q = probability of ascending r = probability of staying
Greek Letters oo2 = variance prior to mixing u2 = mixture variance Literature Cited Bard, Y., IBM Scientific Center, New York, N.Y., 1967. Barber, M. N., Ninham, B. W., "Random and Restricted Walks", Gordon and Breach Science Publishers, New York, N.Y.. 1970. Christy, R. E., Milling, 17 (1972). Coulson. J. M., Maita. N. K.. lnd. Chem., 26, 55 (1950). Draper, N. R., Smith, H., "Applied Regression Analysis", Wiley, New York, N.Y., 1966. Eckhoff. N. D., Hill, T. R., Kimel, W. R., Trans. Kansas A&. Sci., 71, 101 (1968). Fan, L. T., Chen, S. J., Eckhoff, N. D., Watson, C. A., Powder Technol., 4,345 (1971). Greenspan, D., "Discrete Models", Addison-Wesley Publishing Co., Reading, Mass., 1973.
Inoue. U.,Yamaguchi. K., Kagaku Kogaku, 33, 286 (1969). Kuprits, Y. N., "Technology of Processing and Provender Milling", published by U S . Dept. of Agriculture and the National Science Fwndation, Washington, D.C., 1965. Lacey, P. M. C., J. Appl. Chem., 4, 257 (1954). Oleniczak, A. T., Ph.D. Thesis, Princeton University, 1962. Oyama, Y., Ayaki, K., Kagaku Kogaku, 20, 148 (1956). Parzen, E., "Stochastic Processes", Holden-Day, San Francisco, Calif., 1962. Rose, H. E., Trans. Inst. Chem. Eng., 37, 4 (1959). Yamaguchi, K., J. Powder Technol. (Japan), 6, 292 (1969). Walawender, W. P.,Oelves-Arocha, H. H.,Fan, L. T., Watson, C. A,, "A Study on Axial Mixing and Demixing of Grain Particles in a Motionless Mixer", Proceedings of the First lntematknal Conference in Particle Technology, 1973. Wang, R. H., Fan, L. T., Chem. Eng., 81, (11). 88 (1974). Weidenbaum, S. S., Bonilla, C. F., Chem. Eng. Pfog.. 51, 27 (1955). Wolf, D., White, D. H.. "Experimental Study of the Residence Time Distribution in Plasticating Screw Extruders", 78th National A I M Meeting, Salt Lake Ci. Aug 18-21, 1974.
-
Receiued for review March 31,1975 Accepted January 7,1976
Design Considerations for Fixed-Bed Reactors Using Pellets of Nonuniform Catalytic Activity Thomas Gordon Smith Department of Chemical Engineering and Materials Science, University of Minnesota, Minneapolis, Minnesota 55455
An approximate analytical method is suggested for incorporating pellet effectiveness and yield calculations into nonisothermalfixed bed reactor simulation procedures when the use of nonuniformlyactivated pellets is of interest. A step distribution of catalyst is included as a special case. The effect of nonuniformly activated pellets on reactor volume and performance is also discussed.
Introduction The concept of distributing a catalytically active material nonuniformly within an inert porous support has received increased attention over the past few years. Two distinct cases can be imagined: first, uniform activation to a fractional depth with an inert core which, following Minhas and Carberry (1969), will be referred to as partial impregnation, and second, a distribution of catalyst which changes gradually with depth (nonuniform impregnation), the most common case involving a continuous decrease from the surface to the center of the pellet. The former approach has been recommended by Delancey (1973) as the optimal policy for a reactor experiencing uniform activity decay and has been used by Horvath and Engasser (1973) in connection with enzyme reactors and by Friedrichsen (1969) and Smith and Carberry (1975a) with phthalic anhydride reactors. Nonuniform impregnation has been treated most recently by Corbett and Luss (1974)relative to yield and poisoning phenomena in spherical pellets. In addition to the way in which the catalyst is deposited, the level of activation relative to the maximum possible should be kept in mind; i.e., an upper limit on the number of active sites per unit area of support must be established as a reference point. T o utilize these developments in nonisothermal fixed-bed reactor simulation and design, it is usually mandatory, in the interests of computer time, that a simple analytical solution be available for calculation of the pellet effectiveness and/or yield. For catalyst distributions other than a step function, it happens that even in the isothermal case with simple boundary conditions, analytical solutions are not generally 388 Ind. Eng. Chem., Process Des. Dev., Vol. 15, No. 3, 1976
available, and when they are, they occur in forms that are not easily manageable, particularly if yield calculations are anticipated. It is therefore the purpose of this work to review the relationship between research involving individual pellets containing a nonuniform distribution of catalyst and its subsequent use in nonisothermal fixed bed catalytic reactor design and, further, to suggest an approximate analytical method for incorporating effectiveness or yield calculations into reactor simulation procedures when the use of nonuniformly activated pellets is of interest. S t e p Distribution In his paper correlating effectiveness factors for the infinite flat plate, the infinite cylinder, and the sphere, Aris (1957) included a discussion of depositing catalyst only in an outer spherical shell. He showed that use of the volume to surface ratio as the characteristic particle dimension again allowed the effectiveness factor to be calculated with reasonable accuracy from the formula for an infinite flat plate. Since then (Rester and Aris, 1969), finite particle shapes other than the sphere have been shown, under fully impregnated circumstances, also to obey this rule. Consequently, by induction, their partially impregnated counterparts will likewise have effectiveness factors which are well approximated by the infinite plate formula
v = - tanh 4 4 where
lor'" " " ' ~ " ' '
"""'
'
,Approximate '
' '""'1/
Analytical
1
095'-
This permits any isothermal first-order irreversible reaction occurring in a partially impregnated catalyst, or by further suitable modification of q5 (Petersen, 1965), any nth-order reaction, to be easily incorporated into an overall reactor simulation algorithm. It should be emphasized that up denotes only the particle volume containing catalytic material. If yield reactions of the type A B C are of interest, the derivation of Carberry (1962), which includes the effect of external mass transfer, may be invoked. For flat plate geometry, it was shown that the isothermal yield for equal diffusivities of A and B is given by
--
CBO Y = - s - -mA?B -+s mBVA[cAO
1
+s
v
a(x)dV
CL
4 '/e, '/2, 1,2,6; 4~'. - X ; 2.5 - 2~ '3 where x = 0 at the center and 1at the surface. Although all of the above distributions except the declining one (2.5 - 2 x ) permit an analytically determinable effectiveness factor, the resulting Bessel series solution, if used in the computation of yield, eventually requires the intervention of numerical techniques. Furthermore, for the case of declining activity, or for arbitrary distributions such as those displayed by Harriott (1969), numerical solutions of the steady-state material balance equation are necessary to even determine the x";
LY
= 0,
(Corbett and Luss,
080Approximate Analytical
1 100
to00
effectiveness. Thus, the work of Kasaoka and Sakata or a modification thereof allows reactor simulations with pellets having nonuniform impregnation to be undertaken without having to solve a boundary value problem for each particle in the reactor. By invoking the same arguments as developed for the step activity distribution (partial impregnation), the KasaokaSakata modified Thiele modulus can be used with Carberry's yield formula to compute yield-Thiele modulus curves such as those presented by Corbett and Luss and Shadman-Yazdi and Petersen, all of whose curves represent numerical results. T o demonstrate this, the numerically calculated yield resulting from the steepest activity distribution used in either of the above two articles is compared in Figure 1 with the analytical technique proposed here. The results for this extreme case are not identical because the use of a universal Thiele modulus constitutes an approximation. However, considering the accuracy with which the basic physiochemical parameters implicit in the figure can be determined, the agreement is satisfactory for design purposes, particularly comparative designs. The following calculations illustrate the details of the proposed procedure. Example. For the yield reaction A B C, the ordinary differential equations describing mass transfer within a given pellet are
--
(3)
where a ( x ) represents the particular variation of activity with depth. Their results can be adapted to other geometries by the usual redefinition of the Thiele modulus in terms of u,/A and by a modification of a ( x ) which will be described later. In subsequent articles, Shadman-Yazdi and Petersen (1972) (SYP) and Corbett and Luss (CL) evaluated the effectiveness and yield of spherical pellets for the following activity distributions: SYP
,- Numerical 1
Figure 1. Comparison of approximate and exact yield curves for a spherical pellet having a catalyst distribution of 4x9: mi = 1; CBO/CAO = 0; s = 0.1.
Arbitrary Distribution Effectiveness factors for flat plate geometry have been calculated analytically by Kasaoka and Sakata (1968) for many catalyst distributions, including cases of activity declining or increasing with distance from the center. They found that the effectiveness factors for all situations could be brought into satisfactory agreement by employing a modified Thiele modulus of the following form
s
085-
10
Now since Y = Y ( u )it immediately follows that eq 2 must also apply to any shaped catalyst, fully or partially impregnated, for which effectiveness factors, vi, for linear kinetics are available. The generality of eq 1therefore permits the calculation of yield via eq 2 for a variety of catalyst dispositions. Additionally, since Carberry's work was developed for a general macro-micropore catalyst model, this generality extends to the partially impregnated case, a desirable feature in view of the recent pellet optimization work reported by Delancey (1974). A mathematical proof of the above reasoning is given by Aris (1974). Note, however, that since eq 2 was derived for linear kinetics, the introduction of Peterson's modified TJ does not guarantee the accuracy of eq 2 under nonlinear circumstances.
v
Y
(2)
where
4* = !f
0 90t
(4)
with boundary conditions:
dCA- dCB -- 0 dr
dr
CA = CAO;CB = CBO
(r=O) (r = R )
This assumes spherical geometry, equal diffusivities of A and B, negligible film mass transfer resistance, and an isothermal pellet. These conditions were selected so that a comparison could be made between the proposed technique and the results of Corbett and Luss. Equation 4, written in terms of plane geometry, was solved analytically by Kasaoka and Sakata for many different catalyst activity profiles and the resulting effectiveness factors versus Thiele modulus were correlated by eq 3, as discussed previously. The result is shown in Figure 2 (their Figure 8). Corbett and Luss, on the other hand, solved eq 4 and 5 numerically for several activity distributions and produced the Ind. Eng. Chem., Process Des. Dev., Vol. 15,No. 3, 1976
389
given pellet operating conditions. It can be seen from the form of eq 2 that if the effectiveness factors are accurately predicted then the yield itself is accurately predicted. The problem thus reduces itself to the use of Figure 2 to simulate Figure 3. This can be done as shown below. The most severe distribution appearing in Figures 3 or 4 is a ( x ) = 4x9. For s = 0.1, Cg = 0 (corresponding to the conditions of Figure 4), and mA = mg = 1, eq 2 reduces to
Noting that "
03
0 4 0 5 060708091
2
3
4
5
6 7 8 9 1 0
and using equation 1 there results
@* Figure 2. Effectiveness factor, 7, of a first-order isothermal irreversible reaction in an infinite flat plate versus modified Thiele modulus 6' for a variety of activity distributions.
0
05
L1
3
tanh
Y = 1.11 - 0.35
(E)
tanh +A*
Corbett and Luss's definition of q5A = R -for spherical geometry and a volume averaged rate constant must now be converted, for use with Figure 2, to the nomenclature of Kasaoka and Sakata corresponding to plane geometry and a rate constant defined at the pellet surface. The former conversion is easily handled in the usual manner by letting the characteristic dimension L = R/3. The latter conversion is accounted for by noting that since a ( x ) = 4x9, K A must be replaced by 4h.4. Thus, the Thiele modulus becomes
0'4i3
4x/3
0
0'
0
a = 10
A
0.25-2x
I
I
I
1
5
I
2
3
I
5
I
I
IO
20
Y
I 30
/
= R f i
@A
Figure 3. The effect of activity distribution on the effectivenessfactor of a first-order isothermal irreversible reaction in a spherical pellet.
Finally, the normalized activity distribution of a ( x ) = x9 in spherical geometry must be replaced by an equivalent distribution in plane geometry. This is accomplished by observing that the volume averaged rate constants should be the same. Thus
where
k ( r ) = Ra(r)
(11)
Carrying out this integration for a s p h ( x ) = x9 leads immediately to the result that the equivalent activity distribution in plane geometry is a,(+) = x3. This considerable difference in profiles is of course due to the drastically different volume vs. depth functionality in the two geometries. For instance, 50% of a sphere's volume is available at a depth of only 20% of the radius. Now, using eq 3, the Thiele modulus becomes
Substituting in the yield equation gives I
I
I00
IO R
1000
4A tanh -
m
Figure 4. The effect of activity distribution on the yield of intermediate: CBO/CAO= 0; negligible external mass transfer resistance.
Y = 1.11 - 0.35
19
~
(13)
4A
tanh 6
results shown in Figures 3 and 4 (their Figures 1 and 2). If one were interested in simulating the behavior of a fixed bed reactor containing hundreds of pellets of a given activity profile, the computer time required to repeatedly solve eq 4 and 5 would be unacceptedly high. Thus it is suggested that eq 1,2, and 3 be combined to approximate the results of the exact solution of eq 4 and 5 and thus to predict the yield for any 390
Ind. Eng. Chem., Process Des. Dev., Vol. 15, No. 3, 1976
where 4~ is the Thiele modulus as defined by Corbett and Luss. The proposed technique can now be verified by comparison of the results with Figures 3 and 4. For example, for 4~ = 10 Figure 3 gives 7 N 0.5. For = 10, $A* = 1.67 and using the mean curve of Figure 2 gives 7 N 0.55. Similar accuracy is achieved over the complete range of Thiele modulus. NOW using these results together with eq 13 gives the approximate
yield curve shown in Figure 1. The exact curve of Figure 1 is the same as the upper curve of Figure 4. It should be emphasized that this technique is applicable to arbitrary activity distributions in any geometry because such distributions can always be fitted by polynomials of some order and the subsequent integration indicated by eq 10 easily follows. Such arbitrary activity distributions would ordinarily demand the employment of numerical techniques for the computation of effectiveness and yield. Also, the technique is not restricted to situations with negligible external mass transfer resistance or equal diffusivities, as eq 2 was originally derived to include both of these effects. For either the step or gradual distribution, the effect of nonisothermal conditions must be included if the method is to be of practical value. Fortunately, there is considerable evidence in the literature (Smith, 1976) indicating that the temperature increase within the pellet is small. In particular it is usually much smaller than the temperature increase across the external film and can conveniently be ignored with virtually no effect on the overall reactor design. Thus, the above techniques described for the isothermal behavior of partially or nonuniformly impregnated pellets can be readily used, in conjunction with the simple trial and error heat balance for each pellet, to determine the pellet temperature corresponding to the localized conditions within the reactor (see, e.g., Carberry and White). This avoids the repeated solution of a nonlinear boundary value problem for heat and mass transfer around and within the pellets. Finally, it should be mentioned that even though only cases where the catalyst activity decreases toward the interior of the pellet are usually of interest, there are a few instances where an increasing activity could be of value. For example, CO oxidation over platinum catalyst follows negative first order behavior a t CO concentrations greater than a fraction of 1% (McCarthy and Carberry, 1975). Under these circumstances, the observed rate, and hence the effectiveness, is enhanced by the pellet concentration gradient (Smith and Carberry, 1 9 7 5 ~ )Motivated . by considerations similar t o those for positive order kinetics, one would anticipate locating catalyst primarily in the region of the highest reaction rate, thus pointing to a catalyst distribution which increases toward the pellet interior. This situation is handled equally well by the design procedure proposed herein. However, the method becomes inaccurate for large values of the Thiele modulus if the catalytic activity on the surface approaches zero.
Other Design Considerations As has been illustrated not only by Kasoaka and Sakata for plane geometry, by Basmadjian (1963) for hollow pellets of the Raschig ring type, but also by Shadman-Yazdi and Petersen for spheres, a reduction in overall catalyst activity can provoke a decrease in the net rate of consumption of A and production of B in spite of increased pellet effectiveness. At low values of the Thiele modulus this decrease is appreciable. Note that this occurs whether the activity is increasing or decreasing with depth because the reference activity is presumably the maximum possible for the particular catalyst and operating conditions. It is of interest to know how serious this problem is in actual reactors. This answer can only be found by direct simulation and design because the large number of parameters involved, e.g., kinetics, diffusional intrusions, relative rates, relative external concentration levels of products and reactants, absolute and relative rates of external heat and mass transfer, all depend on reactor operating conditions, but can only be examined parametrically in studies of single pellets. For situations where only conversion is of interest, the work of Minhas and Carberry should provide some guidelines for
answering this question. They considered the design of an adiabatic fixed bed reactor which used partially impregnated catalyst to oxidize SO2 over Pt or V205, an exothermic reaction. Their results, if analyzed carefully and with due regard for the reference point of maximum catalytic activity, indicate that partial impregnation increases reactor volume for similar conversion requirements. However, as lately pointed out by Smith and Carberry (1974, 1975a, 1975b3, due to the many parameters involved, the nonlinearity of the problem, and the opportunity for considerable interaction among parameters, nonisothermal fixed bed reactor designs which purport to compare the effects of specific parameter variations are best carried out on the basis of optimized comparisons using an appropriate objective function. Thus a general statement regarding the effect of partial impregnation on reactor volume cannot be extracted from the Minhas-Carberry results. On the other hand, Smith and Carberry (1975a) have recently evaluated, under optimized conditions, the effect of partial impregnation on yield in a nonisothermal, nonadiabatic fixed bed reactor for the production of phthalic anhydride, also an exothermic reaction system. They found that for the same reactor volume and the same conversion level a reactor containing spherical pellets uniformly impregnated to 37% of the total radius at an activity level of 30% of the base case, exhibited not only higher yield but improved thermal sensitivity relative to an optimized reactor employing fully impregnated catalyst pellets. The factor compensating for the drastically reduced amount of catalyst was a 10% increase in the inlet gas temperature. Although both of the above cases involved step distributions of activity, the efforts of Corbett and Luss and ShadmanYazdi and Petersen point to similar benefits for pellets within which the catalyst activity varies gradually with catalyst depth. Indeed, in view of the experiments of Harriott which illustrate some of the unusual distributions that occur, this seems the more likely situation and renders especially useful the approximate analytical technique advocated above.
Summary It has been shown that the “universal” effectiveness factor chart for pellets of nonuniform catalytic activity developed by Kasaoka and Sakata for infinite flat plate geometry can be generalized to other geometries by virtue of two easily implemented modifications. First, the usual modification of the characteristic particle dimension for a fully impregnated pellet is invoked; Le., L = u,/A is incorporated into the Thiele modulus. Second, the activity distribution for the given geometry is converted into the mathematically equivalent distribution in plane geometry by requiring that the volume averaged rate constants in the two geometries be equal. This implies an equivalent accessibility of the catalytic sites in the face of reaction induced diffusional intrusions. For the reaction A B C, the yield occurring in pellets of nonuniform activity can now be determined by using, in combination with the above, an equation derived by Carberry which accounts for pellet geometry and activity distribution solely in terms of the ratio of the appropriate effectiveness factors VB/VA. The proposed technique is valid and easily usable for virtually any distribution of catalyst thereby eliminating the need for the extensive numerical calculations which would ordinarily be required for determination of either effectiveness or yield. Since it can accurately accomodate situations involving both nonisothermality and finite film resistances, the method is particularly suitable for use in computer simulations of fixed bed reactors. Finally, it is pointed out that, when compared on an optimized basis, the overall performance of fixed bed reactors utilizing catalyst pellets of nonuniform activity is probably
--
Ind. Eng. Chem., Process Des. Dev., Vol. 15, No. 3, 1976
391
superior to that of reactors relying on uniformly impregnated pellets.
Nomenclature A = external surface area of catalyst pellet C = concentration of reactant in pellet CAO,CBO = bulk fluid-phase concentrations of species A and B D = effective pellet diffusivity k = reaction rate constant per unit active pellet volume K = volume averaged reaction rate constant k , = interphase mass transfer coefficient L = characteristic pellet dimension R = radiusof pellet r = radial coordinate of pellet V = total pellet volume u p = volume of pellet containing catalyst x = dimensionless distance from center to surface of pellet,
rlR
Y = yield of intermediate = effectiveness factor I#I = Thielemodulus Subscripts i = species A o r B p = infinite flat plate sph = sphere
Literature Cited Aris, R., Chem. Eng. Sci., 6, 262 (1957). Aris, R., "The Mathematical Theory of Diffusion and Reaction in Permeable Catalysts," pp 349-353, Clarendon Press, Oxford, 1974. Basmadjian. D., J. Catal., 2, 440 (1963). Carberry, J. J., Chem. Eng. Sci.. 17, 675 (1962). Carberry, J. J.. White, D., lnd. Eng. Chem., 61, 27 (1969). Corbett, W. E., Luss, D.. Chem. Eng. Sci., 29, 1473 (1974). Delancey, G.. Chem. Eng. Sci., 28, 105 (1973). Delancey, G., Chem. fng. Sci., 29, 1391 (1974). Friedrichsen. W., Chem. lng. Techn., 41, 967 (1969). Harriott, P., J. Catal., 14, 43 (1969). Horvath. L., Engasser, J., lnd. Eng. Chem., Fundam., 12,229 (1973). Kasaoka, S., Sakata, Y., J. Chem. Eng. Jpn., 1, 138 (1968). McCarthy, E. F.. Zahradnik, J., Kuczynski. G. C., Carberry, J. J., J. Catal., in press ( 1975). Minhas, S., Carberry, J. J., J. Catal., 14, 270 (1969). Petersen, E. E., "Chemical Reaction Analysis." pp 48-66, Prentice-Hall, New York, N.Y., 1965. Rester, S., Ark, R., Chem. Eng. Sci, 24, 793 (1969). Shadman-Yazdi, F., Petersen, E. E.. Chem. Eng. Sci., 27, 227 (1972). Smith, T. G.,Carberry, J. J., Adv. Chem. Ser., No. 133, 362 (1974). Smith, T. G., Carberry, J. J., Can. J. Chem. Eng., 53, 307 (1975a). Smith, T. G., Carberry, J. J.. Chem. Eng. Sci., 30, 221 (1975b). Smith, T. G., Carberry, J. J., Chem. Eng. Sci., 30, 763 ( 1 9 7 5 ~ ) . Smith, T. G.. Chem. Eng. Sci., in press (1976).
Received for review April 14,1975 Accepted February 9,1976
This work was supported by the National Science Foundation, Grant GK-42149.
Catalytic Coal Liquefaction Using Synthesis Gas Yuan C. Fu' and Eugene G. lllig Pittsburgh Energy Research Center, U.S. Energy Research and Development Administration, Pittsburgh, Pennsylvania 15213
High sulfur bituminous coal is liquefied and desulfurized effectively by hydrotreating with synthesis gas at temperatures of 400-450 O C and operating pressures of 3000-4000 psi in the presence of cobalt molybdate and sodium carbonate catalysts, steam, and a recycle oil. Comparison with coal liquefaction using pure hydrogen and cobalt molybdate catalyst has shown that, in both systems, the optimum liquefaction temperature is in the range of 425-450 OC and there is no significant difference in the coal conversion and the oil yield. The sulfur content and the viscosity of the oil product both decrease with the amount of hydrogen consumed in each system, but less total hydrogen is required for the same oil product quality in the synthesis gas than in the hydrogen system. The catalytic hydrotreating of coal using synthesis gas, unlike that using hydrogen, removes large amounts of the oxygen in coal as carbon dioxide. The off-gas, with increased H2:C0 ratio, is a low-Btu gas which could be burned or utilized for methane production or methanol synthesis. The cost of hydrogen production is saved.
Introduction Recent energy shortages and environmental problems emphasize the great importance of desulfurizing coal to produce low-sulfur fuels. Efforts are in progress in many organizations to develop catalytic hydrogenation processes for producing liquid fuel from coal. Unfortunately, most hydrogenation processes require enormous amounts of expensive hydrogen and some significant breakthrough is needed before an economical process can be realized. A major effort must be made to reduce the high cost of hydrogen. One approach is to utilize low cost synthesis gas as the process gas to achieve equal or better results. There have been some attempts to use carbon monoxide or carbon monoxide-containing gas for liquefying lignite (Appell et al., 1972), hydrotreating hydrocarbonaceous liquids (Ver392 Ind. Eng. Chem., Process Des. Dev., Vol. 15, No. 3, 1976
non and Pennington, 1973), and desulfurizing heavy liquid hydrocarbon (Pitchford, 1973). The process of liquefying lignite with carbon monoxide and water does not work well on bituminous coals and does not desulfurize coal very effectively. The hydrotreating and desulfurization of heavy liquid hydrocarbons appear to proceed well with carbon monoxide or carbon monoxide-containing gas in the presence of steam and an active metal catalyst. We have also reported that organic wastes have been hydrotreated with synthesis gas and cobalt molybdate-sodium carbonate catalyst and converted to oil (Fu et al., 1974). The present work deals with an effort to develop an economical process for coal liquefaction-hydrodesulfurization using low cost synthesis gas in the presence of cobalt molybdate and sodium carbonate catalysts. While hydrogenation