Article

pubs.acs.org/IECR

Limiting Flux and Critical Transmembrane Pressure Determination

Using an Exponential Model: The Effect of Concentration Factor,
Temperature, and Cross-Flow Velocity during Casein Micelle
Concentration by Microfiltration
Carolina L. Astudillo-Castro*
Escuela de Alimentos, Pontificia Universidad Católica de Valparaíso, Waddington 716, Valparaíso, Chile

ABSTRACT: During microfiltration, limiting (JL) and critical flux (JC) are essential. Using these concepts, the operational zones,
have been established for flux (J) versus transmembrane pressure (ΔPT). However, the resistance model cannot be used for an
accurate description of these operational zones. The aim of this work was to evaluate the application of an equation for process
parameters’ determination during skim milk microfiltration. The exponential model, J = JL(1 − exp(−ΔPT/(ΔPT)C), allowed a
direct determination of JL and the critical ΔPT ((ΔPT)C) and allowed the calculation of JC and the limiting ΔPT ((ΔPT)L). The
effect of temperature, concentration factor, and cross-flow velocity over JL and (ΔPT)C were evaluated. For JL and (ΔPT)C, a good
fitting was obtained (Radj2 = 98.12%; Radj2 = 97.83%, respectively). These results allow to predict the limiting and critical points
for given values of the concentration factor, cross-flow velocity and temperature. Moreover, determination of the operational
zones could be established.

1. INTRODUCTION Zone 1, for low ΔPT values, only the concentration polarization
During the last decades, the industry has shown great interest in phenomenon exists, and it is known as a subcritical zone, where
membrane separation technology, for example, in the dairy the permeate flux is lower than the critical flux (JC). Zone 2 is
industry with the use of microfiltration (MF) for macro- where cake formation and consolidation occur, as well as pore
molecules from milk and whey concentration and separation1 blocking or protein adsorption onto the membrane. Finally,
and in water treatment processes due to its high rate of Zone 3 is where the cake compacts, which is an undesirable
rejection of particles and bacteria.2 Milk as a process fluid is phenomenon, because this irreversible fouling type is difficult to
interesting in membrane processing. Its multicomponent and remove even using chemical membrane cleaning. The opera-
complex nature with high concentration of several components tional curves show two important points: the critical point and
with different particle sizes, such as microorganisms, fat, casein the limiting point. The critical point [(ΔPT)C, JC] is where the
micelles, soluble proteins, lactose, makes its fractioning a linear relationship with the transmembrane pressure is lost,16
challenge.3 MF with ceramic membranes has been used for fat and the limiting point [(ΔPT)L, JL] is where the maximum flux
separation,4 casein concentration,3,5,6 and microorganism value is obtained under some process condition sets;17 it cannot
removal from milk7−12 and sweet whey,13 because it is a be increased by varying ΔPT.
milder technique in comparison to thermal processing, such as Several studies for the prediction of the maximum limiting
pasteurization or evaporative concentration, which both cause flux (JL) have been conducted using back-transport mod-
protein denaturation.14 els.18−20 The limiting flux depends on the shear stress applied,
Nevertheless, fouling of the membrane, which can be the feeding stream properties and the module geometry. The
reversible or irreversible in nature,1,14 has been the main critical flux theory (JC) was introduced for the first time by
limiting factor for enhancing the productivity of the membrane Field et al.,16 and it describes the flux for which an irreversible
processes. Short-time reversible fouling, that is, pore blocking fouling does not form on the membrane. This fouling type is a
or cake formation,3 occurs by the deposition of milk protein consequence of the dispersed phase transition, from the
and colloids, and concentration polarization in nanofiltration concentration polarization to a condensed phase, cake
and reverse osmosis membranes.14 Reversible fouling can be formation, or pore blocking.18,21 If the system is operated
removed through effective membrane cleaning, but there are under a J < JC condition, the possibility of generating
some costs involved related to labor, energy, and the use of irreversible fouling of the membrane can be eliminated or at
chemicals and water. In the case of milk microfiltration for least reduced.22
casein concentration and native whey production, fouling is When a module is operated under subcritical conditions, a
mainly produced by milk proteins such as casein micelles and continuous process can persist for several hours without a
serum proteins.
For operational curves of permeate flux (J) versus trans- Received: August 21, 2014
membrane pressure (ΔPT), three areas or zones related with Revised: December 3, 2014
membrane fouling have been described.3,15 Each has a different Accepted: December 4, 2014
flux behavior as a function of the transmembrane pressure. In Published: December 4, 2014

© 2014 American Chemical Society 414 dx.doi.org/10.1021/ie5033292 | Ind. Eng. Chem. Res. 2015, 54, 414−425
Industrial & Engineering Chemistry Research Article

decrease in the permeate flux.23 However, experimental zone be determined? And, for each condition, what is the flux
evidence of low fouling formation during subcritical operation value?
conditions of a membrane bioreactor have been observed.24 In this context, the aim of this work was to propose the use
This shows that when the term subcritical operation is used, it of an analytical equation to describe the whole curve of the flux
usually refers to “low fouling” rather than “zero fouling”. Even if versus transmembrane pressure and to describe its use for the
the operation of a system under a J < JC condition diminishes determination of the limiting flux. A case of study skim milk
the fouling of a membrane, it is not feasible to eliminate it microfiltration was carried out for casein micelle concentration,
entirely.25 The subcritical operation zone means decreased which had, in average, a particle size over the membrane cut off
production costs26 due to the energy savings during operation, tested.
and it requires fewer membrane cleaning cycles.21,23,27,28
The highest cost in membrane systems operation is related to 2. MATERIAL AND METHODS
membrane fouling,23 and it is the limiting step in the 2.1. Reagents. As a raw material, powdered skim milk from
development of current membrane technology.29 To obtain a Chilean milk factory (Colún) was used. This powdered milk
the same flow process at J = JC, a larger than J = JL membrane all belonged to the same batch (315/69) and was dissolved in
surface is required,30 an element that is related to the plant warm water (40 °C) by mild stirring for 30 min. Then, it was
investment cost. However, during the last several years, a left standing for 30 min to reach an adequate hydration level of
decrease in the cost of membrane devices has been observed, the casein micelles. The reconstituted skim milk showed an
and hence, the operation costs now deserve greater attention.29 average particle diameter (D[3,2]) of 0.41 μm; the analysis was
For this reason, the subcritical operation zone is considered a performed by the Centro Chileno de Energiá Nuclear
valid option for the industry, and it is used as a valid tool for (CCHEN) using laser diffraction equipment (Mastersizer X,
process optimization in terms of fouling control.31 When the Malvern Instruments, 0.63 μm laser wavelength, MSX1, U.K.).
critical flux hypothesis is used with the production cost Several reconstitutions of powdered skim milk were
optimization, it is possible to obtain a sustainable flux that performed with different solids contents to evaluate the
represents the lower flux value where the fouling generation is concentration factor (CF) effect during the skim milk
economically acceptable for the plant operation.31 This is the microfiltration process. Whereas CF = 0.5 represents diluted
reason why a subcritical operation, that is, low constant flux, skim milk with half of the normal solids and casein
low transmembrane pressure, low resistance and low fouling, concentration, CF = 1.0 represents regular skim milk, with
seems to be the key for sustainable membrane operation.26,30 the typical solids and casein content. CF = 1.5 represents a
When cake formation is lower, operating the system within concentrated skim milk with an extra 50% of solids and casein
Zone 2 optimizes the process capacity and minimizes the contents when compared to regular milk.
membrane area used. However, in this zone, the selectivity is The skim milk and cleaning solutions were prepared with
not optimal, due a cake layer formation,3 usually called deionized water (conductivity ≤ 10 μS/cm). Ultrasil11 was
“dynamic membrane”, which can reject more molecules than supplied by Henkel-Ecolab (U.S.A.).
the original membrane. Until now, the industry has opted to 2.2. Experimental Setup. Figure 1 shows the experimental
operate at the critical flux level to concentrate whey protein, for setup used for all the trials. A Membralox module (Pall, U.S.A.)
bacteria and spore removal and to operate at limiting flux
conditions for casein concentration.3
The classical model for membrane resistance14,32 based in
the Darcy’s law33 is used for the determination of the total
resistance of any membrane process. This model is shown in eq
1:
ΔPT
J=
μRT (1)
where μ is the permeate viscosity and RT is the total membrane
resistance. RT depends on the resistance offered by the
membrane (RM) and by the irreversible (RIF) and reversible Figure 1. Full recirculation mode setup.
fouling (RRF), as shown in eq 2.
RT = RM + RIF + RRF was used with a tubular ceramic membrane of 0.14 μm (pore
(2)
size), 3.6 mm diameter, and 0.0094 m2 area (Tami, France).
from where the total fouling term stems and is summarized in Stainless stell manometers (Wika, Chile) coupled to the
eq 3: membrane module and rotatory pumps (Fluid-o-Tech, U.S.A.)
RTF = RIF + RRF were included.
(3)
The pumps were connected to a 0.5 HP engine for all the
Normally, RTF is higher than RM because of a layer deposited trials. The pump characteristic curve describes the flow in L/
onto the membrane.34,35 Equation 1 is unable to describe the min (Q) that is capable to impulse the fluid, and the inlet
flux versus transmembrane pressure behavior when the linear pressure to the module (Pi) is in bars. All the equations of the
correlation between the flux and transmembrane pressure is lost pump characteristic curves were performed with deionized
due to the concentration polarization and fouling effects. The water at 20 °C, and they are shown in Table 1.
correlation is valid only for low transmembrane pressures. The milk temperature was kept in a narrow range (±0.1 °C)
However, what would happen in the case of a transmembrane using an immersion bath circulator (LabTech, Iran), and the
pressure close to the limiting value? How can the trans- process fluid temperature was maintained with a precision of
membrane pressure required for the operation of each specific ±0.5 °C.
415 dx.doi.org/10.1021/ie5033292 | Ind. Eng. Chem. Res. 2015, 54, 414−425
Industrial & Engineering Chemistry Research Article

Table 1. Pump Characteristic Curves and Cross-Flow large enough volume for the determination and concentration
Velocities of protein.
Then, as ΔPT was increased for a following trial, the
pump characteristic curves
procedure was repeated, until the entire working range for the
avg. flow avg. cross-flow coefficient of transmembrane pressure given for each process condition was
pump (L/min) velocity (m/s) equation determination
analyzed.
PA 0711 2.01 1.13 Q = −0.145Pi R2 = 0.9959 2.5. Membrane Cleaning. After the skim milk micro-
+ 2.121
filtration processes, the leftover milk was removed from the
PA 2511 5.31 2.90 Q = −0.080Pi R2 = 0.9923
+ 5.410 system using water rinsing at 50 °C. This was performed in 2
PA 411 7.85 4.29 Q = −0.275Pi R2 = 0.9956 cycles: (i) first 5 min with the permeate valve closed for rinsing
+ 8.296 the membrane surface and (ii) 5 min with the permeate valve
open for rinsing the pores of the membrane.
The permeate flux was determined by direct permeate mass After the rinsing procedure, cleaning was performed using
registration over time using an electronic scale (Shimadzu, BX 0.5% w/v Ultrasil11 at 50 °C and 1.3 bar inlet pressure in 2
4200H, Japan). Full recirculation mode was used in all cycles: (i) 20 min with the permeate valve closed and, (ii) 20
processes, that is, returning both the concentrate and the min with the permeate valve opened. The Ultrasil11 pH value
permeate streams to the feed tank, as shown in Figure 1. was 12.9. After cleaning, the membrane was rinsed several times
Actually, the permeate was continuously measured and it was using water at 50 °C until the water was clear and its
returned to the feed tank at discrete intervals. Therefore, all conductivity was lower than 10 μS/cm.
experiments were performed in quasi-stationary state because Using this methodology for each case, the membrane
the concentration of the feed stream was almost constant performance recovery (MPR) was higher than 90%;37 there-
during the process time. fore, the cleaning procedure was considered successful. Also,
2.3. Equipment Start-Up. At the beginning of each the new membrane was conditioned using a shorter Ultrasil11
process, the permeate pressure was increased and adjusted cleaning procedure, that is, using 10 min in each step instead 20
using compressed air. The skim milk began to circulate through min. Then, the curve J vs ΔPT was measured and was used as
the membrane with a transmembrane pressure equal to zero. starting point for cleaning assessment.
Then, the permeate pressure was slowly decreased to reach the 2.6. Protein Determination. The proteins were classified
needed value for the desired ΔPT for each experiment. This into two types: casein micelles and soluble proteins. For that
ensured that the fouling onto the membrane, in each process, reason, before protein quantification, it was necessary to
was in correspondence to the work transmembrane pressure. develop a protein sample preparation protocol to separate the
This way, the membrane was never exposed to higher proteins into two fractions and to eliminate the lactose
transmembrane pressure than set point, and therefore, it was interference.
prevented an undesired higher fouling and/or hysteresis For the sample preparation, the casein micelles and soluble
effect.36 Moreover, this type of start-up procedure avoids protein fractions were obtained from milk prepared with
rapid ΔPT increase across the membrane and the process can different concentration factors, and their respective permeates
starts with a subcritical ΔPT. were obtained by microfiltration. This was performed using the
2.4. Flux versus Transmembrane Pressure Curve following steps: first, acetic acid (1.2 M) was added to
Determination. For each process condition, for a given precipitate the casein micelles up to reach the isoelectric point
temperature, velocity and concentration factor, 2 L of skim milk (pH = 4.6), and the sample was centrifuged for 5 min at 5000
were processed. When starting, the lowest ΔPT in the working rpm. Second, the supernatant that contained the whey proteins
range was selected. The pressure working operation ranges for and lactose was subjected to precipitation with trichloroacetic
each pump and cross-flow velocity are shown in Table 2. acid (30% w/v). To reach the separation level desired, the
During all the experiments working in these ranges, no suspension was centrifuged at 15000 rpm for 30 min. Both
significant flux diminishing was found. precipitates were then resuspended in a Na2HPO4 buffer (500
mM) solution. The sample was appraised to a known volume
Table 2. Working Range for the Transmembrane Pressure for protein determination. All the centrifugation steps used a
microcentrifuge (Heraeus Sepatech, Biofuge 15 model, U.S.A.).
pump cross-flow velocity (m/s) transmembrane pressure (bar)
The protein samples were analyzed by the bicinchoninic acid
PA 0711 1.13 0.02−0.7 method using the Protein Research Reagents Kit (Pierce: BCA
PA 2511 2.90 0.1−1.5 Protein Assay Reagent, U.S.A.). The measurements were
PA 411 4.29 0.1−1.8 performed at 562 nm using a Plate Lector (ELx 800, BIO-
TEK Instrument Inc., USA).
The apparent retention coefficient (R) was determined for
The flux versus time curves were drawn until J∞ was the casein micelles and soluble proteins for microfiltration
reached, using the criteria described in eq 4: selectivity. The definition is presented in eq 5.
Jt + 1 − Jt
< 5% CP
Jt (4) R=1−
CF (5)
Each step lasted 30 min, and for all of the experiments, J∞ was
achieved within 30 min. The time required for the J calculation where C is the concentration, and the superindexes P and F are
was 2 min, and each protein determination sample was the permeate and feed streams, respectively. The same equation
obtained by collecting permeate fractions at 5 min intervals can be used for determination of protein retention coefficient,
for each transmembrane pressure. A 6 mL sample was taken, a for both soluble and casein micelle.
416 dx.doi.org/10.1021/ie5033292 | Ind. Eng. Chem. Res. 2015, 54, 414−425
Industrial & Engineering Chemistry Research Article

Table 3. Variables and Levels Used for Flux versus Transmembrane Pressure Curve Construction
level pseudocomponents value cross-flow velocity (v) concentration factor (CF) temperature (T)
equation applied v − 2.71 CF − 1.0 T − 50
vPC = CFPC = TPC =
(4.29 − 1.13)/2 (1.5 − 0.5)/2 (60 − 40)/2
low −1 1.13 m/s 0.5 40 °C
intermediate 0 2.90 m/s 1.0 50 °C
high 1 4.29 m/s 1.5 60 °C

2.7. Experimental Design. The variables analyzed on the dJ

flux versus transmembrane pressure curve behavior were the = α(JL − J )
dΔPT (6)
cross-flow velocity, the concentration factor and temperature.
The statistical analysis was performed using the transformation According to eq 6, when the transmembrane pressure is null,
of the original values to pseudocomponents. The levels used for there is no permeate flux (ΔPT = 0 → J = 0). Therefore, eq 7
each variable and the equations for the corresponding can be obtained as follows:
pseudocomponents are shown in Table 3. J dJ ΔPT
A factorial experimental design 33 with central points was ∫0 (JL − J )
= ∫0 α ·dΔPT
performed.39 Because three variables were used, three replicates (7)
of each central point were performed for error determination. Integrating this expression results in the following equation.
An experimental design with center points aids in the
investigation of the curvature in the region around the center. ⎛ J ⎞
These points also stabilize the estimation of the variance in and ln⎜⎜ L ⎟⎟ = α ·ΔPT
around the center. ⎝ JL − J⎠ (8)
The value of 2.90 m/s for the cross-flow velocity is not Then, when clearing the flux value, eq 9 is obtained:
exactly a center point. In fact, the corresponding pseudocom-
ponent value was 0.12, which means that the orthogonality J = JL (1 − exp(−α ·ΔPT)) (9)
design was not perfect. However, the largest variance inflation
factor (VIF) equals 1.056, a low enough value to consider the If eq 9 is derivate and the resulting expression is evaluated at
orthogonal design with no confounding effects.38 the origin (0,0), then the slope of a tangent straight line from
Moreover, an extra concentration factor (CF = 3.0) was the origin is obtained (JLα). Therefore, the tangent straight line
tested at the low and high levels of temperature and cross-flow to the curve in the origin is represented by the following
velocity, generating four additional experiments in addition to equation:
the original experimental design. This was performed to check J = JL α ·ΔPT (10)
whether the flux versus transmembrane pressure curves had a
good agreement with the exponential model at highly This can be interpreted as the behavior that a microfiltration
concentrated milk but was not considered for the experimental process should have if the phenomenon of fouling did not exist.
design. Rather, no particle accumulation takes place on the membrane,
2.8. Data Fitting. The data obtained during the skim milk and fouling does not build up, according to the theory of critical
microfiltration processes were adjusted to the exponential flux.16
pattern proposed in this work (i.e., eq 14) using the minimum When ΔPT → ∞, the limit for eq 9 can be calculated as
square methodology.39 From these results, two important follows:
parameters were determined during microfiltration, that is, the LIM J = LIM JL (1 − exp(−α ·ΔPT))
limiting flux and the critical transmembrane pressure under ΔPT →∞ ΔPT →∞ (11)
different conditions of the cross-flow velocity, concentration
factor, and temperature. Then, the asymptote equation can be written as follows:
2.9. Statistical Analysis. The effects of the cross-flow J = JL (12)
velocity, concentration factor and temperature on the limiting
flux and the critical transmembrane pressure were calculated Equation 12 is the curve obtained when the limiting flux, that is,
using ANOVA. Following this procedure, a regression model the maximum flux that is possible to obtain for a previously
(quadratic response surface) was determined. Table 4 shows established set of conditions, has already been reached. The
some parameters for the determination of goodness-of-fit, that same interpretation of the asymptote has been shown by Yeh.41
is, how well the data fit to the model. To find the value for the critical transmembrane pressure, eqs
10 and 12 were intersected. The intersection is shown in Figure
2 and the interception point (ΔPT*, J*) can be observed. Then,
3. THEORETICAL CALCULATION
if the (ΔPT)* is evaluated on eq 9, a new intersection point can
In this section, an equation was proposed for modeling the flux be found ((ΔPT)C, JC) which represents this critical value
versus transmembrane pressure curves and using it to where a transition exists among zones, that is, between the zone
determine operational parameters during microfiltration. To when “low fouling” exists on the membrane (J < JC, and J is
model this phenomenon, it is proposed that the variation in near to J = JC/(ΔPT)CΔPT) and another zone (J > JC).
permeate flux, as a factor of the transmembrane pressure Therefore, the value at the intersection point (J*) is derived
exerted, is proportional to the difference between the maximum from eqs 10 and 12.
permeate flux obtained experimentally (JL), and the flux
observed at a certain transmembrane pressure, that is, JL = JL α ·(ΔPT)* (13)

417 dx.doi.org/10.1021/ie5033292 | Ind. Eng. Chem. Res. 2015, 54, 414−425

 Table 4. Parameters for Goodness-of-Fitting Determination
param. meaning equation
residual (RES)39 It represents the numerical difference between the observed value and the predicted one. It can be interpreted
as a distance between the each experimental points to the corresponding estimated point in the model. RES = f (xi) − yi
where f(xi) is the estimated value, and yi the observed one.

mean absolute error It is a numerical measure of the model ability to predict the outcome. A zero value indicates a perfect fit and n
(MAE)40 this last one becomes progressively worse as long as the value increases (MAE > 0). 1
MAE = ∑ |f (xi) − yi |
n i=1

where n is the number of data points.

standard error of estimate It determines how well the model fits the experimental data. The smallest possible value is 0 with no upper n
(SEE)39 limit (SEE > 0). The lower is the SEE value, the more adequate is the model. 1
SEE = ∑ (f (xi) − yi )2
n−k−1 i=1

where n is the number of data points, and k is the number of independent variables in the
regression model.
Industrial & Engineering Chemistry Research

coefficient of determination It represents how well the data variability is explained by the model. All the model variables are there included.
⎛ n
(R2)39 An improved regression value could be obtained by increasing the number of coefficients, but this action may ∑ (f (xi) − yi )2 ⎞
R2 = ⎜⎜1 − i =n1 ⎟100
lead to an overfitting risk.39 In consequence, the coefficient of determination adjusted by degrees of fredom ⎝ ∑i = 1 (yi − yi )2 ⎟⎠
is better to describe the goodness-of-fit in this kind of models.
where y ̅ is the average number of observations, n is the number of data points, and k is the number
of independent variables included in the regression model.

418
coefficient of determination It is similar to the coefficient of determination but it excludes some effects, typically, the nonsignificant ones.
⎛ 1 n
adjusted by the degrees of Finally, 0 < Radj2 < R2 < 100 is always true. (∑i = 1 (f (xi) − yi )2 ) ⎞
2 n−k−1 ⎟100
fredom (Radj2)39 R adj = ⎜⎜1 − 1 n
⎝ n−k−h
∑i = 1 (y ̅ − yi )2 ⎟⎠

where y ̅ is the average number of observations, n is the number of data points, k is the number of
independent variables in the regression model, and h is the number of nonsignificant variables or
those ones that have been eliminated from the model.
Article

dx.doi.org/10.1021/ie5033292 | Ind. Eng. Chem. Res. 2015, 54, 414−425

Industrial & Engineering Chemistry Research Article

higher at 40 °C than 50 or 60 °C; moreover, for low

transmembrane pressure, lower rejection coefficients can be
achieved. For transmembrane pressure higher than 0.5 bar the
average retention coefficient, at 60 °C, was in the range from 30
to 40%. When the transmembrane pressure was lower than 0.5
bar, at 60 °C, the average retention coefficient was in the range
from 20 to 40%. Therefore, operation under low trans-
membrane pressure and high temperature leads to lower
soluble protein retention, which is appropriate and expected for
this kind of microfiltration.10,44−46
4.2. Flux versus Transmembrane Pressure Curves
Determination. During all the trials, the Reynolds number
inside the module was higher than 2100, and hence, a turbulent
regime was obtained. Figure 3 presents the flux versus

Figure 2. Intersection of eqs 10 and 12.

where (ΔPT)* is the transmembrane pressure at the

intersection point among curves, and rearranging, (ΔPT)* =
1/α.
Therefore, it was assumed that the latter value represents a
critical transmembrane pressure value (ΔPT)C. For trans-
membrane pressure values under (ΔPT)C, the flux is propor-
tional to the transmembrane pressure applied. This condition is
known as the subcritical region.16
Then, the exponential model for all transmembrane pressure
range is expressed by eq 14. Exponential models such as eq 14
have been previously used in membrane technology for Figure 3. Effect of temperature on the flux versus transmembrane
describing the flux decline along time, such as J = J0·exp(−at), pressure curves obtained during skim milk microfiltration using
considered among a family of exponential−decay curves.42 ceramic membranes of 0.14 μm and 3.6 mm of hydraulic diameter. (v
= 2.9 m/s and CF = 1.0)
⎛ ⎛ ΔPT ⎞⎞
J = JL ⎜⎜1 − exp⎜ − ⎟⎟⎟
⎝ ⎝ (ΔPT)C ⎠⎠ (14) transmembrane pressure curves at different temperatures. It
shows that when the temperature was increased, the maximum
Additionally, by evaluating eq 14 at ΔPT = (ΔPT)C, eq 15 is value for the flux increased as well. Moreover, a good fit of the
obtained, and it directly relates the critical flux value with the experimental curves to the exponential model in all the trials
limiting flux. (R2 ≥ 98.56%). The limiting flux was achieved at pressures
J[(ΔPT)C ] = JC = 0.632JL higher than 0.64 bar.
(15)
To clarify how to compute the limiting transmembrane
This analytical result for the critical flux value is close to that pressure, the following example is shown. Using the data
obtained by Bacchin.43 The author related the critical flux with obtained for the curve shown in Figure 3 at 40 °C and fitting it
a thickness of the critical boundary layer and also established to eq 14, eq 17 is obtained:
that the critical flux is equal to 2/3 of the limiting flux value.
⎛ ⎛ ΔP ⎞⎞
For the calculation of the limiting transmembrane pressure J = 45.80⎜1 − exp⎜ − T ⎟⎟
(ΔPT)L, it was considered that this value was reached when the ⎝ ⎝ 0.21 ⎠⎠
flux value was at least 95% of the limiting flux value. This value
agrees with the experimental data. For this determination, eq 15 R2 = 99.30% (17)
was used, obtaining eq 16. 2
In this case, JL is 45.80 L/m /h, and (ΔPT)C is 0.21 bar. Then,
(ΔPT)L = −(ΔPT)C ·ln(1 − 0.95) ≈ 3(ΔPT)C (16) using eq 15, the critical flux can be computed as JC = 28.94 L/
m2/h, and using eq 16, the limiting transmembrane pressure
can be calculated as (ΔPT)L = 0.64 bar. Therefore, the critical
4. RESULTS AND DISCUSSION and the limiting point were [28.94 L/m2/h; 0.21 bar] and
4.1. Membrane Selectivity. When using membrane of [45.80 L/m2/h; 0.64 bar], respectively.
0.14 um cutoff, a high casein micelle retention and low Figure 4 shows the effect of the cross-flow velocity on the
retention for soluble proteins are expected during skim milk flux versus transmembrane pressure curves. These curves also
microfiltration.10,44−46 In all experiments, the casein micelle had a good fit with the exponential model (R2 ≥ 99.42%). As
retention coefficient was higher than 99.99% regardless of expected, when the cross-flow velocity was increased, an
cross-flow velocity, concentration factor, temperature, and increase in the limiting flux was observed.14,47 Another effect of
transmembrane pressure. Even for low transmembrane this parameter on the curves can be observed: when the cross-
pressure, no significant casein amounts were detected in the flow velocity is increased, the limiting transmembrane pressure
permeate stream. Additionally, the soluble protein rejection was increased significantly from 0.08 bar for 1.13 m/s to 1.05 bar
419 dx.doi.org/10.1021/ie5033292 | Ind. Eng. Chem. Res. 2015, 54, 414−425
Industrial & Engineering Chemistry Research Article

Figure 6 shows the experimental limiting flux (JL) and the

flux calculated (JL̂ ) using eq 14 for each experiment. A simple

Figure 4. Effect of the cross-flow velocity on the flux versus

transmembrane pressure curves obtained during skim milk micro-
filtration using ceramic membranes of 0.14 μm and 3.6 mm of Figure 6. Correlation between the experimental and predicted limiting
hydraulic diameter. (T = 50 °C and CF = 1.0) flux results obtained during skim milk microfiltration using ceramic
membranes of 0.14 μm and 3.6 mm of hydraulic diameter. Data for all
for 4.29 m/s. That means that when a low cross-flow velocity is concentrations, temperatures, and cross-flow velocities used in this
research (n = 34).
used for skim milk microfiltration, a low transmembrane
pressure is needed to achieve a balance between the convection
and erosion within the membrane−solution interface.48 linear regression model was used for testing the relationship
Figure 5 shows the effect of the concentration factor on the between the results. The corresponding correlation is presented
flux versus transmembrane pressure curves for skim milk in eq 18:
JL̂ = 1.0064JL (18)
2
For this case, the correlation coefficient (R ) was higher than
99.88%. That means that the variables are highly correlated. An
ANOVA test was performed using the data for JL and JL̂ . Before
that, normality, independence and homoscedasticity were
corroborated.38 The results showed that there were no
significant differences between the populations of the JL and
JL̂ data at a 95% confidence level. Both data populations showed
the same mean value (p-value > 0.05) and the same standard
deviation (p-value > 0.05). Additionally, the equality for both
distributions was checked using the Kolmogorov−Smirnov test
(p-value > 0.05). It was concluded that the experimental value
is equal to the value obtained by fitting the data using eq 14,
that is, JL=JL̂ .
Figure 5. Effect of the concentration factor on the flux versus Further, a residual analysis was performed for a goodness of
transmembrane pressure curves obtained during skim milk micro- fit evaluation. The hypothesis, which indicates that residues
filtration using ceramic membranes of 0.14 μm and 3.6 mm of come from a normal distribution, was accepted at the 95%
hydraulic diameter. (T = 50 °C and v = 2.9 m/s) confidence level (p-value > 0.05) using the Kolmogorov−
Smirnov test. Therefore, eq 14 is valid for limiting flux
microfiltration. In this case, a good fit between the experimental determination.
data and the exponential model was observed (R2 ≥ 99.42%). 4.4. Effect of the Cross-Flow Velocity, Concentration
As expected, the concentration factor had an impact on JL as Factor, and Temperature on Limiting Flux. For the
previously reported.49,50 When the diluted skim milk was determination of the effects of those variables on limit flux, a
processed, a higher limiting flux was achieved. A similar multiple regression model39 for three factors was fitted. The
observation was noted by Carić et al.,51 who found that the general model is showed in eq 19. In this case x1 is the cross-
limiting flux decreases when the protein concentration flow velocity, x2 is the concentration factor, and x3 is the
increases. temperature.
Related to the limiting transmembrane pressure, the lowest
value of 0.43 bar was observed for the concentrated milk (CF = f (x1, , x 2, , x3) = β0 + β1x1 + β2x 2 + β3x3 + β4 x12 + β5x 2 2
1.5), and the highest value of 0.67 bar was observed for the
+ β6x32 + β7x1x 2 + β8x1x3 + β9x 2x3 + ε
diluted milk (CF = 0.5).
4.3. Comparison between the Limiting Flux Exper- (19)
imental Value and the Value Determined through the where β0 is the constant coefficient; β1, β2, and β3 are the
Exponential Model. For each trial, the limiting flux was coefficients of the linear effects; β4, β5, and β6 are the
determined in two ways: first, as the maximum flux value that coefficients of the quadratic effects; β7, β8, and β9 are the
can be reached for one experimental condition set and, second, coefficients of the interaction effects, and ε is the model error.
by fitting eq 14 to all the experimental data collected for the During the experimental design analysis, all the studied
same experimental condition and deriving JL. variables, their quadratic effects, and their interactions were
420 dx.doi.org/10.1021/ie5033292 | Ind. Eng. Chem. Res. 2015, 54, 414−425
Industrial & Engineering Chemistry Research Article

considered. The analysis presented a good fit, obtaining a Pressure. Performing a similar analysis to the one performed
coefficient of determination corrected by the degrees of with the limiting flux, eq 21 shows the results for the effect of
freedom (Radj2) of 98.18%. Moreover, the analysis revealed the cross-flow velocity, concentration factor and temperature
that all the studied variables and their interactions were on the critical transmembrane pressure. Again, during the
significant (p-value < 0.05), except the temperature and cross- analysis of all the studied variables in the experimental design,
flow velocity quadratic effects (p-value > 0.05). The temper- their quadratic effects and interactions were considered. The
ature and cross-flow velocity quadratic effects were excluded to analysis presented a good fit, obtaining a coefficient of
obtain a more precise model. determination corrected by the degrees of freedom of
The eq 20 represents the quadratic model fitted for the 97.83%. Moreover, the analysis revealed that all the studied
permeate limiting flux in L/m2·h as a function of the variables had significant effects (p-value < 0.05), but in this
concentration factor, cross-flow velocity and temperature. case, only the cross-flow velocity quadratic effect and its
This is for variables as pseudocomponents inside the working interaction with the concentration factor and temperature were
range (0.5 ≤ CF ≤ 1.5; 1.13 m/s ≤ v ≤ 4.29 m/s, and 40 °C ≤ significant (p-value > 0.05). For that reason, the temperature
T ≤ 60 °C) and excluding the quadratic effects previously and concentration factor quadratic effects, and their inter-
mentioned. actions, were excluded to obtain a more precise model.
Equation 21 represents the quadratic model fitted for the
JL = 50.97 − 22.76CFpc + 41.37vpc + 9.86Tpc critical transmembrane pressure in bar as a function of the
concentration factor, cross-flow velocity, and temperature. This
+ 11.44CFpc 2 − 17.27CFpcvpc − 3.98CFpcTpc
was for variables as pseudocomponents inside the working
+ 7.84vpcTpc range (0.5 ≤ CF ≤ 1.5; 1.13 m/s ≤ v ≤ 4.29 m/s; and 40 °C ≤
(20)
T ≤ 60 °C), excluding the effects previously mentioned.
From eq 20, CFpc, vpc, and Tpc are the variables as
pseudocomponents (−1 ≤ CFpc ≤ 1; −1 ≤ vpc ≤ 1; and −1 (ΔPT)C = 0.1597 − 0.0316CFpc + 0.1650vpc − 0.0183Tpc
≤ Tpc ≤ 1). During the adjustment, it was verified that Radj2 =
98.12%. The standard error of estimate was 5.61, and the mean − 0.0438CFpcvpc + 0.0319vpc 2 − 0.0152vpcTpc
absolute error was 3.98. Both values were considered adequate. (21)
The analysis confirmed that the temperature, the concen-
tration factor and the cross-flow velocity effects were significant In eq 21, CF pc , v pc , and T pc are the variables as
(p-value < 0.05). Moreover, all the interactions and the pseudocomponents (−1 ≤ CFpc ≤ 1; −1 ≤ vpc ≤ 1; and −1
concentration factor quadratic effects were significant (p-value ≤ Tpc ≤ 1). During the adjustment, the Radj2 was improved to
< 0.05). In this case, a residual analysis was performed for a 98.08%. The standard error of estimate was 0.02, and the mean
goodness of fit evaluation. Using the Kolmogorov−Smirnov absolute error was 0.01. Both values were considered
test, the hypothesis, which indicates that residues come from a satisfactory.
normal distribution, was accepted at the 95% confidence level The analysis confirmed that the temperature, the concen-
(p-value > 0.05). Consequently, this model can be used for tration factor and the cross-flow velocity had significant effects
limiting flux determination during skim milk microfiltration (p-value < 0.05). Moreover, all the interactions and the
under different conditions for cross-flow velocity, concentration quadratic effect of the cross-flow velocity were significant (p-
factor, and temperature; and as such, it is useful for limiting flux value < 0.05).
predicting on various conditions.52 For the critical transmembrane pressure, it is clear that the
By observing eq 20 and considering the working range for most significant effect is the cross-flow velocity, as can be
each variable, the effect of the velocity seems to be the most observed in eq 21. Thus, if the velocity inside the module
significant. This is in agreement with the fact that an increase in changes, the operational conditions will also change.
the cross-flow velocity is the simplest way of creating In general, the limiting flux has been described by several
turbulence, changing hydrodynamics conditions,53,54 and, authors18−20,58 who related this concept (or its value) to the
therefore, reducing membrane fouling.55 Its immediate effect system hydrodynamic and fluid properties.20,25 For its
is the limiting flux improving, even at low cross-flow prediction, several back-transport models have been use-
velocities. 56 In general, this flux can be modified by d59−66and several authors have studied the critical flux since
manipulating the hydrodynamic conditions and the system the concept was introduced for the first time by Field et al.16
physical−chemical properties.25 For example, Samuelsson et The critical flux is dependent on the system hydrodynam-
al.20 determined that it was a linear Reynolds’ number function. ics,21,23,30 as was previously mentioned, and is a function of
Similar observations have been made for the critical flux, other variables, such as solids concentration50 and particle
which increases with increasing cross-flow velocity.21,30 Addi- size.30,50
tionally, it is considered a function of system hydrodynamic23,30 During the skim milk microfiltration using the membrane of
and a Reynold’s number function.23,50 This agrees with the fact 0.14 μm cutoff and 6 mm hydraulic diameter, a positive effect of
that eq 15 relates JC and JL. the temperature and velocity increase were observed on the
Other investigations showed that JC increases with an permeate limiting flux. Likewise, the velocity has a positive
increase of surface repulsion and membrane-colloid inter- effect on the limiting and critical transmembrane pressure, but
actions.21,30,57 The higher the potential energy required for the the temperature has a negative effect on those parameters. The
particles interaction is,57 the lower the possibility that the concentration effect is also important. When the milk solids
particles coagulate or are deposited onto the membrane surface. concentration increased, a limiting flux decrease was observed,
Therefore, the critical flux is related to the particle stability. as well as the transmembrane pressure that is necessary to
4.5. Effect of the Cross-Flow Velocity, Concentration obtain it. The critical flux and transmembrane critical pressure
Factor and Temperature on the Critical Transmembrane diminished in the same way.
421 dx.doi.org/10.1021/ie5033292 | Ind. Eng. Chem. Res. 2015, 54, 414−425
Industrial & Engineering Chemistry Research Article

Some authors have observed these effects on the critical or

limiting fluxes working with skim milk. During microfiltration
using ceramic membranes (0.14 μm cutoff, 6 mm hydraulic
diameter at 55 °C), an increase in the limiting flux was
observed when the cross-flow velocity was increased.20,46 A
positive effect of the cross-flow velocity over the critical flux was
observed by other authors as well.23,30 Yourvarong et al.36 and
Baruah et al.67 noted the same effect of the concentration factor
that has been verified in this work; that is, the critical flux
increases when the protein concentration decreases. Yourvar-
ong et al.36 also proposed a relationship stating that critical flux
is proportional to the shear stress and inversely proportional to
the concentration. Figure 7. Relationship between J and ΔPT during microfiltration. The
Few studies have been published relating the values of the three fouling zones are shown, in addition to the critical flux (JC) and
critical and limiting transmembrane pressures, and fewer are limiting flux (JL).
available concerning the effect of the cross-flow velocity,
temperature, and concentration factor on the limiting and In relation to the methodology used for the determination of
critical transmembrane pressures during skim milk micro- the critical transmembrane pressure, the literature shows
filtration. Some data can be obtained by the inspection of another graphical method for the determination of the critical
figures already published in the literature,22,25,68−70 from which flux that was proposed by Bacchin et al.71 In that method, they
further discussion can be made. In a study conducted by Gésan- drew the curve J vs JW for the same transmembrane pressure,
Guiziou et al.,68 when using a multitubular membrane of 0.1 determined as JC, the value where J = JW/2. The term JW is the
μm at 50 °C, it was observed that the critical transmembrane flux obtained using pure water at the same temperature and
pressure was lower than 0.5 bar. In another work with the same cross-flow velocity. They could have used the same concept by
cutoff and temperature,69 the critical and limiting trans- graphing the J vs ΔPT curves and JW/2 vs ΔPT. The graphic
membrane pressures were lower than 0.2 bar; however, the method by Bacchin et al.71 will always give higher values for
cross-flow velocity at which the experiments were performed (ΔPT)C than those reported in this research. That is because
was not reported. during all the trials conducted in this work, the critical flux was
Piry et al.70 performed experiments for module length found to be of the weak type. 17 Even at very low
evaluation at 6 m/s and 55 °C using a 0.1 μm zirconia transmembrane pressures, the permeate flux was always smaller
membrane. Their results, as determined by inspecting their than the water flux for the same operating conditions.
curves, showed that the limiting transmembrane pressure was Consequently, in this research, the critical flux values were
lower than 0.4 bar. smaller because the curves of J vs ΔPT are strictly increasing.
Similar results can be observed in the work of Grandison et Moreover, using the Bacchin et al.71 methodology, the values
al.22 The critical transmembrane pressure was 0.5 bar or lower, for the critical flux could be equal or larger than the limiting
and the limiting pressure was under 1 bar at 3.4 m/s and 50 °C flux, leading to possible misunderstanding.
Finally, some advantages and disadvantages for the use of eq
using a tubular membrane (200 kDa, PVDF). In the work of
14 can be pointed out. The model’s main advantage is that it
Yourvarong et al.,25 the critical transmembrane pressure was
allows the efficient determination of four of the most important
lower than 1 bar (200 kDa, PVDF, 50 °C, and 3.4 m/s).
parameters in membrane module operation. These parameters
Moreover, Yourvarong et al.36 showed that using a PVDF
are the limiting flux, critical flux, limiting transmembrane
membrane of 200 kDa at 50 °C and 3.4 m/s, the critical pressure and critical transmembrane pressure. Additionally, the
transmembrane pressure was 0.24 bar. However, none of the model offers an excellent adjustment for experimental data, and
above-mentioned works discussed the effect of the cross-flow the JL determination is estimated using all of the data (all the
velocity, temperature, or concentration factor on the critical or transmembrane pressures for a condition set and not just the
limiting transmembrane pressure. maximum value). This results in a lower error in the
In summary, under a set of experimental conditions, by data determination of JL.
adjustment to eq 14, the limiting flux (JL) and critical The only model disadvantage is that it does not allow
transmembrane pressure ((ΔPT)C) were estimated. Then, predicting when cake compaction takes place nor does it
using eqs 15 and 16, respectively, the critical flux (JC) and explain how the flux behaves in that condition. Fortunately,
limiting transmembrane pressure ((ΔPT)L) can be calculated. cake compaction is not interesting in productive terms because
In this way, two important points in the flux vs transmembrane it is the worst operational condition, that is, the flux decreases
pressure curves are established: [JC, (ΔPT)C] and [JL, (ΔPT)L]. and modifies the selectivity, retaining more whey proteins
If eq 14 and these previously mentioned points are graphed, during skim milk microfiltration. Operating a microfiltration
Figure 7 is obtained. module within this transmembrane pressure range would lead
Figure 7 depicts the three operational zones for the to more severe membrane fouling, and hence, the use of higher
transmembrane pressure using the analysis of eq 14. Despite concentrations of cleaning agents, and more frequent
the fact that the real value of the limiting transmembrane restoration cycles, leading to reduced system productivity.
pressure is higher than that estimated by using eq 16, operating In summary, eq 14 is useful as a model for estimating JL and
the process under the latter might prevent, or at least diminish, (ΔPT)C, and eq 20 and 21, allow JL and ΔPT predicting on
long-term reversible fouling. This type of fouling causes a slow various conditions of temperature, concentration factor and
flux decrease over time, typically during hours of operation cross-flow velocity, in the studied ranges, during skim milk
before the process has to be stopped for cleaning.3 microfiltration using ceramic membranes of 0.14 μm and 3.6
422 dx.doi.org/10.1021/ie5033292 | Ind. Eng. Chem. Res. 2015, 54, 414−425
Industrial & Engineering Chemistry Research Article

mm of hydraulic diameter. Looking further, this equation may RT = total membrane resistance (m−1)
allow automatic control system to adjust skim milk micro- RRF = reversible fouling resistance (m−1)
filtration parameters to operate at critical or limiting point in a R2 = coefficient of determination (dimensionless)
concentration process. Radj2 = coefficient of determination corrected by degrees of
freedom (dimensionless)
5. CONCLUSIONS SEE = Standard error of estimate (dimensionless)
The use of the equation J = JL·(1 − exp(−PT/(ΔPT)C) works μ = permeate viscosity (Pa·s)
efficiently for a direct determination of the limiting flux and t = time (min)
critical transmembrane pressure from the experimental data
using the complete J versus ΔPT curve during skim milk
microfiltration. Moreover, using simple mathematical relations,
■ REFERENCES
(1) Daufin, G.; Merin, U. Cleaning of Inorganic Membranes after
the critical flux and limiting transmembrane pressure can also Whey and Milk Ultrafiltration. Biotechnol. Bioeng. 1991, 38, 82.
be determined, and therefore, the operational zones can be (2) Ishigami, T.; Fuse, H.; Asao, S.; Saeki, D.; Ohmukai, Y.; Kamio,
described. E.; Matsuyama, H. Permeation of Dispersed Particles through a Pore
Additionally, in this work, the critical and limiting trans- and Transmembrane Pressure Behavior in Dead-End Constant-Flux
membrane pressure values were reported, as well as their Microfiltration by Two-Dimensional Direct Numerical Simulation. Ind.
correlation with variables such as the concentration factor, Eng. Chem. Res. 2013, 52, 4650.
cross-flow velocity and temperature for the case of skim milk (3) Brans, G.; Schroën, C. G. P. H.; van der Sman, R. G. M.; Boom,
microfiltration. Using this information, an appropriate range for R. M. Membrane Fractionation of Milk: State of the Art and
the working transmembrane pressure can be estimated to Challenges. J. Membr. Sci. 2004, 243, 263.
prevent any undesirable membrane fouling. There are strong (4) Goudédranche, H.; Fauquant, J.; Maubois, J. Fractionation of
grounds to believe that this methodology could decrease Globular Milk Fat by Membrane Microfiltration. Lait 2000, 80, 93.
membrane fouling and optimize membrane performance. (5) Henning, D. R.; Baer, R. J.; Hassan, A. N.; Dave, R. Major

■
Advances in Concentrated and Dry Milk Products, Cheese, and Milk
Fat-Based Spreads. J. Dairy Sci. 2006, 89, 1179.
AUTHOR INFORMATION (6) Skrzypek, M.; Burger, M. Isoflux Ceramic Membranes-Practical
Corresponding Author Experiences in Dairy Industry. Desalination 2010, 250, 1095.
*E-mail: [email protected]. Tel: +56-322274218. Fax: (7) Beolchini, F.; Vegho, F.; Barba, D. Microfiltration of Bovine and
+56-322274205. Ovine Milk for the Reduction of Microbial Content in a Tubular
Notes Membrane: A Preliminary Investigation. Desalination 2004, 161, 251.
The authors declare no competing financial interest. (8) Guerra, A.; Jonsson, G.; Rasmussen, A.; Nielsena, E. W.;

■
Edelsten, D. Low Cross-Flow Velocity Microfiltration of Skim Milk for
Removal of Bacterial Spores. Int. Dairy J. 1997, 7, 849.
ACKNOWLEDGMENTS (9) Hoffmann, W.; Kiesner, C.; Clawin-Rädecker, I.; Martin, D.;
The author would like to acknowledge CONICYT, Eihoff, K.; Lorenzen, P.; Meisel, H.; Hammer, P.; Suhren, G.; P, T.
FONDECYT/Iniciación 11110402 Project, for the financial Processing of Extended Shelf Life Milk Using Microfiltration. Int. J.
support provided. The author also wishes to thank to Professor Dairy Technol. 2006, 59, 229.
Jacqueline Reveco, Professor Jorge Saavedra and Professor Ariel (10) Lawrence, N. D.; Kentish, S. E.; O’Connor, a. J.; Barber, a. R.;
Ochoa for the valuable help in the writing of the manuscript. Stevens, G. W. Microfiltration of Skim Milk Using Polymeric

■
Membranes for Casein Concentrate Manufacture. Sep. Purif. Technol.
ABBREVIATIONS 2008, 60, 237.
(11) Pafylias, I.; Cheryan, M.; Mehaia, M.; Saglam, N. Microfiltration
a = constant (min−1) of Milk with Ceramic Membranes. Food Res. Int. 1996, 29, 141.
CP = concentration in permeate stream (g/L) (12) Tan, T. J.; Wang, D.; Moraru, C. I. A Physicochemical
CF = concentration in feed stream (g/L) Investigation of Membrane Fouling in Cold Microfiltration of Skim
D[3,2] = particle mean diameter (μm) Milk. J. Dairy Sci. 2014, 97, 4759.
J = flux (L/m2/h) (13) Barukčić, I.; Božanić, R.; Kulozik, U. Effect of Pore Size and
JL = limiting flux (L/m2/h) Process Temperature on Flux, Microbial Reduction and Fouling
JC = critical flux (L/m2/h) Mechanisms during Sweet Whey Cross-Flow Microfiltration by
Jt = flux at time t Ceramic Membranes. Int. Dairy J. 2014, 39, 8.
Jt+1 = flux at time t+1 (next time) (14) Cheryan, M. Ultrafiltration and Microfiltration Handbook; CRC
JW = flux of water (L/m2/h) Press: Boca Raton, FL, 1998; pp 1−615.
J0 = initial flux (L/m2/h) (15) Brans, G.; van der Sman, R. G. M.; Schroën, C. G. P. H.; van der
MAE = mean absolute error (dimensionless) Padt, A.; Boom, R. M. Optimization of the Membrane and Pore
MPR = membrane performance recovery (dimensionless) Design for Micro-Machined Membranes. J. Membr. Sci. 2006, 278, 239.
Q = flow (L/min) (16) Field, R. W.; Wu, D.; Howell, J. a.; Gupta, B. B. Critical Flux
Concept for Microfiltration Fouling. J. Membr. Sci. 1995, 100, 259.
Pi = inlet pressure (bar)
(17) Bacchin, P.; Aimar, P.; Field, R. Critical and Sustainable Fluxes:
ΔPT = transmembrane pressure (bar) Theory, Experiments, and Applications. J. Membr. Sci. 2006, 281, 42.
(ΔPT)C = critical transmembrane pressure (bar) (18) Baruah, G. L.; Belfort, G. A Predictive Aggregate Transport
(ΔPT)L = limiting transmembrane pressure (bar) Model for Microfiltration of Combined Macromolecular Solutions and
R = apparent retention coefficient (dimensionless) Poly-Disperse Suspensions: Model Development. Biotechnol. Prog.
RES = Residual (dimensionless) 2003, 19, 1524.
RF = fouled membrane resistance (m−1) (19) Belfort, G.; Davis, R.; Zydney, A. The Behavior of Suspensions
RM = hydraulic membrane resistance (m−1) and Macromolecular Solutions in Crossflow Microfiltration. J. Membr.
RIF = irreversible fouling resistance (m−1) Sci. 1994, 96, 1.

423 dx.doi.org/10.1021/ie5033292 | Ind. Eng. Chem. Res. 2015, 54, 414−425

Industrial & Engineering Chemistry Research Article

(20) Samuelsson, G.; Huisman, I.; Trägårdh, G.; Paulsson, M. (42) Mallubhotla, H.; Belfort, G. Semiempirical Modeling of Cross-
Predicting Limiting Flux of Skim Milk in Crossflow Microfiltration. J. Flow Microfiltration with Periodic Reverse Filtration. Ind. Eng. Chem.
Membr. Sci. 1997, 129, 277. Res. 1996, 35, 2920.
(21) Bacchin, P.; Aimar, P. Critical Fouling Conditions Induced by (43) Bacchin, P. A Possible Link between Critical and Limiting Flux
Colloidal Surface Interaction: From Causes to Consequences. for Colloidal Systems: Consideration of Critical Deposit Formation
Desalination 2005, 175, 21. along a Membrane. J. Membr. Sci. 2004, 228, 237.
(22) Grandison, A.; Youravong, W.; Lewis, M. Hydrodynamic (44) Hurt, E.; Zulewska, J.; Newbold, M.; Barbano, D. M. Micellar
Factors Affecting Flux and Fouling during Ultrafiltration of Skimmed Casein Concentrate Production with a 3×, 3-Stage, Uniform
Milk. Lait 2000, 80, 165. Transmembrane Pressure Ceramic Membrane Process at 50 °C. J.
(23) Espinasse, B.; Bacchin, P.; Aimar, P. On an Experimental Dairy Sci. 2010, 93, 5588.
Method to Measure Critical Flux in Ultrafiltration. Desalination 2002, (45) Lawrence, N.; Kentish, S.; O’Connor, A.; Stevens, G.; Barber, A.
146, 91. Microfiltration of Skim Milk for Casein Concentrate Manufacture.
(24) Pollice, A.; Brookes, A.; Jefferson, B.; Judd, S. Subcritical Flux Desalination 2006, 200, 305.
Fouling in Membrane BioreactorsA Review of Recent Literature. (46) Samuelsson, G.; Dejmek, P.; Trägårdh, G.; Paulsson, M.
Desalination 2005, 174, 221.
Minimizing Whey Protein Retention in Cross-Flow Microfiltration of
(25) Youravong, W.; Grandison, A. S.; Lewis, M. J. The Effects of
Skim Milk. Int. Dairy J. 1997, 7, 237.
Physicochemical Changes on Critical Flux of Skimmed Milk
(47) Nordin, A.-K.; Jönsson, A.-S. Flux Decline along the Flow
Ultrafiltration. Membr. Sci. Technol. 2002, 24, 929.
Channel in Tubular Ultrafiltration Modules. Chem. Eng. Res. Des. 2009,
(26) Guo, W.; Vigneswaran, S.; Ngo, H. H.; Ben Aim, R. Evaluating
the Efficiency of Pretreatment to Microfiltration: Using Critical Flux as 87, 1551.
a Performance Indicator. In 5th International Membrane Science and (48) Le Berre, O.; Daufin, G. Skimmilk Crossflow Microfiltration
Technology Conference, Sydney, Australia; 2003. Performance versus Permeation Flux to Wall Shear Stress Ratio. J.
(27) Bessiere, Y.; Abidine, N.; Bacchin, P. Low Fouling Conditions in Membr. Sci. 1996, 117, 261.
Dead-End Filtration: Evidence for a Critical Filtered Volume and (49) Jö nsson, A.-S.; Jö nsson, B. Ultrafiltration of Colloidal
Interpretation Using Critical Osmotic Pressure. J. Membr. Sci. 2005, DispersionsA Theoretical Model of the Concentration Polarization
264, 37. Phenomena. J. Colloid Interface Sci. 1996, 180, 504.
(28) Stoller, M.; Chianese, A. Influence of the Adopted Pretreatment (50) Wu, D.; Howell, J.; Field, R. Critical Flux Measurement for
Process on the Critical Flux Value of Batch Membrane Processes. Ind. Model Colloids. J. Membr. Sci. 1999, 152, 89.
Eng. Chem. Res. 2007, 46, 2249. (51) Carić, M.; Milanović, S.; Krstić, D.; Tekić, M. Fouling of
(29) Espinasse, B.; Bacchin, P.; Aimar, P. Filtration Method Inorganic Membranes by Adsorption of Whey Proteins. J. Membr. Sci.
Characterizing the Reversibility of Colloidal Fouling Layers at a 2000, 165, 83.
Membrane Surface: Analysis through Critical Flux and Osmotic (52) Vincent-Vela, M.-C.; Cuartas-Uribe, B.; Á lvarez-Blanco, S.;
Pressure. J. Colloid Interface Sci. 2008, 320, 483. Lora-García, J. Analysis of an Ultrafiltration Model: Influence of
(30) Howell, J. Subcritical Flux Operation of Microfiltration. J. Operational Conditions. Desalination 2012, 284, 14.
Membr. Sci. 1995, 107, 165. (53) Zhang, W.; Zhu, Z.; Jaffrin, M.; Ding, L. Hydraulic Conditions
(31) Guglielmi, G.; Saroj, D. P.; Chiarani, D.; Andreottola, G. on Effluent Quality, Flux Behavior, and Energy Consumption in a
Subcritical Fouling in Membrane Bioreactor for Municipal Waterwaste Shear-Enhanced Membrane Filtration Using Box−Behnken Response
Treatment: Experimental Investigation and Mathematical Modeling. Surface. Ind. Eng. Chem. Res. 2014, 53, 7176.
Water Res. 2007, 41, 3093. (54) Alexiadis, A.; Bao, J.; Fletcher, D. F.; Wiley, D. E.; Clements, D.
(32) Schutyser, M.; Belfort, G. Dean Vortex Membrane Micro- J. Analysis of the Dynamic Response of a Reverse Osmosis Membrane
filtration Non-Newtonian Viscosity Effects. Ind. Eng. Chem. Res. 2002, to Time-Dependent Transmembrane Pressure Variation. Ind. Eng.
25, 494. Chem. Res. 2005, 44, 7823.
(33) Ho, C.; Zydney, A. L. Protein Fouling of Asymmetric and (55) Krstić, D. M.; Koris, A. K.; Tekić, M. N. Do Static Turbulence
Composite Microfiltration Membranes. Ind. Eng. Chem. Res. 2001, 40, Promoters Have Potential in Cross-Flow Membrane Filtration
1412. Applications? Desalination 2006, 191, 371.
(34) Bouzid, H.; Rabiller-Baudry, M.; Paugam, L.; Rousseau, F.; (56) Gebreyohannes, A. Y.; Mazzei, R.; Curcio, E.; Poerio, T.; Drioli,
Derriche, Z.; Bettahar, N. E. Impact of Zeta Potential and Size of E.; Giorno, L. Study on the In Situ Enzymatic Self-Cleansing of
Caseins as Precursors of Fouling Deposit on Limiting and Critical Microfiltration Membrane for Valorization of Olive Mill Wastewater.
Fluxes in Spiral Ultrafiltration of Modified Skim Milks. J. Membr. Sci.
Ind. Eng. Chem. Res. 2013, 52, 10396.
2008, 314, 67.
(57) Bacchin, P.; Aimar, P.; Sanchez, V. Model for Colloidal Fouling
(35) Gésan, G.; Daufin, G.; Merin, U.; Labbé, J.; Quemerais, A.
of Membranes. AIChE J. 1995, 41, 368.
Fouling during Constant Flux Crossflow Microfiltration of Pretreated
(58) Lipnizki, J.; Casani, S.; Jonsson, G. Optimization of Ultra-
Whey. Influence of Transmembrane Pressure Gradient. J. Membr. Sci.
filtration of a Highly Viscous Protein Solution Using Spiral-Wound
1993, 80, 131.
(36) Youravong, W.; Lewis, M.; Grandison, A. Critical Flux in Modules. Desalination 2005, 180, 15.
Ultrafiltration of Skimmed Milk. Food Bioprod. Process. 2003, 81, 3. (59) Chang, D.-J.; Hsu, F.-C.; Hwang, S.-J. Steady-State Permeate
(37) Astudillo, C.; Parra, J.; González, S.; Cancino, B. A New Flux of Cross-Flow Microfiltration. J. Membr. Sci. 1995, 98, 97.
Parameter for Membrane Cleaning Evaluation. Sep. Purif. Technol. (60) Huisman, I.; Trägårdh, G.; Trägårdh, C. Particle Transport in
2010, 73, 286. Crossflow Microfiltration II. Effects of Particle−Particle Interactions.
(38) Ryan, T. P. Modern Experimental Design; John Wiley & Sons, I., Chem. Eng. Sci. 1999, 54, 281.
Ed.; Wiley: NJ, 2007; p 601 pp. (61) Huisman, I.; Trägårdh, C. Particle Transport in Crossflow
(39) Montgomery, D. Analysis and Experimental Design, 5th ed.; John Microfiltration I. Effects of Hydrodynamics and Diffusion. Chem. Eng.
Wiley & Sons: Hoboken, NJ, 2001; p 699. Sci. 1999, 54, 271.
(40) Willmott, C. J.; Matsuura, K. Advantages of the Mean Absolute (62) Hwang, K.-J.; Chou, F.-Y.; Tung, K.-L. Effects of Operating
Error (MAE) over the Root Mean Square Error (RMSE) in Assessing Conditions on the Performance of Cross-Flow Microfiltration of Fine
Average Model Performance. Clim. Res. 2005, 30, 79. Particle/Protein Binary Suspension. J. Membr. Sci. 2006, 274, 183.
(41) Yeh, H. M. Exponential Model Analysis of Permeate Flux for (63) Sethi, S.; Wiesner, M. Modeling of Transient Permeate Flux in
Ultrafiltration in Hollow-Fiber Modules by Momentum Balance. Cross-Flow Membrane Filtration Incorporating Multiple Particle
Chem. Eng. J. 2009, 147, 202. Transport Mechanisms. J. Membr. Sci. 1997, 136, 191.

424 dx.doi.org/10.1021/ie5033292 | Ind. Eng. Chem. Res. 2015, 54, 414−425

Industrial & Engineering Chemistry Research Article

(64) Vera, L.; Delgado, S.; Elmaleh, S. Dimensionless Numbers for

the Steady-State Flux of Cross-Flow Microfifiltration and Ultra-
filtration with Gas Sparging. Chem. Eng. Sci. 2000, 55, 3419.
(65) Field, R. W.; Pearce, G. K. Critical, Sustainable and Threshold
Fluxes for Membrane Filtration with Water Industry Applications. Adv.
Colloid Interface Sci. 2011, 164, 38.
(66) Wemsy Diagne, N.; Rabiller-Baudry, M.; Paugam, L. On the
Actual Cleanability of Polyethersulfone Membrane Fouled by Proteins
at Critical or Limiting Flux. J. Membr. Sci. 2013, 425−426, 40.
(67) Baruah, G. L.; Couto, D.; Belfort, G. A Predictive Aggregate
Transport Model for Microfiltration of Combined Macromolecular
Solutions and Poly-Disperse Suspensions: Testing Model with
Transgenic Goat Milk. Biotechnol. Prog. 2003, 19, 1533.
(68) Gésan-Guiziou, G.; Boyaval, E.; Daufin, G. Critical Stability
Conditions in Crossflow Microfiltration of Skimmed Milk: Transition
to Irreversible Deposition. J. Membr. Sci. 1999, 158, 211.
(69) Gésan-Guiziou, G.; Daufin, G.; Boyaval, E. Critical Stability
Conditions in Skimmed Milk Crossflow Microfiltration: Impact on
Operating Modes. Lait 2000, 80, 129.
(70) Piry, A.; Kuhnl, W.; Grein, T.; Tolkach, A.; Ripperger, S.;
Kulozik, U. Length Dependency of Flux and Protein Permeation in
Crossflow Microfiltration of Skimmed Milk. J. Membr. Sci. 2008, 325,
887.
(71) Bacchin, P.; Espinasse, B.; Aimar, P. Distributions of Critical
Flux: Modelling, Experimental Analysis, and Consequences for Cross-
Flow Membrane Filtration. J. Membr. Sci. 2005, 250, 223.

425 dx.doi.org/10.1021/ie5033292 | Ind. Eng. Chem. Res. 2015, 54, 414−425