Minerals Engineering 205 (2024) 108487

Contents lists available at ScienceDirect

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

Stress-based approach for predicting and improving large-scale HIG

mill performance
Genzhuang Li a, *, Fisher Wang b, Bern Klein a, Michael Deal b, Aaron Wright b, Hector Mezquita b
a
NBK Institute of Mining Engineering, University of British Columbia, 517-6350 Stores Road, Vancouver, BC V6T 1Z4, Canada
b
First Majestic Silver Corp., 1800-925 West Georgia Street, Vancouver, British Columbia V6C 3L2, Canada

A R T I C L E I N F O A B S T R A C T

Keywords: As a fine and ultrafine grinding technology, the High Intensity Grinding (HIG) Mill is relatively new and there is
High intensity grinding opportunity to improve its performance by adjusting operational conditions. A laboratory HIG5 mill is commonly
Mill speed used for HIG mill sizing and scale up, however the clear correlation between the operational variables of lab­
Flow rate
oratory mill to their production level is somewhat lacking. This paper presents a stress-based approach for
Laboratory HIG5 mill
Production mill
predicting and improving large-scale HIG mill performance. A total of 27 experimental tests were performed with
a HIG5 mill over a range of operational variable set points, including media density, media filling level, feed solid
content, mill speed and flow rate. The influence of the variables on product size, mill operating power and
specific grinding energy was characterized via multivariable linear regression modelling. Based on theoretical
stress analysis and empirical calibration with available survey data, the model derived from laboratory testing
was calibrated to predict the operation of the commercial HIG1600 mill. An approach was proposed for deter­
mining the effect of operational variables using a HIG5 that can be directly applied to reduce energy con­
sumption at a given mill feed rate and product size target for the production scale mill. Results demonstrated that
the stress-based approach integrated with laboratory testing can be used for predicting and improving large-scale
HIG mill performance.

1. Introduction opportunity to improve its operational performance. Based on labora­

tory, pilot and full-scale measurements taken over a range of process
Over the past decades, declining ore grades and more complex conditions, Heath et al. (2017) developed a power model for HIG mills
mineral deposits have led to an increased number of fine and ultrafine where the influence of mill speed, feed flow, bed depth, bead size and
grinding applications in the minerals industry. Stirred media mills have density, and feed slurry density were considered. The most recent study
been commonly used for such applications that enable liberation of by Altun et al. (2023) investigated the influences of operational pa­
finely disseminated minerals with relatively lower energy input (Gupta rameters (solid content, the type of grinding media, the alignment of the
and Yan, 2016). agitators, agitator material, and bead filling) on HIG mill energy effi­
The High Intensity Grinding (HIG) mill is used for fine and ultrafine ciency based on 6 circuit surveys. However, when it comes to industrial
grinding application and was first introduced by Swiss Tower Minerals practices, it is necessary to adopt a more robust approach that encom­
AG (STM) into the mining and mineral industry in 2012 and hence passes the varied operational ranges and predicts the mill performance
marketed by Metso (now Outotec). Until July 2023, a total of 87 HIG in a more comprehensive manner. Furthermore, there remains a need for
mills have been installed in hard rock mineral processing plants, ac­ leveraging the developed understanding of the influence of operational
counting for a total installed power of 203 MW. It was reported that the variables and establishing a practical method for optimizing the HIG
advantages of the HIG mill grinding technology include high energy mill operations.
efficiency, steep product particle size distribution, open-circuit config­ The standard laboratory HIG5 mill has similar breakage fundamen­
uration (no external classification required) and high flexibility for tals to industrial HIG mills and generates energy size reduction that
precise process control (STM, 2022). represent the full production units. Scale-up tests by Outotec showed
As the HIG mill is a relatively new technology, there is an that the laboratory HIG5 mill is sufficiently accurate in obtaining energy

* Corresponding author.
E-mail address: [email protected] (G. Li).

https://doi.org/10.1016/j.mineng.2023.108487
Received 17 June 2023; Received in revised form 20 September 2023; Accepted 31 October 2023
0892-6875/© 2023 Elsevier Ltd. All rights reserved.
G. Li et al. Minerals Engineering 205 (2024) 108487

size reduction results of the production HIG700 mill (Lehto et al., 2016).
Ideally the effect of operating variables should be assessed at a labora­
tory or pilot scale before applying to full scale operation. HIG5 mill
testing requires smaller efforts compared to industrial trials and allows
for studying the influence of operational variables under controlled
conditions. However, a critical challenge exists in comparing the effect
of operational variables on the full-scale production. While the mill
speed of laboratory HIG5 testing can be up to 1200 rpm, the industrial
mills operate at speeds less than 200 rpm. The flow rate of the laboratory
HIG5 and the production mill is also significantly different with units of
L/h and m3/h, respectively. In terms of power, the HIG5 mill has a
nominal power of 7.5 kW while the nominal power of production mills is
generally hundreds or thousands of kilowatts. Although several me­
chanical and empirical approaches have been proposed for other types
of stirred media mills for lab/production correlation, a specific approach
for the HIG mill is presently not available in the public (PCI, 2000,
Stalder, 2014, Mazzinghy et al., 2015, Breitung-Faes and Kwade, 2019,
Mende, 2022).
This paper presents a stress analysis-based testing approach with the
Fig. 1. Particle size distribution of the test feed sample.
HIG5 mill to predict and improve the large-scale HIG mill performance.
The approach includes the following steps:
in Table 1. The equivalent tip speeds are 3.4, 5.2 and 6.9 m/s corre­
sponding to mill shaft speeds of 600, 900 and 1200 rpm. In all tests, the
• Conduct Design of Experiment-based laboratory HIG5 mill testing to
size of the media was constant, comprising a mixture of 50 % 3 mm
study and model the influence of operational variables on product
beads and 50 % 4 mm beads.
size, mill operating power and specific energy consumption of the
A two-level design of experiment (DOE) approach was used to define
process.
the testing program. Preliminary tests were conducted to nominate three
• Determine correlation factors for the operational variables to that of
most significant variables among the 5 operating variables. In the sec­
the production mill, based on theoretical stress analysis and empir­
ond step, a response surface design was used to investigate the influence
ical calibration with available survey data. The theoretical analysis
of the three most significant variables in more detail.
was based on the concepts of stress intensity and number of stress
As a result, a total of 28 tests were conducted including 13 screening
events developed by Kwade et al. (1996).
design tests and 15 response surface design tests. The testing conditions
• Based on the calibrated model, determine operational parameters for
are summarized in Table 2. One test (Test #13) failed due to extreme
the production mill to improve the energy consumption at given
testing conditions, as a result, results from 27 test were used in the final
product size and target throughput.
analysis. A, B, C are the three most significant variables to be determined
by the screening tests. No. 26, 27 and 28 tests were designed to be
2. Materials and methods
repetitions with a purpose to assess reproducibility. Compared to a full-
factorial experimental design, for which a total of 35 tests are required,
2.1. Test sample
the combination of screening and response surface experimental design
significantly reduced the number of tests and sample requirements.
The Santa Elena gold-silver mine is located 150 km northeast of
Hermosillo, near Banamichi in Sonora, Mexico. The processing plant
uses the campaigning method and separately treats ore materials from 2.3. HIG5 laboratory mill testing
two adjacent deposits (Santa Elena mine and Ermitaño mine). The
processing circuit consists of three-stage crushing, ball milling and HIG The laboratory HIG5mill has a grinding chamber (φ143mm × 520
milling followed by cyanide leaching. mm) installed with 9 grinding rotors (φ110mm) of castellated shape.
Approximately ~ 300 kg (dry weight) of HIG mill feed samples were The measured net mill volume is 7.2L. The mill is equipped with a slurry
collected at the operation during the Ermitaño campaign. The samples pump for feeding the material and a control system for adjusting the mill
were split into representative sub-samples for laboratory HIG5 testing. speed and flow rate.
During sampling, operational information was collected for the pro­ The 27 grinding tests were performed with the HIG5 mill using the
duction mill. Particle size analysis was conducted by Malvern laser sizer, single pass testing procedure. For each test, grinding media with a target
with results shown in Fig. 1. The sample has a P80 size of 138 μm. media density were added to the mill to target filling level. The flow rate
and mill speed were adjusted to the set point values. The feed slurry was
2.2. Experimental design prepared to the target solid content and passed through the HIG mill in
open circuit. Mill performance under varying operating conditions was
Five operating variables were investigated to assess their influence evaluated in terms of mill operating power (W), specific grinding energy
on the HIG mill performance. The variables include media density, (SGE, kWh/t) and obtained product size (P80, μm). Mill operating power
media filling level, feed solid content, mill speed and flow rate. The was also recorded. The specific grinding energy (SGE) was thus obtained
media density is the density of the ceramic beads (unit as g/cm3); media
filling level is the fraction of the media true volume with respect to the Table 1
net volume of the mill (unit as vol%); feed solid content is the solids’ Operating variables and their tested value ranges.
weight to total weight of feed slurry (unit as wt.%); mill speed refers to Variable Unit High (þ1) Medium (0) Low (¡1)
the shaft rotational speed (unit as rpm); flow rate refers to the volume of Media density g/cm3 3.8 4.5 5.0
slurry that passes through the mill per unit of time (unit as L/h for Media filling level vol% 51.7 43.0 34.4
laboratory mill or m3/h for production mill). Each variable was tested at Feed solid content wt.% 42 47 52
three distinct levels (high, medium, low) to reflect the range of operating Mill speed rpm 600 900 1200
Flow rate L/h 90 120 150
conditions. The determined levels of the 5 parameters are summarized

2
G. Li et al. Minerals Engineering 205 (2024) 108487

Table 2
Test conditions defined by DOE approach.
Screening Experimental Design

Test order Media density, g/cm3 Media filling level, Feed solid content, Mill speed, rpm Flow rate, L/h
vol% wt.%

1 3.8 34.4 42 900 90

2 3.8 34.4 52 600 150
3 3.8 43.0 42 1200 150
4 3.8 51.7 47 600 90
5 3.8 51.7 52 1200 120
6 4.5 34.4 52 1200 90
7 4.5 43.0 47 900 120
8 4.5 51.7 42 600 150
9 5.0 34.4 42 600 120
10 5.0 34.4 47 1200 150
11 5.0 34.4 52 600 90
12 5.0 51.7 42 1200 90
13 5.0 51.7 52 900 150

Response Surface Experimental Design

Test order Variable A Variable B Variable C

14 0 − 1 − 1
15 0 − 1 +1
16 0 +1 − 1
17 0 +1 +1
18 − 1 0 − 1
19 − 1 0 +1
20 1 0 − 1
21 1 0 +1
22 − 1 − 1 0
23 − 1 +1 0
24 1 − 1 0
25 1 +1 0
26 0 0 0
27 0 0 0
28 0 0 0

from the recorded power, volumetric flow rate and calculated slurry 2.5. Determination of correlation factors
density.
The mill operation was considered stabilized when a minimum 4 mill However, the developed MLR models for the laboratory mill cannot
active mill volume (defined as the mill net volume minus the grinding be used directly for predicting the operation of production mills.
media volume) of slurry were fed to the mill. For each test, five slurry Particularly, mill speed and flow rate of laboratory and production mills
samples were collected from the discharge stream at the mill’s steady are not directly comparable. To address this issue, an approach was
state of operation. The five slurry samples were analyzed by Malvern developed for determining the correlation factors for flow rate and mill
laser sizer and particle size distribution (PSD) results of the five samples speed based on the stress intensity and stress number analysis.
were averaged to represent the product size. Theoretically, the stirred media milling process can be treated as a
series of stress events varying stress intensities. The energy efficiency of
2.4. Modelling the grinding process is mainly determined by the stress intensity and
stress number (Kwade et al., 1996, Becker et al., 2001, Jankovic, 2003).
Multivariable linear regression (MLR) technique was applied to Therefore, to ensure the same SGE and product size results for the lab­
obtain the relationship between the operational variables and mill per­ oratory HIG5 and production HIG1600 mill, stress intensity and stress
formance indicators. The p-value of the variables indicated the signifi­ number should be the same.
cance of variables in influencing the mill performance. Variables with p- For the laboratory and production mill, to obtain equivalent particle-
values below 0.05 were identified as significant at the 95 % confidence media collision frequency and therefore equivalent stress number per
level. The MLR models were used for predicting mill operating power, time, the number of media beads should have a constant ratio to the
SGE and product size at given operating conditions. number of feed particles. Given a constant media filling level, media size
and the same feed particle size and solid content, the correlation factor
of the flow rate variable is determined as net volume of the production
mill to that of the laboratory mill as per in Equations 1 and 2. As a result,

Table 3
Mill parameters of HIG5 and HIG1600 mills.
Mill scale and mode Nominal power Net volume Disc diameter Torque efficiency

Laboratory HIG5 mill 7.5 kW 7.2 L 110 mm 95 %

Production HIG1600 mill 1600 kW 8194 L 1160 mm 95 %

3
G. Li et al. Minerals Engineering 205 (2024) 108487

power of the production mill would also be multiplied by the same factor calculated by dividing the applied force on the mill shaft by its rotational
the flow rate, as per in Equations 3 and 4. surface area (PCI, 2000). Based on Equations 5, 6 and 7, the mill speed is

Media filling level*Net volume*Packing factor/Averaged volume of individual media beads

constant stress number ∝ (1)
Flow rate*Slurry density*Feed solid content/Averaged volume of individual feed particle

related to power draw, torque efficiency, disc diameter and the rota­
tional surface area, calculated using Equation 8. Assuming that the
laboratory and production mills have the same stress intensity and tor­
Flow rate (production mill) Net volume (production mill)
= (2) que efficiency, the correlation factor of the mill speed can be calculated
Flow rate (laboratory mill) Net volume (laboratory mill)
using Equation 9. This method to estimate the equivalent mill speed is
considered more reliable than the approach of scaling up the mill RPM
Power = SGE*Flow rate*Slurry density*Solid content (3)
speed to achieve equal tip speed (PCI, 2000).
Power(production mill) Flow rate(production mill) Force
= (4) Stress intensity∝ (5)
Power(laboratory mill) Flow rate(laboratory mill) Rotational Surface Area
Stress intensity of the media-particle breakage is considered to be
Torque
proportional to the stress intensity within the milling environment (such Force = (6)
Disc Diameter
as the stress intensity onto the mill shaft). The article “Scaling for ho­
rizonal media mills” suggested that the stress on the shaft can be

Fig. 2. Interface of the optimization program and demo of operating steps.

4
G. Li et al. Minerals Engineering 205 (2024) 108487

Table 4 friendly interface (as shown in Fig. 2).

Constraints of operational variables for production operation. The approach and the resulting program were used to examine his­
Variable Unit Min Max torical production data where the energy savings potential associated
3 with operational improvements were evaluated. A practical operating
Media density g/cm 3.7 5.0
Media filling level vol% 34.4 51.7 strategy was also developed for future production under various product
Feed solid content wt.% 42.0 52.0 size and throughput scenarios.
Mill speed rpm 100 189 It should be noted that the applicability of the developed approach
Flow rate m3/h 50 140 was limited to HIG1600 operation with Ermitaño ores received from the
Santa Elena mine. As summarized in Table 4, the determined opera­
Power tional variables were also constrained to ranges used in the experimental
Torque = /Torque Efficiency (7) testing and actual operation practices. The predicted mill operating
2*PI*Mill speed(rpm)/60
power should also be limited to below the installed power (1600 kW) of
the production mill.

Power*60
Mill speed (rpm)∝ (8)
2*PI*Torque Efficiency*(Stress intensity*Rotational Surface Area*Disc Diameter)

Mill speed (production) Power(production) Disc diameter (laboratory) Rotational surface area (laboratory)
= * * (9)
Mill speed (laboratory) Power(laboratory) Disc diameter (production) Rotational surface area (production)

The net volume, disc diameters and torque efficiency of the HIG5 and 3. Results and discussion
HIG1600 mills are summarized in Table 3. The rotational surface area
can be estimated as the surface area of the discs for the HIG mill. This 3.1. Influence of operational variables
estimate may be difficult to obtain considering the castellated shape of
the HIG mill discs. In practice, where the media regime and slurry vis­ Table 5 summarizes the p-values of the five operating variables in
cosity of the laboratory mill testing may not be the same and where the correlation to the HIG5 mill performance. In terms of mill operating
parameters used in the theoretical calculation may not be accurately power, the significance of the variables was ordered as mill speed > media
estimated, an alternative approach can be to obtain the correlation
factors through empirical calibration with production data. The torque
efficiency was assumed constant as 95% for laboratory and production Table 6
mills. Statistical descriptions of tested operational variables and obtained results.
For this study, the HIG1600 mill survey data was used for deter­ Variables Mean Min Max Q1 Median Q3 Std.
mining the correlation factors for the flow rate and mill speed. Samples
X1-Media 4.4 3.8 5.0 3.8 4.5 5.0 0.5
used for laboratory HIG5 testing were collected during the survey. The density, g/
feed particle size and ore hardness in the production mill were therefore cm3
considered to be the same as in the laboratory testing. The correlation X2-Media 42.7 34.4 51.7 34.4 43.0 51.7 6.8
factors are used to calibrate the developed experimental models so that filling
level, vol%
the models would be applicable to production data. The factors were X3-Feed 46.8 42.0 52.0 47.0 47.0 47.0 2.9
determined at which the experimental models most accurately predicted solid
mill operating power, SGE and product P80 size (with the minimum content,
combined Relative Squared Error comparing to the actual production wt.%
X4-Mill 627.1 316.0 1141.0 547.0 622.8 731.0 173.5
data).
speed, rpm
X5-Flow 118.9 90.0 150.0 120.0 120.0 120.0 17.3
rate, L/h
2.6. Prediction of production performance Y1-P80, μm 48.4 11.6 99.1 35.0 44.8 53.2 20.1
Y2-Power, 2383 1160 6199 1703 2227 2936 950
W
The calibrated models were used for predicting production of the
Y3-SGE, 31.4 10.1 90.1 23.2 30.7 34.1 14.4
HIG1600 mill at the Santa Elena mine site. In addition, optimal oper­ kWh/t
ating parameters were determined at which the specific grinding energy
consumption can be minimized at given product size and HIG mill feed
throughput rate. A simple and executable program was also developed
for practical use with the calibrated model integrated with a user- Table 7
Results of the developed P80, power and SGE models for laboratory HIG5 mill.

Table 5 Variables Y1-P80, μm Y2-Power, W Y3-SGE, kWh/t

P-values of operational variables in correlation to the HIG5 mill performance. X1-Media density, g/cm 3
− 8.38 777.90 8.46
X2-Media filling level, vol% − 0.25 81.84 0.95
Independentvariables Mill SGE, Product
X3-Feed solid content, wt.% 1.94 21.50 − 0.44
Dependentvariables power, W kWh/t P80, μm
X4-Mill speed, rpm − 0.09 7.62 0.10
Media density, g/cm3 0.0015 0.0551 0.0438 X5-Flow rate, L/h 0.47 0.40 − 0.26
Media filling level, vol% 0.0012 0.0384 0.0901 Intercept 5.88 − 10382.55 − 57.26
Feed solid content, wt.% 0.2710 0.2528 0.2222 R-squared 0.69 0.98 0.96
Mill speed, rpm <0.0001 0.0001 0.0077 Adjusted R-squared 0.62 0.97 0.95
Flow rate, m3/h 0.7258 0.0062 0.1435 Standard error 12.60 164.56 3.42

5
G. Li et al. Minerals Engineering 205 (2024) 108487

Fig. 3. Correlation of predicted and actual power, SGE and product P80.

Table 8 Table 9
Results of the calibrated P80, power and SGE models for HIG1600 production. Comparison of survey data and model predictions.
Variables Y1-P80, μm Y2-Power, kW Y3-SGE, kWh/t Parameters Operational Model Relative Squared
data prediction Error
X1-Media density, g/cm3 − 8.38 861.12 8.46
X2-Media filling level, vol% − 0.25 84.98 0.95 Media density, g/ 3.7 / /
X3-Feed solid content, wt.% 1.94 23.80 − 0.44 cm3
X4-Mill speed, rpm − 0.28 26.21 0.31 Media filling level, 40.8 / /
X5-Flow rate, m3/h 0.43 0.40 − 0.24 vol%
Intercept 5.88 − 11493.33 − 57.26 Feed solid content, 47 / /
R-squared 0.69 0.98 0.96 wt.%
Adjusted R-squared 0.62 0.97 0.95 Mill speed, rpm 165 / /
Standard error 12.60 182.16 3.42 Flow rate, m3/h 110.0 / /
Product size P80, μm 49.1 54.8 0.013
Mill power, kW 1332 1230 0.015
filling level > media density > flow rate > feed solid content. Over the range SGE, kWh/t 18.3 20.5 0.006
of variable levels tested, the first three were most significant with p-
values less than 0.05. The same three variables were identified as most
3.2. Experimental modelling
significant for product P80 with slightly different order: mill speed >
media density > media filling level > flow rate > feed solid content. In terms
Table 6 summarizes the statistical results for all 27 experimental
of SGE, the order of significance was mill speed > flow rate > media filling
tests. Results showed that the HIG5 milling responses varied signifi­
level > media density > feed solid content.
cantly over the tested ranges of operating variables. The obtained mill
Considering all three mill performance indicators, media density,
power ranged from 1160 to 6199 W, the SGE varied in ranged from 10.1
mill speed and media filling level were determined as the three most
to 90.1 kWh/t, and the product P80 varied from 11.6 to 99.1 μm. The
significant and were investigated further in subsequent response surface
experimental results were anticipated to represent the industrial pro­
experimental testing. The feed solid content and flow rate were kept
duction under wider range of operating conditions.
constant at 47 % and 120 L/h, respectively.
The results of the developed models are summarized in Table 7 and
the regression models are expressed as Equations 10, 11 and 12. The

6
G. Li et al. Minerals Engineering 205 (2024) 108487

comparison of predicted and actual values is shown in Fig. 3. in actual production (6 mm for new media addition; seasoned media size
not known). As a result, it was anticipated that the actual production
yP80(μm) = − 8.38x1 − 0.25x2 + 1.94x3 − 0.09x4 + 0.47x5 + 5.88 (10)
would likely have a smaller stress intensity compared to the theoretical
analysis so as to obtain equivalent energy efficiency. Therefore, the
yPower(W) = 777.90x1 + 81.84x2 + 21.50x3 + 7.62x4 + 0.40x5 − 10382.55
theoretical correlation factor of the mill speed would likely be less than
(11)
0.29 to compensate for the differences caused by using a different media
size. To verify the assumption, the theoretical correlation factor of the
ySGE(kWh/t) = 8.46x1 + 0.95x2 − 0.44x3 + 0.10x4 − 0.26x5 − 57.26 (12)
mill speed could be calculated by measuring the rotational surface area
Results showed that the predicted power and SGE models correlate of the mill discs for both laboratory and production mills.
well with the actual power and SGE values, with adjusted R-squared For future studies, it is recommended laboratory testing should
values of 0.97 and 0.94, respectively. The standard error associated with maintain a consistent media regime (including the media size, media
the model predictions were ± 0.165 and ± 3.42 kWh/t. The P80 model density, media filling level) and feed viscosity (including feed size and
was less accurate, with an adjusted R-squared of 0.62 and a prediction solid content) to that of the production mill. Therefore, the laboratory
error of ± 12.60 μm. and production mills will have more equivalent media-particle stress
It is important to note that some influential variables such as feed breakage behaviours and more comparable energy efficiency results.
size and ore harness were not included in the modelling due to limited The correlation parameters determined from such practices are expected
samples that were available at the time of the study. Consequently, the to be similar to that determined by the theoretical analysis. It is also
model should only apply to the mill feed with similar properties (such as recommended to conduct production tests under the determined
particle size and ore hardness) to the tested material. More test work optimal operating conditions to compare them with model predictions
should be conducted incorporating such variables to obtain a more and thereby validate the developed approach.
comprehensive and robust model.
3.5. Production prediction and optimization
3.3. Calibration of experimental models with correlation factors
The calibrated model was used to predict the HIG1600 mill operation
To apply the developed experimental model for predicting the in­ at the Santa Elena mine site. Given a target product size and HIG mill
dustrial scale production, mill speed and flow rate parameters are cali­ feed throughput, the optimal operational parameters were determined
brated by multiplying them with the correlation factors. In this study, to reduce the specific grinding energy consumption. In the analysis, the
the empirical calibration approach was applied with available survey SGE (kWh/t) was calculated by dividing power (kW, predicted by the
data. The correlation parameters for mill speed and flow rate were power model) by throughput (tph, calculated by feed solid content,
determined where the calibrated model has the minimum combined slurry density and flow rate). This method was used to ensure a mean­
relative squared error in predicting mill operating power, SGE and ingful relationship between SGE, power and throughput.
product P80 size. Historical production data was examined to understand the energy
The correlation parameters for mill speed and flow rate are deter­ savings potential that could be achieved by adjusting the operational
mined to be 0.29 and 1038, respectively. The results of the calibrated variables. Table 10 show three production cases with significantly
models are summarized in Table 8 and mathematical Equations 10, 11 different product sizes and throughputs.
and 12. Table 9 showed the predicted results of the calibrated models Results showed that by use of selected operational parameters, en­
comparing to the survey data. ergy saving of 3.0, 1.5 and 3.7 kWh/t were obtained respectively. The
results indicate the opportunity for considerable energy savings by
yP80(μm) = − 8.38x1 − 0.25x2 + 1.94x3 − 0.28x4 + 0.43x5 + 5.88 (10) selecting the recommended operational variable levels for the HIG mill.
It should be noted that the predicted energy savings are based assuming
yPower(kW) = 861.12x1 + 84.98x2 + 23.80x3 + 26.21x4 + 0.40x5 − 11493.33
the feed properties (such as hardness and particle size) of the production
(11) cases were same to materials that were tested in the laboratory. How­
ever, in reality the ore properties might vary.
ySGE(kWh/t) = 8.46x1 + 0.95x2 − 0.44x3 + 0.31x4 − 0.24x5 − 57.26 (12)
In addition, some of the operational variables might not be as
accurately measured on the mine site such as in the controlled labora­
3.4. Comparing the empirical calibration approach to theoretical analysis tory environment, which could also affect the operation/optimized
comparison. The cases shown here were intended to demonstrate how
The correlation factor of the flow rate is lower compared to its the developed approach can be used to improve operational perfor­
calculated value based on the theoretical analysis (8194L/7.2L ≈ 1138), mance of full scale HIG mills. It should also be noted higher mill speed
indicating a higher stress frequency would likely occur in actual mill would tend to more media and linear wear, which has not been included
operation. This could be due to the smaller media size used in the in the optimization analysis.
experimental testing (3 mm:4mm = 50 %:50 %) compared to that used Table 11 presents a summary of the best results for various

Table 10
Comparison of historical production to optimized operation.
Parameters Production case 1 Production case 2 Production case 3

Operation Optimized Operation Optimized Operation Optimized

3
Media density, g/cm 3.7 3.7 3.7 3.7 3.7 3.7
Media filling level, vol% 40.8 34.4 43.9 34.4 41.5 34.4
Feed solid content, wt.% 47.0 42.0 50.0 42.0 46.0 42.0
Mill speed, rpm 165.0 184.0 160.0 183.3 164.0 183.4
Flow rate, m3/h 110.0 128.4 109.9 140 110.1 125.7
Product size P80, μm 49.1 49.1 54.6 54.6 48.1 48.1
Mill feed throughput, tpd 72.7 72.7 79.3 79.3 71.2 71.2
SGE, kWh/t 18.3 15.3 15.4 13.9 19.7 15.4
Mill power, kW 1332 1112 1224 1098 1402 1068

7
G. Li et al. Minerals Engineering 205 (2024) 108487

Table 11
Optimal HIG1600 results for given mill feed throughput and target product sizes.

combinations of throughput and product size scenarios (assuming for determining the optimal operational variables and minimizing en­
similar feed properties), where the operational variable cells are ergy consumption at a given HIG mill feed throughput and product size.
conditionally highlighted based on variables’ values. The product sizes Experimental modelling showed that the media density, mill speed
and throughputs that can be obtainable by the production mill were and media filling level were the most significant variables that influence
preliminarily defined. For smaller throughput such as 50tph, finer the mill performance. Optimization analysis with the HIG1600 pro­
product sizes of 30 ~ 40 μm could be obtained; while for higher duction showed that under the given operational constraints, low media
throughputs of 80tph, the mill would only be capable of relatively filling level and low feed solid content should be used to minimize the
coarser grinding tasks (product P80 of 50 ~ 65 μm). Such information energy consumption for a target throughput. Meanwhile, the optimal
could be valuable to the mine site personnel in understanding the mill’s mill speed, flow rate and media density should vary depending on the
capacity at optimal operating conditions. target product size and throughput.
Results also showed that under the operational constraints, low To conclude, the laboratory testing and stress analysis-based
media filling level and low feed solid content should be used for the approach can be used for predicting and improving the production
investigated grinding tasks to obtain the best energy efficiency. Mean­ scale HIG operation under complex operational conditions. It should
while, depending on the target product size and throughput, the optimal also be noted that some influential variables such as media size, feed size
mill speed, flow rate and media density would vary. The optimal flow and ore harness were not included in the modelling due to limitations at
rates were determined as 140.0, 123.6, 105.9 and 88.3 m3/h respec­ the time of the study. In addition to the energy efficiency, wear of the
tively for target throughputs of 80, 70, 60 and 50 tph, respectively. It media and rotor should also be considered in determining the optimal
was also observed that at a target throughput of 80 tph, the flow rate operating parameters. More test work can be further conducted in
reached its highest constraint (140 m3/h) and in this case, a slightly obtaining a more comprehensive and robust model that would more
higher feed solid content should be used to meet the throughput target. accurately reflect the complex production in actual practices. It is also
At a specific throughput, the optimal mill speeds and/or media recommended to conduct trail production tests under the determined
densities should be determined differently for different product sizes. optimal operating conditions to compare them with model predictions
For coarse product particle size ranges, as the target product size be­ and thereby validate the developed model.
comes coarser, a lower mill speed should be used. When the target
product size is fine (such as 50 μm at 80tph throughput, and 35 μm at CRediT authorship contribution statement
60tph throughput), the mill speed reached its max operational con­
straints (189 rpm) and, in this case, the use of higher density media is Genzhuang Li: Conceptualization, Investigation, Formal analysis,
recommended. Note that frequent adjustments to the media density Methodology, Writing – original draft. Fisher Wang: Conceptualization,
during mill production would not be practical; rather, such adjustments Resources, Writing – review & editing. Bern Klein: Conceptualization,
can be performed on a long-term basis. Funding acquisition, Resources, Writing – review & editing. Michael
Deal: Resources. Aaron Wright: Resources. Hector Mezquita:
4. Conclusions Resources.

This paper presents a laboratory testing and stress analysis procedure Declaration of Competing Interest
to predict and improve large-scale HIG mill performance. A total of 27
experimental tests were performed with a laboratory HIG5 mill under The authors declare that they have no known competing financial
various media density, media filling level, feed solid content, mill speed interests or personal relationships that could have appeared to influence
and flow rate variable levels. The influence of the variables on product the work reported in this paper.
size, mill power and specific grinding energy was characterized via
multivariable linear regression modelling. Based on theoretical stress Data availability
analysis and empirical calibration with production data, the model
derived from laboratory HIG5 testing was calibrated to predict the The authors do not have permission to share data.
operation of the production HIG1600 mill. An approach was proposed

8
G. Li et al. Minerals Engineering 205 (2024) 108487

Acknowledgement Heath, A., Keikkala, V., Paz, A., Lehto, H., 2017. A power model for fine grinding
HIGmills with castellated rotors. Miner. Eng. 103, 25–32.
Jankovic, A., 2003. Variables affecting the fine grinding of minerals using stirred mills.
The authors would like to thank the First Majestic Silver Corp and Miner. Eng. 16 (4), 337–345.
Mitacs for providing funds for conducting this research project. Kwade, A., Blecher, L., Schwedes, J., 1996. Motion and stress intensity of grinding beads
in a stirred media mill. Part 2: Stress intensity and its effect on comminution. Powder
Technol. 86 (1), 69–76.
References Lehto, H., Keikkala, V., Kurki, P., Paz, A., 2016. Developments in stirred media milling
testwork and industrial scale performance of Outotec Higmill. In Comminution
Altun, O., Darılmaz, Ö., Karahan, E., Sert, T., Altun, D., Toprak, A., Hür, A., 2023. Conference.
Understanding HIGMill operation at copper regrind application; operating Mazzinghy, D.B., Schneider, C.L., Alves, V.K., Galéry, R., 2015. Vertical agitated media
parameters, wear and mineral liberation. Miner. Eng. 191, 107964. mill scale-up and simulation. Miner. Eng. 73, 69–76.
Becker, M., Kwade, A., Schwedes, J., 2001. Stress intensity in stirred media mills and its Mende, S., 2022. "Scale-up of Agitator Bead Mills". from https://analyzing-testing.ne
effect on specific energy requirement. Int. J. Miner. Process. 61 (3), 189–208. tzsch.com.
Breitung-Faes, S., Kwade, A., 2019. Mill, material, and process parameters–A mechanistic PCI, 2000. “Scaling for horizonal media mills ”from https://www.pcimag.com/articles/
model for the set-up of wet-stirred media milling processes. Adv. Powder Technol. 30 82699-scaling-for-horizonal-media-mills.
(8), 1425–1433. Stalder, B., 2014. Scalability of Agitated Bead Mills - From Lab to Production Scale.
Gupta, A., Yan, D.S., 2016. Stirred Mills - Ultrafine Grinding. Mineral processing design Paint&Coatings Industry.
and operations: an introduction. STM, 2022. “HIGMill portfolio.” from https://www.stmminerals.com/#home.