ISSN 2176-3135

SIMULATING A PILOT-SCALE DRY BALL MILL GRINDING

ITABIRITE USING BATCH GRINDING DATA1
Alessandro Luiz de Oliveira2
Luís Marcelo Marques Tavares3
Abstract
Batch grinding tests in lab mills have been widely and, to a great extent, successfully
used in the design and optimization of large-scale continuous mills using a couple of
popular scale-up procedures. Some researchers, however, argue that the size
distribution of ore within a continuous mill is not the same as the one that is tested in
batch mode in the laboratory, questioning the basis of such approaches. Indeed, in
spite of their widespread use, data are surprisingly scarce that demonstrate the
fidelity of the methods that use batch grinding data to predict continuous mill
performance, in particular when considering Brazilian iron ores. The paper compares
predictions of a continuous pilot-scale mill using the population balance model and
the Austin and co-workers scale-up procedure to data from grinding of an Itabirite
iron ore and a spent catalyst used in the oil industry. It shows that the deviations
encountered between measured and calculated 80% passing sizes for the mill
operating under a variety of grinding conditions were, on the average, 7.5 µm,
demonstrating the validity of the method. The benefit, however limited, of using an
improved prediction of mill hold up of solids is also demonstrated.
Key Words: Modeling; Population balance; Simulation; Grinding.

SIMULAÇÃO DE UM MOINHO PILOTO A SECO NA MOAGEM DE ITABIRITO USANDO

DADOS DE MOAGEM DESCONTÍNUA
Resumo
Ensaios de moagem em batelada em moinhos de laboratório têm sido utilizados
amplamente e, em grande parte, com sucesso tanto no projeto quanto na otimização de
moinhos contínuos, usando alguns procedimentos populares de escalonamento. Alguns
pesquisadores, no entanto, afirmam que a distribuição granulométrica do minério dentro de
um moinho contínuo não é a mesma que aquela testada nos moinhos em operação em
batelada, contestando a base de tais abordagens. De fato, apesar do uso disseminado de
métodos de escalonamento que usam os dados da moagem em batelada para prever o
desempenho de moinhos contínuos, dados são surpreendentemente escassos que
demonstram a fidelidade dessas previsões, especialmente considerando minérios de ferro
brasileiros. O trabalho compara as simulações da moagem de um moinho piloto contínuo,
usando o modelo do balanço populacional e o método de escalonamento proposto por
Austin e colaboradores para dados de moagem de um minério de ferro Itabirítico e de um
catalisador gasto. Os resultados mostram que os desvios encontrados entre os valores
medidos e calculados para o tamanho 80% passante para o moinho operando sob uma
variedade de condições de moagem foram, em média, iguais a 7,5 µm, demonstrando a
eficácia do método. O benefício, ainda que limitado, do uso de previsões mais precisas da
massa de hold up de minério no moinho, é demonstrado.
Palavras-chave: Modelagem; Modelo do balance populacional; Simulação; Moagem.

1
Technical contribution to the 6th International Congress on the Science and Technology of
Ironmaking – ICSTI, 42nd International Meeting on Ironmaking and 13th International Symposium on
Iron Ore, October 14th to 18th, 2012, Rio de Janeiro, RJ, Brazil.
2
Materials Engineer, M.Sc. candidate, PEMM-COPPE, UFRJ, RJ, Brazil.
3
Mining Engineer, M.Sc., Ph.D., Associate Professor, Programa de Engenharia Metalúrgica e
de Materiais, COPPE, UFRJ, RJ, Brazil.

733
 ISSN 2176-3135

1 INTRODUCTION

The population balance model (PBM) was first applied 40 years ago to milling as an
alternative to the Bond method, which lumps all parameters that influence grinding in
both a mill and a classifier operating in closed circuit into a single parameter, the
work index. The PBM first had the merit of decoupling the mill and the classifier, and
then of accounting for the different functions that describe breakage, internal
classification and transport within ball mills.(1) As natural consequences of this
approach, a number of researchers have proposed scale-up approaches that rely on
batch grinding tests from which functions that characterize the ore-dependent
breakage response can be estimated.
The present work compares predictions of a continuous pilot-scale mill using the
population balance model and the scale-up procedure proposed by Austin and co-
workers to data from grinding both an Itabirite iron ore and ECAT, a spent catalyst
from petroleum processing.

2 MATERIALS AND METHODS

2.1 Materials

Samples used in the tests included an Itabirite iron ore from the Iron Quadrangle
(Minas Gerais) and ECAT, a spent catalyst used in oil cracking from a plant in Rio de
Janeiro. A summary of the samples characteristics is presented in Table 1. The feed
size of the Itabirite ore is consistent with that of a typical concentrate for pellet feed
production.
The angle of repose of the samples was measured as a means of accounting for the
widely different flowabilies of the materials analyzes, being measured using a
procedure described elsewhere.(2)

Table 1. Summary of the samples characteristics

Specific gravity ρs F80 Angle of repose
Sample % - 10 µm
(g/cm3) (µm) (o)
Itabirite 3.64 177 8.0 31.6
ECAT 2.47 117 0.8 18.6

2.2 Grinding Tests

The mill used in both batch and continuous tests is illustrated in Figure 1. The mill
has 31 cm of diameter (D) and 32 cm of length (L). In batch mode, both the feed and
discharge ends were blinded, so as to prevent the discharge of material while running
the mill. Tests were then conducted for a number of grinding times, all using a
constant steel ball size (15 +/- 2 mm).
In continuous tests, the feed was introduced using a screw vibratory feeder, by
Vibrascrew® Inc, at a constant rate and, when steady-state conditions were reached,
samples of the mill discharge were collected for analyzes. The mill is equipped with a
grate discharge, with 8% open area, with circular openings of 7 mm of diameter. In
order to allow the accurate measurement of the discharge rate of solids leaving the
mills, a precision scale (Gehaka BG 8000) was positioned below the discharge
trough. The attainment of steady-state conditions in the continuous ball milling tests
was verified by both comparing feed and discharge rates, and analyzing the stability
in the mill hold-up value, measured using load cells.

734
 ISSN 2176-3135

After all tests size analyzes were conducted by laser scattering using a Malvern
Mastersizer® 2000 of the samples dispersed in water.

Figure 1. Schematic diagram of the mill.

3 RESULTS AND DISCUSSION

3.1 Batch Grinding

Grinding can be described using the size-mass balance model which, for a batch
grinding process, is given by Austin, Klimpel and Luckie. (1)
dw i i 1
 S i w i   S j bij w j
dt j 1
(1)
Where wi is the fraction of particles contained in size interval i, Si is the specific
breakage rate and bij is the discrete breakage function. By using appropriate values
of these functions the entire size distributions generated after different grinding times
may be fitted to experimental data.
The specific breakage rate is significantly influenced by particle size, being
successfully described by Austin, Klimpel and Luckie. (1)

x 
S i  S1  i 
 xo
(2) 
Where xo is the reference size, taken as 1 mm. In the equation, no correction for
abnormal breakage of the coarse material was introduced, given the fine size of the
feed in the present work.
The breakage distribution function, on the other hand, is well described in its most
general form as: (3)
 
x  x 
Bij   i 1   (1   ) i 1 
 x   x 
 j   j 

for xi-1 > D*(3)

and

735
 ISSN 2176-3135

 
 x i 1   x i 1   x 
Bij       (1   ) i 1 
 x   D*   x 
 j   j 

for x i 1  D *
(4)
The breakage distribution function in density form is simply computed from
bij  B( xi ; x j )  B( xi 1; x j )
bii  0 (5)
Parameters in Equations (2-4) have been fitted to batch grinding data of the Itabirite
iron ore and ECAT in the mill depicted in Figure 1 (Table 2), and results are shown in
Figure 2, which demonstrates the very good fit to the data. The relationship between
the solids hold up in the mill (M) and the interstitial filling (U) is given by Austin and
Tangsathitkulchai(4)
M  0.24  D 2 L J U  s / 4
(6)
Where D is the mill diameter, L the length, J the ball filling and ρs is the solids
density.

Table 2. Summary of breakage characteristics of the materials studied

Breakage rate function Breakage distribution function
Sample
S1 (min-1) α     D* (µm)
Itabirite 2.95 1.90 1.00 0.04 0.56 0.49 5.30
ECAT 11.27 1.66 1.00 0.99 0.60 1.11 2.96

100
Feed

07 min Measured
80
15 min Measured
Cumulative Passing (%)

30 min Measured
60

40

20

0
0,001
0,010 0,100 1,000
Particle Size (mm)
Figure 2. Comparison between measured and fitted size distributions from batch grinding of Itabirite
iron ore (J = 0.30, U = 1.04 and φc = 0.75): symbols are experimental data and lines represent the
model fit.

736
 ISSN 2176-3135

From these, the model can be used to predict the size distributions from grinding,
departing from a feed containing a given size distribution. However, in order to do
that, it is necessary to describe the pattern of mixing of solids within the mill. The
residence time distribution for the mill in question has been measured (5) and Figure 3
shows a comparison between measured results and fitting using a model that is
based on three mixers in series, given by Oliveira, Carvalho e Tavares.(5)
1( 2   3 ) exp(  t 1 )   2 ( 3  1 ) exp(  t  2 )   3 (1   2 ) exp(  t  3 )
E (t ) 
(1   2 )( 3  1 )( 2   3 ) (7)
Where 1, 2 and 3 are the mean residence times in the three mixers. Given as a
function of the total residence time within the mill (), the values were 1/ = 0.1, 2/
= 0.83, 3/ = 0.07.

120
Cumulative weight of Zn in

90
discharge (g)

60

30

0
0 30 60 90
Time (min)

Figure 3. Cumulative residence time distribution of solids within the mill from Figure 1 (5): symbols
represent experimental data and the line represents the fit to the model corresponding to three mixers
in series (Equation 7)

Consistent with this model, it is now possible to describe grinding in three continuous
perfectly mixed regions operating under steady-state conditions as
0.1M i 1 0.1M
w i(1)  w iF  Si w i(1)   bij S j w (j1)
W j 1 W (8)
0.83 M i 1 0.83 M
w i( 2)  w i(1)  Si w i( 2 )   bij S j w (j 2)
W j 1 W (9)
0.07 M i 1 0.07 M
w i  w i( 2)  Si w i   bij S j w j
W j 1 W (10)
F
Where w i is the feed size distribution, w i is the size distribution of the product
(1) ( 2)
leaving the mill, whereas w i and wi are the size distributions of the material leaving
mixing zones 1 and 2, respectively. W is the feed flow rate of solids and M is the
solids hold-up within the mill. Equations (8-10) can be solved recursively and

737
 ISSN 2176-3135

sequentially, and a Matlab® code was written for that purpose. It is important to note
that no exit classification was introduced in the model. This is justified by the large
grate openings present in the mill and the fact that no forced suction of air (air
sweeping) was used during the tests.
Unfortunately, the PBM gives no information a priori that allows predicting how these
functions, in particular the breakage rates, are affected by milling conditions, that
would make it usable for simulating the continuous mill under conditions that differ
from those tested in the batch mill. In order to overcome this limitation, empirical
expressions have been proposed for this purpose.(1,6) The approach used in the
present work is the one proposed by Austin and co-workers. It is essentially based on
the empirical scale-up of the breakage rate function, using the relationship (1)
S1  1  6.6J b2.3   c  0.1  1  exp15.7( cb  0.94 ) 
    0.1  1  exp15.7(  0.94 )  exp1  1.32(U  U b )

S1b  1  6.6J 2.3  cb  
c
(11)
Where subscript b corresponds to experimental conditions used in the batch test mill.
Since the same mill diameter is used both in the batch and continuous grinding tests,
then the mill size correction has been removed from the equation.
One potential challenge identified in the application of the scale-up procedure by
Austin and co-workers is the prediction of the mill hold-up of solids (M) for a given set
of milling conditions and feed rate (W). Although this was actually measured directly
in the tests, it is initially assumed in the simulations that it was maintained constant at
U = 1, leading to values of M that are estimated using Equation 6. As such, the
continuous mill was simulated and Figure 4 compares measured to calculated size
distributions from the continuous mill operating under a range of feed sizes, showing
the good agreement between the two, considering that the model was not fitted to
this data. A large number of tests, covering a range of values of J and φc has been
conducted with the two materials studied, and Figure 5 compares the P80 (80%
passing size) values from experiments and simulations. It is evident that the general
agreement is good, with an average standard error of 7.5 µm in the predictions.

100

80
Cumulative Passing (%)

60

40

J = 0,30 / φc = 0,75 / W = 9 kg/h

20 J = 0,20 / φc = 0,75 / W = 18 kg/h
J = 0,20 / φc = 0,30 / W = 24 kg/h
0
0,01 0,10 1,00
Particle Size (mm)
Figure 4. Comparison of measured (symbols) and predicted (lines) particle size distributions from
continuous grinding tests of Itabirite iron ore: symbols represent experimental data and lines model
predictions.

738
 ISSN 2176-3135

160
J = 0.20 / φc = 0.75
140 J = 0.20 / φc = 0.30
J = 0.30 / φc = 0.75
Predicted P80 (um) 120 J = 0.30 / φc = 0.30

100

80

60

40
40 60 80 100 120 140 160
Measured P80 (um)
Figure 5. Comparison between measured and predicted P80 values for tests for the two
materials studied, assuming U = 1. Dashed lines represent the standard deviations of the
predictions.

Finally, Figure 6 shows the variation of interstitial filling (U) as a function of grinding
conditions, including feed rate. It shows that values obtained for iron ore are, on the
average, higher than those found for ECAT, and also the values differ significantly
from the assumed value of U = 1 consider in the previous simulations. As such,
simulations of batch grinding were then conducted considering the actual values of
mill hold up and comparisons are shown in Figure 7 between measured and
simulated P80 values. It is evident that deviations between actual and simulated
values are reduced if the actual hold-up values are known. Indeed, these errors
reduce to 6.7 μm in that case. Work is ongoing in the authors´ laboratory to develop
an appropriate model that will be able to describe that.

2,5
J = 0.20 / φc = 0.75
J = 0.20 / φc = 0.30
Fraction of Interstitial Filling - U

2,0 J = 0.30 / φc = 0.75

J = 0.30 / φc = 0.30
J = 0.30 / φc = 0.30
1,5

1,0

0,5

0,0
0 10 20 30
Feed rate - W (kg/h)
Figure 6. Interstitial filling (U) values measured in continuous tests for iron ore and ECAT as a
function of feed rate.

739
 ISSN 2176-3135

160
J = 0.30 / φc = 0.75
140 J = 0.20 / φc = 0.75
J = 0.20 / φc = 0.30
Predicted P80 (um) 120 J = 0.30 / φc = 0.30

100

80

60

40
40 60 80 100 120 140 160
Measured P80 (um)
Figure 7. Comparison between measured and predicted P80 values for tests conducted using the two
materials studied (U calculated from measured M). Dashed lines represent the 6.7 µm error band.

4 CONCLUSIONS

The size-mass balance model has been successfully used to describe batch grinding
data of an Itabirite iron ore and a spent catalyst.
Considering the same mill, now operating in continuous mode, as three mixers in
series, experiments and simulations have been compared for a range of fractions of
critical speeds, fillings and feed rates with good agreement, but with deviations as
high as 9.6 µm between measured and calculated P80s. It is demonstrated that
improved predictions can be achieved if accurate estimates of the mill interstitial
filling (U) are available as a function of operating characteristics.

Acknowledgements

The authors would like to acknowledge the support of CNPq and CAPES to this
investigation. Also, the assistance of personnel from the Laboratório de Tecnologia
Mineral staff, that is, Eng. Rodrigo M. de Carvalho in the development of the
computer codes, Eng. Yemcy Flores in the Malvern analyzes and Evandro Ramos in
the angle of repose measurements is deeply appreciated.

REFERENCES

1 AUSTIN, L.G., KLIMPEL, R.R., LUCKIE, PT. Process Engineering of Size Reduction,
AIME, 1984.
2 RAMOS, E., CARVALHO, R.M., TAVARES, L.M. Simulação do movimento da carga em
moinhos de bolas utilizando o método dos elementos discretos. In: XXIV Encontro
Nacional de Tratamento de Minérios e Metalurgia Extrativa, Salvador, v. 1, p. 440-447,
2011.
3 KING, R.P. Modeling and Simulation of Mineral Processing Systems, Butterworth-
Heinemann, 2001.

740
 ISSN 2176-3135

4 AUSTIN, L.G., TANGSATHITKULCHAI, C. Comparison of methods for sizing ball mills

using open-circuit wet grinding of phosphate ore as a test sample, Ind. Eng. Chem. Res.,
v. 26, p. 997-1003, 1987.
5 OLIVEIRA, A.L, CARVALHO, R.M., TAVARES, L.M., Distribuição de tempos de
residência de moinho de bolas operando a seco em escala piloto, XXIV Encontro
Nacional de Tratamento de Minérios e Metalurgia Extrativa, Salvador, v. 1, p. 776-783,
2011.
6 HERBST, J.A., FUERSTENAU, D.W. Scale-up procedure for continuous grinding mill
design using population balance models, Int. J. Miner. Proces., v. 7, p. 1-31, 1980.

741