Fault Diagnostics in PEMFC Stacks by Evaluation of Local Performance and Cell Impedance Analysis

Starvation, flooding, and dry-out phenomena occur in polymer electrolyte membrane fuel cells due to heterogeneous local conditions, material inhomogeneity, and uneven flow distribution across the single cell active area and in between the individual cells. The impact of the load level and of the air feed conditions on the performance was identified for individual single cells within a 10-cell stack. Analysis of the current density distribution across the active area at the cell level was correlated with electrochemical impedance spectroscopy to enable operando fault diagnostic without any impact of the applied analytical tools on the single cell behavior. Moreover, the combination of both technologies allows in-depth analysis of fault mechanisms in fuel cell single cells with improved sensitivity. Current density distribution and the quantitative assessment of the performance homogeneity demonstrated high sensitivity to small humidity changes and allow the detection of critical events such as dry-out in single cells. Originally published on: Fuel Cells 4 (2020), 403-412 DOI: 10.1002/fuce.201900193 https://onlinelibrary.wiley.com/doi/full/10.1002/fuce.201900193 Impedance analysis is more sensitive regarding polarization and diffusion limitations and allows detection of cell flooding. The combination of both techniques is required for reliable identification of air starvation faults.

was combined with EIS on the single cell level. The objective of this work was the analysis of the faulty conditions and of the degradation phenomena in fuel cell stacks operating under system relevant conditions. This study was focused on the behavior of a 10-cell PEMFC stack being subject to different load levels and to critical operating conditions at the cathode side.
This combined study allows reliable detection of different fault modes and identifies reasons for increased local degradation effects in PEMFC stacks. Consequently, this in-depth analysis of fault mechanisms can enable fuel cell diagnostic in single cells with higher sensitivity towards occurring faults during system operation.

Fuel Cell Stack
The presented tests were carried out using a liquid-cooled PEMFC stack manufactured by ZSW (Zentrum für Sonnenenergie-und Wasserstoff-Forschung Baden-Württemberg). This stationary stack of 480 W nominal electrical power output was equipped with ten cells of an active electrode area of 96 cm² using graphitic composite bipolar plates. Three PCBs for measuring the current density distribution were included in the stack. The first PCB was integrated in the first cell close to the anodic current collector (referred to cell with index 01), the second PCB in the middle of the stack (cell 05) and the third PCB in cell 10 close to the cathodic current collector and the media supply ports (Figure 1 (A)). Multiple serpentine flow fields were used in the bipolar plates on the anode and on the cathode side. The resulting media flow configurations in each cell and the applied segmentations are shown in Figure 1

Test Station
Fuel cell tests were carried out using an DLR in-house manufactured 1 kW test station using hydrogen (grade 5.0) and compressed ambient air (particle-filtered, dried and oil-free). The test bench is equipped with programmable logic controllers (PLCs) and commercial electronic loads. It allows automatic control of the operating cell conditions, such as humidity of the reactants, cell temperature, gas flow rates and stack pressure. The relative humidity (RH) of the inlet gases is controlled by direct evaporators and high precision liquid mass flow controllers obtained from 10 kHz to 0.1 Hz. Above 66 Hz ten logarithmically spaced frequencies were measured for each decade using 20 measure periods each. Below 66 Hz five logarithmically spaced frequencies were measured for each decade using 4 measure periods each, resulting in data acquisition time of about 5 minutes. The AC amplitude was adapted to the DC load level to assure linear EIS response with good signal-to-noise ratio (Table 1). Data acquisition and analysis was carried out using the Thales XT 5.2.0 software.
The simplified equivalent circuit presented in Figure 2 (A) was used to fit the EIS data in this work. This equivalent circuit consists of a serial connection of the following elements: (i) an inductivity L for the cables in the setup, (ii) an ohmic resistance Rel mainly dominated by the electrolyte resistance (membrane), (iii) an RC element (parallel element of an ohmic resistance Rpol and a capacitor Cpol) for the polarization reaction, and (iv) an RC element for mass transport limitation dominated by diffusion processes (Rdiff and Cdiff).

Current Density Distribution
The used PCBs for the measurement of the current density and temperature distribution were specially design for parallel use in the examined stack and manufactured by "Helmbold -Messtechnik für Brennstoffzellen". 81 segments for current density measurement and 8 sensors for temperature measurement were included in each PCB. The data acquisition was realized using a 34980A Multifunction Switch/Measure Unit (Keysight Technologies) equipped with six 34922A 70-channel multiplexer modules. Resulting data acquisition time was about 4 seconds. Data processing and visualization was carried out using a DLR in-house programmed LabView application.

Impact of Electrical Load Level
The current-voltage characteristic of the investigated stack is shown in Figure 2 (B). The stack was operating under nominal conditions as defined by the stack manufacturer and as described in the following. The stack temperature was controlled to 80 °C according to the coolant temperature at the stack inlet and the feed gases were heated to 85 °C to avoid water condensation. Hydrogen and air stoichiometry of 1.5 and 2.0 were applied, respectively. The reactant pressure was controlled to 150 kPaabs at the stack outlet and the relative humidity of the reactants was adjusted to 30% at the stack inlet. To avoid starvation issues in case of current densities below 0.2 A cm -2 , the minimum flow rate was fixed at the value corresponding to the For more detailed analysis of the stack behavior at the load levels indicated in Figure 2    to (E)). Consequently, the polarization resistance Rpol decreased with increasing load from 1.92 to 0.15  cm 2 as can be seen from EIS data analysis via fitting which is provided in Figure 5 (G). Simultaneously, a second semi-arc at lower frequencies appears and increases with increasing current density. This semi-arc is caused by mass transport limitations. The resulting diffusion resistance Rdiff, also shown in Figure 5 (G), is increasing from 0.07  cm 2 at 0.2 A cm -2 to 0.48  cm 2 at 1.0 A cm -2 . In the transport control regime at 1.0 A cm -2 ( Figure 5 (F)), this mass transport limitation is dominant. Furthermore, Rdiff increases also for current densities below 0.2 A cm -2 to 0.24  cm 2 at 0.05 A cm -2 and to 0.31  cm 2 at 0.02 A cm -2 . This increase can be explained by the over-stoichiometric operation in this region and the resulting dry-out of the ionomer in the catalyst layer. It is known that the oxygen permeability through ionomer films decreases under dry conditions [39], which limits the oxygen access to the catalyst. The analysis of the electrolyte resistance Rel is presented in Figure 5 (G). Rel decreases with increasing electrical load from 0.23 to 0.07  cm 2 due to increasing water production and membrane humidification. Below 0.2 A cm -2 Rel is significant higher due to membrane dry-out caused by the over-stoichiometric operation in this region.
It is obvious that the impedance spectra of all single cells are similar and differences are negligible. No significant variations in Rel, in the polarization resistance, or in the transport resistance can be detected for a certain single cell under the applied nominal conditions. This indicated that applied nominal operating conditions enable homogeneous, non-critical operation of the examined stack. The inhomogeneous cell behavior as obvious in the current density distribution (Figure 3 and Figure 4) is not detectable by EIS. This shows that the current density distribution is more sensitive to small changes in the individual cell operation, even if these have minor impact on the cell voltage and the impedance response averaged over the entire active area of the cells. Thereby, the sensitivity of the current density distribution is not only accessible by the quantitative changes in the homogeneity, but also by the location of the current density maximum in relation to the air inlet. Furthermore, it can be concluded that nonlinear responses are not an issue for the applied single cell EIS analysis under these conditions.

Impact of Cathode Dry-out and Flooding
To validate the applied methodology for cathodic faults, the stack was operated under critical conditions. First, the impact of cathode dry-out and flooding was studied. These tests were realized at a current density of 0.5 A cm -2 . Operating parameters were the same as specified in section 3.1, but the cathode humidity level was adapted. Therefore, the relative humidity of the applied air feed (RHair) was varied between 0% and 100%.
The visualization of the current density distribution is shown in Figure 6 using relative values.
Single cell dry-out can be detected by current density distribution when non-humidified air is fed to the stack (Figure 6 (A)). A large area of low current density is obtained close to the air inlet (bottom left corner). Also at low relative air humidity of 20% (Figure 6 (B)), the current density is decreased close to the air inlet. This effect may be a result of membrane and catalyst layer dry-out and is in line with further studies on the impact of reactant inlet humidity [10].
Due to water production, the humidity of the air in the flow field increases along the flow channels and the current density maximum is located slightly shifted to the upper part of the cell. This detected dry-out effect is most visible for cell 10 close to the media inlet and less pronounced in cell 01. The measurements demonstrate that the impact of the air humidity is highest in the range of 0% to 50% (Figure 6 (A) to (C)). The maximum current density shifts towards the air inlet if the air inlet humidity is increased. The most homogeneous behavior is reached at 20% relative humidity, i.e. at conditions close to the nominal conditions (see Figure   3). The current density distribution changes in the range of 75% to 100% are not significant, but feed air humidification of 50% and more clearly results in a higher inhomogeneity of the current density compared to the nominal conditions. However, the difference between the individual cells in this region is less pronounced compared to the dry-out effect.
The quantitative assessment of the impact of the air inlet humidity on the performance homogeneity is given in Figure 7 (A). As already visible in Figure 6, cell 01 is almost not affected by the humidity variation. The homogeneity only slightly increases for the box values from ± 6% to ± 5% and for the extreme values from ± 20% to ± 15% with increasing air inlet humidity. In contrast, the homogeneity of cell 05 and cell 10 clearly increases with increasing air inlet humidity. Without humidification, the current density distribution is highly The comparison to the single cell voltages in Figure 7 (B) shows that a high cell voltage and thus a high cell performance are not always connected to high homogeneity in the cell behavior.
Non-humidified conditions result in low single cell voltage and low homogeneity. But the highest single cell voltages are achieved using 20% air inlet humidity, while highest homogeneity is achieved at higher humidity of at least 75%.
The single cell EIS responses during this study are shown in Figure 8. Apparently, the cell spectra are more distinguishable from each other compared to operation under nominal conditions in Figure 5 where all cell behaved similarly. The EIS analysis shows that the differences in Rel between the single cells is not affected by the humidity level, but Rel decreases significantly when the humidification is increased. As already visible in the current density measurements, the impact of the relative humidity on the cell performance is highest between 0% and 50% (Figure 8 (A) to (C)). In this region Rel decreases from 0.12 to 0.06  cm 2 and remains almost constant for high humidification (Figure 8 (G)). Consequently, the relevant diagnostic parameters for the assessment of cell dry-out phenomena are the Rel increase by EIS and the decrease of the current density at the air inlet by PCB. It was shown that the current density measurements are significantly more sensitive regarding the dry-out of individual cells for this application.
The analysis of Rpol (0.27  cm 2 ) and Rdiff (0.30  cm 2 ) shows that both values are also increased if the air feed is not humidified. Again, this seems to be caused by the electrolyte dry- out in the catalyst layer and the significant decrease of oxygen permeability through ionomer films [39], which limits the oxygen access to the catalyst. As soon as the fed air is humidified (Figure 8 (B) to (F)) Rpol increases from 0.16 to 0.30  cm 2 and Rdiff increases from 0.25 to 0.38  cm 2 (Figure 8 (G)) with increasing humidification. This indicates that formation of liquid water in the catalyst layer and in the flow field can also limit the oxygen availability at the catalyst surface. It can be observed that this increase in Rpol and Rdiff is different in the individual cells, which is clearly visible at 50% relative humidity in Figure 8 (C) and in the high standard deviation in Figure 8 (G). While the increase of these resistances in cells 03 to 09 is comparable, cells 01, 02, and 10 shows significant higher changes. Rpol for cells 01, 02, and 10 with (0.33 ± 0.09)  cm 2 is about 2.8-times higher compared to Rpol for cells 03 to 09 with (0.12 ± 0.03)  cm 2 . Rdiff for cells 01, 02, and 10 with (0.40 ± 0.07)  cm 2 is about 1.4-times higher compared to Rdiff for cells 03 to 09 with (0.29 ± 0.03)  cm 2 . There seems to be a transitional behavior regarding flooding phenomena at 50% relative humidity with direct impact on the cell performance. Cells 01, 02, and 10 are affected by flooding while the other cells are not. Above 50% relative humidity all single cells seems to be affected. For this flooding phenomenon, EIS has shown higher sensitivity compared to current density measurements, where the difference in the individual cell behavior was not detectable.

Impact of Air Starvation
To study the impact of air starvation, the stoichiometry of the air supply (air) was stepwise lowered starting from the nominal value 2.0 down to 1.5 using an air inlet relative humidity of 100%. All other parameters remain on the nominal values as specified in section 3.1.
The current density measurement results of this study at a load level of 0.5 A cm -2 are presented in Figure 9. While the current density distribution at an air stoichiometry of 2.0 is quite        Results are shown for RHair of 0% (A), 20% (B), 50% (C), 75% (D), 90% (E), and 100% (F).
Results are presented in relative values and scaled to ± 20% deviation from the set value of 0.5 A cm -2 . Cell and media flow configuration is given in Figure 1.   Results of EIS fitting is provided in (G) using the equivalent circuit given in Figure 2   Results are shown for air of 2.0 (A), 1.9 (B), 1.8 (C), 1.7 (D), 1.6 (E), and 1.5 (F). Results are presented in relative values and scaled to ± 20% deviation from the set value of 0.5 A cm -2 .
Cell and media flow configuration is given in Figure 1.