Basics of Milling and
Classification
Adrian Hinde
[email protected]
Minerals Processing Division, Mintek
Topics of discussion
Bond methods for rod and ball milling
Population balance modelling
Hydrocyclone modelling
Basic concepts of AG/SAG milling
Bond’s third law of
comminution
Bond tried to justify his “law” in terms of
Crack theory. Although his arguments
were fundamentally flawed, his basic
equation has stood the test of time
1 1
W 10Wi
P F
80 80
Bond rod mill test
Locked cycle test in a 12 inch rod mill
Must have stage crushed feed to –12.7 mm
Need to calculate the grams of product
produced per rev. after each cycle to home in
on steady state
Usually requires about 7 iterations
Bond rod mill test
62
Wi
0.23 0.625 10 10
P1 Grp
P F
Bond ball mill test
Locked cycle test in a 12 inch ball mill
Must have stage crushed feed to –3.26 mm
Need to calculate the grams of product
produced per rev. after each cycle to home in
on steady state
Usually requires about 7 iterations
Bond ball mill test
44.5
Wi
0.23 0.82 10 10
P1 Grp
P F
Bond work index is not a
constant
Need to measure at different closing mesh
sizes
Ball Work Index for UG2 ores can vary from 7
to 25 kW/t for 212 µm to 75 µm, respectively
Problems with dry screening below 75 µm
Some useful calculations
42.305
Critical speed Cs
D
Fractional critical speed f Cs
mill speed[rpm]
Cs
Mill diameter inside Per cent of critical speed
liners [m]
Rod mill Ball mill
0.91-1.83 76-73 80-78
1.83-2.74 73-70 78-75
2.74-3.66 70-67 75-72
3.36-4.57 67-64 72-69
4.57-5.49 - 69-66
More useful calculations
D 2
Mass of ball/rod load M V p LB D
4
Mill power per ton of balls – slump factor only applies for
mill>3.3 m diameter
(1 / 3) 0.1
K Wb 4.879D (3.2 3V p ) f Cs 1 9 10 f Cs S s
2
B 12.5 D
S s 1.1023
50.8
Mill power per ton of rods
K Wr 1.752 D 1 / 3 (6.3 5.4V p ) f Cs
Applying correction factors
Dry grinding - EF1
Dry grinding requires 1.3 times more power
than wet for given grind
Open circuit grinding – EF2
Product size control reference per Open circuit inefficiency
cent passing multiplier
50 1.035
60 1.05
70 1.10
80 1.20
90 1.40
92 1.46
95 1.57
98 1.70
Diameter efficiency - EF3
EF3 has a value of 1 for mills smaller than
2.44 m diameter
0.2
2.44
EF3
D
Oversized feed – EF4
Optimum feed sizes
14.33
Fo ( Rod ) 16 000
Wi
14.33
Fo ( Ball ) 4 000
Wi
F80 Fo
Rr (0.9072Wi ) 7)
Fo
EF4
Rr
Fine ball milling - EF5
Applies to P80 less than 75 µm
P80 10.3
EF5
1.145 P80
High or low reduction ratio for
rod milling = EF6
Optimum reduction ratio
L
Rro 8 5
D
Applies for limits of Rro-2 and Rro+2
Rr Rro
2
EF6 1
150
High case only applied when WI>7.7 kWh/t
Low reduction ratio for ball
milling – EF7
Only applies when reduction ratio is less than 6
2( Rr 1.35) 0.26
EF7
2( Rr 1.35)
Rod milling crusher
circuit type – EF8
Milling circuit type Factor EF8
Crusher circuit type
Open Closed
Rod mill only 1.4 1.2
Rod mill – ball mill 1.2 1.0
Optimal makeup rod and ball
sizes
F 0.75 Wi S g
Rmax 25.4 80
160 110 .23 f
Cs 3.281D
If Rr is less than than 8, Rmax is increased by
12 mm
F W S
1/ 3
Bmax 25.4 80 i g
K 110 .23 f
Cs 3.281D
Startup rod Size
distributions
Make-up Mass per cent make-up rod size for
rod sizes
[mm] Rmax [mm]
125 115 100 90 75 65
125 18 - - - - -
115 22 20 - - - -
100 10 23 20 - - -
90 14 20 27 20 - -
75 11 15 21 33 31 -
65 7 10 15 21 39 34
50 9 12 17 26 30 66
Startup ball sizes
Make-up Mass per cent of make-up ball size for
ball sizes
[mm] Bmax [mm]
115 100 90 75 65 50 40
115 23.0 - - - - - -
100 31.0 23.0 - - - - -
90 18.0 34.0 24.0 - - - -
75 15.0 21.0 38.0 31.0 - - -
65 7.0 12.0 20.5 39.0 34.0 - -
50 3.8 6.5 11.5 19.0 43.0 40.0 -
40 1.7 2.5 4.5 8.0 17.0 45.0 51.0
25 0.5 1.0 1.5 3.0 6.0 15.0 49.0
Shortcomings of Bond’s
method
Can’t provide complete size distribution
information and caters only for 0 and 250 %
circulating load
Database for correction factors based on
mills available 50 years ago
Can apply only to rod and ball mills
Can’t apply to grinding-circuit subprocesses
and optimisation tasks
Population balance model
Basics for milling
b52 b42 b32 S2
6 5 4 3 2 1
History of Mill Population
Blance Modelling
Brown in 1941 first suggested the idea
Broardbent and Chalcott in 1956 formalised the
maths using a matrix approach
Gardner and Austin in 1962 demonstrated its
usefulness to industrial applications
Herbst and Fuerstenau in 1973 introduced the
concept of energy-based breakage rates
JKSimMet, Modsim, MinOOcad simulation
packages now available
Cumulative breakage rate
model
Fully mixed
Accumulation =in - out - consumption
dMW P E
i
FFi FWi Ki Wi M
dt M
Plug flow is equivalent to batch
Accumulation=-consumption
dMW P E
i
K iWiM
dt M
Steady State Solutions
Fi K iE
Pi
1 K iE
Pi 1 (1 Fi ) exp( K iE )
Can solve numerically for
dynamic simulation
dMWi P E
FFi FWi K i Wi M
dt M
becomes
P E
Wi (t t ) Wi (t ) t ( F ( Fi Wi ) /M) Ki Wi )
M
Cumulative breakage rate
functions can take simple
forms
K iE (exp(a1 ln( x i ) a 2 (ln( x i )) 2 a n (ln( x i )) n ))
1
K x i
i
E
1 ( xi / )
Shortcoming of cumulative
breakage rate models
Assumes that breakage rates are not affected
by the size distribution above the size of
interest
Model can correlate a single set of data
accurately and predict trends, but cannot
predict accurately.
Need to treat breakage rates and breakage
distributions separately
Grindmill for measuring
breakage rate and breakage
distribution functions
Measuring breakage
parameters by milling
particles in narrow size
ranges
dmi (t ) P E
S i mi (t )
dt M
mi ( ) mi (0)(exp( S )) i
E
Can determine Si by regression
Breakage distributions
can be measured by
milling for short time
Specific breakage rate
function
100
-1
Specigfic breakage rate [kWh/t]
10
0.1
0.01 0.1 1 10
Particle mesh size [mm]
1
S xi
i
E
1 ( xi / )
Breakage distribution
function
Bij j ( x i 1 / x j ) (1 j )( x i 1 / x j ) ; i j
1; i j
j 1 ( x j / x1 ) ; i 2,3, n 1, j ( x j / x1 ) 1
1 , j ( x i / x1 ) 1
Back-calculation of
parameters
Doing one size class at a time is too time
consuming
Back-calculation involves regression or
minimisation of sum-of-squares of difference
between measured size distribution and
calcualted size distribution
Beware of parameter
sensitivity issues
Start with simplest model until sum of
squares changes by marginal amount only
S x i
i
E
Bij ( x i 1 / x j )
Residence time distributions
Production mills don’t approximate plug flow
Can’t use a batch grind to directly determine
power for a production mills
Production mills behave like three mixed
tanks in series
1 4
2 3
1 t
(t ) exp( t / 1 ) exp( t / 2 ) exp( t / 2 )
( 1 2 ) ( 1 2 ) 2
Comparison of open circuit
and closed circuit milling
1 4
2 3
10
9 Mill residence time simulated as three fully mixed mills
1 6
3 4 5
2 7
Open and closed circuit
milling
4500
4000
3500 Open circuit
Closed circuit
3000
Net power [kW]
2500
2000
1500
1000
500
0
fully mixed typical plug flow
Hydrocyclone partition
functions
The performance of a classifying hydrocyclone
is probably best represented by its partition
function giving recoveries as a function of
particle size to the underflow
Popular partition models are those of the form:
Ri ( xi ) r f (1 r f )Ci ( xi )
Where rf is the feed bypass fraction, which
closely approximates the water recovery.
Ci represents the the effects of classification
after correcting for the water split
Popular classification functions
After, correcting for the water split the following two
functional forms for the classification function are probably
the most commonly used:
Rosin-Rammler:
Ci 1 exp( 0.6931( xi / d 50c ) m
x
exp i 1
Double exponential: d 50c
Ci
xi
exp exp 2
d 50c
Ci 1 /(1 1 /( xi / d 50c )) 0
1.54m 0.47
Modelling objectives
To establish how the bypass, cut-size
and sharpness of cut parameters
change with cyclone geometry and
operating variables such as feed
,
flowrate, feed pulp density, and feed
size distribution
Historical background
Hydrocyclones first used in minerals processing
in the 1940s
Early research focussed mainly on small
laboratory units operated at low pulp densities
with results ineterpreted in terms of
fundamental fluid mechanics theory
Breakthroughs came in the 1960s and 1970s
with empirical approaches to modelling cyclone
behaviour at JKMRC with PhD theses by
Rao(1966) and Nageswararao(1975)
Plitt’s model followed in 1976
Defining hydrocyclone
geometry
Calculating Cut-Size
(equilibrium orbit
hypothesis)
Fd 3dw
d3 v2
Fc (s ) Fd 3dw
6 r 18wr
d 50c
( s )v 2
Plitt’s cut-size regression
model
0.459
F1 0.000998Dc Di0.605 Do1.205 exp(0.0635 )
d 50 c
Du0.714 h 0.382 Q 0.448 ( s ) 0.5
Flow and water split
F21.36( Du / Do ) 3.31 h 0.544 ( Du2 Do2 ) 0.357 exp(0.00539 )
S
H 0.236 Dc1.11
n
m
S /( S 1) i
1 f exp[ 0 . 6931( d i / d 50 c ) ]
100 i 1
Rf
n m
1 1 f i exp[ 0.6931(d i / d 50 c ) ]
100 i 1
Pressure drop
Pressure drop based on head loss approach
that should give pressure drop proportional to
square of feed volumetric flowrate
F3 595000.0Q1.78 exp(0.00546 )
P 0.365 0.938 0.281 2
Dc Di h ( Du Do2 ) 0.870
Sharpness of cut
Plitt argued that sharpness should
depend on residence time and volumetric
recovery to underflow
0.15
D h
2
m F4 0.0931 c
exp[1.58(1 Rv )]
Q
JKMRC argue that sharpness depends on
ore type only for given cyclone
Cyclones suppliers say inlet design has
an effect
Model shortcomings
Plitt and Nageswararao models don’t cater
for feed particle size effects
Yet Nageswararao’s developed separate
models for water and volumetric flow
recoveries, which implies cut size is
uniquely defined by feed size distribution
Plenty of evidence from testwork that this
is the case for industrial cyclones
operating at high pulp densities
Models don’t cater for roping and
operational instabilities
Spigot Capacities
Equipment suppliers support the “crowding
theory” hypothesis and publish spigot
capacity relationships
AG/SAG - historical
development
In the 1950s in North America, large
diameter pancake mills with high impact
condition were developed for the primary
milling of ROM ores
At about the same time South Africa,
Jackson tried feeding ROM ore to pebble
mills, leading to first single-stage FAG mill at
Winkelhaak with emphasis on abrasion
AG/SAG Mill Geometry
Charge motion
Cataracting
Cascading
Kidney
Modes of breakage
Attrition/compression
breakage of interstitial fines
Impact spalling
Surface chipping
Surface
abrasion by
rubbing Self breakage
Impact cracking
DEM analysis
Computation tool for analysing individual
particle interactions in mills based on
fundamental classical mechanics theory
Can be coupled with CFD and PB modelling
- called high-fidelity simulation
DEM analysis of charge
dynamics
Charge snapshot Charge velocity map Charge impact energy map
Charge contact force map Charge abrasion wear map
Progeny fragment
distributions
Discharge behaviour
fi wi D ci wi
For fully mixed conditions, discharge rates
for class i are proportional to hold-up mass
and a classification coefficient 0 ci 1
Steady state solution to
population balance model
Picard’s algorithm
i 1
P
f i bij S j wj
j 1 M
wi
E P
D ci Si
M
Make an initial estimate of the hold-ups
from pilot data and iterate using circular
calculations
Cumulative breakage rates
Low load SAG - cumulative rates of breakage High load SAG - cumulative rates of breakage
0.100 0.100
Cumulative rates of breakage [1/minute]
Cumulative rates of breakage [1/minute]
0.010 0.010
Si measured
Si measured
Series2
Si model fit
0.001 0.001
0.01 0.10 1.00 10.00 100.00 1000.00 0.01 0.10 1.00 10.00 100.00 1000.00
Upper limit of particle mesh size Upper limit of particle mesh size
Low charge load High charge load
Grate Discharge Classification
Low load SAG - grate classification model fit High load SAG, grate classification model fit
1.0 1.0
0.9
Grate fractional recoveries to oversize
0.9
Grate fractional recoveries to oversize
0.8 0.8
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0.0 0.0
0.010 0.100 1.000 10.000 100.000 1000.000 0.010 0.100 1.000 10.000 100.000 1000.000
Upper limt of size class [mm] Upper limit of size class [mm]
Population balance based on
separate breakage rate and
breakage distribution functions
Impact breakage distribution function derived from dropweight tests
Low energy abrasion distribution function derived from tumbling test
Combined breakage distribution function taken as weighted sum
t a BijA t10 BijI
Bij
t a t10
Specific breakage rate function obtained by back-calculating from pilot
or plant data
Impact Breakage Function
0.1
Mass fraction less than size
0.01
0.001
0.0001
0.00001
0.010 0.100 1.000 10.000 100.000 1000.000
Particle mesh size [mm]
Normalised abrasion
breakage function
100
Mass percentage less than size
10
1
0.001 0.01 0.1 1
Normalised particle size (size/top size)
Abrasion Breakage
Distribution Function
1
Mass fraction less than size
0.1
0.01
0.001
0.01 0.1 1 10 100 1000
Particle mesh size [mm]
Overall breakage
distribution function
1
Mass fraction less than size
0.1
0.01
0.001
0.01 0.1 1 10 100 1000
Particle mesh size [mm]
Specific breakage rates
estimated by back calculation
from plant data
10
-1
Specific breakage rate [kW/t]
0.1
0.01
0.01 0.1 1 10 100 1000
Particle mesh size [mm]
Dynamic model for SAG mill
using separate B and S
accumulati on in out generation consumptio n
i 1
d ( wi ) P E P
f i D ci wi bij S j w j Si wi
dt j 1 M M
accumulation in out generation consumption
i 1
P E P
wi (t t ) wi (t ) t ( f i D ci wi bij S j w j Si wi )
j 1 M M
Steady state solution to standard
SAG/AG model using Picard
algorithm
i 1
P
f i bij S j wj
j 1 M
wi
E P
D ci Si
M
Make an initial estimate of the hold-ups
from pilot data and iterate using circular
calculations
