See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/225125958

Particle screening phenomena in an oblique multi-level tumbling reservoir: A

numerical study using discrete element simulation

Article in Granular Matter · November 2007

DOI: 10.1007/s10035-007-0042-6

CITATIONS READS

32 2,209

2 authors, including:

Hashem Alkhaldi
University of Jordan
20 PUBLICATIONS 161 CITATIONS

SEE PROFILE

All content following this page was uploaded by Hashem Alkhaldi on 26 February 2015.

The user has requested enhancement of the downloaded file.

Granular Matter (2007) 9:415–429
DOI 10.1007/s10035-007-0042-6

Particle screening phenomena in an oblique multi-level tumbling

reservoir: a numerical study using discrete element simulation
Hashem Alkhaldi · Peter Eberhard

Received: 25 July 2006 / Published online: 30 May 2007

© Springer-Verlag 2007

Abstract Due to their wide usage in industrial and sifting rate of the separated particles and the efficiency of the
technological processes, granular materials have captured segregation process have been studied. It is recognized that
great interest in recent research. The related studies are often the screening phenomenon is very sensitive to the machines
based on numerical simulations and it is challenging to inves- geometrical parameters, i.e. plate inclinations, shaft eccen-
tigate computational phenomena of granular systems. tricities and aperture sizes in the sieving plates at different
Particle screening is an essential technology of particle sepa- levels of the structure. The rotational speed of the machine
ration in many industrial fields. This paper presents a and the feeding rate of the particles flow have also a great
numerical model for studying the particle screening process influence on the transportation and segregation rates of the
using the discrete element method that considers the motion particles. In an attempt to better understand the mechanism
of each particle individually. Dynamical quantities like par- of the particle transport between the different layers of the
ticle positions, velocities and orientations are tracked at each sifting system, different computational studies for achieving
time step of the simulation. The particular problem of interest optimal operation have been performed.
is the separation of round shape particles of different sizes
using a rotating tumbling vertical cylinder while the par- Keywords Discrete element method · Particle
ticulate material is continuously fed into its interior. This simulation · Sorting process · Sorting machine · Particle
rotating cylinder can be designed as a uniform or stepped screening
multi level obliqued vertical vessel and is considered as a big
reservoir for the mixture of particulate material. The finer 1 Introduction
particles usually fall through the sieve openings while the
oversized particles are rebounded and ejected through out- The last decades have witnessed an enormous development
lets located around the machine body. Particle–particle and in the research of granular media and particle simulation [20].
particle–boundary collisions will appear under the tumbling Granular media studies are considered as being of great inter-
motion of the rotating structure. A penalty method, which est and are required in many engineering processes in differ-
employs spring-damper models, will be applied to calcu- ent fields of industry. To improve the performance of such
late the normal and frictional forces. As a result of colli- processes, a good understanding of the behavior of parti-
sions, the particles will dissipate kinetic energy due to the cle motion is important and will contribute in the burgeon-
normal and frictional contact losses. The particle distribution, ing of many different industrial applications of the granular
technology.
H. Alkhaldi · P. Eberhard (B) The term ‘Molecular Dynamics’ refers usually to compu-
Institute of Engineering and Computational Mechanics, tational techniques which use classical mechanics to analyze
University of Stuttgart, Pfaffenwaldring 9,
70569 Stuttgart, Germany
the structure and dynamics of molecular systems including
e-mail: [email protected] polymers and macromolecules but are also applied to par-
URL: www.itm.uni-stuttgart.de ticulate materials. Contact phenomena are among the most
H. Alkhaldi interesting problems in molecular dynamics studies. Contact
e-mail: [email protected] usually involves friction, which may only be neglected for

123
416 H. Alkhaldi, P. Eberhard

simplicity in the case where frictional forces are small com- (a) vibrating screens in which the particulate material moves
pared to the normal ones. The complexity of a contact prob- relative to the screen in a vertical plane, (b) flat screens in
lem is due to at least three aspects. The first aspect is the which the particulate material moves relative to the screen
nonlinear boundary condition at the contact region caused by in the plane of the screen and (c) rotating screens in which
the impenetrability constraint. The parameters of the contact the screen surface is cylindrical and the particulate material
region which include the sliding state, the frictional stress cascades over the inner surface as the screen is rotated.
distribution, the shape and the size of the region, etc. are The research work of Jansen and Glastonbury [8] in
unknown before the analysis. The second aspect is the 1967/1968 in studying particle screening phenomena is
description of friction phenomena, which usually has no sim- among the earliest works in this field. They have tried to
ple solution. The third aspect is the material and geometric analyze the dynamics of screening processes and to study
nonlinearity. It is expensive and difficult to solve such a prob- the factors that affect the screening performance. They have
lem with geometric and physical nonlinearities. built their results on probability theories to understand the
Molecular dynamics (MD) simulation is a commonly used kinetics of sieving. The effect of the non-ideal aperture dis-
tool in physics and material science for modeling solids and tribution of sieving on the sieve residue is studied in [9],
liquids at the atomic level. In granular media often each par- where an algorithm for deducing an effective sieve residue
ticle is treated as a rigid spherical body and Newtons equa- from the rate of powder passage through a sieve is described.
tions of motion are integrated to track the motion of each A detailed study of the sieving kinetics using batch siev-
particle. Interactions between grains are calculated with a ing is described in [25] showing that the rate of sieving and
contact algorithm that forbids interpenetration but allows its efficiency are determined by the numerical values of the
separation, sliding and rolling with friction. The mechani- sieving rate constants of each of the particle sizes in the feed
cally correct description and simulation of contacts between mixture. Furthermore, they found that the near-mesh parti-
many bodies is a very computation-time intensive topic. cles play a major role in the overall kinetic process and the
While by simple methods, a relatively low number of par- presence of the oversized particles in the feed enhances the
ticles can be already computed with sufficient accuracy and sieving rate dramatically. Continuous screening phenomena
acceptable computation times, there remain difficult and and comparisons with batch sieving are discussed in [27]. The
interesting problems as soon as elastic deformations, compli- results showed that the two operations are comparable and
cated particle geometries or a huge number of particles have the effect of the oversized particles is beneficial in speeding
to be considered. Tremendous improvements in computer up the screening of near-mesh material.
power and computational methodology have accelerated the The influence of some operating variables of separating
pace towards simulation of larger and larger systems, so that sifters has been studied in [26]. Over a certain range of oper-
now simulations of millions of particles are possible. Such ating variables, the screening efficiency of two types of parti-
advances also enable researchers to obtain more information cles over a vibrating screen has been observed. The variables
from their simulations. include the flow rate, deck angle, angular velocity and mesh
Particle separation phenomena are important in granular size. The results show that the separation process is sensi-
media studies. Screeners, classifiers, shakers and separators tive to the operating variables. Many other researchers have
are used in a large number of industrial applications requiring studied the radial and axial segregation process of granu-
separation and classification of powders or other bulk mate- lar mixtures in rotating cylinders, see [1,12,19,32]. In [11]
rials by particle size as well as separation of particles by den- a constitutive model for the radial segregation flux in cas-
sity, magnetic properties or electrical characteristics. These cading layers of rotating cylinders is proposed. There, parti-
machines are divided into different categories such as round cle dynamics and Monte Carlo simulations for steady flows
and rectangular screeners, magnetic separators, electrostatic down an inclined plane are used for studying the density
separators, rotary sifters, wet or concentrating tables, rake segregation.
classifiers, classifying hydrocyclones, floatation systems and Three decades ago, the discrete element method (DEM)
trammels. The operation of particle separation is divided into that describes the motion of particles and models the behavior
two main categories, continuous and batch operations. In of dense solid assemblies in soil mechanics was proposed [6].
continuous operation, the particles are continuously fed into The DEM was adapted in [15] to be used for the analysis of
the separation unit during the whole separation process. This the internal dynamics of tumbling mills. An elastic-perfectly
type of particle separation is usually called ‘screening’. On plastic contact model was used, see [16]. Using this model
the other hand, batch operation is used if the particulate mate- requires only material parameters that may be obtained from
rial is charged only once. This kind of batch separation is standard tests. Some other researches studied volume filling
commonly described by the term ‘sieving’. fractions of the particles charged in the rotating drum of the
Screens can be classified into three groups according to separating machine, see [7]. Numerically the dynamics of
their mode of particle movement, see [31]. These groups are the size segregation process of binary particle mixtures in

123
Particle screening phenomena in an oblique multi-level tumbling reservoir 417

rotating drums is studied. A direct dependency between the ent flow rates as well as batch sieving with a limited number
different particle size ratios and the final amount of the radial of particles are analyzed and compared.
segregation flow was marked in their study.
Analytical works, e.g. [17,18,30], consider particles with
or without cohesive, adhesive, frictional forces between them 2 Description of the tumbler screening machine
during contact. The spring-dashpot model or penalty method
uses different contact parameters which essentially embody The investigated tumbler screening machine (TSM) consists
contact properties, see e.g. [16]. None of these parameters of several main parts, see Fig. 1a.
are typical material properties and hence they are difficult
to determine experimentally. In [13] a numerical model for 1. Driving motor: It runs in different speeds which should
studying a screening separation of granular mixture com- be selected depending on the type of material to be sep-
prising two different sizes of particles on an inclined surface arated.
is presented. A two-dimensional transient model has been 2. Machine foundation: It is heavy enough to hold the
developed to calculate the particle motion on and through machine structure and to isolate the environment against
the screen using the DEM. They also discussed the influence vibrations.
of the feeding rate, the depth of the particle bed and the screen 3. Holder and adjustable plates: The holder plate is hori-
inclination on the screening efficiency, see [14]. Their analyt- zontal and held by the main rotating shaft of the machine.
ical results were compared with some experimental studies. To create the tumbling motion, an adjustable plate is fixed
Screeners, which are in the focus of our research, are sift- over the holder plate with certain tilt angles α and β. The
ing units which are rotated as material is fed into their interior. overall sifting unit is connected to the adjustable plate by
The finer particles should fall through the sieve opening and an auxiliary shaft. The inclination angles are chosen to
oversized particles are ejected out through certain outlets. obtain the best efficiency of the separation process.
Screeners are available in three main types: drum sifter, rect- 4. Rotating shafts: The main shaft is driven by a V-belt
angular deck, and round deck. A rotary round deck separat- connected to the driving electrical motor. This shaft and
ing machine for screening round particles of different sizes is the auxiliary shaft have a certain margin of eccentricity
the main problem of interest in this study. To reach our goal in between and they hold the entire body of the machine
of better understanding the processes, a tumbler screening and transmit the motion to the main sifting unit.
machine has been modelled as a multibody system [22] and 5. Sifting meshes: These separating units consist of a group
investigated in our discrete element simulation program. of potentially different-sized sieves along the successive
The efficiency of the sorting process is determined, e.g., decks of the machine, see Fig. 1b. The undersized parti-
by counting the number of both right-sorted and undesired- cles may pass through the holes. The oversized particles
gangue particles of the whole process at the different levels with some of the undersized ones will be rebound from
of the machine. Different parameters affect the machine per- the mesh plane when they strike the obstructed portion
formance, e.g., the machine speed, the feeding rate of parti- of the screen. Of course, the mesh widths are decreasing
cles, the barrel inclination angles and the shaft eccentricity. from the top to the bottom.
Different simulations are conducted to study the effect of 6. Sifting vessel: The main cylinder is charged with the bulk
these parameters. Continuous feeding processes with differ- material. This vertical rotating vessel can be designed

sifting cylinder
auxiliary shaft

adjustable plate
z̄
z ȳ
x̄
α y β
x

c c holder plate
?

machine foundation
main shaft

(a) Main parts of the machine (b) Internal section

Fig. 1 Tumbler screening machine TSM

123
418 H. Alkhaldi, P. Eberhard

Fig. 2 Stepped and uniform TSM models

Fig. 3 Motion of tumbler screening machine

as a uniform or stepped multi level obliqued barrel, see 3 Operation and machine movement
Fig. 2a. The number of these levels depends directly on
the particles to be separated and on the process. Differ- The machine is modelled as a multibody system [22]. It
ent exits are distributed at each sorting level and located consists basically of the machine itself with different lay-
at the outer periphery around the body of the machine, ers of sifters and the particulate material to be screened and
see Fig. 2b. A wide variety of machine sizes is avail- classified.
able, e.g. produced from Allgaier, Uhingen, Germany.
Diameters are ranging in their machines from 60 to
290 cm. 3.1 Machine movement
7. Feeding container: In the simulation of continuous
screening operation, the TSM is charged by the parti- The basic motion of the tumbler screening machine is gyra-
cle flow through a feeding container. The size of the tory, see Fig. 3. The angular velocity ω is assumed to be
exit nozzle of the container determines the particle flow constant during the separation process. Using a special axi-
rate, see Fig. 2a. The machines are charged from the ally spring-mounted graphic recorder, the spatial movement
top and in the middle of the highest level of the sifting of the machine can be represented on paper for repeatable
unit. adjustments. The recording of the screening action permits

123
Particle screening phenomena in an oblique multi-level tumbling reservoir 419

Fig. 4 Flowchart of the start

segregation process
within the reflect by
eject outside
yes area of no applying
the machine
fall down exits? forces
through the yes
mesh holes
yes

touch with contact with contact with

yes mesh no surrounding no ceiling
surface? walls? surface?

yes
no
particle reflect up
undersized
center over yes no from the
particle?
a hole? mesh surface stay inside
the layer
no

the optimum operating data to be reproduced, which will 4 Particle modelling

assist quality assurance.
4.1 Particle-to-particle contact
3.2 Particle movement
In a spatial multibody system, each free body has six degrees
The TSM can be used for materials with particle sizes of freedom in translation and rotation, see [22,24]. In order
between micrometers and millimeters, such as those used in to analyze the system, the discrete element method is used,
the chemical, pharmaceutical, foodstuffs and plastics indus- see [21]. In this method, the motion of each single particle
tries. The complicated three-dimensional screening motion is considered individually. Particle positions, velocities and
of the machine is mainly independent of the material load. orientations are computed at each time step of simulation.
This screening motion is adjustable to improve the separation The equations of motion of the rigid body i in the multibody
process. Experiments are necessary to find optimal process system is governed by six differential equations
parameters. As the screening material is continuously fed m i ai = Fi , Ii α i = Mi , (1)
into the center of the top screen, it is distributed over the
entire screening surface from the center towards the periph- where m i and Ii are the mass and the inertia tensor of particle
ery. The particles will often travel in a spiral motion due to i, Fi and Mi are the force and torque vectors, ai and α i are
the combination of the tangential and radial inclination of the linear and angular accelerations. Considering two bodies
the screen and to the friction between the particles and the i and j in an N particle system, see Fig. 5a, the force Fi and
surface area of the mesh, see [29]. torque Mi acting on particle i can be calculated as
In the center, many fine particles will pass through the


N
mesh holes. Towards the outside, the horizontal and vertical Fi = Fi j + m i g, (2)
accelerations are increasing, causing the particles with near j=1, j
=i
mesh size to be separated. The remaining oversized particles


N

N
are carried to the outlets where the flow is influenced by an Mi = Mi j = ri × Fi j , i = 1, . . . , N ,
adjustable deflector. This deflector is another factor in con- j=1, j
=i j=1, j
=i
trolling the residence time of material on the screen. This
(3)
process is repeated on every screen deck. After the granular
materials have been tumbled in the rotating vessel and then where ri is the vector from the center to a point on the surface
segregated depending on their size, the particles should be of particle i and g is the gravity vector.
classified by ejecting them out through outlets located in the The contact calculations are based on the soft-particle
walls. The particle segregation flowchart, see Fig. 4, clari- model which leads to a deterministic simulation where the
fies the particle transportation between the different layers state of each particle in the system and all particle inter-
of the rotating sifting unit, particle movement, reflections, actions are determined using physical laws. Applying the
falling and sorting of the oversized and undersized types of penalty method, we can determine the normal and frictional
particles inside the tumbling machine during the separation forces between the colliding particles, see Fig. 5b. This model
process. assumes that the contact forces result from an unphysical

123
420 H. Alkhaldi, P. Eberhard

Fig. 5 Particles in overlap

(a) Vector description of particle (b) Spring-damper model

interaction

overlap between the bodies in contact, see [2,3]. As a result 4.2 Particle-to-mesh contact
of collisions, the particles will dissipate kinetic energy due
to the normal and frictional contact losses. In order to derive the mathematical formulation of the system,
The normal force Fi j between two colliding particles i an inertial system K g {O, x, y, z} and a rotating coordinate
and j can be formulated as system K l { Ō, x̄, ȳ, z̄} are defined. The body fixed coordi-
nate system K l is rotating with the tumbling cylinder with its
origin located at the base of the auxiliary shaft on the inclined
Fi j = (k p δi j + c p δ̇i j )n. (4)
adjustable plate, see Fig. 6. To check the contacts and cal-
culate forces, the particle positions and velocities should be
Substitution of Fi j in Eqs. (2) and (3) will yield the resul- expressed in this coordinate system.
tant contact force and torque acting on particle i due to the
particle–particle interaction, where

qi − q j 4.2.1 Location and particles velocity

δi j = (ri + r j ) − (qi − q j )T n, n = , (5)
|qi − q j |
Since we write the particle position qi and the particle veloc-
k p , c p are the spring stiffness and damping coefficient of the ity vi in the rotating coordinate system K l , it is required to
penalty approach, n is the normal unit vector between the two derive the translation vector and the rotation matrix based on
particles, δi j and δ̇i j are the overlap and the relative veloc- the geometrical parameters of the inclined plate and on the
ity between the two colliding particles i and j. The system angular velocity of the machine. The translation vector roō,g
of forces (Fi , Mi ) which acts on each individual particle is from point O to point Ō given in the coordinate system K g
calculated in the global coordinate system. is given as

Fig. 6 Particles in global and

local coordinate systems

(a) Vector description (b) Particle contact models

123
Particle screening phenomena in an oblique multi-level tumbling reservoir 421

⎡ ⎤ 4.2.2 Contact forces with the mesh

u cos(ωt) − u cos(ωt) cos β + e sin(ωt) −  sin(ωt)+
⎢ u sin(ωt) sin α sin β − e sin(ωt) cos α ⎥
⎢ ⎥
roō,g =⎢ u sin(ωt) − u sin(ωt) cos β − e cos(ωt) +  cos(ωt)− ⎥, The particle transportation through the mesh depends directly
⎣ u cos(ωt) sin α sin β + e cos(ωt) cos α ⎦
u sin β cos α + e sin α
on the position of the specified particle with respect to the
(6) hole boundaries. In reality, the mesh is built of a combina-
tion of intersecting wires over which the particles will have
where u and e are two geometrical dimensions related to
a contact. It happens that the particle contacts with the mesh
the adjustable inclined plate,  is the eccentricity between
wires and is reflected in different directions around the hole.
the fixed main shaft and the rotating auxiliary one and ω
This will depend on the exact point on the wire where the
is the angular velocity of the tumbling screening machine
particle hits. Taking into consideration all those precise con-
around a vertical axis of rotation. The rotation matrix Alg
tacts with the mesh wires will lead to many calculations and
between the coordinate systems can be expressed as
⎡ ⎤
cos(ωt) cos β − sin(ωt) sin α sin β sin(ωt) cos β + cos(ωt) sin α sin β − cos α sin β
Alg = ⎣ − sin(ωt) cos α cos(ωt) cos α sin α ⎦, (7)
cos(ωt) sin β + sin(ωt) sin α cos β sin(ωt) sin β − cos(ωt) sin α cos β cos α cos β

where α and β are the two inclination angles of the adjustable infeasible computational efforts. In order to reduce these
plate of the machine. The position of any particle i inside the computations while still reaching physical results, another
rotating vessel, see Fig. 6a, can be expressed by less computationally expensive approach is suggested. This
approach depends on defining a reduced mesh hole which
is only used in simulation, to consider all contact possibili-
qi,g = qi,g − roō,g or qi,l = Alg (qi,g − roō,g ). (8)
ties and adjust the number of passed and reflected particles
in a realistic way. This method assumes that the undersized
Similarly, the particle velocity can be expressed relative particles pass through the mesh if they are small enough
to the rotating coordinate system by differentiating Eq. (8) and their centers are over the reduced hole, otherwise they
once with respect to time are reflected. The mesh thickness is set to zero. The reduc-
tion in the mesh hole diameter can be adjusted to agree
q̇i,l = Alg (q̇i,g − ṙoō,g ) + Ȧlg (qi,g − roō,g ). (9) with experimental observations which are currently under
investigation.
Substituting vi = q̇i and vi = q̇i in Eq. (9) for particle The direction of the velocity vector of the particles has
velocity we get without doubt a major effect on the reflected particles and
then on the number of the particles which will pass down
through the mesh holes. The direction of the velocity vector
vi,l = Alg (vi,g − ṙoō,g ) + Ȧlg (qi,g − roō,g ). (10)
will determine the exact point where the particle will touch
the plane of the mesh. A reduced hole parameter is intro-
The terms ṙoō,g and Ȧlg in Eq. (10) represent the rate of duced in our simulation as a probability factor to adjust the
change of the translation vector and the rotation matrix with number of reflected and passed particles according to their
respect to time. By direct differentiation of Eqs. (6) and (7), sizes and center positions to the reduced hole. This adjust-
respectively, these rates are ment basically follows from experimental observations and
⎡ ⎤
−u sin(ωt) + u sin(ωt) cos β + e cos(ωt) −  cos(ωt)+
⎢ u cos(ωt) sin α sin β − e cos(ωt) cos α ⎥
⎢ ⎥
ṙoō,g =ω ⎢
⎢ u cos(ωt) − u cos(ωt) cos β + e sin(ωt) −  sin(ωt)+ ⎥ ⎥ (11)
⎣ u sin(ωt) sin α sin β − e sin(ωt) cos α ⎦
0

and
⎡ ⎤
− sin(ωt) cos β − cos(ωt) sin α sin β cos(ωt) cos β − sin(ωt) sin α sin β 0
Ȧlg = ω ⎣ − cos(ωt) cos α − sin(ωt) cos α 0⎦. (12)
− sin(ωt) sin β + cos(ωt) sin α cos β cos(ωt) sin β + sin(ωt) sin α cos β 0

123
422 H. Alkhaldi, P. Eberhard

this simplified approach is required to keep simulation time

feasible.
In case of friction, the approaching angle of contact
between the particle and the mesh is different from the
reflection angle over the surface of the mesh. The direction
of the particle velocity with respect to the machine at the
instant of contact determines the direction of the frictional
force in the tangential plane of the mesh surface. Therefore,
the reflection angle is determined according to the normal
and frictional force components between the particle and the
mesh plane.
The contact forces with the machine walls and the ceiling
are computed analoguely to the particle–particle forces. More
interesting is the contact with the mesh surface. For practical
purposes, sifters can be designed in many shapes according
to the desired holes having e.g. circular, square or polygonal
shapes. Here, we are interested in square-patterned meshes.
The number of holes of barrel k in a square mesh depends
directly on the mesh radius Rk , the width of the holes 2w and Fig. 8 Contact detection with mesh surface
width of the solid part g between the different rows of the
holes, see Fig. 7,
will fall through the mesh if it is small enough. At any level k,
Rk consider the points a and b on the borders of two successive
mk = + 1, (13) holes on the mesh surface, see Fig. 8,
2wk + gk
x̄a = w + j (2w + g), x̄b = x̄a + g,
where m k is the number of holes along the radius of the mesh (14)
plate. Since the radial number of holes is always an integer, j = −(n k − 1), . . . , j = (n k − 2),
m k should be rounded to the nearest integer n k .
where j is an integer counter over the number of the gap
The mesh plate consists of (2n k −1) aisles of square holes
aisles in the mesh surface along the x̄ direction. To check the
in the x̄ direction and the same number of aisles holes in the
particle position with respect to the mesh holes in x̄ direction,
ȳ direction. It also consists of (2n k − 2) aisles of gaps and
a flag integer cx is defined as
the identical number of aisles in both x̄ and ȳ directions,
respectively. 
0 if (x̄a − b) ≤ q x̄ ≤ (x̄b + b),
The particle–mesh contact should be detected in both cx = (15)
1 else,
directions x̄ and ȳ. The particle will be considered to be
reflected if it is in contact with the mesh surface and its center where b is the adjustable hole clearance. This parameter is
is over the solid part and not over a reduced hole, otherwise it defined to be used as a simulation parameter to adjust the size
of the mesh openings within certain limits, i.e. 0 ≤ b < w.
Reduced holes with adjustable clearance b help in controlling
the falling particles through the mesh holes during the simula-
tion process. The adjustable parameter b is selected to adjust
the simulation results in order to be in a good agreement with
experimental observations which allow predictions for fur-
ther computations. Experimental investigations are currently
in progress using glass balls and will be compared in the near
future. To run the simulations and check our codes, always
a value of b = 0.8w is used in our study. It will be most
interesting for the future comparisons to see how this value
must be chosen correctly.
In the same way, the ȳ direction is investigated yielding
c y . In order to determine the contact forces with the mesh
surface, it is checked if the particle is in direct contact with
Fig. 7 Geometrical description of the square mesh the solid part of the mesh or if it is over a hole

123
Particle screening phenomena in an oblique multi-level tumbling reservoir 423


1 particle potentially falls down, 4.3 Numerical time integration
cx c y = (16)
0 particle is reflected.
To solve the differential equations numerical time integra-
All oversized particles will definitely reflect up from the mesh tion is necessary. To calculate the trajectories of particles and
surface. Some of the undersized particles will reflect also determine their new positions and orientations, different inte-
upward when they hit the solid part of the mesh and not a grators can be used. Algorithms developed by Verlet in 1967
hole, otherwise they will fall down. are among the most popular in molecular dynamics. Verlet
The friction force affects the particles and tries to reduce integration, velocity-Verlet and leapfrog-Verlet are different
their velocities. Here, the friction force between the parti- versions of Verlet integrators, see [4,17]. These integrators
cle and the mesh surface is usually in the direction of their offer better stability than the simpler Euler integration meth-
relative velocity. The tangential component of the relative ods with sufficient accuracy and stability. The new position
velocity of particle j in the mesh plane can be written as and orientation of particle i at the time step (m + 1) are

v̄t j = v 2x̄ j + v 2ȳ j . (17) Yim+1 = 2Yim − Yim−1 + h 2 Ÿim , (22)

where Yi denotes the generalized coordinates of the body i

The unit vector of tangential velocity can be expressed for and h is the integration time step.
v̄t j
= 0 as As a summary, Fig. 9 presents the DEM procedure. The
⎡ ⎤ ⎡ ⎤ starting conditions are the initial positions of particles and
tx̄ v their initial velocities. Using DEM and soft particle con-
1 ⎣ x̄ ⎦
t j = ⎣ t ȳ ⎦ = v ȳ . (18) tact, the contact forces and moments are calculated. Integrat-
v̄t j
0 j 0 j ing the accelerations will give the new velocities, positions
and orientations in the next time step. This cycle has to be
The force between particle j and the mesh surface in the repeated for all time steps.
normal direction to the mesh plane can be written as
5 Simulation results
N j = N j n = | f¯z̄ j |n, (19)
Different simulations have been performed to investigate
where N j and n are the normal contact force and normal unit screening phenomena in the rotating tumbling machine. Par-
vector to the mesh plane, respectively. Using Coulomb’s law, ticle distribution and sifting rates of the separated particles
the friction force is have been studied. Screening phenomena are very sensitive to
⎡ ⎤ the machine parameters, e.g., plate inclinations, shaft eccen-
−µd | f¯z̄ j | v x̄ tricities and size of the mesh openings and rotational speed
F f j = −µd N j t j = ⎣ v ȳ ⎦
v̄t j of the machine.
0 j
start
⎡ ⎤ (20)
v x̄
−µd
|kw δm j − cw v z̄ j | ⎣ v ȳ ⎦ ,
initialize simulation environment,
=
v̄t j TSM and particle states
0 j
find forces and moments acting
where µd is the dynamic coefficient of friction between par-
on each particle in the system
ticles and the mesh surface, δm is the mesh overlap, kw and cw
are the elastic spring stiffness and the viscous damping coeffi- determine the linear and angular
cient with the mesh plane, respectively. The overall contact acceleration of each particle
m no
force F p−w between particle j and the mesh surface is then integrate the equations of motion to find the new
velocities, positions and orientations of the particles
⎡ −µd ⎤
v x̄ make measurements and record states
⎢ v̄t ⎥
m ⎢ −µd ⎥ in the output files
F p−w = |kw δm j − cw v z̄ j | ⎢ v ȳ ⎥ , (21)
⎣ v̄t ⎦
SIGN(kw δm − cw v z̄ ) j end ? yes stop

where the function SIGN(q) is 1 for q>0 and otherwise −1. Fig. 9 Flowchart of DEM

123
424 H. Alkhaldi, P. Eberhard

Fig. 10 Continuous and batch screening using a TSM

Due to the complicated motions of the particles over the M is the total number of barrels in the machine where i = 1
screen surface and the various factors that influence such is the lowest level and i = M the highest one. Since there
motion, the understanding of the actual mechanisms involved are no gangue particles in the lowest level, we have g1 = 0,
in screening is still in its infancy. Attempts in the past to therefore Eq. (23) can be written as
describe the performance of screening processes mathemati-
cally have adopted either a probabilistic approach or a kinetic
M
100
approach [10]. Most of the attempts have been limited to c= 1+ ci . (24)
M
defining the frequency and amplitude of vibration, and pos- i=2
sibly the direction of motion of the screen surface [28].
In our work, the particles have been tested for both batch The machine efficiency c would be 100% if the number
and continuous feeding in a uniform and a stepped multi level of gangue particles is gi = 0 in the different layers of the
obliqued vertical cylinder. For continuous screening, a pack- machine. Although some decks of the machine are more effi-
age of 3,483 particles has been tested and the efficiency of cient than others during the sorting process, it is difficult to
machine has been recorded for different machine conditions, find a clear prediction explaining this disparity. The weight
see Fig. 10a. Our simulation codes have also been tested and size ratio of the different particles in the mixture, the
for more number of particles which lead to a much compu- material properties and the operation conditions influence
tational effort and more simulation time. These simulations this phenomenon.
are observed carefully through animations which show a nice To study the effect of machine parameters on the sort-
and acceptable separation process over the different layers ing efficiency, an environment of 3,483 round particles have
of the machine, see Fig. 10b, c. The finer particles fall fre- been simulated for the presented investigations. Larger par-
quently through the sieve openings while the oversized parti- ticle numbers can be investigated but the focus is first put
cles should be ejected through certain outlets located around to the basic mechanisms and relations. These particles con-
the machine body. Some of the undersized particles are unin- sists of a mixture of five different sizes with their numbers
tentionally forced to travel outside the machine together with and radii as: (1,210, 10 mm), (832, 12 mm), (611, 14 mm),
the sorted ones. Those unavoidable and also undesired par- (467, 16 mm) and (363, 18 mm). Four squared-pattern sieves
ticles, which we call here “gangue particles” reduce the effi- of hole dimensions w = {11, 13, 15, 17} mm are used in a
ciency of the machine. For an efficient separation process, the uniform-radius TSM with R = 80 cm.
selection of proper operating conditions reduces the number
of these gangue particles. 5.1 Influence of machine speed
Mathematically, to measure machine efficiency, the char-
acteristic value c can be expressed as The tumbling screening machine has the ability to run in
a range of angular velocities. This angular speed which is
100

M
si assumed as being constant during the sorting operation pro-
c= ci , ci = , (23)
M si + gi cess, has a great influence on the sifting rate of the mixed
i=1
particles at each of the different sorting levels of the machine.
where ci is the individual efficiency of layer i of the machine, Using computer simulation enables us to change the machine
si and gi are the number of the oversized sorted and the under- speed over a wide range in order to study its effect on the
sized gangue particles at level i of the machine, respectively, machine performance.

123
Particle screening phenomena in an oblique multi-level tumbling reservoir 425

Fig. 11 Influence of rotational 140 500

level 2, gangue level 2, gangue
speed on the particle sifting level 3, gangue level 3, gangue
rates at different sorting levels 120
level 4, gangue 400 level 4, gangue
of the machine in continuous level 5, gangue level 5, gangue

number of particles

number of particles
100
screening operation, α = 1◦ , level 1, sorted
level 2, sorted
level 1, sorted
level 2, sorted
β = 0.65◦ , w/g = 1, b = 0.8w 80 level 3, sorted 300 level 3, sorted
and feeding rate 147 particles/s level 4, sorted level 4, sorted
60 level 5, sorted level 5, sorted
200
40
100
20

0 0
0 5 10 15 0 5 10 15
time (s) time (s)
(a) w = 1rpm (b) w = 50rpm

350 500
level 2, gangue 1 rpm
level 3, gangue 30 rpm
300

number of sorted particles

level 4, gangue 400 40 rpm
level 5, gangue 50 rpm
number of particles

250 level 1, sorted 70 rpm

level 2, sorted 100 rpm
200 level 3, sorted 300 150 rpm
level 4, sorted 200 rpm
150 level 5, sorted 250 rpm
200
100
100
50

0 0
0 5 10 15 0 5 10 15
time (s) time (s)
(c) w = 150rpm (d) lowest sorting level

The angular velocity ω should not be too high in order c = 72%, see Fig. 11c. Going further, the very high velocity
to avoid too fast motion of particles. Too high velocities of of 250 rpm will decrease the efficiency again to c = 58%. It
the particles will decrease the separation rate of the mixture is observed that the intermediate speed of 50 rpm gives the
and reduce the chance for undersized particles to fall down best sifting rates with maximum efficiency about c = 78%
through the mesh openings. On the other side the rotational during the continuous screening process, see Fig. 12. From
speed should not be too slow since low separation rates this figure, it is also clear that for all machine speeds the effi-
and a big number of undesired gangue particles can then be ciency starts low and approaches its steady operating mode
expected. This is due to the fact that the particles experience in about 10–20 s.
not enough mixing due to their small velocities. This insuffi-
cient mixing reduces the chance of the undersized particles
to fall through the mesh, see Fig. 11. 5.2 Influence of feeding rate
In order to reach the best rate of particle separation, we
have to find the rotational speed of the machine which max- We next study the effect of the feeding rate of the mixed
imizes the number of the sorted particles and minimizes the particulate material on the machine efficiency. Increasing the
gangue ones. In order to do this, we simulate our problem size of the output nozzle of the feeding container will increase
for different angular velocities and measure the efficiency of the feeding rate of the particle flow. Since the number of parti-
each set of geometrical, contact and material parameters. It is cles in the feeding container is limited, the feeding rate of the
observed that the sorting under very low rotational velocity particles through a relatively large orifice will not be kept con-
leads to very bad separation rates. The main reason for this is stant through the whole simulation process. In this case, most
that the mesh is blocked soon. This would be a disastereous of the particles in the feeding container will fall down contin-
situation for an industrial process. The number of undesired uously and in a constant rate just only in the beginning of the
gangue particles are even much higher than those of desired simulation, see Fig. 13a. For small orifice size, e.g. 30 mm,
sorted ones in most of the machine levels, see Fig. 11a. Low the flow rate will stay constant at 147 particle/s over the
efficiency of c = 40% is recorded for this case. whole simulation since there is enough material in the feed-
On the other hand, a relatively high velocity of 150 rpm ing cylinder which keeps the particles flow through the feed-
leads to better results and higher efficiency of about ing nozzle constant after a few seconds. This constant flow

123
426 H. Alkhaldi, P. Eberhard

Fig. 12 Efficiency of TSM 80 80

1 rpm
machine for different speeds in 30 rpm
the continuous screening 70 40 rpm
operation mode, α = 1◦ , 50 rpm 70
β = 0.65◦ , w/g = 1, b = 0.8w

efficiency c (%)

efficiency c (%)
70 rpm
60 100 rpm
and and feeding rate 147 150 rpm
particles/s 50 200 rpm 60
250 rpm

40
50
30

20 40
0 5 10 15 0 50 100 150 200 250
time (s) speed (rpm)

Fig. 13 Influence of the

number of particles through orifice

4000 700
feeding rate on the sorted 200mm level 2, gangue
30mm level 3, gangue
number of particles at different 600 level 4, gangue
35mm

number of particles
sorting levels of TSM machine 3000 45mm level 5, gangue
500 level 1, sorted
in batch sieving and continuous level 2, sorted
screening operation modes, 400 level 3, sorted
α = 1◦ , β = 0.65◦ , w/g = 1, 2000 level 4, sorted
level 5, sorted
b = 0.8w and ω = 50 rpm 300

ca. 147 particle/s 200

1000
100

0 0
0 5 10 15 0 2 4 6 8 10
time (s) time (s)
(a) flow rate for different nozzle radii (b) nozzle radius 200mm
700 700
level 2, gangue 30 mm
level 3, gangue
number of sorted particles

45 mm
600 level 4, gangue 600 60 mm
number of particles

level 5, gangue 100 mm

500 level 1, sorted 500 200 mm
level 2, sorted batch
400 level 3, sorted 400
level 4, sorted
300 level 5, sorted 300

200 200

100 100

0 0
0 2 4 6 8 0 5 10 15
time (s) time (s)
(c) batch sieving (d) lowest sorting level

rate is used in our simulation study for continuous screening In case of the batch sieving, 3,483 particles have been
operations. tested. The machine was started running with a big number
Continuous screening and batch sieving are studied. In of particles in the very beginning of the operating process.
continuous screening, it is observed that low feeding rates Those particles are fed immediately inside the machine bar-
with nozzle radius of 30 mm will lead to a low screening rel. Due to these heavy packed particles and the partially
performance. There, the sorted particles travel slowly to the blocked mesh, it is observed that the gangue particles will
machine exits together with some of the undesired gangue rapidly accumulate in the beginning of the simulation. Many
ones, see Fig. 11b. The rate of the sorted particles is observed collisions between different particles will happen and force
to improve with increasing feeding rate. Opening the nozzle many of these particles to travel rapidly through the machine
output further will increase the feeding rate of the particle exits without even touching the mesh. It is not surprising
flow and improve the screening efficiency of the machine that one of the most dangerous situations is blocking of the
during the constant rate feeding period, see Fig. 13b. mesh and one has to ensure that the feeding rate is not so

123
Particle screening phenomena in an oblique multi-level tumbling reservoir 427

Fig. 14 Efficiency of TSM 80 76

machine for different openings 30 mm continuous screening
45 mm 74 batch sieving
of the feeding nozzle in batch 70 60 mm
sieving and continuous 100 mm 72

efficiency c (%)

efficiency c (%)
screening operation modes, 200 mm 70
60
α = 1◦ , β = 0.65◦ , w/g = 1, batch
68
b = 0.8w and ω = 50 rpm 50
66
40 64
62
30
60
20 58
0 2 4 6 8 0 50 100 150 200
time (s) feeding nozzle radius (mm)

high that this can happen. As the time proceeds, the rate ciency of c = 76% will be obtained when β = 0.65◦ , see
of the gangue particles will decrease to be steady after 5 s Fig. 15b.
simulation time, see Fig. 13c. Fewer gangue particles will The inclination angles should be selected to achieve the
then travel outside the machine exits since most of the par- best separation rates and the maximum performance of the
ticles are already separated at the different levels inside the machine. For continuous screening of 147 particles/s, speed
machine. In the meantime, only the right-sized-particles will of 50 rpm and our setup it was followed that α = 1◦ and
be sorted and directly run away through the machine exits, β = 0.65◦ are the optimal values for a maximum efficiency
which improves the performance of the separation process. of c = 78.2% of the TSM machine, see Fig. 15d.
As an indication of the influence of the feeding rates on
the machine efficiency, the number of sorted particles in the 5.4 Influence of shaft eccentricity
lowest level of the sifting unit is recorded, see Fig. 13d.
The lowest-level-particles are usually opposed to many col- The eccentricity between the main vertical and the inclined
lisions and face different obstacles during their way from holder shafts of the machine has a clear influence on the
the top until they reach their final destination in the lowest efficiency of the machine. In order to get a quantitative idea
deck of the machine. It is observed that very low feeding about this influence, the number of sorted particles in the
rates are not recommended in the case of continuous screen- lowest sorting level of the machine is recorded. It is found
ing, see Fig. 14. Increasing the rate of the feeding material that small eccentricities will often be better than to work
will increase the machine efficiency. Although batch siev- with large shaft eccentricities. High range eccentricities of
ing might reveal better sorting efficiency compared to the  = 300 mm (which are technologically infeasible) show a
continuous screening as long as no blocking occurs, it is small number of 200 sorted particles in comparable with the
not commonly used in industrial processes and screening number of about 500 particles obtained using small range of
technology since most processes have to run without eccentricities, see Fig. 16a. It is clear that a zero eccentric-
interruption for a long time. ity as well as a large eccentricity reveals bad performance,
There exist detailed investigations about practical feed- while a value in between of  = 40 mm leads to a maximum
ing rates for certain nozzle openings, see e.g. [5]. Since this screening performance of about 80%, see Fig. 16b.
depends highly on the used material and many more aspects,
it would require detailed investigations to carefully verify
this by our simplified simulation model.
6 Conclusions

In this paper, the particle screening phenomenon over a rotary

5.3 Influence of inclination angles tumbling screen is studied. The discrete element method is
applied and used as a simulation tool for the separation pro-
The inclination angles α and β of the sifting unit have a great cess. This method proves its ability to be a powerful numer-
influence on the machine performance. To study their effect, ical modeling tool for solving problems in granular media.
we fix one of them and change the other. It is observed that For specific geometrical and contact parameters particles
the steady state of machine efficiency will be reached after transportation, sifting rates and machine efficiency are
13 s of simulation time, see Fig. 15a, c. For an inclination recorded. Particles are simulated in uniform and stepped
condition of α = 0◦ , it is noticed that increasing the angle β models of tumbling cylinders. For both continuous screening
will improve the machine performance. The maximum effi- and batch sieving, it was found that the segregation process

123
428 H. Alkhaldi, P. Eberhard

Fig. 15 Efficiency of TSM 80 77

machine for different inclination β=0
β=0.65
angles α and β of the sifting 70 76
β=1
unit, w/g = 1, b = 0.8w, β=2

efficiency c (%)

efficiency c (%)
ω = 50 rpm and feeding rate 60 75
147 particles/s
50 74

40 73

30 72

20 71
0 5 10 15 0 0.5 1 1.5 2
time (s) β (deg)
(a) a = 0° (b) a = 0°

80 79
α=0
α=0.5 78
70
α=1
α=1.25
efficiency c (%)

efficiency c (%)
60 α=2 77

50 76

40 75

30 74

20 73
0 5 10 15 0 0.5 1 1.5 2
time (s) α (deg)
(c) b = 0.65° (d) b = 0.65°

Fig. 16 Influence of the shaft 500 80

eccentricity on the sorted ε=0
number of sorted particles

number of particles and on the ε=30

400 ε=50 75
machine efficiency, α = 1◦ ,
efficiency c (%)

ε=100
β = 1◦ , w/g = 1, b = 0.8w, ε=150
ω = 50 rpm and feeding rate 300 ε=300
70
147 particles/s
200 65

100 60

0 55
0 5 10 15 0 50 100 150 200 250 300
time (s) ε (mm)
(a) lowest sorting level (b) overall efficiency

is very sensitive to the rotational speed of the machine. This same operating conditions, batch sieving shows better results
speed should be selected to be within certain limits to maxi- compared to continuous screening.
mize the number of sorted particles and to improve the sifting Improving the accuracy of the simulation requires to be
rates for the different machine levels. Too high and too low more realistic in implementing the contact forces and the
speeds will lead to a bad screening performance. associated contact parameters of the dynamical system of the
Furthermore, the particle feeding rates, inclination angles granular system. Physical contacts inside the TSM require
and shaft eccentricity have a great influence on the machine some more detailed investigations. These parameters can be
efficiency. Small angles between 0.5◦ and 1◦ and eccen- obtained from special experiments. For better understand-
tricities between 25 to 50 mm are recommended. For the ing of the particle sorting and transportation between the

123
Particle screening phenomena in an oblique multi-level tumbling reservoir 429

different layers of the machine, experimental studies have to 16. Mishra, B., Thornton, C.: An improved contact model for ball mill
be performed. simulation by the discrete element method. Adv. Powder Tech-
nol. 13(1), 25–41 (2002)
17. Muth, B.: Simulation von Kontaktvorgängen einfacher Körper mit
References Methoden der Molekulardynamik (in German), DIPL-87, Institute
B of Mechanics, University of Stuttgart, Germany (2001)
1. Alexander, A., Shinbrot, T., Muzzio, F.: Granular segregation in 18. Muth, B., Müller, M., Eberhard, P., Luding, S.: Contacts between
the double-cone blender: transitions and mechanisms. Phys. Flu- many bodies. Mach. Dyn. Probl. 28(1), 101–114 (2004)
ids 13(3), 578–587 (2001) 19. Nakagawa, M.: Axial segregation of granular flows in a horizontal
2. Alkhaldi, H., Eberhard, P.: Computation of screening phenomena rotating cylinder. Chem. Eng. Sci. 49(15), 2540–2544 (1994)
in a vertical tumbling cylinder. In: Proceedings in Applied Math- 20. Rhodes, M.: Introduction to Particle Technology. Wiley-Inter-
ematics and Mechanics (PAMM) (2006, submitted) science, Chichester (2005)
3. Alkhaldi, H., Eberhard, P.: Efficient computation of colliding parti- 21. Sawley, L., Cleary, W.: Parallel Discrete-Element Method for
cles in a vertical tumbling sorting machine. In: Proceedings Second Industrial Granular Flow Simulation. Csiro Mathematical & Infor-
International Congress on Computational Mechanics and Simula- mation Sciences, Clayton, Australia (1999)
tion (ICCMS), Guwahati, India (2006) 22. Schiehlen, W., Eberhard, P.: Technische Dynamik (in German).
4. Allen, M., Tildesley, D.: Computer Simulation of Liquids. Claren- Teubner, Stuttgart (2004)
don, Oxford (1987) 23. Schinner, A.: Fast algorithms for the simulation of polygonal par-
5. Beverloo, W., Leniger, H., Velde, J.: The flow of granular solids ticles. Granular Matter 2, 35–43 (1999)
through orifices. Chem. Eng. Sci. 15, 260–269 (1961) 24. Shabana, A.: Dynamics of Multibody Systems. Wiley-Inter-
6. Cundall, P., Strack, O.: A discrete numerical model for granular science, New York (1989)
assemblies. Geotechnique 29, 47–65 (1979) 25. Standish, N.: The kinetics of batch sieving. Powder Tech-
7. Dury, C., Ristow, G.: Competition of mixing and segregation in nol. 41, 57–67 (1985)
rotating cylinders. Phys. Fluids 11(6), 1387–1394 (1999) 26. Standish, N., Bharadwaj, A., Hariri-Akbari, G.: A study of the
8. Jansen, M., Glastonbury, J.: The size separation of particles by effect of operating variables on the efficiency of a vibrating
screening. Powder Technol. 1, 334–343 (1967) screen. Powder Technol. 48, 161–172 (1986)
9. Kaye, B., Robb, N.: An algorithm for deducing an effective sieve 27. Standish, N., Meta, I.: Some kinetic aspects of continuous screen-
residue from the rate of powder passage through a sieve. Powder ing. Powder Technol. 41, 165–171 (1985)
Technol. 24, 125–128 (1979) 28. Subasinghe, G., Schaap, W., Kelly, E.: Modelling the screen-
10. Kelly, E., Spottiswood, D.: Introduction to Mineral Processing. ing process: a probabilistic approach. Powder Technol. 59, 37–
Wiley-Interscience, New York (1999) 44 (1989)
11. Khakhar, D., McCarthy, J., Ottino, M.: Radial segregation of 29. Takahashi, Y., Kataoka, M., Uekusa, M., Terumichi, Y.: Behavior
granular mixtures in rotating cylinders. Phys. Fluids 9(12), 3600– of three kinds of particles in rotary barrel with planetary rota-
3614 (1997) tion. Multibody Syst. Dyn. 13(2), 195–209 (2005)
12. Lee, J., Ladd, A.: Axial segregation of a settling suspension in a 30. Tsuji, Y., Tanaka, T., Ishida, T.: Lagrangian numerical simulation
rotating cylinder. Phys. Rev. Lett. 95, 1–4 (2005) of plug flow of cohesionless particles in a horizontal pipe. Powder
13. Li, J., Webb, C., Pandiella, S., Campbell, G.: A numerical simula- Technol. 71, 239–250 (1991)
tion of separation of crop seeds by screening-effect of particle bed 31. Wessel, J.: Siebmaschinen (in German). Aufbereitungstechnik
depth. Inst. Chem. Eng. IChemE Part C 80, 109–117 (2002) 2, 449–456 (1963)
14. Li, J., Webb, C., Pandiella, S., Campbell, G.: Discrete particle 32. Yamane, K., Nakagawa, M., Tanaka, T., Tsuji, Y.: Steady
motion on sieves—a numerical study using the DEM simula- particulate flows in a horizontal rotating cylinder. Phys. Flu-
tion. Powder Technol. 133, 190–202 (2003) ids. 10(6), 1419–1427 (1998)
15. Mishra, B., Rajamani, R.: The discrete element method for the
simulation of ball mills. Appl. Math. Modell. 16, 598–604 (1992)

123

View publication stats