Minerals Engineering 33 (2012) 1319

Contents lists available at SciVerse ScienceDirect

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

Modelling and dynamic simulation of gradual performance deterioration

of a crushing circuit Including time dependence and wear
Gauti Asbjrnsson , Erik Hulthn, Magnus Evertsson
Chalmers Rock Processing Research, Chalmers University of Technology, SE41296 Gothenburg, Sweden

a r t i c l e

i n f o

Article history:
Available online 4 March 2012
Keywords:
Dynamic simulation
Crushing
Modelling
SysCAD
Wear

a b s t r a c t
The use of steady-state models in process simulation is a well-established method in many process
industries. Designing a large crushing plant by relying on steady-state simulations alone will not generally provide the full picture of possible operational performance. The dynamics and variation between
equipment and stochastic events can signicantly reduce predicted plant performance. In order to
dynamically simulate the crushing circuit, models for process equipment need to be further developed.
The purpose of this paper is to create a wear function for an existing Particle Size Distribution model
(i.e., a Swebrec-function) with data obtained from a real crusher operating at gradually increasing closed
side settings. This is done to create an accurate and updated model of the crusher in which the transient
consequences of wear are captured. The Swebrec-function and correlation model were implemented into
simulation software with simulated events; this simulation was validated with actual process readings.
Improved simulations were then attained with the developed functions.
2012 Elsevier Ltd. All rights reserved.

1. Introduction
Modelling and simulation are important tools for many process
industries. These tools are used to simulate a process to predict
process behaviour under different conditions and constraints. Simulation results can be utilised for plant design and plant optimisation. For accurate simulations, it is essential that a model
represents the reality of a process as closely as possible.

1.1. Comminution
Comminution is a process of progressively reducing particle size
of rock material (Wills, 2006). The mining and aggregates industry
utilise crushing plants to reduce the size of blasted rock from
quarries into aggregate products or ores. A crushing plant is a conguration of different production units. These production units
consist of crushers, screens, conveyors and bins. The number and
conguration of units is dependent on the preferred performance
for which that the plant and equipment were designed. The nal
product or products are produced by multiple reduction stages,
the conguration of each depending on feed from the quarry and
the purpose of the material. This can range from a single crusher
with a pair of conveyors to several crushers in combination with
a complex system of screens and conveyors.
Corresponding author. Tel.: +46 708 876 867.
E-mail address: [email protected] (G. Asbjrnsson).
0892-6875/$ - see front matter 2012 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mineng.2012.01.016

In the aggregate industry, cone crushers are traditionally used

in the last stages of the crushing plant to produce the nal product.
In cone crushers, the rock material is crushed a number of times by
compressive crushing before it leaves the crushing chamber
(Evertsson, 2000). This compressive crushing is made possible by
an eccentrically rotating mantle within a concave. The smallest
distance between the mantle and the concave during crushing
operation is referred to as the Closed Side Setting (CSS) (Fig. 1),
while the largest distance is the Open Side Setting (OSS). The difference between the CSS and the OSS varies between different mantle
types and is often called the Stroke; it represents the eccentric
throw of the mantle.
A cone crusher can be operated both at different CSS and at different eccentric speeds. CSS can be altered using a Crusher Control
Unit (CCU) by changing, depending on the design of the crusher,
either the vertical position of the mantle or the concave. Most
crushers operate at a constant eccentric speed, but controlling
the eccentric speed continuously during operation can have several
benets (Hulthn and Evertsson, 2010b).

1.2. Plant performance

Crushing plant performance is usually dened by the terms
Plant Capacity and Particle Size Distribution (PSD), but additional
terms such as cost can be used (Svedensten and Evertsson, 2005).
In this paper, plant performance will refer to either capacity or
PSD. Plant design and selected production units will have a

14

G. Asbjrnsson et al. / Minerals Engineering 33 (2012) 1319

Feed

Concave
Mantle

CSS

Product

Eccentric
rotation
Fig. 1. Fundamental picture of a cone crusher.

dominating inuence on plant performance, although plant performance is still far from predictable during an operation.
Overall plant performance is affected by several factors that can
change during operation; these factors include machine settings
and material properties. Changes in machine parameters can cause
a change in plant performance (i.e., decreasing CSS will cause an
immediate change in plant performance). Variations in feed material are caused by geological variation and depend primarily on the
source of the material but how it affects the production is often unclear and hard to monitor continuously.
Plants are affected by gradually deteriorating performance over
time in a form of wear. Wear results from abrasion between rock
particles and the crushing chamber. The wear mechanics in compressive crushing is a complex phenomenon where the wear rate
is affected by a number of factors, such as the properties of the
incoming feed, motion of the particles, the PSD and more (Lindqvist
and Evertsson, 2005). Due to wear the crusher properties will
change, causing the CSS to gradually increase, i.e., the void ratio
between the mantle and chamber will change. This will affect the
shape of the PSD curve. During this study, the increase in CSS during
a period of 4 h was observed in relation to change in the plant performance. Due to the relative shortness of the test compared to the
overall lifetime of the mantle, changes in the mantle prole in this
paper have been considered to be too small to contribute to any signicant change in PSD.

1.3. Crushing plant simulation

Crushing plant simulations are used to predict plant
performance by using ow sheets, as illustrated in Fig. 2. Most sim-

ulations perform steady-state simulations to evaluate plant design

according to chosen setup (Svedensten and Evertsson, 2005;
Whiten, 1974). Examples of available software packages that perform steady-state simulations are PlantDesigner (Sandvik), Aggow
(Bedrock Software), JKSimMet (JKMRC) and USIM PAC (Caspeo). In
steady-state simulations, the plant is simulated until equilibrium
is achieved.
In recent years, an interest in more dynamic simulations has
been growing (Liu and Spencer, 2004; Napier-Munn and Lynch,
1992; Smith, 2005). Examples of available software that can perform dynamic simulations include SysCAD (Kenwalt), Aspen
Dynamics (Aspentech) Simulink (Matlab) and Dymola (Dassault
Systmes). Even though plants experience a steady-state condition
under certain circumstances, it is inaccurate to assume that the
system is steady under all circumstances. Crushing is a continuous
_
process; as a continuous system xt,
equipment is subjected to
variations (u0) and changes over time (t).

_
xt
f xt; u0 ; t
Given the initial condition,

xt 0 x0
In a previous project at a crushing plant near Gothenburg, the
change in plant performance was identied to be caused by wear
in the crusher (Hulthn and Evertsson, 2010a). The studied circuit
was quite stable and insensitive to stochastic changes in the upstream or downstream processes. The loss in production due to
wear is illustrated in Fig. 3. The triangles represent the gradual decrease in the amount of nished product exiting the circuit, while
the marked squares represent a loss of production due to adjustments of the mantle. Crusher throughput is not noticeably affected
by short-time wear.
Both steady-state and dynamic simulations struggle to replicate
this behaviour. Most dynamic simulations are based on steadystate models that do not consider wear to be signicant factors
in plant performance. Dynamic simulations are able to simulate
events that can be used to represent disturbances caused by the
calibration sequence. If the crusher is adjusted frequently, the time
needed for calibrations can be up to 5% of the crushers available
time.
The purpose of the present study was to develop a wear function based on the parameters in the Swebrec-function (Ouchterlony, 2005) that can mimic the short-term wear for dynamic
simulations. This would also create a correlation between machine
parameters and the parameters included in the Swebrec-function.
All simulations in this paper were performed with the simulation
software SysCAD, which is capable of performing both steady-state
and dynamic simulations.
2. Material and methods
To determine the correlation between time dependence and
wear, a set of measurements were conducted at an aggregate plant
in Uddevalla, 80 km north of Gothenburg. The crushing plant is
owned and operated by NCC Roads. In its tertiary crushing stage,
the plant produces high-quality aggregate products, ranging in size
from 02 mm to 1632 mm (see Fig. 4). The crusher is a Metso
Nordberg HP4 cone crusher equipped with a medium chamber;
the feed size to the tertiary crushing stage is 880 mm.
2.1. The test site

Fig. 2. Fundamental difference between steady-state and dynamic simulations

the dynamic simulation takes into consideration residence and lag time, while the
steady-state simulation does not.

Due to the design of the crusher, the crushing chamber needed

to run empty before the CSS could be adjusted to compensate for
wear. The CSS was altered manually with an Original Equipment

G. Asbjrnsson et al. / Minerals Engineering 33 (2012) 1319

15

Fig. 3. Loss of production during operation marked with striped triangles; loss of production due to adjustment of the mantle marked with striped squares.

Fig. 4. Flow sheet over the tertiary phase of the test site. Belt scales are marked with a circle.

Manufacturer system by stopping the feed, unclamping the tread,

rotating the upper shell and clamping the tread again. To compensate for wear and guarantee product quality, this manoeuvre was
performed manually in 2- to 3-h intervals and took approximately
46 min. Although a frequency converter was installed on the
crusher that allowed changes to be made to the eccentric speed
in real time, the speed was kept constant during tests. All 10 conveyors in the tertiary phase were equipped with power meters that
monitored and logged electrical power draw. From these data, the
actual mass-ow could be calculated (Hulthn and Evertsson,
2006).

100

Percentage passing (%)

90
80
70

Time
60

b
50
40
30

2.2. Experiment procedures

20
10

X50

Xmax

Sieving aperture (mm)

Fig. 5. Representation of parameters in the Swebrec function and how they are
affected by time.

In order to measure the gradual changes in the PSD in comparison to changes in the crusher, the CSS was measured. Lead cubes
were lowered down into idling crushing chambers to be deformed
to the actual CSS. This procedure was repeated a number of times
with certain time intervals. During the tests and crushing operation, plant performance, production output and power drawn by
the crusher were monitored and logged. From the plant data, the
change in the plant output could be mapped.

16

G. Asbjrnsson et al. / Minerals Engineering 33 (2012) 1319

crusher model. In previous work, the Swebrec-function has been

studied to approximate the PSD from the crusher with dynamic
simulations (Asbjornsson, 2011). The Swebrec function, depicted
below, was developed to describe the PSD of blasted rock (Ouchterlony, 2005) but can be applied to crushed rock (Svedensten and
Evertsson, 2005) with certain accuracy. The Swebrec function is depicted as:

  1b
In X max
@
 X A
Px 1 
In XXmax
50

Fig. 6. Change in the calculated PSD from belt scales over time.

3. Mathematical modelling
For a better understanding of the process dynamics, the product
should be considered to be a time-dependent vector (P(t)). The values will therefore change over time due to various factors.

P1 t

6 P t 7
7
6 2
7
6
@P
7
6
..
Pt 6
7
.
7
6
@t
7
6
4 Pn1 t 5
Pn t
To improve the dynamic simulation, the plants gradually
decreasing performance needs to be modelled with the existing

where X50 represents the aperture point for 50% of the cumulative
weight, Xmax represents the largest particle in the PSD, b is the slope
of the curve and X is the dened PSD interval (see Fig. 5). During this
study, the simplest form of the Swebrec function was used. One of
the drawbacks with this form is the inability to predict nes accurately. To represent the ne tail in this process, the smallest fraction
has been set to 2 mm, which is the nest product screened out by
the plant.
4. Results
From the process reading, the PSD could be estimated while
parameters could be derived from the data collected. Each belt
scale can be considered to hold a certain size interval. From this
assumption, an approximation of the PSD curve can be tted with
the Swebrec function. The PSD curve was changed over time due to
abrasion between the crushing chamber and the rock material. As
more material goes through the crusher, wear gradually moves the
PSD curve in the horizontal direction, as can be observed in Fig. 6.
To compensate for this development, the crusher is calibrated,
which restores it to a narrower CSS.

Fig. 7. Calculating X50 (left gures) and b (right gures) from the process readings at two of the test periods.

G. Asbjrnsson et al. / Minerals Engineering 33 (2012) 1319

17

Fig. 8. The trend of CSS (dotted grey line interpolated between tests) and X50, (black dots process readings) as a function of material ow through the crusher at two of the
test periods (approximately 4 h). The results are close to parallel. Spikes in the X50 curve indicate an interruption in the process due to calibrations or mechanical failure.

The Swebrec-function was tted to the PSD of the belt scales by

interpolating X50 from the process readings. Xmax was estimated by
adding the calculated change in X50 to the dened OCC. Finally, b
was iterated out by comparing the Root Mean Square (RMS) error
of the results from every b and the actual PSD for that moment.
RMS error was calculated as:

s
P
PSDcalculated;i  PSDBeltscales;i 2
RMS
n
By calculating the change in the three parameters, X50, Xmax and
b, the wear function could be created. The data from the calculation
are presented in Fig. 7. Fig. 7 shows how X50 gradually increases
during the test period, while b is inconsistent with a small localised
decrease which varies within a certain interval. Spikes in the diagrams are due to disturbances in the process from measuring the
CSS.
Fig. 8 presents the results from two of the calibrations combined with the calculation of X50 during these test periods. The correlation between the measured CSS and the calculation of X50 was
between 0.6 and 0.8, with the average of all of the test periods
being 0.71. Because the measured CSS and calculated change is
close to parallel, it can be presumed that the wear was concentrated near the point of the CSS. Between some calibrations, the
CSS seems to decrease. This could be caused by many factors, such
as uneven wear on the mantle or inaccurate measurements. To
minimise the risk of uneven measurements, the same procedure
was performed at every test.
By analysing the change in the predened parameters, the wear
function was generated. Fig. 9 shows the collected data, which are
presented as a change at dened intervals, from the calibration.
This gives a simplied indication of the wear trend that occurred
during the tests.
Equations were formulated for the Swebrec parameters to
approximate the product change over time. This was done to create
a wear function based on the amount of crushed material between
mantle adjustments.

X 50 a1  CSS0 a2

m dt

The parameter a1 represents the ratio between the initial CSS and
X50 after a mantle adjustment while a2 represents the wear rate
depending on the amount of crushed material per hour. Xmax was
considered to be directly proportional to the OSS, as the size fraction will be approximately as large as the largest settings. The
OSS increases the same amount as the CSS during drift, with the
average rate shown in Fig. 9. The change in CSS will follow the same

Fig. 9. Wear trend (black line) generated from the calibration results.

rate as the change in X50 during operation. Parameter b was later

considered to be constant during the drift as the b parameter in
the data was inconsistent and only varied within a small interval.
5. Application of wear function
To validate the developed wear function, the plant was modelled with the SysCAD simulation software package. Both steadystate and dynamic simulations were performed and compared to
real-time data. The Swebrec function was modied to include the
wear function for the dened parameters, and a simplied version
of the plant controllers was created. All simulations were initiated
with the same PSD and CSS. The steady-state simulation was run
until equilibrium was achieved; dynamic simulations were run
for the same duration as the real crushing operation. All disturbances during testing were logged and entered as events in the
simulation. To not affect the full-scale test, the starting CSS was
estimated for the simulation by calculating the X50 parameter. In
Fig. 10, the ow sheet of the test plant can be viewed.
The results from the simulation and the test can be viewed in
Fig. 11. No variation was added to the simulation. Although there
are many similarities between the simulated process and the actual process, there are some distinct differences. By analysing the
actual process readings, it became evident that one or more belt
scales needed to be calibrated. This is because, according to the
process readings, more material enters the system than is being
produced. As can be observed in the simulated results, the lines
From secondary and Product should be in line with each other,
not parallel. Also, the amount of Product and Circulating Load

18

G. Asbjrnsson et al. / Minerals Engineering 33 (2012) 1319

Fig. 10. Flow sheet of the modelled plant in SysCAD.

Fig. 11. Representation of steady-state simulation results (upper-left), dynamically simulated ow with calibration events (upper-right), dynamic simulation with events and
wear function implemented (lower-left) and actual reference ow after modication to compensate for un-calibrated belt scales (lower-right).

should equal the Material through crusher, which is, in other words
is the crusher capacity. By studying the process readings while the

plant was idling (i.e., at startups or during calibration), the error

due to un-calibrated belt scales could be minimised.

G. Asbjrnsson et al. / Minerals Engineering 33 (2012) 1319

Table 1
Comparing the error of the product produced from dynamic and steady-state
simulations to the actual ow.
Process

Average performance
(TPH)

Error
(%)

Actual
Dynamic simulation with wear
function
Dynamic simulations without wear
function
Steady-state simulation

161.0
172.6

7.2

185.2

15.0

189.2

17.5

19

for all crushers or processes but rather to identify a way to better

represent the production process with dynamic simulation.
When comparing the dynamic simulation with the actual process, it was possible to identify belt scales that needed to be calibrated. By running dynamic simulations in parallel with the
production it was possible to identify malfunctions before they
could cause severe problems, such as failures on the conveyors.
The next step is to study data from various periods in a mantles
lifecycle to identify how long-term wear affects the PSD and
crusher capacity for the purpose of dynamic plant simulation.
Acknowledgements

By comparing the dynamic simulation, the steady-state simulation and the process data, the error could be calculated. By analysing the amount of product produced during the production and the
simulated time, the simulation error could be calculated, as illustrated in Table 1.
As can be seen the dynamic simulation Fig. 11- Finished Product
and Table 1), the wear function shows signicant improvement in
representing the actual process when compared to the steady-state
simulation, showing only a 7.2% deviation from the actual process.
The error for the dynamic simulation without the wear function is
not much greater than the steady-state simulation with interruptions. The dynamic simulation with implemented wear function
resulted in a 10.3% improvement over the steady-state simulation.
Due to the manual nature of the procedure, the CSS will never return to the initial CSS value. Thus, additional improvement could
be achieved by running the simulation at the same CSS after each
mantle adjustment.
6. Conclusions and future work
In this paper, a wear function for dynamic plant simulations
was presented that is capable of improving dynamic simulations.
Full-scale tests were performed both to collect data and to validate
the wear function. During validations, it has been shown that
dynamic simulations can be improved to better represent the actual process. These types of simulations, which work to improve
dynamic simulation in plant design, can serve to increase plant
insensitivity to variations, such as wear and malfunctions.
Although these tests gave a clear indication that there is a correlation between a change in the PSD and changes in the CSS, it is
not possible to say a denite correlation exists between the PSD
and the CSS from these data alone. As stated in the introduction,
several other factors, such as degree of reduction and material
properties, can contribute to wear. These factors were close to
constant during the testing as the material came from the same
source, and it was crushed and screened in the secondary stage.
The purpose of this paper is not to create a general wear function

This work has been performed within the Sustainable Production

Initiative and the Production Area of Advance at Chalmers; this support is gratefully acknowledged.
The authors wish to thank the Hesselman Foundation for Scientic Research and the Swedish national research program MinBaS (Minerals, Ballast and dimensional Stone) for its nancial
support.
NCC Roads and their personnel in Glimmingen, Uddevalla, are
gratefully acknowledged for all of their support and efforts to make
this work possible.
References
Asbjornsson, G., 2011. Modellering och simulering av dynamiskt beteende i
krossanlggningar. MinBaS.
Evertsson, C.M., 2000. Cone Crusher Performance, PhD Thesis. Chalmers University
of Technology, Gothenburg, Sweden.
Hulthn, E., Evertsson, C.M., 2006. A cost effective conveyor belt scale. In: 11th
European Symposium on Comminution. The Hungarian Chemical Society
(MKE), Budapest, Hungary.
Hulthn, E., Evertsson, C.M., 2010a. Real-Time Optimization of Cone Crushers, PhD
Thesis. Chalmers University of Technology, Gothenburg, Sweden.
Hulthn, E., Evertsson, C.M., 2010b. Two variable real-time algorithm for cone
crusher control. In: Proceedings of XXV International Mineral Processing
Congress. Brisbane, Australia.
Lindqvist, M., Evertsson, C.M., 2005. Wear in Cone Crusher Chambers, PhD Thesis.
Chalmers University of Technology, Gothenburg, Sweden.
Liu, Y., Spencer, S., 2004. Dynamic simulation of grinding circuits. Minerals
Engineering 17 (11-12), 11891198.
Napier-Munn, T.J., Lynch, A.J., 1992. The modelling and computer simulation of
mineral treatment processes current status and future trends. Minerals
Engineering 5 (2), 143167.
Ouchterlony, F., 2005. The Swedbrec function: linking fragmentation by blasting
and chrushing. Mining Technology 114, A29A44.
Smith, H., 2005. Process simulation and modelling. In: Mike, D.A., Wills, B.A. (Eds.),
Developments in Mineral Processing. Elsevier, London, pp. 109121.
Svedensten, P., Evertsson, C.M., 2005. Crushing plant optimisation by means of a
genetic evolutionary algorithm. Minerals Engineering 18, 473479.
Whiten, W.J., 1974. A matrix theory of comminution machines. Chemical
Engineering Science 29 (2), 589599.
Wills, B.A., 2006. Mineral Processing Technology: An Introduction to the Practical
Aspects of Ore Treatment and Mineral Recovery. Elsevier, London.