Separation Science and Technology

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/lsst20

A new Sauter mean diameter correlation suited

for the gas cross flow in packed bed reactors
based on PSO optimization algorithm

Hajer Troudi, Moncef Ghiss, Najeh Ben Guedria, Mohamed Ellejmi & Zoubeir
Tourki

To cite this article: Hajer Troudi, Moncef Ghiss, Najeh Ben Guedria, Mohamed Ellejmi & Zoubeir
Tourki (2023) A new Sauter mean diameter correlation suited for the gas cross flow in packed bed
reactors based on PSO optimization algorithm, Separation Science and Technology, 58:1, 188-211,
DOI: 10.1080/01496395.2022.2098145

To link to this article: https://doi.org/10.1080/01496395.2022.2098145

Published online: 17 Jul 2022.

Submit your article to this journal

Article views: 56

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=lsst20
SEPARATION SCIENCE AND TECHNOLOGY
2023, VOL. 58, NO. 1, 188–211
https://doi.org/10.1080/01496395.2022.2098145

A new Sauter mean diameter correlation suited for the gas cross flow in packed
bed reactors based on PSO optimization algorithm
Hajer Troudia, Moncef Ghissa, Najeh Ben Guedriab, Mohamed Ellejmic, and Zoubeir Tourkia
a
Laboratory of Mechanics of Sousse, National Engineering School of Sousse, University of Sousse, Sousse, Tunisia; bHigher Institute of Transport
and Logistics, University of Sousse, Tunisia; cAlpha Engineering International, AEI, Sousse, Tunisia

ABSTRACT ARTICLE HISTORY

Cross-flow in packed bed reactors (PBRs) was often encountered in industries. However, the effect Received 24 December 2021
of inlet and outlet nozzle position on the droplet diameter was little considered. In this study, the Accepted 29 June 2022
effect of inlet and outlet nozzle angles of the gas on the spray disintegration was assessed by KEYWORDS
changing the angle (i.e., 0°, 30°, 45°, and 60°). A 3D computational fluid dynamic (CFD) model was CFD; cross-flow; Eulerian–
involved to understand the influence of cross-flow of a single and two-phase gas–liquid flow in PBR. Lagrangian approach; PSO
An Eulerian–Lagrangian approach (E-L) was used for solving the flow pattern and the spray
disintegration. The RNG k-ε model was employed to simulate the turbulent flow. First, an indirect
validation of the cross-flow packed bed was proposed based on a single phase. Second, the
variation in parameters such as superficial liquid and gas velocity, mass flow rate, and gas angle
orientation were investigated and graphically represented. A new correlation of Sauter mean
diameter was developed using particle swarm optimization (PSO) with respect to gas angle
orientation. This correlation was done for 64 cases and validated with existing correlations in the
literature, giving an MAPE of ±12% and indicating a good agreement with the data.

Introduction experimental investigation to control the size and mor­

phology of the sprays in contact to the gas flow. Guo
Packed bed reactors (PBRs) with regular mono-sized
packings have been extensively established in chemical et al.[7] have studied the flow regimes by processing
and industrial processes for gas–liquid or gas–liquid– images taken from the wall region at different gas-to-
solid reactions, such as catalytic reaction, distillation, liquid flow ratios. They have highlighted two key influ­
hydrogenation, air and carbon dioxide separation.[1,2] ential variables: air velocity and injection pressure.
Recently, PBRs are more and more used in precombus­ When the ambient pressure remains low, liquid mass
tion technology where the solid particles of the packed would concentrate on the spray axis. However, with the
bed were used to capture the dioxide of carbon in order increase in the ambient pressure, the spray periphery
to achieve a sufficient reduction of gas emissions.[3] would become dominant and creates a full-cone spray.
These reactors provide higher heat and mass transfer The flow in a PBR is very complicated because of the
rates while maintaining lower operation and mainte­ presence of spherical solid particles of different sizes,
nance costs. The effective design and process optimiza­ and it involves multi phases: gas and liquid phases.
tion of the PBRs are strongly linked to the gas–liquid Similarly, information about liquid flow, e.g., liquid
flow within the packed bed. To date, two designs of flow patterns and disintegration modes, generated by
PBRs have unchanged in practical engineering fields, the particles, either in cross-flow or in other modes
which are co-current and countercurrent flow.[4,5] was rare. Some efforts have been made to experimentally
To the best of our knowledge, both systems were study the flow in a densely packed bed, but such an
investigated for long term; however, few have explicitly experimental method is technically difficult and expen­
addressed the cross-flow in PBRs and little information sive. On the other hand, numerical methods based on
was available on its effect on droplet diameter. This computational fluid dynamics (CFD) are comparably
process was often studied either in moving beds (e.g., more cost-effective and easily controllable and can pro­
rotating packed bed) or in diesel engines. For instance, vide richer information. Nevertheless, recent numerical
Muir et al.[6] have firstly tested the inlet-flow configura­ studies have shown some interesting findings in this
tions to study the pressure drop. They carried out an aspect.[8,9] In general, the most extensively used methods

CONTACT Hajer Troudi [email protected] Laboratoire deMécanique de Sousse (LMS), Université de Sousse, Ecole Nationale d’Ingénieurs
de Sousse (ENISo), Bp.264 Erriadh, 4023, Sousse, Tunisie
© 2022 Taylor & Francis Group, LLC
 SEPARATION SCIENCE AND TECHNOLOGY 189

are categorized as Eulerian–Eulerian and Eulerian– phase) is performed as liquid droplets. In order to close
Lagrangian models based on the different treatments of the system of equations, the RNG k-ε turbulent model is
the spray. The Eulerian–Lagrangian (E-L) has shown its selected to model the slightly turbulent flow at the dis­
advantages in more realistically representing the spray persion packed bed.
characteristics since it allows to build uniform particle
size. The similar method has been successfully used to
resolve complex physical phenomena of droplet breakup Gas-phase governing equations
and coalescence.[10]
With regard to numerical simulation, Theologos et ­ @ρg � �
al.[11]
seem to be the first to numerically explore the þ Ñ: ρg ug ¼ 0 (1)
@t
impact of feed injector geometry on flow and cracking
process in fluidized reactor. They used multiple nozzles @� � � �
and applied simulations to predict their effect on the ρg ug þ Ñ: ρg ug ug ¼ ÑP
@t � � �
reaction yield. Recently, Marek et al.[12] investigated and �T � 2 �
þ Ñ: μg Ñug þ Ñug μ Ñ:ug I
3 g
confirmed that cross-flow regime is an efficient way to
þ ρg g þ Finterfaciales
achieve the lowest pressure drop and the shortest mal­
(2)
distribution factor. However, this research was per­
formed using one single phase, and the effect of gas In the above equations, the subscript g refers to the gas
cross-orientation on maldistribution was omitted. The phase. ρg and ug represent the gas density and velocity,
combined PBRs and cross-flow have been interested by respectively. g is the local gravitational acceleration. P is
the work by Kaskes et al.[13] and Liu et al.[14] the pressure, while Finterfaciales represents the interphase
In view of implementing correlations, most of them momentum transfer, which are the drag, the lift, the
are based on the superficial flow velocity with few on the turbulent dispersion forces, etc[16,17]. These terms are
local velocity in the channel. From a different perspec­ computed from the Lagrangian frame by an alternate
tive, the proposed correlations are intended to describe process through volume averaging method and then
the cross-flow by taking into account the effect of gas incorporated into the Eulerian gas Equation [16].
orientation on the mean droplet diameter.
The objective of the present paper is to investigate the
spray distribution at various inlet gas configurations in Turbulence model
a three-dimensional (3D) PBR using the E-L approach.
There are different turbulence methods that can describe
The model is first validated using the available experi­
the small-scale eddies inside the particles. Over the k-ε
mental data of co-current flow and then modified to add
models, the RNG k-ε model was proven to be able to well
the cross-shaped gas inlet into the original PBR in order
describe the turbulent flow characteristics in the regular
to change the local flow field. The same configuration as
packings (Wang et al.,[18] Troudi et al.,[19] Ullah et al.,[20]
in the study by Du et al.[15] is used, and different forms
Bu et al.,[21] Bai et al.[22]). This assumption matches with
of cross-direction are addressed. Second, the effects of
our study as our geometric model is a packed reactor
superficial velocities, mass flow, and gas inlet orientation
with regular spherical particles. Also, the RNG k-ε was
on Sauter mean diameter were analyzed at fixed packing
confirmed to be adopted for a packed bed with low
size and porosity. Using the method of dimensionless
diameter ratio (Nijemeisland et al.[23]) and suitable for
parameters, two correlations of Sauter mean diameter
swirling flow as it saves computer-running time. The
are developed with the aid of standard LINEST and PSO
RNG k-ε model was solved only in the Eulerian phase
methods. Finally, the numerical model is validated by
and includes the following two additional transport
good agreement between the predicted results and
equations for turbulent kinetic energy, k, and dissipation
experimental data.
rate, ε.
8 � h� � i h � �T � i
< μ
Numerical model ρg ug :Ñk ¼ Ñ: μg þ σ kt Ñk þ μt Ñug þ Ñug : Ñug ρg ε
� h� � i h � � � i
: ρ ug :Ñε ¼ Ñ: μ þ μt Ñε þ cε1 ε μ Ñug þ Ñug T : Ñug 2
cε2 ρg εk
g g σε k t
In this study, three-dimensional simulations of the PBR
(3)
were performed by FLUENT-ANSYS 14.0. As afore­
mentioned, the E-L multiphase approach was used, where μt is the turbulent viscosity, σk and σε are the
and the governing equations were summarized in this turbulent Prandtl number for k and ε, and cε1,cε2, and
section. The gas flow (continuous phase) is solved using cμ are constants. These terms are the same as the stan­
Navier-Stokes equations, and the liquid flow (dispersed dard k-ε, and they are represented as follows:
190 H. TROUDI ET AL.

k2 where µd is the dynamic viscosity of the droplets and dd

μt ¼ ρg cμ ; cμ ¼ 0:085: is their Sauter mean diameter:
ε (4)
cε2 ¼ 1:68; σ k ¼ σ ε ¼ 0:718: , !
X
N XN
3 2
However, cε1 has an expression, which is different from dd ¼ ni di ni di (8)
i¼1 i¼1
the standard k-ε model, and it is written as follows:
� � Here, Re is the
ηn 1 ηn 4:38 � relative
� Reynolds number based on the
cε1 ¼ 1:42 fn ; fn ¼ ; ηn slip velocity ud ug �
�
� 1 þ 0:012η3n
� � �
�T
μt Ñug þ Ñug ρg dd �ud ug �
¼ (5) Re ¼ (9)
ρg cμ ε μg

where CD is the drag coefficient calculated for different

droplet Reynolds regions (transition flow and turbulent
Droplet transport model flow) by the following formula:
8 h � � i
In the Lagrangian frame, transient behavior of each < 1
1þμT
24
μT Re 4
þ 0:333 14:9
þ Re 0:76 0:5 � Re � 1000 Transition flow
CD ¼ � Re �
individual droplet in the injected parcel is tracked by : 8 2þ3μT Re > 1000 Turbulent flow
Re 1þμT
the following equations:
(10)
� � .
!
dud � g ρd ρg Ñp ! where μT ¼ μd μg is the viscosity rate of the droplet
¼ FD ug ud þ þ F c (6) and gas.
dt ρd ρd
where the subscript d denotes the droplet. Moreover, in
collision-dominated gas–liquid two-phase flow, only the Taylor analogy breakup (TAP)
effects of gravity force, drag force, and pressure gradient
force are usually considered. Note that forces like virtual In this section, the coalescence and the breakup models
mass force or Basset force are commonly neglected in are introduced following the Euler–Lagrange approach to
this equation, as the gas density is much lower than the capture the droplet motion and deformation along the
liquid density and the droplets considered are not sub­ cross-gas direction. Droplet coalescence and breakup are
micron. In addition, it should be mentioned that in modeled by O’Rourke et al.[24] and characterized in terms
different separation processes, the discrete phase is gen­ of Weber number “We” relating the inertial force and the
erally coupled with the coalescence and the breakage force due to surface tension. However, the latter is
equations. The breakup is occurred at the narrow voids a complex phenomenon that concerns several assump­
due to the turbulent inertia and viscous fragmentation of tions. For this purpose, some following assumptions are
the liquid, while the coalescence occurred at empty wall. used: (1) the droplets are considered as spheres, (2) each
To improve the separation efficiency, especially in PBRs, droplet breaks into four child droplets of different sizes,
droplet coalescence is undesirable because it leads to and coalescence process arises only between two child
polydispersed droplets and heterogeneous mixing con­ droplets, and (3) droplets move in isotropic homogeneous
ditions in droplets. As a result, droplet breakup should turbulence (it means that statistics of small-scale motion
be increased, whereas coalescence effects should be of turbulent eddies is universal).
avoided. The breakup is characterized by the Weber number
defined as follows:
� �2
Momentum conservation for droplet ρg dd �ul ug �
We ¼ (11)
σc
The momentum interfacial exchange between droplets
and the gas due to droplet break and droplet coalescence When the Weber number “We” exceeds the critical
is taken into account. The droplet-induced drag force is Weber number “Wecr,” breakup happens. However,
considered in the present study. The momentum inter­ these critical parameters must be determined experimen­
facial exchange is calculated using the following tally. In the literature, a wide range of values by various
equation: authors have been presented. Lee et al.[25] have indicated
18μd CD Re that Taylor analogy breakup “TAB” model shows a good
FD ¼ (7) agreement with their experiment, and they have deter­
ρd dd2 24
mined the critical value of water droplet as Wecr = 101.
 SEPARATION SCIENCE AND TECHNOLOGY 191

� �2 sffiffiffiffiffiffiffiffiffiffiffiffi
d2 y Cl μd dy Ck σ c CF ρg �ul ug � dd
þ þ y¼ (12) x¼ (17)
dt ρd r2 dt ρd r3 Cb ρd r2 1:2d32

As we only provided a short overview of the TAP

2x method, the reader can refer to the work by
y¼ (13)
r Stefanitsis et al.[26]and Pramidula et al.,[27] for an in-
depth view.
where “y” is the deformation of the droplet, “x” is the
displacement of the droplet from its spherical (undis­
turbed or deformed) position, and “r” is the new radius Description of the physical model development
of the deformed droplet. dy=dt and d2 y=dt are the rate Physical models
and the acceleration deformation of the droplet, respec­
tively. The values of the dimensionless constants “Cb,” The small scale of PBR developed in this study is
“CF,” “Ck,” and “Cd” are 0.5, 0.33, 8, and 5, respectively. shown in Fig. 1. The PBR consists of a packed bed
If the actual weber number “We” is bigger than cri­ filled with an immobile stack of particles. The parti­
tical one “Wecr” (We > Wecr), breakup and a significant cles are modeled as stationary walls of fixed diameter
deformation occur. Then, the deformation “y” of the dp = 41 mm. The height of the packed bed and the
droplet will be bigger than 1 ðy > 1Þ. The rate and accel­ diameter of the particles are fixed at H = 1.1 m and
eration deformation of the droplet also will take a value D = 0.25 m, respectively. However, the top and
�� � �2 ��
dy d y bottom nozzles of the gas phase were used as design
different to 0 dt �0 ; dt �0 . Otherwise, at an
variables. The liquid is fed into the system from
equilibrium point and when no breakup occurs, the a cone spray injector, placed at the top of the packed
deformation is equal to 1 ðy ¼ 1Þ. bed and centered on the symmetry axis. The gas is
For this reason, the next parameter that is investigated fed from a one side-inlet at 0.5 m above the bottom
is theSauter mean radius “r32,child” of the child droplet of the packed bed. The inlet nozzle of the gas phase
number after breakup. This can be derived as follows: is directly exposed to the packed bed, as can be
r32 clearly figured in Fig. 1(b). For the cross-current
r32;child ¼ dy 2
(14) purpose, the gas goes from the left to the right
4y2
3
ρd r32 ðdt Þ
1þ 3 þ 8σ c producing a tangential movement in contact to the
downward liquid. Thus, the two phases flow through
where r32 ¼ d32 =2 is the size distribution, which is the empty space between the packings.
assumed to follow the Rosin–Rammler distribution. In detail, the computational domains of the differ­
“d32” is the Sauter mean diameter, which is defined as ent PBR geometries are each illustrated in Fig. 2 and
the ratio of the number of correctly matched particles to named as Model 1 to 3 according to their gas-flow
the number of extracted particles, as follows: configuration. As seen, the particle packing is very
P 3 common for the six models, and the inlet/outlet
nd nozzles are treated as rectiligne section.
d32 ¼ P d2 (15)
ndd
● Model 1: the gas inlet and outlet are tangent to the
where “dd” represents the diameter of individual droplet wall, along the column radius at the height of the
and “n” is the total number of the droplets at the posi­ packed bed. The gas nozzle exit is placed on the
tion. Following Lee et al.,[25] the diameter of each child right edge of the wall.
droplet “dd” is described as a function of the root- ● Model 2: the gas inlet nozzle is paced in the center
normal distribution: of the wall and perpendicular to the to the spray
� �! axis.
x 1 x μg 2 ● Model 3: the configuration is similar to Model 1,
f ðdd Þ ¼ pffiffiffiffiffi exp (16)
2dd σ c 2π 2 σc except for the gas nozzle exit. The latter is placed on
the left edge of the wall.
Therefore, changes in the droplet’s position is expressed ● The aforementioned models are also discussed but
on the basis of the ratio of the actual child droplet with an inclined inlet duct by an angle of 45° with
diameter to the Sauter mean droplet diameter: the horizontal axis.
192 H. TROUDI ET AL.

Figure 1. Schematic diagrams of three-dimensional calculation domain with the boundary conditions: (a) macro- (b) meso-, and (c)
micro-scales.

Boundary conditions on the wall. Initially, air is selected as primary phase and
liquid as secondary phase. The parameters of spray nozzles
In addition, the different boundaries of the computational
used in the simulations are listed in Table 3.
domain are labeled in Figs. 1 and 2 and identified in
Table 1. Water–air system is chosen for this study, as
successfully done in Du et al.[15] (see Table 2). The sur­
Solver setting
rounding condition is set to atmospheric pressure and
room temperature (25°C) in order to simplify the numer­ The numerical simulations were performed on a desktop
ical procedure. Two inlets are used to generate fully devel­ workstation (Intel® Xeon® CPU E5-2670 v3 @ 2.30 GHz)
oped flow in which one is set to be velocity inlet and other equipped with 2 CPUs with 12 cores. All simulations
to be mass flow inlet. Outflow condition is imposed at the were solved following two fractional steps. In the first
down and right down boundary faces. Symmetry condition step, the Lagrangian spray was set to zero, considering
is set on the top, and finally, no-slip condition is imposed only the conservation equations of gas phase. In
 SEPARATION SCIENCE AND TECHNOLOGY 193

Figure 2. Geometric drawings of inlet/outlet nozzles: (a) top view and (b) side view including α = 0° and α = 45°.

Table 1. Boundary conditions. Table 3. Geometric parameters.

Name Boundary condition Model 1/Model 2/Model 3
Inlet 1 Velocity inlet Nozzle width b 0.062 m
Inlet 2 Mass flow inlet Nozzle height h 0.07 m
Outflow 1 Outflow Length of nozzle l 0.126 m
Outflow 2 Outflow Spacing between the nozzle and the spray 0.03 m
Symmetry Symmetry Angle of attack α = 0° and α = 45°.
Wall Wall

the second step, the spray was allowed to evolve in the

Table 2. Physical properties used for the simulation. loaded velocity fields. The discretization of the gas
Continuous phase: air phase-governing equations is based on the finite volume
Density, ρ (kg/m3) 1.225
Dynamic viscosity, μ (Pa.s) 1.789e−5 method employing a staggered grid and solved by the
Dispersed phase: water SIMPLE algorithm. The under-relaxation factor was set
Density, ρ (kg/m3) 998.2 as 0.3 for the purpose of numerical stability, and the time
Dynamic viscosity, μ (Pa.s) 1.003
Surface tension, σ (N/m) 0.0728 step was set to 10−3. With this setup, each simulation was
194 H. TROUDI ET AL.

very expensive, e.g., a single simulation of 5 s takes 20– a high-quality structured mesh. Multiple blocks corre­
30 days of CPU time using 12 cores to achieve the sponding to the particles, inlet/outlet nozzles, and the
pseudo-steady state depending upon the gas angle orien­ cone spray have been done more precisely, neglecting the
tation and the model. The residuals of all the variables boundary layers. For the four models, most of the cells are
were set to 10−4. To improve numerical stability, linear­ concentrated in the packed bed region where a specific
ization of source terms for the discrete phase momen­ refinement with a factor of 5 is applied in each particle or
tum exchange was applied. block. Another mesh refinement is placed at the cone
spray injector with a factor of 2.
In order to quantify this study, the volume fraction of
Results and discussion
the liquid phase is also examined in Fig. 4(b). Here, this
In this section, two sets of experiments were conducted parameter is adjusted by dividing the summation of the
to validate the dynamic characteristics of gas–liquid dispersed-phase concentration in each computational cell
flows. Initially, we have validated the CFD simulation by the total cell volume to derive the concentration of the
for a single phase flow (only gas) with the experiments liquid phase in this cell, as shown in the following equation:
by Suekane et al.,[28]and then, we have checked the
whole gas–liquid flow with the experiment by Nemec P
cell � �
Cc ρd
et al.[29] Afterward, the section is divided as follows. Liquid volume fraction ¼ c¼1 (18)
First, the vorticity inside the bed is studied. Second, the Vcell
process of the droplet motion in the packed bed is
For validation, the experimental data obtained by
indicated. Finally, the effects of some influencing factors
Nemec et al.[29] are used. In their work, they have
are studied separately.
reported the co-current flowing regime in packed bed
of particle diameter dp = 3mm and superficial gas velo­
Analysis of one single phase city usg = 0.015m/s. So, their predicted volume fraction
value is estimated to take 0.2. To do this, our simulation
Suekane et al.[28] have taken a conventional packed bed
inputs are chosen to closely resemble those of Nemec
with two layers of packings of dp = .028 m of diameter
et al.[29] Fig. 4(b) shows that for the last three meshes
and with ratio equal to D/dp = 8, as shown in Fig. 3(a).
(Mesh #2 to #3), the calculated volume fractions seem to
The bed is of porosity ε = 0.476. The velocity results of
be in better agreement to each other and the experimen­
four mesh sizes are shown in Fig. 3(d). As it can be seen,
tal data. They generally vary from 0.19 to 0.22, resulting
the values are almost identical with differences of 26.5%,
a quite small errors compared to Nemec et al.[29]
0.5%, 1.0%, and 1.5% when the interval size is 0.0021,
(between 1.5% and 12%). In addition, Fig. 4(c) depicts
0.00126, 0.0007, and 0.00042 m, respectively. As a first
the computational time for the four meshes. It can be
comment, the velocity profile monotonically increased
inferred that Mesh #2 has a low computational time.
(−1 < x/r < 0) and monotonically decreased (0 < x/r < 1),
However, the computational time required for Mesh #4
and the slope is symmetrical at the center (x/r = 0). It can
is increased by a factor of 1.12 compared to Mesh #1.
be noticed that a maximum velocity from 1.8 to 2.4uo is
To close the discussion in this section, the volume
obtained around the boundary wall of the cone spray
fraction contours of the dispersed phase is successfully
injector. The flow becomes more irregular in compar­
depicted in Fig. 5 with respect to the four meshes.
ison to the areas of the packed bed.
Generally, although the slight difference between the
contours is observed, they represent well the spray. An
Analysis of two phase improvement on the spray shape is highlighted in the
simulation using Mesh #3 and Mesh #4.
Test grid independency and validation
As a consequence, through Figs. 4 and 5, Mesh #3
Test grid independency has been predicted for four differ­
with ≈ 13 M cells is selected due to its ability to have
ent meshes consisting of Mesh #1 ≈ 8 M, Mesh #2 ≈ 10 M,
a good convergence and is computationally inexpensive
Mesh #3 ≈ 13 M, and Mesh #4 ≈ 21 M cells, respectively.
to effectively describe the volume fraction.
This study is undertaken in order to minimize the errors
caused by the narrow arrangement of the particles. Before
entering to this study, lets take an insight into the mesh. Liquid volume fraction and spray morphology
Figure 4(a) shows an example of a completely structured A qualitative comparison of models is described here
and unstructured mesh with hexahedral and quadrilateral with the help of the volume fraction contours and the
cells generated by the ANSYS ICEM 14.5 software. Due to iso-surfaces of the volume fraction, as illustrated in
the complexity of the geometry, it is impossible to create Figs. 6 and 7. Under the same operation parameters
 SEPARATION SCIENCE AND TECHNOLOGY 195

Figure 3. (a) Two vertical layers of spherical packing. (b) volume element. (c) gas velocity distribution in m/s. (d) Simulation results for
gas velocity in different interval size (horizontal cross-sectional at a height of 0.8 m).

and looking for all the six different cases, we note that time, and no further breakup occurs. Nevertheless, the
there is three main pulse shapes that behave very differ­ spray is collected and tightly packed at the center of the
ently. They are the stable undulation shape, unstable bed. In this case, this observed behavior might not
undulation shape, and the bag shape. In Model 1 promote the breakup and do not correspond to the
(Fig. 6(a)), it is shown that the spray can pulverize reel goal of this study.
more easily the introduced droplets than other models, In Model 2 (Fig. 6(b)), it is shown that there is
regardless of the gas flow. Due to the tangential direction a clearly primary longitudinal instability, created
of the gas flow, a stabilized undulation shape is first directly at the spray outlet, which is linked to the fact
created at the spray outlet and kept on growing in size of the gas shear instability. Here, smaller droplets are
around the spray centerline. The liquid enters the bed shown to be produced by the primary pulverization or
through a larger part within the first two layers of par­ atomization process. Contrary to the model 1, it is
ticles and goes with a uniform distribution. This distri­ observed that there is an irregular distributions of smal­
bution is shown to be compact and stable in space and ler droplets in the axial and radial directions, which is
196 H. TROUDI ET AL.

Figure 4. (a) Generated meshes for the geometry, (b) comparison of the predicted volume fraction by CFD against the experimental
value of Nemec et al.,[29] and (c) computation time.

good for the improvement of PBR performance. The the other models. As a result, the configuration of the
most of the liquid are not directed by gravity into the inlet nozzle constrains the motion of droplets as they
center of the bed. They are flowed on the walls and sprayed and hit the first row of particles, resulting in an
discharged to the opposite outlet (Outflow 1) together arduous flow of the spray.
with the gas cross-flow. The centrifugal force of the gas Finally, Model 3 (Fig. 6(c)) reveals a strong similar
phase and droplets’ inertia modify the droplets’ trajec­ behavior with Model 1, as their configurations are
tories toward the wall, where they coalesce and form equivalent. The difference between them is not promi­
a number of liquid films. This indicates that the spray is nent; however, the undulation shape becomes slightly
accelerated much stronger by the surrounding gas than unstable and comes outside from the spray centerline
 SEPARATION SCIENCE AND TECHNOLOGY 197

Mesh #1 Mesh #2 Mesh #3 Mesh #4

Gas flow

Liquid volume fraction

Figure 5. Comparison of liquid phase volume fraction under different meshes at (y-z) plane and stopped at t = 0.33 s.

due to the declination outflow of the gas velocity. The droplets is assumed to occur according to Weber’s ana­
droplets are shrinking on the bottom layers of the bed lysis for capillary instability. Thus, the outlet gas section
and formed large ones or the so-called bags. The latter that is near the wall can behave as a trapping zone of the
droplet size is much larger in Fig. 6(c) than the former longer ligaments.
one in Fig. 6(a) due to the atomization. Figure 7(c) is roughly the same as that shown in Fig. 7
In the case of oriented gas nozzle (α = 45°) and inclined (a). Larger droplets are mainly collected by the centrifu­
gas distribution, it is found that the spray and droplet gal force in the packed bed body, and they are captured
dispersion responded fast to this change. By careful obser­ less on the gas outlet section and more on the bottom.
vation, it is seen that the volume fraction contours for Taking into account all the figures previously
Models 1 to 3 are not symmetric with respect to the spray explained, Model 2 is the most suitable model
centerline due to the combined effect: the centrifugal among the ones tested. It can be concluded that if
force of the gas phase and the larger inertia of the liquid the gas nozzle inlet is placed in the center of the bed
phase, as seen in Fig. 7. This force is further developed by and is oriented enough, the dominant process is the
the strong gas entrainment in the spray outlet area, which breakup of the spray and the generation of a very
allows therefore the formation of a relevant inner recircu­ high amount of tiny droplets. If the gas nozzle inlet
lation zone, which can be seen in Fig. 5. Besides that, is tangent to the bed, the dominant process is the
a radial and axial dispersion of the droplets differs also atomization of the spray. Furthermore, it is impor­
a lot compared to Fig. 7. Their dispersion plays an impor­ tant to note that the cone spray angles in Models 1
tant role in PBR performance including the evaporation and 3 are larger than the ones observed in Model 2,
and the heat and mass transfer. yielding higher atomization and slightly larger pene­
In Fig. 7(a), the spray and initial atomization are no tration. Earlier work by Jadidi et al.[30] has demon­
longer symmetrical in the radial direction but are mainly strated that the presence of ligaments is very
concentrated in the downward facing direction. The important in order to control the droplet size dis­
droplets in the lower outflow take a bag shape. tribution and spray/droplet trajectories and evapora­
Figure 7(b) shows that the spray is composed of tions. These findings support the idea of Model 2 to
a newly created droplets, which appear in the packed correctly predict droplet dispersion.
bed wall and lead to the formation of the so-called
ligaments. This is shown as the first breakup. A part of
Collection efficiency
the spray is pushed on to the opposite surface of the bed
Nevertheless, the collection efficiency “η” is another
and merged with the neighbor droplets. This merging indicator of wettability on particle surface which is ran­
gives rise to a much thicker liquid that forms ligaments. ged from 0 to 1.
Eventually, such ligaments detach from the wall, con­
tract, and breakup to reform droplets with their spheri­ Traped
η¼ (19)
cal shape. The second breakup from ligaments to Injected Suspended
198 H. TROUDI ET AL.

(c)

Figure 6. Liquid volume fraction iso-surfaces and contours for liquid volume fraction equal to 0.5 as shown in the cross-sectional plane.
(a) Model 1: α = 0°, (b) Model 2: α = 0°, and (c) Model 3: α = 0°.
 SEPARATION SCIENCE AND TECHNOLOGY 199

Figure 7. Liquid volume fraction iso-surfaces and contours for liquid volume fraction equal to 0.5 as shown in the cross-sectional plane.
(a) Model 1: α = 45°, (b) Model 2: α = 45°, and (c) Model 3: α = 45°.
200 H. TROUDI ET AL.

The overall collection efficiency for a specific particle cut General form of Sauter mean droplet diameter
size is defined as the ratio of the droplets removed by the
In developing an optimum design of PBR, droplet dia­
bed to the droplets entering the bed:
meter and interfacial area are the most important out­
� � puts that need to be known to judge the performance.
Cout
η¼ 1 � 100 (20) The goal is to develop a general correlation derived from
Cin
numerical results, which can be used for all the six
models. The correlations depend on the most common
where Cin and Cout are the inlet and outlet droplet
nondimensional parameters, namely the gas feed ratio,
number concentrations, respectively.
liquid feed ratio, gas–liquid ratio, and the contact angle
Figure 8 shows the “η” values as functions of
orientation of the gas entrance. These are the same
Sauter mean droplet diameter from 0.02 to
parameters used by Xie et al.[31] except the orientation
1.7 mm at a fixed liquid flow rate of 0.12 kg/s
angle “α.” Accordingly, the basic correlation represent­
and a fixed superficial gas velocity of 0.42 m/s.
ing the droplet diameter and the interfacial area may be
The mean droplet diameter is varied over the
put in the following form:
range of 0.02–1.7 mm, which covers a wide variety
of spray systems in PBR. It is clear that the collec­ !a2 !a3 � � � �
tion efficiency increases with increasing the droplet usl usg L a4 180 α a5
f ¼ a1 � � � �
size for coarse particles, which, in turn, enhances �usl þ usg � �usl þ usg � G α0
the full uniform distribution. As η values are closer (21)
to 1, the mixing phenomenon of the two phases
becomes more dominant and the particles become Here, the liquid feed ratio defines the ratio between
more wettable. As η values are closer to 0, the latter the liquid flow rate and the sum of gas and liquid
generally characterizes that the particle surface as flow rates at the inlet. The values of gas superficial
non-wettable. It is because that inertial impaction is velocity are in the range of 0.041 and 0.0273 m/s, and
the dominant collection mechanism for coarse dro­ the values of liquid superficial velocity are 0.121,
plets. It is concluded that using oriented gas inlet 0.419, 0.500 ,and 0.857 m/s for all the simulations.
(angle higher than 45°) and a raisonnable gas– These values are regulated by changing the inlet
liquid ratio simultaneously has the most effect on velocities. The effect of changing L/G is investigated
increasing collection efficiency compared to other where the mass flow rate of the liquid is varied from
models. 0.01 to 0.2 kg/s, while the gas mass flow rate is kept

Figure 8. Collection efficiency as a function of d32.

 SEPARATION SCIENCE AND TECHNOLOGY 201

constant at 0.14 kg/s. The angle orientation is also To do this, a logarithm function is further applied to
studied by varying only two angles (α = 0° and Eq. (22) in order to obtain a linear form, which is
α = 45°) and limiting those that higher than described in the following equation:
α > 60°. This is because that the inclination angle !
effect disappears at large angles, and most of the gas usl
logð f Þ ¼ logða1 Þ þ a2 log �� �
inlets in PBRs are arranged at the horizontal position usl þ usg �
! � �
or weakly inclined position. The initial angle orienta­ usg L
tion αo is taken as 180° because minimum results are þ a3 log � � � þ a4 log
u þ usg � G
reached at this angle. The dependency of such para­ � sl �
180 α
meters like the gravity and the void fraction is not þ a5 log (22)
included, as the gravity did not change during the α0
numerical simulations and the void fraction did not Eq. (22) can be also written in terms of:
change along the vertical direction of the packings. In
addition, the correlations are defined in order to logð f Þ ¼ logða1 Þ þ a2 x2 þ a3 x3 þ a4 x4 þ a5 x5
assess the uniform distribution of the liquid phase. (23)
The coefficients (a 1 , a 2 , a 3 , a 4 , and a 5 ) are
In this way, it is possible to estimate the variables (x2, x3,
empirical constants whose values are derived using
x4, and x5) and to calculate the unknown constants (a1,
multivariable linear regression. In this study, two
a2, a3, a4, and a5).
useful methods are explored: the LINEST method
8 � �
using Excel and the particle swarm optimization > usl
(PSO) method using MATLAB. The two methods >
> x 2 ¼ log
>
> ju þu j
>
> � sl sg �
are the most appropriate ones and have an exten­ < usg
sive application in modeling area. The used PSO x3 ¼ log u þu (24)
j j
>
>
sl
�sg
algorithm is developed by Ben Guedria[32] based >
> x4 ¼ log L
>
> � G �
on the flocking and schooling techniques of birds >
: x ¼ log 180 α
5 α0
and fish.

Table 4. Sauter mean diameter “d32” calculated for different gas and liquid superficial velocity.
Gas flow rate Liquid flow rate Gas superficial velocity
Observations G (kg/s) L (kg/s) usg(m/s) Liquid superficial velocity usl (m/s)
1 0.14 0.01 0.121 0.041
2 0.14 0.07 0.121 0.041
3 0.14 0.1 0.121 0.041
4 0.14 0.2 0.121 0.041
5 0.14 0.01 0.121 0.0273
6 0.14 0.07 0.121 0.0273
7 0.14 0.1 0.121 0.0273
8 0.14 0.2 0.121 0.0273
9 0.14 0.01 0.419 0.041
10 0.14 0.07 0.419 0.041
11 0.14 0.1 0.419 0.041
12 0.14 0.2 0.419 0.041
13 0.14 0.01 0.419 0.0273
14 0.14 0.07 0.419 0.0273
15 0.14 0.1 0.419 0.0273
16 0.14 0.2 0.419 0.0273
17 0.14 0.01 0.500 0.041
18 0.14 0.07 0.500 0.041
19 0.14 0.1 0.500 0.041
20 0.14 0.2 0.500 0.041
21 0.14 0.01 0.500 0.0273
22 0.14 0.07 0.500 0.0273
23 0.14 0.1 0.500 0.0273
24 0.14 0.2 0.500 0.0273
25 0.14 0.01 0.857 0.041
26 0.14 0.07 0.857 0.041
27 0.14 0.1 0.857 0.041
28 0.14 0.2 0.857 0.041
29 0.14 0.01 0.857 0.0273
30 0.14 0.07 0.857 0.0273
31 0.14 0.1 0.857 0.0273
32 0.14 0.2 0.857 0.0273
202 H. TROUDI ET AL.

Figure 9. Probability density function PDF for the droplets based on Sauter mean diameter (G = 0.14 kg/s).

In this section, an attempt has been made to predict Brief analyses

Sauter mean diameter as a function of parameters. Figure 9 shows the droplet size distribution, denoted by
!a2 !a 3 � � �
PDF, under different liquid mass flow rates. The PDF is
�
d32 usl usg L a4 180 α a5 defined as the number of droplets of diameter i to the
¼ a1 �� � � �
D usl þ usg � �usl þ usg � G α0 total number of droplets and obtained by using the
(25) following equation:

Having divided by the diameter of the column “D,” the Number of droplets of diameter i
PDF ¼ (26)
Sauter mean diameter has also become dimensionless in Total number of droplets
Eq. (25). Table 4 lists the various simulation cases to
evaluate the Sauter mean diameter. The cases have been As seen in Fig. 9, the Sauter mean diameter is about 0.5
implemented at same average initial droplet size do to 4 mm and in agreement with previous data from the
= 2.3 mm and with four different gas superficial velocity, literature. When the flow rate of the continuous phase is
“usg,” two different liquid superficial velocity, “usl,” one unchangeable, an increase in the mass flow rate of the
gas flow rate, “G,” and four different liquid flow rates, dispersed phase changes the PDF from a unimodal to
“L.” Overall, the number of observations is estimated as a multimodal distribution. Because the increase in “L”
4 × 2 × 1 × 4 = 32 for each angle orientation “α” and 64 broadens, the droplet size distribution shifts to larger
for both. In reality, a set of cases have been implemen­ droplet size and increases the coalescence frequency.
ted, but we conserved only the suited ones and excluded Physically, these trends can highlight the level of turbu­
the repetitive results. lence existing in the dispersed phase and can be
 SEPARATION SCIENCE AND TECHNOLOGY 203

Figure 10. Droplet size distribution under various superficial gas velocity.

Figure 11. Scatter plot of correlation between d32 vs. usg.

explained by the fact that the droplet breakup and coa­ 1.5 to 0.5 mm, with the increase in superficial gas velo­
lescence become insignificant as the droplets in the spray city. For usg = 0.857 m/s, a higher gas entrainment hits
become progressively diluted and dispersed with time. the liquid jet and a change of more than 48% is detected.
Another parameter that influences the droplet size It is observed one peak of PDF around d32 = 0.5 mm,
distribution is the superficial gas velocity of the contin­ which indicated that some small droplets were provided
uous phase. Figure 10 depicts the change of droplet size by the large, trapped ones. As the superficial gas velocity
distribution under different “usg.” At usl = 0.121 m/s, the is less, the PDF becomes wider as a result of difficult
droplet size distribution shifts from larger- to smaller- entrainment of the gas phase. This leads to more liquid
size droplets and the dominant droplet size shifts from holdup and coalescence.
204 H. TROUDI ET AL.

Figure 12. Scatter plot of correlation between d32 vs. usl.

Figure 13. Scatter plot of correlation between d32 vs. L/G.

Simple linear fit variable. Overall, Table 5 shows that the predicted
Figures 11–14 show plots of linear correlations mea­ values agree well with the numerical data, with
sured between two variables. The R-square “R 2 ” relatively high “R 2 ” values. Another criterion,
parameter evaluates the goodness of fit in predicting “p-value,” is defined in the program, which tells
the values of the dependent variable. It takes “1” if about the presence of probability of each variable
the regression fits perfectly and “0” if it fits no in the defined correlation. The value of “p-value” is
better than the simple mean of the dependent also between “0” and “1,” but conversely, the closer
 SEPARATION SCIENCE AND TECHNOLOGY 205

Figure 14. Scatter plot of correlation between d32 vs. α.

Table 5. Simple linear fit for Sauter mean diameter by y = ax + b.

x Model function R2 SE p-Value
�� ��
x2 ¼ Log usl ��usl þ usg �� y ¼ 0:2476x2 1:476 0.485 0.003835371 0.044318065
� �
x3 ¼ Log usg �usl þ usg � y ¼ 0:8429x3 2:298 0.7618 0.094202 0.002703329
x4 ¼ LogðL=GÞ y ¼ 0:9393x4 1:073 0.96 0.012904306 2.1335E-11
x5 ¼ Logð180 α=α0 Þ y ¼ 0:391x5 1:976 0.025 0.049773215 0.000328714

!a2 !a3 � � � �
Table 6. Multiple linear fit for Sauter mean diameter by y = a1 usl usg L a4 180 α a5
+ a2x2 + a3x3 + a4x4 + a5x5. f ¼ a1 � � � �
�usl þ usg � �usl þ usg � G α0
ai SE p-Value
Logða1 Þ �� �� −1.10620524 0.093873 5.91E-13 (27)
x2 ¼ Log ul ��ul þ ug �� a2 ¼ 0:003544481 0.013278 0.836121
� �
x3 ¼ Log ug �ul þ ug � a3 ¼ 0:022360598 0.069039 0.76124 The final regression of “d32” numerical data yields:
x4 ¼ LogðL=GÞ a4 ¼ 0:909898355 0.072731 3.27E-13
x5 ¼ Logð180 α=α0 Þ a5 ¼ 0:08466176 0.10512 0.441689
!0:0035 ! 0:0223
d32 usl usg
¼ 0:3308 �� � � �
D usl þ usg � �usl þ usg �
Table 7. Regression statistics. � �0:9098 � �
Multiple R 0.9833488
L 180 α 0:0846
R2 0.966974863 G α0
Adjusted R-square 0.962419671
SE 0.036849531 (28)
Observations 64
Figure 15 shows the linear functions between y ¼
Log ðd32 =DÞ and x ¼ Log ðxi Þ, where x denotes the
to zero indicates that the variable is more effective, dimensionless parameters, and ai is their coefficients.
i.e., if the variable’s “p-value” was zero, it definitely Here, “a2” of the superficial liquid velocity has kept its
should be present in the correlation. positive sign but with very low value. The positive sign
specifies the higher rise of diameter of droplets, while the
Multiple linear fit very low value of “a2” indicates that the adjant of the
This method is able to predict “d32” using the superficial gas velocity will reduce the rise of droplets by
LINEST method and to combine the results of its velocity and the liquid will have a less influence on the
Figs. 11–14 together. Here, R2 and SE were found droplet size. The opposite, “a3,” has kept its negative sign
to be 0.892 and 0.031, respectively, as illustrated in with a value equal to “a2 = −0.0223.” Here, the normal­
Tables 6 and 7. ized Sauter mean diameter shows an inverse trend with
206 H. TROUDI ET AL.

Figure 15. Linear functions between y ¼ Logðd32 =DÞ and x ¼ Logðxi Þ.

Figure 16. Comparison between d32/D predictions calculated by Eq. 28 with CFD simulations.

the increase in superficial gas velocity. This is due to the parameter as apparent from the power of the fourth
higher local turbulence between phases for all the con­ number “a5 = −0.0846.” This higher local turbulence is
figurations. “a4” is higher than other coefficients, espe­ attributed to the inclined entertainment of the gas acting
cially that the fraction inside the third number is closer on the droplets, which lowers the droplet rising velocity.
to unity “a4 = 0.9098.” A higher liquid flow rate leads to
the generation of a large number of micro-droplet ele­
ments in the packing zone, which then flow into the void Comparison between CFD and correlation results
zone. As a result, a larger gas–liquid contacting area Figure 16 depicts a comparison between CFD and
could be obtained to increase coalescence in the void Eq. 4.8. Here, the error in prediction, MAPE, is calcu­
zone. Also, the angle orientation “α” is an important lated as the mean absolute predicted error, given by:
 SEPARATION SCIENCE AND TECHNOLOGY 207

Figure 17. Comparison between d32 predictions calculated by Eq. 28 with experimental and CFD simulations.

Ntests �� � Song et al.[34]:

1 X �Correlation }Exp} or }CFD}��
MAPE ¼ � �
N tests i¼1 }Exp} or }CFD} ! 0:44 ! 0:08 ! 0:3 !0:16
d32 ρg u2sg deq ρl u2sl ρl μl
(29) ¼ 3:745
do σg ρg u2sg ρg μg
where Ntests is the total number of numerical simulations (31)
or tests.
It is visible from Fig. 16 that “d32” rises with lower “α.” Ingebo et al.[35]:
This agrees with the general trends observed in ! 0:25 ! 0:25
Subsection 3.3.a for “α = 0° and “α = 45°.” d32 usg deq ρg ρg u2sg deq
¼ 3:9 (32)
In order to evaluate the availability of Eq. (28), do μg σg
a comparison between predicted results from the sug­
gested correlation and those selected from experimental Figure 17 shows that there are values that are located above
data is investigated in Fig. 17. The results are verified and below the central area and limited by an interval that
using the data of Berna et al.,[33] Song et al.,[34] and covers almost ≈95% of the data for each correlation. In
Ingebo et al.[35]. These correlations are often studied general, it is worth noting that most experimental data
based not onlyon the flow conditions and the column predict little higher “d32” than the correlation with an error
diameter but also they tend to be specific to the used MAPE = ±75%. The error is so huge that this direct com­
fluid properties (viscosity, density, and surface tension). parison is not valid. As pointed out by Song et al.,[34] the
However, it is important to note that the suggested most popular droplet efficiency in cross-flow problems,
correlation has limited those physical properties and is a relatively moderate accuracy is given within ±30%.
only optimized for use within our particular model. Besides, the model by Berna et al.[33] captures significant
Berna et al.[33] noted that Sauter mean diameters are details of the droplet size distribution. Ingebo et al.[35] fail to
typically correlated using power law equations with the predict “d32,” and most predicted values are higher than the
general form, actual ones and placed out of the confidence interval. So,
� � � � � � !0:31 none of these correlations can make satisfactory predictions
d32 usg deq ρg 0:33 usl dNozzle ρl 0:11 ρl u2sl dNozzle 0:68
ρl
¼ 0:11 for “d32.” This spreading indicates that the proposed correla­
do μl μl σl ρg
tion needs further work. Consequently, a modified correla­
(30)
tion on the basis of the previous one may apply.
208 H. TROUDI ET AL.

� � ��
A new correlation for d32 “ exp a6 180α0 α ” is induced to characterize the effect
Here, the effect of “α” is not really observed. Because of of “L/G.” To do this, further “α” is required to discover
simplicity of the LINEST method, it has some limita­ the optimal range of our results. In addition to α = 180°
tions. The correlations become more evident if the area and α = 45°, two supplementary angles α = 30° and
of the inlet gas section is investigated. An attempt was α = 60° are tested, and the total number of the simula­
made to a new correlation to obtain a balance between tions is incremented to Ntests = 128.
“d32” and “α” and to enhance the precision of results. Here, the velocity is divided into two components: the
The following
� is extended
� by adding a new power-law parallel component and the perpendicular component,
term “sina7 180α0 α ” as a function of the orientation which can be given by:
angle because the superficial gas velocity is expected to
vary with “α.” Also, an exponential function usg ¼ usg ðcos α þ sin αÞ (33)

Figure 18. Relationship between “d32” and “α”.

Figure 19. Comparison of d32/D calculated by the new correlation Eq. 35 and CFD simulations.
 SEPARATION SCIENCE AND TECHNOLOGY 209

Then, the new Sauter mean diameter correlation is Conclusion

obtained by adding Eq. (33) to Eq. (28) and can be finally
The overall goal here is to rely on the cross-flow as
expressed as follows:
a new concept in the PBRs. We proposed a smaller
!a2 !a3 � � � � pilot-scale reactor that has the same configuration as
d32 usl γusg L a4 180 α a5
�
¼ a1 � � � � used in industrial scale. In addition, to simulate the
D usl þ γusg � �usl þ γusg � G α0
� � �� � � breakup and coalescence behaviors of droplets in
180 α 180 α
exp a6 sina7 contact to the cross-gas flow, a new simulation
α0 α0
method applying Euler–Lagrange method is also
(34)
developed. Based on the results of the above section,
In this case, the new coefficients are estimated using the the following conclusions can be derived:
improved algorithm PSO developed by Ben Guedria.[32]
This has not been found in previous studies. ● When the gas nozzle is tangent to the bed, the
The effect of gas inlet orientations on the Sauter mean produced droplets are homogeneously distributed
diameter is also depicted in Fig. 18, in which all variables along the axial direction and much larger. When
are held constant except for gas-to-liquid ratio. On the the gas nozzle is in the center of the bed, better axial
one hand, it is found that an increase in “α” results in and radial dispersions are allowed as well as a larger
marginally smaller drop sizes “d32,” which contribute to effective gas–liquid interfacial area.
a higher flooding velocity. On the other hand, the ● From the simulated results, a specific Sauter mean
increased “L/G” ratios will cause a wavier interface due diameter was proposed using the general form of
to the decreased stabilizing effects of the gravity. correlations and validated with taking into account
However, the impact of α becomes less significant the cross-sectional shape of gas inlet. However, the
when the gas entertainment is inclined by more than best fit arrived when testing more than two angle
45° from the horizontal plane. As a result, an exponen­ orientation.
tial is added to the fourth term of Eq. (34), and the best ● The new correlation was made according to the
fits are obtained, as shown in Fig. 19. As shown in Eq. location of Sauter mean diameter peak to under­
(34), the fourth term and fifth term stand the critical angle orientation. By readjusting
� �a5 � � ��
180 α 180 α the value of α and adding an exponential function,
α0 exp a6 α0 produce a positive effect
a better agreement was obtained.
when α < 45° and a negative effect when α > 45°. “d32” ● Despite the time it took to run each model, it is
increases with “L/G” ratio but becomes lower again at clear that Euler–Lagrange model is shown to be
α > 45° due to the fifth term. Consequently, the final a powerful tool to study the physics of sprays and
modified correlation is shown in Eq. (35): to predict breakup and atomization.
!0:0337 ! 0:4327
d32 usl γusg
¼ 0:259443 �� � � �
D usl þ γusg � �usl þ γusg �
Nomenclature
� �0:2448 � � 0:3442 � � �� � �
L 180 α 180 α 0:0848 180 α
exp 0:5884 sin
G α0 α0 α0
(35)
Notation Description (Units)
A Gas–liquid interfacial area ([m−1])
Ap Specific area of the packing materials ([m−1])
a1,a2,a3,a4,a5,a6 and a7 Exponent of the dimensionless number ([−])
C1ε,C2ε, and Cμ k-ε turbulence model constant ([−])
Comparison of the newly developed correlation with Cb,CF,Ck,Cl and Cd TAP model constant ([−])
the existing correlations CD Drag coefficient ([−])
Cdc Droplet phase concentration ([kg])
Figure 19 shows the performance of the new correlation. Cin Inlet droplet number concentration ([kg])
Cout Outlet droplet number concentration ([kg])
The capability of Eq. (35) is compared by testing the four D Packed bed diameter ([m])
angle orientations, and their predicted results are found d0 Initial droplet diameter ([m])
dd Equivalent diameter ([m])
to be reasonably closer to each other. The MAPE asso­ deq Droplet diameter ([m])
ciated with Eq. (35) is less than ±12% at the confidence dNozzle Nozzle diameter ([m])
dp Particle diameter ([m])
level of 98.6%. The benefit of the symmetrical distribu­ d32 Sauter mean diameter ([m])
tion is that the prediction error of the cumulative flow FD Drag force ([N])
rate can be further reduced. In summary, these results Finterfaciales Volumetric forces due to momentum flux ([N])
fn RNG k-ε turbulence model coefficient ([−])
indicate that the newly developed correlation can predict (Continued)
“d32/D” in high accuracy.
210 H. TROUDI ET AL.

(Continued). Disclosure statement

g Gravitational acceleration ([m/s2])
G Gas mass flow ([Kg/s]) No potential conflict of interest was reported by the authors.
H Packed bed height ([m])
n Number of the droplets at the position ([−])
Ntest Test number ([−])
k Turbulent kinetic energy ([m2/s2]) References
L Liquid mass flow ([Kg/s])
P Pressure ([N/m2]) [1] Theologos, K. N.; Markatos, N. C. Advanced Modeling
Re Reynolds (droplet) number ([−]) of Fluid Catalytic Cracking Riser-Type Reactors. AIChE
r Radius of the larger droplet ([m]) J 1993, 39(6), 1007–1017. DOI: 10.1002/aic.690390610.
r32,child Sauter mean radius of the child droplets ([m])
t Time, time step ([s]) [2] Froment, G. F.; Vasalos, I.; Markatoss, N.;
u Velocity ([m/s]) Skandalisg, N. Advanced Modeling of Riser-type
ug Gas velocity ([m/s]) Cracking Reactors Catalytic. Current 1997, 17, 837–844.
ud Droplet velocity ([m/s]) [3] Bolton, S.; Kasturi, A.; Palko, S.; Lai, C.; Love, L.;
usl Superficial liquid velocity ([m/s])
usg Superficial gas velocity ([m/s])
Love, L.; Parks, J.; Tsouris, C. 3D Printed Structures
Vcell Total cell volume ([m3]) for Optimized Carbon Capture Technology in Packed
Vtot Total packing volume ([m3]) Bed Columns. Sep. Sci. Technol 2019, 54(13),
We Weber number ([−]) 2047–2058. DOI: 10.1080/01496395.2019.1622566.
Wecr Critical Weber number ([−]) [4] Sen, N. Acta Astronautica Microgravity Gas–Liquid
xd Displacement of the droplet ([m])
x1,x2,x3,x4 and x5 Multivariable linear regression ([−]) Flow through Packed Beds. Acta Astronautica 2011, 68
y Droplet deformation ([−]) (1–2), 39–45. DOI: 10.1016/j.actaastro.2010.07.014.
Greek symbols [5] Heidari, A.; Hassan, S. Numerical Evaluation of the Gas–
α Gas inlet angle ([°]) Liquid Interfacial Heat Transfer in the Trickle Flow Regime
αo Initial angle orientation ([°])
ε Turbulent dissipation rate ([m/s3])
of Packed Beds at the Micro and Meso-scale. Chem. Eng. Sci
μ Dynamic viscosity ([Pa.s]) 2013, 104, 674–689. DOI: 10.1016/j.ces.2013.09.048.
μd Droplet dynamic viscosity ([Pa.s]) [6] Muir, L. A.; Briens, C. L.; Theologos, K. N.;
μg Gas dynamic viscosity ([Pa.s]) Markatos, N. C. Low Pressure Drop Gas Distributors
μl Liquid dynamic viscosity ([Pa.s]) for Packed Distillation Columns. Can. J. Chem. Eng
μt Turbulent (or eddy) viscosity ([kg/(m.s)])
μT Viscosity rate of the droplet and gas ([−]) 1986, 64(6), 1027–1032. DOI: 10.1002/cjce.5450640623.
ρ Density ([kg/m3]) [7] Guo, M.; Nishida, K.; Ogata, Y.; Wu, C.; Fan, Q.
ρd Droplet density ([kg/m3]) Experimental Study on Fuel Spray Characteristics under
ρg Gas density ([kg/m3]) Atmospheric and Pressurized Cross-flow Conditions,
ρl Droplet density ([kg/m3])
ηn RNG k-ε turbulence model coefficient ([−])
Second Report: Spray Distortion, Spray Area, and Spray
η Collection efficiency ([−]) Volume. Combustion 2017, 206, 401–408.
σk Turbulent Prandtl numbers for k ([−]) [8] Mohammadpour, K.; Alkhalaf, A.; Specht, E. CFD
σε Turbulent Prandtl numbers for ε ([−]) Simulation of Cross-flow Mixing in a Packed Bed
σc Critical surface tension ([N/m]) Using Porous Media Model and Experimental
Subscripts
c Critical Validation. Comput. Part. Mech 2018, 6(2), 157–162.
d Droplet DOI: 10.1007/s40571-018-0203-x.
D Drag [9] Yoo, Y.-L.; Han, D.-H.; Hong, J.-S.; Sung, H.-G.
g Gas phase International Journal of Heat and Mass Transfer.
h Hydraulic
l Liquid
A Large Eddy Simulation of the Breakup and
p Particle Atomization of a Liquid Jet into a Cross Turbulent
s Superficial Flow at Various Spray Conditions. Int. J. Heat Mass
t Turbulent flow Transfer 2017, 112, 97–112. DOI: 10.1016/j.ijheatmas
0 Initial condition stransfer.2017.04.064.
Acronyms
CFD Computational fluid dynamics [10] Troudi, H.; Ghiss, M.; Ellejmi, M.; Tourki, Z. CFD
CPU Central processing unit Simulation of Multicomponent Mixture within
E-L Eulerian–Lagrangian a Packed Deethanizer Column. Heat Mass Transfer.
MAPE Mean absolute percentage error 2019, 55(9), 2605–2622. DOI: 10.1007/s00231-019-
RNG ReNormalization Group
R2 R-square
02586-1.
SE Standard error [11] Theologos, K. N.; Lygeros, A. I.; Markatos, N. C.
TAB Taylor analogy breakup Feedstock Atomization Effects on FCC Riser Reactors
PBR Packed bed reactor Selectivity. Chem. Eng. Sci 1999, 54(22), 5617–5625.
PDF Particle distribution function DOI: 10.1016/S0009-2509(99)00294-8.
PSO Particle swarm optimization
[12] Marek, M. Gas Flow Maldistribution in Random Packed
Beds of Non-spherical Particles – A CFD Study. Chem. Eng.
Acknowledgements Sci 2019, 197, 296–305. DOI: 10.1016/j.ces.2018.12.032.
[13] Kaskes, B.; Vervloet, D.; Kapteijn, F.; Ommen JR, V.
The authors gratefully acknowledge the support from Alpha Numerical Optimization of a Structured Tubular
Engineering International (AEI) and thank Pr. Mouldi Reactor for Fischer–Tropsch Synthesis. Chem. Eng. J
Chrigui for his technical assistance. 2016, 283, 1465–1483. DOI: 10.1016/j.cej.2015.08.078.
 SEPARATION SCIENCE AND TECHNOLOGY 211

[14] Liu, X.-H.; Jiang, Y.; Qu, K.-Y. Analytical Solution of [23] Nijemeisland, M.; Dixon, A. G. CFD Study of Fluid Flow
Combined Heat and Mass Transfer Performance in a and Wall Heat Transfer in a Fixed Bed of Spheres. AIChE
Cross-flow Packed Bed Liquid Desiccant Air J 2004, 50(5), 906–921. DOI: 10.1002/aic.10089.
Dehumidifier. Int. J. Heat Mass Transfer 2008, 51(17– [24] O’Rourke, P. J.; Amsden, A. A. The Tab Method for
18), 4563–4572. DOI: 10.1016/j.ijheatmasstransfer. Numerical Calculation of Spray Droplet Breakup. SAE
2007.11.059. Technical Paper. 1987, 872089. DOI: 10.4271/872089.
[15] Du, W.; Zhang, J.; Lu, P.; Xu, J.; Wei, W.; He, G.; [25] Lee, M. W.; Park, J. J.; Farid, M. M.; Yoon, S. S.
Zhang, L. Advanced Understanding of Local Wetting Comparison and Correction of the Drop Breakup
Behaviour in Gas-liquid-solid Packed Beds Using CFD Models for Stochastic Dilute Spray Flow. Appl. Math.
with a Volume of Fluid (VOF) Method. Chem. Eng. Sci Modell 2012, 36(9), 4512–4520. DOI: 10.1016/j.apm.
2017, 170, 378–392. DOI: 10.1016/j.ces.2017.02.033. 2012.02.015.
[16] Alkhedhair, A.; Gurgenci, H.; Jahn, I.; Guan, Z.; He, S. [26] Stefanitsis, D.; Strotos, G.; Nikolopoulos, N.;
Numerical Simulation of Water Spray for Pre-cooling of Kakaras, E.; Gavaises, M. International Journal of
Inlet Air in Natural Draft Dry Cooling Towers. Appl. Heat and Fluid Flow Improved Droplet Breakup
Therm. Eng 2013, 61(2), 416–424. DOI: 10.1016/j. Models for Spray Applications. Int. J. Heat Fluid
applthermaleng.2013.08.012. Flow 2019, 76, 274–286. DOI: 10.1016/j.ijheatfluid
[17] JM, M.-V.; MA, J.; R, B.; A, M.-F.-F. A CFD flow.2019.02.010.
Comparative Study of Bubble Break-up Models in [27] Pramudita, D.; Tsotsas, E. A Model of Pulse
a Turbulent Multiphase Jet. Heat Mass Transfer. 2007, Combustion Drying and Breakup of Colloidal
43(8), 787–799. DOI: 10.1007/s00231-006-0160-4. Suspension Droplets. Powder Technol 2019, 355,
[18] Wang, G.; Cai, W.; Xie, L.; Zhang, X.; Wang, Y. CFD 755–769. DOI: 10.1016/j.powtec.2019.07.096.
Modeling and Simulation of the Hydrodynamics [28] Suekane, T.; Yokouchi, Y.; Hirai, S. Inertial Flow
Characteristics of Packed Column with Structured Structures in a Simple-Packed Bed of Spheres. AIChE J
Sinusoidal Corrugated Sheets Packings. Chem. Eng. Res. 2003, 49(1), 10–17. DOI: 10.1002/aic.690490103.
Des 2022, 183, 56–66. DOI: 10.1016/j.cherd.2022.04.038. [29] Nemec, D.; Levec, J. Flow through Packed Bed Reactors:
[19] Troudi, H.; Ghiss, M.; Ellejmi, M.; Tourki, Z. 1. Single-phase Flow. Chem. Eng. Sci 2005, 60(24),
Performance Comparison of a Structured Bed Reactor 6947–6957. DOI: 10.1016/j.ces.2005.05.068.
with and without a Chimney Tray on the Gas-flow [30] Jadidi, M.; Sreekumar, V.; Dolatabadi, A. Breakup of
Maldistribution: A Computational Fluid Dynamics Elliptical Liquid Jets in Gaseous Crossflows at Low
Study. Proceedings of the Institution of Mechanical Weber Numbers. J. Visualization 2019, 22(2),
Engineers, Part E: Journal of Process Mechanical 259–271. DOI: 10.1007/s12650-018-0537-8.
Engineering. 2020, 234(1), 83–97. DOI: 10.1177/ [31] Xie, P.; Lu, X.; Ding, H.; Yang, X.; Ingham, D.; Ma, L.
0954408919889417. A Mesoscale 3D CFD Analysis of the Liquid Flow in
[20] Ullah, A.; Amanat, A.; Imran, M.; Shah, S.; Gillani, J.; a Rotating Packed Bed. Chem. Eng. Sci 2019, 199,
Kilic, M. Chemical Engineering and Processing – 528–545. DOI: 10.1016/j.ces.2019.01.038.
Process Intensification Effect of Turbulence Modeling [32] Ben Guedria, N. Improved Accelerated PSO Algorithm
on Hydrodynamics of a Turbulent Contact Absorber. for Mechanical Engineering Optimization Problems.
Chem. Eng. Process. Process Intensif 2020, 156, 108101. Appl. Soft Comput. J 2016, 40, 455–467. DOI: 10.1016/
DOI: 10.1016/j.cep.2020.108101. j.asoc.2015.10.048.
[21] Bu, S. S.; Yang, J.; Zhou, M.; Li, S. Y.; Wang, Q. W.; [33] Berna, C.; Escrivá, A.; Herranz, L. E. Development of
Guo, Z. X. On Contact Point Modifications for Forced New Correlations for Annular Flow. Comput. Methods
Convective Heat Transfer Analysis in a Structured Multiphase Flow VIII 2015, 1, 451–462.
Packed Bed of Spheres. Nucl. Eng. Des 2014, 270, [34] Song, J.; Cary Cain, C.; Guen Lee, J. Liquid Jets in
21–33. DOI: 10.1016/j.nucengdes.2014.01.001. Subsonic Air Crossflow at Elevated Pressure. J. Eng.
[22] Bai, H.; Gillis, P. A.; Witt, P. M.; Witt, P. M. A Coupled Gas Turbines Power 2015, 137(4), 041502–1. DOI: 10.
DEM and CFD Simulation of Flow Field and Pressure 1115/1.4028565.
Drop in Fixed Bed Reactor with Randomly Packed [35] Ingebo, R. D.; Foster, H. H. Drop-size Distribution for
Catalyst Particles. Ind. Eng. Chem. Res 2009, 48(8), Cross Current Breakup of Liquid Jets in Airstreams,
4060–4074. DOI: 10.1021/ie801548h. NACA, TN 4087, 1957.