**Tolerance and Recovery of Ultralow-Loaded Platinum Anode Electrodes upon Carbon Monoxide and Hydrogen Sulfide Exposure**

#### **Sebastian Prass 1,\*, Kaspar Andreas Friedrich 2,3 and Nada Zamel <sup>1</sup>**


Academic Editors: Jean St-Pierre and Shangfeng Du Received: 22 August 2019; Accepted: 26 September 2019; Published: 27 September 2019

**Abstract:** The effects of carbon monoxide (CO) and hydrogen sulfide (H2S) in concentrations close to their respective limits in the Hydrogen Quality Standard ISO 14687-2:2012 on the performance of proton exchange membrane fuel cells (PEMFCs) with ultralow-loaded platinum anode catalyst layers (CLs) were investigated. The anodic loadings were 50, 25, and 15 µg/cm<sup>2</sup> , which represent the current state-of-the-art, target, and stretch target, respectively, for future automotive PEMFCs. Additionally, the effect of shut-down and start-up (SD/SU) processes on recovery from sulfur poisoning was investigated. CO at an ISO concentration of 0.2 ppm caused severe voltage losses of ~40–50% for ultralow-loaded anode CLs. When H2S was in the fuel, these anode CLs exhibited both a nonlinear decrease in tolerance toward sulfur and an improved self-recovery during shut-down and start-up (SD/SU) processes. This observation was hypothesized to have resulted from the decrease in the ratio between CL thickness and geometric cell area, as interfacial effects of water in the pores increasingly impacted the performance of ultrathin CLs. The results indicate that during the next discussions on the Hydrogen Quality Standard, a reduction in the CO limit could be a reasonable alternative considering future PEMFC anodic loadings, while the H2S limit might not require modification.

**Keywords:** fuel impurities; ISO concentration; ultralow-loaded anode catalyst layer; platinum electrode; shut-down and start-up process

#### **1. Introduction**

Proton exchange membrane fuel cells (PEMFCs) are a promising clean energy alternative for applications in the transport sector, as they combine high-power density and efficiency with the significant advantage of fast system refueling times. Hydrogen (H2) as a fuel might, however, contain low concentrations of impurities stemming from production and infrastructure. Impurities such as carbon monoxide (CO) and hydrogen sulfide (H2S) can deteriorate the performance and lifetime of PEMFCs. Naturally, the severity of an impurity is not only affected by its concentration (or rather, dose), but also by the catalyst type, operational parameters, cross-effects, and active or passive mitigation strategies [1,2]. For example, air-bleeding is an effective strategy to provide oxygen (O2) for the oxidation of adsorbed contaminant species in the anode electrode [3], while catalyst alloys containing platinum (Pt) and other platinum group metals (PGMs) can provide higher tolerances versus certain contaminants [4–7]. Although they are very effective, such mitigation strategies partially come with implications about performance or durability. For example, a fraction of the O<sup>2</sup> introduced by air-bleeding readily reacts with H<sup>2</sup> in the anode compartment and thereby lowers the fuel efficiency while simultaneously accelerating membrane degradation through additional peroxide and radical formation [8]. Moreover, alloy catalysts containing PGMs or metals other than Pt usually offer a lower stability, as the alloying components exhibit higher leaching rates. What typically remains is a catalyst particle with a Pt-enriched surface [9], while the leaching cations eventually have impacts on the protonic conductivity or even integrity of the ionomer in the electrode or membrane [10].

Apart from active or passive PEMFC system internal contamination mitigation techniques, adjusting the allowed impurity limits in the Hydrogen Quality Standard ISO 14687-2:2012 poses an additional layer in accommodating enhanced PEMFC requirements versus fuel contaminants. If electrode design or system internal strategies are exhausted, the allowed impurity level for the respective contaminant could be lowered at reasonable levels based upon tangible experimental PEMFC data. Although this option eventually leads to higher H<sup>2</sup> production costs, it helps to avoid higher PEMFC system costs per vehicle or implications coming from internal tolerance improvement strategies.

Some of the major cost drivers in mass-produced PEMFC vehicles are the catalyst layers (CLs) attached to the membrane. The choice of CL materials, the electrode design, and production are primary levers in reducing PEMFC costs while simultaneously increasing the lifetime. Although substantial reductions in PGM catalyst loading per cell area have already been achieved, further reductions are required as a consequence of increasing PGM prizes with higher FC vehicle market penetration. The stipulated reductions range from 50% to 75% compared to the approximate state-of-the-art, resulting in PGM targets for 2020 of about 125 and 62.5 µg/cm<sup>2</sup> depending on the contemplated scenario [11]. In both cases, the loading of the anode electrode is expected to account for 20% (i.e., 25 and 12.5 µg/cm<sup>2</sup> of PGMs, respectively): this is called ultralow loading in the present study hereafter.

Generally, lower anodic catalyst loadings are less tolerant toward catalyst contaminants, as both fuel and contaminants compete for fewer active sites in the electrode. For pure Pt electrodes, the voltage drop was found to increase by 25% when the Pt-loading decreased from 400 to 50 µg/cm<sup>2</sup> if 1 ppm CO was introduced [12,13]. A similar trend was observed for H2S, where the tolerance of the electrode was found to decreased proportionally with the reduction in the anode loading [14]. It is expected that this trend would continue for ultralow loadings (<50 µgPGM/cm<sup>2</sup> ), but so far there has been no study in the literature that has investigated the tolerance of such ultralow anodic loadings. Additionally, processes such as the shut-down and start-up (SD/SU) of FC vehicles are expected to affect the poisoning phenomenon of the electrodes. During downtime, reactants can diffuse from the anode to the cathode, and conversely, mixed potentials arise at the electrodes and poisoned catalysts eventually recover. However, there are limited experimental data available in the literature on recovery due to SD/SU processes, which is especially of interest in the case of recovery from sulfur contamination. Cyclic voltammetry (CV)-like methods triggering oxidative processes at ~0.9–1.1 V count as a recovery strategy for sulfur-contaminated electrodes [6,15,16], but this strategy also induces carbon corrosion and therefore destruction of the electrode itself.

The study presented here therefore seeks to add to the studies by Hashimasa et al. [12,14] by investigating the tolerance of ultralow-loaded anodic platinum catalyst layers. Two different types of contaminants were selected: CO, as its poisoning effect is fully reversible, while in contrast, H2S typically poisons the catalyst irreversibly during regular fuel cell operation. Additionally, recovery from sulfur poisoning through simple shut-down and start-up (SD/SU) processes was examined in more detail for ultralow anodic catalyst loadings.

#### **2. Materials and Methods** *Molecules* **2019**, *24*, 3514 3 of 14

#### *2.1. Test Station and Contaminant Introduction* **2. Materials and Methods**

Single-cell tests were carried out in an in-house-built test station with an integrated potentiostat (Zahner Zennium Pro) and an electric load (Kikusui PLZ664WA) with fluidics (shown schematically in Figure 1). *2.1. Test Station and Contaminant Introduction* Single-cell tests were carried out in an in-house-built test station with an integrated potentiostat (Zahner Zennium Pro) and an electric load (Kikusui PLZ664WA) with fluidics (shown schematically in Figure 1).

**Figure 1.** Single-cell test station scheme. **Figure 1.** Single-cell test station scheme.

In principle, the test station was comparable to the one used by Hashimasa et al. [12], but with a different humidification system for the anode, a different position of the test gas feed inlet (here, the test gases were not fed through the humidifier), and no gas analysis system. In the present study, a differential cell (Baltic qCF type with automotive linear-channel flow field) with an active area of 20.25 cm² was employed, which allowed for the minimization of in-plane effects such as gradients in partial gas pressures, relative humidity, and temperature and therefore enhanced focus on the contamination effect at a given concentration. Although the effects of very low concentrations of impurities eventually become less visible in such a cell [17], a rather uniform coverage of the contaminant on the catalyst throughout the active area was expected. Low concentrations of impurities were achieved by mixing precontaminated test gases with neat H2. Therefore, carbon monoxide (CO, 10 ppm in H<sup>2</sup> 5.0) and hydrogen sulfide (H2S, 0.5 ppm in In principle, the test station was comparable to the one used by Hashimasa et al. [12], but with a different humidification system for the anode, a different position of the test gas feed inlet (here, the test gases were not fed through the humidifier), and no gas analysis system. In the present study, a differential cell (Baltic qCF type with automotive linear-channel flow field) with an active area of 20.25 cm<sup>2</sup> was employed, which allowed for the minimization of in-plane effects such as gradients in partial gas pressures, relative humidity, and temperature and therefore enhanced focus on the contamination effect at a given concentration. Although the effects of very low concentrations of impurities eventually become less visible in such a cell [17], a rather uniform coverage of the contaminant on the catalyst throughout the active area was expected.

N<sup>2</sup> 5.0) were mixed via mass flow controllers with house-supply high-purity hydrogen (all gases provided by Linde AG) in the required fractions. *2.2. Materials* The variations in the anode-loading on the catalyst-coated membranes (CCMs, provided by Low concentrations of impurities were achieved by mixing precontaminated test gases with neat H2. Therefore, carbon monoxide (CO, 10 ppm in H<sup>2</sup> 5.0) and hydrogen sulfide (H2S, 0.5 ppm in N<sup>2</sup> 5.0) were mixed via mass flow controllers with house-supply high-purity hydrogen (all gases provided by Linde AG) in the required fractions.

#### Greenerity GmbH) were achieved through different thicknesses of the anode catalyst layers (CLs), while the cathode loading was kept constant at 400 µg/cm². The catalyst material for both electrodes, *2.2. Materials*

the anode and the cathode, was pure Pt on carbon. The membrane electrode assembly (MEA) specifications are shown in Table 1. **Table 1.** Membrane electrode assembly (MEA) specifications. **Active Cell Area 20.25 cm²** Catalyst Anode Pt/C The variations in the anode-loading on the catalyst-coated membranes (CCMs, provided by Greenerity GmbH) were achieved through different thicknesses of the anode catalyst layers (CLs), while the cathode loading was kept constant at 400 µg/cm<sup>2</sup> . The catalyst material for both electrodes, the anode and the cathode, was pure Pt on carbon. The membrane electrode assembly (MEA) specifications are shown in Table 1.


Cathode Pt/C **Table 1.** Membrane electrode assembly (MEA) specifications.

#### *2.3. Testing Procedure and Conditions*

For every test with a different type of contaminant gas, a fresh MEA sample was assembled into the test cell. To measure the effect of the impurities, the test cell was operated with a constant load to detect the voltage drop associated with the contaminant species and concentration. In the following figures, the cell voltage drop is defined as the relative change based on the initial cell voltage. The effect of CO was tested at three different concentrations, namely 0.1, 0.2, and 0.4 ppm (50%, 100%, and 200% of the impurity limit noted in the H<sup>2</sup> Quality Standard). Before and after the actual contamination, the fuel cell was operated with neat H<sup>2</sup> to establish a baseline voltage and to detect eventual irreversible degradation of the electrodes. The effect of H2S was tested at two concentrations, which were 4 and 20 ppb (100% and 500% of the limit in the Quality Standard), with neat H<sup>2</sup> operation only at the start of the contaminant test. The conditions during the contaminant tests are shown in Table 2.


**Table 2.** Operating conditions during contamination.

The MEAs were characterized, including cyclic voltammetry (CV) on the anode and cathode side at the beginning and end of life (BOL and EOL), as were the polarization curves at the BOL, to compare the performance between the MEA types before starting the contaminant test. The gas pressure during contamination was selected in reference to the studies by Hashimasa et al. [12,14], while the pressure during the polarization curves was chosen according to in-house standardized testing protocols. The conditions during the polarization curves are shown in Table 3.

**Table 3.** Polarization curve conditions.


CV measurements were performed to determine the electrochemically active surface area (ECSA) of the CLs before and after contamination and recovery procedures, specifically from H2S poisoning. The CVs were performed on both the anode and cathode electrodes under the conditions summarized in Table 4. To conduct an anode CV, the test cell was purged with nitrogen in order to exchange the gas supply and the electric connectors of the anode and cathode compartment and then reconditioned with fully humidified H<sup>2</sup> and N<sup>2</sup> for 12 min prior to the CV. Following the anode CV, the cell was purged again and reconnected in a regular anode/cathode configuration for subsequent tests.



comparison.

area.

were chosen as 0.15 to 0.3 V, as shown in Figure 3.

anodes (as tested in the present study).

anodes (as tested in the present study).

An upper CV boundary of 700 mV was selected to avoid the oxidation of adsorbed foreign species, especially during the H2S recovery tests, and to solely focus on the recovery from SD/SU processes. Moreover, the N<sup>2</sup> flow was stopped during the actual CV to avoid disproportionally high H<sup>2</sup> evolution currents during the anodic sweep, which were observed especially for the lowest anodic loading. Figure 2 shows exemplary BOL CVs of the three different anode electrodes and one cathode electrode for comparison. An upper CV boundary of 700 mV was selected to avoid the oxidation of adsorbed foreign species, especially during the H2S recovery tests, and to solely focus on the recovery from SD/SU processes. Moreover, the N<sup>2</sup> flow was stopped during the actual CV to avoid disproportionally high H<sup>2</sup> evolution currents during the anodic sweep, which were observed especially for the lowest anodic loading. Figure 2 shows exemplary BOL CVs of the three different anode electrodes and one cathode electrode for comparison. An upper CV boundary of 700 mV was selected to avoid the oxidation of adsorbed foreign species, especially during the H2S recovery tests, and to solely focus on the recovery from SD/SU processes. Moreover, the N<sup>2</sup> flow was stopped during the actual CV to avoid disproportionally high H<sup>2</sup> evolution currents during the anodic sweep, which were observed especially for the lowest anodic loading. Figure 2 shows exemplary BOL CVs of the three different anode electrodes and one cathode electrode for comparison.

*Molecules* **2019**, *24*, 3514 5 of 14

*Molecules* **2019**, *24*, 3514 5 of 14

**Figure 2.** Anode CVs of MEA types A, B, and C with an MEA type A cathode CV for reference. The inset expands the H<sup>2</sup> adsorption/desorption regions of the anode catalyst layers (CLs) for visual **Figure 2.** Anode CVs of MEA types A, B, and C with an MEA type A cathode CV for reference. The inset expands the H<sup>2</sup> adsorption/desorption regions of the anode catalyst layers (CLs) for visual comparison. **Figure 2.** Anode CVs of MEA types A, B, and C with an MEA type A cathode CV for reference. The inset expands the H<sup>2</sup> adsorption/desorption regions of the anode catalyst layers (CLs) for visual comparison.

Normally, the ECSA is determined through integration of the charge transfer between voltage boundaries, starting from ~0.08 to 0.1 V to the minima or maxima of the respective double-layer charging current, which typically is somewhere between 0.3 and 0.6 V [18]. However, in this study, these boundaries were considered less suitable for CVs on ultralow-loaded anode CLs. High currents associated with H<sup>2</sup> evolution during the cathodic sweep (*H2,ev*) and the coherent reverse-transport of eventually evolved H<sup>2</sup> during the anodic sweep (*H2,rtr*) would account for relatively large errors in the ECSA. Hence, the voltage boundaries for the determination of the ECSA Normally, the ECSA is determined through integration of the charge transfer between voltage boundaries, starting from ~0.08 to 0.1 V to the minima or maxima of the respective double-layer charging current, which typically is somewhere between 0.3 and 0.6 V [18]. However, in this study, these boundaries were considered less suitable for CVs on ultralow-loaded anode CLs. High currents associated with H<sup>2</sup> evolution during the cathodic sweep (*H2,ev*) and the coherent reverse-transport of eventually evolved H<sup>2</sup> during the anodic sweep (*H2,rtr*) would account for relatively large errors in the ECSA. Hence, the voltage boundaries for the determination of the ECSA were chosen as 0.15 to 0.3 V, as shown in Figure 3. Normally, the ECSA is determined through integration of the charge transfer between voltage boundaries, starting from ~0.08 to 0.1 V to the minima or maxima of the respective double-layer charging current, which typically is somewhere between 0.3 and 0.6 V [18]. However, in this study, these boundaries were considered less suitable for CVs on ultralow-loaded anode CLs. High currents associated with H<sup>2</sup> evolution during the cathodic sweep (*H2,ev*) and the coherent reverse-transport of eventually evolved H<sup>2</sup> during the anodic sweep (*H2,rtr*) would account for relatively large errors in the ECSA. Hence, the voltage boundaries for the determination of the ECSA were chosen as 0.15 to 0.3 V, as shown in Figure 3.

**Figure 3.** Electrochemically active surface area (ECSA) determination from reduced H<sup>2</sup> adsorption **Figure 3.** Electrochemically active surface area (ECSA) determination from reduced H<sup>2</sup> adsorption **Figure 3.** Electrochemically active surface area (ECSA) determination from reduced H<sup>2</sup> adsorption area.

area. Using this narrowed voltage range, the anode ECSA was determined from the anodic sweeps associated with the adsorption of H<sup>2</sup> on the catalyst surfaces. Although this procedure cuts the Using this narrowed voltage range, the anode ECSA was determined from the anodic sweeps associated with the adsorption of H<sup>2</sup> on the catalyst surfaces. Although this procedure cuts the Using this narrowed voltage range, the anode ECSA was determined from the anodic sweeps associated with the adsorption of H<sup>2</sup> on the catalyst surfaces. Although this procedure cuts the measured ECSA compared to integration between regular voltage ranges, it was found that it would

measured ECSA compared to integration between regular voltage ranges, it was found that it would

measured ECSA compared to integration between regular voltage ranges, it was found that it would increase the accuracy of the ECSA determination and its changes in the case of ultralow-loaded

increase the accuracy of the ECSA determination and its changes in the case of ultralow-loaded anodes (as tested in the present study). *Molecules* **2019**, *24*, 3514 6 of 14 *Molecules* **2019**, *24*, 3514 6 of 14

#### **3. Results and Discussion 3. Results and Discussion 3. Results and Discussion**

#### *3.1. Performance and Stability of Ultralow-Loaded Anodic CLs 3.1. Performance and Stability of Ultralow-Loaded Anodic CLs*

of the three different MEAs when neat H<sup>2</sup> was supplied to the test cells.

Before the actual contamination tests, the BOL performance and voltage stability of the MEA samples with ultralow-loaded anodes were established. Figure 4 shows the BOL polarization curves of the three different MEAs when neat H<sup>2</sup> was supplied to the test cells. Before the actual contamination tests, the BOL performance and voltage stability of the MEA samples with ultralow-loaded anodes were established. Figure 4 shows the BOL polarization curves of the three different MEAs when neat H<sup>2</sup> was supplied to the test cells. *3.1. Performance and Stability of Ultralow-Loaded Anodic CLs* Before the actual contamination tests, the BOL performance and voltage stability of the MEA samples with ultralow-loaded anodes were established. Figure 4 shows the BOL polarization curves

**Figure 4.** Polarization curves of MEA types A, B, and C using neat H2, with high-frequency resistance (HFR) as dashed lines. **Figure 4.** Polarization curves of MEA types A, B, and C using neat H<sup>2</sup> , with high-frequency resistance (HFR) as dashed lines. **Figure 4.** Polarization curves of MEA types A, B, and C using neat H2, with high-frequency resistance (HFR) as dashed lines.

As can be seen in the figure, the polarization curves of the different MEAs overlap quite well, indicating that overpotentials arising due to a lack of active catalyst sites for the hydrogen oxidation reaction (HOR) were not significant for ultralow anodic loadings. In fact, MEA type C (15 µg/cm²) even showed a slightly better performance at current densities above 2.5 A/cm², (~15 mV at 3 A/cm²), which might have been a result of minimal differences in humidification characteristics of this specific sample and the lower measured high-frequency resistance (HFR). As can be seen in the figure, the polarization curves of the different MEAs overlap quite well, indicating that overpotentials arising due to a lack of active catalyst sites for the hydrogen oxidation reaction (HOR) were not significant for ultralow anodic loadings. In fact, MEA type C (15 µg/cm<sup>2</sup> ) even showed a slightly better performance at current densities above 2.5 A/cm<sup>2</sup> , (~15 mV at 3 A/cm<sup>2</sup> ), which might have been a result of minimal differences in humidification characteristics of this specific sample and the lower measured high-frequency resistance (HFR). As can be seen in the figure, the polarization curves of the different MEAs overlap quite well, indicating that overpotentials arising due to a lack of active catalyst sites for the hydrogen oxidation reaction (HOR) were not significant for ultralow anodic loadings. In fact, MEA type C (15 µg/cm²) even showed a slightly better performance at current densities above 2.5 A/cm², (~15 mV at 3 A/cm²), which might have been a result of minimal differences in humidification characteristics of this specific sample and the lower measured high-frequency resistance (HFR).

In addition to the BOL performance, the cell voltage stability of the three MEA types over a testing time of 100 h of continuous operation at a constant load with neat H<sup>2</sup> was established, which is shown in Figure 5. In addition to the BOL performance, the cell voltage stability of the three MEA types over a testing time of 100 h of continuous operation at a constant load with neat H<sup>2</sup> was established, which is shown in Figure 5. In addition to the BOL performance, the cell voltage stability of the three MEA types over a testing time of 100 h of continuous operation at a constant load with neat H<sup>2</sup> was established, which is shown in Figure 5.

voltage stabilities of the MEA types. The voltages were normalized to the initial cell voltage at time = 0 h. **Figure 5.** Voltage decay over 100 h of continuous operation in neat H<sup>2</sup> at 1.0 A/cm², showing similar voltage stabilities of the MEA types. The voltages were normalized to the initial cell voltage at time = 0 h. **Figure 5.** Voltage decay over 100 h of continuous operation in neat H<sup>2</sup> at 1.0 A/cm<sup>2</sup> , showing similar voltage stabilities of the MEA types. The voltages were normalized to the initial cell voltage at time = 0 h.

**Figure 5.** Voltage decay over 100 h of continuous operation in neat H<sup>2</sup> at 1.0 A/cm², showing similar

During these stability tests, no significant difference between the voltage drops of the MEA types was observed. A slight voltage drop during the first ~2 h was visible for all three MEA types and was associated with the consumption of reactants, which saturated in the electrode before the current was increased. Overall, the comparability of the different MEA types at the BOL under operation with neat H<sup>2</sup> During these stability tests, no significant difference between the voltage drops of the MEA types was observed. A slight voltage drop during the first ~2 h was visible for all three MEA types and was associated with the consumption of reactants, which saturated in the electrode before the current was increased. During these stability tests, no significant difference between the voltage drops of the MEA types was observed. A slight voltage drop during the first ~2 h was visible for all three MEA types and was associated with the consumption of reactants, which saturated in the electrode before the current was increased.

was considered satisfactory and was accepted for subsequent tests with contaminants. Before each contamination test, the cell was operated for 20 h with neat H<sup>2</sup> to establish a baseline voltage. In the case of CO, the first concentration of contaminant was introduced and increased at time steps of 20 h, before we finally shut off the impurity for an additional 20 h of operation with neat H2. In the case of Overall, the comparability of the different MEA types at the BOL under operation with neat H<sup>2</sup> was considered satisfactory and was accepted for subsequent tests with contaminants. Before each contamination test, the cell was operated for 20 h with neat H<sup>2</sup> to establish a baseline voltage. In the case of CO, the first concentration of contaminant was introduced and increased at time steps of 20 h, Overall, the comparability of the different MEA types at the BOL under operation with neat H<sup>2</sup> was considered satisfactory and was accepted for subsequent tests with contaminants. Before each contamination test, the cell was operated for 20 h with neat H<sup>2</sup> to establish a baseline voltage. In the case of CO, the first concentration of contaminant was introduced and increased at time steps of 20 h,

dissolved O2, which was driven out as soon as it was heated in the bubbler and was consequently

bubbler required a refill with fresh deionized (DI) water every 10 h. This DI water contained dissolved O2, which was driven out as soon as it was heated in the bubbler and was consequently before we finally shut off the impurity for an additional 20 h of operation with neat H2. In the case of H2S, after operation with neat H2, a single concentration of H2S was introduced until the cell voltage broke down, and subsequently SD/SU recovery tests were conducted. For all tests, the anode bubbler required a refill with fresh deionized (DI) water every 10 h. This DI water contained dissolved O2, which was driven out as soon as it was heated in the bubbler and was consequently available for the recovery of poisoned Pt sites, which is visible as voltage peaks in the following figures. *Molecules* **2019**, *24*, 3514 7 of 14 available for the recovery of poisoned Pt sites, which is visible as voltage peaks in the following figures. *3.2. Effect of CO on Ultralow-Loaded Anode CLs*

#### *3.2. E*ff*ect of CO on Ultralow-Loaded Anode CLs* Essentially, CO adsorbs on Pt and thereby competes with the actual HOR for active sites on the

Essentially, CO adsorbs on Pt and thereby competes with the actual HOR for active sites on the catalyst surfaces, as shown in Equations (1)–(3): catalyst surfaces, as shown in Equations (1)–(3): 2 + <sup>2</sup> ↔ 2( − ), (1)

$$2Pt + H\_2 \leftrightarrow 2(Pt - H)\_2 \tag{1}$$

$$Pt + CO \leftrightarrow Pt-CO\_{2} \tag{2}$$

$$2Pt + \text{CO} \leftrightarrow (Pt)\_2 = \text{CO}.\tag{3}$$

Depending on the coverage of CO, each molecule blocks one or two active Pt sites via linear or bridge bonds (Equations (2) and (3), respectively) [5,19]. At lower coverages, a higher fraction of bridge bonds is expected, while at higher coverages, an adlayer with CO linear bonds dominates [20]. However, the adlayer CO structure depends on particle sizes, adsorption potentials, facet orientations, and temperature in a complex way because dipole–dipole interactions are important [21]. The effect of different CO concentrations on the voltage decay rates of the three ultralow-loaded anodic CLs is shown in Figure 6. Depending on the coverage of CO, each molecule blocks one or two active Pt sites via linear or bridge bonds (Equations (2) and (3), respectively) [5,19]. At lower coverages, a higher fraction of bridge bonds is expected, while at higher coverages, an adlayer with CO linear bonds dominates [20]. However, the adlayer CO structure depends on particle sizes, adsorption potentials, facet orientations, and temperature in a complex way because dipole–dipole interactions are important [21]. The effect of different CO concentrations on the voltage decay rates of the three ultralow-loaded anodic CLs is shown in Figure 6.

**Figure 6.** Voltage drops induced by different CO concentrations in MEA types A, B, and C at a constant load of 1.0 A/cm². The voltage peaks (every 10 h) were caused by anode bubbler refills and coherent recovery of Pt sites with O<sup>2</sup> dissolved in DI water. Again, the voltages were normalized to the initial cell voltage at time = 0 h, while the results are shown starting from *t* = 5 h. **Figure 6.** Voltage drops induced by different CO concentrations in MEA types A, B, and C at a constant load of 1.0 A/cm<sup>2</sup> . The voltage peaks (every 10 h) were caused by anode bubbler refills and coherent recovery of Pt sites with O<sup>2</sup> dissolved in DI water. Again, the voltages were normalized to the initial cell voltage at time = 0 h, while the results are shown starting from *t* = 5 h.

As expected, the effect of CO in the fuel generally increased for lower anodic loadings, including both a faster and more severe voltage drop. The leveling of the potentials, i.e., the initial decline toward a plateau, depended on the contaminant concentration and the CL thickness [22,23]. For thinner CLs, the reaction front increasingly corresponded with the actual CL thickness, and therefore the local potential was more uniform while contaminants competed throughout the layer with hydrogen for adsorption sites, which resulted in a lower tolerance for thinner (and lower-loaded) CLs. At the ISO concentration (0.2 ppm), the voltage loss due to CO poisoning accounted for ~8%, 41%, and 51% when the anodic loading decreased from 50 to 25 and 15 µg/cm², respectively. Slight potential oscillations of the ultralow-loaded anode MEA types (type B and As expected, the effect of CO in the fuel generally increased for lower anodic loadings, including both a faster and more severe voltage drop. The leveling of the potentials, i.e., the initial decline toward a plateau, depended on the contaminant concentration and the CL thickness [22,23]. For thinner CLs, the reaction front increasingly corresponded with the actual CL thickness, and therefore the local potential was more uniform while contaminants competed throughout the layer with hydrogen for adsorption sites, which resulted in a lower tolerance for thinner (and lower-loaded) CLs. At the ISO concentration (0.2 ppm), the voltage loss due to CO poisoning accounted for ~8%, 41%, and 51% when the anodic loading decreased from 50 to 25 and 15 µg/cm<sup>2</sup> , respectively. Slight potential oscillations of the ultralow-loaded anode MEA types (type B and especially C) at high CO concentrations between

especially C) at high CO concentrations between normalized voltage ratios of 0.4 and 0.6 were also

normalized voltage ratios of 0.4 and 0.6 were also visible. At these potentials, overpotentials induced by CO poisoning forced the anode potential to shift frequently toward the cathode potential and close to the oxidation potential of CO to CO2, allowing for recovery of the electrode [24,25]. This self-recovery was the reason for the maximum coverage of the catalyst with CO in regular PEMFC operation and a flattening of the relative potential drop for lower anodic catalyst loadings with higher CO concentrations, which is partially visible in Figure 7. *Molecules* **2019**, *24*, 3514 8 of 14 coverage of the catalyst with CO in regular PEMFC operation and a flattening of the relative potential drop for lower anodic catalyst loadings with higher CO concentrations, which is partially visible in Figure 7.

**Figure 7.** Normalized voltages over anodic loading; data adapted from Hashimasa et al. [12]. **Figure 7.** Normalized voltages over anodic loading; data adapted from Hashimasa et al. [12].

In the figure, relative voltage drops due to CO poisoning over the anode Pt loading from the study by Hashimasa et al. and the present study are compared. Although the test cells and the operational parameters between the two studies were different (70% fuel usage in the single cell by the Japanese Automobile Research Institute, JARI, versus 8.3% fuel usage in the differential single cell employed in the present study), a general trend for voltage decay with lower anodic loadings or higher CO concentrations can be seen. The onset of the mentioned flattening of the relative voltage drop at maximum CO coverage is visible for the lowest anodic loading and the highest tested CO concentration, where the relative change between MEA types B and C was less significant compared to types A and B. In the figure, relative voltage drops due to CO poisoning over the anode Pt loading from the study by Hashimasa et al. and the present study are compared. Although the test cells and the operational parameters between the two studies were different (70% fuel usage in the single cell by the Japanese Automobile Research Institute, JARI, versus 8.3% fuel usage in the differential single cell employed in the present study), a general trend for voltage decay with lower anodic loadings or higher CO concentrations can be seen. The onset of the mentioned flattening of the relative voltage drop at maximum CO coverage is visible for the lowest anodic loading and the highest tested CO concentration, where the relative change between MEA types B and C was less significant compared to types A and B.

In general, CO contamination is fairly easy to mitigate by providing O<sup>2</sup> to the anode via the air-bleeding technique [3]. This technique not only mitigates CO poisoning, but also partially mitigates poisoning from other contaminants, such as H2S [16]. However, as discussed above, air bleeding also comes with disadvantages, such as a reduction in fuel efficiency and potential effects on the integrity of the ionomer in the CLs and membrane. Therefore, to minimize potentially amplified side effects from such mitigation strategies, a reduction of the limit for CO in the H<sup>2</sup> Quality Standard could be a reasonable option considering the severity of CO poisoning on ultralow anodic loadings, as they likely will be employed in the near future in automotive PEMFCs. In general, CO contamination is fairly easy to mitigate by providing O<sup>2</sup> to the anode via the air-bleeding technique [3]. This technique not only mitigates CO poisoning, but also partially mitigates poisoning from other contaminants, such as H2S [16]. However, as discussed above, air bleeding also comes with disadvantages, such as a reduction in fuel efficiency and potential effects on the integrity of the ionomer in the CLs and membrane. Therefore, to minimize potentially amplified side effects from such mitigation strategies, a reduction of the limit for CO in the H<sup>2</sup> Quality Standard could be a reasonable option considering the severity of CO poisoning on ultralow anodic loadings, as they likely will be employed in the near future in automotive PEMFCs.

#### *3.3. Effect of H2S on Ultralow-Loaded Anode CLs 3.3. E*ff*ect of H2S on Ultralow-Loaded Anode CLs*

ultralow-loaded anodes.

In contrast to CO, H2S poisons catalyst surfaces irreversibly through dissociative adsorption on Pt via chemical or electrochemical reaction pathways, as indicated by Equations (4) and (5), respectively. The elemental sulfur on Pt cumulatively occupies active catalyst sites also via linear or bridge bonds, which eventually leads to a complete breakdown of the PEMFC performance [6,14,16]: In contrast to CO, H2S poisons catalyst surfaces irreversibly through dissociative adsorption on Pt via chemical or electrochemical reaction pathways, as indicated by Equations (4) and (5), respectively. The elemental sulfur on Pt cumulatively occupies active catalyst sites also via linear or bridge bonds, which eventually leads to a complete breakdown of the PEMFC performance [6,14,16]:

$$Pt + H\_2S \leftrightarrow Pt-S + H\_2 \tag{4}$$

$$Pt + H\_2S \leftrightarrow Pt - S + 2H^+ + 2e^-.\tag{5}$$

$$\dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots \quad \dots$$

Higher catalyst loadings provide a higher nominal ECSA and therefore a larger buffer versus Higher catalyst loadings provide a higher nominal ECSA and therefore a larger buffer versus such a breakdown. This decrease in tolerance with a reduction in platinum loading is partially

such a breakdown. This decrease in tolerance with a reduction in platinum loading is partially

visible in Figure 8, which shows the operation times until the breakdown was observed for ultralow-loaded anodes. *Molecules* **2019**, *24*, 3514 9 of 14

*Molecules* **2019**, *24*, 3514 9 of 14

**Figure 8.** Voltage breakdowns induced by 4 and 20 ppb of H2S during operation at a constant load of 1.0 A/cm². The operation of MEA type A was purposely stopped after ~340 h and ~70 h, while MEA types B and C stopped automatically after voltage breakdowns were observed. **Figure 8.** Voltage breakdowns induced by 4 and 20 ppb of H2S during operation at a constant load of 1.0 A/cm<sup>2</sup> . The operation of MEA type A was purposely stopped after ~340 h and ~70 h, while MEA types B and C stopped automatically after voltage breakdowns were observed. **Figure 8.** Voltage breakdowns induced by 4 and 20 ppb of H2S during operation at a constant load of 1.0 A/cm². The operation of MEA type A was purposely stopped after ~340 h and ~70 h, while MEA types B and C stopped automatically after voltage breakdowns were observed.

The voltage breakdowns for the highest tested anodic loading (50 µg/cm²) were not fully observed. In the case of 4 ppb of H2S, the test was purposely interrupted after 340 h of contaminant introduction, as a voltage breakdown was not expected anymore. However, subsequent CVs revealed an almost completely sulfur-blocked ECSA, which is shown in the following sections. In the case of 20 ppb of H2S, the test station automatically stopped at the onset of the breakdown after about ~70 h, but the start of the breakdown was still visible. The voltage breakdowns for the highest tested anodic loading (50 µg/cm<sup>2</sup> ) were not fully observed. In the case of 4 ppb of H2S, the test was purposely interrupted after 340 h of contaminant introduction, as a voltage breakdown was not expected anymore. However, subsequent CVs revealed an almost completely sulfur-blocked ECSA, which is shown in the following sections. In the case of 20 ppb of H2S, the test station automatically stopped at the onset of the breakdown after about ~70 h, but the start of the breakdown was still visible. The voltage breakdowns for the highest tested anodic loading (50 µg/cm²) were not fully observed. In the case of 4 ppb of H2S, the test was purposely interrupted after 340 h of contaminant introduction, as a voltage breakdown was not expected anymore. However, subsequent CVs revealed an almost completely sulfur-blocked ECSA, which is shown in the following sections. In the case of 20 ppb of H2S, the test station automatically stopped at the onset of the breakdown after about ~70 h, but the start of the breakdown was still visible.

Interestingly, for both MEA types with ultralow anodic loadings (MEA types B and C), voltage breakdowns were detected after almost similar poisoning times for both tested H2S concentrations of 4 and 20 ppb. In Figure 9, which compares the accumulated H2S supplied until a 30-mV voltage loss was detected in the present study versus the study by Hashimasa et al., these similar poisoning times are visible as a nonproportional decline in the amount of H2S supplied with the reduction in anodic loading. Interestingly, for both MEA types with ultralow anodic loadings (MEA types B and C), voltage breakdowns were detected after almost similar poisoning times for both tested H2S concentrations of 4 and 20 ppb. In Figure 9, which compares the accumulated H2S supplied until a 30-mV voltage loss was detected in the present study versus the study by Hashimasa et al., these similar poisoning times are visible as a nonproportional decline in the amount of H2S supplied with the reduction in anodic loading. Interestingly, for both MEA types with ultralow anodic loadings (MEA types B and C), voltage breakdowns were detected after almost similar poisoning times for both tested H2S concentrations of 4 and 20 ppb. In Figure 9, which compares the accumulated H2S supplied until a 30-mV voltage loss was detected in the present study versus the study by Hashimasa et al., these similar poisoning times are visible as a nonproportional decline in the amount of H2S supplied with the reduction in anodic loading.

**Figure 9.** H2S supplied to the cell until voltage dropped by 30 mV over anodic loading; data adapted from Hashimasa et al. [14]. **Figure 9.** H2S supplied to the cell until voltage dropped by 30 mV over anodic loading; data adapted from Hashimasa et al. [14]. **Figure 9.** H2S supplied to the cell until voltage dropped by 30 mV over anodic loading; data adapted from Hashimasa et al. [14].

Although Hashimasa et al. described their observed decline as proportional to the reduction in the loading, their data actually rather showed a slight flattening of the curve with the decrease in the anodic loading, comparable to the data from the presented study. Again, although the test cells and the operational parameters were different (70% fuel usage in JARI's single cell versus 8.3% fuel usage in the differential single cell in the present study), the general trend was still visible. One explanation could be that some of the H2S adsorbed on the surfaces of the test bench and Although Hashimasa et al. described their observed decline as proportional to the reduction in the loading, their data actually rather showed a slight flattening of the curve with the decrease in the anodic loading, comparable to the data from the presented study. Again, although the test cells and the operational parameters were different (70% fuel usage in JARI's single cell versus 8.3% fuel usage in the differential single cell in the present study), the general trend was still visible. Although Hashimasa et al. described their observed decline as proportional to the reduction in the loading, their data actually rather showed a slight flattening of the curve with the decrease in the anodic loading, comparable to the data from the presented study. Again, although the test cells and the operational parameters were different (70% fuel usage in JARI's single cell versus 8.3% fuel usage in the differential single cell in the present study), the general trend was still visible.

cell components before actually reaching the CCM and catalyst sites. Depending on the chronology of the tests, this latency could create delays in the voltage breakdown. On the other hand, in the One explanation could be that some of the H2S adsorbed on the surfaces of the test bench and cell components before actually reaching the CCM and catalyst sites. Depending on the chronology One explanation could be that some of the H2S adsorbed on the surfaces of the test bench and cell components before actually reaching the CCM and catalyst sites. Depending on the chronology of the

For these self-recovery tests, the ECSA of the anode CLs exposed to H2S were determined at the BOL after a simulated SD/SU process, after H2S poisoning, and again after an SD/SU process. The SD/SU included a short purge with dry nitrogen to avoid open circuit voltage (OCV) in

For these self-recovery tests, the ECSA of the anode CLs exposed to H2S were determined at the BOL after a simulated SD/SU process, after H2S poisoning, and again after an SD/SU process. The SD/SU included a short purge with dry nitrogen to avoid open circuit voltage (OCV) in

+ SD/SU).

steps.

tests, this latency could create delays in the voltage breakdown. On the other hand, in the present study, the CVs of lower-loaded anodes also revealed a higher degree of self-recovery from simple shut-down (SD) and start-up (SU) processes.

For these self-recovery tests, the ECSA of the anode CLs exposed to H2S were determined at the BOL after a simulated SD/SU process, after H2S poisoning, and again after an SD/SU process. The SD/SU included a short purge with dry nitrogen to avoid open circuit voltage (OCV) in H2/air-atmosphere, a cooldown of the cell to 20 ◦C, a wait time of 3 h, and finally again heating of the cell to 80 ◦C and the introduction of neat H2/air to the cell, which was kept at a fixed potential of 0.8 V during the heating. Figure 10 presents these anode CVs for the three different anodic loadings. *Molecules* **2019**, *24*, 3514 10 of 14 H2/air-atmosphere, a cooldown of the cell to 20 °C, a wait time of 3 h, and finally again heating of the cell to 80 °C and the introduction of neat H2/air to the cell, which was kept at a fixed potential of 0.8 V during the heating. Figure 10 presents these anode CVs for the three different anodic loadings.

**Figure 10.** Anode CVs for MEA types A, B, and C (50, 25, and 15 µg/cm²) at the beginning of life (BOL), after H2S contamination (+H2S), and after a subsequent shut-down/start-up (SD/SU) process (+H2S + SD/SU). For MEA type A, the CV after SD/SU just before the contamination is also shown (A **Figure 10.** Anode CVs for MEA types A, B, and C (50, 25, and 15 µg/cm<sup>2</sup> ) at the beginning of life (BOL), after H2S contamination (+H2S), and after a subsequent shut-down/start-up (SD/SU) process (+H2S + SD/SU). For MEA type A, the CV after SD/SU just before the contamination is also shown (A + SD/SU).

Clearly visible is the difference between the CVs at the BOL and after H2S contamination (black to yellow CV) for all three MEA types, indicating the reduction of ECSA due to sulfur adsorbed on Pt. For MEA type A, the CV after SD/SU and before H2S contamination (blue CV) is additionally shown to exemplarily demonstrate that the SD/SU process did not significantly affect the CV measurement and ECSA determination, as both CVs overlapped quite well. However, when the SD/SU process was carried out after H2S contamination, the CV and therefore the ECSA gained in area compared to the poisoned ECSA (yellow to green CV), indicating a partial recovery from previously deactivated ECSA. This self-recovery was increasingly observed with the reduction in the Clearly visible is the difference between the CVs at the BOL and after H2S contamination (black to yellow CV) for all three MEA types, indicating the reduction of ECSA due to sulfur adsorbed on Pt. For MEA type A, the CV after SD/SU and before H2S contamination (blue CV) is additionally shown to exemplarily demonstrate that the SD/SU process did not significantly affect the CV measurement and ECSA determination, as both CVs overlapped quite well. However, when the SD/SU process was carried out after H2S contamination, the CV and therefore the ECSA gained in area compared to the poisoned ECSA (yellow to green CV), indicating a partial recovery from previously deactivated ECSA. This self-recovery was increasingly observed with the reduction in the anodic loading. Table 5 presents the nominal ECSAs and percentage changes between the test SD/SU steps.


**Table 5.** ECSA at the BOL and relative change after shut-down/start-up processes (SD/SU) before and after contamination with H2S based on narrowed boundaries (integration between 150 and 300 mV). Note: the nominal ECSA was lower by about 60–70% than what would be typically expected for the specific catalyst material, while the relative ECSA changes were amplified to some degree due to the

While only about 35% of the ECSA from MEA type A (50 µg/cm<sup>2</sup> ) could be recovered, 84% and almost a full recovery of 94% could be achieved for MEA types B (25 µg/cm<sup>2</sup> ) and C (15 µg/cm<sup>2</sup> ), respectively, through a simple SD/SU process.

C 19.6 19.1 (97%) 14.0 (71%) 18.4 (94%)

The reason for the different behavior of ultralow-loaded anodes with respect to their tolerance versus H2S contamination and the improved self-recovery during SD/SU processes might have a dimensional character in combination with the scavenging effect of water versus contaminants [26]. Studies in the literature investigating the recovery of sulfur-poisoned electrodes have often employed CV-like processes to increase the potential and thereby oxidize adsorbed sulfur either on cathode or anode electrodes [27,28]. During this oxidation, sulfur oxides such as sulfur dioxide (and in combination with water-soluble anions such as sulfate (SO2- <sup>4</sup>) or sulfite (SO2- <sup>3</sup>)) develop as shown in Equations (6)–(8) [16]:

$$Pt-S + O\_2 \leftrightarrow Pt + SO\_2.\tag{6}$$

$$Pt-S + 3H\_2O \leftrightarrow Pt + SO\_3^{2-} + 6H^+ + 4e^-, \tag{7}$$

$$Pt-S + 4H\_2O \leftrightarrow Pt + SO\_4^{2-} + 8H^+ + 6e^-. \tag{8}$$

Presumably, during an SD/SU process, the catalyst surfaces and adsorbed species relax, the local potential varies depending on the local equilibrium and the available species on Pt, and chemical reactions occur to the point of the formation of sulfur anions in the presence of water. It should be noted that the potential of the anodic electrode prior to and during the SD can affect the reduction state of the sulfur species, which eventually facilitates their oxidation or desorption [29]. As the different anodic loadings tested in this study were achieved through variations in CL thickness, the anode of MEA type C consequently had the lowest thickness, while the active cell area remained the same for all samples. During an SD, water condensates and eventually is driven out through hydrophobic pores of the microporous and gas diffusion layer (MPL/GDL) or collects in pores and areas, which are energetically favorable. As the interface between the MPL and CLs also contains such pores [30], sulfur in proximity to this interface might dissolve in these water accumulations in the form of soluble sulfur anions [26]. As the active cell area and therefore the CL/MPL interface area should be the same on average for all three MEA types, while the anode CL volumes are different, a higher fraction of anions could get removed for lower-loaded and therefore thinner anode electrodes. These anions dissolved in water eventually are flushed out once the PEMFC is started again. This works better so long as sulfur is weakly bonded to the Pt surface via linear bonds. With time, adsorbed sulfur develops stronger bonds to active sites and is bound more strongly to the catalyst, leading finally to the observed voltage breakdowns of the PEMFCs. Thinner CLs may also be associated with a changed ionomer structure, and the potentials within the layer are generally more homogeneous [31]. However, the differentiation of this effect is beyond the scope of this paper.

Consequently, the reduction of the anodic catalyst material down to ultralow loadings seemed to come with a nonproportional reduction in tolerance versus H2S poisoning and an improved self-recovery during SD/SU processes. Hence, lowering the ISO limit for sulfur-containing compounds might not be necessary with regard to ultralow-loaded anode electrodes. However, these effects should be further confirmed in large- or full-scale cell tests using realistic automotive fuel utilizations.

#### **4. Conclusions**

The key findings from this study are that the H<sup>2</sup> Quality Standard ISO 14687-2:2012 eventually requires partial adaption to accommodate future automotive PEMFC designs, including ultralow-loaded anodic CLs, and that ultralow-loaded anodes exhibited an improved self-recovery from sulfur poisoning from simple SD/SU processes.

As expected, CO poisoning induced significant performance losses at an increasing rate and severity with decreases in the platinum loading. At an ISO concentration of 0.2 ppm CO in the fuel, the cell voltage was about 40–50% lower compared to operation with neat H<sup>2</sup> for ultralow anodic loadings, which raises the question of whether the CO limit in the H<sup>2</sup> Quality Standard needs to be reduced with regard to future anodic loadings.

When H2S was in the fuel, the ultralow-loaded anodic CLs exhibited a nonlinear reduction as opposed to the expected linear reduction in tolerance to the reduction in platinum loading. Simultaneously, these anodic CLs recovered to larger degrees from sulfur poisoning during the SD/SU processes. It is hypothesized that the nonlinear reduction in tolerance and improved self-recovery arose due to the decrease in the ratio between the CL thickness (and coherent ECSA) and the geometric cell area. As the ultralow-loaded anodes were also the thinner CLs, larger fractions of sulfur adsorbed on catalyst surfaces in proximity to pores at the CL–MPL interface could have dissolved in the water present in the form of anions, which were driven out of the cell during operation or during the SU of the PEMFCs.

However, to confirm these findings, the performance of ultralow-loaded anodic CLs in the presence of impurities should be further investigated, ideally in large- or full-scale PEMFCs using automotive fuel consumption rates.

**Author Contributions:** conceptualization, S.P.; data curation, S.P.; writing—original draft, S.P.; writing—review and editing, S.P., K.A.F., and N.Z.

**Funding:** This work was supported by the German Federal Ministry for Economy and Energy within the project HAlMa, contract no. 03ET6098A.

**Acknowledgments:** The authors thank Jean St.-Pierre at the University of Hawaii (Manoa) for technical discussions and suggestions as well as Greenerity GmbH for providing the CCM materials.

**Conflicts of Interest:** The authors declare no conflicts of interest. The funders had no role in the design of the study; in the collection, analysis, or interpretation of the data; in the writing of the manuscript; or in the decision to publish the results.

#### **References**


**Sample Availability:** Not available.

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Impact of the Cathode Pt Loading on PEMFC Contamination by Several Airborne Contaminants**

#### **Jean St-Pierre \* and Yunfeng Zhai**

Hawaii Natural Energy Institute, University of Hawaii—Manoa, Honolulu, HI 96822, USA; yunfeng@hawaii.edu **\*** Correspondence: jsp7@hawaii.edu; Tel.: +1-808-956-3909

Academic Editor: Shangfeng Du Received: 1 February 2020; Accepted: 23 February 2020; Published: 27 February 2020

**Abstract:** Proton exchange membrane fuel cells (PEMFCs) with 0.1 and 0.4 mg Pt cm−<sup>2</sup> cathode catalyst loadings were separately contaminated with seven organic species: Acetonitrile, acetylene, bromomethane, iso-propanol, methyl methacrylate, naphthalene, and propene. The lower catalyst loading led to larger cell voltage losses at the steady state. Three closely related electrical equivalent circuits were used to fit impedance spectra obtained before, during, and after contamination, which revealed that the cell voltage loss was due to higher kinetic and mass transfer resistances. A significant correlation was not found between the steady-state cell voltage loss and the sum of the kinetic and mass transfer resistance changes. Major increases in research program costs and efforts would be required to find a predictive correlation, which suggests a focus on contamination prevention and recovery measures rather than contamination mechanisms.

**Keywords:** proton exchange membrane fuel cells; durability; contamination; cathode; catalyst loading

#### **1. Introduction**

Vehicles propelled by proton exchange membrane fuel cells (PEMFCs) are already commercially available. However, opportunities still exist to improve the technology because it is not expected to mature within the foreseeable future [1]. For instance, research activities are still ongoing to reduce cost while maintaining durability with a lower amount of Pt catalyst [2]. Contaminants in air jeopardize PEMFC operation by increasing the cell voltage degradation rate [3] if the intake filter [4] is saturated or damaged. Therefore, risks associated with contamination of low Pt loaded PEMFCs need to be assessed to support commercialization. Furthermore, fuel cell design robustness could be improved by integrating additional mitigation approaches derived from contamination mechanisms.

Only a few publications discuss the impact of the anode catalyst loading during PEMFC exposures to reformate fuel contaminants, such as CO, CO2, H2S, NH3, and halogenated compounds. All of these species are included in the hydrogen fuel standard [5]. For CO and H2S, a lower Pt or PtRu catalyst loading generally leads to an increase in the anode overpotential [6–10]. However, it was reported that for H2S, the catalyst loading effect disappears for values equal to or below 25 µg cm−<sup>2</sup> [10]. An effect was not observed with the weak contaminant CO2, which is attributed to a concentration that was substantially lower (1%) [7] than in a typical reformate (10–20%) [11]. The same situation was noted for NH3, which is assigned to a rapid conversion to NH<sup>4</sup> <sup>+</sup> in the presence of protons or water [12,13], followed by ion exchange with ionomer H<sup>+</sup> and transport to the cathode away from the anode under the influence of the electric field [14,15]. For halogenated compounds, a decrease in the Pt catalyst loading of both electrodes led to a faster degradation rate in the presence of HCl in both reactant stream humidifiers [16]. The effect of the anode Pt catalyst loading was exploited to develop sensor cells that are more sensitive to contamination by CO, H2S, and NH<sup>3</sup> in H2. These sensors were either based on a PEMFC [17] or a H<sup>2</sup> pump [18] design. Only two PEMFC contamination documents

focusing on the cathode catalyst loading effect were found [19,20]. However, contamination data in [19] are not directly comparable because both the catalyst layer design and catalyst loading were concurrently altered. The authors also refer to 10 ppb SO<sup>2</sup> data obtained by another group that showed more severe fuel cell damage with a catalyst loading decrease from 0.4 to 0.3 mg Pt cm−<sup>2</sup> . In contrast, the effect of 2,6-diaminotoluene, a species that leaches out of the fuel cell system balance of plant materials, was more impactful after the Pt catalyst loading was lowered from 0.4 to 0.1 mg Pt cm−<sup>2</sup> [20]. In comparison to the anode, the higher cathode potential is expected to affect the contamination mechanism with, for example, a different Pt surface charge, altered contaminant adsorbates and reaction intermediates, catalyst coverage, and cell voltage loss. This situation is exacerbated with a catalyst loading change, which affects the overpotential of the irreversible oxygen reduction reaction and the cathode potential. Information about chemical and electrochemical reactions for specific contaminants may be available in the literature. However, the presence of relevant cathode reactants, oxygen and water, may not be considered. For instance, novel intermediates or products were not detected with chlorobenzene in air [21]. However, the presence of acetylene in air led to small amounts of methane [22] that were not expected based on acetylene chemistry and electrochemistry. Therefore, tests completed under these significantly different operating conditions are needed in part because contaminant reactions are not currently predictable in assessing catalyst coverage and cell voltage loss.

This report documents the impact of the cathode Pt catalyst loading effect for PEMFCs contaminated with seven model organic airborne species, which were previously evaluated and selected from a larger pool of 21 contaminants [23]: Acetonitrile (nitrile), acetylene (alkyne), bromomethane (halocarbon), iso-propanol (alcohol), methyl methacrylate (ester), naphthalene (polycyclic aromatic), and propene (alkene). Cell voltage transients obtained under galvanostatic conditions were recorded for this analysis. Additionally, impedance spectroscopy data were acquired to facilitate the development of predictive correlations and contamination mechanisms.

#### **2. Results and Discussion**

#### *2.1. Cell Voltage Transients*

Figure 1a depicts voltage transients for cells temporarily exposed to 20 ppm CH3CN. The cell voltage for the first 5 h is constant and higher for the 0.4 mg Pt cm−<sup>2</sup> catalyst loading. This observation is consistent with previously published data for Gore catalyst coated membranes with the same cathode catalyst loadings and gas diffusion layers (Sigracet 25 BC) [24]. After approximately 5 h of operation, acetonitrile was injected into the cell, which led to a rapid cell voltage decrease that progressively slowed until a steady state was reached. For acetonitrile, the cell voltage loss was larger for the 0.1 mg Pt cm−<sup>2</sup> catalyst loading. Subsequently, the acetonitrile injection was interrupted, which quickly initiated a voltage recovery that gradually decelerated until a new steady state was reached. For acetonitrile, the cell voltage after recovery coincided with the value before contaminant injection. Acetonitrile contamination and recovery transients are qualitatively and quantitatively consistent with prior results [23,25–28]. At irregular intervals and during all baseline, contamination, and recovery stages, cell voltage transients were minimally disrupted for a short period by impedance spectroscopy measurements and the superimposition of a current signal of a small amplitude and variable frequency.

propene.

bromomethane and bromide on the catalyst surface. The removal of bromide from the catalyst surface is equally hindered due to an unfavorable cathode potential that is significantly higher than the potential of zero charge, preventing bromide desorption and Donnan exclusion, which explains the incomplete voltage recovery. During isopropanol contamination, the voltage is characterized by rapid fluctuations (Figure 1d), which were not observed for lower isopropanol concentrations [23,31]. These fluctuations are attributed to isopropanol, a surfactant commonly used to disperse Pt/C catalyst particles in solution [32], which adsorbs on carbon materials (gas diffusion layer, catalyst support) [33] and modifies liquid water management (buildup and release of liquid water drops), as previously proposed for acetylene [34]. A higher number of buildup and release events of water drops and a higher voltage fluctuation frequency for the lower catalyst loading (Figure 1d) may be related to the lower cathode potential (cell voltage compensated by a similar ohmic drop), which leaves a higher proportion of isopropanol surfactant unoxidized (oxidation initiated at a potential above 0.32 V vs. the reversible hydrogen electrode (RHE) [35]) and more hydrophilic carbon surfaces. The effect of naphthalene was rapid and severe for the 0.1 mg Pt cm−<sup>2</sup> catalyst loading (Figure 1f). As a result, the current density was temporarily lowered and the contaminant injection was interrupted before a steady state was obtained to avoid an automatic test station shutdown. Contamination and recovery transients are qualitatively and quantitatively consistent with the prior results for acetylene [23,36–39], bromomethane [23,28–30], isopropanol [23,31], methyl methacrylate [23,31], naphthalene

**Figure 1.** Cell voltage transients resulting from a temporary contaminant injection. (**a**) Acetonitrile; (**b**) acetylene; (**c**) bromomethane; (**d**) isopropanol; (**e**) methyl methacrylate; (**f**) naphthalene; (**g**) **Figure 1.** Cell voltage transients resulting from a temporary contaminant injection. (**a**) Acetonitrile; (**b**) acetylene; (**c**) bromomethane; (**d**) isopropanol; (**e**) methyl methacrylate; (**f**) naphthalene; (**g**) propene.

(23% to 89% in comparison to 1.2% to 43%). After the recovery period, the cell voltage change is minimal and independent of the catalyst loading, varying from −1.7% to 2%, with the exception of bromomethane (−40% to −45%). The larger cell voltage loss during contamination for the low catalyst loading is an important consideration for the selection of tolerance limits for commercially relevant catalyst loadings. Data obtained with a 0.4 mg Pt cm−<sup>2</sup> catalyst loading were used to derive tolerance limits [42]. The data of Table 1 suggest that these tolerance limits require a revision for a 0.1 mg Pt cm−<sup>2</sup> catalyst loading and additional tests carried out over a range of concentrations. In contrast, International Organization for Standardization (ISO) tolerance limits for hydrogen contaminants [5,43], which do not take account of the catalyst loading effect, were deemed too strict for

operating conditions (high fuel utilization, fuel recirculation) [44]. The formaldehyde tolerance limit

formaldehyde and formic acid, a low anode catalyst loading of 0.05 mg Pt cm−<sup>2</sup>

was recently modified from 10 [43] to 200 ppb [5].

Table 1 summarizes steady-state cell voltages before, during, and after contamination as well as

, and automotive

Figure 1b–g illustrates voltage transients for the other contaminants. Most of these transients share common features, including a similar initial baseline voltage, a relatively rapid voltage decrease until a steady state is reached, and a complete voltage recovery after contaminant injection was stopped. However, bromomethane transients were significantly slower and only a small fraction of the voltage loss was recovered (Figure 1c). This behavior is the result of a rapid bromomethane hydrolysis within the cell, producing methanol and bromide [28–30]. The effective bromomethane concentration is lower than the nominal value, whereas bromide is prevented from penetrating the ionomer by Donnan exclusion [12]. This situation delays the stronger and inhibiting adsorption of bromomethane and bromide on the catalyst surface. The removal of bromide from the catalyst surface is equally hindered due to an unfavorable cathode potential that is significantly higher than the potential of zero charge, preventing bromide desorption and Donnan exclusion, which explains the incomplete voltage recovery. During isopropanol contamination, the voltage is characterized by rapid fluctuations (Figure 1d), which were not observed for lower isopropanol concentrations [23,31]. These fluctuations are attributed to isopropanol, a surfactant commonly used to disperse Pt/C catalyst particles in solution [32], which adsorbs on carbon materials (gas diffusion layer, catalyst support) [33] and modifies liquid water management (buildup and release of liquid water drops), as previously proposed for acetylene [34]. A higher number of buildup and release events of water drops and a higher voltage fluctuation frequency for the lower catalyst loading (Figure 1d) may be related to the lower cathode potential (cell voltage compensated by a similar ohmic drop), which leaves a higher proportion of isopropanol surfactant unoxidized (oxidation initiated at a potential above 0.32 V vs. the reversible hydrogen electrode (RHE) [35]) and more hydrophilic carbon surfaces. The effect of naphthalene was rapid and severe for the 0.1 mg Pt cm−<sup>2</sup> catalyst loading (Figure 1f). As a result, the current density was temporarily lowered and the contaminant injection was interrupted before a steady state was obtained to avoid an automatic test station shutdown. Contamination and recovery transients are qualitatively and quantitatively consistent with the prior results for acetylene [23,36–39], bromomethane [23,28–30], isopropanol [23,31], methyl methacrylate [23,31], naphthalene [23,40], and propene [23,28,31,41].

Table 1 summarizes steady-state cell voltages before, during, and after contamination as well as the cell voltage change during and after contamination for both catalyst loadings. The cell voltage decrease during the contamination period is generally higher for the 0.1 mg Pt cm−<sup>2</sup> catalyst loading (23% to 89% in comparison to 1.2% to 43%). After the recovery period, the cell voltage change is minimal and independent of the catalyst loading, varying from −1.7% to 2%, with the exception of bromomethane (−40% to −45%). The larger cell voltage loss during contamination for the low catalyst loading is an important consideration for the selection of tolerance limits for commercially relevant catalyst loadings. Data obtained with a 0.4 mg Pt cm−<sup>2</sup> catalyst loading were used to derive tolerance limits [42]. The data of Table 1 suggest that these tolerance limits require a revision for a 0.1 mg Pt cm−<sup>2</sup> catalyst loading and additional tests carried out over a range of concentrations. In contrast, International Organization for Standardization (ISO) tolerance limits for hydrogen contaminants [5,43], which do not take account of the catalyst loading effect, were deemed too strict for formaldehyde and formic acid, a low anode catalyst loading of 0.05 mg Pt cm−<sup>2</sup> , and automotive operating conditions (high fuel utilization, fuel recirculation) [44]. The formaldehyde tolerance limit was recently modified from 10 [43] to 200 ppb [5].

The magnitude of the cell voltage change during contamination (Table 1) with catalyst loading is species-dependent. For instance, the catalyst loading hardly affected the cell voltage loss for bromomethane (−43% and −47%), whereas for acetylene, the cell voltage loss substantially increased from −1.2% to −85% with a catalyst loading decrease. This observation is attributed to different contamination mechanisms. Impedance spectroscopy data obtained during contamination by all Table 1 species and with a 0.4 mg Pt cm−<sup>2</sup> catalyst loading revealed that kinetic, ohmic, and mass transport overpotentials were impacted [42]. These and additional impedance spectroscopy data acquired with a 0.1 mg Pt cm−<sup>2</sup> catalyst loading were analyzed to evaluate the existence of a correlation between these resistances and the cell voltage loss due to contamination at the steady state.


methacrylate

Methyl methacrylate

Methyl


**Table 1.** Steady-state cell voltages at the end of each contamination period, and steady-state cell voltage changes during and after contamination. Propene 0.1 0.609 0.209 0.621 −66 2.0 Propene 0.1 0.609 0.209 0.621 −66 2.0 0.4 0.673 0.598 0.678 −11 0.74 0.4 0.673 0.598 0.678 −11 0.74 0.4 0.671 0.495 0.677 −26 0.89

0.4 0.671 0.495 0.677 −26 0.89

0.4 0.673 0.598 0.678 −11 0.74

Naphthalene 0.1 0.557 0.060 2 0.566 −89 1.6

0.1 0.599 0.444 0.608 −26 1.5

0.4 0.681 0.623 0.687 −8.5 0.88

0.4 0.681 0.623 0.687 −8.5 0.88

0.4 0.681 0.623 0.687 −8.5 0.88

0.4 0.673 0.598 0.678 −11 0.74

Isopropanol 0.1 0.570 0.439 0.577 −23 1.2

Isopropanol 0.1 0.570 0.439 0.577 −23 1.2

0.4 0.671 0.495 0.677 −26 0.89

*Molecules* **2020**, *25*, x 5 of 15

**2020**, *25*, x 5 of 15

*Molecules* 

*Molecules* **2020**, *25*, x 5 of 15

**Table 1.** Steady-state cell voltages at the end of each contamination period, and steady-state cell

**Table 1.** Steady-state cell voltages at the end of each contamination period, and steady-state cell

*Molecules* **2020**, *25*, x 5 of 15

**Table 1.** Steady-state cell voltages at the end of each contamination period, and steady-state cell

**Table 1.** Steady-state cell voltages at the end of each contamination period, and steady-state cell

voltage changes during and after contamination.

voltage changes during and after contamination.

voltage changes during and after contamination.

**Contaminant** 

**Contaminant** 

**Contaminant** 

**Catalyst Loading/ mg Pt cm–2**

**Contaminant** 

**Catalyst Loading/ mg Pt cm–2**

**Catalyst Loading/ mg Pt cm–2**

voltage changes during and after contamination.

**Catalyst Loading/ mg Pt cm–2**

**Before Contamination 1**

**Before Contamination 1**

**Before Contamination 1**

**Before Contamination 1**

**During Contamination** 

**During Contamination** 

**During Contamination**  Acetonitrile 0.1 0.597 0.198 0.587 −67 −1.7

Acetonitrile 0.1 0.597 0.198 0.587 −67 −1.7

**After Contamination** 

**During Contamination** 

Acetonitrile 0.1 0.597 0.198 0.587 −67 −1.7

0.4 0.666 0.412 0.670 −38 0.60

*Molecules* **2020**, *25*, x 5 of 15 **Table 1.** Steady-state cell voltages at the end of each contamination period, and steady-state cell

Acetonitrile 0.1 0.597 0.198 0.587 −67 −1.7

0.4 0.666 0.412 0.670 −38 0.60

0.4 0.666 0.412 0.670 −38 0.60

Acetylene 0.1 0.607 0.093 0.608 −85 0.16

0.4 0.666 0.412 0.670 −38 0.60

*Molecules* **2020**, *25*, x 5 of 15 **Table 1.** Steady-state cell voltages at the end of each contamination period, and steady-state cell

*Molecules* **2020**, *25*, x 5 of 15 **Table 1.** Steady-state cell voltages at the end of each contamination period, and steady-state cell

Acetylene 0.1 0.607 0.093 0.608 −85 0.16

Acetylene 0.1 0.607 0.093 0.608 −85 0.16

0.4 0.672 0.664 0.672 −1.2 0

0.4 0.672 0.664 0.672 −1.2 0

0.4 0.672 0.664 0.672 −1.2 0

Bromomethane 0.1 0.561 0.299 0.309 −47 −45

Bromomethane 0.1 0.561 0.299 0.309 −47 −45

Acetylene 0.1 0.607 0.093 0.608 −85 0.16

voltage changes during and after contamination.

**Contaminant** 

**Contaminant** 

**Contaminant** 

**Catalyst Loading/ mg Pt cm–2**

voltage changes during and after contamination.

voltage changes during and after contamination.

**Before Contamination 1**

> **Catalyst Loading/ mg Pt cm–2**

**Catalyst Loading/ mg Pt cm–2**

**During Contamination** 

Bromomethane 0.1 0.561 0.299 0.309 −47 −45

**Cell Voltage / V Cell Voltage Percentage Change /** 

0.4 0.672 0.664 0.672 −1.2 0

0.4 0.663 0.376 0.398 −43 −40

0.4 0.663 0.376 0.398 −43 −40

**%** 

0.4 0.663 0.376 0.398 −43 −40

**During Contamination** 

**%** 

**%** 

**After Contamination** 

Isopropanol 0.1 0.570 0.439 0.577 −23 1.2

Isopropanol 0.1 0.570 0.439 0.577 −23 1.2

**After Contamination** 

Bromomethane 0.1 0.561 0.299 0.309 −47 −45

**Cell Voltage / V Cell Voltage Percentage Change /** 

**Cell Voltage / V Cell Voltage Percentage Change /** 

0.4 0.663 0.376 0.398 −43 −40

Isopropanol 0.1 0.570 0.439 0.577 −23 1.2

**After Contamination** 

**After Contamination** 

Acetonitrile 0.1 0.597 0.198 0.587 −67 −1.7

**During Contamination** 

**During Contamination** 

**Before Contamination 1**

**Before Contamination 1**

0.4 0.681 0.623 0.687 −8.5 0.88

0.4 0.666 0.412 0.670 −38 0.60

Isopropanol 0.1 0.570 0.439 0.577 −23 1.2

Acetylene 0.1 0.607 0.093 0.608 −85 0.16

Acetonitrile 0.1 0.597 0.198 0.587 −67 −1.7

Acetonitrile 0.1 0.597 0.198 0.587 −67 −1.7

0.4 0.681 0.623 0.687 −8.5 0.88

**During Contamination** 

**During Contamination** 

**After Contamination** 

**After Contamination** 

0.4 0.681 0.623 0.687 −8.5 0.88

0.1 0.599 0.444 0.608 −26 1.5

0.4 0.672 0.664 0.672 −1.2 0

0.4 0.666 0.412 0.670 −38 0.60

0.4 0.666 0.412 0.670 −38 0.60

Bromomethane 0.1 0.561 0.299 0.309 −47 −45

Acetylene 0.1 0.607 0.093 0.608 −85 0.16

Acetylene 0.1 0.607 0.093 0.608 −85 0.16

0.4 0.663 0.376 0.398 −43 −40

0.4 0.672 0.664 0.672 −1.2 0

0.4 0.672 0.664 0.672 −1.2 0

0.1 0.599 0.444 0.608 −26 1.5

Isopropanol 0.1 0.570 0.439 0.577 −23 1.2

Bromomethane 0.1 0.561 0.299 0.309 −47 −45

0.4 0.663 0.376 0.398 −43 −40

0.4 0.663 0.376 0.398 −43 −40

Bromomethane 0.1 0.561 0.299 0.309 −47 −45

0.1 0.599 0.444 0.608 −26 1.5

0.4 0.681 0.623 0.687 −8.5 0.88

0.1 0.599 0.444 0.608 −26 1.5

0.4 0.673 0.598 0.678 −11 0.74

0.4 0.673 0.598 0.678 −11 0.74

0.4 0.673 0.598 0.678 −11 0.74

Methyl methacrylate

Methyl methacrylate

Methyl methacrylate

Methyl methacrylate

**After Contamination** 

**After Contamination** 

**Cell Voltage / V Cell Voltage Percentage Change /** 

**After Contamination** 

**During Contamination** 

**During Contamination** 

**During Contamination** 

**During Contamination** 

**%** 

**After Contamination** 

**After Contamination** 

**After Contamination** 

**After Contamination** 

**Cell Voltage / V Cell Voltage Percentage Change /** 

**Cell Voltage / V Cell Voltage Percentage Change /** 

**Cell Voltage / V Cell Voltage Percentage Change /** 

**%** 

**%** 

**%** 

The magnitude of the cell voltage change during contamination (Table 1) with catalyst loading is species-dependent. For instance, the catalyst loading hardly affected the cell The magnitude of the cell voltage change during contamination (Table 1) with catalyst loading is species-dependent. For instance, the catalyst loading hardly affected the cell The magnitude of the cell voltage change during contamination (Table 1) with catalyst loading is species-dependent. For instance, the catalyst loading hardly affected the cell voltage loss for bromomethane (−43% and −47%), whereas for acetylene, the cell voltage loss loading is species-dependent. For instance, the catalyst loading hardly affected the cell voltage loss for bromomethane (−43% and −47%), whereas for acetylene, the cell voltage loss substantially increased from −1.2% to −85% with a catalyst loading decrease. This observation The magnitude of the cell voltage change during contamination (Table 1) with catalyst loading is species-dependent. For instance, the catalyst loading hardly affected the cell voltage loss for bromomethane (−43% and −47%), whereas for acetylene, the cell voltage loss The magnitude of the cell voltage change during contamination (Table 1) with catalyst loading is species-dependent. For instance, the catalyst loading hardly affected the cell The magnitude of the cell voltage change during contamination (Table 1) with catalyst 1 For 0.1 mg Pt cm−2 , mean = 0.586 V and standard deviation = 0.022 V. For 0.4 mg Pt cm−2 , mean = 0.671 V and standard deviation = 0.006 V. 2 Not at steady state because the cell voltage was still decreasing.

voltage loss for bromomethane (−43% and −47%), whereas for acetylene, the cell voltage loss substantially increased from −1.2% to −85% with a catalyst loading decrease. This observation is attributed to different contamination mechanisms. Impedance spectroscopy data obtained during contamination by all Table 1 species and with a 0.4 mg Pt cm−2 catalyst loading

loading is species-dependent. For instance, the catalyst loading hardly affected the cell voltage loss for bromomethane (−43% and −47%), whereas for acetylene, the cell voltage loss substantially increased from −1.2% to −85% with a catalyst loading decrease. This observation is attributed to different contamination mechanisms. Impedance spectroscopy data obtained during contamination by all Table 1 species and with a 0.4 mg Pt cm−2 catalyst loading

is attributed to different contamination mechanisms. Impedance spectroscopy data obtained during contamination by all Table 1 species and with a 0.4 mg Pt cm−2 catalyst loading

voltage loss for bromomethane (−43% and −47%), whereas for acetylene, the cell voltage loss substantially increased from −1.2% to −85% with a catalyst loading decrease. This observation is attributed to different contamination mechanisms. Impedance spectroscopy data obtained during contamination by all Table 1 species and with a 0.4 mg Pt cm−2 catalyst loading

substantially increased from −1.2% to −85% with a catalyst loading decrease. This observation is attributed to different contamination mechanisms. Impedance spectroscopy data obtained during contamination by all Table 1 species and with a 0.4 mg Pt cm−2 catalyst loading

voltage loss for bromomethane (−43% and −47%), whereas for acetylene, the cell voltage loss substantially increased from −1.2% to −85% with a catalyst loading decrease. This observation is attributed to different contamination mechanisms. Impedance spectroscopy data obtained during contamination by all Table 1 species and with a 0.4 mg Pt cm−2 catalyst loading

substantially increased from −1.2% to −85% with a catalyst loading decrease. This observation is attributed to different contamination mechanisms. Impedance spectroscopy data obtained during contamination by all Table 1 species and with a 0.4 mg Pt cm−2 catalyst loading

#### *2.2. Impedance Spectra*

Figure 2a shows impedance spectra (Nyquist representation) for a 0.1 mg Pt cm−<sup>2</sup> catalyst loading, before, during, and after acetonitrile contamination. All three spectra share the same features and have two main loops that are respectively attributed to oxygen reduction (medium frequencies) and oxygen mass transfer (low frequencies) [45]. A third loop ascribed to hydrogen oxidation is barely visible as a hump at high frequencies [45]. The high-frequency intercept represents the ohmic resistance, which is mostly caused by the poorly conducting membrane [45]. Multiple explanations were proposed for the inductive impedance values at the lowest and highest frequencies, including electrical cables [46,47] for high frequencies, and processes involving side reactions with intermediate species [47], oxide growth [48], or a slow ionomer water uptake/release [49] for low frequencies. Most of these considerations were either ignored because they did not focus on relevant aspects (electrical cables) or were easily dismissed because, in the absence of contaminants, the cathode potential was too low for Pt oxidation and the sub-saturated air stream did not yield an inductive behavior. For the 0.4 mg Pt cm−<sup>2</sup> catalyst loading, the average cell voltage of 0.671 V (Table 1) compensated with an ohmic loss of 0.1 V for a worst-case scenario (1 A cm−<sup>2</sup> <sup>×</sup> 0.1 <sup>Ω</sup> cm<sup>2</sup> from the high-frequency intercepts in Figure 2a) leads to a cathode potential of 0.771 V vs. RHE, which is lower than the smallest Pt oxidation potential of 0.837 V vs. RHE [50]. Acetonitrile contamination causes an increase in the high-frequency intercept and a diameter increase for both main loops (Figure 2a). An increase in ohmic loss was only observed with acetonitrile, owing to the production of ammonium cations by hydrolysis, which displace protons as the main charge carriers in the ionomer [28,51]. In relative terms, this effect is significantly smaller than the kinetic and mass transfer effects, with an approximate doubling of both oxygen reduction and transport loop diameters. However, because the effect is cumulative, a larger change is observed for a longer exposure duration [26]. After contamination, the high-frequency intercept returns to its original value, and both main loops decrease in size to a diameter that is smaller than the original value. These impedance spectra agree with prior results [25–28]. However, smaller kinetic and mass transfer loops are inconsistent with a complete cell voltage recovery (Figure 1a, Table 1). This observation is possibly due to subtle structural or other changes that are not detectable by cell voltage measurements, such as Pt surface reconstruction in the presence of foreign species [52,53]. The oxygen reduction and mass transfer resistances before, during, and after contamination were generally obtained by curve-fitting an equivalent circuit developed for a PEMFC contaminated with SO<sup>2</sup> (Figure 3a) [54]. Resistances during contamination for acetonitrile and a 0.1 mg Pt cm−<sup>2</sup> catalyst loading were derived using a modified equivalent circuit that accounts for the inductive behavior at low frequencies (Figure 3b) [55,56]. Resistances during contamination for acetonitrile (0.4 mg Pt cm−<sup>2</sup> ) and propene (0.1 mg Pt cm−<sup>2</sup> ) were obtained using a modified version of the Figure 3b equivalent circuit by omitting the cathode resistance Rk (Figure 3b) to limit the number of parameters (Figure 3c). The impedance spectra are accurately represented by the equivalent circuit models (Figure 2a–f). The resistance values are discussed later in this section.

Most of the other impedance spectra for both catalyst loadings and all contaminants are equally well represented by the equivalent circuits shown in Figure 3a,c. For this reason, only a selection is given in Figure 2. A few spectra could not be fitted with any of the equivalent circuits in Figure 3a–c for a few 0.1 mg Pt cm−<sup>2</sup> catalyst loading cases. For acetylene, the impedance spectrum during contamination was approximately a single loop of a large diameter that could not be fitted to a two-loop equivalent circuit. For isopropanol, cell voltage fluctuations during contamination (Figure 1d) created a low frequency artefact that also prevented the use of the equivalent circuits of Figure 3a or Figure 3c. For naphthalene, the cell voltage transient was interrupted before a steady state was obtained (Figure 1f), which also led to a low-frequency artefact that could not be fitted to the equivalent circuits of Figure 3a–c. The impedance spectra agree with the prior results for acetonitrile [25–28], acetylene [36–38], bromomethane [28–30], isopropanol [31], methyl methacrylate [31], naphthalene [40], and propene [28,31,41].

**Figure 2.** Impedance spectra before, during, and after contamination by acetonitrile in (**a**) and (**b**), bromomethane in (**c**) and (**d**), and methyl methacrylate in (**e**) and (**f**) for Pt catalyst loadings of 0.1 mg cm<sup>−</sup><sup>2</sup> in (**a**), (**c**), and (**e**), and 0.4 mg cm<sup>−</sup><sup>2</sup> in (**b**), (**d**), and (**f**). **Figure 2.** Impedance spectra before, during, and after contamination by acetonitrile in (**a**) and (**b**), bromomethane in (**c**) and (**d**), and methyl methacrylate in (**e**) and (**f**) for Pt catalyst loadings of 0.1 mg cm−<sup>2</sup> in (**a**), (**c**), and (**e**), and 0.4 mg cm−<sup>2</sup> in (**b**), (**d**), and (**f**).

propene (0.1 mg Pt cm–<sup>2</sup>

) contamination.

Table 2 collects kinetic and mass transfer resistances before, during, and after contamination for both catalyst loadings. Dimensionless kinetic and mass transfer resistances during and after contamination are also given in Table 2. The dimensionless kinetic and mass transfer resistances concurrently increase during contamination and are 1.05, with the exception of the 0.93 dimensionless mass transfer resistance for isopropanol and a 0.4 mg Pt cm−<sup>2</sup> catalyst loading. The isopropanol anomaly may be related to water management, as discussed in the previous section. A hypothesized connection between kinetic and mass transfer resistances during contamination [34] was recently substantiated [57]. Contaminant adsorbates covering the catalyst surface increase the effective current density closer to the limiting value and mass transfer losses in the ionomer layer covering the catalyst. This situation is similar to a decrease in catalyst loading, which has been shown to also increase mass transfer losses [58,59]. The dimensionless kinetic and mass transfer resistances after recovery, with the exception of bromomethane, indicate an incomplete recovery that is less extensive for the lower catalyst loading. For the dimensionless kinetic resistance, values are 0.832 and 0.95 for, respectively, 0.1 and 0.4 mg Pt cm–<sup>2</sup> catalyst loadings. For the dimensionless mass transfer resistance, values are 0.842 and 0.88 for, respectively, 0.1 and 0.4 mg Pt cm–<sup>2</sup> catalyst loadings. These results are in contrast with the data of Figure 1 and Table 1, showing a complete recovery within experimental error, with the exception of bromomethane. The discrepancy between the recovery extents of cell voltage and kinetic and mass transfer resistances is due to the higher sensitivity of impedance measurements and the movement of the reaction front (current density and catalyst layer effectiveness redistributions over the catalyst layer thickness). The hydrogen peroxide yield is enhanced in the presence of acetonitrile, acetylene, methyl methacrylate, naphthalene, and propene [60–63]. The elevated level of hydrogen peroxide in turn promotes ionomer degradation [64] and structural modifications to the catalyst layer that are relatively more impactful for the lower catalyst loading. Therefore, in view of the lower cell voltage and cathode potential for a lower catalyst loading (Figure 1, Table 1), a higher hydrogen peroxide yield [60–63] and ionomer degradation are expected. Tafel plots obtained before and after contamination with acetylene (Figure 4) support this

**Figure 3.** Equivalent circuit models for a proton exchange membrane fuel cell (PEMFC). (**a**) The model previously derived for SO<sup>2</sup> contamination and used for all 7 organic contaminants investigated in this work; (**b**) the model previously derived to capture low-frequency inductive data in the absence of contaminants and used to fit data during acetonitrile contamination (0.1 mg Pt cm–<sup>2</sup> ); (**c**) the modified model (**b**) to capture low-frequency inductive data obtained during acetonitrile (0.4 mg Pt cm–<sup>2</sup> ) and **Figure 3.** Equivalent circuit models for a proton exchange membrane fuel cell (PEMFC). (**a**) The model previously derived for SO<sup>2</sup> contamination and used for all 7 organic contaminants investigated in this work; (**b**) the model previously derived to capture low-frequency inductive data in the absence of contaminants and used to fit data during acetonitrile contamination (0.1 mg Pt cm−<sup>2</sup> ); (**c**) the modified model (**b**) to capture low-frequency inductive data obtained during acetonitrile (0.4 mg Pt cm−<sup>2</sup> ) and propene (0.1 mg Pt cm−<sup>2</sup> ) contamination.

Table 2 collects kinetic and mass transfer resistances before, during, and after contamination for both catalyst loadings. Dimensionless kinetic and mass transfer resistances during and after contamination are also given in Table 2. The dimensionless kinetic and mass transfer resistances concurrently increase during contamination and are ≥1.05, with the exception of the 0.93 dimensionless mass transfer resistance for isopropanol and a 0.4 mg Pt cm−<sup>2</sup> catalyst loading. The isopropanol anomaly may be related to water management, as discussed in the previous section. A hypothesized connection between kinetic and mass transfer resistances during contamination [34] was recently substantiated [57]. Contaminant adsorbates covering the catalyst surface increase the effective current density closer to the limiting value and mass transfer losses in the ionomer layer covering the catalyst. This situation is similar to a decrease in catalyst loading, which has been shown to also increase mass transfer losses [58,59]. The dimensionless kinetic and mass transfer resistances after recovery, with the exception of bromomethane, indicate an incomplete recovery that is less extensive for the lower catalyst loading. For the dimensionless kinetic resistance, values are ≥0.832 and ≥0.95 for, respectively, 0.1 and 0.4 mg Pt cm−<sup>2</sup> catalyst loadings. For the dimensionless mass transfer resistance, values are ≥0.842 and <sup>≥</sup>0.88 for, respectively, 0.1 and 0.4 mg Pt cm−<sup>2</sup> catalyst loadings. These results are in contrast with the data of Figure 1 and Table 1, showing a complete recovery within experimental error, with the exception of bromomethane. The discrepancy between the recovery extents of cell voltage and kinetic and mass transfer resistances is due to the higher sensitivity of impedance measurements and the movement of the reaction front (current density and catalyst layer effectiveness redistributions over the catalyst layer thickness). The hydrogen peroxide yield is enhanced in the presence of acetonitrile, acetylene, methyl methacrylate, naphthalene, and propene [60–63]. The elevated level of hydrogen peroxide in turn promotes ionomer degradation [64] and structural modifications to the catalyst layer that are relatively more impactful for the lower catalyst loading. Therefore, in view of the lower cell voltage and cathode potential for a lower catalyst loading (Figure 1, Table 1), a higher hydrogen peroxide yield [60–63] and ionomer degradation are expected. Tafel plots obtained before and after contamination with acetylene (Figure 4) support this hypothesis, with a larger cell voltage loss for the 0.1 mg Pt cm−<sup>2</sup> catalyst loading (7.9 mV in comparison to 2.9 mV).

propene (0.1 mg Pt cm–2) contamination.


**Table 2.** Steady-state kinetic and mass transfer resistances at the end of each contamination period, and steady-state dimensionless resistances during and after contamination. **Catalyst Dimensionless** 

and steady-state dimensionless resistances during and after contamination.

*Molecules* **2020**, *25*, x 9 of 15

contaminants and used to fit data during acetonitrile contamination (0.1 mg Pt cm–2); (**c**) the modified model (**b**) to capture low-frequency inductive data obtained during acetonitrile (0.4 mg Pt cm–2) and

**Table 2.** Steady-state kinetic and mass transfer resistances at the end of each contamination period,

<sup>1</sup> Resistance during/after contamination divided by the resistance before contamination. <sup>2</sup> Equivalent circuit models do not fit due to a side surface reaction involving intermediates. <sup>3</sup> Artefact created by flooding or rapid change in cell voltage. <sup>4</sup> Data was not recorded by error. circuit models do not fit due to a side surface reaction involving intermediates. 3 Artefact created by flooding or rapid change in cell voltage. 4 Data was not recorded by error.

**Figure 4.** Tafel plots before contamination (BC) and after contamination (AC) with 100 ppm acetylene for 0.1 and 0.4 mg Pt cm–2 catalyst loadings. The change in cell voltage between plots at a current density of 0.0447 A cm–2, a value located in the middle of the range used to correlate data (0.02 to 0.1 A cm–2), ignores the slight change in slope. **Figure 4.** Tafel plots before contamination (BC) and after contamination (AC) with 100 ppm acetylene for 0.1 and 0.4 mg Pt cm−<sup>2</sup> catalyst loadings. The change in cell voltage between plots at a current density of 0.0447 A cm−<sup>2</sup> , a value located in the middle of the range used to correlate data (0.02 to 0.1 A cm−<sup>2</sup> ), ignores the slight change in slope.

#### *2.3. Contaminant Effect Prediction 2.3. Contaminant E*ff*ect Prediction*

The steady-state cell voltage loss during contamination was correlated with the sum of the kinetic and mass transfer resistance changes during contamination (Figure 5). A significant correlation was not identified, as significant deviations from Ohm's law were noted. Furthermore, it The steady-state cell voltage loss during contamination was correlated with the sum of the kinetic and mass transfer resistance changes during contamination (Figure 5). A significant correlation was not identified, as significant deviations from Ohm's law were noted. Furthermore, it is difficult to argue that there is a catalyst loading effect because the two data sets largely overlap. The absence of a correlation is not surprising, considering the effects of cell design and operating conditions on contamination. Several parameters were mentioned in an earlier attempt to correlate the effect of contaminants on oxygen reduction kinetics [65], including contaminant partial pressure and temperature, exposed

Pt surface features (crystal faces, edges), Pt state (reduced or oxidized), phase in contact with the Pt surface (air, ionomer), adsorption isotherms for O2, contaminants, and related intermediates and products, and elementary chemical and electrochemical reactions and associated rate constants for O<sup>2</sup> reduction and contaminant oxidation or reduction. This list is enlarged by factors affecting ohmic and mass transfer losses, including cation and neutral molecules' absorption isotherms influencing ionomer and membrane ionic conductivity and oxygen permeability by swelling and changing the distance between sulfonate groups, and contaminant scavenging by liquid water modifying the effective contaminant concentration [12,13,66–71]. Although cell design and operating conditions were maintained as constant as possible, with the exception of catalyst loading and contaminant concentration, the change in cell resistance is insufficiently precise to capture all contamination nuances and accurately predict the cell voltage loss (Figure 5). An accurate correlation for the cell voltage loss would be useful. However, given the amount of information that will be required and the complexity associated with the derivation of a detailed mathematical model of contamination, a focus on preventive and recovery measures may be more fruitful. This suggestion is reinforced by considering practical aspects, contaminant mixtures [28], and long-term effects [26] that increase the number of contamination parameters and the difficulty in predicting contaminant effects. related intermediates and products, and elementary chemical and electrochemical reactions and associated rate constants for O<sup>2</sup> reduction and contaminant oxidation or reduction. This list is enlarged by factors affecting ohmic and mass transfer losses, including cation and neutral molecules' absorption isotherms influencing ionomer and membrane ionic conductivity and oxygen permeability by swelling and changing the distance between sulfonate groups, and contaminant scavenging by liquid water modifying the effective contaminant concentration [12,13,66–71]. Although cell design and operating conditions were maintained as constant as possible, with the exception of catalyst loading and contaminant concentration, the change in cell resistance is insufficiently precise to capture all contamination nuances and accurately predict the cell voltage loss (Figure 5). An accurate correlation for the cell voltage loss would be useful. However, given the amount of information that will be required and the complexity associated with the derivation of a detailed mathematical model of contamination, a focus on preventive and recovery measures may be more fruitful. This suggestion is reinforced by considering practical aspects, contaminant mixtures [28], and long-term effects [26] that increase the number of contamination parameters and the difficulty in predicting contaminant effects.

*Molecules* **2020**, *25*, x 10 of 15

phase in contact with the Pt surface (air, ionomer), adsorption isotherms for O2, contaminants, and

**Figure 5.** Cell voltage loss as a function of the sum of the changes in kinetic and mass transfer **Figure 5.** Cell voltage loss as a function of the sum of the changes in kinetic and mass transfer resistance.

#### resistance. **3. Materials and Methods**

**3. Materials and Methods** A single modified Fuel Cell Technologies cell with an active area of 50 cm<sup>2</sup> and triple/double serpentine channels for the cathode/anode was used for all experiments. Gore PRIMEA M715 catalyst-coated membranes with a Pt loading of 0.1 or 0.4 mg Pt cm–<sup>2</sup> (50 % Pt/C) on each side were inserted between SGL Carbon Sigracet 25 BC gas diffusion layers. The cell was operated with a FCATS™ G050 series test station (Green Light Power Technologies). After cell activation, operating conditions were set to air/H2, 2/2 stoichiometry, 48.3/48.3 kPa<sup>g</sup> outlet pressure, 50%/100% relative humidity, 80 °C, and 1 A cm–<sup>2</sup> . Contaminant concentrations varied between 1.4 and ~8000 ppm: Acetonitrile (20 ppm), acetylene (100 ppm), bromomethane (5 ppm), isopropanol (~8000 ppm), methyl methacrylate (20 ppm), naphthalene (1.4 ppm), and propene (100 ppm). Contaminant concentrations were individually and empirically adjusted based on prior experience to cause a perceptible to significant cell voltage decrease at the steady state for the 0.4 mg Pt cm–<sup>2</sup> catalyst loading, and to leave a sufficient cell voltage window for an additional decrease induced by the lower 0.1 mg Pt cm–<sup>2</sup> catalyst loading. Contaminants were injected after the air humidifier using air-based gas mixtures for most cases. However, isopropanol and naphthalene were respectively evaporated and sublimated by employing a thermally controlled and calibrated liquid/solid holder. Contaminant A single modified Fuel Cell Technologies cell with an active area of 50 cm<sup>2</sup> and triple/double serpentine channels for the cathode/anode was used for all experiments. Gore PRIMEA M715 catalyst-coated membranes with a Pt loading of 0.1 or 0.4 mg Pt cm−<sup>2</sup> (50 % Pt/C) on each side were inserted between SGL Carbon Sigracet 25 BC gas diffusion layers. The cell was operated with a FCATS™ G050 series test station (Green Light Power Technologies). After cell activation, operating conditions were set to air/H2, 2/2 stoichiometry, 48.3/48.3 kPa<sup>g</sup> outlet pressure, 50%/100% relative humidity, 80 ◦C, and 1 A cm−<sup>2</sup> . Contaminant concentrations varied between 1.4 and ~8000 ppm: Acetonitrile (20 ppm), acetylene (100 ppm), bromomethane (5 ppm), isopropanol (~8000 ppm), methyl methacrylate (20 ppm), naphthalene (1.4 ppm), and propene (100 ppm). Contaminant concentrations were individually and empirically adjusted based on prior experience to cause a perceptible to significant cell voltage decrease at the steady state for the 0.4 mg Pt cm−<sup>2</sup> catalyst loading, and to leave a sufficient cell voltage window for an additional decrease induced by the lower 0.1 mg Pt cm−<sup>2</sup> catalyst loading. Contaminants were injected after the air humidifier using air-based gas mixtures for most cases. However, isopropanol and naphthalene were respectively evaporated and sublimated by employing a thermally controlled and calibrated liquid/solid holder. Contaminant injection was initiated after 5 h with an exposure that lasted from less than 1 to ~70 h until a steady state was achieved. After the contamination injection was interrupted, the self-induced recovery was recorded until a steady state was obtained, which necessitated between 5 and ~60 h.

During the galvanostatic experiments, impedance spectra were acquired at irregular intervals by superimposing 0.1 Hz to 10 kHz (10 points per decade) current perturbations that caused a voltage change of ~5 mV. The Solartron SI1260 impedance/gain-phase analyzer was operated with ZPlot® software (Version 2.9c, Scribner Associates, Southern Pines, NC, USA). Measurement accuracy was improved by utilizing Stanford Research SR560 low-noise preamplifiers and by winding up both load-bank cables, which have an equal length, to reduce their inductance. The ZView® software (Version 3.5e, Scribner Associates) was employed for fitting impedance spectra to equivalent circuit models. Polarization curves were only recorded before and after acetylene contamination. Polarization curves were measured by decreasing the current density from 2 to 0 (open circuit voltage) A cm−<sup>2</sup> in a stepwise fashion, allowing a stabilization time of 15 min at each stage, and otherwise using contamination test operating conditions.

#### **4. Conclusions**

The effect of Pt catalyst loading on the steady-state cell voltage loss was characterized for seven organic airborne contaminants. Impedance spectroscopy was used to gain mechanistic insight. The steady-state cell voltage loss is mostly attributed to a concurrent increase in both kinetic and mass transfer resistances that is reminiscent of the effect of a decrease in catalyst loading in the absence of a contaminant. Low Pt catalyst loadings generally lead to a larger steady-state cell voltage loss. A significant correlation between the steady-state cell voltage loss and the sum of the kinetic and mass transfer resistance changes was not found, and would only be improved with major increases in cost and effort. For this reason, it is proposed to focus activities on contamination prevention and recovery measures.

For a commercially relevant cathode catalyst loading of 0.1 mg Pt cm−<sup>2</sup> , it would be advantageous to expand the current database to other contaminants and contaminant concentrations for the derivation of tolerance limits to support the design of air filters. Although tolerance limits were previously derived for single contaminants rather than for more practically relevant mixtures [42], for very low contaminant concentrations, tolerance limits may still prove useful because the catalyst surface coverage by contaminant adsorbates may be so small that the different species may not interact. In other words, the effects of all contaminants may be additive. It would also be useful to verify this hypothesis with diluted, single, and multiple contaminant mixtures.

**Author Contributions:** Conceptualization, J.S.-P. and Y.Z.; methodology, Y.Z.; formal analysis, J.S.-P. and Y.Z.; investigation, Y.Z.; resources, Y.Z.; writing—original draft preparation, J.S.-P.; writing—review and editing, J.S.-P. and Y.Z.; visualization, J.S.-P.; supervision, J.S.-P.; project administration, J.S.-P. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the United States Department of Energy, grant number DE-EE0000467, and the Office of Naval Research, grant number N00014-17-1-2206. The APC was funded by the Office of Naval Research.

**Acknowledgments:** The authors are grateful to the Hawaiian Electric Company for their ongoing support of the operations of the Hawaii Sustainable Energy Research Facility.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**


**Sample Availability:** Samples of the compounds are not available from the authors.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Performance Recovery after Contamination with Nitrogen Dioxide in a PEM Fuel Cell**

#### **Yasna Acevedo Gomez , Göran Lindbergh and Carina Lagergren \***

Received: 24 December 2019; Accepted: 1 March 2020; Published: 2 March 2020

Applied Electrochemistry, Department of Chemical Engineering, KTH Royal Institute of Technology, 10044 Stockholm, Sweden; yasna@kth.se (Y.A.G.); gnli@kth.se (G.L.)

**\*** Correspondence: carinal@kth.se; Tel.: +46-8-790-6507

Academic Editors: Jean St-Pierre and Shangfeng Du

**Abstract:** While the market for fuel cell vehicles is increasing, these vehicles will still coexist with combustion engine vehicles on the roads and will be exposed to an environment with significant amounts of contaminants that will decrease the durability of the fuel cell. To investigate different recovery methods, in this study, a PEM fuel cell was contaminated with 100 ppm of NO<sup>2</sup> at the cathode side. The possibility to recover the cell performance was studied by using different airflow rates, different current densities, and by subjecting the cell to successive polarization curves. The results show that the successive polarization curves are the best choice for recovery; it took 35 min to reach full recovery of cell performance, compared to 4.5 h of recovery with pure air at 0.5 A cm−<sup>2</sup> and 110 mL min−<sup>1</sup> . However, the performance recovery at a current density of 0.2 A cm−<sup>2</sup> and air flow 275 mL min−<sup>1</sup> was done in 66 min, which is also a possible alternative. Additionally, two operation techniques were suggested and compared during 7 h of operation: air recovery and air depletion. The air recovery technique was shown to be a better choice than the air depletion technique.

**Keywords:** PEM fuel cell; performance; recovery; nitrogen dioxide; contamination

#### **1. Introduction**

As the world is heading towards clean energy sources, the proton exchange membrane (PEM) fuel cell plays an important role, being a good alternative for the transportation sector and stationary power systems. Automobile manufacturers have been releasing electric vehicles as a viable solution to decrease greenhouse gas emissions [1]. The fuel cell vehicle is becoming popular and may be the right solution to replace internal combustion engine (ICE) vehicles in the near future [2,3]. However, the durability of the fuel cell is still an issue, where one aspect is pollutants in the air that seriously affect the performance. It is well known that the air contains unwanted contaminants that come from ICE vehicles, agriculture, and industries. As the fuel cell market grows, fuel cell vehicles must coexist with ICE vehicles on the roads. The coexistence of these two types of vehicles may lead to a dramatic decrease in the fuel cell performance, thus a recovery strategy must be considered in a real traffic situation.

Among the contaminants in air, nitrogen dioxide is one that seriously affects the performance of the PEM fuel cell but has not been completely studied in the literature. In our previous study [4], severe degradation of the cell performance was shown at different concentrations of NO2. For all the tests, the same total dosage of NO<sup>2</sup> was added, but the possibility for the cell performance to recover after contamination differed. At higher concentrations of 50, 100, and 200 ppm NO2, the performance could only be partially recovered. In the study, a mechanism for NO<sup>2</sup> contamination was proposed based on cyclic voltammetry (CV) observation in which NO<sup>2</sup> is oxidized to NO<sup>3</sup> − at 1.05 V, then in the negative sweep reduced to NO<sup>2</sup> <sup>−</sup> at 0.68 V, followed by a subsequent reduction of NO<sup>2</sup> <sup>−</sup> to N2O and/or NH2OH at potentials lower than 0.5 V. The proposed mechanism was confirmed by the detection of NO as intermediate species and N2O by simultaneous mass spectrometry.

Other authors have shown that the contamination can be fully recovered in some cases [5,6], almost recovered in other cases [6–8], or not recovered [5,9], depending on the NO<sup>2</sup> concentration, exposure time, and operating conditions. Misz et al. [6] and Jing et al. [8] tested the contamination with 1 ppm NO<sup>2</sup> over 1 and 100 h, respectively; the shorter exposure time resulted in fully recovered performance while performance following the longer exposure time was almost recovered after cyclic voltammetry scan as a recovery process. It is seen that long-term exposure produces an unrecoverable effect. Mohtadi et al. [9] and Uribe et al. [5] contaminated the fuel cell with 5 ppm NO<sup>2</sup> over 12 and 15 h, respectively. In these cases, the result from Mohtadi et al. [9] was partially recovered cell performance, while the result from Uribe et al. [5] was fully recovered performance. The difference of these two recovery processes was that Mohtadi et al. [9] operated the cell in the range of 0.68–0.7 V and Uribe et al. [5] at 0.5 V. Our previous results [4], in agreement with the results of Chen et al. [10] and Lin et al. [11], showed that, at lower potentials in the negative sweep, reduction of nitrite occurs, and thus it is removed from the Pt-catalyst. Higher concentrations were tried by Yang et al. [7] (10, 140, and 1480 ppm) and Misz et al. [6] (10 and 15 ppm), in which performance recovery was almost reached in all of the cases after approximately 1 h with NO2. When it comes to long-term operation, Uribe et al. [5] showed that performance following contamination of 0.4 ppm NO<sup>2</sup> for around 520 h was not recovered, probably due to the low amount of catalyst they used (17 µg Pt cm−<sup>2</sup> ) that was quickly damaged.

St-Pierre et al. [12] simulated performance recovery after 500 h of exposure to 0.1 ppm NO2. Even if they used dry air conditions in their simulation, where the performance was dramatically affected, the performance was recovered and reached its initial value. This result is contradictory to the one obtained by Uribe et al. [5] and may be due to different conditions, but unfortunately the operating conditions used were not specified in Uribe's report.

The aim of this study was to contribute to the improvement of the durability of the fuel cell by trying different operating conditions that influence the recovery process after NO<sup>2</sup> contamination. These processes included successive polarization curves and recovery at different flow rates and current densities. In real traffic situations, exposure to high amounts of NO<sup>2</sup> is unavoidable, and recovery methods that can be applied online in real fuel cell vehicles are desired. Therefore, two such realistic operation techniques were suggested and compared: consecutive recovery with air and air depletion.

#### **2. Results and Discussion**

#### *2.1. Performance of the Contaminated MEA*

The degradation of fuel cell performance upon contamination with 100 ppm of NO<sup>2</sup> in air and its subsequent recovery of performance is shown in Figure 1a. The sequence of experiments was to run the cell in a galvanostatic mode at 0.5 A cm−<sup>2</sup> with clean air for 30 min without contaminant, followed by the introduction of 100 ppm NO<sup>2</sup> in the cathode air flow for 3 h, and then recovery of performance with pure air. Polarization curves (Figure 1b) and electrochemical impedance spectroscopy (EIS) (Figure 1c) were done at the beginning of life (BOL), after contamination with NO2, and after recovery with air. Figure 1a shows the dramatic performance degradation of 197 mV after 3 h of contamination. However, after switching off the NO<sup>2</sup> contaminant and running the cell with clean air, the fuel cell performance was completely recovered in 4.5 h.

The polarization curves in Figure 1b show a clear contamination of NO2, mainly at lower current densities, where the Pt-catalyst active sites are affected by NO2. In this part of the curve, the contamination is related to the electrode kinetics, most likely at the cathode, which is the main contributor to the performance loss and where the contaminant is introduced. In the graph, it is also shown that the performance was completely recovered when pure air was added at 110 mL min−<sup>1</sup> with a current density of 0.5 A cm−<sup>2</sup> . Furthermore, the recovered performance was better than at beginning

of life (BOL) at high current densities, which may indicate better conductivity in the membrane due to water being produced by the ORR, while at the same time intermediate species are being reduced in the actual potential range, as mentioned in our previous publication [4]. *Molecules* **2020**, *25*, 1115 3 of 11

**Figure 1.** (**a**) Performance during contamination with 100 ppm NO2 and recovery at 0.5 A cm<sup>−</sup>2 and air flow 110 mL min<sup>−</sup>1; (**b**) polarization curves; and (**c**) galvanostatic EIS measurements at 0.5 A cm<sup>−</sup><sup>2</sup> for BOL, with NO2, and after recovery in cathode air flow. **Figure 1.** (**a**) Performance during contamination with 100 ppm NO<sup>2</sup> and recovery at 0.5 A cm−<sup>2</sup> and air flow 110 mL min−<sup>1</sup> ; (**b**) polarization curves; and (**c**) galvanostatic EIS measurements at 0.5 A cm−<sup>2</sup> for BOL, with NO<sup>2</sup> , and after recovery in cathode air flow.

The polarization curves in Figure 1b show a clear contamination of NO2, mainly at lower current densities, where the Pt-catalyst active sites are affected by NO2. In this part of the curve, the contamination is related to the electrode kinetics, most likely at the cathode, which is the main contributor to the performance loss and where the contaminant is introduced. In the graph, it is also shown that the performance was completely recovered when pure air was added at 110 mL min−<sup>1</sup> with a current density of 0.5 A cm−2. Furthermore, the recovered performance was better than at beginning of life (BOL) at high current densities, which may indicate better conductivity in the membrane due to water being produced by the ORR, while at the same time intermediate species are being reduced in the actual potential range, as mentioned in our previous publication [4]. To better diagnose the performance limitation after contamination with NO<sup>2</sup> and the respective recovery in the fuel cell, EIS spectra were recorded at 0.5 A cm−<sup>2</sup> and shown in Figure 1c. After contamination with NO2, a second semicircle is beginning to be formed at lower frequencies. However, this second semicircle disappears after the recovery process with pure air. Additionally, the polarization resistance decreases, which is in accordance with the polarization curve in Figure 1b and may be related to a better access to platinum sites after the recovery process. There is no change in the high frequency resistance (HFR), showing that the membrane resistance was not affected by contamination with NO<sup>2</sup> and the recovery process.

To better diagnose the performance limitation after contamination with NO2 and the respective recovery in the fuel cell, EIS spectra were recorded at 0.5 A cm−2 and shown in Figure 1c. After contamination with NO2, a second semicircle is beginning to be formed at lower frequencies. However, this second semicircle disappears after the recovery process with pure air. Additionally, the polarization resistance decreases, which is in accordance with the polarization curve in Figure 1b Based on the degradation and time for performance recovery shown in Figure 1a, different air flow rates, different constant current densities, and successive polarization curves were tested during the recovery of the contaminated MEA to investigate and understand the recovery process of this contaminant.

and may be related to a better access to platinum sites after the recovery process. There is no change

#### *2.2. Recovery at Di*ff*erent Air Flow Rates*

To find a shorter performance recovery time for the MEA contaminated with NO2, different air flow rates (110, 165, 220, and 275 mL min−<sup>1</sup> ) were tested for the recovery process at a constant current density of 0.5 A cm−<sup>2</sup> , as shown in Figure 2a. The time required to reach the same cell voltage as before contamination is defined as the recovery time. The same contamination sequence as described above was used and the air flow rate was changed to the desired value for the recovery of the performance. Figure 2a shows that all the curves reached their initial values after the recovery process, but after different periods of time. The faster recovery time was found to be at the highest flow rate, 275 mL min−<sup>1</sup> . This is a clear sign that the NO<sup>2</sup> contaminant is not as well attached to the Pt-catalyst surface as sulfur compounds are [9]. As soon as clean air is introduced into the recovery process, most of the NO<sup>2</sup> is removed from the Pt-catalyst. This is shown in Figure 2a by the sharp increase in cell voltage (~180 mV) within about 30 min, after which a slower relaxation period occurs until steady state is reached. *Molecules* **2020**, *25*, 1115 5 of 11

**Figure 2.** (**a**) Performance with 100 ppm NO2 and after recovery at different airflows 110, 165, 220, and 275 mL min<sup>−</sup>1 at 0.5 A cm−2; (**b**) polarization curves; and (**c**) galvanostatic EIS measurements at 0.5 **Figure 2.** (**a**) Performance with 100 ppm NO<sup>2</sup> and after recovery at different airflows 110, 165, 220, and 275 mL min−<sup>1</sup> at 0.5 A cm−<sup>2</sup> ; (**b**) polarization curves; and (**c**) galvanostatic EIS measurements at 0.5 A cm−<sup>2</sup> for the same conditions as in (**a**).

Another strategy investigated was to recover the contaminated MEA at different current densities, as shown in Figure 3a. The same contamination procedure was done as in Figure 1a and the different controlled current densities for recovery process were 0.2, 0.5, 0.75, and 1 A cm−2. It can be seen that all of the performance recovery measurements reached a steady state at their respective current densities. As also seen in the experiments with different air flow rates, the voltage increases abruptly after the NO2 is switched off and replaced with clean air, which here again may be related to the rapid removal of NO2 from the Pt-catalyst. The necessary time to reach steady state after the recovery process was different for the different current densities, and decreased as the current density increased above 0.5 A cm−2. Surprisingly, the time to reach steady state at the recovery current density of 0.2 A cm−2 was the shortest. This is a sign of a different mechanism that occurs at this specific current density. From our previous study using cyclic voltammetry in inert media with no water production [4], it was seen that around the range of potential that this current density corresponds to (0.65–0.76 V), reduction of NO3− to NO2− may occur. However, in the present

A cm<sup>−</sup>2 for the same conditions as in (**a**).

*2.3. Recovery at Different Current Densities* 

Figure 2b shows that all polarization curves overlap up to the current density of 0.4 A cm−<sup>2</sup> . As the current density increases further, small differences in potential can be seen between the curves at different flow rates, where the two highest air flow rates show the best performance. The performance after the recovery process is higher than at BOL for all the different air flow rates, in the same way as in Figure 1b.

The EIS spectra after the recovery process at different air flow rates are depicted in Figure 2c. It can be seen that there is no significant difference in the HFR where the spectra intercept the real axis. After the recovery process, all of the spectra show a lower polarization resistance when compared with BOL, which is in accordance with Figure 2b. It might be possible that some Pt-sites were activated after the recovery process with pure air.

#### *2.3. Recovery at Di*ff*erent Current Densities*

Another strategy investigated was to recover the contaminated MEA at different current densities, as shown in Figure 3a. The same contamination procedure was done as in Figure 1a and the different controlled current densities for recovery process were 0.2, 0.5, 0.75, and 1 A cm−<sup>2</sup> . It can be seen that all of the performance recovery measurements reached a steady state at their respective current densities. As also seen in the experiments with different air flow rates, the voltage increases abruptly after the NO<sup>2</sup> is switched off and replaced with clean air, which here again may be related to the rapid removal of NO<sup>2</sup> from the Pt-catalyst. The necessary time to reach steady state after the recovery process was different for the different current densities, and decreased as the current density increased above 0.5 A cm−<sup>2</sup> . Surprisingly, the time to reach steady state at the recovery current density of 0.2 A cm−<sup>2</sup> was the shortest. This is a sign of a different mechanism that occurs at this specific current density. From our previous study using cyclic voltammetry in inert media with no water production [4], it was seen that around the range of potential that this current density corresponds to (0.65–0.76 V), reduction of NO<sup>3</sup> <sup>−</sup> to NO<sup>2</sup> − may occur. However, in the present experiments, water is produced at the cathode side and may react with NO<sup>2</sup> producing HNO<sup>3</sup> and NO, as shown in Equation (1). It can be pointed out that nitric acid in water is normally present as NO<sup>3</sup> − [13].

$$2\text{ NO}\_2 + \text{H}\_2\text{O} \rightarrow 2\text{H}^+ + 2\text{ NO}\_3^- + \text{NO} \tag{1}$$

The range of potentials in which the performance is recovered at the current density of 0.2 A cm−<sup>2</sup> , i.e. 0.65–0.76 V (Figure 3a), is almost the same as the one in the inert media [4]; therefore, NO<sup>3</sup> − may be reduced to NO<sup>2</sup> − in the present experiments as well. Additionally, NO contamination is similar to CO contamination in that both contaminants affect the catalyst layer and, at low current densities in presence of O2, NO is removed from the catalyst. This may explain the faster recovery at lower current densities (0.2 A cm−<sup>2</sup> ), while at higher current densities the NO contamination is more severe and oxygen is predominantly producing water through ORR. This suggests that chemical reactions may be present and followed by electrochemical reaction, in the same way as discussed by Chen et al. [10].

Figure 3b shows the respective polarization curves after the recovery process at different current densities. The performance after the recovery process done at the current densities of 0.5, 0.75, and 1 A cm−<sup>2</sup> overlapped with that at BOL until 0.4 A cm−<sup>2</sup> in the polarization curve. At higher current densities, they still overlapped each other but they differed from the BOL, in a similar way as in Figure 2b. However, the behavior of the performance after the recovery process done at 0.2 A cm−<sup>2</sup> was different. It is seen that this performance was not fully recovered; even though it reached a steady state during the recovery process, as shown in Figure 3a, it still had 15 mV left to full recovery. For this recovery current density, a better performance than at BOL was seen at current densities higher than 0.6 A cm−<sup>2</sup> . A possible reason it did not fully recover at the current density of 0.2 A cm−<sup>2</sup> may be the formation of intermediate species around 0.7 V that may have affected the performance.

NO3− [13].

Figure 3c shows the EIS spectra after the recovery process at different current densities. As in Figure 2c, there is no significant difference in the HFR and the polarization resistance decreases after the recovery process. The lowest polarization resistance was observed for the recovery at 0.5 A cm−<sup>2</sup> . severe and oxygen is predominantly producing water through ORR. This suggests that chemical reactions may be present and followed by electrochemical reaction, in the same way as discussed by Chen et al. [10].

densities in presence of O2, NO is removed from the catalyst. This may explain the faster recovery at lower current densities (0.2 A cm−2), while at higher current densities the NO contamination is more

*Molecules* **2020**, *25*, 1115 6 of 11

experiments, water is produced at the cathode side and may react with NO2 producing HNO3 and NO, as shown in Equation (1). It can be pointed out that nitric acid in water is normally present as

The range of potentials in which the performance is recovered at the current density of 0.2 A cm−2, i.e. 0.65–0.76 V (Figure 3a), is almost the same as the one in the inert media [4]; therefore, NO3− may be reduced to NO2− in the present experiments as well. Additionally, NO contamination is

3 NO2 + H2O 2H+ + 2 NO3− + NO (1)

**Figure 3.** (**a**) Performance with 100 ppm NO2 and recovery at current densities 0.2, 0.5, 0.75, and 1 A cm<sup>−</sup>2 and constant 110 mL min−1 air flow; (**b**) polarization curves; and (**c**) galvanostatic EIS measurements at 0.5 A cm<sup>−</sup>2 at BOL, after contamination, and after recovery at each current density. **Figure 3.** (**a**) Performance with 100 ppm NO<sup>2</sup> and recovery at current densities 0.2, 0.5, 0.75, and 1 A cm−<sup>2</sup> and constant 110 mL min−<sup>1</sup> air flow; (**b**) polarization curves; and (**c**) galvanostatic EIS measurements at 0.5 A cm−<sup>2</sup> at BOL, after contamination, and after recovery at each current density.

#### *2.4. Other Types of Recovery*

Thus far, it has been shown that the recovery process time after contamination with NO<sup>2</sup> can be shortened. A summary of times for recovery, from the used recovery methods, is shown in Table 1. The two shortest times were found to be at current density of 0.2 A cm−<sup>2</sup> and air flow of 275 mL min−<sup>1</sup> . Therefore, these two operating conditions were combined to potentially obtain an even shorter recovery time (Figure 4a). Additionally, successive polarization curves after contamination with NO<sup>2</sup> were tried as a recovery method. For this experiment, the polarization curves were conducted in galvanodynamic mode at a step rate of 5 mA s−<sup>1</sup> . Figure 4b shows the polarization curves for the latter two recovery methods compared with the polarization curves at 0.5 A cm−<sup>2</sup> and 110 mL min−<sup>1</sup> , at BOL, and directly after contamination with 100 ppm NO2. The recovery time for the 0.2 A cm−<sup>2</sup> and 275 mL min−<sup>1</sup>

air flow was 66 min, i.e. the time was reduced by 10 min compared with the recovery process at 0.2 A cm−<sup>2</sup> and 110 mL min−<sup>1</sup> . This indicates that the airflow rate is an important parameter; it seems that NO<sup>2</sup> can be removed from the Pt-catalyst by the air, and/or that O<sup>2</sup> is participating in chemical and electrochemical reactions in the removal of NO<sup>2</sup> species, as mentioned in the Section 2.3. The recovery time when performing successive polarization curves was 35 min, which was found to be the fastest *Molecules* way to recover the performance of the fuel cell contaminated with NO2. **2020**, *25*, 1115 8 of 11

**Figure 4.** (**a**) Performance with 100 ppm NO2, recovery at 0.5 A cm<sup>−</sup>2 and 110 ml min−1, and recovery at 0.2 A cm<sup>−</sup>2 and 275 ml min−1; (**b**) Polarization curves at BOL, after contamination with NO2, after recovery at 0.5 A cm<sup>−</sup>2 and 110 mL min−1, after recovery by successive polarization curves with 110 mL min<sup>−</sup>1 as a constant air flow, and after recovery at 0.2 A cm−2 and 275 mL min−1; and (**c**) EIS spectra for the measurements done in (**b**). **Figure 4.** (**a**) Performance with 100 ppm NO<sup>2</sup> , recovery at 0.5 A cm−<sup>2</sup> and 110 ml min−<sup>1</sup> , and recovery at 0.2 A cm−<sup>2</sup> and 275 ml min−<sup>1</sup> ; (**b**) Polarization curves at BOL, after contamination with NO<sup>2</sup> , after recovery at 0.5 A cm−<sup>2</sup> and 110 mL min−<sup>1</sup> , after recovery by successive polarization curves with 110 mL min−<sup>1</sup> as a constant air flow, and after recovery at 0.2 A cm−<sup>2</sup> and 275 mL min−<sup>1</sup> ; and (**c**) EIS spectra for the measurements done in (**b**).

The EIS measurements in Figure 4c show that, even though the shortest recovery time was reached by the successive polarization curves, the spectra of the experiment at 0.5 A cm−2 and 110 mL min−1 together with the spectra at 0.2 A cm−2 and 275 mL min−1 were those that had the lowest polarization resistance. It is worth mentioning that none of the polarization curves after contamination with 100 ppm NO<sup>2</sup> reached values around 0.2 V. The lowest potential (0.35 V) was reached at a current density of 1 A cm−<sup>2</sup> . Therefore, the reduction of NO<sup>2</sup> to N2O and/or NH2OH [4] may not be present in these set of experiments.

50 ppm of NO2 in different ways, as shown in Figure 5a, with the goal to suggest online application in a fuel cell car. This concentration was chosen because it is more probable to find 50 ppm NO2 in air than 100 ppm or higher concentrations. In both experiments, the cell was first stable for 30 min, keeping the same potential. The experiment done with air recovery (blue line) consisted of introducing NO2 with balance of air to the cathode for 20 min, and then recovering the performance with clean air for 2 h. The same sequence was repeated three times. In the experiment with air

*2.5. Comparison of Two Operation Techniques* 


**Table 1.** Summary of the performance recovery time after introduction of 100 ppm NO<sup>2</sup> to the cathode air flow. Pure H<sup>2</sup> was used at the anode.

A theoretical prediction for the recovery of NO<sup>2</sup> was made by St-Pierre et al. [12]; however, in their prediction, they did not include all processes in the fuel cell that may be affected by degradation, such as ohmic losses and mass transport, which explains the results obtained. It would be interesting to investigate performance recovery in a wider current density range.

The EIS measurements in Figure 4c show that, even though the shortest recovery time was reached by the successive polarization curves, the spectra of the experiment at 0.5 A cm−<sup>2</sup> and 110 mL min−<sup>1</sup> together with the spectra at 0.2 A cm−<sup>2</sup> and 275 mL min−<sup>1</sup> were those that had the lowest polarization resistance.

#### *2.5. Comparison of Two Operation Techniques*

Finally, two operation techniques for the cathode were applied and compared by introducing 50 ppm of NO<sup>2</sup> in different ways, as shown in Figure 5a, with the goal to suggest online application in a fuel cell car. This concentration was chosen because it is more probable to find 50 ppm NO<sup>2</sup> in air than 100 ppm or higher concentrations. In both experiments, the cell was first stable for 30 min, keeping the same potential. The experiment done with air recovery (blue line) consisted of introducing NO<sup>2</sup> with balance of air to the cathode for 20 min, and then recovering the performance with clean air for 2 h. The same sequence was repeated three times. In the experiment with air depletion (orange line), the NO<sup>2</sup> contaminated air was fed to the cathode during 20 min, after which the air gas flow was switched off until the potential reached 0.01 V. At that point, the gas was switched on again. This experiment was made with the purpose to sweep the cell voltage within a wide range in order to let the fuel cell to recover quickly. The procedure was repeated 21 times to be comparable in time with the air recovery technique. Figure 5a shows a complete reversibility during the air recovery technique, in which all cycles reached the initial value (0.7 V). In both techniques, a lower cell voltage is seen after the 20 min with NO<sup>2</sup> compared to the first contamination cycle, but no significant difference is shown between the cycles. The outcome of the air recovery is in accordance with the results of Mohtadi et al. [9], who obtained a complete recovery after three cycles with 5 ppm of NO2. On the other hand, the cell performance obtained by Yang et al. [7] did not reach the initial value after recovery. However, they used a different pressure (0.5 bar), and it is known that the pressure is an important parameter concerning recovery of a fuel cell contaminated by NO<sup>2</sup> [6].

At the end of each experiment, a polarization curve was recorded (see Figure 5b). The figure shows that the strategy with air depletion resulted in a lower performance after 7 h of operation, which might be caused by deterioration of the electrode due to peroxide formation at low electrode potentials [14–16]. On the other, the polarization curve after air recovery revealed a complete recovery of the Pt-catalyst, and even better performance at current densities higher than 0.4 A cm−<sup>2</sup> . The air recovery technique suggests that NO<sup>2</sup> is only attached to the Pt-catalyst of the electrode and that it can be easily removed by air, apparently, without affecting other components.

EIS was also conducted at the end of each experiment (Figure 5c). The figure shows no significant difference between the two strategies, although the HFR of the air depletion spectrum increased only corresponding to about 3 mV when compared to beginning of life, but this is in the range of error.

These results show that it is possible to operate a specific technique online in a fuel cell vehicle in order to deal with NO<sup>2</sup> air pollution. However, the technique must be adapted to a more realistic drive cycle. Operating parameters such as air flow rate and current density can also possibly be incorporated in a recovery method to keep good performance after NO<sup>2</sup> contamination.

*Molecules* **2020**, *25*, 1115 9 of 11

depletion (orange line), the NO2 contaminated air was fed to the cathode during 20 min, after which the air gas flow was switched off until the potential reached 0.01 V. At that point, the gas was switched on again. This experiment was made with the purpose to sweep the cell voltage within a wide range in order to let the fuel cell to recover quickly. The procedure was repeated 21 times to be comparable in time with the air recovery technique. Figure 5a shows a complete reversibility during the air recovery technique, in which all cycles reached the initial value (0.7 V). In both techniques, a

**Figure 5.** (**a**) Transient cell voltage when introducing 50 ppm NO2 to the cathode air flow at 0.5 A cm<sup>−</sup>2 for two different strategies, namely air recovery (blue) and air depletion (orange); (**b**) polarization curves; and (**c**) galvanostatic EIS measurement at 0.5 A cm<sup>−</sup>2 at BOL and after testing the **Figure 5.** (**a**) Transient cell voltage when introducing 50 ppm NO<sup>2</sup> to the cathode air flow at 0.5 A cm−<sup>2</sup> for two different strategies, namely air recovery (blue) and air depletion (orange); (**b**) polarization curves; and (**c**) galvanostatic EIS measurement at 0.5 A cm−<sup>2</sup> at BOL and after testing the two strategies.

#### two strategies. **3. Materials and Methods**

At the end of each experiment, a polarization curve was recorded (see Figure 5b). The figure shows that the strategy with air depletion resulted in a lower performance after 7 h of operation, which might be caused by deterioration of the electrode due to peroxide formation at low electrode The experimental set up used in this investigation was the same as used in our previous study with NO<sup>2</sup> [4]. A commercial fuel cell hardware from Fuel Cell Technologies, Inc., and a commercial membrane electrode assembly (MEA) (Gore™Primea® 5641), with catalyst loadings of 0.45 mg cm−<sup>2</sup> Pt-alloy on the anode and 0.4 mg cm-2 Pt on the cathode, were used in all of the experiments. The same type of gas diffusion layer (GDL) (Carbel™) was used at both anode and cathode. The geometric electrode area used was 1.5 cm<sup>2</sup> . The cell temperature was kept at 80 ◦C and 1 atm, and the humidification of the gases was 90% RH. The gas cylinder used was the same as in [4], and the contamination flow was controlled by an Alicat Scientific mass flowmeter.

The electrochemical characterization procedure was the same as in our previous study [4]. For the contamination step, a galvanostatic measurement was done, followed by polarization curve measurement and electrochemical impedance spectroscopy (EIS) by use of a Solartron Interface SI1287 potentiostat together with a 1255 frequency response analyzer, controlled by CorrWare software. For the EIS, an AC amplitude of 60 mA (roughly corresponding to 3–15 mV depending on frequency and operating conditions) was used in the frequency range between 10 kHz and 30 mHz. It was assumed, in all experiments, that the electrical bulk and contact resistances were not affected by the introduction of NO2, and that the high frequency resistance is related to the resistance of the membrane.

#### **4. Conclusions**

The results show that it is possible to find adequate performance recovery methods that can be applied in a fuel cell car in a real traffic situation where large amounts of NO<sup>2</sup> are present. In the experiments done in galvanostatic mode at 0.5 A cm−<sup>2</sup> with air flow of 110 mL min−<sup>1</sup> , a significant potential drop was observed due to the presence of NO<sup>2</sup> in the cathode air. This performance loss was however totally recovered after 4.5 h with clean air. The study shows that it is possible to significantly decrease the time for performance recovery by running successive polarization curves or by applying 0.2 A cm−<sup>2</sup> and an air flow of 275 mL min−<sup>1</sup> . Two operation techniques that can be used online in a fuel cell vehicle were also tested: air recovery and air depletion. The air recovery technique was found to be the best option for recovery of performance. Therefore, we assume that air can pull out the NO<sup>2</sup> molecules that surround the Pt-catalyst to free up the active site at higher current densities; however, at the current density of 0.2 A cm−<sup>2</sup> , possibly a different contamination mechanism occurs.

**Author Contributions:** Conceptualization, Y.A.G.; methodology, Y.A.G.; validation, Y.A.G. and C.L.; formal analysis, Y.A.G.; investigation, Y.A.G.; resources, G.L. and C.L.; writing—original draft preparation, Y.A.G.; writing—review and editing, Y.A.G., C.L., and G.L.; visualization, Y.A.G. and C.L.; supervision, C.L. and G.L.; project administration, Y.A.G. and C.L.; and funding acquisition, C.L. and G.L. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the European project BIOGAS2PEM-FC (FP7), grant number 314940 and the Swedish governmental initiative StandUp for Energy.

**Acknowledgments:** The materials for this work were provided by Powercell AB. The experimental set up was built with the help of Mr. Hongkuan Wang.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**


**Sample Availability:** Samples of the tested MEAS are available from the authors.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Automotive Subzero Cold-Start Quasi-Adiabatic Proton Exchange Membrane Fuel Cell Fixture: Design and Validation**

#### **Antonio O. Pistono and Cynthia A. Rice \***

Department of Chemical Engineering, Tennessee Technological University, Cookeville, TN 38505, USA; aopistono21@gmail.com

**\*** Correspondence: crice@tntech.edu

Academic Editors: Jean St-Pierre and Shangfeng Du Received: 30 January 2020; Accepted: 17 March 2020; Published: 19 March 2020

**Abstract:** Subzero automotive cold-starts of proton exchange membrane fuel cell (PEMFC) stacks require accelerated thermal rises to achieve nominal operating conditions and close-to-instantaneous usable output power. Advances in the material, structure and operational dependence on the balance between the maximum power output and the electrochemical conversion of hydrogen and oxygen into water requires validation with subzero cold-starts. Herein are presented the design and validation of a quasi-adiabatic PEMFC to enable single-cell evaluation, which would provide a more cost-effective option than stack-level testing. At –20 ◦C, the operational dependence of the preconditioned water content (3.2 verse 6.2) for a galvanic cold-start (<600 mA cm−<sup>2</sup> ) was counter to that of a laboratory-scale isothermal water fill test (10 mA cm−<sup>2</sup> ). The higher water content resulted in a faster startup to appreciable power output within 0.39 min versus 0.65 min. The water storage capacity, as determined from the isothermal water fill test, was greater, for the lower initial water content of 3.2, than 6.2, 17.4 ± 0.3 mg versus 12.8 ± 0.4 mg, respectively. Potentiostatic cold-starts produced usable power in 0.09 min. The versatility and reproducibility of the single cell quasi-adiabatic fixture avail it to future universal cold-start stack relevant analyzes involving operational parameters and advanced materials, including: applied load, preconditioning, interchanging flow field structures, diffusion media, and catalyst coated membranes.

**Keywords:** proton exchange membrane fuel cells; subzero cold-starts; automotive; isothermal water fill tests

#### **1. Introduction**

Automotive proton exchange membrane fuel cell (PEMFC) stacks are required to withstand the same environmental extremes, including subzero temperatures, as the internal combustion engine. The U.S. Department of Energy's 2020 automotive PEMFC requirements are survivability from −40 ◦C and cold-starts from −30 ◦C. A PEMFC subzero cold-start is defined as the initiation of PEMFC operation to meet the required nominal operating temperature and power. As of 2015, the −20 ◦C cold-start target of 0.5 min to 50% rated power has only been met for a PEMFC stack when using one-and-a-half times the targeted parasitic shutdown/start-up energy [1]. A PEMFC stack is comprised of as many as 400 non-reactive repeat units (flow fields, coolant channels, and current collectors), each encasing the membrane electrode assembly (MEA) component. These non-reactive components behave as thermal sinks, scavenging generated heat during cold-starts. Presently, a common energy-intensive strategy is to use resistive heating to cold-start a PEMFC stack [2]. At nominal operating temperatures, a fine balance is maintained between the rate of water production and evaporative removal from the PEMFC stack. The process of subzero cold-starting of a PEMFC stack is challenging in that a balance

must be attained between the rate of heat generation and product-water redistribution. At subzero temperatures, product-water accumulation and ice formation result in mass transport losses that can lead to failure if the PEMFC stack does not self-heat to above 0 ◦C before oxygen is completely blocked from the accessible reaction sites. As cathode catalysts and catalyst layers advance and become more efficient to meet cost targets, more parasitic power is required to self-heat non-reactive components and in addition to increased material mechanical performance/durability issues due to subzero operation. Therefore, optimization and material validation require a single-cell rapid testing platform.

Subscale single PEMFC cold-starts in standard laboratory fixtures are limited by the thermal mass of the endplates. The testing of short-stacks, of 20–30 repeat units, is cost prohibitive and strongly influenced by performance losses since the endplates behave as thermal sinks. According to the literature, the dominant subzero PEMFC studies investigating the influence of material and operational parameters have been restricted to subscale single-cell freeze-thaw testing [3–14], isothermal water fill tests [15–37], and non-isothermal water fill tests [3,21,33,38–47]. United Technology Corporation (UTC) Power and its corporate research facility co-developed two quasi-adiabatic single PEMFC fixtures in the mid-2000s with geometric active areas of 25 cm<sup>2</sup> and 320 cm<sup>2</sup> [48–50]. The results demonstrated that it was possible under a galvanic load cold-start to replicate the center cells in a stack's voltage and thermal profile in the quasi-adiabatic fixture. Balliet and Newman validated their two-dimensional liquid water transport cold-start model to the UTC-Power's quasi-adiabatic PEMFC fixture performance profiles [49]. However, the inadequate structural integrity of the quasi-adiabatic PEMFC fixture limited its reusability. The present co-author, cited in references [48,50], changed her surname after these studies were published. Published stack results are minimal and limited mostly to modeling [51–56].

Subzero PEMFC cold-starts are possible due to non-frozen water found in the membrane and catalyst layers of the MEA that support proton conduction (σ*H*<sup>+</sup> ) and rapid exothermic heat generation [57,58]. Ohmic heat generation contributes to cold-start performance and is dependent on the initial water concentration. A portion of water in the ionic domains of the membrane and catalyst layers remains non-frozen at subzero temperatures due to colligative and supercooling effects [59]. Liquid water is retained within the catalyst//ionomer aggregate interfaces due to attractive forces of the charged ionomer end-chain sites (−*SO*<sup>−</sup> 3 ), allowing interconnected transport through the agglomerates between the membrane and catalyst layers. The vapor-saturated water content (λ) at the aggregate interfaces is typically less than 14 and reaches a maximum of 22 for liquid saturated [54,60]. The hydrogen fuel supplied to the anode catalyst layer is electrooxidized to protons and electrons, *<sup>H</sup>*<sup>2</sup> <sup>→</sup> <sup>2</sup>*H*<sup>+</sup> + <sup>2</sup>*<sup>e</sup>* − . The protons are transported through the hydrated ionic domains from the anode layer through the membrane to the cathode layer. In the presence of supplied oxygen, protons and electrons recombining within the cathode catalyst layer to form product water and heat, *O*<sup>2</sup> + 4*H*<sup>+</sup> + 4*e* <sup>−</sup> → 2*H*2*O*.

Herein, a single cell quasi-adiabatic PEMFC fixture was designed to be structurally engineered for reproducible subscale cold-starts. Subzero isothermal water fill tests are shown to inadequately advance the understanding of the operational impact of the initial water content (λ*initial*) on automotive stack relevant cold-starts. At <sup>−</sup><sup>20</sup> ◦C, isothermal water fill tests under 10 mA cm−<sup>2</sup> applied loads were compared to cold-starts with loads set to 600 mA cm−<sup>2</sup> .

#### **2. Results and Discussion**

#### *2.1. Quasi-Adiabatic PEMFC Fixture Design*

The design constraints required for a single-cell subzero cold-start PEMFC fixture are (i) thermal isolation of the flow fields and MEA from the endplates, (ii) impervious humidified gas manifolds, (iii) structural uniformity of the active area under axial load, and (iv) high electrical conductivity between the anode and cathode sides of the PEMFC through an external circuit. Material compatibility issues of the multi-layered testing PEMFC fixture are exacerbated due to frequent thermal cycling between subzero temperatures and up to 80 ◦C. Figure 1 shows one side of the symmetric hardware

(flow fields, heater, gas and coolant manifold, insulation, and end plates) used in the quasi-adiabatic fixture. The geometric active surface area of the quasi-adiabatic fixture was scaled down from the prototypic cell size found in automotive stacks (16 versus 320 cm<sup>2</sup> active area) to simplify development and conservatively minimize any expected issues with flow field deflection; however, it is expected the active area and other geometries are scalable to develop an optimized fixture.

**Figure 1.** One half of symmetric quasi-adiabatic single-cell hardware. The parallel coolant channels are machined into the backside of the graphite flow field plate.

#### 2.1.1. Humidified Gas Manifolds

The nature of material properties makes porous structures more thermally insulated, thus satisfying constraint (i) above, but failing constraint (ii) that requires humidified gas containment. Therefore, two distinct layers were required to insulate and distribute humidified gasses to and from both sides of the MEA. Several ridged, high-density plastics were considered for the humidified gas manifold, see Table 1. All the plastics presented in Table 1 have acceptable high densities (>1200 kg m−<sup>3</sup> ) for excluding H<sup>2</sup> leakage and sufficiently high temperature limits (>121 ◦C) for thermal stability within the operating range of a PEMFC (−40 ◦C ↔ 90 ◦C). The manifold material must be machinable and not brittle for threaded fittings to connect humidified inlet and outlets. The ideal plastic would have a high compressive strength and compressive modulus to maintain uniformed axial load across the PEMFC fixture during thermal cycling and subsequent rebuilds (constraint (iii)), while having low thermal conductivity (constraint (i)) to retain heat generated by the MEA during PEMFC operation. An additional requirement of constraint (ii) is that the manifold material would not allow water adsorption, as it would freeze, fracture the material, and cause structural failure below 0 ◦C. Table 1 highlights the maximum and minimum material property values for the high-density plastics considered, underlined and underlined in shaded gray box, respectively. Only materials with low water adsorption (<1%) were considered to ensure the structure integrity of the fixture. UTC-Power adiabatic fixtures used high-density polyamide-imide (Pyropel-HD) as the internal gas manifold material with the maximum compressive modulus of all the plastics from Table 1. However, due to cost and embrittlement issues that made the inlet and outlet connector junctions prone to stress breakage, it was not selected. The manifold material was selected from the remaining plastics in Table 1 by optimizing the lowest range for both thermal conductivity and water uptake. Polyvinylidene fluoride was not selected although it had the lowest water uptake (0% of lower limit in range), as the thermal conductivity was on the higher end (47.8% of lower limit in range). Polycarbonate (PC 1000)

was selected as the internal gas manifold material for the fixture as it had the lowest combination of thermal conductivity (11.0% of lower limit in range) and water uptake (8.9% of lower limit in range).

**Table 1.** High-density polymers considered for the gas manifold of the quasi-adiabatic PEMFC fixture. The highest and lowest material properties values of each type are underlined and the highest is further identified by a shaded gray box. Not available (N/A).


#### 2.1.2. Insulation

Several porous semi-ridged materials were considered to promote retention of the heat generated by the MEA, as listed in Table 2. The density and thermal conductivity of the materials were an order of magnitude lower than that of the humidified gas manifold, thus improving the insulating properties of the material. All the materials presented in Table 2 have acceptably low densities (< 480 kg m−<sup>3</sup> ) that allow them to be insulative and sufficiently high temperature limits (> 149 ◦C) that enhance thermal stability within the range of PEMFC operation (−40 ◦C ↔ 90 ◦C). However, the decreased density made the material more porous and susceptible to entrainment of condensed water from the interior of the environmental chamber. The first three materials listed (polyisocyanurate, cellular glass, and calcium silicate) are common insulators with low compressive strength, ranging from 0.2 ↔ 0.7 MPa, and low thermal conductivity, averaging around 0.043 <sup>±</sup> 0.022 W m−<sup>1</sup> <sup>K</sup> −1 . However, their open structure leads to high water uptake that would result in structural deformation nonuniformly altering the axial load across the active cell area, thus negatively impacting sealing of the manifold and electric continuity. The subsequent materials were of two distinctly different types: synthetic nonwoven fibrous polyimide and naturally grown balsa wood. Balsa wood is a common low-cost, low-weight construction material used for applications such as aircraft construction. In each of these families of materials, the compressive strength increases with density, and the thermal conductivity adversely increases as well. The natural strength of balsa woods is due to the multilayering of primary and secondary walls, forming a randomly distributed fiber-reinforced composite that resists out-of-plane deformation [61].


**Table 2.** Semi-ridged materials considered as insulation of the quasi-adiabatic PEMFC fixture. The highest and lowest material properties values of each type are underlined and the highest is further identified by a shaded gray box. Not available (N/A).

The percentage of water uptake was evaluated for both Pyropel and balsa wood by immersion in water. Immersion was a severe scenario as the only source of water that could contact the insulating material would be condensed water from the environmental chamber. Water uptake was found to be 245% over the dry weight for Pyropel MD-18, Figure 2. The as-received balsa wood water uptake was 50%, compared to the initial mass. Sealing the wood surface with a thin coating of polyurethane reduced the water uptake to only 14%. Coated balsa wood was selected as the insulating material due to the combination of relatively high compressive strength, low water uptake, and low thermal conductivity.

**Figure 2.** Water uptake of down-selected insulation materials for 15-min water immersion.

#### 2.1.3. Uniformity of Applied Axial Load

The peripheral manifolds and insulation parts of the assembly surrounding the MEA must apply uniform axial load across the faces of the flow fields to retain gasses and ensure electrical continuity between the catalyst layers and the current collectors. To compare the quasi-adiabatic fixture's compression uniformity across the MEA with the standard fixture used in previous studies (Figure 3), pressure paper was used instead of the membrane and catalyst layers. Note that the vertical line in Figure 3a is an artifact from the pressure paper and should be disregarded. The standard PEMFC fixture had a flow field geometry identical to that of the quasi-adiabatic PEMFC fixture. The compression of the quasi-adiabatic fixture (Figure 3a) was mostly uniform across the face of the flow fields, but intensity was less than that of the standard fixture (Figure 3b). The torque was not increased above 70 in·lbs due to concerns with deflection of the endplates compromising the contact in the center of the MEA.

**Figure 3.** Compression paper for contact intensity and uniformity between the 16 cm<sup>2</sup> flow fields (**a**) quasi-adiabatic fixture and (**b**) standard fixture. Each fixture was torqued to 70 in·lbs.

#### *2.2. PEMFC Testing*

Use of both the quasi-adiabatic fixture and a standard fixture allowed for characterization of the MEA at normal operating conditions (both fixtures), cold-starts from −20 ◦C, and isothermal water fill tests at −20 ◦C.

#### 2.2.1. Operating Performance

In Figure 4, the conditioned PEMFC H2/Air polarization curves of the quasi-adiabatic and a standard PEMFC fixture for the same MEA, black and gray, respectively, are compared. The 76 mΩ cm<sup>2</sup> additional resistance of the MEA in the quasi-adiabatic fixture (Figure 4a) is related to the compression issues shown in Figure 3. The cell voltages are corrected using current interrupt resistance (iR-corrected) to compensate for electronic resistance losses due to the reduced axial loading of the quasi-adiabatic PEMFC fixture. The iR-corrected polarization profiles are nearly identical in the ohmic and mass transfer regions. The resistance of the quasi-adiabatic PEMFC fixture is 45% greater than that of the standard fixture.

**Figure 4.** H<sup>2</sup> /21% O<sup>2</sup> polarization curves at 80 ◦C cell temperature and 44.3% relative humidity under ambient pressure for the quasi-adiabatic fixture and a standard fixture (**a**) cell voltage and (**b**) compensated iR free cell voltage.

#### 2.2.2. Water Fill Tests

There are five stages to a low-applied load water fill test: (i) initial supply of reactant gasses elevating the cell voltage, (ii) hydration of the interconnected ionomer domains of the MEA once the initial load is applied, (iii) maintenance of quasi-steady-state cell voltage during the attainment of maximum ionomer hydration, (iv) filling of the large non-hydrophilic pores of the cathode catalyst layer reducing O<sup>2</sup> mass transport, and (v) freeze-out due to ice blockage in the cathode catalyst layer [18]. During these water fill tests, the product water was restricted from entering the diffusion media under the applied test conditions of low water-production rate, low-heat generation rate, and minimal flow rates (no convective transport of water). Under an applied load of 10 mA cm−<sup>2</sup> , the water storage capacity was evaluated for two different pre-conditioned λ*initial*(3.2 and 6.2) at −20 ◦C. The λ*initial* was selected to match multiple published water fill tests—the lower setpoint (λ*initial* = 3.2) represents a dry PEMFC scenario, while the higher setpoint (λ*initial* = 6.2) is closer to an operating PEMFC. Both λ*initial* condition profiles had an initial jump in voltage although the high λ*initial* resulted in a higher initial cell voltage (Figure 5a) due to a lower initial resistance (Figure 5b). The maximum cell voltage was reached at similar times (>1 min) and values (0.81 V). The run with the lower λ*initial* stayed in the quasi-steady-state cell voltage stage >5 min longer due to a high ionomer fill capacity caused by starting at a lower ionomer hydration level and water movement from the ionomer caused by resistive heating [15]. The freeze-out stage was identical for both pre-conditioning hydration levels. The water storage capacity, calculated using Faraday's Law, of the higher λ*initial* of 6.2 preconditioned water fill tests was only 12.8 ± 0.4 mg while that of the λ*initial* runs of 3.2 was 17.4 ± 0.3 mg.

**Figure 5.** Preconditioned initial water content runs (3.2 and 6.2) isothermal water fill test at 10 mA cm−<sup>2</sup> , −20 ◦C and H<sup>2</sup> /21% O<sup>2</sup> (0.05/0.1 lpm, respectively). (**a**) Cell voltage and (**b**) cell resistance versus time on load.

#### 2.2.3. Cold-Starts

The impact of adjacent cell heating, λ*initial*, and galvanic versus potentiostat applied loads were investigated on <sup>−</sup><sup>20</sup> ◦C cold-starts. The galvanically applied load of 600 mA cm−<sup>2</sup> was selected to match the work published by Balliet and Newman on UTRC's quasi-adiabatic PEMFC fixture [49]. The applied galvanic load establishes the PEMFCs maximum attainable current density with the cell voltage approaching 0 V. In a PEMFC stack, it is common for the end cells to reach negative voltages during the first few seconds of a subzero cold-start because the overall voltage of the stack is positive. Table 3 summarizes the cold-start conditions investigated within this study: λ*initial*, adjacent cell heat-adjustment factor, and applied load. The heat-adjustment factor was included to supplement heat that would be provided from neighboring cells in a stack [49,51], as well as thermal losses in the quasi-adiabatic fixture. The heat comes from resistive heating pads located adjacent to the coolant loops of the flow fields enshrouded by the balsa wood. Each type of cold-start was preformed twice to ensure reproducibility. Representative cold-start profiles are shown in Figure 6.

Figure 6a,b compare the power density and cathode flow channel temperature versus time, respectively, for the four types of cold-start presented herein. The power densities were initially low for all of the galvanic cold-starts while the potentiostatic start instantaneously had power because the cell voltage was maintained above 0 V. For all cold-start conditions reported herein, the current density increased with time on load at subzero temperatures until the set point of 600 mA cm−<sup>2</sup> could be supported by the cell voltage. Once the current density exceeded the set point, it was adjusted back down by the Scribner fuel cell software, allowing the cell voltage to rise to higher values. The current density improves as the temperature of the MEA increases, due to increased reaction kinetics and proton conduction through the ionomer. To correlate the subzero dependent current density and cell voltage response during a cold-start, the time scale origin was positioned such that it corresponded with the time the cell temperature reached 0 ◦C, as shown in Figure 6c. The cell-resistance profiles, proportional to proton conduction, were similar for all the cold-starts (Figure 6d) due to similar ionomer water contents, with the exception of the lower adjacent cell heating adjustment factor of 1×.

**Figure 6.** Preconditioned initial water content runs (3.2 and 6.2), cold-start tests under set galvanic load of 600 mA cm−<sup>2</sup> , or applied potentiostatic hold of 0.1 V at −20 ◦C and H<sup>2</sup> /21% O<sup>2</sup> (0.5/0.75 lpm, respectively). (**a**) Power density and (**b**) cathode channel temperature versus time on load. (**c**) Current density and (**d**) cell resistance vs. time on load from 0 ◦C.

#### 2.2.4. Heat Adjustment Factor

In Figure 6, doubling the predicted adjacent cell heating adjustment factor (1× → 2×) significantly impacted the cold-start response profile for an λ*initial* of 3.2 is shown. The output of the heating pads (Q in Watts) was scaled by a multiplier of either 1 or 2 to equation 1 to compensate for the load-dependent fraction of heat that would be lost under applied load (current (i) in Amps) times the overpotential calculated from the difference between the thermoneutral voltage (1.48 V) and the cell voltage (V in volts) [49].

$$Q = Ai(1.48 - V) \tag{1}$$

where A is the geometric surface area of the PEMFC.

The resulting key cold-start performance metrics are summarized in Table 3 and include the average of both runs with standard deviation. The initial applied current density increased by a factor of 3.75 for the higher heating adjustment factor. The non-zero rise in voltage for the 1× heating adjustment factor occurred at a cathode flow field channel temperature around −3.1 ◦C, while for 2×, the transition was near 4 ◦C. The thermocouple point of contact is unknown within the cathode flow field channel and most likely a combination of the solid flow field temperature and the exterior of the cathode diffusion media. The mass of the flow field channels acts as a heat sink, reducing the internal temperature of the cathode flow channel, and hence, yielding the negative non-zero temperature transition for the lower heating adjustment factor. The increased heat adjustment factor suggests that a lower non-reactive thermal mass and lower thermal mass would yield a more successful cold-start. The sluggish heating profile of the lower heating adjustment factor increased the required cold-start time until usable power was available. After 1 min into the cold-start, the lower heating adjustment factor (1×) power density output was only 35% compared to the 2× cold-start (Figure 6a).


**Table 3.** Cold-start parameters used in tests from Figure 6 and corresponding select performance metrics.

Other heating adjustment factors could be used to match other cell designs and materials. The heating pads output is an independent variable enabling the simulation of inner stack cells (symmetric heating case) or end cells (asymmetric heating case). Anomalous cells, arising from partially blocked coolant/flow field channels or degraded materials, can also give rise to other asymmetric heating cases for contiguous cells. Validation of heating adjustment factors with stack data requires significant resources and is not trivial because in-situ heating fluxes for all cells in a stack due to heterogeneous components and locally variable heat fluxes. Even if heating adjustment factors are empirically matched using single cell and stack data, a significant amount of work and resources are still required. For these reasons, validation of the quasi-adiabatic fixture with stack data was deemed outside the scope of this report.

#### 2.2.5. Initial Water Content

Increasing the λ*initial* from 3.2 to 6.2, using a heating adjustment factor of 2×, improved the cold-start performance; however, this result was counter to the two-thirds higher isothermal water storage capacity results (Figure 5) for the lower λ*initial* 3.2. For the higher λ*initial* found upon initially applying the load, the cell voltage could sustain nearly double current density because of the more optimal distribution of interconnected non-frozen water domains within the ionomer previously quantified with subzero electrochemical impedance spectroscopy in Dr. Rice's lab [15]. The measured cold-start time until the voltage increased on average was reduced from 0.65 min to 0.39 min for the higher λ*initial*, translating to appreciable power densities sooner.

#### 2.2.6. Applied Load

The type of applied load controlled the onset of appreciable power densities during the initial phase of the cold-start. Jiang and Wang demonstrated that potentiostatic cold-starts maximized the heat output [47]. The potentiostatic hold of 0.1 V multiplied the current density to get instantaneous power densities. The initial current density was 1.8× greater than that of the galvanic applied load under identical conditions.

#### **3. Materials and Methods**

#### *3.1. PEMFC Assembly*

Tests were performed in two different symmetric subscale PEMFC fixtures with an active area of 16 cm<sup>2</sup> . The proton-conducting membrane used was Nafion HP (Ion Power, New Castle, DE, USA). The anode and cathode catalyst layers were directly sprayed (Badger Airbrush 150) onto the membrane with a final loading of ~ 0.4 mgPt cm−<sup>2</sup> (46.6% Pt on high surface area carbon, Tanaka, Chiyoda-ku, Tokyo) and 30 wt% Nafion (1100EW, Ion Power). The microporous side of the hydrophobic gas diffusion layers (SGL25BC, Ion Power) were positioned adjacent to the catalyst layers. The symmetrically sandwiched gas diffusion layers and catalyst layers around the membrane comprise the membrane electrode assembly (MEA). Polytetrafluoroethylene films (Interplast) sealed the perimeter of the compressed gas diffusion layers against the flow fields.

Figure 1 shows one side of the symmetric hardware (flow field, heater, gas and coolant manifold, insulation, and end plates) used in the quasi-adiabatic fixture. The build layup and dimensions are summarized in Table 4. Two dual-sided flow fields/coolant channels were machined out of graphite (BMC-940, MetroMold, Rogers, MN, USA), the flow fields were comprised of opposing triple serpentine channels (width 0.75 mm, depth 1 mm, and land/channel ratio 1.5) to provide reactant transport, and parallel coolant channels (width 2.54 mm, depth 1.52 mm, and land/channel ratio 1) assisted thermal management through heat generated at nominal operating temperatures of the applied load (80 ◦C circulating 60% ethylene glycol/40% water, Isotemp 9500, Fisher Scientific, Hampton, NH, USA).


**Table 4.** Symmetric lay-up of quasi-adiabatic PEMFC fixture centered around the proton-conducting membrane.

The standard portions of the fuel-cell fixture used in both the standard and quasi-adiabatic fixtures had current collectors (gold-plated copper, electroplated in-house) compressing the flow fields. To maintain electrical continuity between the flow fields and the current collector at the non-reactive interfaces, a compressed non-hydrophobic SGL25AA was placed in the window of the polytetrafluoroethylene seal. Kapton-encased resistive heating arrays (Omega Engineering, KH-608/5-P, Norwalk, CT, USA) were positioned near the coolant side of the flow fields. Aluminum end plates (6061-T6) external to the quasi-adiabatic portion of the fixture and stainless-steel bolts torqued to 40 in-lbs. were used to maintain uniform electrical contact and force across the MEA. The quasi-adiabatic portion of the fixture had internal gas manifolds made of polycarbonate (Quadrant EPP PC 1000,

Reading, PA, USA) and were insulated from the aluminum endplates with spray-polyurethane sealed Balsa wood (Specialized Balsa Wood, LLC, Loveland, CO, USA).

#### *3.2. Instrumentation*

A Scribner 850e fuel cell test system was the central control unit for the PEMFC testing presented herein. The system monitored cell voltage, temperature and high frequency cell resistance, while establishing reactant gas flow with specific relative humidities (RH), applied load, and isothermal temperature. A Labview program and supporting hardware were used to monitor test station/software communication and perform the necessary actions for the freezing and cold-start sequence. The thermocouple used to monitor the PEMFC temperature was a flexible ultra-fine (insulated 0.24 mm diameter) designed for in-vivo applications (T-type, Physitemp IT-24P) with an accuracy of ± 0.1 ◦C and located in the cathode flow field channel. Sub-zero temperatures were established using the Isotemp 9500 lab chiller and a ScienTemp 43–1.7 chest freezer equipped with a bulkhead fitting to allow electrical and feed/exit line connections. The membrane resistance was monitored under non-applied load conditions using a Milliohm meter (Agilent Technologies, 4338B, Santa Clara, CA, USA). During cold-starts, the Labview program monitored the current and voltage measured by the test station to emulate adjacent cell heating. The heating pad output was set to be a multiple of the heat that would be generated from adjacent cells in a stack (1× and 2×) and was controlled by two independent-phase angle fired controllers (Eurotherm Corp., Model-984, Worthing, United Kingdom).

#### *3.3. Materials Characterization*

Water uptake tests were performed on both the manifold and insulation materials by immersing approximately 10 g cubic samples in water at room temperature for 15 min and evaluating mass increases. Contact uniformity under axial load was evaluated using compression paper (super low, Fujifilm, 0.5–2.5 MPa) instead of the membrane and catalyst layers between the flow fields.

#### *3.4. PEMFC BOL Conditioning*

At the beginning of life (BOL), to hydrate and activate the PEMFC, 10 cathode potential cycles were run at 80 ◦C (gas feed dew points 75% RH) by maintaining the anode potential at 0 V vs. DHE (100% H2, 0.75 slpm) and varying the cathode potential by switching between 100% N<sup>2</sup> (~0.12 V, 1.5 slpm) and air (>0.9 V, 21% O<sup>2</sup> in a N<sup>2</sup> balance, 1.5 slpm). The RH of the PEMFC feed steams was calculated from the due points (TDP) of the saturators for the specific cell temperature (T) according to the August-Roche-Magnus approximation [62] (Equation (2)):

$$\%RH = 100\% \left(\frac{\exp^{\left(\frac{17.62ST\_{DP}}{T\_{DP} + 243.04}\right)}}{\exp^{\left(\frac{17.62ST\_{D}}{T\_{CD} + 243.04}\right)}}\right) \tag{2}$$

Then, H2/Air polarization curves were performed from open circuit to 0.3 V until the voltage response profile stabilized. Between all polarization curves, the accumulated surface oxides on the cathode surface were reduced in the presence of N<sup>2</sup> to remove surface oxides.

#### *3.5. Freeze Pre-Conditioning*

The initial water content (λ*initial*) was reestablished prior to each subzero test by (i) repeating five H2/Air and H2/N<sup>2</sup> potential cycles at 80 ◦C (45% RH), (ii) two H2/Air polarization curves at 80 ◦C (45% RH), and (iii) establishing the equilibrium λ*initial* by purging the cell with symmetric N<sup>2</sup> (0.75 slpm) at either 45 ◦C (45% RH) or 35 ◦C (75% RH) for >18 h. After the equilibrium purge, the gas feed/exit lines were closed, an electrical shorting strap was placed across the anode and cathode to protect the

cathode from high carbon corrosion potentials (>0.6 V), and the cell was frozen to −20 ◦C. The λ*initial*'s were calculated from the feed %RH's using the equation developed by Hinatsu et al. [63] (Equation (3)):

$$
\lambda\_{\rm initial} = 14.1 \left( \frac{\%RH}{100} \right)^3 - 16 \left( \frac{\%RH}{100} \right)^2 + 10.8 \left( \frac{\%RH}{100} \right) + 0.3 \tag{3}
$$

The λ*initial* values used within this study were 3.2 and 6.2 (45% RH and 75% RH, respectively). Prior to all subzero testing, the coolant was purged from the PEMFC coolant channels.

#### *3.6. Water Fill Tests*

After the completion of all cold-start variations, the MEA was removed from the quasi-adiabatic fixture and installed in a reference PEMFC fixture, and then subzero isothermal water fill tests were performed at −20 ◦C [15]. Initially, the open circuit voltage was established in the presence of H2/Air (0.05/0.10, 0% RH), on the anode and cathode, respectively. A small constant load of 10 mA cm−<sup>2</sup> was applied until the cell voltage dropped below 0.1 V. Runs were repeated 2–3 times to ensure accuracy.

#### *3.7. Cold-Starts*

Subzero cold-starts were performed at −20 ◦C in the quasi-adiabatic PEMFC fixture. Initially, the open-circuit voltage was established in the presence of H2/Air (0.5 slpm/0.75 slpm, 0% RH), on the anode and cathode, respectively. Under applied load using the upper set point value of 600 mA cm−<sup>2</sup> , the stoichiometry was never less than 2. The applied load was controlled either galvanically or potentiostatically. The galvanic loads were ramped up to the set point in less than 1 min as the non-negative PEMFC voltage could sustain higher currents. The potentiostatic hold was initially set to 0.1 V. The output of the heating pads (Q in Watts) was scaled by a multiplier of either 1× or 2× to Equation (1). Runs were repeated 2–3 times to ensure accuracy.

#### **4. Conclusions**

Single-cell, −20 ◦C cold-starts were attained in a quasi-adiabatic fixture, consisting of polycarbonate gas manifolds and balsa wood insulation. This fixture used heating pads placed on the exterior of the internal flow fields to simulate the anticipated heat from adjacent cells in a stack. A 2× heating factor was used due to adjacent cell heating and thermal losses from the flow-field mass. The quasi-adiabatic single-cell fixture can emulate the thermal temperature rise and product water redistribution during cold-starts. Only a limited number of published, stack-level cold-start results, restricted mostly to simulations, are presented in the literature. The majority of the published subzero PEMFC testing is done on single cells and quantifies the water fill capacity before freezeout using a water fill test. The results presented herein succinctly demonstrate the inadequacies of the commonly used lab scale isothermal water fill tests in validating operational and material subzero cold-start capabilities. The higher rate of water production during the galvanic cold-starts (600 mA cm−<sup>2</sup> ) showed maximum hydration of the membrane within less than 2 min in contrast to the 4–8 min required in the isothermal water fill test (10 mA cm−<sup>2</sup> ). As the internal cell temperature rose above 0 ◦C during a cold-start, nearly 20 mg of water were produced. However, for the isothermal water fill test, the highest water fill capacity (λ*initial* = 3.2) was only 17.4 mg. The higher λ*initial* of 6.2 had a lower isothermal water storage capacity than that of 3.2, but conversely, a galvanic cold-start resulted in a shortened time to usable power. The potentiostatic cold-start (0.1 V) provided useful power immediately, resulting in superior cold-start performance.

**Author Contributions:** Conceptualization was originally provided by C.A.R. stemming from her work at United Technologies Corporation and material selection was led by A.O.P. Both A.O.P. and C.A.R. developed the methodology, while A.O.P. further refined the testing. All experimental work was performed by A.O.P. under the direct supervision of C.A.R. Data analysis was initially performed by A.O.P. and further processed by C.A.R. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding

**Acknowledgments:** The authors would like to thank Chris Wilson from the Mechanical Engineering Department at Tennessee Tech University for discussion on material compliance, and the Center for Manufacturing Research for facilities and financial assistance.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Molecules* Editorial Office E-mail: molecules@mdpi.com www.mdpi.com/journal/molecules

MDPI St. Alban-Anlage 66 4052 Basel Switzerland

Tel: +41 61 683 77 34 Fax: +41 61 302 89 18

www.mdpi.com ISBN 978-3-0365-1543-4