Journal of Membrane Science 527 (2017) 78–91

Contents lists available at ScienceDirect

Journal of Membrane Science

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

The effect of feed spacer geometry on membrane performance and MARK

concentration polarisation based on 3D CFD simulations
⁎
Boram Gu, Claire S. Adjiman, Xiao Yun Xu
Department of Chemical Engineering, Imperial College London, South Kensington Campus, London SW7 2AZ, UK

A R T I C L E I N F O A BS T RAC T

Keywords: Feed spacers are used in spiral wound reverse osmosis (RO) membrane modules to keep the membrane sheets
Reverse osmosis (RO) apart as well as to enhance mixing. They are beneficial to membrane performance but at the expense of
Spiral wound module additional pressure loss. In this study, four types of feed spacer configurations are investigated, with a total of 20
Feed spacers geometric variations based on commercially available spacers and selected filament angles. The impact of feed
Computational fluid dynamics (CFD)
spacer design on membrane performance is investigated by means of three-dimensional (3D) computational
Membrane performance
fluid dynamics (CFD) simulations, where the solution-diffusion model is employed for water and solute
transport through RO membranes. Numerical simulation results show that, for the operating and geometric
conditions examined, fully woven spacers outperform other spacer configurations in mitigating concentration
polarisation (CP). When designed with a mesh angle of 60°, fully woven spacers also deliver the highest water
flux, although the associated pressure drops are slightly higher than their nonwoven counterparts. Middle layer
geometries with a mesh angle of 30° produce the lowest water flux. On the other hand, spacers with a mesh
angle of 90° show the lowest pressure drop among all the filament arrangements examined. Furthermore, the
computational model presented here can also be used to predict membrane performance for a given feed spacer
type and geometry.

1. Introduction spacer-filled channels [1,2,5,7–10,12–20,22–32]. Early CFD work

focussed on flow, solute concentration and pressure distributions in
Mass transfer phenomena in reverse osmosis (RO) processes have simplified two-dimensional (2D) geometries [8,12–16,18–20,30,31],
been studied experimentally and theoretically for many decades in including 2D cross-sectional models for cavity, submerged and zigzag
order to understand how water and solute permeation occur within configurations [12,14,15,19]. Different spacer shapes were also exam-
membrane modules [1–16], so as to identify the factors that determine ined in several studies [13,16,18]. Ahmad et al. compared rectangular,
the performance of membrane modules. One of the most critical mass cylindrical and triangular spacers and found the triangular shape to be
transfer phenomena present in an RO membrane module is concentra- most effective in mitigating CP [16]. Despite the computational
tion polarisation (CP), which is caused by the accumulation of solute(s) efficiency and usefulness of 2D CFD models, they are limited in
rejected by a selective RO membrane, resulting in an elevated solute capturing the complex geometric features of spacers, e.g., the intersec-
concentration on the membrane surface [13–16]. In a spiral wound tion of filaments cannot be incorporated in 2D models.
membrane (SWM) module, which is the type of module most widely With the rapid increase in computational power, three-dimensional
used in large scale RO plants, spacers are inserted between membrane (3D) CFD models of spacer-filled channels have emerged
sheets to keep them apart [37]. Feed spacers also promote mixing by [1,2,5,7,10,22–29]. Several studies reported flow patterns and mass
inducing vortices near the membrane surfaces, which can mitigate CP transfer in 3D models of nonwoven spacers, with a focus on the effects
but at the expense of an increased pressure drop [2,5,10,12,17]. It is of geometric parameters such as the filament distance-to-diameter
therefore important to elucidate the effects of spacers on mass and ratio, the mesh angle and flow attack angle [1,7,23,26]. Picioreanu
momentum transfer in an SWM module with the ultimate aim to et al. improved their 3D CFD model by introducing a more realistic
improve module performance. nonwoven spacer geometry based on microscopic images, where
In recent years computational fluid dynamics (CFD) has been variations in filament diameter were captured [29]. Other types of
increasingly used to study mass and momentum transfer in feed spacers that have been considered in 3D CFD models include ‘Parallel

⁎
Corresponding author.
E-mail addresses: b.gu13@imperial.ac.uk (B. Gu), c.adjiman@imperial.ac.uk (C.S. Adjiman), yun.xu@imperial.ac.uk (X.Y. Xu).

http://dx.doi.org/10.1016/j.memsci.2016.12.058
Received 20 September 2016; Received in revised form 29 November 2016; Accepted 28 December 2016
Available online 28 December 2016
0376-7388/ Crown Copyright © 2017 Published by Elsevier B.V. All rights reserved.
B. Gu et al. Journal of Membrane Science 527 (2017) 78–91

Nomenclature X1 term used in Eq. (10).

X2 term used in Eq. (10).
Symbols y Cartesian coordinate
z Cartesian coordinate
A water permeability [m/(Pa s)]
B solute permeability [mol/(m2 s)] Greek letters
c concentration [mol/m3]
Df filament diameter in nonwoven and woven spacers [m] θa angle between inflow direction and axial filaments (attack
Df,1 filament diameter of thin filaments in middle layer angle) [deg]
spacers [m] θf angle between transverse and axial filaments (mesh angle)
Df,2 filament diameter of thick filaments in middle layer [deg]
spacers [m] θt angle between inflow direction and transverse filaments
Ds diffusion coefficient [m2/s] [deg]
Hc channel height [m] ρ density of fluid [kg/m3]
Jw water flux [m3/m2 s] μ viscosity of fluid [Pa s]
Js solute flux [mol/m2 s] π osmotic pressure [Pa]
Lf filament distance [m] κ osmotic factor [m3 Pa/mol]
M concentration polarisation modulus [dimensionless]
u velocity vector [m/s] Subscripts
n normal vector [dimensionless]
N flux [mol/m2 s] b bulk
p pressure [Pa] f feed side
Δp pressure drop per unit length [Pa/m] l leaking
R gas constant [J/(mol K)] low at lower membrane wall
Rs membrane rejection [dimensionless] m at membrane wall
T temperature [K] p permeate side
x Cartesian coordinate or ratio of reduced water flux to bulk up at upper membrane wall
water flux [dimensionless]

spacer’ [2,40] and ‘Middle layer spacer’ [31], both consisting of thick [32,33,36]. The most widely accepted transport theory for RO mem-
filaments with a diameter equivalent to the channel height and thinner brane is the solution-diffusion model [32,33], where the solute flux is
filaments piercing through the thick ones. However, 3D models of related to the solute permeability and the difference in concentration
woven spacers with realistic spacer geometries are lacking. There is between the two sides of a membrane. The predicted solute fluxes can
clearly a need to investigate flow, pressure and concentration patterns then be used to calculate the solute concentration in the produced
in spacer-filled feed channels for a variety of spacer types accounting permeate, thereby obtaining the solute rejection of the membrane.
for realistic spacer geometries. In this study, the solution-diffusion model is employed in order to
In addition to capturing geometric details, it is equally important to provide a more realistic description of water and solute transport
specify appropriate boundary conditions, especially those on mem- phenomena in an RO membrane spacer-filled channel for various
brane walls, which have a direct influence on the predicted momentum spacer geometries by means of 3D CFD simulations. Four spacer types
and mass transfer behaviour in spacer-filled channels [1,2,5– are selected: nonwoven, partially woven, middle layer and fully woven.
7,10,11,13–16,21,38,39]. There are essentially three types of boundary For each spacer type, the mesh angle and flow attack angle (or filament
conditions for water flux through membranes: no water permeation, orientation) are varied according to information available on commer-
constant water flux and the solution-diffusion model. More options cial nonwoven spacers [28]. The influence of spacer type and config-
exist for solute flux through membranes. In a number of studies, uration on membrane performance is assessed quantitatively. In
membranes have been treated as an impermeable wall but with a addition to evaluating and comparing direct performance indicators,
constant solute concentration to allow for convection-dominated mass such as water and solute fluxes and pressure drop, concentration
transfer near the membrane wall [1,2,5,7,9,38–40]. Although this polarisation at the membrane wall is also compared among different
approach is relatively simple and provides useful information on spacer configurations.
enhanced mass transfer due to the presence of feed spacers, simulation
results with a constant wall concentration boundary condition cannot 2. Methods
be used to predict membrane performance directly. As such, spacers of
different types and geometries have been compared in terms of local This study aims to provide not only a qualitative analysis of
shear stress and mass transfer coefficient [1,2,5,7,9,38–40]. A com- momentum and mass transfer behaviours but also quantitative results
monly used approach to account for solute permeation is to specify a on water and salt permeation in an SWM module with various types of
constant membrane rejection combined with water flux and wall spacers. In this section, we present details of the model geometry, the
concentration [6,12,15,18,21,26,30,31]. This can be expressed as: Js governing equations required to construct a coupled fluid dynamics
= Jwcm(1−Rs) where Js is the solute flux, Jw the water flux (or flow and mass transfer model, the boundary conditions and the numerical
velocity on the membrane wall), cm the solute concentration on the procedures used.
membrane wall and Rs the membrane rejection. Introducing the
rejection term is clearly more useful than imposing a constant solute
2.1. Feed spacer geometry
concentration on the membrane wall due to the more realistic
representation of membrane [10], but it does not describe accurately
There are two basic feed spacer configurations: nonwoven and
mass transport through RO membranes, as this is governed by
woven [23], which are commercially available net spacer types. In this
diffusion due to its small size pores (less than a nanometre)
study four different spacer configurations are created; these include

79
B. Gu et al. Journal of Membrane Science 527 (2017) 78–91

nonwoven, partially woven, middle layer and fully woven spacers, as

shown in Fig. 1. The nonwoven configuration is the simplest type,
which consists of two layers of straight filaments. The woven config-
uration can be divided into two different types depending on the extent
to which the spacer filaments are woven together: partially woven and
fully woven, as shown in Figs. 1(b) and (d), respectively. The fully
woven configuration, consisting of curved strands being woven to-
gether, has a highly complex geometry. In the partially woven config-
uration, one set of filaments remains straight, with each strand being in
contact with the top and bottom plane alternatingly, while the other set
is woven through the gaps between the first set of filaments [30]. The
middle layer configuration is built based on the ‘Parallel spacer’ [2,40]
and ‘Middle layer spacer’ [31] in the literature. The filaments are
arranged in a similar manner to those in the nonwoven configuration,
but one set of straight filaments has a smaller diameter and pierces
through the other set of filaments whose diameter is the same as the
channel height.
For each spacer configuration, the orientation of the spacers and
the intersecting angle between two layers of filaments are also varied. A Fig. 2. Schematic illustration of the filament arrangement and definitions of key angles:
schematic illustration of nonwoven spacers is shown in Fig. 2 where θa is the angle between the inlet flow direction and axial filament (attack angle), θt the
important angles that describe the spacer orientation and filament angle between the inlet flow and transverse filament and θf the angle between the axial
arrangement are defined. Spacers consisting of two distinct layers of and transverse filaments (mesh angle), which depends on θa and θt.

intersecting filaments are the most widely used design in RO modules

[42]. Filaments that are oriented along the inflow direction are called configurations introduced earlier, a total of 20 different spacer
axial filaments while those crossing the inflow are transverse filaments. geometries can be created (Table 1). Each spacer configuration has a
The acute angle between the axial and transverse filaments is defined as reference case (or base-case arrangement) where θa=180° and θt=90°.
the mesh angle, θf, which can be calculated from θa (angle between the
inflow direction and axial filament or attack angle) and θt (angle 2.2. Computational fluid dynamics (CFD) simulations
between the inflow and transverse filament), θf =(θa−θt) or
180−(θa−θt), as shown in Fig. 2. 2.2.1. Model description
Five combinations of θt and θa are selected in this study, as Owing to the repeating pattern in feed spacer geometry as shown in
described in Fig. 3. In order to isolate the effect of θa and θt, the base- Fig. 3, a single unit cell is modelled for each simulation case. Elemental
case arrangement shown in Fig. 3(a) is modified to form two groups by units for 20 geometries are created and representative models are
changing the orientation of the axial or transverse filaments. In the first shown in Fig. 4. It should be noted that the elemental unit model for
group, Figs. 3(b) and (c), the transverse filaments stay perpendicular to the fully woven spacer consists of two units in order to capture a
the inflow direction, while the axial filaments are rotated by −30° (or complete repeating pattern. As can be seen in Fig. 4(d), side openings
+150°) and −60° (or +120°), producing mesh angles of 60° and 30°, on the left and right side of the geometry are symmetrical only when
respectively. In the second group, the axial filaments are parallel to the two units are included.
inflow direction while the transverse filaments have angles of 60° or The feed stream flows through the elemental unit in a direction
30° with the axial filaments as shown in Figs. 3(d) and (e). Therefore, normal to the inlet and outlet faces, while natural flow is allowed
for a fixed mesh angle, the effects of spacer orientations can be studied through side openings. Permeation occurs through the upper and lower
by comparing the two groups. membranes in the z-direction. Elemental unit models for Cases 5–12,
By applying these five angular arrangements to the four spacer where inlet and outlet faces are perpendicular to the inflow, are built in

Fig. 1. Filament configurations of feed spacers: (a) nonwoven (b) partially woven (c) middle layer (d) fully woven spacers.

80
B. Gu et al. Journal of Membrane Science 527 (2017) 78–91

Fig. 3. Selected filament arrangements in this study: (a) reference case, and (b)–(e) different variations. Blue and green lines in the upper panels indicate the transverse and axial
filaments, respectively, while the corresponding elemental units are shown in the lower panel. All sides of the elemental unit have an equal length. (For interpretation of the references to
color in this figure legend, the reader is referred to the web version of this article.)

Table 1
20 simulation cases consisting of 4 space types and 5 angular combinations.

Case No. Name Configuration θt (deg) θa (deg) θf (deg)

1 NW90 Nonwoven 90 180 90 Reference cases

2 PW90 Partially woven 90 180 90
3 ML90 Middle layer 90 180 90
4 FW90 Fully woven 90 180 90
5 NW60T Nonwoven 90 150 60 Transverse filaments perpendicular to inflow
6 NW30T Nonwoven 90 120 30
7 PW60T Partially woven 90 150 60
8 PW30T Partially woven 90 120 30
9 ML60T Middle layer 90 150 60
10 ML30T Middle layer 90 120 30
11 FW60T Fully woven 90 150 60
12 FW30T Fully woven 90 120 30
13 NW60A Nonwoven 60 180 60 Axial filaments parallel to inflow
14 NW30A Nonwoven 30 180 30
15 PW60A Partially woven 60 180 60
16 PW30A Partially woven 30 180 30
17 ML60A Middle layer 60 180 60
18 ML30A Middle layer 30 180 30
19 FW60A Fully woven 60 180 60
20 FW30A Fully woven 30 180 30

a similar way to the reference cases. For Cases 13–20, however, the developed velocity profile and a flat concentration profile are imposed
transverse filaments are not perpendicular to the inlet flow. Therefore, at the inlet (artificial inlet for Cases 13–20). Since the actual inlet faces
an artificial section is created to obtain a developed flow velocity profile are tilted at an angle in Cases 13–20, average inlet flow velocities for
at the actual inlet, as shown in Fig. 5 for Case 19 (FW60A). the permeation units are different from those in Cases 1–12. This
allows for an investigation of the effects of different spacer orientations
2.2.2. Governing equations and boundary conditions in a membrane module. A constant pressure is prescribed at the outlet
For each elemental unit, the flow is assumed to be laminar, boundary. Spacer filaments are treated as no-slip walls with zero
incompressible, Newtonian and steady, and is governed by the velocity and zero solute flux. For the side openings through which
continuity and Navier-Stokes equations: fluid can move, periodic conditions for flow and solute transport are
imposed. All the walls in the artificial section for Cases 13–20 are
∇⋅u = 0 (1)
assumed to be no-slip with no flux conditions.
ρu⋅∇u = −∇p + μ∇2 u (2) In order to predict membrane performance, it is critical to have as
realistic a set of boundary conditions as possible, particularly with
where u is the fluid velocity vector, ρ the density of the fluid, μ the regard to water and solute fluxes for membrane walls. Since water and
viscosity of the fluid and p the pressure. The transport of a single solute solute fluxes vary with pressure and concentration on membrane
in the fluid domain can be described by the convection-diffusion surfaces, boundary conditions for membrane walls should accommo-
equation. date this and constant water and solute flux boundary conditions
∇⋅(Ds ∇u) − ∇⋅(uc ) = 0 (3) should be avoided. The most widely accepted description of water and
solute fluxes for RO membranes is the solution-diffusion model, which
where c is the solute concentration and Ds the diffusion coefficient. The can be expressed as [32,33].
fluid is assumed to have the properties of pure water due to the degree
of dilution of typical feed streams in RO processes. Jw = A (Δp − κΔc ) (4)
Each elemental unit model consists of 5 types of boundaries: the
inlet, outlet, side openings, filament walls and membrane walls. A fully Js = BΔc (5)

81
B. Gu et al. Journal of Membrane Science 527 (2017) 78–91

Fig. 4. Elemental units of the 4 reference cases. (a) nonwoven (NW90), (b) partially woven (PW90), (c) middle layer (ML90) and (d) fully woven (FW90). Lf is the distance between
filaments, Hc the channel height, Df the diameter of filaments in the nonwoven and woven configurations and Df,1 and Df,2 the diameter of transverse and axial filaments in the middle
layer configuration, respectively.

Jw (x, y) = A [{ pf (x, y) − pp } − κ {cf (x, y) − cp (x, y)}] (6)

Js (x, y) = B [cf (x, y) − cp (x, y)] (7)

Js (x, y)
cp (x, y) =
Jw (x, y) (8)
where subscripts f and p denote feed and permeate stream, respec-
tively, x and y are Cartesian coordinates as defined in Fig. 4. The
coordinate z is omitted here since the membrane walls are perpendi-
cular to the z-axis. The second assumption is described by Eq. (8) [35].
Since feed pressure and concentration are obtained from the solutions
of the governing equations in Eqs. (1) to (3) and permeate pressure is
assumed to be a constant, it is possible to derive an explicit expression
Fig. 5. Computational model for a fully woven geometry (FW60A) where the transverse
filaments are not perpendicular to the inlet flow direction. The artificial section contains a for permeate concentration by combining Eqs. (6) to (8).
channel with impermeable walls. The middle axial filament in the artificial section has a
κAcp2 + [A ( p − κc ) + B] cp − Bc = 0 (9)
length of Lf and the dimension of the artificial section is determined accordingly.
Subscripts for the feed channel are omitted from now on so as to
where Jw is the water flux, Js the solute flux, A the hydraulic have consistent notation with that used in Eqs. (1) to (3). The permeate
conductivity, κ the osmotic factor which is constant under isothermal pressure is assumed to be zero for convenience in this study although
conditions, B the solute permeability and Δ indicates that the difference its absolute value is usually slightly above the atmospheric pressure in a
between values of the relevant property (pressure or concentration) in practical operation. The root of Eq. (9) is given by
the feed and permeate streams is taken. It is assumed that the intrinsic
membrane properties, A and B, are constant. It is apparent that four −X1 + X12 + X2
cp =
key variables are required to calculate water and solute fluxes: feed 2κA (10)
pressure, feed concentration, permeate pressure and permeate con- where X1 = A ( p − κc ) + B , and X2 = 4κABc .
centration. This can be simplified by making appropriate assumptions The permeate concentration determined by the above equation can
for the permeate stream to avoid having to solve for permeate pressure then be used to obtain water and solute fluxes using Eqs. (6) and (7).
and concentration distributions. Two assumptions are made for this Under the assumption that water flux is normal to the membrane
purpose: (i) the pressure in the permeate channel is constant, and (ii) surface, boundary conditions for the upper and lower membrane walls
the solute concentration at the wall depends on local water and solute in the feed channel can be prescribed.
fluxes and is not influenced by cross flow [35].
ul, up = [0 0 Jw ] (11)
The first assumption can be justified for a model unit which is very
small compared to a real size RO module. The second assumption can ul, low = [0 0 − Jw ] (12)
also be justified owing to the relatively low permeate flowrate in RO
membrane modules (less than 10% of feed flowrate). Based on these where ul, up and ul, low are the leaking velocity at the upper and
assumptions, Eqs. (4)–(5) can be rewritten as follows: lower membrane walls, respectively. As expressed in Eqs. (11) and
(12), the only non-zero component is in the z direction, and the

82
B. Gu et al. Journal of Membrane Science 527 (2017) 78–91

negative sign in Eq. (12) indicates the direction of water flux leaving the first step, the membrane walls are treated as impermeable, i.e. no
the fluid domain. Solute fluxes through the membrane walls are water and solute fluxes. The obtained solutions are then used as an
prescribed as: initial guess for the second step where the aforementioned boundary
conditions are applied at the membrane walls. The MUltifrontal
−n⋅(Ds ∇c − uc ) = −Js. (13)
Massively Parallel Spacer direct solver (MUMPS) is used to solve the
The formulated boundary conditions for RO membrane walls make coupled momentum and mass transport problem, with a convergence
it possible to examine the effects of feed spacer types on membrane criterion of 10−6. All computations are performed on an Intel Xeon
performance accounting for locally varying water and solute permea- CPU E5-2665 workstation with 96 GB of RAM at 2.40 GHz. Each
tion, which is the most important characteristics of RO membrane simulation takes between 3 and 8 h depending on the size of the
channels. geometry.
Parameter values for boundary conditions and transport properties
are listed in Table 2 along with geometric parameters for spacer 2.2.4. Analysis of results
designs. Permeability parameters are chosen within practical ranges In addition to examining spatial distributions of velocity and solute
found in the literature [34]. The osmotic factor is based on the concentration, the performance of the membrane unit is evaluated in
assumption that temperature is 25 °C and the product of the osmotic terms of average water and solute fluxes through the upper and lower
coefficient and ionisation number is 2. Geometric parameters are also membrane walls. Other important parameters, such as pressure drop
found in the literature [2,12]. Inlet velocity, concentration and outlet and CP modulus, are also calculated to allow for direct comparisons
pressure are within RO operating ranges, and the chosen molar among different spacer configurations. Pressure drop is calculated as
concentration is equivalent to approximately 35 kg/m3 of aqueous the difference in average pressure between the inlet and outlet
sodium chloride solution. Based on the definition of Reynolds number pressures. Since the distance between the inlet and outlet pressures
used in [6,12], the channel Reynolds number corresponding to the in each elemental unit varies depending on the mesh angle, the
conditions adopted in the present study is 224, which is within the pressure drop per unit length is calculated by accounting for the
laminar flow regime. filament length (Lf) and axial filament angle (θa).
Δp
2.2.3. Computational details Δp =
L f cos(180 − θa ) (14)
All CFD simulations included in this study are performed by using
COMSOL Multiphysics 5.0, which employs the Galerkin finite element The CP modulus has been used to measure the degree of CP on the
method with linear shape functions and stabilisation methods based on membrane walls, and it is defined as
streamline diffusion and crosswind diffusion for the numerical solution cw, i
of the governing equations. Computational meshes consisting of Mi = (i = up , low )
cb (15)
tetrahedral elements are generated, with local mesh refinement around
sharp corners where spacer filaments are in direct contact with where Mi is the CP modulus for i membrane and the bulk concentra-
membrane walls, and in the vicinity of the membrane walls to allow tion cb is taken from the centroid of the fluid domain. Average CP
for adequate resolution of the mass transfer boundary layer. A mesh moduli at the upper and lower membrane walls and for the entire
sensitivity test is carried out by increasing the number of mesh elemental feed channel are also calculated for the purpose of compar-
elements until the difference in average water flux between two ison.
successive meshes is below 0.2%. The final mesh contains 1.4–2.9
million elements depending on the volume of the fluid domain. 3. Results and discussions
Since the boundary conditions at the membrane walls are fully
coupled with the governing equations as they are a function of pressure Obtained velocity and concentration distributions from the CFD
and concentration, all simulations are carried out in two steps in order simulations are presented first so that the effects of geometric
to achieve more robust convergence preferably within 25 iterations. In differences can be examined. In order to determine which spacer
design provides superior membrane performance, quantitative com-
Table 2 parisons are made of water and solute fluxes, as well as their spatial
List of parameters and boundary conditions. distributions on membrane walls. In addition, CP moduli and pressure
drops in different geometries are compared, which help to explain why
Symbol Parameter Value Source
membrane performances differ among different spacer types.
[unit]

A [m/s/Pa] Hydraulic conductivity 2.50×10–12 [34] 3.1. Velocity and concentration distributions within fluid domains
B [m/s] Salt permeability 2.50×10−8 [34]
κ [m3 Pa/ Osmotic factor 4958 –
mol]
The influence of spacer geometry on flow and mass transfer can be
Ds [m2/s] Diffusion coefficient 1×10−9 [41] appreciated by comparing velocity and concentration fields in feed
ρ [kg/m3] Density 998.20 – channels filled with different spacers. Fig. 6 shows velocity magnitude
μ [Pa s] Viscosity 8.93×10−4 – contours for the 4 reference cases which have a fixed filament angle and
Lf [m] Distance between filaments 4.5×10−3 [12]
orientation but different filament configurations. 2D velocity magni-
Hc [m] Channel height 1×10−3 [12]
Df [m] Diameter of filaments in 0.5×10−3 [12] tude contours are displayed at selected xz and yz planes, which are
nonwoven and woven parallel and perpendicular to the flow direction, respectively. A
spacers common feature is the presence of low-velocity zones near spacer
Df,1 [m] Diameter of transverse 0.5×10−3 [2] filaments in all cases although their sizes vary. Further examination of
filaments in middle layer
spacer
velocity vectors near the membrane wall reveals the presence of flow
Df,2 [m] Diameter of axial filaments 1×10−3 [2] recirculation. Fig. 7 shows velocity vectors in the xz plane in NW90 as a
in middle layer spacer representative case, where a recirculation zone is clearly formed behind
u0 [m/s] Average inlet velocity 0.1 Selected within the front filament. There is hardly any forced flow through the side
c0 [mol/ Inlet concentration 603.45 operating range
openings in NW90, PW90 and FW90 since no pressure gradient is
m3]
p0 [Pa] Outlet pressure 60×105 applied. Flow patterns in spacer geometries with a mesh angle of 90°
are symmetrical with respect to the central xz plane.

83
B. Gu et al. Journal of Membrane Science 527 (2017) 78–91

Fig. 6. Velocity magnitude contours for the reference cases at selected planes. (a) xz planes in NW90, (b) xz planes in PW90, (c) xz planes in ML90, (d) xz planes in FW90 (for y=[0, Lf]),
(e) yz planes in NW90, (f) yz planes in PW90, (g) yz planes in ML90 and (h) yz planes in FW90 (for y=[0, Lf]).

Fig. 8 shows the concentration contours at selected yz planes. Since

the cross-sections of NW90 and ML90 do not change along the flow
path, a slice at x=0.5Lf is chosen as representative. For PW90 and
FW90, slices are taken at three locations to observe variations of solute
concentration along the channels. By comparing Fig. 8 with Fig. 6, it
can be seen that areas of high concentrations correspond to stagnation
zones. In Fig. 8(a) for NW90, accumulation of the solute at the side
edges is observed with a maximum concentration of 1342 mol/m3
since the spacer filaments are in direct contact with the upper
membrane wall all the way through the feed channel. An even higher
maximum concentration, 1452 mol/m3, is observed in ML90, where
the thick axial filaments completely block the side openings. In PW90
and FW90, however, solutes are more uniformly distributed according
to the concentration contours at x=0.5Lf with a maximum concentra-
tion of 829 mol/m3 and 790 mol/m3 for PW90 and FW90, respectively,
as axial filaments are not in contact with any membrane walls. At
x=0.15Lf and 0.85Lf, axial filaments curve towards either the upper or
lower membrane wall so that the gaps between the filaments and
Fig. 7. Velocity vectors in the xz plane at y=0.5Lf for NW90. membrane walls become very small. This prevents the solute from
diffusing back to the bulk solution, leading to accumulation of solutes

Fig. 8. Concentration contours at selected yz planes for the reference cases. (a) yz plane at x=0.5Lf in NW90, (b) yz planes at x=0.15Lf, 0.5Lf and 0.85Lf in PW90, (c) yz plane at
x=0.5Lf in ML90 and (d) yz planes at x=0.15Lf, 0.5Lf and 0.85Lf in FW90 for y=[0, Lf].

84
B. Gu et al. Journal of Membrane Science 527 (2017) 78–91

in the narrow gaps. An interesting observation can be made for FW90 Simulation results for FW60A and FW30A are presented in Fig. 10
in Fig. 8(d) that there are distinct areas of CP in the middle between in the same manner as in Fig. 9. Concentration contours are displayed
axial filaments, where the cross flow velocity is particularly low as the at locations where the transverse and axial filaments overlay. There are
inflow stream is split into four by the front transverse filament as stagnation zones near the spacer filaments as found in other geome-
shown in Fig. 6(h). tries. Pronounced CP areas can be seen immediately behind the front
The concentration contours indicate that although CP occurs transverse filaments as shown in cross-sections (1) and (2) in
predominantly around sharp edges where spacer filaments meet Fig. 10(b). A similar trend of concentration build-up near the front
membrane walls, it can also occur in areas of low cross flow velocity transverse filament is also found in FW30A as displayed in Fig. 10(d),
due to the geometric feature of spacers. although the corresponding cross-sectional areas are much smaller
In order to examine the effects of mesh angle and attack angle on than those in FW60A. Other configurations exhibit similar behaviours
flow and mass transfer, four cases with the fully woven configuration due to the tilted transverse filaments as can be observed in the fully
(FW60T, FW30T, FW60A and FW30A) are selected for analysis since woven configurations, i.e., larger concentration build-up on the right-
they demonstrate the most complex behaviour. Fig. 9 presents velocity hand side (positive y-direction) of the single elemental unit than on the
and concentration contours at selected yz planes in FW60T and left-hand side.
FW30T. A similar velocity pattern to that in Fig. 6(h) is observed,
which appears to have four distinct regions of high velocity due to the 3.2. Water and solute fluxes through membrane walls
split inlet faces, although velocity contours in FW60T and FW30T are
skewed compared to those in FW90. As the flow proceeds towards the Due to the complex geometrical arrangements of feed spacers,
exit, the flow velocity becomes more uniform. As a result, the extent of water and solute fluxes through membrane walls are not uniformly
CP is reduced near the exit except in the narrow gaps between the distributed. Lower water flux and higher solute flux are expected at
filaments and membrane walls, as shown in Fig. 9(b) and (d) for locations of large CP. Spatial distributions of water and solute fluxes in
x=0.85Lf sinθf. Due to the axial filaments being tilted to the positive y- different spacer geometries are presented first and are discussed in
direction, stagnant zones are formed on the opposite side in the relation to the geometric features of each case. Average fluxes are
negative y-direction near the entrance of the channels as shown in calculated for the upper and lower membrane wall separately, as well as
Fig. 9(b) and (d). A similar behaviour is observed in other spacer for the total membrane area in a unit channel.
configurations, that is, stagnant zones tend to form on the left side
(negative y-direction) of a single elemental unit due to the tilted axial
filaments. Comparing the concentration contours in FW60T and 3.2.1. Distributions of local water and solute fluxes
Distributions of water flux through the upper and lower walls for
FW30T, the locations where high concentrations appear are similar
but their sizes are different; in the contours taken at x=0.15Lfsinθf and the reference cases (NW90, PW90, ML90 and FW90) are presented in
Fig. 11 where flow is in the x-direction, that is, from left to right. For all
0.5Lfsinθf CP areas near the central axial filament in FW30T are wider
than those in FW60T. In nonwoven, middle layer and partially woven four cases, water flux is relatively low in areas where spacer filaments
are in direct contact with membrane walls, implying a strong associa-
configurations, areas of high concentration near sharp edges are also
tion between CP and reduced water flux. In NW90, the transverse
larger in geometries with a mesh angle of 30° than 60°.
filaments are placed on the lower membrane causing CP to occur at the

Fig. 9. Velocity and concentration contours at selected yz planes for FW60T and FW30T. (a) Velocity contour in FW60T, (b) concentration contour in FW60T, (c) velocity contour in
FW30T and (d) concentration contour in FW30T.

85
B. Gu et al. Journal of Membrane Science 527 (2017) 78–91

Fig. 10. Velocity magnitude and concentration contours at selected yz planes for FW60A and FW30A. (a) velocity contour in FW60A, (b) concentration contour in FW60A, (c) velocity
contour in FW30A and (d) concentration contour in FW30A.

left and right edge as illustrated in the lower panel in Fig. 11(a). The shown in Fig. 11(d), distribution patterns are partially similar to those
front and rear transverse filaments in PW90, which are straight in PW90 in that there are isolated corners of low water flux on the
cylinders, are in contact with the lower and upper walls, respectively. upper and lower walls. Interestingly, there are bands of moderate water
As a result, low water flux mainly affects one side as seen in Fig. 11(b). flux (yellow coloured) in the middle of each unit channel in FW90, even
The curved axial filaments in PW90 create spots where CP is though there are no spacer filaments in the vicinity of the walls. These
exacerbated due to the narrow gap between the filaments and are regions where the velocity is very low due to the nature of the fully
membrane walls. Distributions of water fluxes through the upper and woven configuration as explained previously with Fig. 6(h).
lower walls in ML90 are almost identical since the thick axial filaments Average water fluxes obtained in NW90, PW90, ML90 and FW90
are in contact with both the upper and lower walls, while the transverse are 6.57×10−6 m/s, 6.69×10−6 m/s, 6.60×10−6 m/s and 6.97×10−6 m/
filaments do not touch any membrane walls. Comparing distributions s, respectively; with FW90 achieving the highest water flux and ML90
at the upper walls of NW90 and ML90, areas of low water flux are the lowest. When water fluxes through the upper and lower walls are
much larger in ML90 than in NW90, which is caused by the thicker considered separately, differences between the upper and lower
axial filaments in ML90 and the absence of side openings. For FW90 membranes are very small, with the largest discrepancy being observed

Fig. 11. Contours of water flux at the upper and lower membrane walls for the reference cases. (a) Jw, up (top) and Jw, low (bottom) in NW90 with a range between 1.42×10−7 –
7.55×10−6 m/s, (b) Jw, up (top) and Jw, low (bottom) in PW90 with a range between 1.79×10−7 – 7.55×10−6 m/s, (c) Jw, up (top) and Jw, low (bottom) in ML90 with a range between
7.20×10−8 – 7.55×10−6 m/s, (d) Jw, up (top) and Jw, low (bottom) in FW90 for y=[0,Lf] with a range between 1.41×10−7 – 7.55×10−6 m/s. (For interpretation of the references to color
in this figure legend, the reader is referred to the web version of this article.)

86
B. Gu et al. Journal of Membrane Science 527 (2017) 78–91

in NW90 where water fluxes through the upper and lower membranes differences between the nonwoven and middle layer configurations are
are 6.75×10−6 m/s and 6.41×10−6 m/s, respectively. amplified at small mesh angles. These trends are consistent with the
Fig. 12 shows distributions of solute fluxes through the upper and observations that velocity and concentration distributions on mem-
lower membranes in NW90, PW90, ML90 and FW90. Patterns in brane walls are less uniform in geometries with a smaller mesh angle.
Fig. 12 are almost identical to those in Fig. 11 except that the colour Comparisons of results for the upper and lower membranes show
code is reversed, meaning that high water fluxes coincide with low that the most notable differences are found in nonwoven configura-
solute fluxes, and vice versa. Average solute fluxes are 1.70×10−6 mol/ tions, with the upper membrane achieving about 5% higher water flux
(m2 s), 1.68×10−6 mol/(m2 s), 1.70×10−6 mol/(m2 s) and than the lower membrane in NW90. This is due to asymmetric mixing
1.62×10−6 mol/(m2 s) for NW90, PW90, ML90 and FW90, respec- patterns near the upper and lower membranes. On the other hand,
tively; with FW90 having the lowest solute flux and ML90 the highest. middle layer geometries behave symmetrically regardless of mesh
The effects of mesh angle and attack angle are also analysed by angles and attack angles. While partially woven geometries show small
using the fully woven geometries. Fig. 13 shows spatial distributions of differences between water fluxes through the upper and lower mem-
water fluxes through the upper and lower membranes in FW60T, branes, those in fully woven spacers are almost identical although
FW30T, FW60A and FW30A. By comparing Fig. 13(a) and (b) with the FW60T and FW30T show little difference between the upper and lower
reference case shown in Fig. 11(d), it is clear that areas of low water membranes.
fluxes appear more irregularly and their sizes are larger as the mesh For each spacer type, the highest average water flux is achieved with
angle becomes smaller. Isolated areas of low water flux near the middle a mesh angle of 60° and the axial filaments aligned parallel to the
axial filament also tend to shift towards the right-hand side from the inflow (case names ending with 60 A). However, fixed geometric
centre axial filament (negative y-direction) as the mesh angle de- parameters such as the filament distance and diameter are used here
creases. On the other hand, the low water flux spots in FW60A and for all configurations and therefore the results might not be the same
FW30A are slightly skewed to the left-hand side from the centre axial for a different set of geometric parameters.
filament (positive y-direction). Visual comparisons between different Results for average solute fluxes show the opposite trend to those
mesh angles reveal that a larger fraction of the total membrane area is for water fluxes in that spacer geometries that achieve higher water
exposed to low water flux at a smaller mesh angle. Other configurations fluxes have lower solute fluxes (Fig. 15). Therefore comparing solute
also react similarly to different angles; different mesh angle and attack fluxes for all the spacer geometries also leads to the same conclusion
angles lead to asymmetric contours compared to those presented in that NW60A, PW60A, ML60A and FW60A are the best performers for
Figs. 11 and 12. each spacer type in terms of the least solute flux produced.
Although water and solute fluxes are direct indicators of membrane
performance, it is also useful to evaluate other parameters related to
3.2.2. Comparison of average fluxes
transport phenomena near membrane walls, such as the CP modulus
Figs. 14 and 15 depict average water and solute fluxes for all the
which is presented in the next section.
spacer geometries included in this study. While the upper and lower
membrane fluxes are presented separately, the average values for the
total area of membrane in a unit channel are also given in order to 3.3. Concentration polarisation (CP) modulus
compare the performance of different spacer designs.
For the 20 spacer geometries examined here, average water fluxes As defined in Eq. (15), CP modulus measures the extent of
in a unit channel range between 6.1–7.3×10−6 m/s. It is clear that fully accumulating solutes on membrane surfaces relative to the bulk
woven spacer geometries achieve higher water fluxes than other concentration. The spatial distribution of CP moduli is similar to that
configurations; this is followed by partially woven, nonwoven and of solute fluxes shown in Fig. 12, with regions of high solute fluxes near
middle layer configurations. When transverse filaments are perpendi- the edges of membranes corresponding to high CP moduli. This is in
cular to the inflow direction (case names ending with T), reducing the keeping with the solution-diffusion model where solute flux is propor-
mesh angle seems to lower water fluxes although there is little tional to the concentration difference between feed and permeate. The
difference between FW60T and FW30T. An interesting finding is that average and maximum CP moduli for each of the 20 spacer geometries

Fig. 12. Contours of solute flux at the upper and lower membrane walls for the reference cases. (a) Js, up (top) and Js, low (bottom) in NW90 with a range between 1.50×10−5 –
3.00×10−5 mol/(m2 s), (b) Js, up (top) and Js, low (bottom) in PW90 with a range between 1.50×10−5 – 2.99×10−5 mol/(m2 s), (c) Js, up (top) and Js, low (bottom) in ML90 with a
range between 1.50×10−5 – 3.01×10−5 mol/(m2 s), (d) Js, up (top) and Js, low (bottom) in FW90 for y=[0, Lf] with a range between 1.50×10−5 and 3.0×10−5 mol/(m2 s).

87
B. Gu et al. Journal of Membrane Science 527 (2017) 78–91

Fig. 13. Contours of water flux at the upper and lower membrane walls for fully woven configurations. (a) Jw, up (left) and Jw, low (right) in FW60T with a range between 1.38×10−7 –
7.55×10−6, (b) Jw, up (left) and Jw, low (right) in FW30T with a range between 1.08×10−7 – 7.55×10−6, (c) Jw, up (left) and Jw, low (right) in FW60A with a range between 1.39×10−7
– 7.54×10−6, (d) Jw, up (left) and Jw, low (right) in FW30A with a range between 1.18×10−7 – 7.54×10−6.

are given in Fig. 16 which shows that FW60A achieves the lowest The average CP moduli obtained in this study are in the range of
average CP modulus at 1.04 while ML30A shows the most severe CP 1.04–1.2, which are slightly lower than those in the aforementioned
with an average CP modulus of 1.22. Although fully woven spacers studies [14,21]. This can be attributed to differences in boundary
generally have low average CP moduli, the lowest maximum CP conditions and spacer geometries. The assumption of a constant solute
modulus is found with PW90 at 2.26. On the other hand, middle layer rejection by the membrane, as was done in the aforementioned study
geometries with a mesh angle of 30° show much greater maximum CP [14,21] and in others [6,12,15,18,26,30,31], may need to be revisited
modulus, as a result of solute accumulation in the larger stagnant zone since membrane rejection is not an intrinsic property of membranes
than those in other geometries. but an output of an RO process. It would be desirable if this could be
CP moduli obtained from other CFD studies in the literature are predicted as part of a CFD simulation. For this purpose, more realistic
compared with those from this study. Lyster and Cohen investigated CP boundary conditions are implemented in the present study by employ-
in an empty rectangular RO cell with a channel height of 1 mm ing the solution-diffusion model for water and solute transport through
experimentally and numerically using CFD [21]. By employing the RO membranes without assuming a constant rejection rate. Combined
solution-diffusion model for water flux at the membrane wall, together with 3D geometric representations of spacer-filled channels, the CFD
with an imposed concentration gradient for solute flux, they found that models presented here are expected to offer more faithful predictions of
the average CP modulus varied from 1.33 to 1.42 for 80–100% membrane performance than previous models.
membrane rejection. Song and Ma studied the impact of spacer
geometry on CP and permeate flux using 2D CFD simulations [14].
3.4. Comparison of pressure drop
They assumed that the solute was completely rejected by the mem-
brane while water could pass through, with its flux being determined by
Pressure drop is an important parameter since it is directly related
the solution-diffusion model. They found that the average CP modulus
to energy consumption in an RO process in which hydraulic pressure is
achieved in an empty channel was approximately 2.5 while those in
the main driving force for water permeation. Fig. 17 shows pressure
spacer-filled channels were 2.10, 2.33 and 1.73 for filament thickness
drop and pressure drop per unit length in different spacer geometries.
of 0.5 mm, 0.25 mm, and 0.75 mm, respectively. For a spacer-filled
The pressure drop per unit length is included due to the different
channel with a filament thickness of 0.5 mm, which is the spacer size
length of spacer geometries. It is clear that ML30T has the highest
used in the present study, CP would be mitigated by a factor of 0.84
pressure drop among all cases while nonwoven geometries have
according to the study by Song and Ma [14]. By applying the mitigation
relatively low pressure drops compared to other spacer types for a
factor to the results of Lyster and Cohen [21], the average CP modulus
given attack angle and mesh angle. The reference cases where the mesh
in a spacer-filled channel would be approximately 1.2.
angle is 90° result in the lowest pressure drop. The effect of mesh angle

Fig. 14. Comparison of average water fluxes in a unit channel for different spacer geometries. Colour bars display the upper and lower membrane fluxes separately, while stars show the
corresponding average water flux through the total membrane area of each unit channel.

88
B. Gu et al. Journal of Membrane Science 527 (2017) 78–91

Fig. 15. Comparison of average solute fluxes in a unit channel for different spacer geometries. Colour bars display the upper and lower membrane fluxes separately, while stars show the
average solute flux through the total membrane area of each unit channel.

Fig. 16. Comparison of CP moduli for all cases. Average CP moduli in a unit channel are presented in bars (Mave on the left vertical axis) and the maximum CP moduli are indicated by
stars (Mmax on the right vertical axis).

is more notable in cases where transverse filaments are perpendicular end with T and A might not be appropriate since pressure drop is
to the inflow while axial filaments are at an angle with inflow (case strongly influenced by the flow rate in a unit feed channel. Nonetheless,
names ending with T). Results in Fig. 17 show that smaller mesh angles it is clearly demonstrated that spacer type and mesh angle have a
lead to higher pressure drops, due to disturbances caused by the strong impact on pressure drop in a spacer-filled channel.
obstruction of axial filaments in the flow path. It is also observed that
the difference in pressure drop between NW60T and NW30T is much 3.5. Limitations of the current study
larger than that between NW60A and NW30A, implying that pressure
drop is more sensitive to the alignment of axial filaments than that of The current CFD model involves a number of assumptions. Firstly,
transverse filaments. a single unit cell is modelled with periodic boundary conditions, which
On the other hand, for the cases where axial filaments are aligned ignores the development of fluid flow and mass transfer from the
parallel to the inflow direction (case names ending with A), spacers inlet along the channel. It has been reported that fluid flow becomes
with a mesh angle of 60° generate higher pressure drops than those fully developed after 4–5 cells in the presence of spacers [43].
with 30°. This is due to different average inlet velocities at the actual Secondly, the flow is assumed to be laminar and steady. Although this
inlet faces, which are calculated based on the imposed average velocity assumption is justifiable for the relatively low Reynolds number
at the artificial inlet. The average velocity at the actual inlet in FW60A examined here, flow in the narrow spacer-filled channel can become
is 0.19 m/s, which is higher than that in FW30A (0.14 m/s). For the unsteady and transitional at high inlet velocities [10,44,45]. Thirdly,
same reason, cross-comparisons between the two groups whose names physical properties of the fluid are assumed to be constant and the

Fig. 17. Comparison of pressure drop (Pdrop on the left vertical axis) and pressure drop per unit length (Pdrop on the right vertical axis) along a unit spacer-filled channel for all cases.

89
B. Gu et al. Journal of Membrane Science 527 (2017) 78–91

adopted diffusion coefficient corresponds to a relatively low tempera- spacers, J. Membr. Sci. 399–400 (2012) 60–72.
[10] G.A. Fimbres-Weis, D.E. Wiley, Review of 3D CFD modelling of flow and mass
ture range. Finally, the number of parameters assessed is limited. It transfer in narrow spacer-filled channels in membrane modules, Chem. Eng.
would be important to determine the effects of key parameters on Process. 49 (2010) 759–781.
membrane performance, especially the effect of inlet flow velocity on [11] R. Salcedo-Díaz, P. García-Algado, M. García-Rodríguez, J. Fernández-Semprere,
Visualization and modelling of the polarization layer in crossflow reverse osmosis in
concentration polarisation in future studies. a slit-type channel, J. Membr. Sci. 456 (2014) 21–30.
[12] A. Subramani, S. Kim, E.M.V. Hoek, Pressure, flow, and concentration profiles in
4. Conclusions open spacer-filled membrane channels, J. Membr. Sci. 277 (2006) 7–17.
[13] A.L. Ahmad, K.K. Lau, Impact of different spacer filaments geometries on 2D
unsteady hydrodynamics and concentration polarization in spiral wound mem-
This study presents the analysis of various feed spacer geometries brane channel, J. Membr. Sci. 286 (2006) 77–92.
where three geometric features are varied: filament configuration, [14] L. Song, S. Ma, Numerical studies of the impact of spacer geometry on concen-
tration polarization in spiral wound membrane modules, Ind. Eng. Chem. Res. 44
mesh angle and attack angle. 3D CFD simulations of 20 spacer
(2005) 7638–7645.
geometries are performed in order to investigate the effects of spacer [15] S. Wardeh, H.P. Morvan, CFD simulations of flow and concentration polarization in
geometry on membrane performance (water and solute fluxes), CP and spacer-filled channels for application to water desalination, Chem. Eng. Res. Des.
pressure drop. A unique feature of the CFD model used in this study is 86 (2008) 1107–1116.
[16] A.L. Ahmad, K.K. Lau, M.Z. Abu Bakar, Impact of different spacer filament
the incorporation of realistic boundary conditions for the RO mem- geometries on concentration polarization control in narrow membrane channel, J.
brane using the solution-diffusion model, allowing for predictions of Membr. Sci. 262 (2005) 138–152.
membrane performance. Based on the numerical results for the [17] R. Ghidossi, D. Veyret, P. Moulin, Computational fluid dynamics applied to
membranes: state of the art and opportunities, Chem. Eng. Process. 45 (2006)
operating condition and spacer geometries studied, the following 437–454.
conclusions can be drawn: [18] G. Guillen, E.M.V. Hoek, Modeling the impacts of feed spacer geometry on reverse
osmosis and nanofiltration processes, Chem. Eng. J. 149 (2009) 221–231.
[19] J. Schwinge, D.E. Wiley, D.F. Fletcher, A CFD study of unsteady flow in narrow
1. Fully woven geometries outperform other spacer configurations in spacer-filled channels for spiral-wound membrane modules, Desalination 146
terms of higher water flux and lower CP, although the associated (2002) 195–201.
pressure drops are slightly higher than their nonwoven counterparts. [20] V.M. Geraldes, V.A. Semião, M.N. de Pinho, Flow management in nanofiltration
spiral wound modules with ladder-type spacers, J. Membr. Sci. 203 (2002) 87–102.
2. Reducing mesh angle leads to reduced water flux and larger CP.
[21] E. Lyster, Y. Cohen, Numerical study of concentration polarization in a rectangular
3. The pressure drop is very sensitive to the angle between axial reverse osmosis membrane channel: permeate flux variation and hydrodynamic
filaments and the direction of inflow (attack angle). end effects, J. Membr. Sci. 303 (2007) 140–153.
[22] V.V. Ranade, A. Kumar, Fluid dynamics of spacer filled rectangular and curvilinear
channels, J. Membr. Sci. 271 (2006) 1–15.
It is hoped that the CFD model presented here can be used to [23] F. Li, W. Meindersma, A.B. de Haan, T. Reith, Optimization of commercial net
predict membrane performance for a given spacer type and geometry. spacers in spiral wound membrane modules, J. Membr. Sci. 208 (2002) 289–302.
Results on CP can also help to assess the likelihood of membrane [24] Y.L. Li, K.L. Tung, M.Y. Lu, S.H. Huang, Mitigating the curvature effect of the
spacer-filled channel in a spiral-wound membrane module, J. Membr. Sci. 329
fouling since areas with concentration build-up are vulnerable to the (2009) 106–118.
deposition of large particles. In this regard, fully woven configurations [25] A.S. Saeed, R. Vuthaluru, Y. Yang, H.B. Vuthaluru, Effect of feed spacer arrange-
would be the preferred choice in situations when energy consumption ment on flow dynamics through spacer filled membranes, Desalination 285 (2012)
163–169.
is not a major limiting factor. Although a mesh angle of 60° is likely to [26] K.K. Lau, M.Z. Abu Bakar, A.L. Ahmad, T. Murugesan, Feed spacer mesh angle: 3D
produce the highest water flux among the angles studied, it should be modelling, simulation and optimization based on unsteady hydrodynamic in spiral
noted that all simulations are undertaken at a fixed operating condition wound membrane channel, J. Membr. Sci. 343 (2009) 16–33.
[27] Y.L. Li, K.L. Tung, CFD simulation of fluid flow through spacer-filled membrane
and model parameters. Results could vary depending on parameters module: selecting suitable cell types for periodic boundary conditions, Desalination
specified in simulations. 233 (2008) 351–358.
[28] S.K. Karode, A. Kumar, Flow visualization through spacer filled channels by
computational fluid dynamics I. Pressure drop and shear rate calculations for flat
Acknowledgement sheet geometry, J. Membr. Sci. 193 (2001) 69–84.
[29] C. Picioreanu, J.S. Vrouwenvelder, M.C.M. van Loosdrecht, Three-dimensional
The authors would like to acknowledge the funding and technical modelling of biofouling and fluid dynamics in feed spacer channels of membrane
devices, J. Membr. Sci. 345 (2009) 340–354.
support from BP through the BP International Centre for Advanced
[30] A.I. Radu, L. Bergwerff, M.C.M. van Loosdrecht, C. Picioreanu, A two-dimensional
Materials (BP-ICAM) which made this research possible. mechanistic model for scaling in spiral wound membrane systems, Chem. Eng. J.
241 (2014) 77–91.
References [31] A.I. Radu, J.S. Vrouwenvelder, M.C.M. van Loosdrecht, C. Picioreanu, Modeling the
effect of biofilm formation on reverse osmosis performance: flux, feed channel
pressure drop and solute passage, J. Membr. Sci. 365 (2010) 1–15.
[1] C.P. Koutsou, S.G. Yiantsios, A.J. Karabelas, A numerical and experimental study of [32] J.G. Wijmans, R.W. Baker, The solution-diffusion model: a review, J. Membr. Sci.
mass transfer in spacer-filled channels: effects of spacer geometrical characteristics 107 (1995) 1–21.
and Schmidt number, J. Membr. Sci. 326 (2009) 234–251. [33] M. Soltanieh, W.N. Gill, Review of reverse–osmosis membranes and transport
[2] M. Shakaib, S.M.F. Hasani, M. Mahmood, CFD modelling for flow and mass models, Chem. Eng. Commun. 12 (1981) 279–363.
transfer in spacer-obstructed membrane feed channels, J. Membr. Sci. 326 (2009) [34] G. Patroklou, K.M. Sassi, I.M. Mujtaba, Simulation of boron rejection by seawater
270–284. reverse osmosis desalination, Chem. Eng. Trans. 32 (2013) 1873–1878.
[3] M.F.A. Goosen, S.S. Sablani, S.S. Al-Maskari, R.H. Al-Belushi, M. Wilf, Effect of [35] K. Jamal, M.A. Khan, M. Kamil, Mathematical modelling of reverse osmosis
feed temperature on permeate flux and mass transfer coefficient in spiral-wound systems, Desalination 160 (2004) 29–42.
reverse osmosis systems, Desalination 144 (2002) 367–372. [36] K.P. Lee, T.C. Arnot, D. Mattia, A review of reverse osmosis membrane materials
[4] M. Khayet, J.I. Mengual, Effect of salt type on mass transfer in reverse osmosis thin for desalination – development to date and future potential, J. Membr. Sci. 370
film composite membranes, Desalination 168 (2004) 383–390. (2011) 1–22.
[5] J.L.C. Santos, V. Geraldes, S. Velizarov, J.G. Crespo, Investigation of flow patterns [37] J. Schwinge, P.R. Neal, D.E. Wiley, D.F. Fletcher, A.G. Fane, Spiral wound modules
and mass transfer in membrane module channels filled with flow-aligned spacers and spacers: review and analysis, J. Membr. Sci. 242 (2004) 129–153.
using computational fluid dynamics (CFD), J. Membr. Sci. 305 (2007) 103–117. [38] A. Saeed, R. Vuthaluru, H.B. Vuthaluru, Impact of feed spacer filament spacing on
[6] K. Darcovich, M.M. Dal-Cin, B. Gros, Membrane mass transport modelling with the mass transport and fouling propensities of RO membrane surfaces, Chem. Eng.
periodic boundary condition, Comput. Chem. Eng. 33 (2009) 213–224. Commun. 202 (2015) 634–646.
[7] G.A. Fimbres-Weihs, D.E. Wiley, Numerical study of mass transfer in three- [39] A. Saeed, R. Vuthaluru, H.B. Vuthaluru, Investigations into the effects of mass
dimensional spacer-filled narrow channels with steady flow, J. Membr. Sci. 306 transport and flow dynamics of spacer filled membrane modules using CFD, Chem.
(2007) 228–243. Eng. Res. Des. 93 (2015) 79–99.
[8] G.A. Fimbres-Weihs, D.E. Wiley, Numerical study of two-dimensional multi-layer [40] M. Shakaib, S.M.F. Hasani, M. Mahmood, Study on the effects of spacer geometry
spacer designs for minimum drag and maximum mass transfer, J. Membr. Sci. 325 in membrane feed channels using three-dimensional computational flow modeling,
(2008) 809–822. J. Membr. Sci. 297 (2007) 74–89.
[9] C.P. Koutsou, A.J. Karabelas, Shear stresses and mass transfer at the base of a [41] M. Taniguchi, S. Kimura, Estimation of transport parameters of RO membranes for
stirred filtration cell and corresponding conditions in narrow channels with seawater desalination, AIChE J. 46 (2000) 1967–1973.

90
B. Gu et al. Journal of Membrane Science 527 (2017) 78–91

[42] J. Johnson, Engineering aspects of reverse osmosis module design, Desalin, Water in narrow zigzag spacer-filled channels: a numerical study, Ind. Eng. Chem. Res. 45
Treat. 15 (2010) 236–248. (2006) 6594–6603.
[43] A. Saeed, Effect of feed channel spacer geometry on hydrodynamics and mass [45] J. Schwinge, D. Wiley, D.F. Fletcher, Simulation of the flow around spacer filaments
transport in membrane modules (Ph.D. Thesis in Chemical Engineering), Curtin between channel walls, unsteady flows, Ind. Eng. Chem. Res. 42 (2003)
University, Perth, Australia, 2012. 4962–4977.
[44] G.A. Fimbres-Weihs, D.E. Wiley, D.F. Fletcher, Unsteady flows with mass transfer

91