due to molds removal. Subsequently, the specimens were weighed, their length was registered and, in the case of the samples used for measurement of autogenous shrinkage, they were sealed with a plastic film. Section 5 of EN 12390-2:2009, Testing hardened concrete -Part 2: Making and curing specimens for strength tests [51], prescribes leaving the test specimens in the mold for at least 16 hours, protected against shock, vibration and dehydration. Taking into consideration that autogenous deformation of high strength concrete may start very early [19,20,34,52] with this type of concrete the first length measurement should be made at an earlier age. The RILEM recommendation TC 107-CSP [25], for measurement of timedependent strains of concrete, does not provide indications about demolding. The three different demolding periods (about 8, 18 and 26 hours) where chosen after carrying out compressive strength tests, which have shown similar values (2-5 MPa) to those specimens tested at early ages. Shrinkage deformations of each specimen were measured using a length comparator, sensitivity of 1 µm, and gage studs on the end sections of the concrete prisms ( Figure 1). Stability of the length comparator was checked by a reference invar bar. Samples for measurement of autogenous shrinkage were placed on two thin supports and kept sealed. The results of the measurements present fluctuation which should not be taken into account. The manual method of measurement implies some error and variation on laboratory room temperature and humidity could not be avoided. Since first length measurement was performed very early (compressive strength not higher than 5 MPa), and the autogenous shrinkage is not relevant for stress analysis before the solid percolation, it is assumed negligible the difference between the actual autogenous shrinkage and the measured shrinkage on the sealed specimens. At the ages of 1, 2, 3, 5, 7, 14 and 28 days, and 2, 3, 4, 5, 6, 7, 8 and 9 months, the length variation of the samples was evaluated. After mold removal, 2 levels of curing were specified: -Uncured (air curing with the temperature of 20 ± 2 ºC and relative humidity of 555%); -Curing until the concrete reaches near 70% of the average strength at 28 days (3, 5 or 7 days), which satisfies the requirements of curing class 4 specified in the EN 13670 standard [53] (70% of specified characteristic 28 days compressive strength). The specimens were prepared with Portland cement, CEM I 52.5 R, CEM II/A-L 42.5R or CEM II/B-L 32.5N (see Table 1 -Chemical properties and Table 2 -Physical properties), according to EN 197-1, siliceous fly ash from Compostilla in Spain (Tables  3 and 4), siliceous sand and limestone coarse aggregate from Algarve in Portugal (Table  5), potable tap water and three superplasticizers ( Table 6). The work presented in this paper is part of a PhD study, which involved a wide range of tests. In order to limit the amount of work and materials used, it was decided to use small specimens (40x40x160 mm 3 ), since the ratio between smallest size of the specimen and largest aggregate size is about 4, thus the size effects were considered having minor influence on the results.     Water absorption (%) 0.6 1.5  The concrete constituents and dosages adopted were chosen in order to produce mixtures that could be used in commercial production of concrete. Three different superplasticizers were used in order to obtain similar flow characteristics with different W/C, using normal dosages of the commercial chemical admixtures. The three compositions were designed based on the continuity of the work presented in [54]. It was considered beneficial to keep the non-inclusion of fillers in the compositions, except those coming from cement, so that this study corresponds better to possible practical applications, taking into account that most of the Portuguese concrete plants do not use this type I addition in the manufacturing process. The criteria for selection the types and classes of cement, as well as the superplasticizers of each composition, included the use of commercially available standard materials and the required level of concrete strength. Table 8 shows some of the properties measured in fresh and hardened states of the mixtures. In Figure 2, the strength development for the three SCC mixtures is presented.

Eurocode 2 model
According to the EC2 [7] the shrinkage strain εcs is composed of two components: where εcs is the total shrinkage strain, εcd is the drying shrinkage strain, and εca is the autogenous shrinkage strain.
The development of the drying shrinkage strain in time may be computed by: where εcd,0 is the basic drying shrinkage strain, which can be calculated from: In Eq. 2, kh is a coefficient, ranging between 0.7 and 1, which depends on the nominal size (mm) of the cross-section (ℎ0 = 2 ⁄ , where Ac is the concrete cross-sectional area, u is the perimeter of that part of the cross section which is exposed to drying).
And βds calculated from: where: t is age of concrete at the time considered, ts is age of concrete at beginning of drying shrinkage (mostly end of curing), The autogenous shrinkage is obtained from: where: (∞) = 2,5 × ( − 10) × 10 −6 Eq 7 fck is the characteristic compressive cylinder strength of concrete at 28 days, where t is given in days.
The values of fcm and fck used in this study, based on the available experimental results, are presented in Table 9. The following results (see Figure 3), were obtained using the average measurements of seven specimens per mixture. For each mixture, the solid curves present the average values and the dashed curves present the average plus or minus one standard deviation. As can be seen in Figure 3a), the specimens used for the autogenous measurements register no mass loss up to approximately 100 days. However, after that age, some drying occurs. The plastic sealing allows some evaporation, but it remains in very low level after 270 days (<0,3%). The mass loss of the specimens subjected to air drying (without cure or subjected to 7 days of cure) starts immediately after the ambient exposure. At 270 days, the mass loss observed in the specimens is almost the same (≈5%), but is a bit smaller on the specimens cured for 7 days. This result is related with the pore structure and hydration degree of the specimens. Keeping the concrete surface humid, during a longer period of time, provides more time for hydration of the concrete skin, and leads to a more refined pore structure. Between 60 days and 270 days, the weight variations are minimal, showing that the mass changes are stabilized.  Figure 3b) the measured values of autogenous shrinkage are higher than those predicted by EC2. A value of 150 microstrain is reached after 3 months, whereas the model predicts less than 50 microstrain at the same age. Beyond 3 months, the difference keeps increasing, but after this age some drying shrinkage also occurs due to the mass loss, which may have affected the measurements. Nevertheless, the results clearly indicate that the EC2 prediction is far from the actual observed values. Figure 3c) shows the results of the total shrinkage obtained from the Ls specimens without cure (demolded between 24 and 28 hours). In this case, the EC2 overestimates the total shrinkage. At early age, the difference is higher, but is still relevant after 9 months, EC2 providing higher values than the measured average (≈+40 %). Figure 3d) shows the results of total shrinkage obtained from the Ls specimens with 7 days curing (demolded between 24 and 28 hours). The measured values and the EC2 predictions are similar. However, taking into account that the experimental results of the autogenous shrinkage, and computing the drying shrinkage as the difference between total shrinkage and autogenous shrinkage, the actual values of the drying shrinkage are smaller than those calculated from the EC2. Thus, it may be concluded that the EC2 model underestimates the autogenous shrinkage and overestimates the drying shrinkage.
After 9 months, the average of the experimental results is equal to the calculated values from the model. At this age, the total measured shrinkage is 573 microstrains, higher than the total shrinkage of the uncured specimens (435 microstrains, Figure 3c)). Since the mass loss is similar for the two curing conditions (Figure 3a)) this difference is mainly related with the more refined pore structure of the cured specimens [55]. Figure 4 presents the results obtained on the intermediate strength concrete mixture. Figure 4a), the mass loss observed shows that the drying profile is similar to Figure 3a) but, as expected, with smaller values. The mass loss of the specimens subjected to air drying (without cure or 5 days of cure) also starts immediately after the ambient exposure, and at 270 days the mass loss of specimens is similar (≈4%), but is smaller on the specimens cured for 5 days.  (Figure 4b)) is higher than the values calculated from the EC2, reaching 200 microstrain at 90 days, more than double the estimated values at the same age. The autogenous shrinkage on Is concrete is also higher than the autogenous shrinkage of Ls, as a result of the lower W/C. Figure 4c) displays the results of total shrinkage obtained from the Is specimens without cure (demolded between 16 and 19 hours). The EC2 overestimates the total shrinkage, resulting on a difference at 9 months of about 250 microstrains (≈+60%). Figure 4d) presents the results of total shrinkage obtained from the Is specimens 5 days cured (demolded between 16 and 19 hours). The EC2 underestimates the shrinkage at early age and overestimates it at later ages. At early age, the autogenous shrinkage prevails, and the difference between measured and calculated values is due to the underestimation of the autogenous shrinkage strain. At later ages, when the drying shrinkage becomes more relevant, the EC2 estimations surpass the measured values. Bermejo et al [56] measured the total shrinkage on 3 different SCC mixtures (30 MPa compressive strength), and refer that the EC2 overestimates the shrinkage strains, in accordance with the results presented in Figures 4c) and 4d).

High strength concrete mixture
The results obtained on the Hs concrete are shown in Figure 5. The mass variation on the specimens used for the autogenous measurements remains low and the highest mass loss was recorded on the specimens without cure (Figure 5a)). Concerning the specimens exposed to air drying, when comparing Figure 5a) with Figures 4a) and 3a), it is clear that there is a decrease of mass loss with the increase of concrete strength, which is related with the pore size. At latter ages, the influence of the curing period is small, but also increases with the concrete strength.
This last finding may be misunderstood, as usually it is considered that the cure is more relevant in low strength concrete. However, it should be noted that, as example, the Hs uncured specimens (demolded between 7 and 9 hours) were exposed to air drying at very early age, and after 1 day the mass loss was already 2 %. It means that, at that age, about 45 l/m 3 of water has dried from the concrete and is not available for cement hydration. This decrease in the water content of the mixture reduces the amount of hydration products, because the water/powder and the water/cement ratios were reduced from 0.31 to 0.21 and from 0.48 to 0.34, respectively. After one day of exposure, 4 days of age, the cured specimens have dried less than 0.5% of the original weight (≈ 11.4 l/m 3 ), and the corresponding reduction in water/cement is small, 0.48 to 0.44, and still above the Powers limit of ≈0.42 [33]. In fact, with different degrees of hydrations, it is expected to have distinct mass losses on the two curing conditions.
This effect of the cure on the degree of hydration is smaller for mixtures with higher water/cement, in accordance with the observed values on the three concrete mixtures. Autogenous shrinkage of Hs specimens is presented in Figure 5b). As can be seen, the EC2 model does not provide an accurate prediction. The EC2 values are higher at early age and smaller at latter ages. At 9 months the measured autogenous strain is near 300 microstrains, similar to the autogenous shrinkage of Is concrete at the same age and higher than the autogenous shrinkage of Ls concrete (≈ 250 microstrain). The difference observed among the three concrete mixtures is small. This result is probably related with the water/clinker values. Using the loss of ignition as an rough estimator of the clinker content of the cements, we can achieve approximate values of about 75 %, 83 % and 97 % for CEM II-B/L, CEM II-A/L and CEM I, respectively. With this clinker content, the water/clinker of the Ls, Is and Hs concretes are, respectively 1.01, 0.83 and 0.49. According to Powers [33], a water content of about 0.42 is enough to avoid significant self-desiccation. This indicates that, without drying and in the absence of fly ash reaction, there is enough water to avoid significant autogenous shrinkage.
As the binder have a high percentage of fly ash, the reaction of this type II addition becomes more relevant for the autogenous strains, delaying the self-desiccation process. Taking this delay into consideration, at latter ages the slope of the autogenous lines is higher on Hs concrete and smaller on Ls concrete (Is having an intermediate slope), which is in accordance with the expected different self-desiccation of the mixtures. At latter ages, the rigidity of the solid body is high and the creep is low, and, consequently it is reasonable to expect small differences on autogenous shrinkage of the different mixtures.
Figures 5c) and 5d) present the results of total shrinkage obtained from the Hs specimens, uncured (demolded between 7 and 9 hours) and subjected to 3 days curing, respectively. It can be seen that EC2 overestimates the total shrinkage. The difference between experimental and estimated values is smaller on the cured specimens, being about 30 % at 270 days.

Comparative analysis for the 3 SCC classes results
According to the presented results, it is observed that the variation of the water/powder ratio influences the autogenous shrinkage. Low water/powder ratios lead to greater autogenous shrinkage. This is in agreement with the revised bibliography [57][58][59].
However, and as referred above, the difference between the results of autogenous shrinkage depends on the type of binder.
In relation to the total shrinkage, for the three concrete mixtures studied, with and without curing, the specimens subjected to curing exhibit higher shrinkage values (comparisons between Figures 3c) and 3d), Figures 4c) and 4d), Figures 5c and 5d)). An explanation for this apparent contradiction behavior is presented by Oliveira et al [55] and corresponds to a refinement of the concrete porous structure for the specimens subjected to cure.

CONCLUSIONS
This paper presents a study comparing experimental concrete shrinkage results (with three different SCC strength classes) with the values estimated by the EC2 model. Taking into account the limited number of mixtures tested, the following conclusions should not be considered as general trends, but rather indications of possible deviations from the standard values.
The analysis carried out shows that the measured mean values of autogenous shrinkage were higher than the mean values estimated using the EC2 equations, for the three SCC strength levels studied. This conclusion was drawn for W/C values ranging between 0.76 and 0.48 and W/P values ranging between 0.46 and 0.31, which are not too small.
On the other hand, despite the expected higher shrinkage of SCC, when compared with shrinkage of standard vibrated concrete, the measured differences between total and autogenous shrinkage (using different specimens), which may be defined as approximations of the actual drying shrinkage, are smaller than the EC2 estimations. However, in general, at long term the EC2 estimations for total shrinkage were in the safe side, i.e., providing higher values than the measured ones.
Finally, the presented results show that, for thin elements, a short curing period does not have negative results on total shrinkage. The EC2 model for shrinkage does not reproduce this phenomenon.