Powder Technology 403 (2022) 117394

Contents lists available at ScienceDirect

Powder Technology

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

Optimization of operating conditions on ultra-fine coal grinding

through kinetic stirred milling and numerical modeling
Francisco Patino, Deniz Talan, Qingqing Huang ⁎
Department of Mining Engineering, West Virginia University, 365 Mineral Resources Building, 1374 Evansdale Drive, Morgantown, WV, United States of America

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

Article history: This study investigated ultra-fine coal grinding performance of four low- to moderate-cost grinding media in a
Received 16 December 2021 laboratory stirred mill. Kinetic grinding tests showed that silica beads generated the finest product size with a
Received in revised form 23 March 2022 P80 of 5.9 μm from a feed size of 24.4 μm while having a specific energy (SE) input of 309 kWh/ton.
Accepted 12 April 2022
Nonetheless, the least energy consumption of 109 kWh/ton was achieved by alumina beads; however, yielding
Available online 16 April 2022
a coarser product size of 15.5 μm. In addition, test results indicate that optimizing media sizes, solids
Keywords:
concentration, and using viscosity modifiers enhance coal ultra-fine grinding performance. Under the optimized
Ultra-fine grinding testing conditions, a P80 of 2.7 μm was produced with a corresponding SE input of 270 kWh/ton using silica beads
Coal ranging from 420 to 850 μm and a dispersant dosage of 14 kg/ton. Moreover, the mathematical models generated
Grinding media materials based on population balance modeling provided an accurate forecast of the particle-size distribution.
Specific energy © 2022 Elsevier B.V. All rights reserved.
Population balance modeling

1. Introduction difference between these mills is their geometry/shape, their operating

principles in terms of breakage mechanisms are the same. Operating pa-
Grinding circuits have always been a crucial part of mineral process- rameters such as stirring speed, feed rate, media filling, media size, pulp
ing operations. Although it is possible to achieve a good liberation with density, grinding aids, etc., are continuously studied to improve milling
conventional grinding approaches, ultra-fine grinding has become in- efficiency [2,7]. One common indicator for evaluating grinding perfor-
creasingly more critical over the years with increased fine-grained mance is the size-energy relationship [50]. The ultra-fine grinding
ores and the depletion of high-grade deposits [68]. This requires an up- (UFG) mills consume a much higher installed power per mill unit vol-
grade of the comminution operations due to the great demand for run- ume [3]. On the other hand, they use a finer media size (i.e., 2–3 mm)
of-mine material to be processed, finer product size required, and than the conventional tumbler mills, which typically require 12 to
elevated energy consumption [24]. Nonetheless, ultra-fine grinding 100 mm grinding media. The use of smaller grinding media in stirred
processes can require up to 4% of the global electric power on an indus- mills helps achieve fine grinds; however, the kinetic energy imparted
trial scale and represent around 50% of the mine site's total consump- to the media decreases, hence significantly reducing the available en-
tion[40]. Therefore, developing performance-enhancing mechanisms ergy transfer in a media/particle contact.
that make particle size reduction processes economically and energeti- The application of stirred mills in the mining industry has gained im-
cally viable has utmost importance. portance in the large-scale processing of metallic minerals such as zinc,
Stirred mills have been widely used in some specialized industries copper, molybdenum, iron, lead, and gold, as well as coal [76]. The need
and are now gaining importance in the mining industry [31]. The limita- to improve the plant performance in regrind circuits by achieving
tions of conventional grinding equipment are overcome by rotating stir- finer liberation sizes drove mineral processing engineers to produce in-
rers inside a stationary mill shell, and fine sizes, in a range of l μm to 10 novative technologies. For instance, Glencore is a pioneer in ultra-fine
μm, can efficiently be produced by significantly increased surface area. grinding and has continuously been devoting more than 25 years to op-
The increased surface area further enhances the efficiency of subse- timizing their flagship, the IsaMill™ [4]. Moreover, companies such as
quent beneficiation processes. The two common types of stirred mills Metso/Eirich, Metso, FLSmidth, and Outotec have also produced their
are vertical mills (VertiMills) and horizontal mills (IsaMills); however, own prototypes known as Vertmill/TowerMill, Stirred Media Detritor
it is not limited to these two main categories [76]. Although the major (SMD), VXP Mill, and HIGMill, respectively [79]. Table 1 summarizes
several of the size-energy relationships obtained from literature
along with the type of stirred mill employed and the ore material.
Manufacturing companies of particle size reduction equipment such
⁎ Corresponding author.
E-mail address: qingqing.huang@mail.wvu.edu (Q. Huang). as Union Processes are also developing cutting-edge attritors to make

https://doi.org/10.1016/j.powtec.2022.117394
0032-5910/© 2022 Elsevier B.V. All rights reserved.
F. Patino, D. Talan and Q. Huang Powder Technology 403 (2022) 117394

Table 1
Stirred mill applications in mineral and coal processing.

Source Mill Type Mfr. Material F80 P80 Spec. Energy

(μm) (μm) (kWh/t)

23.40 10.29 397.00

23.40 2.33 216.00
Huang et al. [38] Attritor Union Process Coal
23.40 6.04 262.00
23.40 4.42 323.00
21.60 12.80 16.10
Adeniji [1] Attritor Union Process Coal 86.40 51.20 92.90
21.60 5.70 218.40
Wang et al. [78] IsaMill™ Glencore Copper-Molybdenum 341.00 75.00 5.10
66.00 36.00 6.50
66.00 39.00 5.30
Mazzinghy et al. [52] Vertmill™ Metso Iron
62.00 36.00 5.80
62.00 35.00 5.60
2287.00 73.00 8.70
Lead‑zinc
Shi et al. [71] Vertmill™ Metso 1999.00 78.00 10.60
Silicate‑gold 2065.00 52.00 17.40
105.00 42.00 10.00
Glencore [27] IsaMill™ Glencore Zinc
45.00 30.00 12.00
129.00 21.10 54.70
Anderson and Bandarian [4] IsaMill™ Glencore Gold 129.00 21.40 55.00
Graham [29] SMD Metso Gold 35.00 10.00 60.00

grinding a step forward. In the past few years, the concept of cryogenic synthetic compounds, or by-products derived from coal, depending
grinding has received increasing attention, and remarkable advances upon its particle size and composition [45]. For instance, Pinchuk et al.
have been achieved. The fundamentals of “cryomilling” are based on em- [60] improved the combustion of coal-water fuel (CWF) using an elec-
brittling mineral particles by inducing temperatures below 275 K during tromagnetic treatment. The solid component of the fuel was coal with
grinding [46]. Usually, liquid nitrogen is used to prepare the cryogen a particle size distribution ranging between 0 and 250 μm. Moreover,
slurry [61]. This method promotes the production of nanosized particles Shukla and his co-workers [72] used different-sized coal samples
and provides significant advantages, particularly for heat-sensitive ma- (i.e., 108 μm, 75.7 μm, and 62.9 μm) to demonstrate the particle size in-
terials such as coal [82]. fluence on the rheological behavior of multiphase fuels such as coal-oil-
In addition to the research efforts to optimize ultrafine grinding be- water fuel (COW). Novel nanostructures such as polycrystalline nano-
haviors through operating conditions and design parameters, mathe- graphite are also produced from fine coal powders [51]. Graphite with
matical models have also been developed to design, optimize, and a mean size of approximately 1 μm is commonly used for the production
scale up grinding circuits. These include empirical, phenomenological, of solar stills where nano-particles and porous materials are required
and fundamental models. The energy-based models were the first em- [56,69,70]. It has also been demonstrated that fine and ultra-fine coal-
pirical models developed in this field and continue to find widespread based materials are a source of rare earth elements (REEs) and other
use. Phenomenological models include population balance models, critical minerals. Silva and his co-workers [73] found that the spontane-
whereas fundamental models cover those based on discrete element ous combustion of coal produced a significant amount of rare earth ele-
method (DEM), computational fluid dynamics (CFD), and statistical ments, carbon nanotubes, and graphene. These minerals can also be
physics [79]. Energy-based models assume a relationship between the concentrated by applying an ultrafine grinding process followed by a
energy input of the comminution device and the resultant effective par- decarbonization step [37].
ticle size of the product. The main theories underlying energy-based Even though many studies have explored the ultra-fine grinding be-
models were initially introduced and evaluated by Rittinger [65] and havior of various minerals using stirred mills [14,55,62,80,83], the stud-
Kick [43]. The first general energy-based comminution equation is devel- ies on coal ultrafine grinding are limited. In addition, there are still many
oped by Walker et al. [77]. Later, Bond [13] proposed his theory known as areas, including modeling and simulation, that need to be studied to un-
“The third theory of comminution”, which stated that the energy input is derstand the grinding mechanism in depth. Therefore, this study sought
proportional to the new crack tip length produced in particle breakage. to optimize ultra-fine coal grinding performance using a laboratory
Charles [16] then derived an energy-size reduction relationship that cor- scale stirred mill. First, the kinetic grinding tests were performed
related the energy input with size reduction and was validated by several while investigating various parameters involved in grinding. Later, the
experimental results. These pave the foundation for the development of experimental results were analyzed, and numerical simulation using
subsequent energy-based models. On the other hand, considerable population balance modeling was performed. Finally, mathematical
work has been conducted to optimize energy consumption in grinding models were proposed based on the selection function, breakage func-
mills using phenomenological grinding kinetics models based on popula- tion, and particle swarm optimization algorithm to indirectly predict
tion balance considerations. Population balance modeling is based on the particle size distribution resulting from comminution.
first-order kinetics and utilizes the specific rate of breakage (or the selec-
tion function) Si and the breakage function bi,j as the basis of the modeling 2. Experimental work
of the grinding process [5,6,35,57]. The application of Population Balance
Models (PBMs) can assist in designing grinding circuits and the scale-up 2.1. Laboratory studies
of laboratory mill results to full scale-up. Nowadays, various studies
have well adopted the approach to investigate breakage kinetics of differ- 2.1.1. Materials and methods
ent minerals (e.g., quartz, pumice, and platinum group minerals) and op-
timize the milling performance using population balance modeling 2.1.1.1. Coal feedstock sample. Feedstock material used throughout the
[18,22,57]. study was provided by an active coal preparation plant located in the
In the past decades, studies have been developed to beneficiate core of the Appalachian Coal Basin, south of Morgantown, West Virginia.
ultra-fine coal, which is widely used to produce high-purity materials, First, samples of Middle Kittanning coal in slurry form were collected

2
F. Patino, D. Talan and Q. Huang Powder Technology 403 (2022) 117394

materials, even though the glass beads have a perfect roundness while
granular sand particles are angular. Table 2 shows detailed information
on the grinding media materials employed in this study, whereas
Table 3 presents the elemental compositions of the coal ash residual
and the grinding media materials obtained from a Vanta ™ Handheld
X-Ray Fluorescence Analyzer (Olympus, USA).
Based on the initial test results, three size ranges of silica beads were
Fig. 1. Grinding media employed to study ultra-fine coal grinding behavior.
chosen, i.e., fine: 40 × 210 μm; regular: 250 × 420 μm; coarse: 420 × 850
μm, to further study the effect of grinding media size on coal liberation
behavior. Moreover, the effects of solids concentration and sodium sili-
cate as a viscosity modifier on ultra-fine grinding were also addressed
Table 2 using coarse silica beads and ceramic balls. Desired solids concentra-
Grinding media material specifications. tions by weight were achieved by dilution or decanting, and the sodium
silicate with analar grade was purchased from Ward's Science and used
Media Type Size Range Specific Hardness Relative Vendor
(mm) Gravity Cost as the slurry viscosity modifier.
Ceramic Balls 0.25–0.34 3.75 Mohs 9 Moderate Carbo Ceramics
Alumina Beads 0.25–0.34 3.80 Mohs 9 Moderate Union Process 2.1.2. Stirred mill kinetic grinding tests
Granular Sand 0.25–0.35 2.65 NA Low Lowes
Kinetic grinding tests were conducted using a laboratory-scale
Silica Beads 0.25–0.42 2.50 Mohs 6 Low Kleen Blast
Fine Silica Beads 0.04–0.21 2.50 Mohs 6 Low Kleen Blast stirred mill (Fig. 2), Union Process, Akron, Ohio). Also known as Attritor,
Coarse Silica Beads 0.42–0.85 2.50 Mohs 6 Low Kleen Blast this versatile mill provides reproducible results from batch to batch and
can be used for wet or dry grinding processes under controlled operat-
ing conditions. The mill is equipped with a 2.5-gal (9.5 L) tank, which
from the plant thickener underflow, and the solids concentration was provides a working capacity of 1.1 gal (4.2 L). The adjustable shaft ac-
adjusted by dilution or decantation. Later, representative samples commodates different-sized grinding media, and the electronic variable
were taken and subjected to the particle size and proximate analysis speed drive system controls the shaft rotation speed in RPM. It also has a
using a CILAS 1190 Particle Size Analyzer (CPS US, Inc., Madison, Wis- bottom discharge grid for sampling and discharge, water-cooled jackets
consin) and Leco TGA 701 (St. Joseph, Michigan), respectively. Accord- for cooling or heating, and a liquid crystal display (LCD) that shows the
ing to the analysis results, the feedstock coal sample had a cumulative agitator's rotational speed (rpm), motor frequency (Hz), and the power
80% passing size of 24.4 μm and an ash content of 48.0%. In addition, draw in kilowatts (kW).
the moisture content of the coal feedstock was 1.02%, while its volatile The ultra-fine grinding tests followed a similar procedure under-
matter and fixed carbon contents were 22.64% and 28.34%, separately. taken by Adeniji [1] and Huang et al. [38]. In this study, replicate tests
These results correspond to the mean values obtained from multiple were conducted to provide reproducible results and ease the analysis
replicate tests. of variation in product size and energy consumption when one param-
eter was changed and the others were held constant. The four main pa-
2.1.1.2. Grinding media. Four low- to moderate-cost grinding media rameters assessed in this study are grinding media type, media size,
(i.e., ceramic balls, alumina beads, silica beads, and granular sand) solids concentration, and dispersant dosage. Mill specification data are
were chosen to evaluate their performance on the ultra-fine coal grind- provided in Table 4, and detailed test conditions are presented in
ing (Fig. 1). To ensure the commensurability and reliability of the re- Table 5.
sults, similar size ranges were either selected or prepared to analyze The grinding media was introduced into the vessel with the coal
the effect of media type on ultra-fine grinding performance. All the sample in slurry form. The shaft was slightly jogged to mix both mate-
grinding media types were screened using a Sweco Round Separator rials homogenously and ensure voids were fully covered with pulp.
(Sweco, Florence, Kentucky) to ensure that the media particles were The slurry holdup volume and media load were selected based on the
within the specified size range. This pre-screening process eases the re- mill capacity. The coal pulp filled the remaining volume not occupied
covery of the media material after grinding tests and avoids the alter- by grinding media. A freeboard was maintained to avoid overflowing
ation of the coal particle size analysis due to the presence of media due to the vortex produced by shaft rotation. Kinetic grinding tests
particles. were completed in a total of 64 min. Power input was constantly mon-
Alumina beads and ceramic balls correspond to the moderate-cost itored throughout this period, and eight representative samples were
grinding media. Both have the same particle size range of 250 × 340 taken at the 0.5th, 1st, 2nd, 4th, 8th, 16th, 32nd, and 64th minute for
μm, similar mechanical properties, density, and uniform size and sphe- post-grinding particle size analysis. Once the tests were finished, the
ricity. Potters silica beads and granular sand are the low-cost grinding material was discharged and screened to separate the coal from the
media. The granular sand has nearly the same particle size range grinding media using a Sweco Separator (Sweco, Florence, Kentucky).
(i.e., 250 × 350 μm) as alumina and ceramic media, but silica beads The sieve aperture size chosen during the screening process equals the
span a broader size range of 250 × 420 μm. Both silica beads and sand grinding media's bottom size so that the grinding media is retained on
have similar properties and density since they are silica-based top while the coal sample passes through the openings.

Table 3
Elemental compositions of the coal ash residual and the grinding media materials.

Materials Unit, %

LE Si Ca Mg Al Fe S K Al2O3 SiO2 Other Total

Ash Residue 55.49 21.36 0.78 ND 13.95 4.53 0.76 2.01 1.12 100.00
Silica 63.82 28.36 5.59 1.85 0.13 0.13 0.07 0.04 99.59
Sand 60.02 31.03 1.73 0.48 3.02 2.23 0.17 0.90 99.59
Ceramic Balls >90 ≥3 ≥3 99
Alumina Beads 99.9 99.9

Note: LE denotes light elements with an atomic number less than 12.

3
F. Patino, D. Talan and Q. Huang Powder Technology 403 (2022) 117394

2.1.3. Power measurement and specific energy calculation

Power consumption at the 0.5th, 1st, 2nd, 4th, 8th, 16th, 32nd, and
64th minute was recorded during the kinetic grinding tests. Subse-
quently, the specific energy (SE) was calculated based on power input
readings obtained from the tests and the unloaded power input,
which was determined while operating under “free-load” conditions.
This no-load power draw represents the energy required to run the
shaft; therefore, it should be deducted to calculate the specific energy
required to produce the reduction in particle size. Eq. (1) was used to
calculate the specific energy [53]:
Rt
0 ðN ðτ Þ−N 0 Þdτ
SE ¼ ð1Þ
mp

where mp is the mass of solids in the pulp, N(τ) is the power at the
time τ, and N0 is the no-load power. As seen, the specific energy is di-
rectly related to feed rate, which directly impacts the product size and
is commonly expressed in kilowatts-hour per short ton.

2.1.4. Particle size, shape, and surface morphology analyses

Particle size distributions of feed and product samples were ob-
tained using the same CILAS 1190 Particle Size Analyzer. With this
laser sizing technique, the average particle volume is first determined
and then converted to the particle size diameter. Multiple replicate
tests per sample were conducted to ensure the results were statistically
reliable. On the other hand, grinding media wear was evaluated through
a shape analysis before and after the grinding tests using a CILAS Micro-
scope attached to the CILAS 1190 Particle Size Analyzer.
The shape descriptors chosen to analyze the morphological variation
are the particle circularity ratio, sphericity, and aspect ratio. These pa-
rameters seek to measure a particle's degree of roundness compared
to a perfect sphere based on its dimensions, perimeter, area, and vol-
Fig. 2. Laboratory scale stirred mill employed throughout the study. ume. The circularity ratio was calculated based on the following equa-
tion defined by Cox [20]:

4πA
Cicularity Ratio ðCRÞ ¼ ð2Þ
P2
Table 4
Mill specification data. where A is the area and P the perimeter of the particle. According to
Tank Volume, L 9.5
Grace and Ebneyamini [28], sphericity is defined as the quotient of the
Tank Height, mm 200.66 surface area of a sphere of the same volume as the particle over the ac-
Tank Inner Diameter, mm 205.74 tual exterior surface of the particle:
Tank Outer Diameter, mm 279.40
Work Capacity, L 4.2 Surface area of a sphere of the same volume as the particle
Height, mm 1168 Sphericity ðϕÞ ¼
Bench Space, mm 660 × 1270 Actual exterior surface of the particle
Weight, kg 340 ð3Þ
Operational speed, rpm 340

Table 5
Detailed test conditions for stirred mill kinetic grinding tests.

Parameter Unit Alumina Beads Granular Sand Ceramic Balls Silica Beads

Slurry/Pulp Mass g 1360 1360 1360 1360

Solids Concentration % 30 30 20, 30, 40 20, 30, 40
Dispersant kg/ton NA NA 2, 8, 14 2, 8, 14
Stirrer Speed RPM 340 340 340 340
Grind Time min. 64 64 64 64
40 × 210
Media Size μm 250 × 340 250 × 350 250 × 340 250 × 420
420 × 850
Media Mass g 5194 5194 5194 5194
Bed Porosity % 40 40 40 40
Ball Filling Volume % 23.98 34.39 24.30 36.45
Powder Filling Volume % 3.98 3.98 3.98 3.98
Interstitial Filling % 41.46 28.91 40.91 27.28
Feed Size, F80 μm 24.36 24.36 24.36 24.36

Note: Powder filling volume and interstitial filling were calculated based on 30% solids concentration by weight.

4
F. Patino, D. Talan and Q. Huang Powder Technology 403 (2022) 117394

The aspect ratio measures the relationship between a particle's than the upper limit of size interval i resulting from the breakage
width and height. For ellipses, the aspect ratio would be the relationship of size interval j. Rajamani and Herbst [67] proposed that the
between the major and minor axis, and it would be equal to one when experimentally observed selection and breakage functions can be
both axes are the same [75]. Eq. (4) defines the aspect ratio: described with simple functional forms:
(  pffiffiffiffiffiffiffiffiffiffiffiffi)
Major axis lenght of approximate ellipse k xi xiþ1
Aspect Ratio ðARÞ ¼ ð4Þ Si ¼ S1 exp − ∑ ζ i ln pffiffiffiffiffiffiffiffiffiffi i ¼ 1, 2 . . . . . . :ðn−1Þ ð9Þ
Minor axis lenght of approximate ellipse i¼1 x1 x2

 α 2  α3
xi xi
2.2. Numerical grinding simulation Bi,j ¼ α 1 þ ð1−α 1 Þ ð10Þ
x jþ1 x jþ1

Process modeling is commonly used in preliminary and detailed where ζi are parameters of logarithmic polynomial functional form
grinding circuit design, comminution performance assessment, and for the selection function, whereas α1, α2, and α3 are parameters of
geo-metallurgy. Although the energy-based models, as the first empiri- functional form for Bi,j.
cal models developed, are convenient to predict any desired percent Moreover, many researchers have developed various mathematical
passing size based on the specific energy input, they do not deliver the equations to describe the breakage function [5,15,26,42,44,47]. Follow-
complete particle size distribution. For this reason, phenomenological ing Austin et al. [5], Klimpel and Austin's formula comprises a general
models, such as the population balance modeling (PBM), are used to expression [32,44]:
simulate the product's behavior in equipment such as high-pressure
grinding mills and vertical mills. Population balance modeling (PBM)   n1 "  2 #n2 "  3 #n3
di di d
also has widespread use in designing, controlling, and optimizing com- Bðdi Þ ¼ 1− 1− 1− 1− i ð11Þ
dj dj dj
minution circuits [23]. Therefore, this study simulated laboratory com-
minution processes in MATLAB on the basis of population balance
where,
modeling.
dj: Original size being broken
di: Size of the progeny fragment of breakage
2.2.1. Population balance modeling (PBM)
n1–n3: Constants depending on the particle shape and flaw density
Population balance modeling (PBM) is governed by two size-
within the particles
discretized kinetic functions, i.e., the selection function (or the specific
B(di): Cumulative mass fraction finer than di where dj > di > 0.
rate of breakage) and the breakage distribution function. The specific
Eq. (12) shown below further presents the general energy-based
rate of breakage function parameters describe the rate at which the par-
differential equation in vector notation [44]:
ticle is broken from a particular discretized size class, whereas the
breakage function parameters indicate the fraction of the primary  
dm E  
breakage product in the feed size class j, which appears in the smaller ¼ −ðI−BÞSm E ð12Þ
dE
size interval i [18]. In the formulation of PBM, the particulate assembly
that undergoes breakage is divided into several narrow size intervals, Solving Eq. (12) leads to a derived question as followings:
e.g., standard √2 Tyler sieve intervals, with the top size interval 1 and
   
up to the nth interval [67]. Therefore, the rate of the breakage of the m E ¼ exp −½I−BSE mð0Þ ð13Þ
size i (Si) into a smaller size in a fully mixed batch mill is [5]:
where,
d½mi ðt Þ mi: Mass fraction of size class i.
¼ −Si mi ðt Þ ð5Þ
dt m(Ē): Matrix (n x 1) of mass fractions resulting from a specific
energy input Ē.
The above equation can be further integrated to the following: I: Identity matrix (n x n).
B: Breakage function matrix (n x n) with elements bi,j.
mi ðt Þ ¼ mi ð0Þe−Si t ð6Þ
S: Selection function matrix (n x n) with diagonal SEi , which is the
specific selection function and is essentially independent of mill
where mi(t) and mi(0) are the mass fraction of size class i, at times t design and operating conditions.
and 0, respectively. The equation is also referred to as the first-order rate Ē: Specific energy input.
model of grinding. Following the formulas proposed by Austin et al. [5], m(0): Matrix (n x 1) of mass fraction of feed.
the variation of the specific rate of breakage Si with particle size is given n: Number of size intervals
below:

Si ¼ α T ð7Þ 2.2.2. Particle swarm optimization (PSO)

X αi
The method for optimizing non-linear functions using particle
swarm optimization was first developed by Kennedy and Eberhart
where Xi is the upper limit of size interval i and αT and α are model
[84]. Its fundamentals are based on artificial life (A-life), bird flocking,
parameters depending on the mill conditions and the material
fish schooling, and swarming theories. The following steps provide an
properties, separately. On the other hand, the breakage function is
overall review of the application of particle swarm optimization [19]:
usually represented in cumulative form, Bi,j, as:
1. Initialize a population of particles with random positions and veloci-
i ties in the n-dimensional space using a uniform probability distribu-
Bi,j ¼ ∑ bk,j ð8Þ
k¼n tion.
2. Evaluate the fitness value (the number that represents the goodness
where bi,j = Bi,j − Bi+1,j, represents the mass fraction of the of a given solution) of each particle in the swarm.
fragmented products with size interval i out of the breakage of size 3. Compare each particle's fitness value with the current pbest (posi-
interval j. Bi, j is the sum of the fractions of materials with a size less tion in parameter space of the best fitness returned for a specific

5
F. Patino, D. Talan and Q. Huang Powder Technology 403 (2022) 117394

Fig. 5. Specific energy input of each type of media material. Error bars represent one stan-
Fig. 3. Particle size distribution of feed and ground products obtained by each type of dard deviation of three replicate tests.
grinding media materials. Error bars represent one standard deviation of twelve replicate
tests.

Fig. 4. Power input and specific energy required by each type of media materials: (a) Ceramic balls, (b) Alumina beads, (c) Silica beads, and (d) Granular sand.

6
F. Patino, D. Talan and Q. Huang Powder Technology 403 (2022) 117394

In this study, the objective function shown in Eq. (14) was used to
determine the best parameter, which corresponds to the parameter
that returned the minimum objective function.

* 1     T      
θ x ¼ m E model −m E test m E model −m E test ð14Þ
M

where,
θ: Objective function.
*
x : Parameter vector.
 
th
(a) 0.5 minute th
(b) 64 minute m E model : Vector of the product size distribution from the simula-
tion.
 
Fig. 6. Degradation process of granular sand. Samples taken at the 0.5th and 64th minute m E test : Vector of the product size distribution from the test.
during the kinetic grinding test.  
M: Number of predictions/elements in m E :

3. Results and discussion

particle). If the current value is better than pbest, then set it as the
new pbest with its corresponding location. 3.1. Laboratory test results
4. Compare the fitness evaluation with the population's overall previ-
ous best. If the current value is better than gbest (position in param- 3.1.1. Effect of grinding media material
eter space of the best fitness returned for the entire swarm), reset Particle size distribution of various 64-min grind products resulting
gbest to the current particle's array index. from different grinding media is shown in Fig. 3. As seen, the finest
5. Change the particle's velocity and position. product size was obtained by silica beads, with a product size (P80) of
6. Loop to step 2. until a stopping criterion is met, which is the maxi- 5.9 μm. It represents a reduction ratio of 4.2:1 considering the feed
mum number of iterations. size of 24.4 μm. Ceramic balls and alumina beads showed a similar

Fig. 7. Shape analysis of the grinding media materials before and after stirred mill grinding tests. Error bars represent one standard deviation of multiple measurements.

7
F. Patino, D. Talan and Q. Huang Powder Technology 403 (2022) 117394

distribution, achieving a P80 of 12.8 and 15.5 μm, respectively, with a

corresponding reduction ratio of 1.9:1 and 1.6:1. Media degradation
was observed for granular sand, as indicated in Fig. 6. Due to the co-
existence of attritted sand slimes and coal fines, the separation of the
coal particles from the grinding media was unachievable. Therefore,
the particle size distribution for the test conducted with granular sand
was not generated. (a) Ceramic balls – Before grinding (b) Ceramic balls – Post grinding
Power readings and corresponding specific energy values obtained
for each test are shown in Fig. 4. The highest energy was consumed by
granular sand, followed by silica beads, ceramic balls, and alumina
beads in descending order. Specific energy values of 579 and 309
kWh/ton were obtained for granular sand and silica beads, and values
of 109 and 129 kWh/ton were attained for alumina beads and ceramic
balls, separately (Fig. 5). The latter two materials demand around 65%
less energy than silica beads and show a second-order polynomial
(c) Alumina beads – Before grinding (d) Alumina beads – Post grinding
trendline. Differently, silica beads and granular sand demonstrate a log-
arithmic trendline and similarity in energy consumption.
Fig. 6 illustrates the wear process of granular sand throughout the ki-
netic grinding test. Although the grinding media are selected to produce
fine product size and provide economic grind, the media will wear and
have an impact on the process efficiency [33]. Media wear is a normal
process that occurs by interacting media with each other, mineral parti-
cles, and mill chamber; however, severely worn media may significantly
affect the results. According to the samples taken at the 0.5th and 64th (e) Silica beads – Before grinding (f) Silica beads – Post grinding
minute, this phenomenon started from the beginning of the grinding
process and rendered a physical separation by screening unfeasible. Fig. 8. Image analysis of grinding media materials submitted to grinding stress.
High energy consumption and rheological changes were also noticed
at the end of the test, revealing the over-grinding of granular sand
throughout the grinding test. A specific surface area (SSA) generation representing a reduction ratio of 7.3:1. Compared to regular-sized silica
process occurs during comminution, modifying the pulp rheology due beads, i.e., P80 = 5.9 μm, finer product size and lower energy
to water adsorption on the particles' surface [12]. This phenomenon sig- consumption were achieved by employing a greater media size range.
nificantly increases pulp density, viscosity, and heat generation, which These findings are consistent with the results obtained by Mankosa
turns the coal pulp into a solid paste. et al. [48] and Guo et al. [30], who also observed an improvement in
Shape analysis was performed on the media beads to determine the the yield of fine ground products by increasing the grinding media
wear rate of each material after being submitted to grinding stress. For size. However, the particle size distribution of ground products from
ceramic balls, the initial circularity ratio, sphericity, and aspect ratio fine silica beads is not shown in Fig. 9. Separating the coal sample
were 0.90, 0.97, and 0.98, separately, which are very similar to those ob- from the grinding media material was not feasible due to their similar
tained for alumina beads (Fig. 7a and Fig. 7b). For silica beads, these in- sizes. A noticeable amount of unground coal particles was retained on
dicators were around 0.90, 0.98, and 0.99, respectively (Fig. 7c). This top of the sieve with the grinding media, implying the necessity of an
indicates that grinding media particles are quasi-perfect spheres. Post additional separation process to recover coarse coal material. Hence, it
grinding ceramic balls and alumina beads present a similar trend. The was concluded that fine silica beads with a size range of 40 to 210 μm
sphericity and aspect ratio dropped by less than 0.30%, while the circu-
larity ratio slightly increased. On the other hand, silica beads show a
higher shape variation than the other materials, accounting for up to
0.63% in sphericity and 0.23% in aspect ratio. Unlike ceramic balls and
alumina beads, the circularity ratio for silica beads also dropped by
0.19%. Becker and Schwedes [8] state that the wear rate depends pri-
marily on the hardness of the beads, explaining the higher wear rate ob-
tained using silica beads since its hardness is 6 on the Mohs scale, while
ceramic balls and alumina beads' hardness is 9. However, all these var-
iations observed are within one standard deviation of the replicate tests,
indicating minimum or no degradation of all media materials after per-
forming the grinding tests. Moreover, no media degradation was visu-
ally observed, as shown in Fig. 8.

3.1.2. Effect of grinding media size

Given the results obtained for specific energy, particle size, and
shape analyses, silica beads provided the optimum grinding perfor-
mance in terms of product size (i.e., 5.9 μm) and energy consumption
(i.e., 309 kWh/ton). Therefore, the effects of grinding media size were
addressed using silica beads with two other size ranges. Their size
ranges span 40 × 210 μm and 420 × 850 μm, denoted as fine silica
beads and coarse silica beads, respectively.
This study displayed an inverse relationship between the coal prod-
uct and grinding media size; the coarser media size tends to generate a Fig. 9. Particle size distribution of feed and ground products obtained using regular and
finer grind product. Coarse silica beads produced a grind size of 3.3 μm, coarse silica beads. Error bars represent one standard deviation of twelve replicate tests.

8
F. Patino, D. Talan and Q. Huang Powder Technology 403 (2022) 117394

Fig. 10. Power input and specific energy required by silica beads with different sizes: (a) Fine silica beads 40 × 210 μm, and (b) Coarse silica beads 420 × 850 μm.

Fig. 11. Specific energy input of silica beads with different sizes. Error bars represent one Fig. 13. Particle size distribution of feed and ground products obtained using coarse silica
standard deviation of three replicate tests. beads (CSB) and ceramic balls (CB) with different solids concentrations by weight. Error
bars represent one standard deviation of twelve replicate tests.

Fig. 12. Shape analysis of silica beads with different size ranges before and after stirred mill grinding tests. Error bars represent one standard deviation of multiple measurements.

9
F. Patino, D. Talan and Q. Huang Powder Technology 403 (2022) 117394

Fig. 14. Power input and specific energy consumed by coarse silica beads and ceramic balls with different solids concentrations by weight.

are not suitable for ultra-fine grinding mainly due to insufficient grind- Altun et al. [3], the grinding media size plays a vital role in grinding op-
ing, the complexity of separation, and additional steps required to fully erations, and its proper selection can positively impact energy effi-
separate the coal sample from the grinding media. ciency. Likewise, unsuitable grinding media sizes might dramatically
As shown in Fig. 10, both size ranges exhibit a similar trend in power increase the power draw [39].
consumption and specific energy; however, coarse silica beads Shape analyses of silica beads after grinding with additional sizes are
(Fig. 10b) required less energy than fine beads. The specific energy presented in Fig. 12. The data indicate that fine silica beads had an initial
input of 283 kWh/ton for coarse silica beads represents around two- circularity ratio, sphericity, and aspect ratio values of around 0.93, 0.92,
thirds of the 421 kWh/ton required by fine beads. These results present and 0.95, respectively. After grinding, the circularity ratio remained al-
a similar trend as those obtained in the previous section (Fig. 4). The most constant while the sphericity and aspect ratio dropped to 0.90
specific energy decreased as the bead size increased, as shown in and 0.94, which accounts for 2.40% and 1.51% reduction, separately.
Fig. 11. A significant saving of 33% can be achieved by increasing the On the other hand, coarse silica beads had an initial circularity ratio of
grinding media size from +40/−210 to +420/−850 μm. According to 0.89, sphericity of 0.95, and aspect ratio of 0.96. After grinding, they

10
F. Patino, D. Talan and Q. Huang Powder Technology 403 (2022) 117394

decreased by 0.26%, 0.85%, and 0.30%, respectively. It seemed that fine

silica beads have higher media wear rates than coarse beads and all
other grinding media materials (Fig. 12a), which may partially explain
the difficult separation between the coal particles and media materials.
However, all the variations are within the standard deviations gener-
ated from replicate tests.

3.1.3. Effect of solids concentration

In addition to media size, the influence of solids concentration on
ultra-fine coal grinding was also studied using coarse silica beads and
ceramic balls based on the fine product size or low specific energy the
two media generated. A total of six kinetic grinding tests were per-
formed with varying solids concentrations ranging from 20 wt% to 40
wt%. Using silica beads, the product sizes of 2.8, 3.3, and 4.1 μm were
achieved for 20, 30, and 40 wt% concentrations, respectively. Similarly,
the product sizes of 9.9, 12.8, and 13.0 μm were obtained for ceramic
balls, separately, as shown in Fig. 13. Considering the feed size of 24.4
μm and the product sizes obtained, the reduction ratio ranged between
5.9:1 and 8.6:1 for coarse silica beads, and 1.9:1 and 2.5:1 when ceramic
grinding media was employed. According to the results generated, re- Fig. 15. Specific energy input of coarse silica beads and ceramic balls with different solids
ducing solids concentration enhanced the ultra-fine grinding perfor- concentration by weight. Error bars represent one standard deviation of three replicate
mance. These findings are also supported by Bernhardt et al. [9], who tests.
studied the effect of suspension properties on ultra-fine grinding. It
was found that the solids concentration significantly influences the en-
ergy utilization and final product size. Suspensions with low solids con-
centrations provide the low fluid resistance needed to achieve finer high (>80 wt%) concentrations as a result of de-agglomeration and in-
product sizes while requiring higher power consumption. Partyka and creased density and viscosity, respectively [74].
Yan [54] also reported the same grind curves during their experimental
work using a horizontal ball mill and quartz andesite rock as a feedstock.
Moreover, according to the mechanistic mill model developed by de 3.1.4. Effect of viscosity modifier
Oliveira et al. [21] using a vertical stirred mill, the increase in solids One of the shortcomings of the ultra-fine grinding processes is the
concentration resulted in lower breakage rates due to the ore particles presence of excessive fines. They may lead to the formation of agglom-
involved in collision events are submitted to grinding stress as mono- erations or flocculation and demands higher energy consumption. The
layer beds. use of a viscosity modifier can enhance the comminution processes sig-
Overall, establishing low solids concentrations seems to be an alter- nificantly by changing the surface of the particles and improving the
native to reduce the grind product size; however, it is imperative to con- slurry flowability [34]. In this study, three different doses of sodium sil-
sider the substantial drop in throughput and the significant increase in icate (i.e., 2, 8, and 14 kg/ton) were employed to modify the rheological
water demand that this practice represents [81]. Generally, industrial behavior of the pulp and study its effect on ultra-fine grinding perfor-
processing plants seek to operate their grinding equipment with high mance.
solids content as this allows to improve capacity, handle more materials, A finer product size was obtained with a high dosage of the viscosity
and boost ore production [25]. modifier, as shown in Fig. 16. Silica grinding media reduced the
On the other hand, the graphs representing the power draw and spe-
cific energy are shown in Fig. 14. The displayed characteristic trend for
silica beads is consistent with the results previously obtained (Fig. 14a,
Fig. 14c, and Fig. 14e). Nevertheless, ceramic balls present either an
order three or four polynomial trend. These trendlines generally offer
one or multiple hills or valleys [36]. As evidenced in Eq. (1), low specific
energies can be achieved with high solids concentrations due to their in-
verse relationship. However, the best grinding result in product size was
obtained with 20 wt% solids concentration using coarse silica beads.
As seen in Fig. 15, specific energy input considerably increased as the
solids concentration was reduced. Lowering this concentration by half
(from 40 wt% to 20 wt%) resulted in tripling energy consumption
from 155 kWh/ton to 432 kWh/ton using silica beads. The energy con-
sumption increased almost five times for ceramic balls, elevating from
39 to 198 kWh/ton. Under the standard test conditions (X = 30 wt%),
the specific energy input was around 283 and 129 kWh/ton for silica
beads and ceramic balls, respectively. Although both grinding media
materials showed a decreasing trendline, ceramic balls required signifi-
cantly less energy than silica beads. However, this power saving is
reflected in the lower reduction ratio due to the coarser product size.
Similar results were found by Simpson et al. [74] during their study of
the effect of solids concentration on the particle size distribution of Fig. 16. Particle size distribution of feed and ground products obtained using coarse silica
deagglomerated barium titanate in stirred media mills. In addition, the beads with different doses of sodium silicate as a viscosity modifier. Error bars represent
researchers found peaks of specific energy at low (<30 wt%) and very one standard deviation of twelve replicate tests.

11
F. Patino, D. Talan and Q. Huang Powder Technology 403 (2022) 117394

Fig. 17. Power input and specific energy consumed by coarse silica beads and ceramic balls with different doses of sodium silicate as a viscosity modifier.

cumulative 80% passing size from 24.4 μm to 2.7 μm by adding 14 kg/ton The power draw readings and specific energy obtained during the
of dispersant. This P80 value slightly increased to 3.1 and 4.9 μm when study are shown in Fig. 17. Both grinding media materials show their
the dosage was reduced to 8 and 2 kg/ton, respectively. On the other characteristic trend lines that resemble those obtained previously in
hand, the finest product size obtained using ceramic balls with 14 kg/ the study of the effect of solids concentration. As expected, the power
ton of sodium silicate was 11.8 μm, followed by the 12.9 μm obtained consumption and specific energy increased as the dispersant dosage in-
using 8 kg/ton and 13.4 μm using 2 kg/ton of sodium silicate. Thus, the creased (Fig. 18). Under standard conditions, the power demand using
use of this grinding aid in stirred milling with coarse silica beads coarse silica beads with 14 kg/ton of sodium silicate was 270 kWh/ton
attained reduction ratio values between 4.9:1 and 9.1:1. Nonetheless, and dropped almost linearly to 238 and 213 kWh/ton when employing
these values were around 1.8:1 and 2.1:1 when ceramic grinding 8 and 2 kg/ton of dispersant, respectively. A linear trend was also ob-
media was utilized. These results agree with the findings reported in served using ceramic balls, in which the specific energy dropped from
Mankosa and his coworkers' study [49], in which an improvement 137 kWh/ton to 119 and 117 kWh/ton, respectively. The higher energy
was observed in the mean product particle diameter by utilizing consumption by dispersants is due to a lower viscosity maintained dur-
various types of dispersants. ing the comminution process [41]. Therefore, the breakage rate

12
F. Patino, D. Talan and Q. Huang Powder Technology 403 (2022) 117394

Fig. 18. Specific energy input of coarse silica beads and ceramic balls employing different Fig. 19. Overall specific energy input as function of product size. Error bars represent one
doses of sodium silicate as a viscosity modifier. Error bars represent one standard devia- standard deviation of three replicate tests. *It was not possible to obtain the product size of
tion of three replicate tests. tests using granular sand (GS) and fine silica beads (FSB), therefore a P80 of 1 μm was used
to represent the points.

increases and demands more energy to achieve a finer product size [34].
Although the gap between the higher and lower specific energy ob- were modified by changing the percent of solids or adding viscosity
tained seems insignificant, it may represent a significant difference modifiers such as sodium silicate. Finally, the inverse relationship
when dealing with very high energy consumption for extended periods. of product size and energy requirement is displayed in Fig. 20,
However, it is essential to determine an optimal dosage as an excessive where P 80 and specific energy fluctuate in the opposite direction
amount of the dispersant can cause slurry destabilization [17,63]. On the regardless of tests conditions.
other hand, an insufficient amount of the dispersant may cause floccu-
lation by bridging mechanisms and increasing the viscosity of the slurry 3.2. Numerical simulation of kinetic grinding
[34].
As a summary, Table 6 shows the output obtained throughout The grinding simulation was performed using Population Balance
the grinding study in terms of feed size, product size, reduction Modeling (PBM) on the basis of the Breakage Function, Selection Func-
ratio, and specific energy. Furthermore, Fig. 19 illustrates the tion, and Particle Swarm Optimization (PSO) algorithm in MATLAB. The
trend lines obtained when the rheological properties of the slurry input for the simulation was the cumulative percent passing of seven
different particle diameters (i.e., 1.2, 2, 5, 15, 25, 50, and 100 μm) from
the feed and actual product particle size distributions of each test, as
Table 6 well as their corresponding specific energy. These variables are denoted
Overall stirred mill grinding test results summary. as m(Ē), m (0), and Ē, respectively. In addition, the in-built particle
swarm optimization algorithm was used to simulate the batch grinding.
Test Feed P80 Reduction Specific Energy
(μm) (μm) Ratio (kWh/ton) This arrangement involved the lower and upper bounds of the selection
Ceramic Balls 24.36 12.82 1.90 128.76
Alumina Beads 24.36 15.54 1.57 109.49
Silica Beads 24.36 5.86 4.16 308.97
Granular Sand 24.36 – – 579.40
Fine Silica Beads 24.36 – – 420.75
Regular Silica Beads 24.36 5.86 4.16 308.97
Coarse Silica Beads 24.36 3.34 7.29 283.22
Coarse Silica Beads 20 wt% 24.36 2.84 8.57 431.69
Coarse Silica Beads 30 wt% 24.36 3.34 7.29 283.22
Coarse Silica Beads 40 wt% 24.36 4.14 5.88 155.02
Ceramic Balls 20 wt% 24.36 9.95 2.45 197.96
Ceramic Balls 30 wt% 24.36 12.82 1.90 128.76
Ceramic Balls 40 wt% 24.36 13.04 1.87 39.22
Coarse Silica Beads Disp. 2
24.36 4.95 4.92 212.73
kg/ton
Coarse Silica Beads Disp. 8
24.36 3.14 7.76 238.27
kg/ton
Coarse Silica Beads
24.36 2.68 9.07 269.74
Disp. 14 kg/ton
Ceramic Balls Disp. 2
24.36 13.40 1.82 117.54
kg/ton
Ceramic Balls Disp. 8
24.36 12.93 1.88 119.03
kg/ton
Ceramic Balls Disp. 14
24.36 11.82 2.06 137.00 Fig. 20. Trendline of specific energy and product size as a function of various grinding
kg/ton
media materials and test conditions.

13
F. Patino, D. Talan and Q. Huang Powder Technology 403 (2022) 117394

Table 7 function parameters (SEi , ζ1, ζ2) and breakage function parameters (α1,
Confidence interval and optimum values of selection and breakage functions. α2, α3). Their corresponding optimum values obtained from the
Parameter Lower Bound Upper Bound Optimum Optimum CB numerical simulation for all tests conducted using both silica and
CSB ceramic grinding media materials are shown in Table 7.
SEi 0.0 0.1 0.0156 0.0081 The numerical simulation for silica beads and ceramic balls was car-
ζ1 0.0 5.0 0.0016 0.0091 ried out separately. The mean squared error (MSE) for silica media was
ζ2 −4.0 1.0 −0.2766 −0.2827 reduced to 0.003207 (Fig. 21), while for ceramic media, this value
α1 0.0 2.0 0.9392 1.4642
reached 0.002891 (Fig. 22). The particle swarm optimization was initi-
α2 0.0 0.5 0.4976 0.4971
α3 10.0 25.0 15.9953 18.8091 ated with a stopping criterion of 40 iterations.
The predicted particle size distribution of the ground product ob-
tained from the simulations was compared with the actual product

Fig. 21. Mean squared error (MSE) minimization for tests conducted with coarse silica beads using particle swarm optimization (PSO).

Fig. 22. Mean squared error (MSE) minimization for tests conducted with ceramic balls using particle swarm optimization (PSO).

14
F. Patino, D. Talan and Q. Huang Powder Technology 403 (2022) 117394

Fig. 23. Simulation results using population balance modeling for various grinding media materials under standard test conditions.

and feed size distributions. Fig. 23 shows the models in which grinding on the breakage function, selection function, and particle swarm op-
media material was studied under standard test conditions (i.e., 30 wt% timization algorithm.
solids concentration, no dispersant, and similar media sizes in a range of
250 × 420 μm), while Figs. 24 and 25 display the patterns obtained 4. Conclusions
while exploring the effect of solids content and viscosity modifiers. A
good model fit was observed between the model data and the actual The application of stirred mills has continuously gained importance
particle size distribution based on the optimized function parameters. as they can achieve finer product sizes when compared to conventional
To investigate the grinding kinetics of the coal particles, five repre- mills. Stirred mills have several advantages over traditional grinding
sentative sizes, i.e., 0.1, 1, 2, 5, and 15 μm, were selected, and their re- mills, such as requiring less floor space, lower maintenance cost, fewer
maining weights (%) as a function of grinding time were plotted for moving parts, and higher controllability. However, ultra-fine coal pro-
ceramic balls and silica beads under standard test conditions. As seen duction using stirred milling is still a high energy-intensive process.
in Fig. 26, the breakage rate slows down with increasing grinding time Therefore, it is vital to optimize operating conditions to achieve micron-
and deviates from the first order rate model of grinding, which has ized particles at low energy consumption, which are now needed for
been seen in many other studies [10,11,58,59]. The findings in the coal many applications. This study was designed to assess the influence of
grinding kinetics are also supported by Rajamani and Guo [66], who grinding materials on coal grinding performance and investigate the ef-
stated that wet grinding is a nonlinear process by nature. fect of grinding media size, solids concentration, and the use of viscosity
The Pearson correlation tests performed on the bivariate data sets modifiers. The results were also assessed using characterization studies,
are shown in Figs. 27 and 28. This test measures the strength and direc- grinding modeling, and simulation. Specific findings of this study in-
tion of a linear correlation [64]. The obtained Pearson correlation coeffi- clude:
cients (r) above 0.98 indicate a robust correlation between the 1. Low-cost grinding media, such as silica beads, developed a superior
simulation and experimental data, according to [85]. Moreover, grinding performance regarding the final grind size and power con-
the coefficients of determination (R2 ) indicate that the variability sumption compared to moderate-cost media, such as ceramic balls
of the predicted models explains more than 97% of the variability and alumina beads.
of the experimental data. Considering that both variables are line- 2. Finer product size and lower energy consumption can be obtained by
arly correlated, it can be concluded that population balance model- increasing the media size range of silica beads from 250 × 420 μm to
ing provides an accurate forecast of product size distributions based 420 × 850 μm. This means that P80 can be further reduced from 5.9

15
F. Patino, D. Talan and Q. Huang Powder Technology 403 (2022) 117394

Fig. 24. Simulation results using population balance modeling for coarse silica beads and ceramic balls with different solids concentrations.

μm to 3.3 μm while achieving a reduction in specific energy from 309 4. Fine silica beads (40 × 210 μm) have not been found suitable for
kWh/ton to 283 kWh/ton. Moreover, a coarser media size also ultra-fine grinding due to grinding media degradation and the com-
facilitates coal separation from grinding media through wet plexity of separation. Additional steps are required to achieve a clear
screening. separation, which would increase the cost of production and demand
3. The use of granular sand was ruled out due to its degradation during advanced classification techniques.
grinding, which would make any process infeasible due to high 5. Minimum or no grinding media degradation was detected by con-
maintenance costs and additional separation stages. ducting shape analysis on the media grains before and after being

16
F. Patino, D. Talan and Q. Huang Powder Technology 403 (2022) 117394

Fig. 25. Simulation results using population balance modeling for coarse silica beads and ceramic balls with different dosages of sodium silicate as a viscosity modifier.

submitted to grinding stress. All media materials showed a variation concentration based on production goals since throughput may be
in circularity ratio, sphericity, and aspect ratio less than 1% post considerably affected.
grinding, indicating their durability. 7. The use of dispersants such as sodium silicate represents an effective
6. Reducing solids concentration from 40% to 20 wt% is an alternative to mechanism to produce ultra-fine materials at low energy consump-
producing finer product sizes. However, the energy consumption can tion. Its use avoids agglomeration and enhances grinding performance
increase up to 5 times depending on the grinding media material em- by reducing slurry viscosity. Nevertheless, the excessive use of viscos-
ployed. Therefore, it is imperative to select the adequate solids ity modifiers can affect downstream processes such as flocculation.

17
F. Patino, D. Talan and Q. Huang Powder Technology 403 (2022) 117394

Fig. 26. Remaining weight of selected sizes as a function of grinding time.

8. Adjusting the rheological properties of pulp using viscosity modifiers

represents a more viable alternative than reducing the solids content.
A dose of 14 kg/ton of sodium silicate can consume up to 37.5% less
power than lowering solids concentration to 20 wt%. Additionally, a
finer product size (P80) is delivered without affecting throughput.
9. Numerical simulation using population balance modeling is an effec-
tive method to simulate grinding and predict product distributions
indirectly. It provides an accurate forecast of the particle size distri-
bution based on the experimental distribution and specific energy
input.

CRediT authorship contribution statement

Francisco Patino: Conceptualization, Investigation, Methodology,

Validation, Writing-original draft. Deniz Talan: Supervision, Validation,
Writing-review and editing. Qingqing Huang: Conceptualization,
Funding acquisition, Investigation, Methodology, Project Administra-
tion, Resources, Software, Supervision, Validation, Writing-review and
Fig. 27. Correlation analysis between simulated and experimental cumulative percent
editing.
passing obtained using silica beads, ceramic balls, and alumina beads.
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.

Acknowledgment

This research was funded by the Society for Mining, Metallurgy and
Exploration (SME) Academic Career Grant. Their financial contribution
and support are greatly appreciated.

References

[1] A.W. Adeniji, Parametric Study of Coal Liberation Behavior Using Silica Grinding
Media, West Virginia University, 2018.
[2] B. Akkaya, I. Toroglu, M. Bilen, Studying the effect of different operation parameters
on the grinding energy efficiency in laboratory stirred mill, Herpetol. Rev. 46 (2015) 1.
[3] O. Altun, H. Benzer, U. Enderle, Effects of operating parameters on the efficiency of
dry stirred milling, Miner. Eng. 43–44 (2013) 58–66, https://doi.org/10.1016/j.
mineng.2012.08.003.
[4] G.S. Anderson, P.A. Bandarian, Improving IsaMillTM energy efficiency through shaft
spacer design, Miner. Eng. 132 (December 2018) (2019) 211–219, https://doi.org/
10.1016/j.mineng.2018.12.018.
Fig. 28. Correlation analysis between simulated and experimental cumulative percent [5] L.G. Austin, R.R. Klimpel, P.T. Luckie, Process engineering of size reduction: ball mill-
passing obtained using ceramic balls, alumina, and silica beads. ing, SME/AIME. (1984).

18
F. Patino, D. Talan and Q. Huang Powder Technology 403 (2022) 117394

[6] L.G. Austin, P.T. Luckie, The estimation of non-normalized breakage distribution pa- [36] J. Henke, Regression analysis, Life Course Res. Social Policies 11 (2020) 275–286,
rameters from batch grinding tests, Powder Technol. 5 (5) (1972) 267–271, https:// https://doi.org/10.1007/978-3-030-36323-9_26.
doi.org/10.1016/0032-5910(72)80030-5. [37] Q. Huang, A. Noble, J. Herbst, R. Honaker, Liberation and release of rare earth min-
[7] S. Baigereyev, G. Guryanov, New method for increase in product fineness in stirred erals from middle kittanning, fire clay, and West Kentucky no. 13 coal sources, Pow-
mills, Arch. Civil Mech. Eng. 19 (3) (2019) 768–778, https://doi.org/10.1016/j.acme. der Technol. 332 (2018) 242–252, https://doi.org/10.1016/j.powtec.2018.03.063.
2019.03.003. [38] Q. Huang, A. Noble, D. Talan, Exploratory investigation on the use of low-cost alter-
[8] M. Becker, J. Schwedes, Comminution of ceramics in stirred media mills and wear of native media for ultrafine grinding of coal, Int. J. Coal Prepar. Utiliz. 00 (00) (2020)
grinding beads, Powder Technol. 105 (1–3) (1999) 374–381, https://doi.org/10. 1–18, https://doi.org/10.1080/19392699.2020.1809387.
1016/S0032-5910(99)00161-8. [39] C.T. Jayasundara, R.Y. Yang, A.B. Yu, Effect of the size of media on grinding perfor-
[9] C. Bernhardt, E. Reinsch, K. Husemann, The influence of suspension properties on mance in stirred mills, Miner. Eng. 33 (2012) 66–71, https://doi.org/10.1016/j.
ultra-fine grinding in stirred ball mills, Powder Technol. 105 (1–3) (1999) mineng.2011.10.012.
357–361, https://doi.org/10.1016/S0032-5910(99)00159-X. [40] J. Jeswiet, A. Szekeres, Energy consumption in mining momminution, Proc. CIRP 48
[10] E. Bilgili, B. Scarlett, Population balance modeling of non-linear effects in milling (2016) 140–145, https://doi.org/10.1016/j.procir.2016.03.250.
processes, Powder Technol. 153 (1) (2005) 59–71, https://doi.org/10.1016/j. [41] P.C. Kapur, T.W. Healy, P.J. Scales, D.V. Boger, D. Wilson, Role of dispersants in kinet-
powtec.2005.02.005. ics and energetics of stirred ball mill grinding, Int. J. Miner. Process. 47 (1–2) (1996)
[11] E. Bilgili, J. Yepes, B. Scarlett, Formulation of a non-linear framework for population 141–152, https://doi.org/10.1016/0301-7516(96)00003-8.
balance modeling of batch grinding: beyond first-order kinetics, Chem. Eng. Sci. 61 [42] D.F. Kelsall, K.J. Reid, Institue Chem. Engineering Symp, Series. AIChE 1965,
(1) (2006) 33–44, https://doi.org/10.1016/j.ces.2004.11.060. pp. 14–20.
[12] N. Blanc, C. Mayer-Laigle, X. Frank, F. Radjai, J.Y. Delenne, Evolution of grinding en- [43] F. Kick, Das Gesetz der proportionalen Widerstände und seine Anwendungen: Nebst
ergy and particle size during dry ball-milling of silica sand, Powder Technol. 376 Versuchen über das Verhalten verschiedener Materialien bei gleichen
(2020) 661–667, https://doi.org/10.1016/j.powtec.2020.08.048. Formänderungen sowohl unter der Presse als dem Schlagwerk. A. Felix, 1885.
[13] F.C. Bond, Third theory of comminution, Min. Eng. 4 (1952) 484. [44] R.R. Klimpel, L.G. Austin, The back-calculation of specific rates of breakage and non-
[14] S. Breitung-Faes, A. Kwade, Mill, material, and process parameters – A mechanistic normalized breakage distribution parameters from batch grinding data, Int. J. Miner.
model for the set-up of wet-stirred media milling processes, Adv. Powder Technol. Process. 4 (1) (1977) 7–32, https://doi.org/10.1016/0301-7516(77)90028-X.
30 (8) (2019) 1425–1433, https://doi.org/10.1016/j.apt.2019.04.013. [45] M. Lewitt, Opportunities for Fine Coal Utilisation, 2011.
[15] S.R. Broadbent, T.G. Callcott, A matrix analysis of processes involving particle assem- [46] J. Liu, X. Bai, J. Hao, H. Wang, T. Zhang, X. Tang, S. Wang, Y. He, Efficient liberation of
blies, Philos. Trans. R. Soc. A249 (1956) 99–123. electrode materials in spent lithium-ion batteries using a cryogenic ball mill, J. Envi-
[16] R.J. Charles, Energy-Size Reduction Relationships in Comminution, Trans. AIME 9 ron. Chem. Eng. 9 (5) (2021) 106017 https://doi.org/10.1016/j.jece.2021.106017.
(1957) 80–88. [47] A.J. Lynch, Mineral Crushing and Grinding Circuits, their Simulation, Optimisation,
[17] V. Chipakwe, P. Semsari, T. Karlkvist, J. Rosenkranz, S.C. Chelgani, A critical review on Design and Control, Elsevier Scientific Publishing Company, 1977.
the mechanisms of chemical additives used in grinding and their effects on the [48] M.J. Mankosa, G.T. Adel, R.H. Yoon, Effect of media size in stirred ball mill grinding of
downstream processes, J. Mater. Res. Technol. 9 (4) (2020) 8148–8162, https:// coal, Powder Technol. 49 (1) (1986) 75–82, https://doi.org/10.1016/0032-5910(86)
doi.org/10.1016/j.jmrt.2020.05.080. 85008-2.
[49] M.J. Mankosa, G.T. Adel, R.H. Yoon, Effect of operating parameters in stirred ball mill
[18] N. Chimwani, D. Glasser, D. Hildebrandt, M.J. Metzger, F.K. Mulenga, Determination
grinding of coal, Powder Technol. 59 (4) (1989) 255–260, https://doi.org/10.1016/
of the milling parameters of a platinum group minerals ore to optimize product size
0032-5910(89)80084-1.
distribution for flotation purposes, Miner. Eng. 43 (2013) 67–78, https://doi.org/10.
1016/j.mineng.2012.09.013. [50] S. Martins, Size-energy relationship exponents in comminution, Miner. Eng. 149
(January) (2020) 106259 https://doi.org/10.1016/j.mineng.2020.106259.
[19] L.S. dos Coelho, C.A. Sierakowski, A software tool for teaching of particle swarm op-
[51] C.A. Masi, T.A. Schumacher, J. Hilman, R. Dulal, Converting raw coal powder into
timization fundamentals, Adv. Eng. Softw. 39 (11) (2008) 877–887, https://doi.org/
polycrystalline nano-graphite by metal-assisted microwave treatment, Nano-Struct.
10.1016/j.advengsoft.2008.01.005.
Nano-Objects 25 (2021).
[20] E.P. Cox, A method of assigning numerical and percentage values to the degree of
[52] D.B. Mazzinghy, J. Lichter, C.L. Schneider, R. Galéry, J.F.C. Russo, Vertical stirred mill
roundness of sand grains, J. Paleontol. 1 (1927) 179–183.
scale-up and simulation: model validation by industrial samplings results, Miner.
[21] A.L.R. de Oliveira, R.M. de Carvalho, L.M. Tavares, Predicting the effect of operating
Eng. 103–104 (2017) 127–133, https://doi.org/10.1016/j.mineng.2016.11.018.
and design variables in grinding in a vertical stirred mill using a mechanistic mill
[53] S. Ouattara, C. Frances, Grinding of calcite suspensions in a stirred media mill: effect
model, Powder Technol. 387 (2021) 560–574, https://doi.org/10.1016/j.powtec.
of operational parameters on the product quality and the specific energy, Powder
2021.04.057.
Technol. 255 (2014) 89–97, https://doi.org/10.1016/j.powtec.2013.11.025.
[22] V. Deniz, T. Onur, Investigation of the breakage kinetics of pumice samples as de-
[54] T. Partyka, D. Yan, Fine grinding in a horizontal ball mill, Miner. Eng. 20 (4) (2007)
pendent on powder filling in a ball mill, Int. J. Miner. Process. 67 (1–4) (2002)
320–326, https://doi.org/10.1016/j.mineng.2006.12.003.
71–78, https://doi.org/10.1016/S0301-7516(02)00041-8.
[55] C.M. Patel, M. Chakraborty, Z.V.P. Murthy, Enhancement of stirred media mill perfor-
[23] V.P.B. Esnault, H. Zhou, D. Heitzmann, New population balance model for predicting
mance by a new mixed media grinding strategy, J. Ind. Eng. Chem. 20 (4) (2014)
particle size evolution in compression grinding, Miner. Eng. 73 (2015) 7–15, https://
2111–2118, https://doi.org/10.1016/j.jiec.2013.09.040.
doi.org/10.1016/j.mineng.2014.12.036.
[56] G. Peng, S.W. Sharshir, Y. Wang, M. An, D. Ma, J. Zang, A.E. Kabeel, N. Yang, Potential
[24] D.W. Fuerstenau, Grinding aids, KONA Powder Part. J. 13 (13) (1995) 5–18, https:// and challenges of improving solar still by micro/nano-particles and porous materials
doi.org/10.14356/kona.1995006. - A review, J. Clean. Prod. 311 (June 2020) (2021) 127432, https://doi.org/10.1016/j.
[25] P. Gao, W. Zhou, Y. Han, Y. Li, W. Ren, Enhancing the capacity of large-scale ball mill jclepro.2021.127432.
through process and equipment optimization: An industrial test verification, Adv. [57] E. Petrakis, E. Stamboliadis, K. Komnitsas, Identification of optimal mill operating pa-
Powder Technol. 31 (5) (2020) 2079–2091, https://doi.org/10.1016/j.apt.2020.03. rameters during grinding of quartz with the use of population balance modeling,
001. KONA Powder Part. J. 2017007 (2017) https://doi.org/10.14356/kona.2017007.
[26] A.M. Gaudin, T.P. Meloy, Model and comminution distribution equation for single [58] E. Petrakis, V. Karmali, G. Bartzas, K. Komnitsas, Grinding kinetics of slag and effect
fracture, Trans. AIME/SME 223 (1962) 40–43. of final particle size on the compressive strength of alkali activated materials, Min-
[27] Glencore, 2020 Compendium of Technical Papers. 344, 2020 https://www.isasmelt. erals 9 (11) (2019) 714, https://doi.org/10.3390/min9110714.
com/en/download/News/2020ISASMELT Compendium of Papers.pdf. [59] E. Petrakis, K. Komnitsas, Development of a non-linear framework for the prediction
[28] J.R. Grace, A. Ebneyamini, Connecting particle sphericity and circularity, Particuology of the particle size distribution of the grinding products, Min. Metal. Explor. 38 (2)
54 (2021) 1–4, https://doi.org/10.1016/j.partic.2020.09.006. (2021) 1253–1266, https://doi.org/10.1007/s42461-021-00388-w.
[29] D. Graham, Fine grinding applications using the metso vertmill grinding mill and the [60] V.A. Pinchuk, T.A. Sharabura, A.V. Kuzmin, Improvement of coal-water fuel combus-
metso stirred media detritor (SMD), Gold Processing. 38th Cadanian Mineral Pro- tion characteristics by using of electromagnetic treatment, Fuel Process. Technol.
cessors Meeting, 2006. 167 (2017) 61–68, https://doi.org/10.1016/j.fuproc.2017.06.014.
[30] W. Guo, P. Gao, Z. Tang, Y. Han, X. Meng, Effect of grinding media properties and stir- [61] U. Process, Media Milling Advances, Small, November, 2008 12–15.
rer tip speed on the grinding efficiency of a stirred mill, Powder Technol. 382 (2021) [62] P. Prziwara, L.D. Hamilton, S. Breitung-Faes, A. Kwade, Impact of grinding aids and
556–565, https://doi.org/10.1016/j.powtec.2021.01.022. process parameters on dry stirred media milling, Powder Technol. 335 (2018)
[31] A. Gupta, D. Yan, Chapter 10: Stirred mills – ultrafine grinding, Mineral Processing 114–123, https://doi.org/10.1016/j.powtec.2018.05.021.
Design and Operations 2016, pp. 287–316, https://doi.org/10.1016/b978-0-444- [63] Paul Prziwara, A. Kwade, Grinding aids for dry fine grinding processes – Part I:
63589-1.00010-1. Mechanism of action and lab-scale grinding, Powder Technol. 375 (2020)
[32] A. Gupta, D. Yan, Chapter 11: Mathematical modelling of comminution processes, 146–160, https://doi.org/10.1016/j.powtec.2020.07.038.
Mineral Processing Design and Operations, 2016 https://doi.org/10.1016/b978-0- [64] M.T. Puth, M. Neuhäuser, G.D. Ruxton, Effective use of Pearson’s product-moment
444-63589-1.00011-3. correlation coefficient, Anim. Behav. 93 (2014) 183–189, https://doi.org/10.1016/j.
[33] P. Hassall, E. Nonnet, V. Keikkala, T. Komminaho, L. Kotila, Ceramic bead behavior in anbehav.2014.05.003.
ultra fine grinding mills, Miner. Eng. 98 (2016) 232–239, https://doi.org/10.1016/j. [65] P.R. Rittinger, Lehrbuch der Aufbereitungskunde in ihrer neuesten Entwicklung und
mineng.2016.08.016. Ausbindung systematisch dargestellt, Ernst & Kern, 1867.
[34] M. He, Y. Wang, E. Forssberg, Parameter effects on wet ultrafine grinding of lime- [66] R.K. Rajamani, D. Guo, Acceleration and deceleration of breakage rates in wet ball
stone through slurry rheology in a stirred media mill, Powder Technol. 161 mills, Int. J. Miner. Process. 34 (1–2) (1992) 103–118, https://doi.org/10.1016/
(2005) 10–21, https://doi.org/10.1016/j.powtec.2005.08.026. 0301-7516(92)90018-R.
[35] J.A. Herbst, D.W. Fuerstenau, Scale-up procedure for continuous grinding mill design [67] K. Rajamani, J.A. Herbst, Simultaneous estimation of selection and breakage func-
using population balance models, Int. J. Miner. Process. 7 (1) (1980) 1–31, https:// tions from batch and continuous grinding data. Transactions of the institution of
doi.org/10.1016/0301-7516(80)90034-4. mining and metallurgy, Sect. B. Appl. Earth Sci. 93 (1984) C74–C85.

19
F. Patino, D. Talan and Q. Huang Powder Technology 403 (2022) 117394

[68] P. Semsari, M. Parian, J. Rosenkranz, Breakage process of mineral processing commi- [76] L. Taylor, D. Skuse, S. Blackburn, R. Greenwood, Stirred media mills in the mining in-
nution machines – An approach to liberation, Adv. Powder Technol. 31 (9) (2020) dustry: material grindability, energy-size relationships, and operating conditions,
3669–3685, https://doi.org/10.1016/j.apt.2020.08.005. Powder Technol. 369 (2020) 1–16, https://doi.org/10.1016/j.powtec.2020.04.057.
[69] S.W. Sharshir, G. Peng, L. Wu, F.A. Essa, A.E. Kabeel, N. Yang, The effects of flake [77] W.H. Walker, W.K. Lewis, W.H. McAdams, E.R. Gilliland, Principles of Chemical Engi-
graphite nanoparticles, phase change material, and film cooling on the solar still neering, 1937.
performance, Appl. Energy 191 (2017) 358–366, https://doi.org/10.1016/j. [78] C. Wang, S. Nadolski, O. Mejia, J. Drozcliak, B. Klein, Energy and cost comparisons of
apenergy.2017.01.067. hpgr-based circuits, Eng. Min. J. 214 (12) (2013) 102–108.
[70] S.W. Sharshir, G. Peng, L. Wu, N. Yang, F.A. Essa, A.H. Elsheikh, S.I.T. Mohamed, A.E. [79] B. Wills, J. Finch, Chapter 7: grinding mills, Will’s Mineral Processing Technology, 8th
Kabeel, Enhancing the solar still performance using nanofluids and glass cover ed 2016, pp. 147–179, https://doi.org/10.1016/B978-0-08-097053-0.00007-8.
cooling: experimental study, Appl. Therm. Eng. 113 (2017) 684–693, https://doi. [80] W. Yao, M. Li, M. Zhang, R. Cui, H. Jiang, Y. Li, S. Zhou, Effects of grinding media on
org/10.1016/j.applthermaleng.2016.11.085. flotation performance of calcite, Miner. Eng. 132 (December 2018) (2019) 92–94,
[71] F. Shi, R. Morrison, A. Cervellin, F. Burns, F. Musa, Comparison of energy efficiency https://doi.org/10.1016/j.mineng.2018.12.005.
between ball mills and stirred mills in coarse grinding, Miner. Eng. 22 (7–8) [81] J.B. Yianatos, M.A. Lisboa, D.R. Baeza, Grinding capacity enhancement by solid con-
(2009) 673–680, https://doi.org/10.1016/j.mineng.2008.12.002. centration control of hydrocyclone underflow, Miner. Eng. 15 (5) (2002) 317–323,
https://doi.org/10.1016/S0892-6875(02)00027-4.
[72] S.C. Shukla, S. Kukade, S.K. Mandal, G. Kundu, Coal-oil-water multiphase fuel: rheo-
[82] T. Yokoyama, Y. Inoue, Chapter 10: selection of fine grinding mills, Handbook Pow-
logical behavior and prediction of optimum particle size, Fuel 87 (15–16) (2008)
der Technol. 12 (07) (2007) 487–508, https://doi.org/10.1016/S0167-3785(07)
3428–3432, https://doi.org/10.1016/j.fuel.2008.05.027.
12013-3.
[73] L.F.O. Silva, T.J. Crissien, I.L. Schneider, É.P. Blanco, C.H. Sampaio, Nanometric parti- [83] X. Zhang, Y. Han, P. Gao, Y. Li, Effects of grinding media on grinding products and flo-
cles of high economic value in coal fire region: opportunities for social improve- tation performance of chalcopyrite, Miner. Eng. 145 (October 2019) (2020) 106070,
ment, J. Clean. Prod. 256 (2020) https://doi.org/10.1016/j.jclepro.2020.120480. https://doi.org/10.1016/j.mineng.2019.106070.
[74] A.B.G. Simpson, J.A. Byrne, J.A.D. McLaughlin, M. Strawhorne, Effect of solids concen- [84] J. Kennedy, R. Eberhart, Particle swarm optimization, Proceedings of ICNN’95-inter-
tration on particle size distribution of deagglomerated barium titanate in stirred national conference on neural networks 4 (1995) 1942–1948, https://doi.org/10.
media mills, Chem. Eng. Res. Des. 93 (2015) 287–292, https://doi.org/10.1016/j. 1109/ICNN.1995.488968.
cherd.2014.04.006. [85] R. Taylor, Interpretation of the Correlation Coefficient: A Basic Review, Journal
[75] Y. Takashimizu, M. Iiyoshi, New parameter of roundness R: circularity corrected by of diagnostic medical sonography 6 (1) (1990) 35–39, https://doi.org/10.
aspect ratio, Prog. Earth Planet. Sci. 3 (1) (2016) https://doi.org/10.1186/s40645- 1177/875647939000600106.
015-0078-x.

20