Minerals Engineering 43–44 (2013) 29–35

Contents lists available at SciVerse ScienceDirect

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

Dynamic modeling and simulation of cone crushing circuits

Pekka Itävuo a,⇑, Matti Vilkko a, Antti Jaatinen b, Keijo Viilo c
a
Department of Automation Science and Engineering, Tampere University of Technology, P.O. Box 692, FIN-33101 Tampere, Finland
b
Minerals Processing Systems, Metso Automation, P.O. Box 237, FIN-33101 Tampere, Finland
c
Crushing and Screening Equipment, Metso Minerals, P.O. Box 306, FIN-33101 Tampere, Finland

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

Article history: As a common practice, steady-state models are used for simulation and process dimensioning of crushing
Available online 7 September 2012 circuits. However, intended circuit performance is rarely achieved due to constantly fluctuating feed-
material size and characteristics. This gap between theoretical and realized performance has the potential
Keywords: for process control.
Crushing Little scientific attention has been paid to the analytic control system design of crushing circuits. The
Screening current lack of suitable dynamic process models for the task is direct evidence of this. Therefore, it is not
Modeling
surprising that currently existing control applications are biased towards heuristic, model-free, non-ana-
Process control
lytic approaches.
This paper presents an effective way to produce dynamic process models from established steady-state
models. The resulting simulator makes it possible to develop control methods that fully utilize the capac-
ity potential of crushers and facilitates efforts for energy-efficient operation of crushing circuits.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction thorough description of dynamic process models for control sys-

tem design. Section 4 uses the presented models to give a dynamic
Crushing plays an important role in the aggregates and mining simulation example.
industries by reducing the particle size of granular solids, such as
rocks and ores. Before the desired product size is reached, the feed
2. State-of-the-art review
material undergoes 2–4 crushing stages that form a circuit.
Each crushing circuit consists of a unique combination of unit
This section presents a short review about dynamic modeling of
operations for crushing, screening, conveying, feeding, and storing.
crushing circuit unit operations. Borison and Syding (1976) empir-
Contrary to the steady-state modeling that primarily focuses on
ically concluded that material flow dynamics in feeders and crush-
crushing and screening, every unit operation in the circuit contrib-
ers can be approximated using first-order lag transfer functions.
utes to the dynamic presentation.
Whiten (1984) later discussed dynamic expansion of static models
Dynamic modeling of crushing circuits for analytic control sys-
by adding small time delays to crusher and screen outputs.
tem design has not received much attention in scientific literature.
Moreover, he proposed that conveyors can be modeled as simple
Traditionally, interest has been limited to the narrow range of min-
time delays, and ore bins, surprisingly, as a LIFO-queue with
ing applications to control crusher load by manipulating the feed
variable time delay.
rate. Controlling crusher product size and shape by using manipu-
Herbst and Oblad (1985) presented the first dynamic simulation
lated variables (that is, the closed side setting (CSS) or eccentric
model specifically made for control system design that incorpo-
speed (ES)), has not been addressed from the dynamic modeling
rated disturbances such as feed-material size, rate, and properties.
perspective. However, crushing control has major possibilities for
Herbst and Oblad linked feed-hopper material level with crusher
efficiency and profitability.
flow rate, power, and product size distribution, using a static rela-
The purpose of this paper is to: provide models for analytic con-
tionship. Moreover, they approximated that a step change in feed
trol system design of cone crushing circuits, and serve as a starting
rate to the hopper causes first-order lag to the hopper level.
point for future dynamic modeling endeavors.
Sbarbaro et al. (2005) empirically modeled feeder-conveyor
Section 2 gives a state-of-the-art review about dynamic
combinations with first-order with time delay (FOTD) transfer
modeling of crushing circuit unit operations. Section 3 presents a
functions. The model structure is in line with Borison and Syding
(1976). In another paper, Sbarbaro (2005) discussed feed-hopper
⇑ Corresponding author. Tel.: +358 50 3444401. material balance and the relationship between material mass and
E-mail address: pekka.itavuo@tut.fi (P. Itävuo). volume. Moreover, he approximated the dynamics between

0892-6875/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.mineng.2012.07.019
30 P. Itävuo et al. / Minerals Engineering 43–44 (2013) 29–35

feed-material mass flow and crusher power with first-order lag Time delay:
transfer function. He also proposes a modeling procedure for vari-
able-speed conveyors.
G1 ðsÞ ¼ ess ; ð1Þ
Most recently, Johansson (2009) used distributed parameter Integrator:
models to give a very precise dynamic presentation of the cone
crusher. However, this model structure would suit better for finite K
G2 ðsÞ ¼ ; ð2Þ
element method (FEM) simulations than for control system design. s
Moreover, there are some simulation software packages (e.g., First-order lag:
SysCAD, IDEAS and ProSim) that have dynamic properties, but
the structure of the dynamics is not known to the authors. K
G3 ðsÞ ¼ ; ð3Þ
The common missing element in all the previous work is that Ts þ 1
none of the presented models includes dynamics of crusher actua- First-order with time delay (FOTD):
tors (CSS and ES). This is a serious error from the control system
design point of view. Moreover, several important disturbance K
G4 ðsÞ ¼ ess ; ð4Þ
types (such as moisture) are not included in the models, even Ts þ 1
though disturbance rejection is one of most important tasks for Integrating first-order with time delay (FOTDI):
the control systems.
K
G5 ðsÞ ¼ ess ; ð5Þ
sðTs þ 1Þ
3. Modeling
where s is the Laplace operator, K is the steady-state gain, s is the
This section presents the general principles for dynamic model- apparent time delay, and T is the apparent time constant. The aver-
ing of cone crushing circuit unit operations. A detailed description age residence time is given by Tar = s + T (Åström and Hägglund,
of dynamic models for the cone crusher, screen, conveyor, and fee- 2006). The parameter K in this work is mostly equal to unity (where
der is given. not mentioned) because K is actually a linearized counterpart of the
static nonlinearity, and thus already modeled in the corresponding
3.1. Modeling approach steady-state model.

Modeling in this work is mostly carried out using Hammerstein 3.2. Cone crusher
and Wiener-type systems that combine static nonlinearities with
linear dynamics. The advantage of such a hybrid structure is that The cone crusher has two manipulated variables (MVs) that can
modeling can be separated into individual tasks. This makes it par- be arbitrarily changed during operation: closed side setting (CSS)
ticularly suitable for situations where the static process behavior is and eccentric speed (ES). The CSS is defined as the shortest distance
known in advance. In other words, it allows for using static perfor- between crushing liners. Other parameters affecting crusher oper-
mance models made for other purposes (Janczak, 2005). The struc- ation are the stroke length and the crushing cavity geometry. Both
ture of a Hammerstein and a Wiener system is given in Fig. 1. of these can be changed, but not while operating, and are thus con-
stants in this work. Moreover, the cone crusher is subject to a num-
3.1.1. Framework for control system design ber of DVs (Evertsson, 2000; Ruuskanen, 2006; Bearman and
The modeling for control system design is particularly inter- Briggs, 1998). These variables are listed in Fig. 3. In addition, the
ested in two specific relationships: the input/output relationship process equipment experiences time-variant behavior due to wear
and the disturbance/output relationship. These two relationships of crushing liners (Bearman and Briggs, 1998), which is not consid-
form a framework and modeling domain for the control system de- ered in this work.
sign. This approach is illustrated in Fig. 2. From the process control A generic simulation model for a control system design of a
point of view, the variable u refers to control inputs, the variable y cone crusher is presented in Fig. 3. The model consists of eight indi-
refers to process outputs and the variable w refers to disturbance vidual submodels for actuators, material and feed-hopper flows,
inputs (that is, manipulated variables (MVs), controlled variables crusher dynamics, and crusher performance models divided to
(CVs), and disturbance variables (DVs), respectively). In general, product size, capacity, and power models. Flow and performance
the control system will use MVs in order to achieve desired CV val- models are static and nonlinear, directly from Ruuskanen (2006).
ues in the presence of measured and unmeasured DVs. In this The other submodels are dynamic and mostly developed within
sense, the material fed to the crushing process is, in fact, a distur- this work. A detailed description of the submodels is presented
bance. Consequently, the process will never be in a steady state next.
without a control system.
3.2.1. Actuator models
3.1.2. Dynamic model types Actuator models are special because the model is actually a con-
In this work, the linear dynamics are primarily modeled using trol loop similar to Fig. 2. The control input to the system is there-
first-order transfer functions given by the following equations: fore a setpoint for the corresponding control loop.

(a) (b)

Fig. 1. A Hammerstein system (a) and a Wiener system (b).

 P. Itävuo et al. / Minerals Engineering 43–44 (2013) 29–35 31

Modeling domain:

Disturbances (w)

Setpoint (yr) Error (e) Inputs (u) Outputs (y)

Controller Process

Fig. 2. Generic model structure for control system design.

Fig. 3. Dynamic model of a cone crusher.

Use of a frequency converter is the obvious way to control the ratio between the hydraulic valve input and the rate of CSS
crusher ES. In this case, the control loop dynamics are usually so change in both cases. The final closed-loop transfer function can
fast that there is no actual need to take them into account in the be then calculated using the following equation:
input/output presentation. The important thing about frequency
GC
converters is that they come with parameterizable ramps for accel- Gcl ðsÞ ¼ ; ð6Þ
1 þ GC
eration and deceleration. Consequently, the ES follows those ramps
very closely. Thus, the input/output relationship becomes an inte- where G and C are the process and controller transfer functions,
grator, as in Eq. (2), with K equal to the acceleration/deceleration respectively.
rate of frequency converter. The actual input to the system is either
1 or 1 in a transient situation, or zero in a steady state. Note that 3.2.2. Flow model
this model is nonlinear. In this work, a flow model is used to provide crusher- and cav-
The cone crusher CSS is controlled by means of a hydraulic pis- ity-specific parameters for performance model inputs. Flow-model
ton. The actual process model depends on the hydraulic system parameters, as described in Ruuskanen (2006), can be calculated as
configuration. Known configurations are at least a two-way, load- a function of CSS and ES for the whole operating range (Evertsson,
compensated proportional valve, a two-way pump, and a one- 2000), and then fitted with ANFIS neuro-fuzzy networks (Jang,
way pump with an on/off valve. The dynamic model for the 1993). This approach is computationally efficient, as all calcula-
hydraulic piston with proportional valve is given by Eq. (5) (Dorf tion-intensive stages can be performed beforehand. Fig. 4 presents
and Bishop, 2001), where the apparent time constant T and time examples of the fitted flow-model parameters for the Metso
delay s usually range between 50 and 200 ms. Due to fast dynam- GP300M, a tertiary cone crusher with 32-mm stroke. Examples
ics, by assuming frictionless conditions, a transfer function can be are (starting from left) effective stroke length at the outlet (eSTR1),
approximated with an integrator, as in Eq. (2). The parameter K is number of crushing zones (CZs), closed-side volume of the last
32 P. Itävuo et al. / Minerals Engineering 43–44 (2013) 29–35

8
12

11 7.5

10
eSTR1

CZ
9
6.5
8
6
7

400
25
250 25
350
20 300
300 20
350
CSS 15 ES ES
400 250 15 CSS

4
x 10

1.4

1.2
Vc1

0.8

0.6

25
250
20 300
350
15
CSS 400 ES

Fig. 4. Examples of ANFIS fitted flow model parameters for Metso GP300M tertiary cone crusher.

crushing zone (Vc1). Other flow-model parameters are volume ra-

Table 1
tio of first and last crushing zone, average effective stroke length in Equation numbers (#) and corresponding parameter values for dynamic manipulation
the cavity, stroke length at the outlet and cavity angle. of performance models’ input parameters.

Input Capacity Power Product size

DV Eq. (1); s = Td Eq. (4); T,s = 0.5Td Eq. (1); s = Td
3.2.3. Feed hopper model MV Eq. (3); T = 0.5Td Eq. (3); T = 0.5Td Eq. (3); T = 0.5Td
The mass balance in the feed hopper is given by the following
equation:

3.2.4. Crusher-dynamics model

dm
¼ Q in  Q out ; ð7Þ The crusher-dynamics model manipulates the input parameters
dt of crusher performance models in a way that resembles the dy-
namic effect of material flow to the cone crusher model outputs.
where m is the mass of material accumulated in the feed hopper, This parameter manipulation is performed separately for each in-
and Qin and Qout are the mass flows of material entering and exiting put/output combination by using the transfer functions given in
the feed hopper. The mass flow Qout is controlled by the crusher. The Table 1. The average material residence time inside the choke–
relationship between material mass and volume can be calculated fed crushing chamber Td can be calculated using the following
with m = Vq, where V is the material volume and q is the material equation:
bulk density. The formula for bulk density estimation is described
in Ruuskanen (2006) and in Evertsson (2000). It is also possible to ncz
calculate the material level h in the feed hopper, which is a nonlin- Td ¼  60; ð8Þ
x
ear function of V and depends on crusher and feed-hopper geometry
(Sbarbaro, 2005). The material flow in the feed hopper is assumed where ncz is the total number of crushing zones in the cavity, and x
to be non-mixing and follows a plug flow pattern. The justification is the crusher eccentric speed (rpm). The parameter values in Table
is based on material-flow patterns in circular metal silos with 1 are based on the assumption that the crusher is choke–fed and
conical outlets; according to Rotter, the following criteria must be half of the crushing zones are effective. The number of effective
met for the plug flow pattern: adequately steep slope, large enough crushing zones affects the crusher output transients during the
outlet opening, and low wall friction (Rotter, 2001). The first two MV change. It is also possible to analytically solve the amount of
are obviously met simply because of the physical design. Moreover, effective crushing zones, as in Evertsson (2000) and treat the
the material used in a feed hopper is supposed to promote the parameters accordingly. This research furthermore assumes that
smooth material flow, which supports the reasoning. the lowest crushing zone (choke zone) controls the crusher
 P. Itävuo et al. / Minerals Engineering 43–44 (2013) 29–35 33

capacity. Thus, all material above the choke zone exhibits plug flow be iterated in a way that the amount of impurity in Eq. (10) matches
behavior under choke–fed conditions (Evertsson, 2000). If the choke with the calculated screening efficiency J. The cumulative particle-
zone is higher in the cavity, the plug flow behavior only occurs size distribution of impurity transported to the oversize stream
above the choke zone. Below that, and in a non-choke fed situation, can be finally constructed by substituting the upper limit of the
the smaller particles are able to move also during compressions and integral in Eq. (10) with particle sizes 0. . .a, and by dividing the
have therefore shorter residence time than the larger ones. As a re- resulting vector with I.
sult, only the particles larger than CSS are subject to breakage and The final retained Qr and passed Qp material-mass flows can be
obtaining the accurate residence time distribution becomes diffi- easily calculated from the following equations:
cult. In this work, we use a rough estimate for non-choke fed case
Q r ¼ Q  Q u  J; ð12Þ
and substitute the transfer functions in the first row of Table 1 with
Eq. (3), where T = 0.5Td. In a general case, using Eq. (4), where s is Q p ¼ Q u  J; ð13Þ
the dead-time above choke zone and T is the average residence time
where Q is the feed-material mass flow. The following screen decks
in the remaining crushing zones, should provide satisfactory results.
can be calculated in a similar manner.

3.2.5. Performance models

3.3.2. Dynamic expansion
Performance models in this work are log-transformed linear
The resulting steady-state model is then expanded into a dy-
regression models that calculate the interaction between crusher
namic Hammerstein system with Eq. (1) for the material retained
and feed material. Models are of form y = eln(X)b, where X is a non-
from the screen decks, and with Eq. (3) for the undersize material
linear function of MVs, DVs and flow-model outputs, and b is the
of the lowest deck. This is done by assuming non-mixing plug flow
vector of fitted coefficients. There is a separate performance model
behavior and identical material-transport velocity for each deck.
for product size distribution, crusher capacity, and crusher power
The apparent time delay s is given by the ratio of screen length l
calculation. Ruuskanen (2006) goes into greater detail on
and material-transport velocity v. This can either be determined
descriptions, model coefficients, and validation results for crusher
empirically or analytically calculated (e.g., Soldinger, 2002). For
performance models and their precise usage in conjunction with
the stream of undersize particles, a time constant T relative to
flow-model calculation.
the average residence time 0.5s is used.
3.3. Screen
3.4. Conveyor and feeder
This subsection presents a dynamic screen model. Modeling is
carried out using Hammerstein-type system, i.e., static nonlinear- A belt conveyor with constant speed can be modeled as a pure
ity is followed by linear dynamics. delay with Eq. (1) (Whiten, 1984). The apparent time delay s can be
determined as a ratio between conveyor length l and speed v. A
3.3.1. Steady-state model model for a conveyor with variable speed drive is a bit more com-
The steady-state part is modeled using a modification of a clas- plex. The transfer function has to be partitioned into a series of
sic VSMA screen-sizing formula used in a Metso Bruno flowsheet FIFO buffers with respect to the conveyor length in order to main-
simulator (Viilo, 2011). The accuracy of the Bruno-model has been tain the material balance during transients (Sbarbaro, 2005). In this
observed to be similar or slightly better than, e.g., a classic Karra- case, the dynamics of transients should also be considered.
model. Vibrating feeders can be modeled as a first-order lag-transfer
The factor for screening efficiency CJ is first calculated using the function given by Eq. (3) (Sbarbaro et al., 2005; Borison and Syding,
following equation: 1976). The value of the apparent time constant T is usually empir-
ically determined.
Qu
CJ ¼ Q8 ; ð9Þ
As  Q s  i¼1 C i 4. Simulation
where Qu is the mass flow of undersize particles in the feed, As is the
effective screening area, and Qs is the basic screen capacity. The cor- This section presents a dynamic simulation example of the Met-
relation factors C1. . .4 are the oversize, half-size, deck location, and so LT300GPB tertiary mobile cone crushing plant. The simulated
wet-screening factors. C5. . .8 are the material density, effective plant layout is given in Fig. 5. The plant consists of a Metso Nord-
screen-surface open area, screen-opening shape, and the moisture berg GP300 tertiary cone crusher (medium chamber, 32-mm
factors. Corresponding correlation factor values are available in Vii- stroke length, 330-rpm ES and 16-mm CSS), a Metso B2100T
lo (2011). The resulting factor CJ is then scaled into value J, which two-deck screen (length 6.25 m, width 1.6 m, material speed
represents the efficiency of undersize recovery (that is, the amount
of passed undersize material divided by the amount of undersize
material in the feed). By substituting the parameter J into the
following equation:
Z a  2 !n
ax
I¼ 1 du ðxÞ dx; ð10Þ
0 aþw

where
 lnð2Þ
n¼   2  ; I ¼ 1  J; ð11Þ
ln 1  ax 50
aþw

x is the particle size, a is the screen aperture, w is the screen wire

diameter, and du is the particle-size distribution density function
of undersize particles, the value of an unknown parameter x50 can Fig. 5. Metso LT300GPB mobile cone crushing plant.
34 P. Itävuo et al. / Minerals Engineering 43–44 (2013) 29–35

Material flow rate

600
Conveyor 1
400

t/h
Feed Conveyor 2
200
0
0 100 200 300 400 500 600

Mean particle size

14

12
mm

Screen product 8/16 Crusher product

10

8
0 100 200 300 400 500 600

Screen performance
100
Deck 1 efficiency
90
%

Deck 2 efficiency
80

0 100 200 300 400 500 600

Feed hopper material volume

1
m3

0.5

0
0 100 200 300 400 500 600
time [s]

Fig. 6. Simulation results for Metso LT300GPB.

circuits. It also presented the generic method for dynamic expan-

Table 2
Apparent dead times for LT300GPB unit operations. sion of existing steady-state models.
The previous work has shown that current operating modes are
Crusher Feed hopper Screen Conveyors 1–4
not suitable for particle-size distribution control of crushing cir-
Dead time (s) 1.5 0–30 20.8 11.3, 6, 5.3, 5.3 cuits, and therefore unable to provide optimal operation with re-
spect to time (Itävuo et al., 2011). With help from the dynamic
simulation model, it is possible to develop control methods that
0.3 m/s, screen decks: #20-mm, #9-mm) and Conveyors 1–4. fully utilize the crusher’s capacity potential. Moreover, it is possi-
Conveyors are run at a constant speed of 1.5 m/s, and are 17 m, ble to develop control schemes that provide optimal energy effi-
9 m, 8 m, and 8 m in length, respectively. Feed material is granite: ciency of crushing circuits.
feed size 0/60, LA-value 18.5, density 2.7, moisture 0.5%, and flak-
iness index 22. The material is fed to Conveyor 1 using a feeder Acknowledgements
with a constant rate of 395 t/h and 3 s time constant. Two step
changes are performed during the simulation. Feed moisture This work was funded by the Finnish Funding Agency for
changes to 3% at 300 s and CSS setpoint changes to 19 mm at 450 s. Technology and Innovation (Tekes). Metso Minerals (Tampere)
The simulation results are presented in Fig. 6. It takes approxi- provided the necessary data and facilities for the process
mately six full circulations (250 s) until the process stabilizes dur- experiments.
ing start-up. There will be a notable increase to this number with
increased plant complexity. The moisture disturbance causes
References
rather intense changes at different parts of the process due to de-
creases in crusher throughput and product size. Moreover, the Åström, K., Hägglund, T., 2006. Advanced PID Control. ISA-The Instrumentation,
most significant plant dynamics appear to be the result of the Systems, and Automation Society.
material transport in the conveyors, screen, and feed hopper; Bearman, R.A., Briggs, C.A., 1998. The active use of crushers to control product
requirements. Minerals Engineering 11 (9), 849–859.
material only takes 1.5 s to pass the crusher. The material transport Borison, U., Syding, R., 1976. Self-tuning control of an ore crusher. Automatica 12
times in unit operations are summarized in Table 2. According to (1), 1–7.
the simulation, CSS seems powerful and rapid enough to control Dorf, R., Bishop, R.H., 2001. Modern Control Systems, ninth ed. Prentice Hall.
Evertsson, C.M., 2000. Cone Crusher Performance. Ph.D thesis, Chalmers University
material size. The observation of screen separation efficiency re-
of Technology.
veals the strong dependency between feed rate and efficiency. Herbst, J., Oblad, A., 1985. Modern control theory applied to crushing. Part 1:
Development of a dynamic model for a cone crusher and optimal estimation of
crusher operating variables. In: IFAC Automation for Mineral Resource
Development, Brisbane, Queensland, Australia, 9–11 July, pp. 301–307.
5. Conclusions Itävuo, P., Jaatinen, A., Vilkko, M., 2011. Simulation and advanced control of transient
behaviour in gyratory cone crushers. In: Proceedings of the 8th International
Mineral Processing Seminar (Procemin). Santiago, Chile, pp. 63–72.
This paper has presented dynamic simulation models and Janczak, A., 2005. Identification of Nonlinear Systems Using Neural Networks and
modeling techniques for control system design of cone crushing Polynomial Models. Springer-Verlag, Berlin Heidelberg.
 P. Itävuo et al. / Minerals Engineering 43–44 (2013) 29–35 35

Jang, J.-S., 1993. Anfis: adaptive-network-based fuzzy inference systems. IEEE Sbarbaro, D., Barriga, J., Valenzuela, H., Cortes, G., 2005. A multi-input-single-output
Transactions on Systems, Man, and Cybernetics 23 (3), 665–685. smith predictor for feeders control in sag grinding plants. IEEE Transactions on
Johansson, A., 2009. Modelling and simulation of cone crushers. In: Proceedings of Control Systems Technology 13 (6), 1069–1075.
IFACMMM 2009. Viña del Mar, Chile. Soldinger, M., 2002. Transport velocity of a crushed rock material bed on a screen.
Rotter, J., 2001. Guide for the economic design of circular metal silos. Civil Minerals Engineering 15 (1–2), 7–17.
Engineering Series. Spon Press. Viilo, K. (Ed.), 2011. Crushing and Screening Handbook, fifth ed. Metso Minerals.
Ruuskanen, J., 2006. Influence of Rock Properties on Compressive Crusher Whiten, W.J., 1984. Models and control techniques for crushing plants. In: Control
Performance. Ph.D thesis, Tampere University of Technology. 84, Minl. Metall. Process, AIME Annual Meeting, Los Angeles, USA, pp. 217–225.
Sbarbaro, D., 2005. Control of crushing circuits with variable speed drives. In:
Proceedings of 16th IFAC World Congress. Praque, Czech Republic.