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

net/publication/322289513

CFD simulation of the preheater cyclone of a cement plant and the optimization
of its performance using a combination of the design of experiment and multi-
gene genetic programming

Article  in  Powder Technology · March 2018

DOI: 10.1016/j.powtec.2017.12.091

CITATIONS READS

17 6,969

3 authors, including:

Elham Kashani Ali Mohebbi

Shahid Bahonar University of Kerman Shahid Bahonar University of Kerman
4 PUBLICATIONS   19 CITATIONS    153 PUBLICATIONS   1,986 CITATIONS   

SEE PROFILE SEE PROFILE

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

Ion Exchange View project

Molecular Dynamics Simulation View project

All content following this page was uploaded by Ali Mohebbi on 11 January 2018.

The user has requested enhancement of the downloaded file.

 CFD simulation of the preheater cyclone of a cement plant and the

optimization of its performance using a combination of the design of

experiment and multi-gene genetic programming

Elham Kashani1, Ali Mohebbi1,, Mahdi Ghaedi Heidari2

1. Department of Chemical Engineering, Faculty of Engineering, Shahid Bahonar University of

Kerman, Kerman, Iran

2. Kerman cement plant, Kerman, Iran

Abstract

Hurriclon cyclone is a specially designed preheater cyclone with two outlet connector pipes of

cleaned gas in the cement industry. In Kerman cement plant, Iran, the initial structure of this

cyclone was changed. This caused a decrease in the cyclone efficiency. In this study, to optimize

the changed cyclone performance, one of the twin cyclones in the first-stage of the preheater

tower, which had the most significant effect on particle separation from gas was simulated and

validated by computational fluid dynamics. Using the design of experiment based on the

simulation results, the effects of three dimensions (vortex-finder length, cylinder height, and

cone tip diameter) were investigated on cyclone performance. The turbulent gas flow inside the

cyclone was modelled using the Reynolds stress model due to the swirling flow inside the

cyclones. The discrete phase model was used to calculate the trajectory of particles. It was

observed that because of high gas inlet velocity and particle density as well as the geometry of

the preheater cyclone, particles larger than the critical diameter continue spinning in the cyclone.


Corresponding author: Tel & Fax: +983432118298
E-mail addresses: amohebbi2002@yahoo.com, amohebbi@uk.ac.ir
The Multi-Gene Genetic Programming (MGGP) was used to obtain two equations for efficiency

and pressure drop in order to optimize the preheater cyclone performance. For this purpose, two-

objective optimization using the Genetic Algorithm (GA) was performed. The optimization

results showed that by using the optimized dimensions for the preheater cyclone, the pressure

drop decreases by 2.2% and the efficiency increases by 13.4%.

Keywords: Preheater cyclone; Cement plant; Computational fluid dynamics; Design of

experiment; Multi-gene genetic programming; Two-objective optimization

1. Introduction

The application of cyclone separators in separating particles from gas flow began hundreds of

years ago. Owing to the wide range and simplicity of the cyclone structure that leads to a low

cost of investment and maintenance, they are still widely used in the chemical and petrochemical

industries. The main parameters of cyclone performance are separation efficiency and pressure

drop [1]. Many efforts have been made to derive an appropriate mathematical model for the

study and prediction of flow behaviour as well as to find suitable cyclone geometries. However,

due to the hydrodynamic complexity of cyclones, no exact mathematical model for cyclones has

been offered yet. For numerical solutions, CFD is a proper tool [2, 3]. The key to the successful

use of CFD for finding turbulent flow pattern depends on the selection of the right turbulence

model. Selecting a suitable turbulence model has been studied by many researchers [2, 4]. The

result of the Reynolds stress model (RSM) among the turbulence models has a very good

agreement with the experimental data [5–8] since the isotropy of the flow assumption is ignored.

In addition, large-eddy simulation methodology can predict turbulent flow inside the cyclone [4,

9, 10].

The pressure drop and the cut-diameter in cyclones are important objective functions, to be

optimized simultaneously.Cyclone geometry has a significant effect on its performance; various

researches have been conducted to investigate the effect of geometrical parameters on cyclone

performance [11–13]. Safikhani et al. [14] obtained the pressure drop and cut-diameter by CFD

and then optimized the cyclone performance with the assistance of multi-objective optimization;

the artificial neural network was then used to calculate the objective function. Elsayed and Lacor

performed many studies in order to optimize the cyclone [15–17]. They used multi-objective

optimization for the performance of the cyclone [18]. Two radial-basis function neural networks
were used to model the pressure drop and the cut-diameter. Their results showed a noticeable

influence of the vortex finder diameter and its length, the inlet width, and the total height on the

efficiency of the cyclone. Fathizadeh et al. [19] used Leith and Licht theory to prepare some

algorithms to obtain the dimensions of the Stairmand cyclone. Vortex finder length was

optimized using genetic algorithm to prevent the direct escape of particles from vortex finder.

Owing to environmental concerns-and as the cement industry is one of the major producers of

greenhouse gases-strict regulations have been enforced in this industry. In recent years, different

researches on the various techniques that improve the cement production process and reduce

resultant pollution [20] have been carried out. Mikulcic et al. [21] showed the application of

CFD in the implementation of the calcination reaction, which is carried out in calciner in order to

reduce carbon dioxide emissions in the cement industry. Mikulcic et al. [22] numerically studied

gas-solid flows in a cement cyclone using the Eulerian-Lagrangian approach. The heat exchange

phenomenon in the cyclone, arising from the heat exchange between the gas-solid phase and the

endothermic calcination reaction, was investigated. Cristea and Conti [23] studied the heat

exchange between the gas and the solid phase in a cement plant preheater. They concluded that

the most of the heat distribution occurs within the duct, and the temperatures of the gas and

particles in the cyclones remain approximately constant and equal. Ferrite and Ari [9] optimized

the preheater cyclone by means of the Taguchi method. They optimized the cyclone performance

by changing the four parameters of diameter, vortex-finder length, velocity inlet, and the

concentration of particles. By modernizing the design of cyclone preheaters, significant energy

saving can be achieved. The role of the first-stage cyclones is significantly important as

specifying the separation efficiency. Wasilewski and Duda [24] for optimizing the first-stage

cyclones proposed guidelines depending on the objective function. In another study [25] they
optimized the geometry of cyclone. Mariani et al. [26] optimized the length and angle of vortex

finder to increase the separation efficiency and decrease the exit gas temperature.

2. Background

In traditional cement manufacturing technology, the preheater system is positioned before the

rotary kiln in order to increase the heat transfer between the raw materials and hot gases. Raw

materials in the preheater are gradually dried, heated, and calcined by moving against the flow of

the hot gases coming from the rotary kiln. In Fig. 1, a rotary kiln with a suspension preheater

similar to that one from Kerman cement plant is shown. Preheaters are built in several stages in

order to increase their heat efficiency (usually four to five stages).

Fig. 1

The preheater of the Kerman cement plant, Iran, was made by the PMT–Zyklontechnik Gmbh

Company (the company name was changed to 'A TEC' in 2005). The cyclone used by this

company is a special type named Hurriclon. Hurriclon has been patented for its technically

superior characteristic of the double immersion tube principle and its integrated vortex-finder

vanes. By adding vortex finder vanes, this company has reduced the cyclone pressure drop by

30%. The reason for the improvement of its performance is a result of collision prevention of the

cyclone outlet gases with the swirling mixture of gas and particles inside the cyclone [28].

However, after two years of preheater operation, both the vortex-finder vanes broke down, which

resulted in the plant staff having to make changes in the twin cyclones of the first-stage. They

removed the two vortex-finder vanes and the immersion tube that had been driven out from the
bottom. Fig. 2 (a) and (b) illustrate the Hurriclon cyclone before and after modernisation

respectively.

Since the structure of the preheater cyclones of the Kerman cement plant was significantly

varied, the purpose of this study was to simulate one of the twin cyclones in the first-stage of the

preheater by CFD. In order to increase the performance, a two-objective optimization study was

performed. The vortex-finder length, the cylindrical height, and the cone-tip diameter of this

cyclone were optimized by using the Genetic Algorithm (GA). In order to model the

relationships between the performance parameters (i.e., the pressure drop and the efficiency) and

the geometrical dimensions, multi-gene genetic programming (MGGP) was employed. The

effect of these dimensions on the performance was obtained using the design of experiments.

Fig. 2

3. The governing equations

For an incompressible and steady-state fluid flow, the continuity equation for the mean flow

is as follows [29]:

ui
0 (1)
xi

The time-averaged Navier–Stokes equation is as follows [29]:

 p    ui u j  2 uk    
x j
 ui u j       
xi x j   x j

xi
   ij



 uiu j 
 
(2)
 3 xk 
 x j

The last term is defined as the Reynolds stress tensor, which reflects the effects of turbulent

fluctuations in the fluid flow. The RSM used in this study is required for solving equations for
each of the Reynolds stresses. In the Reynolds stress model, the transport equation is written as

follows [29]:

  
  uk uiu j   Dij  Pij  ij   ij  S (3)
xk  

The expression on the left-hand side of the equation represents the convective transport term.

Five terms on the right-hand side of the equation are respectively diffusion, production, pressure

strain, dissipation, and the source terms. The final transport equations of RSM can be written as

follows [19]:

     t        u j u  u  u  u  u 
  uk uiu j     ui u j     uiuk  uj uk i   p( i  j )  2 i j (4)
xk   xk   k  xk    xk xk  x j xi xk xk

where the eddy viscosity t , is calculated by the equation below:

k2
t   C (5)


where, C is 0.09.

4. Simulation procedure

4.1. Cyclone geometry

The cyclone was a scroll type in which the radius of the cylindrical part (R) was 1.26 m and

the radius of the scroll part (Rs) was 1.722 m, which was located at a distance of  r  0.64 m from

the cylinder centre. Cyclone dimensions are provided in Table 1; it consists of cyclone diameter

(D), inlet height (a) and width (b), vortex-finder diameter (De) and length (S), the height of outlet

pipe (l) (it was selected greater than usual, in order to be sure of accessing fully developed flow

in the outlet and preventing recirculation flow), cylinder height (h), part of the scroll (hs) and
conical (Hc) height, the cyclone total height (H) and the cone-tip diameter (B). Fig. 3 shows a

schematic diagram for the cyclone.

Table 1

Hexahedral and tetrahedral meshes were generated by using the GAMBIT commercial

software. Fig. 4 shows simulated cyclone; Fig. 4(b) shows part of the cyclone, which is a gridded

tetrahedral. In this Figure, the Z-axis is in the axial (vertical) direction.

Fig. 3

Fig. 4

4.2. Selection of discretization schemes and boundary conditions

The velocity-inlet boundary condition for the inlet gas, the outflow for outlet gas, and the wall

condition for the outlet particles were considered. The atmospheric pressure was 81395 Pa, the

inlet gas volumetric flow rate was 28.22 m3/s, and the gas inlet velocity was 14.7 m/s. The

density and the viscosity of the gas were 0.43 kg/m3 and 2.1 × 10-5 kg/m.s respectively. The

density of particles was 2710 kg/m3. The turbulent intensity was 2.3% and the hydraulic diameter

was 1.3 m. The gas inlet temperature was 420 ᵒC; heat discussion had not been performed in this

research because the inlet gas temperature was lower than the decomposition temperature [22,

30]. This study was based on the Eulerian-Lagrangian approach using the discrete phase model

(DPM), so that while solving the time-averaged Navier-Stokes equations, the fluid phase was as

a continuous phase and the solid phase was as a discrete phase. More details about the equations

of motion of particle can be found elsewhere [19]. If we plot CD versus Re for Schiller and

Nauman [31], Schuh et al. [32] and Morsi and Alexander [33] on one figure, it is found that the

results are close to each other. Therefore, in this study, the drag coefficient was calculated
according to Morsi and Alexander [33] method. The second phase in DPM has to occupy a low

volume fraction (less than 10–12%). It was assumed that there was no particle-particle

interaction and the particles did not affect the flow field, which is known as one-way coupling

[29, 34]. To consider the effect of turbulence fluctuations on particles, the Discrete Random

Walk (DRW) model [35] was used.

Since the problem convergence with the RSM is difficult and time-consuming, at first the

flow field had been derived with the k–ε standard model and then changed to the RSM [36]. The

residuals for convergence of all terms were considered as 10-3.

The governing equations were solved numerically by the finite-volume method using the

commercial CFD code FLUENT 6.3.26. For pressure-velocity coupling, the SIMPLEC algorithm

was used. Because of highly swirling flows in the cyclone, PRESTO was used to interpolate the

pressure. To measure the flow in the cyclone, the momentum equations’ QUICK scheme was

recommended. The Reynolds stress equations were solved based on the second-order upwind

scheme discretization. In this study, for all the simulations, these discretization schemes were

used.

5. Multi-gene genetic programming

Genetic algorithm is a search technique in computer science used to find approximate

solutions for optimization. Genetic algorithms apply the principle of Darwin’s theory.

(According to Darwin’s theory, the generations that have better characteristics than other

generations will have a better chance of survival and reproduction; their superior characteristics

are also transmitted to the next generation). Genetic algorithm with random initial population

generation checks the problem space stochastically. Then, the fitness of each individual in a
population is investigated and the algorithm selects the fittest individuals, and then, genetic

operators affect some of the population and the next generation is generated. These operators are

selection, crossover, and mutation [37].

This study is performed to optimize the preheater cyclone of the Kerman cement plant since

this cyclone is a special case and there is no empirical relationship between its geometries and

performance (i.e., pressure drop and collection efficiency). Thus, a relationship was derived. In

order to achieve this purpose, Multi-Gene Genetic Programming (MGGP) was used. MGGP is a

modified version for model structure selection, which is combined with the least squares

technique for parameter estimations. MGGP is of the symbolic regression class aimed at finding

an explicit relationship between one or more inputs and one output by using mathematical

symbols, variables, and functions. To find equations in symbolic regression, a large and

continuously improving guided search has to be conducted in an evolving space [38].

First, a brief discussion on genetic programming (GP) is essential to comprehend the

performance of MGGP. GP provides potential solutions structurally on the basis of tree

representation, despite common optimization methods such as the genetic algorithm, where

potential solutions are presented with numbers. Each gene consists of a set of F functions and T

Terminals. F includes the basic arithmetic operators (such as −, +, exp, etc.) while the T

terminals involve input variables and constants. Potential solutions can be labelled as branches in

a labelled tree with ordered branches that function and are used in a function set with arguments

in the terminal set. The GP gives a platform with genetic operators such as random generation,

mutation, and crossover that produce, modify, and select individuals in a given population. Any

symbolic model of MGGP is a weighted linear combination of the output of some GP trees. Each
of these trees can be considered as a gene, which is a representative GP of a tree. More details

about MGGP can be found elsewhere [39, 40].

6. Results and discussion

6.1. The mesh independency and the validation of the model

The computational results of the three different meshes for the preheater cyclone pressure

drop and efficiency are provided in Tables 2 and 3 respectively. Since the pressure drop and

cyclone efficiency difference between both the sequential meshes were minimum (less than 5%),

the CFD results are independent of the mesh size. To reduce computational times, the mesh

number of 2208864 was used in this study. The last row in Tables 2 and 3 represent experimental

pressure drop and efficiency of the cyclone respectively. The experimental pressure drop was

measured by a portable device (testo 512) and of course, the monitoring of operating parameters

was available. The amount of dust returned from the cyclones (11 ton/h) could be weighed by

truck. Therefore, considering the amount of kiln feed (13.1 ton/h), the experimental efficiency

could be calculated. Since there was a little difference between simulation results and

experimental data, the solution was validated.

Table 2

Table 3

Fig. 5 (a) and (b) show the contours of static pressure and velocity magnitude in the sections

z=9.5 m and x=0 respectively. Pressure decreases from the wall to the centre and the area with

the lowest pressure and the highest velocity occurs below the vortex finder, which causes upward
suction at the centre that makes the outer vortex in the cyclone change its direction and move

upward. Thus, the inner vortex is created.

Fig. 5

In order to compare Stairmand and preheater cyclones, Stairmand cyclone was designed using

the algorithm written by Fathizadeh et al. [19]. It was found that the total height of the Stairmand

cyclone was greater than that of the old structure while the efficiency will increase 2.5%.

However, due to technical and economic reasons, preheaters tower should be designed with a

limited height.

6.2. Critical diameter

It was observed that when the inlet cyclone particle diameter becomes greater than a certain

value, the particles spin initially downward and then keep spinning at a certain height on a fixed

path near the wall. This value is called the critical diameter. This phenomenon was also reported

by Wang et al. [41]. A critical diameter of 55 μm was obtained for the simulated cyclone. Fig. 6

demonstrates particle spin with a diameter larger than the critical diameter (a) in operation [41]

and (b) in a simulated cyclone.

Fig. 6

In one-way coupling models, particles do not interact with each other, but due to particle

collisions, particles larger than the critical diameter are removed from the cyclone. At high

temperatures, if the particles stagnate on the wall, the permanent deposition, which decreases the

performance of the cyclone, may be formed. Furthermore, the avalanche phenomenon occurs due

to this behaviour [42].

 To find which parameters affect the critical diameter value, according to Table 4, eight

parameters were investigated. In each case, one parameter was examined while the remaining

parameters were taken at the same value, which were the parameters for preheater cyclone,

according to Table. 1 and section 4.2. In general, it can be concluded that by increasing inlet

velocity (U) and particle density (ρp), a reduction in critical diameter is expected, and with a

decrease in cylinder height (at constant total height) and an increase in the cone-tip diameter, the

critical diameter increases. Thus, by increasing the slope of the cyclone conical part (i.e.,

reducing the angle with the vertical axis, which is the Z-axis here), the critical diameter tends to

higher values.

Table 4

6.3. New cyclone simulation according to the design of experiments

Since the diameter of the cylindrical part, the vortex-finder diameter, the inlet dimensions and

the overall height of the preheater cyclone cannot be changed economically (changes of these

values will cause more changes in the entire system), the selected variables for the design of

experiments (DOE) were vortex-finder length, cylindrical height, and cone tip diameter. The

DOE method was used to investigate the effect of the mentioned dimensions on the performance

of the cyclone. DOE is a data collection process while one or more variables are changed. In

DOE, the designer is usually interested in investigating the effect of some parameters on the

output of a process. The analysis of variance determines the impact of each parameter on the

behaviour of the system; the appropriate method is selected depending on the limitations of a

problem [43]. In this study, the full factorial design method was used: all three mentioned

dimensions were studied in three levels, then, in accordance with the full factorial design
method, 3 × 3 × 3 simulation runs with at least two repetitions in the results were performed.

Table 5 gives three levels of the mentioned dimensions.

Table 5

The averaged pressure drop and efficiency for each simulation run are listed in Table 6. In the

table, the cyclone number is such that the first number represents the vortex-finder length, the

second number is the cylindrical height, and the third number is the cone-tip diameter; for

example, cyclone number 132 means: the vortex-finder length, cylindrical height, and cone-tip

diameter are 1.625, 4.553, and 0.628 m respectively. The rest of the main cyclone dimensions

remain constant, as given in Table 1. Owing to lack of experimental data, the Grid Convergence

Index (GCI) method [44] was used to verify the results of 27 cyclones. All the calculations were

carried out with an independent grid.

Table 6

The design of experiments was performed in the Minitab 16 software; the analysis of variance

(ANOVA) for pressure drop and efficiency are provided in Tables 7 and 8 respectively. A

significant level of P>0.05 (95% confidence) was used in all the tests. If the reported p-value for

any term is smaller than the confidence level, then it is said that the term has a significant effect

on the response [16]. According to two tables, all the three factors (dimensions) and their

interactions have a significant impact on the pressure drop and efficiency.

Table 7

Table 8

The impacts of factors (dimensions) on pressure drop and efficiency with the assistance of the

main-effect plot are shown in Fig. 7. The results were as follows:

a) when the factor S was changed from Level 1 to 3 (with increasing S), both the pressure drop

and the efficiency increased. By increasing the vortex-finder length, the probability of the input

particles exiting from the bottom of the vortex finder is lowered, and this leads to greater

efficiency.

b) when the factor h was changed from Level 1 to 3 (with increasing h), both the pressure drop

and efficiency decreased. By increasing cylindrical height (at constant total height), the conic

height decreased. Particles separation from the gas in the cyclone occurs more in the conical part,

where the radius is reduced. Therefore, by decreasing the height of the cone, the efficiency also

decreased.

c) when factor B was changed from 1 to 3 (by increasing B), both the pressure drop and

efficiency decreased. In Fig. 8, the velocity contours for three cyclones-121, 122, and 123-at

section x=0 are compared (the difference among these three cyclones is only in the cone-tip

diameter). As one can see from this figure, by increasing the diameter, the velocity of the

cyclone decreases. This causes a decrease in the efficiency. This finding is in accordance with

Ref. 45.

Fig. 7

Fig. 8

6.4. Modelling of the Euler number and efficiency using MGGP

In order to model the relationship between the performance of the cyclone and the studied

dimensions, MGGP was used from the toolbox of the computer program GPTIPS, which has

been developed under the MATLAB software. Euler number and efficiency as a function of

three independent dimensionless geometrical variables can be defined as follows:


 Eu  f ( S / D, h / D, B / D )

 2P
 Eu  (6)
  v2
 4Q
v 
  D2

 Ef

 f  S / D, h / D, B / D 
 (7)
 Ef   xii


where ΔP is the pressure drop and v is the velocity characteristic, xi is the mass fraction of the

particles with the same diameter and i is the collection efficiency for that diameter. The inlet

particle size distribution is given in Appendix A. According to Eqs. 6 and 7, the three above

mentioned variables and the two dependent variables (i.e., the Euler number and the efficiency)

respectively were considered as inputs and outputs of the models. A total of 22 data from 27 data

were used as the training set, while five data were used as the testing set. In this study, the

parameter settings applied in MGGP implementation are given in Table 9.

Table 9

The best results on predicting the Euler number and efficiency (based on MGGP) with

minimum errors are obtained from the following equations:

Eu  75.69 X1 X 3  5.918 X 2  54.64 X 3  5.69 X1  3.073X 2 X 3 15.33X12 X 3  93.67 (8)

Ef  2.28X13  (0.4903X 2  16.49) X12  (0.4903X 2  0.4903X 3  3.938) X1  4.108( X 3  X 2 )  81.72 (9)

where X1 = S/D, X2 = h/D and X3 = B/D. Figs. 9 and 10 show MGGP prediction versus the

calculated values by CFD for the Euler number and efficiency respectively in the training and the
testing data set. As one can see from these figures, the linear correlation coefficients for the Euler

number and efficiency are 99 and 99.98%, respectively, in the training data set. These figures

also show that these coefficients for the aforementioned parameters are 96.67 and 99.75%

respectively, in the testing data set. This indicates that the equations obtained have favourable

accuracy.

Fig .9

Fig. 10

6.5. Two-objective optimization

A two-objective optimization was used by the genetic algorithm to find the lowest pressure

drop and the highest efficiency of the preheater cyclone. Two objective functions were defined:

Eq. 8 (Euler number) for minimizing and Eq. 9 (efficiency) for maximizing or the reciprocal of

Eq. 9 for minimizing. The Pareto front (non-dominated points) is presented in Fig. 11. Pareto

front represents the set of responses in the search space that are better than the other responses.

In other words, in the Pareto front, one solution cannot be dominated by any other member of

this set [18]. The final selection of the optimal response from the Pareto front results depends on

the awareness of the problem. The best answer choice obtained for dimensions, pressure drop

and efficiency is given in Table 10, which is the point c in Fig. 11. Table 11 lists the dimension

ratios of the optimized cyclone to the preheater cyclone.

Fig. 11

Table 10

Table 11
 The optimized cyclone was simulated again, and the pressure drop and the efficiency were

obtained as 548.42 Pa and 94.2%, respectively. Fig. 12 shows the comparison between the

tangential velocity of the optimized cyclone and the preheater cyclone at section z=7 m versus

the radius of the cyclones. Tangential velocity is the dominant component of gas flow in the

cyclone, which affects the centrifugal force; with an increase in velocity, the force increases.

Thus, by increasing tangential velocity, the efficiency increases. It is observed from Fig. 12 that

the maximum optimized cyclone tangential velocity is greater than the preheater cyclone;

therefore, greater efficiency for the optimized cyclone is expected. By using the dimensions of

the Kerman cement plant preheater cyclone obtained from the two-objective optimization in

CFD simulation, the pressure drop reduces by about 2.2% and collection efficiency increases by

about 13.4%. The pressure drop and the efficiency of the current cyclone are respectively 560 Pa

and 84%.

Fig. 12

7. Conclusions

The first-stage cyclones' efficiency in the preheater system of a cement plant is very

important. In this study, one of the twin cyclones in the first-stage of the preheater in the Kerman

cement plant was investigated using a combination of CFD, the design of the experiment, and

multi-gene genetic programming. Since its structure had been changed, a decrease in its

efficiency had taken place. In the same pressure drop and volumetric flow rate, the Stairmand

cyclone efficiency was 2.5% greater than the current preheater cyclone while the Stairmand

cyclone was relatively larger. The effects of the three dimensions of the preheater cyclone were

investigated using the design of experiments. By increasing the vortex-finder height, the pressure

drop and the efficiency increase. On the other hand, by decreasing the cylindrical height and the
cone-tip diameter, the pressure drop and the efficiency increase. The MGGP was applied to

obtain two equations for calculating the Euler number and efficiency. These equations are able to

predict results with less error. Finally, these equations were used as objective functions for

preheater cyclone optimization. The optimized cyclone had 2.2% less pressure drop and 13.4 %

greater efficiency than the Kerman cement plant preheater cyclone. It could be concluded that

the modernisation of Hurriclon cyclone performed by the cement plant was rather impractical.

Because it had a negative effect on the cyclone performance. The new proposed structure

obtained by optimization technique was more efficient. Also, it is observed that particles which

are larger than the critical diameter spin initially downward, and then keep spinning at a certain

height on a fixed path near the wall. The critical diameter increases by decreasing the inlet

velocity, particle density, and by increasing the slope of the conical part of the cyclone.

Acknowledgements

The authors acknowledge the Kerman cement plant, Iran in allowing them to use their

experimental data.

Funding

This research did not receive any specific grant from funding agencies in the public,

commercial, or not-for-profit sectors.

Nomenclature

a inlet height (m)

b inlet width (m)

B the cone-tip diameter (m)

D cyclone diameter (m)

De vortex-finder diameter (m)

Dij stress diffusion

Ef efficiency

Eu Euler number

h cylindrical part height[m]

hs part of the scroll height (m)

H the cyclone total height (m)

Hc conical height (m)

L the height of outlet pipe (m)

P pressure (Pa)

Pij shear production

r radius of cyclone (m)

S vortex-finder length (m)

U inlet velocity (m/s)

ui time averaged fluid velocity i (m/s)

u′i fluctuating velocity to direction i (m/s)

u sub grid scale velocity (m/s)

v velocity characteristic (m/s)

X1 S/D

X2 h/D

X3 B/D

Xi the mass fraction of the particle i

Greek symbols

 ij Kronecker delta

 ij source term

i collection efficiency for diameter i

ij pressure-strain

ρ gas density (kg/m3)

ρp particle density (kg/m3)

μ viscosity (kg/m.s)

t eddy viscosity (kg/m.s)

ΔP pressure drop (Pa)

Abbreviations

ANOVA Analysis of Variance

DOE Design of Experiment

DRW Discrete Random Walk

GA Genetic Algorithm

GCI Grid Convergence Index

LES Large Eddy Simulation

MGGP Multi-Gene Genetic Programming

PRESTO Pressure Staggered Option

QUICK Quadratic Upstream Interpolation for Convective Kinetics

RSM Reynolds Stress Model

SIMPLEC Semi-Implicit Method for Pressure Linked Equations-Consistent

Appendix A

Table A.1 Inlet particles size distribution of limestone powder.

Distribution mass percent

of particle

(μm)

0-4.5 20.36

4.5-5.5 4.06

5.5-6.5 3.49

6.5-7.5 2.99

7.5-9 3.73

9-11 3.87

11-13 2.92

13-15.5 2.71

15.5-18.5 2.35

18.5-21.5 1.8

21.5-25 1.75

25-30 3.23

30-37.5 3.22

37.5-45 3.22
 45-52.5 3.18

52.55-62.5 4.08

62.5-75 4.86

75-90 5.28

90-105 4.48

105-125 4.75

125-150 4.38

150-180 3.8

180-215 3.08

215-255 2.12

255-305 1.28
References

[1] H. Zhang, R. Dewil, J. Degrève, J. Baeyens, The design of cyclonic pre-heaters in

suspension cement kilns, Int J Sustainable Engineering 7 (4) (2013) 307-312.

[2] W.D. Griffiths, F. Boysan, Computational fluid dynamics (CFD) and empirical modelling

of the performance of a number of cyclone samplers. Aerosol Science 27 (2) (1996) 281–

304.

[3] G. Wan, G. Sun, X. Xue, M. Shi, Solids concentration simulation of different size particles

in a cyclone separator, Powder Technology 183 (2008) 94–104.

[4] S.K. Shukla, P. Shukla, P. Ghosh, The effect of modeling of velocity fluctuations on

prediction of collection efficiency of cyclone separators, Applied Mathematical Model 37

(8) (2013) 5774-5789.

[5] T.G. Chuah, J. Gimbun, S.Y. Choong, A CFD study of the effect of cone dimensions on

sampling aerocyclones performance and hydrodynamics, Powder Technology 162 (2)

(2006) 126–132.

[6] X. Gao, J. Chen, J. Feng, Numerical investigation of the effects of the central channel on the

flow field in an oil–gas cyclone separator, Computers & Fluids 92 (2014) 45-55.

[7] Y. Su, A. Zheng, B. Zhao, Numerical simulation of effect of inlet configuration on square

cyclone separator performance, Power Technology 210 (3) (2011) 293-303.

[8] A. Kępa, The efficiency improvement of a large-diameter cyclone – The CFD calculations.

Separation and Purification Technology 118 (2013) 105-111.

[9] F. Ferit, V. Ari, Optimization of the preheater cyclone separators used in the cement

industry, Green Energy 10 (1) (2012) 12-27.

[10] Kh. Elsayed, Ch. Lacor, The effect of cyclone vortex finder dimensions on the flow pattern

and performance using LES, Computers & Fluids 71 (2013) 224-239.

[11] Kh. Elsayed, Ch. Lacor, The effect of cyclone inlet dimensions on the flow pattern and

performance, Applied Mathematical Modelling 35 (2011) 1952-1968.

[12] R. Xiang, S.H. Park, K.W. Lee, Effects of cone dimension on cyclone performance, Aerosol

Science 32 (4) (2011) 549–561.

[13] K.S. Lim, H.S. Kim, K.W. Lee, Characteristics of the collection efficiency for a cyclone

with different vortex finder shapes, Aerosol Science 35 (6) (2004) 743–754.

[14] H. Safikhani, A. Hajiloo, M.A. Ranjbar, Modeling and multi-objective optimization of

cyclone separators using CFD and genetic algorithms, Computers & Chemical Engineering

35 (9) (2011) 1064-1071.

[15] Kh. Elsayed, Ch. Lacor, Optimization of the cyclone separator geometry for minimum

pressure drop using mathematical models and CFD simulations, Chemical Engineering

Science 65 (22) (2010) 6048–6058.

[16] Kh. Elsayed, Ch. Lacor, Modeling, analysis and optimization of aircyclones using artificial

neural network, response surface methodology and CFD simulation approaches, Powder

Technology 212 (1) (2011) 115–133.

[17] Kh. Elsayed, Ch. Lacor, Multi-objective optimization of gas cyclone based on CFD

simulation, in: ECCOMAS Thematic Conference, CFD & Optimization, Antalya, Turkey

(2011).

[18] Kh. Elsayed, Ch. Lacor, Modeling and Pareto optimization of gas cyclone separator

performance using RBF type artificial neural networks genetic algorithms, Powder

Technology 217 (2012) 84-99.

[19] N. Fathizadeh, A. Mohebbi, S. Soltaninejad, M. Iranmanesh, Design and simulation of high

pressure cyclones for a gas city gate station using semi-empirical models, genetic algorithm

and computational fluid dynamics, Natural Gas Science and Engineering 26 (2015) 313-

329.

[20] M. Vujanovic, N. Duic, A. Galovic, Three-dimensional numerical simulation of the

nitrogen oxides formation in an oil-fired furnace, Strojarstvo 49 (2007) 165-171.

[21] H. Mikulcic, M. Vujanovic, D. Fidaros, P. Priesching, I. Minic, R. Tatschl, N. Duic, G.

Stefanovic, The application of CFD modelling to support the reduction of CO2 emissions in

cement industry, Energy 45 (2012) 464-473.

[22] H. Miculcic, M. Vujanovic, M. Sami Ashha, N. Duic, Large eddy simulation of a two-phase

reacting swirl flow inside a cement cyclone, Energy 75 (2014) 89-96.

[23] E.D. Cristea, P. Conti, CFD simulation of large dust collection cyclones positioned

vertically in staggered downward cascade arrangement, Proceedings of the ASME 2013

Fluids Engineering Division Summer Meeting, Nevada, USA.

[24] M. Wasilewski, J. Duda, Multicriteria optimization of first-stage cyclones in the clinker

burning system by means of numerical modelling and experimental research, Powder

Technology 289 (2016) 143-158.

[25] M. Wasilewski, L.S. Brar, Optimization of the geometry of cyclone separators used in

clinker burning process: A case study, Powder Technology 313 (2017) 293-302.

[26] F. Mariani, F. Risi, C.N. Grimaldi, Separation efficiency and heat exchange optimization in

a cyclone, Separation and Purification Technology 179 (2017) 393-402.

[27] J.I. Bhatty, Innovations in Portland Cement Manufacturing, Technology & Engineering,

Portland Cement Association, (2011).

[28] Holderbank, Suspension preheater, Process Technology II- Kiln Systems, 2000.

[29] Fluent 6.3.22 users guide. Fluent, Inc., 2006.

[30] B.R. Stanmore, P. Gilot, Review—calcination and carbonation of limestone during thermal

cycling for CO2 sequestration. Fuel Processing Technology 86 (16) (2005) 1707–1743.

[31] L. Schiller, and A. Naumann, Z. Ver Deut. In g., 77, 318 (1933).

[32] M. J. Schuh, C. A. Schuler, J. A. C. Humphrey, Numerical Calculation of particle laden gas

flow past tubes, AIChEJ.35(3) (1989) 466–480.

[33] S. A. Morsi, A. J. Alexander, An Investigation of Particle Trajectories in Two-Phase Flow

Systems, Fluid Mechanics 55 (2) (1972) 193-208.

[34] Oh, S. Choi, J. Kim, Numerical simulation of an internal flow field in a uniflow cyclone

separator. Powder Technology 274 (2015) 135-145.

[35] M. Azadi, M. Azadi, A. Mohebbi, CFD study of the effect of cyclone size on its

performance parameters, Hazardous Material 182 (2010) 835-841.

[36] G. Gong, Z. Yang, Sh. Zhu, Numerical investigation of the effect of helix angle and leaf

margin on the flow pattern and the performance of the axial flow cyclone separator, Applied

Mathematical Modelling 36 (8) (2012) 3916-3930.

[37] Kh. Elsayed, Ch. Lacor, CFD modeling and multi-objective optimization of cyclone

geometry using desirability function, artificial neural networks and genetic algorithms,

Applied Mathematical Modeling 37 (8) (2013) 5680-5704.

[38] H. Kaydani, A. Mohebbi, M.H. Saffaripour, A New Correlation Based on Multi-Gene

Genetic Programming for Predicting the Sweet Natural Gas Compressibility Factor. Gas

Processing Journal 3 (1) (2015) 41-50.

[39] H. Kaydani, A. Mohebbi, M. Eftekhari, Permeability estimation in heterogeneous oil

reservoirs by multi-gene genetic programming algorithm, Petroleum Science and

Engineering 123 (2014) 201-206.

[40] D.P. Searson, D.E. Leahy, M.J. Willis, GPTIPS: an open source genetic programming

toolbox for multi-gene symbolic regression. In: Proceedings of the international multi

conference of engineers and computer scientists, Hong Kong 2010.

[41] B. Wang, K. Chu, A. Yu, Numerical study of gas–solid flow in a cyclone separator, Applied

Mathematical Modelling 30 (2006) 1326-1342.

[42] H.F. Meier, K. Ropelato, M. Mori, Computational fluid dynamics CFD for cyclone

evaluation and design. ZKG Int 55 (6) (2002) 58-64.

[43] D. Montgomery, design and analysis of experiments, sixth edition, GGs Book Services,

Atlantic Highlands.

[44] I. Celik, U. Ghia, P.J. Roache, p. Raad, Procedure of Estimation and Reporting of

Uncertainty Due to Discretization in CFD Applications, Fluids Engineering 130 (7) (2008).

[45] M. Azadi, M. Azadi, an analytical study of the effect of inlet velocity on the cyclone

performance using mathematical models, Powder Technology 217 (2011) 121-127.

 Table 1 Dimensions of preheater cyclone.

b (m) a (m) Hc (m) H (m) h (m) hs (m) l (m) S (m) B (m) De (m) D (m)

1 1.92 3.55 10.448 4.553 2.345 3.5 2 0.428 1.193 2.524

Table 2 Comparison of cyclone pressure drop for three different meshes.

Mesh Pressure
Relative error(%)
number drop(Pa)

1722590 523.5 0.28

2208864 525 0.17

2746807 525.9

Experimental
560 6.25
data
Table 3 Comparison of cyclone efficiency for three different meshes.

Mesh Efficiency Relative error

number (%) (%)

1722590 81 0.8

2208864 81.66 0.049

2746807 81.7

Experimental
84 2.78
data
Table 4 The effect of parameters on the critical diameter.

Factor Value Critical diameter

(μm)

U (m/s) 7 -

12 160

14.7 55

20 20

ρp (kg/m3) 500 150

1500 80

2710 55

3300 48

a (m) 1.5 55

1.92 55

2.1 54

b (m) 0.8 54

1 55

1.2 54

De (m) 0.9 53

1.193 55

1.4 55

S (m) 1.625 55

2 55

2.345 56
h (m) 4.553 55

4.44 60

3.255 75

B (m) 0.314 42

0.428 55

0.5 93
Table 5 Selected dimensions for the design of experiments method.

Dimension (m) S h B

Level

1 1.625 3.255 0.428

2 2 4.44 0.628

3 2.345 4.553 1
Table 6 The averaged pressure drop and efficiency for 27 simulated cyclones.

Cyclone number Efficiency Pressure drop (Pa)

( %)

111 80.31 542

112 80.05 534.5

113 79.3 520

121 78.37 526

122 78.11 511.8

123 77.36 496

131 78.2 523

132 77.9 512.3

133 77.15 492.2

211 83.81 546.5

212 83.55 542.2

213 82.81 536

221 81.87 529

222 81.86 522

223 80.87 514

231 81.45 524.8

232 81.4 520

233 80.66 511

311 88 554

312 87.76 551.3

313 87 544.9

321 86.1 529.5

322 85.82 528.7

323 85.1 524.5

331 85.87 528

332 85.61 525

333 84.86 522.1

 Table 7 Analysis of variance for pressure drop.

source Adj MS F-value P-value

S 1252.76 9163.07 0.000

h 3004.15 21793.27 0.000

B 1112.09 8134.2 0.000

S.h 3.41 24.97 0.000

S.B 170.13 1244.42 0.000

h.B 4.9 35.85 0.000

S.h.B 7.18 52.55 0.000

Error 0.14
 Table 8 Analysis of variance for efficiency.

source Adj MS F-value P-value

S 24.791 183383.4 0.000

h 268.079 1983051.34 0.000

B 5.03 37205.64 0.000

S.h 0.005 37.58 0.000

S.B 0.004 27.82 0.000

h.B 0.006 42.51 0.000

S.h.B 0.006 44.91 0.000

Error 0.000
 Table 9 The parameter settings applied in MGGP.

Parameter Value

Function set ×, ÷, +, -

Fitness function RSME

Population size 200

Number of generation 300

Maximum tree depth 4

Maximum number of gens allowed as individual 3

Selection method Tournament

Tournament size 4
 Table 10 Optimized pressure drop, efficiency and dimensions.

Efficiency ΔP (Pa) B (m) h (m) S (m)

(%)

95.24 548 0.49 3.58 2.87

Table 11 Dimension ratios of optimized cyclone to the preheater cyclone.

(B)optimized / (B)preheater (h)optimized / (h)preheater (S)optimized / (S)preheater

1.145 0.786 1.435

B: cone tip diameter, h: cylindrical height, S: vortex finder height

Fig. 1 Rotary kiln with preheater cyclone [27].
Fig. 2 Hurriclon cyclone (a) before (b) after modernization.
Fig. 3 Schematic diagram of the cyclone.
Fig. 4 (a) 3D view of CFD mesh for the preheater cyclone (b) part of the cyclone, which is

gridded tetrahedral.
Fig. 5 Contour of (a) static pressure (Pa) (b) velocity magnitude (m/s) in sections z=9.5 m and x=0 for the

cyclone.
Fig. 6 Particle spinning in operation [41] (a) and a simulated cyclone (b).
Fig. 7 Analysis of design of experiment (a) main effect plot for pressure drop (b) main effect plot for

efficiency.
Fig. 8 Velocity contours for cyclones 121, 122 and 123 at section x = 0.
Fig. 9 MGGP model prediction versus calculated Euler number by CFD in (a) training data set (b) testing

data set.
Fig. 10 MGGP model prediction versus calculated efficiency by CFD in (a) training data set (b) testing

data set.
Fig. 11 Pareto chart for minimizing Euler number and maximizing efficiency for two-objective

optimization.
 Fig. 12 Comparison of tangential velocity for optimized cyclone and Kerman cement plant preheater

cyclone along the radius.

View publication stats