Formation of Liquid Water Pathways in PEM Fuel Cells: A 3-D Pore-Scale Perspective

We investigated the 3-D pore-scale liquid water distribution within the cathode GDL via in operando synchrotron X-ray tomography during low current density fuel cell operation to capture the early appearance of liquid water pathways. We found that the invasion of liquid water into the GDL only partially ﬁ lled certain GDL pores. Liquid water preferentially ﬂ owed along some GDL ﬁ bers, which was attributed to the hydrophilic nature of carbon ﬁ ber and the presence of pore-scale mixed wettability within the GDLs. Heterogeneous

Growing energy use necessitates the adoption of renewable energy sources and sustainable energy conversion devices with low carbon emissions to mitigate anthropogenic climate change. 1 Polymer electrolyte membrane (PEM) fuel cells offer a robust solution for on-site energy conversion and utilization without local carbon emissions. 2 However, the high cost of the PEM fuel cell has bottlenecked its adoption. Innovative water management strategies aimed at the cathode gas diffusion layer (GDL) have the potential to drastically improve high current density performance of the fuel cell and thereby reduce costs. 2 Understanding liquid water transport within the GDL is crucial for driving this much-needed innovation and design. At the pore-scale, most commercial GDLs exhibit a complex heterogeneous structure with mixed wettability due to the non-uniform dispersion 3,4 of a hydrophobic binder, typically polytetrafluoroethylene (PTFE), on the hydrophilic carbon fibers of the GDL. The hydrophobic binder is commonly applied via a dipcoating procedure which results in a heterogeneous coverage of the binder, where the binder may distribute partially on the fibers, deposit as a thin layer along fibers, or preferentially accumulate near the GDL surfaces. 3 To implement successful water management strategies, we need to understand the effect of the complex heterogeneous nature of the GDL on the pore-scale transport of liquid water.
Recent advances in pore-scale modeling [5][6][7] and visualization [8][9][10][11] have shed light on certain water transport mechanisms existing in the complex heterogeneous pore structure of the GDLs. Y. Nagai et al. 8 used in operando synchrotron X-ray computed tomography to demonstrate that primary liquid water pathways, which were established in the large micrometer-sized pores of their custom microporous layer (MPL), promoted the efficient removal of liquid water by limiting the number of entry pathways in the GDL substrate. S. J. Normile et al. 9 showed that large voids (∼10 μm) at the catalyst layer-MPL interface served as locations for liquid water pooling or membrane swelling. S. Alrwashdeh et al. 10 mapped the water distribution in the nanometer-sized MPL pores using synchrotron X-ray phase contrast tomography to elucidate that liquid water in the MPL preferentially accumulated under the flow-field ribs rather than under the channels. J. Gostick et al. 6 simulated porescale liquid water transport using pore network modeling and calculated transport properties of gas and water in GDLs with varying saturations. J. Hinebaugh et al. 7 studied the effect of empirically-determined heterogeneous porosity profile of GDLs on liquid water saturation and found that smooth porosity distributions and a low porosity at the catalyst layer-GDL interface were effective in reducing the overall GDL saturation.
To further understand liquid water transport, the effect of mixed wettability in the complex GDL structure has gained recent attention. 12-15 L. Hao et al. 12 used the lattice Boltzmann method to demonstrate that liquid water transport exhibited capillary fingering and stable displacement regimes in GDLs with high hydrophobicity and neutral wettability, respectively. They also showed that hydrophilic regions in the GDL could be tailored to create preferential liquid water pathways that remain stable once liquid water breakthrough was achieved. R. Wu et al. 13 incorporated mixed wettability in pore network modeling to find that a non-uniform distribution of hydrophilic pore fractions (higher hydrophilic fraction near catalyst layer) led to a lower liquid water saturation. S. M. Moosavi et al. 14 performed pore-scale simulations on experimentally-characterized partially saturated GDLs and showed that hydrophobic treatments on GDLs led to significant improvements in water permeability without impacts on the effective diffusivity of air. Although there has been considerable work on the topic of mixed wettability, the precise nature of how the initial liquid water pathways form in GDLs is not yet known. Initial pathways become established as preferential pathways for liquid water transport, even with increased water generation rates, and hence play an important role in predicting the overall liquid water distribution within the GDL.
In this study, we investigate the three-dimensional (3-D) porescale liquid water distribution within the cathode GDL in operando via synchrotron X-ray tomography during low current density fuel cell operation. The goal of the study was to capture the early appearance of liquid water pathways to enhance our understanding of how liquid water pathways form within GDLs and how the pathways are affected by the mixed wettability of the GDL structure.

Experimental
Here, we describe the experimental setup used for in operando fuel cell tomography. Fuel cell electrochemical performance was characterized while acquiring X-ray tomographic projections of the fuel cell. We then describe the image processing and visualization procedures used to convert the tomographic projections into spatially resolved 3-D liquid water distributions.
In operando fuel cell tomography setup.-Fuel cell test setup.-A custom miniature fuel cell was designed and built for electrochemical testing and in operando X-ray computed tomography z E-mail: abazylak@mie.utoronto.ca *Electrochemical Society Student Member. **Electrochemical Society Member.
(setup shown in Fig. 1). The 5-layer membrane electrode assembly (MEA), with a circular active area of 0.071 cm 2 , consisted of a Pt/Ccatalyst-coated Nafion ® HP membrane (0.3 mg Pt cm −2 on each side, Ion Power) placed between Sigracet SGL 25 BC (Sigracet ® GmbH) gas diffusion layers (GDLs). The GDLs (0.3 cm in diameter) were compressed to 75% of their original thickness using rigid polyethylene naphthalate (PEN) gaskets. The MEA was compressed between two graphite flow-fields (assembly seen in Fig. 1b), which consisted of three parallel channels that were 0.5 mm in width and depth, separated by 0.5 mm-wide ribs. The flow-field plates were placed between two hard gold-coated copper current collectors (out of field-of-view in Fig. 1b), which conducted electric current between the flow-fields and the external circuit.
Electrochemical testing was conducted using a fuel cell test stand (Scribner 850e, Scribner Associates Inc.) coupled with a potentiostat (885 Fuel Cell Potentiostat, Scribner Associates Inc.). The fuel cell was operated at ambient temperature (21°C) and pressure at two operating current conditions: a) reference open circuit voltage (0 A cm −2 ) and b) low current density step of 0.05 A cm −2 . The low current density step was chosen to specifically study the early appearance of liquid water pathways formed in the GDL. Hydrogen and air were fully humidified and supplied to the anode and cathode channels, respectively, in a counter-flow configuration. High reactant flow rates of 0.375 l min −1 were maintained to avoid liquid water flooding at the channels and maintain high stoichiometry (>5000) along channel lengths.
X-ray tomography setup.-The operating fuel cell was imaged using synchrotron X-ray tomography at the Biomedical Imaging and Therapy Wiggler Insertion Device (05ID-2) beamline facility at the Canadian Light Source (CLS) in Saskatoon, Canada. 16 The fuel cell was oriented with the MEA plane horizontal and parallel to the incoming X-ray beam. The X-ray beam was collimated and monochromatic with an energy level of 30 keV. A total of 1500 two-dimensional (2-D) projections were captured at 0.12°rotation increments (total rotation angle of 180°). 2-D images of the incident beam, i.e., flat-field projections, were captured at the beginning and the end of each tomography scan. Each projection image was captured with an exposure time of 300 ms, a pixel resolution of 6.5 μm per pixel (as used successfully by L. Battrell et al. to identify GDL pore space 17 ), and a field-of-view of 13.3 mm (width) by 1.5 mm (height). Individual GDL fiber diameters range between 1-2 pixels (mean fiber diameter of ∼7.6 μm 18 ); however, binder and PTFE were not distinguished from the fibers in this scan, hence bundled fibers or fibers coated with binder/PTFE could appear larger than 1-2 pixels wide. A 100 μm thick LuAG scintillator (CRYTUR spol. s r. o.) converted the transmitted X-ray irradiance into visible light which was detected by a digital scientific complementary metal-oxide-semiconductor (sCMOS) camera (ORCA-Flash4.0, Hamamatsu Photonics K.K.). For each current density step, tomographic imaging was conducted for 15 min at steady state after a 10-minute stabilization period which was sufficient to stabilize water distributions within the GDL. 19 We used a step-and-shoot CT imaging protocol, in contrast to higher temporal resolution on-thefly CT imaging, 17 and a monochromatic beam to obtain high fidelity images of the stabilized water distributions.  Image processing and liquid water visualization.-The 2-D tomographic projection images were processed to obtain 3-D liquid water and material density distributions. A dark-field image (an average of 10 images taken in the absence of the incident X-ray beam) was subtracted from each projection image to correct for the background noise of the camera. Each projection image was corrected for the non-uniform response of the camera, scintillator screen, and incident X-ray beam 20 using an averaged flat-field image (an average of 20 flat-field projections taken before and after tomographic projections). The corrected projection images were reconstructed into 3-D images using the filtered back-projection algorithm available in the NRecon software (Bruker Corporation). 21 A sample cross-sectional slice of the cathode GDL during operation at OCV and later during the 0.05 A cm −2 step is shown in Figs. 2a and 2b, respectively. The pixel intensities of the image slices represent material density, where darker intensities (black) indicate lower material density (e.g., void space) while brighter intensities (grey) represent solid material (e.g., carbon fiber) and/or liquid water. The 3-D images of the fuel cell during OCV and the 0.05 A cm −2 step were registered for any unwanted translation and rotation using the imregister function in MATLAB (MathWorks). 22 The 3-D image taken at OCV was subtracted from the aligned image taken during the 0.05 A cm −2 step to obtain the liquid water distribution during fuel cell operation (Fig. 2c). The 3-D OCV image represents a dry reference state image where the GDL is devoid of liquid water. The subtracted water distribution image was filtered using a 3-D median filter (width of 2 pixels in each direction) to reduce noise, contrast-enhanced, and segmented into water and background using Otsu's automatic threshold. 23 Unphysical pixelwide holes created in the water distribution during the segmentation process were filled in using the Fill Holes function in Fiji. The resulting 3-D liquid water distribution was overlaid onto the 3-D reference OCV image to visualize liquid water with respect to the GDL materials and pores (Fig. 2d). Two software, namely Dragonfly (Object Research Systems) 24 and Fiji, 25 were used to visualize the liquid water distribution in 3-D (Fig. 2e).

Results and Discussion
3-D liquid water pathways were visualized within the fuel cell at a current density of 0.05 A cm −2 (cell potential of 0.43 V; low cell performance was attributed to a high ohmic resistance of 1.3 Ω cm 2 at 0.05 A cm −2 , low OCV of 0.81 V, and high kinetic losses at room temperature). In this study, we report and focus on two specific porescale phenomena observed within the fuel cell: partial filling of GDL pores and preferential flow of liquid water along GDL fibers and channel boundaries. We then present a discussion of some experimental insights into predicting the formation of liquid water pathways within GDLs.
Partial filling of GDL pores.-We observed that the invasion of liquid water into the GDL only partially filled certain GDL pores. Representative examples are shown in Fig. 3. In the example slice shown in Fig. 3d, the area fraction of invaded pores covered by liquid water was only 85% rather than 100% (fully filled). To better understand the mechanisms for partial filling of GDL pores, we investigate the 3-D distribution of water using cross-sectional orthogonal views. We report two possible explanations for the observed partial filling of pores: a) capillary barrier imposed by narrow constrictions in pore morphology, and b) preferential flow of liquid water along GDL fibers and channel boundaries.
It is important to consider the 3-D pore morphology when examining pore-scale liquid water transport. The shape and size of the pore space (pore/throat diameter and connectivity) and the roughness of the pore walls can have significant effects on water transport. 26,27 Here, we report an effect of the 3-D shape and size of the pore space on water transport. In the representative example  shown (Fig. 4), the top view shows a partially filled pore space (Fig. 4a). The side and front views (Figs. 4b and 4c, respectively) show that the unfilled void space constricts and narrows towards the edges (outlined for clarity in Fig. 4d). The narrower regions in the apparently hydrophobic pore space would exhibit higher threshold capillary pressures and be unfavorable for liquid water transport compared to the wider (water-filled region) in the pore, and consequently lead to a partially filled pore (Fig. 4a). This pore filling phenomena has also been discussed in the literature through numerical 28,29 and experimental 30 studies of multiphase flow through porous media.
Preferential flow of liquid water along GDL fibers and channel boundaries.-Liquid water was observed to preferentially flow along some GDL fibers and channel boundaries (Fig. 5). In certain locations, water preferentially flowed and surrounded GDL fibers (shown using red * in Fig. 5). This preferential flow of liquid water along GDL fibers is highlighted for the respective regions of interest using cross-sectional orthogonal slices (water clusters II and III in Fig. 6 and cluster IV in Figs. 5d-5f) and 3-D isometric views (water clusters II and III in Fig. 6d and cluster IV in Fig. 5g). This observation was attributed to the hydrophilic nature of the constituent carbon fiber (contact angle of ∼80°for plain single fiber 31 ) and the presence of mixed wettability within the GDLs. Most commercial GDLs, including SGL 25 BC (used in this study), consist of hydrophilic carbon fibers that are hydrophobized with a non-uniform dispersion 3,4 of PTFE binder (contact angle of 110°3 2 ), where PTFE may cover the fiber partially, as a thin layer, or preferentially near the surface. 3 In addition, liquid water was observed to wick along channel boundaries (shown using red°in Fig. 5), owing to the hydrophilic nature of constituent graphite plates. Due to the mixed wettability in the GDL, a mixed drainage-imbibition process may occur in the GDL, where liquid water preferentially flows along hydrophilic carbon fibers and graphite flow-field boundaries rather than hydrophobic PTFE. This mixed drainage-imbibition process may lead to the partial filling of GDL pores (as seen in Figs. 3 and 5).
On predicting the formation of liquid water pathways: discussion of experimental insights.-Assuming capillary-dominated liquid water transport (negligible temperature gradient at low current density and negligible vapor transport at low temperature fuel cell operation 33 ), liquid water is transported from the catalyst layer to the flow-field by sequentially filling one pore to the next based on the threshold capillary pressure of the smallest constriction between the pores (i.e., throat). Threshold capillary pressure is affected by a) the morphology of the pore space, which determines the distribution of individual pore/throat diameters and connectivity of pores in the path of liquid water transport (as seen in Fig. 4) and b) the local wettability of the pore (as seen in Fig. 5). The 3-D morphology of the pore space can be accurately modelled when the physical pore space is extracted and segmented into a collection of individual pores/throats, as done in state-of-the-art pore-scale simulations of the GDLs. 5,26 However, to accurately predict and model the effects of heterogeneous mixed wettability on the formation of liquid water pathways in GDLs, we recommend to accurately characterize and incorporate the actual 3-D distribution of heterogeneous wettability within the GDL, as also suggested by M. Sabharwal et al. 34

Conclusions
We visualized initial liquid water pathways in the cathode GDL pores in operando. We found that certain GDL pores invaded by liquid water were partially, rather than completely, filled. Partial pore filling was attributed to the presence of complex 3-D pore morphology and heterogeneous mixed wettability within the GDL. Liquid water was observed to preferentially flow along some GDL fibers and channel boundaries, which was attributed to the hydrophilic nature of the carbon fiber/graphite flow-field and the presence of pore-scale mixed wettability within the GDLs. Our results demonstrate the significance of mixed wettability at the pore-scale for the formation of initial liquid water pathways in the GDL. The phenomena of mixed drainage-imbibition due to mixed wettability should be incorporated and leveraged in GDL modeling and design in order to tailor liquid water transport pathways in next-generation GDLs.