Sunteți pe pagina 1din 31

International Journal of Mineral Processing, 7 (1980) 1--31 1

© Elsevier Scientific Publishing Company, Amsterdam -- Printed in The Netherlands

SCALE-UP PROCEDURE FOR CONTINUOUS GRINDING MILL DESIGN


USING POPULATION BALANCE MODELS

J.A. HERBST and D.W. FUERSTENAU


Department of Metallurgical Engineering, University of Utah, Salt Lake City, Utah (U.S.A.)
Department of Materials Science and Mineral Engineering, University of California,
Berkeley, Calif. (U.S.A.)
(Received January 21, 1979; revised version accepted September 18, 1979)

ABSTRACT

Herbst, J.A. and Fuerstenau, D.W., 1980. Scale-up procedure for continuous grinding mill
design using population balance models. Int. J. Miner. Process., 7: 1--31.

This paper presents the development of a mill scale-up procedure based on population
balance models of grinding circuits. Models containing various degrees of details in their
representation of the kinetics of breakage, material transport through the mill, and per-
formance of the classifier are examined; and the extent to which each may be useful for de-
sign is discussed. Special attention is given to the type of experimental information re-
quired at the bench, pilot-plant and plant scales to enable these models to be applied suc-
cessfully to the design of continuous grinding mills. The grinding subproceases: breakage,
transport, and classification are taken into account in developing the present scale-up/de-
sign procedure. Utilization of the concept of specific energy consumption in kinetic models
is demonstrated. The mechanics of design of a commercial-scale mill within the population
balance model framework presented in this paper are illustrated using batch data obtained
for the dry ball milling of limestone.

INTRODUCTION

F o r the past q u a r t e r c e n t u r y , the size o f mills and mill drives r e q u i r e d f o r


c o m m e r c i a l - s c a l e grinding tasks has b e e n selected based o n t h e specific
energy ( e n e r g y per u n i t mass o f p r o d u c t ) necessary t o r e d u c e a feed material
to the desired p r o d u c t size. T h e c h o i c e o f specific e n e r g y as a scale-up
criterion is based on t w o i m p o r t a n t premises: (1) mills o f d i f f e r e n t sizes de-
livering the same specific e n e r g y will yield identical p r o d u c t s w h e n fed the
same feed material a n d (2) existing mill s i z e / p o w e r d r a f t relationships are ac-
c u r a t e e n o u g h t o a l l o w the selection o f a mill size w h i c h will deliver t h e neces-
sary e n e r g y at the design t h r o u g h p u t . Whenever possible the specific e n e r g y re-
q u i r e m e n t f o r a given feed t o p r o d u c t t r a n s f o r m a t i o n is d e t e r m i n e d in s u c h a
w a y as t o m i n i m i z e the design risk, i.e. t h e value is d e t e r m i n e d f r o m an exist-
ing full-scale o p e r a t i o n or f r o m a p i l o t - p l a n t circuit w h i c h is o p e r a t e d in a
similar fashion to that anticipated for the commercial installation. When com-
mercial- or pilot-scale data are not available, design engineers generally use the
Bond energy-size reduction equation, or its equivalent, to estimate specific
energy requirements using a w o r k index parameter determined experimentally
in laboratory scale batch grinding experiments (Bond, 1952, 1960, 1962). For
the m o s t part the simple Bond equation has served the industry well as a basis
for scale-up from laboratory-scale data; however, serious design errors are
sometimes made using this approach (Blaskett, 1969). In recent years im-
portant advances have been made in the development of detailed mathemat-
ical models for grinding circuits. The accuracy of predictions achievable with
these models, relative to the classical energy-size reduction models, suggests
that an alternative approach to scale-up based on detailed models holds con-
siderable promise for reducing design risk.
It is the opinion of the authors that the major shortcoming of the Bond ap-
proach to mill design is its failure to account for the important grinding-circuit
subprocesses -- breakage kinetics, material transport through the mill, and size
classification -- in an explicit manner in the design equations. This fact makes
it difficult to identify, or virtually impossible to assess, the effect of inevitable
deviations from the design assumptions on the performance of a proposed
commercial system. In contrast, the approach to design examined in this paper
is based on models which do contain explicit representations of the important
grinding-circuit subprocesses. In the context of these more detailed models the
design assumptions are well defined and the effect of deviations from these as-
sumptions can be evaluated b y simulation. The influence of changes in the
size distribution of the feed, breakage characteristics of the feed, the per-
formance of the classifier, etc. on the performance of the commercial mill can
all beinvestigated by simulation. In addition to providing accurate predictions
of mill capacity, these models are capable of describing the details of circuit
product-size distributions.

MATHEMATICAL MODELS F O R SCALE-UP DESIGN

A wide variety of types of mathematical models can, in principle, be used


as a basis for mill scale-up design. These model types can be classified accord-
ing to the amount of physical detail they contain and the experimental and
computational difficulty associated with their usage. Figure 1 depicts such a
classification diagrammatically. Each model t y p e is represented as a level in the
diagram. The levels, I through V, are arranged in ascending order of physical
detail, and associated mathematical and computational complexity. At level I,
all subprocesses are lumped into a simple mathematical model, i.e. breakage
kinetics, particle transport and classification behavior are represented b y a
single equation. The Bond correlation is the best example of level I models. At
level V, models and associated scale-up criteria are developed from first prin-
ciples, i.e. the design would be completely predictive. This level represents the
ultimate in design models. Unfortunately, this level of sophistication is clearly
DEGREES OF P H Y S I C A L D E T A I L
IN
M O D E L S FOR M I L L D E S I G N

T 7
~E COMPLETELY PREDICTIVE MODEL
\ /
19" NONLINEAR BREAKAGE AND
CLASSIFICATION BEHAVIOR
TRANSPORT DISTRIBUTED
\ /
]]T LINEAR BREAKAGE AND
CLASSIFiCATiON BEHAVIOR,
TRANSPORT DISTRIBUTED
\ /
Tr L I N E A R BREAKAGE AND
CLASSIFICATION BEHAVIOR,
TRANSPORT LUMPED
\/
"r BREAKAGE KINETICS, TRANSPORT
AND CLASSIFICATION BEHAVIOR
"REPRESENTED" BY A SINGLE
CORRELATION

Fig. 1. Classification of mathematical models for mill design according to degree of


physical detail.

unattainable owing to the complexity of natural mineral feed materials, the


complexity of the stress application environment which exists in tumbling
mills and the complexity of the mathematical structure which would be re-
quired for such a model. Between the extremes of levels I and V are three
levels of models, now in existence, which have considerable promise as a basis
for accurate mill scale-up design.
The tumbling mill grinding models which constitute levels II through IV are
all derivable from population balance considerations (Kelsall and Reid, 1965;
Herbst and Mika, 1970; Austin, 1971--1972; Herbst et al., 1971a). The distinc-
tion between levels is determined by the degree of generality of the equations
employed in the description of the kinetics of particle breakage (linear or non-
linear), the transport of particles through the mill (lumped or distributed) and
in the case of closed-circuit grinding the performance of the size-classification
device (linear or nonlinear). Linear breakage kinetics are said to prevail in a
mill when neither the probability of breakage of a particular particle (as
measured by the breakage rate function) or the distribution of daughter frag-
ments resulting from the primary breakage of that particle (as measured by
the breakage distribution function) is influenced by the size distribution in the
mill. Linear breakage kinetics are frequently observed in ball mill grinding
(Sedlatschek and Bass, 1953; Gardner and Austin, 1962; Austin et al., 1967;
Mika et al., 1967; Herbst and Fuerstenau, 1968; Herbst and Mika, 1970); how-
ever, significant deviations from linearity have been noted in certain tumbling
mill systems (Grandy and Fuerstenau, 1970; Herbst, 1971; Herbst and
Fuerstenau~ 1972). Distributed parameter descriptions entail the use of de-
tailed models for particle transport down the length of a mill. Lumped-
parameter descriptions make use of residence time distribution (RTD) infor-
marion to characterize material transport rather than a detailed particle trans-
port model. The lumped-parameter descriptions are applicable only in situa-
tions where to a good first approximation the kinetics of breakage are linear
and material transport is size independent. Finally, the use of a linear classifier
description for closed-circuit grinding entails the assumption that the cut size
and classifier efficiency curve do not depend on the size consist of the
material feeding the classifier.
Level III and IV models (distributed parameter models [Horst and Freeh,
1970; Mika, 1970; Mika and Fuerstenau, 1971; SchSrlert, 1971a,b]) are
capable, in principle, of depicting the evolution in time and position in the
mill of every size fraction in a particulate assembly. These models are general-
ly quite complex both mathematically and computationally. Level II models
(lumped-parameter models [Kelsall, 1965; Kelsall et al., 1968, 1968--1969,
1969--1970; Kelsall and Reid, 1969; Herbst et al., 1971a] ) are simply mill or
circuit input--output relationships. Because of the existence of general closed-
form solutions for the descriptive differential equations, parameter estimation
and circuit simulation are relatively simple (Herbst et al., 1971a,b). In
engineering applications the advantage of simplicity offered by the lumped-
parameter models must be weighed against the increase in accuracy achievable
with distributed parameter models. In this regard, the results of two recent
computational studies (Herbst and Mika, 1973; Herbst et al., 1976) have
shown that when properly applied level II can provide very good approxima-
tions to the behavior of complex systems which normally require level III and
IV models.
In the following discussion an approach to scale-up design is developed
based on the level II models. By incorporating the important subprocesses (i.e.
size reduction, particle transport, and size classification) explicitly into the
equations for mill design, a potentially more accurate and flexible basis for
scale-up than is presently available is defined. Future information obtained
from applying this type of model not only promises to yield more accurate
mill design but will also provide a foundation for the use of more detailed
models, if and when they are needed.
Before describing this alternative approach to scale-up, the Bond model and
the lumped-parameter population balance model will be briefly discussed in
order that the two approaches to modeling for scale-up design and the assump-
tions attendant to each can be compared.

BOND MODEL

In the Bond approach to mill design, three parameters are used to calculate
the specific energy requirement for commercial grinding, i.e., work index, a
feed-size parameter, and a product-size parameter (Bond, 1952). Implicit in
the definition of the work index is the fact that it represents a lumping of all
breakage, transport and classification processes into a single parameter which
characterizes the material in a closed-circuit wet-grinding configuration.
Modifying factors to the basic formula are applied to account for dry-grinding,
open-circuit grinding, very fine size-product and oversized feed. In some cases,
a diameter efficiency factor is applied to account for capacity and power
scale-up differences.
Experimentally, the work index is evaluated by dry locked-cycle grinding
tests in a standard Bond grindability mill (Bond, 1952) which has been
"calibrated" with data obtained from a continuous 8-ft. diameter wet mill.
Since the detailed procedure for performing this test is readily available it will
not be presented here. It is important to note, however, that this test proce-
dure uses ideal (or perfect) classification which is not attainable outside the
laboratory. Industrial scale classifiers have extremely varied separation
characteristics and typically deviate substantially from perfect separation be-
havior. In addition, equilibrium in a Bond test is equivalent to steady state in
a continuous plug flow mill operated in closed circuit (Gumtz and Fuerstenau,
1970). Studies of particle transport in continuous mills rarely reveal plug flow
characteristics and it has been found that the particle transport description re-
quires, at the minimum, the introduction of residence time distribution infor-
mation (Mori et al., 1964, 1967; Tanaka, 1964; Molerus, 1966; Kel~all et al.,
1969--1970; Molerus and Paulsen, 1970; Austin et al., 1971).
When applying the Bond model, it is implicitly assumed that all materials
break similarly, i.e. the same as an "ideal Bond material". An "ideal Bond
material" is characterized by a product-size-independent "work index" and
batch grinding size distributions which are described by the Rosin-Rammler
equation with an exponent (i.e., a fine size slope) of 0.5 (Grandy et al., 1971;
Austin et al., 1972). In actual practice, the "work index" is often dependent
on product size and batch product size distributions rarely follow a Rosin-
Rammler distribution with a fine size slope of 0.5. Consequently, the Bond
equation does not, in general, accurately represent the grinding characteristics
of the natural materials that are processed in industrial grinding circuits.
The Bond approach attempts to partially compensate for idealizations in
the test procedure and shortcomings of the model by making use of correla-
tions between Bond grindabllity test results and results obtained in a
"standard" 8-ft. milling circuit. Complete compensation is, of course, impos-
sible since this procedure cannot take into account inevitable deviations in
transport, classification etc., characteristics of the commercial circuit from
those of the "standard" circuit.

LINEAR LUMPED-PARAMETER MODELS

In contrast to the Bond model, which has arisen in a completely empirical


fashion, the lumped-parameter models are derived in a physically meaningful
way from population balance considerations. This type of model has been
shown to provide accurate simulations of the entire size distribution of ground
products produced in batch (Gardner and Austin, 1962; Herbst and
Fuerstenau, 1968, 1972), locked-cycle (Gumtz and Fuerstenau, 1970) and
continuous open- and closed-circuit (Kelsall, 1965; Kelsall et al., 1968,
1968--1969, 1969--1970; Kelsall and Reid, 1969) grinding in laboratory-scale
ball mills and is, therefore, a promising candidate for ball mill scale-up design.
The development of the equations required for scale-up design is sketched be-
low (Herbst et al., 1971a).
The linear, size-discretized model for breakage kinetics is obtained by divid-
ing the particulate assembly being ground into n narrow size intervals, (xi,
xi÷l ), i = 1, 2 , . . . n. The intervals chosen generally correspond to those of
the Tyler x/2 sieve series; in which case xi = x/2 xi+l. A mass balance for the
material in the i-th size interval at time t yields (Herbst and Mika, 1970;
Austin, 1971--1972) for i = 1, 2, . . . . n:

d[Hmi(t)] i-1
- -kiHmi(t ) + ~ biykjgmy(t ) (1)
dt i=1
Here, mi(t) is the mass fraction of material in the i-th size interval and H is the
total mass of material (hold up) being ground. In eq. 1, ki, the breakage rate
function for the i-th size interval, denotes the fractional rate at which material
is broken out of the i-th size interval and bii, the breakage distribution func-
tion, represents the fraction of the primary breakage product of material in
the j-th size interval which appears in the i-th size interval*. When the
breakage rate and distribution functions are independent of both the size con-
sist in the batch mill and time, the kinetic model is linear with constant
coefficients.
The set of n coupled-linear differential equations in eq. 1 are conveniently
represented by a single matrix equation (Herbst and Fuerstenau, 1968;
Grandy et al., 1969)**.
d [Hm (t)]
- [I-B]KHm (2)
dt
For a batch mill with initial size distribution mBATCH(0), H is a constant and
an analytical solution for the product size distribution is readily obtained by
standard matrix techniques:
mBATCH(t ) = exp [ - [ I - B] Kt] mBATCH(0 ) (3)
For the case in which no two selection functions are equal, the matrix ex-
ponential is readily simplified by a similarity transformation to give (Herbst
and Mika, 1970):
mBATCH(t) = T J ( t ) T - 1 mBATCH(0 ) (4)

*In previous publications the authors have used the terms selection function (si) and
breakage function. The terms breakage rate function (ki) and breakage distribution func-
tion are m u c h more descriptive of the actual physical quantities represented by these func-
tions and, thus, the latter have been adopted.
**All quantities are defined in the Nomenclature list (p. 27).
where:
I 0 i<j
1 i=y
Tij= i-1 bilkl
kik] Tl] i> j
l=1
exp (-kit) i =j
Jiy(t) =
0 i~=j
It is important to note that in the development of eq. 4 it has n o t been
necessary to assume a functional form for the p r o d u c t size distribution (such
as Rosin-Rammler, Alyavdin, Gaudin-Schuhmann), rather the evolution of the
batch size distribution is, in this case, completely determined by two sets of
physically identifiable and experimentally measurable parameters (ki, bij).
In addition to its usefulness for batch grinding descriptions, eq. 4 plays a
central role in the description of continuous grinding. When all particle sizes in
a continuous mill are characterized by a single residence time distribution, the
steady-state size distribution from an open-circuit mill can be obtained as an
average of batch responses weighted with respect to the distribution of resi-
dence times of material in the mill:
oo

roMp = f m(t)BATCHE(t)dt (5)


0

In eq. 5, mBATCH(t) is the response of the mill operated in a batch fashion


with an initial size distribution, mBATCH(0), equivalent to the steady-state
size distribution of the mill feed mMF. The quantity E (t)d t, the exit age (resi-
dence time) distribution of the material, is an experimentally determinable
function which denotes the fraction of material in the mill discharge which
has resided in the mill for a time interval t to t + dt..
Substitution of eq. 4 into eq. 5 yields:

mMp=T [ 7 J(t)E(t)dt] T--~mMF (6)


0

It is convenient to express the integrand in terms of the dimensionless time


variable, 0 = t/r, where r, the mean residence time, is the ratio of the mass
holdup in the mill to the mass flow rate of feed to the mill (r = H]MMF). The
result is written c o m p a c t l y as:
roMp = T JC(T) T -1 mMF (7)
where:
O0

/ exp (-[kF]O)E(O)dO i=j


JcJ )= o

0 iCj
It is particularly significant that eq. 6 has been developed in terms of an
arbitrary residence time distribution, E (0); hence the steady-state response of
open-circuit mills can be represented in a more realistic fashion than is pos-
sible with a model which invoked additional assumptions (e.g. the plug flow
assumption).

~l TUMBLING MILL J cLAss, ,E.


MF. m; MMF'mMF -1 T JC (T) r -1 MMp' rnMp -I ¢ i Mp, rnp

Fig. 2. Schematic representation of standard closed-circuit grinding.

A steady-state description for closed-circuit grinding is obtained from eq. 7


and a material balance around the classifier (Fig. 2). Formally no assumption
regarding the cut size and shape of the classifier efficiency curve (as character-
i z e d b y the classifier matrix C) is required to obtain the description, b u t the
assumption of linear classifier behavior simplifies the mathematics consider-
ably. In this case the mass flow rate of the circuit product, Mp = MF, is given
by:

Mp = MMp/eT I Jc(T) T -1 ( I - C T JC(T) T - l ) -1 mF (8)


and the size distribution of the circuit product, rnp, is given by:

rnp = [ I - C ] T Jc(r) T -1 [ I - C T Jc(r) T - 1 ] - l mF (9)


Equations 8 and 9, which are readily programmed for a digital computer,
form the mathematical framework for the mill scale-up design procedure
examined in this paper.

BREAKAGE RATE AND DISTRIBUTION FUNCTION SCALE-UP

In order to apply eqs. 8 and 9 to the problem of design scale-up, it is ap-


parent that the dependence of the kinetic parameters (ki, bii) on mill diameter,
mill speed, media load and size, and particle holdup must be known.
Unfortunately, there is nothing inherent in the model structure which allows
the a priori prediction of these dependencies. There have, however, been
several recent studies which have been devoted to defining such relationships
based on data obtained in laboratory-scale ball mills (Mika et al., 1967; Herbst
and Fuerstenau, 1972, 1973; Olsen and Krogh, 1972; Austin, 1973; Malghan
and Fuerstenau, 1976). One of the correlations is particularly interesting be-
cause it seems to account for most of the important variables with a single cor-
relation (Herbst and Fuerstenau, 1973; Malghan and Fuerstenau, 1976). In ad-
dition, this correlation leads quite naturally to the conclusion that specific
energy input to a mill is a principal parameter for scale-up. This correlation
can be stated simply as follows:
(1) The breakage rate functions, ki, are proportional to the specific power
input to the mill:

where the set of proportionality constants, k E (specific breakage rate func-


tions), are independent of mill design and operating variables*.
(2) The breakage distribution functions, bii, are invariant with changes in
mill design and operating variables.
Incorporation of these relationships into eq. 1, with the recognition that
for a batch mill drawing constant power, the quantity Pt/H is exactly equal to
the specific energy input to the mill, E, yields the normalized batch grinding
model for i = 1, 2 , . . . , n:
- - i-1
dmi(E) =-kEmi (~) + ~_l bijkEmi(E) (11)
dE, j=l
This "detailed energy-size reduction relationship" predicts that for a given
material and feed-size distribution [mi(0)] a necessary condition for identical
product-size distributions in different batch mills is identical specific energy in-
puts into each mill -- independent of mill dimensions and mill operating vari-
ables in the normal operating range.
Figures 3 and 4 show a portion of the data supporting the specific power
correlation for the kinetic parameters (ki, bi]). Figure 3 presents a log-log plot
of feed-size breakage rate function values, k 1, plotted against measured P/H
values for: (a) the dry ball milling of dolomite in a 25.4 × 29.2 cm mill with
different ball loads and mill speeds (Herbst and Fuerstenau, 1973); and (b) the
dry ball milling of limestone in different sizes of mill (Siddique, 1977). The
fact that in each case the k i values fall, to a good approximation, on a single
straight line with unit slope supports eq. 10. Figure 4 shows that in each case
the feed size breakage distribution functions are nearly identical for different
levels of mill design and operating variables.
Figures 5 and 6 show actual scale-up predictions based on eq. 11. In this
case data obtained for the dry ball milling of 10 × 14 mesh limestone in a
*k/E values do appear to depend on ball size (Malghan, 1976).
10

1.0
I I I I

w
i
Z
0.7--

Z
0
0.5--
Z

MILL DIAM. = 2 5 . 4 - 7 6 . 2 C M

0.3 ~- M; M:
w 7x9 MESH DOLOMITE • 0.53-0.90 0.35-0.50 0.8-L6
m
IOxl4 MESH LIMESTONE 0 0.6 0.6 1.0

0.2 I I I 1
0 20 30 40 50
SPECIFIC P O W E R , P/H (KW//T)

Fig. 3. Demonstration of the validity of specific power--breakage rate function relationship


(eq. 10) for changes in operating variables (dolomite [Herbst and Fuerstenau, 1973 ]) and
design variables (limestone [ Siddique, 1977 ] ).

MESH SIZE

j,
Z 400 200 I00 48 28 14 8
0 t.O I I I I I ~ /" I
/Z

~- 0,5

UJ

u~ ~ 0.2

LIMESTONE -~//~, /
Z 0.1
,.q~
,. 0.05 /r" 1] RANGEOF VALUES
_>
I,-
/'~- OOLOMITE
J
:D
.I. /
:3 I I I I i I f
(D 0.02 I0 20 50 100 200 500 I000 2000
PARTICLE S I Z E , MICRONS

Fig. 4. Demonstration of the invariance of breakage distribution function with operating


variables (dolomite [Herbst and Fuerstenau, 1973 ]) and design variables (limestone
[Siddique, 1977] ).
11

MESH SIZE
400 200 1(30 48 28 14 8
LO
25.4 CM
MILL 0~ 0 / ' / ~
BATCH DRY o / .i.o o"/I
0.5 10xl4 MESH FEED / /o / //
bJ LIMESTONE /o / O/O//
_Z
b- ORIND,NGT,ME o///
IM,NO'ES~ / o/ / o/o/
0.2
o/° o / /
LL
0.1
,.o °C-.-" / o///
ILl 4.o o/ o/° o/ o/
__>
o.1°/ / o/
_~ 0.05
:E
2.o o~ o/°/ /
o/ o..1"
1.0 of / oE
XPE
RM
IE
NTA
L
0.02
o o~ -- FITTED
0.5
0.01 I I i I I I I
I0 20 50 I00 200 500 I000 2000
PARTICLE SIZE, MICRONS

Figi 5. Fit of normalized grinding m o d e l (eq" 11) to data o b t a i n e d for the b a t c h ball milling
of 10 X 14 m e s h l i m e s t o n e in a 25.4 cm mill.

MESH SIZE
400 200 I00 48 28 14
n~ 1.0 I i i i 0~--(~ ~ ~/.r v
ILl 76.2CM MILL 0~ /~"
Z
U. BATCH DRY / ~ fJ'
-I0 MESH FEED 0 "~0
Z LIMESTONE 0/ /" /
0 0.5
l- GRINDING TIME O/ /" ~" "
(MINUTES) ~ / /
n,.. /
h 4.0 0 .~
.t
.f
W
~0.2 /- f 0 Experimentol
I'-- /-
FEED / - - - - - FEED
J " - - Predicted
=E

uO.I I I 1 I I I I
I0 20 50 I00 200 500 IO00 2000
PARTICLE SIZE , MICRONS

Fig. 6. Prediction of t h e p r o d u c t size d i s t r i b u t i o n o b t a i n e d by grinding minus 10 mesh lime-


stone in a 76.2 cm mill u s i n g the normalized grinding m o d e l w i t h p a r a m e t e r values o b t a i n e d
f r o m a 25.4 cm mill.

25.4 X 29.2 cm batch mill were used to estimate the kinetic parameters (k E,
bi]). The overall fit of the model to these data is depicted in Fig. 5. In turn,
these k E and hi1 values were used in conjunction with known P/H values for a
76.2 cm × 40.6 cm mill to predict the batch grinding response for a - 1 0 mesh
limestone feed. These predictions are compared with experimental measure-
ments in Fig. 6. These results and others provide convincing support for the
12

validity of specific p o w e r scale-up predictions for small and intermediate size


ball mills.
The breakage rate function scale-up relationship given b y eq. 10 can be ex-
pressed directly in terms of mill dimensions and operating variables by recogniz-
ing that the net p o w e r draft of a mill can be obtained from an equation of the
form (Rose and Sullivan, 1958; Bond, 1962):

P= c~ (N'*, MB*, Mp*, Q * ) ' L D (2"5+8) (12)

and the mass holdup of material in a mill can be written as:


H = ¢2 (MB*, Mp*). LD 2 (13)
where D is the mill diameter, L is its length and ¢,, ¢2 are functions of the
dimensionless mill speed = N * , dimensionless ball load = MB*, dimensionless
particle load = Mp*, and dimensionless ball size and lifter geometry variables =
Q*. The parameter 5 in eq. 12 is zero by dimensional analysis (Rose and
Sullivan, 1958) b u t values ranging from -0.1 to +0.1 are generally used in
empirical correlations (Anonymous, a, b, 1971; Bond, 1962; Rowland, 1976).
Substitution of eqs. 12 and 13 into eq. 10 yields:

k i = k/E~a (N *, MB~, Mp ~, Q*)D (0"5+~) (14)

According to eq. 14, breakage rate functions measured for the same material
in mills of different sizes which possess complete kinematic similarity (N*, MB ~,
Mp*, Q* invariant) should vary as the mill diameter to the (0.5 + 3) power
(Fig. 7). A portion of the data supporting eq. 14 is shown in the log-log plot
of k~ versus D values presented in Fig. 7 for dry limestone grinding in 12.7 ×
14.6 cm, 25.4 × 29.2 cm and 50.8 × 58.4 cm mills with similar lifter geometry
and for fixed N*, MB* and Mp* values. Notice that in each case the kl values
fall on a straight line with slope near 0.5, as predicted b y eq. 14. A similar ob-
servation has been made for the grinding of anthracite in 20.3 cm and 60.9 cm
mills (Austin et al., 1973).
A test of the accuracy of predictions achievable with direct diameter scale-
up of breakage rate functions (eq. 14 and a mill-size independent breakage dis-
tribution function, Fig.8) is shown in Fig.9. Batch grinding experiments per-
formed in a 25.4 × 29.2 cm mill fitted with a torque m e t e r were used to esti-
mate ~3 (0.6, 0.5, 1.0) for 5 = 0, the breakage distribution functions bij, and spe-
cific breakage rate functions, k~, for the dry grinding of limestone (Malghan, 1976).
In turn, these estimates were used in conjunction with eq. 11 to predict the
evolution of the p r o d u c t size distribution in a 50.8 × 58.4 cm mill operated

Fig. 7. Demonstration of the validity of mill diameter--breakage rate function relationship


(eq. 14) for kinematieally similar conditions ~Malghan, 1976).

Fig. 8. Demonstration of the invariance of breakage distribution function with mill


diameter.
13

15
I l I I I I I

LIMESTONE
T 1.0 8 xlO MESH

z"
0 I 0 x 12 M E S H
~_ 0.7
Z
u_

I-
< 0.5
n,.
Ld
(.9
v
W
r..r"

0.3 - .* M; ~¢
n 0.5 0.4 I.O
Z~ 0.6 0.5 1.25

0.2 1 I I I I I I
0 20 50 50 70 I00
MILL DIAMETER, CM

1.0 I I I I I I I I I I 1 I I I [ I

DRY LIMESTONE
Z 2.54 CM BALLS
0
0.64 CM LIFTERS
Z
N"= 0 . 6 , M ; = 0 . 5 , M;=,.25

U_

m
IOx12 MESH
O3

W
0. I 8 x l O MESH

W
m MILL OIAM.
W (CM}
>
[] 50.8
t--
0 25.4
.._1 A t2,7

(.J
0.02t--
30 I00 I000
PARTICLE SIZE, MICRONS
14

with the same fraction of critical speed, fractional ball loading, fractional par-
ticle loading as used in the 25.4 × 29.2 cm mill. The good agreement between
experimental and predicted size distributions shown in Fig. 9 attests to the ap-
propriateness of direct scale-up of breakage rate functions for the dry ball mill
grinding in mills exhibiting complete kinematic similitude.
The use of these approximations for continuous grinding mill design is ex-
plored in a subsequent section.

PARTICLE SIZE, MESH


400 , 200 ICE) 48 28 14 _ 8
1.0 , l ' , ' l ' l ~ . . . ~ r " ~ ' ~ J ~

0.5
- . , d _o'" .,,0-..5 j . , -
n," _,./`- ....- ,.,/-...- ,./o /
bJ
Z GmNO,NQ ~ / J / - / /
,'7
,,,,.UTES, 7 " j ' - ~ ..'= / ."
z 0.2
o
t-
~.)

°.OI.o.-
~ 0.1 -
D o-
"" P/` -
- -

I~ /0/" J ~ EXPERIMENTAL
0.05
Z 0 ~ 5 CMMILI.-3.3 KS OF FEED --
• t~ ) u /
P" y ~ ,O., CM M,LL-26.4 KG OF FEED
0
1-1 2/. / ..... FITTED
f / PREDICTED
_ ,.oI.~
0.02 LIMESTONE
(8xlO MESH)

0.01 I I I I I I
50 I00 ~00 500 I000 2000
PARTICLE SIZE, MICRONS

Fig.9. Direct mill diameter scale-up predictions (eqs. 11 and 14) of 8 X 10 mesh limestone
grinding behavior in a 50.8 cm mill using best fit parameter estimates obtained from a 25.4
cm mill operated under kinematically similar conditions.

CLASSIFIER AND RTD P A R A M E T E R SCALE-UP

Some experimental and theoretical studies have been reported which have
given rise to classifier (Stewart and Restarick, 1967; Molerus, 1967; Lynch et
al., 1967) and residence time distribution (Mori et al., 1964, 1967; Tanaka,
1964; Molerus, 1966; Kelsall et al., 1969--1970; Molerus and Paulsen, 1970;
Herbst et al., 1971a; Karra, 1976) descriptions which are suitable for use in
conjunction with the lumped-parameter models. There is, however, very little
15

information available concerning the scale-up of the parameters of these de-


scriptions. If the lumped-parameter models are to become useful for scale-up
design, classifier efficiency curves (determined from a size fraction by size frac-
tion material balance around the classifier) and mill residence time distribu-
tion information (determined from stimulus-response experiments using ap-
propriate tracers) must be obtained for a variety of types and sizes of com-
mercial scale equipment.
In the absence of scale-up data it will be assumed for the purpose of this dis-
cussion that: (1) for a particular mill type if the ratio of mill length to mill
diameter (L/D) is held constant and kinematic similarity is maintained for the
grinding media (i.e. fixed N*, MB*, etc.) then the dimensionless residence time
distribution E (0) is independent of the size of the mill; and (2) equivalent size
classification performance (cut size and efficiency) can be achieved in high
and low capacity circuits. These assumptions simplify the mathematical
manipulations required for design and further, when these assumptions are
combined with the kinetic parameter scale-up approximations given in the last
section, t h e y do in fact lead to a mill d i a m e t e r - c a p a c i t y relationship which is
frequently employed in current mill design practice, namely:
Mp cc LD (2-5+8) (15)

Equation 15 follows from eqs. 8 and 9 because a necessary condition for


equivalent product-size distributions and circulating loads in two circuits (con-
sisting of mills of diameter D, and D2 with the same L/ D ratio and classifier
with equivalent ds0 size and efficiency) receiving feed materials with identical
size consists is (Herbst and Fuerstenau, 1972):
k i (D,) T ( D,) = ki(D2)'r(D2) (16)
which upon making the substitutions:
T = H/MMF = HI(1 + C.L.)Mp

(N*, MS*, Mp*' Q . ) L D ( 2 . 5 +~)


k i = =
H

and requiring that N*' MB*' Mp*, and dB are the same for each mill yields:

Mp~ = [ MP1 7 L2D2(2. 5+8)


L1D1(2.5+~)J (17)

Equation ! 7 is equivalent to eq. 15 since Mp1/L1 D1 (2"5+8) can be taken as a


constant pertaining to a reference mill of diameter D1. Note that this analysis
suggests that direct scale-up of a commercial mill using eq. 15 is only ap-
propriate when reference mill data are available for a mill having the same
dimensionless residence time distribution and classifier efficiency, as well as
the same dimensionless ball load, mill speed, etc., as the full-scale commercial
16

mill. In other situations, e.g., scaling up from batch or locked-cycle grinding


experiments, more detailed calculations involving the original design equations
are required for accurate scale-up.

COMMERCIAL BALL MILL DESIGN FROM LABORATORY GRINDING DATA

In order to apply any of the linear lumped-parameter models to the design


of a commercial mill, experiments with a reference mill are required to esti-
mate the kinetic parameters which characterize the breakage properties of the
materials to be ground. Figure 10 illustrates the general scale-up procedure. If
the kinetics of breakage are truly linear, simple batch experiments involving
an arbitrary size distribution will suffice for parameter evaluation. Experience

COMMERCIAL MILL DESIGN


FROM LABORATORY GRINDINGDATA

TRADITIONAL INPUTS COMPUTERMODEl_ 1 ADDITIONAL INPUTS


I a. Estimote Breekoge I
L Kinetic Parameters 1 @
"=,,,Jb. Design Mill L.
r q C, TestSensitivity J~.

U
COMMERCIALMILL SIZE AND POWERREQUIREMENTS
PRODUCT SIZE DISTRIBUTIONS
SENSITIVITYANALYSIS
Fig. 10. Schematic representation of the general grinding mill scale-up design procedure
using population balance models.

has shown that in instances where the model is known to be quite approximate,
the grinding tests used for parameter evaluation should be performed in such a
way that the product-size distributions obtained in the laboratory are very
similar to those desired for the commercial circuit (Siddique, 1977). For
closed-circuit design with an approximate model, this would entail performing
locked-cycle experiments using the same feed-size distribution as expected for
the full-scale circuit and size classification which closely approximates that of
the commercial classifier. In the context of the linear size discretized kinetic
17

model, parameters can be readily estimated from either batch or locked-cycle


data b y nonlinear regression (Herbst et al., 1971a; Herbst and Rajamani, 1977).
From a design risk standpoint, it is desirable to use the largest reference mill
possible because the magnitude of scale-up is reduced. Furthermore, with
larger reference mills the ball-size effects are minimized. Note that in a mill
scale-up, the kinematic similitude between ball diameter and mill diameter,
and ball diameter and particle size usually cannot be satisfied simultaneously.
From a practical standpoint, however, experiments in a small mill are obvious-
ly easier to perform and require less material. Therefore, a compromise refer-
ence mill size must be chosen such that the mill diameter is large enough to re-
duce the scale-up risk and yet small enough to easily obtain the necessary data.
Once the kinetic parameters have been evaluated from tests in a reference
mill, the design engineer must select a material transport model and classifier
efficiency model (if closed-circuit) for the full-scale circuit based on his
knowledge of commercial equipment characteristics. In addition, he must
specify a feed-size distribution and desired production rate. The mill dimen-
sions required to obtain the desired product (for a given (N ~, MB~, Mp~, Q*))
can then be calculated from the design equations using an appropriate com-
puter program. Finally, the effect o f deviations from the design assumptions
on the behavior of the commercial circuit can be evaluated b y simulation.
In order to illustrate the mechanics of mill scale-up design within the popula.
tion balance model framework, the parameter scale-up approximations iden-
tified in the last t w o sections were used in conjunction with the linear lumped-
parameter model to design a continuous overflow mill for grinding limestone
based on data obtained in a batch reference mill. In this example the material
transport behavior for the commercial mill is specified to be equivalent to that
obtained from two ideal mixers in series*. The feed material is to have a
Gaudin-Schuhmann size distribution with a top size (size modulus) of 8-mesh
and a distribution modulus of unity. The continuous mill is to be operated in
closed-circuit with a classifier which has a 150-mesh cut size and which by-
passes 30% of the particles finer than 150-mesh into the oversize product, i.e.
Cii = 1.0 for xi >~ 105 pm and Cii = 0.3 for xi < 105 um. This means that the
classifier efficiency is 70%, independent o f the classifier feed-size distribution.
The dimensionless mill speed (N*), ball load (MB*) and particle load (Mp*) are
to be 0.6, 0.5 and 1.0, respectively. The design production rate for the circuit
is to be 100 TPH at 250% circulating load.
For the scale-up design the following steps were required:
(1) Perform batch grinding experiments in a reference ball mill. In this
example limestone was stage crushed to minus 6.35 mm and screened to ob-
tain an 8 × 10 mesh feed material. The feed material was dry ground sequen-
tially in a 25.4 × 29.2 cm ball mill containing 30 kg of 2.54 cm balls. The
material charge was 3.3 kg, the mill speed was 53 rpm and the grinding time

*The residence time d i s t r i b u t i o n ( d e n s i t y ) for t w o ideal m i x e r s in series is E(O) =


40 exp ( - 2 0 ) .
18

followed a geometric progression, i.e., 0.5, 1, 2, 4, etc., min. After each grind-
Lug period, the mill was discharged and a 500 g sample was split out for sieve
analysis. The sieving procedure consisted of wet screening on 400 mesh fol-
lowed by 30 min of dry screening on a Ro-tap Sifter using a complete x/2
Tyler series, 10 through 400 mesh (n = 13 size intervals). The sieved products
were weighed and recombined with the balance of the ground material and the
entire 3.3 kg charge returned to the ball mill for the next interval of grinding.
The net mill power was calculated from the mill rotational speed and the
average torque measured by a Baldwin Lima SR4 torque sensing device (Yang
et al., 1967).
(2) Estimate the feed-size breakage rate function. From the linear model
for batch grinding, eq. 4, the behavior of the top size fraction is predicted to
follow the relationship:
ml (t) = ml (0) exp (-hi t) (18)

1.0( I I I I I
LIMESTONE
(SxlO MESH}
0
v 25.4 CM MILL
E

E- 0.5

d
Z
Z

IE
W
I1:

J 0.2
re"
!"-

T
if)
,,, 0.1

I.L
0
Z
kl:-d[l°gllml(t)]dt ['~

\
o 0.05
0
~r
h

0.02 I l I I I
I 2 3 4 5
GRINDING TIME, MINUTES

Fig. 11. Determination of feed size breakage rate function from 8 X 1 0 mesh limestone
feed disappearance data.
19

Therefore, a semi-logarithmic plot of ml (t)/ml (0) (8 X 10 mesh material)


versus grinding time will yield a straight line with negative slope equal to k l
provided that the breakage kinetics are linear. The results for limestone are
shown in Fig. 11 where it is seen that the linear model accurately describes
the grinding of the feed size material. The value of k l determined from the
plot is 0.780 min -1 .
(3) Estimate the feed-size breakage distribution functions. It has been
shown that the feed-size breakage distribution function can be estimated from
linear-linear production rate plots for the material contained in size fractions
finer than the feed. Equivalently, the cumulative breakage distribution lunc-
h
t i o n (Bil = ~ bkll c a n be estimated b y plotting the cumulative fraction
\
k=i ]
finer than size xi versus time (Fig. 12). From the slope, Fi measured at short
grind times, the cumulative breakage distribution function is calculated accord-
ing to:
Si~ = Fi/k, (19)

I I
- Z8 MESH

0.15
//r
I

/
/
- 4 8 MESH

/
/

/
/ - I 0 0 MESH

/
nr /
W
Z
,T
/ iI
Z
o 0.10
t-
o

LL
i,i
I--
..J

0.05 /,~ 3400 MESH


(J

7:::XS
0 I I
0 R I 2 3
G R I N D I N G TIME, MINUTES

Fig. 12. D e t e r m i n a t i o n of " p r o d u c t i o n rate c o n s t a n t s " for t h e grinding o f 8 x 10 mesh


limestone.
20

and individual breakage distribution functions obtained by difference:


bil = Bil - Bi+I,1 (20)
For limestone, the estimated cumulative breakage distribution functions are
shown in Fig. 13 where it is noted that the slope in the fine size range is ap-
proximately 0.75.

PARTICLE S I Z E , MESH
400 200 I00 48 28 14
I ' I I ' I ' I ' 'i

z 2,5.4 CM MILL
0
LIMESTONE
t--
¢9
Z 0.5
u_
Z
0
C-
m
0.2

Q
bJ
c9
0.I

t~
re
m

tlA
> 0.05 0.75
b-

(.9

0.02
50 IOO 200 500 IO00 2000
PARTICLE SIZE, MICRONS

Fig. 13. Estimated cumulative breakage distribution functions for 8 x 10 mesh limestone.

(4) E s t i m a t e the n o n f e e d - s i z e k i n e t i c p a r a m e t e r s . In principle, the breakage


rate and distribution functions for the other size fractions can be obtained by
performing experiments with each size as a feed material. Alternatively, an
estimation scheme based on experimental observations from grinding other
materials can be used. First it is often observed that the breakage distribution
functions for homogeneous materials can be normalized with respect to the
feed-size material, i.e. for j = 2, n; i = ], n

Bij = B i - j + 1,1 (21)


and therefore all breakage distribution functions can be defined from the esti-
mated feed-size breakage distribution function. Secondly, it has been observed
that the breakage rate function size dependence frequently follows a power-
21

law relationship:
°

with a slope, a, approximately equal to the slope of the cumulative breakage


distribution function plot in the fine size range. In the limestone example the
nonfeed-size breakage distribution function was assumed to be given by eq.
20, and kl = 0.780 and ~ -- 0.75 were used in conjunction with eq. 22 to
provide initial estimates of (ki; i = 2, 13) for the batch mill.
(5) Calculate the specific breakage rate functiorL Initial estimates of the
specific breakage rate function can be calculated from eq. 10, i.e.,
k~- kill
P (23)

using the initial estimates of ki from step 4 and P/H as determined from the
batch experiments. In turn, these initial estimates can be improved by using an
appropriate nonlinear regression algorithm.
In this example, the Gauss-Newton estimation algorithm referred to earlier
(Herbst et al., 1971b) was used to obtain values of k E and a (in eq. 22) which
resulted in the "best fit" (in the sense of least squares) of the linear model to
the batch data. The initial and final estimates of the specific breakage rate
functions for limestone are plotted in Fig. 14.
(6) Calculate specific energy to satisfy design conditions. The specific
breakage rate functions, breakage distribution functions, transport model, clas-
sifier constants and feed-size distribution are inputed to a computer program
which simulates the grinding circuit behavior according to eqs. 8 and 9". An
initial estimate for the specific energy, P/MMF, is introduced into the program
and the corresponding circulating load obtained from the simulation. Since a
log-log plot of circulating load versus specific energy is approximately linear
over a wide range of circulating loads (see Fig. 15), the approximate specific
energy required to obtain a 250% circulating load is easily determined by
linear interpolation between the first simulation and another simulation se-
lected such that the 250% circulating load point is bracketed. By repeating
this procedure the exact specific energy is quickly obtained. This value is
1.82 kWh/ton for the current example.
(7) Calculate the mill size from the specific energy. Once the specific
energy is known, the mill dimensions can be determined in the usual way. For
example, Mp is 100 TPH which means that MMF is 350 TPH (250% circulat-
ing load). Then the mill power is calculated from the specific energy

*When the specific breakage rate f u n c t i o n s are used, k E is identified w i t h k i and P/MMF is
identified wit~ r in eqs. 8 and 9. This can be seen by introducing eq. 10 into the expres-
sions for Tij and J c / j (eqs. 4 and 7 and simplifying).
22

S I Z E I N T E R V A L , TYLER MESH
2"IO/4oo IO0/150 55/48 14/20
5.C I I I ' ' I
!
Z
0
-.t. LIMESTONE
1-
s J
x~
2.C

Z
0

(D 1,0
Z

I,iJ
I--
,-it. 0.5
<
u.I
(.9

hi
n.-"
El
0.2
(_:,
E f
w
a.

O. I I I I I I
50 I00 200 500 I000. 2000
GEOMETRIC MEAN PARTICLE SIZE, MICRONS

Fig. 14. D e p e n d e n c e of the initial and improved estimates o f the specific breakage rate f u n c
tion on particle size.

2000 I i I I i

LIMESTONE
%. (8xlO MESH)
Q "R
,%
IOOC --
J ",,Q
%
Z
l.-

J 500 "o,
0
"%
"o,,
Z
i,I
h.
0
~ - - 250 %
rr

~.ez q.

I00 I I I
0.2 0.5 1.0 2.0 5.0
SPECIFIC ENERGY INPUT, P/MMF , K W H / T O N

Fig. 15. Estimation of the specific energy input required to obtain a 250% circulating load
from simulations of the grinding circuit.
23

(1.82 kWh/ton of ball mill feed) to be 1.82 X 350 or 637.0 kW. The mill dimen-
sions are calculated from eq. 11 written in the form:

P 3.5+~
D = $I(N*, MB*, Mp ~, Q * ) [ L / D ]

Substituting P = 637 kW, LID = 1.15 (chosen to be the same as the batch
mill), 5 = 0 and ~i = 0.1078 kW(ft) -3"s (determined from the power measure-
ment for the batch mill f o r N * = 0.6, MB* = 0.5, and Mp* = 1.0) the required
dimensions for the commercial mill are f o u n d to be D = 11.5 ft. and L =
13.2 ft. *. Figure 16 shows the simulated mill discharge and product-size
distribution (from eqs. 8 and 9) expected for the commercial mill at the design
throughput.

I.O
!CLASSIFIER I / I ~ 1 /
LU
N
EFFICIENCY ~ ~ /
[PERCENT) .~~'-~CLASSIFIER~ /
C3
/ UNOERSIZEf
,,, 0.5
I-
/
Z
,q
"r
~"
W
Z
0.2 _ j°v T
7,:'//
tl.
Z
0 0.1 - / FEE°
F-
U

h.
m 0.0'5

W
_>
/ SIMULATEO EFFECT OF ERROR
~J0.02 _ / IN THE DESIGN ASSUMPTION

L)
0.01 I I I I I I I
20 rio IOO 200 500 IOOO 2000
PARTICLE SIZE, MICRON
Fig. 16. S i m u l a t e d mill discharge and p r o d u c t size distribution for the c o m m e r c i a l mill at
the design t h r o u g h p u t and e x p e c t e d changes in t h e m for changes in classifier e f f i c i e n c y .

*It is interesting to n o t e that if o n e uses the mill d i a m e t e r - - p o w e r draft correlation (Bond,


1962):
P = 2.2 P BALLS" (L/D) D3"4M B * ( 3 - 2 - 3 M B * ) N * ( 1 - 0 . 1 / 2 9 - 1 ° N * )
#

to find the mill diameter w h i c h will yield 6 3 7 kW for the same operating c o n d i t i o n s the
value is f o u n d to be D = 11 ft.
24

(8) Evaluate the sensitivity of mill capacity to deviations from the design
assumptions. Once the mill size and power requirements have been determined
the effect of deviations from the assumed classifier performance, RTD model
and feed-size distribution as well as deviations from the measured kinetic
parameters can be readily determined by simulation. For the example being
discussed here, it was found that: (1) if the efficiency of commercial classifier
lies between 65% and 75% the 250% circulating load. production rate for the
mill chosen in step 7 would be between 95.7 and 104.5 TPH (expected changes
in the mill discharge and product-size distributions are shown in Fig. 16);
(2) if the actual size distribution of the fresh feed lies between - 6 mesh and
- 1 0 mesh (with the distribution modulus equal to unity) the production rate
would be between 97.1 and 105.0 TPH; and (3) if the measured breakage rate
functions are 10% too high or 10% too low, the production rate would lie between
91.0 and 110.0 TPH. Assuming the magnitude of possible deviations from the de-
sign assumptions has been realistically assessed, the appropriateness of the de-
sign can be evaluated in terms of the expected range of production rates.
Before leaving this example a few comments regarding the computer
program used for the computations presented here are warranted. Although
eq~ 8 and 9 may look formidable, the programming is quite straightforward.
The computations presented above were made with a single FORTRAN
program which is capable of both simulation and breakage rate function esti-
mation for batch, locked-cycle and continuous open- and closed-circuit grind-
ing. This program requires less than 40K storage locations. The computational
time required for a simulation is about 0.4 sec and the time required for break-
age rate function estimation is about 3.0 sec on a Univac 1108 computer.
A more general program has recently been developed (Herbst and
Rajamani, 1977)which is capable of simultaneously estimating breakage rate
and distribution functions from data obtained with an arbitrary feed size
distribution. This program should simplify the design procedure described
above by eliminating the need for performing single size fraction feed experi-
ments to determine breakage distribution functions.

DISCUSSION AND CONCLUSIONS

In this paper the potential usefulness of population balance models for con-
tinuous grinding mill design has been examinecL One class of these models,
lumped-parameter models, has been investigated in detail owing to the fact
that: (1) the underlying assumptions of these models frequently represent
reasonable approximations to physical reality; and (2) general closed-form
solutions to the model equations are available for steady-state grinding which
facilitate computer simulation and parameter estimation. The type of experi-
mental information required to apply these models to the design of con-
tinuous mills has been identified and the nature of the measurements and com-
putational procedures has been indicated.
25

By making use of the kinetic parameter scale-up relations:

(2) bi.i invariant,


the mechanics of designing a commercial-scale mill from laboratory-scale
batch data were illustrated. This example demonstrated that this design proce-
dure not only yields information on mill and drive size requirements b u t also
complete circuit-size distribution. Perhaps most importantly, it provides an op-
portunity to evaluate the effect of deviations from the design assumptions in
order that the designer can make a realistic appraisal of the design risk as-
sociated with a new installation.
The primary motivation for developing an alternative to the Bond approach
to mill design is the potential improvement in design accuracy achievable by
introducing the kinetics of breakage, material transport and size classification
behavior -- in an explicit manner -- into the design equations. The importance
of properly accounting for transport and size classification behavior is il-
lustrated computationally in Tables I and II. In Table I the production rate of
- 1 5 0 mesh limestone is calculated for an 11.5 X 13.2 ft. ball mill with
transport characteristics ranging from the RTD equivalent of plug flow to that
of a perfectly mixed mill and a classifier whose efficiency ranges from 100%
to 50%. In Table II the input power and mill diameter (LID = 1.15) required

TABLE I

Production rates in 11.5 ft. mill for various classifier efficiencies and residence time
distributions. Production rate of - 1 5 0 mesh material (TPH) at 250% circulating load

Production rate

C.E. = 100% C.E.= 70% C.E.= 50%

RTD equivalent t o :
Plug flow 128.1 109.8 82.8
Two perfect mixers
in series 119.8 100.0 72.9
One perfect mixer 107.0 88.5 65.0

Note: C.E. = classifier efficiency.

for a 100 TPH production rate at 250% circulating load has been determined
for each RTD-classifier efficiency combination. It is evident from these com-
putations that a mill designed with the assumptions of plug flow and perfect
classification would be completely inadequate if the commercial mill behaved
as a single perfect mixer and the actual efficiency of the classifier was, say, 50%.
The importance of accounting for the details of the kinetics of breakage in
the equations for mill design is illustrated by commercial-scale production rate
26

TABLE II

Input power and mill size required for 100 TPH production rate (at 250% C.L.) for various
classifier efficiencies and residence time distributions. Input power (kW)/mill diameter (ft.)

C.E.= 100% C.E.= 70% C.E.= 50%

input mill input mill input mill


power size power size power size
(kW) (ft.) (kW) (ft.) (kW) (ft.)

RTD equivalent to:


Plug flow 498 10.7 581 11.2 770 12.1
Two perfect mixers
in series 532 10.9 637 11.5 875 12.6
One perfect mixer 595 11.3 720 11.9 980 13.0

Note: C.E. = classifier efficiency.

calculations (Grandy et al., 1971) made for three different materials (an "ideal
Bond material" and naturally occurring quartzite and dolomite) having virtual-
ly the same Bond work index. For identical transport and classification be-
havior the calculated commercial-scale production rates for the three ~
materials varied by about 30%. These computations suggest that a single
parameter representation of material breakage characteristics, such as the
work index, is generally inadequate for accurate mill design.
It is interesting to note that a large amount of data collected in commercial-
scale sulfide grinding (Blaskett, 1969) and cement grinding (Smith, 1959) mills

I I i t ] T
F._I2
7" • SULFIDE ORE ,,
O CEMENT i~l/*" .
// /0
/// ...///

~8
8
,S6 /i~,6 """°
/ ,, " O
0

>
i

//~/// --- + 20% PREDICTION ERROR

of I I I I I t
0 2 4 6 8 I0 12
ESTIMATED ENERGY CONSUMPTION (BOND EQ.), KWH/T

Fig. 17. Depiction of the design risk associated with the Bond scale-up procedure.
27

suggests t h a t the design risk associated with the B o n d scale-up p r o c e d u r e is o n


the o r d e r o f +20% (see Fig. 17). In l o o k i n g b a c k over the c o m p u t a t i o n a l re~
sults r e p o r t e d in the last t w o paragraphs it will be observed t h a t this risk is
close t o t h a t w h i c h o n e w o u l d p r e d i c t if t h e details o f breakage behavior,
material t r a n s p o r t and size classification are n o t p r o p e r l y a c c o u n t e d f o r in the
design.
In p r e p a r i n g this p a p e r the i n t e n t i o n o f t h e a u t h o r s has n o t b e e n t o u n d u l y
criticize c u r r e n t mill design m e t h o d s . In fact, these m e t h o d s have served the
mineral and c e m e n t industries very well. Instead, this p a p e r is m e a n t t o p o i n t
o u t t h a t r e c e n t l y t h e r e has b e e n a great deal o f research c o n c e r n e d with the
d e v e l o p m e n t o f detailed m o d e l s f o r grinding circuits. These m o d e l s are
capable o f providing m u c h m o r e accurate descriptions o f grinding b e h a v i o r
t h a n t h e classical energy-size r e d u c t i o n relationships and t h e r e f o r e provide a
promising alternative as a basis for mill design. It is the o p i n i o n of the a u t h o r s
t h a t if mill designers begin t o t h i n k in t e r m s o f breakage kinetics, material
t r a n s p o r t and classifier p e r f o r m a n c e and initiate the e x p e r i m e n t a l p r o g r a m s
r e q u i r e d f o r e x t e n d i n g these c o n c e p t s t o larger scale mills, t r u l y significant ad-
vances in mill design can be e x p e c t e d in the near f u t u r e .

ACKNOWLEDGEMENTS

T h e a u t h o r s wish t o t h a n k t h e U.S. B u r e a u o f Mines and the National


Science F o u n d a t i o n f o r t h e s u p p o r t o f a p a r t o f the w o r k p r e s e n t e d in this
paper.

NOMENCLATURE

RTD residence time distribution


t,o time and dimensionless time, respectively
T mean residence for material in a mill
E(t),E(o) exit age distribution and dimensionless exit age distribution,
respectively
Xi particle size
n number of size intervals represented in model equations
mi(t), m ( t ) mass fraction of material in the size interval xi to xi+ 1 and the vector
of all n such mass fractions (n × 1)
mBATCH(t) vector of mass fractions mi(t), i = 1, n, in batch grinding (n × 1)
roMp, mMF, mp, mF vectors of mass fractions in the mill discharge, mill feed, circuit
product and circuit feed at steady state, respectively
MMI~ MMF, Mp, M F mass flow rates of material in the mill discharge, mill feed, circuit
product and circuit feed at steady state, respectively
H mass hold-up of material in the mill
bij, Bij individual size discretized breakage distribution function and cumula-

tive breakage distribution function ( ~ ) b ~ ] ,respectively


4=1
hi, k E size discretized breakage rate function for material in the i-th size
interval and corresponding specific breakage rate function (kill~P),
respectively
28

B matrix of individual breakage distribution functions (lower triangular,


n×n)
K matrix of size discretized breakage rate functions (diagonal, n × n)
T matrix of eigenvectors of (I - B ) K having elements Tij (lower tri-
angular, n × n)
Jc(r) modal matrix for continuous system having elements Jci j (diagonal,
n×n)
C classifier matrix (diagonal, n × n ); with elements Ci.i give the fraction of
material in each size interval of the classifier feed which reports to the
classifier oversize
/, e T identity matrix (diagonal, n × n) and transposed unit vector e T =
[1 1 1 . . . 1] (lX n)
Fi production "rate constant" for cumulative fraction finer than size xi
P mill power (net)
L,D tumbling mill length and diameter, respectively
N* dimensionless mill speed -- fraction of critical speed of rotation of the
mill
MB* dimensionless ball load -- fraction of the mill volume occupied by the
static ball mass
Mp* dimensionless particle load -- fraction of the static ball mass occupied
by particles
Q* dimensionless variables associated with ball size and lifter geometry
6 empirical correction to the diameter exponent in eq. 12 for mill power
~),, ¢2, ~3 mill operating variable dependent functions in the power equation,
mass hold-up equation and specific breakage rate function equation,
respectively
C.L. circulating load

REFERENCES

Anonymous, a (date unknown). Grinding Mills for the Rock Products, Cement and
Chemical and Mining Industries. Allis-Chalmers Bulletin No. 07B6718A.
Anonymous, b (date unknown). Grinding Mills. Denver Equipment C o m p a n y Bulletin
No. B2-B34.
Anonymous, 1971. Nordberg Grinding Mills: Ball R o d and Pebbles. Nordberg -- Division of
Rex Chainbelt, Inc.
Austin, L.G., 1971--72. A review introduction to the mathematical description of grinding
as a rate process. Powder Technok, 5: 1--17.
Austin, L.G., 1973. Understanding ball mill sizing. Ind. Eng. Chem. Process Des. Dev.,
12(2): 121--129.
Austin, L.G., Klimpel, R.R. and Beattie, A.N., 1967. Solutions of equations of grinding. In:
Zerkleinern. DECHEMA-Monogr., 57(1): 281--313.
Austin, L.G., Luckie, P.T. and Ateya, B.E., 1971. Residence time distributions in mills.
Cem. Concr. Res., 1(3): 241--256.
Austin, L.G., Luckie, P.T. and Klimpel, R.R., 1972. Solutions of the batch grinding
equations leading to Rosin-Rammler distributions. Trans. Am. Inst. Min. Metall. Pet.
Eng., 252: 87--94.
Blaskett, K.S., 1969. Estimation of the power consumption in grinding mills. Proe. 9th
Commonw. Min. Metall. Congr., 3: 631--649.
Bond, F.C., 1952. The third theory of comminution. Trans. Am. Inst. Min. Metall. Pet. Eng.,
193 : 484--494.
Bond, F.C., 1960. Confirmation of the third theory. Trans. Am. Inst. Min. Metall. Pet. Eng.,
217: 139--153.
29

Bond, F.C., 1962. Crushing and Grinding Calculations. Allis-Chalmers Publications,


Milwaukee, Wisconsin, 16 pp.
Fuerstenau, D.W., 1980. The influence of mill diameter on the parameters of the batch
grinding equations. In prep.
Gardner, R.P. and Austin, L.G., 1962. A chemical engineering treatment of batch grinding,
Part I and Part II. In: D. Behrens and H. Rumpf (Editors), Symposium Zerkleinern, the
First European Symposium on Comminution. Verlag Chemic, Weinheim, pp. 217--248.
Grandy, G.A. and Fuerstenau, D.W., 1970. Simulation of nonlinear grinding systems: rod-
mill grinding. Trans. Am. Inst. Min. Metall. Pet. Eng., 247: 348--354.
Grandy, G.A., Gumtz, G.D., Herbst, J.A., Mika, T.S. and Fuerstenau, D.W., 1969.
Computer techniques in the analysis of laboratory grinding tests. In: A. Weiss (Editor),
A Decade of Digital Computing in the Mineral Industry. Am. Inst. Min. Metall. Pet. Eng.,
New York, N.Y., pp. 765--788.
Grandy, G.A., Herbst, J.A., Mika, T.S. and Fuerstenau, D.W., 1971. A Critical Evaluation
of the Use of Bond Grindability Tests for Mill Design. 100th Annu. Am. Inst. Min.
Metall. Pet. Eng. Meet., New York, N.Y.
Gumtz, G.D. and Fuerstenau, D.W., 1970. Simulation of locked-cycle grinding. Trans. Am.
Inst. Min. Metall. Pet. Eng., 247: 330--335.
Herbst, J.A., 1971. Batch Ball Mill Simulation: An Approach for Wet Systems. Diss., Univ.
Calif., Berkeley, Calif., Berkeley, Calif., 173 pp.
Herbst, J.A. and Fuerstenau, D.W., 1968. The zero order production of fine sizes in
comminution and its implications in simulation. Trans. Am. Inst. Min. Metall. Pet. Eng.,
241: 538--549.
Herbst, J.A. and Fuerstenau, D.W., 1972. Influence of mill speed and ball loading on the
parameters of the batch grinding equation. Trans. Am. Inst. Min. Metall. Pet. Eng., 252:
169--176.
Herbst, J.A. and Fuerstenau, D.W., 1973. Mathematical simulation of dry ball milling using
specific power information. Trans. Am. Inst. Min. Metall. Pet. Eng., 254: 343--348.
Herbst, J.A. and Mika, T.S., 1970. Mathematical simulation of tumbling mill grinding: an
improved method. Rudy, 18(3/4): 70--75.
Herbst, J.A. and Mika, T.S., 1973. Linearization of tumbling mill models involving non-
linear breakage phenomena. 11th Int. Syrup. Comput. Appl. Miner. Ind., Tucson, Ariz.
Herbst, J.A. and Rajamani, K., 1980. Simultaneous estimation of selection and breakage
functions from batch and continuous grinding data. In prep.
Herbst, J.A., Grandy, G.A. and Mika, T.S., 1971a. On the development and use of lumped
parameter models for continuous open- and closed-circuit grinding systems. Inst. Min.
Metall. Trans., 80: C193--198.
Herbst, J.A., Grandy, G.A., Mika, T.S. and Fuerstenau, D.W., 1971b. An approach to the
estimation of the parameters of lumped parameter grinding models from on-line
measurements. In: Zerkleinern. DECHEMA-Monogr., 69: 475--514.
Herbst, J.A., Mika, T.S. and Rajamani, A., 1976. A comparison of distributed and lumped
parameter models for open-circuit grinding mills. In: Zerkleinern. DECHEMA-Monogr.,
79: 467--487.
Horst, W.E. and Freeh, E.J., 1970. Mathematical modeling applied to analysis and control
of grinding circuits: part I, development of comminution models. Pap. presented 99th
Annu. Am. Inst. Min. Metall. Pet. Eng. Meet., Denver, Colo.
Karra, V.K., 1976. Continuous Grinding in Grate-Discharge Ball Mills: Analysis, Simulation
and Scale-up. Thesis, Univ. Calif., Berkeley, Calif., 137 pp.
Kelsall, D.F., 1965. A study of breakage in a small continuous wet ball mill. N. Arbiter
(Editor), Proc. VII Int. Miner. Process. Congr., New York, Gordon and Breach,
New York, pp. 33--42.
Kelsall, D.F. and Reid, K.J., 1965. The derivation of a mathematical model for breakage in
a small, continuous, wet, ball mill. Paper 4.2, Joint Am. Inst. Chem. Eng. Meet., London,
England, pp. 14--20.
30

Kelsall, D.F. and Reid, K.J., 1969. Some effects of a change in environment size
distribution on the grinding behavior in a continuous wet ball mill. Inst. Min. Metall.
Trans., 78: C198--205.
Kelsall, D.F., Stewart, P.S.B. and Reid, K.J., 1968. Confirmation of a dynamic model of
closed-circuit grinding with a wet ball mill. Inst. Min. Metall. Trans., 77: C120--127.
Kelsall, D.F., Reid, K.J. and Restarick, C.J., 1968--69. Continuous grinding in a small wet
ball mill, part II: a study of the influence of hold-up weight. Powder Technol., 2:
162--168.
Kelsall, D.F., Reid, K.J. and Restarick, C.J., 1969--70. Continuous grinding in a small wet
ball mill, part III: a study of distribution of residence time. Powder Technol., 3:
170--178.
Klimpel, R.R. and Austin, L.G., 1970. Determination of selection-for-breakage functions in
the batch grinding equations by nonlinear optimization. Ind. Eng. Chem. Fundam., 9:
230--237.
Lynch, A.J., Whiten, W.J. and Draper, N., 1967. Developing the optimum performance of a
multi-stage grinding circuit. Inst. Min. Metall. Trans., 76: C169--182.
Malghan, S.G., 1976. The Scale-up of Ball Mills Using Population Balance Models. Diss.,
Univ. Calif., Berkeley, Calif., 271 pp.
Malghan, S.G. and Fuerstenau, D.W., 1976. The scale-up of ball mills using population
balance models and specific power input. In: Zerkleinern. DECHEMA-Monogr., 79(II)~
No. 1586: 613--630.
Mika, T.S., 1970. Population Balance Models of a Continuous Grinding Mill as a
Distributed Process. Diss., Univ. Calif., Berkeley, Calif., 424 pp.
Mika, T.S. and Fuerstenau, D.W., 1971. The transient behavior of a distributed parameter
model of a grinding mill. In: Zerkleinern. DECHEMA-Monogr., 69: 389--433.
Mika, T.S., Berlioz, L.M. and Fuerstenau, D.W., 1967. An approach to the kinetics of dry
batch ball milling. In: Zerkleinerr~ DECHEMA-Monogr., 57(I): 205--240.
Molerus, O., 1966. Uber die Axialvermischung bei Transportprozessen in Kontinuierlich
betriebenen Apparaturen. Chem.-Ing.-Tech., 3 8 : 1 3 7 - - 1 4 5 (in German).
Molerus, O., 1967, Stochastisches Model der Gleichgewichtssichtung. Chem.-Ing.-Tech., 39:
792--796 (in German).
Molerus, O. and Paulsen, H., 1970. Axialdispersion des Mahlgutes und Energieausnutzung
bei Durchlaufmahlung in der Kugelmuhle. Chem.-Ing.-Tech., 4 2 : 2 7 0 - - 2 7 7 (in German).
Mori, Y., Jimbo, G. and Yamazaki, M., 1964. Residence time distribution and mixing
characteristics of powders in open-circuit ball mill. Kagaku-Kogaku, 28: 204; Kagaku-
Kogaku (Abr. Ed. Engl.), 2: 173--178.
Mori, Y., Jimbo, G. and Yamazaki, M., 1967. Flow characteristics of continuous ball and
vibration mill: mixing, size distribution, dynamic response of flow rate and applications.
In: Zerkleinern. DECHEMA-Monogr., 57(II): 605--632.
Mular, A.L., 1970. The determination of selection elements and lumped parameters for
grinding mills. Can. Inst. Min. Metall., Bull., July, pp. 821--826.
Olsen, T.O. and Krogh, S.R., 1972. Mathematical model of grinding at different conditions
in ball mills. Trans. Am. Inst. Min. Metall. Pet. Eng., 252: 452--457.
Reid, K.J., 1965. A solution to the batch grinding equation. Chem. Eng. Sci., 20: 953--963.
Rose, H.E. and Sullivan, R.M.E., 1958. A Treatise on the Internal Mechanics of Ball, Tube
and Rod Mills. Chemical Publishing Company, New York, N.Y., 258 pp.
Rowland, C.A., Jr., 1976. On line measurements of the utilization of power delivered to
grinding mills. Instrum. Min. Metall. Ind., 4: 80--89.
SchSnert, K., 1971a. Mathematische Simulation yon Zerkleinerungsprozessen. Teil 1. Das
Allgemeine Modell und der Stationare Sonderfall. Chem.-Ing.-Tech., 43(6): 361--367.
SchSnert, K., 1971b. Eine Mathematische Beschreibung des Kombinierten Prozesses
Konvection-Diffusion-Zerkleinerung. In: Zerkleinern. DECHEMA-Monogr., 69:
361--387.
31

Sedlatschek, K. and Bass, L., 1953. Contribution to the theory of milling processes.
Powder Metall. Bull., 6: 148--153.
Siddique, M., 1977. A Kinetic Approach to Ball Mill Scale-up for Dry and Wet Systems.
Thesis, Univ. Utah, Salt Lake City, Utah, 106 pp.
Smith, R., 1959. 1956 Power Survey. Portland Cem. Assoc. Rep., MP-73.
Stewart, P.S.B. and Restarick, C.J., 1967. Dynamic flow characteristics of a small, spiral
classifier. Inst. Min. and Metall. Trans., 77: C225--230.
Tanaka, T., 1964. Axial Flow in Rotating Cylinders. Miner. Process., 5: 24--26.
Yang, D., Mempel, G. and Fuerstenau, D.W., 1967. A laboratory mill for batch grinding
experimentation. Trans. Am. Inst. Min. Metall. Pet. Eng., 238: 273--275.

S-ar putea să vă placă și