Powder Technology 266 (2014) 240–248

Contents lists available at ScienceDirect

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

DEM/CFD analysis of granular flow in static mixers

Aca Jovanović a, Milada Pezo b, Lato Pezo a,⁎, Ljubinko Lević c
a
Institute of General and Physical Chemistry, University of Belgrade, Studentski Trg 12-16, 11000 Belgrade, Serbia
b
Laboratory for Thermal Engineering and Energy, Institute of Nuclear Sciences “Vinča”, University of Belgrade, P.O. Box 522, 11001 Belgrade, Serbia
c
Faculty of Technology, University of Novi Sad, Bulevar Cara Lazara 1, 21000 Novi Sad, Serbia

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

Article history: Mixing of powders, particles and/or granules obtains an economic importance in various technologies, and in the
Received 9 April 2014 food, pharmaceutical and chemical industries. In many cases, the mixing process greatly influences the quality of
Received in revised form 9 June 2014 the final product. The parameters of the mixing process and the design of equipment have a strong impact on the
Accepted 14 June 2014
mixing efficiency, and the quality and the price of the intermediate or final product. In this study, the discrete el-
Available online 24 June 2014
ement method (DEM) is used for modeling of granular flow in various multiple Komax and Ross mixing applica-
Keywords:
tions. Computational fluid dynamic (CFD) method was used for modeling of fluid flow through the Eulerian
DEM/CFD multiphase model. Coupling of these two methods provides reliable, sufficiently correct and adequate results
Static mixer of the proposed model compared to experimental measurements. The objective of this paper is to predict the be-
Komax havior of particles in different static mixer configurations. The results of the numerical simulation are compared
Ross with appropriate experimental results. Application of this model provides the optimization of the geometry and
Particle tracking parameters of mixing systems taking into account the quality of the mixing process and the cost of the final
product.
© 2014 Elsevier B.V. All rights reserved.

1. Introduction commercial availability [6–8]. There are more then 30 commercial

models currently available.
Static mixers are low energy consumption and efficient mixing de- A detailed review on static mixers, concerning the mechanisms, ap-
vices that can handle a wide range of applications. The static mixer plications and characterization methods focusing on mixing and mass
uses energy from flow to produce the desired mixing results for the pro- transfer performance is given in [9]. A static mixer can be a hollow
cess. All mixer devices consume energy for the mixing process, but static tube or channel with a specific geometrical construction that influences
mixers use gravitational force as the driving mechanism. The type of the flow structure in a manner to promote secondary transverse flows
static mixer, the number of elements, and the velocity of the working that enhance mass and heat transfer in the cross-section. Commercial
fluid are the primary drivers for static mixer pressure drop and mixer static mixers are of various types: open designs with helices (helical
performance. Kenics — Chemineer, Inc.), open designs with blades or vortex genera-
Summaries of the field of static mixers including recent improve- tors (low pressure drop, Ross Engineering, Inc.), custody transfer
ments and applications to industrial processes are analyzed in [1]. The mixer (Komax Systems, Inc.), high-efficiency vortex (Chemineer, Inc.),
most commonly used static mixers are described and compared. Static corrugated-plates (SMV, Sulzer, Inc.), multi-layer designs (SMX and
mixers were established for industrial use in the 1970s. The first exam- SMXL, Sulzer, Inc.), closed designs with channels or holes (interfacial
ple of a static mixer was mentioned in the patent describing a single surface generator, Ross Engineering, Inc.), or designs based on metallic
element, multilayer motionless mixer used to mix air with a gaseous foam inserts, offset strip fins, or microstructured parallel plates. Ref.
fuel [2]. There was also a mixer with helical elements that promotes [10] introduced the study of laminar flow of high viscosity Newtonian
mixing in a tube [3], and another French patent which showed a and non-Newtonian fluids in a Kenics static mixer. The geometry of
multi-elemental design for blending of solids [4]. The staged elements the Kenics mixer consists of a number of helical blade elements
which promoted heat transfer were patented in the early 1950s [5]. mounted in a cylindrical pipe. A quantitative comparison of perfor-
Major petrochemical companies made a lot of effort and presum- mances of static mixers was fully analyzed in [11]. In this paper the glob-
ably utilized their designs internally in the decades proceeding the al performance was compared for different mixers in the regime of
laminar flow to find the optimum design using two criteria, i.e. pressure
drop and length.
The two main variables for a static mixer are the mixer size and the
⁎ Corresponding author. Tel./fax: +381 11 3283 185. number of elements. Equipment which is available on the market sat-
E-mail address: [email protected] (L. Pezo). isfies elementary technical conditionals. However, the mixing process

http://dx.doi.org/10.1016/j.powtec.2014.06.032
0032-5910/© 2014 Elsevier B.V. All rights reserved.
 A. Jovanović et al. / Powder Technology 266 (2014) 240–248 241

result of diffusion, convection and shear, which are the main mecha-
Nomenclature
nisms of the homogenization [12].
The problems related to the optimization of the mixing processes
d diameter, m
can be overcome by using mathematical modeling. The models can
E Young's modulus, Pa
drastically reduce the empirical work necessary for predicting parame-
f force, N
ters of the mixing process. This conclusion stems from the fact that
fc,ij contact forces, N
through several experiments the parameters of the model can be deter-
fD,i viscous drag force, N
mined, and then the model can provide information about the process,
fd,ij viscous damping forces, N
when some of the mixing conditions are changed. Without the mathe-
fp–f volumetric fluid–particle interaction force, N
matical model, this process would require many experiments in order
fp_f,i inter-particle forces between particles i and j, N
to find the best mode of operation, or to determine how the system
fpg,i pressure gradient force, N
responds to changes in mixing conditions. The simulation results are
g acceleration due to gravity, m/s2
reflected in the reduction of labor costs and the reduction of the time
I moment of inertia, kg m
required for the experiments.
mi mass of particle, kg
Models based on DEM (discrete element method) have been devel-
Mij rolling friction torque, Nm
oped in past and have shown to be reliable and efficient in catching par-
n unit vector in the normal direction of two contact
ticle interactions and predicting the mixing process for the investigation
spheres, dimensionless
of solid mixing.
p pressure, Pa
A detailed review of previous results in the area of mixing powders
R radius vector (from particle center to a contact point),
and granular materials is given by Bridgwater [13]. The focus was on
m
works related to mathematical modeling and numerical simulation
Re Reynolds number, dimensionless
using the discrete element method. DEM codes are now commonly
t time, s
being used to describe flow patterns on the scale of 10,000–250,000
Tij torques generated by the tangential forces, Nm
particles with a few workers using particles an order of magnitude
u fluid velocity, m/s
higher. A strategy that embraces the effects of particle size, equipment
u′ fluctuating fluid velocity, m/s
size and internal geometry, is advocated for the future. The aim was to
v velocity of particle, m/s
elucidate the engineering principles of general utility. Zhu et al. [14]
V volume, m3
provided a summary of the studies based on discrete particle simulation
in the past two decades or so. The studies are categorized into three sub-
ject areas: particle packing, particle flow, and particle–fluid flow. The
Greek
major findings are discussed, with emphasis on the micro-dynamics in-
β empirical coefficient defined in Table 1, dimensionless
cluding a packing/flow structure and particle–particle, particle–fluid
δ vector of the particle–particle or particle–wall overlap,
and particle–wall interaction forces. The soft-sphere method originally
m
developed by Cundall and Strack [15] was the first granular dynamic
ε porosity, dimensionless
simulation technique published in the open literature. They developed
ρ density, kg/m3
the linear spring and dashpot model whereby the magnitude of the nor-
μ fluid viscosity, kg/m/s
mal force between two particles was the sum of spring force and
μr coefficient of rolling friction, m
damping force. In such an approach, particles are permitted to suffer
μs coefficient of sliding friction, dimensionless
minute deformations, and these small deformations are used to calcu-
τ fluid viscous stress tensor, N/m3
late elastic, plastic and frictional forces between particles. The motion
ωi rotational velocities of particle i, rad/s
of particles is described by the well-established Newton's laws of mo-
tion. A characteristic feature of the soft-sphere model is that particles
are capable of handling multiple contacts which are of importance in
Indexes modeling quasi-static systems. Zhu et al. [14] provided a discussion of
c contact the inclusion of bulk fluid flow through the array of particles. They pro-
D drag posed coupling DEM to computational fluid dynamic (CFD) approach.
f fluid Pierce [16] carried out a simulation of the process of inter-particle per-
i particle i colation. He employed the Ataska code developed from Cundall and
p particle Strack using their standard modeling equations and particle parameters
p–f fluid–particle in the code. A much greater computer power has been used to follow
pg pressure gradient the behavior in a V-blender employing DEM in the work of Lemieux

Table 1
Simulation conditions and parameters of the modeling process.
is not always designed with appropriate care. This causes significant fi- Parameter Value
nancial loss and may cause bad quality of the mixture by unsatisfactory
Size of the mixer r × l, mm × mm 60 × 280
homogeneity.
Particle number 30,000
The typical industrial application of a static mixer is shown in Fig. 1. Particle diameter, mm 2
Screw conveyors are used to remove powder or grain material from Particle density, kg/m3 650
silos, and transport it to the weighing scale and the mixer. Before the Fluid velocity, m/s 0
CFD time step, s 5 × 10−5
mixing process is performed, a premixing action is done, using a static
DEM time step, s 5 × 10−6
mixer. Fluid density, kg/m3 1.2
The mixing process is very sensitive, and it must be optimally config- Fluid viscosity, kg/ms 1.8 × 10−5
ured. A mixing process that takes too long may cause deformation of the Particle friction coefficient 0.3
particles that are mixed and may also cause an increase of the price of Young's modulus 107
Poisson's ratio of particles 0.25
the technological process and final product. The mixing process is a
242 A. Jovanović et al. / Powder Technology 266 (2014) 240–248

particles. The addition of intermediate particle sizes increases convec-

tive and diffusive particle motion promoting a mixing mechanism that
reduces segregation via the sieving mechanism.
Lagrangian tracking techniques have been used in many studies in
order to characterize the mixing performance in different systems
[23–25].
A detailed review and definitions of the quality of a mixture, the
mixing mechanisms, the possibilities for the choice of solid mixer, the
experimental assessment of homogeneity and mixing indexes are pre-
sented in [26].
In this paper, experimental and numerical comparisons between
various multiple Komax and Ross mixing configurations have been
performed. The fluid is treated as a continuum while the solid phase is
modeled using the discrete element method (DEM). The model is
solved in two stages, first the fluid velocity and pressure are computed,
and then, using a separate study, the particle trajectories of the granular
material are calculated. The fluid (air) velocity and pressure are com-
puted by using the CFD (computational fluid dynamic) approach. In
the DEM, particle–particle and the particle–wall interactions are re-
solved and the time integration is performed using Newton's second
law of motion. The quality of the mixing process is analyzed using rela-
tive standard deviation (RSD) criteria [12]. The computational expense
of the DEM is very high owing to the extensive contact detection algo-
rithm, and solid time step limitations to resolve particle interactions
via collisions. Numerical simulations and mathematical modeling are
very powerful tools for optimization. The focus of this paper was to
optimize the geometry and to compare different static mixer devices.
Both Komax and Ross are commercial products, with known geometry,
used widely in various branches of industry. The main aim of this study
was to demonstrate the use of DEM/CFD simulation coupling in plan-
ning the number of Komax or Ross elements in order to gain desirable
mixing results.

2. Description of the mathematical model

Fig. 1. Schematic design of the static mixer industrial application, (1) silos, (2) screw
conveyor, (3) weighting scale, (4) static mixer, and (5) mixer. In static mixers, a fluid is pumped through a pipe containing sta-
tionary blades. This mixing technique is for laminar flow because it
generates only small pressure losses. This paper studies the flow in
two types of twisted-blade static mixers (Komax or Ross mixing
et al. [12]. They increase the length of the mixing time to 120 s and the elements, linked in a series of 1, 2 or 3 pieces, with or without the
number for particles to 225,000, nearly an order of magnitude greater divider (grid) at the exit). It evaluates the mixing performance by
than that used by others. Neuwirth et al. [17] investigated the fluid calculating the trajectory of suspended particles through the mixer.
and particle dynamics in a rotor granulator system (fluid bed rotor The mathematical model is solved in two stages, first the fluid ve-
processor). The mathematical model is based on a three-dimensional locity and pressure are computed by CFD, and then, using a separate
computational fluid dynamic (CFD) approach for the gas phase coupled study, the particle trajectories of the granular materials are comput-
with a discrete element method (DEM). The physical properties and ed by DEM.
collision behavior of test particles are investigated experimentally and The governing equations of gas phase are the same as those used in
incorporated in a viscoelastic particle contact model. The CFD/DEM the multiphase models. The gas phase is treated as the ideal gas, and the
model was firstly introduced by Tsuji et al. [18]. Liu et al. [19] developed ideal state gas equation for air has been applied in this model.
and tested the CFD/DEM model for complex geometry, where DEM code The conservations of mass and momentum in terms of the local
was realized with ANSYS/Fluent software through its user defined func- mean variables over a computational cell are given by:
tion. The Fluent Eulerian multiphase model is employed to couple with
DEM, whose secondary phase acts as a ghost phase but just an image
copy of DEM field. Chu et al. [20] did a numerical study of the gas– ∂ðρ f εÞ
þ ∇ðρ f εuÞ ¼ 0 ð1Þ
solid flow in a gas cyclone. They used a combined discrete element ∂t
method (DEM) and computational fluid dynamic (CFD) model where
the motion of the discrete particle phase was obtained by DEM which
and
applies Newton's equations of motion to every individual particle and
the flow of continuum fluid by the traditional CFD which solves the
Navier–Stokes equations at a computational cell scale. Su et al. [21] ∂ðρ f εuÞ
coupled DEM with OpenFOAM which is an open source CFD framework. þ ∇ðρ f εu  uÞ ¼ −∇p−Fp − f þ ∇ðετÞ þ ρ f εg ð2Þ
∂t
Remy et al. [22] investigated flow and segregation of polydisperse,
spherical particle mixtures in a bladed mixer using experimental and
computational techniques. The numerical simulations showed that the where ε, u, t, ρf, p, Fp–f, τ and g are: porosity, mean fluid velocity, time,
extent of segregation in the bladed mixer can be reduced by introducing fluid density, pressure, volumetric fluid–particle interaction force, fluid
intermediate particle sizes in between the smallest and the largest viscous stress tensor, and acceleration due to gravity.
 A. Jovanović et al. / Powder Technology 266 (2014) 240–248 243

Fluid particle interaction force is defined by: Tangential viscous damping forces:

1 X
k
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!1=2
i

Fp − f ¼ ρ ð3Þ 1−δt =δtmax

Vcell i¼1 p− f;i f dt;ij ¼ −ct 6mi μ s f cn;ij  vt;ij ð9Þ
δt; max

where Fp–f, i is the total fluid force on particle i and ki is the number of vij ¼ v j −vi þ ω j  R j −ωi  Ri
 
particles in a CFD cell.
where: vn;ij ¼  vij  n  n
The contact between two particles is achieved on a finite area due to
the deformation of the particles, which is equivalent to the contact of vt;ij ¼ vij  n  n:
two rigid bodies allowed to overlap slightly in the DEM [27]. The contact
traction distribution over this area can be decomposed into a compo- Body force gravity:
nent in the contact plane (or tangential plane) and one normal to the
plane, thus a contact force has two components: normal and tangential. Gi ¼ mi g: ð10Þ
It is difficult to describe the contact traction distribution over this area
and then the total force and torque acting on a particle, as it is related In this model, the gas–solid interaction force includes the viscous
to many geometrical and physical factors such as the shape, material drag force (fD,i) and pressure gradient force (fpg,i). Inter-particle forces
properties and movement state of particles. Alternatively, the DEM gen- are summed over the ki particles in contact with particle i.
erally adopts simplified models or equations to determine the forces Particle–fluid interaction force:
and torques resulting from the contact between particles. Linear models
are the most intuitive and simple models. The most common linear !2
4:8 ρ f jui −vi jðui −vi Þ πd2i ‐β
model is the so-called linear spring–dashpot model [15], where the f D;i ¼ 0:63 þ ε ð11Þ
spring is used for the elastic deformation while the dashpot accounts Re0:5
p;i
2 4 i
for the viscous dissipation.
The solid phase is treated as a discrete phase and described by the Rep;i ¼ di ρ f εμi jui ‐vi j
f

so-called discrete element method [15]. According to this model, the ð1:5− logRep;i Þ2
β ¼ 3:7−0:65 exp 2
translational and rotational motions of a particle at any time, t, can be where: kc
described by Newton's law of motion: ∑ Vi
ε ¼ 1− i¼1
ΔV : C

ki 
X  Pressure gradient force:
dvi
mi ¼ f p− f;i þ f c;ij þ f d;ij þ mi g ð4Þ
dt j¼1 f pg;i ¼ Vp;i ∇P: ð12Þ

and Torques, Tij, are generated by the tangential forces and cause particle
i to rotate because the inter-particle forces act at the contact point be-
tween particles i and j and not at the particle center. Mij is the rolling
i  
dωi X
k
Ii ¼ Tij þ Mij ð5Þ friction torque that is in opposition to the rotation of i-th particle.
dt j¼1 Torque rolling:
 
Tij ¼ Ri  f ct;ij þ f dt;ij : ð13Þ
where mi, Ii, vi and ωi are: the mass, moment of inertia, translational and
rotational velocities of particle i, respectively.
The forces acting on solids are the gas–solid interaction force, Torque friction:
fp–f,i, inter-particle forces between particles i and j, include the con- ^ ij :
Mij ¼ −μ r f ct;ij ω ð14Þ
tact forces, f c,ij , and viscous damping forces, fd,ij, and the gravita-
tional force, mig. They are detailedly described in [27].
The equations used to calculate the forces in Eqs. (1)–(14) are very
Normal contact forces:
much standardized now and they are listed in [14].
The fluid flow field can be obtained by solving Eqs. (1) and (2) by
E pffiffiffiffiffiffiffiffi 3=2 the use of a standard CFD method. Solid flow field can be obtained by
f cn;ij ¼ −   2Ri δn n: ð6Þ
3 1−v2 solving Eqs. (4) and (5) by an explicit time integration method. The
modeling of the fluid flow by CFD is performed at the computational
cell level, while the modeling of the solid flow by DEM is accom-
Normal viscous damping forces: plished at the individual particle level. Coupling DEM and CFD is
achieved as follows: DEM gives information about positions and ve-
0 11=2
locities of individual particles at each time step, for the evaluation of
B 3mi E C porosity and volumetric fluid–particle interaction force in a compu-
f dn;ij ¼ −cn @qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 pffiffiffiffiffiffiffiffiffiffiffiffiA vn;ij : ð7Þ
2 1−v2 2Rδn tational cell. CFD then uses these data to determine the fluid flow
field which then yields the fluid forces acting on individual particles.
Incorporation of the resulting forces into DEM will produce informa-
Tangential contact forces: tion about the motion of individual particles for the next time step
[20].
Numerical evaluations were performed for various static mixer
" !3=2 #
μ s f cn;ij configurations. The first segment of the mixer is filled with 30,000
minfjδt j; δtmax g
f ct;ij ¼− 1− 1− δt : ð8Þ particles. The scheme of the mixer is presented in Fig. 2: inlet com-
jδt j δt; max
partment filled with 15,000 red particles and second compartment
244 A. Jovanović et al. / Powder Technology 266 (2014) 240–248

account and it represents the force which leads the particles to

the bottom.
The density of the particles released is normalized to the magnitude
of the fluid velocity at the inlet. This means that there are more particles
released where the inlet velocity is the highest and fewer particles
released where the velocity field is low. It is assumed that the gas (air)
velocity is close to zero, so that the impact of fluid on particle movement
is minimal.

3. Experimental method

The experimental apparatus consists of a static mixer, which is made

from transparent Plexiglas. The different experiments were performed
with 1, 2 or 3 Komax or Ross mixer elements, linked in one column,
with or without the divider at the outlet, in the diameter of 60 mm.
Fig. 2b presents the 3 piece Ross static mixer, with no divider (grid) seg-
ment at the bottom. The upper segment is divided into two compart-
ments with a barrier and a mobile panel. Small spherical painted
zeolite granules are placed in both compartments (red granules in the
first compartment and blue granules in the second compartment). The
next 1, 2 or 3 segments (each with a height of 60 mm) were rotated
by 90° to the initial segment. The last two segment heights were
90 mm and 10 mm, respectively. The first of these segments was hollow,
or with the grid at the exit. This grid was used to quench the motion of
the particles on the rim, due to centrifugal force. In experiments, this
Fig. 2. a) Boundary conditions, and b) experimental mixer (three-segmented Ross grid was used only for configuration consisting of 3 mixing elements.
configuration). The last segment served to accept the zeolite granules and to show
the effect of mixing caused by motion of the granules through the
mixer. The material used for these experiments is a spherical shaped ze-
filled with 15,000 blue particles, outlet-pressure outlet (atmo- olite. The characteristics of granules are presented in Table 1. The condi-
spheric pressure) and wall — the other side of the mixer and blades tions under which the experiments were conducted are the same as in
of the mixer (Fig. 2a). The influence of the gravity is taken into the numerical simulation conditions.

Fig. 3. Numerical mesh used for simulation.

 A. Jovanović et al. / Powder Technology 266 (2014) 240–248 245

4. Simulation conditions while diffusion and source terms are approximated with central differ-
ences. Fully implicit time integration is applied. The set of algebraic
The set of balance equations (Eqs. (1)–(14)) is solved by using the equations is solved iteratively with the line-by-line TDMA (three-diag-
control volume based finite difference method. The SIMPLE (semi- onal-matrix algorithm) [28].
implicit method for pressure-linked equations) numerical method is The calculation error for every balance equation and every control
used for solving pressure-correction equation from the momentum volume is kept within limits of 10−5 by iterative solution of sets of linear
and mass balance equations [28]. Three-dimensional flow field is algebraic equations.
discretized in Cartesian coordinates. Numerical grids are made from The DEM requires a much smaller time step than the CFD. The CFD
109,540 to 111,320 control volumes, as shown in Fig. 3. This figure time step is ten times larger than the DEM time step. The DEM time
shows four static mixer configurations, used for numerical simulations: step is limited by the natural oscillation period of the spring-mass sys-
a) 3 Komax elements, b) 3 Komax elements with quadratic divider seg- tem used to model contacting particles. It should satisfy the following
ment, c) 3 Ross elements and d) 3 Ross elements with quadratic divider equation:
segment. The optimization of numerical grid was performed, and a grid
refinement test showed that there is no significant change in results of rffiffiffiffiffi
the simulation for larger number of cells in control volume. The ele- 1 m
ΔtDEM ≤ 2π ð15Þ
ments used in numerical mesh are tetrahedral and the size of the ele- 10 k
ments is less than 10−8 m3.
A discretization of partial differential equations is carried out by
their integration over control volumes of basic and staggered grids. where: m is the particle mass and k is the stiffness coefficient. Simula-
The convection terms are approximated with upwind finite differences, tion conditions and parameters of the process are given in Table 1.

Fig. 4. The velocity for the fluid phase.

246 A. Jovanović et al. / Powder Technology 266 (2014) 240–248

Fig. 5. Poincare maps of the particle trajectories at different Poincare sections, a) 3-segment Komax configuration and b) 3-segment Ross configuration.

5. Results and discussions configuration), the locations of particles at 11 Poincare sections are
shown. The color parameter is the logical expression used to mark the
Numerical simulations are performed for static mixers with different initial color of particles at positions x b 0 (red) and x N 0 (blue). So, par-
geometries and different parameters of the mixing process (as men- ticles marked as red had an initial position of x b 0 and particles marked
tioned above). The schematic description is shown in Fig. 2a. Results as blue had an initial position of x N 0. The first Poincare section, the one
of the DEM/CFD simulation are compared with experimental results. furthest to the left in Fig. 5a and b clearly indicates which particles start
The mechanical properties of zeolite granules are taken from [29,30]. with coordinates of x b 0. As the particles begin to follow the flow field,
The model is solved in two stages, first the fluid velocity field and they begin to mix together. By the end of the mixer, the particles have
pressure field are computed, and then, using a separate study, the parti- not mixed completely — there are still significant pockets of only red
cle trajectories are computed. and only blue particles.
In the following, we consider eight different representative cases Fig. 5 shows results of numeric simulation, and Fig. 6 represents ex-
(1, 2 and 3-segment Ross element configuration, 1, 2 and 3-segment perimental results. Fig. 5 shows cross-sections along the particle routes
Komax element configuration, 3-segment Ross configuration with of the mixer and only the last figure represents the results of the whole
additional divider and 3-segment Komax configuration with addi- mixing process. The mixing degree is getting better along the path, as
tional divider). It was shown that the numerical results were in shown by Fig. 5.
good agreement with experimental results by comparing the quality Color images of experimental results were captured by a Sony
of the mixing process, regarding RSD criteria [12]. PowerShot A550, which is a common digital camera for home use.
The velocity and the pressure field for the fluid phase were obtained All the acquired images were 24 bit RGB (16.8 millions of colors)
via CFD calculations (Fig. 4). This figure showed the tendency of falling with a 1024 × 768 spatial resolution. The macro function of the
particles to group, and the divider was used to quench the motion of the digital camera has been used, to cover a scene area of approximately
particles on the rim, due to centrifugal force. In experiments, this grid Ø60 mm. Samples were placed on a white paper napkin set on a flat
was used only for configuration consisting of 3 mixing elements. white painted surface, inside the closed chamber, 15 cm below the
Poincare plot is used to visualize how particles mix. The Poincare digital camera. Paper napkins were used in order to avoid undesired
plot places a colored dot for each particle at the location at which the reflection effects from chamber walls. With this setup, it was possi-
particle passes through a cut plane (known as a Poincare section). In ble to capture images with negligible shadows and without specular
Fig. 5 (a — 3-segment Komax configuration and b — 3-segment Ross reflections.

Fig. 6. Three-segmented Ross configuration mixing result, a) experimental, and b) processed photo.
 A. Jovanović et al. / Powder Technology 266 (2014) 240–248 247

The acquired images were transferred to a personal computer in used for these calculations, as a powerful tool to find the optimal geom-
the form of jpeg compressed image files. The size of the imported etry of the mixer. Overall particle trajectories for three-segmented
jpeg images ranged from 900 to 1250 KB. For the elaboration of the Komax and Ross configurations, without a grid divider, gained by nu-
images, every image file is imported in the originally developed merical simulations were: 436 and 430 mm, respectively.
computer program. Each image color data was transformed into a The degree of mixing and the time needed to achieve an acceptable
three-dimensional array of the R (red), G (green) and B (blue) values, mix can be predicted using coupled DEM/CFD algorithms and experi-
ranging from 0 to 255. These data were processed in order to find the mental measurements presented here. The literature has proposed
color of the observed pixel and to change its R, G and B values to the several tools and indices to evaluate the degree of homogeneity of a
closest color (red or blue). Fig. 6 presents the results of these calcula- mixture [12,26]. In this work, the relative standard deviation (RSD)
tions. Fig. 6 presents the mixing experiment results, performed with was used to follow the evolution of mixing uniformity for the static
the three-segmented Ross configuration, described above and present- mixers with different configurations, explained above. RSD, which is a
ed in Fig. 6b. An approximation of experimental results was necessary well-known mixing criterion in the pharmaceutical industry, is defined
to compare experimental results and modeling results. An original as:
code was developed for this purpose.
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
The phase portrait can also be used to visualize the effect of particle uM
uX
mixing but is not as useful as the Poincare map since it applies to all u ðx−xi Þ
2
u
particles at a given snapshot in time, rather than a plane in space. σ t i¼1
RSD ¼  100%; σ ¼ ; ð16Þ
The particle trajectories are plotted in Fig. 7. x M−1
The geometry of a mixer influences the mixing process. An optimiza-
tion of the mixing process was performed for different designs of where: M is the number of samples, xi the concentration of sample i and
mixers, Ross and Komax, with and without a divider. A mixing process x the average concentration of all samples.
depends on several parameters. Particle trajectory affects the mixing The results of numerical simulations and experimental mixing pro-
quality. The longer trajectory means a better quality of the mixing pro- cesses are presented in Fig. 8. Mixing begins after particles leave the
cess, but a mixing process that takes too long may cause a deformation upper segment, between sections 0 and 1 in Fig. 8 (as soon as the mobile
of the mixing particles. It is very important to find the optimal parame- panel is removed, enabling the granules to fall toward the static mixer).
ters of the mixing process. Particle trajectory is being increased by the The particles are rapidly blended in the first section, as seen from the
addition of one or more mixing elements, and numerical simulation is figure, reaching the mixing degree of 20–27% at the outlet. Komax
mixing element shows better blending results in this section (20–22%,
for DEM/CFD simulation and experimental results) compared to Ross
(24–27%), which is expected concerning its twisted surface geometry.
Komax mixing elements remained more effective after the second and
third sections, reaching the mixing degrees of 6–8% and 4–5% at the
outlets, while the mixing qualities of 11–13% and 5–6% were obtain-
ed at the outlets of sections 2 and 3, using Ross elements. The small,
but steady decrease in the mixing degree was observed for both
Komax and Ross blending elements during DEM/CFD simulation,
which can be seen in Fig. 8 (between sections 4 and 5). This is due
to the centrifugal force affecting the motion of granules that exit

Fig. 7. Results of the numerical simulation (particle trajectories). Fig. 8. RSD mixing criteria for Ross and Komax mixers.
248 A. Jovanović et al. / Powder Technology 266 (2014) 240–248

the mixing compartment. The decrease in quality is even more evi- [4] M.J. Bakker, Dispositif pour preparer du beton ou une matiere analogue, French
Patent 959,155(1949).
dent for the Komax configuration, as seen in Fig. 8, due to its specific [5] R.S. Lynn, Turbulator, US Patent 2,852,042, assigned to The Garrett Corporation
geometry. The use of the quadratic grid divider enabled the mixing (1958).
degree to remain constant after granules left the third mixing [6] R.F. Stearns, Method and apparatus for continuous flow mixing, US Patent
2,645,463, assigned to Standard Oil Development Company (1953).
element. [7] T.M. Veasey, Plate type fluid mixer, US Patent 3,382,534, assigned to Monsanto
Generally, the Komax mixer showed better blending results, com- Company (1968).
pared to Ross, but the complexity of Komax geometry makes it hard [8] J.E. Tollar, Interfacial surface generator, US Patent 3,239,197, assigned to The Dow
Chemical Company (1966).
to be manufactured. Komax is more applicable when the installation [9] A. Ghanem, T. Lemenand, D. Della Valle, H. Peerhossaini, Static mixers: mechanisms,
height must be low, but the use of Ross is more financially acceptable. applications and characterization methods — a review, Chem. Eng. Res. Des. 92
The use of a quadratic grid divider is advisable, due to the lowering of (2014) 205–228.
[10] M. Rafiee, M.J.H. Simmons, A. Ingram, E. Hugh Stitt, Development of positron
the segregation of granules.
emission particle tracking for studying laminar mixing in Kenics static mixer,
Chem. Eng. Res. Des. 91 (2013) 2106–2113.
6. Conclusion [11] H.E.H. Meijer, M.K. Singh, P.D. Anderson, On the performance of static mixers: a
quantitative comparison, Prog. Polym. Sci. 37 (2012) 1333–1349.
[12] A. Lemieux, G. Leonard, J. Doucet, L.A. Leclaire, F. Viens, J. Chaouki, Large-scale nu-
The discrete element method is applied to solve a wide range of en- merical investigation of solids mixing in a V-blender using the discrete element
gineering problems in: powder technology, mining processes, pharma- method, Powder Technol. 181 (2008) 205–216.
cology, chemical industry, food industry, power industry and many [13] J. Bridgwater, Mixing of powders and granular materials by mechanical means — a
perspective, Particuology 10 (2012) 397–427.
more areas. Coupled DEM/CFD approach was used to investigate the [14] H.P. Zhu, Z.Y. Zhou, R.Y. Yang, A.B. Yu, Discrete particle simulation of particulate
mechanism of fluid flow and the particle tracking for granular flow. In systems: a review of major applications and findings, Chem. Eng. Sci. 63 (2008)
the DEM, particle–particle and particle–wall interactions are resolved 5728–5770.
[15] P.A. Cundall, O.D.L. Strack, A discrete numerical model for granular assemblies,
and the time integration is carried out using Newton's second law of Geotechnique 29 (1979) 47–65.
motion. CFD analysis was used for the determination of velocity and [16] M.E. Pierce, PFC3D modeling of inter-particle percolation in caved rock under draw,
pressure field for the fluid phase by using a mixture model. The particle numerical modeling in micromechanics via particle methods, in: Y. Shimizu, et al.,
(Eds.), Proceedings of the 2nd International PFC Symposium, Kyoto, Japan, October
trajectories with particle positions were tracked and analyzed in order 2004, Balkema, Leiden, 2004, pp. 149–156.
to estimate the quality of the mixing process. A well-known mixing [17] J. Neuwirth, S. Antonyuk, S. Heinrich, M. Jacob, CFD–DEM study and direct mea-
criterion, named relative standard deviation is used for this purpose. surement of the granular flow in a rotor granulator, Chem. Eng. Sci. 86 (2013)
151–163.
Optimization of the dimension and mixing parameters in a static [18] Y. Tsuji, T. Kawaguchi, T. Tanaka, Discrete particle simulation of two-dimensional
mixer was done by using mathematical modeling. The aim of this fluidized bed, Powder Technol. 77 (1993) 79–87.
study is to predict the behavior of granules in different mixer configura- [19] D. Liu, C. Bu, X. Chen, Development and test of CFD–DEM model for complex geom-
etry: a coupling algorithm for Fluent and DEM, Comput. Chem. Eng. 58 (2013)
tions and to optimize parameters of the mixing process taking into ac-
260–268.
count the price of final product, duration of the mixing process and [20] K.W. Chu, B. Wang, D.L. Xu, Y.X. Chen, A.B. Yu, CFD–DEM simulation of the gas–solid
the quality of the mixture. Komax mixing elements are more applicable, flow in a cyclone separator, Chem. Eng. Sci. 66 (2011) 834–847.
compared to Ross, especially when the height of installation is low. [21] J. Su, Z. Gu, X.Y. Xu, Discrete element simulation of particle flow in arbitrarily com-
plex geometries, Chem. Eng. Sci. 66 (2011) 6069–6088.
However, the use of Ross is more financially acceptable, due to its sim- [22] B. Remy, J.G. Khinast, B.J. Glasser, Polydisperse granular flows in a bladed mixer:
pler geometry. The additional quadratic grid divider, installed after the experiments and simulations of cohesionless spheres, Chem. Eng. Sci. 66 (2011)
static mixer outlet, is used to quench the motion of the granules on 1811–1824.
[23] D.M. Hobbs, F.J. Muzzio, Reynolds number effects on laminar mixing in the Kenics
the rim of tubes, and to lower the segregation of granules. static mixer, Chem. Eng. J. 70 (2) (1998) 93–104.
[24] D.M. Hobbs, P.D. Swanson, F.J. Muzzio, Numerical characterization of low Reyn-
Acknowledgments olds number flow in the Kenics static mixer, Chem. Eng. Sci. 53 (8) (1998)
1565–1584.
[25] V. Kumar, V. Shirke, K.D.P. Nigam, Performance of Kenics static mixer over a wide
This work was supported by the Ministry of Education, Science and range of Reynolds number, Chem. Eng. J. 139 (2) (2008) 284–295.
Technological Development of the Republic of Serbia grant TR31055. [26] M. Poux, P. Fayolle, J. Bertrand, Powder mixing: some practical rules applied to
agitated systems, Powder Technol. 68 (1991) 213–234.
[27] H.P. Zhu, Z.Y. Zhou, R.Y. Yang, A.B. Yu, Discrete particle simulation of particulate
References systems: theoretical developments, Chem. Eng. Sci. 62 (2007) 3378–3396.
[28] S.V. Patankar, Numerical Heat Transfer and Fluid Flow, Hemisphere Publishing
[1] R.K. THAKUR, C.H. VIAL, K.D.P. NIGAM, E.B. NAUMAN, G. DJELVEH, Static mixers in Corporation, 1980.
the process industries — a review, Trans. Icheme 81 (2003) 787–826 (Part A). [29] R. Perry, D. Green, Perry's Chemical Engineering Handbook, 7th ed. McGraw-Hill,
[2] W.S. Sutherland, Improvement in apparatus for preparing gaseous fuel, UK Patent 1997.
1784 (1874). [30] J. Lin, X.F. Shu, J.X. Dong, The experimental determination of mechanical properties
[3] Les Consommateurs de Petrole, Dispositif pour le me'lange de deux ou plusieurs of zeolite ferrierite crystal, Mater. Lett. 59 (12) (2005) 1595–1597.
fluides, French Patent 735,033 (1931).