Measurement of Liquid Water Accumulation in a PEMFC with Dead-Ended Anode

Transcription

1 Donloaded Sep to Redistribution subject to ECS license or copyright; see B68 Journal of The Electrochemical Society, 55 B68-B /8/55 /B68//$. The Electrochemical Society Measurement of Liquid Water Accumulation in a PEMFC ith Dead-Ended Anode Jason B. Siegel, a, *,z Denise A. McKay, a, ** Anna G. Stefanopoulou, a Daniel S. Hussey, b, * and David L. Jacobson b a University of Michigan, Ann Arbor, Michigan 489, USA b National Institute for Standards and Technology, Gaithersburg, Maryland 899, USA The operation and accumulation of liquid ater ithin the cell structure of a polymer electrolyte membrane fuel cell PEMFC ith a dead-ended anode is observed using neutron imaging. The measurements are performed on a single cell ith 5 cm active area, Nafion -IP membrane, and carbon cloth gas diffusion layer. Even though dry hydrogen is supplied to the anode via pressure regulation, accumulation of liquid ater in the anode gas distribution channels as observed in most tested conditions. Moreover, the accumulation of liquid ater in the anode channels is folloed by a significant voltage drop. Anode purges and cathode surges are also used as a diagnostic tool for differentiating beteen anode and cathode ater flooding. The rate of accumulation of liquid ater, and its impact on the rate of cell voltage drop is shon for a range of temperature, current density, cathode inlet RH, and air stoichiometric conditions. Operating the fuel cell under dead-ended anode conditions offers the opportunity to observe ater dynamics and measured cell voltage during large and repeatable transients. 8 The Electrochemical Society. DOI:.49/ All rights reserved. Manuscript submitted June 9, 8; revised manuscript received August 4, 8. Published September, 8. The electrochemical poer generation of a popular category of fuel cells depends on the proton-conducting properties of their polymer electrolyte membranes. The ability of the membrane to conduct protons increases ith increasing ater content. Hoever, polymer electrolyte membrane fuel cells PEMFCs operate belo the boiling point of ater, causing excess ater to condense and restrict gas delivery or block the catalyst active area. The buildup of ater mass, referred to as flooding, in an operating fuel cell as first observed ith neutron imaging in Ref.. The impact of this flooding phenomena is a recoverable reduction in the poer output of the fuel cell, seen by a decrease in cell voltage,, but can also lead to irrecoverable material degradation. 4,5 This paper presents the neutron imaging of liquid ater accumulation in a PEMFC operating ith a dead-ended anode. Modeling and testing for anode flooding conditions is rare because most experimental fuel cells operate under flo-through conditions for hich anode flooding is highly unlikely. Here a solenoid valve placed donstream from the cell allos for an occasional purging event ith high hydrogen flo rate to remove ater from the anode, preventing severe voltage drop and reactant starvation. Optimal scheduling of purge events is necessary to prolong fuel cell life and minimize asted hydrogen. An adjustable purging schedule applied to a stack ith a dead-ended anode can enable high hydrogen utilization and remove the need for costly, heavy, and bulky anode humidification and recirculation hardare. Furthermore, understanding the ater removal ith anode purging is important for designing shutdon procedures to alleviate damage caused by freeze and tha cycling. Therefore models hich accurately predict the accumulation and removal of ater are necessary. If in addition to ater accumulation the model can also predict the resulting cell voltage, it could be used in combination ith voltage measurement to allo real-time adaptation of the purging events in response to component aging, environmental changes, and system faults. Specifically, the estimated voltage is compared ith the actual voltage, and the error is used to adjust the model-based purge schedule. A similar model-based technique as used for the estimation of hydrogen starvation 6 and control of the hydrogen production rate from a fuel processor. 7 Hence, the experiments and experimental results detailed here sought to confirm that a consistent correlation beteen ater accumulation and voltage degradation exists. To this end, our experimental results offer the opportunity to * Electrochemical Society Active Member. ** Electrochemical Society Student Member. z siegeljb@umich.edu observe ater dynamics and measured cell voltage during large and repeatable transients and hence provide useful data for calibrating and validating simple lumped parameter transient models 8,9 during anode flooding conditions. Despite some limitations of the imaging system, ith careful cell design and masking the approximate location of the liquid ater ithin the cell structure can be inferred, similar to the ork of Ref.,, and -6. Neutron imaging experiments ith higher resolution in the membrane through-plane direction could provide more information on the intrinsic mechanism of voltage degradation during anode flooding. Hoever, higher spatial resolution imaging requires longer exposure times 7 and hence, the method is impractical for calibrating or validating transient models ith a time resolution less than min. In our setup, e distinguish anode vs cathode channel flooding using controlled cathode surging and anode purging events. During an anode purge, fed ith dry hydrogen, a high gas flo rate through the cell is used to remove the liquid ater stored in the anode. During a cathode surge the air flo is momentarily and abruptly increased beyond the nominal excess ratio to remove any liquid ater stored in the cathode. It is demonstrated that liquid ater accumulates in the anode gas channels, and this buildup of liquid ater is ell correlated ith the dynamic cell voltage response during the majority of the experiments, as originally predicted in Ref. 8. This data also indicates that there is sometimes significant voltage drop, even if the mass of ater does not increase in the fuel cell. Hence, ater flooding cannot be considered alays responsible for the voltage drop and nitrogen accumulation in the anode should also be considered. 9, Our conclusions about channel ater mass are based on several homogeneity assumptions hich ignore the difference in the stored liquid ater mass in the gas diffusion layer GDL under the channels as compared to areas under lands., Nevertheless, the data and data analysis provide useful ne information regarding PEMFC transient behavior. Experimental Experiments ere conducted at the Neutron Imaging Facility at the National Institute for Standards and Technology NIST Center for Neutron Research. 7 We used the amorphous silicon detector for its Hz image acquisition rate to capture the change in mass of liquid ater over time. A cm diameter aperture as used for the experiment, ith a neutron fluence rate I of 7. 6 cm s and a neutron flight path of 6 m. In this case, the ratio of source-todetector distance over the source aperture diameter L/D as equal to 6. The idth of a pixel in the image corresponds to 7 m, but the resolution of the imaging system is about 5 m as a result of scintillator blooming. 7

2 Donloaded Sep to Redistribution subject to ECS license or copyright; see Journal of The Electrochemical Society, 55 B68-B78 8 B69 Hydrogen Bubbler Pressure Regulator T MFC Compressed dry air S P endplate heater Fuel Cell P,T T T P,T S V Load A P,T S MFC Pressure / Temp Measurement Solenoid Valve Mass Flo Controller Heater Figure. Color online Experimental hardare detailing sensor and actuator locations. The NIST test stand is used to supply a humidified gas stream to the cathode in flo-through operation. The air is humidified using a bubbler, and it is assumed that the gas leaving the bubbler is saturated at the ater temperature. The inlet relative humidity RH can be calculated at the fuel cell temperature by assuming the air has a depoint temperature equal to the bubbler temperature. A portable anode purging system as constructed, hich allos dead-ended operation of the fuel cell. This anode purge system consists of a pressure regulator, hich supplies dry hydrogen to the anode inlet, and a solenoid valve donstream of the anode outlet. A needle valve placed donstream from the solenoid valve is used to set the desired flo rate during an anode purge event. The pressure drop across the needle valve is used to reduce the gas flo rate leaving the channel during an anode purge event. This valve as adjusted prior to beginning and kept fixed for the duration of the experiments. The anode gas flo rate during a purge is about 5 slpm. Beteen purges the anode is supplied ith dry hydrogen via pressure regulation, hich provides a dry hydrogen flo at a stoichiometry equal to one, as shon in Fig.. Voltage, current, pressure, temperature, and cathode flo rate measurements ere recorded continuously at Hz resolution. The image data as selectively recorded hen repeatable large transients ere observed. Due to the large file size it as not desirable to capture data for all times, creating gaps in the measured liquid ater mass data. The cell as comprised of a single 5 cm Nafion -IP membrane c hich is 5.4 m thick ith anode and cathode catalyst layers containing a Pt areal density of. mg cm purchased from Ion Poer. SGL BB nonoven carbon GDLs ere used, hich have an uncompressed thickness of 4 m and a porosity of =.84. The cell hardare, purchased from Electrochem, consisted of aluminum endplates, gold-coated aluminum current collectors, and resin-impregnated graphite flo fields. Resin-impregnated graphite is used to prevent liquid ater from accumulating inside the pore structure of the graphite. The graphite plates ere thinned to. cm to reduce neutron attenuation. The anode gas channels are straight ith a channel idth of.8 mm, depth of.78 mm, and c Certain trade names and company products are mentioned in the text or identified in an illustration in order to adequately specify the experimental procedure and equipment used. In no case does such identification imply recommendation or endorsement by the National Institute of Standards and Technology, nor does it imply that the products are necessarily the best available for the purpose. Figure. Color online Neutron images of the fuel cell active area before and after anode purge events, indicating the actual cell orientation. The cell as operated at 566 ma cm, 55 C, ith fully humidified air at a stoichiometry of % for cathode and dry hydrogen supplied to anode. land idth of.88 mm. A semiserpentine flo path is used on the cathode ith five parallel paths, each having a channel idth of.686 mm, a channel depth of.99 mm, and a land idth of.76 mm. A square 45 W resistive heater, ith a surface area of 58 cm, as attached to each end plate to ensure uniform heating and maintain the desired temperature at lo current density. Quantification of Liquid Water Mass The attenuation properties of a neutron beam passing through an object can be used to form a projected image on a detector, similar to X-ray radiography. Neutrons are strongly scattered by hydrogen atoms and only eakly interact ith the other materials used in PEMFCs, such as aluminum and carbon. The scattering interaction is also density dependent; therefore, neutron imaging is particularly useful for measuring liquid ater present in the cell due to the high density of hydrogen atoms in liquid ater but is unable to detect hydrogen gas or ater vapor. This technique provides a useful tool for in situ measurement of liquid ater content hile employing commercial fuel cell materials ith realistic cell designs. The use of these materials reduces the possibility of temperature gradients induced by selection of nonstandard materials, hich ould otherise greatly impact the formation of liquid ater ithin the cell. Figure shos to images collected before and after a purge event. The areas ith high liquid ater accumulation are clearly visible in the top left frame, hich corresponds to a condition before an anode purge event. The second frame in Fig. shos the drier neutron image collected after the purge. The donard pointing vector g indicates the direction of gravity. The image, I j,k, formed by scattering and absorption of neutrons can be modeled by the Beer Lambert la N I j,k = I j,k exp l t l j,k l= here I j,k is the incident flux rate of neutrons, N is the number of material layers in the object, t l is the thickness of each layer, and l is the linear attenuation coefficient of each layer. In order to quantify the thickness of liquid ater, e must determine the amount of attenuation hich is caused by ater in the image. A section of the image outside the active area, defined by S P,

3 Donloaded Sep to Redistribution subject to ECS license or copyright; see B7 Journal of The Electrochemical Society, 55 B68-B78 8 is used to normalize each image before processing to account for fluctuations in the nuclear reactor poer output and I j,k over time, according to I norm I j,k j,k = j,k SP I j,k To reduce systematic error, care as taken to ensure reference images, I norm dry j,k, ere taken of a completely dry cell, and 5 images ere averaged to reduce noise in the image. In order to account for the effects of thermal expansion, hich cause portions of the cell to move ithin the imaging frame, dry reference images ere taken at 5 C increments over the operating temperature range of the fuel cell. The relative neutron transmission, found by dividing an image by the dry reference image of the cell, can be used to calculate ater thickness, assuming that the attenuation caused by everything but liquid ater remains unchanged since the reference image as captured. A median filter is applied to the images to further reduce random noise before calculating the liquid ater thickness. The thickness of the ater layer as calculated using t j,k = ln Inorm I norm dry j,k and the experimentally obtained attenuation coefficient =.78.8 mm, 7 hich is valid for liquid ater thickness less than mm. Liquid ater quantification is further complicated by instrument broadening caused by the scintillator screen, scattering of neutrons by objects in the beam, and detector noise., The trends in the data ill still be evident ithout correction for these effects. Quantification of the error introduced by each phenomena is currently under investigation at NIST. 7 Temporal averaging to reduce uncertainty. The measured uncertainty in ater volume for a single pixel in the detector system sampled at the Hz rate is. 6 cm using L/D = 45, 7 hich is different from our setup L/D = 6. The measured uncertainty can be scaled to account for the different neutron flux used in the experiment by multiplying by the ratio of the square root of the fluence rates, hich yields a per pixel liquid ater volumetric uncertainty of.57 6 cm. This relationship can be derived from the equation for the theoretical uncertainty 7 t = +e t I o AT 4 I o AT here t is the ater thickness, is the attenuation coefficient, I o is the fluence rate, A is the area, T is the integration time, and is the detector efficiency. Equation 4 also indicates that a reduction in uncertainty of the measured liquid ater volume can be achieved at the expense of temporal resolution by averaging several frames. The uncertainty in an averaged frame of s as compared to s is decreased by a factor of /T.. Because the time constant associated ith liquid ater accumulation in the GDL is on the order of s,,4 the s imaging rate as used. The resulting liquid ater volume uncertainty is.6 6 cm pixel, hich corresponds to a ater thickness of m. Unfortunately, the temporal averaging e performed exacerbates the difficulty in tracking movement of ater droplets and slugs in the channels. Liquid ater droplet movement ith the gas stream in the channel could be much faster than the s imaging time, and therefore during a cathode surge or anode purge liquid ater flo in the channel ould appear blurred, even ithout averaging at the fastest exposure rate of the imaging system. Masking. The processed ater thickness image, t j,k, isa to-dimensional D projection of the liquid ater inside the fuel cell. In order to infer the location of liquid ater in the third dimension, a knoledge of the physical material structure can be combined ith logical arguments. For example, liquid ater cannot be Anode Channels Cathode Channels Figure. Schematic of fuel cell layers. The beam direction is into the page and perpendicular to the membrane. located ithin an impervious solid material. Several masking techniques have been used to analyze neutron images.,,,4 We employ a similar process to estimate the mass of liquid ater in three layers: the anode channel, cathode channel, and combined membrane MB and GDL layer GDLMB, hich consists of both anode and cathode GDLs. The masks are formed using the draings of the graphite plates. Mask A defines the active area of the fuel cell, mask B is for the anode channels, and mask C is for the cathode channels. Four mutually disjoint masks, D G, identify regions of the fuel cell corresponding to the different possible combinations of channels and lands on each side of the membrane as shon in Fig.. The area of each mask and their logical relationships are shon in Table I. The ater thickness in the combined membrane and GDL layer can be directly measured from the images for regions identified by mask G, hich contain both anode and cathode lands and therefore ater could not be present in either of the other to layers. Local spatial averaging. Estimation of the distribution of liquid ater in the third dimension using masking over the entire active area requires consideration of the significant spatial variations in the liquid ater thickness along the channels. Therefore, a local D spatial average is constructed. Specifically, the active area is partitioned into a 9 9 grid of 8 segments similar to Ref.. A 9 9 grid is chosen corresponding to the nine segments of parallel cathode channels as seen in Fig. 4, hich is overlaid on mask C. Within each of the 8 segments, the flo of reactant gases is in the same direction. Due to a manufacturing error the cathode inlet and outlet manifolds ere incorrectly positioned, rendering the first and last serpentine flo pass ineffective. The cathode inlet should be Table I. Masks. escription No. pixels Relationship A Active area,6 D E F G B Anode channel area 4,99 C Cathode channel area 69,9 D Both Ch + MB + GDL 8,47 B C E An Ch + MB + GDL 6,5 B C A F Ca Ch + MB + GDL 4,45 B C A G MB + GDL only 8,55 B C A H Subset of G,9 H G D E F G

4 Donloaded Sep to Redistribution subject to ECS license or copyright; see Journal of The Electrochemical Society, 55 B68-B78 8 B7... Average Over All Sections Mask H Average Over Section 9 (inlet) (a) Figure 4. Color online The cathode channel, mask C, is partitioned into 8 segments for the evaluation of the spatial distribution in cathode channel liquid ater. Five representative segments are highlighted and shon in later analysis. located in segment, for proper operation, but is instead located in segment 9 as shon in Fig. 4. As a result, segments 8 and 74 8 are not included in the flo path; therefore, no excess air is passed through these segments of the cathode during operation of the cell. Hence, the cathode channel exhibits an unusual accumulation of liquid ater in these segments. Due to the orientation of the cell, prior to masking, each thickness image is rotated 44. to the right using a bicubic interpolation method to facilitate registration ith the mask. In the rotated orientation, studies can be performed along the channels by traversing the coordinate axis of the image. Note that segment 9, hich is highlighted in Fig. 4, corresponds to the upper corner of the fuel cell shon in Fig.. Segment 9 is also the location of the gas inlets here the dry hydrogen and humidified air enter the anode and cathode channels, respectively. The fuel cell conditions are assumed to be uniform ithin each of the 8 segments. Five image masks D H are applied to each of the 8 segments individually, and the average liquid ater thickness in each of the masked regions is calculated for each segment. A fifth mask H, hich is a subset of the points in mask G hich are not adjacent to either channel, is used to get a more accurate estimate of the ater content in the combined GDL and membrane layer. This mask reduces the effects of instrument broadening and neutron scattering hen there is liquid ater present in the neighboring channels.,7 Since only points near the center of the lands are counted, the effect of different gas flo distributions in the GDL in the adjacent regions, underneath the channel as compared to the land, on liquid accumulation may not be accurately represented. The top subplot in Fig. 5 shos a comparison of the average ater thickness calculated using masks G and H over the entire active area for the combined membrane and GDL layer. When considering the entire active area, there is little difference beteen the average ater thicknesses hen using mask G as compared using mask H. Hoever, for individual segments, a more significant difference beteen the applications of mask G vs mask H is observed, as indicated in the remaining subplots. The largest difference in average thicknesses is found in segment 7, located at the gas outlet Water Thickness (mm) Average Over Section Average Over Section 7 Average Over Section 6 Mask H std Mask H Average Over Section 7 (outlet) Figure 5. Color online Measured average liquid ater thickness in the combined membrane/gdl layer for the entire active area top subplot and ithin selected segments subsequent subplots from the same data set shon in Fig. 6. near the bottom of the cell. Due to the orientation of the cell and dead-ended anode, liquid ater accumulates in this corner of the anode channel because it is forced by the reactant gas flo and gravity. A detailed description of the experimental testing conditions for this experiment can be found in Fig. 6. The standard deviation in ater thickness for points identified by mask G, shon in Fig. 5, provides a metric for ho uniform the ater content in the combined membrane and GDL layer is over that segment. A large standard deviation in thickness may indicate nonuniform conditions in the membrane and/or GDL. Scintillator blurring and/or scattering hen a significant amount of ater is present in the channel reduces the measured standard deviation but creates a systematic uncertainty. In the limiting case, hen the ater thickness is uniform, the standard deviation in measured thickness should be equal to that of the Poisson counting process given by Eq. 4, hich increases ith increasing ater thickness. The large standard deviation measured for segment 9 beteen t = 75 and 8 min, shon in Fig. 5b, is likely due to condensation and buildup of liquid ater near the cathode inlet manifold. A cathode surge conducted at around t = 78 min pushes liquid ater out of segment 9, reducing the measured standard deviation in that segment. Segment 7 consistently contained the most ater and therefore also has the highest standard deviation. Channel liquid mass. To calculate the mass of liquid ater in the gas channels, assume that the ater thickness in the combined (b) (c) (d) (e) (f)

5 Donloaded Sep to Redistribution subject to ECS license or copyright; see B7 Journal of The Electrochemical Society, 55 B68-B78 8 CaCh AnCh Volt (mv) i (A cm ) T( C) CA RH (f) (h).5 (i) membrane and GDL layers is uniform over a segment, and then it is equal to the averaged thickness in the region defined by mask H. In this case the average liquid ater thickness in the anode channel can be estimated by taking the difference beteen averaged ater thickness in masked regions H GDLMB and E GDLMB + AnCh for each segment. Similarly, the average cathode channel liquid ater thickness can be estimated by the difference beteen averaged ater thickness in regions H and F. The total estimated liquid ater mass in each layer of the cell is found by summing the masses calculated for each of the 8 segments, according to W M AnCh M CaCh W Fig. 7(b) Fig. 7(a) Fig. 7(c)Fig. 7(d) W M GDLMB 8 = l A P N P,AA i H i i= 8 = l A P N P,AnCh i E i H i i= 8 = l A P N P,CaCh i F i H i i= Figure 6. Color online Data set : Measured total liquid ater mass, estimated GDL MB and channel liquid ater masses, voltage, current density, temperature, CA RH, and ; fully humidified cathode gas air and dry hydrogen inlet. The red vertical lines in the voltage subplot indicate anode purges. here l is the density of ater, A P is the area of the fuel cell corresponding to a single pixel in the detector, N P,AA i, N P,AnCh i, and N P,CaCh i are the number of pixels defining the active area, anode channel, and cathode channel, respectively, and H i, E i, and F i are the measured average ater thickness in segment i for each of the corresponding masks, here the thickness is given by (a) (b) (c) (d) (e) Table II. Cell operating conditions. Parameter Operating range Anode inlet RH % Cathode inlet RH 4 % Anode pressure.6 kpa absolute Cathode pressure 5 kpa absolute Cell temperature 4, 5, and 6 C Cell current A Cell current density 566 ma cm Eq.. The results of this image processing and data analysis are shon in Fig. 6. Three-dimensional computational models have demonstrated the possibility of a nonuniform liquid ater distribution inside the GDL, ith a buildup of ater in regions underneath channel lands., If the true ater distribution in the GDL is concentrated under the lands, then the previous assumptions ould lead to an underestimation of the amount of liquid ater in the channel, and an overestimation of the GDL ater content. Therefore this data analysis method represents a loer bound on the mass of accumulated ater in the channels and an upper bound on the mass of ater in the GDL. Local averaging combined ith masking yields more accurate results than simply averaging the masked regions over the entire active area due to the nonuniform distribution of ater ithin the cell. The percent difference beteen the total liquid ater mass and W the estimated mass calculated via masked local averages, M AnCh + M CaCh W W + M GDLMB, is less than %, indicating that the relative uncertainty introduced by the assumption of a uniform ater thickness in the combined membrane and GDL layer for each segment is small. Therefore e have a high level of confidence using the locally averaged and masked data to infer the mass of liquid ater in each of the three layers: anode channel, cathode channel, and combined membrane GDL. Results and Discussion The neutron-imaging data ere collected over 4 continuous days of testing 4 h for the range of operating conditions summarized in Table II and shon in Fig. 6-. The same cell as used throughout all the experiments, ith pressure regulated dry hydrogen feed at various current densities, air stoichiometric ratios, cathode inlet RHs, and cell temperatures. It as assumed that the air supplied to the cathode had a depoint equal to the bubbler temperature. Control of the fuel cell endplate temperature and cathode bubbler temperature can therefore be used to achieve the desired cathode inlet RH. Relatively lo current densities ere used as compared to those typically reported in the literature. These lo-to-medium current density conditions are interesting in portable and mobile applications here the system typically idles for a significant time. They are also conditions for hich anode flooding is most predominant. Hoever, at higher current density electro-osmotic drag tends to dry out the anode. 5 One can notice that our operating temperature is also loer than PEMFCs are intended for, but this range as not intentional. Due to the softare temperature calibration, the intended operating conditions ere not attained and the cell as operated at a temperature colder than desired. The thermocouple as recalibrated after the experiment, and the corrected temperature is presented in the data. Because the bubbler temperature as higher than the fuel cell temperature for some of the experiments, in this case the depoint of the air entering the cathode channel as higher than the cell temperature, hich is indicated by a calculated RH greater than, as shon in Fig. 6. In the to subsections belo e first present the results of the neutron image processing and demonstrate ho the effects of the anode purge and surge events on the measured voltage can be used

6 Donloaded Sep to Redistribution subject to ECS license or copyright; see Journal of The Electrochemical Society, 55 B68-B78 8 B7 Figure 7. Color online Processed neutron images shoing liquid ater thickness mm under a variety of conditions. as a diagnostic tool to distinguish beteen operation ith anode channel ater flooding anode plugging or cathode channel ater flooding cathode plugging conditions. Snapshots of the liquid ater thickness images are shon to verify the purging and surging diagnostic capability. These snapshots also clarify differences in the D flo pattern, hich affects the quality or ability of the purge event to remove ater accumulated in the anode channels. Finally, e point to portions of the data here the ater visualization indicates no anode plugging, hoever, repeatable voltage improvement is seen after every purge. These data sets indicate that there is i an anode nitrogen plugging problem and ii a GDL ater mass value belo hich there is no liquid ater flux to the anode channels and hence, anode ater plugging is avoided. This information can be used to tune a value for the immobile saturation of liquid ater inside the GDL, hich is used in later sections to calibrate a simple model. Channel ater plugging. Figure 6 shos the data from one of our full days of testing 6 h continuous operation, and Fig. 8- shos the rest of the data collected on three other days. A detailed discussion of the data shon in Fig. 6 is presented first. This data illustrates ho changing the operating conditions can drive the system from a state of cathode channel flooding to anode channel flooding at lo-to-medium current density. The first subplot of Fig. 6a shos the mass of liquid ater in each of the four mutually disjoint masked regions, described previously, over the entire active area. The next three subplots sho the estimated liquid ater mass in each of the three layers anode channel, cathode channel, and the combined membrane/gdl layer. The last four subplots of this figure display the important operating conditions hich are typical model inputs: 8 current density A cm, end plate cell temperature C, cathode inlet RH, and cathode stoichiometric ratio. The cell is initially operated at lo current density Fig. 6f and high cathode stoichiometric ratio Fig. 6i, and heaters are used to maintain the endplate temperature at 5 C Fig. 6g. Figure 6h shos the estimated cathode inlet RH CA RH hich is the ratio of the saturation pressure at the bubbler temperature over that of the cell temperature. At t = 78 min the cathode stoichiometry is decreased and ater begins to accumulate in the cell, as seen in Fig. 6b and c. Cathode channel flooding is evident in the neutron radiographs, hich can be seen in Fig. 7a at t = 8 min, and can be inferred by the rapid but small voltage fluctuations. This fluctuation in voltage occurs as the cathode channels are intermittently plugged ith ater, and the voltage is insensitive to the anode purges indicating the absence of anode plugging. We use the terms flooding and plugging to differentiate beteen the accumulation of ater mass in the GDL and channel. Water accumulating in the channels forms slugs hich block the flo of reactants along the channel; hence, the adoption of the term plugging. Figure 7a also shos the buildup of liquid ater after the 8 bends in the cathode semiserpentine flo path as observed in Ref. 6. Hoever, the ater in the cathode channel only accumulated after the 8 bends on the left side of the image, hich corresponds to the donhill turnarounds hen vieed in the original orientation as shon in Fig.. The gravity vector, g, delineates the image rotation relative to the cell orientation. The orientation of the cell and the effects of gravity may have a non-negligible effect on the

7 Donloaded Sep to Redistribution subject to ECS license or copyright; see B74 Journal of The Electrochemical Society, 55 B68-B78 8 CaCh AnCh Voltage (mv) i (A cm ) T( C) CA RH Fig. Fig (f) (h).5 (i) Figure 8. Color online Data set : Measured total liquid ater mass, estimated membrane/gdl and channel liquid ater masses, voltage, current density, temperature, CA RH, and. Subsaturated cathode gas air and dry hydrogen inlet. The red vertical lines in the voltage subplot indicate anode purges. (a) (b) (c) (d) (e) CaCh AnCh Voltage (mv) i (A cm ) T( C) CA RH Fig. 4.6 (f) (h).5 (i) Figure 9. Color online Data set : Measured total liquid ater mass, estimated membrane/gdl and channel liquid ater masses, voltage, current density, temperature, CA RH, and. Subsaturated cathode gas air, sitching to fully humidified at t = 8 min and dry hydrogen inlet. The red vertical lines in the voltage subplot indicate anode purges. (a) (b) (c) (d) (e) location of liquid ater buildup ithin the channels. 7 This is different from common modeling assumptions of transport in the GDL, here gravitational effects are negligible. 8 Just before t = 85 min the cathode is surged, removing liquid ater from the cathode channels, and a sustained voltage recovery of 5 mv is observed. Folloing the surge, the current density and stoichiometry are both increased. The higher cathode gas velocity more readily removes liquid ater from the cathode channel, as seen in Fig. 7b at t = 8 min. This ne operating condition eventually leads to anode channel flooding, shon in Fig. 7c and clearly detected by the large periodic voltage increase folloing every purge. At t = 858 min the cathode stoichiometry is again decreased, leading to a condition in hich both anode and cathode channel flooding occurs, as seen in Fig. 7d and detected by the small voltage fluctuations superimposed by purge-induced voltage improvements. Data at loer cathode inlet RH 7% also resulted in anode channel flooding, shon in Fig. 9 near t = 7 min. A good example of cathode channel flooding can be seen shortly after startup in Fig. 8 at t = 5 min, here a 5 mv sustained voltage recovery occurs folloing a cathode surge. Cathode flooding is also present at t = 554 min in Fig. 9, under similar conditions of lo cathode inlet RH and lo current density 89 ma cm. At the to higher current densities hich ere tested, cathode surges did not provide any sustained voltage recovery, indicating that cathode channel flooding as not a problem under these conditions. Anode channel nitrogen plugging. In this section e discuss the portions of the data and delineate the limitations of the purging and surging methodology used for detecting the approximate location of ater in the direction perpendicular to the membrane plane. The first limitation is apparent hen the D distribution of ater along the channels is such that a purge or a surge cannot remove a significant amount of ater. At loer current density, 89 ma cm, the liquid ater as more evenly distributed, rather than concentrated at the end of the channel. Due to the lo rate of ater generation and subsequently small droplet groth rate, the liquid droplets on the surface of the anode GDL did not gro large enough during the observed time beteen purges so that they ould be pulled by gravity to the end of the channel. It as observed that only liquid ater at the end of the anode channel is effectively removed by purging. This is seen in the buildup of liquid ater mass in the anode channel, Fig. 8 beteen t = 75 min and in Fig. 9 beteen t = 55 and 65 min, hich is insensitive to anode purges. At higher current densities, there is a greater liquid production and crossover rate; hence, ater collects near the end of the anode channels, as shon in Fig. 7c, and repeatable liquid ater transients occur. The second limitation in using the voltage-purge relation for diagnosing anode ater plugging arises from the effects of nitrogen crossover and eventual nitrogen accumulation in the anode. In the event of nitrogen flooding, an anode purge can also improve the cell voltage. In the collected data there are cases here the image processing indicates that there is no anode ater plugging; hoever, there is a repeatable voltage improvement after every purge event. Specifically, Fig. t = min shos data at higher temperature 6 C and lo cathode inlet RH 4%, here anode channel flooding as no longer detected in the neutron images. For these conditions, the total ater content in the GDL decreases sig-

8 Donloaded Sep to Redistribution subject to ECS license or copyright; see Journal of The Electrochemical Society, 55 B68-B78 8 B75 CaCh AnCh Voltage (mv) i (A cm ) T( C) CA RH (f) (h).5 (i) Figure. Color online Data set 4: Measured total liquid ater mass, estimated membrane/gdl and channel liquid ater masses, voltage, current density, temperature, CA RH, and. Both saturated and subsaturated cathode gas air conditions ith dry hydrogen. At t = 8 min the bubbler is filled ith colder ater, causing a drop in inlet CA RH. The red vertical lines in the voltage subplot indicate anode purges. nificantly, about half that of the case ith a fully humidified cathode inlet, because the ater carrying capacity of air increases exponentially ith temperature and therefore can remove far more liquid ater under these conditions. 6,9 Nonater anode plugging conditions are also shon in Fig. 8 t = min hile operating at relatively lo current density 77 ma cm, high cathode stoichiometric ratio, high temperature, and subsaturated cathode inlet flo. Nonater anode plugging conditions ere also encountered in the first day of testing shon in Fig. 6 t = 7 78 min and t = min here fully saturated air flo as fed to the fuel cell. Some discussion of these to cases is arranted, as the membrane ater dynamics and the thermal dynamics 6 might explain hy e do not have ater anode plugging in these to cases. The period t = 7 78 min in Fig. 6 as after start-up here the membrane might not have been fully hydrated and hence, might have been absorbing ater from the surrounding GDLs. At t = 95 min a slo temperature rise of about 5 C as measured at the end plates, folloing the current density increase to 566 ma cm. It is expected that the current increase causes a large spatial gradient in temperature. The thermal history of the membrane impacts the membrane ater uptake and conductivity., This thermal history may account for the absence of anode flooding folloing the return to loer current density and nominal temperature at t = min in Fig. 6, conditions hich previously t = 8 85 min exhibited anode channel flooding. In all the cases here nonater anode plugging is observed, except the start-up period of Fig. 6 t 78 min, the combined (a) (b) (c) (d) (e) GDL plus membrane ater mass, estimated through masking, as less or equal to.5 g, hich translates to a volume fraction of liquid ater of s =.. Because no liquid ater is floing to the anode channel, e assume this value to be equal to the immobile saturation limit s im. A linear GDL compression of % is assumed for this calculation, hich yields a compressed porosity of =.67, and e assume that the ater is distributed evenly beteen the anode and cathode GDL layers. Flooding, Plugging, and Voltage Response A consistent correlation beteen voltage and anode purging as observed at lo-to-medium current density under a variety of operating conditions. This relationship can be used to establish a phenomenological model that connects the accumulation of ater and nitrogen in the channel ith an apparent active area available for the reaction and hence, establish an easily tunable model ith lo computational complexity for control applications. 8,9 The hypothesis of a direct effect of anode channel ater and nitrogen mass accumulation on voltage is controversial, as most existing models use the decrease of reactant pressure and the effect of liquid ater accumulation in the cathode GDL-catalyst interface on voltage,, assuming a thin-film model hich predicts the blockage of the cathode catalyst interface. A similar argument is used here to formulate an anode channel interfacial blockage hich leads to an apparent active area, hence, an apparent current density and consequently, a voltage degradation, as shon in the Appendix. We first focus on the testing periods here voltage decay is observed but no channel, nor GDL ater mass is accumulating. A simple model using published membrane crossover values for nitrogen 9 confirms our hypothesis that nitrogen crosses over from the air supplied cathode through the membrane, accumulates in the anode, and contributes to the measured voltage degradation. The confirmation of the hypothesis is based on the good match beteen the predicted and the measured voltage degradation beteen purges. The exact accumulation of nitrogen in the anode channel has to be confirmed ith independent nitrogen concentration measurements, hich ill be the subject of future ork. We then focus on certain periods of testing that sho anode flooding and voltage decay and demonstrate that a simple onedimensional D channel-to-channel through the membrane, isothermal, and lumped parameter model can be used to predict the transient anode ater accumulation and voltage behavior ell during anode flooding conditions under dead-ended operation. 8,8,4 For the model validation e first use test periods ith anode flooding ithout concurrent cathode flooding. Another detailed graph shos a period here anode flooding and voltage decay exhibit a characteristic to-slope behavior, corroborating the hypothesis of voltage decay due to anode plugging instead of cathode catalyst flooding. Finally, a detailed vie of a testing period ith concurrent anode and cathode flooding is used for the model validation. Anode nitrogen plugging and voltage response. Figures 6, 8, and contain boxes hich highlight a portion of the experiments here a small voltage degradation, mv min, is not correlated to any liquid ater accumulation. As one can see, the voltage pattern is repeatable beteen purges despite the lack of any similar pattern in the ater observed. The portions of the experimental data indicate that nitrogen accumulation in the anode channels should be responsible for the relatively small mvmin vs the larger 5 8 mv min voltage drop that e typically observe during anode ater plugging conditions. To investigate the effects of nitrogen a channel-to-channel lumped parameter mass balance can be used to calculate the rate of nitrogen accumulation using the permeation coefficient k N = 4 mol m s Pa. 9 At the pressure conditions highlighted in Fig. 6 t = min, a predicted nitrogen accumulation of.5 mg min in the anode leads to a drop of hydrogen partial pressure of kpa min. Assuming homogeneous channel condi-

9 Donloaded Sep to Redistribution subject to ECS license or copyright; see B76 Journal of The Electrochemical Society, 55 B68-B78 8 tions, the reduction in hydrogen partial pressure affects the cell voltage through Eq. A-4 see the Appendix, and this leads to a voltage drop of only.6 mv min instead of mv min observed here. Because a large change in hydrogen partial pressure alone cannot account for the significant voltage drop observed in the data, e model an effective active area, hich represents the nonblocked area that the reactants can easily reach the catalyst sites. Using the effective area, e can calculate the apparent current density, hich is caused by displaced hydrogen in the anode channel. Specifically, in a pressure-regulated cell ith dead-ended anode, nitrogen ill be pushed by the feed gas and accumulate at the end of the channel. 9 So if complete anode channel plugging is assumed to occur hen % of the anode cross-sectional area has been filled ith nitrogen, then the predicted nitrogen accumulation of.5 mg min ill result in a voltage decay of. mv min using Eq. A- and A- see the Appendix from A app, by replacing t * ith the channel depth, m * an,ch ith the anode channel accumulated nitrogen mass, and * ith nitrogen density. It is most likely that accounting for the diffusion and mixing beteen hydrogen and nitrogen in the channel using an along-the-channel model ill correct for the estimated higher voltage drop. Nevertheless, the simplicity of the model provides a remarkable match ith the experimental observations of nitrogen plugging effects on voltage response. Anode ater plugging and voltage response. A summary of the imaging data indicates that for the majority of the tested conditions anode ater plugging should be considered in order to model the cell performance. Nitrogen should clearly also be considered, but the accelerated rate of voltage decay hen ater plugging is observed requires a more rapid control action and better modeling to prevent potential damage to the cell. A model hich includes the effect of both nitrogen and liquid ater on voltage requires a more detailed channel model to determine the reduction in active area. Standard diffusion and density values for the hydrogen and nitrogen gases and a simple along-the-channel model, not-yet published, indicate that the anode constituents form stratified layers ith liquid ater, folloed by a nitrogen-rich and then a hydrogen-rich area as e approach the channel inlet. The resulting fuel cell area, hich is unavailable to support the reaction, is given by the sum of the nitrogen and ater-covered areas. The mass of liquid ater accumulated in the anode channel is calculated using ell-knon equations for the membrane ater transport 8,5-7 and the D channel-tochannel, through-membrane plane, isothermal, to-phase liquid and vapor flo of ater through the GDL. For a detailed description of the ater crossover through the membrane, the to-phase flo through the GDL, and the accumulation of liquid in the channel, see Ref. 8. The neutron imaging data, ith the applied masking techniques, confirm the presence of liquid ater in the anode channel, as shon in Fig. for a typical purge cycle hich as taken from the third day of testing, shon in Fig. 9. The vertical lines shon in the voltage subplot indicate anode purge events. Folloing an anode purge, all of the liquid ater in the anode channel is removed and then sloly accumulates until the next purge event. Note that there is little change in the cathode channel ater mass beteen or during purges. Folloing the anode purge, a cathode surge is conducted at approximately 96 min. During the surge, the cell voltage increases due to the higher oxygen partial pressure. Hoever, after the cathode flo rate is restored, the voltage returns to its previous value. The mass of liquid ater in the cathode channel decreases as a result of the increased ater removal rate ith higher gas flo rate during the surge. In some experiments, even hen the cathode surge event occurs just preceding the anode purge event, the voltage continues to degrade at the same rate as experienced before the cathode surge. These results indicate that liquid ater in cathode channel has little effect on voltage under these conditions. In contrast, folloing the to anode purges at 947 and 96 min, the voltage recovery is significant and sustained until liquid ater begins to accumulate in CaCh AnCh Volt (mv) Model Data anode purges cathode surge Figure. Color online Measured cell voltage degradation and liquid ater mass accumulation beteen anode purges, taken from the third data set. These experiments ere conducted ith fully humidified air at a current density of 78 ma cm, a cell operating temperature of 5 C, and an air stoichiometry of %. the anode channel again. There exist other operating conditions, here accumulation of ater in the cathode is detected and a sustained voltage recovery occurs folloing a cathode surge. Figure shos no liquid ater accumulation in the cathode channel, ith anode channel ater accumulation beteen purges. Folloing the anode purge at t = 45 min, the imaging data shos a period ith no ater accumulation in the channel folloed by a linear increase ith time. This distinct, to-slope behavior of liquid ater accumulation can be explained by a strong anode purge hich removed liquid ater from the GDL. In this case, ater accumulation in the channel has to modes, a slo mode hich represents vapor transport through the GDL ith condensation along the GDL. Once the accumulation of ater in the GDL has surpassed the immobile saturation limit, 8,8,8 then both capillary liquid and vapor flo into the channel are present, yielding a faster rate of liquid ater accumulation in the channel. Hence, folloing an anode purge hich removes ater from the GDL, it takes time before ater begins to accumulate in the channel again. The observed voltage degradation under this condition is very ell correlated ith the to-slope liquid ater accumulation in the anode channels and has excellent repeatability. The simulated response using the channelbased apparent current density shos an excellent match despite the to-slope behavior. It ould be interesting to investigate hether voltage models that depend on catalyst flooding ater accumulation at the membrane catalyst interface alone can capture the observed behavior. Figure shos the voltage during a substantial change in liquid ater mass for both the anode and cathode channels. Specifically,