Chemical Engineering Science 61 (2006) 5057 – 5069

www.elsevier.com/locate/ces

Ultrafiltration modeling of multiple solutes system for continuous

cross-flow process
A.L. Ahmad ∗ , M.F. Chong, S. Bhatia
School of Chemical Engineering, Engineering Campus, Universiti Sains Malaysia, Seri Ampangan, 14300 Nibong Tebal, Penang, Malaysia

Received 19 August 2005; received in revised form 17 February 2006; accepted 10 March 2006
Available online 12 March 2006

Abstract
A mathematical model suitable for the multiple solutes system in continuous cross-flow ultrafiltration is developed. This model is based on
mass balance analysis coupled with the filtration theory (osmotic pressure model, Kozeny–Carman equation), resistance-in-series model and
gel polarization model. This model is characterized by the parameters Rm , Pm , Kb , Kbi and ki . These parameters are estimated by using
the Levenberg–Marquardt method coupled with the Gauss–Newton algorithm based on the experimental data obtained from the treatment of
pretreated palm oil mill effluent (POME) as a feed in the pilot plant scale ultrafiltration system. The pretreated POME is composed of a ternary
system with the solutes of carbohydrate constituents, crude protein and ammoniacal nitrogen. The simulation results show a good agreement
with the experimental data. The proposed model is suitable for predicting the performance of multiple solutes in an ultrafiltration process. The
concentration of each solute present is correlated with the COD value of the permeate stream.
䉷 2006 Elsevier Ltd. All rights reserved.

Keywords: Membranes; Modeling; Filtration; Mathematical; POME; COD

1. Introduction
Ultrafiltration process uses a porous membrane to separate
water and microsolutes from the solution containing macro-
molecules and colloids. The average pore diameter of the ul-
trafiltration membrane lies from 10 to 1000 Å (Baker, 2000).
Ultrafiltration has become an increasing important separation
process for the concentration, purification or dewatering of
macromolecules and colloidal species in the solution (Yeh et al.,
2004). It is gaining importance for water and industrial wastew-
ater treatment especially in the recovery of chemicals from the
industrial wastewater, desalination, drinking water purification,
removal of oil from oil–water emulsions (Tansel et al., 2000).
The key factor determining the performance of the ultrafil-
tration membranes is the gel polarization where the rejected
solutes deposited on the membrane surface. The gel polariza-
tion causes membrane fouling and leads to the drop in permeate
∗ Corresponding author. Tel.: +60 4 5937788; fax: +60 4 5941013.
E-mail address: chlatif@eng.usm.my (A.L. Ahmad).
0009-2509/$ - see front matter 䉷 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2006.03.017
flux. This gel layer of rejected solutes on the membrane surface
forms a second barrier to the flow through the membrane.
A number of mathematical models are available in the litera-
ture to describe the transport mechanism through the ultrafiltra-
tion membranes by considering the gel polarization phenomena.
The most common model used in describing the ultrafiltration
is the resistance-in-series model (Tansel et al., 2000; Kumar et
al., 2004; Gehlert et al., 2005; Bhattacharjee and Datta, 2003).
The resistance-in-series model assumes that the flux of perme-
ate is proportional to the transmembrane pressure and inversely
proportional to the viscosity of the suspending solution where
the parameter of hydraulic resistance is being introduced. The
hydraulic resistance is the total resistance exerted by the mem-
brane and solutes.
Damak et al. (2004) developed a fluid dynamic model
for the cross-flow filtration tubular membranes. A method
of coupling the Navier–Stokes and Darcy equations, using
a finite difference technique to simulate laminar fluid flow
in the tube and in the porous wall was presented. Djurić
et al. (2004) proposed the flux (Jv ) models as function of
concentration factor (CF), flowrate (Q), temperature (T ) and
5058 A.L. Ahmad et al. / Chemical Engineering Science 61 (2006) 5057 – 5069
transmembrane pressure (
P ). The one-, two- and three-
variable functions are suggested: Jv−1 (CF), Jv−1 (CF, Q),
Jv−1 (CF, T ), Jv−1 (CF,
P ), Jv−1 (CF, Q, T ) and Jv−1 (CF, Q,

P ). These mathematical models predicted the separation time
if the initial and final concentrations of permeate are known.
A general membrane fouling model accounting the internal
clogging, partial and total clogging, cake deposition and cake
deposition with retroflux for ultrafiltration system was devel-
oped by Ghaffour (2004). The model predicted the permeate
flux and a relationship for the limiting flux against the bulk
concentration and it was developed based on the experimental
data. Katsikaris et al. (2005) combined the thin film theory and
Spiegler–Kedem’s model to obtain the mathematical model for
the simulation of ultrafiltration system and accounted the con-
centration polarization phenomena.
Yeh et al. (2004) derived the equations for the prediction of
permeate flux in a tubular membrane module from the momen-
tum balance with declining flowrate and transmembrane pres-
sure along the membrane tube. The model predicted the effects
of solute concentration, transmembrane pressure, tube radius
and tube length on the flux and pressure drop along the mem-
brane tube. Sondhi and Bhave (2001) combined the d’Arcy’s
permeability model and Kozeny–Carman equation to predict
the permeate flux and the cleaning time of ceramic ultrafiltra-
tion with back pulsing.
However, the models proposed in the literature are only suit-
able to predict the performance of ultrafiltration with the sin-
gle solute system. When multiple solutes system is used, these
models can only predict the total permeate concentration of the
system. In the industrial application, multiple solutes systems
are often encountered and the mass transport of each solute in
the multiple solutes system is important. In the present study,
the ultrafiltration system is used to separate the organic mat-
ters from pretreated palm oil mill effluent (POME). POME is
the liquid waste discharged from the palm oil mill with high
chemical oxygen demand (COD). The transport of the organic
matters (mainly carbohydrate constituents, crude protein and
ammoniacal nitrogen) through the ultrafiltration is very impor-
tant. The transport of these organic matters will finally deter-
mine the COD and ammoniacal nitrogen concentration in the
permeate. The COD and ammoniacal nitrogen concentration of
the final discharge should meet the standard discharge limit of
the Environmental Quality Act 1976 set by the Department of
Environment (DOE) of Malaysia.
The objective of the present study is to develop a time-
dependent mathematical model to predict the performance of
an ultrafiltration process in the separation of POME (multiple
solutes system). The model is based on the mass balance anal-
ysis coupled with the filtration theory (osmotic pressure model,
Darcy’s law), resistance-in-series and gel polarization models.
The measurable objectives in the present study are:
(1) To develop equations (time-dependent) for prediction of the
volume flux of permeate and gel layer resistance.
(2) To develop model for the prediction of concentrations
of each solute in the permeate of the multiple solutes
system.
(3) To estimate the parameters of the model from the experi-
mental data obtained from the ultrafiltration system.
(4) To validate the mathematical model by comparing the sim-
ulation results with the experimental data.
(5) To correlate the concentration of each solute with the COD
of the permeate obtained from the ultrafiltration system for
the treatment of POME.
2. Model development
A mathematical model is developed to provide a conceptual
framework for understanding the phenomenon responsible for
flux decline for the multiple solutes system in the ultrafiltration
process. During filtration process, the solvent is transported by
pressure-driven convective flow through the pores. Separation
occurs because the solvent is filtered through some of the pores
in the membrane. The ultrafiltration membrane rejects the so-
lutes by sieving action.
The membrane rejects the multiple solutes by steric hin-
drance where the membrane is considered a sieving barrier that
retains molecules according to their size. Besides van der Waals,
electrostatic or chemical interactions between membrane and
solutes or even among solutes themselves play a large part in
building of filtration selectivity. The total resulting effects are
complex. A typical case is the industrial wastewater containing
complex solutions, where all these interactions are involved on
numerous and often poorly understood solutes.
The POME, for example, is a complex mixture of organic
matters and all specific components of the organic matters
could not be determined. The carbohydrate constituents com-
prise of low molecular weight monosaccharide (180 kg/kmol) to
high molecular weight polysaccharide (400,000 kg/kmol). The
crude protein comprises from the simplest compounds of amino
acids (75 kg/kmol) to the most complex compounds of proteins
(450,000 kg/kmol). Nitrogen in POME is originally present in
the form of organic (protein) nitrogen, and as time progresses,
the organic nitrogen is gradually converted to ammoniacal ni-
trogen which has the molecular weight of 17–35 kg/kmol.
Darnon et al. (2002) had proposed a model based on the the-
ory of thermodynamic of irreversible process for ultrafiltration
system of complex biological solutions. The ultrafiltration of
a single protein (
-lactoglobulin) solution was compared with
a mixture of yeast extract and
-lactoglobulin. Based on the
obtained results, solute–solute interactions in the mixture were
negligible especially when the solutes were of biological type
and the system was operated at the pH away from their isoelec-
tric point. Therefore, solute–solute interaction can be neglected
in the pretreated POME system as the pretreated POME was of
biological origin (palm oil fruit) and the system was operated
at pH 6 which was away from its isoelectric point.
The membrane continuously rejects the solutes and the re-
jected solutes are deposited on the membrane surface. At the
same time, due to stirring action caused by the superficial ve-
locity, some solutes are removed from the membrane surface
and go into the bulk (back transport effect). The deposited layer
of rejected solutes on the membrane surface is termed as “gel
layer”. The rejected solutes are assumed to be deposited on the
 A.L. Ahmad et al. / Chemical Engineering Science 61 (2006) 5057 – 5069 5059

VbCb solutes in the gel layer, Cp is the total concentration of solutes

in the permeate and Cb is the total concentration of solutes in
Bulk Mbt Cb (Vp + Vg) the bulk.
The total mass balance of the system is the summation of
mass balance of each individual solute. Therefore, Eq. (1) can
Gel Layer also be written in terms of concentration of solute i in the gel
layer (Cgi ), concentration of solute i in the permeate (Cpi ),
concentration of solute i in the bulk (Cbi ) and back transport
mass of solute i(Mbti ) up to time t:
Membrane
VpCp n

n

n

n


Vg Cgi = (Vp + Vg ) Cbi − Vp Cpi − Mbti
Fig. 1. Overall mass balance over the multiple solutes system of continuous i=1 i=1 i=1 i=1
cross-flow ultrafiltration. for i = 1, 2, 3, . . . , n. (2)
It has been assumed that the rate of back transport mass for
membrane surface and resistance caused by internal clogging, each solute i is proportional to the superficial velocity, v, and
partial and total clogging are neglected. the concentration of the solute in the gel layer,Cgi :
Thus, in the multiple solutes system of continuous cross-flow
ultrafiltration, the assumptions include: (1) the membrane re- dMbti
∝ vC gi . (3)
jects the multiple solutes by sieving action and the solute–solute dt
interactions are neglected. (2) Since the solute–solute interac-
By introducing the back transport coefficient for solute i (Kbi )
tions are neglected, the concentration of each solute can be ob-
into Eq. (3), Eq. (3) can be written as (Bhattacharjee and Datta,
tained by mass balance analysis with the consideration of back
2003)
transport effect. (3) In the gel layer, each solute will have its in-
dependent value of diffusion, mass transfer and back transport dMbti
= Kbi vC gi . (4)
coefficient. dt
In the multiple solutes system of continuous cross-flow ul-
Therefore, integration of Eq. (4) with the initial condition of
trafiltration with n solutes (i = 1, 2, 3, . . . , n) in a solvent, the
t = 0 and Mbti = 0 gives
boundary layer is first formed over the membrane within a very
short period of time (within 1 s). Subsequently, the buildup of Mbti = Kbi vC gi t for i = 1, 2, 3, . . . , n. (5)
some extra resistance caused by the deposition of solutes over
the membrane surface results in the flux decline. The deposited Substituting Eq. (5) into Eq. (2) and rearranging in terms of total
layer over the membrane is termed as “gel layer” and the con- volume of permeate (Vp ) up to time t, the equation becomes
centration of each solute in the gel layer is termed as “gel n n n
concentration”.



Vp (Cbi − Cpi ) = Vg (Cgi − Cbi ) + vt Kbi Cgi . (6)
Consider the overall mass balance over the multiple solutes
i=1 i=1 i=1
system of continuous cross-flow ultrafiltration as shown in
Fig. 1. During the filtration process, some of the solutes in Rearranging Eq. (6) in terms of gel layer thickness, z, with Vg =
the bulk are transported into the permeate stream through the Az as substitution, where A is the effective area for filtration,
boundary layer and gel layer. However, for the simplicity of the equation gives
the model, the mass balance analysis in the boundary layer is  n
Vp ni=1 (Cbi − Cpi ) Kbi Cgi
ignored by assuming that the total mass of solute obtained in z= n − n i=1 vt. (7)
the boundary layer is negligible compared to the total mass of A i=1 (Cgi − Cbi ) A i=1 (Cgi − Cbi )
solute obtained in the gel layer. This assumption is applica- The total volume flux of permeate, Jv , can be calculated by
ble for the system operated under convection current created using the osmotic pressure model (Karode, 2001):
by cross-flow velocity or stirring action. After filtration time, t,
the total solutes rejected by the membrane result in the solutes 1 dVp
P −

Jv = = , (8)
concentration rise in the gel layer and the remainder is being A dt (RH )
transported back into the bulk by convection current created by
the superficial velocity. where
P is the transmembrane pressure,
 is the osmotic
The mass of the total solutes in the gel layer at any filtration pressure difference across the membrane,  is the viscosity of
time, t, is the bulk, RH is the hydraulic resistance, of the system. The hy-
draulic resistance, RH , is the sum of the membrane resistances,
Vg Cg = (Vp + Vg )Cb − Vp Cp − Mbt . (1) Rm , and the gel layer resistance, Rg (Kumar et al., 2004). Thus,
Eq. (8) can be written as
The term Vg is the total volume of gel layer up to time t, Vp is
the total volume of permeate up to time t, Mbt is the total back 1 dVp
P −

Jv = = . (9)
transport mass up to time t, Cg is the total concentration of A dt (Rm + Rg )
5060 A.L. Ahmad et al. / Chemical Engineering Science 61 (2006) 5057 – 5069

The total volume flux of permeate can also be calculated by Thus, the gel layer thickness will remain constant and the rate
Darcy’s law (Bhattacharjee and Datta, 2003): of solutes deposition over the membrane can be written as
 
1 dVp
Pc dz
Jv = = Pm , (10) = 0. (20)
A dt H dt

where
Pc is the pressure drop across the gel layer, Pm is the Substitution of Eq. (20) into Eq. (19) and rearrangement in
permeability coefficient and H is the effective filtration thick- terms of the concentration of solute i in the permeate, Cpi , gives
ness. Thus, the resistance-in-series model can be obtained by Kbi vC gi
combining Eqs. (9) and (10): Cpi = Cbi − for i = 1, 2, 3, . . . , n, (21)
AJ v,ss

P −
 where the symbol Jv,ss is the volume flux of permeate at the
Jv = . (11)
(Rm + z/Pm ) steady-state condition. Eq. (21) is the equation used to esti-
Substitution of z from Eq. (7) into Eq. (11) and rearrangement mate the concentration of solute i in the permeate, Cpi , of
in terms of 1/Jv gives the multiple solutes system of continuous cross-flow ultrafil-
tration. However, in order to calculate the concentration of
n
1 Rm  (Cbi − Cpi ) solute i in the permeate, the concentration of solute i in the
= + i=1 Vp gel layer, Cgi , must be known. The concentration of solute i
Jv
P −
 AP m (
P −
) ni=1 (Cgi − Cbi )
n in the gel layer can be estimated by using the gel polarization
v Kbi Cgi
− n i=1 t, (12) model (Bhattacharjee and Datta, 2003):
AP m (
P −
) i=1 (Cgi − Cbi )  
(Cgi − Cpi ) Jv,ss
or = exp , (22)
(Cbi − Cpi ) ki
1
= a1 + a2 Vp − a3 t, (13) where the term ki is the mass transport coefficient of solute i
Jv
in the system. The mass transport coefficient, ki , describes the
where solutes transport in the gel layer based on concentration gra-
Rm dient (diffusion coefficient) and flow velocity (hydrodynamic
a1 = , (14) properties). Therefore, mass transport coefficient, ki , exhibits a
(
P −
)
close relationship with the back transfer coefficient, Kbi . When
n
 (Cbi − Cpi ) the mass transport coefficient, ki , increases, the back transfer
a2 = i=1 , (15) coefficient will also increase, Kbi . However, mass transport co-
AP m (
P −
) ni=1 (Cgi − Cbi )
n efficient, ki , describes the solutes transport which are bounded
v Kbi Cgi to the gel layer only.
a3 = n i=1 Eqs. (13)–(16) and (21)–(22) define the multiple solutes sys-
AP m (
P −
) i=1 (Cgi − Cbi )
for i = 1, 2, 3, . . . , n. (16) tem for continuous cross-flow ultrafiltration. Once the operat-
ing parameters (
P , v, Cb , Cbi ) and the membrane parameters
Eqs. (13)–(16) represent the volume flux of permeate in the (Rm ,
, Pm , , Kbi , ki ) are known, the volume flux of perme-
multiple solutes system of the continuous cross-flow ultrafiltra- ate, concentration of the each solutes in the permeate and the
tion with n solutes (i = 1, 2, 3, . . . , n) in a solvent. gel layer resistance can be predicted.
In order to estimate the concentrations of each solute in the In the application of membrane technology in the wastewater
permeate of the multiple solutes system of the continuous cross- treatment, the COD is often used to monitor the efficiency
flow ultrafiltration, the mass of solute i where i = 1, 2, 3, . . . , n of the treatment method. The COD is closely related to the
in the gel layer at filtration time, t, can be written as concentrations of solute in the system. A linear correlation is
used in the present studies to relate the concentrations of all
Vg Cgi = (Vp + Vg )Cbi − Vp Cpi − Mbti . (17) solutes in the system with the COD as follows (Miorin, 1977):
Substituting Eq. (5) into Eq. (17) and rearranging in terms of n


total permeate volume up to time t gives CCOD = bi Ci for i = 1, 2, 3, . . . , n, (23)
i=1
Vp (Cbi − Cpi ) = Az(Cgi − Cbi ) + Kbi vC gi t. (18)
where the term CCOD is the total COD concentration in the
Differentiation of Eq. (18) with respect to time t gives system, Ci is the concentration of solute i in the system and bi
is the dimensionless coefficient.
dz Kbi vC gi
(Cbi − Cpi )Jv = (Cgi − Cbi ) + . (19)
dt A
2.1. Algorithm for the solution of the model
During steady-state filtration, the rate of solutes deposition over
the membrane is equivalent to the rate of solutes being trans- Eqs. (13)–(16) and (21)–(22) are solved using the soft-
ported back to the membrane (Bhattacharjee and Datta, 2003). ware package of Matlab 7.0 following the algorithm as
 A.L. Ahmad et al. / Chemical Engineering Science 61 (2006) 5057 – 5069 5061

START
lected up to time t can be obtained by integrating Eq. (13) in
the form of first order ordinary differential equation (ODE)
as below:
Input parameters of feed concentration of each
solute, operating and membrane parameters. dVp A
= . (24)
dt a1 + a2 Vp − a3 t

Initial guess of the concentrations of each The Vp can be obtained by numerical integration of Eq. (24)
solute in the permeate using the higher-order Runge–Kutta method at equipaced
time interval (Chapra and Canale, 1998).
4. The volume flux of permeate at steady state is obtained when
Initial guess of volume flux of permeate in
the steady-state
the calculated value of volume flux of permeate remained
constant over a period of time.
5. The concentrations of each solute in the permeate is calcu-
Calculate the concentrations of each solute in the gel lated from Eq. (21) when the comparison between the calcu-
layer and its total concentration lated value and the initial guess of volume flux of permeate
at steady state show an error less than 5%.
6. The calculation is terminated when the error between the
Calculate the volume flux of the permeate and its
initial guess of the concentrations of each solute in permeate
steady-state volume flux value
and the calculated value is less than 5%.
7. The total COD concentration in the permeate is calculated
Comparison between Jv,ss guess and Jv,ss calculated from Eq. (23).

2.2. Parameter estimation method

No
≤5%?
Consider the multiple solutes system of continuous cross-
Yes
flow ultrafiltration with n solutes (i=1, 2, 3, . . . , n) in a solvent.
Calculate the concentrations of each solute in the permeate The number of parameters needed to be estimated is determined
as follows:

Comparison between Cpi guess and Cpi calculated N = 4 + 2n, (25)

where N is the number of parameters to be determined in the

No system.
≤5%?
The membrane resistance (Rm ) can be experimentally deter-
Yes mined from the pure water run data. The slope in the linear plot
Calculation of the COD concentration in the permeate. of pure water flux (Jv ) versus increasing transmembrane pres-
sure (
P ) indicates the value of membrane resistance (Rm ).
The mass transfer coefficients (ki ) of each solute can be
END
estimated using the velocity variation method (Ahmad et al.,
2005a). This is done by linearization of gel polarization model
Fig. 2. Flow diagram of the algorithm for solution of the model.
of Eq. (21) which yields
     
1 1 Jv,ss 1
ln −1 = + ln − 1 . (26)
shown in Fig. 2: Roi q v Ri
The term Roi is the observed rejection of solute i and is defined
1. The feed concentration of each solute and the parameters,
as
operating pressure, feed flowrate are generally fixed in the
experiment. The membrane parameters, which are Rm ,
, Cpi
Roi = 1 − . (27)
Pm , , Kbi , ki , are obtained from the experimental data Cbi
using the parameter estimation method.
The term Ri is the true rejection of solute i and is defined as
2. Since the procedure is iterative, the initial estimation of
the concentrations of each solute in the permeate and the Cpi
volume flux of permeate at steady state are needed. The Ri = 1 − . (28)
Cgi
concentrations of each solute in the gel layer are obtained
from the gel polarization model (Eq. (22)). The coefficients q and  are obtained from the best linear
3. The volume flux of permeate is estimated by solving Eqs. fit in the plot of ln(1/Roi − 1) versus Jv,ss /v  and the mass
(13)–(16) simultaneously. The total permeate volume col- transfer coefficients can be expressed in term of superficial
5062 A.L. Ahmad et al. / Chemical Engineering Science 61 (2006) 5057 – 5069

velocity, v:
ki = qv  .
The back transport coefficients for each solute, Kbi , can be
determined from the linear plot of Cpi versus vC gi /AJ v,ss .
From Eq. (21), the value of Kbi can be obtained from the slope
of plot. The experimental data can be obtained by measuring the
change in the concentrations of each solute in the permeate at
steady state with increasing transmembrane pressure at constant
feed flowrate.
For the determination of unknown parameters of a1 a2 , a3 in
Eq. (13), several experimental data with Vp and t as independent
variables and 1/Jv as dependent variable are collected. These
data are fitted into Eq. (13) and the parameters are estimated
using the Levenberg–Marquardt method with Gauss–Newton
algorithm. Once the a1 , a2 , a3 values are known, the unknown
parameters
, Pm can be determined from Eqs. (14)–(16), re-
spectively, when the viscosity () of the bulk is obtained exper-
imentally. These parameters are the characteristic of the mem-
brane separation system under a constant temperature, pressure
and with the same solutes membrane combination.
3. Experimental
3.1. Sample preparation
POME was obtained from the United Palm Oil Mill, Penang,
Malaysia. The POME is the liquid waste generated by the palm
oil mill. The pretreatment of POME is required to remove the
oil and grease and total solids before the POME is fed into
the ultrafiltration system (Ahmad et al., 2003, 2005b). Hundred
(29) liters of POME was transferred to a chemical treatment tank for
the coagulation and flocculation process. Three liters of coag-
ulant, modified industrial grade alum purchased from Envilab
Sdn. Bhd. was added into the POME at the stirring speed of
50 rpm. The pH value was adjusted to the pH 6 by using in-
dustrial grade sodium hydroxide 65% (NaOH). Three liters of
flocculant (0.1%), cationic polymer (FO 4190) was added into
the POME at the stirring speed of 10 rpm for 15 min. The mix-
ture was allowed to settle and the sludge was removed using a
filter press. The supernatant is the pretreated POME for ultrafil-
tration. Table 1 presents the characteristic of raw and pretreated
POME. The details of the pretreatment of POME are presented
elsewhere (Ahmad et al., 2003). Table 1 shows that the oil and
grease as well as the suspended solids were removed (with 70%
removal of COD) by the coagulation and flocculation process.
The remaining COD (30%) was removed by the ultrafiltration
and reverse osmosis processes. The ultrafiltration in this case
serves as the pretreatment for reverse osmosis. From the chem-
ical composition analysis of the pretreated POME, the major
components (solutes) contribute to the COD are carbohydrate
constituents, crude protein and ammoniacal nitrogen (Hwang
et al., 1978; Ho et al., 1984). Therefore, the pretreated POME
forms a ternary solutes system for ultrafiltration. The pretreated
POME samples at various concentrations were prepared by di-
luting the sample with distilled water. The details of the exper-
imental run are listed in Table 2.
3.2. Membrane system

Table 1 Fig. 3 shows a schematic diagram of the pilot plant scale

Characteristic of raw POME and the treated POME
equipment setup used in the present study. The ultrafiltra-
Parameter Raw POME Pretreated POME tion membrane was the ceramic membrane module P16-60
pH 4.7 6.0 (PCI-Memtech) with stainless steel housing. The membrane
Oil and grease, mg/L 4000 Not detectable module had a nominal molecular weight cut-off (MWCO) of
Chemical oxygen demand 50,000 15,000 2500–10,000 g/mol with 19 flow channels. The inner diameter
(COD), mg/L of each tube was 6.0 mm and its length was 1020 mm. The
Suspended solids, mg/L 40,500 Not detectable
effective area for filtration was 0.36 m2 and could be operated
Ammoniacal nitrogen, mg/L 35 34
up to maximum pressure of 6 bar.

Table 2
Details of the experiment

Run
P (bar) QF (m3 /h) Ctotal (mg/L) Ccarbohydrate (mg/L) Cprotein (mg/L) Cammonia (mg/L) COD (mg/L) Ro

1 1 4 5720 3015 2675 34 15,000 0.17

2 2 4 5720 3015 2675 34 15,000 0.21
3 3 4 5720 3015 2675 34 15,000 0.25
4 4 4 5720 3015 2675 34 15,000 0.26
5 4 3 5720 3015 2675 34 15,000 0.11
6 4 5 5720 3015 2675 34 15,000 0.41
7 4 7 5720 3015 2675 34 15,000 0.56
8 4 4 1431 754 669 8 3750 0.08
9 4 4 2862 1508 1337 17 7500 0.13
10 4 4 7155 3770 3344 42 18,750 0.33

P = transmembrane pressure; QF = feed flowrate; Ctotal = total bulk concentration; Ccarbohydrate = bulk concentration for carbohydrate constituents;
Cprotein = bulk concentration for crude protein; Cammonia = bulk concentration for ammoniacal nitrogen; COD = COD in the feed stream; Ro = overall observed
rejection.
 A.L. Ahmad et al. / Chemical Engineering Science 61 (2006) 5057 – 5069
Pretreated POME Permeate recycle
Retentate recycle
Feed Heat exchanger
Tank TI
Pressure control
valve
Pressure control PI FI PI FI
valve
Valve Feed Pump Ultrafiltration
Fig. 3. Schematic diagram of the pilot plant scale equipment set up for
ultrafiltration studies. The symbol PI is the pressure indicator, FI is the
flowmeter and TI is the temperature indicator.
The pretreated POME was placed in the feed tank of the
membrane system and the temperature was maintained at
room temperature (25 ◦ C) using a water cool heat exchanger.
The feed was pumped through the membrane module using a
centrifugal pump (Grundfos CRN-8-50). The required cross-
flowrate and transmembrane pressure were adjusted using
pressure control valves (Gemu-diaphragm type). The inlet and
outlet pressures of the module were measured using the pres-
sure gauges. The turbine function flowmeters (Burkert) were
used to measure cross-flowrates of the feed and permeate. The
permeate and the retentate streams were recycled back to the
feed tank (total recycle mode) to maintain a constant feed
concentration.
3.3. Analysis
The viscosity of pretreated POME was measured with
a Hoeppler falling ball viscometer. The carbohydrate con-
stituents were determined using the colorimetric method with
the phenol–sulfuric acid 98% reaction (Dubois et al., 1956).
The crude protein was measured by using the colorimetric
method with a detergent-compatible formulation based on
bicinchoninic acid (BCA Protein Assay, Pierce). The ammo-
niacal nitrogen was determined using preliminary distillation
step followed by titrimetric method with standard sulfuric acid
titrant, 0.02 N, and the COD was measured by using the colori-
metric method at wavelength 600 nm with spectrophotometer
CECIL 1000 series, Cambridge, UK (APHA, 1999).
4. Results and discussion
In the present study, the experimental data were obtained
from the pilot plant scale membrane system with the pretreated
POME as the feed into the ultrafiltration module. The pre-
treated POME was composed of mainly a ternary solutes sys-
tem (carbohydrate constituents, crude protein and ammoniacal
nitrogen).
5063
Table 3
Estimated parameters for the model
Parameter Value
 (103 Pa s) 1.0240
Rm (10−12 m−1 ) 2.1576
Pm (1016 m2 ) 8.6464
Carbohydrate Protein Ammoniacal
nitrogen
Kbi (108 m2 ) 19.2414 4.3795 8.3812
ki (106 m/s) 4.9940 2.9375 3.5777
4.1. System parameters
By using the experimental data, the parameters of , Rm ,
Pm , Kbi , ki were estimated and are listed in Table 3. The val-
ues of the estimated parameter shown in Table 3 matched with
their typical values (Bhattacharjee and Datta, 2003; Karode,
2001; Mehta and Zydney, 2005). The viscosity () of pretreated
POME as 1.0240 × 10−3 Pa s was slightly higher than the vis-
cosity of water (8.7790 × 10−4 Pa s) due to the presence of
organic matter (carbohydrate constituents, crude protein and
ammoniacal nitrogen) in the pretreated POME. The calculated
value of membrane resistance (Rm ) was in good agreement with
the typical value reported in the literature for ceramic mem-
brane (Mehta and Zydney, 2005). The permeability coefficient
(Pm ) of the system in the present study depends on the type
of solutes and the type of membrane used. However, the or-
der of magnitude of Pm value (10−16 ) was in good correspon-
dence with the value reported in the literature (Bhattacharjee
and Datta, 2003).
The typical value for back transport coefficient (Kbi ) of ev-
ery solute estimated in the present study was 10−8 m2 but the
typical value reported in the literature (Bhattacharjee and Datta,
2003) using continuous stirred ultrafiltration was 10−10 m2 . The
deviation of the value shows that the turbulence created by con-
tinuous cross-flow ultrafiltration resulted in higher back trans-
port of the solutes compared to the turbulence created by stirred
ultrafiltration. The estimated value for back transport coefficient
of carbohydrate constituents (Kb,carbohydrate ) was higher com-
pared with other solutes because the carbohydrate constituents
had the largest molecular size compared to other solutes.
The mass transfer coefficient, ki (2 × 10−6 –5 × 10−6 m/s),
value has the similar order of magnitude with the volume flux
of permeate (5 × 10−6 –35 × 10−6 m/s). Therefore, with the
calculation of transport parameters, the gel polarization effect
taken into consideration in the present study was valid. The
estimated value of ki for all the solutes in the system was in
good agreement with the typical value reported in the literature
(Bhattacharjee and Datta, 2003; Karode, 2001).
4.2. Flux and fouling behavior of the membrane
Fig. 4 considers the effect of various operating conditions
(transmembrane pressure, feed flowrate and bulk concentration)
5064 A.L. Ahmad et al. / Chemical Engineering Science 61 (2006) 5057 – 5069

35 Experimental
Table 4
30 1 bar
Statistical analysis of Fig. 4
Jv x 106 (m3/m2s)

2 bar
25 Fig. 4(a) Fig. 4(b) Fig. 4(c)
3 bar
20 4 bar Transmembrane r2 Feed r2 Bulk r2
Simulated
15 pressure (bar) flowrate concentration
1 bar
(m3 /h) (mg/L)
10 2 bar

5 3 bar 1 0.9706 3 0.9844 1431 0.8118

4 bar 2 0.9923 4 0.9822 2862 0.9482
0
0 1000 2000 3000 4000 3 0.9715 5 0.9884 5724 0.9822
4 0.9822 7 0.9503 7155 0.9887
(a) Time (s)

35 Experimental
30 3 m3/hr
brane pressure until a critical value of transmembrane pressure
Jv x 106 (m3/m2s)

4 m3/hr
25
5 m3/hr (3 bar). At higher value of transmembrane pressure, the steady-
20 7 m3/hr
state volume flux of permeate did not increase further and the
Simulated
15
3 m3/hr
membrane system operated at its critical flux value.
10 4 m3/hr Fig. 4(b) shows the volume flux of permeate versus filtra-
5 5 m3/hr tion time curve with the increasing feed flowrate varying from
7 m3/hr 3 to 7 m3 /h. With the same bulk concentration, the transmem-
0
0 1000 2000 3000 4000 brane pressure was kept constant at 4 bar. The coefficients of
(b) Time (s) determination (r 2 ) of 0.9844, 0.9822, 0.9884 and 0.9503 for
the feed flowrate of 3, 4, 5 and 7 m3 /h, respectively, as shown
45 Experimental in Table 4 indicate that the simulated results are able to rep-
40 1431 mg/L resent the experimental data. The results depicted in Fig. 4(b)
35
Jv x 106 (m3/m2s)

30
25
20
15
10
5
0 feed flowrate.
0 1000 2000 3000 4000
(c) Time (s)
Fig. 4. Volume flux of permeate against filtration time at different (a) trans-
membrane pressure, (b) feed flowrate and (c) bulk concentration.
on the volume flux of permeate. Fig. 4(a) shows the volume
flux of permeate versus filtration time curve with the increas-
ing transmembrane pressure varying from 1 to 4 bar. The feed
flowrate and velocity were kept constant at 4 m3 /h and 2 m/s,
respectively. The concentration of each solute in the bulk was
kept at 3015 mg/L for carbohydrate constituents, 2675 mg/L for
crude protein and 34 mg/L for ammoniacal nitrogen, respec-
tively. The combination of these concentrations contributed to
the total bulk concentration of 5720 mg/L and the COD value
of 15,000 mg/L.
The simulation results show a good agreement with the
experimental data with the statistical analysis as shown in
Table 4. Table 4 shows that the coefficients of determination
(r 2 ) are more than 0.97 for the transmembrane pressure of
1–4 bar. Fig. 4(a) indicates that the fouling of the membrane
caused the volume flux of permeate to decrease until the sys-
tem reached a steady-state value. At the same time, the volume
flux of permeate increased with the increasing transmem-
2862 mg/L
show that initial and steady-state permeate fluxes were greater
5720 mg/L
7155 mg/L
at higher feed flowrate. Increasing feed flowrate (or cross-flow
Simulated velocity) results in slower accumulation of deposited solutes
1431 mg/L on the membrane surface due to higher back transport of the
2862 mg/L rejected solutes into the bulk. Consequently, thinner gel layer
5720 mg/L
7155 mg/L
was formed and higher permeate flux was obtained at higher
Fig. 4(c) shows the volume flux of permeate versus filtra-
tion time curve with the increasing bulk concentration varying
from 1431 to 7155 mg/L. In these experimental runs, the op-
erating conditions were identical; the transmembrane pressure
was 4 bar and the feed flowrate was maintained at 4 m3 /h. The
simulation results indicate a good agreement with the experi-
mental data by having the coefficients of determination (r 2 ) of
more than 0.81 for the increasing bulk concentration as shown
in Table 4. The results observed in these experimental runs in-
dicate that lower fluxes of permeate were experienced at higher
bulk concentration due to greater solutes deposition on the
membrane surface.
The comparison between the simulation results and experi-
mental data of the total volume of permeate collected against
time for the various operating conditions (transmembrane pres-
sure, feed flowrate and bulk concentration) is shown in Fig. 5.
The statistical analysis of these experimental runs is tabulated
in Table 5. The figure shows a very good fit with the coeffi-
cient of determination (r 2 ) more than 0.91 for all the operating
conditions. The curves for the total volume of permeate col-
lected at any filtration time demonstrate a similar trend with
the volume flux of permeate as shown in Fig. 4.
Fig. 6 shows the gel layer resistance as a function of time
for various operating conditions: transmembrane pressure, feed
flowrate and bulk concentration. The gel layer resistance is
 A.L. Ahmad et al. / Chemical Engineering Science 61 (2006) 5057 – 5069 5065

10 Experimental Experimental
50
1 bar 1 bar
8

Rg x 10-11 (m-1)
40 2 bar
2 bar
Vp x 103 (m3)

3 bar
3 bar
6 4 bar
30 4 bar
Simulated Simulated
4 20 1 bar
1 bar
2 bar
2 bar 10
2 3 bar
3 bar
0 4 bar
4 bar
0 0 1000 2000 3000 4000
0 1000 2000 3000 4000
(a) Time (s)
(a) Time (s)
Experimental
Experimental 50
14 3 m3/hr
3 m3/hr
4 m3/hr

Rg x 10-11 (m-1)
12 4 m3/hr 40
5 m3/hr
Vp x 103 (m3)

10 5 m3/hr
30 7 m3/hr
8 7 m3/hr
Simulated Simulated
6 20 3 m3/hr
3 m3/hr
4 m3/hr
4 4 m3/hr 10
5 m3/hr
2 5 m3/hr
7 m3/hr
7 m3/hr 0
0 0 1000 2000 3000 4000
0 1000 2000 3000 4000 (b) Time (s)
(b) Time (s)
Experimental
Experimental 50
14 1431 mg/L
1431 mg/L
12 40 2862 mg/L
Rg x 10-11 (m-1)

2862 mg/L
5720 mg/L
Vp x 103 (m3)

10 5720 mg/L
30 7155 mg/L
8 7155 mg/L
Simulated
Simulated
6 20 1431 mg/L
1431 mg/L
2862 mg/L
4 2862 mg/L 10
5720 mg/L
2 5720 mg/L
7155 mg/L
7155 mg/L 0
0 0 1000 2000 3000 4000
0 1000 2000 3000 4000
(c) Time (s)
(c) Time (s)
Fig. 6. Gel layer resistance against the filtration time at different (a) trans-
Fig. 5. Total permeate volume against filtration time at different (a) trans- membrane pressure, (b) feed flowrate and (c) bulk concentration.
membrane pressure, (b) feed flowrate and (c) bulk concentration.

Table 5 lation, respectively. The gel layer resistance increased sharply

Statistical analysis of Fig. 5 at the initial filtration time. After a sharp increase during the
initial filtration time, this increase became sluggish and grad-
Fig. 5(a) Fig. 5(b) Fig. 5(c)
ually approached asymptotic value. At initial filtration time,
Transmembrane r2 Feed r2 Bulk r2 more solutes were deposited on the feed surface of the mem-
pressure (bar) flowrate concentration brane than those transported back to the bulk.
(m3 /h) (mg/L)
However, after the initial filtration period, the rate of solutes
1 0.9138 3 0.9809 1431 0.9937 deposition on the feed surface of the membrane was balanced
2 0.9987 4 0.9959 2862 0.9934 by the rate of solutes transported back to the bulk. Fig. 6(a)
3 0.9982 5 0.9852 5724 0.9959 demonstrates that the resistance due to the deposition of
4 0.9959 7 0.9335 7155 0.9835
solutes increased with the transmembrane pressure. The feed
flowrate was kept constant at 4 m3 /h with the total bulk concen-
tration of 5720 mg/L (carbohydrate constituents: 3015 mg/L,
calculated from crude protein: 2675 mg/L and ammoniacal nitrogen: 34 mg/L).

P −
 High transmembrane pressure resulted in the higher volume
Rg = − Rm . (30) flux of permeate due to the increase in driving force. This
Jv
would result in the more accumulation of retained solutes and
To obtain the experimental and simulation value of Rg , the ex- hence the gel layer resistance increased. The developed model
perimental and simulation values of Jv were used in the calcu- was successful in predicting the gel layer resistance, Rg , as a
5066 A.L. Ahmad et al. / Chemical Engineering Science 61 (2006) 5057 – 5069

Table 6 0.5
Statistical analysis of Fig. 6
Experimental
Fig. 6(a) Fig. 6(b) Fig. 6(c) 0.4 carbohydrate
nitrogen
Transmembrane r2 Feed r2 Bulk r2 0.3
protein
pressure (bar) flowrate concentration

Ro
(m3 /h) (mg/L) Simulated
0.2
carbohydrate
1 0.9168 3 0.9995 1431 0.9988 nitrogen
2 0.9612 4 0.9190 2862 0.7533 0.1
protein
3 0.9337 5 0.9742 5724 0.9190
4 0.9190 7 0.8032 7155 0.9458 0.0
0 1 2 3 4 5
(a) Trans Membrane Pressure (bar)

1.0
function of time at different transmembrane pressures. Based
on the statistical analysis from Table 6, the coefficients of deter- 0.8
mination (r 2 ) were more than 0.91 for transmembrane pressure
of 1–4 bar. 0.6

Ro
Fig. 6(b) depicts the variation of gel layer resistance with
increasing feed flowrate under the same bulk concentration. 0.4
The transmembrane pressure was maintained at 4 bar. With the
0.2
increase in feed flowrate (or cross-flow velocity), the gel layer
resistance decreases because more solutes were removed from
0.0
the deposited layer to bulk by the action of back transport 2 3 4 5 6 7 8
mechanism. Simulation results show a good agreement with (b) Feed Flow rate (m3/hr)
the experimental data by having coefficients of determination
0.6
of 0.9995, 0.9190, 0.9742 and 0.8032 for the feed flowrate of
3, 4, 5 and 7 m3 /h as tabulated in Table 6. 0.5
Fig. 6(c) shows the effect of increasing bulk concentration on
gel layer resistance at constant transmembrane pressure (4 bar) 0.4
and feed flowrate (4 m3 /h). As the bulk concentration increased,
Ro

0.3
the gel layer resistance increased due to the higher deposition
of rejected solute on the membrane surface. This phenomenon 0.2
lead to higher thickness of the gel layer and thus adding more
resistance to solvent flow (lower flux of permeate) as shown in 0.1
Fig. 4(c). 0.0
0 2 4 6 8
4.3. Permeate quality for the ternary solutes system (c) Feed Concentration (kg/m3)

Fig. 7 shows the rejection of each individual solute (carbo-
hydrate constituents, crude protein and ammoniacal nitrogen)
against the increasing transmembrane pressure, feed flowrate
and bulk concentration. Fig 7(a) demonstrates the effect of
increasing transmembrane pressure on the rejection of each
individual solute at constant feed flowrate (4 m3 /h) and total
bulk concentration of 5720 mg/L (carbohydrate constituents:
3015 mg/L, crude protein: 2675 mg/L and ammoniacal nitro-
gen: 34 mg/L). Fig. 7(a) shows the simulation results by using
Eq. (21) and gave a good agreement with the experimental data
2
(rcarbohydrate = 0.8402, rprotein
2 = 0.9750, rnitrogen
2 = 0.8633). The
statistical analysis was given in Table 7. Both simulation results
and experimental data demonstrated an increase in observed re-
jection when the transmembrane pressure increased. At higher
transmembrane pressure, the higher driving force would force
more solvent (water) of lower molecular weight to pass through
the pores of the membrane. From Fig. 7(a), the carbohydrate
constituents gave the highest observed rejection and followed
Fig. 7. The steady state observed rejection for individual solute against (a)
transmembrane pressure, (b) feed flowrate and (c) bulk concentration.
by the ammoniacal nitrogen and the crude protein. Fig. 7(a)
also indicated that the crude protein passed through the mem-
brane at the highest rate followed by ammoniacal nitrogen and
carbohydrate constituents.
Fig. 7(b) shows the rejection curves of each individual solute
against increasing feed flowrate at constant bulk concentration
and transmembrane pressure. Comparisons made between the
simulation results and the experimental data based on statisti-
cal analysis (Table 7) show that the model is in good agreement
with the experimental data. The results from Fig. 7(b) indi-
cated that the rejection of each solute increased when the feed
flowrate increased. Increasing feed flowrate (or cross-flow ve-
locity) leads to lower gel layer resistance (Fig. 6(b)) and thus
higher rejections were observed.
 A.L. Ahmad et al. / Chemical Engineering Science 61 (2006) 5057 – 5069 5067

Table 7 20000
Statistical analysis of Fig. 7
Experimental
Fig. 7(a) Fig. 7(b) Fig. 7(c) Simulated
15000
Solute r2 Solute r2 Solute r2

COD (mg/L)
Carbohydrate 0.8402 Carbohydrate 0.7903 Carbohydrate 0.9887
Protein 0.9750 Protein 0.8620 Protein 0.9096 10000
Nitrogen 0.8633 Nitrogen 0.9180 Nitrogen 0.9876

5000
20000
COD calculated (mg/L)

15000 0
0 1 2 3 4 5
(a) Trans Membrane Pressure (bar)
10000
20000
5000

0 15000
0 5000 10000 15000 20000

COD (mg/L)
COD experiment (mg/L)
10000
Fig. 8. Comparison of experimental COD against calculated COD.

Fig. 7(c) shows the effect of increasing feed concentration 5000

on rejection of each individual solute at constant feed flowrate
(4 m3 /h) and transmembrane pressure (4 bar). The statistical
2
analysis as listed in Table 7 (rcarbohydrate = 0.9887, rprotein
2 = 0
0 2 4 6 8 10
0.9096, rnitrogen = 0.9876) shows that the simulation results
2
(b) Feed Flow rate (m3/hr)
were in good agreement with the experimental data. For all the
solutes, the increase of rejection with increasing bulk concen- 20000
tration exhibits that the gel layer resistance formed on the mem-
brane surface was actually assisted in the solutes rejections.
The correlation in the present study related the concen- 15000
trations of all solutes (Cp,carbohydrate , Cp,protein , Cp,nitrogen )
COD (mg/L)

with the COD value in the membrane system. The correla-

tion was obtained by using the multivariable regression of
Levenberg–Marquardt method. The values of the dimension-
less coefficients b1 , b2 and b3 were 0.3535, 4.9279 and 0.5595,
respectively, obtained using Eq. (23). The final correlation is
given by
CODcalculated = 0.3535Cp,carbohydrate + 4.9279Cp,protein
+ 0.5595Cp,nitrogen . (31)
The comparison of the experimental data and the model pre-
dictions is shown in Fig. 8. It shows that the correlation was
successfully applied in the prediction of COD concentration in
the membrane system with the coefficient of determination (r 2 )
of 0.8554.
The simulation values obtained from Eq. (31) showed a good
fit with the experimental data with the coefficient of determina-
tion (r 2 ) as 0.9999, 0.7368 and 0.9467 for Figs. 9(a)–(c), respec-
tively. Figs. 9(a) and (b) show the COD of the permeate against
increasing transmembrane pressure and feed flowrate, respec-
tively, at constant bulk concentration. Both figures demonstrate
a decrease of COD reading in the permeate stream when the
10000
5000
0
0 2 4 6 8 10
(c) Feed Concentration (kg/m3)
Fig. 9. Variation of steady-state COD reading in the permeate with (a)
transmembrane pressure, (b) feed flowrate and (c) bulk concentration.
transmembrane pressure and feed flowrate increased, respec-
tively. This is in line with the observed rejection shown in
Figs. 7(a) and (b). The increase of observed rejection at the
increasing transmembrane pressure and feed flowrate resulted
more solvent to pass through the pores of membrane com-
pared to the solutes and thus lower COD reading was obtained.
Fig. 9(c) demonstrates an increase in the COD of the permeate
5068 A.L. Ahmad et al. / Chemical Engineering Science 61 (2006) 5057 – 5069

stream when the bulk concentration increased at constant trans- Jv volume flux of permeate, m3 /m2 s
membrane pressure (4 bar) and feed flowrate (4 m3 /h). This Jv,ss volume flux of permeate at the
was due to the higher COD in the feed stream when the bulk steady-state condition, m3 /m2 s
concentration increased. k mass transport coefficient, m/s
Fig. 9 shows that the efficiency of COD removal using ultra- Kb total back transport coefficient, m2
filtration was low (10–20% of COD removal). The low rejection Mbt total back transport mass up to time
of carbohydrate constituents (30–60%), crude protein (5–30%) t, kg
and ammoniacal nitrogen (10–40%) showed in Fig. 7 indicated n total number of solutes
that most of the solutes were able to pass through the mem- N number of parameters
brane and thus contributed to low efficiency of COD removal. Pm permeability coefficient, m2
Therefore, the carbohydrate constituents, crude protein and am- q,  constant in Eq. (26)
moniacal nitrogen contained large portion of micromolecules r2 coefficient of determination
which need to be further removed by reverse osmosis. R true rejection
Rg gel layer resistance, m−1
5. Conclusions RH hydraulic resistance, m−1
Rm resistances due to the membrane,
A mathematical model suitable for the multiple solutes m−1
system in ultrafiltration based on mass balance analysis Ro observed rejection
coupled with the filtration theory (osmotic pressure model, t filtration time, s
Kozeny–Carman equation), resistance-in-series model and gel v superficial velocity, m/s
polarization model has been developed for a complex biolog- Vg total volume of gel layer up to time
ical solution of pretreated POME. The comparisons between t, m3
the proposed model and the experimental data of ternary so- Vp total volume of permeate up to time
lutes system (pretreated POME) for the continuous cross-flow t, m3
ultrafiltration showed a good agreement and proved its utility z gel layer thickness, m
in predicting the performance of the multiple solutes system,
Greek letters
even solute–solute interactions were ignored in the model.
The proposed model successfully predicted the volume flux
P transmembrane pressure, bar
of permeate and the gel layer resistance as a function of filtration
Pc pressure drop across the gel layer,
time. In addition, the proposed model predicted the observed bar
rejections of each solute in the system. The correlation for
 osmotic pressure difference across
the concentrations of all solutes with the COD value in the the membrane, bar
permeate stream of the membrane system has been successfully  viscosity of the bulk, Pa s
developed. The model is suitable in predicting the performance
Subscripts
of single or multiple solutes ultrafiltration system.
The ultrafiltration gave low efficiency of COD removal for carbohydrate carbohydrate constituents
pretreated POME due to the low rejection of carbohydrate con- i solute
stituents, crude protein and ammoniacal nitrogen. Thus, the nitrogen ammoniacal nitrogen
reverse osmosis is needed to remove the remaining micro- protein crude protein
molecules to achieve the desired water quality where the ultra-
filtration was served as the pretreatment to the reverse osmosis Acknowledgments
process.
The authors would like to gratefully acknowledge Federal
Notation Land Development Authority Foundation (Yayasan Felda) of
Malaysia for their financial support. The authors would also like
to thank United Oil Palm Industry, Nibong Tebal, Pulau Pinang
a constant in Eq. (13), defined by Eqs. for providing the sample of POME to conduct this research.
(14)–(16)
A effective area for filtration, m2 References
b constant in Eq. (23)
Cb total concentration of solutes in the Ahmad, A.L., Ismail, S., Bhatia, S., 2003. Water recycling from palm oil mill
bulk, kg/m3 effluent (POME) using membrane technology. Desalination 157, 87–95.
Cg total concentration of solutes in the Ahmad, A.L., Chong, M.F., Bhatia, S., 2005a. Mathematical modeling and
simulation of the multiple solutes system for nanofiltration process. Journal
gel layer, kg/m3
of Membrane Science 253, 103–115.
Cp total concentration of solutes in the Ahmad, A.L., Ismail, S., Bhatia, S., 2005b. Ultrafiltration behavior in
permeate, kg/m3 the treatment of agro-industry effluent: pilot scale studies. Chemical
H effective filtration thickness, m Engineering Science 60, 5385–5394.
 A.L. Ahmad et al. / Chemical Engineering Science 61 (2006) 5057 – 5069
American Public Health Association (APHA), 1999. Standard methods for
the examination of water and wastewater. APHA, Washington, DC.
Baker, R.W., 2000. Membrane Technology and Applications. McGraw-Hill,
New York.
Bhattacharjee, C., Datta, S., 2003. Analysis of polarized layer resistance
during ultrafiltration of PEG-6000: an approach based on filtration theory.
Separation and Purification Technology 33, 115–126.
Chapra, S.C., Canale, R.P., 1998. Numerical Methods for Engineers. McGraw-
Hill, New York.
Damak, K., Ayadi, A., Zeghmati, B., Schmitz, P., 2004. A new Navier–Stokes
and Darcy’s law combined model for fluid flow in crossflow filtration
tubular membranes. Desalination 161, 67–77.
Darnon, E., Belleville, M.P., Rios, G.M., 2002. Modelling ultrafiltration of
complex biological solutions. AI.Ch.E. Journal 48, 1727–1736.
Djurić, M., Gyura, J., Zavargo, Z., Šereš, Z., Tekić, M., 2004. Modelling of
ultrafiltration of non-sucrose compounds in sugar beet processing. Journal
of Food Engineering 65, 73–82.
Dubois, M., Gilles, K.A., Hamilton, J.K., Rebers, P.A., Smith, F., 1956.
Colorimetric method for determination of sugars and related substances.
Analytical Chemistry 28, 350–356.
Gehlert, G., Adbulkadir, M., Fuhrmann, J., Hapke, J., 2005. Dynamic
modeling of an ultrafiltration module for use in a membrane bioreactor.
Journal of Membrane Science 248, 63–71.
Ghaffour, N., 2004. Modeling of fouling phenomena in cross-flow
ultrafiltration of suspensions containing suspended solids and oil droplets.
Desalination 167, 281–291.
Ho, C.C., Tan, Y.K., Wang, C.W., 1984. The distribution of chemical
constituents between the soluble and particulate fractions of palm oil mill
5069
effluent and its significant on its utilization/treatment. Agricultural Wastes
11, 61–71.
Hwang, T.K., Ong, S.M., Seow, C.C., Tan, H.K., 1978. Chemical composition
of palm oil mill effluents. Planter 54, 749–756.
Karode, S.K., 2001. Unsteady state flux response: a method to determine
the nature of the solute and gel layer in membrane filtration. Journal of
Membrane Science 188, 9–20.
Katsikaris, K., Boukouvalas, C., Magoulas, K., 2005. Simulation of
ultrafiltration process and application to pilot tests. Desalination 171, 1–
11.
Kumar, N.S.K., Yes, M.K., Cheryan, M., 2004. Ultrafiltration of soy protein
concentrate: performance and modeling of spiral and tubular polymeric
modules. Journal of Membrane Science 244, 235–242.
Mehta, A., Zydney, A.L., 2005. Permeability and selectivity analysis for
ultrafiltration membranes. Journal of Membrane Science 249, 245–249.
Miorin, A.F., 1977. Wastewater Treatment Plant Design. Water Pollution
Control Federation, USA.
Sondhi, R., Bhave, R., 2001. Pole of backpulsing in fouling minimization
in crossflow filtration with ceramic membranes. Journal of Membrane
Science 186, 41–52.
Tansel, B., Bao, W.Y., Tansel, I.N., 2000. Characterization of fouling kinetics
in ultrafiltration systems by resistance in series model. Desalination 129,
7–14.
Yeh, H.M., Dong, J.H., Shi, M.Y., 2004. Momentum balance analysis of flux
and pressure declines in membrane ultrafiltration along tubular modules.
Journal of Membrane Science 241, 335–345.