Powder Technology 383 (2021) 244–255

Contents lists available at ScienceDirect

Powder Technology

journal homepage: www.elsevier.com/locate/powtec

Modeling comminution of iron ore concentrates in industrial-scale HPGR

Túlio M. Campos a,⁎, Gilvandro Bueno a,b, Luís Marcelo Tavares a
a
Department of Metallurgical and Materials Engineering, Universidade Federal do Rio de Janeiro – COPPE/UFRJ, Cx. Postal 68505, CEP 21941-972 Rio de Janeiro, RJ, Brazil
b
Vale S.A., Complexo de Tubarão, Vitória, ES, Brazil

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

Article history: High-pressure grinding rolls have an important role in several iron ore pellet feed preparation circuits. Modeling
Received 26 June 2020 and simulation are potentially useful tools to improve their performance, but the applicability of mathematical
Received in revised form 30 November 2020 models from the literature in this particular application has not yet been demonstrated, although the model by
Accepted 1 December 2020
Torres and Casali has been identified as having good potential. The present work relies on several surveys on
Available online 24 January 2021
four HPGRs with variable diameters (1.4 to 2.25 m) and aspect ratios (0.88 to 1.45) operating in pelletizing plants
Keywords:
to test the model originally proposed by Torres and Casali, as well as modifications proposed to it.
Roller press Inadequacies have been found in the original model, which gave poor predictions of power, throughput and
Modeling product size distribution along the roll's length. Modification of the power equation, incorporating a machine-
Simulation dependent correction factor for the nip angle, as well as several modifications to the throughput model, including
High-pressure grinding an expression meant to account for the ejection of material from the edge of the rolls and extrusion, resulted in
Iron ore very good agreement between data from surveys and the model. Finally, incorporation of an empirical correction
equation to account for saturation and a more general expression for the roll profile led to good agreement be-
tween measured and predicted product size distributions, including their variation along the roll's length.
© 2021 Elsevier B.V. All rights reserved.

1. Introduction available to describe accurately the performance of industrial-scale

Since its first application by the Swedish company LKAB in 1993 [1],
the High-Pressure Grinding Rolls (HPGR) has reached an outstanding
position in pellet feed preparation prior to green ball formation. Vale
S.A. was responsible for one of the main successful applications that
followed, when it introduced the technology to some of its pelletizing
plants of Complexo de Tubarão (Vitória, Brazil), being one of the pio-
neers in the application of HPGRs integrated with ball milling [1].
The success from the introduction of HPGRs in iron ore pelletizing
processes can be attributed to a number of reasons. Some authors
have attributed it to its ability to reduce the specific energy consump-
tion in size reduction [1–3], its ability to preferentially generate ultra-
fines, with their high associated specific surface area [3,4] as well as
their ability to operate with feeds containing high moisture contents
[3]. Application of HPGRs in this task is normally associated to ball mill-
ing, either as pre-grinding or as regrinding, although alternative circuit
configurations have also demonstrated potential of application [3,5].
In spite of its success in the pelletizing industry, some challenges still
remain, including the need for enabling additional improvements
within the HPGR iron ore pellet feed pressing [1,3,6], improving scale-
up [7], and analyzing the potential of applying HPGRs in multiple stages
[3]. All these challenges can be met if mathematical expressions become
⁎ Corresponding author.
E-mail address: tulio_uca2013@poli.ufrj.br (T.M. Campos).
HPGRs in iron ore pellet feed pressing.
Mathematical models of the HPGR have been developed in parallel
to the HPGR technology and the machine improvements along the
years [8]. Such models have been proposed primarily to predict power
consumption, throughput and product size distribution during opera-
tion. These models establish relationships between machine settings,
operating conditions and material characteristics to predict the HPGR
performance. Indeed, in the last 30 years or so several mathematical
models [9–17] have been proposed to predict the HPGR performance
and describe particle breakage behavior in a confined particle-bed.
Among these models, three of them [14,15,17] reached popularity in ap-
plication in the minerals industry, being incorporated in commercial
mineral processing plant simulators. However, in spite of their rela-
tively widespread use in the minerals industry, these models still con-
tain limitations in describing the HPGR performance. For instance,
models from Morrell et al. [14] and Dundar et al. [17] are unable to pre-
dict the product size distribution along the rolls length, being unable to
discriminate between centre and edge products. On the other hand, the
model proposed by Dundar et al. [17] does not contain an expression to
predict the HPGR power consumption. Among them, the model pro-
posed by Torres and Casali [15] has the advantage of allowing a detailed
description of the particle breakage along the rolls length, besides pro-
viding physical description of all HPGR performance variables.
In the particular case of pressing iron ore concentrates, Campos et al.
[5] reviewed these models and recognized that the one proposed by

https://doi.org/10.1016/j.powtec.2020.12.075
0032-5910/© 2021 Elsevier B.V. All rights reserved.
T.M. Campos, G. Bueno and L.M. Tavares Powder Technology 383 (2021) 244–255

List of symbols γ breakage function parameter

δ percentage of the feed ejected from the edge of rolls
Aij, k matrix used in the solution of model equations of popu- η breakage function parameter
lation balance model κ nip angle parameter
bij breakage function in distributed form λ fraction of roll surface area corresponding to the
Bij cumulative breakage distribution function autogenous layer
D roll diameter μ shape parameter of the axial roll profile equation model
Ei specific input energy τ throughput parameter
E′ material parameter υ throughput parameter
fobj objective function value φ throughput parameter
Fm compressive force ω breakage function parameter
G BSA model parameter ϕ breakage function parameter
Hk holdup in each block k
J BSA model parameter
L roll length
Torres and Casali [15] described well the performance of a pilot-scale
n number of periods in the Fourier Transform
HPGR, but only after modifications were made to some of the model
N number of size classes
equations. Besides not yet having been tested to industrial-scale HPGR
NB number of blocks
operation with iron ore pellet feed, the Torres and Casali model modi-
P power consumption
fied by the authors [5] has additional shortcomings that would likely
Pk′ axial normalized pressure profile
limit its successful application at this stage, including the assumption
Pk axial power profile
of axial parabolic pressure profile along the rolls and lack of an account
pm operating pressure
for breakage saturation.
Q throughput
The axial pressure profile along the HPGR rolls is associated to the
sEi specific breakage rate
well-known edge effect, which has been widely observed by several au-
sE1 specific breakage rate parameter
thors over the last 30 years of application of the technology [6,14,18–20].
Si, k breakage rate for each class i in each block k
Selected models that describe the HPGR performance account for this ef-
SSA equivalent spherical specific surface area
fect in size reduction when considering different applied forces in the
U roll peripheral velocity
particle bed along the roll's length [14,15]. Although the parabolic profile
U0 initial sheet velocity
proposed by Torres and Casali [15] has demonstrated good ability to pre-
Uf final sheet velocity
dict breakage along the axial roll position [5], some authors argue that
Ug material velocity
variations will appear in the actual profile due to roll wear [21,22], as
Umax maximum roll peripheral velocity achieved in the HPGR
well as variations in moisture content and roll aspect ratio [6].
vz constant velocity in the z direction
Moreover, one phenomenon that can occur both in confined bed
xi particle size
breakage and HPGRs is breakage saturation. As pressures increase be-
xi harmonic mean size
yond a certain level, the bed of particles becomes progressively less
b
xi representative size
able to dissipate the strain energy in generating new fragment surfaces,
wi fraction retained in each size class in BSA model
with a larger fraction of the strain energy of the bed either being stored
wi, k fraction retained in each size class i and block k
as elastic energy or being dissipated by friction and plastic deformation.
WHPGR
calc calculated cumulative passing
All of these contribute to a reduction in energy efficiency under these
WHPGR
exp experimental cumulative passing
conditions [23]. From experiments in a piston-and-die apparatus, Liu
yk normalized position of each block k
and Schönert [24] showed that breakage saturation increases with the
yk normalized average position of each block k
input energy in the particle bed, whereas other authors [25,26] showed
yk′ axial position of each block k in respect to the center of
the increase in elastic deformation of the particle bed with the applica-
the rolls
tion of higher vertical stresses. Recently, Campos et al. [23] presented a
z∗ distance between the interparticle breakage zone en-
detailed study assessing the breakage response of fine iron ore concen-
trance and the HPGR extrusion zone
trates in a piston-and-die apparatus, demonstrating the appearance of a
difference in breakage response as a function of particle size in the bed.
Greeks letters
The study also proposed an expression to characterize the propensity of
Δf final thickness
a material to break under confined conditions, which showed potential
Δ0 initial thickness
to be applied describing a relationship between the input energy and
αip nip angle
the generation of new surface area [23]. This effect of saturation, how-
ξ1 specific breakage rate parameter
ever, is yet to be incorporated in mathematical models of the HPGR.
ξ2 specific breakage rate parameter
Relying on experiments on four industrial-scale HPGRs with differ-
ρa bulk density
ent diameters and aspect ratios, the present work initially analyses the
ρg flake density
application of the Torres and Casali model to describe roller pressing
ρsp specific gravity
of iron ore pellet feed fines. Afterwards, modifications to the model pro-
χc critical size
posed in an earlier work [5], as well as additional ones, are used in com-
χg operating gap
parison to the industrial-scale data.
χgstud operating gap measured from the distance between the
top of the rolls studs
2. Background
χp stud penetration in the particle bed
Λ efficiency in creating new surface model parameter
2.1. Torres and Casali model
Ψ efficiency in creating new surfaces
β breakage function parameter
Torres and Casali [15] proposed a phenomenological approach to de-
scribe the HPGR performance as a function of operating conditions,

245
T.M. Campos, G. Bueno and L.M. Tavares Powder Technology 383 (2021) 244–255

machine design and material characteristics. The model contains ex-

Pk E
pressions to predict the three main HPGR performance variables: Si,k ¼ s ð8Þ
throughput, power consumption and product size distribution, and is Hk i
briefly presented as follows.
where Si,k is the breakage rate for each block and size class, Pk is the par-
The HPGR throughput is based in the plug flow model and can be de-
abolic power profile given from Eq. (9) and Hk is the holdup of each
scribed by [27]:
block [15]:
Q ¼ ULχ g ρg ð1Þ  
α  L2 −4y0k 2
ip
P k ¼ 2F m sin U N   ð9Þ
where Q is the throughput, U is the peripheral roll velocity, L is the roll 2 B
∑ L2 −4y0j 2
length, χg is the operating gap and ρg is the flake density. j¼1
Torres and Casali [15] used an approach based on the compressive
force applied to the particle bed, the torque on both rolls and the angu- where yk′ is the axial position of each block k in respect to the center of
lar velocity to predict the HPGR operating power P: the rolls [15].
The size distribution of the HPGR product for each block is given by:
α 
ip
P ¼ 2F m sin U ð2Þ i
 
2 S j,k ⁎
wi,k ¼ ∑ Aij,k exp − z ð10Þ
j¼1 vz
where αip is the nip angle and Fm is the compressive force, calculated
from: where, z ∗ is the distance between the extrusion zone pressing and the
beginning of the interparticle compression zone [15] and Aij,k is Reid's
D
F m ¼ pm L ð3Þ matrix [30] used in the solution of the population balance equation.
2
2.2. Modified Torres and Casali model
where pm is the operating pressure and D is the diameter of the rolls. The
model also requires the nip angle, which varies according to the operat-
The Torres and Casali model is robust and has demonstrated to pro-
ing conditions and may be calculated from [10]:
vide a valid description of the HPGR performance in size reduction of
" sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi# copper ore [15] also has been successfully used as part of a model pre-
1    2 4χ ρ D
g g dictive control system [31]. Nevertheless, recent studies by the authors
cos α ip ¼ χg þ D þ χg þ D − ð4Þ
2D ρa have identified limitations when predicting iron ore pellet feed pressing
[5]. With the aim of addressing some of these limitations, Campos et al.
where ρa is the bulk density. [5] proposed and validated several modifications to improve the model
Size reduction in Torres and Casali model [15] is described on the prediction in a pilot-scale HPGR, reviewed briefly as follows.
basis of the population balance model. As already proposed by other au- As observed by Campos et al. [5] and already reported by other au-
thors [14], this model also allows discriminating between center and thors [12,13,20,32] a measurable part of the feed material may be
edge products, since it discretizes the roll in Nb blocks along the axial ejected between the edges of the rolls and the cheek plates during
roll position [15]. The pressure profile along the rolls length is assumed HPGR operation. With this in mind, as well as other inadequacies iden-
to have parabolic shape, which has been considered an accurate de- tified, Campos et al. [5] proposed an empirical modification to the com-
scription in earlier studies [14,18]. monly used (Eq. 1) throughput model as:
As such, assuming steady-state conditions and also a plug flow with  
100
a constant speed vz in the coordinate z (vertical), the authors used the Q ¼ ULχ g ρg ð11Þ
population balance model to calculate the HPGR product, taking into ac- 100−δ
count the different breakage rates Si,k for each particle size class i and
where δ is a parameter that estimates the percentage of material from
block k:
the feed that is ejected along the edge of the rolls, being given by [5]:
i−1    
vz
d
w ðzÞ ¼ ∑ S j,k bij w j,k ðzÞ−Si,k wi,k ðzÞ ð5Þ δ χg U τ
dz i,k ln ¼ −υ ð12Þ
j¼1 φ D U max

In order to solve the system of differential equations, the authors used where Umax is the maximum peripheral roll velocity in operation for the
the equations for the cumulative breakage function [28] (Eq. 6) and HPGR in question, and φ, υ and τ are dimensionless parameters that
the specific selection function [29] (Eq. 7), given by: should be fitted from data. An additional modification introduced was
in the calculation of the operating gap, which has been proposed to be
 γ  β
xi x estimated from the distance between the top of the studs on the rollers,
Bij ¼ ϕ þ ð1−ϕÞ i ð6Þ
xj xj added to the depth of penetration of the studs in the particle bed [13],
giving:
      2
ln sEi =sE1 ¼ ξ1 ln b
xi =b
x1 þ ξ2 ln bxi =b
x1 ð7Þ χ g ¼ χ gstud þ 2λχ p ð13Þ

where ϕ, γ, and β are breakage function parameters and sE1, ξ1 and ξ2 are
selection function parameters that must be fitted to data, with bij =
Bi−1, j − Bi, j. b
xi is the representative size of particles contained in size
pffiffiffiffiffiffiffiffiffiffiffiffiffi
range and given by b xi ¼ xi xi−1 .
Moreover, the authors also considered breakage rates that vary
along the axial roll position according to the parabolic assumed power
profile and a constant holdup for each block. The breakage rates are
then given by:
where χgstud is the distance between the rollers measured from the top
of the studs, χp is the stud penetration in the particle bed and λ is the
fraction of roll surface area corresponding to the autogenous layer. For
the pilot-scale HPGR work the optimal parameter values were
φ = 100, υ = 140, τ = 0.1 and λ = 0.65 [5].
In analogy to the throughput model, Campos et al. [5] also reported
poor agreement between the measured and the predicted power con-
sumption in a pilot-scale HPGR pressing iron ore pellet feed. Indeed, it

246
T.M. Campos, G. Bueno and L.M. Tavares Powder Technology 383 (2021) 244–255

is important to recognize that some factors not accounted properly in

the power consumption model (Eq. 2) may have been responsible for
limiting its predictive capabilities. Indeed, the nip angle, estimated
using Eq. (4), has been known to be also influenced by other variables,
including feed size distribution, moisture content and rheological prop-
erties of the material [19,33,34], which are not explicitly accounted for
in Eq. (4). As such, the introduction of a fitting parameter κ, given by
the ratio between the back-calculated nip angle from a reference test
and the calculated nip angle (Eq. 4), has been proposed [5]. As such,
Eq. (2) may be simply rewritten as:
 
κα ip
P ¼ 2F m sin U ð14Þ
2

In the case of the pilot-scale HPGR pressing iron ore pellet feed, the
value of κ was estimated as 2.8, suggesting that Eq. (4) systematically
underestimated the nip angle in this operation [5].
In the case of the product size distribution, Campos et al. [5] intro-
duced a relatively minor modification to the Torres and Casali model,
by replacing the standard size-normalizable breakage function equation
(Eq. 6) by a non-normalizable breakage function when the particle size Fig. 1. Mean size distributions of the feed to the various HPGRs studied.
xi is lower than a critical size ω, given by [35]:

x η x γ  β
x
i i
Bij ¼ ϕ þ ð1−ϕÞ i for xi <ω
ω xj xj
where η and ω are additional fitting parameters.
3. Materials and methods
3.1. HPGR industrial-scale tests
High-pressure grinding rolls operating in different pelletizing plants
at the Complexo de Tubarão from company Vale (Vitória, Brazil) were
the object of the present study. The pelletizing plants are fed with a
blend of iron ore concentrates from different mines, being composed
predominantly by iron oxides, mainly hematite, with minor amounts
of quartz [36]. A total of four experimental campaigns were conducted
in four different industrial-scale machines (#1 to #4) to assess the per-
formance of HPGRs in pellet feed pressing. Two different circuit designs
were surveyed: roller presses #1 to #3 operated after ball milling, in
regrinding pre-pelletizing process, whereas roller press #4 corresponded
to an HPGR in a pre-grinding operation, that is, prior to ball milling.
Moreover, campaigns were conducted with a moisture content of 8 ±
0.5%, which is recognized to be close to the maximum value that is toler-
ated in HPGR operation involving iron ore pellet feed [1].
Among the HPGRs, three of them (#1 to #3) were manufactured by
Thyssenkrupp Polysius and the remaining one (#4) was manufactured
by KHD/WEIR. All machines were connected to a supervisory system,
which allowed capturing information on roll peripheral velocity, oper-
ð15Þ
ating gap, operating pressure, power consumption and throughput.
Care was taken to guarantee that the cheek plates were closely adjusted,
with a gap smaller than 1.5 mm in respect to the rolls, and the autoge-
nous main frame and wear parts were in very good condition, with no
noticeable signs of wear. Process information from each survey was
gathered when the supervisory system indicated that the HPGR was op-
erated under steady-state conditions.
Table 1 summarizes the main operating conditions as well as the
main material characteristics for the experimental campaigns on each
of the HPGRs, while Fig. 1 compares the mean size distributions of
each feed. The coarser feed size corresponding to the HPGR #4 is consis-
tent with its pre-grinding role in the circuit. The feed size distributions
of the feed to HPGRs #1 and #2 are finer and consistent with the
regrinding pre-pelletizing role of the machines downstream from ball
mills operating in closed-circuit. Although with the same regrinding
role, the HPGR #3 is fed with material with slightly coarser top size,
since it operates downstream from ball mills that operate in open cir-
cuit. A snapshot of HPGR #2 in operation is presented in Fig. 2.
The specific gravity of the feed in the various campaigns was mea-
sured using a Helium pycnometer and is given as 4.9 t/m3. The bulk den-
sity was obtained from the ratio between the sample mass and a known
volume and is given by 3.0 t/m3. Finally, the flake density was measured

Table 1
Summary of the main characteristics of the HPGR as well as of the feed material in the var-
ious surveys.

HPGR

#1 #2 #3 #4

Roll diameter (m) 1.7 2.25 2.0 1.4

Roll length (m) 1.4 1.55 1.5 1.6
Aspect ratio (−) 1.21 1.45 1.33 0.88
Specific force (N/mm2) 0.5–2.1 0.5–1.5 1.5–1.7 2.0–2.5
Operating pressure (bar) 40–100 25–75 60–80 50–65
Roll velocity (m/s) 0.5–1.7 0.8–1.0 1.5–1.9 1.3
Operating gap (mm) 9–22 9–12 9–12 6
Maximum roll velocity (m/s) 1.83 2.01 2.19 1.95
Feed 80% passing size (μm) 70 52 84 143
Feed % passing 45 μm 62 73 55 26
Feed BSA (cm2/g) 1750 ± 10 1700 ± 10 1650 ± 10 470 ± 10
Fig. 2. HPGR #2 in operation.

247
T.M. Campos, G. Bueno and L.M. Tavares Powder Technology 383 (2021) 244–255

using preserved flakes from one operation, and the average value was Table 2
3.5 t/m3, which was used throughout the simulations. Summary of the base case conditions selected to fit parameter κ, as well as its optimal
value (Eq. 14) for each HPGR.
During the course of the surveys, samples were collected along the
axial roll position during the HPGR discharge in order to assess the var- HPGR
iation in size reduction along the roll length. Samples were collected in #1 #2 #3 #4
seven different points using a manual sample cutter introduced under-
Operating pressure (bar) 41 59 82 48
neath the rolls, which was quickly removed. Immediately after that, the Roll velocity (m/s) 0.78 0.83 1.54 1.30
HPGR operation was interrupted and samples from the feed and the Measured power (kW) 525 897 2325 814
product conveyors were collected from belt cuts. κ (−) 2.0 2.5 2.6 2.6
All samples were carefully quartered, dried, and subjected to wet
size analysis by laser scattering in a Malvern Mastersizer 2000 (Malvern
Instruments Inc) and Blaine specific surface area (BSA) in a PCBlaine-
Star (Zünderwerke Ernst Brün GmbH).

3.2. Model parameter estimation

The modified Torres and Casali model [5] was implemented in

Matlab® (version R2018a, Mathworks Inc.). The simultaneous estima-
tion of parameters of the breakage and selection functions (Eqs. 7 and
15) was carried out using the nonlinear optimization method available
in Matlab®, called fminsearch. It uses a direct search algorithm and its
task is finding the minimum of a scalar function of several variables
from an initial estimate. The objective function consisted of the sum of
the differences in the logarithms of the experimental and the fitted
values of the particle size distribution of a reference test (Base Case) in
cumulative form using the least squares method:

N h    i2
f obj ¼ ∑ log W HPGR HPGR
calc ðiÞ − log W exp ðiÞ ð16Þ
i¼1

where N is the number of size classes, WHPGR HPGR

calc (i) and Wexp (i) are, re-
spectively, the calculated cumulative passing and experimental cumula- Fig. 4. Relationship between specific throughput and specific force for tests in all
industrial-scale HPGRs. Results are discriminated by machine.
tive passing at size class i.

4. Results and discussions

4.1. Power consumption
Fig. 3 presents a comparison between measured and predicted
power consumptions using the original Torres and Casali model (a) as
well as the modified Torres and Casali model (b) for all HPGR experi-
ments. As shown in the previous study that dealt with pressing iron
ore concentrate in a pilot-scale HPGR [5], the original Torres and Casali
model underestimated in about 60% the power consumption in
industrial-scale HPGRs (Fig. 3a). On the other hand, Fig. 3b showed
that the modified Torres and Casali model (section 2.2) was able to pro-
vide predictions which agreed very well to experiments. This resulted in
fitting the machine-specific constant κ in Eq. (14) to each individual
HPGR (Table 2).

Fig. 3. Comparison of experimental and predicted powers calculated with the Torres and Casali model (a) and the modified Torres and Casali model (b) for all experimental surveys,
discriminated by machine.

248
T.M. Campos, G. Bueno and L.M. Tavares Powder Technology 383 (2021) 244–255

Fig. 5. Comparison between experimental and calculated throughputs using the original Torres and Casali model (a) and the model modified in the present work (b) for all industrial-scale
HPGRs.

4.2. Throughput
Uρa χ c
Ug ¼ ð18Þ
Experimental results are shown in Fig. 4 that illustrate the relation- ρg χ g
ship between the specific throughput and the specific operating force
for all surveys in the different HPGRs. As already observed by Van der where Ug is the material velocity and χc is the critical aperture calcu-
Meer and Leite [37] and confirmed by Campos et al. [6] for pressing lated from [15]:
iron ore concentrates in pilot-scale HPGR in pre-grinding operation, an  
χ c ¼ χ g þ D 1− cos α ip ð19Þ
inverse relationship appears between the specific throughput and the
specific force. Campos et al. [6] suggested that such variation is a conse-
where χg is the operating gap from Eq. (13).
quence of the variation in operating gap. However, different from the
Incorporating the modifications associated to the increase in mate-
results presented by Campos et al. [6] for the pilot-scale HPGR, Fig. 4
rial velocity along the extrusion zone, besides the ejection of material
shows that no single simple relationship exists that is valid for the dif-
between the edge of the rolls and the cheek plates, parameters of
ferent industrial-scale HPGRs surveyed.
Eq. (12) can now be fitted from data. Fig. 6 shows the reasonable agree-
With the aim of assessing the applicability of the Torres and Casali
ment between model fit (line) and measured values for the different
[15] model to the industrial survey data, predictions from Eq. (1) are
HPGRs over the entire range of operating operating conditions sur-
compared to experiments in Fig. 5a, which shows very poor agreement
veyed. As such, the model suggests that between about 10 and 35% of
between model and experiments, as already observed in a previous
study involving the pilot-scale HPGR pressing iron ore pellet feed [5].
However, the application of the modified version of the model (omitted
for brevity) only improved marginally the predictions, demonstrating
the need for further changes in the Torres and Casali model, beyond
those already presented in section 2.2.
One aspect that was neither included in the original model nor in its
modified version is the acceleration of the material along the compres-
sion zone, recognized by some authors [12,13,32] to be important. This
effect has been incorporated in the model by using an analogy between
the HPGR and a lamination process used in metallurgy. Indeed, a mass
balance of a lamination process may be described as [38]:

LΔ0 ρsp U 0 ¼ LΔf ρsp U f ð17Þ

where the terms on the left of the equal sign refer to the material be-
fore the lamination and those on the right refer to the material after
the process, in which L is the roll length, Δ0 is the initial thickness,
ρsp is the specific gravity, U0 is the initial sheet velocity, Δf is the final
thickness and Uf is the final sheet velocity. In order to keep the mass
balanced on both sides of the equation, Dieter and Bacon [38] sug-
gested that Uf > U0. The present work uses Eq. (18) to calculate the
material velocity as a constant value along the extrusion zone in the
HPGR. For pressing in a confined bed an additional modification must
be incorporated by using the packing density before (bulk density) Fig. 6. Relationship between the product of the dimensionless gap and the dimensionless
and after the operation (flake density). As such, rearranging Eq. (17) rolls speed raised to parameter τ and the estimate of the proportion of material ejected
gives: from the edges of the rolls and the cheek plates. Data identified by machine.

249
T.M. Campos, G. Bueno and L.M. Tavares
Fig. 7. Feed and product size distributions from surveys with HPGR #1 discriminated by
specific force (and energy) at a constant throughput of 400 t/h.
material would have been ejected from the edge of the rolls in opera-
tion, which is in general agreement with estimates by Lim et al. [12]
and Van der Ende et al. [20]. The optimal values of the parameters,
valid for all industrial-scale HPGRs, were τ equal to 0.1, υ equal to 190
and φ equal to 100, in which only υ needed to be refitted from the esti-
mates obtained from the pilot-scale study [5].
A comparison between the survey results and the improved model
fit is presented in Fig. 5b, which demonstrates the very good agree-
ment between them. As such, although Fig. 4 did not show a straight-
forward relationship between the operating conditions and the
measured specific throughput, the model was able to incorporate
these effects.
4.3. Product size distribution
Typical size reduction results from the industrial surveys are pre-
sented in Fig. 7, which shows the significant effect of specific compres-
sive force on the product size distribution, demonstrating the
importance of this variable in HPGR performance.
In order to use the modified Torres and Casali model for describing
the breakage response from the surveys (Table 1), the present work as-
sumes validity of the non-normalized breakage function (Eq. 15) pa-
rameters fitted by Campos et al. [5] from pressing iron ore
concentrates in the pilot-scale HPGR. Besides these, two of the parame-
ters in the selection function (Eq. 7) were also assumed to be constant
for all industrial-scale HPGRs. These values are listed in Table 3. The re-
maining parameter sE1 in Eq. (7) has been fitted on the basis of a base
case for each HPGR, selected between those in which the specific energy
is intermediate to low.
Fig. 8 presents the fitted breakage function [5], which shows the
point of inflexion at 13 μm, as well as the specific selection functions
fitted for each HPGR. Results from Fig. 8b are able to highlight the rela-
tively higher specific breakage rates for tests performed in the HPGR #1
(Pre-grinding), showing smaller variations among the other surveys.
In order to assess the model predictive capabilities to deal with
different operating conditions, Fig. 9 compares experimental and simu-
lated product size distributions from surveys with HPGRs #1 and #4.
These results demonstrate the very good ability of the model in
predicting the product size distribution both when the HPGR is fed
with a coarser feed size distribution, such as in a pre-grinding operation
(Fig. 9b), and a finer feed size distribution in the regrinding pre-
pelletizing process (Fig. 9a).
250
Powder Technology 383 (2021) 244–255
Table 3
Summary of the fitted parameters for all HPGRs.
Parameter Values
Selection function ξ1 −0.60
ξ2 −0.16
Breakage function γ 0.89
β 5.46
ϕ 0.86
η 0.46
ω 0.013
Number of blocks Nb 50
Very good agreement in both the coarser and the finer parts of
the size distribution are required in pelletizing, since both extremes
have a significant impact on iron ore pellet characteristics [39]. A
well-known way to capture these features of the size distribution
in the iron ore pelletizing industry is through measurement of the
amount of material passing the 45 μm sieve and the Blaine specific
surface area [39]. A model, presented in the appendix, has been
used to estimate the BSA from the predicted size distributions [5].
As such, Fig. 10 compares measured and predicted values of the
BSA (a) and of the percent passing the 45 μm sieve (b) in the prod-
uct for all surveys with HPGRs #1 to #3. The results demonstrate
good agreement for surveys carried out with specific compressive
forces below about 2 N/mm2. Indeed, when higher specific forces
and specific energies were used, the model tended to overestimate
the product fineness.
These results suggest the appearance of a well-known phenome-
non in confined breakage that is associated to high compressive
forces, in which the energy input to the bed is no longer effectively
used in generating new surfaces [23–25,40]. Such saturation is influ-
enced by a number of variables, but can be studied under controlled
conditions in a piston-and-die apparatus. In order to illustrate such
saturation, the concept of energy utilization, defined as the ratio be-
tween the increase in surface area of the material and the input en-
ergy [23], has been used. Results from testing particles, contained
both in narrow size ranges and as complete size distributions in a
piston-and-die apparatus and carried out by Campos et al. [23] are
presented in Fig. 11. It shows that energy utilization is maximum at
the lowest input specific energies studied, which was just high
enough to surpass initial bed rearrangement [23]. Defining an effi-
ciency variable as the ratio between the actual energy utilization
reached at a given input specific energy and the maximum energy uti-
lization, results are given in Fig. 12, which have been fit to the empir-
ical equation:
"   #
Ei Λ
ΨðEi Þ ¼ exp − ð20Þ
E0
where Ψ is the proposed efficiency variable, Ei is the input energy, E' is a
material parameter, given in kWh/t, and Λ is a dimensionless fitting pa-
rameter. Results from Fig. 12 show Eq. (20) can describe the data rea-
sonably well for pressing iron ore concentrates in a piston-and-die
test. Moreover, a recent work by the authors showed the great similarity
in the energy utilization for pressing different iron ore concentrates in a
piston-and-die apparatus and a lab-scale HPGR, thus justifying the ap-
plication of these results to calibrate the model for industrial-scale
HPGR operations [36]. Nevertheless, Campos et al. [36] also showed that
the energy utilization varies with particle size and size distribution. This
suggests that caution should be exercised in use of Eq. (20) to simulate
changes in feed size distribution.
As such, it is hereby proposed that Eq. (8) of the Torres and Casali
model is modified in order to account for the breakage saturation, by
T.M. Campos, G. Bueno and L.M. Tavares Powder Technology 383 (2021) 244–255

Fig. 8. Cumulative breakage function for selected parent particle size classes (a) and specific selection functions (b) fitted for all HPGRs using sE1 = 0.20 t/kWh (#1), 0.26 t/kWh (#2), 0.15 t/
kWh (#3) and 0.40 t/kWh (#4).

applying the constant Ψ to reduce the breakage rates at each block and length. Results are presented in Fig. 14, showing the significant variation
size class, giving: of both BSA (Fig. 14a) and percent passing the 45 μm sieve (Fig. 14b)
  along the length of the rolls. As already reported by Campos et al. [6]
Pk P for pressing iron ore pellet feed in a pilot-scale HPGR and suggested
Si,k ¼ sEi Ψ ð21Þ
Hk Q by other authors [14,18], it shows the edge effect, with a coarser product
formed close to the edges of the rollers and a finer product being gener-
Fig. 13 then compares experimental and predicted values of BSA (a) and ated in the central zone. However, Fig. 14 does not show the parabolic
percentage passing the 45 μm sieve in the product of tests carried out profile suggested in Eq. (9) as precluded by the Torres and Casali
with HPGRs #1 to #3 after applying Eq. (21). Results demonstrate the model, since a region of approximately constant fineness appears in
very good description of both measures using the proposed model mod- the central part of the rolls. The figure also shows that in the edges
ification, which yielded good predictions even at high compressive nearly no size reduction occurs, since both BSA and the proportion pass-
forces. ing the 45 μm sieve in the product are only marginally higher than in
the feed.
4.4. Axial variation of product size distribution along the rolls´ length As such, an alternative expression to Eq. (9) has been sought to de-
scribe such observed profile. It is proposed that the axial normalized
In selected surveys, samples from underneath the rolls were col- pressure profile (Pk′) can be described as a periodic function based on
lected, so as to map the axial variation of the product along the rolls the Fourier Transform:

Fig. 9. Comparison of experimental and predicted product size distributions in regrinding pre-pelletizing at 700 t/h and specific force of 0.9 N/mm2 (HPGR #1) (a) and pre-grinding at
500 t/h and specific force of 2.5 N/mm2 (HPGR #4) (b).

251
T.M. Campos, G. Bueno and L.M. Tavares Powder Technology 383 (2021) 244–255

Fig. 10. Comparison between measured and predicted BSA (a) and percent passing the 45 μm sieve (b) in the product from surveys with HPGRs #1 to #3 using the modified Torres and
Casali model presented in section 2.

Fig. 11. Variation of the energy utilization as a function of specific input energy for pressing
Fig. 12. Comparison between experimental and fitted values for the efficiency in creating
different iron ore concentrates from a narrow feed size range of 150–125 μm (A-C) and
new surface for different iron ore concentrates from a narrow feed size range of
complete feed size distribution (D) in a piston-and-die apparatus.
150–125 μm (A-C) and complete feed size distribution (D) in a piston-and-die
apparatus, with E´ = 9.8 kWh/t and Λ = 0.81.
4 100 1− cos nπ −μ ðn2 π2 Þ
P 0k ¼ ∑ e sin nπyk ð22Þ
π n¼1 2n

where μ is a single fitting parameter that can be selected to match the Fig. 15 compares experimental and predicted values of BSA (a) and per-
profile shape, varying in the range of 0.001 to 0.1. yk is the normalized centage passing the 45 μm sieve in the product using Eq. (24), compar-
average position along the roll length given by yk ¼ yk−12þyk , in which yk ing to results obtained using the parabolic relationship proposed by
is given by: Torres and Casali (Eq. 9). It clearly shows that the new proposed ex-
k pression is able to more closely describe the shape of the profile ob-
yk ¼ ð23Þ served in the plant survey. Minor deviations were found in Fig. 15b
Nb
with Eq. (24) overestimating the measured percent passing in 45 μm,
where Nb is the number of blocks. Therefore, the power profile along even though still showing a relatively good agreement. These results
the axial roll position, previously described in Eq. (9), would be replaced showed that the value of μ of about 0.01 is able to describe the approx-
by: imately trapezoidal profile observed in Fig. 15. It is important to
  emphasize that when μ is approximately equal to 0.1, Eq. (22) ap-
κα ip P0 proaches the parabolic profile given by Eq. (9), thus demonstrating its
P k ¼ 2F m sin U N k ð24Þ
2 B wide applicability.
∑ P 0j
j¼1

252
T.M. Campos, G. Bueno and L.M. Tavares Powder Technology 383 (2021) 244–255

Fig. 13. Comparison between measured and predicted BSA (a) and percentage passing the 45 μm sieve (b) in the product from surveys on HPGRs #1 to #3 with different specific
compressive forces using the modified Torres and Casali model with the description of the breakage saturation effect given by Eq. (20).

Fig. 14. Variation of BSA (a) and percentage passing the 45 μm sieve (b) along the axial roll position for surveys in HPGR #1 with a constant throughput of 400 t/h at different specific forces.

5. Conclusions
Direct application of the model proposed by Torres and Casali
[15] to data from surveys of four HPGRs operating in pelletizing
plants led to poor predictions of power, throughput, product size
distributions at high specific forces and product fineness along
the roll's length.
Modification of the equation to predict operating power by incorpo-
rating a machine-dependent constant to be fitted from data allowed
matching the predicted power with the values collected in the plant
surveys. Modification of the throughput model by incorporating both
extrusion of material between the rolls and ejection of material be-
tween the edge of the rolls and the cheek plates allowed the model to
produce very good fit to the survey data.
The use of a non-normalizable breakage function, coupled to a func-
tion that accounted for the inefficiency in usage of input energy in par-
ticle breakage, proposed on the basis of piston-and-die tests, led to
improved predictions of product size distributions for all HGPRs over
the range of specific forces studied. Finally, an expression for the
power profile along the axial roll length, which allowed to describe
the profile observed in the surveys that approaches a trapezoid, pro-
vided a very good fit to the survey data.
Declaration of Competing Interest
The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to influ-
ence the work reported in this paper.
Acknowledgements
The authors would like to thank Vale for financial and technical sup-
port to the investigation, as well as the Brazilian Agencies CNPq (grant

253
T.M. Campos, G. Bueno and L.M. Tavares Powder Technology 383 (2021) 244–255

Fig. 15. Comparison between experimental and predicted BSA (a) and percentage passing the 45 μm sieve (b) in the product along the roll axial position using Eq. (24) with μ = 0.01
(Eq. 22) and the parabolic profile (Eq. 9). Survey with HPGR #1 with a specific force of 1.5 N/mm2 and a constant throughput of 700 t/h.

"  #1
x2i þ x2i−1 ðxi þ xi−1 Þ 3
xi ¼ ðA:2Þ
4

where xi and xi−1 are, respectively the lower and the upper sizes of the
interval.
The authors then proposed an empirical expression to calculate
Blaine specific surface area (BSA) [41]:

BSA ¼ GSSA þ J ðA:3Þ

Fig. 16. Comparison between experimental and calculated values for the Blaine specific
surface area (BSA) using Eqs. (A.1) to (A.3).
number 310293/2017-0), CAPES and FAPERJ (grant number E-26/
200.592/2018) for partially funding the work.
Appendix A
Given that the model equations provide as output the size distri-
bution in the HPGR discharge, an equation must be available to esti-
mate the BSA from it. Zhang and Napier-Munn [41] proposed an
expression to predict the BSA according to the size distribution,
given by:
6 N wi
SSA ¼ ∑ ðA:1Þ
ρsp i¼1 xi
where SSA is the equivalent spherical specific surface area, ρsp is the spe-
cific gravity, wi the fraction retained in each size class and xi is the har-
monic mean of each size class, given by:
where G and J are constants that must be fitted to data. On the basis of
data from pressing iron ore concentrates in a pilot-scale campaign, Cam-
pos et al. [6] estimated the optimal values of G = 0.996 and J = 18.0.
However, given the differences in material characteristics between that
study and the feed to the various HPGRs in the present work, given by a
blend of different ore type in the Complexo de Tubarão from Vale S.A.
(Vitória, Brazil), a need for re-estimating the parameters became evi-
dent. This resulted in the constants G equal to 1.398 and J equal to
27.97. The constants G and J were calibrated based in the equivalent
spherical specific surface area (SSA) given in m2/kg and, therefore, the
BSA (Eq. A.3) will be calculated in m2/kg. Transformation of this value
to cm2/g requires multiplying the values achieved in Eq. A.3 by 10. A
comparison between experimental data and the model fit is given in
Fig. 16, which shows that the average difference between measured
and fitted results was 100 cm2/g.
References
[1] F.P. Van der Meer, Roller press grinding of pellet feed, Experiences of KHD in the
Iron Ore Industry, AusIMM Conference on Iron Ore Resources and Reserves Estima-
tion, Perth 1997, pp. 1–15.
[2] G. Ehrenraut, R. Rao, Experience with roller press in the pellet plant Kudremukh Iron
Ore Company Ltd, Aufber. Tech. 10 (2001) 469–476.
[3] F.P. Van der Meer, Pellet feed grinding by HPGR, Miner. Eng. 73 (2015) 21–30.
[4] A. Abazarpoor, H. Halali, R. Hejazi, M. Saghaeian, HPGR effect on the particle size and
shape of iron ore pellet feed using response surface methodology, Physicochem.
Probl. Miner. Process. 53 (2017) 1–9.
[5] T.M. Campos, G. Bueno, G.K. Barrios, L.M. Tavares, Pressing iron ore concentrate in a
pilot-scale HPGR. Part 2: modeling and simulation, Miner. Eng. 140 (2019) 105876.
[6] T.M. Campos, G. Bueno, G.K. Barrios, L.M. Tavares, Pressing iron ore concentrate in a
pilot-scale HPGR. Part 1: experimental results, Miner. Eng. 140 (2019) 105875.
[7] T.M. Campos, G. Bueno, L.M. Tavares, Scale-up of HPGR in size reduction of iron ore
pellet feed, Proc. XXX Int. Miner. Process. Congr. SAIMM, Cape Town 2020, pp. 1–10.
[8] S. Rashidi, R.K. Rajamani, D.W. Fuerstenau, A review of the modeling of high pres-
sure grinding rolls, KONA Powder and Part. J. 34 (2017) 125–140.

254
T.M. Campos, G. Bueno and L.M. Tavares
[9] D.W. Fuerstenau, A. Shukla, P.C. Kapur, Energy consumption and product size distri-
butions in choke-fed, high-compression roll mills, Int. J. Miner. Process. 32 (1991)
59–79.
[10] L.G. Austin, R. Weller, I. Lim, Phenomenological Modelling of the High Pressure
Grinding Rolls, Proc. XVIII Int. Miner. Process. Congr, AUSIMM, Sydney, 1993 87–96.
[11] L.G. Austin, M.P. Trubeljal, H.M. Von Seebach, Capacity of high-pressure grinding
rolls, Min. Metall. & Expl. 12 (1995) 65–73.
[12] W. Lim, J. Campbell, L. Tondo, The effect of rolls speed and rolls surface pattern on
high pressure grinding rolls performance, Miner. Eng. 10 (1997) 401–419.
[13] W. Lim, K. Weller, Some benefits of using studded surfaces in high pressure grinding
rolls, Miner. Eng. 12 (1999) 187–203.
[14] S. Morrell, W. Lim, F. Shi, L. Tondo, Modelling of the HPGR crusher, in: S.K. Kawatra
(Ed.), Comminution Practices, SME 1997, pp. 117–126.
[15] M. Torres, A. Casali, A novel approach for the modelling of high-pressure grinding
rolls, Miner. Eng. 22 (2009) 1137–1146.
[16] C.L. Schneider, V.K. Alves, L.G. Austin, Modeling the contribution of specific grinding
pressure for the calculation of HPGR product size distribution, Miner. Eng. 22 (2009)
642–649.
[17] H. Dundar, H. Benzer, N. Aydogan, Application of population balance model to HPGR
crushing, Miner. Eng. 50-51 (2013) 114–120.
[18] U. Lubjuhn, K. Schönert, Material Flow in the Acceleration Zone and Throughput of
High Pressure Roller Mills, Proc. XVIII Int. Miner. Process. Congr, AUSMM, Sydney,
1993 161–168.
[19] M.J. Daniel, HPGR Model Verification and Scale-up, Masters Thesis Julius
Kruttschnitt Mineral Research Centre, University of Queensland, 2002.
[20] R. Van der Ende, H. Kanapp, F. Van der Meer, Reducing edge effect and material by-
pass using spring-loaded cheek plates in HPGR grinding, 7th Int. Conf. Autogenous
Grinding, Semi-autogenous Grinding and High Pressure Grinding Roll Technol, Van-
couver 2019, pp. 1–12.
[21] J. Quist, C.M. Evertsson, Simulating pressure distribution in High Pressure Grinding
Rolls using the discrete element method, 8th Int. Comminution Symp., Cape Town
2012, pp. 1–10.
[22] G.K. Barrios, L.M. Tavares, A preliminary model of high pressure roll grinding using
the discrete element method and multi-body dynamics coupling, Int. J. Miner. Pro-
cess. 156 (2016) 32–42.
[23] T.M. Campos, G. Bueno, L.M. Tavares, Confined bed breakage of fine iron ore concen-
trates, Minerals 10 (2020) 666.
[24] J. Liu, K. Schönert, Modelling of interparticle breakage, Int. J. Miner. Process. 44
(1996) 101–115.
255
Powder Technology 383 (2021) 244–255
[25] J.T. Kalala, H. Dong, A.L. Hinde, Using piston die tests to predict the breakage behav-
ior of HPGR, 5th Int. Conf. Autogenous and Semi-Autogenous Grinding and High-
Pressure Grinding Roll Technol, Vancouver 2011, pp. 1–14.
[26] T. Mütze, Modelling the stress behaviour in particle bed comminution, Int. J. Miner.
Process. 156 (2016) 14–23.
[27] M.J. Daniel, S. Morrell, HPGR model verification and scale-up, Miner. Eng. 17 (2004)
1149–1161.
[28] L.G. Austin, P.T. Luckie, The estimation of non-normalized breakage distribution pa-
rameters from batch grinding tests, Powder Technol. 5 (1972) 267–271.
[29] J.A. Herbst, D.W. Fuerstenau, Scale-up procedure for continuous grinding mill design
using population balance model, Int. J. Miner. Process. 7 (1980) 1–31.
[30] K.J. Reid, A solution to the batch grinding equation, Chem. Eng. Sci. 20 (1965)
953–963.
[31] E. Vyhmeister, L. Reyes-Bozo, R. Rodriguez-Maecker, C. Fúnez-Guerra, F. Cepeda-
Vaca, H. Valdés-González, Modeling and energy-based model predictive control of
high pressure grinding roll, Miner. Eng. 134 (2019) 7–15.
[32] K. Schönert, U. Sander, Shear stresses and material slip in high pressure roller mills,
Powder Technol. 122 (2002) 136–144.
[33] H. Kellerwessel, High pressure particle bed comminution, state of the art, applica-
tion, recent developments, Eng. Min. J. (1996) 45–52.
[34] R. Klymowsky, N. Patzelt, J. Knecht, E. Burchardt, An overview of HPGR technology,
Proceedings of the SAG Conference 2006, pp. 11–26.
[35] R.P. King, Modeling and Simulation of Mineral Processing Circuits, 1st ed.
Butterworth-Heinemann, 2001.
[36] T.M. Campos, G. Bueno, V.R. Alfonso, A.-C. Böttcher, F. Mayerhofer, A. Kwade, L.M.
Tavares, Relationships between particle breakage characteristics and comminution
response of fine iron ore concentrates, Miner. Eng. (2020) (submitted for publica-
tion, under review).
[37] P.F. Van der Meer, I.A. Leite, Aspects of HPGR in iron ore pellet feed preparation, 6th
Brazilian Iron Ore Aggl. Symp, ABM, São Paulo 2018, pp. 102–115.
[38] G.E. Dieter, D. Bacon, Mechanical Metallurgy, vol. 3, McGraw-Hill, New York, 1986.
[39] K. Meyer, Pelletizing of Iron Ores, 1st ed. Spring-Verlag, New York, 1980.
[40] T. Mütze, Energy dissipation in particle bed comminution, Int. J. Miner. Process. 136
(2015) 15–19.
[41] Y.M. Zhang, T.J. Napier-Munn, Effects of particle size distribution, surface area and
chemical composition on Portland cement strength, Powder Technol. 83 (1995)
245–252.