ABSTRACT 112

HPGR modelling with mineral liberation for plant

optimization
Manuel Gonzalez Fernandez & Stephane Brochot*
CASPEO, Santiago, Chile, Process Engineer, +56-2-22255547, [email protected]
CASPEO, Orleans, France, Scientific Manager, +33-238643615, [email protected]

ABSTRACT
HPGR (High-Pressure Grinding Rolls) is widely used today for grinding of various kinds of ores
such as iron ore or base and precious metal ores. It can be used as first stage of grinding replacing
tertiary crusher and SAG mill, but also as regrinding stage to enhance mineral liberation. In
addition to better energy efficiency, the advantage of HPGR is to produce much more internal
particle fractures allowing faster leaching as it has been demonstrated for some gold ores. This
grinding mode of compression and fracture generation also promotes mineral liberation without
overgrinding.
Mineral processing modelling and simulation associated to detailed ore characterization constitute
a powerful tool for plant design and optimization. The mathematical models of the units of
operations (grinding and separation) have to reproduce the mineral deportment for which the
mineral liberation plays a central role. The present paper is focused on the HPGR modelling with
the objective of plant optimization or advanced design from pilot plant tests. The traditional
approach of the population balance linked to the energy dissipation and the residence time
distribution is generalized in association with a mineral liberation model. The effect of the specific
HPGR breakage matrix and grinding function on the mineral liberation rate is discussed and
compared to the tumbling mill. The implementation of this mathematical model into a plant
simulation tool is illustrated by the example of the optimization of an iron ore concentrator.

*Corresponding author: CASPEO, Scientific Manager, 3 avenue Claude Guillemin – BP 36009 –

45060 ORLEANS CEDEX 2 - FRANCE. Phone: +33 2 38643615. Email: [email protected]
INTRODUCTION
HPGR (High-Pressure Grinding Rolls) is widely used today for grinding of various kinds of ores such as
iron ore or base and precious metal ores. It can be used as first stage of grinding replacing tertiary crusher
and SAG mill, but also as regrinding stage to enhance mineral liberation. In addition to better energy
efficiency, the advantage of HPGR is to produce much more internal particle fractures allowing faster
leaching as it has been demonstrated for some gold ores (Klymowsky et al., 2002; Morley, 2006). This
grinding mode of compression and fracture generation also promotes mineral liberation without
overgrinding.
Mineral processing modelling and simulation associated to detailed ore characterization constitute a
powerful tool for plant design and optimization. The mathematical models of the units of operations
(grinding and separation) have to reproduce the mineral deportment, for which the mineral liberation
plays a central role. The population balance approach answers this objective. The population balance
between the different particle size and particle type classes during size reduction takes into account the
breakage rate per particle size and type and the progeny size distribution. Depending on the size
reduction technology, the breakage rate can be described by a simple probability function or by a first
order kinetics. The progeny size distribution is described by a breakage matrix including the distribution
of produced particles in the different types (Brochot et al., 2006; Brochot & Botané, 2012). Various
approaches have been proposed to describe the mineral liberation (Barbery, 1991; King, 1994; 2001;
Wiegel, 2006; 2011) and how it occurs during comminution. They are mainly based on non-preferential
breakage (Barbery, 1991; Gay, 1999; King, 1994; 2001; Stamboliadis, 2008; Wiegel, 2006; 2011) but recent
works try to take into account the preferential breakage which can occur for some ores (Gay, 2004; Khalesi
et al., 2009). In the case of non-preferential breakage, the size reduction and mineral liberation models are
more or less independent. The breakage model describes the progeny fragment size distribution and the
liberation model distributes these fragments into the various liberation classes regardless of the breakage
events.
The USIM PAC software package offers a comminution model library based on these principles (Brochot
et al., 2006). It is a user-friendly steady-state simulator that allows mineral processing engineers and
scientists to model plant operations using available experimental data and determine the optimal plant
configuration that meets production targets. The simulator can also assist plant designers with sizing unit
operations required to achieve given circuit objectives. The present paper presents the case of the HPGR
model. The traditional approach of the population balance linked to the energy dissipation and the
residence time distribution is generalized in association with a mineral liberation model. The effect of the
specific HPGR breakage matrix and grinding function on the mineral liberation rate is discussed and
compared to the tumbling mill.

THE HPGR MODEL

The model described in this paper represents a steady state HPGR and is developed in the basis of
previous works from Morrel, Tondo & Shi (1997) and Torres & Casali (2009). These original models are
divided in three different calculation structures: throughput, power and product size distribution.
This new model introduces several improvements like different calculation modes, the calculation of
performance indexes (specific throughput, specific energy) for benchmarks purpose and the computing of
the mineral liberation to take into account the micro-fractures that occurs in the progeny particles (Morley,
2006).

Figure 1 HPGR diagram

Throughput model
The throughput of the HPGR is a function of its physical specifications and some operational variables.
The theoretical maximum throughput, (kg/s) is given by Equation 1 where (t/m3) is the density of the
compacted cake when leaving the extrusion zone, (m) the operational gap between rolls, (m) the roll
length and (m/s) the linear speed at the periphery of the roll.
= (1)

If the operating parameters are known, this equation calculates the maximum theoretical throughput of
the HPGR. When the throughput is a known parameter, it could be interesting to calculate the other
parameters like the expected product density or the operational gap . These calculation options have
been added to the algorithm.

= (2)

= (3)
Power model
The HPGR power available for breakage is the product of the average torque generated by the pressure
force with the angular speed of the rolls. The average torque is supposed to be at half the nip-angle for
inter-particular comminution . The Equation 4 gives the power (W) as a function of the operating
pressure (Pa), the roll diameter (m) and length (m), and the linear speed (m/s).

= sin
2
(4)

The material conservation in the cake gives a relationship between the nip-angle for inter-particular
comminution, the bulk density of the ore at the feed (t/m3), the density of the compacted cake (t/m3),
the gap (m) and the roll diameter (m).

1 4
cos = + + + − !
2
(5)

The breakage kinetics is directly linked to the power draw.

consumption " (J/kg).

Knowing both the throughput (kg/s) and the power (W), is it possible to calculate the specific energy

"= (6)

The specific energy consumption (Klymowsky et al., 2002) is a key operational parameter and can be used

# (kg.s/m3s), which can be used as a benchmark tool too.

to benchmark several HPGR operations. Related to this, is it possible to know the specific throughput

#= (7)

The specific throughput is independent of the HPGR size and could be used as a sizing tool (laboratory or
pilot to plant scaling) if both the ore and the roll surface don’t change (Neumann, 2006).

Particle size distribution model

The size reduction in the HPGR includes two different phases, the single particle breakage and the bed
particle breakage (inter-particular comminution). A schema of the HPGR breakage mechanisms is shown
in Figure 2.
 Figure 2 HPGR breakage stages

nip-angle for inter-particular comminution. This critical size $ (m) is given by Equation 8.
The single particle breakage appears for particles larger than the distance between rolls at the level of the

$ = + %1 − cos & (8)

Only the particles larger than the critical size are immediately and totally broken generating particles
under the critical size following the Whiten breakage model (Whiten, 1974). The product of the single
particle breakage is given by:
-

' (
= ) * + ,+( (9)
+./

The quantity of progeny material appearing in the size class 0, ' ( , is the product of the quantity of
material greater than the critical size in the size class 1, ,+( , by the individual breakage matrix * + . The

where the size class 0 is characterized by its upper size 2 .

function proposed by Austin and Luckie (Austin, Klimpel & Luckie, 1984) for the cumulative form is used

* 4 = 5 4 − 5 6/4
3 2 2
: ;
54 = Φ8 9 + 1−Φ 8 9
246/ 246/
(10)

The particle bed breakage feed , < is then calculated as the addition of the product of the single particle
breakage phase and the original feed smaller than the critical size.
The bed particle breakage mechanism has shown two different size distributions for the product
(Klymowsky et al., 2002; Daniel & Morrel, 2004), one for the center and a second one for the edges of the
roll. The edge distribution is coarser than the center distribution because the pressure distribution over
roll and the minimum in both edges. The length of the rolls is then divided into = discretized blocks
the edge of the roll has a parabola shape (Torres & Casali, 2009), with the maximum in the middle of the

containing the same quantity of material >? . For each block @, the material is supposed to travel down
with a constant speed from the beginning of the compression zone (the level A = 0 corresponding to the
angle ) to the level A ∗ of the roll axes. The residence time during this plug-flow transfer is D = A ∗ ⁄ . The

then be written in terms of level z for the F size classes of the block @:
differential equation system governing the population balance and breakage kinetics (Whiten, 1974) can

I# ? A
J/

GH =) 4? * 4 #4? A − ?# ? A
IA
(11)
4./

# ? A is the mass of material in the size class 0 at the level A. The breakage matrix * 4 is supposed to be the
same than for single particle breakage and given by Equation 10. The selection function ? is given by
Equation 12:

=
? K
?
>?
(12)

The power draw ? in the block follows the same parabolic shape as the pressure. The sum over all the
blocks gives the total power of Equation 4. The analytical form proposed by Herbst & Fuerstenau (1980) is
used for the specific selection function (Equation 13).

2R 2R
K
= K
exp 8O/ ln Q S + O/ ln Q S 9
/
2 2
(13)

MINERAL LIBERATION MODEL

In USIM PAC, the phase model describes the set of data used to characterize the material flowing through
the circuit (Brochot et al., 1995; 2006). It is based on the material properties (particle size, particle density,
chemical content) and the way to classify the material successively using these criteria (size distribution,
density distribution, assays per size class, mineral composition, or liberation data). As shown in the
previous section, the HPGR model is mainly based on the particle size distribution of the ore. This
material property is easily obtained by direct measurement. The chemical composition is directly obtained
by laboratory analysis and assaying. However, process modeling generally works on a mineral basis. The
chemical composition can be converted to the mineral composition when the transformation is possible.
The comminution, classification and separation stages process the minerals, not the chemical elements.
The selective grindability, the density dependent classification, the magnetic separation or the particle
flotation are based on the minerals. In the field of iron ore, only a description in terms of minerals allows a
predictive simulation of the grinding circuit including magnetic separation. In that case, the measurement
of magnetic iron is required to identify magnetite and other iron minerals. Due to the size effect in the
magnetic separation stage, a mineral composition per size class is necessary.

Mineral liberation data

To clarify the previous sentence, “the comminution, classification and separation stages process the
particles, not the minerals, nor the chemical elements”. Particles are generally a mixture of various
minerals. The mineral composition of the particles can be measured by image analysis, density
distribution measurement, or Davis tube tests combined with magnetic iron measurements in case of
magnetite-bearing iron ore. The composite particles can be classified regarding their mineral composition.
In the case of only two phases (e.g. magnetics and non-magnetics) the composite particle classes are
defined by a range of first phase content. The composite particle classes are called “particle types”.
Due to mineral liberation, the particle type distribution depends on particle size. The coarser particles are
mainly in the particle types around the mean mineral composition of the massive rock. The finer particles
are mainly in the liberated particle types. The transition between these two extreme distributions occurs
around the liberation size. In USIM PAC, such particle types can be defined with their mineral and
chemical compositions. The particle type distribution by size class is given for each stream, allowing the
calculation of the mineral and chemical compositions per size class and globally. Generally, particle type
distribution cannot be directly obtained from the available measurements, but necessitates a liberation
model for interpretation.

The Gaudin Random Liberation Model

Conceptually, the Gaudin Random Liberation Model (GRLM) is based on starting with an infinitely large
portion of hard rock ore, which on a micro scale consists of an ordered arrangement of uniformly sized
cubic mineral grains (Wiegel, 2006; 2011). In the simplest case, the grains are of two types, valuable
mineral and waste mineral. There is a breakage grid which is superimposed on the grains in a direction
which is parallel to the grain surfaces and which produces uniformly sized cubic particles.
From a liberation standpoint, the important size parameter is the dimensionless ratio of “effective”
mineral grain size to particle size. For a given ore, as particle size decreases, the value of this ratio
increases and the amount of liberation increases for both values and waste.
A quantitative mathematical description of particle composition changes that occur as particle size is
decreased has been developed. It has been possible to generate particle size - composition information for
the GRLM by a combination of analytical mathematics and simulation. In this solution, the product from
an individual locked composition range is distributed to only three composition ranges of a finer particle
size. A portion of the broken material stays in the same composition range; one portion goes to a lower
grade composition range and another portion enters a higher grade composition range (Wiegel 2006).
These portions are termed directional coefficients.
The application of the GRLM to the size reduction-liberation modeling of an ore requires the specification
of three liberation parameters: the “effective” mineral grain size, the valuable mineral grade of the
mineralized crude ore in the feed, and the amount of barren waste dilution, which is included with that
feed.

Coupling the liberation model with the HPGR model

This liberation model is coupled with the HPGR model to produce a comminution model with mineral
liberation. For high mineral grade ores (between 15% and 85%) such as iron ore, 12 particle types have
been defined corresponding to 10 composite particle classes with a volumetric composition of the

population balance has then to be done on particle size classes, but also on particle types. For size class 0
valuable mineral in ranges of 10%, and two liberated classes for valuable mineral and gangue. The

and particle type ', Equation (9) becomes:

- /

' ( = ) ) * 4T U 4T ,4T
(
(14)
4./ T./

And the Equation (11) becomes:

I# ? A
J/ /

GH = )) 4T? * 4T U 4T #4T? A − ?# ? A
IA
(15)
4./ T./

The liberation matrix U 4T gives the distribution over the child particle types ' for the progeny of the
parent particle type V in size class W reporting to the size class 0. These coefficients are calculated assuming
the particles are successively passing through all intermediate size classes between W and 0 following the
recursive relation:
/

U 4T = )U J/+ U J/+4T W <0−1

+./
/ (16)
)U J/T =1
./

The elements U J/+ are the directional coefficients obtained from GRLM (Wiegel, 2006).
The breakage matrix * 4T and the selection function 4T? are given by particle type to simulate the
selectivity. That is to say their function parameters ΦT , γT , βT , /T
K
, O/T and O T can be given per particle
type.

FLOWSHEET SIMULATION
This section describes an example of a size reduction-mineral liberation simulator for the grinding,
classification and concentration of a magnetite ore. The flowsheet is represented in the Figure 3.
 OS

1
HPGR feed

2 3
1
2
HPGR product 3
Feeder
HPGR Screen
4
US

Pre-concentrate

7
LIMS
4
6
Tails

Figure 3 Flowsheet configuration

The feeder unit is associated to a mathematical model which generates a particle type distribution per size
class using the GRLM. The HPGR model described above, associated with the GRLM, is used to simulate
the grinding stage. The screen model describes the partition curve in function of screen openings, the
efficiency and the short circuit of fines. The last unit is a magnetic separator and describes the recovery of
the magnetite per size class and particle type (liberation class). The phase model selected for this
simulation has the ability to describe a solid and a liquid phase. The solid phase is characterized by the
size classes, the mineral composition with magnetite and waste, the chemical composition and the
liberation classes.

100
90
80
70
Recovery, %

60 Magnetite
50 recovery
40
30 Waste
20 recovery
10
0
0.01 0.1 Particle size, mm 1 10

Figure 4 Magnetite and waste recovery vs particle size

Figure 4 shows the recovery of magnetite and waste versus the particle size. In this case the magnetite
recovery is much lower for the middle-sized particles which usually are mixed particles. At the same time,
as expected, the waste recovery becomes lower as the particle size decreases due to the higher percentage
of liberated particles. This kind of information is very useful in the optimization of a process as it show
what are the size classes with a better performance in the concentration stages.
This HPGR approach combined with the GRLM model will help in the study of one of the biggest
advantage of this technology, the effect of the micro-fractures in the concentration process. In this way,
studying the effects of these fractures per size-class and/or particle type could improve the knowledge of
the HPGR technology and its benefits.

CONCLUSION
The library of comminution models of USIM PAC software package has been enriched by an improved
model for HPGR based on the population balance approach. As this platform offers a great variety of
material description, including mineral liberation classes, it has been possible to associate the Gaudin
Random Liberation Model to the published breakage model. This coupling offers the opportunity to
optimize full grinding and separation circuits taking into account mineral liberation and to compare the
performances of a HPGR with the ones of a more traditional grinding circuits with tumbling mills.

ACKNOWLEDGEMENTS
The topic of this paper emerged from various studies conducted by Caspeo for its clients. The author
thanks the partners for supporting these works.

REFERENCES
Austin, L.G., Klimpel, R.R. & Luckie, P.T. (1984) Process engineering of size reduction: Ball milling, Society of Mining
Engineers, New York.

Barbery, G. (1991) Mineral liberation, measurement, simulation and practical use in mineral processing, Editions GB, Quebec.

Brochot, S., Durance, M.V., Fourniguet, G., Guillaneau, J.C. & Villeneuve, J. (1995) ‘Modelling of the Minerals
Diversity: a Challenge for Ore Processing Simulation’ Proceedings EUROSIM’95 Conference, pp. 861-866.

Brochot, S., Wiegel, R.L., Ersayin, S. & Touze, S. (2006) ‘Modelling and simulation of comminution circuits with USIM
PAC’, Advances in Comminution, ed. S.K. Kawatra, SME Inc., Littleton, pp 495-511.

Brochot, S. & Botané, P. (2012) ‘Modelling of crumbling and mineral liberation during crushing using “natural grain
size distribution” for friable ores’, Proceedings XXVI International Mineral Processing Congress, New Delhi,
India, pp 668-682.

Daniel, M.J. & Morrel, S. (2004) ‘HPGR model verification and scale-up’, Mineral Engineering, vol. 17, no. 11-12, pp.
1149–1161.

Gay, S.L. (1999) ‘Numerical verification of a non-preferential-breakage liberation model’, International Journal of
 Mineral Processing, vol. 57, pp. 125-134.

Gay, S.L. (2004) ‘A liberation model for comminution based on probability theory’, Minerals Engineering, vol. 17, pp.
525-534

Gupta, Y. & Yan, D.S. (2006), Mineral processing design and operation: An introduction. Elsevier Science, Amsterdam.

Khalesi, M.R., Bazin, C., Hodouin, D. & Bellec, S. (2009) ‘A grinding-liberation model for the size reduction of gold
ore’, Proceedings World Gold Conference 2009, The Southern African Institute of Mining and Metallurgy,
Johannesburg, pp 61-73.

King, R.P. (1994) ‘Comminution and liberation of minerals’, Minerals Engineering, vol. 7, pp. 129-140.

King, R.P. (2001) Modeling and simulation of mineral processing systems, Butterworth-Heinemann, Oxford.

Klymowsky, R., Patzelt, N., Knecht, J. & Burchardt, E. (2002) ‘Selection and sizing of high pressure grinding rolls’,
Mineral processing plant design, practice, and control - Proceedings, ed. A.L. Mular, D.N. Halbe and D.J. Barratt,
SME Inc., Littleton, vol. 1, pp. 636–668.

Morley, C. (2006) ‘High-pressure grinding rolls – a technology review’, Advances in Comminution, ed. S.K. Kawatra,
SME Inc., Littleton, pp 15-39.

Neumann, E.W. (2006) ‘Some basics on high-pressure grinding rolls’, Advances in Comminution, ed. S.K. Kawatra, SME
Inc., Littleton, pp 41-49.

Stamboliadis, E.T. (2008) ‘The evolution of a mineral liberation model by the repetition of a simple random breakage
pattern’, Minerals Engineering, vol. 21, pp. 213-223.

Torres, M. & Casali, A. (2009) ‘A novel approach for the modelling of high-pressure grinding rolls’, Mineral
Engineering, vol. 22, no. 13, pp. 1137–1146.

Whiten, W.J. (1974) ‘A matrix theory of comminution machines’, Chemical Engineering Science, vol. 29, pp. 589–599.

Wiegel, R.L. (2006) ‘The rationale behind the development of one model describing the size reduction / liberation of
ores’, Advances in Comminution, ed. S.K. Kawatra, SME Inc., Littleton, pp 225-241.

Wiegel, R.L. (2011) ‘The solitary grain liberation model: a description of a mineral liberation model for low-grade
ores’, Minerals and Metallurgical Processing, vol. 28, pp. 159-168.