Você está na página 1de 7

Fuel 117 (2014) 11561162

Contents lists available at ScienceDirect

Fuel
journal homepage: www.elsevier.com/locate/fuel

Coal breakage characterisation Part 2: Multi-component breakage


modelling
Fengnian Shi
The University of Queensland, Sustainable Minerals Institute, Julius Kruttschnitt Mineral Research Centre, 40 Isles Road, Indooroopilly, Brisbane, Qld 4068, Australia

h i g h l i g h t s
 A multi-component model to describe energysize reduction for various coal particle sizes and densities.
 Model parameters calibrated with JKFBC characterisation testing data.
 Using tn-family of curves to determine product size distribution from the model predicted t10 parameter.
 Model was validated with over 100 sets experimental data using an Australian and a Chinese coal sample.

a r t i c l e

i n f o

Article history:
Received 14 May 2013
Received in revised form 7 July 2013
Accepted 9 July 2013
Available online 1 August 2013
Keywords:
Coal
Breakage characterisation
HGI test
Multi-component breakage model

a b s t r a c t
A multi-component breakage model has been developed at the Julius Kruttschnitt Mineral Research
Centre (JKMRC) to describe the energysize reduction in relation to particle size and density for coal
M
breakage characterisation. The model takes the following form: t 10 RD=RD
 f1  expfmat  x  Eg.
c
min

The model incorporates four parameters and ts 60 JKFBC (JK Fine-particle Breakage Characteriser) test
data for each of the Australian coal and Chinese coal samples, with R2 = 0.982 and 0.978 respectively. The
multi-component model can be switched into a single component model by setting c = 0.
A set of tn-family of curves for coals ground in the JKFBC are presented. It was found that the data from
various particle sizes and densities of the two coal samples, collected from the Australian and Chinese
power stations, all fall on similar tn-curve trend lines. These tn-family of curves can be used in the
multi-component model to estimate the product size distribution from the predicted t10 values. A procedure has been developed to calibrate the multi-component model with seven tests based on a combination of various particle sizes, coal densities and grinding energy levels, using the JKFBC device. Over 100
sets of data have been used to validate the calibration procedure.
2013 Elsevier Ltd. All rights reserved.

1. Introduction
Particle breakage characterisation aims to establish the relationship between a specic energy input and the resultant product,
through some type of laboratory test on a given sample [1]. With
this energysize reduction relationship characterised for a particular coal sample, users can investigate how particles break when
subjected to a given input energy; or determine what energy is required in order to achieve a desired breakage degree. Obviously,
the traditional HGI test on a single size coal particle with a single
level of grinding energy cannot provide this level of detail.
An improved coal breakage characterisation method has been
developed at the Julius Kruttschnitt Mineral Research Centre
(JKMRC), which is presented in Part 1 of this paper [2]. An experiment was conducted on various narrowly-sized and density
Tel.: +61 7 3365 5913; fax: +61 7 3365 5999.
E-mail address: f.shi@uq.edu.au
0016-2361/$ - see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.fuel.2013.07.026

franctionated groups of coal samples, using a JKFBC (JK Fine-particle Breakage Characteriser) consisting of a standard HGI grinding
device and a precision torque meter. Each sizedensity group sample was ground in the JKFBC with various input energies. The data
show a strong size effect and density effect on coal breakage, with
the larger and lower density particles being broken more easily under the same specic energy.
Part 2 of the paper presents a mathematical model which describes the multi-component breakage behaviour observed in the
JKFBC tests. Part 3 will demonstrate the applications of the model
for HGI predictions and coal breakage simulations.
2. Particle breakage modelling
2.1. The JKMRC prior-art model
The JKMRC has been using a breakage model (Eq. (1)) to
describe the energysize reduction relationship for a long time:

F. Shi / Fuel 117 (2014) 11561162

t 10 A1  eb:Ecs

where t10 is a size distribution neness index (%), Ecs is the specic
comminution energy (kWh/t), and A and b are the ore impact breakage parameters. The product form of the model parameters Ab is
used as an indicator of ore resistance to breakage, with a large
Ab indicating less resistance to breakage [1]. The Ab indicator for
ore breakage is currently used by researchers and mining engineers
worldwide.
Previous work has found that the crack density of larger particles is much greater than that of smaller particles [3,4]. In view
of this, larger particles tend to be weaker and therefore easier to
break than smaller particles. In the standard data reduction procedures, however, Eq. (1) is used to t the Drop Weight Tester (DWT)
data with one set of A and b parameters for all particle sizes. This
average set of A and b parameters assume that particles of different sizes would be broken in the same way when subjected to the
same impact energy. However, this assumption is questionable.
A new breakage model incorporating particle size effect has
been reported [5] to describe the degree of breakage. It was modied from Vogel and Peukerts work [6] describing the probability
of breakage. The breakage degree model takes the following form:

t 10 Mf1  expfmat:  x  kE  Emin g

where M (%) represents the maximum t10 for a material subject to


breakage, fmat (kg J1 m1) is the material breakage property, x
(m) the initial particle size, k is the successive number of impacts
with a single impact energy, E (J kg1) the mass-specic impact energy, and Emin (J kg1) is the threshold energy. Analysis of the tted
fmat from many DWT and JKRBT test data enables the development
of a sub-model of fmat, which allows determination of the particle
size effect on the impact breakage result. Fig. 1 demonstrates the
model tting quality by Eqs. (1) and (2) respectively, using the same
set of DWT data. It is apparent that the size-dependent breakage
model (Eq. (2)) can describe the data better.

2.2. A multi-component model for coal breakage


2.2.1. Analysis of the JKFBC data
Eq. (2) was developed to treat impact breakage data, and has
been tested with hundreds of sets of DWT and JKRBT data to prove
its validity. A question was raised regarding whether or not Eq. (2)
can be applied for coal breakage characterisation, knowing the following differences exist between the JKFBC and the DWT/JKRBT
test procedures:

1157

(1) The breakage mode utilised in the JKFBC is not by impact as


in the DWT/JKRBT, but rather by compression grinding.
(2) The cumulative energy controlled by the grinding time in
the JKFBC tests is much larger than the energy threshold
(Emin in Eq. (2)) of coal particles; while in the JKRBT incremental low energy breakage, each impact energy is in similar order to the energy threshold of rock particles.
(3) Eq. (2) was developed as a size-dependent model for DWT/
JKRBT ore data reduction. The JKFBC tests were conducted
on a multi-component basis, and the data suggest that both
particle size and density affect breakage results [2].
Regarding the rst difference, the breakage modes will certainly
affect the degree of breakage (t10), resulting in different breakage
parameters when the model is tted to the experimental data.
The same model structure can be applied, allowing the model
parameters to take care of the difference in breakage mode (either
impact or grinding).
In the case of the second difference, the energy threshold Emin in
Eq. (2) can be ignored for coal. For a 60-revolution grinding test,
similar to the standard HGI test, the total specic energy is approximately 7000 J/kg. The energy threshold, or the minimum energy
required to cause breakage for a coal particle, is in the order of
10 J/kg that was estimated from single-particle dropping on an anvil from a known height, using large amount of coal particles for
the test. This energy threshold is minor compared with the
7000 J/kg input energy. Therefore the term Emin in Eq. (2) can be
dropped without any signicant inuence on the model tting
results.
The major challenge in developing the coal breakage model is to
describe the effects of both particle size and density. The simplest
way is to treat the multi-component data with a single component
model, i.e. to t one set of breakage parameters for each density
group, using the size-dependent model as presented in Eq. (2). In
an application for a particular density, the set of parameters for
the given density can be used. However, as the individual sets of
parameters are independent of the density effect, they may predict
an inconsistent trend in the density effect on breakage. In addition,
there is a difculty in selecting the right set of parameters for a
density that was not measured.
Analysis of the data presented in Fig. 6 in Part 1 of this paper [2]
found that the JKFBC data show a regular trend in the size effect,
i.e. the large particle size is consistently on top of the small particle
sizes on the t10Ecs plots. This indicates that the size-dependent
model (Eq. (2)) will be able to handle the data; working from the
expectation that all points of various sizes will fall on one trend
line for each density group.

Fig. 1. Comparison of model tting results to the same DWT data of a quarry material (after Shi and Kojovic [5]).

1158

F. Shi / Fuel 117 (2014) 11561162

the size effect was modelled separately in the exponential term.


However, as the same test data were used to t the density-related
and size-related parameters simultaneously, these parameters
would represent the joint effects of coal density and particle size.
2.2.2. A multi-component breakage model
A multi-component breakage model was developed to describe
the JKFBC data. The model takes the following form:

t10

Fig. 2. Coal density effect on the energysize reduction relationship for the 2.8
4.0 mm Chinese coal ground in the JKFBC.

The analysis also pointed out that particle density affects the
positions of the trend lines. Fig. 2 depicts the density effect on
the energysize reduction relationship for the 2.84.0 mm Chinese
coal ground in the JKFBC. Obviously, as density increases (indicating more mineral matter associated with the coal particles), the
breakage degree (t10) decreases.
The physical meaning of parameter M in Eq. (2) is the maximum
t10 that a sample can reach as the other variables (specic energy E
or particle size x) change. Mathematically, changing M in Eq. (2)
will change the position of the exponential line. Based on this analysis, therefore, the density effect was modelled in the M-term, and

M
 f1  expfmat  x  Eg
RD=RDmin c

where t10, M and x have been dened in Eq. (2), RD is the relative
density of the particle, RDmin is the minimum relative density of
the sample (RDmin = 1.25 for coal), c is a parameter determining
the trend line position, fmat is a model parameter that is described
by a size-dependent equation with parameters p and q, E (J kg1)
is the mass-specic energy in the JKFBC that is determined by Eq.
(4):

2pNT
m

where N is the mill revolution number, T (Nm) is the mean net torque, and m (kg) is the coal mass being ground.
The model incorporates four parameters, M, c, p and q, which
can be calibrated by tting the model to the measured data. Figs. 3
and 4 show the tting results for the Australian coal and the Chinese coal respectively. There are nominal 64 data points (4

Fig. 3. The multi-component breakage model tted to the JKFBC data for a nominal 4 sizes  4 densities  4 energies (61 data points sharing 4 parameters) for the Australian
coal.

Fig. 4. The multi-component breakage model tted to JKFBC data for a nominal 4 sizes  4 densities  4 energies (59 data points sharing 4 parameters) for the Chinese coal.

1159

F. Shi / Fuel 117 (2014) 11561162

2.3. Modelling of breakage distribution


The breakage model (Eq. (3)) predicts a single variable t10 which
represents the cumulative per cent of the product passing 1/10th of

100
t2
80

t n (% Passing)

sizes  4 densities  4 energies), minus a couple of data points


which are missing due to insufcient samples for those tests. To
avoid over-crowded plots in one graph, only the two extreme densities are presented on the t10E graph (left), with the measured
data in dots and the model tting in lines. The overall tting quality can be seen from the predicted versus the measured plots
(right).
The model tting results suggest that the multi-component
breakage model (Eq. (3)) can well describe the particle size effect
and the density effect on the energysize reduction relationship.
With four model parameters tted, the model is able to represent
the coal breakage behaviour of the 60 tests from each of the two
coal samples, covering a wide range of coal sizes and densities typically found in coal-red power station grinding facilities. The
model has been validated with an Australian coal (HGI = 53) and
a Chinese coal (HGI = 80). The HGI range of the two coal samples
is typical in the marketplace of thermal coals used in power stations worldwide.
A question may be raised regarding the exibility of the model
structure. Fig. 7 in Part 1 of the paper [2] presents the density effect
on breakage, in which the coarse particles show a very pronounced
density effect while the ne particles do not (with the four sets of
density data falling on one similar line for the 0.61.18 mm size
particles). Apparently there is a strong interaction between the
density effect and the particle size effect on coal breakage, and
these effects are not presented in a xed pattern. Can the model
present such variations correctly? Fig. 5 shows the model tting
results (in line) and the measured data (in dots) for the four density
groups in the 0.61.18 mm and 2.363.35 mm size fractions
respectively. With one set of four parameters shared by the 61 Australian coal tests, the model demonstrates its robustness to replicate the complex patterns of coal density and size effects on
breakage.
It is worth emphasising that although the model structure is designed for multi-component data, the model can be used easily for
single component data of a bulk sample (without oat-sink tests to
fractionate the size-by-size feed into size-by-density bases). Simply setting the parameter c = 0 will make the denominator in Eq.
(3) equal to one, thereby converting the multi-component model
to a single component model with three parameters M, p and q tted to the size-by-energy test data. This feature will give users exibility to apply the model for coal breakage characterisation
according to their available data.

t4

60
t10
40
t25
20

t50
t75

0
0

10

20

30

Breakage Index, t

10

40

50

(%)

Fig. 6. Determination of size distribution parameter tn from the breakage index t10
(Redraw after Narayanan [10]).

the feed size. This t10 parameter indicates the neness of a product
when subjected to breakage with a certain input energy. Sometimes it is not sufcient knowing the neness of a product. It
may be necessary to predict the product size distribution. For
example, for optimum combustion efciency, the pulverised fuel
(PF) has a certain size distribution requirement, such as 75% passing 75 lm and 99.5% passing 300 lm.
Narayanan and Whiten [7] found that the t10 parameter is uniquely related to other points on a family of size distribution
curves, with tn, dened as the cumulative percentage passing a given fraction of the initial size, x/n. The t10 can then be used to generate a size distribution from relationships between t10 and tnfamily of curves established from the drop weight test database.
The tn-family of curves have been independently conrmed by
Pauw and Mar [8] and by King and Bourgeois [9] using different
materials over a wide range of fracture energies. Fig. 6 depicts
the tn-family of curves for a range of ore types [9]. Spline regression
analysis can be carried out to describe each of the relationships
t10t2, t10t4, t10t25, t10t50 and t10t75.
Using the tn-family of curves to produce a size distribution is
straightforward. For a given t10 (e.g. t10 = 20), Fig. 6 indicates that
t2 = 75, t4 = 42, t25 = 10, t50 = 8, t75 = 6. For a feed geometric mean
size of 50 mm, t2 = 75 means 75% passing 25 mm (=50/2 mm),
t4 = 42 means 42% passing 12.5 mm (=50/4 mm), and so on. The
size distributions at the required size fractions can be determined
from the tn knots using spline regression. Thus the whole size

Fig. 5. Modelling the coal density effect for various feed sizes, with one set of four parameters shared by 61 Australian coal tests.

1160

F. Shi / Fuel 117 (2014) 11561162

Fig. 7. The plots of tn-family of curves using the JKFBC data from grinding the
Australian and the Chinese coal samples.

Fig. 9. The tn-family of curves used for coal grinding in the JKFBC.

Table 1
R2 values for various regression lines between the measured and the predicted values
using the generalised tn-family of curves presented in Fig. 9.

Fig. 8. The plot of t25 versus t10 for various densities of the Australian and Chinese
coals, consisting of 120 measurement points.

distribution of the progeny can be determined once the breakage


index t10 is known.
Considering that the tn-family of curves presented in Fig. 6 were
derived from impact tests on rock particles, and the JKFBC data
were generated from grinding coals, an investigation was carried
out using the JKFBC data to generate the tn-family of curves.
Fig. 7 shows the plots of tn versus t10 using the JKFBC data from
grinding the Australian coal and the Chinese coal samples together.
Each tn-curve consists of 120 data points.
The data illustrate that the Australian coal and the Chinese coal
fall on similar trend lines, despite the signicant differences in
their HGI values, ash content, and ash compositions (refer to Part
1, Section 2.3.1). Note that each tn-curve comprises different density and different feed size data, appearing to all fall on similar
trend lines. Fig. 8 is an enlarged plot of Fig. 7, with only one t25
curve formed on a density by density basis. The graph supports
the notion that both the coal density effect and the particle size effect on breakage have been eliminated in the plot of tn versus t10.
Although the plots look promising, only two coals have been used.
Further data is desirable, particularly for a low HGI coal, to conrm
the approach of tn-family of curves. More coal test data across the
commercial HGI range are required.
Regression was performed on the data displayed in Fig. 7 to produce the matrices for plotting the tn-family of curves for coals
ground in the JKFBC, which are presented in Fig. 9.
The multi-component breakage model and the tn-family of
curves provide a useful tool for coal breakage characterisation.
Once the model is calibrated with the JKFBC test data, the model
will be able to predict the t10 for a given particle size and density.
The corresponding tn values to this predicted t10 can be found from

t2

t4

t25

t50

t75

0.94

0.97

0.95

0.90

0.88

the tn-family of curves (Fig. 9), regardless of particle size and density; thus the whole product size distribution can be determined
from a single breakage index t10.
Using this approach to determine the product size distribution
is helpful when there is no sizing data available, as in the case of
numerical simulations. However, it is noticed from Fig. 7 that the
data on each trend line are rather scattered, similar to the tn-curves
plotted from the rock impact tests [10]. This implies that errors are
associated with the generalised tn-family of curves. Table 1 gives R2
values for each tn curve, when plotting the measured t-values
against those calculated from the tn-family matrices. R2 is the statistical term of coefcient of determination, which provides a measure of how well observed outcomes are replicated by the
regression model. R2 varies from zero to 1. R2 = 1 indicates that
the two sets of data are exactly matched, with a perfect diagonal
line on the plot of the predicted versus the measured t-values, on
the same x-axis and y-axis scales.
For an understanding of the R2 values, readers are referred to
Fig. 8 (the t25 plot). In this plot R2 = 0.95. The t2 (R2 = 0.94) has similar results to the t25, the t4 (R2 = 0.97) is slightly better than the t25,
but both the t50 and the t75 are not as good as the t25. This implies
that for a feed size of 10 mm, predictions of the cumulative percentage passing 5 mm (t2), 2.5 mm (t4) and 0.4 mm (t25) would
be accurate if the predicted t10 is accurate, however larger errors
may be associated with the cumulative percentage passing
0.2 mm (t50) and 0.133 mm (t75). To minimise the errors associated
with product size distributions, using the sizing data to establish a
set of coal-specic tn-family of curves for the tested coal sample is
recommended, rather than using the generalised tn-family of
curves, particularly if experimental data is available, as was the
case in the JKFBC tests.

3. Model calibration and validation


To study coal breakage behaviour, the experiment, as
reported in Part 1 of the paper, was designed for 4 particle sizes  4

F. Shi / Fuel 117 (2014) 11561162

1161

Fig. 10. The multi-component model predictions for the Australian coal, using seven tests to calibrate the model parameters.

Fig. 11. The multi-component model predictions for the Chinese coal, using the seven Australian coal testing conditions to calibrate the model parameters.

densities  4 specic energies; generating a nominal number of 64


tests for each coal sample being conducted. Based on the size-bydensity experimental results, the effects of particle size and density
on the energysize reduction relationship have been established,
and a multi-component breakage model has been developed. The
model incorporates four parameters to dene these effects. To generate a set of unique parameters, the minimum number of tests required is four, but if more test data were available the calibration
would likely generate more robust parameters.
However, using a complete set of 64 tests to calibrate the four
model parameters is more than adequate. It is hypothesised that
the multi-component model can be calibrated with a reduced
number of tests on a given coal, and the calibrated model will still
be able to predict the coal breakage behaviour on a sizedensity
energy basis. Many different combinations of the particle size, density and grinding energy were investigated. Eventually, seven tests
were selected by trial-and-error, covering the whole range of particle size, density and grinding energy. Fig. 10a shows the model
tting results using the data of the seven tests on the Australian
coal. The four model parameters t the seven data points well.
For the Australian coal sample, there are 61 sets of available
data. Seven tests were used for calibration, and the remaining 54
tests were used to validate the model. The model was employed
to predict the t10 values from the tested size, density and energy,

using the calibrated four model parameters. Fig. 10b presents the
predicted t10 versus the measured t10. Comparing Fig. 10b with
Fig. 3, where all 61 sets of data were used to t the four model
parameters, the new characterisation procedure using seven tests
to calibrate the model parameters predicts almost the identical results as those using the 61 sets of data, with R2 of 0.976 for the 7
tests and 0.982 for the 61 tests.
The same seven test conditions were applied to the Chinese coal
sample, with minor variations in the feed particle sizes tested. The
similar results are presented in Fig. 11. Note that the prediction results for the Chinese coal, using the same seven testing conditions
as for the Australian coal, appear more scattered (Figs. 11b compared with 10b). This indicates that more work can be done to
optimise the set of 3D test conditions (sizedensityenergy) used
for the model calibration.
A statistical t-test was performed on paired comparisons between the measured and the predicted t10, using the model parameters calibrated by the seven tests. Table 2 summarises the
statistical test results. The null hypothesis is that each pair of results is equal, or that each difference is zero. The null hypothesis
is assumed true unless proved otherwise. The result given in
Table 2 indicates that the differences in t10 values determined by
the two methods for both the Australian and Chinese coal
samples do not reach the 95% signicant level threshold. The null

1162

F. Shi / Fuel 117 (2014) 11561162

Table 2
Summary of t-test results on a paired comparison between the measured and the
predicted t10 from the model parameters calibrated by the seven tests.
Coal source

Australia

China

Mean difference
Standard deviation
Degree of freedom
t
Distribution
Signicance level (%)

0.35
2.14
60
1.26
2-Tailed
79

0.05
3.69
58
0.096
2-Tailed
8

hypothesis is therefore accepted. This conrms that the multicomponent coal breakage model calibrated with seven reduced
tests can represent the coal breakage behaviour on a size-bydensity-byenergy basis.
Future work remains to develop standardised testing conditions
for calibration when more data are available. Nevertheless this pioneering work demonstrates that using reduced coal breakage tests
with the JKFBC to calibrate the model parameters is a viable approach, and the multi-component model will predict the full
breakage characteristic response of the coal.
4. Conclusions
Based on the multi-component JKFBC test results and the analysis of the effects of particle size and density on the energysize
reduction relationship, a multi-component breakage model for coal
has been developed. The model ts the approximately 60 sets of
JKFBC tests on each of the Australian coal and Chinese coal samples
well, with only four parameters needing to be calibrated. The model demonstrates its robustness to replicate the complex patterns of
coal density and size effects on breakage. The multi-component
model can be switched into a single component model.
A set of tn-family of curves have been established for coals
ground in the JKFBC. It was found that the data of various particle
sizes and densities, from the two coal samples collected at the Australian and Chinese power stations, all fall on similar tn-curve trend
lines. These tn-family of curves can be employed to estimate the
product size distribution from the predicted t10 values by the multi-component model.

The multi-component model can be calibrated with seven


JKFBC tests on a combination of various feed sizes, densities and
specic energies. Predictions for the other testing conditions using
the calibrated model were compared with the measured data to
validate the model. Statistical analysis conrms that the model
predictions are similar to the measured data for both the Australian and Chinese coal samples.
Acknowledgements
The coal breakage modelling work was funded by the Australian
Government Department of Resources, Energy and Tourism as part
of the Asia-Pacic Partnership on Clean Development and Climate
program (APP). The Chinese coal sample was provided by the APP
Project collaborative research team from the China University of
Mining and Technology led by Prof. Yaqun He. The contributions
made by Dr Stephen Larbi-Bram and Mr Weiran Zuo in the experimental work to provide the JKFBC data for the model development
were gratefully acknowledged. The support from Prof. Emmy Manlapig of JKMRC in the APP project was much appreciated.
References
[1] Napier-Munn TJ, Morrell S, Morrison RD, Kojovic T. Mineral comminution
circuits: their operation and optimisation. Julius Kruttschnitt Mineral Research
Centre; 1996. ISBN 0 646 28861x [chapter 4].
[2] Shi F, Zuo W. Coal breakage characterisation Part 1: Breakage testing with
the JKFBC. Fuel 2014;117:114855.
[3] Krajcinovic D. Damage mechanics. Oxford (UK): Elsevier; 1996. p. 15966.
[4] Tavares LM, King RP. Single-particle fracture under impact loading. Int J Miner
Process 1998;54:128.
[5] Shi F, Kojovic T. Validation of a model for impact breakage incorporating
particle size effect. Int J Miner Process 2007;82:15663.
[6] Vogel L, Peukert W. Determination of material properties relevant to grinding
by practicable labscale milling tests. Int J Miner Process 2004;74S:32938.
[7] Narayanan SS, Whiten WJ. Determination of comminution characteristics from
single particle breakage tests and its application to ball mill scale-up. Trans
Inst Miner Metall 1988;97:C11524.
[8] Pauw OG, Mar MS. The determination of optimum impact breakage routes for
an ore. Powder Technol 1988;54:313.
[9] King RP, Bourgeois F. A new conceptual model for ball milling. In: Proceeding
XVIII international mineral processing congress, Sydney, 1993. p. 816.
[10] Narayanan SS. Development of a laboratory single particle breakage technique
and its application to ball mill scale-up. PhD thesis, University of Queensland
(JKMRC), Australia, 1985.

Você também pode gostar