You are on page 1of 7

DESALINATION

Desalination 146 (2002) 195-201

ELSEVIER

www.eisevier.com/locate/desal

A CFD study of unsteady flow in narrow spacer-filled channels


for spiral-wound membrane modules
J. Schwinge, D.E. Wiley, D.F. Fletcherb
UNESCO Centre for Membrane Science and Technology, School of Chemical Engineering and Industrial Chemistry,
The University of New South Wales, Sydney, NSW 2052, Australia
Tel. +61 (2) 9385-4304; Fax +61 (2) 9385-5966; email: D. Wiley@unsw.edu.au
bDepartment of Chemical Engineering, The University of Sydney, Sydney, NSW 2006, Australia

Received 7 February 2002; accepted 21 February 2002

Abstract

In spiral-wound membrane modules, spacers are used to enhance wall shear stress and to promote eddy mixing,
thereby reducing wall concentration and fouling. Insights into the effect of spacer filaments on flow patterns in narrow
channels were obtained using a computational fluid dynamics (CFD) code. The flow patterns were visualized for
different filament configurations incorporating variations in mesh length, filament diameter and for channel Reynolds
numbers up to 1000. The simulated flow patterns revealed the dependence of the formation of recirculation regions on
the filament configuration, mesh length, filament diameter and the Reynolds number. When the channel Reynolds
number is increased above 300, the flow becomes super-critical showing time-dependent movements for a filament
located in the center of a narrow channel; and when the channel Reynolds number is increased above 500, the flow
becomes super-critical for a filament adjacent to the membrane wall. For multiple filament configurations, flow transition
can occur at channel Reynolds numbers as low as 80 for the submerged spacer at a very small mesh length (l,,,/hch= 1)
and at a slightly larger Reynolds number at a larger mesh length (l,,/h=,,= 4). The transition occurs above Rech of 300 for
the cavity spacer (ln/hd, = 4) and above Re, of 400 for the zigzag spacer (l,,/hch = 4).
Keywords: Feed spacer; Computational fluid dynamics (CFD); Unsteady fluid flow; Spiral-wound membrane module

1. Introduction

Treatment of water and other process fluids


with membranes are growing applications. A
*Corresponding author.
Presented at the International
July 7-12, 2002.
001 l-9164/02/$-

Congress on Membranes

major problem for membrane applications is


concentration polarization which reduces the flux
and which also can induce fouling. According to
the boundary layer theory [l], an increase in
velocity reduces concentration polarization due
and Membrane

Processes

See front matter 0 2002 Elsevier Science B.V. All rights reserved

PII: SO0 1 I-9 164(02)00470-8

(ICOM),

Toulouse, France,

196

J. Schwinge et al. /Desalination

to a higher wall shear stress as velocity increases,


but the changes in hydrodynamic conditions in
spiral-wound modules are limited by module
telescoping. Therefore, net-type spacers are used
in spiral-wound modules to enhance mass transfer and to reduce concentration polarization and
to maintain the channel height.
Many researchers have examined flow around
cylinders placed in the middle of a wide freeflow channel where interactions with the channel walls are neglected. They found that for
cylinder Reynolds numbers above 40, the
recirculation region starts to show periodic
movements. For a cylinder Reynolds number
greater than 200, the flow is found to be unsteady
and a vortex street develops behind the cylinder

[21*
For flow in narrow empty channels, the flow
is laminar for Reynolds numbers up to 2000 [ 11.
However, the flow in narrow obstructed channels
becomes unsteady at much smaller Reynolds
numbers. The flow can become unsteady and
show periodic movements for Reynolds numbers
as low as 200, depending on the geometry of the
obstructions [3,6]. Flow over a backward-facing
rectangular step is stable for much higher Reynolds number than for other obstacles submerged
in the channel, the flow being time independent
for Reynolds numbers as high as 2500 [7].
However, small disturbances in the upstream
region propagate unsteadiness downstream.
In this paper we examined the sub- and supercritical flow patterns in narrow two-dimensional
channels for single and multiple filaments. As the
paper focused on cylindrical spacer filaments
orientated transverse to the main flow direction,
two-dimensional CFD simulations were used.
These two-dimensional calculations are an excellent screening device for future evaluations of
more complex spacer geometries and allow the
effect of numerical parameters and modeling
assumptions to be studied for situations in which
grid independence can be guaranteed.

146 (2002) 195-201

2. Numerical modeling procedure


The fluid used is water at a temperature of
293 K which is assumed to be incompressible,
isothermal and to have constant fluid properties.
The fluid motion is described by the NavierStokes equations. The governing Navier-Stokes
equations are valid for all Reynolds numbers, but
under laminar conditions they can be solved
without the need to consider the resolution of
turbulent eddies [2,8]. These equations were
solved using a CFD code (CFX4, Version 4.3,
AEA Technology). A small enough grid size was
chosen, based on the comparison of results from
a succession of finer meshes, to ensure that the
results were independent of the grid.
Five filament configurations were examined:
(1) a single cylindrical filament adjacent to a
membrane wall, (2) a single cylindrical filament
placed in the center of a membrane channel, (3) a
cavity spacer with multiple cylindrical filaments
adjacent to one membrane wall, (4) a zigzag
spacer with multiple cylindrical filaments located
alternately adjacent to the top and bottom membrane wall, and (5) a submerged spacer with
multiple cylindrical filaments placed in the center
of a membrane channel [9, lo].
3. Flow patterns in narrow and obstructed two
dimensional

channels

3.1. Flow projiles for sub-critical

Reynolds

numbers in the channel


Fig. 1 shows steady flows at sub-critical Reynolds numbers for a section of the flow channel
in the region of the spacer filament for the five
filament configurations.
For a single filament adjacent to the bottom
wall, a small recirculation region is formed in
front of the filament and a much larger recirculation region is formed behind the filament. The
size and length of the recirculation region
increases with Reynolds number. Similarly, for a
single filament located in the center of the

Single filament adjacent to the wall


Single filament centered in the channel
Cavity spacer
Zigzag spacer

Fig. 1. Steady flow distributions along the channel for five filament configurations
= 4 for multiple filament configuratiks).

channel, a large recirculation region is formed


behind the filament and the size and length of the
recirculation region increases with Reynolds
number.
At very small Reynolds numbers or large
mesh lengths [9,10], the multiple filament cavity
and the submerged spacers produce large
recirculation regions behind each filament similar
to the regions formed for single filaments.
However, with an increase in Reynolds number
or decrease in mesh length, the recirculation
regions in front of and behind sequential
filaments influence each other and merge to form
one large recirculation region between filaments.
In contrast, the multiple filament zigzag
spacer forces the channel flow into a zigzag
pattern. The recirculation regions behind each
filament are forced to reattach to the membrane
wall by the consecutive filament at the opposite
membrane wall. Only at very small mesh lengths
does the separation streamline of the recircu-

at Re, equal to 100 (d/h, = 0.5, I,,/h,

lation region detach from the membrane wall


forming one large recirculation region between
sequential filaments at the same membrane wall.
However, such very small mesh lengths are
impractical in spiral-wound modules because the
spacer filaments would cover a high amount of
membrane area, significantly reducing the flux
[9,101.
3.2. Flow patterns for super-critical

Reynolds

numbers

Practical application of spiral-wound modules


often requires flow rates which exceed subcritical Reynolds numbers, leading to transient
flow conditions in the feed channel. To check our
method of modeling, transient flows were first
simulated for the flow in a free flow channel
using a single cylinder with the same diameter as
the spacers simulated earlier. The flow channel
was wide enough to avoid any effects of the walls

198

J Schwinge et al. /Desalination

on the flow around the cylinder and long enough


to avoid any interference of the channel entrance
and exit on the formation of vortices behind the
filament.
Our simulations of the single cylinder in a
free flow channel agree with accepted literature findings reported earlier [2]. A vortex street
is formed for Re, greater than 200. In addition,
the shedding frequency obtained here of 139 Hz
shows good agreement with a value of 143 Hz
predicted by Strouhal equation [2]:
-f?f_
urn

- 0.198.

l-19.7
(

%yt

(1)
I

The cylinder Reynolds number is defined by:

(2)
Having validated our method of modeling
transient flows, we then modeled flow in narrow
spacer-filled channels at super-critical Reynolds
numbers. To characterize the flow, a modified
channel Reynolds number Re, was used which
accounts for changes in the superficial velocity
uc,,and hydraulic diameter dh due to the presence
of spacer filaments [9]:

146 (2002) 195-201

In the transient mode, the filament Reynolds


number was varied from 38 to 1000. Time steps
of 5 us were used and convergence was achieved
within a maximum of 20 interations for each step.
For a single filament located in the center of the
narrow channel, the flow does not show any
transient movements for Re, below 200 despite
using small time steps and a fine mesh. Fig. 2
shows the transition to unsteadiness that occurs
above Re, = 300. At the Reynolds number where
the transition begins the recirculation region just
begins to flap behind the filament, but with an
increase in Reynolds number, the time dependent
movements increase and convect downstream
indicating a full transition to unsteady flow.
For a single filament adjacent to the bottom
wall, the flow is steady for Re, below 600.
Above Rech= 600, small disturbances caused by
the flow recirculation
propagate unsteadiness
downstream, as shown in Fig. 3. Despite the
different flow profile of the obstruction used
here, it is interesting that our value for the transition to unsteadiness agrees with the findings of
Kaiktsis et al. [7] for flow over a backwardfacing step.
Comparison of the transient results suggests
that the walls near a cylinder damp the transition
to unsteadiness.
In addition, location of a
cylinder at a wall further stabilizes the flow.
Fig. 4 shows the flow pattern for multiple
filaments located in the center of the channel at a
mesh length of Z,,/h,,,= 1 and at Re, = 195. A

Fig. 2 Unsteady flow caused by a single filament located in the center of the obstructed and narrow channel (d,/h, = 0.5)
at Re,, equal to 500.

J. Schwinge et al. /Desalination

large and fully formed but unsteady recirculation region exits between sequential filaments.
The recirculation regions show periodic movements at a frequency identical to that for a single
filament in the center of the channel.
For the submerged spacer at Z,,/hch= 4, as
shown in Fig. 5, large eddy movements exist
between sequential filaments for Re, = 360.
However, the flow becomes transient at Rech

. _

146 (2002) 195-201

199

above 200 which is similar to the value for


transition to unsteadiness for a single filament
located in the center of the channel. In addition,
a small disturbance caused by an upstream
filament causes an increase in the amplitude of
flow movements for consecutive filaments.
For the cavity spacer at Z,,/hc,,= 4, the flow
becomes transient at Rech above 300 which is a
lower Re, than for transition to unsteadiness for

.I

.,.-

_..

._

.,

I
/
/

_o,oo2

200

150

.,.

,..,..,

.._.

....

....._.
.

220

120
time [ms]

time [ms]

Fig, 3. Unsteady flow for Re,,, equal to 1000 when a filament is adjacent to the wall and small flow disturbances
convected downstream (d//h, = 0.5).

40

60

are

80
lime [ma]

Fig. 4. Periodic movement of the recirculation region between sequential filament for a spacer with filaments located in
the center of the channel at a small mesh length, 1,/h=,,= 1 and Re, = 195 (d/h, = 0.5).

200

J. Schwinge et al. /Desalination

I46 (2002) 195-201

Fig. 5. Propagation of large eddy movements between sequential filaments located in the center of the channel at a large
mesh length, l,Jh, = 4, and Re, = 360 (d,/h, = 0.5).

Fig. 6. Propagation of large eddy movements between sequential


disturbances at Re, = 718 (d//h, = 0.5).

cavity spacer filaments caused by small upstream

Fig. 7. Propagation of large eddy movements behind zigzag spacer filaments caused by small upstream disturbances at
Re, = 718 (d//h, = 0.5).

a single filament adjacent to the wall. For a


higher Reynolds number, a fully formed recirculation region appears between the first two
sequential filaments, but further downstream, as

shown in Fig. 6, the unsteady flow breaks the


fully formed recirculation region and eddies
move along between sequential filaments.
Fig. 7 shows the unsteady recirculation

J. Schwinge et al. /Desalination

regions that exist at Re, above 400 for the zigzag


spacer at Z,,/h, = 4. The transition to unsteadiness
for the zigzag spacer occurs at a smaller Re, than
for a single filament adjacent to the wall but at a
larger Re, than for a cavity spacer.

146 (2002) 195-201

Recy, Rech 24

24ch

~co

4. Conclusions
Computational fluid dynamics calculations
reveal complex relationships between filament
configurations, mesh length, filament diameter,
Reynolds number and the formation of recirculation regions. The transition to time-dependent
flows in an obstructed channel occurs at much
smaller Reynolds numbers than in an empty
narrow channel. For a single filament in a narrow
channel the wall damps the transition which
occurs at slightly higher Reynolds numbers than
for a cylinder in a free flow channel. Different
stages are observed for the transition to unsteadiness from slight movement to downstream
convection of recirculation regions.
Accurate evaluation of the two-dimensional
time-dependent flows demands extremely fine
numerical grids and small time steps which
increases the computational costs. Coarse grids
and large time steps suppress the development of
the time-dependent flow movements.
The complexity ofthe two-dimensional results
suggest that extension of the calculations to three
dimensions will require careful and systematic
identification of appropriate modeling procedures. Time-dependent movements of threedimensional recirculation regions are likely to be
extremely complex and sensitive to small
changes in modeling conditions.
5. Symbols
df

>

h
lLh

Filament diameter, m
Hydraulic diameter, m
Shedding frequency, Hz
Channel height, m
Mesh length, m

201

Cylinder Reynolds number


Channel Reynolds number
Velocity, m/s
Superficial velocity in the spacer
filled channel, m/s
Velocity at an infinite distance from
the filament, m/s
Kinematic viscosity, m%

References
PI H. Schlichting, Boundary Layer Theory, McGrawHill, New York, 1960.
PI B.S. Massey, Mechanics of Fluids, Chapman&Hall,
London, 1989.
[31 G.E. Kamiadakis, B.B. Mikic and A.T. Patera, Minimum-dissipation transport enhancement by flow
destabilization: Reynolds analogy revisited. J. Fluid
Mech., 192 (1988) 365.
[41 M. Greiner, R.F. Chen and R.A. Wirtz, Enhanced
heat transfer/pressure drop measured from a flat
surface in a grooved channel. J. Heat Transfer, 113
(1991) 498.
PI D. Majumdar and C.H. Amon, Heat and momentum
transport in self-sustained oscillatory viscous flow. J.
Heat Transfer, 111 (1992) 886.
PI C.H. Amon, D. Majumdar, C.V. Herman, F. Mayinger, B.B. Mikic and D.P. Sekulic, Numerical and
experimental studies of self-sustained oscillatory
flows in communicating channels. Int. J. Heat Mass
Transfer, 3.5 ( 1992) 3 115.
L.
Kaiktsis, G.E. Kamiadakis and S.A. Orszag,
171
Unsteadiness and convective instabilities in twodimensional flow over a backward facing step. J.
Fluid Mech., 321 (1996) 157.
PI CFX 4.3 User Manual, CFX International, AEA
Technology, Harwell, Didcot, Oxfordshire, UK.
191 J. Schwinge, D.E. Wiley and D.F. Fletcher, Simulation of the flow around spacer filaments in a small
channel. Part I: Hydrodynamics. Ind. Eng. Chem.
Res., in press.
Cl01 J. Schwinge, D.E. Wiley and A.G. Fane, Flux
improvements with a novel spacer design for ultrafiltration. Proc. 6th World Congress of Chemical
Engineering, Melbourne, Australia, 2001.