Electrode Edge Cobalt Cation Migration in an Operating Fuel Cell: An In Situ Micro-X-ray Fluorescence Study (2025)

The μ-XRF experiment was conducted at the Advanced Photon Source (APS) beamline 2-ID-D, Argonne National Lab. The MEAs used in the μ-XRF study were made by hot pressing the 50μm thick cobalt-doped membranes with electrode decals at 140°C, 4.3MPa for 5 minutes. Both the cathode and anode included 47 wt% Pt/Vulcan (TKK) with geometric Pt loadings of 0.2 mg_Pt cm⁻², and 10μm in thickness. The electrodes were fabricated with the traditional decal transfer method at an ionomer-to-carbon ratio (by mass) of 0.75 for both electrodes. A commercially available ionomer solution (Nafion D2020, EW1000 g eq⁻¹, DuPont) was used for both electrodes. The gas diffusion layers (GDL) were 234μm thick Sigracet 29 BC. All tests were conducted at constant flow rates: cathode at 0.19 slpm, and anode at 0.1 slpm. Humidification, gas flow rates, and heating were controlled by a Scribner 850e test station. Owing to the small size of the test cell, an Autolab PGSTAT302N was utilized in the galvanostatic mode to manually apply the selected current loads to the cell. The cell potentials, temperatures, and relative humidities were logged by the Scribner test stand.

The X-ray beam energy was monochromated to 8.5keV via double Si(111) crystals providing an X-ray flux of approximately 4×10⁹ photons second⁻¹. Zone plate optics and a pinhole in a Pt foil allowed the final beam spot size to reach 0.25μm in diameter. The chosen fluorescence signals (Co Kα₁, S Kα₁, and Pt Mα₁ at 6.93032, 2.30784, and 2.0505keV, respectively), were collected by a Vortex-EM/ASIC silicon drift detector (single-element, energy dispersive). The two dimensional (2-D) line scan data was reduced and fitted via a software suite (MAPS) developed at the APS.

An overview of the cell and sample geometry during μ-XRF data collection is presented in Figure 1. Figure 1A presents an image of the test cell aligned with the X-ray beam and the Si drift detector. During data collection, the X-ray beam impinges on the front of the cell. The cell is then moved so that the beam rasters from anode to cathode to complete a single 90 second line scan (0.5μm steps at 0.2s dwell time per spot). To minimize MEA damage by the beam at these energies, the cell was shifted vertically by 5μm prior to every line scan to image a different portion of the MEA. Figure 1B provides a frontal view of the experimental cell where the MEA is the thin, vertical dark stripe between the two TiN-coated titanium plates. A single flow channel (1mm x 1mm) was machined into each titanium plate. The MEA active area was approximately 1 cm² (5.6mm x 17.8mm).

Figure 1C provides a schematic of the MEA and includes a representation of the total data collection depth (yellow shading). Figures 1D and 1E present top and side views of the MEA/cell components. The dashed box in Figure 1D provides a relative boundary for Co Kα₁ fluorescence collection, and clearly encompasses portions of the non-conductive gasket and gas diffusion layer (GDL). The 'front-end' gasket, 0.5mm deep in the direction of the incident X-ray beam (Figure 1E), was necessary to prevent egress of humidified gases during cell operation. While this geometry prevented flooding of the X-ray beam pinhole (Figure 1A), it served to electronically insulate a small portion of the MEA at the front of the cell. As will be discussed later, this front-end gasketed area had important implications regarding data collection and analysis.

With respect to the resolution and element-specificity, EPMA is a powerful tool for determining the concentration profiles of elemental constituents in a sample. Figure 2A presents a typical EPMA map of a MEA cross section. Simultaneous sulfur mapping provided the boundaries of the PFSA membrane, and the Co cations are clearly evenly distributed in the membrane from the anode to cathode. Despite the superb clarity of the EPMA elemental map, the technique suffers from two distinct disadvantages. First, this is a milled MEA cross section, and the tailing of the Pt signal at the anode side is an artifact resulting from less than perfect sample alignment. As this study is focused on Co cation distributions within the membrane, the smearing of the anode electrocatalyst signal can be reasonably discounted during interpretations of the sum map. The biggest disadvantage for EPMA, however, is that it is an inherently ex situ technique, requiring that the MEA be dried, milled, and cross-sectioned prior to measurement. Because the MEA cross section had dried out, it does not provide a snapshot of the Co distributions at the end of the fuel cell test if the cations migrate during membrane drying.²⁰ While the positional resolution of this elemental map is superb, it likely does not reflect the Co gradients under operating conditions (humidified and under load).

Figure 2B presents 2-D, μ-XRF elemental maps of the same MEA obtained under operating conditions (N₂/N₂, 100% RH, 80°C). The positions of the Pt Mα₁ bands are located at the anode and cathode (left and right Pt bands throughout Figure 2), respectively, and the S Kα₁ map provides the boundaries of the 50μm thick membrane. Although the line scan appears to show a consistent, non-zero Pt concentration across the membrane, this is an artifact driven by the close proximity of the Pt Mα₁ and S Kα₁ emission lines (2.05 and 2.31keV, respectively). Of the greatest interest, however, is the distribution of Co cations in the membrane. Co is distributed evenly throughout the membrane as evidenced by the high signal intensity. Extraction of a single line scan (Figure 2C) reveals an elemental map similar to the EPMA data (Figure 2A), but the μ-XRF line scan reflects the Co distribution under fully-humidified conditions (albeit under N₂/N₂). It should be noted that a small but measureable amount of Co is observed in both the anode and cathode electrode areas. The lower count rates in the electrodes arise from not only lower Co concentrations in the electrodes versus the membrane, but also from the smaller escape depth for the Co Kα₁ emission line through Pt/C than the membrane (see supplemental material).

To examine the impact of electrical load on Co cation movement in the membrane, the cell was operated with different current densities in H₂/air followed by N₂/N₂ operation to redistribute the cations. Figure 3A shows the Co signal evolution over time (from top to bottom) corresponding to the test conditions listed to the left side of the map. The intensity of Co is color-coded according to the scale on the left. Periodically, a 90s time lapse between each 2-D scan set was required for re-positioning the MEA. These repositioning events are represented as the horizontal black lines in Figure 3A. Viewed in their entirety, the tests reveal repeating patterns. Under nitrogen (Step #1), the Co cations redistribute equally across the membrane. Under H₂/air conditions (Step #2), a Co cation concentration gradient begins to form. By the end of the H₂/air tests, the Co concentration at the cathode side of the membrane has drastically reduced while the anode side reveals a smaller concentration reduction. Note that the Co concentration decreases at the cathode under load. This is the exact opposite of all reasonable expectations. Before addressing this anomaly, however, it is important to establish if the Co migration is reversible. Upon the reintroduction of N₂/N₂ (step #3) to the cell following the 0.1 A cm⁻² hold in H₂/air, the Co cation concentration gradient gradually relaxed back to the even distribution observed in step #1. To have a clear view of the Co cation movement, 2-D maps of step 3 were converted to overlaying line scans in Figure 3B. It is clear that the initial unbalanced Co distribution at the beginning of the N₂/N₂ test ("BOT" in Figure 3B) has returned to a uniform distribution after a mere ∼1200 seconds (end of test, "EOT"). These results show that these tests can be reliably initiated from the same baseline, uniform Co cation through-plane distribution prior to each load experiment.

Figure 4A shows the extracted Co Kα₁ line profiles from anode to cathode at two different current densities. Under H₂/air, the Co concentration decreases faster at the cathode side than at the anode. Figure 4B compares the changes of the Co concentration with time at a position (marked in Figure 4A) near the cathode under different loads. The "% ΔCo Signal" term was calculated by comparing the Co counts in H₂/air with those under N₂/N₂ at steady state (∼47k counts). For both conditions in H₂/air, the initial Co counts are within 5% of the steady state N₂/N₂ signal, and only begin to deviate after 200s. The slopes in Figure 4B indicate the rates of Co depletion at the cathode position marked by the dashed line in Figure 4A. As expected, a higher load (applied current density) results in a faster depletion rate. At 0.25 A cm⁻² the Co concentration reaches 50% after ∼700s, but requires nearly 900s at 0.1 A cm⁻².

At first review, these results contrast with the expectation that, under load, metal cations would migrate toward the cathode. Rather, the observed cation concentration toward the cathode decreases. A further complication is the lack of Co material balance in the cross-section. The membrane in this study was exchanged with Co to a level representing 12% of the exchange capacity. This means that the ∼47k Co Kα₁ counts under N₂/N₂ (Figures 2C and 3B) are indicative of a 12% Co cation loading. At the conclusion of the load tests, however, the total Co Kα₁ counts at any position across the membrane are well below 47k indicating that Co is being lost from the observed system during the load tests. This cannot be the case, however, as the counts return to 47k during the open circuit N₂/N₂ conditions. Two fundamental questions have been raised by the results of the load tests: 1) why do the Co concentrations decrease on the cathode side under load when they should be increasing, and 2) why does the observed total cumulative Co cation concentration decrease across the entirety of the membrane under load?

A rational explanation for this deviation is based on the measurement location where edge of the MEA is being probed. In our experimental cell, represented in Figure 1D, the Co concentration gradients are a combination of both through-plane (anode to cathode) and in-plane (front edge to interior) migrations. The cell has a thin (500μm) insulating gasket at the front edge of the MEA (anode and cathode) to prevent escape of humidified gases into the beamline optics and detector during data collection. However, this electronically-insulated (inactive) area outside of the GDL edge is devoid of through-plane Co migration owing to the gas feed restriction and subsequent lack of appreciable through-plane current. As will be discussed below, Co that migrates through-plane (from the anode to cathode) can also migrate in-plane (from the inactive area into the active area) near the cathode to reach the exposed active area of the cathode. Hence the presence of this insulated, inactive area generates an additional potential gradient from the active area to the MEA region under the electrically insulating and gas-impermeable gasket. It is important to consider that the Co Kα₁ escape depth through Nafion is only ∼800μm (see supplemental material). Under open circuit (N₂/N₂) conditions it is reasonable to assume that the Co concentration is uniform through the entire 800μm sampling depth. However, if Co depletion occurs over the front 500μm depth of the gasket under load, the registered Co Kα₁ counts will arise only from the rear ∼300μm past the gasket thereby significantly reducing the total intensity.

It should be noted that previously-reported cation mathematical models^19,22,23 are 1-dimensional (1-D) models, describing cation migration in only the through-plane direction. They also must assume the conservation of the cation concentration in that plane. It is clearly shown in these experiments that the Co cation concentrations also change in the in-plane direction, and that a 1-D model is insufficient to describe the phenomenon.

Co cation transport can be driven by concentration gradients (diffusion), potential gradients (migration), and water flow (convection) in the ionomer phase. To provide a better understanding of the Co cation movement observed in these experiments requires elucidating the ionomer phase potential distributions. These particular experiments started with uniform Co concentration and water content (that is, no initial driving force for diffusion and convection). We therefore have developed a model to illustrate potential distributions within an MEA under normal operation. This model considers reactant species transport in the gas phase in addition to proton transport in the electrolyte, while neglecting water transport for simplicity. As it is intended to provide only a qualitative explanation to the observed Co movement based on potential gradients, Co transport itself is not modeled. Figure 5 shows a schematic of the MEA edge appearing in the μ-XRF experimental cell where the X-rays impinge the MEA at x = 0. The front edge of the MEA is exposed to ambient air. During fuel cell operation, hydrogen is oxidized into protons at the anode, while oxygen is reduced in the presence of protons (transported from the anode through the membrane) to form water at the cathode. In order to provide a semi-quantitative rationale for the observed Co movement from the region under gasket to the active area, a two-dimensional, steady-state, isothermal, electrochemical and transport-coupled MEA model is developed based on the laws of conservation of species and charge:

where c_j is the molar concentration of species j (j = H₂, O₂), a_k is the specific area (units of cm²_Pt cm⁻³_electrode) for reaction k that consumes species j, ϕ_e is the electrolyte phase potential, D_j is the effective diffusivity of species j, and κ_eff is the effective proton conductivity of the electrolyte phase accounting for the effects of the electrolyte volume fraction (ɛ_e) in the electrode (i.e. κ_eff = ε_eκ. The volumetric current density of an electrochemical reaction that involves n_k electrons, a_ki_k, is evaluated via the electrode kinetics of HOR²⁴ and ORR²⁵ at a given cathode potential that refers to zero anode potential for the solid phase. Gas velocity v is assumed to obey Darcy's law:

where K is the gas permeability of a catalyst layer, μ is the gas viscosity, and P is the total gas pressure. It is further assumed that gas flows only in-plane in an electrode under the gasket due to the pressure difference between the cell and ambient environment, and convection is negligible elsewhere. The in-plane gas velocity under the gasket is approximately constant, around 0.5 μm s⁻¹ kPa⁻¹ in the cathode and twice as high in the anode, assuming a linear pressure profile under the gasket.

Equations 1 to 2 are discretized by the finite-volume method and solved numerically to obtain species, potential, and current distributions with the following boundary conditions:

It is assumed that gas flow through the electrodes is fully developed at x = 0 and symmetry applies at x = x₂. The gasket is an insulating layer, impermeable to gas. Furthermore, no protons transport out of the MEA.

Additionally, at electrode/membrane interfaces, the reactant gas flux reads:

where p_j is the partial pressure of gaseous species j, δ_m is the thickness of membrane, and K_j is the permeability of species j (j = H₂, O₂).²⁶ Eq. 10 assumes that H₂ (or O₂) permeating through the membrane is completely consumed by the respective reactant species (O₂ or H₂) to form water on the other side of membrane. The applied current density I is calculated by integrating the ORR current over the entire cathode divided by the active area A.

The approximation results from imperfect symmetry.

Under load, the model predicts that the insulating gasket between the cell front and the active area results in a smaller ORR reaction current in that region (Figure 6A). This reduction of current occurs because H₂ and O₂ transport are less efficient in the in-plane direction of the electrodes without the gas-diffusion layer. This produces a smaller electrolyte potential drop in the insulated area across the membrane from anode to cathode than that present in the active area. The simulated ionomer electrolyte potential distribution (contour plots in Figures 6B and 6C) clearly show that an in-plane potential gradient from cell front to the active area develops under load due to the difference in the through-plane potential across the membrane between the insulated and active areas. While the normalized current density distributions look similar under different loads, the through-plane potential gradient increases with load. Consequently, higher loads result in greater in-plane potential gradients. Under load, the through-plane potential gradients would drive cation (e.g. Co²⁺) migration from anode to cathode, while the greater in-plane potential gradients drive cations to migrate from the insulated area to the active area. These gradients counter balance diffusion driven by the concentration gradients within the cathode developed under load, approaching a steady-state. The degree of the potential gradient increases with the applied current density as does the Co migration rate. This predicted in-plane potential gradient qualitatively correlates well with the Co migration versus load relationship observed from the μ-XRF measurements.

This model only provides a qualitative rationale of the Co movement in a MEA under fuel cell operating conditions using this cell design. To produce a truly quantitative model of the Co cation migration will require careful measurement of Co diffusivity and mobility. These studies are currently underway in our laboratories. Nonetheless, these μ-XRF measurements, interpreted using a 2-D transport model, clearly indicate accumulation of Co cations at the active-area cathode (membrane and electrode) under load. We are now aiming to refine the μ-XRF cell design to minimize the MEA edge architecture effects and thus to allow for direct spectroscopic observations of Co accumulation in the active area and putting model validation in reach.

These μ-XRF studies suggest that potential gradients can also be induced in the in-plane direction, and that this is dependent on physical factors such as cell design and operating conditions. Method improvements are needed via modification of the cell design to eliminate the impact of the insulating gasket and allow direct measurement in the active area. Future experiments in this area will explore operating conditions and cell design sensitivities on the migration of Co and other cations (e.g. Ce, Fe, etc.) and their impact on fuel cell operation and durability.