Using modelling approach to validate a bench scale forward osmosis pre-treatment process for desalination

Using modelling approach to validate a bench scale forward osmosis pre-treatment process for desalination

Desalination 350 (2014) 1–13 Contents lists available at ScienceDirect Desalination journal homepage: Using modelling...

2MB Sizes 1 Downloads 14 Views

Desalination 350 (2014) 1–13

Contents lists available at ScienceDirect

Desalination journal homepage:

Using modelling approach to validate a bench scale forward osmosis pre-treatment process for desalination François Zaviska, Linda Zou ⁎ SA Water Centre for Water Management and Reuse, University of South Australia, Mawson Lakes Campus, Adelaide, SA 5095, Australia

H I G H L I G H T S • • • • •

Evaluation a batch laboratory-scale FO system as a pre-treatment for the RO process FO can help to avoid RO fouling and achieving higher overall water recovery Modelling takes into account flux, water recovery and the final draw concentration An experimental design methodology has been successfully applied for FO optimisation The optimal FO conditions are determined and validated by real brackish water

a r t i c l e

i n f o

Article history: Received 27 February 2014 Received in revised form 13 June 2014 Accepted 1 July 2014 Available online xxxx Keywords: Forward osmosis Optimization Modelling Brackish water Experimental design

a b s t r a c t Forward osmosis (FO) has recently attracted growing attention in wastewater, brackish groundwater and seawater desalination, and power generation. This study evaluates the potential of using a batch laboratory-scale FO system as a pre-treatment for the reverse osmosis (RO) process. FO is a low pressure-driven process that offers many advantages compared to the conventional pre-treatment for RO especially for brackish water with high potential of scaling and fouling. FO can help to reduce the RO process cost by avoiding RO membrane fouling and achieving higher water recovery. An experimental modelling has been employed to describe the FO process taking into account water flux, water recovery and final draw solution. Based on this experimental modelling, the energy consumption for RO has been estimated. It has been found that the treatment time for the FO process and the initial draw solute concentration are important parameters that have an interrelated effect on FO and RO efficiency. The optimal conditions for this FO pre-treatment process are determined by modelling and are experimentally validated by using real brackish water as feed in a bench-scale FO system. © 2014 Elsevier B.V. All rights reserved.

1. Introduction Over the last decades, as the demand for fresh water has increased and the available resources have decreased, desalination has become one of the most promising methods of producing drinking water. Where sufficient water cannot be provided using conventional resources or by recycling, the desalination of seawater or brackish groundwater offers an alternative solution [1]. Water desalination has been practised since the 1950s but a wider application at that time has been limited by technology, high capital costs and high energy consumption leading to high unit cost compared to conventional processes for freshwater production. Now, many improvements in reverse osmosis (RO) membrane development and recovery of energy from brine depressurisation means that the capital cost and energy consumption of the RO process have been significantly ⁎ Corresponding author. Tel.: +61 8 8302 5489; fax: +61 8 8302 3386. E-mail address: [email protected] (L. Zou). 0011-9164/© 2014 Elsevier B.V. All rights reserved.

reduced, making it a feasible alternative method for freshwater production [2,3]. Among the different desalination technologies, RO represents the most economically and commercially significant technology for seawater and brackish water desalination [4]. Nevertheless, three key obstacles remain in RO technologies: high energy consumption, low water recovery and membrane fouling [5]. Most of the energy consumption in the RO process is related to the application of hydraulic pressure to overcome the osmotic pressure of sea or brackish water. Typically, energy consumption represents 44% of the total water cost of an RO plant [6]. The water recovery of a single-stage RO desalination system ranges from 40 to 60%. Low water recovery produces a large volume of concentrated brine, which needs to be disposed of and causes environmental concerns, especially for inland desalination plants where brine-discharging outlets are unavailable [7]. Finally, RO membranes are highly sensitive to organic, inorganic (scalant) and biofouling. Membrane fouling reduces the RO process efficiency (water recovery), reduces the life span of the membranes and increases the energy cost of the system. An efficient pre-treatment process is needed


F. Zaviska, L. Zou / Desalination 350 (2014) 1–13

2. Materials and methods Nomenclature 2.1. Feed and draw solutions ANOVA CCD CTA DI ECP FD FO HU NTU ICP PRO RO RSM TDS TFC TOC

analysis of variance central composite design cellulose triacetate deionised external concentration polarisation factorial design forward osmosis Hazen units nephelometric turbidity unit internal concentration polarisation pressure-retarded osmosis reverse osmosis response surface methodology total dissolved solids thin-film composite total organic carbon

to keep membranes in good condition and to maintain them for a reasonable period [8]. For treating the feedwater that contains high level of scaling ions, a new membrane process called forward osmosis (FO) appears to be one of the most promising pre-treatments for protecting RO membrane from scaling and fouling by this “difficult” water. FO is an osmotically driven membrane process that takes advantage of the osmotic pressure gradient to drive water across the semipermeable membrane from the feed-solution (low osmotic pressure) side to the draw-solution (high osmotic pressure) side [9–11]. As a result of the very low hydraulic pressure required, FO provides many potential advantages over pressuredriven processes like RO, such as less energy input [12], lower fouling tendency, easier fouling removal [13–15] and higher water recovery [16,17]. The objective of using FO as a pre-treatment is to remove bacteria and viruses as well as organic and inorganic compounds such as polysaccharides, proteins and scaling ions by filtration, as these compounds are largely responsible for RO membrane fouling and scaling. Following the FO process, the solution contains only mostly single species of salt (draw solute) in fresh water, which has low scaling and fouling potential for the RO membrane. Recently, the combination of FO and RO has been used in some laboratory-scale experimental study and in limited industrial applications [18–22] but no systematic analysis has been conducted to optimise the operational parameters. FO pretreatment offers many advantages compared to other pre-treatment but generally results to higher salinity solution that required RO process, and leading to higher energy cost. As a result, FO seems to be a suitable pre-treatment for desalinating water with high fouling and/or scaling potential such as brackish ground water [20]. The objective of this study is to develop a modelling approach that can describe the FO process when it is used as a pre-treatment for brackish water desalination. An experimental design methodology has been used for evaluating the FO process in terms of water flux, water recovery and final draw solution concentration. Considering the integrated FO– RO system, the draw solution, which is diluted by extracted water from the feed (FO process), will be regenerated by using RO process. In a perspective of optimisation, the RO energy consumption has been estimated (by models) using the experimental FO results. Theses developed models are used to optimise water flux and water recovery of FO process and energy consumption of RO process. The optimal operational parameters predicted by the model are validated by the experimental results and the best conditions are applied to real brackish water.

The synthetic feed solution was prepared by dissolving the appropriate quantities of sodium chloride (NaCl) in 1.5 l of deionised water and stirring the mixture for at least 20 min. Two types of salt (NaCl and Na2SO4) were used to prepare the draw solutions. The appropriate amount of salt was added to 1.5 l of deionised water to obtain a salt concentration varying from 0.66 M to 2.34 M. Sodium chloride and anhydrous sodium sulphate were obtained from Rowe Scientific, Australia. Once the optimal operational conditions for this FO process were determined, real brackish groundwater was used as feed to validate these conditions. The feed–water samples were obtained from a groundwater bore at Mawson Lakes (Mawson Lakes, South Australia) and contains of 1.77 ± 0.03 g · L− 1 TDS (total dissolved solids), 0.5 ± 0.3 mg · L − 1 TOC (total organic carbon), 86.5 ± 0.1 mg · L − 1 calcium and 56.2 ± 0.05 mg · L − 1 magnesium. The pH of this groundwater was 7.90 ± 0.25, with a colour and a turbidity of b 1 HU and 5.4 NTU respectively. Samples were collected and stored in polypropylene containers at 4 °C until use. Before each experiment, the temperature of the samples was adjusted to 22 °C and the concentration of TDS was increased to 10 g · L−1 by adding NaCl. 2.2. Bench-scale FO experiments The experiments were conducted in batch mode using a bench-scale FO membrane system as depicted in Fig. 1. The FO system consisted of a flat-sheet membrane contact cell with two identical compartments separated by a FO membrane, two peristaltic pumps, flow metres and two containers: one for the feed and the other for the draw solution. Flatsheet cellulose triacetate (CTA) and thin-film composite (TFC) FO membranes (Hydration Technology Innovations, Albany, Oregon, USA) were used in this study. The circular membrane contact cell has symmetric channels on both sides of the membrane. The FO cell has spiral baffles creating channels with 3.2 cm width. The active surface area of each cell is 105 cm2. Plastic mesh spacers were used in both channels to support the membrane and also to increase turbulence and reduce external concentration polarisation (ECP) on both sides of the membrane. Depending on the mode of filtration, the orientation of the membrane is different. In FO mode, the active layer of the membrane is oriented towards the feedsolution channel, while in PRO mode, the active layer is oriented towards the draw solution. The feed-solution container was placed on a balance (MS 1600l, Mettler-Toledo, Switzerland) to determine water flux by measuring the variation in the weight of the feed solution according to Eq. (1):

Jw ¼

Δweight : water density  effective area  Δtime


Water recovery was used to determine the concentration increase of the feed water and was calculated using Eq. (2):

Water recoveryð%Þ ¼

Vpermeate  100 Vfeed


where Vpermeate and Vfeed are the permeate and initial feed volumes respectively. The reverse salt diffusion has been determined by measuring the concentration of the solute in the feed solution (pure DI water). The concentration of salt in solution has been calculated by conversion of the conductivity of the solution.

F. Zaviska, L. Zou / Desalination 350 (2014) 1–13


Fig. 1. Schematic diagram of the laboratory-scale FO system.

2.3. Model experimental design A model based on experimental design methodology was employed to evaluate the effect of the inherent parameters on the efficiency of the FO desalination process and also to describe and optimise the process in a defined experimental domain. Experimental design methodology is a statistical approach that enables by planning a study or a process (with the minimum of experiments) to meet the specified objectives. The development of empirical models based on experimental data enables a precise description of the process taking into account the eventual interactions between the different factors affecting the process. The methodology allows the statistical deduction of the experimental error and lack of fit of the model, and also provides an optimisation perspective so that whatever value is in the experimental domain can be predicted with accuracy. The model experimental design of the FO process was conducted using a response surface methodology (RSM) and more particularly a central composite design (CCD). This design is created by distributing points uniformly within the space of the coded variable (Xi). Two of the advantages offered by CCD are the opportunity to explore the whole experimental region and the advantage of interpolating the response. The CCD is composed of a factorial design (FD), axial matrix and replicate in the centre of the domain. In this study, three responses were taken into account: water flux (Y1), water recovery (Y2) and the final concentration of the draw solution (Y3). Water flux and water recovery characterise the FO process efficiency and the final draw solution

Table 1 Experimental range and levels of independent process variables. Coded variables (Xi)

X1 X2 X3

Experimental field Factor (Ui)

Min. value Max. value Ui,0 (−1) (+1)

Time (mins) 120 Initial draw sol. conc. (M) 1.0 250 Recirculation flow rate (mL · min−1)

240 2.0 450


180 60 1.5 0.5 350 100

is a determinant parameter for the subsequent RO process (integrated FO–RO system). The influence of the three main parameters inherent in the FO process—time (U1), initial draw solution concentration (U2) and recirculation flow rate (U3)—was investigated. Table 1 shows the experimental domain of the design with values for each variable. The values of the process variables and their variation limits were selected carefully on the basis of previous experiments. The FD is a combination of the extreme values (levels −1 and +1) of each parameter leading to eight (23) experiments. It was used to evaluate the main effects and correlating effects of the factors on the responses. Six additional axial runouts plus six replicates in the centre of the domain were realised to complete the CCD. Axial runouts allow more information to be obtained regarding the effect of each parameter independently, while centrepoint runouts provide more information in the centre of the domain and also determine the experimental error (pure error) of the process. For the axial-runout matrix, α was chosen to have iso-variance properties by using rotation, with α = (Nf)1/4 = 1.682 (Nf being the number of points required for the factorial matrix). A total of 20 experiments were conducted for the response surface modelling (CCD); Fig. 2 shows a representation of the CCD matrix. 3. Results 3.1. Characterisation of the FO system 3.1.1. Effect of the initial draw solution concentration The effect of the initial draw solution concentration was investigated using concentrations of NaCl varying from 0.5 M to 2.5 M. Experiments were conducted over 2 h using a CTA FO membrane, with DI water as the feed solution, without keeping the draw solution concentration constant. The results are presented in Fig. 3. First, it can be observed that water flux increases with the increase in initial concentration of the draw solution. The osmotic pressure is directly proportional to the solute concentration on both sides of the membrane. Secondly, as a result of the dilution of the draw solution over time, the difference in osmotic pressure between the two compartments decreases, leading to a decrease in water flux. The initial draw solution concentration is a


F. Zaviska, L. Zou / Desalination 350 (2014) 1–13

Fig. 2. Cube representation of the CCD experimental matrix.

significant factor in the FO process because of its direct correlation with the osmotic pressure; this parameter is the driving force of the process. 3.1.2. Effect of the type of draw solute The effect of the type of draw solute was investigated by comparing three different types of salts: sodium chloride, sodium sulphate and magnesium sulphate. Water flux and salt leakage were monitored over 2 h of FO filtration of DI water using 1 M of salt as the draw solute. The results are presented in Fig. 4. The average water flux recorded was 22.0, 17.8 and 14.5 L · h− 1 · m−2 for Na2SO4, NaCl and MgSO4 respectively.

Concerning the salt leakage or reverse solute diffusion, a significant difference is evident between the diffusion of monovalent salt such NaCl and divalent salt such MgSO4 and Na2SO4. The reverse solute diffusion for NaCl is about eight times higher than the reverse diffusion of Na2SO4 and 20 times higher than the reverse diffusion of MgSO4. Hancock and Cath showed that divalent ions have much lower reverse permeation rate than monovalent ions [23]. Furthermore, reverse salt diffusion can be related to the diffusion coefficient of the solute which also depends on the size of the solute. NaCl exhibits the highest diffusion coefficient with a value approximately 2 times higher than Na2SO4 and 4 times higher than MgSO4 (1.48 · 10−9, 0.76 · 10−9 and 0.37 · 10−9 for

Fig. 3. Effect of the initial concentration of the draw solute on water flux (CTA membrane, PRO mode).

F. Zaviska, L. Zou / Desalination 350 (2014) 1–13


Fig. 4. Evolution of water flux and salt leakage for three different types of draw solute: NaCl, Na2SO4 and MgSO4 (CTA membrane, PRO mode, 1 M of draw solute, DI water as feed solution).

NaCl, Na2SO4 and MgSO4 respectively) [24]. This characteristic is even more influential when the experiments have been conducted in PRO mode. Sulphate ions have larger hydrated size than chloride ions resulting to lower reverse salt diffusion in this study. Achilli et al. revealed that the reverse salt diffusion through a negatively charged CTA membrane is likely controlled by the anion hydrated size [24]. 3.1.3. Effect of the mode of filtration Owing to the asymmetrical structure of the FO membrane, there are two possible membrane orientations: FO mode and PRO mode. In FO mode, the active selective layer faces the feed solution, while in PRO mode, the active selective layer faces the draw solution. These two modes of filtration were compared using a CTA membrane with 2 M of NaCl as draw solute and DI water as the feed solution. The results are presented in Fig. 5.

The water flux for PRO and FO modes is 31.9 and 21.6 L · h−1 · m−2 respectively. In these conditions, the water flux in PRO mode is approximately 48% higher than in the FO mode. This difference of water flux can be explained by the internal concentration polarisation (ICP). ICP is a critical phenomenon which occurs in the porous layer of the membrane and affects the efficiency of osmotically driven membrane processes. It has been demonstrated that ICP can reduce water flux in FO significantly (over 80%) [25, 26]. Dilutive ICP occurs when the porous layer faces the draw solution (FO mode) and water permeates through the membrane to dilute the draw solution. In contrast, concentrative ICP occurs when the porous layer faces the feed solution (PRO). Since the water passes throw the membrane and the solute from the feed cannot penetrate the active layer, it will result in a concentrative polarisation layer within the pore of the support layer resulting in a concentrative ICP. In this case, DI water was used as feed solution, only negligible

Fig. 5. Water flux evolution as a function of time for FO and PRO modes (CTA membrane, 2 M of NaCl as draw solute, DI water as feed solution).


F. Zaviska, L. Zou / Desalination 350 (2014) 1–13

Fig. 6. Effect of the type of membrane on the evolution of water flux and salt leakage (PRO mode, 1 M of NaCl as draw solute, DI water as feed solution).

concentrative ICP occurs in PRO mode. Zhao et al. [27] have shown that the feed solution and the degree of concentration are the determinant parameters on the efficiency of these two orientations. Water flux is higher in PRO than in FO mode at low feed concentrations but this tendency reverses at high concentrations. As DI water or a low salt concentration was employed for the feed solution, PRO mode was selected for the remainder of this study. 3.1.4. Effect of the type of membrane Two types of FO membrane commercially available from Hydration Technology Innovations were compared in this study. The first type has been used widely in various FO processes and studies, and is made of CTA that is mechanically supported by embedded polyester mesh. The second type has become available more recently and is made of polyamide on a polysulphone porous layer (TFC).

Experiments were conducted over 2 h with 1 M of NaCl as the draw solution and DI water for the feed. The results for water flux and salt leakage are presented in Fig. 6. It can be observed that a TFC membrane elicits better results in terms of water flux and salt rejection than a CTA membrane; this result is in agreement with the membrane manufacturer's specifications. Water flux recorded with the TFC membrane was 34.5% ± 2.4 higher than that recorded with the CTA membrane, with a maximum flux of approximately 54 L · h− 1 · m − 2. Concerning the reverse salt diffusion, the value recorded for CTA was approximately 21.1 mg · min− 1, and for the TFC membrane, approximately 6.83 mg · min− 1 . The salt leakage was three times higher with the CTA membrane than with the TFC membrane. The TFC membrane was selected for the remainder of the study because it exhibits better performance in terms of water flux and salt rejection.

Table 2 Central composite matrix with experimental results. Experimental design Exp. no.

X1 (mins)

X2 (M)

X3 (mL · min−1)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

120 240 120 240 120 240 120 240 79.1 280.9 180 180 180 180 180 180 180 180 180 180

1.0 1.0 2.0 2.0 1.0 1.0 2.0 2.0 1.5 1.5 0.66 2.4 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5

250 250 250 250 450 450 450 450 350 350 350 350 182 518 350 350 350 350 350 350

Y1: water flux (L · h−1 · m−2)

Y2: water recovery (%)

Y3: final draw sol. conc. (M)

11.26 10.41 20.19 17.16 11.91 10.64 23.57 20.54 16.88 13.6 8.77 23.19 14.23 17.9 16.2 15.95 15.65 14.87 14.82 15.37

15.9 29.4 27.3 51.5 16.8 30.1 33.1 57.3 15.7 44.8 18.6 47.2 30.1 38 35.1 33.8 33.2 31.5 31.4 32.6

0.863 0.773 1.550 1.333 0.856 0.769 1.488 1.272 1.297 1.036 0.557 1.569 1.154 1.074 1.108 1.121 1.126 1.141 1.141 1.131

F. Zaviska, L. Zou / Desalination 350 (2014) 1–13

3.2. Modeling of the FO process Owing to the contradictive correlated effect of some factors on the different responses, it was difficult to evaluate precisely the optimal conditions to obtain the highest water flux and water recovery and the lowest final draw solution concentration. Indeed, if we consider the whole FO–RO process, parameters such as the concentration of the draw solute can have the two-sided effect, depending on the response. A high concentration of the draw solution will have a positive effect on the FO system by increasing water flux (see Section 3.1.2) and water recovery but a negative effect on the RO process by decreasing the efficiency and/or increasing the energy cost of the process. The same contradictive effect was observed with the treatment duration. Thus, a critical balance must be achieved in order to establish and maintain efficient pre-treatment and RO processes in terms of energy cost. The objective of this section is to describe by modelling and then optimise the FO process using a RSM in order to find the best experimental conditions. The model experimental design and results recorded for each response are presented in Table 2. Table 2 shows that the highest water flux was recorded as 23.57 L · h− 1 · m− 2 at a high initial concentration of draw solution (+ 1) and a high flow rate (+1), and after a short period of filtration (−1). Concerning water recovery, the best performance was recorded as 57.3% at a high salt concentration (+1) and a high flow rate (+1), and after a long period of filtration (+ 1). For the final concentration of draw solution, the lowest value was recorded during axial runout with the lowest initial concentration of draw solution (− 1.68), and with treatment time and recirculation rate kept at their medium value (0). All three responses are linked but vary in terms of parameter function. Using Table 2, the regression models for these three responses in terms of coded variables have been expressed by the following equations: Y1 ¼ 15:66−1:00X1 þ 4:50X2 þ 1:01X3 þ 0:49X1 X2 þ 0:75X2 X3


Y2 ¼ 32:67−9:09X1 þ 9:16X2 þ 1:94X3 þ 2:70X1 X2 þ 1:25X2 X3


Y3 ¼ 1:12−0:077X1 þ 0:3X2 þ 0:02X3 −0:032X1 X2 −0:014X2 X3 2


þ0:014X1 −0:023X2



where Xi varies from − 1 to + 1, and Y1, Y2 and Y3 are water flux in L · h− 1 · m− 2, the percentage of water recovery and the final draw solution concentration in M respectively. The three models were developed using the Design-Expert 8 software (Stat-Ease 2010). A linear model with two factor interactions (2FI) has been selected to describe the water flux (Y1) and the water recovery (Y2). Concerning the final draw solution concentration (Y3), a second-order polynomial model (quadratic model) was chosen to describe this response. It can be observed from these models that the three responses present the same/mutual interactions in common: a strong interaction between the time and the initial draw concentration (X1X2) and a weaker interaction between the initial draw solution and the recirculation flow rate (X2X3). Indeed, the time of treatment and the initial draw solution concentration are directly correlated. The kinetics of water transfer depends on both parameters. In other words, the effect of time depends on the initial draw solution concentration and vice-versa. Concerning the second interaction, it can be explained by the external concentration polarisation (ECP). ECP is stronger when the concentration is higher and ECP can be decreased by the fluid velocity. In this way, the effect of the recirculation flow rate is even more important at high draw solution concentration. Table 3 shows the analysis of variance (ANOVA) of these three response surface models. The three models exhibit high F-values (the mean square of the model divided by the mean square of the residual) of 267.3, 198.9 and 1649.7 for Y1, Y2 and Y3 respectively, indicating that the models are significant. (According to Design-Expert 8, there is less than a 0.01% chance that F-values this large could occur due to noise.) The ‘lack of fit’ of these three models is not significant. The Rsquare of these models is 0.99, 0.986 and 0.999 for Y1, Y2 and Y3 respectively, indicating that only a small portion of the corrected total sum of squares is not explained by the models (low residual). It can be observed in Eq. (3) that the most important parameter influencing water flux corresponds to the initial concentration of draw solution (X2), and then the time and the recirculation flow rate (X1 and X3). It has been demonstrated previously that water flux is directly proportional to osmotic pressure, which is dependent on the concentration of the draw solute. The time (X1) has a negative coefficient, which means that water flux decreases with the treatment duration. Experiments have been conducted without maintaining the constant draw solution and the feed-solution concentrations; thus, the osmotic pressure from the draw solution compartment decreases with time while it increases inside the feed compartment, resulting in a decrease of water

Table 3 Analysis of variance results for each response mathematical model (Y1, Y2 and Y3). Analysis of variance (ANOVA) Source

Degree of freedom

Sum of square

Mean square


Pr N F

ANOVA for Y1: water flux Model Residual Lack of fit Pure error

5 14 9 5

310.84 3.26 1.67 1.59

62.17 0.23 0.19 0.32

267.28 – 0.58 –

b0.0001 (significant) – 0.7726 (not significant) –

ANOVA for Y2: water recovery Model 5 Residual 14 Lack of fit 9 Pure error 5

2396.5 33.73 2.37 1.003

479.31 2.41 2.633 2.007

198.92 – 1.31 –

b0.0001 (significant) – 0.4009 (not significant) –

ANOVA for Y3: final draw solution concentration Model 7 Residual 12 Lack of fit 9 Pure error 5

1.33 1.38 · 10−3 5.801 · 10−4 8.00 · 10−4

0.19 1.15 · 10−4 8.287 · 10−5 1.6 · 10−4

1649.73 – 0.52 –

b0.0001 (significant) – 0.7928 (not significant) –


F. Zaviska, L. Zou / Desalination 350 (2014) 1–13

Fig. 7. (a) Effect of the initial concentration of draw solution and time on water flux (X3 = 350 mL · h−1) and (b) comparison of the actual and predicted values of Y1.

flux. The recirculation flow rate has a positive effect on water flux. Higher velocity permits a faster renewal of the solution on both sides of the membrane and also limits ECP. From this equation, it is possible to calculate the percentage effect of each factor on the response, according to the following relation: 0

1 2 b i Pi ¼ @X 2 A  100 ði≠0Þ bi


bi represents the estimation of the principal effect of the factor i. The effects of time, initial draw solution concentration and recirculation flow rate represent 4.33%, 87.8% and 4.38% of the investigated response (Y1 ) respectively. It has to be specified that the effect of

these parameters is specific to the defined experimental domain. The response surface was developed as a function of time and initial draw solution concentration while the recirculation flow rate was kept at 350 mL · min−1. Fig. 7 presents the response surface and a comparison of actual (measured) and predicted values for Y1. The similarity/ likeness between the actual and predicted values of the water-flux production shows a high level of model adequacy, which is in accordance with the statistical significance of the model presented in Table 3. Concerning water recovery (Y2), time and the initial draw solution concentration are the most important parameters with pretty much the same effect, followed by the recirculation flow rate, with an effect of 46.1%, 46.8% and 2.10% respectively. Water recovery is directly related to the quantities of water passing through the membrane and is dependent on the duration of filtration. Moreover, owing to the positive influence of the initial concentration of draw solute and the recirculation rate

Fig. 8. (a) Effect of the initial concentration of draw solution and time on water recovery (X3 = 350 mL · h−1) and (b) comparison of the actual and predicted responses.

F. Zaviska, L. Zou / Desalination 350 (2014) 1–13


Fig. 9. (a) Effect of the initial concentration of draw solution and time on the final draw solute concentration (X3 = 350 mL · h−1) and (b) comparison of actual and predicted responses.

on water flux, they also influence water recovery. Fig. 8 presents the response surface for Y2 as a function of time and initial concentration of draw solution, and a comparison of actual and predicted values. Finally, except for X2, X1 and X3 have a negative effect on the final draw solution concentration. In other words, a higher initial draw solution concentration leads to a higher final draw solution concentration, and a longer treatment time and higher recirculation flow rate leads to a lower final draw concentration. The response surface represented by a quadratic model was developed as a function of time and initial concentration of draw solute for Y3. The results are presented in Fig. 9. 3.3. Integration of FO with RO process: theoretical modelling of RO energy consumption In the perspective of coupling FO with RO process, it is important to take into account the energy efficiency. FO offers many advantages compared to other RO pre-treatments but needs the subsequent RO process

to reconcentrate the draw and produce the fresh water, due to the higher salinity of the draw solution (higher osmotic pressure) which often leads to high energy costs. The objective of this section is to evaluate the energy consumption of RO process using the previous FO experimental data by using a theoretical modelling. Fig. 10 shows the integrated FO and RO process. In this simulated FO–RO process, the diluted draw solution from FO process will be pumped with high pressure through the RO membrane to obtain fresh water and will be reconcentrated in order to be reused for FO process. The theoretical model for RO energy consumption determination assumed that 1000 m3/day of diluted draw solution (outlet of FO process) will be treated using 1000 m2 of RO membrane. An energy recovery device (ERD), like a pressure exchanger, will be employed to recover the energy from the high pressure of the RO retentate. In order to maintain the equilibrium in terms of water flux and concentration, the model assumed the same dilutive and concentrative

Fig. 10. Integrated FO–RO system for modelling.


F. Zaviska, L. Zou / Desalination 350 (2014) 1–13

factor for FO and RO respectively. The dilutive factor of the draw is related to the FO process water recovery which has been determined in the previous section. When the water recovery increases, the permeate flux increases leading to higher dilutive factor of draw. The energy consumption will be calculated/evaluated based on the RO pump work and more specifically on the determination of the operating pressure that should be applied to produce the permeate flux resulting from this concentrative factor. The specific energy consumption (SEC) is often used to characterise the energy cost of RO process and correspond to the energy (from the RO pump) needed to produce 1 m3 of permeate at a desired water recovery (Eq. (7)) [28]:

has been estimated using the empirical equations reported by Taniguchi and Kimura [30]: 8

ΠðC; TÞ ¼ ð0:6955 þ 0:0025TÞ  10 

ρ ¼ 498:4M ¼

ΔP  ðQ f ‐ηERD Q d Þ ηpump

Q p Q f ‐Q d Q ¼ ¼ 1− d : Qf Qf Qf

Cf Cr




where Cf and Cr are the feed and retentate concentration respectively. Using Eqs. (7), (8) and (10), the SEC can be expressed as: SEC ¼

ΔP  ðQ f −ηERD Q d Þ : ηp  YRO  Q f


The RO operating pressure, ΔP, has been determined using the following approximated equation of the reverse osmosis flux [28]: Q p ¼ Sm Am ðΔP‐σπÞ


where Sm and Am are the membrane surface area and the membrane water permeability coefficient, respectively, σ is the reflection coefficient (assumed 1 for high rejection membrane) and π is the average of osmotic pressure of the concentrate stream (assuming the permeate osmotic pressure negligible). The osmotic pressure is directly correlated to the TDS concentration and can be expressed as a function of the salt concentration [29]: Πi ¼ fCi :

 ln Π ¼ fC f

For the purpose of simplification, the RO salt rejection (RT) is considered as 1 (permeate concentration fixed at 0). In this case, the water recovery YRO can be given by: YRO ¼ 1−

T is the temperature in degree Celsius and M is dimensionless. The average osmotic pressure can be approximated by the log-mean average along the membrane module [28,31]:


The water recovery of the RO process is defined as: YRO ¼




where ΔP is the pressure (in Pa) applied to the raw feed water, Qf and Qd is the feed and the retentate flow rate respectively, ηpump and ηERD are the RO high pressure pump efficiency and pressure exchanger efficiency respectively. The ηpump and ηERD have been fixed at 80% and 85% respectively. ΔP represents the difference between the water pressures at the entrance of the RO membrane module (Pf) and the raw feed water at the source (P0, assumed at atmospheric pressure). ΔP ¼ P f −P0 :


where Π is the osmotic pressure, C the concentration of salts (TDS), T the temperature, ρ the density and M is given by: M ¼ 1:0069−2:757  10

where Wp is the pump work (in J · s−1), Qp is the permeate flow rate (in m3 · s−1) and ηp is the pump efficiency. Wp is given by: Wpump ¼


pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 248400M2 þ 752:4MC


Wpump SEC ¼ Qp

  C ρ


Different methods have been suggested to calculate the osmotic pressure at high TDS concentration. In this study, the osmotic pressure

 1 1−YT : YT


The SEC for this simulated RO process has been calculated based on the previous FO results (final draw solution and water recovery). The estimated SEC values have been incorporated in the CCD design and a mathematical model has been developed to describe the potential energy consumption of RO process (Y4) in function of FO parameters: 2


Y4 ¼ 3:00−0:37X1 þ 0:57X2 −0:073X3 þ 0:15X0 −0:061X2


where Y4 is the estimated SEC for RO, X1, X2 and X3, the FO treatment time, the initial draw solution concentration and the flow recirculation rate respectively. This model shows very good fitting and correlation between the calculated and the predicted data. If we compare the effect of the different parameters, we can conclude that the initial draw solution will have the most important impact on the energy consumption of the subsequent RO process (with the higher coefficient). However, the higher the initial draw solution concentration, the higher the FO water flux, at the same time, the higher the final draw solution will be too, and RO requires higher pressure to overcome its osmotic pressure. The treatment time of FO is the second most important parameter. Indeed, increasing the treatment time leads to higher water recovery and thus higher dilution of the draw solution. High FO water recovery leads to high RO water recovery (in an equilibriate system) which decreases the energy cost of the RO process. 3.4. Optimisation of the FO process and validation using real brackish water This study aims to determine the optimal operational parameters for the FO process as an efficient pre-treatment option for the RO process. Owing to the contradictive effects of some parameters on the different responses, the determination of the optimal condition cannot be determined easily by using a conventional iterative modelling method. Indeed, increasing the retention time in the membrane module leads to higher water recovery and lower final draw solute concentration (decreased RO energy cost) but decreased water flux as well. High water flux is a deciding condition for the application of this FO process on a large scale, especially in continuous mode. Water flux will determine the membrane surface area needed in FO continuous mode and thus will impact on the investment cost. In the same way, a high initial concentration of draw solute led to higher water flux and higher water recovery but also a high final concentration of draw solution, which incurs higher energy costs for the next step (RO). Thus, a balance needs to be reached to satisfy the objective of this study. Fig. 11 presents the effect of the parameters on the responses (Y1, Y2 and Y4), considering the objective of maximising FO process efficiency

F. Zaviska, L. Zou / Desalination 350 (2014) 1–13


Fig. 11. Effect of the treatment time (X1), the initial draw solution concentration (X2) and recirculation flow rate (X3) on the water flux (Y1), water recovery (Y2) and RO energy consumption (Y4).

(water flux and water recovery) and minimising RO energy consumption. It is important to quantify the effect of these parameters by using the defined experimental domain. The contractive effect of the treatment time and initial draw solution concentration on water flux and energy consumption can be observed. This study presents a modelling method which permits, by applying specific criteria on the responses (using the developed models and Design-Expert 8 software), to evaluate precisely the desired optimal condition. Three responses have been taken into account for the optimisation: the water flux, the water recovery and the energy cost involved in the RO process. The water flux has been maximised (with importance of 2/5) and the energy consumption has been minimised (with importance of 3/5). The optimal responses have been recorded for different water recovery varying between 25 and 55% (Fig. 12):

It can be observed in Fig. 12 that the variation of energy consumption is not constant/linear in function of water recovery varying from 2.67 to 3.14 kWh · m−3. There is an optimal percentage of water recovery that corresponds to the lowest RO energy consumption. In these conditions, depending on the applied criteria, the lowest RO energy consumption was reached at 2.67 kWh · m−3 around 30% of water recovery. Concerning the water flux, the optimal values increase linearly with the increase of water recovery. A changing slope can be observed after 47.5% of water recovery, indicating that an experimental boundary has been reached. Indeed, if we look after the corresponding optimal parameters shown in Fig. 13, this slope break at 47.5% coincides with the evolution of the treatment time, reaching a maximum of 240 min. This limitation (treatment time) has an influence on the initial draw solution evolution and consequently on the water flux and energy consumption. In order to satisfy the applied criteria as much as possible,

Fig. 12. Optimised responses (water flux and energy consumption) in function of water recovery (criteria number 1).


F. Zaviska, L. Zou / Desalination 350 (2014) 1–13

Fig. 13. Optimal parameters (time and initial draw solution) in function of water recovery (criteria number 1).

some compromises have to be done to determine the best operating condition (best desirability). When a boundary condition is reached, the variation of the other parameters is intensified in order to respect the objective of this optimisation based on the applied criteria. Beyond 47.5% of water recovery, the optimization is restricted by the treatment time, resulting not only to higher energy cost but also to higher water flux. The treatment time is a very important parameter in this process and directly relates to the hydraulic retention time and indirectly to the membrane surface area. By extrapolation and considering the integrated FO–RO system, the membrane surface area required for FO process has been calculated in order to produce 1000 m3 · d−1 of draw solution for 30, 40 and 50% of water recovery. The results are recorded in Table 4: FO is a low pressure driven process which generally requires higher membrane surface area than RO. In this study, the estimated FO surface areas are quite small and lower than RO system (1000 m2). Consequently, the investment cost will be lower but the energy consumption for the next step (RO process) will be higher. The initial volume of feed water, the treatment time and the membrane surface area are inter-correlated factors giving the relative water flux in volume of water treated per unit of time and per unit of surface area. In this study, the initial volume of feed and the membrane surface area have been fixed with the treatment time the only varying factor. The limitation of the treatment time has Table 4 Optimal values obtained from the integrated FO–RO modelling based on the applied criteria. Water recovery (%)




Time (min) [NaCl]i (M) Recirc. flow rate (mL · min−1) Water flux (L · h−1 · m−2) Final draw sol. concentration (M) Estimated RO applied pressure (bar) Estimated RO energy cons. (kWh · m−3) Estimated FO surface area (m2) Surface area ratio (FO/RO) Estimated permeate production (m3/h)

179.1 1.28 450 14.41 0.98 52.1 2.673 669.7 0.67 9.83

216.6 1.49 450 15.99 1.06 58.6 2.747 752.1 0.75 12.11

240 1.74 450 17.95 1.16 65.9 2.962 777.8 0.78 13.94

The estimated values are based on 1000 m3/d of RO feed solution and 1000 m2 of RO membrane surface area.

the consequence of limiting water recovery and thus the final draw concentration which both determine the estimated RO energy consumption. This lack of flexibility shows a weakness of the defined experimental domain at high water recovery leading to higher energy consumption for RO. Nevertheless, lower energy cost can be obtained at lower water recovery (where in this applied criteria the time boundary has not been reached). For example, if stronger criteria on minimising energy consumption are applied (importance 5/5), the estimated energy consumption can drop at 2.169 kWh · m−3 using 1.07 M of draw solution, a recirculation flow rate of 450 mL · min−1 and a treatment time of 238 min. In this optimal condition (for this domain), 32.2% of water recovery can be reached, using a FO membrane surface area of 876.6 m2 producing 11.59 L · h−1 · m−2 of permeate. Experiments were carried out using optimal conditions to validate the results predicted by the models. Apart from using synthetic water in the validation experiments, real Mawson Lakes brackish groundwater water sample was also used, this enabled us to evaluate the effects of the naturally occurring organic, inorganic and biological compounds in real brackish water. The optimal conditions were applied to both water samples which TDS levels were adjusted to the same concentrations. The results have been summarised in Table 5. First, it is evident that the experimental results obtained from both samples are very close to those projected by the models, demonstrating the good accuracy of the models (the experimental values are inside the error ranges of the actual values obtained in the centre of the domain). For the experiments conducted using real brackish water, the values of water flux and water recovery are slightly lower than those obtained with synthetic water but slightly higher for the final draw solute concentration. The FO process is thus slightly less efficient when treating real brackish water but these differences are inside the experimental error range of the process. The small difference in efficiency can be explained by the presence of organic, inorganic and biological compounds that cause membrane fouling and thus reduce water flux and water recovery. However, because FO is a low-pressure membrane process, the fouling tendency of FO is much lower than for the RO process [9,15, 32]. In fact, it is still difficult to conclude on the economic feasibility of using FO as pre-treatment of RO process. On one side, FO presents many advantages such as less energy consumption, simplified pretreatment with reduced chemical usage and reduced fouling/scaling

F. Zaviska, L. Zou / Desalination 350 (2014) 1–13 Table 5 Predicted and experimental optimisation results (synthetic and real water) for water flux, water recovery, final draw solution concentration and estimated RO energy consumption.

Predicted Synthetic water Real water

Water flux (L · h−1 · m−2)

Water recovery (%)

Final draw sol. conc. (M)

RO energy cons. (kWh · m−3)

11.59 12.29

32.2 34.42

0.808 0.796

2.169 2.216





and extended life of the subsequent RO process [18]. On the other side, FO requires inevitably high energy consumption for reconcentrate draw by RO and exhibits low water flow rate and flexibility. Owing to these characteristics, FO process is suitable to be hybridized with RO as a pre-treatment for desalinating the water with high scaling and fouling potential [20]. This study reports an optimisation method of the integrated FO–RO process using experimental and theoretical modelling. This method can be applied on larger scale, in continuous mode, and with different FO hydrodynamic conditions (spiral wound membrane). Other parameters such as the quality of feed, the FO membrane surface area or the applied pressure (in a pressure assisted FO process) can be taken into account. 4. Conclusion The objective of this study was to evaluate the possibility of using the FO system as a pre-treatment for the RO process. The different parameters affecting the FO process were investigated using a batch laboratoryscale system. Time and initial concentration of draw solution are important parameters that influence the efficiency of the FO process. However, if we consider this process as a pre-treatment for the RO process, these parameters have contradictive effects. A response surface methodology was used to describe the FO process by modelling; it determined the optimal outcomes in terms of water flux, water recovery and final draw solute concentration prior to RO filtration. A theoretical modelling has been conducted to estimate the energy consumption of RO process using the experimental data from FO process. The optimal conditions have been determined by coupling FO water flux and water recovery and RO energy consumption models. The FO system with lower fouling and scaling tendency appears to be a very efficient onestep RO pre-treatment particularly for desalinating high scaling potential brackish water. The permeate obtained from this pre-treatment is free of organics, scaling agents and biological compounds, and can thus protect RO membranes from fouling as well as achieve higher RO water recovery. This study has demonstrated that the modelling methodology employed for the optimisation of the experimental conditions was successful. The application of the FO process on real brackish water has validated the modelling results. A continuous-mode FO operation will be tested in a future study by directly coupling these two processes (FO– RO) and taking into account the energy cost. Acknowledgements The authors would like to acknowledge the financial support of the National Centre of Excellence in Desalination Australia, which is funded by the Australian Government through the ‘Water for the Future’ initiative. The authors are also grateful to technical advisories and comments received from Dr. Con Pelekani at SA Water, Mr. Peter Nicoll at Modern Water and Dr. Emile Cornelissen at KWR Watercycle Research Institute.


References [1] X. Nicos Tsiourtis, Desalination and the environment, Desalination 141 (2001) 223–236. [2] M. Li, Reducing specific energy consumption in Reverse Osmosis (RO) water desalination: an analysis from first principles, Desalination 276 (2011) 128–135. [3] C. Fritzmann, J. Lowenberg, T. Wintgens, T. Melin, State-of-the-art of reverse osmosis desalination, Desalination 216 (2007) 1–76. [4] A. Maurel, Dessalement de l'eau de mer et des eaux saumâtres et autres procédés non conventionnels d'approvisionnement en eau douce, 2nd edition Lavoisier, 2006. [5] M.A. Shannon, P.W. Bohn, M. Elimelech, J.G. Georgiadis, B.J. Mariñas, A.M. Mayes, Science and technology for water purification in the coming decades, Nature 452 (2008) 301–310. [6] H. Cooley, P.H. Gleick, G. Wolff, Desalination, With a Grain of Salt: A California Perspective, Pacific Institute, 2006. [7] J.R. McCutcheon, R.L. McGinnis, M. Elimelech, Desalination by ammonia–carbon dioxide forward osmosis: influence of draw and feed solution concentrations on process performance, J. Membr. Sci. 278 (2006) 114–123. [8] Jacques Andrianne, Felix Alardin, Thermal and membrane process economics: optimised selection for seawater desalination, Desalination 153 (2002) 305–311. [9] S. Zhao, L. Zou, C.Y. Tang, D. Mulcahy, Recent developments in forward osmosis: opportunities and challenges, J. Membr. Sci. 396 (2012) 1–12. [10] C. Klaysom, T.Y. Cath, T. Depuydt, I.F. Vankelecom, Forward and pressure retarded osmosis: potential solutions for global challenges in energy and water supply, Chem. Soc. Rev. 42 (2013) 6959–6989. [11] T.Y. Cath, A.E. Childress, M. Elimelech, Forward osmosis: principles, applications, and recent developments, J. Membr. Sci. 281 (2006) 70–87. [12] R.L. McGinnis, M. Elimelech, Energy requirements of ammonia–carbon dioxide forward osmosis desalination, Desalination 207 (2007) 370–382. [13] A. Achilli, T.Y. Cath, E.A. Marchand, A.E. Childress, The forward osmosis membrane bioreactor: a low fouling alternative to MBR processes, Desalination 239 (2009) 10–21. [14] E.R. Cornelissen, D. Harmsen, K.F. de Korte, C.J. Ruiken, J.-J. Qin, H. Oo, L.P. Wessels, Membrane fouling and process performance of forward osmosis membranes on activated sludge, J. Membr. Sci. 319 (2008) 158–168. [15] B. Mi, M. Elimelech, Organic fouling of forward osmosis membranes: fouling reversibility and cleaning without chemical reagents, J. Membr. Sci. 348 (2010) 337–345. [16] C.R. Martinetti, A.E. Childress, T.Y. Cath, High recovery of concentrated RO brines using forward osmosis and membrane distillation, J. Membr. Sci. 331 (2009) 31–39. [17] S. Zhao, L. Zou, Effects of working temperature on separation performance, membrane scaling and cleaning in forward osmosis desalination, Desalination 278 (2011) 157–164. [18] D.L. Shaffer, N.Y. Yip, J. Gilron, M. Elimelech, Seawater desalination for agriculture by integrated forward and reverse osmosis: improved product water quality for potentially less energy, J. Membr. Sci. 415–416 (2012) 1–8. [19] T.Y. Cath, N.T. Hancock, C.D. Lundin, C. Hoppe-Jones, J.E. Drewes, A multi-barrier osmotic dilution process for simultaneous desalination and purification of impaired water, J. Membr. Sci. 362 (2010) 417–426. [20] O.A. Bamaga, A. Yokochi, B. Zabara, A.S. Babaqi, Hybrid FO/RO desalination system: preliminary assessment of osmotic energy recovery and designs of new FO membrane module configurations, Desalination 268 (2011) 163–169. [21] N.T. Hancock, P. Xu, M.J. Roby, J.D. Gomez, T.Y. Cath, Towards direct potable reuse with forward osmosis: technical assessment of long-term process performance at the pilot scale, J. Membr. Sci. 445 (2013) 34–46. [22] N.A. Thompson, P.G. Nicoll, Forward osmosis desalination: a commercial reality, IDA World Congress, 4–9 September (2011), Perth, Australia, 2011. [23] N.T. Hancock, T.Y. Cath, Solute coupled diffusion in osmotically driven membrane processes, Environ. Sci. Technol. 43 (2009) 6769–6775. [24] A. Achilli, T.Y. Cath, A.E. Childress, Selection of inorganic-based draw solutions for forward osmosis applications, J. Membr. Sci. 364 (2010) 233–241. [25] G.T. Gray, J.R. McCutcheon, M. Elimelech, Internal concentration polarization in forward osmosis: role of membrane orientation, Desalination 197 (2006) 1–8. [26] J.R. McCutcheon, M. Elimelech, Influence of concentrative and dilutive internal concentration polarization on flux behaviour in forward osmosis, J. Membr. Sci. 284 (2006) 237–247. [27] S. Zhao, L. Zou, D. Mulcahy, Effects of membrane orientation on process performance in forward osmosis applications, J. Membr. Sci. 382 (2011) 308–315. [28] A. Zhu, P.D. Christofides, Y. Cohen, Minimization of energy consumption for a twopass membrane desalination: effect of energy recovery, membrane rejection and retentate recycling, J. Membr. Sci. 339 (2009) 126–137. [29] M. Mulder, Basic Principles of Membrane Technology, Kluwer Academic Publishers, Boston, 1997. [30] M. Taniguchi, S. Kimura, Estimation of transport parameters of RO membranes for seawater desalination, AIChE J 46 (10) (2000) 1967–1973. [31] Standard Practice for Standardizing Reverse Osmosis Performance Data, ASTM international, April 2000. [32] S. Lee, C. Boo, M. Elimelech, S. Hong, Comparison of fouling behavior in forward osmosis (FO) and reverse osmosis (RO), J. Membr. Sci. 365 (2010) 34–39.