Siegel, R., and J. M.

Savino, “An Analysis of the Transient , “Die asyrnptotischen Losungen fur Vorgange
Solidification of a Flowing Warm Liquid on a Convectively des Erstarrens,” Intern. I. Heat Mess Transfer, 19, 597
Cooled Wall,” Proc. 3rd Intern. Heat Transfer Conf., Vol. 4, (1976).
pp. 141-151, Chicago, 111. ( 1966). Tao, L. C., “Generalized Numerical Solutions of Freezing a
Stephan, K., “Influence of Heat Transfer on Melting and Solid- Saturated Liquid in Cylinders and Spheres,” AIChE J., 13,
ification in Forced Flow,” Intern. J. Heat Mass Transfer, 12, 165 ( 1967).
199 (1969).
-, and B. Holzknecht, “Warmeleitung heim Erstarren
geometrisch einfacher Korper,” Wurme-und StofFiibertragung, Manuscript received December 6, 1976; revision received September
7,200 ( 1974 ). 27, and accepted October 12, 1977.

Droplet Size Spectra Generated in

Turbulent Pipe Flow of Dilute
Liquid/Liquid Dispersions
A. J. KARA 0E LAS
Experiments were carried out with water dispersed in hydrocarbons at Westhollow Research Center
various flow rates. The measured size spectra can be well represented by Shell Development Company
either an upper limit log-probability function or a Rosin-Rammler type of P.O. Box 1380
Houston, Texas 77001
equation with constant parameters and only one variable, the maximum drop
diameter d,, or dss.The latter can be predicted satisfactorily by a correla-
tion based on the Hinze/Kolmogorov model of droplet breakup.

Reliable estimates of the distribution of droplet sizes Sleicher (1962) predict maximum droplet sizes d,,, in
in flowing liquid/liquid dispersions are required in many turbulent pipe flow, which differ almost by an order of
areas of chemical engineering practice, for example, in the magnitude for flow conditions and liquid properties of
design of chemical reactors and of physical separation practical interest. Moreover, according to the Hinze model,
processes. At present, it is impossible to make such a priori d, is approximately inversely proportional to the mean
estimates even in common problems, including pipe flow velocity whereas in the Sleicher correlation, d,, varies
of dilute dispersions, owing to insufficient information with u - 2 . 5 . It is clear, therefore, that in the problem con-
about the effect of flow conditions on the size spectrum. sidered here, the literature provides very little guidance of
Collins and Knudsen (1970) have published the only questionable reliability.
available drop size distribution data, obtained in well-
The need to describe dispersions of water in flowing
defined turbulent pipe flow with water as the continuous
hydrocarbons provided the motivation for this work. The
phase. These authors reached a rather disturbing conclu-
main objective was to make accurate measurements of
sion, namely, that their data did not conform to any of
droplet size spectra generated in turbulent pipe flow. In
the presently available distribution functions, such as the
lognormal, upper limit log-normal, etc. In order to in- the absence of a satisfactory theory, such measurements
terpret the measured size spectra, Collins and Knudsen are necessary in order to select the most representative dis-
developed a computer aided stochastic model by postu- tribution function and to study the influence of flow con-
lating a breakup mechanism in which the maximum stable ditions and liquid properties on some physically significant
drop size d,, is a basic input parameter. However, the parameters of the size spectrum, for example, the Sauter
presently available correlations due to Hinze (1955) and mean a2 or the maximum droplet diameter d,,.

CONCLUSIONS AND SIGNIFICANCE

Accurate data of water drop size spectra have been ob- and 4), the only variable being the characteristic size
tained by a relatively new droplet encapsulation technique. d,, or ds5. The latter can be predicted independently as
Two liquid hydrocarbons of viscosity approximately 2 and a function of flow conditions and liquid physical properties.
20 mN s/m2 have been used as the continuous phase in a The maximum drop sizes, or dg5,appear to be approxi-
5.04 cm I D pipe loop with a straight test section 32 m mately inversely proportional to the mean flow velocity,
long. The measured spectra show a remarkable similarity in agreement with the theory of Kolmogorov (1949) and
and can be satisfactorily represented by either an upper Hinze (1955). A correlation proposed by the latter author,
limit log-probability function or a Rosin-Kammler type of essentially untested so far for pipe flow of dilute disper-
equation with nearly constant parameters (see Tables 2 sions, has been reexamined and slightly modified to im-
0 The American Institute of Chem-
0001-1541-78-9955-0170-$1.3~
prove its accuracy at high Reynolds numbers. Our data
ical Engineers, 1978. are generally in good agreement with this correlation

Page 170 March, 1978 AlChE Journal (Vol. 24, No. 2)

[Equation (23)J. Additional data (Kubie and Gardner, are expected to provide guidance in future modeling
1977) that became available after this paper was sub- studies. For engineering calculations in particular, the re-
mitted for publication are also in agreement with this sults of this study suggest that the drop size spectrum can
correlation. be estimated by using the Rosin-Rammler equation with
In general, the data reported in this paper help delineate a constant slope n N 2.5 and with dg5 computed from
the basic features of drop size spectra in pipe flow and Equation (23).

PREVIOUS WORK tively short test section of 3.81 cm ID which were not
in agreement with predictions based on Equation ( 5 ) . A
Basic studies of droplet breakup in homogeneous and transient type of technique was employed in these tests;
isotropic turbulent flow were first presented by Kolmo- that is, large drops of equal size were first introduced into
gorov (1949) and Hinze (1955). The main assumption in the pipe when the flow rate was very small, and then the
these studies is that drop fragmentation is caused by flow was accelerated, very fast, to the desired higher
dynamic pressure forces, due to rapid turbulent fluctua- level. It was considered that the maximum stable diame-
tions in the vicinity of the drops, which overcome inter- ter d,,, had been obtained when less than 20% of the
facial tension forces. The mean square of velocity fluctua- photographed drops had diameters smaller than the
tions z, over a distance equal to the drop diameter d, original size. Data obtained by Collins (1967) suggest
provides a measure of the dynamic pressure forces. There- that the line length used in this study may have been
fore, an estimate of the maximum stable drop diameter insufficient (6.7m or l'i6 pipe diameters of fully devel-
dma,can be obtained from the condition oped flow) for the large drops to reach a stable maximum
size.
Pr 3 4u
-=c'- (1) Sleicher took into account viscous forces and correlated
2 &ax his data by the following expression
For an isotropic and homogeneous flow field, it has been
shown that z i s proportional to ( ~ d ) 2 / if3 the length d
is much greater than the Kolmogorov microscale TJ =
(6)
( v3/6) lI4. It follows that
which is very different from Equation ( 5 ) . For instance,
it suggests that d,,, is independent of pipe diameter,
which is in sharp contrast to the dependence of
Equation ( 5 ) . Paul and Sleicher (1965) published later
where the constant C must be determined from experi- a few data points taken in a 1.27 cm ID pipe, which
mental data. Hinze (1955) interpreted data taken by showed only a slight diameter effect (d,,, cc D-".'). The
Clay (1940) in a Couette flow field and obtained the constant C = 43 in Equation ( 6 ) gave a better fit of these
value C = 0.725 corresponding to dg5.The main assump- data than C = 38. In subsequent calculations with this
tions made in the derivation of Equation ( 2 ) are, of equation, the value C = 40 will be used.
course, open to criticism. For instance, turbulent shear Levich (1962) provides an interesting discussion of
fields of practical interest are far from isotropic and homo- drop fragmentation mechanisms in his book and reviews
geneous. Nevertheless, Equation ( 2 ) has been found to related Russian work. He points out that the Kolmogorov
describe quite satisfactorily experimental data obtained model can give good estimates of stable drop diameters
in agitated vessels; see, for example, Mlynek and Resnick in the core of turbulent pipe flow but not in the region
(19i2), Shinnar (1961), and Sprow (1967). close to the wall, In this region of large velocity changes,
In order to estimate d,,, or dg5 in pipe flow, the mean much smaller drops will develop, with the smallest drop
rate of energy dissipation per unit mass of fluid Tmay be size to be found at the edge of the sublayer where
used
(7)
(3)
Obviously, during the transient period of dispersion for-
with the friction factor calculated from the Blasius equa- mation in a pipe, the drop diameter will vary between a
tion for smooth Ltmes
L maximum, possibly given by the Kolmogorov-Hinze model
0.0791 for the core, and a minimum resulting from fragmentation
f=- (4)
Re114 at the wall, as suggested by Levich. At present, it is im-
possible to predict the equilibrium drop size distribution
and the constant C = 0.725 obtained in Couette flow. The as a function of flow conditions and liquid physical prop-
result is erties, and accurate experiments are necessary to gain
some insight into this problem.
Collins and Knudsen (1970) have published the only
available drop size distribution data,' obtained in a 11.3
where We = ( DpeT2)/ ( U ) is a dimensionless Weber num-
0 AAer this paper was submitted for publication, a similar study
ber based on the pipe diameter D and mean velocity was published by Kubie and Gardner (1977). These authors, using a
Equations ( 2 ) and ( 5 ) will be reexamined in this paper photographic technique, measured drop size spectra but mede no at-
tempt to correlate them. They did conclude, however, that the mea-
and compared with data obtained in our study. sured maximum drop sizes were in good agreement with the Hinze
Sleicher (1962) made measurements of dmax in a rela- correlation [Equation (5)l.

AlChE Journal (Vol. 24, No. 2) March, 1978 Page 171

m long, 1.89 cm ID pipe, The dispersion of three organic
liquids i n water was studied at relatively high velocities
(4.3 to 6.1 m / s ) . D r o p size distributions were obtained
from photographs. For the prediction of d,, Collins
and Knudsen (1967, 1970) preferred the Sleicher corre-
lation, although they did not systematically compare their
d a t a to existing models. In order to explain the measured
d r o p size distributions, they developed a computer-aided
stochastic model in which d,, is a basic input parameter.
Owing to their significance, some of Collins' d a t a will
be reexamined and compared with the results of this study
in the Appendix. In the following sections we shall first
describe our experiments a n d then present and interpret
the data.

DESCRIPTION OF EXPERIMENTS
Fig. 1. Schematic of flow system.
Flow Loop
A schematic of the flow system is shown in Figure 1. The test A sampling vessel (No. 8, Figure 1) was made out of a
section was made of acrylic pipe sections, with very smooth transparent acrylic pipe section, -1.15 m long, with an inside
joints, and had a total length of -32.3 m. The true inner di-
ameter was 5.04 cm. Fluid was circulated by a rotary type of
pump (manufacturer Bowie Company, 500 series). A Schroeder
diameter of 13.9 cm. This vessel was equipped with a gau e
and a regulator (11) in order to adjust the pressure and wit a-
draw fluid from the pipe under nearly isokinetic conditions. It
LF-1 in-line filter (No. 2, Figure 1) was installed at the dis- was filled up to about 7'0% of its capacity with kerosene con-
charge end of the pump to remove small particles or droplets. taining 0.2% wt dissolved terephthaloyl chloride. This com-
A pressure relief diaphragm ( 3 ) and a pulsation dampener ( 4 ) pound was first dissolved in 300 to 400 cm3 of carbon tetra-
were necessary to protect the pipe against pressure surges and chloride and then mixed with an appropriate amount of kero-
to eliminate flow pulsations, respectively. A calming section ( 5 ) , sene. The short sampling tube had a conical end ( 9 ) , immersed
made of a 15 cm I D pipe, served the dual purpose of dis- approximately 2 em into the fluid, in order to reduce the dis-
engaging gas bubbles from the fluid and providing a uniform charge velocity of the droplets into the vessel, thus increasing
stream into the main line. their settlin time. Upon contact of the droplets with the
A special water injection facility ( 6 ) was used. Water was sampler fluii, the piperazine dissolved in the water reacted
released at a steady rate ( 5 to 12 cm3/s) into the main stream, with the terephthaloyl chloride forming a protective coating.
through a 6 mm steel tube, Under these conditions, that is, low The bottom of the vessel was covered with a layer of viscous
relative jet velocity and fairly large jet diameter, it is expected silicone oil ( Dow Corning 200, 10100 centistokes) approximately
that the jet breakup is controlled almost entirely by the tur- 7 mm thick. The function of this layer was twofold: to terminate
bulence in the main stream. Moreover, it is reasonable to the reaction between droplets and the continuous phase which,
expect that the largest droplets developing upon disintegration beyond a certain time, tends to produce rough coatings on the
of this jet are greater than the ultimate maximum stable droplet surfaces and to result, occasionally, in agglomerates; to
diameter generated by the pipe flow; that is, the process of prevent water wetting of, and droplet sticking onto, the bottom
breakup continues downstream of the injection tube. Experi- plate and to preserve the spherical shape of the settled drops.
ments by Collins (1967) with similar injection tubes, and The density of the silicone oil was between that of kerosene
larger main flow velocities than those of our experiments, show and water. The bottom of the vessel on which droplets de-
exactly this trend. posited was made of optical quality, thick glass plate. As
Pictures of the flowing dispersion were taken at a special indicated in Figure 1, the lower section of the sampling vessel
test section ( 7 ) , and samples were withdrawn simultaneously ( 1 0 ) could be disengaged by using a Victaulic connector.
through a short 1.14 cm ID tube, located approximately 29 This short section was transferred to a stand without disturbing
and 30 m, respectively, downstream of the injection point. the already settIed drops, and pictures were taken from the
The center line of the sampling tube was approximately bottom.
1.45 cm below the pipe center line. This technique was in general very satisfactory. Sampling
A IIalliburton turbine meter ( 1 2 ) was used to monitor the over a period of 3 to 5 s provided a sufficiently large number
flow rate. TWO: a l p steel tanks were employed. One of them of droplets which were distributed on the bottom plate without
(14) was connected to the suction side of the pump and con- excessive crowding and overlapping. Pictures with total magni-
tained clean fluid; the other (13) was used only to collect the fication five to twenty-three times the real object were used to
contaminated fluid while water was injected into the system. measure droplet size distributions under various pipeline flow
This arrangement prevented reentrainment of water droplets conditions. Usually all the droplets in a picture were measured
into the test section. with an accurate ruler under additional magnification. In most
Droplet Size Measurement Techniques cases more than one picture was used to obtain a total count
Droplet Encapsulation. With this method it is possible to of more than 300 drops.
prevent droplet coalescence and preserve the size distribution, Photographic Technique. During the tests with kerosene,
in a properly taken sample, as follows. A protective coating is photographs of the dispersions were taken, concurrently with
formed quite fast on the surface of each droplet, when a small the droplet encapsulation data, in order to compare the two
quantity of a certain monomer, present in the dispersed phase, techniques. A test section very similar to that used by Collins
comes in contact with another monomer dissolved in the con- (1967) was located approximately 0.8 m upstream of the
tinuous phase. Recently, Mlynek and Resnick (1972) suc- sampling tube. Two diametrically opposite holes were drilled
cessfully employed this technique to measure drop size dis- with their centers on the vertical axis of the pipe. Optical
tributions in an agitated vessel. The reacting compounds in quality glass disks were mounted essentially flush with the
their experiments were 0.05% wt piperazine in the aqueous, pipe surface and held in place by a Plexiglas collar.
continuous phase and 0.05% wt terephthalic acid chloride in The camera was focused at a distance of 6.4 mm from the
the dispersed phase, which was a mixture of carbon tetrachlo- inner pipe surface. The viewing ports and camera were aligned
ride and iso-octane. The same additives were used in our study. on an axis approximately 15 deg from the vertical pipe diameter.
A concentration 0.75% wt anhydrous piperazine dissolved in A General Radio Strobotac, type 153-AB light source was
deirnized water gave the best results. Indigo-caramine dye was located diametrically opposite to the camera, 23 cm from the
added to the water to obtain a deep blue color before injection pipe. The flash duration was -0.5 +s. A Nikon F camera was
into the line. Two different liquids, that is, kerosene and used with 1.4 Nikon lens and a specially made 43 cm extension
viscous ti-ansformer oil, were used as the continuous phase. between camera body and lens to obtain the desired magnifica-

Page 172 Match, 1978 AlChE Journal (Vol. 24, No. 2)

 TABLE1. EXPERIMENTAL
CONDITIONS

Run Mean Temperature, Density, pc Viscosity, pc Water injection Continuous

number velocity, m/s "C g/cm3 mN s/m2 rate, cm3/s Phase

1 1.52 22.2 0.798 1.82 6.0 Kerosene

2 1.52 22.2 0.798 1.82 6.0
3 2.98 22.2 0.798 1.82 11.9
4 1.18 22.5 0.798 1.81 4.7
5 2.57 21.1 0.799 1.86 10.3
6 2.22 23.3 0.797 1.77 8.9
7 1.84 21.9 0.798 1.83 7.5
11 3.00 20.6 0.892 16.00 11.8 Transformer oil
12 1.19 21.4 0.892 15.42 6.3
14 1.52 17.5 0.894 18.27 6.1
15 1.86 18.6 0.893 17.43 6.1
16 2.24 21.4 0.892 15.42 8.9
17 2.60 26.1 0.888 12.47 10.3
18 2.08 23.3 0.890 14.12 8.4

tion. The camera setting was F-4 at 1/30 with F X synchroniza- quired for droplet encapsulation. Interfacial tension measure-
tion. Panatomic X film was used and developed in HC-110 ments with a DuNouy tensiometer gave the following results:
mixture B. The negatives were enlarged six times and prints
were made on Polycontrast F pa er. Kerosene/water u = 33.1 mN/m at 22.8%
The same type of camera an{ film were used to photograph 31.7 mN/m at 24.4"C
the samples of encapsulated droplets. Nikon bellows and a Trans. oil/water u = 35.0 mN/m at 17.8"C
Nikon F-2 lens were employed. A camera setting of F-4 gave 33.7 mN/m at 23.3"C-
the best depth of field and sharpness in this case. In order to 32.7 mN/m at 26.7"C
light the sample tray, a photo-flood was mounted 0.6 m above Experimental conditions are summarized in Table 1. Mean
the sample. The light was filtered through a sheet of diffusion velocities varied between -1.2 and 3.0 m/s. After a pipeline
paper located 30 cm above the sample. flow rate was selected, the water injection system was adjusted
Experimental Conditions to provide a steady injection rate of 0.20% by volume. Only
Experiments were carried out with kerosene (Dispersol by in runs No. 12 and 15 the water volume concentration was
Shell Oil Company) and a more viscous transformer oil 0.26% and 0.16%, respectively.
(Texaco Transformer Oil No, 55) as continuous phases. The Samples and line photographs were taken simultaneously.
dispersed phase was deionized water with the additive re- The raw data obtained from direct droplet size measurements
were digitized and processed in the computer. In order to
obtain size distributions and-&2, size intervals of 100 pm
( u p to 1000 pm) and 200 pm (above 1OW pm) were
selected.

-
- 001
- 005
- 010
- 0 20
- 030
- 040 >
- 0 50 z.
- 060 0
;
0
- 010 2
LL

- 080 UJ
I
- 085 3
- 090 2
0

1
c
- 0 9 5 43
2
3
0

01 02 05 1 2 5 10

PARTICLE DIAMETER, d. rnm

PARTICLE DIAMETER, d, mm
Fig. 2 Drop size distribution of water in kerosene, represented by a Fig. 3. Drop size distribution of water in kerosene, represented by a
Rosin-Kammler type of function. = 2.98 and 2.22 m/s. Rosin-Rammler type of function. = 2.57 and 1.18 m/r.

AlChE Journal (Vol. 24, No. 2) March, 1978 Page 173

RESULTS
Figures 2 and 3 show data obtained from samples of 0 RUN 65
0005
water dispersed in kerosene at various velocities. The
0 RUN 35
-
DO1
linearity of these data, plotted in logarithmic coordinates,
RUN 35
suggests that the Rosin-Rammler ( 1933) equation 0 02

v = exp [- (;)n] 0 05

0 10

provides a very satisfactory representation of the measured

0
z 020
distributions, especially at the large diameter end of the V
F
spectra. In Equation (8), V is the cumulative volume frac- a 030

tion of particles with diameters greater than d, and the LL

Lu 040
parameters n and d* are the slope of the line and the 5 050
diameter corresponding to V = 0.3679, respectively.
Alternatively, one can use the fraction V corresponding
9 060

070
to any other characteristic diameter to specify the distri- L'
3
bution. If do5 is used, Equation (8) becomes +-4 080
3

V = exp[ - 2.996 ($-)" ] (9)

2
2 090

0 95
An equally good, and in many cases better, fit of the
data presented here is obtained by the upper limit log- 0 98
probability function, suggested by Mugele and Evans 0 99
(1951). The volume distribution function in this case is 0 995
given as
d(1-VV)
--6 exp(-
- S2z2) (10) 01
1 1
02
I 1
04
I
06
I I I l l
1
1
2
1
4
1 1
6
dz G d/(d,,,-dl

and upon integration Fig. 4. Drop size distribution of water in kerosene, represented by an
1 upper limit log-probability function. u=
2.98 and 2.22 m/s.
exp( - x2)dx = - [I - erf(Sz)]
2 the data. Drop size distribution parameters for all the
(11) runs made with water-in-kerosene dispersions are sum-
where
z=~n[ ad dl marized in Table 2. This table also includes &, calcu-
lated directly from the raw data, as well as the diameter
of the largest drop in a sample.
dmax -
and
a=
&ax - d50
d50
Three parameters are required to specify this distribution,
that is, dmaX,
a, and S, and can be obtained from the data R U N 45
as
&ax - d50(d90 + dl0) - 2d90dlo 0 05
-- (14)
d50 d502 - dgodio > 010
and 2

S=
0.394 ; 020

[ ""1
0
2 030
log10 LL

050 Lu 040
I
3 050
where ui = di/(d,,, - di) . Given the three parameters,
9 060
the Sauter mean diameter 2 3 2 , and dy5 are calculated as Lu
2 070
F
&,X 5 080
T2 = (16) 3
I
2 090

and 0 95

0 98
a
d95 = 0 99

+ -a1 exp [+I

1 163
(17)
0 995
1
1 1 I I I I 1 1 1 1 1 I I
The data shown in Figures 2 and 3 have been replotted 01 02 05 1 2 5
dl(dmax-dl
in log-probability coordinates in Figures 4 and 5 , re-
spectively. It is evident that the upper limit log-probability Fig. 5. Drop size distribution of water in kerosene, represented by an
function provides a very satisfactory representation of upper limit log-probability function. r=2.57 and 1.18 m/r.
Page 174 March, 1978 AlChE Journal (Vol. 24, No. 2)
 TABLE2. DROPSIZEDISTRIBUTION
PARAMETERS
FROM DATAWITH Low VISCOSITY
CONTINUOUSPHASE(KEROSENE).
DATAOBTAINED
FROM SAMPLES

Rosin-Rammler

Run number
-U , m/s
Velocity
a
Upper limit log-normal distribution
6 &ax, pm
distribution Data

3s 2.98 1.174 0.915 1074 808 416 2.90 820 407 756
5s 2.57 1.339 0.827 1404 1057 479 2.53 1075 455 1111
6s 2.22 1.163 0.965 2087 1547 828 3.07 1550 794 1547
7s 1.84 1.535 0.809 3600 2638 1108 2.31 2650 1066 2603
lso 1.52 0.938 0.953 4767 3733 2133 3.30 3800 2083 4242
2s 1.52 1.495 0.881 5338 3814 1743 2.52 3850 1887 4310
4s 1.18 1.269 0.845 5491 4 158 1960 2.58 4250 1942 5172
Arithmetic average 1.329 0.874 2.65

* Excluded from averaging.

f Diameter of largest measured drop.

A number of interesting observations can be made, TABLE3. COMPARISON

BETWEENDATAFROM SAMPLES
AND

based on the results of Table 2. The characteristic diam- PHOTOGRAPHSOF FLOWING

DISPERSIONS.
CONTINUOUS
PHASE:
KEROSENE
eters dg5obtained by fitting the data with an upper limit
log-probability function are at most 2% smaller than d g s
obtained with the Rosin-Rammler equation. This close
agreement is due to the fact that in the range V N 0.05
Run
number
-u,m/s
Velocity
n
Samples
d95, fim
Line photographs
n d 9 5 , pm

to 0.95 the deviation of the Rosin-Rammler from the log-

probability distribution is small and ( omparable to the 3 2.98 2.90 820 3.88 830
5 2.57 2.53 1 075 4.03 1100
standard deviation of the data points. The simple Rosin- 3.07 1550 3.86 1700
6 2.22
Rammler equation, with only t h o parameters, is oov,ously 7 1.84 2.31 2 650 4.25 2 450
more convenient for engineering calculations. It should be 2 1.52 2.52 3 850 3.64 4 700
stressed, however, that the upper limit log-probability func- 4 1.18 2.58 4 250 4.27 6 250
tion is more accurate at both ends of the size spectrum. Arithmetic
In particular, the parameter d,,,, has an obvious physical average 2.65 3.99
significance, and it is very appropriate for the description
of processes and/or data characterized by a maximum The good agreement, shown in Table 2, between $2
stable size. Other advantages of the upper limit log-prob- computed directly from the data and from Equation (11)
ability function are discussed in detail by Mugele and is also a measure of success of the log-probability func-
Evans ( 1951). tion. However, the diameters of the largest measured
drops are closer to d 9 5 than to d,, obtained from Equa-
tion ( 11). This can be explained as follows. Assuming that
the upper limit log-probability equation describes pre-
0 01 cisely drop distributions, the probability of having a parti-
cle of size d approaching d,, is infinitely small. More-
0 05
0 10
over, for most of the distributions measured in this study,
the upper 5% by volume, associated with d 9 5 , corre-
0 20
sponds to a number percentage of only -0.5. In our ex-
0 30
periments, the total number of drops in a sample was 250
040 >
0 50 2
to 500. Therefore, in a sample representative of the true
0 distribution, the largest one to two drops should have
060 0
k
diameters close to 4 5 .
070 $ One of the most interesting results emerging from the
Y

080 Lu data of Table 2 is that the parameters a, 6, and n, perhaps

085
5
_I
with the exception of run 1, are fairly constant and show
0 no influence of the flow rate on the shape of the distri-
090 bution. Obviously, this result is of significance at least
1
k- from the viewpoint of correlating the data.
095 2 The above data were obtained with the special droplet
I
3
encapsulation technique. Figure 6 shows drop size distri-
0
butions obtained from photographs of flowing water-in-
0 98 kerosene dispersion, at mean velocities = 1.52 to 2.98
m/s. These distributions are self-consistent and can be
0 99 satisfactorily described by the Rosin-Rammler and the
upper limit log-probability functions as in the previous
case. The Rosin-Rammler parameters for data obtained,
3 995
n under the same conditions, with the two different tech-
I I l l 1 I I I I I Ill1 I I l l l l l niques are listed in Table 3. The drop diameters dY5 are
0.1 0.2 0.5 1 2 5 10 in good agreement, except at the two smallest flow rates.
PARTICLE D I A M E T E R , d. rnrn The slope n is fairly constant and apparently independent
Fig. 6. Drop size distribution of water in kerosene, obtained from line of flow rate in each case, but quite large in the case of
photographs. line photographs ( n N 4).Obviously, the mean diameters

AlChE Journal (Vol. 24, No. 2) March, 1978 Page 175

 10
- 1
0 RUN 11s I 0
0RUN
RUN 11s
15s
- 001
- 0 02

/
- 005
0 05
-0 10

- 020
-030 ,
-040 .
z
-050 0
- 0 6 0 I-v
Q
-070 E
UI
-000 5
-085 6>
- 090 2
-
+
I:
-095 2
3
V

- 098

O '
0.005 O F /" I I I 1 1 1 1 1 I I
1
I
0'99
0.995
01 02 05
d/(d,,, di
1 2

Fig. 8. Drop size distribution of water in transformer oil, represented

by an upper limit log-probability function. 8 =3.00 and 1.86 m/s.
5

Collins (1967), such poor picture quality may have

biased the distributions toward the larger drops.
3. The instantaneous velocity of a drop is expected to
increase with decreasing size. Therefore, for a fixed flash
duration, a relatively higher percentage of small drops may
appear to be out of focus even if they are within the pho-
tographed depth of field.
The above possible sources of error (especially the last
two) were absent in the case of still photographs of sam-
ples withdrawn from the pipeline. Furthermore, we ob-
served a very large number of small drops (below -200
pm) in the discharge tank (13, Figure l ) ,which did not
appear to be represented in the spectra obtained from
line photographs. For these reasons, the encapsulation
technique was judged more reliable and the data ob-
tained from samples much more representative of the true
distributions. Therefore, in subsequent experiments with
the more viscous transformer oil as the continuous phase,
only data from samples were collected systematically.
Figure 7 shows drop size spectra of water in transformer

TABLE
4. DROPSIZE DISTRIBUTION
PARAMETERS FROM DATA WITH VISCOUSCONTINUOUS
PHASE (TRANSFORMEROIL). DATAOBTAINED FROM SAMPLES

Rosin-Rammler

Run number
Velocity
-
U, m/s a
Upper limit log-normal distribution
6 dmax,pm d95, clm
-
d32, ~m
Data
&z, fim dmaxt, pm n
distribution
d95, pm

11s 3.00 1.369 0.818 1030 775 345 347 730 2.42 790
17s 2.60 1.286 0.888 1 097 814 397 401 951 2.73 825
10s 2.24 1.231 0.914 1383 1 028 520 512 1110 2.82 1045
18.5 2.08 1.819 0.792 1 579 1113 426 428 1217 2.13 1130
15s 1.86 1.283 0.839 2 009 1521 710 700 1381 2.54 1540
14s 1.52 1.230 0.858 2 498 1896 916 840 1805 2.63 1935
12s 1.19 1.573 0.828 2 625 1 894 804 844 1987 2.34 1925
ArithMetic average 1.399" 0.848 2.51

Average a = 1.329 if run 18 is excluded.

f Diameter of largest measured drop.

Page 176 March, 1978 AlChE Journal (Vol. 24, No. 2)

oil, at mean velocities U = 3.00 and 1.86 m/s. The portant observation is that if a, 6, and n are nearly con-
same data have been replotted in log-probability coordi- stant over a range of conditions, one needs only a reliable
nates in Figure 8. As in the case of water/kerosene dis- estimate of d,, or dgs in order to predict the entire dis-
persions, both the Rosin-Rammler and the upper limit tribution.
log-probability functions give a very good fit of the data. CORRELATION OF M A X I M U M DROP SIZE DATA
The distribution parameters for all the runs with trans-
former oil are listed in Table 4. In Figures 9 and 10, the data on dg5 are compared with
These distributions display exactly the same character- the Sleicher correlation and the Hinze model predic-
istics as those of the water/kerosene system. The diameters tions. The data obtained with transformer oil are in good
d95 obtained basis the Rosin-Rammler and log-probability
agreement with Equation (5), whereas predictions of d95
equations are almost identical; the largest measured drop- based on the Sleicher correlation are five times smaller at
lets are closer to dg5than to the parameters d,,; the Sauter high velocities. The disagreement of this correlation with
mean z2 directly computed from the data are in good
agreement with those obtained from Equation ( l l ) , and
data on d,, for which it was originally developed, is
even greater. Furthermore, the data of Figure 9 show an
almost linear dependence of dg5 on velocity that is far
the parameters a, 6, and n appear again to be independent
of flow rate. Furthermore, the mean values of these pa- from the r-2.5 dependence suggested by Sleicher.
rameters are nearly equal to those obtained with water- The data of Figure 10, obtained with kerosene as the
in-kerosene dispersions. A direct consequence of this im- continuous phase, at first look appear to agree somewhat
better with the Sleicher than with the Hinze correlation.
10
- I However, after careful examination of the conditions under
0 DATA. dg5 which these data were obtained, we concluded that the
-.--- effect of gravity is responsible for the deviation of mea-

.I
-
----
dgS/D = 4.0
HINZE. EQUATION 15)
SLEICHER, EQUATION ( 6 ) I
sured d95 from Equation (5), at velocities smaller than
2.4 m/s. As shown elsewhere (Karabelas, 1977), if the
carrier liquid viscosity is small, the largest droplets tend
to concentrate at the bottom of a horizontal pipe at low
velocities. This tendency may have affected our data as
follows:
1. Since the sampling tube was located below the pipe
center line, the samples became less representative of the
average distribution with decreasing flow rate. The con-
sistently greater deviation of d95 from the predictions of
Equation (5) with decreasing flow rate, shown in Figure
10, is indicative of this effect.
2. The increasing mean concentration at the pipe bot-
tom, with decreasing flow rate, coupled with a possible
velocity lag of the largest droplets, may have promoted
coalescence and formation of drops somewhat larger than
01 110 12 14 16 18 20 22 24 26 28 30 those encountered in dilute, homogeneous dispersions.
At this point we must conclude that the Hinze/Kol-
MEAN VELOCITY, G, mls
mogorov model is more reliable than the Sleicher correla-
Fig. 9. Effect of velocity on maximum drop diameter. W a t e r dispersed tion because it has a more solid theoretical basis and it is
in transformer oil (uC = 14.2 to 20.5 mm2/s). generally in better agreement than Equation (6) with the
data presented here. Next, we shall briefly review the con-
ditions under which this model [Equation ( 2 )] applies to
pipe flow. In particular, we shall examine whether the
mean rate of energy dissipation; substituted into Equa-
tion ( 2 ) , provides satisfactory estimates of d, or d95.
The mam assumption involved in the development of
Equation ( 2 ) is that the flow is homogeneous and iso-
tropic. In such a case, most of the kinetic energy of turbu-
lent fluctuations is contained in the range of wave numbers
(inertial subrange) where the Kolmogorov distribution
law is valid. As is well known (see, for example, Tennekes
and Lumley, 1972), the same law describes energy distri-
bution in the inertial sublayer of pipe flow, that is, at y+
greater than 30 to 35 where Equation ( 2 ) is expected to
be valid. If the Blasius Equation (4)is employed to esti-
mate ti*, y + = 35 corresponds to a distance from the wall
PIPELINE
---- dg$3 = 4 0 We-'

--- SLEICHER, EQUATION ( 6 ) where R is the pipe radius. Figure 11 shows that, for a
given pipe diameter, the thickness of the buffer layer is
greatly reduced at high Reynolds numbers; that is, the
lower boundary of the inertial sublayer is very close to the
wall. For example, in a 25.4 cm ID pipe at Re = lo6,
Fig. 10. Effect of velocity on maximum drop diameter. W a t e r dispersed the inertial .sublayer (3' N 35) is within 250 p from the
i n kerosene (vC = 2.2 to 2.3 mm2/s). pipe wall.

AlChE Journal (Vol. 24, No. 2) March, 1978 Page 177

 Using Equation ( 5 ) one can show that, in almost all
cases of practical interest, the maximum stable droplet sizes
generated in the turbulent core are larger than the wall
layer thickness corresponding to y' = 35. Moreover, it where We = Dp,C2/~is a dimensionless Weber numbei
is very likely that these relatively large droplets will have based on pipe diameter and mean velocity, AS shown in
a tendency to migrate away from the wall owing to Figures 9 and 10, this equation fits our data somewhat
lateral forces similar to those observed in laminar pipe flow better than Equation ( 5 ) . It is also expected that Equation
(SegrB and Silberberg, 1962). It appears therefore, that (23) provides reliable estimates of 4 5 throughout the Re
breakup of the largest droplets will take place primarily range of practical interest. However, it should be used
inside the inertial sublayer. with caution in the case of a very viscous dispersed phase
The local rate of energy dissipation per unit mass EL in in which breakup may be influenced by inertial as well as
the inertial sublayer is inversely proportional to the dis- viscous forces.
tance from the wall; that is
€1 N -
u.3
(19)
DISCUSSION
kY For measurements of size distribution, the droplet en-
if we assume (for example, Tennekes and Lumley, 1972) capsulation technique appears to be more accurate than
a balance of turbulence energy production by viscous dis- the photographic method also employed in this as well as
sipation. The von Karman constant k is approximately in previous studies. All the data obtained with the former
0.4. It can easily be shown that the distance from the technique are self-consistent and provide useful informa-
wall at which the local rate of dissipation €1 is equal to the tion about the effect of flow rate on the size spectrum.
Of particular interest is the similarity of the measured
mean'; [Equation ( 3 ) ] is given by
spectra, which permits their representation by a single
0.2486 well-known distribution function such as the upper limit
N-
(20) Iog-probability or the Rosin-Rammler equation.
Re1I8
Collins (1967) and Collins and Knudsen (1970) reached
This function has been plotted in Figure 11. different conclusions by interpreting size distribution data
It is evident now that the use of the mean rate f i n obtained with a photographic technique. This led to re-
Equation ( 2 ) would give conservative estimates of d,,,, evaluation of some of Collins' data and comparison with
OT d95 at high flow rates. Indeed, for usual liquids at high our results (see Appendix). We concluded that most of

-
Reynolds numbers, the largest droplets, of relative size
d,,,/D, that can survive at a distance from the wall
y/Rbmay be comparable in size or larger than the thick-
Collins' data, plotted in terms of cumulative volume
fractions, can indeed be fitted with a Rosin-Rammler or

101
ness of the buffer layer, but they are small compared to
y/RI, These droplets can move deeper into the inertial
sublayer, that is, closer to the wall, where the local energy 5

dissipation rate E L is larger than :6 Further breakup will,

therefore,
- result in d,,,,, smaller than that corresponding
2
to E.
Using the local dissipation rate el and the value of C
= 0.725, suggested by Hinze, into Equation ( 2 ) we ob- 106
tain
-= 2.645 [ y ]ReoJ5[+]
~ 0.6 0.4
d95 (21) 5
D DPJJ a
a
Following the previous arguments, it is reasonable to as- a.
w 2
sume that there is a characteristic distance from the wall m
ZZ
y/R], associated with the largeit stable droplets. At high 3
z
Re, this distance is likely to be between y + c: 35 and the lo5
n
distance corresponding to t l = 2 _I
0
At present, it is very difficult to provide a more ac-
curate definition of y/R(,, on physical grounds, although 2nz 5

one could develop speculative theories relating y/R], to

the kinematics of large droplets near the wall and their
2
mean residence time relative to the characteristic time re-
quired for breakup. In order to simplify Equation (21)
and provide a correlation more accurate than (5),we have 1o4
selected the following expression:
5
2.8

AS shown in Figure 11, the slope of this line is the mean

difference of the slopes of the lines y+ = 35 and EL = E
and essentially eliminates the Re dependence from Equa-
tion (21). The constant in Equation (22) has been ob-
-
tained by assuming that the lines y/RI,, el = E and y f
-
2l
103
10-3 2
I 1 I l l l l l l
5 10-2 2
I I I l l l l l l
5
DIMENSIONLESS DISTANCE FROM THE WALL, y/R
loe1
I
02
I
0.4

= 35 have a common point of intersection. Fig. 11. Effect of Reynolds number on wall layer thickness and on
-
Substituting Equation (22) into (21), we obtain of €local = E.
IOCUS

Page 178 March, 1978 AlChE Journal (Vol. 24, NO. 2)

 an upper limit log-probability type of distribution function, p c , pd = dynamic viscosity of continuous and dispersed
as is the case with our data, However, the characteristic liquid phase, respectively, ML-lT-’
slope of these distributions seems to be high ( n si 4.0) v = kinematic viscosity, p/p, L2T-’
and probably unrepresentative of the true size spectra. pc, pd = density of continuous and dispersed liquid phase,
This is attributed to inaccuracies inherent in the pho- respectively, ML-3
tographic technique. Data on &5 obtained from the above u = interfacial tension, MT-2
distributions are in better agreement with the correlation
proposed here [Equation ( 2 3 )] than with Sleicher’s cor-
LITERATURE CITED
relation. Finally, Collins’ data indicate that the test sec-
tion length used in his as well as in our study (-576 and Clay, P. H., “The Mechanism of Emulsion Formation in Tur-
600 pipe diameters, respectively) is sufficient to form the bulent Flow,” Akademie van Wetenschappen (Amsterdam),
maximum stable drop diameter. Proceedings, 43, 852-965 (1940).
At present, the development of a rigorous theory to pre- Collins, S. B., “Drop Size Distribution Produced by Turbulent
Pipe Flow of Immiscible Liquids,” Ph.D. thesis, Ore. State
dict equilibrium size spectra in pipe flow appears to be a University, Corvallis ( 1967).
very difficult task. The mathematical framework suggested -, and J. G. Knudsen, “Drop Size Distributions Produced
by Valentas et al. (1966) is certainly useful in this re- by Turbulent Pipe Flow of Immiscible Liquids,” AIChE J.,
spect. However, a great deal of information regarding 16, No. 6, 1072-1080 (Nov., 1970).
basic fluid mechanical aspects of the problem (mainly, Hinze, J. O., “Fundamentals of the Hydrodynamic Mechanism
particle kinematics in a turbulent shear field) is required of Splitting in Dispersion Processes,” ibid., 1, No. 3, 289-295
before real progress can be made. The data presented in ( 1955).
this paper are generally considered encouraging in that Karabelas, A. J., “Vertical Distribution of Dilute Suspensions
in Turbulent Pipe Flow,” ibid., 23, No. 4, 426-434 ( 1977).
they reveal some basic characteristics of drop size spectra. Kolmogorov, A. N., “On the Breaking of Drops in Turbulent
Additional data, under conditions different than those of Flow,” Doklady Akad. Nauk. U.S.S.R., 66, 825 ( 1949).
our study, are of course required to firm up the conclu- Kubie, J, and G. C. Gardner, “Drop Sizes and Drop Dispersion
sions reached here and to guide future modeling studies. in Straight Horizontal Tubes and in Helical Coils,” Chem.
Eng. Sci., 32, No. 2, 195-202 (1977).
ACKNOWLEDGMENT Levich, V. G., Physicochemical Hydrodynamics, Translated
from the Russian by Scipta Technica, Inc., Prentice Hall,
The author is rateful to Shell Development Company for Inc., Englewood Cliffs, N.J. ( 1962).
permission to p u b h this paper. Mlynek, Y., and W. Resnick, “Drop Sizes in Agitated Liquid-
Liquid Systems,” AIChE J., 18, No. 1,122-127 (Jan., 1972).
NOTATION Mugele, R. A,, and H. D. Evans, “Droplet Size Distribution in
Sprays,” Ind. Eng. Chem., 43, No. 6, 1317-1324 (June 1951).
a = parameter in upper limit log-probability distribu- Paul, H. I., and C. A. Sleicher, Jr., “The Maximum Stable Drop
tion function, Equation ( 13), dimensionless Size in Turbulent Flow: Effect of Pipe Diameter,” Chem.
C’ = proportionality constant Eng. Sci., 20, 57-59 (1965).
Rosin, P., and E. Rammler, “The Laws Governing the Fineness
D = pipe inside diameter, L of Powdered Coal,” J. Inst. Fuel, 7, 29-36 (Oct., 1933).
- = particle diameter, p, L
d SrgrB, G., and A. Silberberg, “Behaviour of Macroscopic Rigid
d32 = Sauter mean diameter; volume to area mean di- Spheres in Poiseuille Flow. Part 2,” J. Fluid Mech., 14,
ameter, L 136-157 (1962).
A,, = maximum stable drop diameter; also parameter Shinnar, R., “On the Behavior of Liquid Dispersions,” J. Fluid
in upper limit log-probability Equation ( 11), L Mech., 10,259 (1961).
Sleicher, C. A., Jr., “Maximum Stable Drop Size in Turbulent
d50, dg5 = particle diameter corresponding to 50 and 95%
Flow,” AIChE J., 8, No. 4, 471-477 (1962).
vol, respectively, of particles smaller than d50 and Sprow, F. B., “Distribution of Drop Sizes Produced in Tur-
d95, L bulent Liquid-Liquid Dispersions,” Chem. Eng. Sci., 22,
a!,‘ = designates characteristic drop diameters based on 435 ( 1967).
cumulative number fractions, L Tennekes, H., and J. L. Lumley, A First Course in Turbulence,
d” = drop diameter corresponding to V = 0.3679, L The MIT Press, Cambridge, Mass. ( 1972).
AP D Valentas, K. J., et al., “Analysis of Breakage in Dispersed Phase
f = friction factor; - -,dimensionless Systems,” Ind. Eng. Chem. Fundamentals, 5, No. 2, 271-279
AL 2pCV (1966).
R = pipe inside radius, L
-
Re = Reynolds number, DUpc/pc,dimensionless APPENDIX: COMPARISON OF DATA BY COLLINS
U = mean flow velocitv.,’LT-’ W I T H THE RESULTS OF THIS STUDY
u = local fluctuating velocity, LT-’
[-AL -140,
The data obtained by Collins (1967) will be briefly re-
u. = friction velocity, aP D % LT-I
. - ..
examined in the light of the results obtained in this stud
The most complete set of data was taken with 1.3% lig t
hydrocarbon ( Shellsolv ) dispersed in water. Although not
K.
V = cumulative volume fraction, dimensionless
clearly stated in Collins’ thesis, all the data are presented in
We = Weber number, p,DU2/cr, dimensionless terms of cumulative number fractions instead of the more
yf = dimensionless distance from pipe wall, u* y / u conventional cumulative volume fractions. For the types of
distributions measured by Collins, it can be easily shown that
Greek Letters only 0.5 and 9% of the droplet population corresponds, re-
AP/AL = pressure drop, ML-2T-2 spectively, to -5 and -50% of the dispersed phase volume,
8 = parameter in upper limit log-probability, Equation that is, d’99.5 N _ d95 and d’gi N _ d50.
(lo), dimensionless Table 5 shows data on the variation of d’gg.5 and d’gi with
e = energy dissipation per unit mass and time, L2T-3 axial distance which were obtained from distributions measured
= local rate of energy dissipation per unit mass, at three different points in the pipe cross sections, that is, y / R
el
N 0.05, 0.10, and 0.40. The data suggest that at 576 pipe
L2T-3
- = mean rate of energy dissipation per unit ma5s; in
diameters downstream of the injection point, the largest stable
droplets were already formed. The scatter in these data, and
pipe flow 7 = 2 f v 3 / D ,L2T-3 especially in d’99.5, does not permit further analysis. It will be

AlChE Journal (Vol. 24, NO. 2) March, 1978 Page 179

 1000
TABLE
5. EFFECTOF AXIAL DISTANCE
FROM INJECTION
POINT 900 0 dg5 BASED ON VOLUME F R A C T I O N
x/D CHARACTERISTIC
ON DROPSIZES,OBTAINED FROM 800 - 0 dg5 BASED ON NUMERICAL F R A C T I O N
COLLINS’DATA( 1967, PP. 56,70, AND 71 ) --- FOUATION 12%
1.3% SHELLSOLV
DISPERSED IN WATERAT 4.88 m/s - SLEICHER E Q U A T I O N I61

y/R = 0.05 y/R = 0.10 y/R = 0.40

$99.5, d‘91, $99.5, $91, $99.5, $91,
x/D I*m I*m /Lm I*m Ilm Pm

27 610 325 640 320 590 320

209 415 280 460 285 430 300
42 1 415 270 390 285 380 300
576 360 280 380 265 400 285
Note: d’w.5 and d’ol are obtained from cumulative number frac-
tions and correspond very closely to do5 and dw, respectively.
The latter are obtained from data on cumulative volume
fractions. loo I I I I I I I
32 36 10 4 4 48 52 56 60 64 68 7 2

MEAN VELOCITY rms

Fig. 13. Comparison of data by Collins (1967) with predicted maxi-

mum drop sizes.

- 1 4 0 005 dicted by the Sleicher correlation.

- 001 The latter argument does not have any basis because the
A U = 6.10 m/s, n = 3.74 - 002 Sleicher correlation is empirical and probably inaccurate as
- 005
shown in this study. With regard to the former argument, it
is possible that very stringent criteria were used in statistically
0 u = 4.88 mis. II = 4.27 - 010 analyzing the data, as will become apparent below.
- 015 In order to compare the data reported by Collins with the
- 020 results of this study, we have replotted several distributions
- 030
using cumulative volume fractions instead of number frequen-
cies. Figure 12 shows the only available data set taken at four
-040 > different velocities, 421 pipe diameters from the in‘ection point.
z‘
Obviously, the Rosin-Hanlmler Equation (8) dts the data
quite satisfactorily in the range 0.02 < V < 0.98. A good fit
-050
0
t-
-060 U is also obtained with the upper limit log-probability Equation
2
U
(11).
-070 Lu The characteristic diameters d95 based on volume fractions
I and d’95 based on number frequencies are compared in Figure
3
-0080 g 13, with predictions obtained from Equation ( 2 3 ) and from
> Sleicher’s correlation. The physical properties corresponding
- 085 w to these data are as follows: t = 20”C, pCLeN 1.00 mN s/mz,
3 u = 40.3 mN/m, fid = 1.1 mN s/m2, Pd = 0.782 g/cm3.
-088 +
- 090 2 Both correlations predict smaller than the measured d95, but
the modified Hinze correlation, Equation ( 2 3 ) , is in better
-092 5 agreement with the data.
V
I t will be noticed that the slope n of the distributions in
- 095 Figure 12 is around 4.0, which is surprisingly close to the
average slope obtained in our experiments with the same tech-
nique, that is, direct line hotographs (see Table 3 ) . This
similarity may not be accilental. It is possible that Collins’
photographic technique had the same drawbacks as ours, sys-
- 098 tematically favoring the larger droplets. This was shown in
our case by comparison with the more accurate data obtained
by means of the droplet encapsulation technique. If similar
errors were involved in Collins’ data, the maximum drop size,
0 99 or d95, obtained from the measured distributions would be
20 50 100 200 500 1000 greater than the true maximum size. This would also explain
PARTICLE DIAMETER, d. p m the discrepancy between data and predictions shown in Fig-
ure 13.
Fig. 12. Data by Collins (1967, p. 77). 1.3% light hydrocarbon dis- The results of the above comparisons can be summarized as
persed in water. follows:
1. Collins’ data indicate that the test section length used
recalled here that our data were obtained at -590 pipe diam- in this study (-590 pipe diameters) is sufficient to form the
eters downstream of the water injection point. maximum stable drop diameter.
Collins concluded that none of the existing size distribution 2. Most of his data plotted in terms of cumulative volume
functions could satisfactorily describe his data. He did, how- fractions can be fitted with a Rosin-Rammler or an upper limit
ever, recognize that the upper limit log-probability function log-probability type of distribution function, as is the case
gave the best fit. It was rejected, apparently for two reasons: with our data.
1. “A chi square test of the data for several systems indi- 3. The characteristic slope of these distributions is high
cated that in all cases the hypothesis that distribution obeyed (n e 4) and probably unrepresentative of the true size spectra.
the upper limit law could be rejected with only a 0.5% chance 4. Data on d95 appear to be in better agreement with the
of error. The primary contribution to the chi square seemed correlation proposed here than with Sleicher’s correlation.
to occur in the regions of largest and smallest drops” (Collins,
1967, p. 5 7 ) .
2. The parameter d, which gave the best fit using the Manuscript received March 21, 1977; recision received October 26,
upper limit function was considerably greater than dm, pre- andaccepted October 27, 1977.

Page 180 March, 1978 AlChE Journal (Vol. 24, No. 2)