See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/222206529

Flowpattern in Reverse-Flow Centrifugal Separators

Article  in  Powder Technology · November 2002

DOI: 10.1016/S0032-5910(02)00148-1

CITATIONS READS

70 1,056

7 authors, including:

Huub W Dries
Shell Global
51 PUBLICATIONS   310 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Catalytic Cracking View project

fcc powder downflow in diplegs and standpipes View project

All content following this page was uploaded by Huub W Dries on 13 November 2017.

The user has requested enhancement of the downloaded file.

 Powder Technology 127 (2002) 212 – 222
www.elsevier.com/locate/powtec

Flow pattern in reverse-flow centrifugal separators

W. Peng a, A.C. Hoffmann b,*, P.J.A.J. Boot c, A. Udding c, H.W.A. Dries d, A. Ekker d, J. Kater d
a
Department of Engineering Physics, Tsinghua University, Beijing, 100084, China
b
Department of Physics, University of Bergen, Allegt. 55, 5007 Bergen, Norway
c
Department of Chemical Engineering, University of Groningen, Nijenborgh 4, 9747, AG, Groningen, The Netherlands
d
Shell Global Solutions, Shell Amsterdam P.O. Box 38000, 1030 BN Amsterdam, The Netherlands
Received 1 June 2002; received in revised form 1 July 2002; accepted 24 July 2002

Abstract

Experimental flow patterns, determined by Laser Doppler Anemometry (LDA) for two types of reverse-flow centrifugal separators, are
presented. The flow patterns in (a) a conventional cylinder-on-cone cyclone with tangential inlet and (b) a swirl tube with vane-generated
swirl and a cylindrical body are compared. The experimental data are also used to test the validity of the flow assumptions of some widely
used cyclone separation models for each of the two devices. The measured flow patterns are also compared with the results of computational
fluid dynamics (CFD) simulations wherever this was helpful for the interpretation. Although the data globally support the standard flow
assumptions, some features were quite surprising; for instance, it appears that the surface delimiting the central region of upward flow is
largely determined by the diameter of the body and not that of the vortex finder. The similarities and differences between the flow patterns in
cyclones and swirl tubes are discussed, as is the suitability of using cyclone separation models for the prediction of swirl tube performance.
D 2002 Elsevier Science B.V. All rights reserved.

Keywords: Flow pattern; reverse flow; centrifugal separators

1. Introduction flow assumptions made in most cyclone performance mod-

els and the reliability of CFD simulations.
Centrifugal separators, both conventional cylinder-on- The present paper sheds light on these issues, not only
cone cyclones and cylindrical swirl tubes with swirl vanes, for conventional cylinder-on-cone cyclones with tangential
are playing a dominating role in dedusting and demisting in inlets, but also notably for cylindrical swirl tubes with swirl
industry. Efficient, reliable separation is needed to meet ever vanes. The latter device is widely used, often with many
more stringent environmental requirements. Meeting these tubes operating in parallel arranged in so-called ‘swirl
new emission limits using improved cyclone technology is decks’, for instance, for dedusting after the FCC process.
superior in terms of process operation and process econom- In spite of their wide use, neither dedicated performance
ics to using more expensive and cumbersome dedusting models nor detailed information about the flow, which
alternatives. might point to the best model assumptions, has been
Since the emergence of signal analysis also suitable for published.
high velocities, Laser Doppler Anemometry (LDA) has The flow pattern in reverse-flow centrifugal separators
proven a useful tool to study the swirling flow pattern in can roughly be visualized as sketched in Fig. 1. In a
centrifugal dedusting devices without having to insert conventional cyclone, the swirling motion in the separation
probes, which may disturb the swirl. Some new data are space is generated by leading the gas in tangentially. In the
therefore emerging about the flow pattern in reverse-flow outer part of the body, the axial flow is downward. As the
dedusters, but many questions remain unanswered, in par- gas swirls down, it also flows inward into the central core,
ticular, questions about the appropriateness of the sweeping where the axial velocity is upward. The gas leaves through
the vortex finder in the cyclone roof. In the swirl tube, the
gas is injected axially, and the swirl is generated by swirl
* Corresponding author. Tel.: +47-55582876; fax: +47-55589470. vanes. The flow pattern in the separation space is qualita-
E-mail address: [email protected] (A.C. Hoffmann). tively similar to that in the conventional cyclone.

0032-5910/02/$ - see front matter D 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 0 3 2 - 5 9 1 0 ( 0 2 ) 0 0 1 4 8 - 1
 W. Peng et al. / Powder Technology 127 (2002) 212–222 213

globally more successful for modelling various types of

cyclones than the model approaches mentioned above.
Turning from modelling to experimental investigations of
the flow pattern in cyclones. These have historically been
made in four ways:

(a) flow visualization by introducing a tracer in the form of

a very fine dust or smoke into the air stream entering the
cyclone, or at points within the cyclone,
(b) by means of streamers suspended in the cyclone,
(c) by probe measurements of the local velocity, either with
Pitot tubes or hot-wire anemometers, and
Fig. 1. Sketches of a cylinder-on-cone cyclone with a tangential inlet and a
cylindrical swirl tube with swirl vanes, with the flow pattern indicated.
(d) by Laser Doppler Anemometry (LDA).

Experimental studies go back to the 1930s. Often, Pitot

tubes have been used, and workers have found that insertion
Thus, the flow is often viewed as consisting of an outer, of the probe often causes a reduction in the cyclone pressure
downward-spiralling vortex and an inner, upward-spiralling drop. This reduction is likely to be caused by a reduction in
one. The dust is separated from the gas by being centrifuged the swirl velocity in the cyclone due to the insertion of the
to the wall, along which it is transported to the dust outlet by probe. Leith [4] reviews experimental investigations of the
the downwardly directed flow along the wall. flow in cyclones.
A number of predictive models for the separation effi- Patterson and Munz (1996) [5] in a recent study used a
ciency of cyclones have been proposed in the literature. The five-hole Pitot tube to detect the radial profiles of the gas
models can be divided into two main groups. velocity in a cylinder-on-cone cyclone. They also drew
One class, originating from the work of Barth [1], is conclusions about the flow based on observing the striation
based on a force balance on a particle swirling on the pattern of a dust on the cyclone wall. Wakelin [6] built a hot
interface between the inner and outer vortex. Inward drag film anemometer and lineariser into a cyclone. Wakelin did
is proportional to the particle diameter (in Stokes region), not resolve the total velocity he obtained into components,
while outward ‘‘centrifugal force’’ is proportional to the as he could not obtain the yaw and pitch angles accurately.
cube of the diameter. Large particles are therefore separated, Yaodong et al. [7] also measured the velocity distribution in
while small particles are dragged into the inner vortex and a swirl tube with a Pitot tube. Garwood et al. [8] used a two-
lost. The models of Barth and those building on it (e.g. Refs. component, two-colour fiber optic LDA system with an
[2,3]) all make similar assumptions concerning the flow in Argon ion laser to measure the tangential and axial veloc-
the cyclone: axially constant tangential and radial velocities ities in a cyclone.
in the surface separating the inner and outer vortices. This A literature survey in the framework of a large project
surface is assumed to coincide with the cylindrical surface aimed at improving cyclone and swirl tube modelling and
obtained when prolonging the vortex finder wall to the performance concluded that very little has been done to
bottom of the cyclone or swirl tube. chart the velocity distributions in cylindrical swirl tubes
The other class of model considers the time it takes a with axial inlets, and that the information about the flow
particle, injected into the cyclone at a certain radial pattern in conventional cyclones remains patchy.
position, to reach the cyclone wall as it spirals downward. Faced with insufficient information about the flow pat-
In this approach, it is assumed that there is little inward tern, particularly for swirl tubes, the object of the present
velocity in the separation space except in the bottom, work was:
where a strong inward velocity is assumed to sweep the
particles that did not manage to reach the cyclone wall into – Map the velocity fields in cyclones and swirl tubes with
the inner vortex. LDA, and make a qualitative comparison of the two;
Another technique for predicting the flow field and the – Use the results enhanced by CFD simulations to assess
separation efficiency of cyclone is by computational fluid the validity of the flow assumptions made in cyclone
dynamics (CFD), which has gained enormous popularity. The separation models;
cyclone separation efficiency is mostly predicted by Lagran- – Use the results to identify the best modeling approach for
gian particle tracking in a precalculated flow field. CFD can swirl tubes.
be very helpful for detailed investigation of the gas –solid
flow pattern inside the separator body. CFD simulations have, The objective of carrying out CFD simulations was to
until now, not been checked extensively by comparison with further shed light on the experimental results wherever
experimental results, and there is as yet no convincing appropriate, rather than to perform detailed comparisons
evidence that one particular CFD technique has proved of experiment with CFD simulations.
214 W. Peng et al. / Powder Technology 127 (2002) 212–222

A separate objective was to study the asymmetry of the

flow in the inlet region of tangential inlet cyclones. The
results of this are reported in a separate article [9].

2. Velocity profiles in centrifugal separators

Since the flow is strongly swirling, the tangential veloc-

ity component is more important than the axial and radial
components. The radial component is the smallest one and
can be neglected for basic calculations. Fig. 3. Axial velocity profile in cyclones and swirl tubes.
The tangential velocity profile can be seen as two
regions with an outer region of quasi-free vortex flow
(outside R1 in Fig. 2) surrounding an inner region of The surface of zero axial velocity between the outer and the
quasi-forced vortex flow. This type of vortex is often inner vortices is often assumed to coincide with the cylin-
referred to as a ‘‘Rankine’’-type vortex. The velocity drical surface created when prolonging the vortex finder
profiles in both regions can be expressed with the equation wall to the bottom of the cyclone or swirl tube. In this paper,
[9]: the axial flow is taken as negative when downwardly
C directed, as shown in Fig. 3. The axial velocity often
vh ¼ ð1Þ exhibits a dip near the centerline, which can be so severe
rn
that the flow direction is reversed there. This was discussed
where vh is the tangential velocity, C is a constant, and r in Hoffmann et al. [11].
denotes the radial coordinate with origin in the centerline The radial velocity, vr, is the smallest of the three
of the cyclone. For ideal forced vortex flow (or ‘‘solid components and is directed inward almost everywhere. In
body rotation’’) n =  1, and for ideal free (or ‘‘loss-free’’) conventional cyclones and in cyclone models, vr is often
vortex flow, n = 1. In cyclone modeling, the flow in the assumed constant over the area of zero axial velocity. In
outer region is the interesting part, and this is often swirl tubes, the radial velocity distribution on the surface of
described with Eq. (1) using a value for n somewhat less zero axial velocity has been unknown until now. In the work
than unity. The exact value of n depends on a number of presented in this project, only the tangential and axial gas
factors such as wall roughness, cyclone geometry and velocity components were measured directly, but the radial
particle concentration and normally varies between 0.5 velocity distribution over the surface of zero axial velocity
and 0.8. was determined indirectly and is shown in this paper.
The radius R1, corresponding to the maximum in tangen-
tial velocity, is normally assumed to be less than the radius
of the vortex finder Rx. At the wall and at the centerline, the 3. Experimental procedure
tangential velocity is zero.
The axial velocity is directed downward near the wall 3.1. Test rig
and upward in the region around the axis. The outer region
of downwardly directed flow is sometimes referred to as the The nonintrusive, optical technique of Laser Doppler
‘‘outer vortex’’, and the inner region as the ‘‘inner vortex’’. Anemometry (LDA) was used to measure the mean fluid
velocities in one cyclone and one swirl tube. The rig was
situated at the Shell Research and Technology Center in
Amsterdam (SRTCA). A schematic picture of the rig is
shown in Fig. 4.
In LDA, the velocity of the gas is measured as the speed
of small seed particles, considered to follow the gas flow
faithfully. A laser light beam is split in two, and the two
parts are made to cross, which creates an interference pattern
(the measuring volume) through which the seed particles fly.
The light scattered by the particles is detected as a ‘‘doppler
burst’’, the frequency of which is a measure of the speed
with which the particles traverse the measuring volume. In
classical LDA, each of the components of the gas velocity is
measured consecutively. In newer systems, different veloc-
ity components can be measured in the same volume
Fig. 2. The tangential velocity profile in cyclones and swirl tubes. simultaneously using different colored laser light, which is
 W. Peng et al. / Powder Technology 127 (2002) 212–222 215

Fig. 4. Experimental setup.

detected separately. If the direction in which the particle In the present experiments, the LDA measurements were
crosses the measuring volume is unknown, it can be performed with a DANTEC four-beam two-component
determined by shifting the frequency of one of the beams high-performance 5-W Ar-ion backscatter laser doppler
(using a ‘‘Bragg cell’’) to create a moving, rather than a anemometer. This laser produced light of two different
stationary, interference pattern. wavelengths: 514.5 and 488 nm. During the experiments,
The main advantage of LDA is that it is nonintrusive. only the 514.5-nm wavelength was used. The frequency of
Problems can be caused by the seed particles not faithfully one of the laser beams was shifted in a Bragg cell by 40
following the gas flow and by the breaking of the laser MHz.
beams in the cylindrical wall of the cyclone or swirl tube. Dry air was fed into the cyclone from a 6-bar gauge
network. The flow was controlled by a ‘Bronkhorst’ mass

Fig. 6. The tangential gas velocity at a series of axial stations in a cylinder-

Fig. 5. The construction of the cyclone (left) and the swirl tube with the on-cone cyclone. The highest station is just in the cylinder; the rest are in
axial measuring stations indicated. the cone. The darker the points, the lower the station.
216 W. Peng et al. / Powder Technology 127 (2002) 212–222

Fig. 7. The tangential gas velocity at a series of axial stations in a cylindrical swirl tube. The highest station is just under the vortex finder.

flow meter. In order to determine the volumetric flow, the stations are given as distance below the cyclone roof divided
pressure and the temperature in the inlet of the cyclone were by the total length of the cyclone, roof to dust exit: AzA/L.
monitored with a ‘Druck PTX-610-0’ manometer and a The stations were:
‘Digitron, PT100’, respectively.
Diagrams of the cyclone and swirl tube used are shown – four in the gas inlet area (these stations are not shown
in Fig. 5. Their main dimensions are given in the diagrams; here, the results at these station are given in another paper
their detailed design is commercially sensitive. focussing on the flow in the inlet region),
– one in the cylindrical part, near the junction between the
3.2. Test program cylinder and the cone and
– four in the conical part.
Only the tangential and the axial velocity components
were measured. The measurement stations are indicated in In the swirl tube, measurements were performed at 10
Fig. 5. axial stations, each at 20 points in the radial direction. The
In the tangential inlet cyclone, LDA measurements were axial position of the stations are given as distance below the
performed at nine different axial stations each at 20 points in vane pack divided by the total length of the swirl tube, vane
the radial direction. In the cyclone, the axial position of the pack to dust exit: AzA/L.

Fig. 8. The axial gas velocity at a series of axial stations in a cylinder-on-cone cyclone. The highest station is just in the cylinder; the rest are in the cone.
 W. Peng et al. / Powder Technology 127 (2002) 212–222 217

– four between the vanes and the entrance of vortex finder

(these are not shown or discussed here),
– one just beneath the entrance of vortex finder and
– five between the vortex finder and the dust outlet.

The air flow in the conventional cyclone was approx-

imately 200 N m3/h, corresponding to an inlet velocity of
about 10 m/s at inlet conditions of 1.04 bar and 20.4 jC.
The air flow in the swirl tube was 175 N m3/h and a 3% gas
under flow was drawn from the dust collection box, the exit
angle from the vanes with the horizontal is 30j. The
configurations of both dedusters are given in Fig. 5.

3.3. CFD Fig. 10. Axial distribution of the radial velocity in the cyclone at the locus
of zero axial velocity.
The CFD package used in this work is 2-D axisymmetric
using a skew upwind differencing scheme (SUDS). The
turbulence model is a hybrid between an algebraic stress the curves is similar between the two devices and for both
model and a full Reynolds stress model [12]. devices, the shape is more or less independent of the axial
position. Also here, the wall moves inward as we move
down the cyclone. In the cyclone, the locus of zero axial
4. Results velocity seems to move inward as we move down the
conical section.
4.1. Tangential velocity
4.3. Radial velocity
Figs. 6 and 7 show the tangential velocity profiles in the
cyclone and the swirl tube, respectively. As we move down As mentioned, the radial velocity component could not
the conical section of the cyclone, the position of the wall be measured directly. It is, however, possible to calculate the
can be seen to move inward. In the cyclone, the character- local radial flow across the locus of zero axial velocity from
istic shape of the profile mentioned in the Introduction can the outer to the inner vortex. This can be done by numeri-
be recognized, with a near loss-free vortex surrounding a cally integrating the negative part of the curves in Figs. 8
core of near-solid body rotation. In the swirl tube, measure- and 9, which gives the axial flow downwards at each
ments very close to the axis were impossible, since no tracer station, and then computing the differences.
particles were present there. The difference in axial flow between two successive
stations gives the average radial flow from the outer to the
4.2. Axial velocity inner vortex between the two stations. If we integrate the
axial velocity between the cyclone wall, R, until the locus of
Figs. 8 and 9 show the axial velocity profiles in the zero axial velocity, Rtp at a flow station, we get the total
cylinder-on-cone cyclone and the swirl tube, respectively. downflow in the outer vortex at that station. Subtracting the
Also here, the characteristic shape of the profile shown in downflow in the outer vortex at two successive stations
the Introduction can be recognized. In the cyclone, the dip in gives the radial flow to the inner vortex between these
axial velocity near the axis is clearly visible. The shape of stations. If we now assume that the locus of zero radial

Fig. 9. The axial gas velocity at a series of axial stations in a cylindrical swirl tube. The highest station is just under the vortex finder.
218 W. Peng et al. / Powder Technology 127 (2002) 212–222

While the tangential velocity remains closely the same in

the axial direction in the cylinder-on-cone cyclone, it tends
to decrease with depth in the swirl tube. In both types of
apparatus, we have to do with attenuation of swirl as we
move down further from the inlet. Since the flow is not
constricted lower in the cylindrical swirl tube, this attenu-
ation gives rise to a decrease in the tangential velocity with
depth in this device.
One important difference between cyclone and swirl tube
is in the axial distribution of the inward radial flow. Fig. 10
Fig. 11. Axial distribution of the radial velocity in the swirl tube at the locus shows that the radial velocity is almost uniform along CS in
of zero axial velocity. the cylinder-on-cone cyclone (although we may be missing
small-scale differences). The values are very close to the
mean value of 0.46 m/s calculated from: Q/(2pRCSHCS).
However, in the swirl tube (Fig. 11), there are regions of
velocity is the surface CS (which is not exactly true), we can
strong inward velocity just under the mouth of the vortex
calculate the radial velocity in CS:
finder and in the bottom of the swirl tube. Over the rest of
"Z # "Z # CS, the inward velocity is very low; at one point, it even
R R appears to be negative. This would indicate outward flow at
2prvz ðrÞdr  2prvz ðrÞdr this point. This could be consistent with the view of some
Rtp Rtp
i iþ1 workers that a recirculatory flow pattern exists in the
¼ Qi  Qiþ1 ¼ 2pRCS  Aziþ1  zi A  vr ð2Þ separation space, with inward flow low in the body and
outward flow under the vortex finder (below the region of
where, Qi and Q i + 1 are the total flow downward at the ith lip leakage). The mean value of the radial velocity should be
and (i + 1)th axial flow stations in the outer vortex, respec- 1.1 m/s, which is higher than the mean of the values in Fig.
tively, zi and zi + 1, their axial height and vz and vr, the axial 11. The reason is probably that the two regions of high
and the radial velocity, respectively. RCS is the radius of the inward velocity just under the vortex finder and in the
CS, Rtp, the radius at which the axial velocity is zero, and R bottom of the tube are not covered completely by the
the radius of the wall of the separator. measurements.
The results of these calculations are shown in Figs. 10 Computational fluid dynamics (CFD) can help in eluci-
and 11. dating the difference in radial velocity distribution between
the two devices. Fig. 12 shows profile plots of the radial
velocity distribution in both devices from CFD simulations.
5. Discussion The CFD simulations show a strong inward radial flow
just under the vortex finder wall in both devices, (this is a
5.1. Comparison of the flow patterns in the cylinder-on-
cone cyclone and the swirl tube

Since there is very little information about the flow

pattern in cylindrical swirl tubes with vanes in the literature,
we will concentrate on this here and use the data from the
cyclone as comparison.
The results show that the global flow pattern in the
swirl tube is the same as in the cylinder-on-cone cyclone.
Also in the swirl tube, the tangential velocity distribution is
a Rankine-type vortex with a quasi-free vortex in the outer
part. Although we could not measure in the inner part to
find a region of solid body rotation, we know that the
tangential velocity must be zero on the axis of rotation.
The axial flow is also similar to that in conventional
cyclones: downward in the outer part of the body and
upward in the inner. Also in the swirl tube, gas flows from
the outer to the inner vortex.
There are, however, differences in detail between the
flows in cylinder-on-cone cyclones and cylindrical swirl Fig. 12. Profile plots from CFD simulations of the radial velocity
tubes. distribution in the cylinder-on-cone cyclone and the cylindrical swirl tube.
 W. Peng et al. / Powder Technology 127 (2002) 212–222 219

known feature of the flow in cyclones and is often referred Löffler model, they appear to have made an error in the
to as ‘lip leakage’). A region of strong inward flow can also calculation of D, the momentum exchange parameter. For
be seen at the bottom of the swirl tube, but not in the their cyclone, we get a value of 0.3 rather than their 0.04,
cylinder-on-cone cyclone. Both measurements and simula- which leads to prediction of a less intense swirl, more in line
tions thus indicate that the radial flow is more evenly with experiment.
distributed in a cylinder-on-cone configuration than in a
cylindrical one. 5.3. The flow assumptions made in cyclone performance
Although the lip leakage appears to be more localized in models tested by LDA and CFD
the cyclone than in the swirl tube, it is in fact stronger: the
CFD simulations show a maximum radial velocity in the Sweeping assumptions are made about the flow pattern
cyclone 4.41 times greater than the mean axial body in cyclones in the models for separation performance. We
velocity, while the same ratio is only 1.69 in the swirl tube. can test whether these assumptions are reasonable from the
measurements and numerical simulations.
5.2. Comparison of the results with model predictions One important assumption is that the locus of zero axial
velocity, separating the regions of downward and upward
Similar to Patterson and Munz (1996) [5], we can flow, coincides with the surface CS formed by prolonging
compare our measured profiles of the tangential velocity the vortex finder wall to the bottom of the cyclone.
vh with the predictions of models. Patterson and Munz [5] Fig. 13 shows results of both LDA measurements and
compared with the n-law (Eq. (1)) of Alexander [10] and CFD in a swirl tube with a cylindrical body for two different
with the predictions of Meissner and Löffler [13] for a vortex finder diameters. In Fig. 13a, a series of LDA
cyclone of similar geometry to ours. They found that the measurements are shown where the boundary between the
prediction of Alexander of n = 0.5 agreed well with their up- and downwardly directed flow have been pinpointed. In
results, while Meissner and Löffler predicted too little loss Fig. 13b, the loci of zero axial velocity have been made
in the cyclone body, and therefore, much too steep an visible in a CFD flow field by plotting contour plots of ln
increase in vh with decreasing radius, and much too intense (1 + v2z), which shows up the locus as a black region (low
a vortex. contour value).
Our results do not agree with their findings. We find that The qualitative picture is the same in the two figures:
the n-law agrees with the results in Fig. 6 for the much for the larger vortex finder (Dx a little larger than 1/2D),
higher value of n of 0.7, and that the model of Meissner and the surface appears to be cylindrical in agreement with
Löffler actually predicts a slightly less intense vortex than the model assumption, but for the smaller vortex finder,
we see experimentally. the inner vortex widens under the vortex finder wall and
As mentioned in the Introduction, insertion of probes in becomes the same diameter in the main part of the body
cyclones can lead to a decreased pressure drop, which as with the larger vortex finder. The agreement between
indicates a decreased swirl intensity [14]. The further the LDA and CFD in predicting the surface of zero axial
probe is inserted, the more significant this is. Patterson and velocity is excellent, which was verified by superimpos-
Munz [5] may therefore have measured a swirl less intense ing the results from the former on the profile plots from
than in the undisturbed cyclone. For the Meissner and the latter.

Fig. 13. The boundary between upwards and downward flow in the body of swirl tubes by (a) LDA and (b) CFD.
220 W. Peng et al. / Powder Technology 127 (2002) 212–222

Although the general shape of the surface of zero axial tube, there is a slight tendency for the tangential velocity to
velocity agrees with that assumed in the performance decrease as we move down. Overall, the measurements
models, the results in Fig. 13 indicate two interesting things: confirm the validity of this model assumption, at least for
– the diameter of the inner vortex appears to be deter- somewhat normal geometries.
mined by D rather than Dx, except right under the vortex A third assumption in the equilibrium orbit models is that
finder wall. This agrees with the results of Patterson and the radial velocity is uniform over CS. The radial velocity is
Munz (1996) [5] who found very little change in the locus the smallest velocity component, and it is difficult to
of zero axial velocity when Dx was changed substantially. measure with LDA: one cannot assume the seed particles
This would again mean that: to follow the gas motion in the radial direction in the strong
– the influence of Dx on the separation efficiency is centrifugal field. For the cylinder-on-cone cyclone, Figs. 10
mainly determined by the local flow pattern just under the and 12 show that the radial velocity is effectively constant
vortex finder wall. over CS, except for the lip leakage. In the cylindrical swirl
In much of the cyclone literature, the radial position of tube, on the other hand, we have two regions of strong
the boundary between up- and downwards flow is held to inward flow under the vortex finder and in the bottom of the
determine the cut size of the cyclone or swirl tube. However, tube, while through the rest of the separation space, the
for a particle caught in the upward flow to be lost, it still has radial velocity is very low.
to enter the vortex finder. Thus, although we see that the Thus, the flow assumptions made in the cyclone separa-
radial position of this boundary does not correspond with tion models are largely confirmed for the cylinder-on-cone
Dx, Dx remains the parameter determining the cyclone cut geometry. For a swirl tube, however, they are less consistent
size, in accordance with the equilibrium orbit cyclone with the real flow.
separation models. This difference between the two types of configuration
We do not have direct LDA measurements of the locus of indicates that the approach for modelling of separation
zero axial velocity in the cylinder-on-cone cyclone. Fig. 8 performance for cylindrical swirl tubes perhaps should be
indicates that the locus moves slightly inward as we move different to that for cylinder-on-cone cyclones. Further work
down the conical section. In Fig. 14, a CFD profile plot will be undertaken to formulate a reliable model for the
showing the locus is shown. It confirms that the locus separation efficiency of cylindrical swirl tubes.
converges as we move down the conical section of the
cyclone. A very similar shape of the locus of zero velocity in
a cylinder-on-cone cyclone is shown in Mothes and Löffler 6. Concluding remarks
[15].
Another assumption made in cyclone performance mod- The locus of zero axial velocity is at r/R = 0.53 in the
els is that the swirl velocity in the cyclone is constant cylindrical part of the cyclone and at r/R = 0.59 in the swirl
axially. Fig. 6 shows that the tangential velocity is effec- tube. In the cyclone, it moves inward as we move down the
tively constant at all axial positions in the cylinder-on-cone conical section. In the equilibrium orbit modelling concept,
cyclone, even until deep in the conical section. In the swirl this position is important in determining the cyclone cut
size, since a balance between apparent centrifugal force and
drag force on the critical particle with diameter dp,c rotating
in equilibrium in this locus at radius Rtp, with a tangential
velocity, vh,tp, gives:

3 2
pqp dp;c vh;tp g 2 18lf ðvr;tp Þ Rtp
¼ 3plf dp;c ðvr;tp ÞZ d p;c ¼
6Rtp qp g v2h;tp
ð3Þ

where qp is the particle density and lf is the viscosity of the

gas. vr,tp, the radial velocity of the fluid at the position of the
axial velocity turning point is proportional to 1/r if we
assume uniform radial flow, so Rtp cancels out in Eq. (3) for
a given gas flow rate. However, since the tangential gas (and
particle) velocity is higher at smaller radii, Eq. (3) would
indicate that, according to this concept, the separation in the
cylinder-on-cone cyclone should be slightly better than in
the swirl tube.
Fig. 14. The boundary between upward and downward flow in the body of Although the flow assumptions made in the cyclone
the cylinder-on-cone cyclone by CFD. separation models are generally supported by this study,
 W. Peng et al. / Powder Technology 127 (2002) 212–222 221

the details show some interesting features. The equilibrium g acceleration due to gravity
orbit models have been shown in the past to be quite HCS height of CS
successful for a wide variety of centrifugal separators, but L length of the cyclone or swirl tube
this study indicates that the locus of the axial velocity r radial coordinate
turning point is determined more by the diameter of the n exponent in Eq. (1)
body than the diameter of the vortex finder. Thus, the Qi volumetric flow at flow station i
influence of the vortex finder on the separation efficiency, R radius of the cyclone body = D/2
which is known to be strong, may be determined by the flow R1 radius delimiting the inner region of forced rotation
pattern just under the lip of the vortex finder more than in Rtp radius delimiting the inner region of upward axial
the core of the separation space. This information may velocity
give scope for formulating improved cyclone designs and RCS radius of CS, equals Rx
models. Rx radius of the vortex finder = Dx/2
The radial velocity distribution over the length of the vh tangential gas velocity
separation zone is reasonably uniform in the cylinder-on- vh,tp tangential gas velocity at the locus of zero axial
cone cyclone except for the lip leakage. In the cylindrical velocity
swirl tube, the radial velocity is, in addition to the lip vr radial gas velocity
leakage, also high in the bottom of the tube, and therefore vz axial gas velocity
weak in the rest of the body. The question arises: which z axial coordinate
radial velocity distribution is good for the separation effi-
ciency? This is not immediately evident. If the particle Greek
concentration is relatively low in the zones of high radial lf gas viscosity
velocity, a nonuniform radial velocity may be good for the qp particle density
separation.
Based on the results, we can draw the following con-
clusions: Acknowledgements
(1) The tangential velocity is axially constant in the
cylinder-on-cone cyclone, even until deep in the conical The authors would like to thank Ir A. Berrino for expertly
section. In the swirl tube, there is a slight tendency for the carrying out the CFD simulations. One of the authors, ACH,
tangential velocity to decrease as we move down. would like to thank Dr Louis Stein for valuable discussions.
(2) The radial velocity is fairly uniform over the length
of the separation zone in the cylinder-on-cone cyclone
except for a strong, but localized, lip leakage. In the
References
cylindrical swirl tube, little inward velocity is seen in
much of the separation zone, while regions of strong [1] W. Barth, Berechnung und Auslegung von Zyklonabscheidern auf
inward velocity were found under the vortex finder and Grund neuerer Untersuchungen, Brennst, Wärme Kraft 8 (Heft 1)
in the bottom of the body. LDA and CFD agreed very well (1956) 1 – 9.
in these respects. [2] P.W. Dietz, Collection efficiency of cyclone separators, AIChE J. 27
(3) The locus of zero axial velocity was seen to be (6) (1981) 888 – 892.
[3] H. Mothes, F. Löffler, Prediction of particle removal in cyclone sep-
determined more by the diameter of the body than that of arators, Int. Chem. Eng. 28 (2) (1988) 231 – 240.
the vortex finder. Also in this respect, LDA and CFD [4] D. Leith, Cyclones, in: M.E. Fayed, L. Otten (Eds.), Handbook of
simulations agreed closely. This also agrees with the results Powder Science and Technology, Van Nostrand-Reinhold, New York,
of Patterson and Munz [5]. 1984, Chap. 16.
[5] P.A. Patterson, R.J. Munz, Gas and particle flow patterns in cyclones
(4) The radial profiles of the tangential velocity were
at room and elevated temperatures, Can. J. Chem. Eng. 74 (1996)
well described by Eq. (1) with n = 0.7 and by the model of 213 – 221.
Meissner and Löffler. [6] R. Wakelin, Vortex breakdown in dust-collecting return-flow cyclo-
(5) The flow assumptions of the ‘‘equilibrium orbit’’ nes. PhD thesis, University of Canterbury, Christchurch, NZ, 1992.
models were shown to be reasonably realistic in the cylin- [7] W. Yaodong, J. Youhai, J. Zhongli, S. Mingxian, Research and devel-
der-on-cone cyclone, while the flow pattern—particularly opment of the third stage multicyclone separator in FCC power re-
covery system, Proceedings of INTERPEC CHINA’91.
the radial velocity from the outer to the inner vortex—was [8] D.R. Garwood, et al., Flow Analysis of a High Efficiency Cyclone
somewhat different in the swirl tube. Separator Operating Under Base Suction, Filtech Conference, Karl-
stryhe, vol. 2, Filtration Society, Horsham, UK, 1993.
List of symbols [9] W. Peng, P.J.A.J. Boot, A.C. Hoffmann, H.W.A. Dries, J. Kater, A.
C constant in Eq. (1) Ekker, Flow in the inlet region in tangential inlet cyclones, Ind. Eng.
Chem. Res. 40 (2001) 5649 – 5655.
dp,c critical particle diameter [10] R. Mck. Alexander, Fundamentals of cyclone design and operation,
D momentum exchange parameter in the Meissner Proc.—Australas. Inst. Min. Metall. 203 (1949) 152 – 153.
and Löffler model [11] A.C. Hoffmann, M. de Groot, A. Hospers, The effect of the dust
222 W. Peng et al. / Powder Technology 127 (2002) 212–222

collection system on the flow pattern and separation efficiency of a [14] A.C. Hoffmann, M. de Groot, W. Peng, H.W.A. Dries, J. Kater, Ad-
gas cyclone, Can. J. Chem. Eng. 74 (1996) 464 – 470. vantages and risks in increasing cyclone separator length, AIChE J. 47
[12] F. Boysan, J. Swithenbank, W.H. Ayers, Mathematical modelling of (2001) 2452 – 2460.
gas-particle flows in cyclone separators, in: N.P. Cheremisinof (Ed.), [15] H. Mothes, F. Löffler, Motion and deposition of particles in cyclones,
Encyclopedia of Fluid Mechanics, vol. 4, Gulf Publ., Book Division, Ger. Chem. Eng. 8 (1985) 223 – 233.
Houston, 1986, pp. 1307 – 1329.
[13] P. Meissner, F. Löffler, Zur Berechnung des Strömungsfeldes im Zy-
klonabscheider Synopse, Chem. Eng. Technol. 50 (1978) 471.

View publication stats