See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/272552690

Grinding in Ball Mills: Modeling and Process Control

Article in Cybernetics and Information Technologies · June 2012

DOI: 10.2478/cait-2012-0012

CITATIONS READS

43 15,612

3 authors, including:

Vladimir Monov
Bulgarian Academy of Sciences
26 PUBLICATIONS 139 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Sensor data integration from intelligent sensor systems View project

All content following this page was uploaded by Vladimir Monov on 27 June 2016.

The user has requested enhancement of the downloaded file.

BULGARIAN ACADEMY OF SCIENCES

CYBERNETICS AND INFORMATION TECHNOLOGIES • Volume 12, No 2

Sofia • 2012

Grinding in Ball Mills: Modeling and Process Control

Vladimir Monov, Blagoy Sokolov, Stefan Stoenchev
Institute of Information and Communication Technologies, 1113 Sofia
E-mails: [email protected] [email protected] [email protected]

Abstract: The paper presents an overview of the current methodology and practice
in modeling and control of the grinding process in industrial ball mills. Basic
kinetic and energy models of the grinding process are described and the most
commonly used control strategies are analyzed and discussed.
Keywords: Ball mills, grinding circuit, process control.

I. Introduction
Grinding in ball mills is an important technological process applied to reduce the
size of particles which may have different nature and a wide diversity of physical,
mechanical and chemical characteristics. Typical examples are the various ores,
minerals, limestone, etc. The applications of ball mills are ubiquitous in mineral
processing and mining industry, metallurgy, cement production, chemical industry,
pharmaceutics and cosmetics, ceramics, different kinds of laboratory studies and
tests. Besides particle size reduction, ball mills are also widely used for mixing,
blending and dispersing, amorphisation of materials and mechanical alloying
[1, 49, 51].
As a construction, a ball milling device usually consists of a cylindrical vessel
mounted on an appropriate basis at both ends which allows rotation of the vessel
around the center axis. The mill is driven by a girth gear bolted to the shell of the
vessel and a pinion shaft moved by a prime mover. The prime movers are usually
synchronous motors equipped with an air clutch or gear transmission. After the mill
is charged with the starting material (ore, rock, etc.) and the grinding media (balls),

51
the milling process takes place during rotation as a result of the transfer of kinetic
energy of the moving grinding media into the grinding product.
The design of a ball mill can vary significantly depending on the size, the
equipment used to load the starting material (feeders), and the system for
discharging the output product. The size of a mill is usually characterized by the
ratio “length to diameter” and this ratio most frequently varies from 0.5 to 3.5. The
starting material can be loaded either through a spout feeder or by means of a single
or double helical scoop feeder. Several types of ball mills are distinguished
depending on the discharge system and these types are commonly known as
overflow discharge mill, diaphragm or grate discharge mill and centre-periphery
discharge mill, e.g. see [23]. In industrial applications, the inner surface of the mill
is lined with mill liners protecting the steel body of the mill and incorporating mill
lifters which help to raise the content of the mill to greater heights before it drops
and cascades down [36].
There are three types of grinding media that are commonly used in ball mills:
• steel and other metal balls;
• metal cylindrical bodies called cylpebs;
• ceramic balls with regular or high density.
Steel and other metal balls are the most frequently used grinding media with
sizes of the balls ranging from 10 to 150 mm in diameter [30]. Cylpebs are slightly
tapered cylindrical grinding media with rounded edges and equal length and
diameter with sizes varying from 8×8 to 45×45 mm. Their shape is developed to
maximize the grinding efficiency due to their high density and specific surface area
[17]. Ceramic balls with regular density are usually porcelain balls and the high
density balls are made with a high alumina oxide content and they are more
abrasion resistant. The basic properties of the milling bodies are their mass and size,
ware rate, influence on the particle breakage rate and energy efficiency of the
grinding process [15, 18, 22, 28]. A comparison between ball mils and cylpebs is
made in [ 42].
The speed of rotation of the mill determines three basic types of operation
modes: slow rotation (cascading), fast rotation (cataracting) and very fast rotation
(centrifugation). Each type is characterized by a specific trajectory of motion of the
charge in the mill and a different impact of the milling bodies on the ground
material. The grinding process can also take place in dry or wet conditions
depending on whether wetting agents are added to the starting material. Some
important characteristics of dry and wet grinding are studied in [29]. The particle
size reduction depends on the following basic factors:
• characteristics of the material charged in the mill (mass, volume, hardness,
density and size distribution of the charge);
• characteristics of the grinding media (mass, density, ball size distribution);
• speed of rotation of the mill;
• slurry density in case of wet grinding operation.
Quantitative estimations of these parameters can be found in [4, 5, 23].
An important characteristic of an industrial ball mill is its production capacity
which is measured in tons of production per hour. The production capacity depends
52
on mill dimensions, the type of the mill (overflow or grate discharge), the speed of
rotation, the mill loading, the final product size required from a given feed size
(coefficient of reduction), the work index of the material, the mill shaft power and
the specific gravity of the material. These parameters are thoroughly studied in
[4, 5] and an empirical relation is suggested expressing the mill capacity as a ratio
of the mill shaft power and the energy consumed in the grinding process.
In order to achieve the desired particle size, the milling under industrial
conditions is usually performed in grinding circuits including classifiers that
separate the material according to particle sizes. The simplest cases of open- and
closed-circuit systems are shown in Fig. 1.

Fig. 1

In the first case the output material is simply separated in fractions with
different particle sizes and the grinding process is actually not affected by the
classifier. In the second case the classifier returns coarse material back to the mill
feed and only the fine product is obtained at the output of the grinding circuit. In
practice various types of interconnections between mills and classifiers are possible
aimed at increasing the grinding efficiency of the overall process.
The main objectives of the grinding process include obtaining a desired
particle size distribution in the final product without metal or other possible
contamination, increasing the throughput of the grinding circuit and reducing the
production cost of the overall process. To achieve these objectives various
mathematical models and control methods are developed and applied in practice. In
this paper we provide a brief survey of the basic principles in modeling of the
grinding process and analyze several control strategies applied in the design of a
control structure and the implementation of an appropriate process control method.

II. Modeling of the process

II.1. Basic fragmentation mechanisms
The main idea in modeling all comminution processes, including the grinding
process, is to obtain mathematical relations between the size of the feed and the size
of the product. Particles in the feed repetitively reduce their size due to the
imparting energy of the grinding media which disrupts their binding forces. The
size reduction is a result of the following three basic fragmentation mechanisms
[25, 46].

53
 − Abrasion occurs when local low intensity stresses are applied and the result
is fine particles taken from the surface of the mother particle and particles of size
close to the size of the mother particle (Fig. 2a).

a b

c
Fig. 2

− Cleavage of particles occurs when slow and relatively intense stresses are
applied (compression) which produce fragments of size 50-80 % of the size of the
initial particle (Fig. 2b).
− Fracture is a result of rapid applications of intense stresses (impact) which
produce fragments of relatively small sizes with a relatively wide particle size
distribution (Fig. 2c).
In practice the three different mechanisms never occur alone and the process
of particle size reduction involves all of them with possible predominance
depending on the type of the mill, the operating conditions and the type of the
material being ground.
Several basic concepts are commonly used in modeling of the grinding
process. The starting material naturally consists of particles which differ
significantly in size which makes it necessary to define different size classes.
Standard sieves are used to determine quantitatively the size of the particles in each
size class. In the theory of breakage of solids, the fragmentation process is
decomposed into a series of steps consisting of two main operations [46]:
(i) selection of a fraction of the material to be broken,
(ii) breakage of the selected material producing a given distribution of
fragment sizes.
These operations are characterized by two functions: the selection function Si,
i=1, 2, ..., n, and the breakage function bij, n ≥ i ≥ j ≥ 1 , where n is the number of
the size classes. The selection function Si, (also called probability of breakage or
specific breakage rate [23, 45]) represents the probability for a particle of size of x i
to be broken at a given fragmentation step, where xi is the lower limit of the
particle size class i. Thus, S1, S2, … , Sn are the mass fractions of the material in
each size class that are selected for size reduction. The breakage function bij (also

54
known as the distribution function [45, 46]) describes the distribution of fragment
sizes obtained after a breakage of particles of size xj. Thus, b1j, b2j, …, bnj are the
mass fractions of particles in size classes 1, 2, …, n after a breakage of particles in
size class j.
The mechanism of breakage is illustrated in [23] by a diagram shown in Fig. 3.
The left column of the figure shows the size distribution of the feed. The application
of forces on the particles in different size classes is shown by solid arrows and the
movement of fragments to the same or lower size of classes is indicated by dotted
arrows. The breakage functions are shown in the third column and the forth column
contains products obtained after a number of size reductions. During the process,
the mass of feed in size class 1 is distributed in the lower size classes. At certain
time, the mass fraction in size 1 will disappear as the particles are broken and
distributed in the smaller sizes. It should be noted however, that the total mass
remains constant.

Fig. 3

The selection function and the breakage function are basic parameters in the
modeling of all communition processes. Three types of models are commonly
accepted in literature [23, 45]: matrix, kinetic and energy models. A general
principle in the development of each model is to establish mass balance or energy
balance equations relating to the mass components or the energy involved in the
process. The matrix type models are used when the size reduction is predominantly
considered as a discrete process with each discrete step including the three
operations selection-breakage-classification. For modeling of the batch grinding in
continuous steady state mills, the size reduction is treated as a continuous process
and the corresponding mathematical models take into account the time dependant

55
parameters of the process. In this case the kinetic and energy type of models are
most frequently used.

II.2. Kinetic and energy models

Several basic assumptions are made in modeling of the grinding process in various
types of mills and milling circuits. Most frequently, it is assumed that the mill's
content is uniform and thoroughly mixed by the rotation of the mill and the
movement of the grinding media [44]. In this case the model is known as a perfectly
mixed model. In some cases the mill’s charge is considered to be perfectly mixed in
the radial direction and only partially mixed in the axial direction. Another
important assumption is that the particles of different sizes are broken in a similar
way (normalized breakage) and that no agglomeration processes take place during
the size reduction [23, 45].
The kinetic models of the grinding process are based on mass-balance
equations describing the process in the different size intervals. Assuming that the
mill is perfectly mixed in the radial direction and partially mixed in the axial
direction, a kinetic model of second order is given in [45] in the form
i −1
dwi (l , t ) d 2 wi ( l , t ) dwi ( l , t ) ,
(1) = − S i wi ( l , t ) + ∑ S j bij w j (l , t ) + Di 2
− ui
dt j =1 dl dl
where
t is the grinding time;
l – the space coordinate in the axial direction;
wi (l , t ) – the mass fraction of material in the i-th size class;
bij – the breakage function;
S i – the selection function;
Di – a mixing coefficient;
ui – the velocity of convective transport of particles in the axial direction.
The left hand side of equation (1) represents the variation of the mass fraction
of the material in size class i within a time interval [t, t+dt]. The first and second
term in the right hand side represent the mass of disappearing and appearing
particles in this class, respectively. The third term describes the axial dispersion and
the last term represents the convective transport of particles in the axial direction
with velocity ui. The differential equation (1) has the following boundary
conditions:
(2) wi (l , 0) = f i (l ) ,
dw (l , t )
(3) wi (l , t ) = ui wi (l , t ) − Di i for l = 0,
dl
dwi (l , t )
(4) = 0 for l = L,
dl
where f i (l ) is the mass fraction of the feed in size class i and L is the length of
the mill.

56
 Equation (1) with conditions (2)-(4) represents the basic kinetic model of the
process. Depending on the specific operational conditions of the mill, different
variants of this model are also known. Most frequently, the perfectly mixed model
is used under the assumption that the charge is thoroughly mixed and uniform in
both radial and axial direction. In this case, the third and fourth term in (1) can be
neglected and the grinding kinetics is described in the form [3, 25, 46]
i −1
= − Si wi (t ) + ∑ S j bij w j (t ) .
dwi (t )
(5)
dt j =1

That equation can be written in a matrix form as

dw(t )
(6) = ( B − I ) Sw(t ) ,
dt
where S is a diagonal matrix with diagonal elements Si , i = 1, 2, ..., n, B is a lower
triangular matrix with elements bij , n ≥ i > j ≥ 1 , w(t) is a vector with elements
wi (t ), i = 1, 2, ..., n, and I denotes the identity matrix. Matrix ( B − I ) S in (6) has a
lower triangular form with diagonal elements − S1 , − S2 , ..., − Sn . Under the
assumption that functions bij and S i are known and time independent, the solution
of (6) is given by
(7) w(t ) = exp[( B − I ) St ]w(0),
where exp[(B − I ) St ] is the matrix exponent and w(0) is the vector of initial
conditions with elements equal to the mass fractions of the feed in the respective
class sizes. The explicit formulas for wi (t ) , i = 1, 2, ..., n, are known as the Reid
solution to the batch grinding equation and can be found in [3, 23].
A cumulative form of equation (5) is also used in modeling of the grinding
process, i.e.
i −1

∑
dRi (t )
(8) = − Si Ri (t ) + R j (t )[S j +1Bij +1 − S j Bij ] ,
dt j =1

where Bij = ∑k =i +1 bkj is the cumulative breakage function and Ri (t ) = ∑ j =1 w j (t )

n i

is the cumulative mass fraction of particles with size greater than x i , the lower limit
of the particle size class i.
Finding the solutions of equations (5) and (8) pre-supposes a preliminary
knowledge of the breakage and selection functions bij and Si. However, for a
particular process these functions are not known apriori and they are usually
determined by experimental tests and a consecutive treatment and estimation of the
experimental results. Different methods for determination of these functions and
some typical graph plots of breakage functions are given in [23, 25, 46].
Approximate solutions of the cumulative grinding equation (8) in an explicit form
are also shown in [3].
Mathematical models based on energy-balance equations are also used in order
to describe the grinding process. A linear model which is analogous to (5) is

57
developed in [21] where the batch grinding kinetics is expressed in terms of the
specific energy as an independent variable instead of the grinding time. The authors
of this reference conducted a series of experiments accurately measuring the
specific energy consumed by the ball mill under various operating conditions and
different ground material. An analysis of the results obtained in dry milling
conditions shows that the size-discretized breakage rate functions are proportional
to the specific energy input to the mill and that the breakage distribution functions
can be considered invariant [21]. In this case an energy-balance equation modeling
the grinding process can be given in the form
i −1
dwi ( E )
(9) = − SiE wi ( E ) + ∑ S Ej bij w j ( E ) ,
dE j =1

where E is the specific energy input to the mill and S iE is the energy-normalized
breakage rate parameter defined as
Si .
(10) SiE =
P /W
In equation (10) P is the power input to the mill and W is the mass of the feed
material in the mill.
A simple analysis and comparison between the mass-balance model given by
(5) and the energy-balance model in the form (10) is as follows. Both equations (5)
and (9) represent linear models which are relatively simple and easy to use in
studying the first-order breakage kinetics of the process. The model (5) is
thoroughly studied and solutions of the differential equation in explicit forms are
obtained in literature under different assumptions and degrees of approximation. In
general, a solution of (5) describes the change in particle size distribution of the
ground material as a function of the grinding time. On the other hand, the model (9)
shows that the breakage kinetics can be accurately analyzed in terms of the
consumed specific energy instead of time. Furthermore, the power input to the mill
can be accurately measured which makes it possible to use the measured data as an
efficient control parameter in the process. It is pointed out in [21] that the energy
model can also be very useful with regard to mill scale-up and the analysis of other
types of comminution systems such as roll mills. Except for the linear models, it
should be noted that a subject of particular interest is the development of more
precise and complicated mathematical models of the grinding process including
nonlinearities and time-dependant selection and breakage functions [7, 20].
Computer simulations based on the discrete element method [31, 32, 38] are also
widely used in studying dynamical properties of the milling process.

III. Process control methods

The control of a grinding circuit is a difficult task due to many factors such as
nonlinear and undetermined character of the process, inaccuracies in the
mathematical model, the presence of interacting process variables with substantially
different dynamics, the influence of unmeasured disturbances and large time delays,
rough operating conditions and inability to use precise and reliable sensors. On the
58
other hand, an efficient control of the process is of great importance for increasing
the throughput of the circuit and quality of the final product as well as for a
significant reduction of the production costs.

III.1. Process variables and characteristics

From a control point of view, a ball mill grinding circuit represents an
interconnected multivariable system with strong interactions among process
variables. A typical structure of a closed-loop circuit for wet grinding consists of a
ball mill, sump and classifier [10, 13, 33, 39] and it is schematically shown in
Fig. 4.

Fig. 4

The process input variables in the figure are: u1 − mill feed water flow rate,
u2 − fresh ore feed rate, u3 − mill critical speed fraction, u4 − sump dilution water
flow rate and u5 − sump discharge flow rate. The values of these variables can be
manipulated in order to control the output variables: y1 − product mass fraction with
size of particles less than a given value, y2 − product solids concentration,
y3 − product flow rate, y4 − slurry level in the sump, y5 − sump solids concentration.
The most important disturbances to the process are ore hardness changes and feed
size variations. An input-output model of the process can be written in a vector
matrix form:
…
(11) . ,
…
where is the transfer function relating the i-th input and j-th output for
i, j = 1, …, 5. The transfer functions are usually experimentally determined by
applying step changes in the input and measuring the output responses. In order to
obtain correct results usually a sufficiently large number of experiments should be
carried out. Also, depending on the grinding circuit configuration and equipment,
59
different sets of input and output variables can be used in the control design [8, 33,
40].
The process control in a ball mill grinding circuit faces severe difficulties due
to the following well known characteristics:
• the process is nonlinear with immeasurable disturbances and unmodelled
dynamics;
• there are strong interconnections among variables so that each input
variable interacts with multiple output variables;
• the time constants of the process have values in a wide range and there are
significant time delays in some input-output pairs;
• the system model contains a number of integrators;
• the process parameters vary in time as the circuit ages;
• there are technological constraints on the manipulated and controlled
variables;
• the measurements are unreliable and noisy.
The main control objectives are as follows. At the first place, it is necessary to
maintain a stable operation at fixed set-points of the output variables. Within this
objective, it is most important to maintain a stable product size distribution
measured as a percentage of the output material with size of particles less than a
given value. Alternatively, the control objective can be formulated as an
optimization problem including a performance criterion which is to be optimized
subject to certain constraints. It can be either the maximization of the grinding
circuit throughput or minimization of the production costs. Various constraints in
the optimization problem are the minimal or maximal values of the input, output
and internal process variables which represent equipment limits, safe operation
requirements or environmental regulations.
Due to the inherent process characteristics and constraints, the control design
for a grinding circuit appears to be a challenging problem for most of the control
methods applied in practice. In the rest of this section, two of the most frequently
used control approaches are briefly described and an analysis of their main
advantages and drawbacks is presented.

III.2. Decentralized control

According to the results from a statistical study [48], the decentralized control
approach is most frequently used in ball mill grinding circuits. The main advantage
of this approach is the distributed control structure and its easy implementation in
practice. Such a structure may consist of several single-input single-output control
loops involving different process variables. The schematic diagram of a two-input
two-output system is shown in Fig. 5, where the transfer functions of the two open
loops with interconnections are denoted by , , 1, 2, and the transfer
functions of the controllers are and .

60
 Fig. 5

A frequency-based method for the design of extended PID controllers is

described in [35] where the transfer functions and are determined in
the form
(12) , .

From (12) it is seen that the two controllers are interdependent due to the
system interconnections and hence, the individual tuning of each controller depends
on the parameters of the other. To overcome this problem, a procedure to determine
the exact values of and is outlined in [35] by using a supplementary
information from the system closed-loop specifications.
An important problem in the design of a distributed grinding process control is
the choice of appropriate pairs of manipulated and controlled variables. When the
input-output pairs are not properly selected, undesirable interactions between the
control loops take place resulting in a poor control performance. Two most common
variants of variable pairings in a grinding circuit with a sump and a hydrocyclone
classifier are shown in Table 1 [13].
Table 1
Pairing Controlled Variable Manipulated Variable
Particle size hydrocyclone overflow Sump water dilution rate
I
Hydrocyclone feed rate Fresh solids feed (& dilution rate)
Particle size hydrocyclone overflow Fresh solids feed (& dilution rate)
II
Hydrocyclone feed rate Sump water dilution rate

Analysis of the dynamic responses to a set point change of the product size
shows that there is a significant interaction between the control loops in both cases
of pairing I and pairing II. Due to the strong interactions among the variables, a
proper choice of the manipulated-controlled pairs should be based on information
and analysis of both the steady state and the dynamic behavior of the control loops.
In case of more than two input and two output variables, the existing control theory
offers a valuable tool for selection of manipulated-controlled variable pairing which
is known as the relative gain technique [6]. The method involves construction of a
relative gain array associated with the system and the choice of pairs is based on the
analysis of its structure and properties.
61
 Another frequently used approach to compensate the system interactions in a
decentralized control structure consists of including additional decoupling
controllers between the control loops of the system. An illustration of a system
with two control loops and decouplers is shown in Fig. 6, where , , 1, 2,
are transfer functions of the decoupling controllers.

Fig. 6

The system is described by the vector-matrix equations ,

, which give the input-output equation
(13) ,
where
0
(14) , , , .
0
Since is a diagonal matrix, the aim is to determine the transfer functions of
decoupling controllers such that is a diagonal matrix, i.e.
0
(15) .
0
From (15), we have
0
(16) .
0
If and are determined as det / and det / ,
then the transfer functions of decoupling controllers are obtained from (16) in the
form

(17) 1, 1, , .

The above equations show that the decoupling controllers are independent on
the forward path controllers and . This is an important advantage of
the method as far as the controller modes of operation and tuning of and
can be changed without loss of decoupling of the control loops.

62
III.3. Multivariable control
Multivariable control methods generally dominate in industrial processes with
strong interactions among process variables. In the control of ball mill grinding
circuits, a multivariable approach based on Model Predictive Control (MPC)
strategy is successfully used [26, 33, 39]. This approach has numerous industrial
applications where the control is designed to drive the process from one constrained
steady state to another. The main objectives of the MPC system are outlined in [37]
as follows:
• prevent violation of input and output constraints;
• drive the controlled variables to their steady-state optimal values (dynamic
output optimization);
• drive the manipulated variables to their steady-state optimal values using
remaining degrees of freedom (dynamic input optimization);
• prevent excessive movement of manipulated variables;
• when signals and actuators fail, control as much of the plant as possible.
The MPC utilizes control structures incorporating an internal process model
and a predictive control computation. The internal model is used to predict the
future output of the process. The future actions of the manipulated variables are
determined by solving a finite horizon optimization problem of minimizing the
difference between the desired reference trajectory and the predicted output. The
optimization problem is solved at each sampling interval and as a solution a
sequence of future control actions is obtained. At the next sampling interval only
the first control action is applied and the procedure is repeated. Reference [37]
presents an extensive survey of control algorithms employing the MPC approach.
An up-to-date review of the current practice and challenges of MPC can be found in
[14] and an exposition of the theory of MPC is presented in a monograph [16].
One of the main MPC methods is based on the Dynamic Matrix Control
(DMC) algorithm which was developed in the late 70’s of the last century and it
was initially intended for use in petroleum industry. Subsequently, control strategies
utilizing the DMC algorithm have become a powerful tool for process control in
various ball mill grinding circuits [8, 10, 39]. In [8] an implementation of the DMC
algorithm is presented as it is shown in Fig. 7.

Fig. 7

63
 If denotes the length of output prediction horizon,then the predicted output
vector
T
1 1 ,…,
is obtained as
(18) 1 1 ,
T
where 1 1 ,…, is the model output, G0 is a
coefficient matrix and

is the difference between the process and model output vectors. The predicted
control vector is in the form
∆ ∆ , …, ∆ 1 T,
where M is the length of control prediction horizon. The values of ∆ are
determined by a minimization of the following quadratic objective function [8]
(19) 1 1 T 1 1 +
T
∆ ∆ ,
where is an output weighting coefficient matrix and is a control weighting
coefficient matrix. Constraints on the process variables are given in the form
, 0, 1, … , 1,
∆ 1 ∆ , 0, 1, … , 1,
, 0, 1, … , 1.
In applying MPC algorithm to ball mill grinding process, it is necessary to
properly select the values of several important control parameters, such as the
length of model output prediction horizon, the length of control prediction horizon,
the weighting coefficients of output deviations and manipulated input variations.
The length of output prediction horizon is generally set long enough in order to
capture the steady state behavior of the output. Concerning the control prediction
horizon M, it is clear that an increasing of M will improve the control performance
but the price will be an increased amount of computations. Generally, M is shorter
than P. On the other hand, the desired output behavior and the control cost
determine the choice of weighting coefficients in the objective function. If a tighter
control of some output variable is desired, it can be achieved by a higher value of
the respective weighting coefficient. Similarly, excessive variations in the
manipulated variables can be suppressed by increasing the values of control
weighting coefficients.
The surveyed simulation studies and practical applications of decentralized
and multivariable control systems in ball mill grinding circuits reveal the following
important characteristics of the two control strategies.
• Decentralized systems with local PI or PID controllers predominate in
practical applications due to the simple control system implementation. The process
variables are coupled in single-input single-output control loops, such that one
manipulated variable is used for one controlled variable. The main problems in this
approach are the choice of variable pairing and the decoupling of control loop
64
interactions by an appropriate tuning or by introducing decoupling controllers. On
the other hand, the multivariable approach avoids these problems at the expense of
a more complicated control system.
• Multivariable methods using MPC algorithms are becoming increasingly
popular with successful implementations reported in the literature. Advantages of
this approach are its capability to cope with the strong interactions among process
variables and the possibility for a better control system tuning. However, the MPC
requires an accurate process modeling at the design stage and substantially more
complicated computations during the process control.
• Decentralized PID controllers normally give satisfactory results with
respect to the steady state error and closed loop responses to small set point changes
provided that a proper variable pairing is used and the coupling among control
loops is taken into account. However, results from simulation studies show
significant advantages of MPC as compared to decentralized PID controllers. In
particular, MPC schemes explicitly take into account the technological constraints
of the manipulated and controlled variables and allow for achieving optimal
operating conditions of the control system. Also this approach successfully
overcomes problems associated with unmeasured disturbances, time-delays and
nonlinearities of the process.
• The choice of a control method and control system structure for a particular
ball mill grinding process should normally take into account certain economic value
indicators. It turns out that it is difficult to estimate the potential economic profit in
using more complex control structures and computations instead of simplified
control schemes. Furthermore, an investigation [47] estimates the impact of particle
size distribution and comes to a conclusion that a larger economic profit can be
achieved due to the set point move to a better operating point than due to a tighter
process control. In general, the energy efficiency and economic assessment of the
process control in industrial ball mills remains an open problem motivating future
research activities and comprehensive studies [2, 19, 26, 27, 48].
We shall conclude this section by mentioning several other advanced
techniques applied to the process control in ball mills. An approach involving
simultaneous decoupling of interactions and closed loop pole assignment is
proposed in [24]. A control system incorporating disturbance observer and MPC is
designed in [50]. Innovative control platforms are described in [43] and a
comparison of multivariable PI, fuzzy and model predictive control is presented in
[40]. A survey of adaptive control methods and their application to the grinding
process is given in [34]. Simulation results with expert system based control,
supervisory control and observer based control are reported in [9, 11, 12]. An
integrated automation system for monitoring and grinding process control is
designed in [41] and a neurocontrol approach is developed in [13].

65
IV. Conclusion
The existing mathematical models of the grinding process are developed on the
basis of mass balance or energy balance equations which describe the particle size
reduction of the ground material as a function of the grinding time or in terms of the
consumed specific energy. In both cases the main parameters of the model are
selection and breakage functions which are generally not known apriori and their
determination requires additional experimental studies. A subject of particular
interest in the process modeling is the development of more precise and
complicated mathematical models including nonlinearities and time-variant
selection and breakage functions. Concerning the process control, decentralized and
multivariable control methods predominate in the surveyed simulation results and
practical applications. Nevertheless, theoretical and simulation studies of control
systems employing advanced control techniques, such as adaptive, expert system
and fuzzy logic control are gaining increased interest. Important open problems
which motivate further research activities in this area are related with the energy
efficiency of the ball mill and the economic assessment of the process control.

References
1. A j a a l, T., R. W. S m i t h, W. T. Y e n. The Development and Characterization of a Ball Mill for
Mechanical Alloying. – Canadian Metallurgical Quarterly, Vol. 41, 2002, No 1, 7-14.
2. B a u e r, M., I. K. C r a i g. Economic Assessment of Advanced Process Control – A Survey and
Framework. – Journal of Process Control, Vol. 18, 2008, 2-18.
3. B e r t h i a u x, H., C. V a r i n o t, J. D o d d s. Approximate Calculation of Breakage Parameters
from Batch Grinding Tests. – Chemical Engineering Science, Vol. 51, 1996, No 19,
4509-4516.
4. B o n d, F. C. Crushing and Grinding Calculations. Part I. – British Chemical Engineering, Vol. 6,
1961, No 6, 378-385.
5. B o n d, F. C. Crushing and Grinding Calculations. Part II. – British Chemical Engineering, Vol. 6,
1961, No 8, 543-548.
6. B r i s t o l, E. G. On a New Measure of Interaction for Multivariable Process Control. – IEEE
Trans. Auto. Control, AC-11, Vol. 1, 1966, 133-134.
7. C a p e c e, M., E. B i l g i l i, R. D a v e. Identification of the Breakage Rate and Distribution
Parameters in a Non-Linear Population Balance Model for Batch Milling. – Powder
Technology, Vol. 208, 2011, 195-204.
8. C h e n, X. S., Q. L i, S. M. F e i. Constrained Model Predictive Control in Ball Mill Grinding
Process. – Powder Technology, Vol. 186, 2008, 31-39.
9. C h e n, X. S., S. H. L i, J. Y. Z h a i, Q. L i. Expert System Based Adaptive Dynamic Matrix
Control for Ball Mill Grinding Circuit. – Expert Systems with Applications, Vol. 36, 2009,
716-723.
10. C h e n, X. S., J. Y. Z h a i, S. H. L i, Q. L i. Application of Model Predictive Control in Ball Mill
Grinding Circuit. – Minerals Engineering, Vol. 20, 2007, 1099-1108.
11. C h e n, X. S., Q. L i, S. M. F e i. Supervisory Expert Control for Ball Mill Grinding Circuits. –
Expert Systems with Applications, Vol. 34, 2008, 1877-1885.
12. C h e n, X. S., J. Y a n g, S. H. L i, Q. L i. Disturbance Observer Based Multi-Variable Control of
Ball Mill Grinding Circuits. – Journal of Process Control, Vol. 19, 2009, 1205-1213.
13. C o n r a d i e, A. E., C. A l d r i c h. Neurocontrol of a Ball Mill Grinding Circuit Using
Evolutionary Reinforcement Learning. – Minerals Engineering, Vol. 14, 2001, Issue 10,
1277-1294.

66
14. D a r b y, M. L., M. N i k o l a o u. MPC: Current Practice and Challenges. – Control Engineering
Practice, Vol. 20, 2012, 328-342.
15. D e n i z, V. A Study on the Specific Rate of Breakage of Cement Materials in a Laboratory Ball
Mill. – Cement and Concrete Research, Vol. 33, 2003, 439-445.
16. D i n g, B. C. Modern Predictive Control. Taylor and Francis Group, 2010.
17. DOERING International GmbH.
http://www.doering-sinn.de/en/grinding-media.html
18. E r d e m, A. S., S. L. E r g u n. The Effect of Ball Size on Breakage Rate Parameter in a Pilot
Scale Ball Mill. – Minerals Engineering, Vol. 22, 2009, 660-664.
19. F u e r s t e n a u, D. W., A. Z. M. A b o u z e i d. The Energy Efficiency of Ball Milling in
Comminution. – Int. Journal Miner. Process., Vol. 67, 2002, 161-185.
20. F u e r s t e n a u, D. W., A. D e, P. C. K a p u r. Linear and Nonlinear Particle Breakage Processes
in Comminution Systems. – Int. Journal Miner. Process, Vol. 74S, 2004, 317-327.
21. F u e r s t e n a u, D. W., P. C. K a p u r, A. D e. Modelling Breakage Kinetics in Various Dry
Comminution Systems. – KONA Powder and Particle Journal, Vol. 21, 2003, 121-132.
22. F u e r s t e n a u, D. W., J. J. L u t c h, A. D e. The Effect of Ball Size on the Energy Efficiency of
Hybrid High-Pressure Roll Mill/Ball Mill Grinding. – Powder Technology, Vol. 105, 1999,
199-204.
23. G u p t a, A., D. Y a n. Mineral Processing Design and Operations: An Introduction. Elsevier, May
2006. 718 p.
24. H a u k s d ó t t i r, A. S., M. I e r a p e t r i t o u, S. L. J. J o u n e l a, U. Z ü h l k e. A Decoupling
Grinding Circuit Control System. – In: 15th IFAC Triennial World Congress, Barcelona,
Spain, 2002.
25. H e n n a r t, S. L. A., W. J. W i l d e b o e r, P. v a n H e e, G. M. H. M e e s t e r s. Identification of
the Grinding Mechanisms and Their Origin in a Stirred Ball Mill Using Population Balances.
– Chemical Engineering Science, Vol. 64, 2009, 4123-4130.
26. H o d o u i n, D. Methods for Automatic Control, Observation, and Optimization in Mineral
Processing Plants. – Journal of Process Control, Vol. 21, 2011, 211-225.
27. H o d o u i n, D., S-L. J a m s a-J o u n e l a, M. T. C a r v a l h o, L. B e r g h. State of the Art and
Challenges in Mineral Processing Control. – Control Engineering Practice, Vol. 9, 2001,
995-1005.
28. K o t a k e, N., K. D a i b o, T. Y a m a m o t o, Y. K a n d a. Experimental Investigation on a
Grinding Rate Constant of Solid Materials by a Ball Mill – Effect of Ball Diameter and Feed
Size. – Powder Technology, Vol. 143/144, 2004, 196-203.
29. K o t a k e , N., M. K u b o k i, S. K i y a, Y. K a n d a. Influence of Dry and Wet Grinding
Conditions on Fineness and Shape of Particle Size Distribution of Product in a Ball Mill. –
Advanced Powder Technology, 2010, doi:10.1016/j.apt.2010.03.015.
30. K u z e v, L., T. P e n c h e v, D. K a r a s t o y a n o v. New Shape Milling Bodies for Ball Mills. –
Problems of Engineering Cybernetics and Robotics, Vol. 61, 2009, 11-19.
31. M i s h r a, B. K., C. T h o r n t o n. An Improved Contact Model for Ball Mill Simulation by the
Discrete Element Method. – Advanced Powder Technology, Vol. 13, 2002, No 1, 25-41.
32. M i s h r a, B. K. A Review of Computer Simulation of Tumbling Mills by the Discrete Element
Method. Part II – Practical Applications. – Int. Journal Miner. Process., Vol. 71, 2003,
95-112.
33. M u l l e r, B., P. L. de V a a l. Development of a Model Predictive Controller for a Milling
Circuit. – Journal of the South African Institute of Mining and Metallurgy, Vol. 100, 2000,
No 7, 449-453.
34. N a j i m, K., D. H o d o u i n, A. D e s b i e n s. Adaptive Control: State of the Art and an
Application to Grinding Process. – Powder Technology, Vol. 82, 1995, 59-68.
35. P o m e r l e a u, A., D. H o d o u i n, A. D e s b i e n s a, E. G a g n o n. A Survey of Grinding
Circuit Control Methods: From Decentralized PID Controllers to Multivariable Predictive
Controllers. – Powder Technology, Vol. 108, 2000, 103-115.
36. P o w e l l, M. S., I. S m i t, P. R a d z i s z e w s k i, P. C l e a r y, B. R a t t r a y, K. E r i k s s o n, L.
S c h a e f f e r. The Selection and Design of Mill Liners. – In: Advances in Comminution.
S. K. Kawatra, Ed. ISBN-13:978-0-87335-246-8, Society for Mining, Metallurgy, and
Exploration, Inc., Colorado, USA, 2006, 331-376.
67
 37. Q i n, S. J., T. A. B a d g w e l l. A Survey of Industrial Model Predictive Control Technology. –
Control Engineering Practice, Vol. 11, 2003, 733-764.
38. R a j a m a n i, R. K., B. K. M i s h r a, R. V e n u g o p a l , A. D a t t a. Discrete Element Analysis
of Tumbling Mills. – Powder Technology, Vol. 109, 2000, 105-112.
39. R a m a s a m y, M., S. S. N a r a y a n a n b, C. D. P. R a o. Control of Ball Mill Grinding Circuit
Using Model Predictive Control Scheme. – Journal of Process Control, Vol. 15, 2005,
273-283.
40. R e m e s, A., J. A a l t o n e n, H. K o i v o. Grinding Circuit Modeling and Simulation of Particle
Size Control at Siilinjärvi Concentrator. – International Journal of Mineral Processing,
Vol. 96, 2010, 70-78.
41. R e m e s, A., J. K a r e s v u o r i, H. P e k k a r i n e n, S-L. J ä m s ä-J o u n e l a. Integrated
Approach to Monitoring and Control of Mineral Grinding Processes. – In: 1st IFAC
Workshop on Applications of Large Scale Industrial Systems, Finland, 2006.
42. S h i, F. Comparison of Grinding Media – Cylpebs Versus Balls. – Minerals Engineering, Vol. 17,
2004, 1259-1268.
43. S m i t h, G. C., L. J o r d a a n, A. S i n g h, V. V a n d a y a r, V. C. S m i t h, B. M u l l e r, D. G.
H u l b e r t. Innovative Process Control Technology for Milling and Flotation Circuit
Operations. – Journal of the South African Institute of Mining and Metallurgy, May 2004,
353-365.
44. T a v a r e s, L. M., R. M. D. C a r v a l h o. Modeling Breakage Rates of Coarse Particles in Ball
Mills. – Minerals Engineering, Vol. 22, 2009, 650-659.
45. T o n e v a, P., W. P e u k e r t. Modelling of Mills and Milling Circuits. – Handbook of Powder
Technology. Vol. 12. 2007, 874-911. Particle Breakage. A. Salman, M. Ghadiri, M.
Hounslow, Eds. Elsevier, 2007. 1240 p.
46. V a r i n o t, C., S. H i l t g u n, M. N. P o n s, J. D o d d s. Identification of the Fragmentation
Mechanisms in Wet-Phase Fine Grinding in a Stirred Bead Mill. – Chemical Engineering
Science, Vol. 52, 1997, No 20, 3605-3612.
47. W e i, D., I. K. C r a i g. Economic Performance Assessment of Two ROM Ore Milling Circuit
Controllers. – Minerals Engineering, Vol. 22, 2009, 826-839.
48. W e i, D., I. K. C r a i g. Grinding Mill Circuits – A Survey of Control and Economic Concerns. –
In: Proceedings of the 17th World Congress of the International Federation of Automatic
Control, Seoul, Korea, 6-11 July 2008, 1000-1005.
49. W i l l s, B. A., T. J. N a p i e r-M u n n. Mineral Processing Technology. Elsevier Science &
Technology Books. 7th Edition. 2006. 450 p.
50. Y a n g, J., S. L i, C. X i-S o n g, Q. L i. Disturbance Rejection of Ball Mill Grinding Circuits
Using DOB and MPC. – Powder Technology, Vol. 198, 2010, 219-228.
51. Z o z, H., R. R e i c h a r d t, J. S. K i m. Application and Design of Drum Mills, PM²TEC’2002. –
In: Proceedings of the World Congress on Powder Metallurgy & Particulate Materials, 16-21
June 2002, Orlando, FL, USA.

68

View publication stats