Desalination 385 (2016) 148–157

Contents lists available at ScienceDirect

Desalination

journal homepage: www.elsevier.com/locate/desal

Eulerian–Lagrangian modeling and computational fluid dynamics

simulation of wire mesh demisters in MSF plants
Hala Al-Fulaij a,⁎, Andrea Cipollina b, Giorgio Micale b, Hisham Ettouney a, David Bogle c
a
Department of Chemical Engineering, College of Engineering and Petroleum, Kuwait University, P.O.Box 5969, Safat, 13060, Kuwait
b
Dipartimento di Ingegneria Chimica, Gestionale, Informatica, Meccanica, Università di Palermo, Italy
c
University College London, Gower Street, London WC1E 6BT, UK

H I G H L I G H T S

• The manuscript has been reviewed and all the reviewer's comments were taken into consideration.
• There was an error in writing the mathematical model and the error was corrected.
• More details were added to clarify some points as was asked by the reviewers such as the boundary conditions.

a r t i c l e i n f o a b s t r a c t

Article history: This study focuses on computational fluid dynamics model (CFD) for simulation of demisters in multistage flash
Received 13 June 2015 desalination (MSF). The Eulerian–Lagrangian model (steady-state and two-dimensional) was developed to sim-
Received in revised form 7 February 2016 ulate the demister. The model was used to simulate the flow of water vapor and brine droplets in the demister.
Accepted 10 February 2016
The computational domain includes the following three zones: the vapor space above the demister, the vapor
Available online xxxx
space below the demister, and the demister. The demister zone was modeled as tube banks. A sensitivity analysis
Keywords:
of the model revealed that vapor velocity is the main parameter that affects demister performance. Additionally,
Desalination the analysis indicated that vapor temperature had no effect on the pressure drop across the demister. The devel-
Eulerian/Lagrangian modeling oped model was validated using data from both the literature and real MSF plants. Analysis indicated that model
Multistage flashing predictions and experimental data were consistent. This work gives extensive details for CFD modeling of the
CFD MSF demister. The work is based on a previous study by Al-Fulaij et al. (2014). This work is part of a group effort
Demister to develop a comprehensive CFD simulation for the entire flashing stage of the MSF process, which would provide
an extremely efficient and inexpensive design and simulation tool to the desalination community.
© 2016 Elsevier B.V. All rights reserved.

1. Introduction physical properties and demister characteristics, which may include

wire diameter, demister height, and demister area [6]. Additional exam-
Wire mesh demisters are an essential element in all unit processes ples for the empirical models of vapor demisters can be found in the
that involve separation of micron scale liquid droplets from the flow study of El-Dessouky et al. [9] and Al-Dughaither et al. [2]. In both stud-
of a mixture of vapor and liquid droplets. Examples include evaporators, ies, the correlations for the pressure drop show positive dependence on
absorbers, dryers, scrubbers, flash drums, etc. These processes are found the vapor velocity and the packing density and negative dependence on
in most major industries such as petroleum refining, pulp and paper, the wire diameter. Although the empirical models are simple to use,
petrochemicals, gas processing, food, and pharmaceuticals. The Multi their common drawback is the need for experimental measurements
Stage Flash (MSF) desalination process, presently the main source of to determine the values of the empirical constants.
fresh water in the Gulf countries, utilizes wire mesh demisters to re- Several CFD studies for simulating the performance of demisters can
move brine droplets carried over by the flashed off water vapor, as be found in the literature. The study by Rahimi and Abbaspour [20]
shown in Fig. 1 [3]. developed a general CFD model to simulate wire mesh demister in var-
Simple models of wire mesh demisters are usually written in empir- ious processes. Their simulation was carried for a vapor velocity range of
ical form that may relate the vapor release velocity to vapor and liquid 1–7 m/s, packing density of 200 kg/m3, pad thickness 0.2 m and wire
densities or the removal efficiency as a function of the vapor and liquid diameter of 0.31 mm. CFD predictions were found to simulate closely
the experimental data (from a real unit operating within a refinery)
⁎ Corresponding author. and to match the predictions of the empirical model of El-Dessouky
E-mail address: [email protected] (H. Al-Fulaij). et al. [9], especially at low vapor velocities. The study by Galletti et al.

http://dx.doi.org/10.1016/j.desal.2016.02.019
0011-9164/© 2016 Elsevier B.V. All rights reserved.
 H. Al-Fulaij et al. / Desalination 385 (2016) 148–157 149

Fig. 1. Schematic representation of a typical MSF flashing stage.

[11] focused on CFD simulation of wave-plate demister. Their model uti- when fast acceleration phenomena occur. On the other hand, the
lized either the Eulerian or Lagrangian approaches to simulate the vapor Eulerian approach has several advantages in flow cases where high
and droplet phases, respectively. The use of the Lagrangian approach to brine droplet concentrations occur and where high void fraction of the
model the liquid droplets resulted in close matching of the model pre- flow becomes a dominating flow controlling parameter [8]. In other
dictions and laboratory data. words, the Eulerian method can be thought of as providing a picture
Al-Fulaij et al. [4] performed CFD modeling and analysis for the MSF of the spatial distribution of fluid velocity (and of any other continuous
demister using the Eulerian approach to simulate two dimensional phase quantities such as density and pressure) at each instant during
steady state conditions for flow of vapor and brine droplets. The the motion. While in the Lagrangian method, the flow quantities are de-
model results were validated against experimental and plant data. The fined as function of time and of the choice of a material element of par-
analysis shows that the main parameters that affect the demister per- ticle and describe the dynamical history of this selected fluid element
formance are the vapor velocity and the demister permeability. The au- [17].
thors also concluded that the vapor temperature has limited effect on Therefore, the model equations are divided into two parts: 1) contin-
the pressure drop across the demister. uum model for the vapor phase, which is modeled by the Eulerian
Several other CFD studies, having similar features to demister method and 2) discrete model for the brine droplets, which is modeled
analysis though focused on the analysis of different systems, are by the Lagrangian method.
found in simulations of condenser units [12–14,23]. In this case the
condenser tubes have similar features to the demister wires and
droplet condensation and accumulation around the condenser
tubes resemble the capture of liquid droplets carried by the vapor
phase occurring in demister wires. Review of these studies focused
on methods of mesh generation around the condenser tubes, type
of the turbulence model, and effects of non-condensable gases on
phase flow and condensation rates.
This study continues the previous work of Al-Fulaij et al. [4] by
extending the simulation model to employ the Lagrangian approach
in modeling the flow of the brine droplets within the demister. Si-
multaneously, the vapor flow is modeled by the Eulerian approach.
The following sections include discussion of model structure, as-
sumptions and governing equations, domain geometry and grid
analysis. Simulation results referred to the case of unknown separa-
tion efficiency of the brine droplets are presented and thoroughly
discussed. This case provides a useful and inexpensive tool for design
purposes.

2. Mathematical models

Schematic of the tube bank configuration of the demister model is

shown in Fig. 2. The Eulerian approach is used to model the vapor
phase as a continuum and the Lagrangian approach is used to model
the brine droplets as a disperse phase. The Lagrangian approach has
some advantages for predicting the brine droplet flow, especially Fig. 2. Schematic diagram of the tube bank configuration.
150 H. Al-Fulaij et al. / Desalination 385 (2016) 148–157

2.1. Models assumptions represent the viscous terms (shear stress described by Newtonian clo-
sure) and the sixth term represents the pressure gradient due to buoy-
Assumptions used to model the vapor phase by the Eulerian method ancy. The effective viscosity is μev, which is the sum of the turbulent
include the following: viscosity, μtv, and the fluid viscosity, μv.
The standard k-ε model was used for all simulations in this study.
1. Pressure is assumed equal in both phases.
This model uses two differential equations: one for the turbulent kinetic
2. The turbulent diffusivity is equal to the turbulent viscosity,
energy, k, and the other for the turbulent dissipation rate, ε. For the
i.e., Turbulent Schmidt number is equal to one.
vapor phase, the model has the following form [3]:
3. There is no heat transfer between the two phases.
4. In each grid element used to calculate the pressure profile within the
∂ ∂ ∂
computing zone, it is assumed that the pressure within each grid ðρ kv Þ þ ðρ uv kv Þ þ ðρv vv kv Þ
∂t v  ∂x v  ∂y   
element is uniform. μ μ
∂ ∂kv ∂ ∂kv
5. The pressure drop caused by the brine droplets is included during ¼ μ v þ tv þ μ v þ tv þ ðGv  ρv ε v Þ ð4Þ
∂x σ k ∂x ∂y σ k ∂y
their flow within the demister, however, the effect of the drops accu-
mulated on the demister wires is assumed negligible. This is because
∂ ∂ ∂
these drops are assumed to vanish as they hit the demister wires. ðρ εv Þ þ ðρ uv εv Þ þ ðρv vv εv Þ
∂t v  ∂x v   ∂y   
The assumptions used to model the disperse phase (brine droplets) ∂ μ ∂εv ∂ μ ∂εv
¼ μ v þ tv þ μ v þ tv
by the Lagrangian method include the following: ∂x σ k ∂x ∂y σ k ∂y
εv
1. The model assumes that the dispersed second phase occupies a low þ ðC 1 Gv  C 2 ρv εv Þ ð5Þ
kv
volume fraction, even though high mass loading is significant. The
trajectories of the brine droplets are computed individually at speci- where μtv is the turbulent viscosity of the vapor phase and is defined as:
fied intervals during the fluid phase calculation. The discrete phase
formulation used by FLUENT contains the assumption that the sec- 2
C μ ρv kv
ond phase is sufficiently dilute as mentioned previously that drop- μ tv ¼ : ð6Þ
εv
let–droplet interactions are negligible as well as the effects of
droplet volume fraction on the vapor phase. In practice, these issues
The standard turbulence model, without any modifications, was
imply that the discrete phase must be present at a fairly low volume
used with standard wall functions. The five constants of the model
fraction, usually less than 10–12% [10].
were set to the standard k-ε model values: σk = 1, σε = 1.3, C1 =
2. Brine droplets behave as hard spheres.
1.44, C2 = 1.92, and Cμ = 0.09 ([12,20].
3. The brine droplets are assumed to vanish as soon as they touch the
The boundary conditions used to solve the above equations include:
demister wires and do not accumulate on the outer surface.
1) Flow inlet conditions: This is defined on the inlet surface. This condi-
tion is used to define the velocity and scalar properties of the flow at
2.2. The Eulerian model equations for the vapor phase
inlet boundaries. The velocity magnitude and direction or velocity com-
ponents, the turbulence parameters (for turbulent calculations) and the
The general equation for conservation of mass of the vapor phase is
discrete phase boundary conditions (for discrete phase calculations) in-
given by [15]:
formation should be entered for this boundary condition. 2) Pressure
  outlet conditions: It is defined on the outlet surface. 3) Geometric sym-
∂ρv ∂ ∂
þ ðρv uv Þ þ ðρv vv Þ ¼ Sm ð1Þ metry: It is defined on both sides of the geometry. 4) The wall boundary
∂t ∂x ∂y
condition at the wire surface: The demister wires are defined as walls.
where Sm is the mass added to the continuous phase from an eventually Wall boundary condition is used to bound fluid and solid regions. In
dispersed second phase which is the vaporization of the entrained brine viscous flows, the no-slip boundary condition is enforced at walls by de-
droplets. In this study Sm = 0 because it is assumed that the brine drop- fault. The shear stress and heat transfer between the fluid and wall are
lets vanish as they hit the demister wires and no change in phase will computed based on the flow details in the local flow field. When a par-
happen. ticle strikes a boundary face (wire surface), it will escape through the
For the vapor phase, the equations for the conservation of momen- boundary. In this case, the particle is lost from the calculation at the
tum for the two dimensional model (x-direction, horizontal) and (y-di- point where it impacts the boundary. This is similar to the real case
rection, vertical) are given by [12]: where the particle accumulates on the wire surface and then it drops
back to the brine pool. Therefore, the droplet overall balance states
∂ðρv uv Þ ∂ ∂ that the inlet mass of the droplets is equal to the outlet mass of the
þ ðρ uv uv Þ þ ðρv vv uv Þ droplets leaving the top of the demister plus the mass of the droplets
∂t  ∂x v  ∂y     
∂ ∂ uv ∂ ∂ uv ∂ ∂ uv ∂ ∂ vv escaping through the demister wires.
¼ μ ev þ μ ev þ μ ev þ μ ev
∂x  ∂x ∂y ∂y
 ∂x ∂x ∂y ∂x
2 ∂ ∂ uv ∂ vv ∂P 2.3. The Lagrangian model equations for the brine droplets
 μ ev þ  ð2Þ
3 ∂x ∂x ∂y ∂x
Brennen [5] suggested the use of the Lagrangian approach [21] to
∂ðρv vv Þ ∂ ∂ model the flow of the liquid droplets with a particulate volume fraction
þ ðρ uv vv Þ þ ðρv vv vv Þ
∂t  ∂x v ∂y     
less than 10%, which is the case in his study (≪0.1%). In this approach,
∂ ∂vv ∂ ∂vv ∂ ∂uv ∂ ∂vv the flow field is first solved using a continuum approach and subse-
¼ μ ev þ μ ev þ μ ev þ μ ev
∂x  ∂x  ∂y ∂y ∂x ∂y ∂y ∂y quently brine droplets are tracked using a discrete phase model,
2 ∂ ∂uv ∂vv ∂P which is derived from force balances based on Newton's (turbulent
 μ ev þ  ð3Þ
3 ∂y ∂x ∂y ∂y and transitional regimes) and Stokes' (laminar regimes) laws for brine
droplet motion.
On the left side of Eqs. (2) and (3), the first term represents the tran- FLUENT predicts the trajectory of a discrete phase brine droplets
sient unsteady acceleration and the second and third terms represent (droplet) by integrating the force balance on the brine droplets, which
the convective acceleration. On the right side, the first five terms is written in a Lagrangian reference frame. This force balance equates
 H. Al-Fulaij et al. / Desalination 385 (2016) 148–157 151

the brine droplet inertia with the forces acting on the brine droplets, and
can be written (for the y direction in Cartesian coordinates) as [10]:

 
dvl ρ  ρv
¼ F D ðvv  vl Þ þ g y l þ Fy ð7Þ
dt ρl

where Fy is an additional acceleration (force/unit mass) term in the y di-

rection, [FD(vv - vl)] is the drag force per unit mass of brine droplets and
for sub-micron brine droplets, a form of Stokes' drag law is available
[19]. FD is defined as

18 μ v C D Re
FD ¼ ð8Þ Fig. 3. Schematic diagram of the types of meshing in the tube bank.
ρl D2do 24

here, vv is the gas phase velocity, vl is the brine droplet velocity, μv is the
molecular viscosity of the gas, ρv is the gas density, ρl is the density of 4. Domain geometry and grid analysis
the brine droplets, and Ddo is the brine droplet diameter. Re is the rela-
tive Reynolds number of the droplet, which is defined as: The grids were created using Gambit software. Meshing of the tube
bank grid required the use of triangular elements around the tubes
(demister wires) and quadratic elements in the inlet and outlet zones
ρv Ddo jvl  vv j
Re ¼ : ð9Þ and in the space between the tubes. The quadratic element is known
μv
to provide high accuracy and rapid convergence [10]. Fig. 3 shows an ex-
pansion of the meshing around the tubes (demister wires) and the loca-
An additional force arises due to the pressure gradient in the fluid: tions of the triangular and quadratic cells.
The grid analysis was performed to predict the minimum number of
ρv ∂vl elements (or computational cells) that should be used to obtain results
Fy ¼ v : ð10Þ that are independent of the grid size. This was achieved by simulating
ρl l ∂y
identical geometries (demister), boundary conditions and operating
variables. The simulations were performed for the conditions shown
The drag coefficient, CDC, can be taken from [18]: in Table 1, which included a laboratory scale demister and an industrial
scale demister operating at both high and low temperatures. The simu-
a2 a3 lations were based on a comparison of the calculated separation effi-
C DC ¼ a1 þ þ ð11Þ
Re Re2 ciency, pressure drop and velocity (both x and y components). Results
of the grid analysis for the three cases given in Table 1 indicate that so-
where a1, a2, and a3 are constants that apply to smooth spherical brine lution convergence is obtained for a total of 300,000 grid elements. A
droplets over several ranges of Re. sample for solution convergence is shown in Fig. 4. As shown, the opti-
mum number of grid cells for the tube banks approach was 204,481
3. Solution method cells.

The model equations were solved numerically using the commercial 5. Results and discussion
software CFD package FLUENT 6.3. The numerical technique converts
the transport equations of the model into a set of algebraic equations Validation of MSF demister model was made against laboratory and
in each element using a first-order upwind approximation. In each ele- industrial data. This is to demonstrate the generality and reliability of
ment it was assumed that the center value of the field variables, which the model. Subsequently, a new demister was designed with the same
includes velocities, pressure, etc., gives the average value of the variable geometry (spacing between the wires and wire arrangement) but
for this element (Mazzei et al. [16]). Selection of the first-order upwind with different wire diameter. The first demister has 0.24 mm diameter
approximation was motivated by its high accuracy in predicting the wires while the other has 0.20 mm diameter wires. Such choice is not
field variables. The use of higher order approximation (second order) random but depends on the available wire specifications in the industry.
was tested and the test results showed no change in the numerical re-
sults. Examples for the use of this technique can be found in the simula-
tion of flow around condenser tubes by Hu and Zhang [12] and flow Table 1
through demisters in chemical reactors by Rahimi and Abbaspour [20]. Input operating conditions and design parameters used for grid analysis in FLUENT code.
The convergence factor used for all calculated variables was set to
Lab scale Industrial scale, Industrial scale,
1 × 10–7. demister high temperature low temperature
The discrete phase flow, which is modeled through the Eulerian– demister demister
Lagrangian approach, utilized the solution obtained by the Eulerian
Velocity range (m/s) 2.44 1.252 9.718
model for the vapor phase only. The brine droplets, which form the Packing density (kg/m3) 80.317 80.317 80.317
discrete phase, are modeled by the Lagrangian approach. Calculations Water droplet volume 7.34 × 10−5 1.37 × 10−5 3.78 × 10−5
of the droplet trajectory using a Lagrangian formulation include the dis- fraction in the inlet
stream (−)
crete phase inertia, the hydrodynamic drag, and the force of gravity for
Droplet diameter 12b 10a 8a
transient flows. It also predicts the effects of the turbulence on the dis- (micron-m)
persion of brine droplets due to turbulent eddies present in the contin- Porosity (−) 0.9899 0.9899 0.9899
uous phase. It should be stressed that the continuous phase follows a Temperature (K) 373.15 377.548 313.1
steady state model while the discrete phase follows an unsteady state a
[6].
model to track the brine droplet movement with time. b
[1].
152 H. Al-Fulaij et al. / Desalination 385 (2016) 148–157

Fig. 4. CFD predictions for the effect of varying number of elements of tube banks geometry based on lab scale demister on (a) separation efficiency, (b) pressure drop, (c) velocity (y-
component), (d) velocity (x-component).

That will reduce the cost of the demister rather than producing a new It should be noted that the volume fraction of inlet brine droplets is
demister with specific wire diameter. calculated based on six main factors, which are the salinity of the brine
Comparison of the MSF demister model and laboratory data re- entering and leaving the stage, the mass flow rate of the brine entering
ported by El-Dessouky et al. [9] was made at the conditions shown and leaving the stage, the average value of the salinity of the distillate
in Table 2. The laboratory data by El-Dessouky et al. [9] were collected product, the separation efficiency of the demister and the operating
at steady state conditions and it included measurements of the pres- temperature of the stage which effects the densities of both vapor and
sure drop across the demister and removal efficiency of the water brine droplets. The estimate starts with assuming the salinity of the
droplets. product water, which may vary between 2 and 20 ppm [1]; therefore,
 H. Al-Fulaij et al. / Desalination 385 (2016) 148–157 153

Table 2 As shown in Fig. 6, the pressure drop values predicted by the CFD
Input operating conditions and design parameters for the FLUENT code. model are less than the experimental values. That is because in this
Parameter This study El-Dessouky model it was assumed that the droplets escape and vanish as soon as
et al. [9] they reach the demister wires. In reality, the droplets accumulated on
Velocity range (m/s) 1.13–10.4 0.98–7.5 the wires of the demister and then settle down causing more resistance
Packing density (kg/m3) 80.317–176.350 80.317–208.16 to the upcoming flow and resulting with more pressure drop. This
Thickness of demister pad (mm) 100–200 actual resistance cannot be determined or included in the CFD model
Wire diameter (mm) 0.2–0.32
because it is not possible to have upward flow (flashed off vapor and
Droplet diameter (micron-m) 10 1–5
Volume fraction of inlet water droplets (−) 1 × 10–5 NA entrained brine droplets) and downward flow (settled droplets) in the
same zone which is the void space between the wires of the demister
and accounts the interaction between them.
and with knowledge of the total production capacity, the demister sep- CFD FLUENT code was used to simulate a wire mesh demister
aration efficiency, and the average distillate flow rate per stage it is sim- installed in three operating multistage flashing (MSF) desalination
ple to obtain the volume fraction of inlet brine droplets in the flashed off plants. The first plant is a once through MSF plant, the second is a
vapor. MSF-BC plant operating at low temperature and the last is a MSF-BC
As shown in Figs. 5–6, validation of the model showed very good plant operating at high temperature. The demister used in these plant
agreement between the CFD results and the experimental data with had a packing density = 80.317 kg/m3 and 0.28 mm wire diameter.
an error less than 6.3% while for the empirical correlation it reached Table 3 represents the operating conditions for each flashing stage in
16.5%. That indicates how the CFD model is more accurate than the one of the three case study plants, which were used as input values in
model by El-Dessouky. the simulator.

Fig. 5. Variation in pressure drop as a function of vapor velocity for a demister with a packing density = 80.317 kg/m3 and 0.28 mm wire diameter.

Fig. 6. Variation in pressure drop as a function of vapor velocity for demister with a packing density = 176.35 kg/m3 and 0.24 mm wire diameter.
154 H. Al-Fulaij et al. / Desalination 385 (2016) 148–157

Table 3
Operating conditions for the flash stages of a MSF-BC real plant operating at low temperature.

Stage Vapor temperature Vapor density Droplet density Flashed off vapor Inlet volume fraction of
below demister (kg/m3) (kg/m3) velocity droplet in flashed off vapor
(K) (m/s) (−)

1 361.690 0.417 1013.3 1.829 3.181E−06

2 359.256 0.382 1015.1 2.805 2.898E−06
3 356.852 0.350 1016.8 3.006 2.638E−06
4 354.477 0.320 1018.5 3.183 2.402E−06
5 352.131 0.293 1020.1 3.408 2.186E−06
6 349.814 0.268 1021.7 3.649 1.990E−06
7 347.526 0.246 1023.2 4.051 1.813E−06
8 345.267 0.225 1024.7 4.183 1.650E−06
9 343.038 0.206 1026.2 4.498 1.501E−06
10 340.838 0.188 1027.6 4.587 1.365E−06
11 338.667 0.172 1029.0 4.630 1.242E−06
12 336.525 0.157 1030.3 4.724 1.132E−06
13 334.413 0.144 1031.6 4.777 1.030E−06
14 332.330 0.132 1032.8 4.925 9.391E−07
15 330.276 0.120 1034.0 5.015 8.543E−07
16 328.251 0.110 1035.2 5.553 7.778E−07
17 326.255 0.101 1036.4 5.923 7.086E−07
18 324.288 0.092 1037.5 6.318 6.458E−07
19 322.351 0.085 1038.5 6.637 5.896E−07
20 320.443 0.077 1039.6 7.178 5.378E−07
21 318.564 0.071 1040.5 7.628 4.910E−07
22 316.714 0.065 1041.5 8.305 4.484E−07
23 314.894 0.060 1042.4 8.612 4.107E−07
24 313.103 0.055 1043.3 9.718 3.756E−07

As shown in Fig. 7, the validation showed very good agreement be- the demister lower surface and the surface of the brine. As a result,
tween the CFD results and the MSF-BC low temperature plant data the entrainment rate of brine droplets significantly increased compared
with an error less than 6%, except for the last stage where it reached to the previous flashing stages, where the brine height is always less
14.62%. On the other hand the empirical correlation results were far than 0.6 m. Therefore, more brine droplets accumulate on the demister
away from the real data with an error exceeding 52%. This model is wires causing reduction in the flow area, which results in higher
also validated against other two plants to prove its general validity, velocity and larger pressure drop. In contrast, the mathematical model
the first is the MSF-OT and the other is the MSF-BC operating at high assumes zero accumulation or instant disappearance of the brine drop-
temperature. As shown in Figs. 8–9, also in this case the validation lets as it hits the demister wires.
showed good agreement between the CFD results and real plant data As shown in Figs. 10–11, as the wire diameter decreases the pressure
with a difference less than 8.7% while the empirical correlation differ- drop decreases. The separation efficiency remained almost unaffected
ences exceed 98%. for the 0.24 mm diameter demister but it decreased for the 0.20 mm
It should be noted that there is a large increase in the pressure drop diameter allowing the droplets to escape through the demister. Particle
occurred in the last flashing stage in the MSF. The main cause for the in- distribution within each demister type is shown Fig. 12. The pressure
crease in the pressure drop is the increase in the specific volume due to drop kinks at stage 15 and that is due to many reasons such as the in-
the decrease in the stage temperature in addition to the effect of the crease in vapor velocity, decrease in vapor density and increase in drop-
level controller of the last stage that maintains the brine height at a let density. Among these factors, the main cause for the increase in the
fixed value of 0.7 m. This decreased the free board distance between pressure drop is the increase in the specific volume due to the decrease

Fig. 7. Variation in pressure drop as a function of MSF stage number for demister with packing density = 80.317 kg/m3 in MSF-BC plant operating at low temperature.
 H. Al-Fulaij et al. / Desalination 385 (2016) 148–157 155

Fig. 8. Variation in pressure drop as a function of MSF stage number for demister with packing density = 80.317 kg/m3 in MSF-OT plant.

Fig. 9. Variation in pressure drop as a function of MSF stage number for demister with packing density = 80.317 kg/m3 in MSF-BC plant operating at high temperature.

Fig. 10. CFD predictions of the variation in pressure drop as a function of the demister wire diameter for stages 1, 5, 10, 15, 20, 24 in the MSF-BC plant.
156 H. Al-Fulaij et al. / Desalination 385 (2016) 148–157

Fig. 11. CFD predictions of the variation in the demister separation efficiency as a function of the demister wire diameter for stages 1, 5, 10, 15, 20, 24 in the MSF-BC plant.

in the stage temperature. Therefore, a new driving force is initiated due with increase in the void volume within the demister causing easier
to an increase in the rate of droplet re-entrainment at higher velocities. escape routes for the brine droplets. As for the 0.28 mm wire diameter,
Thus the inlet droplet concentration in the inlet vapor to the demister the decrease in the void space within the demister results in increase in
increases and that is directly proportional to the pressure drop in the the pressure drop. Thus, 0.24 mm wire diameter demister is more
demister. preferable.
From the obtained results, 0.24 mm wire diameter demister will be
used to improve the efficiency of the MSF plant because it reduces the 6. Conclusions
pressure drop across the demister with a negligible effect on the separa-
tion efficiency (within the design and acceptable range). It should be Simulation of the MSF demister was carried out adopting an
noted that for the 0.2 mm wires the separation efficiency is much Eulerian–Lagrangian approach, which modeled the two phase flows of
lower than the other two diameters. This is because of the decrease of water droplets (modeled as a discrete phase by the Lagrangian model)
the wire area per unit volume of the demister. Also, this associated and the water vapor (modeled as a continuous phase by the Eulerian
model). The model equations were solved by the commercial CFD soft-
ware package FLUENT. Simulations included both transient and steady
state calculations.
Several numerical tests were made to determine the sensitivity of
the numerical predictions as a function of the grid size. In addition,
the model was validated against real plant data (which covers the
spans of the entire range of operating temperature in the MSF process,
i.e., 105 °C to 40 °C) and experimental data. The results of this test
were necessary to ensure validity and reliability of the model. The CFD
model predictions were found consistent with the experimental as
well as the real plant data. Close examination and comparison of the
empirical model predictions and the CFD model results show that the
CFD simulation results are more accurate than the available empirical
correlation.
The model and results presented in this study add to the knowledge
and understanding to design and operation of demister pads, which are
found in a wide range of industrial application. This model provides an
inexpensive tool to simulate and troubleshoot existing demisters and
to design new demisters. This can help in predicting the effect of
newly designed demisters on the unit operation and the quality of the
vapor product.

References
[1] Al-Deffeeri, N. (2009). Chemical Section Head, Doha West Power Station, Ministry of
Water and Electricity. Kuwait, Personal Communication.
[2] A.A. Al-Dughaither, A.A. Ibrahim, W.A. Al-Masry, Investigating pressure drop across
wire mesh mist eliminators in bubble column, J. Saudi Chem. Soc. 15 (2011) 1–9.
[3] H.F. Al-Fulaij, Dynamic Modeling of Multi Stage Flash (MSF) Desalination PlantPhD
Thesis University College London (UCL), United Kingdom, 2011.
[4] H.F. Al-Fulaij, A. Cipollina, G. Micale, H. Ettouney, D. Bogle, Eulerian–Eulerian model-
ling and computational fluid dynamics simulation of wire mesh demisters in MSF
plants, Eng. Comput. 31 (7) (2014) 1242–1260.
[5] C. Brennen, In Fundamentals of Multiphase Flow (pp. 1–29), Cambridge University
Press, New York, 2005.
Fig. 12. Droplet distribution in the demister with wire diameter: (a) 0.28 mm, [6] E. Brunazzi, A. Paglianti, Design of wire mesh mist eliminators, AICHE J. 44 (1998)
(b) 0.24 mm, (c) 0.20 mm as predicted by CFD. 505–512.
 H. Al-Fulaij et al. / Desalination 385 (2016) 148–157 157

[8] F. Durst, D. Miloievic, B. Schonung, Eulerian and Lagrangian predictions of par- C2 Constant in k-ε model, −
ticulate two-phase flows: a numerical study, Appl. Math. Model. 8 (1984) Cf Interfacial friction coefficient, kg/m3/s
101–115. Cμ Constant in k-ε model, −
[9] H.T. El-Dessouky, I.M. Alatiqi, H.M. Ettouney, N.S. Al-Deffeeri, Performance of wire Ddo Brine droplet diameter, m
mesh mist eliminator, Chem. Eng. Process. 39 (2000) 129–139. f Friction factor, −
[10] Fluent_Inc., Fluent 6.3 Manual. Fluent Inc., Centerra Resource Park, 2006. fd Friction factor, −
[11] C. Galletti, E. Brunazzi, L. Tognotti, A numerical model for vapor flow and droplet g Gravity acceleration, m/s2
motion in wave drainage channels, Chem. Eng. Sci. 63 (2008) 5639–5652. G Generation of turbulent kinetic energy, kg/m/s3
[12] H.G. Hu, C.A. Zhang, A modified k-ε turbulence model for the simulation of two- k Turbulent kinetic energy, m2/s2
phase flow and heat transfer in condensers, Int. J. Heat Mass Transf. 50 (2007) P Pressure, Pa
1641–1648. ΔP Pressure drop between stages, Pa
[13] J. Je, C. Choi, S.J. Kim, M.H. Kim, Numerical Study of Gap Size Ratio Effect for Noncon- ΔPp Demister pressure drop, Pa
densable Gas Ventilation in Condensers, 362012 67–74. Re Reynolds number, −
[14] T. Karlsson, L. Vamling, Flow fields in shell-and-tube condensers: comparison of a t Time, s
pure refrigerant and a binary mixture, Int. J. Refrig. 28 (2005) 706–713. u Velocity component in the x-direction, m/s
[15] E. Kufås, Mathematical Modeling of Coalescence of Oil Droplets in Water V Volume, m3
FlowMaster of Science in Energy and Environment Norwegian University of v Velocity component in the y-direction, m/s
Science and Technology, Department of Energy and Process Engineering,
2008. Greek Letters
[16] L. Mazzei, A. Casillo, P. Lattieri, P. Salatino, CFD simulations of segregating fluidized
bidisperse mixtures of brine droplets differing in size, Chem. Eng. J. 156 (2010) ε Turbulent dissipation rate, m2/s3
432–445. ρ Mass density, kg/m3
[17] G. Micale, Introduction to COMPUTATIONAL Fluid Dynamics (CFD) Physical Models ψ Volume fraction, −
and Numerical Methods with Applications to Engineering Problems, 1993 (Presen- μ Laminar dynamic viscosity, Pa·s
tation at the UCL). μ t Turbulent viscosity, Pa·s
[18] S.A. Morsi, A.J. Alexander, An investigation of particle trajectories in two-phase flow μ ev Effective viscosity, Pa·s
systems, J. Fluid Mech. 55 (1972) 193–208. σk, σε Constants, −
[19] H. Ounis, G. Ahmadi, G.B. McLaughlin, Brownian diffusion of submicrometer parti-
cles in the viscous sublayer, J. Colloid Interface Sci. 143 (1991) 266–277.
[20] R. Rahimi, D. Abbaspour, Determination of pressure drop in wire mesh mist elimina- Subscripts
tor by CFD, Chem. Eng. Process. 47 (2008) 1504–1508.
[21] B.G. Van Wachem, A.E. Almstedt, Methods for multiphase computational fluid dy- v Vapor phase (water vapor)
namics, Chem. Eng. J. 96 (2003) 81–98. l Liquid phase (entrained liquid droplets)
[23] J. Zhao, B. Jin, Z. Zhong, Study of separation efficiency of a demister vane with re- x x-Direction
sponse surface technology, J. Hazard. Mater. 147 (2007) 363–369. y y-Direction

Nomenclature
Ado Total projected area of droplets in a given control volume, m2/m3
C1 Constant in k-ε model, −