Minerals Engineering 86 (2016) 66–74

Contents lists available at ScienceDirect

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

A specific energy-based ball mill model: From batch grinding

to continuous operation
Fengnian Shi ⇑, Weiguo Xie
The University of Queensland, Sustainable Minerals Institute, Julius Kruttschnitt Mineral Research Centre, Qld 4068, Australia

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

Article history: A specific energy-based size reduction model for batch grinding ball mills was reported in a previous
Received 6 October 2015 paper (Shi and Xie, 2015). A discharge function modified from the Whiten classification efficiency equa-
Revised 17 November 2015 tion has been incorporated in the size reduction model to extend its applications from batch grinding to
Accepted 1 December 2015
continuous operation. Five sets of the industrial ball milling survey data were used to validate the ball
Available online 14 December 2015
mill model. The data were acquired from four full scale ball mills covering primary and secondary grind-
ing duties in a gold concentrator and a PGM concentrator. In all cases, the model fits the ball mill oper-
Keywords:
ational data well.
Comminution
Grinding
Features of the specific energy-based ball mill model include the use of an ore-specific and size-
Ball mill model dependent breakage function, whose parameters are independently measured with a fine particle break-
Ore breakage characterisation age characterisation device, the JKFBC. This allows simulations of the effect on ground product size dis-
tribution of changing ore breakage characteristics. The model utilises separate selection function and
discharge function, which permits the investigation of the influences of mill operational conditions on
grinding performance.
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction readings during the experiments. The computer logged torque

readings are converted to power by mill geometry and operational
A specific energy-based size reduction model for batch grinding condition, and integral with grinding time to give the consumed
ball mills has been developed (Shi and Xie, 2015). The objective of energy. The JKFBC is suitable to test
4.75 mm ore particles. The
this work was to overcome the limitations in the existing JKMRC particle size (between 0.106 mm and 4.75 mm) tested in the JKFBC
(Julius Kruttschnitt Mineral Research Centre) perfect mixing ball represents approximately 70–80% of particles by mass feeding to a
mill model, including the use of a set of default breakage appear- typical secondary grinding ball mill. For each ore sample, three to
ance functions for all ores, and the lumped model parameter r/d⁄ five narrowly sized fractions are tested, each with three different
for breakage rate and mass discharge rate. grinding revolutions. A constant volume (60 ml, approximately
As a first step approach to upgrading the ball mill model, a 40–50 g depending on solids density) of particles is used for each
specific energy-based size reduction model was developed. In this test. A size-dependent breakage model (Shi and Kojovic, 2007;
batch grinding size reduction model, no mass transport or dis- Shi et al., 2015) was used for the JKFBC testing data reduction to
charge mechanism occurs, and the product size distribution solely generate breakage characteristic parameters, M, p and q, for the
represents the size reduction results in relation to specific energy ore of interest. Thus the default breakage appearance function in
input and material breakage characteristics. the existing perfect mixing ball mill model is replaced by the
A Fine-particle Breakage Characteriser (JKFBC, Shi and Zuo, ore-specific breakage characteristic parameters.
2014; Shi, 2014a,b) was employed to measure ball mill feed break- A model to describe size reduction in a batch grinding ball mill
age characteristic parameters. The JKFBC is modified from the tra- was developed. This model is based on the mass-size balance, or
ditional Hardgrove ball-race mill popularly used by the coal mass conservation, which is expressed by Eq. (1).
industry to determine coal grindability (HGI). A precision
torquemeter was installed on the HGI mill shaft to record torque
X
i
1
pi ¼ f i  mij ð1Þ
j¼1

⇑ Corresponding author. The variable mij is a lower-triangular breakage matrix, indicating

E-mail address: [email protected] (F. Shi). the mass fraction of appearance of size i material produced by

http://dx.doi.org/10.1016/j.mineng.2015.12.004
0892-6875/Ó 2015 Elsevier Ltd. All rights reserved.
 F. Shi, W. Xie / Minerals Engineering 86 (2016) 66–74 67

Nomenclature

Ab ore competence indicator, the large value indicating the Hi the mass holdup of the mill ground product in size i (t)
less resistance to breakage M, p, q size dependent breakage model parameters, M in %
C the maximum probability of particles reporting to the mij lower-triangular breakage matrix (–)
fine component (–) Pi mass flow rate of the ground product in size i exiting the
Di the discharge rate for size i (h
1) mill, (t h
1)
di geometric mean particle size in fraction i (mm) pi mass fractions of size i in the mill product
Dmax the maximum discharge rate for the ground product t10i product cumulative passing 1/10th of the mean particle
exiting the ball mill like water (h
1) size in feed size i (%)
d50c the particle size at which classification efficiency is 0.5 tn product cumulative passing 1/nth of the initial particle
(mm) size (%)
Ei mass specific energy (J kg
1) xi particle size in the size-dependent breakage model (m)
EF classification efficiency for material reporting to the fine
component (–) Greek letters
fi mass fractions of size i in the mill feed a classification efficiency parameter (–)
fmat parameter in the size-dependent breakage model
(kg J
1 m
1)

fracture of size j material. For each size i, the summation sign sums energy. The assumption behind is that the size-specific energy
up the total material from sizes larger than i to size i
1. The break- selection function is related to the ball size distribution, charge vol-
age matrix mij is ore-specific, size-dependent and specific energy- ume, mill diameter, mill speed, liner condition, feed size, etc, and
based. can be called ‘‘scale-up parameters”. Currently, the selection func-
The breakage matrix mij is calculated from the particle breakage tion at three or four size knots is fitted to the experimental data of
index, t10, and the product size distribution using the t10–tn rela- mill product size distribution. Ultimately, the selection function
tionships. The breakage index t10 for each feed size i is calculated may be calculated from the mill geometry and operational
using Eqs. (2) and (3) based on the size-specific energy and the conditions.
given ore characteristic parameters: A number of batch grinding test datasets using a standard Bond
ball mill were used to validate the size reduction model. The model
t 10i ¼ M  f1
exp½
f mat  xi  Ei g ð2Þ
fits the batch grinding data well. The calibrated model from one

q
ore sample was used to predict the breakage results for another
f mat ¼ p  di ð3Þ ore with the measured ore characteristic parameters. The same
where M, p and q are ore breakage characteristic parameters deter- approach was applied to the JKSimMet ball mill model. Compar-
mined from the JKFBC tests, and Ei is size-specific energy that is cal- ison of the predictions by the two models shows better results
culated by mill power draw and a breakage selective function. from the specific energy-based ball mill model.
It is worth noting that the ore testing data are not simply used As a second step in the ball mill model development, the speci-
in the ball mill model as a single number (e.g. t10), or a single col- fic energy-based size reduction model was extended from batch
umn of appearance vector, which may be questioned on the ratio- grinding to full scale ball milling in continuous operation mode,
nale due to the differences between the two breakage devices. In which is reported in this paper.
modelling the ball mill size reduction, Eqs. (2) and (3) are
employed to calculate the ground product t10 by the given ore
2. Modelling discharge function
characteristic parameters of M, p, q from the feed size xi (in narrow
size fractions) and the energy applied to each particle size fraction
In essence the size reduction model can be applied as a basic
Ei. Thus for each feed size and its associated specific energy level
structure for both batch grinding and continuous grinding ball
the breakage results can be calculated, and a lower-triangular
mills. For a continuous grinding mill, however, a classification
appearance matrix (mij) determined. It is emphasised that for a ball
mechanism and a discharge function must be added into the size
mill to treat the same ore with different mill feed size distributions
reduction model to allow discharge of the ground product based
and in different mill operational conditions (reflected in the differ-
on the particle size.
ent size-specific energy levels), the appearance matrix (mij) is dif-
ferent. This approach was developed from the successful exercise
in hammer mill model (Shi, 2002; Shi et al., 2003) and the vertical 2.1. The classification function
spindle models (Shi et al., 2015; Kojovic et al., 2015).
The size-specific energy term Ei in Eq. (2) explicitly takes into The literature on tumbling mill classification mechanisms was
account the mill operational conditions, such as mill geometry, ball reviewed. A set of classification data collected by Man (2001) from
charge volume, and mill throughput, if the mill power draw is cal- the internal sampling of an operational ball mill reveals that the
culated by Morrell’s power model (Morrell, 1992) or DEM; or classification in a ball mill (Fig. 1a) is similar to that in an AG/
implicitly if the mill power is directly measured. In the specific SAG mill. The classification function is characterised by two
energy-based ball mill model the mill is no longer treated as a regions. The first extends up to a critical particle size, where the
‘black box’; instead both the machine factor and material factor discharge rate is largely constant and equal to that of water. For
have been mechanistically incorporated in the size reduction particles above the critical size the discharge rate progressively
model. Note that the ball mill model does not incorporate breakage reduces. Inspection of the ball mill classification data found that
rate as in the population balance model. However, the model does the shape of the classification function is similar to the Whiten
incorporate a selection function in determine the size-specific expression for corrected efficiency to overflow (Fig. 1b, refer to
68 F. Shi, W. Xie / Minerals Engineering 86 (2016) 66–74

(a) Ball mill classification function (Man, 2001) (b) Efficiency curves to overflow
Fig. 1. The measured ball mill classification function by mill internal and external sampling (Man, 2001) in comparison with the Whiten efficiency curves to overflow
(Napier-Munn et al., 1996).

Napier-Munn et al., 1996). Therefore the Whiten efficiency equa- 3. Model validation
tion was adopted to describe the ball mill classification function
The Whiten efficiency curve can be expressed by Eq. (4): 3.1. Data acquisition
2 3
Five sets of full scale ball mill operational data were used to val-
expð aÞ
1
EF ¼ C  4   5 ð4Þ idate the specific energy-based ball mill model. The five datasets
exp a  d50c
d
þ expðaÞ
2 were collected during plant surveys conducted in a gold ore con-
centrator in Australia and a PGM (platinum group minerals) con-
centrator in South Africa. The gold ore concentrator comminution
Whiten’s efficiency equation incorporates three parameters, C, a
circuit surveys were conducted in the AMIRA P9P project by a
and d50c. Parameter C represents the maximum probability of parti-
research group comprising JKMRC, Chalmers University and site
cles reporting to the fine component, a is classification efficiency
personnel. The PGM concentrator grinding circuit surveys were
(variation of efficiency curves with a is shown in Fig. 1b), and d50c
conducted by MPTech of the University of Cape Town (UCT).
is the particle size at which classification efficiency is 0.5.
The gold ore grinding circuit comprises three ball mills, one
used for primary grinding and the other two for secondary grind-
ing. The PGM concentrator employs one ball mill for primary
2.2. The discharge rate
grinding. Two surveys on the ball mill grinding circuit in the
PGM concentrator were conducted when the ball mill treated the
To describe ball mill discharge rate the classification function
crusher product and the HPGR (High Pressure Grinding Rolls) pro-
(Eq. (4)) was modified by introducing a maximum discharge rate
duct respectively. Specifications of the four ball mills are sum-
in the classification function, as shown in Eq. (5):
marised in Table 1.
" # In the gold ore concentrator the primary ball mill was operated
expðaÞ
1 in a closed circuit with a DSM screen and a cluster of hydrocy-
Di ¼ Dmax  ð5Þ
expða  dd50ci Þ þ expðaÞ
2 clones. The primary ball mill feed comprises the new feed, DSM
screen oversize and hydrocyclone underflow. The primary ball mill
ground product (ie. the trommel undersize), together with the
The parameter Dmax replaces parameter C in the classification func- DSM screen undersize and the primary hydrocyclone overflow,
tion (Eq. (4)). For simplicity, Dmax is treated as a model parameter were pumped to the secondary grinding circuit. The secondary
fitted to ball mill operational data. The second parameter, a, deter- grinding circuit consisted of two ball mills in parallel and one clus-
mines classification efficiency. For coarser and denser particles ter of secondary hydrocyclones. The primary grinding circuit prod-
ground in a ball mill, the settling velocity is high and the efficiency ucts and the trommel undersize from the two secondary ball mills
a is small. For fine and light particles, on the other hand, the effi- were combined to feed the secondary hydrocyclones. The sec-
ciency parameter a in a ball mill is large. This will give a sharp rise ondary hydrocyclone underflow material was split to feed the
in the classification curve. The third parameter d50c (mm) is also fit- two ball mills in parallel.
ted to the ball mill survey data.
For continuous ball mill operation, the mass flow rate of mill
product, Pi, is the result of mill ground product, Hi and discharge
rate, Di, expressed by Eq. (6): Table 1
Specifications of the four ball mills used for the model validation.

P i ¼ H i  Di ð6Þ Ball mill code BM1 BM2 BM3 BM4

Ore ground Gold Gold Gold PGM
The product size distribution can be calculated as a ratio of the mass Grinding duty Primary Secondary Secondary Primary
Mill diameter (m) 5.02 4.08 4.1 4.99
flow rate by size (Pi) to the total mass flow rate. The mass holdup of Mill length (m) 8.45 6.43 6.7 6.69
the mill ground product Hi can be determined by mill feed mass (t) Trunnion diameter (m) 1.41 0.76 0.76 1.21
multiplying the ground product mass fraction of size i (pi) using the Critical speed (%) 75 75 75 72
batch grinding ball mill size reduction model (Eq. (1)). Therefore Top ball size (mm) 80 40 40 80
Ball charge volume (%) 42 42 44 23
Eqs. (1), (5) and (6) constitute the ball mill model for continuous
Gross power (kW) 3924 1908 1933 1856
operation converted from the batch grinding model.
 F. Shi, W. Xie / Minerals Engineering 86 (2016) 66–74 69

In the PGM concentrator the primary ball mill was operated in a 19.6 kWh t
1). However, the JKFBC tests indicated that PL2 feed ore
closed circuit with a screen. The screen oversize returned to the (A  b = 9.7) were slightly weaker than the PL1 feed ore
ball mill. Two surveys were conducted. PL1 was conducted when (A  b = 7.6). As PL2 feed ore was from HPGR product, perhaps
the ball mill treated a feed from a special mobile crusher. Survey the slight difference in the A  b values reflected the pre-
PL2 was performed when the ball mill received a feed from HPGR, weakening effect of HPGR. More data are required for further val-
which had a much finer mill feed than that of the crusher product idation of this observation.
in PL1. In the size reduction model, it predicts particle breakage distri-
During the plant surveys, samples from the grinding circuits bution from the breakage fineness index t10 value using the tn–t10
were collected. Mass balancing was performed on the tonnage, relationships. The tn-family of curves can be established from the
percent solids and size distributions of various streams in the cir- JKFBC tests. Table 3 shows examples of the tn-family of curve val-
cuits. The ball mill feed was reconstituted by the new feed and ues at three t10 knots (t10 = 10, 30 and 50 respectively) for the two
the mass balanced circulations. The reconstituted mill feed and primary grinding ball mill feed samples.
product data from the five surveys were used to fit the parameters It was found from the single-particle impact breakage tests of a
of the specific energy-based ball mill model respectively. number of ore samples that the tn-family of curves are closely
related to the breakage fineness index t10 value (Narayanan,
3.2. Ball mill feed breakage characterisation 1985). The tn-family of curves have been independently confirmed
using different materials over a wide range of fracture energies by
Five ball mill feed samples were collected during the plant sur- Pauw and Maré (1988) and by King and Bourgeois (1993). In the
veys. For the primary ball mill circuit, the new feed samples were JKFBC test under particle bed compression breakage mode, the
tested; while for the secondary ball mills, the hydrocyclone under- investigation suggests that the tn-family of curves may not be ‘uni-
flow samples were tested. versal’ for all ores. Table 3 shows that at small t10 values (e.g.
The JKFBC tests were performed on the five ball mill feed sam- t10 = 10), the tn-family of curve values are similar for the two ore
ples. Four narrow particle size fractions (2.8–4.0 mm, 2.0–2.8 mm, samples; but when t10 increases (e.g. at t10 = 50), the difference
1.18–2.0 mm and 0.6–1.18 mm) for each sample were tested. increases significantly, and the ‘universal’ relationships may be
Three grinding energy levels (by variation in grinding revolutions) questioned. Therefore, in the ball mill model validation process,
were used for each size fraction. The real energy consumption in the measured tn-family of curves from the JKFBC tests for individ-
each testing condition was recorded with a precision torquemetre. ual ore sample were used.
The size-dependent breakage model was employed for the JKFBC
testing data reduction to generate breakage characteristic parame- 3.3. Model fitting
ters, M, p and q, for each of the five ball mill feed samples. Fig. 2
presents the results of the size-dependent breakage model fitted The specific energy-based size reduction model and the dis-
to the five JKFBC testing data. In all cases, the model fits the data charge function have been coded in the SDK (Software Developer
well. Kits) version of JKSimMet. JKSimMet is commercial comminution
The fitted M, p, q parameters are incorporated in the size reduc- software. Fig. 3 shows the ball mill model SDK window interface.
tion model. Since the mining industry is familiar with the A  b All functions of JKSimMet (flowsheet, mass balance, model fitting,
value as an ore competence indicator (Napier-Munn et al., 1996), simulation, reporting, etc.) are available in the SDK version of the
the size-dependent breakage model parameters can be converted specific energy-based ball mill model. There are four basic sections
to the A  b value using Eq. (7): in the SDK window interface. The two sections in the middle are for
ore breakage characteristic parameters: The Breakage Function is
ð1
qÞ for inputting M, p, q parameters, which predicates the product fine-
A  b ¼ 3:6M  p  d ð7Þ
ness index t10 for a given specific energy and a give feed size. The
Due to the particle size effect on breakage results, A  b values vary section tn-family of curves predicts product size distribution from
with particle size. The traditional A  b value is determined for a the t10 index. These two sections present ore breakage property
mean particle size in the Drop Weight Test (DWT). Since this size that is determined by the JKFBC tests on the ball mill feed ore sam-
(DWT testing 13.2–63 mm particles in narrow size fractions) is ple. The top and bottom sections present the machine/operation
too coarse for a ball mill feed and the A  b value at this size is irrel- related functions. The top section is the selection function govern-
evant to ball milling, a nominal size of 2 mm is used to compare ore ing the size-specific energy level. The selection function is defined
competence for ball mill feed. Table 2 gives the A  b values of the at four size knots and extended to the full size range by cubic
five ball mill feed samples calculated using Eq. (7) for d = 2 mm. The spline method. The bottom section is the Discharge Function with
smaller A  b value indicates the more competent ore, or the higher three parameters. During the model fitting process, the two sec-
resistance to grinding. For comparison, the Bond ball mill work tions related to ore properties are kept constant; while the
indices of the five ore samples are also given in Table 2 as a machine/operation related parameters (the Selection and Dis-
reference. charge functions) are adjusted to minimise the fitting error
For the three gold ore samples (BM1, BM2 and BM3), the JKFBC between the measured and the model predicted ball mill product
tests indicated that they were similar in ore competence. This is size distributions.
expected as the three ball mill feed samples were collected in The results of the upgraded ball mill model fitted to the five sets
the same survey when the primary and secondary grinding circuits of full scale ball mill operational data are presented in Fig. 4. The
treated the same ore. The Bond work indices showed large varia- fitting results indicate that the model works well for the continu-
tions among the three ball mill feed samples. Particularly BM2 ous grinding ball mills, after the discharge function is incorporated
and BM3 were in parallel with both feeds coming from the same in the specific energy-based size reduction model.
hydrocyclone underflow, the ore competence was unlikely to vary Table 1 shows that ball mills BM2 and BM3 are similar in geom-
from 22.3 kWh t
1 (BM2) to 18.8 kWh t
1 (BM3). Comparing the etry and mill operational conditions (speed, ball top size, ball
three ball mill feed Bond work indices, 22.3 kWh t
1 (BM2) may charge volume). The selection function parameters fitted to the
be questioned. BM2 survey data was kept constant for BM3. Only three discharge
For the two PGM ore samples, on the other hand, the Bond work function parameters were used to fit the BM3 measured product
indices indicated that they were similar in ore competence (20.0 vs size distribution. As shown in Fig. 4, the predicted BM3 product
70 F. Shi, W. Xie / Minerals Engineering 86 (2016) 66–74

(a) Gold ore from Australia

(b) PGM ore from South Africa

Fig. 2. The size-dependent breakage model fitted to the JKFBC testing data for the five ball mill feed samples.

Table 2 Table 3
Ore competence indicator A  b values calculated by JKFBC characterisation param- tn-family of curve data for the primary ball mill feed determined by JKFBC tests.
eters for a particle size 2 mm and Bond work indices of the five ball mill feed samples.
Gold ore BM1 PGM ore PL1
Ball mill feed BM1 BM2 BM3 PL1 PL2
t10 = 0 10 30 50 10 30 50
A  b (JKFBC @ 2 mm) 7.4 7.7 7.8 7.6 9.7
t2 0 27.53 60.85 87.99 29.13 60.82 85.64
Bond Wi @ 0.125 mm (kWh t
1) 18.7 22.3 18.8 20.0 19.6
t4 0 16.52 42.91 66.88 17.46 43.84 67.25
t25 0 6.89 20.25 33.61 5.99 17.66 29.32
t50 0 5.08 15.83 26.58 4.22 11.89 19.55
size distribution by only fitting the discharge function is equally as t75 0 4.02 14.38 24.75 3.45 11.35 19.25

good as BM2 fitting both selection and discharge functions. The

similar approach was taken in fitting the PGM ball milling survey using PL1 data were applied to PL2 without change. The difference
data. Since PL1 and PL2 were conducted on the same ball mill with in the feed particle size is taken into account in the size reduction
the same ball charge volume, the calibrated selection function model, and the difference in the grinding density and mill through-
 F. Shi, W. Xie / Minerals Engineering 86 (2016) 66–74 71

Fig. 3. The SDK window interface of the ball mill model in JKSimMet.

put is expected to be reflected in the fitted discharge function. employed to demonstrate the effect on ground product size distri-
Therefore only the discharge parameters were fitted to the PL2 bution of changing ore breakage characteristics.
data. Fig. 4 shows that despite the large difference in the ball mill Two sets of ore breakage characteristic parameters were used
feed conditions, the model is able to fit the PL2 product size distri- for the demonstration, which are given in Table 4. The first set is
bution with the similar accuracy as that by fitting both selection from the BM1 feed with an ore competence indicator A  b of 7.4
and discharge functions for PL1. determined by JKFBC and calculated by Eq. (7) from the M, p, q
parameters at a nominal particle size 2 mm. The second set of
breakage characteristic parameters is from a copper ore, which
4. Discussion contains the softer ‘‘DOME” material. The A  b value is 40.2 deter-
mined in the same way as the first set of data. Apparently the cop-
4.1. Ore-specific breakage function per ore used for the simulation is less resistant to ball mill grinding
compared with the gold ore in the BM1 feed.
One of the major features of the specific energy-based ball mill Another distinct difference between the two datasets is the q-
model is the use of ore-specific breakage function determined by parameter values. The q-parameter is used to describe the particle
the JKFBC test for an individual ball mill feed sample. Thus the fit- size effect on breakage result in the size-dependent breakage
ted breakage selection function and discharge function would model (Shi and Kojovic, 2007; Shi et al., 2015). When q is large
reflect the influence of the machine and operational conditions. (0.939 for BM1 feed), the exponent (1
q) of particle size d is close
Once calibrated, the model allows simulation of changes in ore to zero, indicating an insignificant size effect on breakage. When q
breakage property. By way of example, the model calibrated to is small (0.262 for the simulation copper ore), (1
q) is much lar-
the BM1 in the gold concentrator primary grinding circuit was ger than zero, indicating a significant particle size effect on break-
72 F. Shi, W. Xie / Minerals Engineering 86 (2016) 66–74

Fig. 4. Ball mill product size distributions fitted by the ball mill model in comparison with that measured in the five full scale ball mill surveys.

Table 4 isation does show the different breakage behaviour of the two
Two sets of breakage characteristic parameters used for simulations of the effect on types of ore.
the ground product size distributions of changing ores.
Simulations were performed for the ball mill to treat the two
Code Ore M p q A  b (JKFBC @ A  b (JKFBC @ feed ores respectively using the calibrated selection and discharge
2 mm) 0.1 mm) functions to the BM1 survey data and the same feed size distribu-
BM1 Gold 63.3 0.031 0.939 7.4 6.2 tion data. The simulated product size distributions by the specific
Sim Copper 63.8 0.105 0.262 40.2 4.4 energy-based ball mill model are presented in Fig. 5.
The simulations predict that at a particle size coarser than
0.2 mm, the copper ore is likely to be ground finer than the gold
age. As a result of this strong decaying size effect nature, the A  b ore. However in the sizes finer than 0.2 mm, the reverse trend is
value at 0.1 mm for the copper ore is even smaller than the gold predicted. This simulation result reflects the difference in ore
ore (comparing the last column in Table 4). The breakage charac- breakage characteristics. The specific energy-based ball mill model
teristics of the two ore samples indicate that the copper ore has seems able to correctly predict the product size distribution in
less resistance than the gold ore to grinding at coarse particle sizes, agreement with the breakage characterisation results. In compar-
but exhibits a higher resistance to grinding down to 0.1 mm pro- ison, the existing ball mill model incorporated in JKSimMet uses
duct size. This is perhaps related to the ore texture or grain size Bond work index to scale the r/d⁄ (a lumped model parameter rep-
influence of the copper ore sample. Nevertheless, despite the study resenting the ratio of breakage to discharge rates) for the ore
of the size effect on breakage result of ore texture and grain size change simulations. Since Bond work index is a single parameter
being out of the scope of this modelling work, the JKFBC character- that does not relate to the particle size effect, it would be difficult
 F. Shi, W. Xie / Minerals Engineering 86 (2016) 66–74 73

Fig. 4. The fitted discharge functions for the two pairs are presented
in Fig. 6 for comparison.
As shown in Fig. 6, the curve shape of the discharge rates is sim-
ilar to the measured ball mill classification function displayed in
Fig. 1a. The major difference of the fitted discharge rates between
the paired comparisons is in the Dmax values that are represented
as the intercepts of the curves on the y-axis. The fitted Dmax values
were 321 and 249 h
1 for BM2 and BM3 respectively, and 312 and
231 h
1 for PL1 and PL2 respectively. The PL2 survey had finer feed
and larger overall throughput than PL1, but the discharge rate for
fine particle size less than 0.06 mm was smaller than PL1. Analysis
of the fitted discharge rates suggests that this may be attributable
to the higher grinding density in PL2 (77.2% solids in PL2 vs 74.3%
solids in PL1). Associated with the high grinding density, slurry
rheology may become an issue affecting the maximum discharge
rate (Dmax) for fines in PL2. The similar trend was observed in the
gold concentrator survey data presented in Fig. 6, with the larger
Dmax value (321 h
1 in BM2) associate with the lower grinding den-
sity (72.5% solids).
Fig. 5. The simulated ball mill product size distributions for the two ores with their Interestingly, when the results of the ball mill model fitting to
breakage characteristics presented in Table 4.
the PGM concentrator grinding circuit survey data were reported
back to the UCT, the following comment was received (Sweet,
to predict the size effect on the grinding result as shown in Fig. 5 2015, Personal communication):
using the upgraded ball mill model. ‘‘I’m intrigued by the lower Dmax for the HPGR feed survey. We
have done a lot of work on the mill looking at slurry density and we
know that above 78% solids the mill has a slurry pool owing to the
4.2. Separate breakage selection and discharge function viscosity of the material. At 77% solids the mill was on the verge of
forming a pool, and I’m very pleased that without any prior knowl-
The specific energy-based ball mill model incorporates two sep- edge (which could have led your interpretation of the data) the dis-
arate functions: the selection function and discharge function. The charge rates you fitted to the fines show this.”
selection function, presented in the previous paper (Shi and Xie, In summary, incorporation of the discharge function in the
2015), governs the size-specific energy level that describes the mill specific energy-based size reduction model has been proved to
specific energy being applied to particles in different sizes. The dis- be a successful approach for modelling continuous ball mill opera-
charge function, as presented in this paper, represents the ground tion, in overcoming the limitation of the lumped breakage rate and
product being classified and discharged out of the mill as a func- discharge rate used in the prior-art perfect mixing ball mill model.
tion of particle size. It is hypothesised that the selection function
is machine-specific. The mill geometry and the operational condi-
tions such as mill rotational speed, ball charge volume, and ball 5. Conclusion
size will affect the selection function (Austin et al., 1984). In the
two secondary gold ball mill surveys (BM2 and BM3) and the A discharge function was incorporated in the specific energy-
two PGM concentrator surveys (PL1 and PL2), the mill operational based size reduction model. The discharge function takes a form
conditions were similar between the pairs. The calibrated selection of the Whiten classification efficiency equation, with its C-
functions using BM2 and PL1 were applied to BM3 and PL2 respec- parameter replaced by a Dmax parameter. The Dmax parameter
tively. The discharge function parameters (Dmax, a and d50c) were describes the maximum discharge rate for the ground product exit-
fitted to the BM3 and PL2 survey data. Note that there were signif- ing the ball mill like water. Using the size reduction model to cal-
icant differences in the ball mill feed sizes in the PGM concentrator culate the breakage result of the feed particles inside a ball mill,
surveys (F80 = 17.1 mm for PL1, and 4.2 mm for PL2) and mill and using the discharge function to calculate the ground product
throughput (89 t h
1 for PL1 and 113 t h
1 for PL2). The fitted exiting the mill, the specific energy-based ball mill model is able
and the measured product size distributions have been given in to extend its applications from batch grinding to continuous oper-

Fig. 6. The model fitted discharge functions, with the same selection functions being used in BM2 and BM3, and in PL1 and PL2 respectively.
74 F. Shi, W. Xie / Minerals Engineering 86 (2016) 66–74

ation. The ball mill model has been implemented in the SDK ver- References
sion of JKSimMet.
Five sets of industrial ball mill survey data were used to validate Austin, L.G., Klimpel, R.R., Luckie, P.T., 1984. Process Engineering of Size Reduction:
Ball Milling. SME, The Pennsylvania State University.
the ball mill model. The data were acquired from four full scale ball King, R.P., Bourgeois, F., 1993. A new conceptual model for ball milling. In:
mills covering primary and secondary grinding duties in a gold Proceedings of the XVIII International Mineral Processing Congress. Sydney, pp.
concentrator and a PGM concentrator. The ball mill feed samples 81–86.
Kojovic, T., Shi, F., Brennan, M., 2015. Modelling of vertical spindle mills part 2:
collected during the surveys were characterised using a JKFBC to integrated models for E-mill, MPS and CKP mills. Fuel 143, 602–611.
determine their breakage characteristic parameters, based on Man, Y.T., 2001. Model-based procedure for scale-up of wet, overflow ball mills, Part
which the size reduction results were calculated from the given 1: outline of the methodology. Miner. Eng. 14 (10), 1237–1246.
Morrell, S., 1992. Prediction of grinding mill power. Trans. Inst. Min. Metall., Sect. C
size-specific energy input levels. The ball mill model fits all the five
101, C25–C32.
sets of ball mill operational data well. Napier-Munn, T.J., Morrell, S., Morrison, R.D., Kojovic, T., 1996. Mineral
Since the model uses the measured ore-specific breakage char- Comminution Circuits: Their Operation and Optimisation. Julius Kruttschnitt
acteristics, it has the capacity to do simulations of the effect on Mineral Research Centre, ISBN 064628861x.
Narayanan, S.S., 1985. Development of a Laboratory Single Particle Breakage
ground product size distribution of changing ore breakage charac- Technique and its Application to Ball Mill Scale-Up, Ph.D. Thesis. The University
teristics. The model utilises separate selection function and dis- of Queensland (JKMRC), Australia.
charge function, which permits the investigation of the Pauw, O.G., Maré, M.S., 1988. The determination of optimum impact breakage
routes for an ore. Powder Technol. 54, 3–13.
influences of mill operational conditions on grinding performance. Shi, F., 2002. Development of a power draw model for estimation of the dynamic
An example of the effect of grinding density, and implicitly the rhe- recirculating load of swing hammer mills with internal classifiers. Trans. Inst.
ology, on the maximum discharge rates of fines is demonstrated Min. Metall. Sect. C, 111–119 (Proceedings Australasian Institution of Mining
and Metallurgy, vol. 307, September–December 2002).
through the model fitted discharge function. Shi, F., 2014a. Coal breakage characterisation – part 2: multi-component breakage
modelling. Fuel 117, 1156–1162.
Shi, F., 2014b. Coal breakage characterisation – part 3: applications of the multi-
component model for HGI prediction and breakage simulations. Fuel 117,
1163–1169.
Acknowledgements Shi, F., Kojovic, T., 2007. Validation of a model for impact breakage incorporating
particle size effect. Int. J. Min. Process. 82, 156–163.
Shi, F., Kojovic, T., Esterle, J., David, D., 2003. An energy-based model for swing
This work has been supported by the AMIRA P9P project. The
hammer mills. Int. J. Min. Process. 71, 147–166.
authors appreciate the assistance provided by JKMRC pilot plant Shi, F., Kojovic, T., Brennan, M., 2015. Modelling of vertical spindle mills part 1: sub-
staff in conducting the breakage characterisation tests for the pro- models for comminution and classification. Fuel 143, 595–601.
ject. Dr Grant Ballantyne of the JKMRC in providing the gold ore Shi, F., Xie, W., 2015. A specific energy-based size reduction model for batch
grinding ball mill. Min. Eng. 70, 130–140.
concentrator survey report and Ms Jenni Sweet of the UCT in pro- Shi, F., Zuo, W., 2014. Coal breakage characterisation – part 1: breakage testing with
viding the PGM ore concentrator survey data are gratefully the JKFBC. Fuel 117, 1148–1155.
acknowledged.