Minerals Engineering 24 (2011) 987–994

Contents lists available at ScienceDirect

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

Real-time algorithm for cone crusher control with two variables

Erik Hulthén ⇑, C. Magnus Evertsson
Department of Product and Production Development, Chalmers University of Technology, SE 41296 Göteborg, Sweden

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

Article history: Cone crushers are used in the mineral, mining, and aggregate industry for fragmentation of rock materi-
Received 11 October 2010 als, minerals and ores. Systems used for controlling the Closed Side Setting (CSS) on cone crushers, and
Accepted 7 April 2011 thereby the size reduction, are widely used to compensate for wear of the manganese crushing liners
Available online 23 May 2011
and to protect the machines from overloads. With a frequency converter also the eccentric speed in a
cone crusher can be adjusted in real-time in addition to the CSS. The eccentric speed affects the dynamic
Keywords: interaction between the rock material and the crusher liners. Especially the number of compressions the
Crushing
material is exposed to is affected and also the local compression of the rock material is affected, thus the
Process control
Process optimization
particle-size distribution of the product. Eccentric speed also affects crusher capacity. Real-time feedback
Eccentric speed data on the sellable product streams can be obtained by applying mass-flow sensors to the process. The
CSS adjustment of these two online parameters in real-time can result in an increased potential for produc-
tion yield; however, a nontrivial optimization problem with a large solution space also arises. As the feed
material also varies, the optimal setting for the parameters varies in time.
Herein, we report the development of a monitoring and control system including a two variable online
algorithm for the selection of the setpoint for eccentric speed with respect to the current CSS. The differ-
ent product yields from the crushing plant were monitored by mass-flow meters and continuously eval-
uated by a fitness function. A model for the outcome of the crushing stage, with the two parameters
eccentric speed and CSS, was fitted mathematically to the measurement data. However, since the process
varies continuously, due to the wear of crushers and screens and feed material variations, the perfor-
mance landscape is also continuously varying. Therefore, an Evolutionary Operation (EVOP) approach
was adopted, wherein the variations are instead used to continuously find an operating point closest
to the optimal.
The developed algorithm was tested and evolved at a crushing plant for aggregates that produces
around 400,000 tonnes aggregates per year. The algorithm was implemented in a computer that commu-
nicated with the frequency converter and retrieved data from ten mass-flow meters in the process. The
operator was able to interact and supervise the system through a Human Machine Interface (HMI). The
result is an algorithm that can determine the position and direction of a dynamic speed control to con-
tinuously improve the process-operation point. The magnitude of the improvement potential compared
to a fixed speed operation is from 5% to 20%.
Ó 2011 Elsevier Ltd. All rights reserved.

1. Introduction of two ways, depending on the crusher-type. In the first type, the
cone, on which the mantle is fixed, is vertically adjustable by a
Cone crushers are common production units used for size hydraulic system. The hydraulic system also serves as a safety sys-
reduction of rock materials into finer fractions. Their main operat- tem by dropping the cone to its lowest position if tramp iron comes
ing principle is the same today as when first developed over a cen- into the chamber. Here, the concave is fixed in position. In the
tury ago. A mantle rotates and moves eccentrically in the crushing- other type of cone crusher, the concave is attached to an upper part
chamber and the rock material is crushed between the mantle and which has a thread on the outside, which allows the distance be-
the concave several times while falling downwards through the tween the mantle and the concave to be vertically adjusted by
crusher. As the mantle and concave surfaces become worn, the dis- turning the upper part. In this case, the mantle is fixed vertically.
tance between them must be adjusted to maintain the reduction Both types of crusher can be equipped with an automatic control
ratio and control the top size of the product. This is done in one system that compensates for wear. Such a system helps the user
to keep the size setting constant.
Another parameter of a cone crusher, the eccentric speed, has a
⇑ Corresponding author. Tel.: +46 31 772 5854. great impact on the product properties. The speed affects the
E-mail address: [email protected] (E. Hulthén).

0892-6875/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mineng.2011.04.007
988 E. Hulthén, C. Magnus Evertsson / Minerals Engineering 24 (2011) 987–994

number of compressions that the material is exposed to and thus the flow to a computer. Unfortunately, the cost of installing these is
particle-size distribution of the product. Similarly, the speed also relatively expensive, leading to a very scarce usage in the aggre-
indirectly affects the shape of the product. However, this issue is be- gates industry. A more cost-effective way of measuring the current
yond the scope of this work. Capacity is also affected by the speed. capacity is to measure the power draw on a conveyor-belt that is
Whereas changing the CSS moves the product cumulative particle- performing a lifting work (Hulthén and Evertsson, 2006). Such a
size-distribution curve horizontally, changing the eccentric speed measurement device can be obtained for a tenth of the cost of a
tends to rotate it. The eccentric speed of a cone crusher can be chan- traditional belt scale. By directly measuring properties of the sell-
ged by shifting pulleys on the belt drive. This is labour- and time- able products after a screen, as shown in Fig. 1, given that the qual-
consuming, and is therefore not done unnecessarily. It is possible ity aspects, e.g. shape, are acceptable, the control system
to adjust the speed continuously and more easily by applying a fre- (computer) is able to measure, and thus to control, the plant output
quency converter. These have decreased in cost in recent years, mak- particle-size distribution.
ing them more available for use in standard crushing applications.
Practical crushing operations encounter a wide range of vari- 2. Earlier work
ables: natural variations of the rock material properties in the feed,
wear of the equipment, weather, unwanted stops, etc. So, when In the 1970s, Karra performed a large number of tests on cone
implementing real-time control of a crushing plant, monitoring of crushers with different parameter settings, including various
the status of the actual process is crucial. It is difficult to measure eccentric speeds (Karra, 1983). However, those tests showed no
product properties in real-time at a plant due to the high capacities, significant effects of changing the eccentric speed on the parti-
the dirty environment and the wide range of product sizes. cle-size distribution, nor on the capacity. These results were later
The purpose of a crushing plant is to produce rock material with contradicted by those reported by the authors in more recent stud-
certain properties rather than maintain a constant setting on the ies (Hulthén and Evertsson, 2008, 2010). The lack of observed effect
crushers. Therefore, from a user’s perspective of an optimal control in the older study may be attributable to any of a number of rea-
system, the parameters to adjust should be the different sellable sons, one being that the effects of other parameters were much lar-
product capacities, i.e. size distribution and shape. Therefore, we ger. The lack of an accurate procedure for long-term evaluation is
herein propose an approach in which the properties of the crushing another explanation.
plant products are the goal as the two variables described above However, there are models where the eccentric speed affects
are varied in concert. The inspiration comes from the factorial de- both the particle-size distribution and the capacity (Evertsson,
sign of statistics. 2000). Nonetheless, studies on finding the best constant speed
The main focus of this work was the use of online product yields for cone crushers have been rare. Moreover, controlling the crusher
as measures of plant performance and the corresponding adjust- continuously during operation is even rarer. To our knowledge, no
ment of the crusher control parameters. A standard conveyor-belt study regarding online eccentric speed optimization has yet been
scale, for instance, can easily transmit information about the mass- published.

Fig. 1. A mass-flow meter, e.g. a conveyor-belt scale, is attached to each stream from the screen. The computer then estimates the particle-size distribution (PSD), seen in the
box to the left. The above diagram is a real PSD and the ‘‘Time’’ arrow shows how the PSD changes to a coarser product as the mantle becomes worn.
 E. Hulthén, C. Magnus Evertsson / Minerals Engineering 24 (2011) 987–994 989

The authors previously showed that a crusher with an auto- It can take several parameters into consideration, but these indus-
matic-setting regulation can be improved by implementing a tries often have many to optimize, and the focus is normally there-
closed-loop feedback from the process; the improvement was fore on which parameters to deal with. It is recommended that up
3.5% (Hulthén and Evertsson, 2009). In a following work, the to three parameters can be simultaneously used for optimization
authors implemented a real-time optimization of a cone crusher (Box and Draper, 1998). The principle of climbing the ‘‘process
without a control system for CSS by instead controlling the eccen- mountain’’ is outlined in Fig. 2.
tric speed dynamically (Hulthén and Evertsson, 2008). This im- Holmes has successfully optimized a cement plant with an
proved the output from the crushing stage by 4.2%. It also automatic EVOP system without any explicit model (Holmes,
increased liner lifetime dramatically with 27%. In a subsequent 2003). The result was that the desired variable increased by 37%
study, the setting and the eccentric speed parameters were com- at the same time as the mixture of raw-materials used was chan-
bined and formed a model which predicts the output from the ged to a more profitable one. While a cement plant is among the
crushing stage (Hulthén and Evertsson, 2010). The model was then allied process industries, unlike crushing plants they have several
used in designing an algorithm which improved the process by parameters to vary in real-time.
6.9%. However, this algorithm assumes that the constants of the
model are kept updated. It cannot be assumed that the constants
3. Model
have the same values on other crushing stages, thus it must be
re-fitted again and again.
The models used in the real-time algorithm for CSS (Hulthén
Evolutionary Operation (EVOP) is a method wherein the process
and Evertsson, 2009) and eccentric speed (Hulthén and Evertsson,
variations are used for process improvements (Box and Draper,
2008) both build upon the behavior of the process with a closed-
1998). If the process parameters do not vary, and they seldom
loop return flow of the oversize particles. In the authors most re-
are before EVOP is introduced, variations are purposefully intro-
cent study, it was clearly shown that a single fixed speed is not
duced. The method is not an automated method, rather on the con-
optimal over a short time frame (Hulthén and Evertsson, 2010).
trary; it is used manually in manufacturing and process industries.
The assumed appearance of the model is:

^ ¼ a þ bx1 þ cx2 þ dx21 þ ex22 þ fx1 x2

y ð1Þ

where y ^ is the crushing stage output (product yield), x1 is the eccen-

tric speed, x2 is the time since the last CSS adjustment and a–f are
constants. In Fig. 3 an example of how the model fitted to real data
predicts the performance of the crushing stage performance. In this
work, the model in Eq. (1) is applied, but the issue here is how to
use it to improve a process automatically. The net throughput of
the circuit, measured in tph (tonnes per hour), is the height of the
process hill.
The main reason why fixed settings are not optimal is that sev-
eral factors in crushing plants vary with time. In addition to the
short-term wear period of the crushing-chamber discussed above,
the factors that are beyond operator control are raw-material var-
iation, screen-cloth wear, and total crushing-chamber wear over its
useful lifetime.
Fig. 2. The principle of Evolutionary Operation (EVOP). P is the original process
point at which the optimization is started. The variables A and B are then varied in a
pattern during the first phase. The levels of A and B for the next phase are decided 4. Algorithm
upon the results from the current phase. The process performance is represented
with elliptic altitude lines and the optimum is close to the common corner of phases The challenge in this work was to fit the model described above
3 and 4. and simultaneously use it for optimal crusher operation. Therefore,

Fig. 3. The prediction of the crushing stage performance as a function of speed at three different times since CSS adjustment. It is obvious that one fixed speed cannot be
optimal during the short-time wear of the liners.
990 E. Hulthén, C. Magnus Evertsson / Minerals Engineering 24 (2011) 987–994

an EVOP approach was implemented on a Metso crusher from the plant performance for each run respectively are updated for every
Nordberg HP range (the latter of the two types described in the cycle. The cycle is repeated until a sufficiently low level of devia-
introduction). The CSS is adjusted regularly, but not dynamically, tion is reached. Now, a phase is completed and the next step is
by a hydraulic motor, either on manual command or automatically. to decide in which direction the cycle-square should be moved.
This is typically done every two or three hours of operation (this An example of the squares is shown in Fig. 2. This is normally done
depends, of course, on the application). When the liners become by estimating the steepest decent by looking at the effects of the
worn the crusher must be empty of material before the CSS can parameters.
be adjusted, because of the design of the crusher. The CSS is ad-
justed by stopping the feed, unclamping the thread, turning the 6. Full-scale tests
top shell, clamping, and restarting the feed. This takes six to ten
minutes before the process is up and running in steady-state mode 6.1. The plant and equipment
again. Therefore, this adjustment can only be done, at most, a cou-
ple of times during a shift. Every operation time period between The developed algorithms were tested at a crushing plant in
adjustments can therefore be defined as a run in an EVOP context. Uddevalla, 80 km north of Gothenburg. The crushing plant is
Taking the model and the crusher-type limitations into consider- owned and operated by NCC Roads. The plant produces high-qual-
ation, three parameters have been resolved here: ity aggregate products ranging in size from 0–2 mm to 16–32 mm
in its tertiary crushing stage. The crusher is a Metso Nordberg HP4
1. Power draw when adjusted, P. If the power draw is high, the CSS equipped with a fine chamber. The feed size to this tertiary crush-
is small. This implies high reduction but decreased capacity. ing stage is 16–70 mm. The crusher is equipped with a system for
2. Eccentric speed when the run is started, n0. manual adjustment of the CSS.
3. Speed change, Dn, i.e. how much the eccentric speed should be The cone crusher was equipped with a Control Techniques Uni-
changed, e.g. every ten minutes. drive SP frequency converter, which allowed step less changes of
These three parameters are illustrated in Fig. 4 and correspond the eccentric speed of the crusher in real-time. The pulleys were
to positioning of the model in x and y direction a rotation in the x–y configured so that the nominal net frequency, 50 Hz, which then
plane. Because P (power draw) is dependent on how much the gap corresponds to 1500 rpm on the motor, would give 952 rpm on
is tightened, this was excluded as a controlling parameter in the the crusher’s drive shaft.
EVOP algorithm. After the crusher, a series of two classifying screens sort the
product into size fractions of 0–2 mm, 2–5 mm, 5–8 mm, 8–
11 mm, 11–16 mm, 16–22 mm and +22 mm. The +22 mm material
5. EVOP calculation is always returned to the crusher in a closed circuit. All products at
sizes of 5 mm and larger can be returned in a closed circuit at 0%,
The two controlling parameters, n0 and Dn, are varied in a pat- 50% or 100% re-crushing rates. During the final test runs, 100% of
tern at a low and high level of the two parameters, respectively, the 8–11 mm product was re-circulated back to the crusher.
forming a square in the 2-parameter plane. Also the center point An outline of the plant is shown in Fig. 5. In total, ten conveyor-
is included. This is called a cycle and thus a cycle consists of five belts had mass-flow meters monitoring the electrical power draw.
runs. The cycle is repeated and an average and a deviation of the The power was measured with power transducers, which can

Fig. 4. Parameters in the model correspond to positioning in two dimensions and one orientation of the model.
 E. Hulthén, C. Magnus Evertsson / Minerals Engineering 24 (2011) 987–994 991

Fig. 5. Tertiary crushing stage at NCC Roads’ plant in Uddevalla. The plant produces six profitable products from the tertiary stage. The oversize particles are re-circulated to
the crusher.

deliver real-time information to a computer, where the actual The adjustments of the eccentric speed were performed auto-
capacities are calculated. matically, both during each run and to a new value before starting
The algorithm was implemented in a computer that the next run. The CSS was measured indirectly from the power
communicated with the frequency converter, retrieved data from draw. While the power draw cannot be controlled accurately, the
ten mass-flow meters in the process, and also interacted with the level of CSS adjustment is roughly reflected by the power draw.
operator. In addition to this, an HMI/Scada system was developed Therefore, the desired power draw was provided to the operator
for communication with users, see Fig. 6. Using the ten mass-flow on the HMI. The performance of the plant is here defined as the
sensors, all material flows can be monitored by the operators in net crushing stage throughput, i.e. the crusher throughput minus
real-time. the circulating load, measured in tph.

Fig. 6. An HMI/Scada system for the monitoring and control system was developed for convenience.
992 E. Hulthén, C. Magnus Evertsson / Minerals Engineering 24 (2011) 987–994

7. Test period regardless of whether or not a run was continued for a much longer
time; runs of less than 5000 s were neglected.
The initial experiments were conducted during 2009. The
implementation and tests of the new algorithm described above 8. Results
were conducted between March 23rd, 2010 and July 12th, 2010,
see Table 1. During this period, all operation of the crushing stage The EVOP was run in five phases. The result of the EVOP runs
was recorded. On March 23rd, the set of liners were about 20% can be seen in Table 2. The center point in each phase is denoted
worn. A change of liners took place on June 14th, between Phases by ‘P’ in Fig. 2. The parameters were varied around this point.
III and IV. A period constant operation at 52 Hz took place between
Phases II and III. Every run between two CSS adjustments was re-
9. Discussion
corded, and a performance value (mean), the standard deviation,
and the length of the run were calculated. The decision when to ad-
The standard deviations during the runs were relatively high, on
just the CSS was mostly due to the fact that the power value went
average 14% of the measured values. Box and Draper, however,
below a stated limit, 200 kW, but there were also other reasons,
state that large experimental error often indicates a lucrative
e.g. shut-down time, tramp, or lack of feed material. This time var-
source of improvement and it is probable that large effects can
ied greatly, from a couple of minutes to three hours or more. The
be detected (Box and Draper, 1998). Note that the improvements
shortest time corresponds to repeated adjustments of the CSS,
according to the EVOP results shown in Table 2 are often of the
while the longest corresponds to times when the operator was
same magnitude.
too busy with other maintenance. An economical optimum is cal-
The correlation coefficient between the average power draw
culated between one and two hours. Therefore, a time of 5000 s
and the performance value is 0.845, which clearly indicates a link
was chosen as the period for which performances were compared,
in the form of the amount of size reduction achieved. However,
the correlation coefficient between the maximum power draw
and the performance value was only 0.588, which is low.
Table 1
The dates of the EVOP phases.
The differences between the worst and best runs internally in
each phase were 30%, 29%, 23% and 36% in Phases I–IV, respec-
Phase Start date End date tively. This shows a large potential for working actively with the
I March 23rd, 2010 April 29th, 2010 speed. The performance values varied even more over the entire
II April 30th, 2010 May 11th, 2010 crushing-chamber lifetime, as can be seen in Fig. 7. The differences
III May 20th, 2010 June 9th, 2010
IV June 18th, 2010 July 12th, 2010
were systematically more than 100 tph. This is also in accordance
with our prior findings and at the same time underlines the

Table 2
The EVOP runs and their results.

Run Parameter 2 Parameter 3 speed Average performance Change compared Average power Maximum
start speed (Hz) change (Hz/h) after 5000 s (tph) to center point draw (kW) power draw
(kW)
Original settings 50 0
Optimal settings according to 42 2.5
(Hulthén and Evertsson, 2010)
Tested setting in (Hulthén and 42 2.25 274
Evertsson, 2010)
Phase I
Center point 0 47 3 183 151 216
Phase I 1 49 4.5 151 17% 148 188
Phase I 2 49 1.5 192 +5% 174 285
Phase I 3 45 1.5 207 +13% 195 309
Phase I 4 45 4.5 188 +2% 164 245
Phase II
Center point 0 46 1.3 199 162 240
Phase II 1 48 3 172 13% 149 216
Phase II 2 48 0 212 +7% 161 235
Phase II 3 44 0 230 +16% 193 258
Phase II 4 44 3 225 +13% 167 242
Constant 52 151
Phase III
Center point 0 45 0.7 165 144 230
Phase III 1 47 1.3 128 23% 121 162
Phase III 2 47 0 151 9% 130 218
Phase III 3 44 0 166 0% 146 226
Phase III 4 44 1.3 147 11% 162 205
Liners change
Phase IV
Center point 0 45 0.8 170 148 220
Phase IV 1 47 1.6 130 23% 118 233
Phase IV 2 47 0 182 7% 158 265
Phase IV 3 44 0 192 13% 169 264
Phase IV 4 44 1.6 163 4% 138 286
 E. Hulthén, C. Magnus Evertsson / Minerals Engineering 24 (2011) 987–994 993

Fig. 7. The performance varies greatly during the service lifetime of HP-crusher liners.

Fig. 8. The different speed strategies.

994 E. Hulthén, C. Magnus Evertsson / Minerals Engineering 24 (2011) 987–994

importance of measurement over a longer period of time (Hulthén interface on the control computer, the operators’ handling of the
and Evertsson, 2008). It is also clear from Fig. 7 that individual runs plant will subsequently improve. There are also great improve-
cannot be compared with runs from other phases. The differences ment possibilities in adjusting the liners at the right time, which
observed during the service lifetime of a set of liners underscores good instrumentation can facilitate. Also, the time at which the lin-
the fact that the operating conditions vary, and therefore different ers are replaced should be further investigated. It remains an open
speed strategies should be investigated. question if it is acceptable that the liners can sustain an almost 50%
A clear path for the EVOP algorithm is towards lower speeds systematic performance decrease, which has been seen in several
(Parameter 2) and smaller speed changes (Parameter 3). In Phase currently operating crushers and confirmed by talking with people
I, and perhaps Phase II, this was expected. However, in Phase III well-grounded in the field.
the EVOP continues to point in a downward direction in both speed The EVOP algorithm clearly pointed out the direction for the dy-
and speed change. The same pattern was seen in Phase IV, although namic optimal speed. Compared with earlier algorithms, this EVOP
the liners were changed here. To understand this, new adaptations algorithm is less noise sensitive. However, it cannot react to short-
of the models were made from data collected during the EVOP term variations, e.g. changes in the raw-material feed. Once de-
tests. The results, as shown in Fig. 8, were that all runs were per- tailed modeling has provided sufficient process insight the EVOP
formed at higher speeds than the model optimum. The EVOP algo- can be replaced by another algorithm for setpoint selection.
rithm has clearly pointed out the direction for the dynamic optimal
speed. In comparison with the earlier algorithm, Finite State
Machine (FSM), the EVOP algorithm is less sensitive to noise. Acknowledgements
However, it cannot react to short-term variations, e.g. changes in
raw-material properties. The authors wish to thank SBUF, MinFo, the Hesselman Founda-
When lowering the eccentric speed of an HP-crusher, the elec- tion for Scientific Research and the MinBaS-program for their
tric-power demand normally increases. This is evident by inspec- financial support. This work has been carried out within the
tion of the maximum power draw in Table 2. HP-crushers are Sustainable Production Initiative and the Production Area of
known for stalling when too heavily loaded. Stalling occurs when Advance at Chalmers. The support is gratefully acknowledged.
the crusher takes in more material than it can handle, and it simply NCC Roads and their personnel in Glimmingen, Uddevalla are
stops and the top shell starts to float with the chamber full of gratefully acknowledged for all their support and efforts to make
material because of overloading of the pretension hydraulics. This this work possible.
is the reason why the continued lowering of the speed was not ex-
plored; because the EVOP is programmed in an automatic script References
which interacts with the crusher automatically, no risk for over-
loading the crusher could be taken. Future work will include man- Box, G., Draper, N.R., 1998. Evolutionary Operation. Classics edition. John Wiley &
ually testing several runs around the optimum settings. Sons, New York, NY.
Evertsson, C.M., 2000. Cone crusher performance. Ph.D. thesis. Chalmers University
of Technology, Goteborg, Sweden, pp. 1–49.
10. Conclusions Holmes, D.S., 2003. Model-free optimization in cement plants. In: IEEE Cement
Industry Technical Conference (Paper), Dallas, TX, pp. 159–173.
Hulthén, E., Evertsson, C.M., 2006. A cost effective conveyor belt scale. In: 11th
An inevitable development in the rock-crushing industry, espe- European Symposium on Comminution. The Hungarian Chemical Society
cially the aggregate industry, is a more process-oriented focus (MKE), Budapest.
where new and more cost-effective measuring methods will be Hulthén, E., Evertsson, C.M., 2008. On-line optimization of crushing stage using
speed regulation on cone crushers. In: XXIV International Mineral Processing
introduced. The monitoring will be done on both production equip- Congress. Beijing, China, pp. 2396–2402.
ment and product yields. In this study, a two variable algorithm Hulthén, E., Evertsson, C.M., 2009. Algorithm for dynamic cone crusher control.
utilizing a previously developed model was tested for continuously Minerals Engineering 22 (3), 296–303.
Hulthén, E., Evertsson, C.M., 2010. Two variable real-time algorithm for cone
improving the process-operation point. The level of improvement crusher control. In: XXV International Mineral Processing Congress. Brisbane,
in each phase was in the region of 15%. Australia, pp. 813–819.
With an algorithm that automatically chooses the setpoint val- Karra, V.K., 1983. Process performance model for cone crushers. In: XIV
International Mineral Processing Congress. CIM, Montreal, Que, Canada,
ues, crusher operation requires less attention from the operators.
Toronto, Canada, pp. III 6.1–III 6.14.
In addition, when given more instruments, such as a graphical user