Sie sind auf Seite 1von 11

Chemical Engineering Science 59 (2004) 2253 2263

www.elsevier.com/locate/ces

A robust iterative method of computing e%ectiveness factors


in porous catalysts
Dong Hyun Kim , Jietae Lee
Department of Chemical Engineering, Kyungpook National University, Daegu 702-701, South Korea
Received 19 August 2003; received in revised form 29 December 2003; accepted 15 January 2004

Abstract

A robust, e0cient numerical method for computing the e%ectiveness factor of a heterogeneous reaction in a catalyst is developed in this
study. The method is based on shooting at the outer surface of the catalyst and is optimized for an accurate estimation of the concentration
gradient at the outer surface. The shooting at the outer surface, however, is inherently unstable for a cylindrical or a spherical catalyst,
and it also can become unstable with an increasing Thiele modulus even for a slab catalyst. From an analysis of the governing equation,
however, three criteria are developed to make the method work e0ciently in the presence of instability. The numerical method is shown
to be e%ective for diverse kinetic expressions, from simple power-law to sophisticated LangmiurHinselwood kinetics, and for isothermal
or non-isothermal catalysts. The method is also shown to be easily applicable to a more complex case of multiple steady states, estimating
all the corresponding e%ectiveness factors. In this method, a simple rule that determines the stability of each steady state is proposed.
? 2004 Elsevier Ltd. All rights reserved.

Keywords: E%ectiveness factor; Shooting method; Intraparticle di%usion; Heterogeneous reaction; Multiple steady states; Stability

1. Introduction can be found in Aris (1975) and in Schneider (1976). Also,


a recent book (Wijnggaarden et al., 1998) gives a summary
The reaction rate in a porous catalyst is a%ected by in- on various formulae of di%ering complexities and their ap-
traparticle mass and heat transfer in addition to the intrinsic plications.
kinetics. The ratio of the observed reaction rate to the rate Approximation formulae relate the e%ectiveness factor
in the absence of intraparticle mass and heat transfer resis- with the Thiele modulus . For very large or small values of
tances is de:ned as the e%ectiveness factor. The factor can the modulus, all formulae accurately predict the asymptotic
be directly measured experimentally or rather commonly, to behavior of the e%ectiveness factor, but at moderate  val-
predict the observed rate, it is estimated from the solution ues, their predictions often di%er more than 20% from the
of the intraparticle mass and heat balance equations which exact values. Furthermore, for LangmuirHinselwood ki-
are formulated as di%erential equations with boundary con- netics and/or non-isothermal catalysts, approximation of the
ditions. Except for an isothermal :rst-order reaction and a e%ectiveness factor becomes much di0cult, and often yields
zero-order reaction, the balance equations are non-linear and unsatisfactory estimates. As present-day computers become
are usually solved numerically to calculate the e%ectiveness more powerful, numerically solving the balance equations
factor. Since the numerical solution of the problem is re- can be a viable alternative to the approximations.
garded as tedious and time consuming, approximation of The usual numerical methods for the boundary-value
the e%ectiveness factor has been extensively investigated in problem are the :nite-di%erence methods, the shooting
the past and various simple approximate formulae are avail- methods (Carnahan et al., 1969) and the orthogonal col-
able in textbooks (for example, Froment and Bischo%, 1990; location methods (Villadsen and Michelsen, 1978). When
Fogler, 1999; Hill, 1977). In-depth mathematical theory with the problem is non-linear, the methods become necessar-
a comprehensive list of references on the e%ectiveness factor ily iterative ones, :nding an improved solution based on
the results of the previous iterations with a prospect that
Corresponding author. Tel.: +82-53-950-5617; the iterative procedure will lead to the desired solution.
fax: +82-53-950-6615. The :nite-di%erence method converts the boundary-value
E-mail address: dhkim@knu.ac.kr (D.H. Kim). problem to a system of non-linear algebraic equations, the

0009-2509/$ - see front matter ? 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2004.01.056
2254 D.H. Kim, J. Lee / Chemical Engineering Science 59 (2004) 2253 2263

solution of which can be very di0cult to obtain, especially relation in the catalyst
when many base points are used in the method. For a large i De1
Thiele modulus, the concentration pro:le developed in the Ci Cis = (C1 C1s ); i = 2; : : : ; n (2)
Dei
catalyst is steep and the number of base points can be ex-
cessively large since a very small step size is required to is essentially linear, that is, De1 =Dei does not appreciably de-
represent concentration derivatives accurately by the :nite pend on the concentration pro:le developed inside the cata-
di%erence (Chang, 1982). In many cases the non-linear lyst and may be replaced by the ratio evaluated for the bulk
equations are linearized and solved iteratively but without composition and temperature at the outer surface of the cat-
a good initial guess, the iteration may fail to converge. alyst. The linear relationship among the individual species
Orthogonal collocation methods also have similar di0- allows to express the reaction rate in the catalyst, a function
culties as the methods involve solutions of a system of of the n species concentrations, in terms of the key compo-
non-linear equations. The collocation methods are e0cient nent concentration C1 . When the catalyst is non-isothermal,
when successful, but they are often unstable when many the temperature inside the catalyst can be related to C1 as
collocation points are used and the Thiele modulus is large (IH1 )De1
(Shiraishi et al., 1995). The shooting methods convert the T = Ts + (C1s C1 ): (3)
ke
boundary-value problem into an initial-value problem, in
Using this equation enables us to write the non-isothermal
which the missing boundary condition at the initial point is
reaction rate in terms of the concentration C1 .
assumed. Through an iterative procedure, the methods try
We de:ne the following:
to produce a solution that agrees with all the given bound- 
ary conditions (Miyakawa et al., 1999). When the catalyst C1 R(C1 ; T ) R(C1s ; Ts )
center is chosen as the initial point, the methods can only y= ; f(y) = ; =L ;
C1s R(C1s ; Ts ) C1s De1
be used with the condition that the dependent variable,
usually the concentration of the key reactant, at the center (4)
is not zero, since zero initial conditions can only generate a where  is the Thiele modulus and L is the radius of the
trivial solution. The concentration of the key reactant can, sphere or cylinder, or the half-thickness of the slab. The
in fact, become zero at the center when the Thiele modulus dimensionless equation describing the mass transport of the
is large and the apparent reaction order is less than one. On key component in a porous catalyst is
the other hand, if the catalyst outer surface is chosen as the
d2 y s dy
initial point, the resulting initial-value problem becomes + 2 f(y) = 0; (5)
inherently unstable due to a singularity at the catalyst center d x2 x dx
unless the catalyst is of a slab shape, and hence generation where s is the shape factor of the catalyst (s=0 for an in:nite
of a solution that matches the boundary condition at the slab, s = 1 for an in:nite cylinder, and s = 2 for a sphere). It
center is very di0cult. is assumed that the concentration and the temperature at the
The purpose of this study is to develop a robust method outer surface of the catalyst are prescribed. The boundary
for computation of the e%ectiveness factor without short- conditions are
comings of the previous methods. In this regard, we exam- dy
ine the problems associated with the shooting method and y=1 at x = 1; =0 at x = 0: (6)
dz
propose an algorithm based on that method, which exploits
the inherent instability of the method and enables e0cient The e%ectiveness factor  is given by

computation. s + 1 dy 
= : (7)
2 d x  x=1

2.1. A remark on f(y)


2. Mass transfer in a porous catalyst
The formal range of y is [0,1]. During numerical compu-
A general reaction involving n chemical species can be
tation of Eq. (5), however, the range is not often respected,
written as
particularly when numerical instability occurs. Obviously,
n
 numerical results violating the range cannot be accepted as
i Ai = 0; (1) a solution of the equation and are discarded, but the results
i=1 can be used to improve the solution in iterative procedures.
Before arriving at an acceptable solution, the computed y
where i is the stoichiometric coe0cient of the species Ai . may deviate considerably from the formal range, and thus
We choose a key reactant and denote the reactant as A1 . we need to make a provision for y 1 and y 0. We may
Using a modi:ed MaxwellStefan equation to formulate de:ne f(y) outside the range as
the multicomponent di%usion in the catalyst, Schneider
(1976,1998) has shown that the mutual concentration f(y) = 1 for y 1; f(y) = 0 for y 6 0: (8)
D.H. Kim, J. Lee / Chemical Engineering Science 59 (2004) 2253 2263 2255

The :rst condition for y 1 arbitrarily equates the rate to condition, the method tries to produce a solution which gives
the rate at the bulk surrounding the catalyst and may make dy=d x = 0 either at x = 0 if y(0) 0 or at some 0 x 1
f(y) continuous, and the second condition prevents f(y) where y = 0. The resulting initial problem, however, often
becoming negative. This de:nition will simply aid the nu- becomes inherently unstable, that is, its solution is very sen-
merical computation to arrive at the :nal solution which is sitive to the initial guess as x moves from 1 to 0. We may
within the permitted y range (0 6 y 6 1). call this method reverse shooting, in contrast to the shoot-
Numerous kinetic studies have reported reaction rates ing at x = 0.
with negative reaction orders (for an example, see Jiang et We let z = 1 x and rewrite Eq. (5) in terms of z, then
al., 1993). In this case, as y decreases to zero, the reac- d2 y s dy
tion rate f(y) increases without bound, which is physically 2
= + 2 f(y); 0 z 1: (9)
dz 1 z dz
unrealistic and a source of error such as division by zero
in numerical computations. In reality, every reaction ceases The boundary conditions are
when there is no reactant, y = 0, and the reaction rate is :- 
dy 
nite. The negative reaction order should be interpreted as an y = 1 at z = 0; = 0: (10)
dz  z=1
approximation of the real reaction rate for y down to some
value near zero, but not to zero itself. In this regard, we We let ya (z) be the solution of this boundary-value problem,
may assume that f(y) is bounded above, that is f(y) 6 K Eqs. (9) and (10), and also let the gradient of ya at z = 0 be
in 0 6 y 6 1, where K is a su0ciently large number. a. Then, theoretically, the integration of Eq. (9) from z = 0
with the following initial conditions is expected to give the
correct solution ya (z) of the boundary-value problem
3. Shooting method 
dy 
y = 1 at z = 0; = a: (11)
In the well-known shooting method, the boundary-value dz  z=0
problem becomes an initial-value problem with all the
In the shooting method, we solve Eq. (9) with an assumed
conditions speci:ed at the initial point. The missing ini-
initial guess for dy=d z at z =0 and the given initial condition
tial conditions are assumed and the initial-value problem
y = 1 at z = 0. We let the assumed initial guess be
is solved iteratively until the computed solution agrees 
with all the boundary conditions. This approach involves dy 
= b: (12)
a trial-and-error procedure. In the given boundary-value dz  z=0
problem, one of the two boundary points, x = 0 or 1, can
The purpose of the method is to improve b through an itera-
be used as the initial point.
tive procedure and to arrive at an approximation of a within
3.1. Shooting method at x = 0 a speci:ed accuracy.
If b is not close to a, the computed y(z) may behave un-
When x=0 is chosen as the initial point, dy=d x=0 is given realistically in the z interval: y decreases to negative values
and y(0) is assumed. The initial-value problem is solved or y decreases initially and increases later. If y(z) 0 at
using a step-by-step procedure from x = 0 to 1, and the com- some point, it can be shown that, after that point, y(z) con-
puted solution y at x = 1 is compared with the boundary tinues to grow in negative numbers. Or if y starts to increase
condition y(1) = 1. This is repeated by adjusting the ini- at some point, it continues to grow with an increasing pos-
tial guess until the computed value agrees with the bound- itive slope after the point. Since the reactant is consumed
ary condition within a speci:ed tolerance. This method can everywhere in the catalyst, the concentration must be mono-
be successful only when y(0) 0 is assured beforehand, tone decreasing with respect to z and, nevertheless, should
since with y(0) = 0 the resulting computed solution would be non-negative. Consequently, when dy=d z 0 or y 0
be always a trivial solution, y(x) = 0, and cannot agree with occurs at a step during the course of the computation by a
y(1) = 1. In general, however, y(0) 0 is not guaranteed, step-by-step method, the concentration pro:le computed af-
for example, if f(y) = 1, a zero-order reaction, it can be ter the step will be certainly erroneous, and thus the compu-

shown that y(0)=0 when  6 and s=2 (spherical cata- tation may be terminated at the step. Based on the predicted
lyst). For an apparent reaction order less than one and large behavior of y(z) after the step, the initial guess b can be
, y can be zero at 0 6 x 1. Therefore, the applicability accordingly adjusted in the next iteration for an improved
of this shooting method is limited. We denote this method solution. Therefore, using dy=d x 0 and y 0 as the cri-
as forward shooting to distinguish it from the shooting at teria in the iteration can considerably reduce the amount of
x = 1, discussed next. calculations.
Next, we examine how the error in the initial guess propa-
3.2. Shooting method at x = 1 gates in the z interval and also the instability associated with
Eq. (9), since the equation is not analytic at z = 1 when s = 1
The missing initial condition in this case is dy=d x at x = 1 or 2. Let yb be the solution of the initial-value problem, Eqs.
and by suitably adjusting the initial guess for the missing (9) and (12) with y(0) = 1, and de:ne u(z) = yb (z) ya (z),
2256 D.H. Kim, J. Lee / Chemical Engineering Science 59 (2004) 2253 2263

Table 1 behavior of u(z) revealed by the method is


Solutions of Eqs. (14) and (15) for f(y) = 1 and y and for various
catalyst geometry u(z) z for s = 0; (16)
f(y) = 1 f(y) = y 1
 u(z) ln for s = 1; (17)
s=0 z sinh(z) 1z

(slab) 1
u(z) for s = 2: (18)
1 I0 ()K0 ((1 z)) K0 ()I0 ((1 z))
1z
s=1  ln 
1z [I0 ()K1 () + I1 ()K0 ()] This behavior agrees with u(z) listed in Table 1.
(cylinder) Instability or a rapid growth of the error is normally unde-
 sinh(z) sirable in calculating the concentration pro:le y(z). In our
s=2  
1z (1 z) case, however, where an accurate estimation of a is sought
(sphere) for, the instability inevitably gives rise to dy=d z 0 or
y(z) 0 in the computed pro:le and, as a result, can make
the initial guess as close as desired to a by simply checking
dy=d z 0 or y(z) 0 in the pro:le.
the error of yb . As yb satis:es Eq. (9), we obtain
In the initial-value problem, information on whether u(z)
d 2 yb s dyb keeps its sign (+ or ) throughout the z range is helpful. If it
2 f(yb )
d z2 1 z dz does not change its sign, the initial guess with a positive error
( 0) generates a concentration pro:le which is always
d 2 ya s dya 2 d2 u greater than the correct solution over the whole interval, and
=  f(y a ) +
d z2 1 z dz d z2 vice versa. This will simplify correction of the initial guess
s du in the iterative procedure for improving b.
2 (f(ya + u) f(ya )) = 0: (13) Table 1 shows that the sign u(z) is the same as the sign of 
1 z dz
in (0,1) for f(y)=1 or y. It is easy to show that when f(y) is
Since ya is the solution, Eq. (13) becomes non-decreasing with respect to y, u(z) and du=d z keep their
d2 u s du signs (the same with each other) throughout the z interval.
= + 2 (f(ya + u) f(ya )): (14) For example, if (=du=d z at z = 0) 0, then u(z) 0 near
d z2 1 z dz
the initial point z =0, and this makes f(ya +u)f(ya ) 0
The initial conditions for u(z) are since f(y) is non-decreasing. Consequently, the right-hand
 side of Eq. (14) is greater than zero, and this in turn keeps
du 
u(0) = yb (0) ya (0) = 0; = b a : (15) du=d z 0 in the next step. The same pattern in the sign of
d z z=0
u(z) and du=d z continues in successive steps until z =1. The
u(z) is the error in the computed solution when the initial opposite is true when  0.
guess for dy=d z at z = 0 (x = 1) is incorrect by . When A negative reaction order gives rise to f(y) mono-
Eq. (9) is unstable, u(z) can grow very rapidly as z goes to 1. tone decreasing with y. In the cases where f(y) is not
The analytic solution of Eqs. (14) and (15) is possible non-decreasing with y, we are unable to theoretically de-
for a few special forms of f(y). Table 1 lists the solution cide the sign of u(z) over the whole z interval. Based on the
for f(y) = 1 and y. In Table 1, except for the case of s = 0, behavior of y(z) afterward dy=d z 0 or y(z) 0, how-
u(z) grows without bound as z 1, indicating instability, ever, it is clear that if u(z) does not change its sign until
and thus it is impossible to compute y(z) which is :nite at dy=d z 0 or y 0, u(z) keeps its sign over the entire z
z = 1, let alone that it agrees with the boundary condition interval (0,1).
dy=d z = 0 at z = 1. Fig. 1 shows the e%ect of  on u(z) for We applied the shooting method for a number of func-
f(y) = y. u(z) increases more rapidly with increasing  for tional forms of f(y) and in all the tested cases except the
all s. Thus, when   1, Eq. (9) can become unstable well cases of multiple steady states, we found u(z) did not change
before z = 1, whereas for a small , u(z) is moderate when its sign, i.e. the error grew in positive direction if the error
s = 0 or begins to grow rapidly near z = 1 if s = 1 and 2. in the initial guess dy=d z at z = 0, , was positive or grew in
On the other hand, when f(y) = 1, u(z) is independent of  the negative direction if  0. We demonstrate this with an

and modest if s = 0 or grows rapidly near the singular point example where f(y) = 1= y,  = 2, and s = 1. Fig. 2 shows
z = 1 when s = 1 or 2. the computed pro:les of y(z) for various initial guesses. At
For f(y) other than those in Table 1, Eq. (14) becomes the start, y(z) decreases for all b, but eventually either re-
non-linear and, in general, it is not possible to solve the Oects upward when b is greater than a certain value, 3:08
equation analytically over (0,1), but the behavior of u(z) in the present case, or continues to decrease when b is lower
in the vicinity of z = 1 can be inspected by the method of than the value. The pro:les for b = 3:08 and 3:10 are vir-
perturbation. Details of the method as applied to Eq. (14) tually identical in 0 6 z 0:4, whereas in 0:4 z 6 1, the
are given in Appendix A. In close proximity to z = 1, the pro:les deviate from each other considerably with di%erent
D.H. Kim, J. Lee / Chemical Engineering Science 59 (2004) 2253 2263 2257

20 20
s=0 s=1

=20
15 15
=10
=20
u(z)/ =10

u(z)/
10 10 =5

=5
5 5 =2
=2
=1
=1
0 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
(a) z (b) z

20
s=2

15
=20 =10
=5
u(z)/

10
=2
=1
5

0
0.0 0.2 0.4 0.6 0.8 1.0
(c) z

Fig. 1. E%ect of Thiele modulus  on error u(z) for a :rst-order reaction f(y) = y. (a) s = 0, (b) s = 1 and (c) s = 2.

2 correct solution ya (z), whereas a negative error results in


yb (z) ya (z).
1
b= -1 We add a few more examples to show the possible out-
come of y(z) with the reverse shooting method. Fig. 3 shows
-2
-3.08 the computed pro:les for f(y) = y, s = 0, 1 and 2 and
y(z)

0  = 1 and 10. In each case, the pro:les are obtained for two
-4
-3.1 values of b, which are close to each other but with an op-
-1 posite sign in their errors with respect to the exact b(=a).
-5
When  = 10, the pro:les for all three cases of s are simi-
=2, s=1 lar. Each of them drops to zero at a point in the interval and
-2
0.0 0.2 0.4 0.6 0.8 1.0
then moves either upward or downward, depending on the
sign of the error in b. On the other hand, when  = 1, the
z
characteristics associated with s are visible: s = 0 gives pro-
Fig. 2. Variation of computed concentration pro:le y(z) with respect to :les almost identical to the end of the interval without any

initial guess b. f(y) = 1= y. appreciable growth of the error in the initial guess; s=1 gen-
erates pro:les with a discernable di%erence at z = 1; while
signs. In this interval, 3:10 b 3:08, the correct s = 2 yields one pro:le going to and the other going to
initial guess of the problem exists, which is calculated as near z = 1. Although, in theory, the pro:le for s = 1 and
3:08007 : : : : Due to the instability near z = 1, it is not  = 1 should also be either or at z = 1, it is seen that
possible to obtain the pro:le that agrees with the boundary computation with a :nite precision and a small dy=d z near
condition at the catalyst center however accurate the initial z = 1 fails to show clearly the exact asymptotic behavior of
guess may be, but it is entirely possible to reduce the inter- ln(1=(1 z)) near z = 1. When the error growth is not dra-
val of b arbitrarily small, and yet the pro:les computed with matic or in the absence of the instability, it can be shown
the end points of the interval have opposite signs near the that either dy=d z 0 or dy=d z 0 is established near or at
catalyst center. By imposing y(z) = 0 for 0:4 z 6 1, one z = 1.
may consider the pro:le for either b = 3:08 or 3:10 as We may conclude that during the reverse shooting with
the solution ya (z). With thus determined ya (z), it is seen in an initial guess b, one of the three cases occurs, namely,
Fig. 2 that over the whole z range, a positive error in dy
Case A 0; b = b+ ; (19)
the initial guess results in yb (z) always greater than the dz
2258 D.H. Kim, J. Lee / Chemical Engineering Science 59 (2004) 2253 2263

2 2

=10, b=-11.54
1 1 =1, b=-0.465
=1, b=-0.84
=10, b=-10.96 =1, b=-0.475
=1, b=-0.85

y(z)
y(z)
0 0
=10, b=-11.55
=10, b=-10.97
-1 -1

s=0 s=1
-2 -2
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0

(a) z (b) z

2
=10, b=-10.39

1 =1, b=-0.322
=1, b=-0.323
y(z)

=10, b=-10.40
-1

s=2
-2
0.0 0.2 0.4 0.6 0.8 1.0
(c) z


Fig. 3. Computed concentration pro:les for  = 1 and 10 with initial guesses close to the exact values f(y) = y. (a) s = 0, (b) s = 1 and (c) s = 2.

Case B y 0; b = b ; (20) is a b . The second stage is to reduce the size of the inter-
 val (b ; b+ ) su0ciently small, so that any b in the interval
dy  is accurate within a speci:ed tolerance.
Case C y(1) 0 and 0; b = b : (21)
d z z=1 The main part of the algorithm is to determine whether
In Case A, yb (z) increases after dyb =d z becomes positive, the b used in the computation is b+ or b . To make the de-
and then yb (z) ya (z) near z = 1. In Case B, after yb (z) cision, y(z) and dy=d z are computed by a step-by-step pro-
becomes negative, it continues to be negative to z = 1, and cedure and checked at each z increment whether dy=d z 0
thus obviously yb (z) ya (z) near z=1. Case C occurs when (b = b+ ) or y 0 (b = b ). Once the decision is made, the
the reverse shooting does not encounter either case A or B computation with the given b may be stopped there. If the
until z = 1 and indicates yb (z) ya (z) at z = 1. We may decision cannot be made until z = 1, then dy=d z|z=1 0 and
classify b as either b+ if Case A occurs or b if Case B or C y(1) 0 and thus b = b . Here we may also check whether
occurs in the reverse shooting. Then the exact concentration dy=d z|z=1 can be regarded as zero within a speci:ed, toler-
gradient at z = 0, a, is located at the border between b+ and ance.
b . The remaining part consists of getting a better guess for
It is shown that when f(y) is non-decreasing with respect b. At the start of the :rst stage, the begining of the algo-
to y, then u(z) and du=d z have the same sign, either positive rithm, we do not have any information on where b changes
or negative, and their sign does not change in the entire from b to b+ . Since dya =d z at the surface (=a) always de-
z interval. Consequently, for this f(y), b+ has a positive creases (its magnitude increases because of its minus sign.)
error ( 0) and b has a negative error, and thus b+ is with increasing , we may begin with b =  or any other
always greater than b . As shown in Fig. 2, the relation reasonable guess and run the main part to determine whether
b+ b can also be observed even when f(y) is monotone b = b+ (Case I) or b = b (Case II).
decreasing with respect to y. In Case I, we have a b+ and search for a b , which is
less than b+ . We may assign the trial b as 2b+ and run the
main part (2b+ b+ , since b+ 0). If a b is found in
4. An algorithm for the reverse shooting this, then we have an interval (b ; b+ ) and can move to the
second stage below. Otherwise, we have found another new
The proposed algorithm is based on the assumption b+ . Replace the old b+ with this newly found b+ and try
b+ b and consists of two stages. The :rst stage is to again with b = 2b+ . Repeating these steps will eventually
:nd two initial guesses, one of which is a b+ and the other yield a b and an interval (b ; b+ ).
D.H. Kim, J. Lee / Chemical Engineering Science 59 (2004) 2253 2263 2259

100 1.2

1.1
10 b+
1.0
1 b- 0.9 b-
-b -b
0.1 0.8
b+
0.7
b+
0.01
0.6

0.001 0.5
0.1 1 1.00 1.05 1.10 1.15 1.20
(a) (a)
10 2.0

1.8

1.6

1.4

1
1.2

1.0
0.1 1 1.00 1.05 1.10 1.15 1.20
(b)
(b)
Fig. 4. (a) Plot of boundary between b+ and b and (b) plot of e%ec-
tiveness factor corresponding to the boundary points, f(y) = 1=y0:9 . The Fig. 5. Magni:ed view of the marked regions in Fig. 4. (a) Plot of
marked regions are magni:ed in Fig. 5. boundary between b+ and b and (b) plot of e%ectiveness factor.

In Case II, we have a b and look for a b+ . In this, the the other hand, it is obvious that b = belongs to b .
new trial b should be greater than b and hence may try Consequently, on decreasing b from 0 to , b changes
with b = b =2. If this new b is determined to be a b+ , then its class from b+ to b at least once. Multiple steady states
we move to the second stage. Otherwise, replace the old b exist when b changes its class, either from b+ to b or
with the new trial b and try again with b = b =2. We repeat from b to b+ , more than once on the b-axis for a given
these steps until we obtain (b ; b+ ). . On decreasing b, if there are two changes of the type
The interval (b ; b+ ) obtained above is equivalent to b+ b , there must be a change of the type b b+ in
(b ; b =2) or (2b+ ; b+ ). In the second stage, we use the

between the two. In view of this, the number of the steady
method of interval halving or bisection to reduce the size states can only be an odd number (2m + 1), among which
of the interval. We solve the initial-value problem, Eq. (9) m + 1 states occur at b+ b , while m states occur at
with y = 1 at z = 0 and b = (b+ + b )=2, and check whether b b+ .
the present b is a b+ or a b . We replace b+ or b with the Fig. 4(a) shows the regions for b+ and b for f(y) =
newly found b+ or b , respectively, and continue interval 1=y0:9 and s = 1. On the line that separates the two regions,
halving until the interval becomes su0ciently small, so any b changes its class and a steady state exists. Normally, upon
b in the interval has the desired accuracy. In all, 1114 it- decreasing b from 0, b changes from b+ to b once for a
erations are needed to compute the e%ectiveness factor with given . In Fig. 4(a), however, when 0:684  1:175,
0.1% accuracy. b changes its class apparently three times as b decreases
from 0 : b+ b , b b+ , and :nally b+ b . Hence,
three steady states exist in the  range. Fig. 4(b) shows a
5. Multiple steady states plot of the corresponding e%ectiveness factors. Furthermore
Fig. 5(a), a magni:cation of the small marked region in
The initial guess b is in the range (0; ). If b = 0, Fig. 4(a), reveals additional two more such changes in the
then Eq. (9) gives d 2 y=d z 2 0 at z = 0 since 2 f(y) 0. region 1:049  1:065, and hence a total of :ve steady
Hence dy=d z 0 for z 0 and b = 0 belongs to b+ . On states exist in this small  interval.
2260 D.H. Kim, J. Lee / Chemical Engineering Science 59 (2004) 2253 2263

When there is only a single steady state, it is a sta- Table 2


ble steady state and exists where b+ b , as shown in Methanol steam reforming kinetics and other necessary data used in Fig. 6
Figs. 4(a) and (b). If three steady states are possible, two of Catalyst property: pellet density=2350 kg=m3 , void fraction = 0:47,
these will normally be stable, whereas the third state which mean-pore radius = 8 109 m
normally exists between the two stable states is unstable. Reaction rate (Lim, 2002): 
In the  range where three steady states are possible, two kK1 PCH3 OH = PH2
RCH3 OH =   (mol/gcat/h)
steady states are obtained at b+ b and the third state (1 + K1 PCH3 OH = PH2 )(1 + K2 PH2 )
 
located between the two on the b axis, exists at b b+ . 105:3 kJ=mol
k = 1:16 1011 exp
This suggests that the steady states at which b+ b are RT
 
stable, while the states at which b b+ are unstable. In 20:06 J=mol=K 20 kJ=mol
K1 = exp +
view of this, the proposed algorithm based on the assump- R RT
tion that b+ b is e%ective for computation of the e%ec-  
57:18 J=mol=K 50 kJ=mol
K2 = exp +
tiveness factor of the stable steady states. However, if the R RT
algorithm is modi:ed to :nd the position at which b changes Partial pressures are in atm unit
its class, the example in Fig. 4 demonstrates that all the pos-
Reaction temperature: 503 K, total pressure: 1 atm
sible steady states and their stabilities as well as the cor- Bulk composition (mol %):
responding e%ectiveness factors can be easily found by the A: 20% CH3 OH, 30% H2 O, 50% N2 (inert)
reverse shooting method. This is another advantage of the B: 50% H2 , 20% CH3 OH, 30% H2 O
present method when compared to other methods such as
the :nite-di%erence methods and the orthogonal collocation E%ective di%usivities (m2 =s)
Composition A
methods, which by themselves are normally unable to :nd H2 : 1:80 106 , CO2 : 3:87 107 , H2 O : 6:04 107 ,
all the possible steady states and moreover require an initial CH3 OH : 4:49 107
estimate of the concentration pro:le that is close to the cor- Composition B
rect pro:le to compute the pro:le of an unstable steady state. H2 : 1:84 106 , CO2 : 3:96 107 , H2 O : 6:17 107 ,
We may assert that the classi:cation of the initial guess b CH3 OH : 4:60 107
according to the criteria mentioned before is not only useful
in computing the e%ectiveness factors, but also provides a
simple means of making a decision on the stability of the Cu=ZnO=Al2 O3 catalyst
steady states. CH3 OH + H2 O CO2 + 3H2 :
In the previous section, we discussed that when f(y)
is non-decreasing, a positive error in b( 0) always re- The reaction rate reported by Lim (2002) and other data for
sults in yb (z) ya (z) and dyb =d z dya =d z, whereas a computation are listed in Table 2. Methanol was taken as
negative error gives yb (z) ya (z) and dyb =d z dya =d z. the key component and its e%ective di%usivity in the catalyst
If dyb =d z dya =d z, then dyb =d z must become positive at was estimated using the modi:ed MaxwellStefan equation
a point in the z range 0 z 6 1, and thus the initial guess (Schneider, 1976) and the cross-linked pore model (Feng
b belongs to b+ . On the other hand, if yb (z) ya (z) and et al., 1974), recommended by Haugaard and Livbjerg (1998)
dyb =d z dya =d z, then it is evident that either yb (z) 0 based on an experimental comparison of pore-di%usion mod-
at some point in the z range or dyb =d z|z=1 0, giving that els. The catalyst particle diameter varied from 1 105 to
the initial guess belongs to b . This gives that b+ is always 0:1 m, which resulted in  in the range 0.11000 for the
greater than b , and hence, upon decreasing b from zero, composition A and 0.027270 for composition B. Fig. 6
there is only one point on the b-axis (0; ) at which b shows the e%ectiveness factor curves computed by the shoot-
moves from b+ to b . Consequently, the condition that ing method for the two bulk compositions. As it turns out,
f(y) is non-decreasing is a su0cient condition for the the curve also depends on the bulk composition, indicating
uniqueness of the steady state. complexity in the dependence of the e%ectiveness factor on
reaction conditions and kinetics. Fig. 6 also compares the
prediction of an approximation of the e%ectiveness factor,
6. Examples ,
recommended as the best in Wijnggaarden et al. (1998).
The approximation is expressed as an implicit equation:
In the shooting method, a numerical routine for the 1
 =  ;
initial-value problem is needed. Routines based on the 1 + An
1 + (1 )An
0
RungeKutta method or its variants are readily available in
commercial subroutine packages or an internet repository 2
An0 = 1 ;
(www.netlib.org). Any of those will work :ne in the (s + 1)2 [2
f(y) dy]
0
method. 
As a practical example for application of the present !2 df(y) 
An1 = ; (22)
method, we consider methanol steam reforming over a (s + 1)2 dy y=1
D.H. Kim, J. Lee / Chemical Engineering Science 59 (2004) 2253 2263 2261

exact 10
approximation
relative error
=0.1
B
1 0.3 1 0

A -0.1

Relative error
0.2

0.1
0.1 A
0.1 =30
B
0.01
0.01 0.0 0.1 1 10 100
0.01 0.1 1 10 100 1000

Fig. 7. E%ectiveness factor of nonisothermal spherical catalyst for
Fig. 6. Comparison of exact e%ectiveness factor and its approximation by half-order kinetics.
Eq. (22) for methanol steam reforming reaction over a Cu catalyst.

where ! is a geometry factor, which depends only on the 7. Conclusion


shape of the catalyst like s and the value of which is given
as 23 for slab, 1 for cylinder, and 65 for sphere. The approxi- A robust numerical method for computing the e%ective-
mation is seen to predict the e%ectiveness factor of compo- ness factor has been developed in this study. The method is
sition B within 10% error, while relatively poor agreement an iterative one and based on the shooting at the outer surface
is observed for composition A with a maximum of 30% er- of the catalyst, denoted as reverse shooting. We have shown
ror. Furthermore, the approximation can be used only when that during the reverse shooting with an initial guess b, one
An1 is positive, that is df(y)=dy|y=1 0. This limits the of three cases occurs in the computed pro:le yb (z): namely,
applicability of the approximation. dyb =d z 0, yb (z) 0, or yb (1) 0 and dyb =d z|z=1 0.
As another example, we consider a non-isothermal spher- We classi:ed b as either b+ if dyb =d z 0 occurs or b
ical catalyst in which a half-order reaction takes place. The otherwise. With this classi:cation, a steady state, a solution
temperature inside the catalyst is related to the concentration of the problem, has been shown to exist at the boundary
by Eq. (3), enabling to write the reaction rate in terms of between b+ and b . An algorithm useful for locating the
the concentration variable. The dimensionless reaction rate boundary point when there is a unique steady state is pro-
f(y) is given by posed. Also in the case of multiple steady states, the method
 E  is shown to be e%ective and e0cient to determine all the
A exp RT C1
f(y) =  steady states and the corresponding e%ectiveness factors. A
A exp RT E
C1s simple rule that b+ b at the boundary point is suggested
s


for the stability of a steady state.
#$(1 y)
= exp y; (23)
1 + #(1 y)
where # = (IH1 )De1 C1s =ke Ts and $ = E=RTs . For $ = 30 Notation
and # = 0:1; 0 and 0.1, the e%ectiveness factors were com-
puted by the reverse shooting method and shown in Fig. 7. A chemical species involved in reaction
Within a limited Thiele modulus range for a given # and a exact concentration gradient at the catalyst sur-
$, the same e%ectiveness factor can be computed by the face
well-known method proposed by Weisz and Hicks (1962), b initial guess of the concentration gradient at the
being similar to the forward shooting method and thus hav- surface
ing the same limitation that the center concentration must be b+ initial guess b for a concentration pro:le with
greater than zero. For the half-order reaction and for a large dy=d z 0 at a point
Thiele modulus, however, the concentration can decrease to b initial guess b for a concentration pro:le with
zero at some interior location and remains zero thereafter to either y 0 at a point or dy=d z|z=1 0
the center of the catalyst. In this case the method of Weisz C concentration, mol=m3
and Hicks fails. In Fig. 7, the Thiele modulus ranges within De e%ective di%usivity, m2 =s
which the method of Weisz and Hicks can be applied suc- f(y) dimensionless reaction rate (=R(C1 ; T )=
cessfully are marked with arrows. R(C1s ; Ts ))
2262 D.H. Kim, J. Lee / Chemical Engineering Science 59 (2004) 2253 2263

ke thermal conductivity of catalyst, W/m/K f(ya + u) can also be expanded as



L radius of a cylindrical or spherical catalyst, or 
2 df 
f(ya + u) = f(ya + u0 ) + & u1 + : (A.6)
half-thickness of a slab catalyst, m dy y=ya +u0
N mole Oux, mol=m2 =s
R(C1 ; T ) reaction rate, a function of C1 and T , mol=m3 =s Substitution of Eqs. (A.5) and (A.6) into Eq. (A.3) and
s shape factor of the catalyst (in:nite slab s = 0; collecting the terms of order &0 , we obtain
in:nite cylinder s = 1; sphere s = 2)
d 2 u0 s du0
T temperature, K 2
= ; (A.7)
d( 1 ( d(
u(z) error in the computed pro:le (=yb (z) ya (z)) 
x dimensionless position variable in the catalyst du0 
u0 (()|(=0 = q1 ; = q2 : (A.8)
(center, x = 0; outer surface, x = 1) d( (=0
y dimensionless concentration of the key compo-
nent (=C1 =C1s ) The terms of order &2 are
ya (z) exact dimensionless concentration pro:le in the
d 2 u1 s du1
catalyst 2
= + 2 (f(ya + u0 ) f(ya )); (A.9)
yb (z) dimensionless concentration pro:le obtained d( 1 ( d(

with b du1 
u1 (()|(=0 = 0; = 0: (A.10)
z dimensionless position variable (=1 x) d( (=0
Greek letters As u0 is the dominant term in Eq. (A.5) as & 0, we solve
 error in the initial guess b (=b a) for only u0 and obtain
 e%ectiveness factor s = 0 (slab)

 stoichiometric number z1
u0 = &q2 ( + q1 = q1 + q2 +1 ; (A.11)
 Thiele modulus, de:ned in Eq. (4) &
Subscripts s = 1 (cylinder)
1 key component 1
u0 = q2 ln + q1
i chemical species 1(
s catalyst outer surface 1
= q2 ln + q1 + q2 ln &; (A.12)
1z
Appendix A s = 2 (sphere)
q2
Here we are interested in the behavior of u(z) in the z u0 = + q1
1(
interval (1 &; 1). We assume that the initial-value problem,
&q2
Eqs. (14) and (15), has been computed from z = 0 to 1 &, = + q1 : (A.13)
and the result at z = 1 & is 1z

du 
u(1 &) = q1 ; = q2 =&: (A.1)
d z z=1&
References
Now we let
z (1 &) Aris, R., 1975. The Mathematical Theory of Di%usion and Reaction
(= : (A.2) in Permeable Catalysts. Volume 1. The Theory of the Steady state.
& Clarendon, Oxford.
Then Eq. (14) becomes Carnahan, B., Luther, H.A., Wilkes, J.O., 1969. Applied Numerical
Methods. Wiley, New York.
d2 u s du
2
= + &2 2 (f(ya + u) f(ya )) (A.3) Chang, H.N., 1982. Numerical calculation of e%ectiveness factor for the
d( 1 ( d( MichaelisMenten type kinetics with high Thiele moduli. A.I.Ch.E.
with Journal 28 (6), 10301032.
 Feng, C.F., Kostrov, V.V., Stewart, W.E., 1974. Multicomponent di%usion
du  of gases in porous solids. Models and experiments. Industrial
u(()|(=0 = q1 ; = q2 : (A.4)
d( (=0 Engineering Chemistry Fundamentals 13 (1), 59.
Fogler, H.S., 1999. Elements of Chemical Reaction Engineering, 3rd
As we have assumed f(y) is bounded
above, i.e. f(y) 6 K, Edition. Prentice-Hall, Upper Saddle River, New Jersey.
we may require 0 & 1= K, and seek u(() in the Froment, G.F., Bischo%, K.B., 1990. Chemical Reactor Analysis and
form of Design, 2nd Edition. Wiley, New York.
Haugaard, J., Livbjerg, H., 1998. Models of pore di%usion in porous
u(() = u0 (() + &2 u1 (() + &4 u2 (() + : (A.5) catalysts. Chemical Engineering Science 53 (16), 29412948.
D.H. Kim, J. Lee / Chemical Engineering Science 59 (2004) 2253 2263 2263

Hill Jr., C.G., 1977. An Introduction to Chemical Engineering Kinetics Schneider, P., 1998. E%ectiveness factor for a non-isothermal simple
and Reactor Design. Wiley, New York. catalytic reaction with combined transport processes: MaxwellStefan
Jiang, C.J., Trimm, D.L., Wainwright, M.S., 1993. Kinetic study of steam approach. Collection of Czechoslovak Chemical Communications 63,
reforming of methanol over copper-based catalysts. Applied Catalysis 252270.
A: General 93, 245255. Shiraishi, F., Hasegawa, T., Nagasue, H., 1995. Accuracy of the numerical
Lim, M.S., 2002. Production of hydrogen for fuel cell by methanol partial solution of a two-point boundary value problem by the orthogonal
oxidation and steam reforming. Ph.D. Thesis, Kyungpook National collocation method. Journal of Chemical Engineering of Japan 28 (3),
University, Daegu, Korea. 316323.
Miyakawa, H., Nagasue, H., Shiraishi, F., 1999. A highly accurate Villadsen, J., Michelsen, M.L., 1978. Solution of Di%erential Equation
numerical method for calculating apparent kinetic parameters of Models by Polynomial Approximation. Prentice-Hall, Englewood
immobilized enzyme reactions: 1. Theory. Biochemical Engineering Cli%s, New Jersey.
Journal 3 (2), 91101. Weisz, P.B., Hicks, J.S., 1962. The behavior of porous catalyst particles in
Schneider, P., 1976. Intraparticle di%usion in multicomponent catalytic view of internal mass and heat di%usion e%ects. Chemical Engineering
reactions. In: Heinemann, H., Carberry, J.J. (Eds.), Catalysis Reviews, Science 17, 265275.
Science and Engineering, Vol. 12. Marcel Dekker, New York, Wijnggaarden, R.J., Kronberg, A., Westerterp, K.R., 1998. Industrial
pp. 201278. Catalysis. Wiley-VCH, Weinheim.

Das könnte Ihnen auch gefallen