Vous êtes sur la page 1sur 9

Desalination 217 (2007) 242250

CFD modelling of reverse osmosis membrane flow and


validation with experimental results
A. Alexiadisa*, D.E. Wileya**, A. Vishnoib, R.H.K. Leea, D.F. Fletcherc, J. Baoa
a

School of Chemical Sciences and Engineering, The University of New South Wales, Sydney, NSW 2052, Australia
Tel. +61 (2) 9385 4382; Fax: +61 (2) 9385 6955; email: sersunzo@yahoo.com
b
Department of Chemical Engineering, Indian Institute of Technology, Bombay, India
c
School of Chemical and Biomolecular Engineering, The University of Sydney, Sydney, NSW 2006, Australia
Received 16 February 2006; Accepted 2 February 2007

Abstract
Recently, computational fluid dynamics (CFD) models of flows in membrane channels have been proposed and
validated against published simplified models available in the literature. In this paper, experimental and numerical
analysis is performed for a reverse osmosis (RO) module and results compared in order to provide further validation
of the model from a practical prospective. Our results show that, in order to improve the accuracy of the calculations
at high pressure, a membrane permeability that varies with the operational pressure must be taken into account.
Keywords: Reverse osmosis; CFD; Modelling

1. Introduction
For many decades, the models describing the
solute concentration in membrane channels (see
[1] for a review) were based on simplified velocity patterns (e.g. [2]) because the equations of
the motion were too complicated to solve for
generic conditions. For this reason the results
*Current address: Marie Curie Transfer of Knowledge
Center for the Computational Sciences, Department of
Mechanical and Manufacturing Engineering, University
of Cyprus, PO Box 20537, 1678 Nicosia, Cyprus.
**Corresponding author.

were restricted to particular cases where the


adopted simplifications held. Nowadays, with the
availability of CFD software, the hydrodynamics
can be computed with higher accuracy and for a
wider range of conditions. Consequently, CFD
models describing the flow and the concentration
at the porous wall separating the feed and permeate channel in different types of membrane
modules has been recently proposed (e.g. [35,
913] and [1] for a review). This paper compares
the CFD model results against data from a small
bench scale flat-sheet RO system.

0011-9164/07/$ See front matter 2007 Published by Elsevier B.V.


doi:10.1016/j.desal.2007.02.014

243

A. Alexiadis et al. / Desalination 217 (2007) 242250

2. Theoretical model
The model equations are similar to those used
in previous work [5] except for the gravitational
forces, which are not included in this paper
because their effect is negligible for short
channels.
2.1. Geometry
The experimental module is composed of two
rectangular channels: the feed channel and the
permeate channel (see Fig. 1). The feed solution
enters from the feed channel inlet. The permeate
is collected in the lower channel and exits from
the permeate outlet; the retained feed exits from
the feed outlet. In Table 1, the channel dimensions, the membrane parameters and the operating
conditions are reported.

from the surface and the mesh expanded smoothly away from the membrane.
2.3. Experimental method
The experiments were performed on a benchscale, flat sheet reverse osmosis membrane
module. A flat sheet GE Osmonics AG membrane was pre-treated by soaking in 70% ethanol
solution for about 15 minutes to render the
membrane hydrophilic. The membrane was then
rinsed with pure water to remove the ethanol.
Before any experiments were performed, a pure
water experiment was conducted to measure the
water flux. This water flux was used to calculate
the membrane permeation coefficient K. This
Table 1
Geometric constants, membrane parameters and operating
conditions

2.2. Numerical method


As in our previous work [35], the simulations
were performed using the CFX4 code. The same
numerical techniques described in the earlier
study [3,4] were used. Following from that work
[4,5], a mesh of 100 cells across the feed channel
and 50 cells along the membrane section of the
channel was found to give grid independent
solutions. The near wall node on the feed-side of
the membrane was located at a distance of 5 mm

Fig. 1. Channel geometry.

276 mm

h1
h2
W
K
R
c0
u0
P

2.5 mm
2.0 mm
50 mm
6.939.72 10!7 m s!1 bar!1
0.995
2 g L!1
0.030.3 m s!1
914 bar

Geometric
constants

Membrane
parameters
Operating
conditions

244

A. Alexiadis et al. / Desalination 217 (2007) 242250

parameter comes from the pseudo-Darcy law,


vW = K(P!P), used as the boundary condition
at the feed side of the membrane [4]. The membrane was cut into several samples with dimensions of 30080 mm in order to fit into the
membrane cell. Not all the samples had the same
value of K. The membrane samples used in the
experiments could be divided in two categories:
six samples with K1 = (6.930.33)10!1 ms!1 bar!1
and eight samples with K2 = (9.720.77)10-7
ms!1 bar!1. The samples with permeability K1 and
those with permeability K2 were located in two
different areas of the original membrane sheet

Fig. 2. Experimental set-up.

(the value K1 was found at the edges of the


membrane sheet and K2 at the centre). This suggests that the average pore size and the permeaility were not uniform across the membrane sheet.
This fact can be related to the way the membrane
is manufactured, although the exact cause is
unknown.
The experiments were conducted with a
solution of salt (NaCl) at 2 g/L concentration.
The feed solution could be delivered to the
membrane at a maximum pressure of 15 bars (i.e.
1500 kPa), by varying the pump speed and the
retentate flow rate. The cross-flow velocity could

A. Alexiadis et al. / Desalination 217 (2007) 242250

also be altered by adjusting the retentate flow rate


and pump speed. Therefore, the two manipulated
variables are DP (trans-membrane pressure) and
u0 (cross-flow velocity). The measurements that
can be taken are system pressure, differential
pressures along the membrane channel, exit flow
rate, permeate pressure and permeate flux.
In Fig. 2 the experimental set-up is shown.
System pressure and differential pressure are
measured by CB1310 Labom pressure transducers (420 mA output, Tempress Controls) and
permeate pressure is measured with a CB1020
Labom pressure transducer (420 mA output,
Tempress Controls). Retentate flow rate was
measured by a flow sensor (256225 RS
Components) that outputs square waves (voltage
00.1 V), whose frequency is linearly related to
the flow rate. This waveform was recorded and
used to calculate the flow rate off-line. The
permeate flux was measured by a balance
connected to a serial port of a data acquisition
computer. By using the permeate flux, the
retentate flow rate and the channel geometry, the
cross-flow velocity was inferred. All sensor
outputs other than the permeate flux were
conditioned by using a signal-conditioning
module before being sent to a DAQ(National
Instruments PCI-6024E) card. The measurements
were recorded using LABVIEW.
All experiments were performed for 900 s to
allow the system to achieve constant permeate
flux. Pump speed varied from 30 to 46 rpm, and
the operating pressure was in the range of 1000
kPa to 1500 kPa (P ranging from 898.7 to
1398.7 kPa).
2.4. Membrane compaction
After 10 to 12 experiments with a single piece
of membrane it was found that the water flux
started to decline. Tests revealed that the decline
was not due to fouling or crystallization of salt in
or on the membrane but rather was due to
compaction. Further tests showed that the flux of

245

the membrane declined from 3840 LMH for the


unused membranes to 2628 LMH for the compacted membrane.

3. Results
The permeate flux was determined under the
operating conditions reported in Table 1. The
same conditions were simulated in the CFD code
and compared with the experimental results. In
Fig. 3, the comparison between experimental
J(exp.) and numerical J(num.) permeate flux data is
shown for the two sets of membranes at different
operating pressures. Results show good agreement between experiments and calculations
especially for P <1198.7 kPa. The error range
was 1.610.4% with an average deviation of 2%
for the K1-membranes and 0.517.8% with an
average deviation of 4% for the K2 membranes.
Experiments with the K1-membranes are less
scattered, probably because it was possible to
determine K1 with higher accuracy (smaller
standard deviation) than K2. Moreover, the flux at
high pressure (P >1198.7 kPa) is slightly overestimated by the model. The most likely reason
for this is due to the assumption that the K
coefficient is constant. This assumption implies
that the membrane characteristics remain the
same at different pressures. It is known, however,
that the membrane experiences compaction and,
therefore, its characteristics cannot remain the
same.
Additional simulations were carried out considering K2 as an adjustable parameter that varied
with P. The variable values of K2, determined in
order to fit the experimental data, are reported in
Fig. 4 as a function of the transmembrane pressure. The optimal value of K2 (Fig. 4) decreases
with P; this fact is consistent with the observation that the membrane undergoes compaction.
The revised numerical estimates of the flux,
computed with these variable K2, are also shown
in Fig. 3 (dotted lines).

246

A. Alexiadis et al. / Desalination 217 (2007) 242250

Fig. 3. Comparison between experimental J(exp.) and numerical J(num.) results for different cross-flow velocities at different
transmembrane pressures; J(opt.) is computed using variable K2 values given in Fig. 4.

247

A. Alexiadis et al. / Desalination 217 (2007) 242250

Fig. 4. Comparison between constant and


variable K2.

Table 2
Average error between experimental and numerical
results computed with constant and variable K2
P [kPa]

Constant-K2
error [%]

Constant-K2
error [%]

898.7
998.7
1098.7
1198.7
1298.7
1398.7

6.86
6.10
7.80
7.94
11.39
11.28

5.13
5.96
6.62
4.96
3.87
3.29

In Table 2, the comparison between the


experimental and numerical average error is
shown for variable and constant K2. The match
between experimental and numerical values
improves considerably, especially for P $
1198.7 kPa when K2 is varied with P.
3.1. Wall concentration
One of the advantages of using a CFD model
is that it permits evaluation and manipulation of
variables that usually cannot be measured experimentally and that are fundamental to the RO
process. The wall concentration, in particular,
plays a significant role because it directly affects
the osmotic pressure and therefore the permeate

flux. Empirical correlations have been proposed


in the literature (see [6] for a review) in order to
calculate the wall concentration in an RO system.
The results of two widely used correlations are, in
this section, compared with those obtained using
the CFD model.
3.2. Film model
The film model is used as a starting point for
many simplified laws used in membrane science.
The film model assumes one-dimensional flow, a
fully-developed boundary layer [8] and takes the
form of an ordinary differential equation that,
generally, has the following analytic solution

cW cP
J
= exp
cB cP
k

(1)

The mass transfer coefficient k is unknown and


must be calculated from empirical correlations.
For laminar flow, two expressions frequently
used (see [6, 7]) are

k d
d
= 0.664 Re 0.5 Sc 0.33
Sh =
D
L

0.5

when L #0.029 Re d (entry region) and

(2)

248

A. Alexiadis et al. / Desalination 217 (2007) 242250

Table 3
Estimated wall concentration at different conditions: variables indicated with * are from the CFD results, variables
indicated with are from the empirical correlation (Eq. 1); the last column refers to the respective curves in Figs. 5 and 6
P [kPa]

Re

LER [mm]

*J [L/m2/hr]

*cP [g L!1]

cw [g L!1]

Ref.

898.7
998.7
1098.7
1198.7
1298.7
1398.7

666
666
508
547
508
389

48.3
48.3
36.8
39.7
36.8
28.2

24.74
26.85
29.85
33.04
35.91
38.17

0.16
0.17
0.19
0.20
0.22
0.25

3.05
3.15
3.46
3.61
3.85
4.24

(a)
(b)
(c)
(d)
(e)
(f)

Fig. 5. Wall concentration on the retentate side of the membrane (see conditions in Table 3).

Sh =

k d
d
= 1.62 Re0.33 Sc 0.33
D
L

0.33

(3)

when L >0.029 Re d.
3.3. Comparison with CFD results
In Table 3, the values of cW are computed
using Eq. (1). The mass transfer coefficient k is
calculated with Eq. (3) since L #0.029 Re d in all

the cases considered. Fig. 5 compares the CFD


wall concentration profiles on the concentrate
side (continuous lines) with the values computed
using the correlation (dotted lines). In Fig. 6, the
CFD wall concentration profiles on the permeate
side cPW are shown. The average permeate
concentration cP and permeate flux J data listed in
Table 3 are computed from the CFD data. Note
that, in Fig. 5, the first and the last 8 mm of the
channel are non-permeable (see Fig. 1). For this

A. Alexiadis et al. / Desalination 217 (2007) 242250

249

Fig. 6. Wall concentration on the permeate side of the membrane (see conditions in Table 3).

reason, the profile of cW is constant over the first


part of the channel and decreases over the last
section. The same applies to Fig. 6 except that the
permeate channel is shorter than the concentrate
one and begins 8 mm after the latter.
The correlation gives an approximately constant value of the wall concentration but it must
be remembered that additional information about
cP and J is required in order to use Eq. (1). These
variables can be provided either by experiments
or (as in this case) by numerical results but in
both cases Eq. (1) is not a stand-alone equation.
The CFD model, on the other hand, is based on
first principles and requires only the operating
conditions values in order to produce results.
4. Conclusions
This paper compares results obtained from
experiments involving membrane filtration of a
sodium chloride solution with our previously

developed CFD model of membrane systems [4].


The comparison between experimental and theoretical data is good (average error ~7%) for
P <1198.7 kPa. For higher values of P, the
results are less convincing (average error ~11%).
This is probably due to the fact that the membrane undergoes compaction so that one of the
assumptions of the model (K = constant) does not
hold at high pressures. Additional simulations
using a variable K, a function of P, were compared with the previous values. These new results
show better agreement with the experimental data
and suggest a possible direction for improving
CFD modelling in membrane applications. The
pseudo-Darcy law used to model the local flow
through the membrane is based on constant
permeability, but our results show that the permeability does not remain constant during the
experiments and a state-equation of the type K =
K(P) is required to improve calculation
accuracy.

250

A. Alexiadis et al. / Desalination 217 (2007) 242250

In this paper, wall concentration values calculated with the CFD model and a semi-empirical
correlation based on the film theory are also
compared. The results show that the correlation
gives a reasonable first approximation of the wall
concentration but loses all the information regarding the concentration profile in the channel. In
addition, the correlation, unlike the CFD model,
can be used only in association with experiments
since certain variables, such as cP and J, are
required in order to close Eq. (1).
5. Symbols
c0
cB
cP
cW
cPW
d
D
h1
h2
J
k
K
L
P
P
R
Re
Sc
Sh
u0
vW
W

Inlet salt concentration, kg m!3


Bulk salt concentration, kg m!3
Permeate salt concentration (average), kg m!3
Wall salt concentration at the retentate side of the membrane, kg m!3
Wall salt concentration at the permeate side of the membrane, kg m!3
Hydraulic diameter, m
Diffusivity, m2 s!1
Concentrate channel height, m
Permeate channel height, m
Permeate flux, L m!2 h!1
Mass transfer coefficient, m s!1
Membrane permeation coefficient,
kg!1 m2 s
Concentrate channel length, m
Transmembrane pressure, kPa
Osmotic transmembrane pressure,
kPa
Membrane constant rejection
Reynolds number
Schmidt number
Sherwood number
Cross-flow velocity, m s!1
Liquid velocity through the membrane, m s!1
Membrane width, m

Acknowledgments
The authors gratefully acknowledge the financial support of the Australian Research council
(Discovery Project No. DP0343073).
References
[1] C. Kleinstreuer and G. Belfort, Mathematical model-ling
of fluid flow and solute distribution in pressure-driven
membrane modules, in: G. Belfort, ed., Synthetic Membrane Processes Fundamental and Water Applications, Academic Press, New York, 1984, pp. 131190.
[2] A.S. Berman, Laminar flow in channels with porous
walls, J. Appl.. Phys., 24 (1953) 1232.
[3] D.E. Wiley and D.F. Fletcher, Computational fluid dynamics modelling of flow and permeation for pressuredriven membrane processes, Desalination, 145 (2002)
183186.
[4] D.E. Wiley and D.F. Fletcher, Techniques for computational fluid dynamics modelling of flow in membrane channels, J. Membr. Sci., 211 (2003) 127137.
[5] D.F. Fletcher and D.E. Wiley, A computational fluid
dynamics study of buoyancy effects in reverse osmosis,
J. Membr. Sci., 245 (2004) 175181.
[6] A.R. Da Costa, A.G. Fane, C.J.D. Fell and A.C.M.
Franken, Optimal channel spacers design for ultrafiltration, J. Membr. Sci., 62 (1991) 275291.
[7] D.E. Wiley, C.J.D. Fell and A.G. Fane, Optimisation of
membrane module design for brackish water desalination, Desalination, 52 (1985) 249265.
[8] W.S.W. Ho and K.K. Sirkar, Membrane Handbook, Van
Nostrand Reinhold, New York, 1992.
[9] A.L. Ahmad, L.L. Lau and M.Z. Abu Bakar, Impact of
different spacer filament geometries on concentration
polarization control in narrow membrane channel. J.
Membr. Sci., 262 (2005) 138152.
[10] C.P. Koutsou, S.G. Yiantsios and A.J. Karabelas, Numerical simulation of the flow in a plane-channel containing a periodic array of cylindrical turbulence promoters.
J. Membr. Sci., 231 (2004) 8190.
[11] D. Dendukuri, S.K. Karode and A. Kumar, Flow visualization through spacer filled channels by computational
fluid dynamicsII: improved feed spacer designs. J.
Membr. Sci., 249 (2005) 4149.
[12] V. Vivek, V. Ranade and A. Kumar, Fluid dynamics of
spacer filled rectangular and curvilinear channels. J.
Membr. Sci., 271 (2006) 115.
[13] L. Song and S. Ma, Numerical studies of the impact of
spacer geometry on concentration polarization in spiral
wound membrane modules, Ind. Eng. Chem. Res., 44
(2005) 76387645.

Vous aimerez peut-être aussi