You are on page 1of 40

478

a.

Ind. Eng. Chem. Process Des. Dev., Vol. 17, No. 4, 1978 Literature Cited
Coulaloglou, C. A., Tavlarides, L. L., AIChE J., 22, 289-297 (1976). Hixson, A. W., Tenney, A. H., Trans. Am. Inst. Chem. En$, 31, 113-27 (1935). Miller, S. A,, Mann, C. A,, Trans. Am. Inst. Chem. Eng., 40, 709-745 (1944). Skelhnd, A. H. P., Seksafia, R., I d . fng. Chem. ProcessDes. Dev., 17, 56-61 (1978).

= viscosity, Ns/m2

= constant

density, kg/m3 & , = positive density difference between continuous and disperse phase, kg/m3 u = interfacial tension, N/m C$= volume fraction of organic liquid, dimensionless Subscripts c = continuous d = disperse
p =

Received for reuiezu September 15, 1977 Accepted June 29, 1978

This work was partially supported by National Science Foundation Grant No. ENG74-17286.

Complex First-Order Reactions in Fluidized Reactors: Application of the KL Model


Octave Levenspiel"
Chemical Engineering Department, Oregon State University, Corvallis, Oregon 9733 1

Nlels Baden
Kemiteknik, Technical University, 2800 Lyngby, Denmark

B. D. Kulkarni
National Chemical Laboratory, Poona 4 11 008, India

Conversion and product distribution equations are developed for the Denbigh reactions
A - R - S

and all its special cases taking place in a bubbling bed of fine catalyst particles. The final expressions are much like the equations for fixed bed reactors except that the reaction rate constants must be suitably modified to account for the mass transfer effects in the bed.

In 1968 Kunii and Levenspiel developed a model (the KL model from now on) to account for the main features of the vigorously bubbling fluidized bed with its fast rising bubbles. They then applied this model to predict reactor behavior for first-order reactions. Since then there have been a few extensions to other first-order reactions (Kunii and Levenspiel, 1969; Kunii, 1975; Carberry, 1976). In this paper we develop the performance equations for both conversion and product distribution in fluidized bed reactors according to this model for the rather general first-order reaction scheme commonly known as the Denbigh reactions (Denbigh 1958)
A - R - S

F i r s t - O r d e r Irreversible Reaction

The simple KL model assumes bubbles of one size which are fast enough (ub >> u,f) so that gas flow through the bed via cloud and emulsion is negligible compared to flow via bubbles. Figure 1then sketches the main features of the model and shows that there are five resistance steps in series-parallel arrangement for the reactant gas to contact and react on the surface of the solid. For a first-order catalytic reaction with rate given as A R, -rA = kCA,mol of A converted/s kg of cat., we find the following: for ideal plug flow

By putting the appropriate rate constants equal to zero we can obtain at the same time the performance equations for all the following reaction schemes
R

C A O ln-=kT=kCA for the fluidized bed C A O In - = C A


0 1978 American Chemical Society
KT

W
UO

/
'T

A -

R -

S.

A - R - S ,

\
T

W
K-

UO

where the effective rate constant K for the fluidized bed

0019-7882/78/1117-0478$01.00/0

Ind. Eng. Chem. Process Des. Dev., Vol. 17, No. 4, 1978 479
,REACTANT
/

ENTERS HERE

reac'tion

redtion

reac'tion

bubble

cloud

emulsion

(b)

Figure 2. Sketch showing the 22 reaction and mass transfer steps representing the reaction

Figure 1. Main features of the simple KL model for a bed of fast rising bubbles: (a) the model used for chemical conversion calculations. (b) the five mass transfer and reaction steps for the one step reaction A R.

reactor accounts for the five mass transfer and reaction resistance in the array of Figure l b , and is given by

taking place in a fluidized bed.

1
1
1

I
I

YckPs+
I -

1
U O

lx
I

movement of all species; thus we must follow 22 reaction and resistance steps (actually only 20 steps since material balances will give the composition of the last species), as shown in Figure 2. For any differential slice of bed of height dl the material balance equation for any specie can be written as (overall disappearance) = (reaction in bubble) + (transfer to cloud-wake) (transfer to cloud-wake) = (reaction in cloud-wake) + (transfer to emulsion) (transfer to emulsion) = (reaction in emulsion) Now the interchange coefficients for A and R differ because of differing diffusion coefficients. Thus
Kb; = 4.50

(1- E m f b b r P s

(3)

where K b c and K,, represent the mass transfer resistances between bubble, cloud, and emulsion regions. Comparing eq 1 and 2 we find the efficiency of fluidized contacting to be e = - =
K

(7) + (
5.85
h f DiUb

Dil/2 gl/4
db5/4

i = A,R

weight of catalyst needed for plug flow weight needed in the fluidized bed

SBme output

Kck = 6.78

( 7 i =) A, R

(4)

First-Order Irreversible Reaction System: the Denbigh Reactions Consider the reaction scheme
(5)
T U

Since the diffusion coefficients rarely differ greatly and only influence the K values a t most by the 'Iz power, we often can reasonably take the K b c and K,, values for all components to be identical, and thereby drop the superscript. However, to keep the derivation general we here retain the distinction between these different interchange coefficients. Next, calling kl + k z = kl2 and k 3 + k4 = h34 we then have for species A: in bubbles
-rAb

Here we are interested in more than the disappearance of reactant; we may want to follow the concentration changes of the other reacting species, study product distribution, find out how much catalyst is needed t o maximize forto the mation of intermediate R, and then compare CR,max expected in plug flow. For this we must keep track of the

YbklZPsCAb

+ Kb/(CAb

CAc)

in clouds
Kb/(CAb
-

CAc)

YcklZPsCAc

+ KceA(CAc

- CAe)

in emulsion

480

Ind. Eng. Chem. Process Des. Dev., Vol. 17, No. 4, 1978

KceA(CAc- CAe) = Yekl2PsCAe For specie R we have: in bubbles -rRb = -7bklPsCAb + Ybk34PsCRb in clouds KbcR(CRb - CRc) = -Y&lPsCAc in emulsion KceR(CRc- CRe) = -YeklPsCAe

+ KbcR(CRb- CRc)

+ Yck34PscRc +
KceR(CRc - CRe)

+ Yek3lPsCRe

Similar expressions can be written for all the other reacting species. Taking a differential slice of the bed, manipulating and substituting repeatedly to eliminate all cloud and emulsion concentrations leads us eventually to the following expressions, all in terms of bubble concentrations.
- = same as

cu

C A O

- but with k3

CS

. CAO

k 4 and Cso

Cu0

(7) The value of CRm= and the amount of catalyst needed to achieve this when R is absent in the feed, or C R O = 0, is given by

and where

1
1 1
Yck34Ps

1 1

1
- - - -

KAR

= - K 1 2 - KA

kl kl2

(14)

By Laplace transformation of eq 6 to 10, starting with a feed CAO,CRO,Cs0, CTO, and Cuo, we find a t the exit of the reactor

Discussion (a) Fluidized vs. Fixed Bed Operations. In the fluidized bed equations if we put all K h = Kce= m we then k12, K34 k34,KA 0, KAR k l and all the find that K~~ fluidized reactor expressions, eq 15 to 21, reduce to the corresponding fixed bed (plug flow) expressions. (b) Special Cases. The above fluidized bed expressions

Ind. Eng. Chem. Process Des. Dev., Vol. 17, No. 4, 1978 481

reduce directly for all the special cases listed a t the beginning of this paper. Thus
for A

Ubr

= 0.711 (gdb)'l2 = 0.711 (9.8 x 0.32)1'2 = 1.26 m/S = ug - umf

ub

+ Ubr = 0.30

0.03 + 1.26 = 1.53 m/S

S p u t k, = k4 = 0

R
for A
f 2

put k4 = 0 a n d a d d Cs t

C , to give Cu

A-U

K,, = 6.78
and so on. (c) Character of the Solution. By putting numbers into these expressions we come up with three important but not unexpected generalizations. First, with no intrusion of pore diffusion effects, the fluidized bed always needs more catalyst than the fixed bed to achieve a given conversion or to reach CRmax. Second, for reactions in series and with no intrusion of pore diffusion effects, fluidized beds always give a lower yield of intermediate compared to fixed bed reactors. Third, for reactions in parallel, fluidized bed operations do not affect the product distribution. (a). In most situations we would not be justified in retaining the distinction between the various interchange coefficients. Thus we would drop the superscripts and use an average Kbcand K,, throughout.

( 7 =0.147 ) s-l
mfD Ub

l/2

Ye =

(1 - e m f ) ( l - 6) - yc - y b = 2.09 6

kp, =

(400 x 10-6)(2000)= 0.8 s-l

Next, calculate the needed K values. Since k2 = k4 = 0, we have K~~ K~ (put k12 k,) and K~~ k 3 (put k34 k3). Thus from eq 11
K 1 2 = K l =

Conclusion We here have developed the performance expressions for the Denbigh reaction scheme
A - R - S

1
0.614
0.24(0.8)

1
1 2.09(0.8)

+0.147

and all its special cases occurring in a bubbling fluidized bed of fine particles, as pictured by the KL model. The resulting equations are somewhat like the corresponding plug flow expressions, except that the effective rate constants K , to be used, include both mass transfer and reaction resistance steps. Illustrative Example. In the presence of a particular catalyst ( p , = 2000 kg/m3) reactant A decomposes by first-order reactions as follows

(0.3) = 51.4 X (1- 0 . 5 ) ( 1 . 2 6 ) ( 2 0 0 0 )


Similarly, from eq 1 2
K34

m3/kg. s

Jx

K3

= 15.6 x

m3/kg.s

From eq 13
KA

= 13.2 X lo* m3/kg.s

Inserting into eq 14 gives


KAR

A-R-S

ki

kz

~1 - K A

= 38.2 X lo* m3/kgs

Thus eq 20 and 21 give

where k l = 400 X 10Bm3/kg.s and k2 = 25 X lo* m3/kgs. Using a catalyst bed 2 m in diameter with a feed of A in inert bo = 0.30 m/s) find C ,, and the amount of catalyst needed to achieve CR,max (a) for fluidized bed operations (ud = 0.03 m/s, emf = 0.5) if the estimated effective bubble size is d b = 0.32 m, a rather large value, and (b) for fixed bed operations (assume plug flow of gas). Additional Data. From the literature take D = 20 X mz/s for both A and R and a = 0.33. Solution. Let us work everything in SI units. (a) Fluidized Bed. In essence the KL model says that

CRmax CAO

-KAR
K1

(
-

~)K3'(K3-R1)

= 0.44

and

W=

uo ln
K3

(K3/K1) K1

= 31400 kg

(b) Fixed Bed. Here eq 20 and 21 reduce to In (400/25) uo In ( k l / k 3 ) - 0 . 3 ~ = 7000 kg Wfixed = (400 - 25)10-6 kl - k 3

Going through the details of the model we find

Note: The formation of intermediate is drastically lowered in fluidized bed operations (44% vs. 83%) even

482

Ind. Eng. Chem. Process Des. Dev., Vol. 17, No. 4, 1978

though the bed is much larger (31 tons vs. 7 tons). This all comes from the severe bypassing of reactant in the large bubbles. Reducing the size of bubbles will greatly improve the performance of the fluidized reactor.

Greek Letters
a = wake volume/bubble volume y i = volume of solids in region i/volume of bubbles 6 = volume of bubbles/volume of bed emf = void fraction of bed at minimum fluidizing conditions K = effectivereaction rate constant in the fluidized bed,m3/kgs ps = density of solids, kg/m3 7 = W/uo = weight-time, the capacity measure for catalytic

Nomenclature
A, R, S, T, U = reaction components C = concentration, moi/m3 d b = effective or mean bubble diameter, m D = diffusion coefficient for reacting gases, m2/s e = efficiency of contacting in the fluidized bed for a particular reaction k = rate constant for a catalytic reaction, m3/kg.s Kbc = gas interchange coefficient between bubble and cloud,
S-1

flow reactors, kgs/m3

Superscript A, R = refers to components A, R

K , = gas interchange coefficient between cloud and emulsion,


S-1

b = bubble phase c = cloud phase e = emulsion phase mf = at minimum fluidizing conditions Literature Cited
Carberry, J. J., "Catalytic and Chemical Reaction Engineering", McGraw-Hill, New York, N.Y., 1976. Denbigh, K. G., Chem. Eng. Sci., 8, 125 (1958). Kunii, D., "Notes on Fluidized Reactor Design." A short course given at Monash University, Clayton, Victoria, Australia, 1975. Kunii, D., Levenspiel, O., Ind. Eng. Chem. Fundarn., 7, 446 (1968). Kunii, D., Levenspiel, O., Ind. Eng. Chem. Process Des. Dev., 7, 481 (1968). Kunii, D., Levenspiel, O., "Fluidization Engineering", Wiley, New York, N.Y., 1969.

-rA = rate of catalytic reaction, mol/kg.s ub = rise velocity of bubbles in a fluidized bed, m/s Ubr = rise velocity of a single bubble in a bed which is otherwise

bubble-free, m/s umf = minimum fluidizing velocity, m/s uo = superficial entering gas velocity, m/s uo = volumetric flow rate of entering gas, m3/s W = weight of catalyst in the bed, kg

Received for reuiew October 14, 1977 Accepted April 11, 1978

Multiple Hydrodynamic States in Cocurrent Two-Phase Downflow through Packed Beds


Kin-Mun Kan and Paul F. Greenfield*
Department of Chemical Engineering, University of Queensland, St. Lucia, Australia, 4067

Evidence for the existence of multiple hydrodynamic states in trickle bed reactors with small particles is produced. These states are characterized by significantly different pressure gradients and different liquid holdups for identical gas and liquid flowrates. The determining factor is the maximum gas flowrate to which the packed bed has been subjected.

Introduction There is presently considerable interest in trickle bed reactors particularly with regard to their application to the hydrodesulfurization of crude oil. They are also being considered for biochemical systems where enzymes, immobilized on the pores and surfaces of packings, are employed to catalyze reactions between the gas and liquid phases. Satterfield (1975) has provided an excellent review of the current state of the art in the design of trickle reactors while Charpentier (1976) and Hoffman (1977) have specifically discussed the hydrodynamics found in such reactors. Studies on pressure drop and holdup and their relationship have been carried out by Larkins et al. (1961), Weekman and Myers (19641, Turpin and Huntington (1967), Charpentier and Favier (1975) and Specchia and Baldi (1977). Little attention, however, has so far been given to beds of small packings (less than 3 mm diameter) which, though of little industrial importance a t present, are essential for biochemical systems where very high Thiele moduli are encountered because of the high enzyme activity and diffusional resistances. Additionally, they are
0019-7882/78/1117-0482$01.00/0

often used in model trickle bed reactor systems. The small packings, together with the very high gas flowrates generally used, cause an early onset of flooding unless the fluids flow cocurrently. In this work, evidence is provided for the existence of a multiplicity of hydrodynamic states, the attainment of each being dependent on the past history of operating conditions of the packed bed. These states were characterized by different pressure drops which were measured. The existence of various hydrodynamic states has important implications in the design and operation of trickle bed reactors. Current design procedures are invariably based on liquid holdup in the reactor which determines the residence time of the reactants in the liquid phase. Since the holdup varies as a function of the particular hydrodynamic state, so must the performance of the reactor. Additionally, since the pressure drop in one hydrodynamic state can differ from that of another by as much as 5073, the operating costs must also be affected. Finally, the existence of these states implies that scale-up from small reactors presents problems and may help
0 1978 American Chemical Society

504

Ind. Eng. Chem. Process Des. Dev. 1986, 25, 504-507

Spiral Distributor for Fluidized Beds


Fan Ouyang' and Octave Levenspiel
Department of Chemical Engineering, Oregon State University, Corvallis, Oregon 9733 1

Performance characteristics of a new type of distributor, the spiral, are evaluated as a function of the number of slits used in the spiral and the fraction of open area. The quality of fluidization, pressure drop characteristics, and heat-transfer coefficients between the distributor and bed are measured and correlated with the geometry of the unit. Comparison with the sintered-plate distributor shows that depending on the fluidizing condffions, sometimes one, sometimes the other is preferred.

The selection of an appropriate distributor plate is an important consideration in the design of fluidized beds. First of all, the quality of fluidization and the amount of gas bypassing can be strongly influenced by the type of distributor used. Second of all, the pressure drop through the distributor may significantly increase the power consumption of the blower, often a major cost factor for operations. These two considerations are well-known (see: Kunii and Levenspiel, 1978). Certain processing operations also require having a cool distributor plate to support a hot fluidized bed. To prevent the sintering of solids during the reduction of iron ore is one example of such an operation. A second example is in the production of very pure silicon by the decomposition of silane SiH,(g)

Table I. Geometry of Spiral Distributors Tested distributor A B C 24 24 32 no. of blades 0.243 0.214 av gap, mm 0.813 1.43 4.1 70 open area 1.22 Table 11. Properties of Particles Fluidizedn
mean diameter, d,, M m 296 204 158 161 204 heat capacity at 5a o c , J/(kp.K)

US standard mesh no.


sand silicon zirconia
( I

heat

Si(s) + 2Hz

-40+50 -50+70 -70+80 -40+200 -50+70

density, kp/m3 2610 2610 2610 2330 5900

aoo aoo aoo


ai5 445

Here, cold silane enters a bed of hot silicon particles. The gas heats up and decomposes, and the silicon which is released then deposits and fuses onto the bed particles which grow in size. In this operation, it is imperative that the silane does not decompose while passing through the distributor plate, to deposit silicon therein and plug it. So, in the design of reactors of this type, it is important to know the heattransfer coefficient between the bed and its distributor plate. The purpose of this research is to evaluate a new type of bed support, called the spiral distributor, and compare its characteristics, such as the pressure drop, quality of fluidization, and heat-transfer coefficient, with that of a sintered-plate distributor. Experimental Section The apparatus consisted of a 152 mm i.d. fluidized bed resting on either a spiral or sintered-plate distributor, room-temperature air supply, heating lamps, and dust collection system. The spiral distributor was made of N overlapping plates shaped as sectors of a circle with a gap between the plates, as shown in Figure 1. This gap was largest at the bed wall and decreased to zero a t the center such that the fraction of open area is the same for all concentric circles. A t the center, the plates were welded together. Three versions of the spiral distributor of geometry given in Table I were tested and compared with a 1.6 mm thick sintered stainless steel plate with 5 - ~ m pores obtained from Mott Metallurgical Corp.
*On leave, Institute of Chemical Metallurgy, Academia Sinica, Beijing, China.

For nonspherical particles with no particular short or long dimension, Levenspiel (1984) gives d, = $dscleen.

Three types of bed particles were used in this study with properties given in Table 11. Silicon, obtained from the Union Carbide Corp., had a wide size distribution; the other materials had narrow size distributions. The pressure drops across the bed and across the distributor plate were measured with PIN connections and recorded with a Hewlett-Packard 7402A recorder, and the bed was heated radiantly, by spotlights, from above. Finally, the temperatures of the incoming air, distributor plate, and bed were measured by 0.254-mm chromel-alumel thermocouples and recorded by using an Esterline Angus PD 2064 data logger. The location of the thermocouples is also shown in Figure 1.

Results and Discussion 1. Pressure Drop vs. u o for the Spiral and the Porous-Plate Distributors. Figure 2 shows that the pressure drop across the spiral distributor is from 1 to 2 orders of magnitude smaller than for the sintered plate tested; however, this pressure drop increases more rapidly with an increase in gas velocity than for the sintered plate. 2. Quality of Fluidization with Spiral Distributors. Gas enters the bed tangentially through the many slots of the spiral distributor, and in shallow bed operations, this imparts a swirling motion to the solids. In deep beds (height > bed diameter), this action is restricted to the lower portion of the bed. Normal bubbling fluidization is seen in the main portion of the bed, above. 3. Comparison of the Quality of Fluidization with Beds Using Sintered Plates. The pressure fluctuation across a fluidized bed is a convenient, if rough, indication

0196-4305/86/1125-0504$01.50/0 0 1986 American Chemical Society

Ind. Eng. Chem. Process Des. Dev., Vol. 25, No. 2, 1986
thermocouple locations ,&.,
u .
(cm/secl
I85

505

u , / u , f
177-

With Sintered Plate Distributor

273 2 2 0 360 290

440 362

5 6 4 455

Figure 1. Spiral distributor showing the location of the test thermocouples.

620

500

With Spirol Distributor

Figure 4. Pressure fluctuations across fluidized bed of zirconia particles.

1 fluidized bed
L (cm)
V Thermocouple being lowered A Thermocouple being raised

center of

Figure 5. Bed temperature directly above the weld point of the spiral distributor.

pressure fluctuations, hence no consistent advantage of more slots or less open area. uo u u / ,, With Sintered Plate Distributor With Spiral Distributor Visual observations just above the spiral distributor (cm/secl show a horizontal convective movement of particles as 496 165 opposed to essentially stationary particles above the sin735 245 tered plate. I04 347 Finally, the plates of the spiral are welded together a t 136 453 v the center of the distributor, and this may result in particles heaping up a t this point. To measure the magnitude 190 633 7 of this effect, a thermocouple was affixed t o the weld point, 234 7 8 0 while a second, movable thermocouple was located directly 336 I12 above along the axis of the bed. Typical temperature measurements for these two thermocouples are shown in 460 153 Figure 5. The fact that the bed temperature stays constant to within 5 mm of the weld point indicates that very 510 I70 little solids, if any, are piled up a t the weld point. Figure 3. Pressure fluctuation across fluidized bed of silicon par4. Mechanism of Heat Transfer between Bed and ticles. Distributor Plates. Heat transfer between fluidized beds of the quality of fluidization; small fluctuations represent and immersed tubes or bed walls has been explained in small bubbles and less gas bypassing, while large fluctuaterms of various mechanisms: conduction through gas tions represent poorer contacting of the gas with solids. films, direct contact of bed particles with the surface, packets of emulsion resting on the surface, etc. Figures 3 and 4 show the recorded pressure fluctuations Heat transfer between fluidized beds and their distriacross the bed when supported by either the spiral or the bution plates has not been discussed until recently (Lesintered-plate distributor. A comparison of these traces venspiel et al., 1983). Here, the mechanism is somewhat shows that for low density solids (silicon, in Figure 31, the different in that gas flows normal to the surface, rather sintered plate gives better fluidization at low uo,but at uo than sliding along the surface. Thus, no gas film is present > loud, the advantage shifts to the spiral distributor. On at the heat-transfer surface, and transfer is only by direct the other hand, for high density solids (zirconia, in Figure impingement of particles with the surface. 4), the spiral distributor seems to give a better quality of The behavior of the spiral distributor is somewhat as fluidization a t all gas velocities. shown in Figure 6. A t the mouth of the slot, gas flows Similar pressure trace measurements for different spiral horizontally along the surface, entraining and dragging geometries (24 and 32 slots, 1.4% and 4.1% open area) and solids along with it; then the gas turns upward. Thus, heat using sand (narrow size distribution) and then using silicon transfer represents a complex situation somewhere be(wide size distribution) show roughly no difference in
Figure 2. Pressure drop across distributor plates vs. air velocity.

506

Ind. Eng. Chem. Process Des. Dev., Vol. 25, No. 2, 1986
/

,Sintered

"0

plate distributor tibed = m c m u . :9 7 cm/s

q ,
1

,
90

, .
1 0 0

50

60

7 0

80

0 e d Temperoture ( T I

Figure 8. Heat-transfer coefficient independent of temperaturedriving force. Figure 6. Bed behavior at the spiral distributor.

2Ocm

1500W

"}

coil heater

m:mj bed nth sprd dstnbutor

l5Zmm roundbed with sintered plate distributor

I
c

8 ; 10cy
"

zoo:
0

/
=
kgCp,g(Tg,out

,
20

,
30

,
40

IO

20

30

40

uo (cm/sec)

u. ( c d s e c )

Figure 7. Heat-transfer coefficient as a function of gas velocity for different distributors and different bed heights.

Figure 9. Heat-transfer coefficient increasing with a decrease in particle size.

tween the above two cases. Also, we may expect that the number of particles dragged along the surface and giving up heat is proportional to the flow rate or velocity of the gas issuing from the slots. If the heat given up by each particle is constant, then we would expect h to be directly related to uo. Assuming an isothermal distributor plate, a heat balance about the plate gives
Qtransferred

-0

500-

a
c

2 2 c

400-

300-

Tg,in)

= hA(Tbed

- Tdistributor)
I

Assuming, in addition, that the gas leaves the distributor plate a t the temperature of the distributor plate itself, or Tg,out = Tdistributor, we obtain

- 6 0

1 0

20

30

40

u,

(cm/sec)

Figure 10. Heat-transfer coefficient very different for different kinds of solids of the same size.

For sintered-plate distributors, the latter assumption should be quite reasonable; however, for the spiral distributor plate with its few large open channels, this assumption is questionable because Tg,out is likely to be Thus, the values of h somewhat lower than Tdiatributor. calculated by eq 1 will be somewhat higher than the true value of h, giving a conservative (higher) estimate for the temperature of the spiral distributor plate. Nonetheless, adopting eq 1 while recognizing its limitation and measuring the flow rate of gas and the temperatures of incoming gas, distributor, and bed give h directly. These measurements were made for a variety of situations, and the values were compared to the findings with a sintered-plate distributor, as discussed below. 5. Effect of Air Velocity and Bed Height on h . The results of Figure 7 show that h is independent of the bed height and that h rises linearly with the air velocity uo,for uo > ump This result is quite different from the findings with the sintered-plate distributor, where h rises to a maximum at about 2umfand then decreases slowly.

Thus, at low multiples of umf,the spiral distributor has a lower h than has the sintered plate; the reverse is true a t high gas velocity. For the distributor of Figure 7 , the transition occurs a t uo = 6ump Considering the discussion directly following eq 1,the true h for the spiral distributor plate should be somewhat lower than that reported in Figure 7, giving it added advantage over the sintered-plate distributor. 6. Effect of Temperature Driving Force on h If the assumptions underlying eq 1 are reasonable and if h given by this equation is to be useful, then the measured h should be independent of A T between bed and distributor plate. Figure 8 shows the measured h values for both the sintered-plate and spiral distributor for different AT, all other variables kept constant. The h value is found to be independent of AT, as expected. 7. Effect of Particle Diameter. Figure 9 shows the plot of h vs. uo for narrow size cuts of sand. The results are straight lines similar to that for silicon. The h value is found to be very sensitive to the particle size, with beds of smaller particles giving higher h values.

Ind. Eng. Chem. Process Des. Dev., Vol. 25, No. 2, 1986 507
Nx

5 - 5oo6oo. c
3
v1

24 blade distributor (4 I % open area)

24 blade distributor (I 22% open area)

P k

0 32 blade distributor (I 43% open area)

400c

f
V

d 5 y
c

300-

200-

8
100-

e
G O

IO

20

30

"

'

(2) Note that D and were not varied in our experiments. Figure 12 is a graphical representation of this correlation. Conclusion A new type of distributor for fluidized beds, the spiral distributor, is examined and its behavior is compared with the sintered-plate distributor. At high gas velocities, the spiral distributor gives a better quality of fluidization; a t low velocities, it gives lower heat-transfer coefficients between the bed and distributor. Equation 2 correlates the upper bound of the heat-transfer coefficient (see discussion after eq 1) with the properties of the fluidizing material and with the geometry of the distributor.
7

Distributor &

Silicon

Silicon and Sand

z
002-

00102

L 05
1 0

(T) dP PP (uo-unlf)
Figure 12. Dimensionless correlation for the heat-transfer coefficient between the bed and spiral distributors.

8. Effect of Particle Properties. Figure 10 shows the h vs. uocurves for narrow cuts of particles of the same size but having different properties and shows that the particle properties have a strong effect on the heat-transfer coefficient. 9. Effect of Open Area and Number of Slots in the Spiral Distributor. Figure 11 shows that for beds of sand, changing the fraction of open area from 1.2% to 4.1 % does not seem to affect the h value, while changing the number of blades in the spiral from 24 to 32 may have a slight effect on h. Unfortunately, since the fractional change in the number of blades is small, its effect on h is not certain. The same result is found with the other solids. 10. Empirical Correlation for Heat Transfer between the Bed and Spiral Distributor. Correlating the measurements with three kinds of solids, different particle sizes, and three spiral geometries and putting in dimensionless form gives

Acknowledgment We thank Riley Chan and Nick Wannenmacher for their technical assistance in constructing the experimental apparatus, to UNIDO for providing one of us (0.F.) the opportunity to study in Oregon, to JPL (Contract 956133) for the idea which led to this study, and to NSF (Grant CPE-8026799) for support (for 0. F.) while doing this project. Nomenclature A = cross-sectional area of distributor, m2 C , = specific heat, J/(kg.K) D = diameter of distributor, m d p = mean particle size for fluidization purposes, ni h = heat-transfer coefficient, W/(m2.K) k , = thermal conductivity of gas, W/(mK) m = mass flow rate, kg/s N = number of blades in the spiral distributor T = temperature, K umf = minimum fluidizing velocity, m/s uo = superficial velocity of fluidizing gas, m/s
p

Greek Letters = density, kg/m3 = viscosity, kg/(m-s) 4 = sphericity of particles

Subscripts g = gas s = solid

Literature Cited
Kunii, D.; Levenspiel, 0. "Fluidization Engineering": Krieger: Melbourne, FL,
1978

-.

Levenspiel, 0.; Larson, M.; Zhang, G. T.: Ouyang, F. Final Report to JPL, No.

956133,1983.
Levenspiel, 0. "Engineering Flow and Heat Exchange"; Plenum Press: New York, 1984;p 122.

Received for review February 6, 1984 Revised manuscript received September 3, 1985 Accepted September 19, 1985

CORRESPONDENCE

Reaction Rate Constant M a y Modify the Effects of Backmixing


SIR: I n a recent article Levenspiel and Bischoff ( 2 ) discussed the effect of backmixing on conversion in chemical reactors. Reactions of first and second order were considered, and graphs were
given showing the dependence of - = V O -L b D on the degree of conversion with _f: as UL parameter. I t is of some interest to investigate the effect of the reaction rate constant on these variables. The fractional conversion in a tubular reactor LOlong in case of piston flow is given by
xo = 1

- e-klPBLo/u

(1)

if back mixing occurs this modifies to


The reaction rate constant has a

profound effect on L / l o especially if DL

i s large

The boundary conditions used to obtain Equation 2 are not inconsistent with those of Wehner and Wilhelm ( 4 ) (their solution being quoted as Equation 8 in the original article), but represent a special case. Substituting the boundary conditions x + - 0 a t L + m in equation 8 of the original paper :

Using these equations calculated values of L/Lo were plotted against klpB with DL as parameter, a t a constant gas velocity of 1 foot per second, for 99% conversion as shown. I t is evident that k l p B has a profound effect on L/Lo especially if DL is large.

If the performance of a fluidized bed reactor is considered


DL = f(Re) suppose D L
= a

then
L

For L +

.this simplifies to
4a +
a ) 2 e 2oL

x = - (1

- (1 m.

UL

Considering Equation 3b it is evident L that - -t 1 if DL -+ o further if kip,<< 1


a)

U2--n

(dl+ 2..> > I,


2

2akiP~

1)

L O

If
lCIPBa Urn

so

that 4klPBDL
U2

< <

1.

(It is unlikely

i.e.-asa>

1,x-OifL-

-n

I n order to obtain a relation between

L and LOfor the same degree of conversion, equating x = xo gives


k@BLO
U

for any practical case that DL/u2 would be large enough to invalidate the argument). The denominator can be expanded in series to give

This may explain the practical ob-

or Le. the effect of back mixing is negligible if the reaction rate is slow. servation that with high gas velocities in fluidized bed reactors the conversions
VOL. 53, NO. 4
APRIL 1961

31 3

1 0 0

1 0 0

lo

6*

1 0

I 0 0 0 1

001

f=

I-x

0 1

00 1
f a

0 1

I-x

Sizes of actual vs. ideal reactors as influenced by the pertinent variables: extent of conversion, of backmixing, . magnitude reaction rate, and holding time o f fluid. For a -given holding time the decrease in conversion caused by backmixing ;s shown b y movement along dotted lines

obtained approximated to those for piston flow, whereas a t low velocities the conversion was poor.
Conclusions

I t has been shown that the effect of back mixing in a chemical reactor is appreciably influenced by the magnitude of the reaction rate constant. This has been shown analytically for a particular case of a first order reaction. For reactions of higher order numerical solution of the differential equation must be performed. The reaction rate constant had not been included in the correlation proposed by Levenspiel and Bischoff and their results are only applicable for a particular unspecified reaction rate constant.
Nomenclature
x

V L

= fractional conversion = volume of reactor, cubic ft.

= depth of reactor bed, ft. k,pB = reaction rate constant, set.-' D L = back mixing coefficient, square ft. per sec. u = linear gas velocity, ft. per sec.

These various families of lines allow the graphical solution of a wide variety of J. S Z ~ K E L Y problems with a minimum of trial and Imperial College of Science error; in addition, for most cases, D is and Technology proportional to u. University of London The second point wed like to mention London S.W. 7, England pertains to the application of the dispersion model to fluidized beds. Because the assumptions on which the SIR: Mr. SzCkely correctly points out dispersion model is based are far from that certain problems cannot be solved fulfilled in fluidized beds (in contrast to directly with our charts but require packed beds or tubular vessels), the additional calculations. He then goes dispersion model may not well characon to solve one such problem presenting terize such systems. The realization of a plot of the results. However, it should this fact has led to alternate treatments be noted that Mi-. SzCkelys plot is a of flow characterization o fluidized dimensional relationship whose use is beds in the last decade, such as the limited to a specific problem. We feel mixed model approaches of Mathis with the large number of variables inand Watson, A . I . Ch. E.Journal 2, 518 L, k , CO:ti, D, X , that it is volved; V , VO, (1956); May, Chem. Eng. Progr. 55, 49, best to retain a dimensionless plot Dec. (1959); and Lewis, Gilleland, which is more general, hence is more Glass, A.I.Ch. E.Journal 5 , 419 (1959). useful in a wide variety of problems. Thus we caution against the possible Our design charts are, as shown above, improper extension of the dispersion hereby extended by the family of curves model to fluidized beds. representing the reaction rate group k L / u OCTAVE LEVEXSPIEL A N D K. B. BISCHOFF for first order reactions and kCoL/u or Illinois Institute of Technology second order reactions. Other families Chicago 16, Ill. of curves can easily be drawn from these charts, such as the velocity independent group

(4) Wehner, J. F., Wilhelm, R. H., Chern. Eng. Sci. 6, 89 (1956).

Subscript 0 = piston flow References


(1) Lanneau, K. P., presented at Westminster, London, to Institution of Chemical Engineers, M a y 11, 1959. (2) Levenspiel, Octave, Bischoff, K. B., IND. ENG. CHEM. 51, 1431 (1959). (3) Smith, ,,J. M., Chemical Engineering Kinetics, pp, 365, McGraw-Hill, New

(!g)

(!A)

(g)

Correction

or

Effect of Vibration on Natural Convective Heat Transfer


I n this article by Robert Lemlich

or the size independent group

(g) (F) . (2)


=

[IKD. ENG. CHEM. 47, 1175 (1955)], the abscissa of Figure 6 (p. 1179) should read 0.1 Re,. In Equation 5 (p. 1178), the multiplying constant
should be changed from 0.00265 to 0.0000197.

York, 1956.

( 7 = (5%) ) .

kCaD

(5)

3 14

INDUSTRIAL A N D ENGINEERING CHEMISTRY

OCTAVE LEVENSPIEL and KENNETH B. BISCHOFF Illinois institute of Technology, Chicago 16, 111.

Backmixing in the Design of Chemical Reactors


This article provides a useful perspective on the importance of backmixing in several types of reactors. Design charts are given which set limits for the practicing engineer in the field of reactor kinetics and design
A

---

'4'
ADEQ~ATE design of chemical flow reactors rests on knowledge of two factors-the over-all rate a t which the reaction proceeds and the extent of backmixing occurring in the reactor. T h e first factor involves the determination of the rate equation, which for homogeneous reactions may be found most conveniently by using small scale laboratory batch equipment. As backmixing does not occur in such setups, the effect of reaction rate may be isolated and examined alone. Backmixing may also be studied independently, as in hydrodynamic studies in which the progress of the fluid is examined as it passes through the reactor vessel. h-umerous techniques have been employed and the results of the studies reported in many different ways. I n flow reactors backmixing, reaction rate and degree of conversion are tied together in a complex fashion. Because the role of backmixing is difficult to evaiuate in such situations, much present design is based on the assumptions that no backmixing occurs or that the contents of the system are well mixed and uniform in composition. Complete Backmix and Nonbackmix Reactors T h e idealized situation in which there is no backmixing is called plug flow, piston flow, slug flow, tubular flow, or nonbackmix flow, and is characterized by the fact that flow through the reactor is orderly and the residence times of all fluid elements are alike. For this situation, the volume of reactor required to effect a fractional conversion, x , of reactant A is given by

where r is the reaction rate of A and F is the feed rate of A into the reactor. T h e other extreme in flow conditions is characterized by so great a n extent of backmixing that any fluid in the reactor has an equal chance of being found a t the reactor outlet. This idealized situation is called complete or total backmix
VOL. 51, NO. 12

DECEMBER 1959

1431

Backmixing and Local Longitudinal Dispersion Number

Flow patterns in reactors may vary greatly; however, the resulting backmixing may often be characterized by a
VrO -

single dimensionless group--" the locaI longitudinal dispersion number" defined by Dlud, where D is the longitudinal dispersion coefficient. In the local longitudinal dispersion number, u is the fluid velocity and d is

V .

IO2

STREAMLINE FLOW

IN PIPES
D

Udt
00 1
f = I- X 0 1
IO
= FRACTION OF REACTANT REMAINING

Figure 1. The volume of a backmix reactor i s greater than that of the plug flow reactor required for the same duty. The size ratio rises rapidly with increase in both fractional conversion and reaction order

I
10'
Dv

IOZ
R e . Sc

lo3
E ?

lo4
=- udt

d t UP .

PDV

I O
flow, or stirred tank flow; the volume of reactor required for conversion x of reactant A is given by
V , = Fx/r (2 1

TURBULENT FLOW IN PIPES

T h e progress of many reactions may be approximated by the simple rate law


r = kC"

D udt

(3)

where n is the order of the reaction. For such cases, when fluid density remains unchanged, the comparison of sizes of reactors for a given feed rate is found from these equations to be

0.I

lo3

lo4

IO

'
I

IO6

(4)

PACKED BEDS

This result (Figure 1) shows that except for a zero-order reaction the complete backmix reactor always requires a larger volume than a plug flow reactor for a given feed rate? and that the effect of backmixing becomes increasingly important for higher reaction order and for approach to complete conversion of reactant. Hence approximations to ideality which may be permissible a t low fractional conversions would lead to large errors a t high fractional conversions. Partial Backmix Reactors Because a real reactor exhibits some degree of backmixing, the requiredvolume for a given duty should lie somewhere between the two extremes given by Equations l and 2. T h e problem then is twofold: to determine the extent of backmixing by a quantitative measure and then to use this measure with rate data to determine the necessary reactor size.

GASES

I
I

0 1 1 10-

lo-'
dP G o P

IO

IO2

lo3
I

IO,
D UdpE
I
I

PACKED BEDS LIQUIDS

I1 0.I

I IO

IO2

lo3

Figure 2.

Experimental findings on backmixing b y numerous investigators

local dispersion coefficients vary greatly with Reynolds numbers for flaw in pipes ( A and B) but are rather insensitive over wide ranges of Reynolds numbers far flow through packed beds ( C and D)

1432

INDUSTRIAL AND ENGINEERING CHEMISTRY

B A C K M I X I N G I N REACTORS
some size characteristic-i.e., tube diameter for flow in pipes or particle size in packed beds. D/ud varies from 0 for plug flow reactors to for complete backmix reactors; a n increase in D/ud corresponds to a rise in the degree of backmixing. Dispersion groups in general are reciprocals of the corresponding mass transfer Peclet groups. For pipes and packed beds, D / u d correlates simply with the dimensionless groups characterizing the flow conditions. Figure 2 , A , is the correlation for streamline flow in pipes; both Reynolds and Schmidt numbers affect the local dispersion number. This relationship was obtained theoretically and verified by three experimental runs (70). Figure 2, B, the recommended correlation for turbulent flow in pipes, is based on numerous laboratory and field studies with both liquids and gases. Preparation and detailed discussion of Figure 2, A and B, may be found elsewhere (7). Figure 2, C, shows packed bed local dispersion numbers found by various investigators (7-4, 6, 8). T h e results suggest a larger local dispersion number for liquid systems as compared to gas systems. Figurc, 2, D. represents a comprehensive study of liquid dispersion in ordered and randomly packed columns and shows two distinct regimes of flow (5). Reactor longitudinal Dispersion Number Though D/udcharacterizes the intensity of backmixing a t any point in a reactor, the reactor longitudinal dispersion number, D/uL, must be used in conjunction with the reaction rate equations to determine the reactor size. L is some measure of the length of fluid path through the reactor. I n cylindrical reactors, this is measured most conveniently by the length of packed or
200
100

fluidized bed or of vessel in an all-fluid reaction chamber. For vessels in which the intensity of fluid mixing varies from position to position, such as the spherical reaction chamber or the vessel with large end effects, the effective length has yet to be determined. If the local dispersion number may be considered uniform throughout the reactor, D,uL may easily be found. For a packed bed reactor D l u L = (D/ud,)(d,/L) (5) and for a tubular reactor D/uL (D/udi)(dt/L) (6) Chemical Reaction and Backmixing T h e differential equation governing the fractional conversion of reactant as a function of axial distance 1 in a reactor of length L is given by d2f - u df D- - kCon-f = 0 (7) dlz dl in which f = 1 - x is the fraction of reactant remaining (9). First-Order Reaction. For a firstorder reaction, or rl = 1, Equation 7 has been solved ( 7 7 ) under the appropriate boundary conditions corresponding to any homogeneous reactor vessel or any catalytic reactor bed in which the intensity of backmixing is uniform (constant D,ud throughout). In dimensionless form the solution relates the fractional conversion with both the reactor dispersion number, D / u L , and the rate group for the first-order reaction, kL,u. Thus

Figure 3, a graphical presentation of the above result in useful form, was prepared by combining Equations 1 and 8 and compares sizes of reactors required for a given feed rate for nonbackinix and partial backmixing conditions. Second-Order Reactions. T h e efFect of backmixing on reactor size requirement when a second-order reaction is taking place can be found by solving Equation 7 for n = 2. An analytical solution for cases of reaction order other than 1 is not kno1z.n because of the nonlinearity of the differential equation. Therefore, the differential equation was solved numerically on an IBM 650 digital computer. A fourth-order Runge-Kutta method was used, and i t was found that 25 increments were sufficient for the desired accuracy. Because the problem was of a boundary value nature rather than initial valuei.e., both the slope, df,dl, and the ordinate, f, were not known a t either boundary-the method of solution was necessarily trial and error. The process was started a t the reactor outlet where the slope was known to be zero ( 7 7 ) . A value of the ordinate a t this position was estimated and the Runge-Kutta method was used to calculate back to the reactor inlet boundary, a t which point the boundary condition

where f ( O + ) is the fraction of reactant remaining just within the reactor entrance, had to be satisfied to within a given error of 0.01-Le.. absolute values 1 - f(0-t) D/uL d f ( O + ) d ( l L) ab-

60 40

v 20 V.

V V O

1 0

6
4

2
I 0 001

001

01

1 0

0.0I

0.I

1 . 0

f = I - X = FRACTION OF REACTANT REMAINING

Figure 3. The volume o f an actual reactor with a given D/uL compared t o that of the ideal reactor required for the same duty
For a first-order reaction, analytical solution of the differential equation is possible (J J )

f = I-X=FRACTION OF REACTANT REMAINING Figure 4. The volume of an actual reactor with a given D/uL compared to that of the ideal reactor required for the same duty
For a second-order reaction, a numerical solution of differential equation was obtained on an IBM 650 computer

VOL. 51, NO. 12

DECEMBER 1959

1433

solute values L 0.01, If it was satisfied, the answers were punched out and a new set of values were calculated for a different value of D/uL. If not satisfied, the initial assumed ordinate was corrected by the computer, and a new curve was calculated. This process was repeated until the boundary condition at the reactor inlet was satisfied. The last correctedvalue of the reactant concentration a t the outlet was used as the total conversion for plotting. Figure 4 compares ideal and actual reactor sizes for a given feed rate. I t was prepared by combining Equation 1 with the result of the trial and error numerical analysis described above. Fractional-Order Reaction. Interpolation between Figures 3 and 4 allows estimation of reactor size to effect a given conversion for fractional order reactions involving a single reactant. Backmixing does not affect a zero-order reaction. Example A gaseous material undergoes a complex series of changes when in contact with a solid catalyst. To investigate the kinetics of this reaction a small laboratory reactor is constructed consisting of a 1-inch schedule 40 pipe packed to a depth of 1.5 inches with solid /l-inch diameter catalyst pellets. For a feed rate of 2.7 cubic feet per hour or GO = 49 Ib./hr.-sq. ft. the reactant is 99% decomposed. T h e reaction is of the first order with respect to the reactant. What depth of V,-inch diameter catalyst is required to yield a 9970 decomposition in a larger reactor to be constructed of 12-inch schedule 30 pipe if the feed rate is 4000 cubic feet per hour or Go = 547 Ib.,hr.-sq. ft.? Additional information and assumptions: p = 0.01 cp. ; isothermal conditions throughout and no net change in the number of moles of material passing through the reactor; neglect any nonuniform flow patterns due to channeling a t the pipe wall. Solution. FOR I-INCH REACTOR.The particle Reynolds number

Therefore, for the ideal case of plug flow 570 = 1.35 (3600) = 4860 hr.-1

CO
d,
d,

= initial concentration of reactant

A , moles A/cu. ft.


= particle diameter in a packed

FOR12-INCH REACTOR. T h e particle Reynolds number

bed, ft.
= diameter of pipe or tubular

reactor, ft. coefficient or the effective axial diffusion coefficient, sq. ft./ hr. D, = molecular diffusion coefficient, sq. ft./hr. Dlud = local longitudinal dispersion number, a reciprocal Peclet group for mass transfer, dimensionless D/uL = reactor longitudinal dispersion number, a reciwocal Peclet group for mass transfer, dimensionless f = 1 - X, fraction of reactant A remaining, dimensionless F = feed rate of A into reactor, moles A/hr. = superficial mass velocity, 1b.l Go (hr.) (sq. ft.) k = reaction rate constant, moles A l--rr hr,-l = axial distance from entrance of 1 reactor, ft. L = length of pipe or reactor, ft. = order of reaction as defined in n Equation 3 1 = reaction rate, rate of disappearance of reactant A , moles A / (hr.) (cu. ft.) = F/VCo space velocity, hr.-I U = average velocity of flow, ft./hr. V = volume of reactor, cu. ft. X = fraction of reactant A converted into product, dimensionless e = porosity, dimensionless = viscosity, lb./(ft.) (hr.) J ! = density, lb./cu. ft. P

= longitudinal dispersion

and from Figure 2, C, the local dispersion number D/ud, = 0.55. At this point the unknown length, L, enters in both variables S V and DjuL, so the solution involves successive trials. Assume to start that no backmixing occurs in this reactor. For this case using the value of S V o found above vel. feed/hr. SVO = 4860 h r . 7 = vol. reactor 4000
(0.797)L

Therefore the reactor depth L = 1.032 feet = 12.4 inches Check the assumption of plug flow made above. D/uL = (Djudp)(d,/L) = (C.55)(0.25/12.4) = 0.0111 From Figure 3

(2T)

Therefore the assumption is justified, and L = 12.4 inches. From this example, we see that backmixing usually plays only a minor role in large packed bed reactors, although it may be significant in small laboratory reactors. Precautions A number of precautions must be observed in applying this procedure in the scale-up of process equipment. The first involves the isothermal requirement. As heat effects depend on the surface-volume ratio of the reactor, scale-up will result in larger heat effects and probably nonisothermal temperature distributions which must be accounted for by a complex analysis or a mean reaction rate constant. Another precaution involves the unjudicious extrapolation of backmixing data from small to large equipment. Gross flow patterns could vary considerably; this probably is the case between laboratory and industrial-sized fluidized units with their different degrees of bypassing of the fluid in the form of bubbles. T h e degree of backmixing may not be uniform throughout a reactor because of entrance effects, nonuniform cross section, etc. This may be dealt with by using a n average D/uL or going directly to the distribution functions from which the D values are obtained. Nomenclature = defined by Equation 9, dimensionless A = reactant C = concentration of A , moles A / cu. ft.
a

SUBSCRIPTS = nonbackmix, slug or plug flow 0 situation = complete backmix or stirred tank situation literature Cited
(1) Carberry, J. J., Bretton, R. H., A.I.CI1.E. Journal 4, 367 (1958). (2) Danckwerts, P. V., Chem. Eng. Sci. 2, 1

From Figure 2, C, the local dispersion number


D/udp = 0.5

Thus the reactor dispersion number


D/uL = (D/udp)(dp/L) = (0.5)(0.25/1.5) = 0.0833

iYow from Figure 3

-SVo/SV

VIVO

1.35

but for the experimental run - = vol. feed/hr. sv vol. reactor [2.7/(7.5 X

(1953). (3) Deisler, P. F., Jr., Wilhelm, R. H., IND. END.CHEM. 45, 1219 (1958). (4) Ebach, E. A,, Ph.D. dissertation, University of Michigan, Ann Arbor, 1957. (5) Jacques, G. L., Vermeulen, T., U. of Calif., Berkeley, Rept. UCRL-8029 (November 1957). (6) Kramers, H., Alberda, G., Chem. Eng. Sci. 2,173 (1953). (7) Levenspiel, O., IND. ENG.CHEM. 50, 343 (1958). (8) McHenry, K. W., Jr., Wilhelm, R . H., A.1.Ch.E. Journal 3, e3 (1957). (9) Smith, J . M., Chemical Engineering Kinetics. Chap. 11, McGraw-Hill, New York, 1956. (10) Taylor, G. I., Proc. Roy. Soc. 219A, 186 (1953). (11) Wehner, J. F., Wilhelm, R. H., Chem. Eng. Sci. 6,89 (1956). RECEIVED for review May 1, 1959 ACCEPTED August 28, 1959

3600 hr.-l

1434

INDUSTRIAL A N D ENGINEERING CHEMISTRY

78

Ind. Eng. Chem. Fundam., Vol. 18, No. 1, 1979

istence of an optimum sound intensity above which the rate of reaction decreases rather than increasing. As indicated in eq 9, AK increases with the energy t released by the collapse of the cavitation bubbles. A bubble will collapse if the collapse time is equal to or shorter than one-half the period of oscillation of the sound wave. With an increase of sound intensity or applied voltage to the sound transducer, the cavitation bubble size becomes larger, increasing the collapse time and reducing the chance of collapse at a fixed frequency. A reduction in the number of bubbles that collapse will result in a lesser amount of energy release. On the other hand, larger drops release greater energy after their collapse; the released energy is therefore optimum a t the applied voltage which gives the largest bubble size with collapse time equal to one-half the period of sound oscillation. Equation 14 is considered valid when V is within the voltage corresponding to the optimum energy release.

rl, r2 = radius of reacting ions R = radius of the sphere of diffusion Ti = period of an eddy, s U = mean eddy velocity, cm s-l U , = eddy velocity, cm s-l V = transducer voltage, V V , = transducer voltage just before the onset of cavitation Greek Letters
t

= energy dissipation rate, cm2

= kinematic viscosity, cm2 s-l

X = eddy length, cm

Literature Cited
Barren, E., Porter, C., J. Am. Chem. Soc., 63, 3434 (1941). Batchelor, G. K., J. Fluid Mech., 5 , 113 (1959). Benson, D., "Mechanisms of Inorganic Reactkms in SduUofts", p 204, W a w H i Y , London, 1968. Chen, J. W., Kalback, W. M., Ind. Eng. Chem. Fundam., 6 , 175 (1967). Couppis, E. C., Klinzing, G. E., A1Ch.E J., 20, 485 (1974). Fogler, S . . Barnes, D., Ind. Eng. Chem. Fundam., 7 , 222 (1968). Hinze, J. O., "Turbulence", p 165, McGraw-Hill, New York, N.Y., 1959. Levich, V. G., "F'hysim-chemical Hydrodynamics", p 215 RenticeHall, Englewocd Cliffs, N.J., 1962. Moelwyn-Hughes, E. A., "The Chemical Statics and Kinetics of Solutions", p 99, Academic Press, New York, N.Y., 1971. Plesset, M. S., "Bubble Dynamics and Cavitation Erosion", in L. Bjorno, Ed., "Proc. Symp. Finite Amplitude Wave effects in Fluids, Copenhagen", IPC Science and Technology Press, Surrey, England, 1971. Shinnar, R., Church, J. M., Ind. Eng. Chem., 52, 523 (1960). i e l d in Cavitation" in L. D. Rozenberg, Sirotyuk, M. G., "Energy b h n c e of Sound F Ed., "High Intensity Ultrasonic Fields", Plenum Press, New York, N.Y., 1971.

Nomenclature

C, C1 = dimensionless constants C2 = constant in eq 7 C3 = constant in eq 9 C4 = constant in eq 14 D = molecular diffusion coefficient, cm2 s-l De = eddy diffusion coefficient, cm2 s-l E = energy released through cavitation k = wave number kB = Batchelor characteristic wave number kK = Kolmogoroff characteristic wave number K'R = reaction rate constant in cavitation fields KR = specific rate constant, mo1-ls-l k 1 = constant in eq 1 AK = increase in specific rate constant Nsc = Schmidt number, u/D

Department of Chemical Engineering T h e Polytechnic of Wales Pontypridd, Mid Glamorgan CF37 1D L Wales, United Kingdom

M. Seraj-ud Doulah

Received for review August 15, 1977 Accepted September 29, 1978

CORRESPONDENCE
Correspondence on Analysis of Gas Flow in Fluidized Bed Reactors
Equation 1of Bukur's paper 11978) assumes that all gas in excess of minimum fluidization goes through the fluidized bed as bubbles alone, while eq 2 assumes that these fast rising bubbles are accompanied by clouds which take up some of the dense phase. These represent two distinctly different physical pictures which if used together lead to a logical contradiction. Potter et al. and Kunii and Levenspiel explained gas reversal in fluidized beds as a consequence of bubble wake action and solid circulation. But by cleverly manipulating this one contradiction Bukur was able to come up with negative gas flow and then duplicate Potter's equations without even introducing the concept of bubble wake and solid circulation! This shows the immense power of contradiction. Allow one and you can prove anything.

Literature Cited
Bukur, D.

B.,Ind.

Eng. Chem. Fundam.. 17, 120 (1978).

Department of Chemical Engineering Oregon State University Corvallis, Oregon 97331

Octave Levenspiel

Professor Levenspiel's correspondence which concerns my recent paper actually represents a critique of Partridge and Rowe's (196613) model. Equations 1 and 2, as well as eq 3 , 7 , and 10 of my paper, are the basic equations of this model and have been used by many investigators up to the present time. Professor Levenspiel claims now that Partridge and Rowe's model is based on a logical contradiction, which is embodied in eq 1 and 2.

Equation 1 does not assume that all gas in excess of minimum fluidization goes through the bed as bubbles alone as stated by Professor Levenspiel. It represents a general statement which gives the physical description of the bed, and it says that the total gas flow is divided between the bubbles a n d associated clouds and the interstitial gas. It is essentially a material balance equation and no assumptions are made about the actual gas flow

Ind. Eng. Chem. Fundam., Vol. 18, No. 1, 1979

79

rates. Equation 2 specifies the flow rate of gas which passes through the bed in form of bubbles alone (visible bubble flow rate). This expression was derived by Partridge and Rowe (1966,a),and their analysis produced the result in agreement with one of the basic postulates of the simple two-phase theory regarding the bubble flow rate. However, in the simple two-phase theory one doesnt take into account the existence of bubble clouds as is done in Partridge and Rowes model (1966a,b). There are no logical contradictions in this development, and Partridge and Rowe are not using two distinctly different pictures of the bed. However, there is some evidence, both theoretical and experimental, that eq 2 tends to overestimate the visible bubble flow rate. A theoretical analysis of Lockett e t al. (1967) predicts smal1e.r values of GB than eq 2 . Experimental measurements of Grace and Harrison (1969), Geldart (1971), and Chavarie and Grace (1975a,b) indicate that the visible bubble flow rate is smaller than eq 2 predicts. Equation 20 of my paper, which is proposed for

the visible bubble flow rate, is in qualitative agreement with these experimental results. It seems that Professor Levenspiel is not familiar with Potters analysis (1971) of Partridge and Rowes model, which was cited in my paper. Potter obtained eq 9 of my paper, with f , being substituted by eq 10 without introducing concepts of bubble wake and solid circulation.

Literature Cited
Chavarie, C., Grace, J. R., Ind. Eng. Chem. Fundam., 14, 75 (1975a). Chavarie, C., Grace, J. R., Ind. Eng. Chem. Fundarn., 14, 79 (1975b). Geldart, D., Ph.D. Dissertation, University of Bradford, U.K., 1971. Grace, J. R., Harrison, D., Chem. Eng. Sci., 24, 497 (1969). Lockett, J. J., Davidson, J. F., Harrison, D., Chem. Eng. Sci.. 22 1059 (1967). Partridge, B. A., Rowe, P. N., Trans. Inst. Chem. Eng., 44, T349 (1966a). Partridge, B. A., Rowe, P. N., Trans. Inst. Chem. Eng., 44, T335 (1966b). Potter, 0. E., Fluidization, J. F. Davidson and D. Harrison, Ed., 332-333, Academic Press, London, 1971

Faculty of Technology University of Novi Sad 21000 Novi Sad, Yugoslavia

Dragomir B. Bukur

CORRECTION

In the Correspondence, Catalytic Oxidation of Phenol in Aqueous Solution over Copper Oxide, by A. I. Njiribeako, R. R. Hudgins, and P. L. Silveston [Ind.Eng. Chem. Fundam., 17, 234 (1978)], the reply by Professor Katzer is correct: the value of 148 g/L is a thousandfold too high. It should be 148 mg/L.

OCTAVE LEVENSPIEL
Bucknell University, Lewisburg, Pa.

Longitudinal Mixing of Fluids Flowing in Circular Pipes


Design charts incorporating data from the literature can be applied to pipeline studies and design of chemical reactors

problem of fluid self-mixing in vessels has been studied by Danckwerts ( 3 ) ,who considered it in the light of the distribution of, or spread in, the residence time of the fluid in the equipment. Danckwerts ( 4 ) also summarized experimental techniques for the measurement of the distribution of residence times. These methods all involve introducing tracer material into the incoming fluid stream and measuring its concentration in the outgoing fluid . stream, but they differ in the way the tracer is introduced. It may be introduced continuously but in amounts varying sinusoidally with time (6, 70) or in any cyclic manner. It may also be introduced continuously in concentration Co after an initial time, before which no tracer is introduced a t all. Finally it may be injected in a slug, essentially instantaneously. Typical tracer input and output concentrationtime diagrams illustrating these methods of tracer introduction are shown (Figure 1). Danckwerts called the dimensionless form of the output signal (Figure I$), in which the ordinate is measured as C/Co and the abscissa as vO/V, the F-curve. He also called the dimensionless form of the output signal (Figure l,c), in which the ordinate is measured as CV/Q and the abscissa as &/V, the C-curve. The variable nO/V, which may be called reduced time, is unity for the period required to pass one void volume of fluid through the vessel. With the above choice of variables the F-curve rises from 0 to 1, the area under the C-curve is always 1, and when integrated the C-curve yields the corresponding F-curve. Though originally developed for vessels, these tracer diagrams are also applicable to flow in pipes. Numerous investigators (3, 6, 9-12, 78-22) have pointed out that longitudinal fluid mixing may be characterized

THE

by a diffusion-type model governed by Ficks law of diffusion

found by methods suggested by cyclic tracer injection experiments (Figure 1,a).

ac a2c 8 = D -dX2

(1)

Fluid Mixing in Pipes


The diffusion-type model rests on the assumption of homogeneous fluid mixing, and under certain conditions it may be applicable to pipe flow. When applicable unique functional relationships would be expected between the system and fluid variables and the longitudinal dispersion coefficient, D. If mixing is not homogeneous, as in the case of a short pipe (end effects) or in a pipe with bends, average D values result. These would not be expected to correlate directly with the system variables unless the actual flow patterns were accounted for. Because of the complexity of such situations only flow through straight pipe is considered. Dimensional Analysis. Consider the variables affecting the mixing process during steady state flow in a straight pipe from the point of view of their dimensions. These would include the variables usually pertinent to flow problems: the fluid density, p ; the viscosity, y; the velocity of flow, U; the pipe diameter, d, and the roughness, E , which is a measure of the depth of its surface irregularities. The roughness has been shown to be important only in turbulent flow (76). In streamline flow the radial interchange of material between adjacent fluid layers moving at different velocities may also be expected to affect D. This effect involves the molecular self diffusion of the fluid as characterized by the molecular diffusion coefficient D,. Thus, in general, for both streamline and turbulent flow
= *~[P,P,U,d,e,Dvl

where D , which may be called the longitudinal dispersion coefficient, uniquely characterizes the mixing process., Assuming Ficks law to be applicable, equations have been derived (73, 79) relating D to the expected C-curve. Thus

(2)

which is a family of C-curves with D/uL as parameter. Similarly the Fcurve, which is the integral of the C-curve, is an unique function of DluL. I t has been shown (73) that the spread or width of the C-curve as measured by its variance d is related to D/uL by

UL

=;

( d * m -1)

(3)

For the special case that D/uL < < 1, the C-curve as given by Equation 2 approaches the normal or Gaussian error curve
(1
; > t

($)

(4)

in which case Equation 3 reduces to

D _
UL

-1
-

zu2

(5)

Equations 3 and 5 afford a convenient method for finding D from experimental F- and C-curve data; D can also be

According to the Buckingham TI-theorem, this functional relationship may be


VOL. 50, NO. 3
MARCH 1958

343

expressed groups:

in terms of dimensionless

..s

.a
.d

3
2
'B

Q,

5 .E
tj

Equation 6 shows that a dimensionless group involving the longitudinal dispersion coefficient should be a function of the Reynolds number, the Schmidt number, and a relative roughness number which are measures of the viscous, molecular diffusion, and roughness effects, respectively. Streamline Flow. As pipe roughness is not a significant variable in streamline flow Equation 6 in this case should reduce to

D
U d =

+,C(?)> (A)]

(7)

and consequently a plot of D/ud us. dup/p (Figure 2) should result in a family of curves with p / D , p as a parameter. In a series of articles (78-20) Taylor considered the self-mixing of fluids flowing in pipes from a theoretical point of view. For streamline flow he showed (78, 20) that when

u %mE
L

d2

Q)

P
Ly X

Lo

bn

+
0

radial mixing is great enough compared to longitudinal mixing to assure a uniform cross-sectional concentration of material everywhere. For this situation the diffusion-type model is applicable, and the longitudinal dispersion coefficient, D , can be found by relating the radial to the longitudinal mixing of material. The radial mixing of material is assumed to be caused solely by molecular diffusion, while the longitudinal mixing of material is assumed to be caused

C Q
N

9
;?;;?; 3 o o

10 .m t
0

?
a

z
lW

to

U ,

. .
m

DI b-

10

0
o

'9
0 0

Y
I m n

Z O 0

2 %

s a

9
2 . C 0
N .

I
0

S
m
m

" 8 9 2

0 ' 0

0:

k
INPUT
V

- TIME

Figure 1. Typical recordings for various modes of tracer injections


a.

b.
C.

Cyclic input Step input Jump of instantaneous input

344

INDUSTRIAL AND ENGINEERING CHEMISTRY

solely by the parabolic velocity distribution of flow in a pipe with no end effects. The final result of Taylor's analysis is

in turbulent flow, Taylor (79) considered the diffusion model to be universally applicable, in which case the longitudinal dispersion coefficient is given by

D=-- d2u2 192 D,

(9)

e = 3.57 dj
ud

which, in the dimensionless form of Equation 7, yields

The condition of applicability Equation 10 is that

of

This is Equation 8 in dimensionless form. When plotted (Figure 3), Equation 10 predicts that all experimental data should fall on a straight line of slope 1 and intercept 1/192. When Equation 10 is plotted as in Figure 2 a family of curves of slope 1 and intercept (1/192) (p/pD,) should result. Turbulent Flow. In turbulent flow molecular diffusion should not be expected to contribute significantly to the transport of material, this process being greatly overshadowed by turbulent eddy mixing. Thus, Equation 6 should reduce to

where f is the Fanning friction factor. The analysis is based on the assumption that Reynolds' analogy holds. According to this assumption the intraphase transfer of heat, momentum, and matter by a fluid in turbulent flow are analogous. Hence the extent of fluid mixing (transfer of matter) can be calculated by shear stress and pressure drop information (transfer of momentum). To find D from Equation 13 the value of f must be known. Experimentally, f has been found to be a function of the flow and system variables, or

2 ud = a 4 [(?),

( ; ) I

(12)

and consequently a plot of D/ud us. dup/fi should result in a family of curves with E / D as a parameter. In his theoretical analysis of mixing
Figure 2. Correlation for both streamline and turbulent flow ranges of D/ud, a longitudinal dispersion number, as a function of Reynolds number
@,0,0.ideal conditions

thus checking Equation 12. A number of analytical expressions of the above relationship are available (75, 76). However, as these are relatively complex compared to a graphical representation, the latter is used. The result of Taylor's analysis is shown in Figure 2 by a curve based on Equation 13 and a value of f for commercial pipe given in Perry (75). A family of curves for different values of the parameter e/d are obtained if the charts for f presented by Moody are used (74); however, for the present purposes the one curve found in Perry (75) is sufficient.

cient information to allow them to be reanalyzed are considered. Experiments in the streamline flow region, done by Taylor (78) and Fowler and Brown (7), are shown in Figure 3. Taylor's three runs were made by measuring the dispersion of potassium permanganate in water flowing in a tube 0.0504 cm. in diameter. The conditions of Equation 11 were satisfied, and Figure 3 shows that the points are close to the theoretical curve. In plotting the points a problem arose in choosing a value from the molecular diffusion coefficient, Do, for the potassium permanganate-water system as this varies considerably with concentration (8). The value chosen was that reported at the average potassium permanganate concentration. By contrast to the good agreement found by Taylor, the data of Fowler and Brown are much below the theoretical curve. However, this is not surprising as the condition for applicability of the diffusion model, Equation 11, was far from being satisfied. For the points further away from the curve L 1 28.8 d - w 220

r?)(&)

while for the points closest to the curve

Discussion
To test the reliability of the predictive equations they should be checked against experiment. For this purpose all published investigations which present suffi-

&&A.

Data obtained in pipe with few bendse.g., commercial pipeline V,V. Mixing between fluids of different physical properties 0. Flow in artificially roughened pipe 0. Curved pipe

As the requirements of Equation 11 are approached the experimental points approach the curve; however, in no case are the conditions of Equation 11 satisfied. The transition from streamline to turbulent flow is shown by Fowler and Brown to occur a t a Reynolds number of about 2300. A marked change in the magnitude of mixing as measured by D/uL accompanies this transition.

OhE

,
l
R

0 ALLEN 8 TAYLORIIJ 0 FOWLER 8 BROWN171 # TAYLOR(I91

BENDS O N E FLUID
W E WITH BENDS T W O FLUIDS ~
CURVED PIPE

2 ~~),!H8,",",:':E,,,,

A DAVIDSOH E T AL 151

W HULL 8 KENT I91

V SMITH 8 S C H U L Z E l I 7 I
~ ~ ~ ~ p E

o FOWLER a BROWN 171

~ 0 TAYLOR ~ (191 ~

TAYLOR 1191

I
I

1 0

VOL. 50, NO. 3

MARCH 1958

345

For turbulent flow conditions considerable data are available under widely varying flow conditions. Figure 2 and Table I summarize this information. However, not all the experiments were done under the ideal conditions of straight pipe and uniform fluid properties, for much of the information was obtained in commercial pipelines. When fluids of different physical properties were used the Reynolds number chosen was that for the 50 to 50 mixture. Figure 2 shows that as the Reynolds number decreases from 10,000 to 2300, experimental findings deviate increasingly from the theoretical curve. This discrepancy may be explained by considering the assumption on which the analysis of dispersion in the turbulent flow region is based. The assumption is that molecular diffusion plays an insignificant role in the mixing process. Now in the laminar sublayer, molecular diffusion does play a role in the mixing process so that when this layer has an appreciable thickness, as it does for Reynolds number below 10,000, deviations from theory might be expected. Thus one might expect the Schmidt number to appear as another parameter with increasing importance as the Reynolds number decreases to 2300. The only data available in this region are for a narrow range of Schmidt numbers, so the significance of this parameter cannot be determined until further experiments are carried out, Beyond a Reynolds number of 20,000 there is but one point obtained under ideal conditions. Nevertheless. such data that are available allow the inference to be drawn that the theoretical curve is the best estimate for mixing in this region. S o t e that most data scattering occurs in experiments with mixing of two different fluids. Additional data (2) not included in Table I on fluid mixing as a function of pipe length show that the length of gasoline-gasoline and kerosine-gasoline contamination in successive flow varies, respectively, as the 0.482 and 0.529 power of the pipe length for pipes from 4 to 12 inches in diameter. This checks closely with the diffusion mode1 which predicts (see Equation 5) that the length of contaminated section varies as the 0.5 power of the pipe length. Example. A pipe 1 foot in diameter and 100,000 feet long is to pump two Auids successively. The Reynolds number of a 50 to 50 mixture of the two fluids is 20,000. Determine the amount of fluid at the pipeline outlet which is contaminated due to intermixing of components. Fluid is considered to be contaminated if the composition of the components is between 5 and 95%. From Figure 2 for the given flow conditions _- 0.3
ud

But
d

L =
Therefore

$ = ($)

(e)

10-6

concentration of tracer a t point of introduction into fluid, dimensionless d = diameter of pipe, L D = longitudinal dispersion coefficient defined in Equation 1, L2/ T D, = molecular diffusion coefficient,
=

c o

For this low value of D/uL, Equation 5 may be used to find the length of the mixing zone. Thus
=

f =
-

L2/T
Fanning friction factor, dimensionless, defined by T~ =
fPU2/2

2.45

x 10-3

L = a length of pipe
dimension of length M = dimension of mass Q = volume of tracer of unit concentration which would correspond to actual amount of tracer introduced into fluid, L3 T = dimension of time u = vL/V, average flow velocity, L / T v = volumetric fluid flow rate, L2/T volume of pipe, L3 x = space coordinate surface roughness, L E = B = time, measured from time of injection of tracer into the flowing fluid, T P = viscosity, M / L T P = density, MIL3 uz = variance of C-curve. in terms of pipe lengths, dimensionless T O = shear at pipe wall, M / L T i

where u is measured in pipe lengths of fluid. But the central 90y0 of the area under the normal curve will lie within 3.29 u. Therefore, the section contaminated because of intermixing is
= = =

(3.29)(2.45 X 10-3) 8.06 X 10-3 i e lengths of fluid (8.06 X IO+? ?105)(~/4) 634 cubic feet of fluid

v =

Conclusions

If the longitudinal dispersion coefficient is a variable which actually characterizes the mixing process, unique functional relationships with usual flow variables should yield definite correlations among the variables. Thus charts based on the assumption that longitudinal fluid mixing is analogous to the process of diffusion should yield definite correlations among the variables. If the longitudinal dispersion coefficient does not characterize the mixing process the charts should show a scattering of data. Results of numerous investigations under widely varying conditionse.g., pipe diameter, a factor of 300; Schmidt number, a factor of 1000-are brought together into a narrow band thus validating the use of the diffusion model. O n the basis of theory and experiment these charts are universally applicable in the turbulent region, but they can only be used under certain conditions in the streamline region. In the streamline region the few data available indicate that the theoretical predictions are reliable. However, much further experimental work needs to be done, The scatter in the turbulent region data can be explained by the fact that the conditions of applicability of the diffusion model, that of using straight pipe, were not satisfied. Further experiments should be carried out in this region to narrow the range of uncertainty in the recommended curves, to determine the possible roIe of the Schmidt number at low Reynolds numbers, and to investigate the effects of pipe roughness and curvature; these variables have been shown to increase the extent of mixing above that found in straight smooth pipe.
Nomenclature

literature Cited

C = concentration of tracer, fluid volume of tracer per volume of fluid, dimensionles:

(1) Allen, C. M., Taylor, E. A., Trans. Am. SOC. Mech. Engrs. 45, 285 (1923). ( 2 ) Birge, E. A,, O i l Gas J . 176 (Sept. 20, 1947). (3) Danckwerts, P. V., Chem. Ens. Sci. 2, l(1953). (4) Danckwerts, P. V., 2nd. Chemist 3, 102 (1954). ( 5 ) Davidson, J. F., Farquharson, D. C., Picken, J. Q., Taylor, D. C., Chem. Eng. Scz. 4,201 (1953). ( 6 ) Deisler, P. F., Jr., Wilhelm, R. H., IND. ENG. CHEY. 45,1219 (1953). (7) Fowler, F. C., Brown, G. G., Trans. Am. Inst. Chem. Engrs. 39, 491 (1943). (8) Furth, R., Ullmann, E., Kolloid-Z. 41,307 (1927). ( 9 ) Hull, D. E., Kent, I. Mi., IND. ENG. CHEM. 44, 2745 (1952). (10) Klinkenberg, A,, Sjenitzer, F., Chem. Eng. Sci. 5 , 258 (1956). (11) Kramers, H., Alberda, G., Zbid., 2, 173 (1953). (12) Lapidus, L., Amundson, N. R., J . Phys. Chrm. 56, 984 (1952). (13) Levenspiel, O., Smith, W. K., Chem. Eng. Sci. 6 , 227 (1 957). (14) Moody, L. F., Trans. Am. Sac. Mech. Engrs. 66, 671 (1944). (15) Perry, J. H., Chemical Engineers Handbook, 3rd ed., Sect. 5, McGraw-Hill, New Tork, 1950. (16) Rouse, H., Elementary Mechanics of Fluids,: chap. 7, Wilep, New York, 1946. (17) Smith, S. S., Schulze, R. K., Petroleum Engr. 19, 94 (1948); 20, 330 (1948). (18) Taylor, G. I., Proc. Roy.Soc. 219A, 186 (1953). (19) Zbid., 223A, 446 (1954). (20) Ibid., 225A,473 (1954). (21) Wehner, J. F., Wilhelm, R. H., Chem. Eng. Sci. 6,89 (1956). (22) Wilhelm, R. H., Chern. Erg. Prarogrr. 49,150 (1953). RECEIVED for review March 25, 1957 ACCEPTED June 28, 1957

346

INDUSTRIAL AND ENGINEERING CHEMISTRY

TWO-PHASE FLOW IN PACKED BEDS


Evaluation of Axial Dispersion and Holdup by Moment A n a l s i s
V. E. SATER1 AND OCTAVE LEVENSPIEL

Illinois Znstiute of Technology, Chicago 76, Ill.

An experimental investigation of the extent of axial dispersion in both gas and liquid phases in a bed packed with /*-inch Raschig rings or Berl saddles is presented. Radioisotopes, argon-41 in the gas phase and iodine-1 31 in the liquid phase, were used as tracers. The concentration at a position along the bed could then be measured from outside the bed without disturbing the flow patterns by using scintillation detectors. By comparing the moments of the responses measured a t two points along the column, the velocity and dispersion coefficient were evaluated for that section of the bed between the two points. This use of the moments of the curves circumvented the need for an exact tracer input (such as a step function) and greatly simplified the experimental problems by allowing an arbitrary sloppy pulse input. Correlations show the effect of flow rates and packing geometry on the axial dispersion numbers of both phases.

M are solved by making simplifying assumptions concerning


the flow behavior of the system. Probably the most familiar example is in heat transfer. I n calculating film coefficients for laminar flow through pipe, the conduction term in the axial direction is neglected on the assumption that it is small compared to the convective transfer term. This greatly simplifies the problem and leads to no great error in predicting film coefficients. I n the same manner, axial dispersion or mixing in the axial direction has been neglected in the calculation of mass transfer coefficients in packed towers. True values of mass transfer coefficients, those based on the actual existing concentrations of materials, can be evaluated only when the extent of this axial dispersion is known. There are two useful models for describing the mixing in a packed bed. The first (the mixing cell model) assumes that the packing can be characterized by a series of completely mixed cells. The mixing in a bed is therefore a function of only one parameter, the number of mixing cells in the bed. The other approach (the dispersion model) assumes that the various factors causing axial mixing can be described by a diffusional-type process superimposed on plug flow. This is reasonable if the length over which a single mixing effect acts is small and if a large number of such events take place in the vessel. In laminar floiv, the factors causing axial mixing are molecular diffusion and the overtaking of fluid elements due to the velocity profile. For turbulent flow, an additional factor, turbulent eddy diffusion, plays a role in causing axial mixing. With the dispersion model then, deviation from plug flow is accounted for by a flux term

OST

chemical engineering problems involving fluid flow

transfer of material in and out of stagnant pockets and local channeling, in addition to the effects of turbulent eddy diffusion. Admittedly, the use of a single term to account for these many effects may be an oversimplification, but it should describe the flow characteristics in a packed bed better than the plug flow model.
Technique of Evaluating Effective Diffusivity

T o describe the flow through a packed bed mathematically, some simplifying assumptions must be made. The first is that the axial velocity is constant over the cross section of the bed. If the ratio of tower diameter to packing size is greater than 8 to 1, this assumption is reasonable. The second assumption is that no concentration gradients exist in the radial direction. If the axial velocity is uniform and any material is introduced over the entire cross-sectional area, thi3 also is a realistic assumption. To evaluate the mixing term, a tracer is injected into the main flow stream and its behavior as it moves through the bed is analyzed. By applying a material balance to a cross section of the bed, the following partial differential equation is obtained :

-D - $I u bX2

b 2 C

bc

ax

bC

dt

Putting the equation into a dimensionless form gives :

-($)(>+-=-bC

dz2

bz

bc be

-D-

dc
dx

where D is termed the axial dispersion coefficient and accounts for all the factors causing mixing in the axial direction, such as
1

Present address, Arizona State University, Tempe, k i z .


I&EC FUNDAMENTALS

T o solve this equation, appropriate boundary conditions must be used. I n most cases, the concentration is taken to be zero at time equal to zero. The form of tracer input serves as the second boundary condition-e.g., a change in concentration of tracer entering the bed in the form of a pulse, step, or sine function of time can be used. The third boundary condition specifies conditions at the end of the bed-e.g., for a n infinitely long bed, the concentration must be finite throughout the bed. The solution of Equation 2 will give the concentra-

86

tion at any point in the bed as a function of time with the axial dispersion group, D / u d , as a parameter. Thus by comparing this function with the experimental concentration-time data, the dispersion group can be determined. The technique has been used xvith the three forcing functions mentioned above (3-8, 70, 72-74, 7'9). The method described above is easily handled mathematically but difficult to realize experimentally. To inject a tracer into a flowing stream so that the concentration-time relationship corresponds to a siinple mathematical function such as a sine wave without disturbing the flow conditions is not an easy task. This problem can be circumvented by injecting an arbitrary amount of tracer into the bed and measuring the concentration-time response a t two positions far enough downstream so that the flow patterns are re-established (Figure 1). One of the responses would be equivalent to a boundary condition, even though it could not be represented by an analytical function of time. I t then becomes a problem of relating the shapes of the two response curves to the dispersion occurring in the bed. Aris ( I ) , Bischoff and Levenspiel (Z), and Sater (77) have developed the solution for the physical model illustrated in Figure 1 based on a comparison of the moments of the two response curves. The restrictions on the tracer input are that the concentration must be equal to zero for time less than zero and return to zero after some finite time interval. Thus any nondescript pulse can be used as a n input. As the tracer moves through the packing, the pulse will flatten out as shown in Figure 1 as a result of axial mixing. The variance of each curve is an indication of the spread of the curve and, therefore, the difference in the variances of the two curves is a measure of the mixing occurring between the two measurement positions. Likeivise, the dij'ference in the first moments (or the centers of gravity) of the two curves is directly related to the flo\v velocity in the test section. The mathematical relationships between the moments and the physical parameters of the model are given by the following [see ( 2 ) and (77) for details on the derivations] : (3)

where

pi=--

-_

cit at

cidO
0

''

J,

Pp

(5)

c,dt .

and ct is the ratio of length of packing downstream of the second measurement point to the length of the test section (Figure 1). The functions, f and g, are complex and account for the fact that the packed bed has a finite length and thus end effects must be considered. If CY can be made large enough, the values o f f and g in the above equations can be set equal to zero and Equations 3 and 4 reduced to:
Ap =

( 7 )

Aa2 = ):2 (

(f)

Using Equations 5 and 7 lvith the experimental data will directly yield a value for the axial velocity of the measured phase, u , and indirectly, the fraction of the bed occupied by that phase, E , since:
u = uo

(9)

Equations 4 and 6 allow the axial dispersion group, D l u d , and the axial dispersion coefficient, D,to be calculated for the test section.
Tracers

INJECT

MEASUREMENT

ENTRANCE

PAC KING

EXIT

T o avoid disturbing flow patterns by inserting probes or drawing off samples, radioactive tracers measurable from outside the column could be used. But because the radiation had to pass through the packing and the tolver wall, it was necessary to use a strong gamma emitter. Also, a water-air system was studied, so the gas-phase tracer had to have a low vapor pressure to avoid any transfer between phases. (This experiment was performed not to measure adsorption rates but to describe the flow characteristics of each phase.) Unfortunately, the only insoluble gaseous gamma emitter is argon-41, which has a half life of only 2 hours and therefore is not commercially available. So quartz ampoules were filled with high purity argon gas, irradiated by the Argonne Sational Laboratory for a 6-hour period using a flux of 6 X 1 O I 2 neutrons/sq. cm.-sec. (the gas had a specific activity of 1.9 millicuries per milliliter a t the time of removal from the reactor), rushed to the Illinois Tech campus, and used immediately. They were sized to contain about 0.5 millicurie of argon-41 each at the time of use. Iodine-131 in the form of a sodium iodide solution was available from Tracerlab, Inc. It has a low vapor pressure and a half life of 8.1 days, making it a good liquid-phase tracer. The iodide solution \vas purchased in glass vials containing about 50 microcuries each at time of use.
VOL. 5

INJECTION

CURVE I
't-

CURVE 2

Figure 1. Schematic of experimental setup with general response curves

NO. 1

FEBRUARY 1966

87

AIR T O V E N T

Apparatus

POULE

INSERTION

DEVICE, S E E F I G , 3

SUPPORT

PLATE

TO

SEWER

Figure 2.

Packed tower

A packed tower was constructed using 4-inch i.d. glass pipe with tees fitted on the end as shown in Figure 2. The feed streams and tracers were introduced through the branch sides of the tees. The pipe was packed to a depth of about 1 2 feet with '/?-inch ceramic Berl saddles or '/?-inch Raschig rings, both obtained from the hl. A . Knight Co. This height of packing probably caused channeling but was necessary in order to ignore end effects and use Equations 7 and 8. The use of a redistributor would have introduced disturbances in the floiv patterns in the test sections. A support plate for the packing was made by perforating a '/*-inch brass plate so that the perforations accounted for 507, of the total plate area. The air leaving the top of the tower \vas passed through a rubber hose to the laboratory's ventilation system and there diluted with more air before being vented from the roof. Because of the radioactivity, the system was made leakproof. A U-bend at the bottom of the tower served as a liquid seal. Air was taken from the building supply lines and the flow rates were measured Tvith rotameters. At low liquid flow rates, the \vater \vas taken from an overhead tank in order to maintain a steady flow. .4t the higher rates, the water was taken directly from the building main. Because of the relatively small cross-sectional area of the toxver, a pipe with a hole drilled in the wall served as a liquid distributor. The device used to insert the gas-phase tracer into the tower is shoivn in Figure 3. The ampoule was placed in the smasher by opening the gate valve and lo\vering the ampoule into position with a pair of tongs. The valve was then closed and the smasher \cas slowly pushed into the tower against the butt plate. After the flow was established, the rod !vas given a hard push and the piston in the smasher shattered the glass or quartz ampoule. The smasher \vas perforated ivith small holes so that the tracer would quickly be flushed from the smasher.
Instrumentation

AMPOULE IN

ER

GLASS

FRAGMENTS

The radioactivity 1% as detected by two Baird-.i\tomic Model 812B scintillation probes positioned and shielded with lead as shoivn in Figure 4. The shielding allowed only the radiation from a thin cross section of the packing to be detected. The probe outputs Xvere fed into research ratemeters which converted the pulses into a voltage proportional to the rate of pulses received. This voltage was recorded with a two-channel Offner Dynagraph recorder and the strip charts \vere used as the response curves.
Results

Figure 3.

Device for loading ampoules into tower

42"

rh

60"

7 1e
SCINTILLATION

LEAD

SHIELDING

Figure 5 represents a typical pair of response curves. The lines drawn by the recorder are not smooth but fluctuate about a general trend because of the random nature of the radiation. The tails of the curves approach the zero level very slo\vly and at this low radiation level the data are not accurate. To facilitate the processing of these curves, a smooth line was drawn by eye to average out the fluctuations. The tails have a heavy weight when the second moment is calculated. Because the values of the concentration are not reliable at large values of time, a cutoff point was established (Figure 5). The contributions to the moments of the curve to

39"

!UTECTOR
Cut- o f f

Po i n t

Figure 4. detectors

Position of scintillation

+t---Figure

5. Reproduction of actual recorder response for

a typical run

aa

I&EC FUNDAMENTALS

the left of the cutoff point were calculated by a numerical technique. Levenspiel and Smith (73) have shown that the tail of the response to a true impulse would decay exponentially. It was assumed that the response to an arbitrary pulse would also decay exponentially and this was confirmed by plotting the data to the right of the maximum of the response curves on semilogarithmic coordinate paper and obtaining a straight line. This line was then extrapolated beyond the cutoff point and used to caiculate the contributions of that portion of the curve to the total moments. The area under each curve and the first and second moments were calculated using this method. Details are given elsewhere (77). Single Phase. Eighteen runs were made with the tower operating under single-phase conditions-Le., with water only flowing through the bed-as a check on the experimental method, since an abundance of data is available in the literature for single-phase sysi.ems. Because the results substantiate previous work, the numerical values are not presented in this paper (77). Two-Phase Flow, Liquid Phase. The fraction of the bed occupied by a phase can be determined from the difference in the first moments of the response curves for that phase. In the

case of a liquid phase, this fraction is known as the total holdup, H2.Holdup data found by this method are compared with literature values in Figures G and 7 . Although holdup appears to vary with the gas rate, a statistical analysis of the data indicates that any such dependence is masked by the experimental error (77). This substantiates the findings of previous workers that the liquid phase is not affected by the gas flow rate at conditions below the loading point. The total holdup is the sum of the operating holdup, Hop, and the static holdup, Hst. The static holdup is the amount of liquid remaining in the bed after the liquid inlet is shut off and the column allowed to drain. It is therefore independent of the liquid flow rate and is a constant depending only on the packing used. Shulman, Ullrich, and Wells (78) reported values of the static holdup as 0.0325 and 0.0317 for l/Z-inch Raschig rings and l/2-inch Berl saddles, respectively. Thus the operating holdup, the holdup dependent on flow rate, can be obtained by difference. Otake and Okada (76) obtained a good correlation of their holdup data along with points obtained by other investigators by using the following equation :

Hop = 1.295 ( R e ) ~ ~ f ~ ~ ~ ( G a ) ~ - ~ * ~ ~(1 ( 0) ad)

0 v 0

Present Work Jesser a n d E l g i n ( l l ) S h u l m a n , U l l r i c h , a n d W e l l r (18) E l g i n and W e i s s ( 9 )

Ht

0.04

400

1000
L - (Ib./hr,-sq.ft.)

10,000

60,000

Figure 6. Comparison of liquid holdup data with literature values, '/Z-inch Berl saddles

A
0

Present Work Jesser and Elgin(1l) Shulrnan,UIIrich, and Wello(l8)

A A

0
I
I I

I l l

400

1000

10,000

60,000

L- ( I b./ hr.7

sq. f 1 . )

Figure 7. Comparison of liquid holdup data with literature values, 1/2-inch Raschig rings
VOL. 5

NO, 1 F E B R U A R Y 1 9 6 6

89

1 00

ro

IO

4 IO

100

1000

(+IFigure 8. Comparison o f operating holdup values with the correlation of Otake and Okada (16)

- - - - i15%
where

Correlation of Otake and O k a d a (76) (see Equation 10) range

(Re), = dL/,u, Reynolds number of the liquid phase (Ga), = d3gp2/pc",Gallileo number of the liquid phase a = surface area per unit volume of packed bed d = nominal particle diameter Figure 8 shows the data of the present work according to this correlation. Because the dispersion coefficient, D, depends on the flow patterns in the bed, the dispersion group should be a function of the same factors that determine the holdup, and because of the success of Equation 10 in correlating holdup data, the same form was used to correlate the dispersion group.

T o evaluate exponents B, C, and E, a t least three of the variables, one in each dimensionless number, must be varied. Only two variables, L and a, were deliberately varied in the present work. For the Berl saddles, a was equal to 155 sq. feet per cu. foot and for Raschig rings, a was equal to 114. The viscosity varied between 1.1 and 1.35 centipoises, depending on the temperature of the tap water when the run was made. A least squares analysis of the data gave the following values for the constants along with the 95% confidence limits:

The 95% confidence limit on the exponent is zt0.238 and includes the error incorporated by assuming the viscosity and surface area to be constant. Otake and Kunugita (75) obtained liquid phase axial dispersion coefficients by measuring the response to a step change introduced a t the top of the tower. I n the correlation of their results, they use a Reynolds number containing the interstitial velocity. Thus the total holdup must be known before a Reynolds number can be calculated. I n order to determine the axial dispersion group, two correlations, one for the holdup and one for the dispersion group, must be used. T o compare their results with the present work, their raw data were obtained and plotted on Figure 9. Also included are three points reported by McHenry and Wilhelm (74). Two-Phase Flow, Gas Phase. The amount of liquid in a packed bed will determine the shape and size of the channels through which the gas can pass. The greater the holdup, the smaller these channels will be. Thus the flow pattern of gas is a function of the holdup, which in turn is a function of the Reynolds and Gallileo numbers of the liquid phase and also the packing characteristic, ad. And the patterns will also depend on the Reynolds number of the gas phase. Therefore,

A B

= = C = E =

19.4 0.747 i 0.147 -0.693 zk 1.095 1.968 =k 0.997

(1lb)

Equation 11 is shown in Figure 9. Because of the slight variations in viscosity and surface area, the values of C and E are unreliable, as shown by the wide confidence limits. Therefore, the viscosity and surface area were considered constant and the dispersion group was correlated with the Reynolds number only, with the following result:

Not enough variation was obtained in the Gallileo number to determine its functionality. By using data at constant gas flow rates for each packing material, the dependence of the gas-phase dispersion group on the liquid phase Reynolds number was found to be of the form :

(g)
90

= 7.58

dL

0.703

10-3

( ) L

It was found that the value of n does not vary significantly with either rate of gas flow or type of packing. A similar treatment of the data for constant liquid rates showed that the effect of gas flow rate could be described as:

I&EC FUNDAMENTALS

0.7

0-

1/2 INCH BERL SADDLES

4Figure 9.

EQUATION I I

Liquid-phase axial dispersion

0.0 I 0.4

io

0 '/n-inch Berl saddles, present work


'/n-inch Rcischig rings, present work + 10-mm. Raschig rings, McHenry and Wilhelm ( 1 4)

1.0
IO
-Oa6

x1 0
Figure 10.

- 0.00 259 (F) L

- - 0.785-cm.
V 1.55-cm.

I?aschig rings, Otoke and Kunugita ( 1 5 ) Raschig rings, Otake and Kunugita ( 1 5 )

Gas-phase axial dispersion

These two results can be combined as:

(:)>.

K(ReG)-m

form of this dependency cannot be determined but in this case is assumed to be of the form (ad)'. This is strictly an arbitrary choice with no foundation, but is used for convenience. Grouping the Raschig ring and Berl saddle data and fitting them to
ReL

lo-"

(12)

A least squares analysis performed on the data for each packing separately yielded the values and 95% confidence limits shown in Table I. The confidence ranges of both m and n overlap, so there is a possibility that the packing shape has no effect on these constants and that the difference in results is due to experimental error. T o test this hypothesis, a Student's t test as described by Volk (20) was performed on the data, with the result that if there were no difference in the real values of the constants, there would be one chance in six that experimental error could cause differences as great as shown in Ta.ble I. This is not a small enough chance to justify treating the Berl saddle and Raschig ring data separately. Although confidence limits on the coefficients, K , were not evaluated, it is obvious that the coefficient depends on the packing geometry. Since only two geometries were used, the

c:)c

= K(ad)S(ReG)-mX

(13)

gave the following results and 95y0 confidence limits. Number of points = 34 m = 0.668 i 0.184 n = 0.00259 0.00053 s = 2.58 f 0.78 K = 0.0585 Figure 10 shows the data plotted according to Equation 13. Because of the uncertainty as to the effects of packing geometry, the above result should be applied only to systems similar to those used in this work. DeMaria and White (7) arrived at the following correlation for I/d-, 3/g-, and 1/2-inch Raschig rings:
(:)c =

2.4 (Rec)-0.20 x 10

- (0.013 -

0.088 d ) R e L dc

(14)

Table 1. Constants from Analysis of Individual Sets of Data '/2-Inch Bed Saddles '/n-Inch Raschig Rings Constant

K
No. of data points
m n

3.72

6.23

o.oo228 0.54 f 0.20 o,ooo52 o.oo299 0.79 f 0.29 o.ooo93 15 19

Because they used only one type of packing, a term to describe the effect of packing geometry was not included. H ~ they did vary the packing size and found that coefficient n in Equation 13 is a function of packing diameter. Since DeMaria and White used only one tower diameter, 4 inches, they are not justified in including the tower diameter, d,, in their correlation.
VOL. 5

NO. 1

FEBRUARY 1 9 6 6

91

2*o

1
a
G

Nomenclature

= surface area of packing per unit volume of packed bed,

sq. ft./cu. ft.

= concentration of tracer
= = = =

d
dl

nominal packing size, ft. axial dispersion coefficient

= column diameter, ft. = acceleration of gravity, ft./(sec.)2

D
g G Ga h

mass flow rate of gas phase, lb./(hr.-sq. ft.) Gallileo number, defined in Equation 10 = length of test section, ft. Hop = operating holdup, cu. ft./cu. ft. H,, = static holdup, cu. ft./cu. ft. H t = total holdup, cu. ft./cu. ft. L = mass flow rate of liquid phase, lb./(hr.-sq. ft.) Re = Reynolds number, d L / p or dG/p t = time, sec. u = axial velocity, ft./sec. u, = superficial velocity, ft./sec. x = position along bed, ft. z = dimensionless position, x / h GREEK LETTERS = ratio of length of end section to length of test section C Y 0 = defined in Equation 5 = fraction of bed volume occupied by a particular phase e e = dimensionless time, ut/h = first moment, defined by Equation 5 or viscosity when p used in Reynolds or Gallileo number p = density, lb./cu. ft. = second moment, defined by Equation 5 u SUBSCRIPTS = position of response curve = end section G = gas phase L = liqud phase
literature Cited

1 0 0

zoo
R e L = (L)
t L

300

Figure 1 1 . Comparison o f results o f present work with data o f DeMaria and White (7), /Z-inch Raschig rings

To compare Equation 14 with Equation 13, the appropriate values of d, a, and dt for l/Z-inch Raschig rings are used in the equations, with the result:
This work:

i E

DeMaria and White:


=

e)G

2.4 (ReG)-0.20 X 10-o*oo2 ReL

(144

A graphical comparison of the equations and the ranges covered are shown in Figure 11. For similar conditions, Equation 14 gives dispersion groups that are approximately one eighth the values given by Equation 13 based on the present investigation. Further work is needed to resolve this large disagreement. I n the meantime, since the results of the present study on Equation 13 indicate a greater deviation from plug flow, its use will give a conservative performance estimate in design applications.
Acknowledgment

The authors gratefully acknowledge the financial support of the Esso Research and Engineering Co. and the cooperation of the M. A. Knight Co., which supplied the packings.

(1) Aris, R., Amundson, N. R., A.Z.Ch.E. J . 3, 280 (1957). (2) Bischoff, K. B., Levenspiel, O., Chem. Eng. Sci. 17, 245 (1962). J., Bretton, R. H., A.Z.Ch.E. J . 4, 367 (1958). P. V., Chem. Eng. Sci.2, 1 (1953). (5) Danckwerts, P. V., Jenkins, J. W., Place, G., Zbid., 3, 26 (1954). .,(6 Deisler, P. F., Wilhelm, R. H., Znd. Eng. Chem. 45, 1219 (1953). (7 DeMaria, F., White, R. R., A.Z.Ch.E. J . 6 , 473 (1960). (8 Ebach, E. A., White, R. R., Zbid., 4,161 (1958). 9) Elgin, J. C., Weiss, F. B., Znd. Eng. Chem. 31,435 (1939). 10) Jacaues, G. L.. Vermeulen, T., University of California Radiation Laboratory, UCRL-8029 (November 1957). (11) Jesser, B. W., Elgin, J. C., Trans. Am. Znst. Chem. Engrs. 79.277 (1943). .-,(l$ Kramers, H., Alberda, G., Chem. Eng. Sci.2, 173 (1953). (13) Levenspiel, O., Smith, W. K., Zbid., 6,227 (1957). (14) McHenrv. K. W., Wilhelm, R. H., A.Z.Ch.E. J . 3, 83 (1957). 15) Otake, T.; Kunugita, E., Chkm. Eng. (Japan) 22,144 (1958). 16) Otake, T., Okada, K., Zbid., 17, 176 (1953). 17) Sater, V. E., Ph.D. thesis in chemical engineering, Illinois Institute of Technology, Chicago, lll., 1963. (18) Shulman, H. L., Ullrich, C. F., Wells, N., A.Z.Ch.E. J . 1, 247 (1955). 19 Strang, D. A., Geankoplis, C. J., Znd. Eng. Chem. 50, 1305 (1958). (20 Volk, W., Applied Statistics for Engineers, p. 260, Mcdraw-Hill, New York, 1958.

\ - - -

7 -

\--

RECEIVED for review March 29, 1965 ACCEPTED August 11, 1965

92

I&EC FUNDAMENTALS

BUBBLING BED MODEL FOR KINETIC PROCESSES IN FLUIDIZED BEDS


Gas-Solid Mass and Heat Transfer and Catalytic Reactions
DAIZO K U N I I

Chemical Engineering Department, University of Tokyo, Tokyo, Japan


OCTAVE LEVENSPIEL

Chemical Engineering Department, Illinois Institute of Technology, Chicago, Ill.

606 16

Reported physical and chemical rate data in fluidized beds are correlated and explained b y the recently proposed bubbling bed model. In gas-solid mass transfer, where the measured rates are much lower than expected, the model with its flows and gas-solid contacting fits the reported data, including the effect of gas velocity and particle size. This treatment is extended to gas-solid heat transfer with similar results. In catalytic reactions, where conversion often falls far below the extreme of backmix flow, this model satisfactorily explains the reported data. In matching the model to these three kinds of reported data, the one parameter, effective bubble size, was adjusted to fit the data. In all cases the sizes selected were physically reasonable.

bubbling bed model, recently proposed by Kunii and (1968b), described the flow of gas and the contact of gas and solid in fluidized beds in terms of one parameter, the effective bubble size in the bed. I n this paper this model is used to explain and correlate the observed rate data on physical and chemical phenomena occurring in fluidized beds. We consider in turn gas-solid mass transfer, gassolid heat transfer, and solid catalyzed chemical reactions.

TLevenspiel

HE

Gas-Solid M a s s Transfer

For gas-solid interchange between a single sphere and the surrounding fluid the Sherwood number for mass transfer and the Nusselt number for heat transfer are given by the well known expressions Sh = 2 Nu = 2

+ 0.6 ( S ~ ) l / ~ ( R e ) l / * + 0.6 (Pr)ll3(Re)l/*

(1)

These transfer groups asymptotically approach a lower limit of 2 as the Reynolds number becomes very small and this limiting value can be calculated directly from conduction theory. For an assemblage of particles, such as a fixed bed, the lower limiting value for the Sherwood or Nusselt numbers has been calculated by theory to be 2 or somewhat above 2, depending on the bed voidage (Rowe, 1963; Rowe and Claxton, 1965; Zabrodsky, 1963). I n direct opposition to these theoretical predictions, the measured Sherwood and Nusselt numbers of over a dozen investigators (Kunii and Suzuki, 1967) do not level off but keep falling continuously as the Reynolds number is lowered. For fluidized beds the observed mass and heat transfer behavior is similar to that of fixed beds, in that the Nusselt and Sherwood numbers drop continuously as the Reynolds number is lowered. Figure 1 illustrates these findings. Although the findings of theory and experiment may seem to be contradictory, this is not necessarily so if we recognize that two different kinds of coefficients are being considered. First, there are the local coefficients which refer to the true

. 10-1

IO

102

io3

Re
Figure 1 . General form of correlation for gas-solid mass transfer and heat transfer

driving forces for transfer in terms of the actual concentration or temperature of gas bathing the individual particles. Then there are the over-all coefficients, which are measured for the experimental section of bed with its channeling and bypassing of solids by flowing gas. The theories which predict a leveling off of Sherwood or Nusselt numbers concern local coefficients; the coefficients reported by experiment are those measured in beds with their bypassing and channeling but which assume plug flow of gas through the bed. Zabrodsky (1963) recognized that the large extent of byVOL. 7

NO. 4

OCTOBER 1 9 6 8

481

passing of solids by bubbling gas could be the reason for the lowered observed coefficients, and he proposed a model which viewed that the surplus gas beyond that needed for minimum fluidization bypassed one or more rows of solids, then mixed completely with the percolating gas. This process is repeated throughout the bed. Kunii and Suzuki (1967) developed a somewhat similar model for fixed beds and showed that it could quantitatively account for the reported data by proper selection of the number of layers bypassed by the gas flowing through the fixed bed. I n essence, these analyses suggest that if the detailed flow pattern of gas through the fluidized bed is known it should be possible to account properly for the observed coefficients in terms of the local or true coefficients. The bubbling bed model, developed recently by Kunii and Levenspiel (1968b), gives a simple representation of the bubble flow, the emulsion flow, and the interaction of these streams in a fluidized bed, and the aim of this paper is to show how the flow pattern of this model with the local transfer coefficients given by Equation 1 can account for all the gassolid heat and mass transfer data which have been reported in the literature. Bubbling Bed Model. When a bed is fluidized by gas a t a superficial velocity, uo, in excess of the minimum needed, urn,, gas voids, called bubbles, are seen to rise through a denser continuous region, called the emulsion. Assuming a constant bubble size in the bed or section of bed under consideration, the bubbling bed model then described the flow in terms of one parameter, the effective bubble size, db. Only consider flow conditions where u, > 2umf. This model first shows that only large fast bubbles (ub > 5urn,) will be present in the bed. Each such bubble contains gas which circulates within the bubble and in a thin layer of surrounding emulsion, called the cloud. This bubble gas stays distinct and segregated froin the gas rising through the emulsion. The fraction of bed consisting of bubbles is then given by

solids remains at urn,/m,; hence the upward velocity of gas flowing through the emulsion is

(7)
I n vigorously bubbling beds the circulation of solids is high enough (downward in the emulsion) so that the gas flow will reverse itself and be downward in the emulsion. For typical values of CY between 0.25 and 0.40 flow reversal in the emulsion occurs when - > 6 w l l
UO

urn,

(8)

I n beds with fast rising bubbles, the volume of cloud surrounding each of the bubbles is given by

The interchange of gas between bubble and cloud is given by a transfer coefficient volume of gas going from bubble to emulsion and vice versa Kbc = (volume of bubble) (time)
= 4.5

( T )+ (T)
5.85

and analogously for heat transfer we can show that

For nonporous and nonabsorbing solids, the interchange of gas between cloud and emulsion is governed by gas diffusion alone and is given by

6=- uo

- urn1
ub

where the velocity of rise of bubbles ub is given by

Since these quantities are transfer coefficients, the over-all interchange of gas between bubble and emulsion

If the emulsion remains at minimum fluidizing conditions and the bubbles are essentially solid-free, the bed voidages are related by

Now, rising bubbles are observed to drag a wake of solids behind them, consequently setting u p a circulation pattern in the bed with downward solid flow in the emulsion. This downward velocity is given by

C Y 8 u b
UQ =

For a given physical situation (fixed hf, emf, 9, Be, a ) and imposed flow rate uo, this model gives the complete flow pattern of gas through the bed in terms of one measured quantity which can be controlled to some degree, the bubble size db. This then is the one parameter of the model which determines all else, Details of the derivations of these equations are given by Kunii and Levenspiel (1968 b). Distribution of Solids in a Fluidized Bed. T o describe the kinetic phenomena of gas-solid heat and mass transfer, the distribution of solids among bubble, cloud, and emulsion regions is needed. So, in terms of volumes, define

1-6-ffcUs

(5)

where CY is given by volume of wake dragged upward behind a rising bubble volume of a bubble

Yb =

volume of solids dispersed in bubbles volume of bubbles

CY=

)= (0.25 to 0.4)

volume of solids within cloud and wake Yc = volume of bubbles (6)


Ye =

v.
v b

volume of solids within emulsion volume of bubbles

Next, assuming that the emulsion remains a t minimum fluidizing conditions, the relative velocity between gas and
482
I & E C PROCESS D E S I G N A N D DEVELOPMENT

By material balance we can then relate these quantities by

Thus, any two y values with this relationship give all the solid fractions needed to treat the kinetics of physical and chemical operations in fluidized beds. Experiments in beds of uniformly sized bubbles indicate that
70

1 surface of particle

dhA

kd*

dt

(true local concentration difference

g 0.001

0.01

(16)

Values of y c can be estimated by considering a spherical bubble and accounting for solids in both cloud and wake. This gives

Additional experimental values for some of these solid fractions and other related quantities are given by Kunii and Levenspiel (196 8a). Definitions of Mass Transfer Coefficients. For convenience consider the mass transfer process to be a n absorption by solid of a component A , which is present in the fluidizing gas stream. Since the flow of gas through the emulsion is very small, or even downward in vigorously bubbling beds, we can reasonably ignore its minor contribution to total flow. Thus fresh gas enters the bed only in the form of gas bubbles and the decrease in concentration of A in the bubble is a measure of its absorption. T h e mass transfer coefficient for such processes can be reported in a number of ways. First, based on the total surface area of particles we may define an over-all coefficient (kd)over-all [cm./sec. ] by

I n fluidized beds we want to use two different kd* values, one for solids dispersed in rising gas bubbles where the relative velocity between gas and solid can be approximated by the terminal velocity of solids, u t , the other for solids in the emulsion phase where the slip velocity corresponds to minimum fluidizing conditions. We call these coefficients ( k d * ) and (kd*)ml, respectively. Each of these coefficients (kd)oyer-all, (kd*) t , and @a*), f has its corresponding Sherwood number, (Sh)over.all,(Sh*) t , and (Sh*)ml, and all of these quantities are used below. Evaluation of Mass Transfer Coefficient. Consider what happens to gaseous A initially present in the bubble as the bubble rises in the bed. Some of the A is absorbed by solids within the bubble; another portion is transferred to the cloud where a part of it is absorbed and the rest is transferred further into the emulsion. I n the emulsion we assume that gaseous A is completely absorbed, and this is reasonable in the light of the long contact time in the emulsion. Thus we have bubble absorption by solids in where

)+

(transfer) to cloud

(24a)

transfer to cloud

absorption i n ) (cloud by solids

(transfer to) emulsion

(24b)

and where

(
or in terms of the falling concentration of A in the rising bubbles (kd)O W - S ~ surface of solids volume of bubble

transfer to emulsion

) (
=

absorption in emulsion by solids

I n symbols these word equations are

(Cab

- CAS)
Kbc(CAb

Yb(kd*)l a(CAb

- (;A31

Kbc(CAb

- CAC)
KCe(cAC

(25a)

- CAC)

?C(kd*)ml a(CAC

- cA3)

- CAe)
(25b)

and where CAbis the concentration of A in the bubble, CASis the proper concentration measure of A a t the surface of the solid, and surface of solid a = volume of solid
KcdCAc

- CAe)

Ye(kd*)m/a(CAe

- CAS)

(254

6 _
psdp

(20)

where K d is an over-all rate coefficient which accounts for all these phenomena. Eliminating intermediate concentrations CAcand CAefrom the above three equations gives the general expression for the mass transfer coefficient

Second, we may define a mass transfer coefficient Kd[sec.-] based on bubble volume by

These definitions are related by


Kd

T
Kbe YoBd

volume of bubbles

(kd)omr-all particles where

-+Kce

1
YeBd

or

Finally in terms of the actual concentration of A bathing the particles we have the local coefficient k d * (cm./sec.) defined by

This expression links the observed transfer coefficient in a fluidized bed as given by (Sh),,er-all with the true or local coVOL. 7
NO. 4

OCTOBER 1 9 6 8

483

efficients and the flow pattern of gas through the bed. This expression can be simplified as follows. SPECIAL CASE1. \\'hen large bubbles rise in beds of fine particles absorption is rapid compared with gas interchange, and all gaseous A entering the cloud will be completely absorbed. More precisely this condition is expressed by

Preliminary calculations indicate that this simplification may be used with little error in normal gas fluidized beds where
d,

< 0.1 cm. and db > 0.5 cm.

YcBd and YeBd

> > Kbc and Kce

For a given bed of solids and constant bubble size the illustrative calculation of Appendix A shoivs that Equation 27 reduces to an expression of the form

I n this case Equation 26 reduces to

where the constant A increases with particle size and B is a small number. Thus for a series of runs \vith different sizes of the same type of solid this model predicts the relationship ofvariables somewhat as shown in Figure 2. SPECIAL CASE2. In shallow beds of coarse particles-in other words for small bubbles and slow absorption compared to gas interchange-gaseous A will be able to move into the emulsion phase without significant absorption and the concentration of A will be the same every\there. This condition is expressed by

log (Re)

In this case Equation 26 reduces to (Sh)over-aii E (Sh*),l

(30)

Figure 2. Shape of curves predicted from Equation 2 7 for a given type of solid
Individual lines have a slight deviation steeper if 0, flatter if from 45':

>

B<O

Experimental Verification. Figure 3 presents the gassolid mass transfer data reported in the literature. Some of these studies were made in beds of coarse solids and at high Reynolds numbers (Chu e t al., 1950; Riccetti and Thodos, 1961; Rowe and Claxton, 1965) ; another study by Richardson and Szekely (1 961) used carefully controlled conditions-e.g., shallow beds up to 5 particles deep-so as to avoid the bubbling phenomenon. Although some of these conditions may satisfy

Figure 3. Comparison of calculated curves A fluidized beds

I with the reported data on gas-solid mass transfer in

Far calculated lines see Appendix A

484

I&EC PROCESS DESIGN A N D DEVELOPMENT

the restrictions leading to Equation 30, none of these studies is suitable for checking the predictions of the bubbling bed model. TWO of the reported studies were made in normal bubbling beds of fine particles (Kettenring e t al., 1950; Resnick and White, 1949); their reported data are shown in detail in Figure 3, and are used to test the predictions of the bubbling bed model. First we notice that these data seem to lie in the predicted pattern of Figure 2, indicating that the simplified expressions of Equation 27 or 28 can probably be used. T o fit the data with Equation 27 two quantities must be specified: first, the fraction solids in the bubbles, y b , and second, the bubble size in that part of the bed which is of interest and where most of the transfer occurs. Since neither quantity is reported in the original studies, let us select for our calculations: For Resnick and White:
y b = 0.005, db = 0.35 cm., for d, = 0.028, 0.040, 0.047 cm.

IO

Nu
0. I

0.0 I
I I
I

db = 0.41 db = 0.55

0.80 cm., for u, = 30 140 cm./ sec. and d p = 0.074 cm.


0.85 cm., foru, = 40 140cm./sec. and d, = 0.106 cm.

O.OO'

0 Richardson and Ayers(1959) V Donnadleu (1961)


0 Heertjes and McKibblns0956 0 Walton et al(1952) A Kettenrlng et ol(I9501
I

y 1 1

'

' ' I I l ' I

I I I I / l I

I O

""U

I00

For Kettenring e t ai. :


y b = 0.001,
db

I( 3 0

Re
Figure 4. Correlation of data for gas-solid heat transfer, for Re 100

0.8, 1.2, 1.6, 2.0 cm. for d, = 0.036, 0.050, 0.072, 0.100 cm.

<

The values for yo were chosen in the range reported in Equation 16 and its influence is secondary. The bubble sizes were adjusted to fit the theory to the data. The results of calculations based on Equation 27 are shown as lines A E for the data of Resnick and White, and as lines F Z for the data of Kettenring et ai. A sample calculation is shown in Appendix A. The theory fits the data. Discussion and Conclusions. Equations are derived to account for the low observed gas-solid mass transfer coefficient in terms of the local values of the transfer coefficient and the flow pattern of gas through the bed as predicted by the bubbling bed model. I t was possible to obtain a good fit of the data by selecting y b in the range of values observed, and then adjusting the parameter of the model, the effective bubble size, db. From physical considerations it turns out that the small effective bubble sizes which had to be used were not unreasonable, because the region where most mass transfer occurs is close to the distributor. Here bubble sizes can be expected to be small, sometimes dependent on gas flow rate, but certainly influenced by the type of distributor used. Also, one experimental study, that of Kettenring e t ai. (1950), reports on both heat and mass transfer in the same bed (see below for the heat transfer analysis), and the fact that the same effective bubble size is used to fit these two types of data is additional evidence that the effective bubble size obtained from the bubbling bed model has physical significance. For predictive purposes the weakness of this approach is that values of y b and db are not known and must be estimated. I t is hoped that future studies will compare and relate these values with the corresponding experimental quantities. We would expect some simple relationship between these sizes to hold.

From Kothari ( 1 967)

- -

cool bed (Frantz, 1958; Heertjes et ai., 1953; Heertjes and McKibbins, 1956; Kettenring e t al., 1950; Walton et al., 1952) or an unsteady-state technique-e.g., following the temperature rise of a cold bed fluidized by hot gas (Donnadieu, 1961; Ferron, 1958; Fritz, 1956; Wamsley and Johanson, 1954). All the steady-state studies assumed plug flow of gas through the bed and reported the heat transfer coefficients on this basis. Most of the unsteady-state studies took the gas to be completely and uniformly mixed in the bed, thus in backmix flow, and reported the coefficients on this basis, although the comprehensive unsteady-state study of Donnadieu (1961) assumed plug flow. The value of the reported coefficient depends on the type of flow assumed, and the coefficients calculated on plug flow or backmix flow bases should differ. O n a plot of Nusselt number us. Reynolds number the reported results of the unsteady- and the steady-state experiments which assumed plug flow of gas were found by Kothari (1967) to fall in a narrow band which was correlated by

Nu = 0.3(Re)1.3 R e

< 100

(31)

Gas-Solid Heat Transfer

Gas-solid heat transfer in fluidized beds has been studied using either a steady-state experimental technique-e.g., measurement of the temperature profile of hot gas entering a

This is shown in Figure 4, and is the flow model used in all the mass transfer studies. O n this same plot the results reported in terms of backmix flow are scattered one from the other, suggesting that this flow model is not a good one to represent this phenomenon. Consequently these reported coefficients will not be considered further. Additional analyses and discussions of these data are reported by Kothari (1967). When the Reynolds number is lowered, the observed h-usselt number falls continuously to as low as instead of leveling off at 2. As in mass transfer, the sharp drop in observed Nusselt number to less than one thousandth that expected may be attributed to the bypassing and poor contacting in fluidized beds. The purpose of this part of the paper is to explain these low observed coefficients, calculated on the
VOL. 7

NO. 4

OCTOBER 1 9 6 8

485

assumption of plug flow of gas in the bed, in terms of the true or local heat transfer coefficient with the proper flow pattern of gas in the bed as given by the bubbling bed model. Definition of Heat Transfer Coefficient. A number of heat transfer coefficients may be defined, depending on the temperatures used. First, when the temperature of gas bubbles in the bed or the temperature of leaving gas, which consists primarily of gas bubbles, is used in the driving force term, we have what may be called an over-all driving force

IO

I I I I I

Heertjes and

0.36 0.5 I 0.73


1.02

a
0
0

Nuapparent
I-

with corresponding coefficient hover-all and dimensionless group Nuover-alI. When the temperatures of bubble gas Tob and emulsion gas Toe a t a given level in the bed are averaged to give Toand used in the driving force term, we then have what may be called an apparent driving force
ATappwent

dp[mml

0.36 0.50 0.72


1.00
0.1,
I

0
C,

0 0
1 1 1

io0

I I I I l

To - T s

(33) (34)

IO

where

Re

To

6TOb f (1

6)TOe

and with corresponding coefficient happarentand dimensionless group Nuapparent. Since the temperatures of gas and solid in the emulsion phase are close to each other, we can take T,, = T,; hence combining the above three equations gives A T a p p a r e n t = 6A T o v e r - a i l
6happarent

Figure 5. Comparison of the calculated curves A' H' with reported data on gas-solid heat transfer in fluidized beds
For calculated lines see Appendix B

hover-all

6NUapparent = N~over-,11

I n the mass transfer studies reported above, all concentration measurements were made at the bed exit; hence over-all coefficients were used throughout. I n all the steady-state heat transfer experiments gas temperatures were measured in the bed ; hence apparent coefficients were used. Finally we have the local or true heat transfer coefficient, h*, and its corresponding dimensionless group, Nu*, defined in terms of the temperature of gas actually bathing the individual particles. Evaluation of Heat Transfer Coefficient. The treatment for heat transfer closely parallels that for mass transfer. Consider hot gases entering and cooling as they pass through a cold fluidized bed. Then assuming plug flow of gas through the bed we have for any section of bed (heat ;ngFtion or, in symbols, CpoGodTo =
happarent

(35)

This expression is a parallel but simplified version of Equation 25, which assumes that the approach to equilibrium is rapid enough in the cloud so that additional heat transfer by gas from cloud to emulsion can be ignored. This assumption is reasonable for normal fluidized bed ; however, where it does not hold the complete equation analogous to Equation 26 must be developed. The term Hoc (cal./cc..sec..'c.) plays the same role in heat transfer as does Kbc in mass transfer; this analogy provides a means for its evaluation, and its value is given by Equation 11. Matching Equation 36 with 37, using Equation 35, and replacing U b by u o / 6 (a simplification of Equation 2, reasonable for high u o / u m j ) then gives

(38) where Hbc is given by Equation 11 and (Nu*) is the true or local Nusselt group for particles immersed in bubbles, far from interfering particles, in free fall conditions, hence approximated by the terminal velocity of the solids. This is the required relationship between the observed coefficient based on plug flow of gas and the local coefficient with the pattern of flow of the bubbling bed model. I t is analogous to Equation 27 for mass transfer. As shown in Appendix B, for a given bed of solids, it is of the form Nuapparent = A'@)

)-(

heat out of section by gas

(heat transferred to solids

a(To

Ts)dl

(36)

Now, if the gas flow rate is high enough, gas enters the bed only as hot bubbles. Then an accounting of the heat loss by bubble gas in terms of the flow pattern of the bubbling bed model and the local or true heat transfer coefficient gives (heat lost by gas) in bubble
=

+ B'

(39)

heat taken up by solids in bubble

)+

heat transferred to cloud where it is all absorbed or, in symbols,

where A' is a constant which increases with particle size and where B' is a very small constant. Its behavior is similar to that shown in Figure 2 for mass transfer. Experimental Verification. As illustration, to show that Equation 38 can fit the reported data, select the data of Kettenring e t al. (1950) and of Heertjes and McKibbins (1956), as shown in Figure 5. The latter study is of particular interest, since these investigators made simultaneous heat and mass transfer measurements in their equipment. Since the bubble sizes and solid content of bubbles are not reported in these studies let us choose:

486

I & E C PROCESS D E S I G N A N D DEVELOPMENT

For Heertjes and McKibbins:


Yb =

0.001 and do = 0.50, 0.55, 0.68, 0.85 Cm. for d p = 0.036, 0.051, 0.073, 0.102 cm.

For Kettenring et al. :


)b =

where Vs is the volume of catalyst pellets which are considered as nonporous. If diffusional effects intrude, we include an effectiveness factor with the rate constant. Integration of this expression for a packed bed, assuming isothermal plug flow and constant density conditions, gives for the extent of reaction 1
tiA - xA= =

0.001and

db =

0.8, 1.2, 1.6, 2.0 Cm. for d p = 0.036, 0.050, 0.072, 0.100 cm.

e-(l-dKmt

(41) (42)

CAt

Again, as with the mass transfer studies, for the results of Heertjes and McKibbins Yb was chosen in the range of reported values of Equation 16, while the bubble size was selected to fit theory to the data. For Kettenring e t al., however, the values for Yb and db were not chosen arbitrarily but were identical to those used for the mass transfer study above. Replacing all quantities in Equation 38 gives the theoretical lines A N D in Figure 5 for the conditions of Heertjes and McKibbins, and lines E N H for the conditions of Kettenring et al. Appendix B shows how one of these curves is obtained. The fit of theory to the data is satisfactory. Discussion and Conclusions. The discussions and conclusions on mass transfer are directly applicable here. The fact that the bubbling model can fit the reported heat transfer data in addition to the reported mass transfer data strengthens the evidence that this model is a useful and simple representation of a normal bubbling fluidized bed.
Solid-Catalyzed Gas-Phase Reactions

e-Km~m

= ,-Xm

where the space time for the fixed bed is volume of fixed bed (43) where the reaction rate constant for the fixed bed is

Km

(1

- e,)K, - em)K,rm

(44)

and the dimensionless rate group for the fixed bed is


X m = Kmrm = (1

For comparison, for backmix flow of gas, the corresponding expression is

Kumerous studies have tried to explain the observed conversion of solid-catalyzed gas-phase reactions in fluidized beds in terms of the reaction kinetics as measured in packed beds or for isolated particles. I n all these efforts the main difficulty centers about obtaining a satisfactory flow model to represent the flow pattern of gas in fluidized beds. The earliest approach to this problem tried to fit the observed behavior with the tanks-in-series or dispersion models; however, this approach was soon abandoned because these models could fit only conversions between plug and backmix flow, while fluidized beds often exhibited conversions well below the backmix extreme. The most widely used approach has been to devise what is considered to be a reasonable two-region model-one region for the bubble phase, the other for the emulsion-and then to adjust the parameters of this model to fit the observed conversion in fluidized beds. At least a dozen different flow models, having from one to six parameters, have been devised (Lanneau, 1960; Lewis et al., 1959; Mathis and Watson, 1956; May, 1959; Muchi, 1965; Orcutt et al., 1958; Partridge and Rowe, 1966; Shen and Johnstone, 1955; Toor and Calderbank, 1967 ; Van Deemter, 1967) ; however, in every case the fit has been limited to the data of that particular study. Until a comparison of fits of a number of studies is made, these models remain oflimited utility. The purpose of this paper is to show that the catalytic conversion measured by the experiments of others can be explained and fitted directly in terms of the reported kinetic constants combined with the gas flow of the bubbling bed model. Definition of Reaction Rate. Consider a first-order solidcatalyzed reaction of component A which is present in the gas stream flowing through a fixed bed. With porous catalyst pellets and no internal mass transfer resistance we can then define the reaction rate constant, K,, by

I n our treatment of reaction in fluidized beds it is assumed that X, or K , is known or can be found from kinetic studies with isolated particles, with recycle reactors, with beds of diluted catalyst, or in isothermal packed beds. Such data are normally available. Conversion from Bubbling Bed Model. Consider at first a vigorously bubbling bed of Figure 6 which has a downflow of emulsion gas; hence u,/u,,

>6

11

Since fresh reactant gas enters the bed only as gas bubbles, an accounting for reactant A gives over-all disappearance

) (
=

reaction in bubble

) (
+

transfer to cloud and wake (47a)

Leaving g a s , C m
Gas from leavlng bubble, CAb6

CAb CAc

I t Enterlng gas, CAI


Figure 6. Sketch of flow pattern in a fluidized bed for downflow of emulsion gas
VOL. 7
NO.

4 OCTOBER 1 9 6 8

487

transfer to (cloud and wake)

reaction in (cloud and wake

) (
+

transfer to the emulsion (47b)

SPECIAL CASE1. For very fast reaction YbK, is comparable to Khcand reaction takes place primarily in the bubble and in the cloud, with practically no reactant entering the emulsion. For this extreme Equations 49 and 50 reduce to

When the downflow of emulsion is comparatively small we can take

transfer to emulsion

) (
~

reaction in emulsion

For an irreversible first-order reaction, and with the nomenclature previously introduced, these three expressions become, in symbols,

-dcAb -dt

-ub

dAh = (KT)bCAb
dl

This simplification rarely applies; however, Equation 55 is presented primarily to show the equivalence in equation forms for chemical and physical kinetics. This can be seen by comparing Equation 55 with Equation 27 for mass transfer and Equation 38 for heat transfer. SPECIALCASE 2. For catalytic reactions normally encountered, the rate is slow enough so that reaction with particles dispersed within the rising bubbles is negligible and Yb can be ignored. I n this case Equations 49 and 50 simplify to

X where ( K T ) bis an over-all rate coefficient which accounts for all the processes occurring in the bed. Combining the above expressions and eliminating concentrations then give a dimensionless rate group, X for fluidized beds in terms of the individual rate constant; thus,
I

2 ):(
1

- emf

.;'I..

K, +

yc

+1

K, +- 1 Kce

1-

(56)

Ye

X I = (KT)b ( 4 ) = K T

' k)[
):(
UbT

Yh

+
'+YC+&]

1
1

+Kce

Ye

T o find when it is reasonable t o use this simplification we simply insert the largest expected value of Y b in Equation 49 (from Equation 16 this would be Y b = 0.01) and then see if this changes XI appreciably. I t so happens that this expression for Xf correlates the catalytic conversion data reported below. SPECIAL CASE3. For the extreme where reaction is very slow and bubbles are small

(49) From Equations 2, 3, 4, and 45 the relation to the packed bed rate group is

KT

< < (Kho and Kce)

and Equations 49 and 50 reduce to

KT
where

l-'nzf

(2)

Xm

With Equation 15 this simplifies to give (51)


Xf E

= 0.711 (gdb)'l2

xm

(57)

Integrating Equation 48a for the boundary condition

gives the concentration ofreactant A in the rising bubbles as (53) The contribution of the cloud-wake region to the exit gas stream can be estimated. However, this happens to be minor, changing the fraction unconverted by 10% at most. Considering the approximate nature of this treatment we will ignore this factor. Thus we may consider the exit gas to consist only of bubble gas, in which case the fraction of reactant remaining in the exiting gas stream becomes (54) where XIis given by Equations 49 and 50. The five terms in the large bracket of Equation 49 represent the contributions of the individual steps toward the disappearance of reactant. I n any particular situation these are likely to be of different magnitudes; some terms will drop out and Equation 49 should simplify considerably. Consider the following cases.
488
I & E C PROCESS D E S I G N A N D DEVELOPMENT

This expression shows that a t this extreme the fluidized bed approaches the close-to-plug-flow behavior of packed beds. For upflow of emulsion gas, or uo/um < 6 11, both bubble and emulsion gas leave the bed as the exit stream. The conversion of bubble gas is given by Equations 54 and 49, while the conversion of emulsion gas will have to be found by an extension of the present analysis. If we recognize, however, that the emulsion contribution to the total gas flow is still very small as long as uo/urnf > 3, then the over-all conversion can still be approximated by that of the bubble stream. The reasonableness of this crude approximation is shown by the check with experiment given below. Check with Theory. EXPERIMENTAL VERIFICATION. The predictions of the bubbling bed model are checked with the reported data of various experimental studies in Figures 7 to 9. Figure 7 presents the data of Orcutt et al. (1958) for ozone decomposition; Figure 8 presents the data of Iwasaki et al. (1965) for the decomposition of cumene and the data of Ogasawara et al. (1959) for the synthesis of acetonitrile. All these studies were made at high enough uo/umf (u,/umf = 5 N 88) to have downflow of emulsion gas, and it is for these conditions that the above equations were strictly derived. O n the other hand, Figure 9 presents the recent data of Kobayashi et al. (1966) for ozone decomposition a t somewhat lo), where a slight upflow lower gas velocities (u,/urnf = 3

1.0

1.0 0.6
a4
0.2
c

gal
I
c

0.1

0.06

0.04

4(m=(1 - m ) L m K r / U o

aoi
Figure 7. Conversion vs. dimensionless reaction rate group for the catalytic decomposition of 6 ozone in beds where u,/u,~

Figure 9. Conversion vs. dimensionless reaction rate group for catalytic decomposition of ozone
Experimental points from Kobayashi and Arai ( 1 966) for beds where 3 u o / u r n f 10 dr = 20 cm., 1, = 34 cm., o,f = 2.1 crn./sec.

<

<

>

uo, Cm./Sec.

uo/UmJ

0
0

Data from Orcutt ef al. ( 1 958)

6.6 9.9 12.3, 1 6 . 5 , 2 0

3.14 4.71 6.28'9.55

I,,, Cm. 0 61.0 0 61 .O

u o , Cm./Sec.

0 71.3

3.66 14.1 9.15, 14.7


I I I 1 Calcd. by Eqs. 54and56:

1.0

x' 0.1

I
50

10
flm=

20
(1

30

40

-LmlLmKr/Uo

Figure 8. Conversion vs. dimensionless reaction rate group for catalytic reactions in 6 beds where u,/u,J

>

0
0

Decomposition of cumene (Iwaraki, 1965) 7.6 cm., U,J = 0.18 cm./sec.,u, = 2.5 16 crn./sec. Synthesis of acetonitrile (Ogasawara, 1959), closely similar conditions to above

I,,

of emulsion gas may be expected. A comparison of theory with these experimental results should then indicate how well the emulsion downflow equations derived above may be expected to extrapolate to these slightly upflow conditions. The conversion predictions of the bubbling bed model are included in Figures 7 to 9 for various assumed bubble sizes, and Appendix C shows how one of these curves is obtained. I n these calculations the magnitudes of the various rate terms

Kbo etc., are such that 7 0 could be taken to be zero. Thus Equation 56 with one parameter, do, was used to obtain the theoretical curves. Examining these results we see that the actual performance drops progressively below plug flow, and even below backmix flow, as the conversion rises. Also, at high uo/u,, (Figures 7 and 8) bypassing is more severe and performance is poorer than at low uo/u,, (Figure 9). The theoretical curves of Figures 7,8, and 9 follow this general pattern. Consider the variation in performance with bubble size. Little can be said about Figure 7. The data show considerable scatter, but the general trend is matched by the theoretical curves. I n Figure 8 the high values were obtained by using low uo, thus with small bubbles; small values were obtained with high uo, thus with large bubbles. The theoretical curves cross the data in this manner. Finally, Figure 9 shows three series of runs, each at a fixed uo and L,, but with X , varied by changing K,. Thus each series should correspond to a fixed bubble size ; the higher the velocity the larger the bubble size. As a result these data should provide a rather sensitive test of the model. Comparing results we find that the theoretical curves fit the data extremely well, and naturally, the progression of bubble sizes is in the right direction. Over-all Evaluation and Summary. One of the major problems in fluidization is to develop a flow model for the contact of gas and solid which is accurate enough to represent the behavior of real beds, but at the same time simple enough so that its physical basis is clearly seen and it can easily be used. I n a previous paper a simple model was proposed, and there it was shown that this model was in general accord with the observed flow characteristics of gas and solid in fluidized beds. This paper shows that this flow model can also explain the reported kinetics for various gas-solid contacting phenomena, such as heat and mass transfer and catalytic reactions, while correctly predicting the effect of the variables such as particle size and gas velocity. This is seen by the match of the family of calculated curves with the reported data on the dimensionless plots of Figures 3, 5, 7, 8, and 9.

K,,

VOL. 7

NO. 4

OCTOBER 1 9 6 8

489

I n these kinetic phenomena a gas-phase component (hot gas to be cooled, component to be adsorbed, or reacted) is to be transported to the solid. I t enters the bed in a gas bubble, and if it is to come in contact with the solid:

Bed.

A. I t either comes in contact with the solid dispersed in the bubble ( 7 6 is the significant rate quantity for this process), or B. I t is transported to the cloud surrounding the bubble (Kb, is the significant rate quantity).
I n the cloud :

umf = 1.21 cm./sec., emf = 0.5 7 0 = 0.005 (estimated) db = 0.35 (estimated) SOLUTION For the terminal velocity in free fall
(Re), = (0.028)(1.18 X 10-3)(41) = 7.54 1.8 x 10-4

and from Equation 1 the corresponding Sherwood number is (Sh*), = 2

C. I t either comes in contact with the solid in the cloudwake region ( y cis the significant rate quantity), or D. I t is further transported to the emulsion (KE8 is the significant rate quantity). I n the emulsion:

+ 0.6(2.39)'/a(7.54)'/Z = 4.20
(0.065)'/2(980)1/4 = 46.5 sec.-l (0.3 5)6/4

Then from Equation 10


Kbc

= 4.5

(g)

f 5.85

E. I t comes in contact with the solid ( y e is the significant rate quantity).


I n heat and mass transfer the gaseous component has to be transported only to the outer surface of the particles. This process is very rapid and an analysis of the reported data shows that only steps A and B need to be considered. Thus for all practical purposes all of the gaseous component entering the cloud comes in contact with the solid there. I n catalytic reactions the reactant has to be transported to the surface and then into the pores of the solid. This is inherently a slower process and for reactions normally encountered the reported numerical values show that step A can be ignored while steps B, C, D, and E must be considered. Thus for catalytic reactions we can assume Tb = 0. The predictions of the model are dependent on one parameter, the effective bubble size db, as well as the quantities must be 7 0 , yo y e , KOo and Kce. Of these five quantities measured experimentally; the other four quantities can be calculated directly from the model using an experimental value for a. First estimates for ya and a are available from observations. This paper shows that when the effective bubble size is properly chosen, all the kinetic phenomena reported for fluidized beds can be correlated satisfactorily. There is an urgent need a t present to relate this "effective bubble size" obtained from this model with the size of bubbles actually observed in fluidized beds. With this available it is hoped that this model will be useful for predictive purposes. Finally, many simplifying assumptions were made in developing this model (the Davidson bubble, bed with single size of bubbles, emulsion at minimum fluidizing conditions a t all gas velocities, etc.). Why were these assumptions chosen and not others? Actually various sets of assumptions were examined for simplicity and goodness of fit with the reported data on heat and mass transfer and catalytic reactions and it was found that the set leading to the bubbling bed model was the simplest, physically reasonable, and easy to use to fit the data.
Appendix A

from Equation 3

from Equation 2

and from Equation 4

Replacing the above quantities in Equation 27 then gives (Sh)over.all =


u,
uo

+ 12

1.21

. u,

+ 12 [(0.005)(4.20) +
6.6

(0.9) (0.4) (0.028)* (46.511 6 (0.065)


=

0.00829 U,

- 0.0101

Also, for these solids Re = (0.028)(1.18 X 10-a)uO = 0.184 u, 1.8 x 10-4

Combining the above two expressions gives (Sh)over-all = 0.045 Re

- 0.0101

which is shown as the slightly curved line A in Figure 3.


Appendix

From the bubbling bed model predict how the apparent Nusselt numbers should vary with Reynolds number for the smallest size of solid used by Heertjes and McKibbins.

DATA

From the bubbling bed model predict how the observed Sherwood number should vary with Reynolds number for the smallest size of solid used by Resnick and White (1949).

DATA
Solids. ps = 1.06 g./cc., cps = 0.40, d p = 0.028 cm., terminal velocityin free fall, u t = 41 cm./sec. (calculated) g./cc., p = 1.8 X g./cm.-sec., Gas. pe = 1.18 X = 0.065sq.cm./secS,y = 0.9, Sc = 2.39 De =
490

= 0.036 cm., terminal Solid, ps = 1.3 g./cc., cps = 0.806, velocity in free fall, u 1 = 86 cm./sec. g./cc., p = 1.8 x 10-4g./cm. sec. p g = 1.18 X Gas. k, = 6.25 X 10-6cal./cm..sec. C. C ,, = 0.24 cal./g..' C. Pr = 0.69. Bed. u,, = 10 cm./sec., emf = 0.5 = 0.001 (estimated) db = 0.5 cm. (estimated) SOLUTION For the terminal velocity in free fall

zp

I&EC PROCESS D E S I G N AND DEVELOPMENT

and from Equation 1 the corresponding Nusselt number is

from Equation 2
6 =

(Nu*)$= 2

+ 0.6(0.69)1~3(20.3)1~z = 4.39

Then from Equation 11


Hbc =

13.2 - 2.1 = 0.206 53.9

4.5( 10) (0.24) (1.1 8 X

0.50

+
C.

From Equation 12

K,, = 6.78

5.85 [(6.25 X 10-5)(0.24)(1.18 X 10-3)]1/2(980)1/4 (0. 50y4


=

0.5 X 0.204 X 53.9 3.78

from Equation 17
7 , = (1

3.60 X 10+ cal./cc.. sec..

- 0.5)

[0.711 (3.7 3X X 980)'/2 2.1/0.5- 2.1/0.5

+ 0.471
=

from Equation 3 from Equation 15 from Equation 2


6 = - - -u,
Ye =

0.40

(1

- 0.5)(1 - 0.206)
0.206

- 0.40 = 1.53

ub

10

24,

uo

+ 5.8

- 10

from Equation 45

and from Equation 4

K, =

uox,

(1

- c,)L,

1.32 X, (1

- 0.45)(34)

0.706 K ,

Replacing the above quantities in Equation 38 then gives


u,
(W,pparent

Replacing all these above quantities in Equations 54 and 56 gives for the fluidized bed
1

8.70

+ 5.8 [(0.001)(4.39) +

- XA
1

= f'-e

where

XI =
or in dimensionless form
= 0.070

r0.706 x ,

+ 0.051 R e
'This is the required relationship between the reaction rate group and the conversion, and a curve representing it is shown in Figure 9.
Nomenclature

which is shown as the slightly curved line A' in Figure 5.


Appendix C

From the pattern of gas flow of the bubbling bed model show how the conversion should change with the dimensionless reaction rate group for the catalytic decomposition of ozone in a fluidized bed for conditions comparable to those reported by Kobayashi et al. (1966) and shown in Figure 9.

DATA
Gas. D De= 0.204 sq. cm./sec. Bed. d t = 20 cm., L, = 34 cm., c, = 0.45, emf = 0.50 umy = 2.1 cm./sec. U, = (6.6 9.9 13.2 20)/5 = 13.2 cm./sec. db = 3.7 cm. (estimated) y b = 0 (any value between 0 and 0.01 will not change the outlet concentration by more than 1%. This can be seen by the following calculations) a = 0.47; estimated from Rowe and Partridge (1965).

- ,)a' = specific surface area of fluidized bed, sq. cm./cc. a' = 6/(p,dP) = specific surface area of solids, sq. cm./ cc . CAb, C A ~ C, A ~ C,A , = concentrations of A in bubble, in cloudwake region, in emulsion; and the proper concentration of A at the surface of solids, respectively, g.moles/cc. = Cab at top of fluidized bed, g.-moles/cc. CAbo = concentration of A in inlet gas stream, g.-moles/ CAi cc. = specific heat of gas at constant pressure, cal./g.. CP,
=

(1

c.

P, D e

= diffusion coefficient and effective diffusion co-

efficient in emulsion, respectively, sq. cm./sec.

SOLUTION
Determine 1 - X A us. K, from Equations 54 and 56From the given data we find from Equation 10

db
dP
G O

= effective bubble diameter, cm. = diameter of particles, cm. = pouo = superficial mass velocity of flowing gas,

g./sq. cm. .set.

Hac
h (h*)t

= acceleration of gravity, cm./sec.' = volumetric heat transfer coefficient, cal./cc.. sec..


O

= heat transfer coefficient, cal./sq. cm.. sec.. O C. = heat transfer coefficient between gas and solids

c.

from Equation 3
ub, = 0.711 (980 X 3.7)1/2 = 42.8 cm./sec. up = 13.2

dispersed in bubble, cal./sq. cm. .set.. O C. Kbc, K,,, Kbs = interchange coefficients between bubble and

- 2.1 + 42.8

53.9 cm./sec.

cloud-wake region, between cloud-wake region and emulsion, and between bubble and emulsion, respectively, 1/sec.
VOL. 7

NO. 4

OCTOBER 1 9 6 8

491

mass transfer coefficient, 1/sec. reaction rate constant, l/sec. mass transfer coefficient, cm./sec. mass transfer coefficient between solids and bathing gas stream, cm./sec. thermal conductivity of gas, cal./cm. sec. O C. height of fluidized bed and static bed, respectively, cm. distance from distributor, cm. g.-moles A in a bubble hd,/k, = Nusselt number Cpop/kg= Prandtl number reaction rate, g.-moles/sec. d p p o u o / p = Reynolds number kdd,y/a) = Sherwood number u / p & = Schmidt number average temperature of gas at any level in bed,
O

literature Cited

c.

temperature of bubble gas, of emulsion gas, and of solids, respectively, C. t = time, sec. u b , &, umf, uo = velocity of rising bubble, velocity of rising bubble with respect to the emulsion solids, minimum fluidizing velocity, superficial gas velocity (based on an empty tube), respectively, cm./sec. = downward velocity of emulsion solids, cm./sec. Ur vb, V,, Vw = volume of bubble, of cloud, and of wake, respectively, cc. V,, V, = volume of fixed bed and of catalyst pellets, respectively, cc. = fractional conversion of gaseous reactant A X A Y = logarithmic mean fraction of inert or nondiffusing component
=

Tab, Toe, T,

GREEK SYMBOLS
= volume of emulsion transported upward behind

a bubble/volume of a bubble
tion 14

vw/vb

= volumetric fractions of solids, defined by Equa= volume fraction of bed consisting of bubbles

= void fraction of bed at normal fluidizing condition, at static condition, and at minimum fluidization, respectively = dimensionless reaction rate group for a fluidized bed, given by Equation 49 = (1 - e,)&, = (1 - em)K,L,/uo, dimensionless reaction rate group for fixed beds = viscosity of gas, g./cm. .set. = densities of gas and solid, respectively, g./cc. = L,/uo = sphericity

SUBSCRIPTS
( (
)mf
It

Chu, J. C., Kalil, J., Wetteroth, W. A,, Chem. Eng. Progr. 46, 139 (1950). Donnadieu, G., Rev. Znst. Franc. Petrole 16, 1330 (1961). Ferron, J. R., Gas-Solid Heat Transfer and Gas-Mixing in Fluidized Beds, Ph. D. thesis, University of Wisconsin, 1958. Frantz, J. F., Evaporation of Brine Solution in Fluidized Salt Beds, Ph. D. thesis, Louisiana State University, 1958. Fritz, J. C., Heat Transfer between Air and Solid Particles in Sixteen-Inch Diameter Fluidized Beds, Ph. D. thesis, University of Wisconsin, 1956. Heertjes, P. M., de Boer, H. G. J., de Haas van Dorsser, A. H., Chem. Eng. Sci. 2, 97 (1953). Heertjes, P. M., McKibbins, S. W., Chem. Eng. Sci. 5,161 (1956). Iwasaki, M., Kooya, I., Sueyoshi, H., Shirasaki, T., Echigoya, E., Kagaku-Kogaku (Chem. Eng., Japan) 29,892 (1965). Kettenring, K. N., Manderfield, E. L., Smith, J. M., Chem. Eng. Progr. 46, 139 (1950). Kobayashi, H., Arai, F., collaborators, Kagaku-Kogaku (Chem. Eng., Japan) 30, 656 (1966); preprint for Annual Symposium on Chemical Reaction Engineering, Nagoya, November 1966. Kothari, A. K., Analysis of Fluid-Solid Heat Transfer Coefficients in Fluidized Beds, M. S. thesis, Illinois Institute of Technology, Chicago, 1967. Kunii, D., Levenspiel, O., Fluidization Engineering, Wiley, New York, 1968a. Kunii, D., Levenspiel, O., Znd. Eng. Chem. Fundamentals, 7, 446 (1968b). Kunii, D., Suzuki, M., Intern. J . Heat Mass Transfer 10, 845 (1967). Lanneau, K. P., Trans. Znst. Chem. Engrs. 38, 125 (1960). Lewis, W. K., Gilliland, E. R., Glass, W., A.Z.Ch.E. J . 5 , 419 (1959). Mathis, J. F., Watson, C. C., A.Z.Ch.E. J . 2, 518 (1956). May, W.G., Chem. Eng. Progr. 5 5 , 49 (December 1959). Muchi, I., Mem. Fac. Eng. Nagoya Uniu. 17, No. 1, 79 (1965). Ogasawara, S., Shirai, T., Morikawa, K., Saikin-no Hanno Kogaku (Recent Reaction Eng.), No. 1, p. 102, Maki Pub. Co., Japan, 1959. Orcutt, J. C., Davidson, J. F., Pigford, R. L., Chem. Eng. Progr. Symp. Ser. 58 (38), 1 (1958). Partridge, B. A., Rowe, P. N., Trans. Znst. Chem. Engrs. 44, T335 (1966). Resnick, W. E., White, R. R., Chem. Eng. Progr. 45, 377 (1949). Richardson, J. F., Ayers, P., Trans. Znst. Chem. Engrs. 37, 314 (1959). Richardson, J. F., Szekely, J., Trans. Znst. Chem. Engrs. 39, 212 (1961). Riccetti, R. E., Thodos, G., A.Z.Ch.E. J . 7,442 (1961). Rowe, P. N., Intern. J . Heat Mass Transfer 6, 989 (1963). Rowe, P. N., Claxton, K. T., Trans. Znst. Chem. Engrs. 43, T321 (1965). Rowe, P. N., Partridge, B. A., Trans. Znst. Chem. Engrs. 43, T157 (1965). Shen, C. Y . ,Johnstone, H. F., A.Z.Ch.E. J . 1, 349 (1955). Toor, F. D., Calderbank, P. H., Proceedings of International Symposium on Fluidization, Netherlands University Press, Amsterdam, 1967. Van Deemter, J. J., Chem. Eng. Sci. 16, 105 (1961); Proceedings of International Symposium on Fluidization, Netherlands University Press, Amsterdam, 1967. Walton, J. S., Olson, R. L., Levenspiel, O., Znd. Eng. Chem. 44, 1474 (1952). Wamsley, W. W., Johanson, L. N., Chem. Eng. Progr. 50,347 (1954). Zabrodsky, S. S., Intern. J . Heat Mass Transfer 6,23, 991 (1963). RECEIVED for review November 13, 1967 ACCEPTED April 17, 1968

= corresponding to minimum fluidization = corresponding to free falling condition

492

I & E C PROCESS D E S I G N A N D D E V E L O P M E N T

You might also like