Utilization of volcanic ejecta as a high ‐ performance supplementary cementitious material by gravity classification and

The reaction of natural pozzolans is caused by volcanic glass composed of amorphous silicate; however, volcanic ejecta also contains crystal mineral, pumice, and sometimes weathered clay fraction in their natural conditions. By focusing on the differences of physical properties between these components, high-purity volcanic glass powder (VGP) was manufactured by dry gravity classification and pulverization. This paper reports the results of investigations to utilize pyroclastic flow deposits as a supplementary cementitious material (SCM). 
Through this method, the glass content of VGP increased to 88% with a mean particle size of 1 μm, when that of the raw material is about 60%. Chemical analysis indicated that VGP is principally composed of silica (about 72%) and alumina (about 13%). 
The performance of VGP as a SCM was evaluated by conducting tests on concrete mixtures, replacing 0% to 30% by weight of portland cement by VGP with a 20% to 60% water to cement ratio. VGP concrete showed better results of 7-and 28-day compressive strength compared to control concrete in all experiments. In particular, VGP demonstrated better flowability and strength development in concrete with a low water-binder ratio in comparison to silica fume.


Introduction
Ancient concrete used in the Pantheon in Rome contains volcanic ejecta, which is known as pozzolana.One popular technique to reduce CO 2 emissions stemming from concrete production and construction involves the use of these natural pozzolans as SCMs [1].Volcanic activity is common in Japan.And one of the earliest researches to use volcanic ejecta as SCM was started more than 110 years ago in Hokkaido, the northern part of Japan [2,3].Subsequently, many studies were done on the applications of domestic natural pozzolan as SCM, and have been recently reviewed by Cai et al. [4].Although concrete made with volcano-related materials had been manufactured practically until 1960's, artificial materials have been chosen as a replacement.The reaction of natural pozzolans is due to volcanic siliceous glass [5], but volcanic ejecta generally comprises not only glass but also crystal mineral and clay mineral in their natural conditions [6], which may lead to typical undesirable properties of natural pozzolans; large variety, variability, and high water demand [7].
Pyroclastic flow deposits referred to as "shirasu" cover a wide area, forming extensive pyroclastic plateaus with a layer about 10 meters to 200 meters thick in the southern regions of Kyushu, Japan.The greatest amount of shirasu sediment, which is estimated to have a volume of 75 billion m 3 , is the Ito pyroclastic flow (A-Ito) that erupted out of the Aira caldera 29,000 years ago.This "Ito-shirasu" covers an area of about 3,427 km 2 , and it contains non-welded pyroclastic flow sediment, which includes crystalline mineral, amorphous silicate, and clay fraction (CF).It is said that to utilize Itoshirasu as industrial resource is very difficult for these impurities.
By the classification based on the origin of the SCMs [5], Itoshirasu is included in the group of unaltered pyroclastic material.There are a few practical trials to use Ito-shirasu as fine aggregate [8], and a house built with Ito-shirasu concrete won the 2017 ACI overall "Excellence" award [9].Despite their environmental advantage, significant barriers remain to widespread adoption of Ito-shirasu in place of natural sand and crushed sand.Major barrier is related to the typical properties of natural pozzolans; low fluidity due to high fineness and angular particles.Replacing 70% by volume of crushed sand with Ito-shirasu, both slump and slump flow values decreased in spite of increasing superplasticizer dosage and water as shown in Table 1.It was also reported that Ito-shirasu concrete causes higher shrinkage than control [19].In addition, low density (2.0~2.2 g/cm 3 ) and high water absorption (2.5~9.2%) for fine aggregate are other barriers by laws and standards (JIS).In their previous research [10][11][12][13] on the total utilization of pyroclastic flow deposits as construction materials, the authors clarified that dry gravity classification is effective in dividing Ito-shirasu into crystalline mineral, pumice, highpurity volcanic glass, and clay fraction.Through a separating machine made by combining a gravity separator with a winnowing sorter, the amorphous content of volcanic glass increased to 88% when that of raw material is about 60%.As the mechanism of the machine is shown in Fig. 1, winnowing with the action of the vibration fluid bed can sort particles by density, particle size, and terminal velocity with low energy consumption.
This paper presents the results of investigation on the suitability of using volcanic glass powder (VGP), manufactured by dry gravity classification and pulverization, as SCMs in concrete production.

Manufacturing process of VGP
Ito-shirasu used in this investigation was delivered from Kanoya-city, Kagoshima Prefecture in Japan by a mine operator without any pretreatment as shown in Fig. 2. The mineral composition of particle size fractions and the chemical composition of the raw material reported in the previous research [12] are presented in Fig. 3.As to mineral components, the fraction over 2.4 g/cm 3 is defined as crystalline, under 2.4 g/cm 3 defined as amorphous, and in particular under 1.5 g/cm 3 as pumice in this figure.After sieving Ito-shirasu to under 5 mm, Ito-shirasu is divided into five substances by dry gravity separator.Volcanic glass powder was manufactured through dry gravity classification and pulverization as shown in Fig. 4. First, classified volcanic glass is crushed by roller mill to a mean particle size of 5.1 μm (VGP 5).Then, VPG 5 is pulverized by a jet mill.Fine powder with a mean particle size of 1 μm (VGP 1) and coarse powder, 3 μm (VGP 3), are recovered from the bag filter and cyclone separator, respectively.

Properties of VGP
The basic properties of three types of VGP are shown in Table 2; the BET specific surface area measured by N 2 -adsorption, amorphous content based on heavy liquid separation using zinc bromide (ρ=2.4 g/cm 3 ), and chemical composition by fluorescent X-ray analysis.The SiO 2 content of VGP increased to 71.4% to 73.7% from 69.6% of that of the raw material.Besides, total alkali content of VGP is very high as well as raw material.But, previous research clarified that effective soluble amount for ASR is law when large amount of alkali is contained in SCMs [5], and that volcanic glass contained in Itoshirau; raw material, can prevent alkali-silica reactions [18].Whereas the average particle diameter of VGP 5 is the largest, the BET specific surface of VGP 3 is the smallest, presumably due to being classified by jet mill during recovery by cyclone.
As shown in Fig. 5, the particles of VGP 5 have an angular shape and the surface is almost smooth.Hackle marks created by pulverizing process are also observed.On the other hand, lamellar surface is observed on the surface of CF, which might be caused by the weathering of glass.
In regard to setting a density of 2.4 g/cm 3 as a threshold for the glass percentage, X-ray diffraction (XRD) was measured to ascertain the mineral phase and estimate the amount of amorphous materials.XRD was measured using corundum (α-Al 2 O 3 ) with a mean diameter of 3 m as an internal standard substance at an internal ratio of 20%.In Rietveld analysis, the microabsorption was corrected assuming the diameter of the target minerals to be 5 m.The target minerals were those contained in the raw material, such as quartz and albite.After selecting the target phase, parameters were refined, and the mineral phase was quantified.
Samples for measurements were prepared by centrifuging VGP 5 and the CF at 10,000 rpm, each into phases under and over a density of 2.4 g/cm 3 using a heavy liquid with a density of 2.4 g/cm 3 , filtering the separated phases with membrane filter, and washing repeatedly.Table 3 and Fig. 6 show the XRD patterns and results of mineral phase quantification, respectively.The amorphous content of VGP 5 with a density of 2.4 g/cm 3 or less is 98.7%, demonstrating that it is high purity volcanic glass.On the other hand, the amorphous content of VGP 5 with a density of more than 2.4 g/cm 3 is 73%, but crystalline minerals such as quartz are also contained.The amorphous content of the CF is also high, but lamellar minerals (clay minerals), such as antigorite, are contained.The density of individual volcanic particles may vary; but it was reported that the ones of pumices are generally between about 0.7 and 1.2 g/m 3 , glass shards commonly have densities of 2.35 to 2.45 g/m 3 , lithic fragments range from 2.7 to 3.2 g/m 3 , and crystals vary from about 2.6 to 5.2 g/m 3 [6].These results demonstrate that sorting out particles less than 2.4 g/cm 3 by dry gravity concentration and removal of smaller clay fraction by bag filter is technically effective.

Mix proportions and test method
The experimental program for three types of VGP was divided into three series, and tests were conducted on concrete mixtures.Table 4 lists the materials used in the experiments.Table 5 shows the factors, levels and mixing proportions.In the first series, to assess the suitability of using VGP to produce high performance concrete, the water-to-binder ratio was kept at 0.2 replacing 10% by mass of low heat Portland cement with VGP.The performance of concrete was evaluated by tests on its fresh and hardened properties compared with that of silica fume concrete.Slump-flow tests were carried out as per JIS A 1150, while the air content was determined by a pressure meter as per JIS A 1128.
Compressive strength tests were performed on 100×200-mm water-cured cylinders at 7, 28, 56, and 91 days as per JIS A 1108.Three specimens were used for each test at each age and the average values were reported.
Series II was performed to investigate the effect of VGP on the long-term strength.The water-to-binder ratio was kept at 0.

Series I: high performance concrete
Table 6 gives the fresh properties immediately after mixing and chemical admixture dosage.Fig. 7 shows changes in the slump flow and air content over time.Fig. 8 shows the time to the end of flow and time to 50 cm diameter.VGP 1 is found to provide a slump flow comparable to SF with a smaller chemical admixture dosage.The time to the end of flow and time to 50 cm diameter implied similar viscosity, with changes in fresh performance being also similar.On the other hand, VGP 3 and VGP 5 provided greater slump flow than SF with a lower superplasticizer dosage, but the viscosity of the resulting fresh concretes was too high for practical use as suggested by the time of flow stop and 50 cm flow.Fig. 9 shows the particle size distribution of the binders used measured by laser equipment.Fig. 10 shows the packing ratio of binders calculated from the particle size distribution using a particle packing simulation program [14].When compared with the binder solely consisting of low-heat Portland cement, the packing ratio of VGP 1 was high up to a replacement ratio of 10%, whereas those of VGP 3 and VGP 5 was almost the same level with control.Water retained between binder particles decreases as the packing ratio of the binder increases, improving the workability of the mixture such as UHPC by the micro filler effect.
These calculation results suggest that VGP 1 in place of 10% to 30% of LPC may produce such an effect when water to binder ratio is less than 0.2, but VGP 3 and VGP 5 will not achieve the effect even by adjusting the replacement ratio.Fig. 11 shows the results of interparticle distance (edge to edge) distribution of a mixture with a W/B of 0.2, in which 10% of LPC was replaced with VGP, using the same simulation program.Table 7 shows the results of average interparticle distance and average coordination number of particles.While the mean interparticle distance of VGP 1 was calculated to be 0.33 μm, those of VGP 3 and VGP 5 were 0.24 μm, being 72% of VGP 1.And the coordination number of VGP1 mixture was the least.The combination of binders with a longer time to the end of flow turned out to be the combination with a smaller interparticle distance and larger coordination number.
A short interparticle distance and a large number of particles therefore leaves a limited space for the rotational motion of binder particles to make the paste flow.It is therefore inferred that the interparticle distance and coordination number is a key factor for the fluidity improving effect (micro filler effect) in addition to the packing ratio.In comparison with SF with a mean diameter of 0.1 μm, the size of VGP is large.However, it can be said that VGP is a supplementary cementitious material that is expected to produce a sufficient effect of improving the workability of low W/B mixtures as demonstrated in the present results, provided that its physical properties including grading are rectified during the process of crushing and classification and that impurities including clay minerals are removed.
Fig. 12 shows the results of compression tests to an age of 13 weeks.The compressive strength of VGP 1 with a BET of 15m 2 /g, which was similar to SF, exceeded SF at 1 week and was equivalent to SF at 4 weeks and thereafter.The SiO 2 content of VGP 1 is 70%, being lower than SF's 95%, the content of SiO 2 + Al 2 O 3 + Fe 2 O 3 , which are active ingredients of natural pozzolans specified in ASTM C 618, is as high as 86%.This composition and the BET specific surface comparable to SF, as well as the filler effect due to the enhanced packing ratio, presumably led to the strength development of VGP 1 beginning form 1 week.
Fig. 13 shows the result of mercury intrusion porosimetry to ascertain the pore structure and porosity.Specimens were mortar mixture replacing 0% and 10% by mass of normal Portland cement with VGP1 with 0.3 water to binder ratio, and tests were performed at water curing ages of 7days and 28 days.VGP1 mortar showed slightly lower porosity compared with NPC mortar from 7 days.And, cumulated partial pore volume ratio of 3 nm-10 nm ranged diameter started increasing from 7 days, whereas that of 10 nm-300 μm raged diameter started decreasing.This is presumably caused by both the filler effect and pozzolanic reaction, and this denser structure may lead to greater compressive strength.The strength test results of VGP 3 and VGP 5 were lower than those of SF at all ages.The chemical compositions of the three types of VGPs are nearly the same, and the only difference of their physical properties is the fineness.VGP 3 and 5 with low BETs are presumably prone to slow reaction, while their low packing ratios may adversely affect the strength development.

Series II: Long-term strength of concrete
Fig. 14 shows the compression test results to 54 weeks.The compressive strength of VGP 1 exceeded that of control concrete from an age of 1 week, kept active development until 4 weeks, and retained moderate gains until 26 weeks.The strength scarcely increased from 26 to 54 weeks, suggesting that the ultimate strength is reached.The strength of VGP 3 was lower than that of control concrete up to 4 weeks but surpassed the control at 13 weeks, and retained marginal gains until 54 weeks.Its reaction was therefore considered to proceed over a long time.The strength of VGP 5 developed to a level comparable to the control at 13 weeks and kept slight increase thereafter similarly to NPC.Recent study showed that compared with 17 μm, using a smaller sized 6 μm volcanic ash result in a denser pore structure, and thus a greater compressive strength at curing age of 4 weeks [16].These results along with the results of series I demonstrate that, within the range of the present tests, on one hand the reaction of mixtures with a BET specific surface of around 10m 2 /g or less proceeds more slowly than that of SF but their pozzolanic reaction keeps proceeding over a long time after 4 weeks.On the other hand, when crushed to a mean diameter of 1 μm and BET of 15 m 2 /g, it turns into a supplementary cementitious material that contributes to strength beginning from 1 week.

Series III: Basic durability properties (VGP 1)
Fig. 15 shows the results of compression tests until 13 weeks.The strength increased as the VGP 1 replacement ratio increased.Fig. 16 and 17 show SEM image and EDX mapping of NPC specimens and VGP1 specimens just after compression test at 13 weeks, respectively.Table 8 shows results of Ca/Si ratio calculated by SEM-EDX analysis.
It was reviewed that the C-S-H phase in blended cements usually displays Ca/Si ratios in the range of lower than 1 to 1.8, i.e., lower than the observed C-S-H Ca/Si range in normal Portland cement [5].In the present tests, it is presumed that reduced Ca/Si ratio result in finer pore size distribution, lower permeabilities and chemically more resistant on conventional concrete mixtures.Fig. 18 shows the total chloride ion concentration profile in concrete.When the replacement ratio is 10%, the surface concentration is slightly higher than that with 0%.This is presumably because chloride ions scarcely permeate deeper and accumulated near the surface.With a replacement ratio of 20%, however, chloride ion penetration is inhibited even in the surface region.Here, the total chloride ions on the concrete surface and the apparent diffusion coefficient of chloride ions by immersion tests were simultaneously calculated by regression analysis of the values of the total chloride ions measured at each depth of each specimen using the solutions of diffusion equation based on Fick's second law.The apparent diffusion coefficients to three significant digits by rounding off the fourth digit shown in the figure are as low as 15% and 10% with replacement ratios of 10% and 20%, respectively, demonstrating VGP 1's excellent chloride penetration resistance.According to a study [15] dealing with 2-year immersion tests on concrete containing silica fume with a W/B of 0.35 to 0.50 and replacement ratios of 0, 4, and 8% conforming to the same JSCE specifications, the apparent diffusion coefficient decreases to 40% and 20% with replacement ratios of 4% and 8%, respectively, with small scatters depending on the W/B.Analysis by MIP reveals correlation between reductions in the pore volume 10 nm or more in diameter and diffusion coefficients, which are presumed to be due to pozzolanic reaction and the micro filler effect.
JSCE specifies that the period of water curing prior to immersion be 4 weeks.Therefore, the reaction of VGP 1 is considered to have sufficiently proceeded by the end of 4 week curing similarly to SF.The results of strength tests including series I and II are consistent with these results.
Fig. 19 and 20 show the results of accelerated carbonation tests.The square root of the age during acceleration is found to be linearly related to the carbonation depth.The equations of the approximated straight lines shown in the figure demonstrate that the carbonation rate coefficient of specimens with 10% replacement is slightly lower, and that of specimens with 20% replacement is 8% higher, than the control specimens, both being nearly of the same level.It is pointed out that pozzolanic materials in place of part of cement generally reduces the carbonation resistance of concrete due to the lower quantity of Portland cement and calcium hydroxide consumption by pozzolanic reaction [17], however the reduced permeability might render blended cements counteracts the loss of buffering Ca(OH) 2 [1].In the present tests, it is presumed that the effect of pore structure densification offsets the effect of CH consumption by pozzolanic reaction, when reaction begins at an early age.

Conclusions
The following were found in this study: (1) When utilizing volcanic ejecta for engineering purposes in the concrete field, the fraction with a density 2.4 g/m 3 or less is found effective from the aspect of the inclusion of clay minerals.High-purity volcanic glass can be separated from natural pyroclastic flow deposits using equipment referred to as a gravity separator.
(2) Separated volcanic grass contains 70% of SiO 2 , and the fraction with a density of 2.4 g/m 3 or less is 85% or more.When crushed to a mean diameter of 1 μm and BET of 15 m 2 /g, it turns into a supplementary cementitious material that contributes to strength beginning from 1 week in the W/B range of 0.2 to 0.6.
(3) With a low W/B of around 0.2, concrete containing volcanic glass powder with optimized physical properties demonstrates strength and mobility performance equal to or higher than concrete containing SF.With a high W/B of around 0.6, it improved basic durability performance of concrete.
In Japan, all silica fume is imported from abroad.Assuming recoverable reserves of Ito-shirasu (estimated to have a volume of 75 billion m 3 ) to be 20%, as recovery rate of volcanic glass from raw material is around 45%, we can manufacture 7.4 billion tons of VGP (bulk density 1.1 ton/m 3 ).In 2017, about 83 million m 3 of concrete is manufactured in Japan.Regarding average cement content as 300 kg/m 3 , 7.4 billion tons of VGP can be replaced 30% by mass of Portland cement for 100 years.Additionally, volcanic ejecta like Itoshirasu is found in various regions in Japan.If these volcanic ejecta turns from unutilized natural resource to highperformance SCMs, it will contribute to reduction of CO 2 emissions and durability of concrete in the future.

Figure 2 .
Figure 2. Distribution area of Ito-shirasu and sample site.

Figure 9 .
Figure 9. Particle size distribution of binders.

Figure 17 .
Figure 17.SEM picture of VGP1 specimens and EDX maps for Ca and Si at ×500 magnification.

Figure 19 .
Figure 19.Results of phenolphthalein tests at accelerated age of 26 weeks.

Table 1 .
Mixing proportions and fresh properties of concrete comparing Ito-shirasu with crushed sand as fine aggregate.

Table 3 .
Results of Rietveld Analysis.

Table 5 .
Factors and level, mixing proportions.

Table 6 .
Results of fresh concrete.
Figure 7. Slump flow and air content.

Table 7 .
Results of calculation.

Table 8 .
Results of fresh concrete.