The Chemical Engineering Journal, 48 (1992) 17-29 17

Development and verification of a simulation model for a non-

isothermal water-gas shift reactor

Riitta L. Keiski”, Tapio SaImib and Veikko J. Pohjola”

‘Department of Process Engineering, University of Ouh, SF-90570 Oulu (Finland)
bDepatiment of Chemical Engineering, Abo Akademi, SF-20500 Turku (Finland..

(Received December 29, 1989; in final from July 5, 1991)

Abstract

A fixed-bed test reactor suitable for studying non-isothermal reaction kinetics was developed. The reactor
allows axial and radial temperature measurement a.s well as online gas analysis. The water-gas shift
reaction over a commercial iron-based catalyst was chosen as the subject of a case study. A non-isothermal
reactor operating at temperatures between 575 and 675 K and with feed compositions corresponding to
industrial conditions was used. A method of sequential regression analysis was applied to determine the
kinetic parameters from the temperature and conversion profiles of the bed. The experimental data could
be fitted by a power-law type of reaction rate expression. The rate equation combined with a plug flow
model of the bed was successfully used to predict the fixed-bed behaviour within large temperature,
concentration and space velocity intervals.

1. Introduction the mass and heat transfer phenomena essential for

To study the kinetics of catalytic reactions several
experimental techniques have been developed. All
these techniques such as gradientless reactors, pulse
microreactors, isotope labelling, IR microreactors,
while good for fundamental kinetic studies, prove
to be less useful iu providing information about the
macrokinetics of a catalytic process. In many cases
sophisticated equipment cannot be used in studying
a particular system because of problems in online
analysis or in temperature and pressure control.
Therefore for macrokinetic studies a simple tubular
packed-bed reactor provides a very attractive al-
ternative because of its easy set-up and operation.
The main difficulty with a tubular reactor is the
non-isothermality: temperature variations of several
tens of degrees in the catalyst bed are normal.
However, the kinetic parameters of a catalytic
reaction can be determined using a packed bed
operating in conditions close to those of the cor-
responding industrial process. With a reactor
equipped with a device for sampling of the reaction
mixture in various locations along the tube and for
temperature measurements inside the tube it is
possible to obtain detailed information of the con-
centration and temperature gradients of the catalyst
bed. A quantitative description of the kinetics and
the modelling of industrially operating reactors can
be made by analysing the experimental concentration
and temperature profiles.
The water-gas shift reaction studied in the present
work is a reversible exothermic reaction which is
of particular interest owing to its industrial im-
portance:
CO+H,O = CO2 + H2
M(298 K)= -41.09 kJ mol-’ (I)
AS(298 K) = - 42.39 J K-’ mol-’
Iron oxides are used to catalyse the water-gas shift
reaction in the temperature range 320-450 “C. The
iron oxide catalyst contains chromium oxide as a
stabilizer which retards sintering and loss of the
surface area. The catalytically active component is
believed to be magnetite: before use Fez03 is reduced
to Fe304. Numerous studies of the kinetics of the
water-gas shift reaction over iron oxide-chromium
oxide high temperature catalysts (ferrochrome cat-
alysts) have been reported in the past 20 years [ 11.
There is, however, little or no agreement concerning
either the precise form of the rate equation or the
values of the rate constants or the activation ener-
gies. Both first-order [ 21 and second-order [3] ki-
netics have been suggested but many other more

0300-9467/92/$5.00 0 1992 - Elsevier Sequoia. All rights reserved

18 R. L. Keiski et ul. / Water-gas
complex rate equations have been proposed [ 4-81.
Podolski and Kim [9] found that both the Lang-
muir-Hinshelwood and the power-law models could
adequately describe the reaction behaviour. Fott et
al. [lo] report that 20 different kinetic equations
have been proposed during the last 40 years.
The confusion is claimed to be due to three
reasons: impurities in the reaction gases, such as
H2S [5], change the catalyst activity; diffusional
limitations [ 1 l] falsify the kinetics; and catalysts
from different manufacturers give different rate equa-
tions under identical conditions [ 121. Some of the
contradictory results can be explained by different
experimental conditions and catalyst deactivation.
The measured activation energies range from 40.6
kJ mol-’ [13] through 100 kJ mall’ [9], 115 kJ
mol-’ [14], 134 kJ mol-’ [2], 142 kJ mol-’ [15]
to 150 kJ mol- ’ [ 16 1. The low values are probably
caused by diffusional limitations, whereas the values
obtained using small catalyst particles are generally
in the range 100-113 kJ mol-‘.
In the present work the kinetics of the water-gas
shift reaction were studied using the non-isothermal
laboratory scale packed-bed reactor. The experi-
mental conditions, the inlet temperatures and the
dry gas space velocities are chosen to replicate
current industrial practice. The inlet gas compo-
sitions correspond approximately to the output from
a steam reformer. From axial conversion and tem-
perature data, the parameters of a kinetic model
can be estimated using non-linear regression analysis
and the behaviour of the reactor can be numerically
simulated.
2. Experimental details
2. I. Chemicals
A commercial ferrochrome catalyst (CCE C12)
was used in all experiments. The experiments were
performed with two different pellet sizes: the original
cylindrical catalyst pellets (3.2 mm X 3.2 mm) and
crushed and sieved particles (1.2-2.0 mm). The
catalyst contained typically 89% Fe,O, and 9%
CraOa; the rest was A1203 and graphite. AlaOa-Si02
spheres of the same diameter as the catalyst pellets
were used as an inert dispersive material before
and after the catalyst bed.
The gases used in these experiments were obtained
from cylinders and purified in molecular sieve 3A
columns before the reactor. The purity quoted by
the manufacturers was better than 99.9% for Nz,
Hz and COa and better than 99.0% for CO. Deionized
water was used in all experiments.
shifi reactor model
2.2. Equipment
The equipment comprised a gas flow regulation
and mixing system, a reactor section and an online
analytical section. The equipment is shown in Fig.
1.
The reaction gases were metered and regulated
by electric mass flow controllers (Brooks mass flow
controller 5850-TR) and purified and mixed in the
molecular sieve beds before entering the reactor
section. Deionized water was fed to the steam gen-
erator with a Varioperex peristaltic pump (LKB
model 12000).
The reactor section consisted of a preheater (G.
W. Berg tubular reactor) and a non-isothermal re-
actor. The non-isothermal reactor was constructed
in our laboratory from stainless steel tubing of inner
diameter 27.3 mm. The reactor consisted of two
parts: a preheating section of 140 mm length without
a vacuum jacket and a reactor section of 260 mm
length wit,h a vacuum jacket. The reactor section
was surrounded by two massive ovens: one for the
preheating section and the other one for the reactor
section. The whole assembly was insulated with a
50 mm layer of mineral wool. The vacuum jacket
pressure was maintained at the value of 0.001 Pa
by a vacuum pump constructed at the University
of Oulu.
Thermocouples and gas sampling lines were con-
nected via the vacuum jacket to the top and the
bottom of the reactor as shown in Fig. 1. Tem-
peratures were measured at six axial positions using
Ni-Cr-Ni thermocouples. Three thermocouple wires
were connected into one axial sampling point to
measure the radial temperature profiles. These po-
sitions were normally at the axis and at the distances
of 5 and 10 mm from the axis. All thermocouples
were connected to a YEW 3087 portable hybrid
recorder for temperature read-outs. Temperatures
were also measured at the preheater outlet and at
the two ovens around the reactor. Tyristor regulators
(Eurotherm) were used to control the oven tem-
peratures.
Gas samples of 120 cm3 min- ’ were taken from
the inlet, outlet and from four different axial positions
of the catalyst bed at a distance of 5 mm from the
reactor wall. This position represents approximately
the mean value of the radial concentration distri-
bution (( l/4)%( 2 /3)R) [ 171. The gas samples were
taken by switching on the valves at the different
outlets one at a time and throttling the main valve
after the reactor to get a slight overpressure in the
reactor. A condenser was placed at the reactor outlet
to remove the major part of the water. Cold water
was used as a cooling medium. The remaining
water was removed from the gas samples by the
 R. L. Keiski et al. / Water-gas
Flow control Reactor
GAS SAMPLING
FEED -_,
THERMOCOUPLES
Fig. 1. Details of the reactor system: 1, mass flow controller, 2, steam generator; 3, preheater; 4, regulator; 5, diffusion pump;
6, reactor; 7, temperature recorder; 8, condenser; 9, molecular sieve; 10, IR analyser; 11, gas chromatograph;
13, flow meter; 14, mercury manometer.
molecular sieve 3A columns before the gas
analysis.
Reaction and product gas samples were analysed
by a Hewlett Packard 571 OA gas chromatograph
connected to a Hewlett Packard 33SOA integrator.
The gas chromatograph contained an active carbon
column (l/S in) to separate the N2, CO and CO2
responses. The column length was 4 m and helium
was used as a carrier gas. The helium flow rate
was 35 cm3 ruin ‘. The gas chromatograph was
equipped with a TC detector operating at 250
“C. The column temperature was 75
“C. With this system the responses of NB, CO and
CO2 could be recorded after retention times of 1.42
min, 1.81 min and 9.67 min respectively. The hy-
drogen peak had a negligible area since the thermal
conductivity of Hz is very close to that of helium.
At each experimental point three or four parallel
gas samples were taken at 2 min intervals. The
volume of the sample coil of the gas chromatograph
was 0.5 cm3. The CO content of the dry product
gas was analysed continuously by an IR analyser.
The gas chromatograph and the IR analyser were
calibrated with four premixed gases containing CO,
CO2 and Na or CO and Nz. The conversions and
molar flow rates of CO as well as the inlet and
outlet gas compositions of each outlet were cal-
shifi reactor mQd.el 19
Analysis
VACUUM
12, integrator;
culated by a Hewlett Packard 85 desk computer
using the peak areas in chromatograms and response
factors obtained from calibration gas analysis.
2.3. Experimental procedure
The catalyst was loaded into the reactor from the
top of the tube and the bed depth was 410 mm
with 240 mm and 10 mm layers of inert A1203-Si02
spheres above and below the bed respectively. At
the start-up of an experiment the reactor was heated
to 500 K at atmospheric pressure while nitrogen
was flowed through the catalyst bed. After achieving
500 K the catalyst was reduced at 506-675 K using
the following dry gas composition: 13.6% CO, 9.2%
COB and 77.2% Hz. The molar ratio of Hz0 to dry
gas was 0.83. After the catalyst reduction non-
isothermal experiments were carried out using five
reaction gas inlet temperatures (575-675 K) and
different flow rates and gas compositions. Between
the experiments the catalyst bed was kept at the
reaction temperature while nitrogen was flowed
through the bed.
The catalyst stability was analysed by repeating
some experiments (basic experiments) after certain
periods of time. Experiments using Nz adsorption,
mercury porosimetry, scanning electron microscopy
@EM) and X-ray photoelectron spectroscopy (XI’S)
20 R. L. Keiski et al. / Water-gas shifi reactor mm!&

to investigate the physical properties of the fresh r=kfc) (2)

and the used catalysts were made, too. A fast decay
in the catalyst activity was detected during the first where the rate constant k depends on the temper-
150 h of operation, after which the decay became ature according to Arrhenius’ law
much slower. The slow decay has been found to
be due to a sintering process: a loss of the surface k=ko exp(-EIR,n (3)
area accompanied by a growth in the mean pore
The kinetic parameters in eqn. (3) were deter-
size.
mined using the pseudohomogeneous plug flow
The kinetic experiments were performed at at-
model for the catalytic fixed-bed reactor. The con-
mospheric pressure during the slow decay of the
tinuity equation for CO is given by
catalyst activity. The inlet temperatures of the cat-
alyst bed were 575, 600, 625, 650 and 675 K and dr&, IdV = rcop,, (4)
the inlet gas contained CO, COz, hydrogen, nitrogen
and water vapour. The molar ratio between inlet where tic0 is the molar flow rate of CO.
H,O and CO was varied from 2.4 to 12.1 and the A Langmuir-Hinshelwood type of reaction rate
dry reaction gas contained typically 7-30% CO, expression could be derived starting from a set of
3-20% COa, 25-70% Ha, the rest being Nz. The elementary reaction steps. Such rate equations, how-
basic dry gas contained 13.6% CO, 9.2% COz, 54.4% ever, usually include many adjustable parameters.
Hz and 22.8% Nz and the H,O-to-dry-gas mole ratio Therefore we prefer here to use a power-law type
was 0.83. of rate expression (eqn. (5)) which includes a min-
Four different kinds of experiment were per- imum number of parameters:
formed: changing the inlet temperature
(To = 575-675 K), changing the dry gas space ve- -~co=kcZoGoc”co,cW -PI (5)
locity (S= 795-4009 1 h-r), changing the H,O-to-
dry-gas mole ratio (H,O/dry gas= 0.33-1.24) and where p is the reversibility factor: ~=ccozcnJ
changing the CO, COz and/or Ha concentrations in KT GO CHzO.
the reaction gas (Table 1). The gas samples and The molar flow jLee, the concentration cc0 and
temperature read-outs were taken from each outlet the conversion X of CO are related to each other
after the reactor steady state was achieved, ap- by the equation [lico/ritot = cc0 = c,(l -X)] where co
proximately after 1 h. Catalyst number 1 (3.2 and Vt,,, denote the inlet concentration of CO and
mm X 3.2 mm) and catalyst number 2 (1.2-2.0 mm) the total volumetric flow rate respectively.
were used for 420 h and 610 h respectively, before Replacing V by rR2L and expressing molar flows
they were removed from the reactor for physical and concentrations as functions of X, eqns. (3)-(5)
analysis. give

dXldL = (p,,rR21rio)koexp( -EIR, T)con+m+p+q

3. The reactor model
x (I -x)“(“HzO -x)“~Kzo2 +v

X (J&z +KJq[l - (3&z +-WM,, +-V/K,

The rate expression of a catalytic reaction can
often be given in the form 4 1 --w0&?0 -x)1 031

TABLE 1. Experiments performed

Experiment number TCI 7io S H20/C0 co,/co Hz/CO

0 (mol h-‘) 01-V (mol/mol) (mol/mol) (mol/mol)

1 575, 600, 1.116 1997 6.10 0.67 3.99

625, 650,
675
2 575-675 0.442-2.237 795-4009 6.10 0.67 3.99
3 575-675 1.116 1997 2.43-9.12 0.67 3.99
4a” 575-675 0.566-2.481 1997 2.74-12.03 0.25-1.39 1.97-5.31
4bb 600, 650 0.566, 1.116 1997 4.58-12.03 0.25-1.32 1.97-7.87

‘Catalysts 1 and 2, changing the concentration of one component at a time. bCatalyst 2, changing the concentration of one
component or more than one component at a time (full fractional design) [ 171.
 R. L. Keiski et al. / Water-gas shifi reactor model 21

where liO and c0 are defined by the following equa- Li

tions: X(Li) = Q(LY T> a (11)
s
0
co(T) = (~co,o&O~,o)(Pl&T) (7)
where g(L, T) denotes the right-hand side of eqn.
ti,=V tot,OcO(at 293 K) ‘(8) (6). A direct numerical solution of the differential
The equation for the equilibrium constant KT is equations is thus avoided and computing time is
given by Moe [3]: saved. The sum of square errors is minimized by
the appropriate choice of the kinetic parameters.
KT= exp(4577.8/T- 4.33) (9) The optimization method used in the improvement
of the parameters is the Gauss-Newton-Marquardt
where temperature T is expressed in kelvins.
procedure. The observed and estimated conversion
The non-isothermal approach [IS] requires data
profiles are finally compared with numerical sim-
for CO conversion and reaction temperature VS.
ulations solving the differential equations (eqn. (6))
catalyst bed length. The continuity equation (eqn.
by a semi-implicit Runge-Kutta method [20] using
(6)) is therefore integrated numerically along the
a simulation program developed in our laboratory.
tube length coordinate for each experiment, with
estimated values of the kinetic parameters /co,E,
n, m, p and q. The experimental and calculated
CO conversion values are then compared with each 4. Results and discussion
other to guide the choice of improved parameter
values. 4.1. Physical investigation of the catalyst
The axial temperature profile of the reactor is Physical studies of the fresh and used catalysts
approximated by an empirical fourth-order poly- showed that sintering occurred during the early
stage of catalyst use. Similar results have been
nomial
reported by previous investigators [ 2 1, 221. The
loss in surface area was minimal during the kinetic
(10) runs and according to the SEM and XPS investi-
gations no carbon formed on the catalyst surface
to avoid the use of the energy balance during the during the experiments. The physically determined
estimation of the kinetic parameters. As the radial catalyst properties are summarized in Table 2.
temperature profiles were within experimental error
they were neglected, and the simple axial model 4.2. The temperature pro$iks
(eqn. (10)) was taken. To allow the determination of the kinetic param-
The parameter estimation is performed by a non- eters, the temperature profile has to be known at
linear regression program called Reproche [ 19 1. The every axial point of the reactor. Therefore the po-
program can be applied to both algebraic and dif- lynomial model (eqn. (10)) was fitted to every
ferential models and it requires user-supplied sub- experimental temperature profile. The fourth-order
routines to calculate the reactor models. polynomials approximate very well the experimental
The algebraic model is used to determine the data. Typical examples of the estimated temperature
parameters of the axial temperature profile (eqn.
(10)) and the differential model to determine the TABLE 2. Catalyst (CCE C12) properties and reactor parameters
kinetic parameters of the plug flow reactor model
(eqn. (6)). In the algebraic model the empirical 3.2 mmX3.2 mm 1.2-2.0 mm
pellets particles
parameters ao, . . .,a4 are determined by the
Gauss-Newton-Marquardt least-squares procedure Fresh Used Fresh Used
to minimiie the objective function. The empirical
parameters (eqn. (10)) are determined for each SBm (m’ g-l) 53.7 35.5 52.5 35.4
experiment separately to allow a very accurate in- VP (cm” g-l) 0.133 0.109 0.142 0.100
r&l(nm> 4.9 5.7 5.4 6.0
terpolation of the temperature profile.
lP 0.549
In the differential part of the regression package pb (g cmm3) 1.155 0.955
the direct integral least-squares method [ 19 ] is used. pe (g cm-“>” 3.73
The differential equations (eqn. (6)) are transformed m, (g) 108.1 89.45
to integral equations and the integrals are approx- Reactor volume V (cm3) 93.66 93.66
imated by cubic spline functions through the ex- Obtained from Nz adsorption~esorption measure-
perimental points L1, LB, . . ., L6: ments. bObtained from helium pycnometer measurements.
22 R. L. Keiski et al. / Water-gas sh@ reactor rrwakl

profiles are given in Fig. 7(b) below. Usually the

deviations between the estimated and the experi-
mental temperatures were below 1 K. The empirical
constants ao, . . ., a4 for each data set were then
used in kinetic modelling.

4.3. First-order rate constant - pore di$Gsion

effects
The modelling work was started by the deter-
mination of the first-order rate constant parameters
(n=l, m=p=q=O in eqn. (5)) for the crushed
particles (1.2-2.0 mm) and the original pellets with
different ages. The first-order rate approach was
chosen because according to literature cited above
the water-gas shift reaction has often been found
to be close to the first order and so this approach
can be used as a good diagnostic tool. The rate
parameters are listed in Table 3 and the logarithmic
plots of the calculated constants are shown in Fig.
2. As can be seen from Fig. 2 a minor decrease in
the catalyst activity still occurs between 200 and
300 h: the rate constant obtained using the data -2.00
0.0012 0.0015 0.0018
set c (350-400 h) has a slightly lower value than -1
T/K
the constants obtained using the data sets a and
Fig. 2. The calculated temperature dependence of the first-order
b in Table 3 with the catalyst ages of 179-243 h
(n = 1) rate constants: - -, original pellets; - - -, crushed
and 204-278 h respectively. pellets with an age of 350-400 h; . . . , crushed pellets with
The rate parameters determined for the original an age of 204-278 h; -, crushed pellets with an age of
pellets (3.2 mmX 3.2 mm) are also listed in Table 179-243 h.
3. Slightly lower values for the activation energy
and the frequency factor are obtained for the original
pellets than for the crushed particles. A comparison
conditions the expression of the effectiveness factor
between the predicted values of the first-order rate
for irreversible first-order reactions is used:
constants ZIS.inverse temperature is shown in Fig.
2. At the lowest temperatures (below 643 K) the
3 1 1
rate constant obtained with the original pellets is (12)
higher, whereas at the highest temperatures (above q=. ( a-- )
663 K) the constant obtained using the crushed
particles is higher. In the intermediate temperature where 4, denotes the Thiele modulus defined as
region 643-663 K the k values almost coincide.
To investigate the role of pore diffusion in the 4, = rp(kpp/D,>“2 (13)
overall reaction rate the effectiveness factor ap-
proach was used. Since the influence of the factor The equivalent radius of the cylindrical pellets r,
p on the rate law (eqn. (5)) is minor in actual is 1.96 mm. The estimation of the effective diffusion
coefficient D, was performed according to
TABLE 3. Parameters of the first-order rate law determined for
different catalyst particles D,=(E~/T)(~/‘D+~/D&~ (14)

Data ln k. s EIRG s N Catalyst age where D and DK denote the molecular and Knudsen
set (cm3 g-’ s-l> (96) 0 (%) (h)
diffusion coefficients respectively. The molecular
a 17.72 2.4 10954 3.6 20 179-243 diffusion coefficient of CO was obtained from the
b 18.53 1.6 11455 2.6 19 204-278 binary diffusion coefficients using the Wilke ap-
20.33 1.3 12760 2.1 20 350-400 proximation valid for dilute solutions [23]. The
: 12.79 0.1 7812 0.1 21 201-311 binary diffusion coefficients for gas pairs CO-i
“Data sets a-c, crushed catalyst; data set d, original pellets. N, (i= H,O, Ha, COa, N,) were computed using the
number of conversion profiles in estimation. Fuller-Schettler-Giddings equation [23]. The Knud-
 R. L. Ketiki et al. / Water-gas
sen diffusion coefficient for gases in a porous solid
was calculated from the equation given by Salter-field
[24]. The calculations showed that the molecular
diffusion coefficient of CO is practically independent
of the gas composition at the actual conditions.
The molecular diffusion coefficient of CO varies
between 0.93 and 1.40 cm2 s-’ in the temperature
range 573-723 K, whereas the Knudsen diffusion
coefficient of CO varies between 0.095 and 0.107
cm2 s-i in the same temperature range. Thus the
diffusion process is strongly dominated by Knudsen
diffusion in the actual concentration and temperature
domain.
To obtain the effective diffusion coefficient from
eqn. (14) the particle porosity and tortuosity are
needed. The porosity of the catalyst pellet was
determined as 0.55 (Table 2). For the tortuosity of
the iron oxide-chromium oxide catalyst, contro-
versial values appear in the literature [25, 261: the
values of the tortuosity factor +rrange from 2 to
9. Therefore a calculation of the effective diffusion
coefficient using these limiting values of r was
performed. For the lowest limit of r (r=2) the
effective diffusion coefficient of CO would vary
between 0.024 and 0.027 cm2 s-l at temperatures
of 573-723 K. For the highest value of r (r=9)
the effective diffusion coefficient would have values
between 0.0053 and 0.0060 cm2 s-l in the same
temperature range. The higher values of D, indicate
that the effectiveness factor has the values 0.97
and 0.45 at temperatures of 573 K and 723 K
respectively. The lower values of D, indicate ef-
fectiveness factors of 0.87 and 0.24 at 573 K and
723 K respectively. The calculations indicate that
the reaction on the original catalyst pellets is some-
what diffusion limited, at least in the hot spot region
of the catalyst bed. Some values of diffusion coef-
ficients and effectiveness factors are listed in Table
4.
In principle the apparent first-order rate constant
k’ is obtained from the intrinsic rate constant and
the effectiveness factor according to
k’=qk (15)
TABLE 4. Comparison of the rate constants obtained with two pellet sizes
TO D (cm2 s-‘) DK (cm2 s-l) k (cm3 g-’
723 1.40 0.107 14.4
703 1.33 0.105 9.36
683 1.27 0.104 5.81
663 1.20 0.102 3.50
k, the rate constant obtained for the crushed particles. k’, the apparent rate constant obtained according to eqn. (15). k”, the
rate constant obtained for the original pellets.
shi,ft reactor model 23
Calculation of k’ using the effectiveness factors
obtained with r= 2 gives values which are slightly
lower than the apparent rate constant obtained for
the original pellets as can be seen in Table 4. The
correction of the rate constant can be considered
to be satisfactory, taking the uncertainty in the
tortuosity factor into account. At the lowest tem-
peratures there remains, however, a minor dis-
crepancy between the rate data for large and small
particles: the slightly higher rate constants obtained
with the original pellets at the lower temperatures
cannot be explained by diffusional effects.
4.4. Detminuttin of the kinetic parameters
in the power law
The modelling work was continued by considering
the different forms of the power law (eqn. (5)). The
frequency factor ko, the activation energy parameter
E/R, and the concentration exponents of CO (n)
and Hz0 (m) were determined from the experimental
conversion profiles. In the first attempt, the exponent
of Ha0 was fixed at zero and three parameters (ko,
E/R, and n) were estimated. In the following stage
all four parameters were determined simultaneously.
In the sequel the first-order rate model (n= 1)
including only k. and E will be called the two-
parameter model, whereas the models with one (n)
and two (n and m) concentration exponents are
called three- and four-parameter models respec-
tively.
One part of the data obtained with the original
pellets was chosen for regression: 12 experimental
runs within the inlet temperature range 600-650
K and with catalyst ages between 2 12-251 h were
used in parameter estimation. This temperature
range was chosen to minimize the diffusional in-
fluence on the data (Table 4). These data also cover
the concentration domain studied well since the
extreme inlet concentrations of CO, C02, Ha0 and
H2 are included.
The results of the regression analysis are shown
in Table 5. It can be concluded that the three- and
four-parameter models give similar values for the
activation energy and the CO concentration
s-l) 7) k’ (cm3 g-l s-l) k* (cm3 g-’ s-l)
0.449 6.59 7.31
0.528 4.94 5.37
0.615 3.57 3.88
0.705 2.47 2.75
24 R. L. Keiski et al. / Water-gas
TABLE 5. Determination of the parameters in the power-law
13.30 1.8 9243 2.5 0.77
15.93 1.4 9407 1.7 0.73
14.78 1.9 9598 1.8 0.74
MRS, 100 X mean residual square. MRS= 0.28 for the two-parameter
exponent n. The activation energy is 77-78 kJ
mol-’ and the concentration exponent of CO is
between 0.7 and 0.8 according to both models. The
Hz0 concentration exponent m becomes less ac-
curate since Hz0 was present to a large excess in
most experiments; the H,O-to-CO ratio at the inlet
was varied between 2.7 and 12. The influence of
Hz0 on the reaction rate can thus be included in
the frequency factor. The standard deviation of the
CO concentration exponent is typically about 2-3%,
whereas that of the Ha0 exponent is over 7% (Table
5). The mean residual squares are also listed in
Table 5. The decrease in the mean residual square
when switching from the two-parameter model to
the three-parameter model is more significant than
the corresponding difference between the three- and
four-parameter models.
Furthermore, the effect of the COZ concentration
was investigated by determining the COa concen-
tration exponent p simultaneously with the other
exponents n and m. The results are listed in Table
5. The frequency factor, the activation energy and
the CO concentration exponent remain practically
unchanged, whereas a slightly lower value of the
Ha0 concentration exponent was obtained than in
the four-parameter model (Table 5). A small negative
value (p = - 0.2) of the COZ concentration exponent
was obtained which indicates the inhibitory effect
of COa on the reaction rate. Compared with the
four-parameter model, the inclusion of the fifth
parameter p did not essentially improve the overall
fit, and the mean residual square remained on the
same level.
Attempts to incorporate also the H2 concentration
exponent Q in the model gave ambiguous results:
depending on the data set used, the Hz concentration
exponent converged to very small positive or neg-
ative values during the regression analysis. It seems
to be obvious that the effect of the Ha concentration
is very minor on the forward shift reaction rate.
The Ha concentration exponent was therefore not
included in the linal model.
The parameters listed in Tables 3 and 5 give
activation energies of 77, 78 and 80 kJ mol- ’ for
the three- four- and five-parameter models respec-
shifl reactor rrwdel
rate equation (eqn. (5))
3.3 _ - _ 0.17
2.5 0.55 7.6 - - 0.08
2.5 0.47 10.1 -0.18 25.0 0.08
model (experiments in Table 3, data set d). [k,] = (dm3)n+mcp+q
tively. Our results are thus within the values pub-
lished previously for iron-based catalysts. For ex-
ample, Bohlbro and Jorgensen [27] report 106 kJ
mol- ’ for small particles and 60 kJ mol- ’ for large
particles for a similar rate law to our five-parameter
model. The first-order rate law parameters listed in
Table 3 give an activation energy of 91-106 kJ
mol-’ for the crushed particles, whereas the ac-
tivation energy of the original pellets is about 65
kI mall’. As discussed previously, the literature
values range from 40 to 150 kJ mol-’ [2, 9, 13,
15, 16,281. The differences between the pellet sizes
are reflected in the measured activation energies in
the literature: the smallest pellets give the highest
activation energies. Similarly in this study this ob-
servation is confirmed.
The most exhaustive studies of water-gas shift
steady state kinetics have been performed by Bohl-
bro [ 4, 5, 14, 271. The concentration exponents
reported by Bohlbro and Jorgensen [ 2 7 ] are typically
n= 0.75-0.9 for CO, m = 0.24-0.45 for Hz0 and
p=(-0.6)-(-0.18) for COZ, depending on the
particle size and the catalyst composition. In their
most recent study Bohlbro and Jorgensen [27] give
n=0.87, m=0.26 and p= -0.18 for large catalyst
particles and n=0.93, m=0.24 and p = - 0.31 for
small particles. In practice, taking the normal ex-
perimental scattering into account and knowing the
difficulties in determining the concentration ex-
ponents exactly, the small- and large-particle data
of Bohlbro and Jorgensen [ 2 71 can be regarded as
identical. Goodridge and Quazi [ 121 used a power-
law rate expression without the factor p. They
obtained n=l.l, m=0.5-0.8, p=(-0.9)-0.5 and
4 = ( - 0.5)-O (4 is the Hz concentration exponent);
thus the influence of the reverse reaction was shifted
to the exponents of Hz and COZ which have slightly
more negative values than the exponents obtained
by us and by Bohlbro and Jorgensen. We can
conclude that the concentration exponents deter-
mined by us (Table 5) are in fairly close agreement
with the data of Bohlbro and Jorgensen [27] and
also with the data of Goodridge and Quazi [ 121.
The present type of experiments are not very
well suited for investigation of the microkinetic steps
 R. L. Keiski et al. / Water-gas sh@ reactor nwclel 25

proceeding on the catalyst surface. However, re-

cently isotope labelling techniques [ 29 ] and transient
response studies [ 301 have revealed new information
concerning the shift reaction mechanism over iron-
based catalysts. It is claimed that the reaction con-
sists of a regenerative mechanism involving slow
CO oxidation and COa desorption, fast Hz0 ad-
sorption and interconversion followed by slow hy-
drogen desorption steps. These assumptions result
in a steady state rate model, where the CO and
Ha0 concentrations appear in the numerator and 0.60

CO, HzO, COa and Ha in the denominator of the

mathematical expression [ 291. Thus the empirical
exponents obtained by us are in accordance with
the mechanistic rate equation: the exponents of CO
and Ha0 are less than unity because of adsorption.
The adsorption afllnity of Hz0 is strong and its
concentration is higher than the concentration of
CO; therefore the exponent of Ha0 is lower than
that of CO; COa and H2 appear as inhibitors. It
should be emphasized, however, that the present
study cannot be used to confirm the results of the
previous mechanistic studies [29, 301 but the em-
pirical model obtained here is not in conflict with
models based on the intrinsic kinetics.
4.5. Compariscm of the model with
experimental data
The models are further compared by considering
the fit of the simulated conversion profiles to the
corresponding experimental prollles. It is essential
that the model is also valid at the extreme regions
of the experimental data and that no systematic
deviations occur in model predictions. The extreme
data sets in the sense of the water-to-dry-gas ratio
and the inlet CO, COZ and Hz concentrations are
considered first.
The experimental and calculated conversion pro-
files for the lowest H,O-to-CO ratio are given in
Fig. 3. For the highest H,O-to-CO ratio ( = 12) the
three- and four-parameter models give a satisfactory
prediction of the reactor performance. However,
with decreasing Hz0 excess and temperature the
differences between the models increase: the two-
parameter model with first-order kinetics is not able
to describe the reactor behaviour well at the inlet
temperatures of 575-600 K, too high conversions
being predicted as can be seen in Fig. 3. For the
inlet temperature 650 K the outlet conversions are,
however, predicted quite well (Fig. 3) by the two-
parameter model also. The tendencies described are
observable in similar experiments performed with
the H,O-to-CO ratios 3.1 and 4.6. With an increasing
excess of Hz0 the difference between the two- and
three-parameter models gradually disappears; for
0.40
0.20
I I I I I I I I I I I I I I
0.00
0.00 0.50 1.00 1.50 L/dm
Fig. 3. The experimental and estimated conversion profiles at
a low H,O-to-dry-gas ratio (0.33): 0, T0= 600 K; A, 2’0= 650
K; -9 two-parameter model; . . . . . , three-parameter model;
- - -, four-parameter model. The inlet conditions were lice = 1.1
mol h-’ and S= 1997 h-l; the inlet mole ratios were HzO/
CO = 2.4, CO&O = 0.67 and H&O =4.0.
example, the two-parameter model gives a quite
good fit to the observed conversion profiles at the
HaO-to-CO ratio 9.1.
The predictions of the three- and four-parameter
models are quite similar, except the cases with the
lowest Ha0 excess, the HzO-to-CO ratios 2.4 and
3.1. The major difference between the three- and
four-parameter models (Fig. 3) is caused by the
decrease of the HZ0 concentration; in these con-
ditions the influence of the Ha0 exponent becomes
significant and the reaction rate cannot be explained
by the concentration of CO solely (Fig. 3). With
an increasing excess of Ha0 the difference between
the three- and four-parameter models gradually di-
minishes, and at the highest Ha0 excess the pre-
dictions become practically identical.
An example of the model predictions at a high
CO inlet concentration can be seen in Fig. 4 which
represents the inlet composition 30.3% CO in the
dry gas, the corresponding HaO-to-CO ratio being
6.1. The four-parameter model gives again the best
fit to the experimental curves. Figure 4 shows that
26 R. L. Keiski et al. / Water-gas shiI reactor model

II I I ] IllI ) III II II

0.40

0.20

0.00 0.00
0.00 0.50 1.00 1.50 L/dm 0.00 0.50 1.00 1.50 L/dm

Fig. 4. The experimental and estimated conversion profiles at Fig. 5. The experimental and estimated conversion profiles at
a high inlet concentration of CO: El, To=600 K; A, 7’,,=650 a nigh concentration of CO,: 0, T0 = 600 K; A, T, =650 K;
K; -, two-parameter model; . . . . ., three-parameter model; -, two-parameter model; . . . , three-parameter model; - - -,
- - -, four-parameter model. The inlet conditions were ri,,,, = 2.5 four-parameter model; - -, five-parameter model. The inlet
mol h-’ and S= 1997 h-‘; the inlet mole ratios were HaO/ conditions were +A,,ce = 1.1 mol h-’ and S= 1997 h-r; the inlet
CO=6.1, CO,/CO=O.30 and H&0=1.8. mole ratios were H,O/CO=6.1, CO,/CO= 1.4 and HJCO=4.0.

it is useful to account for the Hz0 concentration
when the HZ0 excess is only moderate (H,O/CO = 6.1
at the inlet in the present case).
The model behaviour at higher COz inlet con-
centrations is considered in the next stage. The
reactor inlet dry gas contains 19.2% CO2 in the
experiments depicted in Fig. 5. The inlet H,O-to-
CO ratio was 6.1. The results indicate that the
proposed model is sufficient to characterize the
reaction kinetics also at higher COz concentrations
than those used in most experiments (9.2% COz in
the dry gas at the inlet). This conclusion was further
confirmed by experimental results obtained at 15%
CO2 in the dry gas. Simulations with the five-pa-
rameter model including the COz exponent (Table
5) are also depicted in Fig. 5. The predictions of
this extended model are very similar to those of
the four-parameter model.
The model behaviour at the highest Hz inlet
concentration (72% H, in the dry gas) is illustrated
in Fig. 6. In these experimental conditions the
predictions of the three- and four-parameter models
are practically identical The simulation results in-
dicate that the rate model developed also describes
well the reactor behaviour at a wide range of inlet
hydrogen concentrations and that the rate of the
forward shift reaction is practically independent of
the H2 pressure.
4.6. The predictive properties of the model
The predictive properties of the model were in-
vestigated by comparing simulations with new ex-
perimental data which were not used in parameter
estimation. Two aspects in particular are considered:
the model behaviour at a high temperature and a
high space velocity domain.
An experimental set obtained at the inlet tem-
peratures 575-675 K is depicted in Fig. 7. The
space velocity was 1997 h- ’ as in the previous
cases described. The independent simulations using
the two-, three- and four-parameter models are also
shown in Fig. 7. The figure shows that the reactor
behaviour is very well predicted by the three- and
four-parameter models over the entire temperature
domain. The two-parameter model explains the re-
actor performance well at the highest inlet tem-
peratures (650-675 K), whereas it predicts too high
conversions at the lowest temperatures. In the
 R. L. Keiski et al. / Water-gas shZft reactor model 27

X
III1 1 II l 1, l I l l , , l
1.00 III1 1 II I 11 I I I I, 11

0.80

0.60

0.00 0.50 1.00 1.50 L/dm

0.00 0.50 1.00 1.50 L/dm (a)
Fig. 6. The experimental and estimated conversion profiles at 750.00
, I I , I III 1 III II l 11
a high concentration of Hz: 0, T,,= 600 K; A, Z’,,= 650 K, -,
two-parameter model; . . . . . , three-parameter model; - - -, four-
parameter model. The inlet conditions were A,,,co = 1.1 mol h-’
and S= 1997 h-‘; the inlet mole ratios were H,O/CO=6.1,
CO&O = 0.67 and H&O= 5.3.

previous section the model parameters were de-

termined with the low and medium temperature data
(600-650 K); nevertheless, the model is also able
to predict the conversion correctly at the highest
temperatures (675 K at the inlet). This may suggest
that the effect of pore diffusion is quite small even
if we could not exclude it according to the effec-
tiveness factor considerations. A possible expla-
nation is that only a very narrow section of the
reactor is warmer than 673 K (Fig. 7(b)); thus pore
diffusion has only a minor influence on the overall
reactor performance. Our conclusion is that the
suggested model can be used over the whole tem-
perature domain 575-675 K. This conclusion was
further conkned by other simulations and by com-
parisons with experimental data; one example is
shown in Fig. 8. The conversion pro6les in Fig. 8 550.00 c

have been obtained in experiments with a space @> 0.00 0.50 1.00 1.50 Lfdm
velocity of 2985 h-l. Fig. 7. The model predictions compared with (a) independent
The model behaviour at different space velocities conversion and (b) temperature data: Cl, To = 576 K; A, T,,= 600
is illustrated in Fig.7 and Fig. 8 which represent K; 0, T,,=625 K; n , T,,=650 K; A, 2’,,=675 K; -, two-
parameter model; . . . . . , three-parameter model; - - -, four-
the velocities 1997 h-r and 2985 h-r respectively. parameter model. The inlet conditions were T&,co= 1.1 mol h-’
A comparison of the experimental and computed and S=1997 h-l; the inlet mole ratios were H,O/CO=6.1,
conversion profiles reveals that the plug flow model CO,/CO = 0.67 and H&O = 4.0.
28 R. L. Keiski et al. / Water-gas shzs reactor model

X
rate model (eqn. (6) Table 5) within large con-
1.00 III1 1 IllI 1 II I iI I I centration, temperature and space velocity intervals
was confirmed by several shift reactor simulations
(Figs. 3-8). The non-isothermal approach provides
a feasible method for practical kinetic studies which
are intended to produce a fast determination of the
rate parameters.

Acknowledgments
0.60

This work is part of the Computer Aided Reactor

Design (CARD) project financed by the Technology
Development Centre (ITEKES). The financial support
given to one of the authors (R. Keiski) from the
Academy of Finland, Eemil Aaltosen &ratio, Keski-
Pohjanmaan Kulttuurirahasto and Teollisen Kult-
tutu-in Tutkimussaatio is gratefully acknowledged.

0.20

References

D. S. Newsome, Catal. Rev., 21 (1980) 275.

0.00
0.00 0.50 1.00 1.50 L/&l
Fig. 8. Comparison of the model predictions with experimental
data obtained at the highest space velocity: 0, 7’,,=625 K; A,
Z’, = 650 K; 0, T,= 675 K; --, two-parameter model; . . . . . ,
three-parameter model; - - -, four-parameter model. The inlet
conditions were a,,, o=1.67 mol h-i and S=2985 h-i; the
inlet mole ratios were HaO/CO=6.1, COa/CO=O.67 and Ha/
co = 4.0.
is sufficient to describe the reactor flow pattern
since the model seems to be valid within a large
space velocity interval. The results obtained at 1997
h-l can thus be safely extrapolated to other space
velocities.
5. Conclusions
A simple fixed-bed test reactor system (Fig. 1)
was developed to obtain non-isothermal rate data
for catalytic processes. The amount of kinetic in-
formation obtained in one experiment was increased
by measuring the conversion and temperature pro-
files inside the reactor tube. The steady state ex-
perimental conversion profiles were then fitted to
a reactor model to determine the kinetic parameters
of the actual catalytic reaction. The integral least-
squares method (eqn. (11)) was applied in the
stepwise build-up of the kinetic model. The
water-gas shift reaction over an industrial ferro-
chrome catalyst was chosen to a model reaction in
the present study. The validity of the power-law
P. Mars, Chem. Eng. Sci., 14 (1961) 375.
J. M. Moe, Chem. Eng. Prog., 58 (1962) 33.
H. BoNbro, J. Catal., 3 (1964) 207.
H. Bohlbro, An Investigation on the Kinetics of the Con-
version of Carbon Monoxide with Water Vapour over
Iron based Catalysts, Gjeilerup, Copenhagen, 2nd edn.,
1969.
6 N. V. Kul’kova and M. I. Temkin, Zh. Fiz. Khim., 23 (1949)
695.
7 H. M. Hulburt and C. D. S. Vasan,AZChE J., 7 (1961) 143.
8 V. K. Puri, H. Mahapatra, R. M. Cursetji, N. C. Ganguli and
S. P. Sen, Technology {Coimbato-re, Zndia) 10 (1973) 224.
9 W. F. Podolski and Y. G. Kim, Znd. Eng. Chem., Process
Des. Dev., 13 (1974) 415.
10 P. Fott, J. Vosolsobe and V. Glaser, Coll. Czechoslov. Chum.
Comm., 44 (1979) 652.
11 D. M. Ruthven, Can. J. Che-m. Eng., 47 (1969) 327.
12 F. Goodridge and H. A. Quasi, Trans. Inst. Chem. Eng.,
45 (1967) T274.
13 A. L. Kontsevoi, B. A. Zhidkov, G. P. Chipiga and I. V.
Popik, Kin&. Katal., 14 (1976) 72.
14 H. Bohlbro, Acta Chem. Stand., 15 (1961) 502.
15 G. G. Shchibrya, H. M. Morozov and M. I. Temkin, Kin&.
Katal., 6 (1965) 1057.
16 F. P. Ivanovskii, G. E. Braude and T. A. Semenova, Kin&.
Katal., 5 (1964) 563.
17 L. M. Rose, Chemical Reactor Design in Practice, Elsevier,
Amsterdam, 1983.
18 V. Pexdir, J. Cemy and J. Pasek, Proc. 4th Eur. Symp. on
Chemical Reactor Engineering, Brussels, 1968, 239-253.
19 8. Vsjda and P. Valko, Rqn-ache, Regression proSram for
Chemical Engineers, Eurecha Manual, Budapest, 1985.
20 M. L. Michelsen, AZChE J., 22 (1976) 594.
21 J. Hoogschagenand P. Zwietering, J. Chem. Phys., 21(1953)
2224.
22 G. C. Chinchen, R. H. Logan and M. S. Spencer, Appl.
Catal., 12 (1984) 89.
G. C. Chinchen, R. H. Logan and M. S. Spencer, Appl.
Catal., I2 (1984) 69.
 R. L. Keiski et al. / Water-gas shifi reactor model 29

23 R. C. Reid, J. M. Prausnitz and T. K. Sherwood, TheProperties nco molar flow rate of CO (mol h- ‘)
of Gases and Liquids, McGraw-Hill, New York, 3rd edn., inlet molar flow rate of CO (mol h- ‘)
fi0
1977.
N number of experimental sets
24 C. Satterfield, Mass Transfer in Heterogeneous Catalysis,
Massachusetts Institute of Technology Press, Cambridge, P concentration exponent of CO2
MA, 1970. P total pressure (kPa)
25 E. L. Furen, D. V. Gemet, T. A. Semenova and M. P. 4 concentration exponent of H,
Shmigirovskaia, Kin&. Katal., 12 (1971) 1283. rate (mol gg ’ s- ‘)
26 C. Bokhoven and H. Hoogschagen, J. Chem.. Phys., 21 ii reactor tube radius (dm)
(1953) 159. gas constant (=8.314 J K-’ mol-‘)
27 H. Bohlbro and M. H. Jorgensen, Chem. Eng. World, 5
RG
rP catalyst particle radius (dm)
(1970) 46.
28 T. Salmi, L.-E. Lindfors and S. Bostrom, Chem. Eng. Sci., rP catalyst pore radius (nm)
41 (1986) 929. estimated standard deviation
29 M. Tinkle and J. A. Dumesic, J. Catal., 103 (1987) 6. i space velocity (1 hh ‘)
30 T. Salmi, S. Bostrom and L.-E. Lindfors, J. Catal., 112 S BET the BET surface area (m2 gg ‘)
(1988) 345. T temperature (K)
V volume (dm3)
3 volumetric flow rate (cm3 mm- ‘)
Appendix A: Nomenclature VP catalyst pore volume (cm3t g- ‘)
X conversion of CO
a empirical temperature coefficient (eqn. (10))
Greek letters
concentration (mol dmw3) . ..
L molecular diffusion coefficient (cm2 s- ‘) P reversrbrhty
(eqn. (5))
factor, p = ccozc Hz/GO c~zo&

D, effective diffusion coefficient (cm’ s- ‘)

EP particle porosity
DK Knudsen diffusion coefficient (cm2 s-l)
77 effectiveness factor
E activation energy (kJ mol- ‘) catalyst bulk density (g cme3)
Pb
g function defined in eqn. (11) catalyst skeletal density (g cmm3)
PC
k rate constant ((dn~~)~+~+=+~ kg-’ s-l tortuosity factor
mall-“-“-p-
9
;S Thiele modulus
ko frequency factor ((dm3)lzcm+pcq kg-’ s-r
mall --n--m-p-
“> Subscripts
KT equilibrium constant 0 inlet conditions
L reactor length coordinate (dm) co carbon monoxide
concentration exponent of H,O (eqn. (5)) i temperature coefficient index and temper-
i? parameters in eqn. (6): MHz0 = co, H20/~0,
co; ature exponent (eqn. (10))
M coz = co. co&o, co; and %z = co. H&O, co pore property
m, catalyst mass (g) FJ particle property
n concentration exponents of CO (eqn. (5)) tot total