Hindawi

Mathematical Problems in Engineering

Volume 2020, Article ID 7575392, 10 pages
https://doi.org/10.1155/2020/7575392

Research Article
The Liquid Maldistribution Analysis of the Trickle Bed
Reactor with the CFD Method

Jinjin Liu , Liu Kai, Tong Zhao , and Chuanxin Bai

Mechanical and Precision Instrument Engineering, Xi’an University of Technology, Xi’an 710048, China

Correspondence should be addressed to Tong Zhao; [email protected]

Received 9 April 2020; Revised 14 June 2020; Accepted 8 July 2020; Published 5 August 2020

Guest Editor: Wenqing Wang

Copyright © 2020 Jinjin Liu et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The liquid phase maldistribution factor has been investigated in trickle bed reactor, and the results are compared with the previous
measurement data from literature by using the Electrical Resistance Tomography. The simulation results are in agreement with the
experimental results to some degree. The flow rates and particle sizes have been simulated with the method of multiphase flow.
There are two different particles with average diameters of 3.4 mm and 5.3 mm. The flow rate has been studied ranging from
100 ml/min to 1100 ml/min. It has been found that the changes of the particles and liquid flow rates have a significant impact on
the distribution of the liquid volume fraction. The internal liquid holdup is more serious, and the wall-flow phenomenon is more
obvious in a bigger flow rate. The prediction of the liquid volume fraction distribution is a key research technique. Regression
predictions have also been researched on the section near outlet, which can predict the internal flow state of the trickle bed under
the condition of high temperature and high pressure. The average liquid volume fraction is linear with flow rates. The mal-
distribution factor is the index correlation with the flow rates. The results and main conclusions can be used to predict the
distributions and get the properties in a trickle bed reactor.

1. Introduction random trickle bed reactors. A model based on the three-

phase Eulerian approach is developed and a two-fluid model
The trickle bed reactor has been applied in various fields, is utilized to perform the interphase momentum exchanges.
such as petrochemical, environmental protection, and en- Then, Beni and Khosravi [5] made a research on the pressure
ergy conversion [1]. There are three phases in the trikle bed drop and liquid holdup in high- and low-pressure trickle-
reactor which are contacted with each other, and the gas and bed reactors based on CFD. Rupesh and Jyeshtharaj [6] have
liquid pass together through the solid phase, which stands investigated the effect of the trickle bed reactor on particle
for the catalyst particles at work [2]. Trickle bed reactors play diameter ratio and verified that the predicted values of drag
an important role in most catalytic reactions in various coefficient are consistent with the experimental results in the
industries. Since the efficiency and production in catalytic literature. Mohammad et al. [7] presented a combination of
reaction are always decided by the distribution and velocity CFD and Adaptive Network-based Fuzzy Inference System
of gas and liquid phases, it is useful to get the properties of developed for flow characterization inside a cylindrical
multiphase phases in TBR to evaluate and control the re- bubble column reactor. And the liquid flow pattern and gas
action. However, it is hard to obtain the status of internal dynamics were predicted. Amir and Faical [8] have used a
flow in a TBR, and that is because TBR cannot be observed Euler-Euler porous media CFD model to elucidate the dy-
outside directly. Detection with normal sensors is also not namic properties of gas-liquid flows with moving frame and
easy due to its high temperature and pressure inside [3]. sliding mesh techniques. And from these works, it can be
To solve the problem mentioned above, CFD methods seen that CFD is an effective method and used in TBRs.
are usually chosen by many experts. Bazmi et al. [4] in- In order to get a real distribution in TBRs, some medical
vestigated a numerical model for liquid maldistribution in imaging techniques have been applied to make visualization
2 Mathematical Problems in Engineering

for liquid distribution, such as x-ray computed tomography, average diameter of 5.3 mm have a porosity of 0.431 and a
gamma-ray CT, and magnetic resonance imaging (MRI) deviation of 0.3, and conditions not given are the same as
method. X-ray computed tomography have been used in a [17]. The shape of the particles is spherical with uniform
research on two-phase flow in two different types of square- diameters. Besides, the permeability and the coefficient of
channel monolith structures by Schafer et al. [9]. The inertial principal force need to be chosen in later CFD
gamma-ray tomography was used to study the distribution computations, and they can be calculated as follows:
and flow rate of liquid and gas phases in a trickle bed by 1 150 (1 − ε)2
Christophe et al. [10]. Gladden et al. [11] took MRI to � , (1)
measure the liquid holdup and the wetting efficiency in two α D2 ε3
different reactor environments. Fuhe et al. [12] have mea- 3.5 (1 − ε)
sured the spatial liquid holdup in TBRs also by the gamma- C� , (2)
D ε3
ray tomography, and the result of CFD simulation agreed
with the experimental one. Wang et al. [13] have detected the where D is the diameter of particles here, ε means the
gas-liquid pulsating flows in the trickle bed reactor with the porosity, α means the permeability, and C is the inertia loss
electrical capacitance volume tomography system. These coefficient.
measurement techniques are useful in some situation but A mixture model of multiphase flow is used in CFD. As
always have a low temporal resolution and very high cost. to the mixture model, changing of gas and liquid is the main
These complex setups are hard to monitor or use in daily factors, since the catalytic agent is always in a relatively static
work. Therefore, ERT is getting more attention in TBR due position and its moving can be neglected here. Then, the gas
to its convenience and receptive price. Mcmanus et al. [14] and liquid are set as continuous phases. The mass and
have investigated the distribution of only a single liquid momentum volume mean conservation equations for gas-
phase under different flow rates in TBR by ERT. Then, Eda liquid phases are solved. The application of conservation
et al. [15] have investigated the gas-liquid flow distribution equations for the mass and momentum is done by averaging
of spherical particles in the packed bed by ERT and found the summing of the local instantaneous equilibriums in each
that the liquid flow rate has an influence on the radial phase [18]. In general condition of TBR, the gas phase and
diffusion of liquid phase. liquid phase can be considered as the incompressible fluids.
However, the experimental measurement of the amount Then, the continuity equation can be presented as follows:
for liquid distribution has a time lag. Simultaneously, the z ⇀
environmental and operational errors have a great influence ρ 􏼁 + ∇ · 􏼐ρm ] m 􏼑 � 0. (3)
zt m
on results. The internal working conditions of the trickle bed
�→
reactor are always under high temperatures and pressures, Here, vm is the averaged velocity of mass and expressed
and it is difficult to reveal the distribution of liquid phase in as equation (4). ρm means the mixture density and can be
time. In order to overcome the difficulties mentioned above, expressed as equation (5). αk is the volume fraction of phase k.
a simulation for macroscopic multiphase phase flow of gas- ⇀
liquid in TBR is studied based on CFD, and the results have ⇀ 􏽐nk�1 αk ρk ] k
]m � , (4)
also been compared with the experiment of ERT in [16]. ρm
The contents of this paper are as follows. Firstly, a CFD
n
model is established, and its boundary conditions are set
according to an ERT work. Secondly, the simulation results, ρm � 􏽘 αk ρk . (5)
k�1
with two different particle sizes and three different liquid
flow rates, have been obtained. The maldistributions of the The momentum conservation equation can be derived by
liquid phase are compared with the experiment. Further- summing the momentum equations of all phases. It can be
more, regression models have been taken and the effect on expressed as equation (6).
the distribution of the liquid phase from different flow rates z ⇀ ⇀ ⇀ ⇀ →
and particle diameters is also analyzed. Finally, the model 􏼐ρ ] 􏼑 + ∇ · 􏼐ρm ] m ] m 􏼑 � −∇p + ∇ · 􏼔μm 􏼒∇ ] m + ∇T⇀] 􏼓􏼕 + ρm g
zt m m m
can establish a relationship between the maldistribution of
liquid volume fraction and various flow rates. → n
⎝􏽘 α ρ →
+ F +∇·⎛
→ ⎞ ⎠
k k ] dr,k ] dr,k ,
k�1
2. Method (6)
2.1. The Theory of the Numerical Simulation. The simulation n
method in TBR mainly consists of two steps. Firstly, a μm � 􏽘 α k μk , (7)
distribution with proper porosity should be set in the trickle k�1
bed reactor, and the distribution of porosity in simulation
should be consistent with the measured in experiment. Then, → → →
] dr,k � ] k − ] m . (8)
a mixture model of multiphase flow is built to simulate the →
gas-liquid two-phase flow. Here, n is the number of phases and F is a body force. μm
The particles with an average diameter of 3.4 mm have a is the viscosity of the phases. The relationship between the
porosity of 0.408 and a deviation of 0.3, while those with an drag force and mass transfer can be treated as closed terms to
Mathematical Problems in Engineering 3

calculate the flow state of each phase in the multiphase flow work. The bottom is treated as the exit, which is free flow
and written as follows [19]: model. The gas is fed into the porous medium to the packing
2 2/3 uniformly. Some boundary conditions of simulation not
E1 μ g ε g 􏼐 1 − ε g 􏼑 εs mentioned here are set according to the experiments in [16].
Fgl � 􏼢 􏼣
ε2g d2p 1 − εg Viscous resistance and inertial resistance are defined
(9) according to the result of equations (1) and (2). Face per-
1/3 meability reflects the state of being permeable in the section,
E2 ρg εg 􏼐1 − εg 􏼑􏼐ug − ul 􏼑 ε
+ 􏼢 s 􏼣 , while porous permeability reflects the ratio of permeable
ε g dp 1 − εg
volume to whole immersed volume in porous medium.
2
Multiphase flows in trickle-beds are often characterized
2/3 1/3
E1 μg εs 􏼐1 − εg 􏼑 εs E2 ρg εs 􏼐1 − εg 􏼑 εs by low interaction regimes so that the flow is often assumed
Fgs � 􏼢 􏼣 + 􏼢 􏼣 , laminar either at the reactor level or at the catalyst particle
ε2g d2p 1 − εg εg dp 1 − εg
scale [20]. Therefore, laminar flow model is chosen to study
(10) the multiphase of spaces between particles in some works
[21]. Considering that the situation in inlets and outlets is
⎝ E1 μl ε2s E2 ρl μg εs ⎞
⎠. more complex, turbulent model is also used in this work, as a
Fls � ε1 ⎛ + (11)
ε2l d2p εl dp supplement, besides the laminar flow mode.

where E1 and E2 are Ergun constants [19]. E1 is set to 1.5 and

E2 is 1.75 in a normal simulation by experience. ρl and ρg are 2.2.3. The Experiment Setup of Electrical Resistance
the liquid density and the gas density, respectively. ul and ug Tomography. The Electrical Resistance Tomography is one
are the velocity of liquid and gas, respectively. dp is the of the electrical tomographic techniques; the cross-sectional
diameter of the porous medium region. εl and εg are the electrical conductivity distribution can be obtained with
liquid phase holdup and gas phase holdup, respectively. εs ERT. The ERT setup consists of four sensors, a data ac-
means the volume fraction of solid. quisition system, and related image reconstruction software.
In ERT, an array of electrodes is arranged on the border of
2.2. Model and Condition Settings the trickle reactor to obtain a sensitive field of the object field
in different directions. The electrodes are measured to obtain
2.2.1. Simulation Model. The CFD model has the same size projection data in the different scanning directions. Then,
as the experimental apparatus as shown in Figure 1(a). Four the conductivity distribution in the porous media field is
sections are chosen for analysis. The symbols of P1, P2, P3, estimated from the vector of projection data using an image
and P4 present the different sections with a distance of reconstruction algorithm to realize the visualization of the
60 mm, 180 mm, 300 mm, and 420 mm, respectively, from liquid distribution in the object field. The adjacent mea-
the bottom of TBR. The inlet is on the top, and there are 9 surement strategy is used here, which is to inject current
round holes with a diameter of 3 mm. The middle round hole between two adjacent electrodes and measure the voltage of
is the inlet of liquid and the other 8 round holes are inlets of all other two adjacent electrode pairs. The pairs of injecting
air around it. The relationship and sizes between inlets are current are moved to next adjacent pair and voltage mea-
described in Figure 1(b). surements are repeated [22].
The mesh node spacing of the area between P1 and P4 is The distributions of inlets, sections, and outlet are nearly
set as 10, and both inlets and outlets parts are set as 3 with a the same in Figure 1. The positions of data obtained are
more precise result. Constrained by computing power, it is shown in Figure 2. The height of the trickle bed is 600 mm.
difficult to set all meshes with a spacing of 3. The total The particle packing part of the trickle reactor has a length of
number of meshes is 280536. Considering the volume of 480 mm and an inner diameter of 100 mm. The experimental
whole model, the mesh size is smaller than a particle with a setup and its structure sizes are recorded in [16]. The
diameter of 3.4 mm or 5.3 mm. A workstation is used to spherical activated alumina particles are used for catalyst
simulate the model: the CPU is Inter Xeon E3-1231 v3 and support in the reactor, and the average diameters of packing
computing memory is 16 G. The operating system is Win- particles are 3.4 mm and 5.3 mm, respectively. Water was
dows 10. The hexahedral mesh is used in the entire model. used as liquid. In order to provide enough conductivity for
The convergence of calculation is determined by the residual detection, NaCl of 0.5% was mixed in water. The conduc-
of the mass continuity equation steadied less than 10−6. tivity value is 10 mS/cm. The data acquisition system has an
inject current of 50 mA and an excitation frequency of
2.2.2. Simulation Setup. General information on the sim- 9600 Hz. The cross-sectional images were acquired every 10
ulation setups is presented in Table 1. The simulation is seconds at each of the four sensors.
based on the unsteady state solver of pressure-continuous
with FLUENT. The discretization scheme of second order is 3. Results
also used on the momentum and phase fractions. The liquid
phase is injected in the fluid zone from the inlet. The particle In this paper, the liquid phase volume fraction in the TRB is
filling area is treated as the porous medium which is laminar treated as the research parameter. Four sections are selected
flow model, and the rest is set to a turbulent model in this and represented by the symbol P1, P2, P3, and P4,
4 Mathematical Problems in Engineering

Inlet 9 × ϕ3 mm
Air inlet
ϕ50 mm

60 mm
P1
100mm Liquid inlet

Porous 120mm
media P2 ϕ100mm

120mm
P3

120mm
3 mm
P4 50mm –
Y
60mm

Outlet X

(a) (b)

Figure 1: The 3D simulation model. (a) Meshed model. (b) Distribution of inlets (mm).

Table 1: The parameter values of the multiphase flow and porous area.
Gas phase density (kg/m3) 1.225
Gas phase viscosity (kg/m−s) 1.7894e−5
Reference temperature (k) 298.15
Turbulent kinetic energy 10
Turbulent dissipation rate 0.4
Liquid phase density (kg/m3) 998.2
Simulation setup
Liquid phase viscosity (kg/m−s) 0.001003
Face permeability (m2) 1.4925e−8
Porous permeability 0.5
Viscous resistance 6.7e+7
Inertial resistance 2.17e+3
Porous medium thickness (m) 0.05
Cylindrical diameter (mm) 100
Packing height (mm) 480
Experimental setup Single point source inlet diameter (mm) 3
Porosity 0.408/0.431
Liquid flow rate (ml/min) 100–1100

respectively. The position of four sections is 60 mm, 180 mm, Here, n is the number of selected points. 108 points were
300 mm, and 420 mm. In the experiment of ERT, 316 points obtained from each section every 10 seconds, and the total
are selected due to the test accuracy in experimental ap- time is 4 minutes. c represents the average value of the
paratus. In the CFD work, 108 points in a section are selected points. The average value is given by
extracted. The numbers are different but all points distribute 􏽶������
􏽴
evenly. The maldistribution factor is introduced to evaluate 1 n 2
quantitatively the uniformity distribution of liquid phase c� 􏽘c , (13)
n i i
volume fraction. It is often used in similar problems. The
maldistribution factor is the ratio of variance to mean, and
the calculation formula is shown as follows: where ci is the single value from the sample date. The value of
Mf will become zero when the values of the liquid phase
􏽱��������������
2 volume fraction are the same in sections. The larger value of
1/n 􏽐ni�1 c − ci 􏼁 (12)
Mf � . Mf represents the uneven distribution of the liquid volume
c fraction.
Mathematical Problems in Engineering 5

0.8

Inlet 0.7

Maldistribution factor (–)

0.6
0mm
60 mm
0.5
P1
0.4

0.3
180 mm

P2 0.2

0.1

0
300mm

60 180 300 420

P3
Postion (mm)
CFD 300 ml/min
Experiment 300 ml/min (ref. [16])
CFD 700 ml/min
420 mm

Experiment 700 ml/min (ref. [16])

P4
CFD 1100 ml/min
Experiment 1100 ml/min (ref. [16])
300 ml/min 700 ml/min 1100 ml/min Figure 3: The maldistribution factor with 5.3 mm particles.
Outlet 1.8e – 04 4.4e – 04 7.0e – 04

0.5
Color bar
0.45
Figure 2: The image of liquid volume fraction in simulation with
5.3 mm particles. 0.4
Maldistribution factor (–)

0.35
3.1. The Analysis of Reconstruction Image. The visualization 0.3
of those four sections in CFD models is taken after calcu-
lation and the result is shown in Figure 2. The reconstructed 0.25
images of conductive are recorded in [16]. In this part, the 0.2
particles with a diameter of 5.3 mm are studied under three
0.15
flow rates including 300 ml/min, 700 ml/min, and 1100 ml/
min. In order to avoid an over-discipline in calculation, the 0.1
board condition in outlet is set to be free. 0.05
Figure 2 reflects the value of liquid phase volume fraction 60 180 300 420
and most of the volume fraction is taken by particle phase. Postion (mm)
Then, there is a certain height between the inlets and sec- CFD 300 ml/min
tions; the gas phase takes a large part of volume fraction Experiment 300 ml/min (ref. [16])
since it is easy to be trapped in the particle gap. The flow of CFD 700 ml/min
Experiment 700 ml/min (ref. [16])
liquid phase has also been resisted in a large degree due to CFD 1100 ml/min
the complexity of porous media penetration and the in- Experiment 1100 ml/min (ref. [16])
teraction between gas and liquid. Therefore, the volume
fraction of liquid phase in the whole cross section is rela- Figure 4: The maldistribution factor with 3.4 mm particles.
tively small. Considering the reasons mentioned above and
the obtained distribution of ERT in the following part, it is Meanwhile, wall-flow can be found in both CFD and
still reasonable and reflects the real properties of liquid phase ERT, and it will be more obvious with a higher flow rate.
to some degree. Comparing sections P1, P2, and P3, it can be seen that wall-
Some common trends are found in both results of CFD flow is more obvious near inlet, and a central region with a
and ERT. Firstly, with the increase of flow rate, the volume higher volume fraction, which reflects the position of inlet
fraction grows in all sections obviously. The trends of av- for liquid, can be found in them and it. In section P2, a
erage values and maldistributions are similar to a certain deviation occurred in both kinds of images and the central
extent. Maldistributions are recorded in Figures 3 and 4 and region turns to wall. Then, the central area gradually spreads
it can be found that trends are similar. Although the values and a wall-flow happens in P3.
are different, they are still in an order of magnitude. A more The wall-flow appears in the same section of CFD and
specific discussion is in the next part. ERT but the positions are quite different, and this is because
6 Mathematical Problems in Engineering

it is hard to simulate the actual form of particles and the The average values are also compared between CFD and
surface tension of liquids in CFD. Actually, resistance wires ERT. Since the principles in two methods are quite different,
have to be used in ERT; thus the distribution will certainly be the units, volume fraction in CFD and conductivity in ERT,
changed to some degree. Therefore, there is surely a dif- are far away from each other, and thus they cannot be
ference from actual work. In addition, there is a more ob- compared directly. In order to resolve this problem, average
vious deviation of P4 in the ERT and it tends to be more values are normalized and then can be studied reasonably.
uniform later. This deviation is caused by the differences The original data in CFD is compared with the result in
between real condition and simulation, and it only can be [16]. Calculate mean values from both CFD and ERT based
reduced but hard to eliminate completely. on the min-max normalization. The min-max normalization
is commonly used to get dimensionless parameters. It
changes data into a range from 0 to 1. The calculation can be
3.2. The Distribution under Different Diameters. The distri- described as equation (14). xi means a value to be calculated
bution of particle has an influence on the volume fraction of and x′i means the normalized data. x′max and x′min were set as
liquids and thus two different particles, 3.4 mm and 5.3 mm, 1 and 0 before calculation.
are both studied in three different flow rates. Four sections x − xmin
are taken, and the maldistribution factors are shown in x′i � x′max − x′min􏼁 × i + xmin . (14)
xmax − xmin
Figure 3. The results of the liquid phase distribution at the
3.4 mm particle diameter are presented in Figure 4. Relative error between ERT and CFD is used to evaluate
From the figures, it can be seen that all the values of the uniformity. Since a true value cannot be obtained here,
5.3 mm, in both CFD and ERT, are higher than 3.4 mm. It the denominator of relative error is replaced by a larger value
can be concluded that liquid phase flow is prevented in a between xERT and xCFD as equation (15). Therefore, the
greater extent with a smaller diameter; still the trends of two relative error is still in a range from 0% to 100% and negative
samples are very similar to each other. values will not appear. In addition, a value of 0 was added to
In order to compare the results from different fields, a the data before normalization to keep the minimum as 0
dimensionless unit, maldistribution factor, has to be used. here.
􏼌􏼌 􏼌
The degree of dispersion for distribution is described by the 􏼌􏼌 xCFD − xERT 􏼌􏼌􏼌
maldistribution factor. The experiment data in Figures 3 and error � 􏼌􏼌􏼌 􏼌􏼌 × 100%. (15)
􏼌Max􏼈xCFD , xERT 􏼉􏼌􏼌
4 comes from [16]. From Figure 3, it can be seen that
maldistribution factor declines at the beginning of P1 and The result is shown in Table 2. From the result, it can be
then reaches a minimum near section P2. However, it in- seen that the predictions of trends in CFD have a small error,
creases after minimum value until it leaves the TBR. In and thus there is an agreement with experiments under three
Figure 4, the minimum value is near section P3 due to the different flow rates. Most errors are under 30% and all are
difference of diameters. The process means that the liquid is under 50%. The average of error is 25.21%.
relatively uneven near inlet and then is blocked by the And then, the correlation coefficient is used to evaluate
particles. As the process continues, bigger pressure comes the agreement of CFD and ERT as a supplement to the
out between P2 and P3 and therefore causes the liquid to relative error; it can be seen that the correlation coefficient is
flow more slowly and uniformly. From this part, a clearer 0.7686 to 3.4 mm and 0.7647 to 5.3 mm. As to the correlation
process can be observed and the effect of maldistribution is coefficient from equation (16), a value over 0.8 means a
also proved. In the process, values of 5.3 mm particles reduce significant correlation and the result is close to 0.8; thus there
more quickly than the 3.4 mm particles; meanwhile, the is certainly a similar trend between the average values from
steep rises happen with larger particles, and this phenom- CFD and ERT. The correlation coefficient will keep the same
enon reflects that a wall-flow is more obvious with a larger from both normalized and original data; thus the result has a
diameter. high reliability:

n
􏽐 xi−CFD − xCFD 􏼁 xi−ERT − xERT 􏼁
i
Correlation coefficient � 􏽱������������������ 􏽱������������������ (16)
2 2
􏽐i�1 xi−CFD − xCFD 􏼁 􏽐ni�1 xi−ERT − xERT 􏼁 .
n

3.3. The Distribution under Different Flow Rates. The flow factor. The uniformity is reflected mainly by the maldis-
rate has a deep influence on the distribution of liquid in TBR; tribution factor while the trend of volume fraction is shown
thus it is important to make a comparative analysis. In this more clearly in mean values. In fact, these two parameters
work, the flow rate changes from 100 ml/min to 1000 ml/min have always been used together. P4 section is chosen to study
and 11 different values are included in CFD. To observe the and the relationship between flow rates and mean values is
properties more comprehensively, average value and stan- shown in Figure 5. The relationship between flow rates and
dard deviation are added to analyze the maldistribution standard deviations is shown in Figure 6.
Mathematical Problems in Engineering 7

Table 2: Comparison of normalized average values.

3.4 mm 5.3 mm
Section (flow rate)
CFD (–) ERT (–) Error (%) CFD (–) ERT (–) Error (%)
P1 (300 ml/min) 0.25 0.25 0.67 0.24 0.22 11.88
P2 (300 ml/min) 0.26 0.38 30.29 0.24 0.33 29.30
P3 (300 ml/min) 0.24 0.44 45.18 0.24 0.35 31.66
P4 (300 ml/min) 0.23 0.45 48.68 0.23 0.42 44.75
P1 (700 ml/min) 0.58 0.44 24.68 0.65 0.39 40.53
P2 (700 ml/min) 0.59 0.65 9.02 0.63 0.60 5.26
P3 (700 ml/min) 0.56 0.73 22.84 0.60 0.64 5.60
P4 (700 ml/min) 0.54 0.80 32.63 0.49 0.77 35.90
P1 (1100 ml/min) 0.94 0.57 39.57 0.95 0.52 44.76
P2 (1100 ml/min) 1 0.83 17.37 0.99 0.82 17.46
P3 (1100 ml/min) 0.95 0.91 4.45 1 0.86 14.38
P4 (1100 ml/min) 0.86 1 13.99 0.90 1 9.58
Correlation coefficient 0.7686 0.7647

0.0005 1000 ml/min while it decreases obviously with lower flow

rates.
0.0004 From Figure 7, it can be observed that the liquid phase
R2 = 0.9936
will penetrate catalyst particles well with a higher flow rate.
Average value

0.0003 An ideal state for TBR is that there is a higher volume

R2 = 0.9978 fraction but a lower maldistribution factor; therefore, a
0.0002 higher flow rate is needed. However, the liquid holdup will
also increase if the flow rate reaches a certain value, and then
0.0001 flooding begins seriously [23]. According to the results of P4,
the maldistribution factor keeps stable in the area from
0 500 ml/min to 1000 ml/min; meanwhile serious wall-flow
Flow rate (ml/min) phenomenon occurs when the flow rate reaches 1100 ml/
3.4mm
min according to Figure 3 and this should be avoided. From
5.3mm
the result, it can be found that the flow rate of 800 ml/min is a
Figure 5: The average value with different flow rates. good choice in actual work for this trickle bed reactor.

0.00009
3.4. The Correlations Formula under Different Flow Rates.
0.00008
In order to predict the liquid volume fraction, regression
0.00007 R² = 0.9673 predictions have been researched. The correlations formula
Standard deviation

0.00006 of the relationship between the mean value and the flow rate
0.00005 with 5.3 mm particle diameter is expressed as equation (17)
0.00004 R² = 0.9724 and with 3.4 mm particle diameter is expressed as equation
0.00003 (18). R2 means the coefficient of determination, and it is a
0.00002 statistical measurement to know the approximate degree
0.00001 between the prediction and real data. R2 can be calculated by
0 equation (19).
100 200 300 400 500 600 700 800 900 1000
Flow rate (ml/min) X5.3 � 5e−7 v − 4e−5 , (17)
5.3mm
3.4mm X3.4 � 5e−7 v − 3e−5 , (18)
Figure 6: The standard deviation with different flow rates.
􏽢 i 􏼑2
􏽐ni�1 􏼐Xi − X
2
R �1− 2.
(19)
It can be concluded that the liquid volume fraction 􏽐ni�1 Xi − X􏼁
increases with flow rates. There are linear relationships
between the mean values and standard deviations, and an Here, Xi is the observed data and X means its mean
exponential relationship is in the maldistribution factors for value. X􏽢 i is the predicted value calculated from the coeffi-
both kinds of particles. The inlet flow rates have a significant cient of determination. If R2 is 1, it indicates that there is no
effect on the distribution of liquid phase volume fraction, error in prediction [24]. R2 is 0.9936 with 5.3 mm and 0.9978
which are higher than the particle diameter. Maldistribution with 3.4 mm in Figure 5, respectively. The correlations
factor keeps nearly the same in the range of 700 ml/min to formula of standard deviation with 5.3 mm is expressed as
8 Mathematical Problems in Engineering

0.6 Plot of Mf
1.4
0.5
Maldistribution factor

1.2
0.4

0.3 1
0.2
0.8
0.1

Mf
0.6
0
100 200 300 400 500 600 700 800 900 1000
Flow rate (ml/min) 0.4

5.3mm
0.2
3.4mm
Figure 7: The maldistribution factor with different flow rates. 0
100 200 300 400 500 600 700 800 900 1000
v
equation (20) and its R2 is 0.9724. The correlations formula Mf 1 Mf 3
of standard deviation with 3.4 mm particles is expressed as Mf 2 Mf 4
equation (21) and its R2 is 0.9673. The mean value and
standard deviation have a good linear correlation, and the Figure 8: The relationship between flow rates and Mf .
errors are within 1%.
σ 5.3 � 7e−8 v + 5e−6 , (20) Table 3: The intuitive analysis and weight of Mf .
Flow rates (ml/min) Mf (5.3 mm) Mf (3.4 mm)
σ 3.4 � 6e−8 v + 8e−6 . (21) 100 0.508 0.412
200 0.320 0.338
The correlations formulas of the maldistribution factor 300 0.221 0.209
are expressed as equations (22) and (23). Here, R2 is 0.8539 400 0.220 0.212
in 5.3 particle diameter, and the other one is 0.9486. The two 500 0.182 0.1967
correlation formulas have a good index correlation, and the 600 0.198 0.190
errors are within 15%. 700 0.151 0.163
800 0.133 0.173
Mf5.3 � 5.95v− 0.552 , (22) 900 0.146 0.201
1000 0.149 0.172
Mf3.4 � 2.25v− 0.382 . (23) S 0.495 0.505

In order to prove the reasonability for the correlation

prediction of exponential formula between Mf and the flow Mf3.4
S3.4mm � , (24)
rates, here Mf1 is the value of equation (20) divided by Mf3.4 + Mf5.3
equation (17), defined by equation (12). Similarly, Mf3 is the
value of equation (21) divided by equation (18). Mf2 and Mf4 Mf5.3
represent equations (22) and (23), respectively. It can be seen S5.3mm � . (25)
Mf3.4 + Mf5.3
from Figure 8 that the four functions have the same
movement trend, and therefore the prediction formulas of S can be used as the coefficient to describe the weighting
Mf on P4 section are reasonable. Besides, different particle of two diameters and obtained by equations (22) and (23)
diameters have little effect on the movement trend of Mf on under different flow rates. These two factors will compose
P4 section, while the values may be different. the prediction model as coefficients. Results are shown in
Then, the intuitive analysis method is also introduced to Table 3. Finally, the maldistribution of P4 section with
study the weighting of influence on the flow rate from di- known inlet flow rate F can be predicted by equation (26)
ameter more accurately. Take S as the extreme deviation that with S.
reflects the influence of this factor on the mean square error
of flow rate, and it is the maldistribution of average diameter
F � Mf5.3 × S5.3mm + Mf3.4 × S3.4mm � 2.95]− 0.552 + 1.14]− 0.382 .
and can be noted as S3.4mm and S5.3mm in this work. The
calculation can be expressed as follows: (26)
Mathematical Problems in Engineering 9

TRB always works under high temperature and pressure (4) The correlation formulas under different flow rates
and it is almost impossible to directly observe the flow state were analyzed. In order to predict the liquid volume
of internal liquid phase from the outside. Thus, the given fraction, regression predictions have been
prediction model has a significance for the design and researched. The correlation formulas of the rela-
optimization in TBR. In addition, results in this part may be tionship between the mean value and the flow rate
effective ranging within 300–1100 ml/min in experiment and were calculated, and the results were evaluated. The
100–1100 ml/min in CFD. mean value and standard deviation have a good
linear correlation, and the errors are within 1%.
Then, the intuitive analysis method is also intro-
4. Conclusion duced to study the weighting of influence on the flow
rate from diameter more accurately. The weighting
In order to find the distribution and property of liquid phase of two diameters was described under different flow
in TBR, a series of researches are taken based on both CFD rates. These two factors will compose the prediction
and ERT. The distribution of liquid phase was studied with model as coefficients, therefore, a more specific
particles of different average diameters and a series of flow formula to predict flow rate with different
rates. The maldistribution factor is used to evaluate the maldistributions.
impact factor in trickle bed reactor. Then, the volume
The obtained regression model can be used for the design
fractions of liquid phase have been predicted with CFD. The
of the trickle bed reactor and the evaluation for properties of
main conclusions are as follows:
the internal liquid phase. In addition, this work established a
(1) The distribution of multiple phases was simulated for reasonable model to get the properties of TBR based on CFD
a TBR with a mixed model based on CFD. A laminar method, and then most steps and conclusions can be used to
flow part is used to describe the particle filling area save cost in experiment and design of TBR. Since there are
and a turbulent model for the rest. A series of pa- many different conditions from ideal state, this paper mainly
rameters in CFD is calculated and chosen. Boundary focuses on the verification of CFD model, and most ex-
conditions are set based on the ERT experiment in periments are finished before.
[16]. The distribution of flow volume fractions was
obtained effectively. In order to compare the results Data Availability
from different fields, a dimensionless unit, maldis-
tribution factor, was used. The data used to support the findings of this study are true
(2) The visualization for CFD models was also made to and reliable and are available from the corresponding author
describe the distribution of liquid phase, and images upon request.
in sections were obtained. The degree of dispersion
for distribution is described by the maldistribution Conflicts of Interest
factor in different sections and it is found that the
liquid is relatively uneven near the inlet and then is The authors declare that they have no conflicts of interest.
blocked by the particles. A clearer process can be
observed directly and the effect of maldistribution is Acknowledgments
also proved. The maldistribution of 5.3 mm is higher
This work was supported by the National Natural Science
than 3.4 mm, and trends of maldistributions pre-
Foundation of China (Grant 51876175) and the Technology
dicted in CFD were in an agreement with experi-
Innovative Leading Program of Shaanxi (Grant 2019
mental results. Besides, wall-flow also can be found
CGXNG-030). The authors wish to thank the reviewers for
and it will be more obvious with a higher flow rate.
their advice and opinions as they were very important and
(3) Average values were normalized with a min-max useful.
normalization and then compared between CFD and
ERT. From the result, it can be seen that the pre- References
dictions of trends in CFD have an average error of
25.21%, and thus there is an agreement with ex- [1] A. Arnab, R. Shantanu, and L. Faı̈çal, “Cyclic operation of
periments under three different flow rates. From the trickle bed reactors: a review,” Chemical Engineering Science,
correlation coefficient, there is certainly a similar vol. 115, pp. 205–214, 2014.
trend between the average values from CFD and [2] D. I. A. Dhanraj and V. V. Buwa, “Effect of capillary pressure
ERT. The distribution under different flow rates was force on local liquid distribution in a trickle bed,” Chemical
studied. The liquid volume fraction increases with Engineering Science, vol. 191, pp. 115–133, 2018.
[3] L. Katja, M. Mikko, and A. Ville, “CFD modeling of radial
flow rates. There are linear relationships between the
spreading of flow in trickle-bed reactors due to mechanical
mean values and standard deviations, and an ex- and capillary dispersion,” Chemical Engineering Science,
ponential relationship is in the maldistribution vol. 64, pp. 207–218, 2009.
factors. Maldistribution factor keeps nearly the same [4] M. Bazmi, S. H. Hashemabadi, and M. Bayat, “CFD simu-
in the range of 700 ml/min to 1000 ml/min while it lation and experimental study of liquid flow mal-distribution
decreases obviously with lower flow rates. through the randomly trickle bed reactors,” International
10 Mathematical Problems in Engineering

Communications in Heat and Mass Transfer, vol. 39, no. 5, [20] J. G. L. Rodrigo and M. Q. Rosa, “CFD modelling of mul-
pp. 736–743, 2012. tiphase flow distribution in trickle beds,” Chemical Engi-
[5] A. H. Beni and M. R. Khosravi-Nikou, “Modeling hydro- neering Journal, vol. 147, pp. 342–355, 2009.
dynamics of trickle-bed reactors at high and low pressure [21] J. G. L. Rodrigo and M. Q. Rosa, “Turbulence modelling of
using CFD method,” Petroleum Science and Technology, multiphase flow in high-pressure trickle-bed reactors,”
vol. 33, no. 20, pp. 1770–1779, 2015. Chemical Engineering Science, vol. 64, pp. 1806–1819, 2009.
[6] K. R. Rupesh and B. J. Jyeshtharaj, “CFD modeling of pressure [22] Z. Ren, A. Kowalski, and T. L. Rodgers, “Measuring inline
drop and drag coefficient in fixed beds: wall effects,” Par- velocity profile of shampoo by electrical resistance tomog-
ticuology, vol. 8, no. 1, pp. 37–43, 2010. raphy (ERT),” Flow Measurement and Instrumentation,
[7] P. Mohammad, Z. Mohammadjavad, and G. Poobalan, vol. 58, pp. 31–37, 2017.
“Prediction of multiphase flow pattern inside a 3D bubble [23] E. Hansuld, L. Briens, and C. Briens, “Acoustic detection of
column reactor using a combination of CFD and ANFIS,” flooding in absorption columns and trickle beds,” Chemical
RSC Advances, vol. 104, p. 1039, 2015. Engineering and Processing: Process Intensification, vol. 47,
[8] M. D. Amir and L. Faical, “CFD study and experi-mental no. 5, pp. 871–878, 2008.
validation of multiphase packed bed hydrodynamics in the [24] Z. M. Yaseen, S. O. Sulaiman, R. C. Deo, and K.-W. Chau, “An
context of rolling sea conditions,” AIChE Journal, vol. 65, enhanced extreme learning machine model for river flow
pp. 385–397, 2019. forecasting: state-of-the-art, practical applications in water
[9] T. Schäfer, C. Meitzner, R. Lange, and U. Hampel, “A study of resource engineering area and future research direction,”
two-phase flow in monoliths using ultrafast single-slice X-ray Journal of Hydrology, vol. 569, pp. 387–408, 2019.
computed tomography,” International Journal of Multiphase
Flow, vol. 86, pp. 56–66, 2016.
[10] B. Christophe, A. Koudil, P. Chen, and M. P. Dudu-kovic,
“Study of liquid spreading from a point source in a trickle bed
via gamma-ray tomography and CFD simulation,” Chemical
Engineering Science, vol. 60, pp. 6279–6288, 2005.
[11] L. F. Gladden, M. H. M. Lim, M. D. Mantle, A. J. Sederman,
and E. H. Stitt, “MRI visualisation of two-phase flow in
structured supports and trickle-bed reactors,” Catalysis To-
day, vol. 80, no. 30, pp. 203–210, 2003.
[12] Y. Fuhe, A. Artin, and K. Nandakumar, “Liquid holdup
distribution in packed columns: gamma ray tomography and
CFD simulation,” Chemical Engineering and Processing,
vol. 41, pp. 473–483, 2002.
[13] A. Wang, Q. Marashdeh, B. J. Motil, and L.-S. Fan, “Electrical
capacitance volume tomography for imaging of pulsating
flows in a trickle bed,” Chemical Engineering Science, vol. 119,
pp. 77–87, 2014.
[14] R. L. Mcmanus, G. A. Funk, M. P. Harold, and K. M. Ng,
“Experimental study of reaction in trickle-bed reactors with
liquid maldistribution,” Industrial & Engineering Chemistry
Research, vol. 32, no. 3, pp. 570–574, 1993.
[15] T. Eda, A. Sapkota, and M. Takei, “Investigation of liquid
dispersion in the packed-bed by electrical resistance to-
mography,” in Proceedings of the International Conference on
Nuclear Engineering & the Asme Power Conference, pp. 447–
450, Anaheim, CA, USA, July 2012.
[16] T. Eda, A. Sapkota, J. Haruta, M. Nishio, and M. Takei,
“Experimental study on liquid spread and maldistribution in
the trickle bed reactor using electrical resistance tomography,”
Journal of Power and Energy Systems, vol. 7, no. 2, pp. 94–105,
2013.
[17] T. Zhao, T. Eda, S. Achyut, J. Haruta, M. Nishio, and M. Takei,
“Investigation of pulsing flow regime transition and pulse
characteristics in trickle-bed reactor by electrical resistance
tomography,” Chemical Engineering Science, vol. 130, pp. 8–
17, 2015.
[18] A. Arnab, R. Shantanu, and K. D. P. Nigam, “Prediction of
pressure drop and liquid holdup in trickle bed reactor using
relative permeability concept in CFD,” Chemical Engineering
Science, vol. 62, pp. 5870–5879, 2007.
[19] A. Heidari and S. H. Hashemabadi, “CFD study of diesel oil
hydrotreating process in the non-isothermal trickle bed re-
actor,” Chemical Engineering Research and Design, vol. 94,
pp. 549–564, 2015.