Professional Documents
Culture Documents
Fluid Dispersion-Generalization and Comparison of Mathematical Models-I Generalization of Models
Fluid Dispersion-Generalization and Comparison of Mathematical Models-I Generalization of Models
Fluid Dispersion-Generalization and Comparison of Mathematical Models-I Generalization of Models
Abstract-Many models have been used to characterize dispersion of fluids in flowing systems. Some
account for both transverse and longitudinal dispersion, while others account for longitudinal dispersion
alone. Inaddition, thevariety of experimental tracer methods used to find the parameters of the models
makes the resultant analyses seemingly unrelated.
We show here the relationship between these models and generalize all the previously presented
measurement techniques.
This gives a comprehensive picture of the field of dispersion, allows a quantitative evaluation of the
error incurred when a simple model is used in place of a more exact but more unwieldy model, and gives
future experimenters the necessary information to decide what combination of experimental set-up and
mathematical model yields, within prescribed limits of error, an analysis of desired accuracy.
THE field of fluid mixing in chemical processing data. LEVENSPIEL and SMITH [2] showed that the
vessels has received much attention in recent years. variance or second moment of a measured concen-
The most popular methods for handling the mixing tration curve resulting from the injection of a pulse
have been based on models that use diffusion of tracer across a plane of the flowing system could
equations with modified diffusion coefficients. These be used for this purpose. In their solution they
are called dispersion models, and the coefficients are assumed a perfect delta-function input and a doubly
called dispersion coefficients. Table 1 lists the mathe- infinite system (Table 2a). The equations relating
matical equations corresponding to the various the axial dispersion coefficient and the (dimension-
dispersion models from the most general and less) mean and variance are also given there. VAN
unwieldy [l] to the most restricted but most DEX LAAN [3] extended this treatment for finite
easily handled. Equations I.4 and I.5 are the most vessels (Table 2b) using boundary conditions as
commonly used because of their simplicity. discussed by WEHNER and WILHELM [4]. Since a
The information necessary to use the various perfect delta-function input is impossible to achieve
equations in Table 1 for design work are values of the experimentally, ARIS [5], corrected by BISCHOFF [6],
dispersion coefficients. The usual way of finding showed that with the use of two measuring points
these coefficients is through tracer injection experi- this restriction could be removed (Table 2c, d).
ments. This paper will be concerned with methods of
calculating dispersion coefficients from tracer 1. Extensions and generalizations
injection data, and the second paper will show the These previous workers considered many different
interrelationships between the various models. cases, but they all assumed that the measurements
could be taken inside the vessel. We shall consider
the case when measurements are taken outside the
A. AXAL DISPERSED PLUG FLOW MODEL
vessel. This case along with the previous ones should
We shall be concerned here with only pulse or cover all possible physical set-ups of interest.
delta-function injections. Other methods such as Table 2(e) and (f) shows these cases.
step or periodic injections have been used, but the Since the injection point is not important, it is
pulse method is probably the most convenient for convenient to base the dimensionless quantities on
calculating dispersion coefficients from experimental the length between measurement points rather than
t Presently at Dept. Chem. Eng., Univ. of Texas, Austin
.
245
K. B. BEHOFF and OCTAVE
LEVENSPIEL
Table11 Expermeniol schemes used m relotlon to the owol dispersed plug flow model
Experimental scheme
any Input
d)
x=x.
7,, ,, /
-al x-x, x:x,
f
/; / /I/ Aris[S]ond EtschOff b]
L
any Input
---I
e) x:0 x=x.
Y////I AWL from equ (31
; xy,,j This work
A&m equ (41
1
any input ’
III
(0
s
x=0 x=x.
4,u fromequl51
4r’trom equ (6) Th,s work
and we call this cc,. The solution of equations (1) + 4(1 - a)Px, exp[P,x,] -
and (2) is accomplished by the use of Laplace - 4(1 - a’) exp[P,x,] +
transforms following the same scheme as given in
+ 4(1- b2) exp[-Px,] x
[5] and [6]. For details see [7]. Thus, for the second
measurement point within the test section, or x (exp[P2bl- e~~C%J)+
0 5 x, I x, + 2(1 - a)(1 - b) exp[-Px,] x
x (1 - exp[2P,x,]) + (1 - a)2 exp[2POxO] +
A~=/.J,,,-,u,,=l+ - expCPaxol)
+
+ 4(1- b) exp[-fi,] x
l-b
+ p expCP(x, - x31x x {P(x, exd3%J-
- ~3
x (exp[P,x, - Pxm] - 1)
- P(x,- ~3expCPa4+
(3)
Acr = 0; - 0:
+ (1 - b)2 exp[ - 2Px,] x
5a2 + 2(1- a)Px, + x (exp[2Pge] - exp[2Px,l) (4)
247
and OCTAVELEVENSPIEL
K. B. BISCHOFF
For the second measurement point in the exit type of measurement is to be accurately made, the
section, or x, 2 x, full equation must be used. Equation (4) in its
general form is not too useful since usually if one
A,u = pm - ,no = 1+ 9 (1 - exp[P,x,]) - measurement can be taken inside the vessel then so
can both. Therefore, the general form of equa-
tion (4) in which one measurement is inside and the
- 7 (1 - exp[P,x, - Px,]) (5)
other outside is included only for completeness.
However, a special case of equation (4) is of
interest. This is the case when we take both measure-
ments inside the test section; i.e. a = 1 (Table 2d).
5a2 + 2b + 3b2 + This type of experiment is the best one to use if at all
possible, as will be discussed shortly. If we set
+ 2(1 - b)P(x, - x,) + a = 1, we find
+ 2(1 - a)Pxo +
Ap==,,,--~~=l-- J$ (1 - exp[ -P]) x
-t- 4(1 - a)(Px, - 1 - a) exp[P,x,] +
x exp[P(x, - xJ1 (7)
+ 4(1 - b){ 1 + b - P(x, - x,)} x
2 l-b
x exp[P,x, - PxJ + A.02= 0; - iTg= P -t- p2 exp[P(x, - x3] x
248
Fluid dispersion-generalizationand comparisonof mathematicalmodels-I. Generalizationof models
difference in variances. This problem will be con-
sidered presently. There is another reason why the
measurements should be made such that an infinite
vessel may be assumed, if at all possible. Theoreti-
cally, the end effects could be corrected for by the
use of equations (4), (6) or (8). However, these
equations were all derived assuming no special flow
characteristics at the boundary of the different
sections. Actually, complex flow patterns do exist
at such locations. These may be caused by obstruc-
tions such as flanges, reducers and elbows, among
other things. In packed beds the change from a
packed tube to an empty tube would probably
cause some swirling to occur. Since these effects
cannot be handled mathematically at the present 0.01
o., 0’2 0.4 0.6 00 I.0 z.0
249
0 0. I 0.2 0.3 04
(5 - ‘J
the end of the vessel, then the error again rapidly axial dispersed plug flow model. In the present case,
rises to appreciable values. So in order to avoid a however, the method must be modified in order to
large error, it is necessary to have the measurement keep both axial and radial dispersion in the equa-
device inside the vessel, although usually the distance tions. It will be recalled that the experimental
does not have to be very large. In other words, the equipment is supposed to consist of three parts : the
disturbance right at the boundary may be quite large, upstream section X = - co to X = 0, the test section
but it is not propagated very far into the vessel. X = 0 to X = X,, and the downstream section
Figs. 3 and 4 are cross plots made for convenience. X = X, to X = + co. In the present derivation
the injection takes place at X,(0 < X,, I XJ and
the measurement is at X,(X, -C X,,,).
B. DISPERSEDPLUG FLOW MODEL
Equation (1.4) in dimensionless form is written
For certain systems a knowledge of radial dis- for each section, and thus the boundary value
persion is needed in addition to axial dispersion. problem to be solved is:
Tracer experiments are again used to determine the 2
radial dispersion coefficients. However, since we $%+_-cypEp),
inject over a plane, as was done for the axial measure- XI0 (lOa)
ments, but rather on the axis of the tube. Many i a2c i 1 a ai ac
pg+---r---= -&x-x&f(r),
previous investigators have shown how to utilize L PR ri% ar ax
I.0
04
0,s
0..
04
0.1
0.0s
0.06
0.0.
P
Fxci. 3. Distance from measurlment point to end of vessel FIG. 4. Error in assuming an infinite vessel when
for a 1 percent error in assuming an Smite vessel. measurement is made at end of vessel.
TableDI Experimental scheme used in relation to the dispersed plug flaw model
I
Where first
Experimental scheme Results
derived
Towle and
Sherwood [II
-_
Restriciion D,=D,
Eermrrd and
Wilhelm [a]
Klinkenberg
et at [IO]
Fahien and
Smith [S]
This work
Eauotion (I 20) and (I 2~1
251
K.B.BI~CHOFFand OCTAVELEVENSPIEL
and the rest of the boundary conditions the same as and equation (111.4) follows. Thus their result is a
equation (2) except that equation (2b) becomes special case of equation (12).
c, PL and x throughout. Equation (11) means The result of KLINKENBERG may be obtained
etal.
c(- co) = finite, and c, P and x are replaced by by specifying the condition for a doubly infinite
physically that no mass is transferred through the tube, DLa = DLb = D, and DRo = DRb = DR or
solid pipe wall. All of the other boundary conditions PLa = PLI, = PL and PRa = PRL = PR, giving for
have been discussed previously [4]. These simul- equation (12b) or (12~)
taneous partial differentialequations were solved by a Nim= ewC(*- dP~(x,- X011 (13)
combination of finite Hankel [12] and Laplace
In addition for a point source at the origin x0 = 0
transforms for the two cases of interest: with the
and e + 0. If these are substituted into equation (12)
measurement point within the test section and in the
we get equation (111.3), so it is also included in the
exit section. For details see [7]. The solutions
general equation. BERNARD and WILHELM'S result
evaluated at the measurement point X = X,,, are
(equation 111.2)is a special case of the equation of
Jo(V) Ji(%e) N,
KLINKENBERG et al., so it is also included. Thus
c=l+ c r- equation (12) is a general solution which incorporates
ai>~ JO(ai> 4w ’
252
Fluid dispersion-generalization and comparison of mathematical models-I Generalization of models
- x718)1) (16)
1 +q~exp[-2qPLxo]) (15)
a
where xo is the (dimensionless) distance from the beginning FIG. 6. Error in assuming an infinite vessel when radial
of the vessel to the injection point. Similarly, for a vessel that measurements are made at end of vessel.
253
K. B. BLWHOFFand OCTAVELEVENSPIEL
vessel that were not really taken into account, so a fL= l(P - PmYPI
fR = ICC- cm)lcl
safety factor should be used for the practical appli- h Defined by equation (7b) of Part II
cation of these criteria. I Injection rate of tracer
In order to get an’idea of the magnitude of the Jo, JI Bessel functions
k Moment order, see equation (4) of
errors involved, Fig. 6 was prepared which shows the Part II
error in assuming a doubly infinite vessel for measure- k’ = k”Ro/v Dimensionless parameter for first-
ment at the end of the vessel, (x, - x,,,) = 0. As the order reaction
k” First-order rate constant
experimental section, (x,, - x0), gets shorter, the L = X, - x0 Distance between measurement
error caused by end effects gets larger, as is to be points, or length of experimental
expected. section
mk k-th moment of tracer distribution
averaged over the cross-section of
NOTATION tube, defined by equation(4) of Part11
a = P/P,
Defined by equations (12b) and (12c)
Roots of h(ar) = 0 0 The order symbol
p = vLID’L Dimensionless parameter
b =P/P:
PL = VRO/DL Dimensionless parameter
c Tracer concentration
P’r, = vRo/D’r. Dimensionless parameter
C&i Integral average tracer concentra-
tion in vessel during steady state PL,,, = vRo/D~m Dimensionless parameter
injection PR = VRO/DR Dimensionless parameter
PRCZ = VRO/DR~ Dimensionless parameter
cm Mean concentration. of pulse of
tracer if uniformly distributed in g = l/l* + (“~~pf18~
experimental section of vessel of Radial position
length L. Ro = $dt Tube radius
c = ClCM Dimensionless concentration r = R/Ro Dimensionless radial position
c = c/cm Dimensionless concentration rc Rate of chemical reaction (mole/
d, Effective diameter, defined by equa- time-volume)
tion (12) of Part II Re = d,v/v Reynolds number
Source term
dp Particle diameter
SC = yID: Schmidt number
dt Tube diameter
T Time
D Dispersion coefficient
t = vt/Ro Dimensionless time
DL Axial dispersion coefficient, dis-
persed plug flow model . t = vt/L Dimensionless time
2%
D’L Axial dispersion coefficient, axial v
Velocity vector
dispersed plug flow model V
Mean velocity in axial direction
DL~ Mean value of DL(R) v(R) Velocity as a function of radial
WL?II Axial dispersion coefficient, uniform position
dispersion model vo = EV Mean velocity in a packed bed based
Da(r) = DL(R) Axial dispersion coefficient at on empty tube
radial position r or R, general X Axial position measured from start
dispersion model in cylindrical co- of test section
ordinates x = XiRo Dimensionless axial variable
DR Radial dispersion coefficient, dis- x = X/L Dimensionless axial variable
persed plug flow model 6 Dirac delta function
Darn Mean value of DR(R) 6’ Dirac delta function in cylindrical
D’R~IZ Radial dispersion coefficient, uni- co-ordinates
form dispersion model Fraction voids in packed bed
DR(r) = DR(R) Radial dispersion coefficient at Eigenvalues in equations (6) and (7)
radial position r or R, general of Part II
dispersion model in cylindrical
P Mean of tracer curve
co-ordinates
A/J = pm - PO Difference in means of the tracer
DV Molecular diffusivity
curves at the two measurement
E Radius of injector tube
points Xm and Xo
e = E/R0 Dimensionless radius of injector
tube Kinematic viscosity of fluid
h(R) = [v(R) - VI/v Measure of deviation of v(R) from
Variance of tracer curve
its mean value measurement point
fzU0 = DR(R)/DR, Measure of variation of DR(R) from Difference in variance of the tracer
its mean value curve at the two measurement points
A(R) = DL(R)/DL~ Measure of variation of DL(R) from XOand Xm
its mean value Tortuosity factor
254
Fluid dispersion-generalization and comparison of mathematical models-I. Generalization of models
REFERENCES
BIRD R. B., STEWART W. E. and LIGHTF~~TE. N., Transport Phenomena. John Wiley, New York 1960.
:;; LEVENSPIEL 0. and Sr+nrrr W. K., Chem. Engng. Sci. 1957 6 227.
131 VANDERLAANE. TH., Chem. Engng. Sci. 1958 7 187.
141 WEHNERJ. F. and WILHELMR. H., Chem. Engng. Sci. 1956 6 89.
[51 Ares R., Chem. Engng. Sci. 1959 9 266.
I61 BI~CHOFFK. B., Chem. Engng. Sci. 1960 12 69.
r71 BI~CHOFFK. B., Ph.D. Thesis, Illinois Institute of Technology 1961.
PI BERNARDR. A. and WILHELMR. H., Chem. Engng. Progr. 1950 46 233.
r91 FAHIENR. W. and SMITHJ. M., Amer. Inst. Chem. Engng. J. 1955 128.
IlO1 KLINKENBERG A., KRAJENBRINKH. J. and LAUWERIERH. A., Zndustr. Engng. Chem. 1953 45 1202.
illI TOWLEW. L. and SHERWOOD T. K., Zndustr., Engng. Chem. 1939 31457.
1121 SNEDDON I. N., Fourier Transforms. McGraw-Hill, New York 1951.
Resume-Plusieurs mod&s mathematiques ont CtButilis& pour caracttriser la dispersion des fluides
dans des systemes en ecoulement. Certains tiemrent compte de la dispersion transversale et longitudi-
nale, tandis que d’autres ne tiem-rent compte que de la dispersion longitudinale. De plus la variete des
methodes experimentales avec traceurs, utilisQs pour la recherche des parametres rendent les analyses
qui en resultent, incoherentes.
Nous montrons ici les relations entre ces modeles et nous generalisons toutes les techniques de
mesure present& anterieurement.
Ceci donne une representation commode du champ de dispersion, permet une evaluation
quantitative de l’erreur commise quand on utilise un modele simple au lieu dun autre plus pr&is,
mais mains maniable. L’experimentateur futur aura a sa disposition les don&es necessaires pour le
choix de la combinaison montage experimental-modele mathematique qui donnera une analyse de
precision determinee dans la limite d’erreur permise.
255