Sie sind auf Seite 1von 10

Modelling of a fixed-bed water-gas shift reactor

1. Steady-state model verification

N. H. Bell and T. F. Edgar

Department of Chemical Engineering, University of Texas at Austin, Austin, TX 78712, USA


(Received 3 July 1990; revised 31 October 1990)

This paper describes the development and verification of a non-linear, steady-state model of a laboratory-scale
water-gas shift reactor. The objective of the work was to develop a model of the laboratory reactor for use in
simulation and model-based control strategies. Heat transfer parameters were determined from the experi-
ments without reaction. Parameter estimates for the pre-exponential factor, the activation energy and the
coefficient for heat transfer between the catalyst bed and reactor wall were determined from experiments with
reaction. Non-linear optimization with a least-squares objective function was used to determine the parameter
values. With six fitted parameters, the simplified reactor model accurately predicted the performance of the
laboratory reactor system. A second-order reversible rate expression successfully modelled the water-gas shift
reaction over a cobalt-molybdenum catalyst.

(Keywords: model; non-linear dynamics; control)

Non-linear, dynamic models of chemical tradi- Experimental system and operating procedures
used for analysis and are
increasingly being considered for on-line The water-gas shift reaction, shown below, is mildly
mization strategies. Recent in computer exothermic and reversible.
have lowered of the
to application of models. These CO(g) + H,O(g> = CO,(g) + Hz(g)
based on and principles, AH,, = - 9.8kcal mol-I
often capable predicting plant behaviour more accu-
over a range of operating than In typical industrial applications, the reaction is run in a
linear models in conventional controller The single adiabatic fixed-bed catalytic reactor or in multiple
use better models one means control adiabatic beds in series when high conversion of carbon
performance without the of monoxide is required. The process gas contains carbon
system. monoxide, carbon dioxide, hydrogen and steam. The
Fixed-bed reactors represent class inlet steam flow rate and inlet temperature (feed pre-
non-linear chemical that in heater power input) are two available manipulated vari-
controller Non-linear arises the ables for a single reactor system. Intermediate quenches
exponential of rate of the process gas could provide additional manipulated
Further modelling are due complicated variables for multiple bed operation.
heat mass transfer and the distributed struc- Disturbances that could enter a water-gas shift reac-
of fixed-bed tor system include changes in inlet gas composition, feed
A reactor constructed to a flow rate, and feedstock temperature. The inlet gas com-
non-linear system characteristics similar indus- position for this work, similar to the composition in an
trial for advanced control strategies. industrial system, was nominally 20% CO, 20% COP,
inlet disturbances can introduced so that 10% HZ, and 50% H20. Large disturbances, of the order
and ultimately algorithms could be of 50% for the inlet mole fraction of CO, CO*, or Hz,
over a range of operating conditions. The following were of interest here because of the non-linearity exhi-
include description of laboratory reactor bited by the system for such changes. These disturbances
system, discussion of the development and of cause drastic changes in the reactor performance due to
non-linear, steady-state model results from equilibrium and gas composition effects on the reaction
meter estimation experiments. In 2 of paper, we rate. Feed flow rate changes affect the performance
the effects variable catalyst and through the residence time. Feedstock temperature
mic behaviour of the reactor’. changes would normally be handled by the reactor inlet

J. Proc. 1997, Vol January

0959-1524/9l/OlOO22-10
0 1991 Butterworth-Heinemann Ltd
Modelling of a fixed-bed water-gas shift reactor: N. H. Bell and T. F. Edgar

was continually added to the gas stream to maintain


catalyst activity. The total flow rate ranged from 15 to 30
SLPM. Nitrogen was available for startup, purging and
inert blanketing of the catalyst. Complete details of the
operating conditions and experimental apparatus have
Fixed -bed been given by Bell*.
reactor
The reactor inlet temperature was maintained by
manipulating the feed preheater, which consisted of a
resistance heater in direct contact with the reactor inlet
fittings. This heater was controlled via a solid-state relay
that was switched by a computer-controlled voltage-to-
frequency converter. Inlet temperatures ranged from 290
to 350°C. The measured temperature was used in a PI
control loop to determine the power input to the feed
preheater.
The reactor consisted of two sections of 1.25 in tubing
with a 0.035 in wall. The ratio of the reactor tube dia-
meter to the catalyst particle diameter was x 9.5. Each
Caustic catalyst bed was 0.36 m (14 inches) long, and each con-
Water scrubber
knockout tained four axially distributed thermocouples positioned
Sample to
on the centre line of the reactor for observation
gas 0n0lysers purposes. The thermocouples were inserted through mul-
tiple thermocouple feed-through fittings. A thermocou-
Water
knockout ple was located at the entrance and exit of each catalyst
Figure 1 Laboratory water-gas shift reactor
bed. The second and third thermocouples in each bed
were located 0.15 m (6 inches) and 0.30 m (12 inches)
heater. Even so, the reactor performance could be affec- respectively from the beginning of the active catalyst bed.
ted because of inlet heater dynamics. Use of an Omega feed-through fitting allowed the ther-
The laboratory reactor system used in this study was mocouple positions to be changed if the catalyst was
designed to operate in a manner consistent with indus- removed from the reactor. The system was operated with
trial practice. Guard heaters were employed to reduce one bed or two beds in series, with or without secondary
radial heat losses so that the reactor operated in a near- steam injection. Insulation and heaters along the entire
adiabatic manner. The catalyst tested was a commer- length of the reactor were used to reduce heat losses.
cially available, sulphur-tolerant cobalt-molybdenum Inert packings at the inlet and exit of the reactor
WGS catalyst. A schematic diagram of the reactor provided flow distribution and physical support for the
system is shown in Figure 1. Carbon monoxide, carbon catalyst.
dioxide, and hydrogen were obtained from regulated gas The reactor effluent was sampled and analysed for
cylinders of each component. Flow of each gas was mea- carbon monoxide and carbon dioxide using infrared gas
sured and controlled by mass flow controllers. Steam analysers. This information along with knowledge of the
flow was regulated by a computer-controlled metering inlet mass flow rate of each component, allowed the
pump. The steam was generated in low volume, high complete exit composition to be determined. After
efficiency heater that allowed for quick responses to removing most of the steam in a water knock-out pot,
changes in steam demand. The heater consisted of two the effluent was passed through a 10% caustic scrubber
3.0 m (10 feet) long sections of l/16 in. stainless steel that removed hydrogen sulfide before the gas was
tubing in parallel coiled around a 750 W cartridge heater. exhausted to the atmosphere.
This assembly was placed inside a steel tube and the void The reactor system was interfaced to a distributed
spaces were packed with magnesium oxide. The maxi- computer system consisting of an IBM PC-AT that acted
mum capacity of the steam generator was 20 standard as the supervisor and an IBM PC-XT that served as the
litres per min (SLPM). slave. The XT handled data acquisition and control
A known quantity of water was vaporized to provide signal output using 4 IBM Data Acquisition and Control
steam. The water was superheated enough to prevent adapters for A/D and D/A conversion. A total of 10
condensation downstream of the generator. This was thermocouples, three mass flow rates and two gas ana-
accomplished by manipulating the power input to the lysers were monitored. Three gas flow rates, two water
generator to control the generator temperature. A flow rates and the feed preheater power input were mani-
bypass at the steam generator exit was used for startup pulated. Data were transferred between computers via an
and shutdown of the system. IEEE-488 interface. The AT was responsible for control
Carbon monoxide, carbon dioxide, hydrogen, hydro- calculations and parameter estimation. Programming on
gen sulphide, and steam were mixed and heated to the XT was done in FORTRAN with IBM DOS while a
approximately 250°C in a gas heater. Hydrogen sulphide combination of FORTRAN and C was used on the AT

J. Proc. Cont. 1991, Vol 1, January 23


Modelling of a fixed-bed water-gas shift reactor: N. H. Bell and T. F. Edgar

which ran under the XENIX System V operating system. cobalt-molybdenum WGS catalysts have been pre-
See Bell’ for more hardware and software details. sented by Newsome” and Lee’*. Newsome reported on
The cobalt-molybdenum catalyst tested has several work using a hydrotreating catalyst impregnated with an
advantages when compared with the older iron oxide and alkali metal (e.g. caesium, lithium, sodium, and potas-
copper-zinc versions. Unlike both of the older cata- sium). Lee used several commercial catalysts similar to
lysts, the cobalt-molybdenum catalyst is not poisoned the one used in this work.
by sulphur. It actually requires sulphur in the gas stream Based on the recommendation of the catalyst manu-
to remain active. Another advantage of the cobalt- facturer, (D. Brown, United Catalysts Inc., private com-
molybdenum catalyst is its wide operating temperature munication), the following rate expression9 was used to
range. model the WGS reaction in this work.
The catalyst was usually activated according to the
procedures described in the United Catalysts C-25 Oper-
ating Instruction9. This manual suggested two different (- rco>
= k (ycoymo- ‘T) (1)
methods. In both methods, nitrogen was used to purge
the reactor of oxygen prior to introduction of any reac-
tive gases. In one method, the catalyst was activated by Ampaya and Rinkefl and Bonvin9 did extensive steady-
feeding the process gas containing H2S at the same rate state and dynamic modelling of a WGS shift reactor
and concentration as those used during normal ope- utilizing an iron oxide catalyst. The reactor used in their
ration. By keeping the inlet temperature below 260°C work exhibited multiple steady-states due to the heat
and the maximum temperature below 425°C the catalyst exchange from the catalyst bed to the reactor feed; the
could be safely sulphided and reduced. Injection of addi- feed gas was passed countercurrently through an annulus
tional H2S into the process gas was suggested for increas- around the catalyst bed. They developed a two-dimen-
ing the rate of the activation process. The second sional reactor model and used orthogonal collocation to
method, called presulphiding, used a hydrogen-contain- discretize the model equations. The dynamic studies in
ing gas at an inlet temperature of 150 to 180°C. Hydro- their work did not address the effects of steam as a
gen sulphide was also used in this gas at a concentration manipulated variable.
of approximately 3 vol %. In this procedure, the gas The equations used to model fixed-bed reactors nor-
contained roughly 1.0 SLPM Nz and 0.5 SLPM HZ. The mally consist of mass and energy balances with terms for
H2S flow rate was increased until the catalyst bed temper- convection, dispersion and generation. Both axial and
atures began to increase. The maximum bed temperature radial gradients are often included. Review articles of
was maintained at less than 230°C by carefully monitor- steady-state’3 and dynamic fixed-bedI reactor modelling
ing the H2S flow rate. After a maximum temperature have been published previously. As mentioned above,
between 220 and 230°C was attained at each thermocou- Ampaya and Rinkera and Bonvin9 presented two-dimen-
ple, indicating that most of the sulphiding had occurred, sional pseudo-homogeneous models for the water-gas
the bed temperatures were increased at a rate of approxi- shift reaction. Although the above researchers used the
mately 50°C h-l, up to a maximum of 3 15°C. The temper- iron oxide catalyst, their presentations are useful starting
atures were controlled by manually manipulating the points for modelling a WGS reactor with the cobalt-
inlet, the guard and the exit heaters. This process was molybdenum catalyst.
continued for 1 to 2 h after the maximum temperature A very complete fixed-bed reactor modelling study was
had been reached. Process gas was then introduced. recently presented by Windes et al.15. This paper covered
In both methods, the catalyst would slowly become dynamic modelling of a methanol partial oxidation reac-
more active over a period of approximately 20 h of tor using one and two dimensions as well as single phase
operation. The activity can be related to the temperature (lumped gas and catalyst model-pseudo-homogeneous
history of the catalyst. Activity was seen to be pro- model) and multiple phases (separate gas and catalyst-
portional to the temperature the catalyst had exper- heterogeneous model).
ienced. In the laboratory reactor that operated at close to For this work the quasi-steady state approximation
adiabatic conditions, the increasing temperature profile for the gas concentration was used. The following dimen-
from inlet to exit resulted in increasing activity along the sionless equations make up the pseudo-homogeneous,
length of the reactor. dynamic model of the laboratory WGS reactor; the
steady-state model corresponds to the case in which the
time derivatives are zero:
Model development
Carbon monoxide balance
Most published work on the water-gas shift reaction
has dealt with iron oxide catalysts. These high-tempera-
6X
o=p- &--f$- D,(- rc0)*
ture shift catalysts were used in previous kinetic and
reactor modelling studiesk9 Rasel” presented a case study
involving iron oxide and copper-zinc catalyst beds in Catalyst bed energy balance
series. Newsomeil reviewed the literature for both iron 6Tc
6Tc
oxide and copper-zinc catalysts. Rate equations for the
’ 62T*
- a,(T* - T,*) + D(- rco)*(3)
%F= -6z*+Peh,w*

24 J. Proc. Cont. 1991, Vol 1, January


Modelling of a fixed-bed water-gas shift reactor: N. H. Bell and T. F. Edgar

Reactor wall energy balance


problem must be determined from the steady-state solu-
sI*w=s*w+6 -S2r*W+ UHWP- T*w)
ifit* w 5z*2
tion using the inlet conditions at time t* = 0.
The dynamic model consists of a set of non-linear,
- awAT*w- F*A) (4)
partial differential equations in time and the axial dimen-
Inlet heater energy balance sion. To solve the equations they were simplified to
dF ordinary differential equations using orthogonal
--J = s*, - %4(T*,- PA) - (Y,T*, - YPHT*,ff) collocation.
dt* (5)

Exit heater energy balance


Solution of the steady-state model
dTC
EH=s* - dT*E,Y- pA) - (yET*E-YEHT*EH) (6)
dt* EH
Orthogonal collocation, which uses polynomials to
approximate the true solution of a function, was used to
The equations were made dimensionless by using the obtain both steady-state and dynamic solutions. As dis-
following variables: cussed previously’G20, the solution is approximated at N
collocation points that are the roots of an orthogonal
polynomial. For a fixed-bed reactor, the points are found
on the interval of [O,l]; that is, from the inlet to the exit in
the axial direction.
The reference temperature, T,,, was the reactor inlet The following substitutions were made to remove the
temperature. The reference reaction rate, (- rco,,&, was spatial derivatives from the reactor model equations. In
based on the inlet conditions.
these equations, x can represent conversion or tempera-
Dimensionless groups:
ture.

=XC
DuGYCO.,
- rcO.refl NZ+2

AZjixi
( - rc0) i= I
(15)

(- rco)*
=( - r,,,&
NZ+Z
hw 62X
u=Gc, _=
622 IJ c
B’jiXi
(16)
1’1

NZ was the number of interior collocation points in the


axial dimension. The solution was also specified at the
SW= k, inlet and exit for the axial coordinate. Thus the total
VOPWCPWL
number of collocation points was (NZ+ 2).
hfia,L pw = SWL
The subscripts were defined as follows:
a -p
w - PWCPWVO pwc~wvoT,qf i= index on axial collocation points
j= current axial collocation point
s**= S& s* _ sEHL
Jacobi polynomials were selected and the AL and Bz
P~CPIVO Twf EH - PEHCPEHVOL~ (8) matrices were calculated by using the subroutines
provided by Villadsen and Michelsen19 (01= 0, 13= 1).
The dimensionless boundary conditions are:
Assuming that the second derivative terms were used
in the model, the collocation process produced a set of
3 x NZ non-linear equations. The values of the depen-
(9) dent variables at the boundaries were determined from
1 6x
the discretized boundary conditions. For the steady-state
x=z@ (10) problem, all of the equations were algebraic. HYBRDl,
6T*w_o
-- from the MINPACK package of Mor6 et al.22, was used
6Z* (11) to solve this set of equations.
67* 6x 6T*w_o
z*=1 6z*=6z*= 6z
(12)
p, = pEH with exit heater (13) Parameter estimation

The dimensionless initial conditions are: The ultimate goal of the model development effort was to
obtain a model that accurately predicted the behaviour
t* = 0 Tr =fi(z*),X =fz(z*),pw =f3(z*) (14) of the laboratory WGS reactor. Accuracy was measured
by the sum of the squares of the difference between
where f,, f2,and f3were known functions. For dynamic measured experimental values and corresponding values
simulation, the initial conditions for the dynamic obtained from the model. Parameters were estimated to

J. Proc. Cont. 1991, Vol 1, January 25


Modelling of a fixed-bed water-gas shift reactor: N. H. Bell and T. F. Edgar

minimize the weighted differences between experimental objective function. This program employs a successive
and model values, using a weighted least squares objec- quadratic programming algorithm.
tive function, 6, shown in Equation (17). In addition to determining optimal estimates for the
parameters, the GREG package also provided statistical
information about them. This information was in the
form of a linearized approximation of the parameter
(17) covariance matrix. The diagonal elements of the matrix
that approximated the parameter variances, helped to
determine if the data or observations contained enough
The measurements included three catalyst bed and three information to provide accurate estimates of all elements
reactor wall temperatures and the final conversion of of the parameter set. The off-diagonal elements were
carbon monoxide. used to investigate dependencies, or correlations among
Use of an objective function such as Equation (22) the parameters.
assumes that measurement errors are distributed nor- The covariance matrix V@,for a parameter vector (or
mally, that measurement errors are independent of one set) 0, was obtained from the Hessian or second deriva-
another, and that the variances of the measurements are tive of the objective function with respect to the para-
known and constant. These common assumptions were meters. For this work, the parameter vector contained a
made for convenience. The weights, or and o,, can be maximum of four elements during any one optimization
used to account for differences in scale for different mea- run.
surements and for differences in reliability. The values GREG calculated VB by using the Gauss approxima-
for the measurements in this work were dimensionless tion for H. According to Bard24
and therefore already scaled. Because the reliability of
each measurement was assumed to be similar, the value
of o for all measurements was normally set to unity. In
certain cases, however, values other than unity were used
to investigate measurement effects. The independent var-
iables including inlet temperature, pressure, and flow
rate of each component were assumed to be known The Gauss approximation for the Hessian of objective
exactly. functions based on residuals (differences between experi-
The parameters that were estimated include mental and model observations) eliminates second deri-
vatives because they contain the error terms which are
Kinetic rate parameters assumed to be small. In the equation above, fu is the uth
l A, the pre-exponential factor observation with n total observations, The standard
l E,, the activation energy deviation, a, is approximated as the standard error of the
To improve the condition of the estimation problem, a residuals as shown below.
modified pre-exponential factor, A’, was estimated
rather than A. The relationship between A and A’ is 02L
shown in Equation (21). n-l (19)

The number of degrees of freedom is n - I where I is the


Heat transfer parameters
number of parameters.
l hW, the catalyst bed-wall heat transfer coefficient
l K_,,, the reactor wall-ambient heat transfer coeffi-
cient Model verification
l Key, the end heater-ambient heat transfer coeffi-
Experiments without reaction
cient
The heat transfer and heater parameters were first esti-
mated from data obtained with no reaction occurring in
Heater variables the system. By eliminating the need for the reaction rate
l S,, the end heater power input equation, fewer parameters need to be estimated and the
The parameter estimation problem was split into two complexity of the phenomena that can occur in a fixed-
parts: the heat transfer parameters and heater input were bed reactor is reduced. For example, with reaction occur-
determined from experiments without reaction and the ring in the bed, any maldistribution of the catalyst parti-
reaction rate parameters were determined from experi- cles, variable catalyst activity, or other unmodelled dis-
ments with reaction. The bed-wall heat transfer coeffi- turbances can cause changes in axial and radial
cient was estimated from both sets of data. Predicted temperature profiles that are difficult to interpret. It is
values of the measurements were obtained by solving the therefore advantageous to obtain some parameter values
non-linear reactor model described in the previous using a simplified system.
section. The GREG (General REGression) package Unfortunately, two of the heat transfer parameters
developed by Caracotsiosz3 was used to minimize the in two-dimensional models, namely the bed-wall

26 J. Proc. Cont. 1991, Vol 1, January


Modeling of a fixed-bed water-gas shift reactor: N. H. Bell and T. F. Edgar

obtained when the end heater was ignored. The para-


Table 1 Model predictions and experimental values for bed and wall meter estimates are shown in Table 2.
temperatures for runs with no reaction (end heater model included).
For the model that includes the end heater, the shapes
Bed temperatures Wall temperatures of the axial temperature profiles near the end of the
No. 1 2 3 4 1 2 3 reactor depended upon the temperature of the end heater
1 Model 348 327 309 307 342 318 303 section. The model assumed that the reactor wall temper-
Exp. 348 329 310 307 342 318 301 ature at the exit (z = 1) was equal to the temperature of
2 Model 285 272 263 267 282 268 260 the end heater. This assumption was reasonable because
Exp. 285 274 264 267 282 269 261
4 Model 406 383 362 354 400 371 352 the stainless steel of the end heater section was in contact
Exp. 406 385 360 351 400 370 345 with the steel of the reactor wall.
The estimated values of both hw and ~~ compare rea-
sonably well with literature values. The correlation for
coefficient, hw, and the effective radial thermal conduc- the overall heat transfer coefficient for the catalyst bed to
tivity, h,, must usually be reestimated to fit data with wall suggested by Li and Finlayson26 gave a range of
reaction (vs. no reaction). Hofmann*5, Li and Finlayson26 values from 0.0023 to 0.0034 compared with the value of
and Schwedock et aLz7 have documented this effect on 0.00368 estimated from the no-reaction runs.
experimental fixed-bed reactors. Paterson and CarberryZs In an attempt to account for the change in h, with
suggested that part of this problem is due to the effect of flow rate, h,/G was estimated instead of a constant h,.
axial dispersion on the effective radial conductivity, 1,. However, in this case the standard error of the residuals
Other causes could be the changes in physical properties was 0.00425, 25% higher than the value when constant
of the gas and solid phases such as thermal conductivity h, was used. Hence this simple treatment of the depen-
associated with temperature and concentration profiles dence of h, on flow rate, which the Li and Finlayson
in the reacting catalyst bed. These profiles are usually correlation supports, did not improve the model fit for
quite different from those that occur under conditions these data. This is not surprising because Li and
with no reaction. This re-estimation of h, for the one- Finlayson reported an average error of 27% for their
dimensional model was also done in this work. correlation. The flow rates used in the parameter estima-
Although some of the parameters estimated with no tion varied by only 28% from the average value.
reaction must be checked again in the presence of reac- Table 2 shows the optimal values for K,PA and S,,
tion, it is still useful to obtain these estimates. For exam- corresponding to end heater effects. The 20 intervals
ple, good initial guesses for the heat transfer parameters were much greater for both of these parameters than for
can be obtained and their values can be compared with K~ and hw and the normalized covariance was 0.99. These

correlations. results were not unexpected because of the lack of a


The four parameters that were estimated from the no- temperature measurement for the end heater and the fact
reaction data were that increases in SE,, the heat generated, can be offset by
l h, bed-wall heat transfer coefficient increases in heat losses, Key.
0 KA,wall-ambient heat transfer coefficient
l S&V,the end heater power input Experiments with reaction
l K,cA, the end heater heat loss value
Equation (1) was initially chosen to model the reaction
The portion of the total power input to the end heater rate. In this rate model, two parameters--A, the pre-
was not known accurately a priori because only part of exponential factor and E,, the activation energy-must
this heater actually affected the reactor exit. be estimated from reactor data. The initial set of para-
A total of 13 experiments was analysed. The flow rates, meters estimated from reactor data with reaction con-
ranging from 10 to 16 standard 1min- I, were in the same sisted of these two parameters and h,, the catalyst bed-
range as those used for experiments with reaction. Ope- reactor wall heat transfer coefficient.
ration of the system without reaction was accomplished In the actual estimation problem, the rate constant, k,
by not feeding any CO. The resulting gas mixture of was reformulated*7J9JO
steam, carbon dioxide and hydrogen was similar to the
mixture used in experiments with reaction, except that
steam and carbon dioxide flow rate were generally higher Table 2 Optimal parameter estimates from experiments with no reac-
than normal to make up for the absence of carbon tion
monoxide. The inlet temperatures were set from 260 to No end heater End heater
480°C so that the catalyst bed would be subject to the
same range of temperatures experienced during runs with Parameter Estimated 20 Estimated 2a
value Interval value Interval
reaction.
We first evaluated the need to include end heater hw 0.368 x 10-Z 0.335 x 10-j 0.312 x IO-’ 0.302 x IO-’
effects. Comparison of model predictions with the data KA 0.847 x lo-3 0.296 x IO-4 0.914 x lo-‘0.350 x 10-d
GA - 0.104 x 10-l 0.354 x 10-4
(see Table I) showed that end heater effects were import- s - 0.290 x 10-I 0.939 x 10-2
ant at the lower temperatures. The standard error of the Stan%d 0.443 x 10-Z 0.345 x 10-Z
residuals was reduced by approximately 25% when the error of
residuals
end heater model was used compared with the value

J. Proc. Cont. 1991, Vol 1, January 27


Modelling of a fixed-bed water-gas shift reactor: N. H. Bell and T. F. Edgar

mance well, Table 4. The average difference between the


Table 3 Operating conditions for experiments with reaction-1st model predictions and experimental measurements for
catalyst load
conversion was 0.02 with a maximum of 0.05. Given the
Inlet Inlet mole fraction Total wide range of conversions and temperature profiles
No. temp. flow rate resulting from the operating conditions of the 10 experi-
(“C) CO H,O CO, H2 SLPM
ments, the agreement between the model predictions and
1 308.7 0.119 0.538 0.239 0.104 16.74 experimental data was remarkable. One area of concern,
2 275.7 0.127 0.507 0.254 0.112 15.76 however, was an apparent bias in the temperature predic-
3 304.4 0.127 0.508 0.254 0.112 15.76
4 313.6 0.127 0.508 0.253 0.112 15.76 tions. The model consistently overpredicted the second
5 300.0 0.181 0.476 0.238 0.106 16.82 ‘bed temperature and under-predicted the fourth or exit
6 308.7 0.240 0.474 0.237 0.050 16.87 bed temperature. These prediction errors will be
7 310.9 0.217 0.488 0.244 0.051 16.39
8 298.3 0.217 0.488 0.244 0.051 16.39 addressed in the next section.
9 318.6 0.139 0.557 0.208 0.096 14.37 The optimal parameter estimates are shown in Table 5.
10 318.2 0.139 0.557 0.209 0.096 14.37
Note that the modified value, A’, is shown rather than A,
the pre-exponential factor. All three of the parameters
were well-determined as is evident by the 20 intervals
which were less than 5% of the parameter values for A’
and E,, and approximately 17% of the heat transfer
coefficient, h,.
The parameter estimation results also indicated that
the modified pre-exponential factor is correlated with
both the activation energy and the bed-wall heat transfer
coefficient. The correlation of the rate parameters was
already reduced by the reparameterization of A as A’.
The correlation between A’ and h, is understandable
because increases in A’ corresponding to increases in the
rate of reaction could have been balanced by increases in
the radial heat loss given by h,. The ill-conditioning of
t I I I
0
the estimation problem caused by parameter correlation
0.5 I
was overcome to some extent by the information con-
z* (Normalized axial distance)
tained in the bed temperature, wall temperature, and exit
Figure 2 Steady-state model predictions and experimental obser- conversion measurements. Thus, small 2a intervals for
vations for exp. 7-first catalyst load. Model conversion = 0.64, 0, these parameters were obtained.
bed experiment; 0, wall experiment: -, bed model; ....... wall model A two-dimensional version of the model was also com-
pared with the experimental data, however, no improve-
k = A’exp s - ST) ment in the model predictions was realized. This result
( m was not unexpected given the relatively small radial tem-
where perature gradients.

A’= Aexp (-&) Comparison with other work


(21)
As mentioned earlier, Bonvin9 presented results for para-
The mean temperature, T,, around which the data are meter estimation of a water-gas shift reactor utilizing an
centred was based on the minimum and maximum reac- iron oxide catalyst. With rate parameters estimated from
tor bed temperatures. A value of 397°C for T,,, was used other kinetic experiments using the same iron oxide cata-
for each parameter estimation run presented here. lyst31, Bonvin estimated three heat transfer parameters.
Two sets of steady-state data will now be presented. The parameters were the bed-wall transfer coefficient, a
The first data set consisted of 10 experiments with reac- heater input and an overall energy loss. For seven steady-
tion. The second set was obtained using a different cata- state experiments with the three parameters re-estimated
lyst loading than for the first set. The second set, consist- for each experiment, Bonvin reported an average absol-
ing of 25 experiments, was run at conditions similar to ute difference between model and experimental conver-
the first set. sions of 0.05 with a maximum deviation of 0.06. His
carbon monoxide conversion values ranged from 0.59 to
First catalyst loading 0.81. For the work presented in the previous section the
The first set of data consisted of 10 steady-state experi- average absolute difference was 0.02 with a maximum
ments with a wide operating range. The operating con- value of 0.05. The range of conversions in the set of data
ditions for these experiments are listed in Table 3. Figure from the first catalyst loading was 0.19 to 0.69. The
2 compares the resulting model with experimental data catalyst bed temperature deviations reported by Bonvin
for a typical steady-state run. were less than those in this work.
In general the model predicted the reactor perfor- The data published by Schwedock et a1.27J2 for a

28 J. Proc. Cont. 1991, Vol 1, January


Modelling of a fixed-bed water-gas shift reactor: N. H. Bell and T. F. Edgar

Table 4 Model predictions and experimental values for catalyst bed and wall temperatures and final conversion for runs with reaction-1st catalyst
loading

Bed temperatures Wall temperatures Conv.


(“C) (‘C)
No. 1 2 3 4 1 2 3

1 Model 309 325 338 341 303 316 329 0.45


Exp. 309 320 338 348 301 312 331 0.45
2 Model 276 282 288 290 271 277 282 0.22
Exp. 276 279 287 295 272 274 286 0.19
3 Model 304 320 334 337 299 312 325 0.42
Exp. 304 315 335 346 299 307 328 0.45
4 Model 314 334 350 354 308 325 340 0.50
Exp. 314 331 354 364 308 322 344 0.55
5 Model 300 322 346 352 296 314 336 0.40
Exp. 300 316 346 365 297 308 339 0.41
6 Model 309 355 415 425 306 345 400 0.62
Exp. 309 342 408 428 306 336 387 0.61
7 Model 311 355 408 416 308 345 393 0.64
Exp. 311 348 413 429 309 343 392 0.66
8 Model 298 329 369 379 295 321 358 0.49
Exp. 298 320 360 384 295 314 352 0.46
9 Model 319 355 384 388 315 345 372 0.70
Exp. 319 338 377 387 313 334 369 0.69
10 Model 318 354 383 387 314 344 371 0.69
Exp. 318 338 376 385 312 334 366 0.69

Table 5 Optimal parameter estimates from experiments with reac- Table 6 Optima1 parameter estimates from experiments with reac-
tion-1st catalyst loading tion-2nd catalyst loading

Parameter Estimated value 20 Interval Parameter Estimated value 2a Interval

A’ 0.202 x 10-1 0.110 x 10-d A 0.171 x 10-1 0.126 x 10-r


E. 0.140 x 10s 0.571 x 10’ E, 0.919 x lo” 0.114 x 101
h, 2.51 x 10-l 0.424 x 10-l h, 0.287 x lo-* 0.592 x 10-3
Std. error of residuals 0.142 x 10-l Std. error of residuals 0.359 x 10-I
T, = 397°C T, = 397°C

methanol oxidation reactor showed results for differ-


dock et ~1.32,Li and Finlaysonz6, and Paterson and Car-
ences between model predictions and experimental
berryzs. However, this comparison is not necessarily
values for conversion and temperatures that are similar
valid because the data without reaction were obtained
to the differences found in this work. They estimated the
with a different catalyst loading than was used for these
radial thermal Peclet number, Biot number, two pre-
experiments with reaction. A change in the catalyst load
exponential factors, and the activation energy for metha-
can affect several factors, most important of which are
nol oxidation in a two-dimensional model. WindesiQ3,
probably the catalyst distribution and the distribution of
using a heterogeneous, two-dimensional model, obtained
energy output from the reactor guard heater. In this case,
less deviation for temperatures on the same methanol
the change was made to replace catalyst damaged by
oxidation reactor by additionally estimating temperature
steam condensation.
and flow rate dependencies of the radial thermal Peclet
number and the tortuosity factor. Windes’ improved
results for temperature predictions might have been due Second catalyst loading
to the constant reactor wall temperature that resulted To further investigate the laboratory reactor perfor-
from use of a circulating heat-transfer oil. This may have mance, the catalyst used to obtain the data presented
given a more consistent wall-effect in the experimental previously was removed and replaced with a fresh
data than that resulting from the use of guard heaters in charge. The catalyst was activated in the same manner as
our research. Our desire to operate the laboratory water- the first loading.
gas shift reactor at near-adiabatic conditions precluded This data set consisted of 25 experiment9. The opti-
the use of a reactor with a constant wall temperature. mal parameter estimates are shown in Table 6. A typical
The optimal value of h,, the bed-wall heat transfer comparison of model predictions and experimental data
coefficient, determined from these experiments with reac- is shown in Figure 3. The discrepancies between the
tion, was 2.51 x 10e3 compared with the value of 3.12 x model and experimental data were more evident in this
1O-3 estimated from data without reaction. The decrease set of data. This could have been due to different activa-
in h, for runs with reaction compared with runs without tion procedures, a different operating history, or the
reaction is the opposite of results published by Schwe- different catalyst loading. Further experimentation using

J. Proc. Cont. 1991, Vol 1, January 29


Modelling of a fixed-bed water-gas shift reactor: N. H. Bell and T F. Edgar

the laboratory WGS reactor presented in this paper


450
accurately predicted the steady-state conversion of car-
bon monoxide at the reactor exit over a wide range of
operating conditions. Unfortunately, discrepancies
400
between the experimental catalyst bed temperatures and
2
e the model predictions for some data sets could not be
2 explained by experimental error.
: 350
g
I-”
References
300
1 Bell, N. H. and Edgar, T. F. J. Proc. Cont. 1990, submitted for
publication
I
2 Bell, N. H., ‘Steady-state and Dynamic Modeling of a Fixed-Bed
I I
Water-Gas Shift Reactor’. PhD Thesis. Universitv of Texas. Austin.
0 0.5 I
USA, 1990
z * (Normalized axial distance) United Catalysts, Inc, Post Office Box 32370, Louisville, KY 40232,
USA, United Catalysts C-25 Operating Instructions
Figure 3 Comparison of model predictions and experimental data Moe, J. M. Chem. Eng. Prog. 1962,58(3), 33
using constant catalyst activity rate expression for the second catalyst Ruthven, D. M. Can.-J. Chem. Eng. 196$,47, 327
loading. Model conversion = 0.77; experimental conversion = 0.75; Podolski. W. F. and Kim. Y. G. Ind. Ena. Chem. Proc. Des. Dev.
symbols as in Figure 2 1974,13(4), 415
Singh, C. P. P. and Saraf, D. N. Ind. Eng. Chem. Proc. Des. Dev.
high temperatures showed that the errors were due to 1977,16(3), 313
variable catalyst activity. Ampaya, J. P. and Rinker, R. G. Chem. Eng. S’ci. 1977,32, 1327
The difference between the experimental bed and wall Bonvin, D., ‘Dynamic Modeling and Control Structures for a
Tubular Autothermal Reactor at an Unstable State’. PhD Thesis.
temperatures at the first and second thermocouple University of California, Santa Barbara, USA, 1980
locations was much less than for the first catalyst load- 10 Rase, H.. F. in ‘Chemical Reactor Design For Process Plants’,
ing. Possible reasons for this included different guard Volume 2 John Wilev and Sons. New York. USA. 1977
11 Newsome, D. S. C&l. Rev.- Sk Eng. 198d,21(2j, 275
heater and insulation effects and different catalyst distri- 12 Lee, A. L. Effect of raw product gas Composition on shift catalysts.
butions which could have been caused by the catalyst Technical Reuort Proiect 61015. DOE contract DE-AC21-
reloading procedure. The inlet heater position could also 78ET11330, I&t. of Gas-Tech., Chicago, USA, 1980
13 Froment, G. F. Proc. Fifth European/Second Int. Symp. on Chemi-
have changed and affected the difference between the cal Reaction Eng., Elsevier Science Publishers, Amsterdam, 1972,
inlet bed and inlet wall temperatures. Catalyst distribu- A5-1
tion changes could have affected the bed-wall tempera- 14 Ray, W. H. Proc. Fifth European/Second Int. Symp. on Chemical
Reaction Eng., Elsevier Science Publishers, Amsterdam, 1972, A8-1
ture difference by causing different flow distributions. 15 Windes, L. C., Schwedock, M. J. and Ray W. H. Chem. Eng.
Channelling would cause the flow rate to be higher near Comm. 1989,78(l), 43
the wall than in the interior of the bed, resulting in higher 16 Villadsen, J. V. and Stewart, W. E. Chem. Eng. Sci 1967,22, 1483
17 Finlayson, B. A. in ‘The Method of Weighted Residuals and Varia-
carbon monoxide conversion in the region of channell- tional Principles’, Academic Press, New York, USA, 1972
ing, and thus more energy production and higher tem- 18 Finlayson, B. A. Catal. Rev.- Sci. Eng. 1974, 10(l), 69
peratures near the wall. 19 Finlayson, B. A. in ‘Nonlinear Analysis in Chemical Engineering’,
McGraw-Hill Inc., New York, USA, 1980
The one-dimensional model with Equation (1) as the 20 Villadsen, J. and Michelson, M. L. in ‘Solution of Differential
reaction rate was only partially successful at predicting Equation Models by Polynomial Approximation’, Prentice-Hall,
the performance of the reactor bed for the second data Inc., Englewood Cliffs, New Jersey, USA, 1978
21 Michelsen, M. L. and Villadsen, J. The Chemical Engineering Jour-
set. For the 25 experiments, the average difference nal, 1972,4,64
between model and experimental conversions was 0.033 22 Mart, J. J., Garbow, S. and Hillstrom, K. E., User guide for
with a maximum difference of 0.09. Conversion differ- minpack-1 Technical Report ANL-80-74, Argonne National
Labbratory, USA, 1980 _
ences greater than 0.04 all occurred during the final two 23 Caracotsios. M. ‘Model Parametric Sensitivitv Analvsis and Non-
days of the time period during which this set of data was linear Parameter Estimation -Theory and Ap&icati&‘, PhD The-
sis, University of Wisconsin, Madison, USA, 1986
collected. These larger errors might have resulted from a
24 Bard. Y. in ‘Nonlinear Parameter Estimation’. Academic Press,
change in catalyst activity or an analyser calibration NewYork, USA, 1974
error. 25 Hofmann, H. German Chemical Engineering 1979,2,258
While exit conversion predictions were quite accurate, 26 Li, C. H. and Finlayson, B. A. Chemical Engineering Science 1977,
32, 1055
catalyst bed temperature predictions did not follow the 27 Schwedock, M. J., Windes, L. C. and Ray, W. H. ‘Steady state and
experimental temperature profile for many of the experi- dynamic modeling of a packed bed reactor for the partial oxidation
ments. Most noticeable were the large discrepancies for of methanol to formaldehyde-Experimental results compared
with model predictions’, Presented at the AIChE Annual Meeting,
the catalyst bed and wall temperatures at z* = 0.43. As Chicago, USA, November 1985.
mentioned previously, the difference between the experi- 28 Paterson, W. R. and Carberry, J. J. Chemical Engineering Science,
mental bed and wall temperatures was less than expected 1983,38 (l), 175
29 Bates, D. M. and Watts, D. G. in ‘Nonlinear Regression Analysis
at this location. and Its Applications’, John Wiley and Sons, New York, USA, 1988
30 Van Damme, P. S., Narayanan, S. and Froment, G. F. AIChE
Journa/ 1975,21 (6), 1065
Conclusions 31 Lee, R. W. ‘Dynamic Behavior of an Autothermal Reactor with
Internal Counter-Current Heat Exchange’, PhD Thesis, University
With six fitted parameters, the one-dimensional model of of California at Santa Barbara, USA, 1978

30 J. Proc. Cont. 1991, Vol 1, January


Modelling of a fixed-bed water-gas shift reactor: N. H. Bell and T. F. Edgar

32 Schwedock, M. J., Windes, L. C. and Ray, W. H. Chem. Eng. aw transfer coefficient, catalyst bed to wall (cal/cm2/s/K)
Comm. 1989,78,45 aWH transfer coefficient, wall to heater in wall balance (Cal/
33 Windes, L. C. ‘Modeling and Control of a Packed Bed Reactor’, @/s/K)
PhD Thesis, University of Wisconsin at Madison, USA, 1986. axw transfer coefficient, catalyst bed to wall in heater balance
(cal/cnV/s/K)
transfer coefficient, heater to ambient (cal/cm+/K)
transfer coefficient, inlet heater to ambient (cal/cm+/K)
Nomenclature transfer coefficient, exit heater to ambient (cal/cm*/s/K)
catalyst bed void fraction
preexponential factor &moles CO/g cat/s) dimensionless sensible heat term
reparameterized pre-exponential factor &moles CO/g reactor-ambient heat transfer term (Cal/cm heater/s/K)
cat/s) exit heater heat transfer term (Cal/cm) heater/s/K)
collocation matrices inlet heater heat transfer term @al/cm3 heater/s/K)
area for heat transfer from catalyst bed to reactor wall etTective radial conductivity (Cal/s/cm/K)
per unit wall volume viscosity (g/cm/s)
concentration of CO (g cm)) objective function
gas heat capacity (Cal/g/K) bulk density of catalyst (g cm’)
catalyst heat capacity (Cal/g/K) wall density (g cm-l)
Damkoehler number inlet heater density (g cn-3)
catalyst particle diameter (cm) exit heater density (g cm-J)
tube diameter (cm) objective function weights
axial diffusivity (cm*/s) standard deviation
activation energy (cal/gmole/K)
functionals refers to ambient conditions
gas mass velocity (g/cmQ) refers to experiment
heat of reaction (cal/gmole) refers to exit heater
transfer coefficient at wall (cal/cm*/s) refers to gas
equilibriumconstant . ’ refers to inlet conditions
reaction rate constant (gmoles CO/g cat/s) refers to mass or model
effective axial thermal conductivity (Cal/cm+) refers to energy
reactor length (cm) refers to guard heater
Lewis number refers to inner dimension or inlet heater
average molecular weight of gas (g gmol) refers to pre heater
number of experiments refers to outer dimension
number of interior axial collocation points reference value
partial pressure or percentage heater on-time pressure
axial thermal Peclet number refers to solid catalyst
axial mass Peclet number refers to reactor wall
inner reactor radius (cm) axial
gas constant (1.987 cal/gmole/K) Superscripts
reaction rate of carbon monoxide (gmoles CO/cm+) * dimensionless value
heat generation term for guard heater (Cal/cm+)
heat generation term for inlet heater (cal/cm+)
heat generation term for end heater (cal/cm’/s)
particle surface area Physical dimensions and properties
reactor bed temperature (K)
time (s) Catalyst heat capacity, cp 0.28 Cal/g/K
ambient temperature Catalyst particle density, pp 1.05 g/cm’
exit heater temperature Catalyst particle size, D, 0.125 in
reactor wall temperature at z = 1 Axial diffusivity, D, 0.0243 g/set/cm
inlet heater temperature Heat of reaction, AH,,,,, - 9.8 kcal/gmole
preheater temperature temperature Effective axial thermal conductivity, k. 0.00208 Cal/cm+
guard heater temperature (K) Reactor length, L 35 cm
mean temperature for rate reparameterization (K) Reactor wall thickness 0.035 in
reactor wall temperature (K) Inner reactor radius, R 3.0 cm
interstitial gas velocity (cm/s) Gas constant, R, 1.9871 cal/gmole K
particle volume Heat generation term for guard heater, S,, 0.001 cal/cm’/s
overall heat transfer coefficient at wall (cal/cm+/K) Heat generation term for end heater, S,, 0.002 Cal/cm+
carbon monoxide conversion Heat generation term for inlet heater, S, 0.0 13 Cal/cm+
mole fraction of species i Ambient temperature, T, 27°C
axial position (cm) Mean temperature for rate reparameterization, TIT? 397°C
thermal Damkoehler number Catalyst bed void fraction, E 0.34
dimensionless wall thermal conductivity Wall density, pv 7.92 g/cm’
dimensionless heater thermal conductivity Wall thermal conductivity, k, 0.05 Cal/cm/s

J. Proc. Cont. 1991, Vol I, January 31

Das könnte Ihnen auch gefallen