Evaluating Prediction Models of Creep and Drying Shrinkage of Self-Consolidating Concrete Containing Supplementary Cementitious Materials/Fillers

: Supplementary cementitious materials (SCMs) and ﬁllers play an important role in enhanc-ing the mechanical properties and durability of concrete. SCMs and ﬁllers are commonly used in self-consolidating concrete (SCC) mixtures to also enhance their rheological properties. However, these additives could have signiﬁcant effects on the viscoelastic properties of concrete. Existing models for predicting creep and drying shrinkage of concrete do not consider the effect of SCM/ﬁller on the predicted values. This study evaluates existing creep and drying shrinkage models, including AASHTO LRFD, ACI209, CEB-FIP MC90-99, B3, and GL2000, for SCC mixtures with different SCMs/ﬁllers. Forty SCC mixtures were proportioned for different cast-in-place bridge components and tested for drying shrinkage. A set of eight SCC mixtures with the highest paste content was tested for creep. Shrinkage and creep test results indicated that AASHTO LRFD provides better creep prediction than the other models for SCC with different SCMs/ﬁllers. Although all models underestimate drying shrinkage of SCC with different SCMs/ﬁllers, the GL2000, CEB-FIP MC90-99, and ACI 209 models provide better prediction than AASHTO LRFD and B3 models. Additionally, SCC mixtures with limestone powder ﬁller exhibited the highest creep, while those with class C ﬂy ash exhibited the highest drying shrinkage.


Introduction
Self-consolidating concrete (SCC) is highly flowable, non-segregating concrete that can spread into place, fill the formwork, and encapsulate the reinforcement without any mechanical consolidation [1]. To enhance the stability of SCC, supplementary cementitious materials (SCMs)/fillers are used to improve the viscosity and quality of paste in addition to mechanical and durability properties. The binder composition of SCC, in addition to many other factors, affects its viscoelastic properties, primarily shrinkage and creep. However, existing creep and drying shrinkage prediction models do not account for the effect of SCM/filler type on the predicted values for SCC. Therefore, the objective of this study is to evaluate creep and drying shrinkage prediction models including AASHTO LRFD [2], ACI 209 [3,4], CEB-FIP MC90-99 [5], B3 [6], and GL2000 [7] for SCC mixtures containing different types of SCM/filler. To achieve this objective, a literature review was conducted to determine the different prediction models for shrinkage and creep of SCC as well as the effect of SCMs/fillers on its viscoelastic properties. Fresh, early-age, and hardened concrete properties were evaluated in a laboratory investigation of forty SCC mixtures proportioned using: two types of coarse aggregate-crushed limestone and natural gravel; three nominal maximum sizes of aggregate (NMSA)- 3 4 , 1 2 , and 3 / 8 in.; three SCMs and one filler-25% Class F fly ash, 25% Class C fly ash, 30% ground granulated blast-furnace slag (GGBFS), and 20% Class

Drying Shrinkage Models
Higher drying (free) shrinkage of SCC is expected due to the denser matrix of the system, which leads to small capillary voids and allows faster removal of water than large voids [18]. Additionally, using finer cement leads to higher drying shrinkage due to the pore refinement [19]. On the other hand, using fly ash and GGBFS reduces the drying shrinkage of SCC, while the silica fume increases the drying shrinkage when used in binary blends [20,21]. The shrinkage of high early-strength SCC is similar to or less than that of CVC and there is no significant effect of fine aggregate ratio [22]. Naito et al. [17] presented higher viscoelastic properties of SCC than CVC due to the higher fine aggregate volume in SCC.
Khayat and Mitchell [15] reported that all models underestimate the drying shrinkage of SCC; however, the CEB-FIP MC90 model provides the best prediction of drying shrinkage of SCC as it considers the effect of cement type. Landsberger and Fernandez-Gomez [16] reported that the ACI 209R model provides the best prediction of drying shrinkage of SCC, while the CEB-FIP MC90 and GL2000 models substantially underestimate it. Schindler et al. [22] reported that the ACI 209R model accurately predicts the shrinkage of SCC at later ages (56 and 112 days), while the AASHTO LRFD (2004) model underestimates SCC shrinkage at early ages (7 and 14 days) and overestimates it at later ages (56 and 112 days). Naito et al. [17] reported that the ACI 209 model overestimates the drying shrinkage for both SCC and high early strength concrete. Table 1 and Figure 1 show, respectively, the chemical composition and particle size distribution of the cement type I/II, SCMs (class F fly ash, class C fly ash, GGBFS), and filler (limestone powder) used in the experimental investigation. Two different types of coarse aggregate, crushed limestone and gravel, were used in this investigation. The two types were combined with fine aggregate (natural sand) using three different fine-to-coarse aggregate ratios of 0.45, 0.47, and 0.50. All physical properties and particle size distribution of fine and coarse aggregates are shown in Table 2 and Figure 2, respectively, and the combined aggregate gradations used in SCC mixtures are listed in Table 3. Chemical admixtures included polycarboxylate type high range water reducing admixture (HRWRA) that meets the requirements of ASTM C494 type F admixture; viscosity-modifying admixture (VMA) that meets the requirement of ASTM C494 type S admixture; and air-entraining admixture (AEA) that meets the requirements of ASTM C 260. All materials used in this investigation were obtained from suppliers in the Midwest states of Nebraska, Iowa, and Minnesota.

Mixture Proportioning
Two groups of SCC mixtures were proportioned: one with crushed limestone coarse aggregate (LS), and the other with gravel (G). Each group had 20 mixtures as follows: 5 mixtures had 25% powder replacement with class C fly ash (C), 5 mixtures had 25% powder replacement with class F fly ash (F), 5 mixtures had 30% powder replacement with GGBFS (S), and 5 mixtures had 35% powder replacement with class F fly ash (20%) and limestone powder (15%) (FLP). Each group had NMSA of 19 mm ( 3 4 in.), 12.5 mm ( 1 2 in.), and 9.5 mm ( 3 / 8 in.) with two levels of filling ability: high flow (HF), where slump flow is less than 750 mm (30 in.) but greater than or equal to 650 mm (26 in.); and low flow (LF), where slump flow is less than 650 mm (26 in.) but greater than or equal to 550 mm (22 in.). Table 4a,b present the proportions of the forty SCC mixtures containing limestone and gravel aggregate, respectively.

Workability Testing
All SCC mixtures were proportioned to achieve acceptable levels of filling ability, passing ability and stability (static and dynamic). These properties, except dynamic stability, were assessed using standard test methods to assure the quality of the fresh SCC. Filling ability was evaluated using the slump flow test of the inverted cone in accordance with ASTM C1611. As an indication of the viscosity of the mixtures, the time of reaching 500 mm spread diameter (T 50 ) was also measured. The passing ability of fresh SCC was determined using the J-ring test method according to ASTM C1621. Two parameters were used to describe the passing ability of fresh SCC: (1) the difference between average slump flows (∆D) in restrained (with J-ring) and unrestrained conditions (without J-ring); (2) the difference between the height of concrete patty in the middle of the J-ring, and the average height of the patty at four points around the perimeter of J-ring (∆H) according to AASHTO T 345. The higher the ∆D and ∆H, the higher the probability of blockage when SCC flows around reinforcing bars. The filling capacity of fresh SCC was determined using the caisson test method and according to AASHTO T 349. The measured filling capacity represents the ability of fresh SCC to fill the forms while passing through cross bars. The static stability of SCC was determined using the four standardized test methods: penetration test according to ASTM C1712, column test according to ASTM C1610, visual stability index (VSI) according to ASTM C1611, and hardened visual stability index (HVSI) according to AASHTO PP 58. Dynamic stability was evaluated using the flow-through test according to Lange et al. [23], as no standard test method is available for this property. Additionally, mortar and concrete rheometers were used to characterize the rheological properties of SCC mixtures. Dynamic yield stress and plastic viscosity (i.e., Bingham model parameters) were determined using Brookfield mortar rheometer according to ASTM C1749, while yield torque and slope were determined using IBB concrete rheometer according to Hu and Wang [24].

Creep Testing
Creep strain was measured according to ASTM C512 for only eight SCC mixtures due to the availability of testing frame and length of test duration. SCC mixtures containing limestone and gravel aggregates with NMSA of 9.5 mm ( 3 / 8 in.) were chosen because they have the highest paste volume and, consequently, are expected to have the highest creep strains. A set of two 150 × 300 mm (6 × 12 in.) cylinders was obtained from each mixture and loaded to 40 percent of their 28-day average compressive strength after 28 days from the casting date, and another set of two similar cylinders was unloaded and monitored for deformations due to shrinkage and temperature effects as shown in Figure 3. The average temperature and humidity of the room are 20 degrees Celsius and 38%, respectively.
All cylinders were instrumented using three pairs of detachable mechanical (DEMEC) gauges distributed around the cylinders to measure the longitudinal deformations over 8 in. distance using a dial gauge. The deformations for both sets were recorded every day for a week, then every 7 days for a month, and then every 30 days up to 360 days after time of loading for all mixtures except for mixture with gravel and class C fly ash (G-C) up to 270 due to erroneous readings after 270 days. Average creep strains were calculated by subtracting the average deformation of the unloaded cylinders from those of the loaded cylinders to eliminate shrinkage strain. Additionally, measurements from the three pairs of gauges were compared to check the uniformity of loading.  All cylinders were instrumented using three pairs of detachable mechanical (DEMEC) gauges distributed around the cylinders to measure the longitudinal deformations over 8 in. distance using a dial gauge. The deformations for both sets were recorded every day for a week, then every 7 days for a month, and then every 30 days up to 360 days after time of loading for all mixtures except for mixture with gravel and class C fly ash (G-C) up to 270 due to erroneous readings after 270 days. Average creep strains were calculated by subtracting the average deformation of the unloaded cylinders from those of the loaded cylinders to eliminate shrinkage strain. Additionally, measurements from the three pairs of gauges were compared to check the uniformity of loading. Table 5 lists the five creep prediction models used to estimate the creep coefficient of SCC mixtures. Descriptions of all the model parameters are presented in the notations section at the end of the paper.

Drying Shrinkage Testing
The drying shrinkage was measured in accordance with ASTM C157 for all forty SCC mixtures as shown in Figure 4. Three concrete prisms that are 76 × 76 × 286 mm 3 × 3 × 11 1 4 in.) from each mixture were moist cured for 7 days and maintained at 50% ± 4% relative humidity and 23 ± 2 • C temperature until 56 days. The readings were made at 3, 7, 14, 28, and 56 days after the curing period. Table 6 lists the five shrinkage prediction models used to estimate the drying shrinkage strains of SCC mixtures and compared then with the measured values. Descriptions of all the model parameters are presented in the notations section at the end of the paper. mixtures as shown in Figure 4. Three concrete prisms that are 76 × 76 × 286 mm 3 × 3 × 11 ¼ in.) from each mixture were moist cured for 7 days and maintained at 50% ± 4% relative humidity and 23 ± 2 °C temperature until 56 days. The readings were made at 3, 7, 14, 28, and 56 days after the curing period. Table 6 lists the five shrinkage prediction models used to estimate the drying shrinkage strains of SCC mixtures and compared then with the measured values. Descriptions of all the model parameters are presented in the notations section at the end of the paper.  . εshu CEB-FIP MC90-99 [5] εcso βs(t − tc) B3 [6] εsh ͚ kh S(t − tc) GL 2000 [7] β(h) β(t − tc) Table 7 summarizes the workability properties of all SCC mixtures considered in this investigation. SCC mixtures designed for low filling ability had slump flow between 550 and 650 mm (22 and 26 in.), while those designed for high filling ability had slump flow between 650 and 750 mm (26 and 30 mm). T50 was found to be ≤2 s for all mixtures, which indicates the low viscosity of the tested mixtures. Most mixtures had satisfactory passing ability as ∆D is ≤50 mm (2 in.) and ∆H ≤ 15 mm (0.6 in.). Only a few mixtures, mostly with NMSA = 19 mm ( 3 /4 in.), presented higher probability of blockage. All mixtures had adequate filling capacity more than 70%. Most SCC mixtures had adequate static stability as the penetration values (average of two measurements) were less than 25 mm (1 in.) and column segregation percentage was less than 15%. A few mixtures, mostly with NMSA = 19 mm ( 3 /4 in.), had lower static stability as penetration was equal to 25 mm (1 in.) and column segregation percentage was between 15% and 20%, which might be acceptable for some cast-in-place components. The VSI and HVSI for all SCC mixtures were either 0 or 1, which indicated adequate stability. It should be noted that VSI and HVSI are qualitative test methods that depend on the operator judgment; however, the guidelines presented in test standards were followed to minimize the subjectivity of the assessment. Dynamic stability was measured using the flow-through method for only SCC mixtures with high slump flow. Results indicated that most mixtures had exhibited either high dynamic stability (segregation ≤20%) or moderate dynamic stability (segregation ≤30%). Most SCC  Table 6. Drying shrinkage strain prediction models.

Model Name Drying Shrinkage Strain Prediction Equation, ε(t,t c )
AASHTO LRFD [2] k s k hs k f k td 0.48 × 10 −3 ACI 209 [3,4] (t−tc) α f +(t−tc) α . ε shu CEB-FIP MC90-99 [5] ε cso β s (t − t c ) B3 [6] ε sh∞ k h S(t − t c ) GL 2000 [7] ε shu β(h) β(t − t c ) Table 7 summarizes the workability properties of all SCC mixtures considered in this investigation. SCC mixtures designed for low filling ability had slump flow between 550 and 650 mm (22 and 26 in.), while those designed for high filling ability had slump flow between 650 and 750 mm (26 and 30 mm). T50 was found to be ≤2 s for all mixtures, which indicates the low viscosity of the tested mixtures. Most mixtures had satisfactory passing ability as ∆D is ≤50 mm (2 in.) and ∆H ≤ 15 mm (0.6 in.). Only a few mixtures, mostly with NMSA = 19 mm ( 3 / 4 in.), presented higher probability of blockage. All mixtures had adequate filling capacity more than 70%. Most SCC mixtures had adequate static stability as the penetration values (average of two measurements) were less than 25 mm (1 in.) and column segregation percentage was less than 15%. A few mixtures, mostly with NMSA = 19 mm ( 3 / 4 in.), had lower static stability as penetration was equal to 25 mm (1 in.) and column segregation percentage was between 15% and 20%, which might be acceptable for some cast-in-place components. The VSI and HVSI for all SCC mixtures were either 0 or 1, which indicated adequate stability. It should be noted that VSI and HVSI are qualitative test methods that depend on the operator judgment; however, the guidelines presented in test standards were followed to minimize the subjectivity of the assessment. Dynamic stability was measured using the flow-through method for only SCC mixtures with high slump flow. Results indicated that most mixtures had exhibited either high dynamic stability (segregation ≤20%) or moderate dynamic stability (segregation ≤30%). Most SCC mixtures with high slump flow and 3 4 in. NMSA had shown poor dynamic stability, making them inappropriate for long or deep components.  Two parameters were measured to evaluate the rheology of SCC mixtures: (1) yield torque, which represents yield stress; (2) slope of the rheological model, which indicates plastic viscosity. The effects of different types of SCM/filler on the rheological properties were not significant. However, the SCC mixtures with larger coarse aggregate (NMSA = 19 mm) represented higher yield torque compared to those with smaller aggregate (NMSA = 9.5 mm). Additionally, SCC mixtures containing gravel aggregate had higher yield torque and lower viscosity than SCC mixtures containing limestone aggregate. Figure 5 plots the measured creep strain for tested SCC mixtures, while Figure 6 plots the creep coefficient curves for these mixtures. Creep coefficient represents the ratio of the creep strain to elastic strain at a stress level of 40% of the average 28-day compressive strength. The first readings were recorded at the first day after loading. Statistical analysis was conducted to show whether there was a significant difference between predicted and measured creep coefficient ratios when different types of SCMs/fillers were used. Table 8 shows the results in terms of average and variance of predicted-to-measured creep coefficient ratio. Comparing the results of all models indicates that ACI 209 and AASHTO LRFD models slightly underestimate the creep coefficient, while CEB-FIP MC90-99 and GL2000 models significantly overestimate the creep coefficient. shows the results in terms of average and variance of predicted-to-measured creep coefficient ratio. Comparing the results of all models indicates that ACI 209 and AASHTO LRFD models slightly underestimate the creep coefficient, while CEB-FIP MC90-99 and GL2000 models significantly overestimate the creep coefficient.  shows the results in terms of average and variance of predicted-to-measured creep coefficient ratio. Comparing the results of all models indicates that ACI 209 and AASHTO LRFD models slightly underestimate the creep coefficient, while CEB-FIP MC90-99 and GL2000 models significantly overestimate the creep coefficient.    Figure 7 also plots the measured and predicted creep coefficient using different models regardless of the type of SCM/filler. The slope of the line of best fit for data points presents the level of prediction accuracy of each model (slope of 1.0 is the highest accuracy). The coefficient of determination, R 2 , represents the goodness of fit (the higher R 2 , the lower the scatter of the model predictions). This figure indicates that AASHTO LRFD model has the lowest scatter in its predictions, while the GL2000 model has the highest. Table 8 also indicates that in model prediction models, the type of SCM/filler does not have a significant effect on creep coefficient with the exception of mixtures with limestone powder (FLP) that induce higher creep strains, which is in agreement with Heirman [10]. Therefore, it is recommended to use a modification factor greater than 1.0 to adjust creep coefficient prediction for SCC mixtures with limestone powder. The value of this modification factor varies depending on the model used (e.g., 1.2 for AASHTO LRFD model).

SCM/Filler Prediction Model
Number of  Figure 7 also plots the measured and predicted creep coefficient using different models regardless of the type of SCM/filler. The slope of the line of best fit for data points presents the level of prediction accuracy of each model (slope of 1.0 is the highest accuracy). The coefficient of determination, R 2 , represents the goodness of fit (the higher R 2 , the lower the scatter of the model predictions). This figure indicates that AASHTO LRFD model has the lowest scatter in its predictions, while the GL2000 model has the highest. Table 8 also indicates that in model prediction models, the type of SCM/filler does not have a significant effect on creep coefficient with the exception of mixtures with limestone powder (FLP) that induce higher creep strains, which is in agreement with Heirman [10]. Therefore, it is recommended to use a modification factor greater than 1.0 to adjust creep coefficient prediction for SCC mixtures with limestone powder. The value of this modification factor varies depending on the model used (e.g., 1.2 for AASHTO LRFD model).

Drying Shrinkage Strains
Tables 9 and 10 list the results of measured drying shrinkage at different ages in addition to the average compressive strength at 56 days for mixtures with limestone and gravel aggregate, respectively. Compressive strength at different ages was predicted using the ACI 209 equation if the model does not provide a prediction equation. A graphical presentation of the drying shrinkage strains for all 40 mixtures can be found in the Appendices of NCHRP Report 819 [25]. LS = crushed limestone G = gravel F = class F fly ash C = class C fly ash S = GGBFS; FLP = class F fly ash plus limestone powder HF = high flow LF = low flow. Figure 8 shows that all models underestimate the drying shrinkage; however, GL2000 model provides the closest prediction to measured values, which is in agreement with Mokarem [18]. This model shows higher scatter, as evident in the low R 2 value, compared to the other models as reported by Khayat and Mitchell [15]. The B3 model has the lowest prediction accuracy, which is attributed to low sensitivity to compressive strength. Table 11 shows the statistical data for predicted-to-measured drying shrinkage ratios of SCC mixtures with different types of SCM/filler using each of the five prediction models. It indicates that CEB-FIP MC90-99 and GL2000 models do not have a significant difference in shrinkage prediction, and AASHTO LRFD, ACI 209 and B3 models provide approximately similar drying shrinkage predictions.  Evaluation of creep and drying shrinkage prediction models for SCC was conducted using forty SCC mixtures containing different types of coarse aggregate, NMSA, levels of filling ability and types of SCM/filler. Five prediction models were compared using measured data and the following conclusions were made: 1. The AASHTO LRFD model provided better prediction for creep coefficient of SCC with lower scattering of data when different types of SCM/filler were used. On the other hand, using limestone powder increased measured creep strains more than