membranes

Article
Analysis of Concentration Polarisation in Full-Size Spiral
Wound Reverse Osmosis Membranes Using Computational
Fluid Dynamics
Wenshu Wei 1 , Xiang Zou 2 , Xinxiang Ji 2 , Rulin Zhou 1 , Kangkang Zhao 1 and Yuan Wang 2,3, *

1 Beijing Tianma Electro-Hydraulic Control System Company Ltd., Beijing 100013, China;
[email protected] (W.W.); [email protected] (R.Z.); [email protected] (K.Z.)
2 UNSW Centre for Transformational Environmental Technologies, Yixing 214200, China;
[email protected] (X.Z.); [email protected] (X.J.)
3 School of Civil & Environmental Engineering, UNSW Sydney, Sydney 2052, Australia
* Correspondence: [email protected]

Abstract: A three-dimensional model for the simulation of concentration polarisation in a full-scale

spiral wound reverse osmosis (RO) membrane element was developed. The model considered the
coupled effect of complex spacer geometry, pressure drop and membrane filtration. The simulated
results showed that, at a salt concentration of 10,000 mg/L and feed pressure of 10.91 bar, permeate
flux decreased from 27.6 L/(m2 h) (LMH) at the module inlet to 24.1 LMH at the module outlet as
a result of salt accumulation in the absence of a feed spacer. In contrast, the presence of the spacer
increased pressure loss along the membranes, and its presence created vortices and enhanced fluid



velocity at the boundary layer and led to a minor decrease in flux to 26.5 LMH at the outlet. This


paper underpins the importance of the feed spacer’s role in mitigating concentration polarisation in
Citation: Wei, W.; Zou, X.; Ji, X.; full-scale spiral wound modules. The model can be used by both the industry and by academia for
Zhou, R.; Zhao, K.; Wang, Y. Analysis improved understanding and accurate presentation of mass transfer phenomena of full-scale RO mod-
of Concentration Polarisation in
ules by different commercial manufacturers that cannot be achieved by experimental characterization
Full-Size Spiral Wound Reverse
of the mass transfer coefficient or by CFD modelling of simplified 2D flow channels.
Osmosis Membranes Using
Computational Fluid Dynamics.
Keywords: spiral wound module; reverse osmosis; RO; feed spacer; concentration polarisation;
Membranes 2021, 11, 353. https://
computational fluid dynamics; CFD; desalination
doi.org/10.3390/membranes11050353

Academic Editor: Biplob Pramanik

Received: 13 April 2021 1. Introduction

Accepted: 5 May 2021 Reverse osmosis (RO) has been widely used in desalination, water purification, re-
Published: 10 May 2021 claimed water recovery and industrial wastewater treatment while concentration polarisa-
tion (CP) remains a challenge that causes increases in feed pressure, energy consumption,
Publisher’s Note: MDPI stays neutral permeate salt concentration and membrane fouling [1,2]. The analysis of concentration
with regard to jurisdictional claims in polarisation is often based on the use of the Boundary Layer Film Model [3] which relies
published maps and institutional affil- on experimental or mathematical determination of the mass transfer coefficient of feed
iations.
solute and the boundary layer thickness, which is difficult to assess for commercial spiral
wound modules with spacers. In addition, CP factors calculated from the Boundary Layer
Film Model cannot be used to represent the variation in CP and flux change along the axial
direction of the membranes. The effect of pressure drops, which affects the feed pressure
Copyright: © 2021 by the authors. along different locations in the axial direction and subsequently affects the mass transfer of
Licensee MDPI, Basel, Switzerland. ions in the boundary layer, cannot be fully captured by the Boundary Layer Film Model.
This article is an open access article Computational fluid dynamics (CFD) is a powerful tool for the investigation of flow
distributed under the terms and behaviour inside water channels and has been widely used to study the effect of spacers
conditions of the Creative Commons on mass transfer and flow behaviour [4–6], particle colloid contamination distribution and
Attribution (CC BY) license (https://
bacterial film distribution inside the feed channel of RO membranes [7,8]. Schwinge and
creativecommons.org/licenses/by/
Wiley et al. [9–11] conducted 2D simulations of three typical spacer configurations and
4.0/).

Membranes 2021, 11, 353. https://doi.org/10.3390/membranes11050353 https://www.mdpi.com/journal/membranes

Membranes 2021, 11, 353 2 of 13

discovered that the wall shear stress increased by either decreasing the spacer spacing or
increasing the Reynolds number. At the same Reynolds number, turbulent wakes appeared
more readily by decreasing the critical Reynolds number (ReCR ), which were beneficial for
reducing contaminant accumulation at the membrane surface and for mitigating the CP
phenomenon at a cost of increasing the pressure drop across the membrane module, which
further caused an increase in energy consumption. Ahmad et al. [12] performed CFD simu-
lations of circular, triangular and rectangular spacer filaments at the same cross-membrane
pressure condition and found that in comparison to rectangular filament, circular and
triangular filaments could produce vortices at lower Reynolds numbers. The CP factor of
circular filaments was smaller than that of the rectangular filament at the same Reynolds
number. Cao et al. [13] used CFD to simulate the effect of the spacing of spacers on fluid
flow inside the RO membrane flow channel. The simulation showed that, when the spacer
spacing decreased, the distance between the shear stress peak values at the RO membrane
surface decreases, i.e., the peaks appeared more frequently, which improved the mass
transfer capacity of the membrane. However, this simultaneously increased the pressure
drop, thus increasing the energy consumption. On the other hand, when the spacer spacing
increased, the pressure drops decreased while the distance between the shear stress peak
values at the membrane surface became larger, i.e., the peaks appeared less frequently, thus
aggravating the CP phenomenon. Therefore, the study suggested that the mass transfer
capacity and energy consumption should both be considered in the selection of optimal
spacer spacing. Steady-state laminar and turbulent flow models were used by Ranade
and Kumar [14] to simulate the fluid flow with Reynolds numbers between 50 and 1500.
The k-ε model was used as the turbulent flow model, and the effect of several optimized
spacer designs on fluid flow inside the flow channel was studied in detail; this includes the
cylindrical and curvilinear spacer designs. Koutsou et al. [15,16] used a non-steady-state
laminar flow model to perform CFD simulations of fluid flow and mass transfer with
Reynolds numbers between 70 and 740. The study focused on investigating the effect of
different internal angles and the angle of the filaments on fluid flow in the membrane flow
channel across several spacer designs. Ruiz-García and Pestana developed [17] a 2D model
to simulate the CP of three different Dupont Filmtec RO membrane elements at different
salt concentrations, feed flowrates and feed pressure. This model, however, ignored the
spatial distribution of salts on the membrane surface of the spiral wound module.
Three-dimensional CFD models have been developed to investigate the effect of
geometrical characteristics of spacers on the performance of RO membranes [18]. Abdel-
baky et al. [19] performed CFD analysis on RO membrane modules with feed spacers
having fixed or variable diameter under different inlet salinity and Reynolds number
conditions. The results indicated that the RO membrane modules with variable-diameter
feed spacers can reduce CP. Toh et al. [20] modelled the installed perforations with various
locations, diameters and number of perforations on traditional feed spacers and its impact
on the hydraulic and mass transport. Advanced characterisation techniques such as Particle
Image Velocimetry (PIV) [21] and in situ microscopic observations have also been used to
examine the flow behaviour in 3D feed channels incurred by spacers [22].
As the full-size spiral wound RO membrane module has a much more complex
geometry and is several orders of magnitude larger than a crossflow flat sheet flow channel
in size, most simulations have primarily focused on a single feed spacer or a few feed
spacers [23] in order to reduce computational workload. This results in being unable to
reveal the hydraulic performances of the full-scale RO membrane module. Moreover, most
of the studies neglected permeate flow. Although the impact of permeate on CP can be
considered minimal, these models cannot be used to predict the full profile and spatial
distribution of flux along the axial direction of the spiral wound membrane modules.
This paper established a 3D CFD model to simulate a full-scale commercial spiral
wound RO (SWRO) membrane module (Dupont Filmtec-BW30-400). Previous work per-
formed by Gu et al. developed a one-dimensional model capturing the spiral characteristics
of SWRO membranes by coupling values of geometric parameters such as flow path length,
Membranes 2021, 11, 353 3 of 13

variation of the flow channel height and cross-sectional area [24]. The current work further
developed the spiral equation used in the work by Gu et al. by adding the topologies of
the feed spacer to establish a complete 3D geometric model of the spiral wound membrane
module. The effects of the feed spacer on flow distribution, mass transport and CP were
analysed utilizing the comparison of RO modules with a feed spacer and RO modules
without a feed spacer. The spatial distribution of flux along the spiral wound membranes
in the absence and presence of the spacer was calculated.

2. Model Development and Validation

2.1. Governing Equations
CFD simulations of RO processes involve modelling of the mass transfer of the solute
(inorganic salts) and solvent (water) at the feed channel and permeate channels, respec-
tively [25,26]. The governing equations for fluid flow are the conservation of mass and
conservation of momentum equations for incompressible fluids (Navier–Stokes equations):

∇·(ρu) = Sv (1)

∇·(ρuu) = −∇ p + ∇·(µ∇u) (2)

∇·(ρum) = ∇·(ρD ∇m) + Sm (3)
where u denotes the fluid velocity vector, p denotes the pressure, ρ denotes the fluid density,
µ denotes the fluid viscosity and m denotes the solute mass fraction. At the feed side of a
RO membrane, solvent (water) permeation through RO membranes produces a mass sink
term Sm (4) and a momentum source term Sv (5):

J ·a
Sm = − (4)
V
J · a·v
Sv = − (5)
V
where a is the effective membrane area for solution passage (m2 ); V denotes the corre-
sponding effective volume in the computational domain (m3 ); v denotes the flow veloc-
ity perpendicular to the membrane surface (m/s); and J represents the permeate flux
of the corresponding solution (m3 /(m2 ·s)), represented by (6) according to the Kedem–
Katchalsky Method:
J = A × (∆P − ∆Π) (6)
where A is the permeability coefficient of water through the RO membrane (m/(s·Pa)), ∆P
denotes the pressure difference between the two sides of the membrane (Pa) and ∆Π is the
osmotic pressure difference caused by the salt concentration difference between both sides
of the membrane.
The transport of inorganic salts within the entire flow field is expressed through the
mass conservation Equation (7):

∇·(uc) = ∇·( D ∇c) + Ss (7)

where c denotes the concentration of the corresponding inorganic salt, u is the velocity of
water flow and D is the diffusion coefficient of the corresponding organic salt. Ss is the
solvent mass source sink term and is calculated as follows:
At the feed channel:
Ss = 0, (8)
On the membrane feed surface:
Js · a
Ss = − (9)
V
 𝑆 0, (8)

On the membrane feed surface:

𝐽 ∙𝑎
𝑆 (9)
𝑉
Membranes 2021, 11, 353 4 of 13
where 𝐽 corresponds to the permeate flux of inorganic salt 𝑗 (kg/(m2∙s)), which is repre‐
sented by (10):
𝐽 𝐽 𝐶 𝐵 𝐶 𝐶 (10)
where Js corresponds to the permeate flux of inorganic salt j (kg/(m2 ·s)), which is repre-
sented 𝐵
where byis(10):
the permeability coefficient of the corresponding

inorganic salt through the
Js = J × Cp = B × Cm − Cp (10)
RO membrane (m/s), and 𝐶 and 𝐶 represent the concentration of the corresponding
inorganic
where B issalt
theon the feed side
permeability and on of
coefficient the permeate
the side ofinorganic
corresponding the membrane, respectively
salt through the RO
(kg/m 3).
membrane (m/s), and Cm and Cp represent the concentration of the corresponding inor-
ganic salt on the feed side and on the permeate side of the membrane, respectively (kg/m3 ).
2.2.
2.2. Geometry
Geometry
2.2.1.
2.2.1. Cross‐flow
Cross-FlowFlatFlatSheet
SheetModule
Module
A
A3D 3DCFD
CFDmodel
model was
wasdeveloped
developed to simulate the the
to simulate flowflow
and and
salt concentration in a 16.86
salt concentration in a
mm 6 mm××60.86
16.86× mm mmmm × 0.86
(L ×Wmm ×H) ×W
(L RO × domain
cell (Figure
H) RO cell 1). The
domain feed 1).
(Figure spacer
Theconsisted
feed spacerof
fine filaments
consisted with
of fine varyingwith
filaments thicknesses from 0.22 mm
varying thicknesses fromto0.22
0.45mmmmto(Figure
0.45 mm2a(Figure
and Figure
2a,b).
2b).
GridGrid inflations
inflations at 0.02
at 0.02 mmmm thicknesswere
thickness wereapplied
appliedforforthe
the liquid–membrane interfaces,
interfaces,
which ensures
which ensures CP
CP and
and transport
transport phenomena
phenomena that
that occurred
occurred were
were simulated
simulated using
using aa higher
higher
mesh resolution.
mesh resolution. A Agrid
gridsize
sizeof
of 0.05
0.05 mm
mm was
was employed
employed universally
universally across
across the
the remainder
remainder
of the
of the domain
domain andand resulted
resulted inin approximately
approximately 7,293,005
7,293,005 elements
elements (determined
(determined by by mesh
mesh
independence test
independence test from
from 3,000,000
3,000,000 to
to 15,000,000
15,000,000 elements).
elements).

Membranes 2021, 11, x 5 of 15

Figure 1. Computational domain of the feed channel.

Figure 1. Computational domain of the feed channel.

(a) (b)

Figure 2. Physical(a) (b)

geometries and 3D CAD model of the spacer filament. (a) Physical geometries of
the spacer filament; (b) Three-dimensional CAD model of the spacer filament.
Figure 2. Physical geometries and 3D CAD model of the spacer filament. (a) Physical geometries of
the spacer filament; (b) Three-dimensional CAD model of the spacer filament.
2.2.2. Geometry of a Full-Size RO Membrane Module
A full-size
2.2.2. Geometry spiral wound RO
of a Full-Size ROmembrane
Membranemodule
Module (Dupont Filmtec-BW30-400) con-
sisting of 10 membrane sheets that were 2.05 m long and 0.9 m wide (with an effective
A full-size spiral wound RO membrane module (Dupont Filmtec-BW30-400) consist-
membrane area of 400 ft2 ) and wounded along an Archimedes spiral [27] was modelled in
ing of 10 membrane sheets that were 2.05 m long and 0.9 m wide (with an effective mem-
this work. Adjacent membrane sheets shared a common feed channel and ran parallel with
brane area of 400 ft2) and wounded along an Archimedes spiral [27] was modelled in this
one another with some independency; hence, the module’s geometries could be simplified
towork. Adjacent
possess membrane
only a single sheets shared
feed channel a common(Figure
for the simulation feed channel
3). Theand ran parallel
geometry with
of a single
one another with some independency; hence, the module’s geometries could
feed channel was created using ANSYS SpaceClaim (2019R3) (Ansys, Inc., Canonsburg,be simplified
to possess only a single feed channel for the simulation (Figure 3). The geometry of a single
feed channel was created using ANSYS SpaceClaim (2019R3) (Ansys, Inc, USA). Looking
down from the axial direction, the Archimedean spiral for the single feed channel rolled
up along the length of the membranes had the following equations:
𝑥(𝜃) = (𝛼 + 𝛽𝜃)cos (𝜃) (11)
𝑦(𝜃) = (𝛼 + 𝛽𝜃)sin (𝜃) (12)
where 𝑥 and 𝑦 are the coordinates of points on the spiral, 𝜃 is the angle variable and α
 the spacer filament; (b) Three‐dimensional CAD model of the spacer filament.

2.2.2. Geometry of a Full‐Size RO Membrane Module

A full‐size spiral wound RO membrane module (Dupont Filmtec‐BW30‐400) consist‐
Membranes 2021, 11, 353 ing of 10 membrane sheets that were 2.05 m long and 0.9 m wide (with an effective mem‐ 5 of 13
brane area of 400 ft ) and wounded along an Archimedes spiral [27] was modelled in this
2

work. Adjacent membrane sheets shared a common feed channel and ran parallel with
one another with some independency; hence, the module’s geometries could be simplified
PA, USA). Looking down from the axial direction, the Archimedean spiral for the single
to possess only a single feed channel for the simulation (Figure 3). The geometry of a single
feed channel rolled up along the length of the membranes had the following equations:
feed channel was created using ANSYS SpaceClaim (2019R3) (Ansys, Inc, USA). Looking
down from the axial direction, thexArchimedean
(θ ) = (α + βθ )spiral
cos(θ )for the single feed channel rolled
(11)
up along the length of the membranes had the following equations:
y𝑥
(θ )𝜃 = (α𝛼+ βθ𝛽𝜃 cos
) sin (θ )𝜃 (11)(12)
𝑦 𝜃 𝛼 𝛽𝜃 sin 𝜃 (12)
where x and y are the coordinates of points on the spiral, θ is the angle variable and α and
where 𝑥 and 𝑦 are the coordinates of points on the spiral, 𝜃 is the angle variable and α
β are constants. The constant α is the distance from the start of the spiral to the vertical
and 𝛽 are constants. The constant 𝛼 is the distance from the start of the spiral to the ver‐
axis, i.e., the radius of the tube and the constant β is the ratio between the radial expansion
tical axis, i.e., the radius of the tube and the constant 𝛽 is the ratio between the radial
speed of the spiral to its rotational angular speed, which is dependent on the number
expansion speed of the spiral to its rotational angular speed, which is dependent on the
of wounded membrane sheets, the thickness of the membrane sheets and the thickness
number of wounded membrane sheets, the thickness of the membrane sheets and the
of the corresponding feed channel. For the Dupont Filmtec-BW30-400 RO membranes,
thickness of the corresponding feed channel. For the Dupont Filmtec‐BW30‐400 RO mem‐
the corresponding α and β values were 0.02 m and 0.0062 π m, respectively.
. The measured
branes,
thicknesstheofcorresponding 𝛼 and
feed spacers, i.e., 𝛽 values
the feed were
channel 0.02was
width m and
0.86 mm. m, respectively. The
measured thickness of feed spacers, i.e., the feed channel width was 0.86 mm.

Figure 3. Geometries of the simulated SWRO membrane feed channel.

Figure 3. Geometries of the simulated SWRO membrane feed channel.
The simulated geometries were meshed using ANSYS Meshing (2019R3) (Ansys, Inc.,
Canonsburg,
The simulatedPA, USA). In order
geometries to accurately
were meshed using simulate
ANSYS theMeshing
CP phenomenon of inorganic
(2019R3) (Ansys, Inc,
salt contents
USA). In orderintothe boundary
accurately layer, five
simulate hexahedral
the CP phenomenon inflation layers that
of inorganic saltwere parallel
contents to
in the
the membrane
boundary layer,surface were created.
five hexahedral The first
inflation layers layer
thathad
were a height of to
parallel 0.02
themm and a growth
membrane sur‐
ratewere
face of 1.5created.
for eachThe
of the
firstfollowing
layer hadlayers.
a heightThe rest of
of 0.02 mm theand
geometries
a growthwere meshed
rate of 1.5 for using
each
of the following layers. The rest of the geometries were meshed using tetrahedral13
tetrahedral elements with an average mesh size of 0.05 mm. Approximately million
elements
elements
with were present
an average after
mesh size ofpassing
0.05 mm.theApproximately
mesh independence test. elements were present
13 million
after passing the mesh independence test.
2.3. Model Setup
The CP of the RO membrane module in the absence and presence of a feed spacer
2.3.Model Setup
was simulated using the commercial CFD package ANSYS CFX (2019R3) (Ansys, Inc.,
The CP ofPA,
Canonsburg, theUSA)
RO membrane
with 10,000 module
ppm NaCl in as
thefeed
absence and For
solution. presence of a feed
the full-size spacer
RO module
was simulated
simulation, using
three the commercial
different CFD package
NaCl concentrations wereANSYS CFXThe
modelled. (2019R3) (Ansys,
physical Inc,
properties
USA) withwere
of water 10,000
usedppm NaCl
as the as feed
impact solution.
of salt For the full‐size
concentrations RO module
on fluid physical simulation,
properties, such
three different
as viscosity NaCl
and concentrations
density, were modelled.
were negligible. The feedTheinletphysical
was setproperties of water
as the velocity were
inlet, and
used as the impact of salt concentrations on fluid physical properties, such as
the feed outlet was set as the pressure outlet. The surface of the feed spacer was set as a viscosity
and density,
non-slip were negligible.
boundary [28]. The The
k-ωfeed inlet
model was
was set as
used to the
modelvelocity
liquidinlet,
flow.and the feed outlet
A high-resolution
scheme was used to solve for fluid flow control equations. The membrane permeability
coefficient A (9.56 × 10−12 m/(s·Pa)) obtained from Dupont Filmtec-BW30-400’s published
data and salt (NaCl) permeability coefficient B (5.58 × 10−8 m/s) determined by utilizing
experimental measurement at various feed concentrations and pressure [17] were used in
this study. All simulated flow velocities and pressures converged to 10−4 and inorganic
salt concentrations converged to 10−6 .
Membranes 2021, 11, 353 6 of 13

2.4. Model Validation

A small crossflow RO membrane test apparatus was established for the validation of
CFD models. The feed channel had a length of 145.0 mm, a width of 35.0 mm and a height
of 3.5 mm with an effective membrane area of 48.12 cm2 (Figure A1). Simulation of the
test apparatus was conducted using the aforementioned methods. By changing the feed
velocities and the concentrations of the NaCl solution, the simulated permeate flowrate
was compared with the experimentally obtained data.

3. Results and Discussion

3.1. Model Validation
A comparison between simulated permeate flowrate and experimentally measured
data showed that the inconsistencies were less than 5% across three different experimental
conditions (Table 1). This suggests that the CFD approach used in this investigation can
capture the fluid flow and transport phenomena with a high degree of accuracy.

Table 1. Comparison of the simulation results with experimentally measured data at three different conditions.

Experimental Simulated
Experiment Feed Pressure Feed Flowrate Feed Concentration,
Permeate Flowrate Permeate Flowrate Error
Number (MPa) (mL/min) NaCl (mg/L)
(mL/min) (mL/min)
1 1.8 169.68 4509.4 4.67 ± 0.1 4.50 3.58%
2 1.4 177.24 4509.4 3.75 ± 0.08 3.67 2.08%
3 1.7 177.24 2089.6 6.44 ± 0.17 6.16 4.44%

3.2. Impact of Feed Spacer on Lab Scale Flat Sheet Module

A wave-like flow pattern was observed in the flow field due to the presence of the
Membranes 2021, 11, x spacer filaments (Figure 4a). Strong vortices appeared between the spacer filament 7 of and
15 the
membrane surface, which enhanced the flow velocities on the membrane surface (in the
Y direction). These vortices disappeared in the absence of the spacer and led to turbulent
flow
the Yin the middle
direction). of the
These channel
vortices (in bulk in
disappeared solution) while
the absence stagnant
of the spacerflow
and ledin the boundary
to turbu-
layer where mass transfer relies on diffusion (Figure 4b). The presence of
lent flow in the middle of the channel (in bulk solution) while stagnant flow in the bound- the counter-
flow
ary layer where mass transfer relies on diffusion (Figure 4b). The presence of the counter- and
vortices inside the spacer-filled channel reduced the boundary layer thickness
consequently increased the averagechannel
mass transfer × 10 −4 compared to
flow vortices inside the spacer-filled reducedcoefficient to 1.59
the boundary layer thickness and
7.95 − 5
× 10 inincreased
the absence
consequently the of the spacer
average massand reduced
transfer the CP.toThe
coefficient 1.59mass
× 10-4transfer
compared coefficient
to
on the
7.95 membrane
× 10 surfaceofcan
-5 in the absence the be calculated
spacer by thethe
and reduced following:
CP. The mass transfer coefficient
on the membrane surface can be calculated by the following:
D
𝐷 k= (13)
𝑘= δ (13)
𝛿
where k𝑘isisthe
where themass
masstransfer coefficient, 𝐷D is
transfer coefficient, is diffusion
diffusioncoefficient
coefficientand
and𝛿 δisisthe
the thickness of
thickness
the boundary layer.
of the boundary layer.

(a) Velocity contour plot with feed spacer. (b) Velocity contour plot without feed spacer.

Figure 4. Velocity contour plot with the feed spacer (a) and without the feed spacer (b).
Figure 4. Velocity contour plot with the feed spacer (a) and without the feed spacer (b).

It is also interesting to observe that, in the presence of the spacer, the salt concentra-
tion in the feed channel was lower in the area that is further away from the spacer fila-
ments while a higher concentration profile appeared over the back side of these filaments
Membranes 2021, 11, 353 7 of 13

It is also interesting to observe that, in the presence of the spacer, the salt concen-
tration in the feed channel was lower in the area that is further away from the spacer
Membranes 2021, 11, x filaments while a higher concentration profile appeared over the back side of these 8 of fila-
15
ments (Figure 5a). In contrast, salt concentration is uniform across the membrane surface
in the absence of the spacer and increased steadily along the direction of the flow.

mg/L

(a) Salt concentration distribution in the presence of the spacer on the x-y plane where z = 0.41 mm.

mg/L

(b) Salt concentration distribution in the absence of the spacer on the x-y plane where z = 0.41 mm.

Saltconcentration
Figure5.5.Salt
Figure concentrationdistribution
distribution
onon
thethe
x-yx-y plane
plane where
where z =z0.41mm
= 0.41 mm inpresence
in the the presence
and and
absence of the feed spacer: (a) with the spacer; (b) without the spacer.
absence of the feed spacer: a) with the spacer; b) without the spacer.

To further
To further investigate
investigatethe
theroot
rootcause
causeof
ofthe
theCP
CPphenomenon,
phenomenon,an ananalysis
analysiswas
wascarried
carried
out on the cross section of the feed channel (the y-z plane). It can be seen that
out on the cross section of the feed channel (the y-z plane). It can be seen that the feed the feed
spacer and the incoming flow created backflow vertically to the membrane surface,
spacer and the incoming flow created backflow vertically to the membrane surface, which which
enhanced the transfer of salt into the feed channel. Consequently, this greatly decreased
enhanced the transfer of salt into the feed channel. Consequently, this greatly decreased
the CP phenomenon at the flow facing side. In contrast, the local stagnant zone at the
the CP phenomenon at the flow facing side. In contrast, the local stagnant zone at the flow
flow leaving side of the spacer increased the local concentration in the boundary layer
leaving side of the spacer increased the local concentration in the boundary layer and a
and a stronger CP phenomenon was observed (Figure 6). This is consistent with the
stronger CP phenomenon was observed (Figure 6). This is consistent with the flow behav-
flow behaviour where higher vortices enhanced the back transport of salt away from the
iour where higher vortices enhanced the back transport of salt away from the membrane
membrane surface.
surface.
 leaving side of the spacer increased the local concentration in the boundary layer and a
enhanced the transfer of salt into the feed channel. Consequently, this greatly decreased
stronger CP phenomenon was observed (
the CP phenomenon at the flow facing side. In contrast, the local stagnant zone at the flow
Figure 6). This is consistent with the flow behaviour where higher vortices enhanced
leaving side of the spacer increased the local concentration in the boundary layer and a
the back transport of salt away from the membrane surface.
stronger CP phenomenon was observed (
Membranes 2021, 11, 353 8 of 13
Figure 6). This is consistent with the flow behaviour where higher vortices enhanced
the back transport of salt away from the membrane surface.

mg/L

mg/L
Figure 6. Salt concentration in the feed channel in the presence of the spacer at y = 1/2 of the width
of the computational domain.
Figure 6. Salt concentration in the feed channel in the presence of the spacer at y = 1/2 of the width of the computational domain.

A quantitative analysis showed that

A quantitative the presence
analysis of the
showed that thefeed spacerofled
presence thetofeed
a periodic
spacer led to a peri-
Figure 6. Salt concentration
fluctuation of saltodic in the feedinchannel
concentration the in the presence
boundary layer of the spacer
between at ymg/L
10,196 = 1/2 and
of the width
11,504
fluctuation of salt concentration in the boundary layer between 10,196 mg/L and
of the computational
mg/L while the salt domain.
concentration in the
11,504 mg/L while the feed channel without
salt concentration in thea spacer increased
feed channel alongathe
without spacer increased
flow direction from 10,195
along the mg/L to 11,694from
flow direction mg/L (Figure
10,195 7). to 11,694 mg/L (Figure 7).
mg/L
A quantitative analysis showed that the presence of the feed spacer led to a periodic
fluctuation of salt concentration in the boundary layer between 10,196 mg/L and 11,504
12,100
mg/L while the salt concentration in the feed channel without a spacer increased along the
Without feed spacer
NaCl Concentration in bounadry layer

flow direction from 10,195 mg/L to 11,694 mg/L (Figure 7).

11,700 With feed spacer

12,100
11,300 Without feed spacer
NaCl Concentration in bounadry layer

11,700 With feed spacer

(ppm)

10,900

11,300
10,500
(ppm)

10,900
10,100
0 0.005 0.01 0.015 0.02
10,500 Distance from inlet (m)

Figure 7. Salt concentration profiles in the boundary layer of the crossflow flat sheet module in the
10,100
absence and presence of the feed spacer.
0 0.005 0.01 0.015 0.02
Although the spacer enhances mass transfer and creates higher turbulence, it also
Distance from inlet (m)
causes higher pressure drops, which is adverse to the filtration process. With the feed
pressure of 1.55 MPa in the simulated conditions, the total pressure loss was approximately
369.75 Pa in the presence of the spacer compared to the total pressure loss of 56.5 Pa in the
absence of the spacer (Figure 8). Taking into consideration the effects of mass transfer and
pressure loss, total flux was 27.03 L/(m2 h) (LMH) in the presence of the spacer compared
to 25.23 LMH in the absence of the spacer (Figure 9).
 causes higher pressure drops, which is adverse to the filtration process. With the feed
Although the spacer enhances mass transfer and creates higher turbulence, it also
pressure of 1.55 MPa in the simulated conditions, the total pressure loss was approxi‐
causes higher pressure drops, which is adverse to the filtration process. With the feed
mately 369.75 Pa inMPa
pressure of 1.55 the in
presence of the spacer
the simulated compared
conditions, the totaltopressure
the totalloss
pressure loss of 56.5
was approxi‐
Pa mately
in the absence
369.75 Paof
in the spacer (Figure
the presence 8). Taking
of the spacer into to
compared consideration the loss
the total pressure effects of mass
of 56.5
transfer
Membranes 2021, 11, and
353 pressure loss, total flux was 27.03 L/(m 2 h) (LMH) in the presence of the spacer
Pa in the absence of the spacer (Figure 8). Taking into consideration the effects of mass 9 of 13
compared to 25.23
transfer and LMH
pressure in total
loss, the absence of theL/(m
flux was 27.03 spacer (Figure
2 h) (LMH) in9).
the presence of the spacer
compared to 25.23 LMH in the absence of the spacer (Figure 9).
400
400
Without feed spacer
Without feed spacer
300 With feed spacer
300 With feed spacer

Pressure drop (Pa)

Pressure drop (Pa)
200
200

100
100

0
0
0 0.004 0.008 0.012 0.016
0 0.004 0.008 0.012 0.016
Distance from inlet (m)
Distance from inlet (m)
Figure 8. Pressure loss along the crossflow flat sheet module in the absence and presence of the
Figure
Figure 8. Pressure loss 8. Pressure
along loss along
the crossflow the crossflow
flatspacer.
feed sheet moduleflat
insheet moduleand
the absence in the absenceof
presence and
thepresence of the
feed spacer. feed spacer.

29
Without feed spacer
29
With feed spacer
Without feed spacer
28 With feed spacer
28
Water flux (LMH)

27
Water flux (LMH)

27
26

26
25

25
24
0 0.005 0.01 0.015

24 Distance from inlet(m)

Figure 9. PermeateFigure
0
flux of9.thePermeate
0.005
crossflowflux
flat of themodule
sheet
0.01in the
crossflow 0.015
flatabsence
sheet module in the of
and presence absence
the feedand presence of the
feed spacer. Distance from inlet(m)
spacer.

3.3. Impact 3.3. Impact of Spacer

of Spacer on Full-Size RO Module
Figure 9. Permeate fluxonofFull‐Size RO Module
the crossflow flat sheet module in the absence and presence of the feed
Simulations were conducted
spacer. on the full-size RO module (Dupont Filmtec-BW30-
400) with a feed pressure of 1.09 MPa. The simulation results on 10,000 mg/L of feed
3.3. Impact of Spacerconcentration
on Full‐Size RO showed
Modulethat permeate flux reduced linearly along the feed direction in the
absence of the feed spacer (from 27.6 LMH near the inlet to 24.1 LMH near the outlet)
(Figure 10). The flow velocity had a minimal component that was perpendicular to the feed
direction and tangential to the Archimedes curve; it had a magnitude of less than 0.003 m/s,
which was only 1.5% of the average axial flow velocity. Thus, the flow velocity’s impact
on the flow field was limited and this is consistent with previous researchers’ conclusion
that the curvature of the membrane surface has a negligible effect on the flow distribution
within the feed spacer via the simulation of a single feed spacer module [29,30].
 was only 1.5% of the average axial flow velocity. Thus, the flow velocity’s impact on the
flow field was limited and this is consistent with previous researchers’ conclusion that the
curvature of the membrane surface has a negligible effect on the flow distribution within
the feed spacer via the simulation of a single feed spacer module [29,30].
Membranes 2021, 11, 353 10 of 13

Figure 10. Permeate flux contour plot of the full-size spiral wound RO membrane module.

It can
Figure 10.be observed
Permeate that,
flux for theplot
contour full-size
of the module, feed spacers
full‐size spiral wound played a more important
RO membrane module.
role mitigating CP. Under the same feed concentration conditions (feed pressure of 1.09 MPa
and saltIt concentration of 10,000
can be observed that,ppm),
for thea severe concentration
full‐size increase
module, feed was played
spacers observed a (the
more im‐
outlet concentration increased by 84.67% compared to the inlet concentration) in the absence
portant role mitigating CP. Under the same feed concentration conditions (feed pressure
of the spacer while, in the presence of the spacer, the increase was only 15.30% (Figure 11).
nes 2021, 11, x
of 1.09 MPa and salt concentration of 10,000 ppm), a severe concentration increase
11 of 14 in
was
The corresponding fluxes decreased from 27.6 LMH at the inlet to 24.1 LMH at the outlet
observed (the outlet concentration increased by 84.67% compared to the inlet concentra‐
the absence of the spacer while there was only a minor drop to 26.5 LMH in the presence
oftion) in the(Figure
the spacer absence
12).of the spacer while, in the presence of the spacer, the increase was
only 15.30% (Figure 11). The corresponding fluxes decreased from 27.6 LMH at the inlet
to 24.1
5500
LMH at the outlet in the absence of the spacer while there was only a minor drop
to 26.5 LMH in the presence of the spacer (Figure 12).
RO module with spacer
NaCl equivalent Concentration(ppm)

RO module without spacer

4500

3500

2500
0 0.2 0.4 0.6 0.8

Distance from inlet (m)

Figure 11. Salt concentration distribution of the spiral wound RO module along the feed direction in
Figure 11. Salt concentration distribution of the spiral wound RO module along the feed direction
the absence and presence of the feed spacer.
in the absence and presence of the feed spacer.

28

27
H）
 0 0.2 0.4 0.6 0.8

Distance from inlet (m)

Figure
Membranes 2021, 11.
11, 353Salt concentration distribution of the spiral wound RO module along the feed direction
11 of 13
in the absence and presence of the feed spacer.

28

27

Water flux（LMH） 26

25

24
RO module with feed spacer
RO module without feed spacer
23
0 0.2 0.4 0.6 0.8

Distance from inlet (m)

Figure 12. Flux distributions of the spiral wound RO module along the feed direction in the absence
Figure 12. Flux and presenceof
distributions of the
thefeed spacer.
spiral wound RO module along the feed direction in the ab‐
sence and presence
4. of the feed spacer.
Conclusions
A 3D CFD model was developed in this work to provide insights into the role of feed
4. Conclusions spacers in full-scale spiral wound RO modules. The mechanisms of how flow velocities
and mass
A 3D CFD model transfer
was differ in ainfluid
developed thischannel
work to in the absenceinsights
provide and presence
intoofthea feed
rolespacer were
of feed
presented. The variations in salt concentration and pressure drops caused by the spacer
spacers in full‐scale
from spiral
the inletwound RO modules.
to the outlet The mechanisms
were quantified. of how
This investigation flow
provides thevelocities
foundation
and mass transferfor differ
using 3Din CFD
a fluid
toolschannel in theand
for the design absence and presence
optimization ofRO
of full-scale a feed spacer
modules. The
were presented.accurate
The variations
prediction ofinCPsalt
andconcentration and
flux profiles on the pressure
different drops
locations of a caused
membrane bymodule
the
spacer from the could enable
inlet to the aoutlet
better were
understanding of fouling
quantified. potential and anprovides
This investigation improved the module
foun‐and
spacer design. This investigation enables the evaluation and comparison of full-scale RO
dation for using 3D CFD tools for the design and optimization of full‐scale RO modules.
modules by different manufactures at different operating conditions, which is still a subject
The accurate prediction
of on-goingofwork.
CP and flux profiles on the different locations of a membrane
module could enable a better understanding of fouling potential and an improved module
Author
and spacer design. ThisContributions:
investigation Conceptualization,
enables the Y.W. and W.W.;and
evaluation methodology,
comparisonX.Z. andofY.W.; software,
full‐scale
X.Z. and Y.W.; validation, X.Z.; formal analysis, X.Z. and X.J.; investigation, X.Z.; resources, W.W., R.Z.
and K.Z.; data curation, X.Z. and Y.W.; writing—original draft preparation, X.Z. and X.J.; writing—
review and editing, Y.W. and W.W.; visualization, X.Z. and X.J.; supervision, W.W. and Y.W.; project
administration, K.Z.; funding acquisition, W.W. and R.Z. All authors have read and agreed to the
published version of the manuscript.
Funding: This research was funded by the Science and Technology Innovation and Entrepreneurship
Fund Project of Tiandi Technology Co., Ltd., grant number 2019-TD-GH001.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: All data presented in this study are available in the current article.
Acknowledgments: The authors would like to acknowledge Xuefei Liu for her contribution on the
CFD modelling approach and Shunzhi Lv for setting up the crossflow rig for model validation.
Conflicts of Interest: The authors declare no conflict of interest. The funders had no role in the design
of the study; in the collection, analyses or interpretation of data; in the writing of the manuscript; or
in the decision to publish the results.
 Membranes 2021, 11, 353 12 of 13

Nomenclature

A permeability coefficient of water through the corresponding membrane (m·s−1 ·Pa−1 )

a effective membrane surface area for solution passage (m2 )
B permeability coefficient of inorganic salt (m·s−1 )
D diffusion coefficient (m2 ·s−1 )
δ boundary layer thickness (m)
Cm concentration of the inorganic salt on the feed side of the membrane (kg·m−3 )
Cp concentration of the inorganic salt on the permeate side of the membrane (kg·m−3 )
K mass transfer coefficient
J permeate flux, (m3 ·m−2 ·s−1 )
Js salt flux (kg·m−2 ·s−1 )
µ fluid viscosity (kg·m−1 ·s−1 )
∆P pressure difference across both sides of the membrane (Pa)
∆Π osmotic pressure difference across both sides of the membrane (Pa)
p pressure (Pa)
ρ fluid density (kg·m−3 )
Membranes 2021, 11, x 13 of 14
S source term
u −
fluid velocity vector (m·s ) 1

V corresponding effective computational domain volume (m3 )

v flow velocity (m·s−1 ) (u is the vector)

Appendix
Appendix A A

Figure
Figure A1. A1. Schematic
Schematic diagram
diagram of the
of the crossflow
crossflow RO RO
testtest apparatus
apparatus for for model
model validation.
validation.
References
1. Fritzmann, C.; Löwenberg, J.; Wintgens, T.; Melin, T. State-of-the-art of reverse osmosis desalination. Desalination 2007, 216, 1–76.
[CrossRef]
2. Qasim, M.; Badrelzaman, M.; Darwish, N.N.; Darwish, N.A.; Hilal, N. Reverse osmosis desalination: A state-of-the-art review.
ReferencesDesalination 2019, 459, 59–104. [CrossRef]
1. 3. Fritzmann,
Mulder, C.;
J.; Mulder,
Löwenberg, C. Basic Principles of
J.; Wintgens, T.;Membrane
Melin, T. Technology, 2nd ed.;
State‐of‐the‐art Kluwerosmosis
of reverse Academic: Dordrecht,
desalination. The Netherlands,
Desalination 1996.
2007, 216, 1–
4. 76. Fimbres-Weihs, G.; Wiley, D. Numerical study of mass transfer in three-dimensional spacer-filled narrow channels with steady
2. Qasim, J. Membr.
flow.M.; Sci. 2007,
Badrelzaman, M.;306, 228–243.
Darwish, [CrossRef]
N.N.; Darwish, N.A.; Hilal, N. Reverse osmosis desalination: A state‐of‐the‐art review.
Li, F.; Meindersma,
5. Desalination W.;
2019, 459, 59–104. De Haan, A.; Reith, T. Optimization of commercial net spacers in spiral wound membrane modules.
3. J. Membr. Sci. 2002, 208, 289–302. [CrossRef]
Mulder, J.; Mulder, C. Basic Principles of Membrane Technology, 2nd ed.; Kluwer Academic: Dordrecht, The Netherlands, 1996.
4. 6. Fimbres‐Weihs,
Shakaib, M.; Hasani,
G.; Wiley,S.; Mahmood,
D. Numerical M. study
Study ofonmass
the effects of spacer
transfer geometry in membrane
in three‐dimensional feed channels
spacer‐filled using three-dimensional
narrow channels with steady
computational flow modeling.
flow. J. Membr. Sci. 2007, 306, 228–243. J. Membr. Sci. 2007, 297, 74–89. [CrossRef]
5. 7. Li, F.;
Vrouwenvelder,
Meindersma, W.; J.S.;De
Picioreanu,
Haan, A.;C.; Kruithof,
Reith, J.C.; Van Loosdrecht,
T. Optimization of commercial M.C.M. Biofouling
net spacers in spiral
in spiral wound wound membrane
membrane systems:
modules. J.
Three-dimensional
Membr. CFD model based evaluation of experimental data. J. Membr. Sci. 2010, 346, 71–85. [CrossRef]
Sci. 2002, 208, 289–302.
6. 8. Shakaib,
Picioreanu, C.; Vrouwenvelder,
M.; Hasani, S.; Mahmood, J.; M.van Loosdrecht,
Study M. Three-dimensional
on the effects of spacer geometrymodeling
in membraneof biofouling and fluid
feed channels usingdynamics in feed
three‐dimen‐
spacer
sional channels offlow
computational membrane
modeling.devices. J. Membr.
J. Membr. Sci. 2009,
Sci. 2007, 345, 340–354. [CrossRef]
297, 74–89.
7. 9. Vrouwenvelder,
Schwinge, J.; J.S.;Neal, P.R.; Wiley,
Picioreanu, C.;D.E.; Fane, J.C.;
Kruithof, A.G.Van
Estimation of foulant
Loosdrecht, M.C.M. deposition
Biofouling across the leaf
in spiral of amembrane
wound spiral-wound module.
systems:
Desalination 2002, 146, 203–208. [CrossRef]
Three‐dimensional CFD model based evaluation of experimental data. J. Membr. Sci. 2010, 346, 71–85.
8. 10.Picioreanu,
Schwinge, C.;J.;Vrouwenvelder,
Wiley, D.E.; Fletcher,
J.; vanD. Loosdrecht,
A CFD studyM. of unsteady flow in narrow
Three‐dimensional spacer-filled
modeling channels
of biofouling andforfluid
spiral-wound
dynamicsmem-brane
in feed
modules. Desalination 2002, 146, 195–201. [CrossRef]
spacer channels of membrane devices. J. Membr. Sci. 2009, 345, 340–354.
9. Schwinge, J.; Neal, P.R.; Wiley, D.E.; Fane, A.G. Estimation of foulant deposition across the leaf of a spiral‐wound module.
Desalination 2002, 146, 203–208.
10. Schwinge, J.; Wiley, D.E.; Fletcher, D. A CFD study of unsteady flow in narrow spacer‐filled channels for spiral‐wound mem‐
brane modules. Desalination 2002, 146, 195–201.
11. Schwinge, J.; Wiley, D.E.; Fletcher, D.F. Simulation of Unsteady Flow and Vortex Shedding for Narrow Spacer‐Filled Channels.
Ind. Eng. Chem. Res. 2003, 42, 4962–4977.
Membranes 2021, 11, 353 13 of 13

11. Schwinge, J.; Wiley, D.E.; Fletcher, D.F. Simulation of Unsteady Flow and Vortex Shedding for Narrow Spacer-Filled Channels.
Ind. Eng. Chem. Res. 2003, 42, 4962–4977. [CrossRef]
12. Ahmad, A.; Lau, K. Impact of different spacer filaments geometries on 2D unsteady hydrodynamics and concentration polarization
in spiral wound membrane channel. J. Membr. Sci. 2006, 286, 77–92. [CrossRef]
13. Cao, Z.; Wiley, D.E.; Fane, A.G. CFD simulations of net-type turbulence promoters in a narrow channel. J. Membr. Sci. 2001,
185, 157–176. [CrossRef]
14. Ranade, V.V.; Kumar, A. Fluid dynamics of spacer filled rectangular and curvilinear channels. J. Membr. Sci. 2006, 271, 1–15.
[CrossRef]
15. Koutsou, C.P.; Yiantsios, S.G.; Karabelas, A.J. A numerical and experimental study of mass transfer in spacer-filled channels:
Effects of spacer geometrical characteristics and Schmidt number. J. Membr. Sci. 2009, 326, 234–251. [CrossRef]
16. Koutsou, C.P.; Yiantsios, S.G.; Karabelas, A.J.A. Direct numerical simulation of flow in spacer-filled channels: Effect of spacer
geometrical characteristics. J. Membr. Sci. 2007, 291, 53–69. [CrossRef]
17. Ruiz-García, A.; de la Nuez Pestana, I. Feed spacer geometries and permeability coefficients. Effect on the performance in BWRO
Spiral-wound membrane modules. Water 2019, 11, 152. [CrossRef]
18. Toh, K.; Liang, Y.; Lau, W.; Weihs, G.F. A Review of CFD Modelling and Performance Metrics for Osmotic Membrane Processes.
Membranes 2020, 10, 285. [CrossRef]
19. Abdelbaky, M.M.A.; El-Refaee, M.M. A 3D CFD comparative study between torsioned and non-torsioned net-type feed spacer in
reverse osmosis. SN Appl. Sci. 2019, 1, 1059. [CrossRef]
20. Toh, K.; Liang, Y.; Lau, W.; Weihs, G.F. 3D CFD study on hydrodynamics and mass transfer phenomena for SWM feed spacer
with different floating characteristics. Chem. Eng. Res. Des. 2020, 159, 36–46. [CrossRef]
21. Haidari, A.; Heijman, S.; Uijttewaal, W.; van der Meer, W. Determining effects of spacer orientations on channel hydraulic
conditions using PIV. J. Water Process. Eng. 2019, 31, 100820. [CrossRef]
22. Radu, A.; van Steen, M.; Vrouwenvelder, J.; van Loosdrecht, M.; Picioreanu, C. Spacer geometry and particle deposition in spiral
wound membrane feed channels. Water Res. 2014, 64, 160–176. [CrossRef]
23. Ishigami, T.; Matsuyama, H. Numerical Modeling of Concentration Polarization in Spacer-filled Channel with Permeation across
Reverse Osmosis Membrane. Ind. Eng. Chem. Res. 2015, 54, 1665–1674. [CrossRef]
24. Gu, B.; Xu, X.Y.; Adjiman, C.S. A predictive model for spiral wound reverse osmosis membrane modules: The effect of winding
geometry and accurate geometric details. Comput. Chem. Eng. 2017, 96, 248–265. [CrossRef]
25. Liang, Y.Y.; Chapman, M.B.; Fimbres-Weihs, G.; Wiley, D.E. CFD modelling of electro-osmotic permeate flux enhancement on the
feed side of a membrane module. J. Membr. Sci. 2014, 470, 378–388. [CrossRef]
26. Wiley, D.E.; Fletcher, D.F. Techniques for computational fluid dynamics modelling of flow in membrane channels. J. Membr. Sci.
2003, 211, 127–137. [CrossRef]
27. Schwinge, J.; Neal, P.R.; Wiley, D.E.; Fane, A.G. Spiral wound modules and spacers: Review and analysis. J. Membr. Sci. 2004,
242, 129–153. [CrossRef]
28. Bucs, S.S.; Linares, R.V.; Marston, J.O.; Radu, A.I.; Vrouwenvelder, J.S.; Picioreanu, C. Experimental and numerical charac-
terization of the water flow in spacer-filled channels of spiral-wound membranes. Water Res. 2015, 87, 299–310. [CrossRef]
29. Li, Y.-L.; Tung, K.-L.; Lu, M.-Y.; Huang, S.-H. Mitigating the curvature effect of the spacer-filled channel in a spiral-wound
membrane module. J. Membr. Sci. 2009, 329, 106–118. [CrossRef]
30. Ranade, V.; Kumar, A. Comparison of flow structures in spacer-filled flat and annular channels. Desalination 2006, 191, 236–244.
[CrossRef]