Minerals Engineering 74 (2015) 41–50

Contents lists available at ScienceDirect

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

Optimization of a fully air-swept dry grinding cement raw meal ball mill
closed circuit capacity with the aid of simulation
Ö. Genç ⇑
Muğla Sıtkı Koçman University, Faculty of Engineering, Dept. of Mining Engineering, Kötekli, Muğla 48000, Turkey

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

Article history: Production capacity of a fully air-swept industrial scale two-compartment KHD Humboldt WedagÒ
Received 19 August 2014 cement ball mill was optimized with the aid of simulation. It was proposed to operate the mill as a single
Accepted 9 January 2015 compartment by eliminating the pre-drying compartment. In this respect, grinding performance of the
air-swept ball mill was evaluated and modelled as a perfectly mixed single tank using the perfect mixing
ball mill modelling approach (Whiten, 1974). Static separator was modelled by efficiency curve model
Keywords: (Whiten, 1966). The empirical breakage function required in the estimation of average specific breakage
Grinding
rates was measured by drop-weight technique. The full scale model parameters were used to simulate
Classification
Modelling
the raw meal mill grinding circuit with the aid of JKSimMet Steady State Mineral Processing Simulator.
Simulation Simulation results indicated 23% production capacity increase in cement throughput in case the pre-
Optimization drying compartment was used in grinding.
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction parameters are similar except of the material transport function in

Air-swept raw meal ball mills introduced by the cement mill
Ò Ò
manufacturers F.L.Smidth (Smidth, 2002), Polysius (Polysius,
Ò
2002) and KHD Humboldt Wedag are the most commonly used
ones. KHD Humboldt WedagÒ manufactured fully air-swept raw
meal mills which have two compartments used for drying and
grinding processes. In these mills drying and grinding are per-
Ò
formed in a single mill as similar to the Polysius fully air-swept
mill (Polysius, 2002). First compartment is used as a pre-drying
compartment where it is equipped with lifters and operated with-
out grinding media in order to increase the drying efficiency. In
such systems, kiln discharge gases are used as a drying air. Drying
compartment consumes more energy as compared to the other
systems due to the high level of moisture in the feed. In air-swept
mills circulating load is carried pneumatically. Thus, the energy
consumption for a fully air-swept grinding circuit is higher by
approximately 10–12% as compared to the grinding circuit with
bucket elevator (Duda, 1985). Modelling of fully air-swept ball
mills used in the cement industry were studied with different
approaches in the literature (Austin et al., 1975, 1984;
Viswanathan, 1986; Viswanathan and Narang, 1988;
Viswanathan and Reddy, 1992; Zhang et al., 1988; Zhang, 1992;
Ergin, 1993; Apling and Ergin, 1994; Benzer, 2004). Grinding model
⇑ Tel.: +90 252 2111938; fax: +90 252 2111912.
E-mail addresses: ogenc@mu.edu.tr, omurdeng@gmail.com
the related models. The population balance model requires resi-
dence time which is difficult to determine for the full-scale mill.
Value of residence time distribution is determined experimentally.
Perfect mixing model (Whiten, 1974) simplifies the discharge
(transport) function by assuming a particle size dependent dis-
charge rate function. The discharge of any particle fraction from
the mill can be calculated on the basis of the mass of size fraction
in the mill hold-up and mass flow rate of that particle fraction out
of the mill as product. Perfect mixing model does not constitute
many grinding parameters which needs to be scaled up. The model
could be used directly to predict the performance of full-scale
mills. The relation between particle size and discharge rate depen-
dent breakage rate parameter which was defined as a ratio of
breakage rate to discharge rate function was established to mea-
sure the ball milling performance based on perfect mixing model-
ling approach by Zhang (1992), Benzer (2000) and Hashim (2003).
Breakage function and breakage rate parameters are deter-
mined by laboratory experiments in Austin’s approach (Austin
et al., 1984) and the resulting mathematical equations relating
the breakage function and breakage rate to particle size constitute
many parameters. Thus, more than one parameter set could be
produced in the solution of these equations each of which define
different breakage rate-particle size relationships. For this reason,
it is difficult to relate the effects of operating variables of ball mills
on specific breakage rates. Design and operational parameters
were studied on laboratory scale mills which need to be

http://dx.doi.org/10.1016/j.mineng.2015.01.006
0892-6875/Ó 2015 Elsevier Ltd. All rights reserved.
42 Ö. Genç / Minerals Engineering 74 (2015) 41–50
Nomenclature
i particle size fraction i
j particle size fraction j
fi mass flowrate of mill feed (ton/hour)
pi mass flowrate of mill discharge (ton/hour)
ri specific breakage rate of size fraction i (h1)
di specific discharge rate of size fraction i (h1)

di normalized discharge rate of size fraction i
a single column step triangular breakage function matrix
si mass of size fraction i (ton)
Q volumetric feed rate (m3/h)
D mill diameter (m)
L mill length (m)
scaled-up. There had been a few attempts to relate their model
with air flow through the mill, feed rate, feed size distribution,
material filling and ball filling (Viswanathan, 1986; Zhang, 1992).
Air swept ball mill model proposed by Austin et al. (1975) was val-
idated by Apling and Ergin (1994) using the industrial scale data
from a cement grinding circuit.
In this study, production capacity of a fully air-swept dry grind-
ing raw meal ball mill circuit was evaluated by modelling the mill
using the perfect mixing modelling approach (Whiten, 1972). Sta-
tic separator in the circuit was modelled by efficiency curve model
(Whiten, 1966). JKSimMet Steady State Mineral Processing Simula-
tor was used in the simulation stage. Simulation results indicated
23% capacity increase in cement throughput at the steady state
condition. However, the static separator is expected to operate
with the maximum tonnage that can be handled.
2. Methods
2.1. Sampling survey
The simplified process flowsheet of the sampled circuit with the
sampling points is given in Fig. 1. Air-swept ball mill is operating in
closed circuit with a static separator. The static fines are collected
in product cyclones where the separation of particles from the air
is performed. Product of electrofilter is combined with the cyclone
products to form final cement. Design specifications of the fully
air-swept ball mill and static separator are given in Table 1. Design
ball size distribution applied in the ball mill is given in Table 2.
Steady state condition of the circuit was verified by examining
the variations in the values of operational variables of the ball mill
and the static separator in the process control room system. Sam-
pling was started when the steady state condition was achieved.
Representative amount of samples were collected from the shown
sampling points in Fig. 1. Samples from the raw meal feed were
collected for the determination of moisture content of the mill feed
materials. Values of the operational variables were recorded in
every 5 min from the process control system to be used in the cir-
cuit performance assessment during sampling Control room
recordings and related standard deviation values at the steady
state condition are tabulated in Table 3.
2.2. Experimental
Samples were prepared by using a riffler for dry sieving from
the top size down to 150 lm. Sub-sieve sample (150 lm) was
sized in wet mode in a SYMPATHECÒ laser diffractometer. Dry
sized material (+150 lm) and wet sized sub-sieve sample
(150 lm) were combined to define the full size distribution from
the top size down to 1.8 lm. Raw meal materials were dried at
r/d⁄ ratio of breakage rate to normalized discharge rate
Eoa fraction of feed reporting to overflow
C fraction undergoing ‘‘real’’ classification (1-bypass frac-
tion)
B reduced efficiency curve fish hook parameter
d50c size of a particle in feed which has equal probability of
going to underflow or overflow (cut size)
b⁄ model parameter to preserve the definition of d50c
d particle size
x ratio of di to d50c
a reduced efficiency curve sharpness parameter
approximately 100 °C before sizing in order to carry out an efficient
screening operation. Calculated moisture contents and dry flow-
rates of mill feed materials are given in Table 4.
3. Results and discussions
3.1. Mass balancing
Measured particle size distributions and operational tonnage
flowrates were used to perform mass balance calculations around
the circuit with the aid of mass balance module of the JKSimMet
simulator to calculate the best fit estimates of the size distributions
and tonnage flowrates. Mass balanced flowrates and calculated
fineness as 0.045 mm passing % are given in Table 5. Circulating
load ratio was defined as the ratio of static separator reject tonnage
to static separator fine tonnage and calculated as 75.34%. The
results of mass balance calculations were checked out by plotting
the experimental and calculated particle size distributions
(Fig. 2). Experimental versus mass balanced particle size distribu-
tions were found to be fitted satisfactorily which indicated that,
sampling was successful and the data could be used for modelling
purpose. Experimental size distributions of final cement cyclone
collectors were presented in Fig. 3. Particle size distributions indi-
cated no segregation in the cyclones verifying the sufficient level of
air flow and balanced air distribution within the cyclones.
3.2. Mill inside sampling and granulometry
The circuit was crash-stopped to collect samples from inside of
the mill after completing sampling of the circuit streams. A view of
mill inside at the crash-stop condition is given in Fig. 4. Average
material height above the ball surface level (18 cm) and free height
of the mill (2.27 m) were measured to be used in mill powder load
(hold-up) calculation ahead of collecting the samples along the
long axis of the mill at the crash-stop condition. Mill filling was cal-
culated to be 32% using the mentioned geometrical measurements.
Photograph of the lifter bar design in the drying compartment is
presented in Fig. 5. Considerable abrasion and damage on lifters
were recognized. Whole length of the grinding compartment was
lined with classifying liners. Classifying liner configuration is pre-
sented in Fig. 4.
Sample collection dips were formed by digging out the mill
charge (mill powder + balls) approximately 40 cm below the
charge level. Samples were collected along the long axis of the mill
towards the end of the discharge grate in order to demonstrate the
size reduction performance using the inside mill size distributions
(granulometry). Samples were collected by one meter up to the
sixth meter of the grinding length whereas by half meter at the rest
of the mill length. Mill inlet and outlet temperatures were recorded
 Ö. Genç / Minerals Engineering 74 (2015) 41–50 43

Fig. 1. Simplified flowsheet of a raw meal grinding-classification circuit. Streams/sampled: (1) iron ore bunker belt; (2) clay bunker belt; (3) limestone bunker belt; (4) total
fresh feed; (7) static separator reject (coarse); (9a) product cyclone-1 underflow; (9b) product cyclone-2 underflow; (10) product cyclone combined; (12) electrofilter return;
(13) dust from cooler. Streams/not sampled: (5) mill feed; (6) mill discharge; (8) static separator fine; (11) cyclone dust.

Table 1 Table 3
Design specifications for air-swept raw meal ball mill and static separator. Control room recordings during the sampling survey.

Raw meal ball mill Operational variables Value Standard deviation

Diameter (m) 3.8
Limestone (t/h) 65 1.38
Drying compartment length (m) 2.935
Clay (t/h) 26 1.37
Grinding compartment length (m) 6.935
Iron ore (t/h) 1.72 0.13
Mill power (kW) 1600
Total fresh feed wet flowrate (t/h) 92.72 1.55
Mill rotational speed (rev/min) 15
Static separator reject (t/h) 64 14.40
Critical speed % 69
Ball mill filling % 83 1.44
Ball filling % 27
Ball mill inlet temperature (°C) 325 2.96
Discharge diaphragm middle grate aperture size (cm) 88
Ball mill discharge temperature (°C) 93 3.31
Static separator
Ball mill inlet pressure (mmSS)a 25 4.04
Separator diameter (m) 5.2
Ball mill discharge pressure (mmSS) 360 22.46
Ball mill ventilation pressure (mmSS) 767 27.79
Static separator pressure difference (mmSS) 335 19.49
Ball mill (Amper) 120 0.00
Table 2
Ball mill elevator (Amper) 24 0.00
Design ball size distribution of the grinding compartment.
Ball mill motor (kW) 1240 7.90
Ball size (mm) Weight (kg) Weight % Cumulative weight % Mill specific energy consumption (kW h/t) 14.57 0.26
Kiln capacity (t/h) 71 –
80 2853 4 100.00
a
70 17,552 22 96.45 Millimeters of water column.
60 18,594 23 74.59
50 17,040 21 51.44
40 17,442 22 30.22
30 6827 9 8.50 related to the difficulty of digging of the sample collection dip at
Total 80,308 100 the first meter due to the existing coarse balls such as 90 mm
and 80 mm which could have affected the quality of sampling. Par-
ticle size distribution of the mill inlet was found to be finer than
the first meter sample as shown in Fig. 6. This condition could be
as 325 °C and 93 °C respectively at the crash-stop condition. The due to the accumulation of static separator reject material at the
mill was cooled down for 6–7 h before inside mill sampling by mill inlet which affected the particle size distribution at the
opening the mill inlet. Air flow through the was not allowed as fine crash-stop condition. Particle size distributions of the inside mill
particles will discharge from the mill. samples and the mass balanced mill feed and discharge size distri-
It should be mentioned that, it is crucial to collect representa- butions are presented on log–log scale in Fig. 6. Particle size distri-
tive samples in any sampling operation. The technique used in this bution of the mill hold-up (mill load) was assumed to be calculated
study provided collecting representative inside mill samples at the using the average size distribution of the inside mill samples which
regarding sample collection dip. Collection of material and ball is denoted by the average mill content size distribution in Fig. 6.
samples just above the charge surface (which is common in such The mill modelling approach was to use average mill hold-up
sampling procedures) will not give statistically representative particle size distribution when calibrating the model parameters
results for the evaluation of ball charge load and distribution which of perfect mixing model proposed by Whiten (1974). Inside mill
affects the size reduction performance of the mill. Sample amount particle size distributions (Fig. 6) indicated a consistent size reduc-
collected at each sample collection dip were tabulated in Table 6. tion towards the mill discharge end such that, particle size distri-
Mill length given in Table 6 refers to the measured length at the bution of the samples became finer towards the discharge grate.
sampling condition. Particle size distribution at the mill inlet was Mill inside fineness curve established using the 0.045 mm
found to be coarser than that of the following sampling dips cumulative passing % size is given in Fig. 7. Amount of fine material
excluded of the particle size distribution of the sample collected production in the first meter decreased. However, fine material
from the first meter of the mill length. This condition could be production increased in the following two meters. No more
44 Ö. Genç / Minerals Engineering 74 (2015) 41–50

Table 4 the work of (Kolacz, 1999). Effect of air flowrate on the dis-
Moisture contents of mill feed and calculated dry flowrates. charge rate of material in an air swept ball mill was studied
Raw meals Moisture % Measured wet Dry flowrate by Kolacz (1999). It was concluded that, transportation of mate-
flowrate (t/h) (t/h) rial through the mill by air sweeping becomes more difficult if
Limestone 2.08 65 63.65 the mill content is finer which is due to the agglomeration of
Clay 22.64 26 20.11 very fine particles falling back into the mill bed,
Iron ore 4.12 1.72 1.65  material coating observed at the discharge grate could
Total raw meal 6.64 92.72 85.41
have affected the fine material accumulation amount in the
mill and decreased the grinding performance of the grinding
considerable size reduction was achieved at the rest of the mill media.
length which could be due to a series of operational factors as
given below: Particle size distribution of the mill discharge estimated by
mass balance calculations was found to be finer than that of the
 probable increase in amount of fine material due to the low air sample collected at the mill discharge end which corresponded
flow rate, such that, less fines extracted from the mill, to the sample at the seven point fourth meter of the grinding
 increase in mill inside temperature which could lead to cush- length. This condition is expected under sufficient screening effect
ioning effect as explained by Austin et al. (1984). Coating of ball of the discharge diaphragm (Fig. 6). ‘‘Screening effect’’ was
surface with material is expected to have an adverse effect on explained as the rejection of coarse particles to the last meter of
grinding performance of the grinding media thus will result in the compartment length after screening at the diaphragm and dis-
lower specific breakage rate, cussed in the literature (Benzer, 2000; Genç, 2008; Genç and
 probable agglomeration of fine particles inside the mill which Benzer, 2009) for intermediate and discharge diaphragms of over-
could have decreased the transportation (discharge) rate of flow (gravity discharge) type multi-compartment cement grinding
particles through the mill. This claim could be supported by ball mills.

Table 5
Mass balanced flowrates and fineness as 0.045 mm passing %.

Stream No Stream identification Sample amount (kg) Calculated flowrate (t/h) 0.045 mm passing %
1 Iron ore bunker belt 64.86 1.65 3.57
2 Clay bunker belt 47.36 20.07 2.79
3 Limestone bunker belt 55.77 63.41 1.58
4 Total fresh feed – 85.13 1.91
5 Mill feed – 149.26 8.00
6 Mill discharge (static separator feed) – 149.26 52.32
7 Static separator reject (coarse) 5.36 64.14 16.29
8 Static separator fine – 85.13 79.58
9a Product cyclone-1 underflow 2.56 – 80.00
9b Product cyclone-2 underflow 2.60 – 78.80
10 Product cyclone combined 2.65 82.73 78.10
11 Cyclone dust – 2.40 100.00
12 Electrofilter return 3.24 4.97 100.00
13 Dust from cooler 4.55 2.57 100.00
14 Final cement 2.42 87.70 80.71

Fig. 2. Agreement between experimental and mass balanced size distributions of the circuit streams.
 Ö. Genç / Minerals Engineering 74 (2015) 41–50 45

Fig. 3. Experimental size distributions of final cement cyclone collector products

and dust from cooler upstream.

Fig. 6. Axial mill inside particle size distributions towards the mill discharge end.

100

0.045mm cumulative passing % 80

Grinding Classifying
compartment liners 60

40
Fig. 4. Photographs of mill inside and classifying liners in the grinding
compartment.
20

0
0 1 2 3 4 5 6 7 8
Grinding compartment length (m)

Fig. 7. Fineness variation along the grinding compartment length.

enough (8  8 cm) to allow transportation of finely ground raw

meal powder by only air sweeping in the studied mill. Mill inside
size distributions demonstrated consistent size reduction. Particle
size distributions of the sample at the seventh meter of the grind-
ing compartment length was found to be considerably coarser than
that of the mill discharge (Fig. 6). Both size distributions should be
closely similar under the effective air flowrate operational
Fig. 5. A view of lifter bar liners in the drying compartment. conditions.
Another observation was the existence of coarse particle accu-
mulation in the mill. Certain amount of coarse particle accumula-
Table 6 tion within the size range of 25 + 19 mm, 19 + 13.2 mm,
Mill inside sample amounts.
13.2 + 9.5 mm was observed at the fourth meter of the mill
Length (m) Sample (kg) length. Such operational inefficiencies were attributed to the hard-
Mill inlet 10.42 ness of these particles, material coating at the discharge diaphragm
1 6.60 and low air flowrate condition as the operational air flowrate at the
2 7.96 mill outlet was recorded to be 25.9 m/s. Typical range for the air
3 5.79
flowrate at the mill outlet is 24.4–35.1 m/s for air swept ball mills
4 3.91
5 3.43 (Duda, 1985). Recorded low air flowrate at the mill outlet could
6 3.83 have decreased the grinding capacity due to the transportation of
6.7 3.72 fine material through the mill. On the otherhand, air flowrate
7.4 6.30
through the mill was calculated as 5.02 m/s using the measured
mill filling (32%) at the sampling condition. This figure was found
to be higher than the typical air flowrate range suggested for air-
Mill powder was expected to discharge through the middle swept mills which is 3–4 m/s inside the mill as given by Duda
grate of the discharge diaphragm as the grate opening was wide (1985).
46 Ö. Genç / Minerals Engineering 74 (2015) 41–50

3.3. Ball size classification

In order to assess the classifying performance of the mill liners,

ball samples were collected during the inside mill powder sam-
pling by screening out the balls over a screen with 25  25 mm
aperture size in order to separate the raw meal powder and the
grinding media at the sampling dips. Ball samples were collected
every meter, up to the fourth meter of the mill. The sampling pro-
cedure was to collect some amount of mixture of raw meal powder
and balls and then screening. Balls were retained on the screen and
collected in a sampling bag to be weighted and sized to determine
the ball size distribution along the mill length. On the otherhand,
raw meal powder which was the screen undersize was collected
in another sampling bag. Collected ball sample mass along the mill
length was tabulated in Table 7. It should be mentioned that, the
presented values are not representative of the whole ball load at Fig. 8. Weighted measured ball size along the mill length.

the sampling dip. However, the results clearly indicated the ball
size classification along the long axis of the mill. Ball size distribu-
tion was found to get finer towards the mill discharge end, except
for the sample collected at the second meter of the compartment
which indicated true ball size classification. The concept of true
ball size classification was discussed for cement grinding multi-
compartment ball mills by Genç et al. (2008). This condition shows
the affect of classifying liners. Weighted average ball size was cal-
culated using the collected ball samples at each sampling location
which demonstrated the true ball size classification along the con-
sidered mill length and given in Fig. 8.

3.4. Material characterization

Drop weight technique was used to characterize breakage dis-

tribution function of the mill feed material so as to reflect breakage
characteristics to the model parameters of the mill. Breakage test
was conducted on single particles in the size fraction of
9.5 + 8 mm at an energy level of 1 kW h/ton. A modified manual
version of a JK Tech drop weight test device (Napier Munn et al.;
Brown and Grimes, 2005) was used in the characterization tests. Fig. 9. Normalized single particle breakage functions (replotted after Genç et al.,
Specifications of the drop weight tester which was used was given 2008).
by Genç (2002) and Genç et al. (2004). It was proposed to use a
combined breakage function that was determined by combining
Standard Bond work index value of the mill feed material was
the single particle impact breakage functions of individual compo-
also experimentally determined as 11.03 kW h/ton according to
nents of the mill feed using the weight percentages of the mill feed
TS 7700 standard (TS 7700, 1989) using a 90 lm test sieve.
components (Genç and Benzer, 2008) in modelling of cement
grinding mills. The combined breakage function determined on
the basis of the mentioned assumption and is shown in Fig. 9 as 3.5. Ball mill model
total feed combined. Mill feed is composed of 60% clinker, 24%
trass, 11% limestone and 5% gypsum by weight and used to deter- Air-swept ball mill was modelled using the perfect mixing mod-
mine the combined breakage function. Combined breakage distri- elling approach (Whiten, 1974) which defines the comminution
bution was found to be shifted towards the breakage function of process in terms of three parameters; breakage rate, discharge rate
the dominant component of the mill feed which was clinker. and breakage function Eq. (1). On the other hand, discharge rate
Investigated raw meal mill feed constitutes 74% limestone, 24% (di) of particles were defined to be a function of mill product (pi)
clay and 2% iron ore by weight. According to the recorded findings and mill hold-up (si) as given by Eq. (2) (Napier Munn et al.)
(Genç and Benzer, 2008), the approach was to use single particle
breakage distribution function of limestone which is the major X
i
f i  pi ðr i =di Þ þ aij pi ðr i =di Þ  pi ¼ 0 ð1Þ
component of the raw meal mill feed to estimate the average j¼1
breakage function and presented in Fig. 9.
di ¼ pi =si ð2Þ
Table 7
In these equations, fi and pi are the mass flowrates (t/h) of size
Ball sample amounts along the mill length.
fraction i in mill feed and product respectively, aij is the breakage
Grinding compartment length (m) Total sample weight (kg) function (in the form of single column step triangular matrix), ri
1 26.27 is the specific breakage rate of size fraction i (tonnes broken per
2 33.14 hour per tonne in the mill which is h1), di is the specific discharge
3 28.80
rate of size fraction (i) (tonnes discharged per hour per tonne in the
4 15.43
mill which is h1), and si is the mass of size fraction (i) inside the
 Ö. Genç / Minerals Engineering 74 (2015) 41–50
mill as tons. Perfect mixing model was used by Benzer (2004) in
modelling of an air-swept raw meal grinding ball mill by consider-
ing the single compartment mill as three perfectly mixed tanks
whereas air-sweeping through the mill was modelled by a classi-
fier at the mill discharge. In the model, tank-1 corresponded to
the mill length where lifting liners were applied whereas tank-2
and tank-3 lengths corresponded to the mill length where classify-
ing liners were applied. In the related study, mill performance was
evaluated through particle size versus r/d⁄ combined breakage rate
parameter which normalized the discharge rate effect.
In order to correct the variations in residence time, di is scaled
in terms of the mill volume and volumetric feed rate (Q) to the

term di using Eq. (3), where D and L are the diameter and the
length of the mill respectively. Then, r/d⁄ model parameter is calcu-

lated. Normalized discharge rate (di ) is a function of particle size
Eq. (3) (Napier Munn et al.)
 di
di ¼
4Q =D2 L
Normalized discharge rate function variation established using
the estimated mill hold-up (si) was given for the investigated air-
swept raw meal mill in Fig. 10. Experimentally determined values
(measured) are denoted by the scatter plot and compared with the
typical trend observed in semi-autogenous grinding mills (SAG)
(Napier Munn et al.; Leung, 1987) which is denoted by the dotted
lines in Fig. 10. This function was calculated by eliminating the
classification effect of the discharge grate. The discharge rate func-
tion (di) was considered to be the product of two mechanisms;
transport and classification by the discharge grate as explained
for SAG mills by Leung (1987).
There is a critical particle size in the mill which is denoted by xc

and can be determined using normalized discharge rate (di ) func-
tion as shown in Fig. 10. Particles finer than this size (xc) behave
like a fluid medium in the mill and discharge at a constant rate
through the mill. The rate of discharge for particles coarser than
this size was found to decrease systematically in wet grinding
conditions (Napier Munn et al.; Morrell and Man, 1997). Particles
coarser than the grate size (xg) remain in the mill for further size
reduction where the discharge rate equals to zero. In the investi-
gated air-swept mill, the fluid medium corresponded to air and
the critical particle size (xc) was expected to be highly depended
on the airflow rate through the mill (Fig. 10).
In this study, the modelling approach was to consider the mill
as a perfectly mixed single tank as the whole length of the mill
was lined with classifying liners. Specific discharge rate functions
(di) were calculated from Eq. (2) using the estimated mill
hold-up. Specific breakage rate (ri) function was estimated using
the calculated discharge rate functions from Eq. (1). Mill hold-up
(tons in each size fraction) in grinding compartment was
 E = (1/2)C where E denotes the fraction of feed.
Fig. 10. Measured normalized discharge rate function (di ) in a full scale fully air-
swept raw meal mill (xc = 50 lm). Replotted after (Genç et al., 2008).
47
Fig. 11. Specific breakage rates (ri) in the air-swept raw meal ball mill. Replotted
after (Genç et al., 2008).
ð3Þ
calculated using the size distribution of average mill content and
measured mill filling data at the crash-stop condition. The specific
breakage rate calculation procedure was formulated on ExcelÒ
spreadsheets.
Specific breakage rate function is presented in Fig. 11. Agree-
ment between experimental and back-calculated mill product size
distributions are given in Fig. 12. The experimental data was found
to be fitted to the model satisfactorily. Specific breakage rates were
assumed to not change along the mill in the modelling approach.
The r/d⁄ combined breakage rate parameters of the perfect mix-
ing model were calculated as ln(r/d⁄) in the model fit module of the
JKSimMet simulator considering the mill as a perfectly mixed sin-
gle tank. The fitted values were of the best values that defined the
mill discharge size distribution. The r/d⁄ breakage rate parameters
fitted to the perfect mixing model were tabulated in Table 8 and
used in the simulation step which characterized the specific break-
age rates in the mill. It should be mentioned that, spline function
knot values, which could be defined usually by maximum of four
data points, were selected from the whole set of specific breakage
rate values calculated for each particle size given in Fig. 11.
3.6. Static separator model
Grinding efficiency in ball mills depends on the classifying per-
formance of air separators as explained in the study of (Klumpar
and Slavsky, 1989) and (Kolacz, 1999). Their findings indicated
that, energy consumption in ball milling can be reduced if the clas-
sification efficiency is sufficiently high. The classification behavior
of air separators are described using the efficiency curve concept in
the literature (Austin et al., 1975; Zhang et al., 1988; Zhang, 1992;
Benzer, 2000; Luckie and Austin, 1975; Schneider et al., 1983;
Kuhlmann, 1984; Dunn, 1985; Plank, 1985; Kellett and Rock,
1986; Benzer et al., 2001; Hashim, 2003; Günlü, 2006; Altun,
2007). The mathematical equation of the efficiency curve model
is given in Eq. (4) (Napier Munn et al.).
 
ð1 þ bb xÞðexpðaÞ  1Þ
Eoa ¼ C ð4Þ
expðab xÞ þ expðaÞ  2
where,
Eoa: fraction of feed reporting to overflow.
C: fraction undergoing ‘‘real’’ classification (1-bypass fraction).
a: reduced efficiency curve sharpness parameter.
b: reduced efficiency curve fish hook parameter.
b⁄: parameter to preserve the definition d50c, i.e. d = d50c when
x: ratio of particle size d to corrected size d50c.
48 Ö. Genç / Minerals Engineering 74 (2015) 41–50
Fig. 12. Agreement between experimental and calculated (model fitted) mill
product size distributions.
Table 8
ln(r/d⁄) combined model parameters of the ball mill.
Particle size (mm) ln(r/d⁄)
1.18 4.00
0.425 1.78
0.15 0.83
0.045 0.23
Fig. 13. Efficiency curve (tromp) for static separator (d50 = 0.099 mm; by-
pass = 11.85%; fish-hook = 2.39%).
d50c: size of a particle in feed which has equal probability of
going to underflow or overflow (cut size)
The fraction of feed reporting to underflow (EUA) was defined as
1Eoa (Napier Munn et al.). The separator performance can be
modelled in terms of d50c, C, a and b. It was stated that, b controls
the initial rise in the efficiency curve at fine sizes, while a deter-
mines the slope at larger values of d which is around d50c. b⁄ is cal-
culated iteratively during the fitting of Eq. (4) Whiten, 1966. Effects
of operational parameters on efficiency curve model parameters
were given for air separators used in the cement industry by
Günlü (2006), Altun (2007) and Benzer et al. (2001). The efficiency
curve (tromp curve) for the static separator established on the
basis of the mass balanced size distributions is presented in
Fig. 13. The characteristic efficiency curve parameters which are
d50, by-pass and fish-hook are also given in Fig. 13.
Fish-hook parameter characterizes the difference between the
maximum percentage of fine material amount that appears in
coarse stream (underflow of the separator) and the by-pass per-
centage. Model fitted efficiency curve parameters used in the sim-
ulation of the circuit are given in Table 9. The separator
performance is not at maximum as 11.85% of feed reports to sepa-
rator coarse product. However, this value is reasonable and classi-
fication performance of the static separator is sufficiently high.
4. Simulation
Simulation model of the circuit was designed in simulation
module of the JKSimMet simulator by defining the perfect mixing
model parameters of the air-swept ball mill and efficiency curve
model parameters of the static separator given in Tables 8 and 9
respectively. The ball mill was simulated as a single compartment
mill by eliminating the mill length of 2.935 m which was used in
drying stage, such that the full length (L = 9.87 m) of the mill was
used in grinding. Thus, drying of the raw meal outside the mill
by an appropriate dryer was assumed. Static separator perfor-
mance was sufficiently high and assumed to not change at the sim-
ulated condition. Cyclones are used to separate static fines from
gas and store static separator fine product (cement). There is not
any classification. Thus, cyclones were excluded in the simulation
model whereas electrofilter return was identified as a stream.
The circuit response to the proposed operational condition in
terms of tonnage flow rates and fineness (0.045 mm passing per-
centage) is presented in Table 10. Mass balanced particle size dis-
tributions in comparison to those obtained after simulation
(simulated) at 23% capacity increase case in the cement through-
put are given in Fig. 14. Simulation parameters were kept constant
during the optimization study.
As a consequence of the proposed modification in the mill and
the expected capacity increase, a series of operational modifica-
tions will be required such that, regulation of static separator oper-
ational parameters. For instance, particle size distribution of the
static separator feed (mill discharge) is estimated to become finer
as indicated by the simulated particle size distributions which will
require the optimization of the static separator. Parameters that
can be adjusted in the classification process to attain the target
fineness were recorded in the literature by Kohlhaas (1983) as:
 varying of the air flow rate; increase in air flow rate will
decrease the cut size (d50),
 adjusting of the deflector over the bottom of the inlet duct
through which the powder carrying air enters the separator;
position of the deflector can be adjusted which will effect the
cut size (d50),
 adjusting of the top outlet duct; where the cut size can be var-
ied by vertical adjustment of the air outlet duct at the top of the
separator. For a constant air flow rate, increase in the length of
the duct will lead to decrease the cut-size (finer product) or vice
versa.
Air flow rate in the duct of the mill should be increased before
the adjustment of the static separator parameters (i.e., angle set-
ting adjustable vanes, deflector) by controlling the by-pass
amount. The cyclone performance will change depending on the
cyclone geometry such that, as the cyclone diameter decreases
and the length of the conical section increases, centrifugal force
Table 9
Model fitted efficiency curve parameters used in the simulation of circuit.
Model parameter Value
d50c 0.1069
C (1-by-pass) 85.15
a 3.74
b 0.3633
b⁄ 1.16
 Ö. Genç / Minerals Engineering 74 (2015) 41–50
Table 10
Comparison of crash-stop and simulated cases.
Crash-stop condition (Calc)
Stream flows t/h
Total fresh feed 85.13
Mill discharge 149.26
Static separator reject 64.14
Electrofilter return 4.97
Final cement 90.10
Fig. 14. Agreement between mass balanced and simulated particle size distribu-
tions of streams.
effect on the particle flow pattern will increase and will lead to
effective separation of powder carrying air as explained by
Kohlhaas (1983). Separation efficiency of the cyclones will
decrease at very low or high grain concentrations. Based on the
simulation results, cyclone and electrofilter capacities are expected
to handle 23% capacity increase in addition to the increase in the
dust concentration in the product cyclone overflow. However, the
cyclone will be operated at full capacity.
5. Conclusions
Conventional two-compartment fully air-swept KHD Humboldt
WedagÒ raw meal ball mill operating in closed circuit with a static
separator was modelled and simulated to evaluate the probable
capacity increase in the circuit in case the pre-drying compartment
was used in the grinding stage. The mill was modelled as a per-
fectly mixed single tank as the material discharge was provided
only by air-sweeping. Performance of the separator was assumed
to not change in the simulation stage.
Simulation results indicated that, 23% capacity increase in the
cement throughput could be achieved at the steady state condition
by operating the pre-drying compartment at the same ball charge
level and ball size distribution, without any change in the product
cyclone capacity, and by assuming that the process of pre-drying is
performed in the ball mill upstream. However, air flowrate through
the mill should be critically regulated as the velocity of the air con-
trols the particle size distribution of the mill product in addition to
the operational parameters of the static separator at the new oper-
ational condition for a stable and optimum production rate. Grind-
ing heat generated could increase which may lead to
agglomeration of particles unless reduced. The new design may
require larger dust collectors, larger ventilation fans which will
bring additional cost.
49
Simulated condition (Sim)
0.045 mm passing % t/h 0.045 mm passing %
1.91 105.00 1.91
52.32 165.27 58.21
16.29 60.27 21.43
100.00 4.97 100.00
80.71 109.97 82.41
Acknowledgements
Author’s appreciation goes to SET Italcementi Group Balıkesir
Plant for providing the access to the plant and their valuable sup-
port during the sampling survey. Prof. A. Hakan Benzer for his valu-
able discussions and contributions, Assistant Prof. Okay Altun and
Assistant Prof. Hakan Dündar from Hacettepe University are also
gratefully acknowledged.
References
Altun, O., 2007. Comparison of different efficiency curve approaches in modelling of
air classifiers. MSc Thesis, Mining Engineering Department, Hacettepe
University, Ankara, Turkey.
Apling, A.C., Ergin, H., 1994. Validation of a grinding model for a full-scale air swept
cement mill. Demirel-Ersayın (Ed.), Progress in Mineral Processing Technology,
1994, pp. 527–538.
Austin, L.G., Luckie, P.T., Wightman, D., 1975. Steady state simulation of a cement
milling circuit. Int. J. Miner. Process. 2, 27–150.
Austin, L.G., Klimpel, R.R., Luckie, P.T., 1984. Process Engineering of Size Reduction:
Ball Milling. AIME Publication, NY, pp. 205–290, 51–119.
Benzer, A.H., 2000. Mathematical Modelling of Clinker Grinding Process. PhD Thesis,
Hacettepe University, Ankara, Turkey.
Benzer, A.H., 2004. Modelling and simulation of a fully air swept ball mill in a raw
material grinding circuit. Powder Technol. 150, 145–154.
Benzer, A.H., Ergün, S ß .L., Öner, M., Lynch, A.J., 2001. Simulation of open circuit
clinker grinding. Miner. Eng. 14 (7), 701–710.
Benzer, H., Ergun, L., Lynch, A.J., Oner, M., Gunlu, A., Celik, I.B., Aydogan, N., 2001.
Modelling cement grinding circuits. Miner. Eng. 14 (11), 1469–1482.
Brown, D., Grimes, A., 2005. JK tech drop weight testing device. Operating Manual,
JKMRC, The University of Queensland, Brisbane, Australia.
Duda, W.H., 1985. Cement-Data-Book, International Process Engineering in the
Cement Industry, vol. 1,, French & European Publication, pp. 80–656.
Dunn, M.R., 1985. A method for analysing the performance of a mechanical air
separator. World Cem., 327–332
Ergin, H., 1993. The Simulation of an Air Swept Cement Grinding Mill. PhD Thesis,
University of Leeds.
Genç, Ö., 2002. An Investigation of the Breakage Distribution Functions of Clinker
and Additive Materials. MSc Thesis, Hacettepe University, Mining Engineering
Department, Turkey (in Turkish).
Genç, Ö., 2008. An Investigation on the Effect of Design and Operational Parameters
on Grinding Performance of Multi-compartment Ball Mills used in the Cement
Industry. PhD Thesis, Hacettepe University, Mining Engineering Department,
Turkey (in English).
Genç, Ö., Benzer, A.H., 2008. Analysis of single particle breakage characteristics of
cement clinker and cement additives by drop-weight technique. J. Chamber
Min. Eng. Turkey 47 (1), 13–26.
Genç, Ö., Benzer, A.H., 2009. Single particle impact breakage characteristics of
clinkers related to mineral composition and grindability. Miner. Eng. 22 (13),
1160–1165.
Genç, Ö., Benzer, A.H., Ergün, S ß .L., 2004. Single particle breakage characterization of
materials by drop weight testing. Physicochem. Probl. Miner. Process. Pol. 38,
241–255.
Genç, Ö., Benzer, A.H., Ergün, S ß .L., 2008. Effect of ball size classification on specific
breakage rates of particles in full-scale cement grinding tube mills. In: 24th
International Mineral Processing Congress Proceedings, Beijing, China, pp. 529–
539.
Günlü, A., 2006. Mathematical modelling of air separators used in cement grinding
circuits. PhD Thesis, Mining Engineering Department, Hacettepe University,
Ankara, Turkey.
Hashim, S., 2003. Mathematical modelling the two-compartment mill and
classification. PhD Thesis, Julius Kruttschnitt Mineral Research Centre, The
University of Queensland, Australia.
Kellett, D., Rock, H.G., 1986. Operating experience with the O-Sepa air separator.
Zement-Kalk-Gips, (Translation of No. 6/86), 234–235.
Klumpar, I.V., Slavsky, S.T., 1989. Designing and testing air classifiers. Int. J. Eng.
Fluid Mech. 2 (2), 119–140.
Kohlhaas, B., 1983. Cement Engineers’ Handbook, fourth English ed.
50 Ö. Genç / Minerals Engineering 74 (2015) 41–50
Kolacz, J., 1999. Control of the mill charge behaviour in dry tumbling mills. Miner.
Eng. 12 (1), 51–64.
Kuhlmann, K., 1984. Significance of classification in cement grinding results of a
balance analysis of grinding circuit. Zement-Kalk-Gips (Translation of No. 9/84),
257–260.
Leung, K., 1987. An energy-based ore specific model for autogenous and semi-
autogenous grinding. PhD Thesis, JKMRC, The University of Queensland,
Australia.
Luckie, P.T., Austin, L.G., 1975. Mathematical analysis of mechanical air separator
selectivity curves. Trans. IMM 84, 253–255.
Morrell, S., Man, Y.T., 1997. Using modelling and simulation for the design of full
scale ball mill circuits. Miner. Eng. 10 (12), 1311–1327.
Napier Munn, T.J., Morrell, S., Morrison, R.D., Kojovic, T., 2005. Mineral
comminution circuits their operation and optimization. JKMRC Monograph
Series in Mining and Mineral Processing, No. 2, The University of Queensland,
Brisbane.
Plank, F.W., 1985. Influence of the separating effect of the classifier on the capacity
of the closed-circuit grinding plant. Zement-Kalk-Gips, (Translation of No, 1/
85), 62–65.
Polysius, 2002. Tube mills for dry grinding. Krupp Polysius Manufacturer’s
Catalogue.
Schneider, L.T., Eickholt, H., Blasczyk, G., 1983. Influence of separators on grinding
systems operating results. Zement-Kalk-Gips (Translation No. 12/83), 26–31.
Smidth, F.L., 2002. TiraxÒ Raw Mill. F.L. Smidth Manufacturer’s Catalogue.
TS 7700, 1989. Determination Method of Grinding Work Index, Ankara.
Viswanathan, K., 1986. Computer based models for grinding and industrial case
studies. Aufbereitungs Technik 27, 560–572.
Viswanathan, K., Narang, K.C., 1988. Computer simulation and optimization of ball
mills/circuits. World Cem. 19, 143–148.
Viswanathan, K., Reddy, C.S., 1992. Mathematical modelling of the effect of air
ventilation on solids recycle rate and ball mill performance. World Cem., 20–26
Whiten, W.J., 1966. Winter School on Mineral Processing. Dept. of Min. and Met.
Eng, The University of Queensland, Australia.
Whiten, W.J., 1972. Simulation and Model Building for Mineral Processing. PhD
Thesis, The University of Queensland, Australia.
Whiten, W.J., 1974. A matrix theory of comminution machines. Chem. Eng. Sci. 29,
588–599.
Zhang, Y.M., 1992. Simulation of Comminution and Classification in Cement
Manufacture. PhD Thesis, JKMRC, The University of Queensland, Australia.
Zhang, Y.M., Napier-Munn, T.J., Kavetsky, A., 1988. Application of comminution and
classification modelling to grinding of cement clinker. Trans. Inst. Min. Metall.
Sect. C 97, 207–213.