Вы находитесь на странице: 1из 16

Powder Technology 217 (2012) 8499

Contents lists available at SciVerse ScienceDirect

Powder Technology
journal homepage: www.elsevier.com/locate/powtec

Modeling and Pareto optimization of gas cyclone separator performance using RBF
type articial neural networks and genetic algorithms
Khairy Elsayed , Chris Lacor
Vrije Universiteit Brussel, Department of Mechanical Engineering, Research Group Fluid Mechanics and Thermodynamics, Pleinlaan 2, 1050 Brussels, Belgium

a r t i c l e

i n f o

Article history:
Received 26 July 2011
Received in revised form 3 October 2011
Accepted 9 October 2011
Available online 18 October 2011
Keywords:
Cyclone separator
Articial neural network (ANN)
Genetic algorithm (GA)
Multi-objective optimization
Pareto front

a b s t r a c t
Both the pressure drop and the cut-off diameter are important performance parameters in the design of the cyclone separator. In this paper, a multi-objective optimization study of the gas cyclone separator is performed. In
order to predict accurately the complex non linear relationships between the performance parameters (pressure
drop and cut-off diameter) and the geometrical dimensions, two radial basis function neural networks (RBFNNs)
are developed and employed to model the pressure drop and the cut-off diameter for cyclone separators. The articial neural networks have been trained and tested by the experimental data available in literatures for the pressure drop and the Iozia and Leith model for the cut-off diameter. The results demonstrate that articial neural
networks can offer an alternative and powerful approach to model the cyclone performance parameters. The analysis indicates the signicant effect of the vortex nder diameter Dx, the vortex nder length S, the inlet width b and
the total height Ht. The response surface methodology has been used to t a second-order polynomial to the
RBFNN. The second-order polynomial has been used to study the interaction between the geometrical parameters.
The two trained articial neural networks have been used as two objective functions to get new optimal ratios for
minimum pressure drop and minimum cut-off diameter using the multi-objective genetic algorithm optimization
technique. Sometimes, the main concern is minimizing the pressure drop, so a single objective optimization study
has been performed to obtain the cyclone geometrical ratio for minimum pressure drop. The comparison of numerical simulation of the new optimal design and the Stairmand design conrms the superior performance of
the new design.
2011 Elsevier B.V. All rights reserved.

1. Introduction
Cyclones are one of the most widely used separators, which rely
on centrifugal forces to separate particles from a gas stream. The primary advantages are the economy, simplicity in construction and
adaptability to a wide range of operating conditions.Reversed ow
cyclones with a tangential inlet are the most common cyclone design
[1](Fig. 1). It consists of seven main geometrical parameters: inlet
section height a and width b, cylinder height h, cyclone total height
Ht, dust exit diameter (cone tip diameter) Bc, gas outlet diameter
(also, called the vortex nder diameter) Dx and vortex nder height
S. All these parameters always given as a ratio of the cyclone body diameter D. It is generally known that these seven dimensions characterize the collection efciency (cut-off diameter) and pressure drop
of the cyclone separator [2-5]. Both the pressure drop and the cutoff diameter in a cyclone separator can be decreased or increased by
varying the cyclone dimensions. For an accurate optimal design of a
cyclone, it is quite necessary to use a reliable model for its performance parameters. Optimization of gas cyclone is, indeed, a multi-

Corresponding author. Tel.: + 32 26 29 2368; fax: +32 26 29 2880.


E-mail address: kelsayed@vub.ac.be (K. Elsayed).
0032-5910/$ see front matter 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.powtec.2011.10.015

objective optimization problem rather than a single objective optimization problem that has been considered so far in the literature [5,6].
Both the pressure drop and the collection efciency in gas cyclones
are important objective functions to be optimized simultaneously in
such a real-world complex multi-objective optimization problem
[7]. These objective functions are either obtained from experiments,
empirical models or computed using very timely and high-cost computational uid dynamic (CFD) approaches. Modeling and optimization of the parameters are investigated in the present study, by
using radial basis function articial neural networks and multiobjective genetic algorithm optimization technique in order to maximize the collection efciency (minimize the cut-off diameter) and
minimize the pressure drop.
1.1. Previous optimization studies
In 1951, Stairmand [8] presented the geometrical ratios for high efciency cyclones. Until now, these ratios are still in use (cf. Fig. 1 and
Table 1). Elsayed and Lacor [3] reported the following shortages in the
Stairmand model for pressure drop calculation [9] which has been
used to obtain these geometrical ratios: (1) the velocity distribution
has been obtained from a moment-of-momentum balance, estimating
the pressure drop as entrance and exit losses combined with the loss

K. Elsayed, C. Lacor / Powder Technology 217 (2012) 8499

(a) 3-D view

85

(b) 2-D view

Fig. 1. Schematic diagram for cyclone separator in 3-D and 2-D.

of static pressure in the swirl, i.e., neglecting the entrance loss by assuming no change of the inlet velocity occurs at the inlet area; (2) assuming
a constant friction factor; (3) the effect of particle mass loading on the
pressure drop is not included [10].
Due to the wide range of industrial applications of cyclone separators, they are already the subject of detailed studies for several decades. However, optimization studies are quite limited in the
literature. Moreover, many of these studies are not coherent studies.
Ravi et al. [11] carried out a multi-objective optimization of a set of
N identical reverse-ow cyclone separators in parallel by using the
non-dominated sorting genetic algorithm (NSGA). Two objective
functions were used: the maximization of the overall collection efciency and the minimization of the pressure drop. Non-dominated
Pareto optimal solutions were obtained for an industrial problem in
which 165 m 3/s of air was treated. In addition, optimal values of several decision variables, such as the number of cyclones and eight geometrical parameters of the cyclone, are obtained. Their study shows
that the barrel diameter, the vortex nder diameter, and the number
of cyclones used in parallel, are the important decision variables
inuencing the optimal solutions. Moreover, their study illustrates
the applicability of NSGA in solving multi-objective optimization
problems involving gas-solid separations. The main drawbacks of
their study are: (1) They used the model of Shepherd and Lapple
[12] for predicting the dimensionless pressure drop (Euler number).
In Shepherd and Lapple model, the Euler number depends on only
three factors (Eu = 16ab/Dx2) and they used it to optimize the seven
geometrical parameters. (2) The barrel diameter, number of parallel
cyclones and the gas velocity have been included into the optimization design space. Consequently, it is not devoted to the geometrical
ratio. (3) They used many side constraints on the geometrical values
(0.4 a/D S/D, 0.15 b/D (1 Dx/D)/2 if 0.5 Dx/D 0.6) these
constraints prevent searching for the global optimization geometrical
ratios for the seven geometrical parameters. (4) No table for the nondominated Pareto front points is presented from which the designer
can select a certain geometrical ratio set (optimal solution).

Table 1
The geometrical parameters values for Stairmand design.
Cyclone

a/D

b/D

Dx/D

Ht/D

h/D

S/D

Bc/D

Stairmand design

0.5

0.2

0.5

1.5

0.5

0.375

Swamee et al. [13] investigated the optimum values of the number


of cyclones to be used in parallel, the diameter of cyclone barrel D and
exit pipe Dx, when a specied ow rate of gas is to be separated from
solid particles, and the cut-off diameter is already specied. They
used Stairmand model for calculation of pressure drop and Gerrard
and Liddle formula for the cut-off diameter [13] which is not a widely
used model. Instead of handling two objective functions, they blended
the two objectives into a single objective problem which is not a suitable
method when considering two conicting objectives (the pressure drop
and cut-off diameter).
Sakhani et al. [14] performed a multi-objective optimization of
cyclone separators. First, they simulated many cyclones to obtain
the pressure drop and the cut-off diameter and then used the articial
neural network approach to obtain the objective function values. Finally, multi-objective genetic algorithms are used for Pareto based
optimization of cyclone separators considering two conicting objectives. However, the design variables are only four (instead of seven):
the barrel height, the cone height, the vortex nder diameter and
length. So they ignored the effect of inlet dimensions, which has
been acknowledged by other researchers as signicant geometrical
parameters for the cyclone ow eld and performance (cf. Elsayed
and Lacor [3-5,7]). Moreover, they did not explain why they selected
these particular parameters. Furthermore, they applied four side constraints on the four tested variables, which prevents searching for the
global optimization.
Pishbin and Moghiman [15] applied the genetic algorithm for optimum cyclone design. They studied the seven geometrical parameters.
The data used for optimization was obtained from 2-D axisymmetric
simulations. However, the ow in the cyclone separator is 3-D unsteady.
Instead of using multi-objective genetic algorithm (e.g., non-dominated
sorting genetic algorithm II (NSGA-II) [16]) they used the weighted-sum
genetic algorithm. In this technique, a weighting factor is assigned for
each objective function based on the user preference. The main shortages of the Pishbin and Moghiman [15] study are: (1) How to select
the weighting factor, in scientic and engineering problems, it is a
non-trivial task to nd the one solution of interest to the decision
maker [17]. The decision maker's weight (no matter how dened)
could be greater than necessary as more acceptable solutions are missed.
Optimizing mostly prot could lead to poor quality or reliability, not a
good compromise [17]. The weighted-sum genetic algorithm usually
does not nd all Pareto front points of interest. But this approach is a
simple approach for handling multi-objective optimization problem,

86

K. Elsayed, C. Lacor / Powder Technology 217 (2012) 8499

another simple but better result can be obtained using the desirability
function approach [7,18]. (2) No table for the non-dominated Pareto
front points is presented from which the designer can select a certain
geometrical ratio.
Sakhani et al. [19] carried out a multi-objective optimization
using the genetic algorithm technique to obtain the best vortex nder
dimension (diameter and length) and shape (convergent and divergent). Four design variables have been investigated; the vortex nder
diameter, angle, upper-part length and lower-part length of the vortex nder. They applied neural networks to obtain a meta-model for
the pressure drop and collection efciency from CFD dataset. The
main shortages of the Sakhani et al. [19] study are: (1) They used dimensional values instead of dimensionless, and applied side constraints, which prevent the optimization procedure from obtaining
global optimization. (2) The selection of only the vortex nder dimension as the design variables and neglecting the interaction with
the vortex nder diameter with the other dimensions, especially the
inlet dimensions [4,5].
Elsayed and Lacor [3,5,7,20-22] have performed several optimization
studies (single and multi-objective) on cyclone separator geometry.
They presented new designs for optimum performance, but in all of
their studies, they used the NelderMead technique [23] which suffers
from two drawbacks, (1) the nal solution may be affected by the start
point, (2) the obtained optimum may be a local minimum. To avoid
this, the application of evolutionary methods like the genetic algorithm
is a must.
1.2. Study objectives
There are four objectives of this study. (1) Investigation of the effect of the seven geometrical parameters on the cyclone separator
performance (the pressure drop and cut-off diameter) based on the
experimental data for the pressure drop and the most robust mathematical models for the cut-off diameter. (2) Study the possible interaction between the seven geometrical parameters affecting the
cyclone performance using response surface methodology.
(3) Multi-objective optimization to obtain new geometrical ratios
for optimum performance (minimum pressure drop and minimum
cut-off diameter). (4) Obtaining the optimum design (geometrical ratios) of the cyclone separator for minimum pressure drop using the
genetic algorithm optimization technique, followed by a comparison
of the numerical simulations of the optimal design and the Stairmand
design using the Reynolds stress turbulence model.
2. Articial Neural Network (ANN) approach
With the development of modern computational technologies, articial neural networks (ANNs) have become an attractive approach
for modeling highly complicated and nonlinear system [5,24,25]. The
most widely used types of ANNs for solving the regression problem
are back propagation neural network (BPNN), radial basis function
neural network (RBFNN) and generalized regression neural network
(GRNN). Zhao and Su [25] conducted a detailed comparison between
these three articial neural networks. They developed and employed
these ANNs to model the Euler number for cyclone separators. The
main conclusions of Zhao and Su are: (1) The BPNN results are close
to that of the RBFNN in the testing process but BPNN takes a larger
computational cost. (2) The GRNN has fast conversion but the predicted result for the tested sample was not satisfactory. (3) Compared
with the BPNN and GRNN, the RBFNN provides superior prediction
and a high robustness. In this study, the radial basis function neural
network (RBFNN) has been used to model the effect of cyclone dimensions on the cyclone performance parameters. For more details about
the radial basis function neural network, the interested reader can
refer to Elsayed and Lacor [5].

2.1. The Euler number


The pressure drop across the cyclone essentially depends on the
cyclone dimensions and operating conditions. Generally, it is proportional to the average dynamic pressure at the inlet and is often dened as [25]
P Eu



1
2
g V in
2

Where Eu is the Euler number (the dimensionless pressure drop


also called pressure drop coefcient [25]).
In order to determine the Euler number more accurately, all eight
dimensions of the cyclone are selected to establish the ANN models.
They all affect the Euler number but to different extents [25,26]. Usually, these dimensions can be characterized by the barrel diameter D
and expressed as seven dimensionless geometric ratios [25]:
Eu f



Dx a b S H t h B c
; ; ; ; ; ;
D D D D D D D

According to Eq. (2), seven independent dimensionless geometrical variables and one dependent variable (the Euler number of the cyclone) are selected as respectively the input and output parameters in
theANN model. For simplicity, the division of each factor by the barrel
diameter D will be dropped.
A dataset of 98 samples obtained from the measurements of pressure drop for different cyclone designs [25-27] is used in the present
investigation to evaluate the prediction performance of the ANN
models. Table 2 presents more details about the used dataset including the minimum, mean, maximum, range and the standard deviation
of the seven dimensionless geometrical ratios. Due to the large difference in the order of magnitude of the value (cf. Table 2), the available
dataset is transformed into 1 to 1 interval using the Matlab intrinsic
function; mapminmax in order to avoid solution divergence [25], cf.
Elsayed and Lacor [5] for more statistical details for the used dataset.
The ANN calculations have been performed using the neural network
toolbox available from Matlab commercial software 2010a.
2.2. The cut-off diameter (Stokes number)
The source of the training data for the cut-off diameter has been
obtained from the application of Iozia and Leith model [28]. This
model has been approved as an acceptable approach for calculating
the cyclone cut-off diameter [28,29]. The cut-off diameter x50 is the
particle diameter which produces 50% collection efciency.
A question may appear here, why the authors employed the cutoff diameter instead of the collection efciency, like in the study of
Ravi et al. [11]. Firstly, For low mass loading cyclone separator, the
cut-off diameter can replace the collection efciency, since one can
t the grade efciency curve using the cut-off diameter via some correlations, cf. Hoffmann and Stein ([10], page 91) for more details.
Moreover, many models can predict well the cut-off diameter but exhibit different grade efciency curves ( [10], page 97). Secondly, the
cut-off diameter (instead of the collection efciency) has been used
as an objective function in many recent publications e.g. [3,14]. Moreover, the selection of the cut-off diameter or the collection efciency
for low mass loading cyclones can be considered as a researcher
choice.
Based on Iozia and Leith model [28], the cut-off diameter is a function of the inlet gas velocity (i.e., a function of both gas volume ow
rate, Barrel diameter, inlet section height and width), gas viscosity
and particle density.
The cut-off diameter x50 for a cyclone separator is always given in
units of m. Another way to represent x50 is using a dimensionless
number, the Stokes number. The Stokes number based on the cut-

K. Elsayed, C. Lacor / Powder Technology 217 (2012) 8499

87

Table 2
Descriptive statistical parameters for the seven geometrical factors used to train the RBFNN.
Variable

Dx

Ht

Bc

Minimum
Mean
Maximum
Range
Standard deviation

0.25
0.428653
0.667
0.417
0.110117

0.113
0.629653
1.0
0.887
0.261828

0.067
0.21148
0.4
0.333
0.0935936

0.39
0.89101
3.052
2.662
0.428851

1.158
3.28309
10.97
9.812
2.09556

0.501
1.18908
3.5
2.999
0.672901

0.14
0.341847
1.0
0.86
0.149841

off diameter; Stk50 p x250 V in =18D[30]. It is the ratio between the


2
/(18) and the gas ow integral time
particle relaxation time; px50
scale; D/Vin.
2.3. Fitting the ANNs
Tables 3 and 4 present more details about the validation of the used
RBFNNs. Both the average, standard deviation, minimum, maximum
and range of the input and the predicted values are given. It is clear
from Tables 3 and 4 that the ANNs preserved the descriptive statistical
parameters of the input data. The correlation coefcient between the
input and the output and the mean squared error are given for each
RBFNN. The intercept and the slope of the adjusted line between the
input and the predicted value of the ANN are also given.
The congured RBFNN predictions versus experimental data for
the Euler number and the Iozia and Leith mathematical model for
the Euler number and cut-off diameter are shown in Fig. 2. According
to Fig. 2, it can be seen that the ANN models are able to attain the high
training accuracy. The training mean square errors for the experimental values and the Iozia and Leith model have the values 1.311E-4 and
3.258E-4 respectively (Fig. 2). This indicates that, compared with traditional models of curve tting, the models based on an articial intelligence algorithm have a superior capability of nonlinear tting.
Especially, the RBFNN has its unique and optimal approximation
characteristics in the learning process [25].
Fig. 2 illustrates the agreement between the ANNs input and output. The obtained relation is a typical linear relation with a coefcient
of correlation close to 1 (R > 0.999). The agreement between the
input and output of the ANN is also clear from the value of the
mean squared error E 2. That means, the trained neural networks predict very well both the Euler number and cut-off diameter values and
can be used in cyclone design and performance estimation. Tables 3, 4
and Fig. 2 present different performance indicators as a validation of
the proposed model for experimental values.

diameter Dx and the vortex nder length S, the inlet width b and the
total height Ht. Less effect is due to the cylinder height h (for h > 2.5)
and the inlet height a (for a > 0.55) (cf., Elsayed and Lacor [5]Fig. 6(a)).
For this particular study presented in Fig. 3, the following values
have been used: Barrel diameter D =0.1 m, air ow rate =0.8333 l/
s, air viscosity 1.0E-5 Pa s and particle density 860 kg/m 3.This means
that the obtained results will be valid for this particular case. But,
the authors believe the variation of cut-off diameter due to variations
of cyclone geometrical dimensions is superior to the effect of these
operating parameters, which is quite difcult to cover their range of
operating conditions.
The effects of the geometrical parameters on the cut-off diameter
are depicted in Fig. 3. To study the effect of each parameter, the tested
RBFNN model has been used by varying one parameter at a time from
its minimum to maximum values of the available 98 dataset, while the
other parameters are kept constant at their mean values (cf. Table 2).
Fig. 3 indicates the signicant effect of the vortex nder diameter Dx
and the vortex nder length S, the inlet width b, the inlet height a
and the total height Ht. Less effect is due to the cylinder height h and
the cone tip diameter Bc. More analysis is given in Table 5.
2.5. The signicant geometrical parameters on the cut-off diameter (Stokes
number) using the response surface methodology (RSM) approach

The effect of the geometrical parameters on the Euler number


based on the trained RBFNN has been investigated by Elsayed and
Lacor [5]. They acknowledge the signicant effect of the vortex nder

In this study only the effect of the geometry was taken into account. The effect of ow rate on the performance was not considered.
Overcamp and Scarlett [33] studied the effect of changing Reynolds
number on the cut-off diameter (Stokes number) and found that for
Reynolds number values beyond 1E4, the effect of increasing Reynolds
number is very limited. Furthermore, Karagoz and Avci [34] studied
the effect of increasing the Reynolds number on the pressure drop
and found that beyond Reynolds number of 2E4 any increase in the
Reynolds number has nearly no effect on the pressure drop. As the
Reynolds number for all cases considered is above 2E4. (All the tested
cyclones have the same ow rate.), the effect of ow rate can safely be
neglected. As this optimization study does not include changing the
cyclone diameter or the number of cyclones to get the optimum cyclone diameter. So no need to add the capital cost of the cyclone separator as a design parameter in this study [13]. Consequently, only
the effect of changing the geometrical parameters on the performance
will be considered.

Table 3
Validation of the used RBFNN to model the Euler number.

Table 4
Validation of the used RBFNN to model the cut-off diameter.

2.4. The effect of geometrical parameters on the cut-off diameter based


on RBFNN

Average
Standard deviation
Minimum
Maximum
Range
Correlation Coefcient, R
Mean squared error, E2
Intercept
Slope

23.2684
32.8858
2.3
155.3
153.0
0.99964
1.311E-4
0.0167
0.999

23.2684
32.874
1.74544
155.985
154.24

Both x and y represent the Euler number.


The values of R, E2, intercept and slope are that for the testing stage.
x is the input to the RBFNN and y is the predicted value.

Average
Standard deviation
Minimum
Maximum
Range
Correlation Coefcient, R
Mean squared error, E2
Intercept
Slope

8.21939
2.55998
3.64
15.3
11.66
0.99915
3.258E-4
0.014
0.999

8.21939
2.5578
3.7157
15.4048
11.6891

Both x and y represent the cut-off diameter.


The values of R, E2, intercept and slope are that for the testing stage.
x is the input to the RBFNN and y is the predicted value.

88

K. Elsayed, C. Lacor / Powder Technology 217 (2012) 8499

(a) The Euler number

Table 5
The variation of the cut-off diameter with cyclone dimensions using the RBFNN model
(cf., Fig. 3).

Predicted value (y)

Data point
Linear fit

Factor

120

y = 0.999 x + 0.0167
R = 0.99964
2
E = 1.311E-4

80

40

40

80

120

Input value (x)

(b) The cut-off diameter


Data point
Linear fit

Predicted value (y)

14

y = 0.998 x + 0.014
R=0.99915

12

E2 = 3.258E-4

10
8

Due to the interaction between the geometrical parameters, especially between Ht


with S and Dx (cf., Fig. 5 and Table 7), the obtained conclusions may not be applicable
generally and the application of response surface methodology to analysis the effect of
each particular parameter must.

6
4
4

Analysis

The vortex nder diameter has the most signicant effect on the cut-off
diameter x50 (the highest slope in Fig. 3). The slope is very high until
Dx = 0.5 and any further increase in Dx produces a small change in x50. In
general, increasing Dx increases x50 (decreasing the collection efciency),
this is one of the main reasons of the trade-off between the Euler number
and the cut-off diameter objectives. This makes the optimization of cyclone geometry a multi-objective procedure.
b
The variation of x50 with the inlet width is similar in trend and
signicance to that for Dx but here the slope changes at b = 0.25.
S and a The effect of the vortex nder length and the inlet section height on the
cut-off diameter is almost paralleled up to S = 1.5 and a = 0.6 afterwards
they lose their signicance and become nearly constant.
h
Increasing the barrel height slightly decreases the cut-off diameter with
nearly linear relation. This trend has been reported by other researchers
using CFD simulations, e.g., Elsayed and Lacor [31].
The effect of the cyclone total height is basically due to two effects the
Ht
cone height and barrel height. The curve can be subdivided into four main
regions. Sharp decrease in x50 up toHt = 3.25, no valuable difference
between 3.25 and 5.25, sharp increase between 5.25 and 8, and
insignicant effect beyond 8.
The effect of the cone-tip diameter on the cut-off diameter is quite small.
Bc
First, increasing the cone-tip diameter slightly decreases the cut-off diameter up to Bc = 0.55 and any further increment increases the cut-off
diameter. This trend has been reported by other researchers, e.g., Elsayed
and Lacor [22,32].
Dx

10

12

14

Input value (x)


Fig. 2. Linear regression of the RBFNNs for the Euler number and the cut-off diameter.

2.6. Why RSM?


The usual method of optimizing any experimental set-up is to adjust
one parameter at a time, keeping all others constant, until the optimum
working conditions are found. Adjusting one parameter at a time is necessarily time consuming, and may not reveal all interactions between

Ht
2

10

h, S
0.5

1.5

2.5

3.5

14

the parameters. In order to fully describe the response and interactions


of any complex system a multivariate parametric study must be conducted [35]. As there are seven geometrical parameters to be investigated, the best technique is to perform this study via the response surface
methodology (RSM). Response surface methodology has been used to
prepare a new design of experiment (due to the correlation between
the 98 input data set used for training of the ANNs, cf. Elsayed and
Lacor [5]).
In order to conduct a RSM analysis, one must rst design the experiment, identify the experimental parameters to adjust, and dene
the process response to be optimized. Once the experiment has been
conducted and the recorded data tabulated, RSM analysis software
models the data and attempts to t second-order polynomial to this
data [35]. The generalized second-order polynomial model used in
the response surface analysis was as follows [3,5]:

12

Y 0

i1

X50 [micron]

10

6
Dx
a
b
S
Ht
h
Bc

4
2

0.2

0.4

0.6

0.8

i X i

7
X
i1

ii X i ij X i X j
ibj

where 0, i, ii, and ij are the regression coefcients for intercept,


linear, quadratic and interaction terms, respectively. While Xi and Xj
are the independent variables, and Y is the response variable (Stokes
number Stk50 ).

7
X

Dx, a, b, Bc
Fig. 3. The effect of geometrical parameters on the cut-off diameter based on the Iozia
and Leith model [28]. Note: The plotted curves are obtained for a test case with the following settings, Barrel diameter = 0.1 m, air ow rate = 0.8333 l/s, air viscosity = 1.0E5 Pa s, particle density = 860 kg/m3.

2.6.1. Design of experiment (DOE)


The statistical analysis is performed through three main steps. Firstly,
construct a table of runs with a combination of values of the independent variables via the commercial statistical software STATGRAPHICS
centurion XVI by giving the minimum and maximum values of the
seven geometrical factors under investigation as input. Secondly, perform the runs by estimating the cut-off diameter (Stokes number)
using the trained articial neural network (based on the Iozia and
Leith model [28]). Thirdly, ll in the values of the Stokes number in
the STATGRAPHICS worksheet and obtain the response surface equation

K. Elsayed, C. Lacor / Powder Technology 217 (2012) 8499

with main effect plot, interaction plots, Pareto chart and response surface plots (we refer to Antony [36] chapter 4 for more details about
the denition of these plots and how they have been calculated).
Table 6 represents the parameters ranges selected for the seven
geometrical parameters. The study was planned using BoxBehnken
design, with 64 combinations. A signicant level of P b 0.05 (95% condence) was used in all tests. Analysis of variance (ANOVA) was followed by an F-test of the individual factors and interactions [5].
2.7. Analysis of variance (ANOVA)
Analysis of variance (ANOVA) showed that the resultant quadratic
polynomial models adequately represented the experimental data
with the coefcient of multiple determination R 2 being 0.984099
(cf., Table 7). This indicates that the quadratic polynomial model
obtained was adequate to describe the inuence of the independent
variables studied [37]. Analysis of variance (ANOVA) was used to
evaluate the signicance of the coefcients of the quadratic polynomial models (see Table 7). For any of the terms in the models, a
large F-value (small P-value) would indicate a more signicant effect
on the respective response variables [3,5].
Based on the ANOVA results presented in Table 7, the variable
with the largest effect on the Stokes number (cut-off diameter) was
the linear term of vortex nder diameter, followed by the linear
term of the cyclone total height, the vortex nder length and the
inlet width (P b 0.05); the other three linear terms (inlet height, barrel
height, and cone tip diameter) did not show a signicant effect
(P > 0.05). The quadratic term of cyclone total height, vortex nder
diameter and vortex nder length also had a signicant effect on
the pressure drop; however, the effect of the other four quadratic
terms was insignicant. Furthermore, the interaction between Dx
with (Ht, S, b) and between S with (h, Ht) also had a signicant effect
on the Stokes number, while the effect of the remaining terms was
insignicant.
2.8. Analysis of response surfaces
For visualization of the calculated factor, main effects plot, Pareto
chart and response surface plots were drawn. The slope of the main
effect curve is proportional to the size of the effect and the direction
of the curve species a positive or negative inuence of the effect
[3,38] (Fig. 4(a)). Based on the main effect plot, the most signicant
factors on the Stokes number are: (1) the vortex nder diameter Dx,
with a secondorder curve of direct relation. (2) the cyclone total
height Ht inversely related to the Stokes number. (3) the vortex nder
length S with direct relationship. (4) the inlet dimensions width b and
height a inversely related to the Stokes number. Whereas the other
factors have an insignicant effect. The main effect plot supports the
analysis given in Table 5, except for Ht where the strong interaction
between the cyclone total height and the vortex nder length affected
the trend given in Fig. 3.
Pareto charts were used to summarize graphically and display the
relative importance of each parameter with respect to the Stokes
number [3]. The Pareto chart shows all the linear and second-order
effects of the parameters within the model and estimates the
Table 6
The values of the independent variables used in the design of experiment.
Variables

Minimum

Center

Maximum

Inlet height, a = X1
Inlet width, b = X2
Cone tip diameter, Bc = X3
Vortex nder diameter, Dx = X4
Barrel height, h = X5
Total cyclone height, Ht= X6
Vortex nder length, S = X7

0.4
0.14
0.2
0.2
1.0
3.0
0.4

0.55
0.27
0.3
0.475
1.5
5.0
1.2

0.7
0.4
0.4
0.75
2.0
7.0
2.0

89

Table 7
Analysis of variance and the regression coefcients of the tted quadratic equation for
the Stokes numbera.
Source

Regression coefcient

0.0470554

F-ratio

P-value

Linear
1
2
3
4
5
6
7

0.942933
1.3178
2.10188
8.3493
0.843633
0.527695
1.46453

6.65
24.12
0
1145.05
0
243.98
137.75

0.0154
0.0000
1
0.0000
1
0.0000
0.0000

Quadric
11
22
33
44
55
66
77

1.17696
1.44904
3.50314
2.33086
0.281211
0
0.243505

0.21
0.18
0.37
9.38
1.49
39.06
7.33

0.6489
0.6737
0.5476
0.0048
0.232
0.0000
0.0114

Interaction
12
13
14
15
16
17
23
24
25
26
27
34
35
36
37
45
46
47
56
57
67
R2

0.726218
0
2.09219
0
0.121646
0.187708
0
4.84457
0
0.262495
0.466053
0
0
0
0
0
0.869946
1.44156
0
0
0.370128
0.984099

0.03
0
1.25
0
0.22
0.09
0
5.03
0
0.78
0.39
0
0
0
0
0
38.41
16.87
0
39.06
58.83

0.8558
1
0.2732
1
0.6401
0.7726
1
0.0330
1
0.3842
0.5352
1
1
1
1
1
0.0000
0.0003
1
0.0000
0.0000

a
Bold numbers indicate signicant factors as identied by the analysis of variance
(ANOVA) at the 95% condence level.

signicance of each with respect to maximizing the Stokes number


response. A Pareto chart displays a frequency histogram with the
length of each bar proportional to each estimated standardized effect
[35]. The vertical line on the Pareto charts judges whether each effect
is statistically signicant within the generated response surface
model; bars that extend beyond this line represent effects that are
statistically signicant at a 95% condence level. Based on the Pareto
chart (Fig. 4(b)) and ANOVA table (Table 7) there are ve signicant
parameters at a 95% condence level: the vortex nder diameter Dx,
the total cyclone height Ht, the vortex nder length S and the inlet dimensions a and b. Therefore, the Pareto chart is a perfect supplement
to the main effect plot.
To visualize the effect of the independent variables on the dependent ones, surface response of the quadratic polynomial models were
generated by varying two of the independent variables within the experimental range while holding the other factors at their central values
(cf., Table 6) [37] as shown in Fig. 5. Thus, Fig. 5(a) was generated by
varying the total height Ht and the vortex nder length S while keeping
the other ve factors constant. The response surface plots presented in
Fig. 5 illustrate the strong interactions between Ht with (S and Dx) and
Dx with (S and b).

90

K. Elsayed, C. Lacor / Powder Technology 217 (2012) 8499

(a) Ht versus S

(a) Main effects plot

(b) Dx versus Ht
(b) Paretochart. A=a,B=b, C=Bc, D=Dx, E=h, F=Ht, G=S,
FG=Ht S, etc.

(c) Dx versus S

(d) b versus Dx
Fig. 4. Analysis of design of experiment for the Stokes number.

3. Optimization using genetic algorithms


The genetic algorithm is an optimization technique for solving
both constrained and unconstrained optimization problems that is
based on natural selection, the process that drives biological evolution
[39]. The genetic algorithm repeatedly modies a population of individual solutions. At each step, the genetic algorithm selects individuals
at random from the current population to be parents and uses them to
produce the children for the next generation. Over successive generations, the population evolves toward an optimal solution. The genetic
algorithm can be used to solve a variety of optimization problems
that are not well suited for standard optimization algorithms, including problems in which the objective function is discontinuous, no differentiable, stochastic, or highly nonlinear [39].
In case of cyclone separator geometry optimization for minimum
Euler number and minimum cut-off diameter, the objectives are conicting with each other. There is no solution for which all objectives
are optimal simultaneously [40]. The increase of one objective will
lead to the decrease of other objectives. Then, there should be a set
of solutions, the so-called Pareto optimal set or Pareto front, in
which one solution cannot be dominated by any other member of
this set [40].
Recently, a number of multi-objective genetic algorithms (MOGAs)
based on the Pareto optimal concept have been proposed. The well
known nondominated sorting genetic algorithm II (NSGA-II) proposed by Deb et al. [16] is one of the most widely used MOGAs since

Fig. 5. The response surface plots for the Stokes number. Note: the stokes number
values are multiplied by 1000.

it provides excellent results as compared with other multi-objective


genetic algorithms proposed [41].
4. Results and discussion
4.1. Optimal cyclone design for minimum pressure drop
The genetic algorithm optimization technique has been applied to
obtain the geometrical ratios for minimum pressure drop (Euler
number). The objective function is the Euler number (using the
trained radial basis function neural network). The design variables
are the seven geometrical dimensions of the cyclone separator (cf.,
Fig. 1(a)).
4.1.1. GA settings
Table 8 presents the settings used to obtain the optimum design
for minimum pressure using global optimization Matlab toolbox in

K. Elsayed, C. Lacor / Powder Technology 217 (2012) 8499


Table 8
Genetic operators and parameters for single objective optimization.
Population type:
Initial range:
Fitness scaling:
Selection operation:
Elite count:
Crossover fraction:
Crossover operation:
Mutation operation:
Maximum number of generations:
Population size:

Double vector
[0.2 0.1 0.1 0.3 2.0 0.65 0.05 ;
0.8 0.8 0.6 0.8 8.0 2.5 0.75]
Rank
Tournament (tournament size equals 4)
2
0.8
Intermediate crossover with the default
value of 1.0
The constraint dependent default
1400
200

Matlab 2010a commercial package. Table 9 gives the optimum values


for cyclone geometrical parameters for minimum pressure drop estimated by the articial neural network using the genetic algorithm optimization technique. It is clear from Table 9 that the new optimized
design is very close to the Stairmand design in many geometrical parameters, whereas the new ratios will result in minimum pressure
drop. To understand the effect of this small change in the geometrical
ratios on the ow eld pattern and performance, a CFD study for the
two designs is needed [5].
4.2. Comparison between the two cyclone designs using CFD

91

Table 10
The values of geometrical parameters for the two designs (D = 0.205 m).
Cyclone

a/D

b/D

Dx/D

Ht/D

h/D

S/D

Bc/D

Stairmand design
New design

0.5
0.595

0.2
0.201

0.5
0.549

4
4.549

1.5
1.411

0.5
0.595

0.375
0.275

The outlet section is above the cyclone surface by Le = 0.618D. The inlet section located at a distance Li = D from the cyclone center.

the axial and tangential velocity in addition to the pressure drop


and cut-off diameter, we refer the reader to Elsayed and Lacor [4, 5].
4.2.2. The pressure eld
Fig. 6 shows the contour plot at Y = 0. In the two cyclones, the
time-averaged static pressure decreases radially from the wall to center. A negative pressure zone appears in the forced vortex region
(central region) due to high swirling velocity. The pressure gradient
is largest along the radial direction, while the gradient in the axial direction is very limited. The cyclonic ow is not symmetrical as is clear
from the shape of the low-pressure zone at the cyclone center (twisted
cylinder). However, the two cyclones have almost the same ow pattern, but the highest pressure of the Stairmand design is nearly 1.5
times that of the new design, implying that the new design has a
lower pressure drop.
The pressure distribution presented in Figs. 7 and 8 of the two cyclones at sections S1S6 depict the two parts pressure prole (for

4.2.1. Solver settings


The air volume ow rate Qin=0.08 m 3/s for the two cyclones
(inlet velocity for Stairmand design is 19 m/s and 16 m/s for the
new design), air density 1.0 kg/m 3 and dynamic viscosity of 2.11E5 Pa s. the turbulent intensity equals 5% and characteristic length
equals 0.07 times the inlet width [42]. Velocity inlet boundary condition is applied at inlet, outow at gas outlet and wall boundary condition at all other boundaries [5].
The nite volume method has been used to discretize the partial
differential equations of the model using the SIMPLEC (Semi-Implicit
Method for Pressure-Linked Equations-Consistent) method for pressure velocity coupling and QUICK scheme to interpolate the variables
on the surface of the control volume. The implicit coupled solution algorithm was selected. The unsteady Reynolds stress turbulence
model (RSM) was used in this study with a time step of 0.0001 s.
The grid renement study using different levels of grid shows that
a total number of 134759 hexahedral cells for the Stairmand cyclone
and 378963 hexahedral cells for the new design are sufcient to obtain a grid-independent solution, and further mesh renement yields
only small, insignicant changes in the numerical solution. The hexahedral meshes have been obtained using the GAMBIT commercial
software. These simulations were performed on an 8 nodes CPU
Opteron 64 Linux cluster using FLUENT 6.3.26 commercial software.
The geometrical values for the two cyclones (cf. Fig. 1(b)) are given
in Table 10.
In order to validate the obtained results, the predictions are compared with the measurements of Hoekstra [43] on the Stairmand cyclone using Laser Doppler Anemometry (LDA). For the validation of
Table 9
The optimized cyclone separator design for minimum pressure drop.
Factor

Low

High

Stairmand design

Optimum design

Dx
a
b
S
Ht
h
Bc

0.2
0.5
0.14
0.4
3.0
1.0
0.2

0.75
0.75
0.4
2.0
7.0
2.0
0.4

0.5
0.5
0.2
0.5
4.0
1.5
0.375

0.549
0.595
0.201
0.595
4.549
1.411
0.275

Fig. 6. The contour plots for the time averaged ow variables at sections Y = 0 (cf. Fig. 1
(b)). From top to bottom: Stairmand design and the new design respectively. From left
to right: the static pressure (N/m2), the tangential and axial velocity (m/s). Note: both
cyclones have the same barrel diameter and air volume ow rate.

K. Elsayed, C. Lacor / Powder Technology 217 (2012) 8499

Table 11
The position of different sectionsa.
Section

S1

S2

S3

S4

S5

S6

z`/Db

2.75

2.5

2.25

2.0

1.75

1.5

Sections S1S5 are located in the conical section, section S6 at the cylindrical part.
z is measured from the top of the inlet section (cf. Fig. 1(b)).

Rankine vortex). Again, the highest static pressure for Stairmand design is more than 1.5 times that of the new design at all sections while
the central value is almost the same for the two cyclones irrespective
of the section location. This indicates that, the new design has a lower
pressure drop with respect to the Stairmand design.

Static pressure (N/m2)

2500

New design
Stairmand design

2000

1500

1000

500

0
-0.1

-0.05

0.05

0.1

Tangential velocity/ Inlet velocity

4.2.3. The velocity eld


Based on the contour plots of the time-averaged tangential velocity
presented in Fig. 6, and the radial proles at sections S1S6 shown in

1500

1000

500

0.05

Tangential velocity/ Inlet velocity

Static pressure (N/m2)

New design
Stairmand design

-0.05

2
1.5
1
0.5
0
-0.1

1000

500

0.05

Distance from center (m)

0.1

Tangential velocity/ Inlet velocity

Static pressure (N/m2)

1500

0.05

-0.5
-0.1

0.1

1.5
1
0.5
0
0

0.05

2.5

1.5
1
0.5
0
0

0.05

Distance from center (m)

0.05

0.1

New design
Stairmand design

0.5

-0.5
-0.1

-0.05

0.05

0.1

Distance from center (m)

New design
Stairmand design

-0.05

0.1

-0.1

-0.05

Distance from center (m)

New design
Stairmand design

-0.05

New design
Stairmand design

0.5

Distance from center (m)

New design
Stairmand design

-0.05

-0.1

0.1

2000

0
-0.1

-0.05

2.5

Distance from center (m)


2500

New design
Stairmand design

Distance from center (m)

2000

0
-0.1

The discrete phase model in Fluent follows the EulerLagrange approach. The uid phase is treated as a continuum by solving the timeaveraged NavierStokes equations, while the dispersed phase is
solved by tracking a large number of particles through the calculated
ow eld. A fundamental assumption made in this model is that the
dispersed second phase occupies a low volume fraction (usually less
than 1012%, where the volume fraction is the ratio between the

2.5

Distance from center (m)


2500

4.3. Discrete phase modeling (DPM)

Axial velocity/ Inlet velocity

0.1

Axial velocity/ Inlet velocity

Figs. 7 and 8, the following conclusions can be drawn. The tangential velocity distributions for the two cyclones are approximately nearly identical in pattern and values (dimensionless), with the highest velocity
occurring at 1/4 of the cyclone radius for both cyclones. This implies
nearly equal collection efciency for both cyclones, as the centrifugal
force is the main driving force for particle collection in the cyclone separator. The axial velocity proles for the two cyclones are also very close,
exhibiting the inverted W axial velocity prole.

Axial velocity/ Inlet velocity

92

New design
Stairmand design

0.5

-0.5
-0.1

-0.05

0.05

0.1

Distance from center (m)

Fig. 7. The radial prole for the time averaged tangential and axial velocity at different sections on the XZ plane (Y = 0) at sections S1S3 (cf., Table 11). From top to bottom: section
S1S3. From left to right: time-averaged static pressure, tangential and axial velocity respectively.

2000

1500

1000

500

-0.05

0.05

0.1

New design
Stairmand design

2
1.5
1
0.5
0
-0.1

Static pressure (N/m2)

2500

New design
Stairmand design

2000

1500

1000

500

0
-0.1

-0.05

0.05

2.5

1.5
1
0.5
0
-0.1

1000

500

0.05

Tangential velocity/ Inlet velocity

Static pressure (N/m2)

1500

-0.05

0.05

2.5

1.5
1
0.5
0

Distance from center (m)

-0.05

0.05

0.05

0.1

New design
Stairmand design

0.5

-0.5
-0.1

-0.05

0.05

0.1

Distance from center (m)

New design
Stairmand design

-0.1

0.1

0.1

-0.05

Distance from center (m)

Distance from center (m)

New design
Stairmand design

-0.05

-0.5
-0.1

0.1

0.1

2000

0
-0.1

0.05

New design
Stairmand design

Distance from center (m)


2500

New design
Stairmand design

0.5

Distance from center (m)


Tangential velocity/ Inlet velocity

Distance from center (m)

-0.05

Axial velocity/ Inlet velocity

0
-0.1

2.5

Axial velocity/ Inlet velocity

New design
Stairmand design

93

Axial velocity/ Inlet velocity

Static pressure (N/m2)

2500

Tangential velocity/ Inlet velocity

K. Elsayed, C. Lacor / Powder Technology 217 (2012) 8499

0.1

New design
Stairmand design

0.5

-0.5
-0.1

Distance from center (m)

-0.05

0.05

0.1

Distance from center (m)

Fig. 8. The radial prole for the timeaveraged tangential and axial velocity at different sections on the XZ plane (Y = 0) at sections S4S6 (cf., Table 11). From top to bottom: section S4S6. From left to right: time-averaged static pressure, tangential and axial velocity respectively.

total volume of particles and the volume of uid domain), even


though high mass loading is acceptable. The particle trajectories are
computed individually at specied intervals during the uid phase
calculation. This makes the model appropriate for the modeling of

Table 12
The performance parameters for the two cyclones.

Table 13
Genetic operators and parameters for multi-objective optimization.

Design

Method

p
(N/m2)

Euler number

x50
(m)

Stokes
number 103

New design

CFD
ANN
Ramachandran
model [26]
CFD
ANN
Ramachandran
model [26]

803
584.4
877.98

6.338
4.613
5.523

0.804
2.938

0.114
1.815

1190
1015.8
699.66

6.567
5.606
4.846

1.0
3.314

Stairmand
design

particle-laden ows. The particle loading in a cyclone separator is


small (35%), and therefore, it can be safely assumed that the presence of the particles does not affect the ow eld (one-way coupling).
Collection efciency statistics were obtained by releasing a specied

0.209
1.931

Population type:
Population size:
Initial range:
Selection operation:
Crossover fraction:
Crowding distance fraction
Crossover operation:
Number of generations (iterations):

Double vector
105 (i.e., 15* number of variables)
[0.2 0.1 0.1 0.3 2.0 0.65 0.05 ;
0.8 0.8 0.6 0.8 8.0 2.5 0.75]
tournament (tournament size equals 2)
0.8
0.35
Intermediate crossover with the default
value of 1.0
1400 (i.e., 200* number of variables)

K. Elsayed, C. Lacor / Powder Technology 217 (2012) 8499

Table 14
The diameters, air ow rates and the particle densities for the sixteen test cases.
Case

D [mm]

Q [l/min]

p [kg/m3]

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

205
205
205
205
205
205
205
205
31
31
31
31
31
31
31
31

50
60
70
80
50
50
50
50
50
60
70
80
50
50
50
50

860
860
860
860
1000
1500
1750
2000
860
860
860
860
1000
1500
1750
2000

number of mono-dispersed particles at the inlet of the cyclone and by


monitoring the number escaping through the outlet. Collisions between particles and the walls of the cyclone were assumed to be

(a)

Case 1
Case 2
Case 3
Case 4
Case 5
Case 6
Case 7
Case 8
Case 9
Case 10
Case 11
Case 12
Case 13
Case 14
Case 15
Case 16

Stokes number x 103

3.5
3
2.5
2
1.5

4.3.1. The DPM results


In order to calculate the cut-off diameters of the two cyclones, 10 4
particles were injected from the inlet surface with zero velocity and a
_ of 0.001 kg/s (corresponding to inlet dust
particles mass ow rate m

p
_ p =Q 11:891 gm=m3 ). The particle density p
concentration C in m
is 860 kg/m 3 and the maximum number of time steps for each injection was 200000 steps. The DPM analysis results and the pressure
drops for the two cyclones are depicted in Table 12. Although, the difference between the two cyclone cut-off diameters is small, the saving in pressure drop is considerable (nearly 32.5% the value of
Stairmand cyclone).
Based on the ow pattern analysis and the DPM results, one can
conclude that the cyclone collection efciency for the two cyclones
is very close, with the advantage of low pressure drop in the new design. The authors want to emphasis that only small changes in the
geometrical dimensions of the two designs lead to this improvement
in the performance.

(c)

Pareto chart for 16 test cases, linear scale

perfectly elastic (coefcient of restitution is equal to 1). For the equation of particle motion and the DPM settings, the interested reader
can refer to Elsayed and Lacor [5].

Pareto chart for test case 1

3.5

Stokes number x 103

94

3
2.5
2
1.5

0.5

0.5

10

15

20

25

10

Euler number

15

20

25

Euler number

(b) Pareto chart for 16 test cases in log scale with the curve fitting

(d)

Pareto chart for test case 9

formula, Stk50 = 10 0.3016(log10 (Eu))2 0.9479log10 (Eu) 2.5154,


R2 = 0.98643
Pareto points
Stk50=10^{0.3016[LOG10(Eu)]^2-0.9479 LOG10(Eu)-2.5154}

3.5

Stokes number x 103

Stokes number x 103

7
6.5
6
5.5
5
4.5
4
3.5
3
2.5
2
1.5

3
2.5
2
1.5

1
1

0.5
0.5
5

Euler number

10

15 20 25 3035

10

15

Euler number

Fig. 9. Pareto charts for different test cases.(Continued) Pareto charts for different test cases.

20

25

K. Elsayed, C. Lacor / Powder Technology 217 (2012) 8499

4.4. Optimal cyclone design for best performance

95

of minimum Euler number and maximum Stokes number. Point B indicates the point of maximum Euler number and minimum Stokes
number. Point C indicates an optimal point for the multi-objective optimization problem.
In order to obtain the Euler numberStokes number relationship,
Fig. 9(b) has been drawn. It indicates a general relationship (trend)
between the two dimensionless numbers irrespective to the barrel diameter, gas ow rate, particle density. A second-order polynomial has
been tted between the logarithms of Euler number and Stokes number, Eq. (4). The obtained correlation can t the data with a coefcient
of correlation R 2 = 0.98643 as shown in Fig. 9(b).

4.4.1. NSGA-II settings


Table 13 presents the genetic operators and parameters for multiobjective optimization. The Euler number values have been obtained
from the articial neural network trained by experimental values. The
Stokes number values are obtained from Iozia and Leith model [28]. In
order to investigate the effect of different geometrical and operational
parameters on the Pareto front, sixteen test cases with different barrel diameter, gas ow rate and particle density have been tested, cf.
Table 14. The sixteen test cases covers: 1) Two barrel diameters,
31 mm and 205 mm. 2) Four levels of air ow rates, 50, 60, 70 and
80 l/min. 3) Five values of particle density, 860, 1000, 1500, 1750
and 2000 kg/m 3.

Stk50 10

4.4.2. Pareto front


The Pareto front (non dominated points) for the sixteen test cases
are presented in Fig. 9(a). Fig. 9(a) clearly demonstrate tradeoffs in
objective functions Euler number and Stokes number from which an
appropriate design can be compromisingly chosen by the designer
[14]. All the optimum design points in the Pareto front are nondominated and could be chosen by a designer as optimum cyclone
separator [14]. The corresponding geometrical ratios of the Pareto
front shown in Fig. 9(a) are given in Table 15 for test case 1 and
Table 16 for test case 9. Three points A, B and C are indicated in
Figs. 9(c) and (c) and Tables 15 and 16. Point A indicates the point

4.4.3. Bubble plots for Pareto front


For visual inspection of the effect of the seven geometrical parameters on the two conicting performance parameters, the bubble plots on
Pareto front points have been drawn for each geometrical parameter.
However, only gures for test case 1 (Fig. 10) and 9 (Fig. 11) are presented, but all other cases depict the same results (trend).
Fig. 10 indicates that: a) Decreasing the vortex nder diameter Dx
decreases the Stokes number and increases the Euler number, Fig. 10
(b). b) Generally speaking, increasing the inlet height a increases the
Euler number and decreases the Stokes number. c) A similar trend is

0:3016log 10 Eu2 0:9479log 10 Eu2:5154

Table 15
The seven geometrical parameters and the obtained Euler number and Stokes number for the nondominated points (Pareto-front) for test case 1 (cf. Table 14).
Point

Dx

Ht

Bc

Euler number

Stokes number 103

1
2B
3
4
5C
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36A
37
Minimum
Maximum

0.306
0.306
0.618
0.326
0.360
0.670
0.585
0.306
0.306
0.305
0.303
0.476
0.312
0.308
0.592
0.320
0.598
0.309
0.430
0.306
0.306
0.326
0.308
0.680
0.307
0.307
0.514
0.310
0.559
0.399
0.516
0.413
0.306
0.307
0.646
0.692
0.558
0.303
0.692

0.659
0.688
0.229
0.331
0.295
0.230
0.226
0.685
0.666
0.516
0.286
0.240
0.622
0.686
0.229
0.570
0.229
0.441
0.307
0.667
0.392
0.331
0.686
0.229
0.653
0.368
0.235
0.596
0.246
0.308
0.244
0.256
0.507
0.644
0.229
0.228
0.216
0.216
0.688

0.385
0.398
0.213
0.240
0.253
0.213
0.217
0.299
0.361
0.271
0.318
0.217
0.277
0.379
0.213
0.273
0.213
0.266
0.231
0.389
0.277
0.209
0.348
0.213
0.296
0.338
0.217
0.275
0.221
0.229
0.241
0.280
0.352
0.333
0.219
0.213
0.230
0.209
0.398

0.410
0.404
0.411
0.451
0.443
0.410
0.419
0.401
0.407
0.439
0.449
0.434
0.419
0.427
0.413
0.423
0.412
0.450
0.438
0.417
0.441
0.451
0.427
0.409
0.421
0.439
0.429
0.424
0.453
0.438
0.422
0.444
0.437
0.423
0.409
0.408
0.439
0.401
0.453

6.957
6.993
6.774
6.668
6.664
6.777
6.763
6.985
6.965
6.840
6.611
6.743
6.943
6.974
6.831
6.898
6.770
6.769
6.735
6.950
6.708
6.668
6.974
6.787
6.972
6.708
6.709
6.918
6.785
6.707
6.728
6.667
6.811
6.960
6.797
6.819
6.802
6.611
6.993

1.779
1.779
1.885
1.885
1.910
1.859
1.901
1.789
1.782
1.868
1.930
1.901
1.909
1.781
1.895
1.909
1.893
1.881
1.898
1.779
1.882
1.900
1.807
1.857
1.816
1.895
1.901
1.899
1.898
1.891
1.904
1.927
1.848
1.807
1.863
1.855
1.922
1.779
1.930

0.387
0.317
0.495
0.444
0.459
0.496
0.492
0.425
0.398
0.428
0.462
0.471
0.429
0.459
0.494
0.434
0.494
0.435
0.465
0.367
0.443
0.471
0.459
0.495
0.420
0.430
0.479
0.429
0.455
0.460
0.486
0.472
0.460
0.401
0.497
0.498
0.486
0.317
0.498

23.843
27.322
1.026
6.485
4.892
0.879
1.159
18.474
22.300
12.977
8.404
1.915
14.816
23.145
1.117
12.767
1.095
10.629
3.211
24.941
9.963
5.515
21.132
0.849
17.606
11.399
1.596
14.387
1.486
3.732
1.797
3.510
16.110
19.929
0.963
0.815
1.311
0.815
27.322

0.552
0.540
2.785
0.798
0.937
3.478
2.089
0.576
0.558
0.633
0.699
1.524
0.619
0.554
2.491
0.658
2.565
0.674
1.229
0.550
0.676
0.823
0.564
3.640
0.588
0.659
1.718
0.622
1.915
1.102
1.674
1.162
0.602
0.575
3.113
3.805
1.945
0.540
3.805

A
indicates the point of minimum Euler number and maximum Stokes number. B indicates the point of maximum Euler number and minimum Stokes number. C indicates an optimal
point for the multi-objective optimization problem. (cf. Fig. 9(c)).

96

K. Elsayed, C. Lacor / Powder Technology 217 (2012) 8499

Table 16
The seven geometrical parameters and the obtained Euler number and Stokes number for the nondominated points (Pareto-front) for test case 9 (cf. Table 14).
point

Dx

Ht

Bc

Euler number

Stokes number 103

1A
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26B
27
28
29C
30
31
32
33
34
35
36
37
Minimum
Maximum

0.686
0.308
0.308
0.585
0.314
0.309
0.578
0.309
0.310
0.473
0.308
0.432
0.321
0.308
0.308
0.407
0.309
0.452
0.678
0.625
0.309
0.313
0.664
0.309
0.608
0.308
0.309
0.308
0.330
0.658
0.308
0.520
0.631
0.549
0.316
0.686
0.380
0.308
0.686

0.236
0.655
0.692
0.235
0.330
0.486
0.237
0.272
0.538
0.238
0.375
0.238
0.327
0.606
0.643
0.255
0.333
0.236
0.236
0.237
0.596
0.325
0.237
0.365
0.235
0.692
0.436
0.427
0.290
0.245
0.362
0.251
0.236
0.242
0.270
0.236
0.239
0.235
0.692

0.236
0.390
0.390
0.235
0.236
0.335
0.235
0.347
0.369
0.235
0.382
0.236
0.236
0.383
0.381
0.240
0.353
0.236
0.236
0.236
0.379
0.262
0.236
0.345
0.235
0.390
0.381
0.360
0.252
0.239
0.365
0.244
0.236
0.240
0.341
0.236
0.237
0.235
0.390

0.478
0.423
0.423
0.418
0.431
0.426
0.421
0.424
0.429
0.478
0.423
0.445
0.431
0.424
0.424
0.470
0.424
0.448
0.473
0.449
0.425
0.430
0.476
0.425
0.438
0.423
0.424
0.424
0.465
0.477
0.424
0.470
0.449
0.430
0.425
0.478
0.477
0.418
0.478

6.910
6.902
6.902
6.925
6.837
6.869
6.684
6.897
6.882
6.923
6.901
6.785
6.839
6.896
6.895
6.899
6.882
6.793
6.893
6.800
6.894
6.847
6.911
6.876
6.921
6.902
6.899
6.886
6.890
6.910
6.883
6.906
6.796
6.909
6.891
6.910
6.890
6.684
6.925

1.915
1.995
1.995
1.727
1.864
1.931
1.975
1.967
1.984
1.971
1.989
1.947
1.862
1.990
1.986
1.907
1.958
1.945
1.919
1.944
1.988
1.884
1.914
1.946
1.797
1.995
1.989
1.962
1.955
1.924
1.668
1.684
1.945
1.783
1.966
1.915
1.922
1.668
1.995

0.473
0.471
0.472
0.471
0.470
0.471
0.471
0.472
0.467
0.471
0.471
0.472
0.470
0.471
0.472
0.472
0.471
0.472
0.473
0.472
0.471
0.471
0.473
0.471
0.472
0.472
0.464
0.471
0.471
0.473
0.467
0.472
0.472
0.471
0.472
0.473
0.473
0.464
0.473

1.008
21.688
22.898
1.355
6.568
13.947
1.356
8.025
16.799
2.105
12.168
2.494
6.223
19.736
20.775
3.119
10.021
2.255
1.029
1.187
19.039
7.197
1.076
10.740
1.261
22.898
14.120
13.112
5.639
1.146
11.924
1.996
1.164
1.613
7.482
1.008
3.327
1.008
22.898

3.743
0.558
0.551
2.436
0.750
0.621
2.008
0.697
0.597
1.481
0.634
1.300
0.777
0.571
0.565
1.165
0.665
1.394
3.615
2.891
0.578
0.730
3.404
0.656
2.691
0.551
0.616
0.626
0.818
3.290
0.648
1.671
2.964
1.823
0.726
3.743
1.071
0.551
3.743

A
indicates the point of minimum Euler number and maximum Stokes number. B indicates the point of maximum Euler number and minimum Stokes number. C indicates an optimal
point for the multi-objective optimization problem. (cf. Fig. 9(d)).

exhibited by the inlet width b but due to interaction with other geometrical and operational variables, one could see a range of bubble
sizes in the region of best performance (lower values for both the
Euler and Stokes numbers). d) The higher values of total cyclone
height Ht will produce less Stokes number, intermediate values
could produce less Euler number, smaller-intermediate values could
produce the optimum performance due to interaction with other variables. e) Short barrels will produce better collection efciency (low
Stokes number) and higher Euler numbers. Intermediate values results in low Euler number values. Long barrels can produce the best
performance. f) Short vortex nder may produce higher values of
Euler numbers or higher values of Stokes number due to strong interaction with other variables. Long vortex nder can produce the optimum performance. g) Generally speaking, the variation of the conetip diameter Bc has no effect on the performance parameter. The
above comments is restricted to the range of each geometrical variables located on the Pareto front and not for the whole range of values
(cf., Fig. 10 for the range of each geometrical parameters).

5. Conclusions
To predict the complex non-linear relationships between the performance parameters and the geometrical dimensions, two radial basis
neural networks (RBFNNs) are developed and employed to model the
Euler number and Stokes number for cyclone separators. The neural

networks have been trained and tested by the experimental data available in literature for Euler number (pressure drop) and Iozia and Leith
model [28] for the Stokes number (cut-off diameter). The effects of
the seven geometrical parameters on the Stokes number have been investigated using the trained ANN. To declare any interaction between
the geometrical parameters affecting the Stokes number, the response
surface methodology has been applied. The trained ANN has been
used as an objective function to obtain the cyclone geometrical ratios
for minimum Euler number using the genetic algorithms optimization
technique. A CFD comparison between the new optimal design and
the Stairmand design using the Reynolds stress turbulence model has
been performed. A multi-objective optimization technique using
NSGA-II technique has been applied to determine the Pareto front for
the best performance cyclone separator.
The following conclusions can be drawn from analysis of the
obtained results:
The result demonstrates that articial neural networks can offer an alternative and powerful approach to model the cyclone performance.
The analysis indicates the signicant effect of the vortex nder diameter Dx and the vortex nder length S, the inlet width b, the
inlet height a and the total height Ht on the cyclone performance.
The response surface methodology has been used to t a secondorder polynomial to the RBFNN for the cut-off diameter. The analysis of variance of the cut-off diameter indicates a strong interaction
between Dx with (Ht, S, b) and between S with (h, Ht).

K. Elsayed, C. Lacor / Powder Technology 217 (2012) 8499

(a) Dx, range: 0.303 - 0.692

(b) a, range: 0.216 - 0.688

(c) b, range: 0.209 - 0.398

(d) Ht, range: 6.611 - 6.993

(e) h, range: 1.779 - 1.930

(f) S, range: 0.401 - 0.453

(g) Bc, range: 0.317 - 0.498

Fig. 10. Bubble plots for different geometrical parameters for test case 1 (cf. Fig. 9(c) and Table 15).

97

98

K. Elsayed, C. Lacor / Powder Technology 217 (2012) 8499

(a) Dx, range: 0.308 - 0.686

(b) a, range: 0.235 - 0.692

(c) b, range: 0.235 - 0.390

(d) Ht, range: 6.684 - 6.925

(e) h, range: 1.668 - 1.995

(f) S, range: 0.418 - 0.478

(g) Bc, range: 0.464 - 0.473

Fig. 11. Bubble plots for different geometrical parameters for test case 9 (cf. Fig. 9(d) and Table 16).

K. Elsayed, C. Lacor / Powder Technology 217 (2012) 8499

The trained RBFNN has been used to get a new optimized cyclone
for minimum pressure drop (Euler number) using the genetic algorithm optimization technique.
A comparison between the new design and the standard Stairmand
design has been performed using CFD simulation with the Reynolds
stress turbulence model to get a clear vision of the ow eld pattern
and performance in the new design.
CFD results shows that, the new cyclone design are very close to the
Stairmand high efciency design in the geometrical parameter ratio,
and superior in low pressure drop at nearly the same cut-off diameter.
The new cyclone design results in nearly 68% of the pressure drop
obtained by the old Stairmand design at the same volume ow
rate. This conrms that the obtained design using the genetic algorithm is better than that obtained using NelderMead technique
which results in 75% of the Stairmand pressure drop [5].
The two trained RBFNNs have been used in a multi-objective optimization process using NSGA-II technique. Sixteen test cases with
different barrel diameter, gas ow rate and particle density have
been tested and plotted. The Pareto fronts for the 16 test cases are
very close. A second-order polynomial has been tted between
the logarithms of Euler number and Stokes
number to obtain a gen2
eral formula, Stk50 100:3016log10 Eu 0:9479log10 Eu2:5154 with a coefcient of correlation R 2 = 0.98643. This formula can be used to
obtain the Stokes number if the Euler number is known.
As a future extension of this study, the following issues still need
more investigation. (1) Comparison between the suggested geometrical parameters ratios and the other cyclone designs available in literatures e.g. Elsayed and Lacor design [3] (they used MM model for the
estimation of the pressure drop). (2) Create a neural network model
as a design tool for the cyclone separator and estimate its performance
parameters. (3) Generate performance curves for each geometrical
and operating parameters that affect the cyclone performance to
help the designer in predicting the change of the performance due to
change in the cyclone geometrical ratios and operating conditions.
(4) Perform a Robust parameter design study to investigate the effect
of uncertainty in the geometrical values and the optimization process
Acknowledgments
Part of this study has been published in Evolutionary and Deterministic Methods for Design, Optimization and Control (Eurogen
2011) conference, Italy 2011.
References
[1] F. Kaya, I. Karagoz, A. Avci, Effects of surface roughness on the performance of tangential inlet cyclone separators, Aerosol Science and Technology 45 (8) (2011) 988995.
[2] R. Xiang, K.W. Lee, Exploratory study on cyclones of modied designs, Particulate
Science and Technology 19 (4) (2001) 327338.
[3] K. Elsayed, C. Lacor, Optimization of the cyclone separator geometry for minimum
pressure drop using mathematical models and CFD simulations, Chemical Engineering Science 65 (22) (2010) 60486058.
[4] K. Elsayed, C. Lacor, The effect of cyclone inlet dimensions on the ow pattern and
performance, Applied Mathematical Modelling 35 (4) (2011) 19521968.
[5] K. Elsayed, C. Lacor, Modeling, analysis and optimization of aircyclones using articial neural network, response surface methodology and CFD simulation approaches,
Powder Technology 212 (1) (2011) 115133.
[6] H. Sakhani, M.A. Akhavan-Behabadi, N. Nariman-Zadeh, M.J.M. Abadi, Modeling
and multi-objective optimization of square cyclones using CFD and neural networks, Chemical Engineering Research and Design 89 (3) (2011) 301309.
[7] K. Elsayed, C. Lacor, Multi-objective optimization of gas cyclone based on CFD simulation, ECCOMAS thematic conference, CFD & Optimization, Antalya, Turkey, 2011.
[8] C.J. Stairmand, The design and performance of cyclone separators, Industrial and
Engineering Chemistry 29 (1951) 356383.
[9] C.J. Stairmand, Pressure drops in cyclone separators, Industrial and Engineering
Chemistry 16 (B) (1949) 409411.
[10] A.C. Hoffmann, L.E. Stein, Gas Cyclones and Swirl Tubes: Principle, Design and Operation, 2nd Edition Springer, 2008.
[11] G. Ravi, S.K. Gupta, M.B. Ray, Multiobjective optimization of cyclone separators
using genetic algorithm, Industrial and Engineering Chemistry Research 39
(2000) 42724286.

99

[12] C.B. Shepherd, C.E. Lapple, Flow pattern and pressure drop in cyclone dust collectors cyclone without intel vane, Industrial and Engineering Chemistry 32 (9)
(1940) 12461248.
[13] P.K. Swamee, N. Aggarwal, K. Bhobhiya, Optimum design of cyclone separator,
American Institute of Chemical Engineers (AIChE ) 55 (9) (2009) 22792283.
[14] H. Sakhani, A. Hajiloo, M. Ranjbar, N. Nariman-Zadeh, Modeling and multiobjective optimization of cyclone separators using CFD and genetic algorithms,
Computers and Chemical Engineering 35 (6) (2011) 10641071.
[15] S.I. Pishbin, M. Moghiman, Optimization of cyclone separators using genetic algorithm,
International Review of Chemical Engineering (I.RE.CH.E.) 2 (6) (2010) 683690.
[16] K. Deb, A. Pratap, S. Agarwal, T. Meyarivan, A fast and elitist multiobjective genetic
algorithm: NSGA-II, IEEE Transaction Evolitionery Computation 6 (2002) 182197.
[17] C.A.C. Coello, G.B. Lamont, D.A.V. Veldhuizen, Evolutionary Algorithms for Solving
Multi-Objective Problems, 2nd Edition Springer, 2007.
[18] S.H.R. Pasandideh, S.T.A. Niaki, Multi-response simulation optimization using genetic algorithm within desirability function framework, Applied Mathematics
and Computation 175 (1) (2006) 366382.
[19] H. Sakhani, S. Nourbakhsh, N. Nariman-zadeh, Modeling and multi-objective optimization of cyclone vortex nder using CFD and neural networks, 2nd International
Conference on Engineering Optimization, September 69, 2010, Lisbon, Portugal,
2010, pp. 19.
[20] K. Elsayed, C. Lacor, Application of response surface methodology for modeling
and optimization of the cyclone separator for minimum pressure drop, Fifth European Conference on Computational Fluid Dynamics (ECCOMAS CFD 2010), Lisbon, Portugal, 2010.
[21] K. Elsayed, C. Lacor, Optimization of the cyclone separator geometry for minimum
pressure drop based on articial neural network model and CFD simulation,
ECCOMAS thematic conference, CFD & Optimization, Antalya, Turkey, 2011.
[22] K. Elsayed, C. Lacor, Investigation of the geometrical parameters effects on the
performance and the ow-eld of cyclone separators using mathematical models
and large eddy simulation, 13th Aerospace Sciences & Aviation Technology
(ASAT-13), Military Technical College, Cairo, Egypt, 2009.
[23] J.A. Nelder, R. Mead, A simplex method for function minimization, The Computer
Journal 7 (4) (1965).
[24] L. Wang, C. Shao, H. Wang, H. Wu, Radial basis function neural networks-based
modeling of the membrane separation process: hydrogen recovery from renery
gases, Journal of Natural Gas Chemistry 15 (2006) 230234.
[25] B. Zhao, Y. Su, Articial neural network-based modeling of pressure drop coefcient for cyclone separators, Chemical Engineering Research and Design 88
(2010) 606613.
[26] G. Ramachandran, D. Leith, J. Dirgo, H. Feldman, Cyclone optimization based on a
new empirical model for pressure drop, Aerosol Science and Technology 15
(1991) 135148.
[27] J. Dirgo, Relationship between cyclone dimensions and performance, Ph.D. thesis,
Harvard University, USA (1988).
[28] D.L. Iozia, D. Leith, The logistic function and cyclone fractional efciency, Aerosol
Science and Technology 12 (3) (1990) 598606.
[29] D.L. Iozia, D. Leith, Effect of cyclone dimensions on gas ow pattern and collection
efciency, Aerosol Science and Technology 10 (3) (1989) 491500.
[30] J.J. Derksen, S. Sundaresan, H.E.A. van den Akker, Simulation of mass-loading effects in gassolid cyclone separators, Powder Technology 163 (2006) 5968.
[31] K. Elsayed, C. Lacor, A CFD study of the effect of cyclone barrel height on its performance parameters, 8th International Conference on Computational Fluid Dynamics in the Oil & Gas, Metallurgical and Process Industries, Trondheim,
Norway, 2011.
[32] K. Elsayed, C. Lacor, Numerical modeling of the ow eld and performance in cyclones of different cone-tip diameters, Computers & Fluids 51 (1) (2011) 4859.
[33] T.J. Overcamp, S.E. Scarlett, Effect of Reynolds number on the Stokes number of
cyclones, Aerosol Science and Technology 19 (3) (1993) 362370.
[34] I. Karagoz, A. Avci, Modelling of the pressure drop in tangential inlet cyclone separators,
Aerosol Science and Technology 39 (9) (2005) 857865.
[35] J.S. Cowpe, J.S. Astin, R.D. Pilkington, A.E. Hill, Application of response surface
methodology to laser-induced breakdown spectroscopy: inuences of hardware
conguration, Spectrochimica Acta Part B 62 (2007) 13351342.
[36] J. Antony, Design of Experiments for Engineers and Scientists, Elsevier Science &
Technology Books, 2003.
[37] Y. Yuan, Y. Gao, L. Mao, J. Zhao, Optimisation of conditions for the preparation of
-carotene nanoemulsions using response surface methodology, Food Chemistry
107 (2008) 13001306.
[38] M. Gfrerer, E. Lankmayr, Screening, optimization and validation of microwaveassisted extraction for the determination of persistent organochlorine pesticides,
Analytica Chimica Acta 533 (2005) 203211.
[39] Matlab, Global Optimization Toolbox User's Guide, Matlab R2010a, The MathWorks, Inc, 2010.
[40] X. Wang, C. Hirsch, S. Kang, C. Lacor, Multi-objective optimization of turbomachinery using improved NSGA-II and approximation model, Computer Methods
in Applied Mechanics and Engineering 200 (912) (2011) 883895.
[41] C. Coello, Twenty years of evolutionary multi-objective optimization: a historical
view of the eld, IEEE Computational Intelligence Magazine 1 (2006) 2836.
[42] A.J. Hoekstra, J.J. Derksen, H.E.A. Van Den Akker, An experimental and numerical
study of turbulent swirling ow in gas cyclones, Chemical Engineering Science 54
(1999) 20552065.
[43] A. J. Hoekstra, Gas ow eld and collection efciency of cyclone separators, Ph.D.
thesis, Technical University Delft (2000).

Вам также может понравиться