Role of Operating Conditions in a Pilot Scale Investigation of Hollow Fiber Forward Osmosis Membrane Modules

Although forward osmosis (FO) membranes have shown great promise for many applications, there are few studies attempting to create a systematization of the testing conditions at a pilot scale for FO membrane modules. To address this issue, hollow fiber forward osmosis (HFFO) membrane modules with different performances (water flux and solute rejection) have been investigated at different operating conditions. Various draw and feed flow rates, draw solute types and concentrations, transmembrane pressures, temperatures, and operation modes have been studied using two model feed solutions—deionized water and artificial seawater. The significance of the operational conditions in the FO process was attributed to a dominant role of concentration polarization (CP) effects, where the selected draw solute and draw concentration had the biggest impact on membrane performance due to internal CP. Additionally, the rejection of the HFFO membranes using three model solutes (caffeine, niacin, and urea) were determined under both FO and reverse osmosis (RO) conditions with the same process recovery. FO rejections had an increase of 2% for caffeine, 19% for niacin, and 740% for urea compared to the RO rejections. Overall, this is the first extensive study of commercially available inside-out HFFO membrane modules.


Introduction
Forward osmosis (FO) is an emerging technology utilizing a gradient of osmotic pressure across the membrane as a driving force for water transport. Pressure-less operation results in a unique process with a low fouling propensity [1], a high rejection of solutes [2], and low energy consumption [3]. Osmotically driven membrane processes have been intensively studied for decades in academia, but it was not until recently that progress in commercialization of this technology has manifested through multiple industrial applications of commercial FO membranes in pilot and full-scale installations (e.g., [4][5][6][7][8][9]). Due to the relatively early stage of this technology's development, FO membranes with diverse form factors can be found on the market. Thus, Fluid Technology Solutions (FTSH 2 O) and Toray offer spiral wound membrane elements, Porifera provides plate-and-frame modules, and Toyobo and Aquaporin A/S both provide hollow fiber modules. Modules offered by Toyobo are based on outer-selective cellulose acetate fibers, whereas Aquaporin A/S manufactures inner-selective biomimetic modified HFFO modules will be referred to as chlorinated membranes and denoted HF-C. They will be compared with unmodified HFFO modules, referred to as original membranes and denoted HF-O.

Membrane Characterization
The morphology of the selective layer was investigated with scanning electron microscopy (SEM). Membrane samples were extracted from dried modules and coated with gold using a Leica EM ACE600 coater (Leica Microsystems, Wetzlar, Germany). SEM images from the cross-section and inner surface of the membranes were taken using a high-resolution QuantaTM FEG 250 SEM microscope (Thermo Fisher Scientific, Erlangen, Germany). The surface charge of the investigated membranes was measured with the streaming potential technique. To facilitate the analysis of the surface charge and due to the equipment available, dried fibers extracted from the HFFO modules and mini-modules (consisting of 300 fibers) were prepared. The mini-modules were only created and tested for the surface charge analysis and the HFFO modules used to extract the fibers were discarded. The inner surface zeta potential was determined with a SurPASS™ 3 electro kinetic analyzer (Anton Paar GmbH, Ostfildern, Germany). The streaming potential of the mini-modules was assessed by flushing 1 mM of KCl through the lumen of the fibers. Calculations for the zeta potential using the streaming potential data were done using the Smoluchowski equation [44]. The water permeability coefficient (A), solute permeability coefficient (B), and structural parameters (S parameter) for the investigated membranes were estimated using an adapted Bui et al. [45] approach, where various concentrations of NaCl, MgCl 2 ·6H 2 O, and MgSO 4 ·7H 2 O were used as DS and water flux (J w ), and the reverse solute flux (J s ) of the two investigated membrane types was measured and used for the calculations. Detailed explanations of the method used for these calculations can be found in the Supplementary Materials Section S2.

FO Tests with HFFO Modules and Performance Evaluation
Two chlorinated membranes (HF-C) and two original membranes (HF-O) modules were individually tested in an FO setup using the Titan OSMO Inspector developed by Convergence Industry B.V. (Enschede, The Netherlands). The schematic flow diagram of the system is shown in Figure 1. The FO system consisted of two diaphragm pumps (model NF 1.6 by KNF Neuberger Inc., Trenton, NJ, USA), each to supply DS and FS solutions to the HFFO module. The flow of the solutions was controlled by two high precision flow meters (IR-OPFlow 100-11, Tecflow International, Velp, The Netherlands). Pressure at the inlets and outlets to the modules were monitored with 4 manometers. The concentrations of the draw solutes in the DS and concentrate solutions (CS) were monitored with two digital conductivity meters (CT) (model LTC 0.35/23 and LGT 1/23, respectively by Sensortechnik Meinsberg, Waldheim, Germany). A digital palette balance (model KFB-TM 1.5T by KERN, Balingen, Germany) was used to measure accumulation of the CS in the tank. Both FS and DS were stored in tanks and their temperature was controlled by heating jackets (IBC-J90 by Frank Berg Supplies, Delft, The Netherlands). At the beginning of each experiment, membrane modules were rinsed with DI water for 5 min through both the shell and the lumen side. The FO setup was then started with the specific set of operational conditions and was running for 45 min since J w and J s /J w were stabilized in the system after a few minutes. Thus, data obtained in 45 min was found to be sufficient for the required analysis. After each experiment, the membrane module was cleaned for 45 min with DI water through both the shell and lumen side. The FO performance of tested modules was analysed by measuring Jw, Js and specific reverse solute flux (Js/Jw). The concentrate weight recorded on the balance was used to calculate Jw applying Equation (1): where QF,in and QF,out represent the volumetric flow rate of the feed and concentrate (feed outflow) respectively of the module and SA represent the surface area of the membrane. Js was measured experimentally using Equation (2): where σ is the average conductivity (µS·cm −1 ) of the concentrate and β is a proportionality coefficient describing the relation between conductivity and salt concentration. Here, β was experimentally evaluated for each salt type used as a DS and equals 0.47, 0.43, 0.49 mg·L −1 per µS·cm −1 for NaCl, MgCl2, and MgSO4, respectively.
Js/Jw can be used to describe the efficiency of the FO process, calculated by division of the Js (Equation (1)) by Jw (Equation (2)).
All experiments were carried out in a single-pass mode following the standard conditions listed in Table 1. Each parameter was modified one-at-a-time, while keeping the rest of the parameters constant. Additionally, for tests with feed flow rates, draw flow rates, and TMP and operation modes, two different FSs, varying in osmotic pressure, were selected ( Table 1). The standard conditions described in Table 1 were based on the recommendations of the HFFO membrane module manufacturer and Jw and Js characteristics of the membranes to avoid damaging the membranes during all operating conditions tested. A detailed list of the parameters modified during the FO performance tests are listed in Table 2.   The FO performance of tested modules was analysed by measuring J w , J s and specific reverse solute flux (J s /J w ). The concentrate weight recorded on the balance was used to calculate J w applying Equation (1): where Q F,in and Q F,out represent the volumetric flow rate of the feed and concentrate (feed outflow) respectively of the module and S A represent the surface area of the membrane. J s was measured experimentally using Equation (2): where σ is the average conductivity (µS·cm −1 ) of the concentrate and β is a proportionality coefficient describing the relation between conductivity and salt concentration. Here, β was experimentally evaluated for each salt type used as a DS and equals 0.47, 0.43, 0.49 mg·L −1 per µS·cm −1 for NaCl, MgCl 2 , and MgSO 4 , respectively. J s /J w can be used to describe the efficiency of the FO process, calculated by division of the J s (Equation (1)) by J w (Equation (2)).
All experiments were carried out in a single-pass mode following the standard conditions listed in Table 1. Each parameter was modified one-at-a-time, while keeping the rest of the parameters constant. Additionally, for tests with feed flow rates, draw flow rates, and TMP and operation modes, two different FSs, varying in osmotic pressure, were selected ( Table 1). The standard conditions described in Table 1 were based on the recommendations of the HFFO membrane module manufacturer and J w and J s characteristics of the membranes to avoid damaging the membranes during all operating conditions tested. A detailed list of the parameters modified during the FO performance tests are listed in Table 2.  For the test with various DSs, four concentrations of each salt type were applied. In order to compare performance of the DS based on the osmotic pressure they generate, various concentrations of the salts were prepared. Table 3 lists the osmotic pressure of the applied DS as a function of the salt concentration. The calculations were performed using the Van't Hoff equation [46]. For all the experiments and both membranes (HF-C and HF-O), a statistical analysis was performed to confirm significant differences between the membranes and the experiments. The probability value (p-value) was calculated for J w and J s /J w and used to confirm the hypothesis explained during examination of the results. Information on the p-values obtained and the statistical analysis used can be found in the Supplementary Materials Section S3.

Forward Solute Rejection in FO and Low Pressure Reverse Osmosis (LPRO)
During the FO process, the permeate flow is diluted with the DS. Therefore, rejection calculations of a target compound must take this dilution into account. Following the literature studies [30,49], if the concentration of the target compound is obtained from the draw outlet flow, instead of the permeate flow, the dilution will create an overestimated rejection, which does not represent the real membrane rejection. Hence, using the mass balance of the system, concentration of the target compound in the permeate is calculated using Equation (3): where c D,out , c D,in and c permeate represent the concentrations of the target compound in the draw outflow, draw inflow, and permeate stream. Q D,out , Q D,in and Q permeate stand for draw outflow, draw inflow and permeate volumetric flow rate. Subsequently, membrane rejection can be calculated according to Equation (4): where R FO represents the membrane rejection in the FO process. A solution of 200 mg·L −1 of caffeine, niacin and urea were used as FS to analyse membrane rejection in both FO and LPRO processes. Table 4 summarizes the testing conditions applied in these experiments. Lastly, tests to investigate correlation between recovery and urea rejection were carried out. Different recoveries were obtained by changing the feed flow rate in the FO set up for the original and HF-C modules ( Table 5).  Niacin and caffeine concentration analysis used in membrane rejection experiments was carried out with a Genesys™ 10S UV-Vis spectrophotometer (Thermo Fisher Scientific, Erlangen, Germany). Both compounds were analysed at 240 nm. Urea analysis was done by the Urea/Ammonia Assay Kit from Megazyme (Bré, Ireland). This kit uses urease enzyme to degrade urea into ammonia, which is later reacted by glutamate dehydrogenase to produce a fluorescent indicator. Results were analysed with the UV-Vis spectrophotometer.
HF-C and HF-O modules were also tested in LPRO process for rejection of caffeine, niacin, and urea. The schematic diagram of the setup is shown in Figure 2. The LPRO system consisted of a gear pump (model GC-M25.JVS.6 by Micropump, Vancouver, WA, USA) suppling the FS, an analogue flow meter, two analogue manometers situated at the inlet and outlet of the modules, and a needle valve for pressure adjustment. The temperature of the FS was controlled and maintained at 25 • C by a heat exchanger immersed in the feed tank. Similarly, during LPRO tests, FS contained 200 mg·L −1 of the niacin, caffeine, or urea. To obtain the same process recovery (23 ± 4%) for both types of studied membrane modules, flow rates of the FS were set to 1 L·min −1 and 0.5 L·min −1 for HF-C and HF-O modules, respectively. For all tests, TMP applied was established at 2 bar, calculated according to Equation (5): where Pin and Pout are the pressures at the inlet and at the outlet of the HFFO modules. After approximately one hour of equilibration, tests were carried out and permeate samples were collected for permeability and rejection measurements. The water permeability coefficient was calculated according to Equation (6): where V is volume of the collected permeate sample and ∆t is sample collection time. The collected permeate samples were used for concentration analysis with the UV-Vis spectrophotometer or urea kit analysis mentioned previously. Rejection of the niacin, caffeine, or urea were calculated with Equation (7): where RRO is the solute rejection in the RO process. For all the experiments, rejection (Jw and Js/Jw) was calculated from the individual tests and later averaged for HF-O and HF-C modules. Standard deviations were calculated from these values obtained from duplicates for each test.

Results and Discussion
In this section, characterization and FO performance tests for two types of HFFO membranes are presented and discussed. The analysis of the HFFO membrane provides an insight into the differences in polyamide layer structure between the HF-C and HF-O membranes. Awareness of the difference in the membrane characteristics helps to understand membrane response to the operating conditions and the outcome in FO performance.

Membrane Characterization
The two types of HFFO modules were analyzed to understand the impact of the chlorination treatment on the polyamide layer. Specifically, the difference in surface charge and surface morphology between HF-C and HF-O membranes are reported here. Furthermore, the results of the mathematical model described in the Supplementary Materials (section S2), used to estimate the intrinsic characteristics of the FO membranes (A, B, and S parameter) for three different draw solutes, are discussed in this section. Similarly, during LPRO tests, FS contained 200 mg·L −1 of the niacin, caffeine, or urea. To obtain the same process recovery (23 ± 4%) for both types of studied membrane modules, flow rates of the FS were set to 1 L·min −1 and 0.5 L·min −1 for HF-C and HF-O modules, respectively. For all tests, TMP applied was established at 2 bar, calculated according to Equation (5): where P in and P out are the pressures at the inlet and at the outlet of the HFFO modules. After approximately one hour of equilibration, tests were carried out and permeate samples were collected for permeability and rejection measurements. The water permeability coefficient was calculated according to Equation (6): where V is volume of the collected permeate sample and ∆t is sample collection time. The collected permeate samples were used for concentration analysis with the UV-Vis spectrophotometer or urea kit analysis mentioned previously. Rejection of the niacin, caffeine, or urea were calculated with Equation (7): where R RO is the solute rejection in the RO process. For all the experiments, rejection (J w and J s /J w ) was calculated from the individual tests and later averaged for HF-O and HF-C modules. Standard deviations were calculated from these values obtained from duplicates for each test.

Results and Discussion
In this section, characterization and FO performance tests for two types of HFFO membranes are presented and discussed. The analysis of the HFFO membrane provides an insight into the differences in polyamide layer structure between the HF-C and HF-O membranes. Awareness of the difference in the membrane characteristics helps to understand membrane response to the operating conditions and the outcome in FO performance.

Membrane Characterization
The two types of HFFO modules were analyzed to understand the impact of the chlorination treatment on the polyamide layer. Specifically, the difference in surface charge and surface morphology between HF-C and HF-O membranes are reported here. Furthermore, the results of the mathematical model described in the Supplementary Materials (Section S2), used to estimate the Membranes 2019, 9, 66 9 of 28 intrinsic characteristics of the FO membranes (A, B, and S parameter) for three different draw solutes, are discussed in this section.

Zeta Potential
Electrostatic interactions between membrane surface and solute contribute to the solute's rejection. Zeta potential measurements are the most common technique for evaluating the surface charge of the membrane. In Figure 3, the inner surface zeta potentials as a function of pH for both HF-C and HF-O membranes are shown. Electrostatic interactions between membrane surface and solute contribute to the solute's rejection. Zeta potential measurements are the most common technique for evaluating the surface charge of the membrane. In Figure 3, the inner surface zeta potentials as a function of pH for both HF-C and HF-O membranes are shown. For both membrane types, zeta potential is reduced with a reduction of pH. The isoelectric point was measured at 3.7 and 2.8 for HF-O and HF-C membranes, respectively. The isoelectric point around pH 2-4 was expected due to the nature of the polyamide and the presence of the carboxylic groups at the surface of the polyamide [50]. The shift in the isoelectric point after membrane chlorination could be explained as a side effect of the chlorination post-treatment. The chlorinationpromoted hydrolysis of the polyamide C-N bond and led to the cleavage of the polyamide layer. Consequently, the hydrolysis resulted in a rise of the carboxyl group number at the surface of the selective layer, and thus, an increase in the negative surface charge manifested by a more negative zeta potential.

Scanning Electron Microscopy
The morphology of the selective layer for both HF-C and HF-O membranes was studied with SEM imaging. SEM images of the inner surface and of the fibers cross-section are shown in Figure 4.  For both membrane types, zeta potential is reduced with a reduction of pH. The isoelectric point was measured at 3.7 and 2.8 for HF-O and HF-C membranes, respectively. The isoelectric point around pH 2-4 was expected due to the nature of the polyamide and the presence of the carboxylic groups at the surface of the polyamide [50]. The shift in the isoelectric point after membrane chlorination could be explained as a side effect of the chlorination post-treatment. The chlorination-promoted hydrolysis of the polyamide C-N bond and led to the cleavage of the polyamide layer. Consequently, the hydrolysis resulted in a rise of the carboxyl group number at the surface of the selective layer, and thus, an increase in the negative surface charge manifested by a more negative zeta potential.

Scanning Electron Microscopy
The morphology of the selective layer for both HF-C and HF-O membranes was studied with SEM imaging. SEM images of the inner surface and of the fibers cross-section are shown in Figure 4.
Based on the cross-section images (Figure 4c,d), the thickness of the polyamide layer was estimated to be 70.8 ± 9.8 nm for HF-O membranes and 49 ± 4.9 nm for the HF-C membranes. In contrast, the inner surfaces of the investigated membranes (Figure 4a,b) did not exhibit any significant difference between each other. Chlorine usage is a common method for disinfection in water treatment as it inhibits the growth of harmful microorganisms [51]. However, it is widely known that polyamide membranes are highly sensitive to chlorine solutions, and Simon et al. [52] highlighted the need for chlorine-resistant membranes for water purification processes. The difference in thickness between the membranes can easily be explained by the chlorine attack of NaClO to the polyamide. Under alkaline conditions, two competing mechanisms might occur in the presence of a chlorine species ( Figure 5). Based on the cross-section images (Figure 4c,d), the thickness of the polyamide layer was estimated to be 70.8 ± 9.8 nm for HF-O membranes and 49 ± 4.9 nm for the HF-C membranes. In contrast, the inner surfaces of the investigated membranes ( Figure 4a,b) did not exhibit any significant difference between each other. Chlorine usage is a common method for disinfection in water treatment as it inhibits the growth of harmful microorganisms [51]. However, it is widely known that polyamide membranes are highly sensitive to chlorine solutions, and Simon et al. [52] highlighted the need for chlorine-resistant membranes for water purification processes. The difference in thickness between the membranes can easily be explained by the chlorine attack of NaClO to the polyamide. Under alkaline conditions, two competing mechanisms might occur in the presence of a chlorine species ( Figure 5). The first mechanism postulated by Orton et al. [53] is known as N-chlorination. Here, the chlorine radicals attack the amidic nitrogen in the polyamide structure, which has a lone electron pair and is prone to share it with the partially positive chlorine radicals. This reaction is reversible and can be further aided in acidic conditions where a direct aromatic substitution will occur, producing a chlorinated aromatic ring (Orton arrangement). Once conditions are alkaline, N-chlorination results in the hydrolysis of the polyamide are promoted by the chlorine attack. In that case, after Nchlorination, polarization of the C-N bond facilitates its hydrolysis by the OHgroups present in the elevated pH of the NaClO solution. This results in the cleavage of the polyamide. As schematically The first mechanism postulated by Orton et al. [53] is known as N-chlorination. Here, the chlorine radicals attack the amidic nitrogen in the polyamide structure, which has a lone electron pair and is prone to share it with the partially positive chlorine radicals. This reaction is reversible and can be further aided in acidic conditions where a direct aromatic substitution will occur, producing a chlorinated aromatic ring (Orton arrangement). Once conditions are alkaline, N-chlorination results in the hydrolysis of the polyamide are promoted by the chlorine attack. In that case, after N-chlorination, polarization of the C-N bond facilitates its hydrolysis by the OH − groups present in the elevated pH of the NaClO solution. This results in the cleavage of the polyamide. As schematically described in Figure 5, the cleavage might result in the reduction of cross-linking and an increase in the number of hydroxy groups in the polyamide structure, which enhances the hydrophilicity of the selective layer. Thus, in accordance with the literature [41,42,52] and our observations, HF-C membranes exhibit reduced polyamide layer thickness, an increase in water permeability, and a reduction in solute rejection. As described by Verbeke et al. [42], severe chlorine-promoted hydrolysis of the polyamide will result in dissolution or separation of the selective layer from the support layer, thereby damaging the membrane. Since the HF-C membranes did not exhibit a drastic decrease in solute retention, and Figure 4a,b showed no difference in membrane surface morphology, we concluded that 20 ppm NaClO solutions can be effectively used to improve J w for FO membranes.

Evaluation of the A, B, and S Parameters
LPRO tests with HF-O and HF-C were carried out at 2 bars of TMP and 500 mg·L −1 NaCl solution as a feed. The LPRO tests resulted in A values of 2.46 ± 0.12 L·m −2 ·h −1 ·bar −1 and 1.59 ± 0.13 L·m −2 ·h −1 ·bar −1 for HF-C and HF-O modules, respectively. Calculated B parameter was measured as 0.18 ± 0.03 L·m −2 ·h −1 and 0.15 ± 0.01 L·m −2 ·h −1 for HF-C and HF-O modules, respectively. The LPRO findings agree with our expectations, and hollow fibers after chlorination indeed exhibit higher water permeability and higher solute (NaCl) permeability. In order to estimate the A, B, and S parameters in the FO process, the modified Bui et al. [45] model for randomly-packed bundle of fibers was applied. Results obtained from FO tests with 4 different concentrations of NaCl, MgCl 2 , and MgSO 4 as draw solutions were used as inputs for estimation of the A, B, and S parameters. The summary of the results generated is listed in Table 6. Please note that the applied method is quite sensitive to the quality of the experimental data; experimental errors in J w and J s values may lead to calculations of A, B and S parameters that are not feasible. For example, the tests with the HF-C module and NaCl as a draw solution obtained an A value of 4.04 L·m −2 ·h −1 ·bar −1 , which is far from the results generated with the same membrane for tests with MgCl 2 and LPRO. Therefore, the A value was fixed to a more feasible 2.46 L·m −2 ·h −1 ·bar −1 (LPRO results), and the calculation was run again. The calculation resulted in a slightly higher root mean square error (RMSE) value (from 0.77 to 1.22), a slightly lower R 2 -J w (from 0.98 to 0.92) and significantly better R 2 -J s (from 0.79 to 0.95). The fact that the RMSE was reduced and R 2 -J s was improved indicates that the method requires subjective judgment and the calculated values should be treated as estimations. The same logic was behind the adjustment of the applied calculation constraint for experiments, where HF-C modules were tested with MgSO 4 . Here, the poor fit was related to the barely measurable reverse solute flux.
Nevertheless, after subjective treatment of the generated data, we can conclude that the A value ranges from 1.56 L·m −2 ·h −1 ·bar −1 to 1.86 L·m −2 ·h −1 ·bar −1 for the HF-O module and from 2.46 L·m −2 ·h −1 ·bar −1 to 2.74 L·m −2 ·h −1 ·bar −1 for HF-C modules. As expected, the values are close to the results obtained for the LPRO test and an increase of water permeability with chlorination is clear for both tests. The S parameter estimation resulted in values of 0.09 mm to 0.18 mm for all the membranes. The obtained results are in agreement with the S parameter reported for Aquaporin Inside TM membranes by Xia et al. [54]. The resulting low S parameter contributes significantly to reduction of the ICP and better utilization of the available driving force. Finally, as expected, the estimated B value for NaCl was the highest from all the draw solutions (0.24 L·m −2 ·h −1 and 0.80 L·m −2 ·h −1 for HF-O and HF-C, respectively) while the lowest was from MgSO 4 (0.01 L·m −2 ·h −1 and 0.12 L·m −2 ·h −1 for HF-O and HF-C, respectively). Table 6. Parameters A, B, and S estimated for NaCl, MgCl 2 , and MgSO 4 draw solute systems. RMSE and R 2 represent the root mean square error and the coefficient of determination, respectively, for the model. J w , water flux; J s , reverse solute flux.

Draw Solute NaCl
Membrane Type

FO Tests with Hollow Fiber Modules
The measured J w and J s /J w of the HF-O and HF-C modules were compared under different feed and draw flow rates, TMP, draw solutes, operational modes (PRO vs. FO and co-current vs. countercurrent), and temperature. In addition, the effect of two types of FS (DI water and seawater solution) were also compared. Tests with an artificial seawater solution were analyzed using only J w due to the limitations in accurate measurements of the solute permeation from the draw to the feed solution. Moreover, rejection of the model compounds (caffeine, niacin, and urea) with different molecular weights was analyzed when modules were operated in an FO or LPRO setup. The relation between membrane rejection and process recovery was studied using the example of urea as a feed contaminant.

Role of the Feed Flow Rate
Feed flow rate was varied to investigate the influence of the ECP on the active layer side of the membranes. Here, the feed flow rate varied from 60 to 140 L·h −1 , and 1 M NaCl was employed as a DS and operated at 25 L·h −1 . The results for both membrane types are summarized in Figure 6.
For DI water as FS, the HF-O module exhibited an J w increase from 17.4 to 20.2 L·m −2 ·h −1 (about 16.2%) in comparison to HF-C module that showed J w to increase from 20.5 to 26 L·m −2 ·h −1 (about 26.7%) by varying feed flow rate in the tested range. As shown in Figure 6a, the J w of the HF-C membranes is higher than that of the HF-O membranes at any testing conditions. J s /J w is also increased with feed flow rate for the HF-C membranes but remains relatively constant for HF-O membranes for all tested feed flow rates. Furthermore, J s /J w is much lower for HF-O membranes (J s /J w = 0.15 ± 0.02 g/L) than it is for HF-C membranes (J s /J w = 0.4 ± 0.1 g/L).
An increase of J w for the HF-C membranes was expected. As described in the literature [42] and in accordance with Section 3.1.2 results (Figure 4), chlorination of the TFC polyamide-based layer results in the reduction of the cross-linking density and thickness of the selective polyamide layer. Thus, chlorine treated membranes are more permeable and less selective. The main reason for the higher J s /J w of the HF-C membranes is the increase of the salt permeation coefficient B caused by this decrease in solute retention of the polyamide (see Section 3.  Figure 6a, the Jw of the HF-C membranes is higher than that of the HF-O membranes at any testing conditions. Js/Jw is also increased with feed flow rate for the HF-C membranes but remains relatively constant for HF-O membranes for all tested feed flow rates. Furthermore, Js/Jw is much lower for HF-O membranes (Js/Jw = 0.15 ± 0.02 g/L) than it is for HF-C membranes (Js/Jw = 0.4 ± 0.1 g/L) An increase of Jw for the HF-C membranes was expected. As described in the literature [42] and in accordance with section 3.1.2 results (Figure 4), chlorination of the TFC polyamide-based layer results in the reduction of the cross-linking density and thickness of the selective polyamide layer. Thus, chlorine treated membranes are more permeable and less selective. The main reason for the higher Js/Jw of the HF-C membranes is the increase of the salt permeation coefficient B caused by this decrease in solute retention of the polyamide (see section 3.1.3).
An increase of the Jw with feed flow rate was quite surprising since we do not expect a contribution of the ECP on the active layer side for DI water that is free of any solutes. Nevertheless, the results summarized in Figure 6a clearly indicate that Jw for both membranes increases with feed flow rate. Such behavior can only be attributed to the reduction of the ECP by reduction of the mass transfer coefficient with the flow rate. We speculate that here, by introduction of higher feed velocity, a stagnate layer of the fluid at the active layer gets disturbed, and diffused salt from the DS can be more effectively transported to the bulk of the FS. Therefore, the net driving force-the difference in osmotic pressure between the feed side and draw side of active layer-is enhanced. A similar rationalization can be applied to the explanation of the difference in development of Js/Jw for the HF-C and HF-O membranes with feed flow rate. As such, the low retention HF-C membranes can transport more salt to the feed side, locally increasing its concentration at the feed side of the selective layer. By facilitating the transport of this accumulated salt to the bulk solution via a higher feed flow rate, the driving force for salt transport from DS to the FS is also enhanced. As a result, we can observe an increase in Js/Jw with feed flow rate for low retention HF-C membranes. In contrast, Js/Jw for HF-O membranes that exhibit higher retention for salt is not affected by the feed flow rate since salt transport from the draw to the feed side is negligible. We think that this effect is valid for all types of FO membranes, and DS and can be mitigated by selecting a larger molecular size of draw solutes or more selective membranes. This effect also confirms the postulate by Werber et al. [13] that more selective membranes, not more water permeable FOs, are critically needed. An increase of the J w with feed flow rate was quite surprising since we do not expect a contribution of the ECP on the active layer side for DI water that is free of any solutes. Nevertheless, the results summarized in Figure 6a clearly indicate that J w for both membranes increases with feed flow rate. Such behavior can only be attributed to the reduction of the ECP by reduction of the mass transfer coefficient with the flow rate. We speculate that here, by introduction of higher feed velocity, a stagnate layer of the fluid at the active layer gets disturbed, and diffused salt from the DS can be more effectively transported to the bulk of the FS. Therefore, the net driving force-the difference in osmotic pressure between the feed side and draw side of active layer-is enhanced. A similar rationalization can be applied to the explanation of the difference in development of J s /J w for the HF-C and HF-O membranes with feed flow rate. As such, the low retention HF-C membranes can transport more salt to the feed side, locally increasing its concentration at the feed side of the selective layer. By facilitating the transport of this accumulated salt to the bulk solution via a higher feed flow rate, the driving force for salt transport from DS to the FS is also enhanced. As a result, we can observe an increase in J s /J w with feed flow rate for low retention HF-C membranes. In contrast, J s /J w for HF-O membranes that exhibit higher retention for salt is not affected by the feed flow rate since salt transport from the draw to the feed side is negligible. We think that this effect is valid for all types of FO membranes, and DS and can be mitigated by selecting a larger molecular size of draw solutes or more selective membranes. This effect also confirms the postulate by Werber et al. [13] that more selective membranes, not more water permeable FOs, are critically needed.
In these experiments, DI water represents the ideal FS (no osmotic pressure, no solutes), whereas artificial seawater represents the problematic FS (high osmotic pressure, high concentration of solutes). Hence, the use of these two FSs is able to display an overview of the effect of the FS osmotic pressure and its impact on the membrane performance. In the case of artificial seawater as an FS, the obtained J w values are much lower than those obtained for the DI water case (Figure 6b). The lower J w was expected as the osmotic pressure of the feed is significantly higher, and the difference in osmotic pressure between FS and DS is only ∆Π = 20.5 bar. The J w for HF-O membranes increased from 2.6 ± 0.1 L·m −2 ·h −1 to 3.5 ± 0.3 L·m −2 ·h −1 and for HF-C membranes from 2.0 ± 0.1 L·m −2 ·h −1 to 3.8 ± 0.1 L·m −2 ·h −1 . Interestingly, in the conditions where the FS exhibits high osmotic pressure, the difference in J w between HF-C and HF-O membranes is diminished. Furthermore, similar to the DI water experiments, an increase in the feed flow rate resulted in an increase in flux, proving the dominance of ECP at the active side of the membrane.

Role of the Draw Flow Rate
Similar to the feed flow rate, the applied draw flow rate has also influenced the performance of HFFO modules. FO performance obtained at various draw flow rates for the two types of investigated HFFO modules is summarized in Figure 7. As listed in Table 1, the feed flow rate was kept constant at 100 L·h −1 whereas the flow rate of 1M NaCl DS was varied from 25 to 100 L·h −1 .
osmotic pressure between FS and DS is only ΔΠ = 20.5 bar. The Jw for HF-O membranes increased from 2.6 ± 0.1 L·m −2 ·h −1 to 3.5 ± 0.3 L·m −2 ·h −1 and for HF-C membranes from 2.0 ± 0.1 L·m −2 ·h −1 to 3.8 ± 0.1 L·m −2 ·h −1 . Interestingly, in the conditions where the FS exhibits high osmotic pressure, the difference in Jw between HF-C and HF-O membranes is diminished. Furthermore, similar to the DI water experiments, an increase in the feed flow rate resulted in an increase in flux, proving the dominance of ECP at the active side of the membrane.

Role of the Draw Flow Rate
Similar to the feed flow rate, the applied draw flow rate has also influenced the performance of HFFO modules. FO performance obtained at various draw flow rates for the two types of investigated HFFO modules is summarized in Figure 7. As listed in Table 1, the feed flow rate was kept constant at 100 L·h −1 whereas the flow rate of 1M NaCl DS was varied from 25 to 100 L·h −1 . When DI water was applied as an FS, the HF-O modules exhibited J w of 19.0 ± 1.0 L·m −2 ·h −1 at 25 L·h −1 and J w systematically increased with the draw flow rate to 25.4 ± 1.0 L·m −2 ·h −1 at 100 L·h −1 (J w increase of 33.5%). For the HF-C membrane, J w was higher than for the HF-O membrane at any draw flow rate investigated, and it increased from 24.9 ± 0.6 L·m −2 ·h −1 at 25 L·h −1 to 31.3 L·m −2 ·h −1 at 100 L·h −1 , respectively (J w increase of 25.4%). The rate of J w increase with the draw flow rate does not have a linear character, as it was for the variations of the feed flow rate (Figure 6a), and reaches a plateau value (Figure 7a).
The larger impact of the draw flow rate on the change in J w in comparison to the feed flow rate can be explained by the more severe effect of the draw flow rate on the dilutive ICP and the dilutive ECP at the shell side of the membrane. The cause of this is simple-dilution of the DS with permeated water contributes significantly to the reduction of the available driving force along the length of the HFFO module. As a result, the average concentration of the DS inside the module is no longer 1 M NaCl and it depends heavily on the draw flow rate and water permeation. The higher the draw flow rate, the less dilution is obtained in the module. In an ideal situation, the DS should be operated at very high flow rates. However, the hammer effect of the liquid entering the module and the pressure build-up at the shell side might cause mechanical damage of the fibers. Therefore, to extend the lifetime of the membrane, low or moderate draw flow rates are more practical. When the draw flow rate increases, an average concentration of draw solute increases because of the shorter residence time of the DS in the module. Moreover, an increase of the mass transfer coefficient makes the boundary layer at the support-bulk interface thinner, allowing accelerated diffusion of the salt from the outer surface of the fiber into the support. Subsequently, dilutive ECP is decreased. An increase in the solute concentration inside the module elevates concentration of the solute in the support, alleviating the detrimental effect of the ICP. Consequently, J w increases as the net osmotic pressure difference across the active layer is higher. The maximum concentration in the module that can be reached is the initial concentration of the DS, and by increasing draw flow rate, we asymptotically approach this concentration. Therefore, the plateau in J w can be observed at higher draw flow rates. Hence, there is an optimal draw flow rate for each HFFO module, above which an increase in J w does not occur, and operation above this draw flow rate makes no sense from a practical and economical point of view.
In the case of artificial seawater as an FS, a maximum J w obtained was 5.7 L·m −2 ·h −1 . Herein, for tests with various draw flow rates, no difference between HF-C and HF-O modules was observed in terms of J w . The unfavorable conditions (concentration polarization effects and low permeate flux) produced by a high osmotic pressure FS continues to cancel all differences in J w between HF-C and HF-O. Yet, by increasing the draw flow rate, the average concentration of the DS increased and the net driving force for water transport was enhanced. Therefore, by reducing the ICP effect, the J w showed a greater increase (Figure 7b) than in the case of the experiments with various feed flow rates (Figure 6b).

Applied TMP
FO is a process driven by the difference in osmotic pressure between DS and FS, and, as such, there is no need to apply pressure to obtain water permeation. Additionally, we would not expect a great impact of the TMP on the performance of the HFFO membranes since the hydraulic pressure applied is much lower than the osmotic pressure difference applied in the FO process. Nevertheless, multiple studies have reported both negative and positive impacts of the applied TMP in FO processes [23,38,55,56]. In those studies, elevated TMP was used mostly to study fouling; to facilitate permeate flow rate; or to normalize the draw and feed flow rates, when there are several modules connected in series. In this study, the effect of elevated TMP has been analyzed using the HF-C and HF-O modules. FO performance of the investigated modules at systematically elevated TMP are summarized in Figure 8. For all the testing conditions, where TMP varied from 0.1 to 1.1 bar, the feed and draw flow rates were kept constant and were 100 L·h −1 and 25 L·h −1 , respectively.
in the solute concentration inside the module elevates concentration of the solute in the support, alleviating the detrimental effect of the ICP. Consequently, Jw increases as the net osmotic pressure difference across the active layer is higher. The maximum concentration in the module that can be reached is the initial concentration of the DS, and by increasing draw flow rate, we asymptotically approach this concentration. Therefore, the plateau in Jw can be observed at higher draw flow rates. Hence, there is an optimal draw flow rate for each HFFO module, above which an increase in Jw does not occur, and operation above this draw flow rate makes no sense from a practical and economical point of view.
In the case of artificial seawater as an FS, a maximum Jw obtained was 5.7 L·m −2 ·h −1 . Herein, for tests with various draw flow rates, no difference between HF-C and HF-O modules was observed in terms of Jw. The unfavorable conditions (concentration polarization effects and low permeate flux) produced by a high osmotic pressure FS continues to cancel all differences in Jw between HF-C and HF-O. Yet, by increasing the draw flow rate, the average concentration of the DS increased and the net driving force for water transport was enhanced. Therefore, by reducing the ICP effect, the Jw showed a greater increase (Figure 7b) than in the case of the experiments with various feed flow rates (Figure 6b).

Applied TMP
FO is a process driven by the difference in osmotic pressure between DS and FS, and, as such, there is no need to apply pressure to obtain water permeation. Additionally, we would not expect a great impact of the TMP on the performance of the HFFO membranes since the hydraulic pressure applied is much lower than the osmotic pressure difference applied in the FO process. Nevertheless, multiple studies have reported both negative and positive impacts of the applied TMP in FO processes [23,38,55,56]. In those studies, elevated TMP was used mostly to study fouling; to facilitate permeate flow rate; or to normalize the draw and feed flow rates, when there are several modules connected in series. In this study, the effect of elevated TMP has been analyzed using the HF-C and HF-O modules. FO performance of the investigated modules at systematically elevated TMP are summarized in Figure 8. In this range of tested TMPs, J w increased from 17.9 ± 0.5 L·m −2 ·h −1 to 19.2 ± 0.5 L·m −2 ·h −1 (7.3% increase in J w ) for HF-O modules and from 24.0 ± 1.3 L·m −2 ·h −1 to 27.7 ± 0.2 L·m −2 ·h −1 (15% increase in J w ) for HF-C modules. The results obtained are similar to the ones reported by Coday et al. [23], where minimal changes in J w and J s /J w were obtained at elevated TMPs between 1 to 2 bar.
On one hand, the obtained results are expected since TMP does not contribute to a significant improvement of J w for both membrane types. On the other hand, the applied pressure range is quite narrow and performance of the tested HFFO modules could be further increased by increasing TMP. Nevertheless, the fact that FO membranes are purposefully designed with thinner and more-fragile support restricts the possibilities to significantly increase TMP. The difference in the rate of the J w increase with the applied TMP between the HF-O and HF-C modules originates from the difference in the A parameter between those two modules (see Section 3.1.3). A more permeable selective layer, like in the case of the HF-C membrane, will always be more responsive to the applied TMP than a membrane with a less permeable selective layer, like an HF-O membrane. J s /J w was significantly reduced with applied TMP for HF-C membranes (from 0.40 ± 0.02 g·L −1 at 0.1 bar to 0.31 ± 0.2 g·L −1 at 1.1 bar) and negligibly for HF-O modules (from 0.14 ± 0.02 g·L −1 at 0.1 bar to 0.12 ± 0.2 g·L −1 at 1.1 bar). We speculate that the greater increase in J w for HF-C membranes caused a more severe dilutive ICP at the draw side of selective layer. As a result, the solute concentration difference between the draw side and feed side of the active layer was reduced; therefore, the driving force for solute diffusion was diminished. Consequently, we observed reduced J s /J w with the applied pressure.
The same minor effects of various TMPs applied to J w were observed for artificial seawater as an FS (Figure 8b). As shown in previous experiments, no differences in J w between HF-O and HF-C membranes were observed when using artificial seawater as an FS.

Influence of the Applied DS
DS selection has been abundantly studied [20,31,[57][58][59] in order to tune the FO process for a specific application, to reduce the costs of the process, or to obtain a more environmentally friendly technology. Here, three common inorganic salts (NaCl, MgCl 2 , MgSO 4 ) at various concentrations were tested as a draw solute. Concentrations of draw solutes were selected to obtain the same osmotic pressure and compare the DS based on the driving forces that they generate. NaCl and MgCl 2 osmotic pressures varied from 24.7 to 99.0 bar, whereas MgSO 4 osmotic pressure was investigated from 7.9 to 74.2 bar due to the limits in solubility of the MgSO 4 . For all the tests, DI water was used as an FS. Feed flow rate and draw flow rate were kept constant at 100 L·h −1 and 25 L·h −1 , respectively. The summary of the results is shown in Figure 9.
Moreover, for all tests with various DS, the J w obtained with HF-C modules was higher here than the J w obtained for HF-O modules for any DS and for all concentrations investigated. Obviously, for each draw solute, J w increases with an increasing DS concentration since the driving force (osmotic pressure difference) is raised across the membrane. As shown in Figure 9a, by applying NaCl concentration varying from 0.5M to 2M (24.7 bar to 99.0 bar), the HF-O membrane exhibited J w increases from 13.8 ± 0.7 L·m −2 ·h −1 to 28.3 ± 1.1 L·m −2 ·h −1 . In comparison, the HF-C membrane showed a J w increase from 18.6 ± 0.6 L·m −2 ·h −1 to 30.1 ± 0.5 L·m −2 ·h −1 . When MgCl 2 was applied as a draw solute in the concentrations equivalent in osmotic pressure to the 0.5-2 M range of NaCl (24.7 bar to 99.0 bar), The HF-O membrane showed an increase in J w from 12.6 ± 0.9 L·m −2 ·h −1 to 22.5 ± 1.0 L·m −2 ·h −1 , whereas the HF-C membrane showed a J w increase from 15.6 ± 0.9 L·m −2 ·h −1 to 28.3 ± 0.5 L·m −2 ·h −1 . Finally, for MgSO 4 as a draw solute, the HF-O membrane exhibited a J w increase from 11.8 ± 0.9 L·m −2 ·h −1 to 15.0 ± 0.8 L·m −2 ·h −1 in comparison to the HF-C membrane that showed a J w increase from 13.8 ± 0.8 L·m −2 ·h −1 L to 21.3 ± 0.1 L·m −2 ·h −1 . Note that for MgSO 4 , the maximum applied osmotic pressure was 74.2 bar.
The osmotic pressure of the tested DS is comparable, and, therefore, the differences in J w between them is solely due to the intrinsic properties of the salt and the ICP development for each salt type. Based on the results presented in Figure 9 and the available literature [2,60], we can distinguish two regions of the flux increase with osmotic pressure: 1-a linear increase of the J w with osmotic pressure and 2-an asymptotic increase of J w with osmotic pressure. Obviously, the first regime where J w increases linearly with osmotic pressure represents the most efficient operation regime since applied potential in osmotic pressure is directly proportional to the flux obtained. The second regime with asymptotic increase of the flux with osmotic pressure is related to an extensive ICP. In this regime an increase of the osmotic pressure is not directly proportional to the obtained J w , and the available driving force is not efficiently utilized to generate J w . As shown in Figure 9a, all tests with NaCl show a close to linear correlation between J w and the applied osmotic pressure of the DS. High diffusivity of small NaCl ions (Table 3) and a low S parameter of the support (Table 6) enhance the transport of salt ions into the support, minimizing dilutive ICP. As a result, a high concentration of the NaCl is experienced by the selective layer, and the highest J w values were obtained for both membrane types using this salt. The diffusion coefficient of MgCl 2 is slightly lower than NaCl (Table 3), and thus non-linear J w behavior was exhibited by the HF-O membranes at the highest salt concentration. HF-C membranes could maintain a linear J w behavior due to their higher A parameter, delaying severe ICP. However, for MgSO 4 tests, J w displayed a clear non-linear behavior for both membrane types and here we can explain it as a significantly lower diffusion coefficient (Table 3) of the MgSO 4 , compared to the other draw solutes, which makes this system more prone to ICP contribution at high salt concentrations. The diffusion coefficient is not the only driver for the J w differences between the draw solutes, as it is closely dependent on the molecule/ion size of the solutes [20,61]. Na + and Cl − ions are both small enough to easily diffuse through the membrane support without hampering water transport, thus minimizing ICP. However, for MgCl 2 , the fast-diffusive Cl − ions are impaired by the bigger and slower Mg 2+ ions to maintain electroneutrality. This causes a slight blockage, by the Mg 2+ ions, of the water transport from the FS to the DS, and an increase of the ICP. These effects are only further aggravated when Mg 2+ ions are introduced as MgSO 4 with SO 4 2− ions, which have the biggest ion size from all the draw solutes tested. Consequently, both of these ions, due to their size and low diffusion rate, have more difficulty travelling through the membrane support, lowering the osmotic pressure difference across the selective layer, and obtaining the lowest J w in this test. Moreover, for all tests with various DS, the Jw obtained with HF-C modules was higher here than the Jw obtained for HF-O modules for any DS and for all concentrations investigated. Obviously, for each draw solute, Jw increases with an increasing DS concentration since the driving force (osmotic pressure difference) is raised across the membrane. As shown in Figure 9a Nevertheless, compared to NaCl, the lower diffusion coefficients and bigger size of MgCl 2 and MgSO 4 can be advantageous in terms of obtaining reduced J s and J s /J w . More severe ICP for MgCl 2 and MgSO 4 results in a lower concentration of the draw solutes at the support-selective layer interface, so a lower driving force for solute transport from the feed to the draw solution is generated. Moreover, TFC active layers generally have a higher retention for divalent ions like Mg 2+ and SO 4 2− in comparison with monovalent ions, such as Na + and Cl − . Those two effects summarized are responsible for the reduced J s and J s /J w with MgCl 2 and MgSO 4 . In an extreme case, a combination of high retention HF-O and MgSO 4 resulted in a barely measurable J s /J w (close to 0 g·L −1 for all MgSO 4 concentrations).
In comparison, the tests with NaCl as a DS generated the highest J s /J w values.

Operation Mode of the Membrane
In this section, two membrane operation modes were compared. Firstly, we compared the FO performance of the tested modules during operation in an FO mode (FS facing active layer) and a PRO mode (DS facing active layer solution). Secondly, we investigate the impact of the operation in a co-current and counter-current mode on FO performance of the modules.

(a) Co-current vs. counter current for HFFO modules
In counter-current operations (used in all experiments in this study), FS and DS enter and exit the HFFO module from opposite sides. In the co-current mode, FS and DS enter and exit the module from the same side of the module. The comparison of the counter-current and co-current operation with the HFFO modules was carried out with 1 M NaCl as DS applied at 25 L·h −1 . DI water and artificial salt solution were used as an FS and operated at 100 L·h −1 . As shown in Figure 10a, for the DI water as an FS, for both the HF-C and HF-O modules, the differences in performance between operation in a co-current or counter-current mode were not significant. counter-current and co-current modes, respectively. As schematically described in Figure 11, for an FS with low osmotic pressure, such as DI water, the osmotic pressure of the FS does not change significantly during the up-concentration inside the HFFO module. A lack of solutes in the FS and a relatively low diffusion of the salt from the DS to the feed side keeps almost constant osmotic pressure on the FS. Thus, for operation in co-current and counter-current modes, the same average driving force is obtained inside the module, and no difference between operation in co-current and countercurrent mode can be measured. Also negligible is the difference in Jw between counter-current and co-current operation when artificial seawater is applied as an FS (Figure 10b). As reported in the literature [17,38], operation with counter-current mode with high osmotic pressure FS should generate higher flux in comparison to operation in a co-current mode. Due to the maintained constant osmotic pressure difference between FS and DS, the available difference in osmotic pressure is used more efficiently than during operation in the co-current mode ( Figure 11). However, this effect cannot be seen in our case due to the low permeation in the process (Jw < 4 L·m −2 ·h −1 ). We speculate that a higher difference in osmotic pressure or operating conditions that can enhance Jw (see sections 3.2.1, 3.2.2 and 3.2.4) could reveal the advantage of counter-current operation. HF-O modules exhibited a J w of 18.2 ± 1.1 L·m −2 ·h −1 and 19.0 ± 1.4 L·m −2 ·h −1 for co-current and counter-current operations, respectively, whereas HF-C membranes generated higher flux for both operation modes: J w of 23.8 ± 0.8 L·m −2 ·h −1 and J w of 24.9 ± 0.6 L·m −2 ·h −1 for co-current and counter-current modes, respectively. Similarly, no significant change was measured in J s /J w when the two operational modes were compared (Figure 10a). HF-O modules obtained a J s /J w of 0.14 ± 0.02 g·L −1 and 0.12 ± 0.01 g·L −1 , and HF-C modules achieved a J s /J w of 0.36 ± 0.05 g·L −1 and 0.35 ± 0.04 g·L −1 for counter-current and co-current modes, respectively. As schematically described in Figure 11, for an FS with low osmotic pressure, such as DI water, the osmotic pressure of the FS does not change significantly during the up-concentration inside the HFFO module. A lack of solutes in the FS and a relatively low diffusion of the salt from the DS to the feed side keeps almost constant osmotic pressure on the FS. Thus, for operation in co-current and counter-current modes, the same average driving force is obtained inside the module, and no difference between operation in co-current and counter-current mode can be measured.

(b) PRO vs FO for HFFO modules
Operations in the FO mode (used in all the experiments in this study) and PRO mode were compared with the same concentration of DS (1 M NaCl) and the same feed and draw flow rates (100 L·h −1 and 25 L·h −1 , respectively) for both operating modes. As shown in Figure 12a, when DI water was applied as a DS, HF-O modules exhibited a Jw of 17.9 ± 0.5 L·m −2 ·h −1 in the FO mode, whereas, in the PRO mode, Jw increased to 20.9 ± 1.3 L·m −2 ·h −1 . Similar behavior was observed for HF-C modules. Here, Jw increased from 24.1 ± 1.3 L·m −2 ·h −1 in the FO mode and to 26.8 ± 1.3 L·m −2 ·h −1 in the PRO mode. The obtained results are in good agreement with the available literature [17,38,60], describing that the dilutive ICP of the DS inside the support side (FO mode) causes a larger Jw decline than concentrative ICP of the FS inside the support (PRO mode). However, under similar conditions, the obtained Jw improvement (13.9 ± 3.8% increase) during operation in PRO mode is not as significant as reported in the literature [38] (60-80% increase in Jw). We correlate this rather low impact of the FS and DS orientation on the FO performance to the low S parameter (see section 3.1.3) of the membrane and generally low ICP when operated in FO mode (see section 3.2.4). Also negligible is the difference in J w between counter-current and co-current operation when artificial seawater is applied as an FS (Figure 10b). As reported in the literature [17,38], operation with counter-current mode with high osmotic pressure FS should generate higher flux in comparison to operation in a co-current mode. Due to the maintained constant osmotic pressure difference between FS and DS, the available difference in osmotic pressure is used more efficiently than during operation in the co-current mode ( Figure 11). However, this effect cannot be seen in our case due to the low permeation in the process (J w < 4 L·m −2 ·h −1 ). We speculate that a higher difference in osmotic pressure or operating conditions that can enhance J w (see Sections 3.2.1, 3.2.2 and 3.2.4) could reveal the advantage of counter-current operation.

(b) PRO vs. FO for HFFO modules
Operations in the FO mode (used in all the experiments in this study) and PRO mode were compared with the same concentration of DS (1 M NaCl) and the same feed and draw flow rates (100 L·h −1 and 25 L·h −1 , respectively) for both operating modes. As shown in Figure 12a, when DI water was applied as a DS, HF-O modules exhibited a J w of 17.9 ± 0.5 L·m −2 ·h −1 in the FO mode, whereas, in the PRO mode, J w increased to 20.9 ± 1.3 L·m −2 ·h −1 . Similar behavior was observed for HF-C modules. Here, J w increased from 24.1 ± 1.3 L·m −2 ·h −1 in the FO mode and to 26.8 ± 1.3 L·m −2 ·h −1 in the PRO mode. The obtained results are in good agreement with the available literature [17,38,60], describing that the dilutive ICP of the DS inside the support side (FO mode) causes a larger J w decline than concentrative ICP of the FS inside the support (PRO mode). However, under similar conditions, the obtained J w improvement (13.9 ± 3.8% increase) during operation in PRO mode is not as significant as reported in the literature [38] (60-80% increase in J w ). We correlate this rather low impact of the FS and DS orientation on the FO performance to the low S parameter (see Section 3.1.3) of the membrane and generally low ICP when operated in FO mode (see Section 3.2.4). As shown in Figure 12a, Js/Jw is reduced when the process is carried out in PRO mode. In our tests, for HF-C membranes Js/Jw decreased from 0.4 ± 0.02 g·L −1 in FO mode and to 0.27 ± 0.02 g·L −1 in PRO mode, respectively. For HF-O membranes, this reduction in Js/Jw was much lower (0.14 ± 0.01 g·L −1 in FO mode and 0.11 ± 0.01 g·L −1 in PRO mode, respectively). In PRO mode, diffusion of the solute from the active layer to the bulk of the FS is restricted by the support, similar to the limited solute diffusion into the support when operated in FO mode. We speculate that this limited solute diffusion from the support to bulk solution contributes to a reduction of Js/Jw in PRO mode.
Like the tests with DI water, when artificial seawater was applied as an FS, the use of the PRO mode results in a slight Jw increase. Jw increased from 3.6 ± 0.4 to 4.9 ± 0.1 L·m −2 ·h −1 and from 4.8 ± 0.1 to 5.1 ± 0.2 L·m −2 ·h −1 when switching from the FO to PRO mode for HF-O and HF-C modules, respectively. It can be argued that this Jw increase is negligible due to an error originating from the accuracy of the applied measuring technique. In fact, it is expected that the advantages of PRO mode are limited for high osmotic pressure solutions. Due to concentrative ICP of the feed solute, the net difference in osmotic pressure between both sides of the active layer is reduced, and, thus, the PRO mode's advantages are minimized.

Role of the Temperature
Multiple studies show that an increase in temperature can effectively enhance water transport in membrane processes [24,25,[33][34][35][36]. For example, Seker et al. [25] highlighted the importance of the temperature for viscous FS and described how the efficiency of the FO process can be enhanced by an increase of the FS temperature (reduced viscosity) while keeping the DS cold (reduced draw solute diffusion to the FS). In addition, Xie et al. [24] demonstrated that an increase in Jw with temperature can come with a negative side effect of a decreased solute rejection of trace organic compounds, due to the increase in solute diffusivity. Therefore, it is crucial to find the right balance between rejection and Jw when choosing the process temperature.
Here, the temperature of both DS and FS varied from 15 °C to 45 °C to determine the temperature impact on the performance of HFFO modules. As before, 1 M NaCl was used as the DS, and feed and draw flow rates were kept at 100 L·h −1 and 25 L·h −1 , respectively. The results for both membrane types are summarized in Figure 13. As shown in Figure 12a, J s /J w is reduced when the process is carried out in PRO mode. In our tests, for HF-C membranes J s /J w decreased from 0.4 ± 0.02 g·L −1 in FO mode and to 0.27 ± 0.02 g·L −1 in PRO mode, respectively. For HF-O membranes, this reduction in J s /J w was much lower (0.14 ± 0.01 g·L −1 in FO mode and 0.11 ± 0.01 g·L −1 in PRO mode, respectively). In PRO mode, diffusion of the solute from the active layer to the bulk of the FS is restricted by the support, similar to the limited solute diffusion into the support when operated in FO mode. We speculate that this limited solute diffusion from the support to bulk solution contributes to a reduction of J s /J w in PRO mode.
Like the tests with DI water, when artificial seawater was applied as an FS, the use of the PRO mode results in a slight J w increase. J w increased from 3.6 ± 0.4 to 4.9 ± 0.1 L·m −2 ·h −1 and from 4.8 ± 0.1 to 5.1 ± 0.2 L·m −2 ·h −1 when switching from the FO to PRO mode for HF-O and HF-C modules, respectively. It can be argued that this J w increase is negligible due to an error originating from the accuracy of the applied measuring technique. In fact, it is expected that the advantages of PRO mode are limited for high osmotic pressure solutions. Due to concentrative ICP of the feed solute, the net difference in osmotic pressure between both sides of the active layer is reduced, and, thus, the PRO mode's advantages are minimized.

Role of the Temperature
Multiple studies show that an increase in temperature can effectively enhance water transport in membrane processes [24,25,[33][34][35][36]. For example, Seker et al. [25] highlighted the importance of the temperature for viscous FS and described how the efficiency of the FO process can be enhanced by an increase of the FS temperature (reduced viscosity) while keeping the DS cold (reduced draw solute diffusion to the FS). In addition, Xie et al. [24] demonstrated that an increase in J w with temperature can come with a negative side effect of a decreased solute rejection of trace organic compounds, due to the increase in solute diffusivity. Therefore, it is crucial to find the right balance between rejection and J w when choosing the process temperature.
Here, the temperature of both DS and FS varied from 15 • C to 45 • C to determine the temperature impact on the performance of HFFO modules. As before, 1 M NaCl was used as the DS, and feed and draw flow rates were kept at 100 L·h −1 and 25 L·h −1 , respectively. The results for both membrane types are summarized in Figure 13.  (Figure 13). The increase in Jw was anticipated, as temperature influences diffusion rates and fluid viscosity. Firstly, higher diffusivity of the solute enhances its transport inside the support, thereby reducing ICP, and consequently generates higher Jw [36]. Secondly, liquid viscosity drops with temperature, thereby reducing resistance for solute diffusion, again improving Jw by reduction of ICP.
Enhanced diffusion of the draw solute, and thus reduced of ICP caused by elevated temperature, also results in an increase in Js. As shown in Figure 13, Js/Jw for HF-O slightly increases with the temperature from 0.11 ± 0.01 g·L −1 at 15 °C to 0.17 ± 0.01 g·L −1 at 45 °C. This result indicates that Js increases faster than Jw with the temperature and facilitates solute diffusion that favors salt transport from the draw side to the feed side over water transport from the feed side to the draw side. We speculate that the lower A value of the HF-O membranes, compared to the HF-C membranes, contributes to this effect. For HF-C membranes, Js/Jw is relatively constant for tested temperatures (Js/Jw of about 0.32 g·L −1 ), indicating uniform contribution of the temperature to the enhancement of the salt and water transport.

Rejection: FO vs LPRO
The mechanism behind the solute rejection in pressure driven processes (e.g., RO) and osmotically driven processes (e.g. FO) is discussed in multiple studies [49,[62][63][64]. This mechanism is described as a combination of interactions based on steric exclusion, electrostatic repulsion, system transmembrane pressure, process recovery, concentration polarization, and the characteristics of the membrane and the solute. In this section, rejection of the model compounds caffeine, niacin, and urea is compared between two processes-RO and FO. These compounds were chosen due to their differences in molecular weight and molecular charge ( Figure 14) and their relevance in food and beverage applications. Urea is an abundant and common contaminant in wastewater, and, as a small In brief, the HF-O module exhibited a J w increase from 12.7 ± 1.0 L·m −2 ·h −1 to 22.0 ± 1.0 L·m −2 ·h −1 , and HF-C modules showed an increase in J w from 16.1 ± 0.2 L·m −2 ·h −1 to 25 ± 0.1 L·m −2 ·h −1 by increasing the temperature of FS and DS from 15 • C to 45 • C. Interestingly, for both HF-C and HF-O modules, J w increases to the same extent (about 9.0 L·m −2 ·h −1 ) within the tested temperature range ( Figure 13). The increase in J w was anticipated, as temperature influences diffusion rates and fluid viscosity. Firstly, higher diffusivity of the solute enhances its transport inside the support, thereby reducing ICP, and consequently generates higher J w [36]. Secondly, liquid viscosity drops with temperature, thereby reducing resistance for solute diffusion, again improving J w by reduction of ICP.
Enhanced diffusion of the draw solute, and thus reduced of ICP caused by elevated temperature, also results in an increase in J s . As shown in Figure 13, J s /J w for HF-O slightly increases with the temperature from 0.11 ± 0.01 g·L −1 at 15 • C to 0.17 ± 0.01 g·L −1 at 45 • C. This result indicates that J s increases faster than J w with the temperature and facilitates solute diffusion that favors salt transport from the draw side to the feed side over water transport from the feed side to the draw side. We speculate that the lower A value of the HF-O membranes, compared to the HF-C membranes, contributes to this effect. For HF-C membranes, J s /J w is relatively constant for tested temperatures (J s /J w of about 0.32 g·L −1 ), indicating uniform contribution of the temperature to the enhancement of the salt and water transport.

Rejection of Target Compounds
Rejection: FO vs. LPRO The mechanism behind the solute rejection in pressure driven processes (e.g., RO) and osmotically driven processes (e.g., FO) is discussed in multiple studies [49,[62][63][64]. This mechanism is described as a combination of interactions based on steric exclusion, electrostatic repulsion, system transmembrane pressure, process recovery, concentration polarization, and the characteristics of the membrane and the solute. In this section, rejection of the model compounds caffeine, niacin, and urea is compared between two processes-RO and FO. These compounds were chosen due to their differences in molecular weight and molecular charge ( Figure 14) and their relevance in food and beverage applications. Urea is an abundant and common contaminant in wastewater, and, as a small and neutral molecule (Figure 14), urea is the most interesting compound to examine due to challenges that membrane technology faces in rejection of this compound. and neutral molecule ( Figure 14), urea is the most interesting compound to examine due to challenges that membrane technology faces in rejection of this compound. Feed flow rates for LPRO and FO process were varied to obtain uniform recovery of about 23 ± 4% ( Table 4). Concentration of the model compounds in the FS were 200 mg·L −1 for all tests. As shown in Figure 15, caffeine rejection of 99 ± 1% was obtained in the FO process, whereas roughly 95 ± 1% caffeine rejection was obtained in pressure driven LPRO for both HF-C and HF-O modules. For niacin with a slightly lower molecular weight, also 100 ± 1% rejection was obtained in FO for both types of HFFO modules. However, in LPRO process, we observed a difference in rejection between HF-C and HF-O module. In particular, niacin rejection in LPRO was 77 ± 1% and 90 ± 1% for HF-C and HF-O modules, respectively. Finally, for the low molecular weight urea, calculated rejection in the FO process was 78 ± 1% and 84 ± 1% for HF-C and HF-O membranes, respectively. Again, for LPRO operations, urea rejection was much lower and for HF-C membrane, we did not observe any urea rejection while HF-O modules showed only 19 ± 1% urea rejection. Feed flow rates for LPRO and FO process were varied to obtain uniform recovery of about 23 ± 4% ( Table 4). Concentration of the model compounds in the FS were 200 mg·L −1 for all tests. As shown in Figure 15, caffeine rejection of 99 ± 1% was obtained in the FO process, whereas roughly 95 ± 1% caffeine rejection was obtained in pressure driven LPRO for both HF-C and HF-O modules. For niacin with a slightly lower molecular weight, also 100 ± 1% rejection was obtained in FO for both types of HFFO modules. However, in LPRO process, we observed a difference in rejection between HF-C and HF-O module. In particular, niacin rejection in LPRO was 77 ± 1% and 90 ± 1% for HF-C and HF-O modules, respectively. Finally, for the low molecular weight urea, calculated rejection in the FO process was 78 ± 1% and 84 ± 1% for HF-C and HF-O membranes, respectively. Again, for LPRO operations, urea rejection was much lower and for HF-C membrane, we did not observe any urea rejection while HF-O modules showed only 19 ± 1% urea rejection.
In FO, the HF-C and HF-O membrane showed nearly identical rejection values for caffeine and niacin, and similar urea rejection values at low recoveries. The results confirm that HF-C modules are still selective towards compounds present in the FS regardless of higher J s /J w values obtained during previous FO tests. When operated in FO, for the largest compounds, such as caffeine (Figure 14), we cannot distinguish any difference in solute retention between two membrane types, but the difference becomes more pronounced with low molecular weight compounds such as urea. In general, rejection in LPRO is lower for the same compounds than in FO. Thus, LPRO tests reveal more differences in membrane retention for the model contaminants between HF-C and HF-O membranes and the effect of the molecular weight and charge becomes more visible (Figure 15).
Similarly to the reported studies touching on the comparison between FO and RO solute rejections [31,49,64], solute rejections are also higher for FO process. A larger permeate flux in FO operation can be responsible for this effect. However, Xie et al. [65] compare FO and RO operations at the same permeate fluxes and, in some cases, observed elevated rejection of solutes in FO when compared to RO. Xie suggests that this effect is attributed to the delayed forward diffusion of feed solutes through the selective layer pore when operated in FO mode. However, to fully understand this complex topic, more systematic studies on the interactions between membrane, feed solute, and DS are necessary. Table 4). Concentration of the model compounds in the FS were 200 mg·L for all tests. As shown in Figure 15, caffeine rejection of 99 ± 1% was obtained in the FO process, whereas roughly 95 ± 1% caffeine rejection was obtained in pressure driven LPRO for both HF-C and HF-O modules. For niacin with a slightly lower molecular weight, also 100 ± 1% rejection was obtained in FO for both types of HFFO modules. However, in LPRO process, we observed a difference in rejection between HF-C and HF-O module. In particular, niacin rejection in LPRO was 77 ± 1% and 90 ± 1% for HF-C and HF-O modules, respectively. Finally, for the low molecular weight urea, calculated rejection in the FO process was 78 ± 1% and 84 ± 1% for HF-C and HF-O membranes, respectively. Again, for LPRO operations, urea rejection was much lower and for HF-C membrane, we did not observe any urea rejection while HF-O modules showed only 19 ± 1% urea rejection.

Urea Rejection vs. Recovery
Urea rejection was analyzed under different process recoveries in the FO operation to further understand the correlation between recovery and membrane retention. As mentioned before, urea was selected as a model compound due to the challenges it causes in rejection on the membranes. Feed flow rates were varied from 60 to 140 L·h −1 to obtain various recoveries in the module, while the draw flow rate was maintained at 25 L·h −1 , and the DS used was 1M NaCl. Summarized results on the influence of process rejection on urea rejection are shown in Figure 16. At the lowest feed flow rate (60 L·h −1 ), the highest process recovery (74.1%) was achieved by the HF-C module while the HF-O membrane generated 59.9% recovery. With the highest feed flow rate applied (140 Lh −1 ), a higher process recovery was also obtained for the HF-C membranes (34.2%) than for the HF-O module (25.5%).
Reduction of the rejection with increasing recovery is expected, as high recovery results in an increase of the solute concentration in the module. At the same time, ECP of the feed solute is enhanced by the rising solute concentration but also by the reduced feed flow rate, decreasing the mass transfer coefficient. Combined, these two effects yielded a higher solute diffusion rate than the feed to the draw side and the lower solute rejection. Interestingly, as shown in Figure 16, HF-C membranes seem to be more sensitive than HF-O membranes to increasing recoveries and reduction of the rejection is more severe for those modules. Here, lower retention of the chlorinated polyamide layer is more pronounced at higher process recoveries. The opposite phenomenon is seen with low recoveries, where HF-C and HF-O membranes have similar rejection. Figure 16. Rejection of urea in FO operation for HF-C and HF-O modules as a function of process recovery. The urea concentration was 200 mg·L −1 . The feed flow rate for both modules was modified from 60-140 L·h −1 to obtain a wide range of process recoveries. The draw flow rate was maintained at 25 L·h −1 with a DS of 1M NaCl. Red data points are urea rejections from the FO tests shown in Figure  15, where a different draw flow rate, 16 L·h −1 , and a different DS concentration, 0.5 NaCl, was used.
Reduction of the rejection with increasing recovery is expected, as high recovery results in an increase of the solute concentration in the module. At the same time, ECP of the feed solute is enhanced by the rising solute concentration but also by the reduced feed flow rate, decreasing the mass transfer coefficient. Combined, these two effects yielded a higher solute diffusion rate than the feed to the draw side and the lower solute rejection. Interestingly, as shown in Figure 16, HF-C membranes seem to be more sensitive than HF-O membranes to increasing recoveries and reduction of the rejection is more severe for those modules. Here, lower retention of the chlorinated polyamide layer is more pronounced at higher process recoveries. The opposite phenomenon is seen with low recoveries, where HF-C and HF-O membranes have similar rejection.

Conclusion
In this study, the significance of the operating conditions to the performance of the FO process for two types of HFFO modules, HF-C and HF-O membranes, has been demonstrated by analyzing their behavior under various testing conditions. The following conclusions can be made from these investigations: • Jw can be enhanced for the HFFO module by chlorination due to a decrease in the thickness of the polyamide layer. • Dilutive ICP was found to be one of the most significant parameters responsible for membrane performance as it acutely limits Jw when it becomes severe. This confirmed the dominant role of ICP for FO membranes, independent of membrane characteristics. Consequently, draw concentration and draw solute type have the biggest impact on the membrane performance. Additionally, optimal flow rates and osmotic pressure differences between FS and DS reduce the effect of other operating conditions, such as membrane orientation (PRO/FO mode, co-current/counter-current mode).  Figure 15, where a different draw flow rate, 16 L·h −1 , and a different DS concentration, 0.5 NaCl, was used.

Conclusions
In this study, the significance of the operating conditions to the performance of the FO process for two types of HFFO modules, HF-C and HF-O membranes, has been demonstrated by analyzing their behavior under various testing conditions. The following conclusions can be made from these investigations: • J w can be enhanced for the HFFO module by chlorination due to a decrease in the thickness of the polyamide layer.

•
Dilutive ICP was found to be one of the most significant parameters responsible for membrane performance as it acutely limits J w when it becomes severe. This confirmed the dominant role of ICP for FO membranes, independent of membrane characteristics. Consequently, draw concentration and draw solute type have the biggest impact on the membrane performance. Additionally, optimal flow rates and osmotic pressure differences between FS and DS reduce the effect of other operating conditions, such as membrane orientation (PRO/FO mode, co-current/counter-current mode). • A low osmotic pressure difference between DS and FS drastically reduces any differences between the chlorinated HFFO modules and the original HFFO modules. However, changes in operating conditions still alter membrane performance following the same mechanism explained for the large osmotic pressure difference between DS and FS.

•
The RO process displayed lower solute rejection compared to the FO process. High process recoveries have a detrimental effect on solute rejection, due to an increase of solute concentration inside the module, which facilitates solute diffusion to the permeate.
This work has demonstrated that operational parameters heavily influence membrane performance for HFFO modules. Consequently, our results can serve as a guideline to take maximum advantage of the HFFO membrane modules in pilot plant or full-scale processes. Furthermore, guidelines can aid FO installation design, thus paving the way for implementation of FO in technological applications.
Supplementary Materials: The following are available online at http://www.mdpi.com/2077-0375/9/6/66/s1, Figure S1: Illustration of the solute concentration profile at steady state across a thin film composite membrane in FO mode (active-layer facing FS), Adapted from Bui et al. [45], Table S1: Specifications of the Aquaporin Inside TM HFFO2 module used in all the tests. ID, inner diameter; OD, outer diameter, Table S2: Probability values (p-value) from the statistical analysis ANOVA of J w between the two types of HFFO (HF-C, HF-O) and between the tests studied in this article. "Membranes" represents the significant differences in J w between HF-C and HF-O. "Test" represents the significant differences in J w for each test conditions for both HF-C and HF-O. p-values < 0.05 show a significant difference between the groups, Table S3: Probability values (p-value) from the statistical analysis ANOVA of J s /J w between the two types of HFFO (HF-C and HF-O) and between the tests studied in this article. "Membranes" represents the significant differences in J s /J w between HF-C and HF-O. "Test" represents the significant differences in J s /J w for each test condition for both HF-C and HF-O. p-values < 0.05 show a significant difference between the groups.