Chemical

Engineering Science
PERGAMON Chemical Engineering Science 54 (1999) 1431-1439

Factors Controlling the Performance of

Catalytic Dehydrogenation of Ethylbenzene
in Palladium Composite Membrane Reactors
R. Dittmeyerl*, V. H611ein/ P. Quicker2, G. Emig2, G. Hausinger3 , F. Schmide
1 DECHEMA e.V., Karl-Winnacker-Institut, Theodor-Heuss-Allee 25, D-60486 Frankfurt am Main
2 Lehrstuhl fur Technische Chemie I, Universitat Erlangen-Niirnberg, Egerlandstr. 3, D-91058 Erlangen
3 Siid-Chemie AG, Research & Development, Waldheimer Str. 13, D-83052 Bruckmiihl-Heufeld

Abstract - Ethylbenzene dehydrogenation to styrene in a palladium composite membrane reactor was investigated.
The reaction kinetics have been studied in a gradientless recycle reactor to develop a suitable kinetic model, which
was subsequently used for the simulation of a hypothetical industrial-scale packed-bed membrane reactor. For this
simulation, the measured hydrogen permeability of palladium composite membranes, prepared by different methods
on asymmetric porous ceramic and porous sinter metal supports, was used. The results demonstrate that the
performance of the membrane reactor is controlled both by the membrane permeability and by the reaction kinetics,
i.e. 4 - 27 % savings of the ethylbenzene feed at equal styrene output are predicted for industrially relevant
operating conditions. Besides savings of raw materials, a second advantage of the membrane reactor is seen in a
reduced ethylbenzene load of the ethylbenzene/styrene fractionation column, thanks to the increased conversion.
Moreover, a simplified heat management seems to be possible by utilizing the combustion of the permeated
hydrogen to supply the heat required for the dehydrogenation. Concerning membrane permeability, the simulation
demonstrates a noticeable contribution of the support to the overall hydrogen transport resistance when going for a
tube size suitable for an industrial-scale reactor. Hence, care has to be taken not only of the permeability of the
hydrogen permselective layer, but also of the thickness, porosity, and mean pore size of the support. All in all, the
simulation results show that a packed-bed multitubular membrane reactor in fact offers the potential for substantially
increasing the styrene yield in ethylbenzene dehydrogenation. What is needed are both high hydrogen permeability
and high catalyst efficiency. However, a large-scale application obviously means a very big challenge not only in
respect of the manufacture of highly permeable and resistive membranes but also in terms of maximizing catalyst
efficiency and developing a suitable reactor design. © 1999 Elsevier Science Ltd. All rights reserved.

Keywords: Dehydrogenation, Membrane reactor, Palladium composite membrane, Simulation study.

INTRODUCTION reactions and various reactor configurations. Both

Combining reaction with separation into mUltifunctional experimental investigations and simulation studies have
reactors becomes more and more an interesting concept been carried out. A recent survey has been given for
to both academic and industrial research. Among various example by Hsieh, 1996. Our work is focused on the
potential advantages of such concepts above all the use of membranes for hydrocarbon dehydrogenation,
perspective to circumvent existing thermodynamic limits i.e. ethylbenzene to styrene and propane to propene,
on the yield of the desired reaction products certainly is while trying to keep close to the operating conditions of
the main incentive. This is also the central motivation for the industrial processes. Metal composite membranes,
applying membrane reactors to hydrocarbon made by deposition of thin metallic layers on a
dehydrogenation. Given the removal of hydrogen persistent porous sinter metal support, appear most
produced by dehydrogenation, the reaction can be attractive for this application due to their superior
continued beyond equilibrium conversion. However, the permeability and permselectivity. For coating these
use under typically harsh conditions of industrial supports, palladium and palladium alloys are well
dehydrogenation processes not only calls for high established metals with excellent hydrogen
permeability and high selectivity of the membranes permeability (Gryaznov and Orekhova, 1998). Other,
employed, but also for sufficient thermal, mechanical, less costly metals that might be used include group IVa
and chemical stability. Successfully linking membranes metals Ti and Zr as well as group Va metals V, Nb, and
to conventional chemical reactor design is not a trivial Ta (Buxbaum and Kinney, 1996). Besides preparation
task either. In particular, gas-tight sealing of ceramic and testing of the membranes, modeling of the relevant
membranes to metallic reactor parts is difficult. heat and mass transport steps occurring in a palladium
composite membrane reactor along with the chemical
The work published in the field of membrane reactors for reactions is part of our research. This paper is focused
dehydrogenation of hydrocarbons covers all common on the results of a simulation study carried out to

* Corresponding author

0009-2509/99/$-see front matter © 1999 Elsevier Science Ltd. All rights reserved.
PH· !,:OOOQ-?"iOQ(QQ)0004R-2
1432 R. Dittmeyer et al.lChemical Engineering Science 54 (1999) /431-1439

identify crucial parameters that are likely to control the permselective membranes which are able to remove the
performance of an industrial-scale membrane reactor for produced hydrogen from the catalyst bed.
dehydrogenation of ethylbenzene.
With reference to hydrogen removal by membranes,
most of the experimental studies deal with ceramic
DEHYDROGENATION OF ETHYLBENZENE ultrafiltration membranes. Wu et al. (1990) could
Ethylbenzene dehydrogenation (EBDH) is a well- improve ethylbenzene conversion from 48 % in a
established industrial reaction with more than 60 years of conventional packed-bed reactor to 65 % at 640°C by
development in the past, and a worldwide styrene using ceramic membranes of 4 nm pore size. At the
production today of more than 15 million tons per year same time, a slight increase of styrene selectivity was
(Kochloefl, 1997). Figure 1 shows a simplified reaction observed. It should be mentioned that the experiments
scheme. Besides the main products styrene and were performed with a feed diluted with nitrogen.
hydrogen, toluene and benzene are formed as side Various membrane reactor configurations for porous
alumina tubes were investigated by Yang et al. (1995),
+~ who reported an absolute increase of styrene yield of
11 % at 620°C in the membrane reactor compared to
the conventional packed-bed. Tiscareno-Lechuga and
Hill (1993) were able to reach the equilibrium
conversion of ethylbenzene by employing a porous
membrane of 5 nm pore size. However, they faced
tremendous problems with coking and the low
permselectivity of the membrane employed. Again, the
feed has been diluted by nitrogen. A hybrid
configuration with a conventional packed-bed reactor
followed by a membrane reactor (4 nm pore size) was
investigated by Gallaher et al. (1993), who achieved an
Figure 1. Simplified reaction scheme of ethylbenzene
absolute increase of styrene yield of 7 %. The
dehydrogenation on a K2COJiFe203-catalyst.
experiment was performed at LHSV =0.4 h· l , a
temperature of 575°C, and a feed composition
products accompanied by methane, C2-hydrocarbons, N2:H20:EB of 9:9:1. Quicker et al. (1998) observed in
and carbon oxides which are due to oxidation of e.g. a palladium composite membrane reactor at 580°C an
carbon and tar deposits by steam. Steam is added in large increase of styrene yield by 14 %, compared to a
excess to suppress the reverse reaction, to stabilize the conventional packed-bed reactor. In a patent, Bitter
active phase of the catalyst, and to remove the deposits (1986) claimed the use of porous ceramic membranes
by oxidation (Kochloefl, 1997). The operating conditions for dehydrogenation of ethylbenzene and other
of conventional EBDH-reactors are characterized by hydrocarbons. At 625°C and 4 bar an improvement of
temperatures of 540 - 650°C, normal or subatmospheric ethylbenzene conversion from 50.7 % in a packed-bed
pressure, a liquid hourly space velocity (LHSV) of 0.5 - to 65.2 % in the membrane reactor is reported at 94 %
1.2 h·\ and a molar steam to oil ratio (S/O) of 6 - 12 selectivity to styrene.
depending on the process conditions (,isothermal' or Simulation studies on the application of porous ceramic
adiabatic). Commercial adiabatic units consist of two or membrane reactors for dehydrogenation of
three reactors in series. Parallel reactors are used for the ethylbenzene have been carried out by Wu and Liu
less common 'isothermal' process. The conversion (1992) and Becker et al. (1993). Abdalla and Elnashaie
reaches 60 - 70%, while the selectivity to styrene is (1994) and Hermann et al. (1997a, 1997b) reported
around 95 % (James and Castor, 1994; Kochloefl, 1997). simulation results on the use of palladium composite
membranes to ethylbenzene dehydrogenation.
New developments
Given the fact that the potential for styrene yield
improvement by modification of the catalyst appears KINETICS OF ETHYLBENZENE
more or less limited, considerable effort is being put into DEHYDROGENATION
the improvement of the process design. The ultimate Detailed kinetic studies were carried out for a
goal is to overcome the thermodynamic limit on conventional styrene catalyst provided by Siid-Chemie
conversion which would significantly bring down the AG using a BERTY-type gradientiess recycle reactor,
fractionation cost. This can be achieved for example by specially designed for normal pressure and high
selectively oxidizing the produced hydrogen using an temperatures up to 650°C. The reactor is integrated
additional noble metal catalyst after the first EBDH- into a laboratory plant comprising several high-
reactor in the SMART Process (UOP/Lummus). An precision peristaltic pumps for dosage of organic
alternative, not yet commercially established, is to use liquids (ethylbenzene, styrene, benzene, toluene), a
oxidative dehydrogenation, either in co-feed mode of HPLC-pump and an evaporator unit to provide the
ethylbenzene and oxygen or by successive reaction and necessary steam, mixing devices, thermal mass flow
regeneration cycles, i.e. reacting ethylbenzene with the controllers for dosage of various gases (nitrogen,
oxidized catalyst in a first step and reoxidizing the hydrogen, carbon dioxide), a condenser linked to a
reduced catalyst with oxygen in the second step. Finally, cryostat to separate the liquids from the product stream,
a third possibility is seen in the application of hydrogen an organic/aqueous phase splitter, and a gas
 R. Dittmeyer et al.lChemicai Engineering Science 54 (1999) /431-1439 1433

chromatograph for analysis of the composition of gas balances. Hence, the 'faulty' measured reactor
and liquid products. Details of the experimental setup concentrations are not used as independent variables,
will be reported elsewhere (Quicker et aI., 1999). which, strictly speaking, would be in conflict with the
common assumption of error-free observations of the
Modeling strategy: independent variables in classical regression.
Initial results obtained in the recycle reactor showed a It should be stressed however, that this approach relies
significant influence of thermal reactions as a matter of on the assumption that the measurement error of the
the high gas volume. The recycle reactor has a total operating variables appearing in the CSTR mass
volume of 554 cm 3 and a catalyst charge of 0.5 - 20 g balance (cF, T, p, ry is in fact small compared to the
(typically 5 g), whereas the amount of catalyst per unit of error of the measured reactor concentrations c;. A
reactor volume is roughly 1 glcm 3 for a conventional second advantage is that a possible change of volume
packed-bed reactor. Hence, the contribution of thermal due to the reaction is described more accurately.
reactions in the recycle reactor is enhanced roughly by a Regression methods implemented in SiAMoD include
factor of 25 - 110 compared to the packed-bed, BFGS, conjugated gradients, and a Gauss-Newton
depending on the amount of catalyst used. Despite this method with step-size control. Moreover, a special
problem, the recycle reactor has several advantages, i.e. feature of the program is the estimation of parameter
an improved temperature control, constant partial confidence intervals by an error back-propagation
pressure of the reacting species within the entire reactor method that makes use of generalized SVD (singular
volume, and an easy evaluation of the measured kinetic value decomposition).
data. The kinetic models were compared in terms of the
A stepwise strategy was applied to determine the weighted residual error S, defined in Eq. (1), as a global
stationary kinetic properties of the catalyst, assuming indication of the quality of the fit (Oij denotes the
that a simple superposition of thermal and catalytic estimated measurement error of variable Y;j). In
reactions is permissible. In other words, no interaction addition, the single-component sums of the weighted
between gas phase and surface reactions other than residual error s;, and the significance of the parameter
through the measured gas phase species may occur, i.e. estimates, expressed by the approximate confidence
intermediates such as adsorbed species and gas phase intervals of the transformed parameters, are used to
radicals are not explicitly considered. Given this judge the quality of the models (Zander, 1999). The
assumption, the thermal reaction system was analyzed general guideline was to keep the models as simple as
thoroughly in advance, before the catalytic reactions possible in order to promote their use for practical
were studied. Both, for thermal and catalytic reactions, purposes.
the experimental design included variations of
temperature, pressure, and residence time, as well as the
addition of products, e.g. styrene, benzene, toluene, S =2>; =II-1 (
n n m sUn -
y;)
yex
;}
p
1 (1)
hydrogen, and carbon dioxide (catalytic reaction only). i=1 i=1 }=1 2 (iu
Moreover, experiments with different type and amount
of inert filling material as well as runs with varying
catalyst charge and particle size were carried out. The Thermal reaction system:
results provided neither evidence of catalytic effects of Model discrimination finally yielded three candidates
the filling material, nor indications of diffusion of different complexity. Model TM1 is the most
limitation using catalyst particles of 0.8 - 1.0 mm grain detailed model. It contains 8 reactions and all 11
size, which were obtained from crushing and sieving of reacting species mentioned in Figure 1 except 'coke'.
the industrial-size pellets. TM2 is obtained by lumping the carbon oxides to CO 2,
A total of 92 experiments in absence of catalyst and 65 motivated by a large molar excess of CO 2 over CO
experiments in presence of different amount of catalyst observed in the range of 10 - 20. TM3 (cf. Table 1) is
were carried out in the recycle reactor. In addition, a the result of further lumping ethylene and ethane to
number of kinetic tests were performed in a plug-flow C2H4, justified again by a remarkable excess of
tubular reactor on catalyst particles of 0.8 - 1.0 mm size. ethylene over ethane found in the experiments. Despite
The results, together with literature information (Wenner its simpliCity, model TM3 shows a good fit of the
and Dybdal, 1948; Carra and Forni, 1965; Lebedev et aI.,
1977; Hirano, 1986; Bhat, 1988; Abdalla et aI., 1994), Table 1. Thermal reaction model TM3.
served to formulate various kinetic models. These were
subsequently fitted to the experimental data using the Reaction equation Kinetic expression
regression-code SiAMoD, developed together with
PSTPH z J
Siemens AG, Miinchen (Zander, 1999). The program can EB <=> ST+H2 'I =kl · ( PEB-K;-
handle static and dynamic optimization problems with
EB -+ B+C 2H 4 '2 =k2· PEB
constraints. This feature allows to combine the kinetic
EB+H2 -+ T+CH 4 '3 = k3 · PEB
equations of the various models with the CSTR element
EB+3H2 -+ 4C 2H 4 '4 = k4 · PEB
mass balances that apply to the gradientiess recycle
reactor. The solution of these non-linear equations is C2H 4 +4HP -+ 2C02 +6H 2 '5 =ks· PCzH • . PHzO
obtained by a robust generalized Newton-method. The C2H 4 +2H 2 -+ 2CH 4 '6 =k6 · PCzH.
main advantage is that the kinetic equations are not -E j _(122700_126.3.T_2.194xlO- l .T z)
evaluated using the measured concentrations but the 5
k1.=kO'eRT·K
'p
=e RT x 10 [Pal
concentrations derived from the solution of the mass J
434 R. Dittmeyer et al.lChemical Engineering Science 54 (/999) 1431-1439

experimental data and was therefore selected to represent Based on a hypothetical reaction mechanism, various
the thermal reactions in the subsequent analysis of the kinetic expressions were developed and tested for the
catalytic reaction system (Quicker et aI., 1999). main reaction and the formation of toluene and
benzene, respectively. The details will be reported
Catalytic reaction system: elsewhere (Quicker et aI., 1999). Adsorption of styrene
Analyzing the entire data set showed that 6 catalytic turned out to have an inhibiting effect on the main
reactions are necessary to describe the experimental data reaction, as found already by Lebedev et al. (1977),
obtained from the recycle reactor and the packed-bed although the data obtained from the recycle reactor and
reactor with reasonable accuracy. These are summarized the packed-bed reactor do not provide strong evidence
in Table 2. Within experimental uncertainty benzene of the kinetic importance of this effect, in a sense that a
formation in the recycle reactor proved to be due to the similar model ignoring the inhibition by styrene has
thermal reaction only. However, the data of the only a moderately higher residual error (Quicker et aI.,
additional kinetic tests in a tubular packed-bed reactor at 1999).
high conversion proved an influence of the catalyst on
benzene formation too. The ratio of ethylene to benzene The kinetic expression for the main reaction given in
found in these packed-bed experiments furthermore Table 2 can be interpreted visualizing a dual site
suggested to formulate two catalytic reactions yielding mechanism for the rate-determining surface reaction
benzene, by dealkylation (8) and by reaction of adsorbed where ethylbenzene is adsorbed at site A whereas site B
ethylbenzene with surface hydrogen (9) similar to the serves as a hydrogen acceptor. Site A would be subject
formation of toluene (10). Experiments with addition of to inhibition by styrene adsorption, site B would be
styrene did not provide sufficient evidence to include reversibly blocked by CO 2, Adsorption of all other
consecutive reactions of styrene. The observed increase components is ignored. Table 3 gives the final
of C02-production in presence of the catalyst could be parameter estimates for the thermal and catalytic
explained in different ways. After testing several reactions. For the thermal reactions, frequency factors
alternatives, steam reforming of ethylbenzene as well as and apparent activation energies are listed whereas in
methane on the catalyst surface gave the best fit of the case of the catalytic reactions only the apparent
data. Experiments with controlled addition of CO 2 activation energies and relative rate constants are
showed that the formation of styrene and toluene is given, because the estimates of the frequency factors
suppressed reversibly by CO 2, Although no direct have to be treated as confidential. The relative rate
evidence of a suppression of benzene formation by CO 2 constants were obtained by multiplication of the
could be detected within the experimental limits, CO 2 frequency factors with a typical packed-bed density and
inhibition was introduced for reaction (9) analogous to scaling by the rate constant of the thermal
toluene formation (10). Reversible blocking of active dehydrogenation (1) evaluated at a certain reference
sites by CO 2 has been demonstrated already by Muhler et temperature. Hence, they provide information about the
al. (1992). To quantify the inhibiting effect of CO 2, relative rate of the catalytic reaction compared to the
adsorption of CO 2 on the catalyst was studied in the thermal reaction under typical conditions.
relevant temperature range of 400 - 600 "C using an All estimates are significant in a sense that the
ASAP 2000 instrument (Micromeritics). Assuming a confidence intervals of the transformed parameters are
Langmuir-type adsorption isotherm, the adsorption sufficiently small (Quicker et aI., 1999), i.e. do not
equilibrium constant for CO 2 was calculated from include the origin. The only exception is the formally
measured adsorption/desorption curves at different introduced activation energy of the styrene adsorption
temperatures. The data were interpreted using a formal equilibrium constant, which was set zero for this
Arrhenius statement resulting in the estimates specified reason.
in Table 2.

Table 2. Catalytic reaction model (thermal reactions as of Table 1)

Reaction equation Kinetic expression

( PEB----
PST PH, )
Kp
EB (:) ST+H2 r7 =k7 ·
(1 + KST PST)' (1 + KeD, PeD,)
EB ~ B+C 2H 4 's =ks ' PEB
EB+2H2 ~ B+2CH 4 _~ .
"9- PEB
(1+ Kco,PeD,)
EB+H2 ~ T+CH4 -k
'10-10'
PEB
(1 + KeD, PeD, )
EB +16H 2O ~ SC0 2 + 21H2 '11 =k11 . PEB' PH,D
CH 4 +2H zO ~ CO 2 +4H 2 '12 =k12 . PeH • . PH,D
-E j -15930 -EST
kj =k~.e liT; KeD, =3.45 x 10-3 ·e"R'f[Pa- I ); KST =K~T·e liT; Kp asof Table 1.
 R. Dittmeyer et al.lChemical Engineering Science 54 (1999) 1431-1439 1435

Table 3. Parameter estimates for thermal and catalytic The program supports different process options, i.e. use
reactions. of an inert sweep gas, evacuation of the permeate, and
oxidation of the permeated hydrogen. Isothermal,
Reaction Frequency factor Activation adiabatic, and polytropic operation can be treated by
or rate constant energy introducing a lumped overall transmembrane heat
[mole·m·3s· l pa· X ] [kl/mole] transfer coefficient. Mass transport through the
No.
membrane is described based on the different transport
Thermal 1 1.13.107 220.5
mechanisms: diffusion through the stagnant gas films is
2 2.11,10 10 282.0 treated according to the Stefan-Maxwell equations,
9
3 7.38'10 279.6 while the dusty gas model is used to describe effective
11
4 1.12'10 306.7 bulk and Knudsen diffusion and viscous flow through
5 5.25'10.9 48.8 the macro porous support and the micro porous
intermediate layer. The packed-bed can be described
6 1.07-l0·8 -30.6
either using a pseudo-homogeneous or a heterogeneous
Catalytic 7 814* 158.6 model to account for the influence of external and
8 0.18' 114.2 internal heat and mass transport at the catalyst pellet.
9 6.0* 160.2
10 22.3* 208.0
MEMBRANES AND PERMEABILITY
11 1.29* 170.4 Composite palladium membranes were prepared by
12 676" 37.1 electro less plating on tubular a-alumina microfiltration
4
KST 1.30'10 [Pa· l ] not membranes and by various other methods, e.g.
significant electro less plating and electroplating, PVD, CVD, and
thermal spraying on asymmetric porous sinter metal
* dimensionless rate constant, evaluated at T = Tr• tubes. The experimental work has been described in a
k/ = p"k/Tr)/klTr) for: j=7-10
separate publication (Quicker et aI., 1998). Of these
k/ = p"k/Tr)PHlO/klTr) for: j=11,12 methods electro less plating on ceramic supports so far
gives the best results in terms of membrane flux and
perm selectivity. Single gas separation factors for HYN2
of 250 - 300 have been obtained. The measured
MEMBRANE REACfOR MODEL 7
hydrogen permeability is in the range of 4 - 5xlO·
The membrane reactor model has been described
mole·m- 2·s· I ·Pa· l ; the thickness of the palladium hiyer is
previously (Hermann et aI., 1997a; Hermann et aI.,
around 3 f.lm. A considerably higher hydrogen
1997b). It has been developed for steady state tubular
permeability, i.e. 3.5xlO-6 mole·m- 2·s- I ·Pa-\ has been
membrane reactors consisting of a conventional fixed-
reported by Kikuchi (1998) for similar palladium
bed of catalyst surrounded by an inert hydrogen
composite membranes.
permeable membrane. The location of the catalyst may
be either on the tube side or on the shell side. Plug flow
of retentate and permeate are assumed down the reactor
long axis. Composite membranes consisting of 3 layers
SIMULATION RESULTS
The above described kinetic model, the membrane
can be treated, i.e. macroporous support, microporous
reactor model, and the measured permeability were
intermediate layer, and impervious metal skin-layer
coated either on the tube side or on the shell side. In used together to predict the performance of an
industrial-size membrane reactor. A straightforward
addition, stagnant gas films on both faces of the
concept for an IMCR (inert membrane catalytic
membrane are taken into account (cf. Fig. 3).
reactor) is a multitubular packed-bed reactor, similar to
the design used in the 'isothermal' styrene process.
r This would mean that catalyst pellets are packed into a
number of membrane tubes, while the permeated
hydrogen on the shell side would be either oxidized by
air or removed using inert sweep gas or vacuum.
Important variables of this design are the tube diameter,
the pellet size, the number of tubes, and the tube length.
The pellet size determines the influence of pore
diffusion as well as the pressure drop, while the tube
diameter fixes the available membrane area per unit of
reactor volume. Moreover, a certain ratio of tube
diameter to pellet size, i.e. 10 or higher appears to be
Retentate necessary to allow a feasible loading of the reactor with
z catalyst. Catalyst pellets less than 4 - 5 mm in size
cannot be used under the high throughput requirements
of industrial packed-bed reactors.
Figure 3. Structure of a metal composite membrane as Given the capacity of a world-scale styrene plant of
treated by the membrane reactor model. about 300.000 tons per year, the total amount of
1436 R. Dittmeyer et al.lChemicai Engineering Science 54 (1999) 1431-1439

hydrogen produced would be 360 kmole/h, as a first straight cylindrical pores of the support and the
order estimate. Taking out the hydrogen from the reactor intermediate layer are supposed, i.e. a tortuosity of
via membranes having a permeability of , = 1.0 is assumed for calculation of the effective
180 mole/m 2·h·bar (5x10· 7 mole·m· 2·s· I ·Pa· l ) at an diffusion coefficients. For the time being, the hydrogen
average driving force of 0.1 bar hydrogen partial permeability of the palladium layer is defined based on
pressure difference would obviously require 20000 m2 of the measured hydrogen permeability of the above
membrane area. For the hydrogen permeability reported mentioned composite palladium membranes related to
by Kikuchi (1998), the necessary membrane area would the thickness of the palladium layer. To account for the
be only 2300 m2• The amount of catalyst required may influence of diffusion inside the catalyst pellets, an
be estimated from typical performance data of an overall catalyst effectiveness of 30 % is introduced, as
industrial unit, i.e. conversion of 65 % and styrene estimated for 5 mm pellet diameter based on
selectivity of 95 % at LHSV = 0.5 h· l . From this, the calculations using the heterogeneous model. Moreover,
ethylbenzene feed is calculated to 61.85 t/h. Taking the the reactor shall be isothermal at 600°C and operating
density of ethylbenzene as 0.868 tlm3 we arrive at at 1.2 bar inlet pressure. For the above ethylbenzene
142.5 m3 packed-bed volume. Table 4 shows the number feed of 61.85 t/h and a weight steam-to-oil ratio of 1.2,
of tubes and the wall area (membrane) that would be the calculated outlet pressure is around 0.7 - 0.8 bar.
available if one would place this volume in tubes of 4 m To remain instructive, the discussion is restricted to
length and different diameter. ethylbenzene conversion and styrene selectivity/yield.
The simulation of the conventional packed-bed reactor
Table 4. Tube diameter, number of tubes, and wall area predicts 66.2% conversion and 94.8% selectivity while
of a hypothetical multi tubular packed-bed membrane the performance of the membrane reactor under the
reactor charged with 142.5 m3 catalyst (4 m tube length). same conditions shows a modest increase of conversion
and selectivity to 67.9 % and 94.9 %, respectively (cf.
d t [em] Nt At [mz] Fig. 4). This means an increase of styrene output from
3 50441 19016 304500 to 312700 t/a (+2.68 %). Reducing the
ethylbenzene feed by approximately 4 % brings the
4 28373 14262
styrene output back to the design case of the
5 18159 11410 conventional packed-bed reactor. Hence, a potential of
6 12610 9508 4 % reduction of the ethylbenzene feed is predicted for
7 9265 8150 the membrane reactor under these conditions.
8 7093 7131
10 4540 5705 100
12 3153 4754
15 2018 3803
80

60
. . . . . . ,......::::.~.j;: 300

~
iii
40 ~
................ ........... / .. 250 ~
From Table 4 we notice that 20000 m2 is in fact not a E ..
-X-8m
o
<I 20 • m
realistic figure, as this would imply a tube diameter even iii
200 g
below 3 cm which is certainly too low to accept typical >- 0 +---~r-~-T-----r~~~~ ~
catalyst pellets of 5 - 6 mm. Moreover, more than 50000 hio iii

tubes would be required, which would be very costly.

·······Tr······ 150 E

40 ....:..............:.........~ ..~ ..... .

Now, one could argue that in fact not the entire hydrogen
has to be removed. For a closer look on the performance ·60 ....:.............:............1Savings (4%) 100

of the membrane reactor, the pseudo-homogeneous ·80 L---~~--~----~--~~~50

model is used. For these calculations, we assume Fractional feed F/Fd... [-]

maximum efficiency of hydrogen removal, i.e. the

permeated hydrogen being oxidized by air to supply the Figure 4. Simulated ethylbenzene conversion, X,
heat required for the dehydrogenation. Moreover, styrene selectivity, Ss" yield, Ys" output, ms" and
isothermal operation is supposed to mark the ideal case relative output change, L1mSb compared to a
of an infinite heat exchange between retentate and conventional packed-bed, of a multi tubular membrane
permeate. The membranes may be asymmetric sinter reactor for ethylbenzene dehydrogenation as a function
metal tubes with a reasonable diameter of 6 cm and a of the fractional feed (1.0 means design feed). Tube
wall thickness of 6 mm as a base case. The tubes may number: 12610. Tube diameter: 60n2 mm. Tube
consist of a support with 35 % porosity and a mean pore length: 4 m. 600" C, 1.2/0.7 bar. Catalyst efficiency:
diameter of 10 -20 !4m and be coated by a 20 !4m thick 30 %. H2 diffusivity (palladium): 2.0x1O· 7 m2·s· 1•
intermediate layer with 20 % porosity and a pore 10 !4m palladium coating.
diameter below 0.5 !4m. On top of this structure, a thin
metal diffusion barrier is assumed, and finally a 10 !4m Several reasons were identified for this limited
thick film of a hydrogen permeable metal, e.g. potential. First, the hydrogen permeability of the
palladium. The model takes into account the diffusion membranes has been estimated in a conservative way
resistance of the support, the intermediate layer, and the by assuming a linear decrease of the permeability with
palladium layer as well as the stagnant gas films adjacent the thickness of the palladium layer. Figure 5 shows the
to both faces of the membrane. As a first order estimate, influence of the palladium thickness on ethylbenzene
 R. Dittmeyer et al.lChemicai Engineering Science 54 (/999) 1431-1439 1437

and low diffusion resistance of the metal layer.

100 ,-----,-----,-----,---....,.., 330
Moreover, even though conversion may increase at
AAA higher catalyst efficiency in the case of a large tube
90 .............. !............. ;.............. ;.. 325 diameter too, the selectivity to styrene is reduced
~ -+-X
320 :>
m substantially, if the produced hydrogen cannot be
: -t.- S removed fast enough. All in all, a tube size of 6 cm
~
~
80 .......... ;.............. ;. -o-Y .. !.
___ m 8 seems to be a reasonable compromise.
(I) 315 ~ CI)
Besides membrane permeability, the kinetic properties
X E of the catalyst playa key role for the utilization of the
70 ..
membrane effect. This is demonstrated in Figure 8
showing the influence of catalyst efficiency on
60 .I..-_ _;...-_ _; . - - _ - - - '_ _--:....l 305 ethylbenzene conversion and styrene selectivity, yield,
0 5 10 15 20
and output. Note that the data in Figs. 6, 7a, and 8 have
Palladium layer thickness [~m]

Figure 5. Influence of the palladium layer thickness on Catalyst efficiency: :

90 90 .30% .40% .;
membrane reactor performance. Operating conditions .50% .60%
as of Figure 4. .100%
~80 ~ 80
conversion and styrene selectivity, yield, and output.
;;
>- .;
Note that a reduction of the thickness from 10 J.lm to 70 70
3 J.lm increases the styrene output to 320800 tla . .
compared to 304500 tla for the conventional packed-bed H2d~SiVity: 1.75x ,~.em2/s: 60 H2 diffusivity: 2.0 x , 0. m'~ls
7
60 "--,'--,_':""'_~'-:..J

reactor (+5.35 %). Based on the hydrogen permeability 6 7 8 9 6 7 8 9

reported by Kikuchi (1998) the styrene output would Tube diameter [em] Tube diameter [em]
increase even further to 323800 tla (+7.46 %).
Figure 6 illustrates the influence of the wall thickness of Figure 7. Influence of the tube diameter on the styrene
the support tube. Going down from 6 mm to 4 mm wall yield of the membrane reactor. 4 mm wall thickness.
thickness would bring additional 3500 tla styrene, i.e. a 3 J.lm palladium coating. H2 diffusivity (palladium)
total of 330800 tla (+8.61 %). However, a further Left: 1.75xlO-6 m 2·s- 1, Right: 2.0xlO- 7 m2·s- 1• Catalyst
efficiency as parameter. Other operating conditions as
100 , - - - - - , - - - - - - , - - - - - - " 340 of Figure 4.
95 ..

90 ". 335 been obtained for the 'best-case' scenario of a high

~ -+-X
~ hydrogen diffusivity of 1.75xlO-6 m2·s- 1 together with a
o
-o-Y 330 g palladium thickness of 3 J.lm and a support thickness of
J80 ___ m . -..... ........... .
~
::.
iii
4 mm. As catalyst efficiency approaches unity, the
styrene output reaches 405300 tla, representing a

. . ..
X 75 ....................:.................... ,........ E
__. •• : 325 33.1 % increase compared to the conventional packed-
70 ........ ~ ......:.................... , ......... ~ ..... :::.;. bed. In Figure 8 an additional curve &' is plotted
~ ~ showing the %-savings on the ethylbenzene (and
65 ~----i-----i----.t-1 320
o 2 4 6
steam) feed that are possible while maintaining the
design output of 304500 tla styrene. Note that an
Wall thickness of the support [mm]

Figure 6. Influence of the wall thickness of the support 100 16=;::~=:;::~=:;::~=r-1420

on membrane reactor performance. 3 J.lm palladium
coating. H2 diffusivity (palladium): 1.75xlO-ti m2·s- 1 80 400
(derived from Kikuchi, 1998). Tube diameter and m
U.
operating conditions as of Figure 4. <I 60 .. -+- X -t.- S .... 380 ;

reduction of the wall thickness appears unrealistic due to

.:
~ 40 ..
-o-y -"'-t.F
.. ___ m
360
g
~
CI) : : : : E
CI)

the reduced mechanical stability.

The effect of the tube diameter is shown in Figure 7. A X 20 .......... :~,,~t=,,=t="~::t . . 340
tube size below 5 cm can be ruled out because of the
requirement of an unproblematic catalyst loading. On the o -L-_ _- ' -_ _- ' -_ _+--_ _ +__~ 320
other hand, the wall area per unit of packed-bed volume 0.2 0.4 0.6 0.8 1.0
decreases with increasing tube diameter, leading to an Catalyst efficiency [-]
imbalance of the amount of hydrogen produced versus
the amount of hydrogen removed by permeation per unit Figure 8. Influence of catalyst efficiency on the
of packed-bed volume and time. The effect of the tube performance of the membrane reactor. 4 mm wall
diameter is more pronounced for high catalyst efficiency thickness. Operating conditions as of Figure 6.
438 R. Dittmeyer et al.lChemical Engineering Science 54 ( 1999) 1431-1439

increase of the catalyst efficiency from 30 % to 60 % between permeate and retentate of 10 - 15 K, although
allows for about 25 % reduction of the feed. It should be a reliable prediction of the temperature behavior would
stressed that this is not observed in a conventional require the use of a two-dimensional reactor model.
packed-bed reactor because there the deceleration of the Provided the catalyst efficiency can be increased, e.g.
main reaction at high conversion causes a loss of styrene by a moderate reduction of the pellet size, all in all the
selectivity. In the membrane reactor, the predicted advantages of the membrane reactor are seen in a lower
conversion is 84.2 %, the selectivity to styrene amounts ethylbenzene consumption, a reduction of the
to 94.4 %, giving a styrene yield of 79.5 %. fractionation costs, thanks to the lower ethylbenzene
Figure 9 gives an example of the calculated hydrogen load of the fractionation column, and in the potential
partial pressure profiles across the membrane. The for a simplified heat management. However, based on
profile refers to the reactor exit plane and to the 'best- the current state of the art, in particular concerning
case' scenario concerning the hydrogen permeability catalyst efficiency, the potential of a multitubular
(DH2 =1.75xlO-6 m 2·s-1, 3 lAm palladium coating) as well membrane reactor appears limited and would probably
as to an improved catalyst efficiency of 60 %. Note that not justify the additional investment required.
under these conditions, the support already controls the
overall hydrogen transport resistance.
ACKNOWLEDGEMENT
The authors thank H.J. Zander for valuable assistance
in parameter estimation, GKN Sinter Metals Krebsoge
Filters GmbH and Inocermic GmbH for the supply of

• • 1• 1.
1200
membrane supports, and the Institute for Solid State
r0-
e:.
N 800
and Materials Research, Dresden for performing PVO-
coating experiments. Financial support by the Bavarian
J:
0. Science Foundation under contract C3.2 (FORKAT) is
400 .... 1" ........ ,......... gratefully acknowledged.

o ~T---~--~---+--~---4~ NOTATION
d, Symbols
(4mm) (20I.lm) (3I.1m) At =Wall area [m2]
Figure 9. Hydrogen partial pressure profile across the
CjO =Feed concentration [mole.m-3]
composite membrane (cf. Fig. 3). Tube diameter:
Cj =Reactor concentration [mole.m·3]
d = Thickness (layer) [lAm, mm]
60/68 mm. 3 lAm palladium coating. Catalyst efficiency:
60 %. H2 diffusivity (palladium): 1.75xlO-6 m 2·s- 1• Other
dt =Tube diameter [cm]
operating conditions as of Figure 4.
L1m St =%-Change of styrene output [%]
L1F =%-Savings on feed (Eb and H20) [%]
Ej =Apparent activation energy [kJ/mole]
CONCLUSION
kj , =Rate constant [mole.m-3.s- I.Pa-X ]
The simulation results presented in this paper
k/ =dimensionless rate constant (cf. Table 3)
demonstrate the dependency of the performance of a k()j = Frequency factor [mole.m·3.s· I.Pa· X ]
hypothetical isothermal industrial-scale membrane Kj = Adsorption equilibrium constant [pa- I]
reactor for ethylbenzene dehydrogenation on hydrogen Kp =Reaction equilibrium constant [Pal
mSI = Styrene output [1000 t/a]
permeability and catalyst efficiency. Both have to fit
together in order to achieve a reasonable increase of Nt =Number of tubes
styrene yield. For this reaction, the highest reported CI;j =Estimated standard deviation
hydrogen permeability of a lab-scale composite Yij =Regression variable
palladium membrane of about 3.5xlO-6 mole·m- 2·s· I·Pa· 1 p =Pressure [bar]
(Kikuchi, 1998) seems to be sufficiently high. However, rj =Rate ofreactionj [mole·m-3·s· l ]
for what concerns industrial-size membranes, not only R = General gas constant []·mole·I.K I]
the permselective metal layer but also the support has to Ph =Packed-bed density [kg·m-3]
be optimized in order to minimize the diffusion So S =Weighted sum of squared residual errors [-]
resistance. To make use of the hydrogen removal effect Sj =Molar selectivity of product i [%]
requires both high permeability and high reaction rate. T = Temperature [0C]
High reaction rate and insufficient permeation rate leads r = Residence time [s]
to a decrease of styrene selectivity whereas high u =Flow velocity [m/s]
permeation rate and insufficient reaction rate prevents X = Molar conversion of ethylbenzene [%]
high conversion. In practice the limited heat transfer Yj =Molar yield of species i [%]
through the membrane tube and the fixed bed as well
may cause a significant temperature decrease towards the Indices
centre line which reduces the reaction rate to some extent r,z = Radial, axial
below the 'ideal' isothermal case. Assuming an overall R,S = Reaction side, sweep side
heat transfer coefficient of 100 - 200 W/m2 KI the P,R = Permeate, retentate
model indicates a moderate temperature difference P =Pore
 R. Dittmeyer et al.lChemical Engineering Science 54 (1999) 1431-1439 1439

St = Styrene Membrane Assisted Catalytic Dehydrogenation of

sim =Simulated Ethylbenzene to Styrene. Proc. 4th ESF-Workshop
exp = Experimental Optimisation of Catalytic Membrane Reactor Systems,
Oslo, May 30 - 31, pp. 171-179
Abbreviations Hirano, T., 1986, Roles of Potassium in Potassium-
B =Benzene Promoted Iron Oxide Catalyst for Dehydrogenation of
BFGS =Broyden-Fletcher-Goldfarb-Shanno Ethylbenzene. Appl. Catal. A:GeneraI26, pp. 65-79
CSTR = Continuous stirred-tank reactor Hsieh, H.P., 1996, Inorganic Membranes for
CVD = Chemical vapor deposition Separation and Reaction. Elsevier, Amsterdam
EB = Ethylbenzene James, D.H., Castor, W.M., 1994, Styrene, in:
HPLC =High pressure liquid chromatography Ullmann's Encyclopedia of Industrial Chemistry
IMCR = Inert membrane catalytic reactor Volume 25, 5th Ed., pp. 329-344, VCH, Weinheim
LHSV =Liquid hourly space velocity [m3EB/m·3cat.·hj Kikuchi, E., 1998, Membrane Reactor Application to
PVD = Physical vapor deposition Hydrogen Production. 3'd Int. Conference on CatalYSis
SiAMoD= Siemens Adaptive Modeling of Dynamic in Membrane Reactors. Copenhagen, September 8 - 10
SYstems - - -
KochloefJ, K., 1997, Dehydrogenation of
S/O = Steam to oil ratio Ethylbenzene, in: Ertl, G., Knazinger, H., Weitkamp,
ST =Styrene J., Handbook of Heterogeneous Catalysis Volume 5.
T =Toluene pp. 2151-2159, VCH-Wiley, New York
SVD = Singular value decomposition Lebedev, N.N., Odabashyan, G.V., Lebedev, V.V.,
Makorov, M.G., 1977, Mechanism and Kinetics of
Ethylbenzene Dehydrogenation and By-Product
Formation, Kinet. Katal. 18, pp. 1441-1447
Liu, Y., Dixon, AG., Ma. Y.H., Moser, W.R., 1993,
REFERENCES Permeation of Ethylbenzene and Hydrogen through
Abdalla, B.K., Elnashaie, S.S.E.H., 1994, Catalytic Untreated and Catalytically-Treated Alumina
Dehydrogenation of Ethylbenzene to Styrene in Membranes. Sep. Sci. Technol. 25, pp. 1511-1521
Membrane Reactors. AlChE J. 40, pp. 2055-2059 Muhler, M., Schlagl, R., Ert!, G., 1992, The Nature of
Abdalla, B.K., Elnashaie, S.S.E.H., Alkhowaiter, S. the Iron Oxide-Based Catalyst for Dehydrogenation of
Elshishini, S.S., 1994, Intrinsic Kinetics and Industrial Ethylbenzene to Styrene. J. Catal. 138, pp. 413-444
Reactor Modeling for the Dehydrogenation of Quicker, P., Hallein, V., Dittmeyer, R., 1998, Catalytic
Ethylbenzene to Styrene on Promoted Iron Oxide Dehydrogenation of Hydrocarbons in Composite
Catalysts. Appl. Catal. A:GeneraI1l3, pp. 89-102 Palladium Membrane Reactors. Submitted for
Becker, Y.L., Dixon, AG., Moser, W.R., Ma, Y.H., publication in Catal. Today
1993, Modelling of Ethylbenzene Dehydrogenation in a Quicker, P., Zander, H.J., Hausinger, G., Schmidt, F.,
Catalytic Membrane Reactor. J. Membr. Sci. 77, pp. 233- Dittmeyer, R., 1999, Studies on the Kinetics of
244 Ethylbenzene Dehydrogenation to Styrene on a
Bhat, Y.S., 1988, Kinetics of Dehydrogenation of Promoted Iron Oxide Catalyst using a Gradient!ess
Ethylbenzene to Styrene over a Promoted Iron Oxide Recycle Reactor. In preparation
Catalyst. Ind. Eng. Chern. 30, pp. 43-48 Tiscareno-Lechuga, F., Hill jr., e.G., 1993,
Bitter, J.G.A., 1986, Process and Apparatus for the Experimental Studies of the Non-Oxidative
Dehydrogenation of Organic Compounds, UK Patent Dehydrogenation of Ethylbenzene to Styrene using a
Application,2,201,159A Membrane Reactor. Appl. Catal. A General. 96, pp. 33-
Buxbaum, R.E., Kinney, AB., 1996, Hydrogen 51
Transport through Tubular Membranes of Palladium- Wenner, R.R., Dybdal, E.C., 1948, Catalytic
Coated Tantalum and Niobium. Ind. Eng. Chem. Res. 35, Dehydrogenation of Ethylbenzene. Chem. Eng. Progr.
pp.530-537 44, pp. 275-286
Carra, S., Forni, L., 1965, Kinetics of Catalytic Wu, J.e.S., Gerdes, T.E., Pszczolkowski, J.L., Bhave,
Dehydrogenation of Ethylbenzene to Styrene. Proc. Des. R.R., Liu, P.K.T., 1990, Dehydrogenation of
Dev. 4, pp. 281-285 Ethylbenzene to Styrene using Commercial Ceramic
Gallaher jr., G.R., Gerdes, T.E., Liu, P.K.T., 1993, Membranes as Reactors. Sep. Sci. Technol. 25, pp.
Experimental Evaluation of Dehydrogenations using 1489-1510
Catalytic Membrane Processes. Sep. Sci. Tech. 28, pp. Wu, I.C.S., Liu, P.K.T., 1992, Mathematical Analysis
309-326 of Catalytic Dehydrogenation of Ethylbenzene using
Gryaznov, V.M., Orekhova, N.V., 1998, Reactors with Ceramic Membranes. Ind. Eng. Chem. Res. 31, pp.
Metal and Metal-Containing Membranes. in: Cybulski, 322-327
A, Moulijn, J.A, Structured Catalysts and Reactors. Yang, W.S., Wu, J.e., Lin, L.W., 1995, Application of
Marcel Dekker, New York Membrane Reactor for Dehydrogenation of
Hermann, Ch., Quicker, P., Dittmeyer, R., 1997a, Ethylbenzene. Catal. Today 25, pp. 315-319
Mathematical Simulation of Catalytic Dehydrogenation Zander, H.-J., 1999, Dynamische Modellierung
of Ethylbenzene to Styrene in a Palladium Composite reaktionskinetischer Systeme mit Neuronalen Netzen
Membrane Reactor.J. Membr. Sci. 136, pp. 161-172 und hybriden Modellen. Doctoral thesis, University of
Hermann, Ch., Quicker, P., Dittmeyer, R., 1997b, Erlangen-Niirnberg, In preparation