High Performance Hydrogen/Bromine Redox Flow Battery for Grid-Scale Energy Storage

Bromine redox exchange-current densities were calculated from polarization data collected with the RDE. The kinetically limited current (I_K) was obtained from the total current (I_T) in the IR-corrected polarization data using²⁶

where I_D is the diffusion-limited current. Log I_K was plotted versus the IR drop-corrected overpotential (η). The linear region of this plot was extrapolated to zero η to obtain the log of the exchange current, as shown in Figure 3. Exchange currents measured with the three electrode materials in 1 M HBr, 0.24 M Br₂ solution are given in Table III.

Zoom In Zoom Out Reset image size

From Table III, bromine redox-reaction rates are much higher on platinum than on either carbon, although they are still high on carbon compared to many other redox chemistries.⁵ However, platinum is expensive, was found to dissolve in HBr,²⁷ and the kinetics are sufficient on carbon alone. Thus, the decision was made to use carbon without Pt as the bromine electrode.

The bromine reaction on carbon can be improved by increasing the surface area available for reaction. It is of interest to estimate the surface area that would be required in a porous-carbon electrode to obtain an energy efficiency equivalent to that of a typical platinum-catalyzed RFB electrode. In the best-case scenario of kinetically-limited polarization (i.e., no ohmic or mass-transfer limitations), an estimate can be made using the kinetically-limited current versus IR-corrected voltage data obtained from the RDE measurements.

The calculation of the bromine-electrode voltage can be exemplified in the case of the Pt-catalyzed CL. A typical platinum-catalyzed CL has a roughness factor on the order of 100 cm² Pt surface area per cm² geometric area.²⁸ To account for the Pt loading, we multiplied I_K by 100 in the I_K vs η current-density data for platinum derived from the Pt RDE polarization experiments in 1 M HBr, 0.24 M Br₂, and obtained the I_K vs η data for a CL with a Pt loading of 100 cm² Pt per cm² geometric area. By taking the iR-corrected bromine electrode potential measured in RDE experiments (0.997 V at 0.01 A/cm²) and subtracting the hydrogen-electrode potential (−0.012 V vs. NHE at STP, obtained from Nernst equation with an assumption of 0.4 M H⁺ in Nafion), the corresponding voltage of a hydrogen bromine RFB charging at 1 A/cm² (given the multiplication factor due to Pt loading discussed above) was calculated to be 1.009 V. The corresponding data for a GC-catalyzed CL is obtained again by multiplying the experimentally obtained GC I_K vs η current-density data by 100 to 20000 to compare the performance of the two catalysts at loading ratios (GC/Pt) of 1 to 200. The result of this comparison at 1 A/cm² is shown in Figure 4.

Zoom In Zoom Out Reset image size

As indicated in the figure, to obtain a kinetically limited performance within 2% of that for a Pt-catalyzed CL, a GC electrode requires about 20 times more surface area per cm², or 2000 cm² carbon surface area per cm² geometric area. Carbon blacks have specific surface areas ranging up to nearly 1000 m²/g.²⁹ Thus a 2000 cm² carbon surface area per cm² geometric should be easily achievable; a CL carbon loading of only 2 mg/cm² with a specific surface area of only 100 m²/g would meet this requirement. Finally, it should also be noted that although Pt is used on the hydrogen electrode, it has been shown in hydrogen fuel cells that the fast kinetics allow for a much smaller amount to be used (∼0.05 mg/cm²) than what is used in this study with negligible effect on performance up to 2 A/cm².³⁰

Three different electrode structures were prepared for the (+) side to investigate their effect on cell performance, as illustrated in Figure 5: Pt/C-CL electrode (i.e. PM∥Pt/C|MEM|Pt/C∥PM), C-CL electrode (i.e. PM∥Pt/C|MEM|C∥PM), and C-PM electrode (i.e. PM∥Pt/C|MEM∥PM)^c where Nafion NR212 was used for proton-exchange membrane (MEM), the left and right sides of MEM represent the (−, H₂) and (+, Br₂) sides, and for PM, SGL 25BC and 25AA were used for the (−) and (+) sides, respectively. As shown in Figure 5, the CL-type electrode has catalyst particles mixed with ionomer to form a laminate, which is then thermally bonded onto the membrane. The PM-type electrode is a porous carbon-medium sheet that contacts the bare membrane physically via a compressive force during cell assembly. 0.9 M Br₂/1 M HBr solution was fed to the (+) side at 200 mL/min while fully humidified hydrogen at a stoichiometric ratio of 3 was provided to the (−) side, and cell tests were conducted at 20°C and ambient pressure.

Zoom In Zoom Out Reset image size

As shown in Figure 6 and in agreement with the RDE results in the previous section, there was no significant performance drop in the low-current-density region (i.e. less than 0.2 A/cm²) for the case of the Pt/C-CL electrode, indicating activation loss (i.e. kinetic loss) is nearly negligible due to the fast kinetics of bromine on Pt. The maximum power density achieved was 0.7 W/cm² which is 3 to 4 times greater than that of typical RFBs in the literature,³¹ and comparable with recent reported work with the all-vanadium system at elevated temperatures.³²

Zoom In Zoom Out Reset image size

For the case of the C-CL electrode in which there is no platinum and only carbon particles, again, no significant voltage drop in the kinetic region was observed, while there was substantial voltage drop in the higher-current-density region (beyond 0.9 A/cm²) wherein mass-transfer becomes the dominant cell-performance limitation. The kinetic-related performance can be explained with the reaction area of each electrode type, as described in the RDE test results. The carbon particle in the C-CL electrode is calculated to have almost 20 times greater area than platinum in the Pt/C-CL electrode.^d Thus, the bromine reaction on carbon became comparable to platinum due to the greater reaction area (see Figure 4). The significant mass-transfer loss at higher current densities is attributed to the surface wettability of the electrode to the aqueous bromine electrolyte, where Pt-containing CLs seem to be more hydrophilic.³³ Thus, the hydrophobic property in the C-CL electrode may impede the transport of aqueous bromine to the reactions sites in the CL, thus resulting in lower cell performance.

For the case of the C-PM electrode (i.e. SGL 25AA carbon paper), there was a significant kinetic loss at low current densities. This loss is probably due to the smaller reactive surface area of the C-PM electrode, which has a highly porous structure (porosity of 90%) with carbon-fiber diameters of 7 to 10 μm, which are several orders of magnitude greater than the diameter of the Pt (2 to 6 nm) and carbon particles (30 to 50 nm) in the CLs.^34,35 In addition, the carbon-paper fibers are more graphitic than the carbon in the CL, which may result in slower kinetics. The reaction area for the C-PM electrode was calculated to be less than 4% of that of the C-CL electrode when using a fiber-filament model³⁶

where a is the specific surface area (cm²/cm³), ε is the porosity, and d_f is fiber diameter (cm).

Although the low-current-density performance is less with the C-PM electrode, the high current-density region was not significantly affected by additional mass-transfer losses, in contrast to the case of the Pt/C-CL electrode. This lack of change is probably due to better bromine accessibility to the reaction sites due to the highly porous structure and relatively good hydrophilicity of the C-PM electrode.

To examine the mass-transfer effects, two different flow-field structures were examined; namely, a single serpentine flow field (SPFF) and a solid flow field (SFF), as shown in Figure 7. In the SPFF, the electrolyte flows by the PM and the aqueous bromine accesses the surface of the electrode mainly by diffusion and perhaps some convection around the U-shaped bends (i.e. Flow-By Mode), whereas for the SFF, the electrolyte is forced to flow convectively through the PM to reach the reaction site (i.e. Flow-Through Mode).

Zoom In Zoom Out Reset image size

The cell performance of each flow-mode structure was compared with the cell consisting of 25BC∥Pt/C|NR-212∥25AA. For the case of flow-by mode (i.e., Figure 7a), the maximum cell performance was 0.47 to 0.51 W/cm², and the maximum current density was around 1.1 A/cm², as shown in Figure 8a. As the electrolyte flowrate increased, the cell performance was enhanced by 8.5% due to improved mass transport, which probably resulted in more convective transport within the PM (i.e., convection under the ribs around the serpentine bends). By changing the flow mode from flow-by to flow-through (i.e., Figure 7b), the maximum cell power density was enhanced by 23% (from 0.51 to 0.63 W/cm²), and the maximum current density was improved significantly from 1.1 to 1.4 A/cm², as illustrated in Figure 8b. The increase in performance is probably due to the forced convection of the electrolyte toward the membrane, which facilitates the transport of Br₂ and HBr to and from the C-PM electrode, respectively. This is also seen in that there was no noticeable enhancement in the kinetic-dominant region (i.e. at low current densities), thus indicating the reaction kinetics was not improved significantly with the change of flow mode.

Zoom In Zoom Out Reset image size

To elucidate the differences between the two designs, one can correlate the reaction rate to the underlying bromine transport phenomena³⁷

where, i is current density, n is equivalent electrons per mole of reactant, F is Faraday constant, D_j is diffusivity of component j, C_j is concentration of component j, and u_z is velocity of component j in the direction to the electrode. In the flow-by system, the bulk motion and convection near the electrode is small relative to the diffusion, and the consumption rate of reactant can be generally correlated with diffusive transport. For the case of flow-through mode, transport of reactant and product take place due to forced convection. Therefore, the bulk motion and convection near the electrode becomes dominant in Eq. 6. This idea is also in-line with the simplified simulation results shown in Figure A2, where the diffusive boundary layer is visible for the flow-by mode and there is more convection and reaction in the flow-through one.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To understand further the differences in mass transfer due to flow mode, additional analysis of the polarization curves is conducted. HFR measured during operation yields the total ohmic resistance associated with the cell components (i.e. electronic resistance through solid components, interfacial or contact resistance between them, and proton resistance in the membrane) and hardware connections between the cell and load bank. Therefore, the ohmic drop related to proton transport in the membrane can be determined by subtracting the total ohmic resistance with the resistance obtained from the same-cell test-setup but without membrane, which was measured in DC mode separately. It should be noted that ohmic losses associated with ionic conduction in the mixed-conducting or porous electrodes are not measurable using HFR as it measures the path of least resistance for the current (i.e., via electron conduction).³⁷

As shown in Figure 9a, the proton resistance was around 0.15 and 0.22 Ω-cm² for flow-through and flow-by modes, respectively, which is about 2 or 4 times greater than that expected for a liquid-equilibrated Nafion membrane.⁶ This behavior is considered to be related to the fact that Nafion conductivity decreases as its free-acid content increases,^6,11,38–40 and thus the effect of flow mode and electrolyte flowrate on the conductivity might be interpreted with this characteristic behavior. As the flow mode changes from flow-by (i.e. diffusive transport) to flow-through (i.e. forced convective transport) and the flowrate increases, the local buildup of the generated HBr in the cell could be mitigated, generating a relatively less acidic environment surrounding the membrane, and thus increasing the membrane conductivity, consistent with Figure 9a. In fact, this is what the simulation results demonstrate in Figure 9b, which is calculated according to the simplified model described in the Appendix. In Figure 9b, one can see that although more HBr is produced in the flow-through case, the flow-by one results in a substantial HBr concentration next to the membrane due to the fact that removal of HBr is essentially only by diffusion for the flow-by mode.

Zoom In Zoom Out Reset image size

To understand the non-measurable components of the polarization curve, the kinetic and linear parts of the curve were fit by

Where η_{IR, mem} and η_{IR, HW} are the losses associated due to membrane conduction, and contact resistances and electron conduction, respectively, which together compromise the measured HFR, η_{IR, elect} are the losses associated with proton transport in the liquid electrolyte, and η_{act, j} is activation (kinetic) loss of component j on the electrode,

where, R is universal gas constant, T is the cell temperature, a is the specific interfacial area, L is the electrode thickness, and α is the transfer coefficient which is set to a value of 1 for the two reactions.^41,42 It should be noted that the solution concentration change was only about 1% after discharge operation due to the use of the high volume of electrolyte; therefore, the effect of concentration change in this analysis is not considered explicitly. In equation 7, both the activation potentials for the H₂ and Br₂ reactions are accounted for, where ai₀L for H₂ is taken from literature (0.1 A/cm²)⁴¹ and for Br₂ it is fit to the experimental polarization curve (0.0017 A/cm²). Also, the Tafel expression in equation 8 may overestimate the reaction overpotential since for fast reactions a full Butler-Volmer expression is expected; for this reason, the very beginning points of the curves were not fit. It should be noted that while η_{IR, HFR} is given by the HFR, η_{IR, elect} is the pseudo-iR loss associated with mass transport (i.e. ionic transport) within the porous electrodes which cannot be measured with HFR, as mentioned above.

Figure 10 shows that the cell performance for the flow-through-mode case (see Figure 8b) can be separated into the various constituent overpotentials using Eq. 7; where, for high current densities, the nonlinear portion of the curve results from additional mass-transfer limitations that are probably related to Br₂. Due to the use of Tafel expressions for the kinetics, the overpotentials at low current densities go higher than observed as mentioned above. From the figure, it is clear that at mid-range current densities, the resistance is due both to kinetic and ohmic factors, with the bromine kinetics and membrane conduction being the largest fraction of each. To understand the flow-mode and mass-transport effects leading to the pseudo-IR response better, it is instructive to do a simple analysis. If one assumes that Br₂ is limiting the reaction, then the reaction proceeds at the back of the electrode up to a maximum penetration depth of bromine into the electrode. In this case, this mass transfer will result in a pseudo-iR response that is due to the proton movement from the membrane to this penetration depth. From fitting the two flow-mode-case curves using Eq. 7, the pseudo-ohmic electrolyte resistances for the flow-through and flow-by modes were calculated to be 0.05 and 0.09 Ωcm², respectively. If those values are considered to be solely due to the ohmic loss due to proton conduction in the electrolyte, then, using Ohm's law and the conductivity of the HBr/Br₂ solution (0.15 S/cm),⁴³ the maximum bromine penetration depths are determined to be 75 and 145 μm from the membrane for the flow-by and flow-through cases, respectively. Thus, this simple analysis shows that there is perhaps twice as much penetration of Br₂ in the flow-through than the flow-by mode, which is consistent with the observed higher limiting current and Figure A2. This also agrees with the fact that the liquid flows in the flow-through design penetrate further and perhaps remove the HBr better to reduce the membrane ohmic resistances as discussed above (see Figure 9a).

Zoom In Zoom Out Reset image size

It is worthwhile to mention the scalability of the flow-through mode for actual application in a RFB system. The pressure drop and channeling/preferential flow of electrolyte are key parameters to be considered for the actual application. In the tests reported herein, the flow-by and flow-through modes had similar pressure drops (i.e 16 versus ∼12 psig at 200 mL/min of electrolyte, respectively). The similarity is due to the longer flow path for the flow-by serpentine channel (668 versus 32 mm) and the highly porous nature of the carbon electrode (∼90% porosity). As one scales up, there may be some pressure-drop concerns, which can be mitigated by moving toward perhaps interdigitated^44,45 or rectangular flowfields. The other potential problem of the flow-through mode for actual application is related to channeling and preferential flow of electrolyte, which minimizes electrode utilization. However, this issue can be mitigated by proper manifold design, as has been shown in liquid-fed fuel cells.⁴⁶

The above two sections describe improvements to the flowfield (i.e. flow mode) and the ability to use C-PM as an alternative to Pt for the bromine electrode. These two studies were done concurrently, and it is of interest to put them together for an improved cell design using a lower-cost electrode material. As discussed above, the surface area of the C-PM needs to be increased to offset its slightly lower kinetics. In addition, as seen in Figure 10, there is still a substantial kinetic overpotential for the bromine electrode. To increase the active surface area, two sequential procedures were undertaken. First, a stack of 3 SGL 10AA PM was used as the bromine electrode to increase the surface area for reaction. Second, the electrode was pretreated by submersion in 99.99% H₂SO₄ for 5 hours at ambient conditions, which is thought to help increase the catalytically active area as well as perhaps the wetting characteristics of the electrode.⁴⁷ These electrodes in combination with the SFF give the cell performances displayed in Figure 11. As can be seen, the stack of SGL 10AA electrode is on par with that of the Pt/C CL (i.e., ∼0.7 W/cm² peak power density), but with the higher maximum current density (see Figure 6) afforded by the flow-through design. In addition, the pretreatment of the electrode further enhanced both the kinetic and mass-transport parts of the polarization curve with a maximum power density of 0.92 W/cm² and a limiting current density greater than 1.5 A/cm²; the exact effects of pretreatment are currently under investigation.

Zoom In Zoom Out Reset image size

Cell performance is expected to increase due namely to changes in the transport properties (conductivity and diffusion coefficients), and, to a lesser extent, the electrochemical kinetics. To investigate the impact of temperature, the above best-performing cell (i.e. multi-layered C-PM highly activated by pretreatment and flow-through flow mode) was tested additionally at 40 and 55°C. As shown in Figure 12, the maximum power density increased from 0.85 W/cm² at 20°C to 1.13 W/cm² at 40°C, and to 1.4 W/cm² at 55°C, and the maximum current density increased to 2.5 A/cm² at 55°C; this cell performance is some of the highest reported values for RFBs.

Zoom In Zoom Out Reset image size

To investigate the cell losses, the effect of temperature on the HFR and the subsequent HFR-corrected polarization curves are compared further in Figure 13. As shown in Figure 13a, the HFR, which is dominated mainly by conduction through the membrane, significantly decreased as the temperature increased, with the average membrane resistance (i.e. after compensating for the ohmic resistance associated with the hardware) going from 0.06 Ωcm² at 20°C, to 0.04 Ωcm² (33% decrease) at 40°C, and to 0.03 Ωcm² (50% decrease) at 55°C. This change is slightly higher, but consistent with the expected change in Nafion conductivity from 20 to 55°C.⁴⁸ It is also interesting to note that the HFR increases with current density, which may be due to anode dryout or the issues with HBr dehydrating the membrane as discussed above, where the latter mechanism may be more likely since the increase decreases with temperature.

Zoom In Zoom Out Reset image size

To investigate the temperature effect on the rest of the cell losses, the HFR-compensated performance is shown in Figure 13b. Due to the linearity of the data, a line was fit to them and the change in slope was compared, which is a measure of the other, primarily mass-transfer, resistance. The calculated resistances are: 0.19 Ωcm² at 20°C, 0.12 Ωcm² (37% decrease) at 40°C, and 0.08 Ωcm² (58% decrease) at 55°C. These values are higher than the HFR ones, showing that for the improved cell the membrane and contact resistances are roughly a third of the total resistance. Similar to the membrane resistance, increasing temperature greatly reduced the other cell resistances.

The above studies focus on the discharge performance of the RFB cell. To understand charge operation, additional experiments were conducted using the improved RFB cell. Constant current mode was utilized for discharge and charge experiments with an electrolyte of 0.9 M Br₂/1 M HBr. The polarization tests of charge and discharge were repeated three times while recirculating the Br₂/HBr electrolyte, and the repeatable performances are compared in Figure 14.

Zoom In Zoom Out Reset image size

As illustrated, the polarization curves for discharge and charge are nearly symmetric around the OCV potentials. This indicates that the charge and discharge behavior of the H₂/Br₂ RFB are nearly reversible, in contrast to other RFB systems where charge and discharge were asymmetric.³¹ In addition, the charging curves do not demonstrate the presence of significant side reactions (a sudden upturn in current with increase in voltage) even up to 2.5 A/cm² for 55°C operation, thus indicating that the potential is still below that of oxygen evolution and carbon oxidation on these materials and under these conditions and concentrations.^49,50 From Figure 14, one can also estimate the round-trip voltaic efficiency of the cell

For example, at a constant-power charge and discharge of 0.4 W/cm², V_η = 82, 88, and 91% at cell temperatures of 20, 40, and 55°C, respectively, which is one of the highest such efficiencies reported to our knowledge for a RFB at reasonable power.

To understand the RFB rate capabilities, full discharges at various current densities were conducted under ambient conditions. An amount of solution of 186 mL of 0.9 M Br₂/1 M HBr was utilized so that it could be discharged in 1 hour at 0.9 A/cm², i.e. a 1C rate; thus a discharge rate at 4.5A would take 2 hours (i.e., C/2). As illustrated in Figure 15, the cell potential dropped gradually with discharge, following a Nernstian potential, and then dropped precipitously at around the theoretical time for the full discharge. The change of concentration can be seen in the color change of the storage solution before and after the discharge (Figure 15). The bromine utilization is calculated by

where t_actual is actual time for discharge and t_theoretical is discharge time calculated from Faradary's law. The calculated values are about 93% for 0.5 C and 1 C, 87% for 1.2 C, and 4% for 1.4 C. Thus, at modest rates, the RFB uses most of the solution, which is consistent with the polarization-curve analysis above (see Figures 9b and 10). Furthermore, the RFB cell demonstrates hardly any utilization loss over several rates, thus widening the operating window for its applicability to energy-storage requirements for different applications and ramp-rate and charge/discharge requirements.

Zoom In Zoom Out Reset image size