Computers & Fluids 113 (2015) 53–64

Contents lists available at ScienceDirect

Computers & Fluids

j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / c o m p fl u i d

Drag, lift and torque coefficients for ellipsoidal particles: From low to
moderate particle Reynolds numbers
Rafik Ouchene a, Mohammed Khalij a, Anne Tanière a,⇑, Boris Arcen b
a
CNRS, LEMTA, UMR 7563, Université de Lorraine – ESSTIN, 2 rue Jean Lamour, 54500 Vandoeuvre-les-Nancy, France
b
CNRS, LRGP, UMR 7274, Université de Lorraine, Nancy F-54000, France

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

Article history: An accurate prediction of the translational and rotational motion of ellipsoidal particles can be only given
Received 3 December 2013 if a complete set of correlations of the drag, lift and pitching torque coefficients is known. The present
Received in revised form 18 November 2014 study is thus devoted to the assessment of the available correlations in the literature through a
Accepted 4 December 2014
comparison with numerical results of the forces acting on a particle given by a full body-fitted direct
Available online 19 December 2014
numerical simulation (DNS) in the case of a uniform flow, for three different ellipsoidal particles, and
for Reynolds number ranging from 0.1 to 290. The comparison between the computed force (or hydrody-
Keywords:
namic coefficients) and the literature correlations shows clearly that certain precautions must be taken.
Non-spherical particles
Drag
The mean deviation of the most accurate correlation considered in the present study from our full DNS
Lift results can be of the order of 20% for the drag coefficient. For the lift and pitching torque coefficients,
Torque the deviations can increase up to roughly 25%. This comparison shows that further works are definitively
Direct numerical simulation necessary to develop a complete set of correlations for ellipsoidal particles outside Stokes regime.
Body-fitted method Ó 2014 Elsevier Ltd. All rights reserved.

1. Introduction particles were previously proposed and summarized in Clift et al.

The assumption that particles are spherical has been exten-
sively used in the modeling of particle transport in fluid flows for
the past 20 years (Marchioli et al. [1], Arcen et al. [2]). However,
in natural processes as in a variety of engineering applications such
as dispersion of pollutant, pulverized coal combustion, pneumatic
transport, or in all aspects of the pulp and paper industry, many
different shapes are possible. Owing to the significant effect of
the particle shapes on their motion and their practical importance
in hydrodynamic applications, the non-sphericity is more and
more considered in the modeling and simulations of particles
transport in fluid flows (Mandø and Rosendahl [3], Lain and
Sommerfeld [4], Njobuenwu and Fairweather [5], van Wachem
et al. [6]). Unfortunately, it is not possible to consider each shape
in the implementation of numerical models, because of the nonex-
istence of a single approach describing accurately the sizes and
shapes of non-spherical particles. Therefore, irregular particle
shapes are commonly idealized to some regular shapes such as
spheroids, cuboids or cylinders (see Fig. 1). In addition, several
methods of size measurement, shape categorization, and shape
parameters used to describe regular and irregular non-spherical
⇑ Corresponding author.
E-mail address: anne.taniere@univ-lorraine.fr (A. Tanière).
[7], Allen [8], Mandø [9], and Li et al. [10].
Even if the particle shapes are idealized, a complete description
of the behavior of these particles requires the modeling of their
translational and rotational motion taking its orientation into
account, in contrast to the spherical particle case.
Several studies have been performed over the past 20 years, but
they are mainly devoted to the drag force acting on non-spherical
particles. Chhabra et al. [13] were the first to compile and classify
the most used correlations. According to their study they showed
that the correlation derived by Ganser [14] is the most accurate.
They distinguished two types of approach that were used in later
studies. The first one is to define a single correlation for all shapes
and orientations (Tran-Cong et al. [15], Yow et al. [16], Holzer and
Sommerfeld [17]). Indeed, Tran-Cong et al. [15] proposed a corre-
lation from their experimental data obtained from agglomerates
of ordered packed spheres. Yow et al. [16] used data available in
the literature to extract a new correlation which unfortunately,
does not take the orientation into account. This parameter was
introduced in the correlation of Holzer and Sommerfeld [17],
through two shape parameters which are the sphericity (/) and
the crosswise sphericity. The sphericity is defined by Wadell [18]
as the ratio between the surface area of the volume equivalent
sphere and the surface area of the considered particle (S). The
crosswise sphericity, /? , is defined by Holzer and Sommerfeld

http://dx.doi.org/10.1016/j.compfluid.2014.12.005
0045-7930/Ó 2014 Elsevier Ltd. All rights reserved.
54 R. Ouchene et al. / Computers & Fluids 113 (2015) 53–64

[17] as the ratio between the cross-sectional area of the volume
equivalent sphere and the projected cross-sectional area of the
considered particle perpendicular to the flow (S? ). The second
approach consists in obtaining a drag coefficient expression for a
fixed shape and any orientations. Rosendahl [19] suggested to
determine the drag coefficient at two incidence angles (a), i.e. 0°
and 90° ðC D;a¼0 ; C D;a¼90 Þ, from the best experimental results or
from existing correlations. Then, a function linking these two lim-
iting cases, with an inflection point at 45°, is used to represent the
whole range of incidence angles for a non-spherical particle.
Recently, Zastawny et al. [20] proposed new correlations to predict
the drag, lift, and torque (pitching and rotational) coefficients for
non-spherical particles from data given by a direct numerical sim-
ulation carried out with an immersed boundary method. Their cor-
relation for the drag force is derived following a functional form
similar to that proposed by Rosendahl [19], nonetheless C D;a¼0
and C D;a¼90 are given from formulas which are function of the par-
ticle Reynolds number and particle-specific fit parameters. In their
study, Zastawny et al. [20] compared their drag coefficient for
ellipsoidal particles with the theoretical expression provided by
Happel and Brenner [21] at low Reynolds numbers. At higher
Reynolds numbers, the comparison is conducted with the formula
proposed by Rosendahl [19], the drag coefficient at 0° and 90°
being given by the correlation proposed by Holzer and Sommerfeld
[17]. At low and higher Reynolds numbers, some significant differ-
ences were noted. Currently, given this uncertainty on the accuracy
of the exiting drag coefficient correlations for ellipsoidal particles,
it is difficult to make a choice about the drag correlation to use.
This choice is all the more difficult for the lift or torque coefficients
since there are few studies devoted to these coefficients in the lit-
erature. In some studies, a proportionality of the lift coefficient
with the drag one is assumed (Yin et al. [22], Hoerner [23]).
Nevertheless, Zastawny et al. [20] showed the weakness of this
assumption through their DNS results and provided a new correla-
tion valid for ellipsoidal particles and for two aspects ratios. For
instance, they proposed the following relation for the lift coeffi-
cient (Table 1 includes all the correlations proposed by Zastawny
et al. [20]):
!
b1 b3 b7 b10
CL ¼  sin ðaÞb5 þb6 Rep cos ðaÞb8 þb9 Rep ;
Rebp2 Rebp4
where the coefficients, bi, are specific to the aspect ratio. At the
present time, there is no correlation of the lift coefficient which
can be applied for any particle aspect ratio. Concerning the pitching
torque, it is generally accepted that it is intimately linked to the
drag and lift forces which act at the pressure center, its location
being different from the gravity center, this induces the pitching
torque. The only formula for the pitching torque coefficient was
given by Zastawny et al. [20] for ellipsoidal particles and for specific
values of the aspect ratio.
Up to now, there is no set of correlations of the drag, lift and
pitching torque coefficients for ellipsoidal particles outside the
Stokes regime which takes into account the incidence angle,
aspect ratio, and Reynolds number. Therefore, when the motion
of ellipsoidal particles has to be predicted, some choices have to
be made among available correlations even if they were derived
under strong assumptions. For instance, several authors studied
the motion of inertial ellipsoidal particles using the theoretical
expression for the force and torque derived under the assumption
of low Reynolds number (Zhang et al. [24], Mortensen [25],
Marchioli et al. [26], Tian and Ahmadi [27], Marchioli and Soldati
[28], Zhao et al. [29]). The torque was obtained from the expres-
sion derived by Jeffery [30] while the force was computed from
the expression given in Happel and Brenner [21]. However, when
the particle inertia is large, the low Reynolds number assumption
is not valid anymore. Due to the lack of correlations outside the
Stokes regime, researchers are therefore generally forced to study
the non-spherical particle motion under the creeping flow
assumption. An accurate prediction of the complete motion of
non-spherical particles in a turbulent flow by a Lagrangian
approach and outside the Stokes regime would be possible if an
accurate set of hydrodynamic coefficient correlations was avail-
able. The aim of the present work is thus twofold. Firstly, to guide
the reader through the choice of correlations of the drag, lift and
torque coefficients outside the Stokes regime. Secondly, to pro-
vide data on the drag, lift, and pitching torque coefficients for dif-
ferent ellipsoidal particles by a full direct numerical simulation
using a body-fitted approach. It is hoped that this database will
help to develop models of these coefficients.
The document is organized in three sections. The first one is
devoted to the description of the model of the motion of non-
spherical particles and to the main available correlations. In the
second section, the numerical simulation technique and some
ð1Þ the validation cases are presented. Then, the results obtained
for the non-dimensional hydrodynamical parameters (drag, lift,
and pitching torque coefficients) are given, for three different par-
ticle shapes, and discussed in the third section before the
conclusion.

Table 1
Drag, lift, and torque correlations.

Authors Correlations
Holzer and Sommerfeld [17] C D ¼ Re8p p1ffiffiffiffiffi þ Re16 p 1ffiffiffi
þ p3ffiffiffiffiffiffi /3=4
1
0:2
þ 0:42  100:4ð log /Þ /1 / is the sphericity and /? the crosswise sphericity
/? p / Rep ?
8
Rosendahl [19] 3
>
> C D ¼ C D; a ¼90  þ ðC
D; a ¼90   C
D; a ¼0  Þ sin a
>
>
>
< Where : 0:2
C D;a¼0 ¼ Re8p p1ffiffiffiffiffi þ Re 16 p 1ffiffiffi
þ p3ffiffiffiffiffiffi /3=4
1
þ 0:42  100:4ð log /Þ /1
>
> /? p / Rep ?
>
> 0:2
>
: C D;a¼90 ¼ Re8p p1ffiffiffiffiffi þ Re 16 p 1ffiffiffi
þ p3ffiffiffiffiffiffi /3=4
1
þ 0:42  100:4ð log /Þ /1?
/? p / Rep
8 a
Zastawny et al. [20] >
> C ¼ C D;a¼90 þ ðC D;a¼90  C D;a¼0 Þ sin 0 a
> D
>
< where
C D;a¼0 ¼ a1a2  a3a4
>
> Rep Rep
>
>C a5 a7
: D;a¼90 ¼ a  a
Rep6 Rep4
  b7 b10
CL ¼ b1
b  b3
b sin ðaÞb5 þb6 Rep cos ðaÞb8 þb9 Rep
Rep2 Rep4
  c7 c10
CT ¼ c1
c  c3
c sin ðaÞc5 þc6 Rep cos ðaÞc8 þc9 Rep
Rep2 Rep4

Hoerner [23] CL
¼ sin2 a cos a
CD
 R. Ouchene et al. / Computers & Fluids 113 (2015) 53–64 55

Fig. 1. Example of spheroid coal–particle shapes with different aspect ratios. Picture was taken from Brix et al. [11] and Powers [12].

Fig. 2. 3d Flow configuration in the inertial frame (x, y, z) (left). 3d Flow configuration in, respectively, co-moving (x0 , y0 , z0 ) and co-rotating (x00 , y00 , z00 ) frames references (right).
The angle of incidence a is defined as the angle between the direction of the fluid velocity and the longest axis of the particle. a and b are respectively the longest and smallest
semiaxes.

2. Motion of a non-spherical particle Table 2

Particle shapes.
There are few models that describe the motion of non-spherical Ellipsoid w = 5/1 Ellipsoid w = 5/2 Ellipsoid w = 5/4
particles suspended in a fluid flow. Indeed, this type of particle is
not only subjected to translation forces but also to a rotational
motion around itself. Let ðx; y; zÞ, ðx0 ; y0 ; z0 Þ, ðx00 ; y00 ; z00 Þ be respec-
tively the inertial, co-moving and co-rotating frames of reference
(see Fig. 2). The translational motion of a particle is described in
the inertial frame of reference by the conservation of momentum.
In addition, for ‘‘heavy’’ solid particles in a gas flow, the drag ~ F D , lift
~
F L and volume forces control the particle motion [24]:
d~
up ~
mp ¼ FD þ ~
F L þ V p ðqp  qf Þ~
g; ð2Þ where Ix00 , Iy00 , Iz00 represent the moment of inertia; xx00 , xy00 , xz00 are
dt
the angular velocities according to the co-rotating frame, and T x00 ,
where mp , ~ up , V p , qp , represent respectively the particle mass, T y00 , T z00 are the torque relative to the three directions.
velocity, volume, and density, qf is the fluid density, and ~ g is the Usually, the forces acting on a non-spherical particle are
gravitational acceleration. characterized by the dimensionless drag CD, lift CL and torque CT
The rotational motion for non-spherical particles is induced by coefficients which depend on the particle Reynolds number
the different location of the center of mass and the center of pres- Rep ¼ k~ uR kdp =m, where ~
uR ¼ ~
up  ~
uf is the relative velocity between
sure. Therefore, an additional torque is added and the following the particle and the fluid, ~ uf represents the velocity of the fluid at
equations have to be solved:
8 dx 00
>
> I 00 x  xy00 xz00 ðIy00  Iz00 Þ ¼ T x00 ;
> x dt
< Table 3
dxy00 Parameters of the simulations.
Iy00  xz00 xx00 ðIz00  Ix00 Þ ¼ T y00 ; ; ð3Þ
>
> dt
>
: dxz00
Rep Particle Ratio w = a/b Orientation a
Iz00 dt
 xx00 xy00 ðIx00  Iy00 Þ ¼ T z00 ;
0.1, 10, 60, 240, 290 Ellipsoids 5/1, 5/2, 5/4 0–90°
56
the particle location, dp is the volume equivalent spherical diame-
ter, and m is the fluid kinematic viscosity. Thus, these coefficients
are obtained using the following formulations:
k~
FDk k~
FLk
CD ¼ ; CL ¼ ; CT ¼
1
2
qf u2R p4 d2p 1
2
qf u2R p4 d2p 1
2
qf u2R p8 d3p
For a particle which is not symmetric (unlike the sphere), the
drag and lift coefficients depend not only on particle Reynolds
number but also on its incidence angle a with respect to the flow
(see Fig. 2). This effect plays an important role in the motion of
non-spherical particles [23] through the drag and lift forces. Hence,
for an accurate description of non-spherical particles motion, the
drag, lift and torque coefficients have to be precisely predicted.
Table 1 reports the correlations that will be compared to our
numerical results in order to assess their accuracy. Concerning
the drag coefficient three correlations have been chosen for the
comparison. They were developed by Rosendahl [19], Holzer and
Sommerfeld [17] and Zastawny et al. [20]. The lack of lift and
torque correlations has conducted us to use the correlations pre-
sented in the introduction. For the lift, the expressions by Hoerner
[23] and Zastawny et al. [20] will be assessed, and for the torque,
we will consider the correlation by Zastawny et al. [20].
3. Numerical overview
3.1. Numerical technique
We want to examine the hydrodynamic forces acting on ellip-
soids of revolution (Table 2), i.e. the minors semi-axes b and c
 
are equal (b = c). The range of aspect ratios w ¼ length
width
¼ ba consid-
ered in this study is w 2 ½1  5 (see Table 2). This is the typical
range encountered in many technical processes such as coal com-
bustion (see Fig. 1). In addition, the particle Reynolds number in
many technical processes is, generally, lower than 300. Thus, in
this study, we studied the hydrodynamic forces acting on ellipsoi-
dal particles for Rep 2 ½0:1  300½. In Table 3, the aspect ratios and
particle Reynolds numbers computed are summarized.
Besides, it is known that the flow past such particles becomes
non-axisymmetric for Rep > 200 [20,31–35]. Hence, it was required
to consider simulation in a complete 3D configuration. Thus, the
particle is supposed fixed and immersed in a tridimensional uni-
form flow.
3.2. Governing equation and numerical details
In the present study, it is assumed that the flow is isothermal,
incompressible, the fluid is considered as Newtonian. The unsteady
Fig. 3. Geometrical configuration and boundary conditions.
R. Ouchene et al. / Computers & Fluids 113 (2015) 53–64
term in the Navier–Stokes equations has been neglected. The final
form of governing equations reads as:
r ~
u ¼ 0;
k~
Tk 1 ð5Þ
: ð4Þ u  rÞ~
ð~ u¼ rp þ mD~
u;
qf
where p is the pressure. In this study, the flow was thus predicted
by solving steady-state equations although it is known that the flow
regime can be unsteady for the highest particle Reynolds number
considered (Rep = 290). The unsteadiness of the flow past a sphere
occurs for Rep 2 ½270  280 [32], while for an ellipsoidal particle
with w = 2 aligned or perpendicular to the flow, the flow remains
steady for Rep < 250 [35]. Nevertheless, as can be noted from the
data given in Johnson and Patel [32] for a sphere at Rep = 300, the
fluctuations of the drag coefficient around its time-averaged value
are weak (roughly 0.5% of the time-averaged drag coefficient). From
a preliminary steady computation of flow past a sphere at Rep = 300,
it was shown that the difference between the present drag coeffi-
cient and that of Johnson and Patel [32] is of 3.7%. It is believed that
the unsteadiness of the flow has little effect on the drag, lift, and
torque coefficients for the particle Reynolds number range consid-
ered in the present study. The drag and lift are the components of
the fluid forces (pressure and friction) acting on the considered par-
ticle in the three principal directions. These forces are obtained by
integrating the total stress acting on the particle surface area. Thus,
both forces are:
Z Z
k~
F D k ¼ F D;pressure þ F D;friction ¼  p~ x ~
ndS þ ~ n  s ~
xdS
Z S ZS ð6Þ
k~
F L k ¼ F L;pressure þ F L;friction ¼  p~ y ~
ndS þ ~ n  s ~
ydS;
S S
where s and ~ n are respectively the viscous stress tensor and the
normal vector to the particle surface. The pitching torque is the
torque induced by the drag and lift forces acting on the center of
pressure ~
xcp :
~
T ¼ ð~
FD þ ~
FLÞ  ~
xcp : ð7Þ
Simulations are performed using the CFD code Ansys FluentÓ. Cen-
tered scheme are used for discretizing all the numerical fluxes in
the Navier–Stokes equations. The problem of the coupling between
velocity and pressure is tackled with a Semi-Implicit Method for
Pressure-Linked Equations (SIMPLE) [31]. The geometrical configu-
ration (Fig. 3) is a parallelepiped fluid box surrounding the particle.
The dimensions of the box were determined to prevent any distur-
bance of the flow due to boundary conditions, particularly in the
boundary layer. From a practical point of view, these dimensions
were imposed by the case Rep = 0.1 resulting in a greater boundary
layer where the particle is located transversely to the flow. In addi-
tion, the length of the computational box was set to have an ade-
quate distance for the development of the wake. The mesh of the
Fig. 4. Computational mesh for flow past ellipsoidal particle.
 R. Ouchene et al. / Computers & Fluids 113 (2015) 53–64
fluid domain consists of tetrahedral elements. It was verified that
the mesh surface of the particle is sufficiently fine to give an accu-
rate integration of the forces acting on the particle surface (Fig. 4).
The influence of the number of mesh cells on the hydrodynamic
coefficients for an ellipsoid (w = 5/1) and an incidence angle equals
to 90° at Rep = 290 is shown in Table 4. As it can be seen, the drag
coefficient can be considered mesh independent with a number of
cells equals to 650 000. The difference with the finest grid result
is less than 1%. For this aspect ratio and incidence angle, the com-
puted values of the lift and torque coefficients are close to zero
since the wake is axisymmetric. Thus, the number of cells chosen
for all the simulation is always greater than 500 000 (this number
varies with w and a). It has to be noted that the boundary layer is
always represented by at least 20 nodes.
The boundary conditions are presented in Fig. 3. A constant
velocity inflow condition is imposed at the inlet boundary whereas
homogeneous Neumann conditions are chosen at the outlet. On
lateral boundaries, no-slip Dirichlet condition is specified.
3.3. Validation cases
Before examining the results obtained for the cases given in
Table 3, the computational methodology is validated by comparing
the results obtained from the present DNS results for a sphere and
different ellipsoidal particles with a well-known drag correlation,
another simulation at particle scale, and theoretical results. The
two validation cases are described below.
 In the first one, we computed the drag coefficient of a sphere for
Rep 2 ½0:1  280, the present results are compared to the well-
known correlation by Schiller and Naumann [36],
24Re1 0:687
p ð1 þ 0:15 Rep Þ, in Fig. 5. The results given by the
recent DNS of Zastawny et al. [20] are also reported.
For this first validation test case, the present direct numerical
simulation applied at the scale of a spherical particle gives very
good results with respect to the well-known drag correlation of
Schiller and Naumann [36] for a large particle Reynolds range. In
Table 4
Influence of the number of mesh cells on the drag, lift on torque coefficient for w = 5/1
at Rep = 290.
Number of mesh cells CD CL CT
344,134 0.853 8.32  105 5.87  104
664,911 0.830 1.53  105 8.56  105
978,559 0.828 3.90  105 1.14  104
Fig. 5. Drag coefficient as a function of Rep. Present results: ( ); Schiller and
Naumann [36]: (–). Results of Zastawny et al. [20] are also reported ( ).
57
comparison with the recent DNS (based on a mirroring immersed
boundary method) results of Zastawny et al. [20], it is noted that
they obtained a similar trend. Nonetheless, their prediction of the
drag coefficient is less accurate to that given by the present DNS
whatever the value of the Reynolds number. The mean relative
deviation from the present DNS results with the drag coefficient
correlation is 1.5% whereas from the DNS results of Zastawny
et al. [20], it is about 17%.
 The second case is performed with three different ellipsoids
(w = 5/1, w = 5/2 and w = 5/4) in a Stokes flow. The drag and lift
coefficients results are compared to the analytical solution
given in Happel and Brenner [21] in Fig. 6. The deviations from
the theoretical results are shown in Table 5. We also reported
the DNS numerical results of Zastawny et al. [20] (drag and lift
coefficients) in Fig. 6 and Table 5 in order to compare two differ-
ent DNS approaches (body-fitted and immersed boundary
methods). Note that the case w = 5/1 was not considered by
Zastawny et al. [20].
From the second validation test case results, it can be noted that
the DNS results for the drag and lift coefficients are qualitatively in
good accordance with the theoretical results of Happel and Bren-
ner [21]. However, a slight overestimation of the present numerical
results is observed for the drag and lift results whatever the value
of the aspect ratio. The DNS, using an immersed boundary method,
by Zastawny et al. [20] underestimates the drag coefficient while
their lift coefficient is always greater than the theoretical results.
As shown in Table 5, the mean relative deviation of the present
results from the theory [21] does not exceed 5% for the drag coef-
ficient. This deviation is less than 10% for the lift coefficient what-
ever the value of the aspect ratio. It is noted that the mean
deviation increases when w ? 1 but it never exceeds 10%. In addi-
tion the maximum deviations are of the same order as the mean
deviation. The increase of the mean relative deviation as the aspect
ratio tends to unity is also noted for the results given by Zastawny
et al. [20]. However, the mean relative error is at least two times
higher for a shape close to a sphere than the one obtained from
the present DNS results.
Zastawny et al. [20] also observed this deviation from the theo-
retical expression by Happel and Brenner [21], they explained it by
the use of a finite domain. Nevertheless, the reasons for these dis-
crepancies at low Reynolds number and for large aspect ratio are
not clear since no study of the effect of the computational domain
size, or of the number of mesh cells was provided in their study. It
is surprising that the results of Zastawny et al. [20] are better for
w = 5/2 than for w = 5/4. It is also surprising that the lift coefficient
is better predicted than the drag coefficient for w = 5/2.
In order to understand these particular trends, we present in
Fig. 7, the pressure and viscous contributions to the drag and lift
coefficients computed from Eq. (6) with the present numerical
results for the three different ellipsoidal particles at Rep = 0.1. In
this figure, it is noticed that the viscous stress is the main contribu-
tion to the drag coefficient whatever the angle of incidence and
aspect ratio. This trend is completely different for the lift coeffi-
cient which is mainly driven by the pressure. Therefore, it could
be assumed that the DNS of Zastawny et al. [20] did not accurately
capture the viscous effect for w = 5/2 since better results are
obtained for the lift coefficient than for the drag coefficient. How-
ever, for w = 5/4, this explanation does not hold since their drag
coefficient results, which are mainly governed by the viscous
effects, are in better accordance with the theory than their lift coef-
ficient. Finally, it can be only concluded that their results are less
accurate than the present body-fitted computation.
To sum up, the present DNS based on a body-fitted approach
gives very satisfactory results of the forces acting on a sphere over
58 R. Ouchene et al. / Computers & Fluids 113 (2015) 53–64

Fig. 6. Drag (left) and lift (right) coefficients obtained for different ellipsoids at Rep = 0.1. First row: w = 5/1, second row: w = 5/2, third row: w = 5/4. Present results: ( );
theoretical results of Happel and Brenner [21]: (–); results of Zastawny et al. [20] are also reported ( ).

Table 5
Deviation of the computed drag and lift coefficients from the theoretical results by Happel and Brenner [21] in Stokes flow regime.

Drag coefficient Lift coefficient

Mean relative deviation (%) Maximum relative deviation (%) Mean relative deviation (%) Maximum relative deviation (%)
w = 5/1 Present simulation 2.14 2.45 4.2 4.29
w = 5/2 Present simulation 2.89 3.07 4.13 4.21
Zastawny et al. [20] 13.34 17.75 2.67 4.07
w = 5/4 Present simulation 4.04 4.13 8.04 8.65
Zastawny et al. [20] 11.31 13.11 23.68 54.57

a large Reynolds number range, as well as for ellipsoidal particles Brenner [21] for ellipsoidal particles is acceptable (never exceeds
at low Reynolds number. The mean relative deviation from the 10%). The present DNS results have also been compared to DNS
correlation of a widely used sphere drag correlation (Schiller and data obtained using a different numerical approach (mirroring
Naumann [36]) or from the theoretical results by Happel and immersed boundary method). This comparison shows that similar
 R. Ouchene et al. / Computers & Fluids 113 (2015) 53–64 59

Fig. 7. Pressure (dashed lines) and friction (continuous lines) drag (left) and lift (right) coefficients as a function of the angle of incidence a in a Stokes regime. ( ): w = 5/1;
(–): w = 5/2; ( ): w = 5/4.

Fig. 8. Comparison of the computed drag coefficient with the existing correlations for different particles Reynolds numbers. (–): Rosendahl [19]; ( ): present results;
( ): Zastawny et al. [20]; ( ): Holzer and Sommerfeld [17]. The dashed lines are referred to Rep = 10 and the continuous lines are referred to Rep = 290.

tendencies are given by both methods, nevertheless, the results
from the mirroring immersed boundary method DNS are less
accurate. The present computational methodology (domain size,
number of mesh points, Navier–Stokes solver, and computation
of the hydrodynamic force) used to compute the hydrodynamic
forces acting on spherical and non-spherical particles can be thus
considered to be valid. In the next section, results obtained outside
Stokes flow regime are shown for the ellipsoidal particles
described in Table 2.
4. Results and discussion
Simulations were performed to compute the drag CD, lift CL and
pitching torque CT coefficients for ellipsoidal particles at different
angles of incidence and for different particle Reynolds numbers.
Several aspect ratios were computed (see Table 3).
4.1. Drag coefficient
In Fig. 8, the computed drag coefficients are presented and
compared with the correlations of Holzer and Sommerfeld [17],
Rosendahl [19] and Zastawny et al. [20] given in Table 1. Concern-
ing the CD results, it has to be noted that the value of CD,a=0°, and
CD,a=90°, used for the correlation of Rosendahl [19] were computed
from the correlation of Holzer and Sommerfeld [17]. The drag
coefficient is presented as a function of the angle of incidence.
Zastawny et al. [20] results are not presented for the case w = 5/1
since they did not consider this aspect ratio in their study.

Table 6
Deviation of the computed drag coefficients from the present results at Rep = 10.

Mean relative deviation (%) Maximum relative deviation (%)

Holzer and Sommerfeld [17] Rosendahl [19] Zastawny et al. [20] Holzer and Sommerfeld [17] Rosendahl [19] Zastawny et al. [20]
w = 5/1 6.73 3.41 12.49 7.3
w = 5/2 3.18 4.7 11.69 6.82 6.98 15.86
w = 5/4 15.35 15.73 21.32 16.78 16.78 21.54
60 R. Ouchene et al. / Computers & Fluids 113 (2015) 53–64
Table 7
Deviation of the computed drag coefficients from the present results at Rep = 290.
Mean relative deviation (%)
Holzer and Sommerfeld [17] Rosendahl [19]
w = 5/1 116.1 83.73
w = 5/2 71.7 59.01
w = 5/4 18.99 17.50
Qualitatively, the present results are in good agreement with
those issuing from the literature since a similar trend is observed.
The first main observation lies in the decrease of the drag values
when the Reynolds number increases whatever the aspect ratio.
The second one is linked to the evolution of the drag value as a
function of incidence angle for a given Reynolds number. The drag
coefficient increases for increasing incidence angle whatever the
values of aspect ratio. An inflection point is noted around a = 45°.
Note that this inflection point is not captured by the correlation
of Holzer and Sommerfeld [17]. For a shape near to a sphere
(w = 5/4), the drag coefficient increase is attenuated whatever the
Reynolds value. In other words, for this aspect ratio near to a
sphere, the drag coefficient does not significantly depend on the
incidence angle as expected.
Concerning the quantitative comparison of the results given, it
is noticed that there is a great disparity in the results plotted.
The difference between our drag coefficient values and those
obtained from the correlations by Holzer and Sommerfeld [17],
Rosendahl [19], and Zastawny et al. [20] are reported in Tables 6
(Rep = 10) and 7 (Rep = 290) for each aspect ratio. These deviations
were averaged on the whole range of angles of incidence. At low
Reynolds number (Rep = 10, Table 6), we clearly note that the most
important deviation from our results is obtained for a shape close
to a sphere (w = 5/4) whatever the correlations used. It is noticed
that the mean relative deviation is 20% for the results of Zastawny
et al. [20] and around 15% with the results of Holzer and
Sommerfeld [17] and of Rosendahl [19]. A similar trend is noted
for the calculated maximum relative deviation. The similarities
noted for the correlations proposed by Holzer and Sommerfeld
[17] and Rosendahl [19] were expected since the values of CD,a=0°,
Fig. 9. Normalized drag coefficient, CD/CD,a=90°, as a function of the angle of incidence a. (
Maximum relative deviation (%)
Zastawny et al. [20] Holzer and Sommerfeld [17] Rosendahl [19] Zastawny et al. [20]
140.7 99.22
22.08 84.59 79.46 37.56
7.19 25.02 25.02 10.07
and CD,a=90° used for the correlation of Rosendahl [19] were com-
puted from the correlation of Holzer and Sommerfeld [17]. It is
therefore not surprising to have similar deviations.
At higher Reynolds number (Table 7), we note that the devia-
tion previously noted at low Reynolds number is lower in the case
where the shape is close to a sphere. However, when the aspect
ratio increases, the deviation does not decrease in contrast to the
low Reynolds number case. The deviation is much larger for
w = 5/2 at Rep = 290 while this was noticed for w = 5/4 at
Rep = 10. Moreover, important deviations from the present results
are remarked when the correlation of Holzer and Sommerfeld
[17] or of Rosendahl [19] is used. Again, similar mean and maxi-
mum deviations are obtained. The deviation can be found to be
more than 100% using the correlation from Holzer and Sommerfeld
[17] whereas the deviation is slightly smaller than 100% (but
remains around 100%) when the correlation of Rosendahl [19] is
used for the highest aspect ratio (w = 5/1). If the correlation of
Holzer and Sommerfeld [17] does not provide an accurate predic-
tion of the drag coefficient, the prediction by the correlation of
Rosendahl [19] will be also inaccurate. These inconsistencies with
respect to our results are certainly due to the fact that their
correlation was developed for arbitrary shaped particles from
experimental and numerical data obtained for various forms like
isometric particles, cuboids, cylinders or disks. This correlation is
not dedicated to ellipsoidal particles, a lack of accuracy is thus
expected. Holzer and Sommerfeld [17] found that the accuracy of
their correlation depends on the shape of the particles (the mean
relative deviation between the correlation and experimental data
is about 14%, we have found a similar deviation at Rep = 10). In
comparison to the results from the correlations of Holzer and
): Rep = 10; ( ); Rep = 60; ( ): Rep = 240; ( ): Rep = 290.
 R. Ouchene et al. / Computers & Fluids 113 (2015) 53–64 61

Fig. 10. Pressure (dashed lines) and friction (continuous lines) drag coefficients normalized by CD,a=90° as a function of the angle of incidence a for different aspect ratios.
( ): w = 5/1; ( ); w = 5/2; ( ): w = 5/4. Left: Rep = 10. Right: Rep = 290.

Sommerfeld [17] or Rosendhal [19], the accuracy of the correlation
provided by Zastawny et al. [20] is better at Rep = 290. At Rep = 10,
the mean deviation given by this latter correlation is higher than
those provided by the two other correlations. Moreover, the mag-
nitude of the deviation depends on the particle aspect ratio. More
accurate results are obtained for w = 5/2 at Rep = 10 while a better
prediction is found for w = 5/4 at Rep = 290. However, the mean
relative error remains high. Unfortunately, we have no clear expla-
nation for these particular trends.
In order to better characterize the evolution of drag coefficient
as a function of the Reynolds number, aspect ratio, and angle of
incidence, we have plotted in Fig. 9 the computed normalized drag
coefficients CD/CD,a=90° as a function of a for different aspect ratio
and for several particle Reynolds numbers.
In Fig. 9, it is clearly seen that CD decreases when the Reynolds
number increases as observed before. It is also interesting to note
that when Rep is higher than 240 the ratio CD/CD,a=90° does not sig-
nificantly evolves as a function of the Reynolds number. As already
noticed, the incidence angle has a less important influence when
the aspect ratio decreases.
To more thoroughly examine the drag coefficient, the pressure
and viscous drag contributions (the total drag force is the sum of
pressure and friction drag forces) normalized by CD,a=90° are given
in Fig. 10 for two values of Rep (10 and 290). The pressure drag
coefficient is seen to be significantly influenced by the angle of
incidence while the friction drag coefficient does not greatly vary
as a function of the particle orientation. Therefore, the evolution
of total drag coefficient as a function of a is mainly due to the pres-
sure drag coefficient contribution. This trend is observed at both
particle Reynolds numbers. It should be noted, that the friction is
the main contribution to the drag at Rep = 10, and whatever the
angle of incidence. At Rep = 290, this contribution is higher than
that of pressure at low angle of incidence, then the pressure contri-
bution dominates. Finally, from this figure, it can be seen that the
friction contribution is less influenced by the particle aspect ratio
than the pressure contribution. For instance, at Rep = 290, the pres-
sure contribution for w = 5/4 is 4 times higher than that obtained
for w = 5/2 while the friction contributions are almost identical.
To sum up, it has been shown that the present computation can
accurately predict the drag coefficient of a spherical particle from
Stokes flow up to Rep = 290. A very good accordance between the
present DNS computations and the theoretical results in Stokes
flow regime has been also observed for the drag coefficient of ellip-
soidal particles. For higher Reynolds numbers, the comparison

Fig. 11. Comparison of the computed lift coefficient with the existing correlations for different particle Reynolds numbers. ( ): present results; ( ): Zastawny et al.
[20]; ( ): Hoerner [23]. The dashed lines are referred to Rep = 10 and the continuous lines are referred to Rep = 290.
62 R. Ouchene et al. / Computers & Fluids 113 (2015) 53–64
Table 8
Deviation of the computed lift coefficients from the present results at Rep = 10.
Mean relative deviation (%)
Hoerner [23]
w = 5/1 84.90
w = 5/2 134.86
w = 5/4 601.16
with existing drag coefficient correlations (derived from numerical
and/or experimental data) shows that they do not accurately
predict the evolution of the drag coefficient as a function of the
Reynolds, aspect ratio, and angle of incidence. From a physical
point of view, it has been shown the friction part is not greatly
influenced by the angle of incidence in contrary to the pressure
contribution.
4.2. Lift coefficient
The lift coefficient CL at Rep = 10 and Rep = 290 is shown in
Fig. 11 as a function of the incidence angle. When w = 5/2 or 5/1,
the lift coefficient is higher for Rep = 10 than for Rep = 290, as
already noticed for CD. This trend is also noted for w = 5/4 when
a < 70°, but for higher angle of incidence the lift coefficient at
Rep = 290 does not tend toward zero. The lift coefficient is thus
greater than that at Rep = 10 for this range of incidence angle. This
nonzero value of the lift force at Rep = 290 is induced by the non-
axisymmetry of the wake. Johnson and Patel [32] also observed
this effect for a sphere (w = 1) when Rep > 210. In addition, except
for w = 5/4 at Rep = 290 and a > 70°, the computed lift coefficient
is seen to decrease as w tends toward unity (sphere). This was
expected since in the present study, the lift force is only induced
by the particle shape. Finally, it has to be noted that, whatever
the aspect ratio and particle Reynolds number, the lift coefficient
is maximum near an incident angle of a = 45°.
These trends are similar to that obtained by Zastawny et al. [20].
In Fig. 11, the prediction of the lift coefficients given by the corre-
lations derived by Hoerner [23] and Zastawny et al. [20] are also
plotted. For sake of clarity, the curve predicted by Hoerner [23] is
cut for w = 5/4. From this comparison, it is seen that the evolution
of CL predicted by Hoerner [23] does not coincide with the present
results nor with those obtained by Zastawny et al. [20]. Moreover,
the maximum lift coefficient predicted by Hoerner [23] is located
at a = 60° while Zastawny et al. [20] and the present results show
that it is located near a = 45°. It can be definitively concluded that
the lift coefficient cannot be only a function of incidence angle a
and drag coefficient CD as suggested by Hoerner [23]. A better
prediction of the lift coefficient is obtained with the correlation
presented by Zastawny et al. [20] as can be seen in Fig. 11. In Tables
8 and 9, the deviation between the prediction by Hoerner [23] or
Zastawny et al. [20] and the present results are given. It confirms
that the correlation by Hoerner [23] is not able to predict the lift
force acting on ellipsoidal particles since the mean deviation
ranges from 35% to 600%. The maximum mean relative deviation
for the results of Zastawny et al. [20] is about 27%. It is noted that
this deviation error increases with increasing particle Reynolds
numbers and/or decreasing aspect ratio in contrast to the results
for the drag coefficients. It is also observed that the maximum
deviation is obtained for a shape near a sphere whatever the value
of the Reynolds number. It has to be noted that in their simula-
tions, the number of mesh cells depends on the Reynolds number
considered. This number was varied from 8 to 12 cells on the diam-
eter of the volume equivalent sphere, it can be thus assumed that
this number is not enough high to capture the boundary layer at
high Reynolds number (keeping in mind that the boundary layer
thickness varies approximately as Rep1/2).
Maximum relative deviation (%)
Zastawny et al. [20] Hoerner [23] Zastawny et al. [20]
173.23
8.70 263.23 11.72
19.94 978.4 23.59
In order to better understand the trend observed for the lift
coefficient, we computed the contribution to the lift coefficient
induced by the pressure and the viscous stress. The pressure and
viscous contributions are plotted in Fig. 12 for Rep = 10 and
Rep = 290. It can be first noticed that the incidence angle has a weak
effect on the pressure and viscous stress lift coefficients. In
addition, the pressure is the main contribution to the lift force
whatever the particle Reynolds number and aspect ratio. Nonethe-
less, there is an important difference between these two particle
Reynolds number cases. At Rep = 10, the viscous contribution is
always negative while this contribution is positive at Rep = 290
and very close to zero. The pressure plays definitely a major role
in the lift coefficient. It is interesting to note that Cherukat et al.
[37] also found, from numerical simulations, that the viscous stress
contribution becomes negative for a spherical solid particle
embedded in a linear shear flow when the particle Reynolds num-
ber is greater than unity. A similar trend was observed by Legendre
and Magnaudet [38] for a spherical bubble in a linear shear flow
when the bubble Reynolds number is greater than one. In the pres-
ent study, this negative contribution is induced by the particle
shape since the ellipsoidal particles considered are embedded in
a uniform flow.
4.3. Pitching torque coefficient
In Fig. 13, the computed pitching torque coefficient is plotted as
a function of a for two aspect ratios and two particle Reynolds
numbers (Rep = 10 and Rep = 290). It is noticed that the pitching
torque decreases as the particle Reynolds number increases. As
expected, this torque also decreases when w tends toward unity.
These data are compared to the results given by the correlation
derived by Zastawny et al. [20]. It is noted that a better accordance
is obtained for w = 5/2 than for w = 5/4. The deviation between the
prediction by the correlation given by Zastawny et al. [20] and the
present results are shown in Table 10. It should be kept in mind
that the pitching torque is obtained by the combination of drag
and lift forces acting on the center of pressure. Therefore, as
expected, the differences between the present results and those
given by the correlation of Zastawny et al. [20] are of the same
order to those already noticed for the drag and lift coefficients.
For the pitching torque, the mean relative deviation is always
lower than 25%. The deviation increases with increasing aspects
ratios. It also increases with increasing particle Reynolds numbers
for w = 5/2 ellipsoids. On contrary, it is seen that the mean devia-
tion does vary as a function of the particle Reynolds number for
w = 5/4. Again, these trends are different from those noticed for
the drag and lift coefficients and cannot be easily explained.
5. Conclusion
The first objective of the present study was to provide, using a
body-fitted DNS, the drag, lift and pitching torque coefficients for
three different ellipsoidal particles and for Reynolds number rang-
ing from 0.1 to 290, and to compare these results to some existing
correlations (Holzer and Sommerfeld [17], Rosendahl [19] and
Zastawny et al. [20]). From this comparison, it was shown that
 R. Ouchene et al. / Computers & Fluids 113 (2015) 53–64 63

Table 9
Deviation of the computed lift coefficients from the present results at Rep = 290.

Mean relative deviation (%) Maximum relative deviation (%)

Hoerner [23] Zastawny et al. [20] Hoerner [23] Zastawny et al. [20]
w = 5/1 38.83 86.21
w = 5/2 36.64 21.34 85.15 30.25
w = 5/4 54 27.25 89.50 67.82

Fig. 12. Pressure (dashed lines) and friction (continuous lines) lift coefficients, normalized by the total lift coefficient, as a function of the angle of incidence a for different
aspect ratios. ( ): w = 5/1; (–); w = 5/2; ( ): w = 5/4. Left: Rep = 10. Right: Rep = 290.

Fig. 13. Comparison of the torque coefficients for different particles Reynolds numbers. Present results: ( ); results of Zastawny et al. [20]: ( ). The dashed lines are
referred to Rep = 10 and the continuous lines are referred to Rep = 290.

Table 10
as already noted by Zastawny et al. [20], that the lift coefficient
Deviation of the computed drag, lift, and pitching torque coefficients given by
Zastawny et al. [20] from the present results at Rep = 10 and 290. cannot be modeled as a function of only the angle of incidence
and the drag coefficient. The correlation by Zastawny et al. [20]
Mean relative deviation (%) Maximum relative deviation (%)
gives more accurate results, nevertheless, it was observed that
Rep = 10 Rep = 290 Rep = 10 Rep = 290 the mean relative deviation from our DNS results increases with
w = 5/2 2.24 15.17 4.35 20.28 increasing Reynolds number and ranges from 8% to 27%. As for
w = 5/4 24.75 23.43 26.54 30.51 the drag coefficient, they provide a correlation which is only valid
for two ellipsoid aspect ratios. The lift coefficient was also divided
into a pressure part and a viscous part. From this decomposition, it
these correlations cannot accurately predict the drag coefficient for has been noticed that the pressure is the main contribution to the
the three aspect ratios and over the entire particle Reynolds lift force whatever the aspect ratio, angle of incidence, and Rey-
number range considered. The correlation by Zastawny et al. [20] nolds number. Moreover, as also observed for a spherical particle
provides the best prediction, nevertheless, the deviation from our in a linear shear flow, the viscous part can be negative (i.e. in the
full DNS results can be of the order of 20% and their correlation opposite direction to the total lift force). Finally, we have presented
for ellipsoidal particles is valid for only two aspect ratios. From the pitching torque coefficient and compared it to the correlation
the DNS results, it was noted the viscous stress and pressure con- derived by Zastawny et al. [20]. As for the drag and lift coefficients,
tributions to the drag force vary with the Reynolds number, aspect the mean deviation from our DNS results ranges from 2% to 25%.
ratio, and angle of incidence. Generally, the viscous part is the main Therefore, the study by Zastawny et al. [20] is the only one to pro-
contribution. In addition, it was noted that the functional form of vide a complete set of correlations for ellipsoidal particles. How-
the model proposed by Rosendahl [19] to express the drag coeffi- ever, these correlations are valid for only two aspect ratios. This
cient as a function of the angle of incidence is acceptable. Concern- is one of the major drawbacks of their models. The second one is
ing the lift force, the comparison with the correlations by Hoerner that we have found that the DNS used by Zastawny et al. [20] to
[23] or Zastawny et al. [20] have shown that the correlation by derive their models does not provide an accurate prediction of
Hoerner [23] is not able to predict the lift coefficient. This means, the hydrodynamic force acting on ellipsoidal particle at low
64 R. Ouchene et al. / Computers & Fluids 113 (2015) 53–64
Reynolds number. In addition, some significant deviations have
been also noted between the correlations derived by Zastawny
et al. [20] and our DNS results outside Stokes flow regime. For all
the reasons given above, we deeply believe that further efforts
are definitively necessary to develop a complete set of correlations
for ellipsoidal particles outside Stokes flow regime. These
correlations should be function of the angle of incidence, particle
Reynolds number, and aspect ratio of particles in order to
accurately model the forces and torque acting on non-spherical
particles. We hope that the present existing database on drag, lift
and torque coefficients will help to the future developments of
such correlations.
Acknowledgments
This work received funding by ANR-PLAYER; their support is
gratefully acknowledged. We would like to thank our partners
IMFT and I2M laboratories.
References
[1] Marchioli C, Soldati A, Kuerten JGM, Arcen B, Taniere A, Goldensoph G, et al.
Statistics of particle dispersion in direct numerical simulations of wall-
bounded turbulence. Results of an international collaborative benchmark test.
Int J Multiphase Flow 2008;34:879–93.
[2] Arcen B, Taniere A, Zaichik L. Assessment of a statistical model for the
transport of discrete particles in a turbulent channel flow. Int J Multiphase
Flow 2008;34(4):419–26.
[3] Mandø M, Rosendahl L. On the motion of non-spherical particles at high
Reynolds number. Powder Technol 2010;202(25):1–13.
[4] Lain S, Sommerfeld M. Kinetic simulations of non-spherical particle response
behavior. In: 8th International conference on multiphase flow; 2013.
[5] Njobuenwu DO, Fairweather M. Effect of shape on inertial particle dynamics in
a channel flow. Flow, Turbul Combust 2014;92:83–101.
[6] van Wachem B, Zastawny M, Zhao F, Mallouppas G. Modelling of gas–solid
turbulent channel flow with non-spherical particles with large Stokes
numbers. Int J Multiphase Flow 2015;68:80–92.
[7] Clift R, Grace JR, Weber ME. Bubbles, drops and particles. New York: Academic
Press; 1978.
[8] Allen T. Particle size measurement. Chapman & Hall; 1981.
[9] Mandø M. Turbulence modulation by non-spherical particles. Aalborg
University; 2009.
[10] Li T, Li S, Zhao J, Lu P, Meng L. Sphericities of non-spherical objects.
Particuology 2012;10(1):97–104.
[11] Brix J, Jensen PA, Jensen AD. Modeling char conversion under suspension fired
conditions in O2/N2 and O2/CO2 atmospheres. Fuel 2011;90:2224–39.
[12] Powers MC. A new roundness scale for sedimentary particles. J Sediment
Petrol 1953;23:117–9.
[13] Chhabra R, Agarwal L, Sinha N. Drag on non-spherical particles: an evaluation
of available methods. Powder Technol 1999;101:288–95.
[14] Ganser GH. A rational approach to drag prediction of spherical and non-
spherical particles. Powder Technol 1993;77(2):143–52.
[15] Tran-Cong, Gay M, Michaelides E. Drag coefficients of irregularly shaped
particles. Powder Technol 2004;139:21–32.
[16] Yow HN, Salman A, Pitt MJ. Drag correlations for particles of regular shape.
Powder Technol 2005;16(4):363–72.
[17] Holzer A, Sommerfeld M. New simple correlation formula for the drag
coefficient of non spherical particles. Powder Technol 2008;184:361–5.
[18] Wadell H. The coefficient of resistance as a function of Reynolds number for
solids of various shapes. J Franklin Inst 1934;217:459–90.
[19] Rosendahl L. Using a multi-parameter particle shape description to predict the
motion of non-spherical particle shapes in swirling flow. Appl Math Model
2000;24:11–25.
[20] Zastawny M, Mallouppas G, Zhao F, van Wachem B. Derivation of drag and lift
force and torque coefficients for non-spherical particles in flows. Int J
Multiphase Flow 2012;101:288–95.
[21] Happel J, Brenner H. Low Reynolds number hydrodynamics. Englewood
Cliffs: Prentice-Hall; 1965.
[22] Yin CG, Rosendahl L, Kaer SK, Sorensen H. Modelling the motion of cylindrical
particles in a non uniform flow. Chem Eng Sci 2003;58(15):3489–98.
[23] Hoerner SF. Fluid dynamic drag. Hoerner Fluid Dynamics; 1965.
[24] Zhang H, Ahmadi G, Fan F-G, McLaughlin JB. Ellipsoidal particles transport and
deposition in turbulent channel flows. Int J Multiphase Flow
2001;27(6):971–1009.
[25] Mortensen PH, Andersson HI, Gillissen JJJ, Boersma BJ. Dynamics of prolate
ellipsoidal particles in a turbulent channel flow. Phys Fluids 2008;20:093302.
[26] Marchioli C, Fantoni M, Soldati A. Orientation, distribution and deposition of
elongated, inertial fibers in turbulent channel flow. Phys Fluids
2010;22:033301.
[27] Tian L, Ahmadi G. Fiber transport and deposition in human upper
tracheobronchial airways. J Aerosol Sci 2013;60:1–20.
[28] Marchioli C, Soldati A. Rotation statistics of fibers in wall shear turbulence.
Acta Mech 2013;224:2311–29.
[29] Zhao L, Marchioli C, Andersson HI. Slip velocity of rigid fibers in turbulent
channel flow. Phys Fluids 2014;26:063302.
[30] Jeffery GB. The motion of ellipsoidal particles immersed in a viscous fluid. Proc
R Soc 1922;102:161–79.
[31] Lázaro BJ, Lasheras JC. Particle dispersion in a turbulent, plane, free shear layer.
Phys Fluids 1989;A1(6):1035–44.
[32] Johnson TA, Patel VC. Flow past a sphere up to a Reynolds number of 300. J
Fluid Mech 1999;378:19–70.
[33] Leonard BP. A stable and accurate convective modeling procedure based on
quadratic upstream interpolation. Comput Meth Appl Mech Eng
1979;19:59–98.
[34] Hölzer A, Sommerfeld M. Lattice Boltzmann simulations to determine drag, lift
and torque acting on non-spherical particles. Comput Fluids 2009;38:572–89.
[35] Richter A, Nikrityuk PA. Drag forces and heat transfer coefficients for spherical,
cuboidal and ellipsoidal particles in cross flow at sub-critical Reynolds
numbers. Int J Heat Mass Transfer 2012;55:1343–54.
[36] Schiller L, Naumann A. Fundamental calculations in gravitational processing. Z
Vereines Deutscher Ingenieure 1933;77:318–20.
[37] Cherukat P, McLaughlin JB, Dandy DS. A computational study of the inertial lift
on a sphere in a linear shear flow field. Int J Multiphase Flow
1999;25(1):15–33.
[38] Legendre D, Magnaudet J. The lift force on a spherical bubble in a viscous linear
shear flow. J Fluid Mech 1998;368:81–126.