See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/222641214

Modeling of Fixed Bed Catalytic Reactors

Article  in  Computers & Chemical Engineering · December 1985

DOI: 10.1016/0098-1354(85)80028-4

CITATIONS READS
21 2,407

3 authors, including:

Osvaldo M. Martínez Norberto O. Lemcoff

Universidad Nacional de La Plata Rensselaer Polytechnic Institute
81 PUBLICATIONS   700 CITATIONS    67 PUBLICATIONS   1,182 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Municipal solid waste incineration in fluidized bed. International Cooperation CONICET-CNRS View project

Selective hydrogenation View project

All content following this page was uploaded by Norberto O. Lemcoff on 10 December 2017.

The user has requested enhancement of the downloaded file.

Co,,,pum~ and Chemrcal Engmeerq Vol. 9. No. 5. pp 535-545. 1985 0098&13S4/85 $300 + .oO
Pnnted in Great Britam C; 1985 Per&mm Press Ltd

MODELING OF FIXED BED CATALYTIC REACTORS

0. M. MARTINEZ and S. I. PEREIRA DUARTE

Departamento Ingenieria Quimica, Universidad National La Plats, 1900 La Plata, Argentina
and
N. 0. LEMCOFF+
PINMATE, Departamento de Industrias, Facultad de Ciencias Exactas y Naturales, Ciudad
Universitaria, 1428 Buenos Aires, Argentina

(Received 1 February 1984; revision received 15 November 1984; received for publication 29 May 1985)

Abstract-The simulation of a fixed-bed catalytic reactor requires the selection of a model, which is a set
of balance equations that describes the reactor, as well as correlations for the model parameters involved.
In this work general criteria, leading to a better choice of a model that fulfills the objectives of the simu-
lation, are established. Different ways in which the parameters can be obtained are analyzed, and the
numerical methods for solving the model equations are discussed.

Scope-For the simulation of a fixed bed catalytic reactor, it is necessary to establish the model that best
represents the reactor behaviour. The adequacy of a model will depend on the simulation objectives, namely
design, control, optimization, scale-up, etc.
Two ways of modeling can be differentiated: o priori and (I posteriori. Wilhelm [I] set forth the status
of LIpriori designs of fixed bed catalytic reactors by recognizing the role of transport phenomena in the
reactor performance. In a priori modeling, both kinetics and transport parameters are measured indepen-
dently and then brought together without adjustment and replaced into the model. In a posteriori modeling,
parameter values (activation energy, preexponential factor, effective radial conductivity or wall heat transfer
coefficient) are suitably adjusted by comparing them with experimental results.
Once the model and the corresponding parameters are established, it is necessary to determine a
mathematical method that could be applied to solve the model equations. The compromise between the
simulation objectives, precision of the results and computing time must generally be analyzed.
Several aspects involved in the modeling of fixed bed reactors are analyzed. Conditions for the use of
different models and methods for the evaluation of the parameters and for solving the model equations
are discussed with the view of establishing a methodology for the simulation of a fixed bed catalytic reactor.

Conclusions and Significance-Models are developed in order to represent the actual behaviour of a system.
The degree of complexity of a model depends on the simulation objectives. The models for a fixed bed
catalytic reactor can generally be broken up into a kinetic model, representing the reaction at the catalyst
active site, a pellet model, representing the intraparticle phenomena, and the reactor model itself. In general,
the kinetic model is empirical and the other two can be derived from basic principles. Although the pellet
and reactor models are interconnected, they can be studied separately. In this work and taking into account
this division, only reactor models are considered, without analyzing kinetic or pellet models.
Different types of reactor models have been proposed in the literature. If dispersion-type reactor models
are considered, eight models can be defined, whether or not axial, radial dispersion and differences between
fluid- and solid-phase conditions are taken into account.
With the present knowledge it is very difficult to set up an a priori reactor model. A model is normally
developed in stages, and, therefore, it is important to establish a criterion for reducing the iterative process
involved. From the analysis carried out in the present work, steps toward the development of standard
routine packages to be included in a process simulator have been taken. According to the characteristics
of the reaction system and the operating conditions, the convenience of using a certain model can be
established, and, therefore, the simulator should select the most appropriate model.
Just as important as the model selection is the existence of reliable values of the parameters involved.
The difference that exists between elaborated and simple models is, in many cases, of the same order of
magnitude as the difference generated by the uncertainty of the parameters. It follows that more precise
correlations to evaluate heat transfer parameters are needed. An analysis of the effect of data dispersion
on the model predictions was carried out. Similarly, the effect of the chemical reaction on the heat transfer
coefficient, when the heat transfer to the surroundings is evaluated from parameters obtained in experiments
without chemical reaction, is also analyzed.
The different reactor models analyzed in the present work generate sets of ordinary, partial differential
and/or algebraic equations. Several numerical methods proposed in the literature are briefly discussed.
Although most of them are adequate enough for solving steady-state and dynamic problems, one of the
limitations may be the speed of solution when complex models are incorporated into on-line state estimation
and control algorithms.

‘Author to whom correspondence should be addressed.

535
536 0. M. MARTINEZ et al.

1. INTRODUCTION model, representing the intraparticle phenomena, has

to be included.
The concept of reactor modeling is normally used in
Although in many cases the basic knowledge nec-
a rather broad sense by different authors, but it can
essary to derive an approximation for both the pellet
be described as a method for translating existing in-
and reactor models is available, this very seldom is
formation and data into useful predictions for new
true for the kinetic model, and therefore we have to
conditions [2]. Such predictions might involve
accept that reaction rate expressions are empirical
(a) better understanding of the system, which may correlations.
lead to improvements in the design; In some cases the influence of the pellet phase var-
iables is expressed in terms of an effectiveness factor,
(b) prediction of the system’s dynamics for pur-
poses of control; whose calculation may be more or less complicated,
Cc) optimization of steady-state operating condi- ranging from the solution of simple algebraic equations
tions; to the solution of coupled nonlinear boundary value
(d) analysis of the effect of different feedstocks, problems. In all circumstances one should try to reduce
catalysts and reaction conditions on product the complexity of the pellet model as far as possible,
distribution; since the reactor model is in itself quite complicated.
(e) scale-up from pilot plant to a large reactor. In such a way the influence of the reactor macro var-
iables can be studied with sufficient accuracy without
In the process of modeling it is assumed that there too many details in the pellet phase description.
exists a true mathematical relation among the output Although modeling is normally presented as a
variables and all the input variables. In many fields of straightforward process, actually it is an iterative pro-
engineering, such as distillation and electronic equip- cess involving considerable judgement. The final model
ment, a model can be developed that closely describes is not set up immediately, but it is developed in stages.
the actual system. However, owing to the complexity In the first stage a simple model is considered, and its
of the process, in chemical reaction modeling the true parameters are evaluated. From the comparison be-
relationship is in most cases inaccessible and therefore tween experimental and calculated concentration and
it is necessary to settle for an approximate description. temperature profiles, the performance of the model is
In spite of this, chemical reactor modeling and, specifi determined. If the fitting is not adequate, a more com-
tally, fixed bed catalytic reactor modeling have made plex model has to be analyzed. But models developed
tremendous progress in the last 30 years [3]. Perhaps in succesive stages may involve contribution of addi-
the main limitation is the existence and accuracy of tional or more detailed transport processes; therefore,
experimental data. Any prediction that can be made the evaluation of the corresponding parameters has to
will be as exact as the physical and chemical data that be considered. As a matter of fact, our purpose is to
are introduced in the model. When there is not enough be able to reduce this iterative process by determining
data available, once the model is proposed, the whole (I priori the more appropriate model for the reactor.
effort has to be concentrated on a more precise esti- Although one of the goals of reactor modeling can
mation of the parameters in already known models. be the prediction of the reactor dynamics, the addition
The desired accuracy in the knowledge of the param- of the nonsteady term introduces extra difficulties, in
eters depends, to a certain extent, on their influence addition to those already described. Therefore, the
on the overall process. In a general way, it can be present analysis is restricted to the steady-state case.
established that more accurate parameters are needed Recently, some fundamental work on chemical reactor
when their influence is greater. modeling for the purpose of control has been carried
Another important limitation may reside in the ex- out [5,6].
istence of mathematical methods for solving the model
equations. In order to represent all the phenomena
occurring in a fixed bed catalytic reactor, it may be
2. FIXED BED REACLOR MODELS
necessary to develop a very complex model. However,
if this model cannot be handled mathematically, that Only continuous-type models are considered, since
is, if no calculation techniques are available, the model they have been used more frequently than cell models
does not seem to be adequate [4]. Therefore, the model [3,7,8]. Therefore, the simplest model is the pseudo-
has to be reduced in complexity by making use of the homogeneous one-dimensional model without axial
processing objectives so that it can be solved in a dispersion. Interfacial gradients are neglected, and heat
reasonable way. However, the simplification of the and mass balances are written for a single pseudo-
model must be done realistically so that it still rep- homogeneous phase (Table 1). At the other extreme,
resents the actual reactor closely. a heterogeneous two-dimensional model with axial dis-
Shinnar [2] proposed that the chemical reactor persion of heat and mass can be written. Separate heat
model be broken up into a kinetic model which relates dispersion terms for each phase are considered (Table
reaction rates to local conditions, and the reactor 2). Traditionally, heat dispersion terms were lumped
model itself, which describes transport processes and into the fluid phase, and only recently has the right
heat and mass balances in the reactor. However, when approach been taken [9- 131. The boundary condition
dealing with a heterogeneous catalytic reactor, a pellet for the heat balance equation at the reactor wall is
 Modeling of fixed bed catalytic reactors 537

Table 1. One-dimensional pseudohomogeneous model for the evaluation of these parameters [l 11,not very
reliable values are available for some of them.
(without axial dispersion)
d(u, c, There are some additional phenomena that may
A- = - q r,<
dZ have to be considered in establishing a reactor model:
d(u, pp C, 7) pressure drop in the fixed bed, nonuniform velocity
=(-AH)?p-+ Tw)
dZ profile, variable porosity, dependence of certain prop-
Initial conditions erties on temperature and composition, etc. However,
c= c,
T= T,, I @Z=O
we consider these secondary phenomena that can even-
tually be incorporated in any of the basic models.
In this analysis we have separated the kinetic and
pellet models, since they can be incorporated into any
of the basic models. It is assumed that the reaction
generally expressed in terms of an additional resist- rate can be evaluated, given uniform temperature and
ance. Some authors have proposed, instead of a wall concentration values at the pellet surface. It has been
heat transfer coefficient, a lower value of the radial shown that this assumption does not introduce sig-
effective thermal conductivity close to the wall [14]. nificant errors in the evaluation of the reaction rates,
On the other hand, boundary conditions for the at least when temperature gradients inside the pellet
system with axial dispersion depend on the regions are not important [16,17].
that precede and follow the reactor [11,15]. In Table In order to optimize both the effort and time
2, those corresponding to the system, including the needed, it is obvious that the most adequate model for
pre- and postreactor regions, are shown. representing a system is the simplest possible one,
We can establish, in principle, eight basic reactor which considers all the phenomena that are important
models, four with and Four without axial dispersion, in the system. Therefore, it is necessary to analyze
which correspond to the combination of the one- and whether criteria can be established to determine apriori
two-dimensional versions of the pseudohomogeneous which phenomena have to be considered.
and heterogeneous models (Fig. 1). These basic models Comparisons between one- and two-dimensional
are related among themselves, and, under simplifying models have been carried out [18] in order to obtain
assumptions, the simpler models can be derived from a qualitative criterion to limit the region of the vari-
the more complex ones. ables space where each model should be used. It was
The simplest model that we consider involves a found that the performance of a one-dimensional
single heat transfer parameter, the overall heat transfer model depends mainly on the maximum (? - 7’,) and
pseudohomogeneous coefficient. The more complex ;jl. Similar results have been obtained by Hlavacek and
model involves, when a single reaction is considered, Marek [19]. It follows that a one-dimensional model
seven heat transfer and three mass transfer parameters. is recommended for a reactor operating under low
Although correlations are available in the literature thermal sensitivity conditions.

Table 2. Two-dimensional heterogeneous model

Radial boundary conditions:

5 _ 0; !$$Lo at r = 0 all Z
ar -
dc, _ 0
ar -

-A@Jr = a/,(T,- T,) at r = R all Z

-A:.$ = a:(T, - T,)

Axial boundary conditions:

c,= c, r, = T, = To all R Z - - 00
C,is finite T, = T, = T, allR Z-+m
538

Fig. 1. Fixed-bed reactor models and heat transfer parameters

In order to select between pseudohomogeneous and behaviour of a natural gas steam reforming reactor
heterogeneous models, it is necessary to evaluate the using a pseudohomogeneous one-dimensional model
importance of interphase gradients. Mears [20] and and studied the effect of the heat transferred from the
Gonzo and Gottifredi [21] developed criteria valid for furnace. However, Li and Finlayson [25] found, from
a single pellet. However, this cannot be applied to a published data, an uncertainty of &25% in the heat
reactor unless local conditions are known. Recently, transfer coefficient. It follows from Fig. 2 that the
Pereira Duarte and Lemcoff [ 181 developed a criterion estimation of the reformer tube lifetime varies by al-
for predicting the deviations between the models as a most one order of magnitude, according to the value
function of the inlet conditions. Although the criterion used for this coefficient, and therefore has an important
was developed for a single reaction, it could be ex- economic significance.
tended to more complex reactions schemes by an ad- Another important aspect in nonadiabatic reactors
equate lumping. is the hot-spot value, since too high temperatures can
When the axial dispersion is negligible, pseudo- sinter the catalyst or lead to runaway. The maximum
homogeneous models can be applied for a ratio d,/d, axis and mean temperature rise are shown in Table 3
> 10. For smaller ratios, the decision will depend on for different inlet concentrations. The results in the
the reaction sensitivity to the interfacial gradients, first column correspond to a pseudohomogeneous two-
which in turn depend on the adiabatic temperature dimensional model, the second to a pseudohomoge-
rise, activation energy, reaction order, difference be- neous one-dimensional model and the third to a simple
tween inlet and wall temperatures, etc., and can be approximate equation that has recently been proposed
evaluated from the mentioned criterion. [26]. In the last two cases, in order to evaluate the
It is also important to determine under which con- axis temperature, a radial parabolic temperature profile
ditions axial dispersion is negligible, since this com- has been assumed [27]. The fourth column corresponds
plicates to a large extent the numerical solution of the
model equations. Fortunately, for most of the com-
mercial reactions its effect is small, since Re > 100
[22,23].
Therefore, it follows from the analysis presented
here, that the information available permits the a priori
selection of the more appropriate reactor model.

3. TRANSFER PARAMETERS IN MODELING

OF PACKED BED REACTORS
Once a model has been selected, the parameters
involved have to be evaluated. We can differentiate
between those corresponding to the pellet and those
to the reactor. In what follows, the heat transfer pa-
rameters corresponding to the reactor will be analyzed. L.-L_-_ _l___

Their importance in reactor modeling and the accuracy 0 05 1.0Thfdl%hl‘m, 1.5

with which they should be known are reflected in Fig. 2. Influence of the heat transfer coefficient on the reactor
several examples. Demicheli et al. [24] simulated the tube lifetime [24].
 Modeling of fixed bed catalytic reactors 539

Table 3. Maximum axis and mean temperature rise (d,/d, = 12.5, n = 1, y0 = 20,
Re = 500, To = T, = 64310

Two-dim.
P AT Two-dim. One-dim. Prediction (modified parameter)

0.4 AT 42.14 41.02 42.16 32.83

0.4 AL 70.22 64.84 66.59 54.14
0.5 AT 63.48 59.61 57.94 47.63
0.5 AL: 108.14 94.5 91.52 79.94
0.6 AT 96.43 85.85 75.71 69.0
0.6 AT,,,, 168.08 135.18 119.6 118.0

to results obtained with the same model as in the first action, that is, from heat exchange experiments. As a
column with all the parameters remaining constant, matter of fact, this is the method most commonly used.
except the effective radial conductivity and the wall Although more reliable data seem to arise, a certain
heat transfer coefficients. According to the uncertainty care has to be taken when these data are incorporated
found by Li and Finlayson [25], these are increased into the reactor model owing to the influence of the
by 25%. For the conditions in Table 3 and using the chemical reaction on the parameters of the simpler
criteria developed by Van Welsenaere and Froment models. This aspect can be better understood by ana-
[28] and by Rajadhyasksha et al. [29], the limit for lyzing the relationship between the parameters of the
parametric sensitivity is found for fl = 0.58. It can different models.
be seen that the difference that exists between elabo-
rated and simple models is, in this case, on the same
5. RELATIONSHIP BETWEEN HEAT TRANSFER
order of magnitude as the difference generated by the
PARAMETERS
uncertainty of the parameters. Although in certain
cases results arising from simple models are qualita- In order to understand better the physical meaning
tively adequate for understanding the reactor’s behav- of the parameters of the simpler models, it is conve-
iour, more complex models may be needed under ex- nient to determine the relationship between the dif-
treme conditions. However, in order to obtain reliable ferent reactor models. Several authors analyzed this
quantitative results, accurate values of the parameters problem in the last decades [9,11- 131. As an example
involved are necessary. let us analyze the pseudohomogeneous one-dimen-
sional heat transfer coefficient. Traditionally, it has
been obtained from experiments without chemical re-
4. DETERMINATION OF HEAT TRANSFER
action by measuring the inlet and exit mean temper-
PARAMETERS
atures [34,35]. In this way a mean coefficient for the
Let us briefly analyze the different ways in which whole bed is evaluated. However, from the analysis of
heat transfer parameters can be evaluated. a pseudohomogeneous two-dimensional model, Li and
(1) From correlations. This is perhaps the simplest Finlayson [25] establish that the coefficient depends
and cheapest way, since a number of correlations are on the bed length, although for sufficiently long beds
available in the literature [3,30]. However, the disper- it reaches an asymptotic value. Similarly, from the
sion of the data is so large that much care must be analysis of the relationship with the heterogeneous
taken in choosing a particular correlation. Previous two-dimensional model, a way of determining the
experience can help in reaching a decision. If a reactor asymptotic value of the coefficient has been developed
that is operating has to be modeled, the correlation [36, 371. Dixon and Cresswell [ 111 defined the param-
that produces a better fitting of the results should be eters of the heterogeneous two-dimensional model as
chosen. Hyman [31] modified already published cor- basic parameters, and, therefore, we shall try to es-
relations in order to get a better fitting. This is ques- tablish their relationship with the parameters of the
tionable, and the resulting correlation is certainly valid other models. Generally, the relationship is established
only for a limited range of the variables. for systems without chemical reaction and this is the
(2) From experimental measurements. When both case to be analyzed.
transfer and kinetic parameters are simultaneously ob-
tained by regression of experimental results from a 5.1 Heterogeneous and pseudohomogeneous
system with reaction, a good fitting can be expected two-dimensional models
[32]. However, a large interdependence of the param- In order to find the equivalence between these
eters can exist, and extrapolation of the parameter models, the pseudohomogeneous temperature has to
values can be dangerous. In many cases parameter be defined. If we analyze the convective term in the
adjustment may be necessary, but it is important that energy balance equation, it has to be set equal to the
as many parameters as possible be independently mea- fluid temperature; but if we analyze the reaction term,
sured, since otherwise a high parameter cross-corre- it has to be set equal to the solid temperature. One of
lation may render such adjustment physically meaning- the first papers in this area is that by Olbrich [9], who
less [33]. Alternatively, the heat transfer parameters obtained explicit expressions for the solid and fluid
can be determined in a system without chemical re- temperatures. Later, Dixon and Cresswell [11] in-
540 0. M. MARTINEZet al.

chided the axial dispersion term and obtained the re- 5.3 Pseudohomogeneous one-dimensional
lationship between the parameters from the matching and heterogeneous two- dimensional models
of approximate solutions of the fluid phase and pseu- The overall coefficient of the pseudohomogeneous
dohomogeneous temperature. The expression for the one-dimensional model can be related to the basic
effective axial conductivity was objected to by Vort- parameters of the heterogeneous two-dimensional
meyer and Berninger [38] since, in a previous work, model in two ways: through the heterogeneous one-
Vortmeyer and Schaefer [15] found a different rela- dimensional model or through the pseudohomoge-
tionship between the axial dispersion coefficients of neous two-dimensional model.
the pseudohomogeneous and heterogeneous models. For a heat exchanger, when the heterogeneous and
In addition, they conclude that the pseudohomoge- pseudohomogeneous one-dimensional models are com-
neous temperature should be equal to the solid one. pared, the following equation can be obtained [37]:
The difference between the results can perhaps be at-
tributed to the fact that Dixon and Cresswell analyze
a steady-state case, whereas Vortmeyer and Schaefer a, = j-$ + a_i,
a nonsteady state case [39]. It can be concluded that
the meaning of the pseudo homogeneous temperature
where a: and a{ are evaluated from Eqns (5) and (6).
is not yet clearly established.
The second way, that is, the evaluation of the over-
If the solid, fluid and pseudohomogeneous tem-
all heat transfer coefficient from the parameters of the
peratures are assumed to be equal, a simple relation-
pseudohomogeneous two- dimensional model, has al-
ship between the parameters arises:
ready been discussed in the literature [25]. The heat
transferred to the external media is set to be equal in
h,,=h:,+ hL, (1)
the heat exchanger case, and the following equation
a, = a; + a’,. (2) arises:

This is a reasonable assumption, since the pseudohom-

a, = a, J,(X,) x, (8a1
ogeneous model is applicable when the temperature
2J,(X,) ’
difference between phases is not very large.

where X, is the first eigenvalue of

5.2 Heterogeneous one- and two-dimensional
models
The correct one-dimensional version has recently Bi J,(X,) = XJ,(X,). (8b)
been proposed by Pereira Duarte et al. [13], and if we
set the heat transferred to the surroundings in both The same result was obtained by Crider and Foss [41]
models equal, that is if using a model with axial dispersion, since its inclusion
does not affect the expression for the heat transfer
a{(T, -- T,) = a’, CT& - T,), (31 coefficient. In any case, the equations of a pseudo-
homogeneous two-dimensional model without chem-
a: (7, -- TJ = a: (T& - TJ, (4) ical reaction can be solved, and an infinite series arises.
Except in a region close to the inlet, only the first term
we can obtain the one-dimensional coefficients of the in the series is important, and therefore the asymptotic
heterogeneous model: value of the coefficient can easily be obtained.
In order to avoid the root search, an approximate
a;= 5, (5)
explicit equation has been proposed:
4
1
a’=
a:
-.
-= ‘+ + (9)
R, a, a, er

In order to obtain R, and R,, the expressions for For a radial parabolic temperature profile a value
the radial variation of the solid and fluid temperatures of b = 4 arises. Similarly, when a one-point collocation
are required. Those derived by Olbrich [9] can be used method is applied, b = 3 or b = 4 is obtained, ac-
(see Appendix), and asymptotic values of the one- cording to the orthogonal polynomial used [42]. On
dimensional heat transfer coefficients arise. Alterna- the other hand, in searching for the minimum differ-
tively, expressions obtained by Dixon and Cresswell ence between the values of the one-dimensional heat
[l l] for a system with axial dispersion can be used. transfer coefficient arising from Eqns. (8) and (91, Cri-
These are not affected by the axial dispersion and der and Foss [41] found a value of b = 3.0665.
simply represent a limiting case [40]. This is a result From the simulation of heat exchange experiments,
of having considered a single collocation point in the using a heterogeneous two-dimensional model, Pereira
radial direction, since radial parabolic temperature Duarte et al. [37] determined that Eqn (7) gives the
profiles arise for both the solid and fluid phases, each best values for the one-dimensional heat transfer
one determined by its respective Biot number. coefficient.
 Modeling of fixed bed catalytic reactors 541

6, USE OF THE HEAT TRANSFER PARAMETERS chemical reaction. It can also be seen that for a heat
IN A CHEMICAL REACTOR MODEL exchanger Qg = 0 and Eqn (11) simplifies to Eqn (7).
The values of a, obtained from Eqn (10) or (11)
When a model is proposed, the driving forces will
by numerically solving the heterogeneous two-dimen-
be average values, and the corresponding coefficients
sional model equations have been plotted in Figs 3
will be defined in such a way that they represent a
and 4. A first-order irreversible reaction has been con-
series of mechanisms, and are therefore called effective
sidered, as well as the fact that the pseudohomoge-
values. Since effective heat transfer coefficients are gen-
neous temperature of the one-dimensional model is
erally obtained in a system without chemical reaction,
equal to the radial average fluid temperature of the
it is necessary to analyze whether they still represent
heterogeneous two-dimensional. The deviations of the
exactly the heat transferred to the surroundings when
curves from unity are a measure of the error introduced
there is heat generated or consumed in the solid phase
in the heat transfer term when the asymptotic overall
by the reaction.
heat transfer coefficient arising from experiments with
Let us consider the heat transferred to the sur-
no chemical reaction is used (a,,). The singularity
roundings according to a heterogeneous two-dimen-
found at a certain reactor length for an exothermic
sional model and a pseudohomogeneous one-dimen-
reaction is similar to that found by Sundaram and
sional model:
Froment [43] for empty tubular reactors. These sin-
gularities are due to the fact that the mean fluid and
a,@ - 7-J
wall temperatures coincide at a certain axial position.
= aLCT,l, - TJ + a’,(T,I, - TJ. (10) This depends on the ratio T,,/T,, and the value of the
heat of reaction.
On the other hand, from the equations of the het- It follows that, in order to express the heat trans-
erogeneous one-dimensional model, we can derive an ferred correctly, a variable coefficient should be used
expression for the overall heat transfer coefficient [37]: in the simpler model.
The differences between the pseudohomogeneous
one-dimensional model and the heterogeneous two-
a, = a{+ -- a: R Qg&/(l + 69
(11) dimensional model are due to the facts that (i) the
l + 65 + 2 CT,- T,) ’
reactor rate is evaluated at the pseudohomogeneous
where temperature and not at the solid temperature, (ii) the
reaction rate is evaluated at radially averaged temper-
ature and concentration values, and (iii) the heat trans-
fer to the surroundings is represented by a single coef-
ficient and expressed in terms of the radial mean
Pereira Duarte et al. [37] have shown that a< and temperature. When variable heat transfer coefficients
a:, defined in Eqns (3) and (4), are affected by the are used, only the last aspect is considered, and, by

2.5,
I I

O(ia
2. -

0 01 0 02 0 05 01 02 05 1. 2. 5.
_z_ lo.

LR
Fig. 3. Deviation of the one-dimensional heat transfer coefficient with respect to the asymptotic value
(To < TJ. 1. IpI = 0.5; 2. IpI = 1.5. exothennic reaction, - - - - - endo-
thermic reaction, - . - - - . - - heat transfer (no reaction).
542 0. M. MARTINEZ et al.

3.5
&

(xi,
3.

2.5

2.

1.5

I.

0.5

I.0 1 0.02 0.05 0.1 0.2 0.5 1. 2. 3. z

LR
Fig. 4. Deviation of the one-dimensional heat transfer coefficient with respect to the asymptotic value
(T, = T,,). 1. IpI = 0.5; 2. IpI = 1.5. exothermic reaction, - - - - - endo-
thermic reaction.

no means, are all the differences between the models models of fixed bed reactors by taking into account
eliminated. In Table 4 the results of simulations using the different factors that generate deviations. Mc-
both models are compared, and the discrepancies in Greavy and Turner [44] took into account the effect
conversion and temperature at the hot spot are cal- of the reaction by using a variable one-dimensional
culated. First, a variable coefficient in the one-dimen- heat transfer coefficient. Since the expressions obtained
sional model, arising from Eqn (lo), is considered. were empirically adjusted, no general validity can be
Also, an asymptotic heat transfer coefficient evaluated claimed. Later Ahmed and Fahien [45] set up an im-
from Eqns (5)-(7) and (A3)-(Al?), and a simple proved model by accepting variable radial profiles of
expression arising from Eqns (l), (2) and (9), with temperature and concentration. Heat and mass balance
b = 4, are used. As expected, the best results are equations are written at three radial positions, and
obtained when variable coefficients are used, and larger very good results, compared with a two-dimensional
discrepancies are found at more extreme conditions. model, are obtained. These papers represent the ap-
The asymptotic coefficients normally produce higher plication of interesting ideas and show a useful ap-
temperature values, since they are the lower boundary proach toward obtaining good responses with low
(Figs 3-4) and constitute, therefore, a conservative computing time.
value [25]. The coefficient arising from the sum of the One final aspect that remains to be analyzed is
parameters can adequately be used under mild con- whether the basic parameters of the heterogeneous
ditions. The use of variable coefficients has been pre- two-dimensional model are modified by the chemical
sented in order to allow a better understanding of the reaction. As more detailed models are proposed, the
system, but in the way they were obtained here, they parameters involved will represent more elementary
are not of any practical use. Very few authors have mechanisms, and the effect of chemical reaction is
intended to improve the response of one-dimensional expected to become completely isolated.

Table 4. Differences between the pseudohomogeneous one-dimensional and the heterogeneous two-
dimensional models (d,/d, = 10, n = 1, y0 = 15, Re = 500, T0 = 673K)

Asymptotic coeffi-
Variable coefficients Sum of parameters
cients

ToI Tw P e,(%) AT e,(n) AT e,@) AT

1.0466 0.5 0.28 0.18 0.53 0.62 0.47 0.52
1.0466 1.0 -3.9 - 3.0 2.5 6.1 -4.3 -8.2
1 0.5 -0.32 -0.058 2.3 4.0 0.24 -0.45
1 1.0 -4.0 -4.0 2.2 8.1 -4.3 - 11.4
0.9573 0.5 0.047 0.14 2.1 5.2 0.71 - 0.023
0.9573 1.0 -3.4 -3.3 2.6 12.5 -2.9 - 11.0
 Modeling of fixed bed catalytic reactors 543

7. NUMERICAL METHODS TO SOLVE MODEL for representing steady- state problems, in the simu-
EQUATIONS lation of reactor dynamics, more complex models may
be needed [6]. If the final objective is to implement
The fixed bed catalytic reactor has been treated on-line state estimation and control algorithms, effi-
here as a hypothetical nonisotropic continuum, where cient methods of solution have to be used so that
heat and mass dispersion are represented, respectively, computing times are smaller than the real times in-
in a form analogous to Fourier and Fick laws. Sets of volved. These are not referred to here, since we have
ordinary or partial differential and algebraic equations restricted this analysis to the steady-state case.
arise, and several mathematical methods of solution
are available in the literature [27,46-481. Acknowledgement-The authors are grateful to the Consejo
The simplest model considered, the pseudohomo- National de Investigaciones Clientificas y Tecnicas, Argen-
tina, for their financial assistance, which made this work
geneous one-dimensional model, leads to a set of or-
possible.
dinary differential equations of the initial-value type.
The number of equations involved is equal to N + I,
where N is the number of independent components.
NOMENCLATURE
For the corresponding heterogeneous model, N + 1
additional algebraic equations are needed. Finlayson external particle surface area per unit
[27] presented a comparison of the different methods reactor volume
for integrating ordinary differential equations, based A, constants in Eqn (11)
on the accuracy, stability, tendency to oscillate and b constant in Eqn (9)
b, constants in Bqn (Al)
work effort and related to function and Jacobian eval-
Bi,,Bi ,,Bi Biot numbers for the fluid phase (a{. RI
uations and matrix decompositions. Although the final A!,), solid phase (a; R/h:,),
choice depends on the particular goal and the problem pseudohomogeneous (a! + a;) RI
to be solved, the general criteria are to use implicit (A’,, + A:,)
methods for stiff problems and explicit methods for reactant molar concentration
specific heat of fluid
nonstiff problems. In the present work a fourth-order
pellet diameter
Runge-Kutta-Gill method was used. axial and radial mass dispersion
When the axial dispersion is considered, two-point, coefficients
boundary-value, nonlinear differential equations arise. e, percentual conversion difference
Two-dimensional models lead to sets of nonlinear el- E activation energy
hl heat transfer coefficient between fluid and
liptic or parabolic partial differential equations, de-
pellet
pending on whether or not axial dispersion is taken (-AH) heat of reaction
into account. For the heterogeneous two- dimensional I<>,1, modified Bessel functions of the first kind
model with axial dispersion, N + 2 elliptic equations of order 0, 1
plus N algebraic equations are involved, whereas for Jo. J, Bessel functions of the first kind of order
0,1
the corresponding homogeneous model, N + 1 elliptic
k, mass transfer coefficient between fluid and
equations are needed. pellet
In principle, two different kinds of methods can be L bed leneth
used to solve partial differential equations. One dis- Lll referenci length, u,p,C,R / 2a,,
eigenvalues in Eqn Al
cretizes the differential equations in both the axial and
; dimensionless parameter, h,a,L’/h,,
radial direction (finite-difference, global collocation or Pe Peclet number, I&, pIR /XL,
Galerkin methods).
Q* radial mean value of heat generated per
Finlayson [42] compared the solution with a two- unit bed volume
dimensional catalytic reactor model and concluded radial coordinate
rate of reaction per unit bed volume at the
that, especially for stiff systems, a collocation method
particle surface conditions
is faster and more accurate than an implicit finite- R tube radius
difference method. Similar results were obtained by RX gas constant
Mihail and Iordache 1491, who also found that an Rk parameters defined in Bqns (3)-(6),
explicit scheme is the most sensitive to step size and r, -T,)I(Tk,, - T,)
Re Reynolds number, d,u,p, Iv,
that its results deviate from the other two solutions.
T temperature (K)
Recently, Hlavacek and Van Rompay [50] reviewed u* superficial fluid velocity
the numerical methods for simulation of reactor prob- x dimensionless axial coordinate, Z/R
lems and suggest solving elliptic equations by applying XI first eigenvalue in Eqn (8b)
orthogonal, double collocation or false transient meth- dimensionless radial coordinate, r IR
k axial coordinate
ods, while the Crank-Nicolson finite-difference Greek Symbols
scheme and orthogonal collocation are reliable for a a,. af, a: one-dimensional heat transfer coefficient
great deal of reactor problems involving parabolic dif- on the bed side: lumped, for fluid phase,
ferential equations. In the present work the Crank- for solid phase
wall heat transfer coefficient: lumped, for
Nicolson technique was used, since stiff problems were
fluid phase, for solid phase
not dealt with. dimensionless adiabatic temperature rise,
Although simple models may be adequate enough (-hH)COlp,C,To
544 0. M. MARTINEZ et al

dimensionless activation energy, (R I&T) 16. K. B. Bischoff, Effectiveness factors and temperature
dimensionless parameters defined in Eqns distributions for catalyst particles in non-uniform envi-
(A51 and (A61 ronments Chem. Engng Sci. 23, 451 (1968).
dimensionless parameter, (2af IR )I h,a, 17. V. Hlavacek, & M. Kubicek, Modeling of chemical re-
void fraction of packing actors-XX. Heat and mass transfer in porous catalyst.
effectiveness factor for solid particle The particle in a non-uniform external field. Chem. Engng
effective thermal conductivity with respect Sci. 25, 1527 (1970).
to axial direction: lumped, for fluid phase, 18. S. I. Pereira Duarte & N. 0. Lemcoff, Analysis of fixed
for solid phase bed catalytic reactor models, ACS Symposium Series no.
effective thermal conductivity with respect 237, p. 239, ACS, Washington, D.C. (1984).
to radial direction: lumped, for fluid 19. V. Hlavacek & M. Marek, Modeling of chemical reactors.
phase, for solid phase II. Tubular non-isothermal, non-adiabatic packed bed
fluid phase viscosity reactor; problems in the design of a reactor with radial
fluid density heat transfer for the reaction A + B. Collection Czech.
Subscripts and superscripts Chem. Comun. 32, 3309 (1967).
asymptotic value 20. D. E. Mears, Tests for transport limitations in experi-
7 fluid phase mental catalytic reactors, Indust. Engng Chem. Process
0 inlet condition Des. Develop. 10,541 (1971).
s solid surface or solid phase 21. E. E. Gonzo & J. C. Gottifredi, Rational approximations
w at the wall of effectiveness factor and general diagnostic criteria for
_ radial mean value heat and mass transport limitations, Cat. Rev. Sci. Engng
25, 119 (1983).
22. L. C. Young & B. A. Finlayson, Axial dispersion in
nonisothermal packed bed chemical reactors. Indust.
REFERENCES Engng Chem. Fundam. 12, 412 (1973).
23. D. E. Mears, On criteria for axial dispersion in noni-
1. R. H. Wilhelm, Progress towards the “a priori” design sothermal packed bed catalytic reactors, Indust. Engng
of chemical reactors. Pure Appl. Chem. 5, 403 (19621. Chem. Fundam. 15, 20 (1976).
2. R. Shinnar, Chemical reactor modelling-The desirable 24. M. Demicheli, 0. A. Ferretti & M. A. Laborde, Heat
and the achievable, in Chemical Reaction Engineering transfer effect on the behaviour of a natural gas steam
Reviews, ACS Symposium Series, No. 72, p. 1. Wash- reformer. Latinoam. J. Heat Mass Transfer7, 281 (1983).
ington (1978). 25. C. Li & B. A. Finlayson, Heat transfer in packed beds.
3. L. Lapidus & N. R. Amundson, Eds., Chemical Reactor A reevaluation. Chem. Engng Sci. 32, 1055 (1977).
Theory, a Review. Prentice-Hall, Englewood Cliffs, NJ 26. S. I. Pereira Duarte, 0. M. Martinez & N. 0. Lemcoff,
(1977). Prediccidn de1 hot spot en reactores cataliticos de lecho
4. D. M. Himmelblau & K. B. Bischoff, Process Analysis fijo. XIIdas. Jornadas sobre Investigaciones en Ciencias
and Simulation: L)eterministic Systems. John Wiley, New de la Ingenieria Quimica y Quimica Aplicada, Tucuman,
York (1968). Argentina (1983).
5. R. Shinnar, Chemical reactor modeling for purposes of 27. B. A. Finlayson, Nonlinear Analysis in Chemical Engi-
controller design. Chem. Engng Commun. 9, 73 (19811. neering. McGraw Hill, New York (1980).
6. L. C. Windes, M. J. Schwedock & W. H. Ray, Dynamic 28. R. J. Van Welsenaere & G. F. Froment, Parametric
models for the control of a highly exothermic packed sensitivity and runaway in fixed bed catalytic reactors.
bed reactor. AlChE Meeting, Los Angeles, CA, (Nov. Chem. Engng Sci. 25, 1503 (1970).
1982). 29. R. A. Rajadhyasksha, K. Vasudeva & L. K. Dorais-
7. H. A. Deans & L.. Lapidus, A computational model for wamy, Parametric sensitivity in fixed bed reactors. Chem.
predicting and correlating the behaviour of fixed bed Engng Sci. 30, 1399 (1975).
reactors: I. Deviation of model for nonreactive systems. 30. G. F. Froment & K. B. Bischoff, Chemical Reactor Anal-
AIChE J. 6, 656 (1960). ysis and Design. John Wiley, New York (1979).
8. H. A. Deans & L.. Lapidus, A computational model for 31. M. H. Hyman, Simulate methane reformer reactions.
predicting and correlating the behaviour of fixed bed Hyd. Proc. 47, 131 (1968).
reactors: II. Extension to chemically reactive systems. 32. G. Emig, H. Hofmann & H. Friedrich, Ermittlung kin-
AIChE .I. 6, 663 (1960). itischer parameter aus integralen labordaten, in Proc. 5th
9. W. E. Olbrich, A two phase diffusional model to describe European Congress in Chemical Reaction Engineering. El-
heat transfer processes in a non-adiabatic packed tubular sevier, Amsterdam (19721.
bed, in Proc. Chemeca 70 Conference, p. 101, Butter- 33. W. R. Patterson & J. J. Carberry, Fixed bed catalytic
worths, London (1970). reactor modeling. The heat transfer problem. Chem.
10. A. P. De Wasch & G. F. Froment, A two-dimensional Engng Sci. 38, 175 (1983).
heterogeneous model for fixed bed catalytic reactors. 34. M. Leva, M. Weintraub, M. Grummer & E. L. Clark,
Chem. Engng Sri. 26, 629 (197 11. Cooling of gases through packed tubes. Indust. Engng
11. A. G. Dixon & D. L. Creswell, Theoretical prediction Chem. 40, 747 (19481.
of effective heat transfer parameters in packed beds. 35. H. Verschoor & G. C. A. Schuit, Heat transfer to fluids
AIChE J. 25, 663 (1979). flowing through a bed of granular solids. Appl. Sci. Res.
12 S. I. Pereira Duarte, G. F. Barreto & N. 0. Lemcoff, AZ, 97 (1952).
Comparison of two-dimensional models of fixed bed cat- 36. T. Wellauer, D. L. Cresswell & E. J. Newson, Heat
alytic reactors, Chem. Engng Sci. 39, 1017 (19841. transfer in packed reactor tubes suitable for selective
13. S. I. Pereira Duarte, 0. A. Ferretti & N. 0. Lemcoff, A oxidation. ACS Symposium Series 196, paper No. 41, p.
heterogeneous one-dimensional model for non-adiabatic 257 (1982).
fixed bed catalytic reactors. Chem. Engng Sci. 39, 1025 37. S. I. Pereira Duarte, 0. M. Martinez, 0. A. Ferretti &
(1984). N. 0. Lemcoff, Coeficiente de transferencia de calor un-
14. M. Ahmed & R. W. Fahien, Tubular reactor design- idimensional seudohomogeneo para un reactor cata litico
I. Two-dimensional model. Chem. Engng Sci 35, 889 de lecho fijo. XIIdas. Jornadas sobre Investigaciones en
(1980). Ciencias de la Ingenieria Quimica y Quimica Aplicada,
15. D. Vortmeyer & R. J. Schaefer, Equivalence of one- and Tucuman, Argentina (1983).
two-phase models for heat transfer processes in packed 38. D. Vortmeyer & R. Berninger, Comments on the paper:
beds: One-dimensional theory. Chem. Engng Sci. 29, 485 Theoretical prediction of effective heat transfer param-
( 1974). eters in packed beds. AIChE I. 28, 508-525 (1982).
 Modeling of fixed bed catalytic reactors 545

39. D. L. Cresswell & A. G. Dixon, Reply to comments. Except close to the inlet, only the first term is important
AIChE J. 28, 511 (1982). and the equation depends only on the first eigenvalue. There-
40. S. I. Pereira Duarte, N. 0. Lemcoff, 0. M. Martinez, 0. fore, it follows that
A. Ferretti, Coeficientes de transferencia de calor para
un modelo heterogtneo unidimensional, in Proc. 1.x Con-
T,- T,
greso Latinoamericano de Transferencia de Calor y Ma- R, =
T ,,R - T,
feria, Vol. 2, p. 1011 (1982).
41. J. E. Crider & A. S. Foss, Effective wall heat transfer
= ZJ,(y,,)ly,, + b, I,(y,,)iy,,l
coefficients and thermal resistances in mathematical
models of packed beds. AIChE J. 11, 1012 (1965). [Jo(y,,) + b, Myz,)] ’ (A3)
42. B. A. Finlayson, Packed bed reactor analysis by or-
thogonal collocation. Chem. Engng Sci. 26, 1081 (1971).
43. K. M. Sundaram & G. F. Froment, A comparison of T>- Tw
simulation models for empty tubular reactors Chem. R, =
T,,, - T,
Engng Sci. 34, 117 (1979).
44. C. McGreavy & K. Turner, Model reduction of fixed ~[(Y:I - N,) (J,(yr,)Iy,,) - b, (r:r + N,) (I,(yzr)~y>,)]
=
bed catalytic reactors. Canad. J. Chem. Engng 48, 200
[(Y:, - N,) Jo(r,,) - b, (r:, + N,) Io(yz,)l
(1970).
(A4)
45. M. Ahmed, & R. W. Fahien, Tubular reactor design-
II. A modified one-dimensional model. Chem. Engng Sci. where
35, 897 (1980).
46. J. Villadsen & M. L. Michelsen, Solution of Differential

’ %
y:, = _ (N, + N, - n:)
Equation Models by Polynomial Approximation. Prentice-

+
Hall, Englewood Cliffs, New Jersey (1978). 2

) 1
47. J. Villadsen & M. L. Michelsen, Numerical Methods in N, + N, - n:
Reaction Engineering. Danmarks Tekniske Hojskole, + N, n: , (A5)
[( 2
Lyngby, Denmark (1982).
48. W. D. Seider, Model and algorithm synthesis in process
analysis and design. Conferencia international sobre nue-
vos desarrollos hacia tecnologias de bajo consume y:, = (N, + NJ - n!)
energetico, Santa Fe, Argentina (1983).
2
49. R. Mihail & C. Iordache, Performances of some numer-
ical techniques used for simulation of fixed bed catalytic
reactors. Chem. Engng Sci. 31, 83 (1976). + (~6)
[(
50. V. Hlavacek & P. Van Rompay, Current problems of
multiplicity, stability and sensitivity of states in chemi-
cally reacting systems. Chem. Engng Sci. 36, 1587 (1981).
b = YII JI(yII) - Bi, J,(Y,,)
I (A71
y21 II(yII) + Bi,I,(y,,)

APPENDIX and n, is the first root of

Analytic expressions of the solid and fluid temperature

yII Ji(rr,) - Bi, JO(rll)
profiles for a system without reaction can be obtained from
the equations of the heterogeneous two-dimensional model ( yzI I, (y2r) + Bi,I, (rl,) 1
without axial dispersion by assuming that the axial temper-
ature variations are similar in both phases [9] (~8)

T,- T,
p=
To- T,

+ h, I,(Y,, ~11 (Al)

Jo(Y,, Y)

_ b (Y,, +- N,) -n’x

,--I,(y,, y) exp* G42)
N. 1

View publication stats