Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
1025
Ó 1999 Kluwer Academic Publishers. Printed in the Netherlands.
Modelling of mass transfer within the PEM fuel cell active layer:
limitations at the particle level
Key words: active layer, mass transfer, modelling, particle level, PEM fuel cells
Abstract
A microscopic approach is proposed for predicting the behaviour of the active layer from the local mass transfer
equations at the catalytic particle level. The model takes into account diffusion and electrochemical reaction without
ohmic drop limitation and is numerically performed for three geometric descriptions of the active layer using the
®nite element method. Diffusion limitations within the whole active layer are con®rmed, but diffusion and
competition effects at the particle level are also pointed out. As a practical conclusion, these effects at the particle
level, almost negligible for oxygen reduction, are signi®cantly in¯uent for hydrogen oxidation.
Ionic conductivity strongly depends on temperature and Low current 10)10±10)8 60±70 jgo j < 400
water content and its values can vary between 1 to density
High current 10)7±10)6 115±125 jgo j > 400
10 S mÿ1 for a Na®onâ membrane [10] or recast
density
Na®onâ [11]. In the same way, the product of diffus-
1027
Table 2. Variation ranges considered for electrochemical parameters of Use of this model requires knowledge of the active
hydrogen oxidation [13] layer thickness, L, and the roughness factor, c, which are
Exchange Tafel slope, b Total k easily measured and range, respectively, from 10 to
current, io /mV dec)1 overpotential, go /m s)1 50 lm and from 10 to 500 cm2 cmÿ2 . On the other hand,
/A cm)2 /mV the size of ¯ooded agglomerate
2 1 can be roughly
H2 anode 10)3±10)2 30±120 go < 150 10)5±1
estimated about 1 to 5 lm [19].
For an isothermal system under steady state condi-
tions, the mass balance equation leads to a one-dimen-
sional equation:
For the slowest reactions, this factor is close to unity
and all the catalyst particles work in a uniform way. For @ 2 C cl C
D ÿ k 0
6
the fastest reactions, tends to zero and only a small @2 l C0
number of the particles is effectively active.
where the local roughness factor, c1 , de®ned for one
2.2. Classical model of the active layer parallelepiped ¯ooded agglomerate, can be estimated
using the following relation:
Several models [18±22] are available to predict and
Total platinum surface c c
simulate the behaviour of a porous gas-diffusion elec- l
7
trode. Among them, the classical agglomerate model Agglomerate volume L
1 ÿ l 2l
(Figure 1(a)) has been proposed to describe mass
Equation 6 can be reduced to dimensional equation:
transport limitations [22]. It considers gas pores within
the active layer, which surround a set of parallelepiped
agglomerates. d2 C
ÿ UC 0
8
Moreover, the model is based on two other assump- dZ 2
tions. First, the Na®onâ electrolyte and the carbon
which contains the reduced concentrations C C=C0 ,
supporting the electrocatalyst are both intimately mixed
the coordinate Z z=l and the dimensionless parameter
into a single phase distributed in a volume. Secondly,
U [17] de®ned as a DamkoÈhler number (ratio of the
diffusion and migration occur in perpendicular
kinetics rate at the interface to the diffusion ¯ux density
directions. Thus the dissolved gas diffuses through the
referred to the characteristic length l):
Na®onâ and is simultaneously consumed. This model,
in the absence of ohmic drop, is equivalent to the totally cl kl
¯ooded model. U
9
DC0
Z 0; C1 10a
dC
Z 1; 0
10b
dZ
U U u a/d
Fig. 2. (a) Hexagonal 3D network of spherical particles: modi®ed
O2 cathode )4
10 ±10 2 )7
10 ±10 )8
10 ±10 )1
>1±20 agglomerate model; (b) hexagonal 2D network of hemispherical
H2 anode 10)1±107 10)4±104 10)4±102 particles: modi®ed thin ®lm model; (c) two isolated spheres: compe-
tition eect between two particles.
1029
ZZ
3.2. Numerical formulation
ar
ÿDrC dS 0
18
For the modi®ed models, resolution requires a numer- X
ical approach and simulations were performed using
Flux-Expertâ software and the ®nite element method. which leads to the strong integral from using the Ostro-
Concerning the discrete 3D or 2D distribution of the gradski theorem [3]:
catalyst phase, six planes of symmetry can be considered ZZ Z
around a given particle and its neighbours (Figure 2(a) DrarC dS ÿ akC dl
19
and (b)). This hexagonal symmetry is approximated by a X @X
cylindrical symmetry. Then the resolution is performed
using a 2D grid (Figure 3) with axisymmetric equations. Solving this equation allows derivation of the concen-
The ®nal numerical formulation consists in writing tration pro®le in the active layer and, ®nally, the ¯ux by
soft-coupled equations, the Laplace equation (Equation integration.
14) and a one-order kinetic law at the electrolyte±
particle interface (15):
4. Discussion
~ r
C
r
D ~ 0
14
As previously indicated, the mass transport limitations
r k
C=C0
15 have been studied by using the dimensionless parameters
U ; u; a=d for the modi®ed agglomerate and thin ®lm
models. The results predicted by the Flux-Expert soft-
These equations lead to a system of partial differential ware (modi®ed model) were compared to those obtained
equations with the following boundary conditions: with the classical models.
(i) a concentration is imposed on the gas±electrolyte Figures 4 and 5 present the contour plots character-
interface (Dirichlet condition): izing the concentration distribution in the agglomerate
or thin ®lm at the anode (hydrogen oxidation) predicted
C C0
16 either from the modi®ed agglomerate and the thin ®lm
models. As expected, the contour plots have a quasi-
(ii) a null ¯ux density passes through every symmetry linear shape in the whole active layer but the discrete
surface (Neumann condition): distribution of catalyst signi®cantly acts on the concen-
tration distribution close to the catalyst particles. These
/0
17 pro®les demonstrate a planar ¯ux density distribution
toward the active layer changing to a spherical ¯ux
For implementing the ®nite element method, the local density close to the particles. For the relatively slow
Laplace equation (Equation 14) is replaced by the oxygen reduction, a planar concentration pro®le is
integral form (Equation 18), using a projective function obtained at any location within the active layer.
a according to Galerkin's method [3]:
4.1. Limitation at the agglomerate or thin ®lm level
Fig. 4. Equiconcentration curves distribution in the agglomerate for the modi®ed agglomerate model in the case of anodic hydrogen oxidation.
Fig. 7. Eectiveness factor as a function of U for constant u
a=d 5 for the modi®ed and classical thin ®lm model. Modi®ed thin ®lm model;
u 0:01 (d); u 1 (m) and u 100 (j). Classical thin ®lm model: u 0:01 (s); u 1 (n) and u 100 (h).
value which is determined by U . On the other hand, for for the discrete model is lower than that predicted by the
a constant U value, the effectiveness factor sharply homogeneous model.
decreases if u increases beyond unity because the local However, the effect of diffusive mass transport
spherical diffusion effect at the particle level is no longer hindrance at the particle level is always much weaker
negligible. Moreover, the effectiveness factor predicted at the cathode due to the slow oxygen reduction kinetics.
by the classical model is always higher than that In contrast, anodic hydrogen oxidation may be in¯u-
obtained from the discrete model. This demonstrates enced by local spherical diffusion. Recent results
that spherical diffusion at the particle level can induce a obtained for hydrogen oxidation at room temperature
strong limitation at the particle level and therefore con®rm that a slight diffusion effect already occurs at
reduces the ef®ciency. Consequently, diffusion hin- the particle level [23]. This local effect becomes more
drance at the particle level adds to diffusive hindrance signi®cant for an anode working at higher temperature,
in the agglomerate or thin ®lm. for instance 353 K. Moreover, the current densities
Figure 9 shows that the conclusions for the thin ®lm predicted by the classical model always remain larger
model are similar to those obtained with the agglomer- than the modi®ed ones because of the local diffusion
ate model. Once again the effectiveness factor predicted effect which is not negligible.
Fig. 8. Eectiveness factor as a function of u for constant U Fig. 9. Eectiveness factor as a function of u for constant U
a=d 5
a=d 10 for the modi®ed and classical agglomerate models. Mod- for the modi®ed and classical thin ®lm models. Modi®ed thin ®lm
i®ed agglomerate model; U 1 (d); U 10 (m) and U 100 (j). model; U 8 10ÿ3 (d); U 0:8 (m) and U 80 (j). Classical thin
Classical agglomerate model: U 1 (s); U 10 (n) and U 100 (h). ®lm model: U 0:8 (n); and U 80 (h).
1032
Fig. 10. Experimental and simulated current densities of oxygen reduction for a thin nonporous active layer (l=5 lm, c 27 m2 mÿ2 ,
d 4:1 nm). Key: (ÐÐÐ) experimental data; ( d ) classical model; ( - - -j- - -) modi®ed model.
For oxygen reduction at the cathode, no diffusion powders or by comparing two kinds of particles for the
effect appears at the particle level (Figure 10). Thus the same total loading: powders having a low Pt/C ratio
current densities predicted by both the classical and (low values of a=d) and mixtures of pure carbon and
modi®ed models are similar and close to the experimen- powders having a high Pt/C ratio (high values of a=d).
tal data [24]. No effect of the geometrical ratio a=d was found down
On the one hand, the working of the anode should to values lower than 10 for oxygen reduction in contrast
depend on the mean particle diameter. As a consequence, to hydrogen oxidation [23, 24] on Pt nanoparticles.
the local diffusion effects provide an added reasons for
reducing the size of catalyst particles to increase the
surface to volume ratio of the noble metal and to 5. Conclusion
minimise its loading. On the other hand, for the cathode
the optimal size of catalyst particles in mainly deter- The models presented show the in¯uence of parameters
mined by electrode kinetic considerations [6, 7]. characterizing the active layer on the working behaviour
of PEM electrodes. The results provided by the classical
4.3. Competition effect between neighbour particles models concerning the effect of planar diffusion have
been con®rmed. However, numerical modelling also
The competition of neighbour particles can be demon- points to a local effect of spherical diffusion at the
strated by studying how the effectiveness factor varies as particle level which had been neglected by classical
a function of the dimensionless ratio a=d. As a ®rst
approach, the competition effect is analysed by only
considering two isolated sphere surrounded by the
electrolyte. The results are shown in Appendix 1 and
lead to the following conclusions: under kinetic control
u < 10ÿ2 , the competition effect is masked by kinetics
(Figure 11), whereas, under mixed kinetic and diffusion
control, the effectiveness factor decreases for a=d values
smaller than 10.
The modi®ed agglomerate model predicts a decreasing
effectiveness factor when a=d becomes lower than 10
while the modi®ed thin ®lm model demonstrates a rather
unexpected, but explainable, increase in the effectiveness
factor when a=d decreases [25].
As a practical conclusion for H2 =O2 PEMFC elec-
trodes, the competition effect is masked at the cathode
by the slow oxygen reduction kinetics, but it should
appear at the anode for hydrogen oxidation, even if its
Fig. 11. Eectiveness factor for two isolated particles versus a=d under
in¯uence is relatively limited. mixed kinetic and diusion control
u. Two isolated spheres; u 0:01
This effect was experimentally studied in our labora- (d); u 1 (m) and u 100 (j). One isolated sphere: u 100 (). One
tory for well-characterized Pt nanoparticles on graphite isolated equivalent sphere u 100 (+).
1033
models. Local spherical diffusion actually acts in addi- As a consequence, it can be stated that under diffusion
tion to planar diffusion. Moreover, a slight competition control, the effectiveness factor signi®cantly decreases
effect appears under mixed kinetic and diffusion control down to about 70% when the interparticle distance
while it is masked under kinetic control. becomes smaller than 10 times the mean particle
For the H2 =O2 PEM electrodes, the parametric study diameter.
demonstrates that for oxygen reduction at the cathode
only planar diffusion prevails, while hydrogen oxida-
tion is affected by diffusion limitations both at the References
layer and the particle scales. In practice the modi®ed
1. M. Watanabe, H. Sei and P. Stonehart, J. Electroanal. Chem. 261
models were used to investigate the kinetic parameters,
(1981) 375.
corrected for diffusion both at the electrolyte and 2. Y. Bultel, P. Ozil, R. Durand and D. Simonsson, in Proceedings of
particle scales [23]. the ®rst international symposium on `Proton Conducting Mem-
However, the major inconvenience of this approach brane Fuel Cells', edited by S. Gottesfeld, G. Halpert and A.
is to require complete numerical computations for Landgrebe. The Electrochemical Society Proceedings series, PV 95-
23, Pennington, NJ (1995), p. 34.
improving the modelling of PEM electrodes by taking
3. Y. Bultel, R. Durand and P. Ozil, J. Appl. Electrochem. 28 (1998) 269.
into account the local spherical diffusion at the particle 4. K. Kinoshita, `Electrochemical Oxygen Technology' (J. Willey &
level. Therefore, we developed two modi®ed thin ®lm Sons, New York, 1992), p. 275.
and agglomerate models which have been successfully 5. S. Gottesfeld and T.A. Zawodzinski, Polymer electrolyte fuel cells,
shown to be equivalent to the numerical models in Advanced Electrochemistry Science Engineering', Vol. 5, edited
by R.C. Alkire, H. Gerisher, D.M. Kolb and C.W. Tobias (Wiley/
presented here. These models are presented elsewhere
VCH, 1997), p. 197.
[25, 26]. 6. M. Watanabe, M. Makita, H. Usami and S. Motoo, J. Electroanal.
Chem. 195 (1986) 195.
7. M. Uchida, Y. Aoyama, N. Eda and A. Ohta, J. Electrochem. Soc.
Appendix 1: Two isolated spheres ¯ooded in electrolyte 142 (1995) 4143.
8. A. Kabbabi, F. Gloaguen, F. Andolfatto and R. Durand,
J. Electroanal. Chem. 375 (1994) 251.
The competition effect is studied by considering two 9. A. Gamez, D. Richard, P. Gallezot, F. Gloaguen, R. Faure and R.
isolated spherical particles. It obviously depends on the Durand, Electrochim. Acta. 41 (1996) 307.
two dimensionless parameters a=d and u, as shown by 10. A. Parthasarathhy, S. Srinivasan, A.J. Appleby and C.R. Martin,
Figure 11, which describes the variations of the effec- J. Electrochem. Soc. 139 (1992) 2530.
11. S. Gottesfeld, J. Electrochem. Soc. 139 (1992) 2980.
tiveness factor.
12. S. Gottesfeld, I.D. Rastrick and S. Srinivasan, J. Electrochem. Soc.
Under kinetic control
u < 10ÿ2 , the effectiveness 134 (1987) 1455.
factor does not depend on the dimensionless ratio a=d 13. M. Watanabe, H. Igarashi and K. Yosioka, Electrochim. Acta 40
and is close to unity, and the competition effect is (1995) 329.
masked when the limiting step is the interfacial kinetics. 14. P. BjoÈrnbom, Electrochim. Acta 32 (1987) 115.
15. E-TEK, Inc., Catalog., (1995)
On the other hand, under mixed kinetic and diffusion
16. K. Broka and P. Ekdunge, J. Appl. Electrochem. 27 (1997) 281.
control
u > 10ÿ2 , the effectiveness factor decreases 17. P. Stonehart and P. Ross, Electrochim. Acta 21 (1976) 441.
from a constant value when the ratio a=d becomes 18. D.M. Bernardi and M.W. Verbrugge, J. Electrochem. Soc. 139
smaller than 10. Moreover, the closer to diffusion (1992) 2477.
control the more important the competition effects 19. S. Srinivasan, H.D. Hurwitz and J.O'M Bockris, J. Chem. Phys. 46
(1967) 3108.
are. Generally, two limiting behaviours can be pointed
20. S. Srinivasan and H.D. Hurwitz, Electrochim. Acta 12 (1967) 495.
out from numerical computations. First, when a=d is 21. J. Giner and C. Hunter, J. Electrochim. Soc. 116 (1969) 1124.
greater than 10, each particle acts as if isolated. 22. F. Gloaguen and R. Durand, J. Appl. Electrochem. 27 (1997) 1029.
Secondly, when a=d tends to unity, two particles acts 23. O. Antoine, Y. Bultel, R. Durand and P. Ozil, Electrochim. Acta 24
as a single one, characterized by an equivalent diameter (1998) 3681.
24. O. Antoine, PhD thesis, INPG (1998) Grenoble.
deq and developing the same active area:
25. Y. Bultel, P. Ozil and R. Durand, Electrochim. Acta 43 (1998) 1077.
ÿ ÿ 2 p 26. Y. Bultel, P. Ozil and R. Durand, AIDIC Conference Series, Vol.
2 pd 2 p deq $ deq d 2
21 22 (1997), p. 403.