Electrochemistry Communications 9 (2007) 23402345 www.elsevier.

com/locate/elecom

A microuidic approach for measuring capillary pressure in PEMFC gas diusion layers
Joseph D. Fairweather a, Perry Cheung a, Jean St-Pierre b, Daniel T. Schwartz
a b

a,*

University of Washington, Seattle, WA 98195-1750, USA University of South Carolina, Columbia, SC 29208, USA Received 27 June 2007; accepted 29 June 2007 Available online 12 July 2007

Abstract A dearth of experimental capillary pressure data limits our understanding and optimization of liquid water transport in PEMFC gas diusion layers (GDLs). A microuidic device and method is described for measuring the capillary pressure as a function of liquid water saturation for these thin porous materials with complex, heterogeneous wetting properties. A sample sandwich (hydrophilic membrane GDLhydrophobic membrane) is key for probing the entire hydrophilic and hydrophobic pore volume of the GDL during sequential liquid intrusion and gas intrusion experiments. The capillary pressure curves for an as-purchased Toray 090 and two dierentially-processed Avcarb P75T GDLs were evaluated; each material displayed highly repeatable, but quantitatively dierent, room temperature capillary pressure curves that matched qualitative dierences in their macroscopic wettability. The measurements show that hysteresis between the liquid intrusion and gas intrusion curves is signicant. For example, both the Toray and fully wet-proofed Avcarb GDLs appear hydrophobic during most of the liquid intrusion curve and hydrophilic during most of the gas intrusion curve. The implications of this work for water management, and future device designs and experiments are described. 2007 Elsevier B.V. All rights reserved.
Keywords: Capillary pressure; Multiphase; Fuel cell; Gas diusion layer; Hysteresis

1. Introduction Gas diusion layers (GDLs) are engineered porous media with a critical role in proton exchange membrane fuel cells (PEMFCs). The GDL is responsible for supporting the reaction sites in the thin catalyst layer, distributing reactant gases to the reaction sites, conducting the electrical current produced, and removing the product water. The removal of liquid water is critical, because liquid can block reactant transport and limit the performance of a PEMFC [1]. The most common porous GDL materials are 200500 lm thick and composed of carbon paper sheets or woven carbon bers [2]. In order to increase the wicking of water away from the reaction sites, Teon is incorporated into the carbon substrate to make the sur*

faces hydrophobic [2]. A comprehensive review of the existing research in this eld is provided by Wang [3]. Enhancing the performance of PEMFCs through improved liquid transport is an ongoing subject of research [46]. Electrochemical measurements of cells and stacks are used to indicate when liquid water leads to performance degradation [58], but interest has been increasing for a more rened understanding of liquid water transport and the role of GDL manufacturing on performance [914]. One of the critical constitutive relationships for describing capillary ow in a porous material is the capillary pressure vs. saturation (Pc vs. Sl). For this work, we dene the capillary pressure as P c P l P g; 1

Corresponding author. Tel.: +1 206 543 8388; fax: +1 206 685 3451. E-mail address: dts@u.washington.edu (D.T. Schwartz).

where Pl refers to the liquid water phase pressure and Pg refers to the gas phase pressure. Capillary pressure is a thermodynamic quantity governed by the liquidgas

1388-2481/$ - see front matter 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.elecom.2007.06.042

J.D. Fairweather et al. / Electrochemistry Communications 9 (2007) 23402345

2341

interface and the properties of the conning porous medium (geometry and surface energetics), which can be described by the modied YoungLaplace equation [15] Pc r cos h Reff 2

ASTM porous plate apparatus [19], but it has been adapted for both liquid intrusion and gas intrusion experiments carried out over the full range of saturation. 2. Experimental methods 2.1. Materials 2.1.1. GDL samples Three commercially available GDL materials were investigated. The rst was an as-received Toray TGP-090 sample (Toray CFA, Flower Mound, TX). The second was an unsintered Avcarb P75T sample (Ballard Materials Products, Lowell, MA) that has been inltrated with hydrophilic surfactant dispersion containing Teon. The third was the treated Avcarb P75T after sintering at 400 C for 15 min to remove the surfactant. Table 1 lists the physical properties of each sample. The unsintered Avcarb is listed as hydrophilic because a water droplet placed onto the unsintered Avcarb was spontaneously imbibed into the GDL, and the others are listed as hydrophobic because they repelled the droplet. 2.1.2. Fluidic system The microuidic device, photographed in Fig. 1A, is designed to measure the pressure and control the saturation in the thin GDL samples we use. The body and transducer have a small total volume (<250 lL) that is comparable to the pore volume of the GDL samples (Table 1). The syringe has continuous hydraulic contact to the sample through the rigid body; this continuous contact allows the syringe pump to directly control the sample saturation level. The exploded schematic of the microuidic device is shown in Fig. 1B. The acrylic ow cell (layers bf) directs liquid water from the syringe (layer a) to the GDL sample housing (h) and pressure transducer (g). The acrylic ow cell is built using a laminar construction technique alternating between layers of acrylic (grey layers) and 150 lm double-sided Macbond adhesive (white layers, West Coast Paper Company, Kent, WA). All the acrylic and adhesive layers are precisely cut using an M-300 CO2 laser platform (Universal Laser Systems, Scottsdale, AZ) and are press-t together using a jig. The pressure dierence is recorded from the transducer (Omega Engineering, Stanford, CT) through custom Labview 8 control software (100 Hz sampling rate), that also
Table 1 Physical properties for the 24 mm diameter GDL samples Toray TGP 090 Thickness (lm) Porosity, e Vpore (lL) Wettability 280 0.78 99 Hydrophobic Unsintered Avcarb P75T 240 0.83 91 Hydrophilic Sintered Avcarb P75T 240 0.83 91 Hydrophobic

where r is the surface tension between the liquid waterair interface, h is the contact angle of the liquidair interface on the porous surface (taken with respect to the liquid phase), and Re is the eective pore radius. The wettability is indicated by the contact angle on a smooth surface: a hydrophilic surface will have h < 90, and a hydrophobic surface will have h > 90. GDL materials typically have mixed wettability from the combination of hydrophobic Teon and mildly hydrophilic carbon bers [16]. Thoroughly describing the Pc vs. Sl relationship of a GDL requires probing both the hydrophilic and hydrophobic pore spaces (negative and positive capillary pressures, respectively). For a complex porous medium like a GDL, h and Re are both functions of the liquid saturation Sl (the fraction of total pore volume occupied by liquid). Moreover, h depends on whether liquid is being imbibed into or drained from the media [15]. This makes capillary pressure a member of the interesting class of hysteretic thermodynamic properties [15], and it means both liquid intrusion and gas intrusion curves must be probed to fully characterize the Pc vs. Sl relationship. Both h and Re are determined by the GDL pores where the liquid/gas interfaces reside at a given saturation, so the Pc vs. Sl relationship provides a ngerprint for a given GDL material. The relationship allows a window into how GDL manufacturing and fuel cell operation aects the interaction of the material with multiphase water. Experimental data for GDLs was rst obtained by Gostick et al. [9], who measured the Pc vs. Sl relationship for an octaneair and waterair system using the method of standard porosimetry (MSP) [17]. They measured Pc vs. Sl for a portion of the pore volume during gas intrusion, but MSP cannot be used to determine liquid intrusion curves or probe pore volume that does not spontaneously wet. Koido et al. [10] and Nguyen et al. [11] applied a pressure dierence to force saturation changes in GDL samples. Nguyen has measured Pc vs. Sl curves for both imbibing and draining, while Koido has only examined imbibing. Both used methods that are limited by the natural breakthrough pressures of the GDL materials, so they only probed a relatively small pressure range. An alternative approach has been to estimate the Pc vs. Sl relationship from a materials wetting and structure, either using a simple model geometry [12] or through pore network simulations [13,14]. Here we discuss the construction and use of a microuidic device that is capable of measuring the Pc vs. Sl relationship for thin porous materials such as GDL materials. An earlier version of this device has been described previously [18]. The design is inspired by the

2342

J.D. Fairweather et al. / Electrochemistry Communications 9 (2007) 23402345

A 100 lm thick hydrophilic membrane (layer 5; nylon, 1.2 lm characteristic pore size; GE Osmonics, Minnetonka, MN) on the liquid side of the GDL provides intimate liquid contact between the GDL sample and the liquid water, and inhibits air from entering the body during gas intrusion by increasing the bubbling pressure. A 100 lm hydrophobic membrane (layer 3; treated nylon, 0.45 lm pore size; GE Osmonics) on the gas side similarly inhibits water from breaking through the top of the sample during liquid intrusion by increasing the breakthrough pressure. The use of paired membranes to access pores with dierent wettabilities has previously been reported [20]. The hydrophilic membrane is 100% saturated with liquid and the hydrophobic membrane is 0% saturated for the pressure range used in these experiments, so the membranes will not aect the measured Pc vs. Sl curve. The acrylic plates on the liquid and gas side of the GDL support the sample and are vented to allow phase contact and redistribution. Because of airs low viscosity and the small ow rates used, it is assumed that the gas gauge pressure in the GDL is always zero, so at equilibrium the gauge pressure of the liquid Pl equals Pc, according to Eq. (1). During construction, the outer perimeter of the membranes and GDL sample are sealed with epoxy. The housing then is assembled and compressed at approximately 70 psi to permanently seal the adhesive layers. The compression of the GDL was not varied for this study, but this is an obvious parameter to investigate in the future to see the eect on the Pc vs. Sl relationship [21]. 2.2. Operation Before testing each GDL sample, the acrylic ow cell is completely lled with liquid water, to eliminate trapped air and ensure continuous hydraulic contact between the syringe and the GDL sample. The ow cell and GDL housing are wetted separately, then bolted together and sealed by the neoprene gasket (layer i in Fig. 1B). This leads to an uncertainty in the initial saturation, which is corrected as explained below in Calculating Pc vs. Sl. After cell assembly, the pressure response of the GDL is tested as the pump moves liquid water into and out of the sample. Liquid water is repeatedly injected into and drained from the GDL sample at a constant ow rate of 5 lL/min, while the liquid gauge pressure is maintained between 30,000 Pa. The range is chosen to lie well within the bubbling and breakthrough pressure of the membranes. This cycling procedure is done to evenly wet the sample and to establish the total volume change between the positive and negative pressure limits. Once the pressure response is repeatable for constant ow liquid intrusion/gas intrusion cycles (usually after 2 3 cycles), a stepwise liquid intrusion/gas intrusion experiment is performed. The concept for each step is sketched in Fig. 2. Each step takes place in four stages: in part A, equilibrium is initially present in the liquid pressure at a given syr-

Fig. 1. A photograph (A) and exploded schematic (B) of the microuidic device. The body (layers bf) and sample housing (layer h) were built using laminar construction with acrylic layers (grey) bonded by adhesive layers (white). Individual layers (ai, 17) are discussed in the text.

controls the syringe pump. A NE-500 syringe pump (New Era Pump Systems, Farmingdale, NY) with a 250 lL gastight syringe (Hamilton Company, Reno, NV) controls uid displacement. De-ionized water at room temperature ($25 C) is used after being de-gassed at 85 C for 30 min. 2.1.3. Sample housing Each GDL sample is permanently encased in a separate acrylic housing with its components numbered in Fig. 1B. The GDL sample (4) is sandwiched between two membranes (3 and 5) and the resulting GDL sandwich is sealed between two adhesive layers (2 and 6) and acrylic plates (1 and 7) to complete the acrylic sample housing.

J.D. Fairweather et al. / Electrochemistry Communications 9 (2007) 23402345

2343

2.3. Calculating Pc vs. Sl Because the initial saturation of the GDL is unknown, correlating Vsyr with Sl requires comparison to a welldened reference state. The reference state chosen is the irreducible liquid volume, Vir, of the GDL sandwich. In porous media, the capillary pressure asymptotes toward negative innity when it approaches the irreducible liquid volume of the sample [15]. Thus, by using the irreducible volume as a reference, any syringe volume (Vsyr,i) can be adjusted to a total volume (Vsyr,i + Vir). To determine Vir, the average initial dry weight (Mdry) is rst measured for a set of GDL material samples. The samples are then wetted with water under vacuum. To drain the samples down to their irreducible volume, several layers of highly hydrophilic tissue paper (VWR International, West Chester, PA) are rmly pressed to the samples for 15 min in a humidied air chamber (to minimize evaporation). The samples are again weighed (Mir) and the irreducible volume is determined through
Fig. 2. The four stages (AD) for each stepwise liquid/gas intrusion experiment, with the top showing the volume imbibed from the syringe and the bottom showing the corresponding local dierential liquid pressure measured by the pressure sensor.

V ir

M ir M dry qwater

inge volume. At this stage a Pc,i vs. Vsyr,i point is known. During part B, liquid water is injected into the system at a constant rate for 30 s while pressure is measured. The ow is stopped in Part C and the local liquid pressure relaxes to the equilibrium value at (Pc,i+1,Vsyr,i+1). The process is repeated to construct the Pc vs. Vsyr curve. After 20 steps of liquid intrusion, the ow direction is reversed, and gas is intruded (while liquid is drained) for 20 steps. Each stepwise liquid intrusion/gas intrusion cycle took place over 100 min. The initial liquid pressure is always set to about 30,000 Pa, while the ow rate is chosen such that the highest pressure is +30,000 Pa after 20 liquid intrusion steps. The capillary number can be used to predict phase displacement behavior, and ensure that the experiment is run in the same regime as an operating fuel cell. The capillary number is a ratio of the viscous forces to capillary forces, dened as [22] Ca huii li r

where qwater is the density of water. This measurement has indicated that the irreducible volume is less than 1 lL/cm2 for all samples (irreducible saturation is less than 5%), so we assume that the saturation at our lowest liquid pressure is eectively zero. Our liquid saturation (Sl) is then simply Sl V syr ; V pore 4

3. Results and discussion Fig. 3 shows a typical stepwise liquid intrusion/gas intrusion experiment for a sintered Avcarb sample. Fig. 3a illustrates how the controlled variable, Vsyr, is changed during an experiment. The volume of water imbibed into the GDL sample is stepped up 20 times until the target maximum pressure ($30,000 Pa) is achieved, and afterwards the water is drained in a similar stepwise fashion. The corresponding pressure response due to the applied volume change is shown in Fig. 3b. Negative liquid pressure corresponds to the liquid waterair interface present in hydrophilic-like (wettable) pores whereas positive dierential liquid pressure corresponds to hydrophobic-like (non-wetting) pores. The hysteresis between the liquid intrusion and gas intrusion branches of the liquid pressure curve is dramatic, as shown in Fig. 3b. During liquid intrusion, positive liquid pressure must be applied to force liquid into the sintered Avcarb GDL. However, to drain the same pore volume, negative liquid pressure must be applied. While the GDL appears hydrophobic (positive capillary pressure, implying airwetting) when liquid water is being moved into the material, it acts as a hydrophilic material (negative capillary pressure, water-wetting) when water is being removed.

where hui i is the supercial velocity of the invading phase, li is the viscosity of the invading phase, and r is the surface tension between the two phases. The ow rates are varied between 1.3 and 2.1 lL/cm2/min, which for water invading at 25 C yields a capillary number range of 2.64.3 106. This corresponds roughly to the capillary number of a fuel cell operating with a current of 1 A/cm2 at 80 C (around 5 106). For this capillary number range, capillary forces are important, and signicant ngering of the intruding phase is expected, as opposed to a smooth displacing front [23].

2344

J.D. Fairweather et al. / Electrochemistry Communications 9 (2007) 23402345

Fig. 3. Typical experimental data for sintered Avcarb. The syringe volume (a) controls the sequence of stepwise liquid/gas intrusion as the liquid pressure response (b) is measured. Insets show some expanded steps.

All measurements are taken after reproducible cyclic behavior is observed, so the response can be replicated repeatedly. Hysteresis is common in the porous materials literature, and is largely due to ink-bottle eects and hysteresis in the contact angle h [15,24,25]. Pc vs. Sl data are calculated from dynamic data like that in Fig. 3, using the methods described in Section 2.3. Fig. 4ac shows a set of liquid intrusion and gas intrusion Pc vs. Sl curves for the Toray 090, treated Avcarb P75T, and treated and sintered Avcarb P75T GDL samples, respectively. Each Pc vs. Sl curve in Fig. 4 corresponds to three replicate experiments, showing good reproducibility of the hysteretic eects between liquid intrusion and gas intrusion. The Pc vs. Sl curves obtained for each sample matches well with qualitative observations; the GDLs that repel water (Fig. 4a and c) have largely positive liquid intrusion curves, whereas the sample that spontaneously imbibes water has a largely negative liquid intrusion curve. The carbon bers used in the GDL support are of intermediate wettability, with an advancing contact angle of around 90 [26], which means samples can easily appear hydrophobic during liquid intrusion and hydrophilic during gas intrusion. Hydrophobic agents such as Teon can only do so much to mitigate this eect, since the receding contact angle of these agents is also often close to 90 [24,27]. Thus, the hydrophilic character during gas intrusion is not surprising even for GDLs with hydrophobic liquid intrusion curves.

Fig. 4. A set of liquid intrusion (lled points) and gas intrusion (outline) Pc vs. Sl curves for the Toray 090 (triangles), treated Avcarb P75T (circles), and treated and sintered Avcarb P75T (squares) sample. Solid curves have been added to guide the eye.

Quantitatively, the primary change in saturation occurs over 104 < Pc < + 104 Pa, which is similar to the range observed by Gostick and Koido [9,10]. However, we note that Nguyen has reported that the primary change in Sl occurs over a much narrower range, 200 < Pc < + 200 Pa [11]. We are examining the eect of relaxation time on these measurements, since Fig. 3b clearly shows the transient can extend longer than the 120 s we allowed after each step.

J.D. Fairweather et al. / Electrochemistry Communications 9 (2007) 23402345

2345

All the curves in Fig. 4 display nearly asymptotic behavior as the irreducible liquid saturation and total saturation are approached. As discussed above in Section 2.3, separate gravimetric measurements indicate that the irreducible liquid saturation is less than 5%; that is, the liquid phase remains continuous even at very low saturation, and almost all of the liquid can be removed through ow to the surface. It is not surprising that the liquid phase has continuous access to the surface in these very thin samples. 4. Conclusions and implications Careful control of GDL design and manufacturing could lead to marked improvements in liquid water transport. We have shown that a microuidic approach is capable of measuring the liquid intrusion and gas intrusion Pc vs. Sl curves for common GDL materials with dierences in each material clearly displayed. It was also shown that it is possible to probe both hydrophilic-like and hydrophobic-like pore volumes as well as approach the irreducible water and irreducible air phase saturations. Because of the devices ability to measure the pressure response during both liquid intrusion and gas intrusion, valuable insight can be gained into the eect of hysteresis on the apparent wettability of the GDL materials. The membranes used to sandwich the GDL samples allow a wider range of saturations to be probed than was previously possible in GDL experiments. As changes to GDL design and manufacturing are tested, the use of a microuidic tool like this can provide Pc vs. Sl relationships that ngerprint the materials fundamental capability to transport liquid water. We are working on improved device designs, longer operating windows (to allow complete equilibration), different operating parameters (temperature, GDL compression pressure, etc.), use of the dynamic pressure data to understand transport properties as well as capillary thermodynamics of GDL materials, and providing an appropriate empirical Pc vs. Sl relationship. Acknowledgments This study was funded by Ballard Power Systems, the Boeing-Sutter endowment, and the University of Washing-

ton Fuel Cell Infrastructure Equipment Grant. We thank Colleen Legzdins of Ballard Power for helpful conversations and feedback. References
[1] W. Vielstich, A. Lamm, H.A. Gasteiger, Handbook of Fuel Fells: Fundamentals Technology and Applications, Wiley, Chichester, England, New York, 2003. [2] A.L. Dicks, J. Power Sources 156 (2006) 128. [3] C.Y. Wang, Chem. Rev. 104 (2004) 4727. [4] Z.G. Zhan, J.S. Xiao, D.Y. Li, M. Pan, R.Z. Yuan, J. Power Sources 160 (2006) 1041. [5] J.M. Song, S.Y. Cha, W.M. Lee, J. Power Sources 94 (2001) 78. [6] C. Lim, C.Y. Wang, Electrochim. Acta 49 (2004) 4149. [7] M.G. Santarelli, M.F. Torchio, Energy Convers. Manage. 48 (2007) 40. [8] J. Ihonen, M. Mikkola, G. Lindbergh, J. Electrochem. Soc. 151 (2004) A1152. [9] J.T. Gostick, M.W. Fowler, M.A. Ioannidis, M.D. Pritzker, Y.M. Volfkovich, A. Sakars, J. Power Sources 156 (2006) 375. [10] T. Koido, T. Furusawa, K. Moriyama, K. Takato, ECS Trans. 3 (2006) 425. [11] T.V. Nguyen, G. Lin, H. Ohn, D. Hussey, D. Jacobson, M. Arif, ECS Trans. 3 (2006) 415. [12] A.Z. Weber, R.M. Darling, J. Newman, J. Electrochem. Soc. 151 (2004) A1715. [13] C.-Y. Wang, P. Sinha, ECS Trans. 3 (2006) 387. [14] V.P. Schulz, J. Becker, A. Wiegmann, P.P. Mukherjee, C.-Y. Wang, J. Electrochem. Soc. 154 (2007) B419. [15] J. Bear, Dynamics of Fluids in Porous Media, American Elsevier Pub. Co., New York, 1972. [16] D. Bevers, R. Rogers, M. vonBradke, J. Power Sources 63 (1996) 193. [17] Y.M. Volfkovich, V.S. Bagotzky, V.E. Sosenkin, I.A. Blinov, Colloids Surface A. 187188 (2001) 349. [18] J. Fairweather, P. Cheung, D.T. Schwartz, J. St-Pierre, ECS Trans. 3 (2006) 981. [19] ASTM D 2325-68 (Reapproved 2000). [20] A. Salehzadeh, A.H. Demond, J. Environ. Eng. 125 (1999) 385. [21] A. Bazylak, D. Sinton, Z.S. Liu, N. Djilali, J. Power Sources 163 (2007) 784. [22] M. Kaviany, Principles of Heat Transfer in Porous Media, second ed., Springer, Verlag, New York, 1995. [23] R. Lenormand, J. Phys.: Condens. Mat. 2 (1990) SA79. [24] N.R. Morrow, J. Can. Petrol. Technol. 14 (1975) 42. [25] W.G. Anderson, J. Petrol. Technol. 39 (1987) 1283. [26] N.I. Kim, H.M. Kang, Y.T. Hong, T.H. Yoon, J. Adhes. Sci. Technol. 16 (2002) 1825. [27] C.W. Extrand, Y. Kumagai, J. Colloid Interf. Sci. 191 (1997) 378.