Powder Technology 330 (2018) 152–163

Contents lists available at ScienceDirect

Powder Technology

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

Evaluation of drag models for particles and powders with non-uniform

size and shape
H. Gerhardter a,⁎, R. Prieler a, B. Mayr a, M. Knoll a, M. Mühlböck b, P. Tomazic b, C. Hochenauer a
a
Graz University of Technology, Institute of Thermal Engineering, Inffeldgasse 25/b, 8010 Graz, Austria
b
M. Swarovski Gesellschaft m.b.H., Industriestraße 10, 3300 Amstetten, Austria

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

Article history: One of the key issues with the application of drag models for non-spherical particles is the incorporation of dif-
Received 19 October 2017 ferently shaped particles. Under real world conditions, every single grain in a powder or granule has a unique ge-
Received in revised form 1 February 2018 ometry which strongly affects its drag coefficient. To incorporate the effects of different particle shape factors on
Accepted 12 February 2018
their trajectories in the flow field, a drag analysis of crushed coal slag particles with sizes between 400 μm and
Available online 17 February 2018
850 μm was performed within this work. The particle shapes were examined using brightfield microscopy and
Keywords:
an optical particle analyzer. Also, the drop velocities of particles in still air were measured using a “Sony DSC-
Particle drag coefficient RX100 IV” standard digital camera at a frame rate of 1000 pictures per second instead of more costly special-
Varying particle shape purpose high speed camera equipment. A novel particle classification was introduced based on the obtained
Polydisperse powders data. This data was also used as an input for three different drag models from literature. In order to evaluate
Multiphase flow the practicality of each drag model, a simple test rig in form of an air tunnel separator was built. It basically
Computational fluid dynamics consisted of a box where freely falling slag particles were deflected by a free stream of air. The particle mass
spreading at the bottom of the air flow separator was compared to numerical results. Numerical simulations of
the setup were performed using Computational Fluid Dynamics (CFD) and a numerically efficient Euler-
Lagrange approach in order to calculate the particle trajectories. It was concluded that the simple particle classi-
fication method provided satisfying calculated results. A special drag formulation for crushed coal slag particles
was developed on the basis of two drag models from literature. The customized drag model improved the calcu-
lated results significantly. It can be added to nearly every commercial CFD code on the market without significant
increase on computational cost and the proposed procedure also appeared to be well suited for the determination
of drag model characteristics of other particulate materials in many fields of research.
© 2018 Elsevier B.V. All rights reserved.

1. Introduction which seems not acceptable for industrial applications. In general it

In order to describe the interaction between disperse particles and
the flow of a continuous phase, numerous models for the calculation
of particle drag coefficients were proposed in literature. The applicabil-
ity of each model mostly depends on the particle Reynolds number and
its shape. This research is restricted to non-spherical particles. Experi-
ments and numerical calculations within this work were carried out
for particle Reynolds numbers in the range ReP b 1000.
An evaluation of most commonly used models was performed by
Chhabra etal. [1]. They compared the predictions of different drag
models at N1900 data points and concluded, that the drag model pro-
posed by Ganser [2] provided the most accurate results but still incorpo-
rated an average error of 16.3% and a maximum error of almost 181%
⁎ Corresponding author.
E-mail addresses: hannes.gerhardter@tugraz.at (H. Gerhardter),
rene.prieler@tugraz.at (R. Prieler), bernhard.mayr@tugraz.at (B. Mayr),
mario.knoll@tugraz.at (M. Knoll), christoph.hochenauer@tugraz.at (C. Hochenauer).
was noted that the error increased with lower particle sphericity, for ex-
ample needles and disks, while the results were acceptable for arbitrary
shaped, isometric particles. Based on the extensive experimental data of
Schulz etal. [3], one of the first drag models which incorporated the par-
ticle shape was developed by Swamee etal. [4]. Here, the drag coefficient
was calculated in dependence of the particle Reynolds number and the
Corey shape factor which is a function of three particle main dimen-
sions. This basic idea was adopted by a number of researchers and
more recently developed drag models are still based on it. A very mod-
ern model was proposed by Hölzer and Sommerfeld [5]. Besides the par-
ticle Reynolds number and the particle sphericity, the drag formulation
was based on the lengthwise and crosswise sphericity. In their work, the
crosswise sphericity is 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. The lengthwise sphericity
is the ratio between the cross-sectional area of the volume equivalent
sphere and the difference between half the surface area and the mean
projected longitudinal cross-sectional area of the considered particle.

https://doi.org/10.1016/j.powtec.2018.02.036
0032-5910/© 2018 Elsevier B.V. All rights reserved.
 H. Gerhardter et al. / Powder Technology 330 (2018) 152–163
The mean relative deviation between the correlation of Hölzer and
Sommerfeld and 2061 data points from literature was reported to be
at 14.1% [5]. According to Bagheri and Bonadonna [6], the characteriza-
tion of particles based on length measurements is easier, more precise
and less time consuming in comparison to the determination of their
sphericity. They developed a drag formulation where the particle
shape was incorporated by two shape factors corresponding to Newtons
and Stokes regime in the flow [7]. Both shape factors are a function of
the equal volume diameter, the particle flatness and the particle elonga-
tion. They found that the average error of the novel equation was at ap-
proximately 10%.
One characteristic, that the drag models proposed by Hölzer and
Sommerfeld etal. [5] and Bagheri and Bonadonna [7] have in common
is, that not only the particle shape, but also the particle orientation is
taken into account. The orientation of freely falling disks was in-
vestigated by Willmarth etal. [8]. They concluded that particles
orienttheir largest projection area perpendicular to the flow up to a
critical particle Reynolds number which, in the case of disks, was at
ReP ≈ 100. The same effect was observed by Carranza etal. [9] and fur-
ther, they investigated the influence of tumbling motions and the
angle of attack on the particle drag coefficient. In the case of disks, tum-
bling motions of the particle had a major impact on the drag coefficient.
Richter and Nikrityuk [10] performed numerical calculations in the
range 10 b ReP b 200 in order to verify the correlation between particle
orientation, drag, lift and torque coefficients and the particle Nusselt
number and provide closure relations for cubic and ellipsoidal particles.
Their results showed that at different angles of attack the drag coeffi-
cient of ellipsoidal particles changes up to 80% and up to 23% for cubical
particles.
The precise prediction of the drag coefficient of nonspherical parti-
cles is of major importance in many fields of research. Especially when
modelling the gasification and combustion of particulate fuels, the influ-
ence of particle size and shape has a large impact on the calculated com-
bustion characteristics, pollutant formation and aerosol emissions. The
trajectories of biomass particles in a 1 MW furnace were numerically in-
vestigated by Jones etal. [11]. They concluded in application of an Euler-
Lagrangian approach that non-spherical particles with a sphericity of
0.1 and a diameter of 750 μm penetrate to the fly ash while spherical
particles of the same size would end up in the bottom ash of the boiler.
The impact of particle shape on the combustion characteristics of co-
fired coal and biomass particles was investigated by Gubba etal. [12].
They used an Euler-Lagrangian multiphase model in order to calculate
the particle trajectories, devolatilization and char combustion and also
compared the numerical to experimental results of an industrial scale
test facility. Their numerical work involved two test cases. In the first
case, the particles were assumed to be spherical while in the second
case, 50% of the particles had a cylindrical shape in order to represent
the fibrous texture of ground biomass. The experimentally determined
flame shape was in close agreement with the second test case and it
was concluded that the combustion characteristics of biomass particles
were highly influenced by the particle geometry. A comprehensive
study on the combustion of biomass particles was carried out by Yin
etal. [13]. In the first part of their research, they developed a model
which allowed not only the calculation of drag, but also of other forces
like torque, lift and a virtual mass force for cylindrical biomass particles
[14]. To validate the model, the rotation and translation of a PVC particle
during sedimentation in water was investigated by numerical calcula-
tions and experimental work. The results showed good correlation
and it was therefore concluded that the validated model can be ex-
tended for calculations on the combustion characteristics of biomass
particles. In a further work [15], Yin etal. performed a detailed numerical
analysis on the co-firing of biomass particles and natural gas in applica-
tion of their specialized model and an Euler-Lagrangian tracking algo-
rithm. Calculations were performed on a 10 m long combustion
chamber, 22 shape and size groups of particles were incorporated and
the results were compared to a second case where particles were
153
assumed to be totally spherical. The results brought up that, due to the
lower surface-to-volume ratio, spherical particles heated up slower.
The devolatilization, pyrolysis and combustion characteristics of parti-
cles showed significant differences and thus, the temperature dis-
tribution as well as the mass fractions of volatiles and oxygen within
the burner chamber were strongly influenced by the shape of fuel par-
ticles. Only the tracks of the smallest cylindrical particles with a length
of 1.3 mm, a diameter of 0.2 mm and an initial density of 139 kg/m3
were almost similar compared to those of spherical particles with the
same initial mass. While the work of Yin etal. [13–15] was focused on
the motion of particles, numerous other researchers [16–19] focused
on the physical and chemical processes within the particle and also con-
cluded, that the particle shape also should be taken into account by this
field of research. The influence of particle porosity on heat and fluid flow
past and through porous particles was investigated by Wittig etal. [20].
They performed numerical calculations using agglomerates of differ-
ently shaped grains and an immersed boundary method in order to re-
solve porous particles. The particle porosity, as well as the particle
surface ratio were varied in order to identify the impact on drag and
heat transfer coefficients. Based on the numerical data, new correlations
for the drag coefficient and the Nusselt number of porous particles were
proposed.
In summary, the numerical and experimental methods are highly
optimized and have been in the focus of researchers for a long time.
On the other hand, the proposed methods lack of simple feasibility in in-
dustrial applications. Tamura etal. [21] investigated the grinding and
combustion characteristics of woody biomass. They found that the
resin content in the biomass had large impact on the stability of
thegrinding process. Without any major research effort, none of the
above-mentioned methods can determine the effect of quality fluctua-
tions of biomass on combustion. A highly simplified method for drag
calculations was proposed by Dunnu etal. [22–24]. They investigated
the aerodynamic lift velocity of solid recovered fuel (SRF) particles
and found, that satisfying results were achieved if a constant particle
drag coefficient had been assumed. The validity of this assumption
was limited to Newtons flow regime. The lift velocities of different SRF
particles were calculated and experimentally validated, the closest cor-
relation between experiments and calculated lift velocities was
achieved with a drag coefficient cD = 1.5.
The number of applications, where particles are in Newtons flow re-
gime and a constant drag coefficient is a valid approximation, is limited.
Hence, a simple, time- and cost-efficient method for the determination
of the drag coefficient of crushed coal slag particles was presented
within this paper. The influence of the particle geometry on the aerody-
namic properties of the particles and the applicability of different drag
models in combination with this particle type were also investigated.
2. Drag models
The development of novel drag correlations was not the main focus
of this research but instead, three different drag models from literature
were evaluated. The drag coefficient cD is a dimensionless quantity for
drag force FD calculations and is given by Eq.(1).
2∙F D
cD ¼ ¼ f ðRe; GeometryÞ ð1Þ
ρ∙u2rel ∙A
2
Here, ρ is the fluid density, A ¼ deq;V π=4 is the cross-sectional area of
! !
the volume-equivalent sphere and urel ¼ j u p − u j is the magnitude of
relative velocity between the particle and the fluid. Within this work,
the particle Reynolds number is calculated via ReP = deq, V ∙ c ∙ ρ/μ
where the equal-volume diameter deq, V is the characteristic length
scale and μ is the dynamic fluid viscosity. As stated earlier, the drag coef-
ficient is a function of the particle Reynolds number and the geometric
properties of the particle. Depending on the selected drag model, the
154 H. Gerhardter et al. / Powder Technology 330 (2018) 152–163
influence of the particle geometry is taken into account by different
shape factors. A very common and widely used model for drag calcula-
tions was provided by Haider and Levenspiel [25]. The particle drag coef-
ficient was calculated via Eq.(2) and the coefficients A, B, C and D were
given in [25] as a function of sphericity as the governing particle shape
descriptor in this model.
24
 C
cD;HL ¼ 1 þ A∙ ReB þ   ð2Þ
ReP D
1þ
ReP
The sphericity Ψ is the ratio between the surface area of the equal
volume sphere and the surface area of the non-spherical particle. Spher-
ical particles have a sphericity Ψ = 1 while the value is close to zero for
highly apsherical particles. The model of Haider and Levenspiel is valid
for ReP b 2.6 ∙ 105 and provides reasonable results for Ψ N 0.67 [25].
The second model for drag force calculations within this work was
developed by Richter and Nikrityuk [26] on the basis of numerical calcu-
lations. The influence of the particle geometry on the drag coefficient,
given by Eq.(3), was incorporated by the particle length parallel to the
flow direction S normalized with the equal volume sphere diameter
deq, V. Within this work, this proportion was referred to as the size
ratio SR = S/deq, V.
20 6:9
cD;RN ¼ 0:21 þ SR0:58 þ pffiffiffiffiffiffiffiffiffi ∙SR−1:4
ReP ReP
The model was able to predict the drag coefficient of spherical, ellip-
soidal and cubic particles with a maximum relative deviation of 8.58%
compared to numerical calculations in the range 10 ≤ ReP ≤ 250 [26].
The third drag model, that was used within this work, has been pro-
posed by Kishore and Gu [27]. On the basis of numerical calculations, the
drag law given by Eq.(4) was developed.
24∙AR0;49

cD;KG ¼ ∙ 1; 05 þ 0; 152∙ ReP 0;687 ∙AR0;671
ReP
The model was specially designed for ellipsoidal particles. The parti-
cle geometry was taken into account by the aspect ratio AR which is the
ratio between the equatorial and polar radii of the particle [27]. The fact
that particles orient themselves with their largest projection area nor-
mal to the flow [8,9] must be considered by the calculation of the aspect
ratio. The model is valid in the range 1 ≤ ReP ≤ 200 and the authors con-
sidered particles with aspect ratios 0.25 ≤ AR ≤ 2.5, but no strict limits of
the particle aspect ratio were given [27]. The drag model proposed by
Haider and Levenspiel [25], as well as the drag work of Richter and
Nikrityuk [26] represent more general drag formulations while the
drag model of Kishore and Gu [27] was primarily tested on ellipsoidal
particles. Due to the fact that the ellipsoid was proposed as a simplifica-
tion for a wide range of particle shapes [28], the application of a special-
ized drag formulation with caution to the input parameters still
appeared as a valid approach.
3. Experimental setup
In basic research, precise measurements with high resolution are
often mandatory. For example, Bagheri and Bonadonna [6,7] measured
the particle velocity in a settling column using a high speed camera
and also presented 3D-models of particles which were generated
using SEMmicro-CT. Carranza and Zhang [9] proposed the usage of ste-
reo vision and Schlieren photography to study the orientation of parti-
cles and their drag coefficients in settling column experiments.
Unfortunately, these methods are either very costly or time consuming.
For that reason, the results in the present study were determined in ab-
sence of such techniques. The equipment used for experimental work
was either from the market segment of consumer electronics or already
standard in many industrial applications. One of the objectives of this
study was to point out, that the rising performance and decreasing
cost of consumer electronics allows researchers in many fields the at-
tainability of results with very reasonable quality under the use of less
resources.
3.1. Particle geometry analysis
Different slag particles were examined using a digital brightfield mi-
croscope and 50× magnification. The software package “ImageJ” [29,30]
was used for post-processing of the images and the measurement of
particle dimensions. Fig. 1 shows the minimum and maximum Feret di-
ameters of the largest projection area and a second lateral view of the
particle, these measurements were determined by the software.
The microscopic characterization of particles was carried out in con-
sideration of the “PA protocol” as proposed by Bagheri and Bonadonna
[6] in order to prevent operator-dependent measurements. Averaged
particle dimensions were obtained by measuring 50 individual particles
and forming the arithmetic mean value. The drag model of Kishore and
Gu [27] was specially developed for elliptical particles and the particle
aspect ratio, which is the particle shape descriptor within this model,
was as well only defined for elliptical particles. Hence, the geometrically
determined particle aspect ratio ARGeo was defined via Eq.(5) within this
work.
ð3Þ
2∙D f ;max
ARGeo ¼ ð5Þ
D f ;min þ d f ;max
In addition to the characteristic particle dimensions, the mean sphe-
ricity and the particle size distribution of particle samples were deter-
mined by a particle analyzer based on the principle of dynamic image
analysis [31]. The apparatus analyzed at least 50,000 individual particle
ð4Þ images of each sample in less than 2 min which obviously saved an
enormous amount of time in comparison to microscopy.
3.2. Drop velocity measurement
As already mentioned in Section2, the drag models within this work
rely on different particle shape descriptors as input parameters. As Loth
[32] concluded, the use of geometrically determined shape descriptors
provides drag calculations with reasonable results, but their quality
can still be improved by experimental data of the same particles at the
same particle Reynolds – numbers. Here, the experimental data was ob-
tained from particle velocity measurements after a drop from 3 m
through still air. The particles fell through a grounded stainless steel
pipe with a diameter of 50 mm. Shortly after exiting the pipe, the parti-
cles were tracked with a frame rate of 1000 pictures per second. A “Sony
DSC-RX100 IV” digital camera was used for imaging, light was provided
by two 400 W halogen headlamps. The major advantage of the setup
displayed in Fig. 2 was, that the cost for the camera had only been a frac-
tion of the cost for usual high speed cameras. The camera axis was posi-
tioned exactly 3 m below the particle inlet and pictures with a height of
126 mm were taken. The distance δP between the camera lens and the
pipe axis was set to 250 mm, the scale was mounted at the same dis-
tance. In order to prevent external influences on the measurements,
the setup including the camera was shielded with clear polycarbonate.
Due to the fact, that the particles passed the view field of the camera
at an arbitrary distance δP, 1 ≥ δ ≥ δP, 2 but the scale was mounted at
the distinct distance δP as outlined in Fig. 2, the velocity of each particle
varied in the range z_ P;1 ≥ z_ P ≥ z_ P;2. Thus, a random error ε = ± 8, 77% was
induced to the velocity measurements. With respect to the random
error, the highest and lowest velocity measurements were neglected
as outlined further in Section5.1.
 H. Gerhardter et al. / Powder Technology 330 (2018) 152–163 155

Fig. 1. Particle from two different views.

3.3. Air tunnel separator
Simplifications such as free fall and sedimentation are rare in indus-
trial applications. Particularly in the thermal utilization of granular fuels,
the particles are exposed to highly turbulent flows. Particle trajectories
and concentrations are difficult to measure in such environments, but
an important point of this research was to perform application-
oriented tests on the above-mentioned drag formulations. For this rea-
son, a test rig in form of an air tunnel separator, displayed in Fig. 3
was built.

Fig. 2. Camera setup.

156 H. Gerhardter et al. / Powder Technology 330 (2018) 152–163

Fig. 3. Test rig.

The volume flow rate of supplied compressed air was adjusted by a The median aspect ratio of the three computational grids was below
pressure regulator. A rotameter was used for adjustments while the vol- 1.4 and the median orthogonal quality above 0.93 which seemed suffi-
ume flow rate was determined using a simple diaphragm gas meter and a cient. The details of mesh 2 are displayed in Fig. 4.
stop watch. The air stream was fed to the air tunnel separator through a The 4 mm thick plywood baffles were resolved by the mesh. Here,
slotted hole with a width of 6 mm and a length of 46 mm in the left particles were reflected. Due to the lack of reliable literature data for
side wall. The heart of the test stand was a box which was 815 mm the pairing of coal slag particles and plywood walls, a coefficient of res-
long, 700 mm high and 216 mm in depth. The clear polycarbonate front titution of 0.6 was assumed in tangential and normal direction. This
wall, which is missing in Fig. 3, was removed for documentation only. value was a rough approximation of the data published by Sommerfeld
Particles were fed through a bore which was 5 mm in diameter and lo- and Huber [35] for rough quartz particles and polished steel. The varia-
cated in the top wall. The particles were deflected by the air jet and set- tion of the coefficient of restitution between 0.5 and 0.8 brought up the
tled in the plastic containers on the bottom. Baffles made from 4 mm same results. The portion of particle mass flow into each of the 14 rows
thick plywood were installed to prevent the particles from bouncing be- was determined by numerical samples at the marked surfaces in Fig. 4.
tween the 14 plastic container rows. The rows were numbered, starting
with number one on the left and number 14 on the right side of the box. 4.2. Modelling of fluid flow and turbulence
The basic idea was that particles with lower drag coefficients will
settle on the left side of the box while highly aspherical particles A steady-state solution of the isothermal flow within the test bench
would be further deflected to the right. After the experiment, the parti- was obtained in application of the commercial CFD solver Ansys Fluent
cle mass distribution on the bottom was determined by weighing the 17.0. For each cell in the computational grid, the governing laws, namely
plastic containers. Later, it was compared to the numerical result. The the conservation of mass and momentum, given by Eqs.(6)and (7) in
parameters of the experiment, which have also been used for further form of the Reynolds Avereged Navies Stokes equations (RANS), were
numerical work, are listed in Table1. The material properties of air solved numerically.
were taken from [33], the particle density was set to ρP = 2500 kg/m3
according to the work of Aineto etal. [34]. ∂
ðρui Þ ¼ 0 ð6Þ
∂xi
4. Numerical setup
" !#
4.1. Numerical grid and boundary conditions ∂   ∂p ∂ ∂ui ∂u j 2 ∂ui
ρui u j ¼ − þ μ þ − δij
∂x j ∂xi ∂x j ∂x j ∂xi 3 ∂xi
To ensure the mesh-independence of the numerical simulation, cal- ∂
0 0

þ −ρui u j ð7Þ
culations were performed on three different numerical grids. The grid ∂x j
parameters are given in Table2.
Table 1
Parameters of the experimental setup.

Duration of the experiment 484 [s] Table 2

Air volume flow rate 297,52 [l/min] Mesh parameters.
Air temperature 20 [°C]
Mesh 1 Mesh 2 Mesh 3
Air density 1204 [kg/m3]
Air dynamic viscosity 0,00001825 [kg/m s] Number of elements 0.86 Mio. 1.68 Mio. 2.22 Mio.
Air mass flow rate 0,0059702 [kg/s] Max. cell aspect ratio 17.39 16.76 20.92
Particle mass flow rate 0,002360 [kg/s] Min. orthogonal quality 0.1534 0.1809 0.1199
Particle density 2530 [kg/m3] Max. skewness 0.88 0.88 0.89
 H. Gerhardter et al. / Powder Technology 330 (2018) 152–163 157

Fig. 4. Mesh and boundary conditions.

The Transition-SST turbulence model [36–38] was used to incorpo- each particle was tracked 50 times. ARosin-Rammler distribution
rate turbulence. The basic equations were solved in application of a [40,41] was fitted to the cumulative size distribution measured by the
pressure-based coupled solver. The pressure-velocity coupling was particle analyzer, both curves are displayed in Fig. 5.
solved in application of the PRESTO! – scheme, other transport equa- The parameters of the Rosin-Rammler distribution were a mean di-
tions were discretized using a second order upwind scheme. A Least ameter deq;V ¼ 700 μm and a spread parameter of 6. 10 diameter classes
Squares Cell-Based method was applied to calculate gradients [36]. in the range between 400 μm and 850 μm were incorporated by the
numerical model. The particle volume fraction was well below 1% and
4.3. Discrete phase modelling hence, a one-way-coupled approach, where the particles had no
influence on the flow field of the continuous phase was neglected,
While the basic equations of fluid mechanics have been solved in an was chosen [36].
Eulerian reference frame, a Lagrangian reference frame was used to cal-
culate the particle trajectories. The equation of particle motion is basi-
5. Results and discussion
cally a force balance, here displayed in form of Eq.(8).
! ! ! 5.1. Determination of drag parameters
dup u−up !
¼ þ a ð8Þ
dt τr
The first objective was to determine a valid formulation for the par-
! ticle drag coefficient by the measurement of the drop velocity of 50 dif-
dup ferent, randomly picked particles. The measured drop velocities are
The temporal change of particle momentum is caused by drag
dt displayed in Fig. 6.
! !
u−up ! The drop velocities were spread widely in the range between
τr and other forces a , for example gravity [36]. An Euler-implizit
discretization scheme was applied to solve the equation of motion for −4.47 m/s and −2.04 m/s. The variation of the fall speed could
!nþ1 not be explained by the size distribution alone, the influence of differ-
each particle. The particle velocity u p after the timestep Δt is calcu-
ently shaped particles had to be taken into account. Thus, the particles
lated via Eq.(9).
were divided into four classes and the mean drop velocity z_ P was
!
!n
! ! u
u p þ Δt a þ
!nþ1 τr
up ¼ ð9Þ
Δt
1þ
τr

The particle relaxation time τr is given by Eq.(10).

2
ρP d P 24
τr ¼ ∙ ð10Þ
18μ cD ∙ ReP

In addition to the particle density ρP, the dynamic viscosity of the

continuous phase μ, the volume-equivalent particle diameter deq, V and
the particle Reynolds – number ReP, it is mainly influenced by the parti-
cle drag coefficient cD. The characteristic length scale for particle track-
ing was set to 10−4 m, further refinement didn't bring up significantly
different results. A stochastic tracking approach proposed by Gosmann
and Ioannides [39] was used to incorporate the interaction between
the slag particles and turbulent fluctuations in the flow field. This way, Fig. 5. Particle size distribution.
158 H. Gerhardter et al. / Powder Technology 330 (2018) 152–163

Fig. 6. Drop velocity measurements and particle classification.

determined for each class. To account for the random error, the results mean drop velocities of the four particle classes and the free fall of a
of measurement 1–3 and 48–50 were neglected in further work while spherical particle with the same size are compared in Fig.7a. Here, the
44valid results remained. It was assumed that the four particle drag model proposed by Kishore and Gu [27] was applied.
classesshare the same particle size distribution as displayed in Fig.
5and thus, the particle mean diameter of each class was set to deq;V ¼
700 μm. Theproperties of the classes are outlined in Table3. The mea-
surements 4–10 were picked to represent particle class 1. The particle
mass fraction of each class was determined by the ratio between the
number of measurements corresponding to one particle class and the
total number of measurements, e.g. class 1: x1 = 7/44 = 15.91%.
In the second step, the drag model parameters, namely the mean
particle sphericity Ψ for the drag model of Haider and Levenspiel [25],
the mean size ratio SR for the model of Richter and Nikrityuk [26] and
the mean aspect ratio AR for the model of Kishore and Gu [27] were de-
termined. The velocity of a particle with a diameter deq, V = 700 μm after
3 m of free fall in still air was calculated with Eqs.(9)and (10) and a
timestep Δt = 10−4 s. The drag models [25–27] were used to calculate
the drag coefficient at each timestep and the corresponding particle
shape descriptors were varied in order to fit the calculated drop velocity
to the measured mean velocities in Table3, the resulting mean particle
shape descriptors of each class are given in Table4. The calculated

Table 3
Particle classification.

deq;V ½μm Measurement no. z_ P ½m=s xi

Class 1 700 4–10 −3.94 15.91%

Class 2 11–25 −3.47 34.09%
Class 3 26–40 −2.85 34.09%
Class 4 41–47 −2.45 15.91%

Table 4
Particle shape descriptors.

Ψ ½− SR ½− AR ½−

Class 1 0.786 0.674 1.365

Class 2 0.714 0.546 1.69
Class 3 0.599 0.414 2.295
Class 4 0.503 0.343 2.875
Fig. 7. Calculated drop velocities.
 H. Gerhardter et al. / Powder Technology 330 (2018) 152–163 159

Table 5
Calculated drop velocity [m/s] and Reynolds number [-] variation.

Drag model Haider and Richter and Kishore and

Levenspiel [24] Nikrityuk [25] Gu [26]

deq, V 400 850 400 850 400 850

Class 1 −2.69 −4.32 −3.19 −4.10 −2.41 −4.51
Class 2 −2.46 −3.80 −2.90 −3.60 −2.09 −4.03
Class 3 −2.08 −3.13 −2.47 −2.94 −1.71 −3.35
Class 4 −1.80 −2.68 −2.17 −2.52 −1.48 −2.89
ReP,max 242 230 253
ReP,min 47.5 57.3 39

The calculated results of particle class 1 and 4 and all applied drag
models are compared with each other in Fig.7b. The deviation between
the results was small and it has to be noted, that the particles in class 3
and 4 already reached N99% of their terminal velocity after 3 m. The cal-
culations were performed with a diameter deq, V = 700 μm, the variation
between the velocity of each particle class at a distinct position was only
caused by the influence of the particle shape. The drop velocities of
differently sized and shaped particles were summarized in Table5. The
maximum and minimum particle Reynolds numbers ReP,max and ReP,
min were calculated with the highest and lowest drop velocity magni-
tude given by the corresponding drag model. When the particle size
was incorporated by the calculations, the different drag models
[25–27] resulted in different velocity variations. The lowest spreading
was predicted by the drag model of Richter and Nikrityuk [26]. In-
corporating the particle size and shape, the values were spread between
−2.17 m/s and −4.10 m/s which almost covered the measured spec-
trum between −4.47 m/s and −2.04 m/s in Fig. 6. The drop velocity
predicted by the drag model proposed by Kishore and Gu [27], varied
between −1.48 m/s and −4.51 m/s which is significantly wider than
the measured range displayed in Fig. 6.
5.2. Air tunnel separator
After the determination of valid drag model parameters, the numeri-
cal and experimental results of the air tunnel separator were evaluated.
The main emphasis was to investigate the applicability of the dragmodels
and the corresponding model parameters in practical applications.
5.2.1. Fluid flow and mesh independence
The numerically calculated velocity profile within the air tunnel sep-
arator is displayed in Fig. 8, the results were obtained on the three dif-
ferent meshes. The freestream of air as well as the circulating flow are
Fig. 9. Velocity profiles for mesh independence test.
clearly visible in all three pictures. For further investigations, the veloc-
ity profiles at the dashed lines in Fig. 8 were compared with each other.
The velocity profiles are displayed in Fig. 9.
Mesh 2 and mesh 3 brought up almost identical results in the rest of
the body while small deviations occur for mesh 1. The velocity profile
within the freestream was almost equal in all three simulations. The
comparison of the numerically obtained particle mass distributions
within the container rows of the test stand in Fig. 10 brought up a sim-
ilar trend. Here, the drag model of Kishore and Gu [27] was applied. The
deviations between mesh 1 and the other grids were relatively high
while the differences between mesh 2 and 3 were almost negligible.
Hence it was concluded, that the resolution of mesh 2 and mesh 3 was
sufficient for further work while mesh 1 appeared to be too coarse. Fur-
ther results in this work were obtained on the basis of mesh 2.
5.3. Multiphase flow
The particle content in each of the rows on the bottom of the test
stand was weighted and compared to the numerical results. The
amount of particles in the rows 1–4 and 12–14 was relatively small.
For that reason, the filling of the containers in these rows was deter-
mined by a single weighing.

Fig. 8. Calculated velocity profile within the test stand.

160 H. Gerhardter et al. / Powder Technology 330 (2018) 152–163

Fig. 10. Particle mass distribution within the test stand on different meshes.

Fig. 11. Particle Classes in the test stand (numerical result).

The importance of particle classification is displayed in Fig. 11 which
was obtained by numerical calculations and the drag model proposed
by Kishore etal. [27], similar results were obtained in application of
the other drag models. As lined out in Table3 and Table4, the drop
velocity decreased from class 1 to class 4. Slower particles had higher
aspect ratios or lower size ratios and sphericity values, respectively.
This significantly affected the particle drag coefficient. As displayed in
Fig.11, highly aspherical particles in class 4 were caught further down-
stream within the test stand while more spherical particles settled in
the first rows. If only a single particle class with an averaged drag coef-
ficient of the coal slag particles was used in numerical calculations, the
calculated particle dispersion would have been significantly lower.
The experimentally determined particle mass distribution in the
container rows and the numerical results are displayed in Fig. 12. The
drag model proposed by Haider and Levenspiel [25], in combination
with the custom drag model parameters given in Table4, generally
overestimated the drag coefficients. As a result, the calculated particle
accumulation in rows 8–14 was too high. The drag models proposed
by Kishore and Gu [27] and Richter and Nikrityuk [26] provided more
reasonable results in the left section of the test stand while they
underestimated the particle content in the downstream rows. The
reason for the deviations between the models was found, when the
particle drag coefficients as a function of the particle Reynolds number,
displayed in Fig. 13, and the Reynolds numbers of 100 particles,

Fig. 12. Particle mass distribution within the test stand – experimental and numerical results.
 H. Gerhardter et al. / Powder Technology 330 (2018) 152–163
Fig. 13. Drag coefficient as a function of particle Reynolds number.
Fig. 14. Particle tracks coloured by the particle Reynolds number.
displayed in Fig. 14, were investigated. The results in Fig. 14 were ob-
tained by the drag model proposed by Kishore and Gu [27].
The drag formulation of Haider and Levenspiel [25] predicted an in-
creasing drag coefficient in the region 100 ≤ ReP ≤ 1000 while a decreas-
ing curve was calculated by the drag formulations of Richter and
Nikrityuk [26] and Kishore and Gu [27]. As pointed out in Section2,
the latter two drag models only offer a limited validity in the range
Fig. 15. Particle mass distribution in application of the custom drag model.
161
ReP N 200. Fig. 14 shows that the particle Reynolds numbers were
below this limit in most regions of the airflow separator. Higher particle
Reynolds numbers only occurred directly in the freestream. It was as-
sumed, that the particle drag coefficient was overestimated by [25],
while the values calculated by [26,27] were too low for 100 ≤ ReP ≤
1000 and that the problem could be overcome by a simple solution:
The particle drag coefficients were calculated in application of the
drag models proposed by Haider and Levenspiel [25] via Eq.(2) in com-
bination with the drag model proposed by Kishore and Gu [27] and Eq.
(4). The custom drag formulation is given by Eq.(11). The best fit was
found for the weighting factors C1 = 0.3 and C2 = 0.7.
cD;HL KG ¼ C 1 ∙cD;HL þ C 2 ∙cD;KG ð11Þ
The particle mass distribution within the container rows is displayed
in Fig. 15 and the corresponding deviation between experimental and
numerical results is displayed in Fig. 16. The results on the basis of the
weighted average drag model due to [25,27] showed the slightest devi-
ation in all rows. Even though the amount of particles in row 12–14 was
underestimated by N50%, the custom drag model provided a significant
improvement compared to the other approaches.
5.3.1. Particle shape factor measurement
The four particle classes were characterized by the particle shape de-
scriptors given by Table4. These shape factors were obtained by fitting
calculated drop velocities to measurements and as a consequence, they
constituted empirically determined drag model parameters. As already
displayed in Fig. 11, particles of each class settled in distinct sections of
162 H. Gerhardter et al. / Powder Technology 330 (2018) 152–163
Fig. 16. Error of the numerical results.
the test stand. According to the numerical results, only the particle classes
1 and 2 were trapped in the first four rows, whereas in the rows 12–14
almost exclusively the particle class 4 was to be found. All four particle
classes settled in row 7. In these sections the empirically determined
drag model parameters, namely the mean sphericity Ψ and mean aspect
ratio AR, were compared to the geometrical shape descriptors ΨGeo and
ARGeo . The geometric mean particle aspect ratio ARGeo was calculated by
Eq.(5), the corresponding particle dimensions were obtained by micro-
scopic measurements. The particle sphericity ΨGeo was determined by dy-
namic image analysis of the coal slag particles. The results, which were
obtained in application of the weighted average drag model due to
[25,27], are listed in Table6.
In general, the drag model parameter Ψ was lower than the geomet-
rical shape descriptor ΨGeo. Calculating the particle drag coefficient with
the geometrical shape descriptor would have returned values which
were too low and as a result, the particle deflection would have been
underestimated. The drag model parameter AR was significantly lower
than the geometric value ARGeo in rows 1–4 while the opposite was
the case in the rows 12–14. In row 7, a relatively close fit was obtained.
The determination of the particle drag coefficient by means of the geo-
metric particle aspect ratio would have resulted in too low particle dis-
persion. Loth's proposal [32] to determine the model parameters
experimentally instead of using geometric shape parameters as input
variables was thus confirmed.
6. Conclusion and outlook
The performance of the three evaluated drag models was improved by
measuring the settling velocity of powder grains. The measurements
were divided into four velocity classes and, instead of using geometrically
determined particle shape descriptors for drag calculations, the drag
model parameters of each particle class were determined empirically by
fitting them according to the mean velocity of each class. A standard dig-
ital camera with a frame rate of 1000 pictures per second was used for
particle velocity measurements. The camera enabled measurements
with sufficient precision and it was shown that products from the con-
sumer electronics segment can very well support research in a very
cost-efficient way. The visibility of very small or, for example, transparent
Table 6
Comparison of particle shape descriptors.
Row Ψ ΨGeo AR ARGeo
1–4 0.763 0.844 1.469 1.845
7 0.540 0.761 2.083 1.929
12–14 0.505 0.646 2.863 2.217
particles on the images was identified as the only limiting factor. Real
powders consist of differently sized and shaped particles. It was shown
that it is necessary to incorporate not only the particle size distribution,
but also different particle shapes in the drag calculations in order to pre-
dict the motion of the powder grains in a flow. The empirically deter-
mined drag model parameters were validated by measurements and
numerical calculations of an air tunnel separator. Two of the three evalu-
ated drag models, namely the model of Kishore and Gu [27] and the drag
model proposed by Richter and Nikrityuk [26], were specialized on ellip-
soidal particles while the third model, proposed by Haider and Levenspiel
[25], was a general drag model for nonspherical particles. Even though the
drag parameters of all three models were determined in the same way,
the specialized models for ellipsoidal particles provided satisfying results.
It was concluded that the results of drag calculations can be improved fur-
ther by selecting a drag model which is specialized on the particle shape.
Based on the experimental and numerical results of the air tunnel separa-
tor, the empirically determined particle shape descriptors were compared
to the geometric values. The geometrically determined particle shape fac-
tors deviated significantly from the empirically determined drag model
parameters. It was concluded that both the particle dispersion and the po-
sition of the peak particle concentration at the bottom of the test rig can
only be calculated by means of empirically determined model parame-
ters. The proposed procedure for the determination of drag model param-
eters provided a significant improvement on the calculated results. Due to
its simplicity, it can be easily adapted for the determination of drag pa-
rameters of almost any powder or granular matter. This way, the results
of numerical calculations of particle laden flows in various fields of re-
search, for example waste incineration, biomass gasification or in the
chemical industry can be highly improved.
Acknowledgement
This work was financially supported by the Austrian Research
Promotion Agency (FFG) (Project 861148, eCall 9336196).
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, https://doi.org/10.
1016/S0032-5910(98)00178-8.
[2] G.H. Ganser, A rational approach to drag prediction of spherical and nonspherical
particles, Powder Technol. 77 (1993) 143–152, https://doi.org/10.1016/0032-5910
(93)80051-B.
[3] E.F. Schulz, R.H. Wilde, M.L. Albertson, Influence of Shape on the Fall Velocity of
Sedimentary Particles, 1954.
[4] P. Swamee, C. Ojha, Drag coefficient and fall velocity of nonspherical particles, J.
Hydraul. Eng. 117 (1991) 660–667, https://doi.org/10.1061/(ASCE)0733-9429
(1991)117:5(660).
[5] A. Hölzer, M. Sommerfeld, New simple correlation formula for the drag coefficient of
non-spherical particles, Powder Technol. 184 (2008) 361–365, https://doi.org/10.
1016/j.powtec.2007.08.021.
 H. Gerhardter et al. / Powder Technology 330 (2018) 152–163
[6] G.H. Bagheri, C. Bonadonna, I. Manzella, P. Vonlanthen, On the characterization of
size and shape of irregular particles, Powder Technol. (2015) 141–153, https://doi.
org/10.1016/j.powtec.2014.10.015.
[7] G. Bagheri, C. Bonadonna, On the drag of freely falling non-spherical particles,
Powder Technol. 301 (2016) 526–544, https://doi.org/10.1016/j.powtec.2016.06.015.
[8] W.W. Willmarth, N.E. Hawk, R.L. Harvey, Steady and unsteady motions and wakes of
freely falling disks, Phys. Fluids 7 (1964) 197, https://doi.org/10.1063/1.1711133.
[9] F. Carranza, Y. Zhang, Study of drag and orientation of regular particles using stereo
vision, Schlieren photography and digital image processing, Powder Technol. 311
(2017) 185–199, https://doi.org/10.1016/j.powtec.2017.01.010.
[10] A. Richter, P.A. Nikrityuk, New correlations for heat and fluid flow past ellipsoidal
and cubic particles at different angles of attack, Powder Technol. 249 (2013)
463–474, https://doi.org/10.1016/j.powtec.2013.08.044.
[11] J.M. Jones, A.R. Lea-Langton, L. Ma, M. Pourkashanian, A. Williams, Mathematical
modelling, Pollut. Gener. by Combust. Solid Biomass Fuels, Springer London,
London 2014, pp. 71–97, https://doi.org/10.1007/978-1-4471-6437-1_6.
[12] S.R. Gubba, L. Ma, M. Pourkashanian, A. Williams, Influence of particle shape and in-
ternal thermal gradients of biomass particles on pulverised coal/biomass co-fired
flames, Fuel Process. Technol. 92 (2011) 2185–2195, https://doi.org/10.1016/j.
fuproc.2011.07.003.
[13] C. Yin, Biomass co-firing, Biomass Combust. Sci. Technol. Eng, Elsevier 2013,
pp. 84–105, https://doi.org/10.1533/9780857097439.2.84.
[14] C. Yin, L. Rosendahl, S. Knudsen Kaer, H. Sorensen, Modelling the motion of
cylindrical particles in a nonuniform flow, Chem. Eng. Sci. 58 (2003) 3489–3498,
https://doi.org/10.1016/S0009-2509(03)00214-8.
[15] C. Yin, L. Rosendahl, S.K. Kær, T.J. Condra, Use of numerical modeling in design for
co-firing biomass in wall-fired burners, Chem. Eng. Sci. 59 (2004) 3281–3292,
https://doi.org/10.1016/j.ces.2004.04.036.
[16] H. Lu, E. Ip, J. Scott, P. Foster, M. Vickers, L.L. Baxter, Effects of particle shape and size
on devolatilization of biomass particle, Fuel 89 (2010) 1156–1168, https://doi.org/
10.1016/j.fuel.2008.10.023.
[17] B. Peters, C. Bruch, Drying and pyrolysis of wood particles: experiments and simula-
tion, J. Anal. Appl. Pyrolysis 70 (2003) 233–250, https://doi.org/10.1016/S0165-
2370(02)00134-1.
[18] R. Sampath, D.J. Maloney, J.W. Zondlo, S.D. Woodruff, Y.D. Yeboah, Measurements of
coal particle shape, mass, and temperature histories: impact of particle irregularity
on temperature predictions and measurements, Symp. Combust. 26 (1996)
3179–3188, https://doi.org/10.1016/S0082-0784(96)80163-9.
[19] M. Kestel, D. Safronov, A. Richter, P.A. Nikrityuk, Pseudo-steady-state approach for
carbon particle combustion/gasification, Gasif. Process. Wiley-VCH Verlag GmbH &
Co. KGaA 2014, pp. 205–242, https://doi.org/10.1002/9783527673186.ch8.
[20] K. Wittig, P. Nikrityuk, A. Richter, Drag coefficient and Nusselt number for porous
particles under laminar flow conditions, Int. J. Heat Mass Transf. 112 (2017)
1005–1016, https://doi.org/10.1016/j.ijheatmasstransfer.2017.05.035.
[21] M. Tamura, S. Watanabe, N. Kotake, M. Hasegawa, Grinding and combustion charac-
teristics of woody biomass for co-firing with coal in pulverised coal boilers, Fuel 134
(2014) 544–553, https://doi.org/10.1016/j.fuel.2014.05.083.
[22] G. Dunnu, T. Hilber, U. Schnell, Advanced size measurements and aerodynamic clas-
sification of solid recovered fuel particles, Energy Fuel 20 (2006) 1685–1690,
https://doi.org/10.1021/ef0600457.
163
[23] G. Dunnu, J. Maier, T. Hilber, G. Scheffknecht, Characterisation of large solid recov-
ered fuel particles for direct co-firing in large PF power plants, Fuel 88 (2009)
2403–2408, https://doi.org/10.1016/j.fuel.2009.03.004.
[24] G. Dunnu, J. Maier, U. Schnell, G. Scheffknecht, Drag coefficient of solid recovered
fuels (SRF), Fuel 89 (2010) 4053–4057, https://doi.org/10.1016/j.fuel.2010.06.039.
[25] A. Haider, O. Levenspiel, Drag coefficient and terminal velocity of spherical and non-
spherical particles, Powder Technol. 58 (1989) 63–70, https://doi.org/10.1016/
0032-5910(89)80008-7.
[26] A. Richter, P.A. Nikrityuk, Drag forces and heat transfer coefficients for spherical, cuboi-
dal and ellipsoidal particles in cross flow at sub-critical Reynolds numbers, Int. J. Heat
Mass Transf. 55 (2012) 1343–1354, https://doi.org/10.1016/j.ijheatmasstransfer.
2011.09.005.
[27] N. Kishore, S. Gu, Momentum and heat transfer phenomena of spheroid particles at
moderate Reynolds and Prandtl numbers, Int. J. Heat Mass Transf. 54 (2011)
2595–2601, https://doi.org/10.1016/j.ijheatmasstransfer.2011.02.001.
[28] M.E. Clift, R. Grace, J.R. Weber, Bubbles, Drops and Particles, 1978https://doi.org/10.
1017/S0022112079221290.
[29] J. Schindelin, C.T. Rueden, M.C. Hiner, K.W. Eliceiri, The ImageJ ecosystem: an open
platform for biomedical image analysis, Mol. Reprod. Dev. 82 (2015) 518–529,
https://doi.org/10.1002/mrd.22489.
[30] J. Schindelin, I. Arganda-Carreras, E. Frise, V. Kaynig, M. Longair, T. Pietzsch, S.
Preibisch, C. Rueden, S. Saalfeld, B. Schmid, J.-Y. Tinevez, D.J. White, V. Hartenstein,
K. Eliceiri, P. Tomancak, A. Cardona, Fiji: an open-source platform for biological-
image analysis, Nat. Methods 9 (2012) 676–682, https://doi.org/10.1038/nmeth.
2019.
[31] ISO 13322-2:2006-11, Particle Size Analysis – Image Analysis Methods – Part 2:
Dynamic Image Analysis Methods, 2006.
[32] E. Loth, Drag of non-spherical solid particles of regular and irregular shape, Powder
Technol. 182 (2008) 342–353, https://doi.org/10.1016/j.powtec.2007.06.001.
[33] Çengel, Introduction to Thermodynamics and Heat Transfer, 2013, https://doi.org/
10.1017/CBO9781107415324.004.
[34] M. Aineto, A. Acosta, J.M. Rincón, M. Romero, Thermal expansion of slag and fly ash
from coal gasification in IGCC power plant, Fuel 85 (2006) 2352–2358, https://doi.
org/10.1016/j.fuel.2006.05.015.
[35] M. Sommerfeld, N. Huber, Experimental analysis and modelling of particle-wall col-
lisions, Int. J. Multiphase Flow 25 (1999) 1457–1489, https://doi.org/10.1016/
S0301-9322(99)00047-6.
[36] A. Inc, Ansys Fluent Theory Guide 17.0, 2016.
[37] F.R. Menter, T. Esch, S. Kubacki, Transition modelling based on local variables, Eng.
Turbul. Model. Exp. 5 (2002) 555–564, https://doi.org/10.1016/B978-008044114-
6/50053-3.
[38] F.R. Menter, R.B. Langtry, S.R. Likki, Y.B. Suzen, P.G. Huang, S. Völker, A correlation-
Basedtransition model using local variables—part I: model formulation, J. Turbomach.
128 (2006).
[39] A.D. Gosman, E. Ioannides, Aspects of computer simulation of liquid-fueled combus-
tors, J. Energy 7 (1983) 482–490, https://doi.org/10.2514/3.62687.
[40] P. Rosin, E. Rammler, The laws governing the fineness of powdered coal, J. Inst. Fuel
7 (1933) 29–36.
[41] A. Vesilind, The Rosin-Rammler particle size distribution, Resour. Recover. Conserv.
5 (1980) 275–277.