Vous êtes sur la page 1sur 11

Numerical Investigation of Laboratory Knelson Concentrator to Study the

Effects of Operational Parameters Using LIGGGHTS DEM Solver


M. Fatahi1, A. Farzanegan2
1-School of Mining Engineering, College of Engineering, University of Tehran, P.O. Box
11155-4563, Tehran, Iran, Mohammadreza.fatahi@yahoo.com
2- School of Mining Engineering, College of Engineering, University of Tehran, P.O. Box
11155-4563, Tehran, Iran

Abstract
In this research, the effect of operational parameters including centrifugal force intensity, i.e.,
the rotational velocity of the bowl, feeding rate and separation cycle time on separation
mechanism of a 7.5-cm laboratory Knelson Concentrator (LKC) at particle scale is investigated
by applying a new approach based on DEM modeling. Separation performance of LKC is
simulated using an open-source DEM solver ‘LIGGGHTS’. The simulations were run using
5000 mono size particles. Feed was a binary mixture of iron or platinum and quartz respectively
as heavy and light particle with specific weight ratios.
Simulation results are expressed qualitatively and quantitatively in terms of performance
indices, i.e., concentrate grade and total recovery. In order to validate simulator predictions,
experimental tests on synthetic sample of pure quartz and magnetite under specific conditions
close to simulations conditions were carried out using the 7.5-cm LKC. DEM simulations were
validated based on both qualitative (visual) and quantitative observations. For qualitative
validation, the visual performance of DEM-simulated LKC is compared with the actual
performance observed in the real experiments with LKC. For quantitative validation, grade and
recovery predictions resulted from DEM simulations are compared with experimental
measurements obtained from laboratory tests.

Keyword: Laboratory Knelson Concentrator, operational parameters, DEM simulation,


LIGGGHTS, separation performance

1. Introduction
Enhanced gravity separation is a mature technology that had its beginnings in the late 1980s
and early 1990s and now has had more than 20 years of development and application [1].
Enhanced gravity concentrators (EGCs) are gravity concentrators that employ a centrifugal
force to enhance the settling rate of particles. The use of centrifugal force to increase the settling
rate of particles has been applied for many years [2]. In an EGC like Knelson concentrator the
centrifugal force is generated by rotating the separating bowl itself. Therefore, the separation
is primarily based on the relative settling velocity differential between the particles differing in
size and density. With enhancing the velocity differential, separation of particles in finer size
ranges is possible [3].
Much of the published research performed about Knelson concentrator has been experimental.
Previous experimental investigations on the Knelson concentrator demonstrated that different
operational variables have important impacts on the separation performance of the KC [4-7].
Understanding and optimizing equipment like KC is expensive and time consuming and it also
requires considerable manpower and sample size. The discrete element method (DEM) which
has been developed since 1970s is a numerical method which is suitable for calculating
mechanical behavior of granular medium systems. DEM has got successful applications in
granular media engineering fields, such as geotechnical engineering, mining engineering,
mineral processing and material separation. DEM has become a multidisciplinary research
method which core is to help people investigate the relationship between microscopic and
macroscopic characteristic of the discrete particles [8]. However, to date there has been no
report in the literature about DEM simulation of the separation process in Knelson concentrator.
In this article we investigate the effects of operational variables including centrifugal force
intensity, feeding rate and separation cycle time on the concentrate grade and total recovery of
the LKC by a number of controlled numerical simulations. These studies present a better
understanding of the concentration process in the Knelson concentrator helpful for the optimal
design of an Enhanced gravity separation process.

2. Theoretical background
2.1. DEM model
Discrete element method simulation involves following the motion of every particle (coarser
than some resolved cutoff size) in the system and modelling each collision between the particles
and between the particles and their environment, which in this article is the Knelson bowl, by
the rule of contact forces. This provides essentially unlimited flexibility in specifying the
complex three-dimensional geometries of real machines with which the particles interact. The
general DEM methodology has been well established and is described in early review articles
by [9-11].
The contact model considers the force of gravity and the normal and tangential forces acting
on a particle during the separation process. In addition, two torques that are produced by a
tangential force and a rolling friction force also are considered to act on the particles. The
translational and rotational motions for i’th particle, at any time, t, are determined by Newton's
second law of motion. These can be written as:
dvi
mi = ∑(Fijn + Fijs ) + mi g (1)
dt
j

dωi
Ii = ∑(R ij × Fijs − μr R i |Fijn |ωi ) (2)
dt
j

Where m and I are the mass and moment of inertia of the i’th particle, vi, transfer speed, ω i,
angular velocity, t, time, g, the acceleration of gravity, Rij, a vector from center of particle i to
contact point with particle j, Fijn, normal force of contact, Fijs, tangential force of contact.
Simulations to become a versatile tool for industrial applications [12]. Recent advances in
discrete element modelling have resulted in this method becoming a useful simulation tool that
can provide detailed information not easily measured during experiments [13]. With the
maturing of DEM simulation, it is now becoming possible to run simulations of millions of
particles with complex shapes and inter-particle cohesive forces in tolerable times on single
processor, desktop computers.

2.2. The DEM solver “LIGGGHTS”


LIGGGHTS which stands for “LAMMPS Improved for General Granular and Granular Heat
Transfer Simulations” is an open-source software package for modeling granular material by
means of DEM. [14]. LAMMPS (Large-scale Atomic/Molecular Massively Parallel Simulator)
is an open-source classical molecular dynamics code by Sandia national laboratories for
massively parallel computing on distributed memory machines. It also includes coarse-grained
particles and interaction potentials, so it can model materials at the mesoscale as well [15].
For the DEM simulations, we used a version of the LIGGGHTS package adapted to our needs.
Visualization of simulations is possible by translating the data of LIGGGHTS with PIZZA and
loading them into ParaView [16]. The application of LIGGGHTS software in the field of
mineral processing is limited. Recently M.jahani et al [17] investigated the screening
performance of banana screens with LIGGGHTS. He examined the effects of design
parameters incline angle of panels, vibration amplitude and vibration frequency and operational
parameters including simulation time and feed particle size distribution on screening recovery
of both laboratory and industrial scale screens.

3. Knelson concentrator geometry and simulation conditions


The Knelson concentrator used in this investigation was a 7.5-cm laboratory Knelson
concentrator from IMPRC, Iran. The Knelson concentrator bowl is made from polyurethane
and consists of 5 collecting rings with different sizes in diameter. The bowl is installed
concentrically within a cylindrical stainless steel outer bowl. The detailed geometry of this
bowl is given in Table 1. The bowl geometry was built using a CAD package (CATIA V5R20)
and imported as a triangular surface mesh in the DEM simulation software. Figure 1 shows a
picture and 3D view of geometry for this bowl. Figure 2 shows the STL mesh created on this
bowl geometry by Gmsh software, a three-dimensional mesh generator. This mesh is
constituted of 90000 triangular cells.

Table 1. Inside dimensions of the Knelson bowl from the bottom to the top ring [6].
Ring No 𝐷1 (cm) 𝐷2 (cm) 𝐷3 (cm) 𝐴(cm2) 𝑆(cm) 𝑉(cm3) Number of holes
5 7.5 6.2 5.0 0.88 1.95 17.2 34
4 6.8 5.6 4.4 0.86 1.76 15.1 32
3 6.2 5.0 3.8 0.75 1.57 11.8 30
2 5.6 4.4 3.2 0.72 1.38 09.9 28
1 5.0 3.8 2.6 0.71 1.19 08.4 26
Average 6.2 5.0 3.8 0.78 1.57 12.5

In the table𝐷1 , 𝐷2 and 𝐷3 are diameter of the bottom of a ring, diameter of the central line of a
ring and diameter of the inner rim of a ring respectively, 𝐴 is the average of cross-sectional
area of a ring, which is almost the same for each of the five rings, 𝑆 is the length of the central
line of ring and 𝑉 is Volume of ring. The slope angle of the bowl is about 15° and there are
totally 150 fluidization water holes with average 0.3 mm diameter on the bowl.
The Knelson concentrator separation process was simulated using 5000 feed particles in the
form of a binary mixture of heavy and light particles. In order to investigate the effect of
changes in relative centrifugal force and feeding rate, iron and quartz particles mixture was
selected as feed while for separation cycle time simulations, feed was constituted of heavy
platinum and light quartz particles. In these simulations, all particles were spherical and had
the same size of 0.72 mm in diameter. The material properties and model physical parameters
are shown in Table 2 and Table 3. The values of operational variables and feed properties used
in simulations are listed in Table 4.
Two different heavy particles contents in the feed was considered to mimic feed grade. For
feeding rate simulations this content was 10% and for other simulations it was 15%. In all
simulations, the separation cycle time which was the sum of feeding and collecting times,
actually depended on the particles number and feeding rate. The separation cycle time was
between 1 to 4 seconds. For calculation of the separation performance, the concentrate grade
refers to the mass of collected heavy particles in the bowl riffles to the total mass of all particles
inside the bowl after separation. Total recovery is also defined as the mass of remained heavy
particles inside the bowl to their mass in the feed.
The movement of the particles in the Knelson concentrator bowl is three dimensional. Feed
particles were inserted randomly into the bowl from a constant height of 3 cm from top of the
bowl. As soon as particles are fed on the bottom of the rotating conical bowl, they are
accelerated radially, axially and tangentially and move in a spiral motion along the near vertical
sides of the inner bowl that is 75o to the horizontal. Most of heavy particles are thrown
outwards because of the effect of the centrifugal force acting on the particles to maintain a
circular path. As a result some of the particles get trapped within the grooves of the bowl
forming a concentrate bed while a big part of light particles with a few number of heavy ones
continuously flows upward into the tailings stream. After the complete insertion of particles,
we stopped the rotation of bowl meaning end of concentration process and ran the simulation
for more 0.5 s allowing particles out of the bowl to settle down. Figure 3 demonstrates
snapshots of simulations of separation process in the Knelson concentrator bowl at the
beginning, middle and end of simulation time.

Figure 1. A picture and 3D transparent illustration of LKC bowl from side and top views

Figure 2. Triangular mesh of LKC bowl used in DEM simulations in STL format
Table 1. Material properties used for the DEM simulations of the LKC
Young’s modulus
Material Density(kg·m-3) Poisson ratio
(MPa)
light particle (quartz) 2650 0.30 5
heavy particle (iron) 5800 0.40 71
heavy particle (platinum) 21450 0.38 168
Knelson bowl
1200 0.50 0.025
(polyurethane)

Table 3. Model physical parameters used for the DEM simulations of the LKC

Physical parameter value


coefficient of restitution 0.30
sliding friction coefficient 0.50
rolling friction coefficient 0.05
1
time step (s)
× 10−5

Table 4. Operational parameters of the feed and Knelson concentrator

Parameter Value
Feed particles number 5000
Particle size (mm) 0.720
Heavy particles content in feed 10, 15
(%)
Feed rate (g·min-1) 90, 170, 250, 500
relative centrifugal force (g- 40 ×g, 50 ×g, 60 ×g, 80 ×g, 120
force) ×g,
bowl rotation speed (rpm) (1194, 1335, 1460, 1715, 2070)
separation cycle time (s) 1, 2, 3, 4

3. Results and discussion


In overall, 16 simulation runs were designed and performed in this research to investigate the
effects of the variables (Table 5). Among these simulations 3 runs are replicated to check the
reproducibility of the results. We ran our simulations in parallel using 8 number of 3.2 GHz
processors, 16 GB of RAM, and message-passing techniques (MPI) as well as spatial-
decomposition of the simulation domain. In these simulations, the grade of concentrate
accumulated in the bowl and total recovery of heavy particles were calculated. In Table 5 the
bold values show the differences between the simulation conditions and the star sign represent
the replicated simulations. The replicated tests showed there is just a maximum 2.5% difference
in the results which shows the reproducibility is appropriate.
4.1.Effect of relative centrifugal force
Figure 5 shows the effect of relative centrifugal force intensity on the concentrate grade and
total recovery. It is obvious by increasing centrifugal force intensity, the grade of concentrate
reduces whereas the total recovery increases. For example by doubling the relative centrifugal
force intensity from 60 × g to 120 × g, concentrate grade dropped 5 % while total recovery
grows about 5 %. When the bowl rotation speed is higher, more centrifugal force is on the light
quartz particles so they remains in the bowl and decrease the concentrate grade. With more
relative centrifugal force intensity, further number of feed heavy particles are captured by the
bowl rings so the process total recovery increases.
4.1.Effect of feeding rate
As it is shown in Figure 6, when the feeding rate increases, the total recovery decreases due to
the reduction in retention time of particles inside bowl. By reducing retention time, all particles
have less opportunity to react with centrifugal force, therefore less particles are collected in the
rings. Despite of the total recovery, the concentrate grade has no distinct trend. It is at the lowest
value when the recovery is maximum and vice versa it is at the most when the recovery is
minimum.
4.2.Effect of separation cycle time
To investigate the influence of separation time on the concentrate grade and total recovery we
continued simulation number 1 in Table 4 for more separation times of 2, 3 and 4 seconds.
Figure 7 demonstrates that with increasing the separation time from 1 s to 2 s, both grade and
total recovery increased more than 8 % and 11 % respectively, but after 2 s any further increase
in separation time has a negligible impact on the concentrate grade and total recovery and
almost they reach to constant values. This phenomenon shows that at 2 s after the beginning of
process, separation completes nearly and there is no more light particles to be outpoured. At
this moment, it can said that Knelson concentrator has reached to a steady state.
4.3.Validation
It is essential that both DEM simulations and the DEM solver are validated properly and
adequately. In general a comprehensive validation of DEM solver and simulations is not
possible and in most cases it can be done only partially. In this study, in order to validate the
simulation results, a number of experimental separation tests on synthetic sample of pure quartz
and magnetite under specific conditions (Table 8) close to simulations conditions were carried
out using a laboratory 7.5 cm Knelson concentrator from IMPRC, Iran. Both samples were
provided by local silica and iron ore producers. The chemical compositions of both samples
are shown in Table 6 and Table 7. Quartz and magnetite had specific gravities of 2.65 g·cm-
3and 4.8 g·cm-3 respectively. Experimental results are listed in Table 8. Figure 8 demonstrates
proper and adequate validation occurred by making a comparison between the concentrate
grade and total recovery of simulations 3, 5 and 11 conducted by LIGGGHTS software and
obtained results from experiments. The differences between experimental and simulation
results are less than 10 % which shows a relatively high agreement indicating both data and
DEM solver validity.

Table 6. Chemical composition of magnetite sample

Unit SiO2 MgO Al2O3 P2O5 SO3 K2O CaO Fetotal FeO L.O.I
% 2.15 1.07 2.12 0.03 0.09 0.21 1.65 67.2 21.19 1.46

Table 7. Chemical composition of quartz sample


Unit SiO2 Al2O3 K2O CaO Fe2O3 L.O.I
% 93.61 4.15 0.64 0.52 0.36 0.73

Table 8. Conditions and results of experimental tests


Feed size Magnetite Feed rate centrifugal Fluidization Grade Recovery
Test No
𝑑50 (μm) content (%) )g·min-1) force (×g) water (l·min-1) (%) (%)
1 720 15 170 60 15 20.25 89.30
2 720 15 170 120 15 17.30 95.50
3 720 10 170 50 15 18.30 87.90

Figure 3. Snapshots showing the motion of iron and quartz particles inside the LKC with
particles colored by their mass, feed particles number: 5000, weight ratio: 0.10 to 0.90, feed
rate: 500 g·min-1, particles diameter: 0.72 mm, bowl speed: 1335 rpm: (a) at the beginning of
the simulation (b) at the middle of the simulation (1.5s); (c) at the end of the simulation
Table 5. Separation performance of LKC under different operational conditions

Type of Heavy Relative Feed


Simulation Particles separation Feeding rate Concentrate Total
heavy particles centrifugal particle size
number number cycle time (s) (g·min-1) grade (%) recovery (%)
particles content (%) force (×g) (mm)
1 5000 iron 15 40 1 170 0.72 21.70 91.50
2 5000 iron 15 50 1 170 0.72 20.25 92.35
3 5000 iron 15 60 1 170 0.72 19.45 93.8
4 5000 iron 15 80 1 170 0.72 17.30 96.50
5 5000 iron 15 120 1 170 0.72 16.50 98.50
6 5000 platinum 15 40 1 170 0.72 41.13 74.45
7 5000 platinum 15 40 2 170 0.72 49.37 85.50
8 5000 platinum 15 40 3 170 0.72 49.60 86.26
9 5000 platinum 15 40 4 170 0.72 49.70 86.50
10 5000 iron 10 50 1 90 0.72 13.45 98.7
11 5000 iron 10 50 1 170 0.72 17.55 92.3
12 5000 iron 10 50 1 250 0.72 14.45 90.6
13 5000 iron 10 50 1 500 0.72 19.7 88.6
14* 5000 iron 10 50 1 500 0.72 19.2 88.2
15* 5000 iron 15 40 1 170 0.72 21.70 93.80
16* 5000 iron 10 50 1 170 0.72 17.75 91.92
Figure 6. The effect of feeding rate on the concentrate grade and total recovery

Figure 7. The effect of separation cycle time on the concentrate grade and total recovery

Figure 5. The effect of centrifugal force intensity on the concentrate grade and total recovery
Figure 8. Comparison between the predicted concentrate grade and total recovery in simulations
3, 5 and 11 and laboratory measurements
Conclusion
DEM simulations of Knelson concentrator were run with LIGGGHTS open source software to
study the effects of operating parameters including relative centrifugal force intensity, feeding rate
and separation cycle time on concentrate grade and total recovery as separation performance
criteria. Our investigations demonstrated that DEM can be used efficiently as a tool for
understanding of Knelson concentrator separation process at particle scale and predicting the
variations induced by operational variables. Based on the presented simulation results, the
following conclusions may be drawn:
1. By increasing the LKC bowl rotational speed, the concentrate grade will declines linearly
due to the collection of more light particles within the bowl by higher RCF while the total
recovery will be added.
2. When the feeding rate increases, the concentrate grade has no distinct trend but the total
recovery decreases because of the reduction in retention time of particles inside the bowl.
3. With increasing the concentration time from 1 s to 4 s, both the concentrate grade and the
total recovery increase until 2 s, but after that further increase in concentration time has no
effect on both separation indices.
In order to investigate the validity of the simulation results, experimental tests on synthetic sample
of pure quartz and magnetite under specific conditions were carried out. Comparison of the results
of simulations conducted by LIGGGHTS software with those from experiments indicates a high
agreement. This validates the simulation results and the DEM solver computations.

References
[1] Knelson, B., the Knelson Concentrator: Metamorphosis from crude beginning to sophisticated
worldwide acceptance, Miner. Eng. 5 (1992) 1091–1097.
[2] Burt, R. O., Korinek, G., Young, S. R., and Deveau, C, 1995, Ultrafine tantalum recovery
strategies, Minerals Eng., 8(8), pp. 859–870.
[3] Majumdera, A.K., Barnwala, J.P, 2006, modelling of enhanced gravity concentrators-present
status, Mineral processing and Extractive Metallurgy Review: An International Journal, Volume
27, Issue 1, 61-86
[4] Koppalkar, S., 2009, effect of operating variables in Knelson Concentrator: a pilot-scale study,
Ph.D. Thesis, McGill University, Montreal, Canada.
[5] Ling, J., 1998, a study of a variable speed 3-in Knelson Concentrator, Ph.D. Thesis, McGill
University, Canada.
[6] Huang, L., 1996, Upgrading of gold gravity concentrates a study of Knelson Concentrator,
Ph.D. Thesis, McGill University, Canada.
[7] Laplante, A.R., Shu, Y., Marois. J., 1996a, Experimental Characterization of a Laboratory
Centrifugal Separator, Canadian Metallurgical Quarterly, Vol. 35.No. 1. pp. 23-29.
[8] Zhao, L.L., Liu, CS., Yan, J.X., Xu, Z.P., 2010, Numerical simulation on segregation process
of particle using 3D discrete element method. Acta Phys Sin, 59(3), 1870–6.
[9] Cundall, P.A., Strack, O.D.L., 1979. A discrete numerical model for granular assemblies.
Géotechnique 29, 47–65.
[10] Walton, O.R., 1994, Numerical simulation of inelastic frictional particle-particle interaction,
Particulate two-phase flow, Roco, M., (Ed.), Butterworth-Heinemann, Boston MA, 884-991.
[11] Barker, G.C. 1994. Computer simulations of granular materials, In Granular Matter: An
Interdisciplinary Approach. Edited by Anita Mehta. New York: Springer-Verlag.
[12] Cleary, P.W., 2009b. Industrial particle flow modelling using discrete element method. Eng.
Computation. 26, 698–743.
[13] Cleary, P.W., Sinnott, M.D., 2008. Assessing mixing characteristics of particle-mixing and
granulation devices. Particuology 6, 419–444.
[14] Goniva, C., Kloss, C., Deen, N.G., Kuipers, J. A.M., Pirker, S., 2012. Influence of rolling
friction on single spout fluidized bed simulation. Particuology 10, 582– 591.
[15] FrantzDale, B., Plimpton, S.J., Shephard, M.S., 2010. Software components for parallel
multiscale simulation: an example with LAMMPS. Eng. Comput. 26, 205–211.
[16] Mechtcherine, V., Gram, A., Krenzer, K., Schwabe J.-H., Shyshko, S., Roussel, N., 2013.
Simulation of fresh concrete flow using Discrete Element Method (DEM): theory and applications.
Mater. Struct. RILEM Publications, DOI 10.1617/s11527-013-0084-7.
[17] M. Jahani, A. Farzanegan, M. Noaparast, Investigation of screening performance of banana
screens using LIGGGHTS DEM solver, Powder Technol. 283 (2015) 32–47.

Vous aimerez peut-être aussi