CFD Modeling of Crossflow Membrane Filtration –

Integration of Filtration Model and Fluid Transport

Model
E.H. Khor*, P. Kumar, Y. Samyudia
Department of Chemical Engineering, Curtin University of Technology, 98000 Miri, Sarawak, Malaysia
E-mail address: khor.ee.huey@stud.curtin.edu.my

the membrane cassette. Using this study, we are able to

design an optimum condition for our oil separation.
Cross-flow membrane filtration has become a promising
technique for waste-water treatment as compared to Most of the works done on modeling are on the flow
conventional treatment methods. One of the reasons is that the across the membrane and flux decline during filtration.
membrane techniques offer separation that can be achieved at Fouling models are based mainly on pore-blocking law,
ambient temperature with minimum energy. It is also an concentration polarization [9] and cake formation [6].
innovation for the application of cross-flow filtration in oil and Particle deposition on the membrane had also been studied
gas industry especially as an integral part for the oil-in-water very extensively, such as in the modeling work of Elimelech
analysis of produced water prior to offshore disposal. and Song [10]. Several authors are also investigating the
However, good fouling control is essential for the efficiency of hydrodynamics of fluid relating to the membrane filtration
the cross-flow filtration unit. With the fact that membrane is process. Many of them model using the combination of
not a passive entity, the understanding of particle deposition Navier-Stokes equations and Darcy’s law. Different
phenomena is vital for reducing fouling. approaches have been performed to simulate the combined
models such as finite element method [7], finite difference
In this paper, filtration will be modeled through the scheme [5], and finite volume method [8]. Finite volume
relationship between hydrodynamics of the cross-flows and the method and SIMPLE algorithm are commonly used in
transfer of flows across the membrane. The results of FLUENT problems dealing with fluid flow. For our simulation, we
simulated model are in good agreement with experimental
want to study the flow pattern which is inside a concealed
results. Simulation results of the model are presented and then
validated using experimental data for distilled (DI) water.
membrane. Our membrane is a rectangular small slits, anti-
From the model, some connecting variables are identified and gravity flow type. We therefore use commercial finite
established in this modeling work. By attaining these volume package FLUENT models to visualize the flow
connections, optimization of membrane filtration can be pattern inside the membrane at steady state.
achieved by adjusting the operating parameters.

Keywords: CFD modeling, Cross-flow membrane filtration, Our research work starts with model searching. From the
Produced water many models available we had chosen two sub-models in
our first stage. The first sub-model describes the fluid
I. INTRODUCTION transport of flow parallel to the membrane while the second
The determination of oil in produced water has been one describes the filtration across the membrane. The idea is
carried out for nearly two decades using solvent based to simulate the hydrodynamic characteristics which
extraction followed by infra-red quantification. However, influence the solute deposition on the membrane. For our
following Freon being banned from used due to ozone first stage, we use the models to simulate distilled (DI)
depletion and the concerns over health and safety of its water flow through membrane and validate with our
replacement Tetrachloroethylene (TTCE), OSLO-PARIS experimental data. For the second stage, we use a third
commission (OSPAR) implemented a new standard method model (film theory model) to simulate the deposition of
across North Sea in 2007. The method is called OSPAR GC- dispersed oil on the membrane which forms the gel layer. In
FID method but it has its own limitation [1]. Our research is this paper, it illustrates the first part of the study where
to incorporate the membrane filtration as part of oil-in water filtration through membrane is studied at standard condition,
analysis so that dispersed and dissolved oil can be separated room temperature and by using DI water.
and thus allowing the oil and gas operators to comply with
the stringent regulations. By doing this, we need to
understand fouling or gel layer built up on the membrane so
that our separation of both the oils can be complete. As such,
we use FLUENT model to simulate the flow pattern inside
the membrane to see the velocity and pressure distribution in
 II. NUMERICAL MODEL FORMULATION κ ∂P
u =−
Recall again, our simulation scope of this model is µ ∂x
defined to be within the region of the membrane cassette
(Fig. 2.1). For the first part of our modeling work, we model Porous wall in the membrane is assumed to be
the flow of fluid through the slit between two membrane homogenous and isotropic and the flow through porous wall
sheets (Fig. 1). The models done by Damak and coworkers can be treated as the boundary condition of the free flow
[5] are modified to fit the membrane system that we are through the tube.
using. The flow in-let is from the bottom and out through the
top of the rectangular slit of two membranes. According to 1) Boundary Conditions
Belfort and colleague [3] turbulent flow started at Re 4000
for porous tubes instead of Re 2100 in non-porous tubes. The a) At the inlet, the inlet pressure is
dimensions in the membrane cassette are complicated and
1 r 2
therefore it is difficult for Re calculations. However, due to p0 = ps + ρv
the high pressure exerted at the retentate outlet, the flow is 2
being disturbed and eddies are assumed to be formed. Flow Where po is the total pressure gauge at the inlet, ps is the
in the porous slits is therefore expected to be in turbulence. static pressure and v is the initial velocity which is 0.

A. Flow regime in the slit

b) At the exit, fully developed profile is assumed.
According to continuity equation and Navier-Stokes
y = L,0 < x < X : ∂v
equation, = 0; u = 0
∂y
∂u ∂v
+ = 0
∂x ∂y c) At the axis of symmetry there are no momentum
fluxes crossing the boundary.
The u-component is
x = 0,0 ≤ y ≤ L : ∂ v = 0 , u = 0
∂u ∂u ∂ u ∂ u
2
∂P
2 ∂x
ρ (u +v ) = µ( 2 + 2 ) − + ρg x
∂x ∂y ∂x ∂y ∂x
d) At the porous wall, the wall suction velocity is
given by Darcy’s law and no slip velocity is applied, as
The v-component is
follows,

∂v ∂v ∂ 2v ∂ 2v ∂P κ P − Pe
ρ (u + v ) = µ( 2 + 2 ) − + ρg y x = X ,0 ≤ y ≤ L : u = ,v = 0
∂x ∂y ∂x ∂y ∂y µ e
y

u = R ( P − Pe ), R = κ .1
µ e
where κ is the permeability of the membrane, µ is the
v viscosity of the fluid and e is the thickness of the porous
wall. These three parameters can be determined empirically
and we conclude them in a term R, resistance. Pe is the
L u Membrane external pressure (including osmotic pressure).

The numerical model assumed that the filtration is at

steady state, with a turbulent flow type. There are six
classical turbulence models in FLUENT i.e. mixing length,
x standard k-ε, RNG k-ε, realizable k-ε, Reynolds’s stress and
Figure 1: Slit between two sheets of membrane algebraic stress models [12].
(diagram is not proportional to the actual setup)
For our case, standard k-ε model [13] was chosen as our
B. Flow regime at the porous wall membrane geometry is not complex and the flow is assumed
to be fully turbulent and the effects of molecular viscosity
are negligible. The standard k-ε model is a semi-empirical
The momentum equation across porous zone, i.e. Darcy’s model based on model transport equations for the turbulence
law, is written as: kinetic energy (k) and its dissipation rate (ε). The model
transport equation for k is derived from the exact equation,
while the model transport equation for ε was obtained using
physical reasoning and bears little resemblance to its
mathematically exact counterpart. The transport equations
for k and ε are as follows:
Membrane
2
sheets
k
As µ t = ρC µ ; 0.165m
ε
y
∂(ρk ) ∂
+ ( ρ ku i ) =
∂t ∂xi z

∂  µt  ∂k  ∂u x
  µ +   − ρ u iu − ρε
j

∂x σk
j 0.02m
j    ∂x j  ∂xi
And 0.02m
Figure 2.2: Side view of
0.03m
∂ ( ρε ) ∂ membrane cassettes with
+ ( ρε u i ) = Figure 2.1: A dimension of illustration of membrane sheets
∂t ∂xi membrane cassette inside
Pp
∂  µt  ∂ε  ε ∂u ε 2 Permeate

  µ +  (− ρ u iu ) − C 2ε ρ
j
 + C 1ε
σε
j
∂x j    ∂x j  k ∂x i k PR
Retentate

C1ε and C2ε are constants. σk and σε are the turbulent Prandtl Membrane
numbers for k and ε, respectively. When differential Feed Tank
pressure (DP) is increased from 0.5bar to 2bar, with an Pin

interval of 0.5; the turbulent intensity decreases as described

in Tab. 2.
Figure 3: Experimental set-up for cross-flow membrane filtration
As mentioned in previous section, we use the Darcy’s
equation in modeling the filtration of distilled water across The empirical data of the permeability of the membrane,
the membrane. In FLUENT, Darcy’s equation is under the κ is calculated based on the formula,
porous jump boundary condition with the numeric model
described as follows: κ = u /( P.µ )
Where u is the permeate flux, P is the trans-membrane
µ 
∆p = − v ∆ m pressure (TMP) and µ is the viscosity of the fluid in this case
α  0.001kgm-1s-1.

P is the Trans-membrane Pressure (TMP), µ is the viscosity Trans-Membrane Pressure,

of the fluid, α is the permeability of the medium, v is the TMP = [(Pfeed + Pret)/2] – Pper
filtration flux, m is the thickness of the membrane. Differential Pressure,
DP = Pfeed – Pret
In our FLUENT geometry, we used a representation of 2
slices of membrane separating 3 rectangular compartments Pfeed, Pret,and Pper, are Feed pressure, Retentate pressure
to model the flow pattern inside the membrane cassette as and Permeate pressure respectively. Thickness of membrane,
illustrated in Fig. 1. e is 0.1mm based on the information given by the
manufacturer. All these information are included in the
simulation. Experiments are performed at TMP = 2.75 bar
III. EXPERIMENTAL SET-UP and Re between 4000 and 10000 by varying the operating
pressures using the experimental set-up shown in Fig. 3.
For experimental validation, Sartocon Slice Cassette Permeate and feed velocities are measured at DP 0.5, 1.0,
containing Polyethersulfone (PESU) membrane with 50kD 1.5, and 2.0. The simulated results from FLUENT are then
molecular-weight-cut-off (MWCO) is used. The validated with experimental data.
dimensions of the cassettes are given in the diagram (Fig
2.1). Each membrane cassette contains 20 slices of
membrane sheets. The distance between each membrane
sheets is approximately 1mm (Fig 2.2).
 IV. RESULTS AND DISCUSSIONS 1.0

Darcy’s equation and Navier-Stokes transport equations

associated with the boundary conditions given in Section II Experim ent vs. CFD perm eate data for DP
were solved by using finite volume method from a 1.5
commercial software FLUENT.
Feed mass Permeate Retentate 0.07

Permeate mass Flowrate

DP TMP
flowrate mass flowrate mass flowrate 0.06
0.5 2.75 0.04952 0.04584 0.00368 0.05
1.0 2.75 0.06910 0.04450 0.02510 0.04
1.5 2.75 0.08504 0.04233 0.04271
2.0 2.75 0.09375 0.04100 0.05280 0.03
0.02
Tab.1 shows the experimental data for mass flowrate for DP 0.5 to 2.0 at
constant TMP 0.01
0
Tab. 1 shows how the experimental feed, permeate and 7 8 9 10 12 13 15 17 18 20 30
retentate mass-flowrate (kg/s) changes under different
operating differential pressure and Trans-membrane Pressure Turbulence Intensity (%)
(TMP). The conditions (turbulence intensity) are adjustable
to fit the experimental data. CFD Experiment

Experim ent vs. CFD perm eate data for DP

Fig 6: CFD vs. Experimental data at different turbulence intensity for DP
0.5
1.5

0.2
0.1
Experim ent vs. CFD perm eate data for DP 2
0
Permeate mass Flowrate

-0.1 10 20 30 40 50 55 60 70 80 90

-0.2 0.08
Permeate mass

-0.3 0.06
Flowrate

-0.4
-0.5 0.04
-0.6 0.02
-0.7
0
-0.8 .5
10

11

12

13

14

20

30
9

Turbulence Intensity (%)

10

CFD Experiment Turbulence Intensity (%)

Fig 4: CFD vs. Experimental data at different turbulence intensity for DP CFD Experiment
0.5

Fig 7: CFD vs. Experimental data at different turbulence intensity for DP

Experim ent vs. CFD perm e ate data for DP 2.0
1.0

0.12
Fig. 4 to 7 shows the effect of the changes in turbulence
0.1
intensity towards the permeate fluxes. However the most
Permeate mass

suitable turbulence intensity conditions to have the CFD data

Flowrate

0.08
0.06
verified with experimental data are identified and listed in
Tab. 2.
0.04
0.02 Tab. 2 shows the CFD and experimental results for feed,
0 retentate and permeate mass flowrates. To avoid confusions,
Fig. 8 was plotted for permeate only for both CFD data and
30

40

60

80

Turbule nce Inte nsity (%)

experimental data. From the CFD results all the DPs fits well
with the experimental results and they are within the range of
CFD Experiment 10% in difference.

Fig 5: CFD vs. Experimental data at different turbulence intensity for DP

 CFD Experiment
DP Condition Feed Permeate Retentate Feed Permeate Retentate
0.5 70 0.04950 0.04214 0.00736 0.04952 0.04584 0.00368
1.0 40 0.06910 0.04812 0.02113 0.06910 0.04450 0.02510
1.5 13 0.08504 0.04457 0.04048 0.08504 0.04233 0.04271
2.0 10 0.09375 0.04557 0.06009 0.09375 0.04100 0.05280
*Condition – turbulence intensity (%)

Table 2: Experimental data and CFD data for mass-flowrates (kg/s) of feed,
retentate and permeate for various DPs

Permeate Mass Flowrate vs. DP

0.05000
0.04800
Mass Flowrate

0.04600
0.04400
CFD Figure 9: Velocity plot at the outlet for condition DP=0.5, TMP=2.75 at
0.04200
Experiment 70% turbulence intensity.
0.04000
0.03800
0.03600 pressure-outlet-12
pressure-outlet-5
pressure-outlet-6
0.5 1 1.5 2 4.50e+00

DP 4.00e+00

3.50e+00
Figure 8: Comparison of permeate flux from experiment and CFD
3.00e+00

2.50e+00
Fig. 9 shows the velocity plot at the outlet for DP 0.5 at Velocity
Magnitude 2.00e+00
TMP 2.75. The velocity plot from -0.5 to 0.5mm are the (m/s)
1.50e+00
retentate fluxes and the rest are permeate fluxes. This
1.00e+00
diagram shows that permeate fluxes are higher than the
5.00e-01
retentate fluxes. As DP increases from 0.5 to 2 (Fig 9 to 12),
we can see that the trend is reversed, as retentate fluxes 0.00e+00
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
increases whereas the permeate fluxes decreases. This trend Position (mm)
is the same as the trend from our experimental data. As DP
increases, fluid flow in the retentate slit becomes faster and Velocity Magnitude Nov 20, 2009
FLUENT 6.3 (2d, pbns, ske)

thus less permeation through the membrane. The CFD result Figure 10: Velocity plot at the outlet for condition DP=1.0, TMP=2.75 at
at DP 1.0 is much lower than the experimental data which 40% turbulence intensity.
may be attributed to the turbulent flow assumption.
Nevertheless, the simulation is good enough as it predicts the
same trend as the experimental data.
From Tab. 2, it can be seen that as DP increases, the
turbulence intensity decreases. This indicates that lower DP
has higher turbulence intensity than higher DP. The
calculation of turbulence intensity is by the equation, [12]
1
−
I = 0.16(Re) 8

Therefore from the equation, we will be able to calculate

and find out the Re for the flow in the retentate slit.

Figure 11: Velocity plot at the outlet for condition DP=1.5, TMP=2.75 at
13% turbulence intensity.
 [11] G. Van den Berg, C.A. Smolder, “The boundary-layer resistance
model for unstirred ultrafiltration. A new approach”, Journal of
Membrane Science, vol. 40, 1989, pp. 149-172.
[12] FLUENT 6.3 User-guide, Sept 2006
[13] B.E. Launder, D.B. Spalding “Lectures in Mathematical Models of
Turbulence”, Academic Press, London, England, 1972.

Figure 12: Velocity plot at the outlet for condition DP=2.0, TMP=2.75 at
10% turbulence intensity.

V. CONCLUSIONS

In conclusions, the CFD can predict the permeate flux

with a reasonable accuracy for the deionized water filtration.
From the simulation it can be seen that the differential
pressure and velocity is interconnected. Lower DP
contributes to higher turbulence intensities. Therefore to
obtain a higher permeate flux we need to operate at lower
DP and vice versa.

VI. REFERENCES

[1] E.H. Khor, Y. Samyudia, “The Study of Mass Transfer Coefficient in

Membrane Separation For Produced Water”, International Journal
Chemical Engineering, 2009, in press
[2] R.W. Baker, “Membrane Technology and Applications”, McGraw-
Hill, 2004.
[3] G. Belfort., N. Nagata “Fluid mechanics and crossflow filtration:
some thoughts”, Desalination, vol. 53, 1985, pp. 57-79.
[4] P.L.T. Brian “Mass Transport in Reverse Osmosis. Desalination by
Reverse Osmosis. U. Merten, ed. ed., MIT Press, Cambridge, MA,
1966.
[5] K. Damak, A. Ayadi, B. Zeghmati, P. Schmitz “A new Navier-Stokes
and Darcy's law combined model for fluid flow in crossflow filtration
tubular membranes. Desalination, vol. 161, 2004, pp. 67-77.
[6] L. Huang, M.T. Morrissey “Fouling of membranes during
microfiltration of surimi wash water: Roles of pore blocking and
surface cake formation”, Journal of Membrane Science, vol. 144,
1998, pp.113-123.
[7] V. Nassehi, “Modelling of combined Navier-Stokes and Darcy flows
in crossflow membrane filtration”, Chem. Eng. Sci., 53, 1998, pp.
1253-1265.
[8] A. Pak, T. Mohammadi, S.M. Hosseinalipour, V. Allahdini “CFD
modeling of porous membranes”, Desalination, vol. 222, 2008, pp.
482-488.
[9] M.C. Porter, “Concentration Polarization with Membrane
Ultrafiltration”, Ind. Eng. Chem. Prod. Res. Develop., vol. 11, 1972,
pp. 234 - 248.
[10] L. Song, M. Elimelech “Particle deposition onto a permeable surface
in laminar flow”, Journal of Colloid and Interface Science, vol. 173,
1995, pp. 165-180.