You are on page 1of 9

Desalination 369 (2015) 156–164

Contents lists available at ScienceDirect

Desalination

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

Numerical simulation of brackish water desalination by a reverse


osmosis membrane
Ali E. Anqi a,b, Nawaf Alkhamis a,c, Alparslan Oztekin a,⁎
a
Lehigh University, Dept. of Mechanical Engineering & Mechanics, USA
b
King Khalid University, Dept. of Mechanical Engineering, Saudi Arabia
c
King Abdulaziz University, Dept. of Mechanical Engineering, Saudi Arabia

H I G H L I G H T S

• The brackish water desalination by reverse osmosis membranes is investigated.


• Membrane performance is enhanced with addition of spacers in the feed channel.
• Spacers in the feed channel help in mitigating concentration polarization.
• Spacing and arrangement of spacers profoundly influence the permeate flux.
• This work aids in design and optimization of reverse osmosis membrane.

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

Article history: The brackish water desalination by reverse osmosis membranes has been widely used to purify water. This study
Received 4 February 2015 focuses on how the momentum mixing in the feed flow affects the membrane performance. Steady and transient
Received in revised form 5 May 2015 two dimensional flows in a channel containing circular shaped spacers are investigated. Different arrangements
Accepted 7 May 2015
and the spacing of cylinders are considered for the Reynolds number of 400, 800, and 4000. The feed channel is
Available online 16 May 2015
bounded by membranes. The velocity, the pressure, and the concentration fields are obtained by solving the
Keywords:
Navier–Stokes and the mass transport equations. Laminar and turbulent flow models are employed in this
Brackish water desalination study. The Shear Stress Transport (SST) k-ω turbulence model is utilized when the flow is turbulent. The mem-
Reverse osmosis brane is treated as a permeable wall, and it is modeled as a functional surface. The water permeate is calculated
Spacers based on the local osmotic pressure and the concentration at the membrane surface. Momentum mixing promot-
Mass transport ed by spacers enhances the membrane performance significantly.
© 2015 Elsevier B.V. All rights reserved.

1. Introduction dimensional flow and mass transfer in feed channel including spacers.
They assumed that the membrane was an impermeable surface along
The reverse osmosis (RO) desalination is gaining more attention to which the concentration was constant. They showed that the mass
solve water scarcity in the world. The reverse osmosis membrane sys- transfer in the feed channel is strongly influenced by the spacer geom-
tems use less energy than the thermal distillation plants that results in etry. Geraldes et al. [4] studied concentration polarization inside the
the reduction of overall desalination cost. In reverse osmosis desalina- feed channel with ladder-type spacers. They employed momentum
tion, seawater or brackish water flows with high pressure while and mass transfer equations and imposed no-slip, no-penetration con-
contained by the semi-permeable membrane through which pure ditions at the membrane surface. They calculated the salt flux through
water passes and salt is rejected. The rejected salt accumulates and in- the membrane by artificially introducing the permeate velocity. They
creases the salt concentration near the membrane wall which causes a concluded that the spiral wound membrane performs better when the
phenomenon called concentration polarization. The concentration spacer-filament is adjacent to the membrane surface. Both Shakaib
polarization reduces the pure water production by increasing the osmotic et al. [3] and Geraldes et al. [4] considered laminar flows.
pressure near the membrane wall and reduces membrane life time [1,2]. Recently, Srivathsan et al. [5] have investigated the membrane
Several investigators have studied the flow and the mass transfer in performance by conducting three dimensional steady computational
reverse osmosis feed channel. Shakaib et al. [3] considered three fluid dynamics simulations. They utilized periodic condition in the
streamwise direction by including two crossing spacers in the computa-
⁎ Corresponding author. tional cell. They have treated the membrane as a permeable surface
E-mail address: alo2@lehigh.edu (A. Oztekin). and also as an impermeable surface and calculated Sherwood number

http://dx.doi.org/10.1016/j.desal.2015.05.007
0011-9164/© 2015 Elsevier B.V. All rights reserved.
A.E. Anqi et al. / Desalination 369 (2015) 156–164 157

in the channel containing square cross-sectioned spacers with different


Nomenclature
configurations. Laminar flow at Re ≈ 150 is studied by utilizing the pen-
alty method and the finite element method to solve the Navier–Stokes
S distance between spacers [m]
equations. The penalty method decouples the pressure and the velocity
fx force component in x-direction [N]
field; which reduces the number of unknown to be solved. The penalty
Uave average velocity [m/s]
number becomes too large when the Reynolds number exceeds 400;
Δp pressure difference [Pa]
that causes a numerical instability. It has been reported that the penalty
a1 turbulent model parameter
method is not reliable at high Re flows [8,9].
cb feed bulk concentration [kg/m3]
In this paper we present a numerical simulation of high salinity
CL lift coefficient
brackish water desalination in a spiral wound RO membrane. A channel
cw membrane salt concentration [kg/m3]
containing spacers in different configurations and spacings were con-
d spacer diameter [m]
sidered for different values of the Reynolds number in the range of
F1, F2 blending functions
400 to 4000. The channel walls were treated as permeable, and the
h channel height [m]
water flux and concentration at the walls were coupled. A commercial
k turbulent kinetic energy [J/kg]
software utilizing finite volume method was used to perform the
Re Reynolds number, Re = Uave 2h/ν
simulation.
(u,v) components of velocity [m/s]
Sc Schimdt number, Sc = ν/D
1.1. Governing equations
fy force component in y-direction [N]
vw water flux [m/s]
Two dimensional steady and transient flows are considered in the
A water permeability [m/s Pa]
feed channel of the RO membrane system. The flow is incompressible,
c0 inlet salt concentration [kg/m3]
and the physical properties of the fluid are assumed as constant. It is
cp production concentration [kg/m3]
also assumed that the diffusion coefficient is constant. The equations
CD drag coefficient
governing the fluid motions and the concentration field are:
D diffusivity [m2/s]
e distance from spacer to wall [m] the continuity
Sh Sherwood number, Sh = hm h/D
hm mass transfer coefficient [m/s] ∂ui
¼ 0; ð1Þ
p pressure [Pa] ∂xi
t time [s]
St Strouhal number, St = fd/Uave the conservation of momentum

2
Greek letters ∂ui ∂u 1 ∂p ∂ ui
μ dynamic viscosity [Pa s] þ uj i ¼ − þν ; ð2Þ
∂t ∂x j ρ ∂xi ∂x j ∂x j
μt eddy viscosity [Pa s]
ρ density [kg/m3]
and the mass transport equation
ω specific dissipation rate [1/s]
α normalized spacing, α = s/d 2
∂c ∂c ∂ c
β, β* turbulent model parameters þ uj ¼D : ð3Þ
∂t ∂x j ∂x j ∂x j
σω,σ ω2 turbulent model parameters
ν kinematic viscosity [m2/s]
νt kinematic eddy viscosity [m2/s] Here ρ is the density, μ is the viscosity and D is the diffusion coeffi-
κ osmotic coefficient [kPa m3/kg] cient. i and j are the summation indices, the components of the velocity:
τ scaled time τ = 2 t Uave/d u1 = u and u2 = v, the spatial coordinates: x1 = x and x2 = y, t is the
γ turbulent model parameter time and p is the pressure and ν is the kinematic viscosity (ν = μ/ρ).
σk1 turbulent model parameter Several investigators studied the flow inside the channel with
spacers and they found that the transition to turbulence occurs at
Subscripts Re ≥ 400 [11–13]. Recently, the present authors [10,11] investigated
i and j index notation flows between two parallel membranes including spacers for Re in the
w properties at the membrane surface range of 400 to 1000. They compared results obtained by LES simula-
tions against those predicted by k-ω baseline and they concluded that
k-ω baseline can predict the time averaged flow field very accurately.
using an empirical relation. Saeed et al. [6] have performed numerical
k-ω Shear Stress Transport (k-ω SST) turbulence model proposed by
simulations to study the mass transport in spacer filled membrane.
Menter [12] is employed here. This model predicts the turbulent flow
They solved the Navier–Stokes equations and the mass transport equa-
structure near boundaries and the onset of flow separation well. The
tion in a feed channel bounded by an impermeable wall. Quasi three di-
equations for k-ω SST turbulence model yields
mensional simulations are conducted by using periodic boundary
conditions in the stream-wise direction. The mass fraction of sodium !
∂ui ∂u 1 ∂p 1 ∂ ∂u
chloride is assumed to be constant along the membrane surface. Both þ uj i ¼ − þ ðμ þ μ t Þ i : ð4Þ
Srivathsan et al. [5] and Saeed et al. [6] concluded that the spacing be- ∂t ∂x j ρ ∂xi ρ ∂x j ∂x j
tween spacers highly influences the membrane performance and the
pressure drop. Here the eddy viscosity is μ t ¼ ρ maxðaa11ω;Ω
k
F 2 Þ. Equations for the turbu-
Concentration polarization has strong influence on the RO mem-
lent kinetic energy, k, and the specific dissipation rate, ω, are of the form
brane performance, and it is important to characterize it in the flow
and the mass transfer models of membrane systems. Ma and Song [7] " #
∂ðρkÞ ∂ðρkÞ ∂ui  ∂ ∂k
performed a numerical study in which the water flux and the concen- þ ui ¼ τi j −β ρωk þ ðμ þ σ k1 μ t Þ ð5Þ
∂t ∂xi ∂x j ∂x j ∂x j
tration at the membrane surface were coupled. They considered flows
158 A.E. Anqi et al. / Desalination 369 (2015) 156–164

" #
∂ðρωÞ ∂ðρωÞ γ ∂ui ∂ ∂ω The local Sherwood number (Sh) is calculated from
2
þ ui ¼ τi j −βρω þ ðμ þ σ ω μ t Þ
∂t ∂xi νt ∂x j ∂x j ∂x j 
ð6Þ ∂c 
D 
∂y y¼h
1 ∂k ∂ω hm h
þ 2ρð1− F 1 Þσ ω2 : Sh ¼ ; hm ¼ ð8Þ
ω ∂x j ∂x j D ðcb −cw Þ

where hm is the local mass transfer coefficient, cb is the bulk


concentration.
The parameters and constants in Eqs. (4)–(6) are presented in Ref
[12]. Here Ω is the vorticity magnitude, a1, β, β*, σk1, σω, σω2, and γ The lift and drag coefficients of a spacer is determined from C L ¼
are closure coefficients, and F1, F2 are the blending functions. Details of 2 fy
and C D ¼ 2fx
2 2. Here fx and fy are the component of the force in
the closure parameters can be found in Ref [12]. ρ U 2sp d ρ U 2sp d

The schematic of the computational domain is depicted in Fig. 1. The the stream-wise and the span-wise direction, respectively. Usp is the av-
solution of salt water is studied for separation of salt from water using a erage velocity of the flow past a spacer.
RO membrane. Flow inside the channel with turbulators is considered
for various arrangements: the inline geometry (e = 1/4 h), the stag- 1.2. Numerical method and mesh independency
gered geometry (e = 1/8 h), and the staggered geometry with spacers
touching the membrane (e = 0). Here d is the diameter of the spacer The simulations are performed by employing the commercial soft-
and e is the distance between the edge of the spacer and the membrane ware CFX 14.5. The Semi-Implicit Method for Pressure Linked Equations
surface, as shown in Fig. 1. Simulations are conducted with α ¼ ds of 10 (SIMPLE) algorithm – a fully coupled solver – is used to solve the veloc-
and 20 for each spacer arrangement. Here s is the gap between the ity and the pressure field. The flux conditions at the membrane were
spacers. The distance x/h = 10 is assigned between the inlet and the modeled using CEL expression language. The membrane permeability,
first spacer to eliminate the inlet effect. Similarly, the distance x/h = A, is assumed to be constant at 2.3 × 10−11 m/(s Pa). Since the pres-
20 is assigned between the last spacer and the outlet to minimize the sure difference between the inlet and the outlet in the feed channel is
outlet effect. Steady state simulations were conducted for three values small compared to the transverse pressure difference, Δp, across the
of the Reynolds number: 400, 800 and 4000. In addition, a transient sim- membrane, Δp is assumed to be constant at 1.25 MPa. The osmotic coef-
ulation at Re = 800 is conducted using the large eddy simulation (LES) ficient is fixed as κ = 75 kPa m3/kgand the salt diffusivity in water is
to study the effects of transient effects on the membrane performance in D = 1.5 × 10−9 m2/s. These values are taken from Ref [7] and they rep-
the inline geometry. The Reynolds number is defined based on the hy- resent typical properties for the desalination process of brackish water.
draulic diameter of the channel, Re = Uave 2h/ν. The density of solution is fixed at ρ = 1000 kg/m3, and the dynamic vis-
At the inlet, the fully-developed velocity profile is applied, uðyÞ ¼ cosity of the solution is also assumed to be constant at 1 × 10–3 Pas. The
  Schmidt number is Sc = ν/D = 667. Flow in the channel without
6 U ave hy 1− hy , where Uave is the average velocity. The concentration
spacers is modeled as laminar at Re = 400 and 800, and it is modeled
of the salt at the inlet is constant at c0 = 4000 [g/m3]. No-slip and no-
as turbulent at Re = 4000.
penetration conditions are imposed on the velocity field at the surface
A mesh optimization study is conducted using three meshes:
of spacers. The membrane wall is modeled as a functional surface
1.2 × 106, 1.8 × 106, and 2.4 × 106. For the three meshes, the profile of
through which the water flux is determined from local concentration
stream-wise component of the velocity at the wake of the 1st spacer is
and the pressure difference across the membrane [7]
nearly the same, as shown in Fig. 2. Additionally, the drag coefficient is
calculated for the 1st spacer using three meshes. The drag coefficient
is found to be 1.09, 1.10, and 1.08, respectively for the 1.2 × 106,
   9
@y ¼ 0 and h : u ¼ 0; vw ¼ A Δp−κ cw −cp = 1.8 × 106, and 2.4 × 106 meshes. That ensures that the grid independen-
∂c   ð7Þ cy is satisfactory. The 1.8 × 106 mesh is chosen to carry out both the
D ¼ vw cw −cp ;
∂y steady state and the transient simulations.

1.3. Validation
where vw is the local water permeate at the membrane surface, A is the
water permeability through the membrane, cp is the salt concentration In order to validate the k-ω SST turbulence model employed in the
in the production channel, Δp is the transmembrane pressure, κ is the present work, simulations are conducted in an empty channel and in a
osmotic coefficient, and cw is the local concentration at the membrane channel containing square cross sectioned spacers with an inline ar-
surface. The salt rejection at the membrane is assumed to be 100%. rangement. The results of these simulations are compared to those re-
The present authors [10,11] derived the membrane model that can be ported by Ma and Song [7]. The same parameters used in Ref [7] are
applied to gas separation or desalination. They showed that when the selected: Δp = 800 psi, c0 = 32,000 ppm, A = 7.3 × 10−12 m/(s Pa),
limit salt concentration approaches to zero the rejection rate tends to h = 1 mm, Uave = 0.1 m/s, and Re = 149. The spacing between the cyl-
be 100%, as is the case in the present study. inders is 4.5 mm and the channel length is 10 cm. The flow and the

Fig. 1. Schematic of the flow geometry.


A.E. Anqi et al. / Desalination 369 (2015) 156–164 159

work than in Ref [7], we expect to see slightly different concentration


and water flux profiles between the two works. The pressure drop in
the channel without and with spacers is 106 and 1014 Pa, respectively.
The pressure drop predicted here for each case matches the reported
pressure drop of corresponding cases in Ref [7].

2. Results and discussions

The results of steady flow simulations in the inline and the staggered
geometries are presented at Re = 400, 800 and 4000 for α of 10 and 20.
The results of the transient flow simulations in the inline geometry for
α = 20 is presented and compared against that of the steady
simulations.
Fig. 4 illustrates steady state contours of the stream-wise component
of the velocity and the normalized salt concentration at various loca-
tions along the x-axis for Re = 800 and α = 20. The cylinders are placed
in an inline arrangement. The images are acquired between the first and
second cylinder, the seventh and eighth cylinder, and between eleventh
and twelfth cylinder, respectively. The velocity contours show that the
flow has a repeated pattern following each spacer. Flow past the spacer
Fig. 2. The profiles of stream-wise component of the velocity calculated using three differ-
ent meshes: 1.2 × 106, 1.8 × 106, and 2.4 × 106. expands, that leads to a thickening of a boundary layer in the wake few
diameters away from the spacer. Polarization of concentration occurs at
the same location where the momentum boundary layer thickens. The
concentration field are depicted in Fig. 3. Contours of the velocity and intensity of the polarization increases further downstream, as shown
the salt concentration are very similar to those documented in Ref [7]. in Fig. 4.
On the other hand, the wall concentration and the water flux through Fig. 5 depicts the normalized concentration profiles (c/c0) along the
the membrane exhibit slightly different trend. The spacers are expected upper membrane. The black lines denotes the concentration profiles in
to redirect the flow toward the membrane which results in a thinner the empty channel, the blue and the red lines denote the concentration
boundary layer. In the wake of the spacer, the expansion causes a profiles for α = 20, and α = 10, respectively, in various geometries. The
thicker boundary layer and as the boundary layer becomes thicker it same color coding is applied for the water flux and the Sherwood num-
leads to the concentration polarizations, as seen in Fig. 3(c). Since the ber profiles along the membrane surface depicted in Figs. 6 and 7, re-
pressure in the governing equations is treated differently in the present spectively. The concentration polarization is alleviated at all Re with

Fig. 3. a) Concentration contour of the submerged case. b) Velocity contour of the submerged case. c) Normalized concentration profile along the surface of the upper membrane.
d) Permeate velocity along the surface of the upper membrane.
160 A.E. Anqi et al. / Desalination 369 (2015) 156–164

Fig. 4. Steady flow at Re = 800, α = 20, and e = 1/4 h. a1), a2), and a3) velocity contours and b1), b2 and b3) concentration contours at different locations along x-axis.

the presence of spacers in the channel. In the inline geometry, the best and α = 10. The improvement is better with α = 10 than that with
mitigation of the concentration polarization is obtained at Re = 400 α = 20, as illustrated in Fig. 5(b). On the other hand, the staggered
when it is compared against to the open channel, as shown in spacers hardly mitigate the concentration polarization at Re = 400.
Fig. 5(a). It is also noted that at high speed flow, Re = 4000, the concen- Fig. 5(c) and (d) show the concentration profiles in the staggered geom-
tration profiles are not influenced significantly as the spacing of the cyl- etry when spacers touching the membrane surface for α = 20 and 10,
inders is changed. Fig. 5(b) shows the concentration profiles in the respectively. In this geometries, Re = 800 shows the maximum level
staggered geometry with e = 1/8 h. The staggered spacers exhibit sig- polarization mitigation. There is a spike in the concentration distribu-
nificant level of polarization mitigation at Re = 800 for both α = 20 tion at locations where spacers are touching the membrane. These

Fig. 5. Concentration profiles along the surface of the upper membrane a) e = 1/4 h, α = 20 and α = 10 b) e = 1/8 h, α = 20 and α = 10 c) e = 0, α = 20 d) e = 0, α = 10.
A.E. Anqi et al. / Desalination 369 (2015) 156–164 161

Fig. 6. Water flux along the surface of the upper membrane a) e = 1/4 h, α = 20 and α = 10 b) e = 1/8 h, α = 20 and α = 10 c) e = 0, α = 20 d) e = 0, α = 10.

spikes are results of the stagnant flow in the region front of spacers, and Tables 1a and 1b list the pressure drop in the feed channel, the aver-
they might result in the formation of surface fouling. age value of the water flux and the Sherwood number in all geometries
Fig. 6 illustrates the water fluxes along the upper membrane in the at Re of 400, 800 and 4000. In order to eliminate the inlet and the outlet
inline and the staggered geometries for α of 10 and 20 at Re = 400, effects, the average values of vw and Sh are calculated from the local
800 and 4000. The water suction rate for the membrane system includ- values determined in the region 0.01 m ≤ x ≤ 0.2 m. The average values
ing spacers in all geometries at all flow rates increases when it is com- of the Sherwood number and the suction rate increase as the flow rate
pared against the open channel. Fig. 6(a) shows the water fluxes in increases in all geometries. The Sherwood number for α = 10 is higher
the inline geometry for α = 20 and 10. The largest increase in the than that of α = 20 at all Re. As it is expected the pressure drop increases
water flux through the membrane is obtained at Re = 800. The geome- as the flow rate increases. As it is also expected the pressure drop is
try with α = 10 performs slightly better than that of α = 20. higher for α = 10 in all geometries. Table 1 illustrates that the pressure
Fig. 6(b) depicts the water flux in the staggered geometry for e/h = 1/ drop in the feed channel is strongly dependent on the spacers' configu-
8, α of 10 and 20. Similarly, the improvement in the suction rate is the ration. The highest values of the mass transfer coefficient occurs in the
highest at Re = 800. Slight or no discernible enhancement is seen in geometry of e = 0 for α = 10, yet the pressure drop is not the highest
the water flux at Re = 400. The nominal value of the suction rate is in this geometry. It is noted that the highest pressure drop occurs in
very close to that obtained for the open channel at Re = 400, as seen the staggered geometry with e = 1/8 h for α = 10, as shown in Table 1b.
in Fig. 6(b). Fig. 6(c) and (d) illustrates the water flux through the mem- Fig. 8 illustrates images from the transient simulation at t = 0.2078 s
brane at the upper membrane in the staggered geometry when spacers (or at dimensionless time, τ=148.5). Contours of the stream-wise com-
touch the membrane surface. The best improvement is again attained at ponent of the velocity are shown in Fig. 8(a), and contours of the nor-
Re = 800. malized salt concentration are shown in Fig. 8(b). The simulations are
Fig. 7 depicts the Sherwood number, Sh, determined along the sur- conducted in the inline geometry for α = 20 at Re = 800. The velocity
face of the upper membrane. The nominal value of the Sherwood num- and the concentration field obtained for the transient simulations can
ber in the inline and the staggered geometries is higher than that of the directly be compared against the results of steady simulations present-
open channel at all values of Re. As expected the Sherwood number in- ed in Fig. 4 for the same conditions. The numbers (1, 2, and 3) in the fig-
creases as the Re is increased in all geometries including the open chan- ure caption denote the location where images are acquired and they are
nel. The greatest enhancement in the Sherwood number is reached at identical to those shown in Fig. 4. The vortex shedding from spacers en-
the highest flow rate, as shown in Fig. 7. The local Sherwood exhibits hances the momentum mixing in the region near and away from the
strong dependence on the spacers' configuration. spacers. The flow induced by the Karman vortex street increases the
162 A.E. Anqi et al. / Desalination 369 (2015) 156–164

Fig. 7. Local Sherwood number along the surface of the upper membrane a) e = 1/4 h, α = 20 and α = 10 b) e = 1/8 h, α = 20 and α = 10 c) e = 0, α = 20 d) e = 0, α = 10.

Table 1a
The pressure drop, the average water flux and the Sherwood number in inline geometry.

Channel Channel with inline spacers e = 1/4 h

α = 20 α = 10

|Δpx| [Pa] vw  10−5 [m/s] Sh |Δpx| [Pa] vw  10−5 [m/s] Sh |Δpx| [Pa] vw  10−5 [m/s] Sh

Re = 400 106 1.438 26.7 471 1.484 28.6 822 1.547 31.5
Re = 800 204 1.565 31.8 1309 1.732 42.3 2407 1.798 49.6
Re = 4000 1702 1.965 74.9 22,949 2.078 155.1 42,398 2.084 161

concentration mixing as shown in Fig. 8(b1), (b2), and (b3). The results cylinders. Fig. 9a illustrates that the flow induced by the vortex shed-
presented for the steady simulations denote the time average of the ding in the wake region of spacers is strongly influenced by the proxim-
transient velocity and the concentration fields, as shown in Fig. 4. ity of the walls. The similar flow structures are reported by Singha and
Vorticity contours shown in Fig. 9(a1), (a2) and (a3) reveal the in- Sinhamahapatra [13] and Sahin and Owens [14] in the channel contain-
tensity of the secondary flows induced by the vortex shedding. The vor- ing spacers with the same blockage ratio at the similar flow rates. The
tices shed by the spacers dissipate as they are convected away from the lift coefficient calculated for the spacer on the first row proves that the

Table 1b
The pressure drop, the average water flux and the Sherwood number in the staggered geometries.

Staggered spacers e = 1/8 h Staggered spacers e = 0

α = 20 α = 10 α = 20 α = 10
−5 −5 −5
|Δpx [Pa] vw  10 [m/s] Sh |Δpx| [Pa] vw  10 [m/s] Sh |Δpx| [Pa] vw  10 [m/s] Sh |Δpx| [Pa] vw  10−5 [m/s] Sh

Re = 400 456 1.44 27.2 864 1.468 28.5 300 1.632 38.1 467 1.7 49.9
Re = 800 1337 1.821 53.4 2637 1.92 73.5 1011 1.833 63.5 1672 1.845 73.9
Re = 4000 22,461 2.063 139.7 49,446 2.078 155.8 20,335 2.045 154.8 36,763 2.046 167.6
A.E. Anqi et al. / Desalination 369 (2015) 156–164 163

Fig. 8. Transient simulation at Re = 800, α = 20 and e = 1/4 h @t = 0.2078 [sec] (τ = 148.5). a1), a2), and a3) velocity contours, and b1), b2), and b3) concentration contours at different
locations along x-axis.

flow is nearly perfectly periodic at Re = 800. The power spectral density 0.4 at Re = 800. The Sherwood number calculated at three different
of the lift coefficient signals yield the dimensionless frequency, the axial location, P1, P2, and P3, from the transient and steady state results
Strouhal number, St = fd/Uave = 0.252. Here f is the frequency of the are compared, the axial locations selected are x = 15, 18, and 22.5 mm,
fluctuations manifested by the vortex street behind the spacers. Mettu respectively. The solid lines denote the Sh determined from the tran-
et al. [15] found the Strouhal number to be 0.25 for a blockage ratio of sient simulations and the dashed lines denote the Sh determined for

Fig. 9. Transient simulation at Re = 800, α = 20 and e = 1/4 h. a1), a2), and a3) are vorticity contours at different locations along x-axis @t = 0.2078 [sec] (τ = 148.5), b) local Sh number at
three different points, c) lift coefficient of the first spacer vs. nondimensional time, and d) the power spectral density of the lift coefficient signal.
164 A.E. Anqi et al. / Desalination 369 (2015) 156–164

the steady state results. Sh calculated from the steady state simulations drop should be considered as an important part of the of optimization
is not too different than that calculated from the transient simulations, process.
as shown in Fig. 9(c). These results imply that steady state simulations
provide an accurate assessment of the membrane performance in the
geometries considered here. References
Steady and transient two dimensional flows in a channel containing
[1] C. Fritzmann, J. Löwenberg, T. Wintgens, T. Melin, State-of-the-art of reverse osmosis
circular shaped spacers are considered here. For two dimensional flows desalination, Desalination 216 (2007) 1–76.
spacers should be oriented perpendicular to the primary flow. The effect [2] S. Sablani, M. Goosen, R. Al-Belushi, M. Wilf, Concentration polarization in ultrafil-
of the cross-linked mesh structure of spacer filaments and their orienta- tration and reverse osmosis: a critical review, Desalination 141 (2001) 269–289.
[3] M. Shakaib, S. Hasani, M. Mahmood, CFD modeling for flow and mass transfer in
tion with the primary flow directions is critically important on the con- spacer-obstructed membrane feed channels, J. Membr. Sci. 326 (2009) 270–284.
centration polarization and the performance of the reverse osmosis [4] V. Geraldes, V. Semiao, M.N. de Pinho, Concentration polarisation and flow structure
membrane. Such three dimensional effects in a desalination process within nanofiltration spiral-wound modules with ladder-type spacers, Comput.
Struct. 82 (2004) 1561–1568.
are being investigated by the present authors and the results will be [5] G. Srivathsan, E.M. Sparrow, J.M. Gorman, Reverse osmosis issues relating to pres-
published in a full paper later. sure drop, mass transfer, turbulence, and unsteadiness, Desalination 341 (2014)
83–86.
[6] A. Saeed, R. Vuthaluru, H.B. Vuthaluru, Investigations into the effects of mass trans-
3. Conclusion
port and flow dynamics of spacer filled membrane modules using CFD, Chem. Eng.
Res. Des. 93 (2014) 79–99.
Numerical simulations have been conducted to study the perfor- [7] S. Ma, L. Song, Numerical study on permeate flux enhancement by spacers in a
mance of reverse osmosis membrane for a brackish water desalination crossflow reverse osmosis channel, J. Membr. Sci. 284 (2006) 102–109.
[8] B.N. Jiang, A least-squares finite element method for incompressible Navier–Stokes
process. The momentum and mass transport equations govern the problems, Int. J. Numer. Methods Fluids 14 (1992) 843–859.
flow of the brackish water in the feed channel. The mass flux of water [9] J. Heinrich, C. Vionnet, The penalty method for the Navier–Stokes equations, Arch.
through the membrane varies with the local pressure and the concen- Comput. Methods Eng. 2 (1995) 51–65.
[10] N. Alkhamis, A. Anqi, D.E. Oztekin, A. Alsaiari, A. Oztekin, Gas separation using a
tration. The SST k-ω turbulence model has been selected to perform membrane, ASME 2013 International Mechanical Engineering Congress and
the steady turbulent flow simulations while the LES has been selected Exposition, American Society of Mechanical Engineers, 2013 (V07AT08A039-
to perform the transient turbulent flow simulations. Three flow rates, V007AT008A039).
[11] N. Alkhamis, D.E. Oztekin, A.E. Anqi, A. Alsaiari, A. Oztekin, Numerical study of gas
Re = 400, 800, and 4000, are considered in the inline and the staggered separation using a membrane, Int. J. Heat Mass Transf. 80 (2015) 835–843.
geometries. Momentum mixing resulting from turbulators improves [12] F.R. Menter, Two-equation eddy-viscosity turbulence models for engineering appli-
the performance of the membrane greatly at all flow rates considered. cations, AIAA J. 32 (1994) 1598–1605.
[13] S. Singha, K. Sinhamahapatra, Flow past a circular cylinder between parallel walls at
The concentration polarization is mitigated in the inline and the stag-
low Reynolds numbers, Ocean Eng. 37 (2010) 757–769.
gered geometries at all flow rates. The results clearly indicate that the [14] M. Sahin, R.G. Owens, A numerical investigation of wall effects up to high blockage
Sherwood number is strongly influenced by Re and the arrangement ratios on two-dimensional flow past a confined circular cylinder, Phys. Fluids 16
(2004) 1305–1320.
of the spacers. This study proves that, in order to enhance the brackish
[15] S. Mettu, N. Verma, R. Chhabra, Momentum and heat transfer from an asymmetri-
water desalination by the RO membrane, the arrangement and the spac- cally confined circular cylinder in a plane channel, Heat Mass Transf. 42 (2006)
ing of turbulators must be optimized for a given range of Re. Pressure 1037–1048.

You might also like