Professional Documents
Culture Documents
Demo Day Region Es
Demo Day Region Es
PII: S0032-5910(18)30743-5
DOI: doi:10.1016/j.powtec.2018.09.016
Reference: PTEC 13690
To appear in: Powder Technology
Received date: 5 February 2018
Revised date: 30 August 2018
Accepted date: 6 September 2018
Please cite this article as: Alessandro L.R. de Oliveira, Luís Marcelo Tavares , Modeling
and simulation of continuous open circuit dry grinding in a pilot-scale ball mill using
Austin's and Nomura's models. Ptec (2018), doi:10.1016/j.powtec.2018.09.016
This is a PDF file of an unedited manuscript that has been accepted for publication. As
a service to our customers we are providing this early version of the manuscript. The
manuscript will undergo copyediting, typesetting, and review of the resulting proof before
it is published in its final form. Please note that during the production process errors may
be discovered which could affect the content, and all legal disclaimers that apply to the
journal pertain.
ACCEPTED MANUSCRIPT
1
Department of Metallurgical and Materials Engineering, Universidade Federal do Rio de
T
Janeiro – COPPE/UFRJ, Cx. Postal 68505, CEP 21941-972, Rio de Janeiro, RJ (Brazil)
IP
2
Centro Federal de Educação Tecnológica Celso Suckow da Fonseca, CEFET/RJ, UnED,
CR
Angra dos Reis, RJ (Brazil)
US
ABSTRACT AN
Prediction of continuous steady-state ball milling using the traditional population balance
model requires knowledge of several functions, namely those describing the rate and
M
distribution of breakage products, the mass transfer relationship between the mill hold-up
and the discharge, the mode of transport within the mill, as well as the description of
ED
internal classification, whenever present. In spite of its long track record, there are
comparatively few instances in which the population balance model of continuous ball
PT
milling has been validated in great detail under controlled conditions. The work analyzes
how confidently one can predict continuous milling in a dry open-circuit pilot-scale
CE
operating under a wide range of feed rates, mill speeds, fillings and feed materials (an iron
ore and a spent catalyst) using the population balance model. It relies on measured
AC
residence time distribution of the solids, the Austin model of grinding and the expressions
recently proposed by Nomura to describe mass transfer from the mill to the discharge to
predict continuous grinding. Breakage parameters of the materials studied were estimated
on the basis of batch grinding tests. Parameters in Nomura’s transport model were fitted on
the basis of a few selected continuous tests and were used to estimate the mill hold up
operating under a range of conditions. A comparison between experiments and predictions
demonstrated that errors in the measurements of hold-up were, on the average, 6.0%,
whereas those associated to the 80% passing size of the product were smaller than 5.6%.
1
ACCEPTED MANUSCRIPT
1. Introduction
T
Prediction of the product size distribution in continuous ball milling using the
IP
population balance model (PBM) requires the description of several relevant
microprocesses, namely rate and distribution of breakage products, mass transfer
CR
relationship and internal classification, whenever present. With different simplifications,
the traditional population balance model has been successfully used to describe ball milling
US
in industry [1-3], being also the basis of well-known model-based scale-up procedures.
Modeling approaches that rely on measurements of material breakage
AN
characteristics in batch grinding mills as the basis for predicting continuous milling [2, 3]
became particularly popular. They rely on the assumption that material breakage
M
characteristics, namely breakage rate and distribution functions, may be estimated from
batch grinding tests under controlled conditions, to then simulate the mill operating
ED
the PBM in predicting the capacity of the famous Bougainville Copper Mine mill, which
was not reaching the capacity designed on the basis of the Bond method.
CE
In spite of its importance, there are only a few studies that analyzed in great detail
the confidence in the predictions using the PBM. Mulenga [11] recently carried out a
AC
sensitivity analysis of parameters of the breakage function and the selection function on
predictions on continuous milling. The researcher observed, for instance, that about 20%
deviations in parameters of the breakage rate function resulted in 32% deviation in the mill
product size, thus demonstrating an important error-propagating effect. In the case of
breakage distribution function parameters, deviations of 20% in its parameters resulted in
deviations of up to 23% in the product size. Mulenga [12] also demonstrated that the model
of transport and mixing in the mill had a significant impact on predictions in continuous
grinding. However, studies that analyze in great detail the deviations on the prediction of
2
ACCEPTED MANUSCRIPT
the continuous population balance model in ball mills operating under a wide range of
conditions are not easily found in the literature.
One key component of the continuous mill model is the residence time, that is, the
relationship between mill hold-up and mill feed/discharge rate. The residence time of solids
in the mill influences the mill performance both directly and indirectly; the latter associated
to the influence of mill voids filling on the rates of breakage [3, 7, 13]. Indeed, the mass
transfer in the mill is closely related to the concept of residence time distribution (RTD),
T
which has been the subject of extensive research for more than four decades [14-19]. Such
IP
studies have been useful in proposing descriptions of the number of mixers which can be
CR
used to describe the mass transfer from the mill interior to the discharge. Further, Swaroop
et al. [15] demonstrated a linear relationship between hold-up and feed rate for dry
US
grinding, whereas Austin et al. [3] proposed an empirical relationship in which the rate of
solids discharge follows a power-law relationship of the hold-up of solids, whose
AN
parameters must be determined from experiments. More recently, Nomura [20, 21]
proposed a very detailed model that can be used to determine the relationship between the
M
mass hold-up and the discharge rate for mills operating under a variety of conditions, with
very limited empiricism. This model, however, has not yet been validated in conjunction to
ED
scale-up mill models, such as the one proposed by Austin et al. [3].
The present work deals with the validation of the Austin scale-up model and the
PT
2. Modeling background
AC
3
ACCEPTED MANUSCRIPT
assuming first order breakage kinetics, the size-mass balance equation for batch grinding is
given by [3]
𝑖−1
𝑑𝑝𝑖 (𝑡)
= − 𝑠𝑖 𝑝𝑖 (𝑡) + ∑ 𝑏𝑖𝑗 𝑠𝑗 𝑝𝑗 (𝑡) (1)
𝑑𝑡
𝑗=1
𝑖>1
where pi is the fraction of material contained in size class i and t is the grinding time. A
T
function that has been widely used to describe the effect of size on the rate at which
IP
particles disappear from size class i is [3]
CR
𝑆1 (𝑥𝑖 )𝛼
𝑠𝑖 =
𝑥 𝛬 (2)
1 + ( 𝑖)
𝜇
US
in which xi is the representative size of particles in size class i, S1 is a parameter of the
AN
model, which represents the specific breakage rate of 1-mm particles, and α, μ e Λ are
additional parameters that depend on both material and grinding conditions.
M
Several studies [2-5, 11-13] demonstrated that, for a given material, the breakage
rates vary with design and operating conditions in ball mills. Detailed studies have shown
ED
𝑆1 (𝑥𝑖 )𝛼
𝑠𝑖 = 𝐾𝐾 𝐾 𝐾
𝑥𝑖 𝛬 2 3 4 5 (3)
1 + (𝜇 𝐾 )
1
CE
𝐷 0.2 𝑑𝑏
𝐾1 = ( ) ( )
𝐷𝑇 𝑑𝑏𝑇
𝑑𝑏𝑇
𝐾2 = ( ) (4)
𝑑𝑏
𝐷 0.5
𝐾3 = ( )
𝐷𝑇
4
ACCEPTED MANUSCRIPT
1 + 6.6𝐽𝑇 2,3
𝐾4 = ( ) 𝑒𝑥𝑝(−𝑐 (𝑈 − 𝑈𝑇 ))
1 + 6.6𝐽2,3
𝜑𝑐 − 0.1 1 + 𝑒𝑥𝑝(15.7 (𝜑𝑐𝑇 − 0.94))
𝐾5 = ( )( )
𝜑𝑐𝑇 − 0.1 1 + 𝑒𝑥𝑝(15.7 (𝜑𝑐 − 0.94))
where D is the mill diameter, db is the grinding media size, J the fractional mill filling, U
the void filling fraction and φc the fraction of the critical speed of the mill. The subscript T
T
refers to the parameters used in the standardized batch test, used to fit the model parameters
IP
in Eq. (2). The parameter c contained in the expression for K4 is usually taken to be equal to
CR
1.2 for dry milling and 1.3 for wet milling [18].
The discretized breakage function term bij in Eq. (1), represents the proportion of
US
material contained in size class i that resulted from breakage of particles originally
contained in size class j. It is generally calculated on the basis of the cumulative breakage
AN
distribution function, given by
𝛾 𝛽
𝑥𝑖−1 𝑥𝑖−1
𝐵𝑖𝑗 = 𝜙 ( ) + (1 − 𝜙) ( ) (6)
𝑥𝑗 𝑥𝑗
PT
in which the parameters ϕ , β e γ vary according to the material and are relatively
CE
𝑖−1
𝑀ℎ 𝑀ℎ
𝑝𝑜𝑢𝑡,𝑖 = 𝑝𝑖𝑛,𝑖 − 𝑠𝑖 𝑝𝑜𝑢𝑡,𝑖 + ∑ 𝑏𝑖𝑗 𝑠𝑗 𝑝𝑜𝑢𝑡,𝑗 (7)
𝐹 𝐹
𝑗=1
𝑖>1
5
ACCEPTED MANUSCRIPT
where Mh is the hold-up of solids inside the mill, F is the feed and discharge rate, si is the
rate of breakage of particles contained in size class i and bij is the breakage distribution
function. pin,i is the fraction of particles contained in size i that enter this well-mixed region
in the mil and pout,i is the size distribution of the discharge of this region. The ratio between
Mh and F is the nominal residence time of solids in this perfectly-mixed volume.
Practical application of Eq. (7) first requires knowledge of the breakage rate sj and
distribution functions bij that describe the breakage response of material in the mill.
T
According to Austin´s method, these may be estimated from batch grinding experiments on
IP
a test mill under controlled conditions, fitting parameters of Eqs. (1), (2) and (6) to data,
CR
and then applying Eqs. (3) and (4) to account for deviations from the fitted (test) case.
Further, application of Eq. (7) also requires knowledge of the number of well-mixed
US
regions which may be used to describe the mill active volume, the relative sizes of these
well-mixed volumes, besides the mass of hold-up of solids in the mill and how it relates to
AN
the feed or discharge rates. Eq. (7) directly applies to the case of a mill that is described as a
single perfect mixer with no internal classification.
M
One key component in Eq. (7) is the estimate of the hold-up of solids inside the
mill, for a given set of design and operating conditions of the mill or, alternatively, the
PT
mean residence time of solids in the mill. In the absence of any such estimate, one
alternative has been to assume that the fractional voids filling is equal to one [16]. A large
CE
body of experimental work on continuous mills, however, demonstrates that this is seldom
rigorously valid in practice [3, 14, 15, 17].
AC
Fortunately, a detailed model has been recently developed by Nomura [21], which
establishes a relationship between the hold-up and the feed rate in tumbling mills based on
the movement of the mill load and a model of the motion of balls and particles. The model,
when applied to the particular case of dry and grate-discharge mills, is briefly reviewed as
follows.
The hold-up mass (Mh) of solids in the mill may be expressed as
6
ACCEPTED MANUSCRIPT
where ρ is the apparent density of the material, L is the length of the mill, A is the cross-
sectional area of the mill, J is the fractional mill filling, εb is the nominal fractional porosity
of the static grinding load (taken as 0.4), U is the fraction of the grinding load occupied by
the material in a static mill and Ah is the cross-sectional area of the hold-up.
The expression used to describe the discharge of the material will depend on the
type of discharge used in the mill. Further, the rate of discharge of the material will depend
T
on the speed at which it leaves the milling chamber (υop) and on the fractional open area
IP
available at the mill outlet (φs), in the case of grate-discharge mills. Under steady-state
CR
conditions, any infinitesimal change in hold-up of solids will be translated in change in the
discharge rate, since [21]
𝑑𝐹 = 𝜌υop 𝜑𝑠 𝑑𝐴ℎ
US (9)
AN
Combining Eqs. (8) and (9), the differential equation of the hold-up will be
M
𝐿
𝑑𝑀ℎ = 𝜌𝐿𝑑𝐴ℎ = 𝑑𝐹 (10)
ED
υop 𝜑𝑠
PT
Eq. (10) establishes the linear relation between hold-up and feed rate, under steady-
state conditions, when the material output velocity and percentage of open area in the
CE
output are constant. Integrating Eq. (10) with the initial conditions Mh = Mh0, the mass of
material retained in the mill when there is no discharge (F = 0), and considering that φs and
υop are constant, gives
AC
𝐿
𝑀ℎ = ( ) 𝐹 + 𝑀ℎ0 = 𝑐1 𝐿𝐹 + 𝑀ℎ0 (11)
υop 𝜑𝑠
where c1 is the constant that expresses the discharge properties of the material at the outlet
of the mill. For Nomura [21], the mill is capable of retaining material in two ways: through
their adhesion to the grinding media, and through a volume that will remain in the mill even
7
ACCEPTED MANUSCRIPT
after the mill is no longer discharging – called dead volume. This mass of material adhered
to the grinding media is called Mhb, whereas the mass of material in the mill's dead space is
called Mhd, so that Mh0 will be given by
According to Nomura [21] the mass of material adhered in the grinding media (Mhb)
T
will depend on the coefficient of friction of the material f and also on the total surface area
IP
of the grinding media. The total load of the mill, that is, the grinding media and the
CR
material, is divided into three regions in any cross-section of the mill (Figure 1): grinding
zone, ascending zone and falling zone. When the mill load flows through the three zones, a
US
flow of material is created, favoring the transport of material to the exit of the mill.
However, at the same time as the grinding zone and the ascending zone contribute to the
AN
formation of the material flow, they also hinder the flow since they facilitate the creation of
a bed of grinding media, which is responsible for blocking the passage of the solids. Hence,
M
Mhb is proportional to the friction factor f and to the surface area of the grinding media in
the grinding zone and in the ascending zone (Abg) through the expression
ED
where c2 is a constant, f is the coefficient of friction, Vrb is the volumetric flow of balls in
the circumferential direction of the mill, Δtg and Δta are the residence times in the grinding
AC
and ascending zones, respectively, and db is the diameter of the grinding media. Derivations
of expressions for Vrb, Δtg and Δta, as well as an efficient solution algorithm, are presented
in Appendix A, which follows the work from Nomura [20].
In the case of dry grinding the coefficient of friction may be estimated from the
angle of repose (Θ𝑟 ) using the expression [21]
1 + tan Θ𝑟
𝑓= (14)
3
8
ACCEPTED MANUSCRIPT
The mass of material retained in the dead space (Mhd) will depend on the mill
characteristics, in particular, on the type of discharge control [21]. For instance, in grate-
discharge mills, the grate restricts the passage of material, contributing to the formation of
the dead volume. In this case, Mhd is proportional to the geometry of the estimated dead
space, represented by the area of the dead space (Ahd), i.e.,
T
IP
where c3 is a constant and εb0 is the porosity of the grinding load in the grinding and
CR
ascending zones. For grate-discharge mills, Ahd will be
US
𝑉𝑟𝑏 (∆𝑡𝑔 + ∆𝑡𝑎 )
𝐴ℎ𝑑 = (1 − 𝜑𝑠 ) ⌈ ⌉ (16)
𝐿(1 − 𝜀𝑏0 )
AN
where φs is the fraction of open area in the grates.
M
Nomura's important observation is that the same material can be accounted for in
both Mhb and Mhd. However, the constants c2 and c3 can compensate for the doubled
ED
3. Experimental
9
ACCEPTED MANUSCRIPT
T
177 µm and 8% of material finer than 10 µm The apparent density used in the computations
IP
has been estimated assuming a nominal porosity of 40% of the powder, a simplification
CR
commonly used [3].
The static angle of repose of the samples was measured using the hollow cylinder
US
method, following the procedure used by Barrios et al. [22]. Through this, a lot of 2 kg of
material initially filled a PVC tube, which was then lifted, leaving a pile of material from
AN
which the angle was measured. Such measurement was conducted in triplicate and used to
account for the widely different flowabilities of the samples, evident from Table 1. From
M
the measured angle of repose, the coefficient of friction of the material was estimated
(Appendix A).
ED
The dry milling system used in the present work was composed of a vibratory screw
feeder coupled to a continuous grate-discharge ball mill, which discharges directly on a
CE
scale (Figure 2). The feeder and its storage silo are placed on load cells, with a precision of
20 g. This feeding system is controlled by a computer in order to guarantee an accurate feed
AC
rate. The entire system is monitored and controlled by a supervisory system, developed in
LabVIEW®, that allows monitoring the most important variables of the process.
The continuous ball mill has a single grinding chamber (Figure 3), with 32 cm in
diameter (D) and 31 cm in length (L). The mill-motor assembly is positioned on three load
cells, which allow real-time measurement of the hold-up mass during the experiments,
whose maximum weight is 400 kg with an accuracy of 30 grams.
Discharge of material from the mill occurred through a grate, shown in detail in
Figure 3. It has 7 mm diameter apertures, which represent 8% of the total grate area.
10
ACCEPTED MANUSCRIPT
T
prevent exit of any particle from the grinding chamber during the tests.
IP
The batch tests consisted in milling of the samples at different times, followed by
CR
size analyzes of the ground material for each time by laser scattering in a Malvern
Mastersizer® 2000. The times selected for both spent catalyst and iron ore samples were 7,
US
15 and 30 minutes, and the experiments were run under the operating conditions listed in
Table 2. Different values of voids filling fraction were selected for the materials given
AN
observations of typical values found in the continuous tests, analyzed later in the present
work.
M
In order to limit the influence of grinding media size on the grinding experiments,
experiments were all carried out using 16 mm chrome steel balls.
ED
Continuous tests were performed with the mill under different operating conditions,
varying apparent volume feed rates (68 to 185 cm3/min), fractional mill filling (0.20 and
CE
0.30) and fraction of critical speed (0.30 and 0.75). The experimental procedure adopted in
all tests consisted in feeding the mill with a set feed rate of solids, operating the mill and
AC
monitoring the discharge rate of solids through continuously measuring the rate with the
scale. When the discharge rate matched the feed rate within less 5% of this value during a
period of time of at least 20 minutes, it was identified that steady-state conditions were
reached (Figure 4). During this period, samples of the product were collected and identified
so that their particle size distribution could be later determined using the laser scattering
particle size analyzer (Malvern Mastersizer® 2000). After this, the grinding circuit was
switched off and the mill contents were emptied, allowing a direct measurement of the mill
hold-up of solids. A total of 22 experiments were conducted. Unfortunately, in three of
11
ACCEPTED MANUSCRIPT
these experiments problems found in weighting prevented estimating the mill hold-up, with
the data only used in validation of Austin´s model.
T
demonstrate the very good fit of the model to data. Table 3 summarizes the least-squares
IP
best-fit parameters of the breakage rate and breakage distribution functions. In both cases it
is evident that such good fit could be reached with even simpler functional forms of the
CR
breakage rate and breakage distribution functions than those presented in Eqs. (2) and (6),
respectively. Indeed, parameters in the denominator of Eq. (2) (μ and Λ) were disregarded
US
in the model, considering then the denominator equal to one, so that
𝑠𝑖 = 𝑆1 (𝑥𝑖 )𝛼 (18)
AN
This is explained by the particularly fine feed (Table 1) when compared to the ball
M
size (16 mm) used in the experiments. Further, it was found that the breakage function
could also be described by a simpler form of Eq. (6), with the parameter ϕ equal to one,
ED
𝛾
PT
𝑥𝑖−1
𝐵𝑖𝑗 = ( ) (19)
𝑥𝑗
CE
12
ACCEPTED MANUSCRIPT
The model of Nomura was implemented in Matlab® (Mathworks Inc.) and the least-
squares best fit values of parameters c1, c2 and c3 of Eq. (17) were estimated using the
Nelder-Mead optimization method [23]. From the 22 experiments, 7 were selected for
model fitting, covering different combinations of mill speeds and fillings, so that four and
three of them corresponded to the mill operating at 30 and 75% of critical speed,
respectively. It was found that the same set of values could be used for both materials. This
contrasts with findings from Nomura [21], who observed that such parameters were both a
T
function of material and mill. However, different sets of values were required to fit the data
IP
for the mill operating under the different percentages of critical speed. This requirement can
CR
be explained by the wide range of values of fraction of critical speed used in the tests (0.30
and 0.75), which is wider than typically covered in practice.
US
Figures 7 and 8 compare the experimental data to those from the model, identifying
the seven sets used in fitting the model and the remainder, which corresponded to true
AN
validation of the model. The dashed lines of Figure 8 represent a dispersion of ± 300g,
showing that only four data points presented differences larger than that. The mean relative
M
𝑛
100 𝑀ℎ𝑒,𝑖 − 𝑀ℎ𝑐,𝑖
Mean relative deviation (%) = ∑| | (20)
𝑛 𝑀ℎ𝑒,𝑖
𝑖=1
PT
where Mhe,i and Mhc,i are the measured and calculated values of solids hold-up for the ith
CE
experiment, respectively.
Table 5 summarizes the fitted values of the parameters as a function of fraction of
AC
critical speed. The inverse relationship found by Nomura [21] for the parameters c1 and c2
for different materials ground in the same mill has not been observed here.
13
ACCEPTED MANUSCRIPT
equation describing continuous milling (Eq. 7), continuous grinding in the mill could be
described by
𝑖−1
(1) 𝑀ℎ (1) 𝑀ℎ (1)
𝑝𝑖 = 𝑝𝑖𝑛,𝑖 − 0.07 𝑠𝑖 𝑝𝑖 + 0.07 ∑ 𝑏𝑖𝑗 𝑠𝑗 𝑝𝑗 (21)
𝐹 𝐹
𝑗=1
𝑖>1
𝑖−1
(2) (1) 𝑀ℎ (2) 𝑀ℎ (2)
𝑝𝑖 = 𝑝𝑖 − 0.10 𝑠𝑖 𝑝𝑖 + 0.10 ∑ 𝑏𝑖𝑗 𝑠𝑗 𝑝𝑗
T
𝐹 𝐹 (22)
𝑗=1
IP
𝑖>1
𝑖−1
(2) 𝑀ℎ 𝑀ℎ
CR
𝑝𝑜𝑢𝑡,𝑖 = 𝑝𝑖 − 0.83 𝑠𝑖 𝑝𝑜𝑢𝑡,𝑖 + 0.83 ∑ 𝑏𝑖𝑗 𝑠𝑗 𝑝𝑜𝑢𝑡,𝑗 (23)
𝐹 𝐹
𝑗=1
𝑖>1
Eqs. (21) to (23) for the three perfect mixers in series are solved sequentially using
US
the recursive method presented by King [18]. A total of six fitting parameters (three for the
breakage rate and breakage distribution functions per material and three for the transport
AN
function) were required to predict the size distribution of the mill discharge for a given mill
speed, with additional three parameters required for the other speed investigated.
M
discharge are presented in Figure 9 for the iron ore. They show that the model provides
reasonable predictions, even considering that the simulated value of fraction of mill speed
is below the range used by Austin et al. [3] in fitting the parameter K5 in Eq. (4), which is
PT
0.4 to 0.9.
Figure 10 compares the measured to the predicted 80% passing sizes, which
CE
demonstrate that errors were smaller than 10 µm (with only two exceptions) – a maximum
of 17.8%, with a mean relative error of 5.61%. A summary of the fitting results is presented
AC
in Table 6. It shows that the errors involved in the predictions were significantly larger in
the case of the experiments conducted at 30% of critical speed, which is a highly unusual
condition, which is beyond of the scope of Austin´s scale-up model.
A sensitivity analysis was then conducted of the model of continuous grinding
considering its different assumptions. First, simulations were conducted using the actual
measured value of solids hold-up in Eq. (21) to (23). Although this is not realistic in
practice, it is useful to estimate some of the assumptions in the continuous mill model, that
is, the validity of describing the continuous mill as three mixers-in-series as well as the
14
ACCEPTED MANUSCRIPT
basic assumption of using data from batch grinding experiments to predict continuous
grinding. Table 6 shows that the errors have similar magnitudes using the measured value
of hold-up and using Nomura model to predict it, being even higher for simulations using
the experimental values of solids hold-up. This shows that the accuracy obtained in fitting
parameters of the Nomura model more than sufficed the needs for predicting the mill
product, so that no error-propagating effect from fitting the Nomura model affected
predictions of product size distribution. It thus shows that errors in predicting the mill
T
product were associated to either the breakage parameters, the validity of the mixers-in-
IP
series model and/or the validity of describing continuous milling on the basis of batch data.
CR
Additional simulation scenarios are presented in Table 5. For instance, if instead of
considering the model for the three perfect mixers in series (Eqs. 21 to 23) it is assumed
US
that the mill could be represented by a single perfect mixer (Eq. 7), mean relative
deviations would increase to 6.45%. Table 6 shows that such increase was mainly
AN
associated to the increased deviations of the results obtained with the mill running at 30%
of the critical speed, a condition in which mixing of the charge was less vigorous.
M
Finally, simulations were conducted of the mill assuming that voids filling would be
constant, equal to U = 1, that is, all voids left in grinding media filled with particles. In that
ED
case, mean relative deviations would also increase in comparison to the base case, now due
to higher deviations in the results for 75% of critical speed, resulting in an overall relative
PT
5. Conclusions
A successful implementation of the Austin scale-up procedure coupled to Nomura
AC
transport model, followed with their application to describe continuous dry grinding in an
open-circuit pilot-scale mill, allowed to conclude that:
The Nomura transport model could successfully fit the data, with a mean deviation
between measured and predicted hold-up values of 6.0%.
The same set of fitting parameters were able to describe transport of two widely
different materials (spent catalyst and iron ore) milled in a range of throughputs and
mill fillings using the Nomura model. However, different sets of parameters were
15
ACCEPTED MANUSCRIPT
required to fit the data for the different speeds studied (0.30 and 0.75 of critical
speed).
Batch grinding data for both materials could be successfully fitted to simple
expressions of the breakage rate and breakage distribution functions.
The same mill used for conducted batch grinding tests was used to run continuous
open-circuit tests and predictions with the model considering three mixers in series
were consistent with experimental results, with mean relative deviations of 5.6% in
T
IP
values of 80% passing size in the product.
Additional simulations of the continuous mill allowed to conclude that no error
CR
propagation was identified of the Nomura transport model in predicting the product
size distribution. Also, that simplifying assumptions such as in describing the mill
US
as a perfect mixer or in assuming that all voids left by the grinding charge are filled
with ground material, resulted in larger deviations between experiments and
AN
predictions. The magnitude of increase in these errors, however, was only marginal.
M
Acknowledgements
The authors would like to thank the Brazilian Council of Research CNPq for
ED
financial support to the investigation (grant number 310293/2017-0). The authors would
like to express their appreciation to Dr. Shinishiro Nomura, for his guidance in the
PT
implementation of the transport model. The assistant of Mr. Anderson Chagas in the
implementation of the Austin model was also appreciated.
CE
List of symbols
AC
16
ACCEPTED MANUSCRIPT
T
F coefficient of friction, (-)
IP
fc fractional volume of voids in the grinding charge (-)
CR
G acceleration due to gravity (m/s2)
Hab mean falling distance of balls (m)
US
Hap mean falling distance of particles (m)
I index for size classes (-)
AN
J index for size classes (-)
J mill filling – fraction mill volume occupied by ball bed (-)
M
Mhc hold-up of solids in the mill calculated using the Nomura model (kg)
Mhd mass of material retained in the dead space (kg)
AC
17
ACCEPTED MANUSCRIPT
T
U0 fraction of ball charge voids filled by bulk of particles in grinding
IP
zone (-)
CR
Vm volume of the mill chamber (m3)
Vrb volumetric flow of balls in the circumferential direction of the mill
US
(m3/s)
Vrp volumetric flow of particles in the circumferential direction of the
AN
mill (m3/s)
xi representative size of particles in size class i (mm)
M
Greek letters
ED
18
ACCEPTED MANUSCRIPT
zones (-)
Θ𝑟 angle of repose (°)
θb0 angle for surface level of grinding zone (rad)
θ1 to θ3 mean residence time of the three mixers (min)
Ρ apparent density of the material (kg/m3)
ωr angular frequency of mill revolution (rad/s)
ξi R0/(D/2) (-)
T
ξs Ras/(D/2) (-)
IP
CR
Appendix A. Derivation of expressions for Vrb, Δta e Δtg in the Nomura model
US
In spite of the thoroughness of the presentation of the model by Nomura [21], the
authors found it important to present in greater detail the key part of the solution procedure
AN
required to apply it in practice, which is not straightforward to follow in the original
publication. Further, the equations have been reviewed for minor typos in the original
M
publication.
Prediction of solids hold-up using the model by Nomura [18] is based on volumetric
ED
balances, both for the grinding media and material, in the different regions as presented in
Figure 1. The volumetric balance of grinding media and material of a mill in operation is
PT
given by
𝑉𝑚 𝐽(1 − 𝜀𝑏 ) = (𝑉𝑟𝑏 Δ𝑡𝑔 ) + (𝑉𝑟𝑏 Δ𝑡𝑎 ) + (𝑉𝑟𝑏 Δ𝑡𝑓 ) (A1)
CE
where Vrb and Vrp are the volumetric flowrates of grinding media and particles which
circulate in the mill, fc is the fraction of the mill occupied by particles and Δt are the mean
residence times for each of these regions. In order to find the dynamic values J0, U0 e εb0 it
is necessary to solve simultaneously Eqs. (A1) and (A2) from the static parameters J, U and
εb. Equations for the three zones in Figure 1 are presented as follows.
19
ACCEPTED MANUSCRIPT
T
𝑉𝑟𝑝 = 𝑈0 𝜀𝑏0 𝑤𝑟 ( ) [ ] (A6)
𝜋 2
IP
CR
where R0 is the radius of the surface in the grinding zone and wr is the angular velocity of
rotation in and the mill. The parameter J0 is expressed as a function of the geometry in the
US
grinding zone, such that AN
𝐽0 = [2𝜃𝑏0 − 𝑠𝑖𝑛(2𝜃𝑏0 )]/ 2𝜋 (A7)
M
where θb0 is the angle of the surface level of the grinding zone. It is important to highlight
that cos(θb0) = ξi. The parameters U0 and εb0 are defined according to the amount of
ED
material in the mill. In the case in which the mill is underloaded (U ≤ 1), the voids filling in
the grinding zone would be given by
PT
𝑈0 = 𝛼𝑓 𝑓𝑐 / (𝐽0 𝜀𝑏0 ) ≤ 1
{ (A8)
𝜀𝑏0 = 𝜀𝑏
CE
The fraction of the total of particulate material that is present in the grinding zone
AC
(αf) is given by
𝛼𝑓 = 𝑉𝑟𝑝 ∆𝑡𝑔𝑝 /(𝑉𝑚 𝑓𝑐 ) = ∆𝑡𝑔𝑝 ⁄(∆𝑡𝑔𝑝 + ∆𝑡𝑎𝑝 + ∆𝑡𝑓𝑝 ) (A9)
In case there is overfilling of particles (U > 1), then particles occupy all voids
within the grinding charge, resulting in an expansion of the charge. In this case the
parameters U0 and εb0 will be
𝑈0 = 𝛼𝑓 𝑓𝑐 / (𝐽0 𝜀𝑏0 ) = 1
{ (A10)
𝜀𝑏0 = 𝛼𝑓 𝑓𝑐 / (𝐽0 )
20
ACCEPTED MANUSCRIPT
T
where
IP
1
𝐺(𝜉) = ∫ (Ψ − Ψ0 ) 𝜉 𝑑𝜉
CR
𝜉
𝑤𝑟2 (A13)
𝑐𝑜𝑠(Ψ) = −𝑟 = − 𝑓𝑤 2 𝜉
𝑔
US
{ 𝑐𝑜𝑠(Ψ0 = 𝑅0 𝑟 = 𝜉𝑖 ⁄𝜉 = 2 𝑟⁄(𝐷)
) ⁄ AN
The distance between the surface of the ball layer and the center of the mill in the
ascending zone, Ras or ξs, may also be estimated on the basis of the voids filling value.
M
When U ≤ 1, Ras would be equal to R0, and therefore ξi = ξs. Whenever U > 1, grinding
media tend to move towards the wall of the mill, in the ascending zone, during mill
ED
operation. This would result in segregation of media and particles, so that the porosity of
the layer of grinding media would be equal to a εb. Manipulating Eqs. (A5) and (A11), it
PT
gives
CE
With this relationship between ξi and ξs. it is possible to define Δta and Δtap so that
21
ACCEPTED MANUSCRIPT
1⁄
Δ𝑡𝑓 = (2 𝐻𝑎𝑏 ⁄𝑔) 2 (A17)
1⁄
2
Δ𝑡𝑓𝑝 = (2 𝐻𝑎𝑝 ⁄𝑔) (A18)
in which Hab and Hap are the free mean paths of fall of grinding media and particles,
T
respectively, which were given by
IP
𝐻𝑎𝑏 ⁄(𝐷⁄2) = [(1⁄2) 𝑓𝑤 2 (1 + 𝜉𝑖 2 )] + 𝑐𝑜𝑠𝜃𝑏0
CR
(A19)
US
− [(1 − 𝜖𝑏0 )⁄𝜀𝑏0 ] {[(1⁄2) 𝑓𝑤 2 (1 + 𝜉𝑖 2 )] + 𝑐𝑜𝑠𝜃𝑏0 }
AN
A detailed analysis of the expressions proposed by Nomura [20, 21] allow to
establish that, of all dynamic variables in the model, that is, J0, U0 and εb0, the only variable
M
that represents a challenge in estimating is J0, assuming that the mill is operating under
steady-state conditions. An efficient iterative algorithm that can be used to estimate it is
ED
previous study [24] and are only briefly reviewed here. Experiments were conducted using
manufactured sand rock. Zinc powder (99.9% pure, Merck®) was used as the solid tracer.
AC
The choice of zinc was due to the fact that this element was not detected in the chemical
analysis of the feed material. The experiment was carried out with the mill operating in
open circuit in steady state with 28% filling of 25 mm steel balls, turning at 75% of the
critical speed, corresponding to 58.7 rpm. Mixed with the feed, introduced at an average
rate of 6.9 kg/h, a total of 120 g of zinc powder were added, with 30 grams added every 15
seconds, as a pulse. In parallel to the addition of the tracer, sampling of the discharge was
started, the entire product being discharged every minute during the two-hour period.
22
ACCEPTED MANUSCRIPT
Measurement of the zinc content of the collected and then quartered samples was
performed by X-ray fluorescence (Shimadzu EDX-720).
It was assumed that the results could be fitted, as is commonly found in practice
[18] by considering three mixers in series with different volumes
𝑡 𝑡 𝑡
𝜃1 (𝜃2 − 𝜃3 )𝑒𝑥𝑝 (− 𝜃 ) + 𝜃3 (𝜃1 − 𝜃2 )𝑒𝑥𝑝 (− 𝜃 ) + 𝜃2 (𝜃3 − 𝜃1 )𝑒𝑥𝑝 (− 𝜃 )
𝐸(𝑡) = 1 3 2 (B1)
(𝜃1 − 𝜃2 )(𝜃1 − 𝜃3 )(𝜃2 − 𝜃3 )
T
IP
where θ1, θ2 and θ3 are the mean residence times in each of the mixers.
CR
Results are given in Figure B1, which shows the very good fit of the model to data.
The model was fitted considering the values of the ratio of the three residence times and the
US
total residence time in the mill, giving 1/ = 0.07, 2/ = 0.10, 3/ = 0.83.
AN
References
1. W.J. Whiten, A matrix theory of comminution machines, Chem. Eng. Sci., 29
M
(1974) 589-599.
2. J.A. Herbst, D.W. Fuerstenau, Scale-up procedure for continuous grinding mill
ED
design using population balance models, Int. J. Miner. Proces. 7 (1980) 1-31.
3. L.G. Austin, R.R. Klimpel, P.T. Luckie, Process Engineering of Size Reduction,
PT
AIME-SME (1984).
CE
4. L.G. Austin, K. Brame, A comparison of the Bond method for sizing wet tumbling
ball mills with a size-mass balance simulation model, Powder Technol. 34 (1983)
AC
261-274.
6. J.A. Herbst, Y.C. Lo, Analysis of the performance of large-diameter ball mills at
Boungainville using the population balance approach, Miner. Metal. Process. (1988)
221-226.
23
ACCEPTED MANUSCRIPT
7. L.G. Austin, K. Julianelli, A.S. de Souza, C.L. Schneider, Simulation of web ball
milling of iron ore at Carajas, Brazil, Int. J. Miner. Process. 84 (2007) 157-171.
8. J. Kown, J., Jeong, H. Cho, Simulation and optimization of a two-stage ball mill
grinding circuit of molybdenum ore, Adv. Powder Technol. 27 (2016) 1073-1085.
9. P. Faria, L.M., Tavares, R.K. Rajamani, Population balance model approach to ball
mill optimization in iron ore grinding. In: XXVII Int. Miner. Process. Congr., 2014,
T
Santiago, GECAMIN, 2014.
IP
10. F. Shi, W. Xie, A specific energy-based ball mill model: From batch grinding to
CR
continuous operation, Miner. Eng. 86 (2016) 66-74.
11. F.K. Mulenga, Sensitivity analysis of Austin’s scale-up model for tumbling ball
mills – Part 1. Effects of batch grinding parameters, Powder Technol. 311 (2017)
US
398-407.
12. F.K. Mulenga, Sensitivity analysis of Austin’s scale-up model for tumbling ball
AN
mills – Part 2. Effects of full-scale milling parameters, Powder Technol. 317 (2017)
6-12.
M
batch grinding data for simulation of industrial milling of platinum group minerals
ore, Miner. Eng. 63 (2014) 100-109.
PT
14. V.K. Gupta, D. Hodouin, M.D. Everell, The influence of pulp composition and feed
rate on holdup weight and mean residence time of solids in grate discharge ball mill
CE
15. S.R.H. Swaroop, D.W. Fuerstenau, A.Z.M. Abouzeid, Flow of particulate solids
AC
16. R. Hogg, Mass Transport models for tumbling ball mills, Control´1984, Society of
Mining Engineers, (Ed. J.A. Herbst), Chapter 7 (1984).
17. P. Songfack, R. Rajamani, Hold-up studies in a pilot scale continuous ball mill:
dynamic variations due to changes in operation variables, Int. J. Miner. Process. 57
(1999) 105-123.
24
ACCEPTED MANUSCRIPT
18. R.P. King, Modeling and Simulation of Mineral Processing Systems, Butterworth-
Heinemann (2001).
19. F. Shi, An overfilling indicator for wet overflow ball mills, Miner. Eng. 95 (2016)
146-154.
20. S. Nomura, Dispersion properties for residence time distributions in tumbling ball
mills, Powder Technol. 222 (2012) 37-51.
T
21. S. Nomura, Analysis of Hold-up in continuous ball mills, Powder Technol. 235
IP
(2013) 443-453.
CR
22. G.K.P. Barrios, R.M. Carvalho, A. Kwade, L.M. Tavares, Contact parameter
estimation for DEM simulation of iron ore pellet handling, Powder Technol. 248
US
(2013) 84-93.
23. J.A. Nelder, R. Mead, A simplex method for function minimization, Computer
AN
Journal. 7 (1965) 308–313.
24. A.L. de Oliveira, L.M. Tavares, Simulating a pilot-scale dry ball mill grinding
itabirite using batch grinding data, Proc. 6th Int. Congress on the Science and
M
25
ACCEPTED MANUSCRIPT
T
IP
CR
US
AN
M
ED
PT
CE
AC
26
ACCEPTED MANUSCRIPT
T
IP
CR
US
AN
M
ED
PT
CE
AC
27
ACCEPTED MANUSCRIPT
T
IP
CR
US
AN
M
ED
PT
CE
AC
28
ACCEPTED MANUSCRIPT
T
catalyst
182 16.2 3.55 99
IP
67.5 6.0 1.95 72
0.3 0.30 135.0 12.0 2.40* 84
CR
182 16.2 3.05 89
69 9.0 5.51 116
0.2 0.75 137 18.0 6.66* 127
US
188 24.6 7.98 136
69 9.0 4.31 126
0.2 0.30 142 18.6 4.57 140
Iron ore 183 24.0 5.34* 142
AN
69 9.0 5.45 88
0.3 0.75
147 19.2 6.11* 115
69 9.0 4.10* 115
M
29
ACCEPTED MANUSCRIPT
T
IP
CR
US
AN
M
ED
PT
CE
AC
30
ACCEPTED MANUSCRIPT
Table 6. Values of relative deviations in 80% passing sizes as a function of mill speed and
simulated scenario
T
Austin and Nomura
models for one 3.80 10 8.66 12 6.45 21.1
IP
perfect mixer
CR
Austin and Nomura
models for three 3.90 10 7.03 12 5.61 17.8
perfect mixers
US
Austin models for
three perfect
5.69 10 6.62 12 6.20 14.9
mixers assuming U
AN
=1
M
Modeling and simulation of grinding of spent catalyst and iron ore in pilot dry ball
ED
mill
Hold-up of continuous mill predicted using Nomura model
Product size predicted using Austin/Nomura models
PT
31
Figure 1
Figure 2
Figure 3
Figure 4
Figure 5
Figure 6
Figure 7
Figure 8
Figure 9
Figure 10