Analysis of an Electrochemical Filter for Removing Carbon Monoxide from Reformate Hydrogen

We studied the operation of a twin-cell electrochemical ﬁlter for removing carbon monoxide (CO) from reformate hydrogen by periodically adsorbing and then electrochemically oxidizing CO on the electrode. During the adsorption step, we studied the effects of feed CO concentration, ﬂow rate, electrode catalyst loading, type of feeder gas, and temperature on CO breakthrough. We then applied a ﬁxed bed adsorber model to show that the breakthrough time could be accurately correlated to the adsorption-step operating parameters. Since the oxidation step was found to be much faster than that for CO breakthrough, adsorption time should dictate switching time. This insight was used to predict steady-state ﬁlter performance, and the prediction was validated for an electrochemical ﬁlter operated with CO contaminated hydrogen to decrease the CO concentration from 10,000 to 10 ppm. The model was also used to explore the employability of an electrochemical ﬁlter over a range of operating conditions by considering the comparative electrode area of a ﬁlter with that of a fuel cell. the the

Proton exchange membrane fuel cells (PEMFC) efficiently convert chemical energy in hydrogen (H 2 ) to electrical energy. However, the low volumetric energy density of H 2 makes the fuel storage and transport cumbersome. Alternatively, H 2 can be generated at the point-ofuse from hydrocarbons via catalytic steam reforming. 1 The reformate gas has carbon monoxide (CO) concentrations as high as 10,000 ppm, 2 whereas the platinum (Pt) anode, typically used in PEMFC, is susceptible to poisoning at concentrations as low as 10 ppm. [3][4] In water-gas shift catalysis, typically used for removing CO from H 2 , improving the selectivity of oxidation of CO over H 2 faces thermodynamic limitations below a CO concentration of 1,000 ppm. 5 Mitigation strategies such as operating fuel cells at elevated temperatures, 6 air-bleeding, 7-8 anode alloying [9][10][11] and in situ potential pulsing 4,12-13 have been explored to minimize the impact of CO poisoning. But the tolerance level of PEMFC has been improved only to a CO concentration of 100 ppm. 9,[14][15][16][17][18][19][20] In general, CO contamination in H 2 is handled by employment of CO removal techniques such as pressure swing adsorption (PSA), [21][22] or membrane separation, 23 or conversion of CO into methane in catalytic methanation [24][25][26] or carbon dioxide (CO 2 ) in preferential oxidation (PrOx) with air bleed. [7][8]14,[27][28][29][30][31][32][33][34][35][36][37][38][39] However, the consequent increase in cost due to fuel loss, 7 power loss, 40 components and space requirements 41 hampers the utilization of aforementioned techniques for commercially viable systems, especially in portable power applications. Balasubramanian et al. proposed an electrochemical filtering technique in which CO in reformate is concentrated by preferentially adsorbing on an electrocatalyst bed. Instead of purging as done in PSA, the adsorbed CO is electrochemically oxidized to CO 2 . 42 Also, the ambient-pressure operation of this technique avoids the loss of power in compressing the adsorber bed as in PSA. However, the inability to simultaneously keep both parasitic power losses and exit CO concentrations acceptably low during the application of potential pulses limits the technique's utility. As an improvement over this technique, we developed a twin-cell electrochemical filter that enabled us to achieve both performance metrics. 43 Here, we measure and analyze the effects of temperature, flow rate, feed CO concentration, the type of feeder gas, and catalyst loading on the twin-filter operation. We apply a fixed bed adsorber model to estimate model parameters for describing CO adsorption and then validate the model by comparing simulation results to performance data from an electrochemical filter. Finally, we use the model to explore * Electrochemical Society Fellow. a Present address: Department of Chemical Engineering, Indian Institute of Technology-Madras, Chennai, India. z E-mail: weidner@engr.sc.edu the employability of electrochemical filtering by taking into account the comparative electrode area of a filter and a fuel cell it can feed over a range of operating conditions.

Theory
In the twin-cell filter design (Figure 1), two electrochemical cells (F-1 and F-2) undergo alternating cycles of CO adsorption and electrooxidation. During adsorption mode, CO contaminated reformate H 2 is passed into the anode chamber of F-1 at open circuit. Carbon monoxide in reformate adsorbs onto the catalyst (e.g., Pt) as in the reaction 1.
The gas exiting the anode of F-1 would have decreased levels of CO concentration due to CO adsorption on the catalyst sites. After a certain duration (which we will refer to as the switching time), Figure 1. Schematic conception of a twin cell filter setup describing the oxidation and adsorption steps of the filter cells.

E232
Journal of The Electrochemical Society, 162 (10) E231-E236 (2015) the reformate flow is switched to F-2 for CO adsorption, while a potential pulse is applied to F-1 for oxidizing the adsorbed CO from the previous cycle. The switching time between the cycles of adsorption and oxidation is chosen such that the CO concentration at the filter exit is always maintained at a desirable level for fuel-cell operation. The applied potential is chosen to ensure that Pt anodes are regenerated for continuous operation of filter cells. The application of potential pulse oxidizes the adsorbed CO as in reaction 2.
The oxidation of CO trapped during filter adsorption mode leads to the regeneration of Pt sites for CO adsorption during the adsorption mode of the next cycle. During the oxidation mode, H 2 co-adsorbed or present as gas in the anode compartment also undergoes oxidation as in the reaction 3.
The corresponding cathode reaction is, The H 2 generated at the filter cell's cathode can be sent back to fuel cell with no loss of H 2 as is the case of catalytic methanation or preferential oxidation. Alternatively, oxygen (or air) could also be fed to the filter cathode (as in Figure 1) and the resulting oxygen reduction reaction at the filter cathode would then lower the operating potential of the filter, but will result in a net consumption of hydrogen. The choice of the cathode feed would probably be application specific, and is beyond the scope of this paper. Fixed bed adsorber model.-The cell during CO adsorption on the Pt anode is modeled as a fixed bed adsorber under isothermal conditions. Assuming axial dispersion and porosity effects are negligible, a differential material balance for CO concentration along the serpentine channels of the filter is written as (variables are defined in the List of Symbols), The rate of CO adsorption is limited by various resistances: external film diffusion resistance, internal pore diffusion resistance, surface diffusion resistance and surface adsorption kinetic resistance. A sum of all of these resistances that impede the adsorption of CO can be represented by a parameter 1/k. Assuming a linear relation between gas phase CO concentration(C) and adsorbed phase CO concentration(Q), the adsorption rate equation is then written as Rewriting the governing Equations 5 and 9 in dimensionless form with normalized concentrations of CO in gas phase (c = C/C in ) and adsorbed phase (θ = Q/Q s ) in reference to inlet gas concentration (C in ) and saturated adsorbed phase CO concentration (Q s ), respectively, results in, Q s is estimated from stripping cyclic voltammogram (SCV) of a CO saturated electrode as described in the Experimental section and also in the reference. 42 τ and ξ are dimensionless time and channel length variables and are defined as, and ξ L is evaluated at z = L. A solution is obtained by solving the coupled PDEs in Equation 10 along with the dimensionless form of boundary conditions in Equations 6-8 for the CO concentration at the exit of the filter (i.e. z = L). [44][45][46] In Equation [13], I o is the modified Bessel function of first kind with zeroth order.

Experimental
The membrane electrode assembly (MEA) was procured from Lynntech Inc. The MEA has two platinum black electrodes, each with a geometrical area of 25 cm 2 , coated on either side of a Nafion-115 membrane. Unless specified a Pt loading of 4 mg/cm 2 was used in the electrode. The gas diffusion layer (GDL) used was a woven carbon cloth without any microporous carbon coating on it. The MEA along with two GDLs (one for each electrode) were assembled into filter cells' hardware, which has triple serpentine channeled graphite flow fields (Fuel Cell Technologies Inc.). The gas flow, temperature and pressure were monitored using a test station (Fuel Cell Technologies Inc.), while the relay mechanism used to switch gas flow and potentiostat channels was developed in house. The CO concentration in the gas stream at the exit (C out ) was measured using an online gas chromatograph (GC) (Mfrs: Buck scientific, Model: 910) equipped with a flame ionization detector (FID). As FID cannot detect CO directly, CO was methanized by passing through a methanizer column containing catalyst at elevated temperature in the presence of hydrogen. This FID was calibrated using standard reference gases of various CO concentrations in the range of 1 to 100,000 ppm. All the gases used were procured from Praxair.
The MEAs assembled into filter cell hardware were initialized and humidified by following a break-in procedure, in which each of the filter cells were operated like a fuel cell for 8 hours at 75 • C and 0.5 V with pure hydrogen and oxygen as fuel and oxidant feeds, respectively. A polarization curve was taken to characterize the performance of each MEA in comparison to that of a standard MEA. To obtain Q s , the electrode was exposed to CO containing gas for certain duration sufficient enough to saturate the filter anode. Then 100% humidified gas (N 2 or He) was fed for a minute to purge out the CO in gas phase. The remaining CO adsorbed on the filter anode was quantified using CO stripping cyclic voltammetry (SCV). The CO covered anode was scanned from 0.05 to 1.1 V at a scan rate of 50 mV/s for two cycles, while 4% H 2 /N 2 flowing through the filter cathode acted as the counter and reference electrodes for the SCV. The resulting current response during the forward sweep of the first cycle was corrected for background current by subtracting out the current response of the forward sweep of the second cycle. The background current corrected response shows a peak that corresponds to the CO oxidation current. The area under the CO oxidation current peak over time indicates the quantity CO adsorbed on the filter anode. [47][48] CO breakthrough.-For determining the CO breakthrough, a conditioned MEA was exposed to the CO containing gas and the change in the CO concentration in the gas at the filter anode's exit with time was quantified using GC-FID. To quantify the concentration of CO at the filter exit, the gas was sampled using a purge and trap system connected with a solid state relay. No sampling bags were used. At the precise time, controlled by a computer, a relay kicks in and collects the E233 gas sample from the exit. This sample was then fed to a GC column, which was on line with FID (i.e. the filter exit → purge-trap → GC column → FID). After the sampling for a GC run was completed, the CO containing gas flow was stopped. The remaining adsorbed CO was oxidized and quantified using CO-SCV. As each GC run takes about 10 minutes, it is impossible to collect all of the samples for different adsorption time in one adsorption experiment. So the adsorption run was repeated for sampling at different adsorption times. Despite repeated runs of adsorption and oxidation the CO adsorption capacity of the MEA was constant and the breakthrough curves were consistent. Unless specified, the general conditions for the experiments were a gas flow rate of 100 cm 3 /min, a temperature of 25 • C and a pressure of 1 atm.
Choosing oxidation potential.-The suitable potential to oxidize the adsorbed CO within the switching time estimated from the CO breakthrough curves, was obtained by carrying out a combination of chronoamperometry and CO-SCV on a CO covered filter anode. The clean filter anode was exposed to CO gas for a certain switching time and then the gas flow was stopped. With 4%H 2 /N 2 flowing through the cathode, a pulse potential, for e.g. 0.7 V, was applied for a time interval, followed by a CO-SCV to quantify the leftover CO and recovered active sites. The lowest potential at which all of the adsorbed CO can be oxidized within the switching time was chosen as the suitable oxidation potential.
Filter setup and operation.-In a twin-cell filter, each of the cells undergo alternating cycles of adsorption and oxidation. To demonstrate the operation of a filter, we used a single filter cell and simulated the cycles of CO adsorption and oxidation. A set of solid state relays were used to switch gas flows and potentiostat, according to preset switching time. The gas stream exiting filter anode was sampled and the CO concentration in it was measured using GC-FID. The filter experiments were carried out for various switching times and exit CO concentrations were measured.

Results and Discussion
CO concentration effects.- Figure 2 shows the experimental breakthrough curves (symbols) and model fit (lines) for different CO concentrations in hydrogen (10,000, 5,000 and 1,000 ppm of CO in H 2 ) at 25 • C, 100 cm 3 /min and 4 mg Pt/cm 2 . The initial CO break  through (i.e., the CO concentration at the exit reached 10 ppm) occurred at around 25, 45 and 190 seconds as the CO feed concentration increased, and leveled to the feed concentration at around 43, 90 and 285 seconds, respectively. The model fit for the respective feed concentrations was predicted using a single value for the mass transfer coefficient (k = 414 s −1 ) fitted for the entire set of data shown in the Figure 2. Hence the breakthrough curve scales with the CO concentration and does not show significant deviation in mass transfer resistance with a change in CO concentration.
CO flow rate effects.- Figure 3 shows the experimental breakthrough curves (symbols) and model fit (lines) for different flow rates (50, 100 and 150 cm 3 /min) at 25 • C, 10,000 ppm CO in H 2 and 4 mg Pt/cm 2 . The initial CO break through occurred at around 50, 25 and 15 s as the flow rate increased, and leveled to the feed concentration at around 80, 43 and 30 s, respectively. The decrease in breakthrough time scaled with a corresponding increase in the molar flow rate, since more CO is being fed in a shorter about of time. However, one mass transfer coefficient value (k = 414 s −1 ) fit all three breakthrough curves at the different flow rates. If the mass transfer rate in the filter was affected by the film diffusion resistance, then increase in flow rate should have improved the mass transfer (i.e., increased k). A constant value of k suggests that the dominant resistance came from either or both of the internal mass transfer resistance (macro-and/or micro-pore diffusion) and surface adsorption reaction resistance.
Effects of catalytic loading.- Figure 4 shows the experimental breakthrough curves (symbols) and model fit (lines) for different Pt loadings (1.5, 4.0 and 8.0 mg/cm 2 ) at 25 • C, 10,000 ppm CO in H 2 and 100 cm 3 /min. The initial break through occurred at around 15, 25 and 45 seconds as the loading increased, and leveled to the feed concentration at around 30, 43 and 90 seconds, respectively. The decrease in breakthrough time scaled with a corresponding decrease in the Pt loading since there are fewer sites to adsorb the CO. Again, one mass transfer coefficient value (k = 414 s −1 ) fits the breakthrough curves at the three different catalyst loadings. If the mass transfer rate in the filter was affected by macro-pore diffusion, than decreasing the Pt loading (i.e., decreasing the film thickness) should have improve mass transfer. A constant value of k suggests that the dominant resistance came from either or both micro-pore diffusion and surface adsorption reaction resistance. Since the fit to the breakthrough curves in Figs. 2-4 result in the same mass-transfer coefficient (i.e., k = 414 s −1 ), the performance of the filter could be described by a characteristic performance curve. Therefore, the breakthrough curves shown in Figs. 2 -4 were replotted on a dimensionless scale (i.e., C in /C out vs τ) in Figure 5. Indeed, all the data followed a single dimensionless breakthrough curve. The only outliers were two data points for the shortest sampling times at the highest flow rates. Considering the concentration is on a log scale, the error at those short times is reasonable. Therefore, the filter performance for CO in H 2 at 25 • C can be accurately predicted over a range of feed CO concentrations, flow rates and catalyst loadings with no adjustable parameters.
Effect of temperature.- Figure 6 shows the experimental breakthrough curves (symbols) and model fit (lines) for different temperatures (25,45 and 60 • C) at 10,000 ppm CO in H 2 , 100 cm 3 /min and 4 mg Pt/cm 2 . The initial break through occurred at around 25, 16 and  13 seconds with increasing temperatures, and the respective k values are 414, 128.6 and 94.6 s −1 . That is, the resistance to CO adsorption is increasing with temperature. Diffusional resistance of a gas should decrease with an increase in temperature. 49 However, CO adsorbs more strongly at lower temperature. 50 The contradictory increase in mass transfer resistance indicates that the CO adsorption resistance is more dominant than the diffusion resistance over this temperature range. Figure 7 shows the experimental breakthrough curves (symbols) and model fit (lines) for different carrier gases (N 2, H 2 , and He) at 25 • C, 1000 ppm CO, 100 cm 3 /min, and 4 mg Pt/cm 2 . The initial break through occurred at around 160, 200, and 230 seconds as the carrier gas changed from N 2, H 2 , and He, respectively. The k values obtained from the fit changed from 3115, 414, and 114.6 s −1 respectively. The diffusional resistance for CO in N 2 , H 2 and He should decrease in that order since the diffusion coefficient increases with decreasing molecular weight of the carrier  gas. 49 This trend is consistent with the k values of CO/N 2 and CO/He breakthrough. However, the CO/H 2 resistance is in between those two, indicating another resistance dominates for this case. This additional effect is attributed to the competing adsorption reaction between CO and H 2 , which slows down the adsorption of CO and hence contributes to an increase in the overall resistance to the CO removal. Neither N 2 nor He competes with CO for adsorption sites.

Effects of inert gas.-
Electro-oxidation of carbon monoxide.-A suitable value for the oxidation potential (0.7 V with a 5 A current maximum) was chosen so that sufficient adsorbed CO oxidized during the pulse without causing undesired side reactions. Figure 8 shows the current response (solid line) and the corresponding applied potential (dashed line). The 5 A current maximum was used to simulate more closely how an actual filter would be run to avoid current spikes. The initial high current observed from 0 to 7 seconds is due to the oxidation of hydrogen trapped in the cell. After the oxidation of the trapped hydrogen, the potential increases to 0.7 V, where CO oxidation starts. The current decays to zero as the adsorbed CO is full oxidized between 7 to 20 seconds. CO filter demonstration.-The exit CO concentration from a continuously operating filter cell, which undergoes alternating cycles of adsorption and oxidation of CO, is shown in Figure 9. The feed to the cell was 10,000 ppm CO in H 2 at 100 cm 3 /min. The cell operated 25 • C and 1 atm, and the catalyst loading was 4.0 mg of Pt/cm 2 . The corresponding breakthrough curve for these conditions is shown in Figure 2. As seen in this figure, to reach an outlet concentration of 10 ppm CO, the adsorption time must be less than 25 seconds. Therefore, a switching time of 20 seconds was chosen for this filter demonstration.
Comparing the coulombs pass during the oxidation step to the amount of CO adsorbed shows that 82% of the charge went to H 2 oxidation, with the remaining going to CO oxidation. This means that 5 H 2 molecules are pumped across the cell for a net gain of one H 2 molecule for every CO molecule oxidized. This comes at the expense of parasitic power loss to the overall system. The power loss can be reduced by using air as the cathode gas for the filter cell, in which the oxygen reduction reaction at the cathode will decrease the overall potential required for the oxidation of CO. However, this will result in a net consumption of hydrogen (approximately 1.5% for a CO concentration of 1,000 ppm). For an exit CO concentration level of 10 ppm, the hydrogen loss in electrochemical filtering compares favorably with other technologies like PSA (8%) 22 and catalytic methanation (11%). 26 The tradeoff in decreasing the losses between power and hydrogen would need to be analyzed for a particular application.
While the fuel and power loss may further be decreased with optimization and design improvements to filter, the other important factor is the volume occupied by the CO handling equipments.
The significant decrease in the volume of filter cell required for a low feed CO concentration is exemplified in the Figure 10. Assuming a switching time of 20 seconds, it relates the ratio of filter cell area to fuel cell area with the CO adsorption capacity per cm 2 of the electrode. Hydrogen concentration in the fuel is assumed to be 40%. For a feed CO concentration of 10,000 ppm fed to a fuel cell operating at 1.0 A/cm 2 requires the filter cell (at 4 mg of Pt/cm 2 ) to be almost 10 times the fuel cell area. However, the required filter area decreases to one fifth of the fuel cell area, if the feed CO concentration is 1,000 ppm and an operating fuel cell current density is 0.2 A/cm 2 , which is typical for stationary fuel cells operating at high voltage efficiency. In case of direct methanol fed fuel cells require a catalyst loading of the order of 4-8 mg/cm 2 , 51 which is comparable to the catalyst loading expected for a filter cell.
The volume required for PrOx or catalytic methanation units increases by a factor of 10, if the desired exit CO concentration decreases from 250 to 10 ppm. 7,26 This is due to the limitation of catalyst selectivity at low CO concentrations. In the electrochemical filter described here, the filter volume scales with the quantity of CO to be removed. Therefore, decreasing the CO concentration from 1000 to 10 ppm requires 10 times less filter volume than from going to 10,000 to 10 ppm. Hence, instead of treating the entire reformate containing 10,000 ppm of CO with either a PrOx reactor or an electrochemical filter, CO removal can be done in stages. The reformate gas with higher CO concentrations (∼ 10,000 ppm) can be treated initially in a PrOx reactor to bring the CO concentration down to 1,000 ppm. The outlet from the PrOx can then be fed to an electrochemical filter to further decrease the CO concentration from 1000 to 10 ppm with a significantly reduced overall volume of CO handling equipments.
Though a CO concentration of 10 ppm in H 2 is still a considerable level of contamination to poison a PEMFC anode, recent advances in anode backing 52 and catalysts 11 lay path to a moderately CO tolerant PEMFC anodes. New catalysts for filter with stability under continuous potential cycling and increased CO adsorption capacity may significantly reduce the power and volume requirements of CO handling equipments for a movable PEMFC system.

Conclusions
The CO adsorption breakthrough on a Pt electrode was studied for different experimental conditions. It is observed that the CO removal is not significantly affected by the feed CO concentration, flow rate and catalyst loading. Observable difference in mass transfer resistance was found, while varying the temperature and the filler gas exhibiting different diffusivity. The breakthrough data were fitted against a fixed bed adsorber model to extract the lumped mass transfer coefficient, which is used to predict the suitable switching time for achieving the desired CO concentration at the filter exit. The model was validated by predicting a suitable switching time for an MEA with known active site concentration and mass transfer coefficient. The power and fuel losses associated with the filter operation were compared with other technologies and possible roles and future improvements were discussed.

A filter
Electrode area of a filter cell -cm 2 A fuel cell Electrode area of a fuel cell -cm 2 C CO gas phase concentration -mol/cm 3 C in Inlet CO gas phase concentration -mol/cm 3 C out Outlet CO gas phase concentration -mol/cm 3 I 0 Modified Bessel function of first kind with zeroth order L Total channel length -cm Q CO adsorbed phase concentration -mol/cm 3 Q s Saturated CO adsorbed phase concentration -mol/cm 3 c Dimensionless CO concentration in gas phase, c = C / C in i fuel cell Current density of a fuel cell k Lumped mass transfer coefficient -1/s t Absolute time -s v Gas velocity -cm/s z Channel length variable -cm θ Dimensionless CO concentration in adsorbed phase,