Powder Technology 277 (2015) 222–230

Contents lists available at ScienceDirect

Powder Technology

journal homepage: www.elsevier.com/locate/powtec

A new shape dependent drag correlation formula for non-spherical rough

particles. Experiments and results
Fabio Dioguardi a,b,⁎, Daniela Mele a
a
Dipartimento di Scienze della Terra e Geoambientali, University of Bari, Bari, Italy
b
Dipartimento di Meccanica, Matematica e Management, Politecnico di Bari, Bari, Italy

a r t i c l e i n f o a b s t r a c t

Article history: The drag of non-spherical rough particles has been investigated in a wide range of Reynolds numbers (0.03–
Received 16 December 2014 10,000). The study is based on experimental measurements of the terminal velocities of irregular particles falling
Received in revised form 2 February 2015 in fluids of different densities and viscosities. The particle shape is described by a shape factor that takes into
Accepted 28 February 2015
account both sphericity and circularity, which are measured via image particle analysis techniques. This shape
Available online 16 March 2015
factor is particularly suitable for non-spherical and highly irregular particles. The drag coefficient has been corre-
Keywords:
lated to the particle Reynolds number and the shape factor and a new correlation law has been found; the corre-
Drag coefficient lation has the functional form of a power law. Due to the mutual dependency of the particle terminal velocity on
Shape factor the drag coefficient, which in turn depends on the particle shape and Reynolds number, an iterative procedure
Terminal velocity needs to be designed for calculating the terminal velocity of particles of a specific size and shape. Such a proce-
Non-spherical particles dure is adopted herein and a spreadsheet and a Fortran 90 code allowing the iterative calculation are provided
Multiphase flow in the Supplementary Material. The fitting of experimental measurements with our model calculations show
Settling that our new law predicts the drag coefficients and the terminal velocity of irregularly shaped particles, as
volcanic ash, more accurately than other shape-dependent drag laws.
© 2015 Elsevier B.V. All rights reserved.

1. Introduction where μ is the fluid viscosity. From numerous experiments on spherical

The knowledge of how irregularly shaped particles settle in quies-
cent or moving fluids is fundamental in a wide range of research fields
and applications, from industrial (e.g. chemical engineering) to natural
processes (sedimentation of solid particles in rivers and during explo-
sive eruptions) [1–6]. The equilibrium velocity wt at which particles set-
tle in a static Newtonian liquid can be calculated by balancing the
surface (drag) and body forces acting on the particle [7], thus obtaining
the Netwon's impact law:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u  ffi
u4gd ρ −ρ
t p p
wt ¼
3C d ρ
where g is the gravitational acceleration, dp is the particle size, ρp is the
particle density, ρ is the fluid density and Cd is the drag coefficient
(refer to Table 1 for Symbol notation), which, for a spherical particle, is
solely a function of the particle Reynolds number:
particles it was possible to define a Cd vs. Re experimental relationship,
which was however not easily interpolated to obtain a single drag law
over the entire Re range (from 0 to 107). For this reason several correla-
tions have been proposed in the literature, each having a different accu-
racy and range of applicability (e.g. Schiller and Naumann [8]; Dallavalle
[9]; Clift et al. [10]; White [11]). For non-spherical particles these correla-
tions are no longer valid and the drag coefficient is a function of both
particle Reynolds number and shape.
In the multiphase fluid dynamics literature there is not a unique way
to define the shape of a particle. Some authors use the ratio between the
volume-equivalent-sphere and the surface-equivalent-sphere diame-
ters (dn/dA), where the surface is that of the projected area [2,12–14].
ð1Þ
Other authors, in order to focus on the effects of particle elongation, de-
fine particles shape in terms of the aspect ratio (particle length along the
symmetry axis over the largest diameter of the cross section) [15,16].
Another important shape factor is the sphericity, which is the ratio
between the surface area of the equivalent sphere Asph and that of the
actual particle Ap:

ρwt dp Asph
Re ¼ ð2Þ Φ¼ : ð3Þ
μ Ap

⁎ Corresponding author at: Dipartimento di Scienze della Terra e Geoambientali, via E.

Orabona 4, 70125, Bari, Italy. Tel.: +39 3403733928; fax: +39 0805442625. Sphericity is probably the most widely used among the shape
E-mail address: fabio.dioguardi@uniba.it (F. Dioguardi). factors, as it is generally considered to be an accurate shape descriptor

http://dx.doi.org/10.1016/j.powtec.2015.02.062
0032-5910/© 2015 Elsevier B.V. All rights reserved.
 F. Dioguardi, D. Mele / Powder Technology 277 (2015) 222–230 223

for isometric non-spherical particle [3,17–21]. However it is very diffi- Table 1

cult to be evaluated for highly irregular particles, since it is not easy to Symbol notation.

measure Ap. Therefore other shape factors have been devised, e.g. the Symbol Description Units
circularity c [2]: Latin
a Exponent applied to Re –
πdA Ap Surface area of the actual particle cm2
c¼ ð4Þ
P Asph Surface area of the equivalent sphere cm2
c Particle circularity (as defined in [2]) –
where P is the projected perimeter of the particle in its direction of Cd Fluid-particle drag coefficient –
motion. Another possibility is to use combinations of the three principal Cd,norm Normalized fluid-particle drag coefficient –
Cd,sphere Drag coefficient of the spherical particle –
axes of an irregular particle (dl, dm, ds: long, medium and short axes,
dA Surface-equivalent-sphere diameter cm
respectively), e.g. the Corey shape factor [22,23]. Finally, for highly irreg- dl Long axis of the particle cm
ular particles, Dellino et al. [3] found that the ratio between sphericity Φ dm Medium axis of the particle cm
and a newly defined circularity X was an effective shape descriptor: dn Volume-equivalent-sphere diameter cm
dp Particle dimension cm
Φ ds Small axis of the particle cm
Ψ¼ : ð5Þ err% Relative error of wt,calc with respect to wt,meas –
X
exp Exponent applied to the particle shape factor –
g Gravitational acceleration cm s−2
The circularity X is more suitable in the case of experiments in which K1 First Ganser's shape factor –
the secondary motions of the falling particles are not taken into account K2 Second Ganser's shape factor
in the analysis. Indeed parameters like c (Eq. (4)) can be measured in m Particle mass g
experiments in which the particle shape is regular, although not spher- P Projected perimeter of the particle in its direction of cm
motion
ical, thus allowing to have a complete and constant control of its posi-
Pmp Maximum projection particle perimeter cm
tion relative to the surrounding fluid during the fall experiment and, Pp Perimeter of the circle equivalent to the maximum cm
therefore, to know the value of the projected area and perimeter. Details projection area
on Ψ will be discussed in the following section. Re Particle Reynolds number –
Re* Initial guess for particle Reynolds number in the iterative –
Basing on these diverse shape factors and on experimental data from
procedure
free-fall experiments (e.g. Tran-Cong et al. [2]; Dellino et al. [3]), wind X Particle circularity –
tunnel testing (e.g. Bagheri et al. [24]) or by compiling data from the wt Terminal velocity cm s−1
literature, previous authors (e.g. Chhabra et al. [1]; Hölzer and wt,calc Terminal velocity calculated by drag laws cm s−1
Sommerfeld [21]) have found different correlations between Cd, Re wt,meas Terminal velocity measured during the experiments cm s−1
wt,Stokes Terminal velocity calculated with the Stoke's law cm s−1
and particle shape. Generally these relationships are valid among spe-
cific Re ranges, and are obtained from experiments on non-spherical Greek
particles but with well-defined shapes (e.g. cylinder, disks, octahedrons, β Corey shape factor –
μ Fluid viscosity P
etc.), which allow a good control on the effective area exposed by the
ρ Fluid density g cm−3
particle to the fluid during settling. However, this case is not represen- ρp Particle density g cm−3
tative of irregular natural cases, such as those transported by wind or Φ Particle sphericity –
water or pumice particles produced during explosive eruptions, in Ψ Particle shape factor –
which the grains are far to be described by well-defined shapes. In
this sense the law of Dellino et al. [3] represented a useful improvement,
as it was derived from free-falling experiments of highly irregular parti-
cles of volcanic origin of different sizes, densities and shapes. The rela- The size of a particle, dp, was taken as equal to the diameter of the
tionship found by Dellino et al. [3] is valid for Re N 50, although it was equivalent sphere, dn:
later verified that its validity could be extended down to 10 [25]. In
this paper new experiments on the same particles falling at different sffiffiffiffiffiffiffiffi
Re numbers were designed in order to cover lower Re values down to 3 6m
dp ¼ dn ¼ ð6Þ
less than 1 and a new drag law covering the enlarged Re range was ob- πρp
tained. This represents, in our opinion, an important step forward for
the characterization of fine particulate solids transportation by fluid
where m is the particle mass. Particle density ρp was measured, depend-
currents.
ing on the particle size, by two standard Gay–Lussac picnometers of 25
and 5 ml capacities.
2. Particle shape factors and settling experiments
In order to calculate particle sphericity Φ, which is needed for the
evaluation of Ψ, Asph and Ap had to be measured (Eq. (3)). The surface
In order to obtain a model to predict the terminal velocity of irregu-
area of the equivalent sphere Asph is:
larly shaped particles, the particle parameters have been measured and
experiments performed on them following the procedure described in
 2
Dellino et al. [3] and Mele et al. [26]. The complete experimental dataset dp
Asph ¼ 4π : ð7Þ
is reported in the Microsoft Office Excel file table.xlsx (in the first sheet 2
“Experiments”) available as Supplementary Material.

As an approximation of Ap, we took the area of a scalene ellipsoid [3]:

2.1. Particle parameters
! 1
Experiments were performed using the same set of particles ðdl =2Þ2 ðdm =2Þ2 þ ðdl =2Þ2 ðds =2Þ2 þ ðdm =2Þ2 ðds =2Þ2 1:6075

employed in Dellino et al. [3]. The particles were sampled from the Ap ¼ 4π ð8Þ
3
volcanic material erupted during explosive eruptions at Vesuvius and
Campi Flegrei volcanoes, Italy. They are characterized by a wide range
of size, density and shape. where dl, dm, ds were defined in the Introduction section.
224 F. Dioguardi, D. Mele / Powder Technology 277 (2015) 222–230
Circularity X is defined as:
P mp
X¼
Pp
where Pmp is the maximum projection perimeter and Pp is the perimeter
of the circle equivalent to the maximum projection area Amp, i.e. the area
containing dl and dm. Circularity X is sensible to the irregularity of the
contour of particles [27], as described in detail in Dellino and La Volpe
[28].
The three particle axes (dl, dm, ds) and circularity (X) were esti-
mated by means of image particle analysis techniques on high-
resolution digital images using the software Optilab Pro 2.6. Each
particle was mounted on a goniometric universal stage under a ste-
reomicroscope equipped with a camera, in a way that the maximum
projection area Amp was exposed to the camera (Fig. 1). Upon ana-
lyzing the picture taken in this position, we measured X, dl and dm
(Fig. 1a). Afterwards the goniometric universal stage was tilted 90°
and a second picture was taken in order to measure ds (Fig. 1b). Since
the circularity is more sensitive to the perimeter (Eq. (9)), which in
turn strongly depends on the more external pixels of the particle
image, we ensured (by magnifying the pictures) that all the particles
of different dimensions were covered by at least 5000 pixels in order
to avoid or reduce the resolution effect when measuring the particle
perimeter [28].
The shape factor ranges between 0 and 1; in the limit of a spherical
particle, Ψ = 1. The lower the shape factor, the more irregular is a par-
ticle shape. In Fig. 2 some of the particles we used in the experiments are
shown, with the corresponding value of Ψ. From the images it can be
observed that also particles that do not look very different from a sphere
(e.g. the fourth particle of group a) have a shape factor that is signifi-
cantly lower than 1. This is due to the smallest scale irregularities of
the surface, which are related to the morphology of broken gas bubbles
contouring the volcanic glass particle. The shape factor proposed by
Dellino et al. [3] is indeed able to capture such shape irregularity,
which renders particle surface rough, and influence drag force upon set-
tling throughout a fluid.
2.2. Experiments
We used the same experimental setup employed by Dellino et al. [3],
i.e. a vertical glass cylinder of height 1.5 m and inner radius of 5 cm
(Fig. 3). In the previous work [3], the authors used distilled water. In
order to use the same particles in our new experiments, the only way
to change the Re range was to vary the fluid characteristics. In particular
we used two different solutions of glycerine and distilled water, one of
glycerine diluted by 13.5% (solution 1) volume of distilled water and
dl
dm
dl = 17.95 mm, dm= 16.60 mm, ds= 11.80 mm, Χ = 1.48, Ψ = 0.36, Φ = 0.53
Fig. 1. Example of the determination of dl, dm and ds of a particle mounted on a goniometric universal stage under a stereomicroscope. a. In this position the particle exposes the maximum
projection area Amp, thus dl and dm can be measured. b. The particle was tilted 90° with respect to the position in a, so that dl and ds are visible and ds measurable.
the other with 40% of distilled water (solution 2). Experiments were
performed at atmospheric pressure and at the controlled room temper-
atures of 19 and 22 °C for the solutions 1 and 2, respectively, as to insure
ð9Þ
that the rheological properties of the solution remained constant. Under
these conditions, the densities and viscosities of the two solutions were
1.235 g cm−3 and 2.088 P for solution 1, and 1.172 g cm−3 and 0.1499 P
for solution 2 [29]. It is to note that in table.xlsx of the Supplementary
Material, where both data of the older Dellino et al. [3] experiments
and those of the new experiments are reported, data are sorted by Re
and not in chronological order. Therefore the reader will find values of
μ and ρ of distilled water (1 g cm−3 and 0.0102 P, respectively), which
is the fluid employed in the older experiments, intercalated in the list
with the values of fluid solutions used in the new experiments. The cyl-
inder was filled with the fluid solution at least one day prior to each ex-
perimental session, as to allow the escape of air bubbles, which could
have modified the solutions physical properties.
Because of the high viscosity of the solutions and the high irregular-
ity of particles surfaces, as to insure a perfect solid–fluid contact,
particles stayed immersed in the solution for two days before each ex-
perimental run. The solutions were heated to about 90 °C to allow the
fluid to fill the open vesicles, thus avoiding the formation of bubbles
when immersing the particle in the solution. After the first experiment
with solution 1, the set of particles was cleaned both with hot distilled
water and ethanol to remove the old solution. The particles were then
re-weighed to verify their perfect cleanliness before their use with
solution 2.
The settling velocity was measured by a high-definition video cam-
era (720 × 1280 pixels) with a recording rate of 50 frames s− 1
(Fig. 3). Frame by frame and multi-frame particle tracking techniques
were employed (Fig. 4). The high-definition video format allowed a
discretization of the scene at a scale smaller than 0.02 cm pixels−1, so
the precision of spatial measurements was about ±0.01 cm. The record-
ing rate of 50 frames s−1 resulted in a maximum typical distance trav-
eled by a particle between two successive frames of about 2.5 cm,
which obviously decreases at decreasing settling velocity. The relative
error on distance measurement between two successive frames is
therefore about ±1%. The error on the time interval between two suc-
cessive frames is linked to the precision of the internal digital clock of
the video camera and is negligible compared to the distance error.
Therefore, the relative error of settling velocity measurements is about
±1%.
3. Data analysis
Upon analyzing the video footages of the experiments, we deter-
mined the terminal velocity of each settling particle wt. From wt and
a b
dl
ds
 F. Dioguardi, D. Mele / Powder Technology 277 (2015) 222–230 225

Fig. 2. Picture of some of the particles used in the experiments. The value of the shape factor Ψ of each particle is reported. It can be observed how, due to the small scale surface roughness
induced by open vesicles, Ψ is significantly lower than 1 (the value of a sphere) even for particles apparently not very different from a sphere (e.g. the fourth one of group in a).

particles physical parameters, it was possible to calculate the drag coef-

ficient by rearranging the Newton's impact law (Eq. (1)):
 
4gdp ρp −ρ
Cd ¼ : ð10Þ
3ρw2t

This is the first step for calculating the quantity CdRe2, which is inde-
pendent from the terminal velocity, as it can be verified by combining
Eqs. (1) and (10):
 
2
4gd3p ρ ρp −ρ
C d Re ¼ : ð11Þ
3μ 2

This procedure is appropriate for spherical particles [3]. In order to

Fig. 3. Photo taken in the laboratory before an experimental session. The glass cylinder
filled with a solution is visible, together with the video camera mounted on a tripod and
connected to a separate monitor.
apply it also to irregularly shaped particles, a shape parameter needs
to be introduced in the equation. In Dellino et al. [3] the quantity
CdRe2 was multiplied by the shape factor and data were plotted in the
Re vs. CdRe2Ψexp diagram, and the best correlation was searched by the
least-square method. In this way the authors included the effect of the
particles irregular shape; the functional form (Ψexp multiplied by
CdRe2) was found to be the most appropriate for obtaining the best fit.
The term exp is the exponent applied to the shape factor: the higher
its value, the bigger the effect of the particle shape on fluid drag. With
226 F. Dioguardi, D. Mele / Powder Technology 277 (2015) 222–230
5 cm
0 cm
Fig. 4. Two successive frames of a video footage of falling experiments. The scale bar is shown on the left side of the two pictures. The black lines represent the lowest position reached by
the falling particle; this position is monitored for measuring the particle terminal velocity wt.
the aim to increase the correlation coefficient, the authors considered
different values of exp. With this procedure the following power law
was obtained:
 
2 1:6 0:5206
Re ¼ 1:0387 C d Re Ψ
which is valid for the Re range of 50–104, although it was later stated
that this range could be extended down to a value of 10 [25].
Thanks to our experiments with fluids of higher viscosity, we
expanded the dataset and extended the Re values down to 0.03 (see
table.xlsx file). To look for optimum correlation, we applied a similar
procedure to that employed by Dellino et al. [3], i.e. we multiplied the
quantity CdRe2 by Ψexp, but we did not search for a fixed value of exp
because of the following:
1) The fitting law should be appropriate also for very low Re numbers,
meaning that Cd varies significantly with Re.
2) Cd is also dependent on Ψ.
As we want the law to be a function of the particle shape factor, we
checked that the shape influence on particle terminal velocity is evident
even at the lowest Re values, i.e. in the Stokes regime. We then calculat-
ed the particle terminal velocity with the Stokes's law [7], wt,Stokes, and
compared it to the measured terminal velocity wt,meas (see the sheet
“Experiments” in the table.xlsx file). The difference between the two
terminal velocities is evident, with an almost systematic overestimate
of wt,Stokes with respect to wt,meas, meaning that the non-spherical
shape, which is not taken into account in the Stoke's law, has a strong
influence on particles terminal velocity even at the lowest Re flow
regimes.
Thus it was convenient to let exp vary as a function of Re in the whole
investigated Re range. Therefore we searched for a relationship similar
to Eq. (12), but of this functional form:
 
2 f ðReÞ
Re ¼ f C d Re Ψ : ð13Þ
5 cm
0 cm
4. Results and discussion
By means of the processing of our dataset, which with the addition
of the new experiments now consists of 340 experimental data, we
searched for the best correlation law expressed in the form of Eq. (13).
In order to find the best fit, we had first to find the best functional rela-
ð12Þ
tionship between the shape factor and the Reynolds number, which
eventually resulted in the form:
a
exp ¼ f ðReÞ ¼ Re ð14Þ
where a is an exponent. In agreement with what is reported in the liter-
ature, it was not possible to find a unique correlation law of the type
(Eq. (13)) for the entire range of Re. However, upon normalizing the par-
ticle drag coefficient Cd by the spherical particle drag coefficient Cd,sphere
at the corresponding value of Re, it was possible to find a unique corre-
lation law and, thanks to the new experimental data, we modified the
previous law of Dellino et al. [3] and extended its validity also in the
range of 0–50, which was not originally covered by Eq. (12). Cd,sphere
was calculated by applying the law of Clift and Gauvin [30], which effi-
ciently cover the entire Re range of the standard drag curve for spherical
particles. The data and fitting law are shown in the logarithmic plots in
Fig. 5. The best fit was obtained with a power law:
 
2 exp 0:4826
Re ¼ 1:1883 C d;norm Re Ψ ð15Þ
where exp = f(Re) is equal to Re−0.23 and Re0.05 for the Re ranges of 0–50
and 50–10,000, respectively, and Cd,norm is the ratio between Cd and
Cd,sphere. The relationship (Eq. (15)) can be easily written in the usual
form Cd = f(Re):
  1
C d;sphere Re 0:4826
Cd ¼ : ð16Þ
Re2 Ψexp 1:1883
With Cd it is then possible to calculate the particle terminal velocity
by using Eq. (1). In this way, since from the experiments we know Re
(otherwise an iterative procedure should be employed, as explained
 F. Dioguardi, D. Mele / Powder Technology 277 (2015) 222–230 227

10000.00
y = 1.1883x0.4826 where K1 and K2 are complex functions of particle sphericity. Finally the
R² = 0.998
law of Chien [20] is much simpler:
1000.00
30 −5:03Φ
Cd ¼ þ 67:289e : ð21Þ
Re
100.00

After the calculation of Cd,calc with the three aforementioned laws,

for each experiment we calculated the terminal velocity wt,calc by
Re

10.00
using Eq. (1) and we compared it with the measured one wt,meas and
then the error err% with Eq. (17). Data of these comparisons are report-
1.00 ed in the second sheet “Law comparisons” in the file table.xlsx available
in the Supplementary Material. In Fig. 6, a plot of wt,calc vs. wt,meas is
displayed for each law, including ours (a: Chien [20]; b: Swamee and
0.10
Ojha [23]; c: Ganser [18]; d: our model). It is worth noting that with
our law (Fig. 6d) the measured terminal velocities are better predicted
0.01 than with the other models; this is demonstrated not only by the
1E-04 1E-02 1E+00 1E+02 1E+04 1E+06 1E+08 value of the correlation coefficient R2 (0.937), but also from the equation
Cd,normRe2Ψexp
of the fitting line. It was forced to pass through the origin of the axes,
thus the perfect agreement between wt,calc and wt,meas would be the
Fig. 5. Correlation law for the entire Re range between Re and Cd,normRe2Ψexp. The fitting
case of an angular coefficient equal to 1 (y = x line). This value is vari-
law is reported together with the correlation coefficient R2.
ously approximated by the literature models (the best one being that
of Ganser, with an angular coefficient of 0.949), while with our model
hereinafter), we re-calculated the terminal velocity for all the experi- the angular coefficient is almost equal to 1 (0.99) (a certain degree of
mental runs (wt,calc) and we compared them with the measured ones scatter is visible with all the models at the highest terminal velocities).
(wt,meas). For each run we calculated the error introduced with this Concerning errors, with our model the average |err%| is 11.14%, which
procedure as: is lower than that resulting by applying the models of Chien (15.03%)
and Swamee and Ojha (27.76%), and of the same order of that of Ganser
wt;rec −wt;meas
err% ¼ 100: ð17Þ (11.06%). It can be thus stated that Ganser's model works quite well
wt;meas compared to other laws, and gives an average error |err%| comparable
with ours, but the linear correlation between wt,calc and wt,meas differs
The average absolute |err%| resulting by the use of Eqs. (1) and (16) is significantly from the equality line. With our model, instead, despite a
11.14%. As the instrumental error in the measurement of wt was quan- certain scatter, data are well distributed around this line, thus it can be
tified in the order of 1%, we think that most part of this error is to be at- concluded that our new model is able to predict quite well the experi-
tributed in the secondary movements (like rotations, tumbling, etc.) mentally measured particle terminal velocities of irregularly shaped
affecting the fall trajectory of the irregular particles. This effect cannot particles over a wide range of Re, and it compares favorably with other
be accounted for by Eq. (1). These secondary movements were unavoid- models of the literature.
able in our experiments with highly irregular particles. Another source In Fig. 7 plots of Cd vs. Re are shown, as obtained with the same
of error could be the wall effect, although we discarded and repeated models of Fig. 6. The black circles represent the measured value of the
experiments in which particles deviated too much from the central drag coefficient Cd,meas (Eq. (10)), and the gray squares represent the
vertical line of the glass cylinder. calculated value Cd,calc (Eqs. (18), (20) and (21)). With our model, Cd,calc
In order to evaluate how the average |err%| we obtained compares was obtained by using Eq. (16). The analysis of Fig. 7 allows understand-
with other models, we evaluated Cd and wt of our particles with other ing the origin of the error in the settling velocity prediction of each model.
well known laws available in the literature. Unfortunately some litera- The worst performance is clearly represented in Fig. 7b, which is the plot
ture models were not directly usable with our data, because in some for the model of Swamee and Ojha: the gray squares representing Cd,calc
cases it was not possible to evaluate the particle shape factor as defined follow a trend that is significantly different from that of the black circles
in those laws [2,21]. Our comparison was possible with three other representing Cd,meas. This is the reason of the large error resulting by the
models, those of Swamee and Ojha [23], Ganser [18] and Chien [20]. application of this law in the prediction of particle terminal velocity,
These models are based on relationships that are mainly a function of especially for the lowest terminal velocities (b 15 cm s− 1). In the
particle sphericity Φ [18,20] or on the three major axes dl, dm, ds [23], other three cases Cd,calc almost follows the trend of Cd,meas, but in the
all available in our dataset. The law of Swamee and Ojha [23] is as case of Chien (Fig. 7a) and Ganser (Fig. 7c) the gray squares are more
follows: dispersed than the black circles. With our model (Fig. 7d) the gray
"  0:32
# square representing Cd,calc are distributed over the same dispersal area
48:5 Re 1 as Cd,meas.
Cd ¼  0:8 0:64 þ
1 þ 4:5β0:35 Re Re þ 100 þ 100β β18 þ 1:05β0:8 Despite the model of Dellino et al. [3], of which our new model can
ð18Þ be considered as an evolution, with the formulation we obtained
Eq. (15), it is not possible to obtain a law for the particle terminal
in which β is the Corey shape factor, defined as: velocity that is not explicitly dependent on the particle Reynolds
number Re, as it can be inferred by analyzing Eq. (16), where exp is a
ds function of Re unlike Eq. (12), where the exponent of the shape factor
β ¼ pffiffiffiffiffiffiffiffiffiffi : ð19Þ
dl dm was a fixed number. For these reasons, the use of this new law for calcu-
lating the terminal velocity of a particle falling in a fluid requires itera-
The law of Ganser [18] is a somewhat more complicated: tive calculations. We provide a Microsoft Office Excel spreadsheet
(wt_calculator.xlsx) and a Fortran 90 source code (wt_calculator.f90)
Cd 24 n 0:6567
o 0:4305 as Supplementary Material, which calculate recursively Re, Cd and wt.
¼ 1 þ 0:1118ðReK 1 K 2 Þ þ ð20Þ
K 2 ReK 1 K 2 3305 The user should give as input particle (dp, ρp, Ψ) and fluid (ρ, μ) charac-
1þ
ReK 1 K 2 teristics. The iterative procedure, which is schematized in the flow chart
228 F. Dioguardi, D. Mele / Powder Technology 277 (2015) 222–230

Chien a Swamee & Ojha b

70 70

60 60
y = 0.931x y = 0.943x
R² = 0.833 R² = 0.923
50 50
wt,calc (cm s-1)

wt,calc (cm s-1)

40 40

30 30

20 20

10 10

0 0
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
-1
wt,meas (cm s ) wt,meas (cm s-1)

Ganser c Our law d

70 70

60 60 y = 0.99x
y = 0.949x
R² = 0.882 R² = 0.937
50 50

wt,calc (cm s-1)

wt,calc (cm s-1)

40 40

30 30

20 20

10 10

0 0
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
wt,meas (cm s-1) wt,meas (cm s-1)

Fig. 6. Plot of wt,calc vs. wt,meas for three different laws and our new law. The linear interpolation law is also reported with R2. a. Chien [20]. b. Swamee and Ojha [23]. c. Ganser [18]. d. Our law.
It can be observed that with our law the linear fitting is very close to the perfect agreement (y = x) between wt,calc and wt,meas, although with a certain degree of scatter at the highest
terminal velocities.

Chien a Swamee & Ojha b

10000.0 10000.0

1000.0 1000.0

100.0 100.0
Cd

Cd

Cd,meas Cd,meas
Cd,calc 10.0 Cd,calc

1.0 1.0

0.1 0.1
0.01 0.10 1.00 10.00 100.00 1000.00 10000.00 0.01 0.10 1.00 10.00 100.00 1000.00 10000.00
Re Re

Ganser c Our law d

10000.0 10000.0

1000.0 1000.0

100.0 100.0
Cd
Cd

Cd,meas Cd,meas
10.0 Cd,calc 10.0 Cd,calc

1.0 1.0

0.1 0.1
0.01 0.10 1.00 10.00 100.00 1000.00 10000.00 0.01 0.10 1.00 10.00 100.00 1000.00 10000.00
Re Re

Fig. 7. Plot of Cd vs. Re for three different laws and our new law. The black circles represent Cd,meas, and the gray squares represent Cd,calc. a. Chien [20]. b. Swamee and Ojha [23]. c. Ganser
[18]. d. Our law. It can be observed that with our law Cd,meas is better predicted than with the other models, as the gray squares are distributed over the area occupied by the black circles.
 F. Dioguardi, D. Mele / Powder Technology 277 (2015) 222–230
Guess an initial
Re*
Calculate Cd
with eq. 16
Calculate wt
with eq. 1
Calculate the new
value of Re
Calculate residuals:
Renew-Reold
wt,new-wt,old
Cd,new-Cd,old
NO
Residuals < Tolerance?
YES
end
Fig. 8. Flow chart of the iterative procedure to be implemented for the calculation of Re, Cd
and wt.
shown in Fig. 8, is easy to implement also in source codes of other
programs: given a guessed value of the particle Reynolds number Re*,
Cd is calculated by Eq. (16), from which wt can be computed with the
Newton's impact law (Eq. (1)). With this terminal velocity, a new
value of Re is calculated, and the subsequent operations are repeated
until the difference between the latest updated values of Re, Cd and wt
and those of the previous steps drops to machine precision, or to a
user defined tolerance.
5. Conclusions
We presented a new drag law for non-spherical particles based on a
series of settling experiments at different fluid dynamics regimes. The
experiments cover a Re range of 0.03–10,000, thus our law is suitable
for non-spherical particles falling into both laminar and turbulent con-
ditions. The law depends on a particle shape factor Ψ, which is defined
as the ratio between sphericity Φ and circularity X, both obtained by
image particle analysis. Our model compares favorably with literature
laws when predicting terminal velocity, and it also performs well in
predicting the drag coefficient.
The drag law formulation does not allow to obtain a relationship for
the particle terminal velocity independent of the particle Reynolds
number, a condition that was instead satisfied in Dellino et al. [3]. For
this reason we propose an iterative procedure for calculating Re, Cd
and wt, given particle and fluid physical parameters.
This law, thanks to the extension of the Re range down to very low
values (0.03), can now be used in fluid dynamics regimes that were
not covered by the previous law of Dellino et al. [3], e.g. the very slow
settling of volcanic ash or pollutants in the atmosphere or water [31,32].
229
Because of the use of rough non-spherical particles, which cannot be
described by specific geometries unlike many other examples in the
literature (e.g. [18,21]), it was not possible to obtain a law that collapse
on the standard drag curve for perfect spherical particles at any value of
Re. The user should be aware of this limitation, thus avoiding to use our
law for spherical particles especially in the Stokes region.
In the future the objective is to speed up the measure of the particle
shape factor, that to date is a time-consuming laboratory procedure
merging particle analysis and image analysis. Furthermore, it will be of
interest to try to extend this single particle drag-law to a mixture of
particles, with the aim of obtaining a multiphase drag law suitable for
the calculation of interphase drag force in multiphase flows where the
solid phase is made of irregular particles.
Supplementary data to this article can be found online at http://dx.
doi.org/10.1016/j.powtec.2015.02.062.
Acknowledgments
We would like to thank Prof. Pierfrancesco Dellino for his valuable
suggestions during the experiments and the writing of the manuscript.
This work was partly supported by National Civil Protection and INGV
within the DPC-INGV agreement 07-09, Project V1: Probabilistic volcanic
hazard evaluation.
References
[1] R.P. Chhabra, L. Agarwal, N.K. Sinha, Drag on non-spherical particles: an evaluation
of available methods, Powder Technol. 101 (1999) 288–295, http://dx.doi.org/10.
1016/S0032-5910(98)00178-8.
[2] S. Tran-Cong, M. Gay, E.E. Michaelides, Drag coefficients of irregularly shaped parti-
cles, Powder Technol. 139 (2004) 21–32, http://dx.doi.org/10.1016/j.powtec.2003.
10.002.
[3] P. Dellino, D. Mele, R. Bonasia, G. Braia, L. La Volpe, R. Sulpizio, The analysis of the
influence of pumice shape on its terminal velocity, Geophys. Res. Lett. 32 (2005)
L21306, http://dx.doi.org/10.1029/2005GL023954.
[4] F. Dioguardi, P. Dellino, D. Mele, Integration of a new shape-dependent particle–
fluid drag coefficient law in the multiphase Eulerian–Lagrangian code MFIX-DEM,
Powder Technol. 260 (2014) 68–77, http://dx.doi.org/10.1016/j.powtec.2014.03.
071.
[5] F. Dioguardi, P. Dellino, PYFLOW, a computer code for the calculation of the impact
parameters of dilute pyroclastic density currents (DPDC) based on field data,
Comput. Geosci. 66 (2014) 200–210, http://dx.doi.org/10.1016/j.cageo.2014.01.013.
[6] T. Dürig, D. Mele, P. Dellino, B. Zimanowski, Comparative analyses of glass fragments
from brittle fracture experiments and volcanic ash particles, Bull. Volcanol. 74 (3)
(2012) 691–704, http://dx.doi.org/10.1007/s00445-011-0562-0.
[7] C.T. Crowe, M. Sommerfield, Y. Tsuji, Multiphase Flows With Droplets and Particles,
CRC Press, Boca Raton, Fla., 1998
[8] L. Schiller, A. Naumann, Über die grundlegende Berechnung bei der
Schwekraftaufbereitung, Ver. Dtsch. Ing. 77 (1933) 318–320.
[9] J.M. Dallavalle, Micrometrics, the Technology of Fine Particles, Pitman Publishing,
New York, 1948.
[10] R. Clift, J.R. Grace, M.E. Weber, Bubbles, Drops and Particles, Academic Press, New
York, 1978.
[11] F.M. White, Fluid Mechanics, McGraw-Hill, New York, 1991.
[12] A.N. Singh, K.C. Roychowdhury, Study of the effects of orientation and shape on the
settling velocity of non-isometric particles, Chem. Eng. Sci. 24 (1969) 1185–1186.
[13] A. Unnikrishan, R.P. Chhabra, An experimental study of motion of cylinders in
Newtonian fluids: wall effects and drag coefficient, Can. J. Chem. Eng. 69 (1991)
729–735, http://dx.doi.org/10.1002/cjce.5450690315.
[14] D. Rodrigue, D. DeKee, R.P. Chhabra, Drag on non-spherical particles in non-
Newtonian fluids, Can. J. Chem. Eng. 72 (1994) 588–593, http://dx.doi.org/10.
1002/cjce.5450720406.
[15] E.K. Marchildon, A. Clamen, W.H. Gauvin, Drag and oscillatory motion of freely falling
cylindrical particles, Can. J. Chem. Eng. 42 (1964) 178–182, http://dx.doi.org/10.
1002/cjce.5450420410.
[16] E. Loth, Drag of non-spherical solid particles of regular and irregular shape, Powder
Technol. 182 (2008) 342–353, http://dx.doi.org/10.1016/j.powtec.2007.06.001.
[17] A. Haider, O. Levenspiel, Drag coefficient and terminal velocity of spherical and non-
spherical particles, Powder Technol. 58 (1989) 63–70, http://dx.doi.org/10.1016/
0032-5910(89)80008-7.
[18] G.H. Ganser, A rotational approach to drag prediction of spherical and nonspherical
particles, Powder Technol. 77 (1993) 143–152, http://dx.doi.org/10.1016/0032-
5910(93)80051-B.
[19] M. Hartman, J.G. Yates, Free-fall of solid particles through fluids, Collect. Czechoslov.
Chem. Commun. 58 (5) (1993) 961–982.
[20] S.F. Chien, Settling velocity of irregularly shaped particles, SPE Drill. Complet. 9
(1994) 281–288.
230 F. Dioguardi, D. Mele / Powder Technology 277 (2015) 222–230
[21] A. Hölzer, M. Sommerfeld, New simple correlation formula for the drag coefficient of
non spherical particles, Powder Technol. 184 (2008) 361–365, http://dx.doi.org/10.
1016/j.powtec.2007.08.021.
[22] J. Baba, P.D. Komar, Settling velocity of irregular grain at low Reynolds numbers, J.
Sediment. Petrol. 51 (1981) 121–128.
[23] P.K. Swamee, C.P. Ojha, Drag coefficient and fall velocity of nonspherical parti-
cles, J. Hydraul. Eng. 117 (1991) 660–669, http://dx.doi.org/10.1061/(ASCE)0733-
9429(1991)117:5(660).
[24] G.H. Bagheri, C. Bonadonna, I. Manzella, P. Pontelandolfo, P. Haas, Dedicated vertical
wind tunnel for the study of sedimentation of non-spherical particles, Rev. Sci.
Instrum. 84 (2013) 054501, http://dx.doi.org/10.1063/1.4805019.
[25] P. Dellino, M.T. Gudmundsson, G. Larsen, D. Mele, J.A. Stevenson, T. Thordarson, B.
Zimanowski, Ash from the Eyjafjallajökull eruption (Iceland): fragmentation pro-
cesses and aerodynamic behaviour, J. Geophys. Res. 117 (2012) B00C04, http://dx.
doi.org/10.1029/2011JB008726.
[26] D. Mele, P. Dellino, R. Sulpizio, G. Braia, A systematic investigation on the aerody-
namics of ash particles, J. Volcanol. Geotherm. Res. 203 (2011) 1–11, http://dx.doi.
org/10.1016/j.jvolgeores.2011.04.004.
[27] R. Büttner, P. Dellino, L. La Volpe, V. Lorenz, B. Zimanowski, Thermohydraulic explo-
sions in phreatomagmatic eruptions as evidenced by the comparison between
pyroclasts and products from molten fuel coolant interactions experiments, J.
Geophys. Res. 107 (2002). http://dx.doi.org/10.1029/2001JB000511.
[28] P. Dellino, L. La Volpe, Image processing analysis in reconstructing fragmentation
and transportation mechanisms of pyroclastic deposits. The case of Monte Pilato-
Rocche Rosse eruptions, Lipari (Aeolian islands, Italy), J. Volcanol. Geotherm. Res.
71 (1996) 13–29.
[29] N.S. Cheng, Formula for viscosity of glycerol–water mixture, Ind. Eng. Chem. Res. 47
(2008) 3285–3288.
[30] R. Clift, W.H. Gauvin, Proc. CHEMECA '70, vol. 1, Butterworth, Melbourne, 1970,
pp. 14–28.
[31] R. Sulpizio, A. Folch, A. Costa, C. Scaini, P. Dellino, Hazard assessment of far-range
volcanic ash dispersal from a violent Strombolian eruption at Somma-Vesuvius
volcano, Naples, Italy: implications on civil aviation, Bull. Volcanol. 74 (2012)
2205–2218, http://dx.doi.org/10.1007/s00445-012-0656-3.
[32] A. Folch, A review of tephra transport and dispersal models: evolution, current
status, and future perspectives, J. Volcanol. Geotherm. Res. 235–236 (2012) 96–115,
http://dx.doi.org/10.1016/j.jvolgeores.2012.05.020.