Advanced Powder Technology 22 (2011) 294–299

Contents lists available at ScienceDirect

Advanced Powder Technology

journal homepage: www.elsevier.com/locate/apt

Translated paper: SPTJ Young Researchers’ Award 2009

Classification of particles by centrifugal separator and analysis of the fluid behavior q

Tetsuya Yamamoto ⇑, Tomohiko Shinya, Kunihiro Fukui, Hideto Yoshida
Department of Chemical Engineering, Hiroshima University, 1-4-1, Kagamiyama, Higashi-Hiroshima 739-8527, Japan

a r t i c l e i n f o a b s t r a c t

Article history: This study gives fundamental knowledge on the particle classification performance by centrifugal sepa-
Received 13 December 2010 rator.
Accepted 15 December 2010 It is found that the cut size of a centrifugal separator decreases as the rotational speed increases and the
Available online 1 January 2011
liquid flow rate decreases. Fitting our experimental results with the theory, they agree with each other at
high flow rate. However, the difference between them generates at low flow rate. This is because dead
Keywords: spaces are generated in the centrifugal separator at the low flow rate. Also, the computer simulation of
Centrifugation
the fluid behavior in the centrifugal separator can find the decrease of the velocity near the wall under
Particle classification
Particle size
the low flow rate, which suggests the possibility of the formation of dead spaces in the separator.
Computational fluid dynamics Ó 2010 The Society of Powder Technology Japan. Published by Elsevier B.V. and The Society of Powder
Technology Japan. All rights reserved.

1. Introduction As Eq. (1), sedimentation velocity is proportional to the square

of particle diameter, hence, larger particles are collected by the
Recently, to enhance the value of powders in the process of wall of the separator; smaller particles go through the separator.
powder technology, the powders with the smaller size and narrow Because the conventional method can classify particle with the size
size distribution have been needed. Especially, the emergence of from 1 to 800 lm [1], it is necessary to develop the classification
high-value added industrial products using nanoparticles requires technique for the nanoparticles.
development of classification technique for nanoparticles. This paper includes the basic data about the performance of
Wet type of classification system is effective for classification of centrifugal separator and discussion comparing the experimental
the particles which are difficult to be dispersed in the air, and is data with theory and computational fluid dynamics (CFD)
broadly classified into gravitational and centrifugal sedimenta- simulation.
tions. Gravitational classification making use of the difference of
terminal velocity between particles in the slurry is effective for 2. Experimental apparatus and method
classification of particles with the size from 40 to 200 lm [1]. On
the other hand, classification using centrifugal force, such as hy- Fig. 1 illustrates the centrifugal separator system used in the
dro-cyclone [2–4] and centrifugal separator, makes use of the bal- present study. The centrifugal separator was able to be rotated at
ance between centrifugal force and fluid drag. In the present study, extremely high velocity by the motor attached to the top of the
centrifugal separator, which was able to classify particle with the separator. The entire length and inside diameter of the centrifugal
size of much smaller, was used. separator were 302 and 44.4 mm, respectively. The details of the
Centrifugal separator, which has the same classification mecha- separator are shown in Fig. 2(a). Various blades were inserted in
nism as gravitational sedimentation method, is compact and has a the centrifugal separator, the effect of number of blades, n, on clas-
high performance in an operating condition, hence, it is used in the sification performance was investigated. The details of blades are
field of mining and chemical industry. Centrifugal separators can shown in Fig. 2(b).
exert forces up to approximately 50,000g on the fluid and particles Silica spherical particle with the density of 2300 kg/m3 and
that comprise liquid systems to classify particles. The terminal median size of 0.5 lm (SO-C2, Admatechs Co. Ltd.) was used as a
velocity of particle in the centrifugal force field, mt, represents Eq. (1). tested powder. The size distribution of the tested powder was
ðqp  qÞD2p r x2 shown in Fig. 3. This powder was dispersed by ultrasonic homoge-
mt ¼ ð1Þ nizer in the water whose temperature was controlled to be 30 °C
18l
by the heater to make the slurry. The concentration of the slurry
q was 0.5 wt.% in the slurry tank, this original slurry was moved from
Japanese version published in JSPTJ, Vol. 44, No. 5 (2007); English version for
APT received on 13 Dec. 2010. the slurry tank to the bottom of the centrifugal separator by the
⇑ Corresponding author. Tel.: +81 824 24 7853; fax: +81 824 24 5494. pump. Centrifugal separator was operated to classify particles un-
E-mail address: [email protected] (T. Yamamoto). der the conditions of the rotation speed from 6000 to 12,000 rpm

0921-8831/$ - see front matter Ó 2010 The Society of Powder Technology Japan. Published by Elsevier B.V. and The Society of Powder Technology Japan. All rights reserved.
doi:10.1016/j.apt.2010.12.009
 T. Yamamoto et al. / Advanced Powder Technology 22 (2011) 294–299 295

Nomenclature

C dimensionless particle concentration () r, z dimensionless radial and axial coordinates ()
D particle diffusion coefficient (m2/s) V inner volume of the centrifugal separator (m3)
Dp particle diameter (lm) mg terminal velocity in gravitational force field (m/s)
Dp95 95% cut size of centrifugal separator (lm) mt terminal velocity in centrifugal force field (m/s)
fo(Dp),ff(Dp) particle size distributions of feed and classified fine uz,ur,uh dimensionless axial, radial and tangential fluid veloci-
sides, respectively (/lm) ties ()
g gravity acceleration (m/s2) mz, mr, mh dimensionless axial, radial and tangential particle veloc-
mo, mf mass of the collected particles for feed and classified ities ()
fine sides, respectively (kg) Dg partial separation efficiency ()
n number of blades () h circumferential coordinate ()
p dimensionless pressure () l fluid viscosity (Pas)
Q liquid flow rate (ml/min) q,qp fluid and particle density, respectively (kg/m3)
Re fluid Reynolds number () x rotational speed of centrifugal separator (rpm)
T time

and flow rate of the slurry from 85 to 1000 ml/min. After twice the
residence time passed supplying slurry to the separator, the slurry
with fine particles was sampled from the top of the separator. The
weights and size distributions of fine and original particles were
measured, and partial separation efficiency, Dg, as defined by Eq.
(2) was calculated. Partial separation efficiency curve was drawn
with x-axis of particle size and y-axis of partial separation effi-
ciency, Dg, to evaluate the classification performance of the cen-
trifugal separator. In Eq. (2), mo and mf are the weight of original
and fine powders, respectively, fo(Dp) and ff(Dp) represent volumet-
ric size distributions of original and fine particles, respectively. Vol-
umetric particle size distributions of the original and classified fine
slurries were measured using laser diffraction/scattering particle
size distribution analyzer (LA-950, HORIBA Co. Ltd.). Because the
coarse particles were deposited in the separator, it was impossible
to operate this system continuously without cleaning.
mo fo ðDp ÞDDp  mf ff ðDp ÞDDp
Dg ¼ ð2Þ
mo fo ðDp ÞDDp

3. Results and discussion

3.1. Classification experiments and the theory

3.1.1. Results by classification experiment

Fig. 4 shows the influence of the number of blades inserted into
the separator, n, on classification performance under the condi-

Fig. 2. Schematic diagrams of centrifugal separator and blades.

Fig. 1. Centrifugal separator system. Fig. 3. Particle size distribution of the tested silica particles.
296 T. Yamamoto et al. / Advanced Powder Technology 22 (2011) 294–299

Fig. 6. Influence of flow rate at x = 6000 rpm on partial classification efficiency

Fig. 4. Influence of the number of blades inserted into the separator on classifi-
curve.
cation performance under the conditions of x = 6000 rpm and Q = 300 ml/min.
3.1.2. Comparison of experimental result with theory
When the particle size on the partial separation efficiency
tions of x = 6000 rpm and Q = 300 ml/min. At n = 0, the classifica-
curve, where Dg was equal to 0.95, was defined as the 95% cut size
tion experiment was carried out without blade. Fig. 4 indicated
of the centrifugal separator, Dp95, the relationships between Q and
that it was impossible to classify particles without blade. That
Dp95 at rotation speed of 6000, 9000 and 12,000 rpm are shown in
was to say, the blade was necessary for classification of particles
Fig. 7. The particle size at Dg = 1.0 on the partial separation
because it enabled the fluid in the separator to rotate either almost
rigidly or rigidly. Generally, it is predicted that classification accu-
racy is enhanced with the increase of n because the fluid is ex-
pected to be rotated strongly by increase of n. However, good
classification performance was not always obtained increasing
the number of blades probably because frictional loss of the fluid
became larger. Fig. 4 demonstrated that classification performance
using the blade of n = 4 was better than that of others, hence, the
forward experiments was carried out using the blade of n = 4.
Fig. 5 shows the influence of rotation speed on classification
performance at Q = 300 ml/min. Because particles were subjected
to be a stronger centrifugal force field increasing rotation speed
and smaller particles were easy to be collected by the wall of the
separator, particle separation efficiency curve shifted to smaller re-
gion. Fig. 6 shows the effect of flow rate at x = 6000 rpm on partial
classification efficiency curve. Decrease of flow rate made the res-
idence time of particle in the separator longer. Hence, particles
were subjected to a centrifugal force field for longer time and eas-
ily collected by the wall, and partial separation efficiency curve
moved to the smaller size. These results demonstrated that partial
separation efficiency curve shifted to smaller when the centrifugal
separator was operated under the conditions of stronger force gi-
ven to the particles and longer time for particle’s being in the sep-
arator, which was understood intuitively.

Fig. 5. Influence of rotation speed on classification performance at Q = 300 ml/min. Fig. 7. Comparison between experimental data of Dp95 with theoretical equation.
 T. Yamamoto et al. / Advanced Powder Technology 22 (2011) 294–299 297

efficiency curve was used generally to evaluate the classification

performance, however, Dp95 was utilized to enhance the accuracy
of evaluation of partial separation efficiency curve in this paper.
To compare experimental results with theory, the theoretical
equation was derived using the schematic representation illus-
trated in Fig. 8. At z = 0, r-coordinate of the particles nearest of
the central axis and the minimum value of r-coordinate of the col-
lected particles were defined as r1 and r0 , respectively. Also, r-coor-
dinate of wall of the separator was defined as r2. It was assumed
that particles moved in the direction of r at constant terminal
velocity, mt, and in the direction of z at constant average velocity
of QL/V, which was linear velocity obtained from dividing the total
flow rate by the average cross-sectional area of the separator. As a
result, Eq. (3) was derived, furthermore, Eq. (4), where mg was ter-
minal velocity of particle in the gravitational field defined as Eq.
(5), was obtained integrating of Eq. (3) from (r0 ,0) to (r2,L) on the
assumption that the collected particles moved on the dotted line
in Fig. 8. Because the centrifugal separator collected the particles
in the region between r0 and r2 at z = 0 of the particles existing be-
tween r1 and r2 at z = 0, partial separation efficiency was expressed
as Eq. (6). Using Eq. (4), Eq. (6) was rewritten as Eq. (7) [5]. r1/r2 in
Eq. (7) was regarded as fitting parameter and the value of it was
selected for Eq. (7) to be fitted with the experimental data shown
in Fig. 7. The solid lines drawn in Fig. 7 represent fitting lines.

dr mt V
¼ ð3Þ
dz QL
 
r 0
V x2
Fig. 8. Schematic representation of centrifugal separator for the derivation of
¼ exp  mg ð4Þ
r2 Qg
theoretical calculation.

Fig. 9. Photos of the sheets with copper particles after classification experiment under the various experimental conditions.
298 T. Yamamoto et al. / Advanced Powder Technology 22 (2011) 294–299

ðqp  qÞD2p g
mg ¼ ð5Þ
18l
pðr22  r02 Þ
Dg ¼ ð6Þ
pðr22  r21 Þ
" (  2 )#
9Q l r1
D2p ¼  ln 1  D g 1  ð7Þ
V x2 ðqp  qÞ r2

It was clear that the fitting parameter as shown in Fig. 7, r1/r2,

decreased with the increase of x. This was valid result because
particles were easy to be near the wall of the separator by stronger
centrifugal force with rotation speed increasing. At Q 6 200 ml/
min and x = 6000 rpm, however, the difference between experi-
mental data and fitting curve was generated, as indicated by
Fig. 7(a). Similarly, under the conditions of lower flow rate at
x = 9000 and 12,000 rpm, the difference between experimental
data and fitting curve was found.

3.2. Visualization of the fluid behavior in the centrifugal separator

It was found that the difference between experimental data and

fitting curve according to Eq. (7) was larger at lower flow rate in
the Section of 3.1.2. To clarify the reason, the fluid behavior in
the separator was visualized and analyzed by experiments and
CFD simulation.

3.2.1. Fluid behavior investigated by experiment

Sheet with a thickness of 0.35 mm, as shown in Fig. 9(a), was
prepared between the blade and wall of the separator. Particles
were deposited on the sheet after the classification experiment,
hence, observation of the positions of the deposited particles on
the sheet were able to predict the fluid behavior of the slurry
in the separator. Silica particle used as tested powder in the pres-
Fig. 10. Schematic representations of the dead space in the centrifugal separator
ent experiment, however, showed the poor adhesive on the sheet under the low flow rate.
not to stay at the position collected by the sheet. Hence, copper
particle slurry with the density of 9500 kg/m3 and the concentra-
tion of 1.0 wt.%, was supplied to make the collected particles ad- 300 ml, the sample was obtained from the top of the separator. Be-
here to the sheet strongly. Fig. 9(b) and (c) show photos of the cause the volume of the dead spaces was difference between the
sheets after classification experiments under the conditions of supplied amount of the slurry and the inner volume of the separa-
(x, Q) = (6000, 360) and (6000, 780), respectively. It was found tor, the dead space with the volume of 35 ml was generated in the
that almost all areas of the sheets were covered with the copper separator. Because of the dead spaces, the theoretical equation was
particles. Similarly, at the condition of (x, Q) = (12,000, 780), the not fitted with the experimental data at flow rate near Q = 360 ml/
copper particles deposited on the sheet uniformly as indicated in min in Fig. 7(c). The influence of flow rate at x = 12,000 rpm on
Fig. 9(e). On the other hand, Fig. 9(d) showed that some areas of formation of dead space in the separator was investigated experi-
the sheet were not covered with cooper particles at the condition mentally in this section. Similarly, it was possible to think that
of (x, Q) = (12,000, 360). The sheet was divided into four parts by dead spaces also were generated in the separator at the flow rates,
the blade of n = 4, the three parts of them, which were indicated where experimental data and fitting curve did not agree with each
by the arrows in Fig. 9(d), clearly showed the region without par- other in the case of x = 6000 and 9000 rpm. Under the condition
ticle deposition. Because cooper particles did not cover all areas where such a dead space was not generated, namely, all areas of
of wall of the centrifugal separator at (x, Q) = (12,000, 360), it the wall of the centrifugal separator were covered with samples,
was possible to think that fine slurry was obtained from the experimental data agreed well with fitting curve. From the above
top of the separator with dead spaces. Fig. 10 show the imaginary discussion, at low flow rate, experimental results are approximated
illustrations of dead space in the centrifugal separator. Fig. 10(a) to the fitting curve and to make the cut size smaller if inner struc-
shows three dimensional illustration of dead space in the space ture of the separator is improved to prevent the formation of dead
surrounded by the blades and wall of the separator and space.
Fig. 10(b) shows two dimensional image of the separator with
dead space at a h. 3.2.2. Fluid behavior investigated by CFD simulation
To confirm the formation of the dead spaces in the separator, To investigate the experimental results in Section of 3.2.1, the
mass balance between the total amount of the supplied slurry fluid behavior in the separator was analyzed by CFD simulation.
and the inner volume of the separator was investigated. Under As indicated by Fig. 9(d), the flow in the separator including such
the experimental condition of (x, Q) = (12,000, 360), where dead a dead space was not regarded as the two dimensional axisym-
spaces would be generated in the separator, the time between ar- metric flow. In the present study, however, direct calculation
rival of the sample at the inlet of the separator and sample’s reach- assuming the flow was two dimensional axisymmetric was se-
ing at the outlet was found to be 50.1 s. That was, when the total lected for the sake of simplicity. To calculate the velocity distri-
amount of the sample slurry supplied to the separator reached butions in the centrifugal separator, continuous equation,
 T. Yamamoto et al. / Advanced Powder Technology 22 (2011) 294–299 299

Fig. 11. Velocity distributions by numerical simulation under the conditions of Q = 100, 200, and 330 ml/min at x = 12,000 rpm.

Navier–Stokes equations of z and r, diffusion of particle equation 4. Conclusions

[6] as shown in Eqs. (8)–(11) were discretized and calculated
using control volume method [7]. To calculate the field of pres- Classification performance using the centrifugal separator was
sure, SIMPLE (Semi-Implicit Method for Pressure-Linked Equa- investigated by experiments and theory, and visualization experi-
tion) was used. The numbers of grid partitioning on z and r ment and CFD simulation were carried out to clarify the fluid
axis were 53 and 51, respectively. Fluid velocities, uz, ur, on the behavior in the separator. The following conclusions were
wall were assumed to be zero. To give the fluid in the separator obtained:
rigid rotation, uh was rx. Calculation was carried out at the re-
gion of 100 6 Re 6 500. (1) Setting flow rate and rotation speed low and high, respec-
tively, gave particles stronger centrifugal force to make the
separation size smaller.
@rur @ruz
þ ¼0 ð8Þ (2) The experimental data and theory fitted well at high flow
@r @z
  "   # rate. On the other hand, they did not agree with each other
@uz @uz @uz @p 1 @ @uz @ 2 uz at low flow rate probably because of dead spaces generated
q þ ur þ uz ¼ þl r þ 2 þ qg
@t @r @z @z r @r @r @z easily in the centrifugal separator.
(3) CFD simulation of the fluid behavior in the centrifugal sepa-
ð9Þ rator clarified that the stronger velocity field moved from
  "   #
@ur @ur u2h @ur @p @ 1 @rur @ 2 ur the bottom to the top of the wall of the centrifugal separator
q þ ur  þ uz ¼ þl þ 2 as the flow rate increased. This result agreed with experi-
@t @r r @z @r @r r @r @z
mental results qualitatively about the generation of dead
ð10Þ spaces in the separator.
    
@rC @rC v r @rC v z @ @C @ @C
þ þ ¼D r þ r ð11Þ References
@t @r @z @r @r @z @z
[1] The Society of Powder Technology, Japan, Huntai Kogaku Binran, second ed.,
Fig. 11 shows the velocity distributions by numerical simulation Nikkan Kogyo, 1998, p. 318.
under the conditions of Q = 100, 200, and 330 ml/min at [2] H. Yoshida, K. Fukui, Y. Isshiki, Control of particle size separation by use of
hydrocyclon, J. Soc. Powder Technol. Japan 34 (1997) 690–696.
x = 12,000 rpm. From these results, it was found that the magni- [3] H. Yoshida, S. Akiyama, K. Fukui, A. Kumagaya, Particle classification with
tude of velocity near the wall of the separator increased as the inlet improved hydro-cyclone separator, J. Soc. Powder Technol. Japan 38 (2001)
flow rate, Q, increased. Fig. 11(a) indicated that strong velocity field 626–632.
[4] Y. Norimoto, K. Fukui, H. Yoshida, Particle classification performance of hydro-
was at the bottom of the separator and the magnitude of velocity
cyclone with forced-vortex type, J. Soc. Powder Technol. Japan 43 (2006) 666–
near the wall becomes smaller, which suggested the possibility 675.
of the formation of dead spaces at the wall near the top of the sep- [5] H. Masuda, K. Higashitani, H. Yoshida, Powder Technology Handbook, CRC Press,
2006.
arator at low flow rate. Hence, experimental data agreed well with
[6] K. Takahashi, Kiso Earozoru Kogaku, Yokendo, 1982. pp. 41–42.
the result by numerical calculation about the prediction of dead [7] S.V. Patankar, Numerical heat transfer and fluid flow, Hemisphere Pub, 1980. pp.
spaces in the separator. 113–137.