On the optimization of BOF slag hydration kinetics

In this study, the early age hydration kinetics of BOF slag is investigated. Isothermal calorimetry and in-situ X-ray diffraction (XRD) are used to examine the effects of K3-citrate concentration and nanosilica addition on slag reactivity. The results reveal that brownmillerite is the most reactive slag phase during the first 24 h of hydration. The kinetics of brownmillerite hydration is controlled by K3-citrate dosage, and the reaction is strongly accelerated with highly concentrated solutions (0.6 M). A similar phenomenon is observed for amorphous C2S at the early hydration stages. Among hydration products, crystalline siliceous hydrogarnet is identified. Precipitation of hydrogarnet (as observed with the XRD technique) coincidences well with the occurrence of a shoulder in the heat flow curves after the heat flow maximum due to brownmillerite dissolution. It is shown that nanosilica slightly delays the hydration of brownmillerite while enhancing the precipitation of C-S-H gel. In consequence, the early age mechanical properties of BOF slag mortars with nanosilica addition are improved.


Introduction
Annually, around 100-130 million tonnes of Basic Oxygen Furnace BOF slag (a by-product from steel manufacture) are generated worldwide (80-120 kg per tonne of steel) [1][2][3]. Therefore, the application of BOF slag in the concrete industry is of high interest [4]. Recent studies show that the BOF slag reveals a partial hydraulic activity and can be applied as a cementitious binder either when it is sufficiently ground or chemically activated [5][6][7]. BOF slag is composed of C 2 S, wuestite, brownmillerite and magnetite, and minor phases including C 3 S, free lime, portlandite and calcite [1,5,[8][9][10]. The hydraulic properties of BOF slags mainly depend on the contents of C 2 S and C 2 (A,F) and the reactivity of these phases. Even though C 2 S constitutes about 50 wt% and brownmillerite about 20 wt% of slag, the reactivity of BOF slag tends to be low due to late hydraulic activity of C 2 S and most likely due to its contamination with Fe 3+ , P 5+ , as well as the entrapment of reactive phases within the inert slag particles [5,7,11].
In cement systems, the reaction of brownmillerite takes place simultaneously with the reaction of C 3 A at the very early hydration stages, while the reaction of C 2 S contributes to the late strength development. Due to the late hydraulic activity of C 2 S, early age mechanical properties of BOF slag-based materials might be insufficient for practical applications. In the work of Kaja and Yu [5], it was shown that the chemical activation of BOF slag with potassium citrate enables the increase of the overall hydration degree of BOF slag, which, together with the superplasticizing effect of potassium citrate, lead to the superior mechanical properties of BOF slag pastes, up to 75 MPa, after 28 days of hydration [5]. However, the impact of the early age phase development on mechanical properties has not yet been investigated.
A commonly applied strategy to increase the early-age strength development of hydrating cement and its composites with slag or pozzolanic materials is the addition of nanomaterials, particularly nanosilica [12,13]. Due to the amorphous nature and high specific surface area, nano-silica (nS) features a very high pozzolanic activity [14][15][16]. The accelerating function of nS is attributed to the provision of additional nucleation sites, and, as revealed more recently [17], to the rapid depletion of calcium ions from the pore solution. Nano-silica contributes not only to early strength development but also causes structural densification [18][19][20][21]. The structural refinement in the presence of nano-silica and the consequent durability improvement can be crucial for BOF slag-based systems, which are often contaminated with heavy metals (mainly vanadium and chromium) [1,5,22]. In the BOF slag system, nano-silica can potentially strengthen the structures of C-S-H gel and siliceous hydrogarnet-the main hydration products, which presumably host heavy metals [5,23].
Consequently, in this study, we aim to investigate the combined effects of nanosilica addition and tri-potassium citrate concentration on the early age hydration of BOF slag. The reaction kinetics and early phase assemblage, analysed with in-situ and ex-situ XRD measurements, isothermal calorimetry and thermogravimetric analyses, are related to the mechanical properties of BOF slag mortars. This strategy is used to optimize the strength development of slag-based materials and, at the same time, to increase their durability.

Materials
A representative batch of BOF slag from the steel manufacture was collected and provided by Tata Steel (The Netherlands) and used in this study to perform the experiments. The slag (2-5.6 mm grains) was firstly ground with a vibratory disc mill (RS 300, Retsch). The particle size distribution of the milled slag, measured with the laser diffraction spectroscopy method (Mastersizer 2000, Malvern), is presented in Fig. 1.
The BOF slag was chemically and mineralogically analysed with the X-ray fluorescence (XRF) and the quantitative X-ray diffractometry (XRD), respectively. The results are presented in Table 1. The XRDamorphous content in the BOF slag accounts mostly for the C 2 S phase, as shown in our previous study [5].
Nanosilica, with 99.7% purity, was obtained through the dissolution of olivine (as described in detail in Refs. [24,25]). The specific surface area (measured with the BET method) and the specific density of the synthesized nano-silica were about 400 m 2 /g and 2.0 g/cm 3 , respectively. Tri-potassium citrate monohydrate (K 3 C 6 H 5 O 7 ⋅H 2 O, GPR REC-TAPUR®, purity > 99%) was employed with a dual role of activator and superplasticizer [5]. The CEN-standard sand was used for the preparation of mortars.

Mix design and sample preparation
The BOF slag pastes with and without the addition of nanosilica were prepared according to the mix design shown in Table 2. Nanosilica was applied by substituting 2 vol % of the slag. This relatively low dosage of nS was selected considering the high specific surface area (400 m 2 /g) of nS and the content of the potentially reactive phases in BOF slag [26]. The concentrations of tri-potassium citrate were chosen based on our previous study [5].
For the preparation of pastes, nanosilica was firstly dispersed in 2/3 of water using the Hielscher UP400S ultrasonic device (15 min, with an amplitude of 75%). The temperature of the solutions was controlled with the ice water bath at the level of 21 ± 0.2 • C before application. The tripotassium citrate monohydrate was dissolved in the remaining water. Both solutions were mixed with slag for 3 min using a high-speed mixer. Pastes were subsequently poured in the polyethylene vials and sealed cured (20 • C, RH>95%) until the designed testing ages.
For the compressive strength evaluation, mortar prisms were prepared. To ensure sufficient workability of mortars (S4, according to EN 206-1), an additional amount of water was added, resulting in a water/ binder mass ratio of 0.25. The sand content was adjusted to obtain a similar paste to sand volumetric ratio as for standard cement mortars . In consequence, a sand to slag mass ratio of 2 was used. Mortars were prepared using a Hobart mixer following the procedure described in EN 196-1 but with extended mixing time (a total mixing time of 10 min was employed). Due to the high density of slag and the limited/late reactivity of the slag phases, bleeding problems were encountered during the samples' preparation. The impact of this phenomenon on the final results should be considered and requires further investigation as some signs of bleeding were also observed on the paste level.

Isothermal calorimetry
The heat flow and cumulative heat release from the reacting BOF slag were monitored with an isothermal conduction calorimeter (TAM Air, Thermometric). The external mixing method was applied. Firstly, the tri-potassium citrate solution was prepared, and nanosilica was dispersed in water with an ultrasonic device. Then, both solutions were mixed with BOF slag for 2 min. Pastes were subsequently transferred to the glass ampoules and loaded into the calorimeter. The measurement temperature was set to 20 • C. For the evaluation of the total heat release, the integration of the heat flow curve was made from 45 min after the sample insertion, when the measurement conditions were stabilized.

XRD
Changes in the phase composition of the hydrating BOF slag pastes were monitored with the in-situ XRD method within the first 24 h of hydration. An X'Pert Pro PANalytical diffractometer in theta-theta geometry was used. For the preparation of pastes, the mixing procedure described in Section 2.2 was employed. Pastes were subsequently placed in the sample holders, flattened with a spatula, and covered with a   29.0 GOI 1000 0.14 Table 2 The compositions of the mixtures for the preparation of the paste. Kapton foil to minimize water evaporation and CO 2 ingress. The measurements were carried out with the settings listed in Table 3. Each diffraction pattern was recorded for about 15 min. At later hydration stages, the hydration of BOF slag pastes was arrested with the solvent exchange method (isopropanol, diethyl ether [27]). The XRD measurements were performed on the powder samples using a Bruker D4 ENDEAVOUR X-ray Diffractometer.

Thermogravimetric analysis
After hydration stoppage, the powders (around 50 mg) were thermally treated using a Jupiter STA 449 F1 Netzsch instrument, with the heating rate of 5 • C/min, in the temperature range between 40 and 1000 • C. A nitrogen atmosphere was employed.

Compressive strength
The compressive strength was evaluated on the mortar prisms (4 × 4 × 16 cm 3 ). Samples were demoulded after 24 h, and then sealed with plastic film, and cured in the controlled conditions of 20 • C, RH>95% until further testing. The strength measurements were performed following the EN 196-1 standard after 3, 7, and 28 days of curing, respectively.

XRD interpretation
Quantitative phase analysis was performed utilizing Rietveld refinement with the use of the TOPAS Academic software v5.0. The external standard method (corundum) was employed. This method enables the quantification of phases independently, which is especially important when hydration products reveal low crystallinity. The mass absorption coefficients of pastes were calculated based on the International Tables of Crystallography [28]. The crystal structures used for the Rietveld refinement are catalogued in Table 4, while described in more detail in our previous work [29]. The lattice parameters, crystallite size, and microstrain of the slag phases were determined on anhydrous slag and then used unrefined for the phase quantification in hydrating slag. The broad XRD hump caused by the Kapton film, free water, and amorphous content was modeled by introducing individual peaks. To account for the changes in the phase contents, the peak areas were refined, whereas the positions of the peaks were kept fixed. With the in-situ XRD method, only the pastes with 0.2 M K 3 -citrate were investigated. Due to the high fluidity of the pastes with 0.6 M K 3 -citrate and strongly alkaline pH of the pore solution [30], slight bleeding and deposition of carbonates (caused by partial destruction of the Kapton foil) were observed on the surface of the samples. It has to be noted that these factors might affect the quantitative interpretation of the results.

Heat flow calculation
To validate the findings regarding the early age reactions taking place in BOF slag pastes obtained with calorimetry and in-situ XRD, the measured heat flow curves from the isothermal calorimeter were compared with the heat flows calculated from the in-situ XRD experiments [31]. This method enables verification of the contribution of the dissolution/precipitation processes assigned to the particular phase in BOF slag to the overall heat release during the investigated period of the hydration. Based on the previous studies [32,33], the following equation was used to describe the reaction of brownmillerite, knowing that the composition of brownmillerite in BOF slag is slightly closer to C 2 F than to C 4 AF [8]: The precipitation of iron (III) hydroxide was assumed to take place during the first hours of hydration, as indicated in Refs. [34,35], and later confirmed by Dilnesa et al. [36] for the C 4 AF phase. During the hydration of C 2 F, similar behaviour is expected. The formation of iron (III) hydroxides was previously observed upon hydration of C 2 F phase [37].
The enthalpy of the C 2 F reaction was determined using the thermodynamic data listed in Table 5. In the next step, the heat flow was calculated according to equation (2) [32]. For the calculations, the phase contents derived from the XRD measurements were smoothed with a Fourier filter [31]. Where:

Hydration kinetics
The hydration kinetics of BOF slag is strongly affected by the dosage of tri-potassium citrate, as already shown in more detail in our previous studies [5,29]. From Fig. 2, it can be seen that the change of K 3 -citrate concentration from 0.2 M to 0.6 M results in the elimination of the induction period, which lasts about 4 h for paste with 0.2 M K 3 -citrate. The shape of the "main peak" of the heat flow curve, associated with the reaction of BOF slag phases, is broader for a higher activator dosage, and   A.M. Kaja et al. the total amount of the heat released during the first 24 h is slightly higher than for paste with 0.2 M citrate. With the addition of 2% of nanosilica, for both K 3 -citrate concentrations, the hydration of BOF slag is delayed. The retarding effect of nanosilica observed in our systems contradicts the accelerating function of nS in Portland cement systems [40]. During the hydration of Portland cement, nanosilica accelerates the reaction by offering additional nucleation sites and/or boosting the depletion of calcium ions from the pore solution. However, these effects seem to be outbalanced by other phenomena when BOF slag is activated by K 3 -citrate. One of the postulates is that potassium citrate interacts with nanosilica, and therefore, its availability to enhance the dissolution of BOF slag phases is lowered. Since at the early hydration stages, the pH of the pore solution of BOF slag paste is strongly affected by the K 3 -citrate dosage (ranging from ~12.8 for 0.1 M to 13.4 for 0.6 M during the first 30 min of hydration [30]), the retardation effect can also be caused by a slight decrease of pH in the presence of nanosilica.

Phase assemblage
The changes in the phase composition of BOF slag pastes upon hydration are presented in Table 5. The quantitative XRD analyses reveal that at the early hydration stages, brownmillerite is the most reactive phase in BOF slag. The reaction of brownmillerite occurs mainly within the first day of hydration regardless of the concentration of potassium citrate. A greater extent of brownmillerite reaction is observed for the pastes with 0.6 M K 3 -citrate than pastes with 0.2 M K 3 -citrate. The addition of nanosilica results in a slightly delayed dissolution of this phase.
Next to brownmillerite, partial dissolution of wuestite is observed. The reactivity of wuestite, examined with the XRD method, might be overestimated due to the sample preparation procedure, as discussed in detail in Ref. [29]. In all pastes, the magnetite is relatively inert during the investigated hydration period. The crystalline C 2 S reacts to a limited extent within 28 days of hydration, showing a slightly higher reaction degree when a lower dosage of K 3 -citrate is applied. The main crystalline product of BOF slag hydration is siliceous hydrogarnet (Hg) [5]. Additionally, the formation of pyroaurite (Mg 6 Fe 3+ 2 (OH) 16 [CO 3 ]⋅4H 2 O) was detected with a broad reflection at 2θ = 13.5 • , in agreement with our previous study [5]. Due to the very low contents and poorly crystalline structure of this phase, it is not included in Table 6.

Hydration of brownmillerite
Since brownmillerite is the most reactive crystalline phase in the examined systems and its reaction occurs mostly during the first 24 h of hydration, the reaction kinetics of this phase becomes crucial for the setting time of pastes and their early age mechanical properties. In-situ XRD experiments were carried out to reveal the impact of nanosilica addition on the reaction kinetics of brownmillerite. Fig. 3 shows the phase evolution in hydrating BOF slag paste with 0.2 M potassium citrate with and without the addition of nanosilica. It can be seen that brownmillerite shows the highest reactivity among other phases and that the precipitation of hydrogarnets follows its hydration. A slight delay is observed in the system with nanosilica. Similar behaviour is expected for BOF slag paste with 0.6 M K 3 -citrate.  Calorimetry data indicate that the dissolution of brownmillerite in that system takes place immediately after contact with water. A comparison between the calorimetry and in-situ XRD results (Fig. 4) shows that the shoulder on the heat flow curve on the right side of the main reaction peak (at around 13 h for paste without nS and 15 h for paste with nS) coincidences well with the precipitation of hydrogarnets. The exact composition of hydrogarnets (Ca 3 (Al x Fe 1− x ) 2 (-SiO 4 ) y (OH) 4(3− y) ) is difficult to establish as it can form solid solutions [23]. The gradual uptake of SiO 4 − in the structure of hydrogarnets is commonly observed [41]. Therefore, simplifications are required for the quantification of this phase, as described in Ref. [5]. The presented data show clear evidence for the formation of hydrogarnets in the investigated BOF slag systems. However, further studies are required to better understand the exact nature of this hydration product. Interestingly, the precipitation of crystalline hydrogarnets was not observed in the previous studies on the early hydration of synthesized brownmillerite [32,33]. Even though it was shown that the majority of brownmillerite underwent dissolution during the initial 24 h of hydration (either with or without the presence of gypsum), the formation of silicon-free hydrogarnets was not observed. On the other hand, for Portland cement paste, the precipitation of hydrogarnets was detected after 16 h of hydration [36]. This behavior indicates that the presence of silicate ions and/or pH of the pore solution are critical factors for the precipitation of crystalline hydrogarnets (at the early hydration stages). For matured Portland cement systems, siliceous hydrogarnets are considered the most stable iron-containing hydrates, being the reaction products of brownmillerite and C 2 S/C 3 S [42,43].

Hydration of silicate
Since part of C 2 S in BOF slag is XRD-amorphous [5] and its potential reaction is not detectable with the XRD method, the heat flow curves from calorimetric measurements were compared with the curves calculated from in-situ XRD data. The brownmillerite was considered the only hydrating crystalline phase during the first 24 h of hydration (see Fig. 3). The slight dissolution of other crystalline phases (as shown in Table 6) was not taken into account, being assigned mostly to the initial dissolution (after contact with water), which is not analysed in this study. Fig. 5 reveals good agreement between the measured and calculated heat flow curves for pastes with 0.2 M of potassium citrate. The satisfactory correspondence regarding the positions (time) and shapes of the curves confirms that the evolved heat is mainly due to the reaction of brownmillerite. The difference between the measured and calculated heat can be explained by the presence of aluminum in brownmillerite (which is not considered in the proposed reaction) but also likely originates from the dissolution of amorphous C 2 S and precipitation of hydrogarnet. It is generally accepted that α ′ -C 2 S polymorph, which coexists in BOF slag with β-C 2 S (commonly observed in Portland cement), reveals a higher reactivity than β-polymorph. Knowing that the potassium citrate is boosting the pH of the pore solution [30], the early age dissolution of this phase is expected to occur, especially in samples with 0.6 M K 3 -citrate [5,44].
To further support this hypothesis, thermogravimetric measurements were conducted. The mass loss due to the dehydration of C-S-H gel after selected hydration times was used to indicate the differences in the extent of C 2 S reaction in the investigated pastes. Similarly, the precipitation of hydrogarnets, resulting from the reaction of   brownmillerite and C 2 S, was followed. An exemplary DTG curve is shown in Fig. 6 a).
For the hydrated BOF slag, the mass loss occurs due to the decomposition of C-S-H gel, hydrogarnets, pyroaurite, portlandite, and carbonates/citrate compounds. It is widely reported that the dehydration of C-S-H gel in cement paste takes place within a broad temperature range (up to 400/500 • C), with the main water release at around 100 • C [45,46]. For the gel products of alkali activation, with aluminum and alkali metals in the structure, the water loss is observed within a similar temperature range [47]. At the same time, depending on the chemical composition, the decomposition of hydrogarnets occurs between 200 and 400 • C [23]. Pyroaurite decomposes between 180 and 410 • C [48] and portlandite between 400 and 500 • C [27]. Overlapping of the temperature ranges in which the water is released from different hydration products precludes the quantitative analysis of the results. Nevertheless, based on the literature and the shape of DTG curves observed for hydrated BOF slag, the temperature intervals assigned to the water loss from different hydration products were selected to demonstrate the differences in the hydration kinetics of slag phases. As presented in Fig. 6 a), it was assumed that mass loss between 40 and 180 • C originates mostly from the dehydration of C-S-H gel, whereas between 180 and 410 • C from the dehydration of hydrogarnet and, to a smaller extent, pyroaurite. Additionally, the total mass loss due to the water release was analysed in the range between 40 and 500 • C, as recommended in Ref. [27]. It has to be noted, however, that the presented temperature ranges cannot be exclusively assigned to the decomposition of a particular hydration product (e.g., decomposition of C-S-H gel takes place up to 500 • C) and therefore have only indicative character.
From Fig. 6 b), it can be seen that the mass loss between 40 and 180 • C is generally higher for BOF slag pastes with 0.6 M of K 3 -citrate during the first 7 days of hydration, indicating a greater extent of C 2 S reaction in these samples. This behaviour can be directly related to the higher pH of the pore solution during the early hydration of BOF slag in pastes with 0.6 M of K 3 -citrate in comparison to the pastes with 0.2 M of K 3 -citrate [30]. Furthermore, the addition of nanosilica results in the enhanced precipitation of C-S-H gel. This observation strongly suggests that nanosilica is consumed in the pozzolanic reaction, resulting in lower portlandite contents (as shown in Table 6), rather than being incorporated in the structure of hydrogarnet.
However, the accelerating effects of pH and nS addition are less pronounced after 28 days of hydration, which is especially evident for the pastes with 0.2 M of K 3 -citrate, in which a higher mass loss is observed for the sample without nanosilica. Similarly, the differences in the mass loss between the pastes with the two analysed dosages of potassium citrate are minimized or eliminated at this age. The abovementioned observations bring us to the conclusion that the enhanced early age hydration of C 2 S adversely affectss the late reaction of this phase.
The mass losses in the temperature range between 180 and 410 • C, mostly due to dehydration of hydrogarnets, correlate well with the findings from the XRD analyses. The mass loss is substantially higher for the pastes with 0.6 M of K 3 -citrate than for pastes with 0.2 M of K 3citrate. For both activator dosages, the delayed reaction of brownmillerite and consequent precipitation of hydrogarnets are also reflected in the presence of nanosilica.
At the same time, the total mass loss due to the water release from the hydration products shows the combination of the two opposite effects of nanosilica, which enhances the precipitation of C-S-H gel while having a negative impact on the hydration of brownmillerite at the early hydration ages. In consequence, no substantial differences in the bound water content between the reference samples and the samples containing nanosilica are observed. Nonetheless, it is worth mentioning that after 28 days of hydration, the samples without nanosilica addition tend to contain more chemically bound water.

Mechanical performance
In our previous study, we have shown that the BOF slag paste can reach a strength of up to 75 MPa after 28 days of hydration (when 0.2 M potassium citrate is applied) [29]. However, it has been often reported that for alkali/chemically activated binders, the transition from paste to mortar causes a significant decline in performance due to higher water demand [49]. In this study, the mechanical performance was investigated on the mortar samples, and therefore, additional water had to be added to the mixtures to provide sufficient flowability. In consequence, the potassium citrate solutions were diluted, and the kinetics of BOF slag hydration was affected. Nevertheless, the observed changes in the mechanical performance provide significant insight into the effects of early slag hydration kinetics on the later phase development, especially the hydration of C 2 S (the most abundant phase in BOF slag). Fig. 7 shows the compressive strength results for the BOF slag mortars. The mortars with 0.6 M potassium citrate reveal accelerated strength development in comparison to the mortars with 0.2 M K 3 -citrate. After 3 and 7 days of curing, the difference in the strength of 64% and 35% is observed, respectively. After 28 days, however, higher compressive strength is reported for mortars with a lower concentration of potassium citrate. The addition of nanosilica improves the compressive strength of mortars with 0.6 M K 3 -citrate after all the investigated curing periods. For mortars with 0.2 M K 3 -citrate, strength increase is observed after 3 and 7 days, while after 28 days, mortars with nanosilica show a decline in strength compared to the reference samples.
The XRD and thermogravimetric analyses disclose that nanosilica delays the hydration of brownmillerite and enhances the precipitation of C-S-H gel. Furthermore, it is revealed that in the pastes with 0.6 M of K 3citrate, the early hydration of amorphous C 2 S is promoted. The improved early-age performance of BOF slag mortars with the higher activator dosage and with nS addition at the early hydration stages indicates that the C-S-H gel is a principal strength-giving phase, whereas hydrogarnets have a lesser impact on the mechanical properties of BOF slag mortars. Furthermore, the increased early age reactivity of C 2 S seems to have a negative influence on its overall hydration degree, and therefore, after 28 days of hydration, the highest mechanical performance is observed for the mortar without nanosilica and with a low concentration of K 3 -citrate.

Conclusions
This work focuses on the early age hydration of BOF slag. Special attention is paid to the impact of nanosilica and K 3 -citrate concentration on the hydration kinetics of the slag phases. The compositional changes in the slag pastes are related to the mechanical properties of mortars. The following conclusions are reached.
• The early hydration of BOF slag activated with K 3 -citrate is dominated by the reaction of brownmillerite and, to a lesser extent, of amorphous C 2 S, leading primarily to the formation of hydrogarnets and C-S-H gel. • The kinetics of the brownmillerite reaction can be controlled by adjusting the dosage of potassium citrate (where hydration is accelerated about 4 h for samples with 0.6 M K 3 -citrate compared to the samples with 0.2 M K 3 -citrate). • Highly concentrated solutions of K 3 -citrate increase the early age reactivity of C 2 S. • Silica nanoparticles (at the dosage of 2 vol %) delay the reaction of brownmillerite while enhancing the formation of C-S-H gel. • Mortars with nS addition and highly concentrated potassium citrate solutions (0.6 M) exhibit improved early age mechanical properties. Increased compressive strength is associated with accelerated hydration of C 2 S / precipitation of C-S-H gel. However, up to 28 days, the strength development becomes insubstantial for these samples, likely due to the suppressed reaction of C 2 S.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.