Early structural build-up, setting behavior, reaction kinetics and microstructure of sodium silicate-activated slag mixtures with different retarder chemicals

.


Introduction
Alkali-activated cements (AACs) are seen as a promising alternative to Portland Cement (PC) [1].This is because a high amount of some industrial wastes, such as ground granulated blast furnace slag (GGBFS), fly ash and metakaolin, can be utilized in their production, and they present better mechanical properties, heat and acid resistance [2][3][4][5].In addition, in the context of global warming, PC is reported to be responsible for 7-9 % of anthropogenic CO 2 production and 3 % of energy use worldwide [6].As a result, alternative binders need to be developed to reduce the environmental impact.
Understanding the rheological behavior of the AACs is important for a successful application of the AACs as it addresses the flowability, stability, pumpability, pouring and casting process of fresh concrete.From the rheology point of view, the structural build-up of AACs materials is an important property to characterize the fresh behavior.In addition, higher structural build-up rates are preferable for reduced formwork pressure and better fresh state stability.It is also desirable for some applications such as slip-form paving and 3D printing of concrete.Conversely, a lower structural build-up is preferred in some cases, such as multi-layer casting applications to increase the bond strength between the layers [7,8].
As well known, the setting time of cementitious materials is of high importance for a successful casting process.The sodium silicate-based activators (sodium silicate and sodium hydroxide) were found to show the best performance on strength development; however, the main shortcoming of the sodium silicate-activated slag cements (AASC) is the rapid set [9].Despite the superior mechanical performance of the AASC activated by sodium silicate-based activators, their quick and uncontrolled setting significantly hinder their common usage in real practice.Therefore, it is vital to find an appropriate chemical admixture to retard the rapid setting of these types of AASC.Chang [9] found that H 3 PO 4 can work as a retarder in sodium silicate-activated GGBFS systems.The author pointed out that incorporating 0.87 M phosphoric acid (H 3 PO 4 ) into the AASC mixture could retard the setting time; however, it reduced the early age compressive strength and increased the drying shrinkage.Some studies also investigated the role of phosphates in the setting of AAC.Kalina et al. [10] reported that the phosphate anion from the trisodium phosphate (Na 3 PO 4 ) used as a retarder would bond to the Ca 2+ ions released from the GGBFS in a highly alkaline environment, thereby forming calcium dihydrogen, later hydrogen phosphate structures.Therefore, it is theorized that the lack of calcium ions in the solution reduces the nucleation and development of the C-S-H phase and hence the initial setting time delays.Gong and Yang [11] also observed the same phenomenon that relatively high sodium phosphate concentration had a strong retarding effect on the setting of the alkali silicate-activated GGBFS-red mud blended system.However, Shi and Li [12] reported no notable retardation effect when Na 3 PO 4 was added to the alkaliactivated phosphorus slag systems.Besides, the use of borates as retarders for PC is well known [13].Yousefi et al. [14] found that borax can improve the fresh state properties of alkali-activated slag/fly ash systems in terms of workability and setting time without negatively affecting the mechanical properties.They also reported that a dosage of 6 % borax was optimal in improving the fresh state properties while keeping the mechanical performance.Revathi et al. [15] also pointed out that both initial and final setting times were prolonged, and the yield stress of AACs decreased in the presence of borax.The same authors revealed that the presence of borate ions significantly influenced the early dissolution of alumina and silicate species due to the reaction of BO 4 tetrahedron with [SiO 4 ], causing the retardation of [AlO 4 ] in the network resulting in increased setting time.Furthermore, some researchers also tried to use organic acids as the retarder in AACs [16,17].Sun et al. [16] investigated the effect of tartaric acid on the early hydration of sodium hydroxide-activated slag pastes.They pointed out that tartaric acid can prolong the initial setting time due to the complexation reaction of Al rather than Ca.
To improve the understanding of the roles of different retarders on the rheology, structural build-up, setting time, strength development and the microstructure of AASC, sodium-silicate activated GGBFS mixtures with citric acid, sodium tetraborate decahydrate, and sodium triphosphate pentabasic were produced and tested in this study.The structural build-up of the AASC pastes with different types of retarders in different dosages was assessed by small-amplitude oscillation shear tests (SAOS).The initial workability and the workability loss of AASC with different retarders were evaluated by a mini-slump test.X-ray diffraction (XRD) analysis and pore solution chemistry by inductively coupled plasmaoptical emission spectrometry (ICP-OES) were applied to find out the mechanism behind the setting characteristics.The porosity and the pore size distribution of the hardened samples were assessed by mercury intrusion porosimetry (MIP).The compressive and flexural strength of the mixtures were also determined to see the effect of retarding chemicals on the mechanical properties.Scanning electron microscopy (SEM) was also employed to observe the morphology of hydrates.

Materials and mix proportions
Ground granulated blast furnace slag (GGBFS) provided by ORCEM, the Netherlands, was used as the precursor in all mixtures.The chemical composition of GGBFS determined by X-ray fluorescence (XRF) is given in Table 1.X-ray diffraction pattern and particle size distribution of GGBFS, determined by a powder diffractometer and laser diffraction, respectively, are shown in Fig. 1.The GGBFS shows a broad diffraction hump between 20 and 35 • 2θ due to its amorphous component structure.The volume-mean particle size (d 50 ) of GGBFS is around 9 μm.The morphology of GGBFS particles observed by scanning electron microscope (SEM) is presented in Fig. 2. As shown in Fig. 2, GGBFS shows an irregular shape with high angularity.
A mixture of liquid sodium silicate solution and sodium hydroxide (NaOH, caustic soda) was used as the alkaline solution.The liquid sodium silicate consists of 28.5 % SiO 2 and 17.95 % Na 2 O with a Ms (the mass ratio of SiO 2 /Na 2 O) value of 1.6.NaOH has 97 % purity, and it was used to reduce the Ms value to 1.2 in the activator solution.It was reported that the sodium silicate activators with a Ms value of around 1.0-1.5 could result in the highest mechanical and durability performance for alkali-activated slag cement [18].For this reason, the sodium silicate activator with a Ms value of 1.2 was used in this study.The Na 2 O content and water to solid binder (w/b) ratio were kept constant as 5 % and 0.45 for all mixtures, respectively.Three types of chemicals, citric acid (CA, C 6 H 8 O 7 ) with purity of 99.0 %, sodium tetraborate decahydrate -Borax (SB, Na 2 B 4 O 7 ⋅10H 2 O) with purity higher than 99.5 % and sodium triphosphate pentabasic (SP, Na 5 O 10 P 3 ) with purity higher than 98.0 % provided by Sigma-Aldrich, were used as retarding chemicals in this study.These three chemicals were added to the activator solution in different amounts (3 %, 4 % and 5 % by weight of the total precursors) one day before, prior to mixing with precursors.The detailed mix proportions are shown in Table 2.The AASC pastes for the rheological measurements were mixed by using a rotational rheometer with a helix geometry [19].The mixing protocol consisted of two stages: firstly, the shear rate was gradually increased from 0 to 3000 min − 1 within 30 s, and then the mixing speed was kept constant at 3000 min − 1 for 120 s.For the other tests, the paste samples were prepared in a Hobart mixer.GGBFS was added to the activator solution in the Hobart mixing bowl and mixed at low (140 ± 5 rpm) and high (285 ± 5 rpm) speeds for 90 s each.

Rheological properties and setting times 2.2.1. Setting times
The setting times were measured in accordance with EN 196-3 [20] with an automatic Vicat needle apparatus.The initial setting time was determined when the distance from the needle to the bottom plate was ± 3 mm.The elapsed time from the mixing was recorded as the final setting time when the penetration was only 0.5 mm.

Flow table
The flow spread of the pastes was determined by the mini-slump tests according to the ASTM C1437-15 [21].A truncated cone with a height of 40 mm and top and base diameters of 46 mm and 68 mm, respectively, was used in the mini-slump test.Mini-slump spread diameter (%) values were calculated by Eq. (1).

Structural build-up
A rheometer (MCR 102, Anton Paar, Austria) with a six-blade vane geometry was used for structural build-up tests.A plastic lid was used to minimize the evaporation during the measurement time.The rheological measurements were conducted under a constant temperature of ± 0.5 • C and each mixture was tested on three newly prepared samples to check the repeatability of the tests.The presented curves are the representative curves having the closest rheological values to the average of three tests.
The structural build-up of AASC is the consequence of the transformation of the solid precursors (GGBFS in this study) into synthetic calcium aluminosilicate gels when the alkaline activator reacts with precursors.The phenomenon of this structural build-up can be evaluated by small-amplitude oscillation shear tests (SAOS) in a non-destructive regime within the linear viscoelastic region (LVER).The storage modulus (G'), which represents the elastic portion of the viscoelastic behavior of the pastes were determined to evaluate the structural buildup of cementitious materials [22][23][24].Therefore, SAOS tests were performed at a constant frequency of 1 Hz and a strain amplitude of 0.005 %, which was consistent with the order of 10 − 5 used in Yuan et al. [7] and Mostafa et al. [22].

Calorimetric tests and mechanical properties
The hydration kinetics of the paste samples were measured using an eight-channel TAM air isothermal calorimeter (TA Instruments, USA).After mixing, 14 g paste sample was immediately loaded into a glass ampoule bottle and then the sealed bottle was put into the isothermal calorimeter.Calorimetric tests were conducted at a constant temperature of 20 ± 0.02 • C for 7 days.
Mechanical properties of the mixtures were determined on mortar mixtures.All mixtures had the same aggregate to binder ratio of 3. Standard CEN sand was used as aggregate.The fresh mortars were poured into plastic molds of 40 × 40 × 160 mm 3 and vibrated for 1 min, then covered with a plastic film for 24 h.After demolding, all mortar samples were cured under 20 • C temperature and 95 % relative humidity conditions until the testing day.Flexural and compressive strength of the samples were determined according to EN 196-1 [25] at the age of 2, 7 and 28 days.

Microstructural characterization
In this study, four mixtures (REF_Ms1.2,CA4, SB4 and SP4) were selected for the microstructural characterization of AASC mixtures.The following tests were carried on these mixtures.

Fourier transform infrared spectroscopyattenuated total reflectance (FTIR-ATR) and pH value measurements
FTIR-ATR measurements were conducted on some selected activator solutions.Before each measurement, the background was first scanned to check the cleanness of the ATR.During measurements, several drops of the activator solution were placed on a 3 mm diameter diamond/ZnSe crystal of attenuated total reflector (ATR).The spectra resolution was 1 cm − 1 and frequencies were scanned in the range of 4000-400 cm − 1 .The pH of the activator solutions were determined by a pH meter.

Pore solution chemistry
To investigate the retarding mechanism of the used chemicals, the pore solution chemistry of AASC mixtures at the early ages was analyzed by Inductively Coupled Plasma -Optical Emission Spectrometry (ICP-OES, 720 ES Varian/Agilent) on selected four AASC mixtures (REF_Ms1.2,CA4, SB4 and SP4).The pore solutions of these mixtures were obtained by the centrifugation of paste samples at different early ages.The obtained pore solutions were filtered by a syringe connected to a disposable 0.45 μm syringe filter.The pore solutions were diluted with   pure water.The dilution ratios for determining low concentration elements (Al, Mg, Ca and Fe) and high concentration elements (Na, Si, B and P) were 1:20 and 1:4000, respectively.ICP -OES analyses were conducted on the pore solution samples as quickly as possible on the same day to prevent any salt precipitation.The ion concentrations of the activator solutions were shown at zero time in the ICP analysis results.

X-ray diffraction (XRD)
The samples were taken from the hardened paste samples at the age of the initial set and 7 days.Following the RILEM TC-238 methodology, 3 g of the crushed pieces with a size between 125 μm and 1 mm were mixed with 100 ml isopropanol for 15 min.The suspension was filtered and rinsed with isopropanol and diethyl ether before drying at 40 • C for 8 min.Then the dried samples were stored in a low vacuum desiccator before the analysis.The XRD measurements were conducted on a Rigaku D/Max-2200/PC X-ray diffractometer with CuKα radiation (λ = 0.1542 nm) at 40 kV and 36 mA, scanning from 5 • to 70 • 2θ with 0.02 • step size.

Scanning electron microscopy (SEM) and energy dispersive X-rays spectroscopy (EDX)
Paste samples at the age of 28 days were immersed into isopropanol for one week to stop the reaction and dried in a 40 • C oven for 1 h for the SEM studies.The dried samples were immersed into epoxy resin and then polished up to 0.25 μm surface fineness using the diamond paste.
The surface of the samples was coated by carbon subsequently dried in a low vacuum desiccator prior to analysis.The specimen preparation protocol used in this study is reported elsewhere in detail [8,26].Afterwards, the fractured and polished samples were observed by SEM with the secondary electron (SE) mode and the backscattered electron (BSE) mode at an acceleration voltage of 15.0 kV under a low vacuum, respectively.BSE images were used for the EDX analysis.

Mercury intrusion porosimetry
At the testing ages of 28 days, the paste samples were crushed into small pieces with dimensions of around 1 cm 3 , and then these small sample pieces were immersed in isopropanol for one week to stop the reaction of the slag and dried in a 40 • C oven for 1 h.The dried samples were stored in a low vacuum desiccator before the analyses.A Pascal 440 mercury porosimeter with a maximum load capacity of 420 MPa was used in the MIP test.However, the maximum pressure was limited to 200 MPa in order to avoid the cracks induced by the mercury pressure [27,28].The adopted mercury surface tension and contact angle between the mercury and the solid surface were 0.482 N/m and 142 • , respectively.

Workability loss, setting times and mechanical performance
The influence of different retarders on the spread diameter (%) of AASC pastes is given in Fig. 3.As shown in Fig. 3, the AASC mixture without any retarding chemicals (REF_Ms1.2) presented a rapid workability loss within 20 min and lost all workability at the age of 27 min.The sodium triphosphate pentabasic (SP) addition provided a higher initial spread diameter as compared to the reference mixture, but SP addition could not provide negligible workability retention at all used dosages.Increasing the SP dosage from 3 % to 5 %, the time to complete loss of workability just prolonged from 33 min to 39 min.These findings are also consistent with the previous studies [14], showing that the SP can only slightly increase the time to retain the flowability of the pastes.
The organic citric acid (CA) addition improved the initial workability and provided longer workability retention as compared to the control mixture and the SP.The highest initial workability among CA mixtures was obtained at 3 % dosage (CA3).However, CA3 presented a sudden workability loss after around 25 min.At 4 % CA dosage, the initial workability was slightly lower than that of CA3, but the mixture kept its initial workability longer than CA3.In addition, the workability loss observed in CA4 was more gradual as compared to CA3.Increasing the CA dosage to 5 % (CA5) caused a slightly lower initial workability compared to REF_Ms1.2; however, the workability loss was more gradual, and the complete loss of workability in CA5 mixture lasted the longest among all tested mixtures (around 115 min).On the other hand, it should be noted here that the flow diameter of CA5 was significantly lower than CA4 in the first hour.
Sodium tetraborate decahydrate (SB) led to a relatively appropriate initial flow diameter and workability retention as compared to SP and CA.The initial spread diameter and the workability retention within the first 40 min of SB3, SB4 and SB5 were significantly higher than REF_Ms1.2AASC mixture.SB3 showed a sudden workability loss after 40 min and lost its all flowability at 58 min.SB4 and SB5 also showed significant workability loss upon 40 min as well but their workability loss rate were slower than SB3.The time for the complete loss of workability was prolonged from 58 min to 75 min when the dosage of SB increased from 3 % to 5 %.Yousefi et al. [14] also found that the mixtures with SB had a better workability retention as compared to the mixtures with SP addition, which was also consistent with the results of storage modulus (as presented in Section 3.4) in this study.
Table 3 presents the initial and final setting times of the tested AASC mixtures.As can be seen from the Table 3, REF_Ms1.2showed rapid initial and final setting times of 46 min and 150 min, respectively.Although there is no well-defined or suggested minimum initial setting time for AASC yet, it is for sure that an AASC paste with a setting time shorter than 60 min would not be appropriate for general use.Furthermore, even much longer setting times would be more suitable for the common use of the AASC for general purposes.In this respect, revealing the performance of different types of retarding chemicals in extending the short setting times of sodium silicate-activated AASC  without a detrimental effect on strength development is important.Table 3 clearly shows that all of the tested three different types of chemicals retards the setting times of sodium silicate (Ms 1.2) activated AASC to different extents.It is observed that their influence on final setting time was more pronounced than the initial setting time.
As can be seen in Table 3, similar to the results of mini-slump, the incorporation of SP at 3 % dosage could prolong the initial setting time from 46 min to 70 min, and the further increase up to 5 % did not significantly prolong the initial setting, ranging from approximately 70 min to 76 min.The same tendency was also observed in the final setting times, which fell between 270 min and 277 min.The CA and SB were found to be more dosage-sensitive in increasing setting times.The CA with 3 % dosage prolonged the initial and final setting times 23 min and 243 min, respectively, compared to REF_Ms1.2.These extensions in initial and final settings were 80 min and 275 min, respectively, at 4 % CA dosage.The further increase of the CA dosage to 5 % caused extremely long setting times (exceeding four days for the initial setting time).The addition of SB at 3 % extended the initial and final setting times 40 min and 242 min compared to the reference.The further dosage increase to 5 % resulted in the extension of the initial and final settings 74 min and 526 min compared to the reference.An overall evaluation of the setting times results presented in Table 3 reveals that the boraxbased retarder presents more reasonable setting times at different usage dosages.It also allows the dosage-based arrangement to reach the desired setting properties in AASC design.In previous literature, Yousefi et al. [14] also indicated that the setting time increased almost linearly as the borax content was increased from 2 % to 8 % of the weight of precursors.
The flexural and compressive strengths of AASC mixtures are presented in Table 4.It was obvious that the flexural and compressive strength for all mixtures increased with age.It could be seen that the CA incorporated mixtures did not gain any measurable strength at 2 days.These mixtures could gain flexural and compressive strength at the ages of 7 and 28 days, but the obtained strength levels were much lower than the REF_Ms1.2.The low mechanical performance of CA incorporated mixtures was more pronounced in high dosage cases, probably due to the decrease in pH value of the activator solution by the dosage increase.Therefore, it can be concluded that the mixtures with CA addition were far inferior to REF_Ms1.2 in terms of mechanical performance.
Interestingly, among all mixtures, the best mechanical performances were obtained from SP incorporated samples.The compressive strength of SP mixtures was higher than 40 MPa at the age of 2 days in all used dosages.The flexural and compressive strength values even slightly increased with a higher dosage of SP.The mixtures with the addition of SB showed a lower flexural and compressive strength than that of REF_Ms1.2 at 2 and 7 days.However, SB3, SB4 and SB5 showed higher mechanical properties than REF_Ms1.2 at 28 days.These results are also parallel with the results of isothermal calorimetric tests (Section 3.2).

Reaction kinetics of AASC pastes
Calorimetric curves for AASC pastes with different retarders are presented in Fig. 4. It has been reported that the calorimetric curve for sodium silicate-based GGBFS mixtures has five distinctive stages: dissolution, induction, acceleration, deceleration and steady-state [26,29,30].
In general, during the first 20 min a sharp local peak is observed for all sodium silicate-based GGBFS mixtures.The heat flow evolved rapidly upon contact between the activator and the GGBFS was mainly attributed to wetting and dissolution of GGBFS particles and partly the interaction of silicate units and sodium with calcium dissolved from GGBFS [31,32].After the dissolution stage, the heat flow for the AASC pastes starts to show an induction period.The main calorimetric peaks after the induction period for the AASC pastes are usually associated to the strength and microstructural development [33].
It could be seen that CA was the most effective retarder to prolong the induction period of AASC mixtures.As compared to REF_Ms1.2, the onset of the acceleration stage was delayed from 14.9 h to 30.5 h and 56.5 h for 3 % and 4 % CA, respectively.However, such a pronounced retarding effect of CA costs the loss of strength development (Table 4).Following the compressive strength development trend of CA incorporated mixtures, the induction period and time to reach the end of acceleration period were longer than the REF_Ms1.2 and the intensity of the main calorimetric peak was much lower as compared to that of REF_Ms1.2.It should be noted that when the dosage of CA reached to 5 %, no second calorimetric peak was observed during the measurement time.Therefore, the heat evolution of the mixtures was very sensitive to the CA dosages, which was also consistent with the mechanical performance of the mixtures incorporated with CA.
SP showed a converse trend as compared to CA in terms of heat flow.As shown in Fig. 4, in the SP incorporated mixtures, the onset of the acceleration period occurred much earlier than that of REF_Ms1.2 and this was followed by a very sharp increase in the intensity of the main calorimetric peaks.These observations agree well with the better mechanical performance of SP incorporated mixtures as compared to REF_Ms1.2.
No very significant heat flow differences as compared to REF_Ms1.2 were observed in SB incorporated mixtures in the first 24 h.The time of the end of the acceleration period was achieved earlier than that of REF_Ms1.2;however, the intensity of the main peaks of SB incorporated mixtures was slightly lower as compared to the mixture REF_Ms1.2.These were also consistent with the compressive strength results at early ages.
Regarding the cumulative heat release, the cumulative heat release of REF_Ms1.2 reached to 131 J/g at 7 days (Fig. 4b) and the cumulative heat release of the mixtures with CA was significantly lower than the reference mixture without CA.As can be seen from Fig. 4b, increasing the CA dosage from 3 % to 5 % resulted in the decrease of cumulative heat release at 7 days from 107 J/g to 13 J/g.This shows that the CA addition is very effective in slowing down the GGBFS reactions and its dosage should be limited in a certain range, otherwise the activator solution would not activate the reaction of GGBFS.The inclusion of SB from 3 % to 5 % can only slightly decrease the cumulative heat at the age of 7 days from 121 J/g to 114 J/g.The trend in the case of SP was inversed.The cumulative heat at the age of 7 days increased from 141 J/ g to 149 J/g, supporting its better mechanical properties as compared to REF_Ms1.2.
To better correlate the initial setting time with the reaction kinetics of AASC pastes with different types of chemical retarders, the first three hours of heat flow and cumulative heat curves of four mixtures (REF_Ms1.2,CA4, SB4 and SP4) were measured and plotted in Fig. 5.As shown in Fig. 5a, after the dissolution peak, an additional exothermic peak was observed for each AASC mixture, as also reported by Gebregziabiher et al. [26] and Shi et al. [33].It was observed that these additional exothermic peaks occurred essentially before the initial setting times of AASC pastes.The arrows in Fig. 5b show the cumulative heat values of AASC pastes with different types of chemical retarders when they reached initial setting times.As could be seen from Fig. 5b, REF_Ms1.2,SB4 and SP4 exhibited their cumulative heat around 17 J/g at the initial setting time.However, the trend was slightly different for CA4 as its cumulative heat showed only 12 J/g at the initial setting time possibly due to the low reaction rate of the mixture leading to the physical effects, such as water loss by evaporation, more dominant in setting time test.Overall, it is therefore likely that these additional exothermic peaks made an important contribution to the initial setting of AASC mixtures.

FTIR-ATR and pH value analysis of activator solutions
The FTIR-ATR and pH value measurements were conducted on four selected activator solutions (REF_Ms1.2,CA4, SB4 and SP4), and their results are presented in Fig. 6 and Table 5, respectively.As shown in

Table 5
The pH values of selected activator solutions.The spectra shift towards higher wavelengths compared to the REF_Ms1.2spectra shows the increase in Si -O -Si bond strength for the activators with chemical retarders [34].New peaks were formed for the solutions CA4 and SP4 while some broad humps were appeared in the spectra of SB4 solution, showing the formation of new bonds.In Fig. 6a, the additional peak located at 1590 cm − 1 could possibly linked to C--C aromatic stretching vibrations as reported in ref. [35,36].Another peak at 1392 cm − 1 could be attributed to Si -OH bonding [36].Besides, the observed peak at 1280 cm − 1 was normally related to the Si-C bonding [37], indicating the citric acid was modified by silicate species in the alkaline solution.It should be noted that there is a significant decrease in the intensity of the band region between around 900 cm − 1 and 950 cm − 1 .This is possibly due to the H + released from CA consumes small silicate anions in the sodium silicate solution [34].
As shown in Fig. 6b, SB4 showed traces of shoulders at around 950 cm − 1 , 1100 cm − 1 and 1200 cm − 1 .The wide absorption band at around 1100 cm − 1 can be attributed to tri-, tetra-, pentaborate, and diborate groups belonging to the BO 3 and BO 4 groups, as well as asymmetric stretching from Si-O-Si [38].The band at around 950 cm − 1 should be associated with the B -O vibration of BO 4 units and was also related to the stretching frequency of Si-O-B [39].While the region centered at 1200 cm − 1 was possibly assigned to the stretching vibration of B -O of the BO 3 units from the boroxol rings.As for the SP4 activator shown in Fig. 6c, three additional peaks at 907 cm − 1 , 1104 cm − 1 and 1209 cm − 1 were also shown for the SP4 activator.The band at 1104 cm − 1 can be related to the bridging stretching Si -O -P and Si -O -Si vibrations [40].While the band at 907 cm − 1 was possibly assigned to symmetric P -O vibrations [41].Furthermore, presence of the band at 1209 cm − 1 can be assigned to the P--O structure [41].These evidences showed that the addition of chemical retarders changed the various types of bridging bonds in the activator solution.
Table 5 presents the pH value of these four mixtures.CA4 activator had the lowest pH value as compared to other mixtures, indicating that CA4 led to less dissolution of GGBFS, which was also consistent with its longest setting times results.REF_Ms1.2 showed the highest pH value, and it had the quickest setting times.The pH value of SB4 and SP4 was 13.22 and 13.35, respectively, which was lower as compared to REF_Ms1.2.This also indicates the addition of SB and SP has retarding effect on dissolution of GGBFS.

Structural build-up of AASC pastes
It has been reported that SAOS measurement can successfully evaluate the structuration process of cementitious materials [19,42].Thus, the mixtures used in this study were tested by the SAOS method to monitor the structuration process of AASC mixtures through the storage modulus measurements.Fig. 7 shows the storage modulus evolution of CA, SB and SP incorporated mixtures comparatively with REF_Ms1.2.As can be seen from Fig. 7, the storage modulus evolution of AASC pastes has two phases: Phase one defines a gradual development of storage modulus evolution until the onset of another regime in storage modulus evolution which is characterized by a sharp and continuous increase of storage modulus (phase two).For the mixture REF_Ms1.2, the first phase occurred before 20 min, where there was negligible growth in storage modulus.In this phase, the colloidal interaction forces between particles were dissipated owing to the viscous nature of the activator solution [43].In the meantime, due to the high silicate concentration in the Ms1.2 solution, Fig. 7.The evolution of structural build-up of AAC pastes (a) CA addition, (b) SB addition and (c) SP addition.
X. Dai et al. the silicate anions would adsorb on the particle surface, increasing the magnitude of repulsive double-layer electric forces [44].Thus, the storage modulus was negligible at an early age due to the plasticizing and deflocculating effects.Phase two was between 20 min to 45 min.A sharp increase in the storage modulus was observed during such a short period of time.This is probably due to the continuous dissolution of GGBFS in phase one, letting the ions in the suspension gradually saturate so that the interactions between different ions in phase two occurred rapidly and started to form a rigid network.The measurement was stopped when the storage modulus reached the limits of the measurement system of 400 kPa [45,46].
As can be seen from Fig. 7, the addition of CA, SB and SP made a significant change on the onset of the sharp increase in storage modulus due to their retarding effect.As can be seen from Fig. 7a, the period of phase one slightly increased and the rigidification rate of storage modulus in phase two significantly decreased by the increasing dosage of CA.Previous studies reported that H + provided by CA reduced the alkalinity of activator solution [17], slowing down the heat evolution.Similar to this, the CA addition can also delay the increase in storage modulus and setting times in this study.
The incorporation of SB significantly increased the period of phase one and delayed the main increase in storage modulus in phase two as compared to that of CA addition (Fig. 7b).On the other hand, unlike the CA addition, the incorporation of SB did not influence the structuration rate significantly, indicating that SB would only cause a delay on the onset of structuration but not change the microstructure of AASC pastes during the structuration phase.Yousefi et al. [14] also found that a higher amount of SB (i.e., 8 % weight of precursors) can also improve the workability and setting times without any deterioration on compressive strength.This also indicates that SB has great potential to retard the hardening rate of AASC even at a high dosage.
In the presence of SP, AASC pastes also exhibited a more extended period of phase one.However, the retarding effect of SP on AASC pastes was not as strong as SB.The mixtures with 4 % and 5 % SP showed an almost similar trend of structural build-up, indicating that the excessive amount of SP cannot effectively retard the structuration process of AASC.

Pore solution chemistry and its relation to the structural build-up of AASC pastes
Four mixtures (REF_Ms1.2,CA4, B4 and SP4) were selected to investigate the reaction process of AASC at an early age.The relationship between the ion concentrations of Si, Na, Al, Ca, Mg, B, P and Fe in the pore solution and structural build-up of AASC as a function of time are presented in Fig. 8.The ion concentrations at zero time show the ion concentrations in the activator solution itself.As well known, on the one hand, the measured ion concentrations of the pore solution show the remaining part of the dissolved ions from the GGBFS or the existing ions in the pore solution after their consumptions to produce the reaction products [8].On the other hand, it should be noted that the early structural build-up of AASC paste is a physico-chemical process with a combined result of colloidal interactions and chemical reaction processes, which is similar to PC system [47,48].Nevertheless, the Fig. 8.The relationship between the evolution of structural build-up and pore solution chemistry of AAC pastes (a) REF_Ms1.2,(b) CA4, (c) SB4 and (d) SP4.
X. Dai et al. correlation between pore solution chemistry and early structural buildup is very helpful to understand the dissolution kinetics and setting process of AASC.
According to Fig. 8, the pore solution chemistry of all AASC pastes was dominated by Na and Si, whereas much lower P, B, Al, Ca, Mg and Fe were identified.As explained in Section 3.4, the evolution of storage modulus can be divided into two phases.In phase one, there was a negligible increase in storage modulus as the GGBFS particles were well dispersed in the activator solution.This dispersion effect leads to more interaction of the alkaline solution and GGBFS grains, thereby promoting the dissolution.Consequently, the concentration of Al, Ca, Mg and Fe increased to a peak value at the end of phase one due to the ongoing dissolution of GGBFS.While the main elements in pore solution, Na and Si, did not change significantly in phase one.Once the dissolved ions gradually reached the oversaturated state, the solid reaction products started to form, leading to the onset of the second phase.In this phase, a significant increase in storage modulus and the concomitant dramatically decrease in the concentration of Ca and Si was observed, probably due to the fast precipitation of C-S-H consuming Si and Ca in the pore solution [49].Palacios et al. [45] also determined the saturation indices from the elemental concentration determined by ICP-OES to predict the solid reaction products that would form.Their thermodynamic calculations showed that M-S-H, C-N-A-S-H, hydrotalcite, siliceous hydrogarnet (Ca 3 Fe 2 (SiO 4 ) 0.84 ⋅(OH) were at all-time oversaturated, reflecting that these solids could potentially form.These solid reaction products could also possibly form in this study as the composition of their AAC mixtures was quite similar to REF_Ms1.2.Correspondingly, it could be seen that the concentration of Na, Fe, Al and Mg also decreased in phase two as some potential solids started to form in this stage.Favier et al. [50] also showed that after some time, the decreasing trends of Al and Na in the liquid phase were concomitant, indicating the Al was taken up in the tetrahedral position of aluminosilicate phase and Na compensated for the charge balance.Their findings support the results of this study.
In general, the mixtures with the addition of CA did not significantly prolong phase one, and the concentration of Ca and Al were lower than that of REF_Ms1.2.Also, the mixtures with CA addition spent more time in phase two than that of REF_Ms1.2, indicating that CA4 had a lower structural build-up rate.Correspondingly, it could be seen that the concentrations of Ca and Si kept constant or slightly decreased in this stage, denoting less reaction products would form.This indicates that CA addition reduced the pH value of the activator solution, and thus the dissolution rate of GGBFS decreased.Eventually, the reaction process of CA4 was delayed.
When the pore solution chemistry of REF_Ms1.2 and SB4 mixtures were compared (Fig. 8a and c), it could be seen that the occurrence of the peak concentrations of Ca, Al, Mg and Fe was delayed in SB4 mixture, and their concentrations first quickly increased continuously in phase one.Herein, although the peak of Al was not easily visible in the log scale, its concentration actually increased to the maximum value (45 mmol/l) at the age of 45 min, and then decreased to 28 mmol/l at the age of 120 min.While the concentration of B rapidly decreased initially and continuously declined through the measurement time, more specifically, the B concentration rapidly decreased from 649 mmol/l (at zero time) to 444 mmol/l (at the age of 10 min), and then slightly decreased to 417 mmol/l (at the age of 120 min).In previous literature, it was reported that the reason for the retardation effect of SB in Portland cement systems had been shown as the slowing down of hydration due to the calcium ion consumption by borax to form a calcium-based borate layer, which prevents the formation of reaction products [51].Revathi et al. [15] reported that the higher early dissolution of alumina and silicate species was significantly disturbed by the presence of borate ions in the AACs pastes due to the reaction of BO 4 tetrahedron with [SiO 4 ], so that the retardation of [AlO 4 ] in the network was effected and thus setting time of the mixtures prolonged.For this reason, the borate in the pore solution would also possibly interact with Ca and Si ions, thereby delaying the structural build-up and setting of AACs.
The Mixture SP4 presented a similar retarding mechanism with mixture SB4, characterized by a delayed reaction product formation due to the consumption of Ca ions from precursors by the SP.The consumption rate of Ca decreased for SP4 as compared to REF_Ms1.2 as shown in Fig. 8.This finding also supports the retarding effect of SP.As shown in Fig. 8d, the concentration of P slightly decreased first and then almost kept constant through the measurement time.This could be because the Ca ions dissolved from the GGBFS would bond with the phosphate anion from the SP.Previous studies [52] stated that the phosphate anions have strong affinities to Ca cations.Shi and Day [33] furthermore indicated that the formation of Ca 3 (PO 4 ) 2 retarded the activation of GGBFS as usually observed during the hydration of PC.Besides, Kalina et al. [10] used Raman spectroscopy and X-ray photoelectron spectroscopy to investigate the effects of Na 3 PO 4 on the reaction process of alkali-activated GGBFS system and showed that the calcium hydrogen phosphate phases, which prolong the initial setting time by delaying the growth of C-S-H due to Ca ion consumption, was first formed in the system.However, it was reported that these phases were not stable and they dissolved by time and the formation of less soluble phases such as secondary C-S-H gel and calcium hydroxyapatite took place.
Moreover, Dupuy et al. [53] recently used magic angle spinning nuclear magnetic resonance (MAS NMR) to study the structural evolutions of geopolymers.The authors tried to correlate the setting time evolutions with the NMR spectral deconvolution results.They analyzed 31 P results and found that the presence of phosphate groups (PO 4 3− ) and P Q 1 (bonds to Al and/or Si) were supposed to be at the terminal of the aluminosilicate chains, leading to partial geopolymerization.Therefore, this can induce the formation of a secondary network, resulting in an extended setting time.Although the situation in the boron-based system was more complicated as the boron atoms in a tetrahedral environment (B IV ) were connected to silicon tetrahedrons, the increasing setting time was attributed to discrepancies between the geopolymer network and the secondary network, which were related to the B IV quantity.Furthermore, the presence of asymmetric B III with one nonbridging oxygen among the residual B III is recognized in relation to the declined pH value, which also contributed to the increase in setting time.

X-ray diffraction of AASC pastes
The XRD patterns of AASC pastes at the age of initial set and 7 days are shown in Fig. 9.As shown in Fig. 1a, GGBFS pastes are mainly amorphous with a diffraction hump between 25 and 35 2θ, similar to their precursor (Fig. 1a).The main diffraction peak corresponding to C-A-S-H type gel for the AASC pastes are difficult to observe at the age of the initial setting time.Previous studies [54] also reported the same observation that the C-A-S-H gel was difficult to be identified at an early age.No new phase formations around the setting time of the mixtures REF_Ms1.2 and CA4 could be identified by the XRD.However, some crystalline traces were observed for sample SB4, indicating the formation of a type of calcium-based borate mineral, colemanite (Ca 2 B 6 O 11 ⋅5H2O).This is also consistent with the results of pore solution chemistry that the concentration of B (boron) decreased in the first 10 min (Fig. 8c).Similarly, whitlockite (Ca 3 (PO 4 ) 2 ) traces were also observed for the mixture SP4, which was also in agreement with the decrease in P as shown in Fig. 8d.In previous studies, the researchers also pointed out that the formation of Ca 3 (PO 4 ) 2 has a retarding effect on the reaction of AACs [11].
As can be observed from Fig. 9b, the peaks around 29.5 • in 2θ were visible for all the pastes [14,30,55], indicating the formation of poorlycrystalline C-A-S-H at the age of 7 days [56].It could be seen that CA4 showed a poorly crystalline structure for C-A-S-H as compared to the other mixtures.This was also consistent with the finding that CA4 could    develop much lower strength at the age of 7 days (Table 4).

Porosity of AASC pastes
Fig. 10 shows the pore size distribution of AASC pastes determined by MIP at the age of 28 days.In general, AASC paste had a very fine pore structure except for mixture CA5 and CA4.It could be seen that with the increasing addition of CA, the cumulative pore volume of AASC pastes increased.In particular, when the dosage of CA reached 5 %, the microstructure of AASC pastes became very porous as compared to other mixtures.This is also consistent with the results of mechanical performance as presented in Table 4.As can be seen in Fig. 10b, the mixtures with the addition of SB showed much smaller cumulative pore volume than that of mixtures with CA addition.In addition, these mixtures incorporating SB exhibited very similar or slightly smaller cumulative pore volume as compared to that of REF_Ms1.2, indicating there was no degradation of the microstructure of the AASC with the addition of SB.However, as can be seen in Fig. 10b, Borate incorporated mixtures have a somewhat coarser pore size distribution as compared to REF_Ms1.2.
Furthermore, SP showed a significant positive effect on the reduction of porosity, enhancing the microstructure of the AASC pastes.Unlike the other two chemical retarders, with increasing dosage of SP, the cumulative pore volume of AASC mixtures even decreased, which was also in agreement with the compressive strength.A denser microstructure of AASC pastes could be formed due to the addition of SP.As shown in Table 4 and Fig. 10, the addition of SP or SB can increase the mechanical performance and reduce the porosity of AASC pastes.This is possibly because these chemical admixtures would also participate in the alkali-activation system [14].Antoni et al. [57] have pointed out that the flash set of two-part class-C fly-ash-based geopolymers could be prevented by adding anhydrous borax with a significant increase in the compressive strength of the geopolymer.A few researchers also confirmed that when borax was used as a part of alkaline activator for geopolymer pastes, the flexural strength increased due to the formation of B-O bonds in the reaction products [58,59].Furthermore, Revathi and Jeyalakshmi [15] reported that using an appropriate dosage of borax into the GGBFS-Fly ash system could slightly increase the compressive strength, implying that competition between [BO 4 ] and [AlO 4 ] in the network of the silicate ended up with a lower degree of Al incorporation.Additionally, the very high reaction rate of Ref_Ms1.2 may have retarded the subsequent reactions and produced a nonuniform distribution of reaction products as compared to those mixtures with chemical retarders (SB and SP).

SEM/EDX of AASC pastes
SEM/EDX analyses were carried out on the samples REF_Ms1.2,CA4, SB4 and SP4.Backscattered electron imaging (BSE) and secondary electron imaging (SE) of AASC pastes at the age of 28 days are shown in Fig. 11.It should be noted that the highlighted white regions in SEM (BSE) images indicate the unhydrated GGBFS particles, the grey regions between the particles denote reaction products and the black regions represent the pores and micro-cracks.In general, from the morphology point of view, it seems that the binding paste in REF_Ms1.2,SB4 and SP4 are denser and present a more homogeneous morphology, while the CA4 sample shows heterogeneously distributed cavities, indicating a more porous microstructure.Besides, the amount of unhydrated GGBFS particles of SP4 seemed to be lower than in other mixtures.These are also in agreement with the MIP results as presented in Section 3.6.Similarly, as shown in SE images, SP4 showed a better packing and a denser microstructure, and this was followed by SB4, REF_Ms1.2 and CA4.
The EDX results obtained for at least 15 points selected from within the binder region (BSE image) of AASC pastes are shown in Table 6 and Fig. 12.In literature [60,61], researchers have demonstrated that a reduction of the Ca/Si ratio has a markedly beneficial impact on the mechanical properties of cementitious materials.It is believed that the higher molar quantities of C-S-H gel per volume were obtained when the Ca/Si ratio is low [60].Indeed, the SP4 mixture had the lowest Ca/Si ratio of 0.91 but presented the best mechanical performance among all mixtures.
The Al/Si ratio of SB4 mixture was the lowest as compared to other mixtures, meaning that less Al was incorporated into C-A-S-H structure.This may be related to the incorporation of 4-coordinated boron into the C-A-S-H structure.These interactions have been characterized by NMR studies by Du et al. [62,63], who identified the chemical shift related to the B -O -Si bonds according to the number of silicon atoms associated with each boron.In the case of SP4, some researchers [64] also reported that Al 3+ ions induce a modification of P -O -P bonds that are partly replaced by Al-O-P bonds.Other authors [65,66] also pointed out that phosphorous atoms could be bonded to a different number of Aluminum atoms which could also be linked to the silica network.This might be the reason why SP enhanced the mechanical performance of AASC.Indeed, the SP4 also showed the highest Na/Si ratio of all mixtures as the phosphate groups required a large amount of alkali for charge compensation [65].
As for magnesium, it is generally known that they would participate in the formation of hydrotalcite type sub-micrometer phase as reported previously by Wang and Scrivener [67].However, the hydrotalcite reaction products were not detected by XRD carried out at the age of 7 days, as presented in Section 3.5.It could be possibly due to the fact that the hydrotalcite would start to form at a later age when a high degree of GGBFS reaction was reached, as reported by Song and Jennings [68].
Calcium, aluminum and silicon contents of samples with 28 days of curing are also renormalized to 100 % on an oxide basis for plotting on the ternary graph (i.e., neglecting the other elements present).These plots clearly show the variation in the gel composition in terms of the effects of different chemical retarders.It could be seen that the major reaction product for these four mixtures was a chained-structure C-A-S-H, which was commonly recognized in the alkali-activated GGBFS system [69].

Conclusions
This study compares the effects of different chemical retarders on the early structural build-up, setting behavior, reaction kinetics, microstructure and mechanical properties of sodium silicate-activated GGBFS mixtures.The following conclusions can be drawn from this study: • The retarding effect of citric acid (CA) was found the most predominant, and it is followed by sodium tetraborate decahydrate (SB), then sodium triphosphate pentabasic (SP) in terms of flowability and setting times.However, the addition of CA had a negative effect on the mechanical performance, porosity and microstructure.The inclusion of SP could promote more heat release and significantly enhance the compressive strength at an early age.By considering fresh and hardened state properties together, SB has been found as the most appropriate chemical retarder as compared to CA and SP for sodium-silicate activated AASC mixtures tested under the experimental conditions used in this study.• The induction period of AASC paste could be significantly extended by the incorporation of CA and shortened by the addition of SP.The use of SB showed a very similar heat flow curve as compared to the reference sample but exhibited a lower cumulative heat at the age of 7 days.• Some meaningful relationships between early structural build-up and pore solution chemistry have been observed.The main increase in the structural build-up is delayed by the addition of chemical retarders.During the period of the main increase in the structural build-up, Ca, Al and Si decrease rapidly and interact together to form reaction products.• The MIP results and BSE images showed that the addition of SP could lead to a denser microstructure with lower porosity, while a porous microstructure was formed by the incorporation of CA.The XRD and EDX results confirmed that the main reaction product of AASC with the addition of chemical retarders was C-A-S-H.

Fig. 4 .
Fig. 4. The evolution of heat flow and cumulative heat release of AASC pastes (a), (c) and (e) are the heat evolution for CA, SB and SP addition, respectively.(b), (d) and (f) are the cumulative heat release for CA, SB and SP addition.

X
.Dai et al.

Fig. 6 ,
Fig.6, all activator solutions showed a main band with a peak at around 1010 cm − 1 , which was related to the overlapped Si-O-Si bands of different Q n species in the activator solution.The peak position for REF_Ms1.2 is located at 1006 cm − 1 , while the peak shifted to 1016 cm − 1 , 1008 cm − 1 and 1008 cm − 1 for CA4, SB4 and SP4 activator, respectively.

Fig. 5 .Fig. 6 .
Fig. 5.The evolution of heat flow (a) and cumulative heat release (b) of AASC pastes in the first three hours.

X
.Dai et al.

Fig. 10 .
Fig. 10.Pore size distributions determined by MIP at the age of 28 days (a) CA, (b) SB and (c) SP.

X
.Dai et al.

Table 1
Chemical composition of the GGBFS used in the study.

Table 2
Mixture proportions of the paste samples.
X.Dai et al.

Table 3
Setting times of AASC pastes.
X.Dai et al.

Table 4
Mechanical performance of AASC pastes.
a Standard errors.X.Dai et al.

Table 6
Average atomic ratios determined by EDX analyses.