Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Toward Understanding Catalyst Layer Deposition Processes and Distribution in Anodic Porous Transport Electrodes in Proton Exchange Membrane Water Electrolyzers

Toward Understanding Catalyst Layer Deposition Processes and Distribution in Anodic Porous... IntroductionHydrogen from water electrolysis will play a key role as a clean and sustainable energy storage solution in the upcoming decades in the context of a decarbonized economy.[1] According to the study “Hydrogen Roadmap Europe: A sustainable pathway for the European Energy Transition”[2] of the fuel cells and hydrogen joint undertaking of the year 2019, nearly one‐quarter of the total energy demand in the European Union could be covered with hydrogen by 2050. So‐called proton exchange membrane water electrolyzers (PEMWEs) will be one of this transition's mainstays. They are already available at the multi‐MW scale and currently approaching the GW scale.[3,4] Nevertheless, it is still necessary to decrease their price per kWh for the proliferation of PEMWEs. This economization must be accompanied by a simultaneous increase in their performance, efficiency, and durability.[5,6]The centerpiece of a PEMWE is the membrane electrode assembly (MEA). It consists of an anode and a cathode, each with a porous transport layer (PTL) and a catalyst layer (CL), and a solid polymer electrolyte membrane in between. In principle, the thin film electrodes of the MEAs can be processed as catalyst‐coated membranes (CCMs) or as catalyst‐coated substrates.[7,8] In the former case, the catalyst ink is either coated directly (e.g., via an ink‐spray‐printer or a roll‐to‐roll process)[9] or indirectly onto the membrane via the so‐called decal transfer method. In the other case, the catalyst ink is coated onto the PTL. Afterward, the membrane is sandwiched between the anodic and the cathodic porous transport electrode (PTE) to obtain an MEA.Although CCMs are the state‐of‐the‐art MEA manufacturing technique for PEMWE, PTEs could be a promising alternative concept for MEA structures due to a different fabrication approach and the related advantages. A technological advantage of PTEs is the reduction of working steps in the manufacturing process compared to the CCM approach by directly processing the catalyst layer onto the substrate, which should lower manufacturing costs. Moreover, the PTE setup allows a higher variety of membranes in the MEA composition, for example, if the glass‐transition temperature of the membrane exceeds the decomposition temperature at the decal transfer.[10] Fluctuating temperature and humidity can change the degree of membrane swelling. As a consequence of the decoupling of the membrane and the catalyst layers, the PTE configuration reduces the deformation of the catalyst layers in comparison to CCMs. On the one hand, the direct spray coating on the well‐conducting titanium also enhances the PTE configuration's electronic conductivity compared to CCMs. On the other hand, the ionic conductivity worsens because parts of the catalyst layer are no longer connected in the ionomer‐membrane network.[11] Furthermore, the concept of PTEs has been successfully implemented also in other electrolyzer systems, like anion exchange[12] and bipolar membrane water electrolyzers.[13,14] Likewise, PTE manufacturing has already been successfully implemented on an industrial level. For example, Toshiba Corporation recently developed a new sputtering technology for the manufacturing of PTEs for large‐scale production.[15,16] In general, the catalyst layer can be either directly coated[17–19] or electrochemically deposited.[20–22] Bühler et al. showed that the performance of PTEs fabricated via spray coating on titanium fibers is similar to CCM with an identical loading of 1.5 mg cm−2 iridium oxide, indeed with worse kinetics, but a tendency to improve mass transport properties.[18] However, Leonard et al.[11] observed a 200 mV worse performance of a sintered titanium PTE at 1 A cm−2 compared to a CCM in a micro‐computed tomography (Micro‐CT) setup capable of measuring operando X‐ray computed tomography (CT) and radiography. They traced the performance loss to a lower connection between catalyst particles and membrane. The anodic electrode had a loading of 3 mg cm−2 iridium oxide ultrasonic spray‐coated onto a sintered titanium paper. Based on their previous publication,[11] the same group performed a systematic study on the influence of the interface between catalyst and PTL for a CCM setup and between catalyst and membrane for a PTE setup.[23] They used sintered (mean pore radius 13.31 µm) and fiber Ti PTLs (mean pore radius 8.07 µm) and varied the iridium oxide loading between 0.5 mg cm−2 and 2.2 mg cm−2. They showed that PTEs perform worse than CCMs at the same catalyst loading and using identical PTLs. Sintered and fiber Ti PTEs had a similar performance at high loadings. Cloe et al. demonstrated that electrochemically deposited iridium oxide on a titanium mesh showed not only a high current density of 0.91 A cm−2 at 1.6 V, but also a protective function against titanium corrosion.[20]Whereas CCMs and PTEs show comparable electrochemical data, their structures, especially the catalyst distribution, deviate from each other due to their different manufacturing techniques (see Figure 1). The catalyst layer of a CCM can be described as a flat homogeneous layer with a nearly constant thickness (see Figure 1c,d). In contrast, for the PTEs, the morphology of the catalyst layer and the structure of the PTL are directly connected because of the directly coated catalyst on the titanium PTL (see Figure 1a,b).1FigureIllustration of differences in the catalyst layer distribution between CCMs and PTEs. a) Surface morphology and b) embedded cross‐section of IrO2‐coated titanium fibers. c) Surface image of the anode catalyst layer of a CCM. d) Embedded cross‐sectional image of a CCM. The lower white stripe is the anode catalyst layer, and the upper is the cathode. A Nafion 117 membrane is sandwiched in between the layers. Epoxy resin and PTFE spacers from embedding are visible above and below the catalyst layers.To the best of our knowledge, no literature exists about a complete quantitative determination of catalyst distribution in anodic porous transport electrodes in PEMWEs. Leonard et al.[11] and Kulkarni et al.[23] imaged in operando PTE PEMWEs via X‐ray micro‐tomography. Due to the resolution limit of above 1 µm of the X‐ray CT imaging, they did not fully resolve the catalyst distribution of the electrodes. Nevertheless, they resolved a majority of the catalyst at different loadings. Thereby, they demonstrated that the catalyst layer follows the morphology of the titanium fibers (fiber diameter of ≈20 µm). With increasing loading, the catalyst covers the top of the fibers and the curved surface area and underlying fibers. They determined the triple‐phase contact area (TPCA) using the tomographies. The TPCA is the area where the catalyst layer overlaps with the membrane and the PTL after an interfacial projection.[11] In general, the TPCA increases with higher loadings, and the TPCA of titanium fiber PTEs is higher than the TPCA of sinter titanium PTEs.[23] Furthermore, electron micrographs of PTEs are found in previous literature reports.[13,18] These images can only convey a qualitative impression of the morphology of respective PTEs. As mentioned above, the catalyst distribution of a pristine CCM is trivial. The catalyst layer can be simply described by its thickness. The underlying nanoporous structure of a PEMWE anode catalyst layer can be analyzed via FIB‐SEM tomography.[24]Numerous works have been published on porous transport layers in CCM setups and their structure.[25,26] In general, the PTL is responsible for transporting water and oxygen and establishing electrical contact with the catalyst. Titanium is mainly chosen as the PTL material due to its good electric conductivity and non‐corrosive bulk properties, which are built up due to a passivating oxide layer under the operating conditions of a PEMWE.[27] There are different types of Ti PTLs, like sintered titanium powders[28,29] or sintered titanium fibers.[29,30] They vary additionally in their structural properties, for example, porosity, thickness, or pore diameter. Overall, the choice of a PTL is always a tradeoff between mass transport, electrical conductivity, and connectivity of the catalyst layer and the PTL. Therefore several studies were performed in the last years to investigate the influence of the structure of a PTL on the electrochemical performance in CCM setups.[25,31–33] For instance, Weber et al.[34] suggested an optimum ratio of 0.5 between PTL thickness and flow field rib width. Furthermore, the interface between PTL and the catalyst layer has been the subject of several studies. The interfacial properties influence the mass transport resistance and the high‐frequency resistance (HFR).[28,35,36] Schuler et al.,[30,35] for example, investigated morphology, topology, and correlated performance of titanium fiber meshes‐based PTLs with different fiber diameters and porosities in a CCM‐based setup. They concluded that a higher interfacial contact area, including a higher catalyst utilization and a lower interfacial resistance, correlates with better performance. One similar titanium fiber mesh (2GDL40‐1.0, Bekaert) is also used in our study. With this knowledge, Schuler et al. designed PTLs with a porous backing layer which increased the catalyst layer utilization and improved mass transport properties.[31]However, the insights from the numerous investigations of PTLs in CCM setups cannot be directly translated into the PTE approach. To improve and optimize the anodic PTEs, it is necessary to understand their morphology, especially the interplay between PTL and the catalyst layer. The publications of the Zenyuk group[11,23] already visualized parts of the catalyst distribution and determined the TPCA for PTEs. In our paper, the main focus is on the through‐plane catalyst distribution. The characterizing system and manufacturing way follows the titanium fiber mesh aPTE approach of Bühler et al., who used spray coating as a deposition technique.[18] Based on this publication,[18] a PTE with a catalyst loading of 1.5 mg cm−2 iridium oxide and a binder content of ≈10 wt.% was chosen as a system to be examined because of its good reproducibility and performance. The catalyst distribution of this coated electrode is determined by the analysis of cross‐sectional images of embedded samples and confirmed via Micro‐CT analysis. Furthermore, the previous two methods give deeper insights into the structural changes occurring in the PTE design regarding surface roughness, volume fraction gradients, and catalyst layer thickness. Finally, the results are implemented in an empirical spray coating model. This model allows us to understand better the influence of the manufacturing processes on the catalyst layer morphology and thus to make predictions for future optimized PTE designs.Results and DiscussionElectrochemical DataIn the following, we verify the ionomer content of the manufactured PTE. Additionally, we present the electrochemical data and compare them to the literature values.ManufacturingThe ionomer content of the spray‐coated PTE was determined to be 10.7 wt.%, which matched the intended ionomer content of ≈10 wt.%. With this ionomer content, a good tradeoff between reproducibility and performance can be achieved, according to Bühler et al.[18] The corresponding thermogravimetric analysis (TGA) can be found in the supplementary information (see Figure S1, Supporting Information).Electrochemical AnalysisThree samples were electrochemically characterized according to the protocol described in the experimental part. The average and the standard deviations of the polarization curves are displayed in Figure 2. The three samples showed similar behavior. The highest current densities were reached at ≈3 A cm−2 because the safety shutoff limit of 2.3 V was reached at the respective galvanostatic electrochemical impedance spectroscopy (EIS) measurements. Additionally, the HFR was determined via galvanostatic EIS (see Figure S2, Supporting Information). The HFR‐free cell voltage was calculated with the knowledge of this resistance (see Figure 2). The obtained electrochemical data conform to the results of Bühler et al., who used an identical MEA configuration but a different testing setup.[18] The HFR‐free cell voltage at 1.5 A cm−2 was at 1.606 ± 0.016 V, which lies between the voltages Bühler et al. measured for the aPTEs with 9 wt.% (1.59 V) and 12 wt.% (1.61 V) ionomer content.[18] The HFR at higher current densities is also in a similar range as in Bühler et al.[18] (between 0.19 and 0.20 Ω cm2).2FigureMean polarization curve with the standard deviations of the three measured aPTE, including the measured cell voltage and the determined HFR‐free cell voltage in a current density range from 5 mA cm−2 to 2.9 A cm−2.Structural AnalysisIn this section, we will discuss the structure of the manufactured PTEs. We used PTEs with a size of 5 mm × 5 mm for the structural analysis. We analyze their structure, draw conclusions about the catalyst layer thickness and calculate the surface roughness.PTE StructureThe complete tomogram of the PTE, received by the segmentation of the Micro‐CT dataset (volume of ≈3.00 × 4.25 × 1.16 mm3 with a small void volume at the bottom and top part in the z‐direction), can be seen in Figure 3a. The iridium oxide catalyst layer and the titanium fibers are not distinguished. The catalyst layer is on top of the segmented volume (close to the z = 0 plane in Figure 3a). Some big ink droplets can be observed at the surface. By analyzing the through‐plane volume fraction gradient (see Figure 3b), the spray coating process raises the solid volume content by ≈7% compared to the mean average of the plain PTL. The structural and transport parameters of the titanium fibers are determined by cropping the segmented PTE to a core part with a volume of 3.00 × 4.25 × 0.75 mm3 to only consider the titanium fibers not affected by the spray coating process. Therefore no catalyst covers the considered fibers, and edge effects can be neglected. The solid volume content of this section is φbulk = 49.16%, corresponding to a porosity of ρ = 50.84%. A representative volume analysis of the bulk volume confirms the global validity of the porosity value (see Figure S3, Supporting Information). The pore size distribution is displayed in the supporting information (see Figure S4, Supporting Information). Mean and median values are slightly above 30 µm. The applied fiber find algorithm's result is illustrated in Figure 3c. The algorithm was capable of allocating the fibers, but not as a whole, just piecewise. The fibers have a diameter of 20.59 ± 3.15 µm (see Figure 3d). They are uniformly orientated in the in‐plane direction. However, they have a strong horizontal alignment in the through‐plane direction (see Figure 3d and fiber orientation tensor in Table S1, Supporting Information).3Figurea) Segmented PTE of the Micro‐CT data. The top side shows the coated titanium fibers. b) Through‐plane trend of the volume fraction (along the z‐direction) and resultant determination of the minimum catalyst layer thickness or coating dependence on the through‐plane position. c) Bulk‐section of the PTE, which is equivalent to a titanium fiber PTL, with the identified fiber fragments. d) Fiber diameter and in‐plane and through‐plane fiber orientation distribution of the previous bulk volume.The structural parameters, porosity, and mean fiber diameter, deviate slightly from the manufacturer's data (ρManufacturer = 56% and dManufacturer = 20 µm). Likewise, literature values also differ from the specified parameters of the same manufacturer. Schuler et al. determined for a similar PTL (nominal ρManufacturer = 56% and dManufacturer = 22 µm) a porosity of 54% and a mean diameter of 15.4 µm.[30] Immerz et al. stated the porosity as 50% of a Bekaert PTL with the product name 2GDL40‐1,00.[37] The reasons for this deviation can be diverse. For instance, the differences can be due to the choice of the analyzing method and the associated limitations, for example, the resolution limit of the Micro‐CT or voxel size. Also, variations in the manufacturing process of the fiber felts can be a reason. For instance, minor variances in the fiber diameter significantly impact the porosity: Assuming a diameter of 20 µm instead of the here‐determined diameter, the porosity would already increase by 3% (ρ20 µm  = ρMicro − CT rMicro − CT2/rManufacturer2≈54%${\rho _{20\;\mu {\rm{m}}\;}}\; = \;{\rho _{{\rm{Micro}}\; - \;{\rm{CT}}}}\;r_{{\rm{Micro}}\; - \;{\rm{CT}}}^2{\rm{/}}r_{{\rm{Manufacturer}}}^2 \approx 54\). A more precise way to resolve the fibers would be a Nano‐CT. However, a higher resolution would lead to a smaller segmented volume and thus sample size. The chosen region of the segmentation influences the porosity. If the segmented volume contains edge areas of the PTL, the overall porosity will increase (see Figure 3b). In addition, variations in the manufacturing process cannot be excluded. Therefore samples from different batches would need to be examined for statistical reasons.Catalyst Layer ThicknessThe volume fraction of the bulk area of the PTE Micro‐CT data is very homogeneous and lies between 48% and 50% (see Figure 3b). However, an increase in the solid content can be found close to the top of the PTE (relative to z = 0, compare Figure 3a,b). This peak in volume fraction can be ascribed to the catalyst layer covering the titanium fibers. Consequently, the catalyst layer thickness can be estimated from the above results. We propose two models for the calculation of the catalyst layer thickness. One assumes the fiber is fully covered, and the other assumes the fiber is half covered (see Figure 3b).In the first model, it is assumed that the catalyst covers the whole fiber with a homogeneous layer and the fibers possess a cylindrical shape (see Figure 3b). We define volume elements Vi,complete, specified by the product of the total in‐plane area and the through‐plane height of a voxel length of 2.5 µm. Each volume element Vi,complete has mi fibers of length li and a total radius composed of the sum of the titanium fiber radius ri and the catalyst layer thickness ci,complete:1Vi,complete = mi π(ri + ci,complete)2li\[\begin{array}{*{20}{c}}{{V_{i,{\rm{complete}}}}\; = \;{m_i}\;\pi {{({r_i}\; + \;{c_{i,{\rm{complete}}}})}^2}{l_i}}\end{array}\]In the second model, only the upper half of the cylindrical‐shaped fiber is homogeneously coated with a constant catalyst layer thickness ci,half (see Figure 3b). The i‐th volume element Vi,half can be calculated as:2Vi,half = 1/2mi π(ri + ci,half)2li + 1/2miπ(ri)2li\[\begin{array}{*{20}{c}}{{V_{i,{\rm{half}}}}\; = \;1{\rm{/}}2{m_i}\;\pi {{({r_i}\; + \;{c_{i,{\rm{half}}}})}^2}{l_i}\; + \;1{\rm{/}}2{m_i}\pi {{({r_i})}^2}{l_i}}\end{array}\]For both models, the average bulk volume of the PTL without catalyst with the same height, n fibers of the length lbulk, the titanium fiber radius rbulk can be expressed as:3Vbulk =  nπrbulk2lbulk\[\begin{array}{*{20}{c}}{{V_{{\rm{bulk}}}}\; = \;\;n\pi r_{{\rm{bulk}}}^2{l_{{\rm{bulk}}}}}\end{array}\]Assuming now the same numbers of fibers per volume element (n = mi), the same fiber length ( lbulk = li = l) and titanium fiber radius ( rbulk = ri = r) the minimum catalyst layer thickness ci,complete, min and ci,half, min of the i‐th volume element can be calculated:4ci,complete, min = (Vi, completeVbulk − 1) r  = (φiφbulk − 1) r\[\begin{array}{*{20}{c}}{{c_{i,{\rm{complete}},\;\min }}\; = \;\left( {\sqrt {\frac{{{V_{i,\;{\rm{complete}}}}}}{{{V_{{\rm{bulk}}}}}}} \; - \;1} \right)\;r\;\; = \;\left( {\sqrt {\frac{{{\varphi _i}}}{{{\varphi _{{\rm{bulk}}}}}}} \; - \;1} \right)\;r}\end{array}\]5ci,half, min = (2Vi, halfVbulk − 1  − 1) r  = (2φiφbulk − 1 − 1) r\[\begin{array}{*{20}{c}}{{c_{i,{\rm{half}},\;\min }}\; = \;\left( {\sqrt {\frac{{2{V_{i,\;{\rm{half}}}}}}{{{V_{{\rm{bulk}}}}}}\; - \;1\;} \; - \;1} \right)\;r\;\; = \;\left( {\sqrt {\frac{{2{\varphi _i}}}{{{\varphi _{{\rm{bulk}}}}}}\; - \;1} \; - \;1} \right)\;r}\end{array}\]φi is here the solid volume content of the i‐th volume element and can be extracted from the Figure 3b dataset. The calculated results are displayed in Figure 3b. The estimated minimum of the fully covered fiber model is given by the peak in Figure 3b and is accordingly ≈0.66 µm. The estimated minimum of the half‐covered fiber model is ≈1.29 µm and is reached close to the surface of the PTE and strongly declines within the next 100 µm.The simplification that the titanium fibers are homogeneously covered with catalyst doesn't correspond to reality. The majority of the catalyst is concentrated on the top part of the titanium fibers (see Figure 4a; Figure S5, Supporting Information). This observation corresponds with FIB‐SEM (focused ion beam scanning electron microscopy) cross‐sections of other PTEs manufactured.[13,14,18,19] However, some fully covered fibers can also be observed, mostly present due to clogging of the catalyst ink occurring at some narrow pores within the titanium fiber system. Therefore, the second model, which only assumes that half of the fiber is homogeneously coated, is closer to reality but with some limitations (e.g., partially inhomogeneous coating and partially fully covered fibers due to clogging). Generally, the derived catalyst layer thickness should only be considered as a lower bound catalyst layer thickness of the top part of the titanium fibers. Especially the simplification that the fiber density is equal at the bulk and the edge area probably does not match with real PTLs. The difference in fiber density is highlighted in the publication of Peng et al.,[38] where the through‐plane porosity profiles of several Bekaert PTLs showed a strong increase in porosity at the edges of the substrate. Consequently, the mean catalyst layer thickness of our PTE would be higher. Nevertheless, the estimation is in good agreement with literature data: Bühler et al. showed a FIB‐SEM cross‐sectional image of a similar PTE (same recipe, only with 12 wt.% Nafion and a loading of 1.4 mg cm−2 iridium) with a catalyst layer thickness of ≈2 µm.[18] The images of Bühler et al. correspond well to our minimum catalyst layer thickness of 1.3 µm at the edge area, considering the rough surface structure of the PTEs (see Figure 1a). Additionally, Kulkarni et al.[23] resolved a majority of the catalyst layer of PTEs with different loadings with a Micro‐CT with a voxel size between 1.16 and 1.73 µm. Therefore the minimum catalyst layer thickness should be close to the resolution of their tomographic data. Furthermore, this group also observed a covering of the entire top area and a curved surface area of the fiber. This strengthens the assumption of the half‐coated fibers. Nevertheless, our simplified assumptions are not qualified to reproduce the catalyst layer thickness and distribution in real PTEs. Therefore, more precise imaging and tomography techniques are necessary.4FigureDetermining the catalyst distribution via embedded cross‐sectioning. a) A segment of one stitched BSD image. In total, three cross‐sections, each with a total image width of several mm, were analyzed to obtain a reliable catalyst distribution. b) Segmentation of the image in (a). c) Normalized catalyst distributions of the three different embedded cross‐sections and their weighted average. Additionally, a lognormal distribution was fitted to the average catalyst distribution.The volume fraction gradient also leads to a local pore gradient. From the literature, it is known that an increasing porosity gradient from CCM to the flow field improves the reacting transport, and additionally, less oxygen saturation is observed.[36,39] This could be an additional reason for the better mass transport properties in comparison to CCMs Bühler et al.[18] discovered. The higher volume fraction could also reduce ohmic losses because the additional connection through the catalyst layer could increase the electrical conductivity.Surface RoughnessThe spray coating of the titanium fibers affects the structure and the surface roughness in two ways. On the one hand, void space on the surface shrinks or is even closed due to the deposition, which could reduce surface roughness. On the other hand, the coating alters the surface of the titanium fibers, for example, big catalyst droplets can occur that raise the surface roughness. Additionally, the surface roughness increases due to the porous catalyst layer itself. This effect happens on the nm scale and, therefore, cannot be detected by the Micro‐CT.Here, the surface roughness parameters of the catalyst‐coated PTE and its non‐coated backside were calculated based on the Micro‐CT data (see Table 1 and Supporting Information) to compare a PTE with an uncoated PTL. The backside of the PTE can be considered equal to an uncoated PTL since the PTL does not show anisotropy in the through‐plane direction. Further, the majority of the deposited catalyst can be found within 100 µm from the surface while the full PTL has a thickness of 1 mm (see Figure 3b). Consequently, the uncoated backside of the PTE resembles a pristine PTL and can therefore act as a reference. The surface roughness profile height maps (see Figure 5) visualize these surface roughness parameters. The visible fibers of the PTE roughness profile height map have a larger diameter than the pure titanium fibers. Furthermore, some big ink droplets on the surface of the PTE can be detected (see Figure 5b: dark circles). These droplet hills can have a height of up to 50 µm. Therefore, the coated fibers have a higher profile height. This visual observation is consistent with the mean surface roughness parameter: the mean surface roughness parameter of the PTE is more than twice the value of the PTL. Whereas the arithmetic mean and root mean square surface roughness are nearly the same. When we chose a surface area without big catalyst droplets for the calculations (see Figure 5b gray outlined area and Figure 5c), the surface roughness parameters of PTE and PTL are similar.1TableCalculated surface roughness parameters of the PTL, specifically the PTE backside, the catalyst‐coated PTE, and a reduced section of the PTE, where no big catalyst droplet can be found. The table includes the mean surface roughness RmPTL/PTE$R_{\rm{m}}^{{\rm{PTL}}/{\rm{PTE}}}$, the arithmetic mean surface roughness RaPTL/PTE$R_{\rm{a}}^{{\rm{PTL}}/{\rm{PTE}}}$, and the root mean square roughness RqPTL/PTE$R_{\rm{q}}^{{\rm{PTL}}/{\rm{PTE}}}$Surface roughness parameters [µm]PTLPTEPTE‐section (gray outlined area)RmPTL/PTE$R_{\rm{m}}^{{\rm{PTL}}/{\rm{PTE}}}$42.486.748.7RaPTL/PTE$R_{\rm{a}}^{{\rm{PTL}}/{\rm{PTE}}}$26.226.426.0RqPTL/PTE\[R_{\rm{q}}^{{\rm{PTL}}/{\rm{PTE}}}\]36.236.435.95FigureRoughness profile height maps of the PTE backside (a) and the spray‐coated front side (b). The gray outlined section (b) is an area without any big catalyst droplets and is displayed separately in (c). The profile height color bar is depicted beside (c).Indeed, the mean surface roughness of the PTL is in good agreement with the literature values from Schuler et al.[30] They postulated a linear relationship between fiber diameter and mean surface roughness based on their experimental results:6mean surface roughness (µm)=2.1×fiber diameter(µm)−3 µm\[\begin{array}{*{20}{c}}{{\rm{mean}}\;{\rm{surface}}\;{\rm{roughness}}\;\left( {\mu {\rm{m}}} \right) = 2.1 \times {\rm{fiber}}\;{\rm{diameter}}\left( {\mu {\rm{m}}} \right) - 3\;\mu {\rm{m}}}\end{array}\]Using our determined fiber diameter, the mean surface roughness is 40.24 µm according to Equation 6, close to the determined value of 42.4 µm. Therefore, our calculated surface roughness is in good agreement with literature values, especially considering the broad fiber diameter distribution. Possible error sources could still be a variation in the manufacturing process, missing statistics, or a not perfectly aligned Micro‐CT.As observed above, in the absence of big dried catalyst droplets, the mean and the arithmetic mean surface roughness are very similar (see Figure 5, Table 1). The presence of this up to 50 µm high droplets may cause a mechanical degradation of the membrane caused by puncture and mechanical stress. In general, in operation times under 1000 h, mechanical stress is one of the main durability problems and causes of failures of PEMWEs.[40–42] In particular, pinholes can lead to short‐circuits and therefore reduce the efficiency of PEMWEs or even lead to their failure.[43,44]Additionally, from PEM fuel cells (PEMFC), it is known that the roughness of a gas diffusion layer (GDL, which is the equivalent of a PTL) can strongly affect the lifetime of a membrane.[45,46] Additionally, a rougher GDL increases the open‐circuit voltage and the hydrogen crossover in PEMFCs.[45]For this reason, the manufacturing process of the PTEs should be further optimized to avoid big catalyst droplets and thereby reduce the mechanical stress on the membrane.Catalyst Distribution and UtilizationCatalyst Distribution via Embedded Cross‐SectionThe cross‐sectional imaging of embedded samples allowed us to get more profound information about the catalyst layer and its distribution. In total, three datasets of the embedded cross‐section were taken. They resulted in total image width of 11.41 mm (catalyst distribution 1: 4.11 mm, catalyst distribution 2: 3.16 mm, and catalyst distribution 3: 4.14 mm). A segment of the stitched BSD image (see Figure S5, Supporting Information) can be seen in Figure 4a. Here, pore (black), fiber (grey), and catalyst (white) can be clearly distinguished with the applied imaging conditions. Additionally, the catalyst regions were verified via energy‐dispersive X‐ray spectroscopy measurements (see Figure S6, Supporting Information). The 2D catalyst regions were clearly segmented with the AI‐based algorithm Unet 2D[47] implemented in GeoDict (see Figure 4b). A through‐plane catalyst gradient and fibers surrounded by a catalyst were observed. Additionally, some catalyst agglomerations, probably due to clogging caused by pore throats, were detected. The amount of catalyst found on the backside of fibers was small. No average catalyst layer thickness was determined due to the 2D nature of the data. Overall, the embedded PTE's cross‐sectional images show a comparable catalyst distribution to the samples of the publication of Mayerhöfer et al.[13]The through‐plane catalyst gradient can be confirmed by plotting the normalized catalyst amount dependent on the penetration depth (see Figure 4c). Based on this data, we obtained a 1D catalyst distribution along the z‐direction (through‐plane direction). Whereas the single data set catalyst distributions are rough, averaging the three data sets smooths the catalyst distribution. These peaks in the single data sets occurred (see Figure 4c catalyst distribution 1) if a certain imaging plane contained long horizontal fibers with a catalyst layer on top. The related count of pixels at this penetration depth is consequently overproportioned and causes outliers in the distribution. This effect is reduced by taking the average of these three distributions (see Figure 4c). By normalizing the three distributions, the similarity of the distributions becomes visible (see Figure 4c). The three comparable distributions suggest the statistical correctness and representative character of this data. However, the catalyst distributions could be distorted by imaging artifacts, segmentation errors, statistical deviations, a crooked alignment of the stitched images, or damaging the catalyst layer during the grinding process.The average catalyst distribution is similar to a lognormal distribution. The according fit is also plotted in Figure 4c (see Table S3, Supporting Information for the fitting parameter). A coefficient of determination R2 > 97% confirms the lognormal behavior of the data. The penetration depth of the catalyst can be quantified by the P(0.90)/P(0.95) interval value. The distribution revealed that 90%/95% of the catalyst had a penetration depth of 98.5 µm/122 µm (see Table 2).2TableP(90) and P(95)‐interval of all catalyst distributions. The R2 value of the BSD and Micro‐CT catalyst distribution relates to the respective experimental data. The R2‐value of each model catalyst distribution is based on the reference lognormal distribution of the embedded sampleMethodBSDMicro‐CTModel 1Model 2Model 3Model 4Penetration depth P(0.90) interval [µm]98.51098087.595100Penetration depth P(0.95) interval [µm]122126105112.5120122.5R20.97360.9840−1.02070.70820.93610.9607Catalyst Distribution via Micro‐CTThe Micro‐CT data set was also used to determine the catalyst distribution. First of all, in each image slice of the Micro‐CT, a brighter region at the edge exists. This is attributed to the iridium oxide catalyst layer (see Figure 6a). The differences in grayscale are due to the different mass attenuation coefficients of titanium and IrO2 at the applied operation voltage and current of the Micro‐CT. The brightness and the catalyst layer become clearer by applying a maximum intensity projection on the Micro‐CT data set. The maximum intensity protection results in an imaging plane with the brightest pixels in the through‐plane direction of every point. The catalyst layer extends over 100 µm into the PTL (see Figure 6b). The through‐plane behavior of the mean grey value clarifies this observation (see Figure 6c). The baseline of the curve is determined at a grey value of 79. The first 200 µm after the intersection between the baseline and mean grey value curves are plotted magnified (see Figure 6c). The P(0.90) interval value of this section is ≈109 µm (see Table 2). The decay of the mean grey value was additionally fitted with a lognormal distribution (R2‐value of over 98%). The Micro‐CT gradient has its origin on the one hand, through the higher volume content, on the other hand, due to the higher attenuation of iridium. The little intensity peak of the through‐plane gradient on the edge of the PTL side (see Figure 6c) suggests that some beam‐hardening effects are present.6Figurea) Single slice of the raw Micro‐CT data. The brighter pixel on the left side can be attributed to the catalyst layer. b) Maximum intensity projection of the Micro‐CT data. This projection highlights the catalyst layer and makes the penetration depth of more than 100 µm visible. c) Through‐plane plot of the mean grey value of the Micro‐CT. The blue section that starts when the mean grey value exceeds the baseline is magnified to get information about the catalyst penetration depth.Comparison of Catalyst DistributionsThe P(0.90) and P(0.95) intervals are very similar (see Table 2). To compare the catalyst distribution of the embedded cross‐sectional imaging and the Micro‐CT more quantitatively, the respective data was normalized that the fitted lognormal distribution had its peak at 1.0 (see Figure 7a, the fitting parameters can be found in Table S3, Supporting Information). Both data sets have a similar trend. The Micro‐CT data is slightly above the embedded cross‐sectional image curve. That is either explained by missing statistics or probably by a not completely correct one‐to‐one translation of the mean grey value to the catalyst distribution. Nevertheless, the Micro‐CT data is qualified to gain inside into the catalyst distribution. For a quantitative usage of the Micro‐CT for the catalyst distribution, more experiments with different catalyst loadings and PTLs with different porosities would be necessary to calibrate the results.7Figurea) Direct comparison between the catalyst distribution and the according lognormal fits of the data received via the embedded cross‐section and the Micro‐CT. b) Catalyst utilization, in our publication defined as the ionically well‐connected catalyst, as a function of the penetration depth of the membrane. The catalyst utilization is calculated as the cumulative sum of the catalyst distribution. The maximum penetration depth is taken from the literature.[30]The above‐calculated catalyst distributions are only relative. The total amount or thickness of the catalyst layer is assessed in three ways: by the TGA and the catalyst loading, by the segmented BSD images, or by the through‐plane trend of the volume fraction.The first method calculates the thickness by considering the loading IrO2 (1.5 mg cm−2), the determined weight ratios of the binder and IrO2 by TGA (YIonomer = 0.1071,${Y_{{\rm{Ionomer}}}}\; = \;0.1071,$YIrO2=  0.8929)${Y_{{\rm{Ir}}{{\rm{O}}_2}}} = \;\;0.8929)$, the density of IrO2 (11.7 g cm−3) and Nafion D520 (0.92–0.94 g cm−3),[48] and an assumption of the porosity of the catalyst layer based on PEMFC and PEMWE literature values[24,49–52] (εCL = 0.5 ±  0.1):7tTGA=(loadingIrO2ρIrO2+loadingIrO2×YIonomer/YIrO2ρNafion D520)/εCL=6.44µm\[\begin{array}{*{20}{c}}{{t_{{\rm{TGA}}}} = \left( {\frac{{{\rm{loadin}}{{\rm{g}}_{{\rm{Ir}}{{\rm{O}}_2}}}}}{{{\rho _{{\rm{Ir}}{{\rm{O}}_2}}}}} + \frac{{{\rm{loadin}}{{\rm{g}}_{{\rm{Ir}}{{\rm{O}}_2}}} \times {Y_{{\rm{Ionomer}}}}{\rm{/}}{Y_{{\rm{Ir}}{{\rm{O}}_2}}}}}{{{\rho _{{\rm{Nafion}}\;{\rm{D}}520}}}}} \right){\rm{/}}{\varepsilon _{{\rm{CL}}}} = 6.44\mu {\rm{m}}}\end{array}\]Considering the estimation of the solid volume fraction of the catalyst layer εCL, the total thickness tTGA lies in the interval {tTGA|6.44 µm − 1.07 µm ≤ tTGA ≤ 6.44 µm + 1.61 µm}.The thickness can also be estimated via the three stitched BSD images (see Figure S5, Supporting Information for one stitched BSD image dataset). The average thickness tBSD is the weighted sum of all three thicknesses of the datasets. The respective weighting wi is determined by the individual length li of each dataset. The individual thickness of each segmented BSD dataset is calculated as the total area of all n Pixels with a pixel width a of 97.7 nm divided by the total length li of the stitched image:8tBSD=∑i=13wintotal,ia2li=5.01 µm±0.33 µm\[\begin{array}{*{20}{c}}{{t_{{\rm{BSD}}}} = \mathop \sum \limits_{i = 1}^3 {w_i}\frac{{{n_{{\rm{total}},i}}{a^2}}}{{{l_i}}} = 5.01\,\mu {\rm{m}} \pm 0.33\,\mu {\rm{m}}}\end{array}\]The total amount can also be determined by summing up the deviations of the volume fraction φi with the volume fraction of the bulk φbulk multiplied by the voxel size hi. The limits of the sums are the points where the volume fraction exceeds and fall below the volume fraction of the bulk for the first time:9tMicro−CT=∑i=1N(φi−φbulk) hi= 3.30µm \[\begin{array}{*{20}{c}}{{t_{{\rm{Micro}} - {\rm{CT}}}} = \mathop \sum \limits_{i = 1}^N ({\varphi _i} - {\varphi _{{\rm{bulk}}}})\;{h_i} = \;3.30\mu {\rm{m}}\;}\end{array}\]The latter calculation should only be understood as a lower limit of the total thickness of the catalyst layer. The assumption that the volume fraction at the edge is equal than φbulk is an approximation and will underestimate the actual total thickness. Therefore, it is concluded that the equivalent CCM catalyst layer thickness is correspondingly between 5.01 µm and 6.44 µm. With this total catalyst layer thickness, an absolute catalyst distribution could be determined.Catalyst UtilizationWith the known structure, it is now possible to estimate the penetration depth of the membrane into the PTE in a working PEMWE. The membrane infiltration depth and the catalyst distribution give information about the catalyst utilization. The catalyst utilization, in our publication defined as the ionically well‐connected catalyst, is calculated by integrating the catalyst distribution over the membrane infiltration depth. In general, the infiltration depth of the membrane will depend on the mechanical properties of the membrane and the PTE, the clamping pressure of the system, and the structural properties of the PTE, especially the surface roughness and the pore and fiber diameters. Schuler et al.[30] determined the membrane or rather the CCM deformation of operated PEMWE MEAs by imaging the system after operation via X‐ray tomography. They used a symmetric PTL/CCM setup, consisting of titanium fibers (Bekaert) and Greenerity E400 CCMs with a Nafion 117CS membrane (Greenerity), with a contact pressure of 2.5 MPa. The analysis was performed with several titanium fiber substrates. The most relevant PTLs for our publication, for which the membrane deformation data is available, have a nominal fiber diameter of 14 µm (L1) and 50 µm (L3) with a common porosity of 56%. They determined the mean surface roughness Rm, the surface roughness at 90%cum R0.9, and the maximum surface roughness Rmax of the membrane deformation. The maximum surface roughness Rmax of L1 was 30 µm and of L3 was 47.6 µm. Our investigated aPTE‐setup has a titanium fiber PTL with a nominal fiber diameter of 20 µm and a porosity of 56%. Therefore, the maximum deformation of our used Nafion 117 membrane should lie between these literature values of Rmax (L1) and Rmax (L3), assuming a similar contact pressure. Thus we assume that a maximal 38% to 60% of the catalyst is in contact with the membrane, considering Rmax (L1) and Rmax (L3) (see Figure 7b). A more precise estimation of the catalyst utilization would require the exact membrane deformation within the system either via simulation or experimental data from in situ computed tomography of an assembled cell.This amount of ionically well‐connected catalyst is also consistent with the results of Kulkarni et al.[23] They calculated the TPCA value of ≈50% for a fiber titanium PTE with a loading of 1.75 mg cm−2 and a TPCA value of ≈20% for an identical PTE with a loading of 1.1 mg cm−2. Linear approximation of these TPCA values leads to a TPCA value of ≈40% for our loading of 1.5 mg cm−2. The used PTL had a similar fiber diameter to our substrates. In our work, the proton conductivity of the binder in the catalyst layer and the different mechanical properties between a CCM and a pure membrane are not considered. Both of them should raise the catalyst utilization. The proton conductivity of the binder will let deeper catalyst‐coated fibers contribute to the OER reaction in the catalyst layer. Although Leonard et al.[11] and Kulkarni et al.[23] showed that the proton conductivity is a bottleneck in the aPTE setup and leads to high ionic potential losses. From PEMFC literature, it is known that the elastic modulus of the platinum catalyst layer is higher than that of the membrane.[53,54] Therefore, it can be assumed that a CCM as a compound of CLs and membrane is stiffer than the pure membrane. As a result, the pure membrane should penetrate deeper into the fibers. However, a deep membrane penetration could hinder water and oxygen transport at the top fibers, reducing the catalyst efficiency.In a first‐order approximation, only the catalyst particles contacted via membrane contribute to the catalyst utilization (see Figure 7b). Consequently, at least 40% of the catalyst in this setup are not used. As a result, future PTEs designs should also improve catalyst utilization by optimizing their structure.Additionally, the interfacial contact area between PTL and CCM is estimated. Schuler et al.[30] determined the interfacial contact area RICA of L1 and L3 (RICA (L1) = 1.81 m2/mgeo2${R_{{\rm{ICA}}}}\;({\rm{L}}1)\; = \;1.81\;{{\rm{m}}^2}{\rm{/m}}_{{\rm{geo}}}^2$ and RICA (L3) = 1.11 m2/mgeo2)${R_{{\rm{ICA}}}}\;({\rm{L}}3)\; = \;1.11\;{{\rm{m}}^2}{\rm{/m}}_{{\rm{geo}}}^2)$. The RICA of our configuration lies probably between these values. Suppose the upper titanium fibers are completely covered with catalyst. In that case, the interfacial contact area of a PTE should always be higher than a similar CCM configuration with the same components and experimental conditions because of the previously mentioned different mechanical properties and a different membrane swelling behavior.Modelling of the Spray Coating ProcessThe upper sections revealed a specific PTE system's structure and catalyst distribution. With this knowledge, the aim was to develop a spray‐coating model for PTEs on titanium fiber substrates.In reality, the coating behavior of the titanium fibers will depend on numerous variables: the ink (surface tension, viscosity, loading,[23] composition of the ink, and evaporation temperature of the solvent[14]) and settings of the ultrasonic nozzle (nozzle amplitude and nozzle frequency[55]), the surface properties of the substrate (structure, roughness, and adhesion) and the surrounding parameters (used equipment, atmosphere, temperature of the heating plate, solution flow rates[56]). In the literature, some models can predict the size of ink droplets created by ultrasonic spray coating, depending on nozzle amplitude and frequency, the surface tension, and the ink's viscosity.[55] Furthermore, there exists literature[56] that correlates process parameters, like deposition parameters and solution properties, and the formation of thin and smooth films. Kulkarni et al.[23] observed that depending on the catalyst loading, different coverages of the titanium fibers were found. While for low loadings, only the top surface of the fibers is covered with catalyst, parts of the curved fiber surface are already deposited with a catalyst for medium loadings. For higher loadings, the catalyst covering even becomes denser. In general, there exists nearly no literature about the correlation between Ir‐based catalyst inks and the catalyst layer formation on titanium substrates. Whereas the Pt/C‐based catalyst layer formation in dependence on the process and ink parameters in PEMFCs has been broadly investigated.[57]As too many physical variables and influencing factors exist, our spray coating model tried to simplify the reality as much as necessary. It is a discrete model using a voxelized implementation of the physical world. The goal of the model was to calculate a two‐dimension catalyst distribution that reproduces the catalyst distribution of the embedded sample.In the applied simulation, the ink droplets have only the size of a voxel, 2.5 µm. According to the manufacturer, isopropanol has a medium droplet size of 27 µm using a nozzle frequency of 48 kHz. The isopropanol droplet size suggests that the catalyst ink has a droplet size of a similar magnitude. Furthermore, the titanium fibers are not composed of only horizontal and vertical edges. They instead have a round or elliptical shape (see Figure 1c). Surface parameters, like roughness and adhesion, and conditions, like temperature, are expressed in the upcoming model (see models 3 and 4) by a not immediate drying of ink and an occurrence of a downward flow. Additionally, clogging of the catalyst ink can be observed (see Figure 4a,b). The clogging probably occurs at some bottleneck throats of the pore between the titanium fibers. However, clogging is not considered in our model.With these restrictions in mind, all four model scenarios described exactly in the experimental part, are illustrated and compared to the catalyst distribution of the embedded sample in Figure 8. The main model assumptions are displayed in Table 3. Additionally, the P(90) and P(95) are calculated (see Table 2). In this calculation, the total catalyst amount was taken as the catalyst amount present in this illustrated 200 µm. As a result, the catalyst located deeper into the PTL was neglected. Additionally, the coefficient of determination was determined between the model data and the reference catalyst distribution (see Table 2).8FigureCatalyst distributions displayed in (a–d) were received by four different modeling approaches. The sketches on the right‐hand side illustrate the functioning of the respective models, which were calculated on the full micro‐CT tomogram of the PTL. The titanium fibers are shown in green, and the deposited catalyst layer is in yellow.3TableAssumptions for the four different cases in the presented spray coating modelScenarioHorizontal facesVertical facesDownward flowDripping1✓✗✗✗2✓✓✗✗3✓✓✓✗4✓✓✓✓Models 1 and 2 are directly connected to the surface roughness of the PTL. Model 1 is just taking into account the horizontal edges (see Figure 8a). This leads to a non‐lognormal distribution with a negative R2‐value of −1.0207 and too‐low penetration depths (see Table 2). The catalyst distribution of model 2 (see Figure 8b) was calculated assuming that both the horizontal and vertical edges are coated with the same amount of catalyst. Model 2 was based on the assumption of the half‐coated titanium fibers presented in the previous section “Catalyst Layer Thickness” (see Figure 3b, Equation (5)). Additionally, as a simplification in model 2, no differentiation between horizontal and vertical surfaces was assumed because of the round shape of the titanium fibers in reality. Model 2 already reveals a lognormal shape of the catalyst distribution and deeper penetration of the catalyst (see Table 2). Overall, model 2 reinforces the assumption of the half‐coated titanium fibers to calculate the catalyst layer thickness (see Figure 3b). However, neither the R2‐values nor the P(90) or P(95) is in good agreement with the reference distribution.The deeper penetration depth can only be reached considering a certain flow of the catalyst while drying. Model 3 allows this downward flow of the catalyst by considering a specific evaporation time of solvents in the ink. In this scenario, horizontal and vertical edges are still coated with the same amount of catalyst. However, different flow rates of the catalyst ink of the horizontal and vertical edges can be present with the restriction that the horizontal has to be less or equal to the vertical flow due to the presence of gravitational forces. A parameter sweep of the horizontal and vertical flows between 1 and 10 voxels led to the highest congruence with the reference distribution with a horizontal flow rate of 4 voxels and a vertical flow rate of 6 voxels (see Figure 8c). The coefficient of determination (R2‐value of 0.9361) and a visual check were the reasons for this. A horizontal flowrate of 3 voxels and a vertical flowrate of 5 voxels directed into a similar distribution. Overall, this simple model already reproduces a similar catalyst distribution to the reference distribution. Nevertheless, in reality, the evaporation rate of water in an aqueous solution decays with time in a square root dependency of time.[58] Therefore, the assumption of the homogenous contribution of each catalyst droplet along the through‐plane direction is probably oversimplified. In reality, the ink distribution will be an interplay between diffusion, migration, and interfacial forces.Also, model 3 misses some amount of catalyst at deeper penetration depth. Therefore model 4 tries to compensate for that with so‐called dripping that accounts for penetration into pores below fibers, which are otherwise protected from a direct deposition. A certain amount of the ink droplet travels through the pore space and is deposited in deeper regions of the titanium substrate. Model 4 was applied to the best data set of model 3. In this process, a parameter sweep of the dripping catalyst amount in 5% steps was performed. The optimum dripping parameter was 10%. This means 10% of the ink droplets that hit the vertical and horizontal edges splashes (see Figure 8d). Therefore, the catalyst distribution is shifted and comes very close to the reference catalyst distribution (R2 = 0.9607).In summary, significant knowledge can be gained from this catalyst spray coating model. First, all models depend on the surface structure directly connected to the mean fiber diameter and the porosity. Schuler et al.[30] showed empirically that a lower mean fiber diameter leads to a lower PTL surface, assuming a constant thickness. Furthermore, the surface roughness of structures with the same mean fiber diameter increases with rising porosity. Second, a flow of the catalyst ink was evidenced and therewith, a deeper infiltration of the catalyst into the porous substrate. The drying time will mainly be influenced by the temperature of the heating plate, the choice of the solvent and its evaporation time, and the interaction between titanium fiber and catalyst ink. So in terms of improved catalyst utilization in aPTEs titanium fibers with a lower mean fiber diameter, lower porosity, and surface roughness should be preferred. This can also be achieved using a porous backing layer or a micro porous layer. It has to be noted that this model requires experimental validation for other PTL structures, like sintered titanium powders, and different electrolyzer systems, like anion exchange membrane water electrolyzers to ensure the translatability to these substrates.ConclusionIn this work, for the first time, the structure and catalyst distribution of an aPTE, produced by spray coating an IrO2 catalyst on a titanium fiber PTL in PEMWE was determined quantitatively. For this purpose, we used two different methods. First, the segmentation and evaluation of several cross‐sectional images of the PTE revealed a through‐plane catalyst layer distribution with a lognormal shape. 90% of the catalyst amount lies within the first 100 µm. The distribution was confirmed by considering the through‐plane grey‐value gradient of a Micro‐CT of a PTE. In conjunction with the membrane deformation measurement of Schuler et al.[30] (of a comparable setup), we concluded that the upper limit ionically well‐connected catalyst in the investigated system is 60%.The analysis of the Micro‐CT also revealed a porosity gradient of the PTE. The catalyst layer thickness was estimated. The mean catalyst layer thickness around the upper half of the fibers at the upper PTE surface was ≈1.3 µm. The determined values of the titanium fiber thickness and the porosity were in good agreement with the literature values.Moreover, the Micro‐CT dataset showed that, in general, the coating of the titanium fibers does not change the surface roughness of a PTE compared to the pure titanium fiber PTL. A local surface roughness increase only occurs where big catalyst droplets are present.In addition, a first model of the deposition process was developed that matches the catalyst distribution found in the tomographic data. We showed that the surface roughness properties, combined with a downward‐flowing ink, are the main responsible factors for the catalyst distributions in PTEs. We expect that the catalyst utilization will increase with lower surface roughness, lower porosity, and lower mean fiber diameter.Based on these findings, it will be possible in future works to investigate the interplay between manufacturing and resulting PTE structure. The presented models can thus allow the finding of optimum titanium structures in future works.In summary, our paper provides a first impression of the structure of PTEs and helps design a roadmap for improving and optimizing PTEs.Experimental SectionPTE FabricationThe anodic PTEs were prepared using an ultrasonic spray coater (Exacta Coat, Sonotek) with a 48 kHz AccuMist nozzle. The procedure was performed according to Bühler et al.[18] Sintered titanium fibers (2GDL40‐1.0, Bekaert) with a porosity of 56% and a mean fiber diameter of 20 µm were used as substrates laser cut to either 5 cm2 for electrochemical testing or 0.25 cm2 for structural analysis. Spray coating was performed employing the following parameters: 5 W ultrasonication power, a nozzle height of 37 mm, a hot plate temperature of 120 °C, a meander‐shaped spray pattern with a pitch of 1.5 mm, a flow rate of 0.45 mL min−1, a path speed of 170 mm s−1 and a shaping air flow of 0.6 kPa, consistent with Mayerhöfer et al.[13]The ink was prepared according to Mayerhöfer et al.[13,18] with a solid content of 1 wt.% containing 98 wt.% IrO2 (Premion, Alfa Aesar) and 2 wt.% Nafion (D520, Chemours) in a 4:1 solvent mixture of water and isopropanol. The aim was to obtain a final catalyst layer with a loading of 1.5 mg cm−2 iridium oxide and a binder content of ≈10 wt.%. The Ti PTLs were mounted onto the spray coater hot plate using frames made of 1 mm thick virginal PTFE foil (HighTechFlon) to prevent slipping. To minimize the precipitation of the catalyst particles, the ink was stirred during the spray coating process. Nevertheless, a reproducible deviation between binder content in the ink (2 wt.%) and the final catalyst layer (10 wt.%) could be observed, as already described by Bühler et al.[18] For PTE quality control, the final binder content in the catalyst layers was determined via thermogravimetric analysis (STA 449 F3 Jupiter, Netzsch). For this purpose, the residual (dried) catalyst layer was scratched off the spray coating masks and heated to 600 °C (10 K min−1 steps) in Al2O3 crucibles. With this method, the detected mass loss represents the ionomer content. The determination of the binder content is shown in Figure S1, Supporting Information). Additionally, the catalyst loading was determined by weighing a 4 cm2 reference sheet before and after the spray coating process. The catalyst loading was calculated by dividing the mass difference of the GDL by the area of the spray‐coated GDL, and by taking into account the binder content of the catalyst layer.Electrochemical CharacterizationA combination of a commercial electrolyzer test system (600‐ETS, Scribner Associates) and an external potentiostat (VSP‐300 with three additional 5 V/10 A booster boards, BioLogic) was used for the electrochemical characterization. The 600‐ETS controlled and managed the pure anode water feed for the anode (40 mL min−1, 80 °C), the purging of the cathode with nitrogen (100 mL min−1, 1 atm), and the heating of the cell and the water reservoir (80 °C). The measurement cell was the associated electrolyzer cell test fixture (600 Electrolyzer Cell Fixture, Scribner Associates).The cell was assembled similarly to Bühler et al.[18] The same carbon cloth‐based cPTE as Bühler et al. (SL‐GDE, FuelCellEtc) with a catalyst loading of 0.5 mg cm−2 platinum (60% platinum on Vulcan) surrounded by a 150 µm tick PTFE frames (HighTechFlon) was used as a cathode. This PTFE frame ensured the same high compression of ≈60%. The anodic PTE was placed within a 1 mm thick PTFE sheet (HighTechFlon). The Nafion 117 membrane (Chemours) was sandwiched between the two PTEs. Finally, eight 1/4" screws tightened the MEA and the cell fixture with a final torque of 8.5 Nm applied in three steps before heating the cell.After heating the cell to 80 °C, a break‐in procedure was performed to stabilize the system. The break‐in procedure consisted of a 1 h constant voltage of 1.8 V. Then, three polarization curves were recorded, including galvanostatic EIS measurements. The electrochemical characterization based on galvanostatic EIS measurements was similar to Suermann et al.[59] employing the following steps for the acquisition of polarization curves: 1 mA cm−2 between 1  and 10 mA cm−2, 10 mA cm−2 between 10 and 100 mA cm−2, and 100 mA cm−2 between 100 and 4 A cm−2. If the voltage exceeded 2.3 V, the polarization curve measurement was terminated to prevent titanium corrosion. Each step was held for 20 s and followed by a shortened EIS sequence to determine the high‐frequency resistance. The EIS scan consisted of a frequency sweep between 200 kHz and 100 Hz with 13 measuring points per decade and a waiting point of one period between two impedance measurements. The x‐intersection of the Nyquist plot determined the current‐dependent HFR. The final break‐in measurement was a complete EIS from 200 kHz till 0.1 Hz with 13 points per decade and a holding time of 10 min of the following current densities: 10, 20, 50, and 100 mA cm−2 – 2 A cm−2 in 100 mA cm−2 intervals. To verify the stabilization, polarization curves were recorded until two subsequent curves were observed to be identical. A minimum of three curves was recorded. Finally, the last curve was taken for further performance evaluation. Overall three PTE samples were measured following this protocol.Structural AnalysisFor all structural measurements, PTEs with a size of 5 mm × 5 mm were used, fabricated with the same ink batch as the electrochemical samples. This sample measurement was chosen as a compromise between two considerations. On the one hand, the sample size should be as small as possible because of the better resolution of the X‐ray tomography. On the other hand, the sample size was limited by the accuracy and edge effects of the laser cutting of the titanium fibers. The same sample examined via Micro‐CT was later used for all electron imaging.Micro‐CT Data AcquisitionThe Micro‐CT measurement of the coated titanium fibers was done by the company RJL Microanalytic. They used a SkyScan 2211 Multiscale X‐Ray Nanotomograph (Software Version 2.5.1) from Bruker microCT and applied a source voltage of 130 kV and a source current of 200 µA. The difference in mass attenuation coefficient of titanium and iridium oxide led to grayscale gradients in the Micro‐CT data.[60,61] The imaging volume was 1282 × 1820 × 588 pixels with a voxel size of 2.5 µm. The reconstruction was done via the software NRecon (Software Version 1.7.5.0) from Bruker microCT. Hereby, ring artifact correction, beam hardening correction, and filter cutoffs were applied.Micro‐CT ReconstructionBefore the segmentation, a 3D anisotropic diffusion filter based on Perona and Malik[62] and Gerig et al.[63] and implemented by Lopes[64] was applied to the dataset to reduce noise by simultaneously preserving the edges of the fibers. The anisotropic diffusion filter was executed in Matlab (Matlab R2018b, The MathWorks). No gradient corrections were applied. Because of the grayscale gradients, the segmentation was done with a Local Otsu method implemented in GeoDict (GeoDict Version 2022, Math2Market) to perform a region‐based segmentation. While doing so, the segmentation only distinguished between pore and material. In consequence, the titanium fibers and the catalyst layer were not distinguished. Afterward, the binary dataset was cropped at the edges to obtain finally a reconstructed PTE with a volume of 1200 × 1700 × 465 pixels. This volume still contained void areas at the edges in a through‐plane direction.Micro‐CT Surface RoughnessSchuler et al. defined several amplitude‐based surface roughness parameters such as surface roughness, arithmetic mean surface roughness, root mean square roughness, and maximum profile height.[30] These parameters are key numbers to compare the roughness of different sintered titanium fibers. Their mathematical definition can be seen in the supporting information, and the calculations were implemented in Matlab. The roughness parameters were calculated for the non‐coated backside as well as for the catalyst‐coated front side of the segmented PTE. Additionally, the surface roughness was illustrated in roughness profile height maps.Micro‐CT Maximum Intensity ProjectionThe maximum intensity projection is an algorithm that projects a 3D data set into a 2D projection plane regarding the maximum intensity of a given direction. Therefore the voxel with the maximum intensity along a specific projection path was selected and displayed in a 2D projection image.[65] This algorithm was implemented in Matlab.Micro‐CT Catalyst DistributionEven though the resolution of the Micro‐CT was not high enough to resolve the catalyst layer, a slice mean gradient of the mean gray value in the through‐plane direction could resolve the catalyst distribution qualitatively. Therefore, a mean gray value of each through‐plane plane was calculated and plotted versus the PTE depth. Afterward, the baseline of the mean gray value of the inner part of the PTE was determined. An interval could be determined in which the mean gray value exceeded and approached the baseline again. The new origin was chosen as the intersection of the mean grey value curve with the baseline. The mean grey value curve was fitted with a lognormal distribution done with the nonlinear fitting tool in Origin (OriginPro 2019 Version 9.6.0.172, OriginLab Corporation).Electron ImagingA Zeiss Crossbeam 540 FIB‐SEM microscope (focused ion beam scanning electron microscope) with a Gemini II column was used for all electron imaging. All samples were attached on aluminum SEM specimen stubs (G301 and G399, Plano GmbH) with conductive carbon pads (G3347 and G3348, Plano GmbH). Additionally, the samples were sputter‐coated either with gold (108 Manual Sputter Coater, Cressington) or carbon (MED 010, Balzers Union) before imaging to obtain better conductivity.SEM Surface ImagingThe surface of the PTE was imaged via a secondary electron detector with an accelerating voltage of 3 kV and a current of 750 pA. The sample was both carbon‐coated and contacted with silver conducting paint (G3692, Plano GmbH) at the edge of the sample to ensure better conductivity.Embedded Cross‐Section Sample Preparation and ImagingOne 5 mm × 5 mm PTE sample was sandwiched between two PTFE sheets with two plastic clips (SAMPL‐KLIP PLASTIC HOLDER, ITW Test & Measurement GmbH) and embedded in epoxy resin (Araldite 502 Epoxy Resin, Electron Microscopy Sciences) into an embedding cup (SamplKup, ITW Test & Measurement GmbH). Afterward, the sample was cured overnight at 60 °C. The subsequent grinding and polishing steps were performed with an automatic polishing machine (LaboForce‐100, Struers GmbH). Therefore, the sample was manually ground with 220‐grain size SiC sanding paper (Struers GmbH) until the edge area of the sample was removed. Next, the samples were ground successively with finer SiC sanding paper from 500 to 4000‐grain size (Struers GmbH). Finally, the embedded sample was polished using two different polishing plates (MD‐Mol and MD‐Nap polishing plate, Struers GmbH) with corresponding diamond solutions (DiaPro Mol B 3 µm and DiaPro Nap ¼ µm, Struers GmbH). In addition, the sample was ultra‐sonicated before and after the polishing procedure to clean the cross‐section surface area. Besides the conductive carbon pad and the gold sputter coating, the conductivity was also improved using aluminum conductive tape (Plano GmbH) connecting the specimen holder and the embedded sample.The imaging was performed with a four‐quadrant backscattering detector in compositional mode, applying an accelerating voltage of 20 kV and 1 nA beam current. The pixel size was 97.7 nm. The embedded sample was aligned horizontally regarding the PTE. Afterward, a sequence of images along the catalyst‐coated part of the titanium fibers was recorded with an overlap of ≈20% and a magnification of 381X (image width of 300 µm). The images were then stitched with the Grid/Collection stitching tool implemented in the imaging software Fiji.[66] Thereby, a large cross‐sectional area with a width of several millimeters was accessed with a high resolution. The horizontal alignment of the stitched image was checked by inspection and, if necessary, corrected with the Rotate‐function in Fiji and a bilinear interpolation of the image.Afterward, the embedded sample was ground, polished, and imaged two more times according to the protocol described above to obtain two more datasets for improved sample size.Embedded Cross‐Section Segmentation and Data ProcessingEach stitched gray value image was segmented with an AI‐based algorithm. The deep learning method Unet 2D[47] implemented in GeoDict was used in an iterative training process to distinguish fiber, catalyst, and pore. After each iteration, the segmented image section was visually inspected, and finally, the AI model was applied to the entire stitched image.The three segmented images were used to determine the catalyst contribution dependent on the penetration depth. The data analysis was performed in Matlab. The catalyst distribution was determined individually for each of the stitched images. Therefore, all pixels assigned to the catalyst layer were summed up for each horizontal line (thickness of 97.7 nm of each horizontal line due to the pixel size). The beginning of the catalyst layer was defined when a horizontal line exceeded a threshold of 50 catalyst pixels. Afterward, the amount of catalyst layer pixels of each horizontal line was related to the penetration depth in the through‐plane direction. Following this, the underlying next 200 µm were displayed in a normalized catalyst contribution plot. Eventually, the average catalyst contribution was determined by taking into account all single catalyst contributions weighted by their image width. Additionally, the curve was fitted with a lognormal distribution (y(x)=y0+A2πwxe−(ln(xxc))22w2)\[\left( {y(x) = {y_0} + \frac{A}{{\sqrt {2\pi } wx}}{e^{ - \frac{{{{\left( {\ln \left( {\frac{x}{{{x_c}}}} \right)} \right)}^2}}}{{2{w^2}}}}}} \right)\]. The fitting was done with the nonlinear fitting tool in Origin.For better comparability, a left‐sided confidence interval of P(0.90) or P(0.95) with a confidence level of 90% or 95% was calculated. This interval specifies the maximum penetration depth, to which 90% or 95% of the catalyst lies.Energy‐Dispersive X‐Ray SpectroscopyThe Zeiss Crossbeam 540 with an energy‐dispersive X‐ray‐detector (X‐Max 150 silicon drift detector, Oxford Instruments; Software: Aztec Version 4.2, Oxford instruments) was used for elemental analysis of the embedded samples. The elemental mapping was performed by applying an accelerating voltage of 20 kV and 1 nA beam current.Catalyst UtilizationIn general, the catalyst utilization of the anode catalyst depends on the local ionic and electronic conductivity, the water transport, and the oxygen removal. In this publication, catalyst utilization and the amount of ionically well‐connected catalyst were equated. Catalyst utilization was simplified and defined as the percentage of catalysts that could be in direct contact with the membrane. The interface area between the catalyst layer and membrane varies depending on the membrane deformation, the associated membrane infiltration depth, and the membrane's swelling. The catalyst utilization was resulted from the catalyst distribution and was the cumulative distribution.Structural ParameterFor the structural analysis, the Software GeoDict was used. The segmented Micro‐CT (volume of 1200 × 1700 × 465 Pixel) was the dataset processed here. The variation in porosity and the volume fraction in through‐plane direction was determined via the package MatDict.The through‐plane edge areas were cropped to determine the basic structural properties of the titanium PTL (volume of 1200 × 1700 × 300 pixel). Thereby only pure titanium fibers were examined, and edge effects were excluded. The porosity was determined using the package MatDict. The pore size distribution was calculated via granulometry with a bin size of two voxels employing the package PoroDict. Fiber diameter distribution and fiber orientation were determined using the package FiberFind. The fibers were fitted as curved circular fibers and identified with a machine learning method. In addition, the fiber fragments at the domain boundary were removed to exclude edge effects.Spray Coating ModelAn algorithm was developed to model the spray coating process for aPTEs to get a deeper understanding of the deposition process. The output of the algorithm was a 2D catalyst distribution. The lognormal fit of the embedded sample serves as a reference catalyst contribution to benchmark the model output.The spray coating process was simulated on a blank titanium fiber PTL. Therefore, the segmentation of the uncoated backside of the Micro‐CT titanium fiber sample was used. The surface of the fibers was not triangulated. Therefore, the surface consisted of horizontal and vertical faces with a side length of 2.5 µm.The droplet size of the catalyst ink was one voxel. Each grid point was spray‐coated. As a consequence, the spray beam consisted of 1200 × 1700 individual and independent beams perpendicular to the titanium fiber plane. This product was the total number of voxels of each plane. The spray beam only hit from above visible fibers. The voxel adjacent to a fiber was then filled with a specific amount of catalyst.Depending on the model assumptions, different catalyst distributions were calculated. The 2D model output was the amount of catalyst in dependency on the through‐plane penetration depth. The model did not provide any information about the exact 3D structure of the catalyst layer.The distribution width was limited to 200 µm to directly compare it to the lognormal catalyst contribution of the embedded sample. The starting point of the distribution of the spray coating models was chosen as follows: The origin of the distribution was the position when the position's value, equaled the catalyst amount of the individual plane, exceeded the value of the catalyst distribution that was 200 µm, equal 80 voxels, apart.Four spray‐coating cases are presented based on these previous assumptions (see Table 3 and Figure 8). Each case included the assumptions of the previous cases:The spray beam is considered to deposit on the horizontal faces of the fiber surfaces visible from above. The fibers are only coated with catalyst on their top surface.The spray beam is now capable of coating specific vertical faces with a catalyst. Considering a straight perpendicular spray beam, all vertical surfaces passed by the beam are coated. The amount of catalyst is equal on horizontal and vertical surfaces.The spray‐coated catalyst layer does not dry immediately and flows downwards. The downward flow is independent of the fiber structure and surrounding spray beams. Depending on whether the spray beam hits a horizontal or vertical face, the maximum depth of the downward flow varies. The downwards flow is regardless of the impact position of the droplet and should be interpreted as a statistical flow. The amount of ink that hits the face is equally contributed between all reached through‐plane planes.Additionally, pores are considered. In the spray coating process, a specific amount x of the catalyst ink will drip or splash through the voids after it has hidden the horizontal or vertical faces. The remaining amount of the deposited material behaves like the spray‐coated layer of model 3. The additional flying distance of the catalyst ink amount x is equal to the mean pore diameter. After flying through the void, it will hit an imaginary voxel with the same horizontal or vertical face as the hidden face and will, from that point on, also follow model 3.To better classify the obtained catalyst distributions by upper models, the coefficient of determination R2= 1−residual sum of squaresmean total sum of squares${R^2} = \;1 - \frac{{{\rm{residual}}\;{\rm{sum}}\;{\rm{of}}\;{\rm{squares}}}}{{{\rm{mean}}\;{\rm{total}}\;{\rm{sum}}\;{\rm{of}}\;{\rm{squares}}}}$[67,68] was calculated between the distributions and the lognormal fit of the embedded sample. However, a good R2‐value (worst value=− ∞, best value =+ 1) was insufficient to determine the quality of a fit. Therefore the catalyst distribution was also checked via eye. To compare the catalyst distribution and the lognormal fit, both the maximum of the lognormal fit of the catalyst distribution of the model and the maximum of the lognormal fit of the embedded reference sample were normalized to 1.AcknowledgementsThe authors gratefully acknowledge the financial support by the Federal Ministry of Education and Research of Germany in the framework of the StacIE project (BMBF/03HY103H) and the Federal Ministry for Economic Affairs and Energy in the framework of HoKaWe (BMWi/ 03EI3029A). Furthermore, the authors want to thank Johannes Bender for the TGA and Thomas Böhm for the helpful discussions.Open access funding enabled and organized by Projekt DEAL.Conflict of InterestThe authors declare no conflict of interest.Data Availability StatementThe data that support the findings of this study are available from the corresponding author upon reasonable request.S. S. Kumar, V. Himabindu, Mater. Sci. Energy Technol. 2019, 2, 442.Fuel Cells and Hydrogen 2 Joint Undertaking, Hydrogen roadmap Europe : a sustainable pathway for the European energy transition, https://data.europa.eu/doi/10.2843/341510 (accessed: March 2021).S. A. Grigoriev, V. N. Fateev, D. G. Bessarabov, P. Millet, Int. J. Hydrogen Energy 2020, 45, 26036.Silyzer 300: The next paradigm of PEM electrolysis, https://assets.siemens-energy.com/siemens/assets/api/uuid:a193b68f-7ab4-4536-abe2-c23e01d0b526/datasheet-silyzer300.pdf (accessed: November 2021).International Renewable Energy Agency (IRENA). Green Hydrogen Cost Reduction: Scaling up Electrolysers to Meet the 1.5°C Climate Goal, https://www.irena.org/-/media/Files/IRENA/Agency/Publication/2020/Dec/IRENA_Green_hydrogen_cost_2020.pdf (accessed: March 2021).C. van Pham, D. Escalera‐López, K. Mayrhofer, S. Cherevko, S. Thiele, Adv. Energy Mater. 2021, 11, 2101998.D. G. Bessarabov, P. Millet, PEM water electrolysis, Vol. 2, Academic Press an imprint of Elsevier, London, UK 2018.Manufacturing of membrane electrode assemblies for fuel cells, International conference on manufacturing research (IMCR), https://pdfs.semanticscholar.org/fc3d/c1538c2eafd0ea27df7b930cd13613869f05.pdf (accessed: March 2021).J. Park, Z. Kang, G. Bender, M. Ulsh, S. A. Mauger, J. Power Sources 2020, 479, 228819.J. Bender, B. Mayerhöfer, P. Trinke, B. Bensmann, R. Hanke‐Rauschenbach, K. Krajinovic, S. Thiele, J. Kerres, Polymers 2021, 13, 3467.E. Leonard, A. D. Shum, N. Danilovic, C. Capuano, K. E. Ayers, L. M. Pant, A. Z. Weber, X. Xiao, D. Y. Parkinson, I. V. Zenyuk, Sustainable Energy Fuels 2020, 4, 921.B. Mayerhöfer, F. D. Speck, M. Hegelheimer, M. Bierling, D. Abbas, D. McLaughlin, S. Cherevko, S. Thiele, R. Peach, Int. J. Hydrogen Energy 2022, 47, 4304.B. Mayerhöfer, D. McLaughlin, T. Böhm, M. Hegelheimer, D. Seeberger, S. Thiele, ACS Appl. Energy Mater. 2020, 3, 9635.B. Mayerhöfer, K. Ehelebe, F. D. Speck, M. Bierling, J. Bender, J. A. Kerres, K. J. J. Mayrhofer, S. Cherevko, R. Peach, S. Thiele, J. Mater. Chem. A 2021, 9, 14285.Toshiba's New Large‐scale Production Technology for Electrolysis Electrodes for Hydrogen Production from Renewables Cuts Iridium Use to 1/10, https://www.global.toshiba/ww/technology/corporate/rdc/rd/topics/22/2210-01.html (accessed: December 2022).TOSHIBA REVIEW 2018, No. 3, Technology for Reduction of Precious Metals Used in Anode Catalyst of PEM Electrolyzers, https://www.global.toshiba/content/dam/toshiba/migration/corp/techReviewAssets/tech/review/2018/03/73_03pdf/a03.pdf (accessed: December 2022).M. Bühler, P. Holzapfel, D. McLaughlin, S. Thiele, J. Electrochem. Soc. 2019, 166, F1070.M. Bühler, F. Hegge, P. Holzapfel, M. Bierling, M. Suermann, S. Vierrath, S. Thiele, J. Mater. Chem. A 2019, 7, 26984.C. van Pham, M. Bühler, J. Knöppel, M. Bierling, D. Seeberger, D. Escalera‐López, K. J. J. Mayrhofer, S. Cherevko, S. Thiele, Appl. Catal., B 2020, 269, 118762.S. Choe, B.‐S. Lee, M. K. Cho, H.‐J. Kim, D. Henkensmeier, S. J. Yoo, J. Y. Kim, S. Y. Lee, H. S. Park, J. H. Jang, Appl. Catal., B 2018, 226, 289.H. Kim, J. Kim, J. Kim, G. H. Han, W. Guo, S. Hong, H. S. Park, H. W. Jang, S. Y. Kim, S. H. Ahn, Appl. Catal., B 2021, 283, 119596.B.‐S. Lee, S. H. Ahn, H.‐Y. Park, I. Choi, S. J. Yoo, H.‐J. Kim, D. Henkensmeier, J. Y. Kim, S. Park, S. W. Nam, K.‐Y. Lee, J. H. Jang, Appl. Catal., B 2015, 179, 285.D. Kulkarni, A. Huynh, P. Satjaritanun, M. O'Brien, S. Shimpalee, D. Parkinson, P. Shevchenko, F. DeCarlo, N. Danilovic, K. E. Ayers, C. Capuano, I. V. Zenyuk, Appl. Catal., B 2022, 308, 121213.F. Hegge, R. Moroni, P. Trinke, B. Bensmann, R. Hanke‐Rauschenbach, S. Thiele, S. Vierrath, J. Power Sources 2018, 393, 62.T. L. Doan, H. E. Lee, S. S. H. Shah, M. Kim, C.‐H. Kim, H.‐S. Cho, T. Kim, Int. J. Energy Res. 2021, 45, 14207.X.‐Z. Yuan, N. Shaigan, C. Song, M. Aujla, V. Neburchilov, J. T. H. Kwan, D. P. Wilkinson, A. Bazylak, K. Fatih, Sustainable Energy Fuels 2022, 6, 1824.D. Bessarabov, H. Wang, H. Li, N. Zhao, PEM Electrolysis for Hydrogen Production: Principles and Applications, CRC Press, Boca Raton, FL 2016.M. Suermann, K. Takanohashi, A. Lamibrac, T. J. Schmidt, F. N. Büchi, J. Electrochem. Soc. 2017, 164, F973.O. Panchenko, E. Borgardt, W. Zwaygardt, F. J. Hackemüller, M. Bram, N. Kardjilov, T. Arlt, I. Manke, M. Müller, D. Stolten, W. Lehnert, J. Power Sources 2018, 390, 108.T. Schuler, R. de Bruycker, T. J. Schmidt, F. N. Büchi, J. Electrochem. Soc. 2019, 166, F270.T. Schuler, J. M. Ciccone, B. Krentscher, F. Marone, C. Peter, T. J. Schmidt, F. N. Büchi, Adv. Energy Mater. 2019, 285, 1903216.J. O. Majasan, F. Iacoviello, P. R. Shearing, D. J. L. Brett, Energy Procedia 2018, 151, 111.B. J. M. Etzold, U. Krewer, S. Thiele, A. Dreizler, E. Klemm, T. Turek, Chem. Eng. J. 2021, 424, 130501.C. C. Weber, T. Schuler, R. de Bruycker, L. Gubler, F. N. Büchi, S. de Angelis, J. Power Sources 2022, 15, 100095.T. Schuler, T. J. Schmidt, F. N. Büchi, J. Electrochem. Soc. 2019, 166, F555.J. K. Lee, C. Lee, A. Bazylak, J. Power Sources 2019, 437, 226910.C. Immerz, B. Bensmann, P. Trinke, M. Suermann, R. Hanke‐Rauschenbach, J. Electrochem. Soc. 2018, 165, F1292.X. Peng, P. Satjaritanun, Z. Taie, L. Wiles, A. Keane, C. Capuano, I. V. Zenyuk, N. Danilovic, Adv. Sci. 2021, 8, 2102950.R. Omrani, B. Shabani, Int. J. Hydrogen Energy 2017, 42, 28515.Q. Feng, X. Yuan, G. Liu, B. Wei, Z. Zhang, H. Li, H. Wang, J. Power Sources 2017, 366, 33.F. N. Khatib, T. Wilberforce, O. Ijaodola, E. Ogungbemi, Z. El‐Hassan, A. Durrant, J. Thompson, A. G. Olabi, Renewable Sustainable Energy Rev. 2019, 111, 1.M. K. Debe, S. M. Hendricks, G. D. Vernstrom, M. Meyers, M. Brostrom, M. Stephens, Q. Chan, J. Willey, M. Hamden, C. K. Mittelsteadt, C. B. Capuano, K. E. Ayers, E. B. Anderson, J. Electrochem. Soc. 2012, 159, K165.A. Albert, T. Lochner, T. J. Schmidt, L. Gubler, ACS Appl. Mater. Interfaces 2016, 8, 15297.L. Placca, R. Kouta, Int. J. Hydrogen Energy 2011, 36, 12393.T. Matsuura, M. Kato, M. Hori, J. Power Sources 2006, 161, 74.V. Radhakrishnan, P. Haridoss, Int. J. Hydrogen Energy 2010, 35, 11107.O. Ronneberger, P. Fischer, T. Brox, in Medical Image Computing and Computer‐Assisted Intervention – MICCAI 2015 (Eds.: N. Navab, J. Hornegger, W. M. Wells, A. F. Frangi), Springer International Publishing, Cham 2015, p. 234.Fuel Cell Store: Unit B College Station, TX. Nafion™ PFSA 5% Dispersions D520: SDS Number: 1334079‐00035 [Online], https://www.fuelcellstore.com/msds-sheets/nafion-dispersion-d520-sds.pdf (accessed: March 2022).S. Ghosh, H. Ohashi, H. Tabata, Y. Hashimasa, T. Yamaguchi, Int. J. Hydrogen Energy 2015, 40, 15663.S. Thiele, R. Zengerle, C. Ziegler, Nano Res. 2011, 4, 849.L. Zielke, S. Vierrath, R. Moroni, A. Mondon, R. Zengerle, S. Thiele, RSC Adv. 2016, 6, 80700.S. Thiele, T. Fürstenhaupt, D. Banham, T. Hutzenlaub, V. Birss, C. Ziegler, R. Zengerle, J. Power Sources 2013, 228, 185.F. Rong, C. Huang, Z.‐S. Liu, D. Song, Q. Wang, J. Power Sources 2008, 175, 699.K. K. Poornesh, C. Cho, D.‐Y. Kim, Y. Tak, Energy 2010, 35, 5241.W. Li, X. Bi, M. Luo, P.‐C. Sui, J. Electrochem. Soc. 2021, 168, 034502.G. S. Lonakar, M. S. Mahajan, S. S. Ghosh, J. V. Sali, Org. Electron. 2012, 13, 2575.H. Liu, L. Ney, N. Zamel, X. Li, Appl. Sci. 2022, 12, 3776.K. Roger, E. Sparr, H. Wennerström, Phys. Chem. Chem. Phys. 2018, 20, 10430.M. Suermann, T. J. Schmidt, F. N. Büchi, Electrochim. Acta 2016, 211, 989.Tables of X‐Ray Mass Attenuation Coefficients and Mass Energy‐Absorption Coefficients, NIST Standard Reference Database, https://www.physics.nist.gov/PhysRefData/XrayMassCoef/cover.html (accessed: January 1995).M. E. Wieser, N. Holden, T. B. Coplen, J. K. Böhlke, M. Berglund, W. A. Brand, P. de Bièvre, M. Gröning, R. D. Loss, J. Meija, T. Hirata, T. Prohaska, R. Schoenberg, G. O'Connor, T. Walczyk, S. Yoneda, X.‐K. Zhu, Pure Appl. Chem. 2013, 85, 1047.P. Perona, J. Malik, IEEE Trans. Pattern Anal. Mach. Intell. 1990, 12, 629.G. Gerig, O. Kubler, R. Kikinis, F. A. Jolesz, IEEE Trans. on Med. Imaging 1992, 11, 221.Anisotropic Diffusion (Perona & Malik), https://www.mathworks.com/matlabcentral/fileexchange/14995-anisotropic-diffusion-perona-malik (accessed: March 2022).F. A. Merchant, in Microscope Image Processing (Eds.: Q. Wu, F. A. Merchant, K. R. Castleman), Academic Press, Burlington 2008, Ch. 14.S. Preibisch, S. Saalfeld, P. Tomancak, Bioinformatics 2009, 25, 1463.S. Wright, J. Agric. Res. 1921, 20, 557.D. Chicco, M. J. Warrens, G. Jurman, PeerJ Comput. Sci. 2021, 7, e623. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Advanced Energy Materials Wiley

Toward Understanding Catalyst Layer Deposition Processes and Distribution in Anodic Porous Transport Electrodes in Proton Exchange Membrane Water Electrolyzers

Download PDF
16 pages

Loading next page...
 
/lp/wiley/toward-understanding-catalyst-layer-deposition-processes-and-xPqZFTjJTH

References (54)

Publisher
Wiley
Copyright
© 2023 Wiley‐VCH GmbH
ISSN
1614-6832
eISSN
1614-6840
DOI
10.1002/aenm.202203636
Publisher site
See Article on Publisher Site

Abstract

IntroductionHydrogen from water electrolysis will play a key role as a clean and sustainable energy storage solution in the upcoming decades in the context of a decarbonized economy.[1] According to the study “Hydrogen Roadmap Europe: A sustainable pathway for the European Energy Transition”[2] of the fuel cells and hydrogen joint undertaking of the year 2019, nearly one‐quarter of the total energy demand in the European Union could be covered with hydrogen by 2050. So‐called proton exchange membrane water electrolyzers (PEMWEs) will be one of this transition's mainstays. They are already available at the multi‐MW scale and currently approaching the GW scale.[3,4] Nevertheless, it is still necessary to decrease their price per kWh for the proliferation of PEMWEs. This economization must be accompanied by a simultaneous increase in their performance, efficiency, and durability.[5,6]The centerpiece of a PEMWE is the membrane electrode assembly (MEA). It consists of an anode and a cathode, each with a porous transport layer (PTL) and a catalyst layer (CL), and a solid polymer electrolyte membrane in between. In principle, the thin film electrodes of the MEAs can be processed as catalyst‐coated membranes (CCMs) or as catalyst‐coated substrates.[7,8] In the former case, the catalyst ink is either coated directly (e.g., via an ink‐spray‐printer or a roll‐to‐roll process)[9] or indirectly onto the membrane via the so‐called decal transfer method. In the other case, the catalyst ink is coated onto the PTL. Afterward, the membrane is sandwiched between the anodic and the cathodic porous transport electrode (PTE) to obtain an MEA.Although CCMs are the state‐of‐the‐art MEA manufacturing technique for PEMWE, PTEs could be a promising alternative concept for MEA structures due to a different fabrication approach and the related advantages. A technological advantage of PTEs is the reduction of working steps in the manufacturing process compared to the CCM approach by directly processing the catalyst layer onto the substrate, which should lower manufacturing costs. Moreover, the PTE setup allows a higher variety of membranes in the MEA composition, for example, if the glass‐transition temperature of the membrane exceeds the decomposition temperature at the decal transfer.[10] Fluctuating temperature and humidity can change the degree of membrane swelling. As a consequence of the decoupling of the membrane and the catalyst layers, the PTE configuration reduces the deformation of the catalyst layers in comparison to CCMs. On the one hand, the direct spray coating on the well‐conducting titanium also enhances the PTE configuration's electronic conductivity compared to CCMs. On the other hand, the ionic conductivity worsens because parts of the catalyst layer are no longer connected in the ionomer‐membrane network.[11] Furthermore, the concept of PTEs has been successfully implemented also in other electrolyzer systems, like anion exchange[12] and bipolar membrane water electrolyzers.[13,14] Likewise, PTE manufacturing has already been successfully implemented on an industrial level. For example, Toshiba Corporation recently developed a new sputtering technology for the manufacturing of PTEs for large‐scale production.[15,16] In general, the catalyst layer can be either directly coated[17–19] or electrochemically deposited.[20–22] Bühler et al. showed that the performance of PTEs fabricated via spray coating on titanium fibers is similar to CCM with an identical loading of 1.5 mg cm−2 iridium oxide, indeed with worse kinetics, but a tendency to improve mass transport properties.[18] However, Leonard et al.[11] observed a 200 mV worse performance of a sintered titanium PTE at 1 A cm−2 compared to a CCM in a micro‐computed tomography (Micro‐CT) setup capable of measuring operando X‐ray computed tomography (CT) and radiography. They traced the performance loss to a lower connection between catalyst particles and membrane. The anodic electrode had a loading of 3 mg cm−2 iridium oxide ultrasonic spray‐coated onto a sintered titanium paper. Based on their previous publication,[11] the same group performed a systematic study on the influence of the interface between catalyst and PTL for a CCM setup and between catalyst and membrane for a PTE setup.[23] They used sintered (mean pore radius 13.31 µm) and fiber Ti PTLs (mean pore radius 8.07 µm) and varied the iridium oxide loading between 0.5 mg cm−2 and 2.2 mg cm−2. They showed that PTEs perform worse than CCMs at the same catalyst loading and using identical PTLs. Sintered and fiber Ti PTEs had a similar performance at high loadings. Cloe et al. demonstrated that electrochemically deposited iridium oxide on a titanium mesh showed not only a high current density of 0.91 A cm−2 at 1.6 V, but also a protective function against titanium corrosion.[20]Whereas CCMs and PTEs show comparable electrochemical data, their structures, especially the catalyst distribution, deviate from each other due to their different manufacturing techniques (see Figure 1). The catalyst layer of a CCM can be described as a flat homogeneous layer with a nearly constant thickness (see Figure 1c,d). In contrast, for the PTEs, the morphology of the catalyst layer and the structure of the PTL are directly connected because of the directly coated catalyst on the titanium PTL (see Figure 1a,b).1FigureIllustration of differences in the catalyst layer distribution between CCMs and PTEs. a) Surface morphology and b) embedded cross‐section of IrO2‐coated titanium fibers. c) Surface image of the anode catalyst layer of a CCM. d) Embedded cross‐sectional image of a CCM. The lower white stripe is the anode catalyst layer, and the upper is the cathode. A Nafion 117 membrane is sandwiched in between the layers. Epoxy resin and PTFE spacers from embedding are visible above and below the catalyst layers.To the best of our knowledge, no literature exists about a complete quantitative determination of catalyst distribution in anodic porous transport electrodes in PEMWEs. Leonard et al.[11] and Kulkarni et al.[23] imaged in operando PTE PEMWEs via X‐ray micro‐tomography. Due to the resolution limit of above 1 µm of the X‐ray CT imaging, they did not fully resolve the catalyst distribution of the electrodes. Nevertheless, they resolved a majority of the catalyst at different loadings. Thereby, they demonstrated that the catalyst layer follows the morphology of the titanium fibers (fiber diameter of ≈20 µm). With increasing loading, the catalyst covers the top of the fibers and the curved surface area and underlying fibers. They determined the triple‐phase contact area (TPCA) using the tomographies. The TPCA is the area where the catalyst layer overlaps with the membrane and the PTL after an interfacial projection.[11] In general, the TPCA increases with higher loadings, and the TPCA of titanium fiber PTEs is higher than the TPCA of sinter titanium PTEs.[23] Furthermore, electron micrographs of PTEs are found in previous literature reports.[13,18] These images can only convey a qualitative impression of the morphology of respective PTEs. As mentioned above, the catalyst distribution of a pristine CCM is trivial. The catalyst layer can be simply described by its thickness. The underlying nanoporous structure of a PEMWE anode catalyst layer can be analyzed via FIB‐SEM tomography.[24]Numerous works have been published on porous transport layers in CCM setups and their structure.[25,26] In general, the PTL is responsible for transporting water and oxygen and establishing electrical contact with the catalyst. Titanium is mainly chosen as the PTL material due to its good electric conductivity and non‐corrosive bulk properties, which are built up due to a passivating oxide layer under the operating conditions of a PEMWE.[27] There are different types of Ti PTLs, like sintered titanium powders[28,29] or sintered titanium fibers.[29,30] They vary additionally in their structural properties, for example, porosity, thickness, or pore diameter. Overall, the choice of a PTL is always a tradeoff between mass transport, electrical conductivity, and connectivity of the catalyst layer and the PTL. Therefore several studies were performed in the last years to investigate the influence of the structure of a PTL on the electrochemical performance in CCM setups.[25,31–33] For instance, Weber et al.[34] suggested an optimum ratio of 0.5 between PTL thickness and flow field rib width. Furthermore, the interface between PTL and the catalyst layer has been the subject of several studies. The interfacial properties influence the mass transport resistance and the high‐frequency resistance (HFR).[28,35,36] Schuler et al.,[30,35] for example, investigated morphology, topology, and correlated performance of titanium fiber meshes‐based PTLs with different fiber diameters and porosities in a CCM‐based setup. They concluded that a higher interfacial contact area, including a higher catalyst utilization and a lower interfacial resistance, correlates with better performance. One similar titanium fiber mesh (2GDL40‐1.0, Bekaert) is also used in our study. With this knowledge, Schuler et al. designed PTLs with a porous backing layer which increased the catalyst layer utilization and improved mass transport properties.[31]However, the insights from the numerous investigations of PTLs in CCM setups cannot be directly translated into the PTE approach. To improve and optimize the anodic PTEs, it is necessary to understand their morphology, especially the interplay between PTL and the catalyst layer. The publications of the Zenyuk group[11,23] already visualized parts of the catalyst distribution and determined the TPCA for PTEs. In our paper, the main focus is on the through‐plane catalyst distribution. The characterizing system and manufacturing way follows the titanium fiber mesh aPTE approach of Bühler et al., who used spray coating as a deposition technique.[18] Based on this publication,[18] a PTE with a catalyst loading of 1.5 mg cm−2 iridium oxide and a binder content of ≈10 wt.% was chosen as a system to be examined because of its good reproducibility and performance. The catalyst distribution of this coated electrode is determined by the analysis of cross‐sectional images of embedded samples and confirmed via Micro‐CT analysis. Furthermore, the previous two methods give deeper insights into the structural changes occurring in the PTE design regarding surface roughness, volume fraction gradients, and catalyst layer thickness. Finally, the results are implemented in an empirical spray coating model. This model allows us to understand better the influence of the manufacturing processes on the catalyst layer morphology and thus to make predictions for future optimized PTE designs.Results and DiscussionElectrochemical DataIn the following, we verify the ionomer content of the manufactured PTE. Additionally, we present the electrochemical data and compare them to the literature values.ManufacturingThe ionomer content of the spray‐coated PTE was determined to be 10.7 wt.%, which matched the intended ionomer content of ≈10 wt.%. With this ionomer content, a good tradeoff between reproducibility and performance can be achieved, according to Bühler et al.[18] The corresponding thermogravimetric analysis (TGA) can be found in the supplementary information (see Figure S1, Supporting Information).Electrochemical AnalysisThree samples were electrochemically characterized according to the protocol described in the experimental part. The average and the standard deviations of the polarization curves are displayed in Figure 2. The three samples showed similar behavior. The highest current densities were reached at ≈3 A cm−2 because the safety shutoff limit of 2.3 V was reached at the respective galvanostatic electrochemical impedance spectroscopy (EIS) measurements. Additionally, the HFR was determined via galvanostatic EIS (see Figure S2, Supporting Information). The HFR‐free cell voltage was calculated with the knowledge of this resistance (see Figure 2). The obtained electrochemical data conform to the results of Bühler et al., who used an identical MEA configuration but a different testing setup.[18] The HFR‐free cell voltage at 1.5 A cm−2 was at 1.606 ± 0.016 V, which lies between the voltages Bühler et al. measured for the aPTEs with 9 wt.% (1.59 V) and 12 wt.% (1.61 V) ionomer content.[18] The HFR at higher current densities is also in a similar range as in Bühler et al.[18] (between 0.19 and 0.20 Ω cm2).2FigureMean polarization curve with the standard deviations of the three measured aPTE, including the measured cell voltage and the determined HFR‐free cell voltage in a current density range from 5 mA cm−2 to 2.9 A cm−2.Structural AnalysisIn this section, we will discuss the structure of the manufactured PTEs. We used PTEs with a size of 5 mm × 5 mm for the structural analysis. We analyze their structure, draw conclusions about the catalyst layer thickness and calculate the surface roughness.PTE StructureThe complete tomogram of the PTE, received by the segmentation of the Micro‐CT dataset (volume of ≈3.00 × 4.25 × 1.16 mm3 with a small void volume at the bottom and top part in the z‐direction), can be seen in Figure 3a. The iridium oxide catalyst layer and the titanium fibers are not distinguished. The catalyst layer is on top of the segmented volume (close to the z = 0 plane in Figure 3a). Some big ink droplets can be observed at the surface. By analyzing the through‐plane volume fraction gradient (see Figure 3b), the spray coating process raises the solid volume content by ≈7% compared to the mean average of the plain PTL. The structural and transport parameters of the titanium fibers are determined by cropping the segmented PTE to a core part with a volume of 3.00 × 4.25 × 0.75 mm3 to only consider the titanium fibers not affected by the spray coating process. Therefore no catalyst covers the considered fibers, and edge effects can be neglected. The solid volume content of this section is φbulk = 49.16%, corresponding to a porosity of ρ = 50.84%. A representative volume analysis of the bulk volume confirms the global validity of the porosity value (see Figure S3, Supporting Information). The pore size distribution is displayed in the supporting information (see Figure S4, Supporting Information). Mean and median values are slightly above 30 µm. The applied fiber find algorithm's result is illustrated in Figure 3c. The algorithm was capable of allocating the fibers, but not as a whole, just piecewise. The fibers have a diameter of 20.59 ± 3.15 µm (see Figure 3d). They are uniformly orientated in the in‐plane direction. However, they have a strong horizontal alignment in the through‐plane direction (see Figure 3d and fiber orientation tensor in Table S1, Supporting Information).3Figurea) Segmented PTE of the Micro‐CT data. The top side shows the coated titanium fibers. b) Through‐plane trend of the volume fraction (along the z‐direction) and resultant determination of the minimum catalyst layer thickness or coating dependence on the through‐plane position. c) Bulk‐section of the PTE, which is equivalent to a titanium fiber PTL, with the identified fiber fragments. d) Fiber diameter and in‐plane and through‐plane fiber orientation distribution of the previous bulk volume.The structural parameters, porosity, and mean fiber diameter, deviate slightly from the manufacturer's data (ρManufacturer = 56% and dManufacturer = 20 µm). Likewise, literature values also differ from the specified parameters of the same manufacturer. Schuler et al. determined for a similar PTL (nominal ρManufacturer = 56% and dManufacturer = 22 µm) a porosity of 54% and a mean diameter of 15.4 µm.[30] Immerz et al. stated the porosity as 50% of a Bekaert PTL with the product name 2GDL40‐1,00.[37] The reasons for this deviation can be diverse. For instance, the differences can be due to the choice of the analyzing method and the associated limitations, for example, the resolution limit of the Micro‐CT or voxel size. Also, variations in the manufacturing process of the fiber felts can be a reason. For instance, minor variances in the fiber diameter significantly impact the porosity: Assuming a diameter of 20 µm instead of the here‐determined diameter, the porosity would already increase by 3% (ρ20 µm  = ρMicro − CT rMicro − CT2/rManufacturer2≈54%${\rho _{20\;\mu {\rm{m}}\;}}\; = \;{\rho _{{\rm{Micro}}\; - \;{\rm{CT}}}}\;r_{{\rm{Micro}}\; - \;{\rm{CT}}}^2{\rm{/}}r_{{\rm{Manufacturer}}}^2 \approx 54\). A more precise way to resolve the fibers would be a Nano‐CT. However, a higher resolution would lead to a smaller segmented volume and thus sample size. The chosen region of the segmentation influences the porosity. If the segmented volume contains edge areas of the PTL, the overall porosity will increase (see Figure 3b). In addition, variations in the manufacturing process cannot be excluded. Therefore samples from different batches would need to be examined for statistical reasons.Catalyst Layer ThicknessThe volume fraction of the bulk area of the PTE Micro‐CT data is very homogeneous and lies between 48% and 50% (see Figure 3b). However, an increase in the solid content can be found close to the top of the PTE (relative to z = 0, compare Figure 3a,b). This peak in volume fraction can be ascribed to the catalyst layer covering the titanium fibers. Consequently, the catalyst layer thickness can be estimated from the above results. We propose two models for the calculation of the catalyst layer thickness. One assumes the fiber is fully covered, and the other assumes the fiber is half covered (see Figure 3b).In the first model, it is assumed that the catalyst covers the whole fiber with a homogeneous layer and the fibers possess a cylindrical shape (see Figure 3b). We define volume elements Vi,complete, specified by the product of the total in‐plane area and the through‐plane height of a voxel length of 2.5 µm. Each volume element Vi,complete has mi fibers of length li and a total radius composed of the sum of the titanium fiber radius ri and the catalyst layer thickness ci,complete:1Vi,complete = mi π(ri + ci,complete)2li\[\begin{array}{*{20}{c}}{{V_{i,{\rm{complete}}}}\; = \;{m_i}\;\pi {{({r_i}\; + \;{c_{i,{\rm{complete}}}})}^2}{l_i}}\end{array}\]In the second model, only the upper half of the cylindrical‐shaped fiber is homogeneously coated with a constant catalyst layer thickness ci,half (see Figure 3b). The i‐th volume element Vi,half can be calculated as:2Vi,half = 1/2mi π(ri + ci,half)2li + 1/2miπ(ri)2li\[\begin{array}{*{20}{c}}{{V_{i,{\rm{half}}}}\; = \;1{\rm{/}}2{m_i}\;\pi {{({r_i}\; + \;{c_{i,{\rm{half}}}})}^2}{l_i}\; + \;1{\rm{/}}2{m_i}\pi {{({r_i})}^2}{l_i}}\end{array}\]For both models, the average bulk volume of the PTL without catalyst with the same height, n fibers of the length lbulk, the titanium fiber radius rbulk can be expressed as:3Vbulk =  nπrbulk2lbulk\[\begin{array}{*{20}{c}}{{V_{{\rm{bulk}}}}\; = \;\;n\pi r_{{\rm{bulk}}}^2{l_{{\rm{bulk}}}}}\end{array}\]Assuming now the same numbers of fibers per volume element (n = mi), the same fiber length ( lbulk = li = l) and titanium fiber radius ( rbulk = ri = r) the minimum catalyst layer thickness ci,complete, min and ci,half, min of the i‐th volume element can be calculated:4ci,complete, min = (Vi, completeVbulk − 1) r  = (φiφbulk − 1) r\[\begin{array}{*{20}{c}}{{c_{i,{\rm{complete}},\;\min }}\; = \;\left( {\sqrt {\frac{{{V_{i,\;{\rm{complete}}}}}}{{{V_{{\rm{bulk}}}}}}} \; - \;1} \right)\;r\;\; = \;\left( {\sqrt {\frac{{{\varphi _i}}}{{{\varphi _{{\rm{bulk}}}}}}} \; - \;1} \right)\;r}\end{array}\]5ci,half, min = (2Vi, halfVbulk − 1  − 1) r  = (2φiφbulk − 1 − 1) r\[\begin{array}{*{20}{c}}{{c_{i,{\rm{half}},\;\min }}\; = \;\left( {\sqrt {\frac{{2{V_{i,\;{\rm{half}}}}}}{{{V_{{\rm{bulk}}}}}}\; - \;1\;} \; - \;1} \right)\;r\;\; = \;\left( {\sqrt {\frac{{2{\varphi _i}}}{{{\varphi _{{\rm{bulk}}}}}}\; - \;1} \; - \;1} \right)\;r}\end{array}\]φi is here the solid volume content of the i‐th volume element and can be extracted from the Figure 3b dataset. The calculated results are displayed in Figure 3b. The estimated minimum of the fully covered fiber model is given by the peak in Figure 3b and is accordingly ≈0.66 µm. The estimated minimum of the half‐covered fiber model is ≈1.29 µm and is reached close to the surface of the PTE and strongly declines within the next 100 µm.The simplification that the titanium fibers are homogeneously covered with catalyst doesn't correspond to reality. The majority of the catalyst is concentrated on the top part of the titanium fibers (see Figure 4a; Figure S5, Supporting Information). This observation corresponds with FIB‐SEM (focused ion beam scanning electron microscopy) cross‐sections of other PTEs manufactured.[13,14,18,19] However, some fully covered fibers can also be observed, mostly present due to clogging of the catalyst ink occurring at some narrow pores within the titanium fiber system. Therefore, the second model, which only assumes that half of the fiber is homogeneously coated, is closer to reality but with some limitations (e.g., partially inhomogeneous coating and partially fully covered fibers due to clogging). Generally, the derived catalyst layer thickness should only be considered as a lower bound catalyst layer thickness of the top part of the titanium fibers. Especially the simplification that the fiber density is equal at the bulk and the edge area probably does not match with real PTLs. The difference in fiber density is highlighted in the publication of Peng et al.,[38] where the through‐plane porosity profiles of several Bekaert PTLs showed a strong increase in porosity at the edges of the substrate. Consequently, the mean catalyst layer thickness of our PTE would be higher. Nevertheless, the estimation is in good agreement with literature data: Bühler et al. showed a FIB‐SEM cross‐sectional image of a similar PTE (same recipe, only with 12 wt.% Nafion and a loading of 1.4 mg cm−2 iridium) with a catalyst layer thickness of ≈2 µm.[18] The images of Bühler et al. correspond well to our minimum catalyst layer thickness of 1.3 µm at the edge area, considering the rough surface structure of the PTEs (see Figure 1a). Additionally, Kulkarni et al.[23] resolved a majority of the catalyst layer of PTEs with different loadings with a Micro‐CT with a voxel size between 1.16 and 1.73 µm. Therefore the minimum catalyst layer thickness should be close to the resolution of their tomographic data. Furthermore, this group also observed a covering of the entire top area and a curved surface area of the fiber. This strengthens the assumption of the half‐coated fibers. Nevertheless, our simplified assumptions are not qualified to reproduce the catalyst layer thickness and distribution in real PTEs. Therefore, more precise imaging and tomography techniques are necessary.4FigureDetermining the catalyst distribution via embedded cross‐sectioning. a) A segment of one stitched BSD image. In total, three cross‐sections, each with a total image width of several mm, were analyzed to obtain a reliable catalyst distribution. b) Segmentation of the image in (a). c) Normalized catalyst distributions of the three different embedded cross‐sections and their weighted average. Additionally, a lognormal distribution was fitted to the average catalyst distribution.The volume fraction gradient also leads to a local pore gradient. From the literature, it is known that an increasing porosity gradient from CCM to the flow field improves the reacting transport, and additionally, less oxygen saturation is observed.[36,39] This could be an additional reason for the better mass transport properties in comparison to CCMs Bühler et al.[18] discovered. The higher volume fraction could also reduce ohmic losses because the additional connection through the catalyst layer could increase the electrical conductivity.Surface RoughnessThe spray coating of the titanium fibers affects the structure and the surface roughness in two ways. On the one hand, void space on the surface shrinks or is even closed due to the deposition, which could reduce surface roughness. On the other hand, the coating alters the surface of the titanium fibers, for example, big catalyst droplets can occur that raise the surface roughness. Additionally, the surface roughness increases due to the porous catalyst layer itself. This effect happens on the nm scale and, therefore, cannot be detected by the Micro‐CT.Here, the surface roughness parameters of the catalyst‐coated PTE and its non‐coated backside were calculated based on the Micro‐CT data (see Table 1 and Supporting Information) to compare a PTE with an uncoated PTL. The backside of the PTE can be considered equal to an uncoated PTL since the PTL does not show anisotropy in the through‐plane direction. Further, the majority of the deposited catalyst can be found within 100 µm from the surface while the full PTL has a thickness of 1 mm (see Figure 3b). Consequently, the uncoated backside of the PTE resembles a pristine PTL and can therefore act as a reference. The surface roughness profile height maps (see Figure 5) visualize these surface roughness parameters. The visible fibers of the PTE roughness profile height map have a larger diameter than the pure titanium fibers. Furthermore, some big ink droplets on the surface of the PTE can be detected (see Figure 5b: dark circles). These droplet hills can have a height of up to 50 µm. Therefore, the coated fibers have a higher profile height. This visual observation is consistent with the mean surface roughness parameter: the mean surface roughness parameter of the PTE is more than twice the value of the PTL. Whereas the arithmetic mean and root mean square surface roughness are nearly the same. When we chose a surface area without big catalyst droplets for the calculations (see Figure 5b gray outlined area and Figure 5c), the surface roughness parameters of PTE and PTL are similar.1TableCalculated surface roughness parameters of the PTL, specifically the PTE backside, the catalyst‐coated PTE, and a reduced section of the PTE, where no big catalyst droplet can be found. The table includes the mean surface roughness RmPTL/PTE$R_{\rm{m}}^{{\rm{PTL}}/{\rm{PTE}}}$, the arithmetic mean surface roughness RaPTL/PTE$R_{\rm{a}}^{{\rm{PTL}}/{\rm{PTE}}}$, and the root mean square roughness RqPTL/PTE$R_{\rm{q}}^{{\rm{PTL}}/{\rm{PTE}}}$Surface roughness parameters [µm]PTLPTEPTE‐section (gray outlined area)RmPTL/PTE$R_{\rm{m}}^{{\rm{PTL}}/{\rm{PTE}}}$42.486.748.7RaPTL/PTE$R_{\rm{a}}^{{\rm{PTL}}/{\rm{PTE}}}$26.226.426.0RqPTL/PTE\[R_{\rm{q}}^{{\rm{PTL}}/{\rm{PTE}}}\]36.236.435.95FigureRoughness profile height maps of the PTE backside (a) and the spray‐coated front side (b). The gray outlined section (b) is an area without any big catalyst droplets and is displayed separately in (c). The profile height color bar is depicted beside (c).Indeed, the mean surface roughness of the PTL is in good agreement with the literature values from Schuler et al.[30] They postulated a linear relationship between fiber diameter and mean surface roughness based on their experimental results:6mean surface roughness (µm)=2.1×fiber diameter(µm)−3 µm\[\begin{array}{*{20}{c}}{{\rm{mean}}\;{\rm{surface}}\;{\rm{roughness}}\;\left( {\mu {\rm{m}}} \right) = 2.1 \times {\rm{fiber}}\;{\rm{diameter}}\left( {\mu {\rm{m}}} \right) - 3\;\mu {\rm{m}}}\end{array}\]Using our determined fiber diameter, the mean surface roughness is 40.24 µm according to Equation 6, close to the determined value of 42.4 µm. Therefore, our calculated surface roughness is in good agreement with literature values, especially considering the broad fiber diameter distribution. Possible error sources could still be a variation in the manufacturing process, missing statistics, or a not perfectly aligned Micro‐CT.As observed above, in the absence of big dried catalyst droplets, the mean and the arithmetic mean surface roughness are very similar (see Figure 5, Table 1). The presence of this up to 50 µm high droplets may cause a mechanical degradation of the membrane caused by puncture and mechanical stress. In general, in operation times under 1000 h, mechanical stress is one of the main durability problems and causes of failures of PEMWEs.[40–42] In particular, pinholes can lead to short‐circuits and therefore reduce the efficiency of PEMWEs or even lead to their failure.[43,44]Additionally, from PEM fuel cells (PEMFC), it is known that the roughness of a gas diffusion layer (GDL, which is the equivalent of a PTL) can strongly affect the lifetime of a membrane.[45,46] Additionally, a rougher GDL increases the open‐circuit voltage and the hydrogen crossover in PEMFCs.[45]For this reason, the manufacturing process of the PTEs should be further optimized to avoid big catalyst droplets and thereby reduce the mechanical stress on the membrane.Catalyst Distribution and UtilizationCatalyst Distribution via Embedded Cross‐SectionThe cross‐sectional imaging of embedded samples allowed us to get more profound information about the catalyst layer and its distribution. In total, three datasets of the embedded cross‐section were taken. They resulted in total image width of 11.41 mm (catalyst distribution 1: 4.11 mm, catalyst distribution 2: 3.16 mm, and catalyst distribution 3: 4.14 mm). A segment of the stitched BSD image (see Figure S5, Supporting Information) can be seen in Figure 4a. Here, pore (black), fiber (grey), and catalyst (white) can be clearly distinguished with the applied imaging conditions. Additionally, the catalyst regions were verified via energy‐dispersive X‐ray spectroscopy measurements (see Figure S6, Supporting Information). The 2D catalyst regions were clearly segmented with the AI‐based algorithm Unet 2D[47] implemented in GeoDict (see Figure 4b). A through‐plane catalyst gradient and fibers surrounded by a catalyst were observed. Additionally, some catalyst agglomerations, probably due to clogging caused by pore throats, were detected. The amount of catalyst found on the backside of fibers was small. No average catalyst layer thickness was determined due to the 2D nature of the data. Overall, the embedded PTE's cross‐sectional images show a comparable catalyst distribution to the samples of the publication of Mayerhöfer et al.[13]The through‐plane catalyst gradient can be confirmed by plotting the normalized catalyst amount dependent on the penetration depth (see Figure 4c). Based on this data, we obtained a 1D catalyst distribution along the z‐direction (through‐plane direction). Whereas the single data set catalyst distributions are rough, averaging the three data sets smooths the catalyst distribution. These peaks in the single data sets occurred (see Figure 4c catalyst distribution 1) if a certain imaging plane contained long horizontal fibers with a catalyst layer on top. The related count of pixels at this penetration depth is consequently overproportioned and causes outliers in the distribution. This effect is reduced by taking the average of these three distributions (see Figure 4c). By normalizing the three distributions, the similarity of the distributions becomes visible (see Figure 4c). The three comparable distributions suggest the statistical correctness and representative character of this data. However, the catalyst distributions could be distorted by imaging artifacts, segmentation errors, statistical deviations, a crooked alignment of the stitched images, or damaging the catalyst layer during the grinding process.The average catalyst distribution is similar to a lognormal distribution. The according fit is also plotted in Figure 4c (see Table S3, Supporting Information for the fitting parameter). A coefficient of determination R2 > 97% confirms the lognormal behavior of the data. The penetration depth of the catalyst can be quantified by the P(0.90)/P(0.95) interval value. The distribution revealed that 90%/95% of the catalyst had a penetration depth of 98.5 µm/122 µm (see Table 2).2TableP(90) and P(95)‐interval of all catalyst distributions. The R2 value of the BSD and Micro‐CT catalyst distribution relates to the respective experimental data. The R2‐value of each model catalyst distribution is based on the reference lognormal distribution of the embedded sampleMethodBSDMicro‐CTModel 1Model 2Model 3Model 4Penetration depth P(0.90) interval [µm]98.51098087.595100Penetration depth P(0.95) interval [µm]122126105112.5120122.5R20.97360.9840−1.02070.70820.93610.9607Catalyst Distribution via Micro‐CTThe Micro‐CT data set was also used to determine the catalyst distribution. First of all, in each image slice of the Micro‐CT, a brighter region at the edge exists. This is attributed to the iridium oxide catalyst layer (see Figure 6a). The differences in grayscale are due to the different mass attenuation coefficients of titanium and IrO2 at the applied operation voltage and current of the Micro‐CT. The brightness and the catalyst layer become clearer by applying a maximum intensity projection on the Micro‐CT data set. The maximum intensity protection results in an imaging plane with the brightest pixels in the through‐plane direction of every point. The catalyst layer extends over 100 µm into the PTL (see Figure 6b). The through‐plane behavior of the mean grey value clarifies this observation (see Figure 6c). The baseline of the curve is determined at a grey value of 79. The first 200 µm after the intersection between the baseline and mean grey value curves are plotted magnified (see Figure 6c). The P(0.90) interval value of this section is ≈109 µm (see Table 2). The decay of the mean grey value was additionally fitted with a lognormal distribution (R2‐value of over 98%). The Micro‐CT gradient has its origin on the one hand, through the higher volume content, on the other hand, due to the higher attenuation of iridium. The little intensity peak of the through‐plane gradient on the edge of the PTL side (see Figure 6c) suggests that some beam‐hardening effects are present.6Figurea) Single slice of the raw Micro‐CT data. The brighter pixel on the left side can be attributed to the catalyst layer. b) Maximum intensity projection of the Micro‐CT data. This projection highlights the catalyst layer and makes the penetration depth of more than 100 µm visible. c) Through‐plane plot of the mean grey value of the Micro‐CT. The blue section that starts when the mean grey value exceeds the baseline is magnified to get information about the catalyst penetration depth.Comparison of Catalyst DistributionsThe P(0.90) and P(0.95) intervals are very similar (see Table 2). To compare the catalyst distribution of the embedded cross‐sectional imaging and the Micro‐CT more quantitatively, the respective data was normalized that the fitted lognormal distribution had its peak at 1.0 (see Figure 7a, the fitting parameters can be found in Table S3, Supporting Information). Both data sets have a similar trend. The Micro‐CT data is slightly above the embedded cross‐sectional image curve. That is either explained by missing statistics or probably by a not completely correct one‐to‐one translation of the mean grey value to the catalyst distribution. Nevertheless, the Micro‐CT data is qualified to gain inside into the catalyst distribution. For a quantitative usage of the Micro‐CT for the catalyst distribution, more experiments with different catalyst loadings and PTLs with different porosities would be necessary to calibrate the results.7Figurea) Direct comparison between the catalyst distribution and the according lognormal fits of the data received via the embedded cross‐section and the Micro‐CT. b) Catalyst utilization, in our publication defined as the ionically well‐connected catalyst, as a function of the penetration depth of the membrane. The catalyst utilization is calculated as the cumulative sum of the catalyst distribution. The maximum penetration depth is taken from the literature.[30]The above‐calculated catalyst distributions are only relative. The total amount or thickness of the catalyst layer is assessed in three ways: by the TGA and the catalyst loading, by the segmented BSD images, or by the through‐plane trend of the volume fraction.The first method calculates the thickness by considering the loading IrO2 (1.5 mg cm−2), the determined weight ratios of the binder and IrO2 by TGA (YIonomer = 0.1071,${Y_{{\rm{Ionomer}}}}\; = \;0.1071,$YIrO2=  0.8929)${Y_{{\rm{Ir}}{{\rm{O}}_2}}} = \;\;0.8929)$, the density of IrO2 (11.7 g cm−3) and Nafion D520 (0.92–0.94 g cm−3),[48] and an assumption of the porosity of the catalyst layer based on PEMFC and PEMWE literature values[24,49–52] (εCL = 0.5 ±  0.1):7tTGA=(loadingIrO2ρIrO2+loadingIrO2×YIonomer/YIrO2ρNafion D520)/εCL=6.44µm\[\begin{array}{*{20}{c}}{{t_{{\rm{TGA}}}} = \left( {\frac{{{\rm{loadin}}{{\rm{g}}_{{\rm{Ir}}{{\rm{O}}_2}}}}}{{{\rho _{{\rm{Ir}}{{\rm{O}}_2}}}}} + \frac{{{\rm{loadin}}{{\rm{g}}_{{\rm{Ir}}{{\rm{O}}_2}}} \times {Y_{{\rm{Ionomer}}}}{\rm{/}}{Y_{{\rm{Ir}}{{\rm{O}}_2}}}}}{{{\rho _{{\rm{Nafion}}\;{\rm{D}}520}}}}} \right){\rm{/}}{\varepsilon _{{\rm{CL}}}} = 6.44\mu {\rm{m}}}\end{array}\]Considering the estimation of the solid volume fraction of the catalyst layer εCL, the total thickness tTGA lies in the interval {tTGA|6.44 µm − 1.07 µm ≤ tTGA ≤ 6.44 µm + 1.61 µm}.The thickness can also be estimated via the three stitched BSD images (see Figure S5, Supporting Information for one stitched BSD image dataset). The average thickness tBSD is the weighted sum of all three thicknesses of the datasets. The respective weighting wi is determined by the individual length li of each dataset. The individual thickness of each segmented BSD dataset is calculated as the total area of all n Pixels with a pixel width a of 97.7 nm divided by the total length li of the stitched image:8tBSD=∑i=13wintotal,ia2li=5.01 µm±0.33 µm\[\begin{array}{*{20}{c}}{{t_{{\rm{BSD}}}} = \mathop \sum \limits_{i = 1}^3 {w_i}\frac{{{n_{{\rm{total}},i}}{a^2}}}{{{l_i}}} = 5.01\,\mu {\rm{m}} \pm 0.33\,\mu {\rm{m}}}\end{array}\]The total amount can also be determined by summing up the deviations of the volume fraction φi with the volume fraction of the bulk φbulk multiplied by the voxel size hi. The limits of the sums are the points where the volume fraction exceeds and fall below the volume fraction of the bulk for the first time:9tMicro−CT=∑i=1N(φi−φbulk) hi= 3.30µm \[\begin{array}{*{20}{c}}{{t_{{\rm{Micro}} - {\rm{CT}}}} = \mathop \sum \limits_{i = 1}^N ({\varphi _i} - {\varphi _{{\rm{bulk}}}})\;{h_i} = \;3.30\mu {\rm{m}}\;}\end{array}\]The latter calculation should only be understood as a lower limit of the total thickness of the catalyst layer. The assumption that the volume fraction at the edge is equal than φbulk is an approximation and will underestimate the actual total thickness. Therefore, it is concluded that the equivalent CCM catalyst layer thickness is correspondingly between 5.01 µm and 6.44 µm. With this total catalyst layer thickness, an absolute catalyst distribution could be determined.Catalyst UtilizationWith the known structure, it is now possible to estimate the penetration depth of the membrane into the PTE in a working PEMWE. The membrane infiltration depth and the catalyst distribution give information about the catalyst utilization. The catalyst utilization, in our publication defined as the ionically well‐connected catalyst, is calculated by integrating the catalyst distribution over the membrane infiltration depth. In general, the infiltration depth of the membrane will depend on the mechanical properties of the membrane and the PTE, the clamping pressure of the system, and the structural properties of the PTE, especially the surface roughness and the pore and fiber diameters. Schuler et al.[30] determined the membrane or rather the CCM deformation of operated PEMWE MEAs by imaging the system after operation via X‐ray tomography. They used a symmetric PTL/CCM setup, consisting of titanium fibers (Bekaert) and Greenerity E400 CCMs with a Nafion 117CS membrane (Greenerity), with a contact pressure of 2.5 MPa. The analysis was performed with several titanium fiber substrates. The most relevant PTLs for our publication, for which the membrane deformation data is available, have a nominal fiber diameter of 14 µm (L1) and 50 µm (L3) with a common porosity of 56%. They determined the mean surface roughness Rm, the surface roughness at 90%cum R0.9, and the maximum surface roughness Rmax of the membrane deformation. The maximum surface roughness Rmax of L1 was 30 µm and of L3 was 47.6 µm. Our investigated aPTE‐setup has a titanium fiber PTL with a nominal fiber diameter of 20 µm and a porosity of 56%. Therefore, the maximum deformation of our used Nafion 117 membrane should lie between these literature values of Rmax (L1) and Rmax (L3), assuming a similar contact pressure. Thus we assume that a maximal 38% to 60% of the catalyst is in contact with the membrane, considering Rmax (L1) and Rmax (L3) (see Figure 7b). A more precise estimation of the catalyst utilization would require the exact membrane deformation within the system either via simulation or experimental data from in situ computed tomography of an assembled cell.This amount of ionically well‐connected catalyst is also consistent with the results of Kulkarni et al.[23] They calculated the TPCA value of ≈50% for a fiber titanium PTE with a loading of 1.75 mg cm−2 and a TPCA value of ≈20% for an identical PTE with a loading of 1.1 mg cm−2. Linear approximation of these TPCA values leads to a TPCA value of ≈40% for our loading of 1.5 mg cm−2. The used PTL had a similar fiber diameter to our substrates. In our work, the proton conductivity of the binder in the catalyst layer and the different mechanical properties between a CCM and a pure membrane are not considered. Both of them should raise the catalyst utilization. The proton conductivity of the binder will let deeper catalyst‐coated fibers contribute to the OER reaction in the catalyst layer. Although Leonard et al.[11] and Kulkarni et al.[23] showed that the proton conductivity is a bottleneck in the aPTE setup and leads to high ionic potential losses. From PEMFC literature, it is known that the elastic modulus of the platinum catalyst layer is higher than that of the membrane.[53,54] Therefore, it can be assumed that a CCM as a compound of CLs and membrane is stiffer than the pure membrane. As a result, the pure membrane should penetrate deeper into the fibers. However, a deep membrane penetration could hinder water and oxygen transport at the top fibers, reducing the catalyst efficiency.In a first‐order approximation, only the catalyst particles contacted via membrane contribute to the catalyst utilization (see Figure 7b). Consequently, at least 40% of the catalyst in this setup are not used. As a result, future PTEs designs should also improve catalyst utilization by optimizing their structure.Additionally, the interfacial contact area between PTL and CCM is estimated. Schuler et al.[30] determined the interfacial contact area RICA of L1 and L3 (RICA (L1) = 1.81 m2/mgeo2${R_{{\rm{ICA}}}}\;({\rm{L}}1)\; = \;1.81\;{{\rm{m}}^2}{\rm{/m}}_{{\rm{geo}}}^2$ and RICA (L3) = 1.11 m2/mgeo2)${R_{{\rm{ICA}}}}\;({\rm{L}}3)\; = \;1.11\;{{\rm{m}}^2}{\rm{/m}}_{{\rm{geo}}}^2)$. The RICA of our configuration lies probably between these values. Suppose the upper titanium fibers are completely covered with catalyst. In that case, the interfacial contact area of a PTE should always be higher than a similar CCM configuration with the same components and experimental conditions because of the previously mentioned different mechanical properties and a different membrane swelling behavior.Modelling of the Spray Coating ProcessThe upper sections revealed a specific PTE system's structure and catalyst distribution. With this knowledge, the aim was to develop a spray‐coating model for PTEs on titanium fiber substrates.In reality, the coating behavior of the titanium fibers will depend on numerous variables: the ink (surface tension, viscosity, loading,[23] composition of the ink, and evaporation temperature of the solvent[14]) and settings of the ultrasonic nozzle (nozzle amplitude and nozzle frequency[55]), the surface properties of the substrate (structure, roughness, and adhesion) and the surrounding parameters (used equipment, atmosphere, temperature of the heating plate, solution flow rates[56]). In the literature, some models can predict the size of ink droplets created by ultrasonic spray coating, depending on nozzle amplitude and frequency, the surface tension, and the ink's viscosity.[55] Furthermore, there exists literature[56] that correlates process parameters, like deposition parameters and solution properties, and the formation of thin and smooth films. Kulkarni et al.[23] observed that depending on the catalyst loading, different coverages of the titanium fibers were found. While for low loadings, only the top surface of the fibers is covered with catalyst, parts of the curved fiber surface are already deposited with a catalyst for medium loadings. For higher loadings, the catalyst covering even becomes denser. In general, there exists nearly no literature about the correlation between Ir‐based catalyst inks and the catalyst layer formation on titanium substrates. Whereas the Pt/C‐based catalyst layer formation in dependence on the process and ink parameters in PEMFCs has been broadly investigated.[57]As too many physical variables and influencing factors exist, our spray coating model tried to simplify the reality as much as necessary. It is a discrete model using a voxelized implementation of the physical world. The goal of the model was to calculate a two‐dimension catalyst distribution that reproduces the catalyst distribution of the embedded sample.In the applied simulation, the ink droplets have only the size of a voxel, 2.5 µm. According to the manufacturer, isopropanol has a medium droplet size of 27 µm using a nozzle frequency of 48 kHz. The isopropanol droplet size suggests that the catalyst ink has a droplet size of a similar magnitude. Furthermore, the titanium fibers are not composed of only horizontal and vertical edges. They instead have a round or elliptical shape (see Figure 1c). Surface parameters, like roughness and adhesion, and conditions, like temperature, are expressed in the upcoming model (see models 3 and 4) by a not immediate drying of ink and an occurrence of a downward flow. Additionally, clogging of the catalyst ink can be observed (see Figure 4a,b). The clogging probably occurs at some bottleneck throats of the pore between the titanium fibers. However, clogging is not considered in our model.With these restrictions in mind, all four model scenarios described exactly in the experimental part, are illustrated and compared to the catalyst distribution of the embedded sample in Figure 8. The main model assumptions are displayed in Table 3. Additionally, the P(90) and P(95) are calculated (see Table 2). In this calculation, the total catalyst amount was taken as the catalyst amount present in this illustrated 200 µm. As a result, the catalyst located deeper into the PTL was neglected. Additionally, the coefficient of determination was determined between the model data and the reference catalyst distribution (see Table 2).8FigureCatalyst distributions displayed in (a–d) were received by four different modeling approaches. The sketches on the right‐hand side illustrate the functioning of the respective models, which were calculated on the full micro‐CT tomogram of the PTL. The titanium fibers are shown in green, and the deposited catalyst layer is in yellow.3TableAssumptions for the four different cases in the presented spray coating modelScenarioHorizontal facesVertical facesDownward flowDripping1✓✗✗✗2✓✓✗✗3✓✓✓✗4✓✓✓✓Models 1 and 2 are directly connected to the surface roughness of the PTL. Model 1 is just taking into account the horizontal edges (see Figure 8a). This leads to a non‐lognormal distribution with a negative R2‐value of −1.0207 and too‐low penetration depths (see Table 2). The catalyst distribution of model 2 (see Figure 8b) was calculated assuming that both the horizontal and vertical edges are coated with the same amount of catalyst. Model 2 was based on the assumption of the half‐coated titanium fibers presented in the previous section “Catalyst Layer Thickness” (see Figure 3b, Equation (5)). Additionally, as a simplification in model 2, no differentiation between horizontal and vertical surfaces was assumed because of the round shape of the titanium fibers in reality. Model 2 already reveals a lognormal shape of the catalyst distribution and deeper penetration of the catalyst (see Table 2). Overall, model 2 reinforces the assumption of the half‐coated titanium fibers to calculate the catalyst layer thickness (see Figure 3b). However, neither the R2‐values nor the P(90) or P(95) is in good agreement with the reference distribution.The deeper penetration depth can only be reached considering a certain flow of the catalyst while drying. Model 3 allows this downward flow of the catalyst by considering a specific evaporation time of solvents in the ink. In this scenario, horizontal and vertical edges are still coated with the same amount of catalyst. However, different flow rates of the catalyst ink of the horizontal and vertical edges can be present with the restriction that the horizontal has to be less or equal to the vertical flow due to the presence of gravitational forces. A parameter sweep of the horizontal and vertical flows between 1 and 10 voxels led to the highest congruence with the reference distribution with a horizontal flow rate of 4 voxels and a vertical flow rate of 6 voxels (see Figure 8c). The coefficient of determination (R2‐value of 0.9361) and a visual check were the reasons for this. A horizontal flowrate of 3 voxels and a vertical flowrate of 5 voxels directed into a similar distribution. Overall, this simple model already reproduces a similar catalyst distribution to the reference distribution. Nevertheless, in reality, the evaporation rate of water in an aqueous solution decays with time in a square root dependency of time.[58] Therefore, the assumption of the homogenous contribution of each catalyst droplet along the through‐plane direction is probably oversimplified. In reality, the ink distribution will be an interplay between diffusion, migration, and interfacial forces.Also, model 3 misses some amount of catalyst at deeper penetration depth. Therefore model 4 tries to compensate for that with so‐called dripping that accounts for penetration into pores below fibers, which are otherwise protected from a direct deposition. A certain amount of the ink droplet travels through the pore space and is deposited in deeper regions of the titanium substrate. Model 4 was applied to the best data set of model 3. In this process, a parameter sweep of the dripping catalyst amount in 5% steps was performed. The optimum dripping parameter was 10%. This means 10% of the ink droplets that hit the vertical and horizontal edges splashes (see Figure 8d). Therefore, the catalyst distribution is shifted and comes very close to the reference catalyst distribution (R2 = 0.9607).In summary, significant knowledge can be gained from this catalyst spray coating model. First, all models depend on the surface structure directly connected to the mean fiber diameter and the porosity. Schuler et al.[30] showed empirically that a lower mean fiber diameter leads to a lower PTL surface, assuming a constant thickness. Furthermore, the surface roughness of structures with the same mean fiber diameter increases with rising porosity. Second, a flow of the catalyst ink was evidenced and therewith, a deeper infiltration of the catalyst into the porous substrate. The drying time will mainly be influenced by the temperature of the heating plate, the choice of the solvent and its evaporation time, and the interaction between titanium fiber and catalyst ink. So in terms of improved catalyst utilization in aPTEs titanium fibers with a lower mean fiber diameter, lower porosity, and surface roughness should be preferred. This can also be achieved using a porous backing layer or a micro porous layer. It has to be noted that this model requires experimental validation for other PTL structures, like sintered titanium powders, and different electrolyzer systems, like anion exchange membrane water electrolyzers to ensure the translatability to these substrates.ConclusionIn this work, for the first time, the structure and catalyst distribution of an aPTE, produced by spray coating an IrO2 catalyst on a titanium fiber PTL in PEMWE was determined quantitatively. For this purpose, we used two different methods. First, the segmentation and evaluation of several cross‐sectional images of the PTE revealed a through‐plane catalyst layer distribution with a lognormal shape. 90% of the catalyst amount lies within the first 100 µm. The distribution was confirmed by considering the through‐plane grey‐value gradient of a Micro‐CT of a PTE. In conjunction with the membrane deformation measurement of Schuler et al.[30] (of a comparable setup), we concluded that the upper limit ionically well‐connected catalyst in the investigated system is 60%.The analysis of the Micro‐CT also revealed a porosity gradient of the PTE. The catalyst layer thickness was estimated. The mean catalyst layer thickness around the upper half of the fibers at the upper PTE surface was ≈1.3 µm. The determined values of the titanium fiber thickness and the porosity were in good agreement with the literature values.Moreover, the Micro‐CT dataset showed that, in general, the coating of the titanium fibers does not change the surface roughness of a PTE compared to the pure titanium fiber PTL. A local surface roughness increase only occurs where big catalyst droplets are present.In addition, a first model of the deposition process was developed that matches the catalyst distribution found in the tomographic data. We showed that the surface roughness properties, combined with a downward‐flowing ink, are the main responsible factors for the catalyst distributions in PTEs. We expect that the catalyst utilization will increase with lower surface roughness, lower porosity, and lower mean fiber diameter.Based on these findings, it will be possible in future works to investigate the interplay between manufacturing and resulting PTE structure. The presented models can thus allow the finding of optimum titanium structures in future works.In summary, our paper provides a first impression of the structure of PTEs and helps design a roadmap for improving and optimizing PTEs.Experimental SectionPTE FabricationThe anodic PTEs were prepared using an ultrasonic spray coater (Exacta Coat, Sonotek) with a 48 kHz AccuMist nozzle. The procedure was performed according to Bühler et al.[18] Sintered titanium fibers (2GDL40‐1.0, Bekaert) with a porosity of 56% and a mean fiber diameter of 20 µm were used as substrates laser cut to either 5 cm2 for electrochemical testing or 0.25 cm2 for structural analysis. Spray coating was performed employing the following parameters: 5 W ultrasonication power, a nozzle height of 37 mm, a hot plate temperature of 120 °C, a meander‐shaped spray pattern with a pitch of 1.5 mm, a flow rate of 0.45 mL min−1, a path speed of 170 mm s−1 and a shaping air flow of 0.6 kPa, consistent with Mayerhöfer et al.[13]The ink was prepared according to Mayerhöfer et al.[13,18] with a solid content of 1 wt.% containing 98 wt.% IrO2 (Premion, Alfa Aesar) and 2 wt.% Nafion (D520, Chemours) in a 4:1 solvent mixture of water and isopropanol. The aim was to obtain a final catalyst layer with a loading of 1.5 mg cm−2 iridium oxide and a binder content of ≈10 wt.%. The Ti PTLs were mounted onto the spray coater hot plate using frames made of 1 mm thick virginal PTFE foil (HighTechFlon) to prevent slipping. To minimize the precipitation of the catalyst particles, the ink was stirred during the spray coating process. Nevertheless, a reproducible deviation between binder content in the ink (2 wt.%) and the final catalyst layer (10 wt.%) could be observed, as already described by Bühler et al.[18] For PTE quality control, the final binder content in the catalyst layers was determined via thermogravimetric analysis (STA 449 F3 Jupiter, Netzsch). For this purpose, the residual (dried) catalyst layer was scratched off the spray coating masks and heated to 600 °C (10 K min−1 steps) in Al2O3 crucibles. With this method, the detected mass loss represents the ionomer content. The determination of the binder content is shown in Figure S1, Supporting Information). Additionally, the catalyst loading was determined by weighing a 4 cm2 reference sheet before and after the spray coating process. The catalyst loading was calculated by dividing the mass difference of the GDL by the area of the spray‐coated GDL, and by taking into account the binder content of the catalyst layer.Electrochemical CharacterizationA combination of a commercial electrolyzer test system (600‐ETS, Scribner Associates) and an external potentiostat (VSP‐300 with three additional 5 V/10 A booster boards, BioLogic) was used for the electrochemical characterization. The 600‐ETS controlled and managed the pure anode water feed for the anode (40 mL min−1, 80 °C), the purging of the cathode with nitrogen (100 mL min−1, 1 atm), and the heating of the cell and the water reservoir (80 °C). The measurement cell was the associated electrolyzer cell test fixture (600 Electrolyzer Cell Fixture, Scribner Associates).The cell was assembled similarly to Bühler et al.[18] The same carbon cloth‐based cPTE as Bühler et al. (SL‐GDE, FuelCellEtc) with a catalyst loading of 0.5 mg cm−2 platinum (60% platinum on Vulcan) surrounded by a 150 µm tick PTFE frames (HighTechFlon) was used as a cathode. This PTFE frame ensured the same high compression of ≈60%. The anodic PTE was placed within a 1 mm thick PTFE sheet (HighTechFlon). The Nafion 117 membrane (Chemours) was sandwiched between the two PTEs. Finally, eight 1/4" screws tightened the MEA and the cell fixture with a final torque of 8.5 Nm applied in three steps before heating the cell.After heating the cell to 80 °C, a break‐in procedure was performed to stabilize the system. The break‐in procedure consisted of a 1 h constant voltage of 1.8 V. Then, three polarization curves were recorded, including galvanostatic EIS measurements. The electrochemical characterization based on galvanostatic EIS measurements was similar to Suermann et al.[59] employing the following steps for the acquisition of polarization curves: 1 mA cm−2 between 1  and 10 mA cm−2, 10 mA cm−2 between 10 and 100 mA cm−2, and 100 mA cm−2 between 100 and 4 A cm−2. If the voltage exceeded 2.3 V, the polarization curve measurement was terminated to prevent titanium corrosion. Each step was held for 20 s and followed by a shortened EIS sequence to determine the high‐frequency resistance. The EIS scan consisted of a frequency sweep between 200 kHz and 100 Hz with 13 measuring points per decade and a waiting point of one period between two impedance measurements. The x‐intersection of the Nyquist plot determined the current‐dependent HFR. The final break‐in measurement was a complete EIS from 200 kHz till 0.1 Hz with 13 points per decade and a holding time of 10 min of the following current densities: 10, 20, 50, and 100 mA cm−2 – 2 A cm−2 in 100 mA cm−2 intervals. To verify the stabilization, polarization curves were recorded until two subsequent curves were observed to be identical. A minimum of three curves was recorded. Finally, the last curve was taken for further performance evaluation. Overall three PTE samples were measured following this protocol.Structural AnalysisFor all structural measurements, PTEs with a size of 5 mm × 5 mm were used, fabricated with the same ink batch as the electrochemical samples. This sample measurement was chosen as a compromise between two considerations. On the one hand, the sample size should be as small as possible because of the better resolution of the X‐ray tomography. On the other hand, the sample size was limited by the accuracy and edge effects of the laser cutting of the titanium fibers. The same sample examined via Micro‐CT was later used for all electron imaging.Micro‐CT Data AcquisitionThe Micro‐CT measurement of the coated titanium fibers was done by the company RJL Microanalytic. They used a SkyScan 2211 Multiscale X‐Ray Nanotomograph (Software Version 2.5.1) from Bruker microCT and applied a source voltage of 130 kV and a source current of 200 µA. The difference in mass attenuation coefficient of titanium and iridium oxide led to grayscale gradients in the Micro‐CT data.[60,61] The imaging volume was 1282 × 1820 × 588 pixels with a voxel size of 2.5 µm. The reconstruction was done via the software NRecon (Software Version 1.7.5.0) from Bruker microCT. Hereby, ring artifact correction, beam hardening correction, and filter cutoffs were applied.Micro‐CT ReconstructionBefore the segmentation, a 3D anisotropic diffusion filter based on Perona and Malik[62] and Gerig et al.[63] and implemented by Lopes[64] was applied to the dataset to reduce noise by simultaneously preserving the edges of the fibers. The anisotropic diffusion filter was executed in Matlab (Matlab R2018b, The MathWorks). No gradient corrections were applied. Because of the grayscale gradients, the segmentation was done with a Local Otsu method implemented in GeoDict (GeoDict Version 2022, Math2Market) to perform a region‐based segmentation. While doing so, the segmentation only distinguished between pore and material. In consequence, the titanium fibers and the catalyst layer were not distinguished. Afterward, the binary dataset was cropped at the edges to obtain finally a reconstructed PTE with a volume of 1200 × 1700 × 465 pixels. This volume still contained void areas at the edges in a through‐plane direction.Micro‐CT Surface RoughnessSchuler et al. defined several amplitude‐based surface roughness parameters such as surface roughness, arithmetic mean surface roughness, root mean square roughness, and maximum profile height.[30] These parameters are key numbers to compare the roughness of different sintered titanium fibers. Their mathematical definition can be seen in the supporting information, and the calculations were implemented in Matlab. The roughness parameters were calculated for the non‐coated backside as well as for the catalyst‐coated front side of the segmented PTE. Additionally, the surface roughness was illustrated in roughness profile height maps.Micro‐CT Maximum Intensity ProjectionThe maximum intensity projection is an algorithm that projects a 3D data set into a 2D projection plane regarding the maximum intensity of a given direction. Therefore the voxel with the maximum intensity along a specific projection path was selected and displayed in a 2D projection image.[65] This algorithm was implemented in Matlab.Micro‐CT Catalyst DistributionEven though the resolution of the Micro‐CT was not high enough to resolve the catalyst layer, a slice mean gradient of the mean gray value in the through‐plane direction could resolve the catalyst distribution qualitatively. Therefore, a mean gray value of each through‐plane plane was calculated and plotted versus the PTE depth. Afterward, the baseline of the mean gray value of the inner part of the PTE was determined. An interval could be determined in which the mean gray value exceeded and approached the baseline again. The new origin was chosen as the intersection of the mean grey value curve with the baseline. The mean grey value curve was fitted with a lognormal distribution done with the nonlinear fitting tool in Origin (OriginPro 2019 Version 9.6.0.172, OriginLab Corporation).Electron ImagingA Zeiss Crossbeam 540 FIB‐SEM microscope (focused ion beam scanning electron microscope) with a Gemini II column was used for all electron imaging. All samples were attached on aluminum SEM specimen stubs (G301 and G399, Plano GmbH) with conductive carbon pads (G3347 and G3348, Plano GmbH). Additionally, the samples were sputter‐coated either with gold (108 Manual Sputter Coater, Cressington) or carbon (MED 010, Balzers Union) before imaging to obtain better conductivity.SEM Surface ImagingThe surface of the PTE was imaged via a secondary electron detector with an accelerating voltage of 3 kV and a current of 750 pA. The sample was both carbon‐coated and contacted with silver conducting paint (G3692, Plano GmbH) at the edge of the sample to ensure better conductivity.Embedded Cross‐Section Sample Preparation and ImagingOne 5 mm × 5 mm PTE sample was sandwiched between two PTFE sheets with two plastic clips (SAMPL‐KLIP PLASTIC HOLDER, ITW Test & Measurement GmbH) and embedded in epoxy resin (Araldite 502 Epoxy Resin, Electron Microscopy Sciences) into an embedding cup (SamplKup, ITW Test & Measurement GmbH). Afterward, the sample was cured overnight at 60 °C. The subsequent grinding and polishing steps were performed with an automatic polishing machine (LaboForce‐100, Struers GmbH). Therefore, the sample was manually ground with 220‐grain size SiC sanding paper (Struers GmbH) until the edge area of the sample was removed. Next, the samples were ground successively with finer SiC sanding paper from 500 to 4000‐grain size (Struers GmbH). Finally, the embedded sample was polished using two different polishing plates (MD‐Mol and MD‐Nap polishing plate, Struers GmbH) with corresponding diamond solutions (DiaPro Mol B 3 µm and DiaPro Nap ¼ µm, Struers GmbH). In addition, the sample was ultra‐sonicated before and after the polishing procedure to clean the cross‐section surface area. Besides the conductive carbon pad and the gold sputter coating, the conductivity was also improved using aluminum conductive tape (Plano GmbH) connecting the specimen holder and the embedded sample.The imaging was performed with a four‐quadrant backscattering detector in compositional mode, applying an accelerating voltage of 20 kV and 1 nA beam current. The pixel size was 97.7 nm. The embedded sample was aligned horizontally regarding the PTE. Afterward, a sequence of images along the catalyst‐coated part of the titanium fibers was recorded with an overlap of ≈20% and a magnification of 381X (image width of 300 µm). The images were then stitched with the Grid/Collection stitching tool implemented in the imaging software Fiji.[66] Thereby, a large cross‐sectional area with a width of several millimeters was accessed with a high resolution. The horizontal alignment of the stitched image was checked by inspection and, if necessary, corrected with the Rotate‐function in Fiji and a bilinear interpolation of the image.Afterward, the embedded sample was ground, polished, and imaged two more times according to the protocol described above to obtain two more datasets for improved sample size.Embedded Cross‐Section Segmentation and Data ProcessingEach stitched gray value image was segmented with an AI‐based algorithm. The deep learning method Unet 2D[47] implemented in GeoDict was used in an iterative training process to distinguish fiber, catalyst, and pore. After each iteration, the segmented image section was visually inspected, and finally, the AI model was applied to the entire stitched image.The three segmented images were used to determine the catalyst contribution dependent on the penetration depth. The data analysis was performed in Matlab. The catalyst distribution was determined individually for each of the stitched images. Therefore, all pixels assigned to the catalyst layer were summed up for each horizontal line (thickness of 97.7 nm of each horizontal line due to the pixel size). The beginning of the catalyst layer was defined when a horizontal line exceeded a threshold of 50 catalyst pixels. Afterward, the amount of catalyst layer pixels of each horizontal line was related to the penetration depth in the through‐plane direction. Following this, the underlying next 200 µm were displayed in a normalized catalyst contribution plot. Eventually, the average catalyst contribution was determined by taking into account all single catalyst contributions weighted by their image width. Additionally, the curve was fitted with a lognormal distribution (y(x)=y0+A2πwxe−(ln(xxc))22w2)\[\left( {y(x) = {y_0} + \frac{A}{{\sqrt {2\pi } wx}}{e^{ - \frac{{{{\left( {\ln \left( {\frac{x}{{{x_c}}}} \right)} \right)}^2}}}{{2{w^2}}}}}} \right)\]. The fitting was done with the nonlinear fitting tool in Origin.For better comparability, a left‐sided confidence interval of P(0.90) or P(0.95) with a confidence level of 90% or 95% was calculated. This interval specifies the maximum penetration depth, to which 90% or 95% of the catalyst lies.Energy‐Dispersive X‐Ray SpectroscopyThe Zeiss Crossbeam 540 with an energy‐dispersive X‐ray‐detector (X‐Max 150 silicon drift detector, Oxford Instruments; Software: Aztec Version 4.2, Oxford instruments) was used for elemental analysis of the embedded samples. The elemental mapping was performed by applying an accelerating voltage of 20 kV and 1 nA beam current.Catalyst UtilizationIn general, the catalyst utilization of the anode catalyst depends on the local ionic and electronic conductivity, the water transport, and the oxygen removal. In this publication, catalyst utilization and the amount of ionically well‐connected catalyst were equated. Catalyst utilization was simplified and defined as the percentage of catalysts that could be in direct contact with the membrane. The interface area between the catalyst layer and membrane varies depending on the membrane deformation, the associated membrane infiltration depth, and the membrane's swelling. The catalyst utilization was resulted from the catalyst distribution and was the cumulative distribution.Structural ParameterFor the structural analysis, the Software GeoDict was used. The segmented Micro‐CT (volume of 1200 × 1700 × 465 Pixel) was the dataset processed here. The variation in porosity and the volume fraction in through‐plane direction was determined via the package MatDict.The through‐plane edge areas were cropped to determine the basic structural properties of the titanium PTL (volume of 1200 × 1700 × 300 pixel). Thereby only pure titanium fibers were examined, and edge effects were excluded. The porosity was determined using the package MatDict. The pore size distribution was calculated via granulometry with a bin size of two voxels employing the package PoroDict. Fiber diameter distribution and fiber orientation were determined using the package FiberFind. The fibers were fitted as curved circular fibers and identified with a machine learning method. In addition, the fiber fragments at the domain boundary were removed to exclude edge effects.Spray Coating ModelAn algorithm was developed to model the spray coating process for aPTEs to get a deeper understanding of the deposition process. The output of the algorithm was a 2D catalyst distribution. The lognormal fit of the embedded sample serves as a reference catalyst contribution to benchmark the model output.The spray coating process was simulated on a blank titanium fiber PTL. Therefore, the segmentation of the uncoated backside of the Micro‐CT titanium fiber sample was used. The surface of the fibers was not triangulated. Therefore, the surface consisted of horizontal and vertical faces with a side length of 2.5 µm.The droplet size of the catalyst ink was one voxel. Each grid point was spray‐coated. As a consequence, the spray beam consisted of 1200 × 1700 individual and independent beams perpendicular to the titanium fiber plane. This product was the total number of voxels of each plane. The spray beam only hit from above visible fibers. The voxel adjacent to a fiber was then filled with a specific amount of catalyst.Depending on the model assumptions, different catalyst distributions were calculated. The 2D model output was the amount of catalyst in dependency on the through‐plane penetration depth. The model did not provide any information about the exact 3D structure of the catalyst layer.The distribution width was limited to 200 µm to directly compare it to the lognormal catalyst contribution of the embedded sample. The starting point of the distribution of the spray coating models was chosen as follows: The origin of the distribution was the position when the position's value, equaled the catalyst amount of the individual plane, exceeded the value of the catalyst distribution that was 200 µm, equal 80 voxels, apart.Four spray‐coating cases are presented based on these previous assumptions (see Table 3 and Figure 8). Each case included the assumptions of the previous cases:The spray beam is considered to deposit on the horizontal faces of the fiber surfaces visible from above. The fibers are only coated with catalyst on their top surface.The spray beam is now capable of coating specific vertical faces with a catalyst. Considering a straight perpendicular spray beam, all vertical surfaces passed by the beam are coated. The amount of catalyst is equal on horizontal and vertical surfaces.The spray‐coated catalyst layer does not dry immediately and flows downwards. The downward flow is independent of the fiber structure and surrounding spray beams. Depending on whether the spray beam hits a horizontal or vertical face, the maximum depth of the downward flow varies. The downwards flow is regardless of the impact position of the droplet and should be interpreted as a statistical flow. The amount of ink that hits the face is equally contributed between all reached through‐plane planes.Additionally, pores are considered. In the spray coating process, a specific amount x of the catalyst ink will drip or splash through the voids after it has hidden the horizontal or vertical faces. The remaining amount of the deposited material behaves like the spray‐coated layer of model 3. The additional flying distance of the catalyst ink amount x is equal to the mean pore diameter. After flying through the void, it will hit an imaginary voxel with the same horizontal or vertical face as the hidden face and will, from that point on, also follow model 3.To better classify the obtained catalyst distributions by upper models, the coefficient of determination R2= 1−residual sum of squaresmean total sum of squares${R^2} = \;1 - \frac{{{\rm{residual}}\;{\rm{sum}}\;{\rm{of}}\;{\rm{squares}}}}{{{\rm{mean}}\;{\rm{total}}\;{\rm{sum}}\;{\rm{of}}\;{\rm{squares}}}}$[67,68] was calculated between the distributions and the lognormal fit of the embedded sample. However, a good R2‐value (worst value=− ∞, best value =+ 1) was insufficient to determine the quality of a fit. Therefore the catalyst distribution was also checked via eye. To compare the catalyst distribution and the lognormal fit, both the maximum of the lognormal fit of the catalyst distribution of the model and the maximum of the lognormal fit of the embedded reference sample were normalized to 1.AcknowledgementsThe authors gratefully acknowledge the financial support by the Federal Ministry of Education and Research of Germany in the framework of the StacIE project (BMBF/03HY103H) and the Federal Ministry for Economic Affairs and Energy in the framework of HoKaWe (BMWi/ 03EI3029A). Furthermore, the authors want to thank Johannes Bender for the TGA and Thomas Böhm for the helpful discussions.Open access funding enabled and organized by Projekt DEAL.Conflict of InterestThe authors declare no conflict of interest.Data Availability StatementThe data that support the findings of this study are available from the corresponding author upon reasonable request.S. S. Kumar, V. Himabindu, Mater. Sci. Energy Technol. 2019, 2, 442.Fuel Cells and Hydrogen 2 Joint Undertaking, Hydrogen roadmap Europe : a sustainable pathway for the European energy transition, https://data.europa.eu/doi/10.2843/341510 (accessed: March 2021).S. A. Grigoriev, V. N. Fateev, D. G. Bessarabov, P. Millet, Int. J. Hydrogen Energy 2020, 45, 26036.Silyzer 300: The next paradigm of PEM electrolysis, https://assets.siemens-energy.com/siemens/assets/api/uuid:a193b68f-7ab4-4536-abe2-c23e01d0b526/datasheet-silyzer300.pdf (accessed: November 2021).International Renewable Energy Agency (IRENA). Green Hydrogen Cost Reduction: Scaling up Electrolysers to Meet the 1.5°C Climate Goal, https://www.irena.org/-/media/Files/IRENA/Agency/Publication/2020/Dec/IRENA_Green_hydrogen_cost_2020.pdf (accessed: March 2021).C. van Pham, D. Escalera‐López, K. Mayrhofer, S. Cherevko, S. Thiele, Adv. Energy Mater. 2021, 11, 2101998.D. G. Bessarabov, P. Millet, PEM water electrolysis, Vol. 2, Academic Press an imprint of Elsevier, London, UK 2018.Manufacturing of membrane electrode assemblies for fuel cells, International conference on manufacturing research (IMCR), https://pdfs.semanticscholar.org/fc3d/c1538c2eafd0ea27df7b930cd13613869f05.pdf (accessed: March 2021).J. Park, Z. Kang, G. Bender, M. Ulsh, S. A. Mauger, J. Power Sources 2020, 479, 228819.J. Bender, B. Mayerhöfer, P. Trinke, B. Bensmann, R. Hanke‐Rauschenbach, K. Krajinovic, S. Thiele, J. Kerres, Polymers 2021, 13, 3467.E. Leonard, A. D. Shum, N. Danilovic, C. Capuano, K. E. Ayers, L. M. Pant, A. Z. Weber, X. Xiao, D. Y. Parkinson, I. V. Zenyuk, Sustainable Energy Fuels 2020, 4, 921.B. Mayerhöfer, F. D. Speck, M. Hegelheimer, M. Bierling, D. Abbas, D. McLaughlin, S. Cherevko, S. Thiele, R. Peach, Int. J. Hydrogen Energy 2022, 47, 4304.B. Mayerhöfer, D. McLaughlin, T. Böhm, M. Hegelheimer, D. Seeberger, S. Thiele, ACS Appl. Energy Mater. 2020, 3, 9635.B. Mayerhöfer, K. Ehelebe, F. D. Speck, M. Bierling, J. Bender, J. A. Kerres, K. J. J. Mayrhofer, S. Cherevko, R. Peach, S. Thiele, J. Mater. Chem. A 2021, 9, 14285.Toshiba's New Large‐scale Production Technology for Electrolysis Electrodes for Hydrogen Production from Renewables Cuts Iridium Use to 1/10, https://www.global.toshiba/ww/technology/corporate/rdc/rd/topics/22/2210-01.html (accessed: December 2022).TOSHIBA REVIEW 2018, No. 3, Technology for Reduction of Precious Metals Used in Anode Catalyst of PEM Electrolyzers, https://www.global.toshiba/content/dam/toshiba/migration/corp/techReviewAssets/tech/review/2018/03/73_03pdf/a03.pdf (accessed: December 2022).M. Bühler, P. Holzapfel, D. McLaughlin, S. Thiele, J. Electrochem. Soc. 2019, 166, F1070.M. Bühler, F. Hegge, P. Holzapfel, M. Bierling, M. Suermann, S. Vierrath, S. Thiele, J. Mater. Chem. A 2019, 7, 26984.C. van Pham, M. Bühler, J. Knöppel, M. Bierling, D. Seeberger, D. Escalera‐López, K. J. J. Mayrhofer, S. Cherevko, S. Thiele, Appl. Catal., B 2020, 269, 118762.S. Choe, B.‐S. Lee, M. K. Cho, H.‐J. Kim, D. Henkensmeier, S. J. Yoo, J. Y. Kim, S. Y. Lee, H. S. Park, J. H. Jang, Appl. Catal., B 2018, 226, 289.H. Kim, J. Kim, J. Kim, G. H. Han, W. Guo, S. Hong, H. S. Park, H. W. Jang, S. Y. Kim, S. H. Ahn, Appl. Catal., B 2021, 283, 119596.B.‐S. Lee, S. H. Ahn, H.‐Y. Park, I. Choi, S. J. Yoo, H.‐J. Kim, D. Henkensmeier, J. Y. Kim, S. Park, S. W. Nam, K.‐Y. Lee, J. H. Jang, Appl. Catal., B 2015, 179, 285.D. Kulkarni, A. Huynh, P. Satjaritanun, M. O'Brien, S. Shimpalee, D. Parkinson, P. Shevchenko, F. DeCarlo, N. Danilovic, K. E. Ayers, C. Capuano, I. V. Zenyuk, Appl. Catal., B 2022, 308, 121213.F. Hegge, R. Moroni, P. Trinke, B. Bensmann, R. Hanke‐Rauschenbach, S. Thiele, S. Vierrath, J. Power Sources 2018, 393, 62.T. L. Doan, H. E. Lee, S. S. H. Shah, M. Kim, C.‐H. Kim, H.‐S. Cho, T. Kim, Int. J. Energy Res. 2021, 45, 14207.X.‐Z. Yuan, N. Shaigan, C. Song, M. Aujla, V. Neburchilov, J. T. H. Kwan, D. P. Wilkinson, A. Bazylak, K. Fatih, Sustainable Energy Fuels 2022, 6, 1824.D. Bessarabov, H. Wang, H. Li, N. Zhao, PEM Electrolysis for Hydrogen Production: Principles and Applications, CRC Press, Boca Raton, FL 2016.M. Suermann, K. Takanohashi, A. Lamibrac, T. J. Schmidt, F. N. Büchi, J. Electrochem. Soc. 2017, 164, F973.O. Panchenko, E. Borgardt, W. Zwaygardt, F. J. Hackemüller, M. Bram, N. Kardjilov, T. Arlt, I. Manke, M. Müller, D. Stolten, W. Lehnert, J. Power Sources 2018, 390, 108.T. Schuler, R. de Bruycker, T. J. Schmidt, F. N. Büchi, J. Electrochem. Soc. 2019, 166, F270.T. Schuler, J. M. Ciccone, B. Krentscher, F. Marone, C. Peter, T. J. Schmidt, F. N. Büchi, Adv. Energy Mater. 2019, 285, 1903216.J. O. Majasan, F. Iacoviello, P. R. Shearing, D. J. L. Brett, Energy Procedia 2018, 151, 111.B. J. M. Etzold, U. Krewer, S. Thiele, A. Dreizler, E. Klemm, T. Turek, Chem. Eng. J. 2021, 424, 130501.C. C. Weber, T. Schuler, R. de Bruycker, L. Gubler, F. N. Büchi, S. de Angelis, J. Power Sources 2022, 15, 100095.T. Schuler, T. J. Schmidt, F. N. Büchi, J. Electrochem. Soc. 2019, 166, F555.J. K. Lee, C. Lee, A. Bazylak, J. Power Sources 2019, 437, 226910.C. Immerz, B. Bensmann, P. Trinke, M. Suermann, R. Hanke‐Rauschenbach, J. Electrochem. Soc. 2018, 165, F1292.X. Peng, P. Satjaritanun, Z. Taie, L. Wiles, A. Keane, C. Capuano, I. V. Zenyuk, N. Danilovic, Adv. Sci. 2021, 8, 2102950.R. Omrani, B. Shabani, Int. J. Hydrogen Energy 2017, 42, 28515.Q. Feng, X. Yuan, G. Liu, B. Wei, Z. Zhang, H. Li, H. Wang, J. Power Sources 2017, 366, 33.F. N. Khatib, T. Wilberforce, O. Ijaodola, E. Ogungbemi, Z. El‐Hassan, A. Durrant, J. Thompson, A. G. Olabi, Renewable Sustainable Energy Rev. 2019, 111, 1.M. K. Debe, S. M. Hendricks, G. D. Vernstrom, M. Meyers, M. Brostrom, M. Stephens, Q. Chan, J. Willey, M. Hamden, C. K. Mittelsteadt, C. B. Capuano, K. E. Ayers, E. B. Anderson, J. Electrochem. Soc. 2012, 159, K165.A. Albert, T. Lochner, T. J. Schmidt, L. Gubler, ACS Appl. Mater. Interfaces 2016, 8, 15297.L. Placca, R. Kouta, Int. J. Hydrogen Energy 2011, 36, 12393.T. Matsuura, M. Kato, M. Hori, J. Power Sources 2006, 161, 74.V. Radhakrishnan, P. Haridoss, Int. J. Hydrogen Energy 2010, 35, 11107.O. Ronneberger, P. Fischer, T. Brox, in Medical Image Computing and Computer‐Assisted Intervention – MICCAI 2015 (Eds.: N. Navab, J. Hornegger, W. M. Wells, A. F. Frangi), Springer International Publishing, Cham 2015, p. 234.Fuel Cell Store: Unit B College Station, TX. Nafion™ PFSA 5% Dispersions D520: SDS Number: 1334079‐00035 [Online], https://www.fuelcellstore.com/msds-sheets/nafion-dispersion-d520-sds.pdf (accessed: March 2022).S. Ghosh, H. Ohashi, H. Tabata, Y. Hashimasa, T. Yamaguchi, Int. J. Hydrogen Energy 2015, 40, 15663.S. Thiele, R. Zengerle, C. Ziegler, Nano Res. 2011, 4, 849.L. Zielke, S. Vierrath, R. Moroni, A. Mondon, R. Zengerle, S. Thiele, RSC Adv. 2016, 6, 80700.S. Thiele, T. Fürstenhaupt, D. Banham, T. Hutzenlaub, V. Birss, C. Ziegler, R. Zengerle, J. Power Sources 2013, 228, 185.F. Rong, C. Huang, Z.‐S. Liu, D. Song, Q. Wang, J. Power Sources 2008, 175, 699.K. K. Poornesh, C. Cho, D.‐Y. Kim, Y. Tak, Energy 2010, 35, 5241.W. Li, X. Bi, M. Luo, P.‐C. Sui, J. Electrochem. Soc. 2021, 168, 034502.G. S. Lonakar, M. S. Mahajan, S. S. Ghosh, J. V. Sali, Org. Electron. 2012, 13, 2575.H. Liu, L. Ney, N. Zamel, X. Li, Appl. Sci. 2022, 12, 3776.K. Roger, E. Sparr, H. Wennerström, Phys. Chem. Chem. Phys. 2018, 20, 10430.M. Suermann, T. J. Schmidt, F. N. Büchi, Electrochim. Acta 2016, 211, 989.Tables of X‐Ray Mass Attenuation Coefficients and Mass Energy‐Absorption Coefficients, NIST Standard Reference Database, https://www.physics.nist.gov/PhysRefData/XrayMassCoef/cover.html (accessed: January 1995).M. E. Wieser, N. Holden, T. B. Coplen, J. K. Böhlke, M. Berglund, W. A. Brand, P. de Bièvre, M. Gröning, R. D. Loss, J. Meija, T. Hirata, T. Prohaska, R. Schoenberg, G. O'Connor, T. Walczyk, S. Yoneda, X.‐K. Zhu, Pure Appl. Chem. 2013, 85, 1047.P. Perona, J. Malik, IEEE Trans. Pattern Anal. Mach. Intell. 1990, 12, 629.G. Gerig, O. Kubler, R. Kikinis, F. A. Jolesz, IEEE Trans. on Med. Imaging 1992, 11, 221.Anisotropic Diffusion (Perona & Malik), https://www.mathworks.com/matlabcentral/fileexchange/14995-anisotropic-diffusion-perona-malik (accessed: March 2022).F. A. Merchant, in Microscope Image Processing (Eds.: Q. Wu, F. A. Merchant, K. R. Castleman), Academic Press, Burlington 2008, Ch. 14.S. Preibisch, S. Saalfeld, P. Tomancak, Bioinformatics 2009, 25, 1463.S. Wright, J. Agric. Res. 1921, 20, 557.D. Chicco, M. J. Warrens, G. Jurman, PeerJ Comput. Sci. 2021, 7, e623.

Journal

Advanced Energy Materials – Wiley

Published: Apr 1, 2023

Keywords: catalyst distributions; gas diffusion electrodes; modeling; PEM water electrolysis; porous transport electrodes; tomographies; ultrasonic spray coating

There are no references for this article.