You are on page 1of 12

," ,"

1 to 7 atm. Acta. Chirn. Hung. 53 (1963b).

gen Nitrogen, Carbon Monoxide and Their Binary Mixtures on Adsorption of Gases and Gas Mixtures. II: Investigation of the
, H2 (1981).
'ed mixtur . Sieve Type lOX " J. Chern. Eng. Data
. Mo 1eeU n
Y lar Adsorption Equilibria of Binary Gas Mixtures Acta. Chirn. Hung. , 35,
nt i Sand R. P. Danner A Gas Adsorption Isotherm Equation
SU! AIChE J., 26, 68 (1980a). 245 (1963c).
y solution t: Va ancy Solution Theory, Wilson , R. J. and R. P. Danner Adsorption of Synthesis Gas- Mixture
1)u 0 Yacancy Solution Theory of Adsorption
en Sand R. P. Danner
as Mi tures
frow Land
AIChE J., 26, 76 (1980b).
Y. Iles Adsorption of Gases and Gas Mixtures. 1: Mea-
Components on Activated Carbon
" J. Chern. Eng. Data , 14

szepesy, en
t of the Adsorption Isotherms of Gases on Activated Carbon up
Acta. Chirn: Hung., 37 (1963a).
" of 1000 torr.
of Gases and Gas MIxtures. II: Measurement of the Manuscrpt received August , 1983; reviio receved January 1984 and acepted
sorbed ph under Pressures of January 12.
rption Isotherms of Gases on Activated Carbon
and j
lrface phase

reen species j

as phase

Longitudinal and Lateral Dispersion in

Packed Beds: Effect of Column Length and
Particle Size Distribution
Longitudinal and lateral dispersion coefficients were measured at various axial NEUNG- WON HAN
ium and Ther- . positions in a packed bed in the Peelet number range from 10 to 10
. Three dif-
Div. Chern. ferent types of packings were used: uniform size partieles , a narrow size distribu-
tion, and a wide size distribution. For the case of uniform partieles the longitudinal R. G. CARBONELL
m Thermody-
dispersivities were found to be strong functions of position in the bed unless the
'Iv. Chern. Sci. Department of Chemical Engineering
dispersion length satisfies a constraint dependent on the value of the Peelet number.
North Carolina State University
e Adsorption," Generally, the larger the Peelet number , the larger the required length for constant
Raleigh , NC 27695
, PA (1982). axialdispersivities to be achieved. For the case of the wide size distribution, lon-
,tion of Binary gitudinal dispersivities were larger than in the uniform partiele case , and they
arbon: Experi- required a longer dispersion length to achieve a constant value. This suggests a
. the Ternary characteristic length for dispersion larger than the mean hydraulic radius. The
lateral dispersivities were found to be insensitive to the distribution of partiele
ibria of Ethane sizes or location in the bed.
m. Fund.

Monoxide- Ni- SCOPE

:n Mixtures on

The dispersion equation is used to calculate the average There have been many experimental measurements of the
,tion of Nitro-
ve Type lOX," concentration of a solute in a porous medium under flow con- longitudinal dispersion coeffcient in packed beds and its de-
ditions (Bear , 1972; Dullen , 1979). The dispersive term accounts pendence on the Peelet number (Harleman and Rumer , 1963;
a and Kinetics for the spread of the solute about the mean pulse position due Edwards and Richardson , 1968; Rifai et aI. , 1956; Carberry and
lecular Sieves to molecular diffusion and the coupling of interpartiele velocity Bretton , 1958; Ebach and White , 1958; Pfannkuch , 1963;
and concentration gradients (Whitaker , 1967; Gray, 1975). It Blackwell et aI. , 1959; Gunn and Pryce , 1969). The Peelet number
Chern. Phys. consists of the double scalar product of a dispersion tensor , or dependence of the lateral dispersion coefficient has not been
total dispersivity tensor (Brenner , 1980) with the second deriv- measured as extensively but there are some data at low Peelet
Compounds," ative of the average solute concentration in the porous medium. numbers (Harleman and Rumer , 1963; Gunn and Pryce, 1969;
For the case of an isotropic porous medium, the total dispersivity Grane and Gardner , 1961; Hassinger and von Rosenberg, 1968;
Ethane , Eth-
:tures on 13X - tensor has non-zero components only along the diagonal. The Blackwell, 1962). Implicit in these studies is the assumption that
iagonal element corresponding to the mean direction of flow the dispersion coefficients are constants that depend on the
.erm Data for IS called the longitudinal dispersion coeffcient and it contrib- value of the Reynolds and Schmidt numbers only, and are in-
es (1982). utes to the spread . of solute pulses along the flow direction. The dependent of position in the packed bed or porous medium.
Adsorption other on-diagonal elements are called the lateral dispersion However , G. I. Taylor (1953 , 1954) in his original work on dis-
, 1 319 coeffcients and they contribute to the spread of the solute in persion in capilary tubes was very careful to point out that
the directions orthogonal to the mean flow. dispersion models become valid only after a suffciently long
Ir Mixed Sur- time has passed since the insertion of the solute pulse. How long
this time is in a particular experiment was shown to depend on
lteraction be- ,
73). Conespondence concerning this paper should be addressd to R. G. Carbonell. the value of the molecular diffusivity, the tube radius , and the
uid and Liq- eung- won Han is presently at the Department of Chemical Engineering, Chon-nam National average fluid velocity. More recently, Brenner (1980) showed
mversity, Kwang-ju , Korea.
!53 (1976). ork done while authors were in the Chemical Engineering Depaent , University of California
that dispersion models are valid asymptotically in time for the
m Adsorptio Da VI.
W. case of dispersion in spatially periodic porous media , while

, No. AIChE Journal (Vol. 31 , No. February, 1985 Page 277


Carbonell and Whitaker (1983) demonstrated that this should workers interested in applying dispersion models to p
be the case for any porous medium. Unfortunately, in the design in transport processes in soils , coal gasification , or oil shaler S adjust
of most dispersion experiments in packed beds this possible torting, for example , where there are fairly large distributio values
dependence of the dispersion coefficient on time was not in- of particle sizes. Carbonell (1979; 1980a b) has made Some dispel
pr s
vestigated. The first objective of this work is to measure the axial liminary estimates of the effect of particle size distribution o prove'
dispersion coeffcient at various axial positions in a packed bed dispersion using a capilary tube model for the porous mediu an eXI
for a fairly wide range of Peclet numbers. At very low flow rates, However , with the exception of some early work of Nieman to cia!
the experiments last a suffciently long time so that the asymp- (1969), little experimental information is available on under
totic value of the dispersion coeffcient can develop. This would partiele size distributions affect dispersion coefficients. The porou
be reflected in a constant value of the axial dispersivity at var- second objective of this study is to vary systematically the dis. attent
ious positions along the column length. At very high flow rates tribution of particle sizes in a packed bed and to study the effect carefl
the experiments take a very short time and the dispersion of this change in porous medium structure on the magnitude dispel
coeffcient should be a function of axial position in the column. the longitudinal and lateral dispersion coeffcients. the ef
By analyzing these results , a quantitative criterion can be de- The data on lateral dispersion coefficients in packed beds has latera
veloped for the time required for the dispersion coeffcient to been limited in previous studies to particle Peclet numbers less deper
become constant. than 10 . The third objective of our work was to extend the range Frc
Most experiments on dispersion coefficients , both longitu- of available values of lateral dispersivities to particle Peelet equat
dinal and lateral , have been carried out with partieles with a number of 10 or greater. Both longitudinal and lateral disper. was a
very narrow size distribution. There has been little systematic sion coefficients were measured in the Peelet number range matio
effort to determine the effect of partiele size distribution on to 10 the fo
dispersion coeffcients. This information would be of use to


It was found that the longitudinal dispersivities for uniform molec
The effect of partiele size distribution on longitudinal and
size packings are a function of position in the bed unless the karasl
lateral dispersivities was studied by making up two size distri.
approximate criterion indic:
butions that had the same mean particle diameter and volume functi
fraction that was used in the experiments for uniform size par.
2: 0. tieles. The first size distribution had a ratio of maximum to withil
minimum particle diameter of 2. , the second size distribution
is satisfied. For very high Peelet numbers , one needs a larger had a corresponding ratio of 7. 3. The longitudinal dispersivities
ratio of bed length to partiele diameter in order to treat the its fin
for the uniform partiele size and the first size distribution were spatia
longitudinal dispersivity as a constant. If the criterion above nearly identical. The second size distribution exhibited longi-
is not satisfied , one observes significantly lower values of axial studi
tudinal dispersivities 2 to 3 times larger than the uniform size flow j
dispersion coeffcients that depend strongly on bed location. partieles. It was also observed that the bed length required
Some of the previous results in the literature indicate that the tube'
reach a constant value of the longitudinal dispersivity was much the c:
longitudinal dispersivity depends less strongly on the Peelet longer in the case of size distribution 2 than the case of uniform
number at very high values of than at lower values. This has partieles. Both of these results indicate that the characteristic
been attributed to turbulence effects. However , a similar phe- to tha
length for longitudinal dispersion for the case of nonuniform
nomenon is observed in this work under laminar flow conditions partiele size distributions is much longer than the hydraulic
when the dispersion distances are too short to satisfy the con- radius of the packing. This suggests the presence of larger scale.
straint above. At least some of the curvature of previous data non uniformities that strongly influence the longitudinal dis-
is believed to have resulted from this effect. persion behavior. At the present time there is not enough data wheT!
Lateral dispersivities were measured using a steady-state to provide a quantitative measure of the size of this length scale in the
experiment over a Peelet number range of 10 . This ex-
to 10 or how it correlates with the width of the partiele size distri. is the
tended the range of available lateral dispersion data which bution. The lateral dispersivities , which were measured under a ram
previously had been limited to a Peelet number of 10 . No de- steady state conditions , showed no dependence on the particle Ev,
pendence of the lateral dispersivity on bed location was ob- size distribution. theon

INTRODUCTION compared the results to available experimental data. Of course , in

soils or underground reservoirs , large scale non uniformities can 1. H:
Since the development of the dispersion approximation for the lead to much different values of dispersion coeffcients than those
study of solute transport in capilary tubes by G. 1. Taylor (1953 measured in packed beds , and for these cases spatially periodi 2. E(
1954) and R. Aris(1956), a great deal of work has been done in models cannot be expected to provide excellent results without
applying these principles to the description of solute transport in modifications. There have been other altempts at correlating and (I!
3. Ri
porous media su ch as soils and packed bed reactors (Dullien , 1979; predicting dispersion coeffcients based on a probabilistic approach
Bear , 1972). By using the method of volume or spatial averaging (I'
(Greenkorn and Kessler , 1970; Haring and Greenkorn , 1970; 4. c:
(Slattery, 1972) it is possible to derive the proper form of the Saffman , 1967; De Josselin de Jong, 1958) where the network 0
transport equation for the average concentration of solute in a pores in the porous medium is regarded as an array of cylindrical 5. EI

porous medium (Whitaker , 1967; Gray, 1975; Bear , 1972). Car- capilaries with radii , lengths , and angular orientations governed
bonell and Whitaker (1983) and Brenner (1980) have shown that by probability distribution functions. The agreement of these 6. Pf
for the case of spatially periodic models of porous media , one can models with experimental data is not complete. For example, the
have a predictive theory that can be used to calculate values of the spatially periodic models studies so far tend to overestimate the 7. B\

dispersion coeffcients without any adjustable parameters. Eidsath Peclet number dependence of axial coeffcients an 8. G,
et al. (1983) have computed axial and lateral dispersion coeffcients grossly underestimate the magnitude and Peclet number depen-
in packed beds based on these spatially periodic models , and have dence of lateral dispersivities. The probabilistic models have many

Page 278 February, 1985 AIChE Journal (Vol. 31, No. AICI
() =
() =

Is to probleDl unless a constraint of the type shown in Eq. 2 is satisfied , there are
)r oil shale r djustabl parameters , and it is diffcult to predict a priori what
the no experimental data to verify these results for real porous media.
, distributions values of the parameters should be used to represent accurately
addition to im- In principle there can be a serious error in measured values of
ade some pre. dispersion behavior in a given porous medium. In
dispersion coeffcients in packed beds if the constraint of Eq. 2 is
istribution On
proved theoretical and computational efforts, it is important to have 1(3 in Eq. 2 to be
not satisfied. This can be seen clearly if we allow
an expanded data base of dispersion coeffcients which can be used
rous mediuDl. the characteristic length of pores in a bed filed with particles of
k of Niemann to clarify and support theoretical developments and improve our
diameter and void fraction (Whitaker, 1972)
able on ho understanding of the relationship between the structure of the
ffcients. The porous medium and its dispersive properties. In this paper we focus l(3 =d (3)
ically the dis. attention on three aspects of solute dispersion which have not been 1 -
udy the effect carefully examined. The first is the possible dependence of the axial
If the distance from the inlet is L , and the interstitial velocity is
magnitude of dispersi coeffcient on position in a packed bed. The second is
(v)(3 the characteristic time for a dispersion experiment becomes
the effect of the distribution of particle sizes on longitudinal and
IltS. . Substituting these expressions in Eq. 2 , we see that we
lateral dispersion coeffcients , and the third is the Peclet number
L/ (v)
cked beds has
dependence of the lateral dispersivity.
cannot expect the dispersion coefficient in a packed bed to be a
numbers less constant unless the ratio of the distance from the inlet to the particle
From the form of the exact solution to the convective- diffusion
:end the range
equati in laminar flow in a capilary tube , Taylor (1953 , 1954) diameter is related to the Peclet number by
artiele Peclet
was able to argue quite effectively that the dispersion approxi-
ateral disper-
mation would be valid for times
after the pulse injection such that Pe (4)
nber range of 1 -
the following constraint is satisfied
:It where
Here () is a dimensionless time , R is the tube radius , and :I is the :I

molecular diffusivity of the solute in the fluid. Gil and San-

itudinal and This is a Peclet number based on 1(3 in Eq. 3 as the characteristic
karasubramaniaIl (1970) showed that for times shorter than those
;yO size distri-
length (Whitaker , 1972). For large Peclet numbers this constraint
indicated by Eq. 1 the dispersion coeffcient is an increasing
r and volume can become very diffcult to satisfy in a column of finite length.
function of time that reaches the final Taylor- Aris dispersion value In Table 1 we have summarized some of the experimental efforts
:orm size par- asymptotically. By the time () = 1 , the dispersion coeffcient is
maximum to aimed at studying the Peclet number dependence of the axial di-
within less than 1% of its asymptotic value. However , when () = spersivity. We have listed the experimental conditions as reported
e distribution 01, the dispersion coeffcient is nearly a factor of 10 lower than
dispersivities by the authors and calculated values of L/d and (/l - ) for
its final value. Paine et aL (1983) obtained similar results using the
ribution were each work.
spatial averaging technique for the case of passive dispersion , and The value of L listed in Table 1 is the largest of all the column
Ilibited longi- studied the behavior of transient dispersion coeffcients in laminar lengths used by the investigators if they used more than one
uniform size flow in tubes under conditions of adsorption and reaction at the
h required to packing length. The values of are equivalent sphere diameters
tube waIL Carbonell and Whitaker (1983) concluded that even in
'ity was much as some of the experiments involved non-s pherical particles such
the case of a general three- dimensional porous medium , the dis-
se of unifonn as Raschig rings and cylinders. All volume fractions , fluid velocities
persion coeffcient becomes constant subject to a constraint similar
:haracteristic to that for a capilary tube and particle sizes used in Table 1 came directly from the published
: nonuniform articles. The Reynolds and Schmidt numbers were computed by
he hydraulic :It (2)
estimating physical properties of the fluids used when not given
,f larger scale explicitly in the references. To get the smallest value of for a
itudinal dis- given experiment we used the smallest velocity and particle di-
enough data where 1(3 is a characteristic length associated with the pore spaces ameter in Table 1 , while the largest was computed using the
. length scale, in the fluid-solid system. Brenner (1980) was able to prove that this largest velocity and particle diameter. The characteristic length
Ie size distri- is the case exactly for a spatially periodic porous medium , using used in the definition of is the quantity 1(3 defined by Eq. 3.
asured under a random motion approach to study the solute transport. 11 is clear from
this summary that in several of these previous studies
11 the particle Even though there is a great deal of evidence acquired from the criterion expressed by Eq. 4 is either marginally satisfied , or
theoretical analyses that the dispersion coeffcients are not constant not satisfied at aIL This is especially true of the experiments of


Authors Solvent Solute L(cm) (cm) (v)13 (cm/s) L/d Pe

I -
Of course , in
formities can 1. Harleman and Water NaCI 273. 096 00608- 035- 560 849 11.02- 741
nts than those Rumer (1963) 130 21.4
ially periodic 2. Edwards and Air Argon 100. 0377- 361- 0903- 0133- 72 2 652- 165 0053- 50.
sults without Richardson 607 0.420 32. 96.
,rrelating and (1968)
istic approach 3. Rifai et aL Water NaCI 127. 025- 045 375- 005- 000807- 560 080- 822 0.452- 185
(1956) 395 1600 506
nkorn , 1970;
4. Carberry and 365- 09- 258- 858 226- 102 276- 24 X 10
Water Dye 61.3 05-
Ie network of Bretton (1958) 645 2144
of cylindrical 5. Ebach and Water Dye 152.4 021- 34- 0236- 0275- 858 257- 226. 26. 26 X 10
ons governed White (1958) 673 632 1023
lent of these 6. Pfannkuch Water NaCI 150. 0355- 34- 388 00035- 00069- 560 225- 714 199-3,405
example , the (1963) 6639
restimate the 7. Blackwell et Argon Helium 658. 021 339 023- 00298- 1.82 1.74 X 105 0028- 188
Hicients and aL (1959) 200
8. Gunn and Pryce Air Argon 26. 037- 1.8- 117 234- 88 716. 44. 120- 127.
mber depen-
(1969) 246.
ls have many

, No. AIChE Journal (Vol. 31 , No. February, 1985 Page 279

Carberry and Brelton (1958) and Ebach and White (1958), where previous experiments and to theoretical predictions. In the next
extremely large values of the Pec\et number were reached , as well section of this paper we present a brief summary of the theory of
as those of pfannkuch (1963) where the Lld ratio in some cases dispersion of inert solutes in a porous medium. This is followed by
was significantly smaller than the Pec\et number. Harleman and a detailed description of the experimental techniques and results
Rumer (1963) used a packed bed with one probe at the inlet and as well as a comparison of all the data available on longitudinal and
two other probes downstream , but did not mention any effect of lateral dispersion coeffcients in packed beds.
the probe location on the measured axial dispersion coefficients.
Rifai et al. (1956) also measured axial dispersivities at several 10-
cations in the bed and found no effect , but their experiments satisfy
the length criterion in Eq. 4 throughout the entire range of Pec\et
numbers. Carberry and Brelton (1958) found that the longitudinal
The volume-averaged form of the transport equation for the
dispersion coeffcient decreasd with bed height , but they correctly
average concentration of solute in the fluid phase takes the fo
ascribed this to a flow redistribution problem in their experiment. a (c) (3
(v)(3. "i(c)/3=D* :"i"i(c)/3 (6)
Ebach and White (1958) and pfannkuch (1963) mention that three
different column lengths were used in their experiments but ap- Coord

parently they did not make a systematic effort to study the effect where D* is the total dispersivity tensor (Carbonell and Whitaker Sy,l

of column length on the measured values of the dispersivity. In this 1983). The total dispersivity has contributions from diffusion a
paper we provide data on longitudinal dispersivities as a function well as hydrodynamic dispersion Figure

of axial position in a packed bed for Pec\et numbers ranging from

D* = - (1)(1 + 'T) + DJ + - (1)(1 + (7)
to 10 . This information is of fundamental importance in un-
derstanding some of the data acquired on longitudinal dispersion varyi
Lld does not In the equation above 'T is the tortuosity tensor
especially in those experiments where the ratio of
satisfy the criterion of Eq. 4. It should also be of a practical interest, nEdA (8)
since in the design of packed beds for catalytic reactors and other V /3 J
applications it is often very diffcult to achieve very large ratios of
column length to partiele diameter. and D is the hydrodynamic dispersion tensor Sinc!
Another aspect of dispersion in packed beds and porous media D=-(Vf/3 (9)
in general that has not received systematic attention is the effect
of the porous medium structure on total dispersivity. Of recent with the spatial deviation of the velocity from the average defined
interest has been the effect of partiele size distribution on longi-
tudinal and lateral dispersion coeffcients. This information is v- (v)(3 (10) whe
necessary in attempts to model solute transport in soils as well as colu
in reactors for shale oil retorting and coal gasification and lique- The vector is a function of position that relates the local spatial
faction (Carbonell , 1979; 1980a b). With the single exception of deviations of the point concentration from the average value to the
some work by Niemann (1969), no workers in the field have varied gradient in the average concentration so tl
the width of the distribution of particle sizes in a packed bed and (ll) pro!
then examined the effects on longitudinal and lateral dispersion.
c- (c)/3=c=E. "i(c)/3 ino
Eidsath et al. (1983) performed some calculations on spatially pe- This relation is valid for a homogeneous porous medium when a
riodic arrays of staggered cylinders of two different radii that in- suffciently long time has passed after the introduction of the solute
dicated a strong effect of particle size distribution on dispersion. pulse so that the quasi-steady assumption is satisfied is g
As the ratio of the cylinder radii went from a value of 2 to 5 , the (c)
t1) (12)
axial dispersion coeffcient increased by a factor of 1.5. In contrast,
the lateral dispersion coeffcient decreased by about the same f2 ??
amount. It would be of interest to see to what extent this phe- This assumption makes the spatial deviation field
nomenon would be observed in a real porous medium , since it and this restriction is consistent with the Aris- Taylor approximatio
would provide some insight into how one might apply axial dis- for dispersion in a capillary tube. The vector field
has been
persion data taken in beds of uniform partieles to systems with Frc
computed by Eidsath et al. (1983) for the case of a spatially periodic
highly nonuniform particle size distributions. In this paper we a fj
array of particles. This allowed a direct calculation of the tortuosity
report some observations made on the way that longitudinal and the
and hydrodynamic dispersion tensors. In the section that follows
lateral dispersion coeffcients vary as the width of the particle size we describe how one can measure the elements of the tensor
distribution is increased , but the mean particle diameter and vol- col
experimentally. However , one should keep in mind that if Eq. 12
ume fraction rema\n the same, for the Pec\et number range from thE
is not satisfied , the elements of the total dispersivity tensor cannot
102 to 104 be expected to be constant in time. This would reflect itself
The lateral dispersion coeffcients are the most diffcult com- different dispersivity values observed at different points in the
ponents of the total dispersivity tensor to measure since they require column , even though the volume fraction is constant.
a technique for measuring concentration profies in a packed bed
in a direction normal to the flow direction under either transient
or steady-state conditions. The most accurate experiments have Longitudinal and Lateral Dispersivities
been done by Harleman and Rumer (1963), in which they covered
the partiele Pec\et number range from 10 to 10 . Hassinger and
Let us consider an experiment where the velocity field is uni-
von Rosenberg (1968) measured lateral dispersion coeffcients at directional
lower Pec\et numbers from 10- to 10 , while Cunn and Pryce v) (3 = (vx /3 (13)
(1969) covered the same Pec\et number range as Harleman and
Rumer (1963) but obtained a considerably higher Pec\et number The coordinate system is shown in Figure 1. For the case of uniform
dependence on the lateral dispersivity. The theoretical models of void fraction E the volume-averaged continuity equation reduces
Brenner (1979) and Carbonell and Whitaker (1982) for spatially to the form
periodic porous media allow a priori predictions of lateral d(vx )/3 (14)
dispersivities. However , a very firm data base is not available for "i. (v)/3 = 0
comparin with experimental data. In this investigation we studied
the Pec\et number dependence of lateral dispersion coefficients so that the interstitial velocity remains constant along the flow
up to a value of the of about 10 , and compare our results to direction. At the entrance to the column we introduce a time-

Page 280 February, 1985 AIChE Journal (Vol. 31 , No.

,... ,... == p.

;. In the next
the theory of ( c) /mo (23)
s followed by ( C)
The first absolute moment is the mean residence time for the pulse
s and results
at a given axial position
gitudinal and
(X) mdmo. (24)
Substituting Eq. 20 in Eq. 22 with = 1, we find that
Il (x) (0) x/ (vx (25)
where (0) is the mean residence time for the pulse introduced
ation for the
kes the form at = O. From experimental measurements of the mean residence
time at a given axial position one can experimentally determine
the interstitial velocity. Of course , if the bed has a uniform void
fraction E , this should agree with
Coordinate Step Input Pulse Input Lateral
ad Whitaker System Experiment Experiment Dispersion Q/A
l diffusion a Experiment (Vx (26)
Figure 1. Coordinate system and schematic of longitudinal and lateral dis-
persivity experiments. where Q is the flow rate , A the column area , and is the su-
perficial velocity. The second central moment 1l2(X) is a measure
of the average pulse spread relative to the mean residence time
varying concentration which is independent of the directions
normal to the flow direction 1l2(X)
J: r (t (C) t)dt (1l
)2 (27)
(c) =F(t) x=O (15) If we substitute the expression for the Laplace transform of (c)
into Eq. 22 and calculate m2 and 112 we can show that
Since there is zero flux of solute in the and directions along the
column length , Eq. 6 reduces to
(9) (0)j3 (28)

1l2(X) J.2(0) (1l (X) - 1l

:rage defined
ox (C)
xx OX (16) which can also be written as

(10) where D;x is the longitudinal or axial dispersivity. For very long 1l2 (X) 1l2(0) + 2D;, (( (29)
columns it is appropriate to use the boundary condition (vx
: local spatial
value to the (c) -+O asx-+co (17) The quantity
column at
1l2(0) is the spread of the pulse introduced into the
= O. From measurements of 112 at various axial posi-
so that the column outlet does not influence the concentration tions it is possible to use Eq. 29 to cal"ulatethe longitudinal D;x,

(ll) profies in the bed. The solution to Eq. 16 subject to a step change dispersivity, from pulse inputs into the column. Both the pulse
in concentration at the bed entrance response and the step response are convenient techniques for
Hum when a Iheasuring the axial dispersivity. The accuracy of these methods
I of the solute ;I(t) u(t) at = 0 (18)
wil be compared in the section discussing experimental results.
is given by The most convenient technique to measure lateral dispersivities
(12) (c) x- (vx is to consider a steady-state experiment where a uniform concen-
12erc tration of solute is introduced at the column entrance distributed

uas i-steady, over one half of the bed cross section

12 exp D erfc (19)
I f has been
xx :u: 0-Cy-CW/2 at = 0 for all

ially periodic From measurements of the concentration as a function of time at (C) fco (30)
the tortuosity a fixed axi iI position it is possible to use this relation to compute -Cy-CO
that follows the axial dispersivity.
lie tensor D* 11 is also possible to measure D;x from pulses introduced into the The coordinate system is shown in Figure 1. The concentration at
:hat if Eq. 12 column at = O. These pulse responses can then be analyzed by = 0 is independent of and as a result Eq. 6 reduces to
ensor cannot the method of mom nts. The Laplace transform of the solution to
flect itself in
points in the
Eq. 16 subject to the boundary conditions of Eqs. 15 and 17 is given

(vx ):x
ox = D*
xx ox D*
YY oy

(c) (s) (s) exp

In all cases considered , the concentration gradients are suffciently
steep in the direction to neglect the dispersion contribution along
(Vx: 2 + /2l the flow direction and write Eq. 31 as
, field is uni- xx D xx o(c) (C)
The moments of the concentration response of the system at a given
axial location are defined by the expression
oX = D*
YY oy (32)

(13) The error caused by the neglect of the axial dispersion term is less
se of uniform (c) t)dt n = 0 (21)
than 1% for most values of Furthermore , the boundary con-
ltion reduces = J: dition in Eq. 30 can then be approximated very well by
From a knowledge of the Laplace transform of (c , we can (c) =co u(y) atx=O (33)
compute these moIhents
is a unit step function. Eq. 32 can then be solved subject
where u(y)
= 0 (22)
)fg the flow
(c) =O asy-+::co (34)
luce a time- The absolute moments of (c) with respect to time are obtained
by the simple expression The solution is

31, No.
AIChE JOurnal (Vol. 31 , No. February, 1985 Page 281
y, ,,, .,.,""

Flow Divider
I 16 SPT Fastener
!,:i:' Outlet Manifoid
(.jl:j top
Inlet Distributor dra
\:!I ane
i:ii' (faJ
Support Ring Sealing Material
ROjameter Right Side of Column onl
Lt. Material
Ring Column !in
Left S,de of Column Wail de:
Figure 2. Overall schematic of experimental equipment. pIa
Figure 4. Device for lateral motion of probe.


(35) the density of the NaCl solution. This helped in controlling natural con-
t3 vection effects at the lowest Reynolds nurnbers. in)
The concentration of NaCI was measured using the platinum conduc- lat
tivity probes shown in Figure 3. Platinum wires of 0. 065 cm diameter were
and from measurements of (C) t3as a function of we can calculate wound in grooves on the snrfa e of a ny Ion connector a distance of 0. 18 cm in-
the lateral dispersivity. apart. The ends of the platinum wires were soldered to copper wires which
were then inserted in glass melanine tubes of 1.27 cm outer diameter. The id,
tubes were given a thin wax coating to prevent penetration of water. The SOl
EXPERIMENTAL PROCEDURE platinum rings wound around the connector were in direct contact with sir
the fluid phase. All other portions of the wire were out of contact with the
fluid since the v were inserted in the platic tube. The ends of the tubes ex-
Set- tended throug h the column walls and the signals were recorded on a mul-
tichannel recorder. There were five probes located in the column at the
A plexiglas column of square cross section was constructed in order to following distances from the inlet: 2. 8 cm , 25. 7 cm , 47. 0 cm , 112. 1 cm , and fll
perform the experiments outlined in the previous section and shown 147. 6 cm. These probes are labeled 1 through 5 , respectively. Probes 3 and
schematically in Figure 1. An overall diagram of the equipment is shown 4 were designed so that they could traverse the column cross section along
in Figure 2. The width W of the column is 27 cm and the height of the the direction by means of a screw assemblage shown in Figure 4. The total
packing section is 150 cm. A distributor at the top insures flow uniformity. moving span was 15. 2 cm or about 60% of the column width. For every turn
It consists of a 1.2'7 cm plexiglas plate with 0. 32 cm diameter perforations or rotation of the mechanism the probes translated a distance of 0. 16 cm.
every square centimeter. An inlet manifold 5 cm high was outfitted with As a result , it was possible to determine very accurately the probe location
a flow divider in order to perform the lateral dispersion measurements. At for the lateral concentration measurement.
the outlet , a distributor consisting of 1.9 cm diameter PVC tubes 11.4 cm The particles used in these experiments consisted of urea formaldehyde
long insured straight streamlines at the bottom of the porous bed. Two sets spheres of several diameters: 0. 25 cm , 0. 35 cm , 0. 45 cm , and 0. 55 cm. We
of reservoirs , pumps , and rotameters were necessary in order to make the also used glass spheres of 1.00 and 1.58 cm diameters. The density of the
lateral dispersivity measurements. formaldehyde spheres was 1.463 g/cm 3 and those of the glass spheres was
The inlet tracer consisted of NaCI in de- ionized water at approximately 94 g/cm 3 for the 1.00 cm spheres and 2. 83 g/cm3 for the 1.58 cm spheres.
5 X Hr g eq. /L. The molecular diffusivity of NaCI was estimated to be There was a variance of less than 0. 1 % from the mean particle diameter
equalto 1J = 1.545 X 10- , and the Schmidt number to beSc v/1J for both the glass and urea formaldehyde spheres. The void fractions were
= 560. The density of the water which did not contain NaCI was adjusted computed by carefully weighing the particles before inserting into the bed
by adding small amounts of sucrose so that it matched as well as possible and calculating the volume of particles from the density. All the volume
fractions of fluid in this experiment were in the range E = 0. 39 to E =
The column was packed by letting the particles settle in water after
stirring and then tamping down slightly. The column was filled about 4
-- c - --- cm at a time using this procedure. Air bubbles were eliminated by de-
T ---- -.. gassing the water prior to use , and by carefully stirring the packing as the
column was being filled.
:':e or
cu Wire

In the experiments for measuring the longitudinal dispersivities , both
pulse inputs and step inputs were used. In the pulse input experiments the
column was filed with water and drained so that the manifold above the
distributor only contained about 3 mm of water. The solution containing
NaCI was then carefully and slowly introduced into the manifold. The
column was drained slowly so that the band of NaCl solution was introduc
at the top of the packing section. The manifold was filled with water again
and the pumps turned on to the desired flow rate. A recirculation or bypass
line attached to the inlet line of the column was used in order to set the flow
rate to the desired value prior to the start of the experiment. Once the flow
Connector rate was fixed , the fluid was diverted to the column by turning off the by-
( Nylon)
pass valve and turning on the inlet line valve. This way the liquid flow rate
Figure 3. Platinum conductivity probe. was at the desired level throughout the length of the experiment. ThIs

Page 282 February, 1985 AIChE Journal (Vol. 31 , No.


r ' :'111

teehnique resulted in sharp, nearly symmetric concentration pulsesbyatfirst

the 100. Probe 2
I = 25.7
top of the bed. The step inputs were inserted in a similar fashion 90.
draining the column so that only 3 mm of water wassolution.
left in the
pumps :; 80. Uniform Size Distribution
d then
a ere manually
started and thefiling the manifold
volumetric withsetNaCI
flow rate to the desired value. The - 70.
p = 3000

measured using the multichannel 60.

:anient respons at each probe was then Probe 3
i 50. I = 74. 0 em
The lateral dispersion measurements were made by dividing the mani-
635 cm thick plexiglas plate. NaCI 30.
fold into two regions by means of a 0.
soluti and water were pumped from their respective tanks , the NaCI to &1 20.
, the water to the other side. The flow rates were 10.
one side of the divider
, the sliding probes were
equalized and after a steady state was reached
lateral concentration profies. Seanning
moved in order to measure the
, and concentration measurements were 120 240 360 480 600 720 840 960 1040

started from the fresh water side Time (sees)

taken every 0. 16 cm , again using the recorder. After eaeh probe displace-
the con- Figure 6. Sample raw data for pulse response experiment.
ment the system was aIlowed to come to a steady state prior to
centratio measurement.
Particular care was taken to ensure that the response of the probes was optimal values of these parameters were found using a search
g eq./L. The circuit
linear over the concentration range of 0 to 5 X (c)13 versus time data at four different
Jlumn to process the signals from the routine that analyzed the
designed by M. Ahmon (1977) was used
platium wires. Yoltage was supplied to a probe only when a meaurement
values corresponding to probes 2 , 3 , 4 , and 5. As a result , one can
connecting and triggering the see variations in average fluid velocity and axial dispersivity be-
was being taken. This was accomplished by in the multichannel tween probe locations. Of course the true value of
(v. ) 13 is known
power supply to the channel switching mechanism
prevent stray currents between , and the true void
recorder. This was necessary in order to from the overall flow rate Q, the column area A
one probe and another from
modifying the desired signal. We also made fraction E obtained from weighing the particles. One can compute
sure that the flow in the column was uniform by using the sliding probes. an experimental volume fraction of fluid from the relation
atural con-
The meaured responss and axial dispersion coeffcients from both the step
input and pulse response experiments were essentiaIly independent of the (36)
m conduc- lateral probe location , indicating a flat veloeity profie and uniform inlet
E exp
- -
meter were
conditions. This was confirmed by dye observation experiments which
of 0. 18 cm indieated little channeling anywhere in the bed
, and very small amounts If the data and analysis are-consistent , and the flow is uniform , Eexp
vires which for the dye was should closely approximate the true
of ehanneling near the corners. The mean residence time
meter. The identical to that for NaCI , indicating that there were no appreciable ad- In the pulse response experiments , we computed the zeroth , first
water. The sorption effects between the solute and the particles. This was to be expected and second moments of the time response at probes 1 through 5.
mtact with since the ratio of be width to mean particle diameter was about 77 in thes The absolute first and second central moments of the response at
,ct with the experiments. Natural convection effects prevented us from doing experi- probe 1 were treated as (O) and J.2 (0) respectively. The value of
3. Since we were
Ie tubes ex- ments with particle Reynolds numbers much less than 0.
J.2 (X) at the remaining probe locations were used to
doing experiments with relatively large particles , the Grashoff numbers
(x) and
dona mul-
umn at the calculate (vx ) 13 and V;. from Eqs. 25 and 29. An experimental void
were large enough to cause natural convective currents at these low liquid
1 cm , and flow rates. This made the velocity profie in the bed nonuniform and the fraction can again be computed from Eq. 36.
robes 3 and dispersion measurements unreliable. No runs were made with different In Figure 7 we see E exp as a function of Q / A , the superficial ve-
,ction along fluid viscosities or bed diameters. locity for both the step and input response data. These are average
4. The total void fractions between probes 2 and 5. The results compare very
r every turn well with the true void fraction of 0.41 , especially at large flow
of 0.16 cm.
RESULTS rates , where E exp is approximately 0.43. This indicates good packing
)be location
and a lack of flow nonuniformities. In Figure 8 we show the lon-
maldehyde Longitudinal Dispersivity Measurements- Uniform Particle Size gitudinal dispersivities at the various axial locations as a function
55cm. We of the Peclet number obtained from the pulse response experi-
nsity of the In Figures 5 and 6 we show typical experimental results for step ments. These data exhibit a great deal of scatter due to the inac-
spheres was input and pul response experiments when the bed was filled with curacies of calculating the second moments of the experimental
cm spheres. particles of 0. 35 cm diameter andE = 0.41. As can be seen , the raw data. Small differences in the way the tails of the concentration
,Ie diameter data are very smooth and show little scatter for both experiments. response are truncated and small errors in the value of measured
actions were From these time" dependent concentration measurements, Eq. 19 concentration for very low values of concentrations give extremely
into the bed was used to calculate V;. for the step input experiments and Eq. large errors in computed V;x. This factor contributes significantly
the volume 29 was used to calculate the same quantity from the moments of to the large amount of scatter found in dispersion data. In contrast
39 to E =
the pul response. In using Eq. 19 the value of V;. was found from Figure 9 shows the results for axial dispersivities using the step
water after successive iterations , u ing the best linear fit to the inverse com- response data. It is clearly much less scattered , and it shows clear
lied about 4 plementary error function of the terms in the equation. Conver-
mted by de- gence was obtained when the old and riew values differd by less
,cking as the than 0. 1%. For the case of the step input , both the velocity (v. ) 13
were treated as unknowns , and

and longitudinal dispersivity V;.
o .
. 0
. o 80 O
. 0

sivities , both R -mln

R. 0.
,eriments the mn - Rmin eJp 0.
Id above the
n containing
anifold. The
s introduced Uniform Size Distribution
980 0.1
I water again
ion or bypass
o set the flow 100 500 1.000 500 000
10- 10-
)nce the flow TIME (see)

Ig off the by- ) = Q, A (em/see)

Figure 5. Sample raw data for step response experiment. R is recorded
luid flow rate Figure 7. Volume fractions obtained from probe response.
riment. This

February, 1985 Page 283

~1, No. AIChE Journal (Vol. 31, No.
.,,--= (( ..
. . y.:




Capillary Tube Result (Paine el aI. , 1983)

8 = (
0 Probe 5
. Probe 4
Figure 10. Longitudinal dispersivity dependence on
to Probe 3

Ii .J
1 - E
)) 1. (37

We can develop a quantitative criterion for when the axial dis-

persivity is constant by taking the axial dispersivities shown in
Figure 9 at the probe locations 2 , 3 , and 4 , and dividing them
the dispersivity measured at Jocation 5 at each Peclet number
Figure 8. Longitudinal dispersivities from pulse response experiments-uniform (D;. )oo. Since L is different at the different probes , we can compute
particles. Ii for each probe according to Eq. 37. The results are shown in

Figure 10 for all P ~clet numbers .: 5 X 101 We did not carry

out the data anal s for h lgher Peclet numbers since for these high
trends of the variation of the longitudinal dispersivity with Peclet Peclet numbers It IS possIble that we have not achieved the final
value of D;. even at the last probe location. For values of Ii? , the
number and probe location. We note from Figure 9 that for

-( 1() the values of D;. are nearly identical for all values of =L ratio of D;. to (D;. )oo oscilates about 1.0 with about a 10% varia-
or probe location. However , as the Peclet number nears 10 tion. This is a good measure of the accuracy of the experiment.
begins to notice a significant difference between the dispersion see that for Ii , one can be assured of being within 90% of the
final constant value of the longitudinal dispersivity. We can con-
coefficient calculated at 25. 7 cm and that computed at 147. 6 cm.
Basicaly, one can observe an increase in the value of the dispersion clude that Eq. 37 is too restrictive , and in fact one can expect
coeffcient with increasing distance down the column. The larger constant longitudinal dispersivities as long as
the value of the Peclet number , the larger the discrepancy between .J
Ii (38)
the dispersion coeffcient measured at = 25. 7 cm and = 147.
In Eq. 2 we have expressed the time criterion required in order This is the constraint that should be considered when requiring
to obtain a constant value of the longitudinal dispersivity. If we constant longitudinal dispersivities in a transient axial dispersion
associate with time since injection of the pulse the quantity L/ experiment. In Figure 10 we also show the theoretical prediction

(v. , and substitute the characteristic length ltJ defined in Eq.
of Paine et al. (1983) on the time dependence of the axial dispersion
, we can rewrite the time constraint in Eq. 2 as coeffcient. Even though this calculation was done for a capilary
tube , one can see a very similar trend to that shown by the data.
In Figure 11 we compare our data for probe 5 , the largest dis-
persion distance , to a great deal of axial data in the lit-
erature , mostly from references listed in Table 1. There is a very
o x 0 147. 6 ern Probe 5 good agreement between our results and previous values even
o x 0 112.1 em Probe 4
to X = 74. 0 ern Probe 3
though our data tend to be on the lower side of the previously
'1 X = 25.7ern Probe 2 published results. We believe this is due to the fact that we have
taken great pains to reduce effe ts due to particle size distributions
0 0
0 to
o to

.. Pfannkuch
v Ebach and While
. Carberry and Bretton
o Edwards and Richardson
gij V Ji.
. Blackwell et al.
o Rital
o This Work ;joif: v .




10- 10-
Figure 9. Longitudinal dispersivities from step response experiments-uniform
particles. Figure 11. Longitudinal dispersivily values from various experimenls.

Page 284 February, 1985 AIChE Journal (Vol, 31 , No.

(j p, . .

oll ell
. I =112.1 em - Probe 3 lJ Harleman and Rumer (19&3)
o I = 74.0 em - Probe 2 . Gunnand Pryce (1969)
0.7 .. Hasserman and Von Rosenberi (1968)
o ThisWork
102 s8 8
t 0. ;a e- - - - - - - - - -
-; 0.
0.3 p = 873 .
Uniform size distribution
. b,4J!e

0.1 -t 0
18 ) eeee 00o

1.7 1.7 2.4 o 4

, (em) . i.
Figure 12. Sample raw data from lateral dispersion experiments.

small ratios of column- to-particle diameter , flow nonuniformities
and natural convection induced by density differences. One can 10- 103 104

se a curvature of the collected data at Peclet numbers greater than

104 . Bear (1972) has ascribed this to turbulence effects in the porous Figure 13. Lateral dispersivity values from various experiments.
media. However , note that the two experiments that exhibit this
le axial dis-
curvatue at high Peclet number are those of Carberry and Bretton Effect of Particle Size Distribution on Longitudinal and Lateral
s shown in (1958) and Ebach and White (1958) where the constraint of Eq. Dispersion
ng them by
4 is violated. In Figure 9 we show curvature of the dispersion data
with Peclet number when the column length does not satisfy the In order to study the effect of the distribution of particle sizes
let number on longitudinal and lateral dispersion coeffcients , we repeated the
an compute criterion of Eq. 4. The results of our work are not explainable in
ternis of turbulence since the highest Reynolds numbers we used measurements described above for two different particle size dis-
e shown in tributions that had the same void fraction and the same mean
id not carry
were on the order of 18. This is far below the laminar- to- turbulent
transition in packed bed studies. Careful analysis of the data in particle diameter , but different particle size ranges. Details of the
r these high
Figure 11 indicates that there is curvature also in the data of particles that made up these distributions are given in Table 2. Size
ed the final
pfannkuch (1963) where Table 1 indicates that the criterion of Eq. distribution 1 has E = 0. 39 and a mean particle diameter d of 0i35
)f :; 1 , the
4 is marginally satisfied and the Reynolds numbers are no higher cm. Size distribution 2 has E = 0. 41 and the identical value of
10% varia-
than 10. This is the same diameter as the particle size used in the experi-
riment. We ments described previously in which all the particles had a uniform
, 90% of the
This is very convincing evidence that the curvature seen in
Figure 11 is not dependent on turbulent effects , but simply too size. The mean particle diameter is defined as
Ve can con-

can expect
short a column for the high flow rates used. As Figure 10 indicates
if the column is too short one can measure a dispersion coeffcient = 6 L ni L ni (39)
that is significantly lower than the final asymptotic value.
where V p and A p are the total volume and surface area of the
(38) is the di-
particles in the bed ni is the number fraction , and
Lateral Dispersivity Measurements- Uniform Particle Size
ameter of each size particle. Equation 39 can be rewritten using
rI requiring the weight fractions and densities of the different particles
In Figure 12 we show a typical plot of the lateral concentration
I dispersion
profies measured during steady-state experiments aimed at de-
I prediction (40)
termining the lateral dispersivities. There is an increase in the L (xd Pi L (xd pi
II dispersion
spread of the concentration profie from = 74. 0 cm to = 112.
. a capilary cm. Equation 35 was used to calculate D;y from known values of This is the most convenient way to compute d from measured
y the data.
(vx P and and the concentration profie. This was accomplished weights of each type of particle. The main difference between size
largest dis- distributions 1 and 2 is that size distribution 2 has a small percentage
by plottig the inverse complementary error function of 2 (c ) P /
:a in the lit- of very large particles where the ratio of maximum to minimum
as a function of and finding the slope (Harleman and Rumer
,re is a very particle diameter is 7. 3. Size distribution 1 has a ratio of maximum
1963). The results are shown in Figure 13 , together with some
ralues even to minimum particle diameter of 2. 2. Even though the large par-
collected values from the literature. The results are independent
previously of probe loction. We have plotted in Figure 13 all of the data taken ticles in size distribution 2 are only a very small fraction of the total
,at we have at both probes. Previous results for lateral dispersivities had covered number , they make up about 11% of the total volume.
istributions the Peelet number range from 10- to 10 . Our data extend the
available values up to :r'Peclet number of 10 . As can be seen from Longitudinal Dispersion
Figure 13 there is very good agreement between our results and
those of Harleman and Rumer (1963) in the Peclet number range /;n measured at different
1: . In Figure 14 we show the values of D

probe locations using step input experiments for the case of a uni-
.:1": where they overlap.
." V'
, 0

Distribution 1 (E = 0. = 0. 35 cm)
Particle type , * i
Particle diameter dpl (cm) 0.45 1.00 1.58
Number fraction 649 237 074 040
Weight fraction
Volume fraction
Distribution 2 (E = 0. = 0. 35 cm)
Particle type, * i
Particle diameter dpl (cm) 0.45 1.00 1.58
Number fraction nj 6604 2808 0566 0017 0045
Weight fraction
Volume fraction 3328 3882 1665 0552 0573

erimenls. . Particle type 1- 4 are urea formaldehyde spheres PI = 1.463g/cm

Particle type 5-6 are glas spheres = 2.940 g/cm , Po = 2.83 g/cm

, No. 2) . AIChE Journal (Vol. 31 , No. February, 1985 Page 285

-- .. ':(

12, 000
10, 000 3445
'1 " Size Distribut,on
0 0 Size Distribution 2
000 Uniform VZ,
1455 1a
940 disj
D;y disl
Size Distribution 1 Ion
000 .'1

2174 eXI
1462 the
D;x 803 esti
12, 000 Size Distribution 2
3884 102 rad
000 Figure 16. Effect of particle size distribution on lateral dispersivities.

(Lld niformity near the electrode. This phenomenon was not observed
Probe 2 Probe 3 Probe 4 Probe 5 in either of the other two packings.
Figure 14. Effect of particle size distribution on longitudinal disperslvity at Since size distribution 2 showed a larger required length
various probe locations. achieve constant axial dispersivities , we need to be a little careful
in deciding which are the final asymptotic longitudinal dispersi-
vities measured for this size distribution. Figure 14 indicates that
form particle size , size distribution 1 , and size distribution 2 for for a for 1 642 or less , the longitudinal dispersivity measured
representative Peclet numbers. We note that for size distribution at probe 5 is fairly close to its final value. In comparing the results
1 and the uniform particle size case , the longitudinal dispersivity for the three different size distributions we chose only those particle
values become constant at shorter dispersion lengths than for the Peclet numbers less than 2 000. This way we are more or less as- AtJ
wider size distribution 2. For example , it is not elear that the as- sured of comparing longitudinal dispersivities which have reached
ymptotic value of D /:J has been reached at a Peelet number of their final value at a particular particle Peclet number. In Figure
642 in size distrbution 2. However, for the uniform pareles and 15 we have plotted the highest values of V;, /:J measured for the
size distribution lone sees nearly constant values of longitudinal three size distributions for this small Peclet number range. Note
dispersivities at = 1 455 and 1 462 respectively. It is apparent that there is almost no difference at all between the results for the
that the wider size distribution has a longer characteristic length case of uniform packing and size distribution 1. Since the mean
than that used in Eq. 3. One can conjecture that this is due to a particle sizes are the same , the volume fraction is nearly the same
larger scale nonuniformity in the macroscopic flow field , but there and the ratio of maximum to minimum size is not too far from 1
are not enough data at this time to quantify the effect. Mention one does not observe much of a difference between size distribution
should be made of the apparent slight decrease in the longitudinal 1 and the uniform case. However , the values of D;. /:J for size
dispersivity found at the last probe in the experiments with size distribution 2 are larger by a significant amount (about a factor
distribution 1. We believe this is an artifact in the experiment 2) than those for the uniform particle size case. The two lines of data
caused by either a slight bending of the probe or a packing nonu- are nearly parallel , indicating that one could superimpose them
if there was a re- definition of the characteristic length in the Peclet
number for the case of wide particle size distributions. This
consistent with the results of Figure 14 , which also indicated that
there was a length scale , larger than the mean hydraulic radius,
. Uniform
'1 Size Distribution 1 which dominates the approach to a constant axial dispersivity
. Size Distribution value. ItJ
It should be noted that the three different pac kings used- the
uniform particles and size distributions 1 and 2- have nearly
identical hydraulic radii since they have the same value of and
nearly the same volume fraction. As a result , the observed differ-
ences in longitudinal dispersivities must be caused by the spread
of the size distribution alone. Eidsath et al. (1983) computed lon-
c;x 103 gitudinal dispersivities in two dimensional spatially periodic arrays
of cylinders with two different radii. They found that when the
ratio of cylinder radii was increased from 2 to 5 , the axial dis-
persivity increased by a factor of 1.5 at the same Peclet number.
This is an increase of the same order as was found in the experi-
mental observations presented in this work. Carbonell (1979) also
found a very strong dependence of longitudinal dispersivities
the spread of pore size distributions for a very simple capilary tube
model of the porous medium. If one is wiling to associate an in-
creased particle size distribution with an increased pore size dis-
102 tribution , the trend observed in the experiments just described
at least qualitatively the same as that predicted by the theory.
Additional experiments need to be done before this effect can
understoo fully, but it is clear that one needs to emphasize the role
Figure 15. Longitudinal dispersivities for uniform particles, size distribution of the width of the particle size distribution on the dispersion
1 and size distribution 2. process.

Page 286 February, 1985 AIChE Journal (Vol. 31 , No.

= width of bed
taleral Dispersio
.X = axial position in bed
' In Figure 16 we have collected the results for lateral dispersivities = weight fraction
V' for the uniform particle size case, as well as for size distributions = Cartesian coordinate normal to flow direction
2. There is no evidence to indicate a change in the lateral = Cartesian coordinate normal to flow direction
persion with particle
size distribution. At least , lateral
)iJ = refers to an interstitial average
sensitive to size distribution than
;persivities seem to be much less
longitudinal dispersivities. This conclusion perhaps should be
ualified by the fact that we took steady-state measurements of Greek Letters
. One should expect that
values measured in a transient
= void fraction
eriment would be subject to the same long- time constraint as
= density of particle of type i
e longitudinal dispersivities. There have been only one theoretical
= tortuosity tensor
. stimate of the effect of particle size distribution on lateral dis- = nth absolute moment in time
rsion. Eidsath et al. (1983) found that when the ratio of cylinder J1n
= nth central moment in time
radii in a two- dimensional spatially periodic array was changed J1n
= kinematic viscosity
from 2 to 5, the lateral
dispersivity actually decreased by a factor
= refers to the fluid phase
of about 3 , but this is an artifact of the extremely simple geometry
vites. employed in these computations.


bserved Ahmon , M. One- Chip Conductivity Meter Monitors Salt Concentrations,"

Electronics 132 (Sept. 15 , 1977).
mgth to Aris , R. On the Dispersion of a Solute in a Fluid Flowing through a Tube
This work was supported by NSF Grant ENG. 7913269.
careful Proc. Roy. Soc. A235 , 67 , London (1956).
lispersi- Bear , J. Dynamics of Fluids in Porous Media Elsevier Pub. Co. , New York
ltes that (1972).
easured BlackwelI , R. J. , J. R. Rayne , and W. M. Terry, " Factors Influencing the
results Effciency of Miscible Displacement AlME Petroleum Trans. , 217
1 (1959).
BlackwelI , R. J. " Laboratory Studies of Microscopic Dispersion Phenom-
. less as- Ap" = fluid-solid interfacial area ena," Soc. Petroleum Eng. j., 1 (Mar. , 1962).
reached = column cross sectional area Brenner , H. Dispersion Resulting from Flow through SpatialIy Periodic
I Figure = total surface area of particles in bed Porous Media Phil. Trans. Roy. Soc. 297 , 81 , London (1980).
l for the = point solute concentration Carberry, J. J. , and R. H. Bretton Axial Dispersion of Mas in Flow through
:e. Note = solute concentration in input Fixed Beds AIChE J., 4 , 367 (1958).
, for the = spatial deviation in solute concentration CarbonelI , R. G. Effect of Pore Size Distribution and Flow Segregation
Chem. Engr. Sci. , 1 031 (1979).
e mean (c)P = interstitial average solute concentration on Dispersion in Porous Media,"

(c)iJ Flow Non- Uniformities in Packed Beds: Effect on Dispersion,"

Ie same (c)P = Laplace transform of
Chem. Engr. Sci. , 1 397 (1980a).
from 1 = particle diameter Effect of Particle Size Distribution and Flow Non- Uniformities
ribution = average particle diameter on Dispersion in Porous Media: Application to Oil Shale Retorts Proc.
for size dpi = diameter of particles of type i 2nd Multiphae Flow and Heat Transfer Symp. T. Veziroglu , Ed. , 2379
actor of = hydrodynamic dispersion tensor (1980b).
; of data = total dispersivity tensor Carbonell , R. G. and S. Whitaker Dispersion in Pulsed Systems. Part II:
,e them V;.,V;y = longitudinal and lateral dispersivities Theoretical Developments for Passive Dispersion in Porous Media,"
e Pec\et = .molecular diffusivity Chem. Eng. Sci. 38, 1795 (1983).
This is = closure function for the spatial deviations De Josslin de Jong, G. Longitudinal and Transverse Diffusion in Granular
Deposits Trans. Amer. Geophys. Union 67 (1958).
ted that :1(t) = time- dependent inlet concentrations
Dullen , F. A. 1. Porous Media: Fluid Transport and Pore Structure
radius (s) = Laplace transform of ;; Academic Press , New York (1979).
ersivity = unit tensor
Ebach , E. A. , and R. R. White Mixing of Fluids Flowing through Beds
= characteristic length for fluid between particles of Packed Solids," AIChE j., , 161 (1958).
the = characteristic length for average concentration Eidsath , A. , et aI. Dispersion in Pulsed Systems. Part II: Comparison
nearly changes between Theory and Experiments for Packed Beds Chem. Engr. Sci.
and = nth moment. of the average concentration with 38, 1803 (1983).
I differ- time Edwards , M. F. , and J. F. Richardson Gas Dispersion in Packed Beds
spread = number fraction of particle type i Chem. Engr. Sci. , 109 (1968).
ted lon- = unit normal from fluid to solid Gil , W. N. , and R. Sankarasubramanian Exact Analysis of Unsteady
Convective Diffusion Proc. Roy. Soc. A316 , 341 , London (1970).
e arrays = particle Peclet number
Grane , F. E. and G. H. F. Gardner Measurements of Transverse Dis-
hen the = volumetric flow rate persion in Granular Media," J. Chern. Eng. Data 283 (1961).
dal dis- = capilary tube radius Gray, W. G. A Derivation of the Equations for Multiphase Transport
umber. = radius of the averaging volume Chem. Engr. Sci. , 229 (1975).
experi- = particle Reynolds number Greenkorn , R. A. , and D. P. Kessler Dispersion in Heterogeneous , Non-
79) also = Laplace transform variable Uniform , Anisotropic Porous Media Flow Through Porous Media
ities on = Schmidt number Amer. Chem. Soe. , Washington , D. C. (1970).
lry tube = time Goon , D. J. , and C. Pryce Dispersion in Packed Beds Trans. lnst. Chem.
an in- u(t), u(y) = unit step functions in time and lateral position Eng. , T341 (1969).
Haring, R. E. , and R. A. Greenkorn A Statistical Model of a Porous Me-
,ize dis- = point velocity vector in fluid phase dium with Non- Uniform Pores," AIChE j., 477 (1970).
ribed is = spatial deviation in fluid velocity Harleman , D. R. F. , and R. R. Rumer Longitudinal and Lateral Dispersion
theory. (v)iJ = interstitial average velocity vector in an Isotropic Porous Medium " J. Fluid Mech. 16, 385 (1963).
: can be (vx )iJ = superficial and interstitial average velocity in flow Hassinger , R. c. , and D. U. von Rosenberg, " A Mathematical and Exper-
the role direction imental Examination of Transverse Dispersion Coeffcients Trans. Soc.
persion = volume fraction of particle type i Petroleum Eng. j., 243 , 195 (1968).
= total volume of particles in bed Niemann , E. H. Dispersion during Flow in Non- Uniform Heterogeneous

1,0. AIChE Journal (Vol. 31, No. February, 1985 Page 287
, " , :' , " , ", " ," ,,""

Porous Media , MS Thesis, Chem. Eng. Dept. , Purdue Univ. , Lafayette Slattery, J. c. Momentum , Energy, and Mass Transfer in COntinua
IN (1969). McGraw- Hili , New York (1972).
Paine , M. A. , R. G. Carbonell , and S. Whitaker Dispersion in Pulsed Taylor , G. 1. Dispersion of Soluble Matter in Solvent Flowing Slowly
Systems. Part I: Heterogeneous Reaction and Reversible Adsorption in through a Tube Proc. Roy. Soc. A219 , 186 , London (1953).
a Capilary Tube," Chern. Eng. Sci. 38, 1781 (1983). Conditions under Which Dispersion of a Solute in a Stream 01
pfannkuch , H.- Contribution a J'Etude des Deplacements des Fluides Solvent Can be Used to Measure Molecular Diffusion
Miscibles dans un Milieu Poreux Revue de l'nstitut du Petrole , 18 A225, 473 , London (1954). Proc. Roy. Sac

215 (1963).
Rifai , M. N. E. , W. J. Kaufman , and D. K. Todd Dispersion in Laminar
Flow through Por JUs Media Sanitary Eng. Res. Lab. Rpt. No. , fER
420 (1967).
Whitaker , S. Diffusion and Dispersion in Porous Media,"

Whitaker , S. Forced Convection Heat Transfer Correlations for Flaw

AIChE j.,
13 '

Series , Berkeley, CA (July 1 , 1956). in Pipes , Past Flat Plates , Single Cylinders , Single Spheres , and for Flow
Ryan , D. , R. G. Carbonell , and S. Whitaker A Theory of Diffusion and in Packed Beds and Tube Bundles AfChE j., 361 (1972).
Reaction in Porous Media," AIChE Symp. Ser. , No. 202 77, 46
Saffman , P. G. Dispersion in Flow through a Network of Capilaries Manuscript received April 11 1983; revision received November , 1983 and ac.
Chem. Eng. Sci. , 123 (1959). cepted December , 1983.

Particle Collection in Magnetically

Stabilized Fluidized Filters
A study was conducted on the use of magnetically stabilized fluidized (MSF) R. V. ALBERT and
fiters for removing fine partieles from gases. The effect on filter performance of
a number of variables was examined , ineluding the applied field strength , gas ve- CHI TIEN
locites, and bed height. The dynamic behavior of MSF filters was found to be
Department of Chemical Engineering and
characterized by a decreasing collection effciency with time as a result of sig-
Materials Science
nifcant partiele re-entrainment.
Syracuse University
Syracuse, NY 13210

Granular fitration has been applied extensively in the past vantage of continuous operation , which is desirable in large.
to control particulate emission and has proven effective for scale applications.
partieles with a wide size range. It is particularly suitable for This paper reports results obtained from a systematic exper-
operation under high temperature and pressure conditions. On imental study of the performance of magnetically stabilzed
the other hand, granular fitration is inherently unsteady state. fluidized (MSF) filters. The effect on filter performance of
The accumulation of fine particles within a filter requires pe- number of variables was examined, ineluding the magnetic field
riodic media regeneration. strength, gas velocities , bed height , and the size of fluidized
A possibleway of overcoming the shortcomings arising from partieles. Experiments were also conducted on the extent of
the cyelic nature of granular filtration is to operate filter beds partiele re-entrainment.
in the fluidized mode, thus allowing a continuous withdrawal The concept of the unit collector efficiency proposed by TieD
and introduction of fiter grains. Furthermore, if fluidization and Payatakes (1979) was used to assess the effectiveness of MSF
is carried out with magnetic stabilzation , the presence of gas filters and as a basis for comparing the performance of a MSF
bubbles in the bed can be eliminated and/or reduced. One may filter with its equivalent fixed-bed filter. The feasibilty
speculate that such an arrangement offers an ideal combination employing a phenomenological rate expression to describe the
of high collection effciency of granular fitration and the ad- dynamic behavior of MSF fiters was also demonstrated.


The results of this experimental study demonstrate that sub- finding is the significant partiele re-entrainment that occurs as
stantial partiele collection effciency over long operating periods the specific deposit of the filter increases. As a result, MSF fi.
can be achieved with the use of fluidized fiters under magnetic ters, in most cases, exhibit a decrease in collection effciency
stabilzation. with time.
Compared with fixed- bed filters of comparable bed height The dynamic behavior of MSF filters can be approximately
MSF filters give a somewhat lower collection effciency. The described by the phenomenological equation for deep- bed fil-
most important variable affecting the performance of MSF tration with the inelusion of particle re-entrainment.
fiters is the applied magnetic field strength. Another important

Page 288 February, 1985 AIChE Journal (Vol. 31 , No.