INTELLIGENT GRINDING CONCEPT
Remes A.1), Karesvuori, J.2), Pekkarinen, H. 3), Jämsä-Jounela, S-L.1)
1) Helsinki University of Technology
Department of Chemical Technology
Laboratory of Process Control and Automation
P.O.Box 6100, FIN-02015 HUT, Finland
2) Outokumpu Technology, P.O.Box 84, FIN-02201 Espoo, Finland
3) Outokumpu Chrome, Kemi Mine P.O.Box 172, FIN-94101 Kemi, Finland
Abstract: To enhance the operation of mineral grinding processes, a greater number of
monitoring services and control schemes are nowadays being offered by the equipment
manufacturers. In this paper an intelligent grinding concept is formulated and the
components of the concept for typical grinding processes are proposed. Furthermore, the
benefits of the process monitoring services are studied on the basis of a specific case
study - the Outokumpu Chrome Kemi concentrator. Finally, the results are discussed and
a new control scheme is outlined. Copyright © 2005 IFAC
Keywords: mineral grinding, particle size analysis, process monitoring, extended product,
process control, chromite concentration, gravity separation.
et al., 1999). In addition, the particle size has
also been monitored using model-based soft-
In order to enhance the performance of mineral
sensors, see (Casali et al., 1998) and (del Villar
processing equipment, a greater number of
et al., 1996). Recently, the main interest in the
intelligent functionalities are being integrated
monitoring of mill feed has centered on the ore
into the equipment. The equipment suppliers can
type determination, see (Jämsä-Jounela et al.,
provide advanced operating, maintenance and
1998), and on vision-based ore size and texture
monitoring methods by adding these type determination (Guyot et al., 2004).
functionalities. The concept integrates the
equipment, instrumentation and service A number of authors have presented control
resources in order to perform the defined
strategies for the grinding circuits. Jämsä-
operations. Typically, the main parts of the
Jounela (1990) applied the inverse Nyquist array
intelligent grinding concept are the process
method in multivariable grinding control. Niemi
monitoring and control modules.
et al. (1997) simulated an industrial process
using model predictive control for particle size
In the past, several monitoring methods have
and slurry density. In addition, a control system
been developed for the grinding circuit,
based on the mill charge and the particle size on-
including monitoring of the feed, product and
line estimation in LKAB’s Kiruna iron ore
the mill operating conditions. Mill charge
concentrator is presented in Herbst et al. (1996).
position monitoring has recently gained interest.
Recently, Yianatos et al. (2002) showed
In this area Valderrama et al. (2000), Campbell
significant improvements in circuit throughput
et al. (2001) and Pax (2001) applied signal
using particle size rule based control. Laboratory
processing methods to interpret the mill surface
mill grinding simulations have also been carried
vibrations. There are three main industrial
out in order to compare the PI and MPC control
measurement techniques for performing the
schemes (Ramasasy et al., 2005). Elsewhere,
particle size analysis of mill product:
Radhakrishnan et al. (1999) applied the ball mill
mechanical distance detection, ultrasonic
and hydrocyclone models in order to develop a
attenuation and laser diffraction (Napier-Munn
model-based optimizing control. Fuzzy logic has
been applied in the control of SAG mill feed
size variation in the Ok Tedi Mines, resulting in
higher throughput (McCaffery et al., 2002).
Hybrid neural network MPC control has been
studied in Mathur et al. (1999), where a NN is
used for determining the grinding process state.
Further, Duarte et al. (2001) tested a combined
NN-MPC control in the Codelco Andina
grinding plant simulation.
Fig. 1. Structure of the intelligent grinding
The use of variable rotation speed control in
mineral grinding circuits is increasing. In
addition, high accuracy on-line particle size
In this project the first stage was to define the
distribution measurements enhance the concept components together with the
development of the optimizing control for
equipment manufacturers and the end-users.
grinding circuits. As an early study, Herbst et al.
Two web-based questionnaires were performed
(1983) showed that the mill speed is a major
and personal interviews were made worldwide.
manipulated variable for controlling the
Based on the results of the questionnaire survey,
circulating load. Recently, discrete element
the main mineral grinding operating goals are
simulations have been used to study different
described, new aspects for resource development
aspects of mill behavior. Cleary (1998)
are given and, finally, the main grinding
concluded that the lifter wear rates behave
equipment automation functions are summarized
nonlinearly when the rotation rate is increased. It
in the following.
has also been proposed that, in order to maintain
a steady throughput and to avoid grinding
2.1 Services for monitoring and optimization of
media-liner impacts, the total charge volume
the grinding process
should be continuously assessed (Brodie, 2003).
The advantages of rotation speed control include
In mineral grinding processes the production
better control of product size and downstream
goals, and thereby the operating strategies, vary
processes, power savings and longer liner life,
in each particular case. However, it is typical of
and as a consequence, lower maintenance costs.
grinding processes that the capacity should be
maximized, while keeping the total costs as low
In this paper an intelligent grinding concept for
as possible. The operating strategy should
typical grinding processes is presented and
therefore ensure maximal equipment
discussed. Furthermore, a case study with a
grinding circuit including variable speed control
mills and a particle size analyzer is presented in
It is recommended that the APQ (Availability,
Performance and Quality) measure index is
utilized to maximize equipment availability and
performance (Hagberg, et al., 1998). In a
2. DESCRIPTION OF THE FUNCTIONS OF
grinding circuit, the availability (A) is calculated
THE INTELLIGENT GRINDING CONCEPT
for each of equipment, taking into account mill
lining wearing, stoppages, and process
As a concept, intelligent grinding includes - in
interruptions. The performance (P) factor is
addition to the usual process instrumentation and
calculated from the basic mill feed and power
automation - functions that enable the
draw measurements. The quality (Q) is a
optimization, monitoring and operator support
measure of how accurately the process is kept in
services. The intelligent grinding concept
the product targets or within the desired
utilizes an extended product scheme, in which
the additional functionalities and services are a
part of the physical product (Thoben et al.,
Hence, in order to monitor the equipment
2001). As categorized in Fig. 1, the operating
availability, Condition Based Maintenance
resources for grinding processes are the process
(CBM) methods are utilized to refine the process
equipment itself, and the related instrumentation
and maintenance data, and to predict the
and available services. Utilizing the integrated
remaining availability (Bengtsson, 2003).
intelligence in the equipment, the equipment
Furthermore, in order to construct an efficient
automation should meet the operating goals.
operator support tool, the equipment life-cycle
Based on the goals, the equipment automation
scale Product Data Management (PDM) or
functions are optimization and control, as well
Enterprise Asset Management (EAM) features
as monitoring and operator support.
will be included in the concept.
In addition, in order to achieve the maximal
with respect to the process monitoring and
grinding circuit performance, the mill control strategy. Development was started with
throughput has to be maximized while, at the
process data analysis in accordance with the
same time, minimizing the total operating costs.
control strategy design, which is described in the
The constraints to be taken into account in the
optimization are typically the degree of mineral
liberation and the prevention of over- and under-
grinding. The capacity, as well as the target
3.1 Description of the Kemi concentrator and
values of the slurry properties, is eventually
the grinding circuit
dictated by the following process stages. The
optimization of throughput and operating costs
The Kemi chromium ore deposit is located in
requires estimation of the power curve of the
northern Finland. The ore reserves are 52 Mt
grinding process. Additionally, on-line particle
and the annual production of the Kemi
size distribution measurement and ore type
concentrator is 1.2 Mt of ore. The products are
information have been found to be beneficial for
upgraded lumpy ore with a grade of 35.0 %
grinding optimization. Finally, the goal of the
Cr2O3 and lumpy size of 12-100 mm, and the
intelligent grinding concept is to advise the
metallurgical grade concentrate with a grade of
operators in optimizing the grinding process as a
45.0 % Cr2O3 and average grain size of 0.2 mm.
part of the whole mineral processing chain.
After crushing, separation of the 12-100 mm ore
is carried out at the dense medium separation
As a result, monitoring and optimization are
plant. The undersize is further processed in the
provided as services within the process
concentration plant, where the ore is ground in
equipment. Additional services to be offered are
the grinding circuit. Concentration is
data-mining, control loops tuning, subsequently carried out using gravity and
circuit/equipment process audit, and magnetic separation.
maintenance, as well as training and operator
support services (Jämsä-Jounela et al., 2005).
The grinding circuit, shown in Fig. 3, consists of
a rod mill and a ball mill, with a maximum
Finally, as a summary of the defined
power consumption of 560 kW and 220 kW,
components of the intelligent grinding concept,
respectively. The classification is carried out
the components are categorized according to the
using Derric screens with a 0.8 mm aperture.
desired goals into capacity maximizing, usability
The mills have variable speed drives, which can
and total cost minimizing. These are presented
be used to control the product particle size
in Fig. 2.
distribution, which is measured from the screen
underflow using the laser diffraction based
PSI500 Particle Size Analyzer (Kongas et al.,
3. CASE STUDY: THE KEMI
2003). The size range of 1…500 µm is measured
CONCENTRATOR GRINDING CIRCUIT
to a precision of 1-2 %.
The aims of the Kemi concentrator case were, in
the first phase, to develop the concept modules
OPTIMIZATION AND CONTROL
MONITORING AND OPERATOR SUPPORT
Avaialbility Performance Quality
Goals and conditions estimation
Data analysis indices
Degree of mineral liberation
Preventing over- and under-
difference from situations,
power and input optimum
Product Data Management
TOTAL COST MINIMIZING
vs. the benefit
Fig. 2. Main functions of the intelligent grinding concept.
related to the changes in product particle size
fractions and, subsequently, which variables are
significantly inter-related together. To describe
the width of the particle size distribution and
thus the amount of fine fractions, the slope of
the steepest part of the cumulative size
distribution was determined. This variable was
also included into the PCA study.
Fig. 3. The Kemi grinding circuit with rod and
The PCA analysis showed that the mill rotation
ball mills and screen classification.
speed and the ball mill power (indicating the
amount of circulating load) have the most
significant inverse effect on the size distribution.
3.2 Input variables and training data for the
The greater these variables are, the lower is the
PCA, PLS and SOM models
cumulative size distribution slope value,
meaning a higher production of fines.
The aim of the process study was first to
determine how the mill operating variables
The data were further examined using the partial
affect the product particle size distribution. This
least square analysis. The aim of the partial least
information was subsequently utilized to
squares projections to the latent structures (PLS)
develop new control strategy for the grinding
is to define a linear multivariate model between
circuit. The PCA and PLS methods were
the operating variables and the process output
selected to be tested first. Finally, the SOM
variables. In this case, the goal was to study how
method was also applied.
the particle size distribution is affected by the
operating variables, and which variables
The analyzed process data included the
contribute the most to the product particle size
following mill operating variables: circuit feed
rate (t/h), mill power draws (kW) and rotation
speeds (rpm), and consequently the grinding
The PLS model was made for the variables -10,
energy per ton. The -10, -32, -74 and -125 µm
-74 and -125 µm. From those results it was
particle size fractions where chosen as output
deduced that the rod mill rotation speed and
variables. A total of six data sequences
power draw have the most significant effect on
containing approximately 12 000 rows of minute
the product size fractions. A higher speed
data were analyzed. The data were median
produces more fine material, while the higher
filtered to remove the outliers. The time delays
primary mill power – indicating a higher mill
were determined and taken into account by
charge – reduces the amount of fine material.
shifting the data appropriately. To describe the
ore hardness, a grindability index was
calculated, proposed by Tano et al. (2005),
3.4 SOM analysis
which takes into account the amount of fine
material produced per grinding energy used as
Finally the self-organizing map (SOM) was
applied to get more insight into the process
behavior. The aim of the self-organizing map
(SOM) is to classify a high dimensional process
1 = ( S
data and to compress the information into a two-
where 32 m
S ? and 32 m
S ? are the portions of dimensional plane. In this case the goal was to
material finer than 32 µm in the circuit discharge
determine which process operating conditions
and the feed, F is the feed (t/h) and P is the total
cause a coarse/fine grinding product.
mill power draw (kW). The amount of fine
Consequently, the method provides information
material is assumed to be negligible and
about the process by classifying the data,
considered as a constant in the feed stream.
especially when the combined PCA-SOM
method is applied.
3.3 PCA and PLS analyses
The SOM analysis showed that the process has
clearly two clusters, separated mainly according
The aim of the principal component analysis
to the milling power. The high milling power is
(PCA) is to reduce the number of variables and
correlated with a high rotation speed and a low
detect structures between the variables, and
grindability, whereas the rod mill feed does not
thereby to classify the variables. In this case the
affect the clustering significantly.
goal was to study which process variables are
Utilization of the intelligent grinding concept
enables the addition of operational intelligence
into the process equipment, thereby increasing
the performance of the process. In this paper
definitions for the components and functions of
the concept were proposed. Furthermore, a case
study on the particle size measurement based
monitoring of the Kemi concentrator grinding
circuit was carried out. In the case study, the
mill rotation speed made a significant
contribution to the grinding product size
distribution. The results encourage further
Fig. 4. Combined PCA-SOM, the U-matrix with
development of the control strategy and
a PCA similarity coloring.
monitoring methods for the Kemi process.
To visualize the process data and to determine
the process conditions causing a fine and a
coarse product, the combined PCA-SOM
analysis was performed. The data were clustered
Bascur, O.A. (1982), Modelling and computer
into two principal components, which where
control of a flotation cell, University of Utah,
used in the similarity coloring of the U-matrix,
Salt Lake City, 372 p.
resulting from the SOM training. The
Bengtsson, M., (2003), Standardization Issues in
corresponding process condition were
Condition Based Maintenance, Proceedings of
interpreted from the SOM component planes and
the 16th International Congress of Condition
added to the figure. The resulting graph is
Monitoring and Diagnostic Engineering
shown in Fig. 4. The figure indicates that the
Management (COMADEM), Ed. Shrivastav,
production of fine material has occurred when
O. and Al-Najjar, B., Växjö University Press,
the ore was easily grindable, but also when the
ore was harder and too a high rotation speed was
Brodie, M.N., (2003), Variable speed SAG
applied. This implies that the current control
milling, CIM Bulletin, 96, pp. 67-71.
strategy has not reacted to the changing
Campbel, J.,Spencer, S., Sutherland, D.,
conditions quickly enough.
Rowlands T., Weller, K., Cleary, P., Hinde,
A., (2001), SAG mill monitoring using
surface vibrations, Proceedings of the
4. OUTLINE OF THE GRINDING PROCESS
international conference on autogenous and
semiautogenous grinding technology, ed.
Barrat, D.J., Allan, M.J., Mular, A.L.,
Currently the grinding circuit control at the
University of British Columbia, 2001,
Kemi concentrator is based on visual monitoring
Vancouver, Canada, pp. 373-385.
of the amount of screen overflow and
Casali, A., Gonzalez, G., Torres, F., Vallebuona,
monitoring of the specific size fraction trends.
G., Castelli, L., Gimenez, P., (1998), Particle
However, this procedure may lead to a
size distribution soft-sensor for a grinding
suboptimal operating point, causing excess fines
circuit, Powder Technology, 99, pp. 15-21.
production and thus economic losses.
Cleary, P.W., (1998), Predicting charge motion,
power draw, segregation and wear in ball
The aim of the control strategy proposed is to
mills using discrete element methods,
produce the desired product size distribution,
Minerals Engineering, 11, pp.1061-1080.
while minimizing the energy consumption and
Del Villar, R. G., Thibaultb, J., Del Villara, R.,
the lifter wear. This is obtained by changing the
(1996), Development of a softsensor for
mill rotation speed, thus affecting the material
particle size monitoring, Minerals
impact forces inside the mill. The circuit feed
Engineering, 9, pp. 55-72.
rate is selected according to the downstream
Duarte, M., Suarez, A., Bassi, D., (2001),
requirements, and the decision to change or
Control of grinding plants using predictive
accept the off-spec product is an optimization
multivariable neural control, Powder
problem beyond this control scheme.
Technology, 115, pp 193-206.
Subsequently, the control strategy aims to
Guyot, O., Monredon, T., LaRosa, D.,
stabilize the circulating load or, if this is not
Broussaud, A., (2004), VisioRock, an
possible, bypassing the secondary mill is
integrated vision technology for advanced
control of comminution circuits, Minerals
conference on autogenous and
Engineering, 17, pp. 1227-1235.
semiautogenous grinding technology, ed.
Hagberg, L., et al. Edited, (1998), Keep it
Barrat, D.J., Allan, M.J., Mular, A.L.,
running: industrial asset management,
University of British Columbia, 2001,
Scemm, Vantaa Finland, 198 p.
Vancouver, Canada, pp. 386-393.
Herbst, J.A., Pate, W.T., 1996, On-line
Radhakrishnan, V.R., (1999), Model based
estimation of charge volumes in
supervisory control of a ball mill grinding
semiautogenous and autogenous grinding
circuit, Journal of Process Control, 9, pp.
mills, Proceedings of the international
conference on autogenous and Ramasamy, M., Narayanan, S.S., Rao, Ch.D.P.,
semiautogenous grinding technology, ed.
(2005), Control of ball mill grinding circuit
Barrat, D.J., Mular, A.L. Knight, D.A.,
using model predictive control scheme,
University of British Columbia, Vancouver,
Journal of Process Control, 15, pp. 273-283.
Canada, pp. 817-327.
Tano, K., Pålsson, B.I., Sellgren, A., (2005) On-
Herbst, J.A., Robertson, K, Rajamani, K.,
line lifter deflection measurements showing
(1983), Mill speed as a manipulated variables
flow resistance effects in grinding mills,
for ball mill grinding control, Proceedings of
Minerals Engineering, 18, pp.1077-1085.
the 4th IFAC symposium 22-25 Aug. 1983, ed.
Thoben, K-D., Jagdev, H., Eschenbaecher, J.
T. Westerlund, Helsinki, Finland, pp. 153-160.
(2001), Extended products: evolving
Jämsä-Jounela, S.-L., Laine, S., Ruokonen, E.,
traditional product concepts, Proceedings of
(1998), Ore type based expert system in
the 7th International Conference on
mineral processing plants, Particle & Particle
Concurrent Enterprising: Engineering the
Systems Characterization, 15, pp. 200-207.
Knowledge Economy Through Co-operation,
Jämsä-Jounela, S.-L., (1990) Modern
Bremen, Germany, pp. 429-439.
approaches to control of mineral processing,
Valderrama, W.R., Pontt J.O., Magne, L.O.,
Acta Polytechnica Scandinavica, Mathematics
Hernández, J.S., Salgado, F.I., Valenzuela,
and Computer Science Series No. 57,
J.S. Poze, R.E., (2000), The Impactmeter, e
new instrument for monitoring and avoiding
Jämsä-Jounela, S.-L., Remes, A., Suontaka, V.,
harmful high-energy impacts on the mill liners
(2005), Questionnaire on development of the
in SAG mills, Preprints of IFAC workshop on
grinding process, Internal project report -
future trend in automation in mineral and
Automin, Helsinki University of Technology,
metal processing, 2000, Helsinki, Finland,
Espoo, 50 p.
Kongas, M., Saloheimo, K., Pekkarinen, H.,
Yianatos, J.B., Lisboa, M.A., Baeza, D.R.,
Turunen, J. (2003), New particle size analysis
(2002), Grinding capacity enhancement by
system for mineral slurries, Preprints of IFAC
solid concentration control of hydrocyclone
workshop on new technologies for automation
underflow, Minerals Engineering, 15, pp.
of the metallurgical industry, Shanghai,
China, pp. 384-389.
Mathur, A., Parthasarathy, S., Gaikwad, S.,
(1999), Hybrid neural network multivariable
predictive controller for handling abnormal
events in processing applications, Control
Applications, Proceedings of the 1999 IEEE
International Conference, Hawai, USA, pp.
McCaffery, K.M., Katom, M., Craven, J.,
(2002), Ongoing evolution of advanced SAG
mill control at Ok Tedi, Minerals &
Metallurgical Processing, 19, pp. 72-80.
Napier-Munn, T.J., Morrel, S., Morrison, R.D.,
Kojovic, T., (1999), Mineral Comminution
Circuits: Their Operation and Optimisation,
Niemi, A.J., Tian, L., Ylinen, R., (1997), Model
predictive control for grinding system,
Control Engineering Practice, 5, pp. 271-8.
Pax, R.A., (2001), Non-contact acoustic
measurement of in-mill variables of SAG
mills, Proceedings of the international