Minerals Engineering 18 (2005) 325–331

This article is also available online at:

www.elsevier.com/locate/mineng

Particle breakage kinetics in horizontal stirred mills

J. Yue, B. Klein *

Mining and Mineral Process Engineering, Frank Forward Building, Rm. 517-6350 Stores Road, University of British Columbia,
Vancouver, BC, Canada V6T 1Z4

Received 9 May 2004; accepted 19 June 2004

Abstract

The particle breakage kinetics, product size and size distribution, as well as grinding limit for quartz suspensions in a horizontal
stirred mill are investigated. Results of grinding studies on quartz suspensions in a horizontal stirred mill confirm ‘‘first-order’’ par-
ticle breakage rates. A comparison of equation fits reveal that the product size distribution from stirred mill is best represented
by the Rosin–Rammler equation when compared to the Gaudin–Shuhmann equation, but results fit poorly as particle size
below 5 lm.
A method of determining breakage rates versus particle size is demonstrated which shows that the 10% top size fraction is most
appropriate for the determination of specific breakage rates. The method is used to demonstrate the relationship between particle
breakage rate and particle size. As the feed particle size becomes smaller, the size reduction ratio decreases and approaches a grind-
ing limit which depends on the effective stresses created in the mill.
Ó 2004 Elsevier Ltd. All rights reserved.

Keywords: Mineral processing; Comminution; Fine particle processing; Grinding; Modelling

1. Introduction within the mill depends only on the mass of that size
fraction which is present in the mill contents (Austin,
Stirred bead mills have been applied increasingly to 1990).
ultra-fine grinding of minerals. However, since stirred
mills have only recently been introduced to large-scale Breakage rate of size fraction j ¼ S j wj ðtÞW ð1Þ
metal mines, models have not been developed to the where, Sj is the specific breakage rate of size j, W is the
same degree as for tumbling mills. mill hold-up, wj(t) is the weight fraction of size j material
The most popular mathematical model used for tum- at grinding time t.
bling mills is the ‘‘population balance model’’, which Thus,
was introduced by Epstein (1947), and developed further
by several researchers (Whiten, 1974; Herbst and Fuer- dwj ðtÞ=dt ¼ S j wj ðtÞ or wj ðtÞ ¼ wj ð0ÞeS j ðtÞ ð2Þ
stenau, 1968, 1973; Austin and Shah, 1983). For this
model, a ‘‘first-order breakage rate’’ is assumed. First- For a batch mill the population mass balance equation
order breakage is based on the hypothesis that is:
pffiffiffi for a
given size range of particles, for example a 2 screen X
i1
interval, the production of ground material per unit time dwi ðtÞ=dt ¼ S i wi ðtÞ þ bij S j wj ðtÞ where n > i > j > 1
j¼1
i>1

*
ð3Þ
Corresponding author. Tel.: +1 604 8223986; fax: +1 604 8225599.
E-mail address: bklein@mining.ubc.ca (B. Klein). where, bij, is the primary breakage distribution function.

0892-6875/$ - see front matter Ó 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mineng.2004.06.009
326 J. Yue, B. Klein / Minerals Engineering 18 (2005) 325–331

The primary breakage distribution function describes Table 1

the weight fraction of the products that occur in the size Test program for stirred mill
interval i if material of size interval j is broken. Eq. (3) Test no. Grinding cycles % Solids Flow rate (rpm)
states that the net rate of production of size material 1 1 35 3.10
equals the sum of the rates of appearance from breakage 2 5 40 1.68
of all sizes larger than size i minus the rate of disappear- 3 5 35 1.68
4 5 30 1.68
ance of size i by breakage. Morrel (Napier-Munn et al., 5 3 40 3.10
1996) developed a population balance model for tower 6 3 35 3.10
mills, which are similar to horizontal stirred mills. 7 3 35 3.10
Hogg (1999) developed a theoretical model for the
breakage kinetics for stirred mills that accounts for
50
breakage by massive fracture and attrition. He sug-
45
gested that attrition becomes more important as parti- 40
cles become finer, particularly in the micron range. His 35
model indicates that attrition grinding result in acceler-

Response
30
ation of particle degradation and non-first-order break- 25 Power, kW
age. In this case, the population-balance equation could 20 Temperature, C
% -10 µm
not be applied to grinding. 15
Studies were conducted with a pilot scale horizontal 10
stirred mill to characterize the particle breakage kinetics. 5
The results were analyzed to provide evidence of the types 0
0 30 60 90 120 150 180 210 240 270 300 330
and relative magnitudes particle breakage mechanisms. Grinding Time, s
The study is part of a larger student that investigates rhe-
ological aspect of grinding in horizontal stirred mills. Fig. 2. Equilibrium test for sampling in the stirred bead mill.

2. Procedures
Pilot scale tests were conducted using a continuous
Netzsch LME 4 horizontal stirred bead mill. The mill
is rubber lined and has an effective volume of 2.4 l.
The agitator shaft for the mill has eight polyurethylene
discs. The grinding medium used was 1 mm steel beads
and the loading charge was 80% of the total mill volume,
which corresponded to 5125 g.
Fig. 1 is a schematic showing the experimental system
that was used. The feed material was mixed in a surge
tank (1) and fed to the mill (3) via a pump (2). The flow
rate was regulated from the control panel (4). The mill is
equipped with digital-display panel which showed the
operating conditions (agitator speed, pump speed, mill
pressure, slurry temperature and shaft power).
The tests were conducted at three solid contents of
30%, 35%, and 40% by weight. For each test, a sample
was fed to the mill and the product was collected, sam-
pled and re-fed to the mill. This cycle was repeated up to
five times. Between each cycle, the mill was flushed with
water before starting the next grinding cycle. Table 1
lists the test conditions used for this study.
Plug-flow was assumed to simplify the calculation of
mean resident time s, which was defined as the net mill
volume (2.4 l) divided by the volumetric flow rate. For
all tests, the agitator speed was maintained at 1500
rpm. Particle size analyses (PSA) and rheological tests
were performed on each product sample. The PSA were
performed using either an Elzone 280 or a Malvern
Mastersizer 2000 Laser Diffraction. The feed material
used in this study was quartz with a F80 of 83 lm.
Test 1 was conducted to determine the sampling time
at which equilibrium conditions is achieved in the mill.
Equilibrium refers to stable operation with respect to
mill power, temperature and product size (% passing
10 lm). Fig. 2 plots these parameters versus time and
shows that equilibrium is achieved after about 180 s; this
was therefore used as the minimum operating time be-
fore sampling for each test.

3. Results and discussion

3.1. Product size distribution

Fig. 1. Experimental system for ultra-fine grinding tests. 1–– Surge The Gaudin–Schuhmann (Eq. (4)) and Rosin–
tank, 2–– Pump, 3–– Netzsch horizontal stirred bead mill, 4–– Display Rammler (Eq. (5)) equations were used to characterize
panel, 5–– Malvern Mastersizer, 6–– Haake RV 20 rheometer. the particle sizes of the grinding mill products.
 J. Yue, B. Klein / Minerals Engineering 18 (2005) 325–331 327

Gaudin–Schuhmann 1.00

m
W p ¼ 100ðx=kÞ ð4Þ

Correlation Coefficient, R2
where, Wp is wt% passing, x is the particle size, k is the 0.95
particle size modulus, m is the distribution modulus.
Rosin–Rammler
b 0.90
W r ¼ 100 exp½ðx=aÞ  ð5Þ
where, Wr is wt% retained, x is the particle size, a is the
size at which 36.8% of particles retained, b is the distri-
bution coefficient. 0.85
The model fits were compared using the correlation 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
coefficient R2, which is defined as follows: R-R particle size distribution modulus, b

Yb ¼ b
^ þb
0
^ Xi
1 ð6Þ Fig. 4. Relationship between distribution modulus and correlation
coefficient.
P
n P
n
^ b^ X i Þ2
ðY i  Y Þ2  ðY i  b 0 1 1.00
i¼1 i¼1
R2 ¼ P
n ð7Þ
2
ðY i  Y Þ

Correlation Coefficient, R2
0.98
i¼1

where b ^ is
^ is the intercept of the fitted regression line, b
0 1 0.96
the slope of the fitted regression line, Xi and Yi are the
values of corresponding point (Xi, Yi), Yb denotes the 0.94
estimated response at Xi based on the fitted regression
line, Y is the sample mean of the observations on Y. 0.92
The correlation coefficient R2 was plotted against test
number (Fig. 3) for both models. The average correla-
0.90
tion coefficient R2 for the Gaudin–Shuhmann and 0 10 20 30 40 50 60 70 80 90
Rosin–Rammler models were 0.91 and 0.97, respec- Particle size P80, µm
tively. The results suggest that the Rosin–Rammler fits
the data better than Gaudin–Shuhmann model. There- Fig. 5. Relationship between particle size and correlation coefficient.
fore the Rosin–Rammler equation was used to charac-
terize the size distribution of the feed and stirred mill Rosin–Rammler equation fits the data best at coarser
products. sizes and at narrower size distributions (higher values
In order to assess the suitability of the Rosin–Ramm- of b).
ler equation over the range of particle sizes and size dis- Fig. 6 is a plot of size distribution modulus versus
tributions, the correlation coefficient R2 was plotted product size. The graph indicates that the size distribu-
against the product size (P80) and distribution coefficient tion becomes narrower (b increases) with decreasing
(b). The trend lines shown in Figs. 4 and 5 show that
4.00
R-R Particle Size Distribution

1.00 3.50 40 % solid

Correlation Coefficient, R2

35 % solid
0.95 3.00
30 % solid
Modulus, b

0.90 2.50
0.85
2.00
0.80
1.50
0.75
1.00
0.70 Gaudin-Shuhmann
0.65 Rosin-Rammler 0.50

0.60 0.00
0 5 10 15 20 25 30 35 40 45 50 55 1 10 100
Test No. P 80, µm

Fig. 3. Correlation coefficient of both fitted models versus test Fig. 6. Rosin–Rammler particle size distribution modulus versus
number. product size P80.
328 J. Yue, B. Klein / Minerals Engineering 18 (2005) 325–331

100.00 The breakage rates were determined for two size frac-
tions, the +97 lm fraction and the +64 lm fraction. The
10.00 tests were conducted at three solid contents of 30%, 35%
Cumlative percentage

and 40%.
Figs. 9 and 10 show the relationship of particle break-
passing, %

1.00
age rate and grinding time. The slope of each line repre-
Feed sents the specific breakage rate Sj. The linear
0.10 R-R Fit R2=0.99 relationship suggests ‘‘first-order’’ particle breakage in
the mill. Since first-order breakage is observed for the
0.01 two different size fractions, it is assumed that all size
fractions will experience first-order breakage. The curves
show that breakage rate increases with decreasing solid
0.00
1 10 100 1000 content. By comparing Figs. 9 and 10 it can be seen that
Size, µm the breakage-rate decreases with decreasing particle size.
The results confirm ‘‘first-order’’ breakage rates for
Fig. 7. Particle size distribution feed sample with fitted Rosin–
Rammler equation. stirred mill grinding and support the use of the popu-
lation balance model. As described by Hogg (1999),
attrition breakage would lead to the acceleration of
100 the disappearance and appearance resulting in non-
first-order breakage. The Ôfirst-orderÕ breakage rate
implies massive fracture breakage mechanisms for the
Commulative Percentage

operating conditions used in this study.

Kwade (1999) analyzed the bead trajectories and
Passing, %

10 power density zones for a stirred disc mill. He concluded

that stress resulting from bead collision (impact) is the

1 F 80 66 µm 2 P 80 23 µm
3 P 80 14 µm 4 P 80 9 µm 1.9 40% pulp density, +97 µm
5 P 80 8 µm 6 P 80 6 µm 35% pulp density, +97 µm
30% pulp density, +97 µ m
Breakage Rate, log W(t)

1
1 10 100
Particle Size, Micron

Fig. 8. Particle size distribution products from five cycles of grinding

0.9
with fitted Rosin–Rammler equation.

product size P80. The trend becomes more significant

with decreasing solid content. The results imply that
decreasing the solid content will produce a narrower size -0.1
distribution. 0 10 20 30 40 50 60 70
Grinding Time, s
Fig. 7 shows the size distribution of the mill feed sam-
ple, which is characterized by a size coefficient of 53.92 Fig. 9. Breakage rate versus grinding time for the +97 lm fraction.
and distribution coefficient of 1.38. Fig. 8 shows the fit-
ted Rosin–Rammler equation for through five cycles of
grinding in the stirred mill for test number 2. Examina- 2
tion of the fitted Rosin–Rammler equation reveals a rea- 40% pulp density, +64 µm
35% pulp density, +64 µm
sonable fit down to about 3 lm, below which it deviates 30% pulp density, +64 µ m
Breakage Rate, log W(t)

from the data.

3.2. Breakage rate 1

The ‘‘one size fraction technique’’ (Austin, 1990) was

used to determine the disappearance rate of the top nar-
row size fraction of the stirred mill feed. The specific
breakage rates S were then obtained from plots of 0
0 10 20 30 40 50
breakage rate W(t) versus particle residence time t: Grinding Time, s

log W ðtÞ ¼ log W ð0Þ  St=2:3 ð8Þ Fig. 10. Breakage rate versus time for the +64 lm fraction.
 J. Yue, B. Klein / Minerals Engineering 18 (2005) 325–331 329

main mechanism causing particle breakage in stirred k ¼ S 1 d ap ð9Þ

mills, as long as the particle size is larger than 1 lm
S1 is the selection function at 1 mm and was determined
and the solid concentration is not too high. Breakage
to be 0.0035. The slope of the line is a, which is 0.51.
resulting from bead collision would lead to fracture
rather than attrition, supporting results that suggest
3.4. Grinding limit
attrition plays a minor role in stirred mill grinding.
The size reduction ratio was also determined from the
3.3. Breakage rate versus particle size
five cycle grinding tests. The size reduction ratio refers to
the ratio of feed size to product size for each cycle (F80/
For determination of specific breakage rate based on
P80). Fig. 12 shows the size reduction ratio decreases
Austin (1990) ‘‘one size fraction technique’’, 5%, 10%,
with the feed size and approaches 1.0 at about cycle five.
and 20% were chosen as top size fractions for the parti-
A value of 1.0 indicates that, for the grinding condition
cle size distributions of grinding products from the mill.
tested, a grinding limit exists whereby increasing grind-
The top size fraction W(0) of each grinding product was
ing time no longer results in particle breakage. The exist-
determined and W(t) was obtained from a successive
ence of a grinding limit is supported by the shape of the
grinding product through a series of grinding test using
product size distribution curves, which drop off sharply
Eq. (8). Since first-order breakage rate was confirmed,
below about 5 lm; this is where the Rosin–Rammler
the specific breakage rate could be obtained by the slope
equation deviates from the data.
of a two point line.
Fig. 13 is similar to Fig. 11, but shows the relation-
As shown in Fig. 11, the results of specific breakage
ship between specific breakage rate and particle size,
rate with 5% top size fraction show significant variation
below 10 lm. The results suggest that below about
as indicated by the low R2 value of 0.90. The variation
10 lm, the relationship between breakage rate and par-
can be explained by variable composition of the material
ticle size is not linear. The non-linear relationship
in the 5% top size fraction, which is not representative.
Using a top size fraction of 20% shows same variation
indicated by a R2 value of 0.93. This is likely due to par-
ticle breakage overlap (i.e. top size particle create prog- 3.5
eny that are still within the upper 20% range). It is
Size Reducion Ratio F80/P80

3.0
suggested that the 10% top size fraction of the grinding
products in stirred mills is the most appropriated in that 2.5
it does not suffer from variable composition or breakage 2.0
overlap. This is supported by the R2 value of 0.98.
1.5
Fig. 11 shows a good linear relationship between spe-
cific breakage rate and particle size for a stirred mill. The 1.0
curve shows the same type of relationship found for
0.5
tumbling mills; a decrease in breakage rate as feed size
become smaller. The straight line portion of the Austin 0.0
1 10 100
model can be used to describe the results for the specific F80, µm
breakage rate–particle size relationship.
Fig. 12. Particle size reduction ratio versus feed size F80.

0.10
1.000
Specific Breakage Rate, s-1

10% R2=0.99
Specific Breakage Rate, s-1

2
20% R =0.97 40%
R2=0.95
5%
35%
0.100
30%

0.010

0.01
1 10 100 1000 0.001
Particle Size, Micron 1 10 100
Particle Size P80
Fig. 11. Specific breakage rate versus top particle size fractions
representing 5%, 10% and 20%. Fig. 13. Specific breakage rate versus particle sizes.
330 J. Yue, B. Klein / Minerals Engineering 18 (2005) 325–331
implies a change in the dominant breakage mechanism
below this size for the grinding conditions used in the
testwork.
Tromans and Meech (2002) conducted a theoretical
study of fracture toughness and surface energies. Based
on their results, it is suggested that the grinding limit de-
pends on the square of the critical stress intensity for
crack propagation, provided that the energy is effec-
tively exerted on to the particles. It is reasonable to con-
sider that more energy will be needed to fracture a
particle that lacks flaws. They also calculated the energy
efficiency (theoretical computed value over Bond Work-
ing Index) for many minerals, and showed that all the
energy efficiencies are very low (of the order of 1%).
The low efficiency was attributed to energy losses due
to particle deformation and to inefficient impact from
grinding media. Therefore the practical grinding limit
depends on the effective stresses created in the mill
rather than critical stress intensity for crack propagation
for a material.
3.5. Breakage mechanisms
Attrition and abrasion (no impact) were once be-
lieved the primary breakage mechanisms in stirred bead
mills. Recent observations (Kwade, 1999; Kwade and
Schwedes, 2002; Shinohara, 1999) suggest that all three
breakage mechanisms (attrition, abrasion and impact)
exist in these mills, although the degree may vary from
case to case.
Hogg (1999) assumed that product size would be-
come increasingly bimodal as the relative contribution
from attrition increases. Fig. 14 shows the size distribu-
tion of products through three successive cycles of
grinding. For cycles one and two, the distribution
becomes narrower as the grinding limit is approached.
Cycle 3 results in the production of a very fine fraction
of about 0.2 lm in size creating a bimodal size distribu-
tion. The results suggest that massive fracture was the
main breakage mechanism from cycles one and two.
7.0
6.0 Feed
Product (cycle 1)
5.0 product (cycle 2)
Product (cycle 3)
Volume %
4.0
3.0
2.0
1.0
0.0
0.01 0.1 1 10 100
Particle Size, µm
Fig. 14. Size distribution for stirred mill products following successive
grinds.
However, near the grinding limit, attrition or abrasion
become dominant and there seems to be very little mas-
sive fracture.
4. Conclusions
The breakage kinetics of quartz suspensions with dif-
ferent solid content were investigated for a horizontal
stirred bead mill. The results confirm the ‘‘first-order’’
breakage rate for tested slurries in the mill. The relation-
ship of specific breakage rate and particle size for stirred
mill is similar to those for tumbling mill and therefore
the population balance model can be applied.
A comparison of equation fits reveal that the product
size distribution from stirred mill is best represented by
the Rosin–Rammler equation. The Rosin–Rammler fit
improves for products with a narrower size distribution
and coarser particle size but is poor for particle sizes
below five microns.
A method of determining breakage rates versus parti-
cle size is demonstrated which shows that the 10% top
size fraction is most appropriate for the determination
of specific breakage rates in a stirred mill.
The size reduction ratio decreases with the feed size
and approaches 1.0, indicating that a grinding limit ex-
ists. It is proposed that this limit depends on the effective
stresses created in the mill which are determined by bead
size, agitator speed and pulp density. Based on size dis-
tribution results, the primary breakage mechanism in
the horizontal stirred mills is massive fracture. However,
near the grinding limit attrition is dominant.
Acknowledgements
The authors would like to acknowledge the Natural
Sciences and Engineering Research Council of Canada
for providing funding to conduct the research and the
Canada Foundation for Innovation for the equipment
grant to acquire the testing facilities.
References
Austin, L.G., 1990. Ball Mills, Semi-Autogenous Mills, and High
Pressure Grinding Rolls. The Pennsylvania State University, pp.
4.1–5.14.
Austin, L.G., Shah, I., 1983. A method for interconversion of
microtrac and sieve size distributions. Powder Technol. 35, 271–
278.
Epstein, B., 1947. The material description of certain breakage
mechanisms leading to the logarithmic––normal distribution.
1000
Franklin Inst., 244–471.
Herbst, J.A., Fuerstenau, D.W., 1968. The zero order production of
fines in comminution and its implication in simulation. Trans.
SME/AIME 241, 531–549.
 J. Yue, B. Klein / Minerals Engineering 18 (2005) 325–331
Herbst, J.A., Fuerstenau, D.W., 1973. Mathematical simulation of dry
ball milling using specific power information. Trans. SME/AIME,
254–348.
Hogg, R., 1999. Breakage mechanisms and mill performance in
ultrafine grinding. Powder Technol. 105, 135–140.
Kwade, Arno, 1999. Wet comminution in stirred media mills-research
and its practical application. Powder Technol. 105, 14–20.
Kwade, Arno, Schwedes, Joerg, 2002. Breaking characteristics of
different materials and their effect on stress intensity and stress
number in stirred media mills. Powder Technol. 122, 109–121.
331
Napier-Munn, T.J., Morrell, S., Morrison, R.D., Kojovic, T., 1996.
Mineral Comminution Circuits. Their Operation and Optimisa-
tion. Julius Krutschnitt Mineral Research Centre, Brisbane.
Shinohara, Kunio, 1999. Fine grinding characteristics of hard mate-
rials by attrition mill. Powder Technol. 103, 292–296.
Tromans, D., Meech, J.D., 2002. Fracture toughness and surface
energies of minerals: theoretical estimates for oxides, sulphides,
silicates and halites. Miner. Eng. 15, 1027–1041.
Whiten, W.J., 1974. A matrix theory of comminution machines. Chem.
Eng. Sci. 29, 588–599.