Desalination 354 (2014) 7686

Contents lists available at ScienceDirect

Desalination
journal homepage: www.elsevier.com/locate/desal

Characterization and modeling of radial ow membrane (RFM) module

in ultraltration
Debasish Sarkar a, Debojyoti Chakraborty b, Mithu Naskar a, Chiranjib Bhattacharjee b,
a
Department of Chemical Engineering, University of Calcutta, Kolkata, India
b
Department of Chemical Engineering, Jadavpur University, Kolkata, India

H I G H L I G H T S

Design of a novel radial ow membrane (RFM) module

Performance characterization of the module in ultraltration of BSA
Average permeate ux enhancement is 0.06 10 6 m3 m 2 s 1 per unit kPa rise of TMP.
Mathematical model to predict the steady permeate ux
Maximum absolute deviation of the model is well within 7%.

a r t i c l e i n f o a b s t r a c t

Article history: Design, characterization and modeling of a novel radial ow membrane (RFM) module is reported in the present
Received 31 March 2014 study. The module has been investigated in ultraltration of bovine serum albumin (BSA) with a 30 kDa
Received in revised form 16 September 2014 polyethersulphone (PES) membrane under steady state. The proposed semi-empirical, steady state model was
Accepted 17 September 2014
validated under different parametric conditions of transmembrane pressure and bulk concentration. The
Available online xxxx
model may be viewed as an extension of the well known osmotic pressure model in conjugation with
Keywords:
SpieglerKedem black box model and a system specic, steady state solute balance equation. Model empiricity
Ultraltration was introduced primarily to account for unknown entranceexit effects and effects of membrane roughness on
Radial ow hydrodynamic prole. The empirical parameter was evaluated by minimizing a global objective function. The
Steady state maximum absolute deviation of the model with respect to the experimental data of permeate ux was found
Mathematical model to be well within 7%.
Permeate ux 2014 Elsevier B.V. All rights reserved.
Rejection

1. Introduction Different remedial techniques ranging from feed pretreatment to force

Membrane separation technology is an energy efcient downstream
process either to separate valuable components from an intermediate
process stream or to treat the far downstream efuents in order to
meet the discharge standards [13]. However, rapid ux decline due
to (i) reversible concentration polarization and (ii) irreversible mem-
brane fouling restrict the growth of membrane based techniques as an
alternative to the conventional energy intensive separation processes.
These non idealities triggers higher energy consumption and addition-
ally reduce the effective membrane life [4]. In general, concentration po-
larization can be partially controlled by the use of periodic pulsation,
electric eld, gas sparging or ultrasound [510]. On the other hand,
membrane fouling is a complicated physicochemical phenomenon,
which leads to an irreversible loss of membrane permeability [11].
Corresponding author.
E-mail address: cbhattacharyya@chemical.jduv.ac.in (C. Bhattacharjee).
eld assisted ltration have been proposed to counter fouling phenom-
enon. Recently, in a review article, Hilal et al. [12] have presented a root
cause analysis and necessary remedial techniques of fouling in details. It
has been also established that to limit the membrane fouling, concen-
tration polarization must be reduced to its lowest possible threshold
[13].
It is well recognized that high membrane shear arrests the growth of
polarized layer and the therefore enhances the permeate ux. To create
a high membrane shear, it is necessary to circulate the uid at high
speed i.e. in the range of 36 m s1 [14] in a standard cross ow module.
This leads to a higher pumping cost. Over the last two decades, different
types of cross ow modules with varied geometry and/or shear enhanc-
ing accessories were proposed [1520]. Alternative to passive enhance-
ment strategies, active use of external force elds, for example
ultrasound to enhance the permeate throughput are also reported [21,
22]. Dynamic shear enhanced (DSE) modules, which consist of rotating
or reciprocating parts like rotating, spinning or vibrating disks and/or

http://dx.doi.org/10.1016/j.desal.2014.09.020
0011-9164/ 2014 Elsevier B.V. All rights reserved.
 D. Sarkar et al. / Desalination 354 (2014) 7686
membranes were introduced in late 1980s to further intensify the
membrane shear independent of feed ow rate [2327]. However, the
specic membrane surface area (m2 m3) of even the most advanced
DSE module is practically insignicant compared to the standard cross
ow units. For example, in a typical plate and frame module (the oldest
cross ow unit commercialized), the surface to volume ratio is around
350500 m1, whereas the same for the well known hollow ber mod-
ules is as high as 700013,000 m 1 [28,29]. Naturally the permeate
throughput becomes much higher in cross ow devices compared to
the equivalent DSE units of same size. Instead of several advantages, en-
hanced axial pressure drop in a cross ow module leads to a progressive
reduction of transmembrane pressure. Therefore, it is necessary to
introduce a cross ow design with reduced pressure drop so as to opti-
mize the membrane area usage in terms of uniform TMP distribution.
It is well known that in a radial ow channel with central inlet and
peripheral outlet (source ow type), the pressure drop is lower than a
rectangular/cylindrical conduit of same internal volume and surface
area. Based on this idea, the present work has been undertaken in an at-
tempt to develop and characterize a radial ow module in ultraltration
of bovine serum albumin (BSA). Moreover, a semi-empirical model was
proposed to simulate the present radial ow membrane (RFM) module
under steady state.
2. Materials and methods
2.1. Materials
Bovine serum albumin (BSA) with an average molecular weight
of 66,000 g mol 1 was supplied by E. Merck, Mumbai, India. Moist,
semi-permeable, asymmetric polyethersulphone (PES) membrane
with a molecular weight cut off (MWCO) of 30 kDa was obtained from
Spectrum Medical Industries (USA). The at disk membrane (diameter:
Fig. 1. Sectional view of the radial ow membrane (RFM) module (insert showing the photograph of the module internals).
77
0.045 m) was operable in the pH range of 210 and temperature range
of 554 C.
2.2. Filtration system
The present module was designed to ensure smooth radial ow of
the feed over a at circular membrane. Therefore, the corresponding ki-
nematic pattern must either be source or sink ow type. The primary
drawback of sink ow is the heavy impact of incoming radial jets over
the central region. This may create ow reversal and may hinder the
outow of the retentate. Accordingly, source ow type feed inlet and
retentate withdrawal system was chosen for the present work. Fig. 1
represents the sectional view of the module along with the snapshots
of the membrane holder and the top lid. The membrane holder is a per-
forated disk through which permeates ows out. This also acts as a me-
chanical support of the membrane. Permeate can be collected through a
radial tube tted below the holder. The feed is delivered to the module
through a central opening on the top lid. Just below the opening, a circu-
lar impingement plate of 0.016 m radius (ri) is xed to top lid in order to
avoid the feed jet impact on the membrane. Essentially, the feed enters
the module through a narrow channel of 0.001 m depth between the
impingement plate and the top lid. The retentate ows out of the mod-
ule through nine peripheral holes of diameter 0.005 m, drilled through
the lid. The clearance between the top lid and the membrane (zt) is
0.014 m, as shown in Fig. 1. An O-ring was used to x the membrane
over the porous membrane holder. This eventually reduces the effective
diameter of the membrane diameter (ro) to 0.041 m. For the present
work, the RFM module, made of SS-316 was fabricated by Concept
International, Kolkata, India as per the specied design. SS-316 was
selected as the material of construction.
The schematic of the entire ltration bench is shown in Fig. 2. The
feed tank is tted with a coolant circulation system. The feed is
78 D. Sarkar et al. / Desalination 354 (2014) 7686
Fig. 2. Schematic of the ltration bench.
delivered to the module by a single acting piston pump. The TMP and
the ow rate can be adjusted by operating the back pressure regulator
(BPR) and the bypass valve (V1) respectively.
2.3. Analysis
The concentration of BSA in BSA/water solution was measured by a
Hitachi dual beam UVvisible spectrophotometer (Model No. U-2800,
tted with UV solution software) by the principle of absorbance assay
at 280 nm [30]. The solution pH was monitored by a Systronix make
digital pH meter.
2.4. Experimental procedure
Prior to all experimental runs, the membrane was subjected to water
compaction nearly for 2 h at a transmembrane pressure (TMP) of 600
kPa, which is higher than the highest operational TMP of the present
study. This was followed by the actual experimental runs. Permeate
ux was measured by the direct method. According to this method,
for a permeate volume of V, collected over a xed time interval t,
the respective ux value may be calculated as J A1 Vt , where A is the ef-
fective membrane surface area. Flux values were recorded at an internal
of 10 min. Essentially t 10 min. The observed rejection, dened as
Ro 1 CC P0 , where C0 is the feed concentration, was calculated measur-
ing the corresponding steady state permeate concentration (CP) of a
specic run. A particular run was continued until three successive ux
readings were equal. Once an experimental run was over, the mem-
brane was thoroughly cleansed with distilled water to remove any de-
posits. The water ux was checked again to ensure the same hydraulic
resistance of the membrane. For all the runs, the feed solution was pre-
pared and maintained at the neutral pH, far from the isoelectric point of
BSA (pH 4.71). This was done in order to reduce the solute adsorption
led irreversible membrane fouling. Solute concentration in the feed
tank was continuously monitored and make-up solution was periodi-
cally added to maintain C0 at its set point.
In order to extract the dependence of permeate ux on two process
parameters, namely the TMP and C0, a 22 central composite rotatable
design (CCRD) was carried out because CCRD takes all the curvatures
of the response surface i.e., parametric interactions into account [31].
Thirteen experimental runs were performed with varying TMP (223.31
to 368.69 kPa) and feed concentration (5.86 to 34.14 kg m3). Corre-
sponding data of permeate ux were used to determine an adjustable
parameter of the proposed model as discussed later. Additionally, in
order to extract the dependence of permeate ux on TMP and C0 over
an extrapolated parametric space, 25 experimental runs were designed
with varying TMP (196.2, 244.6, 294.3, 347.4 and 392.4 kPa) and feed
concentration (5, 10, 20, 30 and 40 kg m3). One parameter at a time
was subjected to variation, while the other was held constant to extract
the true nature of the dependence. A constant retentate ow rate of
6.4 10 6 m3 s 1 was maintained throughout all the experimental
runs in order to restrict the number of effective process parameters
within two. The operating temperature of the present study was
30 1 C. Each experiment was conducted in triplicate to ensure
repeatability.
3. Theoretical development
The main objective of the model is to simulate the steady state var-
iations of three major process variables, namely (i) the permeate ux


( J) (ii) radially averaged membrane surface concentration C m and
(iii) the permeate concentration (CP) with the applied transmembrane
pressure (TMP) and feed concentration (C0). Therefore, three simulta-
n
 o
neous equations of the form f i J; C m ; C P 0 are necessary.
i1; 3
From the well known osmotic pressure model [32] and the Spiegler
Kedem black box formulation of the membrane based on irreversible
thermodynamics [33], two simultaneous equations, out of three, may
be directly obtained. In order to get the third equation, a steady state
solute balance within the concentration boundary layer is to be derived.
The assumptions made for the model formulation are as follows:
1. The ow is steady, incompressible and turbulent.
2. Solute concentration is independent of angular coordinate (),
i.e. C = C(r, z).
3. Feed and retentate are Newtonian uids. This assumption may be
justied in terms of dilute feed concentration (=40 kg m3).
4. Diffusive transport of solute along radial direction is negligible com-
pared to convection.
5. Axial permeation velocity is small enough compared to the radial
velocity.
6. Concentration boundary layer is well within the viscous sub layer of
the fully developed momentum boundary layer.
3.1. Steady state concentration prole
For solute with low diffusivity (like BSA in the present study),
Schmidt number (Sc) is quite large and since c  Sc3 , thickness of
1
the concentration boundary layer becomes negligible compared to the
momentum boundary layer. Considering assumptions 5 and 6, the
axial velocity component (vz) within the thin concentration boundary
layer may be regarded equal and opposite to the permeate ux [34] as
the z axis is vertically upward (Fig. 1). So
vz J: 1
On the other hand, the radial velocity component (r) may be
expressed in terms of free stream radial velocity (V0) and the dimen-
sionless axial distance from the membrane ( z/) according to the
approximate formulation proposed by Lin [35]:
r V 0 f : 2
 D. Sarkar et al. / Desalination 354 (2014) 7686 79

However, for the present study, the velocity prole expression is Substituting the velocity components, vr and vz from Eqs. (6) and (1)
required only within the viscous sub layer (assumption 6). From the respectively, the following form is obtained:
universal velocity prole of turbulent ow [36], the radial velocity
 
distribution within the viscous sub layer can be expressed as a linear 1 Az C 2 C
C J D 2: 8
function of z: r r r z z

w z The appropriate boundary conditions of Eq. (8) are,

r : 3

(i) at r r i ; C C 0 for all z 9a
Analytical expression of the wall shear stress (w) for a conned C
radial jet has been reported by Glauert as a function of V0 and [37]. (ii) at z c ; 0 for all r 9b
z
The proposed form was exactly in accordance with the well known Z r
Blasius law [37]: 1 0
(iii) z 0 ; C C m  2 2  C r; z2rdr 9c
r 0 ri ri
 0:25
2
w 0:0225V 0 : 4
V 0
where C m is the radially averaged membrane surface concentration.
Since the thickness of concentration boundary layer is practically negli-
Using this expression, the variation of turbulent boundary layer gible compared to the same of its hydrodynamic counterpart, variation
thickness () with the radial distance was solved by Watson [38] as of the radial velocity component along the axial direction may be
 1 neglected. Accordingly, the axially averaged radial velocity, vr over the
14zt =5 concentration boundary layer can be calculated as
0:205 r: 5
Q
Z c
1
vr r vr z; r dz: 10
Combining Eqs. (3), (4) and (5) and noting that V 0 Q c
2rzt , the ex- 0

pression of radial velocity prole becomes

Using the expression of vr(z, r) as in Eq. (6), vr may be expressed as
z
vr r; z A 2 6 A c
r vr : 11

0:84 2 r2
where A 7:97  103 Q z Q
. To develop the solute concentration
t
c
As c  Sc3 , the nal working form of vr can be obtained using
1
prole in the RFM module, an annular control volume between
[r, r + r] and [z, z + z] is considered as shown in Fig. 3. Neglecting Eq. (5), after introducing an empirical correction factor (k) as
the diffusive transport along radial direction (assumption 4), the steady  1
state solute mass balance over the differential control volume yields Ak zt 5 1 13
vr Sc : 12
2 Q r
   
  C  C 
C2rzvr jr C2rzvr jrr D  2rzD  2rr
  z z z zz It may be noted that in the present module, the effective membrane
vz Cjz 2rrvz Cjzz 2rr 0 7 radius is only 0.041 m. Therefore, the entrance and exit effects cannot be
1 rvr C C 2 C fully neglected. However, incorporation of these effects in the respective
vz D 2: momentum balance equation gives rise to severe complication, such
r r z z
that the analytical solution of the velocity prole becomes impossible.
The relation between wall shear stress and hydrodynamic boundary
layer thickness, as in Eq. (4) is generally accepted, but in case of mem-
brane separation processes, w is reported to be additionally dependent
on membrane roughness, porosity and on entranceexit effects [39]. Ac-
cordingly, the correction factor k, as used in Eq. (12), may be inferred to
compensate for the errors introduced due to (i) entrance-exit effects,
(ii) effects of membrane properties (roughness, porosity) on velocity
prole and (iii) lack of coefcient data in the relation between c
and . So the nal expressions of pertinent velocity components, to be
used in the solute mass balance equation are

vz J 13a

and
a
vr 13b
r

1
zt 5 13
where a Ak
2 Q Sc . Using these expressions, Eq. (7) may be
modied to a much simpler form compared to Eq. (8), which is as
follows

Fig. 3. Different incoming and outgoing uxes over the annular shell between r to r + r 1 aC C 2 C
J D 2: 14
and z to z + z (not in scale). r r z z
80 D. Sarkar et al. / Desalination 354 (2014) 7686

h

The respective boundary conditions remain unchanged, as in Eq. (9). Combining Eqs. (19) and (20) m C m and P = (CP), whereas
Eq. (14) can be solved by the standard variable separation method. The
= m P], the equation of the desired functional form, f 2
solution can be represented as

J; C m ; C P ; 0 is obtained.

C0 m2 zzt m m zzt 2a r2 r2i 2
 3

C r; z m2 e 2e 1 e 15 1 9:22  10
3
C m C P

m1 RT 4
1 P
 5  C m C P
m1 5 2
MW 3:01  10  C m C m C P C P 2

J 0
p
p
Rm
J J 4D
2
J
J 2 4D
where m1 2D m2 ; 2D . is equivalent to the single 21
effective eigen value of Eq. (14). It may be noted that can be evaluated
using the following equation, which is an outcome of the third boundary


condition (Eq. (9c)). 3.4. SpieglerKedem black box model: Formulation of f 3 J;C m ; C P ; 0

h 2 2 i
Permeate ux for UF, NF and RO membranes can be described by the
2aC 0 e2ar0 ri 1 m z m m z

Cm   e 2 t 2 e 1 t 16 standard SpieglerKedem black box model [33]. The model relates the
 2 2 m m1
r 0 r i 1 2 real rejection (Rr) to the permeate ux through solute permeability
m1 (PS) and osmotic reection coefcient () as process parameters

CP 1 F
Rr 1 22
However, C m is unknown
n
beforehand. Sooin order to solve for , three Cm 1 F
equations of the form f i J; C m ; C P 0 ; i 1; 3 as stated previ- n o
ously, are required. where, F exp 1
P
J
. Eqs. (16), (18), (21) and (22) can be simulta-
S

neously solved to predict the steady state values of J; C m ; C P and under

3.2.

Solute mass  balance at the membrane surface: Formulation of
different parametric conditions of TMP and feed concentration (C0).
f 1 J;C m ; C P ; 0

4. Simulation technique
Under steady state, the net solute ux through the membrane must
be balanced by the Fickian back diffusion ux. Accordingly, the solute
4.1. Estimation of process parameters
balance equation at the membrane surface becomes

 Experimental validation of the proposed model requires the numer-

 C  ical values of different process parameters (appearing in Eqs. (16), (18),
J C m C P D  17
z  (21) and (22)), which are D, Rm, , Ps, , and k.
z0
The solute diffusivity (D) may be evaluated by the following empir-
r
ical correlation [41].
where, C C z 1
0 C r; z2rdz is the radially averaged solute
r20 r2i ri 1
9
concentration at any D 2:74  10 M Pol3 23

axial location.  Combining Eqs. (15), (16) and (17),
the nal form of f 1 J; C m ; C P ; 0 is as under
For BSA, the average molecular weight is around 67,000 g mol1. Ac-
 r2
cordingly the diffusivity was estimated to be 6.74 1011 m2 s1,
2A
i
Dm2 C 0 e  m z m z 
e2A 0 i 1 e 1 t e 2 t :
2
r r
2
which is fairly close to the respective diffusivity value reported by
J  18
m2  2 2  Opang and Zydney [42]
1 r 0 r i
m1 2A The osmotic reection coefcient () and solute permeability (PS)
were evaluated by the well known iterative method of Nakao and
Kimura [43]. For the present solutemembrane combination the values
were found to be 0.894 and 2.47 107 m s1 respectively.


3.3. Osmotic pressure model: formulation of f 2 J;C m ; C P ; 0 The hydraulic resistance of the membrane was determined by
making a series of water runs with the present radial ow cell after
In any standard UF process, the osmotic pressure at the membrane allowing the initial membrane compaction. The mean Rm, estimated to
surface cannot be neglected because of high polarization effects. So be 1.69 1012 m1 from different experimental runs, was used in the
the permeate ux may be evaluated by the phenomenological equation present study. The density and viscosity of BSAwater solution depends
of osmotic pressure model [32], described as primarily on the solute concentration.
The reported correlations at 30 C [40], as used in the present study
are as follows
P
J 19

Rm 4 0 2 3 3
1:00021 2:856  10 C  10 kg m 24

where is the osmotic reection coefcient and Rm is the hydraulic re-


sistance of the membrane. Regarding the evaluation , the following 0
4 0 2 3
0:94 0:21C 1:329  10 C  10 Pa s: 25
empirical correlation of as a function of solute concentration (BSA as
used in the present study) is used [40]
The hydrodynamic parameter (k), introduced to account for
(i) entrance-exit effects, (ii) effects of membrane properties on velocity
RT
3 2 5 3

prole and (iii) lack of coefcient data in the relation between c and
C 9:22  10 C 3:01  10 C : 20
MW was impossible to estimate analytically. Therefore, the same was
 D. Sarkar et al. / Desalination 354 (2014) 7686
estimated by minimizing the sum of the mean square error between the
experimental and model predicted permeate ux values under different
parametric conditions. Accordingly, the model may be regarded as a
single-adjustable parameter, semi-empirical, steady state model of the
present RFM module.
4.2. Simulation algorithm
The simulation scheme to obtain the steady state values of permeate


ux (J), membrane surface concentration C m and permeate concen-
tration (CP), under different parametric conditions is shown in Fig. 4.
Eqs. (16), (18), (21) and (22) are simultaneously solved using the stan-
dard predictorcorrector Newton method with a suitable tolerance. As
mentioned previously, the only adjustable parameter of the model, k
was determined by minimizing the following global objective function
E(k), dened as
!2
1X n
J i;model
Ek 1 26
n i1 J i;exp
where, n is the total number of data points marked by different TMP
and C0. It is evident from the denition that the function E(k) actually
represents the sum of the normalized square error between the exper-
imental and model predicted permeate ux over the entire parametric
space of the present study. The direct search technique with accelera-
tion [44] was used to determine the optimum value of the parameter
k. A small tolerance limit of 0.001 for E(k) was used as a stop criterion
of direct search algorithm. C m C 0 , CP = 0, J P
Rm
C 0
, = 1 and
k = 1 were the initial values/guess values used in the present study.
Fig. 4. Flow chart showing the steady state simulation scheme for permeate ux,
membrane surface concentration and permeates concentration of RFM module.
81
5. Results and discussion
5.1. Fixing the adjustable parameter (k)
The only adjustable parameter of the model (k) was evaluated by
minimizing the global objective function, dened in Eq. (26). It was
found that a single k value of 0.37 resulted in a minimum E(k) =
6.24 10 4 (dimensionless), which is the average of normalized
square error for 13 experimental data points according to the CCRD
scheme. The parameter (k) was introduced in the velocity prole equa-
tion primarily to account for the entranceexit effects and the effects of
membrane properties on velocity prole as mentioned earlier. There-
fore, k must be primarily dependent on TMP and on C0 to some extent.
The present module was fed through a narrow channel of thickness
0.001 m, where from the radial jet had to suddenly expand over the
full axial span of zt. At the outlet, the retentate ow was again subjected
to a sudden contraction from zt to the diameter

of the peripheral holes,
i.e., 0.005 m. As axial to radial span ratio r0 zr
t
i
of the present module
was 0.56, the entrance and exit effects must be dependent on the re-
spective TMP. Effects of feed concentration on entrance and exit velocity
proles might be negligible. On the other hand, a change in C0 would
lead to a change in membraneuid interaction through density and
viscosity effects. So in true sense, the parameter k is inferred to be
dependent on TMP and C0. However, the dependence was weak; other-
wise a single adjusted k value would not have resulted into a minimum
E(k) of 4.08 104. In order to conrm the nature of the dependence,
the same k value (= 0.37) was used to simulate the permeate ux,
membrane surface concentration and permeate concentration over an
extrapolated parametric space marked by TMP [196.2 kPa, 392.4
kPa] and feed concentration [5 kg m3, 40 kg m3]. Over the extrap-
olated space, E(k) was found to be 6.24 104, which clearly conrms
the apprehended nature of dependence. The contribution of end effects
(entrance and exit effects) or the effects of membrane roughness on k
cannot be predicted separately because of the gross empiricity inbuilt
in the denition of the same parameter.
5.2. Variation of steady state ux with TMP and C0
Fig. 5 represents the steady state permeate ux as a function of TMP
at different bulk concentration. As evident from the osmotic pressure
model, the permeate ux was observed to increase with TMP. On the
other hand, because of higher osmotic pressure differential, the ux
exhibits a decreasing trend with C0. In both of the cases, the variation
J
was nearly linear. For example, at C0 = 20 kg m3, TMP was found to
be 5.89 107 m3 m2 s1 kPa1 over the TMP range of [196.2 kPa,
294.3 kPa], whereas the same was 5.02 10 7 m3 m 2 s 1 kPa1
over the next higher range i.e. [294.3 kPa, 392.4 kPa]. Similar trend
was observed with respect to C0, though the respective prole was
slightly nonlinear. As a result, the contour plot of the steady permeate
ux becomes marginally nonlinear (Fig. 5 insert). This is exactly consis-
tent with the trends predicted by osmotic pressure model. According to
Eq. (19), the steady ux has to be a linear function of TMP, provided ,
and Rm remains constant. However, is nonlinear in terms of C m and
C m is supposed to increase in a complex, nonlinear fashion with C0. This
may be attributed to the phenomenon of concentration polarization. It
is well known that any UF process is marked by the presence of severe
polarization resulting a sharp concentration gradient between the
membrane surface and bulk uid. Hence the average membrane surface
concentration (C m ) must be a nonlinear function of C0. Subsequently,
the permeate ux becomes essentially nonlinear with respect to C0.
Still the degree of nonlinearity is pretty low, as indicated by the contour
plot. This may be explained once again in terms of osmotic pressure
model in conjugation with the empirical functionality between
and C (Eq. (20)). Eq. (20) reveals that the ratio of quadratic to
linear term is in the order of 102 C. As the experimental range of C0
was 540 kg m3 of BSA in the present study, which may be marked
82 D. Sarkar et al. / Desalination 354 (2014) 7686
Fig. 5. Variation of the steady state permeate ux with TMP at different bulk concentration (inserts showing (i) contour plot of J as a function of TMP and C0 and (ii) parity diagram of
experimental and model predicted permeate ux).
as lower side values of C0, the contribution of quadratic term becomes
negligible compared to the linear one. Essentially the steady ux may
be justied to be a nearly linear function of C m . On the other hand, the
nonlinearity between C m and C0 is negligible at low C0 owing to a
small thickness of the polarized layer. As a consequence, variation of
the steady ux with respect to TMP and C0 becomes approximately
linear as indicated by the contour plot (Fig. 5).
The parity diagram, as shown in the insert of Fig. 5, other than the
contour plot, clearly points out that the proposed model can predict
the variation of steady ux with TMP and C0 with reasonable accuracy.
A closer investigation reveals that the average value of the normalized
s

2
n J
j1 1 iJj:model
standard deviation, dened as n1
i j;exp
for the present
study was found to be 0.025, which is well within the acceptable limit.
5.3. Steady state real rejection as a function of TMP and C0
Real rejection, as described by Eq. (22) is an important parameter
to characterize the performance of any membrane separation unit in
general. Compared to the observed rejection, dened as Ro 1 CC P0
(where, C0 is the bulk uid concentration), real rejection is not marked
by the phenomenon of concentration polarization and hence indicates
the true separation efciency of the membrane. The model predicted
variation of Rr with TMP at different C0 over the extrapolated parametric
space, as mentioned earlier is shown in Fig. 6. As expected, the real re-
jection shows an increasing trend both with TMP and C0. For example,
at C0 = 5 kg m3, real rejection was observed to increase from 0.78 to
0.82 (=5.1%) for a change of TMP from 196.2 kPa to 392.4 kPa. Similarly
at TMP = 392.4 kPa, the same quantity exhibits an increase of 13.2%
with a change of C0 from 5 to 40 kg m3. In standard ultraltration pro-
cesses, at increased TMP and/or C0 the solute rejection increases rapidly
leading to a growth of the polarized layer.
The layer acts as an additional mass transfer resistance promoting
further growth. However, the monotonicity levels off to a steady state
value due to the tangential velocity led membrane shear and the Fickian
back diffusion. Generally in a cross ow module, the membrane shear is
primarily dependent on the feed/retentate ow rate. Similarly in the
9
present module, a monotonic relation between w and Q (w Q 5 )
may be established from Eqs. (4) and (5). It is to be noted that the
present study was conducted at a xed retentate ow rate of Q =
6.4 10 6 m3 s 1. Effect of C0 on w through may be assumed to
be negligible small as the study was conducted with dilute feed solution
(C0 : 5 to 40 kg m3). Additionally, w has got no specic functional
dependence on TMP, which can be justied from Eqs. (4) and (5). There-
fore, a negligibly small variation of w over the parametric range of the
present study may be conrmed. On the other hand, the solute rejection
by the membrane surface, due to the higher rate of transport from the
 D. Sarkar et al. / Desalination 354 (2014) 7686 83

Fig. 6. Model predicted real rejection as a function of TMP at different C0 (inserts showing the variation of experimental and model predicted permeate concentration with TMP at
different C0).

bulk is expected to increase monotonically with TMP and C0. Accordingly
the approximate linear increasing trend of Rr with the two process param-
eters of the present study becomes evident as indicated by Fig. 6.
In order to establish the validity of the proposed model, experimental
CP values were compared with the corresponding model predictions at
different TMP and C0. Insert of Fig. 6 illustrates the respective variations
with TMP at different C0. For all the experimental data points, the nor-
s

2
n
j1 1 CPiPij;model
C
malized standard deviation, dened as C P n1
j;exp
, was
as low as 0.078, which clearly indicates the validity of the proposed
model. The gure also reveals that CP exhibits a sluggish exponential
growth with TMP at different bulk concentrations. The trend may be ex-
plained in terms of growth saturation of the polarized layer at higher side
values of TMP. However, the growth saturation does not mean a constant
mass transfer resistance, with respect to the solvent (water) as well as
the solute. For solvent, it may attain a constant value, leading to a near
linear increase of solvent ux (J) with TMP. But in case of the solute,
the resistance may go on increasing, exactly counter balancing the effect
of enhanced driving force (TMP). For example, at TMP = 196.2 kPa,
C P has increased by 129% for a change of C 0 from 5 to 10 kg m 3 ,
whereas only 18% change of CP was observed for C0 change from 30
to 40 kg m 3.
5.4. Variation of the effective eigen value equivalent term of Eq. (9)
The parameter , equivalent to the effective eigen value of Eq. (9),
was obtained by solving Eqs. (16), (18), (21) and (22) along with J, C m
and Cp. was observed to be strongly dependent on TMP and C0. There-
fore, may be represented as function of these two parameters. In order
to decipher the respective functional form, response surface methodol-
ogy (RSM) was adopted with the previously mentioned 22 central com-
posite rotatable design (CCRD). Standard regression models, which
include (i) linear, (ii) quadratic, (iii) 2FI, and (iv) cubic were tested to
represent the response () as a function of independent/regressor vari-
ables. The software package of Design Expert (version 8.0) was used for
the present RSM analysis. From the t summary, multivariable quadrat-
ic regression model was chosen to be the optimum in representing the
variation of . Cubic and quartic models were also tested to t the
present data, though the test parameters did not show any signicant
improvement over the chosen quadratic model. The corresponding

Fig. 7. (a) Parity diagram of for quadratic regression model. (b) Contour plot of as a function of TMP and C0.
84 D. Sarkar et al. / Desalination 354 (2014) 7686

parity diagram, as shown in Fig. 7(a) also justies model selection. The
nal working equation of , as a function of TMP and C0 becomes
2:28757 0:037577  TMP0:40838  C 0 6:00  10
6 2 3 2
C 0 6:73907  10  TMP 8:70657  10  C0:
The model F value of 2079.4 implies the model is signicant. More-
over, high values of R2 (=0.99) as well as adjusted R2 (=0.99) repre-
sent the gross validity of the regression model. It is well known that
R2 always increases as more terms are added to the model, whereas ad-
justed R2 statistic often decreases upon adding unnecessary terms. For
the present model, up to two decimal places equality of R2 and adjusted
R2 conrms the presence of signicant terms only. Moreover, adequate
precision (signal to noise ratio N 4 is desirable) of 140.583 suggests that
the model equation can be used to navigate the entire parametric space.
So the overall predictive capability of the quadratic regression model
based on these criteria seems to be quite satisfactory. The contour
plot, representing the variation of is shown in Fig. 7(b). However, a
complete characterization of the response cannot be done only with
the contour/response surface plots. Therefore a more formal canonical
analysis was carried out to reveal the true nature of the response. The
model equation (Eq. (27)) may be represented in a more compact form
shear rate arrests the growth of the polarized solute layer and thereby
increases the permeate ow. Naturally, the permeate ux is expected
to be higher in any DSE modules compared to the most advanced
4
 TMP cross ow unit. However, the specic ltration areas of DSE modules
27 are insignicant in relation to the simplest cross ow modules. For ex-
ample, the specic surface area of the reported RDM modules is only
3.62 m2 m 3, whereas the same for the present RFM module is
60.55 m2 m3. Therefore, the gross permeate ow becomes higher in
case of cross ow devices. On the other hand, in case of electric eld en-
hanced cross ow ltration, the maximum steady state permeate ux
was reported to be 4.44 105 m3 m2 s1 at a TMP of 635 kPa [47],
which is only 11.1% higher than the maximum ux of the present
study. For air sparged ultraltration experiments, using a tubular mem-
brane module with human serum albumin (HSA) as the test media, the
reported maximum value of permeate ux was 7.3 105 m3 m2 s1
[48], though the bubble size and frequency have strong inuence on the
permeate ux. From the foregoing discussion, it becomes clear that the
performance of the proposed RFM module is comparable with other
cross ow and DSE units. Moreover, the RFM module is devoid of the
complicated internal arrangements necessary for gas sparging, electric
eld or rotating membrane.

T
b0 Xb XBX
" #
b12
b11
where X = [TMP C 0] T , b = [b1 b 2] and B b12
2 . The eigen
2 b22
values of the matrix B for the present system are [0.0001 0.0087] T.
This implies that the response surface must pass through a mini-
1
hmum, though the corresponding
i stationary point, Xs 12 B b
3
7177:5 kPa 223:9 kg m is far outside the region of exploration.
5.5. Performance comparison with other membrane modules
A legion of studies on ultraltration of BSA and using different mod-
ules with different types of membrane was reported. Performances of
some of them in terms of operating conditions and the respective steady
state permeate ux are reviewed here in comparison to the proposed
RFM module and are presented in Table 1 [40,4548]. Amongst the
other studies, the highest permeate ux as reported by Sarkar et al.
[40] was 2.0 10 4 m 3 m 2 s 1 in the ultraltration of BSA
using a standard rotating disk-membrane (RDM) module. This is
followed by the permeate ux obtained from a single stirred cell
(= 11.11 10 5 m 3 m 2 s 1 ) [46]. It may be noted that both the
studies were conducted in standard DSE modules in which membrane
shear can be increased independent of feed ow rate. The enhanced
23 6. Conclusion
The present investigation illustrates the design, performance charac-
terization and modeling of a new cross ow module. The module has
been named as radial ow membrane (RFM), owing to the directional-
ity of ow. In the present study, bovine serum albumin (BSA) was sep-
arated from its synthetic aqueous solution using the present module
tted with polyethersulfone (PES) membrane of 50 kDa MWCO. The
steady state performance of the proposed module was investigated
under different transmembrane pressure (TMP) and feed concentration
(C0). A maximum permeate ux of 4.0 105 m3 m2 s1 was obtain-
ed at TMP = 392 kPa C0 = 5 kg m3. In general, an average permeate
ux enhancement of 0.06 10 6 m3 m 2 s 1 per unit kPa rise of
TMP was recorded over the parameter range of the present study. Addi-
tionally, a single parameter, semi-empirical model was proposed to
simulate the RFM module. The only adjustable parameter of the pro-
posed model was introduced primarily to account for the unknown
entranceexit effects and the effects of membrane properties (rough-
ness, porosity) on velocity prole. The respective parameter was
evaluated by minimizing a global objective function equivalent to the
sum of the normalized square error between the experimental and
the model predicted permeate ux under steady state. The maximum
absolute deviation of 7% establishes the gross validity of the model in
simulating the steady state performance of the RFM module.

Table 1
Comparison of the proposed RFM module with the reported DSE and cross ow modules in terms of permeate ux.

Module Operating conditions Membrane/solute combination Steady permeate ux Reference

(m3 m2 s1)

Standard cross ow cell C0: 20 kg m3 Polyamide/BSA 1.25 105 Wang and Tang [45]
TMP: 1240 kPa
Stirred cell C0: 0.6 kg m3 Polysulfone/BSA 11.11 105 Gell and Davis [46]
TMP: 71 kPa
Electric eld enhanced rectangular C0: 0.11.5 kg m3 Polyphenylene ethersulfone/BSA 4.44 105 Sarkar et al. [47]
cross ow channel TMP: 220635 kPa
Velocity: 0.060.18 m s1
Electric eld: 01000 V m1
Tubular cross ow module with air C0: 4.5 kg m3 Polyvinylidene uoride/human 7.3 105 Li et al. [48]
sparging facility TMP: 101.32 kPa serum albumin (HAS)
Rotating disk-membrane module C0: 130 kg m3 Polyethersulphone/BSA 2.0 104 Sarkar et al. [40]
TMP: 294882 kPa
Stirrer speed: 5.230.9 rad s1
Membrane speed: 5.230.9 rad s1
Radial ow membrane module C0: 540 kg m3 Polyethersulphone/BSA 4.0 105 Present study
TMP: 196.2392 kPa
 D. Sarkar et al. / Desalination 354 (2014) 7686 85

Nomenclature p permeate
a parameter dened by Eq. (13b) pol polymer
A parameter dened by Eq. (6) 0 bulk
B coefcient matrix dened by Eq. (23)
b vector dened by Eq. (23)
b0 parameter dened by Eq. (23) Superscript
C solute concentration (kg m3)
C' solute concentration (gm cm3) T transpose
C average solute concentration (kg m3)
Cm average membrane surface concentration (kg m3)
CP permeate concentration (kg m3)
C0 feed concentration (kg m3) References
D solute diffusivity (m2 s1)
[1] N.N. Li, A.G. Fane, W.S.W. Ho, T. Matsuura, Advanced Membrane Technology and
E(k) global objective function dened by Eq. (21) Applications, John Wiley & Sons, New York, 2011.
F parameter dened by Eq. (22) [2] B. Nicolaisen, Developments in membrane technology for water treatment, Desali-
J permeate ux (m3 m2 s1) nation 153 (2003) 355360.
[3] S.P. Nunes, K.V. Peinemann, Membrane Technology: In the Chemical Industry, John
Ji,exp experimental permeate ux at ith data point (m3 m2 s1) Wiley & Sons, New York, 2006.
Ji,model model predicted permeate ux at ith data point (m3 m2 s1) [4] S.S. Sablani, M.F.A. Goosen, R. Al-Belushi, M. Will, Concentration polarization in
k parameter dened by Eq. (12) ultraltration and reverse osmosis: a critical review, Desalination 141 (2001)
269289.
m1, m2 parameters dened in Eq.(15) (m1) [5] S. Curcio, G. Scilingo, V. Calabr, G. Lorio, Ultraltration of BSA in pulsating condi-
MW molecular weight (kg kmol1) tions: an articial neural networks approach, J. Membr. Sci. 246 (2005) 235247.
n total number of data points [6] A. Fouladitajar, F.Z. Ashtiani, H. Rezaei, A. Haghmoradi, A. Kargari, J. Ind. Eng. Chem.
20 (2014) 624632.
Ps solute permeability (m s1) [7] B. Venkataganesh, Abhijit Maiti, Subir Bhattacharjee, Sirshendu De, Electric eld
Q feed/retentate ow rate (m3 s1) assisted cross ow micellar enhanced ultraltration for removal of naphthenic
r radial coordinate (m) acid, Sep. Purif. Technol. 98 (2012) 3645.
[8] Biswajit Sarkar, Sirshendu De, Prediction of permeate ux for turbulent ow in cross
R universal gas constant (J mol1 k1)
ow electric eld assisted ultraltration, J. Membr. Sci. 369 (2011) 7787.
Rm hydraulic resistance of the membrane (m1) [9] M. Cai, S. Zhao, H. Liang, Mechanisms for the enhancement of ultraltration and
Rr real rejection (dimensionless) membrane cleaning by different ultrasonic frequencies, Desalination 263 (2010)
133138.
Ro observed rejection (dimensionless)
[10] M. Cai, S. Wang, H. Liang, Optimization of ultrasound-assisted ultraltration of Radix
ri radius of the impingement plate (m) astragalus extracts with hollow ber membrane using response surface methodolo-
r0 membrane radius (m) gy, Sep. Purif. Technol. 100 (2012) 7481.
Sc Schmidt number (dimensionless) [11] D.E. Potts, R.C. Ahlert, S.S. Wang, A critical review of fouling of reverse osmosis
membranes, Desalination 36 (1981) 235264.
t time (s) [12] N. Hilal, O.O. Ogunbiyi, N.J. Miles, R. Nigmatullin, Methods employed for control of
T temperature (K) fouling in MF and UF membranes: a comprehensive review, Sep. Sci. Technol. 40
TMP transmembrane pressure (kPa) (2005) 19572005.
[13] J.G. Wijmans, S. Nakao, J.W.A. Van Den Berg, F.R. Troelstra, C.A. Smolders, Hydrody-
r radial velocity component (m s1) namic resistance of concentration polarization boundary layers in ultraltration, J.
vr average radial velocity component (m s1) Membr. Sci. 22 (1985) 117135.
Z axial velocity component (m s1) [14] H. Choi, K. Zhang, D.D. Dionysiou, D.B. Oerther, G.A. Sorial, Inuence of cross-ow
velocity on membrane performance during ltration of biological suspension, J.
V0 free stream radial velocity (m s1) Membr. Sci. 248 (2005) 189199.
X vector dened by Eq. (23) [15] Y.L. Li, K.L. Tung, CFD simulation of uid ow through spacer-lled membrane
z axial coordinate (m) module: selecting suitable cell types for periodic boundary conditions, Desalination
233 (2003) 351358.
zt axial distance between the membrane and the top lid (m)
[16] J.N. Ghogomu, C. Guigui, J.C. Rouch, M.J. Clifton, P. Aptel, Hollow ber membrane
module design: comparison of different curved geometries with Dean vortices, J.
Membr. Sci. 181 (2001) 7180.
Greek letters [17] K. Clint, Module with self-supporting sheet membrane, patent no: PCT WO 03/059494
A1 (US), PCT/US02/10375 (2002).
density of the feed (kg m3) [18] M.E. Brewster, K.Y. Chung, G. Belfort, Dean vortices with wall ux in a curved chan-
nel membrane system: 1. A new approach to membrane module design, J. Membr.
w wall shear stress (Pa) Sci. 81 (1993) 127137.
osmotic pressure (Pa) [19] J. Schwinge, D.E. Wiley, A.G. Fane, Novel spacer design improves observed ux, J.
solution viscosity (Pa s) Membr. Sci. 229 (2004) 5361.
[20] A.G. Fane, S. Chang, E. Chardon, Submerged hollow bre membrane moduledesign
osmotic reection coefcient (dimensionless) options and operational considerations, Desalination 146 (2002) 231236.
normalized standard deviation used for permeate ux [21] M.Y. Jaffrin, Dynamic shear-enhanced membrane ltration: a review of rotating
(dimensionless) disks, rotating membranes and vibrating systems, J. Membr. Sci. 324 (2008) 725.
[22] A. Saxena, B.P. Tripathi, M. Kumar, V.K. Shahi, Membrane-based techniques for the
CP normalized standard deviation used for permeate concentra-
separation and purication of proteins: an overview, Adv. Colloid Interf. Sci. 145
tion (dimensionless) (2009) 122.
effective eigen value equivalent term of Eq. (14) [23] H.P. Feuerpeil, D. Blas and H. Olapinski, Aaowsystems GmbH, German Patent DE
102 39 247 C1 (2003).
thickness of the hydrodynamic boundary layer (m)
[24] M.Y. Jaffrin, G. He, L.H. Ding, P. Paullier, Effect of membrane overlapping on perfor-
C thickness of the concentration boundary layer (m) mance of multishaft rotating ceramic disk membranes, Desalination 200 (2006)
kinematic viscosity (m2 s1) 269271.
 z [25] G. He, L.H. Ding, P. Paullier, M.Y. Jaffrin, Experimental study of a dynamic ltration
dimensionless axial distance from the membrane sur-
system with overlapping ceramic membranes and non-permeating disks rotating
face (dimensionless) at independent speeds, J. Membr. Sci. 300 (2007) 6370.
[26] D. Sarkar, C. Bhattacharjee, Modeling and analytical simulation of rotating disk
membrane module, J. Membr. Sci. 320 (2008) 344355.
Subscript [27] K.H. Kroner, V. Nissingen, Dynamic ltration of microbial suspensions using an
axially rotating lter, J. Membr. Sci. 36 (1988) 85100.
[28] X. Yang, R. Wang, A.G. Fane, C.Y. Tang, I.G. Wenten, Membrane module design and
s solute dynamic shear-induced techniques to enhance liquid separation by hollow ber
m membrane module: a review, Desalin. Water Treat. 51 (2013) 36043627.
86 D. Sarkar et al. / Desalination 354 (2014) 7686
[29] S.K. Karode, Laminar ow in channels with porous walls, revisited, J. Membr. Sci.
191 (2001) 237241.
[30] E. Layne, Spectrophotometric and turbidimetric methods for measuring proteins,
Methods Enzymol. 3 (1957) 447455.
[31] D.C. Montgomery, Design and analysis of Experiments, 7th edition John Willey &
Sons, Inc., U.K., 2009.
[32] O. Kedem, A. Katchalsky, The physical interpretation of phenomenological coef-
cients of membrane permeability, J. Gen. Physiol. 45 (1961) 143179.
[33] Z.V.P. Murthy, Latesh B. Chaudhari, Separation of cadmium ions and estimation of
membrane transport parameters of a nanoltration membrane, Indian J. Chem.
Technol. 15 (2008) 107112.
[34] S. De, S. Bhattacharya, P.K. Bhattacharya, A. Sharma, Generalized integral and
similarity solution of concentration prole for osmotic pressure controlled ultral-
tration, J. Membr. Sci. 130 (1997) 99121.
[35] C.C. Lin, On the stability of two dimensional parallel ows. III. Stability in a viscous
uid, Quat, Appl. Math. 3 (1946) 277301.
[36] R.W. Fox, A.T. McDonald, Introduction to Fluid Mechanics, Wiley, New York, 1985.
[37] M.B. Glauert, The wall jet, J. Fluid Mech. 1 (1956) 625643.
[38] E.J. Watson, The radial spread of a liquid jet over a horizontal plane, J. Fluid Mech. 20
(1964) 481499.
[39] V. Gekas, B. Hallstm, Mass transfer in the membrane concentration polarization
layer under turbulent cross ow, J. Membr. Sci. 30 (1987) 153170.
[40] D. Sarkar, D. Datta, D. Sen, C. Bhattacharjee, Simulation of continuous stirred rotating
disk-membrane module: an approach based on surface renewal theory, Chem. Eng.
Sci. 66 (2011) 25542567.
[41] T.K. Sherwood, R.L. Pigford, C.R. Wile, Mass Transfer, McGraw Hill, New York, 1980.
[42] W.S. Opong, A.L. Zydney, Diffusive and convective protein transport through
asymmetric membranes, AIChE J 37 (1991) 14971510.
[43] S. Nakao, S. Kimura, Analysis of solute rejection in ultraltration, J. Chem. Eng. Jpn 14
(1981) 3237.
[44] S.G. Beveridge, R.S. Schector, Optimization: Theory and Practice, McGraw-Hill, Inc.,
USA, 1970.
[45] Y.N. Wang, C.Y. Tang, Protein fouling of nanoltration, reverse osmosis, and ultral-
tration membranesThe role of hydrodynamic conditions, solution chemistry, and
membrane properties, J. Membr. Sci. 376 (2011) 275282.
[46] C. Gell, R.H. Davis, Membrane fouling during microltration of protein mixtures, J.
Membr. Sci. 119 (1996) 269284.
[47] B. Sarkar, A. Sengupta, S. De, S. DasGupta, Prediction of permeate ux during electric
eld enhanced cross-ow ultraltrationA neural network approach, Sep. Purif.
Technol. 65 (2009) 260268.
[48] Q.Y. Li, Z.F. Cui, D.S. Pepper, Effect of bubble size and frequency on the permeate ux
of gas sparged ultraltration with tubular membranes, Chem. Eng. J. 67 (1997)
7175.