Sie sind auf Seite 1von 14

Minerals Engineering 85 (2016) 92105

Contents lists available at ScienceDirect

Minerals Engineering
journal homepage: www.elsevier.com/locate/mineng

Cone crusher modelling and simulation using DEM


Johannes Quist , Carl Magnus Evertsson
Dept. of Product and Production Development, Chalmers University of Technology, SE-412 96 Gteborg, Sweden

a r t i c l e

i n f o

Article history:
Received 10 May 2015
Revised 27 October 2015
Accepted 2 November 2015
Available online 7 November 2015
Keywords:
Cone crusher
DEM
Simulation
Calibration
Validation

a b s t r a c t
Compressive crushing has been proven to be one of the most energy efficient principles for breaking rock
particles (Schnert, 1979). In this paper the cone crusher, which utilizes this mechanism, is investigated
using the discrete element method (DEM) and industrial scale experiments. The purpose of the work is to
develop a virtual simulation environment that can be used to gain fundamental understanding regarding
internal processes and operational responses. A virtual crushing platform can not only be used for understanding but also for development of new crushers and for optimisation purposes.
Rock particles are modelled using the bonded particle model (BPM) and laboratory single particle
breakage tests have been used for calibration. The industrial scale experiments have been conducted
on a Svedala H6000 cone crusher operated as a secondary crushing stage. Two different close side settings
have been included in the analysis and a high speed data acquisition system has been developed and used
to sample control signals such as pressure and power draw in order to enable detailed comparison with
simulation results. The crusher has been simulated as a quarter section with a batch of breakable feed
particles large enough to achieve a short moment of steady state operation. Novel methods have been
developed to estimate the product particle size distribution using cluster size image analysis. The results
show a relatively good correspondence between simulated and experimental data, however further work
would be need to identify and target the sources of observed variation and discrepancy between the
experiments and simulations.
2015 Elsevier Ltd. All rights reserved.

1. Introduction
The cone crusher is the most common machine for secondary
and tertiary crushing stages in both the aggregates industry and
minerals processing comminution operations. The performance
and efficiency of these machines are hence of great importance
to the industry. Analytical models for estimating cone crusher performance have been developed by Evertsson (2000) and Eloranta
(1995). Others have contributed to the modelling and understanding of wear (Lindqvist, 2005), product quality (Bengtsson, 2009)
and process optimisation (Hulthn, 2010). These mechanistic or
semi-mechanistic analytical models are derived from first principle
equations as well as empirical data gained from those measurements and observations that are possible to perform in industrial
or laboratory environments. However, the internal mechanics
Abbreviations: BPM, bonded particle model; DEM, discrete element method;
PBRM, population balance replacement model; CSS, close side setting; OSS, open
side setting; CAD, computer aided design; CPUH, CPU processing hours; SPB, single
particle breakage; IPB, interparticle breakage.
Corresponding author.
E-mail address: johannes.quist@chalmers.se (J. Quist).
http://dx.doi.org/10.1016/j.mineng.2015.11.004
0892-6875/ 2015 Elsevier Ltd. All rights reserved.

and flow behaviour of rock particles are difficult to measure


directly in comminution machines. Numerical methods such as
the discrete element method (DEM) enables such further analysis
of the internal conditions and dynamics inside machines handling
particulate material. The DEM was originally proposed by Cundall
and Strack (1979) and has since then contributed to the understanding of many machines and processes in the area of comminution (Weerasekara et al., 2013).
In this paper a cone crusher is modelled and simulated using
DEM in the commercial software EDEM provided by DEMSolutions Ltd. Cone crushers and gyratory crushers have previously
been the subject for DEM modelling and simulation. Lichter et al.
(2009) successfully modelled a laboratory Nordberg B90 cone
crusher. A Hydrocone type cone crusher was modelled by Quist
and Evertsson (2010) using the BPM model with spherical cluster
particles in order to investigate the influence of eccentric speed
on mass flow and hydrostatic pressure. The bonded particle model
approach for rock breakage modelling was further developed with
real rock particle shapes based on 3D scanned rock particles in a
study where the liner wear on a primary gyratory crusher was
investigated (Quist et al., 2011). The same approach for rock

J. Quist, C.M. Evertsson / Minerals Engineering 85 (2016) 92105

93

Nomenclature
F sc
F nc

rcr
scr
s

kb
n
kb
J
~
Fi
i
n
t i
Lb
Rb
~
F i;b
M nb

critical bond shear force


critical bond normal force
critical normal strength
critical shear strength
shear bond stiffness
normal bond stiffness
bond beam moment of inertia
contact resultant force
normal unit vector
tangential unit vector
bond beam length
bond disc radius
bond resultant force
bond normal torque

particle modelling was further on applied to a H6000 cone crusher


(Quist, 2012). Other attempts on modelling the cone crusher have
recently by presented by Li et al. (2014), Delaney et al. (2015) and
Cleary and Sinnott (2015).
There are mainly three different approaches for modelling the
rock material breakage in DEM. The most commonly used is based
on a population balance replacement model (PBRM) approach
where a particle is replaced by a set of progeny particles if a load
constraint is reached. The size distribution of the progeny particles
can be calibrated according to breakage test data. The second
approach is the bonded particle model (BPM) proposed by
Potyondy and Cundall (2004). This approach is based on bonding
a cluster of sub-particles together using imaginary beams between
each particle in contact. The third approach is based on modelling a
particle using tetrahedral mesh elements with compliant contacts
and stress constraints in both normal and tensile directions. A
comprehensive review on different rock particle breakage modelling approaches have been provided by Lisjak and Grasselli
(2014).
All these methods for modelling rock breakage have a set of
drawbacks and advantages. The replacement strategy is less computationally expensive than the BPM model since it requires fewer
particles in total. It is also relatively easy to calibrate as an experimentally derived breakage function can be applied directly. However, since the particle is removed at a breakage event the particle
dynamics of the crushing sequence will be affected. The newly
generated progeny particles in a breakage event will either have
to be created with zero velocity or with some initial velocity. As
a consequence the continuity of a particle element trajectory
through the crusher will be broken at every breakage event.
Another drawback is that it is difficult to implement breakage
sensitivity to the loading condition on a particle in the PBRM. In
reality a particle may be subjected to many different types of loading conditions or patterns depending on how many other particles
that surrounds it. In the case of interparticle breakage the total
load on a particle may be distributed on many different contact
points. The number of contact points and the force level will determine the breakage probability and size distribution of the progeny
material (Schnert, 1979). In a single particle breakage mode with
two contact points a lower force would in most cases be required
to break a rock specimen than if loaded at e.g. 10 contact points,
distributing the load more evenly.
A BPM particle will naturally conform to any loads acting on it
and break if the loading constraint is reached. This means that it
may also be sequentially broken down in several steps. As already
mentioned the main drawback with the BPM approach is that all
sub-particles need to be included from the beginning of the

M sb

aF
rp

Fc
Fc
As
Un
Us
D

ai
rb
sb
rc

bond shear torque


mantle slope angle
estimated particle tensile strength
critical force for fracture
crushing force
hydrostatic bearing area
normal overlap
tangential overlap
SPB particle diameter
response model parameter estimate
response model critical normal strength variable
response model critical shear strength variable
response model particle tensile strength

simulation when creating the meta-particles which limits the total


number of meta-particles that can be simultaneously modelled.
Another limitation is that since a set of sub-particles are packed
together and bonded, the cluster will have a specific packing density. This results in a mass conservation problem that needs to be
compensated for.
1.1. Cone crusher theory
The functional principle of a cone crusher is to compress particles between two surfaces. Particles can be subjected to two different modes of compressive breakage; single particle breakage (SPB)
and inter-particle breakage (IPB). The particle bed in the interparticle breakage mode can either be confined or unconfined.
The compressive action is realised by inflicting a nutational
motion on an inner cone (mantle) while an outer cone (concave)
remains fixed. The nutational motion is actualised by an eccentric
bushing transferring the rotational motion of the drive shaft via a
gear ring to the main shaft, see Fig. 1. The mantle is allowed to
rotate freely around the main shaft axis. Therefore it will act as a
planet wheel in a planetary gear where the mantle is rolling on
the bed of rock particles. In real operations the rolling motion is
sometimes subjected to a slip behaviour. The close side setting
(CSS) is defined as the smallest gap between the mantle and concave and the open side setting (OSS) is defined as the largest gap
on the opposing side on the liners. The difference between the
CSS and OSS is defined as the eccentric throw (Evertsson, 2000).
2. Industrial experiments
The experiments were conducted at an aggregate producer outside of Gothenburg, Sweden. The operation surveyed is a secondary
crushing stage, treating primary crushed material from a jaw
crusher as feed material. The crusher close side setting was carefully calibrated manually at four different positions prior to each
test. The feed and product streams where sampled by 1 m belt cuts.
The moisture content for the feed material was measured to 1.5
2.3% and for the products produced 0.18%. The rock material is a
Granite-Gneiss with a mineralogical composition corresponding
to 46% Feldspar, 19% Amphibolite, 17% Quartz, 12% Biotite and 6%
pot. alkali-reactive materials as well as some traces of sulphides.
Each material sample was first split at a 45 mm aperture screen
in order to process the coarse and fine end separately. The +45 mm
material was sized by hand using specially made square shaped
sizers, see Fig. 3. The 45 mm material was first screened on a
large deck Gibson laboratory screener. The 8 mm material was
split, dried and sized according to European standard for determi-

94

J. Quist, C.M. Evertsson / Minerals Engineering 85 (2016) 92105

100
Feed

Fx

EXP CSS 34 mm

Fc

80

Fz

Cumulative Passing [%]

EXP CSS 50 mm

60

40

20

0
10

-1

10

10

10

Size [mm]
Fig. 2. Particle size distributions from the crushing experiments for close side
setting 34 mm and 50 mm and feed.

2.1. High frequency data sampling

Fig. 1. Schematic illustration of the cross-section of a Hydrocone type cone crusher.

nation of particle size distribution, EN 933-1 (1998). The resulting


particle size distributions for the experiments at 34 mm CSS and
50 mm CSS as well as the incoming feed can be seen in Fig. 2.

A data acquisition system has been developed for sampling


data at high frequency from the available crusher sensors.
Pressure, power draw, shaft position and temperature signals
have been sampled using opto-isolators splitting the signal
from the installed crusher control system. In this work only
the pressure and power draw signals have been analysed.
The motive behind using a secondary data acquisition system
instead of extracting data from the installed control system
is based on the suspicion of signal aliasing due to a low sampling rate.
A multifunctional data acquisition card (model: NI USB-6211)
from National Instruments was used for sampling the signals at
500 Hz. The card is connected via USB 2.0 interface to a laptop with
the NI software LabVIEW.

Fig. 3. Hand sizing of the +45 mm samples.

J. Quist, C.M. Evertsson / Minerals Engineering 85 (2016) 92105

95

2.2. Modelling geometry

3. Bonded particle model

The Svedala H6000 crusher has been modelled in the CAD software Catia. The manganese liner geometries were relatively worn
during the experiments and were replaced a week after the tests.
This opened up the possibility to 3D scan the mantle and concave
geometries in a time frame close to the actual tests. A Faro Focus
3D S 20 scanner was used and the scan point clouds were analysed
and registered using the software SCENE version 4.8. In Fig. 4 the
concave liner is seen as an unfolded view. The 3D scanned mantle
is shown in Fig. 5.
When performing a 3D scan the resulting output is a point
cloud. It is not possible to use a point cloud directly in a DEM simulation as representation of geometry. In order to utilize the scan
data a mesh needs to be created based on the point cloud. In this
work the liner profiles where generated by a circumferential averaging of the scan data and fitting of a spline function to the average
points. This results in rotational symmetrical liner geometries
which is needed when modelling a section of the crusher such as
in this case. The final CAD model including feed geometry arrangements can be seen in Fig. 6. The variation in CSS for the two simulated cases is achieved by moving the main shaft assembly
vertically until a satisfactory setting is attained.

The BPM model was published by Potyondy and Cundall (2004)


for the purpose of simulating rock breakage. The approach has
been applied and further developed by Cho et al. (2007). The concept is based on bonding or gluing a packed distribution of spheres
together forming a breakable body. The particles bonded together
will hereafter be called fraction particles and the cluster created is
defined as a meta-particle. The fraction particles can either be of
mono size or have a size distribution. By using a relatively wide
size-distribution and preferentially a bi-modal distribution the
packing density within the meta-particle increases (Groot and
Stoyanov, 2011). It is important to achieve as high packing density
as possible due to the problematic issue with mass conservation as
the clustered rock body will not be able to achieve full inherent
solid density. Also, when the bonded particle cluster breaks into
smaller fragments the bulk density will somewhat change as the
cluster breaks apart.
For each particle in contact a bonding beam will be created. This
means that if a larger fraction particle is surrounded and in contact
with several smaller fraction particles it will have multiple beams
attached to it in a star pattern. The forces and torques acting on the
theoretical beam can be seen in Fig. 7.

Fig. 4. Image of the 3D scanned crusher concave as an unfolded planar view. The scanner was placed inside the concave which enabled the capture of the full geometry in one
scan.

Fig. 5. Image of the 3D scanned crusher mantle.

96

J. Quist, C.M. Evertsson / Minerals Engineering 85 (2016) 92105

Bond breaks

Te

Bond breaks

ns
ion
Sh

ea

Compression

Fig. 8. Schematic forcedisplacement plot of the different modes of loading on a


bond beam. (Modified from Cho et al. (2007)).
n
d~
F n;b kb ADU n
t
d~
F t;b k ADU t

b
t

dM nb kb J DHn
n

dM sb kb 2J DHn
Fig. 6. Image of the developed CAD model used in the DEM simulations. The main
shaft length is 2180 mm and the mantle maximum diameter is 1440 mm.

The schematic graph in Fig. 8 illustrates the relationship


between different loading modes (tension, shear, and compression), bond stiffness and strength criteria. The line in the
second quadrant corresponds to the relationship between normal deformation Un, and the normal force Fn,b. The bonds
experience a tension force when the particles are pulled apart
and a compressive force when pressed together. The line in
the first quadrant corresponds to the relationship between
tangential deformation US and the tangential force Ft,b. The
slope of the lines corresponds to the stiffness of the bond
beam.Before bond formation and after bond breakage the particles interact according to the HertzMindlin no slip contact
model. The bonds are formed between particles in contact
at a pre-set time tbond. When the particles are bonded the
forces and torques are adjusted incrementally according to
the Eqs. (1) and (2).

(a)

where

DU n mn dt
DU t mt dt
DHn xn dt
DHt xt dt
A pR2b

J 12 pR4b
The normal and shear stresses are computed and evaluated if
exceeding the critical stress constraint values according to Eq. (3).
~

2M
 max F n;total
r
J b Rb < rcr
A
~

2M
smax F t;total
J b Rb < scr
A

In this work the critical strength levels are set to a single value
defining the rock strength. For future work it would be preferable
to define the critical bonding strength according to a probability
distribution corresponding to the distribution of the variance of
experimental breakage tests.

(b)

Fig. 7. Schematic representation of; (a) two particles overlapping when interacting giving a resultant force (~
F i ) according to the HertzMindlin contact model. (b) two
particles bonded together with a cylindrical beam leading to a resultant force (~
F i;b ) as well as normal (M nb ) and shear (Msb ) torques (modified from (Cundall and Strack, 1979;
Potyondy and Cundall, 2004)).

97

J. Quist, C.M. Evertsson / Minerals Engineering 85 (2016) 92105

- --- --

i: Mono distribution

ii: Gaussian distribution

- -- - - -

-- - -

iii: Bimodal distribution

Fig. 9. Schematic illustration of three types of packing structures.

The size distribution of the fraction particles building up the


meta-particles needs to be chosen so that a sufficient quality of
breakage is achieved while keeping the computational economy
within the limiting constraints of the available computing
resource. A schematic illustration of three different packing structures can be seen in Fig. 9. It has been found that the bimodal distribution gives the best possibilities for realising high packing
density, good breakage characteristics using the least number of
spheres. This finding is also supported in the literature
(Brouwers, 2006; Groot and Stoyanov, 2011). The decision on size
distribution and its resulting packing structure could be
100
Fraction Particle Size

Cumulative Passing [%]

80

60

40

20

formulated as a multi-objective optimisation problem. However,


in this work the bimodal size distribution of the fraction particles
was found iteratively until a satisfactory result could be obtained
balancing the total number of meta-particles and the total number
of fraction particles in each meta-particle. In rock mechanics investigations where DEM is used to simulate e.g. a uniaxial strength
test the researcher has the opportunity to include all the particles
in just one rock specimen. In this case however, one specimen is
not enough hence in order to be able to simulate a large enough
batch of feed particles it is not possible to include too many fraction particles in each cluster.
In Fig. 10 the size distribution of the fraction particles used to
build up the meta-particles is shown. The distribution is created
using two normal distributions that are overlapping hence the final
distribution is not distinctly bi-modal. The large part of the bimodel distribution was set to a mean sphere diameter size of
18 mm and a standard deviation of 1.8 mm. The smaller part of
the bimodal distribution was set to a mean size of 4.8 mm and
standard deviation of 1.2 mm. The packing density of the fraction
particle bed was calculated to 0.76.
When a satisfactory packing structure has been created it can
be used to form the meta-particles. A 3D model of a rock is used
as a selection space as seen in Fig. 11. The position and radius of
each fraction particle within the selection space is exported to
form a meta-particle cluster model that can be used in the crusher
simulation. In this work two different meta-particle sizes and
shapes have been developed corresponding to the feed distribution
in the experiments.
4. BPM calibration

0
0
10

10

Size [mm]
Fig. 10. Particle size distribution for the fraction particles in the bonded particle
model.

The first step the calibration is to first performing a series of laboratory compressive breakage experiments to establish the
strength distribution of the rock material. Secondly, a set of simulations are performed using Design of Experiments (DoE) and

Fig. 11. Illustration of how a 3D scanned rock model is used as a selection space for creating meta-particle clusters with realistic shape.

98

J. Quist, C.M. Evertsson / Minerals Engineering 85 (2016) 92105

response modelling. Finally, the error between the experimental


data and the response model output is minimised using an optimisation approach. The optimizer is used as the calibrated strength
variables for the BPM model.
4.1. Laboratory breakage experiments
Single particle compressive breakage experiments of irregular
rock specimens were conducted in order to estimate the strength
of the material studied. When testing rock particles the irregularity
leads to a high variance in terms of the critical force needed to
break a particle of a certain size. This effect can be seen in
Fig. 12b where the particle height is plotted against the critical
breakage force. The data suggests an increased force variance with
increasing particle size.
Hiramatsu and Oka (1966) investigated the tensile strength for
irregular as well as spherical shapes and showed that the tensile
particle strength for an irregular shaped rock can be approximated
by Eq. (4).

0.995
0.99

log normal fitness line

Log Normal Probability

data points

0.95
0.9
0.75
0.5
0.25

2:8F c

pD2

This simple equation is derived from a more complex expression of the stress state of a sphere subjected to compression. The
numerator is defined as the critical force for failure times a factor
given by the loading condition, geometrical features and poisons
ratio. The denominator is defined as a disc-area where D is the distance between the loading points. An average value of 13.3 MPa for
the tensile particle strength was calculated from the data using Eq.
(4). The calculated critical strength data fits well to a log normal
distribution as can be seen in Fig. 12a.
4.2. Calibration breakage simulations
The calibration of the bonded particle model is based on simulations of 33 different factor combinations. The design of experiments (DoE) methodology has previously applied for calibration
of bonded particle models by (Hanley et al., 2011; Yoon, 2007). A
n

stiffness ratio kb =kb has been set to 2.5 according to results presented by Yoon (2007) and Wang and Tonon (2009). The value
intervals for each calibrated factor can be seen in Table 1.
A cylindrical test specimen shape was used resembling a Brazilian splitting tensile strength test (ASTM, 2008). In Fig. 13 the bond
structure and force network between the compression plates can
be seen. Broken bonds are coloured black and can be seen in the
line contact regions.
A least square response model has been fitted to the simulation
data set using the statistics software SAS JMP 10 (SAS, 2012). The
model prediction expression can be seen in Eq. (5). Firstly the statistically significant terms were screened out. Those terms with
significant contrasts are then included as variables in the model.

rc a0 a1 sb a2 knb a3 rb a4 sb  s2 a5 knb  knb sb


 a6 rb  r
 b
s

0.1
0.05
0.01
0.005

10

10 1
p

10

The response model estimate values and statistical significance


are shown in Table 2. The results indicate that the critical shear
strength is the most influential parameter.
The response modelling summary of fit and ANOVA table can be
seen in Table 3 and the statistical fit is reasonably good. Potential
reasons to the observed lack of fit may be too few observations
or that this kind of relatively simple empirical model is not sophisticated enough to represent the observed behaviour. However, for
the purpose of finding a set of parameters that gives a reasonably
good correspondence to the real particle strength the model is
competent enough.
In Fig. 14 the response model is presented as ISO-surface plots
for a set of different levels for rb. The model rationale seen in the
surface plots corresponds well with the observed behaviour from
simulations. An increment in shear and normal bond strength
results in increased particle strength. High stiffness results in a
high force on the bonding beam (given a specific deformation)
and hence lower particle strength compared to the opposite case.
The final step of the calibration is to find the best set of parameters that minimizes the error deviation between the simulation

(a)
2

data [MPa]

60

50

40

Fc [kN]

rp

30

20

10
Table 1
Value intervals for calibrated factors.

(b)
0
0

10

20

30

40

50

60

Particle Height [mm]


Fig. 12. (a) Diagnostic plot of the experimental critical strength data to a log normal
distribution with estimates for scale parameter l = 2.46 and shape parameter
r = 0.51. (b) Scatter plot of particle height and corresponding critical breakage force
for all the SPB tests.

Factor

Notation

Value interval

Unit

Shear stiffness
Normal bond stiffness

G
n
kb

[0.357, 0.536]
[1252, 2004]

(GPa)
(GPa/m)

Normal strength
Shear strength
Stiffness ratio

rb
sb

[10, 50]
[1, 30]
2.5

(MPa)
(MPa)
()

n
t
kb =kn

99

J. Quist, C.M. Evertsson / Minerals Engineering 85 (2016) 92105


Table 2
Response model parameter estimates.
Term ai

Estimate

Std error

T Ratio

Prob >|t|

a0
a1
a2
a3
a4
a5
a6

26.325
0.602
0.0185
0.210
0.0189
0.00279
0.0215

6.277
0.0580
0.0033
0.0647
0.00930
0.000648
0.00855

4.19
10.38
5.59
3.24
2.03
4.3
2.51

0.0003
<.0001
<.0001
0.0033
0.0526
0.0002
0.0186

Table 3
Summary and ANOVA table for the BPM calibration response model.
Summary of fit

Value

R2
R2 Adj
Root mean square error
Mean of response
Observations

0.857
0.824
2.285
10.372
33

Source

DoF

Sum of squares

Analysis of variance (ANOVA)


Model
6
814.2222
Error
26
135.8018
C. Total
32
950.0239

Mean square

F Ratio

135.704
5.223

25.9812
Prob > F
<.0001

At the moment the full calibration procedure needs to be performed for every new mineral type simulated. Future work should
target this by generating a master response model for the BPM
model based on the Brazilian tensile strength test and uniaxial
strength test. In this way the BPM parameters corresponding to
standardized material data can be retrieved, at least for similar
mineralogical compositions and textures.

Fig. 13. The image shows the beginning of the breakage process in a BPM
calibration simulation of a cylindrical disc. The red colour denotes normal force
level on the bonds and black denotes bonds which are broken. (For interpretation of
the references to colour in this figure legend, the reader is referred to the web
version of this article.)

and experiment. The optimisation problem was solved by using


the MATLAB optimisation toolbox and the fmincon solver for
constrained nonlinear minimisation.

5. Particle size analysis of simulated discharge


When meta-particles have been broken into pieces they form
clusters of surviving progeny particles together with free fraction
particles. Some clusters survive and leave the crushing chamber
in the discharge region as the coarse end of the product particle
size distribution. The total product size distribution is composed
by both the surviving clusters as well as the free fraction particles
no longer bonded to other particles. This means that the final

Fig. 14. ISO-surface of the response model rc (applicate) plotted against sb (ordinate) and kb (abscissa) at a set of different levels of rb (colourbar). (For interpretation of the
references to colour in this figure legend, the reader is referred to the web version of this article.)

100

J. Quist, C.M. Evertsson / Minerals Engineering 85 (2016) 92105

(a)

(b)

(c)

Fig. 15. Example of three steps of image analysis processing from the CSS 34 mm simulation. (a) Image of non-broken bonds. (b) Image cropped to discharge region and an
image processing filter applied in order to solidify the clusters. (c) The image analysis algorithm has identified particle outlines and discarded edge particles.

product size distribution is biased by the fraction particle portion


in at least three ways:
 The size distribution of the fraction particles is pre-determined
according to the bi-modal distribution previously presented.
 The fraction particles have a minimum size limitation
 The fraction particles are non-breakable hence any crushing of
these particles that should have happened will not be
accounted for.
5.1. Cluster size analysis
There is no easily available method to determine the size of surviving clusters in the discharge region. Instead the cluster sizes
must be determined using post-processing methods. Two different
methodologies have been evaluated in order to calculate the discharge cluster size distribution.
 Method A Cluster 3D analysis
 Method B Cluster 2D image analysis
In method A the shape and size of the clusters are determined
directly by the geometrical configuration of the cluster. Depending
on how the modelling of the bonded particle model is conducted
this can be more or less complicated to perform. In the BPM model
used in this work the same fraction particle ID-name was used for
all meta-particles. Hence clusters cannot be determined using sorting algorithms alone. Instead the following procedure has been
developed:
i. Create a discharge region selection space
ii. Export particle position (x, y, z) and radius in the selection
space. A time interval should be set so that the same cluster
is not evaluated twice.
iii. Identify each group of fraction particles in contact forming a
cluster
iv. Determine the equivalent size of the clusters
v. Calculate the particle frequency based size distribution
vi. Calculate the particle mass based size distribution
The analysis procedure in method A is relatively computationally expensive as the algorithm has to read through all the particle

data n  m times, where n is the number of rows or number of particles in each exported time-frame and m is the number of
exported time-frames. Step iv in the procedure also elucidates a
common problem within any research field that handles particles.
What is the size of an irregularly shaped particle? The answer is
that there exists no one-dimensional unit that will fully describe
the size of a three-dimensional object. Some kind of one dimensional value together with a shape representation in two dimensions needs to be used.
Table 4
Simulation parameters for the DEM simulations.
Parameter

Value

Unit

DEM material properties


Solid density
Shear stiffness
Poissons ratio

Rock
2630
5.57  108
0.35

Steel
7800
7.0  1010
0.3

(kg/m3)
(Pa)
()

Coefficient of static friction


Coefficient of restitution
Coefficient of rolling friction

Rock-Rock
0.5
0.2
0.001

Rock-Steel
0.7
0.25
0.001

()
()
()

BPM parameters
Normal stiffness
Shear stiffness
Normal critical stress
Shear critical stress
Bond disc radius

1670
667
36
24
6.4

(GPa/m)
(GPa/m)
(MPa)
(MPa)
(mm)

Machine
Crusher model
Chamber type
Eccentric throw
Eccentric speed
Close side setting
Section simulated
Liner geometry

Svedala H6000
CHX
48
5
[34,50]
1/4
3D scanned and modelled

()
()
(mm)
(Hz)
(mm)
()
()

Simulation
Time step
Write out frequency
No. of particles
Simulation time (CSS34)
Simulation time (CSS50)
CPU clock freq.
CPU cores

3e7
500
92,508
418
404
3.33
8

(s)
(Hz)
()
(CPUH)
(CPUH)
GHz
()

J. Quist, C.M. Evertsson / Minerals Engineering 85 (2016) 92105

5.2. Cluster image analysis


The second method applied to determine the size of the clusters
in the discharge region is based on using conventional image processing techniques. The DEM simulation environment makes it
possible to generate snapshots of the discharge region at given

101

time intervals, see Fig. 15. The size of the clusters can then be
determined by image analysis in a similar way as for optical particle size image analysis equipment where a curtain of particles is
photographed sequentially using a stroboscope light source.
The open source software ImageJ (Rasband, 2015) has been
used for the image analysis of the surviving cluster discharge.

Fig. 16. Front image from the 34 mm close side setting simulation with particles displayed as clustered bonds. Broken bonds are coloured black.

Fig. 17. Image from behind of the 34 mm close side setting simulation with particles displayed as clustered bonds. Broken bonds are coloured black.

102

J. Quist, C.M. Evertsson / Minerals Engineering 85 (2016) 92105

The algorithm calculates the projected area, outline circumference,


feret min, feret max and different shape parameters such as sphericity.The equivalent ellipsoid volume is calculated using the feret
max and feret min values on each identified particle in the image
analysis. The general ellipsoid function is given by Eq. (6).

4pabc
3

Since the third dimension is unknown in a 2D image analysis it


is assumed to be the mean between the feret max and the feret min
dimensions. The motivation for not using either the feret max or
feret min directly as representation for the third dimension is that
it is unlikely that an irregular shaped particle has two equal minimum or maximum dimensions. The ellipsoid equivalent volume
for a progeny cluster is then given by Eq. (7).

V elip



Rf Rf 
4p
Rf Rf 
3
2

SIM CSS 34 mm
EXP CSS 34 mm (coarse)

80

Cumulative Passing [%]

100

40

20

(a)
0

Further, the ellipsoid equivalent mass for a progeny cluster is given


by Eq. (8), where q is the solid density of the material.

10

The particle size distribution with respect to mass is given by


size sorting each cluster in the total cluster vector based on the
average between the feret min and feret max dimension. If the cluster size is within the interval xi1 6 xmean 6 xi the mass of the cluster will be added to a mass summation vector of length i, where i
denotes the size class. The mass summation vector can then be
used to calculate the mass based particle size frequency distribution and mass based particle size cumulative distribution. In this
work these mathematical and logical operations are realised using
MATLAB scripts.
6. Cone crusher simulations

10

10

Size [mm]
100
SIM CSS 50 mm
EXP CSS 50 mm (coarse)

80

Cumulative Passing [%]

M elip V elip q

60

60

40

20

In Table 4 a summary of the simulation parameters details is


presented. The simulations are prepared in a series of sequential
steps. First the meta-particle clusters are imported in two different
steps as two different sizes and 3D shapes have been used to model
the feed material. After the meta-particles are in place the bonds
are activated and the dynamics of the mantle are started. The mantle nutation motion is realised by a sinusoidal rotation around the
X- and Y-axis where one of the motions has an offset of 90 degrees.
In Figs. 16 and 17 the breakage process for the 34 mm CSS case
can be seen. Each meta-particle is broken down during several
breakage events. Due to the coarse feed material mostly single particle breakage can be observed. The observed choke level corresponds well to the position of the minimum cross-sectional area
of the chamber.

(b)
0
100

101

10 2

Size [mm]
Fig. 18. Comparison of the product size distribution between simulation and
experiment for (a) CSS 34 mm and (b) CSS 50 mm.

for explaining the discrepancy. Even though the simulated


34 mm case predicts a too coarse distribution the trend is in the
correct direction when comparing the simulated cases, i.e. the
34 mm case is finer than the 50 mm case. At this stage of modelling
development, that can be regarded as a good start.

7. Results
7.2. Hydrostatic pressure
In this section the results from performed experiments and simulations will be presented and compared.
7.1. Product size distribution
In Fig. 18 the experimental and simulated product size distributions for CSS 34 mm and 50 mm are compared. The experimental
size distributions are represented as the coarse end +8 mm. This
is done since the minimum size limit of the fraction particles
makes it impossible to predict the product size below that level.
The 34 mm CSS case displays poor correspondence to the experimental data whereas the 50 mm case shows a relatively good correspondence. Currently no hypothesis have been proven as viable

The hydrostatic pressure is a good indicator for the internal


loads in the crusher. In a Hydrocone type crusher the forces from
breaking the particles are transferred via the main shaft to the
eccentric bushing and bottom plain bearing. The pressure in the
hydraulic system supporting the plain bearing hence gives a good
indication of the loads in the crusher at any given moment. In
Fig. 19 the simulated and measured hydrostatic pressures are displayed for 1.5 s of operation. As can be seen the measured pressure
from the experiments displays a cyclic behaviour with the frequency of 5 Hz. This is the same frequency as the eccentric speed
of the crusher which indicates that the crusher must have been
working harder on one side of the crushing chamber. This is caused

103

J. Quist, C.M. Evertsson / Minerals Engineering 85 (2016) 92105

the hydraulic piston. Since a quarter section of the crusher has


been simulated it is natural that the cyclic behaviour is observed.
This can be regarded as an extreme misaligned feeding situation.
An offset for the simulated pressure has been applied to account
for the mass of the main shaft assembly.

by an unwanted miss-aligned feeding situation where more material reports to one side on the crusher than the other. This phenomenon was also observed visually during the experiments.
The simulated pressure is calculated by exporting the total force
acting on the mantle in the vertical direction divided by the area of

(a)
3

Pressure EXP CSS 34mm

Pressure [MPa]

Pressure SIM CSS 34mm

0
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.1

1.2

1.3

1.4

1.5

1.4

1.5

Time [s]

(b)
3

Pressure EXP CSS 50mm

Pressure [MPa]

Pressure SIM CSS 50mm

0
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.1

1.2

1.3

Time [s]
Fig. 19. The plot displays a comparison of simulated and measured hydrostatic pressure for 1.5 s of operation.

250

Power Draw [kW]

(a)

Power EXP CSS 34mm

200

Power SIM CSS 34mm

150
100
50
0
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.1

1.2

1.3

1.4

1.5

1.4

1.5

Time [s]
250

(b)

Power EXP CSS 50mm

Power Draw [kW]

200

Power SIM CSS 50mm

150
100
50
0
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.1

1.2

1.3

Time [s]
Fig. 20. The plot displays a comparison of simulated and measured power draw for 1.5 s of operation.

104

J. Quist, C.M. Evertsson / Minerals Engineering 85 (2016) 92105

When comparing the simulated and experimental pressure, see


Fig. 19, the data series are in the same value interval which at this
stage of modelling development is a good result. The similarity of
the 5 Hz cyclic behaviour is very distinct. The simulated mean
pressure is slightly higher in the CSS 34 mm case than in the
50 mm case.

correct value range and show the correct trend. The simulated
throughput capacity is very close to the experimental result in
the 50 mm case whereas a relatively large difference can be
observed for the 34 mm case. More work needs to be conducted
to provide further explanation to the observed discrepancies
between simulated and experiment data. Potential sources of
error are listed below:

7.3. Power draw


The power draw is an important performance measurement as
it indicates how much crushing work is done in the crusher. The
experimental power draw data have been measured directly at
500 Hz. When this work was conducted there was no calculation
algorithm for the power draw available in the DEM software. The
power draw can instead be calculated in different ways as presented below:
 Particle breakage energy over time
 Calculation of the total torque on the mantle from particlemantle interactions
 Calculation of the pressure or resultant force on the mantle and
using an external calculation model for the torque and further
on the power draw.
In this work the first and second alternative proved to be unsatisfactory hence the third alternative was used. The total force magnitude was exported from the simulation and the mechanical
model derived by Lindqvist (2005) was used to calculate the power
draw. The model is based on the mechanics and the actual geometrical dimensions of e.g. the angles of the mantle, shaft lengths and
eccentricity. The idling power in the experiment is unknown. However, as an approximation the lowest observed value during the
experiments has been used for applying an offset to the simulated
data.
The measured and simulated power draw can be seen in Fig. 20.
The measured power draw shows a cyclic behaviour with the frequency of 5 Hz in the same way as for the pressure data. This indicates that the misaligned feed also affects the power draw i.e. more
crushing work is done on one side of the crusher predominantly.
The power draw from the simulations display a cyclic behaviour
due to that a quarter of the crusher has been modelled.
No significant difference can be observed for the simulated
34 mm and 50 mm CSS cases. It may be so that the cycle peak values of the simulated power draw are a good indication for what the
level would be if simulating the full crusher.
The simulated and calculated power draw shows lower values
than the measured. However the cycle peak values are within
the correct range. If a full crusher would have been simulated
the data would range with a mean and amplitude of the peak values. The simulated power draw is slightly higher in the 34 mm case
than in the 50 mm case.
In Table 5 the results from simulations and experiments are
summarised. At this stage of modelling development it should
be considered a success that the simulated values are in the

 Error in the mantle dynamics properties


 Error in the power draw estimation model
 Error in the particle-geometry contact force and torque calculation due to influence of geometry mesh tessellation size.
 Too low feed material batch size
 High variance in the image analysis of surviving clusters due to
particle shape and orientation effects

8. Conclusions
A Svedala H6000 cone crusher has been investigated by means
of experimental measurements and DEM simulations. A BPM
model with a bi-modal fraction particle distribution has been
developed and calibrated against laboratory single particle breakage experiments. In order to enable careful measurement of power
draw and hydrostatic pressure data a high frequency data sampling system was developed.
Is has been demonstrated that the throughput capacity, power
draw, pressure and product particle size distribution can be estimated using DEM however more work is needed in order to
improve the accuracy of prediction. The bonded particle model
(BPM) is a good approach for modelling breakage in crushers however it demands a high computational load.
This work shows that it is possible to simulate a complex comminution device that utilizes a compressive breakage mode and
investigate the influence of a change in an operating variable.
The results presented strengthens and are in line with the theoretically derived analytical models provided by Evertsson (2000) in
terms of particle flow behaviour and breakage in the crusher
chamber.
In order to reach an improved correspondence between the
simulation and experimental reference system the following
aspects and actions may be considered:
 Investigate the influence of geometry mesh tessellation size and
its influence on contact force and torque calculations in the
DEM software.
 Perform Brazilian tensile strength experiments on rock drill
cores in order to reduce the variance from the SPB tests
 Further investigate the observed discrepancy between the simulated and experimental particle size distributions in the CSS
34 mm case.
 Perform more simulations with a Design of Experiments
approach in order to investigate the influence of eccentric
speed, eccentric throw and close side setting.

Table 5
Summary of experimental and simulated averaged results.

Pressure (MPa)
Power draw (kW)
Throughput capacity (t/h)
Specific energy (kWt/h)

Mean
Std Dev
Mean
Std Dev
Mean

SIM34

EXP34

SIM50

EXP50

1.105
0.831
85.17757
6.54
361.2
0.235818

2.018
0.606
131.5
30.97
268.3169
0.490092

1.071
0.796
83.25391
10.15
428.64
0.194228

1.314
0.626
96.01
39.57
437.2548
0.219574

J. Quist, C.M. Evertsson / Minerals Engineering 85 (2016) 92105

Acknowledgements
This work has been carried out within the Sustainable Production Initiative and the Production Area of Advance at Chalmers.
The support is gratefully acknowledged. The authors would also
like to acknowledge the support from Ellen, Walter and Lennart
Hesselmans foundation for scientific research and MinFo.
Discrete Element Method (DEM) simulations were conducted
using EDEM 2.5 particle simulation software provided by DEM
Solutions. Ltd., Edinburgh, Scotland, UK.

References
ASTM, 2008, D3967-08, Standard Test Method for Splitting Tensile Strength of
Intact Rock Core Specimens.
Bengtsson, M., 2009. Quality-Driven Production of Aggregates in Crushing Plants.
In: Dep. Product and production Development. Chalmers University of
Technology, Gothenburg, Sweden.
Brouwers, H.J.H., 2006. Particle-size distribution and packing fraction of geometric
random packings. Phys. Rev. 74.
Cho, N., Martin, C.D., Sego, D.C., 2007. A clumped particle model for rock. Int. J. Rock
Mech. Min. Sci. 44 (7), 9971010.
Cleary, P.W., Sinnott, M.D., 2015. Simulation of particle flows and breakage in
crushers using DEM: Part 1 Compression crushers. Miner. Eng. 74, 178197.
Cundall, P.A., Strack, O.D., 1979. A discrete numerical model for granular assemblies.
Gotechnique 29 (1), 4765.
Delaney, G.W., Morrison, R.D., Sinnott, M.D., Cummins, S., Cleary, P.W., 2015. DEM
modelling of non-spherical particle breakage and flow in an industrial scale
cone crusher. Miner. Eng. 74, 112122.
Eloranta, J., 1995. Influence of Crushing Process Variables on the Product Quality of
Crushed Rock. Tampere University of Technology, Tampere.
Evertsson, C.M., 2000. Cone Crusher Performance. In: Dept. of Machine and Vehicle
Design. Chalmers University of Technology, Sweden.
Groot, R.D., Stoyanov, S.D., 2011. Close packing density and fracture strength of
adsorbed polydisperse particle layers. Soft Matter 7, 47504761.

105

Hanley, K.J., OSullivan, C., Oliveira, J.C., Cronin, K., Byrne, E.P., 2011. Application of
Taguchi methods to DEM calibration of bonded agglomerates. Powder Technol.
210 (3), 230240.
Hiramatsu, Y., Oka, Y., 1966. Determination of the tensile strength of rock by a
compression test of an irregular test piece. Int. J. Rock Mech. Min. Sci. Geomech.
Abstr. 3 (2), 8990.
Hulthn, E., 2010. Real-Time Optimization of Cone Crushers. In: Dep. Product and
Production Development. Chalmers University of Technology, Gteborg.
Li, H., McDowell, G., Lowndes, I., 2014. Discrete element modelling of a rock cone
crusher. Powder Technol. 263, 151158.
Lichter, J., Lim, K., Potapov, A., Kaja, D., 2009. New developments in cone crusher
performance optimization. Miner. Eng. 22 (78), 613617.
Lindqvist, M., 2005. Wear in Cone Crusher Chambers. In: Dept. of Applied
Mechanics. Chalmers University of Technology, Sweden.
Lisjak, A., Grasselli, G., 2014. A review of discrete modeling techniques for fracturing
processes in discontinuous rock masses. J. Rock Mech. Geotech. Eng. 6 (4), 301
314.
Potyondy, D.O., Cundall, P.A., 2004. A bonded-particle model for rock. Int. J. Rock
Mech. Min. Sci. 41 (8), 13291364.
Quist, J.C.E., 2012. Cone Crusher Modelling and Simulation, Master Thesis In Product
and Production Development. Chalmers University of Technology, Gteborg.
Quist, J.C.E., Evertsson, C.M., Franke, J., 2011. The effect of liner wear on gyratory
crushing a DEM case study, In Computational Modeling 11, Falmouth.
Quist, J.C.E., Evertsson, C.M., 2010. Simulating Capacity and Breakage in Cone
Crushers Using DEM, In Comminution 10, Capetown, South Africa.
Rasband, W., 2015. ImageJ-Image Processing and Analysis in Java, 1.48v ed. National
Institutes of Health, USA.
SAS, I., 2012. JMP 10 Software.
Schnert, K., 1979. Aspects of the physics of breakage relevant to comminution, In
Fourth Tewksbury Symposium, Melbourne.
Standardization, E.C.f., 1998, EN-933-1, Tests for geometrical properties of
aggregates, Part 1: Determination of particle size distribution Sieving method.
Wang, Y., Tonon, F., 2009. Modeling Lac du Bonnet granite using a discrete element
model. Int. J. Rock Mech. Min. Sci. 46 (7), 11241135.
Weerasekara, N.S., Powell, M.S., Cleary, P.W., Tavares, L.M., Evertsson, M., Morrison,
R.D., Quist, J., Carvalho, R.M., 2013. The contribution of DEM to the science of
comminution. Powder Technol. 248, 324.
Yoon, J., 2007. Application of experimental design and optimization to PFC model
calibration in uniaxial compression simulation. Int. J. Rock Mech. Min. Sci. 44
(6), 871889.

Das könnte Ihnen auch gefallen