Sie sind auf Seite 1von 9

Separation and Purification Technology 70 (2010) 265–273

Contents lists available at ScienceDirect

Separation and Purification Technology


journal homepage: www.elsevier.com/locate/seppur

Optimization of filtration conditions for CIP wastewater treatment by


nanofiltration process using Taguchi approach
Z. Beril Gönder ∗ , Yasemin Kaya, Ilda Vergili, Hulusi Barlas
Istanbul University, Faculty of Engineering, Department of Environmental Engineering, Avcilar Campus, Avcilar, Istanbul 34320, Turkey

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

Article history: In this study, the nanofiltration of cleaning-in-place (CIP) wastewater was studied using a two-step
Received 24 April 2009 nanofiltration (NF) process. In the first step, the effects of operating parameters including pH, tem-
Received in revised form perature, and transmembrane pressure on the membrane fouling were investigated using a loose NF
29 September 2009
membrane. Also, Taguchi method was applied in order to design the experiments and optimize the
Accepted 1 October 2009
experimental results. The optimum conditions providing the lowest flux decline were estimated. L9 (34 )
orthogonal array for experimental planning and the smaller-the-better response category was selected
Keywords:
to obtain optimum conditions. Analysis of variance (ANOVA) was used to determine the most significant
Nanofiltration
CIP wastewater
parameters affecting the flux decline caused by membrane fouling. The optimum conditions were found
Taguchi method as the second level of pH (7), first level of temperature (25 ◦ C) and first level of transmembrane pressure
ANOVA (12 bar). Under these conditions, flux decline caused by membrane fouling was predicted as 24% that
Flux decline was within the range of confidence limit of the observed value of 20%. Transmembrane pressure was
found to be the most important factor on the flux decline. The effect of pH on the membrane fouling
was also investigated with scanning electron microscopy (SEM) and contact angle measurements. It was
found that pore plugging was dominant at pH 4. In the first step of the treatment; 97% chemical oxygen
demand (COD), 100% sulphate, 50% chloride and 38% conductivity removal was obtained under the opti-
mized conditions. In the second step, the quality of permeate obtained from the first step was improved
using a tight membrane at 20 and 30 bar. Even better rejections were achieved especially for chloride
and conductivity in the second step at 30 bar.
© 2009 Elsevier B.V. All rights reserved.

1. Introduction ultrafiltration (UF) membrane processes are preferred for CIP


wastewater treatment if the surfactant concentration is above
In the detergent industry during the production such as sham- the critical micelle concentration (CMC). However, nanofiltration
poo, liquid dishwasher detergent, shower gel and liquid soap, (NF) membrane process is chosen when the surfactant concentra-
cleaning is a crucial operation after every change of product for tion is below the CMC [3]. In most of the studies, NF/UF or only
ensuring hygienic safety of product and for recovering perfor- UF membranes have been chosen for CIP wastewater treatment
mances of plant. Cleaning is performed by a two-step (hot tap [4–7].
water, followed by deionized water) cleaning-in-place (CIP) sys- The application of membrane processes has some drawbacks
tem [1]. CIP wastewater contains considerable amount of surfactant like concentration polarization and fouling that result in flux
which is hazardous and toxic for the surrounding environment decline with time. Fouling of membranes has limited the use
and must be treated before releasing to receiving media. There of nanofiltration in wastewater treatment. However, flux can be
are several methods such as biological treatment, foam fractiona- improved by adjusting the filtration conditions such as pH, tem-
tion, chemical precipitation, oxidation, adsorption and membrane perature, pressure and velocity [8]. Statistical experimental designs
processes which can be used for the treatment of CIP wastewater described as design of experiments (DOE) can be used to investi-
[2]. gate the effect of all the possible combinations of the conditions. In
Membrane processes are widely used in various wastewa- this context, the Taguchi method as a statistical technique is used to
ter treatment applications due to easy operation, lower cost in determine the optimum conditions. This method is introduced by
some cases and high removal efficiency. Microfiltration (MF) and R.A. Fisher in the 1920s and the concept was improved in the 1940s
by G. Taguchi. The aim of this method is to find out the optimal
and robust process characteristic that has a minimized sensitivity
∗ Corresponding author. Tel.: +90 212 4737070; fax: +90 212 4737180. to noises. It is a type of fractional factorial design which uses an
E-mail address: bgonder@istanbul.edu.tr (Z.B. Gönder). orthogonal array to study the influence of factors with only a small

1383-5866/$ – see front matter © 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.seppur.2009.10.001
266 Z.B. Gönder et al. / Separation and Purification Technology 70 (2010) 265–273

parameters through the confirmation experiment and calculation


Nomenclature of the confidence interval [11,12].
Taguchi method is widely employed in various field of science
A membrane filtration area (m2 ) from chemistry to engineering and from microbiology to agricul-
Ai sum of all observations of level i of factor A ture [13–18]. But limited study has been reported in the membrane
ANOVA analysis of variance field. Hesampour et al. [9] employed the influence of different
CI confidence interval operating conditions (pH, oil concentration, feed flow velocity and
CIP cleaning-in-place temperature) on the permeate flux of a model oil wastewater. It
CMC critical micelle concentration was indicated that the rank of important factors changed with salt
CMF concentration mode of filtration concentration. The best optimum conditions were achieved when
COD chemical oxygen demand (mg L−1 ) salt and oil concentrations were at the lowest level whereas tem-
DOE design of experiment perature, transmembrane pressure and flow velocity were at the
DOF degrees of freedom highest levels in this work. Rezvanpour et al. [19] investigated the
DOFe degrees of freedom of error effects of membrane type, transmembrane pressure, the content of
ei random error in the ith experiment oil in the feed, the flow velocity and pH on the UF of an emulsion
FR flux recovery of kerosene in water. They found that membrane type is the most
Fcr critical F value important factor affecting the UF to get the best flux. In another
J permeate flux (L m−2 h−1 ) study conducted by Hesampour et al. [20], they aimed to estimate
Jf pure water flux of fouled membrane (L m−2 h−1 ) the best operational conditions in the treatment of wastewater
Js wastewater flux (L m−2 h−1 ) from metal working fluids by UF. The experimental results showed
Jw pure water flux of clean membrane (L m−2 h−1 ) that an alkaline pH (11), a high transmembrane pressure (3.5 bar),
kA number of the levels of factor A a high temperature (40 ◦ C) and a low amount of CaCl2 concentra-
L9 orthogonal array layout (34 ) tion (0 g L−1 ) gave the optimum conditions. Madaeni and Koocheki
MF microfiltration [21] explored the parameters such as temperature, transmembrane
MS mean of square (variance) pressure and concentration which affect the flux and rejection in
MSA variance of A the reverse osmosis treatment of wastewater containing nitrate,
MSe variance of error nitrite, sulfite and phosphate. Their research findings revealed that
MWCO molecular weight cut-off (Dalton) temperature and transmembrane pressure had the highest con-
n number of repetition performed for an experimental tribution in flux whereas concentration of feed solution was the
combination most important factor in ion rejection. Idris et al. [22] used the
nAi number of all observations at level i of factor A Taguchi method to discuss the relationship between the rejection
N the number of all observation rate coefficient, permeation rate, and dry–wet spinning conditions
NF nanofiltration for the production of cellulose acetate hallow fiber reverse osmosis
P percentage of the product obtained experimentally membrane. The experimental results indicated that the type of bore
RF relative flux fluid and the dope extrusion rate were the most important factors
S sample size for confirmation test among the other factors (polymer contents, acetone/formamide
SS sum of squares ratio, residence time and gas flushing rate) in the performance of RO
SSA sum of squares of A membranes. Also, SEM images of the hallow fibers were presented
SSe sum of squares of error to depict how the morphology of the membrane was affected by
SST total sum of squares these factors.
SEM scanning electron microscopy Based on literature review no other study, considering the effec-
S/N signal-to-noise ratio tive factors on flux decline in nanofiltration of CIP wastewater
t filtration time (min) using Taguchi method, was found. This paper describes a case
T sum of all observations study investigating the parameters that affect the flux decline
UF ultrafiltration caused by membrane fouling for CIP wastewater treatment by
Vc final volume of the concentrate (L) nanofiltration process. The parameters of pH, temperature, and
Vf initial volume of feed (L) transmembrane pressure were chosen for the experiments. The
VRF volume reduction factor main objective is to find a combination of effective parameters
Xi fixed effect of the quantity level combination used to achieve lower flux decline caused by membrane fouling using
in ith experiment Taguchi method. Optimum conditions were determined using the
yi is the observation of i S/N ratio of experimental results. Results obtained both experimen-
 overall mean of the performance value tally and theoretically were given and analyzed. ANOVA was used to
˝ percentage value subject to omega transformation determine the influence and relative importance of the factors. Also,
(dB) membrane fouling on pores and on surface of NF membranes was
characterized by SEM and contact angle measurements, respec-
tively. Furthermore, it was tried to improve the permeate quality
number of experiments. The design of experiment using Taguchi obtained from the first step NF membrane under the conditions
method provides efficient and systematic approach to determine providing the lower flux decline using different NF membrane.
the optimum conditions [9,10].
Taguchi method is based on several steps as follows iden-
tification of the quality characteristics and selection of design 2. Materials and methods
parameters, determination of the number of factor levels, selection
of the appropriate orthogonal array, execution of the experiments 2.1. Membrane filtration
based on the arrangement of the orthogonal array, evaluation of
the results using signal-to-noise (S/N) ratios, ANOVA, selection of Filtration experiments were performed in a lab-scale plant in
the optimum levels of factors, verification of the optimum process cross-flow operation, which was purchased from Osmo, Germany.
Z.B. Gönder et al. / Separation and Purification Technology 70 (2010) 265–273 267

Table 1 2.4. Calculation of flux decline


Characteristics of NF membranes used in the experiments.

FM NP030 FM NP010 The permeate flux is described by Darcy’s law [25]:


Material Polyether-sulfone Polyether-sulfone 1 dV
Maximum pressure (bar) 40 40 J= (2)
A dt
MWCO (Da) 400 1000
Maximum temperature (◦ C) 95 95 where J is the permeate flux, A is the effective membrane area, V
pH range 0–14 0–14 is the total volume of permeate, and t is the filtration time. The
Zeta potential (mV)a permeate fluxes were measured in three steps in order to deter-
pH 4 −5 −2 mine the decline percent of flux. Firstly, the pure water flux was
pH 7 −15 −12 measured and the flux at steady state was defined as Jw . Secondly,
pH 10 −18 −15
wastewater permeate flux was measured and the flux at the end
a
From Ref. [23] (isoelectric point is pH 4 for both membranes; old names of FM of the experiment was defined as Js . Finally, pure water permeate
NP030 and FM NP010 membranes are N 30F and NFPES10, respectively). flux with the fouled membrane was measured again until a stable
permeate flux was occurred and it was defined as Jf .
The relative flux, RF, was defined as
The filtration cell that has an effective area of 80 cm2 was con- J 
structed from stainless steel. Experimental set-up was equipped s
RF (%) = × 100 (3)
with a high pressure pump. Commercial NF membranes, loose Jw
membrane FM NP010 and tight membrane FM NP030 (Microdyn- The flux decline occurring during filtration is (100 − RF). The flux
Nadir, GmbH, Germany) were used in the first and second step, recovery, FR, was defined as
respectively. The characteristics of membranes are summarized J 
f
in Table 1. Before first use, each membrane was precompacted FR (%) = × 100 (4)
Jw
with pure water for 5 h at the transmembrane pressure of 30 bar
to obtain stable membrane structure. Cross-flow rate was kept at In this case that (100 − FR) corresponds to the irreversible flux
2.0 L min−1 throughout the all experiments. The pH was adjusted decline caused by fouling. In addition, (FR − RF) corresponds to the
using 1 M NaOH and 1 M HNO3 , and the temperature of the feed reversible flux decline caused by either concentration polarization
solutions was maintained by using the cooling water. Experimen- or reversible adsorption phenomenon [26].
tal runs were carried out with a feed volume of 7 L at the beginning
of each run using FM NP010 membrane. 2.5. Membrane characterization
All the experiments were conducted in concentration mode of
filtration (CMF). Permeate was collected in a beaker, concentrate 2.5.1. Contact angle
was circulated back to the feed vessel. The permeate flow rate was The hydrophobicity of the membrane surface was analyzed
measured by an electronic balance (Precisa 320 XB-1200 C) and by contact angle measurements, which were carried out using a
recorded by a computer. Experiments were carried out until the Goniometer (KSV Instruments, CAM 101). The sessile drop method
volume reduction factor (VRF) value of 3 was reached. The VRF was was chosen for the measurements. Each contact angle was mea-
calculated using the following equation: sured 5–10 times, and an average value was calculated. All contact
angle measurements were carried out for both clean and fouled
Vf membranes. Contact angle measurements were used to analyze
VRF = (1)
Vc membrane fouling by means of membrane surface change determi-
nations. A larger contact angle corresponds to a more hydrophobic
where Vf and Vc are the initial volume of feed and the final volume
material.
of the concentrate, respectively.

2.5.2. Scanning electron microscopy (SEM)


2.2. Shampoo CIP wastewater The membrane fouling on the pores of membranes was observed
with a scanning electron microscope using a JEOL/JSM-6335F-INCA
The shampoo CIP wastewater used in the experiments was instrument with an accelerating voltage of 10.0 kV.
obtained from the first step of CIP system in a detergent factory
located in Gebze. A detailed analysis of the wastewater is presented 2.6. Experimental design based on Taguchi method
in Table 2.
Taguchi method was used to design the experiments. The
2.3. Analytical procedure Taguchi method applies fractional factorial experimental designs,
called orthogonal arrays, to reduce the number of experiments. The
All the analyses were carried out according to the Standard selection of a suitable orthogonal array depends on the number
Methods [24]. Conductivity was measured by a WTW Level 3 con- of control factors and their levels. The factors and their levels are
ductivity device. The pH of the samples was monitored by a Thermo presented in Table 3.
Orion 3-Star model pH meter. The total degrees of freedom need to be computed to select an
appropriate orthogonal array. Basically the degrees of freedom for

Table 2 Table 3
Characterization of shampoo CIP wastewater used in this study. Parameters and their values corresponding to their levels studied in experiments.
Parameter Unit Average values Parameters Designation Levels
−1
COD mg L 27,500
1 2 3
Chloride mg L−1 1,600
Sulphate mg L−1 586 pH A 4 7 10
Conductivity ␮s cm−1 4,000 Temperature (◦ C) B 25 35 45
pH – 7 Transmembranepressure (bar) C 12 16 20
268 Z.B. Gönder et al. / Separation and Purification Technology 70 (2010) 265–273

Table 4 the estimate of mean))], N is the number of all observation, S is the


A design of L9 (34 ) by the Taguchi method.
sample size for confirmation test, and CI is confidence interval.
Trial no. Factors If the experimental results are given in percentage (%), such as
membrane fouling before evaluating Eqs. (6) and (7), the ˝ trans-
A B C
formation of percentage values should be applied first using the
1 1 1 1
following formula [30]:
2 1 2 2
3 1 3 3 1 
4 2 1 2 ˝ (dB) = −10 log −1 (8)
P
5 2 2 3
6 2 3 1 where ˝ (dB) is the decibel value of percentage value subject to
7 3 1 3
8 3 2 1
omega transformation and P is percentage of the product obtained
9 3 3 2 experimentally. The values which of interest are also later deter-
mined by carrying out reverse transformation by using the same
equation.
the orthogonal array should be greater than or at least equal to those Finally, confirmation experiment must be carried out with opti-
for the parameters. For example, a three level design parameter mal process parameters to verify predicted results. If the predicted
counts for two degrees of freedom (DOF). In this study, there are six results are confirmed, the suggested optimum working conditions
degrees of freedom owing to being three factors. L9 (34 ) orthogonal will be adopted.
array of Taguchi design which involves nine experiments for four
parameters with three levels was used. This array has eight DOF.
2.7. Analysis of variance (ANOVA)
Experimental plan table according to L9 (34 ) is shown in Table 4
[27]. The order of the experiments was made random in order to
In order to determine the influence and relative importance of
avoid noise sources.
the factors, ANOVA was performed. ANOVA results are carried by
With the selection of L9 (34 ) orthogonal array, the number of
separating the total variability into contributions by each of the
experiments required can be reduced to 9. So, 9 experiments should
design parameters and error. Sum of squares (SS), DOF, mean of
be conducted in order to study the main effects and interactions
square (MS) and associated F-test of significance (F) can be calcu-
whereas full factorial experimentation would require 33 = 27 num-
lated as follows [12,22,31]:
ber of experiments.
k 
Taguchi method recommends the use of the loss function A
A2 T2
i
to measure the performance characteristics deviating from the SSA = − (9)
nAi N
desired value. The value of the loss function is further transformed i=1
into signal-to-noise (S/N) ratio. The optimum conditions should be
determined using the S/N ratio of the results obtained from exper- where kA is the number of the levels of factor A, nAi is the number of
iments designed by orthogonal array technique. There are three all observations at level i of factor A, Ai is the sum of all observations
basic S/N ratios, the larger-the-better, the smaller-the-better and of level i of factor A and T is the sum of all observations. SS of error
the nominal-the-better. The larger S/N ratio corresponds to bet- is computed using the following equation:
ter performance characteristic. The lower flux decline is required SSe = SST − (SSA + SSB + · · ·) (10)
for optimization of the membrane fouling. Therefore, the smaller-
the-better option was selected in this study. The performance where SST is the total SS:
characteristics were evaluated using the following equation [28]:
  
N
T2
1
n SST = yi 2 − (11)
N
The smaller-the-better S/N = −10 log Yi 2 (5) i=1
n
i=1
where yi is the observation of i. MS is calculated by dividing the
where n is the number of repetition performed for an experimental sum of squares by the degrees of freedom. DOFA is estimated by
combination and Yi is the performance value of the ith experiment. DOFA = kA − 1. F value is calculated as follows:
In the Taguchi method, the performance value corresponding
to the optimum working conditions can be predicted by using the MSA
FA = (12)
balanced characteristic of orthogonal array. For this purpose, the MSe
following additive model may be used [11,29]: MSe is the variance of error.
Yi =  + Xi + ei (6)
3. Results and discussions
where  is the overall mean of the performance value, Xi is the fixed
effect of the quantity level combination used in ith experiment, and 3.1. Taguchi results
ei is the random error in the ith experiment. Since Eq. (6) is a point
estimation, which is calculated by using experimental data in order Fig. 1 shows the flux declines as a function of VRF during the
to determine whether results of the confirmation experiments are filtration of shampoo wastewater for all the experiments carried
adequate or not, the confidence interval must be evaluated. At the to Taguchi experimental design (Table 4). In addition, wastewater
selected error level, the confidence interval is calculated using the fluxes (Js ) which are used for calculating the total flux decline (Eq.
following equation [11]: (3)) are presented in Fig. 1.
  1   1  As seen from Fig. 1, the highest total flux decline (83%) was
CI = F(1, DOFe ) × MSe × + (7) occurred at Trial 3 whereas the lowest total flux decline (62%) was
neff S
occurred at Trial 1. Variation in permeate flux was more stable at
where F is the value of F table at desired confidence level at degrees Trial 1.
of freedom of 1 and degrees of freedom of error (DOFe ), MSe is mean The measured Jw values, Jf values, the determined flux declines
of square (variance) of error, neff is [N/(1 + (total DOF associated in caused by fouling and the computed S/N ratios for each series of
Z.B. Gönder et al. / Separation and Purification Technology 70 (2010) 265–273 269

Fig. 1. Permeate fluxes as a function of VRF during the filtration of shampoo wastewater for all the experiments. (a) Trials carried out at pH 4. (b) Trials carried out at pH 7.
(c) Trials carried out at pH 10.

Table 5 example, the mean S/N ratio for A factor at levels 1, 2 and 3 can
Experimental results for flux decline and corresponding S/N ratios.
be calculated by averaging the S/N ratios for the experiments 1–3,
Trial no. Flux (L m−2 h−1 ) Flux decline (%) S/N ratio (dB) 4–6 and 7–9, respectively. The mean S/N ratio for every factor is
calculated similarly.
Jw Jf Fouling (100 − FR)
The S/N ratio averages for each factor at three levels are pre-
1 137 93 32 −30.09
sented in Fig. 2. As seen from figure, overall mean value was
2 255 167 34 −30.75
3 435 221 49 −33.84 calculated as −30.78 from all Taguchi experiment results. On the
4 185 134 27 −28.71 other hand, the maximum variation level was determined as −2.75.
5 315 192 38 −31.59 Also it can be seen that the slopes of the lines between different lev-
6 262 184 30 −29.44 els are not the same for pH and temperature factors. So, the levels
7 234 151 35 −30.96
8 190 147 27 −28.63
have different influence on membrane fouling. The increase in foul-
9 358 199 44 −32.96 ing is bigger when the temperature changes from 35 to 45 ◦ C than
when it changes from 25 to 35 ◦ C. So, it can be said that membrane
fouling was lesser affected from temperature when the tempera-
experiments are presented in Table 5. In order to evaluate the influ- ture is lower than 35 ◦ C. However, the slopes of the lines between
ence of each factor on the flux decline, the S/N ratio for each factor 12–16 bar and 16–20 bar are almost the same.
should be computed. The S/N ratio for a single factor can be cal- Fig. 2(a) shows the effect of pH on the membrane fouling. The
culated by averaging the value of S/N ratios at different levels. For pH of minimum membrane fouling was observed at 7, which is the

Fig. 2. Main effect plots for S/N ratios of flux decline caused by fouling: (a) pH, (b) temperature and (c) transmembrane pressure.
270 Z.B. Gönder et al. / Separation and Purification Technology 70 (2010) 265–273

Fig. 3. SEM images of the top layers’ cross-section of membranes. (a) FM NP010 before filtration. (b) FM NP010 after filtration at pH 4. (c) FM NP010 after filtration at pH 10.
(d) FM NP010 after filtration at pH 7.

medium pH value among the three levels tested in this study. The by the deposition of the particles in the pores. So, it could be seen
value of wastewater pH can affect the membrane surface charge that the membrane was indeed fouled after the filtration at pH 4.
due to the disassociation of membrane’s functional groups. Zeta However, SEM images of the membranes after filtration at pH 7
potential is used to quantify the membrane surface charge. The and 10 were almost similar to clean membrane. Flux decline might
membranes are positively charged at pH values below the iso- result from the surface fouling compared to pore plugging.
electric point whereas they are negatively charged at pH values As seen from contact angle measurements (Fig. 4), contact angle
above the isoelectric point of the groups. The surface charge of of the membrane at pH 7 was similar to that at pH 10. But at pH 4,
the membrane increases because of increasing electrostatic repul- contact angle of the membrane was lower than the others. A higher
sion between a negatively charge solute and membrane [32,33]. FM contact angle indicates an increase in surface fouling. So, it can be
NP010 membrane skin layer is made of polyether-sulfone that has concluded that pore plugging was dominant at pH 4.
dissociable sulphonic acid groups in the chemical structure. For the As a conclusion, it can be said that the FM NP010 membrane
FM NP010 membrane, it is known that isoelectric point is 4.2 [23]. showed the best performance for membrane fouling at pH 7.
The membrane surface was nearly no charged at pH 4. Therefore, According to Fig. 2(b), the S/N ratio was decreased with increas-
there was not an electrostatic interaction between the membrane ing of temperature and hence the highest membrane fouling was
and the solutes in the wastewater. Separation of molecules could be occurred at higher temperature (45 ◦ C). It is known that an increas-
occurred due to size exclusion. So, the flux decline caused by foul- ing in temperature increases the permeate flux. On the other hand,
ing at pH 4 was higher than at pH 7 and 10. At pH 7 and 10, which
are above the isoelectric point, the membrane surface was nega-
tively charged. Thus, the electrostatic repulsion occurred between
the negatively charged membrane and the solutes, preventing the
adsorption of solutes onto the membrane. Also, the flux decline
caused by fouling at pH 10 was higher than that at pH 7. At high
values of pH, electrostatic repulsion between the charged groups
can cause a reduction in pore size and the pores can shrink [32]. This
situation resulted in higher membrane fouling at pH 10. Also, SEM
and contact angle measurements of the new and clean membranes
were determined to visualize membrane fouling on the pores and
membrane surface for different pH values. Fig. 3 represents the FM
NP010 membrane before (a) and after filtration experiments which
were carried out at pH 4 (b) that the highest membrane fouling
(49%) occurred, at pH 10 (c) that the intermediate membrane foul-
ing (35%) occurred and at pH 7 (d) that the lowest membrane fouling
(27%) occurred.
As seen from Fig. 3(a), the pores of clean FM NP010 membrane
are clearly visible. But, for the membrane that has been fouled with
wastewater at pH 4 (Fig. 3(b)), the size of the pores was reduced Fig. 4. Contact angle values for clean and fouled membranes at pH 4, 7 and 10.
Z.B. Gönder et al. / Separation and Purification Technology 70 (2010) 265–273 271

at higher temperature values, the flux decline caused by fouling was


greater than at low temperatures. The variation of permeate flux
with temperature is usually explained by solvent viscosity, solvent
diffusion coefficient in the membranes and thermal expansion of
the membrane material. The reduction of solvent viscosity results
increases in solvent diffusion coefficient. Also, due to the higher
temperature the membrane structure may expand and solutes can
easily pass through the membrane [34,35]. Therefore, the increase
in flux decline at higher temperature can be attributed to plugging
within pores and membrane surface by the solutes that can easily
pass at higher temperature.
The applied transmembrane pressure is one of the most crit-
ical factors in operating a membrane process. Experiments with
varying transmembrane pressures were carried out to investigate
whether the phenomenon of fouling exist. It can be seen from
Fig. 2(c), an increasing of transmembrane pressure was decreased
S/N ratio. So, the lowest membrane fouling was occurred at lower
transmembrane pressure (12 bar). The flux decline due to fouling
was increased with increasing pressure. Also, Fig. 3(c and d) sup-
ports that pore plugging was dominant at 20 bar (Trial 7) than that
of 16 bar (Trial 4). As a result of increasing the transmembrane
pressure, a phenomenon known as concentration polarization on Fig. 5. Contribution of each factor on the performance statistics.
the membrane surface occurs and the concentration differences
between the both sides of the membrane increases [36]. Therefore,
the diffusion driven force increases, more particles cross the mem- freedom for the error is 2. Fcr value of the parameters for degrees
brane and pore plugging with free surfactant aggregates occurs that of freedom of 2 and 2 at a confidence level of 95% is 19. According
opposes the permeate flow. It can be thought that pore plugging at to the results, the F-ratio for all factors is smaller than the Fcr value.
higher transmembrane pressure resulted in flux decline. This means that the variance of all factors is insignificant compared
The optimum conditions for flux decline caused by fouling were with the variance of error and none of them have meaningful effect
determined considering the levels which give the biggest S/N ratio on the responses. The use of the F-ratios in an ANOVA is only helpful
for the factors. As seen from Fig. 2, A2 (pH 7), B1 (T = 25 ◦ C) and C1 for the qualitative evaluation of factorial effects. Quantitative eval-
(P = 12 bar) combination is the best filtration condition to mini- uation can be achieved with using percentage contribution (P%)
mizing the flux decline caused by membrane fouling. If Table 4 is [30]. It is calculated by dividing the source’s net variation by SST ,
examined carefully, it can be seen that experiments corresponding which is given as follows:
to optimum conditions for minimum flux decline have not been
carried out during the experimental study. SSA − (DOFA × MSe )
P (%) = × 100 (13)
SST
3.2. ANOVA results
Percent contributions of all factors are presented in Fig. 5.
ANOVA was performed in order to see whether the process The order of importance of factors is as follows: transmembrane
parameters are statistically significant or not. The results of ANOVA pressure > temperature > pH. Transmembrane pressure is the most
are listed in Table 6. The row which is marked as error refers to the important factor for flux decline caused by fouling. Higher pressure
error caused by uncontrollable factors (noise). In general, the value results in higher permeate flux. However, rapid membrane fouling
should be below 50%; otherwise the results would not be reliable. was occurred by higher permeate flux.
As seen from table, the calculated error is about 9% which is far
enough from the limit. It means that the error of the experiment is
3.3. Confirmation experiment
not significant.
F-ratio is a tool to indicate which parameter has a significant
The final step is confirmation testing in the Taguchi method.
effect on the flux decline caused by membrane fouling. The larger
The confirmation experiment is highly recommended by Taguchi
the F-ratio, the greater is the effect on the membrane fouling. An
to verify experimental conclusions. After optimum conditions were
F-ratio was calculated using Eq. (12) from the experimental results.
determined, the confirmation experiment was performed with
Then it was compared to the critical F value (Fcr ) which can be found
combination of the optimum levels to compare the results with
in most of the statistics and experimental design books [37]. If the
the predicted performance.
calculated F-ratio is greater than the Fcr value, the statistical test is
The pure water flux of fouled membrane (Jf ) was found to be
significant at the selected confidence level. In this study, degrees of
109 L m−2 h−1 in the confirmation experiment. The flux decline
caused by fouling was calculated according to Eq. (4) and the
Table 6 observed flux decline was found as 20%. The predicted flux decline
Results of ANOVA for flux decline caused by membrane fouling. caused by fouling and 95% confidence interval of the confirmation
Parameters DOF SS MS F-ratio Fcr %P test were calculated by using Eqs. (6) and (7), respectively. The
results are shown in Table 7.
A 2 66.89 33.44 1.66 19 14.9
B 2 160.22 80.11 3.98 19 35.7
The flux decline caused by fouling is within a ±5% error
C 2 181.55 90.77 4.51 19 40.4 range. Also, there is a good agreement between the predicted and
observed values.
Error 2 40.23 20.11 9
The quality of the composite permeate obtained from confirma-
Total 8 448.89 100 tion experiment was compared with feed values in Table 8.
272 Z.B. Gönder et al. / Separation and Purification Technology 70 (2010) 265–273

Table 7 Table 10
Optimum working conditions, observed and predicted flux decline caused by Process performance of FM NP030 membrane at 20 and 30 bar.
fouling.
Parameter Permeate quality (percent removal %)
Parameter Value Level
20 bar 30 bar
pH 7 2
Temperature ( ◦ C) 25 1 COD (mg L−1 ) 570 (97.9%) 400 (98.5%)
Transmembrane pressure (bar) 12 1 Chloride (mg L−1 ) 425 (73%) 375 (77%)
Observed flux decline caused by fouling (%) 20 Conductivity (␮s cm−1 ) 1285 (68%) 1253 (69%)
Predicted flux decline caused by fouling (%) 24
Predicted confidence interval for fouling (%) 10–45
ate quality since concentration polarization has reversible effect on
flux decline.
Table 8 The process performance of the FM NP030 membrane at 20
Characterization of composite permeate obtained from confirmation experiment.
and 30 bar is shown in Table 10. In addition, the rejections were
Parameter Feed wastewater Permeate quality calculated based on the feed concentrations and the composite per-
quality (percent removal) meate concentrations obtained from the FM NP030 membrane at
COD (mg L−1 ) 27,500 820 (97%) the end of the experiments for evaluating the total performance of
Chloride (mg L−1 ) 1,600 800 (50%) the system.
Sulphate (mg L−1 ) 586 0 (100%) The best quality of composite permeate was obtained at 30 bar.
Conductivity (ms cm−1 ) 4 2.46 (38%)
It was observed that the resulting permeate was free from sulphate.
The reduction observed in COD was greater than 98%. But chloride
and conductivity could not be removed effectively.

4. Conclusions

In this study, Taguchi design of experiment (L9 ) was employed


to optimize the effective parameters on the flux decline caused by
fouling. The smaller-the-better S/N ratio was used to analyze the
results of experiments.
According to the mean S/N ratios at three levels of pH value, min-
imum flux decline caused by fouling was observed at pH 7 which is
above the isoelectric point of the membrane. SEM and contact angle
measurements of the new and fouled membranes were shown that
pore plugging was dominant at pH 4. However, membrane foul-
ing was occurred on the membrane surface at pH 7 and 10. The
S/N ratio was decreased with increasing of temperature from 25
to 45 ◦ C. This case may be caused by plugging within pores and
membrane surface by the solutes that can easily pass at higher
Fig. 6. Permeate fluxes as a function of VRF for FM NP030 membrane at different temperature. S/N ratio decreased with increasing of transmem-
transmembrane pressure values (pH 7, temperature = 25 ◦ C).
brane pressure. At lower transmembrane pressure (12 bar), lower
flux decline caused by fouling was observed than that of higher
3.4. Enhancement of permeate quality transmembrane pressure (20 bar). With increasing concentration
polarization at higher transmembrane pressure, pore plugging can
The composite permeate (4.7 L) obtained from the loose be occurred that resulted in flux decline.
FM NP010 membrane (pH 7, T = 25 ◦ C, transmembrane pres- The optimum conditions within the selected parameter values
sure = 12 bar) was treated using a tight FM NP030 membrane to were found as the second level of pH (7), first level of tempera-
provide the better permeate quality. The experiments were con- ture (25 ◦ C) and first level of transmembrane pressure (12 bar). The
ducted at 20 and 30 bar until the VRF value was reached to 3. The confirmation experiment was carried out at optimum working con-
permeate fluxes depend on VRF are shown in Fig. 6. ditions. Flux decline caused by membrane fouling was decreased
As seen from Figure 6, the flux slowly decreased at 20 bar until to 20% by setting the control factors. Predicted (24%) and observed
the VRF value was reached to 1.5 whereas it decreased fastly at (20%) flux decline values are close to each other. Also, the result
30 bar. The fluxes were almost constant after VRF value of 1.5 for obtained from confirmation experiment is within the calculated
both transmembrane pressures. The flux decline analysis for both confidence interval. So, it may be concluded that the additive model
transmembrane pressure values caused by concentration polariza- is adequate for describing the flux decline process on the various
tion and fouling is also shown in Table 9. parameters.
As seen from Table 9, the contribution of fouling to the total According to the percent contribution of each factor, indicated
flux decline was larger at 20 bar than that of 30 bar. In other words, in the ANOVA table, the most effective parameter for minimum flux
flux decline due to the concentration polarization was dominant at decline was found to be transmembrane pressure while pH had an
30 bar. 30 bar is more suitable for the enhancement of the perme- effect at a lesser degree.

Table 9
The flux decline results of FM NP030 membrane at 20 and 30 bar.

Transmembrane pressure (bar) Flux (L m−2 h−1 ) Flux decline (%)

Jw Js Jf Total(100 − RF) Concentration polarization(FR − RF) Fouling(100 − FR)

20 53 44 48 17 8 9
30 80 56 71 30 19 11
Z.B. Gönder et al. / Separation and Purification Technology 70 (2010) 265–273 273

Under the optimized conditions 97% COD, 50% chloride and [13] Y. Lin, Y. Chen, D. Wang, H. Lee, Optimization of machining parameters in mag-
38% conductivity removal was obtained with the loose NF mem- netic force assisted EDM based on Taguchi method, J. Mater. Process. Technol.
209 (7) (2009) 3374–3383.
brane (FM NP010) whereas the resulting permeate was free from [14] O. Keleş, An optimization study on the cementation of silver with copper in
sulphate. The permeate quality especially for chloride and conduc- nitrate solutions by Taguchi design, Hydrometallurgy 95 (2009) 333–336.
tivity was improved using the tight NF membrane (FM NP030) at [15] K.D. Kim, D.W. Choi, Y. Choa, H.T. Kim, Optimization of parameters for the syn-
thesis of zinc oxide nanoparticles by Taguchi robust design method, Colloids
20 and 30 bar. It was achieved a chloride removal of 77% and a Surf. A 311 (2007) 170–173.
conductivity removal of 69% at 30 bar in which 11% flux decline [16] W.H. Yang, Y.S. Tarng, Design optimization of cutting parameters for turning
caused by fouling occurred. Also, flux decline due to the concentra- operations based on the Taguchi method, J. Mater. Process. Technol. 84 (1998)
122–129.
tion polarization has higher contribution to the total flux decline at [17] M. Yeşilyurt, Determination of the optimum conditions for the boric acid
30 bar. The permeate collected from the membrane at 30 bar is not extraction from colemanite ore HNO3 solutions, Chem. Eng. Process. 43 (2004)
sufficient for discharging it into the receiving water. But organic 1189–1194.
[18] M. Çopur, C. Özmetin, E. Özmetin, M.M. Kocakerim, Optimization study of leach-
load transferred to wastewater treatment plant was reduced by a
ing of roasting zinc sulphide concentrate with sulphuric acid solution, Chem.
two-step nanofiltration process. If further removal of chloride and Eng. Process. 43 (2004) 1007–1014.
conductivity is required, an RO process can be additionally used. [19] A. Rezvanpour, R. Roostaazad, M. Hesampour, M. Nyström, C. Ghotbi, Effective
Furthermore, the concentrate obtained from both membranes (FM factors in the treatment of kerosene–water emulsion by using UF membranes,
J. Hazard. Mater. 161 (2009) 1216–1224.
NP010 and FM NP030) can be used in the batch production of deter- [20] M. Hesampour, A. Krzyzaniak, M. Nyström, Treatment of wastewater from
gents. metal working by ultrafiltration, considering the effects of operating condi-
tions, Desalination 222 (2008) 212–221.
[21] S.S Madaeni, S. Koocheki, Application of taguchi method in the optimization
Acknowledgements of wastewater treatment using spiral-wound reverse osmosis element, Chem.
Eng. J. 119 (2006) 37–44.
This work was supported by Research Fund of the Istanbul Uni- [22] A. Idris, A.F. Ismail, M.Y. Noordin, S.J. Shilton, Optimization of cellulose acetate
hallow fiber reverse osmosis membrane production using Taguchi method, J.
versity (Project number: 514/05052006). Authors would like to Membr. Sci. 205 (2002) 223–237.
thank Evyap Detergent Company for their support. [23] K. Boussu, C. Vandecasteele, B. Van der Bruggen, Study of characteristic and the
performance of self-made nanoporous polyethersulfone membranes, Polymer
47 (2006) 3464–3476.
References [24] APHA/AWWA/WEF, Standard Methods for the Examination of Water and
Wastewater, American Public Health Association, Washington, DC, USA, 1995,
[1] Y. Kaya, H. Barlas, S. Arayici, Nanofiltration of Cleaning-in-Place (CIP) wastew- ISBN 0-87553-223-3.
ater in a detergent plant: effects of pH, temperature and transmembrane [25] Y. Lee, M.M. Clark, Modeling of flux decline during crossflow ultrafiltration of
pressure on flux behavior, Sep. Purif. Technol. 65 (2) (2009) 117–129. colloidal suspensions, J. Membr. Sci. 149 (1998) 181–202.
[2] S.H. Lin, C.M. Lin, H.G. Leu, Operating characteristics and kinetic studies of sur- [26] D. Doulia, I. Xiarchos, Ultrafiltration of micellar solutions of nonionic surfac-
factant wastewater treatment by Fenton oxidation, Water Res. 33 (7) (1999) tants with or without alachlor pesticide, J. Membr. Sci. 296 (2007) 58–64.
1735–1741. [27] M.S. Phadke, Quality Engineering Using Robust Design, Prentice Hall, Engle-
[3] B. Goers, J. Mey, G. Wozny, Optimised product and water recovery from batch- wood Cliffs, NJ, 1989.
production rinsing waters, Waste Manage. 20 (2006) 651–658. [28] P.J. Ross, Taguchi Techniques for quality engineering, 2nd edition, McGraw-Hill,
[4] A. Schreiner, B. Goers, U. Wiesmann, G. Wozny, Combination of ultrafiltration New York, 1996.
and biological treatment for product and water recovery, Eng. Life Sci. 2 (3) [29] S. Kaytakoğlu, L. Akyalçın, Optimization of parametric performance of a PEMFC,
(2002) 86–91. Int. J. Hydrogen Energy 32 (2007) 4418–4423.
[5] B. Goers, G. Wozny, Flexible design and operation of a two-step UF/NF sys- [30] Y.Ş. Yildiz, Optimization of Bomaplex Red CR-L dye removal from aqueous solu-
tem for product recovery from rinsing waters in batch production, Water Sci. tion by electrocoagulation using aluminium electrodes, J. Hazard. Mater. 153
Technol. 41 (10–11) (2000) 93–100. (2008) 194–200.
[6] B. Goers, M. Forstmeier, B. Wendler, G. Wozny, A systematic approach for water [31] M. Sadrzadeh, A. Razmi, T. Mohammadi, Separation of different ions from
network optimisation with membrane processes, Resour. Conserv. Recyc. 37 wastewater at various operating conditions using electrodialysis, Sep. Purif.
(2003) 217–226. Technol. 54 (2007) 147–156.
[7] B. Wendler, B. Goers, G. Wozny, Regeneration of process water containing [32] A.E. Childress, M. Elimelech, Relating nanofiltration membrane performance
surfactants by nanofiltration-investigation and modelling of mass transport, to membrane charge (electrokinetic) characteristics, Environ. Sci. Technol. 34
Water Sci. Technol. 46 (4) (2002) 287–292. (2000) 3710–3716.
[8] A.I. Schäfer, A.G. Fane, T.D. Waite, Nanofiltration Principles and Applications, [33] Y. Shim, H.G. Lee, S. Lee, S.H. Moon, J. Cho, Effects of NOM and ionic species on
Elsevier Ltd., UK, 2005, 1-85617-405-0. membrane surface charge, Environ. Sci. Technol. 36 (2002) 3864–3871.
[9] M. Hesampour, A. Krzyzaniak, M. Nyström, The influence of different factors [34] R.R. Sharma, R. Agrawal, S. Chellam, Temperature effect on sieving characteris-
on the stability and ultrafiltration of emulsified oil in water, J. Membr. Sci. 325 tics of thin-film composite nanofiltration membranes: pore size distributions
(2008) 199–208. and transport parameters, J. Membr. Sci. 223 (2003) 69–87.
[10] S.M. Mousavi, S. Yaghmaei, A. Jafari, M. Vossoughi, Z. Ghobadi, Optimization of [35] I. Kowalska, K. Majewska-Nowak, M. Kabsch-Korbutowicz, Influence of tem-
ferrous biooxidation rate in a packed bed bioreactor using Taguchi approach, perature on anionic surface active agent removal from a water solution by
Chem. Eng. Process. 46 (2007) 935–940. ultrafiltration, Desalination 198 (2006) 124–131.
[11] A.B. Engin, Ö. Özdemir, M. Turan, A.Z. Turan, Color removal from textile dye- [36] P. Banerjee, S. DasGupta, S. De, Removal of dye from aqueous solution using
bath effluents in a zeolite fixed bed reactor: determination of optimum process combination of advanced oxidation process and nanofiltration, J. Hazard. Mater.
conditions using Taguchi method, J. Hazard. Mater. 159 (2008) 348–353. 140 (2007) 95–103.
[12] M. Sadrzadeh, T. Mohammadi, Sea water desalination using electrodialysis, [37] D.C. Montgomery, Design and Analysis of Experiments, 5th edition, John Wiley,
Desalination 221 (2008) 440–447. New York, 2001.