Difficulties in Scaling Up The Fluidized Bed Reactor

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Difficulties in Trying To Model and Scale-Up the Bubbling Fluidized Bed (BFB)

Reactor
The purpose of running experiments in the laboratory is to
obtain information to help the engineer in his design of large-
scale operating units. However, the scaleup and design of
bubbling fluidized bed (BFB) reactors is uncertain, for several
reasons. Primarily, it is because the size of bubbles in the larger
unit strongly influences the conversion of reactant, and this
factor cannot be reliably estimated from small-scale laboratory
experiments. We discuss this problem.
During the second world war, the United States had an urgent
need to produce enormous quantities of aviation gasoline. Fine
particle fluidization (the fluidized catalytic cracking (FCC)
process) was proposed and chosen; commercial plants were built
chop-chop and were wildly successful. Interestingly, all were
built without any basic information on flow behavior, heat and
mass transfer, reaction behavior, and other contacting phenom-
ena in these units. After the war, news of this new gas-solid
(G/S) contacting method was made public, and this spurred the
research community to explore its properties and its unique
features.
The first phenomenon studied was the very simplest: the heat-
transfer rate between particles and the fluidizing gas. Until 1965,
there were at least 14 such studies. These came from all over
the worldsfrom countries such as the United States, Europe,
Russia, and Japan. These researchers measured the temperatures
of gases entering and leaving a reactor and, assuming plug flow
or mixed flow of gas in the bed, calculated values for the kinetic Figure 1. Reported result for nine experimental studies of gas-solid (G/
factor (h). Figure 1 shows some of their results on a plot of S) heat transfer in fluidized beds (see Kunii and Levenspiel1).
Nusselt number versus Reynolds number (Nu vs Re). This figure same mistake, and that all their results were wrong. What is
shows the values for h for the heating of a single particle falling the explanation for this? Let us see.
through a gas and also for particles mixing in bubbling fluidized Now, for any process, there are four inter-related quantities:
bed (BFB) reactors. The experimental values in BFB reactors (1) the inlet gas to the BFB reactor; (2) the outlet gas to the
all lie well below the single particle correlationssome by a BFB reactor; (3) the kinetics, including the heat- and mass-
factor of 1000 or more. transfer coefficients between the heating particle and the bed
It is hard to understand why none of the 14 groups of solids; and (4) the flow pattern of solids in the vessel. If you
experimenters questioned this low result. Would the researcher measure quantities 1 and 2, then 3 and 4 are related. Therefore,
not suspect that, in an environment that contains many buffeting if you make an error in quantity 3, then quantity 4 is also wrong.
and colliding particles and much turbulence, the h value between Here, in these experiments, because the kinetic factor (h) is way
a particle and a gas would be greater than for a single falling off, this means that the flow model chosen (plug or mixed flow)
particle? Why was it actually lower? Why wasn’t the researcher is also wrong. But what is it? That is our problem.
suspicious? What kind of a researcher was he?
Any of the researchers could have conducted a theoretical Chemical Reactor: Plug Flow or Mixed Flow Model?
experiment to determine whether the reported results were Numerous investigators have studied the behavior of solid
reasonable and made sense. Let us do that now. catalyzed BFB reactors. They measured the fluid compositions
Let us introduce a single particle into a fluidized bed at 0 °C entering and leaving the reactor, and they knew the reaction
and ask how long it takes for the particle to heat up to 90 °C. rate constant (the kinetics), so they could check the fit of this
(See Figure 2.) Eight minutes or a quarter of a second: which flow model or that flow model.
sounds more reasonable? Certainly not the 8 min. This example Thirty-nine flow experiments were conducted by five different
suggests that all nine groups of experimenters had made the reactor groups in various flow reactors, and the extent of the
10.1021/ie071355v CCC: $40.75 © xxxx American Chemical Society
Published on Web 12/22/2007 PAGE EST: 4.7
B

Figure 4. Schematic of a general two-region six-parameter flow model.

Figure 2. Depiction of the time needed to heat a single, very tiny, particle
in a fluidized bed; plots from theory and from experiment are shown.

Figure 5. Four special cases of the two-region model: (a) one-parameter


model, (b) two-parameter model, and (c and d) three-paramater models.

Table 1. Number of Parameters Used in the Modelsa


Figure 3. Results of 39 experimental studies of reactor behavior of
model research group
beds of small (Geldart A/B and B) particles, and of very small
(Geldart A) particles (data obtained from Johnstone et al.2 and Kunii and 1 parameter Shen
Levenspiel3). 1 parameter Gomezplata
1 parameter Orcutt
1 parameter Lanneau
conversion of reactant was measured in the experiments. The 1 parameter Massimilla
results of these experiments are shown in Figure 3, and they
2 parameters Mathis
clearly indicate that neither the plug flow model nor the mixed 2 parameters May
flow model even closely matches the actual experimental 2 parameters Lewis
findings. In fact, the behavior is much worse than either model. 3 parameters van Deemter
This behavior is consistent with the finding in heat transfer,
4 parameters Kobayashi
in which the heat-transfer rate is much lower than expected from 4 parameters Muchi
either of these models. 4 parameters Mamuro
Therefore, to conclude, the simple $10 models of plug flow 6 parameters Exxon researchers
and mixed flow do not reasonably represent the flow of gas in a Data obtained from Grace.5
fluidized beds and can be inaccurate by a factor of as much as
10 000! bubbles are free of solids, Figure 5d shows no interchange
between regions, and so on.
Arbitrary Two-Region Flow Models With so many alternatives available, we are left with an entire
basketful of flow modelssat least 34 by 1970sthat extrapolate
Johnstone and his research group2 said, “Let us look at the differently. For example, of the three-parameter models,
BFB to see if there are any clues to their surprisingly bad Shell in Amsterdam chose the model depicted in Figure 5d,
behavior”. They noticed that the bed consisted of a dense region whereas Shell in Baton Rouge, LA, chose the model depicted
through which rose voids, which he called bubbles. in Figure 5c. These two models are definitely contradictory;
Johnstone’s group was the first to try to account for this, and for example, one states that the emulsion gas flows upward,
this led to a new class of model: the two-region six-parameter and the other proclaims that the flow is downward! Did this
model shown in Figure 4. bother them? Not at all, because by properly choosing the values
Many special cases were proposed, and Figure 5 illustrates of the parameters, both these models can be made to fit the
four of these. Figure 5a shows a model with one adjustable data.
parameter (x). Figure 5b shows a two-parameter model (x and Other groups developed even more-complicated models. For
m1), and Figures 5c and 5d show two different three-parameter example, Exxon even said, “We have the best mathematicians
models (with variables of x, m1, and V1 and D1, D2, and Vrec, and the biggest computers in the business so we’ll use the $1000
respectively). Note that Figures 5a and 5d assume that the six parameter model”. The constants varied for their different
C

Figure 6. Fitting the five data points with a first-, second-, and third-
degree polynomial.

Figure 8. Sketches showing the two extremes in bubble behavior in


fluidized beds: (a) fine particle bed and (b) coarse particle bed.

Figure 7. Extrapolation of the models depicted in Figure 6.

reactor vessels, and for their different feeds, regardless of


whether they were from Venezuela, or Nigeria, etc. Some of
their fitted volumes were negative, but that did not matter,
because, through a proper choice of constants, they could obtain Figure 9. Rising bubbles can be modeled in various ways.
wonderful fits. Some samples of these models are presented in
Table 1. use a bit of science to ANALYSE (sic) the behavior of the
Yes, with the proper choice of constants, these 34 models bubbles”. He then invented the very simplest of models, based
were able to fit everything ... but they predicted nothing. on fluid mechanics. He assumed the presence of spherical
bubbles that were free of solids in a bed whose dense region
What is an Engineering Model? moved out of the way of the rising bubbles as an inviscid fluid,
There are dozens of models that are all contradictory, but all and where the gas flowed through the dense region as an
are able to fit the data. incompressible viscous fluid.
Does not that bother you? It bothers me, and leads me to At first thought, one may think this to be ridiculously crude,
ask, “What are the requirements of an engineering model?” I not even a $10 model. However, the consequences of this picture
would like to present the following three criteria: were magical and, when tested against the experiment, were
(1) The model must fit the data. spectacular.
(2) The model must be extrapolatable, theoretically and/or I cannot go into this development here, except to say that (a)
(preferably) experimentally. for fine particle beds, bubbles rise as vortex rings, each
(3) The model parameters must be reasonable, from the surrounded by a thin cloud (or “halo”) of circulating gas (the
physical point of view, and not be arbitrarily chosen to fit each bubble gas in these fine particle beds remains segregated from
specific situation. the remainder of the gas in the bed), and (b) for coarse particle
Let me illustrate what I mean with a simple example. Suppose beds, the behavior is quite different, because the rising gas uses
I give you the five data points shown in Figure 6. First-, second-, the bubbles as a shortcut. Figure 8 shows sketches of these two
and third-order polynomials would all satisfactorily fit the data. extremes.
However, to determine which model more closely represents With this breakthrough, researchers stampeded to this concept
reality, extrapolate the operating conditions and then see which and developed all types of variations of these models. Figure 9
model continues to fit the data. Figure 7 shows that, at larger x shows the five distinct choices available: (i) spherical or
values, the y value of the cubic model zooms up to infinity, the nonspherical bubbles; (ii) bubbles followed by accompanying
linear model predicts a modest rise in y value, and the quadratic wakes or without wakes; (iii) bubbles that are constant in size,
model predicts a decrease in the value of y. This would be a growing in size, or have a size distribution which changes with
discriminating test for these models. height; (iv) unchanging bubbles, or coalescing and splitting
Unfortunately, no one has tried widely different experimental bubbles; or (v) assume that the bed consists of three regions
conditions to test fluidization models, so we are left with an (the bubble, the cloud, and the emulsion) or else combine the
entire basketful of models, all making different predictions, all cloud with the emulsion or with the bubble to give two-region
claiming to represent reality. models.
Figure 10 shows the various combinations chosen. In some
The Davidson Genius
of these combinations, the bubble and the cloud are considered
At this point, Davidson, with Harrison,6 said, “Let us not just to be one region; in other combinations, the cloud and the
LOOK at the fluidized bed and note the bubbles, let us try to emulsion are combined. And then there is the complete three-
D

Figure 13. Sketch of an example depicting theoretical estimation of how


the conversion changes with bubble size.

Figure 10. Sketches showing the variations chosen: (a) two-region


simplification of the Davidson model and (b) the complete Davidson model
with its three regions.

Figure 14. Sketch showing the decrease in conversion, using the three-
region model, with increasing bubble size.
Figure 11. Depiction of the three-region flow model for fine particle
bubbling fluidized beds (BFBs).

Figure 12. Sketch showing the chemical conversion for the three-region
model. Five factors are involved.

region model. However, which of these models should we use?


Bubbles certainly are not spherical; they split apart, they
coalesce, and they certainly are not uniform in size. But it does
not make sense to incorporate into a model complications that Figure 15. Sketch showing the decrease in conversion, using a three-region
cannot be measured. No, at this point, we should heed Einstein’s model where intermediates are the desired products, with increasing bubble
admonition on model building: size.

“Keep things as simple as possible, but not simpler” But does this represent an engineering model? The definitive
answer to this question requires that experiments be performed
With this in mind, we choose the three-region flow model of using different bubble sizes and how the conversion changes
Kunii and Levenspiel,1,4 which contains just one size of bubble with bubble size be monitored. No one has done this yet;
(shown in Figure 11). Only one parameter of this model can be therefore, one must turn to the theoretical estimate.
observed and measured: it is the bubble size. For this estimation, consider the example of Figure 13, which
The chemical conversion for this model involves five represents a bed 2 m in diameter and 2 m high, containing 7
factors: the two transfer rates and the three reaction rates. This tons of catalyst with the rate constant shown.
is shown in Figure 12. One should also note that these rate Let us see what conversion this model predicts for different
factors can be determined without computer calculations. bubble sizes, and let us see if the predictions are reasonable.
PAGE EST: 4.7 E

Problems with Predictions and Scaleup


First, from a laboratory unit, it is almost impossible to
determine the bubble size in a scaled-up reactor, because so
many factors will affect it: type of distributor for the bed,
fraction of fines in the solid mixture, fraction of inert solids
mixed with the catalyst, solid properties especially its density,
etc. (see Figure 16). Also, it is difficult to obtain a reasonable
estimate of the size of the crazily changing shapes of the bubbles
in the bed, and one must remember that the bubble size is the
prime factor that controls the conversion.
Thus, scaleup and design of reaction-controlled BFB is
fraught with uncertainty. Maybe one should build a full-scale
Figure 16. Schematic depiction of the expected progression with the cold (hydrodynamic) model. Certainly one should learn from
Kunii-Levenspiel bubbling bed model. previous experience, or as one of my friendly bridge players
would say, “one peek is worth two finesses”. On the other hand,
We find that, for no bubbles (a packed bed), 95% conversion the BFB is certainly useful for systems where heating effects
is observed; as the bubbles grow in size, from 8 cm to 16 cm are rate-controlling, such as highly exothermic or endothermic
to 32 cm, the conversion decreases to 84%, to 59%, and to 32%. reactions, or where temperature effects dominate.
This is shown in Figure 14.
Thus, larger bubbles allow more gas to pass by, and thus Literature Cited
reducing the gas conversion; this seems reasonable.
(1) Kunii, D.; Levenspiel, O. Fluidization Engineering; Wiley: New
Next, for reactions in series with the intermediate as the York, 1969; Chapter 7.
desired product, we see the results shown in Figure 15. Again (2) (a) Johnstone, H. F.; Batchelor, J. D.; Shen, C. Y. Low-Temperature
we see that larger bubbles allow more gas to pass by, and more Oxidation of Ammonia in Fixed and Fluidized Beds. AIChE J. 1955, 1,
solids are needed to reach the optimum; this decreases the 318. (b) Shen, C. Y.; Johnstone, H. F. Gas-Solid Contact in Fluidized
Beds. AIChE J. 1955, 1, 349.
maximum amount of intermediate obtainable. This is the
(3) Kunii, D.; Levenspiel, O. Fluidized Reactor Models. 1. For Bubbling
expected progression with increased bubble size. Beds of Fine, Intermediate, and Large Particles. 2. For the Lean Phase:
Finally, calculations show that tall narrow beds behave better Freebaord and Fast Fluidiziation. Ind. Eng. Chem. Res. 1990, 29, 1226.
than fat shallow beds. Also, adding inerts improves bed behavior (4) Kunii, D.; Levenspiel, O. Fluidization Engineering, 2nd Edition;
when all else (bubble size, volumetric feed rate, and total amount Butterworth-Heinemann: Boston, 1991; Chapter 12.
(5) Grace, J. R. An Evaluation of Models of Models of Fluidized Bed
of catalyst) remains unchanged. This is the expected progression Reactors. Chem. Eng. Prog. Symp. Ser. 1971, 67 (116), 159.
with the Kunii-Levenspiel bubbling bed model. All this is (6) Davidson, J. F.; Harrison, D. Fluidized Particles; Cambridge
shown in Figure 16. University Press: New York, 1963.

Octave Levenspiel

Emeritus Professor,
Chemical Engineering Department,
Oregon State UniVersity,
CorVallis, Oregon 97331-2702
IE071355V

You might also like