Power Generation Performance of Reverse Electrodialysis (RED) Using Various Ion Exchange Membranes and Power Output Prediction for a Large RED Stack

Reverse electrodialysis (RED) power generation using seawater (SW) and river water is expected to be a promising environmentally friendly power generation system. Experiments with large RED stacks are needed for the practical application of RED power generation, but only a few experimental results exist because of the need for large facilities and a large area of ion-exchange membranes (IEMs). In this study, to predict the power output of a large RED stack, the power generation performances of a lab-scale RED stack (40 membrane pairs and 7040 cm2 total effective membrane area) with several IEMs were evaluated. The results were converted to the power output of a pilot-scale RED stack (299 membrane pairs and 179.4 m2 total effective membrane area) via the reference IEMs. The use of low-area-resistance IEMs resulted in lower internal resistance and higher power density. The power density was 2.3 times higher than that of the reference IEMs when natural SW was used. The net power output was expected to be approximately 230 W with a pilot-scale RED stack using low-area-resistance IEMs and natural SW. This value is one of the indicators of the output of a large RED stack and is a target to be exceeded with further improvements in the RED system.


Introduction
Recently, salinity gradient energy (SGE), which is generated when two salt solutions with different concentrations are mixed, has attracted attention as a renewable energy source. Reverse electrodialysis (RED) is a technology that converts SGE into electrical energy using ion exchange membranes (IEMs) [1,2]. RED power generation uses the electrochemical potential created by IEMs to generate electricity and has access to a nearly inexhaustible supply of seawater (SW) and river water (RW), thus eliminating issues regarding depletion of energy sources. As RED power generation is largely unaffected by weather conditions and has a high operating rate, it has great potential as a new power generation method.
A RED stack comprises alternating layers of cation exchange membranes (CEM) and anion exchange membranes (AEM) between two electrodes. Gaskets with spacer nets are placed between the membranes to maintain a constant distance between the membranes, create a flow path, and agitate the solutions. An electric potential difference is generated by alternately supplying high-and low-concentration solutions between the membranes. When an external load resistance is connected to the two electrodes, ionic diffusion from the high-concentration side to the low-concentration side is converted to an electric current by a redox reaction at the electrodes; hence, the RED stack generates electricity, i.e., converts SGE to electricity.
The objective of this study is to predict the power output of a pilot-scale RED stack with various commercial IEMs from experiments using a lab-scale RED stack. We evaluated and analyzed the power generation performance of a lab-scale RED stack with 40 membrane pairs and a total membrane area of 7040 cm 2 using five types of CEM/AEMs with different membrane properties, including reference membranes; CIMS and ACS-8T. Model SW, model concentrated SW (RO brine), and natural SW were used as feed solutions. To discuss the differences in the open-circuit voltage of the lab-scale RED stack, the membrane potential of each IEM was measured. This study suggests that the output of pilot-scale RED could be more than doubled by changing CIMS/ACS-8T to IEMs with low-area-resistance.

Membranes
CEM and AEM pairs: Neosepta ® CSE/ASE, Neosepta ® CMX/AMX, Neosepta ® CIMS/ACS-8T, C-2/A-2 (ASTOM Corp., Japan), and Fumasep ® FKS-20/FAS-20 (FU-MATECH BWT GmbH, Germany) were used as CEM/AEM for lab-scale RED stacks. CSE/ASE and CMX/AMX are the current-and previous-generation standard membranes, respectively. The CIMS/ACS-8T are monovalent ion-selective membranes. C-2/A-2 and FKS-20/FAS-20 exhibited low-area-resistance. Membrane properties were measured in the same way as previously described [28,29]. The basic properties of the membranes used in this study are listed in Table 1. For the monovalent ion-selective membranes, CIMS and ACS-8T, the membrane potentials were measured with the selective side facing the high-concentration solution. For all basic membrane properties and membrane potential measurements, all reagents were of analytical grade, and solutions were prepared using ion-exchanged water.  [30]. 0.30, 0.45, 1.50, 2.24, and 3.40 mS cm −1 (≈2.3, 3.7, 13.5, 20.3, and 31.1 mM) NaCl solutions were used as model river water (RW). To evaluate the power generation performance of the RED stack using actual solutions, SW collected at an SW desalination plant (Chatan, Okinawa, Japan) was used as the high-concentration solution (HS) and surface water from a water treatment plant (Chatan, Okinawa, Japan) was used as the low-concentration solution (LS). The electrical conductivity and ionic composition of natural SW and surface water are listed in Table 2. Natural SW and surface water were filtered using a filtration system (AF-4, ZEOLITE Co. Ltd., Fukuoka, Japan) to remove impurities before use. Surface water was adjusted to conductivities of 0.45, 1.50, 2.24, and 3.40 mS cm −1 by adding natural SW. 3.0 M NaCl solution was used as the electrode solution in both the model and natural solutions. All NaCl solutions used for RED power generation tests were prepared using 95% NaCl purchased from Diasalt Corporation (Nagasaki, Japan) and tap water.  Figure 1 shows a schematic of the lab-scale RED stack. Forty pairs of CEMs/AEMs were used for the RED stack. The monovalent ion-selective membranes, CIMS, and ACS-8T were stacked alternatively with the ion-selective surface facing the HS. The effective area of the membrane in contact with the solution per IEM was 88 cm 2 (11 cm × 8 cm). The total effective membrane area of the RED stack was 7040 cm 2 (88 cm 2 × 80). To maintain the inter-membrane distance and prevent solution leakage, a gasket with an integrated spacer of 200 µm thickness was placed between the CEM and AEM. Ag and AgCl were used as cathode and anode, respectively.  Figure S1 shows a flow diagram of the RED stack performance evaluation. Solutions were fed to the RED stack using a magnet pump (MD-30RZ-N, IWAKI Co., Ltd., Tokyo, Japan), and the linear flow velocity (v LS ) in the LS channel inside the RED was set to 2.0 cm s −1 (0.77 L min −1 ). The linear flow velocity was calculated using the following equation:

Evaluation of Power Generation Characteristics
where Q LS is the flow rate in the LS channel, d is the distance between membranes (200 µm), L is the width of the flow channel (8 cm), and N is the number of membrane pairs (40). The flow rate of the HS channel was set such that the pressure at the HS channel inlet was 3 kPa higher than that at the LS channel inlet. The pressures of the HS and LS channels were measured using a digital pressure gauge (KDM30, Krone Corporation, Tokyo, Japan). An electronic load device (PLZ164W, Kikusui Electronics Corporation, Kanagawa, Japan) was connected to the electrodes of the RED stack, and the current (I) was varied from 0 A at a rate of 0.2 A min −1 to measure the voltage (V)-I curve. V in the RED stack is expressed as follows: where V OC is the open-circuit voltage and R int is the internal resistance of the RED stack. V OC is defined as the voltage at zero current in the V-I curve. Contrastingly, assuming that the intramembrane diffusion potential is negligible, V OC can be calculated from the Nernst equation [31].
where t CEM and t AEM are the transport numbers of the CEM and AEM, respectively; t ave is the average of the transport numbers of CEM and AEM; R is the gas constant (8.314 J mol −1 K −1 ); T is the absolute temperature; F is the Faraday constant (96,485 C mol −1 ); γ HS is the activity coefficient of NaCl in HS; γ LS is the activity coefficient of NaCl in LS; c HS is the concentration of NaCl in HS; and c LS is the concentration of NaCl in LS. V OC, theory was calculated using T = 298 K and t ave = 0.98, and the activity coefficient values from Table S1. Although R int is ideally obtained from the slope of the V-I curve (see Equation (2)), the slope of the actual V-I curve in the RED performance test was not constant and decreased with increasing current because the higher the current, the higher the conductivity of the solution on the LS side owing to ionic diffusion. Therefore, R int was defined in this study as the slope of the V-I curve at the current value when the maximum output (P max ) was obtained. From Equation (2), and P = V I, P max is given by [32] The maximum output density (PD max ), which is the maximum output per membrane area, was calculated using the following equation: where A is the effective membrane area (88 cm 2 ).

Comparison of the Open Circuit Voltage
The open-circuit voltage (V OC ) and internal resistance of the RED stack (R int ) are important factors for analyzing the performance of the RED stack. The conductivity of SW or RO brine as an HS depends on the site location of the RED plant. Therefore, examining the effect of conductivity at the LS channel on V OC and R int is an important factor when considering the operating conditions for obtaining a high RED output. Figure S2 shows the V-I curves of the RED stacks with various membrane pairs using model SW and model RW. Figure  In previous studies, V OC per CIMS/ACS-8T pair using a pilot-scale RED stack at various feed flow rates was 0.14~0.17 V using 50 mS cm −1 NaCl as HS and 0.34 mS cm −1 surface water as LS [27]. The V OC per CIMS/ACS-8T pair using the lab-scale RED stack in this study showed almost the same values as those obtained using the pilot-scale RED stack.
The experimental values of V OC for all IEM pairs were lower than the values calculated from Equation (3) for all the κ LS values. In particular, V OC was much lower than V OC,theory , at a smaller κ LS . To examine the cause of the deviation between V OC and V OC,theory , the membrane potential of a diffusion dialysis system consisting of an IEM, and 50 and 1.5 mS cm −1 NaCl solution was measured, as shown in Figure S3A. The absolute values of the membrane potential were approximately 88 mV for all the CEMs and 80 mV for all the AEMs. In previous studies, the absolute value of the membrane potential of CMX was higher than that of AMX [33,34]. The difference in the absolute values of the membrane potential between CEM and AEM is expected to be due to the junction potential between the salt bridges (3 M KCl) and the measurement solutions (50 and 1.5 mS cm −1 NaCl solutions) [29]. The membrane potential generated when 40 pairs of CEMs/AEMs are stacked (total φ m ), can be calculated using the following equation: where φ m,CEM and φ m,AEM are the membrane potentials of the CEM and AEM, respectively. The calculated total φ m is shown in Figure S3B. The calculated values of all IEM pairs used in this study were slightly lower than the ideal membrane potential (≈6.8 V) calculated using Equation (3). Here, the junction potential mentioned above was canceled out when |φ m,CEM | and |φ m,AEM | were added together. There were few differences in the total φ m between the IEM pairs. This indicates that the effect of the difference in the membrane properties listed in Table 1 on the membrane potential is small under the measurement conditions. The deviation between the total φ m (≈6.8 V) and V OC of the stack using the IEMs pairs at κ LS = 1.50 mS cm −1 (5.9~6.3 V) was not negligible. The possible causes of the deviation are as follows: In the case of the membrane potential in the diffusion dialysis system, there is no spacer on the membrane surfaces, and the solutions in contact with the membrane were well stirred. However, there is a spacer on the membrane surfaces in the feed channels of the RED stack. Hence, the effect of concentration polarization on the decrease in the membrane potential in the stack [27] will be higher than that in the diffusion dialysis system. FKS-20/FAS-20 had the highest V OC among all CEM/AEM pairs, indicating that FKS-20/FAS-20 may have the lowest osmotic water flow, which is a cause of the decrease in the membrane potential due to the external concentration gradient.

Comparison of the Internal Resistance
The V-I curves in Figure S2 show that the slope is not straight and decreases as the current increases, although the V-I curve is theoretically straight (see Equation (2)). This is due to the increase in the concentration of the LS channel in the stack because of the diffusion of ions from the HS to the LS when an electric current is generated, resulting in a decrease in the R int of the RED stack. Hence, in this study, R int was defined as the slope of the V-I curve at the current when P max was obtained, as described in Section 2.4. Figure 2B shows the κ LS dependence of R int . For all the IEM pairs, R int decreased as κ LS increased, which was due to the decrease in LS resistance. The value of R int ranged from 6 Ω to 23 Ω. In a previous study, the RED stack with a comparable effective membrane area (100 cm 2 × 50 FKD/FAD (FUMATECH BWT GmbH, Bietigheim-Bissingen, Germany) pairs = 10,000 cm 2 and 200 µm inter-membrane distance) using 30 g NaCl L −1 (≈0.51 M NaCl) as HS and 1 g NaCl L −1 (≈0.017 M NaCl) as LS showed a R int value of approximately 18 Ω [35]. The stack had the same inter-membrane distance (200 µm) and almost the same membrane resistance (CEM; FKD 2.14 Ω cm 2 and AEM; FAD 0.89 Ω cm 2 [6]) as those of our stack. Hence, the stack resistance per pair of CEM/AEM (0.345 Ω) [6] was almost the same as that of our stack (0.15-0.575 Ω). The R int increased in the following IEM pairs order: FKS-20/FAS-20 < CMX/AMX < CSE/ASE CIMS/ACS-8T. The stack using FKS-20/FAS-20 had the lowest R int among all the stacks. This is attributed to the lowest area resistance of FKS-20/FAS-20 in all IEM pairs. In contrast, the highest R int value was obtained when the CIMS/ACS-8T pair was used. In previous studies, the R int of CIMS/ACS-8T was higher than that of CMX/AMX [29], which is consistent with the current tendency. This will be due to the fact that membrane resistance was measured by the alternating current (AC) method; on the other hand, direct current was generated in a RED stack. Owing to the monovalent ion-selective surface of CIMS and ACS-8T, the resistance was higher for direct current than for alternating current.
The RED stack comprised one electrode pair and many cell pairs. The cell comprised the flow channels of the HS, LS, CEM, and AEM. Hence, to analyze the dependence of the electrical resistance of each component on R int, the internal resistance of the RED stack (R int,cal, ) was calculated from the summation of the resistances due to the electrode, HS, LS, CEM, and AEM, including the spacer shadow effect [36]: where R el is the electrode resistance; R HS and R LS are the resistances of the channels at HS and LS, respectively; R CEM and R AEM are the resistances of the CEM and AEM, respectively; β sol is the spacer shadow effect in the solution compartment (1.93); β mem is the spacer shadow effect on the membrane (1.10); κ HS is the conductivity of HS. Details of the calculation of β sol and β mem are provided in the Supplementary Materials. Figure S4 shows the κ LS dependence of R int,cal . For all IEM pairs at all values of κ LS , R LS occupied the highest proportion of R int,cal . The second highest proportion in R int,cal was R CEM + R AEM , and the contribution of R el and R HS to R int,cal was low. The experimental values at κ LS = 0.45 mS cm −1 were lower than R int,cal values in all IEM pairs. As mentioned above, the conductivity of the LS channel inside the RED cells will be higher than that of the LS channel inlet because of ionic diffusion via the IEMs. Simultaneously, the conductivity of the HS channel decreases; however, the contribution of R LS is much larger than that of R HS . Therefore, R int was lower than R int,cal due to the increase in the conductivity of the LS channel. At κ LS ≥ 1.50 mS cm −1 , the R int values of CSE/ASE, CMX/AMX, and FKS-20/FAS-20 were lower than R int,cal , while those of CIMS/ACS-8T were almost the same as R int,cal . This will be due to the fact that the former three IEM pairs had lower resistance than CIMS/ACS-8T (see Figure 2B); hence, the RED stack had a higher electric current than CIMS/ACS-8T. Owing to the high current, the resistance of the LS channel is lower than that calculated. Figure 2C shows the κ LS dependence of the PD max . For CSE/ASE, CMX/AMX, and CIMS/ACS-8T, PD max was maximal around κ LS = 1.50 and 2.24 mS cm −1 . For FKS-20/FAS-20, PD max was maximal around κ LS = 2.24, and 3.50 mS cm −1 . This is due to the decrease in V OC and R int with increasing κ LS (see Equation (5)). PD max increased in the following IEM pairs order: CIMS/ACS-8T < CSE/ASE < CMX/AMX < FKS-20/FAS-20. The small PD max in CIMS/ACS-8T is due to the small V OC and high R int , while the large PD max in FKS-20/FAS-20 is due to the large V OC and low R int . 1.3 W m −2 was especially obtained with FKS-20/FAS-20 at κ LS = 2.24 mS cm −1 . In the past, 1.2 W m −2 has been reported with the RED stack having a comparable effective membrane area (100 cm 2 × 25 FKD/FAD pairs = 5000 cm 2 and 200 µm inter-membrane distance) using 30 g NaCl L −1 (≈0.51 M NaCl) as HS and 1 g NaCl L −1 (≈0.017 M NaCl) as LS [37]. The value in this study, 1.3 W m −2 , to our best knowledge, is the highest power density in the world at lab-scale RED stacks (effective membrane area of 5000~7000 cm 2 ). This will be due to the lower membrane resistance of FKS-20/FAS-20 (0.93 Ω cm 2 at CEM/AEM pair) compared to FKD/FAD (3.03 Ω cm 2 [6] at CEM/AEM pair) in the previous study. Compared to CIMS/ACS-8T (0.77 W m −2 ), FKS-20/FAS-20 had 1.7 times higher PD max at κ LS = 2.24 mS cm −1 .

Comparison of the Maximum Power Density
When using model SW as HS and model RW as LS, the values of V OC were lower than the theoretical values at all κ LS . This is due to the concentration polarization on the membrane surface in the RED stack. FKE-20/FAS-20 had the highest V OC in all CEMs/AEMs used in this study, which indicates FKS-20/FAS-20 may have the lowest osmotic water flow. R int was lowest with FKS-20/FAS-20. This is attributed to the low-arearesistance and large current flow. When using FKS-20/FAS-20, the highest V OC and the lowest R int produced the highest output. Figure S5 shows the V-I curves of the RED stacks using the model RO brine and model RW with various membrane pairs. Figure 3A shows the κ LS dependence of V OC obtained from the V-I curves. For all IEM pairs, V OC decreased as κ LS increased owing to a decrease in c HS /c LS (see Equation (3)). For all IEM pairs, because c HS /c LS of model RO brine was larger than that of model SW, the V OC for model RO brine was larger than that for model SW. V OC increased in the following IEM pairs order: CMX/AMX < CIMS/ACS-8T < CSE/ASE < FKS-20/FAS-20, which is of the same order as V OC using model SW (Figure 2A). The V OC per pair of CIMS/ACS-8T was 0.19, 0.17, 0.16, and 0.14 V at κ LS = 0.30, 1.50, 2.24, and 3.40 mS cm −1 , respectively. In a previous study, the V OC per pair of CIMS/ACS-8T using a pilot-scale RED stack at various feed flow rates was 0.17~0.18 V using 90 mS cm −1 NaCl as HS and 0.34 mS cm −1 surface water as LS [27]. This indicates that not only the use of model SW but also model RO brine as HS, the V OC per pair of CIMS/ACS-8T using the lab-scale RED stack in this study showed almost the same values as those using the pilot-scale RED stack. V OC obtained from the experiment in all IEM pairs was 20% to 40% lower than V OC, theory calculated from Equation (3) (11.8, 8.4, 7.6, and 6.8 V).

Comparison of the Open Circuit Voltage
To examine the cause of the differences in V OC between IEM pairs and the deviation between V OC and V OC,theory , the membrane potential of a diffusion dialysis system consisting of an IEM, and 90 and 1.50 mS cm −1 NaCl solutions was measured and is shown in Figure S6A. The absolute values of the membrane potential were approximately 102 mV for all the CEMs and 91 mV for all the AEMs. The membrane potential generated when 40 pairs of CEMs/AEMs were stacked (total φ m ) is shown in Figure S6B. The calculated values of all IEM pairs (≈7.7 V) were lower than the ideal membrane potential (8.4 V) calculated using Equation (3). This could be attributed to a decrease in the transport number of IEMs owing to the use of a high concentration of NaCl solution (≈1.2 M), and the concentration ratio at the two membrane surfaces will decrease because of the external concentration polarization caused by osmotic water flow. CIMS/ACS-8T had the highest total φ m in all IEM pairs; however, V OC was lower than that of CSE/ASE and FKS-20/FAS-20. The difference between the V OC of the RED stack and the total φ m in the diffusion dialysis system will be due to the difference in the concentration polarization between the RED stack [27] and the diffusion dialysis system mentioned in Section 3.1.1. Figure 3B shows the κ LS dependence of R int obtained from the slope of the V-I curve ( Figure S5) at the current when P max was obtained. R int decreased with κ LS due to a decrease in the conductivity of the LS side in the RED stack. R int increased in the following IEM pairs order: FKS-20/FAS-20 < CSE/ASE < CIMS/ACS-8T < CMX/AMX. R int of FKS-20/FAS-20 were 5.4~6.8 Ω and the values were 30~59% lower than the other IEM pairs. This is attributed to the lowest area resistance of FKS-20/FAS-20 in all the IEM pairs and the increase in the conductivity of the LS channel due to the high current (see Figure S5). Additionally, the κ LS dependence of R int of FKS-20/FAS-20 showed strange trends that were different from other IEM pairs, that is, R int increased with the increase of κ LS at κ LS ≥ 1.5 mS cm −1 . This will be due to the following reason: R int was calculated at the current when P max was obtained. At κ LS = 1.5 mS cm −1 , FKS-20/FAS-20 showed the current value of 0.58 A at P max , while the current values were about 0.52 A and 0.47 A at P max at κ LS = 2.24 and 3.40 mS cm −1 , respectively. Thus, the high current will decrease the resistance of LS channel. Therefore, FKS-20/FAS-20 showed the lowest R int at κ LS = 1.5 mS cm −1 . All IEM pairs except for FKS-20/FAS-20 had almost the same R int value for each other at κ LS = 3.40 mS cm −1 . This is because at the high conductivities, the contribution of the resistance of HS and LS channels on R int will be larger than that of the membrane resistance. Except for the CMX/AMX at κ LS = 1.50, 2.24, and 3.40 mS cm −1 , R int values in all the IEM pairs were lower than those with model SW. Model RO brine has lower resistance than model SW, and the increase in κ LS caused by ion diffusion from the HS channel will be greater than from model SW due to the concentration difference, resulting in above lower R int values. When model SW was used in CMX/AMX, internal leakage in the RED stack may have occurred, which caused an increase in κ LS and a decrease in R int .  The κ LS dependence of the calculated values (R int,cal ) is shown in Figure S7. At κ LS = 0.30 mS cm −1 , R int,cal is large (≈60 Ω) because of the low conductivity of the LS channel. For the same reason as the model SW, the experimental values were much lower than R int,cal at κ LS = 0.30 mS cm −1 . At κ LS ≥ 1.50 mS cm −1 , the R int,cal values were not much different from those of model SW because the contribution of the R HS to R int,cal was low. However, the experimental values (R int ) using the model RO brine were lower than the values of R int,cal because of the increase in the conductivity of the LS channel caused by ion diffusion from the HS channel. Figure 3C shows the κ LS dependence of the PD max . For all IEM pairs, PD max reached maximum values around κ LS = 1.50 and 2.24 mS cm −1 . Owing to an increase in the concentration ratio of HS and LS, the maximum PD max with the model RO brine was larger than that with the model SW for all IEM pairs. PD max increased in the following IEM pairs order: CMX/AMX < CIMS/ACS-8T < CSE/ASE < FKS-20/FAS-20. FKS-20/FAS-20 showed maximum PD max (2.6 W m −2 ) at κ LS = 1.50 mS cm −1 because of the lowest area resistances in all the IEM pairs (see Table 1). Additionally, compared to CIMS/ACS-8T (1.3 W m −2 ), FKS-20/FAS-20 showed 2.0 times larger PD max at κ LS = 1.50 mS cm −1 . In the case of model SW, CMX/AMX showed the second highest PD max ; however, in the case of the model RO brine, they had the lowest PD max . This is due to the lowest V OC and highest R int of the IEM pair.

Comparison of the Maximum Power Density
When using model RO brine as HS and model RW as LS, the values of V OC were lower than the values of V OC,theory , and the difference between V OC and V OC,theory was larger than when model SW was used as HS. This is due to the large difference in concentration between HS and LS, resulting in a large osmotic water flow and a large concentration polarization at the membrane surface. FKE-20/FAS-20 area resistance is low, showing the largest V OC and the lowest R int in all CEMs/AEMs, and produced the highest output. Figure S8 shows the V-I curves of the RED stacks using natural SW and surface water with various membranes. Figure 4A shows the κ LS dependence of V OC obtained from the V-I curves. For all IEM pairs, V OC decreased as κ LS increased owing to a decrease in c HS /c LS (see Equation (3)). The values using natural SW were 1~14% smaller than those using model SW. Natural SW contains K + , Mg 2+ , Ca 2+ , and SO 4 2− , in addition to Na + and Cl − contained in the model SW. In particular, the presence of divalent ions causes a decrease in V OC [11,[18][19][20]38,39]. Although the order of V OC for the IEM pairs varied with the value of κ LS , FKS-20/FAS-20 had the highest V OC (7.0, 6.0, 5.5, and 4.9 V) at all κ LS value. This is the same as for the model SW and the model RO brine, which may be ascribed to the low osmotic water flow. V OC of the other IEMs pairs differed little (about 6.8, 5.6, 5.0, and 4.4 V) except for C-2/A-2 at κ LS = 0.45 and 1.50 mS cm −1 . C-2/A-2 is thinner than other IEMs (see Table 1) and may cause high osmotic water flow, which caused a concentration polarization and the low V OC . Almost equal V OC of CMX/AMX and CIMS/ACS-8T is consistent with the result that there was little difference in V OC when natural SW and distilled water were supplied to the RED stack (10 membrane pairs, 200 µm inter-membrane distance, 1760 cm 2 total effective membrane area) using CMX/AMX and CIMS/ACS-8T [29]. V OC per pair of CIMS/ACS-8T using the lab-scale stack was 0.17, 0.15, 0.13, and 0.11 V at κ LS = 0.45, 1.5, 2.2, and 3.4 mS cm −1 , respectively. In a previous study, V OC per CIMS/ACS-8T pair using a pilot-scale RED stack at various feed flow rates was 0.15~0.17 V using 52 mS cm −1 natural SW as HS and 0.34 mS cm −1 surface water as LS [27].  Figure 4B shows the κ LS dependence of R int calculated from the slope of the V-I curve ( Figure S8) at the current when P max was obtained. For all IEMs pairs, R int decreased as κ LS increased because the resistance of LS decreased. Except for κ LS = 0.45 mS cm −1 , R int values were about 10% larger for CSE/ASE, CIMS/ACS-8T, and FKS-20/FAS-20 and 45% larger for CMX/AMX than those using model SW. This is due to the lower molar mobility of divalent ions in a membrane compared to monovalent ions such as Na + and Cl − , resulting in a larger R int using natural SW containing the divalent ions [18,38,40]. C-2/A-2 had the lowest R int (12, 8.6, 6.3, and 5.1 Ω) in all IEM pairs, with FKS-20/FAS-20 and CMX/AMX having the second lowest and the highest R int , respectively. This result is ascribed to the lowest area resistance of C-2/A-2 (see Table 1) and high current value at P max (see Figure S8).

Comparison of the Internal Resistance
The κ LS dependence of R int,cal is shown in Figure S9. Because the calculations do not consider the presence of multivalent ions, the values of R int,cal in natural SW were almost the same as those in Model SW. As in the case of model SW and model RO brine, the experimental values at κ LS = 0.45 mS cm −1 were lower than R int,cal values in all the IEM pairs. The experimental values using CSE/ASE and CIMS/ACS-8T at κ LS ≥ 2.24 mS cm −1 and using CMX/AMX at κ LS ≥ 1.50 mS cm −1 were larger than R int,cal , probably because of the effect of multivalent ions. In the case of FKS-20/FA-20 and C-2/A-2, the experimental values at κ LS ≥ 1.50 mS cm −1 were larger than R int,cal because the decrease in the conductivity of the LS channel caused by the high current (see Figure S8) may be larger than the effect of multivalent ions. Figure 4C shows the κ LS dependence of the PD max . CSE/ASE, CMX/AMX, and CIMS/ACS-8T had almost the same PD max as each other (about 0.66, 0.71, 0.62, 0.55 W m −2 at each κ LS ). For FKS-20/FAS-20 and C-2/A-2, PD max increased as κ LS increased (from 0.78 to 1.3 W m −2 ). For all IEM pairs, the values of PD max using natural SW were 4~39% smaller than those using model SW because of the smaller V OC and larger R int . The decrease in PD max due to the use of a natural SW has been reported [16,17,19,38]. PD max was largest when C-2/A-2 was used because it had the lowest R int among all IEM pairs (see Figure 4B) except for κ LS = 1.50 mS cm −1 . In particular, 1.3 W m −2 was obtained at κ LS = 3.40 mS cm −1 , and this value was 2.3 times higher than that with CIMS/ACS-8T (0.55 W m −2 ).

Comparison of the Maximum Power Density
When using natural SW as HS and surface water as LS, V OC was lower and R int was higher than when model SW was used, which is due to the presence of multivalent ions. Although C-2/A-2 did not show the highest V OC , C-2/A-2 showed the lowest R int due to the lowest area resistance, and thus produced the largest output in all IEM pairs. Table 3 shows a summary of the best performance of the lab-scale RED stack in this study. V OC per one FKS-20/FAS-20 pair was 5.7/40 = 0.143 V and 7.1/40 = 0.178 V when using model SW and model RO brine, and V OC per one C-2/A-2 pair was 4.5/40 = 0.113 V when using natural SW. Compared to CIMS/ACS-8T as a control, FKS-20/FAS-20 showed 1.7 and 2.0 times higher PD max when using model SW and model RO brine, and C-2/A-2 showed 2.3 times higher PD max when using natural SW. We evaluated the performance of the pilot-scale RED stack with CIMS/ACS-8T (299 membrane pairs, 200 µm inter-membrane distance, 179.4 m 2 total effective membrane area) using model SW (50 mS cm −1 NaCl), model RO brine (90 mS cm −1 NaCl), and natural SW as HS and surface water (the value of conductivity was 0.34 mS cm −1 ) as LS, and obtained a power output of 174 W, 263 W and 111 W, respectively [27]. In the pilot-scale RED stack, the flow rate was set to 21 L min −1 of HS and 32 L min −1 of LS to obtain above power outputs. The pumping energies at the flow rate were 44 W with model SW, 41 W with model RO brine, and 31 W with natural SW, respectively. When we assume that osmotic water flow through the IEMs and the internal salt leakage are comparable between the lab and the pilot-scale RED stacks, V OC in the pilot-scale RED stack in a case of using FKS-20/FAS/20 and C-2/A-2 using model SW, model RO brine, and natural SW are expected to be 0.143 × 299 ≈ 43 V, 0.178 × 299 ≈ 53 V, and 0.113 × 299 ≈ 34 V, respectively based on a proportional calculation of the number of IEM pair. Assuming that the SGE conversion efficiency is equal for the lab and the pilot-scale RED stack and using the PD max ratio to CIMS/ACS-8T, power outputs of 174 × 1.7 ≈ 300 W and 263 × 2.0 ≈ 530 W can be expected in the pilot-scale RED stack with the FKS-20/FAS-20 using model SW and model RO brine at the flow rate of 21 L min −1 of HS and 32 L min −1 of LS. When using natural SW, the pilot-scale RED stack with C-2/A-2 can produce 111 × 2.3 ≈ 260 W. Net power outputs calculated by subtracting the pumping energy were 256 W using model SW, 489 W using model RO brine, and 229 W using natural SW. It is possible that residence time of the solution in the pilot-scale RED stack will increase, and the osmotic water flow will become high, resulting in a slightly lower power output than the prediction. However, it may be possible to achieve the output predicted in this study by trying other methods of increasing RED output, such as using profiled membranes [41][42][43][44][45]. Table 3. Summary of the best performance of the lab-scale RED stack in this study (40 membranes pairs, 200 µm inter-membrane distance, and 7040 cm 2 total effective membrane area).

Conclusions
In this study, we clarified IEM properties that are effective for RED power generation. In addition, we predicted that the pilot-scale RED stack can produce large outputs by using appropriate IMEs. The key findings are described below.
IEMs with low-area-resistance, such as FKS-20/FAS-20 and C-2/A-2, produced high power outputs in all feed solutions due to the high current.
The monovalent selective membranes (CIMS/ACS-8T) did not contribute to an increase in the power output, though they may be effective in preventing IEM fouling.
For RED power generation, IEMs that can generate a high current flow even if V OC slightly decreases due to osmotic water flow will be effective. Of course, IEMs with lowarea-resistance and low osmotic water flow would be able to produce higher power output.
Using appropriate IEMs in the pilot-scale RED stack, we predicted 1.7~2.3 times the power output of the previous study. Although these values are only conjectures, they can be indicators of the power output of large RED stacks.
Supplementary Materials: The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/membranes12111141/s1, Figure S1: Flow diagram of the performance evaluation of the RED stack; Table S1: Sodium chloride activity coefficient at the molar concentrations; Figure S2: V-I and P-I curves using model SW (50 mS cm −1 NaCl) as HS and model RW as LS; Figure S3: φ m measured in 50 mS cm −1 and 1.50 mS cm −1 NaCl solutions combination; Figure S4: κ LS dependence of R int,cal using model SW (50 mS cm −1 NaCl) as HS and model RW as LS; Figure S5: V-I and P-I curves using model RO brine (90 mS cm −1 NaCl) as HS and model RW as LS; Figure S6: φ m measured in 90 mS cm −1 and 1.50 mS cm −1 NaCl solutions combination; Figure S7: κ LS dependence of R int,cal using model RO brine (90 mS cm −1 NaCl) as HS and model RW as LS; Figure S8: V-I and P-I curves using natural SW as HS and surface water as LS; Figure S9: κ LS dependence of R int,cal using natural SW as HS and surface water as LS; Figure S10: Set up of electrode resistance measurement; Figure S11: V-I curves of the stack without IEMs before and after a series of power generation test using model SW and model RW; Figure S12: V-I curves of the stack without IEMs before and after a series of power generation test using model RO brine and model RW; Figure S13: V-I curves of the stack without IEMs before and after a series of power generation test using natural SW and surface water; Figure S14: Spacer net geometry. References [36,46] are cited in the Supplementary Materials.