Accepted Manuscript

Separation efficiency and heat exchange optimization in a cyclone

Francesco Mariani, Francesco Risi, Carlo N. Grimaldi

PII: S1383-5866(16)32365-6
DOI: http://dx.doi.org/10.1016/j.seppur.2017.02.024
Reference: SEPPUR 13554

To appear in: Separation and Purification Technology

Received Date: 13 November 2016

Revised Date: 9 February 2017
Accepted Date: 10 February 2017

Please cite this article as: F. Mariani, F. Risi, C.N. Grimaldi, Separation efficiency and heat exchange optimization
in a cyclone, Separation and Purification Technology (2017), doi: http://dx.doi.org/10.1016/j.seppur.2017.02.024

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
 Separation efficiency and heat exchange optimization in a cyclone
Francesco Mariani(a), Francesco Risi(a), Carlo N. Grimaldi(a)
(a)
Department of Engineering, University of Perugia, Italy

Abstract
This work presents a study on a gas-solid cyclone separator used in a complex cement production plant. The objective of the study
consists on the performance evaluation and optimization of the cyclone separator in terms of particle separation and heat transfer
efficiencies, while keeping pressure losses under control. The thermal interaction is between two gas-solid mixtures, one feeded at 850
°C and the other at 600 °C, respectively. The solid phase consists, mostly, of calcium carbonate subsequently intended to the so-called
baking process for the production of clinker and then cement. The main goal of the study is maximizing the separation efficiency,
minimizing the temperature at the exit of the plant and maintain the pressure drop not far from its value in the running plant. In a
previous study, the authors defined the suitable setup of the mesh, physical models and turbulence; the same setup has been adopted in
the present work during all the optimization phase. Vortex finder length (initially of cylindrical shape), its conicity angle and the
pressure losses are the parameters (independent variables), taken into account to optimize the separation efficiency and the
temperature of the gas at the exit of the plant. The optimization process was started with the base case that replies the geometric shape
of the plant in operation.
A detailed overview of the classical and recent optimization methods, is included in the Appendix, together with a brief illustration of
a general pattern concerning an optimal design procedure. The numerical approach, for all the cases, is based on implicit unsteady
simulations using the Eulerian Multiphase model.
Keywords: cyclone; Eulerian multiphase; gas-solid heat exchange; computational fluid dynamics; geometrical optimization

1 Introduction
1.1 The industrial process

The present study is a step of a longer activity aimed at the study of possible technological solutions that allow to optimize the heat
exchange between a flow of gas at high temperature (about 900°C) and the raw material powder (consisting largely of calcium
carbonate), to be submitted to the cooking process for the production of "clinker", and then of cement. Much of this heat exchange
takes place in a preheating tower constituted by cyclones and connection pipes: the flow of hot air from the oven meets, in
countercurrent, the raw material to be preheated. Current technologies provide that the heat exchange between gas and raw materials
takes place in a very limited area of the circuit (the preheating tower), that is in the connecting lines between the various cyclones. In
these areas a mixing takes place (in equi-current) between the gas and the powder of raw material which allows a certain heat transfer.
In the cyclones, which constitute the main part of the structure, does not take place any important heat exchange because the gas-
powdery material separation is almost immediate; on the other hand, the separation process is also very important for a good thermal
efficiency of the system. The introduction of the suspension preheater, in the early thirties, was an important evolution of the
clinkerization process. The dried raw meal is preheated and even partially calcined (dry/semi-wet processes) while it is held in
suspension with the hot gases coming from the rotary kiln. At least in theory, a wider contact surface allows a nearly complete heat
exchange; for this reason there are several suspension pre-heating systems as that shown in figure 1(a). These systems usually have
four to six stages of cyclones, which are arranged one over the other to form a tower with a height varying from 50 to 120 m. The
stadium on top can be constituted by two parallel cyclones for a better separation of the powder. The gas exhausted from the furnace
flows through the various stages of cyclones from the bottom upwards. The finely ground raw and dry meal, mixes with the gas before
the highest stage of cyclones; it is separated from the gas in the cyclone and rejoins it before the next stage of cyclones. This
procedure is repeated for each stage until the material is discharged from the last cyclone (the lowest), into the rotary kiln. This
cyclical process of mixing, separation and re-mixing occurs at high temperature to optimize the heat transfer. The cement is produced
by a particularly energy-intensive process: in fact, each tonne of finished product requires about 120 kWh of electric energy and 6x105
kcal of thermal energy. A standard plant which produces about 106 t/year, employs about 12x104 MWh/y of electric energy and 6x105
Gcal/y of thermal energy (equivalent to 8x104 ton of coal/y). The thermal energy is used for the cooking of the semi-finished product
(clinker) and is mainly obtained by burning petroleum coke (about 8,200 kcal/kg). For a standard production of about 125 tons/hour of
clinker are burned about 13 tons/hour of petroleum coke; therefore, it is very important to optimize the heat consumption of the firing.
The insertion, upstream of a moving grate oven (where takes place the final part of the cooking process), of a system of cyclones
(preheating tower), reduces the heat consumption by more than 1200 kcal/kg clinker to about 850 kcal/kg. This change allows you to
recover some of the heat from the combustion gases (about 2x105 nmc/h at 900 °C). Despite this, the gases leave the circuit at a
temperature of about 330-380 °C.

1.2 State of the art

In general, the task of a cyclone is the removal of solid particles from a gas with a consequent beneficial effect on the air pollution
control. At this aim, a lot of plant configurations, available in literature [2–16], have been widely studied both numerically and
experimentally with the main goal of increasing the separation efficiency. Multiple studies [2–9], demonstrated the decisive influence
of both operating parameters and design on the separation efficiency and pressure drop.
The analyzed models are generally constituted by only one cyclone, with a traditional circular cross section and a tangential inlet.
In [2] the authors studied the influence of cylinder shape on separation efficiency; they highlighted the influence of the size and shape
of the cylinder on the core-annulus shape inside the cyclone. This shape plays a very important role on the global flow and particle
separation. In [5], with both experimental and numerical approaches, the authors analyzed the effect of a prolonged cyclone, obtained
attaching vertical tubes of different length at the bottom of the particle outlet, on the separation performance. In [3], Lim et al. have
highlighted how the vortex finder plays a critical role in defining the flow field inside the cyclone; at this aim, they took measures to
check the particle separation performance employing vortex finders of different lengths. In [5], the authors show the influence of the
bottom-contracted and edge-sloped vent-pipe on the separation efficiency and pressure drop of a cyclone separator under different
vent-pipe insert depth and different orientation of the sloped edge. In the papers [6] and [7], by experimental tests, the authors have
shown that it is possible to decrease the pressure drop and improve the separation efficiency using double inlets. In [8] Zheng et al.
presented an experimental work that evaluates, inter alia, the variation of separation efficiency and pressure drop with different
combinations of inlet forms and inlet air speeds. Bernardo et al. [9] analyzed the effects of the inlet section angles on the fluid
dynamics inside the cyclone and on the performance parameters: collection efficiency and pressure drop. The numerical results show
that the value for overall collection efficiency increased to 77.2% for the 45° inlet section angle, while that for the normal inlet duct
was 54.4% under the same operating conditions. The performance of a water-cooled square cyclone was compared in [17] with that of
other cyclones through fly ash analysis. The results showed that the overall performance of the square cyclone in such capacity is
compatible to that of the round cyclone (pressure drop less than 800 Pa). Yaxin et al. [18] numerically investigated a square cyclone
separator with three types of inlet configurations. The resulting velocity, separation efficiency and pressure drop were verified by
comparison with experimental data. Results showed that inlet configurations influenced the turbulent dynamics in the cyclone and led
to different pressure drop and separation efficiency. Inter alia, the authors proved that, the separator with single normal inlet (SNI),
had the best separation efficiency and the maximum pressure drop. Furthermore, when a baffle was installed in the inlet of separator
SNI, the pressure drop increased by about 191%.
For further improvement of cyclone performance it is mandatory the understanding of the turbulent dynamics and the separation
process. The CFD-3D analysis is an efficient approach of understanding the fluid dynamics process in complex plants, providing a
precious way to evaluate how it is affected by changes of operating conditions and geometric shapes. At this aim, the CFD approach
has been used by many authors [2], [9–16]; the papers deal with issues relative to the flow of gas-solid suspensions predicting solid
separation efficiency, at different operating conditions, in good agreement with the experimental evidences. Furthermore, in [14–16]
the authors proved as the Reynolds Stress Model is the suitable turbulent model for cyclone with strong swirling flow. In a previous
work [1], the authors have analyzed the same plant studied in the present paper, using a traditional optimization approach. In the
present study, the Star-CCM+ 11.04 package [19] was used to solve the equations by the finite volume method.

1.3 Optimization and CFD Synergy

In theory, an optimization problem implies that one should analyse an infinite number of geometric changes and be able to minimize
or maximize the associated objective functions to reach the best geometry shape. On this, Thévenin and Janiga [20] observe that “so
many researchers and engineers still employ the terminology optimization in the title or abstract of their publications when they
simply mean in practice that starting from a non-satisfactory configuration, they have tried two or three other ones and chosen at the
end the best case“. In recent years the availability of affordable computational power has been a determining factor in pushing some
manufacturers to face integration problems between CFD and optimization packages. The coupled simulations of the two codes,
makes the optimization task very interesting in terms of developing time, even if applied to complex systems.

In recent years, 2010-2016, many researchers have published interesting works [21–30], that go beyond the limitations described and
reported by Thévenin and Janiga in [20]. These works have in common the integration task between the most sophisticated
optimization techniques with the most famous CFD-3D packages, employed to solve thermo-fluid-dynamics problems. In all the
above works, the word “integration” is strictly appropriated because all follow the cycle: first geometry definition (base case), mesh
definition, simulation, interpretation of results, re-definition of the geometry applying the suggested variations by the auto-adaptive
optimizer, new CAD generation, re-meshing and so on; through the simulation of a very huge number of cases within the constraints
of the geometric variables to be optimized. The present work follows the same methodology, trying to solve a complex optimization
problem of a multi-cyclone plant. The optimized parameters are concerning the shape and dimensions of some geometric details of the
vortex finders, with a constraint on the pressure-loss value and an objective function able to maximizing the separation efficiency and
able to minimizing the temperature of the gas exiting from the top.

The optimization traditional approach to thermo-fluid-dynamics problems is applied, as an example, in [25], [31–33]. In these cases,
the optimizer and the CFD solver, operate in an asynchronous way implying that the time consuming steps have to be solved manually
limiting the number of cases one can analyse. The frequent consequence of this limitation is the quality of the results that are far from
an acceptable optimized solution.

1.4 Structure and contribution of the present work

The complex plant studied in the present paper is an assembly of cyclones as a part of a process whose industrial goal is the
production of the cement. The CFD simulations aimed to evaluate the particles separation performance and the thermal exchange
between the continuous and the solid phase; the separation efficiency and the gas exit temperature are the main physical quantities to
improve. This type of problem can be classified as a multiphase system with heat exchange where the air is the continuous phase, at a
temperature of 850 °C and the dispersed phase at 600 °C consists of Calcium Carbonate (density equal to 2063 kg/m3), subsequently
intended to the so-called baking process, in order to produce clinker and then cement. The volume fraction (VF) of particles, in the
simulated domain, ranges from 0 to 0.57; as shown in the following paragraphs, the maximum value is reached only in two small
regions where the geometric conformation allows the retention of solid material.
The study was divided into two basic stages: in the first, some simulations were carried out with the aim of obtaining results in
agreement with the available experimental data. This simulation is referred as case #0 and characterized by the absence of the vortex
finders; it reproduces the geometry of the system currently in operation. During this iterative phase, the quality of the mesh and the
model setup were modified to match the main experimental quantities: gas-solid exit temperature and separation efficiency.
Using this consolidated setup, the work continues through the optimization phase consisting in the variation of two geometrical
parameter of each vortex finder: the length and the conicity. The optimization process consist in 120 cases running on 24 Xeon cores
in about 100 hours.
The main contribution of the present work consists in having integrated different aspects in detail before highlighted: the plant is a real
plant of considerable size; the optimization process is integrated with the CFD analysis; the simulation model, in addition to the
classical gas / solid separation, integrates all the heat exchange models of convection, conduction and radiation.

Figure 1

2 Model

2.1 Cyclone separator

Together with the table 1, figure 1(b) shows the geometrical dimensions of the cyclone separator model. Figure 1(c) shows the
volumetric model where A and B represent the main and secondary feed respectively; the same figure shows the mesh constituted by
about 500000 polyhedral volumetric cells with a base size of 10 cm, a minimum cell size of 2 cm and 2 prism layers at the walls. The
transparent visualization enables to see the presence of the vortex finders, the length of which characterizes the six simulated models.

Table 1

2.2 Mesh sensitivity test

The following table summarizes the results obtained from the mesh sensitivity study. For all cases the adopted type of mesh is
polyhedral, with four prism layers. The base size of the medium and fine mesh is a half and a quarter of the coarse base size
respectively. In all the following simulations it has been used the coarse. In effect, as can be seen in Table 2, using a finer mesh
implies an increase of the time needed to converge on the physical parameters Tgas-out, Tsolid-out and σ. In particular, the elapsed-time is
especially important bearing in mind that the optimization phase is composed of 120 cases or geometries.
 Table 2

2.3 Numerical model

For incompressible, unsteady and non-isothermal flow in cyclone separator, the eulerian-multiphase model has been used. This model
treats a dispersed multiphase flow as two (or more) fully interpenetrating quasi-fluid also considering the interpenetrating effect of
each phase by using drag model. The volume fraction of each phase is defined on the basis of the distribution of each phase and
the size of computational volume. The two-fluid model equation for each phase is given by the following equations, where the
subscript k represents the single phase:

(1)

(2)

(3)

The difficulty of solve these sets of equations is twofold. First, these equations are difficult from a numerical point of view, since there
are many coupled equations with one shared pressure. Secondly, to solve the equations, closure models are required for and ,
which represent respectively the rheology of the phase and the interaction force acting on the phase k.
In particular, the drag force has been modelling as:

(4)

where the linearized drag coefficient is defined by the Gidaspow model as:

(5)

and where the drag coefficient is defined by the Schiller-Neumann equation as:

In this work the transition volume fraction in set to 0.2 and the hindered settling exponent is n=-1.65. A 2-Layer Realizable K-ε
model has been applied to the gas phase and a Modified Johnson Frictional model was used to evaluate the stresses in granular
medium.
At present, in literature, it is proved that the most utilized turbulence modelling is the RSM turbulence model [2], [4], [9–15]. In
addition, all these papers simulate isothermal plants while the present work has to take into account the thermal exchange in its various
forms (convection and radiation). At this aim, Table 3 summarizes the results evaluated using three different RANS models. These
simulations have been performed on the baseline geometry (with L=1m and α = 90°, for all vortex-finders), with coarse mesh. It can
be seen how the variation of the turbulence model does not involve substantial differences neither on the elapsed time nor on the
fundamental physical parameters.

Table 3

In all simulations the k-ε turbulence model has been adopted. The time-step of integration has been evaluated basing on the
Courant method, while the stop criterion of the simulations was based on the convergence of the two physical parameters: gas and
solid temperatures at the exit and separation efficiency. The boundary conditions concerning the feed of the cyclone (A and B in figure
1(c)) are summarized in Table 4.
 Table 4

All the previous choices in terms of the adopted spatial discretization and models have been previously validated comparing the
experimental data with the corresponding results obtained from the simulation executed on the geometry that reproduces the currently
running system, namely the case #0, that is without the vortex finder. As shown in the following Table 5, the experimental values are
very close to the computed ones.

Table 5

3 Optimization process
3.1 Optimization Methodology Formulation

There are two distinct techniques of optimization, widely used today: classical and advanced techniques to which correspond two
distinct types of optimization algorithms: deterministic and stochastic. To such respect, in Appendix, a brief description of the
principal methods of optimization is given. In the present work the optimization problem was introduced to find the best geometric
shape of the three vortex finder that minimize the temperature of gas exiting from top and, at the same time, maximize the separation
efficiency. Figure 2 shows the loop performed for each analysed design.
The optimization process consists in a coupled activity between a CFD solver and Optimate+, an add-on to the STAR-CCM+ code by
CD-ADAPCO. SHERPA (Systematic Hybrid Exploration, Robust, Progressive and Adaptive), [34] is one of the complex optimization
algorithms of HEEDS (Hierarchical Evolutionary Engineering Design System), able to perform automated design optimization.
SHERPA is a hybrid algorithm that employs multiple search techniques simultaneously; the automated iterative design process puts
together all these elements taking advantages of the best characteristics from each embedded method. It contains methods with tuning
parameters that are changed during the search analysis in agreement with the experience gained from the previous generated design
spaces; that is, SHERPA learns efficiently about the design space and adapts to be able to search all the design spaces. Each phase
associated to a step of the sketch is an automatic activity that does not require any human action. The initial reference model is created
and employed for the first execution of the optimization study; successively, the input file associated with this model is modified by
the optimization tool for each new design evaluation (case) explored during the study. The optimization algorithm is the focal point of
the described cyclic process that, at each iteration step, controls the direction and type of the search to employ for the next activity. In
[37], Chase et al. developed a work showing how much an optimization algorithm is sensitive to the initial guesses. In particular, they
focus their attention on the hybrid adaptive method SHERPA, embedded in HEEDS professional, that was compared to several
existing optimization algorithms: ASA, GA, NLPQL. All compared algorithms were tested on a broad set of benchmark problems and
it was concluded that the SHERPA algorithm is significantly more efficient and robust for these types of problems than the other ones
in the study. Putting together the value(s) returned for the objectives and the degree to which a design satisfies its constraints,
following the relation (1), SHERPA gives a performance rating to each design, ignoring by what margin they meet those constraints,
and only the objectives contribute numerically to the performance evaluation. The performance value is calculated using the following
equation:

(1)
[34], where No is number of objectives, ± is sign for the i-th objective, Obi is the response value for the i-th objective, Nfi is the
normalization factor for the i-th objective, pfj is the penalty factor for the j-th constraint, Nc is the number of constraints, CntViolj is
the amount by which the j-th constraint is violated, and Tgtj is target value of the j-th constraint. For feasible design (all constraints are
satisfied), the performance function is a sum of the normalized objective values. When one or more constraints are violated, the
performance value of the design is reduced by a values based on the violation of the constraint (the second term in the previous
equation).

Figure 2
3.2 Optimization tool setup

As described before, the goal of the optimization process is the minimization of the temperature of the gas exiting from the upper
cylone and the simultaneous maximization of the separation efficiency (σ) if the whole plant, defined as:

also considering the overall pressure drop, that will be less than 900 Pa.
The optimizer has been used to perform a Pareto analysis accomplished by the simulation of 120 cases. In each case and for each
vortex finder, two geometrical parameters, specifically the length and the conicity (Figure 3), have been varied by SHERPA algorithm
using the geometrical constraints described in Table 6.

Figure 3

Table 6

Referring to the whole 120 cases, Figure 4 shows the probability density function (PDF) of the geometric dimensions to optimize that,
the SHERPA algorithm, associates during the “Cad model generation” (one of the steps shown in Figure 2). Each curve is limited by
the geometric constraints previously defined and allows the evaluation of the probability that a value could be extracted from
SHERPA into the constraint range. For each geometrical parameter, this evaluation has been made using the following general
expression:

Where X is a random variable and fX is the related PDF which measures the probability of X falling within the interval [inf. sup].

Figure 4

4 Results and discussion

To evaluate the influence of each geometrical parameter on the separation efficiency, exit gas temperature and pressure drop, the
Pearson product-moment correlation coefficient (PCC) has been used [35], [36]. The PCC is a measure of the linear correlation
between two generic variables X and Y, giving a value between +1 and -1, where 1 is total positive correlation, 0 is no correlation, and
–1 is total negative correlation. It is defined by:

Where σx and σy represent the standard deviation of X and Y respectively, μ x and μy the mean of X and Y and E the expected value.
Table 7 shows the value of PCC for the six geometric parameters used for the Pareto optimization.

Table 7

From these results it’s possible to observe that the variable that mainly influences the separation efficiency improvement is the
conicity of the upper cyclone, and the most influent parameter for the reduction of the temperature of the gas exiting from the top is
the length of the upper cyclone. In particular, there is a negative correlation between α_t and σ, which means that a reduction of the
outlet section of the upper vortex finder corresponds to an improvement of separation efficiency. Obviously, the required regular
operation of the plant, limits the unconditional decrease of the angle α taking under control the magnitude of the pressure load ΔP (in
the optimizer, its value is constrained to be less than 1000 Pa; a limit imposed from the management of the plant). Furthermore, for the
temperature, there is a negative correlation with the length of upper vortex finder; this means that an improvement of the vortex finder
length corresponds to a reduction of the exit gas temperature, due to an increase of the residence time.
 Figure 5

Figure 5 shows the results of all the 120 simulations performed during the optimization process. The points are colored by the overall
pressure drop, and the red ones exceeds the maximum pressure drop imposed by the manufacturer of the plant. The triangle represents
the actual operating conditions of the plant (#0), without vortex finder and corresponding to a value of separation efficiency of 92.7 %
and to an outlet temperature of 695 °C. The star represents the baseline case, characterized by a vortex finder one meter long (in upper
and lower cyclones); it shows a small increase of separation efficiency (93.2 %) and a significant reduction of exit temperature (686
°C). The other points represent all the variations of length and conicity for the three vortex finder. Finally, the square represents the
chosen best design with a separation efficiency of 97% (corresponding to an improvement of 4.7% compared to the #0 case and of
3.8% compared to the baseline), an exit temperature of 682 °C (corresponding to a reduction of 13 °C compared to the #0 and of 4°C
compared to the baseline), and a pressure drop of 775 Pa. Table 8 summarizes all the previous results.

Figure 6

Figure 7

Table 8.

The contour plots of figures 6 and 7 represent the results of the three main configurations relating to the volume fraction of powder
and temperature of the gas. The main variable that is on the base of the physical interpretation of the showed results, surely, is the
residence time of particles inside the cyclone that increases with the insertion of the vortex finders. As a consequence there is a better
heat transfer between the two phases (air and particles), with a decrease of temperature of the gas at the exit of plant and a consequent
increase of temperature of the powder. At the same time, a greater residence time also has a beneficial effect on the separation
efficiency that in the optimum case increases of about 4.6%, if compared to the #0 case. The results summarized in Table 8, also stress
the fact that the pressure drop remains widely within acceptable values (the experimental pressure drop is 647 Pa).

5 Conclusions
The present study is an optimization study executed on a real plant of considerable size; the optimization process is integrated with the
CFD analysis. The simulation model, in addition to the classical gas/solid separation, implements all the heat exchange models of
convection, conduction and radiation. The optimization is performed on the length and angle of the three vortex finders (one for each
cyclone), introduced to increase the separation efficiency and to decrease the exit gas temperature. The experimental values of
temperature at the exit, the pressure drop and the separation efficiency, measured on the plant in operation (#0), were used to define an
optimal setup of the numerical models used in the simulations. In particular, it has been reached a confidence value of about 1.5% as
deviation between the experimental and calculated data.
The optimization process finds a set of geometric parameters that leads to a significant reduction of the temperature of gas exiting
from the plant and a contemporary improvement of overall separation efficiency. These results allow to reduce the quantity of fuel
used for the chemical reactions that, inside the oven, transform the clinker into the cement: the last phase of the process of
transformation.

Acknowledgements
This work was supported by the precious contribute of Colacem S.p.A, – Gubbio (Italy)

References
[1] F. Mariani, F. Risi, and C. Poggiani, “Numerical study of the separation efficiency and heat exchange performance in a
complex gas–solid separator,” Int. J. Comput. Methods Exp. Meas., vol. 4, no. 2, pp. 142–152, 2016.
[2] J. W. Lee, H. J. Yang, and D. Y. Lee, “Effect of the cylinder shape of a long-coned cyclone on the stable flow-field
establishment,” Powder Technol., vol. 165, no. 1, pp. 30–38, 2006.

[3] K. S. Lim, H. S. Kim, and K. W. Lee, “Characteristics of the collection efficiency for a cyclone with different vortex finder
shapes,” J. Aerosol Sci., vol. 35, no. 6, pp. 743–754, 2004.

[4] F. Qian, J. Zhang, and M. Zhang, “Effects of the prolonged vertical tube on the separation performance of a cyclone,” J.
Hazard. Mater., vol. 136, no. 3, pp. 822–829, 2006.

[5] J. Chen, X. Lu, H. Liu, and C. Yang, “Effect of the bottom-contracted and edge-sloped vent-pipe on the cyclone separator
performance,” Chem. Eng. J., vol. 129, no. 1–3, pp. 85–90, 2007.

[6] B. Zhao, H. Shen, and Y. Kang, “Development of a symmetrical spiral inlet to improve cyclone separator performance,”
Powder Technol., vol. 145, no. 1, pp. 47–50, 2004.

[7] K. S. Lim, S. B. Kwon, and K. W. Lee, “Characteristics of the collection efficiency for a double inlet cyclone with clean air,”
J. Aerosol Sci., vol. 34, no. 8, pp. 1085–1095, 2003.

[8] Z. An-qiao, S. U. Ya-xin, W. A. N. Xin, and E. Science, “Experimental Study of a Square-shaped Separator with Different
Inlet Forms,” p. 2008, 2008.

[9] S. Bernardo, M. Mori, A. P. Peres, and R. P. Dionísio, “3-D computational fluid dynamics for gas and gas-particle flows in a
cyclone with different inlet section angles,” Powder Technol., vol. 162, no. 3, pp. 190–200, 2006.

[10] B. Wang, D. L. Xu, K. W. Chu, and A. B. Yu, “Numerical study of gas-solid flow in a cyclone separator,” Appl. Math.
Model., vol. 30, no. 11, pp. 1326–1342, 2006.

[11] F. Kaya and I. Karagoz, “Numerical investigation of performance characteristics of a cyclone prolonged with a dipleg,” Chem.
Eng. J., vol. 151, no. 1–3, pp. 39–45, 2009.

[12] K. U. Bhaskar, Y. R. Murthy, M. R. Raju, S. Tiwari, J. K. Srivastava, and N. Ramakrishnan, “CFD simulation and
experimental validation studies on hydrocyclone,” Miner. Eng., vol. 20, no. 1, pp. 60–71, 2007.

[13] R. B. Xiang and K. W. Lee, “Numerical study of flow field in cyclones of different height,” Chem. Eng. Process. Process
Intensif., vol. 44, no. 8, pp. 877–883, 2005.

[14] A. J. Hoekstra, J. J. Derksen, and H. E. A. Van Den Akker, “An experimental and numerical study of turbulent swirling flow
in gas cyclones,” Chem. Eng. Sci., vol. 54, no. 13–14, pp. 2055–2065, 1999.

[15] K. Pant, C. T. Crowe, and P. Irving, “On the design of miniature cyclones for the collection of bioaerosols,” Powder Technol.,
vol. 125, no. 2–3, pp. 260–265, 2002.

[16] H. C. H. Sommerfeld, M., “Numerical calculation of particle transport in turbulent wall bounded flows,” Powder Technol.,
vol. 131, pp. 1–6, 2003.

[17] L. Junfu, Z. Jiansheng, Z. Hai, L. Qing, and Y. Guangxi, “Performance evaluation of a 220t/h CFB boiler with water-cooled
square cyclones,” Fuel Process. Technol., vol. 88, no. 2, pp. 129–135, 2007.

[18] Y. Su, A. Zheng, and B. Zhao, “Numerical simulation of effect of inlet configuration on square cyclone separator
performance,” Powder Technol., vol. 210, no. 3, pp. 293–303, 2011.

[19] “CD-adapco STAR-CCM+ Documentation Version 11.02.” p. 11.

[20] D. Thévenin and G. Janiga, Optimization and Computational Fluid Dynamics. Springer-Verlag Berlin Heidelberg, 2008.

[21] K. . Elsayed and C. Lacor, “Optimization of the cyclone separator geometry for minimum pressure drop using mathematical
models and CFD simulations.,” Chem. Eng., vol. 65, pp. 6048–6058, 2010.

[22] H. Safikhani, A. Hajiloo, and M. A. Ranjbar, “Modeling and multi-objective optimization of cyclone separators using CFD
 and genetic algorithms,” Comput. Chem. Eng., vol. 35, no. 6, pp. 1064–1071, 2011.

[23] K. Elsayed and C. Lacor, “Modeling, analysis and optimization of aircyclones using artificial neural network, response surface
methodology and CFD simulation approaches,” Powder Technol., vol. 212, no. 1, pp. 115–133, 2011.

[24] K. Elsayed and C. Lacor, “Modeling and Pareto optimization of gas cyclone separator performance using RBF type artificial
neural networks and genetic algorithms,” Powder Technol., vol. 217, pp. 84–99, 2012.

[25] K. Elsayed and C. Lacor, “CFD modeling and multi-objective optimization of cyclone geometry using desirability function,
Artificial neural networks and genetic algorithms,” Appl. Math. Model., vol. 37, no. 8, pp. 5680–5704, 2013.

[26] O. L. Sgrott, D. Noriler, V. R. Wiggers, and H. F. Meier, “Cyclone optimization by COMPLEX method and CFD simulation,”
Powder Technol., vol. 277, pp. 11–21, 2015.

[27] A. Alves, J. Paiva, and R. Salcedo, “Cyclone optimization including particle clustering,” Powder Technol., vol. 272, pp. 14–
22, 2015.

[28] K. Elsayed, “Design of a novel gas cyclone vortex finder using the adjoint method,” Sep. Purif. Technol., vol. 142, pp. 274–
286, 2015.

[29] K. Elsayed, “Optimization of the cyclone separator geometry for minimum pressure drop using Co-Kriging,” Powder
Technol., vol. 269, pp. 409–424, 2015.

[30] P. Singh, I. Couckuyt, K. Elsayed, D. Deschrijver, and T. Dhaene, “Shape optimization of a cyclone separator using multi-
objective surrogate-based optimization,” Appl. Math. Model., vol. 40, no. 5–6, pp. 4248–4259, 2016.

[31] N. Athanasios, N. Nikolaos, M. Nikolaos, G. Panagiotis, and E. Kakaras, “Optimization of a log wood boiler through CFD
simulation methods,” Fuel Process. Technol., vol. 137, pp. 75–92, 2015.

[32] M. Jia, M. Xie, T. Wang, and Z. Peng, “The effect of injection timing and intake valve close timing on performance and
emissions of diesel PCCI engine with a full engine cycle CFD simulation,” Appl. Energy, vol. 88, no. 9, pp. 2967–2975, 2011.

[33] J. a Samareh, “Aerodynamic shape optimization based on free-form deformation,” vol. 1, pp. 1–12, 2004.

[34] Red Cedar Technology, “HEEDS (Hierarchical Evolutionary Engineering Design System). Getting started manual.,” U.S.A.,
2013.

[35] K. Pearson, “Notes on Regression and Inheritance in the case of two parents.,” in Proceedings of the Royal Society of London.

[36] H. M. Kaltenbach, A Concise Guide to Stocastics. Springer, 2012.

[37] N. Chase, M. Redemacher, E. Goodman, R. Averill, R. Sidhu, A benchmark study of optimization search algorithms, Red Cedar
Tech.; USA 2010. http://www.redcedartech.com/pdfs/SHERPA_Benchmark_0110.pdf.

Contact Information

Francesco Mariani
06128 - Perugia (Italy), Via G. Duranti, 67
Phone: +39-075-585-3732
francesco.mariani@unipg.it

Definitions/Abbreviations
ASA Adaptive Simulated Annealing

CFD Computational Fluid Dynamics

GA Genetic Algorithm

HEEDS Hierarchical Evolutionary Engineering Design System

IVC Intake Valve Close

NLPQL Non-Linear Programming by Quadratic Lagrangian

PCCI Premixed Combustion Compressed Ignition

PCC Pearson Correlation Coefficient

RMS Response Surface Methodology

SHERPA Systematic Hybrid Exploration Robust, Progressive, and Adaptive

vf Vortex Finder

VF Volume Fraction

L_t vf length of upper cyclone

L_b,l vf length of bottom left cyclone

L_b,r vf length of bottom right cyclone

α_t vf semi-conicity of upper cyclone

α_b,l vf semi-conicity of bottom left cyclone

α_b,r vf semi-conicity of bottom right cyclone

σ separation efficiency

ΔP pressure drop
Appendix

Optimization Methods

In the optimization process of a design, the objective could be simply maximize the efficiency of production or minimize the cost of
production. An optimization algorithm is an iterative procedure which tries to find the minimum (maximum) values of mathematical
functions. Among other things, it is used to evaluate design tradeoffs, to assess control systems, and to find patterns in data.
With the rapid progress in computer science, optimization has become an important part of computer-aided design activities. In
particular, the advent of the multi-core processors is allowing ever cheaper and less time-consuming simulations. Figure A1 shows an
outline of the main steps involved in an optimal design.

Figure A1

In the current literature, optimization algorithms are classified in two main types: classical and advanced techniques.

Classical Optimization Technique

These are analytical methods that make use of differential calculus in locating the optimum solution. They are useful in finding the optimum solution
or unconstrained maxima or minima of continuous and differentiable functions. Their restriction is to have limited scope in practical applications as
some of them involve objective functions which are not continuous and/or differentiable. In fact, these methods assume that the function is
differentiable twice with respect to the design variables and the derivatives are continuous. However, despite their limitations, these classical
techniques of optimization constitute a solid foundation for developing most of the numerical techniques that have evolved into advanced techniques
more suitable to today’s practical problems. The classical optimization techniques is an approach generally used to handle three main types of
problems: single variable functions, multivariable functions with no constraints and multivariable functions with both equality and inequality
constraints. In particular, with equality constraints the Lagrange multiplier method can be used. If the study presents inequality constraints, the
optimum solution is identified by using the Kuhn-Tucker conditions [A1], a generalization of the Lagrange multiplier method. These are necessary
conditions for the solution of a problem of nonlinear programming [A2].

Numerical Optimization Technique

Numerical methods of optimization present four main type of programming: linear, integer, quadratic and non-linear. The linear programming
studies the case in which the objective function (f) is linear and the design variable space (A) is specified using only linear equalities and inequalities.
The integer programming method studies linear programs in which some or all variables are constrained to take on integer values. The quadratic
programming allows the objective function to have quadratic terms, while the set A must be specified with linear equalities and inequalities. The
nonlinear programming studies the general case in which the objective function or the constraints or both contain nonlinear parts.
The dynamic programming is one of the five additional later specialized programming techniques. It is a method for solving complex
problems and its optimization strategy is based on splitting the problem into smaller sub-problems, solving each of those sub-
problems, just once, and storing their solutions. The next time the same sub-problem occurs, instead of re-computing its solution, one
simply looks up the previously computed solution, thereby saving computational time. At the end, the algorithm will examine the
previously solved sub-problems and will combine their solutions to give the best solution for the given problem. In the past, this
approach has been intensively used to solve the famous classical traveling salesman problem, the aim of which is to find the shortest
path between two points. The stochastic programming studies the case in which some of the constraints depend on random variables.
It is a mathematical optimization, for modeling real world problems that involve uncertainty. Whereas deterministic optimization
problems are formulated with known parameters, real world problems almost invariably include some unknown parameters. The
problems approached with this technique are of a probabilistic nature, that is to say, use probability theory results to get to their target.
The combinatorial optimization is concerned with problems where the set of feasible solutions is discrete or can be reduced to a
discrete one. It has important applications in several fields, including artificial intelligence, machine learning, mathematics and
software engineering where the goal consists of finding an optimal object from a finite set of objects. Some common problems
involving combinatorial optimization are the traveling salesman problem and the minimum spanning tree problem. The infinite-
dimensional optimization studies the case when the set of feasible solutions is a subset of an infinite-dimensional space, such as a
space of functions. The constraint satisfaction studies the case in which the objective function f is constant; it is used in artificial
intelligence, particularly in automated reasoning.

Advanced Optimization Technique

Adaptive Simulated Annealing (ASA)

ASA has been developed rapidly since the past 20 years as an effective and simple optimization technique. The superiority of SA lies
in its good robustness and convenience to realize global optimization; moreover, it is not dependent on solution gradients. ASA [A3]
is a code that finds the best global fit of a nonlinear cost-function over a so-called D-dimensional space (a space of parameters able to
describe a variety of physical problems). It has over 100 options to provide robust tuning over many classes of nonlinear stochastic
systems. These many options help ensure that ASA can be used robustly across many classes of systems. In particular, in the
benchmark described in [A4], 20 tunable parameters were used.

Non Linear Programming by Quadratic Lagrangian (NLPQL)

NLPQL is a sequential quadratic programming (SQP) method which solves problems with smooth continuously differentiable objective
function and constraints. The algorithm uses a quadratic approximation of the Lagrangian function and a linearization of the constraints. It
is a code for solving constrained nonlinear optimization problems. It is assumed that all problem functions are continuously
differentiable. In [A5] is presented a special implementation of a sequential quadratic programming (SQP) method. Proceeding from a
quadratic approximation of the Lagrangian function and a linearization of constraints, a quadratic programming sub-problem is
formulated and solved by QL. Depending on the number of nodes of the distributed system, objective and constraint functions can be
evaluated simultaneously at predetermined test points along the search direction. The parallel line search is performed with respect to
an augmented Lagrangian merit function. Moreover, a non-monotone line search is performed in error situations where the line search
cannot be stopped within a given number of iterations. All theoretical convergence properties of the SQP algorithm remain satisfied.

Genetic Algorithms

Genetic algorithms (GA) were invented in the 1970s and even today new variations are being created [A3]. They are local search
techniques used to find approximate solutions to optimization and search problems. This class of evolutionary algorithms uses
techniques inspired by evolutionary biology such as inheritance, mutation, selection, and crossover or recombination. They are
typically implemented as a computer simulation, in which a population of abstract representations (called chromosomes) of candidate
solutions (called individuals) to an optimization problem, evolves toward better solutions. The evolution starts from a population of
completely random individuals and occurs in generations. In each generation, the fitness of the whole population is evaluated, multiple
individuals are stochastically selected from the current population (based on their fitness), and modified (mutated or recombined) to
form a new population. The new population is then used in the next iteration of the algorithm.

Final considerations

The three previous described algorithms together with the response surface method and SHERPA (Systematic Hybrid Exploration that is Robust,
Progressive, and Adaptive), were taken under consideration in the benchmark study [A4]. The goal of SHERPA was pushing the limits of the all
monolithic search methods that work effectively only when it is possible to accurately compute gradients of the responses with respect to the
variables. Some methods work only for continuous or discrete variables (but not both), or for a relatively small number of variables. “Some methods
find only the nearest local optimum as opposed to performing a wider search for the global optimum. And some methods require a relatively large
number of design evaluations to be performed in order to find an optimal solution. No single monolithic method or algorithm works best on all, or
even a broad class of problems” [A3]. SHERPA uses the elements of multiple search methods simultaneously in a unique blended manner. This
approach attempts to take advantage of the best attributes of each method. Attributes from a combination of global and local search methods are used,
and each participating approach contains internal tuning parameters that are modified automatically during the search accordi ng to knowledge gained
about the nature of the design space. These characteristics are on the base of the good results obtained during the cited benchmark study.

References
[A1] Kuhn, H. W., Tucker, A. W., “Nonlinear programming - Proceedings of 2nd Berkeley Symposium”, Berkeley, University of
California Press, 1951, pp.481-492.
[A2] Avriel, Mordecai,”Nonlinear Programming: Analysis and Methods”, Dover Publishing – 2003, ISBN 0-486-43227-0.

[A4] Chase, N., Rademacher, M., Goodman, E., Averill, R., Sidhu, R., “A Benchmark Study of Optimization Search Algorithms”;
http://www.redcedartech.com/pdfs/SHERPA_Benchmark_0110.pdf.
[A5] Yu-hong Dai, K. Schittkowski, ”A sequential quadratic programming algorithm with non-monotone line search”, Pacific Journal
of Optimization, 2008, Vol. 4, 335-351
Figure 1: geometry (a) and mesh representation (b) of the solid separator.

Input of initial
geometric
parameters
For case = 1:N

Generation of
CAD model
new geometric
generation
parameters
Steps of the automated loop

Star-CCM+
(Mesh + Solving)

Results
evaluation

No Maximum
case number?

Yes

Best design

Figure 2. Conceptual outline followed by the optimization process.

Figure 3 – Geometrical parameters

Figure 4 – PDF for each geometric parameter

Figure 5 - History of all simulation

Figure 6 – Powder volume fraction contour-plot

Figure 7 - Temperature contour-plot

 Need for optimization

Choose design variables

Constraints definition

Formulate objective function

Setup variable bounds

Choose optimization algorithm

Generate solution(s)

Figure A1: A scheme of the optimal design procedure

 Parameter Symbol Size [mm]
Diameter of upper vortex finder D1 3370
Diameter of upper cylindrical part D2 6250
Diameter of lower vortex finder D3 2320
Diameter of lower cylindrical part D4 4330
Diameter of upper spigot D5 780
Diameter of lower spigot D6 500
Length of upper outlet tube L1 8014
Length of upper vortex finder L2 *
Length of upper conical part L3 5670
Length of lower cylindrical part L4 4583
Length of lower conical part L5 4119
Length of upper cylindrical part L6 2891
Length of lower outlet tube L7 11700
Length of lower vortex finder L8 *
Table 2 - Geometry of the cyclone

Tgas-out ≡ Tsolid-out pressure drop n° of Elapsed- ϵσ ϵT

mesh density σ Cores (*)
[°C] [Pa] volumetric cells time [s] [%] [%]
coarse
686 0.932 626 380000 10 7127 - -
(base- line)
medium 685 0.955 777 772000 22 26831 2.5 0.1
fine 681 0.960 906 1367000 40 32104 2.9 0.7
Table 2 – evaluation of different mesh sizes

(*) To balance the load per core, the number of cores used for each simulation has been evaluated assuming the cells/core ratio
approximately equal to 35000 (from the best-practices criteria).

Tgas-out ≡ Tsolid-out Elapsed-time

Turbulent models σ
[°C] [s]
k-ϵ 685.77 0.932 9060
RSM 685.73 0.926 9201
k-ω 683.67 0.923 9408
Table 3 – Comparison between different turbulent models

Gas mass flow rate Solid mass flow rate Gas/solid

Inlet port temperature.[°C]
[kg/s] [kg/s]
A 31 25.5 850
B 36 23.5 600
Table 4 - Boundary conditions (see figure 1 (c))

Tgas-out ≡ Tsolid-out
σ
[°C]
Exp. 693 91.5
Calc. 697 92.7
Table 5 – experimental value and simulation results during the validation phase

Parameter Min Max Baseline

α_t [°] 70 100 90
α_b,l [°] 85 95 90
α_b,r [°] 85 95 90
L_t [m] 0 4.5 1
L_b,l [m] 0 5.5 1
L_b,r [m] 0 5.5 1
Table 6 – Geometrical constraints
 Vs. σ Vs. TGAS,OUT
α_t -0.8 -0.33
α_b,l 0.04 0.34
α_b,r -0.23 0.30
L_t -0.10 -0.45
L_b,l -0.17 -0.34
L_b,r 0.14 0.27
Table 7 - PCC for the geometric parameters.

Case σ Tgas-out ≡ Tsolid-out Pressure drop [Pa]

[°C]
#0 92.7 697 659
base-line 93.2 686 626
optimum case 97 682 775
Table 8 – the main configurations.
Highlights

 The paper discusses cement plant optimization problem using CFD and optimization simulation;
 Goals: maximize particle separation and gas/solid heat exchange, keeping pressure losses under control;
 The optimization process allows you to perform design exploration studies within the CFD environment.