Performance Assessment of SWRO Spiral-Wound Membrane Modules with Different Feed Spacer Dimensions

: Reverse osmosis is the leading process in seawater desalination. However, it is still an energy intensive technology. Feed spacer geometry design is a key factor in reverse osmosis spiral wound membrane module performance. Correlations obtained from experimental work and computational ﬂuid dynamics modeling were used in a computational tool to simulate the impact of different feed spacer geometries in seawater reverse osmosis spiral wound membrane modules with different permeability coefﬁcients in pressure vessels with 6, 7 and 8 elements. The aim of this work was to carry out a comparative analysis of the effect of different feed spacer geometries in combination with the water and solute permeability coefﬁcients on seawater reverse osmosis spiral wound membrane modules performance. The results showed a higher impact of feed spacer geometries in the membrane with the highest production (highest water permeability coefﬁcient). It was also found that the impact of feed spacer geometry increased with the number of spiral wound membrane modules in series in the pressure vessel. Installation of different feed spacer geometries in reverse osmosis membranes depending on the operating conditions could improve the performance of seawater reverse osmosis systems in terms of energy consumption and permeate quality.


Introduction
Seawater desalination has become one of the main solutions to the worldwide problem of water scarcity. Among the available technologies, reverse osmosis (RO) is widely regarded as being the most efficient and reliable [1]. However, seawater RO (SWRO) is an intensive energy consumption process [2,3]. Optimizing the design and operation of SWRO desalination plants is one of the various ways to reduce the specific energy consumption (SEC) [4][5][6]. A key factor in this respect is the SWRO membrane technology that is used [7,8]. In spiral-wound membrane modules (SWMMs), the permeability coefficients have a significant impact on plant performance in terms of production and solute rejection [9,10]. Important efforts are being made to try to inhibit the effect of fouling on the permeability coefficients during plant operation by improving the pretreatment process [11] and increasing the resistance to fouling [12]. Another important characteristic of SWRO SWMMs is the feed spacer geometry (FSG) [13,14]. The FSG plays an important role in the concentration polarization (CP) phenomena and the pressure drop along the SWMMs [15][16][17].
In the optimal design and operation of SWRO systems, consideration of the water and solute permeability coefficients (A and B, respectively) in conjunction with the FSG is fundamental. the design and performance estimation of SWRO systems. Equations proposed by the membrane manufacturer for pressure losses and PF were used in that work. The same equations were used by E. Ruiz-Saavedra et al. [36,37] in their development of a design method for brackish water RO systems. June-Seok Choi et al. [38] centered their work around the optimization of two-stage SWRO systems. Equations obtained in another published work [39] for pressure losses and k were used. A computational tool for designing brackish water RO systems with SWMMs was developed by A. Ruiz-García and I. Nuez [40]. This tool allowed the use of different FSGs, but simulations using different FSGs in brackish water RO SWMMs were not provided.
To evaluate the impact of the different FSGs in a full-scale PV with commercial SWRO SWMMs requires simple equations that can be used without the high computational requirements of CFD modeling. This is the main motive to use simple correlations such as those proposed by some mentioned authors [18,23,24] are needed. Another important matter that should be considered concerns the membrane permeability coefficients A and B. The values of A and B are key in the optimization of the performance of SWMMs considering different FSGs. In full-scale SWRO desalination plants, slight variations SEC could have large repercussions on operation and maintenance costs. This paper aims to simulate a PV with full-scale SWRO SWMMs under a range of feed flow (Q f ), feed pressure (p f ) and feed concentration (C f ) characteristics, considering different membrane permeability coefficients (A and B) and FSGs.
The following sections are the methodology where characteristics of the selected SWMMs, process modeling and calculation algorithm are described; results and discussion where the performance analysis of the considered PVs that contain SWMMs with different FSGs is presented; and finally the conclusions of this work including future research lines that should be taken into consideration from now on.

Methodology
Two SWRO SWMMs were considered, FILMTEC TM SW30XLE-400 and FILMTEC TM SW30XHR-400 from Dupont R , Wilmington, Delaware, USA. The Water Application Value Engine (WAVE) software from Dupont R company was employed to determine the coefficients A and B of the SWMMs under test conditions. Table 1 shows the calculated coefficients A and B. These SWRO membranes were selected due to their different characteristics (high production (SW30XLE-400) and high rejection (SW30XHR-400)). The mentioned characteristics are key in the operation of full-scale SWRO desalination plants. Both characteristics are related, through the membrane permeability coefficients A and B, with the performance of the desalination process in terms of SEC and permeate quality. To carry out a comparative study of the two full scale SWRO membranes, PVs of 6-8 elements were simulated. A C f range of between 32 and 45 kg m −3 of sodium chloride (NaCl) was used with feed flow (Q f ) and feed pressure (p f ) ranges from 3 to 16 m 3 h −1 and from 40 to 80 bar, respectively. The different FSGs studied by C.P. Koutsou et al. [23] were used. The solution-diffusion transport model [41][42][43], which presumes that the RO membrane does not have porous or imperfections, was utilized. This model is based on considering that each solvent and solute are dissolved in the membrane separately on the feed-brine side and then diffused in individual fluxes through the membrane under the effect of pressure and concentration gradients (Equations (1) and (2)). This is the most extended model and provides results close to the real behavior of RO systems for both seawater and brackish water [44]. The transport equations used the mean values of each SWRO SWMM, and pressure gradients in the permeate carrier as well as temperature changes along the SWRO SWMMs were disregarded. The solvent (permeate) flow is described by Equation (1) and the solute flow by Equation (2).
where Q p is the permeate flow, ∆p is the pressure gradient across the membrane, ∆π is the osmotic pressure gradient across the membrane and S m is the active membrane surface.
where Q s is the solute flow across the membrane, and ∆C is the concentration gradient of solute on either side of the membrane. Coefficient A (Equation (1)) was considered to depend on two variables (Equation (3)): feed temperature (T) and flow factor (related to fouling and operating time (FF)) [40]. The effect of the CP on coefficient A was not considered. As the RO SWMMs get fouled the Q p decreases under a fixed p f which means that the coefficient A decreases. FF is an important parameter that affect the coefficient A due to fouling [45]. Various methods have been proposed to try to predict the value of this parameter [46]. This work regards a comparative study between using different FSGs in two different SWRO SWMMs. It was considered that the flow factor FF was 1 (membrane without fouling). Usually, the FF decreases with operating time as the SWMMs get fouled [45]. A T of 25 • C was considered, so the temperature correction factor (TCF) had a value of 1.
where A 0 is the initial value of A. The net driving pressure (NDP) depends on p f , pressure drop (∆p fb ), permeate pressure (p p ), average osmotic pressure on the membrane surface (π m ) and average osmotic pressure of the permeate (π p ): ∆p fb was calculated as follows [40,47]: where where L is the SWMM length (taken as 1 m), ρ fb is the average feed-brine density (kg m −3 ), M is an empirical parameter, v fb is the average feed-brine water velocity (m s −1 ), d h (m) is the hydraulic diameter of the feed channel, ε is the porosity of the cross-sectional area of the feed channel (assumed 0.89 [18]) and h is the height of the feed channel, which was taken as 7.11 × 10 −4 m (28 milli-inches) for the two SWMMs. In this study, pressure losses in the permeate channel were not considered, and a value of p p = 34, 473.8 Pa (5 psi) was considered [48]. Figure 1 shows the different FSG parameters. The correlations used for λ were those proposed by C.P. Koutsou et al. [23] ( Table 2). λ was multiplied by the parameter K λ , which was introduced by V. Geraldes et al. [19]. Values between 1.9 and 2.9 were obtained in that study.  Due to CP phenomena, the solute concentration on membrane surface increases. This concentration generates a diffusive reverse flow to the feed-brine bulk. Once steady state conditions are established, the PF provides the relationship between the average concentration of solute on the membrane surface (C m ) and the average feed-brine solute concentration (C fb ). Equation (12) was used to calculate π p . This enables calculation of the solute concentration of the permeate (C p ) through Equation (9) [48]: where CF is the concentration factor, π f is the feedwater osmotic pressure and Y the fraction recovery of the SWMM. Equations (13) and (14) were obtained by using the film boundary model for CP [49].
where M s is the molecular weight of NaCl, J p is the permeate flow per unit of S m and k is the mass transfer coefficient, which is obtained by Equation (15) [18]: where a, b and c are parameters, Sc is the Schmidt number, ν fb is the feed-brine velocity (m s −1 ) and η (0.000891 kg m −1 s −1 when T = 25 • C) the dynamic viscosity of pure water. C. P. Koutsou et al. [24] calculated correlations for the Sh for different FSGs (Table 3). Solute diffusivity (D s (m 2 s −1 )) was calculated using the Equation (18) [50]: In order to calculate all the above variables, an algorithm previously proposed by the authors [40] was used and implemented in MATLAB R . The SEC was determined by dividing the energy consumed by the high pressure pump (which was assumed to have 100% efficiency) by the permeate flow. Steps of 0.25 m 3 h −1 , 0.5 bar and 1 kg m −3 were used for Q f , p f and C f , respectively. If some result exceeded the constrains established by the membrane manufacturer (minimum concentrate flow of 3 m 3 h −1 , 15% maximum element recovery), it was removed. L/d = 6 and β = 90 • . The SW30XLE-400 has a higher coefficient A than the SW30XHR-400 (Table 1). Consequently, higher R values are reached with lower p f than with the SW30XHR-400, but the operating window (possible operating points according with the membrane manufacturer constrains) is wider for the SW30XHR-400 than the SW30XLE-400 (Figure 2a,b). This is due to the high permeate production attained by the SW30XLE-400 membrane. As the pressure rises the concentrate flow decreases considerably, reaching the minimum established by the manufacturer with not very high pressures. This factor should be taken into account when this type of membrane is placed in PVs of 7 and 8 elements.    Figure 3a,b show the exponential growth of SEC with the increase of C f . With the increment of C f there was a small increase in the distance of the exponential curves of each FSG. The membrane with the higher coefficient A showed to have a lower SEC. Although, the separation between curves was more pronounced for the SW30XLE-400 than the SW30XHR-400 membrane. This showed that the influence of FSG with the C f was higher for the SW30XHR-400 membrane. This was due to mentioned SWMM allowed to pass less salt (higher C fb ) and spite of having lower coefficient A (which make C fb to decrease), the impact of having lower coefficient B was higher than having a higher A on SEC.  As happened with SEC, C p also showed an exponential growth with the increase of C f for both membranes (Figure 4a,b). Again, slightly bigger differences between curves were reached at higher C f values and were even more pronounced for the membrane with the higher coefficient A (SW30XLE-400). The   Tables 4-6 show the results in terms of SEC, R and C p in four different cases. Cases 1 and 2 with a standard C f and Cases 3 and 4 with a high C f . Table 4 shows the SEC for the two membranes studied in PVs of 6 and 8 SWMMs with different FSGs. The lowest SEC for both membranes and PVs were obtained for L/d = 12 and β = 105 • . As Q f increased the differences between the aforementioned FSG and the rest in terms of SEC grew wider. The highest values of SEC were calculated for L/d = 6 and β = 120 • in both membranes. The difference between maximum and minimum SEC was higher in PVs of 8 SWMMs in both membranes. The FSG had more impact on the SW30XHR-400, with greater differences between the maximum and minimum SEC values than with the SW30XLE-400.  Table 5 shows the R for the two membranes studied in PVs of 6 and 8 SWMMs with different FSGs. The lowest R for both membranes and PVs were obtained for L/d = 12 and β = 105 • as in the SEC. The results obtained for R had the same trend as SEC as both are closely related. Table 6 shows the C p for the two membranes studied in PVs of 6 and 8 SWMMs with different FSGs. Considering the PVs of 6 SWMMs and Cases 1 and 3, the lowest C p for the SW30XLE-400 were obtained for L/d = 6 and β = 120 • , and for the SW30XHR-400 for L/d = 8 and β = 120 • . When Q f was increased, the minimum values of C p were found in FSGs with L/d = 12 and β = 105 • for both membranes and PVs. In Cases 1 and 3, the highest values of C p were obtained for L/d = 6 and β = 90 • in both membranes and PVs, while for Cases 2 and 4, the highest values of C p were obtained for L/d = 6 and β = 120 • in both membranes and PVs. The difference between the maximum and minimum value of C p was higher as Q f was increased for both membranes and PVs.  Tables 7 and 8 show the results in terms of gradients of SEC and C p in four different cases. The C f in these four cases represent the minimum and maximum values of the most common seawater concentrations (32 g L −1 for Cases 5 and 6, and 37 g L −1 for Cases 7 and 8). Table 7 shows the SEC differences between PVs of 6-7 (∆SEC(6 − 7)) and 7-8 (∆SEC(7 − 8)) SWMMs of the SW30XLE-400 and SW30XHR-400 membranes, in different operating points. The highest differences were found in ∆SEC(6 − 7) for both membranes. In all cases, wider differences corresponded to β = 90 • . Depending on the operating point and the membrane, the highest differences were obtained for L/d = 8 and 12. For Case 5 (higher Q f ) the highest differences were obtained in L/d = 8 for the SW30XLE-400 and in L/d = 12 for the SW30XHRE-400. This was due to different permeability coefficients and flow patterns (higher R for SW30XLE-400 implies lower Q fb and Re). The higher the Re the higher the ∆p fb affecting R, SEC and C p . This is the reason why differences in terms of ∆SEC were higher for the SW30XHRE-400. Variations in ∆SEC(6 − 7) (considering the 4 cases) of between 20.2 and 27.3% and of between 10.2 and 17.8% were obtained respectively for the SW30XLE-400 and SW30XHRE-400. ∆SEC(7 − 8) variations of between 23.2 and 38.2% and between 12.7 and 24.9% were calculated for the SW30XLE-400 and SW30XHRE-400, respectively. In terms of C p , higher differences between PVs with 6 and 7 SWMMs and PVs with 7 and 8 elements were obtained for the SW30XHRE-400 than for the SW30XLE-400 (Table 8). This was due to the coefficient B of the SW30XHRE-400 membrane, which has a higher salt rejection. Consequently, C m along the PV was higher (also affected by PF) for this membrane. Considering the 4 cases in terms of percentages, the variations of ∆C p (7 − 6) and ∆C p (8 − 7) were in a range of 4.6-17.5% and 6.3-30.9% for the SW30XLE-400 and in a range of 4.1-18.2% and 6.4-33.4% for the SW30XHRE-400. These percentages in terms of ∆SEC and ∆C p show the impact of FSGs considering different operating points (Cases) in three types of PV (6, 7 and 8 SWMMs). Table 6. C p (mg L −1 ) for PVs of 6 and 8 SWMMs with the two membranes studied with different FSGs.

Conclusions
The impact of different FSGs on SWRO membrane performance with different permeability coefficients was studied. It was observed that the longer the PV the higher the influence of the FSG on SEC. In terms of SEC, the membrane with a lower coefficient A suffered a more pronounced FSG impact. The effect of the FSG increased with Q f . The impact of the FSG on C p was slightly higher for the membrane with the lower coefficient B and increased with Q f , with the differences between the maximum and minimum C p values also increasing for both membranes. Manufacturers of RO SWMMs should take into consideration the installation of different FSGs in the same SWRO membranes. The option of having a membrane with different FSGs could help to improve SWRO plant operation. Normally, membrane manufacturers offer membranes with higher production or higher rejection (different permeability coefficients), different active area (400 or 440 ft 2 ) or feed spacer thickness (28 or 34 milli-inches. . . ) but with a default FSG. It should be noted that this work is based on simulations and that the impact of membrane fouling can have different effects on different FSGs. This aspect was not considered in this study. The decrease of the coefficient A with fouling and operating time as well as due to CP were also not considered as it could be different for each FSG and different operating conditions. In term of costs, it should be considered that manufacturing SWMMs with different FSGs on request could increase the investment cost of the RO system. Operating and maintenance costs regarding the application of different FSGs in SWMMs would depend on the performance decay due to fouling, which depend not only on operating conditions but on the FSGs.

Conflicts of Interest:
The authors declare no conflict of interest.

Abbreviations
The following abbreviations are used in this manuscript: