Concentration Polarization in Spacer Filled Reverse Osmosis Membrane Systems Ma Shengwei

CONCENTRATION POLARIZATION IN SPACERFILLED REVERSE OSMOSIS MEMBRANE SYSTEMS
MA SHENGWEI
(B. Sc., Nanjing Inst. of Meteorology
M. Sc., Chinese Academy of Sciences)
A THESIS SUBMITTED
FOR THE DEGREE OF DOCTOR OF PHILOSOPHY
DEPARTMENT OF CIVIL ENGINEERING
NATIONAL UNIVERSITY OF SINGAPORE
2005
ACKNOWLEDGEMENTS
I wish to express my appreciation and gratitude to my supervisor, Associate

Professor Song Lianfa for his continuous, enthusiastic and invaluable supervision and
encouragement throughout the entire course of this project.
I also wish to express my gratitude to National University of Singapore for

providing all computing resources at Supercomputing & Visualization Unit for this
project.
I am deeply indebted to my parents, my wife and my daughter for their

continuing support and encouragement in the years of research in NUS.
TABLE OF CONTENTS
ACKNOWLEDGEMENTS............................................................................................i
TABLE OF CONTENTS ..............................................................................................ii
SUMMARY...................................................................................................................v
LIST OF TABLE.........................................................................................................vii
LIST OF FIGURES ................................................................................................... viii
LIST OF SYMBOLS...................................................................................................xv
CHAPTER 1 INTRODUCTION..................................................................................1
1.1 Introduction .................................................................................................1
1.2 Aim of the research .....................................................................................3
1.3 Overview of the dissertation........................................................................3
CHAPTER 2 LITERATURE REVIEW.......................................................................4
2.1 Concentration polarization in RO systems.................................................4
2.2 Analytical models for concentration polarization in RO membrane
systems.......................................................................................................6
2.3 Numerical models for concentration polarization in RO membrane
systems.......................................................................................................8
2.4 The impact of spacer on concentration polarization and RO
membrane performance ...........................................................................14
2.5 Finite element method for coupled momentum transfer and solute
transport problems ...................................................................................17
2.6 Summary ...................................................................................................19
CHAPTER 3 NUMERICAL MODEL .......................................................................21
3.1 Introduction ..............................................................................................21
ii
3.2 Model development..................................................................................22

3.2.1 Governing equations......................................................................22
3.2.2 Penalty formulation for Navier-Stokes equations .........................24
3.2.3 Initial and boundary conditions ......................................................25
3. 2.4 SUPG finite element formulation and numerical strategies .........26
3.3 Model validation.......................................................................................33
3.4 Effect of meshing scheme on accuracy .....................................................36
3.5 Summary ...................................................................................................40
CHAPTER 4 CONCENTRATION POLARIZATION IN SPACER-FILLED
CHANNELS ......................................................................................................42
4.1 Introduction ...............................................................................................42
4.2 Velocity profiles in spacer-filled channels................................................45
4.3 Major mechanisms of concentration polarization in spacer filled
channels ...................................................................................................52
4.4 Summary ...................................................................................................65
CHAPTER 5
IMPACT OF FILAMENT GEOMETRY ON
CONCENTRATION POLARIZATION ...........................................................68
5.1 Introduction ...............................................................................................68
5.2 Filament shape...........................................................................................69
5.3 Filament thickness.....................................................................................77
5.4 Summary ...................................................................................................96
CHAPTER 6 FILAMENT CONFIGURATION AND MESH LENGTH ON
CONCENTRATION
POLARIZATION
AND
MEMBRANE
PERFORMANCE ..............................................................................................98
6.1 Introduction ...............................................................................................98
6.2
Concentration polarization patterns for different filament

configurations ........................................................................................101
6.3 Filament configuration on membrane performance ................................111
iii
6.4 Impact of mesh length on membrane performance .................................115

6.4.1 Mesh length on permeate flux ......................................................115
6.4.2 Impact of mesh length on pressure loss........................................123
6.5 Summary .................................................................................................127
CHAPTER 7 CONCLUSIONS AND RECOMMENDATIONS.............................129
7.1 Conclusions .............................................................................................129
7.2 Recommendations for further research ...................................................131
REFERENCES ..........................................................................................................133
APPENDIX A : LIST OF
PUBLICATIONS AND CONFERENCE
PRESENTATIONS..........................................................................................150
iv
SUMMARY
As a phenomenon inherently associated with membrane separations,
concentration polarization has long been identified as a major problem that
deteriorates the performance of RO systems. However, this phenomenon is still not
well understood especially in practical spiral wound modules, where spacer is an
essential part to form the feed channel. The purpose of this study was to study
concentration polarization in spacer filled RO systems and to quantify the impact of
feed spacer on concentration polarization and membrane performance. In this study,
a fully coupled 2-D streamline upwind Petrov/Galerkin (SUPG) finite element model
was developed so that it becomes possible to simultaneously simulate hydrodynamic
conditions, including permeate velocity at membrane surface, and salt concentration
profiles, including wall concentrations in RO membrane channels. The numerical
model was compared with the available experimental data of RO systems in the
literature. With this numerical model, the role of feed spacer on concentration
polarization and system performance can be quantitatively investigated in realistic
conditions.
It was found that concentration polarization in spacer filled membrane channel

was affected by two major mechanisms: concentration boundary layer disruption due
to flow separation and, concentration boundary layer disruption due to the constricted
flow passage. The two mechanisms may work separately or jointly dependent on
spacer configurations. Filament geometry was found to have significant impact on
concentration polarization although it would not change the overall concentration
polarization patterns. Extremely high wall concentrations were found close to the
contact point of membranes with cylindrical filaments. Increasing filament thickness
v
could significantly alleviate concentration polarization at cost of elevated pressure

loss.
It was also found that membrane performance was strongly affected by

filament configurations and mesh length. In most cases, zigzag configuration
provided the best permeate flux enhancement while submerged configuration resulted
in the lowest peak wall concentrations. There was an optimum mesh length with
cavity and zigzag configurations for maximizing permeate flux enhancement
Decreasing mesh length may lead to significant increase of pressure loss especially
for zigzag configurations, and may lead to permeate flux decline in certain cases.
The results suggest that the commonly used overall parameter of spacer (e.g.,
voidage) is inadequate or inappropriate to characterize spacers in a RO system. The
results also imply that a universally optimized spacer design does not exist and
optimization of the spacers has to be carried out particularly for different situations.
Through this study, the understandings of concentration polarization in

spacer-filled RO channels and the effects of spacer on concentration polarization and
system performance have been significantly advanced.
The numerical model
developed in this study can provide a powerful tool to realistically study

concentration polarization in spiral wound RO modules.
The quantitative
visualization and assessment of the impact of spacer on system performance would

provide the technical foundation for the optimum design of RO membrane systems.
vi
LIST OF TABLE
Table 6.1 Membrane properties and operating conditions ........................................100
vii
LIST OF FIGURES
Figure 3.1 Flowchart of the solver and model organization......................................32
Figure 3.2 Comparison of numerical simulation results with experimental data.......34
Figure 3.3 Simulated wall concentration at different values of membrane
permeability (crossflow velocity: 200cm/s; other conditions: Merten et al
(1964))................................................................................................................34
Figure 3.4 Comparing the accuracy of simulation results due to different
meshing schemes. The solution became independent of the meshing
schemes when the non-uniform (exponential) elements increased to 60
and more in the channel height direction. (crossflow velocity: 20cm/s;
membrane permeability: 8.2910-6g/cm2 sec atm) ............................................38
Figure 3.5 A successful mesh scheme (part) to capture the flow direction
transition near membrane surface ......................................................................39
Figure 3.6 Velocity field (part) in the flow direction transition region in an
empty channel (simulation conditions: 2 membranes at y=0 and y=H;
p=429psi; A=510-12m/s Pa; c0=9800mg/l; u0=0.01m/s; H=1mm;
L=5cm) ...............................................................................................................39
Figure 4.1 Illustration of the spacer configuration .....................................................44
Figure 4.2 Illustration of the mesh adjacent to a cylindrical filament .........................44
Figure 4.3 Contour of flow velocity in a feed channel (part) with 0.5mm (in
diameter) cylinder transverse filaments (simulation conditions:
p=800psi; c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s; h=1mm;
lf=4.5mm) ...........................................................................................................46
Figure 4.4 Contour of flow velocity in a feed channel (part) with 0.75mm (in
lf=4.5mm) ...........................................................................................................47
Figure 4.5 Contour of x-component flow velocity in a feed channel (part) with
0.5mm (in diameter) cylinder transverse filaments (simulation conditions:
viii

lf=4.5mm) ...........................................................................................................49
Figure 4.6 Contour of x-component flow velocity in a feed channel with
0.75mm (in diameter) cylinder transverse filaments (simulation
conditions: p=800psi; c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s;
h=1mm; lf=4.5mm).............................................................................................50
Figure 4.7 Velocity field near the reattachment point in a feed channel with
lf=4.5mm) ...........................................................................................................51
Figure 4.8 Salt concentration (c/c0) profiles in an empty feed channel,
disproportional in height and length (simulation conditions: p=800psi;
c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s; h=1mm) .................................54
Figure 4.9 Salt concentration (c/c0) profiles in a feed channel with 0.5mm (in
lf=4.5mm) ...........................................................................................................55
Figure 4.10 Local variation of wall concentration (cw/c0) in an empty channel
and a feed channel with 0.5mm (in diameter) cylindrical transverse
filaments (simulation conditions: p=800psi; c0=32,000mg/l; A=7.31012
m/s Pa; u0=0.1m/s; h=1mm; lf=4.5mm) ..........................................................56
Figure 4.11 Local variations of permeate flux in an empty channel and a feed
channel with 0.5mm (in diameter) cylindrical transverse filaments
(simulation conditions: p=800psi; c0=32,000mg/l; A=7.310-12m/s Pa;
u0=0.1m/s; h=1mm; lf=4.5mm) ..........................................................................57
Figure 4.12 Enlarged view of local wall concentration (cw/c0) profiles (on the
membrane attached to the transverse filaments) in feed channels with
0.5mm (in diameter) cylindrical filaments (simulation conditions:
lf=4.5mm) ...........................................................................................................59
Figure 4.13 Velocity field near the small recirculation regions in a feed channel
with 0.5mm (in diameter) cylinder transverse filaments (simulation
h=1mm; lf=4.5mm).............................................................................................62
ix
Figure 4.14 Velocity field in the upstream of the first filament in a feed channel
with 0.5mm (in diameter) cylinder transverse filaments (simulation
h=1mm; lf=4.5mm..............................................................................................62
Figure 4.15 Local wall concentration (cw/c0) profiles (on the membrane
attached to the transverse filaments) in feed channels with 0.5mm (in
diameter) cylindrical filaments (simulation conditions: p=800psi;
c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s; h=1mm; lf=1.5mm) ................63
lf=1.5mm) ...........................................................................................................64
Figure 4.17 Velocity field in the upstream region of a filament in a feed
channel with 0.5mm (in diameter) cylinder transverse filaments
u0=0.1m/s; h=1mm; lf=1.5mm) ..........................................................................65
Figure 5.1 Illustration of the computing domain........................................................69
Figure 5.2 Salt concentration (c/c0) profiles in a feed channel with 0.5
(thickness)0.392 (width)mm rectangular bar transverse filaments
u0=0.1m/s; h=1mm; lf=4.5mm) ..........................................................................72
Figure 5.3 Contour of flow velocity in a feed channel (part) with 0.5
u0=0.1m/s; h=1mm; lf=4.5mm) ..........................................................................73
Figure 5.4 Wall concentration (cw/c0) profiles in an empty channel and a feed
channel with 0.5 (thickness)0.392 (width)mm
rectangular bar
transverse filaments (simulation conditions: p=800psi; c0=32,000mg/l;
A=7.310-12m/s Pa; u0=0.1m/s; h=1mm; lf=4.5mm)..........................................74
Figure 5.5 Comparison of local wall concentration (cw/c0) profiles (on the
0.5mm (in diameter) cylindrical filaments and 0.392mm0.5mm
(thickness) rectangular bar filaments (simulation conditions: p=800psi;
membrane opposite to the transverse filaments) in channels with 0.5mm
(in diameter) cylindrical filaments and 0.392mm0.5mm (height)
rectangular bar filaments (simulation conditions: p=800psi;
0.50.5mm square bar filaments and 0.392mm0.5mm (thickness)
membrane opposite to the transverse filaments) in channels with
0.50.5mm square bar filaments and 0.392mm0.5mm (thickness)
u0=0.1m/s; h=1mm; lf=4.5mm) ..........................................................................80
Figure 5.10 Salt concentration (c/c0) profiles in a feed channel with 0.50.5
mm square bar transverse filaments (simulation conditions: p=800psi;
u0=0.1m/s; h=1mm; lf=4.5mm) ..........................................................................82
0.25(thickness)0.5mm (width) rectangular bar transverse filaments
u0=0.1m/s; h=1mm; lf=4.5mm) ..........................................................................83
0.50.5mm square bar transverse filaments (simulation conditions:
lf=4.5mm) ...........................................................................................................84
xi
0.75(thickness)0.5mm (width) rectangular bar transverse filaments
u0=0.1m/s; h=1mm; lf=4.5mm) ..........................................................................85
membrane opposite to the transverse filaments) in channels with square
bar filaments with different filament sizes (0.25mm, 0.5mm and 0.75mm
in thickness and 0.5mm in width) (simulation conditions: p=800psi;
u0=0.1m/s; h=1mm; lf=4.5mm) ..........................................................................87
Figure 5.17 Contour of flow velocity in a feed channel (part) with 0.50.5 mm
square bar transverse filaments (simulation conditions: p=800psi;
(thickness)0.5 (width) mm rectangular bar transverse filaments
u0=0.1m/s; h=1mm; lf=4.5mm) ..........................................................................89
rectangular bar filaments with different filament sizes (0.25mm, 0.5mm
and 0.75mm in thickness and 0.5mm in width) (simulation conditions:
lf=4.5mm) ...........................................................................................................92
rectangular bar filaments with different filament sizes (0.25mm, 0.5mm
and 0.75mm in thickness and 0.5mm in width) (simulation conditions:
lf=4.5mm) ...........................................................................................................93
Figure 5.21 Local wall concentration (cw/c0) profiles (on the membrane
attached to the transverse filaments) in feed channels with 0.5mm square
bar filaments with reduced mesh length (simulation conditions:
lf=2.5mm) ...........................................................................................................94
xii
Figure 6.1 Illustration of the computing domain (part) ..............................................99

Figure 6.2 Salt concentration (c/c0) profiles in the feed channel (part) with
cavity spacer (simulation conditions: p=800psi; c0=32,000mg/l;
A=7.310-12m/s Pa; u0=0.1m/s; H=1mm; lf=4.5mm).......................................102
zigzag spacer (simulation conditions: p=800psi; c0=32,000mg/l;
submerged spacer (simulation conditions: p=800psi; c0=32,000mg/l;
Figure 6.5 Contour of X-component flow velocity in the feed channel (part)
with submerged spacer (simulation conditions: p=800psi;
c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s; H=1mm; lf=4.5mm) .............106
Figure 6.6 Contour of flow velocity in the feed channel (part) with submerged
spacer (simulation conditions: p=800psi; c0=32,000mg/l; A=7.31012
m/s Pa; u0=0.1m/s; H=1mm; lf=4.5mm)........................................................107
Figure 6.7 Comparison of local wall concentration (cw/c0) profiles in feed
channels with cavity, zigzag and submerged spacers (simulation
H=1mm; lf=4.5mm)..........................................................................................108
Figure 6.8 Contour of flow velocity in the feed channel (part) with zigzag
m/s Pa; u0=0.1m/s; H=1mm; lf=4.5mm)........................................................109
Figure 6.9 Contour of x-component flow velocity in the feed channel (part)
with zigzag spacer (simulation conditions: p=800psi; c0=32,000mg/l;
Figure 6.10 Comparison of local permeate velocity profiles in 10cm long feed
channels with cavity, zigzag and submerged spacers (simulation
H=1mm; lf=4.5mm)..........................................................................................113
Figure 6.11 Comparison of permeate flux in 10cm long feed channels with
different mesh length and with different filament configurations (cavity,
xiii
zigzag and submerged) (simulation conditions: p=800psi;

c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s; H=1mm) ..............................116
Figure 6.12 Comparison of the impact of mesh length on averaged permeate
flux in channels with submerged spacers.........................................................117
flux in channels with zigzag spacers................................................................118
flux in channels with cavity spacers.................................................................119
m/s Pa; u0=0.1m/s; H=1mm; lf=1.5mm)........................................................121
Figure 6.16 Contour of flow velocity in the feed channel (part) with cavity
m/s Pa; u0=0.1m/s; H=1mm; lf=1.5mm)........................................................122
Figure 6.17 Comparison of pressure loss in 10cm long feed channels with
different mesh length and with different filament configurations (cavity,
zigzag and submerged) (simulation conditions: p=800psi;
c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s; H=1mm) ..............................123
m/s Pa; u0=0.1m/s; H=1mm; lf=0.5mm)........................................................126
xiv
LIST OF SYMBOLS
A
membrane permeability (m/s Pa)
salt concentration (kg/m3)
diffusivity (m2/s)
characteristic dimension of an element (m)
channel height (m)
osmotic pressure constant (Pa m3/kg)
lf
mesh length (distance between two neighboring filaments, m)
channel length (m)
shape function
normal direction of boundary
pressure (Pa)
applied pressure (Pa)
xv
Pe
Peclet number (=Hu0/D, H rather than hydraulic diameter is used in
this research)
Re
Reynolds number (=Hu0/, H rather than hydraulic diameter is used in
this research)
time (sec)
axial velocity (x direction) (m/s)
lateral velocity (y direction) (m/s)
axial coordinate (crossflow direction) (m)
lateral coordinate (channel height direction) (m)
Greek letters
viscosity (m2/s)
density (kg/m3);
the penalty number
natural coordinates (-1~+1) in master elements (mapping of x)
natural coordinates (-1~+1) in master elements (mapping of y)

xvi
osmotic pressure (Pa)
computation domain
boundary of
Superscripts
element
time
transpose
Subscripts
at x/L=0
element
permeate
wall or membrane surface
filament
xvii
CHAPTER 1 INTRODUCTION
1.1 Introduction
Membrane separation technology has become a popular separation technology
in many industries that require separation of solutes from aqueous solutions and water
treatment. Normal osmosis takes place when water passes from a less concentrated
solution to a more concentrated solution of solute through a semipermeable
membrane because of the osmotic pressure differences. In a RO system, a pressure
higher than the osmotic pressure differences is applied to the concentrated solution,
causing a reversed direction of water passage through the membrane. Reverse
Osmosis (RO) is the tightest possible membrane in liquid/liquid separation. Water is
in principal the only material passing through the membrane; essentially most
dissolved and suspended material is rejected. Hence, RO is used in many high-purity
water treatment and reuse systems.
Although current material and chemical engineering technology has made it

possible to manufacture membranes with excellent performance, e.g., high
permeability, low fouling potential and high rejection, for pressure-driven membrane
systems there are still some unsolved practical problems that retard the process of
popularization. Concentration polarization and membrane fouling are the most
important twin problems in most practical RO membrane systems.
In all RO membrane separation systems, concentration polarization is an

inherent phenomenon. When water continuously passes through the membrane as
permeate, part of the rejected solutes and colloids will accumulate near the membrane
surface and form a concentration layer with concentration substantially higher than
that in the bulk. This phenomenon is known as concentration polarization.
Concentration polarization is less pronounced in the crossflow systems than dead-end
systems because the rejected contaminants are continuously carried away from the
membrane surface by the cross flow. However, even in the cross flow RO system,
concentration polarization is inevitable and an important factor for membrane
performance. Concentration polarization significantly deteriorates the performance of
the membrane system both in decreasing the permeate flux and increasing salt
passage. Because of the limited knowledge of concentration polarization in real RO
membrane modules, it is still impossible to optimize the design of RO modules
corresponding to the operating conditions and feed water properties for the best
alleviation of concentration polarization in the membrane channel.
In practical RO membrane separation applications, spiral wound modules
have been widely used due to low operating cost and high packing density. In spiral
wound modules, feed spacer is an essential part to support membranes to form the
feed channel. It has been proven by experiments and numerical analyses that the feed
spacer could alleviate concentration polarization by promoting mixing in the feed
channel. However, because of the geometrical complexity of membrane channel with
the spacers, a direct quantitative linkage of the characteristics of spacer, e.g., filament
geometry, filament configurations and mesh length, to concentration polarization
and/or permeate flux in spiral wound RO modules is still unavailable. This, in turn,
has greatly hindered the progress in spacer design and optimization to alleviate
concentration polarization in spiral wound RO modules.
1.2 Aim of the research

The aim of this research is to study the concentration polarization in spacerfilled RO membrane channels and to quantify the impact of feed spacer on
concentration polarization and membrane performance. This is to be achieved through:
1. developing a numerical model capable of modeling concentration polarization in
more realistic conditions in the spiral wound RO modules under various operating
conditions and with different spacer configurations;
2. identifying and assessing the major mechanisms of concentration polarization in
spacer filled RO channels;
3. investigating the effects of filament geometry on concentration polarization;
4. investigating the effects of filament configurations and mesh length on
concentration polarization and membrane system performance.
1.3 Overview of the dissertation

To achieve the aim of this research, chapter 2 reviews the development of the
research on analytical and numerical models for concentration polarization and the
research on the impact of feed spacer on mass transfer in the feed channel. Chapter 3
describes the development and calibration of the 2-D streamline upwind
Petrov/Galerkin (SUPG) finite element model for concentration polarization in spiral
wound RO modules. Chapter 4 studies the concentration polarization patterns and
major mechanisms in spacer-filled RO channels. Chapter 5 studies the effects of
filament geometry on concentration polarization. Chapter 6 studies the impact of
filament configuration and mesh length on concentration polarization and membrane
performance. Chapter 7 is the conclusions and recommendations for further research.
CHAPTER 2 LITERATURE REVIEW

2.1 Concentration polarization in RO systems
In a RO membrane separation system, due to the permeation of the solvent the
rejected particles or solutes would be accumulated near membrane surface and form a
concentration polarization layer. However, this concentration buildup would, in turn,
decrease permeate velocity because of the elevated osmotic pressure and therefore
limit the solute transport through this convection process. The concentration buildup
would also result in diffusion of solutes from membrane surface towards the bulk
because of the concentration gradient. The cross flow may also transport the
accumulated solutes to the downstream in crossflow systems. Usually in a very short
period, steady state would be achieved (Sherwood et al, 1965; Gill et al, 1966;
Matthiasson and Sivik, 1980; Noble and Stern, 1995; Song and Yu, 1999). Because
permeate velocity interacts with wall concentration to achieve the steady state, for
concentration polarization, the solute transport and momentum transfer are coupled at
the surface of the semipermeable membrane. This makes concentration polarization
substantially different from other mass transfer problems in a channel with
impermeable walls or porous walls.
Because concentration polarization usually has significant adverse impact on

RO membrane performance, e.g., reducing permeate flux, since 1960s many
researchers have attempted to understand and quantify this phenomenon. For example,
Merten et al (1964) experimentally demonstrated the adverse effects of concentration
polarization on permeate flux and the dependency of concentration polarization on
circulation rate (crossflow velocity). Brian (1965) showed that constant permeate flux
may result in significant different local concentration polarization behavior compared

with that of variable flux. Some other studies also suggested that the local variations
of wall concentration and permeate velocity play an important role in concentration
polarization (Srinivasan et al, 1967; Matthiasson and Sivik, 1980; Song and
Elimelech, 1995; Song and Yu, 1999; Wiley and Fletcher, 2003). Quantifying the
local variations of wall concentration and permeate velocity is essential for
concentration polarization minimization and membrane system optimization.
However, because concentration polarization occurs in a very thin layer close to
membrane surface and concentration gradient in this layer is very sharp, it is still a
challenge to capture the details of concentration profiles and permeate flux in
concentration polarization layer either numerically or experimentally. In spiral wound
RO modules, a feed spacer which may alter the mass transfer and momentum transfer
patterns in the feed channel dramatically (Schwinge et al, 2004b), is always present in
order to form the feed channel. This makes it even more difficult to capture the local
variations of salt concentration and hydrodynamic conditions in these realistic RO
systems. Until now, concentration polarization in spacer-filled channels and the
impact of spacer on permeate flux are still not well understood.
It has been extremely difficult through experiments to directly observe and

detect the concentration profiles in concentration polarization layer. Therefore,
numerical simulation has been the major method to study this phenomenon. Many
concentration polarization models for crossflow membranes separation systems have
been proposed. Most of the models deal with particles or colloids and are usually
valid only for MF/UF but not for RO membranes on which concentration polarization
is mainly caused by the buildup of solutes. Concentration polarization models for RO
membranes can be classified as: i) analytical models for permeate rate and/or wall
concentration based on some simplified assumptions and/or analogies; and ii)
numerical models of the governing equations with few assumptions. These two
different types of models are reviewed in Sections 2.2 and 2.3 respectively. In
addition, the spacer in feed channel has been an interesting and challenging research
subject of great practical and theoretical importance. Studies on the impact of spacers
on concentration polarization and membrane performance are reviewed in Section 2.4.
Finite element method is one of the major powerful numerical methods in

computational fluid dynamics to deal with complex computing domains like spacerfilled channels. Streamline upwind Petrov-Galerkin (SUPG) finite element has been
widely reported in solving convection dominated problems such as solving NavierStokes equations and convection-diffusion equation. The recent advances of finite
element method in mass and momentum problems are reviewed in Section 2.5.
2.2 Analytical models for concentration polarization in RO

membrane systems
Although the analytical theoretical models have many different forms and
names, solute transport in the feed channel is essentially modeled by a film model and
solvent passing through the membrane is modeled by a few parameters, including
membrane permeability or resistance, transmembrane pressure and elevated osmotic
pressure at the membrane surface. Spiegler and Kedem (1966) attributed the flux
decline to the increased osmotic pressure induced by the higher concentration of the
rejected solute near the membrane surface. This approach to determine permeate flux
from applied pressure, membrane permeability, reflection coefficient and elevated
6
osmotic pressure due to concentration polarization is now widely accepted and is

essentially equivalent to Darcys law. As for stagnant film models, Zydney (1997)
presented a mathematical justification for this kind of model from the fundamental
governing equation of solute transport equation (convection-diffusion equation) as
well as different forms of film model for concentration-dependent viscosity and
diffusivity.
Most of the analytical models are based on film models and Spiegler-Kedens
permeate flux approach or on some analogies. Mass transfer in film model was either
expressed as an analytical expression of the concentration or a correlation for the
mass transfer coefficient. Johnson and Acrivos (1969) developed an expression for
wall concentration based on the analogy of natural convection boundary layer
problems. Srinivasan and Tien (1970) developed a concentration polarization model
by assuming that the local Sherwood number for a given species of solute is
approximately proportional to the cubic root of its Schmidt number and relatively
independent of other parameters. This makes it possible to predict concentration
polarization for multi-component systems. Gekas and Hallstrom (1987) reviewed the
mass transfer correlations in turbulent ducts and proposed a modified correlation for
mass transfer in concentration polarization layer. The correlation was often cited in
later studies. Bader and Jennings (1992) further developed the correlation for mass
transfer in turbulent flow regime by assuming that the actual rejection is a function of
wall concentration.
In recent years, there have been a few concentration polarization models

incorporating some other factors to enhance the general-purpose film model and/or
permeate flux formulations.
For example, Bhattacharya and Hwang (1997)
introduced a modified Peclet number and separation factor into concentration

polarization model to enhance the modeling ability. Elimelech and Bhattacharjee
(1998) developed a comprehensive model for the concentration polarization
phenomenon during crossflow membrane filtration of small hard spherical solute
particles by combining the hydrodynamic and thermodynamic (osmotic pressure)
approaches. Song and Yu (1999) developed a dimensionless model for concentration
polarization in crossflow RO systems by coupling solute mass balance, film model
and the fundamental permeate rate function.
Because the analytical models were developed under various implicit or

explicit assumptions, these types of models are only applicable to cases in which all
these assumptions are valid. This limitation is often ignored or overlooked. In
addition, due to mathematical limitations the analytical models are usually unable to
deal with domain and/or boundary with complex geometry; therefore, it is almost
impossible for the impact of the spacer on concentration polarization or membrane
performance to be simulated by these types of models although they may better our
understanding of the role of certain parameters in concentration polarization.
2.3 Numerical models for concentration polarization in RO

membrane systems
Concentration polarization in RO membranes is the result of solute and
solvent transport in the feed channel governed by coupled momentum and mass
transfer equations. Concentration polarization in RO membrane systems can be
adequately described by Navier-Stokes equations for water flow coupled with
8
convection-diffusion equation for solute transport (Srinivasan et al, 1967;

Matthiasson and Sivik, 1980; Belfort and Nagata, 1985).
With the development of high performance computing technology and the

urgent demand for better understanding of concentration polarization, attempts have
been made to seek numerical solutions of the coupled Navier-Stokes equations and
convection-diffusion equation. The numerical difficulty of solving the coupled
governing equations in the RO membrane feed channel is characterized by the
extremely large aspect ratio, which is usually less than 1mm in height and several
meters in length. Moreover, the thickness of the concentration polarization layer is
usually of the order lower than 10-4m (Bhattacharyya et al,1990; Huang and
Morrissey, 1999) and a computational mesh height as low as 10-6m or lower is
required to capture the very steep concentration gradient near the membrane surface.
Therefore, it is still very difficult to obtain the detailed velocity and salt concentration
information in the concentration polarization layer and using a too coarse mesh or
improper algorithm may produce erroneous results (Ma et al, 2004).
Currently, there are still very a few literature reports on study of concentration
polarization in RO membranes by numerically solving the coupled Navier-Stokes and
convection-diffusion equations. Most numerical models tried to achieve numerical
solutions of the solute transport equation only, in which the water flow field was
determined by using the simplified analytical velocity field derived for empty channel
(Berman, 1953) and/or empirical velocity profiles for spacer-filled channel (Focke
and Nuijens, 1984; Miyoshi et al, 1982). The decoupling of solute transport from
momentum transfer actually eliminated the role played by the interaction between
solute and momentum transfers. One of the most important factors for concentration
polarization became intangible and the settings for the study of effect of spacers were
unduly oversimplified.
Early numerical studies of concentration polarization were mainly focused on

empty channels (without spacers). For example, Sherwood et al (1965) developed
concentration polarization models in empty channels for both turbulent and laminar
flow conditions in both cylindrical and flat sheet geometry based on Bermans
simplified flow field. In this model, constant permeate flux was assumed and
therefore, solute transport and momentum transfer was uncoupled. Brian (1965)
relaxed the constant flux assumption and solved the solute transport equation with
finite difference method. It was found that it was faster than the infinite series method
as used by Sherwood et al (1965). The impact of variable flux on concentration
polarization was also investigated. The results demonstrated that local variations of
wall concentration were significantly different if variable flux was used. Srinivasan
and Tien (1969) further studied mass transfer characteristics of reverse osmosis in
turbulent flow regime by finite difference method, which would require fewer
assumptions compared with series solution. They found that the implicit assumption
that convective term in crossflow direction is negligible in the solute transport
equation (which was used in most one dimensional concentration polarization models)
was questionable.
Since 1990s with the development of computing technologies, a few twodimensional models for solute transport in membrane systems were developed.
Bhattacharyya et al (1990) developed a two-dimensional Galerkin finite element
10
model for concentration polarization in RO membranes based on the simplified

velocity field (Berman 1953), and modeled the effects of flow condition, diffusivity,
membrane permeability, transmembrane pressure, tapered cell geometry and nonNewtonian fluid on wall concentration and permeate rate. The numerical results
suggested that the convection-diffusion equation was preferable to film model for
concentration polarization modeling in RO membranes. In order to study the solute
transport in turbulent flow, Pellerin et al (1995) modeled turbulent flow features in
empty membrane modules using upwind finite difference package and solved NavierStokes equations and solute transport equation independently, i.e., Navier-Stokes
equations and solute transport equation were not coupled. The model was actually
inapplicable to RO membranes because introducing the inexistent turbulent
dissipation may yield erroneous hydrodynamic results when the crossflow velocity is
only in the order of 10-1m/s in most spiral wound RO modules. In addition, setting
osmotic pressure term to zero in Darcys law for permeate velocity and improper
concentration boundary condition at membrane surface make it difficult for the model
to simulate concentration polarization.
Similar models were developed for UF and nanofiltration (NF) systems. For
example, Lee and Clark (1998) developed a finite difference model to simulate
concentration polarization in crossflow UF systems when the effects of cake
formation are important. The model used simplified velocity field and incorporate
cake resistance into it. Huang and Morrissey (1999) modeled the concentration
polarization in ultrafiltration of protein solutions (BSA) using finite element analysis
package to solve the convection-diffusion equation with simplified velocity field from
Berman (1953). Unsteady models were also reported. For example, using the same
11
simplified velocity field (Berman, 1953), Madireddi et al (1999) numerically solved

the convection-diffusion equation for empty channel and the effect of spacers was
considered through a mixing constant in velocity field. In this model, unsteady state
governing equation was used to obtain the steady state solutions, which may save
some computing resources.
All these models employing simplified velocity profiles are uncoupled, i.e.,
solute transport equation is solved based on pre-determined velocity field. This
essentially omitted the interaction of solute transport and momentum transfer, which
is the characteristic process in concentration polarization. In order to resolve this,
several coupled models were developed. The following segments discuss the
work/research related to this goal.
Srinivasan et al (1967) developed a numerical model that could simulate the

simultaneous development of velocity and concentration profiles in two dimensional
and axisymmetric empty RO channels. However, in this model the quadratic
expression for the concentration profile in the concentration polarization layer was
assumed. This may underestimate the wall concentration because the concentration
profile in the concentration boundary layer is exponential-like rather than quadratic;
thus when the thickness of concentration boundary layers decreases, higher order
polynomial terms may be required to make good approximation. Without relying on
such an assumption, Geraldes et al (1998) studied mass transfer in slits with semipermeable membrane walls with a coupled numerical model. It was found that hybrid
scheme for convection and diffusion terms was more suitable for this type of problem
compared with upwind and exponential schemes. The impact of grid refinement on
12
the accuracy was also studied. However, this model was not tested for spacer-filled
channels.
To address the complex geometry in the feed channel, Richardson and

Nassehi (2003) developed a finite element model to simulate concentration
polarization with curved porous boundaries in an empty channel. This model implies
that finite element method would be suitable for modeling concentration polarization
especially for feed channel with complex geometry.
Numerical models based on commercial CFD (computational fluid dynamics)

software were also reported. For example, Wiley and Fletcher (2003) developed a
coupled concentration polarization model using the commercial computational fluid
dynamic software (CFX). The model was, however, only applied to empty channels
and not tested in spacer-filled channels. However, this research implied that solving
the fully coupled Navier-Stokes equations and solute transport equation could provide
a very effective way to simulate concentration polarization in membrane systems and
to study the effects of spacers on concentration polarization, which is far beyond the
ability of any analytical and empirical models.
Similar coupled models were also reported for NF systems. For example,
Geraldes et al (2002a, 2004) developed a numerical model for concentration
polarization in nanofiltration (NF) spiral wound modules based on coupled solute
transport and Navier-Stokes equations. However, in this model fixed permeate
velocity was assumed, so the interaction between solute transport and momentum
transfer was also neglected. As revealed by Brian (1965), this constant flux
13
assumption would lead to significant errors in concentration polarization simulations

especially when wall concentration profiles are concerned.
2.4 The impact of spacer on concentration polarization and RO

membrane performance
In the most commonly used RO membrane modules, the feed channels are
filled with spacers. However, the impact of spacers on concentration polarization and
membrane performance is only understood qualitatively. Most of the studies on this
subject used a single parameter of eddy constant and the empirical velocity field to
describe the impact of spacers on concentration polarization, while the important
characteristics such as the geometry of spacer filaments and configuration of
filaments could not be accounted. It has been shown that mesh length and the
configuration of filaments have significant impact on velocity field and other
hydrodynamic parameters (Cao et al, 2001; Schwinge et al, 2002a, 2002b). This
implies that concentration polarization may also be significantly affected by different
spacer configurations because of the resulting flow velocity field. However, most of
the spacer studies focused on hydrodynamic conditions in the feed channel.
Quantitative simulations and assessments of the impact of spacers on concentration
polarization and membrane performance were very limited.
Tien and Gill (1966) found that the alternatively placed impermeable sections
and membrane sections could relax concentration polarization and therefore increase
membrane productivity noticeably. This implies that the impermeable spacer
filaments, which invariably block some membrane areas, may relax concentration
polarization noticeably even if the mixing is not enhanced significantly by the spacer.
14
Early studies on the effects of spacers usually focused on obtaining some

overall correlations for flow, pressure drop and/or mass transfer coefficient using one
or more parameters (Winograd et al, 1973; Chiolle et al, 1978; Kang and Chang,
1982; Focke and Nuijens,1984; and Miyoshi et al, 1987; Boudinar et al, 1992; Da
Costa et al, 1994). Such correlations cannot explain mechanisms or manner with
which the spacer filaments of different configurations affect concentration
polarization and membrane performance, which is critical to understand the spacers
impacts on the performance of the membrane (Song and Ma, 2005). For example,
Schock and Miquel (1987) found that for all feed spacers investigated the friction
coefficient is only related to Reynolds number with a power of -0.3. It was also
reported that the measured Sherwood number in the spacer-filled channel was
significantly higher than that in empty channels and the Sherwood number is
dependent on Reynolds number with a power of 0.875 and Schmidt number with a
power of 0.25. In contrast, Kim et al (1983) found that the spacer configurations had
noticeable effects on mass transfer and zigzag type promoters was more effective than
cavity ones in concentration polarization alleviation based on their experiments. In a
similar electrodialysis system, Kang and Chang (1982) studied flow and mass transfer
in an eletrodialysis system with zigzag turbulence promoters and found that mass
transfer was greatly enhanced by these promoters through forming recirculation field
in the downstream of the filaments. The effects of spacer on pressure drop were also
reported. For example, Da Costa et al (1994) developed a novel model to simulate
pressure drop due to cylindrical filaments based on momentum balance in UF
membrane systems.
15
Recently, Cao et al (2001) numerically solved the two-dimensional NavierStokes equations using commercial CFD software package (Fluent) and different
velocity profiles due to different spacers and the possible impact on concentration
polarization were tentatively studied. It was revealed that the geometry of spacers and
configuration of the feed channel had significant effects on flow patterns, velocity
distribution and wall shear stress. Similarly, Karode and Kumar (2001) simulated the
flow and pressure drop in rectangular channels filled with several kinds of
commercial spacers using the commercial CFD package of PHOENICS.
To further study the effects of filament configuration on mass transfer,

Schwinge et al (2002a, 2002b) used the commercial CFD package (CFX) to simulate
flow patterns and mass transfer enhancement in the feed channel with three different
spacer configurations: cavity, zigzag and submerged spacers. Some statistical
relationships for the scale of the recirculation region, the mass transfer enhancement,
channel height, mesh length, filament diameter and Reynolds number were developed.
However, constant (and artificial) wall concentrations and impermeable wall were
assumed on membrane surface in their study; therefore the impact of spacer on
concentration polarization and permeate flux are unable to be simulated.
The effects of feed spacer on concentration polarization in NF systems were

also reported. For example, Geraldes et al (2002a, 2002b, 2004) studied the impacts
of feed spacer (cavity configuration) on flow and concentration polarization in NF
membrane systems. It was reported that the recirculation and concentration
polarization were significantly affected by filament thickness and mesh length.
16
Some researchers also attempted to optimize the design of feed spacer to

achieve the best system performance. For example, Li et al (2002, 2005) studied mass
transfer enhancement and power consumption of net spacers and tried to develop
some optimized spacer configurations. However, in their numerical simulations fixed
wall concentration (zero) and impermeable boundary (vw=0) were assumed for
membrane surface, so it is also impossible to simulate and study the effects of spacer
on concentration polarization and permeate flux.
Although there are some studies on the impact of feed spacer on

hydrodynamics, the quantitative link between spacer filaments (geometry,
configuration, mesh length) and concentration polarization or membrane performance
in RO systems has so far not been reported.
2.5 Finite element method for coupled momentum transfer and

solute transport problems
Finite element method (FEM) is one of the popular and reliable numerical
methods in solving mass, heat and momentum transfer problems. One of the
advantages of FEM over finite difference method (FDM) is its ability to deal with
arbitrary geometries by using unstructured meshes. The ability of naturally
incorporating differential type boundary conditions also makes FEM preferable to
FDM and finite volume method (FVM) when dealing with problem with open and
flux boundaries (Chung, 1978; Zienkiewicz and Taylor, 2000).
Several formulations such as velocity-pressure formulation, penalty

formulation,
mixed
formulation,
streamfunction-vorticity
formulation
and
17
streamfunction formulation, are commonly used in numerical solution of NavierStokes equations. Malkus and Hughes (1978) proved that the solutions obtained from
mixed formulation with bilinear elements and constant pressure are identical to those
obtained from penalty formulation with bilinear elements and one-point reduced
integration of the penalty term. Penalty method for Navier-Stokes equations has been
extensively studied and successfully applied in many different cases (Carey et al 1984;
Dhati and Hubert 1986; Shih, 1989; Funaro et al 1998; Hou and Ravindran 1999; Wei,
2001; Be et al, 2001; Valli, 2002). The advantage of penalty formulation over other
formulations is less computation time required because of the reduced number of
unknowns.
In convection dominated problems, streamline upwind Petrov-Galerkin

(SUPG) finite element formulation can produce wiggle-free solutions without
refinement and losing accuracy. In the last 20 years, SUPG finite element method has
been applied in many kinds of mass, heat and momentum transfer problems and now
is one of the preferable methods for this kind of problem. Brooks and Hughes (1982)
developed and introduced this method systematically for the first time for
incompressible Navier-Stokes equations. Later this method was successfully used in
solute transport problems, solute transport and reaction problems (Hughes and Mallet,
1986; Tezuyar et al, 1987; Ielsohn et al, 1996).
In the study of concentration polarization, because only part of the feed

channel is usually modeled, a proper numerical scheme to treat the open boundary is
essential. Hassanzadeh et al (1994) showed that the inhomogeneous Neumann
boundary conditions were satisfied automatically by direct Galerkin finite element
18
formulation of the derived pressure Poisson equation. Heinrich et al (1996) showed

that for Navier-Stokes equations the natural boundary conditions must be combined
with a procedure to eliminate perturbations on the pressure at the open boundary.
Griffiths (1997) discussed the open boundary conditions for an advection-diffusion
problem and found that the open boundary was superior to a standard no-flux outflow
condition. Renardy (1997) proved that open boundary condition proposed by
Papanastasiou et al (1992) would not lead to the problem being underdetermined at
the discrete level but would yield a well-defined problem superior to that with
artificial boundary condition. Padilla et al (1997) studied open boundary conditions
for two-dimensional advection-diffusion problem and found that open conditions
were not compatible with the conservative formulation for non-conservative and
steady state flow fields.
2.6 Summary
Concentration polarization, which is the result of the interaction of solute
transport and momentum transfer, is an inherent phenomenon in membrane separation
systems and may affect the performance of RO membrane systems significantly.
However, this phenomenon is still not well understood in practical spacer-filled RO
systems. Detailed salt concentration profile, the impacts of spacer on wall
concentration and permeate flux have rarely been reported.
Both analytical and numerical models are commonly used in the study of
concentration polarization. Analytical models are derived based on some simplified
assumptions and/or analogies. They are only suitable for empty channels of simple
geometry. Concentration polarization in the membrane feed channels with spacer
19
filaments is too complex to be adequately described by any analytical models. On the

other hand, numerical models are developed with few assumptions and they are more
suitable for the study of concentration polarization in spacer-filled channels. However,
most of the current numerical models oversimplify or neglect the interaction between
solute transport and momentum transfer on the membrane surface, which is the
essential cause for concentration polarization to occur. This renders these numerical
models unsuitable for accurate simulation of concentration polarization.
To accurately study concentration polarization in a membrane channel of

complex geometry and the impacts of feed spacer on concentration polarization and
permeate flux, a fully coupled model that could adequately handle the interaction of
momentum transfer and solute transport on membrane surface is essential. Such
models have not been reported in the literature for concentration polarization in
spacer filled RO membrane channels. The intrinsic merit of finite element in naturally
dealing with open and flux boundary conditions and dealing with problems with
complex geometry makes it preferable in modeling concentration polarization and the
effects of feed spacers.
20
CHAPTER 3 NUMERICAL MODEL

3.1 Introduction
Concentration polarization is essentially the result of mass and momentum
transfers in the feed channel. In the membrane channel, the permeate flow brings
solutes or particles to membrane surface and leaves them there when the water passes
through the membrane. On the other hand, the formation of high concentration layer
on the membrane surface alters permeate flux by providing an additional resistance
layer. Therefore, understanding of the interactions between mass and momentum
transfers is the key to simulate accurately concentration polarization and permeate
flux in the channel.
This interaction makes concentration polarization in a membrane channel

substantially different from the typical mass transfer problem in channels with
impermeable walls or with sinks/sources in the wall. Any assumptions of pre-scripted
boundary conditions of concentration or permeate flux on the membrane surface
would not reflect the problem appropriately. A realistic solution of the problem must
be sought directly from a coupled model of solute transport and momentum transfer
in the feed channel.
Momentum and mass transfers in a membrane channel are generally described

by Navier-Stokes equations and convection-diffusion equation, respectively. Both
equations have been studied extensively in fluid mechanics and mass transfer
problems. However, difficulties are often encountered in obtaining numerical
solutions of these equations when they are applied to the RO feed channel. Firstly,
21
the geometry of the domain of interest is extremely narrow and long in typical spiral
wound RO modules. The height of the feed channel is usually less than 1 mm and the
length in the order of meters. Moreover, the thickness of concentration polarization
layer is usually far less than the channel height. Therefore, the required number of
elements or cells or grids is enormous in order to satisfy the requirement of aspect
ratio of the elements and to capture the sharp gradient of salt concentration near the
membrane surface. Secondly, for solute transport in the feed channel, Navier-Stokes
equations and convection-diffusion equation are coupled not only in the domain but
also on the boundary (membrane surface); most general-purpose CFD software
packages are not designed for this type of problem. Therefore, it is necessary to
develop a specially designed numerical model for concentration polarization to
reliably simulate this phenomenon in spiral wound modules.
In this chapter, a finite element model specially designed for the study of
concentration polarization and membrane system performance (e.g. permeate flux and
pressure loss) in spiral wound RO modules is presented. The streamline upwind
Petrov/Galerkin (SUPG) method was employed in the model to solve numerically the
coupled governing equations. The model was then tested with published experimental
data and other models.
3.2 Model development

3.2.1 Governing equations
Laminar flow can be generally assumed in practical spiral wound RO modules
because of the small channel height and low fluid velocity (Van Gauwbergen and
22
Baeyens, 1997). Schock and Miquel (1987) and Van Gauwbergen and Baeyens
(1997) have shown that the bent envelope in spiral wound modules can be reasonably
approximated by an unwound flat membrane channel. Therefore, momentum transfer
and solute transport in the spiral wound membrane module can be adequately
described in two dimensions by:
Navier-Stokes equations:
u v
+
=0
x y
u
u
u
1 p
2u 2u
+u
+v
=
+ ( 2 + 2 )
y
t
x
y
x
x
2v 2v
v
v
v
1 p
+ ( 2 + 2 )
+u +v
=
t
x
y
x
y
y
(3.1)
(3.2)
Solute transport equation:
2c 2c
c
c
c
= D( 2 + 2 )
+u +v
y
x
y
x
t
(3.3)
Equations (3.1)-(3.3) are the dynamic models with temporal terms involved.
Numerical solutions of the equations are pursued within a time frame until a steadystate is achieved so that:
u v c
=
=
=0
t t t
23
One of the advantages of seeking the steady-state solutions from a dynamic

model is that it may significantly lower the memory demand if proper time integration
scheme is used.
3.2.2 Penalty formulation for Navier-Stokes equations

Navier-Stokes equations can be numerically solved in several formulations,
e.g., velocity-pressure formulation, penalty formulation, mixed formulation, stream
function-vorticity formulation and stream function formulation. In our model, the
widely used penalty formulation (Carey and Krishnan, 1984; Shih, 1989) is applied.
For penalty formulation of Navier-Stokes equations, the pressure can be expressed as:
v
p = v
(3.4)
The continuity equation will be automatically satisfied, i.e., the divergencefree velocity field will be obtained when penalty function is substituted into NavierStokes equations. Hence, the continuity equation and Navier-Stokes equations would
be written as:
u
u
u u v
2u 2u
+u +v
=
( + ) + ( 2 + 2 )
t
x
y x x y
x
y
v
v
v u v
2v 2v
+u +v
=
( + ) + ( 2 + 2 )
t
x
y y x y
x
y
(3.5)
One of the advantages of this formulation is that the number of unknowns in

the global equations would be significantly reduced. This may save considerable
computing resources.
24
3.2.3 Initial and boundary conditions

Initial conditions are arbitrary in this problem because steady state of
numerical solutions of a dynamic model is not dependent on the initial conditions.
However, properly chosen initial conditions may significantly shorten the CPU time
required to achieve the steady-state solution. The boundary conditions of the problem
are described below.
For the inlet, Dirichlet boundary (prescribed velocity and salt concentration) is
imposed:
u(x=0,y,t)=U0(y,t); v(x=0,y,t)=V0(y,t); c(x=0,y,t)=C0(y,t)

For the impermeable solid wall and the spacer/filaments, no-slip, nopenetration conditions are imposed:
u=0; v=0; D
c
=0
n
At membrane surface velocity and salt concentration are coupled through:
v w = A(p (c w , c p ))
c
= v w (c w c p )
n
(3.6)
(3.7)
25
In the study of membrane filtration, a small part rather than the whole feed
channel of the membrane module is often simulated, therefore, an open boundary
condition is usually imposed on the outlet (Papanastasiou et al, 1992; Heinrich et al,
1996; Padilla et al, 1997).
Penalty formulation of Navier-Stokes equations was used for the numerical

solutions; hence the boundary conditions are not required for pressure.
3. 2.4 SUPG finite element formulation and numerical strategies

The finite element method (FEM) (Chung, 1978; Zienkiewicz and Taylor,
2000) is applied to solve the governing equations. Bilinear quadrilateral isoparametric
elements are used; hence, except for convection terms, bilinear shape functions are
used as weighting functions:
Ni =
1
(1 + i)(1 + i )
4
i=1,2,3,4
(3.8)
For convection terms, the optimal format of a streamline upwind PetrovGalerkin SUPG weighting functions (A. N. Brooks and T.J.R. Hughes, 1982) is
applied:
Wi = N i+
N
N
v (u i + v i )
x
y
2V
i=1,2,3,4
(3.9)
26
where
= coth
; =
r
V h
for Navier - Stokes equations and =
r
V h
D
for solute
transport equation.
The trial functions for variables of u, v and c can be written as:
u( x, y , t ) = N i U it = N e U et
(3.10)
i =1
v ( x, y , t ) = N i Vi t = N e Vet
(3.11)
i =1
c( x, y, t ) = N i Cit = N e Cet
(3.12)
i =1
where,
N e = [ N1 , N 2 , N 3 , N 4 ]
U et = [U 1t, U 2t , U 3t , U 4t ]T
Vet = [V 1t, V2t , V3t , V4t ]T
Cet = [C 1t , C2t , C3t , C4t ]T
27
Using SUPG FEM described above, the elemental governing equation can be
written as:
Ue
t
t
t Ne
t Ne
e Ni Ne t dA+ e Wi Ne Ue x Ue dA+ e Wi Ne Ve y Ue dA
Ni
e
N
2 N
Ne
1
t
t
t
Ue + e Ve ) dA ( Ni 2 e Ue dA+
(
x x
y
Re e
x
Ni
2 Ne
t
Ue dA) = 0
2
y
(3.13)
Ni Ne
Ve
N
N
t
t
dA+ Wi Ne Uet e Ve dA+ Wi Ne Vet e Ve dA
e
e
t
x
y
Ne t
2 Ne t
Ne
1
t
Ni (
Ue +
Ve ) dA ( Ni 2 Ve dA+
e
y x
y
Re e
x
2 Ne t
e Ni y2 Ve dA) = 0
(3.14)
Ce
t
t
t Ne
t Ne
e Ni Ne t dA+ e Wi Ne Ue x Ce dA+ e Wi Ne Ve y Ce dA
2 N
1
t
( Ni 2 e Ce dA+
Pe e
x
Ni
2 Ne t
Ce dA) = 0
y 2
(3.15)
i=1, 2, 3, 4
Using Greens theorem, the elemental governing equations (matrix format)
can be written as:
Me
U e
t
t
t
t
t
t
t
e
+ K ex U e [U e ]T + K ey U e [Ve ]T + K1e U e + K e2 Ve + K 3e U e Fx = 0
t
(3.16)
28
V
t
t
t
t
t
t
t
e
M e + K ex Ve [U e ]T + K ey Ve [ Ve ]T + K e2 U e + K 1'e Ve + K 3e Ve Fy = 0
t
e
(3.17)
t
C e
t
e
t
t
t
t
M
+ K ex C e [U e ]T + K ey C e [ Ve ]T + K '3e C e Fc = 0
t
e
(3.18)
where,
M ije = N i N j d
e
K exijk = Wi N j
Nk
d
x
K eyijk = Wi N j
Nk
d
y
Ni N j
d
e x
x
K 1eij =
K e2ij =
Ni N j
d
y
x
Ni N j
d
e y
y
K 1'eij =
29
K 3eij =
Ni N j Ni N j
1
(
)d
Re e x
x
y
y
K '3eij =
Ni N j Ni N j
1
(
)d
Pe e x
x
y
y
Ne
Ne
Ne Ne
1
t
) U e n x d + N i (
Ue +
Ve ) n x d
+
N i (
e
Re e
y
y
x
x
Ne
Ne
Ne Ne
1
t
) n y Ve d + N i (
Ue +
Ve ) n y d
+
=
N i (
e
Re e
y
y
x
x
Fxi =
Fyi
Fci =
Ne Ne r
1
t
) n C e d
+
N i (
Pe e
y
x
i,j,k=1,2,3,4
After assembling the global algebra equations would take a similar format (T.J.
Chung, 1978; O.C. Zienkiewicz and R.L. Taylor, 2000).
Four-node (22) Gauss quadrature is applied in obtaining numerical

integration of the elements except for the penalty terms where one point quadrature is
used to meet Ladyzhenskaya-Babuska-Brezzi (L.B.B.) stability conditions.
The
implicit Euler method is applied for time integration except for convection and
boundary terms that are treated explicitly. At any time step, solution of NavierStokes equations is first sought for velocity fields; the solution of solute transport
equation is then sought for concentration distribution based on the newly obtained
velocity fields. The boundary conditions would be updated with the newly obtained
30
concentration field for next time step. The procedure is repeated until steady-state is
achieved, i.e., the derivatives
c u
v
,
, and
are smaller than the preset tolerances.
t t
t
The linear algebra equations are solved by direct method (LU factorization) because
penalty formulation is applied to the Navier-Stokes equations.
The SUPG FEM solver was written in standard FORTRAN 90 and was
parallelized using OpenMP. It has been tested on various platforms including Tru64,
IRIX, HP-UX, Solaris, Linux and Windows. The solver was designed to read/import
geometric modeling results from third-party software so that automatic geometry
modeling and meshing can be applied. At the present stage the solver can directly
read neutral files from GAMBIT (Fluent Inc, USA) for geometry and meshing. For
post-processing and visualization, TECPLOT 9 (AMTEC, USA) is used in the solver,
i.e., the solver writes output files for this post-processing software directly. The flow
chart of the model is illustrated in Figure 1.
31
SUPG FEM solver
Pre-processor
Post-processor
Start
TECPLOT
Initialize
GAMBIT
(meshing)
(visualization)
Solve N-S equations
Update velocity field
Solve solute transport

equation
Update concentration
field
Y
Steady state?
N
Continue?
Y
N
End
Figure 3.1 Flowchart of the solver and model organization
32
3.3 Model validation

It is essential to test and validate a numerical model before it is applied to the
problems under study. In this research, I compared the results of our numerical model
with published experimental data on concentration polarization. The results indicated
that the model developed could reliably simulate concentration polarization in RO
systems.
The numerical model was calibrated with data on the effects of concentration
polarization on permeate flux reported by Merten et al (1964).
Concentration
polarization in the channel was simulated with 24,381 nodes and 24,000 quadrilateral
unstructured elements. Fully developed parabolic flow was assumed in the inlet and
constant physical property of the solution (diffusion coefficient of the solute:
D=1.510-9 m2/s; viscosity: =1.010-6m2/s) was assumed.
Simulation results
together with experimental data and previous numerical results reported by Srinivasan
et al (1965) are shown in Figure 3.2.
It was found that, if the value of membrane permeability was set to 7.6106
(g/cm2sec atm), our model would underestimate permeate flux, i. e., overestimate
concentration polarization. This permeability value was estimated by Merten and coworkers by assuming that concentration polarization was negligible when the
crossflow velocity increased to 200cm/s. However, this assumption is rather
questionable. Firstly, in the very narrow empty feed channels, it is difficult for the
flow to develop into turbulence at that crossflow velocity. According to our
simulations, under the experimental conditions the flow is still quite stable and
laminar for crossflow velocity up to 250 cm/sec. Because concentration polarization
33
Figure 3.2 Comparison of numerical simulation results with experimental data
Figure 3.3 Simulated wall concentration at different values of membrane permeability

(crossf low velocity: 200cm/s; other conditions: Merten et al (1964)
is usually significant in RO channels when the flow is laminar, concentration

polarization may be significant when the crossflow velocity is 200cm/s. Figure 3.3
shows the wall concentration along the channel when the crossflow velocity is
34
200cm/sec with different permeability values. It is obvious that concentration

polarization is still noticeable and therefore, should be considered at this crossflow
velocity. Secondly, even in the fully developed turbulent flow, mass transfer in the
viscous sublayer near the membrane surface in the concentration polarization layer is
still dominated by molecular diffusion rather than eddy diffusivity or turbulent mixing;
therefore, the effect of concentration polarization, which is mainly due to molecular
diffusion, may not be negligible. Thus, the value of membrane permeability (7.6106
g/cm2 sec atm) estimated by Mertern and co-workers may be underestimated. In fact,
several researchers have found this value of membrane permeability to be inaccurate.

For example, Gill et al (1966) found the value estimated by Merten and co-workers is
about 5-10% lower than the actual value. Srinivasan and Tien (1969) deduced that the
real value of the membrane permeability is in the range of 7.813-8.592 (10-6 g/cm2
sec atm).
A more accurate value of membrane permeability of 8.29 (10-6 g/cm2 sec atm)
was obtained from one experimental data point (crossflow velocity =200cm/sec) by
trial-and-error method. This value was consistent with those reported in previous
research mentioned above (Gill et al, 1966; Srinivasan et al, 1967). The other data
points of Mertens experiments were independent from the determination of the
membrane permeability and were used for model validation. Much better fitness of
the model simulation with the experimental data was obtained with the new
membrane permeability. The deviation was less than 0.7% at all crossflow velocities
except for u0=5.74cm/sec, at which the deviation was about 4.8%.
35
From Figure 3.2, it can be found that compared with the numerical results
from the model developed by Srinivasan et al (1967), the average flux from the
current model is about 1.5% lower if the same membrane permeability (7.610-6
g/cm2 sec atm) is used and the difference becomes larger when the crossflow velocity
increases. This discrepancy may be attributed to the assumptions used by Srinivasan
and co-workers. Their model was derived from boundary layer equations and a
quadratic expression for the concentration profile in the concentration boundary was
assumed. However, the assumption may underestimate the wall concentration because
the concentration profile in the concentration boundary layer is exponential-like
rather than quadratic. When the crossflow velocity increases, the thickness of
momentum and concentration boundary layers would decrease. Therefore, higher
order polynomial terms may be required to make good approximation, i.e., the error
arising from the assumptions used by Srinivasan and co-workers may increase. In fact,
compared with Brians solution (1965), the model developed by Srinivasan et al
(1967) also consistently underestimated wall concentrations and overestimated
permeate flux.
3.4 Effect of meshing scheme on accuracy

Both stability and accuracy have to be considered in numerical simulations
because the stable solution does not necessarily mean an accurate solution. The
accuracy of numerical solutions can be improved essentially with two approaches:
employing higher order polynomials or using very fine (thereby more) meshes in the
region of interest (Chung, 1978; Zienkiewicz and Taylor, 2000) The second approach
is used in our simulations because it can be easily incorporated into adaptive meshing
without modifying the computation code. The concentration gradient is expected to
36
increase towards the membrane surface; therefore, more elements of smaller sizes
should be used near the membrane surface. Figure 3.4 shows several stable wall
concentrations in the abovementioned experimental channel (with a crossflow
velocity of 20cm/sec) with different meshing schemes.
It is obvious that a
converged and stable solution can be erroneous due to the poor meshing schemes.
It was also interesting to find that if the height of elements adjacent to a

membrane surface was not small enough, wall concentration was always
underestimated. This is consistent with the false diffusion reported by Kang and
Chang (1982). The possible reason is that linear isoparametric element used in the
numerical model may underestimate the sharp concentration gradient if the height of
the element is not small enough to capture the sharp gradient. Because in the regions
immediately close to membrane surface, salt concentration is always monotonously
decreasing from the membrane surface to the bulk and concentration in the bulk is
almost a fixed constant, thus, if the concentration gradient (decline speed) is
underestimated, a converged solution would very likely tend to underestimate the
maximum (peak) concentration at membrane surface.
Proper meshing is usually found by trial and error, i.e., meshing the
computation domain with different schemes to find the mesh-independent solution. It
was found from our simulations that if the mesh could not capture the flow direction
transition in the narrow region near the membrane surface, the mesh was not adequate
to simulate accurately wall concentration. As shown in Figure 3.4, the non-uniform
20 (in channel height direction)300 (in crossflow direction) elements can have the
same or even better accuracy than that of the uniform 40300 elements because of
37
improved (although still inadequate) representation of that region. When the nonuniform elements increase to 60 or more in the channel height direction, the solution
becomes independent of the mesh. For the impact of the incremental time step on the
accuracy of the solution, we found it has negligible impact on the results of wall
concentration when Courant number constraint is satisfied.
Figure 3.4 Comparing the accuracy of simulation results due to different meshing
schemes. The solution became independent of the meshing schemes when the nonuniform (exponential) elements increased to 60 and more in the channel height
direction. (crossflow velocity: 20cm/s; membrane permeability: 8.2910-6g/cm2 sec
atm )
It should be pointed out that this flow direction transition region actually is
very thin and usually in the order of less than 10-5m in typical RO channels. The
thickness of this region can be roughly estimated in empty channels by assuming a
parabolic profile for the crossflow velocity. In this way the mesh can be quickly
checked to ensure that the flow direction transition can be captured and thereby
improving the reliability and accuracy of the simulations. Figure 3.5 shows part of a
38
successful mesh scheme near the membrane surface. Figures 3.6 shows the velocity
profiles of flow direction transition layer with a successful mesh scheme in an empty
channel. Under normal operating conditions, the minimum element height should be
in the order of 10-6-10-7 m or lower to obtain a mesh-independent solution.
Figure 3.5 A successful mesh scheme (part) to capture the flow direction transition
near membrane surface
Figure 3.6 Velocity field (part) in the flow direction transition region in an empty
channel (simulation conditions: 2 membranes at y=0 and y=H; p=429psi; A=51012
m/s Pa; c0=9800mg/l; u0=0.01m/s; H=1mm; L=5cm)
39
3.5 Summary
Concentration polarization is induced and controlled by the interactions
between momentum transfer and solute transport in membrane channel, which can
only be accurately described by the coupled Navier-Stokes and convection-diffusion
equations. Any prescribed concentration or permeate flux on the membrane surface
would not be able to reflect concentration polarization in the membrane channel
properly. In this study, a fully coupled streamline upwind Petrov/Galerkin finite
element model was developed for the coupled governing equations for the complex
geometry found in spiral wound RO modules. Several techniques including using
penalty formulation for Navier-Stokes equations and unsteady state governing
equations, were applied to lower the demands of computing resources so that the
model is able to simulate hydrodynamic and salt concentration profiles in domains
with more than 100,000 elements on PCs. This model was verified with published
experimental data of permeate flux in a RO channel.
Meshing quality is critical for a stable and accurate numerical simulation of

concentration polarization in a membrane channel.
One important criterion for
accurate simulation is that the mesh should be able to capture the flow direction
transition from crossflow direction to the direction towards membrane surfaces. Wall
concentration would be underestimated if a mesh fails to capture the flow direction
transition. Although the velocity in the flow direction transition regions is in the order
of permeate velocity, which is far smaller than that of mean crossflow velocity, the
existence of such regions makes the membrane separation process fundamentally
different from the problems of flow and mass transfer in channels with impermeable
walls.
40
Under common operating conditions, most spiral wound RO systems and

some nanofiltration (NF) systems can be accurately simulated by the model for flow
patterns and concentration distribution.
This model may not be applicable to
ultrafilatration (UF) or mircofiltration (MF) systems because the flow conditions in

these systems are usually in turbulent or transitional regimes, which cannot be
accurately described by the model developed.
41
CHAPTER 4 CONCENTRATION POLARIZATION IN

SPACER-FILLED CHANNELS
4.1 Introduction
As a phenomenon inherently associated with membrane separations,
concentration polarization has long been identified as a major problem that
deteriorates the performance of reverse osmosis (RO) systems by reducing permeate
flux and salt rejection. However, this phenomenon is still not well understood
especially in practical systems.
In most practical RO modules, feed spacer is an essential part to keep

membranes apart to form the feed channel. Although the effects of spacer filaments
on hydrodynamic parameters such as velocity, wall stress and pressure drop, have
been studied and reported, a quantitative link between the spacers/filaments and
concentration polarization is still unavailable. This makes it extremely difficult to
design or optimize the spacer for a better system performance.
The SUPG finite element model developed in Chapter 3 makes it possible to

directly simulate salt concentration profiles including the wall concentrations in RO
channels with spacers so that concentration polarization in spacer-filled channels and
the effects of spacer can be quantitatively studied. It does not require unrealistic
assumptions such as constant wall concentration, constant permeate velocity or
impermeable walls. This chapter investigates concentration polarization patterns in
spacer filled RO channels by applying this numerical model. Moreover, the major
42
mechanisms through which spacer filaments affect concentration polarization are

identified and assessed.
Net type spacers are commonly used in commercial spiral would modules. For
RO membrane systems, due to the low cross flow velocity the angle between
longitudinal filaments and transverse filaments is usually 90 and the mean flow
direction is usually parallel to longitudinal filaments as shown in Figure 4.1. In this
study, sodium chloride (NaCl) solution was used for all simulations and the
concentration in the inlet was 32,000 mg/kg. Constant physical properties of the
solution were assumed (D=1.510-9 m2/s, =1.010-6m2/s). Complete (100%) salt
rejection and linear dependency of osmotic pressure on salt concentration
( = k cw , k=75 kPa m3/kg) were used in the simulations. The applied pressure is
5.5106Pa (800psi). The channel height was set as 1mm. Fully-developed parabolic
flow profile was assumed in the inlet. No spacers were placed in the first 5mm and
the last 5mm so that the entrance effect and exit effect due to the spacer filaments
could be minimized. Proper meshing scheme was achieved by trial and error to obtain
mesh-independent solution. A sample (part) of the mesh adjacent to a filament is
illustrated in Figure 4.2.
43
Figure 4.1 Illustration of the spacer configuration
Figure 4.2 Illustration of the mesh adjacent to a cylindrical filament
44
4.2 Velocity profiles in spacer-filled channels

As shown in Figure 4.3, in a spacer-filled channel the velocity profile is
significantly different from the parabolic-like flow as in empty channels. The flow
velocity increases significantly in the narrowed regions (constricted passages)
between the spacer filaments and opposite membrane/wall compared with that in am
empty channel at the same locations. For example, when the averaged inflow velocity
is 0.1m/s, with 0.5mm (in diameter) cylindrical filaments, the maximum velocity
reaches as high as 0.3m/s, which are roughly about 2 times that in an empty
channel.Obviously increasing the filament thickness (height) would result in a higher
maximum flow velocity in such regions. For example, as shown in Figure 4.4 when
the filament thickness is increased to 0.75mm, the corresponding maximum velocity
reaches about 0.56m/s. This elevated maximum flow velocity is mainly caused by the
constricted passage between the filaments and the opposite membrane/wall.
Increasing the filament thickness, i.e., narrowing this constricted passage, would
result in higher velocity in these regions.
Figure 4.3 also shows that the velocity in the regions immediately in front of
and behind the filament near the membrane surface is relatively low and form the
relatively stagnant regions. This is mainly caused by the obstacle effect of spacer
filaments. It can be expected that mass transfer is less efficient in these regions
because of the low flow velocity. This can be clearly observed in salt concentration
profile discussed in Section 4.3.
45
Figure 4.3 Contour of flow velocity in a feed channel (part) with 0.5mm (in diameter) cylinder transverse filaments (simulation conditions:
p=800psi; c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s; h=1mm; lf=4.5mm)
46
Figure 4.4 Contour of flow velocity in a feed channel (part) with 0.75mm (in diameter) cylinder transverse filaments (simulation conditions:
47
Figure 4.5 shows the contour for the x-component velocity in a feed channel.
It can be found that the attached filaments create recirculation regions (wakes) in the
downstream of each filament. The reversed (relative to the cross flow in the bulk)
flow can be as strong as the main cross flow in the bulk in these recirculation regions.
Because of the recirculation formed in the downstream of each filament, the growth
of boundary layer between two filaments is divided into two opposite directions: one
approaching the upstream filaments and the other approaching the downstream
filaments. It was found that increasing filament thickness significantly would increase
these recirculation regions. For example, as shown in Figure 4.6, when increasing the
filament thickness to 0.75mm, the recirculation reaches the downstream filament
under the same simulation conditions.
The detailed flow field in the regions close to the reattachment point is shown
in Figure 4.7. It can be found that the flow in these regions is relatively slow because
flow is separated into two directions here. Actually divergence fields could be clearly
observed in these narrow regions above the reattachment points. Therefore, this small
relatively stagnant flow region is formed.
48
Figure 4.5 Contour of x-component flow velocity in a feed channel (part) with 0.5mm (in diameter) cylinder transverse filaments (simulation
conditions: p=800psi; c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s; h=1mm; lf=4.5mm)
49
Figure 4.6 Contour of x-component flow velocity in a feed channel with 0.75mm (in diameter) cylinder transverse filaments (simulation
50
Figure 4.7 Velocity field near the reattachment point in a feed channel with 0.5mm (in diameter) cylinder transverse filaments (simulation
51
4.3 Major mechanisms of concentration polarization in spacer filled

channels
In empty RO feed channels, salt concentration increases monotonously
downstream and decreases from the membrane surface to the bulk as shown in Figure
4.8. However, the distribution of the salt concentration can be very complicated when
filaments exist. Knowing details of the concentration profile in the feed channel with
filaments is essential for predicting system performance and understanding
concentration polarization in real systems. Unfortunately, this type of information
has not been reported in literature. It can be demonstrated that these salt concentration
distributions are readily simulated with the newly developed finite element model.
Two major mechanisms through which filaments affect concentration polarization are
also identified and assessed.
The distribution of salt concentrations in a channel with filaments attached to

the membrane surface was studied. Figure 4.8 shows the salt concentration profile in
a part of the channel. It was found that salt concentration increases in some regions
and decreased in other regions compared with that in an empty channel under the
same operating conditions. For example, in the low velocity regions adjacent to the
filaments, salt concentration increases significantly compared with that in an empty
channel under the same operating conditions, while salt concentration decreases in
other regions. This phenomenon implies that concentration polarization could be
alleviated in some regions and aggravated in other regions by the existence of spacer
filaments in membrane channel. Therefore, the overall effects of spacer on
concentration polarization and membrane performance (e.g., averaged permeate flux)
in the whole system may be either negative or positive. The outcome depends on
52
several factors including operation conditions, geometric configurations of the feed

channel and the filaments, and membrane characteristics. The impacts of filament
geometry, mesh length and filament configurations on concentration polarization will
be quantitatively discussed in subsequent chapters.
The longitudinal profile of wall concentrations and distribution of permeate

flux in a channel with multi cylindrical (0.5mm in diameter) filaments are plotted
with those in an empty channel respectively, in Figures 4.10 and 4.11. Figure 4.10
shows that there are periodic vibrations (variations) in wall concentrations on both
membranes. The wall concentration on the membrane opposite to the transverse
filaments shows smaller amplitude and is consistently lower than that in the empty
membrane channel. The amplitude of wall concentration variation is much higher on
the membrane attached to the transverse filaments, with the peak values higher and
the valley values lower than those in the empty membrane channel. The patterns of
wall concentration profiles are oppositely reflected in the distribution of the permeate
fluxes, as shown in Figure 4.11. For the membrane opposite to the transverse
filaments, the location of the minimum value of wall concentrations (or the maximum
value of permeate flux) in each cycle is very close to the middle point of each
transverse filaments.
53
Figure 4.8 Salt concentration (c/c0) profiles in an empty feed channel, disproportional in height and length (simulation conditions: p=800psi;
c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s; h=1mm)
54
Figure 4.9 Salt concentration (c/c0) profiles in a feed channel with 0.5mm (in diameter) cylinder transverse filaments (simulation conditions:
55
Figure 4.10 Local variation of wall concentration (cw/c0) in an empty channel and a feed channel with 0.5mm (in diameter) cylindrical transverse
filaments (simulation conditions: p=800psi; c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s; h=1mm; lf=4.5mm)
56
Figure 4.11 Local variations of permeate flux in an empty channel and a feed channel with 0.5mm (in diameter) cylindrical transverse filaments
(simulation conditions: p=800psi; c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s; h=1mm; lf=4.5mm)
57
From Figure 4.10 it can be found that the peak values of wall concentrations
on the membrane attached to the transverse filaments occurs at the contact points of
the filaments to the membrane. This pattern of concentration polarization is caused by
the periodic boundary layer disruption due to the recirculation flow formed in the
regions between the neighboring filaments. Because tangential flow on the membrane
surface is divided into upstream and downstream at a point (reattachment point)
between two filaments, wall concentration and concentration boundary layer grows in
two opposite directions from the reattachment points. As shown in Figure 4.5 in the
upstream of the reattachment points, the recirculation regions form. The reversed flow
in these recirculation regions is critical in forming the special pattern of concentration
polarization in the membrane channel with filaments. The enlarged flow field around
the reattachment point is shown in Figure 4.7. It can be found that the reattachment
point is a division point where the flow breaks into upstream and downstream two
tangential flows along membrane surface. These tangential flows would move the
retained salt toward filaments and contribute peak concentrations on both sides of the
filaments. These peak values are usually far higher than the corresponding wall
concentrations at the same location in an empty channel under the same operating
conditions. This implies concentration boundary layer disruption may aggravate
concentration polarization in some regions when it alleviates concentration
polarization in other regions.
It should be pointed out for the membrane attached to the transverse filaments,
the minimum wall concentration does not concur with the reattachment point. As
shown in Figure 4.12 the location of the minimum wall concentrations, point B, is
actually on the upstream of the reattachment point, point A. In fact, there is a small,
58
yet noticeable peak of wall concentrations at reattachment point. This slightly higher
wall concentration on the reattachment point is very likely caused by stagnant or
relatively slow tangential flow in the regions adjacent to reattachment point as shown
in Figure 4.7. Because of the low flow velocity in this region as discussed in Section
4.2, mass transfer is less efficient in these regions. In addition, the main flow
direction in this region is directly approaching the membrane surface; this would
enhance the concentration buildup near the membrane surface. For these two reasons,
salt accumulation in such regions is more pronounced compared with that in
neighboring regions and a small peak of wall concentrations is observed.
membrane attached to the transverse filaments) in feed channels with 0.5mm (in
diameter) cylindrical filaments (simulation conditions: p=800psi; c0=32,000mg/l;
A=7.310-12m/s Pa; u0=0.1m/s; h=1mm; lf=4.5mm)
Another
major
mechanism
through
which
spacer
filaments
affect
concentration polarization is periodical boundary layer compression. This is clearly

reflected on the longitudinal wall concentration profile of the membrane opposite to
the filaments. As shown in Figure 4.10, for the membrane opposite to the transverse
59
filaments, wall concentrations shows periodic increase and decrease corresponding to

the locations of the filaments, which are attached to the opposite wall. This is mainly
due to the periodical flow acceleration and deceleration corresponding to the
periodical compression and expansion of cross sectional area: when the flow is
approaching a filament, fluid acceleration is observed because of the narrowed
passage; when the flow is leaving a filament, fluid deceleration is observed because
of the expanded passage. Under normal operating conditions for RO/NF systems,
such acceleration and deceleration is unlikely to cause flow recirculation in these
regions; hence, boundary layer disruption does not occur in such regions. Because of
the evaluated velocity in the constricted passage between transverse filaments and the
membrane surface, concentration boundary layer is periodically compressed and
concentration polarization is alleviated in these regions. This velocity pattern results
in two valley values of wall concentrations in each cycle with the valley value in the
downstream consistently higher than that in the upstream. Similar results were
reported in NF system by Geraldes et al (2004). The location of the minimum wall
concentrations in each cycle is very close to the center of the filament, where the
highest velocity is observed as shown in Figure 4.3. Figure 4.10 also shows that, for
the membrane opposite to the transverse filaments, wall concentrations are always
lower than those at the same location in an empty channel under the same operating
conditions. This suggests that unlike concentration boundary layer disruption,
concentration
boundary
layer
compression
always
alleviates
concentration
polarization. However, because the fluid acceleration in the constricted passage is

more pronounced in the center part while concentration polarization layer is close to
membrane surface, the efficiency of such a mechanism may not be as high as
boundary layer disruption.
60
Figure 4.12 also shows there is a small but noticeable secondary peak of wall
concentrations near the downstream filaments (illustrated as point C in Figure 4.12).
This is very likely caused by the small-scale recirculation formed in this region as
shown in Figure 4.13. The results from Schwinge et al (2002balso showed there
was a noticeable peak for wall shear stress in the similar location when lf=8hf, but it
was not discussed in their papers. This small recirculation may disrupt the boundary
layer and cause it to grow in two directions in this narrow region. As shown in Figure
4.13 this recirculation actually enhances wall concentrations buildup in these
recirculation regions. The main flow in the upstream of this circulation is leaving
from membrane surface towards the bulk; therefore, this recirculation would retard
the mass transfer from membrane surface to the bulk.
In some cases this secondary peak may not appear. For example, Figures 4.10
and 4.12 show that there is no such a peak in the upstream of the first filament. This is
probably because the small recirculation in the upstream of the first filament is
suppressed by the fully developed bulk flow. As shown in Figure 4.14 compared with
the bulk flow, the strength and the range of the recirculation in the upstream of the
first filament are not significant. Compared with the recirculation corresponding to
other filaments (Figure 4.13), the recirculation in the upstream of the first filament is
significantly weaker. Therefore, it is difficult to form a noticeable peak of wall
concentrations in the upstream of the first filament. If the neighboring filaments are
separate far enough so that they cannot sense each other, i.e., the flow becomes
fully developed before it encounters the downstream filament, the secondary peak
would not appear.
61
Figure 4.13 Velocity field near the small recirculation regions in a feed channel with
0.5mm (in diameter) cylinder transverse filaments (simulation conditions: p=800psi;
c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s; h=1mm; lf=4.5mm)
In this study it was also found that if the two filaments are close enough so
that the recirculation region can reach the downstream filament, the secondary peak
would also not appear. Figure 4.15 shows the wall concentration profile with a mesh
length of 1.5mm. It can be found that there is not a secondary peak in the upstream of
each filament, possibly because the recirculation region could reach the downstream
filament when mesh length is 1.5mm, as shown in Figure 4.16. In this case the
secondary recirculation in the upstream of the downstream filaments is negligible or
even disappeared as shown in Figure 4.17. As a result there were no secondary peaks
of wall concentrations.
Figure 4.14 Velocity field in the upstream of the first filament in a feed channel with
A=7.310-12m/s
Pa;
u0=0.1m/s;
h=1mm;
lf=4.5mm
c0=32,000mg/l;
62
Figure 4.15 Local wall concentration (cw/c0) profiles (on the membrane attached to the transverse filaments) in feed channels with 0.5mm (in
diameter) cylindrical filaments (simulation conditions: p=800psi; c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s; h=1mm; lf=1.5mm)
63
Figure 4.16 Contour of x-component flow velocity in a feed channel (part) with 0.5mm (in diameter) cylinder transverse filaments (simulation
64
Figure 4.17 Velocity field in the upstream region of a filament in a feed channel with
c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s; h=1mm; lf=1.5mm)
4.4 Summary
This chapter attempted to investigate concentration polarization patterns in
spacer filled RO channels and to assess the major mechanisms of concentration
polarization in spacer-filled RO channels. It was found spacer filaments in the feed
channel substantially alter hydrodynamic conditions and therefore, changed salt
concentration profiles and concentration polarization in the channel. For example,
recirculation region and reversed flow are often formed in the downstream of each
filament. The periodical constricted passage between the filaments and the opposite
membrane/wall also results in periodical fluid acceleration and deceleration in these
regions.
Concentration boundary layer disruption and compression are the two major
mechanisms that the filaments affect concentration polarization in the feed channel.
Concentration boundary layer disruption corresponds to the recirculation formed in
the downstream of each filament and occurs on the membrane with the transverse
65
filaments attached. Concentration boundary layer compression corresponds to the

periodical fluid acceleration and deceleration in the periodical constricted passage
between filaments and the opposite membrane, and occurs on the membrane opposite
to the transverse filaments.
Concentration boundary layer disruption usually results in two peaks of wall

concentration because concentration boundary layer grows in two opposite directions
from the reattachment points toward the upstream and downstream filaments. These
peak values are usually far higher than those in an empty channel at the same location
under the same operating conditions. This means concentration boundary layer
disruption may aggravate concentration polarization in some regions although it
alleviates concentration polarization in other regions. The valley value of wall
concentration between two neighboring filaments is usually observed in the upstream
of the reattachment point. In some cases secondary peak of wall concentration in the
upstream of each filament may be observed because of the small recirculation formed
there. When neighboring filaments are separated far enough so that they cannot
sense each other or, two filaments are close enough so that the recirculation region
reaches the downstream filament, this secondary peak does not appear. Although
there are peak wall concentrations around the filaments, the overall effect of the
spacers is to alleviate concentration polarization and to increase permeate flux in the
membrane channel.
Concentration boundary layer compression results in two valley values of wall

concentrations in each cycle corresponding to the center of the two neighboring
filaments that are attached to the opposite membrane/wall and the valley value in the
66
downstream is consistently higher than that in the upstream. This means

concentration
boundary
layer
compression
always
alleviates
concentration
polarization and results in lower wall concentrations compared with that in an empty
channel. However, the efficiency of this mechanism is unlikely to be as high as
boundary layer disruption.
As shown in this study, the geometry and arrangement of the filaments may
have notice impact of concentration polarization and system performance. This will
be discussed in Chapters 5 and 6.
67
CHAPTER 5 IMPACT OF FILAMENT GEOMETRY ON

CONCENTRATION POLARIZATION
5.1 Introduction
Concentration polarization in RO channels deteriorates the performance of the
membrane processes by reducing the effective driving pressure. Concentration
polarization is also closely related to membrane fouling because foulants are
generally accumulated in the concentration polarization layer. It was revealed in
Chapter 4 that concentration polarization was significantly affected by the spacer
filaments. This chapter studies the effects of filament geometry on concentration
polarization using the 2-D SUPG model developed earlier. It is expected that this
study will lead to a better understanding of the role of filament geometry on
concentration polarization alleviation in spiral wound RO modules for more realistic
conditions.
Filaments of two typical shapes (cylinder and square bar filaments) and three
different filament thickness (0.25, 0.5, and 0.75mm) were investigated for their
impact on concentration polarization alleviation in the feed channel. The channel
height and length used in this numerical study were 1mm and 10cm, respectively.
Constant parameters typical to sodium chloride solution were used in all simulations,
which were diffusivity D=1.510-9 m2/s and viscosity =1.010-6m2/s. Complete
(100%) salt rejection and linear dependency of osmotic pressure on salt concentration
( = k cw , k=75 kPa m3/kg) were assumed. Fully developed parabolic flow profile
was assumed in the inlet. No spacers were placed in the first 5mm and the last 5mm
68
of the channel so that the entrance effect and exit effect due to the spacer filaments
could be minimized. The computing domain is illustrated in Figure 5.1.
Figure 5.1 Illustration of the computing domain
5.2 Filament shape

To compare the effects of filament shapes on concentration polarization,
cylindrical and rectangular bar filaments were used in the study. The diameter of the
cylindrical filaments was 0.5mm. The thickness (height) and width (longitudinal
direction) of the rectangular bar filaments were 0.5mm and 0.392mm respectively.
This makes that the cross sectional area is identical for these two types of filaments;
and therefore the overall parameters to describe filaments in a channel, e.g., porosity
or voidage, are also identical. The distance between two neighboring filaments was
set as 4.5mm.
Figure 5.2 shows the salt concentration profile in the feed channel with
rectangular bar filaments. It can be found that salt concentration increases in some
regions and decreases in other regions compared with that in an empty channel under
the same operating conditions (Figure 4.8). The concentration profile is very similar
to that in the channel with cylindrical filaments (Figure 4.9). This suggests that
filament shape does not have a significant influence on the overall patterns of salt
concentration profiles. This can be explained by the similar flow patterns for these
69
two spacer filaments. The filament shape does not change the overall flow patterns,
i.e., the main flow structures, and the recirculation in the downstream of each filament
and the periodical fluid acceleration and deceleration in the constricted passage are
similar for these two filament shapes (Figure 5.3 and Figure 4.3). Comparing Figure
4.10 with Figure 5.4, it can be found that the overall oscillation pattern of wall
concentration is also similar for cylindrical and rectangular bar filaments.
Although the overall salt concentration profiles are similar for cylindrical and
rectangular bar filaments, there are significant differences in the peak wall
concentrations for the membrane attached to transverse filaments. Figure 5.5 shows
that the peak values of wall concentrations are much higher with cylindrical filaments
than that with bar filaments under the same operating conditions. In fact, these peak
wall concentration values for cylinder filaments are very close to the wall
concentration upper limit corresponding to the applied pressure. The much higher
wall concentration near the contact point of cylindrical filaments with membrane is
very likely caused by the stagnant cavity formed adjacent to the contact point as
shown in Figure 4.13. For the rectangular bar filaments, most of the regions
corresponding to this stagnant cavity adjacent to the cylindrical filament are blocked
by the filaments and therefore the peak wall concentration is smaller. This implies
that for cylinder filaments the increased membrane area may not significantly
contribute to membrane productivity because of high salt concentrations in these
regions. Numerical simulations revealed that the average permeate flux for a 10cm
membrane channel with 20 cylindrical filaments is about 5.4% lower than that with
rectangular bar filaments with identical cross sectional area, although the net
membrane surface with cylindrical filaments is about 8% more than that with
70
rectangular bar filaments. Figure 5.5 also shows the wall concentrations are generally
slightly lower for rectangular bar filaments than for cylindrical filaments. This
suggests that the rectangular bar filaments may create slightly stronger recirculation
(reversed flow) than the cylindrical filaments.
Figure 5.6 compares wall concentrations on the membrane opposite to the

transverse filaments. Similarly, the wall concentrations with rectangular bar filaments
are slightly yet consistently lower than those with cylindrical filaments. This is
possibly due to the fact that most part of the constricted passage between the filament
and the opposite membrane is slightly narrower for the rectangular bar filaments than
for the cylindrical filaments because of the differences in the shape. Consequently,
the fluid acceleration and the resulting concentration boundary layer compression in
this region are slightly more significant for square bars filaments than for cylindrical
filaments. This shows that for the membrane opposite to the transverse filaments, the
square bar filaments are slightly more effective in reducing concentration polarization
than the cylindrical filaments. Despite the slight differences in concentration profile,
the effects of the filament shape on permeate flux of the membrane opposite to the
transverse filaments are quite limited and the difference in permeate flux (averaged in
10 cm long channel) is less than 2% under the simulation conditions.
71
Figure 5.2 Salt concentration (c/c0) profiles in a feed channel with 0.5 (thickness)0.392 (width)mm rectangular bar transverse filaments
72
Figure 5.3 Contour of flow velocity in a feed channel (part) with 0.5 (thickness)0.392 (width)mm rectangular bar transverse filaments
73
Figure 5.4 Wall concentration (cw/c0) profiles in an empty channel and a feed channel with 0.5 (thickness)0.392 (width)mm rectangular bar
transverse filaments (simulation conditions: p=800psi; c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s; h=1mm; lf=4.5mm)
74
Figure 5.5 Comparison of local wall concentration (cw/c0) profiles (on the membrane
attached to the transverse filaments) in feed channels with 0.5mm (in diameter)
cylindrical filaments and 0.392mm0.5mm (thickness) rectangular bar filaments
(simulation conditions: p=800psi; c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s;
h=1mm; lf=4.5mm)
opposite to the transverse filaments) in channels with 0.5mm (in diameter) cylindrical
filaments and 0.392mm0.5mm (height) rectangular bar filaments (simulation
conditions: p=800psi; c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s; h=1mm;
lf=4.5mm)
The slightly higher permeate flux with rectangular bar filaments was obtained
at the cost of slightly higher pressure drop in the channel than cylindrical filaments. In
the 10cm long channel with 20 cylindrical filaments (0.5mm in diameter), the
75
pressure drop is about 450 Pa (cross flow velocity was 0.1m/s), which is about 15%
lower than that with 20 rectangular bar filaments (0.392mm in width and 0.5mm in
thickness). This implies membrane systems with rectangular bar filaments may
consume slightly more energy than those with cylindrical filaments with identical
voidage.
To study the influence of filament width on concentration polarization, wall

concentrations of the membrane attached to and opposite to the transverse filaments
in channels with 0.5mm0.5mm square bar filaments and channels with
0.392mm0.5mm rectangular bar filaments are compared in Figures 5.7 and 5.8. It
can be found that for these two types of filaments, the overall patterns of
concentration polarization, e.g., the location of valley value, the amplitude of the
valley value and the location of the secondary peak, are almost identical. For the
membrane attached to the transverse filaments (Figure 5.7), slight difference in the
amplitude of wall concentration peak value is noted. The 0.392mm0.5mm filaments
result in a slightly yet consistently higher peak values of wall concentrations. This is
very likely caused by the slightly longer concentration boundary layer growing
distance for the case with 0.3920.5mm filaments. The locations and the amplitude of
the valley values are very close for these two filaments. This implies that slightly
reducing the filament width does not have significant impact on concentration
polarization. This is consistent with the analysis of separation flow due to an
obstruction (Nallasamy, 1986). For the membrane opposite to the transverse filaments,
the wall concentration does not have significant differences. This suggests that
slightly reducing filament width would not affect concentration polarization
significantly.
76
attached to the transverse filaments) in feed channels with 0.50.5mm square bar
filaments and 0.392mm0.5mm (thickness) rectangular bar filaments (simulation
conditions: p=800psi; c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s; h=1mm;
lf=4.5mm)
opposite to the transverse filaments) in channels with 0.50.5mm square bar filaments
and 0.392mm0.5mm (thickness) rectangular bar filaments (simulation conditions:
5.3 Filament thickness

The impact of filament size thickness on concentration polarization was
studied by varying the thickness (height) of rectangular bar filaments from 0.25mm to
77
0.75mm. The width of these filaments was set as 0.5mm so that the longitudinal
distribution of wall concentration could be compared on all locations.
Figures 5.9, 5.10 and 5.11 show the salt concentration profiles in the feed
channel (part) with 0.25 (thickness)0.5mm (width), 0.50.5mm and 0.75(thickness)
0.5mm (width) filaments respectively. It can be found that the overall salt
concentration distribution patterns are very similar. The two major mechanisms
through which filaments affect salt concentration remain the same for all these three
cases: concentration boundary layer disruption and concentration boundary layer
compression. However, it shows that increasing filament thickness would
significantly alleviate concentration polarization. The reason is that increasing
filament thickness would result in stronger recirculation in the downstream of each
filament and stronger fluid acceleration in the constricted passage. For example, when
the filament thickness is 0.25mm, the recirculation only stretches about 1mm in the
downstream of each filament as shown in Figure 5.12. This value increases to about
3mm if filament thickness increases to 0.5mm as shown in Figure 5.13. When
filament thickness is 0.75mm, the recirculation reached the downstream filament as
shown in Figure 5.14.
Figure 5.15 compares wall concentrations on the membrane opposite to the

transverse filaments. It shows that the size of filament has significant effects on
concentration polarization near this membrane. Compared with the wall
concentrations in the empty channel (10cm in length) under the simulation conditions,
the averaged wall concentrations on the membrane opposite to the transverse
filaments are reduced by 3.4% and 31% with the 0.25mm and 0.75mm filaments,
78
respectively. The corresponding increases in permeate flux are 3.9% and 36%.
Concentration polarization alleviation on the membrane opposite to the transverse
filaments is mainly achieved by the periodical acceleration and deceleration of the
feed flow. When the filament thickness is large, the corresponding constricted
passage between the filaments and the opposite membrane would narrow. Therefore,
the acceleration in the narrow regions and decelerations afterwards is enhanced. For
example, as shown in Figure 5.16, the maximum velocity is about 0.18m/s when the
filament thickness is 0.25mm; this maximum velocity increases to 0.31m/s (Figure
5.17) and 0.61m/s (Figure 5.18) when the filament thickness increases to 0.5mm and
0.75mm respectively. It should be noted that although filament thickness has
significant impact on the amplitude of wall concentrations, it does not have notable
effects on the overall patterns of concentration polarization for the membrane
opposite to the transverse filaments. For example, as shown in Figure 5.15, the
oscillation pattern of wall concentrations corresponding to different filament
thickness is almost identical although the amplitude varies sharply. For the membrane
opposite to the transverse filaments, the major mechanism affecting concentration
polarization by filaments is concentration boundary compression. Although
increasing filament thickness would result in a narrower constricted passage and
significantly increased flow velocity and gradient, it does not alter the flow patterns in
these regions as shown in Figures 5.16, 5.17 and 5.18. Therefore, increasing filament
thickness does not affect the oscillation pattern of wall concentrations.
79
80
Figure 5.10 Salt concentration (c/c0) profiles in a feed channel with 0.50.5 mm square bar transverse filaments (simulation conditions:
81
82
Figure 5.12 Contour of x-component flow velocity in a feed channel (part) with 0.25(thickness)0.5mm (width) rectangular bar transverse
83
Figure 5.13 Contour of x-component flow velocity in a feed channel (part) with 0.50.5mm square bar transverse filaments (simulation
84
Figure 5.14 Contour of x-component flow velocity in a feed channel (part) with 0.75(thickness)0.5mm (width) rectangular bar transverse
85
Figure 5.15 Comparison of local wall concentration (cw/c0) profiles (on the membrane opposite to the transverse filaments) in channels with
square bar filaments with different filament sizes (0.25mm, 0.5mm and 0.75mm in thickness and 0.5mm in width) (simulation conditions:
86
Figure 5.16 Contour of flow velocity in a feed channel (part) with 0.25 (thickness)0.5 (width)mm rectangular bar transverse filaments
87
Figure 5.17 Contour of flow velocity in a feed channel (part) with 0.50.5 mm square bar transverse filaments (simulation conditions:
88
Figure 5.18 Contour of flow velocity in a feed channel (part) with 0.75 (thickness)0.5 (width) mm rectangular bar transverse filaments
89
Figure 5.19 shows that filament thickness also has a strong impact on wall
concentration profile on the membrane attached to transverse filaments. A
0.250.5mm filament is unable to cause a very strong recirculation flow between the
filaments as shown in Figure 5.12 and therefore, the minimum wall concentration
occurs close to the upstream filament. On the other hand, a strong recirculation flow
that stretches to the downstream filament could be induced by a 0.750.5mm filament
as shown in Figure 5.14. As a result, the location of the minimum wall concentrations
shifts to the downstream filament. Therefore, increasing filament thickness can shift
the location of the valley value of wall concentrations from near upstream filament to
near downstream filament.
As discussed in section 4.3, for the membrane attached to the transverse

filaments, the major mechanism for concentration polarization alleviation by
filaments is concentration boundary layer disruption. Therefore, between two
neighboring filaments, concentration boundary layer grows in two opposite directions
toward two neighboring filaments and thus there are two peaks in the wall
concentration profile near the upstream and downstream filaments. Figure 5.19 and
5.20 show these peak values are significantly affected by filament thickness. For
example, the reattachment point for 0.25mm (thickness) filament is about 1mm from
the upstream filaments but it is about 3mm for 0.5mm (thickness) filament. Therefore,
the length for the concentration boundary layer to develop in the reversed flow region
of 0.25mm filament is shorter than that of the larger filaments. This leads to a lower
peak of wall concentrations near the upstream filaments (location A in Figure 5.20)
with smaller filaments than with larger filaments. The 0.75mm (thickness) filaments
could create recirculation in the whole region between the two neighboring filaments
90
and result in the highest peak values near the upstream filament, the lowest peak
values near the downstream filaments (location B in Figure 5.20) and the lowest
average salt concentration in the whole interval.
The peak values of wall
concentration near the downstream filament do not show significant difference for the
0.25mm and 0.5mm filaments. This is probably related to the small secondary peak
(location C in Figure 5.20) in wall concentration profile of the 0.5mm filament.
Hence, increasing filament thickness can result in higher wall concentration adjacent
to the upstream filaments because of longer concentration boundary layer growth
distance.
In addition, increasing filament thickness was found to alleviate concentration

polarization more significantly although higher wall concentrations were observed
adjacent to the upstream filament. For example, for the membrane attached to the
transverse filaments, the averaged permeate flux increases about 10% (compared with
that in an empty channel) in a 10cm long channel with 20 0.25 mm0.5mm filaments.
The increase is about 17% and 21% with same number of 0.5mm0.5mm and
0.75mm0.5mm filaments, respectively. These flux improvements suggest that
increasing filament thickness can result in better concentration polarization alleviation
for the membrane opposite to the transverse filaments than that for the membrane
attached to transverse filaments.
91
Figure 5.19 Comparison of local wall concentration (cw/c0) profiles (on the membrane opposite to the transverse filaments) in channels with
rectangular bar filaments with different filament sizes (0.25mm, 0.5mm and 0.75mm in thickness and 0.5mm in width) (simulation conditions:
92
Figure 5.20 Enlarged view of local wall concentration (cw/c0) profiles (on the membrane opposite to the transverse filaments) in channels with
rectangular bar filaments with different filament sizes (0.25mm, 0.5mm and 0.75mm in thickness and 0.5mm in width) (simulation conditions:
93
As shown in Figure 5.20, the secondary peaks do not appear when the
thickness of filaments increases to 0.75mm. In this case the recirculation regions
formed in the downstream of each filament could reach the downstream filament (as
shown in Figure 5.14), which would effectively surpass or even eliminate the
possibility of the small recirculation in the front of the downstream. It can be
predicted that for the 0.5mm0.5mm filaments, there would be no secondary
concentration peaks if the mesh length decreases so that the recirculation in the
downstream of each filaments reaches the downstream filament. To verify this
hypothesis, simulations were carried out for the 0.5mm0.5mm filaments with a
reduced mesh length of 2.5mm and the results are shown in Figure 5.21. There are no
secondary peaks and the peak concentration values near the downstream filaments are
much lower than those near the upstream filaments.
Figure 5.21 Local wall concentration (cw/c0) profiles (on the membrane attached to
the transverse filaments) in feed channels with 0.5mm square bar filaments with
reduced mesh length (simulation conditions: p=800psi; c0=32,000mg/l; A=7.31012
m/s Pa; u0=0.1m/s; h=1mm; lf=2.5mm)
94
Figure 5.19 also shows that there are no secondary peaks if the filament
thickness is 0.25mm. The possible explanation is that the disturbance on the flow
induced by the 0.25mm filament is so small that the flow would be restored to the
fully developed state before it encounters the next filament. Actually, the recirculation
in the downstream of each filament stretches only about 1mm in the downstream as
shown in Figure 5.12. This is very similar to the case of the first filament as
discussed in 4.3. Therefore, secondary peak can be eliminated by either increasing or
decreasing mesh length. However, it should be noted that increasing mesh length
would reduce the effectiveness of boundary layer disruption and decreasing mesh
length would result in a higher pressure loss. As this secondary peak is undesirable in
alleviating concentration polarization, proper choice of mesh length based on filament
size and cross flow velocity is important to eliminate this peak to improve system
performance. A possible impact of the secondary peaks is to promote precipitation of
some sparingly soluble salts such as CaSO4 onto membrane if they are present in the
feed water although their impact on permeate flux is negligible.
It should be pointed out that higher filament thickness would cause higher
pressure drops than smaller ones. Under the simulation conditions, the pressure drop
in a 10cm long channel with 20 0.75mm (thickness)0.5mm rectangular bar filaments
is about 3200 Pa, which is about 6 times of the pressure drop in the same channel
with 0.5mm0.5mm square bar filaments and about 16 times of that with 0.25mm
(thickness)0.5mm bar filaments. This is due to energy dissipation caused by the
drastic change in flow velocity in the membrane channel. Higher filament thickness
can result in more pronounced changes in flow velocity and direction. This finding
suggests that pressure drop along the membrane channel may become a concern when
95
large filaments are used to alleviate concentration polarization in spiral wound RO

modules in some cases.
5.4 Summary
In this chapter, the effects of filament geometry on concentration polarization
in RO channels were studied by investigating the effects of filament shape and
filament thickness.
Filament shape has noticeable impact on concentration polarization in RO

channels. This impact is relatively small for the membrane opposite to the transverse
filaments. However, for the membrane attached to the transverse filaments, the
cylindrical filaments usually result in extremely high wall concentration peaks close
to the filaments. This suggests that cylindrical filaments may facilitate more severe
membrane fouling near filaments than the bar filaments. It was also found that the
membrane channel with cylindrical filaments produces noticeably lower permeate
flux than that with rectangular bar filaments of the identical voidage under the same
operating conditions, even though cylindrical filaments block less membrane area.
Decreasing width of the rectangular bar filaments may slightly increase the peak
values of wall concentrations on the membrane attached to the transverse filaments.
Although filament thickness does not change the overall patterns of salt
concentration distribution in the feed channel, it has very significant effects on
concentration polarization. The larger filaments are more effective in concentration
polarization alleviation and permeate flux enhancement although they usually induce
higher peak wall concentration near the upstream filaments on the attached membrane.
96
Increasing filament thickness usually results in a better concentration polarization

alleviation for the membrane opposite to the transverse filaments than that for the
membrane attached to transverse filaments. However, the improvement of system
performance by the large filaments is obtained at the cost of drastically increased
pressure drop along the membrane channel. The effect of the spacer on alleviating
concentration polarization may be deteriorated by the occurrence of a small secondary
recirculation in the front of filaments, which can be eliminated by a proper choice of
the interval between two neighboring filaments of given filament thickness. The
results suggest that the size (thickness) and mesh length are the two important
parameters in optimizing spacer design for the membrane channel configuration
considered in this study. The effects of mesh length and different filament
configurations on concentration polarization and membrane performance will be
discussed in Chapter 6.
97
CHAPTER 6 FILAMENT CONFIGURATION AND

MESH LENGTH ON CONCENTRATION
POLARIZATION AND MEMBRANE PERFORMANCE
6.1 Introduction
Permeate flux in a reverse osmosis (RO) system is affected by salt
concentration buildup near the membrane surface, i.e., concentration polarization,
which is an inherent phenomenon in membrane separation systems. Such negative
effects become more pronounced when the high permeability RO membranes are
commonly used in most practical systems nowadays. Quantifying the impact of feed
spacer on concentration polarization not only enables a more accurate calculation of
permeate flux in practical systems, but also provides technical foundations to
optimize spacer design for a better system performance.
It has been observed that permeate flux in UF systems can be increased up to

several folds by spacer arrangement optimization (Levy and Earle, 1994; Schwinge et
al, 2004a). However, the effects of the spacer on concentration polarization and
membrane performance have rarely been reported for spiral wound RO modules in
literature. The flow regime in a RO system is quite different from that in UF or even
NF systems because of much lower membrane permeability and crossflow velocity.
This leads to a different pattern of concentration polarization in spiral wound RO
modules and the spacers may affect concentration polarization and permeate flux
differently.
98
In this chapter the effects of filament configurations (e.g., cavity, zigzag and
submerged) and mesh length (longitudinal distance between two neighboring
filaments), which are the two key factors to characterize or identify spacers, are
studied with the fully coupled streamline upwind Petrov/Galerkin (SUPG) finite
element model. To achieve this, concentration polarization, permeate flux, pressure
loss in a typical RO membrane channel with different mesh lengths, three different
configurations (cavity, submerged and zigzag) were investigated.
In the simulations, the feed channel height and length were set as 1mm and
10cm respectively. Square bar filaments of 0.5mm0.5mm were used because this
geometry provides fixed contact area with membrane surface. No spacers were placed
in the first 5mm and the last 5mm so that the entrance effect and exit effect due to the
spacer filaments could be minimized. Fully developed parabolic flow profile of
0.1m/s averaged crossflow velocity was used in the inlet. It was assumed that all
transverse filaments are perpendicular to longitudinal filaments and to the crossflow
direction. This configuration is quite common in RO modules because of low
crossflow velocity in these systems. The computing domain is illustrated in Figure
6.1.
Figure 6.1 Illustration of the computing domain (part)
99
Constant physical properties typical to sodium chloride solution were assumed

(D=1.510-9 m2/s, =1.010-6m2/s). Complete (100%) salt rejection and linear
dependency of osmotic pressure on salt concentration ( = k cw , k=75 kPa m3/kg)
were used in the simulations. Simulations were conducted for three different RO
systems as summarized in Table 6.1.
Mesh lengths investigated were 0 (for
submerged configurations only), 0.5mm (for zigzag configurations), 1mm, 1.5mm,

2.5mm, 4.5mm, 8mm and 13mm.
Table 6.1 Membrane properties and operating conditions
Seawater
High-salinity brackish Low-salinity brackish
desalination
water desalination
water desalination
A (m/s Pa)
7.310-12
2.310-11
2.310-11
p (psi)
800
270
225
C0 (mg/kg)
32,000
6,000
2,000
U0 (m/s)
0.1
0.1
0.1
100
6.2 Concentration polarization patterns for different filament

configurations
As discussed in Chapter 4 a filament attached to a membrane can affect
concentration boundary layer in a membrane channel in two ways. For the membrane
attached to the transverse filaments, concentration boundary layer is disrupted
periodically by the filaments and the induced recirculation. In an interval between
two neighboring filaments, the concentration boundary layer grows in two opposite
directions: one approaching the upstream filaments due to reversed flow and the other
approaching the downstream filaments.
For membrane region opposite to the
transverse filaments, concentration boundary layer is periodically compressed by the

constricted passage between transverse filaments and the opposite membrane.
Therefore, concentration polarization would be alleviated in these regions.
The concentration contours in Figures 6.2 and 6.3 shows that both
concentration boundary layer disruption and compression by the transverse filaments
would play their roles in the cavity and zigzag configurations. The difference is that
the disruption occurs on one membrane and compression occurs on the other for the
cavity configuration, while disruption and compression occurs alternatively on both
membranes for the zigzag configuration.
101
Figure 6.2 Salt concentration (c/c0) profiles in the feed channel (part) with cavity spacer (simulation conditions: p=800psi; c0=32,000mg/l;
A=7.310-12m/s Pa; u0=0.1m/s; H=1mm; lf=4.5mm)
102
Figure 6.3 Salt concentration (c/c0) profiles in the feed channel (part) with zigzag spacer (simulation conditions: p=800psi; c0=32,000mg/l;
103
Figure 6.4 shows that for submerged configuration there are no concentration
boundary layer disruptions because the transverse filaments are not attached to the
membranes. The flow profiles in Figure 6.5 show that the recirculation flow formed
in the downstream of each filament is only restricted in the central part of the channel
and no reversed flow is developed near the membrane surface. Therefore, the only
mechanism for concentration polarization relaxation in submerged configuration is
concentration boundary layer compression due to the periodical constricted passages.
As shown in Figure 6.6 the elevated flow velocity field is formed in the constricted
passage between the filaments and the opposite membranes.
Figure 6.7 shows the wall concentrations for the three spacer configurations.
The dotted line is the wall concentration in an empty membrane channel, which will
serve as the baseline for elaborating the effectiveness of the spacers of different
configurations.
It can be seen that the peak wall concentrations in a small
neighborhood of an attached filament are higher than the baseline. However, the wall
concentrations in most of the region between two filaments are much lower than the
baseline. While for membranes without attached transverse filaments, although the
wall concentrations are always lower than the baseline, the extents of concentration
reduction are much smaller.
104
Figure 6.4 Salt concentration (c/c0) profiles in the feed channel (part) with submerged spacer (simulation conditions: p=800psi; c0=32,000mg/l;
105
Figure 6.5 Contour of X-component flow velocity in the feed channel (part) with submerged spacer (simulation conditions: p=800psi;
c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s; H=1mm; lf=4.5mm)
106
Figure 6.6 Contour of flow velocity in the feed channel (part) with submerged spacer (simulation conditions: p=800psi; c0=32,000mg/l;
107
Figure 6.7 Comparison of local wall concentration (cw/c0) profiles in feed channels
with cavity, zigzag and submerged spacers (simulation conditions: p=800psi;
As shown in Figure 6.7 for zigzag configurations, the valley value occurs on
the location opposite to the filament, where a constricted passage is formed by the
filament on the opposite side. As shown in Figure 6.8 significant fluid acceleration is
observed in this region and therefore concentration boundary layer can be compressed.
It can be found that compared with cavity and submerged configurations, zigzag
configuration results in lower wall concentrations in most areas. This implies that
combined effect of concentration boundary layer disruption and compression is more
efficient in alleviating concentration polarization than individual mechanism alone.
108
Figure 6.8 Contour of flow velocity in the feed channel (part) with zigzag spacer (simulation conditions: p=800psi; c0=32,000mg/l; A=7.310m/s Pa; u0=0.1m/s; H=1mm; lf=4.5mm)
12
109
Figure 6.9 Contour of x-component flow velocity in the feed channel (part) with zigzag spacer (simulation conditions: p=800psi;
110
Figure 6.7 also shows that for zigzag configurations, there is also a noticeable
secondary valley value corresponding to boundary layer disruptions. The location is
closely related to the reattachment point of the recirculation in the downstream of
upstream filaments. For example, in the region between 50mm~55mm from the inlet,
this secondary valley value is observed at the location of about 52.5mm from the inlet,
which is very close to the recirculation boundary (reattachment point) as shown in
Figure 6.9.
From the above discussions it is obvious that the impact of concentration

boundary layer compression is generally less effective on concentration polarization
alleviation than concentration boundary layer disruption. The combination of
concentration boundary layer compression and disruption can result in the greatest
concentration polarization alleviation, as in the case of zigzag configurations. For
submerged configurations, concentration polarization alleviation and permeate flux
enhancement are mainly achieved by concentration boundary layer compression. This
also applies to the membrane opposite to the transverse filaments in cavity
configurations. For cavity configurations, if the membrane is attached to the
transverse filaments, concentration polarization and permeate flux would be mainly
affected by concentration boundary layer disruption. For zigzag configurations these
two mechanisms are combined.
6.3 Filament configuration on membrane performance

As discussed in 6.2, concentration polarization patterns are strongly affected
by filament configurations. It is expected that filament configurations must have
considerable impacts on membrane performance, e.g., permeate flux and pressure loss.
111
If the transverse filaments are attached to a membrane surface in cavity and zigzag
configurations, concentration polarization is aggravated in the stagnant regions
adjacent to filaments. Wall concentrations in these regions are usually much higher
than those in an empty channel on the same location. This implies spacer may
deteriorate membrane performance (permeate velocity) in these locations. Moreover,
the impermeable filaments block noticeable membrane surfaces and further reduce
membrane productivity. For example, 20 0.5mm0.5mm square bar filaments in a
10cm channel blocks about 5% membrane area. Therefore, if concentration
polarization alleviation is not strong enough to surpass the negative impact of the
membrane surface blockage and concentration buildup adjacent to the filaments,
permeate flux may be not increased significantly or even be reduced by the spacers.
The localized permeate fluxes in the membrane channel typical for seawater
desalination with different spacer configurations are plotted in Figure 6.10 with that
in the empty channel as a comparison baseline. On average, the existence of all types
of spacers elevates permeate flux in the membrane channel though permeate velocity
is lower than the baseline in small regions close to the filaments for zigzag and cavity
configurations. This indicates the possibility of reducing membrane productivity by
improper design of spacers.
It can also be found that permeate rate is significantly affected by different

filament configurations. For example, as shown in Figure 6.10, with a 4.5mm mesh
length, membrane productivity in the 10cm long channel increases by about 16% and
34% over that in an empty channel for cavity configurations and zigzag
configurations,
respectively.
For
submerged
configurations,
concentration
112
polarization relaxation is less significant compared to the other configurations and

permeate flux increase is about 14%. The main reason could be that concentration
boundary layer disruption does not occur for submerged configurations and the
concentration boundary layer compression is not very strong when mesh length is at
that value (4.5mm). Therefore, membrane productivity boost is less significant even
though there is about 5% more effective membrane area with submerged
configurations than the other two configurations.
Figure 6.10 Comparison of local permeate velocity profiles in 10cm long feed
channels with cavity, zigzag and submerged spacers (simulation conditions:
p=800psi; c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s; H=1mm; lf=4.5mm)
113
Figure 6.10 shows that zigzag configuration results in significantly higher

permeate flux enhancement than the other two configurations do. This implies that
the alternative occurrence of concentration boundary layer disruption and
compression on one membrane is more effective in concentration polarization
alleviation and flux enhancement than either consecutive disruptions or consecutive
compressions on one membrane.
From Figure 6.10 it can also be found that the local variations of permeate
flux are drastically affected by spacer configurations. For zigzag configuration, two
peak values of permeate velocity are observed between two consecutive filaments:
one peak close to the reattachment point of the recirculation in the downstream of the
upstream filament and the other higher peak in the location opposite to the filament
attached to the opposite wall/membrane. Both peaks are significantly higher than the
corresponding peaks for cavity configurations. Submerged configurations result in the
lowest peak values of permeate flux.
Simulation results also revealed that the pressure loss for submerged spacers is
the highest among the three configurations in otherwise similar conditions. For
example, in the simulation of a 10cm long channel with a 4.5mm mesh length and 20
filaments, the pressure drop is about 1.2KPa for the submerged configuration, which
is about 2.3-2.4 times of the pressure drop for zigzag and cavity configurations. This
pressure loss difference is directly related to the velocity profile in channel height
direction. For submerged configurations, the parabolic velocity distribution are
periodically obstructed by the filaments in the center part where velocity is the
114
highest in the channel, and this would cause elevated viscous and form drags
compared with these for cavity or zigzag configurations.
6.4 Impact of mesh length on membrane performance

6.4.1 Mesh length on permeate flux
The impact of the mesh length of three types of spacers on averaged permeate
flux were simulated under conditions typical for desalination of seawater, high
salinity brackish water, and low salinity brackish water, respectively (Table 6.1).
Figure 6.11 shows that mesh length affects permeate flux significantly for all three
spacers in the seawater desalination case. The impact of mesh length shows different
patterns for different spacer configurations and zigzag configurations usually results
in better flux enhancement in most cases. The same phenomenon was observed in
low-salinity and high-salinity brackish water desalination cases. This implies that the
overall parameters for spacer, e.g., voidage and hydraulic diameter, are inadequate to
describe the impact of spacer on permeate productivity and membrane performance in
a practical spiral wound RO system.
The impact of mesh length of submerged configurations on the permeate flux

is shown in Figure 6.12. Permeate flux is noted to increase consistently with
decreasing mesh length. The main reason is that decreased mesh length always causes
more constricted passages where concentration boundary layer is compressed because
of elevated flow velocity in these regions. For the extreme case when the mesh length
decreases to 0, i. e., the transverse filaments form an impermeable wall in the center
of the feed channel, the effect of spacer would reach the maximum possible level on
115
concentration polarization alleviation and, consequently on flux enhancement.

Compared to the permeate flux in an empty channel under the same operating
conditions, these maximum permeate flux enhancements in the 10cm long channel
were found to be 38%, 33% and 17% for seawater, high-salinity brackish water and
low-salinity brackish water cases. Figure 6.12 also shows that permeate flux
enhancement is more significant for higher salinity cases. A possible reason is that the
impact of concentration polarization on permeate flux is usually more significant in a
high salinity system than that in a low salinity system.
Figure 6.11 Comparison of permeate flux in 10cm long feed channels with different
mesh length and with different filament configurations (cavity, zigzag and submerged)
H=1mm)
116
Figure 6.12 Comparison of the impact of mesh length on averaged permeate flux in
channels with submerged spacers
As shown in Figure 6.13 for zigzag configurations, there is an optimum mesh

length for permeate flux enhancement. When the mesh length is sufficiently large, to
shorten mesh length would effectively increase permeate flux by concentration
polarization alleviation. However, the spacers of too short mesh length may have
negative impact on permeate flux enhancement because of the increased surface
blockage and stagnant regions. It is shown that the optimum mesh length
corresponding for permeate flux enhancement is dependent on operation conditions,
which is about 1.5mm, 2.5mm, and 8mm for the seawater, high-salinity, and lowsalinity cases, respectively. This result shows the relative importance of different flux
controlling mechanisms in different cases. For a given mesh length, the beneficial
effects of increased concentration polarization alleviation on permeate flux
enhancement can be more significant than the adverse effects due to increased surface
117
blockage and stagnant regions in a system with high salinity. The opposite may occur
for the same mesh length in a low salinity system. Therefore, spacers should be
optimized for the specific operating conditions in particular applications.
channels with zigzag spacers
Figure 6.14 shows that in cavity configurations the mesh length has similar
impact on permeate flux as zigzag configurations. The optimum mesh length is
slightly larger for seawater desalination case than that with zigzag configurations. It is
shown that the smaller mesh length below the optimum value has net negative effects
on permeate flux. For example, the 1mm mesh length reduces the average permeate
flux by 6%, 10% and 19.4% in the seawater, high-salinity and low-salinity brackish
water cases respectively, compared with that in an empty channel under the same
operating conditions.
118
channels with cavity spacers
It can be seen by comparing Figures 6.13 and 6.14 that cavity configurations
are less effective than zigzag configurations in permeate flux enhancement if other
conditions are identical, and the difference usually grows with decreasing mesh
length. This can be explained by the difference in the flow patterns. The flow patterns
in zigzag and cavity configurations with 1mm mesh length are presented in Figures
6.15 and 6.16, respectively. Figure 6.15 shows that the recirculation flow is unlikely
to reach the downstream filament in zigzag configurations due to compression by the
filament on the opposite site. This would effectively enhance flux in these regions
because of the disruptions and compression in concentration boundary layer in these
regions. When the mesh length becomes smaller, the combined effect of boundary
119
layer compression and disruption may become more pronounced. As shown in Figure
6.16 the recirculation stretches to the downstream filaments for cavity configurations.
The relatively larger stagnant regions near the filaments would facilitate salt
accumulation and hinder further flux enhancement.
120
Figure 6.15 Contour of flow velocity in the feed channel (part) with zigzag spacer (simulation conditions: p=800psi; c0=32,000mg/l; A=7.31012
m/s Pa; u0=0.1m/s; H=1mm; lf=1.5mm)
121
Figure 6.16 Contour of flow velocity in the feed channel (part) with cavity spacer (simulation conditions: p=800psi; c0=32,000mg/l; A=7.31012
122
6.4.2 Impact of mesh length on pressure loss

It should be noted that decreasing mesh length would inevitably increase the
pressure loss. Pressure drops in the 10cm long channel with different filament
configurations are plotted in Figure 6.17. It can be found that the pressure loss
increases for all cases when mesh length decreases. This is mainly due to the
increased number of filaments corresponding to decreased mesh length.
Figure 6.17 Comparison of pressure loss in 10cm long feed channels with different
mesh length and with different filament configurations (cavity, zigzag and submerged)
H=1mm)
As shown in Figure 6.17 submerged configurations cause much higher

pressure loss than the other two configurations when the mesh length is larger than
1.5mm. The possible reason is that for submerged configurations, the flow
123
obstruction occurs at the regions with higher velocity as discussed earlier. This would
cause more energy dissipation and therefore higher pressure losses. Similar results
were reported by Schwinge et al (2002a).
For cavity and zigzag configurations there are no significant differences in

pressure loss when the mesh length is larger than 2.5mm. The main reason is that
when the mesh length is larger than this value, the downstream filament that is on the
opposite side of the upstream filament would not have significant impact on the
recirculation flow in the downstream of the upstream filaments, because the
recirculation (reversed flow) stretches about 2.5mm in the downstream under the
simulation conditions. Therefore, similar to the situations in cavity configurations, the
constricted passage corresponding to the downstream filament would mainly affect
the flow field after the recirculation regions. This would lead to very close pressure
loss.
However, when the mesh length is lower than 1.5mm, zigzag configurations
result in a significantly higher pressure loss, which is even higher than that for
submerged configurations. For example, when mesh length decreases from 1.5mm to
1mm in zigzag configurations, the pressure loss increases by about 166% although the
number of filaments only increases by 50%; when the mesh length decreases from
1mm to 0.5mm, the pressure loss increases by about 300% and the number of
filaments only increases by 33.3%. This pressure loss hike is unlikely caused by the
increase of the filament numbers. It could be envisaged as a result of the constricted
zigzag movement of main water flow between filaments that would not occur in the
other two configurations. When mesh length decreases, the zigzag flow path would
124
become more tortuous. Moreover, when mesh length is very low, the maximum flow
velocity would increase significantly although the height of the constricted passage is
still unchanged. For example, when the mesh length is 1.5mm (Figure 6.15), the
maximum velocity is about 0.35m/s; when the mesh length is 0.5mm (Figure 6.18),
the maximum velocity reaches 0.76m/s. These two factors would be surely more
pronounced, and therefore cause more energy loss when mesh length decreases.
Therefore, the pressure loss may increase sharply when the mesh length is lower than
1.5mm.
125
Figure 6.18 Contour of flow velocity in the feed channel (part) with zigzag spacer (simulation conditions: p=800psi; c0=32,000mg/l; A=7.31012
126
6.5 Summary
Concentration polarization, permeate flux, pressure loss in typical RO
membrane systems with different mesh lengths, three different configurations were
investigated to study the effects of filament configurations and mesh length on
concentration polarization and the performance of RO systems.
Concentration polarization patterns are drastically affected by filament

configurations. For submerged configurations and for the membrane opposite to the
transverse filaments in cavity configurations, the main mechanism is periodical
concentration boundary layer compression. For the membrane attached to transverse
filaments in cavity configurations, the main mechanism is concentration boundary
layer disruption. While for zigzag configurations two mechanisms are combined and
the recirculation regions in the downstream of each filament would be compressed or
even disrupted by the downstream filament. Flux enhancement is more pronounced
in most cases when the two mechanisms are combined as that in zigzag
configurations.
Permeate flux in a spacer filled membrane channel is affected by mesh length

of all three configurations. Permeate flux increases consistently with decreasing mesh
length for submerged configurations. There is an optimum mesh length for cavity
and zigzag configurations that results in the highest permeate flux enhancement.
Salinity appears as an important factor for the effectiveness of the spacers in flux
enhancement. The impact of spacers is more pronounced for higher salinity and the
optimum mesh length of cavity and zigzag configurations decreases with increasing
salinity.
127
Decreasing mesh length would lead to increased pressure loss. Submerged

configurations would usually result in high pressure loss. Pressure loss for cavity and
zigzag configurations is very close when the mesh length is larger than the length of
the recirculation formed in the downstream of a filament. However, when the mesh
length is lower than this value, zigzag configurations may lead to very high pressure
loss, which may be even substantially higher than that corresponding to submerged
configurations in some cases
The above results suggest that permeate flux enhancement may be achieved
by optimizing the design of the spacer through numerical simulations. In addition, the
results imply that the commonly used overall parameter of spacer (e.g., voidage) may
be inadequate to reflect the impact of spacer on concentration polarization and
permeate flux in a RO system, and that an optimized spacer design, which is suitable
for all RO systems, would unlikely exist.
128
CHAPTER 7 CONCLUSIONS AND

RECOMMENDATIONS
7.1 Conclusions
Concentration polarization is one of the most important factors that affect
fouling and the performance of membrane systems. However, most of the previous
studies on concentration polarization in the spacer filled membrane channel were
conducted with either pre-specified salt concentrations or flow velocities on
membrane surfaces or both. In fact, the salt concentration and flow velocities on the
membrane surfaces in most applications are not a prior knowledge. On the contrary,
the determination of these variables, which are major indicators of the system
performance, is one of the motivations to study concentration polarization. In this
study, a fully coupled SUPG finite element model was developed so that
simultaneous simulation of hydrodynamic conditions, including permeate velocity at
membrane surface, and salt concentration profiles, including wall concentrations in
RO membrane channels, become available. This makes it possible to study
concentration polarization and the role of feed spacer in spacer-filled channels
quantitatively.
Although the velocity is very low the flow direction transition region close to
membrane surface plays an important role in salt accumulation. Failing to capture
solute transport process in such regions by using inadequate meshing schemes or
129
neglecting it by assuming impermeable membrane surface would very likely

underestimate concentration polarization significantly.
It has been numerically demonstrated or visualized that concentration

polarization in spacer filled membrane channel is affected by two major mechanisms:
concentration boundary layer disruption due to flow separation and, concentration
boundary layer disruption due to the constricted flow passage. The two mechanisms
may work separately or jointly dependent on spacer configurations. For submerged
configurations, concentration polarization is solely affected by boundary layer
compression. For cavity configurations, boundary layer disruption works on the
membrane attached to the transverse filaments while boundary layer compression
works on the membrane opposite to the transverse filaments. The membranes with
zigzag configurations are affected by both mechanisms alternatively and this
combination usually leads to the best concentration polarization alleviation.
Filament geometry is demonstrated to have significant impact concentration

polarization although it would not change the overall concentration polarization
patterns. Extremely high wall concentrations are found close to the contact point of
membranes with cylindrical filaments. Increasing filament thickness can significantly
alleviate concentration polarization at cost of elevated pressure loss.
Membrane performance is strongly affected by filament configurations and

mesh length. In most cases, zigzag configuration provides the best permeate flux
enhancement while submerged configuration results in the lowest peak wall
concentrations. There is an optimum mesh length with cavity and zigzag
configurations for maximizing permeate flux enhancement.
The optimum mesh

130
length is found to increase for feed water of lower salinity. Decreasing mesh length
may lead to significant increase of pressure loss especially for zigzag configurations,
and may lead to permeate flux decline in certain cases. The significant difference
between configurations demonstrates that the commonly used overall parameter of
spacer (e.g., voidage) is inadequate or inappropriate to characterize spacers in a RO
system. Furthermore, the different behavior of spacers under different salinity implies
that a universally optimized spacer design does not exist. Optimization of the spacers
has to be carried out particularly for different situations.
The numerical model
developed in this study can serve as a powerful tool to facilitate the optimization
process.
Through this study, the understandings of concentration polarization and the

effects of spacer on concentration polarization and system performance have been
significantly advanced. The numerical model developed in this study can provide a
powerful tool to optimize spacer design in spiral wound RO modules.
7.2 Recommendations for further research

In most RO membrane systems membrane fouling, which is closely related to
concentration polarization, may also affect the performance of the membrane system.
However, it was not within the scope of this study. It is recommended to investigate
membrane fouling in the spacer filled membrane channel to better the understanding
of the role of feed spacer in spacer filled RO systems. The numerical model
developed in this study and the knowledge obtained on concentration polarization
have laid a strong and reliable foundation for the follow-up study on membrane
fouling.
131
In addition, the numerical model can be potentially further refined with the
incorporation of variable solution properties, such as the concentration-dependent
viscosity and density. These variable parameters may further increase the accuracy of
the model especially for high salinity cases. Currently, considerable computing time
is required to simulate flow patterns and concentration profiles in a short segment of
the membrane channel with spacers. Therefore, the development of more efficient
algorithms for numerical computing is also very attractive. That will make it possible
to simulate the whole RO membrane channel in a reasonable time frame.
132
REFERENCES
S. Avlonitis, W.T. Hanbury and M. Ben Boudinar (1993), Spiral wound
modules performance an analytical solution: part II, Desalination, 89: 227-246
M.S.H. Bader, P.A. Jennings (1992), Concentration polarization phenomena

in turbulent flow: Review and modification, Journal of Environmental Science and
Health, Part A: Environmental Science and Engineering, 27: 463-483
Y-N Be, Y-R Hou and L-Q Mei (2001), Global finite element nonlinear
Galerkin
method
for
the
penalized
Navier-Stokes
equations,
Journal
of
Computational Mathematics, 19: 607-616
G. Belfort (1984) Synthetic Membrane Processes: Fundamentals and Water

Applications, Academic Press, Orlando
G. Belfort and N. Nagata (1985), Fluid mechanics and cross-flow filtration:

some thoughts, Desalination, 53: 57-79
G. Belfort (1989), Fluid mechanics in membrane filtration: recent

development, Journal of Membrane Science, 48:231-262
A. S. Berman (1953) Laminar flow in channels with porous walls, Journal of

Applied Physics, 24:1232-1235
133
P. S. Bernard and J. M. Wallace (2002), Turbulent Flow: Analysis,

Measurement and Predication, John Wiley & Sons
D. Bhattacharyya, S.L. Back and R.I. Kermode (1990) Predication of

concentration polarization and flux behavior in reverse osmosis by numerical analysis,
Journal of Membrane Science, 48: 231-262
S. Bhattacharya and S.-T. Hwang (1997), Concentration polarization,

separation factor, and Peclet number in membrane processes, Journal of Membrane
Science, 132: 73-90
G. Bosch, M. Kappler and W. Rodi (1996), Experiments on the flow past a

square cylinder placed near a wall, Experimental Thermal and Fluid Science, 13: 292305
M. Ben Boudinar, W.T. Hanbury and S. Avlonitis, Numerical simulation and

optimization of spiral-wound modules, Desalination, 86: 273-290
P. L. T. Brian (1965), Concentration polarization in reverse osmosis

desalination with variable flux and incomplete salt rejection, Ind. Eng. Chem.
Fundam., 4: 439-445
A. N. Brooks and T.J.R. Hughes (1982), Streamline upwind Petrov-Galerkin

formulations for convection dominated flows with particular emphasis on the
incompressible Navier-Stokes equations, Computer Methods in Applied Mechanics
and Engineering, 32: 199-259
134
O.R. Burggraf (1966), Analytical and numerical studies of the structure of

steady separated flows. J. Fluid Mech., 24: 113151
Z. Cao, D.E. Wiley and A.G. Fane (2001) CFD simulations of net-type
turbulence promoters in a narrow channel, Journal of Membrane Science 185: 157176
G.F. Carey and R. Krishnan (1984), penalty finite element method for the
Navier-Stokes equations, Computer Methods in Applied Mechanics and Engineering,
42: 183-224
J.W. Carter, G. Hoyland and A. P. M. Hasting (1974), Concentration

polarisation in reverse osmosis flow systems under laminar conditions: effect of
surface roughness and fouling, Chemical Engineering Science, 29:1651-1658
A. Chiolle, G. Gianotti, M. Gramondo and G. Parrini (1978), Mathematical

model of reverse osmosis in parallel wall channels with turbulence promoting nets.
Desalination, 26: 3-16
T.J. Chung (1978), Finite Element Analysis in Fluid Dynamics. McGraw-Hill,

New York
A.R. Da Costa and A.G. Fane (1994a), Net-type spacers: effects of

configuration on fluid flow path and ultrafiltration flux, Industrial & Engineering
Chemistry Research, 33: 1845-1851
135
A.R. Da Costa, A.G. Fane and D.E. Wiley (1994b), Spacer characterization
and pressure drop modeling in spacer-filled channels for ultrafiltration, Journal of
Membrane Science, 87: 79-98
G. A. Denisov (1994), Theory of concentration polarization in cross-flow

ultrafiltration: gel-layer model and osmotic-pressure model, Journal of Membrane
Science, 91: 173-187
G. Dhati ; G. Hubert (1986), A study of penalty elements for incompressible

laminar flows, International Journal for Numerical Methods in Fluids, 6: 1-19
M. Elimelech and S. Bhattacharjee, S. (1998) A novel approach for modeling

concentration polarization in crossflow membrane filtration based on the equivalence
of osmotic pressure model and filtration theory, Journal of Membrane Science 145:
223-241
J. Fakova (1991), The pressure drop in membrane module with spacers,

W.W. Focke and P.G.J.M. Nuijens (1984), Velocity profile caused by a high
porosity spacer between parallel plates (membrane), Desalination, 49: 243-253
R. Franke, W. Rodi and B. Schonung (1990), Numerical calculation of

laminar vortex shedding flow past cylinders, Journal of Wind and Industrial
Aerodynamics, 35:237-257
136
D. Funaro, M. Giangi, D. Mansutti (1998), A splitting method for unsteady

incompressible viscous fluids imposing no boundary conditions on pressure, Journal
of Scientific Computing, 13: 95-104
E. J. Gallopoulos (1985), The Massively Parallel Processor for problems in

fluid dynamics, Computer Physics Communications, 37: 311-315
D. Van Gauwbergen and J. Baeyens (1997), Macroscopic fluid flow

conditions in spiral-wound membrane elements, Desalination, 110: 287-299
V. Gekas and B. Hallstrom (1987), Mass transfer in the membrane

concentration polarization layer under turbulent cross flow. I. Critical literature
review and adaptation of existing Sherwood correlations to membrane operations,
V. Geraldes, V. Semiao and M. N. de Pinho (1998), Numerical Modeling of

mass transfer in slits with semi-permeable membrane walls, Engineering
Computations, 17: 192-217
V. Geraldes, V. Semiao and M. N. de Pinho (2001), Flow and mass transfer

modeling of nanofilatration, Journal of Membrane Science, 191: 109-128
V. Geraldes, V. Semiao and M. N. de Pinho (2002a), The effect of the laddertype spacers configurations in NF spiral wound modules on concentration boundary
layers disruption, Desalination, 146: 187-194
137
V. Geraldes, V. Semiao and M. N. de Pinho (2002b), Flow management in

nanofiltration spiral wound modules with ladder-type sapcers, Journal of Membrane
Scicence, 203: 87-102
V. Geraldes, V. Semiao and M. N. de Pinho (2003), Hydrodynamics and

concentration polarization in NF/RO spiral-wound modules with ladder-type spacers,
V. Geraldes, V. Semiao and M. N. de Pinho (2004), Concentration

polarization and flow structure within nanofiltration spiral-wound modules with
ladder-type spacers, Computers & Structures, 82: 1561-1568
W. N. Gill, Chi Tien and D. W. Zeh (1966), Analysis of continuous reverse

osmosis systems for desalination, Int. J. Heat Mass Transfer, 9: 907-923
D. F. Griffiths (1997), The no boundary condition outflow boundary

condition, International Journal for Numerical Methods in Fluids, 24: 393-411
S. K. Hannani, M. Stanislas and P. Dupont (1995), Incompressible NavierStokes computations with SUPG and GLS formulations--A comparison study,
Computer Methods in Applied Mechanics and Engineering, 124: 153-170
F. K. Hebeker and P. Wilde (1992), On missing boundary conditions with

unsteady incompressible Navier-Stokes flows, Mathematical Methods in the Applied
Sciences, 15: 421-432
138
J. C. Heinrich, S. R. Idelsohn, E. Onate and C.A. Vionnet (1996), Boundary

conditions for finite element simulations of convective flows with artificial
boundaries, International Journal for Numerical Methods in Engineering, 39: 10531071
S. Hou, Q. Zou, S. Chen, G. Doolen and A. C. Cogley (1995), Simulation of

cavity flow by the lattice Boltzmann method, Journal of Computational Physics, 118:
329-347
T.J.R. Hughes and M. Mallet (1986), A new finite element method for CFD:
IV. A discontinuity-capturing operator for multidimensional advective-diffusive
systems, Computer Methods in Applied Mechanics and Engineering, 58: 329-336
L. Huang and M.T. Morrissey (1999), Finite element analysis as a tool for
crossflow membrane filter simulation, Journal of Membrane Science, 155: 19-30
S. Idelsohn, N. Nigro, M. Storti and G.Buscaglia (1996) A Petrov-Galerkin

formulation for advection-diffusion, Computer Methods in Applied Mechanics and
Engineering, 136: 27-46
A.R. Johnson and A. Acrivos (1969), Concentration polarization in reverse

osmosis under natural convection, Indus & Eng Chem-Fundamentals, 8: 359-361
I. S. Kang and H.N. Chang (1982), The effects of turbulence promoters on

mass transfer-numerical analysis and flow visualization, International Journal of
Heat and Mass Transfer, 25:1167-1181
139
S. Kang (2003), Characteristics of flow over two circular cylinders in a sideby-side arrangement at low Reynolds numbers, Physics of Fluids, 15:2486-2498
S.K. Karode (2001), Laminar flow in channels with porous walls, revisited,
S.K. Karode and A. Kumar (2001), Flow visualization through spacer filled
channels by computational fluid dynamics I. pressure drop and shear rate calculations
for flat sheet geometry, Journal of Membrane Science, 193: 69-84
D. H. Kim, I. H Kim and H. N. Chang (1983), Experimental study of mass

transfer around a turbulence promoter by the limiting current method, International
Journal of Heat and Mass Transfer, 26: 1007-1016
W. S. Kim, J. K. Park and H. N. Chang (1987), Mass transfer in a threedimensional net-type turbulence promoter, Int. J. Heat Mass Transfer, 30: 1183-1192
C. P. Koutsou, S.G. Yiantsios and A.J. Karabelas (2004), Numerical

simulation of the flow in a plane-channel containing a periodical array of cylindrical
turbulence promoters, Journal of Membrane Science, 231: 81-90
Y. Lee and M.M. Clark (1998), Modeling of flux decline during crossflow
ultrafiltration of colloidal suspensions, Journal of Membrane Science, 149: 181-202
140
P. F. Levy and R.S. Earle (1994), The effect of channel height and channel
spacers on flux and energy requirements in crossflow filtration, Journal of Membrane
Science, 91: 135-143
F. Li, W. Meindersma, A.B. de Haan and T. Reith (2002), Optimization of

commercial net spacers in spiral wound membrane modules, Journal of Membrane
Science, 208: 289-302
F. Li, W. Meindersma, A.B. de Haan and T. Reith (2005), Novel spacers for
mass transfer enhancement in membrane separations, Journal of Membrane Science,
253: 1-12
S. Ma, L. Song, S. L. Ong and W.J. Ng (2004), A 2-D streamline upwind

Petrov/Galerkin finite element model for concentration polarization in spiral wound
reverse osmosis modules, Journal of Membrane Science, 244: 129-139
K. Madireddi, R. B. Babcock, B. Levine, J. H. Kim and M. K. Stenstrom

(1999) An unsteady-state model to predict concentration polarization in commercial
spiral wound membranes, Journal of Membrane Science, 157: 13-34
D.S. Malkus and T.J.R. Hughes (1978), Mixed finite element methods
reduced and selective integration techniques: a unification of concepts, Computer
Methods in Applied Mechanics and Engineering, 15: 63-81
141
B. J. Marinas and R. I. Urama (1996), Modeling concentration-polarization in

reverse osmosis spiral-wound elements, Journal of Environmental Engineering, 122:
292-298
E. Matthiasson and B. Sivik (1980) Concentration polarization and fouling,

Desalination 35: 59-103
U. Merten, H. K. Lonsdale and R. L. Riley (1964), Boundary-layer effects in

reverse osmosis, Ind. Eng. Chem. Fundam., 3: 210-213
J. M. Miranda and J. B. L. M. Campos (2001), Concetration polarization in a

membrane placed under an impinging jet confined by a conical walla numerical
approach, Journal of Membrane Science, 182: 257-270
J. M. Miranda and J. B. L. M. Campos (2004), Mass transport regimes in

laminar boundary layer with suction parallel flow and high Peclect number,
International Journal of Heat and Mass Transfer, 47: 775-785
H. Miyoshi, T. Fukuumoto and T. Kataoka (1982), A consideration of flow

distribution in an ion exchange compartment with spacer, Desalination, 42: 47-55
J. Mizushima and T. Akinaga (2003), Vortex shedding from a row of square

bars, Fluid Dynamics Research, 32:179-191
142
Z. V. P. Murthy and S. K. Gupta (1997) Estimation of mass transfer

coefficient using a combined nonlinear membrane transport and film theory model,
Desalination 109: 39-49
M. Nallasamy (1986), Numerical solution of the separation flow due to an

obstruction, Computers & Fluids, 14: 59-68
R. D. Noble and S.A. Stern (1995), Membrane Separations Technology:

Principles and Applications, Elsevier Science B.V.
F. Padilla, Y. Secretan and M. Leclerc (1997), On open boundaries in the

finite element approximation of two-dimensional advection-diffusion flows,
International Journal for Numerical Methods in Engineering,, 40:2493-2516
T.C. Papanastasiou, M. Malamataris and K. Ellwood (1992), A new outflow

boundary condition, International Journal for Numerical Methods in Fluids, 14: 567608
E. Pellerin, E. Michelitsch, K. Darcovich, S. Lin and C. M. Tam (1995),

Turbulent transport in membrane modules by CFD simulation in two dimensions,
R. Peyret and T. D. Taylor (1983), Computational Methods for Fluid Flow,

Springer-Verlage, New York
143
M. N. de Pinho, V. Semiao and V. Geraldes (2002), Integrated modeling of

transport processes in fluid/nanofiltration membrane systems, Journal of Membrane
Science, 206: 189-200
S. V. Polyakov and F. N. Karelin (1992), Turbulence promoter geometry: its

influence on salt rejection and pressure losses of a composite-membrane spiral wound
module, Journal of Membrane Science, 75: 205-201
J. N. Reddy (1993), An Introduction to the Finite Element Method, McGraw

Hill
M. Renardy (1997), Imposing no boundary condition at outflow: why does it

work? International Journal for Numerical Methods in Fluids, 24: 413-417
C. J. Richardson and V. Nassehi (2003), Finite element modeling of

concentration polarization in flow domains with curved porous boundaries, Chemical
Engineering Science, 58: 2491-2503
S. S. Sablani, M.F.A. Goosen, R. Al-Belushi and M. Wilf (2001),

Concentration polarization in ultrafiltration and reverse osmosis: A critical review,
S. S. Sablani, M.F.A. Goosen, R. Al-Belushi and V. Gerardos (2002),

Influence of spacer thickness on permeate flux in spiral-wound seawater reverse
osmosis systems, Desalination, 146: 225-230
144
J.C. Schippers, J.H. Hanemaayer, C.A. Smolders and A. Kostense (1981),

Predicating flux decline of reverse osmosis membranes, Desalination, 38: 39-348
H. Schlichting and K. Gersten, Boundary-layer Theory, Springer Verlag,

Berlin, 2000
G. Schock and A. Miquel (1987), Mass transfer and pressure loss in spiral
wound modules, Desalination, 64: 339-352
J. Schwinge, D.E. Wiley and D.F. Fletcher (2002a), Simulation of the flow
around spacer filament between narrow channel walls. 1. hydrodynamics, Industrial
& Engineering Chemistry Research, 41:2977-2987
J. Schwinge, D.E. Wiley and D.F. Fletcher (2002b), Simulation of the flow
around spacer filament between narrow channel walls. 2. Mass transfer enhancement,
Industrial & Engineering Chemistry Research, 41: 4879-4888
J. Schwinge, D. E. Wiley and A. G. Fane (2004a), Novel spacer design

improves observed flux, Journal of Membrane Science, 229: 53-61
J. Schwinge, P. R. Neal, D. E. Wiley, D. F. Fletcher and A. G. Fane (2004b)

Spiral wound modules and spacers: Review and analysis, Journal of Membrane
Science, 242: 129-153
145
T. K. Sherwood, P. L. T. Brian, R. E. Fisher and L. Dresner (1965), Salt

concentration at phase boundaries in desalination by reverse osmosis, Ind. Eng. Chem.
Fundam., 4: 113-118
T.M. Shih, C.H. Tan, B.C. Hwang(1989), Equivalence of artificial

compressibility method and penalty-function method, Numerical Heat Transfer, Part
B: Fundamentals, 15: 127-130
L. Song and M. Elimelech (1995) Theory of concentration polarization in

crossflow filtration, J. Chem. Soc. Faraday Trans. 91: 3389-3398
L. Song and S. Yu (1999) Concentration polarization in cross-flow reverse

osmosis, AIChE Journal, 45: 921-928
L. Song, J.Y. Hu, S.L.Ong, W.J.Ng, M. Elimelech and M. Wilf (2003),

performance limitation of the full-scale reverse osmosis process, Journal of
Membrane Science, 214: 239-244
L. Song and S. Ma (2005), Numerical studies of the impact of spacer

geometry on concentration polarization in spiral wound membrane modules,
K. S. Spiegler and O. Kedem (1966) Thermodynamics of hyperfiltration

(reverse osmosis): criteria for efficient membranes, Desalination 1: 311-326
146
K. Sreenivas, P. Ragesh, S. DasGupta and S. De (2002), Modeling of crossflow osmotic pressure controlled membrane separation processes under turbulent flow
conditions, Journal of Membrane Science, 201: 203-212
S. Srinivasan, C. Tien and W. N. Gill (1967), Simultaneous development of

velocity and concentration profiles in reverse osmosis systems, Chemical Engineering
Science, 22 :417-433
S. Srinivasan and C. Tien (1969a), A finite difference solution for reverse

osmosis in turbulent flow, Desalination, 7: 51-74
S. Srinivasan and C. Tien (1969b), Simplified method for the prediction of

concentration polarization in reverse osmosis operation for multi-component systems,
D. Smner, S.S. T. Wong, S. J. Price and M. P. Paidoussis (1999), Fluid

behavior of side-by-side circular cylinders in steady cross-flow, Journal of Fluids and
Structures, 13: 309-338
Y. Taniguchi (1978), An analysis of reverse osmosis characteristics of ROGA

spiral-wound modules, Desalination, 25: 71-88
T. E. Tezuyar, Y.Park and H.A. Deans (1987), Finite-element procedures for

time-dependent convection-diffusion-reaction systems, International Journal for
Numerical Methods in Fluids, 7:1013-1033
147
C. Tien and W.N. Gill (1966), The relaxation of concentration polarization in

a reverse osmosis desalination system. AIChE Journal, 12: 722-727
A.M.P. Valli, G.F. Carey, A.L.G.A. Coutinho (2002), Control strategies for
timestep selection in simulation of coupled viscous flow and heat transfer,
Communications in Numerical Methods in Engineering, 18: 131-139
D. Wei (2001), Penalty approximations to the stationary power-law NavierStokes problem, Numerical Functional Analysis and Optimization, 22: 749-765
J.G. Wijmans, S. Nakao, J.W.A. van den Berg, F.R. Troelstra and C.A.
Smolders (1985) Hydrodynamic resistance of concentration polarization boundary
layers in ultrafiltration, Journal of Membrane Science, 22: 117-135
D.E. Wiley and D.F. Fletcher (2003), Techniques for computational fluid
dynamics modeling of flow in membrane channels, Journal of Membrane Science,
211: 127-137
Y. Winograd, A. Solan (1969), Concentration build-up in reverse osmosis in

turbulent flow, Desalination, 7: 97-109
Y. Winograd, A. Solan, and M. Toren (1973), Mass transfer in narrow

channels in the presence of turbulence promoters, Desalination, 13: 171-186
148
M. Yao, M. Nakatani and K. Suzuki (1989), Flow visualization and heat

transfer experiments in a duct with staggered array of cylinders, Experimental
Thermal and Fluid Science, 2: 193-200
M. B. Zaturska and W.H.H. Banks (2003), New solutions for flow in a

channel with porous walls and/or non-rigid walls, Fluid Dynamics Research, 33: 5771
O.C. Zienkiewicz and R.L. Taylor (2000), The Finite Element Method V3:
Fluid Dynamics, 5th edition, Oxford: Butterworth Heinemann
A.L. Zydney (1997), Stagnant film model for concentration polarization in

membrane systems, Journal of Membrane Science, 130:275-281
149
APPENDIX A : LIST OF PUBLICATIONS AND

CONFERENCE PRESENTATIONS
Shengwei Ma, Lianfa Song, Say Leong Ong and Wun Jern Ng (2004), A 2-D
streamline upwind Petrov/Galerkin finite element model for concentration
polarization in spiral wound reverse osmosis modules, Journal of Membrane Science,
244: 129-139
Lianfa Song and Shengwei Ma (2005), Numerical studies of the impact of

spacer geometry on concentration polarization in spiral wound membrane modules,
Shengwei Ma, Lianfa Song, Say Leong Ong and Wun Jern Ng (2003),
Numerical simulation of the effects of spacers on concentration polarization in spiral
wound reverse osmosis modules, presentated in the 14th Annual Meeting of North
American Membrane Society (NAMS 2003), Wyoming, USA, May 2003
Shengwei Ma and Lianfa Song (2005) Numerical study of the impact of

spacer filament geometry on concentration polarization in a reverse osmosis
membrane channel, presented in International Congress on Membrane and Membrane
Processes 2005 (ICOM2005), Korea, August 2005
150

Concentration Polarization in Spacer Filled Reverse Osmosis Membrane Systems Ma Shengwei

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Concentration Polarization in Spacer Filled Reverse Osmosis Membrane Systems Ma Shengwei

Hochgeladen von

Copyright:

Verfügbare Formate

CONCENTRATION POLARIZATION IN SPACERFILLED REVERSE OSMOSIS MEMBRANE SYSTEMS

I wish to express my appreciation and gratitude to my supervisor, Associate

I also wish to express my gratitude to National University of Singapore for

I am deeply indebted to my parents, my wife and my daughter for their

3.2 Model development..................................................................................22

Concentration polarization patterns for different filament

6.3 Filament configuration on membrane performance ................................111

6.4 Impact of mesh length on membrane performance .................................115

It was found that concentration polarization in spacer filled membrane channel

could significantly alleviate concentration polarization at cost of elevated pressure

It was also found that membrane performance was strongly affected by

Through this study, the understandings of concentration polarization in

The numerical model

developed in this study can provide a powerful tool to realistically study

visualization and assessment of the impact of spacer on system performance would

Table 6.1 Membrane properties and operating conditions ........................................100

p=800psi; c0=32,000mg/l; A=7.310-12m/s Pa; u0=0.1m/s; h=1mm;

Figure 6.1 Illustration of the computing domain (part) ..............................................99

zigzag and submerged) (simulation conditions: p=800psi;

membrane permeability (m/s Pa)

salt concentration (kg/m3)

characteristic dimension of an element (m)

channel height (m)

osmotic pressure constant (Pa m3/kg)

mesh length (distance between two neighboring filaments, m)

channel length (m)

normal direction of boundary

applied pressure (Pa)

Peclet number (=Hu0/D, H rather than hydraulic diameter is used in

Reynolds number (=Hu0/, H rather than hydraulic diameter is used in

axial velocity (x direction) (m/s)

lateral velocity (y direction) (m/s)

axial coordinate (crossflow direction) (m)

lateral coordinate (channel height direction) (m)

the penalty number

natural coordinates (-1~+1) in master elements (mapping of x)

natural coordinates (-1~+1) in master elements (mapping of y)

osmotic pressure (Pa)

wall or membrane surface

Although current material and chemical engineering technology has made it

In all RO membrane separation systems, concentration polarization is an

1.2 Aim of the research

1.3 Overview of the dissertation

CHAPTER 2 LITERATURE REVIEW

Because concentration polarization usually has significant adverse impact on

may result in significant different local concentration polarization behavior compared

It has been extremely difficult through experiments to directly observe and

Finite element method is one of the major powerful numerical methods in

2.2 Analytical models for concentration polarization in RO

osmotic pressure due to concentration polarization is now widely accepted and is

In recent years, there have been a few concentration polarization models

permeate flux formulations.

For example, Bhattacharya and Hwang (1997)

introduced a modified Peclet number and separation factor into concentration

Because the analytical models were developed under various implicit or

2.3 Numerical models for concentration polarization in RO

convection-diffusion equation for solute transport (Srinivasan et al, 1967;

With the development of high performance computing technology and the

Early numerical studies of concentration polarization were mainly focused on

model for concentration polarization in RO membranes based on the simplified

simplified velocity field (Berman, 1953), Madireddi et al (1999) numerically solved

Srinivasan et al (1967) developed a numerical model that could simulate the

To address the complex geometry in the feed channel, Richardson and

Numerical models based on commercial CFD (computational fluid dynamics)