ISIJ International, Vol. 49 (2009), No. 6, pp. 843–850

1. Introduction

Mold fluxes which are based on the CaO–SiO2–Al2O3

system with additions of CaF2 and Na2O have many func-tions. The several essential functions include: (i) protectingthe meniscus of the steel from oxidation; (ii) providingthermal insulation to prevent the steel from freezing; (iii)providing liquid lubrication for the strand; (iv) controllingthe horizontal heat transfer between the strand and mold ac-cording to the steel grade being cast; and (v) absorbing in-clusions from the steel.1–5) When inclusions were picked up,such as Al2O3 and TiO2, it would result in significantchanges in the physical properties of mold fluxes, such asviscosity, break temperature, or degree of solid slag filmcrystallinity.5–10) So the additives of Li2O, MgO and B2O3

were often added to keep the stable physical properties ofmold fluxes.3,11) And the powders were tried to make lowerviscosity, melting point and higher melting rate for highspeed casting by adding Na2O, CaF2 and Li2O. The resultsshow that the effects were obvious with minor addition ofLi2O.12–15) Therefore, Li2O was often added in the moldfluxes for high speed casting.

The physical properties of mold fluxes, such as viscosityand solidification temperature are critical elements in thedesign of optimized mold fluxes for high speed casting.Minor addition of Li2O (�2 mass%) can significantly alter

the physical properties of a mold flux, provide adequate lu-brication to prevent sticking and produce strands with de-fect free. So the productivity and quality of steels are im-proved. The crystallization rate of mold fluxes can be obvi-ously increased by the addition of Na2O to form the high-density crystalline layer,16) so Li2O and Na2O are oftenfound in the commercial mold fluxes. But it was reportedthat Na2O and MgO were substituted by high content ofLi2O to form Li-cuspidine and increase the crystallizationrate.17) The lubrication can be improved by increasing thecontent of Li2O, but the influence of high content of Li2Oon the heat transfer hasn’t been reported. And the heattransfer through the slag film in the mold primarily dependson two parameters of the mold fluxes, namely the thicknessof the slag film layer and crystallization tendency.18–20)

However, the influence of Li2O on the crystallization ten-dency of mold fluxes hasn’t been understood. Therefore,there is a need to investigate the influence of Li2O on thecrystallization behavior of mold fluxes containing high con-tent of Li2O.

It is usually thought that the onset of crystallization inmold fluxes must be a function of cooling rate. The averagecooling rate of the mold fluxes in mold may be about50–100°C/s,21) especially near the meniscus. So the criticalcooling rate required for the glass formation is one of theimportant characteristics for the utilization of mold fluxes.

Crystallization Behaviors of Mold Fluxes Containing Li2O UsingSingle Hot Thermocouple Technique

Hui LIU, Guanghua WEN and Ping TANG

Department of Metallurgy, College of Materials Science and Engineering, Chongqing University, 174 Shazhengjie, Chongqing400044, P. R. China. E-mail: wengh@cqu.edu.cn.

(Received on November 6, 2008; accepted on February 23, 2009)

Minor addition of Li2O (�2 mass%) can significantly alter the melting temperature and viscosity of a moldflux. However, the crystallization tendency of mold fluxes containing Li2O more than 2 mass% hasn’t beenunderstood. In this paper, single hot thermocouple technique, X-ray diffraction, and Back Scattered Electron(BSE) were employed to investigate the crystallization behavior concerned with CCT and TTT diagrams ofmold fluxes containing high content of Li2O.

The critical cooling rate and crystallization rate of samples were increased with the addition of Li2O,whether there is Na2O or not. But the incubation time was reduced obviously with Li2O more than 6 mass%in mold fluxes. The critical cooling rate and incubation time of NL1.5 sample (Na2O�8 mass%, Li2O�1.5 mass%) were close to those of L7 sample (Na2O�0 mass%, Li2O�7 mass%). Moreover, there is no dif-ference in the crystal phase of the two samples, and the crystal phases precipitated at high temperatureand low temperature were respectively Ca2(SiO4) and Ca4Si2O7F2. Li-cuspidine was promoted to precipitatewhen the basicity of the sample was increased. The small, dense grains of cuspidine were observed inNL1.5 sample, and the thickness of slag film was 0.4 mm more than that of L7 sample. The heat flux of NL1.5sample measured was 10% less than that of L7 sample, as expected. Therefore, high content of Li2O canbe substituted by the combination of Na2O and minor of Li2O, as is the composition in the commercial moldfluxes.

KEY WORDS: mold fluxes; single hot thermocouple technique; critical cooling rate; incubation time.

843 © 2009 ISIJ

Moreover, the crystal morphology, chemistry, growth rateand the time evolution fraction of solid are important parameters in the effect on heat transfer of mold fluxes.Therefore, either continuous cooling transformation dia-grams (CCT curves) or isothermal time temperature trans-formation diagrams (TTT curves) must be constructed tocomprehensively understand the crystallization tendency ofmold fluxes.21–24) And hot thermocouple techniques whichcombine video observation and image analysis were devel-oped to meet the requirements of cooling rates. In thispaper, single hot thermocouple technique (SHTT) was cho-sen to investigate the crystallization behavior concernedwith CCT and TTT diagrams of mold fluxes containingLi2O.

2. Experimental

2.1. Sample Preparation

Experimental samples have two systems: NL samples(Na2O�8 mass%, Li2O�0–3 mass%) and L samples(Na2O�0, Li2O�2–8 mass%). The basicity ((CaO�56/78CaF2)/SiO2), Al2O3 and F� of all the samples are 1.25,4 mass% and 6 mass%, respectively. Synthesized sampleswere prepared by mixtures of high purity CaCO3, SiO2,Al2O3, CaF2, Li2CO3 and Na2CO3 powders (99.5 mass%).These powders were weighed at various initial composi-tions and melted in graphite crucibles in air atmosphereusing MoSi2 furnace. The liquid mixture was held at

1 400°C for 10 min. Each sample was quenched in water,and then dried, crushed, grounded and sieved by 200 meshscreen. It was confirmed that the samples were amorphousby X-ray diffraction (Rigaku Co. D/MAX-3C). The meltingtemperature and viscosity were measured by hemispherepoint and rotating viscometer. The chemical compositions(mass%) calculated and physical properties of experimentalsamples are shown in Table 1.

2.2. Experimental Process

The heating and cooling of the SHTT can be controlledby means of a computer program, so continuous coolingexperiments and isothermal experiments can be performed.Pure K2SO4 was used to calibrate the temperature measure-ment occurring from a contact potential difference betweenthe electrode and relay circuit. Figure 1(a) shows the ther-mal cycle employed in the continuous cooling experiment.First of all, a small sample was heated on the tip of the ther-mocouple at the rate of 15°C/s. The sample was held at1 500°C for 60 s to eliminate bubbles, and homogenize itscomposition and temperature. Then the sample was cooleddown to 900°C at different cooling rates. The cooling rateat which there is no crystals precipitated is defined as thecritical cooling rate. CCT diagrams were constructed byrecording the crystallization temperature and the beginningtime of crystallization at different cooling rates.

The thermal cycle employed in the isothermal experi-ment is shown in Fig. 1(b). The sample was held at 1 500°C

ISIJ International, Vol. 49 (2009), No. 6

844© 2009 ISIJ

Table 1. Chemical composition (mass%) and physical properties of experimental mold fluxes.

for 60 s, and then it was rapidly cooled down at the coolingrate of 80°C/s22) to different isothermal temperatures. Thatis the starting of the isothermal experiment. TTT diagramswere constructed by recording the onset time of crystalliza-tion at various isothermal temperatures.

Figure 2 shows the crystallization process of a sample.The mold fluxes start to crystallization when the crystalscan be observed, and the volume fraction of crystallizationis defined as 0.5%22) by image analysis. SEM (TescanVega II LMU) was used on quenched samples which weremelted on the SHTT at 1 500°C and quickly cooled to a de-sired temperature to measure the microstructure of thesesamples. As the amount of sample quenched from SHTTwas very small (�5 mg), it is very difficult to use XRDwith a small sample. Therefore, another quenched experi-ment was tried. Samples (100 g) were melted in graphitecrucibles in air atmosphere at 1 400°C for 5 min, and cooledat 10°C/min to various isothermal temperatures, and heldfor 30 min. The samples were taken out from the furnaceand quenched in water. The phases of the quenched sam-ples were analyzed by XRD.

3. Experimental Results

3.1. The Results of Continuous Cooling Experiment3.1.1. Mold Fluxes Containing Li2O and Na2O

The CCT diagram of NL1.5 (Na2O�8 mass%, Li2O�1.5

mass%) sample was shown in Fig. 3(a). The crystallizationtemperature of mold fluxes lowers as the cooling rate in-creases, as expected. As is shown in Fig. 3(b), the criticalcooling rate of NL samples first appreciably reduces, thenincreases with the increase of XLi2O

/(XLi2O�XNa2O

), which isconsistent with the results of Ota et al.25) NL1.5 sample hasthe minimum critical cooling rate 3°C/s obtained from CCTdiagram, and the critical cooling rate of NL samples wasincreased 6°C/s per 1 mass% Li2O to obviously strengthenthe crystallization tendency.

3.1.2. Mold Fluxes Containing High Content of Li2O

Figure 4 is the continuous cooling results of L sampleswithout Na2O, and the mold fluxes were easily glass forma-tion when the content of Li2O was less than 6 mass%.When Li2O is over 6 mass%, the critical cooling rate of Lsamples increased 4°C/s per 1 mass% Li2O. Also, the re-sults previous reported by Tsutsumi et al.26) were shown forcomparison. It’s indicated that the influence of Li2O on thecrystallization tendency of mold fluxes found by Tsutsumiet al. was more obvious owing to the addition of high con-tent of Li2O in CaO–SiO2–Li2O slag system.

3.2. TTT Diagrams3.2.1. Mold Fluxes Containing Li2O and Na2O

Figure 5(a) shows that the incubation time of NL sam-ples became shorter with the addition of Li2O, and the incu-

ISIJ International, Vol. 49 (2009), No. 6

845 © 2009 ISIJ

Fig. 1. Thermal cycle of experiments.

Fig. 2. The crystallization process of a sample.

bation time was reduced by 25 s when the content of Li2Owas increased from 0.5 to 1.5 mass%. But there is no obvi-ous influence on the crystallization rate to further increasethe content of Li2O.

3.2.2. Mold Fluxes Containing High Content of Li2O

From Fig. 5(b), we can see that the incubation time of Lsamples was shortened by 50 s when the content of Li2Owas increased from 6 to 7 mass%. Considering the result ofOmoto et al.17) indicates that Li2O has a positive effect onthe crystallization rate of mold fluxes. The incubation timeof L8 sample is close to zero to increase the frequency ofsticking type breakout. And there’s no crystals precipitatedat temperature over 1 050°C for L6 sample, so it goesagainst to control the heat transfer in the region of menis-cus. It was reported that after 20 s from the casting the crys-tallization of cuspidine starts and the thermal resistance ofmold flux increases in commercial mold fluxes.24) It is inter-esting to note that the incubation time of the L7 sample was20 s to form appropriate thickness of the crystalline layer.

4. Discussion

4.1. The Critical Cooling Rate

As is shown in Fig. 3 and Fig. 4, the crystallization ten-dency of mold fluxes was strengthened with the addition ofLi2O, whether there is Na2O or not. Besides, the influencesof Li2O on the critical cooling rate in both NL and L sam-

ples were close to each other. The electrostatic potential ofLi� (1.47 Å) is very close to that of SiO4

4� (1.44 Å),27) andthe cationic size order and the order of cationic strength isNa��Li� and Li��Na�,28) respectively. So other cationswould be hindered to combine with SiO4

4� by the additionof Li2O in mold fluxes. But the polarization of Li� to SiO4

4�

was too small to divide the SixOyz� polymerid. Therefore,

the minor additive of Li2O would improve the glass forma-tion of mold fluxes. With the increase of M2O/SiO2 ratio thechain breakdown of silica network continued, and the crys-tallization property of mold fluxes would be strengthened.

It’s interesting to note that the critical cooling rate of NLand L samples started to increase when the value of(XLi2O

�XNa2O)/XSiO2

was over 0.30. It’s indicated that theoxygen ion of alkali oxide divides the silicate flow unit as anon-bridging oxygen with (XLi2O

�XNa2O)/XSiO2

�0.30 in thecomposition range of experimental samples. But the rela-tion needs to be verified in the future work.

4.2. Crystallization Rate

The crystallization rate of mold fluxes was improved bythe addition of Li2O (shown in Fig. 5), and the incubationtime of mold fluxes containing Li2O more than 6 mass%was decreased much more obviously. The cationic strengthof alkali ions increases in the order of Li��Na�, thus thepresence of cations with lower strength results in the inter-action of a higher ratio of bridge oxygen.29) Consequently,the viscosity of L samples was greatly reduced to promote

ISIJ International, Vol. 49 (2009), No. 6

846© 2009 ISIJ

Fig. 3. The influence of Li2O on the critical cooling rate of NL slags.

Fig. 4. The influence of Li2O on the criticalcooling rate of L slags.

Fig. 5. The influence of Li2O on the incubation time of samples.

the precipitation of solids.Cramb et al.16) pointed out that the incubation time of

mold fluxes became shorter with the addition of Na2O, sothere is often some Na2O found in the composition of com-mercial mold fluxes. Alkali oxides have similar influenceon the melting temperature and viscosity of mold fluxes,and the influences of Li2O, Na2O on the critical cooling rateand incubation time were close to each other.30) Therefore,Li2O can be substituted by Na2O to reduce the cost. NL1.5

(Na2O�8 mass%, Li2O�1.5 mass%) and L7 (Na2O�0,Li2O�7 mass%) samples were chosen owing to similar crit-ical cooling rate.

Figure 6 shows TTT diagrams of NL1.5 and L7 samples,and the incubation time of the two samples was close toeach other in the temperature below 1 100°C. But the crys-tals precipitated could be observed over 1 150°C in NL1.5

sample. NL1.5 sample has 100°C higher melting point toprovide more driving force of crystallization. The high den-sity crystalline layer was formed by the fast crystallizationat high temperature to increase the interfacial thermal re-sistance.

4.3. The Results of XRD and BSE

The mineralogical phase of the precipitated crystals pro-vides additional information to understand the crystalliza-tion behavior of mold fluxes. If Li2O was substituted byNa2O, the following substitution reaction (1) should betaken into account according to Yasushi et al.31–35) But itwas found that most of F� ions are fourfold coordinated byCa2� rather than Na� for samples of nNa/(nNa�nCa)�0.3,and the presence of Na2O may not prevent the precipitationof cuspidine from the mold fluxes.36–39) In the compositionrange of experimental samples, the value of nNa/(nNa�nCa)was less than 0.3, and it was confirmed by XRD that therewas precipitation of cuspidine for all the samples. So Li2Osubstitution by Na2O has no effect on the crystal morphol-ogy in the crystalline film layer.

Na2O�CaF2�CaO�2NaF .....................(1)

The TTT diagrams of NL1.5 and L7 were divided into two“C” regions, which suggested two separate nucleationevents. Figures 7(a) and 7(b) show XRD results of the two

samples, and there is no difference in the crystal phase pre-cipitated. The crystal at higher temperature was dicalciumsilicate (Ca2(SiO4)), and cuspidine (Ca4Si2O7F2) was thedominant phase measured at low temperature. But there’sno Li-cuspidine found in the high content of Li2O samplesreported by Omoto.17)

When the basicity ((CaO�56/78CaF2)/SiO2) and Li2O inmold fluxes are 2.2 and 4 mass%, a quantity of 350 g slagwas melted in an apparatus to obtain the slag film layer. Thedetails of the experimental apparatus have been reported inliterature.40) The copper detector was lifted up and the at-tached solid slag was removed after an immersion time of45 s. The solid slag film was grounded and sieved by 200mesh screen to measure the precipitated phase by XRD. Itwas confirmed that Li-cuspidine was found in the solid slagfilm (shown in Fig. 7(c)).

It’s assumed by the authors that reaction (2) took place,where 1 mol cuspidine reacted with 1 mol Li2O to form2 mol Li-cuspidine. The viscosity of basicity 2.2 samplewas 0.019 Pa · s to break the Si–O–Si bond of cuspidine,and Li� was compensated to the Si–O unit owing to thesmaller size and higher strength. The breakdown of the sili-cate network with the addition of Li2O to mold fluxes was

ISIJ International, Vol. 49 (2009), No. 6

847 © 2009 ISIJ

Fig. 6. The comparison of incubation time between observationand calculation.

Fig. 7. X-ray diffraction patterns of samples (Co-Ka).

shown in Fig. 8. The viscosity of L8 was 0.112 Pa · s, toohigh to break the Si–O–Si bond of cuspidine. It is consis-tent with the results that no Li-cuspidine was found in thehigh content of Li2O samples. But the assumption on theformation mechanism of Li-cuspidine needs to be verifiedin the future work.

Ca4Si2O7F2�Li2O�2LiCa2FSiO4 ................(2)

The microstructure of quenched samples melted on theSHTT is shown in Fig. 9 and Fig. 10, and the crystallineratio of NL1.5 and L7 increased with the decrease of theisothermal temperature. The crystals of NL1.5 at 1 100°Ctook dendrite structure, and the crystals became small anddense when the temperature was reduced. It is consideredthat cuspidine with the dendrite structure crystallizes rap-idly from liquid.

But the crystals of L7 sample presented coarse and sparsestructure. The viscosity of NL1.5 is 0.276 Pa · s, two times ofthat of L7, so it was more favorable for the nucleation.

4.4. Heat Flux through the Solid Slag Film

An apparatus for simulating copper mold was also usedto measure the heat flux of the slag film,40) which was cal-culated based on the temperature difference of water be-tween outlet and inlet against immersion time. As is shownin Table 2, NL1.5 sample has 10% less heat flux, and thethickness of slag film was 0.4 mm more than that of L7.

As the critical cooling rates of NL1.5 and L7 samples weretoo small to promote the crystals precipitated in the coolingrate range of the apparatus, no crystal was identified byXRD in the slag film. So the two samples were melted inMoSi2 furnace, held at 1 000°C for 10 min, and then pouredinto water. Figure 11 shows the BSE results of quenchedsamples. The crystal phase which consists of small grainsof dendrite Ca4Si2O7F2 was observed in NL1.5 sample. Butglassy phases and the large grains of Ca4Si2O7F2 were ob-

ISIJ International, Vol. 49 (2009), No. 6

848© 2009 ISIJ

Fig. 8. The reaction process of Li-cuspidine.

Fig. 9. The BSE of NL1.5 sample.

Fig. 10. The BSE of L7 sample.

Table 2. The influence of Li2O on the heat transfer and thick-ness of slag film layer.

Fig. 11. The BSE of sample quenched at 1 000°C.

served in the L7 sample, and the crystals presented bonestructure. The results of quenched samples were consistentwith those of SHTT.

The thickness of the slag film layer and crystallizationtendency of NL1.5 both contributed to the decrease of theheat transfer through the slag film, which provided the bed-stone to increase the speed casting for crack-sensitivesteels. It’s necessary to note that the mold fluxes containinghigh content of Li2O didn’t have obvious influence on theheat transfer through the slag film layer.

4.5. Kinetics Analysis of Crystallization

When the growth and nucleation of crystals were bothconsidered, the Johnson–Mehl–Avrami equation was oftenused, as is shown in Eq. (3), where n represents the mecha-nism of crystal growth and K is a coefficient correspondingto the nucleation and growth mechanism. And the exponentn can be described by a growth constant (q) and a nucle-ation constant (p), according to Eq. (4). Equation (3) can berewritten in the form of Eq. (5), which is suitable of theevaluation of K and n for each sample.

XR�1�exp(�K · tn) ...........................(3)

n�p�q ....................................(4)

ln[�ln(1�XR)]�n · ln t�ln K....................(5)

The comparison of volume fraction of NL1.5 and L7 sam-ples between observation and calculation is shown in Fig.12. And we can see that the exponent of NL1.5 sample is 3,indicating two dimensional growth and the interface reac-tion was the dominant condition. While the exponent n ofL7 sample is close to 4, which means that the limit case ofthree dimensional growth with constant nucleation andgrowth rates. This result is similar to the crystallizationmechanism of the mold fluxes reported by Kashiwaya etal.22) The following expressions (6)–(9) give acceptable cal-culation results that fit the measured incubation time as afunction of temperature.

NL1.5:

...........................................(6)

...........................................(7)

L7:

...........................................(8)

...........................................(9)

5. Conclusions

In this paper, SHTT, XRD and BSE were employed tostudy the crystallization behavior of mold fluxes containingLi2O. And the conclusions are as follows:

(1) The critical cooling rate of NL and L samples in-creased 6°C/s and 4°C/s per 1 mass% Li2O, and the incuba-tion time was reduced by 25 s and 50 s, respectively. Thecrystallization rate was promoted obviously in mold fluxescontaining high content of Li2O.

(2) The critical cooling rate and incubation time ofNL1.5 sample were close to those of L7 sample, and the pre-cipitation of cuspidine wasn’t prevented by addition ofNa2O.

(3) There is no difference in the crystal phase of NL1.5

and L7 samples, and it was confirmed by XRD that thecrystal phases precipitated at high temperature and lowtemperature were Ca2(SiO4) and Ca4Si2O7F2, respectively.Li-cuspidine was promoted to precipitate when the basicityof the sample was increased.

(4) The small and dense grains of cuspidine were ob-served in NL1.5 sample, and the thickness of slag film was0.4 mm more than that of L7 sample. The heat flux of NL1.5

sample was 10% less than that of L7 sample, as expected.Mold fluxes containing high content of Li2O didn’t have ob-vious influence on the heat transfer through the slag filmlayer. Therefore, high content of Li2O can be substituted bythe combination of Na2O and minor of Li2O, as is the com-position found in the commercial mold fluxes.

tT T

T0 005

9

1 1 2 2

1 6 10

1200 700900 1065. . .

.

( ) ( ),�

� �� �° °C C

tT T

T0 005

16

6 3 2 3

1 0 10

1170 10001065 1140. . .

.

( ) ( ),�

� �� �° °C C

tT T

T0 005

15

2 7 3 8

7 8 10

1200 850950 1080. . .

.

( ) ( ),�

� �� �° °C C

tT T

T0 005

10

1 7 1 9

1 0 10

1400 10001080 1220. . .

.

( ) ( ),�

� �� �° °C C

ISIJ International, Vol. 49 (2009), No. 6

849 © 2009 ISIJ

Fig. 12. The comparison of volume fraction between observation and calculation.

Acknowledgements

The authors wish to express their gratitude to the Na-tional Natural Science Foundation (Grant No.: 50574109)for funding the current study. The special efforts of JuntaoBa and Yanhong Zhao to measure the heat flux and thick-ness of slag film are also greatly appreciated.

REFERENCES

1) Y. Meng, B. G. Thomas and A. A. Polycarpou: Can. Metall. Q., 45(2006), 79.

2) K. C. Mills, A. B. Fox and Z. Li: Ironmaking Steelmaking, 32(2005), 26.

3) K. C. Mills and A. B. Fox: ISIJ Int., 43 (2003), 1479.4) J. Diao, B. Xie and N. Wang: ISIJ Int., 47 (2007), 1294.5) A. B. Fox and M. E. Valdez: ISIJ Int., 44 (2004), 836.6) J.-H. Park, I.-H. Jung and H.-G. Lee: ISIJ Int., 46 (2006), 1626.7) C. Orring, S. Sridhar and A. W. Cramb: ISIJ Int., 40 (2000), 877.8) P. Rocabois, J. N. Pontoire, J. Lehmann and H. Gaye: J. Non-Cryst.

Solids, 282 (2001), 98.9) D.-C. Park, I.-H. Jung and P. C. H. Rhee: ISIJ Int., 44 (2004), 1669.

10) S. Sridhar and A. W. Cramb: Metall. Mater. Trans. B, 31B (2000),406.

11) K. C. Mills: Mold Powders for Continuous Casting, Chapter 8, theAISE Steel Foundation, Pittsburgh, (2003), 15.

12) A. B. Fox, K. C. Mills and D. Lever: ISIJ Int., 45 (2005), 1051.13) H. Wang, G. Li and Q. Dai: ISIJ Int., 46 (2006), 637.14) E. Asayama, H. Takebe and K. Morinnaga: ISIJ Int., 33 (1993), 233.15) S. Sukenaga, N. Saito and K. Kawakami: ISIJ Int., 46 (2006), 352.16) C. Mutale, T. Claudon and A. W. Cramb: Metall. Mater. Trans. B,

36B (2005), 417.

17) T. Omoto, H. Ogata and J. Itoh: Shinagawa Tech. Rep., 49 (2006),73.

18) J. W. Cho and H. Shibata: J. Non-Cryst. Solids, 282 (2002), 110.19) M. Hayashi, R. Abdul Abas and S. Seetharaman: ISIJ Int., 44

(2004), 691.20) J. Cho, H. Shibata and T. Emi: ISIJ Int., 38 (1998), 440.21) Y. Kashiwaya, A. W. Cramb and K. Ishi: ISIJ Int., 38 (1998), 348.22) Y. Kashiwaya, C. E. Cicutti and A. W. Cramb: ISIJ Int., 38 (1998),

357.23) Y. Kashiwaya, T. Nanauchi and K. S. Pham: ISIJ Int., 47 (2007), 44.24) H. Nakada and K. Nagata: ISIJ Int., 46 (2006), 441.25) R. Ota, T. Wakasugi, W. Kawamura and B. Tuchiya: J. Non-Cryst.

Solids, 188 (1995), 136.26) K. Tsutsumi, T. Nagasaka and M. Hino: ISIJ Int., 39 (1999), 1150.27) X. Huang: The Principle of Ferrous Metallurgy, Metallurgical Indus-

try Press, Chongqing, (1986), 93.28) T. Watanabe, H. Hashimoto and M. Hayashi: ISIJ Int., 48 (2008),

925.29) A. Angelopoulou, V. Montouillout and D. Massiot: J. Non-Cryst.

Solids, 354 (2008), 333.30) E. Asayama, H. Takebe and K. Morinaga: ISIJ Int., 33 (1993), 233.31) K. Tsutsumi, J. Ohtake and T. Nagasaka: Tetsu-to-Hagané, 84

(1998), 66.32) A. Angelopoulou, V. Montouillout, D. Massiot and G. Kordas: J.

Non-Cryst. Solids, 354 (2008), 333.33) Y. Sasaki, H. Urata and K. Ishi: ISIJ Int., 43 (2003), 1897.34) Y. Sasaki and K. Ishi: ISIJ Int., 44 (2004), 660.35) H. Nakada, H. Fukuyama and K. Nagata: ISIJ Int., 46 (2006), 1660.36) M. Hanao, M. Kawamoto and T. Watanabe: ISIJ Int., 44 (2004), 827.37) T. Watanabe, H. Fukuyama and K. Nagata: ISIJ Int., 42 (2002), 489.38) M. Hayashi, T. Watanabe and K. Nagata: ISIJ Int., 44 (2004), 1527.39) M. Hayashi, T. Watanabe and H. Nakada: ISIJ Int., 46 (2006), 1805.40) G. Wen, S. Sridhar and P. Tang: ISIJ Int., 47 (2007), 1117.

ISIJ International, Vol. 49 (2009), No. 6

850© 2009 ISIJ