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

net/publication/321785872

Experimental and Computational Investigation of Mixing with Contra-

Rotating, Baffle-Free Impellers

Article in Chemical Engineering Research and Design · December 2017

DOI: 10.1016/j.cherd.2017.12.010

CITATIONS READS

17 2,615

8 authors, including:

Pongsarun Satjaritanun Eric Bringley

Hyzon Motors University of Cambridge
56 PUBLICATIONS 530 CITATIONS 14 PUBLICATIONS 163 CITATIONS

SEE PROFILE SEE PROFILE

John W Weidner Yottana Khunatorn

University of Cincinnati Chiang Mai University
338 PUBLICATIONS 6,723 CITATIONS 27 PUBLICATIONS 562 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Sirivatch Shimpalee on 21 December 2017.

The user has requested enhancement of the downloaded file.

 Chemical Engineering Research and Design 1 3 0 ( 2 0 1 8 ) 63–77

Contents lists available at ScienceDirect

Chemical Engineering Research and Design

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

Experimental and computational investigation of

mixing with contra-rotating, baffle-free impellers

P. Satjaritanun a , E. Bringley b , J.R. Regalbuto a , J.A. Regalbuto c , J. Register a ,

J.W. Weidner a , Y. Khunatorn d , S. Shimpalee a,d,∗
a Department of Chemical Engineering, University of South Carolina, Columbia, SC, USA
b Department of Chemical Engineering and Biotechnology, University of Cambridge, Cambridge, United Kingdom
c Perfect Mixing LLC, Columbia, SC, USA
d Department of Mechanical Engineering, Chiang Mai University, Chiang Mai, Thailand

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

Article history: This work experimentally and numerically investigates the intersection of two fields: (1) sin-
Received 28 July 2017 gle axis, contra-rotating impellers and (2) buoyancy of solid suspensions. The main goals of
Received in revised form 18 this study are to (1) create a working model to quantitatively understand particle mixing, (2)
November 2017 characterize and compare contra-rotating single shaft impellers to single shaft co-rotating
Accepted 4 December 2017 dual impellers, (3) improve quantification of particle mixing through image processing for
Available online 13 December 2017 both computational and experimental techniques, and (4) make design decisions with the
computational analysis. Twelve cases were studied by changing the direction of impeller
Keywords: rotation, impeller pumping direction, and the presence of baffles. Particles with specific
Contra-rotating impellers gravities (SG) of 0.866 and 1.050 were introduced into the experimental and computational
Perfect mixing systems in a finite and countable number. The numerical solution was obtained using the
Lattice Boltzmann method Lattice Boltzmann method and the Discrete Particle method. A commercial LBM solver,
Image analysis XFlow, was used for the simulation. The input torques and mixing efficiency with various
Design decisions flow configurations and specific gravities was used to find an optimal design. For the mixing
of the lighter particles, the contra configuration with inward opposing flow gave optimal
performance of highest mixing efficiency at lowest required torque. Co-rotating impellers
with baffles gave the best performance of high mixing efficiency at lowest power input for
the heavier particles.
© 2017 Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.

1. Introduction operation and maintenance. A contra-rotating impeller design poten-

Mixing and agitation operations are widely used in industrial process
such as pharmaceutics, agriculture, and food processing. Mixing is the
manipulation of a heterogeneous physical system with the intent to
make it more homogeneous (Paul et al., 2003). When a mixer uses a
single or two co-rotating impellers to agitate the substances inside
the mixer, a vortex is generally created (Paul et al., 2003; Regalbuto
and Regalbuto, 2014). In order to eliminate the vortex, baffles are usu-
ally inserted into the mixer to interrupt the flow of the contents (Paul
et al., 2003). The use of baffles required higher power input and are
sometimes problematic in construction (as in glass-lined mixers) or in
∗
Corresponding author.
E-mail address: shimpale@cec.sc.edu (S. Shimpalee).
https://doi.org/10.1016/j.cherd.2017.12.010
0263-8762/© 2017 Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.
tially eliminates the need for baffles. The contra-rotating impellers
design consists of two impellers rotating in opposite directions about
the same axis. Two impellers are arranged one behind the other, and
the power is transferred from the motor via bevel gear, planetary gear,
or spur gear transmission (Regalbuto and Regalbuto, 2014; El-Sayed,
2016). Due to this mechanical structure, the contra-rotating impeller
design has the advantages of high static pressure, high turbulent flow,
and good performance in reversing fluid flow. Prior to this work, this
contra-rotating design was studied and now being used in the aircraft
and marine industries to improve thrust, reduce torque, and optimize
aeroacoustics in the gas and/or liquid environment but not for the mix-
ing of solid–liquid (Wang and Meng, 2016; Paik et al., 2015; Min et al.,
2009; Gaggeroa et al., 2016; Grassi et al., 2010; Çelik and Güner, 2007).
64 Chemical Engineering Research and Design 1 3 0 ( 2 0 1 8 ) 63–77

key features which affect the efficiency of mixing processes. Kukukova

et al. (2009, 2011) applied Danckwerts’ method to define mixing and
Nomenclature
segregation based on three variables of segregation: intensity of seg-
a៝ D(f −p) Acceleration of the particle due to the drag force regation, scale of segregation, and rate of change of segregation. They
a៝ Extf,p External acceleration affecting both phase also concluded that with these variables, it is possible to clearly quan-
tify the mixing and segregation in mixing processes. Several studies
a៝ Extp External acceleration affecting only disperse
related to solid–liquid mixing (Lacey, 1954; Larosa and Manning, 1964;
phase
Harnby, 1967; Lacey and Mirza, 1976; Dlugi et al., 2014) also used the
b Probability distribution functions
intensity and scale of segregation to determine mixing efficiency. In
cs Speed of sound in Lattice scale
solid–liquid mixing, there are a large number of techniques which can
CD Drag coefficient provide both qualitative and quantitative information on the disper-
dp Particle diameter sion of solids in the mixer, i.e., conductivity probe, process tomography,
Da Impeller diameter and visual observation (Paul et al., 2003). In this study, we utilized both
e៝ i Particle discrete set of velocities visual observations and simulation of the dispersion of solid particles
eff Mixing efficiency in water to calculate mixing efficiency with the improved intensity of
fact Actual particle fraction segregation method.
fi Particle distribution function in the i direction The twelve cases that were studied are shown in Fig. 1 and include
the little-studied configuration of contra-rotating impellers. The mix-
fideal Ideal particle fraction
ing models were distinguished by rotation patterns and mixer formats.
g៝ Gravity
Cases 1–4 are contra-rotating impellers without baffles. Cases 5–8 are
Mij Transformation matrix to macroscopic
co-rotating impellers without baffles. Cases 9–12 are the co-rotating
moment impellers with baffles. The tank has a rounded bottom and a capac-

n Unit normal vector at control volume surface ity of 25 l. The two impellers have 4 blades pitched at 45◦ and are
nr Impeller rotation rate driven by an adjustable speed motor. Many researchers have studied
Nvis Number of particles visibly drawn into the liq- the variation of flow pattern, power consumption and mixing time
uid, particles with different reactor designs in order to develop an understanding
Nideal Total number of particles were fed into the liq- of the effect of changing the design parameters (Kuboi and Nienow,
uid, particles 1986; Mahmoudi and Yianneskis, 1991; Mishra and Joshi, 1994; Hiraoka
et al., 2001). These included the study of the impeller clearance from
P Macroscopic fluid pressures
the bottom and the distance between the two impellers. The results of
r Radius vector of impeller
these studies showed that the power consumption depends on the dis-
Re Relative Reynolds number
tance between two impellers as well as the interaction of the impellers
ReS Reynolds number for rotation system flows. However, the design optimization and the analysis of the flow
Ŝij Diagonal relaxation matrix paths of the particles in the reactor still need to be studied.
t Discrete times These studies analyzed the motion of the solid particles in the liquid
t Constant time step to find the solid–liquid mixing efficiency and power consumption using
Tr Torque three-dimensional (3D) computational fluid dynamics (CFD) modeling
u៝ Macroscopic fluid velocity with the Lattice Boltzmann method (LBM). The free surface model and
u៝ f Fluid phase velocity Discrete Phase Model (DPM) was also used in this work. LBM has many
u៝ p Particle disperse phase velocity advantages over conventional CFD (Navier–Stoke equation) for mod-

V Fluid velocity
elling moving geometries. This method uses a meshless particle based
LBM instead of the traditional meshing process for moving parts (i.e.,
x៝ Lattice node
Multiple Reference Frame and Sliding Mesh). The LBM simulation relies
on a generated lattice element, which is organized in an Octree struc-
Subscripts
ture, and uses a Large Eddy Simulation (LES) turbulence model, which
f Fluid phase can reduce meshing operation and computational times (Holman et al.,
p Particle disperse phase 2012). There are various modeling works on mixing, using impellers
act Actual where LBM simulation was utilized (Guha et al., 2008; Sungkorn et al.,
vis Visible 2012), which can model the movement of solid particles or bubble
flow dynamics and predict the solids dynamics in a solid–liquid mix-
Greek symbols ing tank. However, the mixing efficiency in these reactors need to be
 Molecular viscosity of the fluid further investigated. The solid–liquid mixing efficiency of the simula-
i Raw of the moment tion were calculated from the numerical results of the DPM by image
eq analysis techniques (Satjaritanun et al., 2016). The relative impact of
i Raw moment at equilibrium
the factors affecting mixing efficiency and power consumption – rota-
i Degree of deviation from ideal mixing
tion mode, rotation speeds and mixer designs – is clearly delineated
ıh Degree of deviation in the horizontal
and the advantages and disadvantages of contra-rotating impellers are
ıv Degree of deviation in the vertical
discussed.
␯ Macroscopic kinematic viscosity
 Relaxation parameter
 Macroscopic fluid density 2. Experimental setup and procedure
i Collision operator
MRT
i
Multiple relaxation time collision operator
The experimental setup is shown in Fig. 2a. It was constructed
from two tanks, a cylindrical tank with a round bottom and
an outer rectangular tank to eliminate distortion. The cylin-
For the past several decades, many researchers have attempted to drical tank is 0.30 m in diameter and 0.35 m in height. The
quantify mixing efficiency. Danckwerts (1952) quantified the “good- rectangular tank is 0.40 m in diameter and 0.40 m in height.
ness of mixing” with two statistically defined quantities, scale and The cylindrical tank that was filled with water was installed
the intensity of segregation, and noted that methods of measuring are inside the rectangular tank. Both tanks were filled with water
 Chemical Engineering Research and Design 1 3 0 ( 2 0 1 8 ) 63–77 65

Fig. 1 – The twelve flow configurations used in this study: Cases 1–4, contra rotating, Cases 5–8, single shaft co-rotating
without baffle, and Cases 9–12, single shaft co-rotating with baffles.

Table 1 – Experimental and computational conditions. 3. Image analysis

Variables Value Unit
Tracking particles as they are drawn down from the liquid sur-
Rotation speeds, nr 150, 200, 250, 300, 400 rpm face through image analysis provides a simple way to calculate
◦
Water temperature 20 C
the particle concentration and mixing efficiency in trans-
Water density,  998.3 kg/m3
parent systems. This method is an extension of the image
Water dynamic viscosity,  1 × 10−3 Pa s
Water diffusion coefficient 2.26 × 10−9 m2 /s analysis method developed from the intensity of segregation
Water volume 21.204 l and scale of segregation method (Kukukova et al., 2009, 2011).
Solid particle density, p 866 and 1050 kg/m3 This study presents a quantification scheme for countable sys-
Solid particles number 500 – tems and the calculation takes into account the homogeneity
Solid particle size 2.500 and 2.050 mm of the system’s concentration through the location of the par-
Gravity, g 9.810 m/s2
ticles. Images of cross sectional planes were taken as seen in
Fig. 3a–f, where there is high contrast between the background
and the particles submerged.
to the same height. Four blade turbine impellers with 45◦
These 2D images are divided into horizontal and vertical
pitched blades were installed 10 cm apart on the same shaft.
regions to measure the dispersion of the particle depth and
The impeller was installed in the center of the tank and located
spread, respectively. The regions are assigned a weight accord-
10 cm above the bottom of tank. The impeller was stirred by
ing to the volume they represent and the anticipated fraction
a Caframo BDC 3030 rotator motor with speed controls and
of visible particles for a perfectly homogeneous system. These
a torque sensor. Particles of higher and lower density than
regions present the ideal particle fraction (fideal ). For example,
water were used; the higher specific gravity was 1.050 and had
a region that represents 25% of the fluid volume is anticipated
a diameter of 2.05 mm and the lower specific gravity particle
to have 25% of the particles when homogeneously mixed. The
of 0.866 had a diameter of 2.45 mm. A total of 500 particles
systems are assumed to be dilute enough that overlap is not
were used for each experiment. Table 1 shows the experimen-
significant. An example can be seen in Fig. 4 with divisions
tal setup including rotation speeds, operating temperature,
and weights.
water volume in the tank, and properties of solid particle used
After image processing, the number of particles visibly
in the experiment. The same conditions were used for the
drawn into the liquid (Nvis ) for the horizontal and vertical
simulation study.
regions are recorded. The deviation from ideal mixing ( i ) is
The particles movement in the tank was captured by two
calculated individually for each horizontal (h) and vertical (v)
Nikon D3200 cameras to obtain two side-planes images as
regions as the square root of the sum of square differences
shown in Fig. 3a–c. When the system started to rotate, a
between the ideal particle fraction and actual particle fraction
photo was taken every 0.50 s from 0 to 40 s. Upon comple-
(fact ) as shown in Eq. (1). This calculated actual particle frac-
tion of the experiment, the captured photographs of the tank
tion from those images accounts for the area of solid particles
were analyzed to find mixing efficiency. Example images from
per the area of fluid. The worst case i∗ is taken when all of
the camera taken at the same point in time for different
configurations are shown in Fig. 3a–c.
66 Chemical Engineering Research and Design 1 3 0 ( 2 0 1 8 ) 63–77

Fig. 2 – (a) Experimental apparatus used to evaluate torque and capture experimental data and (b) dimensions of mixer and
impellers in this study.
the particles are in the smallest region, thus maximizing the greatest penalty is a 75% reduction. A quadratic function was
deviation from ideal. chosen to keep a low gradient for slight deviations and high
for large deviations from homogeneous, making the combined

i = 2 penalty non-linear.
(fideal − fact ) , i = h, v (1)

Nvis
The degree of deviation (ıi ) is calculated as one minus half eff = · ıv · ıh (3)
Nideal
the ratio of  i and i∗ shown in Eq. (2). Because of the maxi-
mized i∗ , the range of this equation will be between 0.5 (poor where Nvis is the number of particles visibly drawn into the
mixing) and 1 (near-perfect mixing). liquid, and Nideal is the total number of particles were fed into
the liquid.
 2
␴i
ıi = 1 − 0.5 , i = h, v (2)
␴∗i 4. Model development

The mixing efficiency (eff) is defined in Eq. (3) as the fraction 4.1. Lattice Boltzmann method (LBM)
of particles visible and the average product of the degree of
deviation in the horizontal (ıh ) and vertical (ıv ) direction over LBM was chosen to perform the numerical analysis in
the two planes. Because the range of ıi is between 0.5 and 1, the this work. It is one of the most powerful techniques for
 Chemical Engineering Research and Design 1 3 0 ( 2 0 1 8 ) 63–77 67

Fig. 3 – Particles inside the stirred tank under different dispersions between experimental and CFD simulation (a)
experimental, Case 1: contra rotating inward impellers configuration at low and (b) high specific gravities, (c) experimental,
Case 5: single rotating inward impellers configuration at low specific gravities, (d) CFD simulation, Case 1: contra rotating
inward impellers configuration at low and (e) high specific gravities, (f) CFD simulation, Case 5: single rotating inward
impellers configuration at low specific gravities.

Fig. 4 – Cross section divisions, with examples from left to right of the assigned weights, for good mixing, poor drawdown,
and poor dispersion. The two right hand images would generate strong deviation terms making their degree of deviation
close to zero because of the poor mixing.

computational fluid dynamics (CFD) for a wide variety of
complex turbulent flow problems including multiphase flow
and free surface models with complex geometries. This
method uses the concept of streaming and collision of par-
ticles which incorporates the physics of microscopic and
mesoscopic processes so that the macroscopic averaged
properties obey the desired macroscopic equations (Frisch
et al., 1986; McNamara and Zanetti, 1988; Chen and Doolen,
1998). The transport equation of this method is shown
below.
fi (x៝ + e៝ i t, t + t) − fi (x៝ , t) = ˝i (f1 (x៝ , t), . . ., (fb (x៝ , t)), i = 1, . . ., b (4)
where fi is the particle distribution function in direction i, x៝ is
the lattice node, e៝ i is the particle discrete set of velocities, t is
the discrete times, t is the constant time step, i is the colli-
sion operator, and b is the probability distribution function of
the particle distribution function mentioned above. As in the
68 Chemical Engineering Research and Design 1 3 0 ( 2 0 1 8 ) 63–77

continuum Boltzmann equation, macroscopic variables, such

as density and velocity u៝ , can be calculated as the moments
of the density distribution function:


b
= fi (5)
i=1


b

u៝ = fi e៝ i (6)
i=1

The macroscopic fluid pressures are calculated from the

equation of stage:

P = cs2 (7)

The multiscale Chapman-Enskog expansion gives us the

relation between the macroscopic kinematic viscosity (␯) and Fig. 5 – Lattice structure of the D3Q27 model. The weight
the relaxation parameter (): factors (wi ) are w0 = 8/27 (the cell-center), wi = 2/27
 1
 (i = 1–6),wi = 1/54 (i = 7–18) and wi = 1/126 (i = 19–29)
␯ = cs2  − (8) (Satjaritanun et al., 2016).
2

where cs is the speed of sound. For the positive kinematic vis- The relation between the PDFs and the raw moments can be
cosity,  > t defined as:
2 is a required stability condition. In addition, the
relaxation time should stay within the range 0.5–1.5. The value
in this study is 0.5. The LBM makes use of statistical distribu- i = Mij fi (12)
tion function with real variables, conserving the conservation
of mass, momentum, and energy (Chen and Doolen, 1998). In 4.2. Computational model
this model, the collision operator can be approximated by the
multiple relaxation time (MRT) as followed: 3D time-dependent simulations of the mixing and agitation
system with the free surface model were run for the twelve
˝iMRT = Mij−1 Ŝij (i − i )
eq
(9) different designs: Cases 1–4 (contra-rotating impellers with-
out baffles), Cases 5–8 (co-rotating impellers without baffles),
and Cases 9–12 (co-rotating impellers with baffles) as shown in
where the collision matrix Ŝij is a b × b diagonal relaxation
eq Fig. 1. The standard geometry involves an impeller and tank.
matrix, i is the equilibrium value of the i , and Mij is a b × b
The impeller has two impellers which can rotate in the oppo-
matrix, which transforms the distribution function to macro-
site or same direction. The impeller part is rotation when the
scopic moment (Shan and Chen, 2007; d’Humieres, 2002).
Euler angle changes with time. The enforced boundary con-
The collision operator is based on a multiple relaxation time
dition (rotating geometry) was used at the impeller part, a
scheme. However, as opposed to standard MRT, the scattering
free-slip wall with no velocity was defined at all of the reac-
operator is implemented in central moment space. The relax-
tor surfaces, and the free-surface boundary was imposed in
ation process is performed in a moving reference frame by
the function of height of the fluid. The rotation rates (nr ) were
shifting the discrete particle velocities with the local macro-
related to the Reynolds number (ReS ) and the Euler angle.
scopic velocity, naturally improving the Galilean invariance
Reynolds number for rotation system (McCabe et al., 2004) can
and the numerical stability for a given velocity set (Holman
be defined as follow:
et al., 2012; Premnath and Banerjee, 2012). Analogically to Eqs.
(5) and (6), raw moments of the probability distribution func-
nr D2a
tion f can be defined as: ReS = (13)



b
where ReS is the Reynolds number for rotation system, Da is
xk yl zm = fi ekix eliy em
iz (10)
the impeller diameter, nr is the rotation rate (rps), and  is the
i
molecular viscosity of the fluid. The system configuration is
shown in Fig. 2b and the dimensional information is listed in
and the central moments can be defined as:
Table 2.
LBM schemes are classified as a function of the spatial

b
 l
˜ xk yl zm =
 fi (eix − ux )k eiy − uy (eiz − uz )m (11) dimensions m and the number of distribution functions n,
i resulting in the notation DmQn. For the three dimensional (3D)
model used in this work, a commercial LBM solver, XFlow 2016
where k, l, and m are the orders of moments taken in x, y, and (Build 98) (Dassault Systèmes Simulia S.L.U. XFlow, 2016), was
z directions, respectively, and eix , eiy , and eiz are the particle chosen to perform the calculation. This solver uses spatial
discrete set of velocities in x, y, and z directions, respectively. dimension of 3 and number of distribution functions of 27 (i.e.,
The raw moment order is therefore k + l + m. Denoting i as a D3Q27) as shown in Fig. 5. The scale of lattice elements was
raw moment xk yl zm of a given combination of k, l, and m. set at 2 mm and a 40 s analysis time was used. The number of
 Chemical Engineering Research and Design 1 3 0 ( 2 0 1 8 ) 63–77 69

Table 2 – Dimensions of the mixers and impellers for the twelve cases.
Contra-rotating, without baffles Co-rotating, without baffles Co-rotating, with baffles
Cases 1–4 Cases 5–8 Cases 9–12

Tank diameter, DT (m) 0.300 0.300 0.300

Tank height, HT (m) 0.350 0.350 0.350
Liquid height, H (m) 0.300 0.300 0.300
Liquid volume, V (liter) 21.204 21.204 21.204
Baffle plate width, BW (m) – – 0.030
Baffle plate length, BL (m) – – 0.280
Baffle plate thickness, BT (m) – – 3.000
Number of baffles plate, NB – – 0.004
Number of impeller, Np 2 2 2
Impeller diameter, Da (m) 0.090 0.090 0.090
Impeller clearance, Ca (m) 0.100 0.100 0.100
Impeller-bottom clearance, Cb (m) 0.100 0.100 0.100
Blade length, L (m) 0.045 0.045 0.045
Blade width, W (m) 0.020 0.020 0.020
Blade thickness, t (m) 0.003 0.003 0.003
Number of blades, Nb 4 4 4
Blade angle, ␣ (◦ ) 45 45 45

the relative Reynolds number. While Eq. (14) includes a force

Table 3 – Lattice properties and times used for
simulations. of gravity on the particle, which is defined as:

Contra-rotating, Co-rotating, Co-rotating,

without baffles without baffles with baffles 
g៝ p − f
Cases 1–4 Cases 5–8 Cases 9–12 a៝ Extf,p = (17)
p
Lattice element 2.00 2.00 2.00
size (mm)
Total number of 2,847,550 2,847,550 2,867,897 where g is the acceleration due to gravity. The external accel-
elements
eration in the particle force balance are the force require to
Time step (s) 6.25e-5 6.25e-5 6.25e-5
Study times (s) 40 40 40
accelerate the fluid surrounding the particle Eq. (18) and the
force arises due to the pressure gradient in the fluid Eq. (19),
which are defined as:

lattice elements, lattice size and time step used in each sim-
ulation are shown in Table 3 and the simulation settings are
provided in Table 1.
The Discrete Phase Model (DPM) was used to simulate
the particles movement through the mass in the tank. This
method solves the transport equation for the continuous
phase and calculates the transport of a discrete phase con-
sisting of spherical particles (i.e., droplets, dust, bubbles, etc.)
dispersed in the continuous phase. The following is the equa-
tion of motion:
du៝ p
= a៝ D(f −p) + a៝ Extf,p + a៝ Extp
dt
where u៝ p is the particle velocity, a៝ D(f −p) is the acceleration of
the particle due to the drag force exerted by the fluid phase (f)
on the particle disperse phase (p), a៝ Extf,p is the external accel-
eration affecting both phase (f,p), (e.g. gravity), and a៝ Extp is the
external acceleration affecting only disperse phase (p).
18 CD Re
a៝ D(f −p) = · (15)
p d2p 24
f dp |u៝ p − u៝ f |
Re =
 pivot point to the point where the force is applied. In this
where u៝ f is the fluid phase velocity,u៝ p is the particle velocity,
CD is the drag coefficient, f is the density of the fluid, p is the
density of the particle, dp is the particle diameter, and Re is

f d u៝ f − u៝ p
a៝ Extp = · (18)
2p dt
up f ∂uf
a៝ Extp = · (19)
p ∂xi
The DPM calculates the particles trajectory with a
Lagrangian formulation consisting of discrete phase inertia,
hydrodynamic drag, and the effect of external forces. The
numerical schemes implicit Euler integration was used for dis-
(14)
cretization equations of motion for the particles. In this study,
the particles have the following properties: spherical shape,
specific gravity of 0.866 or 1.050, particle diameter of 2.45 mm
or 2.05 mm, respectively, and total number of 500. The injector
surface is on the bottom of the tank for high specific gravity
particles and on the top of the liquid surface for low specific
gravity particles. The system starts feeding the particles when
the simulation starts.
In mixing operations, the power consumption or impeller
torque is used for scale-up, scale-down and design optimiza-
tion. Torque is defined as the measure of the force applied to
produce rotational motion. In general, torque is determined
by multiplying the applied force by the distance from the
(16)
numerical simulation, the torque evaluation is based on an
angular momentum balance on the control volume surround-
ing the impeller (Chapple et al., 2002; Wang and Walters, 2012).
This method uses the velocity profiles around the impeller to
70 Chemical Engineering Research and Design 1 3 0 ( 2 0 1 8 ) 63–77
Fig. 6 – CFD simulation and experimental stirring torque compared in different mixer designs at the rotation speeds from
300 to 650 rpm (a) inward impeller pump configuration (b) outward impeller pump configuration (c) both-up impeller
configuration (d) Both-down impeller configuration.
calculate the force, which is used to calculate torque, as shown
in the following equation:
 
Tr =  ·n
 V   dA
r × V
Control surface
where Tr is the torque (N m), V is the fluid velocity (m/s), n
 is
the unit normal vector at control volume surface, and r is the
radius vector of impeller (m).
In experimental validation, comparison of the results from
the computational model and the experimental data can be
useful in identifying modelling problems with the implemen-
tation of a more general modelling approach. Validation of the
mixing efficiency data set with various flow configurations
and specific gravities can be achieved by the comparison of
the data set obtained experimentally with the image analy-
sis. For torque validation, simulation data can be compared
to the experimental data collected by measuring the torsional
deflection induced by the applied forces. This model allows
us to make more precise predictions on the mixing efficiency
that might be part of future design innovations.
5. Result and discussion
Fig. 6a–d presents for the twelve cases the stirring torque
validation between experimental measurements and torque
computations from the force on the blades and the resulting
momentum in the simulations. The standard deviation (SD)
between the experimentally measured data and computed
data is in the range between 0.026 and 0.090. Overall, the
model predictions agree well with the experimental results.
(20) In the single shaft co-rotating impellers without baffles (Cases
5–8), the results show these systems require lower power.
This is associated with the vortex formed and the bulk flow
in the mixer. When baffles are installed in the mixer (Cases
9–12), these baffles eliminate the vortex while increasing the
required torque (Satjaritanun et al., 2016).
Results of the single shaft co-rotating impellers both
with and without baffles cases can explain the relationship
between torque and flow behavior in the mixer. The results
of contra-rotating impellers without baffles (Cases 1–4) show
these systems require higher torque than other cases, which
is caused by the high turbulent flow in the mixer as shown
in Fig. 7a. In all of the cases in this study, the stirring torque
increases with increasing rotation speed. This is due to the
high degree of turbulent fluid flow. The torque required for
each system does not have a direct relation to mixing effi-
ciency. Due to its design significance, mixing efficiency should
be studied as a function of torque for each geometry, so that
the highest mixing efficiency at the lowest torque can be
found.
Fig. 7a–c show the velocity vector inside the different mix-
ers with inward flow configurations. Both Cases 1 and 9 have
similar velocity vector magnitude, but Case 1 can generate
more turbulent flow in the reactor compared to Case 9 as
 Chemical Engineering Research and Design 1 3 0 ( 2 0 1 8 ) 63–77
Fig. 7 – CFD simulation velocity vector in different mixer designs at 40 s and the rotation speeds of 400 rpm (a) Case 1: contra
rotating inward impellers configuration, (b) Case 5: single shaft co rotating inward impellers configuration without baffles,
and (c) Case 9: single shaft co rotating inward impellers configuration with baffles.
shown Fig. 7a and c. Therefore, the torque of the contra rotat-
ing impeller is the highest when compared with other inward
flow configurations at the same rotation speed as shown in
Fig. 6a. This is because of the effect of turbulent flow inside
the tank which while providing the fluctuation of fluid velocity,
the fluctuation on the impeller blades make an inconsistent
force against the impeller increasing torque. In the mixer
for Case 5 (Fig. 7b), shows lower torque and velocity vector
magnitude while having higher fluid velocity than the other
inward flow configurations, which leads to bulk flow inside the
reactor, which is a disadvantage to the mixing process. This
could be due to the result of the vortex in the reactor. The fluid
velocity inside the reactor with the vortex always has equal
value throughout the reactor, which implies the constant force
acting to the impeller.
Fig. 8a–f show the motion of solid particles for Case 1 (Con-
tra shaft inward) for times of 1, 5, 10, 15, 20 and 40 s. The
rotation speed was set at 300 rpm with 500 low specific gravity
particles. The experimental data and predictions were com-
pared using visual observation. Both results reveal that in the
beginning, the solid particles start with the same flow behav-
ior of circularly moving downward from the top of the liquid
surface toward the bottom of the mixer as shown in Fig. 8a–c.
After that the particles disperse into the entire tank with an
inward flow direction as shown in Fig. 8d–f. Visual observa-
tion show a small amount of solid particles dispersed in the
bottom of the mixing tank, which is the result of the effect of
the inward flow and particle specific gravity being lower than
water. The experimental visual observation from 10 and 15 s
(Fig. 8c and d) shows that the dispersion of the solid particles
is lower than the simulated data because the simulation does
not allow control of the seeding location in the setup as shown
in Fig. 8a. This affects the particles dispersion, which is why
the simulation does not match the experiment perfectly from
the beginning thru 20 s. After 20 s, the particles dispersion in
the tank were similar as presented in Fig. 8e and f.
Fig. 9a–f show the movement of the solid particles in the
mixer for Case 5 (Single shaft inward) from time of 1 s to 40 s.
The rotation speed was set at 300 rpm same as Fig. 8–f but all
71
500 particles had the higher specific gravity of 1.050. When
the impeller starts, the solid particles begin to move up from
the bottom of the tank and then scatter throughout as shown
in Fig. 9a–c. After that, the solid particles begin to separate
into two layers, with most of the particles residing in the bot-
tom of the tank as presented in Fig. 9d–f. The solid particles
float up and move down alternately as also shown in those fig-
ures, which is the result of the particle specific gravity being
higher than water on the one hand and the stream flow caused
by the impeller configuration on the other. With this config-
uration at 300 rpm, the vortex appears after 5 s and grows
steadily until it reaches steady state around 30 s. However, the
disadvantages of the vortex presented in the mixer are poor
mixing, thus resulting reduced mass transfer (Satjaritanun
et al., 2016). Though, there is no vortex observed on both con-
tra and co-rotating with baffles at the impeller speed greater
than 300 rpm.
To quantify the visual observations of particle dispersion,
image analysis was used to evaluate the mixing efficiency as
explained previously. This provides the mixing efficiency of
both experimental raw data and simulation data as shown in
Fig. 10. Fig. 10a–f present the transient simulation and experi-
mental mixing efficiency of Cases 1, 5 and 9 at different specific
gravities. The rotational speed was 300 rpm for all experiments
and simulations. These models have the same flow configu-
ration (i.e., inward flow direction). For the SG 0.866 (Fig. 10a, c,
and e), the mixing efficiency from predictions are higher than
experimental data from times of 0 to 20 s. This is because the
simulations could not allow control the seeding location in the
setup as shown in Fig. 8a. The steady state results from Fig. 10
agree with the long-time visual observation in Fig. 8a–e. For
specific gravity of 1.050, the mixing efficiency from both simu-
lation and experiment are similar from the beginning until the
end of agitation as shown in Fig. 10b, d and f. This is because all
particles were seeded at the same location at the bottom of the
tank. The mixing efficiency of both specific gravities reaches
steady state after 30 s. Therefore, the steady state mixing effi-
ciency was calculated using the average data of the mixing
efficiency after 30 s.
72 Chemical Engineering Research and Design 1 3 0 ( 2 0 1 8 ) 63–77

Fig. 8 – Comparisons of experiment and CFD simulation for Case 1 showing a front view of the particle dispersion at
different times with a rotation speed of 300 rpm. (Particle properties: diameter 2.5 mm, specific gravity of 0.866, and 500
particles.)

Fig. 9 – Comparisons of experiment and CFD simulation for Case 5 showing a front view of the particle dispersion at
different times with a rotation speed of 300 rpm. (Particle properties: diameter 2.05 mm, specific gravity of 1.050, and 500
particles.)
 Chemical Engineering Research and Design 1 3 0 ( 2 0 1 8 ) 63–77 73

Fig. 10 – CFD simulation and experimental mixing efficiencies compared in different mixer designs at two different specific
gravities at a rotation speed of 300 rpm, (a) Case 1: contra rotating inward impellers configuration at low and (b) high specific
gravities, (c) Case 5: single rotating inward impellers configuration at low and (d) high specific gravities, (e) Case 9: single
rotating inward impellers configuration with baffles at low and (f) high specific gravities.

Fig. 11 – CFD simulation and experimental mixing efficiencies compared in the twelve mixer designs at (a) low and (b) high
specific gravities at a rotation speed of 300 rpm.
74 Chemical Engineering Research and Design 1 3 0 ( 2 0 1 8 ) 63–77

Fig. 12 – CFD simulation of mixing efficiency at rotation speeds from 150 to 500 rpm for the twelve mixer designs, (a) Cases
1–4: contra rotating shafts at low and (b) high specific gravities, (c) Cases 5–8: single shaft at low and (d) high specific
gravities, (e) Cases 9–12: single shaft with baffles at low and (f) high specific gravities.

Fig. 11a–b compare the simulation and experimental data
of mixing efficiency at steady state for all configurations at
300 rpm. Fig. 11a presents the case for low specific gravity. The
results show similar mixing efficiency for both the experimen-
tal and simulation results. Fig. 11b presents similar mixing
efficiency for each of the three main configurations (contra,
co-rotating with no baffles, co-rotating with baffles) when
the high specific gravity of particles were introduced. The co-
rotating impellers with no baffles, Cases 5–8, show low mixing
efficiency for both specific gravities. Because this tank has no
baffles, a vortex develops in the bulk flow of the tank. Also, the
mixing efficiency for Cases 5–8 at low specific gravity is higher
than that of high specific gravity for the same operating condi-
tions. Most of the high specific gravity particles resided in the
bottom of the tank, because of the lack of sufficient stream
flow in the lower part of the mixing tank from the generation
of the vortex. The low specific gravity particles move down
and float up alternately and follow the direction of the fluid
motion which results because the particle specific gravity is
being lower than water and bulk flow of the tank. The contra-
rotating impellers (Cases 1–4) and co-rotating impellers with
baffles (Cases 9–12) show better mixing efficiency than Cases
5–8, because these designs have more turbulent flow in the
tank, with the highest efficiencies achieved with Cases 1–3
 Chemical Engineering Research and Design 1 3 0 ( 2 0 1 8 ) 63–77
Fig. 13 – CFD simulation of mixing efficiency with torque for twelve reactor designs with particles at lower specific gravity
(a) inward impellers configuration (b) outward impellers configuration (c) both-up impellers configuration (d) both-down
impellers configuration.
of the contra-rotating impellers. However, with higher spe-
cific gravity, the co-rotating impellers with baffles tank gives
higher mixing efficiency than the contra-rotating. Therefore,
in this mixer set up, the contra-rotating impellers is optimal
for low specific gravity solids and the co-rotating impellers
with baffles case is optimal for high specific gravity solids.
Fig. 12a–f present the simulated mixing efficiency of the
twelve configurations as a function of rpm. The contra-
rotating impellers (Cases 1–4) develop approximately twice
the mixing efficiency of a co-rotating impellers without baf-
fles (Cases 5–8). The mixing efficiency for all contra-rotating
impellers gradually increases and reaches steady state after
400 rpm. The simulations reveal that Case 1 (inward flow con-
figuration) gives the highest mixing efficiency for both specific
gravities as shown in Fig. 12a and b. These designs do not need
baffles and can generate strong turbulent flow in the tank. The
co-rotating impellers (Cases 5–8) provide the highest average
mixing efficiency at around 250 rpm then it drops significantly
to reach constant efficiency after 400 rpm as shown in Fig. 12c
and d. The mixing efficiency drops at higher rpm because of
the vortex formation. The co-rotating impellers with baffles
tank (Cases 9–12) show the highest average mixing efficiency
for all cases with high specific gravity as shown in Figs. 12b,
d, and f. Cases 9–12, at lower specific gravity, show a mix-
ing efficiency lower than that of contra-rotating impellers as
shown in Fig. 12a and e. However, with lower specific grav-
ity, both down flow configurations (Cases 4, 8 and 12) provide
lower mixing efficiency than the other flow configurations.
Cases 4 and 8 showed similar mixing efficiency at impeller
75
speeds greater than 400 rpm. Case 12 showed a lower mixing
efficiency compared to case 8 because of the effect of the buoy-
ancy force, lower fluid velocity (Satjaritanun et al., 2016) and
less flow in the lower part of the mixing tank. It was found that
the rotation speed is related to the mixing efficiency. The mix-
ing efficiency for all cases reach to steady state after 400 rpm.
Therefore, appropriate design choices for each system should
consider the size and weight of particles together.
The configuration of a mixer might be considered optimal
when it achieves highest mixing efficiency at lowest power
input. In this vein, simulations were run for the twelve cases
and low and high specific gravity particles of mixing efficiency
versus rpm, which was then converted to torque. Mixing effi-
ciency is plotted versus torque for the twelve cases of low
specific gravity in Fig. 13 and for high specific gravity in Fig. 14.
In the low specific gravity case (Fig. 13) the optimum design
for both inward and outward flow configurations are the
contra-rotating impellers (Cases 1 and 2) which achieve mixing
efficiency of 80% at torques of 0.08 and 0.10 N m (Fig. 13a and b)
respectively. The optimal configuration of the twelve is the lat-
ter, the inward opposing, and contra setup. The advantage of
this configuration does not appear to have been identified pre-
viously in the mixing literature. The optimum design for the
both-up flow configuration is the co-rotating impellers with
baffles tank (Case 11) with a torque of 0.04 N.m. and a mixing
efficiency of 71% as shown in Fig. 13c. The optimum design for
the both-down flow configuration is the co-rotating impellers
without baffles tank (Case 8) with the torque of 0.05 N m. and
a mixing efficiency of 46% as shown in Fig. 13d. In general, the
76 Chemical Engineering Research and Design 1 3 0 ( 2 0 1 8 ) 63–77
Fig. 14 – CFD simulation of mixing efficiency with torque for twelve reactor designs with particles at higher specific gravity
(a) inward impellers configuration (b) outward impellers configuration (c) both-up impellers configuration (d) both-down
impellers configuration.
both-down flow configuration shows the lowest mixing effi-
ciency for all cases and this configuration appears unsuited
for lower specific gravity. This is the common configuration
of contra-rotating mixers in the literature (Wang and Meng,
2016; Paik et al., 2015; Min et al., 2009; Gaggeroa et al., 2016;
Grassi et al., 2010; Çelik and Güner, 2007). The contra-rotating
impellers with inward flow is the best configuration for lower
specific gravity as it shows the highest mixing efficiency with
lowest torque.
Fig. 14a–d present the mixing efficiency versus torque for
the twelve mixer configurations with the higher specific grav-
ity particles. The optimum design for all configurations is the
co-rotating impellers with baffles (Cases 9–12) with mixing
efficiencies of 88, 90, 86 and 87% at torques of 0.07, 0.04, 0.05
and 0.04 N m and as shown in Fig. 14a–d, respectively. The
only potential disadvantage with the co-rotating configura-
tions is that some of the particles were trapped in the dead
zone created by the baffles. The co-rotating impellers without
baffles (Cases 5–8) show moderately high mixing efficiency at
low torque, but performance becomes much poorer at higher
speeds, for all configurations, as vortexes form. The contra-
rotating impellers have a mixing efficiency only slightly less
than the co-rotating impellers with baffles. The absence of
dead zones and baffles with the contra-rotating impellers per-
haps makes up for the slightly lower mixing efficiency in
the mixing of powders or sludge, which are common in the
pharmaceutical industry or biomass mixing (Paul et al., 2003;
Regalbuto and Regalbuto, 2014).
6. Conclusions
Experimentally determined mixing efficiency can be faithfully
simulated by numerical solutions using the Lattice Boltzmann
method combined with image analysis. Simulations are in
agreement with experiment results, yield fundamental under-
standing, provide basis for design decisions and can be used
to better understand mixing characteristics in a wide variety
of industrial applications.
In comparing contra- to co-rotating impellers, rotation
speed and flow configuration play major roles on the mixing
characteristic especially when buoyancy forces are significant.
The bulk motion of a vortex causes poor mass transfer and
poor mixing. However, the vortex can be eliminated by using
contra-rotating impellers or the installation of baffles around
the tank. Contra-rotating impellers eliminate dead zones at
baffles.
For the mixing of suspended solids with specific gravity
slightly below water, the contra configuration with inward
opposing flow gave optimal performance of highest mixing
efficiency at lowest required torque. Co-rotating impellers
with baffles gave the best performance of high mixing effi-
ciency as lowest power input.
Acknowledgments
The authors would like to thank Perfect Mixing LLC for provid-
ing the experimental equipment. The authors also would like
 Chemical Engineering Research and Design 1 3 0 ( 2 0 1 8 ) 63–77
to acknowledge Dassault Systèmes Simulia S.L.U. for providing
the XFlow software. Finally, the authors would like to recog-
nize Mr. Drew Pereira and Mr. Cody Wilkins for proofreading
our manuscript.
References
Çelik, F., Güner, M., 2007. Energy saving device of stator for
marine propellers. Ocean Eng. 34, 850–855.
Chapple, D., Kresta, S.M., Wall, A., Afacan, A., 2002. The effect of
impeller and tank geometry on power number for a pitched
blade turbine. Trans. IChemE 80, 364–372.
Chen, S., Doolen, G.D., 1998. Lattice Boltzmann method for fluid
flows. Annu. Rev. Fluid Mech. 30, 329–364.
Danckwerts, P.V., 1952. The definition and measurement of some
characteristics of mixtures. Appl. Sci. Res. 3, 279–296.
Dassault Systèmes Simulia S.L.U. XFlow, 2016.
http://www.xflowcfd.com/.
Dlugi, R., Berger, M., Zelger, M., Hofzumahaus, A., Rohrer, F.,
Holland, F., Lu, K., Kramm, G., 2014. The balances of mixing
ratios and segregation intensity: a case study from the field
(ECHO 2003). Atmos. Chem. Phys. 14, 10333–10362.
El-Sayed, A.F., 2016. Fundamentals of Aircraft and Rocket
Propulsion. Springer Publishing Company, New York.
Frisch, U., Hasslacher, B., Pomeau, Y., 1986. Lattice-gas automata
for the Navier–Stokes equation. Phys. Rev. Lett. 56,
1505–1508.
Gaggeroa, S., Gonzalez-Adalidb, J., Sobrinob, M.P., 2016. Design
and analysis of a new generation of CLT propellers. Appl.
Ocean Res. 59, 424–450.
Grassi, D., Brizzolara, S., Viviani, M., Savio, L., Caviglia, S., 2010.
Design and analysis of counter-rotating
propellers—comparison of numerical and experimental
results. J. Hydrodyn. B 22, 570–576.
Guha, D., Ramachandran, P.A., Dudukovic, M.P., Derksen, J.J., 2008.
Evaluation of large Eddy simulation and Euler–Euler CFD
models for solids flow dynamics in a stirred tank reactor.
AIChE J. 54, 766–778.
Harnby, N., 1967. A comparison of the performance of industrial
solids mixers using segregating materials. Powder Technol. 1,
94–102.
Hiraoka, S., Tada, Y., Kato, Y., Matsuura, A., Yamaguchi, T., Lee,
Y.S., 2001. Model analysis of mixing time correlation in an
agitated vessel with paddle impeller. J. Chem. Eng. Jpn. 34,
1499–1505.
Holman, D.M., Brionnaud, R., Abiza, Z., 2012. Solution to industry
benchmark problems with the Lattice-Boltzmann code XFlow.
In: Proceeding in the European Congress on Computational
Methods in Applied Sciences and Engineering (ECCOMAS),
Vienna, Austria.
d’Humieres, D., 2002. Multiple relaxation time Lattice Boltzmann
models in three dimensions. Phil. Trans. R. Soc. Lond. A 360,
437–451.
Kuboi, R., Nienow, A.W., 1986. Intervortex mixing rates in
high-viscosity liquids agitated by high-speed dual impellers.
Chem. Eng. Sci. 41, 123–134.
View publication stats
77
Kukukova, A., Aubin, J., Kresta, S.M., 2009. A new definition of
mixing and segregation: three dimensions of a key process
variable. Chem. Eng. Res. Des. 87, 633–647.
Kukukova, A., Aubin, J., Kresta, S.M., 2011. Measuring the scale of
segregation in mixing data. Can. J. Chem. Eng. 89, 1122–1138.
Lacey, P.M.C., Mirza, F.S.M.A., 1976. A study of the structure of
imperfect mixtures of particles. Part I. Experimental
technique. Powder Technol. 14, 17–24.
Lacey, P.M.C., 1954. Developments in the theory of particle
mixing. J. Appl. Chem. 4, 257–268.
Larosa, P., Manning, F.S., 1964. Intensity of segregation as a
measure of incomplete mixing. Can. J. Chem. Eng. 42, 65–68.
Mahmoudi, S.M., Yianneskis, M., 1991. The variation of flow
pattern and mixing with the impeller spacing in stirred
vessels with two Rushton impellers. In: 7th European. Congr.
on Mixing, Brugge, Belgium, pp. 17–24.
McCabe, W., Smith, J., Harriott, P., 2004. Unit Operations of
Chemical Engineering. McGraw Hill Chemical Engineering
Series, New York.
McNamara, G.R., Zanetti, G., 1988. Use of the Boltzmann equation
to simulate lattice-gas automata. Phys. Rev. Lett. 61,
2332–2335.
Min, K.S., Chang, B.J., Seo, H.W., 2009. Study on the
contra-rotating propeller system design and full-scale
performance prediction method. Int. J. Nav. Archit. Ocean Eng.
1, 29–38.
Mishra, V.P., Joshi, J.B., 1994. Flow generated by a disc turbine. IV:
multiple impellers. Trans. IChemE 72, 657–668.
Paik, K.J., Hwang, S., Jung, J., Lee, T., Lee, Y.Y., Ahn, H., Van, S.H.,
2015. Investigation on the wake evolution of contra-rotating
propeller using RANS computation and SPIV measurement.
Int. J. Nav. Archit. Ocean Eng. 7, 595–609.
Paul, E.L., Atiemo-Obeng, V., Kresta, S.M., 2003. Handbook of
Industrial Mixing: Science and Practice. John Wiley & Sons
Inc, Hoboken.
Premnath, K., Banerjee, S., 2012. On the three-dimensional
central moment Lattice Boltzmann method. J. Stat. Phys. 143,
747–761.
Regalbuto, J.R., Regalbuto, J.A., 2014. Method and Apparatus for
Improved Mixing of Solid, Liquid, or Gaseous Materials and
Combinations Thereof, U.S. Patent: 20140078858 A1.
Satjaritanun, P., Khunatorn, Y., Vorayos, N., Shimpalee, S.,
Bringley, E., 2016. Numerical analysis of the mixing
characteristic for napier grass in the continuous stirring tank
reactor for biogas production. Biomass Bioenerg. 86, 53–64.
Shan, X., Chen, H., 2007. A general multiple-relaxation-time
Boltzmann collision model. Int. J. Mod. Phys. C 18, 635–643.
Sungkorn, R., Derksen, J.J., Khinast, J.G., 2012. Euler–Lagrange
modeling of a gas–liquid stirred reactor with consideration of
bubble breakage and coalescence. AIChE J. 58, 1356–1370.
Wang, Q., Meng, D., 2016. A study of the matching of impellers
and motors for contra-rotating fan based on
electromagnetic-fluid coupling analysis. Adv. Mech. Eng. 8,
1–11.
Wang, X., Walters, K., 2012. Computational analysis of
marine-propeller performance using transition-sensitive
turbulence modeling. Trans. ASME 134, 0711071–07110710.