Drag coefficient and settling velocity for particles of cylindrical shape

Jorge Gabitto
a,b
, Costas Tsouris
a,

a
Oak Ridge National Laboratory, Oak Ridge, TN 37831-6181, USA
b
Chemical Engineering Department, Prairie View A & M University, P.O. Box 4229, Prairie View, TX 77446-4229, USA
Received 21 July 2006; received in revised form 13 June 2007; accepted 29 July 2007
Available online 6 August 2007
Abstract
Solid particles of cylindrical shape play a significant role in many separations processes. Explicit equations for the drag coefficient and the
terminal velocity of free-falling cylindrical particles have been developed in this work. The developed equations are based on available
experimental data for falling cylindrical particles in all flow regimes. The aspect ratio (i.e., length-over-diameter ratio) has been used to account for
the particle shape. Comparisons with correlations proposed by other researchers using different parameters to account for the geometry are
presented. Good agreement is found for small aspect ratios, and increasing differences appear when the aspect ratio increases. The aspect ratio of
cylindrical particles satisfactorily accounts for the geometrical influence on fluid flow of settling particles.
2007 Elsevier B.V. All rights reserved.
Keywords: Drag coefficient; Drag force; Settling velocity; Terminal velocity; Cylindrical particles
1. Introduction
Many processes for the separation of particles of different
sizes and shapes depend upon variations in the behavior of the
particles when subjected to the action of a moving fluid. A
particle falling in an infinite fluid under the influence of gravity
will accelerate until the gravitational force is exactly balanced
by the resistance force that includes buoyancy and drag. The
constant velocity reached at that stage is called the terminal
velocity. The resistive drag force depends upon an experimen-
tally determined drag coefficient.
Drag coefficients and terminal velocities are important design
parameters in many separation processes. Many equations have
been developed and presented in the literature relating the drag
coefficient (C
D
) to the Reynolds number (Re) for particles of
spherical shape falling at their terminal velocities. These cor-
relations are of varying complexity and contain many arbitrary
constants. Many of these correlations are listed in Clift et al. [1],
Khan and Richardson [2], and Haider [3].
In the case of nonspherical particles, less information is
found in the literature. Heywood [4] developed an approximate
method for calculating the terminal velocity of a nonspherical
particle or for calculating its size from its terminal velocity. The
method was an adaptation of his method for spheres. Heywood
used an empirical factor (k) to account for deviations from the
spherical shape.
Haider and Levenspiel [5] presented a generalized C
D
-vs.-Re
correlation for nonspherical particles. They used the concept of
sphericity (), originally introduced by Wadell [6], to account
for the particle shape. The authors also reported a correlation to
calculate explicitly terminal velocities for particles of different
shapes. However, cylinders and needles are usually non-iso-
metric particles, therefore, other parameters maybe more ap-
propriate to account for particle shape. There are also several
ways to define the characteristic length to be used in the required
dimensionless numbers. Clift et al. [1] summarized the different
alternatives.
Predictions by Haider and Levenspiel [5] showed relatively
poor accuracy for particles with b0.67, therefore, some
authors [710] attempted to improve the accuracy of the Haider
and Levenspiel [5] correlations. Chien [7] and Hartman et al. [8]
used the sphericity as shape factor. A somewhat different ap-
proach was presented by Thompson and Clark [9]. These authors
defined a shape factor (), which is simply the ratio of the drag
coefficient for the non-spherical particles to that of a sphere, both
evaluated at Re=1000. The problem found in this approach rests
on the prediction of the shape factor. Thompson and Clark [9]
Available online at www.sciencedirect.com
Powder Technology 183 (2008) 314322
www.elsevier.com/locate/powtec

Corresponding author. Tel.: +1 865 241 3246; fax: +1 865 241 4829.
E-mail address: tsourisc@ornl.gov (C. Tsouris).
0032-5910/$ - see front matter 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.powtec.2007.07.031
failed in their attempt to link the shape factor with the sphe-
ricity, the second harmonic, and the Corey shape factor. Ganser
[10] used the fact that every particle experiences a Stokes regime
where the drag is linearly related with the velocity and a Newton
regime where the drag is proportional to the square of the
velocity. Ganser [10] introduced two shape factors, K
1
and K
2
,
applicable in the Stokes and Newton regimes, respectively. The
two shape factors were found to be unique functions of the
sphericity [10]. The author also presented explicit correlations
for C
D
and Re number.
Chhabra et al. [11] collected experimental results of 19 inde-
pendent studies comprising several different particle shapes,
including cylinders. The resulting data base consisted of
1900 experimental points covering wide ranges of physical prop-
erties and kinematic conditions. The authors used the collected
data to compare different available correlations published in the
literature.
In the case of particles of cylindrical shape several authors
have presented correlations of experimental data [1218]. Clift
et al. [1] reviewed several of these correlations.
The main goal of this work is to develop a generalized
correlation for calculation of the terminal velocity of cylindrical
particles for a wide range of geometric and flow conditions. The
proposed correlation is compared with several general correla-
tions available in literature. The choice of an appropriate shape
factor for cylindrical particles is discussed.
2. Shape factors
Natural and man-made solid particles occur in almost any
imaginable shape from roughly spherical pollen and fly ash to
cylindrical asbestos fibers and irregular mineral particles.
Axisymmetric particles are among the most commonly found.
The group comprises bodies generated by rotating a closed curve
around an axis. Spheroidal and cylindrical particles of various
kinds are of particular interest because they correspond closely
to the shapes adopted by many drops and bubbles and to the
shapes of some solids. Axisymmetric particles are conveniently
described by the aspect ratio (E), defined as the ratio of the length
projected on the axis of symmetry to the maximum diameter
normal to the axis [1].
Most particles of practical interest are irregular in shape. A
variety of empirical factors have been proposed to describe
nonspherical particles and correlate their flow behavior. Em-
pirical description of the particle shape is provided by iden-
tifying two characteristic parameters fromthe following list [19]:
1. volume, V;
2. surface area, A;
3. projected area, A
p
; and
4. projected perimeter, P
p
.
The projected area and perimeter must be determined normal
to some specified axis. For axisymmetric bodies, the reference
direction is taken parallel or normal to the axis of symmetry. An
equivalent sphere is defined as the sphere with the same value
of one of the above parameters. The particle shape factor is
defined as the ratio of a characteristic parameter from the above
list to the corresponding value for the equivalent sphere [1].
Heywood [20] proposed a widely used empirical parameter
based on the projected profile of a particle. The volumetric
shape factor is defined as
k V=d
3
A
; 1
where d
A
=(4A
p
/ )
0.5
is the projected area diameter, which is
calculated as the diameter of a sphere with equal projected area
as that of the particle, and A
p
is the projected area of the particle.
The projected area of the particle is a difficult parameter to
determine because it depends upon the orientation of the particle.
A number of methods have been suggested to estimate d
A
without knowing A
p
[1].
Wadell [6] proposed that the degree of sphericity be defined
as
/ A
V
=A; 2
where A
V
is the surface of a sphere having the same volume as
the particle, and A is the actual surface area of the particle.
According to this definition, the sphericity of a true sphere is
equal to 1. The more the aspect ratio departs from unity, the
lower is the sphericity. In the case of irregular particles, it is
difficult to determine directly.
The sphericity () of the particles was used by Haider and
Levenspiel [5] to account for the particle shape of isometric
particles, which are particles with similar sizes for all significant
dimensions. The authors used experimental data from non-
spherical shapes such as cubes, octahedrons, tetrahedrons, other
nonspherical shapes, and free-falling thin disks. For isometri-
cally shaped particles, the sphericity is considered the best
single parameter for describing the shape of falling particles [5].
Wadell [21] also introduced the degree of circularity as
w P
A
=P
p
kd
A
=P
p
; 3
where P
A
is the perimeter of a sphere with equivalent projected
area, and P
p
is the projected perimeter of the particle. Unlike the
sphericity, can be determined from microscopic or photo-
graphic observation. The use of is only justified on empirical
grounds, but it has the potential advantage for allowing the
correlation of flow dependence on particle orientation.
In the case of axisymmetric particles with creeping flow
parallel to the axis of symmetry, Bowen and Masliyah [19]
found that the most useful shape parameter was
R A=A
p
: 4
Some authors [22,23] have used the so called Corey shape
factor, , defined as,
b c=ab
1=2
; 5
where aNbNc are the three principal axes of the particle. The
shortcomings of such shape factor have been discussed by Alger
315 J. Gabitto, C. Tsouris / Powder Technology 183 (2008) 314322
and Simons [23] who found it quite inadequate for their own
experimental measurements.
The use of different characteristic lengths in the definition of
dimensionless numbers is a common source of confusion. The
characteristic dimensions used by different authors are the
diameters of the same projected area sphere (d
A
), the same
volume sphere (d
V
), and, in the case of disks or cylinders, the
diameter of the disk/cylinder (d
c
). Drag coefficient values also
change depending upon the area used in the calculation of the
drag force. Typically, several different drag coefficients are
reported. A drag coefficient calculated using the projected area
of a sphere has been reported by Coulson and Richardson [24],
Clift et al. [1], and Isaacs and Thodos [18], among others. Tek
and Wilkes [25] used the area of the sphere of the same volume,
while Isaac and Thodos [18] reported the use of the total surface
area of the particle. All these drag coefficients are interchange-
able because the product of the drag coefficient times the area
used in the calculations (C
D
A) is a constant [18].
3. Drag coefficient for nonspherical particles
There are many equations in the literature relating the drag
coefficient to the Reynolds number for particles of spherical
shape falling at their terminal velocities. In the case of par-
ticles of non-spherical shapes, several authors have proposed
correlations that are supposed to apply to particles of several
different shapes ([5,7,8,10,22], among others). The generalized
C
D
-vs.-Re correlation presented by Haider and Levenspiel [5] is
an example of this universal approach. The drag coefficient is
calculated using the following equation:
C
D

24
Re
1 exp2:3288 6:4581/ 2:4486/
2
Re
0:09640:5565/

73:69Re exp5:0748/
Re 5:378 exp6:2122/
:
6
Eq. (6) fits experimental data for spheres, isometric solids,
and disks with a 5.8% root-mean-square (RMS) deviation.
Haider and Levenspiel [5] also presented a more complex
equation that fits the experimental data with a 3.1% RMS
error.
Ganser [10] used data from Haider and Levenspiel [5] and
Thompson and Clark [9] to present a new generalized cor-
relation. Ganser's analysis is based on the fact that every particle
experiences a Stokes regime where drag is linearly related to the
velocity and a Newton regime where drag is proportional to the
square of the velocity. He introduced two shape factors, K
1
and K
2
, applicable in the Stokes and Newton flow regimes,
respectively. Ganser [10] proposed the following drag correla-
tion:
C
D

24
ReK
1
K
2
1 0:1118ReK
1
K
2

0:65657

0:4305
1
3305:
ReK
1
K
2
; 7
where C
D
and Re are based on the equal volume sphere
diameter, and K
1
and K
2
are unique functions of the sphericity
for solids of spherical shape. In the case of solids of non-
spherical shape, K
1
and K
2
are functions of the sphericity and the
particle orientation.
Swamee and Ojha [22] also developed a correlation based on
the experimental data published by Schultz et al. [26]. The authors
employed the equal-volume sphere diameter and the Corey shape
factor (). The authors reported expressions for the drag co-
efficient and terminal velocity in the range 1bReb10,000 and
0.3bb1. The resulting errors are of the order of 20%.
Similarly Chien [7] re-analyzed the data available in the pe-
troleum engineering and processing literature and proposed the
following expression for drag:
C
D

30
Re
67:289exp5:03/; 8
where C
D
and Re are based on the equal-volume sphere diameter.
Eq. (8) was stated to be valid in the ranges 0.2bb1 and
Reb5000. The author did not provide much detail about the fluids
and particles used in the experiments making almost impossible to
work with the experimental data.
Hartman et al. [8] used the fact that the drag coefficient is a
function of particle shape alone in the Newton regime of settling.
The authors proposed that the effects of Re number and the
particle shape are simple additive, they wrote:
Log ReY; / log ReY; 1 PY; /; 9
where Y=C
D
/Re is independent of particle diameter. Note that
the first term on the right hand side of Eq. (9) relates to the
settling of a sphere while the effect of particle shape is contained
in the second term. Hartman et al. [8] reported an equation for
settling of spherical particles and another to calculate the second
termin Eq. (9). The authors stated the accuracy of their approach
to be better than 20%.
Chhabra et al. [11] collected a database of 1900 experimental
data embracing wide ranging particle shapes including needles,
cones, prisms, discs, rectangular parallelepipeds, and cubes. The
resulting data base encompassed wide ranges of physical and
kinematic conditions as: 0.09bb1 and the Re number ranging
from 0.01 to 510
5
. The authors used the data base to critically
evaluate the most widely used correlations available in the
literature. Specifically, Chhabra et al. [11] selected five methods
that predict drag coefficient and terminal velocity of settling
non-spherical particles for critical evaluation (5,7,8,10,22).
They concluded that the best method appeared to be that of
Ganser [10] which uses the equal volume sphere as characteristic
length and the sphericity as shape factor. The resulting mean
error of Ganser's method was found to be 16%, though the
maximum error against specific data sets was as large as 100%.
There are fewer studies of cylindrically shaped particles
falling with their main axis perpendicular to the flow direction.
Pettyjohn and Christiansen [12] studied settling of isometric
particles in the three flow regimes (Stokes, transition, and
Newton). The authors studied particles of cylindrical shape,
among others, with aspect ratios (E) ranging from 0.25 to 2.
However, cylinders of aspect ratio equal to 2 should not be
considered isometric.
316 J. Gabitto, C. Tsouris / Powder Technology 183 (2008) 314322
Heiss and Coull [13] studied the effect of orientation and
particle shape in the viscous (Stokes) region for particles of
cylindrical, rectangular parallelepiped, and spheroidal shapes.
They proposed a correlation of their experimental data. McKay
et al. [14] determined the coefficient of resistance for cylinders
of aspect ratios (L/d
c
) ranging from 0.25 to 5. They proposed a
relationship between the Reynolds and Galileo numbers in
the range 1000bReb16,000. Pruppacher et al. [15] presented a
chart showing a curve fitted to the many determinations of C
D
for steady cross flow past long cylinders in the Re range ap-
plicable to free motion. Clift et al. [1] approximated this curve
using the following expressions:
C
D
9:689 Re
0:78
1 0:147 Re
0:82
0:1 b Re V 5; 10
C
D
9:689 Re
0:78
1 0:227 Re
0:55
5 b Re V 40; 11
C
D
9:689 Re
0:78
1 0:0838 Re
0:82
40 b Re V 400: 12
The boundaries among these expressions correspond to
changes in flow pattern. Reynolds numbers were calculated
using the diameter of the cylinder (d
c
) as characteristic di-
mension, and C
D
is based on the area projected normal to the
axis.
Jayaweera and Cottis [16] presented similar curves to cal-
culate the drag coefficient using the Reynolds number for cyl-
inders of finite length based on experimental data reported by
Jayaweera and Mason [17]. The drag coefficient depends upon
the cylinder aspect ratio and the Reynolds number. There is some
discrepancy between the Pruppacher et al. [15] and Jayaweera
and Cottis [16] curves for a long cylinder. Clift et al. [1] rec-
ommended using the Pruppacher et al. [15] curve because it is
based on a more extensive data compilation.
Isaacs and Thodos [18] studied free settling of cylindrical
particles in the turbulent regime, ReN200. The authors used the
equal volume sphere as the characteristic dimension and found
that the drag coefficient is independent of the Re number. The
drag coefficient depends upon the particle/fluid density ratio
(
p
/
f
) and the aspect ratio (E), defined as L/ d
c
. In the case of
EN1, they reported the following correlation of their experi-
mental data:
C
D
0:99q
p
=q
f

0:12
E
0:08
200 b Re V 60000: 13
4. Terminal velocities for nonspherical particles
The terminal velocity of a particle in free settling can be
calculated from the force balance on the particle through the
determination of a drag coefficient as follows:
U
t

2m
p
gq
p
q
f

q
f
q
p
A
p
C
D
;

14
where U
t
is the terminal velocity, C
D
is the particle drag co-
efficient based on the projected area, m
p
is the particle mass,
p
is the particle density,
f
is the density of the surrounding fluid,
A
p
is the projected area of the particle in the direction of motion,
and g is the gravitational acceleration constant.
Determination of the terminal velocity of a given particle
from any of the available C
D
-vs.-Re expressions requires a
tedious trial and error procedure because the terminal velocity is
present in both variables. Multiplication of the drag coefficient
by Re
2
allows one to derive an explicit equation to calculate the
Reynolds number and, subsequently, the terminal velocity from
the Galileo (Ga) number [24].
Heywood [4] developed an approximate method for calculat-
ing the terminal velocity of nonspherical particles. The method is
an adaptation of the method for spherical particles, but the mean
projected diameter of the particle, d
A
, is defined as the diameter of
a circle having the same area as the particle when viewed from
above and lying in its most stable position. If d
A
is the mean
projected diameter, the mean projected volume is k d
A
3
, where k is
a constant whose value depends on the shape of the particle. For a
spherical particle, k is equal to / 6. The value of k for particles of
defined dimensions can be calculated from the definition of the
drag force for spherical and nonspherical particles given by
Coulson and Richardson [24]. In the case of cylindrical particles,
k can be calculated from the cylinder dimensions as
k k=4
2:5
d
c
=L
0:5
: 15
Heywood's [4] method calculates Reynolds numbers for
nonspherical particles using a Ga number given as
Ga 4 k d
3
A
q
f
g q
p
q
f
=kl
2
; 16
where is the fluid viscosity.
After the Ga number has been calculated, a Reynolds number
is calculated by using the chart provided by Coulson and
Richardson [24] or the equation proposed by Khan and Rich-
ardson [2] as
Re 2:347 Ga
0:018
1:52 Ga
0:016

13:3
: 17
Eq. (17) is valid for 10
1
b Ga b 10
7
and 10
2
b Re b10
4
.
The Reynolds number for a nonspherical particle is computed
by multiplying the value obtained from Eq. (17) by a correction
factor. This correction factor can be read fromtables provided by
Coulson and Richardson [24] or from the following equation:
log10 Factor A=1 explog10Ga B=C D:
18
The constants (A, B, C and D) that appear in Eq. (18) are
functions of Heywood's shape factor k and can be estimated
from the following correlations obtained from Heywood's data:
A 13:676 k
3
14:096 k
2
5:152 k 0:7884; 19:a
B 5:8564 k 1:9651; 19:b
C 0:806; 19:c
D 9:3884 k
3
9:1478 k
2
3:1295 k 0:3654: 19:d
317 J. Gabitto, C. Tsouris / Powder Technology 183 (2008) 314322
Eq. (18) was determined by correlating the experimental
values provided by Coulson and Richardson [24] and is valid
for 0.01bGab10
7
and 0.1bkb0.4. Fig. 1 shows the fitting of
Eq. (18) to Coulson and Richardson's data.
In conclusion, the procedure involves four steps. First,
the Galileo number is calculated from Eq. (16). Second, the
modified Reynolds number is obtained from Eq. (17). Third, a
correction factor is calculated from Eq. (18), and finally, the
Reynolds number for the nonspherical particle is calculated by
multiplying the modified Reynolds number times the correction
factor. The corresponding terminal velocity is calculated from
the Reynolds number definition. This method is only approx-
imate because it is assumed that the factor k completely defines
the shape of the particle, whereas there are many different
shapes of particles for which the k value is the same. This is not
a surprising result if one attempts to define the particle shape
using a single parameter.
Haider and Levenspiel [5] followed a similar approach
defining two dimensionless numbers that can be expressed as
functions of the drag coefficient and the Reynolds and Galileo
dimensionless numbers. They proposed an explicit correlation to
calculate terminal velocity from experimental data for isometric
particles. The authors used experimental data from cubes,
octahedrons, tetrahedrons, nonspherical shapes, and free-falling
thin disks. Haider and Levenspiel [5] used the sphericity factor
defined by Eq. (2) to account for the different shapes of the
particles. In this work, it is more convenient to express the
Haider and Levenspiel [5] correlation as a function of a modified
Galileo number (Ga) and the cylinder aspect ratio as
Re
t

21:5486
E
1=3
GaV

2:7322 f2:6733E
2=3
=0:5 Eg
E
1=6
GaV
1=2
_ _
1
;
20
where Re
t
is the Re number based on the terminal velocity, and
Ga is a modified Galileo number defined as
GaV C
D
Re
2
kd
3
c
q
f
gq
p
q
f
=2l
2
; 21
where C
D
is based on the projected area normal to the axial
direction. Eq. (20) can be used to calculate terminal velocities of
isometric particles with N0.67 and non-isometric disks with
b0.26. Haider and Levenspiel [5] presented data in the range
of 1=d

2000, where d

is a dimensionless particle diameter

given by
d

d
V
gq
f
q
s
q
f

l
2
_ _
1=3

3EGaV
k
_ _
1=3
; 22
and d
V
is the diameter of the sphere that has the same volume as
the particle. The agreement between predictions from Eq. (20)
and experimental data is quite good for isometric particles, but it
is poorer for non-isometric disks.
Clift et al. [1] modified the procedure presented by Jayaweera
and Cottis [16] to allowexplicit calculation of terminal velocities
in the case of EN1 and Reb400. Reynolds numbers within this
range correspond to the Stokes and transitional flow regimes.
The Reynolds number in these flow regimes (Re
L
) can be
calculated using the following set of equations:
Log
10
Re
L
a
0
a
1
w a
2
w
2
a
3
w
3
; 23
where
w Log
10
GaV
1=3
; 24
a
0
0:81824 0:55689=E; 25
a
1
2:41227 1:54674=E 0:53872=E
2
; 26
a
2
0:2056 1:34714=E 0:65696=E
2
; 27
a
3
0:82343 0:40625a
0
0:5625a
1
0:75a
2
: 28
The procedure outlined in Eqs. (23)(28) allows for the
explicit calculation of the Reynolds numbers in the Stokes and
transitional flow regimes. The transitional flow regime extends
approximately up to Ga values equal to 200,000.
Isaacs and Thodos [18] calculated Reynolds number values
in the Newton flow regime (Re
T
) as a function of the ratio of
particle-to-water densities (
p
/
f
) and of the cylinder aspect
ratio. The authors reported that
Re
T
q
p
=q
w

0:06
E
0:04
GaV
0:5
GaVN 100; 000: 29
Eqs. (23) and (29) overlap in the range 110
5
bGab1.610
5
.
Eq. (23) provides terminal velocity values that are 20% higher
than those calculated from Eq. (25). Experimental data from
Adams [27] showed that the results from Eq. (29) are more
accurate than those from Eq. (23). However, Eq. (23) provides
good predictions for the Stokes and most of the transitional flow
regimes. Clift et al. [1] recommended using Eq. (23) for cal-
culating lowand intermediate Reynolds numbers and Eq. (29) for
calculating high Reynolds numbers.
5. Equation development
The use of Eqs. (23) and (29) to calculate Reynolds numbers
creates a discontinuity in the results calculated in the overlapping
Fig. 1. Comparison of correction factors calculated from Eq. (18) with the ones
reported by Coulson and Richardson [24].
318 J. Gabitto, C. Tsouris / Powder Technology 183 (2008) 314322
Ga number range. To avoid the use of a discontinuous function
in this region, we propose the following equation to calculate
Reynolds numbers for EN1 and all flow regimes:
Re Re
L
T
1
Re
T
T
2
; 30
where
T
1
1 for GaVb 2 10
3
; 31
T
1
Ga

3
; for 2 10
3
b GaVb 2 10
5
; 32
and
T
1
0 for GaVN 2 10
5
; 33
where
Ga

2 10
5
G a=2 10
5
2 10
3
; 34
and
T
2
1 T
1
: 35
Eq. (30) describes a smooth function in the range 0.01bGa b
10
8
. The procedure involves calculating the Reynolds numbers
using Eqs. (23) and (29) and the values of aspect ratio (E), the
density ratio (
p
/
f
), and the Ga number. Terminal velocity
values are calculated from the corresponding Reynolds values.
Fig. 2 shows a comparison of Eq. (30) with Eqs. (23) and
(29). In a loglog plot, Eq. (29) gives a straight line of slope 0.5.
Eq. (23) is represented by a curve of slopes higher than 0.5 for
Reb400 and slopes lower than 0.5 for higher Reynolds
numbers. Eq. (30) is depicted by a curve of higher slope than
0.5 for Reb400 and constant slope=0.5 for Re400. Fig. 3
depicts the variation of C
D
values according to Eqs. (23), (29),
and (30). A smooth transition is shown for C
D
curves in the
range of 40bReb400.
The exponent in Eq. (32) was determined using three cri-
teria. First, the calculated Reynolds number should be closer to
those predicted by Eq. (23) in the lower end of the interval.
Second, the calculated Reynolds number should be closer to
those predicted by Eq. (29) in the upper end of the interval.
Finally, there should be a smooth transition from one curve
to the other. The C
D
values were calculated from the corre-
sponding Reynolds number values using Eq. (21). Fig. 3 de-
picts the transition between the C
D
values calculated using
Eqs. (13), (21), (23), and (30).
The values of Ga dividing the flow regimes were selected to
approximately correspond to the Reynolds numbers used in Eq.
(8). Eq. (8) is the third equation used by Clift et al. [1] to derive
Eq. (23). This selection assumes that Eq. (23) works accurately
for Reb40 but increasingly deviates from the experimental
values up to Re=400. Eq. (30) is a function of Ga; therefore, it
is more convenient to use Ga values to divide the correlation
intervals than Reynolds numbers. After the Reynolds number
has been determined, the corresponding drag coefficient values
can be calculated from the values of the Reynolds and Ga
numbers (Eq. (21)).
6. Results and discussion
Results were computed using Eq. (30) for aspect ratios
varying from 1.5 to 100. Fig. 4 shows that Reynolds number
values increase as the aspect ratio increases in all the calculated
range. The increase is small in the Newton regime, Ga N210
5
,
and higher in the transitional and Stokes regimes, Ga b210
5
.
For E 10 there is only a small increase in the predicted Reyn-
olds values. This result occurs because the fluid flow differences
among cylinders of different aspect ratios are produced by the
degree of three-dimensional flow around the bases of the cy-
lindrical particles. The three-dimensional flow effects decrease
Fig. 2. Comparison of Reynolds numbers calculated using Eq. (30) with Clift
et al. [1] procedure and Isaac and Thodos [18] equation.
Fig. 3. Comparison of C
D
values calculated using Eq. (30) with Clift et al. [1]
procedure and Isaac and Thodos' correlation [18].
Fig. 4. Reynolds number vs. Galileo number calculated using Eq. (30) for
different aspect ratios (E).
319 J. Gabitto, C. Tsouris / Powder Technology 183 (2008) 314322
as the aspect ratio increases. The flow in an infinitely long
cylinder is practically two-dimensional. In a cylinder with aspect
ratio 10, the three-dimensional fluid-dynamic effects are already
small. The solid resembles an infinitely long cylinder. These
effects become even smaller as the aspect ratio increases further.
These fluid-dynamic effects are also more important at relatively
low than at high Reynolds numbers.
Eq. (30) is representative of the experimental data measured
by Jayaweera and Mason [17] and Isaac and Thodos [18],
therefore, the accuracy of this equation is approximately the one
reported by the original authors. Fig. 5 shows a comparison of
Eq. (30) with experimental data reported by Pettyjohn and
Christiansen [12], Heiss and Coull [13], and McKay et al. [14]
for aspect ratio equal to 2. The values predicted by Eq. (30) agree
well with the experimental results. It is significant that the
experimental data reported in literature [1214] cover all three
settling regimes.
Comparisons of Eq. (30) with several universal correlations
reported in the literature were also carried out. The procedure
followed was to work either with raw experimental data and
develop correlations similar to Eq. (30) or express an already
existing equation as a function of Ga and aspect ratio (E); for
example, Eq. (20) was derived from the original correlation
reported by Haider and Levenspiel [5].
The results calculated using the model presented in this work
(Eq. (30)) were compared with the Haider and Levenspiel [5]
correlation, Eq. (20); Heywood's procedure [4]; Chien's
equation [7]; and Ganser's equation [10]. The RMS deviation
between results for different aspect ratios calculated using the
model and the corresponding data of the aforementioned authors
are shown in Table 1. The RMS deviation was calculated using
the following formula:
RMS

n
i1
log10Re; model log10Re; eq: j
2
n
_

_
_

_
1=2
: 36
Table 1 shows the values of the different parameters used by
the different authors to account for the influence of geometrical
shape. Results presented in Table 1 indicating that RMS de-
viations remain relatively small for Eb10 but increase con-
tinuously for bigger E values. There is good agreement between
the predictions of Eq. (30) and those fromHaider and Levenspiel
[5], Heywood [4], and Ganser [10] for Eb10. Chien's equation
[7] does not predict accurately the experimental results for
cylinders even for lowaspect ratio. The discrepancy between the
predictions of Eq. (30) and those of all the authors increases as
the aspect ratio increases. The correlation by Haider and
Levenspiel [5] shows the best agreement with the predictions
of Eq. (30) (RMS3% for Eb10), while Chien's results show
the highest discrepancy. The results of Haider and Levenspiel
(Eq. (20)) agree remarkably well with the correlation developed
in this work for 0.67 (E 6), but show increased dis-
agreement for lower sphericities (higher aspect ratios).
Figs. 6 and 7 illustrate this behavior. Fig. 6 shows the
calculated results for the five correlations in the range
10
2
Ga 10
8
for E=2. In general, there is good agreement
Table 1
Root-mean-square percent deviations between this work and different models
E k RMS%
[This work Haider
and Levenspiel [5]]
RMS%
[This work
Heywood [4]]
RMS%
[This work
Chien [7]]
RMS%
[This work
Ganser [10]]
1.5 0.86 0.447 3.51 6.03 10.76 6.61
2 0.83 0.387 3.13 7.38 14.78 7.07
5 0.70 0.245 3.98 8.86 17.69 8.69
10 0.58 0.173 8.48 9.69 25.72 12.51
20 0.47 0.122 14.24 14.99 51.83 19.62
50 0.37 0.078 22.49 23.89 44.89
100 0.28 0.055 29.57 31.18 98.15
Fig. 6. Comparison of predicted Reynolds number values using different models
for E=2.
Fig. 5. Comparison of Eq. (30) with experimental data reported in literature for
aspect ratio=2.
320 J. Gabitto, C. Tsouris / Powder Technology 183 (2008) 314322
among the five sets of results. The best agreement is found
between Eqs. (30) and (20). Chien's [7] equation predicts
smaller values than the other correlations in all the Ga range.
The biggest discrepancy between Eq. (30) and the other cor-
relations is found at low Ga numbers. Eq. (30) predicts similar
results to Chien's equation, but smaller than all the others.
Similar results were found for all the E 6 results calculated in
this work.
Fig. 7 shows the calculated results for the five correlations in
the range 10
2
Ga 10
8
for E=10. Chien's equation [7]
predicts smaller Re values than the other correlations in all the
Ga range while the Haider and Levenspiel [5] correlation pre-
dicts the highest Reynolds values in all the Ga range, especially
at low Ga numbers. Ganser's equation predicts the highest Re
values for the transitional regime and the smallest ones for the
Newton regime.
The main difference between the model presented in this
work, the Haider and Levenspiel [5] correlation, Heywood's
procedure [4], Ganser's equation [10], and Chien's equation [7]
is how to account for the influence of the geometry on the
Reynolds number values. This work uses the aspect ratio as the
relevant geometric parameter, while Haider and Levenspiel [5],
Chien [7], and Ganser [10] use explicitly or implicitly the sphe-
ricity while Heywood [4] uses the empirical parameter k. Haider
and Levenspiel [5] and Chien [7] use directly the sphericity in
their correlations while Ganser [10] uses the sphericity to
calculate the K
1
and K
2
correction factors.
The differences among the different ways to compute Reyn-
olds numbers are related to the influence of these three pa-
rameters. The universal correlations agree relatively well with
the specific correlation presented in this work for cylinders with
small aspect ratio, but are not accurate for cylinders with large
aspect ratios.
Figs. 8, 9, and 10 depict the variation of computed Reynolds
numbers with the aspect ratio at three different Ga numbers.
Fig. 8 shows data in the Stokes regime, Ga =0.1. Eq. (30)
predicts increases in Re numbers for the whole aspect ratio
range. The values predicted by Eq. (30) are very similar to the
ones predicted by Ganser's equation [10]. Chien's equation [7]
predicts the smallest Reynolds numbers. Results from Haider
and Levenspiel [5] and Heywood's procedure [4] are the highest
for all E values.
Fig. 9 shows data calculated for Ga =100. The behavior in this
figure is similar to that in the Stokes flow regime. The values
predicted by Eq. (30) are similar to the ones predicted by Ganser's
equation [10]. Chien's equation [7] predicts the smallest Reynolds
numbers. Heywood's procedure [4] and Haider and Levenspiel [5]
results are the highest for all E values.
Fig. 10 shows a similar comparison for values in the turbulent
region, Ga =210
5
. Results from Heywood's procedure [4]
and Eq. (30) are the highest for all E values. All the other authors
predict lower Re values. Results from Haider and Levenspiel's
Fig. 8. Variation of Reynolds number with aspect ratio calculated using different
models in the Stokes flow regime (Ga =0.1).
Fig. 9. Variation of Reynolds number with aspect ratio calculated using different
models in the transitional flow regime (Ga =100).
Fig. 10. Variation of Reynolds number with aspect ratio calculated using
different models in the Newton flow regime (Ga =210
5
).
Fig. 7. Reynolds number values predicted using different models for E=10.
321 J. Gabitto, C. Tsouris / Powder Technology 183 (2008) 314322
work decrease for low Ga, show a minimum value at about
E=20, and for higher E values increase again. Re numbers
calculated using the other correlations increase monotonically in
the whole aspect ratio range.
7. Conclusions
A correlation has been presented to explicitly calculate
terminal velocities of cylindrical particles settling in a liquid.
The correlation predicts terminal velocities for free settling of
cylindrical particles in all flow-field regimes using a smooth,
continuous function. The aspect ratio of the cylinder accounts for
the shape of the particles. This curve fits well experimental data
for cylinders of different aspect ratios in all flow regimes.
Terminal velocities are more dependent on the aspect ratio in the
Stokes regime than in the Newton regime. The predicted solu-
tion approaches the infinitely long cylinder solution at aspect
ratios above 10.
The correlation presented here has been compared to the
general correlation presented by Haider and Levenspiel [5] for
isometric particles and to Heywood's procedure [4], Chien's
equation [7], and Ganser's equation [10] for non-spherical par-
ticles. To implement a comparison with Heywood's procedure, a
correlation was developed for the calculation of Reynolds cor-
rection factors. Good agreement was found with the Haider and
Levenspiel correlation [5], Ganser's equation [10] and Hey-
wood's procedure [4] for aspect ratios equal or lower than 5.
Increasing deviations were found for aspect ratios bigger than 5.
The geometrical parameter used in accounting for the par-
ticle shape is the main difference among the several ways of
calculating terminal velocities. The aspect ratio seems to be
the best way to estimate the geometrical effects for cylinders.
Sphericity and the empirical parameter k give acceptable re-
sults at small aspect ratios, but they increasingly deviate from
the experimental data as the aspect ratio increases. Drag co-
efficients can be calculated directly from the computed ter-
minal velocities.
Acknowledgments
Gratefully acknowledged is the support by the Ocean
Carbon Sequestration Program, Office of Biological and En-
vironmental Research, U.S. Department of Energy under Con-
tract No. DE-AC05-00OR22725 with UT-Battelle, LLC. Also,
support from the U.S. Nuclear Regulatory Commission for
Jorge Gabitto under the Historically Black Colleges and Uni-
versities Faculty Research Summer program is greatly ap-
preciated. The authors are thankful to Ms. P. P. Henson for
editing the manuscript.
References
[1] R. Clift, J.R. Grace, M.E. Weber, Bubbles, Drops and Particles, Chapter 6,
Academic Press, New York, 1978.
[2] A.R. Kahn, J.F. Richardson, The resistance to motion of a solid sphere in a
fluid, Chem. Eng. Commun. (1987) 62135.
[3] A.M. Haider, M.S. Project, Oregon State University, 1987.
[4] H. Heywood, Calculation of particle terminal velocities, J. Imp. Coll.
Chem. Eng. Soc. (1948) 140257.
[5] A. Haider, O. Levenspiel, Drag coefficients and terminal velocity of
spherical and nonspherical particles, Powder Technol. 58 (1989) 6370.
[6] H. Wadell, The coefficient of resistance as a function of Reynolds number
for solids of various shapes, J. Franklin Inst. 217 (1934) 459490.
[7] S.F. Chien, Settling velocity of irregularly shaped particles, SPE Drill.
Complet. 9 (1994) 281.
[8] M. Hartman, O. Trnka, K. Svoboda, Free settling of nonspherical particles,
Ind. Eng. Chem. Res. 33 (1994) 1979.
[9] T.L. Thompson, N.N. Clark, Aholistic approach to particle drag prediction,
Powder Technol. 6 (1991) 57.
[10] G.H. Ganser, A rational approach to drag prediction of spherical and
nonspherical particles, Powder Technol. 77 (1993) 143.
[11] R.P. Chhabra, L. Agarwal, N.K. Sinha, Drag on non-spherical particles: an
evaluation of available methods, Powder Technol. 101 (1999) 288.
[12] E.S. Pettyjohn, E.B. Christiansen, Effect of particle shape on free-settling
rates of isometric particles, Chem. Eng. Prog. 44 (1948) 157.
[13] J.F. Heiss, J. Coull, On the settling velocity of non-isometric particles in a
viscous medium, Chem. Eng. Prog. 48 (1952) 133.
[14] G. McKay, W.R. Murphy, W.R. Hillis, Settling characteristics of discs and
cylinders, Chem. Eng. Res. Des. 66 (1988) 107.
[15] H.R. Pruppacher, B.P. Le Clair, A.E.J. Hamielec, Some relations between
drag and flow pattern of viscous flow past a sphere and a cylinder at low
and intermediate Reynolds numbers, J. Fluid Mech. 44 (1970) 781790.
[16] K.O.L. Jayaweera, R.E. Cottis, Fall velocities of plate-like and columnar
ice crystals, J. R. Meteorol. Soc. 95 (1969) 703709.
[17] K.O.L. Jayaweera, B.J. Mason, The behavior of freely falling cylinders and
cones in a viscous fluid, J. Fluid Mech. 22 (1965) 709720.
[18] J.L. Isaacs, G. Thodos, The free-settling of solid cylindrical particles in the
turbulent regime, Can. J. Chem. Eng. 45 (6) (1967) 150155.
[19] K.O.L. Bowen, J.H. Masliyah, Drag force on isolated axisymmetric
particles in stokes flow, Can. J. Chem. Eng. 45 (6) (1973) 150155.
[20] H. Heywood, Symp. interaction fluids and particles, Inst. Chem. Eng.,
London, 1962, pp. 18.
[21] H. Wadell, Sphericity and roundness of rock particles, J. Geol. 41 (1933)
310331.
[22] P.K. Swamee, C.P. Ojha, Drag coefficients and fall velocity of non-
spherical particles, J. Hydraul. Eng. 117 (1991) 660.
[23] G.R. Alger, D.B. Simmons, Fall velocity of irregular shaped particles,
J. Hydraul. Eng. Div., ASCE 94 (1968) 721.
[24] J.M. Coulson, J.F. Richardson, Chemical Engineering, vol. 2, Pergamon
Press, Oxford, 1977, Chapter 4.
[25] M.R. Tek, J.O. Wilkes, Fluid Flow and Heat Transfer, University of
Michigan, Ann Arbor, 1974.
[26] E.F. Schultz, R.H. Wilde, M.L. Albertson, Influence of Shape on Fall
Velocity of Sedimentary Particles, Report for the Missouri River Div.,
Corps of Engineers, U.S. Army, through Colorado Research Foundation,
Fort Collins CO, 1954.
[27] E.E., Adams, Massachusetts Institute of Technology, private communica-
tion to C. Tsouris, Oak Ridge National Laboratory, February 2006.
322 J. Gabitto, C. Tsouris / Powder Technology 183 (2008) 314322