Desalination
journal homepage: www.elsevier.com/locate/desal
H I G H L I G H T S
a r t i c l e i n f o a b s t r a c t
Article history: In this study, a 2D CFD modeling and simulation of concentration polarization (CP) phenomenon in microltration
Received 15 October 2012 of oilwater emulsion in a narrow rectangular membrane module was presented. As the novelty of this research,
Received in revised form 27 May 2013 an Eulerian based multiphase method was investigated for the rst time to capture the concentration of emulsied
Accepted 28 May 2013
oil droplets in different extents of membrane module. It was shown that this new model is highly compatible
Available online 25 June 2013
with concentration polarization theory and is highly capable for determining CP prole and mass boundary
Keywords:
layer behavior in different operating conditions, as well. In this research it was proved that under developed
Microltration feed cross-ow shows better performance in terms of membrane surface cleaning. Also it was shown that a
Oilwater Emulsion small portion of oil droplets accumulates at the roof of membrane module feed channel due to gravity. The simu-
Concentration polarization lation results proved that for under developed feed cross-ows, lower feed Reynolds number induces a lower CP
CFD modeling prole peak despite increasing the oil droplet concentration on membrane surface and at the feed channel roof.
Multiphase model 2013 Elsevier B.V. All rights reserved.
1. Introduction to their moderate operational conditions and cost efciency [4]. How-
ever these processes are suffered by fouling phenomena causing per-
Oily waste-water is one of the most important pollutants emitted meate ux decline which is undesirable and in some cases may be
into the environment. These oil contaminated waters are generated irreversible [3]. A solute concentration which is higher in regions close
daily worldwide in many diverse industries such as petroleum ren- to the membrane surface in comparison to the bulk free stream, leads
ery, petrochemical, metal and steel, textile and machine industries to the concentration polarization, CP, phenomenon which arises from
[1,2]. For the oily wastewater treatment, the conventional technolo- the nature of microltration and ultraltration processes [5]. This phe-
gies are not efcient enough to treat the stable oilwater emulsions, nomenon affects the permeate ux negatively by introducing a positive
with oil droplets smaller than 20 m, especially when the oil concen- concentration prole toward the membrane surface causing back-
tration is pretty low and the droplets are nely dispersed. To address diffusion of solute rejectants from membrane surface to the bulk [5].
this problem membrane processes have been increasingly investigat- Therefore, numerous studies have been accomplished by re-
ed for treating oilwater emulsions. The rst attempts to treat oily searchers experimentally, mathematically and numerically to achieve
water by membrane separation processes were made in the beginning insight knowledge about membrane fouling, especially concentration
of 70's decade [3]. polarization.
In recent decades pressure driven membrane separation processes Many of these studies employed CFD simulation techniques due to
have been playing an important role in industrial unit operations due the accuracy of the results which are yielded through the numerical
solution of conservation of mass, momentum, and energy [6].
Corresponding author. Tel.: +98 21 6454 3124; fax: +98 2166405847. A large number of these studies have been performed over the
E-mail address: zokaee@aut.ac.ir (F. Zokaee Ashtiani). last two decades and are devoted to investigate the effects of different
0011-9164/$ see front matter 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.desal.2013.05.022
38 M. Zare et al. / Desalination 324 (2013) 3747
types of bafes, spacers and other types of turbulent promoters pressure distribution prole on the membrane surface in the compu-
and their orientation and spatial conguration on ow hydrodynam- tational domain [29,31,3437].
ics and concentration boundary layer disruption. These effects have Pak et al. [29] developed a 2D CFD code to simulate concentration
been considered in terms of parameters such as wall sheer stress, polarization for a porous membrane using Darcy's permeation law as
mass transfer coefcients, friction factor, power number, etc. [617]. a boundary condition for the membrane. They showed that with this
Also some researches are about the CFD calculations for new mem- assumption the CFD model may have a better agreement with exper-
brane module designs or special turbulence promoting modules iment results.
such as gas sparged ones to demonstrate their effects on ow hydro- Following this, some researchers employed Darcy's permeation
dynamics, mass boundary layer, and permeation ux [1827]. In this law, solute continuity equation and lm theory in conjunction with
category, some authors neglected the effects of low permeate ux turbulence promoters to calculate a more accurate solution of turbu-
by considering the membrane as an impermeable wall and fully fo- lent dominated membrane modules [8,10,13]. Using Darcy's perme-
cused on studying the effects of the promoted turbulence on ow ation law, Wardeh and Morvan [13] carried out a CFD simulation
hydrodynamics [7,12,14,15,17,20,21,2327]. Consequently the effect of a spacer-obstructed rectangular membrane module. They proved
of membrane transport mechanism on the ow hydrodynamics was that zigzagging type spacers are more efcient in concentration
neglected. polarization disruption.
Liu et al. [14] conducted a CFD simulation to evaluate the effects Since the rst attempts for treating oily wastewater by membrane
of bafes on turbulent ow in a tubular membrane module. They in early 1970's, several studies have been done to achieve insight
showed that using bafes signicantly increases the wall shear stress knowledge about different membrane separation of oilwater emul-
and improves the separation performance. Furthermore, it was proved sion systems and concentration polarization. Most of these studies
that central bafes are more effective than wall bafes in producing are limited to experiments and mathematical modeling where valu-
wall shear stress and result in a higher permeation ux. able results have been achieved; however numerical modeling of
Karode and Kumar [17] visualized the ow in a spacer lled rectan- such systems has received less attention [3,3947]. Besides, the avail-
gular channel. Despite the previous reports, it was revealed that the ability of CFD modeling and simulations of concentration polari-
bulk of uid ows along the spacer laments and does not change direc- zation in membrane separation of oilwater emulsion was found to
tion at each spacer node. Also, it was proved that spacers with unequal be limited.
lament diameters exert lower pressure drop to the uid ow. Addressing this problem, in the current study, a 2D nite volume
On the other hand, some researchers made the assumption of a model was developed for CFD modeling of laminar ow in cross ow
xed solute concentration, higher than that of the feed bulk, as a microltration of oilwater emulsion in an empty rectangular channel.
boundary condition on the membrane wall (impermeable wall) to The membrane actual permeation ux and its effect on ow hydrody-
consider the mass boundary layer formed due to concentration polar- namics were considered based on the Darcy's permeation law through
ization phenomenon. Then, they evaluated the effects of the pro- the calculation of local pressure distribution in the computational
moted turbulence on the mass boundary layer disruption [6,9,11,16]. domain. The oil was intended to be dispersed in water in the form of
These assumptions might lead to a more detailed but unrealistic eval- spherical micelles. As the novelty of this study, multi-phase ow equa-
uation of concentration polarization phenomenon. tions based on Eulerian multiphase model were investigated in CFD
Santos et al. [6] performed a CFD simulation to investigate the ow calculations to yield more accurate and realistic results.
pattern for 12 different ow-aligned using cyclic boundary condition.
They analyzed the results in terms of shear rate, mass transfer and a 2. Theory
modied form of friction factor. It was proved that the modied fric-
tion factor can be used to determine the best spacer conguration to 2.1. Module geometry and problem description
achieve higher mass transfer.
Shakaib et al. [9] carried out a CFD simulation to investigate the In the current study, the focus was on CFD modeling and simulation
ow and mass transfer in spacer lled channel of a spiral wound of concentration polarization in cross-ow microltration of oilwater
membrane module. They proved that axial and transverse spacer emulsion in an empty rectangular membrane module. The feed channel
laments exert higher shear stress at the top membrane wall but was a rectangular slit with dimensions of 0.3 5 10 cm as height,
the mass transfer rate remained almost the same. width and length of the rectangle, respectively. The porous membrane
In contrast, some studies have been conducted to model the was as the lower 5 10 cm wall of the slit. The permeate channel
empty membrane channel with laminar ow condition [13,2838]. was at the same shape and dimensions, as well. The feed was a mixture
In these studies the attempt was to consider the impact of membrane comprising water as continuous phase, containing emulsied oil as
transport mechanism on ow hydrodynamics. The calculations of con- dispersed phase in the form of micelles of diameters of about 1 m.
centration polarization were conducted through the solution of conti- The oil density and viscosity were intended to be 850 kg/m3 and
nuity, NavierStokes and solute continuity equations by incorporating 0.00332 kg/ms, respectively. The membrane module was fed by differ-
Film Theory which is based on true rejection and permeation ux. ent oil concentrations of 200, 1000, 5000 and 20,000 ppm. Also differ-
In these studies the impact of membrane transport mechanism on ent operating conditions in terms of trans-membrane pressure (TMP)
ow hydrodynamics was calculated by considering constant rejection and cross-ow velocity were applied in the problem. The process was
and constant or variable permeation ux as a boundary condition for conducted in different TMPs of 0.5, 1, and 2 bar, and cross-ow veloci-
membrane in CFD calculations [30,32,33,38]. ties corresponding to the Reynolds numbers of 500, 1300 and 2200.
Ahmad et al. [32] performed a CFD modeling of concentration
polarization in a narrow rectangular membrane channel by assuming 2.2. Geometry reconstruction and grid generation
xed permeate ux for the membrane. They investigated the effects
of different operating conditions on CP and showed that there was a According to geometry dimensions noted in previous section, the
good agreement between simulation results and literature data. ratio of height to width of the feed channel of membrane module
But the assumption of xed/variable permeation-rejection might was far less than unity (H/W 0.06). Hence the effects of side walls
lead to introducing errors to the calculation of concentration polari- on ow hydrodynamics were neglected, and in order to reduce com-
zation. Hence some authors tried to improve the accuracy of these putational costs, the model was intended to be two-dimensional for
calculations by introducing the Darcy's permeation law to the CFD CFD calculations. Therefore, a 0.3 10 cm rectangle (longitudinal
codes to predict the membrane permeate ux locally, based on local cross-section of feed channel) was considered as the computational
M. Zare et al. / Desalination 324 (2013) 3747 39
domain. Also, despite the fact that the study of ow hydrodynamics in condition due to the absence of turbulence promoters. The problem was
the permeate side was out of our scope in this research, a rectangle solved through the computational uid dynamic technique based
with dimension of 0.1 10 cm as permeate channel was considered on the multi-phase ow theory. Hence the governing equations of the
under the feed channel. This consideration was made only due to problem were divided into two main categories of continuous and
membrane transport phenomena calculations. The computational dispersed phases. All governing equations have been presented in
domain has been schematically shown in Fig. 1. Cartesian coordinate, as well.
The 2D computational domain was meshed with quadrilateral ele-
ments in order to minimize numerical diffusion errors in calculations. 2.4. Continuous phase
Finally after mesh dependency tests (Fig. 2), the short and long edges
of the feed side of the domain were divided into 300 and 200 intervals, As described above, the mixture contained water as continuous
respectively. To ensure that the grid is ne enough near the membrane phase and oil as dispersed phase in low concentrations. Consequently,
wall to capture the concentration polarization, a bell-shaped prole was the role of water as continuous phase was dominant in ow hydrody-
applied to the short edges of the feed channel, forcing height of the rst namics. For continuous phase, the ow hydrodynamics can be
rows of the cells adjacent to the membrane wall to be 0.176 m. Conse- achieved through the numerical solution of Continuity (Eq. (1)) and
quently the grid generated for feed side channel contained 60,000 cells NavierStokes (Eqs. (2), (3)) in their discretized form as governing
with minimum cell surface area of 8.8 1011 m2. equations.Continuity equation:
For the permeate side 0.1 10 cm channel, a grid of 20 200
cells was generated introducing 4000 cells to the computational do-
u v
main. This grid was ne enough just to capture pressure distribution 0 1
X Y
prole in permeate side of the membrane. A brief description about
problem parameters and operating conditions is presented in Table 1.
Cell height and vertical position of cell center points are shown in NavierStokes in X-direction:
Fig. 3 for each cell.
u u P u u
u v 2
2.3. Governing equations X Y X X X Y Y
Material properties
Water density 998.2 kg/m3
Fit 60,000 Water viscosity 0.001003 kg/ms
Fit 50,000 Oil density 830 kg/m3
Fit 40,000 Oil viscosity 0.003320 kg/ms
Fit 20,000
Fit 10,000 Operating conditiona
TMP 2, 1, 0.5 bar
0 0.02 0.04 0.06 0.08 0.1 0.12 Feed channel Reynolds no. 500, 1300, 2200
X-Position (m) Feed oil concentration 200, 1000, 5000, 20,000 ppm
a
Based on full factorial method, the simulations were conducted for 36 different
Fig. 2. Mesh dependency test (concentration vs. position for different grids). combinations of operating conditions.
40 M. Zare et al. / Desalination 324 (2013) 3747
where n is the number of phases, v m is the mass-averaged velocity,
m is the mixture density, and k is the volume fraction of phase k.
For a mixture containing continuous phase (q) and dispersed
phases (p), volume fraction () equation is:
p p : p p v m : p p v dr;p 7
t
X
n
v dr;p v pq k v qk 8
k1
k k
ck 9
m
p p m
v pq v p v q a 10
f drag p
2
p dp
p 11
18 q
0:687
f drag 1 0:15Re Re1000 12
0:0183Re Re1000
v
a g v m : v m m 13
t
where v dr;p is drift velocity, ck is mass fraction for the phase k, v pq is
relative velocity, p is particle relaxation time, dp is mean particle
diameter, fdrag is drag coefcient and a is the acceleration.
According to the parameters introduced above, the momentum
equation can be expressed as:
h T
i
m v m : m v m v m p : m v m v m 14
Fig. 3. (a) Height of each cell. (b) Vertical position of each cell center point. The cells
t !
X n
are numbered in Y-direction from membrane surface toward the channel top wall. m g F : k k vdr;k v dr;k
k1
NavierStokes in Y-direction:
X
n
m k k 15
v v P v v k1
u v 3
X Y Y X X Y Y
Where F is a body force, m is the viscosity of the mixture, and again
where is density, is viscosity and, u and are velocity components v dr;k is the drift velocity for dispersed phase k. The last term in
in X and Y directions, respectively. right-hand side of Eq. (14) keeps account for the impact of dispersed
phase on continuous phase hydrodynamics. Once more, it should
2.4.1. Dispersed phase be noticed that all time-dependent terms in equations above were
The calculations for emulsied oil micelles, in order to determine neglected in CFD calculations due to steady state nature of the
their accumulation and concentration in different regions of mem- problem.
brane module, were based on Eulerian multi-phase ow theory. From the governing equations, it can be concluded that the impact
Hence the formulation of governing equations introduced in this sec- of any change in momentum on dispersed phase volume fraction can
tion is based on phasic volume fraction, and concepts such as relative be taken into account through the solution of Eqs. (7) and (14). Sub-
velocity and drag force are incorporated in them. sequently it is expected that membrane transport mechanism
Continuity equation for a mixture is as follows:For a mixture (permeation-rejection) effects on oil micelles deposition on mem-
continuity equation is: brane surface to be captured via this formulation. Since oil micelles
are large spheres (dp 1m) the diffusion of rejected oil micelles
from membrane surface to the bulk was assumed to be zero.
m : m v m 0 4
t
2.5. Boundary conditions
X
n
k k v k
Table 2
CFD simulation results for uxes compared with experimental results.
3.1.1. Permeate ux
Simulation results for permeation ux are shown in Table 2. A
comparison with experimental results shows that there is a good
agreement between CFD simulation and experiments and the maxi-
mum and minimum errors are %15.08 and %7.41, respectively. By
taking a shallow look to the simulation results, it can be found that
the permeate ux increases by increasing the TMP. This can be simply
explained by direct proportionality of permeate ux with TMP in
Darcy's permeation law (Eq. (19)).
Fig. 6. Variation of Y-velocity prole across the membrane feed channel for TMP of 2 bar
and feed Re. of 1300. (a) From 1 cm to 5 cm. (b) from 5 cm to 9 cm. Fig. 7. Variation of Y-velocity prole across the membrane feed channel for TMP of
2 bar and feed Re. of 500. (a) From 1 cm to 5 cm. (b) From 5 cm to 9 cm.
Fig. 9. Variation of Y-velocity prole by TMP for different feed Re. at 0.8 of feed channel
Fig. 8. Variation of Y-velocity prole by feed Re. for different TMPs at 0.8 of feed channel length. (a) Re. 2200. (b) Re. 1300. (c) Re. 500.
length. (a) TMP = 2 bar. (b) TMP = 1 bar. (c) TMP = 0.5 bar.
The effect of different feed Reynolds numbers on CP prole was is more intense for lower feed Reynolds numbers. This is due to the
modeled and the results are shown in Fig. 15(a) and (b), for lower fact that, in ows with lower Reynolds numbers, the oil micelles have
half (membrane side) and upper half of the feed channel, respectively. further residence time, and consequently have more time to proceed
It's shown in Fig. 15(a) that applying lower feed Reynolds numbers their migration from feed bulk toward the feed channel roof. It can be
leads to development of a lower peak in CP prole but increases concluded that the deposition of oil micelles near the roof of feed chan-
the rejectant concentration, as a resistance for permeation, in near- nel reduces the concentration in membrane side, and the more the oil
membrane-wall region. The reverse effect of feed Reynolds number micelles deposited near feed channel roof, the lower the peak of CP
on the peak of CP prole can be explained according to Fig. 15(b). In prole.
Fig 15(b), it can be seen that a large portion of oil droplets (in compar- Applying a lower feed Reynolds number reduces the sweeping
ison to the peak of CP prole) is deposited near the feed channel roof effect of ow, and thus leads to a rise in rejectant concentration in
due to the effect of gravity and density difference. Also this deposition the near-membrane wall region as a resistance to the permeation
M. Zare et al. / Desalination 324 (2013) 3747 45
Fig. 10. The effect of TMP on permeate ux for feed Re. of 2200 and concentration of
20,000 ppm.
Fig. 12. The effect of feed Re. on permeate ux. (a) 2 bar and 20,000 ppm. (b) 0.5 bar
and 20,000 ppm.
4. Conclusions
Fig. 13. The effect of TMP on oil concentration prole (or concentration polarization
Fig. 11. The effect of feed concentration on permeate ux. (a) 2 bar and Re. 2200. effect). For Re. of 2200 and oil concentration of 20,000 ppm. Values presented by the
(b) 0.5 bar and Re. 2200. plots should be added up with corresponding bulk feed concentration.
46 M. Zare et al. / Desalination 324 (2013) 3747
Fig. 15. The effect for feed Reynolds number on CP prole for TMP of 1 bar and concen-
tration of 20,000 ppm. (a) For lower half of feed channel. (b) For upper half of feed
channel. Values presented by the plots should be added up with corresponding bulk
feed concentration.
Nomenclature [20] T. Taha, Z.F. Cui, CFD modelling of slug ow in vertical tubes, Chem. Eng. Sci. 61
(2006) 676687.
CP Concentration Polarization [21] S. Buetehorn, D. Volmering, K. Vossenkaul, T. Wintgens, M. Wessling, T. Melin,
TMP Trans-Membrane Pressure (bar) CFD simulation of single- and multi-phase ows through submerged membrane
u X-Component of Velocity (m/s) units with irregular ber arrangement, J. Membr. Sci. 384 (2011) 184197.
[22] L. Martinelli, C. Guigui, A. Line, Characterisation of hydrodynamics induced by
v Y-Component of velocity (m/s) air injection related to membrane fouling behaviour, Desalination 250 (2010)
Density (kg/m3) 587591.
P Pressure (Pa) [23] S. Smith, T. Taha, Z. Cui, Enhancing hollow bre ultraltration using slug-ow a
hydrodynamic study, Desalination 146 (2002) 6974.
Viscosity (kg/ms) [24] N. Ratkovich, C.C.V. Chan, P.R. Berube, I. Nopens, Experimental study and CFD
vm Mass-Averaged Velocity (m/s) modelling of a two-phase slug ow for an airlift tubular membrane, Chem. Eng.
m Mixture Density (kg/m3) Sci. 64 (2009) 35763584.
[25] T. Taha, W.L. Cheong, R.W. Field, Z.F. Cui, Gas-sparged ultraltration using hori-
ak Volume Fraction of Phase, K
zontal and inclined tubular membranesa CFD study, J. Membr. Sci. 279 (2006)
v dr;p Drift Velocity (m/s) 487494.
ck Mass Fraction of Phase, K [26] A. Sarkar, S. Moulik, D. Sarkar, A. Roy, C. Bhattacharjee, Performance characteriza-
tion and CFD analysis of a novel shear enhanced membrane module in ultraltra-
v pq Relative Velocity (m/s)
tion of Bovine Serum Albumin (BSA), Desalination 292 (2012) 5363.
p Particle Relaxation Time (s) [27] M. Bubolz, M. Wille, G. Langer, U. Werner, The use of dean vortices for crossow
fdrag Drag Coefcient microltration: basic principles and further investigation, Sep. Purif. Technol. 26