Date post: | 16-Nov-2023 |
Category: |
Documents |
Upload: | independent |
View: | 0 times |
Download: | 0 times |
Do
Va
b
c
a
ARRAA
KPHRDV
1
chhidtopsttpaP
pp
0d
Journal of Membrane Science 338 (2009) 111–118
Contents lists available at ScienceDirect
Journal of Membrane Science
journa l homepage: www.e lsev ier .com/ locate /memsci
ehydration of water/dichloromethane/n-butanol mixtures by pervaporation;ptimisation and modelling by response surface methodology
erónica García a,∗, Junkal Landaburu-Aguirre a, Eva Pongrácz b, Paavo Perämäki c, Riitta L. Keiski a
Mass and Heat Transfer Process Laboratory, Department of Process and Environmental Engineering, P.O. Box 4300, FI-90014, University of Oulu, Oulu, FinlandThule Institute, Centre of Northern Environmental Technology, P.O. Box 7300, FI-90014, University of Oulu, Oulu, FinlandDepartment of Chemistry, P.O. Box 3000, FI-90014, University of Oulu, Oulu, Finland
r t i c l e i n f o
rticle history:eceived 21 January 2009eceived in revised form 7 April 2009ccepted 9 April 2009vailable online 3 May 2009
eywords:ervaporation
a b s t r a c t
This paper reports pervaporation of water from the ternary water/dichloromethane/n-butanol systemusing a hydrophilic poly(vinyl alcohol) based membrane (GKSS Forschungszentrum). In order to optimizethe separation process, an experimental design approach with one process and three formulation fac-tors was applied. The parameters studied were feed temperature (process factor), and water, n-butanoland dichloromethane feed initial concentrations (formulation factors). The D-optimal design yieldedquadratic models with excellent fit and predictive power for the responses water flux through the mem-brane, and the enrichment factor towards water. The results showed the impermeability of the membrane
ydrophilic membraneesponse surface methodology-optimal designOC purification
towards dichloromethane. The response surface analysis indicated the water flux through the membraneto be positively affected by temperature and initial water feed concentration. The enrichment factortowards water was positively influenced by the increase of n-butanol and dichloromethane initial feedconcentration. The response surface methodology indicted that a feed temperature of 49 ◦C and initialfeed concentration of water, dichloromethane and n-butanol of around 6.64, 41 and 52.36 wt%, respec-tively, were the optimal conditions for the pervaporation process within the studied range. The validity
ed ex
of the model was confirm. Introduction
Dichloromethane (DCM) and n-butanol are volatile organicompounds (VOCs) widely used as solvents in industry andave hazardous characteristics [1–3]. Pervaporation (PV) usingydrophilic membranes is a well established technique for remov-
ng water from organic solvents. The PV process is driven by aifference in chemical potential. It is traditionally explained withhe solution diffusion model, i.e., sorption of the feed componentsn the active surface layer of the membrane, diffusion of the com-onents through the membrane, and desorption at the permeateide [4]. The PV driving force is achieved by keeping the pressure ofhe downstream side much lower than the saturation pressure of
he permeating components. As a consequence, the selective com-ounds from the liquid mixtures are removed as vapour throughnon-porous perm selective membrane [5]. The advantages ofV include low operating temperatures that avoid the degrada-
Abbreviations: VOCs, volatile organic compounds; DCM, dichloromethane; PV,ervaporation; PVA, poly(vinyl alcohol); RSM, response surface methodology; PLS,artial least squares; ANOVA, analysis of variance.∗ Corresponding author. Tel.: +358 8 553 7862; fax: +358 8 553 2369.
E-mail address: [email protected] (V. García).
376-7388/$ – see front matter © 2009 Elsevier B.V. All rights reserved.oi:10.1016/j.memsci.2009.04.040
perimentally and the results were satisfactory.© 2009 Elsevier B.V. All rights reserved.
tion of the compounds, minimal energy expenditures, no emissionsinto the environment, no requirement of additional components inthe feed and, therefore, no product contamination [6]. In contrast,low fluxes, concentration polarization phenomenon and membraneswelling may limit PV efficiency.
Water removal from n-butanol has been reported by severalauthors using mainly poly(vinyl alcohol) (PVA) membranes [7–12].Dehydration of DCM by PV using inorganic membranes wasreported by Sommer and Melin [13]. Further, the existing studiescorrespond to binary mixtures. Information regarding the PV ofmulticomponent systems using hydrophilic membranes is scarce.
Traditionally, the study of the effects of experimental parame-ters on the PV performance has been conducted using an approachwith one variable at a time. In this method the effect of each experi-mental factor is investigated altering the level of one factor at a timewhile maintaining the levels of the other factors constant. If the pur-pose of the study is not to understand the mechanism of the system,but rather to determine the optimum operating conditions, con-ducting the research using response surface methodology (RSM)
becomes especially convenient. RSM is an efficient methodologywhere all the factors are varied simultaneously over a set of exper-imental runs. The use of RSM reduces the number of experimentsneeded for the analysis of the main effects and interactions betweenfactors. Statistical design proved to be useful in some membrane1 brane Science 338 (2009) 111–118
aitc
sgovwincdtuupwp
2
2
Spuu
kmddtTm
2
secwcipuwicdte
TG
C
DnW
12 V. García et al. / Journal of Mem
pplications [14–16]. However, literature regarding its applicabil-ty to PV is very limited [17–19]. Further, the studies do not refero multicomponent systems where composition of mixtures highlyonstrains the experimental designs.
This study is part of a research aiming at developing a treatmentystem for a certain type of industrial wastewater. The starting tar-eted compounds were n-butanol, DCM and sodium chloride. Onef the objectives was to reclaim VOCs for possible future re-use. Pre-iously, PV of the water–n-butanol–DCM–sodium chloride mixtureas performed using hydrophobic membranes [20,21]. The result-
ng permeates did not contain any electrolyte, however, they wereot concentrate enough in the VOCs for direct re-use. This workontinues from the result of the previous study and aims at theehydration of the VOCs. This paper reports the PV of water fromernary systems at different feed concentrations and temperaturessing a composite hydrophilic membrane. In this research, RSM wastilized to study the effect of experimental parameters on the PVerformance and to find the optimum conditions. The ultimate aimas to find suitable approximating functions that described the PVrocess in order to predict and determine the future responses.
. Experimental
.1. Materials
n-Butanol and DCM were purchased from VWR BDH Prolabo andigma–Aldrich, respectively. n-Butanol and DCM were of GPR recta-ure and purum grade, respectively. Feed solutions were preparedsing distilled Milli-Q water. Some properties of the compoundssed in this study are listed in Table 1.
The composite hydrophilic membrane used in this study wasindly supplied by GKSS Forschungszentrum (Geesthacht, Ger-any). The flat membrane consisted of three different layers: a
ense selective top layer of about 1 �m made of PVA and titaniumioxide, a supporting microporous middle layer of polyacryloni-rile, and a third mechanical support layer of polyphenylene sulfide.itanium dioxide is included as nanoparticles to add stability to theembrane. SME pictures of the membrane are shown in Fig. 1.
.2. Pervaporation experiments
PV experiments were conducted using a cross-flow laboratorycale membrane unit (model P28, CELFA AG, Switzerland) with anffective membrane area of 2.8 × 10−3 m2. All experiments wereonducted maintaining the permeate pressure below 3 mbars andith a feed flow rate of 3 dm3 min−1. Feed temperatures were kept
onstant by means of a thermostatic unit (LAUDA, Ecoline Stared-tion E 103). Details of the PV apparatus are described in previousublications [20,21].Permeate and feed samples were taken at reg-lar intervals and analysed by gas chromatography (Agilent, 6890N)ith different procedures. Permeate was collected by condensation
n liquid nitrogen cold traps, and, if needed, dissolved in a spe-ific amount of deionised water. The permeate composition wasetermined by a gas chromatograph with a flame ionization detec-or. The feed composition was determined by a gas chromatographquipped with a thermal conductivity detector.
able 1eneral properties of water, n-butanol and DCMa.
ompound Molecular formulae BP (◦C) Dipole momentum MW
CM CH2Cl2 40 1.60 4.9-Butanol C4H10O 117 1.66 74.1ater H2O 100 1.85 18
a BP, boiling point; MW, molecular weight.
Fig. 1. SEM photographs of pervaporation membranes at a magnification of 1000×.
2.3. Response surface methodology
RSM is a particular set of mathematical and statistical methodsthat includes experimental design, model fitting and validation andcondition optimisation. The study of the dehydration of the targetedsystem by PV was highly constrained by the membrane toleranceand the homogeneity of the feed. In such cases the experimentalruns are designed in accordance with D-optimal design. A detaileddiscussion of D-optimal design is found in Myers and Montgomery[22] and Eriksson et al. [23].
In this work, the RSM for the pervaporation of water fromthe ternary water/DCM/n-butanol was conducted by using theMODDE 8.0 software (Umetrics). The effect of the following fourfactors on the efficiency of the PV process was studied: the quan-titative factor feed temperature (T) and the formulation factorsinitial feed water concentration (Cw), initial feed DCM concen-tration (CDCM) and initial feed n-butanol concentration (Cb). Theinteraction between the effects of the different factors was alsoconsidered. Based on preliminary trials, the tolerance of the mem-brane and the homogeneity of the feed solution, the formulationfactors varied between 5 wt% < Cw < 20 wt%, 5 wt% < CDCM < 45 wt%and 50 wt% < Cb < 90 wt%. The temperature levels were 30 and 50 ◦C.The responses were the water flux through the membrane (Jw) andthe enrichment factor towards water (ˇ). The responses are calcu-lated as Jw = m/t × A and ˇ = yw/xw, where m is the amount of watercollected in the permeate during a certain time t, A is the effectivemembrane area, xw and yw the weight fraction of water in the feedand permeate. The model recommended was quadratic as follows:
Y = b0 +n∑
i=1
biXi +n∑
i=1
biiX2i +
n−1∑
i=1
n∑
j=i+1
bijXiXj (1)
where Y is the predicted response, bo the constant coefficient, bi the
linear coefficients, bij the interaction coefficients, bii the quadraticcoefficients and Xi, Xj are the coded levels of the factors studied. Inorder for the experimental plan to be viable, the reference mixtureand the candidate set selected by the software were altered man-ually. Runs were carried out in randomised order to break down(g mol−1) Solubility in water at 20 ◦C � (g m−3) p01 at 40 ◦C (kPa)
4 1.32 1.33 102.542 7.70 0.81 2.47
1 7.39
V. García et al. / Journal of Membrane Science 338 (2009) 111–118 113
Table 2D-optimal design arrangements and responses for the removal of water from the water/DCM/n-butanol system by PV.
Experiment number Factorsa Responsesb
T (◦C) Cw (wt%) CDCM (wt%) Cb (wt%) Jw × 105 (kg m−2 s−1) ˇ
N1 30 4.96 4.88 90.16 0.97 19.41N2 30 5.84 42.31 51.86 1.82 18.40N3 30 17.62 4.73 77.65 2.73 5.77N4 30 10.17 4.52 85.31 2.08 9.83N5 30 5.05 29.61 65.34 1.36 9.51N6 30 12.53 16.94 70.53 2.71 8.17N7 50 5.18 42.46 52.36 3.61 27.44N8 50 4.99 4.68 90.33 2.16 23.15N9 50 17.73 4.61 77.66 8.69 6.20N10 50 9.74 4.66 85.59 5.25 11.41N11 50 14.20 4.71 81.09 7.68 7.95N12 50 4.10 17.19 78.71 2.14 28.60N13 50 9.06 27.41 63.53 6.61 12.90N14 40 10.00 16.01 73.99 3.43 10.72N15 40 10.08 18.38 71.53 3.71 10.83N16 40 9.76 18.69 71.54 3.72 11.44N17 40 10.34 18.67 70.97 3.97 10.21N18 40 10.14 18.94 70.91 4.01 10.44N19 40 9.91 18.55 71.54 3.22 10.75N 28 81.49 7.53 7.90N 72
centra
amorwtepTcrwR
todtt9ttbmfi
3
3
fqoqattno
20 50 14.23 4.21 30 4.69 4.
a T, temperature; Cw, initial feed water concentration; CDCM, initial feed DCM conb Jw, water flux through the membrane; ˇ, enrichment factor towards water.
ny occurring systematic time trend. The final experimental designatrix as well as the obtained output responses during the tests
f 240 min are shown in Table 2. The original design that includedeplicates of the reference mixture for estimating the pure erroras augmented with replicates of random runs in order to check
he homogeneity of the error variance.The relationship between thexperimental variables and the responses was investigated usingartial least squares (PLS) with a model selected by the software.he model validation is an important part of the data analysis pro-edure, as the approximating model would give poor or misleadingesults if the fit were inadequate. The goodness of fit of the modelas evaluated by inspecting the coefficients of determination R2,
2adj and, Q2 values. R2, R2
adj and, Q2 indicate the fraction of the varia-ion of the response explained by the model, the fraction of variationf the response explained by the model adjusted for degrees of free-om and the fraction of the variation of the response predicted byhe model, respectively. The model validity was also evaluated byhe analysis of variance (ANOVA). The confidence level used was5%. Finally, the optimal conditions were determined by runninghe optimizer tool of MODDE 8.0 for each system and the assump-ion of the model was also verified. The optimal conditions giveny the software depend on the boundaries of the study. The opti-izer uses a Nelder Mead simplex method with the fitted response
unctions to optimize an overall desirability function combining thendividual desirability of each response.
. Results and discussion
.1. Data analysis and evaluation of the model by RSM
During the conduction of the PV experiments, when DCM wasound in the permeate, its concentration was negligible. Conse-uently, the membrane is considered to be non-permeable to therganic solvent. Results shown in Table 2 were fitted by PLS intouadratic models with four variables. The models were improved by
pplying logarithmic transformation to both responses, water fluxhrough the membrane and enrichment factor towards water. Addi-ionally, the models were simplified by removing the statisticallyon-significant terms from Eq. (1). However, when the exclusionf such terms from the model decreased or kept a constant R2adj
90.59 0.81 20.18
tion; Cb, initial feed n-butanol concentration.
value for any of the responses, the term was included in the model.The statistically non-significant linear terms also remained in themodel when the respective quadratic or interactive effects werestatistically significant. The variable importance plot (Fig. 2) wasused to evaluate the importance of the model terms. The adjust-ment of the model obtained to the target variables for the waterflux through the hydrophilic membrane was excellent, in terms ofR2 (R2 = 0.988), R2
adj (R2adj = 0.982) and Q2 (Q2 = 0.967). The repro-
ducibility obtained for this response was 98.76%. Regarding themodel of the enrichment factor towards water was also excellentlyadjusted. The obtained values of R2, R2
adj, Q2 and the reproducibil-ity were 0.994, 0.991, 0.970 and 99.85%, respectively. According tothe R2 values for both models, more than 98% of the data devia-tion can be explained by the two empirical models. A large valueof R2 does not necessarily imply a good regression model as R2
always increases when adding any variable, significant or not sig-nificant, to the model. R2
adj, however, generally increases whensignificant variables are only added to the model. In our study,R2 values are in good agreement with R2
adj for both models. Con-sequently, all the significant terms are included in the empiricalmodels.
Tables 3 and 4 show the ANOVA for the responses water per-meate flux through the hydrophilic membrane and the enrichmentfactor towards water, respectively. According to the obtained resultsFvalue > Ftabulated and p < 0.05 for both regression models. This meansthat the obtained models for water permeation through the mem-brane and the enrichment factor towards water are both statisticallysignificant with a 95% confidence level in the range studied. Further,the lack of fit of the model for water permeation is not significantwith a 95% confidence level as Fvalue < Ftabulated and p > 0.05. This isto say that the model error is in the same range as the pure erroras is desirable. On the contrary, according to Table 4, the lack offit for the model of enrichment factor towards water is significantwith a 95% confidence level (Fvalue > Ftabulated and p < 0.05) indicat-ing that the model error is significantly larger than the pure error.
However, as stated earlier, a very high reproducibility was obtainedfor this response (99.85%), which means that the pure error is arti-ficially close to zero. In this case, the reproducibility, or pure error,does not represent the true experimental error and the model erroris compared with an unreal pure error. A true lack of fit would114 V. García et al. / Journal of Membrane Science 338 (2009) 111–118
Fig. 2. Variable importance plot (VIP) for the removal of water from the water/DCM/n-butanol system by PV using a PVA membrane.
Table 3ANOVA for the water permeate flux through the hydrophilic membranea.
Source of variation DF SS MS Fvalue Ftabulated (˛ = 0.05)* Probability (p) SD
Total corrected 20 1.57 0.079 0.280Regression 6 1.55 0.259 189.1 2.9 0.000 0.509Residual 14 0.019 0.001 0.037Lack of fit (model error) 8 0.014 0.002 1.7 3.8 0.267 0.041P
iation
hFwamf
J
TA
S
TRRLP
ure error (replicate error) 6 0.006 0.001
a DF, degrees of freedom; SS, sum of squares; MS, mean square; SD, standard dev* 5% significance level.
ave been found if both R2 and Q2 values were small. Additionally,ig. 3 illustrates that the observed responses correlated very wellith the predicted values. Therefore, the models can be considered
dequate for the predictions and for optimisation. The quadraticodels describing the correlation between the responses and the
actors were as follows:
w = −4.454 + 0.179T − 0.023CDCM − 0.067Cb + 0.191Cw
− 0.0004C2DCM − 0.006C2
b − 0.053C2w − 0.009CDCMCb
+ 0.023CDCMCw + 0.035CwCb (2)
able 4NOVA for the enrichment factor towards the permeation of water.
ource of variation DFa SSb MSc
otal corrected 20 0.821 0.041egression 6 0.816 0.136esidual 14 0.005 0.0004
ack of fit (model error) 8 0.005 0.001ure error (replicate error) 6 0.0004 0.0001
a DF, degrees of freedomb SS, sum of squaresc MS, mean square.d SD, standard deviation.* 5% significance level.
0.031
.
ˇ = 1.056 + 0.031T + 0.070CDCM + 0.055Cb − 0.190Cw
− 0.007C2DCM + 0.003C2
b + 0.030C2w + 0.005CDCMCb
− 0.023CDCMCw − 0.017CwCb (3)
The coefficients of the terms included in the models are notexact numbers, but terms which contain uncertainty due to theexperimental errors. The study of the coefficients can give an esti-mate of the effects of the factors on the responses. A positivecoefficient indicates synergistic effect, while a negative coefficient
Fvalue Ftabulated (˛ = 0.05)* Probability (p) SDd
0.203361.9 2.9 0.000 0.369
0.01910.1 3.8 0.006 0.008
0.019
V. García et al. / Journal of Membrane Science 338 (2009) 111–118 115
F he men
itDhmoccin
F5
ig. 3. Plot for relationship between observed and predicted (a) water flux through tumbers presented in Table 2.
ndicates an antagonistic effect. Eqs. (2) and (3) show that, withhe exception of the coefficients of the temperature term, and theCM quadratic term, the coefficients of the factors of one modelave opposite signs in the other. According to the models, theost important terms when describing the effect of the factors
n the water flux and enrichment factor are initial feed wateroncentration and temperature, and initial feed water and DCMoncentration, respectively. The presence of significant square andnteraction terms confirms the existence of quadratic behaviour andon-linear blending effects.
ig. 4. Contour plot for the effects of the initial feed concentration of water (Cw), DCM (CD
0 ◦C.
mbrane and (b) enrichment factor towards water. The numbers indicate experiment
3.2. Response surfaces analyses
From the empirical models described in Eqs. (2) and (3)response surface plots were developed, which can be usedfor a better understanding of the effect of the factors on theresponses. The 3D contour plots describe the variation of both
responses, the water flux through the membrane (Fig. 4) andthe enrichment factor towards water (Fig. 5) as a function ofthe mixture composition at the studied temperatures (30, 40 and50 ◦C).CM) and n-butanol (Cb) on the water flux through the PVA membrane at 30, 40 and
116 V. García et al. / Journal of Membrane Science 338 (2009) 111–118
F M (CD
moswkaocttoceAiod[
mtipcitccqadfFcflpnlcTai
ig. 5. Contour plot for the effects of the initial feed concentration of water (Cw), DC
According to the contour plots the permeate flux through theembrane was positively affected by temperature. This common
bservation is due to the enhancement of the PV driving forceince the vapour pressure and hence the chemical potential increaseith temperature. Diffusion is also favoured because of the greater
inetic energy of the penetrating compounds and the larger avail-ble free volume to permeate through [24]. In addition, the viscosityf the feed liquid decreases with temperature due to weakenedohesive forces between the molecules in the liquid [25]. Therefore,he mass transfer is facilitated. Further, the influence of tempera-ure on the enrichment factor depends on the temperature impactn the sorption of all the species in the membrane and the relativehanges in their diffusion coefficient. Fig. 5 indicates the negligibleffect of the temperature on the enrichment factor towards water.n increase of water flux through the membrane and an insignif-
cant increase of the enrichment factor towards water were alsobserved by Veerapur et al. when separating water from water/1,4-ioxane mixtures by PV using PVA–zeolite T composite membranes26].
Additionally, according to Fig. 4 the flux of water through theembrane benefited from the increase of water concentration in
he feed. The flux enhancement was mainly due to the fact that anncrease in feed concentration caused a raise in activity and partialressure, hence enhancing the PV driving force. As for the otheromponents of the mixture, an increase in their feed compositionmplied a decrease in water flux. The flux decline was caused byhe main constraint of mixture designs. The total concentration ofomponents needs to be 100%. As a result the formulation factorsannot be manipulated completely and independently. In conse-uence, an increase of n-butanol and DCM concentration implieddecrease in water feed concentration, hence decreasing the PV
riving force. The decrease in water flux when decreasing the watereed concentration also indicated the lack of membrane swelling.rom the subplots of Fig. 4 it may be noticed that the initial con-entration of DCM needed to be highly altered to affect the waterux negatively. This observation suggested the DCM being the com-ound least affecting the response, probably caused by the DCMot permeating through the membrane. Eq. (2) also confirms the
atter observation as the coefficient of the initial DCM feed con-entration is the smallest of all the concentration linear terms.he effects of the interaction terms on the response were evalu-ted in the region of the contour plot where the terms involvedn the interaction are altered while keeping the third one con-
CM) and n-butanol (Cb) on the enrichment factor towards water at 30, 40 and 50 ◦C.
stant. As shown in Fig. 4, the interaction between n-butanol andDCM seemed to affect the response the least. The minor impactof the latter term compared to the other interaction terms is alsoreflected in Eq. (2). As for the other interaction terms, the waterflux through the membrane was increased by increasing the waterconcentration while reducing the DCM feed concentration. Thecolours of the plots in Fig. 4 suggest that the optimum regionfor obtaining the highest water flux though the membrane wouldbe at the highest water feed concentration within the studiedrange.
Fig. 5 indicates that the improvement of the enrichment factortowards water was achieved by the decrease of water concentra-tion in the feed. Further, the increase of the DCM concentration andn-butanol in the feed enhanced the enrichment factor. In principle,an increase in water feed concentration implies an enhancementof the water flux through the membrane and a decrease of thepermeation flux of n-butanol. As a result, the enrichment factortowards water should increase with the increase of water feed con-centration. However, the opposite effect was perceived, probablyas a consequence of the definition of the enrichment factor itself.The observed decrease may have been caused to a higher extendby the increase of the denominator of the ratio rather than by aneffect of the relative fluxes of the compounds on the enrichment fac-tor. Regarding the interaction terms, the interaction between bothorganic solvents seemed to be the least term affecting the enrich-ment factor. Eq. (3) also confirms the latter observation. The coloursof the plots in Fig. 5 suggest that the optimum region for obtainingthe highest enrichment factor towards water is at low water feedconcentrations and high concentrations of organic solvents withinthe studied range.
3.3. Optimal conditions and verification of the model
The last step of the RSM is to obtain the optimal conditions forthe removal of water from the ternary system by PV. An optimumPV dehydration process is the process that exhibits the maximumachievable flux of water through the membrane and enrichmentfactor towards water. Comparing Figs. 4 and 5 the competition
between responses becomes evident, i.e., improving one responsewill have an opposite effect on the other one. For optimisation aims,the software balances the global information of the model in termsof response surfaces, main effects and interaction among variablesto obtain the best conditions for the process as a whole.brane
4bstaoprc
mmiam7rtaq
4
ifidficbittca4td
A
t(nfiFi
A
AbbbbBCDDFF
[
[
[
V. García et al. / Journal of Mem
The optimal conditions were found at a feed temperature of9 ◦C and at an initial feed concentration of water, DCM and n-utanol of 6.64, 41 and 52.36 wt%. PV of the water/DCM/n-butanolystem with conditions close to the optimal, initial feed concen-ration of water, DCM and n-butanol of 6.51 ± 0.10, 38.57 ± 0.55nd 54.92 ± 0.49 wt%, respectively, were conducted. The water fluxbserved, 4.39 × 10−5 ± 0.20 × 10−5 kg m−2 s−1, was in the rangeredicted by Eq. (2): 4.04 × 10−5 to 5.07 × 10−5 kg m−2 s−1. Theesulted enrichment factor towards water, 17.95 ± 0.25, was verylose to those predicted by Eq. (3): 18.80–21.18.
Additional experiments with different conditions than the opti-al were carried out in order to further assess the validity of bothodels. When the PV of the mixture water/DCM/n-butanol with a
nitial feed concentration of about 15, 5 and 80 wt%, respectivelyt 50 ◦C, the water flux through the membrane and enrich-ent factor obtained were 6.94 × 10−5 ± 0.11 × 10−5 kg m−2 s−1 and
.01 ± 0.05, respectively. The experimental values observed wereeasonably close to the range predicted by the models; 7.10 × 10−5
o 8.51 × 10−5 kg m−2 s−1 for the water flux through the membranend 7.01–7.70 for the enrichment factor towards water. Conse-uently, the validity and adequacy of the models were verified.
. Conclusions
This study shows that the PVA–titanium dioxide membranes able to separate water from DCM and n-butanol by PV at dif-erent operating conditions. The membrane was concluded to bempermeable to DCM in the studied conditions. Further, RSM wasemonstrated to be effective and reliable in proposing models fortting the experimental data, predicting and finding the optimalonditions in the PV process. The water flux through the mem-rane was established to be positively affected by temperature and
nitial water feed concentration. Additionally, the enrichment fac-or towards water was concluded to be positively influenced byhe increase of the VOCs initial feed concentration. The optimalonditions established by RSM were a feed temperature of 49 ◦Cnd initial feed concentration of water, DCM and n-butanol of 6.64,1 and 52.36 wt%, respectively, which were confirmed experimen-ally. The results reported in this paper can contribute to further theevelopment of recovery of VOCs from waste streams.
cknowledgements
The authors would like to thank Ms. Auli Turkki for her assis-ance in conducting the experiments. The Academy of Finlandproject no. 111416), the Maj and Tor Nessling foundation (projecto. 2008394) and the Thule Institute are acknowledged for theirnancial support. Further, the authors would like to thank the GKSSorschungszentrum for generously supplying the membranes usedn this study and providing the SEM photographs.
ppendix A. Nomenclature
effective membrane area (m2)o constant coefficienti linear coefficientii quadratic coefficientij interaction coefficientP boiling point (◦C)
initial feed concentration (wt%)determinant
F degrees of freedomtabulated Fisher test critical valuevalue Fisher test calculated value
[
[
Science 338 (2009) 111–118 117
J permeation flux (kg m−2 s−1)m mass of permeate (kg)MS mean squareMW molecular weight (g mol−1)N number of experiments of the designp p-value, probabilityp0 vapour pressure of the pure component (kPa)Q2 response variation percentage predicted by the modelR2 coefficient of determinationR2
adj adjusted coefficient of determinationSD standard deviationSS sum of squaresT temperature (◦C)t experimental time interval (s)x liquid weight fraction (kg kg−1)Xi, Xj factor coded levelsY predicted responsey permeate weight fraction (kg kg−1)Z design matrix
Greek letterˇ enrichment factor
SuperscriptT transpose
Subscriptsb n-butanolDCM dichloromethanew water
References
[1] IARC, Re-evaluation of some organic chemicals, hydrazine and hydrogen per-oxide, in: Proceedings of the IARC Working Group on the Evaluation ofCarcinogenic Risks to Humans, Lyon, France, February 17–24, 1998, pp. 251–315.
[2] L. Scheflan, M.B. Jacobs, The Handbook of Solvents, D. Van Nostrand CompanyInc., New York, 1953.
[3] N. Sonoyama, K. Ezaki, T. Sakata, Continuous electrochemical decompositionof dichloromethane in aqueous solution using various column electrodes, Adv.Environ. Res. 6 (2001) 1–8.
[4] J.G. Wijmans, R.W. Baker, The solution-diffusion model: a review, J. Membr. Sci.107 (1995) 1–21.
[5] B. Smitha, D. Suhanya, S. Sridhar, M. Ramakrishna, Separation of organic–organic mixtures by pervaporation—a review, J. Membr. Sci. 241 (2004) 1–21.
[6] D. Panek, K. Konieczny, Preparation and applying the membranes with carbonblack to pervaporation of toluene from the diluted aqueous solutions, Sep. Purif.Technol. 57 (2007) 507–512.
[7] H.I. Shaban, Pervaporation separation of water from organic mixtures, Sep.Purif. Technol. 11 (1997) 119–126.
[8] W.F. Guo, T.S. Chung, T. Matsuura, Pervaporation study on the dehydration ofaqueous butanol solutions: a comparison of flux vs. permeance, separationfactor vs. selectivity, J. Membr. Sci. 245 (2004) 199–210.
[9] T.A. Peters, N.E. Benes, J.T.F. Keurentjes, Hybrid ceramic-supported thin PVApervaporation membranes: long-term performance and thermal stability inthe dehydration of alcohols, J. Membr. Sci. 311 (2008) 7–11.
10] Y.Q. Dong, L. Zhang, J.N. Shen, M.Y. Song, H.L. Chen, Preparation of poly(vinylalcohol)–sodium alginate hollow-fiber composite membranes and pervapora-tion dehydration characterization of aqueous alcohol mixtures, Desalination193 (2006) 202–210.
[11] M.C. Burshe, S.B. Sawant, J.B. Joshi, V.G. Pangarkar, Sorption and through PVApermeation of binary water–alcohol systems membranes crosslinked with mul-tifunctional crosslinking agents, Sep. Purif. Technol. 12 (1997) 145–156.
12] T.A. Peters, C.H.S. Poeth, N.E. Benes, H.C.W.M. Buijs, F.F. Vercauteren, J.T.F.Keurentjes, Ceramic-supported thin PVA pervaporation membranes combin-ing high flux and high selectivity; contradicting the flux-selectivity paradigm,J. Membr. Sci. 276 (2006) 42–50.
13] S. Sommer, T. Melin, Performance evaluation of microporous inorganic mem-branes in the dehydration of industrial solvents, Chem. Eng. Process. 44 (2005)1138–1156.
14] R. Malaisamy, G. Annadurai, D. Mohan, Performance optimization of poly-sulfone ultrafiltration membranes for riboflavin separation using designexperiments, Bioprocess Eng. 22 (2000) 227–232.
15] A. Idris, A.F. Ismail, M.Y. Noordin, S.J. Shilton, Optimization of cellulose acetatehollow fiber reverse osmosis membrane production using Taguchi method, J.Membr. Sci. 205 (2002) 223–237.
1 bran
[
[
[
[
[
[
[
[
[
[
[
and Environmental Engineering at the University of Oulu.Her primary research areas include heterogeneous cataly-sis, advanced separation technologies, reactor separationunit and process design, environmental and green engi-neering, air pollution control engineering, and researchmethodology.
18 V. García et al. / Journal of Mem
16] A. Idris, F. Kormin, M.Y. Noordin, Application of response surface methodologyin describing the performance of thin film composite membrane, Sep. Purif.Technol. 49 (2006) 271–280.
17] M. Khayet, C. Cojocaru, G. Zakrzewska-Trznadel, Response surface modellingand optimization in pervaporation, J. Membr. Sci. 321 (2008) 272–283.
18] F. Xiangli, W. Wei, Y. Chen, W. Jin, N. Xu, Optimization of preparation con-ditions for polydimethylsiloxane (PDMS)/ceramic composite pervaporationmembranes using response surface methodology, J. Membr. Sci. 311 (2008)23–33.
19] M.N. Hyder, R.Y.M. Huang, P. Chen, Pervaporation dehydration of alcohol–watermixtures: optimization for permeate flux and selectivity by central compositerotatable design, J. Membr. Sci. 326 (2009) 343–353.
20] V. García, E. Pongrácz, E. Muurinen, R.L. Keiski, Pervaporation ofdichloromethane from multicomponent aqueous systems containingn-butanol and sodium chloride, J. Membr. Sci. 326 (2009) 92–102.
21] V. García, E. Pongrácz, E. Muurinen, R.L. Keiski, Recovery of n-butanol from saltcontaining chemical wastewaters by pervaporation, Desalination 241 (2009)201–211.
22] R.W. Myers, D.C. Montgomery, Response Surface Methodology, Process andProduct Optimisation Using designed Experiments, Second ed., John Wiley &Sons Inc., New York, 2002.
23] L. Eriksson, E. Johansson, C. Wikström, Mixture design—design generation, PLSanalysis, and model usage, Chemometr. Intell. Lab. 43 (1998) 1–24.
24] C.C. Pereira, C.P. Ribeiro Jr., R. Nobrega, C.P. Borges, Pervaporative recovery ofvolatile aroma compounds from fruit juices, J. Membr. Sci. 274 (2006) 1–23.
25] A.F.M. Barton, CRC Handbook of Solubility Parameters and Other CohesionParameters, Practical Liquid Solubility Scales, CRC, Boca Raton, FL, 1983.
26] R.S. Veerapur, M.B. Patil, K.B. Gudasi, T.M. Aminabhavi, Poly(vinyl alcohol)–zeolite T mixed matrix composite membranes for pervaporation separationof water + 1,4-dioxane mixtures, Sep. Purif. Technol. 58 (2008) 377–385.
Verónica García obtained her M.Sc. (Chemistry) degreein 2001 from the University of the Basque Country, Spain.Since 2003 she has been conducting her Ph.D. studies inthe Mass Transfer Process Laboratory, Department of Pro-cess and Environmental Engineering at the University ofOulu in the field of separation of volatile organic com-pounds by pervaporation.
Junkal Landaburu-Aguirre obtained her M.Sc. (Chem-
istry) degree in 2004 from the University of the BasqueCountry, Spain. Since 2006 she has been working as a Ph.D.student in the Mass and Heat Transfer Process Laboratory,Department of Process and Environmental Engineering atthe University of Oulu regarding the removal of heavy met-als by ultrafiltration.e Science 338 (2009) 111–118
Eva Pongrácz is a Doctor of Science in Technology (Uni-versity of Oulu, Finland, 2002) and Doctor Universitatis inEnvironmental Economics (Budapest University of Tech-nology and Economics, Hungary, 1996). She is the Directorof the Centre for Northern Environmental Technology atOulu University, and Docent of industrial environmentalengineering.
Paavo Perämäki obtained his Ph.D. from the University ofOulu in 1992. He has been a professor of chemistry at OuluUniversity since 1993. His primary research areas includesample preparation and pretreatment methods for traceelement analysis and method development for trace andultra trace analysis utilizing atomic spectrometric tech-niques (ETAAS, ICP-OES) and ICP-MS. In his research hehas emphasized chemometric tools, especially statisticalexperimental design.
Riitta L. Keiski obtained her Ph.D. from the University ofOulu in 1991. Since 2001 she is professor in Mass and HeatTransfer Process Laboratory in the Department of Process