PATTERNS OF FLOW I N CHEMICAL PROCESS VESSELS

Octave Levenspiel
Department of Chemical Engineering
Illinois institute of Technology. Chicago. Illinois
and
Kenneth 8 Bischoff.
Department of Chemical Engineering
University of Texas. Austin. Texas

I. Introduction ........................ ......................... 95

....................... ......................... 95
low . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
C . Stimulus-Response Methods of Characterizing Flow . . . . . . . . . . . . . . . . 98
D . Ways of Using Tracer Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
I1. Dispersion Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
A . General Description . . . . . . . . . . . . . . . . . . . . . . . . ................... 105
B . Mathematical Description . . . . . . . . . . . . . . . . . . ................... 107
C . Measurement of Dispersion Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
D . Relationships Between Dispersion Models . . . . . . . . . . . . . . . . . . . . . . . . . . 134
E . Theoretical Methods for Predicting Dispersion Coefficients . . . . . . . . . . 142
I11. Tanks-in-Series or Mixing-Cell Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
A . Introduction . . . . . . . . . . . . . . . .................................... 150
B. Mathematical Description . . .................................... 151
C . Comparison with the Dispersion Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
IV . Combined Models . . ........................................ 158
A . Introduction . . . . ....................................... 158
B . Definition of Deadwater Regions . ............................. 159
C. Matching Combined Models t o Experiment . . . . . . . . . . . . . . . . . . . . . . . . 161
D . Application to Real Stirred Tanks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
E . Application to Fluidized Beds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
V . Application of Nonideal Patterns of Flow to Chemical Reactors . . . . . . . . 171
A . Direct Use of Age Distribution Information . . . . . . . . . . . . . . . . . . . . . . . . . 173
B . Stirred Tank Reactors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
C . Tubular and Packed Bed Reactors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
D . Fluidized Bed R .......................................
VI . Other Applications .........................................
A . The Intermixing ds Flowing Successively in Pipelines . . . . . . . . . . 187
B. Brief Summary of Applications to Multiphase Flow and Other
Heterogeneous Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
VII . Recent References . . ............................ ............ 189
Nomenclature . . . . . . . . . . . . . . . . . . ................................... 190
Text References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192

.
I Introduction
A . SCOPE
Fluid is passed through process equipment so that i t may be modified
one way or other . It may be heated or cooled. it may gain or lose material
95
96 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

by mass transfer with an adjoining phase (either solid or fluid), or it may

react chemically.
To predict the performance of equipment we must know: ( a ) the
rate a t which fluid is modified as a function of the pertinent variables,
and (b) the way fluid passes through the equipment.
Of all possible flow patterns two idealized patterns, plug flow and
backmix flow, are of particular interest. Plug flow assumes that fluid
moves through the vessel “in single file” with no overtaking or mixing
with earlier or later entering fluid. Backmix flow assumes that the fluid in
the equipment is perfectly mixed and uniform in composition throughout
the vessel. Design methods based on these ideal flow patterns are rela-
tively simple and have been developed for heat and mass transfer equip-
ment as well as for chemical reactors.
All patterns of flow other than plug and backmix flow may be called
nonideal Pow patterns because for these the design methods are not
nearly as straightforward as those for the two ideal flow patterns. The
methods of treating nonideal patterns either have only recently been
developed or are yet to be developed.
I n real vessels flow is usually approximated by plug or backmix flow;
however, for proper design, the departure of actual flow from these
idealizations should be accounted for. Here we intend to consider these
nonideal flow patterns; to characterize them, to measure them and to
use this information in design.
The treatment of these nonideal flow patterns divides naturally into
two parts.
(1) FZoul of Single Fluids. The major application here is the design
of chemical reactors (homogeneous and solid-catalyzed fluid systems
using tubular, packed bed or fluidized bed reactors). Of relatively minor
importance is the application to heat transfer from a single flowing fluid
and contamination of fluids flowing successively in pipelines.
(2) Flow of Two Fluids. The major applications are in absorption,
extraction, and distillation, with and without reaction. Other applica-
tions, also quite important, are for shell-and-tube or double-pipe heat
exchangers, and noncatalytic fluid-solid reactors (blast furnace and ore-
reduction processes).
This paper deals with the nonideal flow of single fluids through
process equipment.

FLOW
B. TPPESOF
Names have been associated with different types of flow patterns of
fluid in vessels. First of all, we have the two previously mentioned ideal
flow patterns, plug flow and backmix flow. Flow in tubular vessels ap-
 PATTERNS OF FLOW I N CHEMICAL PROCESS VESSELS 97

proximates the ideal conditions of plug flow, while flow in agitated or

stirred tanks often closely approximates backmix flow.
Descriptive terms, not mutually exclusive, such as channeling, re-
cycling, eddying, existence of stagnant pockets, etc., are used in connec-
tion with various forms of nonideal flow. I n channeling, large elements
of fluid pass through the vessel faster than others do; however, all fluid
does move through the vessel. Channeling may be found in flow through
poorly packed vessels or through vessels having small length-to-diameter
ratios. Stagnant pockets of fluid may occur in headers, at the base of
AStagnant regions

-
--------. Extreme short-circuiting
_ _ _-- ~. ~ _ ~ and bypassing; a result
of poor design
Channeling; especially
serious in countercurrent
two-phase opcrations

3)
FIG.1. Nonideal flow patterns which may exist in process equipment (L13).

pressure gages, and in odd-shaped corners, and cause bypassing of these

regions. By-passing serves to cu't down the effective or useful volume of
the equipment, is not desirable, and is an indication of poor design.
I n recycling, a certain amount of fluid is recirculated or returned to the
vessel inlet. This type of flow may be desirable, for example, in auto-
catalytic or autothermal reactions, and can be promoted by suitable
baffling arrangements or by proper vessel design. These various types of
nonideal flow are illustrated in Fig. 1.
I n most cases nonideal flow is not desired, and by proper design its
gross aspects can be eliminated. However, even with proper design, some
extent of nonideal flow remains due both to molecular and turbulent
98 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

diffusion and to the viscous characteristics of real fluids which result in

velocity distributions. Hence nonideal flow must be accounted for, prefer-
ably in such a manner that quantitative predictions of performance of
real equipment can be made.

METHODS
C. STIMULUS-RESPONSE FLOW
OF CHARACTERIZING

To be able to account exactly for nonideal flow requires knowledge

of the complete flow pattern of the fluid within the vessel. Because of the
practical difficulties connected with obtaining and interpreting such
information, an alternate approach is used requiring knowledge only of
how long different elements of fluid remain in the vessel. This partial
information, not sufficient to completely define the nonideal flow within
the vessel, is relatively simple to obtain experimentally, can be easily
interpreted, and, either with or without the use of flow models, yields
information which is sufficient in many cases to allow a satisfactory
accounting of the actual existing flow pattern.
The experimental technique used for finding this desired distribution
of residence times of fluid in the vessel is a stimulus-response technique
using tracer material in the flowing fluid. The stimulus or input signal is
simply tracer introduced in a known manner into the fluid stream enter-
Tracer TtEW

Cyclic brcsr
input signal

Step tracer
input signal

signal

Tim Tim

FIQ.2. Stimulus response techniques commonly used in the study of the behavior
of flow systems (L13).
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 99

ing the vessel. This input signal may be of any type; a random signal, a
cyclic signal, a step or jump signal, a pulse or discontinuous signal or
any arbitrary input signal. The response or output signal is then the
recording of tracer leaving the vessel (see Fig. 2).
We will restrict this treatment to steady-state flow with one entering
stream and one leaving stream of a single fluid of constant density.
Before proceeding further let us define a number of terms used in con-
nection with the nonideal flow of fluids.
1. Open and Closed Vessels
We shall define a closed vessel to be one for which fluid moves in
and out by bulk flow alone. Plug flow exists in the entering and leaving
streams. I n a closed vessel diffusion and dispersion are absent a t entrance
and exit so that we do not, for example, have material moving upstream
and out of the vessel entrance by swirls and eddies.
An open vessel is one where neither the entering nor leaving fluid
streams satisfy the plug flow requirements of the closed vessel. When
only the input or only the output fluid stream satisfies the closed vessel
requirements we have a closed-open or open-closed vessel.

2. M e a n Residence T i m e of Fluid in a Vessel

The mean residence time of fluid in a vessel is defined as:
volume of vessel
t- = available for flow - --
V
volumetric flow rate v
of fluid through vessel
3. Reduced T i m e
It is frequently convenient to measure time in terms of the mean
residence time of fluid in the vessel. This measure, called the reduced
time, is dimensionless and is given by

4. I and I ( t ) - T h e Internal Age Distribution of a Fluid in a Closed

Vessel
Taking the age of an element of fluid in a vessel to be the time it
has spent in the vessel, it is evident that the vessel contains fluid of
varying ages. Let the function I be the measure of the distribution of
ages of fluid elements in the vessel and let it be defined in such a way
+
that I dB is the fraction of fluid of ages between 8 and 0 d8 in the vessel.
100 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

Since the sum of all these fractions of fluid is unity (the total vessel
contents) we have
/oW1dB=1 (3)
The fraction of vessel contents younger than age 0 is

The fraction older than 8 is

1"IdB' = 1- LIdB'
Where time rather than reduced time is used let the internal age
distribution function be I ( t ). Then

I = t I(t) (4)
and the relationships corresponding to those using reduced time follow
(see Fig. 3 ) .

ction of vesSel contents

Ags distribution of fluid

A
1 in exn stream; total area
under curyo is unity
Internal age distribution; E
1 lopa is never positive, total
area under curve is unity

e d
FIG.3. Typical internal age distri- FIG.4. Typical distribution of resi-
bution of vessel contents (L13). dence times of fluid flowing through
a vessel (L13).

5. E and E(t)-The Residence-time Distribution of Fluid in a Closed

Vessel or the Age Distribution of Exit Stream
I n a manner similar to the internal age distribution function, let E be
the measure of the distribution of ages of all elements of the fluid stream
leaving a vessel. Thus E is a measure of the distribution of residence
times of the fluid within the vessel. Again the age is measured from the
time that the fluid elements enter the vessel. Let E be defined in such a
way that E dB is the fraction of material in the exit stream which has an
+
age between 0 and 0 do. Referring to Fig. 4, the area under the E vs. 6
curve is
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 101

The fraction of material in the exit stream younger than age 0 is

while the fraction of material older than 8, the shaded area of Fig. 4,is

Where time rather than reduced time is used let the residence time
distribution be designated by E ( t ). Then

E = 'iE(1) (6)
and the relationships corresponding to those using reduced time follow.

6 . F Curve-The Response to a Step Tracer Input

With no tracer initially present, let a step function (in time) of tracer
be introduced into the fluid entering a vessel. Then the concentration-
time curve for tracer in the fluid steam leaving the vessel, measured in
terms of tracer concentration in the entering stream Co and with time in
reduced units, is called the F curve. As shown in Fig. 5 , the F curve
rises from 0 to 1.

" Step function

tracer input signal
1 --

Tracer output
I*

0- t--

FIG.5. Typical downstream response to an upstream step input; in the dimension-

less form shown here this response is called the F curve (L13).

7. C Curve-The Response to an Instantaneous Pulse Tracer Input

The curve which describes the concentration-time function of tracer
in the exit stream of any vessel in response to an idealized instantaneous
or pulse tracer injection is called the C curve. Such an input is often
called a delta-function input. As with the F curve, dimensionless co-
ordinates are chosen. Concentrations are measured in terms of the initial
concentration of injected tracer if evenly distributed throughout the
102 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

vessel, CO, while time is measured in reduced units. With this choice the
area under the C curve always is unity, or

or
Co=lmCdO=~lmCdl

A typical C curve is shown in Fig. 6.

The terms F, C,I and E were introduced by Danckwerts (D4, D6).

I r a c c ~input signal

u
or c C U M
V

0 1
0
FIG.6. Typical downstream response to an upstream delta function input; in the
dimensionless form shown here this response is called the C curve (L13).

8. Relationship between Tracer Curves, Age Distribution, and Residence-

Time Distribution of Fluids Passing through Closed Vessels
By material balance, the experimental response curves F and C can
be related to the I and E distributions. Thus, a t any time t or 6, we have
from Danckwerts (D4) or Levenspiel (L13),

F = 1 - I = 1 - f I(t) = E do’ = /d E(t‘) dt’ = C dd’ (8)

or

Equation (9) is a special case of the easily proven fact that for a linear
system, if input 2 is the derivative of input 1, then output 2 is also the
derivative of output 1.
These relationships show that the F curve is related in a simple way
to the age distribution of material in the vessel, while the C curve gives
directly the distribution of residence times of material in the vessel. I n
addition, the C and E curves represent the slopes of the corresponding
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 103

F and I curves, while the F and I curves a t a given time represent the
area under the corresponding C and E curves up to that time.
Thus we see that the stimulus-response technique using a step or pulse
input function provides a convenient experimental technique for finding

---*
the age distribution of the contents and the residence-time distribution of
material passing through a closed vessel.
ffow Backmix flow Arbitrary flow

23-
t 1"-

1
F

FIG.7. Properties of the F, C, I, E curves for particular patterns of flow (dead-

water regions and bypassing flow absent) in closed vessels (L13).

Figure 7 shows the shapes of these curves for various types of flow.
It is interesting to relate the mean of the E curve

- hm 0EdB
1 m

BE =
LmE d 0 = OEd0 (10)

to the mean residence time of fluid in the vessel, 8= 1 or Tt: By material

balance we find
&=e=1
-
tE = t
1 } only for closed vessels.
For other than closed vessels, this does not hold. This fact, probably
104 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

somewhat strange to the intuition, was first shown by Levenspiel and

Smith (L16) and was proved in a general manner by Spaulding (521).
D. WAYSOF USINGTRACER
INFORMATION
Suppose we are given two separate pieces of information, the resi-
dence time distribution for plain unchanging fluid passing through the
vessel and the kinetics' for a change which is to be effected in the vessel.
Can we predict what will be the performance of the vessel when this
change is occurring in the vessel? The answer depends on the type of
change occurring. If this is simply a linear function of an intensive
property of the fluid, then the performance can be predicted, or
tracer rate information performance
information for a process of equipment
for flowing + with rate linear -+ when this
unchanging in an intensive process is
fluid fluid property occurring
As illustrations, consider the following three cases. ( a ) Isothermal
reactor f o r first order reaction. Knowledge of the rate constants for the
reaction and of response characteristics for the vessel suffices to predict
how the vessel will perform as a reactor. (b) Nonisothermal reactor. If
the temperature, hence, the rate, is a function of position as well as
concentration, then tracer plus rate information is insufficient to predict
performance. (c) Heat transfer to a fluid flowing in a heat exchanger.
Since the heat-transfer rate depends not only on the temperature of fluid
and of the vessel walls but also on the contacting pattern, the perform-
ance of the vessel as a heat exchanger cannot be predicted by having only
the tracer information and the heat transfer coefficient. Applications will
be treated in more detail later.
If the linear requirements on the kinetics of change are not satisfied,
then the performance of equipment cannot be predicted from these
separate pieces of information. I n this case the actual flow pattern of
fluid through the vessel must be known before performance predictions
can be made.
As mentioned earlier, obtaining and interpreting the actual experi-
mental flow pattern is usually impractical. Hence, the approach taken is
to postulate a flow model which reasonably approximates real flow,
and then use this flow model for predictive purposes. Naturally, if a flow
model closely reflects a real situation, its predicted response curves will
closely match the tracer-response curve of the real vessel; this is one
of the requirements in selecting a satisfactory model.
'Kinetics is used here to mean any type of rate process.
 PATTERNS OF FLOW I N CHEMICAL PROCESS VESSELS 105

This is a fruitful approach, and much of what follows concerns the

development and use of such flow models. The parameters of these models
are correlatable with physical properties of the fluid, vessel geometry,
and flow rate; once such correlations are found for all types of fluid
processing, performance predictions can be obtained without resort to
experimentation.
Many types of models can be used to characterize nonideal flow
patterns within vessels. Some draw on the analogy between mixing in
actual flow and a diffusional process. These are called dispersion models.
Others visualize various flow regions connected in series or parallel.
When backmix flow occurs in all these flow regions we have the tanks-in-
series, mixing-cell, or backmix models. When different types of flow
regimes are interconnected, we have combined models. Some models are
useful in accounting for the deviation of real systems (such as tubular
vessels or packed beds) from plug flow, others for describing the devia-
tion of real stirred tanks from the ideal of backmix flow.
Models vary in complexity. One-parameter models seem adequate to
represent packed beds or tubular vessels. On the other hand models in-
volving up to six parameters have been proposed to represent fluidized
beds.
We shall first discuss the dispersion and backmixing models which
adequately characterize flow in tubular and packed-bed systems ; then
we shall consider combined models which are used for more complex
situations. I n connection with the various applications, the direct use
of the age-distribution function for linear kinetics will also be illustrated.

II. Dispersion Models

A. GENERALDESCRIPTION
Dispersion models, as just stated, are useful mainly to represent flow
in empty tubes and packed beds, which is much closer to the ideal case of
plug flow than to the opposite extreme of backmix flow. I n empty tubes,
the mixing is caused by molecular diffusion and turbulent diffusion,
superposed on the velocity-profile effect. In packed beds, mixing is caused
both by “splitting” of the fluid streams as they flow around the particles
and by the variations in velocity across the bed.
When flow is turbulent the resulting concentration2 fluctuations are
rapid, numerous, and also small with respect to vessel size. They might be
considered to be random, which would lead to a diffusion-type equation.
In actual fact, the fluctuations are not independent, and correlations
*Concentration is used here in a general way. It could also represent tempera-
ture, etc.
 w
0
Q,

TABLE I
DISPERSIONMODELSO

Simplifying assumptions
or restrictions in addition
to those for the model Parameters of !2
Name of model above model De&ing differential equation d
General dispersion:
includes chemical Constant density
reaction and
source terms U

General dispersion Bulk flow in axial

in cylindrical direction only.
coordinates Radial symmetry

Uniform dispersion Dispersion coefficients

independent of position
hence constant

Dispersed plug flow Fluid flowing a t mean DR, DL, u

velocity, hence plug flow
0
Axial dispersed No variation in properties D'L, u ac
-
aT
ac azc
+ u - = D' L s 2 + S + r c 3
plug flow in the radial direction (1-5)

* From Bischof and Levenspiel (B14.)

 PATTERNS OF FLOW I N CHEMICAL PROCESS VESSELS 107

exist between them. Unfortunately, the inclusion of such correlations into

the analysis would greatly complicate i t ; instead of a partial differential
equation of the diffusion type, we would obtain an integro-diff erential
equation. Moreover, the detailed theories of turbulence are not yet suf-
ficiently developed to justify their use to describe mixing (especially in
such complicated systems as packed beds). Hence, we shall not discuss
them deeply; a thorough treatment is given by Hinze (H9).
As the alternative, a phenomenological description of turbulent mix-
ing gives good results for many situations. An apparent diffusivity is
defined so that a diffusion-type equation may be used, and the magnitude
of this parameter is then found from experiment. The dispersion models
lend themselves to relatively simple mathematical formulations, analo-
gous to the classical methods for heat conduction and diffusion.
The only real test of such an assumption is its success in representing
real systems. I n general such models seems to work well for both empty
tubes and packed beds; however, recent work has shown limits on the
accuracy with which the mixing may be represented by these models
(B4,R3).
The remainder of this section will be devoted to describing the
methods for measuring dispersion coefficients and the resulting correla-
tions of the data.
B. MATHEMATICAL DESCRIPTIONS
Table I lists the equations corresponding to the various dispersion
models ranging from the most general to the most restricted.

1. General Dispersion Model

Equation (1-1) is the general representation of the dispersion model.
The dispersion coefficient is a function of both the fluid properties and the
flow situation; the former have a major effect a t low flow rates, but
almost none a t high rates. I n this general representation, the dispersion
coefficient and the fluid velocity are all functions of position. The disper-
sion coefficient, D, is also in general nonisotropic. I n other words, it has
different values in different directions. Thus, the coefficient may be
represented by a second-order tensor, and if the principal axes are taken
to correspond with the coordinate system, the tensor will consist of only
diagonal elements.

2. General Dispersion Model for Symmetrical Pipe Flow

The first simplification of importance is for the frequently encoun-
tered situation of symmetrical axial flow in cylindrical vessels. For this
particular geometry, the dispersion coefficient tensor reduces to (H8),
108 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

D = [
DL(R> 0
O
0
DR(R>
0 DR(R>
] (11)

and Eq. (1-1)reduces to Eq. (1-2). With the dispersion coefficients in the
axial and radial directions, D L ( R ) and D R ( R ), and the fluid velocity,
u ( R ), all functions of radial position, analytical solutions of this equa-
tion are impossible. This makes evaluation of dispersion extremely dif-
ficult; hence further simplifications are needed to permit analytical solu-
tions to the differential equation.

3. Uniform Dispersion Model

If the axial and radial dispersion coefficients are each taken to be
independent of position, we get Eq. (1-3) for which an analytical solution
will sometimes be possible. We shall call the coefficients of this model the
uniform dispersion coefficients; thus the parameters of this model are
DRn: DLm’ and u ( R ) .

4. Dispersed Plug-Flow Model

Even with constant dispersion coefficients, accounting for the velocity
profile still creates difficulties in the solution of the partial differential
equation. Therefore it is common to take the velocity to be constant a t
its mean value u. With all the coefficients constant, analytical solution
of the partial differential equation is readily obtainable for various situa-
tions. This model with flat velocity profile and constant values for the
dispersion coefficients is called the dispersed plug-flow model, and is
characterized mathematically by Eq. (1-4). The parameters of this model
are Dn, DL and u.
I n this model, the effect of the velocity profile is “lumped” into the
dispersion coefficients, as will be discussed later. I n comparison, the
coefficients calculated from the uniform dispersion model, or the general
dispersion model, are more basic in the sense that they do not have two
effects combined into one coefficient.

5 . Axial-Dispersed Plug-Flow Model

When there is no radial variation in composition in the fluid flowing
in the cylindrical vessel, the only observable dispersion takes place in
the direction of fluid flow. I n this situation Eq. (1-4) reduces to Eq. (I-5),
and we get the axial-dispersed plug-flow model with parameters D’L
and u.
With this model the mathematical problems are greatly simplified
since the radial variable is eliminated completely, decreasing the number
 PATTERNS OF FLOW I N CHEMICAL PROCESS VESSELS 109

of independent variables by one. Because of this simplification it is con-

venient to use this form whenever possible. The justification and limita-
tions of this model will be discussed later.

C. MEASUREMENT
OF DISPERSION
COEFFICIENTS
1. Tracer Injection
As has been discussed, the usual method of finding the dispersion
coefficients is to inject a tracer of some sort into the system. The tracer
concentration is then measured downstream, and the dispersion coeffi-
cients may be found from an analysis of the concentration data. For
these tracer experiments there are no chemical reactions, and so r, = 0.
Also the source term is given by
I
S = - 6(X
?r
- Xo)f(R) (12)
where
I = injection rate of tracer,
6(X - X o ) = Dirac delta function (S20), simply indicating that tracer is
introduced a t position x,,.
f ( R ) = 1/E2,R 5 El
= 0, E 5 R 5 Rot
E = injector-tube radius.
If Eq. (12) is substituted into Eq. (I-2), it becomes,
ac ac
ax = DL(R)- + R- -
+ u(R)- 1 a RDdR) a~
a2c
at ax2aR

+I - Xo)f(R> (13)
Equation (13) is the starting point for the detailed discussion of meas-
urement techniques to follow.
2. Axial -Dispersed Plug-F 1ow M ode1
a. Preliminary. Three methods are commonly used to find the effec-
tive axial dispersion coefficient, all involving unsteady injection of a
tracer either in the form of a pulse or delta function, a step function,
or a periodic function such as a sine wave, over a plane normal to the
direction of flow. The tracer concentration is then measured downstream
from the injection point. The modification of this input signal by the
system can then be related t o the dispersion coefficient which character-
izes the intensity of axial mixing in the system.
The same information can be found from all methods. The periodic
110 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

methods present advantages in situations with extremely rapid response ;

however, the nonperiodic methods, especially the pulse methods, are
often preferable from the point of view of simplicity of experimental
equipment and ease of mathematical analysis. For this reason, our discus-
sion will therefore be centered about the pulse methods.
If a pulse of tracer is injected into a flowing stream, this discontinuity
spreads out as i t moves with the fluid past a downstream measurement
point. For a fixed distance between the injection point and measurement
point, the amount of spreading depends on the intensity of dispersion
in the system, and this spread can be used to characterize quantitatively
the dispersion phenomenon. Levenspiel and Smith (L16) first showed
that the variance, or second moment, of the tracer curve conveniently
relates this spread to the dispersion coefficient.
I n general, all moments are needed to characterize any arbitrary
tracer curve, the first moment about the origin locating the center of
gravity of the tracer curve with respect to the origin, the second moment
about this mean measuring the spread of the curve, the third moment
measuring “skewness,” the fourth measuring lLpeakedness,”etc. However,
under the assumptions of the dispersion model, the tracer curve is gener-
ated by a random process, so only two moments need be considered
independent and the rest are functions of these. For convenience, we
choose the first two moments, the mean and variance, as independent.
The next problem is to find the functional relationship between the
variance of the tracer curve and the dispersion coefficient. This is done
by solving the partial differential equation for the concentration, with
the dispersion coefficient as a parameter, and finding the variance of this
theoretical expression for the boundary conditions corresponding to any
given experimental setup. The dispersion coefficient for the system can
then be calculated from the above function and the experimentally found
variance.
b. Perfect Pulse Injection. We shall first put Eq. (13) into the form
needed for mathematical solution, and then briefly discuss the boundary
conditions used by previous workers. I n Eq. (13) set the radial terms
equal to zero, make the velocity constant, and substitute DL’for DL(R).
This gives
ac
- + - = DL’a
u ac -2c -I 6(X - XO)-
+ 1
at ax ax2 Ro2
The function f(R)becomes 1/Ro2, since injection is uniform over the
entire plane. The rate of tracer injection, I, is now a function of time.
If we define CLve as the concentration of injected tracer if evenly dis-
tributed throughout the vessel, for a perfect pulse input, we have
 PATTERNS OF FLOW I N CHEMICAL PROCESS VESSELS 111

I = c:,,vs(t) (15)
It is convenient to change the variables to dimensionless form:
e = ut/L = vt/v
x = X/L
P = uL/DL'
L = length of test section
c' = c/c:,,
Equation (14) then becomes,

ac'
-
ae + ax
- = - - + 6(x - x,,)s(e)
ac' 1
pax2
~ Z C

The mean and variance of the tracer curve are defined as

pl = ed dB = mean about e origin

= first moment about e origin (17)

2 = La(e - pJ2c' de = second moment about the mean (18)

As shown by van der Laan (V4) and Aris (A7), if Laplace transforms
are used to solve Eq. (16), the mean and variance may be easily found
from the relations

and in general,

pn = LwenddB = nth moment about e origin

where
p = Laplace transform variable,
F' = Laplace transform of c'.
The only difference now in various treatments comes from the boundary
conditions used to solve Eq. (16).
The simplest type of boundary conditions to use in solving Eq. (16)
are the so-called "infinite pipe" conditions. With these conditions, the
+
vessel is assumed to extend from- 00 to 00. Physically this means that
112 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

the changes in flow a t the ends of the vessel are neglected. Usually the
flowing fluid is led into the vessel in which the dispersion is to be meas-
ured through a pipe that has dispersion characteristics different from t h a t
of the vessel. Likewise, the exit pipe will generally have different charac-
teristics than the vessel. These end effects will affect the measurement of
the dispersion in the main vessel and should be taken into account. It is

TABLE I1
EXPERIMENTAL
SCHEMESUSEDI N RELATION
TO THE AXIAL-
DISPERSED
PLUG-FLOW MODEL
Expression for finding
Experimental scheme Where first derived
the dispersion cocfficieni

Levenspiel ond

Smith
[L161
I FL-
6 function
4
output
inout
x=x,

=4zF x:x,

6finction
X=Xm

'
van der L o a n [V4]

II I
any input
r-'7 Bischoff (Ell)

A p , Eq. (38) Arks (AB)

I any input A+', Eq.(39) Bischoff (81I)

(e) ee reference (814) Bischoff a n d

Levenspiel (814)

;ee reference (614) Bischoff and

Levenspiel (814)
 PATTERNS O F FLOW IN CHEMICAL PROCESS VESSELS 113

found, if tracer is injected and measured far enough from the ends of
the vessel, that the end effects are negligible; the distances from the ends
that are necessary in order that this be true for various cases will be
discussed later.
Levenspiel and Smith (L16) dealt with this simplest of cases which
is shown in the first sketch of Table 11. Using the open vessel assumption
and a perfect delta-function input, they found that the concentration
evaluated a t x = 1 was given by

exp [ -
c' = 51 (Z)lip
P
-
4e e'21
From this equation (or from its Laplace transform) the first and second
moments are found to be,

PI, = 1 p
2
+
= p + pz
2 8
urn2

Van der Laan (V4) extended this for much more general boundary
conditions that took into account the different dispersion in the entrance
and exit sections. These boundary conditions were originally introduced
by Wehner and Wilhelm (W4). They assumed that the total system could
be divided into three sections: an entrance section from X = - co to
X = 0 (designated by subscript a ) , the test section from X = 0 to
X = X, (having no subscript), and the exit section from X = X,to CQ +
(designated by subscript b ) , each section having different dispersion
characteristics. This is illustrated in the second sketch in Table 11.
The boundary-value problem then had the form:

X I 0

with boundary conditions,

114 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

1 act 1 ac;
P-
ct(xo-, e) - - e) = cbt(xe+,e)
ax (x~--, - pb
-- ax 0) (26e)

c’(x~-,e) = c;(x,+, e) (26f)

Cbt(+W, e) = finite (26d
The physical significance of these boundary conditions is as follows.
Equation (26a) represents the fact that just before the tracer is injected
into the system the concentration is everywhere zero, Equations (26b)
and (26g) are obvious since a finite amount of tracer is injected. Equa-
tions (26d) and (26e) follow from conservation of mass a t the boundaries
between the sections (W4) ; the total mass flux entering the boundary
must equal that leaving. Equations (26c) and (26f) are based on the
physically intuitive argument that concentration should be continuous
in the neighborhood of any point. These boundary conditions will be used
extensively in the subsequent derivations.
Van der Laan (V4) solved the above system by Laplace transforms,
and obtained the following result for the transform of the concentration
<
evaluated a t x = xm ( xo < xm x,) :
(q + qa)(q + qb) + (q - qa)(q - qb)e-2qp(4-1)
e(1/2--OP + (q - qJCq + qb)e-2‘lho + (q + q,)fq - qb)e-2qp(L-m)
-cm’ = -
2q (9 + qa>(q+ (13 - (q - qa)(q - qb)e-2qh.
(27)
It can be seen, from the complexity of Eq. (27), that to find the
inverse Laplace transform in the general case would be exceedingly
difficult, if not impossible. Yagi and Miyauchi ( Y l ) have presented a
solution for the special case where D, = Db= 0. Fortunately, the mo-
ments can be found in general from Eq. (27) without evaluating the
inverse transform. They are,

a2 =
P
+- P2 (8 + 2(1 - a)(l - b ) e - h
1

- (1 - a)e-h[4xoP + 4(1 + a) + (1 - a)e--]

- (1 - b)e--P(x1-&)[ 4 ( ~ 1- Xm)P + 4(1 + b) + (1 - b)e-p(xl-xm)]}
(29)
where a = P/Pa and b = P/Pb. Van der Laan (V4) shows how these
equations reduce to simpler forms for many cases. I n particular, they
reduce to Eq. (23) and (24) for the open vessel where a = 1, b = 1.
c. Imperfect Pulse Injection. Both Levenspiel and Smith’s and van
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 115

der Laan’s work depended on being able to represent the tracer injection
by a delta function, a mathematical idealization which physically can
only be approximated since it requires that a finite amount of tracer
be injected in zero time. The closer a physical injection process approxi-
mates a perfect delta function, the greater must be the amount of tracer
suddenly injected. However, since we are trying to measure the proper-
ties of the system, we would like to disturb the system as little as possible
with the injection experiment. Thus we should inject slowly from this
standpoint. Unfortunately, these two requirements are in opposite direc-
tions. To satisfy the mathematical delta function we must inject very
rapidly, but in order to not disturb the system we must inject slowly.
Aris (A8), Bischoff ( B l l ) , and Bischoff and Levenspiel (B14) have
utilized a method that does not require a perfect delta-function input.
The method involves taking concentration measurements a t two points,
both within the test section, rather than a t only one as was previously
done. The remaining sketches in Table I1 show the systems considered.
The variances of the experimental concentration curves a t the two points
are calculated, and the difference between them found. This difference
can be related to the parameter and thus to the dispersion coefficient.
It does not matter where the tracer is injected into the system as long
as it is upstream of the two measurement points. The injection may be
any type of pulse input, not necessarily a delta function, although this
special case is also covered by the method.
Since the injection point is not important, it is convenient to base
the dimensionless quantities on the length between measurement points.
Therefore, we will here call X o the first measurement point rather than
the injection point as in Levenspiel and Smith’s or van der Laan’s work.
The position X , will be taken as the second measurement point. The
injection point need only be located upstream from X o . Equation (16)
is again the basis of the mathematical development. With the test sec-
<
tion running from X = 0 to X = X , we shall measure first a t X o 0 and
then a t X , > 0 where the second measurement point can be either within
the test section, X ,< X,, or in the exit section, X , 2 X,. Tracer is in-
jected a t X < X o . The boundary-value problem that must be solved is
somewhat similar to that of van der Laan:
116 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

with boundary conditions :

c,’(x, 0) = C’(X, 0) = cd(x, 0) = 0
CO’(X~,el = co’(e)

cd(+w, 0) = finite (33d

There are no source terms in Eq. (31), since the injection point is
upstream from the sections over which the equations are to be used. The
boundary conditions are of the same type used by van der Laan (V4),
except for Eq. (33b). This merely states that we are going to measure
the concentration a t X = X o , which we call c4. The solution of Eqs.
(30), (31), (32), and (33) is again accomplished by the use of Laplace
transforms following the same scheme as given in (A8, B11, B14). The
Laplace transform of the concentration a t x = x, is
--I

Fo’
= 2 {(q - q b ) exp [(+ + q)P(x, - x,)]
+ (q + qa) exp [(3 - dP(xm- xJI}, xm ,< x e (34)

where
A = (q + qa){(q - qb) exp [(3 + qa)PaxO - (3 + q)%I
+ (q + q b ) exp [(3 - qa)Plxo - (3 - q)F’~el)
- (q - qa){(P - qb) exp [(+ - q0)p4x0 - (3 + q>P~el
+ (q + qd exp [(3 + q4)p4x0 - (3 - q)PxeI)
Notice that the right-hand side of Eq. (34) is equal to the ratio of the
transformed concentration a t the second measurement point to the trans-
formed concentration a t the first measurement point. I n the terminology
of control engineering, this quantity is the transfer function of the
system between X o and X,. The Laplace-transform method is possible
because the diffusion equation is a linear differential equation. Thus,
the right-hand side of Eq. (34) could in principle be used in a control-
system analysis of an axial-dispersion process.
 PATTERNS O F FLOW I N CHEMICAL PROCESS VESSELS 117

The mean and variance can be found from Eqs. (34) and (35) by
use of van der Laan's method, Eq. (21). The results are of the form
Api = - pi0 = #i(P, Pa, Pb, XO,x, x,)
pim (36)
Au' = am2- UO' = #z(P, Pa, Pb, XO, x,, x,) (37)
where and q2 are cumbersome functions given by Bischoff and Leven-
spiel (B14). I n order to use this method, the tracer concentration is
measured a t two points X o and X,, and the variances calculated. The
difference in variances can be found, and this number is then used in
Eq. (37) along with the physical dimensions of the experimental ap-
paratus to calculate P, hence DIL.
An interesting simplification of these general expressions occurs when
we take both measurements within the test section. In this case Eqs. (36)
and (37) become
l - b
Ap1 = 1-- P (1 - exp [-PI) exp [P(x, - xe)] (38)

=
2
- + p2
l - b exp [P(xm- xe)1{4(1 + b)(exp [-PI - 1)

+ 4P(xm- + (1 - b)(exp [-2P] - 1) exp [P(xm- xe)]

Xe)

+ @(xe - exp [-PI}

XO) (39)
as found by Ark (A8) and Bischoff (B11).
These expressions reduce even further to particularly simple forms for
the case of an infinite tube or where b = 1. Thus letting subscript 00
refer to the infinite tube, or open vessel, case we find as shown in the
third sketch of Table I1
Apim = 1 (40)
2
AuW2= -
P (41)

As far as the actual use of these expressions in their most general

form, Eqs. (36) and (37), probably the only case which need be con-
sidered is that represented by the bottom sketch of Table 11. Here meas-
urements are taken only on either side of the test section. This type of
measurement could be used to find the dispersion characteristics of a
process vessel in a plant where no means are available for inserting a
probe inside the vessel. With these equations, measurements need be
taken only on either side of the text section.
When the probes can be inserted within the test section, then Eqs. (38)
to (41) can be used. Comparing these expressions we see that it is highly
118 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

desirable to make measurements far enough away from the ends of the
vessels so that end effects become negligible, in which case the extremely
simple expressions, Eqs. (40) and (41), can properly be used. Another
even more important reason for using the infinite tube expressions is that
the end effects cannot be exactly accounted for in real systems because
of the complex flow patterns a t these locations. Thus Eqs. (38) and
(39) at best only give approximate end corrections.
However, there is one important use for Eqs. (38) and (39) : to esti-
mate the magnitude of the end effects as represented by the second term
in Eq. (39). This will let us know how far in from the ends of the vessel
the probes must be placed so that end effects can be neglected and the
simpler equation used. Bischoff and Levenspiel (B14) present design
charts which allow making such an estimation. As an example in a typi-
cal packed bed (dt/d, = 15) followed by an empty tube these charts
show that the measurement points should be placed a t least eight particle
diameters into the packed section for less than one percent error. For
more details see (B14).
d. Perfect Step Injection. For a perfect step input, the injection rate,
I, in Eq. (14) will be represented by the unit step function, U ( t ) . The
equation in dimensionless form then becomes:

As an example of the solution of Eq. (42), Lapidus and Amundson (L3)

and Robinson (R4) considered the closed-open (semi-infinite) system
and obtained the solution:

c'= F =- erfc
1-8
+ #L/DL erfc
2d(D~/uL)0
+ (43)
~~(DL/UL)~
]
Since the step response ( F curve) is the time integral of the pulse
response ( C curve), the Laplace transform of the step response for the
general boundary conditions as used by van der Laan will be given by
dividing equation (27) by the transform variable, p ,
zm'[Eq. (27)]
ELrtep = (44)
P
Thus, many other solutions could be generated from Eq. (44) if the in-
verse transforms could be found. As mentioned previously, however, this
would be exceedingly difficult in most cases.
The usual method of relating the dispersion coefficient to the experi-
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 119

mental response has been to compare the theoretical and experimental

slopes a t 8 = 1. This immediately leads to some problems, however. I n
order to have a formula for the slope, the solution in the time domain of
Eq. (42) must be available. Since this solution for general boundary
conditions is not known, and estimation of end effect error is not possible.
Taking the derivative of experimental data is usually very inaccurate,
leading to further errors. Also, only a small part of the experimental
information, that around 6 = 1, is used, It is not an “efficient” method.
A better data analysis method, as yet unused, might be to use the
area under the step response curve. It can be shown that the moments are
related to the step response by

for a “down-step” or purge run, and

for an “up-step” or feed run. Thus the entire range of data could be
used, as is done in the pulse method. The general expressions for the mo-
ments as given by Eqs. (28) and (29) can again be used for calculating
the dispersion coefficients. Hyman and Corson (H17) have devised an on
stream analog computer to perform these calculations.
e. Imperfect Step Injection. In principle, by injecting an imperfect
step function and then measuring twice, the dispersion coefficient could
be calculated. This is so because when solving Eqs. (30), (31), (32), and
(33) no assumptions were made about the form of the response a t the
first measuring point. Thus, C,’(e) could have any form, including im-
perfect pulses and steps. This is also obvious from Eq. (34) which is
merely the transfer function, or the ratio of the Laplace transforms of
the concentrations a t the two measurement points. The method of using
slopes could not be used, though, because the time domain rather than
the Laplace transformed solution would be necessary.
However, the moment method could again be used to give differences
in the moments. This information could then be used with Eqs. (36)
and (37) to find the dispersion coefficient.
f. Comparison of Pulse and Step Methods. A brief summary of the
advantages of the pulse and step injections methods is given below.
(1) Experimental Problems. It is usually easier to approximate a
perfect step than a perfect pulse. However, this is not important when
two measurements are taken. Also, the imperfect pulse might be prefer-
120 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

able for commercial testing since the system would not be disturbed
for as long a time as with the step input.
(2) Characterization of Data. The use of the moments for either
pulse or step data would seem to be best since the data is integrated and,
thus, a smoothed “mean square’’ fit is obtained. Other methods use only
the data from a narrow region such as at the pulse maximum or the slope
of the step response a t a given location.
(3) Computation Errors in Analyzing the Data. Probably the most
serious drawback in using moments is that the tail end of the recorded
curve plays too strong a role and slight errors here are magnified. How-
ever, taking the slope of the step response data is usually a rather inac-
curate process also.
(4) Comparison W i t h Model. Use of moments enables the general
expressions with any desired boundary conditions to be utilized. Other
methods require time domain solutions to be known which can be found
only for limited circumstances.
g . Periodic Injection. The sinusoidal input method has been discussed
by Rosen and Winsche (R5), Kramers and Alberda (K14), Deissler and
Wilhelm (D13), Ebach and White ( E l ) , and McHenry and Wilhelm
0 4 4 )*
The output may be found from Eq. (14) by assuming a periodic solu-
tion of the form
A (x)e -id
where w is a dimensionless frequency, o = wL/u. A simpler and more
general method would be to recall from Laplace transform theory that
the response to a sinusoidal forcing function for a system can be found
by replacing the transform variable, p , by +io. (This method is used
very widely in automatic-control theory, where the frequency response
is found by replacing the transform variable s in the transfer function
(impulse response) by S i w .) With this substitution, the amplitude ratio
can be found from the absolute value of the complex transfer function,
and the phase lag from the argument of the complex transfer function.
Thus the response to a periodic injection for very general boundary
conditions can be found by substituting p = +iw into Eq. (34). The
results for the general case would be very complicated; so, as an illustra-
tion of the form of the periodic response, we will consider only the
simplest case: a doubly infinite system. For such a system, DLal = DL’
= DL{,and Eq. (34) reduces to

The transform variable, p , appears only in q,

 PATTERNS O F FLOW IN CHEMICAL PROCESS VESSELS

1
= ZPl +i
ZP2

Therefore, the response is given by

The amplitude ratio is

and the phase lag is

These equations were used, for example, by Ebach and White (El). The
periodic response for more general boundary conditions could be found
from Eq. (34) by the same method, but the results would of course be
much more complicated.
h. Experimental Findings. Literature data on axial dispersion of
numerous investigators are presented in Figs. 8, 9, and 10. One of the
problems in bringing the data together in a generalized manner is the
choice of a characteristic length which will account for both particle
20

10

s'
$1.0

0.1
0

FIQ.8. Axial dispersion in packed beds, dispersed plug flow model (L13).
122 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

10

1.0

dtuplM

FIG.9. Axial dispersion in pipes, dispersed plug flow model (L13).

size and tube size, and allow for comparison of packed bed data with
empty tube data. For this purpose the hydraulic diameter used by Mott
(M12), Wilhelm (WlO), and Cairns and Prausnitz (C5) was used. This
length is defined similarly to the hydraulic diameter as used for friction
factors in pipes (B10) :
4(free volume of fluid)
d, =
wetted area
or

We notice that, for empty tubes, L = 1 and d , = dt. Therefore d, retains

its significance for empty tubes. The results for each type of system will
now be discussed.
i. Packed Beds. The use of the effective diameter, Eq. (50), requires
the specification of the particle diameter, dp. It was found that for non-
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 123

spherical packing the equivalent spherical diameter as given by Carrnan

(ClO), which is the diameter of a sphere with the same volume as the
particle, could be used. This gave a simple parameter that correlated all
of the data relatively well.
The experimental data was obtained by the investigators through use
of several of the above measuring techniques. Data on liquid systems
have been obtained using pulse inputs with a single measurement point
by Carberry and Bretton (C9) and Ebach and White ( E l ) . No investi-
gators have yet used the method of two measurement points discussed in
Section 11, C,2,c. Step inputs have been used by Ampilogov, Kharin, and
Kurochkina (A4), Cairns and Prausnitz (C5), Danckwerts (D4), Jacques
and Vermeulen ( J l ) , Klinkenberg and Sjenitzer (K11) and von Rosen-

1 0 Sc= 500 (ref. ~ 1 5 ) \ \ Sc = 1000(liquid)

I x a= o,e(.ef,~ 1 9 ) \, SC= 5OO(liquid)
- p.
-
\/
// \p/ x
S / / \
'- 1.0;- / / \
\ X
--
/' /'
0
' X)
/ \-
\
SC= 1(gos) I
/

' ,,
'..
/
\
, ,
- _ _,
*

0.005 0.001 0.1 1.0 10 50 100

dtu/v
. Axial dispersion in laminar flow in pipes, dispersed plug flow model.
F I ~ 10.
Adapted from (B13).

berg (V7). Frequency response methods have been used by Ebach and
White ( E l ) , Liles and Geankoplis (L18), Kramers and Alberda (K14),
and Strang and Geankoplis (S24).
Gas data have been obtained using pulse inputs by Carberry and
Bretton (C9), step inputs by Blackwell e t al. (B17) and Robinson (R4),
and frequency response by MeHenry and Wilhelm (M4).
Inspection of Fig. 8 shows that there is considerable scatter in the
data. Part of this may be due to the fact that we are attempting to repre-
sent a complex phenomenon with a single parameter, the dispersion co-
efficient. Errors would also be caused by the common practice of taking
measurements a t or beyond the exit of the packed section. This neglect
of end conditions could lead to large errors in the calculated dispersion
coefficients, as pointed out by Bischoff and Levenspiel (B14). Also, all
the analyses were based on the assumption of having a perfect pulse, step,
124 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

or sinusoidal input. These assumptions can now be relaxed with the two-
measuring-point method. Thus, even though there is much reported data,
it would seem that further experimental work is needed in which all
possible experimental errors are minimized.
The length parameter used in this correlation to represent the different
types of packing may also contribute to the scatter. The use of the ef-
fective diameter may not be the best way to bring the data together,
and several of the investigators found that they could correlate their
own data quite well by using certain arbitrary functions of the void
fraction, C. However, i t was found by Bishchoff (B13) that the effective
diameter was the most satisfactory size parameter when all of the data
were considered.
A further problem in plotting the data is that the fluid viscosity has
very little effect on the dispersion ( E l ) . Therefore, the Reynolds num-
ber might not be appropriate as a plotting variable (C5). However, no
other dimensionless group has yet been proposed, and so the Reynolds
number was retained in Fig. 8.
Figure 8 shows that the group DL'/ud, is roughly constant for all
Reynolds numbers. Jacques and Vermeulen (Jl) found, however, that
their data with regular arrangements of the particles showed a definite
break in the values. This seems to be caused by a transition from laminar
to turbulent flow.
Rough estimates for axial dispersion coefficients can be made using
random walk techniques, and these will be discussed in Section I1,E.
Also, a theory can be developed for predicting axial dispersion coefficients
from radial dispersion coefficients which is the source of the dotted line
of Figure 8. This will be discussed in Section II,D. Bischoff (B13), Fro-
ment (F9), and Hofmann (H11) have presented summaries of packed-
bed data.
j . Turbulent Flow in Empty Tubes. Data for axial dispersion co-
efficients in turbulent flow in empty tubes has been collected by Leven-
spiel (L10) and Sjenitzer (Sl8). Their results are combined in Fig. 9.
The liquid data have been obtained using single measurements of a pulse
input by Allen and Taylor (A3a), Hull and Kent (H15), Kohl and
Newacheck (K13), Lee (L7), Shipley (S15) and Taylor (T4). A step
input was used by Fowler and Brown (F7), Smith and Schultz (S19),
and no one has used a frequency response method. Gas data was obtained
by Davidson et al. (D11) using a pulse technique.
Figure 9 shows considerable scatter because much of the data were
taken in commercial pipelines with valves, elbows, bends, and other
types of flow disturbances. In the transition region of Reynolds num-
bers below 10,000, the dispersion coefficient rises rapidly. This can be
 PATTERNS O F FLOW I N CHEMICAL PROCESS VESSELS 125

explained by the build-up of the laminar sublayer at these low Reynolds

numbers with its pronounced effect on the velocity distribution. The
dashed curves are based on a theory to be discussed in Section I1,D.
k. Laminar Flow in Empty Tubes. All of the data have been taken
by pulse methods with a single measurement point. The liquid data are
by Blackwell (B15) and Taylor (T2), and the gas data by Bournia et al.
(B19). The dashed lines are again from theory. It is seen that the liquid
data agree quite well with the theory. The gas data, however, do not.
Bournia et al. discussed this and arrived a t the conclusion that the axial-
dispersed plug-flow model may not be a good representation of the sys-
tem for this case. Additional experimentation in gases is suggested.
Crookewit, Honig, and Kramers (C23) have given results for flow in an
annular region.

3. Dispersed Plug-Flow Model

a. Preliminary. The experimental methods for measuring radial dis-
persion coefficients again involve injecting tracer into the system and
measuring its concentration a t a point downstream from the injection
point. In this case, we do not want to inject over a plane or measure over
a plane. Instead we must use an experimental method where the con-
centration varies with radial position allowing us to study the radial
movement of tracer and thereby giving information on the radial mixing
occurring in the system. Usually the tracer is injected on the axis of the
tube. The axis is chosen so that there will be radial symmetry about the
tube axis, which simplifies the mathematics. As the tracer moves down
the tube, i t will spread radially by dispersion, until a long distance down-
stream (theoretically infinite) from the injection point the tracer will be
completely mixed with the flowing fluid. Therefore, the measurement
point must not be too far from the injection point in order to have enough
of a concentration profile to get accurate results. The methods for meas-
uring axial dispersion coefficients all required unsteady-state injection
methods. Now, however, because we are going to inject a t a point rather
than over a plane and measure radial concentration profiles, a steady-
state method can be used, which simplifies the analysis. Unsteady state
methods could also be used, but they would just add unnecessary compli-
cations.
The mathematical developments are based on Eq. (13) with all of the
coefficients assumed to be constant,
ac -I- U -
- ac a2c + D~ a ac + -I 6(X - Xo)f(R)
at dX = DL-
ax2 --R aR R -
aR ?r (51)
We shall change the variables in Eq. (51) to dimensionless forms which
126 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

are different from those used in the previous section on the axial-dis-
persed plug-flow model. Thus

A dimensionless concentration c is defined as follows

where Caveis the mean concentration of tracer far enough downstream

from the point of steady-state tracer injection of rate I. By a mass bal-
ance, then,
(rate of injection) = (rate of flow out of tube)
or
I = rR&&*,e
With these dimensionless quantities introduced, Eq. (51) becomes

This equation is used in finding the dispersion coefficients from experi-

mental data.
b. Solutions for Various Measurement Experiments. Most of the meas-
urement techniques have used the same sort of injection: steady flow a t
the tube axis. Thus the differences in the various contributions were in
the exact experimental setup used (which determines the boundary con-
ditions), and in the completeness of the mathematical model used t o
represent the physical situation. Various simplifying assumptions were
usually made in Eq. (52) or in the boundary conditions. These are shown
in Table 111. Bischoff and Levenspiel (B14) presented a general solution
that used no assumptions beyond those inherent in Eq. (52). We will
discuss this solution first, since all other cases are special cases of this
general solution.
The method of solution is similar to that used for the axial-dispersed
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 127

TABLE I11
EXPERIMENTAL
SCHEMEUSEDIN RELATION
M THE
DISPERSED
PLUG-FLOWMODEL

Where first
Experimental scheme Results used

Touir and
Shrruood bd

Restriction D,=D,

Bernard and
Wilhrim [&I

7 /
Kllnkrnbrr
et a1
I Tmcr in M*awnmml
point

Restriction DL*O

Fohlen and
Smith
rlth J , ( a , ) =O
J

x!k+!y2ixm
Equatlon (55 1 and (56 1

Equation (55 and ( 5 7 1

Tracrr In
Marunm*nl
polnt

plug-flow model. In the present case, however, the method must be modi-
fied in order to keep both axial and radial dispersion in the equations.
It will be recalled that the experimental equipment is supposed to con-
sist of three parts: the upstream section X = --oo to X = 0, the test sec-
tion X = 0 to X = X,, and the downstream section X = X , to X = 00. +
The injection takes place a t Xo(O 5 X o 5 2,) and the measurement is
a t X,(Xo < X,) ; see Table 111.
128 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

Equation (52) is written for each section, and thus the boundary
value problem to be solved is:

-- +-- a- r -
PLa2 P R r a r ar
--
ax
= -6(x - xo)f(r) 0 5 x 5 xe (53b)

with boundary conditions :

ca(- co, r ) = finite

c,(O-, r ) = c(O+, r )
1 ac, 1 ac
ca(o-9 - p~. (0-, r ) = c(O+, r ) -- -
pLax (O+, (544
1 ac 1 aca
c(x,-, r ) - -- (xe-, r )
pLax
= cb(xe+,r ) - --
pLbax
be+,
r) (54e)

Equation (54a) means physically that no mass is transferred through

the solid pipe wall, and all of the other boundary conditions have been
discussed previously. The solution evaluated a t the measurement point
X = X , is:

where
Jl(Ui) = 0

For measurement within the test section (x, 5 xe), also

 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 129

The solutions of the various investigators as listed in Table I11 can be

found by suitable simplification of Eq. ( 5 5 ) as shown in (B14).
Eq. (55) is impractical to use. Hence it would be valuable to know
under what conditions we are justified in using some of the simpler ex-
pressions based on various assumptions. This information can then be
used to guide the design of experimental equipment so that the mathe-
matical analysis can be made relatively easily.
Bischoff and Levenspiel (B14) considered this problem, and have pre-
sented design charts which allow estimation of errors in the calculated
dispersion coefficients for various conditions. It was found that when the
ratio of injection to tube diameter is less than 20%, or e < 0.2, then the
assumption of a point source, or e -+ 0, was good to within 5%. Thus
for many cases, the neglect of the finite size of injection tube is justified.
It was also found that the end-effect error is smaller than for the
measurement of axial dispersion coefficients. Even though the error may
be large for measurement right a t the end of the vessel, the error de-
creases very rapidly on moving the probe into the bed. Hence, in most
typical packed beds, taking measurements one or two particle diameters
into the bed is often sufficient to make the end-effect error negligible.
c. Use of Equations in Interpreting Experimental Data. The experi-
mental setup corresponding to the mathematical model used [Eqs. (53)
and (54)] would be as follows. A steady stream of tracer is injected from
a tube on the axis of the cylindrical vessel. Then a t some point down-
stream from the injection point the tracer concentration is measured, and
Eq. ( 5 5 ) is used to find the dispersion coefficients from the concentration
data. A practical problem that arises in the tracer concentration meas-
urement is that the recording instrument usually has to be calibrated to
give the proper numerical values of concentration. I n most instruments,
recalibration is needed from time to time, which is not very convenient.
For this reason, many investigators (B6, D20, F2) have measured the
experimental average stream concentration, CA, at a point far down
stream, in addition to finding the radial‘ concentration distribution a t the
measurement point, X,. Then if the ratio C / C , is used, any “drifting”
in the measuring device will automatically cancel. Therefore, concentra-
tion measurements are usually reported as C / C A (B6).
If our experimental techniques were perfect, the experimental average
concentration C A would be equal to the integral mean concentration C,,,.
130 OCTAVP: LEVENSPIEL AND XENNETH B. BISCHOFF

The extent that they differ from one another can be taken as a “mass
balance” for the system. If Eq. (55) is multiplied through by Cave/CA
we get:

In order to have an estimate of the experimental precision, it is neces-

sary to take several concentration readings. This is most conveniently
done a t a number of radial positions for a given axial position in the
vessel. Suppose we write Eq. (58) as:

where
+
C/CA = KO KJo(air) + KsJo(w) + *
- (59)

Now if we have data of C/CA as a function of pipe radius, r, we can

use standard least-squares techniques to estimate KO, K 1 , Kz, * . * . I n
addition, we can find the standard deviations of the estimates of Kt by
the least-squares procedure, which gives an indication of the precision of
the data. The first constant, KO,should be unity if we have a perfect
mass balance, and the deviation from this value gives an estimate of the
reliability of the data. Knowing the injection tube size, we can find the
Nd/q from the least squares Kt from Eq. (59).

The dispersion coefficients can now be found from the N 6 / q by a non-

linear calculation procedure, such as the Newton-Raphson method, utiliz-
ing the expressions for N,, Eqs. (56) or (57). In the general case, values of
P L 1 , P R l , P L ~P,R 2 , and the physical dimensions of the apparatus are
substituted into Eq. (56) or (57), and then PL and PR can be found from
the simultaneous (nonlinear) solution of the expressions for N 1 and N z .
The variances of the dispersion coefficients could also be found from the
variances of the K6 by standard statistical methods.
Another method has been proposed by Blackwell (B16) and by Hiby
and Schiimmer (H8) that avoids the necessity of measuring the complete
concentration profile. A pipe with a diameter smaller than the system,
thus forming an annular region, is used a t the sampling point. A mixed
mean sample from the annular region is now sufficient to enable one to
determine the radial dispersion coefficient. From Eq. (55) this concentra-
tion will be, for an annular region of dimensionless radius (Y,
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 131

This is the value for general boundary conditions, and it will reduce to
the equations given by the above authors for each of their special cases.
d. Experimental Findings. Literature data on radial dispersion of
numerous investigators is presented in Figs. 11 and 12. The data of almost

d,u/r
F I ~ 11.
. Radial dispersion in packed beds, dispersed plug flow model. Adapted
from (B13).

all the workers were obtained by a steady "point" source injection of

tracer on the axis of the tube, with subsequent use of Eq. (59), (61),
(56), and (57) to calculate the dispersion coefficients. We will discuss the
results for each type of system separately.
0.01

2-
\

D -- -_ ----_
0.001

O.OOC5
5x105 o4 105
dtu/u

FIQ.12. Radial dispersion in pipes, dispersed plug flow model (B13).

132 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

(1) Packed Beds. Data on liquid systems using a steady point source
of tracer and measurement of a concentration profile have been obtained
by Bernard and Wilhelm (B6), Jacques and Vermeulen ( J l ) , Latinen
(L4), and Prausnitz (P9). Blackwell (B16) used the method of sampling
from an annular region with the use of Eq. (62). Hartman et al. (H6)
used a bed of ion-exchange resin through which a solution of one kind
of ion flowed and another was steadily injected a t a point source. After
steady state conditions were attained, the flows were stopped and the
total amount of injected ion determined. The radial dispersion coefficients
can be determined from this information without having to measure de-
tailed concentration profiles.
Data on gas systems, again using a point source, have been obtained
by Bernard and Wilhelm (B6), Dorweiler and Fahien (D20), Fahien and
Smith (F2), and Plautz and Johnstone (P6). Plautz and Johnstone meas-
ured dispersion coefficients under isothermal and nonisothermal condi-
tions and found that there was a difference between the two only for
low Reynolds numbers.
The data were plotted, as shown in Fig. 11, using the effective diame-
ter of Eq. (50) as the characteristic length. For fully turbulent flow,
the liquid and gas data join, although the two types of systems differ
a t lower Reynolds numbers. Rough estimates of radial dispersion coef-
ficients from a random-walk theory to be discussed later also agree with
the experimental data. There is not as much scatter in the data as there
was with the axial data. This is probably partly due to the fact that a
steady flow of tracer is quite easy to obtain experimentally, and so there
were no gross injection difficulties as were present with the inputs used
for axial dispersion coefficient measurement. I n addition, end-effect er-
rors are much smaller for radial measurements (B14). Thus, more experi-
mentation needs to be done mainly in the range of low flow rates.
(2) Turbulent Flow in Empty Tubes. Figure 12 gives radial dispersion
coefficient data for turbulent flow in empty tubes. The liquid data was
obtained by Flint et al. (F5). Gas data is from Baldwin and Walsh
( B l ) , Flint et al. (F5),Longwell and Weiss (L19), Mickelson ( M l l ) ,
Schlinger and Sage (SS), and Towle and Sherwood (T9).All of the
above data was obtained by measuring concentration profiles in systems
with a steady point source of tracer. Keyes (K4a) used a frequency-
response experiment and interpreted his results in terms of a film thick-
ness a t the wall. Sherwood and Woertz (S14) and Dhanak (D19) have
reported results for flow in rectangular channels.
The radial dispersion coefficient for this case is, of course, the average
eddy diffusivity as discussed in works on turbulence (H9). If the vari-
ous analogies between momentum, heat, and mass transport are used,
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 133

estimates of the eddy diffusivity can be obtained. However, we only

show the directly measured mass-diffusion data in Fig. 12. There is a
fair amount of scatter in the data which is probably caused by two
main factors. Some of the investigators used large cumbersome injection
systems which might have disturbed the flow, and some of the equip-
ment might not have had an upstream section long enough to insure
fully developed flow.
(3) Laminar Flow in E m p t y Tubes. As will be discussed in Section 11,
DR, the radial coefficient for the dispersed-plug flow model for laminar
flow is merely the molecular diffusivity.
4. General Dispersion Model for Symmetrical Pipe Flow
a. Discussion. As was stated previously, the general dispersion model
is much more difficult to use since analytical solutions are not possible.
Very little work has been done with these models, and it is entirely
possible that the extra accuracy gained with this more complex model
does not justify the great increase in the attendant computational dif-
ficulties. The only answer to this will be to test whether the simpler
dispersion models are sufficient to characterize processing systems.
b. Experimental Findings. For turbulent flow in empty tubes, the
problem is one of finding the radial variation of eddy diffusivity, which
is discussed, for example, by Hinze (H9). Lynn et al. (L20) have pre-
sented data for air natural-gas systems, They utilized a relation similar
to Eq. (13), but for a compressible fluid. We will illustrate their method
with the steady-state form of the simpler Eq. (1-2) with no axial disper-
sion,
ac a
Ru(R)- = - RDR(R)
ac
ax aR
If this equation is integrated once with respect to R and rearranged,

The quantities in the numerator and denominator were directly calcu-

lated from the experimental concentration profile data. Some results for
DR(R) are shown in Fig. 13. It is seen that the eddy diffusivity is rela-
tively constant except near the walls. The values seem quite sensitive to
the axial position where the concentration profile was measured; in other
words, the system is not axially homogeneous. These results have been
134 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

discussed in more detail in a previous review by Opfell and Sage (01).

Carrier (C11) has considered the case of a periodic input for laminar
flow in an empty tube.
Fahien and Smith (3'2) and Dorweiler and Fahien (D20) have con-
sidered the variation of D R in packed beds, using a separation-of-vari-
ables technique to solve Eq. (63). The X-dependent part was solved
analytically, and a set of difference equations was used to solve the
R-dependent part. DetaiIs are given in (F1). The velocity profile data
of Schwartz and Smith ( S l l ) was used to calculate values of DR(R) in
the packed column,. typical results from Fahien and Smith (F2) being
shown in Fig. 14. Dorweiler and Fahien's (D20) data, for a lower Rey-

0.030
6.9
OD26 0.08- dt/dp: 11.1
-0: -5 0.06-
13.3

-
= 25.6

'
0.022
0"
0018 0.04
0.014 0.02 ,-+-
0.2 0.4 06 0.8 0 0.25 0.50 0.75 1.00
Rodiol pattion r Rodiol position r

FIG.13. Radial dispersion in pipes, FIG.14. Radial dispersion in packed

general dispersion model (01). beds, general dispersion model. Adapted
from (F2).
nolds' number range, was similar. It is seen that for a sufficiently large
ratio of tube-to-particle diameter, DR(R) is approximately constant
over the radius. These findings suggest that if the dispersed plug-flow
model does not satisfactorily represent the system and a more accurate
model with varying velocity profile is needed, then the uniform dispersion
model could be used.

1. Preliminary
We have discussed methods for experimentally finding dispersion co-
efficients for the various classes of dispersion models. Although the models
were treated completely separately, there are interrelations between them
such that the simpler plug-flow models may be derived from the more
complicated general models. Naturally, we would like to use the simplest
possible model whenever possible. Conditions will be developed here for
determining when it is justifiable to use a simpler plug-flow model rather
than the more cumbersome general model.
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 135

2. Theoretical Deviations
Taylor (T2) and Westhaver (W5, W6, W7) have discussed the rela-
tionship between dispersion models. For laminar flow in round empty
tubes, they showed that dispersion due to molecular diffusion and radial
velocity variations may be represented by flow with a flat velocity pro-
file equal to the actual mean velocity, u, and with an effective axial dis-
persion coefficient DL’= Ro2~2/48a).3 However, in the analysis, Taylor
ignored axial diffusion. Aris (A6) later showed that the true axial effect
+
is additive and thus, more correctly, DL’ = a) Ro2u2/48D. Use of DL’
gives the same results as would be obtained from the more rigorous calcu-
lation involving radial and axial diffusion and true velocity profile, Eq.
(1-2). Aris (A6) generalized the entire treatment to include all types of
velocity distribution with any vessel geometry. He showed that the co-
efficient given as 1/48 by Taylor is really a function of tube shape and
velocity profile.
The mathematical method of Aris assumes a doubly infinite pipe (as
does Taylor), with both the velocity distribution and the diffusion co-
efficients constant in the direction of flow. Hence in any real pipe, the
length would have to be long enough so that the buildup of the velocity
profile a t the entrance would not invalidate the doubly infinite pipe as-
sumption. Thus there are some practical restrictions on the method used
by Aris.
Taylor (T4, T6), in two other articles, used the dispersed plug-flow
model for turbulent flow, and Aris’s treatment also included this case.
Taylor and Aris both conclude that an effective axial-dispersion coef-
ficient DL’ can again be used and that this coefficient is now a function of
the well known Fanning friction factor. Tichacek et a2. (T8) also con-
sidered turbulent flow, and found that DL’ was quite sensitive to varia-
tions in the velocity profile. Aris further used the method for dispersion
in a two-phase system with transfer between phases ( A l l ) , for disper-
sion in flow through a tube with stagnant pockets (AlO), and for flow
with a pulsating velocity (A12). Hawthorn (H7) considered the tempera-
ture effect of viscosity on dispersion coefficients; he found that they can
be altered by a factor of two in laminar flow, but that there is little effect
for fully developed turbulent flow. Elder (E4) has considered open-chan-
nel flow and diffusion of discrete particles. Bischoff and Levenspiel (B14)
extended Aris’s theory to include a linear rate process, and used the
results to construct comprehensive correlations of dispersion coefficients.
*Actually, Taylor originally suggested using this formula in reverse for obtaining
diffusion coefficients. Dt’ could be found simply from experimental data and then
the formula could be used to obtain the diffusivity, 5).
136 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

The most straightforward way of comparing the models would be

to solve the partial differential equations for each case and compare the
solutions. Because we cannot find these solutions in general, we must be
satisfied with a comparison of the moments of the concentration distribu-
tions given by these models. This method is much used in theoretical
statistics, and is useful since most distributions encountered are uniquely
determined by their moments.
If we consider the variation of concentration along the axial direc-
tion as a “distribution” of concentration, we can calculate the moments
for each model in terms of their respective parameters, and then compare
the moments to find the relationship between parameters. Since we shall
consider the concentration as a distribution function along the axis of
the tube, the moments are with respect to axial distance, rather than
with respect to time as used previousIy. Since flow in cylindrical vessels
is so common, we will discuss only this case in detail. Aris (A6) gives
the more general treatment in vessels of arbitrary cross section.
The moments are defined as:

where subscript Ic indicates the moment order. Use of the variable (x - 8‘)
relates the moments to a coordinate system that is moving with the mean
speed of the flowing fluid. By using Eq. (65) with the mathematical repre-
sentations of the various models, Eqs. (I-2), (I-3), (1-4), and (I-5), one
obtains the following expressions for the moments. For details see Aris
(A6) and Bischoff and Levenspiel (B14).
Axial-Dispersed Plug-Flow Model
Mo = 1
MI = 0
28’
Mz = -
Pt’
Dispersed Plug-Flow Model
Mo = 1 (674
MI = 0 (67b)

Mz =
28’
-
PL
+ O(exp [ -X18’/PR]) (67c)

For large values of time (the “ultimate” value) Mz reduces to

 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 137

Uniform Dispersion Model

Mo = 1
MI = constant
Mz = I- 28'
PLm
+ +
2 8 ' h P ~ ~ ' O(exp [ - h e ' / P ~ ] ) (684
For large values of time M z becomes

General Dispersion Model

Mo = 1
MI = constant
28'
+ +
M z = - 28'h P R ~ O(exp [-A18'/PB,])
PLrn
(69c)

For large values of time M z becomes

I n these equations,

and fl, fz, f3, Darn, and DLm are found from u ( R ) = u ( 1 f l ( R ) ) ,+
D R ( R )= D R J z ( R ) , and D L ( R ) = D h f 3 ( R ) . Also, hl is the smallest
of eigenvalues that arise during the solution of the partial differential
equations for the moments. The various models are now related to each
other by comparing the ultimate moments.
Zeroth Moments. The zeroth moment for each of the models is unity.
From Eq. (65) the zeroth moment is defined as:

This is merely the volume integral of the concentration over the entire
vessel. Therefore, the value of unity indicates that the total amount of
solute in the system is constant.
First Moments. For both of the dispersed plug-flow cases M I = 0.
This means that the center of gravity of the solute moves with the mean
speed of the flowing fluid. For the uniform and the general dispersion
models, however, this is not always true. If the solute concentration is
initially uniform over a cross-sectional plane, i t can be shown (A6) that
138 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

the constants in Eqs. (68b) and (69b) are zero; thus, for this special
case, the center of gravity of solute does move with the mean speed of
flow. If the initial solute concentration is not uniform over a plane,
however, the constants in Equations (68b) and (69b) are not zero,
and the center of gravity of solute does not move with the mean speed
of flow. However, if the initial distribution of solute is fairly uniform
over the cross section, as is true in many practical cases, the constants
in Eqs. (68b) and (69b) are small, and the first moments of the various
models will all be approximately zero.
Second Moment. By equating the second moments of the different
models we get the relationships sought between the parameters.
First, comparing Eqs. (66c) and (67d), we see that
P' = PL or DL' = DL (72)
Thus we may drop the primed notation on the coefficient for the axinl-
dispersed plug-flow model and identify this coefficient with the one for
the dispersed plug-flow model.
Next we compare the dispersed plug-flow model with the general
model by equating Eqs. (67d) and (69d). Thus, as found by Aris (A6),

This equation enables us to calculate the value of PL from the velocity

profile using mean values of the coefficients of the general dispersion
model. The constant radial coefficient used in the dispersed plug-flow
model is the same as the mean value of the varying radial coefficient in
the general dispersion model.
Finally by comparing Eqs. (68d) and (69d), we see that the constant
coefficients of the uniform model are the same as the mean values in
the general model, or
DRm' = DRm (74)
and
D L ~=' D h (75)
Summary. For all models with either constant or varying velocities
we conclude that the constant radial coefficients are all alike and equal to
the mean value of the varying coefficient, or

In addition, two distinct axial coefficients exist: those associated with

models using mean velocities (plug-flow models) and those associated
with models using a radial velocity profile. The reason for this difference
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 139

is that the plug-flow models incorporate in their dispersion coefficients

the axial mixing resulting from velocity variations. The general model
and the uniform model treat in separate terms the two phenomena
responsible for axial dispersion. The equations (72) to (76) summarize
the interrelationships between the parameters of the various dispersion
models treated in Table I.
The concentration distributions of the dispersed plug-flow models a t
any time are “normal” (Gaussian), and Aris (A6) has shown that the
general model with its larger number of parameters also approaches
normality for large time. Hence, matching the first two moments is
sufficient to compare distributions and relate parameters as was done
above. However, the approach to normality in the general model is slow.
Consequently, minor deviations between distributions will persist for
some time even though the distributions are essentially similar.

3. Experimental Verification of the Relationships

Bischoff and Levenspiel (B14) present some calculations using exist-
ing experimental data to check the above predictions about the radial
coefficients. For turbulent flow in empty tubes, the data of Lynn et al.
(L20) were numerically averaged across the tube, and fair agreement
found with the data of Fig. 12. The same was done for the packed-bed
data of Dorweiler and Fahien (D20) using velocity profile data of
Schwartz and Smith (Sll), and then comparing with Fig. 11. Unfortu-
nately, the scatter in the data precluded an accurate check of the predic-
tions. In order to prove the relationships conclusively, more precise ex-
perimental work would be needed. Probably the best type of system for
this would be one in laminar flow, since the radial and axial coefficients
for the general dispersion model are definitely known; each is the
molecular diffusivity.
Checks on the relationships between the axial coefficients were
provided in empty tubes with laminar flow by Taylor (T2), Blackwell
(B15), Bournia et al. (B19), and van Deemter, Broeder and Lauwerier
(V3), and for turbulent flow by Taylor (T4) and Tichacek et al. (T8).
The agreement of experiment and theory in all of these cases was satis-
factory, except for the data of Bournia et al.; as discussed previously,
their data indicated that the simple axial-dispersed plug-flow treatment
may not be valid for laminar flow of gases. Tichacek et al. found that the
theoretical calculations were extremely sensitive to the velocity profile.
Converse (C20), and Bischoff and Levenspiel (B14) showed that rough
agreement was also obtained in packed beds. Here, of course, the theo-
retical calculation was very approximate because of the scatter in packed-
bed velocity-profile data.
140 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

4. Uses of t h e Relationships
The most important use of the relationship among models is in
showing t h a t the dispersed plug-flow models are good representations
of the flow system under certain specified conditions. The relationships
can also be used to predict the dispersed plug-flow coefficients from the
general coefficients. We have shown how this is done for flow in empty
tubes and the predictions are given by the dashed lines on Figs. 8, 10,
and 12.

lo-‘ 10 10’ 10’

id!

FIG.15. Comprehensive picture of dispersion-empty tubes (B14).

For packed beds, the use of these equations for predictions is limited
by inaccuracy in the velocity-profile data. Therefore, Bischoff and
Levenspiel (B14) used the equations in a semiempirical way for inter-
polating between the existing data. The results are shown in Figs. 15
and 16, for both empty tubes and packed beds. The heavy lines show
the regions of experimental data, and the dashed lines, the interpolations.
For sufficiently low flow rates, the curves lead into the reciprocal Schmidt
number (modified by a “tortuosity factor” in packed beds). The data
of Blackwell (B16, B17) a t very low flow rates seems t o verify this.
At high flow rates, liquids and gases show no differences because of the
 PATTERNS O F FLOW I N CHEMICAL PROCESS VESSELS 141

-
deu
I

FIG.16. Comprehensive picture of dispersion-packed beds (B14).

turbulent flow conditions. Figures 15 and 16 give a rather comprehensive

picture of dispersion, and point out the areas in which there is a scarcity
of experimental data.
5 . Limitations of the Theory
When we used the "ultimate" values of the moments in Eqs. (67d),
(68d), and (69d), this amounted to assuming that (B14),
O(exp [ - xIe'/PRml)<1 (77)
If a ratio of 1 O : l is used in the inequality, and if an estimate is used
for All the limitation is,
DtRm
> 0.16
Ro2
If the mean residence time of the fluid in the system, L/u, is used as a
measure of the time, we finally obtain,
L > 0.04 ud
- (79)
dt DRm

This is approximately the same as the criterion originally used by Taylor

(T5) in his treatment of laminar flow. Using the values of DRmfrom the
142 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

preceeding correlations, the order of magnitude of the necessary length-

to-diameter ratio needed for the dispersed plug-flow models to be valid in
various types of systems is shown in Fig. 17. This is, of course, only an
order of magnitude estimate and should not be relied upon for accurate
values. It should be used only to give some sort of idea as to the types
of systems where the dispersed plug flow models can be theoretically
justified.

-
do*
V

FIQ.17. Restrictions on length to diameter ratio for dispersed plug flow models
to be valid (B14).

E. THEORETICAL
METHODSFOR PREDICTING
DISPERSION
COEFFICIENTS

1. Introduction
If the theory of turbulence were complete enough, it would be possible
to use i t to predict the dispersion coefficients. Unfortunately, even for the
simple case of homogeneous isotropic turbulence, this cannot yet be done.
For cases of bounded flows in a real pipe and of flow through packed
beds, the situation is even more discouraging. Nevertheless, several ap-
proximate estimates have given surprisingly good results as to the order
of magnitude of the dispersion coefficients. For empty tube turbulence,
which is a field in itself, we refer to Hinee (H9) and to a recent article
by Roberts (R3).Here we will discuss the models for packed beds,
but only very briefly.
 PATTERNS OF FLOW I N CHEMICAL PROCESS VESSELS 143

2. Random-Walk Models
Since the randoni-walk approach is successful in molecular diffusion
(K5) and Brownian motion studies (C14), it would seem that it might
also be useful for the dispersion process. This has been considered by
Baron (B2), Ran2 (Rl), Beran (B5),Scheidegger (S6), Latinen (L4)
and more recently by de Josselin de Jong (D14) and Saffman (Sl, S2,
S3). The latter two did not strictly use random-walk since a completely
random process was not assumed. Methods based on statistical mechanics
have been proposed by Evans e t al. (E7), Prager (P8), and Scheidegger
(S7).
The sample random-walk analyses postulate that the mixing is caused
by “splitting” or side-stepping of the fluid around the particles. Thus, by
analogy with the mixing-length theories of turbulence, one might imagine
that the mass flux would be proportional to the particle diameter and
to the velocity,
D a ud, (80)
Baron (B2) has given a rather simple treatment for radial dispersion.
Let us assume that as a fluid element approaches a particle, it is de-
*
flected by an amount pd, where p is a fraction of the order of one-half.
As the fluid element flows through the packing, this process keeps
occurring. When the fluid element has travelled a distance, L, it has
been deflected n times. The fluid element is deflected essentially each
time it approaches a particle. Thus, n = aL/d,, where a is of the order
of unity. If the deflections are random, the mean-square deviation is
the sum of the squares of each deflection,
-
AX2 = np2dP2 (81)
If the Einstein equation for diffusion (H9) is used (which again assumes
random motion), the dispersion coefficient may be approximated,
ax2= 2Dt np”d,2
= (82)
where t is the diffusion time, and may be taken to be t N L/u. Thus,

If Eq. (83) is rearranged and the approximate numerical values of

(Yand p substituted,
144 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

Comparison of this with the radial dispersion data of Fig. 11 for an

average packed bed (using d,/dp N c = 0.38) shows good agreement.
Notice that Eq. (84) predicts that D / u d p should be independent of flow
rate. From Fig. 11, this is true for the larger Reynolds numbers of fully
turbulent flow. At lower Reynolds numbers, the mixing would become
less random, and so i t would be expected that the theory would break
down.
Prausnitz (P10) has devised an approximate mixing length model
for estimating the axial dispersion coefficient that allows for the inter-
action of the velocity profile. He used

(85)

where A L is an axial and A is a radial scale of turbulence. From con-

centration fluctuation studies (to be discussed later), A R 2: d,/4 and
A L CT 7 A n . If the velocity gradient is approximated by u / d p , the mixing-
length theory predicts

or
D ,-7
A
ud, - 16
which is of the right order of magnitude (see Fig. 8 ) .
Further details may be found in the above quoted references. I n
particular, de Josselin de Jong (D14) and Saffman (Sl, S2, S3) give
relatively rigorous developments that take into account the anistropy
caused by the flow. Thus different estimates are obtained for the disper-
sion coefficients depending on whether or not the direction considered is
perpendicular or parallel to the mean flow.
3. The “Statistical” Models
These models picture the mixing process as consisting of “motion
phases” and “rest phases.” In a model proposed by Einstein (E3) and
discussed by Jacques and Vermeulen (Jl) and Cairns and Prausnitz
(C5), it is assumed that the time represented by a motion phase is much
less than that by a rest phase. For the packed bed, the motion phase
might be taken as the period when the fluid element is passing through
the restriction between particles, and the rest phase as the period when
the fluid element is in the void space. If values are assigned to the
probabilities of motion or rest, consideration of the geometry of the fluid
elements’ motions will lead to
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 145

p ( X , t ) dX dt
e-x-t dX dt
= (87)
as the probability density for any “jump” of the element (see Cairns
(Cl) for details). Consideration of a large number of “jumps” then yields

ptot= e-x’-t’~o(21/X“) (88)

as the probability of finding all particles a t relative position X’ and a t
relative time t’ that entered at time t’ = 0. I n other words, Eq. (88)
gives the impulse response or C curve for the system. The relationship
between the relative position, X’, and physical position, X, depends on
the length of each step. Similarly, there is a relation between t’ and t.
Thus by comparing Eq. (88) with the solution of the axial-dispersed
plug-flow model a t very large X and t, it is found that

Thus Eq. (88) can be used to find the dispersion coefficient, and was
used by Cairns and Prausnitz (C5) with a step input which corresponds
to the time integral of Eq. (88). It would seem that the use of this
model would involve problems similar to those for the ordinary disper-
sion models.
Giddings and Eyring ( G 2 ) , Giddings (Gl) , and Klinkenberg (K10)
have also proposed a model based on similar “rest-phase”-“motion
phase” considerations. However, they used different assignments of the
probabilities.
4. Turner’s Structures
Turner (T14)has proposed two detailed models of packed beds which
try to closely approximate the true physical picture. The first model
considers channels of equal diameter and length but with stagnant
pockets of various lengths opening into the channels. There is no flow
into or out of these pockets, and all mass transfer occurs only by
molecular diffusion. The second model considers a collection of channels
of various lengths and diameters. We will briefly discuss each of these
models, which are probably more representative of consolidated porous
materials than packed unconsolidated beds.
a. Model I . It is assumed that the dispersion in the channel may be
represented by an axial-dispersed plug-flow model,
ac - ac
D L -a~w - u - -- &%
ax ax at ~
= 0

where DL is the axial dispersion coefficient in the channel, pr is the

volume of stagnant pocket of length 1, per unit channel volume, and qr
146 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

is the average concentration in the rth pocket. Turner originally took DL

to be the same as that for flow in a tube with no pockets, such as we
have previously discussed. However, Aris (A10) showed that DL is
influenced by the pockets, and so this modified value should be used.
I n the pockets, ordinary molecular diffusion equation was assumed :

where C, is the concentration in the rth pocket. The solution of Eq. (91)
is used to find dq,/dt, which is needed to solve Eq. (90).
For a periodic sinusoidal input of tracer, Turner (T14) showed how
to find the values of 2, and p, from the experimental data taken at dif-
ferent frequencies, w . Aris (A9) generalized the model by taking p(1) to
be a continuous distribution of pocket volume rather than a set of
discrete values, p r . The problem of finding p (1) then becomes one of solv-
ing the integral equation,

cc(w') = lo- P(Os(w', 0 dl (92)

where
w' = (w/2%)1"/2
~ ( 0 ' )= function determined from amplitude ratio
and phase lag of experimental data.

-1 sinh 2w'l - sin 2w'l

q(w', 1) =
2w'l cosh 20'1 + cos 2w'l
Aris presented a graphical method for the solution of Eq. (92). If the
integral were approximated by a sum, the equations used by Turner
would be generated.
In a later paper, Turner (T15) used experimental data from ap-
paratus with known p in order to see how the method would work. He
found approximate agreement between the experimental and known p's ;
quite accurate data would be needed to obtain good estimates of p.
b. Model 11. It was again assumed that the dispersion in any one
channel could be represented by the axial-dispersed plug-flow model,
Eq. (90) with p, = 0, since there were no pockets. He defined eg,. as that
part of the void fraction contributed by the channel of length 1, and
radius rq. Then, by a procedure similar to that used for Model I, a set
of equations was developed for a sinusoidal input of tracer. For the
measurement a t the frequency wj,
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 147

where
pj = function determined from amplitude ratio
and phase angle of experimental data
qqrj = (zy [7
exp MZ? (1 - d 1 + dl + Ni2r2/M4)]

M = 1929p/p
N j = 3072a>p2@j/p
p = viscosity of fluid
p = pressure drop
From Eqs. (93) the values of cPr could be found, similarly to solving
the integral equation in Model I. Aris (A9) has also generalized this
discrete model with circular channels to a continuous model with chan-
nels of any shape. Unfortunately, the equations for the continuous case
can not easily be solved.
Even though the use of Turner’s structures to represent packed beds
may be too complex, the overall concept of utilizing frequency-response
experiments to construct detailed models is very interesting and might
find use for other situations.
5. Concentration Fluctuations
In order to get a more detailed picture of the processes occurring in
flowing systems, several investigators have directly measured concentra-
tion fluctuations. The basic ideas were introduced by Taylor (Tl) in a
classic paper, and are discussed in detail in books on turbulence such as
those by Hinze (H9) and Pai ( P l ) .
The elements of Taylor’s treatment are as follows. Consider a particle
of fluid with turbulent velocity u in a homogeneous turbulent field. The
distance that the fluid particle travels in time t is,

X(t) = /d u(t’) dt‘ (94)

Since the motion is supposed to be random, the value of X ( t ) could be
positive or negative with equal probability, and so the average for a
large number of particles would be zero,
X(t) = 0
However, the quantity, F,which measures the spreading of the particles
will not be zero:
148 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

If we introduce a Lagrangian correlation coefficient

R(s) =
u(t)u(t s)
-
+ (96)
U2

Equation (95) becomes

KampQde Fbriet ( K l ) showed that this can be changed to the form,

-
Xyt) =22 [(t - s)R(s)as
For very short times, R ( s ) + 1, and

%(t) = 2 2 /o (t -
t
S) ds (98)
-
= UZt2

The dispersion is thus proportional to the square of the time. For very
long times, t >> s,

= 2% + constant
where the Lagrangian integral time scale is

T = /omR(s) ds (99)
For long times, we see that the dispersion is proportional to the first
power of time. The dispersion coefficient can be defined in a way similar
to the Einstein equation for molecular diffusion,

1a
D = -- F
2 dt
-
= u2t short time
-
= U2T long time
Thus, the dispersion coefficient can be taken as a constant only for long
times. This, of course, would mean that the dispersion-type models would
only be valid for long diffusion times.
Unfortunately the correlation coefficient R ( s ) can not be predicted a t
the present time, although empirical relationships have been used. This
means that Eq. (98) is limited to use for the short and long diffusion
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 149

times already discussed, and it is not very useful for the intermediate
(and interesting) time ranges. Thus, other methods are necessary for
determining the dispersion coefficients.
Direct measurement of concentration fluctuations for liquid flow in
packed and fluidized beds have been made by Hanratty e t al. (H4),
Prausnitz and Wilhelm (P12), Cairns ( C l ) , and Cairns and Prausnitz
(C4). Detailed descriptions of electrical conductivity probes used for
measurement of these fluctuations have been given by Prausnitz and
Wilhelm (P11) and Lamb e t al. ( L l ) .
A concentration-fluctuation correlation coefficient, similar to that
proposed by Danckwerts (D3, D5), may be defined,

where Cl’,Cz’ are the concentration fluctuations a t points 1 and 2.

This correlation coefficient has properties similar to the one defined
by Eq. (96). I n particular, a scale of concentration fluctuation may be
defined as

where p is the distance between points 1 and 2. This scale is used as a

measure of the order of magnitude of the size of a particle of fluid that
has a uniform concentration. I n other words, fluid particles of this size
are approximately homogeneous in character. A length scale of radial
turbulent diffusion may be defined by

D R = AR& (104)
In terms of the Lagrangian integral time scale, Eq. (99), we find

AR = & (105)
These various scales, along with the intensity, or strength, of the con-
centration fluctuations,

can be used as a means of characterizing the mixing.

The detailed fluctuation data for packed beds can be found in Praus-
nitz and Wilhelm (P12). They found that the fluctuation intensity was
directly proportional to the radial position in the bed except a t the
center and near the tube wall. The magnitude of the intensity was of the
order of 8%. The radial scale, AR, was found to be about one fourth of
a particle diameter. An estimate was made for an axial scale, AL, and it
was found that approximately,
150 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

AL 2 ~ A B (107)
It was also found that for fully developed turbulent flow the above
values were essentially independent of the Reynolds number.
The mixing in fluidized beds is somewhat more complicated, as
might be expected. The over-all results found by the previously men-
tioned authors are summarized by Cairns and Prausnitz (C4):
“(1) The mixing properties in a fluidized bed are a strong function
of the fraction voids. Minimum values of radial Peclet numbers
(ud,/DR) are observed a t E = 0.7, corresponding to a transition
in the type of particle circulation in the bed.
(2) The packing particle density and fraction voids strongly affect
the radial scale of turbulence ( A R ); larger scales are found for
beds containing denser particles a t a given fraction voids. The
scale of turbulence has a maximum value a t E = 0.7.
(3) The root-mean-square value of the radial velocity fluctuation
[m2]*/C varies only slightly with fraction voids and ap-
pears to be independent of particle density.
(4)The scale of concentration fluctuation (A) is affected by the
fraction voids in the same manner as the scale of turbulence,
but it is not greatly influenced by the particle density. The
scales of concentration fluctuation show that there are isocon-
centration eddies several times the size of a packing particle.”
The above concentration fluctuation information should aid in the
fundamental description of mixing in packed beds and fluidized beds.
Exactly how this information should be used in designing such systems
must be the subject of further research.

111. Tanks-in-Series or Mixing-Cell Models

A. INTRODUCTION
I n the preceding section we discussed the dispersion model which can
account for small deviations from plug flow. It happens that a series of
perfectly mixed tanks (backmix flow) will give tracer response curves
that are somewhat similar in shape to those found from the dispersion
model. Thus, either type of model could be used to correlate experimental
tracer data.
A perfectly mixed tank can also be used, of course, to represent a
real stirred tank. Since the patterns of flow in many real stirred tanks
are rather complicated, more complex models are often required. This
whole topic will be discussed in Section IV on combined models. Thus
we will only discuss here the use of a series of perfectly mixed tanks to
 PATTERNS OF FLOW I N CHEMICAL PROCESS VESSELS 151

represent other types of flow systems, in particular those which deviate

only slightly from plug flow.
B. MATHEMATICAL DESCRIPTION
First we must derive the equations for the perfectly stirred tanks.
I n thcse ideal tanks, it is assumed that the entire contents have the same
composition as the outlet stream. Thus the C curve, or the response t o a
pulse input, can be found quite easily by a material balance.
(material in) = (material out) + (accumulation of material in tank)
COV6(t) = vc + v-
dC
dt (108)
where Co is the initial concentration of the pulse of tracer in the perfectly
mixed stirred tank. For a j-tank system, to be considered next, Co is the
average concentration of tracer if evenly distributed in the j-tank system.
In dimensionless form, Eq. (108) becomes

where

and
c = c/co
Equation (109) can easily be solved using Laplace transforms with the
result,
c=C=e* (110)
1. One-Dimensional Array
The preceding results can easily be generalized to j perfectly mixed
tanks in series. This has been discussed by many authors: Ham and Coe
(H2), MacMullin and Weber (M2, M3), Mason and Piret (M5),
Kandiner (K2), Katz (K3) and Young (Y3). Other authors have dealt
with stirred tanks specifically for use as chemical reactors, and these
will be discussed later. The material balance around the ith tank then
becomes

where Vt is the voIume of the ith tank. For simplicity, we will only deal
with the case where all the tanks have the same volume, Vi.The more
general case of unequal-sized tanks is discussed, for example, by Mason
and Piret (M5).
152 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

The C curve, or pulse response, may be found by solving the set of

Eqs. (111) for i = 1,2,3, . . ., j with the condition that the input t o the
first tank, i = 1, is a delta function of tracer. I n other words the condi-
tions on Eq. (111) are
V
Cizo(0)= 0 and Ci,o = Co; 6(t)
where V = total volume of the j-tank system. Equations (111) can again
be solved by the Laplace transform, as follows:
+
PVtCi vCi = vCi-1
where ci is the transform of Ci. Then,

-
- CO
(p: + l)j

The inverse transform of Eq. (112) is

The total volume of the system is

V = jVi
so t h a t
- v vV .
t = - V= j 2
Thus Eq. (113) may be written in dimensionless form

Equation (114) is the C curve for a series of j stirred tanks, and is shown
in Fig. 18. It has roughly the same shape as the C curve of the dispersion
model. This relationship will be further considered later.
 PATTERNS OF FLOW I N CHEMICAL PROCESS VESSELS 153

18
16
14
12

c 10
08
06
04
02

e
FIG.18.C curve for tanks in series model.

As was done with the dispersion models, the mean and the variance
can be found either directly from Eq. (114) or from the Laplace trans-
form, Eq. (112). The results are,

Pl = 1 (115)

Thus, similarly to the dispersion model, the experimental C curve data

could be used to determine a variance, 2, and then Eq. (116) used to
find j.
If a perfect delta function input is not possible, a method can again be
devised utilizing two measurement points to find the parameter, j, of
this model. Consider the experimental set up of Fig. 19. The concentra-

*...--t;$pu
M4

I
U

INPUT OUTPUT I OUTPUT 2

FIG.19. Determination of tanks-in-series model parameter by two-measurement

method.

+
tion of tracer is measured both entering the (M 1 ) t h tank (or leaving
the Mth tank) and leaving the Nth tank, and let j now be the number
of tanks in the experimental region which is between these two measure-
ment points, or j = N-M. The injection is upstream of both tanks. Now
Eq. (111) can be used for any tank i between these measurement points
giving,
154 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

ac + vci =
Vi 2
,
vci-1 (117)
at
with boundary conditions
CoM(0) = 0 and Ci(t)li-M = C d t )
The algebraic mainipulations will be easier to follow if Eq. (117) is
changed to dimensionless form. Define,
V volume of j-tank experimental section
+ +
=
= VM+i VM+e * * * -k VN

= jVi

V
t = - = mean residence time in the j-tank experimental section
V

C
c=co
Co = average concentration of all entering tracer if evenly distributed in
the j-tank experimental section
Then Eq. (117) becomes for the Nth tank

Solving Eq. (118) by Laplace transforms,

Now if Eq. (119) is manipulated in exactly the same way as that used
for the dispersion model by Aris (A8), Bischoff ( B l l ) , and Bischoff and
Levenspiel (B14) (Section II,2,c), the following relationships are found:
Api = pw - pi^ = 1 (120)
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 155

Thus if an imperfect delta function is injected into a system and the

mean and variance measured a t two measurement locations, the tanks-
in-series model can also be used to interpret the results.
There would seem to be little reason for using the tanks-in-series
model for flow in empty tubes, since little correspondence exists between
the physical picture and the model. However, Coste et al. (C22) found
recently that a system with mass and heat dispersion combined with
chemical reaction was easier to handle with this model than with the
dispersion model. On the other hand, Carberry and Wendel (CS) have
solved a similar problem with the dispersion-model by a finite-diff erence
technique different from the one used by Coste e t al., and found no diffi-
culties. Thus the question of which model is best computationally is still
not answered.
The use of the tanks-in-series model for packed beds can be more
strongly justified. The fluid can be visualized as moving from one void
space to another through the restrictions between particles. If the fluid
in each void space were perfectly mixed, the mixing could be represented
by a series of stirred tanks each with a size the order of magnitude of
the particle. This has been discussed in detail by Aris and Amundson
(A14). The fluid in the void spaces is not perfectly mixed, and so an
“efficiency)’ of mixing in the void spaces has to be introduced (C6).
This means that the analogy is somewhat spoiled and the model loses
some of its attractiveness. In laminar flow the tanks-in-series model may
be still less applicable.
2. Three-Dimensional Array
We have discussed the tanks-in-series model in the sense that the
composition in the system was constant over a cross-section. Recently
Deans and Lapidus (D12) devised a three-dimensional array of stirred
tanks, called a finite-stage model, that was able to take radial as well
as axial mixing into account. Because of the symmetry, only a two-
dimensional array is needed if the stirred tanks are chosen of different
sizes across the radius and are properly weighted. By a geometrical
argument, Deans and Lapidus arrived a t the following equation for the
(i, j ) tank:

where
=
cj - Wf-l,j-1/2 + cj - $)Ci-1,,+1/2
(af-l,j
(2.i - 1)
with boundary conditions,
Cf,, = Co (initial condition)
C0,,= C’ (inlet t o bed)
156 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

By choosing the stirred-tank size about the same as a particle in the

packed bed, Eq. (122) will reproduce experimental mixing data. This is
similar, of course, to what was found for the one-dimensional case
previously discussed.
The reason for constructing this rather complex model was that even
though the mathematical equations may be easily set up using the disper-
sion model, the numerical solutions are quite involved and time con-
suming. Deans and Lapidus were actually concerned with the more
complicated case of mass and heat dispersion with chemical reactions.
For this case, the dispersion model yields a set of coupled nonlinear
partial differential equations whose solution is quite formidable. The
finite-stage model yields a set of differential-double-difference equations.
These are ordinary differential equations, which are easier to solve than
the partial differential equations of the dispersion model. The stirred-
tank equations are of an initial-value type rather than the boundary-
value type given by the dispersion model, and this fact also simplifies the
numerical work,
Thus, i t would seem that calculations using the finite-stage model
might be easier than with the dispersion model. However, i t has been
found by Schechter and Wissler (55) that the set of difference equations
used in the finite-stage model are equivalent to one of the sets of pos-
sible difference equations that could be used to solve numerically the
partial differential equations of the dispersion model. Looking a t the
finite-stage model equations from the numerical-analysis point of view,
it is seen that the finite-difference mesh size is dictated by the particle
size and the experimental value of the dispersion coefficient that is to be
simulated. This lack of flexibility could conceivably cause inaccuracies
in the calculated results. A further problem when the finite-stage model
is used for simultaneous mass and heat dispersion is that the magnitudes
of the two types of dispersion must be taken to be equal, since they are
both fixed by the same mesh size.
Thus, we conclude that the computational effort required will be
approximately the same for either model. On the one hand the finite-stage
model is somewhat inflexible as indicated above, but on the other hand
it might be superior to the dispersion model in systems with large
particles where the continuum approach of the dispersion model would
probably not be appropriate. On the whole, more work is needed to deter-
mine the practicality of the various computation methods.

WITH DISPERSION
C. COMPARISON MODEL
Since both the tanks-in-series model and the dispersion model give
about the same shape of C curve, the question arises as to how similar
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 157

are the predictions of the two models. There are several methods to com-
pare the two. One method uses variances. Kramers and Alberda (K14)
used the variance for the doubly infinite dispersion model, which from
Section 11, C, 2,b is,

Using this with Eq. (116) gives

1=2
3
7 (2)
Unfortunately, Eq. (124) does not extrapolate to j = 1 as D L + a.
Therefore, Kramers and Alberda suggested using
- =12 ( % )
j-1
This expression extrapolates correctly, and is approximately the same as
Eq. (124) for large j , say j > 10.
Levenspiel (L13a) Iater showed that the reason for the incorrect
extrapolation of Eq. (124) was that the doubly infinite vessel was not
the proper one to use for the comparison. Instead, the closed vessel (plug
flow into and out of the vessel) must be used. The expression for the
variance can be found from van der Laan (V4), and when combined
with Eq. (116) gives

Equation (126) does extrapolate properly to j = 1 for DL-+ co. For small
DL/uL (large j ) it approaches Eq. (124).
We also notice, from either Eq. (124) or Eq. (126), that j + 00
as DL+ 0. This is the basis for the statement that an infinite number of
stirred tanks in series is the equivalent to plug flow.
Trambouze (T10) suggests two alternate methods of comparison. By
matching the C curve maxima for these two models, he showed that
DL- (2j - 1)2
UL - 2 j ( j - 1)(4j - 1)
and by matching the C curves a t 0 = 1 he obtained

Both equations extrapolate properly for DL/uL+ 0, but only Eq. (127)
gives j = 1 for DL/uL= CQ. Equation (127) also reduces to Eq. (124)
for small DL/uL (large j) .
158 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

From the above it is seen that there is no unique way of matching

the two models. This conclusion is strengthened when chemical reactions
are considered because the comparisons for each reaction type and de-
gree of conversion are different, and none of them is the same as the
comparison obtained from matching tracer curves. An illustration of the
different methods of comparison is given in Fig. 30.
As a means of choosing between models it has been suggested that
some of the higher moments of the C curve could be used. The skewed-
ness, or third moment, of either the dispersion or tanks-in-series model is
uniquely determined by the value of the dispersion coefficient or the num-
ber of tanks. Thus, the parameter is determined from the second moment
and then used to calculate the third moment. This is compared with the
third moment computed from the experimental data ; whichever model
has the closest third moment would be chosen. Unfortunately, this method
has two drawbacks that severely limit its usefulness. One is that experi-
mental data is not good enough to give meaningful third moments. The
other is that the third moments as calculated from the different models
have almost the same values, as was shown by van Deemter (V2).
As mentioned when discussing the three-dimensional finite-stage
model, by a proper choice of the stage size, experimental dispersion data
may be simulated as discussed previously for packed beds. Aris and
Amundson (A14) originally showed that by taking the size of the indi-
vidual stirred tanks equal to one particle diameter, the value DL/udp
= 0.5 can be reproduced. Inspection of Fig. 8 shows this to be a rough
estimate of the value found from experiment. Deans and Lapidus (D12)
found that for the same size of stirred tanks, the value Dn/udp = 1/8
can be reproduced. Inspection of Fig. 11 shows that this also is a reason-
able estimate of the experimental data. By using a finite-stage size other
than exactly one particle diameter, other values of the dispersion groups
may be generated.
I n conclusion then, it would seem that for small deviations from plug
flow, both the dispersion and the tanks-in-series models will give satis-
factory results. Up to the present, which one is used may be largely a
matter of personal preference.

IV. Combined Models

A. INTRODUCTION
When the gross flow pattern of fluid deviates greatly from plug flow
because of channeling or recirculation of fluid, eddies in odd corners, etc.,
then the dispersion model or the tanks-in-series model can not satis-
factorily characterize flow in the vessel. This type of flow can be found,
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 159

for example, in industrial stirred-tank reactors, shell-and-tube heat ex-

changers, and in fluidized beds. I n these situations it is probably most
fruitful to view the real vessel to consist of interconnected flow regions
with various modes of flow between and around these regions. Models
such as these can be called combined or mixed models.
The following kinds of regions are used in the construction of com-
bined models:
Plug flow regions,
Backmix flow regions,
Dispersed plug-flow regions,
Deadwater regions.
The last-mentioned region accounts for the portion of fluid in the
vessel which is relatively slow moving and, for practical purposes, stag-
nant. As we shall see, there are two ways to deal with deadwater regions:
to assume their contents to be completely stagnant or to view a slow
interchange of their contents with the fluid passing through the vessel.
In the first approach the treatment is quite simple; the second approach
more closely approximates real situations but requires quite involved
analyses.
I n addition to these regimes, combined models may use the following
kinds of flow:
Bypass flow, where a portion of the fluid bypasses the vessel or a
particular flow region,
Recycle flow, where a portion of the fluid leaving the vessel or leaving
a flow region is recirculated and returned to mix with fresh fluid,
Cross flow, where interchange, but no net flow, of fluid occurs between
different flow regions.
With these as the components of combined models the problem then is
to find the volumes of the various regions and the rate of each type of
flow occurring such that the response curves of the model match as closely
as possible the response curves for the real vessel.

B. DEFINITION
OF DEADWATER
REGIONS
1. Completely Stagnant
I n attempting to represent flow in a real vessel by combined models
containing completely stagnant fluid, we meet with difficulties. For ex-
ample the existence of completely stagnant fluid cannot be reconciled
with the assumption of steady flow through the vessel. Again, with this
definition the mean age of the vessel contents would not be useful in
matching models because even if a deadwater region consisted of only
160 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

one molecule, the calculated mean age of the vessel would still be infinite.
The following definition of deadwater regions by Levenspiel (L12)
overcomes these difficulties while still maintaining a concept of these
regions which is useful in matching models with real situations.
“In a vessel the deadwater regions are the relatively slow moving portions of the
fluid which we chose to consider to be completely stagnant. Deadwater regions con-
tribute to the vessel volume; however we ignore these regions in determining the
various age distributions.’’
The cutoff point in residence time between what we chose to con-
sider as active and as stagnant fluid depends on the accuracy of predic-
tions of vessel performance. In most cases material which stays in a
vessel twice the mean residence time can, with negligible error, be taken
as stagnant.
2. Slow Cross Flow
Instead of considering fluid in deadwater regions to be completely
stagnant, an alternate view considers that there is a slow interchange
or cross flow between the fluid in these regions and the active fluid passing
through the vessel. With this approach Adler and Hovorka (A2) treated
the combined model shown in Fig. 20. This consists of j identical units

FIG.20. Combined model used by Adler and Hovorka (A2),

in series, each of which contains a plug flow, a backmix and a deadwater

region. The deadwater region is viewed to be in backmix flow and to be
interchanging fluid slowly with the active backmix flow region. This
model has four parameters; the cross flow rate, the number of units in
series, and two parameters for the ratio of sizes of the three types of
regions. The procedure developed for determining the model’s parame-
ters is as follows:
(1) From the experimental C curve find when tracer just appears
(where the C curve just rises from zero) and also locate the maximum of
this curve.
(2) Specially prepared charts for different integral j then give sets
 PATTERNS OF FLOW I N CHEMICAL PROCESS VESSELS 161

of parameter values, one for each j, consistent with the above C curve
findings.
(3) The tracer curves for these different j are then matched with
the experimental C curve to select the curve of best fit with its corre-
sponding set of parameters.
This fitting procedure is designed to match closely the critical part
(the early section) of the tracer curve, hence the predictions of this
model should correspond to actual performance. This model is quite flexi-
ble in that it is capable of fitting extremely skewed age distributions.
Its disadvantage when compared with the simpler combined models em-
ploying completely stagnant regions is that the third step of the matching
procedure necessitates the use of computers.
In a model for the structure of packed beds, Turner (T14, T15) and
Aris (A9, A10) have also used stagnant pockets with crossflow by only
molecular diffusion.
C. MATCHING COMBINED MODELSTO EXPERIMENT
The following suggestions may be helpful in searching for and devis-
ing flow models to fit given experimental response curves. This section
has particular application to models which employ the simple stagnant-
fluid definition of deadwater regions. The matching procedure for the
particular model of Adler and Hovorka has already been outlined.
1. Existence of Deadwater Regions.
Select a reasonable cutoff point, say 6 = 2, then find the mean of the
C curve up to that point. If no deadwater regions are present, then
measured fc
ec = hrl
v/v
If deadwater regions are present, then
ec <1
The fraction of vessel consisting of deadwater regions is given by the
deviation of Jc from unity. Hence with V d as the volume of deadwater
region, and with v,/v as the fraction of active fluid as given by the area
of the C curve up to the cutoff point,

Danckwerts (D9) discusses this type of method for finding the location
of stagnant regions in systems. Alternatively the area under the I curve
will give the dead volume. Figure 21 summarizes these results.
162 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

Model for vessel

______
J

/-Reg!on of aclive llow

! g m considered lo
1s Ignored in be campletelv
stagnant
devising models

I
I
Measured area only slightly I
different from unity I
I
,'o
I
I
V
I
Shift of $from unity I
measures deadwater I
n
f region I

toff I Area ignwd i n tat! 1s

(very small but contributes
much in shifting 8,
from unity = y , / v

FIG.21. Particular features of age distribution curves for combined models which
include deadwater regions. Adapted from (L13).

2. Existence of Bypass Flow

In bypass flow we may look at the incoming fluid as splitting into two
parallel streams, the fraction passing through the vessel being 211, the
fraction bypassing it instantaneously being 212. Figure 22 shows char-
acteristics of typical response curves when bypass flow occurs. From the
rapid initial drop in the I curve, the shift in mean value for the main
portion of the E curve, or from the area of the main portion of the
E curve the magnitude of such short-circuiting can be estimated. Prob-
ably the role of bypass flow can be evaluated more easily from F or
I curves than from the C or E curves.
 PATTERNS OF FLOW I N CHEMICAL PROCESS VESSELS 163

Bypass flow portion 01 E curve,

CL
area = 1
Main portion of E curve,

&=I 8E=L 2

for the whole E curve 1 e

t-";or main portion of E curve

FIG.22. Particular features of age distribution curves for models which include
bypassing of fluid (L13).

3. Regions in Series
For flow regions 1, 2, . . . connected in series the mean age of vessel
contents is

while the mean age of fluid in the exit stream is

ZE = + ZE,2+ ' * * (131)
Here Vl, Tiz, . . . refer only to the active volumes,
v refers to the total vessel volume including deadwater
regions
&,Zz, . . . refer to the separately measured mean time for the flow
regions 1, 2, . . .
164 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

4. Regions in Parallel
For flow regions 1, 2, . . . connected in parallel the mean age of
vessel contents is

wliilc tlie mean age of fluid in the exit stream is

5. Number of Parameters in a Model

The number of parameters used in a model is an indication of its
flexibility in fitting a wide variety of situations, and in addition suggests,
to some extent, the complexity of the accompanying mathematics. With
more and more parameters, the models are able to fit a wider variety of
conditions. However, we must balance this gain against the unwieldiness
of the accompanying mathematics as well as the possibility that such a
model may have very little correspondence with fact. The latter is a
serious objection because an unrealistic many-parameter model may
closely fit all present data “after the fact,” but may be quite unreliable
for prediction in new untried situations. Hence, in fitting a real situation,
we should aim for the simplest model which fits the facts and whose
various regions are suggested by the real vessel. This way the parame-
ters of the model have physical meaning and may be predicted by inde-
pendent methods. This question is of especial concern in fitting models
to fluidized beds.
The tanks-in-series and dispersion models are one-parameter models.
I n general, the number of parameters in a combined model is

number of )=.(
flow regions in ) z:(flow paths in)
(parameters
+

excess of one excess of one

+ z:(zones
cross flow
) + z (with
Of
flow regions
dispersion )

4 arbitrary restrictions on
flow and volume ratios ) (134)

Using the “completely stagnant” interpretation of deadwater regions,

Fig. 23 illustrates some simple combined models and their tracer-response
curves. I n these models V b , V,, and V d stand for the volume of backmix,
plug flow, and deadwater regions. If V is the volume of vessel we then
have
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 165

vD \

lt 't

FIQ.23. Simple combined models and their corresponding age distribution func-
tions (L13). (Continued on p p . 166 and 167.)
166 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

*\
a = klk +1

a l k 2alk h l k

V
II
bl

a l k 2alk 3alk
8

U J

+ vbf [ Infinite
cross flow

Area = 1

FIG.23 (Continued).
 PATTERNS OF BLOW I N CHEMICAL PROCESS VESSELS 167

e
A

FIQ.23 (Continued).

v = z Va + B v, + z Va (135)
The volumetric flow rates of streams in parallel are designated by vl,
v2 * * -. If v is the flow rate of fluid to the vessel, we then have
v = Vl+VS.+ *.- (136)
Varying the relative sizes of the flow regions as well as the flow rate of
parallel streams allows great flexibility in matching the response curves
of these models to that for the real vessel. Model F of this Figure has
also been extended to j such units in series by Brothman et al. (B22).
The following brief discussion shows how combined models are being
used to characterize flow in two broad classes of process equipment,
stirred tanks and fluidized beds. Other types of mixed models have also
been devised for various purposes; by Bartok e t al. (B3), Cholette and
Cloutier (C16), Handlos e t al. (H3), Pansing (P3), and Singer et al.
(517). Eguchi (E2) presents and discusses some of the models used to
date.
D. APPLICATION TANKS
TO REALSTIRRED

When the time required for an element of entering fluid to achieve

homogeneity with the rest of the vessel contents is small with respect
168 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

to the mean residence time of fluid in the vessel, then the fluid in this
vessel may be considered to be in backmix flow.
Stead et al. (S23) and Johnson and Edwards (53) showed that homo-
geneity can be achieved in as short a time as 0.1 sec., with sufficient agita-
tion in a laboratory sized stirred tank. The relation between this time
and the intensity of agitation was studied by MacDonald and Piret ( M l ) .
Eldridge and Piret (E5) then used kinetic experiments to show that a
series of up to five laboratory-sized stirred tanks with sufficient agita-
tion acted as perfect backmix reactors.
Aiba (A3), Fox and Gex (F8), Kramers, Baars and Knoll (K15),
Metzner and Taylor (MlO), Norwood and Metzner (N3), Van de Vusse
(V5) and Wood et al. (W12) have studied flow patterns and mixing
times. I n addition, Brothman et al. (B22), Gutoff (G9), Sinclair (S16)
and Weber (W3) analyzed flow in a stirred tank in terms of the recycle
flow model of Fig. 23F. This model corresponds to the draft-tube reactor,
and with sufficiently large recycle rate the performance prediction of this
model approximates backmix flow.
A study aimed a t devising a model for the experimentally found
deviations from backmix flow of fluids through vessels was made by
Cholette and Cloutier (C16). Using various low agitation rates, these
investigators explored the nonideal flow of fluid in a 30-in.-i.d. 30-in.-high
stirred tank. Matching F curves, they found that their data were best
described by a combination of a backmix and a deadwater region with a
portion of the fluid shortcircuiting the vessel. The internal age-distribu-
tion function for this model is

or
Va -1 e - t / t a
I(t) = -
v fa
and the exit age-distribution function is

or

The first term of Eq. (140) represents flow through the active portion of
the vessel with a mean residence time of Fa = Vb/vl. The second term of
Eq. (140) represents the fluid which is bypassing the vessel. This model
 PATTERNS OF FLOW I N CHEMICAL PROCESS VESSELS 169

#I=$ 3
FIQ.24. Combined model of Cholette and Cloutier for real stirred tanks ((316, L13).

and its tracer-response curves are shown in Fig. 24. I n this study it was
also found, as expected, that the experimental conditions influenced the
parameters of the model.
I n a follow-up study Cloutier and Cholette (C19) examined in detail
the influence of feed inlet location, impeller size, and impeller speed on
the parameters of the model. Their results, shown in Fig. 25, indicate that:
(1) If the agitator speed falls below some critical value then the ac-
tive volume V b / V drops to and remains a t some constant value.
(2) At a given feed rate this critical agitator speed is a function of

f(W -200
AGITATOR SPEED, r.p. m.
FIG.25. Effect of intensity of agitation on the parameters of the model for real
stirred tanks ((319).
170 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

impeller size, suggesting that the energy input into the fluid could deter-
mine when the deadwater region reaches a maximum.
(3) The minimum active volume depends on feed location but not on
impeller size. This suggests that the maximum size of deadwater region
depends on the geometry of the vessel.
(4) Above the critical agitator speed, the active volume rises linearly
to unity with r.p.m. This rate of rise is a function of impeller blade size
(hence energy input) but is independent of feed location (or vessel
geometry).
( 5 ) Although a smooth line can be drawn through the values of wl/w,
i t is difficult to correlate this factor with the experimental conditions.
Results indicate that while the onset of bypassing and deadwater may be
concurrent they need not be so.
E. APPLICATION
TO FLUIDIZED
BEDS
Numerous investigators have studied mixing of fluids in fluidized beds.
Danckwerts e t al. (DlO), Gilliland and Mason (G4, G5), Gilliland e t al.
(G6), and Huntley e t al. (H16) have given data on the distribution of
residence times. Others have used the dispersion model to characterize
flow in fluidized beds: Askins e t al. (A17), Brotz (B23), Cairns and
Prausnitz (C3), Gilliland and Mason (G4), Handlos e t al. (H3),
Hanratty e t al. (H4), Muchi e t al. (M13), Reman (R2), Trawinski
(T13), Wicke and Trawinski (W9), Wakao e t al. ( W l ) , and Yagi and
Miyauchi ( Y l ) . The inability of this approach to yield broad predictive
correlations, particularly with solid-catalyzed gas-phase reactions, seems
to show that this one-parameter model can not satisfactorily explain
fluidized-bed behavior. A different approach was needed, the basis for
which was found in the observation that gas-solid fluidized beds seemed
to consist of dense regions through which pass bubbles of gas.
A number of models all having a dense or emulsion phase and a lean

@I=
or bubble phase have been proposed. These are all special cases of the

Fraction of all solids

which are present
Dense in this phase, m

phase phase
v1

v
FIG.26. General two-region model of a fluidized bed. Fluid is in dispersed plug
flow in both regions. The six parameters of this model are m, 2, Vl, v1, D1,and 0 2
(L13).
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 171
general two-region, six-parameter model shown in Fig. 26. This general
model has not yet been used, for two reasons: the difficulty in interpret-
ing experimental data so as to evaluate the model parameters, and the
fact that probably fewer parameters could equally well represent reality.
Many different sets of restrictions have been proposed to reduce the gen-
eral model to more tractable form with fewer parameters. The restrictions
used are of the following kind:

(1) Fix the dispersion coefficients of the dispersed plug flow model,
DL= D 1or D2,a t infinity or zero to obtain backmix or plug flow
in the individual regions.
(2) Assume that no solids are present in the lean phase.
(3) Assume that there is no net gas flow upward through the dense
phase.
(4) Assume that the volume of dense phase, the fraction solids in it,
and the gas flow through it remain the same at all gas velocities,
in which case, the lean phase alone expands and contracts to ac-
count for the variation in total volume of fluidized bed with
change in gas flow rate. The dense-phase characteristics are given
by the conditions a t incipient fluidization.

Table IV shows the restrictions which must be placed on this general

model to obtain each of the special cases studied. Also shown are the
number of parameters for each of the models. What is now needed is an
evaluation of these models: to find those models which fit the fluidized
bed in its wide range of behavior, and then to select from these the
simplest model of good fit. Practically every one of these models is flexi-
ble enough to correlate the data of any single investigation; consequently
a proper evaluation would require testing every model under the ex-
tremely wide variety of operating conditions of different investigators.
Other aspects of the behavior of fluidized beds can be found in books
by Leva (L9) and Zenz and Othmer ( Z l ) , and in the literature from
which these books have drawn.

V. Applications of Nonideal Patterns of Flow to Chemical Reactions

Conversion in a reactor with nonideal flow can be determined either

directly from tracer information or by use of flow models. Let us con-
sider each of these two approaches, both for reactions with rate linear in
concentration (the most important example of this case being the first-
order reaction) and then for other types of reactions where information
in addition to age distributions is needed.
 TABLE IV
MODELSFOR FLUIDIZED BEDSO

Restrictions on general model Model parameters for

Model Homogeneous Heterogeneous
Dense phase Lean phase reactions reactions Reference

D1 = 0 (plug flow) D L= 0 (plug flow) Shen and

MI 2 X
01, VI fixedb no solids, m = 0 Johnstone (S12)
Gomesplata
and Shuster (G7)
~ -~
D1 = = (backmix) Dz 0 (plug flow)
= Shen and
M2 vl, V1 fixed* no solids, m = 0 2 2
Johnstone (512)
D1 = 0 (plug flow) Mathis and
M3 4 = 0 (plug flow)
VI, VIk e d b Watson (M7)
Lewis et al. (L17)
M4

Dl= (backmix) Lewis et al. (L17)

M5 D2 = 0 (plug flow)
(
1 :

01 = 0 VI, m,x

M6 Dz = 0 (plug flow) VI, VI, X, DI 01, 2, Di May (M8)

no solids, m = 0
M7 no solids, m = 0 Vi,01, Di,Dz VI, DI,Ds van Deemter (Vl)
no cross flow, 3: = 0

M8 D1 = 0 (plug flow) Dz = 0 (plug flow) V1,01, m, z VI, m, z Lanneau (L2)

z = cross-flow rate.
m = fraction of all solids present in lean phase.
a From Levenspiel, O., “Chemical Reaction Engineering,” John Wiley, New York, 1962.
’ As given by conditions of incipient fluidization.
+ +
Note: V = V I VZand v = UI uz are known and are not parameters of the models.
 PATTERNS OF FLOW I N CHEMICAL PROCESS VESSELS 173

A. DIRECTUSE OF INFORMATION
AGE DISTRIBUTION
1. General
The distribution of residence times gives information on how long
various elements of fluid spend in the reactor, but not on the detailed
exchange of matter within and between the elements. For a reaction with
rate linear in concentration, the extent of reaction can be predicted solely
from knowledge of the length of time each molecule has spent in the
reactor. The exact nature of the surrounding molecules is of no im-
portance. Thus the distribution of residence times yields sufficient infor-
mation for the prediction of the average concentration in the reactor
effluent.
For all other types of nonlinear reactions, however, the extent of the
reaction depends not only on the length of time spent in the reactor but
also on what other molecules were “seen” during the passage through
the reactor. I n this case then, the distribution of residence times is not
sufficient, and detailed information on the degree of mixing would be
required to predict the average concentration in the reactor effluent.
If i t is assumed that each element of fluid passes through the reactor
with no intermixing with adjacent elements (termed segregated flow), the
distribution of residence times can be used to determine the conversion.
Thus
concentration fraction of
mean Concentration of reactant exit stream

age between
t and t + dt

Equation (141) predicts the conversion for what can be termed a “macro-
fluid” in which aggregates of molecules move about in “insulated”
packets. (In plug flow, this model will always apply.) The other extreme,
which can be called a “microfluid,” is a fluid in which mixing occurs on
the molecular scale. A real fluid lies somewhere between these two ex-
tremes, and in normal cases, much closer to the microfluid extreme. The
effects of this mixing on reactions has been studied by Danckwerts (D7,
D8), Greenhalgh et al. (G8), Metaner and Pigford (M9), Gilliland and
Mason (G5), Gilliland et al. (G6), Sherwood (S13), and Zwietering (22).
Further papers in this field were included in two Symposia on “Chemical
174 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

Reaction Engineering," published in Chem. Eng. Sci. 8, (1958) and 14,

(1961). We will discuss these effects for each type of reaction.
2. Linear Rate Equations
As was discussed previously, the exact state of mixing has no effect
for reactions with rate linear in concentration. For such reactions Eq.
(141) may always be used to predict the conversions. Thus, we have
dcelement -
- r c = --
dt - h(Celement - hi) (142)

From a physical standpoint, C can never be negative; thus, we must

restrict k2 2 0 for the following treatment. This restriction eliminates
zero-order reactions, whose conversions do depend on the state of mixing.
With Celement = Co at t = 0, integration gives
(Co - kde-"'
Cclement = + kz (143)
Substituting into Eq. (141) yields

C = kw[(Co - kz)e-"" + kz]E(t)dt (144)

We consider now the special forms of Eq. (144) for a first-order reac-
tion (i.e., k2 = 0) occurring in various types of flow. In plug flow
E(t) = 6 ( t - f) (145)
Hence, as expected from the usual methods of kinetics,

(146)

For backmix flow, in a single backmix reactor,

-f/t
E(t) = t (147)
hence,

Again this is the expected result from the methods of kinetics.

For a series of equal-sized backmix reactors the exit age distribution
function is

hence Eq.(141) becomes

 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 175

For a series of unequal-sized backmix reactors with mean residence

- -
.
times 6, 8,. , , 4, . . . , tk we have

hence Eq. (141) becomes

For flow with arbitrary exit age distribution E(t) Eq. (144) must
be solved directly. A convenient graphical method for doing this has been
devised by Schoenemann (S9) who then discusses the application of this
method to some industrial reactors. The direct use of Eq. (144) is also
illustrated by Levenspiel (L13), Sherwood (S13) and Petersen's (P5)
treatment of catalyst-activity levels in regenerator-reactor systems.

14:
3. Nonlinear Rate Equations
For reactions with rates that are linear in concentration, conversions
cannot be calculated from tracer information alone, since a given tracer
curve can represent a range of flow patterns with earlier or later mixing
EARLY MIXING ONES

REacTlON
OF
RATE

CONCENTRATIW

FIG.27. Characteristic curvature of the rate-concentration curves for reactions

which favor either early or late mixing of fluid.

of fluid. Thus a specific tracer response curve can be consistent with a

range of possible conversions. As shown in Fig. 27 the curvature of the
rate-concentration curve will tell whether early or late mixing will favor
high conversions.
176 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

a. Conversion for Late Mixing. Since Eq. (141) assumes the latest
possible mixing, thus no intermixing of fluid elements, and this in turn
implies that the concentration of reactants remain as high as possible, it
yields the upper bound to conversion for reactions with order greater
than unity, but yields the lower bound for reaction orders smaller than
unity.
b. Conversion for Early Mixing. Zwietering (22) has given a treat-
ment that shows how to calculate the conversion for the earliest possible
mixing consistent with a given age distribution. It is based on a quantity,
J, called the degree of segregation, introduced by Danckwerts (D8) :
variance of ages between points
J =
variance of ages of all molecules in system
1/v ,/ ( a p - a)Zdi,-
(153)

where,
a = age of a molecule in the system
a = mean age of molecules
= /om aI(a)da (154)
a p = mean age within a point P

= /ow a’Ip(a’)da’ (155)

I p ( a ’ )= age distribution within a point P
and the volume integral represents the sum over all points. The term
“point” is taken to mean a region small compared to the scale of segrega-
tion but large enough to contain many molecules (D8).
Danckwerts (D8) discussed the case of a perfect mixer. For the
completely segregated case, all molecules within a point have the same
age, a, and so
Ip(a’) = 8(a’ - a)
and
ap = /ow a’+’ - a) da’ = a ( 156)

Also, the distribution of points can be found from the I(&) curve, and so,

1 / V /v (a - dV hm (a - ~ ) ~ 1 (daa )
J =
Lm -(a ?i)21(a)
da
-
Lrn (a - n)’I(a) d a
= 1 (157)
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 177

For mixing uniform on a molecular scale, the distribution of ages within

any point is the same as that in the entire system, and so
)’.(,I = I(a’)
and,
a, = hm a’I(a’) da’ = 3 (159)
Thus,
1/vIv (a - 5) dl‘ -
J =
hm - (a n)ZI(a) da
(160)

I n summary, for a backmix reactor, J is unity for perfect segregation

and zero for perfect molecular mixing.
Zwietering (22) generalized this treatment to an arbitrary age dis-
tribution. The argument for perfect segregation is still valid for this
case, and so the upper limit of J is again unity. The lower limit is not
zero, however, because there must be a difference in ages a t various
points in the system. The state of “maximum mixedness,” or earliest
possible mixing, is arrived a t by a rather lengthy argument involving
the definition of the life expectation in the vessel, A, of a molecule,

(of a molecule ) + (life expectation of a molecule)

residence time
= (age of a molecule)
t=a+X (161)
Life-expectation distribution functions are derived, and the final result
for earliest mixing, assuming no radial variations, is,

(a, - a)* dV = lm
‘[ 1;
I@,)
I(s) ds - a]” I@,) dX, (162)

Thus knowledge of the I ( t ) curve will permit calculation of the lower

bound of J. Unfortunately there is yet no way to determine the variance
between points for the general case, and so all that we can do is calculate
the two extremes.
For a chemical reactor operating with earliest possible mixing,
Zwietering gives an equation for the calculation of conversion,

The reactor outlet conversion is found from C(0). The conversion for
segregated flow is, of course, found from Eq. (141), discussed previously.
With these equations, the limits between which the conversion must lie
178 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

for any real reactor can be predicted. More work is needed, however,
before any closer predictions within the two extremes may be made.

B. STIRRED
TANKREACTORS
1. Ideal Stirred Tanks
For one perfect stirred tank, the formulation used in Section III,B
is used, modified by adding a term for the chemical reaction,

c0v= clv + v dC
-dt
' - Vr,(C,)
The steady state solutions of these equations are well known and
will not be considered here. Extensions to steady state flow in a chain
of stirred tanks also have been extensively treated in the literature:
Denbigh (D16, D17), Devyatov and Bogatchev (D18), Eldridge and
Piret (E5), Fan (F3), Jenney (J2), Kirillov (K6, K7), Kirillov and
Smirnova (K8), Leclerc (L6), Lessells (L8), MacMullin and Weber
(M2, M3), Stead et al. (S23), and Weber (W3).
Numerous short cut procedures for solving graphically the design
equations for nth order reactions are available in the literature; for
example see Fan (F3), Hofmann (HlO), Jenney (J2), Lessells (L8),
Levenspiel (L13) and MacMullin and Weber (M2, M3). For more
complex reaction types, Eldridge and Piret (E5) have given a catalog
of solutions. Bilous and Piret (BS), Jones (J4), Jungers et al. (J5),
Levenspiel (L13), and Trambouee and Piret (T12) have discussed gen-
eral design methods.
The solutions of the nonsteady-state expression, Eq. (1641, both for
single tanks and chains of tanks have been made by Acton and Lapidus
( A l ) , Mason and Piret (M5, M6), and Standart (522). Ark and
Amundson (A15, A16), Bilous and Amundson (B7), Bilous e t al. (B9),
and Gilles and Hofmann (G3) have studied the stability, control, and
response of a stirred tank reactor.
Nagata, e t al. (Nl, N2), Kawamura et al. (K4), and Yagi and
Miyauchi (Y2)have studied the characteristics of various impeller
agitated multistaged vessels. Such vessels were assumed to be a succession
of plug-flow and backmix units, whose relative sizes were a function of
the impeller speed. The parameter of the model, the fraction of total
volume in a plug-flow, could also be related to a dispersion coefficient.
Verification of the model was then obtained with kinetic experiments.
Aris (A13), Cholette and Blanchet (C15), Cholette e t al. (C17), and
Trambouee and Piret (T12) have discussed using combinations of
backmix and plug-flow reactors.
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 179

2. Nonideal Stirred Tanks

The conversion in a real, hence, nonideal, stirred tank reactor can
be calculated for the model proposed by Cholette and Cloutier (see
Section IV,D). The exit stream consists of reacted fluid from the active
backmix region combined with unreacted bypassing fluid. By material
balance, then,
total exit
(stream )
fluid from the
= (backmix region) h?ring)
unreacted

or
(v1 + v*)C = VlCBM + VZCO

where C,,/Co is found from the design equation for backmix flow, or,

For first order reactions, -rC = klCUJr;thus, combining,

while for second order reactions,

C. TUBULAR BED REACTORS

AND PACKED

Both the dispersion and tanks in series models can be used to represent
the non-ideal flow behavior of fluids in packed bed and tubular reactors.
As mentioned in the previous sections dealing with these models, they
are both good for the slight deviations from plug flow encountered in
the above systems.
General discussions of several aspects concerning the treatment of
chemical reactions with diffusion are given by Damkohler (D2), Horn
and Kuchler (H12), Prager (P7), Schoenemann and Hofmann (SlO),
and Trambouze (T11). Corrsin (C21) has discussed the effects of tur-
bulence on chemical reactions from the fundamental point of view of
turbulence theory. We will first discuss the application of each type of
model to chemical reactors. Then a short comparison will be made
between the different approaches.
180 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

1. Dispersion Model
a. Axial-Dispersed Plug-Flow Model. The mathematical description
of the process is provided by Eq. (1-5) of Table I. For steady flow the
equation reduces to
ac azc
u-=
ax DL-+T,
a
x 2

For a reaction of order n, Eq. (167) becomes,

ac a2c klC"
u-
ax = D L - -
a
x 2

The proper boundary conditions to use with Eq. (167) have been ex-
tensively discussed. Wehner and Wilhelm (W4)gave a general treatment
and used the conditions already discussed in Section II,C,2,b, Eq. (26).
This involved using three sections with different dispersion characteristics
in each: the fore section, X 5 0, the reaction section, O<X I L, and the
after section, X 2L.Similar boundary conditions for the special case of
no dispersion in the fore and after sections have been discussed by
Damkohler ( D l ) , Hulburt (H14), Danckwerts (D4), Pearson (P4), and
Yagi and Miyauchi ( Y l ) . For this case,

co = C(O+) - D-LdC(O+)
u ax
and
dC (L-1
ax = o
where Co = concentration of unreacted feed. Wehner and Wilhelm solved
Eq. (168) with the general boundary conditions for a first-order reaction,
n = 1, and obtained the following result for X = L,

where

Equation (171) turned out to be the same result that had been obtained
using the simpler boundary conditions assuming no diffusion in the fore
and after sections. In other words, the solution of Eq. (168) with the
general boundary conditions gives the same result as with the simpler
boundary conditions. Wehner and Wilhelm used their analytical solutions
for a first order reaction to show that this indeed was true; Eqs. (169)
 PATTERNS OF FLOW I N CHEMICAL PROCESS VESSELS 181

and (170) are valid even with diffusion in the fore and after sections.
Bischoff (B12) later showed that this is so for any order of reaction.
Forster and Geib (F6) derived Eq. (171) by using what we now call the
distribution of residence times for a finite vessel with axial dispersion.
Figure 28 is a graphical representation of Eq. (171). The ratio of
reactor volume actually needed with dispersion to the plug-flow value is

0.001 0.01 01 1
GG
FIG.28. Comparison of performance of reactors for the plug flow and dispersed
plug flow models. Reaction is of first order, a A + products, and constant density,
occurring in a closed vessel (L14,L15).

plotted against the fraction of reactant remaining a t the outlet with the
dispersion group, DL/uL, as a parameter. It is seen that for large values
of the group DL/uL and for high conversions (low fraction of reactant
remaining), a significantly larger reactor would be needed than predicted
using the plug flow analysis. However, the dispersion model might not
be valid for large DL/uL, and so in practice, only the lower section of
the chart could be relied upon. For small DL/uL, Eq. (171) gives ap-
proximately, for equal conversions in the two reactors,
(172a)

(172b)
182 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

Carberry (C7) and Epstein (E6) discussed the magnitudes of the correc-
tions necessary for dispersion in packed beds. It was found that for many
practical cases of interest, the axial mixing effect was very small.
Levenspiel and Bischoff (L14, L15) later extended the above treat-
ment to second order reactions, and presented a chart similar to Fig.
28. Fan and Bailie (F4) presented results for reactions of order
n = 1/4, 1/2, 2, 3. They gave the complete set of concentration profiles
throughout the reactor section.
All of the preceding work was for simple, or one step, reactions. The
more interesting case of multiple reactions has been studied by de Maria
et al. (D15) and by Tichacek (T7). de Maria e t al. considered the
catalytic oxidation of n a ~ h t h a l e n e .They
~ found that the consideration
of the dispersion effects enabled them to obtain a better design. Tichacek
considered the selectivity for several different types of reactions. Natu-
rally, the results were rather complicated, and the statement of general
conclusions is rather difficult. For small values of the reactor dispersion
group, D&L <.- 0.05, i t was found that the fractional decrease in the
maximum amount of intermediate formed is closely approximated by the
value of DL/uL itself. For other ranges of the parameters, we refer to
the original work (T7).
Coste et al. (C22) considered simultaneous mass and heat dispersion
in a tubular reactor. As discussed previously (Section II1,b) they found
that the numerical computations caused some trouble, although Carberry
(C8) used a finite difference scheme that seemed to avoid the difficulties.
Hovorka and Kendall (H13) discussed reactions in a baffled vessel.
One final point that should be mentioned is that for nonlinear reaction
rates, where the distribution of residence times is not sufficient informa-
tion to predict conversions, the validity of the preceding theoretical cal-
culations is questionable. However, in view of the fact that the dispersion
model should only be relied upon for slight deviations from plug flow
(small DL/uL), this problem might not arise since the nonlinear effects in
this range would not be too important. However, not enough experimental
work has been done as of yet to determine whether or not the above
predicted results can actually be measured in a real reactor. These ex-
periments should be performed so that the necessity (or neglect) of
taking dispersion effects into account in reactor design can be determined.
b. Dispersed Plug-Flow Model. For this model, Eq. (1-4) of Table I
is used. For steady flow, the equation reduces to,

‘Their work was actually for a fluidized bed, but since they used the dispersion
model, it is discussed here.
 PATTERNS O F FLOW IN CHEMICAL PROCESS VESSELS 183

The solution of Eq. (173) poses a rather formidable task in general. Thus
the dispersed plug-flow model has not been as extensively studied as the
axial-dispersed plug-flow model. Actually, if there are no initial radial
gradients in C, the radial terms will be identically zero, and Eq. (1731
will reduce to the simpler Eq. (167). Thus for a simple isothermal
reactor, the dispersed plug flow model is not useful. Its greatest use is for
either nonisothermal reactions with radial temperature gradients or tube
wall catalysed reactions. Of course, if the reactants were not introduced
uniformly across a plane the model could be used, but this would not be
a common practice. Paneth and Herzfeld (P2) have used this model
for a first order wall catalysed reaction. The boundary conditions used
were the same as those discussed for tracer measurements for radial
dispersion coefficients in Section II,C,3,b, except that a t the wall,

-DR aC (Ro) = klC(R0)

aR
~

The solution is

where
(174b)

(174c)

The principal use of Eq. (173) is in conjunction with a similar heat

dispersion equation. Unfortunately, a system of coupled nonlinear partial
differential equations then has to be solved, which is very difficult even
with the aid of computers. I n the oxidation of sulfur dioxide, Hall and
Smith ( H l ) found relatively good agreement between theory and experi-
ment near the center of the reactor. Their calculations were based on the
heat-dispersion equation, and they did not take detailed mass dispersion
into account. Baron (B2) later solved the mass and heat dispersion equa-
tions simultaneously by a novel graphical method, and found better
agreement between his calculations and the data of Hall and Smith.
Kjaer (K9) gives a very comprehensive study of concentration and
temperature profiles in fixed-bed catalytic reactors. Both theoretical and
experimental work is reported for a phthallic anhydride reactor and
various types of ammonia converters. Fair agreement was obtained, but
due to the lack of sufficiently accurate thermodynamic and kinetic data,
definite conclusions as to the suitability of the dispersed plug flow model
could not be reached. However, the results seemed to indicate that the
184 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

model did provide a basis for the successful prediction of reactor per-
formance.
Amundson (A5) discussed the analytical solution of the heat disper-
sion equations for a packed bed chemical reactor. The form of the
differential equations is, of course, similar to the mass dispersion equa-
tions for certain cases. A wealth of analytical methods and results are
presented for various types of boundary conditions.

c. General Dispersion Model. The general dispersion model has been

used only for fully developed laminar flow in empty tubes: Chambrh
(C12, C13), Cleland and Wilhelm (ClS), Damkohler (D2), Krongelb
and Strandberg (K16), Lauwerier (L5), Schechter and Wissler (S4),
Walker (W2), and Wissler and Schechter (W11). Most of the work has
been concerned with the computations involved in solving Eq. (1-2) of
Table I with u ( R ) = 2u(Ro2- R 2 ) ,A solution has usually been obtained
by using the separation-of-variables technique to reduce the partial
differential equation to a Sturm-Liouville problem. The Sturm-Liouville
eigenfunctions were then computed by a series expansion or other numeri-
cal method. In order to use this method, the equation must be linear,
and so first-order reactions were considered. Schechter and Wissler (S4)
considered a more complicated case of a first-order photochemical reac-
tion with a position-variable photon intensity.
Cleland and Wilhelm (C18) used a finite-difference technique which
could be used for nonlinear reactions, but they limited their study t o a
first-order reaction. Experiments were also performed to test the results
of the theory. I n a small reaction tube, the two checked quite well. I n a
large tube there were differences which were explained by consideration
of natural convection effects which were due to the fact that completely
isothermal conditions were not maintained. This seems to be the only
experimental data in the literature to date, and shows another area in
which more work is needed. The preceding discussion considered only
isothermal conditions except for Chambrd (C12) who presented a general
method for nonisothermal reactors.

2. Tanks-in-Series Model

As discussed in Section 111, for small deviations from plug flow such
as those occurring in tubular and packed-bed reactors, a model consisting
of a series of tanks can be used to represent the fluid mixing. The con-
version predicted by the model can be found from the equations discussed
in the section on conversion in ideal stirred tanks. Figure 29 shows the
ratio of reactor volume needed with j stirred tanks to the volume needed
 PATTERNS OF FLOW I N CHEMICAL PROCESS VESSELS 185

with plug flow versus the fraction of reactant remaining a t the vessel
outlet for a first order reaction. It is seen that the graph is quite similar
to the one for the dispersion model, Fig. 28.
The discussion in Section II1,C showed that there was no unique way
to compare the stirred tank and dispersion models based on the tracer
curves. Each different basis of comparison gave different results. The two
models have been compared for chemical reactions by van Krevelen
(V6), Trambouae (TlO), and Levenspiel (L13a). Levenspiel used Figs.
28 and 29 to determine the correspondence of the models. His results are
shown in Fig. 30. The various criteria give results that differ increasingly
with rise in reaction order, conversion, and degree of mixing.
30

20

10

1.o
0.01 0.1 1
C/CO

FIG.29. Comparison of performance of reactors for the plug flow and tanks in
series models. Reaction is of first order, aA + products, and constant density, occur-
ring in a closed vessel (L13).

Thus it seems that the type of comparison that should be made

depends on the purpose of the model. For design of mixing vessels the
tracer curves should be matched, and for reactor design conversions
should be matched. Unfortunately, this means that a general approach
is not possible for all cases. However, the various criteria of correspond-
ence approach each other with approach to plug flow ( j + m or
DL/uL+ 0 ) , and so for many practical cases of interest a comparison
is possible. Thus, for small deviations from plug flow, either the disper-
sion model or stirred tanks model may be satisfactorily used depending
on one’s personal preference.
Deans and Lapidus (D12) have also extended their finite-stage model
186 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

to chemical reactions. As was discussed in Section III,B,2, the reason for

devising this model was to avoid, if possible, the mathematical com-
plexity of solving the coupled partial differential equations. However,
since Schechter and Wissler ( S 5 ) showed (see Section III,B,2) that the

p/u'

FIG.30. Ways of comparing the dispersed plug flow and tanks in series models (L13a).

finite stage equations were one of a family of difference-equation rep-

resentations of the partial differential equations, there may not be any
actual computational advantages. Again, all that can be said is that more
work is required in this area.

D. FLUIDIZED BED REACTORS

If the dispersion model is chosen to represent fluidized-bed behavior,
then the expressions found for packed-bed reactors and tubular reactors
 PATTERNS OF FLOW I N CHEMICAL PROCESS VESSELS 187

are applicable. These have been presented by Blickle and Kaldi (B18),
de Maria et al. (D15), Gilliland and Mason (G5), and Wicke (1378). On
the other hand, if the two-region flow models of Table I V are used, the
conversion must be determined for the specific model being used. Each
of the references given in Section IV,E should be consulted for details.
Different types of expressions will be obtained for homogeneous and
solid-catalyzed reactions, since in the catalytic case reaction does not
occur where solid is absent, Thus, for homogeneous systems, the volume-
ratio of phases is a parameter of the model, but in catalytic systems i t
is not. This fact is shown in the number of parameters tabulated for
each of the models listed in Table IV and is further discussed in (L13).
Due to lack of applications, conversion expressions for these two-
region models have not been developed for homogeneous systems. For
heterogeneous systems, which are outside the scope of this article, the
appropriate expressions can be found in the works of the individual
investigators.
VI. Other Applications

A. INTERMIXING FLOWING
OF FLUIDS IN PIPELINES
SUCCESSIVELY
A pipeline may be used to transport a variety of fluids, and in switch-
ing from one fluid to another a region of intermixing (or zone of con-
tamination) forms between the leading and following fluid. For proper
design and operation of a pipeline so as to minimize contamination, it is
necessary to be able to predict the extent of such intermixing. This was
done by Levenspiel (L11) using the dispersed plug-flow model. For the
general findings and design charts see (L11).
B. BRIEFSUMMARY OF APPLICATIONS FLOW
TO MULTIPHASE
AND OTHER HETEROGENEOUSPROCESSES
For two phase systems, deviations of flow patterns from ideality can
be more serious than for single phase systems, and thus errors in design
can be much greater. Recently, much work has been done in this area,
but the treatment is necessarily more cumbersome and difficult. It is not
within the scope of this article to deal with these cases. The following
references are presented simply to indicate the type of work being done
during the last few years.
Acrivos, A,, On the combined effect of longitudinal diffusion and external mass trans-
fer resistance in fixed bed operations. Chem. Eng. Sci. 13, 1 (1960).
Asbjornson, 0. A,, The distribution of residence times in a falling water film. Chem.
Eng. Sci. 14,211 (1961).
Bradshaw, R. D., and Bennett, C. O., Fluid-particle mass transfer in a packed bed.
A.I.Ch.E. Journal 7, 48 (1961).
188 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

de Maria, F., and White, R . R., Transient response study of gas flowing through
irrigated packing. A.I.Ch.E. Journal 6, 473 (1960).
Foss, A. S., Gerster, J. A,, and Pigford, R. L., Effect of liquid mixing on the per-
formance of bubble trays. A.Z.Ch.E. Journal 4, 231 (1958).
Gilbert, T . J., Liquid mixing on bubble-cap and sieve plates. Chem. Eng. Sci. 1%
243 (1959).
Gray, R . I., and Prados, J. W., The dynamics of a packed gas absorber by frequency
response analysis. A1.Ch.E. Journal 9, 211 (1963).
Houston, R . H., Univ. California Radiation Lab. Rept. 3817 (1958). “A Theory for
Industrial Gas-Liquid Chromatographic Columns.”
Hennico, A,, Jacques, G. L., and Vermeulen, T., Longitudinal dispersion in packed
extraction columns. Univ. California Radiation Lab. Rept. 10696 (1963).
Kada, H., and Hanratty, T. J., Effect of solids on turbulence in a fluid. A.I.Ch.E.
Journal 6, 624 (1960).
Kasten, P. R., and Amundson, N. R., An elementary theory of adsorption in fluidized
beds. Ind. Eng. Chem. 42, 1341 (1950).
Lapidus, L., Flow distribution and diffusion in fixed-bed two-phase reactors. Ind.
Eng. Chem. 49, 1000 (1957).
Mar, B. W., and Babb, A. L., Longitudinal mixing in a pulsed sieve-plate extraction
column. Ind. Eng. Chem. 51, 1011 (1959).
Miyauchi, T., and Vermeulen, T., Longitudinal dispersion in two-phase continuous-
flow operations. Ind. Eng. Chem. Fundamentals 2, 113 (1963).
Miyauchi, T., Influence of longitudinal dispersion on tray efficiency. Chem. Eng.
( To k y o ) 24, 443 (1960).
Oliver, E. D., Liquid mixing on bubble trays-a correction. A.I.Ch.E. Journal 5, 564
( 1959).
Oliver, E . D., and Watson, C. C., Correlation of bubble cap fractionating-column
plate efficiencies. A.I.Ch.E. Journal 2, 18 (1956).
Otake, T., and Kunugita, E., Mixing characteristics of irrigated packed towers. Chem.
Eng. ( To k y o ) 22, 144 (1958).
Schiesser, W.E., and Lapidus, I,., Further studies of fluid flow and mass transfer in
trickle beds. A.I.Ch.E. Journal 7, 163 (1961).
Seimes, W., and Weiss, W., FlussigBeitsdurchmischung in engen Blasensiulen. Chem.
Ing. Tech. 29, 727 (1957).
Sleicher, C. A,, Axial mixing and extraction efficiency. A.I.Ch.E. Journal 5, 145 (1959).
Sleicher, C. A,, Entrainment and extraction efficiency of mixer settlers. A.I.Ch.E.
Journal 6, 529 (1960).
Van Deemter, J. J., Zuiderweg, F. J., and Klinkenberg, A., Longitudinal diffusion
and resistance to mass transfer as causes of nonideality in chromatography. Chem.
Eng. Sci. 5, 271 (1956).
van de Vusse, J. G., Residence times and distribution of residence times in dispersed
flow systems. Chem. Eng. Sci. 10, 229 (1959).
Vermeulen, T., Separation by adsorption methods. Advances Chem. Eng. 2, 147
(1958).
Dunn, M’. E., Vermeulen, T., Wilke, C. R., and Word, T . T., Longitudinal dispersion
in packed gas absorption columns. Univ. California Radiation Lab. Rept. 10394
(1963).
Weber, H., Dissertation, Tech. Hochschule, Darmstadt, Germany, 1960; quoted in
Hofmann, Reference (H11).
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 189

VII. Recent References

A list of recent work not included in the preceding discussion is presented here.
The pertinent section in this article is indicated after the reference.
Bailey, H. R., and Gogarty, W. B., Numerical and experimental results on the dis-
persion of a solute in a fluid in laminar flow through a tube. Proc. Roy. SOC.
A269,352 (1962). (II,C,B,h; II,D,2; II,D,3)
Biggs, R . D., Mixing in continuous-flow stirred tanks. Paper presented at A.1.Ch.E.
Meeting, Los Angeles, February, 1962. (IV,D)
Carter, D , and Bir, W. G., Axial mixing in a tubular high pressure reactor. Chem.
Eng. Progr. 58, 40 (March, 1962). (II,C,2,h)
Curl, R. L , Dispersed phase mixing-theory and effects in simple reactors. A.1.Ch.E.
Journal 9, 175 (1963). (IV,D; V,B,2)
De Baun, R. M., and Katz, S., Approximations to residence time distribution in
mixing systems and some applications thereof. Chem. Eng. Sci. 16, 97 (1961).
( I ; IV)
Farrell, M. A,, and Leonard, E. F., Dispersion in laminar flow by simultaneous con-
vection and diffusion. A.I.Ch.E. Journal 9, 190 (1963). (I1,D)
Froment, G. F., Design of fixed-bed catalytic reactors based on effective transport
models. Chem. Eng. Sci. 17, 849 (1962). (VC)
Glaser, M. B., and Litt, M., A physical model for mixed phase flow through beds of
porous catalyst. A.1.Ch.E. Journal 9, 103 (1963). (I1,E)
Glaser, M. B., and Lichtenstein, I . Interrelation of packing and mixed phase flow
parameters with liquid residence time distribution. A.1.Ch.E. Journal 9, 30 (1963).
(II,E)
Gottschlich, C. F., Axial dispersion in packed beds. AJ.Ch.E. Journal 9, 88 (1963).
(II,C,2,h; II,E)
Leonard, E. F., and Ruszkay, R. J , Frequency, transient and moment methods in
process analysis. Paper presented at A.1.Ch.E. Meeting, New York, December,
1961. (I)
Miller, R. S , Ralph, J. L., Curl, R. L., and Towell, G. D., Dispersed phase mixing-
measurements in organic dispersed systems. A.Z.Ch.E. Journal 9, 196 (1963). (IV,D)
Muchi, I., Mamuro, T., and Sasaki, K., Studies on the mixing of fluid in a fluidized
bed. Chem. Eng. ( T o k y o ) 25, '747 (1961). (IV,E)
Rosenweig, R. E., Hottel, H. C., and Williams, G. C., Smokescattered light meas-
urement of turbulent concentration fluctuation. Chem. Eng. Sci. 15, 111 (1961).
(11)
Schugerl, K., Merz, M., and Fetting, F., Rheologische Eigenschaften von gasdurch-
stromten Fliessbettsystemen. Chem. Eng. Sci. 15, 1 (1961). (IV,E)
Stoyanovskii, I . M., Investigation of the longitudial transfer of iodine in various
solvents during the motion of a liquid stream through a granular glass bed. J .
A p p l . Chem. USSR (Engl. Transl.) 34, 1863 (1961). (II,C,2,h)
Yablonskii, V. S., Asaturyan, A. Sh., and Khizgilov, I . Kh., Turbulent diffusion in
pipes. Intern. Chem. Eng. 2, 3 (1962). (II,C,2,h)
Zelmer, R . G., Residence times in continuous multistage reactors. Chem. Eng. Progr.
58,37 (March 1962). (V)
190 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

Nomenclature

a =P/P, Radial dispersion coef-

a* Roots of Jl(ac) = O ficient, dispersed plug
b =P/PB flow model
C Dimensionless response Mean value of D R ( R )
curve to a pulse input, Radial dispersion coef-
defined in Section I ficient, uniform disper-
C Concentration sion model
Co Initial concentration of = D R ( R ) Radial dis-
tracer or reactant enter- persion coefficient, gen-
ing the vessel or reactor eral dispersion model in
Co Average concentration cylindrical coordinates
of tracer in system Molecular diffusivity
C,,, Integral average tracer Exit age distribution
concentration in vessel function, defined in Sec-
during steady state in- tion I
jection (dispersion Radius of injector tube
model) = E/Ro. Dimensionless
C',,, Mean concentration of radius of injector tube
pulse of tracer if uni- Dimensionless response
formly distributed in to step input, defined in
experimental section of Section I
vessel of length L = [ u ( R )- ul/u. Meas-
c = C/Co. Dimensionless ure of variation of
concentration u ( R ) from its mean
= C/C.,.. Dimension- value
less concentration = DR(R)/DR,. Meas-
c' = C/C'.,. Dimension- ure of variation of
less concentration D R ( R ) from its mean
d, Effective diameter, de- value
fined by Eq. (50) = D L ( R ) / D L ~Meas-
.
d, Particle diameter ure of variation of
dr Tube diameter D L ( R ) from its mean
D Dispersion coefficient value
DL Axial dispersion coeffi- =I(P - Pw)/Pl
cient, dispersed plug =I(c - cw/c)I
flow model Defined by Eq. (73)
D'L Axial dispersion coeffi- Internal age distribu-
cient, axial-dispersed tion function, defined in
plug, flow model shown Section I
equal to D L in Eq. (72) Injection rate of tracer
DL, Mean value of D L ( R ) Number of ideal stirred
DL,,; Axial dispersion coeffi- tanks in series
cient, uniform disper- Bessel functions
sion model Moment order, see Eq.
DL(r) = D L ( R ) Axial disper- (65)
sion coefficient, general Reaction rate constants
dispersion model in cy- = X , - Xo. Distance
lindrical coordinates between measurement
 PATTERNS OF FLOW I N CHEMICAL PROCESS VESSELS 191

points, or length of ex- e = u t / L = t / z . Dimen-

perimental section sionless time
M r kth moment of tracer u Velocity vector
distribution averaged u Mean velocity in axial
over the cross section of direction
tube, defined by Equa- u ( R ) Velocity as a function
tion (65) of radial position
N' Defined by Equations uo Mean velocity in a
(56) and (57) packed bed based on
0 The order symbol empty tube
P Laplace transform vari- U ( t ) Unit step function
able w Volumetric flow rate
P = uL/DL'. Dimension- V Volume of vessel
less parameter Va Volume of backmix
PL = u R ~ / D L .Dimension- flow region
less parameter Va Volume of deadwater
P L' = u R ~ / D L 'Dimension-
. region
less parameter Vp Volume of plug flow
P L m = u R ~ / D L ~ . Dimen- region
sionless parameter X Axial position measured
PR = uRo/DR. Dimension- from start of test sec-
less parameter tion
PRm = uRo/DBm. Dimen- x = X / R o . Dimensionless
sionless parameter axial variable
q =d(1/4) + (p/P) x = X / L . Dimensionless
Q =$(1/4) + (u<'/PLPR) axial variable
8 ( t ) Dirac delta function,
R Radial position
see reference (S20)
Ra = dr/2. Tube radius e Fraction voids in
r = R/Ra. Dimensionless packed bed
radial position x1 Eigenvalue in Eq.
To Rates of chemical reac- (67c), (68c), and ( 6 9 ~ )
tion, (moles/time- pl Mean of tracer curve
volume) at measurement point
R (s) Lagrangian correlation (dimensionless)
coefficient Apl = plm - plo Difference in means of
N R ~Reynolds number the tracer curves at
S Time difference used in the two measurement
correlation coefficient points Xm and X O
s Source term, defined by Y Kinematic viscosity of
Eq. (12) fluid
s o Schmidt number 2 Variance of tracer curve
t time at the measurement
- point (dimensionless)
t = V / v . Mean residence
time of fluid in the sys- Ao2 = urn' - uo2 Difference in variance
tem of the tracer curve at
e' = u1/Ro. - Dimensionless the two measurement
points Xo and X m
I -

time
192 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

+. Tortuosity factor measurement point or

to the second of two
SUBSCRIPT measurement points ;
a Refers to entrance or applies to X, x, and x
upstream section 0 Refers to the injection
b Refers to exit or down- point or the first of two
stream section. measurement points;
e Refers to end of test applies to X, x, and X
section; applies to X , 03 Refers to the doubly
x and x infinite tube, the open
m Refers t o the single vessel

TEXTREFERENCES
Al. Acton, F. S., and Lapidus, L., Znd. Eng. Chem. 47, 706 (1955).
A2. Adler, R. J., and Hovorka, R . B., “A Finite-Stage Model for Highly Asym-
metric Residence-Time Distributions,” Preprint 3, Second Joint Automatic
Control Conference, Denver, Colorado, June, 1961.
A3. Aiba, S., A.Z.Ch.E. Journal 4, 485 (1958).
A3a. Allen, C. M., and Taylor, E. A,, Trans. A m . Soc. Mech. Eng. 45, 285 (1923).
A4. Ampilogov, I. E., Kharin, A. N., and Kurochkina, I. S., Zh. Fiz. Khim. 32,
141 (1958).
A5. Amundson, N. R., Znd. Eng. Chem. 48, 26 (1956).
A6. Aris, R., Proc. Roy. SOC.A235 67 (1956).
A7. Ark, R., Proc. Roy. SOC.A245 268 (1958).
AS. Aris, R., Chem. Eng. Sci. 9, 266 (1959).
A9. Ark, R., Chem. Eng. Sci. 10, 80 (1959).
A10. Ark, R., Chem. Eng. Sci. 11, 194 (1959).
A l l . Aris, R., Proc. Roy. SOC.A262, 538 (1959).
A12. Ark, R., Proc. Roy. Soc. A269, 370 (1960).
A13. Aris, R., Can. J . Chem. Eng. 40,87 (1962).
A14. Aris, R., and Amundson, N. R., A.Z.Ch.E. Journal 3, 380 (1957).
A15. Ark, R., and Amundson, N. R., Chem. Eng. Sci.7, 121 (1958).
A16. Aris, R., and Amundson, N. R., Chem. Eng. Sci. 9, 250 (1958).
A17. Askins, J. W., Hinds, G . P., and Kunreuther, F., Chem. Eng. Progr. 47, 401
(1951).

B1. Baldwin, L. V., and Walsh, T. J., A.Z.Ch.E. Journal 7, 53 (1961).

B2. Baron, T., Chem. Eng. Progr. 48, 118 (1952).
B3. Bartok, W., Heath, C. E., and Weiss, M. A., A.Z.Ch.E. Journal 6, 685 (1960).
B4. Batchelor, G. I<., and Townsend, A. A,, in “Surveys in Mechanics” ( G . K.
Batchelor and R. M. Davies, eds.), p. 352. Cambridge Univ. Press, London and
New York, 1965.
B5. Beran, M. J., J . Chem. Phys. 27, 270 (1957).
B6. Bernard, R. A., and Wilhelm, R. H., Chem. Eng, Progr. 46, 233 (1950).
B7. Bilous, O., and Amundson, N. R., A.Z.Ch.E. Journal 1, 513 (1955).
B8. Bilous, O., and Piret, E. L., A.Z.Ch.E. Journal 1, 480 (1955).
B9. Bilous, O., Block, H . D., and Piret, E. L., A.Z.Ch.E. Journal 3, 248 (1957).
BlO. Bird, R. B., Stewart, W. E., and Lightfoot, E. N., “Transport Phenomena.”
Wiley, New York, 1959.
B11. Bischoff, K. B., Chem. Eng. Sci. 12, 69 (1960).
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 193
B12. Bischoff, K. B., Chem. Eng. Sci. 16, 131 (1961).
B13. Bischoff, K. B., Ph.D. Dissertation, Illinois Inst. Technol., Chicago, 1961.
B14. Bischoff, K. B., and Levenspiel, O., Chem. Eng. Sci. 17, 245, 257 (1962).
B15. Blackwell, R. J., An investigation of miscible displacement processes in capil-
laries. Paper presented at local section A.1.Ch.E. Meeting, Galveston, Texas,
October, 1957.
B16. Blackwell, R. J., Laboratory studies of microscopic dispersion phenomena.
Paper presented at A.1.Ch.E.-Soc. Petrol. Eng. Symp., San Francisco, Cali-
fornia, Dec. 6, 1959.
B17. Blackwell, R. J., Rayne, J. R., and Terry, W. M., J. Petrol. Technol. 11, 1
(1959).
B18. Blickle, T., and Kaldi, P., Chem. Tech. (Berlin) 11, 181 (1959).
B19. Bournia, A., Coull, J., and Houghton, G., Proc. Roy. SOC. A26L 227 (1961).
B20. Bosworth, R. C. L., Phil. Mag. 39, 847 (1948).
B21. Bosworth, R. C. L., Phil. Mag. 40, 314 (1949).
B22. Brothman, A., Weber, A. P., and Barish, E. Z., Chem. Met. Eng. 60, 111
(July, 1954); 50, 107 (August, 1943); 50, 113 (September, 1943).
B23. Brotz,, W., (?hem.-1ng.-Tech, 28, 165 (1956).

C1. Cairns, E. J., Ph.D. Dissertation, Univ. of California, Berkeley, 1959

C2. Cairns, E. J., and Prausnitz, J. M., Ind. Eng. Chem. 61, 1441 (1959).
C3. Cairns, E. J., and Prausnitz, J. M., A.I.Ch.E. Journal 6, 400 (1960).
C4. Cairns, E. J., and Prausnitz, J. M., A.I.Ch.E. Journal 6, 554 (1960).
C5. Cairns, E . J., and Prausnitz, J. M., Chem. Eng. Sci. 12, 20 (1960).
C6. Carberry, J. J., A.Z.Ch.E. Journal 4, 13M (1958).
C7. Carberry, J. J., Can. J. Chem. Eng. 36, 207 (1958).
C8. Carberry, J . J., and Wendel, M. M., A.I.Ch.E. Journal 9, 129 (1963).
C9. Carberry, J. J., and Bretton, R. H., A.I.Ch.E. Journal 4, 367 (1958).
ClO. Carman, P. C., “Flow of Gases Through Porous Media.” Academic Press, New
York, 1956.
C11. Carrier, G. F., Quart. Appl. Math. 14, 108 (1956).
C12. ChambrC, P., Appl. Sci. Res. A9, 157 (1960).
C13. ChambrC, P., J. Chem. Phys. 32, 24 (1960).
C14. Chandrasekhar, S., Rev. M o d . Phys. 15, 1 (1943).
C15. Cholette, A., and Blanchet, J., Can. J. Chem. Eng. 39, 192 (1961).
C16. Cholette, A., and Cloutier, L., Can. J. Chem. Eng. 37, 105 (1959).
C17. Cholette, A., Blanchet, J., and Cloutier, L., Can. J. Chem. Eng. 38, 1 (1960).
C18. Cleland, F. A., and Wilhelm, R. H., A.I.Ch.E. Journal 2, 489 (1956).
C19. Cloutier, L., and Cholette, A,, private communication (1962).
C20. Converse, A. O., A.I.Ch.E. Jo,mal 6, 334 (1960).
C21. Corrsin, S., Phys. Fluids 1, 42 (1958).
C22. Coste, J., Rudd, D., and Amundson, N. R., Can. J. Chern. Eng. 39, 149 (1961).
C23. Croockewit, P., Honig, C. E., and Kramers, H., Chem. Eng. Sci. 4, 111 (1955).

D l . Damkohler, G., 2. Elektrochem. 43, 1 (1937).

D2. Damkohler, G., in “Der Chemie-Ingenieur” Eucken, A. and Jakob, M., eds.,
Vol. 111,Part I. Akad. Verlagsges., Leipzig, 1937.
D3. Danckwerts, P. V., Appl. Sci. Res. A3, 279 (1952).
D4. Danckwerts, P. V., Chem. Eng. Sci. 2, 1 (1953).
D5. Danckwerts, P. V., Research 6, 355 (1953).
194 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

D6. Danckwerts, P. V.,Ind. Chemist 3, 102 (1954).

D7. Danckwerts, P. V., Chem. Eng. Sci. 7, 116 (1957).
D8. Danckwerts, P. V., Chem. Eng. Sci. 8, 93 (1958).
D9. Danckwerts, P. V., Chem. Eng. Sci. 9, 78 (1958).
D10. Danckwerts, P. V., Jenkins, J. W., and Place, G., Chem. Eng. Sci. 3, 26 (1954).
D11. Davidson, J. F., Farquharson, D. C., Picken, J. Q., and Taylor, D. C., Chem.
Eng. Sci. 4, 201 (1955).
D12. Deans, H. A,, and Lapidus, L., A.I.Ch.E. Journal 6, 656 (1960).
D13. Deissler, P. F., and Wilhelm, R. H., Znd. Eng. Chem. 4.6, 1219 (1953).
D14. de Josselin de Jong, G., Trans. A m . Geophys. Union 39, 67 (1958).
D15. de Maria, F., Longfield, J. E., and Butler, G., Znd. Eng. Chem. 53, 259 (1961).
D16. Denbigh, K. G., Trans. Faraday SOC.40, 352 (1944).
D17. Denbigh, K. G., Trans. Faraday SOC.43, 648 (1947).
D18. Devyatov, B. H., and Bogatchev, A. N., Zh. Prikl. Khim. 24, 1156 (1951).
D19. Dhanak, A. M., AJ.Ch.E. Journal 4, 190 (1958).
D20. Dorweiler, V. P., and Fahien, R. W., A.I.ChB. Journal 5, 134 (1959).

E l . Ebach, E. A., and White, R. R., A.I.Ch.E. Journal 4, 161 (1958).

E2. Eguchi, W., Kagaku Soshi 2, 43 (1960).
E3. Einstein, H. A., Dissertation, Eidg. tech. Hochschule, Zurich, Switzerland,
1937.
E4. Elder, J. W., J. Fluid Mech. 5, 544 (1959).
E5. Eldridge, J. W., and Piret, E. L., Chem. Eng. Progr. 46, 290 (1950).
E6. Epstein, N., Can. J . Chem. Eng. 36, 210 (1958).
E7. Evans, R. B., Watson, G. M., and Mason, E. A,, J . Chem. Phys. 35, 2076
(1961).

F1. Fahien, R. W., Ph.D. Dissertation, Purdue University, Lafayette, Indiana,

1954.
F2. Fahien, R. W., and Smith, J. M., A1.Ch.E. Journal 1, 28 (1955).
F3. Fan, L., Ind. Eng. Chem. 52, 921 (1960).
F4. Fan, L., and Bailie, R. C., Chem. Eng. Sci. 13, 63 (1960).
F5. Flint, D. L., Kada, H., and Hanratty, T. J., A.I.Ch.E. Journul 6, 325 (1960).
F6. Forster, T., and Geib, K. H., Ann. Physik [51 20, 250 (1934).
F7. Fowler, F. C., and Brown, G. G., Trans. Am. Znst. Chem. Eng. 39, 491 (1943).
F8. Fox, E. A., and Gex, V. E., A.I.Ch.E. Journal 2, 539 (1956).
F9. Froment, G. F., Belg. Chem. Ind. 24, 619 (1959).

GI. Giddings, J . C., J . Chem. Phys. 26, 169 (1957).

G2. Giddings, J. C., and Eyring, H., J. Chem. Phys. 59, 416 (1955).
G3. Gilles, E . D., and Hofmann, H., Chem. Eng. Sci. 15, 328 (1961).
G4. Gilliland, E. R., and Mason, E. A., Ind. Eng. Chem. 41, 1191 (1949).
G5. Gilliland, E. R., and Mason, E. A., Znd. Eng. Chem. 44, 218 (1952).
G6. Gilliland, E. R., Mason, E . A., and Oliver, R. C., Ind. Eng. Chem. 45, 1177
(1953).
G7. Gomezplata, A., and Shuster, W. W., A.I.Ch.E. Journal 6, 454 (1960).
G8. Greenhalgh, R. E., Johnson, R. L., and Nott, H. D., Chem. Eng. Progr. 65,
44 (February, 1959).
G9. Gutoff, E. B., A.I.ChE. Journal 6, 347 (1960).
 PATTERNS OF FLOW I N CHEMICAL PROCESS VESSELS 195

H1. Hall, R. E., and Smith, J. M., Chem. Eng. Progr. 45, 459 (1949).
H2. Ham, A,, and Coe, H. S., Chem. M e t . Eng. 19, 663 (1918).
H3. Handlos, A. E., Kunstman, R. W., and Schissler, D. O., Ind. Eng. Chem. 49,
25 (1957).
H4. Hanratty, T . J., Latinen, G. A., and Wilhelrn, R . H., A.I.Ch.E. Journal 2, 372
(1956).
H5. Harai, E., Chem. Eng. ( T o k y o ) 18, 528 (1954).
H6. Hartrnan, M. E., Wevers, C. J. H., and Kramers, H., C h i m . Eng. Sci. 9, 80
(1958).
H7. Hawthorn, R. D., A.I.ChE. Journal 6, 443 (1960).
H8. Hiby, J . W., and Schiimmer, P., Chem. Eng. Sci. 13, 69 (1960).
H9. Hinze, J. O., “Turbulence.” McGraw-Hill, New York, 1959.
H10. Hofmann, H., quoted in Schoenemann, K., Chem. Eng. Sci. 8, 161 (1958).
H11. Hofmann, H., Chem. Eng. Sci. 14, 193 (1961).
H12. Horn, F., and Kiichler, L., Chem.-Ing.-Tech. 31, 1 (1959).
H13. Hovorka, R. B., and Kendall, H. B., Chem. Eng. Progr. 56, 58 (August, 1960).
H14. Hulhurt, H., Ind. Eng. Chem. 36, 1012 (1944).
H15. Hull, D. E., and Kent, J. W., Ind. Eng. Chem. 44, 2745 (1952).
H16. Huntley, A. R., Glass, W., and Heigl, J. J., Ind. Eng. Chem. 53, 381 (1961).
H17. Hyman, D., and Corson, W. B., Ind. Eng. Chem. Process Design Develop. 1,
92 (1962).

J1. Jacques, G. L., and Vermeulen, T., Univ, California Radiation Lab. Rept.
8029 (1957).
52. Jenney, T. M., Chem. Eng. 62, 198 (December, 1955).
53. Johnson, J. P., and Edwards, L. J., Trans. Faraday Soc. 45, 286 (1949).
54. Jones, R. W., Chem. Eng. Progr. 47, 46 (1951).
55. Jungers, J. C., Balaceanu, J. C., Coussemant, F., Eschard, F., Giraud, A.,
Hellin, M., Leprince, P., and Limido, G. E., “CinCtique chimique appliquCe.”
Technip, Paris, 1958.

K1. Kampd de Fdriet, J., Ann. Soc. Sci. Bruxelles Ser. I., 59, 145 (1939).
K2. Kandiner, H. J., Chem. Eng. Progr. 44, 383 (1948).
K3. Kata, S., Chem. Eng. Sci. 9, 61 (1958).
K4. Kawamura, S.,Suzuki, E., and Saito, Y., Chem. Eng. (Tokyo) 21, 91 (1957).
K4a. Keyes, J . J., A.I.Ch.E. Journal 1,305 (1955).
K5. King, G. W., Ind. Eng. Chem. 43, 2475 (1951).
K6. Kirillov, N. I., Zh. Priklad. Khim. 13, 978 (1940).
K7. Kirillov, N. I., Zh. Priklnd. Khim. 18, 381 (1945).
K8. Kirillov, N. I., and Smirnova, V. P., J . Appl. Chem. USSR (Engl. Transl.1
18, 381 (1945) ; Z h . Priklad. Khim. 18, 393 (1945).
K9. Kjaer, J., “Measurement and Calculation of Temperature and Conversion in
Fixed-Bed Catalytic Converters.” Gjellerups Forlag, Copenhagen, 1958.
K10. Klinkenberg, A., J . Phys. Chem. 59, 1184 (1955).
K11. Klinkenberg, A., and Sjenitzer, F., Chem. Eng. Sci. 5, 258 (1956).
K12. Klinkenberg, A,, Krajenbrink, H. J., and Lauwerier, H. A., Ind. Eng. Chem.
45, 1202 (1953).
K13. Kohl, J., and Newacheck, R. L., Petrol. Engr. 26, D13 (1954).
K14. Kramers, H., and Alberda, G., Chem. Eng. Sci. 2, 173 (1953).
196 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

K15. Kramers, H., Baars, G. M., and Knoll, W. H., Chem. Eng. Sci. 2, 35 (1953).
K16. Krongelb, S., and Strandberg, M. W. P., J . Chem. Phys. 31, 1196 (1959).

L1. Lamb, D. E., Manning, F. S., and Wilhelm, R. H., A.Z.Ch.E. Journal 6, 682
(1960).
L2. Lanneau, K. P., Trans. Znst. Chem. Engrs. (London) 38, 125 (1960).
L3. Lapidus, L., and Amundson, N. R., J . Phys. Chem. 56, 984 (1952).
L4. Latinen, G. A., Ph.D. Dissertation, Princeton Univ., Princeton, New Jersey,
1957.
L5. Lauwerier, H. A., A p p l . Sci. Res. A8, 366 (1959).
L6. Leclerc, V. R., Chem. Eng. Sci. 2, 213 (1953).
L7. Lee, J. C., Chem. Eng. Sci. 12, 191 (1960).
L8. Lessells, G. A., Chem. Eng. 64, 251 (August, 1957).
L9. Leva, M., “Fluidization.” McGraw-Hill, New York, 1959.
L10. Levenspiel, O., Znd. Eng. Chem. 50, 343 (1958).
L11. Levenspiel, O., Petrol. Refiner 37, 191 (1958).
L12. Levenspiel O., Can. J . Chem. Eng. 40, 135 (1962).
L13. Levenspiel, O., “Chemical Reaction Engineering.” Wiley, New York, 1962.
L13a. Levenspiel, O., Chem. Eng. Sci. 17,575 (1962).
L14. Levenspiel, O., and Bischoff, K. B., Znd. Eng. Chem. 51, 1431 (1959).
L15. Levenspiel, O., and Bischoff, K. B., Znd. Eng. Chem. 53, 314 (1961).
L16. Levenspiel, O., and Smith, W. K., Chem. Eng. Sci. 6, 227 (1957).
L17. Lewis, W. K., Gilliland, E. R., and Glass, W., A.Z.Ch.E. Journal 5, 419 (1959).
L18. Liles, A. W., and Geankoplis, C. J., A.Z.Ch.E. Journal 6, 591 (1960).
L19. Longwell, J. P., and Weiss, M. A,, Ind. Eng. Chem. 45, 667 (1953).
L20. Lynn, S., Corcoran, W. H., and Sage, B. H., A.Z.Ch.E. Journal 3, 11 (1957).

MI. MacDonald, R. W., and Piret, E. L., Chem. Eng. Progr. 47, 363 (1951).
M2. MacMullin, R. B., and Weber, M., Trans. A m . Znst. Chem. Engrs. 31, 409
(1935).
M3. MacMullin, R. B., and Weber, M., Chem. Met. Eng. 42, 254 (1935).
M4. McHenry, K. W., and Wilhelm, R. H., A.Z.Ch.E. Journal 3, 83 (1957).
M5. Mason, D., and Piret, E. L., Znd. Eng. Chem. 42, 817 (1950).
M6. Mason, D., and Piret, E. L., Znd. Eng. Chem. 43, 1210 (1951).
M7. Mathis, J. F., and Watson, C. C., A.I.Ch.E. Journal 2, 518 (1956).
M8. May, W. G., Chem. Eng. Progr. 65, 49 (1959).
M9. Metzner, A. B., and Pigford, R. L., in “Scale-Up in Practice” (R. Fleming, ed.),
p. 16. Reinhold, New York, 1958.
M10. Metzner, A. B., and Taylor, J. S., A.Z.Ch.E. Journal 6, 109 (1960).
M11. Mickelson, W. R., Natl. Advisory Comm. Aeron. Tech. Note 3570 (1955).
M12. Mott, R. A,, in “Some Aspects of Fluid Flow.” (H. R. Lang, ed.) Arnold, Lon-
don, 1951.
M13. Muchi, I., Mamuro, T., and Sasaki, K., Chem. Eng. (Tokyo) 25, 747 (1961).

N1. Nagata, S., Eguchi, W., Inamura, T., Tanigawa, K., and Tanaka, T., Chem.
Eng. (Tokyo) 17, 387 (1953).
N2. Nagata, S., Eguchi, W., Kasai, H., and Morino, I., Chem. Eng. (Tokyo) 21,
784 (1957).
N3. Norwood, K. W., and Metzner, A. B., A.Z.ChB. Journal 6, 432 (1960).
 PATTERNS OF FLOW IN CHEMICAL PROCESS VESSELS 197

01. Opfell, J. B., and Sage, B. H., Advan. Chem. Eng. 1, 241 (1956).

P1. Pai, S.,“Viscous Flow Theory,” Vol. 11. Van Nostrand, Princeton, New Jersey,
1957.
P2. Paneth, F., and Herzfeld, K. F., 2. Elektrochem. 37, 577 (1931).
P3. Pansing, W. F., A.I.Ch.E. Journal 2, 71 (1956).
P4. Pearson, J. R. A., Chem. Eng. Sci. 10, 281 (1959).
P5. Petersen, E. E., A.I.Ch.E. Journal 6, 488 (1960).
P6. Plautz, D. A., and Johnstone, H. F., A.I.Ch.E. Journal 1, 193 (1955).
P7. Prager, S., Chem. Eng. Progr. Symp. Ser. 25, 55, 11 (1959).
P8. Prager, S., J . Chem. Phys. 33, 122 (1960).
P9. Prausnitz, J. M., Ph.D. Dissertation, Princeton University, Princeton, New
Jersey, 1955.
P10. Prausnitz, J. M., A.I.Ch.E. Journal 4, 14M (1958).
P11. Prausnitz, J. M., and Wilhelm, R. H., Rev. Sci. Znstr. 27, 941 (1956).
P12. Prausnitz, J. M., and Wilhelm, R. H., Ind. Eng. Chem. 49, 979 (1957).

R1. Ranz, W. E., Chem. Eng. Progr. 48, 247 (1952).

R2. Reman, G. H., Chem. Ind. (London) p. 46 (1955).
R3. Roberts, P. H., J . Fluid Mech. 11, 257 (1960).
R4. Robinson, W. G., Univ. California Radiation Lab. Rept. 9193 (1960).
R5. Rosen, J. B., and Winsche, W. E., J . Chem. Phys. 18, 1587 (1950).

S1. Saffman, P. G., J . Fluid Mech. 6,321 (1959).

52. Saffman, P. G., Chem. Eng. Sci. 11, 125 (1959).
53. Saffman, P. G., J . Fluid Mech. 7, 194 (1960).
S4. Schechter, R. S., and Wissler, E. H., Appl. Sci. Res. A9, 334 (1960).
S5. Schechter, R. S.,and Wissler, E. H., private communication (1961).
S6. Scheidegger, A. E., “The Physics of Flow Through Porous Media.” Macmillan,
New York, 1960.
57. Scheidegger, A. E., Can. J . Phys. 39, 1573 (1961).
58. Schlinger, W. G., and Sage, B. H., Ind. Eng. Chem. 45, 657 (1953).
S9. Schoenemann, K., Dechema. Monograph 21, 203 (1952).
S10. Schoenemann, K., and Hofmann, H., Chem.-hg.-Tech. 29, 665 (1957).
S11. Schwartz, C. E., and Smith, J. M., Ind. Eng. Chem. 45, 1209 (1953).
512. Shen, C. Y., and Johnstone, H. F., A1.Ch.E. Journal 1, 349 (1955).
513. Sherwood, T. K., Chem. Eng. Progr. 51, 303 (1955).
514. Sherwood, T. K., and Woerts, B. B., Ind. Eng. Chem. 31, 1034 (1939).
515. Shipley, J. R., Pipe Line News 23, 31 (December, 1951).
S16. Sinclair, G. G., A.I.Ch.E. Journal 7, 709 (1961).
517. Singer, E., Todd, D. B., and Guinn, V. P., Ind. Eng. Chem. 49, 11 (1957).
S18. Sjenitzer, F., Petrol. Engr. 30, D31 (1958).
S19. Smith, S. S., and Schultz, R. K., Petrol. Engr. 19, 94 (1948); 20, 330 (1948).
S20. Sneddon, I. N., “Fourier Transforms.” McGraw-Hill, New York, 1951.
S21. Spaulding, D. B., Chem. Eng. Sci. 9, 74 (1958).
522. Standart, G. L., Ind. Eng. Chem. 48, 1228 (1956).
S23. Stead, B., Page, F. M., and Denbigh, K. G., Discussions Faraday SOC.2, 263
(1947).
S24. Strang, D. A., and Geankoplis, C. J., Ind. Eng. Chem. 60, 1305 (1958).
198 OCTAVE LEVENSPIEL AND KENNETH B. BISCHOFF

T1. Taylor, G. I., Proc. London Math. SOC. 20, 196 (1921).
T2. Taylor, G. I., Proc. Roy. SOC.
A219, 186 (1953).
T3. Taylor, G. I., Appl. Mech. Rev. 6, 265 (1953).
T4. Taylor, G. I., Proc. Roy. SOC.A223, 446 (1954).
T5. Taylor, G. I., Proc. Roy. SOC.A225, 473 (1954).
T6. Taylor, G. I., Proc. Phys. SOC.(London) B67, 857 (1954).
T7. Tichacek, L. J., Selectivity in experimental reactors. Paper presented at
A.1.Ch.E. Meeting, San Francisco, California, 1959.
T8. Tichacek, L. J., Barkelew, C. H., and Baron, T., A.Z.Ch.E. Journal 3, 439
(1957).
T9. Towle, W. L., and Sherwood, T. K., Ind. Eng. Chem. 31, 457 (1939).
T10. Trambouze, P. J., Rev. Inst. Franc. Petrole Ann. Combust. Liquides 15, 1948
(1960).
T11. Trambouze, P. J., Genie Chim. 84, 189 (1960).
T12. Trambouze, P. J., and Piret, E. L., A.I.Ch.E. Journal 5, 384 (1959).
T13. Trawinski, H., Chem.-Ing.-Tech. 23, 416 (1951).
T14. Turner, G. A., Chem. Eng. Sci. 1, 156 (1958).
T15. Turner, G. A., Chem. Eng. Sci. 10, 14 (1959).

V1. Van Deemter, J. J., Chem. Eng. Sci. 13, 143 (1961).
V2. Van Deemter, J. J., Chem. Eng. Sci. 13, 190 (1961).
V3. Van Deemter, J. J., Broeder, J. J., and Lauwerier, H. A,, Appl. Sci. Res. AS,
374 (1956).
V4. Van der Laan, E. T., Chem. Eng. Sci. 7, 187 (1958).
V5. Van de Vusse, J. G., Chem. Eng. Sci. 4, 178, 209 (1955).
V6. Van Krevelen, D. W., Chem.-Ing.-Tech. 30, 553 (1958).
V7. Von Rosenberg, D. V., A.1.Ch.E. Journal 2, 55 (1956).

W1. Wakao, N., Oshima, T., and Yagi, S.,Chem. Eng, ( T o k y o ) 22, 786 (1958).
W2. Walker, R. E., Phys. Fluids 4, 1211 (1961).
W3. Weber, A. P., Chem. Eng. Progr. 49, 26 (1953).
W4. Wehner, J. F., and Wilhelm, R. H., Chem. Eng. Sci. 6, 89 (1956).
W5. Westhaver, J. W., Ind. Eng. Chem. 34, 126 (1942).
W6. Westhaver, J . W., Ind. Eng. Chem. 39, 706 (1947).
W7. Westhaver, J. W., J. Res. Natl. Bur. Std. 38, 169 (1947).
W8. Wicke, E., Chem. Eng. Sci. 6, 160 (1957).
W9. Wicke, E., and Trawinski, H., Chem.-Ing.-Tech. 25, 114 (1953).
W10. Wilhelm, R. H., Chem. Eng. Progr. 49, 150 (1953).
W11. Wissler, E. H., and Schechter, R. S.,Appl. Sci. Res. A10, 198 (1961).
W12. Wood, J. C., Whittemore, E. R., and Badger, W. L., Chem. M e t . Eng. 27,
1176 (1922).

Y1. Yagi, S., and Miyauchi, T., Chem. Eng. ( T o k y o ) 17, 382 (1953).
Y2. Yagi, S., and Miyauchi, T., Chem. Ens. (Tokyo) 19, 507 (1955).
Y3. Young, E. F., Chem. Eng. 64, 241 (1957).

Z1. Zenz, F. A., and Othmer, D. F., “Fluidization and Fluid Particle Systems.”
Reinhold, New York, 1960.
22. Zwietering, T. N., Chem. Eng. Sci. 11, 1 (1959).