International Journal of Mineral Processing, 8 (1981) 61--78 61

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

MODELLING MINERAL SIZE REDUCTION IN THE CLOSED-CIRCUIT

BALL MILL AT THE PINE POINT MINES CONCENTRATOR

J.A. FINCH and J. RAMIREZ-CASTRO

Department of Mining and Metallurgical Engineering, McGill University, Montreal
(Canada)
(Received January 10, 1980; revised and accepted August 29, 1980)

ABSTRACT

Finch, J.A. and Ramirez-Castro, J., 1981. Modelling mineral size reduction in the closed-
circuit ball mill at the Pine Point Mines concentrator. Int. J. Miner. Process., 8:
61--78.

Using plug flow material transport and a cumulative-basis rate-of-breakage parameter,

overall size reduction through the closed-circuit ball mills at the Pine Point and Gibraltar
concentrators was simulated over a wide range of operating conditions. The rate-of-breakage
parameter was related to particle size by a power law, the exponent (n) being: Pine Point,
n - 1.043 ± 0.026, and Gibraltar, n = 0.747 ± 0.020. The success of this approach probably
stems fron~ the high (> 1.5) circulating load ratios encountered.
By analogy individual mineral size reduction at Pine Point was examined. A similar
rate-of-breakage parameter versus size relationship was found. Pyrite was the hardest
mineral, but fine galena was equally resistant. However, the approximation that mineral
and overall rates of breakage were the same gave an adequate fit to the mineral size re-
duction. This was emphasized by combining with a cyclone model to simulate cyclone
overflow mineral size distribution. A more accurate cyclone model is shown to be more
important in simulating mineral deportment at Pine Point.
Complementary laboratory batch grinding tests were conducted on rod mill discharge
and ball mill feed samples. Sufficient agreement with the first-order hypothesis was ob-
served to analyse the rate-of-breakage parameter. The kinetics was similar for both
samples and in turn similar to the plant-derived kinetics in terms of relative mineral rates-
of-breakage and the relationship of the rate-of-breakage parameter with particle size.

INTRODUCTION

M o d e l l i n g of c l a s s i f i c a t i o n a n d c o m m i n u t i o n u n i t s f o r c i r c u i t e v a l u a t i o n ,
c o n t r o l a n d s c a l e - u p s t u d i e s has s e e n a m a j o r e f f o r t in t h e l a s t t w e n t y years.
U s u a l l y t h e average, o r overall, p e r f o r m a n c e is m o n i t o r e d . F o r h o m o g e n e o u s
feeds (i.e. n o p a r t i c l e t o p a r t i c l e v a r i a t i o n i n p r o p e r t i e s s u c h as d e n s i t y or
h a r d n e s s ) this is a p p r o p r i a t e . H o w e v e r , f o r h e t e r o g e n e o u s feeds, s u c h as
P b / Z n ores, i n d i v i d u a l m i n e r a l b e h a v i o u r is m o r e a p p r o p r i a t e .
M o d e l l i n g i n d i v i d u a l m i n e r a l b e h a v i o u r in c l o s e d g r i n d i n g c i r c u i t s has
b e e n d e s c r i b e d b y C a m e r o n e t al. ( 1 9 7 1 ) , L y n c h ( 1 9 7 7 a ) a n d F i n c h a n d

0301-7516/81/0000--0000/$ 02.50 © 1981 Elsevier Scientific Publishing Company

62

Matwijenko (1977). Ramirez-Castro and Finch (1980) recently reported a

simulation study of the Pine Point Mines grinding circuit (Fig. 1) which
examined the reduction in lead sliming upon incorporating a unit cell on
ball mill discharge. Cumulative rate-of-breakage parameters for each mineral
were determined from laboratory batch grinding tests and plug flow
material transport was assumed for the continuous mill. Fitting was ade-
quate for the purpose of evaluating the competing circuits.

COF

RMD CF

B A L L MILL
3.5x4.9 m

Fig. 1. The No. 3 grinding circuit at the Pine Point Mines concentrator. (Three out of
five cyclones were in operation at time of sampling; all streams identified were sampled.)

The cumulative-basis rate-of-breakage approach eliminates the need for a

breakage parameter. In the case of Cameron et al. (1971) and Lynch (1977a)
for example, a mineral breakage function had to be assumed. The dis-
advantage in the m e t h o d of Ramirez-Castro and Finch (1980) is the use of
laboratory-derived parameters, implying the need for scale-up. This paper de-
scribes a model of mineral size reduction at Pine Point derived solely from
plant data.
The key is a simplified material transport model. Randolph and Ranjan
(1977) and Smith and Guerin (1979) show the fully mixed approximation
is adequate for simulating size reduction across closed-circuit ball mills.
Furuya et al. (1971) show, assuming an ideal classifier, that at high (>1.5)
circulating loads, the plug flow approximation is adequate for simulating
size reduction. Overall size reduction at the Pine Point and Gibraltar con-
centrators is analyzed using the plug flow material transport approximation
which is extended to individual mineral size reduction in the case of Pine
Point. Some complementary laboratory batch grinding studies are also
described.

BALL MILL MODEL

For a cumulative-basis rate-of-breakage parameter (K) with a plug flow

transport model:
 63

R ( x , r ) = R ( x , O ) exp ( - K r ) (1)
where R (x, T), R (x,0) is the cumulative percent retained above size x at
time r and 0, respectively, where r is the mean residence time given by:
VBH volumetric ball mill hold-up of slurry
v= - (2)
VFR volumetric feed rate of slurry to ball mill
By analogy with the relationship between the discrete size rate-of-breakage
parameter and particle size, it is assumed that:

K = Cx n (3)
As written, the equations apply to the overall size reduction. The sub-
script m will be used to denote an individual mineral, so equations
analogous to eqs. I and 3 can be written:

R m ( x , r ) = Rm(x,0) exp (-Kmr) (4)

and

Km = Cmxnm (5)

K can be estimated from:

K= --r In ~x,~!R (6)

Hence r and the ball mill feed and discharge size distribution must be
measured. Failing a direct measure of r, it can be estimated from VBH (as-
sumed constant) and VFR determined from a mass balance.

EXPERIMENTAL PART

Plant tests

Pine Point, N . W . T . , Canada

This concentrator treats a galena/sphalerite ore with principally pyrite
and calcite/dolomite gangue. Grinding is in three parallel circuits consisting
of a rod mill followed by a ball mill in reverse classification with a cyclone.
Figure I shows the No. 3 circuit. The cyclopac consists of 5 cyclones of
which 3 were in use. The circuit was tested at 154.2 and 190.3 mtph fresh
feed rates and each stream from rod mill discharge on was cut every 15 min
for 2 hours to provide a composite sample. Pulp density was determined by
direct weighing and the sample was screened from 28 to 400 mesh in a ~/2
series using the wet/dry .technique. Each size fraction was assayed for Pb,
Zn and Fe which were converted to mineral assays with calcite/dolomite
determined by difference.
64

A mass balance around the cyclone was conducted which gave a better
than 0.95 correlation coefficient for the circulating load estimate (Finch
and Matwijenko, 1977). The circuit was then balanced and data adjusted
using the scheme described by L y n c h (1977b) extended to adjust mineral as-
says. Two additional constraint equations are required to balance mineral
flows, namely: (1) mass of mineral entering circuit equals that leaving; and
(2) mass of mineral entering ball mill equals that leaving, i.e. Rm(0,T ) must
equal Rm (0,0). The mineral mass flows must be balanced across the ball
mill, for, if R m (0,r) 4= Rm(0,0), the calculated rate-of-breakage parameter,
Kin, becomes distorted especially as x + 0. Table I gives some data on the
balanced circuit.

TABLE I

Data on balanced circuit

mtph Fresh feed (%} Ball mill feed (%) Circulating % w/w solids in r s.g. ball
load ratio ball mill feed mill feed
PbS ZnS FeS2 PbS ZnS FeS~ (rain.)

154.2 1.9 8.7 14.1 6.8 12.6 32.9 1.63 70.1 3.11 3.63
190.3 1.8 6.6 6.3 9.3 12.3 27.1 2.00 73.6 2.18 3.50

mtph % - 2 0 0 mesh PbS % -200 mesh overall

RMD CF COF CUF BMD RMD CF COF CUF BMD

154.2 34.3 47.5 94.2 39.4 49.9 34.68 36.68 62.90 20.57 37.89
190.3 29.8 39.6 93.9 34.3 40.6 28.77 33.18 52.80 23.39 35.39

The volumetric ball mill hold-up of slurry was calculated from mill ge-
ometry. The mill is 3.5 m by 4.9 m. Assuming ball filling to the lip of the
overflow, bulk ball filling was estimated to be 18.35 m 3 ( - 40% mill
volume). Assuming the balls pack to give 38% voids (Taggart, 1945) and
slurry filling is 120% of void volume, the mill hold-up volume is 8.37 m 3 or
18% of mill volume. From the volumetric flow rate of cyclone underflow
slurry, r can be calculated; r is given in Table I.

Gibraltar, B.C., Canada

• The ore processed is chalcopyrite/molybdenite with a quartz/feldspar
gangue. The grinding circuit consists of three parallel, identical lines with
the same layout as at Pine Point. The difference is the equipment size, the
ball mills are 4.1 m > 6.1 m and there are six 76 cm cyclones. Routine sur-
veys are conducted giving pulp density and size distribution. Twenty-seven
data sets from June 1972 to August 1974 were made available. Mass balanc-
ing was conducted and 26 sets met the requirement of a correlation coeffi-
 65

cient greater than 0.95 on the circulating load estimate. The size distribu-
tion data was then adjusted (Lynch, 1977b).
A wide range of operating conditions was represented by these data sets.
Table II gives data for the lowest (489 m t ph ) to highest (674 m t p h ) fresh
feed rate. The range in circulating load was 1.75 to 6.45 and in % -28 mesh
in ball mill feed, 68.6 to 78.7. Percent solids of ball mill feed remained ap-
p r o x i m a t e l y co n s t ant at 75% w/w.
Assuming a hold-up volume of 18%, as at Pine Point, the VBH is 14.5 m 3.
The specific gravity of ball mill feed was assumed to be 3 and r was
calculated. The range in r was 0.37 to 1.32 min.

TABLE II

Operating conditions at the low (489 mtph) and high (674 mtph) tonnage Gibraltar data

mtph Circulating %-28 mesh % w/w solids in r

load ratio ball mill feed (min.)
RMD CUF BMD COF

489 3.19 69.44 73.36 82.93 99.86 73.7 0.81

674 2.66 59.93 52.97 66.78 96.48 76.4 0.76

Laboratory tests

During the sampling at 154.2 m t p h at Pine Point, sufficient ball mill feed
and rod mill discharge was cut (-= 30 kg) for l aborat ory batch grinding tests.
Some of the properties of these t w o materials are given in Table III.
A 16.5 cm × 20.3 cm Abb~ porcelain grinding jar rot at ed at 68% of its
critical speed was used. Grinding conditions were based on a void volume of
12% of mill volume (Le Houillier and Marchand, 1977) and slurry filling of
120% o f void volume (cf.assumption for plant mill). Details are given in
Table IV. The r od mill discharge (RMD) and ball mill feed (CUF) samples
had a specific gravity of 3.08 and 3.53, respectively, giving required sample
masses of 1000 g (RMD) and 1150 g (CUF). These samples were obtained
on a spinning riffler. Grinding times were 1 to 12 rain. with a fresh sample
for each time. The products were w e t / d r y screened from 28 to 400 mesh in

TABLE III

Comparison of ball mill feed (CUF) and rod mill discharge (RMD) samples used in
laboratory tests
% +65 mesh % -200 mesh % PbS % FeS2 % ZnS specific gravity

CUF 18.6 22.9 6.8 32.8 12.6 3.53

RMD 40.5 31.2 2.0 14.2 8.6 3.08
66

TABLE IV

Laboratory ball mill grinding conditions

Balls charge Void volume Slurry

Size (cm) No. % mill (of total vol.) % mill cm 3 % voids cm 3 % solids

2.54 196 31.5 12.0 641.0 14.4 769.0 42.1

a x / 2 series a n d each size f r a c t i o n was assayed f o r Pb, Zn and Fe.

D i s a p p e a r a n c e p l o t s were c o n s t r u c t e d . A t sizes 65 m e s h a n d c o a r s e r devia
tions f r o m 1st o r d e r were a p p a r e n t . A p - o r d e r fitting r o u t i n e (Harris, 1 9 7 1 )
was s u b s t i t u t e d f o r eq. 1 where:

R ( x , t ) = R ( x , O ) e x p [ - b t p] (7)

RESULTS

Plant tests

Overall size reduction

Figure 2 shows K vs x m e a s u r e d at Pine Point. A g r e e m e n t b e t w e e n the
t w o t o n n a g e trials is v e r y close. A s s u m i n g the p o w e r relationship:
K ( x ) = 0 . 0 0 0 9 0 x 1"°43 (K, m i n - 1 ; x, p m ) (8)

with 95% c o n f i d e n c e intervals on the p a r a m e t e r s o f ( H i m m e l b l a u , 1970):

0 . 0 0 0 7 9 ~< C < 0 . 0 0 1 0 2 (C, m i n -1 "pm - n )

and
1.017 < n < 1 . 0 6 9
Figure 3 s h o w s K vs x f o r the 26 G i b r a l t a r d a t a sets. This large b o d y of
d a t a r e d u c e s to K e s t i m a t e s with a small c o n f i d e n c e interval a b o u t the as-
s u m e d p o w e r relationship. Regression gives:

K ( x ) = 0 . 0 0 3 8 9 x °'v47 (x ~< 850 p m ) (9)

with 95% c o n f i d e n c e intervals on the p a r a m e t e r s of:
0 . 0 0 3 4 8 ~< C < 0 . 0 0 4 3 0
and
0 . 7 2 7 ~< n < 0 . 7 6 7
 67

I I I I
/

•S .5 o

mtph "g~rain /
d a154.2 3.11 ~/

° / Pine Point Data

=_
.1 /0 Krx)= C'X n
/ C =0"0009-+'00012

/° n=1.043 -+.026

3 /
E

.02 i I i i
20 50 100 200 500 1000

Particle Size, x, microns

Fig. 2. Overall cumulative-basis rate-of-breakage p a r a m e t e r , K, as a f u n c t i o n o f particle
size, x, for Pine P o i n t No. 3 ball mill. ( m t p h refers t o fresh feed rate t o circuit, r t o
m e a n ball mill residence time.)

2 I I t I I I I I [

E 1

cc ~ G i b r a l t a r Data
.~ .2
~ K(x): Cx
C= .00389 -+.00041
.1 n = .747 -+.020

,~E ~ 95pct confidence of single

¢J ~. measurement of K(x) at x
i I i i i I I i i

75 105 149 212 297 425 600 850 2000

Particle Size, x, microns

Fig. 3. Overall cumulative-basis rate-of-breakage p a r a m e t e r , K, as a f u n c t i o n o f particle
size, x, f o r Gibraltar ball mills.
68

The coarsest size (x = 2000 pm or 9 mesh) shows the largest confidence

interval on K and significant deviation f r om the relationship. By com-
parison with eq. 9, K (x = 2000 pm) is lower than expected.
Figure 4 shows the measured vs predicted % -200 mesh in the BMD for
Gibraltar using eq. 9 substituted into eq. 1 and knowing the % +200 mesh
in the ball mill feed and r. The average deviation (measured-simulated) is
0.30% (absolute).

I F I I I

24 •

c~

20 •

16

16 20 24

Measured - 2 0 0 mesh, Pct.

Fig. 4. Measured vs simulated % -200 mesh in ball mill discharge for Gibraltar data.

Mineral size reduction

Figure 5 shows Km vs x for the Pine Point tests with the overall K vs x
(dashed line) for reference. T he average Km for the two tonnage trials is
given as no significant difference was evident. All the minerals e x c e p t galena
show a roughly parallel K m vs x relationship which is parallel to K vs x.
Pyrite is the most resistant to size reduction (KFeS2 is consistently the
lowest). Galena at the coarsest sizes exhibits a high rate-of-breakage but be-
comes comparable to pyrite at the finer sizes.
As an ap p r o x i m at i on it can be assumed that there is no real difference
between K m for each mineral and K overall. The consequences of this as-
sumption are examined in Fig. 6a, b. The ball mill discharge size distribu-
tion of mineral is simulated by substituting eq. 8 into eq. 1, knowing the
ball mill feed mineral size distribution and ~. The simulated p r o d u c t size
distribution agrees well with the measured, here illustrated for galena and
pyrite.

Laboratory data

Figure 7a, b shows the disappearance plots for ball mill feed overall and
galena. Note the deviation from first order at the coarser sizes. E x c e p t for
galena, at sizes less than 65 mesh p was generally in the range 0.9 < p
 69

i I r I /
?,
o /

.S o PbS o //A

o CaI-Dol /
°/ / v
~; --- Overall

oJ

//8
o~ .1 I/ v
CC /o / / o ~ ZnS

// ~ FeS 2

t//
r~ Q

¢J
.01 I I I
20 50 100 200 500 1000

Particle Size, x, microns

Fig. 5. Mineral cumulative-basis rate-of-breakage parameter, Kin, as a function of particle

size, x, for Pine Point case. (The average K m for the two tonnage trials is given and the
K vs x relationship, the dashed line, for reference.)
' L L ' ' ' i , , , i i , , J i
Q b
22 • 22

11] 2 mtph ~ 18

u. " 1C

i 1 i i i i i ~ i 1 i i L ~ i
270/400 150/200 65/100 35/4B 270/400 1501200 551100 35148
Particle Size ,mesh Particle Size, mesh
Fig. 6.a. Simulated (full line) and measured (symbol) mass frequency size distribution of
PbS and FeS 2 in the ball mill discharge at 154.2 mtph fresh feed rate. Assumption is
K m = K. b. As (a) but for 190.3 mtph fresh feed rate case.

1.1 w h i c h w a s t a k e n as first order. T h e average r e l a t i v e 95% c o n f i d e n c e

interval o n K a n d K m w i t h i n this range o f p w a s ~ 3 % . F i g u r e 8 s h o w s K
vs x f o r t h e r o d m i l l d i s c h a r g e a n d ball m i l l f e e d s a m p l e s and F i g u r e 9
s h o w s K m vs x f o r s p h a l e r i t e a n d p y r i t e w i t h t h e overall i n c l u d e d ( d a s h e d
l i n e ) f o r r e f e r e n c e . O m i t t e d p o i n t s are f o r c o n d i t i o n s w h e r e p lies o u t s i d e
0 . 9 t o 1.1. ( N o t e t h e 95% c o n f i d e n c e i n t e r v a l o n K a n d K m is a p p r o x -
imately the size of the symbol.)
70

100 ~ I I I I I 100 J I I I I I

50
I ~ ° ~

•
° ~ °

+200 Mesh
p= 1.03
°
50 " ~ O ~ o

o ~......~..

b •
b= . 0 7 - +200
p=0.98
Mesh

+ 6 5 Mesh
\ •
o WRA L L ~:
5
\
+65 Mesh
"

p = 1.09
o,. • b= .27 o,, ~ p= 1 . 2 2

• ~ \

E +28 Mesh +28 Mesh GA L E N A

tj p= .78 • ~ ~ u p= 0 . 7 4
b= .71 ~ ~ , 9 2
.5 • ,5 ~
• \
o\
.2 I I I I I , I .2 I I I h I I
2 4 6 8 10 12 2 4 6 8 10 12

T i m e , t, Min. T i m e , t, Min.
Fig. 7.a. Disappearance plots for +28, +65 and +200 mesh overall using ball mill feed
(CUF) sample, b. Disappearance plot for +28, +65 and +200 mesh galena using ball mill
feed (CUF) sample.

1
i j i I

"7

E .5 o
-,- o Rod Mill Discharge o

¢ Ball Mill Feed o o

d o

.2

95 pct confidence intervals

•.~ .05 e are approximately size of symbol

E e
¢j

.02 i r ~ i
20 50 100 200 500 1000

Particle S i z e , x, m i c r o n s

Fig. 8. Overall cumulative-basis rate-of-breakage parameter, K, from laboratory tests on

rod mill discharge and ball mill feed samples as a function of particle size, x.
 7]

1 __
I I l [ l

"7, Rod Mill Discharge

E .5 [] Sphalerite o o
• Pyrite o ~+
-'~ Ii I

Ball Mill Feed ," • •

o Sphalerite , •
.2 • Pyrite
]
m o," t
Q.) /0

.1
/
,/ •

Overall K(x), average

/x ~ o f K in F i g u r e 8

.~ .05 ]"'+ •
0 f •
., o

.02 L t £ I ;
10 20 50 100 200 500 1000

Particle S i z e , x, microns
Fig. 9. Mineral cumulative-basis rate-of-breakage parameter, Km, from laboratory tests
on rod mill discharge and ball mill feed samples as a function of particle size, x.

The trends are similar to those observed for the continuous mill, the
mineral rate-of-breakage versus x lying parallel to the overall and pyrite
being the hardest c o m p o n e n t . T he difference between rod mill discharge
and ball mill feed samples appears small. Although galena is difficult to
quantify with a single parameter, comparing the b and p parameters with
the overall (Fig. 7a, b) shows galena is similar to the overall.

SIMULATING MINERAL BEHAVIOUR IN A CLOSED GRINDING CIRCUIT

From the evidence presented, mineral size reduction through the closed-
circuit ball mill at Pine Point can be a ppr oxi m at ed by a plug flow material
transport model with an overall cumulative rate-of-breakage parameter.
Combining this with the cyclone modelling scheme proposed by L ynch
(1977a) and Finch and Matwijenko (1977) a simulator describing mineral
behaviour can be constructed. The two unit models, derived from data at
the 154.2 m t p h test, are:
Ball Mill:
R m ( X , T ) = R m ( x , 0 ) exp (-0.0009xl'°4T) (10)

Cyclone:
Ym = (1 - 0 . 3 1 ) ( 1 - exp [-0.693(~/Rs0m ) ~ " + ] } + 0.31 (11)

In Xs0m = - 1 . 4 2 In (Pro - 1) + 6.14 (12)

72

where Ym is the mass fraction of mineral m of geometric mean size ~ re-

porting to the underflow and ~s0m is the corrected cut-size of mineral m of
density Pm. The cyclone model fitted the data to give a mean square of
residuals on Ym of 0.0018.
Figure 10 shows the comparison of measured and simulated cyclone over-
flow size distribution of galena, pyrite and calcite/dolomite. The fit is good
for the calcite/dolomite but the simulation is too fine for galena and pyrite.
The dashed line simulation is obtained by substituting measured Ym values
for those calculated from eqs. 11 and 12. The fit is n o w much improved in-
dicating that the cyclone simulation introduces the most error; the ball mill
model and the assumption that K m = K did n o t introduce significant error
in circuit simulation. The principal problem is that Ym according to eq. 11
approaches 1 with increasing x t o o rapidly for the high density minerals,
which, due to locking with lower density components, actually attains Ym
more slowly. Another illustration of problems encountered with cyclone
models is that at 190.3 mtph, the short circuiting to underflow factor (0.31
in eq. 11) appeared to be mineral dependent, a not unprecedented observa-
tion (Finch and Matwijenko, 1977).
Figure 10 also shows that Pb is the finest c o m p o n e n t of the cyclone over-
flow because of preferential re-cycling by the cyclone, not preferential
breakage.
t i i 1 i
100 ~ o--- ° - ~ ° / °~°--°

.,,o / : / PbS I , v
m
o. o CYCLONE OVERFLOV~

60 / Measured
FeS2 ,/ o symbols

Simulated
40 /o -- Ym m o d e l
"~ Cal-
(O Dol o
-- - - Ym measured

0 I I I I I
400 200 100 48 28

Particle Size, mesh

Fig. 10. Measured (symbol) and simulated cyclone overflow mineral size distribution.
(Simulation (full line) uses cyclone model, eqs. 11 and 12, while simulation (dashed line)
uses measured Ym values. )

DISCUSSION

Overall size reduction

Figures 2 and 3 show that the plug flow approximation with cumulative-
basis first-order kinetics leads to a simple closed-circuit ball mill size reduc-
 73

tion model. The actual values of K do depend on the estimate of T and

hence VBH but the linearity of K vs x is n o t dependent on r which will
only affect C. (The assumption VBH equals 18% of mill volume is similar
to that made by Hodouin et al. (1978) but may be low due to dynamic
factors (Smith and Guerin, 1979). Thus with C "calibrated" and with a con-
sistent calculation of r, the model is adequate for size reduction simulation.
This is demonstrated in Fig. 4, where, to put in perspective, the deviations
are of the magnitude encountered in calibration of real-time particle size
analyzers (Hathaway and Gunthals, 1976). The applicability of this model-
ling scheme can be readily determined by checking the linearity of
In (R(x,O)/R(x,T)} (i.e. KT) against x. Experience with the Pine Point and
Gibraltar data shows that raw data are adequate for this initial survey. Data
adjustment has the effect of reducing the confidence interval on K (and
hence on C and n).
The success of the plug flow approximation probably stems from the
high feed rate to the ball mill resulting from high circulating loads. It is
emphasized that this modelling approach is for closed-circuit ball mills.
The constancy of K (at a given x) over a wide range of operating condi-
tions deserves comment. Since the discrete size rate-of-breakage parameter is
a function of x, it can be expected that K should depend on the ball mill
feed size distribution. For example, K (x = 75 pm) should depend on the
proportion of +300 pm material. This feed size distribution in turn can be
expected to depend on operational factors such as fresh feed rate and ore
hardness. The apparent insensitivity of K may mean the ball mill feed size
has n o t changed sufficiently. Another possibility is that the size distribution
shape has not changed sufficiently, preserving the relative proportion of each
size and that this is the key to a constant K.
For the Gibraltar data, taking - 2 8 mesh as the size indicator, the range
was from 68.6% to 78.7%. For Pine Point the variation was from 20.6%
- 2 0 0 mesh (154.2 mtph) to 23.4% - 2 0 0 mesh (Table I). To examine the
distribution, the ball mill feed was fitted to the Rosin-Rammler distribution
to extract n~:

R (x) = exp [ - C , x n' ] (13)

where R(x) is the mass fraction coarser than x. For Gibraltar n, varied
from 1.13 to 0.75 and for Pine Point from 1.10 (154.2 mtph) to 0.97
(190.3 mtph).
It is tentatively postulated that these variations, which must be typical of
many circuits, are n o t sufficient to influence K. High circulating loads prob-
ably contribute to damping the variations.
Figure 3 does show K (x = 2000 pm) with a greater uncertainty and fal-
ling below the (extrapolated) linear relationship. The greater uncertainty is
associated with the small mass retained at 9 mesh, which is subject to a
large relative error, and a quotient having to be formed to solve for K in
eq. 6. A decreased K at the coarse size, by analogy with observation on the
74

relationship between the discrete size rate-of-breakage parameter and

(Austin, 1973) may reflect a particle to ball and/or mill size unfavourable
to breakage. If the raw data is divided into two sets, one for r ~< 0.7 min,
reflecting high ball mill feed rates (~> 1850 mtph dry), the other for
r > 0.7 min, the mean K (with 95% confidence interval) is 0.747 + 0.148
and 0.893 + 0.143, respectively'. Thus at higher feed rates the K value is
lower, while for the lower feed rates K approaches the value expected by
extrapolation. The low K may, therefore, represent a mill overload condi-
tion. The division into low and high flow rates produces little difference at
smaller sizes, the " n o r m a l " relationship being recovered by x = 850 pm.
This, incidentally, reflects the small mass retained at x = 2000 pm.
K, then, is not entirely free of operating conditions.
For theoretical understanding, it would be useful if C and n had a
physical meaning. The possible application of C has been discussed by
Furuya et al. (1971) but since C here depends on the VBH assumption
without a direct measurement of VBH present discussion of C is n o t worth-
while. On the other hand, n is independent of VBH. Gaudin and Meloy
(1962) deduced that the probability of breakage by ball impact (e) was
proportional to x. Since K and e are proportional, the present results would
be in agreement if n = 1. This appears to be the case for Pine Point; for
Gibraltar n is somewhat lower. F u r u y a et al. (1971) suggested n was equal
to n l of the Rosin-Rammler distribution. Taking the ball mill feed, the
average nl for Pine Point was 1.04 and for Gibraltar 0.926 + 0.046 (95%
confidence interval). There is some agreement, therefore, with the suggested
equivalence. However, it should be noted that the deduction e a x used a
highly idealized model and the power law and Rosin-Rammler distribution
are simply convenient empirical models.

Mineral size r e d u c t i o n

Figure 5 shows the Km vs x relationship is similar to the overall. Greater

uncertainty enters here because as well as screening errors, there are assay
errors. A relationship between Km and x could be identified, but at least
for this example at Pine Point, the approximation K ~ Km can be made.
The difference in size distribution of galena and pyrite in the ball mill dis-
charge (Figs. 6a, b) depends not so much on grinding as on the classifica-
tion which created differences in the mineral distribution in the f e e d to the
ball mill.
The lack of substantial difference between mineral grinding kinetics is
emphasized by Fig. 10. It is for instance misleading, in this case, to con-
sider galena a " s o f t " mineral. The "sliming" of the Pb in the cyclone over-
flow results from recycling due to the high density of galena. The observa-
tion Km -~ K would also make K independent of mineral mix, as the
limited data here show. However, a change in mineral mix of the heads may
cause a change in overall cyclone overflow fineness. For example, an in-
crease in Pb heads should give a finer overall cyclone overflow. A particle
size monitor interpreting this as " s o f t e r " ore may result in erroneous
control action.
In detail, Fig. 5 shows pyrite to be the " h a r d e s t " mineral, as might be ex-
pected. This is reflected in Fig. 6a, b where the simulation slightly "over-
grinds" the coarse pyrite. The fine galena also appears hard, in contrast to
the usual assumption. There is some evidence that galena above 150 mesh
comprises loose agglomerates of grains, while below the galena is present as
coherent, single grains. The coherent grain will be harder.

Laboratory-derived kinetics

It is of interest to determine if laboratory batch grinds reveal similar rela-

tive mineral Km values and a similar relationship with x. This would imply
possible determination and use at the circuit design stage. Selecting rod mill
discharge and ball mill feed gives samples from the same ore which differ in
particle size and mineral mix, permitting the effect of these differences to
be observed.
The initial step is to show agreement with the first order hypothesis.
Deviation from first order was apparent, and a second parameter, p, was
introduced. Austin (1971/1972) discusses deviation from first-order, the
deviations, here described by p < I and p > 1, are equivalent to his a- and
c-types, respectively. Deviation due to too wide a size interval does not ap-
pear to be the problem as the wider intervals (e.g. +200 mesh) tend to show
less deviation. Horst and Bassarear (1976) also determined cumulative basis
rate-of-breakage parameters and found no effect from size range. The coarse
sizes consistently show p ~ 1. For example, Fig. 7a shows p = 0.78 for the
+28 mesh overall. This is equivalent to K decreasing with time. A physical
interpretation is that initially the soft components are removed leading to
enrichment with the harder minerals. Pyrite is probably the hardest com-
ponent (see Fig. 5). The assay of FeS2 in the +28 mesh fraction increased
from 23.7% at t = 0 min. to 56.6% at t = 12 min., supporting the physical
picture.
Detailed analysis of the laboratory data is n o t essential to the present
study. Taking the range 0.9 < p < 1.1 to define first order within experi-
mental error gave Figs. 8 and 9. These show the same general relationships
as the plant data (Figs. 2 and 5); K m vs x is approximately parallel to K
vs x and the K m / K ratio is similar with pyrite, again, the hardest component.
Considering x ,: 212 pm, then the slope of K vs x is also similar to the
plant~derived relationship. (Comparing Figs. 7 and 2 also suggests K for
plant and laboratory are numerically similar. Without a direct measure of
VBH, no c o m m e n t on this will be offered.) Such similarities are basic to
scale-up and, for example, facilitated the use of laboratory-derived grinding
parameters in the simulation of Ramirez-Castro and Finch (1980). A nota-
ble difference of laboratory and plant data in the former shows K and K m
76

reaching a maximum at x --- 300 pm, whereas no maximum is suggested for

the plant data. This can be expected based on the work of Austin (1973)
where the maximum in the discrete-size rate-of-breakage parameter occurred
at smaller x for smaller diameter mills.
No unequivocal conclusion ,is reached regarding the differences in param-
eters using rod mill discharge and ball mill feed samples. Statistically, differ-
ences can be shown. However, substituting rod mill discharge-deduced param
eters for ball mill feed-deduced ones does n o t lead to a large effect on
mineral size distribution (Ramirez-Castro, 1979). Nevertheless some caution
may have to be exercised in laboratory determination of grinding param-
eters for highly heterogeneous ores as the mill/cyclone interaction may suf-
ficiently alter the mineral mix in the ball mill feed to invalidate parameters
deduced on an ore sample. A potentially important practical problem in ap-
plying mineral grinding kinetics at the design stage is the accuracy of the
parameters as screening and assaying errors are combined. This is felt to
play a role in KpbS falling outside of p = 0.9 to 1.1 on most occasions. To
illustrate, a 100 g screening sample containing 2% PbS means effectively
only a 2 g PbS sample is screened from which to extract the parameters.
No repeat tests were performed, so the uncertainty in Km is unknown.

CONCLUDING REMARKS

Adding mineral behaviour to the usual grinding and classification models

involves a degree of complexity which warrants, at least initially, the use of
simplified unit models. Using a plug flow transport model is supported by
the evidence here and by the analysis of closed grinding circuits by Furuya
et al. (1971). Cumulative-basis kinetics has been successfully applied by
others (Razumov et al., 1968; Horst and Bassarear, 1976) and appears to
have been first applied by Roberts (1950). For control studies often only
one size is monitored, and a cumulative-basis rate-of-breakage for this size
may be sufficient for modelling purposes (Olsen and Lenhartzen, !979). An
implicit simplifcation in the analysis is to ignore locked particles.
The model presented is adequate for investigating a circuit modification
such as a unit cell on ball mill discharge. Ramirez-Castro and Finch (1980)
suggested, with a unit cell in place, that the lead and zinc grinds could be
tuned to their respective flotation requirements. This would also be amen-
able to simulation with the aid of the proposed ball mill model. The more
complex problem appears to be modelling individual mineral classification
behaviour, as the attempted circuit simulation revealed here.

CONCLUSIONS

(1) A cumulative basis rate-of-breakage parameter (K) with plug flow

material transport adequately fitted overall size reduction thrdugh the
closed-circuit ball mill at the Pine Point and Gibraltar concentrators. The
 77

success of this simplified approach probably stems from high circulating

loads.
(2) The rate-of-breakage parameter was related to size (x) by a power
law. For Pine Point the exponent was n = 1.043 -+ 0.026 (all x) and for
Gibraltar n = 0.797 + 0.020 (x ~< 850 pm).
{3) An analogous size reduction model was applied to individual
minerals. Pyrite was the hardest mineral but assuming Km = K did not lead
to substantial error in simulating mineral deportment.
(4) Laboratory batch grinding tests on rod mill discharge and ball mill
feed samples gave similar kinetics which were also similar to the plant re-
sults in terms of K vs x and K m / K relationships.
(5) Modelling mineral classification seems to pose a greater problem in
circuit simulation than mineral size reduction from the experience at the
Pine Point concentrator.

ACKNOWLEDGEMENTS

The authors wish to thank the personnel at the Pine Point Mines con-
centrator for their enthusiastic help in sample gathering and Gibraltar Mines
for making data available. Funding was partly from a Mexican government
grant to Ramirez-Castro, the Natural Sciences and Engineering Research
Council of Canada, the Quebec Department of Education (FCAC programme)
and Dept. of Supply and Services, Canada.

REFERENCES

Austin, L.G., 1973. Understanding ball mill sizing. Ind. Eng. Chem. Process Des. Dev.,
12: 1 2 1 - 1 2 9 .
Austin, L.G., 1971/1972. A review -- introduction to the mathematical description of
grinding as a rate process. Powder Technol., 5: 1--20.
Cameron, A.W., Kelsall, D.F., Restarick, C.J. and Stewart, P.S.B., 1971. A detailed
assessment of concentrator performance at Broken Hill South Limited. Australas. Inst.
Min. Metall., Proc., 240: 53--65.
Finch, J.A. and Matwijenko, O., 1977. Individual mineral behaviour in a closed grinding
circuit. CIM Bull., 70: 164--172.
Furuya, M., Nakajima, Y. and Tanaka, T., 1971. Theoretical analysis of closed-circuit
grinding system based on comminution kinetics. Ind. Eng. Chem. Process Des. Dev.,
10: 449--456.
Gaudin, A.M. and Meloy, T.P., 1962. Model of a comminution distribution equation for
repeated fracture. Trans. AIME, 223: 43--50.
Harris, C.C., 1971. The effect of time on batch grinding. Powder Technoh, 4: 57--60.
Hathaway, R.E. and Gunthals, D.L., 1976. The continuous measurement of particle size
in fine slurry processes. CIM Bull., 69: 64- -71.
Himmelblau, D.M., 1970. Process Analysis by Statistical Methods. Wiley, New York, N.Y.
118 pp.
Hodouin, D., B~rub~, M.E. and Everell, M.D., 1978. Modelling industrial grinding circuits
and application in design. CIM Bull., 71: 138--146.
Horst, H.E. and Bassarear, J.H., 1976. Use of simplified ore grindability technique to
evaluate plant performance. Trans. AIME, 260: 348--354.
78

Le Houillier, R. and Marchand, J.C., 1977. Empirical correlation predicting particulate

mass effect on selection parameters. Powder Technol., 17: 101--107.
Lynch, A.J., 1977. Mineral Crushing and Grinding Circuits. Elsevier, Amsterdam,
a) pp. 117--119 and pp. 64--65; b) Chap. 7.
Olsen, T.O. and Lenhartzen, T., 1979. Grinding parameter estimation using the
maximum likelihood method. Int. J. Miner. Process., 6: 133--153.
Ramirez-Castro, J. and Finch, J.A., 1980. Simulation of a proposed grinding circuit
change to reduce lead sliming. CIM Bull., 73: 132--139.
Ramirez-Castro, J., 1979. Simulation of a Proposed Grinding Circuit Change to Reduce
Lead Sliming. M. Eng. Thesis, McGill University, Montreal, Canada.
Randolph, A.D. and Ranjan, R., 1977. Effect of a material-flow model in prediction of
particle-size distribution in open- and closed-circuit mills. Int. J. Miner. Process., 4:
99--110.
Razumov, K.A., Perov, V.A., Zverevitch, V.V. and Bilenko, L.F., 1968. Some regularities
of ball mill grinding. VIIIth Int. Miner. Process. Congr., Leningrad (in Russian).
Roberts, E.J., 1950. The probability theory of wet ball milling and its application. Trans
AIME, 187: 1267--1272.
Smith, H.W. and Guerin, D., 1979. Dynamic modelling, simulation and control of a
grinding circuit. Pres. CIM Conf. Metallurgists, Sudbury, Aug. 19--23.
Taggart, A.F., 1945. Handbook of Mineral Dressing. Wiley, New York, N.Y., pp. 5--32.