You are on page 1of 12

4428 Ind. Eng. Chem. Res.

2009, 48, 4428–4439

SEPARATIONS

Modeling and Simulation of Stirred Dead End Ultrafiltration Process Using the
Aspen Engineering Suite
Sabuj Das, Prabirkumar Saha,* and G. Pugazhenthi
Department of Chemical Engineering, Indian Institute of Technology, Guwahati, Assam 781 039, India

Simulation of pressure-driven membrane process has been carried out considering two different types of
solute (silica and dextran) and membrane (partially permeable and totally retentive) in a stirred cell using the
Aspen Engineering Suite. One solute, silica, exerts negligible osmotic pressure but the other, dextran, offers
adequate osmotic pressure. Silica is very susceptible to form a gel layer while dextran has no such tendency.
Two types of membrane are considered for this work. One membrane completely separates solute from the
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

solvent (totally retentive membrane) and the other does it partially (partially permeable membrane). The
second type of membrane is regarded as highly compact granulated particle layer which has a different range
of porosity. So the diffusivity within the membrane layer is referred to as hindered back diffusion. Osmotic
pressure model and gel polarization model have been considered for batch and continuous mode operation.
Downloaded via CERTH CPERI on October 3, 2022 at 11:23:51 (UTC).

Silica suspension has been simulated with totally retentive membrane and dextran solution is simulated with
partially retentive membrane in batch and continuous mode, respectively. The effect of applied pressure,
stirring speed, and initial feed concentration on permeate flux, membrane surface concentration, permeate
concentration, and true and observed rejection have been studied.

1. Introduction small noncharged molecules in an aqueous solution using reverse


osmosis and ultrafiltration membranes. The results revealed that
Pressure-driven membrane processes are being increasingly
in a stirred cell with high stirring velocity, the concentration
incorporated in the process industries for the separation of
near the membrane was close to the bulk concentration, resulting
valuable chemical and biological compounds from mixtures. Of
these processes, reverse osmosis (RO) and ultrafiltration (UF) in improved rejection compared to the unstirred batch cell
have become increasingly important during the past decades results. Several studies have reported the experimental results
due to their mild and energetically favorable separation and as well as modeling of limiting flux for the separation of macro
concentration technique, which have been commercialized in solute and colloids by UF that provided strong support to the
pharmaceutical, chemical, and food industries. Among the main gel polarization model.4-6
problems hampering its widespread use, concentration polariza- Karode7 simulated the unsteady-state permeate flux response
tion as well as fouling may be mentioned, because both for three types of solutes, such as (i) solutes which exert an
phenomena give rise to flux decline. The flux decrease could osmotic pressure but do not form a gel; (ii) solutes which do
also be due to an increase in osmotic pressure, formation of gel not exert an osmotic pressure but form a gel, and (iii) solutes
layer, solute adsorption on membrane, and pore plugging. In which exert an osmotic pressure and also form a gel. He showed
order to minimize the concentration polarization, much attention that the concentration at the surface of a membrane was a
has been paid to rather mechanical aspects, such as cross flow, function of operating parameters and it was possible to predict
back pulsing, and promotion of turbulence, e.g., by gas the membrane wall concentration as a function of time. In
purging.1,2 Moreover, any attempt to increase the flux by addition, a step change in transmembrane pressure showed that
increasing pressure causes severe increase in concentration a solute, which exerts osmotic pressure but does not form a gel
polarization, thus giving little increase in flux compared to layer, exhibited a gradual flux decline over time. Experimental
increase in pressure. Eventually, the flux does not increase verification of this work was presented by Zaidi and Kumar.8-10
beyond a certain point (called limiting flux) even if the pressure They experimentally studied unsteady-state flux response to a
is increased further. Types of solute and membrane have great step change in transmembrane pressure for dextran8 and
impact for gel formation and concentration polarization. polyethylene glycol (PEG) solution9 and reported that polariza-
Flux decline rate also depends on the flow direction and tion resistance depended on applied pressures and initial bulk
operating condition. In the dead end filtration, the feed flow is feed concentrations, whereas the thickness of polarization layer
directed perpendicular to the membrane, whereas in the cross- affected by only feed concentration. They have also investigated
flow filtration, feed flow is directed parallel or tangential to the the characteristics of a deposited layer formed in dead end UF
membrane. The former one is most often used for small volume of silica after reaching steady-state flux.10 In addition, the effects
laboratory applications and the later one is used for large volume of operating conditions such as applied pressure and bulk
process applications. Dead end filtration may be carried with concentration on specific resistance, thickness, porosity, and
stirred and unstirred cells. Nicolas et al.3 showed a comparison silica concentration of gel were reported. Zaidi and Kumar11
study of rejection in a stirred and an unstirred batch cell for have also experimentally investigated the influence of ethanol
* To whom correspondence should be addressed. E-mail: p.saha@ concentration on flux in ultrafiltration of PEG and dextran with
iitg.ac.in. Phone/Fax: +91.361.2582257. a solvent resistance polymeric membrane. They observed that
10.1021/ie801293d CCC: $40.75  2009 American Chemical Society
Published on Web 04/09/2009
Ind. Eng. Chem. Res., Vol. 48, No. 9, 2009 4429
both steady-state flux and gel formation were significantly
influenced in the UF of both PEG and dextran from blended
solvent (ethanol-water).
The unsteady-state flux behavior for silica and dextran in an
ultrafiltration cell has been studied experimentally by van Oers
et al.12 During the filtration of dextran, only a polarization layer
was built up and the osmotic pressure model provided a good
description of the flux for the experiments with dextran. For
silica suspension also, a gel layer formation occurred and was
well predicted by the gel polarization model. An unsteady-state
model was developed to predict the feed surface and permeate
concentration for ultrafiltration of macromolecules using the
diffusion coefficient of the solution through the membrane.13
This model was proficient in predicting the feed surface
concentration once the operating conditions were fixed and did
not make an implicit assumption of constant feed surface
concentration. It was further shown that, for a certain range of
system parameters, a negative rejection (based on the feed
concentration) was possible at steady state.
Figure 1. Schematic representation of a concentration polarization layer
Nakao et al.14 reported that no gel layer was found during over gel layer and mass balance over the concentration polarization layer.
the ultrafiltration of dextran even with a very high solute
concentration of 400 kg/m3. The above reason was also further
confirmed by Choe et al.15 They obtained the permeate flux of acid by zeolite-clay composite membrane using a two-
dextran for totally retentive membranes in a stirred UF cell as dimensional space charge model. They reported a new scheme
a function of time for sudden variations of transmembrane for the solution of space-charge model which reduces the
pressures and concluded that only the polarization layer was computational time considerably. An unsteady-state mass
formed. The osmotic pressure model was well fitted with the transfer model was developed taking the effect of reversible
dextran flux data. They also studied the flux decline of dextran pore plugging25 by the diffusing solute molecules into account.
in UF and concluded that the change in the driving force could The osmotic pressure model has been used to predict the
be explained in terms of osmotic pressure at the membrane membrane surface concentration. The rate of available fractional
surface. Bhattacharjee and Bhattacharya16,17 have developed a area of the membrane that blocked at any time due to pore
model to predict the flux using a generalized formulation which plugging phenomenon was assumed to be the function of
takes into account the unsteady-state behavior during the initial dimensionless membrane surface concentration existing at that
stages of continuous stirred ultrafiltration of PEG. They17 time.
showed the effect of operating variables on limiting flux
In this article, we have modeled the effect of various operating
phenomena and the membrane parameters, i.e., reflection
parameters on the performance of membrane for separation of
coefficient and solute permeability, were determined to char-
dextran and silica suspension using a process engineering
acterize the membrane. A correlation was developed to relate
software, viz., Aspen Engineering Suite. The advantage of this
polarized layer resistance with concentration polarization,
software is manifold. First, this has a dedicated user interface
osmotic pressure to applied pressure ratio and Reynolds number.
Bhattacharjee and Datta18 used the experimental data reported for both steady-state and dynamic process simulation. Unlike
in the literature16,17 to integrate the governing partial differential high-level programming languages such as C++ or Fortran
equation and calculate the back transport coefficient by a least- where the user needs to write rigorous codes, Aspen software
squares fit. The model predicted that the flux decline during has inbuilt object modules that can handle codes for ODE/PDE
ultrafiltration of PEG in a continuous stirred cell using a solver, optimization, real-time graphical output, etc. Second, the
cellulose acetate membrane occurred mainly due to the resis- software has a rich library of commonly used unit operations
tance offered by solute molecules during their back transport such as distillation column, reactors, heat exchanger, etc., which
to the bulk. The effects of osmotic pressure and cake/gel are menu-driven and can easily be integrated with various
formation were assumed to be negligible in this study. A three- process operations. And third and most importantly, Aspen has
parameter model based on unsteady-state mass transfer was a huge database of physical as well as thermodynamic properties
developed for the prediction of flux and rejection of solute of various chemical compounds which could be called in while
during ultrafiltration in an unstirred batch cell.19 The main simulation progresses as and when required. In other words,
feature of this model was that it allowed variation of solute even a minute change of physicochemical property of a target
diffusivity with concentration in the boundary layer whose effect chemical can be predicted and used in the simulation when
was more pronounced in the case of unstirred batch cell. process condition such as temperature, pressure, or concentration
Bhattacharjee et al.20 utilized Kedem-Katchalsky (KK) and changes. This is contrary to the poor assumption that is usually
Spiegler-Kedem (SK) models for the parameter estimation like considered by other custom-made simulation code that takes
solute permeability and reflection coefficient for ultrafiltration constant density/viscosity etc. while performing real-time
of black liquor by using an asymmetric membrane in a stirred simulation. Two different types of membrane are considered
batch cell, which was modified to work on a continuous mode. for batch and continuous mode of operation in a stirred dead
The variations of parameters with other process variable like end UF cell. In this analysis, variations of permeate flux,
bulk concentration, pressure difference, and stirrer speed were membrane surface concentration, permeate concentration, ob-
also established. Shukla and Kumar21-24 have modeled the served and true rejection with the transmembrane pressure,
separation of aqueous solution of FeCl3, AlCl3, and chromic stirrer speed, and initial feed concentration have been studied.
4430 Ind. Eng. Chem. Res., Vol. 48, No. 9, 2009

2. Theory
Both the models discussed above, viz., gel polarization model
and osmotic pressure model, incorporate the phenomenon of
concentration polarization (see Figure 1).7 Based on the film
theory, the formation of a polarization layer can be described
with the following partial differential equation (PDE), assuming
flux J(t), and diffusivity D, as independent of concentration c
of the macromolecular solutes

∂c ∂c ∂2c
) -J(t) + D 2 (1)
∂t ∂x ∂x
Within the membrane (δpol < x < δpol + δmem), the concentration
is governed by13

∂c ∂2c
) Dm 2 (2)
∂t ∂x
The gel layer and membrane resistance may be considered
as two resistances in series, and the permeate flux is then
described by Darcy’s law:

1 dVP ∆P - ∆π
J) ) (3)
Am dt µ(Rm + Rg)
where
∆π ) π(cm) - π(cp) (4)
The gel-polarization model assumes that the concentration
at the membrane surface cannot exceed the gel concentration
Cg.12 Initially, polarization layer forms when the membrane
surface concentration reaches the gel concentration. The gel
Figure 2. Schematic representation of continuous (a) and batch (b) setup.
layer thickness increases with time and reduces its hydraulic
permeability. Figure 1 shows the schematic representation of
concentration polarization layer over the gel layer. For totally Both the cases, membranes are considered as black box, i.e.,
retentive membrane (where cp ) 0), the rate of growth of the membrane parameters like permeability, reflection coefficient,
gel layer thickness can be calculated by a mass balance as and charge density are not judged. In partially retentive
follows7 membrane, diffusivity within the membrane layer is referred to

( )
∂δg as hindered back diffusion. To simplify the modeling, the
D ∂c
)J- (5) membrane layer is represented as granulated layer characterized
∂t Cg ∂x |x)δpol by equivalent diffusivity. The same assumption was used by
The hydraulic resistance of the gel layer of particles can be Agashichev26 and Boudreau27 for calculating effective diffusion
calculated using the Kozeny-Carmen equation. coefficient within the granulated cake layer. The diffusivity of
solute within the membrane can be calculated as
(1 - n)2
Rg ) 180 δg (6) ε
dp2n3 Dm ) D (7)
1 - ln ε2
3. Model In this simulation study, we have considered a stirred dead
end UF cell with a membrane area of Am and initial concentra-
In this work, two different types of solute separation are tion and volume of the feed are cf0 and Vf0, respectively. If the
considered such as (i) solute which has negligible osmotic mass arrested into the polarization layer is neglected, then for
pressure in the solution but has a great tendency to form gel a totally retentive membrane (no permeation of solute through
layer (silica suspension) and (ii) solute which has observable membrane), the change of bulk volume (Vf) and concentration
osmotic pressure and no tendency to form gel layer (dextran) (cf) with time in the UF cell for batch mode operation (Figure
using two different categories of membrane, viz., (i) totally 2b) can be written as7
retentive membrane where solutes cannot pass through the
membrane and (ii) partially retentive membrane where solute dVf
may pass with solvent through the membrane. The following ) -JAm (8)
dt
two cases have been considered here for the performance study dcf cf
using stirred dead end UF setup and is shown in Figure 2. ) JAm (9)
Case I: Separation performance of dextran solution using dt Vf
partially retentive membrane in the continuous mode of opera- The mass transfer coefficient for a stirred batch cell is
tion (Figure 2a). calculated by the following empirical equation:16,18,20

( Dr )( Dν ) ( ωrν )
Case II: Separation performance of silica suspension using
0.33 2 0.8
totally retentive membrane in the batch mode operation (Figure
k ) 0.0443 (10)
2b).
Ind. Eng. Chem. Res., Vol. 48, No. 9, 2009 4431
It is also assumed that the time taken for formation of Table 1. Properties of Membrane, Solute, and Operating Conditions
concentration polarization layer is negligible. Polarization layer of the Dead-End UF Cell7
thickness is calculated by film theory. Many researchers have Membrane Parameter
reported that there is no significant influence of molecular weight
membrane radius r ) 0.06 m
of dextran on osmotic pressure.8,9,12 So in this report, osmotic membrane area Am ) πr2 ) 0.0137 m2
pressure of dextran is calculated using the virial expansion membrane porosity ε ) 0-35%
thickness of the ∂mem ) 10-6 m
π ) A1c + A2c2 + A3c3 (11) membrane
hydraulic resistance Rm ) 1013 m-1
The initial and boundary conditions for solving eqs 1 and 2 for
Cell Parameter
both the cases are given below.
transmembrane pressure ∆P ) 1-12 bar
Case I:
operating temperature T ) 298.15 K
t ) 0; 0 e x e δpol ; c ) cf0 stirring speed ω ) 4.8-10000 rpm
Solute and Solvent Parameter
t > 0; x ) 0; c ) cf ; and x ) δpol ; Jc ) D ( ∂x∂c ) - -
x)δpol 1. dextran
water solution
molecular wt 73500 Da (av dia )11.4 nm)
∂c
D ( )
m
initial concentration cf0 ) 28 kg/ m3
∂x +
x)δpol initial volume Vf0 ) 0.02 m3
virial coefficients A1 ) 37.5
Within the membrane, the PDE eq 2 is solved with the A2 ) 0.752
following boundary condition:13 A3 ) 76.4 × 10-4
2. silica viscosity µ ) 0.001 Pa · s
t ) 0; δpol < x < δpol + δmem ; c)0 water solution
∂c ∂c initial concentration cf0 ) 2.5-120 kg/ m3
t > 0; x ) δpol ; Jc ) D ( )
∂x x)δpol
-
- Dm ( )
∂x +
x)δpol
initial volume
particle diameter
Vf0 ) 0.02 m3
dp ) 13 × 10-4 m
∂c particle density from Aspen property file
x ) δpol + δmem ; Jc ) -D ( )
m solution density from Aspen property file
∂x -
x)δpol+mem
diffusivity from Aspen property file
Case II: molecular wt from Aspen property file
gel porosity n ) 0.37 (assuming spherical
t ) 0; 0 e x e δpol ; c ) cf0 particle)
gel concentration Cg ) 230 kg/m3
∂c
t > 0; x ) 0; c ) cf(t); and x ) δpol ; Jc ) D ( )
∂x x)δpol difference method and the nonlinear equations are solved
using mixed Newton method. Implicit Euler method has been
4. Simulation used as integrator.
For modeling, a flat sheet with circular shape membrane is
All the model equations discussed so far are either ordinary considered for both the cases and its properties are given in
differential equations (ODE) or PDE and most of them are Table 1. It is assumed that there is no dead zone in the cell and
nonlinear. Simultaneous solutions of these equations are the stirrer radius (assumed equal to membrane radius) is
computationally extensive and can only be done through sufficient to maintain the uniform mass transfer rate throughout
appropriate numerical techniques. The entire task is gigantic the membrane area. Table 1 gives the summary of all the
and computationally extensive. Aspen Engineering Suite has parameter value and the solution (dextran and silica) properties
been used in this work to overcome the above difficulties. used in this model.12
The suite also provides an extensive database for physical
and thermodynamic property library, viz., DIPPR, that can
5. Results and Discussion
be directly linked in while performing the simulation. It has
readily available GUI, various types PDE, ODE, nonlinear 5.1. Separation Performance of Dextran Solution Using
or linear solvers, or optimization tools. Aspen Engineering Partially Retentive Membrane in the Continuous Mode of
Suite is a chemical process simulation software that provides Operation. It is generally accepted that concentration polar-
a complete library of models for a number of commonly used ization of solute at the membrane surface can lead to solute
unit operations in the chemical processes such as distillation, adsorption, precipitation, and gel layer formation and this
absorption, extraction, heat exchanger, reactor, etc. Process results in flux decline. This increases operating costs due to
industries that use Aspen Engineering Suite include petro- the need for frequent cleaning and possibly reduces mem-
leum, petrochemical, gas processing, polymer, mineral brane lifetime. In this article, we have investigated to
processing, and so on. Unfortunately, the membrane separa- understand this phenomenon so that its effect can be
tion units have not so far been included in the Aspen minimized. Figure 3 shows the steady-state permeate flux of
Engineering Suite as a ready model. In the present work, dextran solution with variation of stirrer speed for three
Aspen Custom Modeler (ACM), version 20.0, is used for different transmembrane pressures (100, 200, and 300 kPa)
the numerical solution of the model equations. All the with feed concentration of 28 kg/m3. It is clear from the figure
membrane model equations have been coded in Aspen that the flux increases with raise of stirrer speed, and after a
Custom Modeler (ACM) and for getting property data (which certain speed, there is no further increase and it becomes
are pressure, temperature, and concentration dependent) are steady. This means that the increasing stirrer speed results
called within the code by using Aspen Property file. For the in the brushing away of solute deposited on the membrane
case I (Figure 2a), both the concentration polarization layer surface and membrane surface acquires in the vicinity to the
and membrane are divided into 100 parts, while in the case bulk concentration. However, higher transmembrane pressure
II (Figure 2b), the concentration polarization layer is divided (300 kPa) requires higher stirrer speed to reach steady-state
into 100 parts. These PDE are solved with backward finite permeate flux compared to the lower pressure (100 kPa). This
4432 Ind. Eng. Chem. Res., Vol. 48, No. 9, 2009

Figure 3. Effect of stirring speed on steady-state permeate flux of dextran Figure 4. Effect of stirring speed on steady-state membrane surface
solution for different transmembrane pressure (at feed concentration ) 28 concentration of dextran solution, Cm, for different transmembrane pressures
kg/m3; membrane porosity ) 10%; membrane thickness ) 1 µm). (at feed concentration ) 28 kg/m3; membrane porosity ) 10%; membrane
thickness ) 1 µm).

is due to the increase of concentration at the membrane


surface as the driving pressure increases and consequently
the osmotic pressure also increases. In a stirred cell, the
growth of concentration polarization layer is usually less due
to turbulence and, at higher speed, the layer resistance drops.
The polarization layer is gradually removed by increasing
the stirrer speed and beyond certain extent of turbulence;
the resistance offered by this additional layer is negligible.
It can be seen from eq 3 that the osmotic pressure is only
for the bulk concentration which turns into fixed (as
continuous mode) and the flux becomes unchanging. As
expected, an increase of pressure stands for increase of flux
which is also confirmed by the above reason. van Oers et
al.12 have proved experimentally that the ultrafiltration of
dextran was found to satisfy the case of only osmotic pressure
limited flux, even if flux at higher concentration was found
to be pressure independent. In our case also, there is osmotic
pressure limited flux established due to no gel formation.
Figures 4 and 5 demonstrate the steady-state decrease of
concentration at the surface of membrane (Cm) and permeate
concentration (Cp), respectively, with variation of stirrer
speed. From these figures, it is understandable that with the
augment of stirrer speed, the membrane surface is obtaining Figure 5. Effect of stirring speed on steady-state permeate concentration
toward the bulk concentration. It is expected that an increase of dextran solution, Cp, for different transmembrane pressures (at feed
of applied pressure raises the concentration at the surface of concentration ) 28 kg/m3; membrane porosity ) 10%; membrane thickness
) 1 µm).
membrane. The membrane used in this analysis retained the
solute partially. This high concentration at the membrane Figure 6 explains the observed and true rejection of dextran
surface increases the true rejection (see Figure 6) and the with variation of stirrer speed for different applied trans-
low diffusion coefficient of dextran (4.6 × 10-11 m2 s-1) membrane pressures (100, 200, and 300 kPa). As the
reduces the rate of back transport at higher applied pressure. permeate concentration diminishes with the higher stirrer
It is confirmed from the results that more stirring is required speed, the observed rejection increases for all the pressures
for closely obtaining the bulk concentration at the membrane studied (Figure 6). For a particular stirrer speed, permeate
surface. concentration increases with the applied pressure, resulting
The permeate concentration also depends on the permeate in the reduction of observed rejection (Ro). It can be seen
flux and mass accumulated on the membrane surface and that, for lower stirrer speed, negative observed rejection is
hence increases with applied pressure for a fixed stirrer speed obtained, which is because the adsorbed solute particles
(see Figure 5). This implies that increasing pressure forces within the membrane pores try to come out at higher pressure
the solute to pass more through the membrane pores. and cause the permeate concentration to become more than
Bhattacharjee and Datta18 have also reported that the the feed concentration. Karode13 has also observed the same
membrane surface concentration depends on stirring speed. and explained the possibility of this condition by simulation.
Ind. Eng. Chem. Res., Vol. 48, No. 9, 2009 4433

Figure 6. Effect of stirring speed on steady state observed and true


rejection of dextran solution for different transmembrane pressures (at
feed concentration ) 28 kg/m3; membrane porosity ) 10%; membrane
Figure 8. Unsteady-state permeate flux of dextran solution with time for
thickness ) 1 µm).
different stirrer speeds (at feed concentration ) 28 kg/m3; transmembrane
pressure ) 200 kPa; membrane porosity ) 10%; membrane thickness ) 1
µm).

Figure 8 depicts the unsteady-state permeate flux variation of


dextran solution with time for different stirrer speed with initial
feed concentration of 28 kg/m3 and transmembrane pressure of
200 kPa. As expected, there is an initial sharp drop in the permeate
flux from the initial value for short filtration time. Zaidi and Kumar8
have experimentally found the similar type of flux decline trend
for the dextran water solution using unstirred batch cell. Chudacek
and Fane29 have also used stirred batch cell with stirrer speed of
1000 rpm for the separation of dextran solution and obtained the
same type of results what we reported in Figure 8. Many researchers
suggested that the permeate flux versus time curve can be divided
into two domains.30-32 Domain 1 is the initial decline of flux that
may be attributed to the membrane fouling. This domain is
characterized by an operationally irreversible deposition of material
in the membrane pores and the sorption on the membrane surface
of the macromolecules initially present in the bulk solution. Domain
2, the remaining flux versus time curve, may be associated with
the concentration polarization and gel layer formation phenomena.
Figure 7. Effect of stirring speed on steady-state permeate concentration This domain is characterized by the reversible deposition of
of dextran solution for different membrane porosities (at feed concentra- macromolecules on the membrane surface. Mehta33 has also
tion ) 28 kg/m3; transmembrane pressure ) 100 kPa; membrane developed a model and reported that the rate of flux decline with
thickness ) 1 µm).
time is a linear function of the difference between the flux at any
The same was experimentally proved by Balakrishnan et al.28 time during the time period and the flux at the end of the time
The observed rejection is found to be lower than the true period investigated. The resulting independent solutions were
rejection for all the ranges of stirrer speed and transmembrane combined to provide the global solution applicable for the entire
pressure studied. This indicates that the concentration flux vs time curve. However, one can minimize the flux decline
polarization effects are significant and leads to considerable due to adsorption of solute layers at the membrane surface by
loss of efficiency. This clearly shows that Ro is strongly changing the membrane material or chemically modifying it. The
dependent on the operating conditions while Rt represents adsorption at the membrane surface occurs due to the strong
the true rejection of the membrane. The porosity of the interaction between the solute and membrane material. The result
membrane has also great impact on permeate concentration. of Cherkasov et al.34 study showed that hydrophobic membranes
It can be seen from Figure 7 that the permeate concentration attracted a thicker irreversible adsorption layer than hydrophilic
increases with porosity of membrane which is due to the more membranes.
permeation of dextran through the pores. For a particular Figures 9 and 10 demonstrate the unsteady-state effect of
porosity, as the stirrer speed increases, membrane surface membrane surface concentration and permeate concentration
concentration decreases and so the permeate concentration with time for different stirrer speed with constant transmem-
also decreases. brane pressure of 200 kPa, respectively. Many studies
4434 Ind. Eng. Chem. Res., Vol. 48, No. 9, 2009

Figure 9. Unsteady-state effect of membrane surface concentration (Cm)


of dextran solution with time for different stirrer speeds (at feed concentra- Figure 11. Unsteady-state effect of permeate flux of dextran solution with
tion ) 28 kg/m3; transmembrane pressure ) 200 kPa; membrane porosity time for different transmembrane pressures (at feed concentration ) 28
) 10%; membrane thickness ) 1 µm). kg/m3; stirrer speed ) 57.3 rpm; membrane porosity ) 10%; membrane
thickness ) 1 µm).

quickly with increase of stirrer speed. This is because of the


augmentation of the back diffusion rate resulting in less
concentration at the membrane surface and so flux reaches
steady state quickly. It elucidates that higher stirrer speed
makes the flux variation small for a particular pressure that
is due to less membrane surface concentration and the
permeate concentration decreases with increasing stirrer
speed. It can be clearly seen from the Figures 9 and 10 that
the time required to reach steady-state permeate concentration
(Cp) and membrane surface concentration (Cm) is not same.
In all the cases, membrane surface concentration requires less
time to obtain steady state than the permeate concentration.
Figures 11-13 explain the unsteady-state effects of permeate
flux, membrane surface concentration, and permeate con-
centration of dextran for different transmembrane pressure
(100, 200, and 300 kPa), respectively, at 57.3 rpm stirrer
speed. For a fixed stirrer speed, increasing applied pressure
enhances the flux. Initially, the change of flux with time (dJ/
dt) is different for different pressures and after a certain time,
it becomes the same and the gradient becomes zero after 10 s.
From Figure 11, one can notice that at 1 s, dJ/dt for the
pressures of 100, 200, and 300 kPa are almost equal. This
Figure 10. Unsteady-state effect of permeate concentration (Cp) of dextran gradient becomes equal for the pressure of 200 and 300 kPa
solution with time for different stirrer speeds (at feed concentration ) 28 at 0.1 s, suggesting that at higher pressure there would be
kg/m3; transmembrane pressure ) 200 kPa; membrane porosity ) 10%; no gradient difference. The experimental results reported in
membrane thickness ) 1 µm).
ref 8 also support the trend presented in Figure 11, though
reported experimentally that, at the start of separation, the time scales do not match well because experiments were
membrane surface concentration is equal to bulk concentra- performed by using unstirred batch cells with low dextran
tion. So, osmotic pressure is contributed only by bulk concentration. The flux decline due to increase of the
concentration of the solution. As the time progress, membrane concentration of membrane surface is also time dependent.
surface acquires more mass from the bulk due to the However, the doubling of applied pressure does not lead to
convection. Some acquired mass on the membrane surface the doubling of permeate flux, because the increase of applied
is returned to the bulk by diffusion due to the higher pressure partially was compensated by an increase of osmotic
concentration than the bulk. After a certain time, membrane pressure due to increase of membrane surface concentration.
surface reaches the equilibrium concentration (see Figure 9), It is observed from Figures 12 and 13 that the membrane
which is more than bulk concentration and flux becomes surface and permeate concentration acquire higher concentra-
steady for that transmembrane pressure and stirrer speed. It tion for higher pressure as is common in filtration practice.
is observed from the Figure 9 that flux attains steady state Zaidi and Kumar8 experimentally proved that the rejection
Ind. Eng. Chem. Res., Vol. 48, No. 9, 2009 4435

Figure 12. Unsteady-state effect of membrane surface concentration of


dextran solution with time for different transmembrane pressures (at feed Figure 14. Unsteady-state effect of membrane surface concentration of silica
concentration ) 28 kg/m3; stirrer speed ) 57.3 rpm; membrane porosity ) suspension with time for different transmembrane pressures for stirred batch
10%; membrane thickness ) 1 µm). mode (initial feed concentration cf0 ) 2.5 kg/m3; stirrer speed ) 9.55 rpm).

Figure 13. Unsteady-state effect of permeate concentration of dextran Figure 15. Unsteady-state effect of permeate flux of silica suspension with
solution with time for different transmembrane pressures (at feed concentra- time for different transmembrane pressures for stirred batch mode (initial
tion ) 28 kg/m3; stirrer speed ) 57.3 rpm; membrane porosity ) 10%; feed concentration cf0 ) 2.5 kg/m3; stirrer speed ) 9.55 rpm).
membrane thickness ) 1 µm).
mode of operation using stirred UF cell (see Figure 2b). The
of dextran was more for lower pressure compared to that of mass of silica in the upstream side is same as the initial mass
higher applied pressure. This implies that the permeate for all the process time, as there is no passage of silica
concentration is low at low applied pressure while higher through the membrane and only water is passing through
pressure gives high permeate concentration. Our simulation membrane. It is well-known that the separation of silica
results reported in Figures 12 and 13 also support this suspension by membrane forms a gel layer on the membrane
experimental phenomenon and show the dynamics of mem- surface. The osmotic pressure effect is negligible and hence
brane surface concentration and permeate concentration. flux is governed by a gel polarization model. Figures 14 and
5.2. Separation Performance of Silica Suspension Using 15 demonstrate the plot of unsteady-state membrane surface
Totally Retentive Membrane in the Batch Mode Opera- concentration (Cm) and permeate flux with time for silica
tion. In this case, we have used a totally retentive membrane suspension with different transmembrane pressures (200-900
to study the separation performance of silica in the batch kPa). The results indicate that there is no reduction of applied
4436 Ind. Eng. Chem. Res., Vol. 48, No. 9, 2009

pressure because of negligible osmotic pressure of silica.


There is no flux reduction with the time until the membrane
surface obtains the gel concentration (shown in Figure 15)
and this time is termed as characteristic time.10 Higher
pressures result in less characteristic time than the lower
pressure. For this simulation, we have taken the gel concen-
tration value of silica solute as 230 kg/m3 from the published
literature12 for different operating pressures, though the
thermodynamic relation of gel concentration depends on both
temperature and pressure35,36 and their experimental work
showed almost constant (230 kg/m3)12 for varying pressure
ranges from 100-200 kPa. The gel layer initiates to form
over the membrane surface and increases with the time of
operation. This layer provides an additional resistance that
results in a flux decline. For higher pressures, the membrane
surface acquires the gel concentration much faster than for
the lower pressure for a fixed initial feed concentration, which
can be seen from Figures 14 and 15. For lower pressure, no
gel layer appears within 2000 s. For an example, at 400 kPa
transmembrane pressure, membrane surface reaches gel
concentration at 1135 s while for 500 kPa pressure it takes
170 s. Before gel formation, there is small flux decline that Figure 16. Unsteady-state effect of membrane surface concentration of silica
is due to the increase of viscosity. For all the pressure ranges suspension with time for different initial feed concentrations (transmembrane
studied, the increase of membrane surface concentration has pressure ) 200 kPa; stirrer speed ) 9.55 rpm).
two slopes (see Figure 14). The steepest slope is for initial
unsteady effect due to the formation of concentration
polarization. The second slope is due to the increase of
upstream side concentration. So membrane surface concen-
tration increases until it reaches the gel concentration. This
high concentration at the membrane surface is due to high
rejection of silica which is found to be nearly 100%. The
concentration at the membrane surface increases with the
transmembrane pressure for a fixed initial feed concentration.
This is mainly due to the low rate of back transport of silica
(because of low diffusion coefficient of silica). This implies
that gel formation is rapid and instantaneous at the beginning
of separation, particularly at higher applied pressure. From
Figure 15, it is unambiguous that the permeate flux follows
the same path after a certain time for 700, 800, and 900 kPa
transmembrane pressure; i.e., the flux becomes independent
of pressure which is known as the limiting flux. Figure 16
depicts the unsteady-state effect of initial feed concentration
of silica suspension on the membrane surface concentration
with time. It demonstrates that pressure of 200 kPa is not
sufficient for the initial feed concentration of 120 kg/m3 to Figure 17. Comparison of unsteady-state permeate flux of silica
acquire the gel concentration within 2000 s at the membrane suspension versus time simulated by using Aspen with experimental data
surface. Figure 17 shows the comparison of permeate flux reported by Zaidi and Kumar10 (initial silica feed concentration of 28
kg/m3, pressure of 270 and 405 kPa).
versus time obtained from simulation by using Aspen with
experimental data reported by Zaidi and Kumar10 (initial the stirrer speed of the UF cell from 4.8 to 47.7 rpm by keeping
silica feed concentration of 28 kg/m3, pressure of 270 and the initial feed concentration and transmembrane pressure of
405 kPa). It can be seen that experimental results match well 30 kg/m3 and 1200 kPa, respectively, and the results are shown
with the simulation done in Aspen. So it is strongly advocated in Figures 19-21. As can be seen from Figure 19, at the
that Aspen Engineering Suite is justified as a modeling tool beginning, the permeate flux is a time invariant for all the stirrer
for NF/UF/MF. speed studied. After a few seconds, a sudden drop in the flux is
The effect of gel layer thickness of silica suspension with observed due to the buildup of polarization at the membrane
time for different transmembrane pressures with initial feed surface. This leads to an increase in the concentration at the
concentration of 30 kg/m3 is shown in Figure 18. At higher membrane surface and the same can be seen in Figure 20. As
pressure, gel layer thickness is more which results in more the stirrer speed increases, the time required to reach the gel
resistance. With the initial feed concentration of 30 kg/m3 and layer formation increases which is due to the removal of silica
stirrer speed of 9.55 rpm, the applied pressure of 200 kPa is particle from surface of membrane. Figure 21 represents the
not enough to form gel within 400 s. It is expected that with deposited gel layer thickness as function of time for an initial
increasing pressure the gel layer formation is faster. To study feed concentration of 30 kg/m3 and transmembrane pressure of
the influence of stirrer speed on flux, concentration of the 1200 kPa. At a particular time, the rate of solute deposition is
membrane surface (Cm), and gel layer thickness, we have varied higher for lower stirrer speed that leads to build up of a thick
Ind. Eng. Chem. Res., Vol. 48, No. 9, 2009 4437

Figure 20. Unsteady-state effect of membrane surface concentration of silica


Figure 18. Unsteady-state effect of gel layer thickness of silica suspension suspension with time for different stirrer speeds in batch mode operation
with time for different transmembrane pressures (initial feed concentration (initial feed concentration cf0 ) 30 kg/m3; transmembrane pressure ) 1200
cf0 ) 30 kg/m3; stirrer speed ) 9.55 rpm). kPa).

Figure 19. Unsteady-state permeate flux of silica suspension with time for Figure 21. Unsteady-state effect of gel layer thickness of silica suspension
different stirrer speeds in batch mode operation (initial feed concentration with time for different stirrer speeds in batch mode operation (initial feed
cf0 ) 30 kg/m3; transmembrane pressure ) 1200 kPa). concentration cf0 ) 30 kg/m3; transmembrane pressure ) 1200 kPa).
layer and is also confirmed from Figure 19. The reason for this
behavior may be due to the increase in the bulk concentration The flux decline rate is larger for higher applied pressure
at the upstream side of membrane that provides additional with constant stirrer speed. As the membrane porosity
resistance. increases, the observed rejection and permeate concentration
decrease. For silica suspension, an increase of stirrer speed
reduces the flux and increases the time required to reach the
6. Conclusion
gel layer formation at the membrane surface. At the begin-
We have successfully modeled the separation performance ning, permeate flux is time invariant for all the stirrer speeds
of dextran and silica suspension in a dead-end stirred UF studied; after a few seconds, a sudden drop in the flux is
using Aspen Engineering Suite. For dextran solution, it is observed due to build up of polarization at the membrane
observed that the permeate flux increases with the stirrer surface. For a fixed transmembrane pressure and stirrer speed,
speed up to a certain range beyond which the flux remains fouling is more for higher initial feed concentration. It is
constant. Higher stirrer speed with a constant transmembrane also concluded from the above study that higher stirrer speed
pressure offers more permeate flux and observed rejection. and less initial feed concentration are favorable for pressure-
4438 Ind. Eng. Chem. Res., Vol. 48, No. 9, 2009

driven membrane filtration process by providing less flux υg thickness of the gel layer
decline. Simulation results of UF of silica suspension using ∂mem thickness of membrane (m)
Aspen Engineering Suite was compared with experimental ∂pol thickness of concentration polarization layer (m)
data reported in the literature and it matches quite well. It
justifies that the Aspen Engineering Suite can also be used
as a modeling tool for NF/UF/MF. Literature Cited
(1) Mulder, M. H. Basic Principles of Membrane Technology; Kluwer
Acknowledgment Academic Publishers: Dordrecht, The Netherlands, 1996.
(2) Scot, K. Handbook of Industrial Membranes; Elsevier: Oxford, UK,
The authors gratefully acknowledge the financial support from 1995.
CSIR project grant (ref no: 22(0435)/07/EMR-II) from which (3) Nicolas, S.; Balannec, B.; Bariou, B. Ultrafiltration and reverse
the Aspen Engineering Suite was procured. osmosis of small non-charged molecules: a comparison study of rejection
in a stirred and an unstirred batch cell. J. Membr. Sci. 2000, 164, 141–155.
NOMENCLATURE (4) Goldsmith, R. L. Macromolecular ultrafiltration with microporous
membrane. Ind. Eng. Chem. Fundam. 1971, 10, 113–120.
Am membrane cross-sectional area (m2) (5) Wang, L.; Song, L. Flux decline in crossflow microfiltration and
A1-3 virial coefficient ultraflltration: experimental verification of fouling dynamics. J. Membr. Sci.
1999, 160, 41–50.
c solute concentration (kg m-3) (6) Porter, M. C. Concentration polarization with ultrafiltration. Ind. Eng.
cp solute concentration of the permeate (kg m-3) Chem. Prod. Res. DeV. 1972, I1, 234–248.
(7) Karode, S. K. Unsteady state flux response: a method to determine
cm solute concentration at the membrane surface (kg the nature of the solute and gel layer in membrane filtration. J. Membr.
m-3) Sci. 2001, 188, 9–20.
cf0 initial feed solute concentration (kg m-3) (8) Zaidi, S. K.; Kumar, A. Experimental studies in dead-end ultrafll-
tration of dextran: Analysis of concentration polarization. Sep. Purif.
D diffusivity of solute (m2 s-1) Technol. 2004, 36, 115–130.
Dm effective diffusivity of the solute through the mem- (9) Zaidi, S. K.; Karode, S. K.; Kirpalani, D.; Kumar, A. A new method
for identifying osmotically limited and gel layer controlled pressure
brane (m2 s-1)
independent flux in ultrafiltration. Can. J. Chem. Eng. 2004, 82, 343–348.
dp solute particle diameter (m) (10) Zaidi, S. K.; Kumar, A. Experimental analysis of a gel layer in
J permeate flux (m3 m-2 s-1) dead-end ultrafiltration of a silica suspension. Desalination 2005, 172, 107–
117.
k mass transfer coefficient (m/s) (11) Zaidi, S. K.; Kumar, A. Effects of ethanol concentration on flux
M molecular weight and gel formation in dead end ultrafiltration of PEG and dextran. J. Membr.
Sci. 2004, 237, 189–197.
MW molecular weight (Da) (12) van Oers, C. W.; Vorstman, M. A. G.; Muijselaar, W. G. H. M.;
MWCO molecular weight cutoff Kerkhof, P. J. A. M. Unsteady-state flux behaviour in relation to the presence
of a gel layer. J. Membr. Sci. 1992, 73, 231–246.
n porosity of cake or gel layer (13) Karode, S. K. A new unsteady state model for macromolecular
NF nanofiltration ultraltration. Chem. Eng. Sci. 2000, 55, 1769–1773.
r radius of the membrane (m) (14) Nakao, S.; Wijmans, J. G.; Smolders, C. A. Resistance to the
permeate flux in unstirred ultrafiltration of dissolved macromolecular
Ro observed rejection solutions. J. Membr. Sci. 1986) , 26, 165–178.
Rt true or intrinsic rejection (15) Choe, T. B.; Masse, P.; Verdier, A.; Clifton, M. J. Flux decline in
batch ultrafiltration: concentration polarization and cake formation. J.
Rg membrane hydraulic resistance Membr. Sci. 1986, 26, 1–15.
Rm gel resistance (16) Bhattacharjee, C. Prediction of limiting flux in ultrafiltration of Kraft
black liquor. J. Membr. Sci. 1992, 72, 137.
RO reverse osmosis (17) Bhattacharjee, C. Flux decline analysis in ultrafiltration of Kraft
UF ultrafiltration black liquor. J. Membr. Sci. 1993, 82, 1.
(18) Bhattacharjee, C.; Datta, S. Analysis of mass transfer during
t time (s) ultrafiltration of PEG-6000 in a continuous stirred cell: effect of back
Vf0 initial feed volume (m3) transport. J. Membr. Sci. 1996, 119, 39–46.
Vf feed volume (m3) (19) Bhattacharjee, C.; Datta, S. A numerical simulation for the
prediction of flux and rejection during ultrafiltration in unstirred batch cell
Vp permeate volume (m3) using variable diffusivity concept. Sep. Purif. Technol. 2001, 24, 13–22.
x distance from the membrane (m) (20) Bhattacharjee, C.; Sarkar, P.; Datta, S.; Gupta, B. B.; Bhattacharya,
P. K. Parameter estimation and performance study during ultrafiltration of
∆P applied pressure difference between permeate and Kraft black liquor. Sep. Purif. Technol. 2006, 51, 247–257.
retenate side (Pa) (21) Shukla, A.; Kumar, A. Modeling of separation of aqueous solutions
∆π osmotic pressure difference between permeate and of FeCl3 and AlCl3 by zeolite-clay composite membranes using a space-
charge model. J. Colloid Interface Sci. 2004, 274, 204–215.
retenate side (Pa) (22) Shukla, A.; Kumar, A. Analysis of separation of chromic acid by
∆t differential time (s) zeolite-clay composite membrane using space-charge model. J. Membr. Sci.
2004, 237, 119–113.
∆x differential thickness (m)
(23) Shukla, A.; Kumar, A. Characterization of chemically modified
υ stoichiometric coefficient of solute zeolite-clay composite membranes using separation of trivalent cations.
µ viscosity of the solution (P) Separ. Purif. Technol. 2005, 41 (1), 83–89.
(24) Shukla, A.; Kumar, A. Separation of Cr(VI) by zeolite-clay
Fp solute density (kg m-3) composite membranes modified by reaction with NOx. Sep. Purif. Technol.
Fg gel density (kg m-3) 2007, 52 (3), 423–429.
(25) Ghose, S.; Bhattacharjee, C.; Datta, S. Simulation of unstirred batch
π osmotic pressure (Pa) ultrafiltration process based on a reversible pore plugging models. J. Membr.
ε porosity of membrane Sci. 2000, 169, 29–38.
(26) Agashichev, S. P. Enhancement of concentration polarization due
ω stirring speed (rad/s) to gel accumulated at membrane surface. J. Membr. Sci. 2006, 285, 96–
υ kinematic viscosity (m2 s-1) 101.
Ind. Eng. Chem. Res., Vol. 48, No. 9, 2009 4439
(27) Boudreau, B. P. The diffusive tortuosity of fine-grained unlithified (33) Mehta, B. “Processing of model compositional whey solutions with
sediments. Geochim. Cosmochim. Acta 1996, 60, 3139–3142. pressure driven membranes”, Ph.D. thesis, Ohio State University, 1973.
(28) Balakrishnan, M.; Agarwal, G. P.; Cooney, C. L. Study of protein (34) Cherkasov, A. N.; Tsareva, S. V. Selective properties of ultrafil-
transmission through ultrafilltration membranes. J. Membr. Sci. 1993, 85, tration membranes from the standpoint of concentration polarization and
111. adsorption phenomena. J. Membr. Sci. 1995, 104, 157–164.
(29) Chudacek, M. W.; Fane, A. G. The dynamics of polarization in (35) Karode, S. K. A method for prediction of the gel concentration in
unstirred and stirred ultrafiltration. J. Membr. Sci. 1984, 21, 145–160. macromolecular ultrafiltration. J. Membr. Sci. 2000, 171, 131–139.
(30) Constenla, D. T.; Lozano, J. E. Hollow fibre ultrafiltration of apple (36) Lele, A. K.; Badiger, M. V.; Hirve, M. M.; Mashelkar, R. A.
juice macroscopic approach. Lebensmittel-Wissenschaft Technol. 1997, 30, Thermodynamics of hydrogen-bonded polymer gel-solvent systems. Chem.
373–378. Eng. Sci. 1995, 50, 3535–3545.
(31) Lahoussine-Turcaud, V.; Wiesner, M. R.; Bottero, J. Y. Fouling in
tangential flow ultrafiltration: the effect of colloid size and coagulation ReceiVed for reView August 25, 2008
pretreatment. J. Membr. Sci. 1990) , 52, 173–190. ReVised manuscript receiVed January 14, 2009
(32) Ben Amar, R.; Gupta, B. B.; Jarin, M. Y. Apple juice clarification Accepted March 16, 2009
using mineral membrane: fouling control by backwashing and pulsating
flow. J. Membr. Sci. 1990, 55, 1620–1625. IE801293D

You might also like