You are on page 1of 13

Journal of Membrane Science 362 (2010) 221233

Contents lists available at ScienceDirect

Journal of Membrane Science


journal homepage: www.elsevier.com/locate/memsci

Hydrogen permeation in palladium-based membranes in the presence of carbon monoxide


Jacopo Catalano, Marco Giacinti Baschetti , Giulio C. Sarti
Dipartimento di Ingegneria Chimica, Mineraria e delle Tecnologie Ambientali (DICMA), Alma Mater Studiorum, Universit di Bologna, via Terracini 28, 40131 Bologna, Italy

a r t i c l e

i n f o

a b s t r a c t
A theoretical model is proposed to describe hydrogen permeation in palladium and silverpalladium membranes in presence of a non-inert gas as CO; it is known indeed that hydrogen ux through palladiumbased membranes drastically decreases when H2 is fed in mixtures containing carbon monoxide due to the interaction of the latter gas with the membrane surface. To model this process, the adsorption step of the well-known approach suggested by Ward and Dao has been suitably modied, since it must be considered as a competitive adsorption of the different non-inert molecules on the metal interface. In particular, the competitive adsorption of CO and H2 has been examined accounting for the spectrum of information available for CO adsorption on palladium, as well as for hydrogen in palladium and palladiumsilver alloys. A validation of the model proposed has been performed through a comparison between several literature data and model calculations, over a rather broad range of operating conditions. A quite good agreement was obtained in the different cases; the model, thus, can be protably used for predictive purposes. 2010 Elsevier B.V. All rights reserved.

Article history: Received 21 April 2010 Received in revised form 10 June 2010 Accepted 26 June 2010 Available online 31 July 2010 Keywords: Hydrogen permeation Palladium Palladiumsilver membrane Carbon monoxide adsorption

1. Introduction Grahams discovery of hydrogen adsorption in palladium around 1866 [1], made the PdH2 system the rst metalhydrogen system studied and started a protable research activity that has led to extensive investigation, recently enhanced by the increased interest in hydrogen as an energy carrier [24]. The high solubility and mobility of hydrogen in the Pd lattice makes palladium an ideal material for separating hydrogen in a highly pure form [2,5,6], which is needed for the direct use of this gas in polymer electrolyte fuel cells (PEM-FCs), the most promising technology to obtain clean energy from hydrogen [7]. At present, hydrogen is largely produced by steam reforming of methane [5,8,9] and, as a consequence, palladium-based membranes operating in such separation processes are exposed to complex gases mixtures comprising H2 , CO, CO2 , H2 O, CH4 , among other species. Many experimental works have studied the behaviour of these membranes and several of them reported a dramatic reduction of the hydrogen permeate ux when carbon monoxide was added to the feed stream [1017]; in that case the hydrogen permeate depletion was attributed mainly to the reduction of the available dissociation sites on the metal surface caused by CO adsorption which competes

with that of hydrogen. Despite the many experimental evidences and the amount of studies presented, the inuence of this gas on hydrogen permeation has not yet been modelled rigorously and effectively, and only very general equations have been proposed in order to account for the effect of CO on the reduction of hydrogen permeance in Pd-based membranes [14,16]. The aim of this work, in order to ll the void still present in the modelling of this system, is to propose a quantitative description of the hydrogen permeation process in the presence of carbon monoxide in the feed stream, following an approach which can be naturally extended to include the possible presence of other compounds, also affecting the H2 permeate uxes, such as H2 S or water vapour. In particular, a modication of the theory based on the sequence of elementary processes proposed by Ward and Dao [18] is considered, by introducing the competitive adsorption on the surface exposed to the feed gas. The model will be described in its main features and the effect of different parameters and operating conditions will be discussed. Several permeation data available from the literature and considering a broad range of operating conditions will then be considered and compared to the model calculations in order to test its reliability.

2. Theoretical background
Corresponding author. Tel.: +39 0512090408; fax: +39 0516347788. E-mail address: marco.giacinti@unibo.it (M. Giacinti Baschetti). 0376-7388/$ see front matter 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.memsci.2010.06.055

Hydrogen transport through Pd-based membranes is commonly described, in the so called diffusion limited regime, by the cou-

222

J. Catalano et al. / Journal of Membrane Science 362 (2010) 221233

pling of two main processes: (i) adsorption of molecular hydrogen and its dissociation into atoms, at the upstream membrane surface (H2 2H) and, vice versa, association and desorption (2H H2 ) at the downstream surface of the Pd-alloy lm and (ii) diffusion of atomic hydrogen in the metallic lattice [2,19]. In particular, in the absence of other gases adsorbing onto the metal surface, when the reaction is fast enough for the diffusion to be the controlling step of the process, Sieverts law holds and the hydrogen ux, N H2 , can be written as [20]: N H2 = PH2
.5 .5 (p0 p0 ) H ,ret H ,per
2 2

modication of Sieverts law: N H2 = 1 (T ) KCO pCO 1 + KCO pCO P0 H2 exp Ea RT


.5 .5 (p0 p0 ) H ,ret H ,per
2 2

(4)

Ea RT
.5 .5 (p0 p0 ) H ,ret H ,per
2 2

P0 H2 exp

(1)

where R is the universal gas constant, T is absolute temperature, pH2 is the hydrogen partial pressure in the retentate (ret) or permeate (per) side, is the membrane thickness, P0 H2 is the pre-exponential factor of hydrogen permeability, PH2 , and Ea represents the apparent activation energy for hydrogen permeability. Still remaining under diffusion limited conditions for the membrane behaviour, other resistances to the ux may also be present and have indeed been considered and modelled in different, more or less detailed ways. For example, different approaches have been used to describe resistances in the porous support, often present in palladium-based membranes, in which the gas can move through Knudsen diffusion, viscous ow or a combination of the two, depending on the average pore dimensions, or in the gas phase due to the concentration polarization phenomenon. Such resistances are frequently considered empirically in a lumped way, by modifying the exponent of the pressure dependence in Sieverts law as follows [17]: N H2 = P0 H2 exp Ea RT (pn H
2 ,ret

where pCO represents the carbon monoxide partial pressure at metal interface, while and KCO are the Langmuir afnity parameters. Both the approaches above retain the overall driving force for hydrogen ux as the difference of the square roots of hydrogen partial pressures at the opposite sides of the membrane and modify only hydrogen permeance, as if the effects of CO were present on both sides of the membrane, while in fact they are only acting on the retentate side. The resulting equations thus appears to some extent justied only when the downstream hydrogen pressure is negligibly small. In addition, these approaches are based on Sieverts law and can obviously be applied only when Eq. (1) holds for pure hydrogen permeation, that is only in the diffusion limited regime, while they cannot describe the desorption limited regime and cannot be applied to the whole permeation phenomenon in a broad range of operating conditions [16]. In fact, the adjustable parameters of both the above models have been estimated from a limited experimental data set, collected ad hoc by the respective authors, without considering a model validation procedure based on a comparison with the several available adsorption data in the presence of carbon monoxide; thus such lumped models represent more the description of a well dened and narrow set of experimental data rather than a general model useful for a broad range of operating conditions. On the other hand, since the parameters used do not refer to the modelling of any single kinetic step, but to the process as a whole, they have no stringent and precise physical meaning and their extension to different experimental conditions, if possible, is certainly not straightforward. 2.1. General model A complete and sound theoretical approach, on the contrary, needs to consider the effect of CO on hydrogen permeation by directly estimating its inuence on the hydrogen adsorption process which is just one of the steps of the whole permeation process. This can be assessed for example by considering the work of Johansson et al. [23] which studied in details the sticking probability of hydrogen, in the presence of carbon monoxide, on several transition metals supported on graphite, and took into account the competitive adsorption on the metal surface of the two gases in the equilibrium coverage. This kind of approach can be coupled with the description of all the other steps contributing to hydrogen permeation, to derive the actual inuence of carbon monoxide on the mass transport resistance of the membrane. To that aim, the fundamental approach originally proposed by Ward and Dao [18] can be followed, considering the sequence of the relevant elementary processes contributing to H2 transport in Pd lms, and their respective kinetic expressions, also on the basis of the works by Holleck [2] and by King and Wells [24]. In particular, Ward and Dao [18] considered the hydrogen transport as the result of the following series of steps: 1. molecular transport from the bulk gas to the gas layer adjacent to the surface; 2. dissociative adsorption onto the metal surface; 3. transition of atomic H from the surface into the bulk metal; 4. atomic diffusion of hydrogen through the bulk metal; 5. transition from the bulk metal to the surface at the permeate side; 6. recombinative desorption of hydrogen from the surface; 7. gas transport away from the surface to the bulk gas phase.

pn H

2 ,per

(2)

where the exponent n is considered to range between 0.5 and 1 depending on the dominant resistance in the system. Alternatively, such effects can be described through more detailed and appropriate approaches which account for the different additional resistances for hydrogen ux, and more correctly describe the hydrogen permeation process [21,22]. When the additional resistance is due to the presence of contaminants interacting with the membrane surface, such as carbon monoxide, the number of modelling works present in the literature becomes denitely lower and in general only a global lumped approach has been considered, adapting Eq. (1) by introducing an empirical correction factor in the proportionality coefcient of Sieverts law. In particular, Wang et al. [14] proposed a correction factor based on a Langmuir isotherm for the CO coverage, assuming that H2 needs two sites for the dissociative chemisorption, obtaining the following expression for hydrogen ux: N H2 = 1 (1 + c e
2 ECO /RT )

P0 H2 exp

Ea RT

.5 .5 (p0 p0 ) H ,per H ,ret


2 2

(3)

where the term in the square brackets is the correction factor for the effective surface area and ECO represents the heat of adsorption for carbon monoxide; the parameter c should depend on the partial pressure of CO, albeit in a non-specied way. Another equation to model the hydrogen ux from H2 CO mixtures was proposed by Barbieri et al. [16] who followed a similar approach, considered a Langmuir isotherm with a single site adsorption for hydrogen for the estimation of the available surface fraction and, based on a heuristic argument, proposed the following

J. Catalano et al. / Journal of Membrane Science 362 (2010) 221233

223

Table 1 Complete set of rate equations for each kinetic step contributing to hydrogen transport, as reported by Ward and Dao [18] under the simplied assumption that the hydrogen adsorption on the metal surfaces is described by a Langmuir adsorption isotherm, with no competing species. Specic kinetic step Dissociative adsorption Molecular desorption Surface-to-bulk Solid-state atomic diffusion Bulk-to-surface Kinetic equation
0.5 ads rH = 2 pH2 ,i 1 2 kTmH
2

Eq. no. S0 n2 (1 H,i ) S n2 2 S H,1 ES B H,i (1 RT (x x ) nb s,1 s,2


2

(I) (II)

des = 2 kd,0 exp rH S B rH = nS nb 0

2ED RT

exp
Ediff RT

xs,i )
EBS RT

(III) (IV) (V)

rH

diff

= D0 exp

B S rH = nS nb xs,i (1 H,i ) 0 exp

The complete theory associating an explicit kinetic equation to each of the steps listed above can be found in Ref. [18] in the case of pure hydrogen feeds, and the corresponding rate and ux equations for steps 26 are reported in Tables 1 and 2, respectively for more immediate reference. To our present aim, the presence of CO in the feed is accounted for by considering the simultaneous competitive chemisorption of hydrogen and CO on the feed side of the membrane surface. That leads to modied expressions for the hydrogen coverage fraction H and for the fraction of surface sites which are available for hydrogen adsorption, which are quantities used in the rate equations for steps 2 and 3, i.e. Eqs. (I) and (IV) of Table 1. Such effects were not included in the original work by Ward and Dao, who, however, pointed out that, in the presence of a contaminant, one expects qualitatively (i) a rise in the temperature at which the desorption limited regime would become the rate controlling step and (ii) a decrease of desorption limited ux. The competitive adsorption of CO and H2 on palladium must be considered and included in the theoretical framework proposed by Ward and Dao and, to that purpose the following reasonable assumptions are considered for simplicity sake: (i) mass transport resistance in the external gas phases is negligible; (ii) hydrogen and carbon monoxide follow a competitive Langmuir adsorption; (iii) carbon monoxide adsorption is lumped into a process involving a single palladium active site and is described by a single effective adsorption rate; (iv) CO does not disproportionate upon adsorption on the metal surface in the temperature range inspected, and only the surface of the retentate side is exposed to CO; (v) there is no diffusion of carbon monoxide through the palladium-based membrane. The rst assumption is considered to focus the attention merely on the membrane behaviour, and the others are introduced following a general consensus in order to simplify the mathematical approach. In particular for assumption (ii) it can be shown that, in the operative condition of interest, the hydrogen concentration

calculated in the bulk, as well as the hydrogen ux, do not vary signicantly using the Langmuir adsorption isotherm for hydrogen involving two sites, instead of the equation derived from the more complete statistical thermodynamic arguments [18]. Assumption (iii) is instead related to the fact that at least two different states of adsorbed CO are known to exist, i.e. linear and bridge-bonded with the metal surface, depending on the number of metal atoms binding the C atom: one for the linear and two for the bridgebonded species, respectively [25]. In the present case, to minimize the number of model parameters and in agreement also with the approach proposed by Johansson et al. [23], the choice has been made to describe the carbon monoxide adsorption through a single rate equation which thus must be regarded as an effective average between the different CO adsorption kinetics actually existing on the metal surface. As far as hypothesis (iv) is concerned, on the contrary, we notice that there are no experimental evidences on carbon monoxide dissociation in polycrystalline Pd lms above 500 K [26], nor for Pd (1 1 1) in the range between 300 and 750 K, even if experiments performed on silica supported Pd nano-clusters indicate that CO dissociates above 600 K at 185 mbar [27]. Assumption (v) is a consequence of the fourth one, given the substantial impossibility of the CO molecule to diffuse in the lattice due to its dimension. On the basis of the assumptions above and of the results obtained by Johansson et al. [23] on the sticking probability for H2 on metal surfaces in the presence of CO, the competitive adsorption of CO and H2 on palladium and Pd-based alloys is described by considering that atomic hydrogen and CO molecules both compete for the same surface active sites. Therefore, the adsorption and ads and r des , can be written as: desorption rates of CO per unit area, rCO CO
ads ads = kCO pint rCO CO (1 CO H ) nS des des rCO = kCO nS CO

(5) (6)

in which nS represents the overall number of adsorption sites available per unit area, CO and H represent the carbon monoxide and atomic hydrogen coverage fractions, respectively, (1 CO H ) is ads and kdes are the CO adsorpthe fraction of available active sites, kCO CO is CO partial pressure at tion and desorption rate constants and pint CO the metal interface. Since there is no CO ux across the membrane, under steady state conditions the adsorption and desorption rates

Table 2 Complete set of ux equations reported by Ward and Dao [18] in the simplied assumption that the hydrogen adsorption on the metal surfaces is described by a Langmuir adsorption isotherm. Kinetic stage Surface adsorption/desorption processes (upstream surface) Surface-to-bulk transition (upstream side) Solid-state atomic diffusion Bulk-to-surface (downstream side) Flux equation
0.5

Eq. no.
1 2 kTmH 0

N H2 = pH2 ,i N H2 = N H2 = N H2 =
1 n 2 S 1 2 1 2

S0 n2 (1 H,1 ) kd,0 exp S


E B S RT

2ED RT

n2 S H,1

2 E S B RT

(VI) (VII) (VIII)

nb

exp
Ediff RT

H,1 (1 xs,1 )
(xs,1 xs,2 ) EBS RT 1 2 kTmH

1 n 2 S

nb xs,1 (1 H,1 ) 0 exp nb exp


2 ES B RT

D0 exp

nb

nS nb xs,2 (1 H,2 ) 0 exp


2ED RT

1 nS 2 0.5

H,2 (1 xs,2 ) (IX) (X)

Surface desorption/adsorption processes (downstream surface) N H2 = kd,0 exp

n2 pH2 ,2 S H,2

S0 n2 (1 H,2 ) S

224

J. Catalano et al. / Journal of Membrane Science 362 (2010) 221233

for CO are identical and thus: CO = (1 H )


int CO pCO 1 + CO pint CO

(7)

ads /kdes , is a funcThe adsorption equilibrium constant, CO = kCO CO tion of temperature and can be calculated by using statistical thermodynamics arguments, as reported by Dulaurent et al. [25], as follows:

difference between the surface-to-bulk term, given by Eq. (III), and the bulk-to-surface term given by Eq. (12); therefore, in the presence of CO in the feed gas, the net surface-to-bulk transport is no longer given by Eq. (VII) used by Ward and Dao [18], but is rather given by: N H2 = nS nb
0

exp

ESB RT

H,1 (1 xs,1 ) nS nb xs,1 EBS RT (14)

CO

h3 k (2 mCO k)
3/2

1 T 5/2

exp

des E ads ECO CO

1 1+
int CO pCO

RT

(8)

(1 H,1 ) 0 exp

in which h is the Plancks constant, k is the Boltzmanns constant, des and E ads represent the activation energies for CO desorption ECO CO and adsorption, respectively, and mCO is the mass of a CO molecule; des E ads . of course the heat of CO adsorption is ECO = ECO CO In view of Eq. (7) the fraction of available active sites of the membrane surface exposed to the COH2 mixture is given by: (1 CO H ) = 1 1+
CO

pint CO

(1 H )

(9)

Eq. (9) represents the new expression needed for the fraction of available active sites which must be used in the presence of CO, in place of the expression (1 H ) which holds true merely when hydrogen is the only species adsorbed onto the metal surface. That expression will then be used to obtain the rate of dissociative adsorption of hydrogen over the upstream surface from COH2 mixtures as follows:
ads ads rH = 2 S0 kH pH2 n2 S (1 CO H ) ads = 2 S0 kH pH2 n2 S 2 2

1 1+
CO

pint CO

(1 H )

(10)

where, S0 is the sticking probability of H2 , and the rate of adsorption ads , has no activation barrier and can be expressed as coefcient, kH [23]:
ads kH =

1 (2 mH2 kT )
1/2

(11)

Also in the presence of carbon monoxide the entire sequence of the elementary kinetic steps which contribute to the hydrogen ux remains the same as indicated by Ward and Dao [18] in the absence of CO. The only difference is due to the fact that the upstream surface is endowed with a CO coverage which affects the rate of hydrogen adsorption/desorption as well as the rate of hydrogen transfer from surface-to-bulk, at the upstream side of the membrane. On the other hand, since CO is not present in the bulk of the metallic lm, the kinetic expressions for all other stages, namely diffusion of atomic hydrogen to the permeate side, hydrogen transport bulk-to surface at the permeate side and the desorption/adsorption stage at the permeate side, all remain expressed by the same equations which hold true in the case of pure hydrogen feeds, and thus Eqs. (VIII), (IX) and (X) of Table 2 are valid also in the present case. Therefore, the complete model for hydrogen transport through Pd-based membranes in the presence of CO in the feed, and in the absence of mass transfer resistance in the gas phase, is represented by Eqs. (13) and (14), describing the steps occurring at the feed side of the membrane, coupled with Eqs. (VIII), (IX) and (X) of Table 2, describing the kinetic steps within the membrane and at the downstream side of it. The set of ve non-linear equations was solved with the Newton method in Matlab programming environment. For model validation the results obtained have been compared with several available experimental data, spanning in particular a broad range of temperature values. 3. Parameters estimation In order to test the model reliability through comparison of simulation results with available experimental data, all the parameters appearing in the model equations presented above, and reported in Table 1, have been determined on the basis of independent information for different Ag contents in the palladium-based membranes in a broad temperature interval [2,18,2833]. The values obtained for Pd/Ag alloys, are reported in Table 3, while in the work by Ward and Dao [18] they were computed only for pure palladium membranes. The procedure used to obtain the values of the parameters appearing in the ux equations of Tables 1 and 2 will be presented for each kinetic step separately. 3.1. Surface processes The presence of silver in the system, affects the number of active sites per unit surface, nS , as well as the activation energy of the adsorption for the different species. The value of nS can be estimated simply from the alloy composition: since only palladium atoms allow hydrogen dissociative adsorption or CO adsorption, the number of active sites on the surface can be related to the Pd surface concentration, nS,Pd , that depends on palladium bulk concentration, nb,Pd through the equation [18]: nS = nS,Pd = nb,Pd NAV
1/3 2/3

in which mH2 represents the mass of a H2 molecule. Eq. (10) will then be used in place of Eq. (I) of Table 1, which holds only for the case in which the only adsorbed gas is hydrogen. Similarly, in the presence of CO in the feed the kinetic expression for the bulk-to-surface hydrogen transfer rate at the upstream membrane interface is no longer given by Eq. (V) of Table 1 but rather by the following equation:
B S rH = nS nb xs,i

1 1+
int CO pCO

(1 H,i ) 0 exp

EBS RT

(12)

Correspondingly, the net ux of molecular hydrogen arriving at the interface exposed to CO is nally calculated as the difference between the adsorption and desorption rates, described by Eq. (10) and by Eq. (II) of Table 1, respectively; one thus obtains: N H2 = pH2 ,1 (2 mH2 kT ) kd,0 exp
0.5

S0

1 1+
CO

pint CO

(1 H,1 )

2ED RT

2 n2 S H,1

(13)

Eq. (13) represents the net ux of molecular hydrogen reaching the upstream interface, and is of course different from Eq. (VI) which was used by Ward and Dao [18] to calculate the H2 ux on the surface of Pd membranes in the absence of CO. In parallel, the net hydrogen ux due to the surface-to-bulk transport at the upstream side of the membrane is given by the

(15)

J. Catalano et al. / Journal of Membrane Science 362 (2010) 221233 Table 3 Model parameters at several silver contents as calculated in this work or retrieved from the literature data indicated. xPd a (1010 ) Ref. [28] m 3.89 3.91 3.93 3.95 3.97 3.99 nb (105 ) Ref. Eq. (16) molPd m3 1.13 1.11 1.09 1.08 1.06 1.05 ns,Pd (105 ) Ref. Eq. (15) molPd m2 2.77 2.56 2.34 2.12 1.90 1.66 ESB Ref. [31] kJ mol1 52.79 48.60 43.89 47.36 47.59 56.65 0 (107 ) Ref. [29,30] m3 mol1 s1 6.78 6.33 5.51 5.26 3.59 4.37 EBS kJ mol1 22.19 22.19 19.47 24.70 26.38 33.50 D0 (107 ) Ref. [31] m2 s1 2.90 2.69 2.33 2.21 1.50 1.82 EAb Ref. [2] kJ mol1 16.75 23.18 28.22 30.69 33.58 29.69

225

0 S H Ref. [2] J mol1 K1 97.53 98.79 100.46 98.37 102.14 100.88

1 0.9 0.8 0.7 0.6 0.5

where NAV represents the Avogadros number; the Pd bulk concentration, nb,Pd , is then calculated from the density, i , and the molar mass, Mi , of the two metals of the alloy: nb,Pd = Pd MPd
Alloy

Pd MPd

Pd
Pd

1 Pd
Ag

(16)

In Eq. (16), in view of the lack of experimental data on alloys densities, the simplifying assumption of volume additivity has been made, which in any event is not expected to crucially affect the quality of the results nor the temperature dependence of the membrane response. A similar procedure is applied to calculate also the bulk concentration of silver atoms as well as the sum of the two, which is the total average concentration of metals atoms in the lattice, nb , so one has: nb = Pd 1 Pd + MPd MAg Pd
Pd

1 Pd
Ag

(17)

The activation energy of the adsorption and desorption processes can be calculated from experimental data or, when that is not possible, can be taken from the values available in the literature [3138]. In particular for hydrogen, the adsorption process is usually considered to have negligible activation energy [18], while the activation energy of desorption, ED , can be estimated from the dependence of hydrogen permeate ux on temperature in the desorption limited regime, as shown for example in Fig. 1. On the other hand, several estimates have already been reported for ED in previous works but they are distributed over a wide range of values with no univocal choice [3436]. For pure palladium, for example, Ward and Dao used ED values between 41 and 50 kJ mol1 to t the majority of literature data; nevertheless in the literature cited in

Ref. [18] values of ED are reported between 15 and 50 kJ mol1 . In addition, in the alloys, ED was found to be substantially unaffected by the presence of Ag in the lattice, through a differential scanning calorimetry (DSC) analysis performed by Artman and Flanagan [37] who investigated pure palladium as well as PdAg alloys containing 20% by weight of silver, while ED was reduced by the addition of silver in the metal surface, according to a rst principle study performed by Lovvik and Olsen [38]. Experimental data concerning 75% Pd 25% Ag alloy can be found in the work by Nguyen et al. [17], who reported a value of ED = 11.0 kJ mol1 , and in the work of Serra et al. [34] who found ED = 11.5 kJ molH 1 . On the other hand Bhargav and Jackson [33] found a coverage dependent activation energy for hydrogen desorption, with ED ranging from 35 to 27 kJ molH 1 , when the hydrogen coverage goes from zero to a complete hydrogen monolayer. In the present work, the choice has been made to consider a constant value of the activation energy for the desorption of hydrogen, obtained directly form pure hydrogen permeation experiments in the desorption limited region, whenever possible. For carbon monoxide, instead, the determination of the activation energies for the surface processes is complicated by a non-negligible dependence on CO surface coverage [25,3942] and by the existence of at least two different families of adsorbed carbon monoxide molecules, namely linear and bridge-bonded with the metal surface, as pointed out by the analysis performed in Ref. [25]. These two archetypal forms involve different numbers of metal atoms binding to the C atom, namely one atom for the linear and two atoms for the bridge-bonded species, respectively [40]. Dulaurent et al. [25] found that the heat of adsorption ( ECO = des E ads ) of carbon monoxide on pure palladium decreases form ECO CO 92 to 54 kJ mol1 and from 168 to 92 kJ mol1 increasing the CO coverage fraction from zero to one for the linear and the bridgebonded species, respectively. The decrease of the heat of adsorption is related to the repulsive interaction between adsorbates [39] as well as to the fact that, due to the surface heterogeneity, different adsorption sites are present on the metal surface and those associated to high adsorption energies are usually more reactive and thus are lled rst [40]. In particular, ECO can be reasonably described by a linear relationship with CO coverage and one can estimate its value, from the analysis of literature data [25,41,42], as about 125 kJ mol1 for CO = 0.5, obtained by neglecting the possible effects due to different co-adsorbed molecules. In view of these ndings, a linear dependence of the heat of adsorption ECO with CO coverage was used in the following to best t the experimental behaviour of the hydrogen permeated ux H2 CO mixture: ECO = aCO + b 3.2. Surface-to-bulk transitions Surface-to-bulk transition involves the hydrogen transfer from adsorption active sites on the surface to the bulk sites in the lattice and vice versa; the parameters entering the rate equations for these kinetic steps, Eqs. (14) and (IX), are the activation energies ESB (18)

Fig. 1. Data tting and model sensitivity analysis versus activation energies for desorption in a 250 m PdAg membrane at 14 kPa of pure hydrogen in the retentate side and vacuum in the permeate side [10].

226

J. Catalano et al. / Journal of Membrane Science 362 (2010) 221233 Table 4 Main characteristics of the membranes and experimental operative conditions of the literature data analyzed in this work. Authors Ref. Silver content by weight Thickness (m) Total feed pressure (kPa) CO partial pressure (kPa) Temperature range (K) Chabot et al. [10] 23% 250 147 0.2914 373723 Nguyen et al. [17] 25% 75, 100, 300 101.325 0.5020 300773

and EBS of the different processes involved, which can be obtained directly from experimental data [18], and the pre-exponential factors 0 and 0 related, respectively, to jump frequency of the hydrogen atoms from the bulk to the surface of the membrane and vice versa. The activation energy for the bulk-to-surface transition, EBS , was set equal to the activation energy of metal bulk diffusion, as previously stated by other authors [18,33], while, ESB , related to the inverse process, was calculated following the procedure suggested by Ward and Dao which estimated its value from the knowledge of EBS and the heats of absorption ( EAb ) and adsorption ( EAd ) of hydrogen in palladium following the idea that the sum of the two activation energies should be equal to the sum of the two absorption and adsorption heats. The values for latter parameters for the different PdAg alloys considered were obtained from the data available in the open literature: in particular the heats of absorption found in the Sieverts limit were taken from Holleck [2] and are reported in Table 3, while EAd was considered independent of silver content and its value was xed at 38 kJ mol1 [37]. Following the same argument presented by Ward and Dao [18], 0 has been calculated through the value of the atomic diffusion coefcient in the metal lattice considering that the jump frequency from the bulk to the surface is similar to that for the diffusion in the metal lattice, since in both cases the hydrogen atom is jumping from a bulk site. By equating the two frequencies it is possible to write, under conditions of hydrogen coverage approaching zero [18]: 0 = 4D0 a2 nb (1 H ) (19)

4. Results and discussion 4.1. Model validation In spite of a large amount of literature data on the hydrogen permeation in the presence of carbon monoxide, only few authors investigated experimentally wide ranges of temperature and carbon monoxide compositions, and in particular the experimental data by Chabot et al. [10] and Nguyen et al. [17] appear relevant and have been selected in order to test the accuracy of the model proposed. The main characteristics of the membranes, as well as the experimental operative conditions used in Refs. [10,17] have been reported in Table 4. In both works the experimental setup used is gas-resistance free and is characterized by extremely low hydrogen permeate pressures (the permeate was removed by means of a vacuum pump); therefore, the model proposed above can be applied by using membrane properties alone, with no need to introduce any other parameter or equation relative to gas phase or support resistances at the downstream side of the membrane. Finally, pure hydrogen permeation rates were considered also in desorption limited regimes, leading to a more stringent validation of the model proposed: indeed, that allows to determine in an accurate way the activation energy for the hydrogen desorption in silverpalladium alloys, for which there is some discrepancy, as mentioned before, among the values reported in different works [3336]. In order to capture the pure hydrogen data by Nguyen et al. and Chabot et al., it is necessary to use values of Edes = 11 kJ molH 1 and Edes = 13.1 kJ molH 1 , respectively, where the rst value was directly suggested by the authors while the second was calculated using Ward and Dao approach to t the experimental data as shown in Fig. 1. Such values are denitely lower than those proposed by Bhargav and Jackson [33] for a 23 wt% Ag alloy (between 27 and 35 kJ molH 1 ) while are comparable to experimental value of 11.5 kJ molH 1 found in 25 wt% Ag membranes by Serra et al. [34]. The experimental hydrogen permeate uxes by Chabot et al. were collected in a 250 m silverpalladium (2377 wt%) tube without any ceramic or metallic support, with a constant upstream hydrogen partial pressure of 14 kPa and in the temperature range between 373 and 723 K. They are reported in Fig. 2, together with the results of our model calculations, for all gas mixtures considered, containing carbon monoxide at partial pressure between 0.29 and 14 kPa. In order to t the experimental data a linear relationship between the CO heat of adsoprtion and carbon monoxide coverage was used: the value of 125 kJ mol1 for CO = 0.5 and a value at zero coverage of 140 kJ mol1 were used for ECO , according to Ref. [42]. The latter value is actually smaller with respect to the maximum value of about 168 kJ mol1 reported for CO = 0 in pure palladium [25]; nevertheless it is reasonable to consider a nonnegligible reduction of the heat of adsorption of CO when palladium silver alloys are used instead of pure palladium [43]. A rather good agreement can be noticed between the experimental data and the results of model calculations, in the entire region of CO [0, 0.95] represented by the solid curves. The agreement appears particularly remarkable considering that the activation energies for diffusion and for hydrogen desorption are

from which one can calculate 0 , using the proper lattice parameter, a, as well as the specic pre-exponential factor of the diffusion coefcient, D0 of the silver alloy [28,31]. The pre-exponential factor, 0 , depends on the thermodynamic properties of the metalhydrogen system. Its value can be calculated following the same procedure used by Ward and Dao [18], by imposing the validity of Sieverts approximation at low hydrogen concentration, in the diffusion limited regime. In this case, it is assumed that the relationship between the jump frequency and the surface-to-bulk rate constant is only dependent on the silver content of the membrane, while it is not inuenced by the presence of carbon monoxide on the metal surface; therefore, the value of 0 can be calculated once the Sieverts constant of the hydrogen in the alloy under consideration is known. By considering Langmuir adsorption for hydrogen atoms and following the same algebra shown in Ref. [18] the nal equation correlating 0 to 0 is: 0
0

= exp

0 S H R

0.5 S0 0.5 nS,Pd kd, (2 mH2 kT ) 0 0.25

(20)

3.3. Atomic diffusion Extensive studies on the diffusion of hydrogen atoms in several silverpalladium alloys have already appeared, in which both pre-exponential factor and activation energy can be found as a function of silver content [2,31]. The pre-exponential factors D0 taken from Ref. [31] and reported in Table 3, for example, have been used in the calculations performed in this work. The values of Ediff on the other hand can also be calculated directly from pure hydrogen permeation experiments, by considering the dependence of permeability on temperature in the diffusion limited regimes. This approach allows to obtain this parameter directly for the membranes of interest giving a more stringent validation of the models capabilities; therefore it was preferred to the use of literature data and was applied whenever possible.

J. Catalano et al. / Journal of Membrane Science 362 (2010) 221233

227

Fig. 2. Comparison between experimental data taken from Ref. [10] collected in 250 m silver palladium membrane with an hydrogen partial pressure on the retentate side of 14 kPa at several carbon monoxide partial pressures and results of model calculations (lines).

Fig. 3. Comparison between experimental data taken from Ref. [17] collected in 100 m silver/palladium (25 wt%/75 wt%) membrane, with 1 atm of total pressure in the retentate side and vacuum on the permeate side, at several carbon monoxide concentrations and results of model calculations (lines).

taken from the best t with pure hydrogen experimental data, so that only the heat of adsorption of CO is used to describe all the curves for different CO compositions; in view of the linear relationship between ECO and CO , all the experimental data taken at different carbon monoxide concentrations are satisfactorily described by the use of just two parameters, that possess a completely dened physical meaning and can be compared to other available literature data; the best t between model calculations and experimental data, over the entire range of temperatures and of CO content in the feed, was obtained by using Eq. (18) with a = 30 kJ mol1 and b = 140 kJ mol1 . Only at very high carbon monoxide coverage (i.e CO > 0.95) the curves predicted by the model for the H2 permeation ux (dotted lines in Fig. 2) deviate from the experimental data. In these conditions, indeed, the interaction between ad molecules becomes signicantly higher and the dependence of the heat of adsorption of carbon monoxide on CO coverage is thus increased, and cannot be described by a simple linear relationship. Experimental data taken by Nguyen et al. [17] for a 25/75 silver palladium membrane of 100 m and collected at 1.013 bar of total retentate pressure and carbon monoxide composition up to 20 vol.%, are reported in Fig. 3. In order to provide a proper comparison with data obtained at different upstream or permeate pressures, in this case the data are not reported in terms of ux but rather in terms of the permeability PH2 dened in Eq. (1); even though PH2 is the proper quantity to consider in the diffusion limited regime when Sieverts law holds, it will be considered also for a general comparison between the different behaviours observed by varying the operating conditions as temperature and CO content in the feed gas. In Ref. [17] an anomalous behaviour of the hydrogen permeability was found in pure H2 experiments, with the presence of a maximum in the permeability curves that was not explained by the authors. Such a phenomenon has not been observed by other authors and cannot be predicted by the simple model proposed here; as a consequence these data have not been considered and the comparison was limited to pure hydrogen data at temperatures higher than 550 K (diffusion limited regime) and lower than 450 K (desorption limited regime). By using the same heat of adsorption for carbon monoxide already obtained from the data by Chabot et al. and reported above, the hydrogen permeate ux calculated from the model underestimates the experimental data and for an optimal t the value of coefcient b in Eq. (18) should be 120 kJ mol1 ,

while the same slope a = 30 kJ mol1 is required as for the best t of the experimental data by Chabot et al. The difference between the heat of adsorption used to describe all the data by Chabot et al. and by Nguyen et al. cannot be attributed only to the different silver content of the membranes used in the two works, which is indeed too small (i.e. 23 wt% against 25 wt%, respectively) to produce a ECO reduction of about 20%. On the contrary, the motivation for the different values of ECO required by the two different sets of data can be associated to the different behaviour of the two membranes, which showed different responses in the process to restore the permeance to pure hydrogen experienced before CO adsorption. Indeed, Nguyen et al. [17] observed that CO adsorption on their membrane was easily reversible and the pure hydrogen permeance before and after the addition of carbon monoxide attained the same value with no need of a particular regeneration procedure. On the contrary, the membranes used by Chabot et al. [10] experienced a non-reversible poisoning upon CO exposure, and the original permeance of pure hydrogen could be restored only by heating the membrane under vacuum or by applying a cleaning process. This fact suggests that CO adsorption was different for the two membranes and in particular it was stronger in the membranes used by Chabot et al. which in fact requires a higher value of ECO to best t the experimental data. The observed differences may reasonably be attributed to the fact that different amounts of the two types of adsorbed carbon monoxide species, linear and bridge-bonded, are present over the membranes considered in Refs. [10,17], and that can lead to the different ECO values required to best t the two different sets of data. Data of Ref. [17] are also useful to investigate the relationship between hydrogen permeability and thickness of the metal layer, because membranes with three different thicknesses were employed, i.e. 75, 100 and 300 m. The hydrogen permeability calculated for the case of pure hydrogen as well as for binary mixtures containing also carbon monoxide are reported in Fig. 4. As pointed out by Nguyen et al., it is immediate to notice that in the presence of CO only slight differences in hydrogen permeability are observed for different membranes thicknesses, while for pure hydrogen the decrease in the hydrogen permeability with decreasing temperature is more marked for the thinner membrane. Interestingly this feature is correctly predicted by the model without changing the parameters found in the initial tting of the data. Therefore, the

228

J. Catalano et al. / Journal of Membrane Science 362 (2010) 221233

tion from the downstream surface. In the presence of CO, on the contrary, one never reaches a desorption limited regime since the effects of CO adsorption onto the upstream interface result in a relevant resistance dominating the overall kinetics already at temperatures where the desorption resistance is still negligible. The different asymptotic behaviours encountered by decreasing signicantly temperature, in presence or absence of CO, are thus associated to different rate controlling steps: desorption on the permeate side for pure hydrogen conditions, adsorption on retentate interface in the case of H2 CO feeds. From a mathematical point of view the situation is described by the different asymptotes due to surface limited regimes in absence and presence of carbon monoxide. In the rst case, the desorption limited ux is obtained by Eq. (X) of Table 2, describing the surface adsorption/desorption process at the downstream side of the membrane, when the hydrogen coverage is unity: N H2 = kd,0 exp
Fig. 4. Calculated hydrogen permeability behaviour for pure hydrogen and mixtures containing also carbon monoxide for membranes with different thickness and comparison with data from Ref. [17].

2ED RT

nS 2

(21)

which is written in terms of hydrogen permeability recalling Eq. (1): PH2 = kd,0 nS 2 .5 0.5 pH ,ret p0 H2 ,per 2 exp 2ED RT (22)

approach used to account for the effects of CO on the hydrogen ux is able to catch the qualitative behaviour of the experimental trend also in this case, and to obtain even a satisfactory quantitative agreement with the use of the same values of the two tting parameters a and b in Eq. (18) for ECO calculation. 4.2. Model analysis and predictions After model validation we can now analyze different aspects affecting hydrogen permeation in order to understand their relative importance at different operating conditions, both with or without CO in the feed. A sensitivity analysis is thus performed by changing the relevant model parameters around the values obtained to t the data reported in Ref. [17], and the expected response of the system is investigated and discussed, even though experimental data are not directly available for comparison. The effects of CO in the feed gas on hydrogen permeation have in all cases a clear feature both at higher temperatures and at lower temperatures: at high temperatures, say above 623 K, the presence of CO gives only very minor deviations from pure hydrogen results, while at lower temperatures a dramatic permeability decrease is observed with respect to the pure H2 case. The behaviour observed at high temperature [10,17] is correctly described by the model that does not predict any ux depletion in that region due to CO presence, as it is clearly shown in Figs. 24. At lower temperatures, on the contrary, two different limiting behaviours are clearly distinguishable in the absence or in the presence of CO; for the pure hydrogen case permeation is controlled at low temperatures by the desorption process at the permeate side, as already discussed for instance in ref [18], while a much steeper decline of permeability is observed in the Arrhenius plot, in presence of even small amounts of carbon monoxide starting already when the desorption resistance downstream is still negligibly small. The model presented accounts for the observed behaviour and indicates that at high temperature the overall kinetics is controlled by diffusion so that surface processes, such those involving CO upstream or desorption downstream, do not inuence the permeation behaviour as actually observed from the experimental data [10,17]. When temperature is decreased, however, the situation drastically changes. For the case of pure hydrogen the change in slope of the permeability in the Arrhenius plot at lower temperatures marks the switch from the regime controlled by diffusion through the membrane and the regime controlled by the desorp-

On the other hand, in the presence of CO the limiting value observed is obtained when the surface process controlling the permeation rate is the hydrogen adsorption on the retentate side; in that case, the asymptotic ux is given by Eq. (13), considering a complete CO coverage or, equivalently, by setting the hydrogen coverage equal to zero: N H2 = pH2 ,1 (2 mH2 kT )
0.5

S0 nS 2

1 1+
CO

pint CO

(23)

The corresponding permeability at low temperatures is then evaluated by Eq. (1) at each retentate and permeate pressures as: PH2 = pH2 ,ret .5 .5 p0 p0 H2 ,ret H2 ,per (2 mH2 kT )
0.5

S0 n2 S

1 1+
CO

pint CO (24)

The use of the permeability dened in Eq. (1), which is consistent with Sieverts law, also outside the region in which the diffusion limited regime holds allows to compare directly various conditions in which the rate controlling steps are different; on the other hand it introduces a dependence on upstream and downstream pressures, as it is clear in Eqs. (22) and (24), which may appear somewhat articial. The maximum limiting values of the permeabilities are reached when the downstream pressures, pH2 ,per , is zero. The slope of the asymptotes presented by the experimental data varies as a function of CO partial pressure, due to the already discussed dependence of ECO on surface coverage, as well as of membranes thickness, and both effects are correctly predicted by the model, as shown in Figs. 24. From the model analysis we can also understand better the effect of membrane thickness: indeed, for the case of pure hydrogen the change in slope of the permeability in the Arrhenius plot marks the switch from the regime controlled by diffusion through the membrane and the regime controlled by the desorption from the downstream surface, at lower temperatures. The desorption resistance is independent of membrane thickness, while the resistance to diffusion increases with membrane thickness so that, for thicker membranes the transition takes place at lower temperatures. In the presence of CO, on the contrary, one never reaches a desorption limited regime and the temperature at which CO adsorption

J. Catalano et al. / Journal of Membrane Science 362 (2010) 221233

229

Fig. 5. Model calculated permeability obtained varying the hydrogen partial pressure on the permeate side, for a 100 m Pd/Ag (75/25 wt%) membrane. The inset points out a difference of 5% in the permeability of pure hydrogen with respect to the theoretical limit of diffusion limited regime, for Pper of 0 and 1 atm.

Fig. 6. Hydrogen permeability calculated from the model at different activation energies for the hydrogen desorption. Data calculated for a 100 m Pd/Ag (75/25 wt%) membrane considering an upstream hydrogen partial pressure of 1 bar and vacuum in the permeate side.

starts dominating the overall kinetics is less dependent on membrane thickness due to the higher energy involved in this process ( ECO > ED ). The model, therefore, clearly indicates that the addition of CO, even in very small quantities, substantially changes the relative importance of the different processes involved in hydrogen permeation; when low temperatures are considered, indeed, diffusion resistance becomes negligible and the rate limiting step is offered by the competitive adsorption of CO and H2 on the retentate side of the membrane, rather than by hydrogen desorption from the permeate side, as in the pure hydrogen case. In the presence of CO, therefore, the system switches from diffusion resistance to adsorption resistance on retentate side, and the downstream processes do not play a role in determining the permeation rate. In order to clarify better this point a sensitivity analysis has been performed by changing the permeate pressure and the activation energy of desorption, to study their inuence on permeability. The results are shown in Figs. 5 and 6, respectively. Remarkably, Fig. 5 shows that when the permeate pressure is increased from 0 to 1 atm, for a given feed pressure (2 atm absolute), the permeability of pure hydrogen dened in Eq. (1) experiences a decrease in the temperature range below 623 K. The change in permeate pressure indeed affects the hydrogen coverage at the downstream side of the membrane and changes the relative importance of the desorption step on the whole permeation process: by increasing Pper the hydrogen coverage increases so that the desorption resistance at the permeate surface gains importance earlier with respect to diffusion. This fact is apparent also in the inset of Fig. 5, in which the temperature value above which permeation is in the diffusion limited regime increases with increasing the permeate pressure from 0 to 1 atm: in the diffusion limited regime Sieverts law holds and the permeability dened in Eq. (1) is constant, unaffected by changes in the operative conditions. With decreasing temperature the other resistances increase their relevance; in that case, changes in permeate pressure inuence the rate of the surface process at the permeate side of the membrane, making the resistance of this step not negligible, even if not dominating, with respect to that due to diffusion or to the competitive adsorption of carbon monoxide. In the presence of CO, the resistance associated

to the complete adsorption of CO is localized at the upstream interface and is not directly affected by the permeate pressure, so that the change in the curves describing COH2 permeation below 600 K is actually due to the fact that now the adsorption becomes soon important and actually dominant in a temperature range where the ux is already decreased due to the desorption resistance. Fig. 6 shows, on the other hand, that hydrogen permeability in presence of carbon monoxide is substantially unaffected by a change in ED since all the curves collapse on one another, indicating

Fig. 7. Comparison between data from Ref. [14] for a 100 m pure palladium membrane at 0.51 bar of hydrogen partial pressure in the retentate side and vacuum applied in the permeate side and model calculations considering three different activation energies for hydrogen desorption. The dashed straight lines indicate the molecular desorption limited uxes.

230

J. Catalano et al. / Journal of Membrane Science 362 (2010) 221233

Fig. 8. Calculated behaviours of hydrogen coverage and atomic hydrogen molar fraction at the two metallic surfaces at the conditions used in the experimental work by Chabot et al. [10]. (a) H2 coverage at retentate side; (b) H2 coverage at permeate side; and (c) H2 molar fractions at the two sides of the membrane.

that in presence of CO the permeation behaviour is quite insensitive to variations of this parameter that instead strongly affects pure hydrogen permeability in the desorption limited regime; changes in desorption activation energy can increase the relative weight of desorption process with respect to diffusion or CO adsorption, but the system behaviour is insensitive to this parameter, contrary to what seen for pure H2 . That is particularly interesting because it makes it possible to apply the model developed also for experimental data relative only to the diffusion limited regime, that are the great majority of those found in literature, with no information on the desorption activation energy. For example, we can consider the experimental data by Wang et al. [14] for a 100 m pure palladium membrane, which were collected at 0.51 bar of hydrogen partial pressure on the retentate side, while vacuum was applied in the permeate side in order of 0.11 Pa. The carbon monoxide partial pressure upstream was xed at 0.66 kPa and the temperature inspected was in the range between 400 and 473 K. From the analysis of the experimental data it is clear that, in the temperature range inspected by Wang et al. [14], there is no slope change in the Arrhenius type plot shown in Fig. 7 and that all permeability data fall in the diffusion limited regime. The desorption limited curves associated to different activation energies have been drawn in the same Fig. 7, allowing to estimate the upper-limit of ED compatible with the experimental

data, that is around 38 kJ mol1 ; however, the use of lower values for the activation energy ED , i.e. 26, 30 and 34 kJ mol1 , does not change substantially the response of the model in the presence of carbon monoxide as it is also shown in the same gure. The agreement of the model results with the experimental data is quite good with the same values of the heat of adsorption for carbon monoxide found for experimental data by Nguyen et al., that is using Eq. (18) with a = 30 kJ mol1 and b = 120 kJ mol1 , which give ECO values denitely closer to the experimental values found in the literature for this parameter [25,38,39,42], rather than the value of 30.7 kJ mol1 indicated in Ref. [14] using the global lumped approach given by Eq. (3). 4.3. Simplied model The complete model formulated above may appear too elaborated and possibly more complex than actually needed to represent the observed behaviour of hydrogen ux in the presence of CO, just as the complete Ward and Dao model is more complex than required to represent the hydrogen ux in the diffusion controlled regime for pure hydrogen. Therefore, it is worthwhile to try to simplify the model as far as possible, by removing from the description those processes that have lower or negligible impact on the ux measured. The most intuitive approach in this case would be

J. Catalano et al. / Journal of Membrane Science 362 (2010) 221233

231

Fig. 9. Comparison between complete and simplied model at two different carbon monoxide partial pressures, for: (a) hydrogen permeate ux; (b) hydrogen coverage in the upstream surface; (c) carbon monoxide coverage in the upstream surface; and (d) hydrogen concentration in the sub-surface. The calculations are made for a 250 m thick 23% Ag/Pd membrane, with ECO = 14030CO kJ mol1 , Edes = 13.1 kJ mol1 , PH2 ret = 14 kPa, Pper = 1 Pa.

to consider the diffusion limited regime, trying to derive a modied Sieverts law accounting for the presence of CO similar to what presented already by other authors [14,16]; however such an approach, albeit very appealing, is impractical for the present purpose. In fact it is easy to notice that if adsorption equilibrium holds for hydrogen on the retentate surface, the simultaneous bulk-to-surface equilibrium would lead to an equilibrium condition between the atomic hydrogen mol fraction xs,1 and the external partial pressure of hydrogen, thus ruling out all effects due to the presence of CO. In the purely diffusion limited regimes, therefore, there is no substantial inuence of CO on hydrogen ux, as it is expected since the competitive absorption is a surface process, and as it is also conrmed by the experimental data (Figs. 2 and 3 above 673 K), which in general show very limited ux reduction in these working conditions [10,17]. A more comprehensive analysis of the role of the different processes involved in hydrogen permeation can be made by considering the behaviours of hydrogen coverage on the membrane surface and of hydrogen concentration in the bulk as a function of the partial pressure of carbon monoxide in the feed (Fig. 8). From Fig. 8a it can be seen that the hydrogen coverage at the retentate side becomes smaller and smaller by increasing the carbon monoxide partial pressure upstream; this behaviour indicates that in the presence of CO the temperature at which the diffusion lim-

ited regime ceases to be the dominant resistance of the process is substantially increased, as it was already pointed out by Ward and Dao. In the presence of carbon monoxide, however, the ux is not limited by desorption from the permeate side of the membrane, but, as already shown, by the adsorption step on the retentate side, where the competitive effect of carbon monoxide plays its important role. That can be conrmed also by analyzing the behaviour of the hydrogen coverage at the permeate side, reported in Fig. 8b; in the presence of CO, H does not reach the value of unity as it happens for pure hydrogen in the case of desorption limited regime (see Eq. (21)); on the contrary, it has a complex trend which becomes independent of CO partial pressure when the surface limited regime is reached, that is when the hydrogen coverage at the retentate side approaches zero. From Fig. 8c it can be seen that, at the same time, also the value of hydrogen mol fraction in bulk palladium at retentate side collapses on that of the permeate side and eventually reaches the value which can be calculated by Sieverts law, considering the equilibrium with permeate hydrogen partial pressure. More precisely, at the permeate side the hydrogen bulk concentration never differs substantially from its equilibrium value so that the kinetics of all the surface and sub-surface processes can be neglected at the permeate side of the membrane. By looking at the model behaviour, therefore, the complete approach can be simplied considering the following conditions:

232

J. Catalano et al. / Journal of Membrane Science 362 (2010) 221233

(a) the downstream sorption/desorption kinetics are extremely fast so that actual equilibrium holds for them; (b) also the bulk-to-surface kinetics is extremely fast, so bulk/surface equilibrium conditions are satised at the downstream side of the membrane; (c) the mol fractions of atomic hydrogen in the membrane bulk are systematically much smaller than unity; (d) similarly to point (b), the bulk-to-surface kinetics is considered extremely fast also at the upstream side. Vice versa, in view of the previous considerations, at the upstream membrane surface equilibrium is considered to hold only for CO adsorption, and not for hydrogen. The equations of the general model, therefore, can be simplied on the basis of the assumptions indicated above, leading to a set of three equations instead of ve as it was for the complete model. In particular the system behaviour can be obtained by solving Eq. (13), which remains unchanged also in the simplied approach, together with the equilibrium condition for the bulk-to-surface ux at the feed side, which can be obtained from Eq. (14) setting the hydrogen ux equal to zero, and the diffusion equation (Eq. (IV) of Table 1) where, due to equilibrium at the permeate side, the value of xs,2 can be directly calculated through Sieverts constant from the value of downstream hydrogen pressure. Considering the heat of adsoprtion for the carbon monoxide independent of CO coverage, the simplied model results in a single implicit equation relating the hydrogen ux to the upstream and downstream H2 partial pressures and to the upstream partial pressure of CO, while considering ECO a function of coverage the above simplications yield to a set of two equations. The resulting equations, even in the case of ECO constant, are clearly more complex than the simplied heuristic models given in Eqs. (3) and (4), respectively; nonetheless all the parameters entering the model have a precise physical meaning and a well dened temperature dependence and can be obtained from independent measurements. In Fig. 9 four parity plots are presented, in which the hydrogen permeate ux, the hydrogen and carbon monoxide coverage as well as the hydrogen concentration at the upstream interface of the membrane calculated from the simplied model are compared with the results obtained by using the complete general model; in particular, the parameter optimized on experimental data from Chabot et al. [10] has been used. The maximum deviation observed between complete and simplied model is less than 0.1%, conrming the possibility to adopt the simplied model to describe the hydrogen permeation in the presence of carbon monoxide. 5. Conclusions A model for hydrogen permeation in palladium and palladiumbased membranes is proposed, in the presence of carbon monoxide in the feed. The well know equations used by Ward and Dao [18] to describe the whole hydrogen permeation process were modied in the adsorption step to account for the simultaneous competitive adsorption of hydrogen and carbon monoxide molecules. The adsorption of the carbon monoxide as well as that of hydrogen were simply described using a Langmuir type adsorption and considering steady state conditions for co-adsorption. The parameters needed for the calculations are endowed with a precise physical meaning and were obtained by best tting the model to the experimental data available over a rather broad range of experimental conditions, in particular of temperature and CO partial pressure, for adsorption on pure palladium, and have been adapted to describe the permeation in PdAg membranes. The calculated behaviour obtained by changing temperature, hydrogen and CO partial pressures as well as membrane thickness were compared with available liter-

ature data and showed a good agreement, at all the temperatures inspected. In this procedure the only tting parameter used was the heat of adsorption of carbon monoxide, that was allowed to vary in the range of values experimentally observed for different Pd-based membranes.

Acknowledgement This work was performed with nancial support of the Italian Ministry of University and Research Contributo del Fondo Integrativo Speciale Ricerca FISR DM 17/12/2002-anno 2001 Progetto: Idrogeno puro da gas naturale mediante reforming a conversione totale ottenuta integrando reazione chimica e separazione a membrana.

Nomenclature Symbols a lattice parameter [m] c pre-exponential factor for the CO coverage in Eq. (3) D0 pre-exponential factor of the diffusion coefcient (m2 s1 ) Ea apparent activation energy for the hydrogen permeation phenomenon (J mol1 ) EAb heat of absorption of hydrogen in palladium (J mol1 ) EAd heat of adsorption of hydrogen in palladium (J mol1 ) ESB activation energy for the surface-to-bulk transition (J mol1 ) EBS activation energy for the bulk-to-surface transition (J mol1 ) ads ECO activation energy for CO adsorption on the surface (J mol1 ) des ECO activation energy for CO adsorption from the surface (J mol1 ) ED activation energy for hydrogen desorption (J mol1 ) Ediff activation energy for the atomic diffusion coefcient (J mol1 ) h Plancks constant (J s) k Boltzmanns constant (J K1 ) KCO Langmuir afnity parameter in Eq. (4) kd,o pre-exponential factor for the molecular desorption constant (m2 mol1 s1 ) ads ki adsorption constant for species i (Pa1 s1 )
des ki mi Mi n NAV nb nb,Pd

N H2 nS nS,Pd pi H P 2 R rCO S0

desorption rate constant for species i (m4 mol2 s1 ) mass of a molecule of species i (kg mol1 ) molecular mass of species i (kg mol1 ) empirical exponent in modied Sieverts law Eq. (2) Avogadros number Pd bulk concentration in palladium (molPd m3 ) Pd bulk concentration in palladium alloys (molMe m3 ) hydrogen ux (mol m2 s1 ) total number of adsorption sites available on the palladium surface (molPd m2 ) total number of adsorption sites available on the alloy surface (molPd m2 ) partial pressure of species i (Pa) hydrogen permeability (mol m1 s1 Pa0.5 ) universal gas constant (J mol1 K1 ) adsorption or desorption rates (s1 ) hydrogen sticking coefcient

J. Catalano et al. / Journal of Membrane Science 362 (2010) 221233

233

T xs,i

absolute temperature (K) H/Pd ratio in bulk metal referred to surface i

Greek symbols parameter in Eq. (4) ECO heat of adsorption for carbon monoxide (J mol1 ) 0 partial molar entropy of dissolution of hydrogen S H atoms at innite dilution (J mol1 K1 ) membrane thickness (m) adsorption coefcient for CO (Pa1 ) CO pre-exponential factor for the surface-to-bulk metal 0 transition (m3 mol1 s1 ) i coverage of species i density of species i (kg m3 ) i i weight fraction of species i Superscripts and subscripts ads refers to adsorption des refers to desorption per refers to permeate side ret refers to retentate side

References
[1] T. Graham, Philos. Trans. Roy. Soc. 156 (1866) 415431. [2] G.L. Holleck, Diffusion and solubility of hydrogen in palladium and silverpalladium alloys, in: G. Alefeld, J. Volkl (Eds.), Hydrogen in Metals II, Springer, Berlin, Heidelberg, 1978. [3] F.A. Lewis, The Palladium Hydrogen System, Academic Press, London, 1967. [4] A. Pisarev, V. Shestakov, R. Hayakawa, Y. Hatano, K. Watanabe, Gas-driven hydrogen permeation in the surface-limited regime, J. Nucl. Mater. 320 (2003) 214222. [5] J. Tong, R. Shirai, Y. Kashima, Y. Matsumura, Preparation of a pinhole-free PdAg membrane on a porous metal support for pure hydrogen separation, J. Membr. Sci. 260 (2005) 8489. [6] K.S. Rothenberger, A.V. Cugini, B.H. Howard, R.P. Killmeyer, M.V. Ciocco, B.D. Morreale, R.M. Enick, F. Bustamante, I.P. Mardilovich, Y.H. Ma, High pressure hydrogen permeance of porous stainless steel coated with a thin palladium lm via electroless plating, J. Membr. Sci. 244 (2004) 5568. [7] N. Itoh, T. Akiha, T. Sato, Preparation of thin palladium composite membrane tube by a CVD technique and its hydrogen permselectivity, Catal. Today 104 (2005) 231237. [8] D.A. Pacheco Tanaka, M.A. Llosa Tanco, S. Niwa, Y. Wakui, F. Mizukami, T. Namba, T.M. Suzuki, Preparation of palladium and silver alloy membrane on a porous -alumina tube via simultaneous electroless plating, J. Membr. Sci. 247 (2005) 2127. [9] Y. Chen, Y. Wang, H. Xu, G. Xiong, Hydrogen production capacity of membrane reformer for methane steam reforming near practical working conditions, J. Membr. Sci. 322 (2008) 453459. [10] J. Chabot, J. Lecomte, C. Grumet, J. Sannier, Fuel clean-up system. Poisoning of palladiumsilver membranes by gaseous impurities, Fusion Technol. 14 (1988) 614618. [11] A.B. Antoniazzi, A.A. Haasz, P.C. Stangeby, Effect of adsorbed carbon and sulphur on hydrogen permeation through palladium, J. Nucl. Mater. 162164 (1989) 10651070. [12] H. Amandusson, L.-G. Ekedahl, H. Dannetun, The effect of CO and O2 on hydrogen permeation through a palladium membrane, Appl. Surf. Sci. 153 (2000) 259267. [13] K. Hou, R. Hughes, The effect of external mass transfer, competitive adsorption and coking on hydrogen permeation through thin Pd/Ag membranes, J. Membr. Sci. 206 (2002) 119130. [14] D. Wang, T.B. Flanagan, L. Kirk, Shanahan, Permeation of hydrogen through preoxidized Pd membranes in the presence and absence of CO, J. Alloys Compd. 372 (2004) 158164.

[15] F.C. Gielens, R.J.J. Knibbeler, P.F.J. Duysinx, H.D. Tong, M.A.G. Vorstman, J.T.F. Keurentjes, Inuence of steam and carbon dioxide on the hydrogen ux through thin Pd/Ag and Pd membranes, J. Membr. Sci. 279 (2006) 176185. [16] G. Barbieri, F. Scura, F. Lentini, G. De Luca, E. Drioli, A novel model equation for the permeation of hydrogen in mixture with carbon monoxide through PdAg membranes, Sep. Purif. Technol. 61 (2008) 217224. [17] T.H. Nguyen, S. Mori, M. Suzuki, Hydrogen permeance and the effect of H2 O and CO on the permeability of Pd0.75Ag0.25 membranes under gas-driven permeation and plasma-driven permeation, Chem. Eng. J. 155 (2009) 5561. [18] T.L. Ward, T. Dao, Model of hydrogen permeation behaviour in palladium membranes, J. Membr. Sci. 153 (1999) 211231. [19] A. Sieverts, W. Danz, Solubilities of D2 and H2 in palladium, Z. Physik. Chem. (B) 34 (1936) 158159. [20] R.E. Buxbaum, A.B. Kinney, Hydrogen transport through tubular membranes of palladium-coated tantalum and niobium, Ind. Eng. Chem. Res. 35 (1996) 530537. [21] A. Caravella, G. Barbieri, E. Drioli, Modelling and simulation of hydrogen permeation through supported Pd-alloy membranes with a multicomponent approach, Chem. Eng. Sci. 63 (2008) 21492160. [22] J. Catalano, M. Giacinti Baschetti, G.C. Sarti, Inuence of the gas phase resistance on hydrogen ux through thin palladiumsilver membranes, J. Membr. Sci. 339 (2009) 5767. [23] M. Johansson, O. Lytken, I. Chorkendorff, The sticking probability for H2 in presence of CO on some transition metals at a hydrogen pressure of 1 bar, Surf. Sci. 602 (2008) 18631870. [24] D.A. King, M.G. Wells, Reaction mechanism in chemisorption kinetics: nitrogen on the {1 0 0} plane of tungsten, Proc. Roy. Soc. London 339 (1974) 245269. [25] O. Dulaurent, K. Chandes, C. Bouly, D. Bianchi, Heat of adsorption of carbon monoxide on a Pd/Al2 O3 solid using in situ infrared spectroscopy at high temperatures, J. Catal. 188 (1999) 237251. [26] D.R. Alfonso, Comparative studies of CO and H2 O interactions with Pd(1 1 1) surface: a theoretical study of poisoning, Appl. Phys. Lett. 88 (2006), 051908051908-3. [27] E. Ozensoy, B.K. Min, A.K. Santra, D.W. Goodman, CO dissociation at elevated pressures on supported Pd nanoclusters, J. Phys. Chem. B 108 (2004) 43514357. [28] C.-Y. Huang, H.-J. Chiang, J.-C. Huang, S.-R. Sheen, Synthesis of nanocrystalline AgPd alloys by chemical reduction method, Nanostruct. Mater. 10 (1998) 13931400. [29] R.J. Borg, G.J. Dienes, An Introduction to Solid State Diffusion, Academic Press, San Diego, 1988. [30] C. Wert, C. Zener, Interstitial atomic diffusion coefcients, Phys. Rev. 76 (1949) 11691175. [31] H. Zchner, Untersuchung der Diffusion of Wasserstoff in Pd- und Pd/AgLegierungen mit einer Stromsto-Methode, Z. Naturforsch. Pt. A 25 (1970) 14901496. [32] J. Behm, K. Christmann, G. Ertl, Adsorption of hydrogen on Pd(1 0 0), Surf. Sci. 99 (1980) 320340. [33] A. Bhargav, G.S. Jackson, Thermokinetic modeling and parameter estimation for hydrogen permeation through Pd0.77Ag0.23 membranes, Int. J. Hydrogen Energy 34 (2009) 51645173. [34] E. Serra, M. Kemali, A. Perujo, D.K. Ross, Hydrogen, Deuterium in Pd-25 Pct Ag alloy: permeation, diffusion, solubilization and surface reaction, Metall. Mater. Trans. A 29A (1998) 10231028. [35] S. Tosti, A. Basile, G. Chiappetta, C. Rizzello, V. Violante, PdAg membrane reactors for water gas shift reaction, Chem. Eng. J. 93 (2003) 2330. [36] H. Yoshida, S. Konishi, Y. Naruse, Effects of impurities on hydrogen permeability through palladium alloy membrane at comparatively high pressure and temperature, J. Less-Common Met. 89 (1983) 429436. [37] D. Artman, T.B. Flanagan, Desorption of hydrogen from palladium and palladiumsilver alloys followed by differential scanning calorimetry, Can. J. Chem. 50 (1972) 13211324. [38] O.M. Lovvik, R.A. Olsen, Density functional calculations of hydrogen adsorption on palladiumsilver alloy surfaces, J. Chem. Phys. 118 (2003) 32683276. [39] M. Myyrylainen, T.T. Rantala, Kinetic Monte Carlo modeling of CO desorption and adsorption on Pd(1 1 0) surface, Catal. Today 100 (2005) 413417. [40] D.M. Ruthven, Principles of Adsorption and Adsorption Processes, John Wiley and Sons, New York, 1984. [41] T.M. Duncan, K.W. Zilm, D.M. Hamilton, T.W. Root, Adsorbed states of carbon monoxide on dispersed metals: a high-resolution solid-state NMR study, J. Phys. Chem. 93 (1989) 25832590. [42] E.H. Voogt, L. Coulier, O.L.J. Gijzeman, J.W. Geus, Adsoprtion of carbon monoxide on Pd(1 1 1) and palladium model catalysts, J. Catal. 169 (1997) 359364. [43] K. Christmann, G. Ertl, Adsorption of carbon monoxide on silver/palladium alloys, Surf. Sci. 33 (1972) 254270.

You might also like