Chemical Engineermg Science, Vol. 45, No. 3. pp. 759-763, 1990. CKW-2509190 $3.00 + 0.

00
Printed in Great Britain % 1990 Pergamon Press plc

Liquid holdup and dispersion in packed columns

(Received6 February 1989; acceptedfor publication 12 July 1989)

INTRODUCTION holdup. The step decrease method can differentiate between

In real flow systems the flow pattern is intermediate to the these two liquid portions and is therefore best suited for this
two extremes of plug flow and backmixed (CSTR) behaviour. purpose. Bennett and Goodridge (1970) and Schubert et al.
A dispersion model is frequently used to describe the flow (1986) obtained the liquid phase axial dispersion characteris-
behaviour of real systems which do not drastically deviate tics along with h, and h, in a packed column using the step
from plug flow. It is well known that the liquid phase flow decrease method. These workers however. used relatively
pattern in a packed column does not deviate drastically from small sized ( < 0.0095 m) packings (Raschig rings, Berl
plug flow and therefore the dispersion model has been widely saddles) in very small ( < 0.075 m diameter) columns. The
used to describe it (Carberry and Bretton, 1958; Sater and lowest nominal packing size used industrially is 0.025 m.
Levenspiel, 1966; van Swaaij et al., 1961; Bennett and Further, highly open packings like Pall rings are used in
Goodridge, 1970; Kan and Greenfield, 1983; Schubert er al.. modern plants rather than the closed Raschig rings. In view
1986). The liquid holdup in a packed column is comprised of of this it was thought desirable to obtain the above-men-
two major regions (Shulman et al., 1955; Hoogendoorn and tioned characteristics using a metal Pall ring (MPR). For
Lips, 1965; Kan and Greenfield, 1983): (a) the relatively fast comparison with Shulman et al., data were also obtained for
moving or dynamic liquid holdup, h,, which accounts for the a 0.025-m ceramic Raschig ring (CRR). The characteristics of
major solute transfer; and (b) the relatively stagnant or static the packings used are given in Table 1.
liquid holdup, h,, which exchanges mass with the surround- Various models have been used for describing the
ing dynamic liquid phase through a convective-type diffusion nonideality of flow in a packed column. The mixing cell
process. The relative proportions of h, and h, are important model postulates flow through a number of perfectly mixed
to the designer inasmuch as it gives the extent of effective cells in series. The flow between any two consecutive cells is
liquid holdup for mass transfer. assumed to be piston-like. This model has only one para-
From the above it is clear that a complete description of meter, i.e. number of perfectly mixed cells (Sater and
the liquid phase flow in a packed column requires a know- Levenspiel, 1966). The drawback of this model is that it is far
ledge oE from reality and does not represent the physical picture of
the flow in a packed bed.
(a) total, dynamic and static liquid holdup, h,, h, and h,, A two-parameter model which is more realistic is the
respectively; piston exchange (PE) model, which assumes piston flow
(b) dispersion effects. through dynamic holdup and exchange with the static hold-
up. There are a number of variations of the PE model. All
Shulman et al. (1955) were the first investigators who these versions are termed crossflow models (Patwardhan,
differentiated between h, and h,. Shulman et al. determined 1978a). In a series ofarticles Patwardhan (1978a,b, 1979, 1980,
h,, h, and h, by the weighing method. They found that the 1981) proposed an extended crossflow model wherein the
static holdup is constant over the entire range of liquid flow exchange between static and dynamic holdup is depicted by
rate. van Swaaij et al. (1969) used impulse tracer approach to convective-type transport. He applied this model to various
study the hydrodynamics and concluded that static holdups cases of mass transfer with chemical reaction to mathemat-
measured by the tracer method and draining method are ically evaluate the effectiveness of static holdup in mass
equivalent, and that all the liquid holdup is accessible to the transfer compared to dynamic holdup, and elegantly
tracer. However, it can be argued that for the impulse tracer brought out the applicability of the extended crossflow
the total period between tracer entry and exit may not be model even for second-order kinetics and reactions where
sufficient to cause interaction of the short lived tracer with gas phase transfer is rate-controlling. Like all other crossflow
the static holdup. The response to an impulse input thus may models this model also assumes piston flow in the dynamic
not include static holdup regions. On the other hand the step holdup. From the foregomg discussion it is evident that the
decrease tracer approach seems more versatile as far as static static holdup is that part of the total holdup which attains
holdup is concerned. Bennett and Goodridge (1970) and a dynamic equilibrium with its surroundings and exchanges
Schubert et al. (1986) used this step decrease method to mass with the dynamic holdup through a convective-type
investigate static and dynamic holdup. It was shown by these process. This description correctly defines the static holdup
authors that the static holdup determined by tracer methods and its coexistence with the dynamic holdup from mass
is lower than that found by the draining method. transfer viewpoint.
The basic difference between the tracer methods and the The PDE model proposed by van Swaaij er al. (1969) is
draining method is that in tracer experiments the column a modification of the crossflow model. It is a three-parameter
hydrodynamics are not disturbed as during the draining model which accounts for the axial dispersion taking place in
method. Secondly, the draming method measures the liquid dynamic holdup. This model is better than the previous
which remains trapped in packing interstices and joints as models since it represents the reality more closely. Bennett
well as the liquid which is held on the packings due to surface and Goodridge (1970) have given the details of the model
forces. Under irrigated condition a significant part of this and solved the equations using the Laplace transform
latter liquid may start moving, thereby reducing the static method. These authors have reported that this solution con-

CES 153.11 759

 Shorter Communications

Table 1.

Geometric Packing
Piece density surface area Voidage factor
Packing (No./ m3) (m-l) (%) (F)

25-mm metal Pall ring 48,000 188 94 48

25-mm ceramic Raschig ring 45,500 180 73 160

verges too slowly for high values of the Peclet number

(Pe > 25) to be satisfactory and that the partial differential
too
equation can be better solved using the Crank-Nicholson
procedure for the entire range of parameters.
Bennett and Goodridge also proposed an innovative
graphical method of estimating parameters from the ob-
served characteristics of the response curves on
a semilogarithmic scale. This method was used by
Patwardhan and Shrotri (1981) to find the exchange coeffi-
cient between the static and dynamic holdup by analysing
the tail portion of the response curve. In the present work
a computer program was written to solve the model equa-
tions and to estimate the model parameters directly from the
experimentally obtained response curves. The constrained
simplex method of nonlinear programming was adopted to
obtain the best match between the experimental and pre-
dicted response curves for the step decrease. The graphical Fig. 1. Transformed experimental response curve for MPR:
method of Bennett and Goodridge was used to arrive at the I/ , = 0.0889 x lo-‘m/ s.
preliminary guesses required for the computer program.

EXPERIMENTAL
RESULTS AND DISCUSSION
The experimental set-up used in the present work was
Table 2 summarises the correlations for Pe and k,/k, for
similar to that of Bennett and Goodridge (1970). A perspex
the packings investigated.
column of 0.2 m diameter and 1.5 m height was used. A lad-
der-type liquid distributor was used to distribute the liquid.
Total holdup, h,
A conductivity probe (cell constant 0.104, time lag < 0.1 s)
The total liquid holdup obtained is shown in Fig. 2. The
was fixed in the column outlet to measure the conductivity of
data obtained for a 0.025 m CRR by Shulman et al. are also
the outlet solution. The procedure used was akin to that of
shown in Fig. 2. The total holdups obtained by the present
Bennett and Goodridge.
Schubert rt al. (1986) used preflooded columns, which is method are 15-20%higher than those obtained by Shulman
et al. This difference may be attributed to the difference in the
an unlikely situation in industrial applications. In the pres-
ent work no attempt was made to wet the packings by two procedures employed.
flooding the column before commencing an experiment.
Holdup ratio, h,/ h,
Also, the liquid velocity was increased for each successive
run to incorporate the effect of improvement in packing Figure 3 shows the values of the holdup ratio determined
by the present method. It has been observed that k, is a much
surface coverage on the holdup as liquid velocity increases.
stronger function of V, than k, and therefore k,/k, decreases
The liquid velocity was varied from 0.0005 to 0.006 m/ s.
with Vi.
Since it is well known that the gas velocity has no effect on
the liquid holdup in the preloading region, no gas was
Static ho ldup, h,
introduced in the column.
Figure 4 shows the variation of static holdup with liquid
TREATMENT OF EXPERIMENTAL DATA velocity. The static holdup for 0.025-m CRR reported by
Shulman et a l. (1955) is also shown for comparison. The
The input to the computer program is required to be in the
following observations can be made:
form of dimensionless concentration vs dimensionless time.
The static holdup determined by the present method is
The experimentally obtained concentration values were much less compared to the values reported by Shulman et al.
made dimensionless on the basis of the conductivity of inlet
(1955). A similar observation has been made by Schubert et
water. The area under the c vs t curve was calculated numer- a l. (1986). This difference can be attributed not only to the
ically to find k,. The area under the curve up to “ D” (Fig. 1) is different methods of measurement used but also to the defini-
equal to the mean residence time of the dynamic liquid
phase. Based on td the time axis was made dimensionless for
further parameter estimation. Table 2.
To find the characteristics of the response curve., a plot of
In (c) YS l was made. The slopes of the first and second linear (a) MPR:
portions and the intercept of the extrapolated second linear k = 0.1024 l’~“ .5z’ (SD = 5.74%)
portion were measured. Using the method of graphical solu- a’= 0.0408 v-o.597 (SD = 7.39%)
tions of Bennett and Goodridge (1970) the preliminary esti- Pe = 43.034 b1.0s3
I (SD = 8.75%)
mates of Pe, k,/k, and q were found. These estimates were (b) CRR:
fed to the computer program which minim&d the sum of h = 0.1088 V” .4887 (SD = 5.07%)
squares due to residuals between experimental and predicted aI= 0.0852 i+J67 (SD = 9.8%)
response curves. The results of the computer program were Pe = 18.011 Y0.‘s5
I (SD = 7.4%)
more accurate than the graphical estimates.
 Shorter Communications 761

the liquid held up in the packed bed. A significant portion of

this held up liquid may have been a part of the dynamic
holdup before stoppage of flow. A s against this in the present
method only that part of the held up liquid which exchanges
mass slowly with the dynamic holdup has been determined
as the static holdup.
The static holdup shows a clear increase with liquid veloc-
ity. A t higher liquid velocities it attains a constant value. A s
liquid velocity increases more and more surface is covered by
the flow of liquid due to increased radial liquid spreading.
Correspondingly, the static holdup, which is also dependent
upon the wetted surface, is found to increase with liquid
velocity. A s mentioned earlier in the present work the pack-
ings were not preflooded or prewetted. Thus, the static hold-
up was not a function of the history of the packing surface. In
0.01 I I 1 I / 1 the method using preflooding (Schubert et al., 1986) all the
0 0.2 0.5
packing is prewetted and therefore the effect of liquid veloc-
LIQUID VELOCITY, VL(10-2m.i’) --
ity on static holdup is not noticed.
Previous workers obtained a constant value for the static
Fig. 2. V ariation of total holdup with liquid velocity: holdup while determining it by the draining or tracer method
(0) CRR, (0) M PR, (A ) Shulman et al. (Shulman et al., 1955; Bennett and Goodridge, 1970;
Schubert et al., 1986). These investigators deduced h, by
extrapolating the plot of h, vs 4 to V ! = 0. Obviously h, vs
y plot is unique for each packing and its intercept on the
Y-axis has also a unique value. Therefore this method can
not differentiate h, for different liquid velocities. Indeed the
value obtained by the extrapolation procedure is h, at V , = 0
or under non-irrigated conditions. It is probably this lacuna
which led these investigators to believe that h, is constant.
In the case of the Pall rings also liquid pools can build at
contact points of two packings. However, because of the
slotted geometry of the Pall ring a larger number of these
pools may interact with the dynamic liquid than in the case
of the Raschig ring. Thus, the M PR exhibits a very low static
holdup as defined by Bennett and Goodridge.

Dynamic holdup, h,
Figure 5 shows the dynamic holdup data obtained for the
packings investigated The data of Shulman et al. are also
shown in Fig. 5. The h, values determined by the tracer
method are again much higher (about 60% ) than the values
of Shulman et al. In both methods h, is determined from
Fig. 3. V ariation of holdup ratio with liquid velocity: a knowledge of h, and h,:
(Cl) M PR, (0) CRR.
h,=h,-hh,. (1)
A s discussed earlier, h, and h, values obtained by the tracer
method are higher and lower, respectively, than those ob-
tained by the draining method. Consequently, h, values
obtained from eq. (1) show considerable diflerence.

Peclet number, Pe
Figure 6 shows the variation of Pe with V ,. The increasing
trend of Pe with V, is similar to that observed by Bennett and
0012
0011
o-010
0009 0 08

0.008

0 007
0.006

OW5
0.002 0004
0 0.2 04
LMJn VELoctlY, v, m’m 6’ 1 -

Fig. 4. V ariation of static holdup with liquid velocity:

(A ) Shulman et al., (0) CRR, (0) M PR.

0 04
LtQUtD “EtoctTV”,:,~ld2~.i’) -

tion of static holdup which is different in the two cases. The Fig. 5. V ariation of dynamic holdup with liquid velocity:
static holdup obtained by the draining method is simply all (0) CRR, (El) M PR, (A ) Shulman et al.
762 Shorter Communications

ition region data for pipe flow (Levenspiel, 1969) where the
following approximate proportionality can be derived:

Pe D a. Re’-‘. (8)
The exponents on Re in eqs (7) and (8) are not significantly
different. It can therefore be concluded that dispersion in
packed beds is equivalent to that in pipe flow when the
Peclet and Reynolds numbers are defined with respect to the
dynamic liquid region. It is likely that eq. (7) may also hold
for other random packings. This possibility, however, needs
td be tested with more extensive data.

Liquid exchange coeficient, T,

The computer program yields values of the dimensionless
liquid exchange coefficient, q, which can be converted into
the dimensional value, k, as follows:
Fig. 6. Variation of Peclet number with liquid velocity:
(El) MPR, (0) CRR. k = TJtd. (9)
It was observed that k increases with V,. This behaviour can
be explained by the fact that as liquid velocity increases the
level of turbulence in the dynamic holdup increases and
therefore the rate ofexchange of mass between the static and
dynamic holdup increases. The k values obtained for the
Goodridge. The present data for the CRR can be compared CRR are in the same range as those reported by Bennett and
with the literature data of Dunn et al. (1977) who also used Goodridge (1970).
a step input to obtain axial dispersion characteristics of It can be argued that k is decided jointly by the turbulence
0.025-m CRR and Berl saddles. Dunn et al. correlated in the bed and the location of the static liquid regions. Both
Peclet number with the liquid mass velocity as follows: these factors are in turn governed by the packing shape and
Pe = 0.038 x 10’4.93” lo-‘L’) bed geometry. For instance, the CRR with its closed ge-
d (2) ometry and continuous surface can build liquid prisms, un-
A similar correlation was derived from the present data: like the MPR. This difference has been discussed earlier. It is
therefore unlikely that a single correlation on the same lines
Pe = 0.0185 x 10’2.15 x 10_4L’)
d (3) as that of Bennett and Goodridge can be obtained for the
The Pe, values obtained in the present work are consider- MPR and CRR. Indeed, all attempts at a single correlation
ably higher than those of Dunn rr ~1. It should be noted that satisfying both the MPR and CRR failed. This failure can be
the model used by Dunn et al. neglected the static holdup attributed to the reasons discussed above and implies that
and therefore the consequent slow exchange of the tracer further characterisation of the static holdup locations and
material in h, with h,. This latter phenomenon causes an the local turbulence for different packings is needed. The
increase in the spread of the distribution. Dunn er al. in- presently available information is insufficient for this pur-
cluded this additional spread in the dispersion in the dy- pose. Therefore the following separate correlations for the
namic region and therefore obtained lower Pe,. The values of MPR and CRR are proposed (Fig. 8):
Pe, obtained in this work are, however, in the same range as MPR: kd,, = 3.843 x lo-“‘Reg 699 (SD = 3.7%) (10)
those of 0.065- and 0.095-m Raschig rings used by Bennett
and Goodridge. CRR: kd,, = 4.83 x 10-8Re~.571 (SD = 7.6%) (11)
The two packings used here have different geometries. where d., = 6(1 - ~),/a~,
Whereas the CRR is a closed packing, the MPR has a highly c: = dry-bed voidage,
open geometry. The beds formed by these packings and the ap = geometric surface area of the packing (mZ/m3).
liquid flow in them is therefore expected to be considerably
different. In the model used for our data analysis the disper-
sion is solely restricted to the dynamic region. Based on the
treatment of dispersion in pipe tlow it was thought that
dispersion in packed beds can also be similarly treated if

‘.Or------
a relevant Reynolds number, Re,, for the dynamic region is
used:

Re, = d, ~‘,P,/P, (4)

1/
A modified Peclet number for the dynamic region was de-
fined as
Pe, = V,d,/D, (9
I
0
0
where O-1

V, = Y/h, (6)
B
and d, is the diameter of the dynamic liquid stream having
a cross-sectional area A, = Q,/k’,.
The above definition of Re, is expected to incorporate the
effect of packing characteristics and the effect of bed ge-
ometry due to the inclusion of h, which is characteristic of
the particular packing for a given K. Data for both the 0 0, 1
packings were subjected to regression analysis and the fol- 100 1000 101 30

lowing correlation was obtained: Reo -

PP, = 2.195 x 10-6(Re,)‘.46 (SD = 15%). (7) Fig. 7. Variation of Peclet number with Reynolds number:
The above correlation can be compared with the trans- (0) CRR, (0) MPR.
 Shorter Communications 763

0.02 . k exchange coefficient between static and dy-

namic holdup, l/s
L length of the packed bed, m
O-01 - ,~,,,, / L liquid flow rate, ib/h/ft2
Pe Peclet number based on the length of the bed
0,005 CLV,lD,)
P%i Peclet number based on the size of packing
Cd,b/D, )
Peclet number defined by eq. (5)
volumetric liquid flow rate, m’/s
liquid phase Reynolds number defined by
0001 eq. (4)
liquid exchange coefficient between static and
dynamic liquid holdups (kc,)
OJX05 time, s
0002 100
; , ,
0 r;ldoo mean residence time of dynamic liquid phase, s
o-a03
liquid velocity in the dynamic holdup as de-
fined by eq. (6), m/s
-3
superficial liquid velocity (Q,/A ), m/s

Fig. 8. V ariation of kd,, with Re,: ( q) CRR, (0) M PR.

Greek letters
a ratio of static holdun to dvnamic holdup
Wk,)
PI density of the liquid, kg/m3
P’r viscosity of the liquid, kg m/s
Comparing the power index of Re, in the above equations
it can be concluded that the mechanism of liquid exchange is
similar in both cases.
REFEREN CES
CONCLUSIONS Bennett, A . and Goodridge, F., 1970, Hydrodynamics and
(a) The data obtained by the present method indicate that mass transfer studies in packed absorption columns.
the static holdup so determined is less than that obtained by Trans. Instn them. Engrs 48, T232.
the draining method. This observation is in agreement with Carberry, J. J. and Bretton R. H., 1958, A xial dispersion of
the conclusions of Bennett and Goodridge (1970) and mass in flow through fixed beds. A.I.Ch.E. J. 4, 367.
Schubert et al. (1986). Dunn, W . E., V ermeulen, T., W ilke, C. R. and W ord, T. T.,
(b) The static holdup is found to be a function ofthe liquid 1977, Longitudinal dispersion in packed gas-absorption
velocity below a certain liquid velocity unlike the findings of columns. Ind. Engng Chem. Fundam. 16, 116.
previous investigators. A rational explanation has been given Hoogendoorn, C. J. and Lips, J., 1965, A xial mixing of liquid
in support of this observation. in gas-liquid flow through packed beds. Can. J. chern.
(c) The present work provides more meaningful holdup Engng 43, 125.
and dispersion data for larger packed columns using com- Kan. K. M . and Greenfield, P. F., 1983, A residence-time
mercial sizes of packings, especially at low liquid flow rates. model for trickle-flow reactors incorporating incomplete
mixing in stagnant regions. A.I.Ch.E. J. 29, 123.
Acknowledgement-One of the authors (KBK) wishes to Levenspiel, O., 1969, Chemical Reaction Engineering. W iley
thank M S Kevin Enterprise, Bombay, for sponsoring him for Eastern, New Delhi.
the research work. Patwardhan, V . S., 1978a, Effective interfacial area in packed
beds for absorption with chemical reaction. Can. J. them.
Engng 56, 56.
K. B. KUSHA LKA R Patwardhan, V . S., 1978b, G as-liquid reactions in a packed
V . G. PA N G A RKA R bed: some implications of the extended crossflow model.
Department of Chemical Technology Can. J. them. Engng 56, 558.
University of Bombay Patwardhan, V . S., 1979, The effectiveness of static hold-up
Matunya, Bombay 400019, India for absorption with a chemical reaction in a packed trickle
bed: general order kinetics. Can. J. them. Engny 57, 582.
Patwardhan, V . S., 1980, G as-liquid reactions in packed
beds: regimes of reaction in the static hold-up. Can. J.
NOTA TION
them. Engng 58, 454.
AD cross-sectional area of dynamic stream, mz Patwardhan, V . S., 1981, G as-liquid reactions in packed
% geometric area of packings, l/m beds: the effectiveness of static hold-up in presence of gas
C concentration of tracer in the dynamic liquid side resistance. Can. J. them. Engng 59, 483.
phase, kmol/m3 Patwardhan, V . S. and Shrdtri, V . R., 1981, M ass transfer
CO initial concentration of the tracer in the dy- coefficient between the static and dynamic holdups in
namic liquid phase, kmol/m3 a packed column. Chem. Engng Commun. 10, 349.
c dimensionless concentration of the tracer in Sater, V . E. and Levenspiel, O., 1966, Two phase flow in
the dynamic liquid phase (C/C,,) packed beds. Ind. Engng Fundam. 5, 89.
liquid phase axial dispersion coefficient, r&/s Schubert, C. N., Lindner, J. R. and Kelly, R. M ., 1986,
nominal packing size, m Experimental methods for measuring static liquid holdup
equivalent diameter of packing as defined by in packed columns. A.I.Ch.E. J. 32, 1920.
eq. (1% m Shulman, H. L., Ullrich, C. F. and W ells, N., 1955, Perform-
total, dynamic and static Liquid phase holdup, ance of packed columns. 1: Total, static, and operating
respectively holdups. A.I.Ch.E. J. 1, 247.
van Swaaij, W . P., Charpentier, J. C. and V illermaux, J.,
1969, Residence time distribution in the liquid phase of
‘A uthor to whom correspondence may be addressed. trickle flow in packed columns. Chem. Engng Sci. 24.1083.