Energy

Procedia

Energy Procedia 00 (2010) 000000
www.elsevier.com/locate/XXX

GHGT-10
2-Dimensional membrane separator modelling:
mass transfer by convection and diffusion
Jurriaan Boon
1*
, Hui Li, Jan Wilco Dijkstra, Johannis A.Z. Pieterse
Hydrogen & Clean Fossil Fuels, Energy research Centre of the Netherlands
P.O. Box 1, NL1755 ZG Petten, The Netherlands
Elsevier use only: Received date here; revised date here; accepted date here
Abstract
Hydrogen selective membranes may present a technologically and economically efficient method for the separation of H
2
from CO
2

in pre-combustion decarbonisation of power production from fossil fuels. Accurate scale-up and performance prediction of
membranes strongly depends on adequate representation of the prevailing resistances to mass transfer, especially for present-day high
flux membranes. In a series of experiments, H
2
/N
2
separation is measured as a function of feed flow and retentate pressure for a
supported palladium membrane enclosed by annular channels for feed/retentante and sweep/permeate flow. Comparison of model
predictions with measured data reveals that mass transfer resistances in the gas phase are significantly reduced by a radial velocity
component in cases of high transmembrane flux, which can only be adequately described by a 2D model. For accurate interpretation
of experiments, scale-up, and design of modules with high flux membranes, 2D modelling is required.
2010 Elsevier Ltd. All rights reserved
Keywords: membrane separation; gas permeation; mass transfer resistance; 2D modelling; Pd membrane
1. Introduction
In precombustion decarbonisation for power production, CO
2
is extracted from the fuel before combustion. Initially,
fuel is converted to synthesis gas: mainly CO and H
2
. Water-gas shift (WGS) further converts this into a mixture
consisting largely of CO
2
and H
2
. A suitable separation technique then yields two streams: a stream of concentrated CO
2

and a stream of H
2
the latter can be used for power generation without concomitant CO
2
emissions. Several techniques
are available for separating H
2
from CO
2
, yet H
2
-selective membranes may offer an advantage in terms of economic and
energy efficiency [1, 2]. Further efficiency benefits may arise if WGS reaction and H
2
separation are integrated at
elevated temperatures in a separation-enhanced WGS reactor [3-5]. Similarly, the fuel reforming reaction, WGS, and H
2

separation may be integrated in a membrane reformer [6]. In all of these cases minimisation of equipment size and
required membrane surface area is crucial to the economic feasibility and this can be realised by the use of high
performance membranes. Therefore development efforts are directed towards achieving a high hydrogen
transmembrane flux. Unfortunately, with higher fluxes the resistances to heat and mass transfer will also become
increasingly important. Therefore, optimal design and operation of membrane units necessitates a fundamental
understanding of the prevailing heat and mass transfer processes [7, 8]. This work describes a series of experiments of
H
2
/N
2
separation with a supported palladium membrane enclosed by a double annular geometry. For interpretation of
the results and studying the effect of mass transfer resistance, 1D and 2D axially symmetric membrane separator models
are developed. The crucial differences between 1D and 2D modelling will be highlighted.

* Corresponding author. Tel.: +31-224-564576; fax: +31-224-568489.
E-mail address: boon@ecn.nl.
c 2011 Published by Elsevier Ltd.
Energy Procedia 4 (2011) 699706
www.elsevier.com/locate/procedia
doi:10.1016/j.egypro.2011.01.108
Jurriaan Boon et al. / Energy Procedia 00 (2010) 000000 2
2. Experimental
Experiments on H
2
/N
2
separation have been performed on ECNs Process Development Unit (PDU), described in
more detail elsewhere [9]. Three membranes are used, each consisting of a commercial 14 mm outer diameter -Al
2
O
3

macroporous membrane support tube, two mesoporous -Al
2
O
3
layers of 45 m each [10], and a 5.6-6.1 m Pd layer
[11]. After sealing [12], the effective length is 44 cm, giving a total surface area for three membranes of 5.810
-2
m
2
.
Membrane tubes are mounted in modules (Figure 1); an insert tube is used for the sweep gas, creating a double annulus
geometry enclosing the membrane. The modules are placed in an electrically heated oven at 400C, with a maximum
temperature gradient over the module length of 25C. For the three modules combined, feed gases were prepared by
mass flow controllers, outlet flows and composition were determined with a mass flow meter and gas chromatograph,
respectively. The experimental results reported here were all obtained with 60.2% H
2
in N
2
feed gas and pure N
2
sweep
gas. The feed flow rate was varied in the range of 30-90 Nlmin
-1
and the countercurrent sweep flow rate was 20 Nlmin
-
1
. Feed-side pressures were varied in the range of 2.0-3.1 MPa, while sweep side pressure was kept at 1.5 MPa. In these
conditions, the Reynolds number varies up to 400 and hydrodynamics will be laminar at all times.


Figure 1. Schematic membrane module layout


Figure 2. Membrane module annular geometry for feed/retentate and countercurrent sweep/permeate side (r
s
= 3 mm,
r
m
= 7 mm, r
f
= 13.3 mm)
3. Modelling
Modelling is done for the axisymmetric module geometry depicted in Figure 2. Both the feed/retentate side and the
countercurrent sweep/permeate side of the module have an annular shape. For description of fluid flow and mass
transfer, 2D steady state balances are solved for both channels in cylindrical geometry (r, , z coordinates). Under the
assumptions of (i) constant density and temperature, (ii) no angular gradients (/ = 0), (iii) Newtonian fluid of
constant viscosity, (iv) ignoring gravity, (v) no radial pressure gradient, and (vi) constant diffusivity, the steady-state
equations of continuity and motion, and the hydrogen material balance reduce to [13]:
0
1
= |
.
|

\
|
c
c
+
c
c
z
v
rv
r r
z
r
p
continuity (1)

c
c
+ |
.
|

\
|
c
c
c
c
+
c
c
= |
.
|

\
|
c
c
+
c
c
2
2
1
z
v
r
v
r
r r z
p
z
v
v
r
v
v
z z z
z
z
r
u p
motion (2)
|
|
.
|

\
|
c
c
+
|
|
.
|

\
|
c
c
c
c
=
c
c
+
c
c
2
2
2 2
2
2 2
1
z
c
r
c
r
r r
D
z
c
v
r
c
v
H H
H
H
z
H
r
material balance (3)
0 = |
.
|

\
|
c
c
c
c
z
p
r
no radial pressure gradient (4)
The boundary conditions are no-slip at the module walls, except for the membrane flux:
( )
n
s H
n
f H H mem
p p Q J
, ,
2 2 2
=
(5)
700 J. Boon et al. / Energy Procedia 4 (2011) 699706
Jurriaan Boon et al. / Energy Procedia 00 (2010) 000000 3
p
RT
J v
mem
r r
r
m
=
=
(6)
and continuity of hydrogen flux at the membrane surface:
mem
r r
H
H H r
J
r
c
D c v
m
=
|
|
.
|

\
|
c
c

=
2
2 2
(7)
The value of n=0.72 for the pressure exponent in the membrane flux equation (5) is obtained from pure H
2

experiments [14], which are not reported here. At the inlets, developed laminar flow velocity profiles are assumed[13,
2.4]. Four dependent variables v
r
, v
z
, c(H
2
), and p/z are determined by the coupled PDEs (1-4) that are solved as
ODEs in Matlab with built-in solver bvp4c after discretisation in finite differences for z, using second-order upwind
approximation for first-order derivatives and a centered-space approximation of second-order derivatives [15]. After
solving the velocity components and hydrogen concentration in two dimensions, the mixing-cup average flow and
concentration are obtained by quadrature, to yield observable model results such as the outlet flow and hydrogen
concentration for the retentate and permeate. Typical 2D model outcome is shown in Figures 3 and 4. The left part of
Figure 3 shows the distribution of hydrogen on both sides of the membrane, the right part shows the calculated gas
velocity. Figure 4 shows cross-sectional profiles at z = 0.2143 m of the hydrogen concentration (left) and the radial
velocity component (right). Concentration gradients on both sides of the membrane indicate that mass transfer
resistances in the gas phase indeed contribute to the overall mass transfer resistance. A radial velocity component is
created by the transmembrane flux.


Figure 3. 2D model results for the H
2
concentration (left) and velocity (right), feed flow 90 Nlmin
-1
and 3.1 MPa
retentate pressure, membrane located at r = 710
-3
m

J. Boon et al. / Energy Procedia 4 (2011) 699706 701
Jurriaan Boon et al. / Energy Procedia 00 (2010) 000000 4

Figure 4. Cross-sectional profiles at z = 0.2143 m of the H
2
concentration (left) and radial velocity component v
r
(right)
from 2D model results, feed flow 90 Nlmin
-1
and 3.1 MPa retentate pressure

More simple one-dimensional models can be of great importance in design calculations for membrane modules [16,
17]. A 1D plug flow model has been constructed for comparison, and involves material balances for hydrogen on both
sides of the membrane, combined with the membrane permeation equation (5). Suitable relations for mass transfer
resistances (concentration polarisation) on both sides of the membrane relate the flux of hydrogen to/from the
membrane surface to the concentration difference of hydrogen and a mass transfer coefficient according to:
|
.
|

\
|
=
=
m
r r
H H H
c c k J
2 2 2
(8)
Mass transfer coefficients (k) are correlated with the hydraulic diameter (d
h
) and molecular diffusivity (D) as
dimensionless Sherwood number:
D
kd
Sh
h

(9)
Combination of equations (8) and (9) allows for Sh to be obtained from the 2D simulation results. For the 1D model,
Sh-values are needed and these can be readily obtained by using the analogy between heat and mass transfer [13] and
finding literature values of Nu for heat transfer in laminar flow. These are available for developed velocity and
concentration profiles (Nu

, Sh

) as a function of the ratio of annular radii [18]. Experimental values for developing
temperature (concentration) profiles are also available [19] and these have been approximated as:
b
h h
b
h b
z
d
D
d v
a
z
d
Sc Re a Gz a Sh
|
|
.
|

\
|
|
|
.
|

\
|
|
|
.
|

\
|
= |
.
|

\
|
= =
p
u
u
p
(10)

s = =
> =

62 18 . 6
62 07 . 1
42 . 0
Gz Sh Sh
Gz Gz Sh
feed/retentate (membrane at the inner wall) (11)

s = =
> =

51 04 . 5
51 88 . 0
44 . 0
Gz Sh Sh
Gz Gz Sh
sweep/permeate (membrane at the outer wall) (12)
Thus, the current work is based upon three separate models: (i) a full 2D model, (ii) a simplified 1D model with
constant Sh, and (iii) a more extended 1D model with Sh = Sh(Gz).
4. Results
Experimental results with three membrane modules for a feed flow rate of 30-90 Nl.min
-1
and a feed/retentate side
pressure of 2-3.1 MPa are depicted in Figure 5, together with model results. The permeance has been determined as
Q = 1.810
-5
molm
-2
s
-1
Pa
-0.72
by fitting of the pressure variation data using the 2D model. With this value of the
permeance, it can be seen that all models can qualitatively describe the experimental results. But for the higher
recoveries, the 1D models underpredict the recovery. This is confirmed by a parity plot of the predicted membrane flux
versus the experimentally measured flux as depicted in Figure 6. The 1D models deviate more than the 2D model at
higher flux, i.e. J
mem
0.23 molm
-2
s
-1
. It can also be concluded from the results that there is no significant difference
between the 1D model with constant Sh, and with Sh(Gz) and therefore that the effect of developing concentration
profiles is sufficiently small to ignore.
702 J. Boon et al. / Energy Procedia 4 (2011) 699706
Jurriaan Boon et al. / Energy Procedia 00 (2010) 000000 5
In order to further compare the models, Sh values were calculated from the 2D model as described above for feed
flow 90 Nlmin
-1
and 3.1 MPa retentate pressure, with a measured recovery of 40% and an average membrane flux of
0.27 molm
-2
s
-1
. An additional computational run was done in which the developed laminar velocity profile was not
disturbed by the membrane flux, i.e. setting v
r
= 0 in equation (6). Results are compared with the literature Sh values
from equations (11) and (12) in Figure 7. The Sh values in 2D simulation are up to about a factor 2 larger than literature
Sh values. The difference appears to be due to the radial velocity component created by the transmembrane flux: when
the radial velocity is set to zero the difference disappears.


Figure 5. Hydrogen recovery versus retentate pressure (for feed flow 90 Nlmin
-1
) and feed flow (for 3.1 MPa retentate
pressure): measured (), obtained with the 2D model (), and obtained with the 1D model with constant () and
varying () Sh


Figure 6. Predicted versus measured values of the length-averaged membrane flux for the 2D model (), and the 1D
model with constant () and varying () Sh

J. Boon et al. / Energy Procedia 4 (2011) 699706 703
Jurriaan Boon et al. / Energy Procedia 00 (2010) 000000 6

Figure 7. Plots of local Sh values for 2D simulation (), 2D simulation with v
r
= 0 (--), and from literature ()
5. Discussion
In the current configuration, two annular channels with laminar flow are separated by a H
2
-selective membrane. The
difference in concentration on both sides creates a driving force for H
2
transport from the feed/retentate to the
sweep/permeate side. The flux of H
2
is subject to three separate mass transfer resistances in series: diffusion on the feed
side, membrane permeation, and diffusion on the sweep side. The diffusion processes in the annular channels can be
described with the 2D model presented above, but they can also be approximated with mass transfer coefficients and Sh
correlations, allowing for more simple 1D models. However, the 1D models increasingly deviate from experiment at
higher membrane fluxes, in particular for J
mem
0.23 molm
-2
s
-1
. Comparison of Sh values for the 1D and the 2D
simulations indeed reveals a large difference in mass transfer coefficients that originates from the assumptions of Nu
correlations in literature, in particular the assumption of developed laminar flow. This assumption ignores the radial
velocity component that is indeed absent in thermal conduction problems, but is created here by permeating H
2
.
Radial convection can significantly add to the transport to/from the membrane surface by convective transport of H
2
. In
contrast, the effect of developing concentration profile near the inlet on both sides of the membrane may be safely
ignored. Because mass transfer resistances are in series, the effect that the radial velocity component increases the Sh
value is only observed with sufficiently high membrane flux. With the development of high flux membranes, 2D
modelling will become ever more important to allow for accurate estimation of the resistances to mass transfer. For
module design and scale-up calculations, the use of 1D models is restricted to a relatively low transmembrane flux and
will otherwise predict too low hydrogen flux and accordingly too high a required membrane surface.
6. Conclusion
Experimental results obtained on H
2
/N
2
separation experiments with high flux membranes cannot be described
accurately with 1D models, including literature based Sh correlations for mass transfer resistances on both sides of the
membrane. Simulations with a 2D model reveal that the high membrane flux creates a radial velocity component that
significantly decreases the transport resistances to/from the membrane. The distorted velocity profiles invalidate the Sh
correlations that can be derived from literature data on Nu for heat transfer in developed laminar flow in identical
geometry. The error that is made in 1D models becomes increasingly large for increasing membrane flux and in this
case a 2D separator model is required for accurate scale-up and design calculations.
Acknowledgement
This work has been performed as part of the Dutch national R&D programme for CO
2
capture, transport and storage
CATO-2. The authors gratefully acknowledge Ir. Peter Veenstra and Prof. Dr. Arian Nijmeijer of consortium partner
Shell Global Solutions, Amsterdam (NL) for fruitful discussions. Mr. Sander de Munck of ECN Hydrogen & Clean
Fossil Fuels is acknowledged for performing the experiments.
704 J. Boon et al. / Energy Procedia 4 (2011) 699706
Jurriaan Boon et al. / Energy Procedia 00 (2010) 000000 7
Nomenclature
c
i
concentration if i, molm
-3

<c
i
> mixing-cup average concentration of i, molm
-3

d
h
hydraulic diameter, m
D
i
diffusivity of i in gas mixture, m
2
s
-1

Gz dimensionless Graetz number (equation 10)
J
mem
transmembrane flux, molm
-2
s
-1

k mass transfer coefficient, ms
-1

N normal gas volume, specified at 273 K and 1 atm
p total pressure, Pa
p
i
partial pressure of i, Pa
Q
i
permeance of i, molm
-2
s
-1
Pa
-n

r radial coordinate, m
R gas constant, Jmol
-1
K
-1

Re dimensionless Reynolds number (equation 10)
r
f
outer radius of feed/retentate side channel, m
r
m
membrane radius, m
r
s
inner radius of sweep/permeate side channel, m
Sc dimensionless Schmidt number (equation 10)
Sh dimensionless Sherwood number (equation 9)
Sh

dimensionless Sherwood number for fully developed flow profile and concentration profile
T temperature, K
v gas velocity vector, composed of (v
r
,v
z
), ms
-1

z axial coordinate, m
angular coordinate,
dynamic viscosity, Pas
gas-phase density, kgm
-3

List of References
[1] Ockwig NW, Nenoff TM. Membranes for Hydrogen Separation. Chemical Reviews 2007; 107(10): 4078-110.
[2] Lu GQ, niz da Costa JC, Duke M, Giessler S, Socolow R, Williams RH, et al. Inorganic membranes for
hydrogen production and purification: A critical review and perspective. Journal of Colloid and Interface
Science 2007;314(2):589-603.
[3] Bracht M, Alderliesten PT, Kloster R, Pruschek R, Haupt G, Xue E, et al. Water gas shift membrane reactor
for CO2 control in IGCC systems: techno-economic feasibility study. Energy Conversion and Management
1997;38(Supplement 1):S159-64.
[4] Bredesen R, Jordal K, Bolland O. High-temperature membranes in power generation with CO2 capture.
Chemical Engineering and Processing 2004;43(9):1129-58.
[5] Kikuchi E, Uemiya S, Sato N, Inoue H, Ando H, Matsuda T. Membrane Reactor Using Microporous Glass-
supported Thin Film of Palladium. Application to the Water Gas Shift Reaction. Chemistry Letters
1989;18(3):489-92.
[6] Pieterse JAZ, Boon J, Van Delft YC, Dijkstra JW, Van den Brink RW. On the potential of nickel catalysts for
steam reforming in membrane reactors. Catalysis Today 2010; In Press, Corrected Proof.
[7] De Falco M, Di Paola L, Marrelli L, Nardella P. Simulation of large-scale membrane reformers by a two-
dimensional model. Chemical Engineering Journal 2007;128(2-3):115-25.
[8] De Falco M. Pd-based membrane steam reformers: A simulation study of reactor performance. International
Journal of Hydrogen Energy 2008;33(12):3036-40.
[9] Dijkstra JW. Development of Membrane Reactor Technology for Power Production with Pre-Combustion CO
2

Capture. 19-23 Sep 2010; 10
th
International Conference on Greenhouse Gas Control Technologies, GHGT10;
2010.
[10] Bonekamp BC, van Horssen A, Correia LA, Vente JF, Haije WG. Macroporous support coatings for molecular
separation membranes having a minimum defect density. Journal of Membrane Science 2006;278(1-2):349-56.
[11] Hou S, Jiang K, Li W, Xu H, Yuan L, inventors; A metal palladium composite membrane or alloy palladium
composite membrane and their preparation methods.WO2005065806. 2005.
[12] Rusting FT, de Jong G, Pex PPAC, Peters JAJ, inventors; Sealing socket and method for arranging a sealing
socket to a tube.WO0163162. 2001.
[13] Bird RB, Stewart WE, Lightfoot EN. Transport Phenomena. New York etc.: Wiley; 1960.
J. Boon et al. / Energy Procedia 4 (2011) 699706 705
Jurriaan Boon et al. / Energy Procedia 00 (2010) 000000 8
[14] Li H, Dijkstra JW, Pieterse JAZ, Boon J, Van den Brink RW, Jansen D. Towards full scale demonstration of
hydrogen-selective membranes for CO
2
capture. Journal of Membrane Science 2010; In Press.
[15] Hoffman JD. Numerical Methods for Engineers and Scientists. 2
nd
edn. Boca Raton: Taylor & Francis; 2001.
[16] Hou K, Hughes R. The effect of external mass transfer, competitive adsorption and coking on hydrogen
permeation through thin Pd/Ag membranes. Journal of Membrane Science 2002;206(1-2):119-30.
[17] Koukou MK, Papayannakos N, Markatos NC, Bracht M, Alderliesten PT. Simulation tools for the design of
industrial-scale membrane reactors. Chemical Engineering Research & Design 1998;76(8A):911-20.
[18] Ebadian MA, Dong ZF. Forced convection, internal flow in ducts. In: Rohsenow WM, Hartnett JP, Cho YI,
editors. Handbook of Heat Transfer. 3 edn. New York: McGraw-Hill; 1998. p. 5.1-5.137.
[19] Heaton HS, Reynolds WC, Kays WM. Heat transfer in annular passages. Simultaneous development of
velocity and temperature fields in laminar flow. International Journal of Heat and Mass Transfer
1964;7(7):763-81.
706 J. Boon et al. / Energy Procedia 4 (2011) 699706