ISIJ International, Vol. 52 (2012), No. 12, pp.

2130–2138

Influence of Liquid Metal Properties on Water Atomised Iron

Powders

Fredrik PERSSON,1)* Anders ELIASSON2) and Pär Göran JÖNSSON2)

1) Global Process Development, Höganäs AB, SE-263 83 Höganäs, Sweden.

2) Department of Materials Science and Engineering, KTH Royal Institute of Technology, SE-100 44, Stockholm, Sweden.
(Received on May 16, 2012; accepted on July 30, 2012)

The main focus of the present study was the influence of liquid metal properties on the particle size
during water atomisation. Experiments for liquid iron showed that alloy additions of carbon and sulphur
decreased the particle size. Moreover, it was indicated that the reduced d50 value at increased %C and
%S contents may be related to a decreased viscosity and surface tension respectively. An alternative
mechanism could be that raised superheats at increased carbon contents increased the total available
time for atomisation. This may also have decreased the particle size. The influence of surface tension and
viscosity on the d50 value was further analysed with a theoretical d50 model proposed in a previous work.
A reduced viscosity from 6.8 to 4.3 mPa s decreased the d50 value with 33%. In addition, the particle size
was estimated to decrease with 27% by decreasing the surface tension from 1 850 to 900 mN m–1.

KEY WORDS: water atomisation; surface tension; viscosity; metal powders; particle size; superheat.

melt was increased from 0.02 to 0.55% at a superheat of

1. Introduction
Water atomisation of liquid metals has been used for
many years to produce a large number of different metal
powders. The scale of operation varies, with melting units
from a few kg up to large production atomisers over 100
tons.1) Examples of commercial applications are iron and
steel powders for sintered components,2) silver powders for
electrical contacts,3) and ultrafine stainless steel powders for
metal injection moulding.4)
In a study by Dunkley,5) it was illustrated that the physical
properties of the liquid metal can have a significant influ-
ence on the particle size for water atomized metal powders.
It was observed that the mass median particle size (d50) can
vary by a factor of 6 for different metals and alloys, which
were atomised under identical conditions. Additional studies
by Dunkley and Palmer6) showed that the superheat of the
liquid metal can have a large effect on the particle size dur-
ing water atomisation. It was concluded that the d50 value
decreased by about 5–10% per 100°C of superheat for alloys
with a liquidus temperature above 500°C. These results
were confirmed six years later by Ankus and Wenter,3) who
investigated the influence of superheat for water atomised
silver powders. The mass median particle size decreased
between 7–9% per 100°C of superheat at water pressures
between 17.2 and 33.8 MPa.
Klar and Shafer7) alloyed liquid copper to study how a
decreased surface tension influenced the particle size for
water atomised copper powders. It was estimated that the d50
value decreased by 13%, when the oxygen content in the
* Corresponding author: E-mail: Fredrik.s.persson@hoganas.com
DOI: http://dx.doi.org/10.2355/isijinternational.52.2130
around 150°C. The influence of surfactants on the particle
size was further investigated by Bergquist,8) who alloyed
sulphur to water atomised iron powders. Very limited influ-
ence on the particle size was seen for sulphur contents up to
0.20%. However, a final test at a water pressure of 12.5 MPa
indicated that the d50 value could be reduced by 22% if pure
liquid iron is alloyed with 1.5%S.
The main objective of the present study is to investigate
the relation between the particle size and the physical prop-
erties of the liquid metal during water atomisation. Design
of experiments (DOE)9) was used to study how the d50 value
was influenced by increased contents of %C and %S in liq-
uid iron. Other parameters included in the DOE study were
the water pressure, water level in the atomising tank, type
of water jet and melt stream diameter. In the final part of the
paper, it is specifically discussed how the particle size may
be influenced by the viscosity, surface tension and superheat
of the liquid metal. An empirical d50 model is also proposed,
which is used to discuss how some important atomising
parameters may be correlated to the particle size.
2. Experimental
2.1. Experimental Facility
The atomising experiments in the present work were car-
ried out in a V-jet pilot water atomiser, Fig. 1. Pure iron was
melted in a 250 kg high-frequency induction furnace using
an alumina refractory lining. Carbon was alloyed with
graphite, which was charged to the furnace before the melt-
ing was started. In addition, Ferro sulphur was added to the
liquid iron just before the start of atomisation. The furnace
was covered with a steel lid during the entire melting

© 2012 ISIJ 2130

 ISIJ International, Vol. 52 (2012), No. 12

Table 1. Experimental programme for the DOE study, see

“Nomenclature” for parameter definitions.

Test Heat pW (MPa) J DM/D0 H/H0 %C %S

1 D1959 8 A 0.714 1 4 0.5
2 D1951 18 A 0.714 1 0.05 0.004
3 D1961 8 B 0.714 0.38 0.05 0.5
4 D1952 18 B 0.714 0.38 4 0.004
5 D1957 8 A 1 0.38 4 0.004
6 D1956 18 A 1 0.38 0.05 0.5
7 D1954 8 B 1 1 0.05 0.004
8 D2134 18 B 1 1 4 0.5
9 D1974 13 A 0.857 0.69 2.03 0.25
10 D1960 13 A 0.857 0.69 2.03 0.25
11 D1953 13 B 0.857 0.69 2.03 0.25
12 D1958 13 B 0.857 0.69 2.03 0.25

Table 2. DOE study, parameters and results for test 1–6, see
“Nomenclature” for parameter definitions.

Test 1 2 3 4 5 6
Heat D1959 D1951 D1961 D1952 D1957 D1956
Fig. 1. Experimental set-up for V-jet water atomisation trials. Parameters
Atomising time (s) 251 222 222 244 136 54
sequence. Nitrogen gas was purged through the lid, to mM (kg s ) –1
0.96 0.95 0.95 0.82 1.62 1.39
decrease the degree of oxidation of the melt. The liquid steel
mW (kg s–1) 4.99 6.51 4.16 5.27 4.99 6.51
temperature was measured with an expendable temperature
Temperature furnace
sensor, which was immersed with a steel lance in the liquid (°C)
1 626 1 624 1 617 1 617 1 620 1 630
steel. When the target steel temperature of 1 620°C was
Temperature tundish
reached, the furnace was tilted and the liquid metal was (°C)
1 560 1 552 1 547 1 530 1 580 -
poured into a tundish with an alumina refractory lining. A
Liquidus temperature
Liquefied petroleum gas burner was used to preheat the 1 236 1 520 1 519 1 167 1 241 1 513
(°C)
tundish, which was red hot before the atomisation was started. μM (m Pa s) 5.03 6.27 6.76 4.34 4.93 6.49
The steel temperature was measured in the furnace just before
γM (mN m–1) 926 1 680 901 1 849 1 712 920
atomisation and after 1–2 minutes of atomisation in the
WeL/ReL (Eq. (13)) 0.59 0.61 0.82 0.38 0.31 1.15
tundish. The liquid iron was poured through a tundish nozzle
into the V-jet water atomising system. The V-jet has in total As atomised powder
eight symmetrically arranged water nozzles. These include LECO (wt %)
two pairs of main and side jets respectively and four support- C 3.38 0.22 0.04 4.22 3.53 0.13
ing jets. The jet angle between the water jet and the metal N 0.002 0.005 0.008 0.001 0.002 0.008
stream was 26.5° for the main jets. Nitrogen was used to flush O 0.41 0.39 0.48 0.29 0.32 0.96
the atomising tank, to avoid the risk for hydrogen explosions.
S 0.513 0.006 0.503 0.004 0.008 0.443
The steel level in the tundish was regulated manually to
around 20 cm during the atomisation. The steel flow rate to ICP-OES
the atomizer was stopped immediately when the furnace was Al 0.004 0.002 0.001 0.000 0.002 0.002
empty, by pressing a steel rod through the melt to block the Cr 0.007 0.009 0.001 0.009 0.003 0.017
tundish nozzle. The powder slurry was dewatered after atom- Cu 0.002 0.004 0.001 0.002 0.002 0.004
isation. The wet iron powder cake was finally dried at 180°C Mn 0.028 0.022 0.012 0.041 0.032 0.023
by using warm compressed air. The drying was interrupted
Mo 0.000 0.001 0.000 0.000 0.000 0.001
when the temperature of the iron powder reached 95°C.
Ni 0.015 0.018 0.014 0.015 0.015 0.018

2.2. Experimental Programme and Procedure P 0.003 0.002 0.002 0.003 0.002 0.003
The experimental programme was planned and evaluated Si 0.025 0.014 0.000 0.076 0.051 0.028
with the software Modde 9 developed by Umetrics AB.9) A Ti 0.019 0.000 0.000 0.106 0.038 0.002
26–3 fractional factorial design with resolution III and four V 0.071 0.063 0.040 0.091 0.079 0.043
centre points was chosen for the study, Table 1. Investigated Particle size
factors were pW (water pressure), J (configuration of the
d50 (μ m) 80.5 59.3 103.6 48.4 131.8 57.6
water jet), DM/D0 (relative melt stream diameter), H/H0 (rel-
σLN 2.16 2.34 2.46 2.12 2.00 2.47
ative water level in the atomising tank), %C (carbon content
in the liquid steel) and %S (sulphur content in the liquid AD (g cm–3) 3.42 2.94 3.38 3.38 3.20 3.27

2131 © 2012 ISIJ

 ISIJ International, Vol. 52 (2012), No. 12

Table 3. DOE study, parameters and results for test 7–12, see Table 4. Influence of melt stream diameter, parameters and results
“Nomenclature” for parameter definitions. for test 13–16, see “Nomenclature” for parameter defini-
tions.
Test 7 8 9 10 11 12
Test 13 14 15 16
Heat D1954 D2134 D1974 D1960 D1953 D1958
Parameters Heat D1981 D1982 D1983 D1984

Atomising time (s) 134 140 178 113 136 114 Parameters

mM (kg s–1) 1.60 1.60 1.26 1.24 1.14 1.18 Atomising pressure (MPa) 15 15 15 15

mW (kg s–1) 4.16 5.27 5.68 5.68 4.77 4.77 Atomising time (s) 195 132 112 70

Temperature furnace DM/D0 0.571 0.714 0.857 1

1 627 1 616 1 625 1 625 1 626 1 620
(°C) mM (kg s ) –1
0.57 0.87 1.12 1.79
Temperature tundish mW (kg s–1) 2.22 3.28 5.00 6.67
1 565 - - 1 571 1 575 -
(°C)
Temperature furnace (°C) 1 620 1 620 1 621 1 619
Liquidus temperature
1 528 1 152 1 382 1 389 1 365 1 383 Liquidus temperature (°C) 1 468 1 458 1 449 1 460
(°C)
μM (m Pa s) 6.50 4.72 5.79 5.80 5.76 5.79 μM (m Pa s) 5.90 5.86 5.84 5.86

γM (mN m ) –1
1 596 965 1 016 1 000 1 015 1 048 γM (mN m–1) 1 639 1 629 1 662 1 649

WeL/ReL (Eq. (13)) 0.44 0.80 0.79 0.81 0.79 0.77 WeL/ReL (Eq. (13)) 0.62 0.62 0.61 0.62

As atomised powder As atomised powder

LECO (wt %) LECO (wt %)

C 0.12 4.17 1.92 1.83 2.11 1.92 C 0.918 1.059 1.181 1.022

N 0.006 0.001 0.004 0.005 0.005 0.004 N 0.007 0.004 0.004 0.003

O 0.44 0.85 0.28 0.21 0.70 0.38 O 0.19 0.35 0.30 0.36

S 0.011 0.46 0.258 0.276 0.268 0.218 S 0.006 0.007 0.006 0.006

ICP-OES ICP-OES

Al 0.002 <0.001 0.002 0.002 0.002 0.001 Al 0.000 0.000 0.001 0.001

Cr 0.007 0.65 0.110 0.002 0.003 0.002 Cr 0.567 0.141 0.081 0.266

Cu 0.003 0.002 0.001 0.002 0.003 0.002 Cu 0.001 0.001 0.001 0.001

Mn 0.018 0.023 0.020 0.022 0.027 0.023 Mn 0.031 0.025 0.027 0.029

Mo 0.001 0.007 0.000 0.000 0.001 0.000 Mo 0.002 0.000 0.000 0.002

Ni 0.019 0.044 0.016 0.015 0.017 0.014 Ni 0.015 0.014 0.014 0.016

P 0.002 0.003 0.002 0.002 0.002 0.002 P 0.001 0.002 0.002 0.003

Si 0.007 0.011 0.053 0.030 0.019 0.034 Si 0.025 0.022 0.028 0.029

Ti 0.000 0.002 0.003 0.006 0.008 0.007 Ti 0.003 0.001 0.002 0.002

V 0.049 0.095 0.074 0.071 0.070 0.069 V 0.061 0.068 0.075 0.076

Particle size Particle size

d50 (μ m) 176.4 35.4 67.4 82.5 79.2 82.7 d50 (μ m) 78.5 83.1 85.0 91.7

σLN 2.69 2.80 2.41 2.57 2.53 2.48 σLN 2.47 2.33 2.57 2.48

AD (g cm ) –3
3.34 3.12 3.71 3.56 3.49 3.74 AD (g cm–3) 3.10 3.20 3.10 3.08

steel). Two different setups of water nozzles were investi-
gated in the study, jet A and B. The water flow rate in jet A
was 20–24% higher than in jet B, depending on the level of
the water pressure, Tables 2 and 3. The water level during
atomisation was defined as the distance from the tank top to
the water surface in the atomising tank. The water level in
the atomising tank (H) and the melt stream diameter (DM)
were related to a reference value, Table 1. The parameters
H0 and D0 represent the lowest water level and the largest
melt stream diameter, respectively.
The experiments were carried out in a random order, to
minimize that time dependent trends influenced the experi-
mental values. Four additional trials were finally carried out,
to further investigate how the particle size was influenced
by an increased melt stream diameter, Table 4. The heat size
during these four trials was 150 kg.
A powder sample of 5 kg was taken from each heat, by
slowly feeding all atomised powders through a powder sam-
pler. A spinning riffler was finally used to split the powder
to the specific amount needed for each analysis. Carbon,
sulphur, oxygen and nitrogen were determined by a LECO
analysis, according to the standards ISO 15350 (percentag-
es of C and S) and ISO 15351 (percentages of O and N).
Inductively plasma atomic emission spectroscopy was used
to determine trace elements, according to ISO 13898 1-4.
The as-atomised powder particle size was determined by a
sieve analysis (EN ISO 24497). The apparent density was
determined with a Hall flow meter (ISO 3923-1). In addi-
tion, scanning electron microscopy (SEM) was used to study
the particle shape and morphology.
3. Results
As mentioned previously, the main purpose of this paper

© 2012 ISIJ 2132

 ISIJ International, Vol. 52 (2012), No. 12
is to study how the physical properties of the liquid metal
influence the particle size for water atomised metal pow-
ders. Carbon and sulphur was alloyed to liquid iron, to spe-
cifically investigate how the d50 value was influenced by the
viscosity and surface tension of the liquid metal. The target
tapping temperature of the furnace was constant at 1 620°C
for all test heats, to be able to select as consistent data as
possible for these properties. The superheat during the
experiments varied from around 100 to 450°C, depending
on the carbon and sulphur contents, Tables 2–4. The super-
Fig. 2. Curve fitting of as-atomised sieve analysis against log nor-
mal particle size distribution, test 2, heat D1951.
Fig. 3. Scanning electron microscope investigation, test 2, heat
D1951, 0.22%C, particle size of 75–106 μ m.
Fig. 4. Scanning electron microscope investigation, test 4, heat
D1952, 4.22%C, particle size of 75–106 μ m.
2133
heat is here defined as the difference between the steel tem-
perature in the induction furnace just before atomisation and
the liquidus temperature. The software Thermocalc10) was
used to calculate the liquidus temperature for all test heats
in this work. The d50 values for the experiments were calcu-
lated by curve fitting of data from a sieve analysis against
the log normal distribution. The correlation between the
cumulative particle size distribution obtained from sieve
analysis and the log normal distribution was very good for
all experiments in the present study, Fig. 2. Atomised pow-
ders contained a mixture of irregular and rounded particles.
Figures 3 and 4 show the particle morphology for a 0.22%C
and a 4.22%C content, respectively. It is clear that the par-
ticle shape was generally more irregular for powders atom-
ised at low carbon contents.
3.1. Factorial Experiments
All d50 values for the factorial experiments were plotted
in Fig. 5, to compare the replicate centre points to the entire
experimental design, Tables 1–3. The replicate error was rel-
atively small, since the variation for all factorial experi-
ments was significantly larger compared to the replicates
(tests 9–12). The condition number of the design was calcu-
lated, to determine the orthogonality for the experimental
matrix. Experimental data for %C and %S from the trials
were used in this calculation (Tables 2 and 3), while values
presented in Table 1 were selected for all other factors. The
condition number was calculated to 1.35, which is within
the interval of 1–3 typically reported for fractional factorial
designs.9) Figure 5 also shows that the distribution of the d50
values was slightly skewed, with only 3 experiments above
100 μ m. The d50 values were therefore logarithmically trans-
formed, which distributed the data closer to the normal dis-
Fig. 5. Replicate plot of d50 data (untransformed values), factorial
experiments.
Fig. 6. Histogram plot of d50 data after logarithmic transformation
(base 10), factorial experiments.
© 2012 ISIJ
 ISIJ International, Vol. 52 (2012), No. 12

tribution, Fig. 6. This transformation is reasonable, since for

instance several studies have shown that the water pressure
has a very strong and non-linear relation to the particle
size.3,5,6,8) Multiple linear regression (MLR) was finally used
to develop a model for the d50 value, based on the factorial
experiments. The goodness of fit and predictive power for
this model was evaluated with R2 and Q2 respectively, which
were defined as follows9)
SS − SSresid
R2 = ............................ (1)
SS
Fig. 7. Scaled and centred regression coefficients for d50 data, fac-
SS − PRESS
Q2 = ........................... (2) torial experiments: Number of experiments = 12, Degree of
SS freedom of residuals = 7, R2 = 0.977, Q2 = 0.927, Residual
standard deviation = 0.036, Confidence level = 0.95.
where SS and SSresid are the sum of squares of Y and resid-
uals respectively, corrected for the mean. The variable
PRESS is the Prediction Residual Sum of Squares, which
was calculated to obtain a cross validation of the model:

( Y − Yl )
2
i i
PRESS = Σ ........................ (3)
(1 − hi )
i 2

where ( Yi − Yl i ) is the residual and hi is the ith diagonal ele-

ment of the Hat matrix, X(X′X)–1X′.
In general, Q2 values above 0.5 and 0.9 are considered as
good and excellent respectively. In addition, Q2 and R2
should not preferably be separated by more than a factor of
0.2 to 0.3 for a model with a good predictive power. Factors
with a significant influence on the d50 value are presented in
Fig. 7, where the error bars correspond to a 95% confidence
interval. A high goodness of fit and predictive power were
obtained for the MLR model, with R2 and Q2 values equal
to 0.98 and 0.93 respectively. The MLR model is presented
in Eq. (4)
DM
log d50 = 2.321 − 0.037 pW + 0.234
D0 ........... (4)
− 0.036[%C ] − 0.321[%S ]
The water pressure showed the largest influence on the
d50 value during the factorial experiments, Fig. 7. A signif-
icant effect was also obtained by alloying carbon and sul-
phur to the liquid steel. Calculations with Eq. (4) and the
centre point values in Table 1 indicated that an increased
carbon content from 0.05 to 4% decreased the d50 value by
28%. Similar estimations for sulphur indicated a decrease of
the particle size by 31%, at increased sulphur contents from
0.004 to 0.5%.
The influence of the melt stream diameter on the d50 value
was small and more uncertain, but still significant at a 95%
confidence interval level. The water level in the atomising
tank and jet configuration were excluded from the model,
Table 1. This since no measurable influence was seen on the
particle size for these parameters.
3.2. Influence of Melt Stream Diameter
Additional tests outside the factorial experiments were
performed, to further investigate how the particle size was
influenced by an increased melt stream diameter. This since
a relatively large standard deviation was seen for this param-
eter during the factorial experiments, Fig. 7. Four heats were
Fig. 8. Relation between the d50 value and the relative melt stream
diameter.
atomised using a Fe–1%C melt and DM/D0 ratios between
0.571 and 1, Table 4. The water flow rate was simultaneous-
ly increased from 2.22 to 6.67 kg s–1, which maintained the
water to metal ratios between 3.7–4.5. Figure 8 shows the
relation between the d50 value and DM/D0 ratio for these
experiments. Similar results were obtained as during the fac-
torial experiments. The particle size was found to increase
with an increased relative melt stream diameter. The relation-
ship between the d50 value and the DM/D0 ratio in Fig. 8 was
also estimated with a regression analysis. An R2 value of 0.94
was obtained for this expression, which is given by Eq. (5):
0.26
⎛D ⎞
d50 = 90.4 ⎜ M ⎟ ......................... (5)
⎝ D0 ⎠
3.3. Estimation of Viscosity and Surface Tension
Estimations of the surface tension (γM) and viscosity (μM)
of the liquid metal for all experimental heats are presented
in Tables 2–4. These properties were calculated at 1 550°C,
which is close to the average steel temperature of 1 560°C
measured in the tundish during the tests. The surface tension
was calculated using models developed by Chung and
Cramb.11) For %C > 0.9, the surface tension was calculated
as follows:
γ Fe −C − S = 1 913 + 0.43 × [1 823 − T ] + 67.75 × [%C ]
− 0.107 × T ln [1 + K S aS ] − 0.1
153 × T ln [1 + KO aO ]
.......................................... (6)
log K S =
10 013
− 2.87 ....................... (7)
T

© 2012 ISIJ 2134

 ISIJ International, Vol. 52 (2012), No. 12
11 370
log KO = − 4.09 ....................... (8)
T with the results from Lee and Morita17) for the Fe–C–S sys-
where KS, KO, aS and aO are adsorption coefficients and
activities for sulphur and oxygen respectively. The oxygen
activity was calculated by assuming equilibrium with car-
bon monoxide. First order interaction coefficients and ther-
modynamic data were taken from Engh12) and Guthrie13)
respectively. The surface tension at %C < 0.2 was finally
estimated by using a model developed for the system Fe–S
at 1 550°C:11)
γ Fe − S = 1 913 − 195 ln [1 + 365aS ] ............... (9)
Reference data for the viscosity were selected based on
a literature review performed by Assael et al.14) Viscosity
values were estimated from studies of the Fe–C system by
Barfield and Kitchener15) and Vatolin, Vostryakov and
Yesin,16) since experimental investigations are missing in the
literature for the system Fe–C–S. Experimental data at
1 550°C from these studies were finally used to calculate
average viscosity values for each experiment.
4. Discussion
4.1. Influence of Surface Tension and Viscosity
Additional evaluations of the surface tension and viscos-
ity were made, to try to explain why the particle size
decreased at increased carbon and sulphur contents, Fig. 7.
Estimations of the surface tension at 1 550°C using Eqs. (6)
and (9) are presented in Fig. 9. The calculations show that
sulphur is a strong surfactant, which leads to a lowered sur-
face tension value. Furthermore, the predictions indicate
Fig. 9. Calculated surface tension at 1 550°C based on Eqs. (6) and
(9), for liquid iron alloyed with sulphur and carbon.
Fig. 10. Estimated viscosity values at 1 550°C versus carbon con-
tent for all test heats in the present study.
2135
that carbon has a much weaker influence on the surface ten-
sion compared to sulphur. These calculations agree well
tem at 1 350–1 550°C. The surface tension for liquid iron
alloyed with 0–4%C was for the present conditions almost
exclusively determined by sulphur, at contents above
0.005%S.
Figure 10 shows calculated viscosity values for all test
heats in this work. Increased carbon contents at tempera-
tures of 1 550°C are estimated to decrease the viscosity for
the system Fe–C. In addition, it can be concluded that there
is no consistent data published for the system Fe–S at
1 550°C. Some investigations report that sulphur decreases
the viscosity for liquid iron, while the opposite is concluded
by other authors.18)
However, Figs. 9 and 10 indicate that the reduced d50 val-
ue observed in Fig. 7 at increased sulphur and carbon contents
may be caused by a decreased surface tension and viscosity
respectively. These properties was further analysed in Fig.
11, which shows the d50 value versus the water pressure for
all test heats in this work. The large influence of the water
pressure on the d50 value is obvious. It can also be noticed
that the particle size generally decreased for heats atomised
at low surface tensions. In addition, the results indicate that
the d50 value was further reduced for experiments with low
estimated viscosity values, see also Tables 1–4.
The influence of the viscosity and surface tension during
the atomisation was finally evaluated by comparing the rela-
tion between the viscous energy dissipation and surface
energy for a single atomised droplet. The relative impor-
tance of viscous and surface tension forces in resisting
atomisation was estimated by calculating the ratio between
the liquid phase Weber and Reynolds numbers according to
Eq. (10)19)
Eμ WeL μ M ΔU
≈ = ....................... (10)
Eγ ReL γM
where Eμ is the viscous energy dissipation, Eγ is the surface
energy and ΔU is the typical velocity difference across the
zone during break up. The parameter ΔU was estimated
using Eq. (11),20) which can be used to calculate the water
velocity at the exit of the water nozzles
1/ 2
⎛ 2p ⎞
vW = φ ⎜ W ⎟ ...........................(11)
⎝ ρW ⎠
where φ is a constant that considers deviations from an ideal
Fig. 11. Influence of water pressure, surface tension and viscosity
on the d50 value, for tests 1–16.
© 2012 ISIJ
 ISIJ International, Vol. 52 (2012), No. 12
frictionless flow and ρw is the water density.
The coefficient φ was estimated to have a value of 0.86
for the nozzles in the current study, based on water flow
rates and pressure values from the nozzle supplier. The
water temperature was assumed to be 20°C in the calcula-
tions, which gives a water density of 0.998 g cm–3. Data
from all experiments in this work (Tables 1–4) were finally
used to estimate the ratio between the viscous energy dissi-
pation and the surface energy according to Eq. (13). A value
of WeL/ReL = 1 would mean that the surface tension and vis-
cous forces are of similar importance during the atomisation
process. WeL/ReL was estimated to have values between 0.3
and 1.2 for the tests in the present study, Tables 2–4. This
indicates that both the viscosity and the surface tension of
the liquid metal may have influenced the atomisation of the
liquid metal in this work.
4.2. Influence of Superheat
The influence of the liquid metal properties can be dis-
cussed further by evaluating how the superheat affects the
particle size. As mentioned earlier, it was estimated using
Eq. (4) that the d50 value decreased by 28% when the carbon
content in the liquid steel increased from 0.05 to 4%. This
increased carbon content also corresponds to a raised super-
heat from around 90 to 430°C, Tables 2–4. Summarized, this
corresponds to a reduction of the d50 value by 8% per 100°C
of superheat. This is in agreement with previous investiga-
tions by Ankus et al.3) and Dunkley et al.,6) who noticed a
reduction of the d50 value by 7–9% and 5–10% respectively
per 100°C of superheat.
The calculated influence of the superheat on the particle
size in this work may be explained by a decreased viscosity
at increased carbon contents, Fig. 10. However, a second
possible mechanism could be that the total available time for
atomisation increases at increased superheats. This can be
illustrated by evaluating the cooling time down to the liqui-
dus temperature for an atomised steel droplet. The cooling
time tC can be estimated with Eq. (12):21)
ρ M cM (T − TSteam )
tC = d d ln ................ (12)
6h (Tm − TSteam )
where dd is the metal droplet diameter in m, ρM is the liquid
metal density in kg m–3, cM is the specific heat for the liquid
metal in J kg–1 °K–1, h is the heat transfer coefficient between
the metal droplet and the surrounding atmosphere in W m–2
Fig. 12. Calculated cooling time for liquid metal droplets from the
tundish temperature down to the liquidus temperature,
tests 2 and 4.
© 2012 ISIJ
°K–1, T is the temperature of the liquid metal in °K, and
Tsteam is the steam temperature in the atomizing zone in °K.
The cooling time was calculated for test 2 and 4 which
were atomised at carbon contents of 0.22 and 4.22% respec-
tively, Tables 1 and 2. The steel temperature T was estimated
with the measured temperature in the tundish. Values for h
were taken from a study by Sanin and Nichiporenko.22) The
heat transfer coefficient at 18 MPa was estimated to decrease
from 8.9×105 to 1.4×105 W m–2 °K–1, when the metal droplet
size increased from 10 to 500 μ m. Values for ρM and cM were
equal to 7 030 kg m–3 and 717 J kg–1 °K–1 respectively.13) The
steam temperature was estimated to 900°C.21)
Figure 12 shows calculated cooling times at 5 different
metal droplet diameters. The cooling time increased with an
increased metal droplet size. It was determined to be 17
times longer at a 4.22%C content compared to a 0.22%C
content. The rounder particle shape for test 4 compared to
test 2 further indicated that the cooling time was significantly
longer at high carbon contents, Figs. 3 and 4. It can be spec-
ulated that the available time for atomisation was increased at
higher carbon contents in the present study, since the superheat
was raised at increased carbon contents. This may improve
the secondary atomisation of primary atomised ligaments,
which could contribute to a reduction of the particle size.
4.3. d50 Model
The experimental results in this work were further evalu-
ated with an empirical d50 model proposed in a previous
work.23) This model is presented in Eq. (13), with calculated
exponents relevant for the present study
−0.39
γ 0.45 μ 0.90 D 0.20 ⎛ m ⎞
d50 = k M 0.72M M ⎜ W ⎟ ............. (13)
pW sin α ⎝ mM ⎠
where k is a model constant, α is the jet angle between the
axis of the water jet and axis of the metal stream, DM is the
melt stream diameter in m, mW and mM is the mass flow rate
of water and metal respectively in kg s–1.
All experimental heats in this work were used to deter-
mine the k value and exponents in Eq. (13), Tables 1–4. The
parameter units defined in ‘Nomenclature’ were used to cal-
culate the k value. The least-square method was used to min-
imize the error between experimental and calculated d50
values. The k value in Eq. (13) for this study was estimated
to 1.85×109. The superheat was not directly considered in
this model. The influence of liquid metal properties was
Fig. 13. Experimental data from the present study versus calcu-
lated d50 values for Eq. (13).
2136
 ISIJ International, Vol. 52 (2012), No. 12
Table 5. Calculated exponents for Eq. (13) compared with previous investigations, see “Nomenclature” for parameter
definitions.
Model I II III
Year 1972 1986 1993
26) 6)
Ref. Kishidaka Dunkley Ternovoi
pW - –0.7 to –1.2 - –0.80
DM 0.21 - -
mW/mM –0.04 - -
γM 0.22 - –0.15
μM 0.57 - 0.35
instead modelled by the surface tension and viscosity. Fig-
ure 13 shows calculated d50 values with Eq. (13) versus
experimental data for the present study. The correlation is
very good with an R2 value of 0.972.
A comparison between Eq. (13) and previous d50 models
are presented in Table 5. A negative exponent value means
here that the particle size is reduced when the specific
parameter is increased. The exponents for the water pressure
for Eq. (13), models IV and V are within the interval of –0.7
to –1.2 proposed for flat water jets by Dunkley and Palmer.6)
In addition, it is interesting to notice that the calculated
exponent for the melt stream diameter in this work is very
close to the value suggested in model I.
Moreover, the influence of the water to metal ratio (mW/mM)
on the d50 value was estimated to be significantly larger in
this work compared to all other d50 models, Table 5. The
present study included several tests atomised at water to
metal ratios below 4, Tables 2–4. This may explain the more
negative exponent value for this work in Table 5, since ear-
lier investigations6,27) have observed that the d50 value can
increase drastically at mW/mM ratios below 3–4.
In general, all models in Table 5 propose that the d50 value
decreases with a decreased surface tension and viscosity of
the liquid metal. This is in agreement with the results for the
current study. Moreover, the exponent for the viscosity in
this work is close to the values proposed in models IV and
V. However, in general the exponents for the surface tension
and viscosity have a relatively large scatter in the different
studies. One possible explanation to this may be that it is
difficult to measure both the viscosity and surface tension,
which has resulted in a large scatter between published data
for liquid metals.14,28)
The extremely oxidising gas atmosphere during water
atomisation may be another source of uncertainty. Both the
surface tension and viscosity of liquid metals are generally
measured under very controlled conditions, typically in
argon17) and helium18) gas atmospheres respectively. These
conditions are very different from the gas atmosphere during
water atomisation, where the liquid metal stream and atom-
ised particles to a large extent are surrounded with hot water
steam. Oxygen is, for example, a very strong surfactant in
liquid iron.28) More specifically, only 100 ppm dissolved
oxygen is estimated to decrease the surface tension by 14%
at 1 550°C. It could therefore be possible that the local sur-
face oxidation of the melt during water atomisation may
change the physical properties of the liquid metal. This may
contribute to the relatively large scatter seen for the viscos-
ity and surface tension exponents in Table 5.
2137
IV V Eq. (13)
1999 2012 2012
24,25) 8) 23)
Bergquist Persson This work
–0.98 –0.72
- 0.21 0.20
–0.04 –0.04 –0.39
0.80 0.80 0.45
1.00 0.93 0.90
Fig. 14. Calculated water pressure with Eq. (13) versus surface
tension and viscosity, for atomisation to a d50 value of 80
μ m, DM/D0 = 0.714, mW/mM = 5, α = 26.5°, μM = 6.0 mPa s
(increased surface tension), γM = 1 650 mN m–1 (increased
viscosity).
However, the results in the present study show that the
liquid metal properties can have a significant influence on
the particle size for water atomised powders. Moreover, it is
shown that empirical models can be useful tools for predic-
tion of the d50 value during the atomisation. This was finally
illustrated by estimating the influence of the viscosity and
surface tension during the atomisation of metal alloys with a
target d50 value of 80 μ m. Equation (13) was used to calculate
the required water pressure, at increased values of the viscos-
ity and surface tension, Fig. 14. These calculations show that
it may be necessary to adjust the water pressure significantly
for different alloys, depending on the viscosity and surface
tension values of the liquid metal. The importance of process
control is also worth to comment. Variations of the surface
tension and the viscosity for a specific alloy should be mini-
mized, by carefully controlling the melt temperature and liq-
uid metal composition before the atomisation. This should
reduce the variation of the particle size between atomised
heats and thereby improve the quality of produced powders.
5. Conclusions
The main focus of the present study was the influence of
physical properties of the liquid metal on the particle size
during water atomisation. Carbon and sulphur were alloyed
to liquid iron, to study how the d50 value was influenced by
changed physical properties of the liquid metal. Design of
experiments was used to plan the study, which also included
tests with different water pressures, melt stream diameters,
water levels in the atomising tank and configurations of the
water jet.
© 2012 ISIJ
 ISIJ International, Vol. 52 (2012), No. 12

Alloy additions of 4%C and 0.5%S to liquid iron KS : adsorption coefficient for sulphur
decreased the d50 value with 28 and 31%, respectively. It is mM : mass flow rate of liquid metal/kg s–1
indicated that the reduction of the particle size at increased mW : mass flow rate of water/kg s–1
carbon and sulphur contents may be related to a decreased pW : water pressure/Pa
viscosity and surface tension of the liquid metal, respectively. ReL: Liquid phase Reynolds number
The relation between investigated parameters and the par- T: temperature of liquid metal/°K
ticle size was further analyzed with an empirical d50 model T m: liquidus temperature of metal/°K
proposed in a previous work.23) The influence on the d50 val- Tsteam: water steam temperature in the atomising zone/
ue was estimated as follows for investigated parameters: °K
i) water pressure: large effect, a 44% decrease for pres- vW : velocity of atomising water/m s–1
sures from 8 to 18 MPa. WeL: Liquid phase Weber number
ii) viscosity: medium effect, a 33% decrease for viscosities α: jet angle, between the axis of the water jet and
from 6.8 to 4.3 mPa s. the melt stream/°
iii) surface tension: medium effect, a 27% decrease for sur- γM : surface tension of liquid metal/N m–1
face tensions from 1 850 to 900 mN m–1. μM : viscosity of liquid metal/Pa s
iv) melt stream diameter, small effect, a 10% decrease for ρM : density of liquid metal/kg m–3
relative diameters (DM/D0) from 1 to 0.57. ρW : density of atomising water/kg m–3
The influence of the surface tension and viscosity should σLN : standard deviation of the log normal distribution
be seen as provisional, since there is a relatively large scatter
between experimental data in the literature14,28) REFERENCES
It may also be possible that the reduced particle size at 1) J. Dunkley: Met. Powder Rep., 57 (2002), No. 11, 18.
2) E. Klar: Met. Powder Rep., 40 (1985), No. 1, 7.
increased carbon contents in this work was caused by an 3) A. T. Ankus and R. D. Venter: Powder Technol., 73 (1992), No. 2,
increased available time for atomisation. Alloy additions of 169.
4%C to liquid iron increased the superheat in the furnace 4) Y. Kato: Int. Conf. on Powder Metallurgy for Automotive Components,
Science Publisher, Enfield, USA, (1997), 295.
from around 90 to 430°C. Model calculations indicated that 5) J. J. Dunkley: Powder Metall. Int., 10 (1978), No. 1, 38.
this raised superheat resulted in around 17 times longer 6) J. J. Dunkley and J. D. Palmer: Powder Metall., 29 (1986), No. 4,
287.
cooling times for atomised droplets, from the tundish tem- 7) E. Klar and W. M. Shafer: Int. J. Powder Metall., 11 (1975), No. 4,
peratures down to the liquidus temperature. It can be spec- 241.
ulated that the significantly longer cooling time at large 8) B. Bergquist: Powder Metall., 42 (1999), No. 4, 331.
9) L. Eriksson, E. Johansson, N. Kettaneh-Wold, C. Wikström and S.
superheats may improve the secondary atomisation of pri- Wold: Design of Experiments, Principles and Applications, Umetrics
mary atomised ligaments. This could contribute to a reduc- AB, Umeå, (2008), 75, 417.
10) J. O. Andersson, T. Helander, L. Höglund, P. F. Shi and B. Sundman:
tion of the particle size. Calphad, 26 (2002), 273.
11) Y. Chung and A. W. Cramb: Metall. Trans. B, 31B (2000), 957.
12) T. A. Engh: Principles of Metal Refining, Oxford University Press,
Acknowledgements New York, (1992), 62.
Special thanks are given to H. Gadmar, L. Boo, M. 13) R. I. L. Guthrie: Engineering in Process Metallurgy, Oxford University
Augustsson and M. Olsson for all their help and support dur- Press, New York, (1992), 483
14) M. J. Assael, K. Kakosimos, R. M. Banish, J. Brillo, I. Egry, R.
ing the pilot scale tests. M. Cengiz is also gratefully Brooks, P. N. Quested, K. C. Mills, A. Nagashima, Y. Sato and W.
acknowledged for all his input and help during this study. A. Wakeham: J. Phys. Chem. Ref. Data, 35 (2006), 286.
15) R. N. Barfield and J. A. Kitchener: J. Iron Steel Inst., 180 (1955),
Many thanks are also given to O. Litström, M. Berntsson, 324.
H. Grosser, A. Heinrich and O. Bergman, for their assistance 16) N. V. Vatolin, A. A. Vostryakov and O. A. Yesin: Fizika Met., 15
during the design and evaluation of the experiments. (1963), 222.
17) J. Lee and K. Morita: Steel Res., 73 (2002), No. 9, 367.
18) M. G. Frohberg and T. Cakici: Arch. Eisenhuttenwes., 49 (1978), No.
Nomenclature 5, 229.
19) A. J. Yule and J. J. Dunkley: Atomisation of Melts, Oxford University
AD: apparent density/g cm–3 Press, New York, (1994), 26.
aO: oxygen activity in the steel 20) R. J. Grandzol and J. A. Tallmadge: Int. J. Powder Metall., 11 (1975),
aS: sulphur activity in the steel No. 1, 103.
21) O. S. Nichiporenko: Sov. Powder Metall. Met. Ceram., 15 (1976),
cM: specific heat of liquid metal/J kg–1 °K–1 No. 9, 665.
d50: mass median particle size of as atomised powders/ 22) A. F. Sanin and O. S. Nichiporenko: Sov. Powder Metall. Met.
Ceram., 27 (1988), No. 9, 677.
μm 23) F. Persson, A. Eliasson and P. Jönsson: Powder Metall., 55 (2012),
dd: metal droplet diameter/m No. 1, 45.
24) Y. F. Ternovoi and O. S. Nichiporenko: Powder Metall. Met. Ceram.,
DM/D0: relative melt stream diameter 32 (1993), No. 1, 1.
H/H0: relative water level in atomising tank 25) Y. F. Ternovoi, O. S. Nichiporenko, V. N. Makogan and V. I. Bilan:
h: heat transfer coefficient, metal droplet to atmo- Powder Metall. Met. Ceram., 32 (1993), No. 2, 99.
26) H. Kishidaka: Proc. Multidisciplinary Meeting on Sintered Metals
sphere/W m–2 °K–1 and Magnetic Properties, Japan Society for Powder and Powder
J: water jet configuration/A = –1, B = 1 Metallurgy, Kyoto, (1972), 19.
27) T. H. Wang, W. H. Lee and W. C. Liu: China Steel Tech. Rep., 7
k: model constant (1993), No. 7, 98.
KO: adsorption coefficient for oxygen 28) B. J. Keene: Int. Mater. Rev., 33 (1988), No. 1, 1.

© 2012 ISIJ 2138