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

Abstract

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.

1. 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 (FTSH₂O) 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 thin film composite (TFC) hollow fiber forward osmosis (HFFO) membranes. In applications where two liquids are in contact with a membrane surface, such as dialysis, membrane contactors, or gas separation, hollow fiber configuration has been proven to have multiple advantages compared to other form factors [10,11,12]. Also, in traditional filtration processes such as ultra or microfiltration, hollow fiber modules present a competitive solution [1,10]. The most prominent advantages of the hollow fiber module are their high packing density, self-supporting structure and uniform flow distribution facilitating mass transfer and cleanability of the membrane [1]. This indicates that the hollow fiber configuration for the FO process is very promising from a practical and commercial point of view.

According to the available literature (e.g., [13,14]), an ideal FO membrane should consist of an ultra-thin and dense selective layer, ensuring high water permeability and solute rejection. The layer should be deposited on a highly porous and thin support, giving sufficient mechanical stability and minimal resistance towards diffusion of the draw solute; thereby allowing maximal utilization of the available osmotic pressure as internal (ICP) and external concentration polarization (ECP) inhibit mass transfer, reducing the available driving force. In addition, high membrane hydrophilicity tends to diminish fouling propensity [15]. Both ICP and ECP are influenced by the properties of the solutes, hydrodynamics of the fluid, and membrane structure and composition [16]. As a result, selection of optimal process parameters [17,18,19], suitable draw solution (DS) [20], and draw recovery strategy [21] are all important to maximize the overall performance of the FO process. To date, assessments of FO membranes have not been standardized and multiple examples of the testing conditions and testing setup configurations have been reported [4,7,17,22,23,24,25,26,27]. Even more challenging is the testing and comparison of FO membranes in commercial modules. The hydrodynamic limitations of individual module geometries might contribute significantly to the reduction of the performance, thereby rendering comparisons with lab scale membrane coupon testing difficult [28,29].

Regardless of the application, the efficiency and attractiveness of the FO process is heavily dependent on the solute rejection and multiple studies have described mechanisms and parameters responsible for solute rejection in the FO operation. For example, ** et al. [30] examined the rejection of pharmaceutical compounds with a 60 cm² TFC–FO membrane and reported that an increase in water flux by double NaCl DS concentration did not influence forward solute rejection. In contrast, Yamamoto et al. [31] reported an increase in solute rejection, with an increased concentration of MgCl₂ used as a DS. Also, for challenging compounds, such as urea and boron, FO rejection has been tested. For instance, lab-scale investigations of Shon et al. [32] revealed that using a configuration where the DS was facing the active layer (pressure retarded osmosis (PRO) mode) reduced membrane rejection for boron in comparison to the system with the DS facing the support layer (FO mode). The effect of these configurations (PRO and FO mode) were also found to be critical by Park et al. [7], as ECP becomes a key factor of membrane performance. Park et al. [7] minimized the negative effects of ECP by develo** an HFFO membrane with a novel interfacial polymerization (IP) process and by changing the module configuration. The temperature is another extensively studied parameter [24,25,33,34,35,36] that affects membrane rejection and water flux. For example, **e et al. [24] described how temperature increased the diffusion coefficient of the contaminants, and thus solute rejection decreased with increasing temperature. The variance in FO testing conditions and membrane behavior in all of these studies only increases the necessity for systematizing the influence of the operational conditions on HFFO membrane performance.

Cath et al. [37] proposed a standard testing method for an osmotically driven membrane processes. However, the proposed testing method and setup are applicable only for flat sheet membrane coupons. Majeed et al. [38] investigated the testing conditions of HFFO membranes with various DS and different configurations (PRO mode vs. FO mode) of the module. They observed that the performance of the investigated HFFO membranes—quantified as water flux, reverse solute flux, and specific reverse solute flux—are strongly dependent on the flow rates of feed solution (FS) and DS. These tests were carried out using lab scale modules with membrane areas <30 cm². The findings from lab-scale systems are difficult to extrapolate to a pilot plant or full-scale systems. Choi et al. [39] recognized this problem and developed a model to predict membrane performance under different operational conditions for large FO plants. But the model results were only compared to experimental data from spiral wound membranes. Another study in a full scale desalination system by Kim et al. [4] described the relevance of the operational conditions on FO processes and a pilot plant design using spiral wound membranes. They found that due to the high flow rates of the pilot plant, co-current operation of FS and DS facilitated fouling of the membrane. It was also concluded that the initial draw flow rate restricted the maximum number of possible modules in a series, as the pressure build-up on the shell side can break the modules. The transition between lab-scale and full-scale is also challenging for hollow fiber modules. For inner-selective hollow fiber membranes, for which the selective layer is deposited after potting of the fiber into the module [40], non-invasive testing of the membrane performance is only feasible with complete and intact modules. Such modules can have from few to several dozen square meters and a high packing density of the fibers affecting flow conditions. Therefore, a design of a suitable testing protocol for commercial scale HFFO modules is necessary to evaluate their performance under realistic conditions.

In this study, we collected the different operating conditions used in a diverse array of published studies and chose a selection of them to create a guideline for testing HFFO modules. Specifically, we studied the effect of draw and feed flow rates; draw solute and concentration; transmembrane pressure (TMP); temperature; membrane orientation (PRO, FO, counter-current and co-current modes); and solute rejection of target compounds (caffeine, niacin and urea), under FO and reverse osmosis (RO) conditions. These compounds were chosen for their different molecular weights, which facilitated solute rejection analysis. Using those operating conditions, we investigated performance of the commercial inner-selective biomimetic HFFO membrane Aquaporin Inside™ HFFO220 with a surface area of 2.3 m². To show the validity and uniformity of the operating conditions impact on the HFFO membrane modules, Aquaporin Inside™ HFFO220 modules were post-treated to increase their water flux. Following the literature studies [41,42], we used sodium hypochlorite (NaClO) to enhance the water flux of the membrane, and we compared its performance to the standard HFFO220 module. Explicitly, we correlated the difference in the membrane performance to the properties of the selective layer and described parameters allowing control of the FO operation when the HFFO configuration is applied to a pilot or full-scale application.

2. Materials and Methods

2.1. Materials

The membrane modules used are commercially available HFFO220 membrane modules provided by Aquaporin A/S (Kongens Lyngby, Denmark), and their specifications can be found in the Supplementary Materials Section S1. The hollow fiber membrane modules have an active membrane area of 2.3 m². The HFFO membranes consist of a biomimetic TFC selective layer [43] supported by polysulfone fibers with an inner diameter of 195 µm. NaClO from Sigma Aldrich (Søborg, Denmark) was used to modify commercially available HFFO membranes to increase water flux. The salts used for the DS were NaCl and MgCl₂·6H₂O from AzkoNobel A/S (Mariager, Denmark) and MgSO₄·7H₂O purchased from Tab-Sol (Straszęcin, Poland)—all were food-grade quality. For rejection tests, urea (≥99%), niacin (≥98%) and caffeine (≥99%) from Sigma Aldrich were used. All the solutions used in the tests were prepared with deionized water (DI) (σ < 10 µS/cm).

2.2. Methods

2.2.1. HFFO Membrane Modification

HFFO membrane modules were exposed to 20 mg·L⁻¹ NaClO solution at pH 10.5. During the exposure, the solution was recirculated for 2 minutes at the lumen side of the fibers. After the exposure, post-treated modules were intensively flushed with DI water for about 1 h. In this study the 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.

2.2.2. 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, MgCl₂·6H₂O, and MgSO₄·7H₂O 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.

2.2.3. 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 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):

$${\mathrm{J}}_{\mathrm{w}}=\frac{{\mathrm{Q}}_{\mathrm{F},\mathrm{in}}-{\mathrm{Q}}_{\mathrm{F},\mathrm{out}}}{{\mathrm{S}}_{\mathrm{A}}}$$

(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):

$${\mathrm{J}}_{\mathrm{s}}=\frac{{\mathrm{Q}}_{\mathrm{F},\mathrm{out}}·\mathsf{\sigma }·\mathsf{\beta }}{{\mathrm{S}}_{\mathrm{A}}}$$

(2)

where $\mathsf{\sigma }$ is the average conductivity (µS·cm⁻¹) of the concentrate and $\mathsf{\beta }$ is a proportionality coefficient describing the relation between conductivity and salt concentration. Here, $\mathsf{\beta }$ 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.

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. Operational conditions for the standard forward osmosis (FO) tests. During the FO performance tests, only one of the listed testing parameters was modified while other parameters were kept constant. TMP, transmembrane pressure; DI, deionized.

Standard Operational Conditions for the FO Performance Tests
Feed		Draw		Applied TMP	Temperature	Operation Mode
Flow	Solution	Flow	Solution	Bar	°C
100 L·h−1	DI H2O 3.5% NaCl	25 L·h−1	1 M NaCl	0.2	25	FO	Counter-current

. Each parameter was modified one-at-a-time, while kee** 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. Operational parameters modified during FO tests. PRO, pressure retarded osmosis.

Operational Conditions Modified During the FO Performance Tests
Feed Flow	Draw Flow	Applied TMP	Operation Mode		Temperature
L·h−1	L·h−1	Bar	Draw and Feed Flow		°C
60	25	0.1	FO	counter current	15
80	50	0.4	PRO	co-current	25
100	75	0.7			35
120	100	1.1			45
140

.

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. Characteristics of draw solution (DS) used for the FO tests and for estimation of water permeability coefficient A, solute permeability coefficient B and structural parameter S.

Draw Solutions
	Concentration	Osmotic Pressure	Diffusivities [45,47,48]
	M	bar	(× 10−9 m2·s−1)
NaCl	0.5	27.7	1.47
	1	49.5	1.42
	1.5	74.3	1.36
	2	99	1.31
MgCl2	0.32	27.7	1.04
	0.58	49.5	1.07
	0.85	74.3	1.09
	1.11	99	1.11
MgSO4 *	0.3	7.4	0.49
	0.77	24.7	0.41
	1.54	49.5	0.35
	2.53	74.3	0.30

* Concentrations for MgSO₄ differ from the other salts due to the solubility limit of MgSO₄.

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.

2.2.4. 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):

$${\mathrm{c}}_{\mathrm{permeate}}=\frac{{\mathrm{C}}_{\mathrm{D},\mathrm{out}}·{\mathrm{Q}}_{\mathrm{D},\mathrm{out}}-{\mathrm{C}}_{\mathrm{D},\mathrm{in}}·{\mathrm{Q}}_{\mathrm{D},\mathrm{in}}}{{\mathrm{Q}}_{\mathrm{permeate}}}$$

(3)

where ${\mathrm{c}}_{\mathrm{D},\mathrm{out}}$, ${\mathrm{c}}_{\mathrm{D},\mathrm{in}}$ and ${\mathrm{c}}_{\mathrm{permeate}}$ 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):

$${\mathrm{R}}_{\mathrm{FO}}=\left(1-\frac{{\mathrm{C}}_{\mathrm{permeate}}·{\mathrm{Q}}_{\mathrm{permeate}}}{{\mathrm{C}}_{\mathrm{F},\mathrm{in}}·{\mathrm{Q}}_{\mathrm{F},\mathrm{in}}}\right)·100\%$$

(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. Operational parameters used for the membrane rejection analysis of niacin, caffeine, and urea in FO and low-pressure reverse osmosis (LPRO) processes. The parameters were modified to maintain the same process recoveries between the experiments. HF–C, chlorinated membrane; HF–O, original membrane.

Niacin - Caffeine - Urea Tests *
Membrane	FO			LPRO		Recovery
HFFO	Feed Flow	Draw Flow	TMP	Feed Flow	TMP	FO	LPRO
	L·h−1	L·h−1	bar	L·min−1	bar	%
HF–C	140	16	0.2	1	2	23 ± 4
HF–O	135	16	0.2	0.5	2	23 ± 4

* All compounds were tested with 200 mg·L⁻¹ of niacin, caffeine or urea in the feed solution.

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. Testing conditions used for investigation of the correlation between recovery and urea rejection in the FO process.

HF–C and HF–O Membrane Modules
Urea FO Test			Recovery
Feed Flow	Draw Flow	TMP	HF–C	HC-O
L·h−1	L·h−1	bar	%	%
60	25	0.2	74.1 ± 3.5	59.9 ± 4.7
80	25	0.2	57.5 ± 3.5	41.0 ± 1.4
100	25	0.2	46.0 ± 0.0	35.0 ± 1.4
120	25	0.2	40.0 ± 0.0	29.5 ± 2.8
140	25	0.2	34.2 ± 1.0	25.5 ± 0.7

).

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):

$$\mathrm{TMP}=\frac{{\mathrm{P}}_{\mathrm{in}}+{\mathrm{P}}_{\mathrm{out}}}{2}$$

(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):

$$\mathrm{A}=\frac{\mathrm{V}}{{\mathrm{S}}_{\mathrm{A}}·\mathrm{TMP}·\Delta \mathrm{t}}$$

(6)

where V is volume of the collected permeate sample and $\Delta \mathrm{t}$ 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):

$${\mathrm{R}}_{\mathrm{RO}}=1-\frac{{\mathrm{C}}_{\mathrm{permeate}}}{{\mathrm{C}}_{\mathrm{F},\mathrm{in}}}$$

(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.

3. 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.

3.1. 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.

3.1.1. 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.

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.

3.1.2. 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).

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.

3.1.3. 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, MgCl₂, and MgSO₄ 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. Parameters A, B, and S estimated for NaCl, MgCl₂, and MgSO₄ draw solute systems. RMSE and R² 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	A	B	S	RMSE	R2-Jw	R2-Js	Remarks
	L·m−2·h−1·bar−1	L·m−2·h−1	mm	-	-	-
HF–O	1.56	0.24	0.15	0.54	0.99	0.76	-
HF–C	4.04	1.29	0.26	0.77	0.98	0.79	Out of range
	2.46	0.80	0.18	1.22	0.92	0.95	Fixed A = 2.46
Draw Solute MgCl2
HF–O	1.86	0.07	0.16	0.40	0.99	0.02	-
HF–C	2.74	0.59	0.15	0.74	0.97	0.64	-
Draw Solute MgSO4
HF–O	1.68	0.01	0.12	0.43	0.97	0.11	-
HF–C	2.04	0.10	0.08	0.21	1.00	0.78	Out of range
	2.46	0.12	0.09	0.44	0.98	0.50	Fixed A = 2.46

. 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 MgCl₂ 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 R²-J_w (from 0.98 to 0.92) and significantly better R²-J_s (from 0.79 to 0.95). The fact that the RMSE was reduced and R²-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 MgSO₄. 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 ** the DS cold (reduced draw solute diffusion to the FS). In addition, **e 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.

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.

3.2.7. 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.

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, **e 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. **e 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.

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.

4. 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.