Effect of Water Granulation Conditions on Density and Grain Size of Granulated Blast Furnace Slag

Abstract

The water granulation conditions for producing high density, coarse granulated blast furnace slag were investigated in a laboratory-scale experiment. The influence of slag temperature, water temperature and nozzle shape on the density and grain size of granulated slag was clarified. The influence of these factors on the density and grain size of the slag was confirmed by using various nozzles in a slag water granulation system.

Neural network computation was applied to estimation of the density and grain size of granulated blast furnace slag. The influence of water granulation conditions on the density and grain size predicted by neural network computation.

Based on the results of this research, we proposed a new slag granulation system and manufactured a high density, coarse blast furnace slag fine aggregate.

1. Introduction

In recent years, use of blast furnace slag as fine aggregate for concrete has increased in Japan due to decreased demand for use in cement and road base material, which had been the main applications. Water granulated blast furnace slag has a latent hydraulic property, and it has been confirmed that the compressive strength of concrete increases after 20 years when blast furnace fine aggregate is used in place of sand.¹⁾ Recently, research on use of blast furnace fine aggregate in high strength concrete has also been promoted,²⁾ and the “Recommendation for Practice of Concrete with Blast Furnace Fine Aggregate”³⁾ was revised.

Blast furnace fine aggregate is rarely used alone. It is normally used as fine aggregate by mixing with natural sand. However, because the particle size distribution of natural sand differs depending on the locality, the particle size distribution required for blast furnace slag fine aggregate also differs according to area. For instance, because fine-grained natural sand with a minute grain size is produced in Japan’s Kanto region, the particle size distribution is adjusted by mixing coarse-grained aggregate.⁴⁾

Water granulated slag for fine aggregate of concrete differs from water granulated slag for cement, in that a high density granulated slag with a unit weight 1.45 kg/l or more is demanded for use in concrete. Since many closed pores exist in water granulated slag, it is necessary to reduce the porosity of water granulated slag in order to achieve high density. According to Monna et al.⁵⁾ and Fuwa et al.,⁶⁾ pore formation is caused by a process in which steam is dissolved in the molten slag during water granulation, the steam reacts with the nitrogen in the slag, and this results in evolution of nitrogen gas and hydrogen gas. Reduction of the slag temperature is well-known as an effective technique for obtaining high density water granulated slag.^5,7)

On the other hand, few studies have examined the grain size of water granulated slag, particularly techniques for obtaining large grain size. It is difficult to satisfy both a coarse grain size and high density because the cooling rate decreases when the grain size of water granulated slag is increased, and foaming occurs easily.

Sano and Sato derived an equation for estimating the grain size of water granulated slag based on laboratory-scale experiments using a low melting point alloy and blast furnace slag.⁸⁾ However, their equation did not consider the influence of the molten slag temperature and blowing water temperature.

The present study was carried out to investigate influence of the molten slag temperature, blowing water temperature, and nozzle shape on the density and grain size of water granulated slag in laboratory-scale water granulation experiments. As a result, the influence of nozzle shape was found, and the influence of the nozzle shape on the density and grain size of water granulated slag was then confirmed by using various nozzles in an actual water granulation plant.

In an actual water granulated slag plant, the slag flow rate and molten slag temperature change constantly during the tapping period, and accompanying those changes, the water/slag ratio and blowing water temperature also change. The water granulation conditions that influence the quality of granulated slag include multiple variables, and those variables also have mutual effects. The neural network is an analytical method for examining the correlation between a target and various conditions when numerous conditions are involved. Prediction of the viscosity of the oxide melt,⁹⁾ the surface tension of the molten silicate,¹⁰⁾ and the sulfide capacity of the molten slag¹¹⁾ have been studied by using the neural network computation technique in order to clarify the interaction between physical properties and the factors that control those properties. Regression calculations of multivariate input-output data can be performed and the results of new conditions can be predicted with good accuracy by neural network calculations. Therefore, in this study, the density and grain size of water granulated slag were estimated by neural network calculations, and the influence of molten slag conditions and blowing water conditions on the density and grain size of water granulated slag were estimated.

Based on these results, we developed a new water granulation system for production of water granulated slag with the unit weight 1.40 kg/l or more and average grain diameter of 2 mm or more as a raw material for coarse, high density concrete aggregate. Although it is impossible to achieve these targets with conventional water granulation systems, these target values were realized with newly-constructed equipment developed on the basis of the present research. This paper presents a detailed description of the basic examination and development outlined above.

2. Experimental Method

2.1. Method of Investigating Quality of Water Granulated Slag

To investigate the relationship between the density and grain size of water granulated slag and the molten slag temperature, blowing water temperature, slag flow rate, and water/slag ratio, water granulated slag samples were taken from an actual water granulation system during the tapping period. Manufacturing data were also collected at the same time.

The granulated slag samples were collected from the water granulation system at No. 6 blast furnace at East Japan Works (Chiba District), JFE Steel Corporation. Figure 1 shows the outline of this water granulation system.¹²⁾ The slag was recovered with sampling equipment set up between the conveyors at the outlet side of the dehydrator. Samples weighing about 20 kg were collected at 20 min intervals. Data on the molten slag temperature, traffic volume of granulated slag on the belt conveyor, flow rate and temperature of the blowing water, and temperature of the agitation tank were collected simultaneously with sampling of the granulated slag. Before the molten slag was blown with the water stream, the surface temperature of the slag was measured with a radiation thermometer. The water/slag ratio was calculated from the flow rate of the blowing water and the measured traffic volume of the granulated slag on the belt conveyor.

Fig. 1.

Schematic diagram of slag water granulation system.¹²⁾

To evaluate the density of the water granulated slag, the unit weight was calculated in accordance with JIS A 1104, Method of test for bulk density of aggregates and solid content in aggregates. To calculate the average diameter of the water granulated slag, the particle size distribution was measured in accordance with JIS A 1102, Method for test of sieve analysis of aggregates.

2.2. Method of Water Granulation Experiment

In the water granulation system, the flow rate and temperature of the molten slag, which flows out of the tapping hole of the blast furnace, change with time from the start of tapping. Therefore, it is difficult to evaluate independently the influence of the water/slag ratio and blowing water temperature on the grain size of water granulated slag. To determine the effect of the molten slag temperature and blowing water temperature on the grain size and density of water granulated slag, a experimental water granulation apparatus was prepared, as shown in Fig. 2. The apparatus consists of a 50 kW induction furnace, a cooling water tank, a booster pump, a nozzle, a cold runner, and a slag recovery tank.

Fig. 2.

Schematic diagram of experimental apparatus.

First, 5 kg of granulated blast furnace slag was melted in a graphite crucible with the induction furnace. Because the N concentration in the molten slag has a large effect on pore generation in water granulated slag, nitrogen gas was used as the atmosphere gas during melting.

The slag was melted at 1673–1823 K and held for 30 minutes or more. The stopper was then removed, allowing the molten slag to flow from the hole (10 mmΦ) in the crucible bottom onto the cold runner. The molten slag flow was granulated by blowing water, which was supplied from a nozzle at one side of the cold runner. The water granulated slag flowed along the cold runner with the cooling water and fell in the collection tank. The flow time of the molten slag was about 1 min.

The blowing water was pressurized by the booster pump, and the pressure was adjusted in the range of 50–60 kPa. The blowing water temperature was adjusted to 293–343 K by blowing steam into the water tank. The water granulated slag was collected from the collection tank by pulling up the net. The unit weight and particle size distribution of the slag were measured after drying for 12 hours or more at 383 K.

The multi-hole nozzle and slit nozzle shown in Fig. 3 were used to examine the influence of nozzle shape. The multi-hole nozzles with the holes diameter of 2–5 mm and holes spacing of 5–12 mm were prepared.

Fig. 3.

Shapes of nozzles used in laboratory scale experiment.

2.3. Experimental Method in Water Granulation System

Based on the results with the experimental apparatus, an experiment was carried out in which nozzle shape in the granulation system were changed. Multi-hole type or slit type nozzle plate were installed in the water granulation system to investigate whether the unit weight and average diameter of water granulated slag can be controlled by changing the diameter, arrangement, and number of holes of the nozzle plate. The hole diameter D (mm) and center-to-center distance L (mm) were changed variously, as shown in Fig. 4. Moreover, the flow velocity of the blowing water was also adjusted by changing the nozzle hole opening area. Samples weighing about 20 kg were collected at 20 minute intervals.

Fig. 4.

Shapes of nozzles used in slag water granulation system.

3. Results

3.1. Relationship between Water Granulation Conditions and Unit Weight and Grain Size of Water Granulated Slag

Figure 5 shows the variation with time of the molten slag temperature, blowing water temperature, slag flow rate, and water/slag ratio (by weight), and the unit weight and average diameter of the granulated slag. Because the taphole diameter is small in the first stage of tapping and the hot metal collected in the bottom of the blast furnace is discharged preferentially from the taphole, the initial slag flow rate is small. With the passage of time, the taphole expands, the slag flow rate increases, and the water/slag ratio becomes small. Due to the influence of the slag flow rate, the slag temperature and blowing water temperature were low in the early stage and rose gradually over time. The unit weight of the granulated slag was large in the early stage and decreased gradually. On the other hand, the average diameter of the slag increased slightly in the latter half of tapping.

Fig. 5.

Change in conditions and qualities of granulated blast furnace slag in a tap.

Figure 6 shows the relationship between the unit weight of water granulated slag and the molten slag temperature before water granulation. When the temperature exceeded 1673 K, the unit weight decreased as the slag temperature increased. When the molten slag temperature was 1673 K or less, the unit weight of the water granulated slag was constant at about 1.4 kg/l.

Fig. 6.

Relationship between unit weight of water granulated slag and slag temperature.

As shown in Fig. 5, the change of grain size was similar to the change of blowing water temperature. Therefore, Figure 7 shows the relationship between the average diameter of water granulated slag and the blowing water temperature. The grain size becomes coarse as the blowing water temperature increases. Figure 8 shows the relationship between the average diameter and unit weight of granulated slag. Coarse grain are characterized by low density, while fine grain display high density. In other words, the conditions for coarse grains and the conditions for high density granulated slag are mutually contradictory.

Fig. 7.

Relationship between grain size of water granulated slag and cooling water temperature.

Fig. 8.

Relationship between grain size and unit weight of water granulated slag.

While it is possible to select either small, high density grains or large, low density grains, it is difficult to produce fine aggregate with large, high density grains with the current water granulation system.

3.2. Influence of Slag Temperature and Blowing Water Temperature on Density and Grain Size

Figure 9 shows the influence of the molten slag temperature on the unit weight and average grain size of the water granulated slag with the experimental water granulation apparatus. Although the absolute values differ depending on the nozzle shape and water temperature, the unit weight increases 0.1–0.2 kg/l and the average diameter increases 0.2–0.7 mm when the slag temperature decreases 100 K from 1773 K to 1673 K.

Fig. 9.

Effect of slag temperature on grain size and unit weight of water granulated slag.

Figure 10 shows the influence of the blowing water temperature on the unit weight and average grain size of the water granulated slag with the experimental apparatus. In these experiments, the molten slag temperature was constant at 1673 K. Grain size increases with increasing water temperature, and conversely, density decreases. When the blowing water temperature decreases 20 K, the unit weight increases 0.05–0.25 kg/l and the average diameter decreases 0–0.3 mm.

Fig. 10.

Effect of water temperature on grain size and unit weight of water granulated slag.

3.3. Influence of Nozzle Shape

In order to investigate the influence of nozzle shape, an experiment was carried out by using the multi-hole and slit nozzles with the experimental water granulation apparatus. The experimental conditions were constant at a water/slag ratio of 12, slag temperature of 1673 K, blowing water pressure of 60 kPa, blowing water temperature of 343 K, and blowing water flow rate of 60 l/min.

Figure 11 shows the effect of the nozzle shape on the unit weight and average grain size of the water granulated slag with the experimental apparatus. With the slit type nozzle, small, low density grains were produced. The largest average diameter was observed at D/L=0.5. However, the optimum value of D/L for unit weight was not clear.

Fig. 11.

Effect of nozzle shape on grain size and unit weight of water granulated slag.

The situation of collision between the molten slag and blowing water by visual and video observation was as follows. With the slit nozzle, when the molten slag collides with water, the slag forms grains and scatters strongly. A large number of scattered grains could be seen flying on the water stream without entering it. On the other hand, with the multi-hole nozzle, the slag grains flowed into the water in the cold runner. Because the diameter of the molten slag stream was about 3–4 mm, the slag invariably collides with water when a nozzle with holes spacing of up to 4 mm is used. However, when the holes spacing is more than 4 mm, the molten slag stream does not necessarily collide with the water. Therefore, it is deduced that there is an optimum value of D/L.

3.4. Results of Nozzle Change Experiments with Water Granulation System

Figure 12 shows the result of a comparison of the relationship between the average diameter and the weight with the slit nozzle and shower nozzle. Points closer to the upper right in this figure indicate that it is more possible to manufacture high density, coarse granulated slag. As in the results with the experimental apparatus, both the unit weight and the average diameter decreased with the slit nozzle compared with the multi-hole nozzle.

Fig. 12.

Effect of nozzle shape on quality of water granulated slag in water granulation system.

Figure 13 shows the results of an examination of the influence of the nozzle hole diameter D, nozzle hole spacing L, and flow velocity of the blowing water with the shower nozzle. Regarding the nozzle hole diameter D, higher density slag was obtained with the 20 mmΦ diameter than with 35 mmΦ. A low water flow velocity tended to produce coarse grains. The highest density and largest grain water granulated slag was obtained when the flow velocity of the blowing water was 11 m/s. However, it was not possible to find the conditions for obtaining an average grain diameter of 2 mm or more in combination with the unit weight of 1.4 kg/l.

Fig. 13.

Effect of multi-hole nozzle and blowing water velocity on quality of water granulated slag in water granulation system.

4. Discussion

4.1. Applicability of Grain Diameter Estimation Equation

The atomization method is used in metal powder production processes as a process for producing particles by the collision of two fluids. Equation (1), which was obtained experimentally by Lubanska as an estimation equation for grain diameter, is the most widely used equation in gas atomization.¹³⁾ The average diameter of the obtained metal powder is expressed by the physical properties of the molten metal and the spraying gas, and by the flow rate and the flow velocity of the spraying gas. Equation (1) can be expressed as Eq. (2) by using the Weber number We. Here, d[m] is a grain diameter, D_s[m] is a diameter of a molten metal stream, ν_m[m²/s] is a kinematic viscosity of a molten metal, ν_g[m²/s] is a kinematic viscosity of a spraying gas, δ_m[N/m] is a surface tension of a molten metal, ρ_m[kg/m³] is a density of a molten metal, W_m[kg/s] is a flow rate of a molten metal, and W_g[kg/s] is a flow rate of a spraying gas.   

$$d/D_s=K{\left[\frac{{\nu }_m}{{\nu }_g\text{W}e}\left(1+\frac{W_m}{W_g}\right)\right]}^{0.5}$$(1)

  

$$\text{W}e=\frac{{\rho }_m{V}^2{D}_s}{{\sigma }_m}$$(2)

To apply Lubanska’s equation to water atomization, Hiraga et al. modified the above-mentioned equation by considering the spraying water rate as energy.¹⁴⁾ Sekino et al. considered the kinetic energy of a spraying gas in a spraying process to be proportional to the surface energy of droplets, and modified the estimation equation for grain diameter based on the results of experiments with annular type nozzles.¹⁵⁾ However, these estimated equations are only valid for identical nozzle shapes.

From Figs. 9 and 10, it was understood that the grain diameter increases with decreases in the molten slag temperature or with increases in the blowing water temperature. Only Lubanska’s equation, which includes a term for kinematic viscosity, considers the influence of these two fluid temperatures. Figure 14 shows the results of calculations when the results in the water granulation experiments (data in Figs. 12 and 13) were arranged by using Lubanska’s equation. Here, the results of measurements by Koshida et al. were used for surface tension.¹⁶⁾ The correlation coefficient between observed values and calculated ones is 0.15. Lubanska’s equation is an experimental equation derived from the values of d/D in reports by multiple researchers, in which d/D differed by two orders of magnitude. Because the average diameter of water granulated slag is 1.2–2.0 mm and the range of d/D is narrow, at 0.02–0.05, the diameters estimated by Lubanska’s equation don’t agree with obserbed values.

Fig. 14.

Comparison between observed d/D and result of calculation by Lubanska’s equation.

4.2. Granulation Situation of Water Granulated Slag

From the results of the nozzle change experiments with the experimental apparatus and the actual water granulation system, the situation of water granulation is considered as shown in Fig. 15. First of all, the molten slag stream collides with the water stream, is broken, and forms grains. These particles take an angular shape due to pulling and tearing by the water stream, rather than a spherical shape due to surface tension, as in the case of air granulation. The particles come into contact with the water in the water stream and are cooled. Steam films form around the particles, which then are dissolved by the steam. In a hot thermocouple experiment, Sakaeda et al. confirmed that wetting between molten slag and water does not occur at high water temperatures.¹⁷⁾ When the film boiling condition occurs at the molten slag surface, N₂ gas and H₂ gas evolve as a result of reaction between the dissolving steam and the N in the molten slag.⁵⁾ These gases form bubbles, foaming occurs in the granulated slag, and as a result, the slag grain diameter increases.

Fig. 15.

Image of slag granulation and cooling by blowing water.

Banya et al. measured the solubility and solution rate of water vapor into CaO–SiO₂–Al₂O₃ system slag, which is close to the composition of blast furnace slag.¹⁸⁾ Water vapor solubility (%H₂O)_s [ppm] can be expressed by Eq. (3) and is given by the partial pressure of water vapor P_H2O [mmHg].   

$$\mathrm{K}\text{'}={(\%{\mathrm{H}}_2\mathrm{O})}_{\mathrm{s}}/{\mathrm{P}}_{\mathrm{H}2\mathrm{O}}{}^{1/2}$$(3)

In the state in which a steam film has formed around the granulated slag, it can be thought that P_H2O=1 atm (=760 mmHg). From the experimental result of Banya et al., substitution of the equilibrium constant K’=10^−1.09 of CaO/SiO₂=1, which is the nearest to the composition of blast furnace slag, in Eq. (3) leads to (%H₂O)_S=813 ppm.

On the other hand, Imai et al. measured the steam solubility of steelmaking slag, which has a high FeO content, and found that steam solubility was approximately 10 ppm at CaO/SiO₂=1.¹⁹⁾ Thus, blast furnace slag in the reduction condition has high water vapor solubility and foams readily.

The amount of dissolved water vapor in molten slag depends on the amount of steam generated around the slag particles and the contact time between the slag and the steam. Steam generation increases with increasing slag temperature or water temperature and with a decreasing water/slag ratio. When the water flow velocity or the water/slag ratio rises, the water peels off the steam surrounding the slag particles, and contact between the water and the slag surface occurs easily. On the other hand, because the molten slag is granulated by blowing off the slag with a water stream, the water/slag ratio and the water flow velocity, namely, the energy of the water stream against the molten slag, affect the grain diameter of the granulated slag. When the blowing water temperature or the slag temperature rises, steam is generated and the energy of the water stream decreases, and in this case, the granulated slag foams and becomes coarse-grained. Therefore, it is considered that the slag temperature, the blowing water temperature, the water slag ratio, and the blowing water flow velocity influence the density and grain size of water granulated slag.

4.3. Estimation of Granulated Slag Density by Neural Network Computation

A neural network computation is an operation technique for imitating the neurons which exist apporximetly 10 billion in the brain. When a signal exceeds a threshold, the signal is transmitted in a signal transfer between neurons. In a neural network computation, this is expressed by using the sigmoid function of Eq. (4).   

$$f(x)=\frac{1}{1+e^{-\eta \cdot x}}$$(4)

The exceeded critical values are outputted through a sigmoid function to the middle layer by inputting many signals. Next, the results that exceed a critical value are outputted through a sigmoid function again to the output layer by inputting signals of the middle layer. The cross-term can be considered by setting up the middle-layer.

An output value to the middle layer a_k is given by using the sigmoid function in Eq. (5).   

$$a_k=f\left(\sum _{\text{i}=\text{1}}^{\text{n}}{x}_i\cdot W_{\text{ki}}-W_{ko}\right)$$(5)

Where, x_i is an input value of unit i in the input layer, W_ki is a connection weight between unit i in the input layer and unit k in the middle layer, and W_ko is a critical value for unit i.

The final result y in the output layer is expressed by using the sigmoid function as follows.   

$$y=f\left(\sum _{\text{k}=\text{1}}^{\text{n}}{a}_k\cdot V_{\text{k}}-V_o\right)$$(6)

Where, a_k is an input value of unit k in the middle layer, V_k is a connection weight between unit k in the middle layer and the output layer, and V_o is a critical value for unit k.

As shown in Fig. 16, measurements are stored beforehand as a teacher signal t, and are compared with the output value y obtained from this calculation. The connection weights are calculated corresponding to differences between y and t, and then a new calculation is conducted by adding the connection weights. This calculation is repeated until the overall error value drops to the setting range. Using the function with the final connection weight, it is possible to estimate an output value from new input data.²⁰⁾

Fig. 16.

Structure of neural network computation.

Here, the unit weight of water granulated slag is estimated by using a neural network computation. The experimental result of 150 data in Figs. 12 and 13 are used in this calculation, except for 3 data, in which the water slag ratio > 80 t/t and slag temperature <1640 K, as these were considered extremely unusual conditions. The assumed input conditions are the water slag ratio, blowing water velocity (flow velocity at nozzle exit), blowing water temperature, molten slag temperature, and D/L of the nozzle. The output value is the unit weight. The middle unit used for the neural network computation is 5 units, and it is calculated until 0.05 kg/l or less with the average error 5% or less. Figure 17 shows the relationship between the observed unit weights and the calculated values. The iteration count converges at 4378 times in this calculation. The correlation coefficient between observed unit weights and calculated ones is 0.93. The calculated values agree well with the observed values. The parameters of this neural network computation are listed in Table 1.

Fig. 17.

Comparison between observed unit weight and result of calculation by neural network computation.

Table 1. List of parameter values for unit weight.

Number of Input Unit			5
Number of Middle Unit			5
Weight Matrix of INPUT to MIDDLE layer: Wki
Components	W1i	W2i	W3i	W4i	W5i
Water/slag	16.091949	3.87006	0.812757	4.705324	3.241892
Water velosity	5.412087	−3.63222	2.93967	1.126513	−2.933721
Water temperature	0.032177	−0.952206	−5.067539	−8.458968	−0.947308
Slag temperature	0.553824	0.239442	−0.843562	13.069121	3.005437
D/L	−5.181974	−1.692928	−2.064522	−0.011525	−15.371742
Threshold for MIDDLE layer: Wk0
	W10	W20	W30	W40	W50
	0.132283	0.626573	−2.623762	0.791736	1.929385
Weight vector of MIDDLE to OUTPUT layer: Vi
	V1	V2	V3	V4	V5
	6.022087	3.320196	1.787458	−6.059613	−2.310761
Threshold for OUTPUT layer: V0			0.288377

The influence on the granulated slag density of each condition of water granulated slag is presumed by using the values in Table 1. Table 2, which shows the average condition of water granulation, is assumed to be the standard condition. The unit weights are calculated by assuming four of five conditions using the values of Table 2, and only one condition is changed. The calculated unit weights are compared with the measured values in Fig. 18. Regarding the influence of slag temperature, unit weight increases 0.22 kg/l when the temperature decreases 50 K from 1723 K to 1673 K. The influence of slag temperature on increased unit weight by the neural network computation is larger than the result of the laboratory experiment.

Table 2. Base conditions to estimate effect of each condition on unit weight by neural network computation.

Water/slag (t/t)	Water velocity (m/s)	Water temperature (K)	Slag temperature (K)	D/L (m/m)
30	14	348	1723	0.571

Fig. 18.

Effect of water granulation conditions on unit weight by neural network computation.

The unit weight increases 0.1 kg/l, which is on the same order as in the lab experiment, when the blowing water temperature decreases 20 K from 343 K to 323 K. The influence of the blowing water velocity is small. As for D/L, unit weight displays its maximum value at D/L=0.45. Unit weight increases as the water/slag ratio increases.

4.4. Estimation of Granulated Slag Grain Diameter by Neural Network Computation

As with density, here, the grain diameter of water granulated slag will be estimated by a neural network computation. Using the results of the nozzle dimension change experiment with the water granulation system in Figs. 12 and 13, the assumed input conditions are the water slag ratio, blowing water velocity (flow velocity at nozzle exit), blowing water temperature, molten slag temperature, and D/L of the nozzle. The output value is the average diameter of water granulated slag, which is calculated until 0.087 mm or less with the average error of 5% or less. The middle units used for the neural network calculation are six units for the smallest iteration counts. Figure 19 shows the relationship between the observed average diameters and the calculated values. The iteration counts converge at 9002 times in this calculation. The correlation coefficient between observed average diameters and calculation values is 0.85. This neural network computation can estimate the grain diameter with higher accuracy than the expressions proposed by Lubanska.¹³⁾ The parameters of this neural network computation are listed in Table 3.

Fig. 19.

Comparison between observed diameter and result of calculation by neural network computation.

Table 3. List of parameter values for grain size.

Number of Input Unit			5
Number of Middle Unit			6
Weight Matrix of INPUT to MIDDLE layer: Wki
Components	W1i	W2i	W3i	W4i	W5i	W6i
Water/slag	−3.863173	8.310304	−7.123713	10.054397	4.095436	3.546344
Water velocity	−13.135866	1.077602	5.991301	−0.755393	0.791933	1.982512
Water temperature	8.394529	0.09753	2.81849	−6.498744	−2.196593	0.068588
Slag temperature	2.66004	−4.678781	4.055833	2.449656	−1.91708	−7.119845
D/L	6.702109	2.391121	3.591397	−2.977008	−1.06546	10.406311
Threshold for MIDDLE layer: Wk0
	W10	W20	W30	W40	W50	W60
	7.65614	2.410095	11.552577	3.82521	3.562869	2.831193
Weight vector of MIDDLE to OUTPUT layer: Vi
	V1	V2	V3	V4	V5	V6
	1.794398	2.520443	3.34451	−2.756549	5.453109	−3.57254
Threshold for OUTPUT layer: V0			0.133437

The influence of each condition on the grain diameter of water granulated slag is presumed by using the values in Table 3. Table 2 is assumed to be the standard condition. The average grain size is calculated by assuming four of five conditions using the values in Table 2, and only one condition is changed. The calculated average diameters are compared with the measured values in Fig. 20. The grain diameters are almost constant at the water/slag ratio of 30 or less, the blowing water temperature of 343 K or more, and the molten slag temperature of 1723 K or more. As with the laboratory experiments, the average diameter increases 0.1 mm as the blowing water temperature increases 20 K at 343 K or less. When the molten slag temperature increases 50 K from 1673 K to 1723 K, the average diameter increases 0.2 mm. This result shows the opposite tendency to that in the lab experiments. If D/L is set to 0.6 or less, the grain diameters become large. When the blowing water velocity is 12 m/s or less, the grain diameter increase with a decrease in blowing water velocity.

Fig. 20.

Effect of water granulation conditions on grain size by neural network computation.

4.5. Manufacture of Granulated Slag for Large-Grained, High Density Fine Aggregate

For use as an admixture in Japan’s Kanto district, the quality target of fine aggregate is assumed to be the average diameter of 1.5 mm or more and the unit weight of 1.5 kg/l or more. Water granulated slag has many edges. Thus, in order to maintain the workability of fresh concrete and the compressive strength of concrete, a grinding process is needed if this material is to be used as fine aggregate. However, after grinding the water granulated slag, its grain size becomes small and its density increases. Therefore, a corresponding increase in the grain size of the granulated slag before grinding is necessary. The quality targets for water granulated slag before grinding are the average diameter of 2.0 mm or more and the unit weight of 1.40 kg/l.

The specification of the equipment for manufacturing water granulated slag with high density and a coarse grain size was determined based on the results of the analysis by the neural network computations.

Figure 21 shows a schematic diagram of the developed equipment. A new blowing box and agitation tank were constructed, and these were connected to the existing granulation system. The granulated slag and water are returned from the new agitation tank to the existing cold runner, and the granulated slag is dewatered with a dehydration filter.

Fig. 21.

Schematic diagram of new slag granulation system.

Reducing the slag temperature to 1673 K or less is effective for adjusting the unit weight to 1.40 kg/l or more. Because the slag temperature is reduced to 1673 K or less in the new granulation system, the molten slag is granulated by blowing water after the slag is charged into the slag ladle. The nozzle plate ratio of D/L=0.44 is used to obtain high density granulated slag. To obtain coarse grains, the water slag ratio is set to 10, and the water temperature is adjusted to 343 K or more.

To secure the average diameter of 2.0 mm or more in water granulated slag, which could not be obtained by the water granulation plant for cement, we tried using a low blowing water velocity. Figure 22 shows the relationship between the average diameter of the water granulated slag and the blowing water velocity in the new equipment. The decrease in blowing water velocity is effective for producing coarse water granulated slag. The estimated value by the neural network calculation from Table 3 is shown by the solid line in Fig. 22. Because the estimation by the neural network calculations is limited to within the range of the measured data, the lower limit of the blowing water velocity is assumed to be 10.5 m/s. The estimated average diameter at this velocity is 1.8 mm. Granulated slag with the average diameter of 2 mm or more is obtained by additionally decreasing the blowing water velocity. The blowing water velocity for manufacturing water granulated slag with the target average diameter of 2 mm was determined to be 7 m/s. The unit weight with the blowing water velocity of 7 m/s is 1.43–1.63 kg/l, which achieved the target value.

Fig. 22.

Effect of blowing water velocity on grain size in new slag granulation system.

After grinding to round the edges, fine aggregate with the unit weight of 1.5–1.7 kg/l, average diameter of 1.5–1.7 mm (fineness modulus: 3.3–3.5), and water absorption of 0.2–0.5% could be obtained, thus satisfying the target quality.

Figure 23 shows cross-sectional photographs of a) water granulated slag manufactured with the equipment for cement and b) fine aggregate made from the developed granulated slag, which is manufactured with the new equipment followed by grinding. Low porosity water granulated slag can be manufactured with the new equipment. In addition, the corners and many pore parts in the water granulated slag can be broken preferentially by grinding. As a result, the developed fine aggregate shown in Fig. 23b) has few corners and low porosity.

Fig. 23.

Cross sectional photographs of water granulated slag. a) Water granulated slag for cement. b) Developed fine aggregate.

5. Conclusions

The conditions for manufacturing high density, coarse water granulated blast furnace slag for use as fine aggregate for concrete were examined. In water granulation systems, the water/slag ratio and blowing water temperature change accompanying changes in the slag flow rate and slag temperature, making it difficult control these values to specific conditions. The influences of the slag temperature and blowing water temperature on the unit weight and average diameter of water granulated slag were clarified based on laboratory-scale experiments and experiments with an actual water granulation system at an integrated steel works. It was also found that high density, coarse water granulated slag can be produced by adjusting the hole diameter and holes spacing of a multi-hole nozzle.

The unit weight and average diameter of water granulated slag could be estimated with good accuracy by neural network computations. It was also possible to understand the influence of individual conditions by neural network computations.

Based on these results, a manufacturing process for high density, coarse water granulated slag was proposed, and a new dedicated system for manufacturing fine aggregate was constructed. High density, coarse water granulated slag, with the unit weight of 1.40 kg/l or more and average diameter of 2.0 mm or more, can be obtained with this equipment by reducing the blowing velocity. The unit weight of 1.5–1.7 kg/l and average diameter of 1.5–1.7 mm, which are required in fine aggregate for concrete, can be obtained by grinding this water granulated slag.

As demonstrated by this research, neural network computation is an efficient method for optimizing the processing conditions and facility design for water granulated slag.

References

• 1)   K.  Kokubu and  Y.  Shimoyama: Cem., Concr., 714 (2006), 27.
• 2)   D.  Shi,  Y.  Masuda and  Y.  Lee: AIJ J. Technol. Des., 17 (2011), 787.
• 3)  Recommendation for Practice of Concrete with Blast Furnace Fine Aggregate, 2nd ed., Architectural Institute of Japan, Tokyo, (2013), 13.
• 4)   Y.  Shimada,  M.  Hattori,  K.  Kajiya and  K.  Kokubu: Cement Science Concrete Tecnology, 53 (1999), 501.
• 5)   I.  Monna,  A.  Okamoto,  S.  Suzuki,  Y.  Nagao,  H.  Tokumaru and  S.  Myozin: Tetsu-to-Hagané, 68 (1982), 81.
• 6)   T.  Fuwa,  S.  Banya,  Y.  Iguchi and  F.  Ishi: Tetsu-to-Hagané, 69 (1983), 371.
• 7)   A.  Ehrenberg: ZKG Int., 66 (2013), 64.
• 8)   K.  Sano and  H.  Sato: Tetsu-to-Hagané, 68 (1982), S747.
• 9)   M.  Hanao,  M.  Kawamoto,  T.  Tanaka and  M.  Nakamoto: ISIJ Int., 46 (2006), 346.
• 10)   M.  Nakamoto,  M.  Hanao,  T.  Tanaka,  M.  Kawamoto,  L.  Holappa and  M.  Hamalainen: ISIJ Int., 47 (2007), 1075.
• 11)   B.  Derin,  M.  Suzuki and  T.  Tanaka: ISIJ Int., 50 (2010), 1059.
• 12)   M.  Kimura,  M.  Saino,  E.  Akimoto,  J.  Furuya,  Y.  Ikeda and  T.  Harada: Kawasaki Steel Giho, 18 (1986), 230.
• 13)   H.  Lubanska: JOM, 22 (1970), 45.
• 14)   Y.  Hiraga,  Y.  Murakami,  T.  Ushiro,  Y.  Nakajima,  Y.  Yashima and  K.  Takahashi: Nisshin Seiko Giho, 67 (1993), 20.
• 15)   S.  Sekino,  M.  Masao and  T.  Nishida: Tetsu-to-Hagané, 67 (1981), 541.
• 16)   T.  Koshida,  T.  Ogasawara and  H.  Kishidaka: Tetsu-to-Hagané, 67 (1981), 1491.
• 17)   M.  Sakaeda,  H.  Ide,  K.  Morinaga and  T.  Yanagase: Tetsu-to-Hagané, 66 (1980), S115.
• 18)   S.  Banya,  Y.  Iguchi and  S.  Nagata: Tetsu-to-Hagané, 71 (1985), 55.
• 19)   M.  Imai,  H.  Ooi and  T.  Emi: Tetsu-to-Hagané, 48 (1962), 111.
• 20)   T.  Tanaka: Electric Furn Steel, 78 (2007), 29.