particle image velocimetry measurements in a model proton...

I

Pwlrcpantslcpr

etnSrimcthbecmwppt

T5

3

Downloa

J. P. Feser

A. K. Prasad1

e-mail: [email protected]

S. G. Advani

Department of Mechanical Engineering,University of Delaware,

Newark, DE 19716-3140

Particle Image VelocimetryMeasurements in a Model ProtonExchange Membrane Fuel CellParticle image velocimetry was used to measure 2D velocity fields in representativeregions of interest within flow channels of interdigitated and single-serpentine protonexchange membrane (PEM) fuel cell models. The model dimensions, gas diffusion layer(GDL) permeability, working fluid, and flow rates were selected to be geometrically anddynamically similar to the cathode-side airflow in a typical PEM fuel cell. The model waseasily reconfigurable between parallel, single-serpentine, and interdigitated flow fields,and was constructed from transparent materials to enable optical imaging. Velocity mapswere obtained of both the primary and secondary flow within the channels. Measurementsof the secondary flows in interdigitated and single-serpentine flow fields indicate thatsignificant portions of the flow travel between adjacent channels through the porousmedium. Such convective bypass can enhance fuel cell performance by supplying freshreactant to the lands regions and also by driving out product water from under the landsto the flow channels. �DOI: 10.1115/1.2744053�

Keywords: PEM, PIV measurements, serpentine, interdigitated, convective bypass

ntroductionConvective flow affects several key processes of hydrogen

EM fuel cell operation. Particularly important in the cathodehere diffusion is less dominant, these processes include the de-

ivery of high-concentration reactant oxygen to the catalyst layer,emoval of product water �vapor and liquid�, and in some appli-ations the removal of heat. There are two primary fuel cell com-onents that determine the effectiveness of convection. The firstre the bipolar plates, which affect the flow by means of the chan-el geometry. By tailoring the channel design, it is possible con-rol the pressure gradients and velocity fields in fuel cells. Theecond component that affects convection is the gas diffusionayer. Despite its name, the gas diffusion layer is porous and thusapable of fluid convection when pressure gradients exist. Theermeability of the gas diffusion layer then plays an importantole in determining convective flow patterns.

Not surprisingly, there is a need to experimentally verify theffectiveness of convective flow in various fuel cell designs. Thus,he development of spatially resolved flow measurement tech-iques for fuel cell applications is a highly active area of research.everal researchers have developed transparent fuel cells prima-ily designed to qualitatively observe the transport of liquid watern hydrogen fuel cells �1,2� and carbon dioxide bubbles in direct

ethanol fuel cells �3–6�. One quantitative technique that has re-ently been developed is thermal neutron imaging �7–9�. Thisechnique operates by directing a collimated neutron beam into aydrogen fuel cell; the neutron beam is scattered by liquid watery an amount related to the thickness of water. By scanning thentire fuel cell, a 2D map can be created of water content in theell. Despite the success and importance of these techniques, itust be noted that they observe the movement of product liquidater as opposed to the gaseous reactant species. Although com-uter models for predicting reactant flow have become quite so-histicated, the current problem would require a 3D computa-ional fluid dynamics �CFD� code �either direct numerical

1Corresponding author.Submitted to ASME for publication in the JOURNAL OF FUEL CELL SCIENCE AND

ECHNOLOGY. Manuscript received February 13, 2006; final manuscript received June
, 2006. Review conducted by Abel Hernandez.
28 / Vol. 4, AUGUST 2007 Copyright © 20

ded 23 Aug 2007 to 128.175.116.163. Redistribution subject to ASM

simulation or coupled with the correct turbulence model� and achannel/GDL interaction model that must simultaneously solvefor flow through the porous GDL. Current simulations showpromise in this regard but require validation against experimentaldata. Particle image velocimetry �PIV� can not only provide de-tailed velocity data for comparison and validation of such CFDmodels, but can also directly provide guidelines for improved flowfield designs as well as choice of GDL materials.

Particle image velocimetry �PIV� is a technique used to obtaininstantaneous 2D velocity fields. This is in contrast to techniquessuch as laser Doppler anemometry and hot wire anemometry,which have high temporal resolution but are pointwise and thushave low spatial resolution. The operating principle of PIV is asfollows �Fig. 1�. A fluid is seeded with passive tracer particles thatare small enough to closely follow the streamlines of the flow.Then an illuminating source �typically, a laser sheet� emits twopulses of light that cast two images of the particles onto a record-ing medium �typically, a CCD camera�. Because of the time delaybetween pulses, particles in the second image are displacedslightly from the first. A process known as interrogation is used toquantify the displacement field between the two images. To ac-complish this, the two images are divided into many subregionsknown as interrogation spots which each contain small clusters ofparticles. Software is used to compute a discrete spatial crosscor-relation between particles within the interrogation region; thepoint of maximum crosscorrelation determines the approximate xand y displacements of the particles �10�. Using subpixel interpo-lation, the displacement can be measured to within 0.1–0.2 pixels�11�. Knowledge of the time delay between illuminating pulsesallows a velocity vector to be obtained in each interrogation re-gion, allowing one to construct the entire 2D velocity field.

The most recent attempt to apply PIV to fuel cell operation wasan investigation of centrifugally driven secondary flow at a U-turnsimilar to flow that would be found within the flow field in a fuelcell �12�. A dimensionally scaled model was constructed fromacrylic in which two channels were connected by a square turn.PIV was able to determine that secondary flows associated withthe U-turn are laminar for Reynolds numbers less than �400;beyond this, turbulence ensues both in the corner and down-stream. It was concluded that the reactant would be well mixed in
the flow channel under such conditions. Although this apparatus
07 by ASME Transactions of the ASME

E license or copyright, see http://www.asme.org/terms/Terms_Use.cfm

wicccs

ersn

E

atpaissTtettmtuapa

fidadaoFcf�boafifcetrma

J

Downloa

as able to model the turbulent nature of the turns of a fuel cell,t did not model other aspects of convection in the fuel cell. Spe-ifically, it could not observe channel convective bypass �flow thatrosses from one channel to another via the porous media� be-ause it did not contain a porous layer to simulate the gas diffu-ion layer.

In this paper, we seek to advance the use of PIV as a tool forvaluating flow channel designs by using it in conjunction withealistic channel designs that include a porous underside. In doingo, it will be possible to directly observe previously elusive phe-omena, such as convective bypass into the gas diffusion layer.

xperimental SetupThe current study employs dimensional analysis in order to

lleviate some of the technical difficulties that would be encoun-ered while observing operational fuel cells. In the current ap-roach, a primary assumption is that the convective flow of air infuel cell can be modeled by flow of a nonreacting, incompress-

ble fluid, such as water, as long as the geometric and dynamicimilarities are maintained. The fact that air is 79% nitrogen givesome justification for this, but it is ultimately an approximation.hen, since no attempt is made to duplicate the reaction physics,

he special materials needed for fuel cell operation become unnec-ssary. Thus, the bipolar plate of the model can be made fromransparent materials that allow the use of the optical imagingechniques used in this study. Likewise, the proton exchange

embrane can be modeled as an impermeable boundary. In ordero simplify the optical requirements of the experiments, it wasseful to geometrically scale up the model by a factor of 10. Then,n important consideration becomes the nature of the modelorous medium, whose permeability should also be scaledppropriately.

An experimental apparatus was constructed to measure velocityelds representative of those in a fuel cell. The flow field wasesigned to be dynamically similar to a 15 cm2 fuel cell operatingt an equivalent current density of 4 A/cm2 �equivalent currentensity is defined as the product of the measured current densitynd the stoichiometry�. The flow field is reconfigurable, capable ofperating as a parallel, serpentine, or interdigitated network �seeig. 2�. The flow channels were constructed by bonding opticallylear acrylic pieces together to form five parallel channels. Theour internal ribs have dimensions of 360 mm�9.5 mm

9.5 mm. The flow channels are 9.0 mm wide and 9.5 mm tall. Auffer of 10 mm was left between the end of each rib and theuter viewing window. The basic channel configuration representsparallel flow field. To create serpentine and interdigitated flow

elds from the parallel configuration, rubber plugs were cut andriction fitted into appropriate regions of the acrylic base as indi-ated in Fig. 2. An aluminum spacer wrapped in polytetraflouro-thylene �PTFE� tape was used to provide the proper thickness ofhe porous medium as well as to seal the model from the sur-oundings. A solid aluminum plate was placed under the porousedium to act as the “impermeable” membrane. Finally, a top

Fig. 1 Basic requirements for a PIV system †10‡

luminum plate with cutouts for optical access was fastened to the

ournal of Fuel Cell Science and Technology


bottom plate �see Fig. 3�. Ports were drilled into the acrylic toserve as the inlet and outlet for fluid flow as shown in Fig. 2. Apump was used to circulate water from a reservoir into the appa-ratus. The water flow rate was controlled using an acrylic flow-meter with metering valve.

The region of interest was illuminated by twin Nd-YAG lasers�Continuum Surelite II�, operating at 10 Hz, and providing 30 mJper pulse �pulse duration=6 ns� at a wavelength of 532 nm.Spherical and cylindrical lenses were used to form a light sheet ofappropriate thickness. Images were recorded on a CCD camera�LaVision ImagerIntense, 1280�1020 pixels� specialized forsingle-frame–double-pulse PIV. Fluorescent particles�20–40 �m� were used to seed the flow. By using fluorescentparticles in conjunction with a long wave filter, background noisecan be significantly reduced while retaining high particle imageintensity. In a further attempt to reduce background noise andincrease particle correlation, a background image �200 images av-eraged together� was subtracted from each PIV image. Image in-terrogation was performed using in-house software. An interroga-tion spot size of 64�64 pixels with a 50% overlap was usedduring post-processing. PIV was performed on the interdigitatedand serpentine flow field geometries.

Fig. 2 Diagram of reconfigurable flow field. Each channel isdenoted by a number, 1–5. Friction-fitted rubber flow stoppersdetermine the path of the fluid.

Fig. 3 Assembly diagram for PIV model. For illustrative pur-
poses, the flow field is shown upside down.
AUGUST 2007, Vol. 4 / 329


dnndamoptcns3apFdttlps

wifitfifidFpottdntfdli

wwnismi

Ettrffitttw

wm

3

Downloa

In order to describe the positions of the measurements, a coor-inate system was defined �see Fig. 2�. The channels were desig-ated by numbers, number 1 being the closest to the inlet andumber 5 being closest to the outlet. The dimension z is used toescribe the distance along the channel, while dimensions x and yre used to describe the width and height of each channel. Experi-ents in the interdigitated configuration focus on the observation

f channel convective bypass. Cross-sectional images in the xylane were taken for several values of z along channels 2 and 4,he exiting “fingers.” Images of the yz plane were also taken in theenter of all five channels in order to observe the primary flowear the finger edges of channels 1–3. PIV was conducted in theingle serpentine configuration along xy cross sections of channelfor several values of z. Channel 3 was chosen because its flow is

nticipated to contain the flow pattern most similar to the spatiallyeriodic flow encountered in the center of the prototype device.or cross-sectional images in the xy plane, the laser sheet wasirected vertically down into the channel through cutouts in theop aluminum plate and the camera viewed the illuminated regionhrough the model end wall. For axial images in the yz plane, theaser sheet was again directed down through the cutouts in the toplate, but the camera viewed the illuminated region through theidewall of the model.

For both sets of experiments, multiple images �typically, 75�ere acquired and the instantaneous velocity fields obtained by

nterrogation were averaged in order to produce smooth velocityeld plots. Images acquired deep within the channels were par-

icularly noisy because of the large number of particles and glassbers that block the optical path; for these regions, 200 velocityelds were acquired and averaged. The optical arrangement wasetermined as a compromise between several competing factors.or maximum accuracy in PIV, it is usually beneficial to allowarticle displacements of several pixels; however, due to the out-f-plane motion of the particles, large particle displacements inhe image plane can cause particles to exit the light sheet prema-urely. This decreases the quality of the correlations performeduring the interrogation stage. In order to correct for this, it wasecessary to properly coordinate the pulse separation, laser sheethickness, and lens f-number. In order to avoid imaging out-of-ocus particles, the laser sheet thickness must be less than theepth of field of the optical device. The depth of field of a singleens and iris optical system is determined by the f-number accord-ng to

�z = 4�1 + M−1�2f2�

here M is the magnification, f is the lens f-number, and � is theavelength of the scattered light. Thus, the depth of field is sig-ificantly improved by increasing the f-number. However, increas-ng depth of field through changes in the f-number has the unde-irable effect of reducing the amount of captured light and perhapsore importantly reducing the quality of the image through an

ncreased diffraction-limited spot size as given by

ds = 2.44�1 + M�f�

ven at the modestly large f-number of 16, this yields the diffrac-ion limited spot size of 34 �m, which is quite large compared tohe geometric particle size of 20–40 �m. Thus, in order to accu-ately resolve particle displacements of several pixels, an-number between 16 and 25 can be used to achieve high depth ofeld while avoiding undesirable diffractive effects. Pulse separa-

ion and light-sheet thickness can then be set to enforce particleso remain in the light sheet. The current study employs the criteriahat at least 75% of particles should remain in the light sheet,hich gives the final condition for the optical requirements that

�tmax � 0.25�z

V

here �tmax is the maximum pulse separation and V is the pri-
ary velocity in the channel, which can be estimated a priori.
30 / Vol. 4, AUGUST 2007


Then, for various regions of the PIV measurements, the pulseseparation differs depending on the primary velocity in the chan-nel and the desired particle pixel displacement. For the currentexperiments, this was between 0.1 ms and 10 ms with light-sheetthicknesses between 0.5 ms and 4 mm.

Care was taken to properly scale permeability of the porousmedium that lies beneath the model “bipolar plate.” Permeabilitycarries the dimensions of L2. Thus, since the geometric scalingfactor for the model is 10, the permeability must be scaled up bya factor of 100. Experimental data show that in-plane permeabilityfor commercially available GDL materials can take on a widerange of values �5�10−13 m2 to 5�10−11 m2� �13,14�. An isotro-pic fabric consisting of randomly oriented e-glass fibers wasfound to be suitable, having an in-plane permeability of 1�10−9 m2 at a thickness of 3.2 mm �determined by the techniquedescribed in �15��. This would then represent a GDL at the upperend of the permeability range with a thickness of �320 �m. Itshould be noted that it is difficult to match all the anisotropicproperties of the GDL. However, we have made the assumptionhere that the only property of the GDL that greatly influences flowis the in-plane permeability �which is matched between the modeland the “prototype” fuel cells�. Our assumption is justified by therelatively longer in-plane path the flow must travel under thelands, compared to the short through-plane travel under the chan-nels. The fiber structures that produce that permeability may bedifferent, but the permeability has been correctly scaled.

Note that although the intention of the experiments was tosimulate a square cell with a 15 cm2 active area under variousconfigurations, the model has only five “passes” instead of the 19or so that would exist in the actual cell. However, it is geometri-cally similar in all other aspects. The reduced number of passeshas the advantages of allowing easier optical to access to thecenter channel and reducing the assembly burden, but has severalrepercussions related to achieving dynamic similarity. The mostobvious shortcoming of having so few channels is that it causes a“finite width” affect. For a cell with 19 passes, it can be expectedthat the velocity field will eventually achieve a periodically re-peating profile; the finite width of the model, however, may not besufficiently long to allow the fluid to settle to it periodically re-peating state. However, it is expected that by the third pass, flowin our model will closely resemble the periodic flow of the full15 cm2 cell. For dynamic similarity, the Reynolds number basedon hydraulic diameter inside an individual channel should matchbetween the model and the prototype. Because of the inexact geo-metric modeling, experimental parameters were modified appro-priately to achieve dynamic similitude as closely as possible. Fordynamic similarity, the interdigitated configuration requires lessflow than an exact geometric model because the flow is beingdistributed through fewer channels; specifically, because the cur-rent model is 9 cm wide, whereas the exact geometric scalingwould be 38 cm wide, the current model requires 0.23 �9/38�times the flow rate of the exact scaling. Based on this rationaliza-tion, a water flow rate of 3.5 cm3/s was used while conductingexperiments in the interdigitated cell configuration. The serpentineconfiguration does not require such a flow rate correction sinceflow is carried along a single channel regardless of the inexactgeometric similarity. Thus, for experiments conducted in the ser-pentine configuration, a water flow rate of 15 cm3/s was used.Note that interdigitated and serpentine fuel cells operate at con-siderably different Reynolds numbers �350 and 1400, respectively,here� even though they have equivalent cell mass flow rates be-cause of the way flow is distributed throughout the cell. Further-more, the ratio of the interdigitated Reynolds number to the ser-pentine Reynolds number increases with cell area.

Results

Interdigitated Flow Field. Because of the symmetry of the
interdigitated flow field, it was anticipated that velocity fields in
Transactions of the ASME


FmG

J

Downloa

ig. 4 Velocity fields for interdigitated cell configuration. „a…– „f… correspond to the cross section and „g…– „k… represent the axialeasurement. In „g…– „k…, the zi=34 cm and zf=35 cm describe the location of the measurement „see Fig. 2 for coordinate system….
ray zones are regions with invalid data due to optical obstructions.
ournal of Fuel Cell Science and Technology AUGUST 2007, Vol. 4 / 331

ded 23 Aug 2007 to 128.175.116.163. Redistribution subject to ASME license or copyright, see http://www.asme.org/terms/Terms_Use.cfm

3

Downloa

Fig. 4 „Continued….

32 / Vol. 4, AUGUST 2007 Transactions of the ASME


Fo

J

Downloa

ig. 5 Velocity fields in channel 3 of the serpentine cell configuration. Gray zones are regions with invalid data due to opticalbstructions.

ournal of Fuel Cell Science and Technology AUGUST 2007, Vol. 4 / 333


ras=isocb=dtfimnwfintvd

i

3

Downloa

eturn channels 2 and 4 should resemble mirror images of onenother for equal values of z. Figures 4�a�–4�f� show a compari-on of channels 2 and 4 for three different values of z. At z34 cm, a clear infusion of fluid from the porous medium is vis-

ble. Although the vector fields in channels 2 and 4 are fairlyymmetric in nature as expected, they are not exact mirror imagesf one another. Flow entering channel 2 from the bottom rightorner is somewhat smaller than flow entering channel 4 from theottom left corner. This trend is repeated at z=25 cm and z30 cm. One possible explanation for this is that the porous me-ium could have a local variation in thickness or in permeability;his is particularly possible because the randomly oriented glassber had visible regions with fewer fibers than others, whichakes the fiber volume fraction on the smaller scale nonhomoge-

eous. In retrospect, a sparsely woven cloth may have been aiser choice for the porous medium. However, the random glassber mat is very similar in pore structure to carbon fiber paper andonwoven gas diffusion layers; thus, perhaps the unevenness isruly representative of the physics in a fuel cell. In the absence ofelocity field data from full size fuel cells, that is difficult toetermine.

Velocity fields taken in the axial yz plane near the end of the

Fig. 5 „Continued….

nlet “fingers” are shown in Figs. 4�i�–4�k�. A comparison of the

34 / Vol. 4, AUGUST 2007


three return channels �1, 3, 5� shows that the velocity fields are allvery similar. It is interesting that the velocity magnitudes near theedge of the fingers are still quite large compared to the cross-sectional components. This is not entirely unexpected since theReynolds number at the inlet is 350, indicating the importance offluid inertia. The stagnation of the flow at the right wall as it abutsthe channel end �rubber stopper� is clearly evident. Also evident isthat the flow then bends downward and exits via the porous me-dium towards the outlet manifold.

Velocity fields in the yz plane of the return channels are shownin Figs. 4�g� and 4�h�. Since at this point of the cell, the twochannels carry almost all of the return flow, the primary velocity isquite high near the center of the channel ��30 mm/s�. It has anear parabolic profile, such as that expected in a fully developedpipe flow. If we compare velocities in the primary direction withthe secondary velocities, it is seen that the secondary velocitymoves at only �1% the speed of primary flow. Because this tendsto promote out-of-plane loss of pairs in the PIV procedure, thismakes measurement of the secondary velocities quite difficult asmentioned in the previous section.

Single Serpentine PIV. Velocity fields were determined in thecross-sectional xy plane of the center channel �channel 3� in aserpentine arrangement at various distances along the channel �de-noted by the coordinate z�. Figure 5�a� shows the flow just afterthe 180 deg turn. As expected, large secondary flows in the formof three vortices are present corresponding to centrifugally drivenflow around the corner. The secondary vortices quickly diminishas seen in Fig. 5�b� just 5 cm from the turn. Flow appears toemerge from the porous medium at the left corner �x /w�0� of theporous medium and exit at the right corner �x /w�1�; this is aclear view of the channel convective bypass, which is predictedby computational �16� and analytic �17� models. However, theemergence from the left side is somewhat surprising, given thatvery little pressure gradient exists between adjacent channels nearthe corner. One possible explanation could be that the inertia fromthe centrifugally driven vortices drives fluid into the porous layer.For similar reasons, the magnitude of the flow exiting at the rightcorner for large z was expected to be small. Again, this is notobserved; flow exit at the right corner is nonzero and relativelyeven over all z locations as evident from Fig. 6. One aspect of theobservations that is consistent with modeling is that in general,flow emerging from the bottom-left side of the velocity fields inFigs. 5�a�–5�h� appears to become more intense with increasing z.

Fig. 6 Vertical velocity at the y location closest to the porousmedia „y=0… for various values of z in channel 3 of the serpen-tine configuration. Note: The data for z=26 cm is taken alongthe line y /h=0.14 because of fibers that protrude into thechannels.

Figure 6 shows the development of the flow along the z coor-

Transactions of the ASME


dmiiapprsssptubtrtncoo

C

acalwetvcciptpitfl

cclestc

J

Downloa

inate. Although concurrent analytical modeling efforts �17� haveade the simplifying assumption that anti-symmetry should exist

n the velocity field along the z coordinate, Fig. 6 shows that thiss clearly not the case. One possible explanation is that sincenalytic efforts have neglected inertial effects, they may be inca-able of predicting the true profile. However, there is anotherossibility. As stated earlier, the porous medium is composed ofandomly oriented in-plane glass fibers that, when viewed macro-copically, has relatively consistent permeability; however, a vi-ual inspection of the fabric shows that the pores are only slightlymaller than the scale of the channel. On this length scale, theermeability of the porous layer may not be a well-defined quan-ity. Thus, local variations in permeability may be to blame for thenevenness in the measured flow. This explanation would apply tooth the serpentine and the interdigitated experimental results. Inhe future, it may be useful to use woven instead of random fab-ics to model the fuel cells porous layer; this would at least dis-inguish the role of the porous layer in determining the smooth-ess of the channel bypass. However, it is possible that real fuelells behave similarly to the current model, since real fuel cellsften use randomly oriented gas diffusion layers with pore sizesnly slightly smaller than the channel width.

onclusionsIn the current study, dimensional similarity was used to designgeometrically and dynamically scaled fuel cell model. The flow

onditions in the model were selected to match the cathode-sideirflow in an equivalent, albeit nonreacting fuel cell. Despite thisimitation, the model allows several advantages that make it over-helmingly simpler to perform PIV measurements than in an op-

rational fuel cell. The first is that all the walls of the model areransparent, which has made it possible to measure channel con-ective bypass for the first time. Another simplification of theurrent method is that is admits the use of the other fluids �in thisase water� as the working fluid in the fuel cell model. The models easily reconfigurable between parallel, interdigitated, and ser-entine flow fields, although velocity data were obtained only forhe latter two. Velocity vector maps were obtained of both therimary and secondary flow in the channels. The secondary flown the interdigitated case is two orders of magnitude smaller thanhe primary flow; thus, successful measurement of the secondaryow is a significant achievement of this study.The PIV velocity vector maps provide confirmation for our con-

urrent analytical study �17�, which predicts the magnitude ofonvective bypass flow through the GDL. By dimensional simi-arity, the PIV velocity maps quantify the flow that would bexpected in to-scale fuel cells. Our data reveal the presence ofignificant bypass flow through the GDL, indicating that convec-ive transport dominates over molecular diffusion for the chosenonditions. Our data also reveal that reactant is not distributed

ournal of Fuel Cell Science and Technology


evenly in serpentine configurations for the chosen conditions,whereas a more even distribution is seen for the interdigitatedconfiguration.

AcknowledgmentThe present work was financially supported by W.L. Gore &

Associates as well as U.S. Department of Energy Grant No.DEFG0204ER63820. A special thanks to Mike Manlove for aid-ing in model construction.

References�1� Tüber, K., Pócza, D., and Hebling, C., 2003, “Visualization of the Water

Buildup in the Cathode of a Transparent PEM Fuel Cell,” J. Power Sources,125, pp. 403–414.

�2� Yang, X. G., Zhang, F. Y., Lubawy, A. L., and Wang, C. Y., 2004, “Visualiza-tion of Liquid Water Transport in a PEFC,” Electrochem. Solid-State Lett.,7�11�, pp. 408–411.

�3� Argyropoulos, P., Scott, K., and Taama, W. M., 1999, “Carbon Dioxide Evo-lution Patterns in Direct Methanol Fuel Cells,” Electrochim. Acta, 44�20�, pp.3575–3584.

�4� Argyropoulos, P., Scott, K., and Taama, W. M., 1999, “Gas Evolution andPower Performance in Direct Methanol Fuel Cells,” J. Appl. Electrochem.,29�6�, pp. 663–671.

�5� Lu, G. Q., and Wang, C. Y., 2004, “Electrochemical and Flow Characterizationof a Direct Methanol Fuel Cell,” J. Power Sources, 134�1�, pp. 33–40.

�6� Yang, H., and Zhao, T. S., 2005, “Effect of Anode Flow Field Design on thePerformance of Liquid Feed Direct Methanol Fuel Cells,” Electrochim. Acta,50, pp. 3243–3252.

�7� Kramer, D., Lehmann, E., Frei, G., Vontobel, P., Wokaun, A., and Scherer, G.G., 2005, “An On-Line Study of Fuel Cell Behavior by Thermal Neutrons,”Nucl. Instrum. Methods Phys. Res. A, 542, pp. 52–60.

�8� Kramer, D., Zhang, J. B., Shimoi, R., Lehmann, E., Wokaun, A., Shinohara,K., and Scherer, G. G., 2005, “In Situ Diagnostic of Two-Phase Flow Phenom-ena in Polymer Electrolyte Fuel Cells by Neutron Imaging—Part A: Experi-mental, Data Treatment, and Quantification,” Electrochim. Acta, 50�13�, pp.2603–2614.

�9� Pekula, N., Heller, K., Chuang, P. A., Turhan, A., Mench, M. M., Brenizer, J.S., and Unlu, K., 2005, “Study of Water Distribution and Transport in a Poly-mer Electrolyte Fuel Cell Using Neutron Imaging.” Nucl. Instrum. MethodsPhys. Res. A, 542, pp. 134–141.

�10� Prasad, A. K. 2000, “Particle Image Velocimetry,” Curr. Sci., 79�1�, pp. 101–110.

�11� Roesgen, 2003, “Optimal Subpixel Interpolation in Particle Image Velocim-etry,” Exp. Fluids, 35�3�, pp. 252–256.

�12� Martin, J., Oshkai, P., and Djilali, N., 2005, “Flow Structures in a U-ShapedFuel Cell Flow Channel: Quantitative Visualization Using Particle Image Ve-locimetry,” ASME J. Fuel Cell Sci. Technol., 2�1�, pp. 70–80.

�13� Mathias, M., Roth, J., Fleming, J., and Lehner, W., 2003, “Diffusion MediaMaterials and Characterization,” Handbook of Fuel Cells—Fundamentals,Technology and Applications, Wiley, New York, Chap. 26.

�14� Bluemle, M., Gurau, V., Mann, J., Zawodzinski, T., De Castro, E., and Tsou, Y.M., 2004, “Permeability and Wettability Measurements for Gas Diffusion Lay-ers of PEMFCs,” Fuel Cell Seminar: Abstracts.

�15� Feser, J. P., Prasad, A. K., and Advani, S., 2006, “Experimental Characteriza-tion of In-Plane Permeability of Gas Diffusion Layers,” J. Power Sources,162, pp. 1226–1231.

�16� Pharaoh, J. G., 2005, “On the Permeability of Gas Diffusion Media Used inPEM Fuel Cells,” J. Power Sources, 144, pp. 77–82.

�17� Feser, J. P., Prasad, A. K., and Advani, S., 2006, “On the Relative Influence ofConvection in Serpentine Flow Fields of PEM Fuel Cells,” J. Power Sources,161, pp. 404–412.

AUGUST 2007, Vol. 4 / 335


particle image velocimetry measurements in a model proton...

Documents