Journal of Membrane Science 349 (2010) 225–230

Contents lists available at ScienceDirect

Journal of Membrane Science

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

The use of factorial design for modeling membrane distillation

Pelin Onsekizoglu a,∗ , K. Savas Bahceci b , Jale Acar a
a
Department of Food Engineering, Hacettepe University, 06532 Beytepe, Ankara, Turkey
b
Department of Food Engineering, Hitit University, Corum, Turkey

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

Article history: A two-level factorial experimental design was used to investigate the influence of the main operating
Received 27 August 2009 parameters on evaporation flux and soluble solid content of apple juice during concentration through
Received in revised form 26 October 2009 osmotic distillation (OD) and membrane distillation (MD) processes. The factorial models have been
Accepted 22 November 2009
obtained from experimental design to study all interactions among the considered parameters (osmotic
Available online 27 November 2009
agent concentration (0–65% CaCl2 ), flow rate (10–30 L/h) and temperature difference between feed and
osmotic agent (10–30 ◦ C)) and validated statistically by analysis of variance (ANOVA). For both responses,
Keywords:
the osmotic agent concentration was the most influential factor. The magnitude of the main influence of
Osmotic distillation
Membrane distillation
CaCl2 concentration was followed by the temperature difference and flow rate, respectively. The anal-
Factorial design ysis of the experimental responses revealed that CaCl2 concentration and temperature difference had
Osmotic agent significant interactive effects on evaporation flux. All interactions between the studied parameters were
Temperature significant in the case of soluble solid content at the 99% confidence level. Although the interaction terms
have significant effects, their levels were only a small amount compared to linear effects. The predicted
responses were compared with the experimental ones. In general, the predicted values were in reason-
able agreement with the experimental data, further confirming the very good prediction ability of the
models.
© 2009 Elsevier B.V. All rights reserved.

1. Introduction peratures and/or compositions. In both processes, a microporous

Fruit juices are beverages of high nutritional value since they
contain high levels of minerals, vitamins and other beneficial
components for human health. In order to obtain chemical and
microbiological stability as well as reducing the transport, pack-
aging and storage costs, fruit juices are generally concentrated.
During the concentration process, the water should be removed
selectively in order to obtain a product with an appearance and
taste as close as possible to the original juice. However, multi-stage
vacuum evaporation processes are generally used for concentra-
tion, resulting significant loss of aroma compounds, irreversible
alteration of nutritional value and color changes due to high oper-
ation temperatures [1]. One of the solutions to this problem is
the use of alternative processes that avoid high temperatures of
operation, such as the membrane processes. Membrane distillation
(MD) and osmotic distillation (OD) are well known methodolo-
gies having great potential as concentration processes carried out
at atmospheric pressure and temperatures near the ambient tem-
perature [2–4]. Both operations involve microporous hydrophobic
membranes that are in contact with fruit juices at different tem-
∗ Corresponding author. Tel.: +90 312 297 71 20; fax: +90 312 297 21 23.
E-mail address: pelins@hacettepe.edu.tr (P. Onsekizoglu).
hydrophobic membrane is in contact with feed solution on one
side. The driving force for the water transport through the gas phase
immobilized within the pores is a water vapor pressure difference
related with the water activity differences between the juice and
an osmotic agent, in the case of the OD process, or by a temperature
difference for the MD process [5,6].
Response surface methodology (RSM), a collection of statisti-
cal and mathematical techniques, is a useful tool for development,
improvement, and optimization of processes. It is used to exam-
ine the relationship between one or more response variables and a
set of quantitative experimental variables or factors. Meanwhile,
use of RSM has gained prominence in food process design and
optimization owing to the ease of operation, reliability and repro-
ducibility of the model parameters. Nowadays, factorial designs
have proved their usefulness, and are widely used in the statis-
tical planning of experiments to obtain empirical models relating
process response to process factors. Khayet et al. [7] studied RSM
in direct contact membrane distillation using salt (NaCl) aqueous
solutions and investigated the operating factors, namely, the stir-
ring rate, feed temperature, and solute concentration. However, the
effects of main process parameters involved in OD and MD dur-
ing concentration of fruit juices need to be carried out in detail.
The objective of this work is to evaluate the effects of temperature
difference between the feed and permeate side of the membrane,
concentration of the osmotic agent (CaCl2 solution) and flow rate

0376-7388/$ – see front matter © 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.memsci.2009.11.049
226 P. Onsekizoglu et al. / Journal of Membrane Science 349 (2010) 225–230

on the evaporation flux (J) and total soluble solid concentration Table 2
Factors, their coded levels and actual values as used in the design.
(SSC) of apple juice reached after a predetermined period of mem-
brane/osmotic distillation process using factorial design. Variable Symbol Real values of coded levels

−1 0 +1
2. Materials and methods
Temperature difference, T (◦ C) x1 10 20 30
CaCl2 concentration, C (% (w/w)) x2 0 32.5 65
2.1. Experimental set-up Flow rate, Q (L/h) x3 10 20 30

The concentration process was performed using a laboratory-

Table 3
size membrane module (MD 020 CP 2N, Microdyn, Germany) Experimental matrix design and results obtained for each of the response variables
having 40 polypropylene capillaries with 2.8 mm outer and 1.8 mm studied.
inner diameter (Table 1).
Run number Input variables Responses
In the direct contact membrane distillation process, the apple
juice, with an initial concentration of 12◦ Brix, was pumped in the T (◦ C) C (%) Q (L/h) J (kg/m2 h) SSC (%)
tube side and the deionised water was recirculated in the shell side 1 20 32.5 20 0.669 17.76
of the membrane in a countercurrent mode by using peristaltic 2 20 32.5 20 0.619 17.37
pumps (Heidolph PD 5001, Germany). The temperature difference 3 10 65.0 30 1.11 27.00
4 10 65.0 10 0.764 18.00
imposed between the feed and the permeate side of the membrane
5 30 65.0 10 1.164 28.69
at the inlets using two heat exchangers (Lauda E100, Germany) 6 30 65.0 30 1.462 44.11
and in every case, the inlet temperature of the juice was main- 7 30 0.0 10 0.723 17.44
tained constant at 10 ± 1 ◦ C. The temperatures were measured at 8 10 0.0 10 0.064 12.80
9 30 0.0 30 1.074 24.97
the inlets and at the outlets of the membrane module using type
10 20 32.5 20 0.671 18.06
J thermocouples. Samples of the apple juice were taken over time 11 20 32.5 20 0.585 16.85
and the respective concentration was determined using a digital 12 10 0.0 30 0.266 13.84
refractometer (Atago PAL-3, Tokyo, Japan). 13 20 32.5 20 0.642 16.96
The same experimental procedure was used in the osmotic
membrane distillation process, however, in this case, calcium chlo-
centre points to establish the experimental errors). The variable fac-
ride dihydrate (Merck, Germany) was used as stripping solution.
tors with the coded and actual values are presented in Table 2. The
The initial weight of the stripping solution was three times higher
experiments were carried out in randomized run order to deter-
compared to that of the juice in order to prevent a significant dilu-
mine two characteristic responses: evaporation flux (J) and soluble
tion with consequent decreasing of the driving force during the
solid content (SSC). The results were taken after 3 h of concentra-
process. After each experimental run, the membrane module was
tion process under steady state conditions. The flux was relatively
cleaned by a five step cleaning process. First, both sides of the
stable throughout the run even when achieving 44◦ Brix of the feed.
membrane were rinsed with deionised water. Then 1% (w/w) NaOH
Table 3 shows the experimental matrix design and the results
solution was circulated for 1 h at 30 ◦ C. After a short rinsing with
of the response variables studied. The experimental design and
deionised water, a 1% (w/w) citric acid solution was circulated for
analysis of data were done using a commercial statistical package,
1 h at 30 ◦ C. Finally, the circuit was rinsed with deionised water. The
Design-Expert version 7.1 (State-Ease, Inc., Minneapolis, MN).
efficiency of the cleaning was ascertained by checking the mem-
The true response surface can often be approximated over a
brane performance parameters (water flux and solute rejection).
small experimental region by a low-order polynomial. A first-order
The evaporation flux was calculated by measuring the increase in
polynomial model is only able to estimate the main effects of the
weight of the stripping solution or deionized water with a digital
experimental factors and does not account for either interactions or
balance (Ohaus AV8101, Germany).
curvilinear effects. If there is little curvature in the limited region,
a first-order model with interaction is appropriate for modeling.
2.2. Factorial design methodology
Adding interaction terms introduces curvature into the response
function [8,9]. The first-order model with interaction terms pro-
A total of 13 experiments were performed according to a full fac-
posed for each response variable (Yi ) was based on the multiple
torial design with three factors (8 points of the factorial design and 5
linear regression method. The empirical model in terms of coded
factors was:
Table 1
Data sheet of Microdyn—MD 020 CP 2N filter module. Y = ˇ0 + ˇ1 x1 + ˇ2 x2 + ˇ3 x3 + ˇ12 x1 x2 + ˇ13 x1 x3 + ˇ23 x2 x3 (1)

Configuration Tubes and shell where ˇi are the values of the regression coefficients. ˇ0 being
Material the constant term, ˇ1 , ˇ2 and ˇ3 the linear effects, ˇ12 , ˇ13 and
Housing Polypropylene ˇ23 the interaction effects while the x1 , x2 , x3 are the independent
Membrane Polypropylene coded variables (temperature difference, CaCl2 concentration and
Potting Polyurethane flow rate of both feed and distillate streams, respectively).
Membrane and module data Stepwise deletion of terms was applied to eliminate the statis-
Number of capillaries 40 tically non-significant terms. The goodness of fit of the model and
Inner diameter 1.8 mm
significance of each regression coefficient was evaluated by regres-
Pore size 0.2 ␮m
Membrane area inside 0.1 m2
sion analysis and ANOVA. 3D surface plots were generated using
Free flow area 1.0 cm2 Design-Expert software.
Housing dimensions (D × L) 25 mm × 500 mm
Feed flow rate at axial velocity of 1 m/s 360 L/h
3. Results and discussion
Membrane operational data
Max. transmembrane pressure inside to outside 1.6 bar The independent and dependent variables were fitted to the
Processing temperature 5–40 ◦ C
first-order model equation with interaction terms (Eq. (1)) and for
 P. Onsekizoglu et al. / Journal of Membrane Science 349 (2010) 225–230 227

Table 4
Test of significance for regression coefficient.

Model term Ja SSCb

Coefficient estimate Sum of squares p-Value Coefficient estimate Sum of squares p-Value

x1 , T 0.280 0.620 <0.0001 5.45 237.30 <0.0001

x2 , C 0.300 0.700 <0.0001 6.09 296.98 <0.0001
x3 , Q 0.150 0.180 0.0002 4.12 136.08 <0.0001
x1 x2 −0.089 0.064 0.0025 1.51 18.13 0.0003
x1 x3 0.013 0.001 0.4626* 1.62 20.87 0.0002
x2 x3 0.011 0.001 0.5058* 1.98 31.36 <0.0001
a
Transmembrane flux.
b
Soluble solid content.
*
Coefficients with p-values greater than 0.01, indicating they are not significant.

each response variable were examined for goodness of fit. Table 4
presents the regression relationships for each response monitored.
The p values were used as a tool to check the significance of each of
the coefficients, which in turn may indicate the pattern of the inter-
actions between the variables. The smaller the value of p, the more
significant was the corresponding coefficient [10]. The effect of lin-
ear coefficients (x1 , x2 , x3 ) on evaporation flux (J) were significant
(p < 0.01). The interaction of CaCl2 concentration and temperature
difference (x1 x2 ) were also found to be statistically significant at
99% confidence level. However, the other interaction effects (x1 x3
and x2 x3 ) were not significant since p values were greater than
0.01. In the case of SSC, both the linear and interaction coeffi-
cients had significant effects (p < 0.01). The main effects of factors
studied in the system are also demonstrated from Pareto charts
in Fig. 1. According to the Pareto chart, in which the bar lengths
are proportional to the absolute value of the estimated effects with
99% confidence level, it was possible to verify that the CaCl2 con-
centration demonstrates the most significant effect on both the
evaporation flux and soluble solid content. The effect of temper-
ature difference on evaporation flux was close to the effect of CaCl2
concentration while the magnitude of the influence of flow rate was
almost half of their effects. The interaction of CaCl2 concentration
and temperature difference effects were very small in comparison
with linear effects but it was also significant at 99% confidence level.
The interaction of CaCl2 concentration and flow rate effects and
the interaction of temperature difference and flow rate effects on
evaporation flux were not significant. For soluble solid content, the
magnitude of the main influence of CaCl2 concentration was fol-
lowed by the main effect of temperature difference and the main
effect of flow rate, respectively. Although the interaction terms
have significant effects on soluble solid content, their levels were
only a small amount compared to linear effects.
The regression Eqs. (2) and (3) represents the best description
after the elimination of non-significant parameters (p > 0.01) from
the results summarized in Table 4. The final empirical models in
terms of actual parameters were determined as follows:

J = −0.501 + (0.0367 + T ) + (0.0146 × C) + (0.0150 × Q )

− (0.0003 × C × T ) (2)

SSC = 11.555 + (0.0710 × T ) − (0.027 × C) − (0.1086 × Q )

+ (0.0046 × T × C) + (0.0162 × T × Q )

+ (0.0061 × C × Q ) (3)

where T is the temperature difference between the feed and the

permeate side of the membrane, C is the CaCl2 concentration and
Q is the flow rate.
The goodness of fit of the model was evaluated by the coeffi-
cient of determination (R2 ), adjusted-R2 , predicted-R2 , coefficient
of variance (CV), prediction residual error sum of squares (PRESS),
adequate precision and the lack of fit test for the model from the
ANOVA table [11]. Table 5 summarizes the statistics used to test
the adequacy of the model.
The p value for the model was less than 0.01, indicating that the
terms in the model have significant effects on the responses. The

Table 5
Statistics used to test goodness of fit of the models.

J SSC

p-Value of model <0.0001 <0.0001

p-Value of lack of fit 0.2826 0.9612
R2 0.9921 0.9986
Adjusted-R2 0.9876 0.9968
Predicted-R2 0.9625 0.9977
CV 5.57 2.19
Fig. 1. Pareto’s chart of standardized effects for variables using the responses: (a)
PRESS 0.059 1.71
evaporation flux; (b) soluble solid content (J: evaporation flux, SSC: soluble solid
Adequate precision 50.627 86.433
content, C: CaCl2 concentration, T: temperature difference, Q: flow rate).
228 P. Onsekizoglu et al. / Journal of Membrane Science 349 (2010) 225–230

Fig. 2. Response surface plot of evaporation flux as a function of: (a) temperature difference (T) and CaCl2 concentration (C) at fixed flow rate of Q = 20 L/h; (b) CaCl2
concentration and flow rate (Q) at fixed temperature difference T = 20 ◦ C; (c) temperature difference and flow rate at fixed CaCl2 concentration C = 32.5%.

Fig. 3. Response surface plot of soluble solute content as a function of: (a) temperature difference (T) and CaCl2 concentration (C) at fixed flow rate of Q = 20 L/h; (b) CaCl2
concentration and flow rate (Q) at fixed temperature difference T = 20 ◦ C; (c) temperature difference and flow rate at fixed CaCl2 concentration C = 32.5%.
 P. Onsekizoglu et al. / Journal of Membrane Science 349 (2010) 225–230 229

lack of fit which is a measure of the failure of a model to represent between the two circulating solutions approximately doubled the
data in the experimental domain at which points were not included evaporation flux, on the other hand, the increase in brine veloc-
in the regression, also was not significant for both the estimated ity slightly increased evaporation flux. Previous investigators have
models at 95% confidence level. also shown the increase in evaporation flux with the effect of flow
Coefficient of determination (R2 ) is the proportion of variation in rate of both in the feed and distillate streams [17]. The increase in
the response attributed to the model. It is suggested that R2 should evaporation flux with the effect of flow rate can be attributed to the
be close to 1 for a good fit model [12]. The estimated models for both reduction in concentration polarization effect on both the feed and
J and SSC had satisfactory R2 values of more than 90%, however, a the osmotic agent side [14].
large value of R2 does not always imply that the regression model is Fig. 3a depicts the 3D surface plot of final soluble solid concen-
good one. R2 always increases with the addition of a new variable to tration at varying temperature and CaCl2 concentration at fixed
the model, regardless of whether additional variable is statistically flow rate of 20 L/h. The rise in feed temperature results in an addi-
significant or not. Thus, it is preferred to use the adjusted-R2 to tional driving force that works synergistically with driving force
evaluate the model adequacy since it is adjusted for the number of generated due to the concentration gradient [13,15,18]. Higher
terms in the model. The adjusted-R2 should be over 90% indicating temperatures also reduce the viscosity of the feed stream causing
a high degree of correlation between the observed and predicted an increase in the liquid mass transfer coefficient [15]. Fig. 3b shows
values. Table 5 shows that R2 and adjusted-R2 values for the models the 3D surface plot of soluble solid content at varying CaCl2 con-
did not differ dramatically indicating non-significant terms have centration and flow rate at a fixed temperature of 20 ◦ C. The effects
not been included in the model. of both the CaCl2 concentration and flow rate results an increase up
The coefficient of variation (CV) is the standard deviation to and beyond their optima. The effect of CaCl2 was more promi-
expressed as a percentage of the mean and is calculated by dividing nent since the increase in CaCl2 concentration results an increase in
the standard deviation by the mean value and multiplying by 100. driving force due to a decrease in vapor pressure on osmotic agent
As a general rule, CV should not be greater than 10% for a good fit side however the effect of flow rate is related with the reduction
to the selected model and these values were found to be 5.57% and in concentration polarization effect due to the reduction in hydro-
2.19% for J and SSC, respectively. dynamic boundary on both the feed and osmotic agent side. The
Predicted-R2 measures the amount of variation in new data effect of temperature difference on soluble solid content was also
explained by the model. Generally, a number closer to one is more prominent than flow rate (Fig. 3c) because the increase with
preferred. PRESS value is an overall measurement of the discrep- temperature difference is mainly due to the increase of the driving
ancy between the data and the estimation model. The smaller force (vapor pressure difference).
the discrepancy, the better the model’s estimations will be. The 10 additional runs were performed in order to check exper-
discrepancy is quantified in terms of the sum of squares of the imentally the adequacy of the developed empirical models. The
residuals. The adequate precision test measures the signal to noise comparison of predicted responses and the experimental ones are
ratio. It compares the range of the predicted values at the design
points to the average prediction error. Ratios greater than 4 indicate
adequate model discrimination and then the model can be used to
navigate the design space.
The results shown in Table 5 indicate that all the fit indices indi-
cated a good fit to the estimated models for both the flux and soluble
solid concentration.
The main and interaction effects of factors upon the responses
are depicted in the three-dimensional surface plots (Figs. 2 and 3).
Fig. 2a shows the 3D surface plot of flux at varying temperature
difference and CaCl2 concentration at fixed flow rate of 20 L/h. The
effect of CaCl2 concentration was more distinct than temperature
difference. The increase in concentration of osmotic agent solution
results a decrease in vapor pressure on osmotic agent side, which
in turn resulted in an increase in driving force for water transport
across the membrane [13,14]. The temperature difference between
the feed and the permeate side of the membrane has a similar
impact on water transport across the membrane. As the temper-
ature difference increases, so does the vapor pressure difference
as the driving force and thus the evaporation flux rises. However,
the flux increases with temperature according to an Arrhenius-type
equation [13]. Fig. 2b shows the 3D surface plot of flux at varying
CaCl2 concentration and flow rate at a fixed temperature differ-
ence of 20 ◦ C. The CaCl2 concentration showed the most significant
effect on evaporation flux. Flux increased rapidly with the increase
of CaCl2 concentration. Similar results obtained for the variation of
flux with the effects of temperature difference and flow rate. How-
ever, the effect of temperature difference was more prominent on
evaporation flux when compared to flow rate (Fig. 2c). The results
were in agreement with previous studies [15,16]. Hongvaleerat et
al. [15] determined a twofold increase in evaporation flux for a
15 ◦ C increase in the temperature of pineapple juice, whereas the
change in the evaporation flux slightly increased (about 7%) with
the increase in brine velocity. Courel et al. [16] also observed similar Fig. 4. Comparison of predicted and experimental responses: (a) evaporation flux
results. These authors found that a temperature difference of 12 ◦ C and (2) soluble solid content.
230 P. Onsekizoglu et al. / Journal of Membrane Science 349 (2010) 225–230
summarized in Fig. 4. The results indicated that the predicted values
were in reasonable agreement with the experimental data, further
confirming the very good prediction ability of the models.
4. Conclusion
The performance of MD process was modeled and expressed
in terms of flux and soluble solid content of apple juice following
membrane distillation assisted by osmotic effect for a predeter-
mined period. A 23 factorial design with five center points was used
for investigating the effects of temperature difference between the
feed and permeate side of the membrane, concentration of the
osmotic agent (CaCl2 solution) and flow rate. The developed model
equations can be used to predict evaporation flux and soluble solid
as influenced by operating factors studied in this system. Results
have clearly shown that all the linear effects of the factors have
significant influences on the evaporation flux. There was a sig-
nificant interaction between CaCl2 concentration and temperature
difference for evaporation flux. In the case of soluble solid content,
both the linear and interaction coefficients had significant effects at
the 99% confidence level. Taking main and interactive effects into
account, the most significant effect on both the evaporation flux and
soluble solid content was the main effect of CaCl2 concentration.
Acknowledgements
The authors would like to thank The Scientific and Technical
Research Council of Turkey (TUBITAK) (Project number: TOVAG
107O096) and Hacettepe University for financial support to this
research project.
References
[1] A. Cassano, E. Drioli, G. Galaverna, R. Marchelli, G. Di Silvestro, P. Cagnasso, Clar-
ification and concentration of citrus and carrot juices by integrated membrane
processes, J. Food Eng. 57 (2003) 153–163.
[2] E. Curcio, E. Drioli, Membrane distillation and related operations—a review,
Sep. Purif. Rev. 34 (2005) 35–86.
[3] K. Belafi-Bako, B. Koroknai, Enhanced water flux in fruit juice concentration:
coupled operation of osmotic evaporation and membrane distillation, J. Membr.
Sci. 269 (2006) 187–193.
[4] M. Gryta, Osmotic MD and other membrane distillation variants, J. Membr. Sci.
246 (2005) 145–156.
[5] H. Valdes, J. Romero, A. Saavedra, A. Plaza, V. Bubnovich, Concentration of noni
juice by means of osmotic distillation, J. Membr. Sci. 330 (2009) 205–213.
[6] V.D. Alves, I.M. Coelhosso, Orange juice concentration by osmotic evapora-
tion and membrane distillation: a comparative study, J. Food Eng. 74 (2006)
125–133.
[7] M. Khayet, C. Cojocaru, C. Garcia-Payo, Application of response surface method-
ology and experimental design in direct contact membrane distillation, Ind.
Eng. Chem. Res. 46 (2007) 5673–5685.
[8] K. Carley, K.M. Kamneva, N.Y. Reminga, Response Surface Methodology, Center
for Computational Analysis of Social Organizational Systems, CASOS Techni-
cal Report, CMU-ISRI-04-136, Carnegie Mellon University, School of Computer
Science, Pittsburgh, PA, 2004.
[9] Y.H. Hui, Handbook of Food Science, Technology and Engineering, CRC Press,
2006.
[10] A. Haber, R. Runyon, General Statistics, third ed., John Wiley & Sons Inc., New
York, 1977.
[11] R.H. Myers, D.C. Montgomery, Response Surface Methodology: Process and
Product Optimization using Designed Experiments, JohnWiley and Sons, New
York, 2002.
[12] F.J. Xiangli, W. Wei, Y.W. Chen, W.Q. Jin, N.P. Xu, Optimization of preparation
conditions for polydimethylsiloxane (PDMS)/ceramic composite pervapora-
tion membranes using response surface methodology, J. Membr. Sci. 311 (2008)
23–33.
[13] V.D. Alves, I.M. Coelhoso, Mass transfer in osmotic evaporation: effect of process
parameters, J. Membr. Sci. 208 (2002) 171–179.
[14] B.R. Babu, N.K. Rastogi, K.S.M.S. Raghavarao, Mass transfer in osmotic mem-
brane distillation of phycocyanin colorant and sweet-lime juice, J. Membr. Sci.
272 (2006) 58–69.
[15] C. Hongvaleerat, L.M.C. Cabral, M. Dornier, M. Reynes, S. Ningsanond, Con-
centration of pineapple juice by osmotic evaporation, J. Food Eng. 88 (2008)
548–552.
[16] M. Courel, M. Dornier, J.M. Henry, G.M. Rios, M. Reynes, Effect of operating
conditions on water transport during the concentration of sucrose solutions by
osmotic distillation, J. Membr. Sci. 170 (2000) 281–289.
[17] F. Lagana, G. Barbieri, E. Drioli, Direct contact membrane distillation: modeling
and concentration experiments, J. Membr. Sci. 166 (2000) 1–11.
[18] F. Vaillant, E. Jeanton, M. Dornier, G.M. O’brien, M. Reynes, M. Decloux, Concen-
tration of passion fruit on an industrial pilot scale using osmotic evaporation,
J. Food Eng. 47 (2001) 195–202.