Chemical Engineering Journal 80 (2000) 295–303

Evaluation of hydrocyclone models for practical applications

Wu Chen∗ , Nathalie Zydek, Frank Parma
Dow Chemical Company, Freeport, TX, USA

Abstract
Hydrocyclone is an important industrial solids–liquid separation equipment. Although widely used nowadays, the selection and design of
hydrocyclones are still empirical and experience based. Although quite a few hydrocyclone models had been developed over the years, the
validity of these models for practical applications was still not clear to all users. In this work, seven hydrocyclone models were evaluated.
They are the more theoretically oriented models by Bohnet, Braun, and Mueller, semi-empirical models by Schubert/Neese and Svarovsky,
plus the empirical models by Plitt and Krebs Engineers. Plant operation data from the Dow Chemical Company were used to compare
with the predictions from these models. It was found that most of the models studied work well for certain cases but none of the models
can predict all applications. The best results are obtained by using more than one model for predictions. Some experimental data are very
important in choosing the models. © 2000 Elsevier Science B.V. All rights reserved.
Keywords: Hydrocyclone; Model; Cut size; Grade efficiency; Solid–liquid separation

1. Introduction from pilot tests and production plants at the Dow Chemi-
cal Company’s production complex at Freeport, Texas. The
One type of the industrial centrifugal separators is the comparisons were done with pressure drop, cut size, reduced
hydrocyclone. It has been used as an industrial separation grade efficiency, particle size distribution, flow split, and
device for more than 50 years. Due to the simple design, concentration in the underflow. The goal was to understand
low cost, easy operation and low maintenance, hydrocy- how well these models could predict separation results for
clones have become important equipment for solid–liquid a wide range of cyclone geometries, operating parameters
separations. Although hydrocyclones are widely used and material properties.
nowadays, the separation phenomena are still not fully
understood. The selection and design of hydrocyclones are
empirical and experience based. There is a need to have a 2. The hydrocyclone
reliable hydrocyclone model that allows design engineers
to predict the performance of hydrocyclones and make the The drawing of a typical hydrocyclone is shown in Fig. 1.
proper selections. A hydrocyclone consists of a cylindrical section joined to a
A few hydrocyclone models have been developed over the conical section. The suspension is fed tangentially through
years. These models can be very empirical in nature, theo- the inlet opening into the cylindrical portion. The fast move-
retically based, or in between. These models often have their ment of suspension develops an intense whirling motion that
limitations due to the specific system the model development causes a separation of solid particles from the liquid by virtue
was based on. So far there has not been a model which can of the centrifugal acceleration. One part of the feed stream
simulate most hydrocyclone operations and earned recogni- is discharged out of the top of the hydrocyclone, through
tion among cyclone users and researchers. a cylindrical pipe called the vortex finder. This stream is
It is the purpose of this study to understand how well the called the overflow which contains more liquid and finer
hydrocyclone models work for industrial processes and de- particles. The second stream is called the underflow which
velop a recommendation for industrial users for using these is discharged through a circular opening (the apex) at the
models. A few hydrocyclone models, both empirical and end of the conical section. Normally, larger and/or heavier
theoretical based, were selected to simulate hydrocyclone solids are discharged through the underflow.
operations and the results were compared with data obtained For industrial applications, it is desired to know the flow
rate, solid concentration, and particle size distribution in
∗ Corresponding author. Tel.: +1-979-2389943; fax: +1-979-2380969. the underflow and overflow for a given feed under certain
E-mail address: wuchen@dow.com (W. Chen). operating conditions.

1383-5866/00/$ – see front matter © 2000 Elsevier Science B.V. All rights reserved.
PII: S 1 3 8 3 - 5 8 6 6 ( 0 0 ) 0 0 1 0 5 - 2
296 W. Chen et al. / Chemical Engineering Journal 80 (2000) 295–303

3.2. Braun’s model

This model is built upon Bohnet’s model. A cyclone is

divided into three sections (entrance, downflow and upflow)
according to Dietz [5]. The solid concentration in each sec-
tion is described with a differential equation. These equa-
tions are solved numerically to obtain the concentration of
each individual particle size in the underflow and overflow.
To calculate the grade efficiency, the underflow or overflow
rate is required. Braun did not propose a correlation to sim-
ulate the flow split ratio of the underflow rate to the feed
rate. The users need to know or guess a underflow to feed
flow ratio to perform the calculation.
Fig. 1. Cross sectional view of a hydrocyclone.
3.3. Mueller’s model

3. Hydrocyclone models Mueller [6] modified Braun’s model by incorporating the

development of Harms [7] and took into account of the sec-
Hydrocyclone models are available to calculate the sep- ondary flow along the top of the hydrocyclone roof and along
aration efficiencies of solid particles and pressure drops in the vortex finder. If the particles following this flow have
hydrocyclones. These models are normally based on one or low specific gravity, the centrifugal forces are not strong
more of the following principles [1–3]. enough to push these particles towards the wall of the hy-
1. The empirical models: These models are determined by drocyclone and these particles may bypass to the overflow
fitting formulas to experimental data. without getting to the downflow area.
2. The equilibrium orbit theory: A particle attains an equi- Above three models have very sound theoretical basis and
librium radial position in the cyclone where their termi- are able to take the cyclone geometry into account. How-
nal settling velocity is equal to the radial velocity of the ever, they adopted in their models some correlations which
liquid. That means, if the liquid flows outward, the parti- might not be appropriate for some systems and parameters
cles will go toward to the wall and are separated through in those correlations might be hard to determine by users.
the underflow. If the liquid flows inward, the particles For example, Eiler’s [8] correlation was used to calculate the
will go with the liquid to the overflow. viscosity of suspensions. Since the prediction of suspension
3. The residence time theory: A particle is considered sep- viscosity is still not an exact science and the use of any cor-
arated if it can travel to the cyclone wall region within relation may lead to erroneous results. Friction factors along
the residence time of that particle in the hydrocyclone. the wall of cyclone inlet, cyclone top and conical section
4. The turbulent two-phase flow theory: The separation is were also used in these models. The values of these friction
caused by the turbulent cross flow flowing in perpendic- factors are empirical and not readily available unless the ac-
ular to the direction of the force field. tual hydrocyclone data is available to fit these parameters.
In this work, seven models were evaluated. The under-
lining principal and the characteristics of each model are 3.4. Schubert/Neesse’s model
briefly discussed in the following. The first three models
were developed at the Technical University of Braun- This semi-empirical model was developed by H. Schubert
schweig, Germany and carried some similarities. and T.H. Neesse [9]. The theoretical part is based on the
turbulent two-phase flow theory. The cut size and pressure
3.1. Bohnet’s model drop are calculated by iterations. Based on the feed con-
centration, this model was divided into dilute flow (<25%
This is based on the equilibrium orbit theory [4]. Bohnet volume fraction) and dense flow (>25% volume fraction)
defined a critical particle size as the size of particles which models.
neither move toward the wall of the hydrocyclone (and thus
being separated) nor move inward and out from the vortex 3.5. Plitt’s model
finder. Therefore, a particle with the critical particle size
has 50–50% chance of either being separated or not. This Plitt [10] developed an empirical model with the con-
critical size is considered equivalent to the d50 cut size. stants and the relationships between the operating parame-
Bohnet used the force balance at the radius of the vortex ters determined from 297 experiments. This model provides
finder to calculate the critical particle size. This model does a simple equation for direct calculations of cut size, grade
not propose a correlation for grade efficiency calculation. efficiency, flow split and pressure drop.
 W. Chen et al. / Chemical Engineering Journal 80 (2000) 295–303 297

Table 1
Required input parameters in each model
Bohnet Braun Mueller Schubert Plitt Svarovsky Krebs

Hydrocyclone Geometry
Diameter * * * * * * *
Vortex finder diameter * * * * *
Inlet geometry * * * * *
Total length * * * *
Vortex finder immersion * * * *
Apex size * * * * *
Cone angle * * *

Material properties
Feed particle size distribution *
Feed solid concentration * * * * * *
Maximum solid volume concentration * * * *
Liquid viscosity * * * * * * *
Liquid density * * * * * * *
Solid density * * * * * * *
Feed rate * * * * * * *
Overflow rate * *

Others (experimental parameters)

Wall friction factor * * *
Friction coefficient at inlet * *
Friction coefficient at top * *
Friction coefficient at cone * *
Wall roughness * *
Empirical constants * *

3.6. Svarovsky’s model The parameters used for model calculation are different
from model to model. Some use more and some use less. A
Svarovsky proposed a dimensionless group correlation for summary of the input parameters for each model is shown
analyzing hydrocyclones [1]. His correlation derived from in Table 1. The cyclone size (diameter of the cylindrical
Rietma’s optimum hydrocyclone proportions [11] were used section) is the only parameter used by all models. Braun and
in the work for calculation of the pressure drop, flow split, Mueller use more input parameters than other models. As
and the cut size. The closer the cyclone’s geometry propor- for the simulation results, all models calculate pressure drop
tions are to the Rietema’s proportions, the better fit of this and cut size. Braun’s and Mueller’s models calculate particle
model. Svarovsky did not propose a correlation to predict size distribution while Schubert’s, Plitt’s and Krebs’ models
the grade efficiency from the cut size. predict grade efficiency. With the knowledge of flow split
(by measuring or guessing), all the data can be calculated
(Table 2).
3.7. Krebs Engineers’ model

This model was described by Besendorfer [12]. It is an 4. Verification of models

empirical model and a flow rate–pressure drop relationship is
required for calculation. This model is very simple and uses The comparisons of the above mentioned models were
the least number of parameters compared to other models. made with data collected from a series of pilot tests and

Table 2
Data predicted by each model
Bohnet Braun Mueller Schubert Plitt Svarovsky Krebs

Pressure drop * * * * * * *
Cut size * * * * *
Grade efficiency * * *
Flow split * * *
Particle size distribution * *
298 W. Chen et al. / Chemical Engineering Journal 80 (2000) 295–303

Table 3
Hydrocyclone systems used for model comparison
System Test dust/water Salt/glycerin Lime/water Salt/organic solvent

Hydrocyclone geometry
Cyclone size (m) 0.0445 0.1016 0.0445 0.0100
Cyclone length (m) 0.4150 0.8382 0.4150 0.0800
Vortex finder diameter (m) Vary 0.0254 0.0140 0.0021
Vortex finder insertion depth (m) Vary 0.0445 0.0165 0.0083
Apex (m) Vary 0.0127 0.0032 0.0024
Inlet diameter (m) – 0.0381 – 0.0021
Inlet height (m) Vary – 0.0100 –
Inlet width (m) Vary – 0.0080 –
Cone angle (◦ ) 6.8 12.0 6.8 6.0

Material properties
Solid density (kg/m3 ) 2770 2207 2557 2160
Liquid density (kg/m3 ) 997 1268 1031 1170
Liquid viscosity (Pa s) 0.0010 0.0040 0.0013 0.0012

some plant operation data. The pilot hydrocyclone tests were The results show that the model performance is system de-
conducted with slurries made of test dust and water. Three pendent. Mueller’s model worked well for the test dust/water
sets of plant data were used, salt in glycerin, lime in water, system but turned out to have very poor prediction on plant
and salt in an organic solvent. The geometry of the hydro- systems. The simplest model is the Krebs model. It does not
cyclone used in each case and the properties of materials appear to be accurate for all test dust runs but predicts ex-
involved are shown in Table 3. tremely well for the plant systems. Up to this point, it can be
For all the data series, the pressure drop across the cyclone seen that model prediction without some operation data ver-
feed and overflow were recorded and the volume flow rates ification is dangerous. A model may work well for certain
of feed, overflow and underflow were measured. Samples of systems but erroneous results can occur for other systems.
feed, overflow, and underflow were collected for measuring
solid concentrations and particle size distributions. For the 4.2. Cut size
reliability of measurement, multiple samples were taken and
the averaged values were used. The cut size is defined as the size of particles which
The correct sample analyses were key to the success of have 50% chances of being separated to the underflow. The
model comparisons. The materials involved in these four cut size is calculated directly in models of Bohnet, Schu-
systems all behave differently, some contain large crystals bert/Neesse, Plitt, Svarovsky, and Krebs. The cut size for
and some have very fine particles. One system was even the Braun’s and Mueller’s models as well as test data were
temperature sensitive. Therefore, the sample analyses were obtained from the reduced grade efficiency curves.
quite tedious and required different techniques for different For practical considerations, all models seemed to work
systems. The majority of the time in this work was spent in well for the test dust/water system but Braun’s and Plitt’s
ensuring the good data quality. models had a little better overall performance. Larger dis-
The properties of interest in hydrocyclone simulations are crepancies between the model prediction and measurement
the pressure drop, particle separation efficiency, and the flow were observed for the plant hydrocyclones system. Oc-
split (underflow to feed ratio). These data allow the calcula- casionally, one model may predict a cut size close to the
tion of the solid concentration and particle size distribution measurement but in general the performance was poor. The
in underflow and overflow. All the data obtained are tabu- poor performance may be due to the difficulties in getting
lated in Tables 4 and 5. the exact dimensions of the operating cyclones in the plant.
For some models, the correct cyclone dimensions are very
4.1. Pressure drop important for the calculation.

Pressure drop is the first design parameter for all cyclone 4.3. Reduced grade efficiency
applications. All the models studied in this work predict
pressure drop across the cyclone. These calculated values Grade efficiency is the separation efficiency of particles
were compared with the pressure drop measured from the with a particular particle size. All the solids reported to the
operating cyclones. The comparisons were tabulated and the underflow are considered separated. Since the feed splits into
sums of the squared deviations were also illustrated (Tables 4 underflow and overflow in a hydrocyclone, the flow splitting
and 5). itself provides a ‘guaranteed’ separation efficiency. If the
 W. Chen et al. / Chemical Engineering Journal 80 (2000) 295–303 299

Table 4
Model comparison for test dust/water system
System Test dust/water

Run no. 1 2 3 4 5 6 7 8 –
Vortex finder (m) 0.0140 0.0140 0.0080 0.0080 0.0080 0.0080 0.0080 0.0080 –
Apex (m) 0.0032 0.0022 0.0032 0.0032 0.0022 0.0022 0.0032 0.0032 –
Feed rate (m3 /h) 3.1580 3.3609 1.4334 1.7475 1.4211 1.7159 1.4790 1.7460 –
Overflow rate (m3 /h) 3.0800 3.3259 1.2827 1.5778 1.3398 1.6264 1.3310 1.5800 –
Underflow rate (m3 /h) 0.0780 0.0350 0.1507 0.1697 0.0813 0.0895 0.1480 0.1660 –
Feed concentration (vol. %) 0.50 0.31 0.47 0.46 0.45 0.41 4.66 4.64 –
Overflow concentration (vol. %) 0.23 0.22 0.19 0.18 0.21 0.19 2.16 2.00 –
Underflow concentration (vol. %) 9.30 9.50 2.85 3.07 4.51 4.32 27.22 29.68 –

P
Comparison of model predictions for ∆p R2
Measured (bar) 2.00 2.00 2.00 3.00 2.00 3.00 2.00 3.00 0.00
Bohnet (bar) 3.29 3.75 1.24 1.98 1.15 1.79 1.28 1.88 10.30
Braun (bar) 3.29 3.75 1.24 1.98 1.15 1.79 1.28 1.88 10.30
Mueller (bar) 2.70 3.12 1.87 2.84 1.86 2.78 2.00 2.84 1.88
Schubert/Neese (bar) 3.26 3.68 3.76 5.01 3.31 4.82 3.85 5.35 25.52
Plitt (bar) 2.36 2.70 1.78 2.53 1.87 2.62 1.92 2.58 1.23
Svarovsky (bar) 1.89 2.16 0.36 0.54 0.35 0.52 0.35 0.49 26.67
Krebs (bar) 2.11 2.39 0.43 0.65 0.43 0.62 0.46 0.64 24.22

P
Comparison of model predictions for cut size R2
Measured (␮m) 9.80 9.70 6.60 5.95 7.60 7.95 7.85 6.80 0.00
Bohnet (␮m) 13.61 13.20 10.86 9.84 10.91 9.93 10.69 9.84 92.23
Braun (␮m) 8.95 9.50 8.45 7.49 8.95 7.90 9.75 8.80 15.99
Mueller (␮m) 5.85 5.65 5.45 4.90 5.70 5.05 5.95 5.40 52.02
Schubert/Neese (␮m) 15.26 14.24 11.43 10.38 12.96 11.97 12.79 11.84 188.07
Plitt (␮m) 10.84 13.63 6.80 6.21 8.89 8.15 8.73 8.08 20.75
Svarovsky (␮m) 9.35 9.67 12.12 10.98 13.33 12.06 15.14 13.89 209.11
Krebs (␮m) 5.46 5.28 8.49 7.60 8.53 7.67 9.40 8.56 51.11

P
Comparison of model predictions for grade efficiency and particle size distribution
Bohnet – – – – – – – – –
Braun 1 1 3 2 2 2 3 3 17
Mueller 3 3 2 2 2 3 3 3 21
Schubert/Neese 3 3 4 3 3 3 4 4 27
Plitt 2 3 1 1 2 2 5 5 21
Svarovsky – – – – – – – – –
Krebs 4 4 3 2 2 1 2 2 20

P
Comparison of model predictions for underflow coacentration R2
Measured (vol. %) 9.30 9.50 2.85 3.07 4.51 4.32 27.22 29.68 0.00
Bohnet (vol. %) – – – – – – – –
Braun (vol. %) 9.17 9.34 2.34 2.50 3.72 3.98 23.03 24.78 0.43
Mueller (vol. %) 10.53 13.89 2.70 3.05 4.03 4.66 25.32 29.19 0.25
Schubert/Neese (vol. %) 6.64 5.40 2.04 2.16 3.10 3.23 20.17 21.44 1.46
Plitt (vol. %) 8.05 5.51 2.62 2.81 3.95 3.99 24.92 26.51 0.33
Svarovsky (vol. %) – – – – – – – – –
Krebs (vol. %) 11.33 115.27 2.41 2.73 3.99 4.13 24.05 25.71 0.64

separation efficiency due to the flow splitting is subtracted, size distribution and the measured underflow and overflow
the resulting separation efficiency is called the reduced grade rates.
efficiency. The separation efficiencies calculated by models Since the reduced grade efficiency is not a single num-
are actually the reduced efficiency. In Schubert/Neese’s, ber, the model comparisons are done with charts. Due to
Plitt’s and Krebs’ models, correlations are offered to cal- the space limitation, only the reduced grade efficiency chart
culate the reduced grade efficiency from the cut size. In for run no. 1 is shown (Fig. 2). For all the other data set,
Braun’s and Mueller’s models, the reduced grade effi- the comparison is represented with numerical numbers in
ciencies were calculated with the model predicted particle Tables 4 and 5. A number of ‘1’ is given to the model which
300 W. Chen et al. / Chemical Engineering Journal 80 (2000) 295–303

Table 5
Model comparison for plant systems
System Salt/glycerin Lime/water Salt/organic solvent

Run no. 9 10 11 12
Feed rate (m3 /h) 13.6270 2.6230 0.2910 0.2430 –
Overflow rate (m3 /h) 8.2450 2.5440 0.1300 0.0900 –
Underflow rate (m3 /h) 5.3830 0.0790 0.16201 0.1520 –
Feed concentration (vol. %) 31.10 1.70 4.45 12.42 –
Overflow concentration (vol. %) 18.50 0.93 0.01 0.04 –
Underflow concentration (vol. %) 50.30 23.2 8.01 19.72 –

P
Comparison of model predictions for ∆p R2
Measured (bar) 3.45 1.40 1.24 1.24 0.00
Bohnet (bar) 0.29 2.00 1.83 5.10 25.57
Braun (bar) 0.29 2.00 1.83 5.10 25.57
Mueller (bar) 1.98 1.85 14.35 9.67 245.18
Schubert/Neese (bar) 2.52 2.37 N/A N/A 1.80
Plitt (bar) 2.01 1.71 4.48 4.49 23.18
Svarovsky (bar) 0.81 1.23 6.13 3.57 36.31
Krebs (bar) 3.80 1.45 1.40 1.04 0.19

P 2
Comparison of model predictions for cut size R
Measured (␮m) 105.00 25.40 13.00 N/A 0.00
Bohnet (␮m) 54.80 18.65 4.15 N/A 2643.93
Braun (␮m) 135.00 13.50 3.98 N/A 1122.97
Mueller (␮m) N/A 8.60 2.55 N/A 391.44
Schubert/Neese (␮m) 231.60 22.17 N/A N/A 16037.99
Plitt (␮m) 205.20 15.76 3.30 N/A 10227.06
Svarovsky (␮m) 143.90 13.68 3.19 N/A 1746.80
Krebs (␮m) 76.70 7.80 2.78 N/A 1215.10

Comparison of model predictions for grade efficiency and particle size distribution
Bohnet – – – – –
Braun 3 3 3 3 12
Mueller 4 4 3 3 14
Schubert/Neese 5 1 – – 6
Plitt 2 5 5 15
Svarovsky – – – – –
Krebs 3 4 3 2 12

P 2
Comparison of model predictions for underflow concentration R
Measured (vol. %) 50.30 23.26 8.01 19.72 0.00
Bohnet (vol. %) – – – – –
Braun (vol. %) 45.13 29.29 8.01 19.49 0.63
Mueller (vol. %) 37.34 34.97 8.01 19.69 3.05
Schubert/Neese (vol. %) 27.13 23.91 – – 5.37
Plitt (vol. %) 42.78 27.53 7.76 17.65 0.79
Svarovsky (vol. %) – – – – –
Krebs (vol. %) 61.13 37.04 8.01 19.74 3.07

shows the best fit to the measured data and the larger the 4.4. Particle size distribution
number the worse the data fitting.
Again, a strong dependency of model performance on the The solutions of Braun’s and Mueller’s models provide
system evaluated can be seen. For example, Braun’s model the particle size distribution of underflow and overflow. In
worked very well for the first two runs of test dust/water other models, the flow splits are required to calculate the
system but not as good as Krebs or Plitt’s models in other particle size distribution from the reduced grade efficiency.
runs. A number of ‘5’ on Tables 4 and 5 represent a really An example of the simulated overflow and underflow parti-
bad fit. In general, Braun’s, Mueller’s and Krebs’ models cle size distribution for the salt/glycerin sysem are shown in
are more reliable in grade efficiency and grade efficiency Figs. 3 and 4. All models seemed to do pretty well in pre-
predictions. dicting the particle size distribution. If only the particle size
 W. Chen et al. / Chemical Engineering Journal 80 (2000) 295–303 301

Fig. 2. Grade efficiency comparison for run no. 1.

distribution is examined, one may conclude that the models cases studied in this work. Therefore, the measured flow
are very effective. However, once the pressure drop, grade rates were used for this study.
efficiency and concentrations are considered, limitations in The lack of capability to predict the flow split ratio is the
models can be realized. Models verified with particle size major deficiency for all models. Without the correct predic-
distributions only need to be treated with caution. tion of the flow ratio, the hydrocyclone simulations are not
very meaningful.
4.5. Flow split
4.6. Underflow and overflow concentration, and solid flow
The flow split ratio between the underflow and overflow is rates
required for the calculation between the grade efficiency and
particle size distribution. Among the models studied, only Concentrations and solid flow can be calculated from the
Schubert/Neesse, Plitt and Svarovsky offered predictions for above discussed data. Braun’s model seems to have a little
the flow split and the predictions were not accurate for the better overall performance. However, the best model for each

Fig. 3. Particle size distribution in the overflow for salt/glycerin system.

302 W. Chen et al. / Chemical Engineering Journal 80 (2000) 295–303

Fig. 4. Particle size distribution in the underflow for salt/glycerin system.

system is again varying. Test data need to be available to authors and involved a bit of uncertainties and system de-
determine which model is the best choice. pendent parameters. Therefore, although these models are
theoretically sound but the uncertainties of some key pa-
rameters make the prediction of hydrocyclone performance
5. Discussion difficult.
Schbert and Neesse considered a homogenous turbulence
It can be seen that none of the models made good predic- field and force field for mathematical simplifications. This
tions for every set of data. Under different operating condi- assumption may not be representative for the real conditions
tions it was a different model which could be considered as in a hydrocyclone. Also a constant value which is used for
the best fit for the experimental data. laminar flow was used in the grade efficiency calculation.
It also can be seen that no models can be considered as the The field in a cyclone is turbulent most of the time and other
best model if both pressure drop and separation performance values may be more appropriate.
are compared. A model may have good prediction on separa- Bohnet calculates the size of particles which cannot be
tion but perform poorly on predicting pressure drop. Braun’s separated because of a balance of forces. In a real separation
model made good grade efficiency prediction, but the cal- all particles go to overflow or underflow. Also the friction
culation of pressure drop was not very good. The pressure coefficients used for the calculation of the tangential velocity
drops were best predicted by Krebs’ and Plitt’s models. were set as a constant value and the wall roughness was not
Braun’s and Mueller’s models are theoretical models with taken into account.
some empirical correlations. The calculations of these cor- During the experiments, an air core could be seen which
relations are not easy and sometimes involve assumptions influenced the flow inside of a cyclone. Only Braun took this
and uncertainties. For example, the friction factors used for air core into account in the calculation of the turbulent ex-
calculating the tangential velocity in a cyclone is not easy to change coefficient. Other models didn’t consider this effect.
get by an industrial user, the factors used in Mueller’s model In the model by Svarovsky, Rietma’s optimum hydrocy-
are based on experimental data obtained from tests with cy- clone proportions were used. The real proportions of hydro-
clones running air, water and oil [7]. In Braun’s model the cyclones evaluated were not necessary close to the values
formula for rough pipes from Colebrook [13] was used and suggested by Rietma.
the coefficient of roughness of the wall was set as a con- Plitt’s and Krebs’ models were experimentally based. The
stant (1 ␮m). These coefficients will not be the same for coefficients used for calculations are obtained by fitting the
all systems. For the calculation of pressure drop in Braun’s models with experimental data. These parameters may not
model (same as Bohnet’s model), a constant friction was be appropriate once dealing with systems outside the do-
used. This friction coefficient may not be the same for the main of their experimental database. However, these param-
real conditions. Both models also include correlations to cal- eters allow the user to adjust the simulation according to
culate tangential velocity, suspension viscosity and turbulent the real data. A distribution formula (also called sharpness
exchange coefficient. These predictions came from others of separation) is also used in Plitt’s and Krebs’ models for
 W. Chen et al. / Chemical Engineering Journal 80 (2000) 295–303 303

grade efficiency calculation. Different distribution formula References

can lead to very different results. Without actual data, it is
difficult to determine the reliability of the simulation. [1] L. Svarovsky, Hydrocyclones, Technomics, Lancaster, 1994.
[2] B.C. Flinthoff, L.R. Plitt, A.A. Turak, Cyclone modeling: a review
of present technology, CIM Bull. 80 (905) (1987) 39–50.
6. Conclusions [3] T. Braun, Theoretische und experimentelle Untersuchungen des
Einflußes der Feststoffkonzentration und der Partikelgrößenverteilung
auf das Trennverhalten von Hydrozyklonen, Doctoral dissertation,
From this study, it was realized that a good simulation of Technical University of Braunschweig, Braunschweig, Germany,
hydrocyclone operations involves the calculation of pressure 1989.
drop, cut size, grade efficiency, and the flow split (underflow [4] M. Bohnet, Neuere Untersuchungen über die Trennwirkung und den
to feed or overflow to feed ratio). It is still impossible to Druckverlust von Hydrozyklonen, Chemische Technik 20 (9) (1969)
have one model which can describes all the above mentioned 376–381.
[5] P. Dietz, Collection efficiency of cyclone separators, AIChE J. 27 (6)
parameters satisfactorily. (1981) 888–892.
Due to the complexity of hydrocyclone operations, most [6] M. Mueller, M. Bohnet, Pressure Drop and Grade Efficiency of a
of the existing models actually are a combination of a group Newly Developed Hydrocyclone for the Separation of Two Different
of models, and each of these ‘submodels’ describes a cer- Solids from a Liquid Flow, third International Conference on
tain property in a cyclone operation. With these different Multiphase Flow, Lyon, France, 1998.
[7] G. Harms, Flüssig/Flüssig-Trennung im Hydrozyklon, Doctoral
considerations, assumptions and adjustable parameters, it is dissertation, Technical University of Braunschweig, Aachen,
possible to select a combination of parameters to fit a partic- 1996.
ular set of hydrocyclone data very well. However, this best [8] H. Eilers, Die viskosität von emulsionen hochviskoser stoffe als
combination of parameters may not necessary be suitable funktion der konzentration, Kolloid Zeitschrift 97 (3) (1941) 313–
for another system or even the same system operated under 321.
[9] T. Neesse, H. Schubert, Praktische und theoretische aspekte der
different conditions. dichtstromklassierung, Aufbereitungstechnik 32 (9) (1991) 459–
The complex nature of the two-phase flow and particulate 472.
system make it difficult to develop a single perfect model [10] L.R. Plitt, B.C. Flinthoff, Cyclone modelling: a review of present
to describe all systems well. However, models can still be a technology, CIM Bull. 80 (9) (1987) 905, 39–50.
good tool for estimating hydrocyclone performance as long [11] K. Rietma, Performance and design of hydrocyclones, parts I, II, III
and IV, Chem. Eng. Sci. 15 (1961) 298–325.
as the models are applied in the right domain. [12] C. Besendorfer, Exert the force of hydrocyclone, Chem. Eng. 9
It is recommended based on the results of this work that (1996) 108–114.
some data should be obtained to select an appropriate model [13] C.F. Colebrook, Turbulenten Drehsenkenstromung im Zyklon-
for calculations. Any model predictions without experimen- abscheider, Doctoral dissertation, University of Karlsruhe, Karlsruhe,
tal data should be treated cautiously. 1977.