https://www.scribd.com/doc/127929066/AdvancedComminutionTechnologiespdf
11/29/2013
text
original
Society for Mining, Metallurgy, and Exploration, Inc.
ADVANCES
IN
COMMINUTION
Society for Mining, Metallurgy, and Exploration, Inc. (SME)
8307 Shaffer Parkway
Littleton, Colorado, USA 80127
(303) 9739550 / (800) 7633132
www.smenet.org
SME advances the worldwide mining and minerals community through information
exchange and professional development. SME is the world’s largest association of mining
and minerals professionals.
Copyright 2006 Society for Mining, Metallurgy, and Exploration, Inc.
All Rights Reserved. Printed in the United States of America.
Information contained in this work has been obtained by SME, Inc., from sources believed to be
reliable. However, neither SME nor its authors guarantee the accuracy or completeness of any
information published herein, and neither SME nor its authors shall be responsible for any errors,
omissions, or damages arising out of use of this information. This work is published with the
understanding that SME and its authors are supplying information but are not attempting to
render engineering or other professional services. If such services are required, the assistance of an
appropriate professional should be sought.
No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any
form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the
prior written permission of the publisher. Any statement or views presented here are those of the
author and are not necessarily those of SME. The mention of trade names for commercial products
does not imply the approval or endorsement of SME.
ISBN13: 9780873352468
ISBN10: 0873352467
Library of Congress CataloginginPublication Data
Advances in comminution / edited by S. Komar Kawatra.
p. cm.
Includes bibliographical references and index.
ISBN13: *9780873352468
ISBN10: 0873352467
1. Stone and ore breakersTechnological innovations. 2. Crushing machineryTechnological
innovations. 3. Mining engineeringTechnological innovations. I. Kawatra, S. K.
TN510.A38 2006
622'.73dc22
2005057533
vii
Preface
This third international symposium and proceedings, Advances in Comminution, have
come at a critical time. Because of rapidly rising energy prices, it is important that the
latest information be made available for improving the efficiency of highly energyintensive
comminution processes.
The contributors and topics for this third international symposium have been care
fully selected to provide a balance between academic and industrial practice so that the
reader can readily find information on current best practices and evaluate future indus
try trends.
Two previous symposiums, also organized by the Society for Mining, Metallurgy,
and Exploration, were great successes. The first conference was held in 1992, at a time
when there was much discussion about switching from traditional rod mill and ball mill
circuits to autogenous grinding. The second comminution symposium, held in 1997,
focused on initial installations of high pressure grinding rolls (HPGRs). Now, in 2006,
the HPGRs are becoming part of hardrock grinding circuits. They have proven to be a
very economical addition to many comminution processes because of lower energy con
sumption and easy integration into existing conventional systems.
The 2006 conference focuses on the dilemma of needing to grind materials to ever
finer sizes while maintaining reasonable energy costs. The selection and sizing of stirred
mills for regrinding and ultrafine grinding applications do not lend themselves to con
ventional methodologies; therefore, new approaches are being developed. There is also
a great deal of activity directed toward improving ore characterization to predict AG/
SAG mill energy requirements, as well as developing improved models and instrumenta
tion for optimization and control of comminution circuits. Instrumentation, modeling,
and control functions in particular have benefited from rapidly advancing computer
technology, with calculations that were formerly extremely timeconsuming becoming
rapid and routine. These advances will keep energy waste to a minimum and will pro
vide the increased energy efficiency needed to maintain ongoing industry success.
It is hoped that the symposium and these proceedings will be useful to those who
are working toward major advances in industrial practice. Appreciation is extended to
members of the organizing committee, who were instrumental in acquiring highquality
papers and reviewing them on very short notice, and to the SME staff, particularly Ms.
Tara Davis and Ms. Jane Olivier, for their assistance in organizing the third international
symposium and publishing these proceedings.
iii
Contents
EDITORIAL BOARD v
PREFACE vii
PART 1 ADVANCED COMMINUTION TECHNOLOGIES 1
HighPressure Grinding Rolls—Characterising and Defining
Process Performance for Engineers 3
HighPressure Grinding Rolls—A Technology Review 15
Some Basics on HighPressure Grinding Rolls 41
HighPressure Grinding Rolls for Gold/Copper Applications 51
Selection and Sizing of Ultrafine and Stirred Grinding Mills 69
Effects of Bead Size on Ultrafine Grinding in a Stirred Bead Mill 87
Specific Energy Consumption, Stress Energy, and Power Draw of
Stirred Media Mills and Their Effect on the Production Rate 99
AG/SAG Mill Circuit Grinding Energy Requirement—How
to Predict It from SmallDiameter Drill Core Samples Using
the SMC Test 115
PART 2 COMMINUTION PRACTICES 129
Causes and Significance of Inflections in Hydrocyclone
Efficiency Curves 131
SimulationBased Performance Improvements in the Ispat
Inland Minorca Plant Grinding Circuit 149
Determining Relevant Inputs for SAG Mill Power Draw
Modeling 161
Cement Clinker Grinding Practice and Technology 169
Extended Semiautogenous Milling: Smooth Operations
and Extended Availability at C.M. Doña Ines de Collahuasi
SCM, Chile 181
PART 3 LIBERATION AND BREAKAGE 191
Shell and Pulp Lifter Study at the Cortez Gold Mines SAG Mill 193
Breakage and Damage of Particles by Impact 205
The Rationale behind the Development of One Model
Describing the Size Reduction/Liberation of Ores 225
Influence of Slurry Rheology on Stirred Media Milling
of Limestone 243
iv
Experimental Evaluation of a Mineral Exposure Model
for Crushed Copper Ores 261
Linking Discrete Element Modeling to Breakage in a PilotScale
AG/SAG Mill 269
Significance of the ParticleSize Distribution in the Quality
of Cements with Fly Ash Additive 285
Modeling Attrition in Stirred Mills Applying Statistical Physics 293
PART 4 MILL DESIGN 307
Design of Iron Ore Comminution Circuits to Minimize
Overgrinding 309
Evaluation of LargerDiameter Hydrocyclone Performance
in a Desliming Application 321
Selection and Design of Mill Liners 331
The Importance of Liner Geometry and Wear in Crushing 377
Bond’s Method for Selection of Ball Mills 385
Developments in SAG Mill Liner Design 399
The Gearless Mill Drive—The Workhorse for SAG and Ball Mills 413
Optimizing Hydrocyclone Separation in ClosedCircuit Grinding 435
PART 5 INSTRUMENTATION, MODELING, AND SIMULATION 445
Use of Multiphysics Models for the Optimization
of Comminution Operations 447
Batu Hijau Model for Throughput Forecast, Mining and Milling
Optimization, and Expansion Studies 461
The Use of Process Simulation Methodology in Process Design
Where Time and Performance Are Critical 481
Modeling and Simulation of Comminution Circuits
with USIM PAC 495
Remote and Distributed Expert Control in Grinding Plants 513
Developments in Sensor Technology for Tumbling Mills 527
Ball Mill Circuit Models for Improving Plant Performance 539
INDEX 547
1
PART 1
Advanced Comminution
Technologies
3
HighPressure Grinding Rolls—
Characterising and Defining Process
Performance for Engineers
Richard Bearman
*
ABSTRACT
Highpressure grinding rolls (HPGRs) are increasingly becoming a part of the hardrock
processing picture through their energy efficiency, the ability to induce microcracks and
preferential liberation, coupled with high throughput and high reduction ratio. Given that
the machine is still not regarded by many as an offtheshelf piece of process equipment,
there is work required to define guidelines for its use and to provide engineers with tools
they can use. This paper examines the current knowledge around the HPGR process perfor
mance and explores key relationships available to engineers, whilst considering current
approaches to simulation.
I NTRODUCTI ON
Highpressure grinding rolls (HPGRs) have struggled for acceptance into the hardrock
mining sector. Many of the issues that restricted their widespread use have now been
conquered, but it is still regarded as an “immature” technology. Why is this the case?
In contemplating an answer to the issue of the “immaturity,” the status of other
accepted technologies must be examined. As an example, the traditional compression
style conegyratory crushers can be considered. When a plant design is being assembled,
every wellequipped engineer will be able to turn to numerous rules of thumb associated
with these crushers—even without reference to textbooks or suppliers. The types of rules
referenced above include
Productsize distribution will be approximately 80% passing the closedside setting—
with poor applications dropping to 50%.
Centralized and circumferentially distributed feed is required to extract the best
performance.
Profile and condition of the crushing liners is critical to deliver the best distribu
tion of energy into the crushing chamber.
Lowbulkdensity feeds reduce throughput.
Maximum product bulk density is 1.9 to 2.1 t/m
3
for average limestone feedstock.
Secondary applications are power driven, whilst tertiary duties are pressure driven.
* Rio Tinto Technical Services, Perth, Western Australia
4 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
Mostly 5%–10% of the feedsize distribution is the maximum less than the closed
side setting—except with modern cones that are trying to generate interparticle
crushing.
Maximum feed size should not exceed 80% of the openside feed opening.
Feed moistures >4% should be avoided.
Given this type of knowledge, it is easy for the designer to determine the position
within the flowsheet and to then calculate the feed rates, type of feed arrangement, and
the pre and postclassification required. Why do these rules of thumb, or guidelines, not
exist for HPGRs? There are several reasons for this lack of clarity, namely:
Number and type of applications
Genesis of the HPGR concept
Industry position on technology
Existence of process models
First, there are very few actual, or operating, applications in hardrock duties. The
only hardrock applications that have been in existence for any length of time are
restricted to the diamond and iron ore (pelletfeed) sectors.
Another consideration is that the HPGR is a very rare breed of machine, in that its
development stemmed from fundamental research. Given the types and focus of early
publications, much was made of the nature of the interparticle breakage at the heart of
the technology. Obviously, given the groundbreaking nature of the invention, this focus
was fully justified, but it led—unfairly—to the HPGR being regarded as an academic
device searching for an industry application. The language used about the HPGR, and
unfamiliar terms such as “mdot” (denoting specific throughput), further led to an air of
mystique around the HPGR. Was it a crusher or a mill? Its place in the world was unclear.
Another element restricting the rate of application was the lack of process models.
Simulation is a large part of the flowsheet design exercise and this inevitably requires
process models to exist for each piece of equipment. In the case of the HPGR, much of
the effort was placed in scaleup procedures. Several organisations did produce process
models of HPGRs, but they have been fragmented in their acceptance. Currently, the
most complete model approach is that reported by Daniel and Morrell (2004), who have
developed an approach from the earlier model of Tondo (1997). It is interesting to note
that the Tondo model came out of the first major process study of HPGRs, namely the
AMIRA P428 that was completed in 1997.
If these points above are added to the naturally conservative stance of the mining
industry, this provides a view of why, even after mechanical/wear issues have been over
come, there is still a slow rate of acceptance.
As of today, the situation has changed. The features and benefits have become clear
to many practitioners, including
Energy efficiency
Preferential liberation at natural grain boundaries
Microcracking and enhanced extraction
Small footprint in terms of throughput and size reduction
Minimal vibration from machine into drive mechanisms and support structure
Of increasing importance is the energyefficiency issue. It was not too long ago that
the mining industry regarded energy consumption as somewhat of a side issue. The
Kyoto Protocol and the greenhouse debate changed this view forever (Ruben 2002).
HPGRS—CHARACTERISING AND DEFINING PROCESS PERFORMANCE 5
CRI TI CAL HPGR PARAMETERS
HPGR roll diameters typically range from 0.5 m to 2.8 m, depending on the supplies, and
roll widths vary from 0.2 m to 1.8 m. The aspect ratio of the rolls also varies as a function
of manufacturer. Typical HPGR throughput rates range from 20 to 3,000 tph, with
installed motor power as high as 3,000 kW per roll. The roll surface is protected with
wearresistant materials, and it has been these that have traditionally stymied HPGR
acceptance, but solutions are now in place (Maxton, Morley, and Bearman 2004).
When operating an HPGR, the two most important operating parameters are
Operating pressure
Roll speed
The two key operating parameters are inherently linked to the following:
Specific throughput
Specific pressing force
Maximum pressure between the rolls
Specific energy input
Detailed descriptions of the derivation and formulation of the parameters are given
in numerous texts, and as such, the following section provides only a précis of the critical
formulas, with some examples of actual relationships from testwork.
Specific Throughput
The specific throughput, mdot, is regarded by many as the key parameter for sizing the
rolls. Specific throughput is defined as the throughput (tph), divided by the roll diameter
(m), roll width (m), and the peripheral roll speed (m/s). For the purposes of brevity,
only the equations for this parameter are reported here. Further details are provided in
earlier works. (Schönert 1991). Part of its importance is that the equation allows com
parison between any size of rolls providing that the surfaces are the same.
m• = M/(D u L u u) (EQ 1)
where
M = throughput rate (tph)
D = roll diameter (m)
L = roll width (m)
u = roll speed (m/s)
m• = specific throughput (ts/hm
3
)
The throughput can also be calculated from the continuity equation as follows:
M = L u s u u u U
c
u 3.6 (EQ 2)
where
s = operating gap (mm)
U
c
= density of the product cake (t/m
3
)
Combining equations (1) and (2), one obtains:
m• = (s/D) u U
c
u 3.6 (EQ 3)
For a given material and operating conditions, the gap scales linearly with the diam
eter of the rolls, and hence the specific throughput can be assumed to be constant.
6 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
It should be noted that recent work by Daniel (2005) has examined the determina
tion of an equivalent diameter for piston press tests. Daniel proposes
D = (U
cf
u x
c
u x
d
) / ((U
cf
u x
c
) – (x
g
u U
g
)) (EQ 4)
where
U
cf
= feed bulk density, lightly compacted
x
c
= initial bed height in piston press
x
d
= displacement of piston
x
g
= final bed height (i.e., operating gap)
U
g
= density of product flake
This relationship has potential to assist in translating piston press results to engineering
parameters.
Variation in Throughput with Key Variables
Figure 1 shows the variation in the specific throughput as a function of the feed bulk
density. The relationship appears to be linear over the range of feeds tested. Given that
the specific gravity of the feed material is 2.85 t/m
3
, it would be unlikely that the loose
feed bulk density would exceed 1.8 t/m
3
; therefore, this graph suggests that the relation
ship is relevant over a vast majority of cases. It should be noted that throughput is high
est at the lowest pressure, with larger changes associated with the allin (high bulk
density) feed types. Figure 2 shows the type of linear increase in specific throughput
associated with increasing operating gap.
Figure 3 shows a plot of all tests versus the specific energy (power) consumed. It is
interesting to note that the data appear in two distinct clusters. The righthand cluster
consists purely of the allin feed types with no truncation of the feedsize distribution at
the lower end, whilst the lefthand cluster is formed from feeds with fines truncation.
1.45 1.40 1.50 1.55 1.60 1.65 1.70 1.75
250
230
210
190
170
150
m

d
o
t
,
t
s
/
h
m
3
Bulk Density, t/m
3
30 Bar
38 Bar
52 Bar
FIGURE 1 Variation in specific throughput as a function of feedbulk density for various operating
pressures using a pilotscale HPGR
HPGRS—CHARACTERISING AND DEFINING PROCESS PERFORMANCE 7
Specific Pressing Force
The specific pressing force is defined as the grinding force applied to the rolls (kN),
divided by the diameter (m) and width (m) of the rolls (Schönert 1988). The specific
pressing force has the unit of N/mm
2
.
F
sp
= F/(1,000 u D u L) (EQ 5)
where
F
sp
= specific pressing force (N/mm
2
)
F = applied grinding force (kN)
D = roll diameter (m)
L = roll width (m)
15 16 17 18 19 20 21
250
230
240
220
200
180
160
210
190
170
150
m

d
o
t
,
t
s
/
h
m
3
Operating Gap, mm
FIGURE 2 Variation in specific throughput as a function of operating gap using a pilotscale
HPGR at an operating pressure of 38 bar
40 90 140 190 240 290
60
56
58
54
50
46
42
52
48
44
40
m

d
o
t
,
t
s
/
h
m
3
Power, kW
FIGURE 3 Variation in specific throughput as a function of operating gap using a pilotscale HPGR
8 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
Ranges for specific pressing force vary considerably in the range 1–9 N/mm
2
, with
studded machines normally restricted to 5 N/mm
2
maximum pressure.
Specific pressing force is a key parameter used in scaleup and for comparison pur
poses between different machine sizes.
Maximum Pressure between Rolls
The maximum pressure applied to the material between the rolls has been estimated by
several workers, and it is generally assumed to be in the range of 40 to 60 times the spe
cific pressing force. It is generally accepted that the following equation (Schönert 1988)
holds true:
P
max
= F/(1,000 u D u L u k u D
ip
) (EQ 6)
where
P
max
= maximum pressure (MPa)
F = applied grinding force (kN)
D = roll diameter (m)
L = roll width (m)
k = material constant (0.18–0.23)
D
ip
= compression angle (6–10 degrees)
The parameter D
ip
can be calculated from the operating gap, with a detailed description
being given by Schönert and Lubjuhn (1990).
Specific Energy Input
The specific energy consumption of an HPGR is a familiar quantity to process engineers.
As with all other instances of the parameter, it is calculated from the net power input to
the rolls divided by the ore throughput rate.
It is important to note that specific energy input (kWh/t) is proportional to the spe
cific pressure applied to the rolls. Typical specific energy values for studded rolls range
from 1 to 3 kWh/t. As with all direct comminution devices, harder material will absorb
more energy compared to a softer material, for a given size reduction.
A rule of thumb is that the ratio of specific pressing force to specific energy input is
1.8–3:1, with this ratio decreasing towards 1.0 for finer comminution. Figure 4 shows
the type of response mentioned. In this case, the slope of the graph indicates a ratio of
1.5:1.
Specific energy consumption is markedly impacted by the feedsize distribution, as
illustrated in Figure 5. As the feed distribution lengthens (i.e., the bulk density
increases), the specific energy consumption drops.
The major impact of specific energy input is the product fineness. As with all commi
nution equipment, a point of diminishing returns will occur where extra energy does not
generate a commensurate increase in fineness. Figure 6 shows a range of energies and
fines generation. At the levels displayed in Figure 6, the point of diminishing returns has
not been reached.
SI MULATI ON OF HPGR PERFORMANCE
As with all modeling and simulation of process equipment, there is a sliding scale from
the simplest spreadsheetbased feedproduct transfer function at one end, through
empirical representations, to mechanistic models, and finally to detailed fundamental
descriptions. The key process issues that need to be estimated, or predicted, during the
design phase of a process plant are
HPGRS—CHARACTERISING AND DEFINING PROCESS PERFORMANCE 9
Throughput
Size reduction (product and oversize)
Power consumption (energy efficiency)
Required hydraulic stiffness
Target gap and operating pressure
Using these parameters, it is then possible to insert the HPGR into a flowsheet and make
sensible comparisons against other types of equipment and flowsheet configurations.
The additional benefits of preferential liberation and enhanced extraction must be
assessed via laboratory tests and incorporated with the full analysis.
1.0 1.5 2.0 2.5 3.0 3.5 4.0
0.5
0.7
0.9
1.1
1.3
1.5
1.7
1.9
2.1
S
p
e
c
i
f
i
c
E
n
e
r
g
y
C
o
n
s
u
m
p
t
i
o
n
k
W
h
/
t
Specific Pressing Force, MPa
FIGURE 4 Relation between specific energy consumption and specific pressing force using a
pilotscale HPGR
1.40 1.45 1.50 1.55 1.60 1.65 1.70 1.75
0.5
0.7
0.9
1.1
1.3
1.5
1.7
1.9
2.1
S
p
e
c
i
f
i
c
E
n
e
r
g
y
C
o
n
s
u
m
p
t
i
o
n
k
W
h
/
t
Feed Bulk Density, t/m
3
30 Bar
38 Bar
52 Bar
FIGURE 5 Relation between specific energy consumption and feed bulk density using a pilotscale
HPGR, at various operating pressures
10 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
Piston Press Testing and Ore Characterisation
The main ore characterisation tests for HPGR modeling are the pistonpress and drop
weight procedures. The dropweight test is the Julius Kruttschnitt Mineral Research Centre
(JKMRC) developed, singleparticle test and is used to examine areas in the HPGR
where the breakage is of a singleparticle nature. The pistonpress test is for characterisa
tion of the packedbed breakage zone in the HPGR. The purpose of the pistonpress test
is to generate an appearance function as per the dropweight test, but for packedbed
breakage. Hence, the pistonpress appearance function is used to characterise the pre
dominant breakage action in the HPGR.
The piston press can be used in an analogous manner to the traditional dropweight
test (i.e., breakage parameters and an appearance function can be determined).
In terms of the breakage characteristics, Table 1 provides an example of the compar
ison of the “b” parameters from the dropweight and pistonpress tests for material from
Argyle Diamonds. The immediate observation regarding the data in Table 1 is that the
piston press “b” parameters are higher than the singleparticle test, with the inference
being that the material appears softer in a packedbed environment.
Given the mode of compression (i.e., slow interparticle versus transient compres
sion), Table 1 could represent an efficiency factor relating the two forms of breakage.
Of more practical importance is that the use of the packedbed, pistonstyle test is
critical to the formation of a representative model of HPGR performance.
Application of Piston Press to Provide ConceptualLevel HPGR Performance Estimates
A variety of workers are now using pistonpress tests to research the action of HPGRs.
The press arrangement at Freiberg University has recently been used to test a copper ore
supplied by Rio Tinto. The aim of the tests is to determine the amenability of the ore to
HPGR treatment and to examine the use of the piston press for conceptuallevel evalua
tions. A series of tests at pressures from 80 to 320 MPa were undertaken with the results
presented in Table 2.
The maximum pressures reported in Table 2 were chosen to mimic those seen in the
HPGR pilot tests, and the results appear to be good approximations to those obtained
0.5 1.0 1.5 2.0
17
15
19
21
23
25
27
29
31
33
35
N
e
t
–
1
1
8
m
m
G
e
n
e
r
a
t
i
o
n
Specific Energy Consumption, kWh/t
FIGURE 6 Relation between specific energy consumption and fines generation using a pilotscale
HPGR
HPGRS—CHARACTERISING AND DEFINING PROCESS PERFORMANCE 11
from pilotscale HPGR work. Given this agreement, it is suggested that the piston press
be used to provide a conceptuallevel envelope of performance.
The suggested sequence is
1. Estimate mdot value from Equation (3), by substitution of the product flake
density, operating gap (final bed depth from piston press), and use of Equation
(4) to determine D.
2. Estimate throughput from the rearranged Equation (1), with assumed values for
roll diameter (D), roll width (L), and roll speed (u) relating to the desired scale
of equipment. These values can be determined in association with manufactur
ers. It should be noted that the scale independence of mdot, due to the linearity
of operating gap versus roll diameter, is a major assumption in this step.
3. Calculate the specific pressing force (F
sp
) from Equation (5) using the applied
grinding force from the piston press and the D and L values used above.
With these key parameters, it is possible to ensure that the size of rolls and the bearing
selection is correct. To estimate comminution performance:
Determine the specific energy consumption from assumed relationship with spe
cific pressing force. Values for the ratio F
sp
:W
sp
can be assumed to vary from 1:1
for very fine comminution through to 3:1 for very coarse duties. A value of 1.5:1,
as shown in Figure 5, is a good general value for moderate comminution of hard
ores. Care should be taken—although particlesize distribution is a major part of
the bulk properties that dictate the relationship between F
sp
and W
sp
, other fac
tors also influence the bulk behaviour including ore hardness, friction, and mois
ture (M.J. Daniel, personal communication, 2005).
Specific energy consumption is inherently linked to productsize distribution via the
traditional breakage and appearance type mapping employed in singleparticle drop
weight tests. Using the A and b parameters from the pistonpress test, these along
with the specific energy consumption can be substituted into the following equation:
t10 = A(1 – e
–b. Ecs
) (EQ 7)
where
t10 = percentage passing one tenth of the feed size
A and b = breakage characteristics from pistonpress tests
Ecs = specific energy consumption (kWh/t)
TABLE 1 Singleparticle breakage parameters
SingleParticle Test PackedBed Test
Sample b b
Unweathered lamproite 0.44 0.940
Siliceous waste 0.40 0.703
TABLE 2 Flake density results from pistonpress tests
Maximum Pressure, MPa Flake Density, t/m
3
77.24 2.14
157.29 2.32
230.53 2.32
310.98 2.38
12 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
Using the standard singleparticle relationships between t10 and the other size
distribution markers (i.e., t2, t4, t25, t50, t75), the entire size distribution of the
product can be generated. Theoretically, this is a combination of packed and single
bed approaches, but, as Tondo (1997) showed, the packedbed t10 versus t
n
rela
tionship underestimates size reduction in coarse sizes, compared to singleparticle
tests. Given that the variable edge effect generates coarser products, it is likely
that any underestimation from the packedbed parameters is simply an approxi
mation to the coarser edge comminution. This approach is backed up by the fact
that various workers have chosen to deal with this in different ways, whilst still
obtaining satisfactory results. Tondo (1997) used both singleparticle and packed
bed A and b parameters with separate appearance functions in his work, whilst
Daniel (2002) assumes a 10% split to edge and uses the singleparticle function
for all breakage with a t10 of 30.
This conceptuallevel approach, although not rigorous, helps engineers to obtain a
“feel” for HPGR performance and at least obtain a quick, firstpass estimate of the opera
tional envelope. It should be noted that no account is taken of precrush or edge effects.
Analysis of this technique suggests that both throughput and product fineness are over
stated, but as the scale of machine increases, the discrepancy lessens. This reduction in
error with scale can probably be assigned to the decreasing proportion of machine per
formance impacted by edge effects.
Detailed HPGR Modeling
For a more complete treatment of performance estimation in a modeling sense, true
models are required. The work of Daniel and Morrell (2004) represents the most com
plete current description. The basis for their work is shown schematically in Figure 7.
Daniel and Morrell outline information required for modeling, as shown in Table 3.
To undertake the simulation, there are a variety of parameters relating to the break
age and classification of material in the three different zones as defined in Figure 7. The
main parameters are listed in Table 4.
This extremely comprehensive treatment is then used in a verification and scaleup
scheme procedure; full details can be found in works by Daniel and Morrell (2004).
CONCLUSI ONS
There is an increasing body of knowledge around the application of HPGRs in hardrock
duties. In terms of selection and sizing, much has already been written, particularly by
the suppliers. For process performance, the increasing application is allowing the devel
opment of some rules and shortcuts that can allow a firstpass evaluation of HPGRs for
flowsheet purposes—a critical element on the pathway to engineering acceptance. In
many ways, this paper seeks to provide a pragmatic engineering basis for the assessment
of HPGR performance. This message was also the theme expressed by Klymowsky and
Liu (1996), where they sought a Bond workindex analogy for HPGRs. There is no doubt
that a standardized, accepted HPGR “work index” would be a great boost to HPGR
acceptance.
Beyond these engineering views of HPGRs, the detailed modeling and simulation of
HPGR process performance is finding common ground, and workers have developed
comprehensive approaches that provide the required accuracy and resolution.
Assimilation of this understanding within the industry, along with simpler measures
and guidelines, will accelerate HPGR implementation, particularly now that mechanical
issues are predominantly of historical interest only.
HPGRS—CHARACTERISING AND DEFINING PROCESS PERFORMANCE 13
ACKNOWLEDGMENTS
The author gratefully acknowledges all practitioners in the field of HPGR technology
that have contributed to this paper through discussions. In particular, the discussions
and advice from Mike Daniels, JKMRC, showed that a considerable amount of effort is
still being applied to the issue of HPGR application.
Entry Zone
SingleParticle
Breakage
Centre Zone
PackedBed Breakage
Edge Effect Single
Particle Breakage
Product from HPGR
Feed to HPGR
After Tondo 1997.
FIGURE 7 Schematic representation of Daniel and Morrell model
Source: Daniel and Morrell 2004.
TABLE 3 Model inputs and outputs
Measured Input Measured Output Calculated Output
Sample mass Working gap (x
g
) Measured throughput (Q
m
)
Roll diameter (D) Flake thickness (x
gf
) Calculated throughput (Q
calc
)
Roll width (L) Flake density (q
g
) Specific energy (Ecs)
Roll speed (U) Productsize distribution (measured) Specific force (F
sp
)
Bulk “compacted” density (qc) Batch process time Critical gap (x
c
)
Feedsize distribution Working pressure (pw), power (kW) Productsize distribution
Source: Daniel and Morrell 2004.
TABLE 4 Model parameters
Fixed Default Parameters Critical Model Parameters
t10p, t10e—breakage for edge and precrusher Kp(HPGR)—power coefficient (compression zone)
K1p, K2p, K3p—precrusher model parameter t10h—breakage for compression zone crusher
K1e, K2, K3—edgecrusher model parameter
K1h, K2h, K3h—compression zone parameter
Split factor (c)
Kp(edge)—power coefficient (edge)
14 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
REFERENCES
Daniel, M.J. 2002. HPGR model verification and scaleup. Master’s thesis. Brisbane,
Australia: Julius Kruttschnitt Mineral Research Centre, Department of Mining and
Metallurgical Engineering, University of Queensland.
———. 2005. Paper submitted to Randol Pacific Gold Forum, Perth, Australia.
Daniel, M.J., and S. Morrell. 2004. HPGR model verification and scaleup. Minerals
Engineering 17:1149–1161.
Klymowsky, I.B., and J. Liu. 1996. Towards the development of a work index for the
roller press. In Comminution Practices, SME Symposium 1996. S.99/105.
Maxton, D., C. Morley, and R. Bearman. 2004. A quantification of the benefits of high pressure
rolls crushing in an operating environment. Minerals Engineering 16:827–838.
Ruben, E.S. 2002. Learning our way to zero emissions technologies. IEA Zero Emission
Technologies Strategies Workshop, Washington, DC, March 19.
Schönert, K. 1988. A first survey of grinding with highcompression roller mills.
International Journal of Mineral Processing 22:401–412.
———. 1991. Advances in comminution fundamental, and impacts on technology. Pages 1–21
in Proceedings of the XVII International Mineral Processing Congress. Volume 1.
K. Schöenert, ed. Ljubijana, Yugoslavia.
Schönert, K., and U. Lubjuhn. 1990. Throughput of high compression roller mills with
plain and corrugated rollers. Pages 213–217 in 7th European Symposium on
Comminution.
Tondo, L.A. 1997. Phenomenological modelling of a high pressure grinding roll mill.
Master’s thesis. Brisbane, Australia: Julius Kruttschnitt Mineral Research Centre,
Department of Mining and Metallurgical Engineering, University of Queensland.
15
HighPressure Grinding Rolls—
A Technology Review
*
Chris Morley
†
ABSTRACT
The development of highpressure grinding rolls (HPGRs) technology is reviewed, with an
emphasis on aspects relevant to hardrock comminution. Case histories are investigated and
lessons learned are discussed in the particular context of the application of the device as a
supplement to, or replacement for, conventional crushing and semiautogenous milling circuits.
The potential for the more widespread use of this technology as a comminution method
in hardrock processing is examined. The use of the technology as a metallurgical tool is
addressed, and future flowsheet concepts are introduced that make progressively greater use
of the energy efficiency of HPGRs.
I NTRODUCTI ON
Highpressure grinding roll (HPGR) technology has its genesis in coal briquetting in the
early twentieth century, but it was not until the mid1980s that it was adopted for com
minution applications, when it was applied in the cement industry to treat relatively eas
ily crushed materials. Since then, it has been applied to progressively harder, tougher,
and more abrasive materials, generally successfully, but as would be expected, not with
out some developmental problems.
Machines are now also in use in the following applications:
Kimberlites in secondary, tertiary, and recrush roles
Iron ores for coarse crushing, autogenous mill pebble crushing, regrinding, pre
pelletising, and briquetting
Limestone crushing
Concentrates fine grinding
Gold ore crushing
Other prospective applications include phosphates, gypsum, titanium slag, copper
and tin ores, mill scale, and coal.
Hardrock operations that use HPGRs as an alternative or supplement to conventional
comminution devices include Argyle, Diavik, Premier, Kimberley, Jwaneng, Venetia and
* Updated from the original paper, “HPGR in Hard Rock Applications,” published in Mining Magazine,
September 2003, www.miningmagazine.com
† Fluor, Australia
16 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
Ekati (diamonds), CMHLos Colorados, CVRD, Empire and Kudremukh (iron ore), and
Suchoj Log (gold ore). Hardrock operations to have considered using HPGR and conducted
pilot testing include Mt. Todd, Boddington, and KCGM, all in Australia. A full plant trial
of an HPGR was conducted on a particularly arduous duty at Cyprus Sierrita between
1995 and 1996; and, more recently, HPGR has been piloted at Lone Tree, Nevada, in the
United States, and Amplats Potgietersrus in South Africa. Currently, HPGRbased com
minution plants are under construction at Bendigo, Australia (gold), and Cerro Verde,
Peru (copper), and at final feasibility study stage for the Soledad Mountain, California
(heap leach gold/silver), and Boddington, Australia (gold/copper), projects.
There are currently three recognised manufacturers of HPGR machines, namely Polysius
(a Thyssen Krupp company), KHD Humboldt Wedag AG, and Köppern, all based in Germany.
THE TECHNOLOGY
Machine Design
The HPGR machine comprises a pair of counterrotating rolls mounted in a sturdy frame.
One roll is fixed in the frame, while the other is allowed to float on rails and is positioned
using pneumohydraulic springs. The feed is introduced to the gap between the rolls and
is crushed by the mechanism of interparticle breakage.
The pressure exerted by the hydraulic system on the floating roll largely determines com
minution performance. Typically, operating pressures are in the range of 5–10 MPa, but can
be as high as 18 MPa. For the largest machines, this translates to forces of up to 25,000 kN.
The rolls are protected with wearresistant surfaces, and the ore is contained at the
roll edges by cheek plates.
Technology Motivators
Generally, the primary motivation for the use of the HPGR as a comminution alternative
is its energy efficiency when compared to conventional crushers and mills. This improved
Courtesy of Köppern.
FIGURE 1 Coal briquetting press—early twentieth century
HIGHPRESSURE GRINDING ROLLS—A TECHNOLOGY REVIEW 17
efficiency is due to the determinate and relatively uniform loading of the material in the
HPGR compression zone, whereas the loading in conventional crushers and (particu
larly) tumbling mills is random and highly variable, and therefore inefficient.
The most energyefficient method of breakage is the slow application of pressure to
individual particles to cause structural failure, such that the energy lost as heat and noise is
minimised. However, until a device is invented that can perform this task on a commercial
scale, the HPGR remains the most energyefficient comminution technology available.
A major operating cost in conventional semiautogenousbased comminution circuits
treating hard and abrasive ores is that of grinding media. One effect of the use of HPGR
based circuits is that semiautogenous mill grinding media is eliminated, and while ball
mill media costs typically are slightly greater (due to the increased transfer size from
HPGRs), the overall media savings are typically of the same order of magnitude as the
energy savings.
In addition to its energy and media benefits, the HPGR may be regarded as a metallur
gical tool offering improved gravity, flotation and leach recoveries, and enhanced thickening,
filtration, and residue deposition performance.
These effects can be attributed to the phenomenon of microcracking of individual
progeny particles due to the very high stresses present in the HPGR compression zone.
Microcracking occurs predominantly at grain boundaries and so increases liberation and
lixiviant penetration, while the effective reduction in milling work index caused by
microcracking reduces overgrinding and slimes generation.
In addition to being ore dependent, the extent of microcracking is a direct function
of the operating pressure—and therefore energy input—of the HPGR, and in any given
operation, the benefits of microcracking must be weighed against the incremental power
required to achieve those benefits.
The HPGR’s mechanism of interparticle breakage is particularly beneficial in the pro
cessing of diamondbearing kimberlites, which undergo a form of differential comminution
Courtesy of KHD Humboldt Wedag AG.
FIGURE 2 HPGR machine
18 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
whereby the host rock is shattered while the diamonds are liberated undamaged—provided,
of course, that the diamonds are smaller than the operating gap of the HPGR. This effect
is also of benefit in the treatment of gold ores containing coarse gravityrecoverable
gold grains, which would be flattened in conventional tumbling mills and rendered more
difficult to recover.
Technology Status
The HPGR, considered a mature technology in the cement industry, is now the norm
rather than the exception in modern diamond plant design and is becoming common in
iron ore processing, particularly in the field of pellet feed preparation.
However, although some of the current diamond and iron ore applications can be
regarded as hardrock duties, HPGR is regarded by many as unproven in true hardrock
mining, and this perception is reinforced by the experience at Cyprus Sierrita in 1995–
1996. This application is widely considered to have been unsuccessful because it did not
lead to a commercial sale; however, the fact that the comminution performance of the
machine was impressive is not in dispute. The difficulties experienced related to the behav
iour of the wear surfaces, and many valuable lessons were learned from this operation
regarding the precautions necessary in circuit design and unit operation for the protec
tion of the studded roll surfaces and the successful application of HPGR technology.
Courtesy of Polysius AG.
FIGURE 3 Cone crusher product particle (conventional crushing)
Courtesy of Polysius AG.
FIGURE 4 HPGR product particle (internal microfractures after Polycom treatment)
HIGHPRESSURE GRINDING ROLLS—A TECHNOLOGY REVIEW 19
The following is a summary of the more important issues arising from observations
of the HPGR operation at Cyprus Sierrita and elsewhere:
The technology is approaching a level of maturity allowing it to be seriously con
sidered for hardrock applications.
HPGRs are sensitive to segregation and tramp metal in the feed.
Mechanical availability of HPGRs is relatively high, and loss of machine utilisation
in hardrock applications is predominantly wear related.
The smooth and profiled hardmetal roll surfaces commonly used in the cement
sector are unsuitable for hard abrasive ores. Instead, the more recently intro
duced autogenous wear layer concept should be used, in which crushed ore is
captured in the interstices between metal carbide studs or tiles.
On hardrock applications in particular, HPGRs are sensitive to feed top size,
which ideally should not exceed the roll operating gap. Oversize material in the
feed can lead to stud breakage.
Roll wear surfaces may be formed as segments or as cylindrical sleeves or tyres.
Segments may be used for softer ores and lower operating pressures, while tyres
are recommended for hardrock duties and higher pressures as they present a uni
form, uninterrupted wear surface to the ore and thereby avoid the preferential
wear that occurs at segment boundaries. In addition, tyres are easier to fabricate
than segments and so are less expensive.
Tyres involve long changeout times due to the need to remove the roll assemblies
from the mainframe, while segments can be changed in situ. Some machine
designs aim to minimise changeout times for tyres by allowing roll assembly
removal without the need for dismantling of the feed system and superstructure.
Wear of the roll edges and cheek plates (the static wear plates used to contain the
ore at the roll edges) remains an issue, and development in this area is ongoing. A
FIGURE 5 Cyprus Sierrita installation
20 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
few operations use rock boxes (chutes at the edges of the rolls) instead of cheek
plates, allowing part of the feed material to flow around the rolls and so relieve
the pressure on, and wear of, the roll edges. This does, however, introduce the
disadvantage of passing uncrushed feed to product.
Technology Hindrances
Hindrances to the adoption of HPGRs in hardrock processing include
The generally conservative nature of the mining industry
A perception of high cost, particularly of the replacement wear parts in abrasive
applications
Uncertainties regarding the reliability of modeling and scaleup from laboratory
or pilot operations to commercial installations
A lack of definition of the requirements for robust flowsheet design of an HPGR
based comminution circuit.
Of these, it is generally acknowledged that high wear rates constitute the major obstacle
to the ready acceptance of the technology in hardrock applications. However, the HPGR
can prove a costeffective comminution device, even when the high cost and frequency of
replacement of wear surfaces in highly abrasive duties are considered.
Scaleup procedures have been the subject of many technical publications and
should now be considered reliable. They are mentioned here only briefly for the sake of
completeness. The characteristics of HPGRs that have a significant impact on flowsheet
design will be considered as the main emphasis of this analysis.
SCALE OF OPERATI ON
A common perception is that a project must be of relatively large scale before the use of
HPGRs can be justified. However, HPGR units of almost any size can be produced (up to
the current practical unit capacity limit of about 2,200 t/h), and this technology deserves
serious consideration over a much wider range of plant capacities than might initially be
imagined.
Ultimately, HPGRs can be justified if they offer benefits to metallurgical perfor
mance and/or project economics, and the potential for such benefits can usually be
assessed at the prefeasibility study phase by conducting preliminary tests. The manufac
turers have test facilities in Germany, and smallscale laboratory facilities are available at
various locations globally. Pilotscale machines are available at several research facilities
in Perth, Western Australia, and a Polysius mobile pilot unit used for trials at an opera
tion in North America in 2003 was subsequently relocated to South Africa for evaluation
on a hardrock mining operation.
THE MANUFACTURERS AND THEI R DESI GNS
Polysius, KHD, and Köppern are widely represented globally, but the machines are man
ufactured exclusively at their respective facilities in Germany.
Polysius favours a highaspectratio design—large diameter, small width—while KHD
and Köppern prefer a lowaspect ratio. The highaspectratio design is inherently more
expensive but also offers an intrinsically longer wear life for a given application, as the
operating gap is larger and the roll surfaces are exposed to a correspondingly smaller
proportion of the material processed. The highaspectratio design also produces a
coarser product due to the greater influence of the edge effect; however, this difference is
relatively slight, particularly with larger units. Nevertheless, for closedcircuit applications,
HIGHPRESSURE GRINDING ROLLS—A TECHNOLOGY REVIEW 21
this additional coarseness does increase the circulating load and tends to offset the wear
life benefits, as a higher total throughput is required for the same net product.
The use of tungsten carbide studs to create an autogenous wear layer on the roll sur
face is covered by a patent held by KHD, from whom this technology is available under
license.
Both Polysius and KHD have experience with minerals applications and studded roll
technology, and are able to supply machines with capacities of up to about 2,200 t/h.
Although Köppern has limited minerals experience, their HPGRs are successfully operat
ing in the cement industry. For highly abrasive materials, Köppern recommends HPGRs
fitted with their Hexadur wear protection.
The Hexadur surface comprises hexagonal tiles of a proprietary abrasionresistant
material set into a softer matrix, which wears preferentially in operation, allowing the
formation of an autogenous wear protection layer at the tile joints. The tiles and matrix
material are fully bonded together and to the substrate in a hightemperature, highpressure
furnace. By contrast, KHD’s studs are inserted into drilled holes. As a result, the tiles are
inherently stronger and more resistant to breakage due to oversize ore or tramp metal.
Köppern supplies patterned and profiled surfaces in both segment and tyre format,
whereas Hexadur is generally available only in tyre format due to the dimensional control
difficulties inherent in the fabrication and furnace treatment of segments. However,
research into the commercial production of Hexadur segments is ongoing.
Meanwhile, the maximum Hexadur roll diameter available currently (and for the
foreseeable future) is 1.5 m, constrained by furnace dimensions. This constraint limits
Köppern’s unit capacity to about 1,000 t/h for hardrock comminution applications using
Hexadur. However, Köppern also offers machines with studded roll surfaces supplied by
KHD, effectively lifting this capacity constraint.
Data of Test Units:
Diameter of Rolls: 0.71 m
Width of Rolls: 0.21 m
Speed of Rolls: 0.29–1.10 m/s
Top Feed Size: 16–35 mm
Diameter of Rolls: 0.30 m
Width of Rolls: 0.07 m
Speed of Rolls: 0.2–0.9 m/s
Top Feed Size: 8–12 mm
REGRO
ATWAL
LABWAL
Courtesy of Polysius AG.
FIGURE 6 Polysius test facility
22 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
Köppern has an established design in which the ends of the mainframe hinge out
wards to allow the roll assemblies to be removed without disturbing the feed system and
superstructure. This allows roll changeout times for tyre replacement of about the same
duration as for insitu segment changeout. Polysius also offers a design that allows rapid
roll assembly removal, but without the need for a hinged frame design. In more recent
developments, KHD has unveiled a rapid changeout concept to be offered on new
Courtesy of KHD Humboldt Wedag AG.
FIGURE 7 Studded roll wear surface
Courtesy of Köppern.
FIGURE 8 Hexadur wear surface for hardore comminution
HIGHPRESSURE GRINDING ROLLS—A TECHNOLOGY REVIEW 23
machines and which can be retrofitted to existing units, and Köppern has introduced
their “Cframe” design that allows the removal of both roll assemblies from one end of
the frame, so offering a maintenance advantage over their earlier design.
KHD uses cylindrical roller bearings that allow the choice of grease or circulating oil
lubrication systems, as there is no relative movement between the bearings and seals.
Polysius and Köppern use greaselubricated, selfaligning spherical roller bearings.
OPERATI NG CHARACTERI STI CS
There are many factors to be considered when specifying an HPGR and selecting an
appropriate flowsheet for a given application. The following subsections summarize the
more important issues.
Courtesy of Köppern.
FIGURE 9 Köppern HPGR
Courtesy of Köppern.
FIGURE 10 Köppern hinged frame
24 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
Ore Characteristics
The compressive strength of the material to be crushed determines the amount of useful
energy that can be absorbed by the material, which in turn dictates the bearing and
motor sizes required for a given duty.
With studded roll wear surfaces, the compressive strength of the ore, in combination
with the feed particle top size and operating pressure, will largely determine the probability
of stud damage—the higher the values of each of these variables, particularly when they
occur together, the higher the likelihood of incurring stud damage. Ongoing development
of stud technology is aimed at reducing the sensitivity of the studs to these variables.
The abrasion index of the material being crushed will determine the wear rate (as
distinct from the breakage rate) of the studs, as well as that of the substrate metal. For
example, the wear life at the iron ore operations at Los Colorados and Empire are about
14,000 and 10,000 hours, respectively, while those at the Argyle and Ekati diamond
mines were about 4,000 hours initially, but increased to 6,000–8,000 hours and beyond
with ongoing development of stud and edge protection configurations.
HPGRs are not generally suitable for the treatment of highly weathered ores or feeds
containing a large proportion of fines. (This of course does not apply to applications
where all the feed material is fine, such as fine grinding of concentrates.) Fine and
weathered material tends to cushion the action of the rolls and so reduces the efficiency
of comminution of the larger feed particles. For example, Argyle bypasses its primary
HPGRs when very fine ore is being mined. On these ore types, the fine or weathered
material should be removed by prescreening if HPGR treatment of the coarser compo
nent is required.
HPGRs are not generally suitable for comminution of feeds containing excessive
moisture, which tends to cause washout of the autogenous layer on studded rolls and
increases slippage on smooth rolls. In both cases, accelerated wear is the result. For
example, Ekati bypasses the –4+1 mm feed fraction around the HPGR when the prevail
ing ore type results in inherently high moistures.
Specific Pressure
The specific pressure (specific press force) is the force (Newtons) divided by the appar
ent (or projected) area of the roll—that is, the product of roll diameter and length:
specific pressure (N/mm
2
) = force (N)/(D (mm) u L (mm))
Typical practical operating values are in the range of 1–4.5 N/mm
2
for studded roll
surfaces and up to 6 N/mm
2
for Hexadur. The required specific pressure determined in
tests is used for scaleup of the required operating hydraulic pressure for the commercial
unit.
Specific Energy Input
The specific energy input (SEI) is the net power draw per unit of throughput:
specific energy input (kWh/t) = net power (kW)/throughput (dry t/h)
Typical operating values are in the range of 1–3 kWh/t. In general, a given ore will
absorb energy up to a point beyond which little additional useful work (i.e., size reduc
tion) is achieved—a zone of diminishing returns is approached.
For equivalent size reduction, a hard, competent ore of high compressive strength
will result in a higher SEI than a softer ore of low compressive strength.
The energy input is governed by the hydraulic pressure, of which it is a roughly linear
function. Generally, specific energy input in coarse crushing applications is numerically
HIGHPRESSURE GRINDING ROLLS—A TECHNOLOGY REVIEW 25
about one half to one third of the specific pressure, so that a specific pressure in the typi
cal operating range of 3–4.5 N/mm
2
can be expected to correspond to a specific energy
input of 1–2.5 kWh/t. In finegrinding duties, this ratio is typically higher—for example,
a ratio of 1.05 applies at the Kudremukh pellet feed operation.
The best method of determining the optimum specific energy is to conduct tests to
derive a graph of product fineness against specific energy. The graph generally displays
an initial steep slope that flattens out to approach the horizontal at high SEI values (e.g.,
3.5–4.5 kWh/t). The optimum SEI can then be selected.
Microcracking
Although the size reduction graph frequently enters an area of diminishing returns with
increasing specific energy, it has been demonstrated on some ores that the reduction in
effective work index due to microcracking (also known as microfracturing or microfis
suring) does not always display the same tendency. As a result, it may be beneficial from
an overall comminution energy perspective to operate at a higher specific energy than
corresponds to the optimum for size reduction in the HPGR stage, to maximise the bene
fits of microcracking. In this regard, the final grind size must also be taken into account,
as the effects of microcracking are felt more in the coarser fractions, so that an applica
tion with a coarse grind will benefit more than one with a fine grind.
It is important to conduct sufficient tests to quantify the optimum point of increased
fines generation and reduced product work index, to ensure an HPGR is specified that is
capable of transmitting the necessary power.
Feed Top Size
For hardrock applications, the feed top size is a critical variable in the successful operation
of an HPGR crusher. For smooth rolls, too large a top size results in reduced nip efficiency,
slippage, and accelerated wear; for studded rolls, tangential forces at the roll surface due
to early nipping—effectively causing singleparticle breakage by direct contact with the
roll surfaces—can cause stud breakage.
Constraints on feed top size have been related in the literature both to roll diameter
and to operating gap. Figures of up to 7% of roll diameter and three times the gap have
been quoted as appropriate limits on feed top size, even though the latter ratio implies
some direct contact of the larger particles with the surfaces of both rolls, leading to single
particle breakage.
These figures are now considered much too optimistic in hardrock applications,
and it is generally accepted that, to minimise the likelihood of stud breakage, feed top
size should not exceed the expected operating gap. This will normally demand a closed
circuit crushing operation upstream to ensure this top size is positively controlled. For
softer materials, this rule can be relaxed—for example, some kimberlite operations suc
cessfully treat opencircuit secondary crushed products with top size–gap ratios of about
1.8–2.0 using studded rolls.
By interpolation, ratios of around 1.3–1.5:1 are tolerable when treating ores of mod
erate hardness. Where uncertainty exists regarding ore hardness categorisation, it is con
sidered prudent to adopt a ratio of close to 1:1 initially, and then relax this incrementally
if and when it is established that stud breakage is not an issue.
As a guide, the directcontact nip angle (for singleparticle breakage and possible
stud damage) is normally in the range of 10˚ to 13˚ while interparticle breakage com
mences at angles of 5˚ to 7˚. By using a scale diagram of an HPGR unit of a given roll
diameter, and showing these angles and an appropriate operating gap, estimates can be
made of the particle size above which singleparticle breakage is likely to occur and
below which interparticle breakage commences.
26 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
Unit Capacity
The capacity of an HPGR is fundamentally a function of the ore characteristics. Capacity
is generally expressed in terms of specific throughput m
x
(mdot), which is a function of
the roll diameter, length, and peripheral speed:
m
x
(t·s/m
3
·h) = throughput (t/h)/(diameter (m) u length (m) u speed (m/s))
The value of m
x
is determined in pilot tests and used in scaleup to the commercial
unit, taking into account the change in the relative proportions of product from the cen
tre of the rolls and from the edges where poorer comminution occurs (the “edge effect”),
and also whether the commercial unit is to be operated with cheek plates or rock boxes
for roll edge protection.
In addition to its fundamental relationship to the ore characteristics, the value of m
x
is a function of many variables. The following should be regarded as general trends for
the majority of ores, rather than as statements of universal fact—there will always be the
exception that proves the rule:
Ore hardness—m
x
increases with ore hardness.
Specific pressure—m
x
decreases slightly with increasing pressure.
Roll surface—m
x
increases with increasing “texture” of the roll surface, due to the
reduced slip (increased kinetic friction) and improved nip between the rolls. Thus,
smooth rolls give the lowest values, with profiled surfaces in the midrange, and
studded surfaces the highest (typically about 50% higher than for smooth rolls).
Roll speed—for smooth rolls, m
x
decreases with roll peripheral speed, so that
actual throughput increases with increasing speed but at a progressively dimin
ishing rate due to increased slippage. The effect is much reduced with profiled or
studded rolls due to the inherently higher kinetic friction of these surfaces.
Feed top size—the available evidence is not conclusive, but it appears that m
x
might increase slightly with an increase in feed top size.
Feed bottom size—m
x
decreases significantly as feed bottom size is increased.
Thus, the highest value of m
x
occurs with a fullfines feed, and this value
decreases progressively as the fines cutoff or truncation size is increased. This is
due to the increased voidage in the truncated feeds, which results in a lower back
pressure on the rolls and a consequent reduction in the operating gap.
Feed moisture—for moisture levels greater than about 1%, m
x
decreases with
increasing moisture due to the replacement of solids with water in the compacted
product flake; higher moisture levels can result in excessive slippage and ultimately
to washout of the autogenous layer on studded rolls. Below 1% moisture, there is
some evidence of reduced m• values with studded rolls due to the difficulty in
generating and maintaining a competent autogenous wear layer with very dry feeds,
as the crushed product is too friable to form a compacted layer between the studs.
Operating Gap
The operating gap is directly related to the unit capacity, all else being equal, so “gap”
can be interchanged with m
x
in the above analysis. Depending on the application, the
ratio of operating gap to roll diameter will normally lie in the range of 0.010 to 0.028.
Circuit Capacity
The capacity of an HPGR circuit, as distinct from the unit capacity discussed above, is
obviously a function of the circuit design. Of the above variables, the feed bottom size
HIGHPRESSURE GRINDING ROLLS—A TECHNOLOGY REVIEW 27
is particularly relevant in this regard, as a truncated feed necessarily implies the pres
ence of a screen or other classification device upstream of the HPGR.
It has been noted that capacity decreases with truncated feeds; however, the capac
ity of the circuit would increase if the amount of fines removed from the HPGR feed
exceeded the reduction in HPGR unit capacity. Whether this occurs in practice remains
the subject of some debate (and in any event is probably ore specific), but recent model
ing of pilot test data for two prospective applications indicates that this is the case, and
this is supported by the limited evidence available in the literature.
However, an increase in circuit throughput achieved in this way may be offset by a
decrease in product fineness and/or reduced microcracking such that, depending on the
downstream processing route, a fullfines HPGR feed may be preferable to a truncated
feed. For any given application, the more efficient flowsheet can be determined only by
comprehensive tests and modeling, but where doubt exists, the circuit should, if possi
ble, be designed with the flexibility to operate with full fines or truncated feed to allow
circuit performance to be optimised. This flexibility normally comprises the prescreening
of the feed and a facility to recycle to HPGR feed a portion of either the HPGR product or,
where the HPGR operates in closed circuit with a screen, the screen undersize.
Product Sizing
As noted earlier, product fineness increases with operating pressure (and therefore
power), generally up to a point of diminishing returns. It has been observed elsewhere
that it is more energy efficient to operate an HPGR at low pressures and in closed circuit
with a screen, so that less energy is wasted on compacting the product. However, this
generally would require more or larger HPGRs to handle the increased circulating load.
Also, it is not clear whether the analysis included the cost of conveying the increased cir
culating load of screen oversize.
Product fineness generally decreases with increasing “texture” of the roll surface; so
smooth rolls give the finest product, with profiled surfaces in the midrange and studded
surfaces the coarsest. This is due to the reduced slip between the rolls and the ore, giving
a higher throughput for a given power draw. For the same product fineness, therefore,
a studded or profiled roll machine would have to be operated at higher pressures than a
smooth roll unit. However, the effect is relatively small, and the benefits of profiled or
studded rolls usually outweigh the reduced product fineness. Furthermore, the effect
appears to be ore specific, and some operations (e.g., Jwaneng) have recorded an
increase in fineness with studded rolls compared to smooth rolls.
Increasing roll speed leads to a reduced product top size and improved F
50
/P
50
reduction ratio, without significantly changing the fine end of the sizing spectrum.
A slight mismatch or differential in roll speeds has been found to enhance grinding
performance, and though this could be considered intuitively plausible, it might also be
expected that adopting this as a deliberate control strategy could lead to increased roll
surface wear rates due to this imposed speed differential. This effect is therefore
regarded as being of academic interest rather than practical significance.
Product sizing is largely independent of feed moisture. Product sizing is a function
of roll aspect ratio. A high aspect ratio gives an inherently coarser product for the follow
ing reasons:
The proportion of edge material in the product is greater.
The pressure peak in the compression zone is lower (for a given specific pressure).
However, the overall effect is generally fairly modest.
The shape of the HPGR product sizing curve is dissimilar to that of conventional
crushers, so that for products with nominally the same P80, the HPGR product contains
28 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
considerably more fines below this size than from a conventional crusher. The implica
tions of this are that, where the product is delivered to, for example, a ball milling oper
ation, mill capacity will be greater when treating HPGR product than predicted by the
standard Bond equation. Milling power requirements are thus reduced by both the sizing
of the HPGR product and the microcracking of the product particles, and are therefore
best determined by pilot testing.
Roll Surface Wear
Increasing roll speed increases turbulence in the feed material and slip of feed against
the roll surfaces, leading to elevated wear rates. This should generally be a concern only
at the top end of the practical speed range. In this respect, Polysius traditionally uses a
rule of thumb to the effect that the peripheral speed of the rolls (in meters per second)
should not exceed roll diameter (in meters), although Köppern does not support this
view and regularly nominates speed–diameter ratios of up to 1.3. KHD also uses these
higher ratios for their smallerdiameter machines but generally uses <1.0 for larger units
and on coarse crushing applications. More recently, Polysius has proposed that it is the
angular velocity rather than peripheral speed that should be used as the roll speed selec
tion criterion, and that maximum speed should be in the range of 18 to 20 rpm for fine
grinding applications and 20 to 22 rpm for typical hardrock coarse crushing duties.
High moisture levels lead to significant increases in wear. It is believed that this
could be due to a combined erosion/corrosion effect analogous to that observed in Nord
berg WaterFlush cone crushers.
In recent studies involving tough, abrasive ores, it was found that wear rates were
significantly higher with truncated feeds than with full fines or untruncated feeds, to the
extent that wear life considerations heavily outweighed the energy efficiency advantages
that had previously been established for these ores using truncated feeds. This illustrates
the importance of conducting comprehensive tests to ensure that decisions made on
flowsheet selection are well informed.
Roll surface wear rates for studded rolls can vary as the operating life of the wear
surface progresses. In one application, the wear rate was observed to increase with time
from 0.006 Pm/t after 200 operating hours to 0.015 Pm/t after 1,000 hours, after which
a plateau was established in the wearrate curve.
It is believed that this effect is due to an initial imbalance in the wear rates of the
studs and the substrate. In the case in point, it would appear that the stud protrusion
above the new roll surface was too small for this particular duty, so that the substrate ini
tially wore more rapidly than the studs. Presumably, had the stud protrusion instead
been too great, then stud wear would have been more rapid initially and declined there
after. This, however, was not demonstrated.
In either case, overall wear rate stabilises when the two components of wear—stud and
substrate—are in equilibrium. The important point is that roll surface wear life should
not be computed from initial wear rates. The wearrate curve must be plotted and the
equilibrium plateau established before wearlife predictions are made.
The “bathtub” effect is a welldocumented phenomenon whereby the central zone of
the rolls wears at a greater rate than the edges, forming a concave wear pattern. For
smooth rolls, this can lead to a requirement for regular grinding of the edges to maintain
parallel roll surfaces and avoid touching of the rolls at the edges with the correct nominal
gap in the central zone. For studded rolls, harder studs can be used in the central zone to
give a more uniform wear pattern across the roll surface. However, harder studs are also
more brittle, and stud breakage, as distinct from stud wear, can become a problem. The
optimum combination of stud hardness levels for the central and edge studs in a given
application must be established by trial and error. Normally, studs of lower hardness are
HIGHPRESSURE GRINDING ROLLS—A TECHNOLOGY REVIEW 29
selected for the initial setup to minimise the incidence of stud breakage, which is more
difficult to accommodate than stud wear. After evaluating initial performance, adjust
ments to stud hardness may be made. Several iterations might be needed to achieve the
optimum configuration. Ekati and Argyle have both achieved significantly improved
results using this approach.
The differential wearrate effect is also well documented—it appears that the floating
roll typically wears at a slightly higher rate than the fixed roll, though the reasons for this
have not been fully investigated. It is believed that the effect is caused by the additional
applied kinetic forces imparted to the floating roll. However, recent experience at a pilot
operation has shown the reverse trend. In any event, the effect is small and of little prac
tical significance.
In a recently commissioned installation, a more irregular wear pattern was observed
on the fixed roll relative to the floating roll, although the overall wear rate for the floating
roll was higher. The reasons for this comparatively irregular wear pattern are not clear, but
it is suspected that it is due to the effect of the presence of a feedregulating gate, which
presents the feed stream preferentially towards the fixed roll. This may result in an
increased level of turbulence in the vicinity of the feed gate tip. The need for the gate at
this operation is not proven (as the variablespeed drives provide adequate turndown), and
it is to be removed as a trial, during which any change in wear patterns will be recorded.
Tramp Steel
Theoretically, the HPGR is equipped to handle tramp steel in that the bearing arrange
ment allows skewing of the floating roll and the hydraulic system is able to relieve exces
sive pressures. However, particularly with larger units, the inertia of the rolls and their
very brief exposure to tramp metal in the compression zone generally results in damage
to the roll surface instead of, or as well as, the relieving action of the floating roll.
Repair of roll damage can be expensive and operationally disruptive, and flowsheet
design should endeavour to locate the HPGR in an intrinsically noncontaminated flow
stream, or ensure that a comprehensive and practical tramp metal detection and removal
system is included. Such a system should preferably be automatic, with contaminated
ore bypassed around the HPGR or rejected from the circuit. It is important to minimise
the need for operator intervention and process interruption.
0
10
20
30
40
50
60
A
T
W
A
L
W
e
a
r
R
a
t
e
,
g
/
t
Truncated 50%
Truncated
Full Feed Truncated 50%
Truncated
Full Feed
MATERIAL 1 MATERIAL 2
FIGURE 11 Impact of feed truncation on roll wear
30 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
Extrusion
Extrusion is a phenomenon whereby the HPGR product stream emerges from the com
pression zone at a speed greater than that of the roll surface. This is similar to the well
known extrusion effect observed in metal rolling and occurs when the product flake
expands as it leaves the compression zone and the applied pressure reduces to zero. This
expansion is typically in the order of 2% to 5%, and the resultant slippage can be the
cause of increased roll wear. Extrusion has generally been observed to increase with roll
diameter, but also has been recorded with pilotsized machines. Extrusion also increases
with applied pressure and feed moisture, and the evidence available suggests the effect
is most pronounced at a roll peripheral speed of about 1.5 m/s. It is most noticeable with
smooth rolls, and decreases markedly with studded or profiled surfaces.
Product Flake Formation and Treatment
The HPGR product emerges from the compression zone as a compacted cake or flake.
The coherence of the flake is a function primarily of the ore type and moisture content,
and also of the operating pressure of the machine. Generally, competent flakes are pro
duced with softer materials or those with a high clay or moisture content—kimberlites,
for example—while hard, primary ores tend to produce fragile flakes, even at relatively
high moistures and pressures.
Depending on flake competency and downstream processing requirements, a dedi
cated unit operation for deagglomeration of the flake product could be required, and this
is a significant consideration in flowsheet development. Kimberlite flakes normally must
be intensively deagglomerated in wet rotarydrum scrubbers, and then screened to
ensure efficient removal of fines before downstream processing, usually in a heavy
media separation operation. By contrast, the flake in a hard, primaryore application
might require no separate deagglomeration, being adequately broken down by handling
in chutes and bins and on conveyors, so that acceptable efficiencies are achieved in nor
mal screening.
The need for, and nature of, a dedicated deagglomeration step in the comminution
flowsheet can normally be assessed by testing. KHD has developed a standard flake com
petency test specifically designed to determine whether separate deagglomeration is
required ahead of further processing.
FLOWSHEET OPTI ONS
The flowsheet for a given ore is driven by the requirements of the process and consider
ation of the above HPGR characteristics. In particular, the possible need for a controlled
feed top size, fines recycling, and separate deagglomeration will have a significant effect
on the formulation of a practical and robust flowsheet.
It is important that the appropriate amount of testing be conducted to determine
flowsheet design requirements. Alternatively, in the absence of adequate tests, a conser
vative approach must be taken to flowsheet design—that is, it must be assumed that top
size control, fines recycling, and deagglomeration will all be needed. This of course has
the potential to impose significant and possibly unnecessary cost penalties on any
project, and a comprehensive test programme generally represents excellent value for
the money in this context.
The selection of flowsheets considered here focuses on alternatives to conventional
hardrock crushing, screening, and milling circuits, either as greenfield projects or as retrofits
to existing operations for the purposes of debottlenecking or plant expansion. HPGR may
also be considered as a beneficial metallurgical tool in heapleach applications.
HIGHPRESSURE GRINDING ROLLS—A TECHNOLOGY REVIEW 31
SAG Mill Precrushing
In this circuit, a portion of the semiautogenous grinding (SAG) mill feed is precrushed in
a secondary crusher followed by an HPGR (with prescreening for topsize control, if nec
essary) before rejoining the main SAG mill feed stream. The total SAG mill feed is therefore
correspondingly finer while some coarse particles are retained in the mill feed as media.
This is suitable for both new and existing operations and has the potential to increase
mill throughput by 50% or more. (The Troilus gold operation in Canada introduced a
screen on the SAG mill feed stream and recorded a circuit capacity increase of about
35% using only a secondary crusher to precrush the –125+25 mm middlings fraction.)
An alternative to this arrangement is to deliver the HPGR product to the SAG mill
discharge screen, so that finished product and material of ball mill feed size bypasses the
SAG mill.
These circuits are flexible in that milling can continue (albeit at reduced rates) with
the precrushing circuit out of service. Also, if an HPGR bypass facility is provided, the
secondary crushing component of the precrushing circuit can continue to operate when
the HPGR is inactive.
SAG Mill Pebble Crushing
Further size reduction of the pebble fraction in an SABC (semiautogenousballcrusher)
circuit can be achieved by passing the pebble crusher product through an HPGR, thus
increasing circuit capacity. Alternatively, the conventional cone crusher may be replaced
entirely by an HPGR, provided the pebble top size is small enough.
In a variation, the pebble crusher and HPGR can be operated in closed circuit with
the screen undersize delivered to the ball milling circuit. This circuit can be used to
opencircuit the SAG mill when this is the circuit bottleneck.
Primary Crushing
Secondary Crushing
Screen
HPGR
Semiautogenous
Milling
Screen
Cyclone
Pebble Crushing
Ball Milling
U/S
U/S
O/F
O/S
O/S
U/F
FIGURE 12 SAG mill precrushing to SAG mill feed
32 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
These circuits have the disadvantage of exposure of the HPGR to tramp steel in the
SAG mill discharge. Also, the use of HPGR in the pebblecrushing circuit can usually be
justified only when pebble arisings are a relatively large proportion of SAG mill new feed.
Multistage Crushing and Ball Milling
In this circuit, the HPGR is used in the tertiary crushing stage immediately ahead of the
ball mills. This can be applied in new projects or as a retrofit to increase crushing capac
ity. However, for hard rock, it is important that the secondary crushing stage be operated
in closed circuit to control HPGR feed top size, and this must be borne in mind when
considering this circuit as a retrofit.
Depending on whether deagglomeration is indicated, the HPGR product screens
may be operated dry (no deagglomeration required) or wet (mild deagglomeration). In
the latter case, the dilute screen undersize slurry must be delivered to the mill sump
rather than mill feed. Where intensive deagglomeration is required, the entire HPGR
product is delivered to the mill, the mill discharge screened, and the screen oversize
returned to the HPGR. In this case, it may be preferable to adopt a twostage milling cir
cuit, with the primary mill designed to minimise pebble generation, so minimising the
return of moist material to the HPGR.
This is inherently less efficient than delivering a controlled feed top size to the milling
circuit, and it might be more efficient to use a dry deagglomerator on the HPGR product,
such as a hammer mill or vertical impactor, followed by conventional dry screening.
OpenCircuit HPGR with Edge Recycle
This option obviates the need for fine screening of the HPGR product and instead uses a
dividing chute below the HPGR to separate the highly reduced centre product from the
Primary Crushing
Secondary Crushing
Screen
HPGR
Semiautogenous
Milling
Screen
Cyclone
Pebble Crushing
Ball Milling
U/S
U/S
O/F
O/S
O/S
U/F
FIGURE 13 SAG mill precrushing to ball mill feed
HIGHPRESSURE GRINDING ROLLS—A TECHNOLOGY REVIEW 33
coarser “edge” material, as typically practised on test units and at a few commercial
operations. The centre product is delivered to downstream processing, while the edge
material is returned to HPGR feed.
This arrangement would typically be used where energy efficiency was not of para
mount importance, such as heapleach applications.
ANALYSI S OF TECHNOLOGY BENEFI TS
The metallurgical benefits of HPGRs have been discussed earlier in a qualitative sense.
These are highly orespecific and should be determined by the appropriate tests.
Primary Crushing
HPGR
Semiautogenous
Milling
Screen
Screen
Cyclone
Pebble Crushing
Ball Milling
U/S
U/S
O/F
O/S
O/S
U/F
FIGURE 14 SAG mill pebble crushing
Primary Crushing
HPGR
Semiautogenous
Milling
Screen
Screen
Cyclone
Pebble Crushing
Ball Milling
U/S
Middlings
U/S
O/F
O/S
O/S
U/F
FIGURE 15 Opencircuit SAG mill
34 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
Likewise, the energy benefits of HPGRs must be properly quantified to allow a realis
tic assessment of HPGRbased circuit options, as comminution energy generally is a
major component of operating costs in hardrock applications.
In a recent study, a comparison was drawn between a highcapacity conventional
SABC circuit and a threestage crushing/singlestage ball milling circuit with both sec
ondary and tertiary crushing stages operating in closed circuit and with the HPGR as the
tertiary step. As part of the analysis, various intermediate circuits were also evaluated in
which the HPGR played a progressively greater role.
Primary Crushing
HPGR Screen Secondary Crushing
Cyclone
Ball Milling
Middlings
U/S
O/F
O/S
U/F
FIGURE 16 Threestage crushing, closedcircuit HPGR
Primary Crushing
HPGR
Screen
Screen
Secondary Crushing
Cyclone
Primary
Ball Mill
Secondary
Ball Mill
U/S
U/S
O/F
O/S
O/S
U/F
FIGURE 17 Threestage crushing, opencircuit HPGR
HIGHPRESSURE GRINDING ROLLS—A TECHNOLOGY REVIEW 35
The comparison showed the following conclusions:
The energy efficiency of the circuit increased with the proportion of comminution
performed by the HPGR.
The specific capital cost (i.e., cost per unit capacity) of the HPGR/ball mill circuit
was 26% lower than that of the SABC circuit. (However, the capacity of the
HPGRbased circuit in this example was higher than that of the SABC basecase
circuit, so this differential figure would be lower in a direct comparison. This has
been supported in subsequent studies on various projects in which the capital
costs of an HPGRbased circuit were found to be about the same or slightly
greater than for the equivalent SAGbased circuit of the same capacity.)
The HPGR/ball mill circuit was 28% more energy efficient than the SABC circuit.
Overall operating costs for the HPGR/ball mill circuit and downstream plant were
22% lower than for the SABC circuit.
Project implementation time was significantly reduced for the HPGR option due
to the removal of the longdelivery SAG mill.
As a result of these conclusions, project viability was considerably enhanced—in fact, it
was determined that, in the absence of HPGR in some part of the comminution circuit,
project viability was at best marginal.
A sensitivity analysis was conducted in which the wear life of the HPGR roll surfaces
was reduced from the 4,000 hours predicted by the manufacturers to a very conservative
figure of 2,000 hours. This had the effect of increasing overall operating costs by only
5%, meaning that the HPGRcircuit operating costs were still 17% lower than those of
the SABC circuit.
COMPARI SON WI TH CONVENTI ONAL TECHNOLOGI ES
Autogenous grinding (AG) and SAG technologies displaced multistage crushing and rod/
ball milling circuits as they were simpler and offered lower capital and overall operating
costs, even though they were often demonstrably less efficient in the use of comminution
energy. AG and SAG mills were also ideal for handling wet, sticky, clayrich, and oxidised
Primary Crushing
HPGR
Screen Secondary Crushing
U/S
O/S
Edge
Centre
Heap
Cement and Water
FIGURE 18 Opencircuit HPGR with edge recycle
36 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
ores, allowing the elimination of the traditional washing plant normally required for
such materials.
With the progressive global depletion of easily treated ores, harder, tougher, and
more abrasive primary ores are being targeted for treatment, and energy efficiency will
become steadily more important from both economic and environmental perspectives.
These ores may be well suited to SAG mill treatment in the context of media competency,
but this treatment path is grossly inefficient in the application of energy for size reduction.
The circuit developed in the study referred to above represents a return to the tradi
tional multistage crushing/ball milling circuit. The difference now is that, with HPGR in
the tertiary crushing step, energy efficiencies are elevated to such a degree that overall
operating costs are much lower than for the equivalent SAG mill circuit. When the poten
tial for lower specific capital costs also is considered, the additional circuit complexity of
the HPGRbased plant can more readily be justified.
Just as the traditional multistage crushing/ball milling circuit has survived in
selected applications, semiautogenous milling will of course also survive and remain the
appropriate choice for some ores. It is believed, however, that HPGR circuits represent
the next generation of hardrock comminution plant design, as semiautogenous milling
did several decades ago.
VI SI ON FOR THE FUTURE
The HPGR is the most energyefficient comminution device currently available to the miner
als processing plant designer, and the focus must be on the development of both machine
and flowsheet to maximise the proportion of total comminution performed by the HPGR.
The initial objective must be to minimise the top size of ball mill feed by reducing
HPGR product screen aperture and recirculating progressively more material to the HPGR.
This will entail changing to wet screening as the separation size falls below about 6 mm,
and this in turn will impact circuit design philosophies, as there will be no opportunity to
stockpile mill feed.
As the mill feed top size falls, there might be some merit in operating tertiary HPGRs
in open circuit and introducing quaternary HPGR crushing to handle screen oversize.
The associated moisture would, however, be detrimental to this process, and some form
of blending with dry material might be necessary.
Ultimately, with very fine mill feeds, the number and size of conventional wet
screens will become unmanageable, while the ball mills will trend ever smaller. The next
step is to abandon screens and ball mills entirely and operate HPGRs in closed circuit
with air separators, with the final product repulped and fed directly to flotation or other
downstream processes.
The technology for this type of circuit already exists and is in operation. An example
is the use in Europe of KHD HPGRs and air separators for the production of dryground
limestone for use in a fluegas desulphurisation process. Typical airseparator perfor
mance in this type of application is 90 Pm P90, while the finest separation is claimed to
be around 20 Pm P90. Product size is adjustable, and the P80 grind sizes of 75 to 150 Pm
common in minerals processing should be readily achievable.
CONCLUSI ON
HPGR technology holds the promise of significant improvements in comminution energy
efficiency in hardrock applications when compared to SAGbased circuits. Properly
designed HPGRbased circuits offer the potential of significant savings in comminution
energy requirements and overall operating costs when compared to SAGbased circuits.
Further energy savings are envisioned as progressively more of the comminution load is
HIGHPRESSURE GRINDING ROLLS—A TECHNOLOGY REVIEW 37
performed by HPGR, culminating in the generation of final ground product by air classi
fication of HPGR product and the elimination of ball milling.
ACKNOWLEDGMENTS
In addition to the technical papers listed in the bibliography, from which much of this
material is drawn, the sourcing of study information and operational data from Bodding
ton Gold Mine and Argyle Diamonds, respectively, is gratefully acknowledged, as are the
contributions from the manufacturers, Polysius, KHD, and Köppern.
BI BLI OGRAPHY
AMIRA Project P428. 1996. Application of High Pressure Grinding Rolls in Mineral Process
ing (Overview). Report P428/11.
Austin, L.G. 1990. Ball Mills, SemiAutogenous Mills and High Pressure Grinding Rolls.
University Park, PA: Penn State University Press.
Austin, L.G., Trubeljal, M.P., and von Seebach. 1995. Capacity of highpressure grinding
rolls. Minerals and Metallurgical Processing (May): 65–73.
Baum, W. 1998. HPGR as a processing tool for gold & copper leaching, flotation and
gravity separation. Paper presented at Raw Material Technology Seminar, Tucson, AZ.
Baum, W., and Knecht, J. 1994. Optimizing refractory and oxide gold ore operations with
high pressure grinding rolls. Paper presented at SME Annual Meeting, Albuquerque,
NM, February 14–17.
———. 2000. HPGR as a processing tool for gold and copper leaching, flotation and gravity
separation. Paper presented at 2nd Annual Crushing and Grinding in Mining Con
ference, Johannesburg.
Baum, W., Patzelt, N., and Knecht, J. 1996. The Use of High Pressure Grinding for Opti
misation of Copper Leaching, SME, Phoenix, AZ, March.
———. 1997. Metallurgical benefits of high pressure roll grinding for gold and copper
recovery. Paper presented at SME Comminution Practices Symposium, Denver, CO.
Bleifuss, R.L., Goetzman, H.E., Benner, B.R. and Zhong, S. 1997. Evaluation of a high
pressure roller press for taconite comminution. Paper presented at SME Comminu
tion Practices Symposium, Denver, CO.
Brachthäuser, M., and Kellerwessel, H. 1988. High pressure comminution with roller
presses in mineral processing. Pages 209–219 in Reprints XVI International Mineral
Processing Congress.
Bruce, R. 1992. Practical experience gathered with new segmented liners. KHD Humboldt
Wedag Symposium.
Burchardt, E. 1998. HPGR: A metallurgical tool for the diamond industry. Paper pre
sented at Proceedings, Randol Diamond Focus, Vancouver, BC, November.
Dunne, R., Goulsbra, A., and Dunlop, I. 1996. High pressure grinding rolls and the effect
on liberation: Comparative test results. Paper presented at Randol Gold Forum ’96,
Olympic Valley, CA.
FischerHelwig, F. 1992. Current state of roller press design. KHD Humboldt Wedag
Symposium.
Fuerstenau, D.W., and Asoke, D.E. 1997. Energy Optimisation in High Pressure Roll/Ball
Mill Hybrid Grinding Systems. Pages 115–128 in Proceedings of XX IMPC. Volume 2.
Aachen, Germany.
38 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
Fuerstenau, D.W., Shukla, A., and Kapur, P.C. 1991. Energy consumption and product
size distributions in chokefed highcompression roll mills. International Journal of
Mineral Processing 32:59–79.
Kellerwessel, A.M. 1996. High pressure particle bed comminution—state of the art, appli
cation, recent developments. Engineering and Mining Journal (February): 45–52.
Kellerwessel, H. 1993. High Pressure Particle Bed Comminution—Principles, Application, Test
ing and ScaleUp, Details of Equipment Design. Germany: KHD Humboldt Wedag AG.
Kellerwessel, H., and Oberheuser, G. 1995. Scaleup of roller presses. Pages 67–70 in
Proceedings of 19th International Mineral Processing Congress.
Klymowsky, I.B., and Liu, J. 1997a. Modelling of comminution in a roller press. Pages 141–
154 in Proceedings of XX IMPC. Volume 2. Aachen, Germany.
———. 1997b. Towards the development of a work index for the roller press. Paper pre
sented at SME Comminution Practices Symposium, Denver, CO.
Klymowsky, R. 1998. Roller press installations in iron ore. World Mining Equipment (April).
———. 2003. High Pressure Grinding Rolls for Minerals. McGill Presentation.
Klymowsky, R., and Cordes, H. 1999. The modern roller press—practical applications in
the ore and minerals industry. Aufbereitungs Technik 40(Nr 8).
Klymowsky, R., Patzelt, N., Burchardt, E., and Knecht, J. 2002. Selection and sizing of
high pressure grinding rolls. Pages 636–668 in Mineral Processing Plant Design. Pro
ceedings, Volume I. Edited by A. Mular, D. Halbe, and D. Barratt. Littleton, CO: SME.
Klymowsky, R., and van der Meer, F. 1999. Roller press grinding—new applications for
the iron ore industry and updates on the latest developments. Paper presented at
5th Annual Iron Ore and Steel Summit, Sydney, Australia, February 22–23.
Lim, W.I.L., Campbell, J.J., and Tondo, L.A. 1996a. The effect of rolls speed and rolls sur
face pattern on high pressure grinding rolls performance. Minerals Engineering
10:401–419.
———. 1996b. Extrusion effects in the high pressure grinding rolls. Paper presented at
SME Comminution Practices Symposium, Denver, CO.
Lubjuhn, U., and Schoenert, K. 1993. Material flow in the acceleration zone and
throughput of high pressure roller mills. Pages 161–168 in XVIII International Min
eral Processing Congress.
Maxton, D. 2003. Argyle Breaks Down HPRC Barriers. Australia’s Mining Monthly (July).
Maxton, D., Morley, C., and Bearman, R. 2002. Recrush HPRC Project—The benefits of
high pressure rolls crushing, crushing and grinding. Kalgoorlie, Australia,
October 30–November 1.
———. 2003. A quantification of the benefits of high pressure rolls crushing in an operating
environment. Minerals Engineering 16(9).
McIvor, R.E. 1997. High pressure grinding rolls—a review. Paper presented at SME Com
minution Practices Symposium, Denver, CO.
McIvor, R.E., Dowling, E.C., Korpi, P.A., and Rose, D.J. 2001. Application of high pressure
grinding rolls in an autogenouspebble milling circuit. Pages 194–201 in Proceedings
SAG 2001. Volume III.
Morrell, S., Lim, W., Shi, F., and Tondo, L. 1997. Modelling of the HPGR crusher. Paper
presented at SME Comminution Practices Symposium, Denver, CO.
Mörsky, P., Klemetti, M., and Knuutinen, T. 1995. A comparison of high pressure roller
mill and conventional grinding. Pages 55–58 in Proceedings of 19th IMPC.
HIGHPRESSURE GRINDING ROLLS—A TECHNOLOGY REVIEW 39
Murilo Mourão, J., and Schwalm, T. Pelletizing at CVRD—three decades of Brazilian
German cooperation. MPT International 3:36–45.
Norgate, T., and Weller, K.R. 1991. Options for incorporating high pressure grinding rolls
into comminution circuits. Pages 19–24 in Fourth Mill Operators’ Conference.
Norgate, T.N., and Weller, K.R. 1994. Selection and operation of high pressure rolls circuits
for minimum energy consumption. Minerals Engineering 7(10):1253–1267.
Parker, B., Rowe, P., Lane, G., and Morrell, S. 2001. The decision to opt for high pressure
grinding rolls for the Boddington expansion. SAG 2001, Vancouver, BC.
Patzelt, N., Klymowsky, R., Burchardt, E., and Knecht, J. 2001. High pressure grinding
rolls in AG/SAG mill circuits. Pages 107–123 in Proceedings SAG 2001. Volume III.
Patzelt, N., Knecht, H., and Baum, W. 1995. Case made for highpressure rollgrinding in
gold plants. Mining Engineering (June): 524–529.
Patzelt, N., Knecht, J., Burchardt, E., and Klymowsky, R. 2000. Challenges for high pres
sure grinding in the new millennium. Pages 47–55 in 7th Mill Operators Conference.
Pietsch, W. 1995. Roller presses—versatile equipment for mineral processing. Pages 59–66
in Proceedings of 19th International Mineral Processing Congress.
Schönert, K. 1979. Energetische Aspekte des Zerkleinerns spröder Stoffe. ZKG 32:1–9.
———. 1988. A survey of grinding with high pressure roller mills. International Journal of
Mineral Processing 22:401–412.
Schönert, K., and Flügel, F. 1980. Zerkleinerung spröder minerale im hochkomprimierten
gutbett. European Symposium of Particle Technology.
Schwechten, D., and Milburn, G.H. 1990. Experiences in dry grinding with high pressure
compression roller mills for end product quality below 20 microns. Minerals Engi
neering 3(1–2):24–34.
Sotillo, F., and Finch, E. 1998. On the beneficiation of high dolomitic pebbles: Exploring
the use of a high pressure roll mill. SME Preprint 9891. Littleton, CO: SME.
Strasser, S. 1992. Current state of roller press technology. KHD Humboldt Wedag Symposium.
Thomsen, L.G. 1997. Operational performance of grinding rolls at Cyprus Sierrita Corpora
tion. Chapter 15 in Comminution Practices. Edited by K.S. Kawatra. Littleton, CO: SME.
Thurat, B., and Wolter, A. 1989. Roller presses for fine grinding of cement raw meal, coal
and lime. (Reprint). Zement, Kalk, Gips Journal 6: 278–285.
Van der Meer, F.P. 1997. Roller press grinding of pellet feed—experiences of KHD in the
iron ore industry. AusIMM Conference on Iron Ore Resources and Reserves Estima
tion, Perth, Australia, September.
Van der Meer, F.P., and Schnabel, H.G. 1997. The effect of roller press grinding on ball
mill energy. Erzmetall 50(Nr 9).
Weller, K.R. 1995. New grinding technologies. Pages 17–26 in Randol Gold Forum.
Weller, K.R., and Norgate, T.E. 1989. The place of high pressure rolls grinding in fine
comminution. Paper presented at Sagsem ’89, Murdoch University and AusIMM,
Perth, Australia.
Weller, K.R., Norgate, T.E., Sterns, U.J., and Housley A.G. 1990. The response of some
Australian ores to high pressure rolls grinding. Pages 821–835 in 7th European Sym
posium on Comminution.
Westermeyer, C.P., and Cordes, H. 2000. Operating experience with a roller press at the
Los Colorados ore dressing plant in Chile. Aufbereitungs Technik 41(Nr 11):497–505.
41
Some Basics on HighPressure
Grinding Rolls
Eberhard W. Neumann
*
ABSTRACT
Highpressure grinding roll (HPGR) technology was first introduced on an industrial scale in
the mid1980s for the grinding of clinker and raw material in cement making. Notwithstanding
early difficulties as would be expected with any new technology, HPGRs quickly proved to be
a very economical addition for comminution processes because of lower energy consumption
and easy integration into existing conventional systems for capacity enhancement.
Subsequent to successes in cement making, HPGRs soon began to be utilized in a variety
of applications in the mining industry to reduce power consumption, increase product yield,
and add production capacity in existing milling systems. With regard to abrasive wear,
material properties, and selection of process parameters, the natural materials used in mining
pose much greater challenges as compared to the relatively constant artificial product,
cement clinker.
This paper describes the principle of comminution in a compacted material bed, pres
sure application to the packed bed, power draw, energy density in the compression zone, and
technical solutions to particular HPGR requirements. The paper will also reflect on typical
process flowsheets with HPGR integration in mining applications.
THE VERY BASI CS
Any rotating machine, from the ancient grain mill (G. Gudat, personal communication)
(Figure 1) to modern highpressure grinding rolls (HPGRs), draws this amount of power:
(EQ 1)
where
P = power (kW)
T = torque (kNm)
Ȧ = angular velocity (sec
–1
)
F = force F (kN)
L = lever length (m)
Ȟ
c
= circumferential speed (m/sec)
R = radius (m)
* Köppern Equipment Inc., Charlotte, North Carolina
P T Z u F L
v
c
R
 u u = =
42 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
This, of course, is true for all unit systems. Using, however, N (Newton) for force, m
(meter) for length, and sec (second) as the time unit gives the power directly in watts:
Our ancestors in the desert already knew that one particle could be cracked by
applying pressure. If our desert mill were to grind more, it either had to grind longer or
the donkey had to run faster. If it ran faster, it increased the Ȧ, thereby increasing power
because the radius stayed the same. If the grind was to be finer, they had to use a heavier
stone and the donkey would have noticed because he would have had to pull harder,
thereby increasing F, and also power, if he could maintain his speed.
The Principle of HighPressure Grinding
This paper presents one type of machine that accomplishes this purpose, the highpressure
grinding roll, or HPGR.
An HPGR consists of two counterrotating rolls (Figure 2), one of which is in a fixed
location while the other one, a moving roll, is supported off hydraulic cylinders acting
against a hydropneumatic spring that allows horizontal movement of the moving roll.
The material to be ground is fed into the gap between the rolls. A small feed hopper,
which always contains material, is installed above the rolls. The surfaces of the rotating
rolls grip the material and pull it into and through the roll gap. For a certain arc length
the material will slip on the roll surfaces until it reaches a point where the circumferen
tial velocity and the material velocity are equal. This is called the nip zone.
Because the horizontal distance between two locations on the roll surfaces
decreases, the material is exposed to an increasing pressure. As it moves down, it reaches
the maximum pressure approximately in the narrowest gap at the rolls’ centerlines. This
pressure is so high that the material particles fracture. The majority of the feed particles
are smaller than the narrowest gap. The maximum pressure p
max
is therefore exerted on
a material bed rather than on single particles. This effect is called interparticle or
packedbed comminution. Not every particle that fractures need come into contact with
the roll surfaces.
FIGURE 1 Ancient packedbed pressure grinding mill
watt
Nm
sec
 =
SOME BASICS ON HIGHPRESSURE GRINDING ROLLS 43
Some Formulas
The specific throughput (O. Knobloch, personal communication) is independent of
machine size and allows up or downscaling for a given feed material and roll surface:
(EQ 2)
where
= throughput
D = roll diameter
L = roll length
The total throughput in mass or weight per time unit is
(EQ 3)
where
s = working gap
ȡ
c
= cake density
The maximum pressure p
max
(K. Schoenert, personal communication) is
(EQ 4)
where c is a constant depending upon machine and material parameters. The required
p
max
to achieve a certain comminution result needs to be determined for each material.
Typical values range from 140 to 300 MPa.
S
lip
Grinding Roll
Grinding Roll
Nip
Cake
Feed Material
PRE
p
max
FIGURE 2 Highpressure grinding—material slip and nip between the rolls
m
·
m
·
M
·
D L v
c
u u
 =
M
·
M
·
M
·
s L v
c
U
c
u u u =
p
max
c
F
s
 =
44 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
Figure 3 shows the force diagram for the grinding force F, the normal force F
n
, and
the tangential force F
u
. The intersection of the three forces is located inside the nip zone.
The total power draw at both roll shafts is
(EQ 5)
with the force acting angle E (K. Schoenert, personal communication).
THE HPGR BONUS
The HPGR bonus describes the ability of the HPGR to replace or supplement kilowatts or
kilowatthours per ton supplied by the reference mill, typically a tube mill, with HPGR
kW or kWh/t:
(EQ 6)
If Wspec
HPGR
< Wspec
mill
, then B > 1. When the HPGR is added to an existing mill circuit
the bonus becomes
(EQ 7)
Wspec
mill
is the kilowatthours per ton for the existing mill before and after adding the
HPGR. The energetic advantage of the HPGR is that the bonus is greater than 1, which
means that 1 HPGR kW or kWh/t will do more grinding work than 1 mill kW or kWh/t.
In other words, adding an HPGR to an existing mill circuit will reduce specific energy
F
u
F
u
F
n
F
R
Grinding Roll Grinding Roll
FIGURE 3 Force diagram, force acting angle ȕ
P 2F v
c
E sin u u =
B
Wspec
mill
Wspec
HPGR
 =
B
Wspec
millbefore
Wspec
millafter
–
Wspec
HPGR
 =
SOME BASICS ON HIGHPRESSURE GRINDING ROLLS 45
consumption and increase production. The savings in specific power consumption at the
mill main drives is
(EQ 8)
The overall system savings are somewhat lower because of added equipment, such as
conveying or screening. Bonus values depend mainly upon feed material, circuit configu
ration, and reference mill.
Some typical values (from Köppern operating and test data) are
Cement clinker 1.8–2.5
Blast furnace slag 2.5–3.8
Limestone 1.7–2.0
Kimberlite 1.6–2.0
The Machine Design
Figure 4 views an HPGR from the side where the hydraulic system is located. The two
grinding rolls are suspended with selfaligning roller bearings in bearing blocks, which
are mounted in the machine frame. Each roll has its own drive train with planetary gear
reducers. Torque arms are provided to neutralize the countertorques generated by the
drives. This particular machine design features a hinged frame that swings open for easy
roll exchange. The machine has the following characteristics:
The energy density in the nip zone is quite high, about 400 times compared to a ball
mill. Correspondingly high are the loads and stresses on the rolls, especially on the roll
surfaces. Figure 5 shows three basic roll designs.
The roll surfaces are of particular importance not only from the wear aspect but also
for their capability to draw in the material. Figure 6 shows a studded roll surface
(according to sources at KHD Humboldt Wedag, Cologne, Germany) where material
builds up between the studs, thereby forming an autogenous wear protection and pro
viding a rough surface for good friction. Figure 7a shows a worn, welded hard surface,
and Figure 7b shows metallurgical powderbased wear elements applied by a hot isostatic
pressure process. These are just three examples; there are several others that have been
developed over the years.
HPGR Applications in Mining
Figures 8 through 11 show some typical applications for HPGRs to increase throughput
and lower specific energy consumption of a grinding circuit. Figure 8 shows the HPGR
after a semiautogenous grinding (SAG) mill for pebble grinding. The HPGR product is
returned either to the SAG mill or to the screen. In Figure 9, the HPGR is located after the
secondary crusher. The HPGR product is screened, and the oversize is returned to the
HPGR. Figure 10 has the HPGR as singlepass pregrinder in front of the milling plant.
Roll diameter 2,140 mm
Roll length 1,300 mm
Circumferential speed variable, max. 1.58 m/sec
Installed power 2 u 1,300 kW
Installed grinding force 19,500 kN
Throughput 850 tph cement clinker
Bonus achieved 2.1
'P Wspec
millbefore
Wspec
millafter
Wspec
HPGR
+ – =
46 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
The HPGR can apply about 2.3 kWh/t to the feed material. At a bonus of, for example,
1.8, it would add 2.3 u 1.8 = 4.14 kWh/t ballmill equivalent to the grinding circuit. The
HPGR can also be located after heavymedia separation (HMS) with or without a crusher
grinding the wet oversize (Figure 11).
The best HPGR location for a given mine needs to be decided for each case; bottleneck
identification, space availability, conveying distances, power grid, and other conditions
must be considered.
CONCLUSI ONS
Versatility and efficient energy utilization have made highpressure grinding an estab
lished comminution technology in the minerals processing industries. Relatively simple
formulae can be used to describe and understand the underlying mechanics of HPGRs.
Careful consideration must be given to the overall grinding process in order to take full
advantage of the special features offered by highpressure comminution.
Hinged Frame Roll
Hydraulic
System
Torque
Arms
Gear
Reducers
FIGURE 4 HPGR assembly at the workshop (view from the hydraulic side)
(a) Solid Roller
1
2 2
3
Grinding Surface
Bearing Journal
(b) TireShaft Roller
Roller Core
1 4
(c) Segments
Roller Core
1 5
Segments
FIGURE 5 Press tools
SOME BASICS ON HIGHPRESSURE GRINDING ROLLS 47
Tungsten Carbide Studs Autogenous Wear Protection
FIGURE 6 Studded roll surface
Circumferential Wear Grooves Hexadur Tiles Softer Interstice Material
(a) Worn Surface Welding (b) Hexadur Roll Surface
FIGURE 7 Roll surface examples
48 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
Primary Crusher
HPGR
SAG Mill
Crusher
Grinding Circuit
Screen
Sorting Section
RunofMine Ore
FIGURE 8 HPGR after SAG mill for pebbles grinding
Primary Crusher
HPGR
Intermediate
Stockpile
Secondary
Crusher
DoubleDeck
Screen
Screen
Concentration
Section
RunofMine Ore
FIGURE 9 HPGR after secondary crusher with screen
SOME BASICS ON HIGHPRESSURE GRINDING ROLLS 49
HPGR
Concentrate
Stockpile
Storage Bin
Grinding Circuit
Pelletizing
Plant
FIGURE 10 HPGR in singlepass grinding before milling plant
HPGR
Ore Storage
Screen
HMS
Sink
Float
Final
Concentration
HPGR
Screen
Screen
HMS
Crusher
Sink
Float
Final
Concentration
Final
Concentration
Fines
Tailings
FIGURE 11 HPGR grinding wet oversize
51
HighPressure Grinding Rolls for
Gold/Copper Applications
Norbert Patzelt,
*
Rene I.B. Klymowsky,
*
Johann Knecht,
*
and Egbert Burchardt
*
ABSTRACT
Successful pilotplant demonstrations carried out in 2003 and 2004 have proven the opera
tional reliability of highpressure grinding rolls (HPGRs) in hardrock applications. As a
result of these breakthroughs, six HPGRs will be commissioned in two copper concentrators
in 2006.
I NTRODUCTI ON
Highpressure grinding rolls (HPGRs) are well established in the diamond and iron ore
industries. Process advantages of HPGRs had been recognised by the minerals industry
for many years. However, unresolved issues pertaining to wear have made the industry reluc
tant to adopt this technology.
Starting in 2003, a successful pilotplant demonstration on an exceptionally hard
and abrasive gold ore proved that the wear issues could be resolved by the design of an
appropriate wearprotection system, and availabilities in excess of 90% could be
achieved. The pilotplant results built up confidence in the minerals industry and a sec
ond pilotplant trial was conducted on another extremely hard ore with the aim of deter
mining if anything could break the machine. The machine demonstrated even higher
availabilities than in the previous case.
A commercial breakthrough then came when one of the world’s leading copper pro
ducers decided to build a new concentrator in South America based on HPGR technology.
Four Polycoms, 24/16 in size, each equipped with two 2,500kW motors, will be used in
tertiary crushing duty in closed circuit with wet screens. Shortly thereafter, a second
major copper producer ordered two large Polycom 20/15 units for an existing copper
concentrator in Indonesia.
In both cases, it was the energy savings and low operating costs of the HPGRs that
attracted the producers. This paper examines the conditions (such as the press force nec
essary) that lead to energy savings, lower operating costs, and the optimum performance
of the HPGRs.
Wide variations occur in ores, even within one deposit. These variations, insofar as
they affect the performance of an HPGR, need to be quantified with meaningful HPGR
indices. Two such indices are the ATWAL Wear Index (ATWI) for wear due to abrasion
and the Polycom Grinding Index (PGI) for quantifying the fines production.
* Polysius AG, Neubeckum, Germany
52 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
A laboratory ball mill test, the Labmill test, is described to overcome the uncertainty
about the energy required for ball milling after an HPGR. The test is aimed to deal specif
ically with the special features of an HPGR product (i.e., microcracking and the high
amount of fines in the product).
EXPERI ENCES WI TH HPGRS I N HARD ROCK APPLI CATI ONS
Cyprus Sierrita
The first serious application of HPGRs in the hardrock mining industry was the installa
tion of an HPGR in a copper concentrator in the United States in 1994 (Figure 1).
The expected performance in terms of throughput, fines production, and energy
consumption was met. However, the hardness and abrasiveness of the ore was by far
higher than that of ores treated in HPGRs previously. It soon became apparent that the
wear protection of HPGRs, in particular, the stud technology, was not advanced enough
at that time to allow for a smooth and easy transition into continuous operation. Different
stud qualities had to be tested and changed in order to suit the requirements of the ore.
The changeouts were facilitated by having the rolls equipped with segments; however,
these also contributed to wear problems. Finally, stud qualities were found that provided
a reasonable lifetime at low cost (
~
0.10 to 0.15 US$/t) even under these difficult circum
stances. In the end, the unit was decommissioned after treating more than 7,000,000 t of
ore when the initial investment plans for the mine were abandoned. Despite the positive
operating results, this installation was widely viewed by the industry as a failure of
HPGR technology, and its acceptance was set back for years.
Newmont Gold, Lone Tree (Nevada)
Following the Cyprus Sierrita demonstration, there was little progress made towards
improving the technology or improving the wear protection for hardrock applications.
The next milestone in HPGR development came in April 2003, when Newmont Mining
Corporation began a 3month trial of a pilotsized HPGR. Polysius designed a new roll
surface specifically for the grinding of hard and abrasive copper and gold ores (Figure 2).
The new roll surface consisted of a replaceable shrinkfitted tyre, armed with a new
design of tungsten carbide studs, and a new edgeprotection system intended to eliminate
FIGURE 1 Installation of an HPGR in a copper concentrator in the United States in 1994
HIGHPRESSURE GRINDING ROLLS FOR GOLD/COPPER APPLICATIONS 53
repair welding of the roll edges. The rolls were also provided with cheek plates to con
tain the material within the gap.
The roll dimensions were 950 mm diameter u 350 mm width. Each roll was driven
by a 160kW motor and was operated at a speed of 21 rpm. The capacity of the unit was
about 80 tph. The unit was run in closed circuit with an 8mm squaremesh screen, and
was protected from tramp metal by an overhead magnet and a metal detector on the
main conveyor belt.
The machine was operated 24 hours a day, 7 days a week, for a period of 87 days
with no mechanical downtimes. The initial operational availability of the unit was 89%
due to a programming glitch, which occurred after startup, after which a 92% availabil
ity was achieved. The properties of the feed are specified in Table 1.
The Lone Tree trial was a true milestone, as no other pilot unit previously had been oper
ated continuously on a 24/7 basis, and no mechanical or welding repairs had been required.
The operators and maintenance staff were encouraged by the operation of the HPGR.
No stud failures occurred during the more than 1,600 operating hours of the demon
stration trial. In a commercialscale application on a similar hard, abrasive ore, an HPGR
would have achieved more than 3,000 hours of service and run considerably longer on a
less competent and less abrasive ore.
Relationships between wear and particlesize distribution were obtained that will
improve future understanding of wear and wear life, benefiting the industry as a whole. The
Polysius ATWAL laboratory abrasion test accurately predicted the wear rate in the trial.
Individual tests, conducted over the course of 1 year on several representative samples
obtained prior, during, and after the trial, were found to be reproducible within 10% of
each other, validating the method used for determining and predicting wear in larger units.
FIGURE 2 A pilotsized HPGR equipped with a new roll surface designed by Polysius
TABLE 1 Material data of Lone Tree ore
Ball Mill Work Index, Wi (BM) 20 kWh/t
Unconfined compressive strength 200 MPa
Silica content 78%–84%
Bond Abrasion Index, Ai 0.64
ATWAL Wear Index, ATWI >40 g/t
54 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
Anglo American Platinum, Potgietersrust Platinum Mine
In October 2004, a pilotplant unit was commissioned in the Potgietersrust platinum
mine. This was a further step forward and the first approach of HPGR technology to the
platinum industry. The HPGR was again operated in closed circuit with 8mm screens.
The feed material was prepared by twostage crushing. Feed size was initially –25 mm
but was increased to –35 mm for extended operating periods. The HPGR treated a total
of 188,000 tph, resulting in 115,000 tph of final product until it was decommissioned in
April 2005. The operating time was in excess of 3,000 hours.
The ore was even tougher than that used in the previous application, which is
reflected in the Bond Work Index (BWI) of 22 kWh/t, and in the operational problems
experienced at the crushers. The HPGR was equipped with the latest wear protection.
Results from abrasion testing indicated low abrasion, which was confirmed in the field.
Availabilities as high as 97% were achieved. The test installation was declared a suc
cess by the operator who, in his words, “had tried very hard to break the machine.”
I MPLEMENTATI ON OF HPGRS I N GRI NDI NG CI RCUI TS
In greenfield installations, HPGRs will have their place as the tertiary crushing stage in
front of ball mills. Even this commitment still leads to many different flowsheet configu
rations. One possible flowsheet is shown in Figure 3.
The secondary crusher and the HPGR, which replaces conventional tertiary crush
ers, are operated in closed circuit with dry screens. The product of the crushing circuit is
stockpiled. In this configuration, the crushing circuit and the ball mill circuit are decoupled,
allowing both circuits to be operated at a different utilisation. However, this decoupling
has two implications. First, an additional stockpile is required. Stockpiling of the HPGR
product, which contains a lot of fines, remains a challenge and requires extensive dust
suppression. Second, screening of the HPGR product has to be done dry because a wet
screen undersize cannot be stockpiled. This entails a coarser product.
Wet screening of the HPGR discharge may provide significant improvements. It is
advantageous from the point of energy efficiency to shift as much grinding work as pos
sible to the HPGR and feed the ball mills with a finer product. This approach requires a
finer mesh size for the screen, 4 to 6 mm. Fine screening usually has a lower efficiency,
especially if the discharge from the HPGR is in the form of highly compacted flakes. Wet
screening will address the disagglomeration of the HPGR discharge and will definitely
improve screening efficiency. It also will facilitate wetting of the material. A flowsheet
illustrating a wetscreen arrangement for an HPGR is shown in Figure 4. The HPGR and
ball mill circuits are combined, whereas the secondary crusher is decoupled. Alterna
tively, if the circuit consists of multiple crushing and grinding units, all three can be com
bined, eliminating the stockpile by oversizing the equipment. This arrangement allows
for the lower availability of the crushers, whereas the HPGR availability is expected to be
high enough for inline operation with the ball mills.
OPTI MUM HPGR PERFORMANCE I N CLOSED CI RCUI T OPERATI ON
In tertiary applications, HPGRs have to be operated in closed circuit. Consequently, the
ball mill feed is not the “discharge” of the HPGR but is the product of the size distribution
of the HPGR discharge and the mesh size of the closing screen. This raises two questions:
first, what influence do HPGR operating parameters have on the feedsize distribution to
the ball mill; and secondly, what is the most efficient way to operate an HPGR?
In open circuit, the operating parameter that manifests the most influence on the
particlesize distribution is the press force applied to the rolls. The energy absorbed by
the material has been shown to be proportional to the applied press force.
HIGHPRESSURE GRINDING ROLLS FOR GOLD/COPPER APPLICATIONS 55
In a closed circuit, however, the influence of the press force on the product size of
the circuit is lost. This is demonstrated with two examples, one taken from tests on a
semiindustrial scale unit and the other from tests on a laboratoryscale HPGR (Figure 5).
Results were taken from singlepass tests on these units, in order to prove that the
findings were independent of the machine size.
The press forces applied were in the range of 2.7 to 4.3 N/mm
2
on the semiindustrial
unit, and from 2.3 N/mm
2
to a higher value of 8.4 N/mm
2
on the laboratoryscale unit.
The impact of the press force on the throughput and energy consumption of the circuit
are also shown.
A screen undersize, representing the circuit product, was calculated on the basis of
100% screen efficiency from the discharge. Cut sizes were 4 mm for the semiindustrial
circuit and 1 mm for the laboratoryscale circuit. It was assumed that the recirculation of
screen oversize did not affect the size reduction in the HPGR significantly. On this basis,
FIGURE 3 HPGR in closed circuit with dry screens
Optional
FIGURE 4 HPGR in closed circuit with wet screens
56 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
a projection of the circuit performance in terms of throughput and energy consumption
was also made. This approach may be considered simplistic but is adequate to explain
some of the principles.
Figure 6 shows discharge size distributions of a copper ore treated in a semiindustrial
test unit. The grinding force was steadily increased from test R4 to R6, resulting in a
finer discharge. The circuit product of an HPGR with a 4mm classifying screen was cal
culated on the basis of 100% screening efficiency.
Figure 7 shows discharge size distributions of a platinum ore treated in a labscale
test unit. The grinding force was steadily increased from test L1 to L4, resulting in a finer
discharge. The circuit product of an HPGR with a 1mm classifying screen was calculated
on the basis of 100% screening efficiency.
The conclusions drawn from Figures 6 and 7 were that the size distribution of the
final circuit product did not vary much, no matter if the HPGR discharge was finer or
not. The fineness of the circuit product was largely determined by the mesh size of the
closing screen.
However, the applied press force had a strong influence on the circulating load and
the circuit throughput with an HPGR of given size, as well as on the energy consumption,
as shown in Figures 8 and 9.
In Figure 8, the specific energy input was increased by 30% while the throughput of
the closed circuit increased by only 10%. In Figure 9, the specific grinding energy was
increased even further by 100% to 8.2 N/mm
2
, whereas the throughput of the closed cir
cuit only increased by 40%. These examples show that operation at high specific press
forces, 8 N/mm
2
, reduces energy efficiency drastically.
The following general conclusions were drawn with regard to optimum operation of
HPGRs in closed circuit:
1. The productsize distribution of an HPGR in closed circuit with screens is not
influenced by the applied press force.
2. The applied press force determines the circulating load and the energy consump
tion of the HPGR circuit.
2.5
2.0
1.5
1.0
0.5
0.0
S
p
e
c
i
f
i
c
E
n
e
r
g
y
,
k
W
h
/
t
0 1 2 3 4 5
Specific Press Force, N/mm
2
Feed 1
Feed 2
FIGURE 5 Absorbed specific energy versus press force
HIGHPRESSURE GRINDING ROLLS FOR GOLD/COPPER APPLICATIONS 57
100
80
60
40
20
0
F
i
n
e
n
e
s
s
C
u
m
u
l
a
t
i
v
e
P
a
s
s
i
n
g
,
%
0.10 0.01 1.00 10.00 100.00
Particle Size, mm
NOTES: Discharge denotes the HPGR discharge from respective tests. S/U denotes the screen undersize
from respective tests.
Feed
R4 Discharge
R4 S/U
R5 Discharge
R5 S/U
R6 Discharge
R6 S/U
FIGURE 6 Semiindustrial HPGR test with copper ore
100
80
60
40
20
0
0.10 0.01 1.00 10.00 100.00
F
i
n
e
n
e
s
s
C
u
m
u
l
a
t
i
v
e
P
a
s
s
i
n
g
,
%
Particle Size, mm
NOTES: D denotes the HPGR discharge from respective tests. S/U denotes the screen undersize from
respective tests.
Feed
L1 – D
L1 – S/U
L3 – D
L3 – S/U
L4 – D
L4 – S/U
FIGURE 7 Labscale HPGR tests with platinum ore
58 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
– For harder ores, increasing the press force increases the circuit throughput, but
the increase in energy consumption is disproportionately higher.
– For softer ores, increasing the press force may even decrease the circuit through
put. The additional fines produced do not compensate for the loss in specific
throughput of soft ores resulting from the reduction in the operating gap.
3. Optimum grinding forces are material specific. Specific grinding forces up to
8 N/mm
2
, such as those applied in cement grinding, are unsuitable for minerals
applications where the final product fineness is substantially coarser.
4. Circuit throughput can be adjusted by varying the applied press force. The
increase in the energy consumption, however, is often disproportionate to the
80
2 3 4 5 6
4
60 3
40 2
20 1
0 0
T
h
r
o
u
g
h
p
u
t
M
,
t
p
h
S
p
e
c
i
f
i
c
E
n
e
r
g
y
I
n
p
u
t
w
(
s
p
)
,
k
W
h
/
t
Specific Grinding Force ϕ, N/mm
2
M
w
(sp)
FIGURE 8 Semiindustrial circuit projection (copper ore <4mm product)
8
0 2 4 6 8 10
8
6 6
4 4
2 2
0 0
T
h
r
o
u
g
h
p
u
t
M
,
t
p
h
S
p
e
c
i
f
i
c
E
n
e
r
g
y
I
n
p
u
t
w
(
s
p
)
,
k
W
h
/
t
Specific Grinding Force ϕ, N/mm
2
M
w
(sp)
FIGURE 9 Labscale circuit projection (platinum ore <1mm product)
HIGHPRESSURE GRINDING ROLLS FOR GOLD/COPPER APPLICATIONS 59
increase in throughput, and varying the roll speed is a far more efficient means
of adjusting the circuit throughput.
These considerations still leave open the question as to what impact the application
of higher press forces may have on the energy requirements in downstream ball milling.
This issue is addressed in the next section.
BALL MI LL ENERGY REQUI REMENTS
Significant savings in grinding energy can be expected on feeding a ball mill with an
HPGR product. The increased grindability can be used to expand the throughput of exist
ing plants or to produce a finer grind if required by the recovery process. In new plants,
ball mills can be sized smaller as compared to a plant where the ball mills receive feed
prepared by conventional crushers.
The reduced energy demand of ball mills fed with HPGR products has been proven
in numerous trials. In ball mill pilotplant tests conducted after the HPGR trials at Kal
goorlie Consolidated Gold Mines PTY Ltd., a reduction in energy consumption of 20%
was observed on HPGR centre product, and a reduction of 16% was obtained on HPGR
total discharge, as compared to a tertiary crushed product. Similar results have been
reported from tests on Boddington ore, and have also been confirmed in practice on iron
ore in Chile.
The energy consumption in a ball mill (w
(sp)
) is often expressed as a function of a
Bond Work Index (BWI), the product size (P80), and the feed size (F80):
w
(sp)
= BWI u (10/ – 10/ ) (EQ 1)
This method is widely used. However, the limitation of the Bond theory is that it does
not consider the feedsize distribution and only refers to a single point on the distribution,
the F80 size. In particular, the Bond theory does not consider any variations in the amount
of fines that may be in the ball mill feed. Figure 10 shows productsize distributions from
crusher and HPGR circuits. It is obvious that the “HPGR circuit” product contains far
more fines than the conventional “crushed” product, although the P80 values may be
similar (see “fine crushed” material in Figure 10). This higher amount of fines will result
in a reduced energy consumption (w
(sp)
) in subsequent ball milling, even if potential
microcracks are ignored and the BWI of the fine crushed and HPGR circuit samples are
found to be identical. According to Bond, the mills would require the same energy.
The two energyreducing features of a “polycomized” product, microcracks and higher
fines content, will not necessarily be reflected in the BWI. The BWI seems to be material
specific and is not affected by how the sample was prepared. Only in a few cases have Bond
tests revealed differences of 10%–18% between conventional crushed and polycomized
products.
Only a grinding test based on the actual feed size distribution that does not require
any further size reduction—which could eliminate microcracks—can provide a realistic
comparison of the ball mill energy required for materials with different size distributions.
Polysius has used the Labmill for decades in order to determine the energy require
ments in ball milling. The test is a dry, opencircuit grinding test, but it can provide
information on the relative energy requirements for different materials and for different
feedsize distributions in wet milling as well.
Typical results from Labmill tests are shown in Figures 11 and 12. In these figures,
feed with different size distributions are compared. The tests were conducted on a “con
ventional crushed” material and on HPGR “discharge” and “centre” products with different
size distributions.
P80 P80
60 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
The graph in Figure 12 shows the product fineness at 90 and 200 Pm as a linear
function of the energy applied according to the following equation:
F
90 Pm
= a
90 Pm
* E + b
90 Pm
(EQ 2)
F
90 Pm
: fineness (cumulative % passing at 90 Pm)
E : energy index or consumption at F
90 Pm
where b
90 Pm
corresponds to the amount of 90Pm material in the feed (F
90 Pm,feed
), and
a
90 Pm
is the slope showing the rate of increase in the 90Pm material produced per
kilowatthour per ton.
Reference to Figures 11 and 12 shows that the a
90 Pm
value is very similar for all feed
size distributions, no matter how the feed material was prepared or how many fines were
present in the feed. The a
90 Pm
value (or a
200 Pm
) reflects the grindability of an ore in the
Labmill and is ore specific similar to the BWI. It is referred to as the Labmill Grinding
Index. The “b” parameter reflects the amount of fines in the feed.
Equation (2) can be rewritten by substituting b
90 Pm
for F
90 Pm,feed
as follows:
E (F
90 Pm
) = (F
90 Pm
– F
90 Pm,feed
) / a
90 Pm
(EQ 3)
This approach allows one to correct the energy requirements for grinding to a certain
fineness according to the feedsize distribution. The relative energy requirements for the
different feedsize distributions tested are summarised in Table 2.
Looking at the 80% <200 Pm size, the difference in the energy index from the HPGR
centre material to the crushed material is 33%, and at 90 Pm, the corresponding differ
ence in the energy index is 25%.
Another series of Labmill tests were conducted on a platinum ore with a high BWI of
22.2 kWh/t. Bond testing showed no difference in the BWI between an HPGR product
100
80
60
40
20
0
0.10 0.01 1.00 10.00 100.00
Particle Size, mm
F
i
n
e
n
e
s
s
C
u
m
u
l
a
t
i
v
e
P
a
s
s
i
n
g
,
%
Crushed
Fine Crushed
HPGR Circuit
FIGURE 10 Conventional crusher products and polycomized product of an HPGR circuit
HIGHPRESSURE GRINDING ROLLS FOR GOLD/COPPER APPLICATIONS 61
100
80
60
40
20
0
0.10 0.01 1.00 10.00 100.00
Particle Size, mm
F
i
n
e
n
e
s
s
C
u
m
u
l
a
t
i
v
e
P
a
s
s
i
n
g
,
%
Crushed
HPGR Discharge
HPGR Centre
FIGURE 11 Feedsize distribution of Labmill test samples (gold ore)
Energy Index, kWh/t
F
i
n
e
n
e
s
s
C
u
m
u
l
a
t
i
v
e
P
a
s
s
i
n
g
,
%
0 5 10 15 20
0
20
40
60
80
100
y
=
5
.
4
9
x
+
1
2
.
8
3
y
=
5
.
4
1
x
+
2
.
3
3
y
=
8
.
2
2
x
+
2
8
.
6
9
y
=
7
.
8
7
x
+
2
0
.
6
2
y
=
7
.
9
8
x
+
4
.
4
7
y
=
5
.
7
2
x
+
1
8
.
4
5
Crushed: % < 90 μm
% < 200 μm
HPGR Discharge: % < 90 μm
% < 200 μm
HPGR Centre: % < 90 μm
% < 200 μm
FIGURE 12 Results from Labmill test with gold ore samples
62 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
and a conventionally crushed product. Labmill tests were conducted to investigate if the
application with higher press forces in an HPGR would have an effect on the energy
requirements in subsequent ball milling.
In the preparation of the samples for the Labmill, two of the samples were pressed in
a laboratory HPGR, and one was crushed conventionally (Figure 13). The specific press
forces applied in HPGR tests were 2.2 N/mm
2
and 8.4 N/mm
2
, respectively.
Figure 14 shows the differences in the energy requirements to grind these samples to
a P80 of 90, and 200 Pm. The results suggest that the main reason for these differences
was the varying amounts of fines <90 and <200 Pm produced in the feed. The Labmill
grindability indices (LGIs) did not appear to be affected by the application of different
press forces. The a
90 Pm
was nearly the same for all the samples, between 4.23 and 4.47.
Also the a
200 Pm
was steady between 6.24 and 6.38. Closedcircuit operation of the
HPGRs at these different pressures would provide productsize distributions that would
be nearly the same as shown before. Therefore, the energy requirements in a subsequent
ball mill may be expected to be the same for products of the high and low press forces.
The overall energy balance for the circuit shows that the application of extremely high
press forces is a waste of energy (refer to Figure 9).
To recapitulate, in a first instance the Labmill test serves as a means of comparing
mill power requirements for feed materials with differences in preparation and fineness.
The test may also be used to provide correction factors to the conventional Bond sizing
considering the particular properties of a highpressurerolls product.
ORE CHARACTERI SATI ON FOR HPGR TREATMENT
Several tests have been developed in order to quantify the behaviour of different ores in
the various crushing and grinding applications. Indices like the Bond work ball and rod
mill index, Bond Abrasion Index, and compressive strength data have a long history.
More recent are tests that refer to autogenous or semiautogenous milling, include the JK
Drop Weight Test and the semiautogenous grinding (SAG) Power Index (SPI) (MinnovEX).
However, none of these tests can be applied to highpressure grinding.
Of greatest importance to HPGRs is
1. Fines production
2. Abrasiveness of an ore
These parameters determine the investment and operating costs of a highpressure
grinding circuit. The more fines that are produced in one pass through the rolls, the
lower the circulating load. Low circulating loads mean small equipment size and lower
capital costs as well as low energy consumption. The abrasiveness of an ore directly cor
relates with availability and wear cost.
TABLE 2 Energy requirements in Labmill tests for different feedsize distributions
Variable Coefficient Crushed HPGR Discharge HPGR Centre
Bond Work Index kWh/t 19.2 n.d. 18.0
Relative % 100 n.d. 94
Energy index from Labmill
80% <90 μm kWh/t 14.4 12.3 10.8
Relative % 100 85 75
80% <200 μm kWh/t 9.4 7.7 6.3
Relative % 100 82 67
HIGHPRESSURE GRINDING ROLLS FOR GOLD/COPPER APPLICATIONS 63
100
80
60
40
20
0
0.10 0.010 1.00 10.00 100.00
Particle Size, mm
F
i
n
e
n
e
s
s
C
u
m
u
l
a
t
i
v
e
P
a
s
s
i
n
g
,
%
Crushed
HPGR (ϕ = 2.2 N/mm
2
)
HPGR (ϕ = 8.2 N/mm
2
)
FIGURE 13 Feedsize distribution of Labmill test samples (platinum ore)
Energy Index, kWh/t
F
i
n
e
n
e
s
s
C
u
m
u
l
a
t
i
v
e
P
a
s
s
i
n
g
,
%
0 5 10 15 20
0
20
40
60
80
100
Crushed % < 90 μm
% < 200 μm
HPGR (ϕ = 2.2 N/mm
2
) % < 90 μm
% < 200 μm
HPGR (ϕ = 8.2 N/mm
2
) % < 90 μm
% < 200 μm
y
=
6
.
3
8
x
+
3
5
.
6
7
y
=
6
.
2
4
x
+
2
1
.
1
8
y
=
6
.
2
7
x
+
1
1
.
5
3
y
=
4
.
2
3
x
+
2
3
.
7
5
y
=
4
.
4
1
x
+
1
3
.
9
0
y
=
4
.
4
7
x
+
6
.
2
6
FIGURE 14 Results from Labmill test with platinum ore samples
64 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
The achievable product fineness in one single pass for a particular ore depends on
the applied grinding force (Figure 15). The fineness at the same grinding forces varies
vastly for different ores, however. The properties of an ore have a far greater impact on
the achievable fines production than the grinding force. Also, wear rates of gold and cop
per ores vary extremely in a range from 2 to 50 g/t.
Two indices have been defined in order to characterise the behaviour of an ore in the
highpressure grinding process (i.e., the PGI and the Polycom Abrasion Index [ATWI]).
Polycom Grinding Index
The PGI is the amount of tons less than 250 Pm and 1 mm that are produced in 1 hour
with a standardised HPGR having rolls of 1m diameter and 1m width and being oper
ated at a speed of 1 m/s. The fines production is referred to a constant specific grinding
force of 3.5 N/mm
2
. The PGI is calculated from test results.
PGI (1 mm) [t/h] = F (1 mm) / 100% u m (EQ 4)
F (1 mm) [%] : net production of –1 mm material
m [t u s/(m
3
u h)] : specific throughput rate of HPGR
The higher the PGI, the higher the fines production and the amenability of an ore to the
HPGR process.
Polycom Abrasion Index
The abrasiveness of ores varies widely with the physical properties of the material and
the operating conditions. Several abrasion tests are available in the minerals industry.
However, none are based on high pressure as the principle for comminution and there
fore cannot be used for reliably predicting wear rates in an HPGR. Inclusion of this princi
ple in the test procedure is a precondition for determining the abrasiveness in an HPGR.
F
i
n
e
n
e
s
s
C
u
m
u
l
a
t
i
v
e
P
a
s
s
i
n
g
,
%
0 2 4 6 8
0
20
40
60
80
100
Specific Grinding Force, N/mm
2
y = 72.967x
0.1483
R
2
= 0.2406
y = 28.787x
0.3479
R
2
= 0.2788
y = 20.35x
0.3031
R
2
= 0.1374
% < 250 μm
% < 1 mm
% < 8 mm
FIGURE 15 Product fineness as a function of the grinding force for different ores
HIGHPRESSURE GRINDING ROLLS FOR GOLD/COPPER APPLICATIONS 65
The ATWAL weartesting, highpressure grinding unit is shown in Figure 16. The
rolls are 100 mm diameter u 30 mm wide. The unit can be equipped with tyres made of
different wear materials. The tyres are weighed before and after a test run. The wear rate
is calculated from the weight loss divided by the amount of material treated and is
expressed in grams per ton, as a wear index designated by the name ATWI. Typically,
100 kg is sufficient for a test run.
Ore Categorisation
Test data collected on a semiindustrial unit and the wear testing unit ATWAL are
included in Figure 17. The PGI (1 mm) and the ATWI are independent. The diagram is
divided into four quadrants. Ores best suited for highpressure grinding are located in
quadrant 4. These ores are low abrasive and produce a lot of fines. Most difficult ores are
FIGURE 16 Wear testing HPGR
A
T
W
I
20
50
25
0
70 120
PGI (1 mm)
Low Abrasive, Low
Fines Production
High Abrasive, Low
Fines Production
High Abrasive, High
Fines Production
Low Abrasive, High
Fines Production
1
2
4
3
FIGURE 17 PGI and ATWI
66 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
found in quadrant 2. This location indicates that these ores are highly abrasive and produce
only a low amount of fines. Test work for PGI testing will be conducted on a labscale
HPGR in the future in order to reduce the amount of material required for testing.
The PGI and ATWI allow different ores to be easily compared with respect to their
behaviour in the highpressure grinding process. They permit a basic comparison of
installations treating different ores. A model is under development based on the PGI that
would allow prediction of how much the throughput may be expected to vary in existing
HPGR grinding circuits, if different ores types have to be treated.
CONCLUSI ONS
A milestone in HPGR development was reached in April of 2003, when Newmont Mining
Corporation began a 3month trial on an HPGR equipped with a new wearprotection
system designed by Polysius specifically for the grinding of hard and abrasive copper and
gold ores. The new wearprotection system proved itself in the field and was accepted by
the operators. Nevertheless, liner development for HPGRs did not stop there. In the
meantime, more than 10 HPGRs in South African diamond mines have been converted
from tyres with a hard facing to a studded design. These installations provide a valuable
source of information on wear behaviour and the suitability of different wearprotection
systems. Experience flowing back from the field ensures further progress in the future.
HPGRs can be operated in closed circuit with either dry or wet screens. Dry screening
allows the crushing section to be decoupled from the ball mills by the interposition of a
stockpile. However, wet screening allows for the implementation of finer cut sizes in the
HPGR circuit, thus producing a finer feed to the ball mills. In addition, wet screening
may be expected to be more efficient than dry screening. A finer HPGR product and
more efficient screening can improve the energy efficiency of a crushing and grinding
circuit.
The press force applied to highpressure rolls determines the circulating load and
the energy consumption of the HPGR circuit but has no influence on the productsize
distribution of a closed circuit. For harder ores, the throughput can be increased by the
application of higher press forces, but the increase in energy consumption will be dispro
portionately higher. For softer ores, circuit throughput may even decrease at higher press
forces. Optimum press forces are material specific. Specific press forces up to 8 N/mm
2
,
such as those applied in cement grinding, are unsuitable for minerals applications where
the required product fineness in tertiary crushing applications is quite coarse. The over
all energy balance for the circuit shows that the application of high press forces is a
waste of energy.
Speed control is a more efficient means of controlling circuit throughput than trying
to adjust the roll gap by varying the applied press force.
Ball mill testing in the Labmill provides an orespecific LGI that is comparable to the
BWI. Feed up to 30 mm in size can be tested directly in the mill without adjustments for
top size, which may alter any properties inherent in the coarser sizes. The Labmill test
can serve as a means of comparing the mill power requirements for feed materials with
differences in preparation and fines content. The test may also provide correction factors
to the conventional Bond sizing considering the particular properties of a highpressurerolls
product.
In the future, the power requirements of wetgrinding ball mills could be also deter
mined by the Labmill test. This approach may allow prediction of the energy requirements
for wet grinding in ball mills according to the actual feed conditions (i.e., feed size and
potentially any microcracks). This approach may be more precise than conventional
Bond sizing, which was developed long before HPGRs came on the scene.
HIGHPRESSURE GRINDING ROLLS FOR GOLD/COPPER APPLICATIONS 67
Two indices have been defined in order to characterise the behaviour of an ore in the
highpressure grinding process (i.e., the ATWI and the PGI). These indices allow different
ores to be easily compared with regard to their behaviour in HPGRs and to benchmark
them with existing installations.
BI BLI OGRAPHY
Baum, W. 1993. Case made for high pressure grinding in gold plants. Mining Engineering
(June): 524–529.
Baum, W., and J. Knecht. 2000. HPGR as a processing tool for gold & copper leaching,
flotation and gravity separation. 2nd Annual Crushing and Grinding in Mining
Conference, Johannesburg.
Baum W., N. Patzelt, and J. Knecht. 1996. The use of high pressure grinding for
optimization of copper leaching. SME Preprint 9668. Littleton, CO: SME.
Burchardt, E. 1998. HPGR: A metallurgical tool for the diamond industry. Proceedings,
Randol Diamond Focus 98. Vancouver, BC, November.
Dunne, R., A. Goulsbra, and I. Dunlop. 1996. High pressure grinding rolls and the effect
on liberation. Comparative test results. Pages 46–54 in Proceedings, Randol Gold
Forum, Olympic Valley, CA, April.
Klymowsky, I.B., and J. Liu. 1997a. Modelling of the comminution in a roller press.
Pages 141–154 in Proceedings of the XX IMPC. Volume 2. Aachen, IMPC, September.
———. 1997b. Towards the development of a work index for the roller press. Chapter 14 in
Comminution Practices. Edited by S.K. Kawatra. Littleton, CO: SME.
Klymowsky, R., and T.C. Logan. 2005. HPGR demonstration at Newmont’s Lone Tree
mine. Pages 325–334 in Proceedings of the Canadian Mineral Processors. Ottawa, ON:
CIMM.
Klymowsky R., N. Patzelt, E. Burchardt, and J. Knecht. 2002. Selection and sizing of high
pressure grinding rolls. Pages 636–668 in Mineral Processing Plant Design.
Proceedings, Volume I. Edited by A.L. Mular, D.N. Halbe, and D.J. Barratt. Littleton,
CO: SME.
Patzelt, N., R. Klymowsky, E. Burchardt, and J. Knecht. 2001. High pressure grinding
rolls in AG/SAG mill circuits. Pages 107–123 in Proceedings, SAG 2001. Volume III,
Mining and Minerals Process Engineering, University of British Columbia, Vancouver.
Rowe, P., B. Parker, G. Lane, and S. Morell. 2001. The decision to opt for high pressure
grinding rolls for the Boddington expansion. Pages 93–106 in Proceedings, SAG
2001. Volume III, Mining and Minerals Process Engineering, University of British
Columbia, Vancouver.
Thompsen, L.G. 1997. Operational performance of grinding rolls at Cyprus Sierrita
Corporation. Chapter 15 in Comminution Practices. Edited by S.K. Kawatra.
Littleton, CO: SME.
Watson, S., and M. Brooks. 1994. KCGM evaluation of high pressure grinding roll
technology. Pages 69–83 in Proceedings of the Fifth Mill Operators Conference. The
Australian Institute of Mining and Metallurgy, October.
Westermeyer, C.P., and H. Cordes. 2000. Operating experience with a roller press at the
Los Colorados ore dressing plant in Chile. Aufbereitungs Technik 41(11):497–505.
69
Selection and Sizing of Ultrafine
and Stirred Grinding Mills
Jens K.H. Lichter
*
and Graham Davey
†
ABSRACT
The selection and sizing of mills for regrind and ultrafine grinding applications do not lend
themselves to conventional methodologies. A more holistic approach is required, one that
considers not only the mill but also the application of the mill within the process. The selec
tion of stirred mills for ultrafine milling requires unique approaches to answer questions
related to selection of the circuit configuration, type of mill, media, and operating condi
tions. Further considerations required for the selection of mills are the inherent difficulties of
particlesize measurement and an accurate definition of the product size required.
I NTRODUCTI ON
Minerals with fine particle intergrowth, either with other metallic minerals or gangue,
are being increasingly mined. These ores have provided new challenges in concentrator
design, requiring fine and ultrafine grinding in order to obtain acceptable grades and
recoveries. The advancement of flotation technologies now permits effective flotation in
the sub10Pm size range, making it possible to separate finely disseminated minerals
from gangue. Similarly, the ability to produce ultrafine feed for various leaching pro
cesses, including bioleach and lowpressure oxidation, often requires fine or ultrafine
grinding to improve reaction kinetics to the level at which these processes become com
mercially viable. Economic ultrafine grinding processes also make it feasible to direct
leach refractory gold ores, rather than the more conventional roasting or highpressure
autoclave routes.
The relationship between energy required and product size is not a proportional
one. Theoretically, the energy required (per unit mass) to break a particle to 1/100th of
its size is 4,000 times greater. As we strive for everfiner grinds, the need to optimize the
comminution process becomes selfevident. In order to achieve the required improve
ments, changes in milling technologies are needed, as well as a better understanding of
the variables that affect them.
The media in a mill generates a particular energy spectrum, which is best defined as
a frequency/magnitude plot of the energy delivered by the mill. It is possible to substan
tially alter this relationship for a mill by changing the operating conditions. Different
mill designs will, however, differ in the range of energy spectra they can generate. The
better that a mill’s energy spectra matches the breakage requirements of an application,
* Metso Minerals Optimization Services, Colorado Springs, Colorado
† Metso Minerals Industries Inc.–Grinding, York, Pennsylvania
70 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
the more efficient the system will be. Jankovic (2001) has demonstrated this where clear
optimum operating points could be determined for a mill by changing the operating
characteristics.
THE SELECTI ON OF MI LLS—EQUI PMENT OPTI ONS
Time and innovation have resulted in numerous different mill designs capable of produc
ing fine and ultrafine products. This paper will concentrate on the few that have seen
mainstream commercial application in the mineral processing industry. In many cases,
these mills are of a unique and proprietary design, do not have generic names, and are
known only by their commercial trademarked names. Mills for fine and ultrafine grind
ing fall into four primary categories:
1. Ball mills
2. Stirred media mills
3. Centrifugal mills
4. Jet mills
The first two categories make up the bulk of the fine and ultrafine grinding duty.
Ball mills still see extensive use for the production of fine powders; however, these
tend to be predominantly dry applications in specific industries. Industrial minerals
applications make extensive use of dry ball mills, often using ceramic grinding media (to
avoid metal contamination of the product) for the production of fine and ultrafine pow
ders. Ball mills in these applications are typically operated in closed circuit with dynamic
classifiers. Long tube mills (lengthtodiameter ratios in excess of 3 to 1) and batch mills
are also used. For wet minerals applications, the application of tumbling ball mills is
declining and limited primarily to very large tonnage applications and relatively coarse
grinds. The efficiency advantages of stirred media mills over ball mills have largely seen the
fine and ultrafine applications move away from conventional tumbling ball mills. Figure 1
shows a typical relationship for specific energy versus grind for a ball mill and a stirred
media mill. At coarser grinds, the stirred media mill requires approximately 30% less
0 50 100 150 200 250
100
10
1
Specific Energy, kWhr/t
Tumbling Mills
Stirred Media Mills
FIGURE 1 Relative performance of tumbling ball mills versus stirred media mills
SELECTION AND SIZING OF ULTRAFINE AND STIRRED GRINDING MILLS 71
energy than a ball mill. For ultrafine grinds, this advantage increases to more than 50%.
The data shown in this example are comparative data from closedcircuit pilot milling
campaigns using a conventional ball mill and a Vertimill. Media size and feed were identical
for both mills. These relationships hold for fullsize mills and are typical of the relative
performance of stirred media mills in general when compared to conventional ball mills.
Stirred media mills can be applied to relatively coarse feeds and grinds, with feeds
up to 6 mm and products as coarse as 300 Pm possible from some of the available mills.
These do, however, represent extremes in the range, and, more typically, feed sizes will
range from 300 Pm down to 50 Pm. Products typically range from 50 Pm down to 5 Pm.
The definition for ultrafine products is not an industry standard, but for the purposes of
this paper, it is defined as sub15 Pm. Stirred media mills can be classified into a number
of different subcategories predominantly defined by the speed, geometry, and orientation of
the media agitator or stirrer.
The mill orientation can be either horizontal or vertical. The media agitator can consist
of a shaft fitted with a spiral screw, pins, or discs, and the media can be either agitated or
fluidized. Although the basic appearance of the mills is often similar, the operating
regime and performance can be very different. There are two fundamentally different
classes of stirred media mills. The first class includes the Vertimill or Tower Mill and con
ventional pin mills. In these mills the agitator “stirs” the media with the fluid, having
limited effect on the interaction of the media with itself. In the second class, typified by
the stirred media detritor (SMD) and the Netzsch/IsaMill, the media size is very fine,
and the speed of the impellor is high enough to effectively fluidize the media. The media
becomes highly mobile and takes on the flow pattern of a viscous fluid.
Stirred media mills such as pin mills or the Vertimill, which use larger media, are
more efficient with coarse, hard feeds. Fluidized media mills using lowdensity silica or
ceramic media have the advantage for ultrafine milling with fine feeds.
One commonly quoted characteristic of stirred mills, the energy intensity, does not
have a strong influence on the relative performance of the mills. Mills such as the
Vertimill and Tower Mill with low energy intensities operate as efficiently as mills, such
as the SMD and the Netzsch/IsaMill, which have very high energy intensities. The key is
correct media selection. The various mills are described briefly in the following sections.
Vertimill or Tower Mill
This is a stirred media mill consisting of a vertical cylinder with a relatively slowspeed
screw media agitator (see Figure 2). The Vertimill/Tower Mill is most often used for con
centrate regrind applications with a typical feed size of around 100 to 300 Pm and typi
cal products of 100 to 15 Pm. Finer products are possible with the use of suitable media.
These mills predominantly use steel ball media ranging in size from 40 mm to 6 mm, and
are also capable of using cast pebble media such as “cylpebs” or “power pebs,” which
become economically attractive for media sizes less then 25 mm. The low impellor speed
aids in reducing component wear but results in a large mill size and volume. These mills
are predominantly wet mills. The finest grinds in commercial applications grind down to
approximately 80% passing 12 Pm, but with suitable media, sub5Pm grinds have been
obtained in pilot installations at comparable efficiencies to other stirred media mills.
Pin Mill
Pin mills have a vertical shaft impellor fitted with pins. The mill body is filled with spheri
cal media, typically steel or ceramic, in the size range of 3 to 12 mm. Pebble or autoge
nous media can also be used. These mills are capable of operating either wet or dry. The
preferred feed size is <300 Pm, and sub10Pm grinds are achievable. Relatively high
72 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
impeller speeds often result in wear issues with these mills, and they are most suitable
for nonabrasive feeds. Figure 3 shows a cutaway view of a Metprotech mill.
Stirred Media Detritor or Sand Grinder
This mill utilizes a vertical shaft pin agitator. Media is typically highgrade alluvial silica
sand or ceramic spheres in the range of 1 to 3 mm in diameter (see Figure 4). The agita
tor speed is high enough to fluidize the media. Screens fitted to the upper circumference
of the mill body allow product to leave the mill while retaining the media. These screens
do not act as a classifier but function only to retain the media. Feed size is typically in the
range of 100 Pm down to 15 Pm, and products as fine as 98% passing 2 Pm are achiev
able. Typical applications in the metalliferous industry have products down to 80% pass
ing 6 Pm. Figure 4 shows the Metso SMD.
Netzsch or IsaMill
This mill utilizes a horizontal shaft media agitator most commonly fitted with discs.
Media utilized ranges from prepared fine slag media through sand media and ceramic
media in the 1to3mm size range. In the case of the IsaMill, the mill is fitted with an
internal centrifugal screen fitted to the impellor for media retention. The application
range is similar to the SMD (see Figure 5).
Centrifugal Mill
This category of mills generates high energy intensity inside the mill by moving the mill
body around a central axis at high speed. It is, therefore, possible to create forces well in
excess of the 1g force available to tumbling ball mills. These mills can be operated with
conventional media or autogenously, and will operate wet or dry. One example is the
HiCom nutating mill. This mill swings the mill body in a nutating motion. These mills
will accept coarse feeds (limited by the throat diameter) and are capable of grinding
below 10 Pm. Media retention can be an issue when small media is required (see Figure 6).
Jet Mill or Fluid Energy Mill
This is a stationary mill that uses the energy contained in a fastmoving fluid to produce
particlesize reduction by impact or abrasion of the particles. Two main types are in use—
either the parallel type where the air is introduced to a circular grinding chamber or the
opposed jet where two opposing fluid streams are impacted. The fluid used to carry the
feed solids is normally compressed air, an inert gas, or steam. No media is used in fluid
energy mills, the feed material and fluid providing the breakage forces.
FIGURE 2 General view of a Vertimill FIGURE 3 General view of a Metprotech mill
SELECTION AND SIZING OF ULTRAFINE AND STIRRED GRINDING MILLS 73
PRODUCT DEFI NI TI ON
The selection of type and size of mill for an ultrafine grinding application has to start
with a thorough understanding of the duty. A mill is not there to provide a product of a
particular size specification, but to provide a product with the desired liberation charac
teristics. To amplify on this statement, different mill characteristics and operating condi
tions will affect the type of comminution that takes place in a mill and will vary the
balance between the fracture, attrition, and abrasion components. This in turn affects
the size distribution, liberation, and surface characteristics of the product. It is important
to consider these points when developing pilotplant flowsheets. It is not advisable to
utilize a conveniently available mill for the generation of feed to a pilot concentrator
plant. The mill product characteristics, and therefore the grade and recovery perfor
mance of the plant, are closely related to the type of mill used. It is important to at least
stay within similar classes of mills. If different types of mills are to be considered, then
due cognizance is required of the differences between the mill products and their effect
on plant performance. At the very least, wholesize distributions should be compared
and evaluated for their recovery characteristics.
Defining Product Size
It is a common practice in the minerals processing industry to define the product size of
a slurry by the particle size at which 80% of the particles by mass are smaller than that
FIGURE 4 General view of a stirred media
detritor
FIGURE 5 General view of a Netzsch/IsaMill
FIGURE 6 Cutaway view of the HiCom nutating mill
74 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
particle size (the P
80
.) This does not give a true picture of the mill productsize distribu
tion. Many industrial minerals producers (e.g., the paper filler suppliers) have long
moved away from such “loose” product definitions. Multipoint productsize distributions
with very tight specifications (sometimes defining the required 99.9% passing size) are
commonplace. Such stringent restrictions are not necessary in most minerals applica
tions and are often not achievable for wet applications. It would, however, be advanta
geous to move away from the customary 80% passing size specification. In many mineral
concentration systems, the P95 or the P10, rather than the P80, will define more accu
rately the grade and recovery possible with that product stream. As an example, consider
the typical flotation grade/recovery characteristics. Recovery of particles below a threshold
size are severely impacted by limitations in the physical chemistry of the system (e.g.,
the ability to depress ultrafine gangue or collect mineralized particles). Similarly, coarse
particles will result in poor liberation affecting both grade and recovery. Recovery losses
in most leach processes (e.g., a cyanide gold leach) are largely determined by the coarse
tail. It is therefore important to consider the whole mill product distribution curve in
relation to the optimumgrade recovery requirements of the downstream concentration
stage.
Consider the decision to mill in open circuit or in closed circuit. Figure 7 shows the
grind versus specific energy characteristics of a Vertimill application. The data depict the
relationship between the grind and the specific energy required for open and closedcircuit
configuration. The grind and energy relationships are shown as functions of the P80 and
the P95. Assume that opencircuit pilot tests preceded leaching tests and a P80 of 10 Pm
was defined as the correct product size for optimum grade/recovery economics. This
would equate to a specific energy requirement of 54 kWhr/t milled. In closed circuit, the
specific energy required to the same grind would be 37 kWhr/t. This is a reduction of
approximately 30%. Considering the cost differences between an open and a closed mill
ing circuit at these product sizes (both capital and operating), the additional specific
energy required might be considered reasonable. If this were a milling circuit preparing
feed for a leach circuit, then the key recovery criteria would probably be a P95 of 22 Pm.
Based on this assessment, an open milling circuit would require 54 kWhr/t, but a closed
milling circuit would require only 18 kWhr/t. If the selection criterion were a P95 rather
than a P80, the energy reduction from open to closedcircuit milling would then be
67%. Reaction kinetics will reduce the difference by a margin, but the basis for a deci
sion would still be substantially different.
An equally important criterion is the product size specified. In the ultrafine product
range, the relationship between the specific energy required and the product size is very
flat. Significant increases in specific energy are required to produce moderate improve
ments in the grind. Figure 8 shows a typical specific energy versus grind relationship for
a stirred media mill. A change in the product specification from 80% passing 7 Pm to
80% passing 6 Pm would require an additional 50% specific energy. In this environment
it is important to be precise as to the required product specification.
ParticleSize Measurement
The definition of the product size also warrants consideration as ultrafine milling adds
an additional level of complication. Unlike typical grinds down to 38 Pm, where screen
ing is commonly employed, there are no “absolute” standards with which to measure
ultrafine particlesize distributions. Current particlesize measurement methods include
laser diffraction, settling, cycloning, optical, and so forth. The most common units are
the Malvern Mastersizer, Microtrac, Cyclosizer, Coulter Counter, and the Sedigraph.
Each of these methods has distinct characteristics and is affected differently by shape,
density, and translucence among other particle properties. For the industrial minerals
SELECTION AND SIZING OF ULTRAFINE AND STIRRED GRINDING MILLS 75
and pharmaceutical industries that have always worked in these fine size ranges, common
industry standards are employed. That is not the case in the metalliferous industry where
many different methods are still used. Table 1 shows comparative results between different
particlesize measurement methods for eight samples over a range of product sizes.
It should be noted that the relative trends (i.e., which machine reports a finer size)
are not always consistent for all powders. The particle characteristics, predominantly
shape, have a strong influence. The discrepancies tend to be more pronounced at the
finer sizes, where machine limitations begin to be encountered.
0 10 120 20 30 40 50 60 70 80 90 100 110
100
10
1
Specific Energy, kWhr/t
Open Circuit P80
Open Circuit P95
Closed Circuit P80
Closed Circuit P95
FIGURE 7 Grind versus specific energy—comparison of open and closedcircuit performance
0 10 20 30 40 50 60 70 80 90
100
10
1
Specific Energy, kWhr/t
FIGURE 8 Typical specific energy versus grind relationship for a stirred media mill
76 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
In addition to potentially large differences in the product sizes “measured” by the
different methodologies, machines of the same type and brand can give significantly dif
ferent results. Even if the type and model of particlesize analyzer has been standardized,
machine setup, maintenance, standard operating procedures, and operator skill all have
significant effects. Agglomeration of particles during measurement is also a concern, and
the use of dispersants adds an additional level to the potential error.
If different milling technologies are being compared, using pilot or batch milling tests
in different locations, it is important to be aware that the particlesize measurement will
be different, and probably by more than the potential difference between the different mill
ing technologies. This would be true even if the same model particlesizemeasurement
device were used. The only reliable comparisons can be made if the products from both
tests are tested on the same particlesize analyzer using the same technician. Failing that,
normal measurement errors may be too high to make meaningful comparisons. Where
the same particlesize analyzer cannot be used, it is essential to at least use similar model
machines and employ the same operating procedures. Comparisons made using dissimi
lar particlesizemeasurement techniques are not very meaningful. Similarly, it must be
remembered that the particlesizemeasurement method must be similar to that
employed when originally determining the grind requirement for optimum liberation.
There are no easy solutions. The key is to exercise caution, and rigorously work
according to bestknown practice. Be realistic about the reliability and accuracy of the
methods used for sizing and comparing equipment.
DETERMI NATI ON OF THE SPECI FI C ENERGY REQUI RED
Stirred media mill designs are generally unique, and mill selection is often based on manu
facturers’ testing, or alternatively, on tests run in thirdparty laboratories using labscale
versions of the mills being considered. Empirical methods, such as the Bond method, are
largely unsuccessful in determining power draw requirements for ultrafine grinds and
are unsuitable for stirred media mills. The Bond method, for example, incorporates a
correction factor (EF5) for fine product sizes. This correction factor was specifically
intended to correct for the inefficiency of ball mills using conventional media sizes when
producing very fine products. With stirred media mills, the media size limitation is
largely overcome and milling efficiencies are dramatically improved. Specific energies
derived from the Bond equations would be unacceptably conservative. It is not possible
to determine Bond Work Indices for the majority of fine and ultrafine grinding applica
tions, as the feed size would not meet the test requirements.
There is also considerable activity in the development of population balance models
(PBMs) for stirred media mills. The primary challenge is to accurately define the breakage
rates and the effects of operating parameters and media on the breakage parameters. It
TABLE 1 Particlesize distributions reported by different sizemeasurement techniques
Sample Screen P80, μm Microtrac P80, μm Malvern P80, μm Malvern P50, μm Sedigraph P50, μm
1 64.3 80.6
2 49.7 61.8
3 13.5 13.6
4 7.4 10.2
5 4.1 6.5
6 3.6 0.83
7 5.6 4.1
8 7.9 4.7
SELECTION AND SIZING OF ULTRAFINE AND STIRRED GRINDING MILLS 77
is unlikely that these techniques will be used for mill sizing in the foreseeable future, as
the challenges are considerable.
One benefit of the finer feed size typical of ultrafine grinding applications is that an
accurate laboratory test is possible. To accurately size a semiautogenous grinding (SAG)
mill, a minimum sample size of 100 kg is required for a laboratory test, and 20–100 t for
a pilot test due to the large feed particle size. Statistical relevance requires significantly
larger samples. For stirred mills, the feed top size is generally <200 Pm, and therefore a
sample mass of 100 g is sufficient for statistical reproducibility.
The only reliable method currently available for the selection of mills for ultrafine
grinding is a wellplanned and executed lab or pilotscale test regime. The test program
should include an evaluation of all the primary operating parameters (listed in the sec
tion on Media Considerations). Small sample size and relative ease of testwork make the
evaluation of multiple operating parameters feasible. The primary data generated with
such a laboratory or pilotscale test involve the relationship between specific energy and
grind. Depending on the type of mill being evaluated, the net specific energy generated
by the test mill can either be used directly without any correction factors or will need to
be adjusted. This task is currently best left to the supplier of the equipment being evalu
ated or to the laboratory where the tests are being executed.
Media Considerations
Use of the correct media is important for all grinding applications, but in the case of
ultrafine grinding in stirred media mills, it becomes the most important variable. Media
parameters that need to be considered include
Size
Type
Competency
Hardness
Stirred media mills use a wide variety of media from 25mmto6mm steel balls and
cylpebs commonly used in mills such as the Vertimill and Tower Mill to 1mmto5mm
highgrade alumina and sand media used in mills such as the SMD and the Netzsch/
IsaMill. Media can affect the specific energy required for an application by an excess of
30%. Any laboratoryscale testwork should therefore include a range of different medias.
Media availability is often regional, and transport costs are a significant contributor to
the total cost. When undertaking laboratory or pilot milling regimes, both local and
known media types should be tested.
Cost and performance are the primary considerations. The cost of conventional steel
media increases rapidly as the size decreases below 25 mm. This has largely restricted
the use of small steel media in mills such as the Vertimill, Tower Mill, and pin mills.
These restrictions are now largely eliminated due to the recent availability of cast media
and steel shot. Size limitations no longer exist, but there are still questions regarding the
media consumption and influence of iron (from the media) on some concentration pro
cesses. Material and sphericity are the strongest influencing factors for the 1mmto3mm
media commonly used in “fluidized” media mills.
Choice of media for laboratory and pilot milling tests should reflect the reasonable
choices available for the proposed plant. Using 2 kg of highalumina ceramic media at
US$70/kg may be acceptable for a pilot mill but would be wholly undesirable for a com
mercial installation requiring 10 t of media. Similarly, using lowcost local media may
provide acceptable specific energies for the application, but media breakage and wear of
mill components (which are not readily measurable in batch laboratory tests) may make
78 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
the media unacceptable for a commercial application. Higher quality media could also
reduce equipment size.
Comparisons between different types of mills should consider media costs and,
where reasonable, use similar media. It is advisable to always include at least one com
mercially available media of known quality in a test regime.
Media Size. Media size has a significant impact on the performance of mills in
fine and ultrafine grinding applications. It is often the primary limitation of the fineness
of grind possible from a type of mill. As feed and product sizes decrease, the energy
required to break a particle also decreases, and the frequency of the breakages per unit
mass increases. Excess energy from breakage events is largely converted to heat and does
not contribute to the grinding process. The most effective way to increase the frequency
of the grinding events and decrease the energy per event is to reduce the media size.
Consider the relationship between media size and the number of balls (or other media
shapes). Table 2 shows the relationship between the media size and the number of balls
per unit mass (or volume). The number of media per unit volume increases by the
inverse ratio of the media size to the third power. As the number of breakage events in a
mill is proportional to the number of media, dramatic improvements are possible
through the selection of the correct media size.
Because of the need to focus energy on everincreasing numbers of smaller particles,
media above a certain size may become completely ineffective in producing ultrafine
particles. A typical media size, specific energy plot for a Vertimill is given in Figure 9.
Significant reductions in the specific energy required to achieve a grind can be obtained
from the use of smaller media. The specific energy required to achieve a grind of 15 Pm
is more than 50% greater with 10mm media than it is with 5mm media. A 15Pm grind
was not achieved with the 18mm media over the range of specific energy tested.
Similar effects are found with the media used in “fluidized” media mills. Figure 10
shows the effect of media size on an SMD. In this particular case, the trend between
energy required and media size reflected in the Vertimill data is reversed. In this exam
ple, the mill requires over 50% more energy to the same grind with the 1mm media
than is required with the 2mm media. The 1mm media is too small to effectively break
down the larger feed particles. A finer, or softer, feed might see this trend reversed. Also
evident from the data is that the use of seasoned media charge is a requirement when
generating design data. Monosized media will not give a true reflection of the eventual
performance of an operation.
Media Type. Media type and shape also affect mill performance. Media can be
either ferrous or nonferrous. Nonferrous media includes highgrade alumina balls and
beads, lowergrade mullite ceramic beads, and silica sand. There is a large variety of
exotic medias available for highly specialized industrial applications where contamina
tion, and not cost, is the primary consideration. These medias are not normally suitable
for hightonnage applications. Table 3 shows some typical medias, as well as examples of
cost and relative wear (where applicable).
Steel media is most commonly used in ball mills, vertical mills with screw agitators,
and pin mills. Fluidized media mills most often use small ceramic beads or silica sand.
One consideration in the selection of a suitable milling technology for an application is
the effect of steel media on flotation recovery characteristics. In ultrafine grinding appli
cations, iron in solution from the media can contaminate sulphide mineral surfaces with
iron oxide, thereby affecting the grade and recovery characteristics of the flotation plant.
Iron in solution will also consume oxygen and affect some downstream processes. Under
these conditions, a nonferrous media may be preferred. Ball mills and stirred mills draw
less power with nonferrous media of a lower density than steel, and the mill sizes must
therefore be increased. Although fluidized media mills can be operated with ferrous
SELECTION AND SIZING OF ULTRAFINE AND STIRRED GRINDING MILLS 79
TABLE 2 Relationship between media size and the number of balls per unit mass
Ball Size,
mm
Surface Area,
m
2
/t
Number of Balls,
per t
Number of Balls,
normalized
20 83.3 66,315 1
15 111.1 157,190 2.4
10 166.7 530,516 8
5 333.3 4,144,132 62
3 555.6 19,648,758 296
2 833.3 66,314,560 1,000
0 10 20 30 40 50
100
10
1
Specific Energy, kWhr/t
18mm Steel Media
10mm Steel Media
5mm Steel Media
FIGURE 9 Effect of media size on mill efficiency (typical results for a Vertimill)
0 10 20 30 40 50 60 70 80 90 100 110
100
10
1
Specific Energy, kWhr/t
100% 2mm Media
100% 1mm Media
50/50 2mm and 1mm Media
FIGURE 10 Effect of media size on mill efficiency (typical results for an SMD)
80 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
media, they are generally designed to operate only with nonferrous media. Design mod
ifications would be required, and as such, this is not typically an option.
Mullite ceramic beads have supplemented highgrade alumina medias. These beads
are typically kaolin based and, while being hard and having good media properties, are
significantly more affordable than more traditional alumina media. This class of media is
seeing increasing application for use in ultrafine grinding.
Sand media should be near spherical, and aspect ratios better than 1:1.1 can be
obtained. With all stirred media mills, the way that media moves over itself affects the
energy utilization. Spherical media moves over itself relatively easily, whereas nonspher
ical shapes will have an increasing tendency to “lock up,” thereby consuming energy.
This effect is most readily seen with “fluidized media” mills. Figure 11 shows a specific
energy relationship for an SMD using sand and ceramic media. The ceramic media has
advantages in both the specific gravity of the media, the hardness, and the sphericity. For
this particular application, the use of ceramic media would reduce the energy required
by almost 50% over the use of sand media. Use of ceramic media for this application
would almost halve the milling capacity required, as well as the energy consumed. This
would need to be factored against the media consumption and cost differences between
the two media types. This level of improvement is not always found and is dependent on the
feed size, hardness, and grind required.
Media Competency. Of particular importance with sand media used in “fluidized
media” mills is the competency. The typical commercial sand media used in “fluidized” media
mills is normally used for filtration or other duties. Mechanical strength is not specified.
Ideal sand media is alluvial sand with rounded edges and is free of flaws. A common
problem associated with lowgrade sand media is that it has internal flaws and is not
able to withstand the forces associated with milling. Flawed media tends to break up
rapidly, degrading the media shape and size. Breakage increases media consumption.
Broken media also presents sharp edges, which increase the wear of the impeller and
wear liners in a mill. These faults are not typically evident during batch laboratory tests.
Care is required therefore when designing a plant based on the use of such media in
batch or pilot milling tests.
Media Hardness. One aspect sometimes not considered is the media hardness.
With steel media, hardness does not affect mill performance, only media consumption.
This is not the case where nonferrous media is used. Consider the environment inside a
typical “fluidized media” mill using sand or other nonferrous media. If the mineral being
ground is harder than the media itself, then the media will in effect be subject to commi
nution, and size reduction of the feed will be reduced (Krause and Pickering 1998).
These data depict the productsize distributions for a fluidized media mill grinding
quartz feed. These are batch grinds for the same duration. Quartz has a Mohs hardness
of approximately 7. The data show dramatically different results for media that is softer
TABLE 3 Media characteristics
Type Main Components Price, US$/kg
Approximate
Consumption, kg/kWhr
Steel media Fe/Cr 0.4–1.5 0.02–0.01
Attrition sand Quartz 0.2 0.02
Mullite ceramic Kaolin 0.8 0.01
Glass beads Quartz 3 0.01
Alumina ceramic Aluminum oxide 15 0.005
Zirconium silicate Zircon 18 0.005
Yttriumstabilized zirconia oxide Zircon 70 0.001
SELECTION AND SIZING OF ULTRAFINE AND STIRRED GRINDING MILLS 81
or harder than the feed. There is also a density effect, but this is primarily due to higher
absorbed powers with the higher specific gravity media. The primary variable is the
media hardness. Whereas sand media may be acceptable for soft minerals, when hard
minerals are encountered, highergradealumina media may be necessary. Figure 12
shows the productsize distributions for the various runs.
Operating Parameters. Outside of the effect of media selection on the perfor
mance of stirred mills, the strongest influencing variable is the slurry percent solids, or
viscosity. Vertimills, Tower Mills, and pin mills are less sensitive to slurry viscosity than
0 10 20 30 40 50 60 70 80 90
70
10
20
30
40
50
60
0
Specific Energy, kWhr/t
Sand Media
Ceramic Media
FIGURE 11 Effect of media type on performance of an SMD
1 10 100
100
10
Size,
P
r
o
d
u
c
t
,
C
u
m
u
l
a
t
i
v
e
%
P
a
s
s
i
n
g
3.56 Specific Gravity  8+ Mohs
3.27 Specific Gravity  8+ Mohs
2.71 Specific Gravity  7+ Mohs
2.68 Specific Gravity  6+ Mohs
FIGURE 12 Effect of media hardness
82 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
“fluidized media” mills. The former category of mills is able to operate over a reasonably
wide range of percent solids.
Fluidized media mills require more careful analysis of the effect of percent solids.
Two sets of trends are presented in Figures 13 and 14, showing typical relationships
between the grind and specific energy at various percent solids. The figures show the effect
of slurry percent solids. As can be seen, the effect of slurry percent solids on the perfor
mance of the mills is significant and trends are not always consistent. Lab or pilot milling
tests over a range of operating conditions are essential in order to determine the opti
mum operating conditions of the selected mill and, therefore, the correct mill selection.
CI RCUI T CONSI DERATI ONS
Classification
The primary options are open or closedcircuit milling, with the possibility to prescalp
the feed ahead of milling. Scalping can be either in dedicated classifiers or, alternatively,
by introducing the new feed to the mill discharge sump. Fine and ultrafine grinding circuits
benefit from classification in much the same way that conventional grinding circuits do.
Use of a classifier will reduce fines generation, produce a tighter product specification,
and reduce the overall energy requirements to achieve a specific grind. These benefits
have to be weighed against the cost, both capital and operating, of classification circuits.
For dry grinding circuits, the most commonly used classifiers are dynamic “whizzer” type
units.
The primary type of wet classifier for fine and ultrafine classification is the hydrocy
clone. As grinds become finer, cyclone sizes need to be reduced, and the percent solids in
the cyclone feed reduced. For sub10Pm grinds, cyclone sizes are around 25 mm to
50 mm. In addition, the product is very dilute, typically no more than 15% solids w/w.
The downstream circuit determines whether this dilution is prohibitive or not. As an
example, where a mill is used to regrind a concentrate, if the concentrate is reintroduced
to the bulk float cell feed, then the dilution may not be an issue. If, however, the mill
product were treated in dedicated float cells, then the dilute feed would not be accept
able. Added dewatering costs need to be factored into the overall economic evaluation.
Another consideration is the slurry percent solids requirement for the mill feed. A
typical concentrate from flotation cells would be too dilute to be used as mill feed for an
opencircuit mill. Good control of the feed density is essential to efficient operation. A
scalping cyclone would be required for this duty. The cyclone would have the combined
benefit of removing finished product in the feed, thereby improving milling efficiency
and producing a tighter product specification, as well as increasing the density of the
mill feed. As the milling circuit is open circuit, the total flow to the cyclones, and there
fore the number, is reasonable, and the cyclone overflow is recombined with the higher
density mill discharge, and further dilution can be limited.
Number of Mills
Milling circuits are always designed for maximum duty. These circuits rarely operate at
these values. During startup of greenfield plants, extended operating periods with signif
icantly reduced feed rates can last for up to 1 or 2 years. Also, where mills are included in
concentration circuits, feed rate to the mill will be affected by both the normal total cir
cuit feed fluctuations due to the hardness of the ore, as well as swings due to the grade of
the ore. In tough applications, these tonnage swings can be on the order of 100% to 200%.
Under these conditions, the turndown ratio (the minimum power at which the mill can
operate) of the selected mills must be considered. Mills installed in fine and ultrafine
SELECTION AND SIZING OF ULTRAFINE AND STIRRED GRINDING MILLS 83
regrind duties are normally constant speed and, therefore, constant power machines.
Ball mills and some stirred mills, such as the screwagitated mill and the pinagitated
mill, can operate at reduced ball levels and, therefore, power draws. These changes in
ball charge can be used to minimize overgrinding during known periods of reduced
throughput, such as startup periods, but typically cannot follow the normal capacity
swings resulting from shortterm changes in the feed rate. Fluidized media mills have
very limited turndown and should be operated at their design power if severe wear prob
lems are to be avoided.
0 10 20 30 40 50 60 70
100
10
1
Specific Energy, kWhr/t
30% Solids
40% Solids
50% Solids
FIGURE 13 Effect of slurry percent solids on fluidised media mill performance (trend 1)
0 10 20 30 40 50
100
10
1
Specific Energy, kWhr/t
30% Solids
40% Solids
50% Solids
FIGURE 14 Effect of slurry percent solids on fluidised media mill performance (trend 2)
84 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
What this means is that if the feed rate should drop below the design values, the
mills will continue to input the full power, and the product will be ground significantly
finer than required. This would be more of a problem for flotation circuits than leach cir
cuits, but the effect on the filters and thickeners in the plant must still be considered. The
most effective method to control overgrinding is with the use of multiple units fed from a
common source. Individual mills can be brought into service, or taken out, as required.
Caution should be exercised when specifying a single mill for an application if feedrate
variations are expected to be a problem.
Vertical mills with screw agitators do have a measure of control over this problem.
The mills include an internal recycle loop that controls the uprising velocity of slurry in
the mill. On specification, product can be removed from the mill and presented to classi
fication, thereby reducing overgrinding.
In some cases, specifications on the allowed coarse material in the product require
the installation of two mills in series (where opencircuit milling is required).
EXAMPLES
Ziniflex Century Ultrafine Milling Circuit
The Ziniflex Century lead–zinc deposit is a massive finegrained ore body located in the
northwestern region of Queensland, Australia. Minable reserves are approximately 100 Mt
at an average grade of 1.7% lead and 12.5% zinc. The flotation plant has a design
throughput of 5 Mtpy, and the first ore was introduced to the plant in November 1999
(Barnham and Kirby 2001).
The deposit was originally identified by Conzinc Riotinto of Australia (CRA) in 1990,
who conducted the definitive feasibility study during 1994 and 1995. Rio Tinto Zinc Cor
poration (RTZ) sold the project to Pasminco after the RTZ merger with CRA during 1997.
The main process problem with the Century ore is that the sphalerite, which is the
main zinc mineral, is very finely interspersed with silica (Hookham and Sutherland
1999). Ultrafine grinding to a P
80
of 6.5 Pm followed by flotation was considered the
most costeffective and practical solution. Fifteen Metso 355kW SMDs were installed at
Century to produce the desired product size with an additional six units in a zinc regrind
duty (Figure 15).
As mentioned previously, small media, smaller than 5 mm, is generally more energy
efficient than coarser media when the material feed size is greater than 100 Pm. This
accounts for the relatively low work input values to mill to a P
80
of 6.5 Pm. Century uses
a sand media with a mean size around 2 mm. Current operating conditions for the zinc
ultrafine grinding SMDs are shown in Table 4.
Individual SMD performance is monitored constantly, and mass flow rates and the
number of SMDs operating is adjusted continually to allow for the varying flotation mass
pull. This minimizes over and undergrinding in the zinc circuit. One limitation of all
fine and ultrafine grinding circuits is the lack of a reliable industrial online particlesize
analyzer capable of measuring sub20Pm particles.
Overall, it was found that the laboratory and pilot studies conducted before the con
struction of the final mineral processing plant provided accurate plant sizing data to the
engineering company charged with designing the plant and Metso process engineers.
This was aided by the quality of the original test samples, a combined effect of the quan
tity of samples mined and processed.
SELECTION AND SIZING OF ULTRAFINE AND STIRRED GRINDING MILLS 85
REFERENCES
Barnham, M., and E. Kirby. 2001. The design and commissioning of the Pasminco
Century process plant. SME Annual Meeting, Littleton, CO.
Hookham, M., and G. Sutherland. 1999. Century grinds its way to saleability. Australian
Mining Monthly (May).
Jankovic, A. 2001. ScaleUp of Tower Mill Performance Using Modelling and Simulation.
JKMRC/Amira P9M project. Third Progress Report. Melbourne, Australia: Amira
International.
Krause, C., and M. Pickering. 1998. Evaluation of ultrafine wet mineral milling using
carboceramics proppant products for attrition grinding media. Colorado Springs,
CO: Metso Minerals Optimization Services.
FIGURE 15 Photograph of 355kW SMD at Century
TABLE 4 Century ultrafine milling data
Operating Parameter Units Value
Mill power draw kW 310
Feed rate tph 8.6
Feed density % solids w/w 50
Feed size F
80
μm 35
Product size P
80
μm 6.5
Sand media consumption kg/kWhr 0.07
Millspecific energy kWhr/t 36
87
Effects of Bead Size on Ultrafine
Grinding in a Stirred Bead Mill
J. Yue
*
and B. Klein
*
ABSTRACT
The effects of bead size and composition on particle breakage rate, product size, and size dis
tribution, as well as the mill power consumption in a stirred mill are investigated and evalu
ated. The results confirm that monosized beads produce better milling performance than
bimodal beads as far as media wear is concerned. There exist optimum bead sizes for certain
feed sizes with respect to particle breakage rate, product size, and size distribution. The opti
mum ratio of bead size to feed size is confirmed at about 20:1.
I NTRODUCTI ON
Stirred bead mills are used widely for ultrafine grinding minerals and other materials to
particle sizes below a few micrometers. The energy needed for breakage is transferred
through the grinding media to the particles. For a given particle size, mineral type, and
specific energy input, the bead size plays a very important role in the comminution pro
cess, particularly with respect to the stress intensity (SI) and the stress number (SN)
(Kwade 1999).
From a physical–mechanical point of view, two conditions need to be satisfied for
breaking particles in a grinding mill: the grinding media must exert sufficient SI to the par
ticles, and there must be direct contact between the media and the particles. For fine
grinding, it has been recognized (Stehr, Mehte, and Herbst 1987; Kwade 1999) that the
specific energy consumption by stirred balls is less than that of tumbling mills. Due to
the highmedia volumetric loading in stirred bead mills, the SI and SN per unit volume
acting on particles is higher than in tumbling mills.
For brittle particles, massive fracture occurs when the overall stress acting on a par
ticle exceeds a critical value, resulting in particle disintegration and creating a large
number of fragments. Attrition is associated with smaller stresses exerted on particle
edges causing a continuous but slow loss of particle mass. It is understood that massive
fracture of brittle solids results in a faster particle breakage rate than attrition.
Kwade and others (Kwade 1999; Kwade and Schwedes 2002) assumed that the SI is
proportional to the kinetic energy of grinding media, and thus showed that it is propor
tional to both stirrer speed and bead size. The SN is also proportional to stirrer speed as
well as to the number of beads in the mill. Kwade (1999) presented the following expres
sions for SI and SN (Equations 1 and 2), which demonstrate these relationships:
* Mining and Mineral Process Engineering, University of British Columbia, Vancouver, Canada
88 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
(EQ 1)
SN = N
C
P
S
/N
P
(EQ 2)
where GM is the grinding media; d is the bead diameter; ȡ is the bead density; v
t
is the
bead tangential velocity; N
C
is the media contact number; P
S
is the probability that a par
ticle is caught and stressed sufficiently by the media; and N
P
is the number of product
particles inside the mill.
Blecher and Schwedes (1996) studied the motion of grinding media with disc stir
rers. They showed that a high tangential velocity gradient exists near the surfaces of the
discs and at the grinding chamber wall. At these areas, they suggested that the power
density is much higher than the mean value for the mill and that more intensive colli
sions take place. These highintensity collisions are believed to account for most of the
particle breakage in the mill.
The size of grinding balls is known to play a very important role in grinding in tum
bling mills. In particular, small balls are important for effective compression and attrition
of fine particles. The motion of mixture of beads, mineral particles, and water in a stirred
mill chamber occurs at a very high speed, but it is not clear if this is also the same sce
nario. So far, rheological concepts have been applied to the aqueous mineral suspension.
These concepts may be applied to the entire mixture within the mill as well. For aqueous
mineral suspensions, the relative viscosity K
r
increases with solid volume fraction )
(Bicerano, Douglas, and Brune 1999; Chong, Christiansen, and Baer 1971), and increas
ing particle size d (Clarke 1967). Similarly, Wang and Forssberg (1997) observed that
increasing bead volumetric loading ()) causes the power draw to intensify greatly. Also,
Herbst (1978) showed that increasing the bead size (d) increases the power draw. Both
of these results demonstrate that the volume content and size of grinding media affect
power in a similar manner to how particles affect rheology. The changes in the physical
properties of suspension that cause the viscosity to decrease correspond to changes in
bead composition that lead to a reduction in power requirements. Therefore, testing was
carried out to demonstrate the effect of bead size on grinding in stirred mills with respect
to breakage kinetics, product size, size distribution, and power consumption.
Research on the rheology of mineral suspensions has demonstrated that selected
compositions of bimodal particlesize distributions produce a “minimum” viscosity (Chong,
Christiansen, and Baer 1971). This bimodal size distribution is also characterized by a
maximum particlepacking fraction. A study was conducted to assess the effect of bimodal
beadsize distribution on power usage, grinding rate, and productsize distribution.
EXPERI MENTAL PROCEDURES
A Netzsch LME 4 horizontal stirred bead mill was used for the study. The mill is 400 mm
in length, 173 mm in diameter, and has a volume of 2.48 L. The stirrer has eight polyeth
ylene discs.
A loading charge of 80% of the mill volume was maintained using steelshot grinding
media with the diameters of 3, 2, 1, and 0.5 mm. The mill is equipped with a digital control
panel on which operating conditions were set and output parameters were displayed,
including temperature, pressure, and power. For tests with bimodal bead sizes, 2mm
and 0.5mm beads were used.
Grinding tests were performed on suspensions of quartz particles with feed sizes
(F
80
) of 83 Pm and 32 Pm (Figure 1). The feed slurry solids content was maintained at
35% solids by weight. Detailed test conditions are shown in Table 1. The slurry feed rate
SI SI
GM
v d
GM
3
U
GM
v
t
2
=
EFFECTS OF BEAD SIZE ON ULTRAFINE GRINDING IN A STIRRED BEAD MILL 89
was kept constant at 3.1 L/min. For tests with constant agitator speed (1,500 rpm), the
power usage was recorded as a response. For tests with constant power, the agitator
speed was adjusted to maintain this constant power and recorded. For tests with bimodal
bead sizes, a series of tests were performed using varying proportions of small beads
(0%–100%).
Particlesize analyses were performed using a Malvern Mastersizer 2000 laser
diffraction particlesize analyzer (PSA). The RosinRammler equation was fit to the data
to characterize the particlesize distributions.
RESULTS AND DI SCUSSI ON
Monosized Grinding Beads
The ultrafine grinding tests were conducted to evaluate the effects of bead size on parti
cle breakage rate, product size, size distribution, as well as agitator power consumption.
The bead size, agitator speed, and feed size were varied, whereas the pulp density was
kept constant during the tests.
Figure 2 shows how the specific breakage rate varies with bead size. At a constant
agitator speed, the specific breakage rate increased almost linearly with bead size. When
the mill power was kept constant, the specific breakage rate was highest at a bead size of
2 mm, and the rates were lower when using smaller and larger beads. Although the
“maximum” breakage rate at a bead size of approximately 2 mm was observed for both
0 1 10 100 1,000
100.0
10.0
1.0
0.1
0.0
C
u
m
u
l
a
t
i
v
e
U
n
d
e
r
s
i
z
e
,
%
FIGURE 1 Feedsize distributions
TABLE 1 Test program for evaluating grinding bead effects on a stirred bead mill
Test No.
Feed Size F
80
,
μm
Agitator Speed,
rpm
Pump Flow Rate,
L/min
Agitator Power,
kW
Bead Size,
mm
8 30 1,500 3.1 — 0.5, 1, 2, 3
9 83 — 3.1 3.7 0.5, 1, 2, 3
10 30 — 3.1 3.7 0.5, 1, 2, 3
11 83 1,500 3.1 — 0.5+2
12 83 — 3.1 3.7 0.5+2
90 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
feed materials (F
80
32 Pm and F
80
83 Pm), it cannot be concluded that this bead size is
optimum for coarser or finer feed materials. The significance of bead size was more pro
nounced for the coarser feed material.
When changing the bead size from 2 mm to 3 mm, the corresponding decrease in
agitator speed had a greater impact on breakage rate than the increase in bead size. Using
a smaller bead size caused a decrease in breakage rate implying a lower SI and a less
effective use of power. There is insufficient information available to make conclusions
about the reasons for the “optimum” bead size. Factors such as SI and SN likely play a
role. However, the results of this study reveal that an optimum bead size can be selected.
Figure 3 shows how bead size affects product fineness (P
80
). At constant agitator
speed, increasing the bead size caused the product to become finer. However, when
grinding power was kept constant, there was an optimum bead size. For the 32Pm feed,
the product size was finest when using a bead size of 1 mm. For the 83Pm feed, the
2mm bead size produced the finest product. In this case, when larger beads were used,
the agitator speed had to be decreased to maintain a constant power. Conversely, when
smaller beads were used, a higher agitator speed was required.
The bead size also affects productsize distributions. Figure 4 is a plot of the Rosin
Rammler particlesize distribution coefficient, b, versus bead size; a high value of b corre
sponds to a narrow size distribution. The productsize distributions were characterized
using the distribution coefficient from the fitted RosinRammler equation. At constant
power, larger beads produced a narrower size distribution. The results agreed with those
obtained by Wang and Forssberg (1997). Larger beads have higher kinetic energy, there
fore, they have greater potential to cause massive fracture, producing narrower product
distribution. The smaller beads produce a wide size distribution, possibly due to the
lower SI promoting attrition over massive fracture.
For mineral processing, a narrower particlesize distribution is usually preferred as it
benefits downstream flotation and dewatering. If the goal is to produce a narrow product
size distribution, the results presented in Figure 4 suggest that there is an optimum bead size.
0 1 2 3 4
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.00
S
p
e
c
i
f
i
c
B
r
e
a
k
a
g
e
R
a
t
e
Bead Size, mm
FIGURE 2 Specific breakage rate versus bead size
EFFECTS OF BEAD SIZE ON ULTRAFINE GRINDING IN A STIRRED BEAD MILL 91
Figure 5 shows the relationship between bead size and agitator power draw (agita
tor speed was kept constant). Tests were performed with and without slurry to demon
strate the background effect of the grinding beads. Increasing the bead size causes the
power draw to increase. In the presence of slurry, the power draw increases more sharply
than without slurry. Such a relationship also can be interpreted by rheological concepts.
Researchers (Chong, Christiansen, and Baer 1971; Clarke 1967) found that slurry viscosity
is lowest at an intermediate particle size and increases above and below this particle size.
0 1 2 3 4
10
20
30
40
50
60
0
Bead Size, mm
FIGURE 3 P
80
versus grinding bead size
0 1 2 3 4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0
0.4
Bead Size, mm
P
a
r
t
i
c
l
e

S
i
z
e
D
i
s
t
r
i
b
u
t
i
o
n
C
o
e
f
f
i
c
i
e
n
t
,
b
FIGURE 4 Particlesize distribution modulus versus bead size
92 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
These results show that adding coarser beads increases the power draw, which implies
an increased viscosity. The particles are filled into the bead voids and nipped, causing a
strong shearing resistance that needs to be overcome by bead kinetic energy.
The experiments showed that at constant agitator speed, increasing the bead size
gave the following results:
Increased breakage rate
Finer product size
Narrower productsize distribution
Increased power usage
These results are supported by Equation 1, which shows that using larger beads will
increase the SI. These results demonstrate that factors that increase the SI will increase
grinding kinetics and produce a finer product. However, there is a tradeoff, as these fac
tors also cause an increase in power usage.
In practice, mills are operated near the upper power limit. It is, therefore, important
to understand how to “optimize” grinding at this constant upper power limit. The results
reveal that at constant power draw, there was an optimum bead size with respect to a
high breakage rate, fineness of the product, and narrow size distribution. The reasons for
the optimum bead sizes cannot be readily explained by Equations 1 and 2.
Mankosa, Adel, and Yoon (1986) suggested that the optimum size ratio between
beads and particles (mean size) is 20:1. Fadhel and Frances (2001) believed the ratio lies
between 20:1 and 200:1. Figure 6 is a schematic diagram in which the ideal geometric
dimensions of both beads and particles are assumed. Figure 6a shows there is one parti
cle nipped by beads; Figure 6b shows four to five particles are captured in the bead void
for optimum breakage. The results calculated in Table 2 confirm that the “optimum”
ratio between bead and particle is close to 20:1. The test data shown in Figures 2, 3, and
4 support Mankosa, Adel, and Yoon’s conclusion. In fact, the maximum breakage rate
occurs when particles are large enough to have a high probability of contacting beads
but are small enough to be effectively caught and broken by the media. Above this limit,
the breakage kinetics and sizereduction ratio decrease. This decrease can be explained
by geometric aspects of particle “nipping” or by a lower SN (in Equation 2, P
S
decreases
and N
P
increases simultaneously), although the SI is kept constant.
0.0 0.5 1.0 1.5 2.0 2 5
1
2
3
4
5
6
0
Bead Size, mm
With Load of Slurries
Without Load of Slurries
P
o
w
e
r
,
k
W
FIGURE 5 Agitator power versus bead sizes
EFFECTS OF BEAD SIZE ON ULTRAFINE GRINDING IN A STIRRED BEAD MILL 93
Figure 7 presents the possible scenarios for impact and compression breakage in
stirred bead mills. Figure 7a shows collision between a particle and a bead. Figure 7b
shows that particles captured in the void between beads are impacted indirectly by
another bead. Figure 7b illustrates particles being “rollcrushed” between beads. For
Figure 7a, massive fracture occurs when the bead impacts the particle, but the probability
of this impact is relatively low due to the vast size difference between beads and particles.
Figure 7b demonstrates the collision as balls hit the bed, in which impact energy is trans
ferred through beads to particles indirectly, resulting in particle fracture. This is perhaps
responsible for most of the particle breakage in stirred mills. The probability of Figure 7c
is also quite high. In this case, compression occurs between beads, which act as numer
ous tiny rollcrushers working in a mill. If particle sizes are too small, compression may
be replaced by attrition.
BeadSize Distribution
There is the theoretical possibility of increasing charge mass load while keeping load vol
ume constant by selecting the beadsize distribution. In particular, the packing density of
r
R
(a) (b)
FIGURE 6 Schematic diagram of beads and particles in a stirred mill
TABLE 2 Relationship between bead diameter (D) and maximum particle size (d) based on
geometries shown in Figure 6 where d = 2(R/cos 30 – R)
D, mm 3 2 1 0.5
(a) d, mm 0.46 0.3 0.15 0.076
(b) d, mm 0.15 0.1 0.05 0.025
(a) (b) (c)
FIGURE 7 Schematic diagram of impact and compression breakage mechanism
94 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
grinding media can be increased by filling the voids between the largest beads with
smaller beads, producing a bimodal beadsize distribution. The maximum packing den
sity for a bimodal distribution is achieved when the ratio of particle sizes (small to large)
is less than about 0.1 and when approximately 30% of the total volume is comprised of
the smaller particles (Furnas and Anderegg 1931). Chong, Christiansen, and Baer (1971)
found that the viscosity of suspensions at a fixed solids concentration was at a minimum
when using this bimodal size distribution. The implication is that a bimodal beadsize
distribution would correspond to a minimum power draw.
For stirred mill grinding, the grinding media typically occupies about 80% of the
mill volume. Introducing fine beads into a bed of coarse beads would theoretically increase
the volume of solid fraction of beads by 24%, from about 60% to about 84%. Therefore,
adding small beads has the potential to increase the charge mass by 40% (24/60) while
maintaining a constant charge volume. There likely would be a consequence of increasing
the charge mass to grinding rate and power usage; however, there are practical limita
tions to maintaining such a beadsize distribution during operation of an industrial mill.
In practice, there is a distribution of bead sizes in grinding mills. During grinding,
the beads wear, reducing their size, and they eventually pass through the mill product
screen and are rejected. There is a tradeoff between optimizing bead size with respect to
grinding (size reduction, power usage, and productsize distribution) and costs associ
ated with media consumption. It is therefore critical to determine the appropriate screen
aperture to satisfy both of these criteria.
A set of experiments was conducted to assess the effect of changing the beadsize
distribution by increasing the proportion of fine beads. In particular, grinding tests were
conducted using mixtures of 2mm and 0.5mm beads with compositions that varied
from 100% coarse beads/0% fine beads to 0% coarse beads/100% fine beads. In order to
maintain a constant charge volume for all experiments, the bulk densities of mixtures
were determined (Figure 8). The graph shows that as the proportion of fine beads is
increased, the bulk density increases; this result can be explained by the packing theory
described above. However, there is an unusual dip in the bulkdensity curve correspond
ing to fine fractions between 30% and 80%. It is suggested that this dip is a result of
imperfect mixing that prevented optimum packing of the beads. Since optimum mixing
likely would not be achieved in an operating mill, the measured bulk densities were used
to prepare media compositions for the study.
The experimental work was aimed at evaluating the effect of media wear. Tests were
conducted using increasing proportions of fine media and determining the specific
breakage rate, product particle size, size distribution, and power consumption. The
results are shown in Figures 9–12.
Under conditions of maintaining constant mill power or maintaining constant agita
tor speed, the specific breakage rate decreases as fine bead fraction increases (Figure 9).
The decrease in breakage rate corresponds to an increase in product size (Figure 10). It
is interesting to note that the productsize distribution becomes wider by increasing the
media fineness. The trend towards a wider size distribution and slower grinding rate can
be interpreted as a change in grinding mechanisms from primarily massive fracture for
the coarse media to attrition for the finer media. Figure 12 shows that the total power
draw and net power draw decreased as the bead size became smaller. Results from Fig
ures 9, 10, and 11 show that the existence of fine beads affects milling performance
greatly with respect to particle breakage rate, product size, and size distribution, even
when the power input is kept constant. Figure 12 shows that replacing fine beads with
coarse ones results in a decrease in power usage. The consequence of the buildup of fine
media is that energy is not efficiently used and mills are not operated under “optimum”
conditions.
EFFECTS OF BEAD SIZE ON ULTRAFINE GRINDING IN A STIRRED BEAD MILL 95
All these results confirmed that the fine beads existing in grinding media have negative
effects in particle breakage rate, product size, and size distribution, and are not optimal
for grinding efficiency in stirred mills. The trends can be attributed to the lack of sufficient
SI in stirred mills when the beads become too fine.
In a ball mill used as secondary grinding, for example, even small balls have suffi
cient energy to break particles through tumbling or rolling due to their relatively large
diameter and heavy mass. However, it is the bead kinetic energy (high speed with a nec
essary mass) that breaks particles in a stirred mill. Where the mass of fine bead is low, it
may not exert sufficient SI onto particles at a certain circumference speed. On the other
hand, the mill power draw (bead viscosity) decreases due to the existence of small
beads, as explained by “the ballbearing effects” theory. This suggests that Chong, Chris
tiansen, and Baer’s (1971) findings for bimodal suspension apply to grinding beads in a
20 0 40 60 80 100
4.7
4.8
4.9
5.0
5.1
5.2
4.6
B
u
l
k
D
e
n
s
i
t
y
,
g
/
c
m
3
Fine Fraction, %
FIGURE 8 Bead bulk density versus fine fraction of bimodal beads
20 0 40 60 80 100
0.14
0.12
0.10
0.08
0.06
0.04
0.02
0.16
0.00
S
p
e
c
i
f
i
c
B
r
e
a
k
a
g
e
R
a
t
e
Fine Bead Fraction (0.5 mm), %
Agitator Power 3 7 kW
Agitator Speed 1,500 rpm
FIGURE 9 Specific breakage rate versus percent fine bead
96 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
stirred mill. When beads move at a high speed, the fine beads behave as bearings and
can alter the directions of coarser ones adjacent to them, thus buffering and lowering the
kinetic energy of coarse beads. The consequence is that the impact and compression
force for breaking down particles from larger beads is weakened, and likely accounts for
the main breakage mechanisms in stirred mills.
CONCLUSI ONS
Monosized and bimodal bead distributions are studied in a stirred bead mill for grinding
quartz suspensions. The effects of bead size and composition on particle breakage rate,
product size, and size distribution, as well as the mill power consumption are investi
gated and evaluated.
20 0 40 60 80 100
70
80
60
50
40
30
20
10
90
0
Fine Bead Fraction (0.5 mm), %
Agitator Power 3 7 kW
Agitator Speed 1,500 rpm
FIGURE 10 Product size P
80
versus grinding bead size
20 0 40 60 80 100
0.90
1.00
1.10
1.20
1.30
1.40
1.50
0.80
Fine Bead Fraction (0.5 mm), %
P
a
r
t
i
c
l
e

S
i
z
e
D
i
s
t
r
i
b
u
t
i
o
n
C
o
e
f
f
i
c
i
e
n
t
,
b
FIGURE 11 Particle distribution modulus versus percent fine bead
EFFECTS OF BEAD SIZE ON ULTRAFINE GRINDING IN A STIRRED BEAD MILL 97
At a constant agitator speed for monosized grinding beads, increasing bead size
increases mill power consumption and thus results in a fast particle breakage rate, fine
product size, and narrow size distribution. When the mill power draw is kept constant,
there exist optimum bead sizes for certain feed sizes with respect to particle breakage
rate, product size, and size distribution. The optimum ratio of bead size to feed size is
confirmed at about 20:1.
Similar to suspensions, bimodal bead viscosity (power consumption) is lower than
monosized beads. The fine beads existing in grinding media have negative effects in par
ticle breakage rate, product size, and size distribution due to ballbearing effects and are
not in favor of grinding efficiency for stirred mills.
REFERENCES
Bicerano, J., J. Douglas, and D. Brune. 1999. Model for the viscosity of particle disper
sions. Journal of Macromolecular Science Reviews C39:561–642.
Blecher, L., and J. Schwedes. 1996. Energy distribution and particle trajectories in a
grinding chamber of a stirred ball mill. International Journal of Mineral Processing
44–45:617–627.
Chong, J.S., E.B. Christiansen, and A.D. Baer. 1971. Rheology of concentrated suspen
sions. Journal of Applied Polymer Science 15:2007–2021.
Clarke, B. 1967. Rheology of coarse settling suspensions. Transactions of the Institute of
Chemical Engineering 45:T251–T256.
Fadhel, H.B., and C. Frances. 2001. Wet batch grinding of alumina hydrate in a stirred
bead mill. Powder Technology 119(2–3):257–268.
Furnas, C.C. 1929. Pages 75–79 in Flow of Gases through Beds of Broken Solid. Bulletin
307. Washington, DC: U.S. Bureau of Mines.
Furnas, C.C., and F.O. Anderegg. 1931. Grading aggregates. Industrial and Engineering
Chemistry 23(9):1052–1064.
20 0 40 60 80 100
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
5.5
6.0
0.0
Fine Bead Fraction (0.5 mm), %
P
o
w
e
r
D
r
a
w
,
k
W
Net Power on Slurries
Total Power on the Shaft
Power on Beads Only
FIGURE 12 Agitator power versus percent fine bead
98 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
Herbst, J.A. 1978. Fundamentals of fine and ultrafine grinding in a stirred ball mill.
Pages 452–470 in Proceedings of the International Powder and Bulk Solids Handling
Conference. Chicago, IL: Industrial and Scientific Conference Management.
Kwade, A. 1999. Wet comminution in stirred media mills—research and its practical
application. Powder Technology 105:14–20.
Kwade, A., and J. Schwedes. 2002. Breaking characteristics of different materials and
their effect on stress intensity and stress number in stirred media mills. Powder Tech
nology 122:109–121.
Mankosa, M.J., G.T. Adel, and R.H. Yoon. 1986. Effect of media size in stirred ball mill
grinding of coal. Powder Technology 49(1):75–82.
Stehr, N., R.K. Mehte, and J.A. Herbst. 1987. Comparison of energy requirement for con
ventional and stirred ball milling of coalwater slurries. Coal Preparation 4:209–226.
Wang, Y., and E. Forssberg. 1997. Ultrafine grinding and classification of minerals.
Pages 203–214 in Comminution Practices. Edited by S.K. Kawatra. Littleton, CO:
SME.
99
Specific Energy Consumption, Stress
Energy, and Power Draw of Stirred
Media Mills and Their Effect on the
Production Rate
Arno Kwade
*
ABSTRACT
Stirred media mills are capable of grinding various products down to sizes in the nanometer
range. The specific energy required to achieve a certain product quality mainly depends
upon the stress energy and can be predicted for different sets of operating parameters using
a stress model and grinding tests. Moreover, the effect of several operating parameters on the
power draw of stirred media mills can be described by the relation between the power num
ber and the Reynolds number. By combining the relations for specific energy and power
draw, the influence of important operating parameters on production rate can be predicted.
Moreover, the production rate of the grinding process can be increased and maximized.
Examples will show that, depending upon the product and the formulation, different operat
ing parameters are important for increasing the production rate of the grinding process.
I NTRODUCTI ON
Currently, the selection, design, and sizing of grinding and dispersion processes are
based mainly on practical experience and empirical relations. For determining optimum
operating parameters and the related production rate of a mill, usually several grinding
or dispersing tests have to be carried out. But still production capacity cannot be deter
mined precisely for operating conditions not tested previously.
A stirred media mill is used most economically if the required product quality is pro
duced with maximum production rate at the lowest possible operating costs. The main
parameters influencing the production rate are given by the following expression:
(EQ 1)
where m
P,Ch
is the solids mass or pigment mass of the charge; t
Ch
is the grinding or dis
persion time of the charge; P
GC
is the power consumed inside the grinding chamber
(motor power minus noload power); and E
m
is the specific energy required for a certain
product quality. The production rate is proportional to the power consumed inside the
* Institute for Particle Technology, Technical University of Braunschweig, Germany
m
·
P
m
·
P
m
P,Ch
t
Ch

P
GC
E
m
 = =
100 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
grinding chamber of the stirred media mill and is inversely proportional to the specific
energy, which is necessary to produce the required product quality. The power con
sumed inside the grinding chamber is the overall power draw decreased by the noload
power. The maximum production rate is gained if the following are true:
The power draw has a maximum value.
The needed specific energy for the production of the required product quality is
minimal.
In order to increase the production rate efficiently, it has to be known which operat
ing parameters have the greatest effects on production rate. To derive a relation between
the production rate and the operating parameters, it must be known how the power
draw and the specific energy requirement for the production of a certain product quality
are affected by the different operating parameters. The power draw of a certain opera
tion condition can be determined by applying the power characteristics of the stirred
media mill (i.e., the relation between power number and Reynolds number). The depen
dency of the specific energy on important operating parameters can be determined using
a socalled stress model, which focuses on the elementary processes taking place in
grinding and dispersion processes.
MODEL OF STRESSI NG PART I CLES I N A MI LL
A socalled “stress model” was developed that describes the physical processes in mills
(e.g., the stressing and grinding of particles). Generally, there are two perspectives from
which stressing and grinding of particles can be observed (Kwade 2002; Kwade 2004).
One perspective is from the view of the product particle (i.e., how intensively and how
often the particle is stressed). The other perspective is from the view of the mill (i.e.,
how strongly and how frequently the mill can stress particles).
The basic principle of the productrelated stress model is that for a given feed parti
cle, the product quality and fineness achieved in a grinding or dispersing process is
determined by
Type of stress event including particle configuration (e.g., compression and shear,
impact)
How often each feed particle and its resulting fragments are stressed and, thus, by
the number of stress events of a feed particle, SN
F
How high the specific energy or specific force at each stress event is and, thus, by the
stress intensity at each stress event, SI
In actual comminution processes, the feed particles and the resulting fragments are
not stressed equally often with the same stress intensity; the number of stress events and
the intensity of these stress events is different for each feed particle. Thus, the number of
stress events and the stress intensity can only be characterized by distributions, not by
single numbers. Both distributions, particularly the magnitude of SN
F
and SI, depend on
the operating parameters. The width of the distribution of the stress number is deter
mined primarily by the residence time distribution of the particles in the mill. The width
of the distribution of the stress intensity depends mainly on how the stress energies dif
fer locally over time. The exact determination of these distributions is difficult but
should be possible using numerical methods in the future. For an engineering approach,
in most cases it is sufficient to use a characteristic value of the stress intensity instead of
an entire distribution (Kwade, Blecher, and Schwedes 1996; Kwade 1996, 1998, 1999).
It could be shown that as long as the following are true—only one particle is sufficiently
stressed; the tangential velocity of the grinding media are proportional to the tip speed of
the discs; the viscosity of the suspension is not too high; the young modulus of the product
ENERGY CONSUMPTION, STRESS ENERGY, AND POWER DRAW OF STIRRED MEDIA MILLS 101
material is low compared to the young modulus of the grinding media; and the geometry
and size of the mill is not changed—then the stress intensity in stirred media mills can be
described by the following expression called stress energy of the grinding media, SE
GM
:
(EQ 2)
where d
GM
is the grinding media size; v
t
is the stirrer tip speed; and µ
GM
is the grinding
media density. The stress energy of the grinding media, SE
GM
, is a measure for the maxi
mum stress energy inside the grinding chamber. As long as the geometry and size of the
grinding chamber are constant and, thus, the stress energy distribution does not
change, it is also a measure for the mean stress energy inside the grinding chamber. If
the geometry and/or the size of the grinding chamber changes, the change in the stress
energy distribution has to be taken into account (Stender, Kwade, and Schwedes
2002).
The stress intensity determines how effective the specific energy transferred to the
product at one stress event is transposed into product quality and product fineness. The rela
tionship between increase in product quality and stress intensity depends on the breakage
characteristics of the grinding material. The principal effect of the stress intensity on the
product quality follows from Figure 1 for the case that single particles are stressed. In
Figure 1, the product quality (e.g., specific surface) is depicted as the function of the rel
ative stress intensity. The relative stress intensity is defined as the ratio of stress intensity
SI to the optimum stress intensity SI
opt
. The stress intensity has an optimum value and
the energy utilization, a maximum value EU
max
, when the energy is just sufficient to
break a particle, to disagglomerate agglomerate or to disintegrate a microorganism. If
the stress intensity is smaller (SI/SI
opt
< 1), the product fineness increases with the stress
SE SE
GM
· d
GM
3
v
t
2
µ
GM
=
1 0.1 10 100
Relative Stress Intensity SI/SI
opt
[–]
S
p
e
c
i
f
i
c
S
u
r
f
a
c
e
Δ
S
m
,
D
i
s
i
n
t
e
g
r
a
t
i
o
n
D
e
g
r
e
e
,
m
2
/
g
%
R
e
a
l C
o
m
m
in
u
tio
n
Disagglomeration and Disintegration
FIGURE 1 Specific surface and disintegration degree as function of relative stress intensity
102 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
intensity for all three applications (with different slopes). The slope of the curve depends
on the material under investigation.
However, if the stress intensity is larger (SI/SI
opt
> 1), there are obvious differences.
In a socalled “ideal disagglomeration” process, the agglomerates are initially broken to
their primary particles at a certain stress intensity. By “ideal disagglomeration,” agglom
erates can also be disagglomerated by abrasion of single primary particles from the
agglomerate. In the case of ideal disagglomeration, the product quality is constant
because the agglomerate is already fully disagglomerated at the stress intensity, called
optimum stress intensity in this paper. The same holds true for the disintegration of
microorganisms if all microorganisms were already disintegrated at optimum stress
intensity. During grinding crystalline materials or during “nonideal” disagglomeration,
the specific surface will further increase with increasing stress intensity, but usually at a
lower slope because the energy utilization is less than at the optimum stress intensity.
The more difficult the feed material is to grind, the greater the slope of the curve. There
fore, two boundary cases can be defined:
1. The upper limit can be assumed to be the case when the product quality (e.g.,
new surface area) increases proportionally to the stress intensity. In this case, the
slope of the curve is 1.
2. The lower limit is equal to an ideal disagglomeration process, in which the size of
the fragments does not depend on the stress intensity, as long as the stress inten
sity is higher than the optimum stress intensity. In this case, the slope of the
curve and the value of exponent is 0.
For SI > SI
opt
, the following equation can be stated:
(EQ 3)
with a = 0 for ideal disagglomeration/disintegration
0 < a < 1 for grinding crystalline materials
If the energy utilization (EU) is used to characterize the effect of the stress intensity,
Figure 2 follows from Figure 1. The EU is defined as the ratio of the newly produced specific
surface ǻS
m
to the specific energy E
m
required to produce ǻS
m
. In general, the energy
utilization also can be considered as the ratio of increase in product quality to the spe
cific energy required to produce the increase in product quality. If a single particle is
stressed once, the stress intensity corresponds to the specific energy for stressing the par
ticle, so that in this case, the energy utilization is equal to ǻS
m
/SI. At the optimum stress
intensity SI
opt
, the energy utilization has its maximum value EU
max
at which a certain
specific surface can be produced with a minimum of specific energy.
In Figure 2 the relative energy utilization EU/EU
max
is plotted versus the relative
stress intensity SI/SI
opt
. At the optimum, both ratios are 1. To the left of the optimum,
the energy utilization increases with increasing stress intensity for all three applications
in a similar, but not equal, mode. To the right of the optimum, the relative energy utiliza
tion decreases. In the case of disagglomeration and disintegration, the slope of the curve
is –1, because the stress intensity or specific energy, respectively, added to the optimum
stress intensity is not used at all and does not affect the product fineness or the disinte
gration rate. Therefore, the energy utilization is inversely proportional to the specific
energy of a single stress event and, thus, to the stress intensity.
If a single crystalline particle is stressed to the right of the optimum stress intensity,
the relative increase in product quality is smaller than the corresponding relative
AS
m,
A
SI
SI
opt

\ .
 
a
·
ENERGY CONSUMPTION, STRESS ENERGY, AND POWER DRAW OF STIRRED MEDIA MILLS 103
increase in stress intensity, so that the EU decreases to the right of the optimum. The
more difficult the feed material is to be ground, the smaller the decrease in energy utili
zation. The upper limit is probable if the energy utilization stays constant while the
stress intensity increases. Thus, for SI > SI
opt
, the following can be stated:
(EQ 4)
with a = 0 for disagglomeration action /cell disruption
0 < a < 1 for realtime grinding
RELATI ON BETWEEN SPECI FI C ENERGY, STRESS ENERGY,
AND STRESS NUMBER
The total energy transferred to the product particles can be determined by the summa
tion of all stress energies SE
i
of the individual stress events. The stress energy is a mea
sure for the stress intensity and independent of the mass of the product stressed between
two grinding media. The sum of all stress energies is again equal to the product of the
number of stress events and a mean stress energy. The specific energy, E
m,P
, actually
transferred to the product particles is obtained by relating the total energy to the total
mass of the product. Due to friction and other losses, the specific energy consumed by
the comminution device or mill, E
m,M
, is not equal but proportional to the specific
energy, E
m,P
, transferred to the product particles. If the losses are taken into account by
an energy transfer factor, v
E
, the two characteristic numbers SN
M
= t
C
· SF
M
and SE can
be related to the specific energy consumed by the mill, as shown in the following:
1 0.1 10 100
Relative Stress Intensity SI/SI
opt
[–]
1
4
0.1
0.01
R
e
l
a
t
i
v
e
E
n
e
r
g
y
U
t
i
l
i
z
a
t
i
o
n
E
U
/
E
U
m
a
x
[
–
]
R
e
a
l C
o
m
m
in
u
tio
n
D
i
s
a
g
g
l
o
m
e
r
a
t
i
o
n
a
n
d
D
i
s
i
n
t
e
g
r
a
t
i
o
n
FIGURE 2 Relative energy utilization as a function of the relative stress intensity
EU
EU
max

SI
SI
opt

\ .
 
a–1
·
104 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
(EQ 5)
where SE
i
is the stress energy at stress event i; SN
M
is the total number of stress events to
achieve a certain product quality; E
m,P
is the specific energy transferred to the product
particles; v
E
is the energy transfer factor of the mill; and E
m,M
is the total specific energy
consumption of the mill. The specific energy transferred to the product E
m,P
is also
named effective specific energy. This specific energy is the part of the total energy con
sumption of the mill, which is actually used for stressing the particles.
Under the assumption of a constant shape of the stress frequency distribution and
the stress energy distribution based on Equation (1), the product quality of a certain
product mass m
P
is already fixed if two of the three parameters—total number of stress
events, SN
M
, mean stress energy, , and specific energy transferred to the product,
E
m,P
—are set. Thereby E
m,P
is fully determined if the energy transfer factor, v
E
, and the
specific energy consumed by the mill, E
m,M
, are known. The energy transfer factor is
among others proportional to the filling ratio of the grinding media, ¢
GM
because the
portion of the energy that can be used for comminution and dispersing inside the grind
ing chamber is proportional to the part of the volume filled with grinding media (Kwade
2002, 2004). In addition, the ratio of surface area to grinding chamber volume affects
the energy transfer factor (Stender, Kwade, and Schwedes 2002) because energy is lost at
the grinding chamber wall due to friction between the grinding media and the wall.
APPLI CATI ON OF THE STRESS MODEL
In order to verify the stress model, batch grinding tests with limestone (median size
approximately 60 µm) in a stirred media mill with discstirrer geometry (see Figure 3)
under different operational conditions (different sizes and densities of the grinding
media as well as different tip speeds) were carried out (Kwade, Blecher, and Schwedes
1996; Kwade 1996; Kwade and Schwedes 1997). In Figure 4, the specific energy used to
produce a median size of 2 µm is shown as function of the stress intensity of the grinding
media, SE
GM
, for different sets of operating parameters. The grinding media size varied
from 399 µm to 4,000 µm, the stirrer tip speed from 6.4 to 12.8 m/s, and the grinding
media density from 2,894 kg/m
3
to 7,550 kg/m
3
. It can be seen that the relation between
specific energy required to obtain a median size of 2 µm and stress energy can be
described by one fitted curve. Thus, the characteristic number SE
GM
can be used to
describe the effect of the operating parameters’ diameter and density of the grinding
media as well as stirrer tip speed on the specific energy required to obtain a certain pro
duct fineness (here, x
50
= 2 µm) in a combined form. From the curve, an optimum stress
energy can be ascertained at which a certain product quality can be achieved with a min
imum of specific energy. If the operating parameters’ size and density of the grinding
media as well as stirrer tip speed are chosen in a way that results in optimum stress
energy, the specific energy requirement for a certain grinding task is the lowest.
As Figure 5 shows, the influence of stirrer tip speed as well as size and material of the
grinding media on the result of different comminution processes can be described well
by two of the three parameters—stress number, stress intensity, and specific energy. The
relations among product quality (fineness or disintegration rate), stress intensity, and spe
cific energy depend on the breakage behavior of the material. If the relations among
product quality, stress intensity, and specific energy are known, statements regarding the
breakage characteristics and comminution behavior of the material can be made.
SE
i
i=1
SN
M
_
m
P tot ,

SN
M
SE
m
P tot ,
 E
m P ,
v
E
E
m M ,
= = =
SE
ENERGY CONSUMPTION, STRESS ENERGY, AND POWER DRAW OF STIRRED MEDIA MILLS 105
FIGURE 3 Schematic drawing of a stirred media mill with discstirrer geometry
0.1 0.01 0.002 1 10
1,000
800
600
400
200
00
k
J
/
k
g
Stress Intensity SI
GM
[10
–3
Nm]
ρ
GM
[kg/m
3
] = 2,894 7,550
v
t
[m/s] = 6.4
v
t
[m/s] = 9.6
v
t
[m/s] = 12.8
d
GM
[μm] = 399–4,000
x
50
= 2 μm
ϕ
GM
= 0.8
c
m
= 0.4
FIGURE 4 Specific energy required for a median size of 2 μm as a function of stress energy
(limestone)
106 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
These relationships are illustrated by comparing comminution results of the follow
ing seven different materials: pigments (Stadler et al. 1990), yeast cells (Bunge 1992),
synthetically produced SiO
2
aggregates (median size of feed particles about 22 ȝm),
waterbased ink (Vock 1997), printing ink, limestone, and fused alumina (median size of
feed particles about 33 ȝm; Becker, Kwade, and Schwedes 2001). The comminution
behavior of the seven materials can be compared by looking at the influence of the
stress intensity on the specific energy required for a certain product quality. Thus, in Fig
ure 5, the ratio of the specific energy required for a certain product quality to the mini
mum specific energy required for the same product quality is presented as a function of
the stress energy related to the optimum stress energy in a log–log diagram. The ratio
SE/SE
opt
is equal to the ratio SI/SI
opt
since the mass of the product stressed at on stress
event is eliminated by dividing the stress intensity by the optimum stress intensity.
In Figure 5, only the results of measurements are shown, at which the stress inten
sity is approximately equal to or greater than the optimum stress intensity. The measure
ment values for the different materials can be described by different approximation
curves, so that the specific energy required for a certain product quality depends more or
less strongly on the stress intensity. The greatest influence of the stress intensity on the
specific energy exists for the disagglomeration and the disintegration processes, where
the measurement values can be described in a first approximation by a straight line with
a slope of nearly 1. Therefore, above all, for a disagglomeration and a disintegration pro
cess, it is advisable that the stress intensity lie in the optimum range. For these two materi
als, the result of a single stress event is independent of the stress intensity as long as the
stress intensity is higher than the optimum stress intensity. At a constant product quality,
the stress number stays constant with increasing stress intensity, therefore, the specific
energy consumption increases proportionally to the stress intensity.
100 1,000 10 1 0.5
SE/SE
opt
[–]
10
100
1
0.8
E
m
/
E
m
,
m
i
n
[
–
]
Pigments [1] x
50
= 1 μm
Yeast Cells A = 60%
SiO
2
Aggregates x
50
= 2 μm
Waterbased Ink [17] Cl = 140%
Printing Ink T = 80%
Limestone x
50
= 2 μm
Al
2
O
3
x
50
= 2 μm
FIGURE 5 Specific energy related to the minimum specific energy as a function of the stress
energy related to the optimum stress energy (constant product fineness or constant
disintegration rate)
ENERGY CONSUMPTION, STRESS ENERGY, AND POWER DRAW OF STIRRED MEDIA MILLS 107
In the case of the synthetically produced SiO
2
aggregates from SiO
2
, the slope of the
straight line is slightly smaller than 1, so that the stress intensity already has a slight
effect on the result of a stress event. Therefore, with increasing stress intensity, the
aggregate is decomposed in smaller agglomerates or primary particles. The smallest
slopes can be found for grinding limestone and fused alumina. In the case of these two
materials, finer fragments are produced with increasing stress intensity. This effect is
somewhat more distinct for fused alumina than for limestone, but the slope of the
approximation curve is clearly greater than 0. For alumina, an increase in stress intensity
is more effective regarding an increase in product quality than for the other materials.
The relationships between the ratio of specific energy and minimum specific energy
and the ratio of stress intensity and optimum stress intensity can be derived from Equa
tions (2) and (3), if the product quality and, thus, the produced specific surface is set
constant. At a constant specific surface, the ratio EU/EU
max
corresponds to E
m,min
/E
m
, so
that the following relation can be given for SE > SE
opt
:
(EQ 6)
with a = 0 for disagglomeration/disintegration
0 < a < 1 for grinding crystalline materials
Therefore, the specific energy required for a certain product quality is a function of the
stress energy, SE. From Figure 5, the values of exponent “a” can be found for the seven
investigated materials. Exponent “a” determines the slope of the curves shown in Figures 1
and 2. The values are shown in Table 1. It can be seen that the value of exponent “a”
increases with the grinding resistance of the material. Summing up the results in Figure 5
and in Table 1 shows that, depending upon the feed material, the specific energy
required for a certain product quality relies more or less on the stress energy. The stron
gest influence of the stress energy on the specific energy exist for pure disagglomeration
and disintegration. In the case of disagglomeration and disintegration, attention should
be given to an optimum setting of the stress energy, because otherwise the specific
energy requirement becomes needlessly high and, thus, at constant power input, the
production capacity becomes needlessly low.
DETERMI NATI ON OF POWER DRAW
The power draw can be determined by simple empirical equations or by a relation between
dimensionless numbers. For stirred media mills, the relation between power number and
Reynolds number will be used to determine the power draw as a function of the most
important operating parameters: The exact description of the threedimensional fluid
flow field in the grinding chamber of a stirred media mill is currently impossible due to
the complexity of fluid–mechanical equations. Thus, the power draw is estimated by
suitable models developed in stirring technology. The basis of this model is the transfer
of power draw behavior of geometrically similar stirring systems (in this case, stirred
E
m
E
m min ,

SE
SE
opt

\ .
 
1–a
·
TABLE 1 Values of exponent “a”
Material Pigments Yeast Cells
Synthetic
SiO
2
Water
based Ink Printing Ink Limestone
Fused
Alumina
1–a [–] ~1 1 0.77 ~0.4 0.37 0.33 0.26
a [–] 0 0 0.23 0.60 0.63 0.67 0.74
108 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
10
0
10
1
10
0
10
–1
10
1
10
2
10
3
10
2
10
3
10
4
10
5
10
6
Re =
N
e
=
1
2
3
d/D = 0.8
L/D = 2.5
Z = 8
FIGURE 6 Relation between Newton number and Reynolds number without grinding media
10
0
10
–1
10
1
10
2
10
3
10
2
10
4
10
5
10
6
Re =
N
e
=
Grinding Media: Glass
d
GM
= 1 mm
ϕ
GM
= 0.0
ϕ
GM
= 0.35
ϕ
GM
= 0.65
ϕ
GM
= 0.90
FIGURE 7 Relation between power number and Reynolds number of stirrer for different filling
ratios of grinding media
ENERGY CONSUMPTION, STRESS ENERGY, AND POWER DRAW OF STIRRED MEDIA MILLS 109
media mills) by means of two characteristic numbers: the power number, sometimes
called the Newton number, Ne; and the Reynolds number, Re
S
. The power number and
Reynolds number are defined as follows:
Power number: (EQ 7)
Reynolds number: (EQ 8)
where P
GC
is the power consumed inside the grinding chamber; d is the diameter of stir
rer discs; n is the number of revolutions; µ
product
is the density of product slurry; and
q
product
is the kinematic viscosity of the product slurry.
In analogy to stirring technology, it is assumed that the power number only depends
on the Reynolds number. Tests with a Newtonian fluid and grinding media were carried
out in a stirred media mill with discstirrer geometry to obtain a theoretical relation
between the Newton number and the Reynolds number (Stehr 1982; Weit 1987). The
results are shown in Figure 6.
In the case of an operation with grinding media, the question arises as to whether
the grinding media belongs to the suspension or to the mill itself. If the grinding media
belongs to the suspension, for the physical characteristics of the density and the viscos
ity, the values of the grinding media–product suspension must be used. In this case, the
relationship between the power and Reynolds numbers shown in Figure 6 is also valid if
the mill is operated with grinding media. If the grinding media are considered to be part
of the mill (like baffles in a stirred vessel), the physical characteristics of the product sus
pension can be taken for the suspension density and the suspension viscosity. In this
case, for each filling ratio of the grinding media, for each grinding media density, and for
each grinding media size, a different relationship between the Newton and Reynolds
numbers exist.
Figure 7 shows the power number as function of the Reynolds number for a mill
with disc stirrer which was operated without grinding media (continuous line) and with
three different filling ratios of grinding media.
As described previously, for each filling ratio of the grinding media, a different rela
tionship between the power number and the Reynolds number exists. At a constant
power number, the Newton number is the greater because of the higher filling ratio of
the grinding media. In detail it can be seen that the curves for the operation with grind
ing media can be distinguished into five regions, each with a different slope instead of
into three regions as in the operation without grinding media (see Table 2).
If the equations for the power and the Reynolds numbers are put in the equations
given in Table 3 (right column), for each region of the Reynolds number, a different rela
tionship arises between the power draw and the parameter number of revolution or stirrer
Ne
P
GC
d
R
5
n
3
µ
product
 =
Re
S
nd
2
µ
product
q
product
 =
TABLE 2 Calculation of power number
Flow Region Reynolds Number Power Number
Laminar region Re
R
< 1.2 · 10
2
Ne
0
= K
laminar
· Re
R
–1
Lower transition region 1.2 · 10
2
< Re
R
< 8 · 10
3
Ne
0
= K
transition,A
· Re
R
–0.5
Upper transition region 8 · 10
3
< Re
R
< 3.5 · 10
4
Ne
0
= K
transition,B
· Re
R
–0.3
Lower turbulent region 3.5 · 10
4
< Re
R
< 2 · 10
5
Ne
0
= K
turbulent,A
· Re
R
–0.2
Upper turbulent region Re
R
> 2 · 10
5
Ne
0
= K
turbulent,B
110 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
tip speed, density of the product suspension, viscosity of the product suspension, and
stirrer diameter:
(EQ 9)
where
Y = –1 for Re
S
< 1,2 · 10
2
Y = –0.5 for 1.2 · 10
2
< Re
S
< 8 · 10
3
Y = –0.3 for 8 · 10
3
< Re
S
< 3.5 · 10
4
Y = –0.2 for 3.5 · 10
4
< Re
S
< 2 · 10
5
Y = 0 for Re
S
> 2 · 10
5
In the following paragraphs, the relationships between the power consumption and
the operating parameters’ filling ratio of the grinding media, ¢
GM
, stirrer tip speed, v
t
,
density of the product suspension, µ
P
, viscosity of the product suspension, q
Susp
, viscosity
of the product suspension, q
P
, and the stirrer diameter, d
s
, are discussed individually.
The symbol of the exponents comes from the following equation:
(EQ 10)
The exponent Y depends on the Reynolds number and results from the relation between
the power number and Reynolds number. The exponents X and Z can only be estimated
from experimental results.
The effect of the filling ratio of grinding media on the relation between power num
ber and Reynolds number is shown in Figure 7, in which the filling ratio was varied from
0 to 0.9. The density and viscosity of the pure Newtonian fluid were used to calculate the
power and Reynolds numbers. Thus, the analog to baffles in stirred vessels in the grind
ing media are considered to be part of the grinding chamber. If the grinding media are
considered to be part of the suspension, the exact influence of the filling ratio of the sus
pension viscosity must be known, which currently is not possible.
The power number, and thus the power draw, increases with increasing filling ratio
of grinding media for a constant Reynolds number. All curves look similar. The curves
approach a horizontal line in the turbulent region. This indicates that the power draw
does not change in the turbulent region. From the shape of the curve and the above
equations, it follows that depending on the number of revolutions and the product vis
cosity, the exponent X of the filling ratio ¢
GM
is in the range of 2.8 for lower Reynolds
numbers and 2.2 for higher Reynolds numbers. The values for the exponent are valid for
mills with a discstirrer geometry. Eventually, other mill geometries may have other val
ues for the exponents.
TABLE 3 Exponents to determine power draw as a function of important operating parameters
Influencing parameter ¢
GM
v
t
µ
Susp
q
Susp
d
S
d
GM
Exponent X Y 3+Y 1+Y –Y 2+Y Z
Low number of revolutions, high viscosity (Re
R
< 10
5
) 2.8 –0.5 2.5 0.5 0.5 1.5 s0
High number of revolutions, low viscosity (Re
R
> 10
5
) 2.2 0 3 1 0 2 >0
P
GC
d
S
5
n
3
µ
Susp

n d
S
2
µ
Susp
q
Susp

\ .

 
Y
·
P
GC
n
3+Y
d
S
5+2Y
µ
Susp
1+Y
q
Susp
Y –
v
t
3+Y
d
S
2+Y
µ
Susp
1+Y
q
Susp
Y –
· · ¬
P
GC
¢
GM
X
v
t
3+Y
µ
Susp
1+Y
q
Susp
Y –
d
S
2+Y
d
GM
Z
·
ENERGY CONSUMPTION, STRESS ENERGY, AND POWER DRAW OF STIRRED MEDIA MILLS 111
Practical experiences show that the influence of grinding media size on power draw
is not uniform. The power draw remains either constant, decreases, or increases when
the media size is changed (Weit 1987). For low Reynolds numbers (low turbulences,
high viscosity), the power draw is higher for smaller grinding media. An explanation for
this behavior can be different mechanisms of power transfer from the stirrer to the grind
ing media. For example, at low Reynolds numbers, the power draw is determined mainly
by the effect of the media size on the viscosity of the grinding media–product suspen
sion, which increases with smaller media sizes. At high Reynolds numbers, the hits
between the grinding media are more important for the power consumption of the mill.
Therefore, in principle, the exponent is less than 0 at low Reynolds numbers and greater
than 0 at high Reynolds numbers.
The dependency of the power draw P
GC
on the operating parameters can be summa
rized by Equation (10) and Table 3.
Exponent values for Reynolds numbers range between 8 · 10
3
and 2 · 10
5
, and thus
values for medium stirrer tip speeds as well as for high tip speeds and simultaneously
high viscosities are between the values given in Table 3. In the case of low stirrer tip
speeds and very high viscosities, the exponent values can also lie outside the range given
in the table (i.e., –1 < Y < –0.5).
The actual value of the exponent Z for the grinding media size d
GM
must be deter
mined experimentally. The tendency is for the exponent to be negative at small Reynolds
numbers (especially in the laminar region) and positive at higher Reynolds numbers.
The simplest way is to carry out two or three grinding tests with different grinding media
sizes. All other parameters have to be held constant. Grinding tests with different grind
ing media sizes are needed for the determination of the optimum grinding media size
and, thus, should always be carried out.
DETERMI NATI ON OF PRODUCTI ON RATE
As described in the introduction to this paper, stirred media mills are used most econom
ically if the required product quality is produced at the maximum production rate and
lowest possible operating costs. The relation between the production capacity and pro
duction rate, respectively, , is given by the following relation (see Equation [1]):
(EQ 11)
The production capacity is proportional to the power consumed inside the grinding
chamber of the stirred media mill and inversely proportional to the specific energy that is
needed to produce the required product quality. The relation between the power draw
and important operating parameters is given by Equation (10). The dependency of the
specific energy on the most important operating parameters can be achieved by rear
ranging Equation (6):
(EQ 12)
The quotient E
m,min
/SE
opt
1–a
is only constant and, thus, is independent of all operat
ing parameters if the energy transfer coefficient v
E
is constant. If the energy transfer
coefficient v
E
changes, the minimum specific energy E
m,min
that is transferred into the
grinding chamber also changes. Therefore, it is advantageous to use the ratio E
m,P,min
/v
E
instead of the specific energy E
m,min
that is transferred into the grinding chamber. The
m
·
P
m
·
P
m
P Ch ,
t
Ch

P
GC
E
m
 = =
E
m
E
m min ,
SE
opt
1 a –
 SE
1 a –
=
112 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
minimum specific energy E
m,P,min
is the minimum specific energy that is actually trans
ferred to the product particles and, thus, can be considered to be constant even at differ
ent operating conditions. By the energy transfer factor, v
E
, the effect of operating
parameters, such as the filling ratio of the grinding media on the production capacity,
can be considered: The energy transfer coefficient is proportional to the filling ratio of
the grinding media (Kwade 2004). Therefore, the relation for the specific energy trans
ferred to the grinding chamber results if the stress energy of the grinding media, SE
GM
(see Equation [2]), is used as a measure for the stress energy:
(EQ 13)
The dependency of the production capacity on the most important operating param
eters results if Equation (13) and Equation (10) are inserted into Equation (1):
(EQ 14)
Equation (14) is the basic equation for determining the effect of the most important
operating parameters on the grinding result. How much the influence of the operating
parameters on the production capacity is affected by the product and formulation of the
grinding or dispersion process is shown by the following two examples.
Example 1: Dispersion of an agglomerate at a high viscosity and relatively low stirrer
tip speed in a mill with discstirrer geometry. In the case of a high viscosity and relatively
low stirrer speed, it can be concluded that the mill is operated in the lower transition
region. Moreover, it is assumed that the grinding media size has no effect on the power
draw in this flow regime. The exponent “a” is 0 in case of disagglomeration. Therefore,
the variables in Equation (14) have the values shown in Table 4.
(EQ 15)
Therefore, in the case of dispersing an agglomerate at high viscosity and relatively low
stirrer tip speed, in addition to a change in the filling ratio of the grinding media, a
change in the grinding media size has the greatest effect on the production capacity.
Thus, the production capacity can be increased most effectively by changing the filling
ratio and the size of the grinding media. Therefore, the stirrer tip speed has only a minor
effect on the production capacity. If the viscosity would be very high and, thus, the mill
would be operated in the laminar flow regime, there would be nearly no effect of the stirrer
tip speed on the production capacity.
Example 2: Real grinding of a hard ceramic raw material at low viscosity and relatively
high stirrer tip speed in a mill with discstirrer geometry. In the case of a low viscosity and
relatively high stirrer speed, it can be concluded that the mill is operated in the turbulent
flow regime. Moreover, it is assumed that the power draw is proportional to the grinding
media size with a power of 0.5, so that the exponent Z is equal to 0.5. The exponent “a”
is assumed to be 0.75, which is close to the value of the 0.74 found for fused alumina.
Therefore, the variables in Equation (14) have the values shown in Table 5.
E
m
E
m P min , ,
SE
opt
1 a –

SE
1 a –
v
E
 C
SE
1 a –
v
E

d
GM
3
v
t
2
µ
GM
( )
1 a –
¢
GM
 · = =
m
·
P
P P
0
–
E
m

¢
GM
X
v
t
3+Y
µ
Susp
1+Y
q
Susp
Y –
d
S
2+Y
d
GM
Z
d
GM
3
v
t
2
µ
GM
( )
1 a –
 · =
m
·
P
¢
GM
2.8+1
v
t
2.5
µ
Susp
0.5
q
Susp
0.5
d
S
1.5
d
GM
0
d
GM
3
v
t
2
µ
GM
( )
1–0
 · ¢
GM
3.8
v
t
0.5
µ
Susp
0.5
µ
GM
 q
Susp
0.5
d
S
1.5
d
GM
3 –
=
ENERGY CONSUMPTION, STRESS ENERGY, AND POWER DRAW OF STIRRED MEDIA MILLS 113
(EQ 16)
Therefore, in the case of grinding a raw ceramic material at low viscosity and relatively
high stirrer tip speed, in addition to a change in the filling ratio of the grinding media, a
change in the stirrer tip speed has the greatest effect on the production capacity. Thus,
the production capacity can be increased most effectively by changing the filling ratio of
the grinding media and the stirrer tip speed. Therefore, the grinding media size has only
a minor effect on the production capacity. If the resistance against grinding would be
even higher (i.e., exponent “a” would be in the range of 0.8), there would be nearly no
effect of grinding media size on the production capacity.
The two examples show how different the effects of the various operating parame
ters on the production capacity and on the power draw can be. This is particularly valid
for the effect of the stirrer tip speed and the grinding media size. How the different oper
ating parameters affect the production capacity is always a function of the product and
formulation of the grinding or dispersion process under consideration. Therefore, it is
not surprising that the influences of the parameters’ stirrer tip speed and grinding media
size on the grinding time and mean residence time can be described very differently
among the various literature. The effect of the filling ratio of the grinding media is
always the only independent factor on the product and formulation of the grinding or
dispersion process.
CONCLUSI ONS
The effect of the important operating parameters on the production rate of stirred media
mills can be determined by using models for stressing the particles and for the power
draw of the mill. For different products and formulations, the relation between the pro
duction rate and the different operating parameters show how sensitive the operating
parameters are regarding the production rate: Two examples of different products and
formulations reveal that the different operating parameters can have totally different
effects: In the case of dispersing agglomerates at high viscosities and low tip speeds, in
addition to the filling ratio of the grinding media, the grinding media size has a great
effect of the production rate, whereas the stirrer tip speed has only a small effect. There
fore, in the case of real grinding of a hard ceramic raw material, in addition to the filling
ratio of the grinding media, the stirrer tip speed is very important, whereas the grinding
media size has only a minor effect.
TABLE 4 Values of the exponents for disagglomeration at high viscosities and low stirrer speed
Variable/exponent X Y Z a
Operating parameters affected ¢
GM
v
t
, µ
Susp
, q
Susp
, d
S
d
GM
d
GM
, v
t
, µ
GM
Value 2.8 –0.5 0 0
TABLE 5 Values of the exponents for disagglomeration at high viscosities and low stirrer speed
Variable/exponent X Y Z a
Operating parameters affected ¢
GM
v
t
, µ
Susp
, q
Susp
, d
S
d
GM
d
GM
, v
t
, µ
GM
Value 2.2 0 0.5 0.75
m
·
P
¢
GM
2.2+1
v
t
3
µ
Susp
1
q
Susp
0
d
S
2
d
GM
0.5
d
GM
3
v
t
2
µ
GM
( )
1–0.75
 · ¢
GM
3.2
v
t
2.5
µ
Susp
µ
GM
0.25
 d
S
2
d
GM
0.25 –
=
114 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
Moreover, using the relations for the determination of the specific energy, the power
draw and the production rate of the mill performance for operating conditions that were
not tested before can be determined as long as the stress energy is greater than the opti
mum stress energy. By that, the production rate can be optimized so that the capacity is
as high as possible.
REFERENCES
Becker, M., A. Kwade, and J. Schwedes. 2001. Stress intensity in stirred media mills and
its effect on specific energy requirement. International Journal of Mineral Processing
61:189–208.
Bunge. 1992. Mechanischer Zellaufschluß in Rührwerkskugelmühlen. Ph.D. dissertation.
Braunschweig, Germany: Technical University of Braunschweig.
Kwade, A. 1996. Autogenzerkleinerung von Kalkstein in Rührwerkmühlen. Ph.D. disser
tation. Braunschweig, Germany: Technical University of Braunschweig.
———. 1998. Wet comminution in stirred media mills—research and its practical applica
tion. Pages 23–32 in Preprints 9th European Symposium on Comminution, Albi,
France. Frankfurt, Germany: European Federation of Chemical Engineering.
———. 1999. Wet comminution in stirred media mills—research and its practical applica
tion. Powder Technology 105:14–20.
———. 2002. Mill selection and process optimization using a physical grinding model. In
Preprints 10th European Symposium on Comminution, Heidelberg, Germany. Frank
furt, Germany: European Federation of Chemical Engineering.
———. 2004. Mill selection and process optimization using a physical grinding model.
International Journal of Mineral Processing 74S:93–101.
Kwade, A., L. Blecher, and J. Schwedes. 1996. Motion and stress intensity of grinding
beads in a stirred media mill. Part II: Stress intensity and its effect on comminution.
Powder Technology 86(1):69–76.
Kwade, A., and J. Schwedes. 1997. Wet comminution in stirred media mills. KONA
15:91–101.
Stadler, N., R. Polke, J. Schwedes, and F. Vock. 1990. Naßmahlung in Rührwerksmühlen.
ChemieIngenieurTechnik 62(11):907–915.
Stehr, N. 1982. Zerkleinerung und Materialtransport in einer Rührwerkskugelmühle,
Ph.D., dissertation. Braunschweig, Germany: Technical University of Braunschweig.
Stender, H.H., A. Kwade, and J. Schwedes. 2002. Stress energy distribution in different
stirred media mill geometries. In Preprints 10th European Symposium on Comminu
tion, Heidelberg, Germany. Frankfurt, Germany: European Federation of Chemical
Engineering.
Vock, F. 1997. Lackherstellung. Termen, Switzerland: Verlag CC Press.
Weit, H. 1987. Betriebsverhalten und Maßstabsvergrößerung von Rührwerkskugelmühlen,
Ph.D., dissertation. Braunschweig, Germany: Technical University of Braunschweig.
115
AG/SAG Mill Circuit Grinding Energy
Requirement—How to Predict It from
SmallDiameter Drill Core Samples
Using the SMC Test
Stephen Morrell
*
ABSTRACT
The SMC (semiautogenous mill comminution) test has been developed to provide a rock
breakage description that can be used to predict autogenous grinding (AG) and semiautoge
nous grinding (SAG) mill performance. The test has been specifically designed to be usable
in situations where only limited quantities of rock samples are available (e.g., smalldiameter
core). The test generates a DropWeight Index (DW
i
) that can be used to estimate the
throughput of AG and SAG circuits through a combination of powerbased and modelbased
approaches. The modelbased approach makes use of the direct relationship between the
DW
i
and the JKTech dropweight test rockbreakage parameters A and b. The powerbased
route uses correlations that have been developed between the DW
i
and the specific energies
of a very wide range of operating AG and SAG circuits. Its usefulness is also shown to extend
to rock mass characterisation in mining applications, as it also is correlated with the point
load index/UCS (unconfined compressive strength). It is therefore ideally suited for mineto
mill studies where it can be simultaneously used to predict comminution circuit performance
and to augment input to blastfragmentation models. This makes it a valuable tool for orebody
profiling in greenfield, brownfield, and established operations. Recent investigations have
shown that the DW
i
is also strongly related to highpressure grinding roll (HPGR) performance.
The ability of the test and associated equations to predict AG/SAG circuitspecific
energy is demonstrated using a wide range of industrial data. This approach is compared to
more traditional ones such as that of Bond, which is also reviewed in the context of its abil
ity to predict AG/SAG circuitspecific energy and energy utilisation efficiency.
I NTRODUCTI ON
As little as 10 years ago, a “conventional” comminution circuit in the minds of many metal
lurgists would have conjured up pictures of crushing–ball mill or rod mill–ball mill circuits.
Today, it is not common to find such circuits in operation, let alone being built. AG and
SAG mills now dominate circuit design in gold and base metals applications and can right
fully lay claim to being conventional, leaving technologies such as highpressure grinding
rolls (HPRGs) the title of “new.” Regardless of how one categorises these technologies,
* SMCC Pty. Ltd., Queensland, Australia
116 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
today a much wider spectrum of proven equipment is available to the circuit designer
than, say, 25 to 30 years ago. Although such choice may be seen as an improvement, it
provides a particular challenge in terms of assessing which circuit is the most energy effi
cient and how in the first place the ore type should be meaningfully described in terms of
its breakage characteristics. For ball milling, Bond’s ball work indices and equations have
become the standard for describing grindability and efficiency. However, to date, there is
no universal equivalent for autogenous/semiautogenous milling. The Julius Kruttschnitt
Mineral Research Centre (JKMRC) dropweight test parameters, A and b, have become
popular when characterising rock for autogenous/semiautogenous milling, but they are
specifically for use in autogenous/semiautogenous modeling and cannot be used for
powerbased calculations. In addition, the JK (JKTech) parameters are obtained from
breaking relatively large quantities of material and hence cannot be obtained from small
samples such as those provided by drill core. The recently developed DropWeight Index
(DW
i
) may provide the solution. In this paper, the DW
i
will be reviewed in terms of how
it relates to AG/SAG mill specific energy as well as to traditional strength measurements
and the JK A,b parameters. The test used to determine the DW
i
(SMC test) is also
described. In the course of this review, traditional ways of determining AG/SAG specific
energy will be analysed.
TRADI TI ONAL APPROACH TO CI RCUI T SELECTI ON
For AG/SAG mill circuit selection, piloting is still regarded as being the best option for
estimating what the performance of the fullscale circuit will be. Tests are normally con
ducted under a range of conditions, the choice of circuit then being made on the basis of
a number of criteria, which normally include factors such as minimum specific energy
and/or maximum power utilisation efficiency. The specific energy is easy to determine as
it is unambiguously defined as the power draw divided by the throughput, with different
operating conditions (e.g., ball load and speed) and circuit configurations (e.g., open
circuit, closed circuit, with or without pebble crusher, etc.) resulting in different specific
energies. However, product grind size from each also varies, leaving the designer with
the problem of determining which is the most energy efficient. Historically, this has often
been done by applying Bond’s equation to determine the operating work index, which is
considered by some to be indicative of the efficiency of the circuit. This has been recently
challenged on the basis that the Bond equation is fundamentally flawed and hence any
conclusions regarding energy efficiency based on its use are likely to be erroneous (Mor
rell 2004a). A further problem is that, whereas the piloting may provide sufficient infor
mation to select the best circuit, it may only apply to the ore that was tested during the
programme. Many deposits have highly variable comminution characteristics leaving
unanswered the very important question: “Will the chosen circuit work as well on other
ore types?” When pilot testing is not carried out at all, this problem is exacerbated as the
circuit design has to rely entirely on laboratoryscale ore characterisation data.
ANALYSI S OF AG/SAG CI RCUI T ENERGY EFFI CI ENCY USI NG BOND’S
EQUATI ON
As mentioned in the previous section, a common choice is to use the Bond equation to
calculate the Bond operating work indices to compare the efficiencies of different cir
cuits. This equation is written as
(EQ 1) OW
i
W
10
1
P

1
F
 –
© ¹
¨ ¸
§ ·
 =
AG/SAG MILL CIRCUIT GRINDING ENERGY REQUIREMENT 117
where
W = specific energy
OW
i
= operating work index
P = 80% passing size for the product
F = 80% passing size for the feed
By way of example to illustrate its use, data from a pilot programme are given in Fig
ure 1 and show a systematic trend in the specific energy as ball charge is varied. It is
pointed out that the AG mill runs were conducted with a pebble crusher in circuit whilst
the SAG mill runs were not. The data indicate that the worst condition (highest specific
energy) is when about 4% of steel balls are used. When the Bond operating work indices
are calculated, a very different picture is obtained, as shown in Figure 2. From these
data, the 4% case is indicated to give the best power utilisation efficiency (lowest OW
i
).
Closer analysis of the data shows that there is also a similar relationship between the
ball charge and the P80 (Figure 3), indicating that the underlying relationship is in fact
one that links operating work index to P80. This is confirmed in Figure 4 where a strong
correlation between the Bond operating work index and the product P80 is seen. This
trend, which is found in many data sets, shows a decreasing operating work index as the
grind becomes finer and is counterintuitive. The expected result would be one in which
either the operating work index remained constant (indicating constant energy effi
ciency and constant material properties) or it increased as product size decreased (i.e.,
the rock became harder as the product size became smaller and/or the mill became less
efficient at producing a finer grind). This result points to a potential error in the Bond
equation and puts into question the conclusion regarding maximum power efficiency
with 4% balls.
Researchers such as Hukki (1962) have challenged the validity of Bond’s equation,
at least outside the range of feed and product sizes treated in ball mills. Recently, an
alternative equation to Bond’s has been proposed (Morrell 2004a). This has the form:
(EQ 2)
where
W = specific energy (kWh/t)
M
i
= index related to the breakage property of an ore (kWh/t)
K = constant chosen to balance the units of the equation
x
2
= 80% passing size for the product
x
1
= 80% passing size for the feed
(EQ 3)
where
a,b = constants
x = 80% passing size
The parameters a and b in Equation (3) have been estimated from analysing a wide range
of size reduction data from industrial grinding mills. Equation (2) can therefore be used
provided M
i
is known. Alternatively, for analysing circuit performance, the equation can be
rearranged such that an operating value for M
i
can be calculated using plant data. This is
the equivalent of the Bond operating work index. When this is done using the data from
Figure 4, the results given in Figure 5 are obtained and show that the operating work
index is in fact largely constant with respect to product size, and hence there is no indi
cated difference in power utilisation efficiency between the different operating conditions.
W M
i
K x
2
f x
2
x
1
f x
1
– =
f x a x
b
+ – =
118 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
EFFI CI ENCY OF AG/SAG AND BALL MI LL CI RCUI TS
Given that the use of Equation (2) indicates that there is little or no difference between
the power utilisation efficiencies of the different modes of AG/SAG mill operation, the
question arises as to whether the equation indicates differences in efficiency between
AG/SAG and ball milling in general. Data from 18 different operations were analysed to
answer this question. The data comprised throughput and power draws as well as feed,
transfer, and ball mill cyclone overflow sizings from each circuit. Initially, Bond operat
ing work indices were calculated for each circuit. These are plotted for each circuit and
shown in Figure 6. The ball mill values largely followed the Bond laboratory work index
results, which were also obtained for each ore type. The AG/SAG operating work indices
0 2 4 6 8 10 12 14
Ball Charge, %
0
2
4
6
8
10
12
14
16
18
20
S
p
e
c
i
f
i
c
E
n
e
r
g
y
,
k
W
h
/
t
With Pebble Crusher
Without Pebble Crusher
FIGURE 1 Trends in pilot SAG mill specific energy
0 2 4 6 8 10 12 14
Ball Charge, %
0
1
2
3
4
5
6
7
8
B
o
n
d
O
W
i
,
k
W
h
/
t
FIGURE 2 Trends in Bond operating work index
AG/SAG MILL CIRCUIT GRINDING ENERGY REQUIREMENT 119
show their usual elevated levels compared to those from the ball mill circuit. This has
often resulted in conclusions concerning lower energy efficiencies of AG/SAG mill cir
cuits compared to ball mills. The correlation between the AG/SAG and ball mill circuit
data is also very poor. Use of Equation (2) shows a very different picture, the results
being illustrated in Figure 7. This shows that on average the “M” operating work indices
of AG/SAG and ball mill circuits are very similar, and hence energy utilisation efficien
cies are similar. Also, the AG/SAG and ball mill circuit operating work indices are highly
correlated.
The conclusion that AG/SAG circuits have, on average, a similar power utilisation
efficiency to ball mill circuits may run counter to much “conventional wisdom.” How
ever, controlled experiments in which very different crushing and grinding circuits have
0 2 4 6 8 10 12 14
Ball Charge, %
0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
P
r
o
d
u
c
t
P
8
0
,
m
m
FIGURE 3 Relationship between ball charge and product P80
0 1 2 3 4 5 6 7
Product P80, mm
0
10
20
30
40
50
60
70
80
B
o
n
d
O
W
i
,
k
W
h
/
t
FIGURE 4 Trend in Bond operating work index with product P80
120 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
been run using identical ores have shown little difference in the energy required to reach
a target grind from a given feed size (Larsen, Cooper, and Trusiak 2001; Morrell,
Johnson, and Revy 1991). The analyses provided in this paper support these results and
lead to the assertion that in many cases, regardless of the processing route, the energy
required to grind an ore from a specific feed size to a specific product size will be similar,
at least to within r5%. It can be concluded from this argument that, at least from an
energy utilisation efficiency viewpoint, all circuits work equally well regardless of ore
type when they are fully optimised. Of course, that is not to say that from a capital cost,
operating cost, and operability standpoint, all circuits are the same—far from it. Ulti
mately, circuit choice should be made on financial grounds. However, differences in
overall power efficiency should not necessarily play a prominent role in decision making
as, when analysed correctly, data show that few differences exist between circuit power
0 1 2 3 4 5 6 7
Product P80, mm
0
10
20
30
40
50
60
70
M
o
r
r
e
l
l
O
W
i
,
k
W
h
/
t
FIGURE 5 Trend in SMCC operating work index with product P80
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Data Set
0
10
20
30
40
50
60
B
o
n
d
O
W
i
,
k
W
h
/
t
AG/SAG
Ball
FIGURE 6 Bond operating work indices for AG/SAG and ball mill circuits
AG/SAG MILL CIRCUIT GRINDING ENERGY REQUIREMENT 121
efficiencies. These arguments relate to conventional crushing and tumbling mill circuits.
The use of HPGRs, however, appears to provide a genuine reduction in power require
ments (Parker et al. 2001).
NEW APPROACH TO PREDI CTI NG AG/SAG SPECI FI C ENERGY
The previous sections have indicated that AG/SAG mill circuit power utilisation efficien
cies are largely similar, regardless of the circuit configuration and operating conditions
such as ball charge, speed, and so forth. If this is the case, then it should be possible to
predict the AG/SAG specific energy of all types of circuits without making any assump
tions and/or corrections concerning energy utilisation efficiency.
The choice of an appropriate measure of the ore breakage characteristics and an
associated technique for predicting the specific energy is obviously very important for
this approach to work. A potential appropriate measure of an ore’s breakage characteris
tics is the socalled DW
i
, which is a parameter derived from the SMC test (Morrell
2004b). The difficulty in determining whether such a relationship exists is that the spe
cific energy of AG/SAG mills does not just depend on ore competence but also factors
such as feed size, ball load, aspect ratio, whether the mill has a pebble crusher or not,
and whether the mill is in closed circuit or not. An equation was therefore developed for
use with the DW
i
for predicting specific energy and has the following form:
(EQ 4)
where
S = specific energy (kWh/t)
F
80
= 80% passing size of the feed
DW
i
= strength index
J = volume of balls (%)
I = mill speed (% of critical)
f(A
r
) = function of mill aspect ratio (length/diameter)
a,b,c,d,e = constants
K = function whose value is dependent upon whether a pebble crusher is in circuit
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Data Set
0
10
20
30
40
50
60
M
o
r
r
e
l
l
O
W
i
,
k
W
h
/
t
AG/SAG
Ball
FIGURE 7 “M” operating work indices for AG/SAG and ball mill circuits
S K F
80
a
DW
i
b
1 c 1 e
dJ –
– +
1 –
I
e
f A
r
=
122 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
A companion equation was also developed for predicting transfer size, as follows,
and works on the basis that the more energy that is input to the mill in relation to the
hardness of the ore, the finer will be the transfer size.
(EQ 5)
where
S = specific energy (kWh/t)
b,f,g = constants
A combination of the two equations gives the AG/SAG specific energy as well as the
transfer size, such that the “M” operating work index from Equation (2) remains fairly
static regardless of autogenous/semiautogenous operating conditions. To develop the
approach, 46 data sets from 30 different operations were used. AG/SAG specific energy
and ball mill specific energy were predicted using DW
i
and Bond ball laboratory work
indices. Ore types represented in the database were from Al, Au, Pt, Cu, Ni, and Pb/Zn
operations. The range of conditions covered is given in Table 1. The results are shown in
Figure 8, indicating a reasonable correlation between observed and predicted specific
energies, the standard deviation of the relative error (precision) being 8.5%.
SMC TEST DESCRI PTI ON
The SMC test, from which the DW
i
is derived, was originally developed to make use of
relatively small samples, both in terms of quantity and particle size, and to be versatile so
as to make as much use as possible of whatever sample(s) is available for testing. As a
result, the test is able to accommodate a wide range of particle sizes, either in core or
crushed form. The test is applied to particles of a particular size, the size being chosen
depending on the type and quantity of sample available. The particle sizes that can be
used in the SMC test are –45+37.5, –31.5+26.5, –22.4+19, and –16+13.2 mm. Sample
sources can be from core sizes as large as PQ (85 mm) and as small as AQ (27 mm). Mostly,
either the 31.5+26.5 mm or –22.4+19 mm sizes are chosen because these are easily
extractable from HQ and NQ cores, respectively, and these tend to be the most popular
choice of core sizes. When sample availability is very limited, quartered (slivered) core
samples are cut using a diamond saw (Figure 9). This results in sample mass require
ments as low as 2–2.5 kg in total. However, where core is available in sufficient quantity
(10–15 kg), it can be crushed instead and the appropriate size fraction extracted.
Once the core has been cut or crushed/sized into the chosen particle size range, 100
specimens are chosen and divided into five equal lots. Each lot is then broken in an
impact device using a range of closely controlled energies. A suitable impact device is
JKMRC’s dropweight tester (NapierMunn et al. 1996). After breakage, the products are
collected and sized on a sieve whose aperture is related to the original particle size. The
percentage of undersize from sieving the broken products is plotted against the input
energy. A typical plot from a test is given in Figure 10 and shows the expected trend of an
increasing amount of undersize as the input energy is increased. The slope of this plot is
related to the strength of the rock, a slope with a larger gradient being indicative of a
weaker rock. The gradient of the slope is used to generate a socalled dropweight index
(DW
i
). The DW
i
has the units of kWh/m
3
, which in turn has the same dimensions as
strength, and hence it is not surprising that the DW
i
is correlated with direct strength
measurements such as the point load index (discussed later in this paper).
The high degree of control imposed on both the size of particles and the energies
used to break them means that the SMC test is very precise and is largely free of the
T
80
f
g S
DW
i
b
 – =
AG/SAG MILL CIRCUIT GRINDING ENERGY REQUIREMENT 123
repeatability problems that plague tumblingmill rock characterisation tests (Angove and
Dunne 1997; Kaya 2001). Such tests usually suffer from variations in feed size, which is
often not closely controlled, as well as energy input per mill revolution, which is often
assumed to be constant but in practice can be highly variable (Levin 1989).
The standard JK dropweight test normally needs about 75 kg of raw material, and
hence its use is normally precluded for small drillcore samples. However, the DW
i
is
highly correlated with the A and b parameters and therefore can be used to estimate
their values with a high degree of accuracy. Figure 11 illustrates this using data from 40
different ore types. The scatter apparent in the figure has an associated standard devia
tion of 6.5%. This is related to the differences in the variation of strength with particle
size that different rocks exhibit. This scatter can be reduced by carrying out full drop
weight tests on selected samples from the ore body in question to better define the size
bysize relationship and hence refine the DW
i
– A,b correlation. Such dropweight tests
TABLE 1 Range of Variables in the Database
Variable Maximum Minimum
JK – A 81.3 48
JK – b 2.56 0.25
Specific gravity 4.63 2.5
DW
i
14.2 1.8
Bond ball working index, kWh/t 26 9.4
F80, μm 176,000 19,400
P80, μm 600 54
Diameter, m 12 3.94
Length, m 8.3 1.65
Ball load, % 25 0
Speed, % 86 68
Aspect ratio, L/D 1.5 0.3
SAG, kWh/t 29.2 2.4
Predicted kWh/t
0
5
10
15
20
25
30
35
O
b
s
e
r
v
e
d
k
W
0 5 10 15 20 25 30 35
AG/SAG Circuit Only
Total Circuit
FIGURE 8 Predicted AG/SAG and total circuit specific energy
124 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
are usually referred to as SMC test “calibrations,” though they can be dispensed with if
the 6.5% precision of the database correlation is deemed to be acceptable.
USE OF DW
i
I N MODELI NG
AG and SAG Mill Circuits
The use of modeling and simulation has become routine in the design and optimisation
of AG and SAG mill circuits. One of the most widely used models for this purpose is the
socalled “variable rates” model (Morrell and Morrison 1996). A more uptodate version
has also been developed with enhanced predictive capabilities (Morrell 2004c). This
uses a twoparameter description of rock breakage that is developed from data obtained
FIGURE 9 Sample pieces cut from 50mm quartered core
0
10
20
30
40
50
60
%
U
n
d
e
r
s
i
z
e
0 50 100 150 200 250
Energy, Joules
FIGURE 10 Typical raw results from an SMC test
AG/SAG MILL CIRCUIT GRINDING ENERGY REQUIREMENT 125
from a dropweight test (NapierMunn et al. 1996). The two parameters (A and b) are
ore specific and are generated as part of the SMC test via their correlation with the DW
i
.
They relate the t
10
(a size distribution index) to the applied specific energy (Ecs). The
equation used for describing the relationship between the t
10
and Ecs is
(EQ 6)
The specific comminution energy (Ecs) has the units kWh/t and is the energy applied
during impact breakage. As the impact energy is varied, so is the t
10
. Higher impact ener
gies produce higher values of t
10
, which are reflected in products with finer size distribu
tions. The A and b parameters, in conjunction with Equation (6), are used in AG/SAG
mill modeling for predicting how rock breaks inside the mill. From this description, the
model can predict what the throughput, power draw, and product size distribution will be.
Apart from being able to predict throughput and power draw of AG/SAG mills, mod
eling and simulation also enables a detailed flowsheet to be built up of the comminution
circuit response to changes in ore type. It also enables optimisation strategies to be
developed to overcome any deleterious changes in circuit performance that are pre
dicted. This is particularly useful during the design stage because the chosen circuit can
be tested under a range of conditions to see whether the circuit will meet its production
targets. Strategies can then be developed to overcome any potential problems. These can
include both changes to how mills are operated (e.g., ball load, speed, etc.) but also
changes to feedsize distribution through modification to blasting practices and primary
crusher operation—the socalled minetomill approach.
MinetoMill Applications
The feed size to AG and SAG mill circuits has been demonstrated to have a significant
impact on throughput. Modifying blast design and primary crusher operation can signif
icantly influence AG/SAG mill feed size, hence giving a potentially costeffective way to
increase comminution circuit throughput. Trialanderror experimentation in this field,
however, can be very costly, and thus it is usual to rely on blastfragmentation modeling
and grindingcircuit simulations to determine what the optimum blast design should be.
0
20
40
60
80
100
120
140
160
P
r
e
d
i
c
t
e
d
A
×
b
0 20 40 60 80 100 120 140 160
Observed A × b
FIGURE 11 Observed versus predicted values of A × b using the DW
i
t
10
A 1 e
b·Ecs –
– =
126 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
This will vary with ore type, and hence it is important not only to have appropriate blast
models but also rockbreakage descriptions. Blasting models require information on rock
mass competence such as provided by the point load strength (Scott, Morrell, and Clark
2002). The DW
i
is correlated with the point load strength (Figure 12) and hence also can
be used in blastfragmentation modeling where direct measurements of point load
strength are not available or are very limited. Conversely, where a significant database of
point load tests are available, these can be used to augment the comminution description
of the ore body by using the correlation from Figure 12 in reverse.
HighPressure Grinding Rolls
Although HPGR technology has become commonplace in the cement and diamond min
ing industries and of late has been making significant inroads in the processing of iron
ore, it has yet to make a major impact in the gold, platinum, and base metals sectors.
However, interest in the technology is now such that the general expectations are that
rapidly increasing numbers of HPGR machines are likely to find their way into these sec
tors. Due to the operation of HPGRs, the more established techniques for breakage char
acterisation, design, and scaleup that have been developed on the basis of tumbling
mills are not applicable. Simulation has helped in this regard, and JKSimMet software
contains a model that has been shown to have good scaleup capabilities (Morrell, Shi,
and Tondo 1997; Daniel and Morrell 2004). This model needs HPGR data to calibrate it,
and although it has been shown that laboratoryscale HPGR results are suitable, separate
tests need to be conducted on every different ore type, as the size reduction and through
put parameters of the model are machine and ore dependent. Ore characterisation there
fore remains a problem, though it is being currently researched in the AMIRA P9 project.
The DW
i
may provide at least part of the answer, as it has been found that it is corre
lated with the operating work index of HPGRs as Figure 13 indicates. The data in this fig
ure have been obtained from 13 different ore types. It is valid for machines operating
with a working pressure in the range of 2.5 to 3 N/mm
2
.
The correlation in Figure 13 is not intended for design purposes but can be used in
conjunction with pilot and/or laboratoryscale HPGR test results to predict the specific
energy requirement of rock samples that cannot be tested in an HPGR. Its value for orebody
0
2
4
6
8
10
12
14
0 2 4 6 8 10 12
DW
i
P
o
i
n
t
L
o
a
d
S
t
r
e
n
g
t
h
,
M
P
a
FIGURE 12 Correlation between point load strength and the DW
i
for a copper ore
AG/SAG MILL CIRCUIT GRINDING ENERGY REQUIREMENT 127
profiling is obvious. Also, the fact that the DW
i
is both applicable to AG/SAG and HPGR
circuits makes the SMC test particularly attractive in greenfield design projects, as its use
for characterising drill core does not compromise the ability of the designer when subse
quently evaluating the response of AG/SAG and HPGR circuits to changes in ore type.
CONCLUSI ONS
The SMC rockbreakage characterisation test has been developed to make use of very
small quantities of sample, such as quartered drill core. The test generates a strength
index (DW
i
) which, via modeling and/or powerbased techniques, can be used to predict
the specific energy of AG and SAG mills as well as HPGR circuits. Its applicability for
modeling stems from its correlation with the JK rockbreakage parameters (A and b). For
powerbased calculations, an equation has been developed that relates it and operating
variables such as feed size, ball load, and speed to AG/SAG mill specific energy with a
precision of 8.5% (1 standard deviation).
The usefulness of the DW
i
also extends to rock mass characterisation in mining
applications, as it is correlated with the point load index/UCS. It is therefore ideally
suited for minetomill studies as it can be simultaneously used as an input to both commi
nution circuit and blastfragmentation models where independent point load/UCS mea
surements are not available.
BI BLI OGRAPHY
Angove, J.E., and Dunne, R.C. 1997. A review of standard physical ore property
determinations. World Gold Conference, Singapore, September 1–3.
Daniel, M.J., and Morrell, S. 2004. HPGR model verification and scaleup. Minerals
Engineering 17(11–12):1149–1161.
Hart, S., Valery, W., Clements, B., Reed, M., Song, M., and Dunne, R. 2001. Optimisation
of the Cadia Hill SAG mill circuit. Pages 11–30 in Proceedings of the International
Conference on Autogenous and SemiAutogenous Grinding Technology. Volume 1.
Vancouver, BC: University of British Columbia.
0
5
10
15
20
25
0 2 4 6 8 10
DW
i
H
P
G
R
O
W
i
,
k
W
h
/
t
FIGURE 13 DW
i
versus HPGR operating work index
128 ADVANCES IN COMMINUTION ADVANCED COMMINUTION TECHNOLOGIES
Hukki, R.T. 1962. Proposal for a solomnic settlement between the theories of von
Rittinger, Kick and Bond. AIME Transactions 223:403–408.
Kaya, E. 2001. Evaluation of bond grindability testing. Pages 339–347 in Proceedings of
the International Conference on Autogenous and SemiAutogenous Grinding
Technology. Volume 1. Vancouver, BC: University of British Columbia.
Larsen, C., Cooper, M., and Trusiak, A. 2001. Design and operation of Brunswick’s AG/
SAG Circuit. Pages 350–367 in Proceedings of the International Conference on
Autogenous and SemiAutogenous Grinding Technology. Volume IV. Vancouver, BC:
University of British Columbia.
Levin, J. 1989. Observation on the Bond standard grindability test, and a proposal for a
standard grindability test for fine materials. South African Institute of Mining and
Metallurgy 89(1):13–21.
Morrell, S. 1996. Power draw of wet tumbling mills and its relationship to charge
dynamics—Part 1: A continuum approach to mathematical modelling of mill power
draw. Transactions of the Institution of Mining and Metallurgy 105:C43–53.
———. 2004a. An alternative energysize relationship to that proposed by Bond for the
design and optimisation of grinding circuits. International Journal of Mineral
Processing 74:133–141.
———. 2004b. Predicting the specific energy of autogenous and semiautogenous mills
from small diameter drill core samples. Minerals Engineering 17(3):447–451.
———. 2004c. A new autogenous and semiautogenous mill model for scaleup, design
and optimisation. Minerals Engineering 17(3):437–445.
Morrell, S., Johnson, G., and Revy, T. 1991. A comparison through observation and
simulation of the power utilisation and performance of two dissimilar comminution
plants. Pages 157–160 in Fourth Mill Operators’ Conference, Australasian Institute of
Mining and Metallurgy, Burnie, Tasmania, March.
Morrell, S., and Morrison, R.D. 1996. AG and SAG mill circuit selection and design by
simulation. Pages 769–790 in Proceedings SAG 96. Volume 2. Vancouver, BC:
University of British Columbia.
Morrell, S., Shi, F., and Tondo, L.A. 1997. Modelling and scaleup of high pressure
grinding rolls. In Proceedings of the XX International Mineral Processing Congress
(IMPC), Aachen, Germany, September 1997.
NapierMunn, T.J., Morrell, S., Morrison, R.D., and Kojovic, T. 1996. Mineral
Comminution Circuits: Their Operation and Optimisation. JKMRC Monograph Series.
Brisbane, Australia: Julius Kruttschnitt Mineral Research Centre.
Parker, B., Rowe, P., Lane, G., and Morrell, S. 2001. The decision to opt for high pressure
grinding rolls for the Boddington expansion. Pages 93–106 in Proceedings of the
International Conference on Autogenous and SemiAutogenous Grinding Technology.
Volume III. Vancouver, BC: University of British Columbia.
Scott, A., Morrell, S., and Clark, D. 2002. Tracking and quantifying value from “Mine to
Mill.” In Proceedings of the Value Tracking Symposium. Melbourne, Australia: AusIMM.
129
PART 2
Comminution Practices
131
Causes and Significance of Inflections
in Hydrocyclone Efficiency Curves
S.K. Kawatra
*
and T.C. Eisele
*
ABSTRACT
Deviations of the shape of hydrocyclone efficiency curves from the ideal “S” shape have long
been reported, appearing as changes in slope, or inflections. These inflections can be divided
into two categories: “coarse” inflections, which are caused by differences in density of the
minerals being separated, and “fine” inflections, for which there are a number of competing
hypotheses concerning their cause. The existing literature addresses either the fine inflection
or the coarse inflection, but no papers have reported both types of inflection occurring at
once. This paper presents hydrocyclone results from both inplant studies and laboratory
experiments that show both coarse and fine inflections, and the industrial significance of
both types of inflections are discussed.
I NTRODUCTI ON
The performance of hydrocyclone classifiers is determined using efficiency curves, which
show the probability of a particle reporting to the hydrocyclone underflow as a function
of its size. This is expressed using selectivity functions, S(d), which represent the fraction
of the feed material in size fraction d that reports to the hydrocyclone underflow. This is
normally divided into (1) a bypass fraction that is taken to be equal to the value of the
water split, Rf, and represents the material that does not undergo classification; and (2) a
classification function, C(d), which has a smooth “S” shape, starting at 100% at the
coarse end and decreasing to zero at the fine end, as shown in Figure 1. The classifica
tion function can be expressed closely by equations such as (Plitt 1976)
(EQ 1)
or (Lynch and Rao 1968)
(EQ 2)
* Department of Chemical Engineering, Michigan Technological University, Houghton, Michigan
C d 1 e
0.693
d
d50c

© ¹
§ ·
m
–
– =
C d
e
D
d
d50c

© ¹
§ ·
1 –
e
D
d
d50c

© ¹
§ ·
e
D
2 – +
 =
132 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
where C(d) = classification function for particles of size d after correcting for particles
that bypass classification; d = particle size (Pm); d50c = particle size that has equal prob
ability of reporting to the overflow or the underflow, after correcting for the particles
that bypass classification; and m, Į = measures of the sharpness of separation.
If a hydrocyclone is expected to produce a standard Sshaped efficiency curve, then
it is possible to predict the corrected d50 size using a relationship such as the one shown
in Equation (3) (Plitt 1976):
(EQ 3)
where d50
c
= corrected d50 (Pm); D
c
= cyclone diameter (cm); D
i
= inlet diameter (cm);
D
o
= overflow diameter (cm); I = volumetric fraction of solids in feed; D
u
= underflow
diameter (cm); h = free vortex height (cm); Q = volumetric flow rate of feed (L/min); U
s
=
solid density (g/cm
3
); and U = liquid density (g/cm
3
).
However, there are a number of cases where the efficiency curve does not follow this
uniform, easily modeled shape. The curve can show inflections where the slope changes
abruptly, and in extreme cases the curve can reverse direction, as shown in Figure 2. This
is commonly referred to as a “fishhook” in the literature. The term comes from the fact
that when the smallest size measured corresponds to the portion where the curve is rising,
it resembles a hook.
A number of possible causes of efficiency curve inflections have been proposed in
the literature. Laplante and Finch (1984) explained the coarse inflection as being a result
of a combination of density and size distribution effects. The causes of the fine inflection
are less clear, and a variety of agglomeration, heavy media, entrainment, and fines recir
culation mechanisms have been proposed (Heiskanen 1993). To date, no investigators
have reported observing both the coarse inflection and the fine inflection in a single
1.0
0.5
0
F
r
a
c
t
i
o
n
o
f
S
i
z
e
t
o
U
n
d
e
r
f
l
o
w
d50
c
Size
FIGURE 1 Shape of an ideal corrected hydrocyclone efficiency curve
d50
c
50.5D
c
0.46
D
i
0.6
D
o
1.21
0.063I exp
D
u
0.71
h
0.38
Q
0.45
U
S
U –
0.5
 =
INFLECTIONS IN HYDROCYCLONE EFFICIENCY CURVES 133
efficiency curve, which has tended to contribute to confusion as to which type of inflec
tion is being addressed in any given paper.
When inflections occur at either coarse or fine sizes, it can be difficult to unambigu
ously determine the d50c size for the hydrocyclone, which greatly complicates efforts to
accurately model comminution circuits that incorporate hydrocyclones.
Coarse Inflections
From Equation (3), it is evident that the cut size is a function of the particle density. The
higher the density of a material, the smaller the d50 (Lynch and Rao 1968). With a hetero
geneous feed containing particles that vary in density, each feed component follows the
appropriate efficiency curve for particles of its density. The effect of this on a synthetic
mixture of lowdensity and highdensity particles is exemplified in Figure 3, which
includes the efficiency curves for pyrite (specific gravity = 5.0) and nonsulfides (specific
gravity 2.7), as well as the combined bulk efficiency curve for the mixture. Both of the pure
components clearly follow portions of classic Sshaped efficiency curves, but the com
bined curve deviates markedly from an S shape.
In order for the differences in particle density to produce a nonSshaped efficiency
curve, the particles must not only differ in density, but also must differ in size distribu
tion, with the heavier mineral being concentrated in the finer size fractions (Laplante
and Finch 1984). This results in the bulk efficiency curve following the shape of the low
density component at the coarser sizes, and then switching to following the highdensity
component at the finer sizes. If the relative percentages of highdensity particles and
lowdensity particles in each size fraction are known, then it is possible to calculate the
bulk efficiency curve based on the weighted averages according to the respective weight
percentages of dense and light components contained in the feed stream (Heiskanen
1993). This is calculated as shown in the following equation (NapierMunn et al. 1996):
1.0
0
F
r
a
c
t
i
o
n
o
f
S
i
z
e
t
o
U
n
d
e
r
f
l
o
w Coarse Inflection
Fine Inflection
FIGURE 2 A hydrocyclone efficiency curve showing two inflections, one at the coarse end and
one at the fine end
134 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
S(d) = f
L
(d)S
L
(d) + [1 – f
L
(d)]S
H
(d) (EQ 4)
where S(d) = bulk efficiency curve value for size fraction d; f
L
(d) = weight fraction that is
the light feed component for size fraction d; S
L
(d) = efficiency curve value of light feed
component for size fraction d; S
H
(d) = efficiency curve value of heavy (dense) feed com
ponent for size fraction d; and d = particle size (Pm)
The degree to which this effect occurs depends upon the difference in density
between types of particles. The most extreme case is reported by Banisi, Laplante, and
Marois (1991), where metallic gold particles in a silicate ore produce a pronounced
inflection at an unusually fine size, due to the extremely high density of gold (19.3 g/cm
3
).
Fine Inflections
Efficiency curve inflections at the coarser sizes are satisfactorily explained by differences
in particle density and size distribution; however, the inflections at the finer sizes are a
completely different phenomenon that has resisted reliable measurement or modeling.
In the case of the fine inflections, there are numerous possible proposed mechanisms
that have not been conclusively demonstrated. It has also not been demonstrated
whether the fine inflection is a significant consideration, because it occurs at the very
fine particle sizes where there is often only a very small fraction of the total material
present. It has even been proposed that the phenomenon of fine inflections does not
even exist, and that its appearance is due mainly to the difficulty of accurately measuring
particle sizes and weights at very fine sizes. The effect is sometimes elusive, and some
authors have not observed the fine inflection even after extensive experiments (Coelho
and Medronho 2001).
The fine inflection is well known in air classifiers and can be expressed mathemati
cally as the result of internal recycling and reclassification of material in the classifier
100
0
10
20
30
40
50
60
70
80
90
%
F
e
e
d
t
o
H
y
d
r
o
c
l
o
n
e
U
n
d
e
r
f
l
o
w
100 10 1,000
Pyrite Feed
Bulk Feed
Nonsulfide Feed
After Laplante and Finch 1984.
FIGURE 3 Example of an inflection at the coarse end of a hydrocyclone efficiency curve due to
the presence of particles of very different densities
INFLECTIONS IN HYDROCYCLONE EFFICIENCY CURVES 135
(Luckie and Austin 1975). However, air classifiers have a significantly different design
than hydrocyclones, with areas that clearly can be identified as producing two stages of
classification. It is the interaction of these two stages that leads to the inflection in air
classifier efficiency curves. This analysis, therefore, does not apply directly to hydrocy
clones, which lack the internal structures needed to produce multiple stages of classifica
tion within a single unit.
A fine inflection can be modeled by assuming that it is due to fines entrainment in
the hydrocyclone underflow (Finch 1983; Del Villar and Finch 1992; Kelly 1991). The
model based on this assumption becomes
S(d) = C(d) + a(d) (EQ 5)
where C(d) = fraction of the feed of size d that is recovered to the underflow as a result of
classification forces only, and a(d) = fraction of the feed of size d that is carried to the
underflow independently of the classification. The bypass function a(d) is normally cal
culated from the water bypass fraction, Rf, as follows:
a(d) = Rf(1 – C(d)) (EQ 6)
If it is assumed that, instead, the value of a(d) instead increases linearly with
decreasing values of d, and becomes equal to the water bypass fraction (Rf ) when d = 0,
then the Plitt equation for the uncorrected efficiency curve becomes
(EQ 7)
where S(d) = selectivity function for particles of size d, uncorrected for the bypass frac
tion; Rf = fraction of the water entering with the feed that reports to the hydrocyclone
underflow; and d
0
= maximum particle size that is entrained by the water flow into the
underflow while bypassing classification.
This model does predict a small inflection in the efficiency curve as long as the value
of d
0
is quite fine, as shown by the example curve given in Figure 4. It is consistent with
the observation that, in many cases, the fine inflection is most pronounced when the
value of Rf is quite high (Pasquier and Cilliers 2000). However, the model does not pre
dict two features that are sometimes reported for the fine inflection: (1) It does not allow
the value of S(d) for the small particles to exceed the value of Rf; and (2) it does not
account for cases where S(d) for the fine particles first rises and then falls again.
A frequently proposed mechanism for the fine inflection is fines agglomerating to
the coarse solids, which are preferentially carried to the underflow, as shown in Figure 5
(Heiskanen 1993; Finch 1983). This mechanism is quite plausible; however, it is difficult
to conclusively prove and even more difficult to reliably model. In any case, the fine
inflection has been observed even in cases where measures have been taken to ensure
thorough dispersion (Rouse, Clayton, and Brookes 1987). Therefore, while fines agglom
eration may be a contributing factor in many cases, it is unlikely to be the cause of all
observed cases of the fine inflection. It would be expected that the fine inflection would
be most pronounced when there are many coarse particles and few fine particles, as then
there would be significant surface area available to carry the fines. It also would be
expected that the magnitude of the fine inflection would vary depending on the proper
ties of the coarse particles that could lead to agglomeration.
Another possible mechanism is based on fine particles being entrained in the wake
behind coarser particles, as shown in Figure 6 (Neesse, Dueck, and Minkov 2004;
S d 1 e
0.693
d
d50c

© ¹
§ ·
m
–
– Rf
d
0–d
d
0

© ¹
§ ·
+ =
136 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
1.0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
F
r
a
c
t
i
o
n
t
o
U
n
d
e
r
f
l
o
w
110 60 10
S(d) Observed
S(d) Calculated
C(d)
a(d)
After Finch 1983.
FIGURE 4 Selectivity curve for a 25cm hydrocyclone, sized by microsieving to 25 μm, and
modeled using Equation (7) with the following parameters: d50c = 70 μm; m = 3.2; d
0
= 100 μm,
Rf = 0.47. In addition to the selectivity function S(d), the entrainment function a(d) and the
classification function C(d) are also shown.
Fluid Flow
Agglomerated
Fines
Particle Motion
Coarse
Particle
FIGURE 5 Schematic of the fines agglomeration effect, with a coarse particle carrying fines that
are attached to its surface
INFLECTIONS IN HYDROCYCLONE EFFICIENCY CURVES 137
Kraipech et al. 2002). This is supported by the observation that when particles of differ
ent sizes are settling together, the settling rate of the finer particles is increased compared
to their settling rate when no coarse particles are present. This effect can be quantified,
and models based on it can show a quite pronounced fine inflection. It is reported that
this effect becomes experimentally noticeable at particle sizes <3.5 ȝm for lime particles,
glass beads, and dusts suspended in water (Kraipech et al. 2002). The quantity of fines
that could be carried to the underflow would depend on the volume of coarse particle
wakes, and would therefore increase as the number of coarse particles increased. It also
would be expected that there would be relatively little dependence on the coarse particle
properties, and the effect would depend only on the size and settling velocity of the
coarse particles.
The objective of this paper is to examine the effects leading to nonSshaped effi
ciency curves from a practical standpoint, and to determine what the presence of these
inflections tells us about the performance of hydrocyclones and the comminution circuits
that contain them. This is accomplished both by examining the results of analyses of plant
samples and by conducting controlled laboratory experiments to confirm these results.
PLANT SAMPLI NG STUDI ES
The plant examined was a magnetite concentrator located in the Lake Superior iron ore
district. This type of plant was selected both because iron ore is one of the most high
volume metallic ores produced and is therefore of considerable practical importance,
and because the feed processed contains approximately equal quantities of a highdensity
mineral (magnetite, ȡ = 5.18) and a lowdensity mineral (silica, ȡ = 2.65), ensuring that
the shape of efficiency curves from the hydrocyclones would be equally affected by the
behavior of both minerals.
The hydrocyclones sampled were part of a standard hydrocyclone/pebble mill circuit,
as shown in Figure 7. The circuit normally operated with 14 cyclones, with 2 additional
cyclones available as spares.
Samples were collected from the cyclone feed, cyclone underflow, and circuit product
as part of an overall survey of the plant performance. Size analyses of the samples were
carried out by three methods:
Coarse
Particle
Fine Particles
Trapped in Wake of
Coarse Particle
FIGURE 6 Fine particles trapped in the wake of a coarse particle, carried along by it to the
hydrocyclone underflow
138 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
1. Wet sieving at 25 ȝm using a wovenwire test sieve, followed by dry sieving of
the +25ȝm particles using wovenwire test sieves in a RoTap sieve shaker
2. Microsieving of dry powders using electroformed nickelfoil sieves in a Sonic
Sifter apparatus, which allowed sieving down to 10ȝm particle size
3. Microtrac laser diffraction particlesize analysis to measure particle sizes down
to 1 ȝm, primarily as a check on the accuracy of the sieve analyses
For the sieved samples, each individual size fraction was assayed using a dichromate
titration method to determine the iron assay for every sieved size fraction for each
stream. The size distribution and assay data were then mass balanced, and the magnetite
concentration in each size fraction was calculated from the iron assays.
Sampling Results
A summary of the results from plant sampling around the hydrocyclones is shown in
Table 1. Note that the –500 mesh fraction of the cyclone underflow product is highly
enriched in magnetite relative to the cyclone overflow. This indicates that the magnetite
is being retained in the grinding circuit until it is ground to a finer size than the silicate
gangue, and as a result is being significantly concentrated into the finer size fractions.
The size analyses of the individual components (magnetite and quartz) were then
used to calculate the hydrocyclone efficiency curves for the magnetite phase, the quartz
phase, and the overall combined result. These efficiency curves are shown in Figure 8.
Discussion—Plant Sampling
The efficiency curves in Figure 8 illustrate the effect of the different mineral densities on
the hydrocyclone separation size. While the hydrocyclone begins to remove quartz from
the circuit at a fairly coarse size (d50 = 39 Pm), it does not remove the magnetite until a
Circuit Product (COF)
Cyclone Feed
Circuit New
Feed (CNF)
Sump
C
y
c
l
o
n
e
U
n
d
e
r
f
l
o
w
(
C
U
F
)
Pebbles
Pebble Mill Discharge (PMD);
Recirculating Load = 250%
Krebs 15in. (38.1cm) Cyclone
16 Total, 2 StandBy
OF: 5.25 in. (13.3 cm); UF: 3 in. (7.6 cm)
Chip Removal
Screen
Pebble
Chips
Pebble Mill
15.5 ft x 32.5 ft
(4.7 m x 9.9 m)
2,650 hp
(1,976 kW)
NOTES:
The total flow rates of the various streams for the particular circuit sampled were as follows:
Circuit New Feed: 120.6 ltph (long tons of dry solids per hour) or 122.5 tph (metric tons per
hour); Cyclone underflow: 299.3 ltph (304 tph); Pebble mill discharge: 306.9 ltph (312 tph);
Circuit product: 128.1 ltph (130.1 tph). The hydroclone feed was 13.2% solids.
FIGURE 7 Configuration of the hydrocyclone/pebble mill circuit sampled for this study
INFLECTIONS IN HYDROCYCLONE EFFICIENCY CURVES 139
significantly finer size (d50 = 20 Pm) is reached. The hydrocyclone underflow between
approximately 20 Pm and 39 Pm is therefore predominantly magnetite. This material is
then returned to the grinding circuit and reground until it finally becomes fine enough to
be removed. This results in a substantial accumulation of magnetite in the hydrocyclone
underflow in this size range, as shown in Figure 9.
The effect of this retention of the higher density fraction in the circuit is seen in the
overall efficiency curve of Figure 8. The magnetite and quartz efficiency curves are very
close to the ideal S shape, however, the overall curve shows an inflection. This is a direct
result of the fact that the cyclone feed consists of approximately equal quantities of mag
netite and quartz at sizes coarser than 39 Pm; but at finer sizes, it is primarily composed
* Flow rates are in long tons per hour. Percentages are relative to the overall flow rate for the corresponding size fraction.
Note that the cyclone underflow, particularly the –500 mesh fraction, is considerably enriched in magnetite compared
to the cyclone feed and overflow.
TABLE 1 Flow rates for total solids, magnetics, and nonmagnetics for the material entering and
leaving the hydrocyclones
*
Stream
Total +500 Mesh Fraction –500 Mesh Fraction
Overall Magnetic Nonmagnetic Overall Magnetic Nonmagnetic Overall Magnetic Nonmagnetic
Cyclone feed 427.5 257.7
(60.3%)
169.8
(39.7%)
226.9 122.0
(53.8%)
104.9
(46.2%)
200.6 135.7
(67.6%)
64.9
(32.4%)
Cyclone
overflow
128.1 73.5
(57.4%)
54.6
(42.6%)
16.1 7.0
(43.5%)
9.1
(56.5%)
112 66.5
(59.4%)
45.5
(40.6%)
Cyclone
underflow
299.3 201.4
(67.3%)
97.9
(32.7%)
207.1 115.8
(55.9%)
91.3
(44.1%)
92.2 85.6
(92.8%)
6.6
(7.2%)
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
F
r
a
c
t
i
o
n
o
f
S
i
z
e
t
o
U
n
d
e
r
f
l
o
w
10 100
Inflection
Magnetics
Nonmagnetics
Overall
FIGURE 8 Hydrocyclone efficiency curves determined from overflow and underflow samples
collected from an operating hydrocyclone in the magnetite concentrator studied
140 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
of magnetite. As a result, the overall efficiency curve is initially the average of the mag
netite and quartz efficiency curves, but then rapidly switches to follow the overall efficiency
curve once it reaches sizes where the quartz has been removed from the circuit. The
result is an inflection in the overall efficiency curve.
The practical significance is that such an inflection only occurs when the cyclone
feed has become enriched in the denser mineral in the fine sizes. If the highdensity and
lowdensity minerals have the same size distribution, then the overall bulk efficiency
curve will simply be the average of the curves for the high and lowdensity minerals.
The inflection can only be seen when the overall curve switches from following the low
density curve to following the highdensity curve, which can only happen if there is a dif
ference in size distributions between the two minerals.
LABORATORY STUDI ES
In order to more closely examine the coarse inflection phenomenon, and to determine
how the relative size distributions affect the degree of the inflection, a series of labora
tory experiments were carried out using controlled size distributions for the hydrocy
clone feed. Also, the plant sampling was unable to determine whether a fine inflection
was occurring, so the laboratory experiments were designed to be able to detect the fine
inflection.
Equipment
A 10.2cmdiameter Krebs hydrocyclone was used, mounted on a test rig with a variable
speed pump, pressure gauge, ultrasonic flowmeter, and sampling mechanism for collecting
simultaneous overflow and underflow samples. The hydrocyclone dimensions were
Feed inlet: 3.5 cm
40
5
10
15
20
25
30
35
0
%
o
f
T
o
t
a
l
H
y
d
r
o
c
y
c
l
o
n
e
U
n
d
e
r
l
f
o
w
10 100 1,000
Magnetics
Nonmagnetics
FIGURE 9 Quantities of magnetite and quartz by size fraction in the hydrocyclone underflow
collected from an operating hydrocyclone in a magnetite concentrator plant. The particles larger
than 20 μm and smaller than 39 μm were primarily magnetite.
INFLECTIONS IN HYDROCYCLONE EFFICIENCY CURVES 141
Vortex finder diameter: 3.18 cm
Spigot diameter: 1.59 cm
The hydrocyclone test rig consisted of a sump with a variablespeed centrifugal
pump circulating material continuously through the hydrocyclone. An oilfilled pressure
gauge was used to monitor hydrocyclone inlet pressure. Flow rates in the hydrocyclone
test rig were monitored using an ultrasonic Doppler flowmeter. Samples were collected
using a specially designed sample cutter that simultaneously collected samples from
both the cyclone overflow and underflow, so that the relative flow rates of the two
streams could be accurately measured. Particlesize distributions were determined using
a Leeds and Northrup Microtrac laser diffraction particlesize analyzer.
Materials and Procedures
Materials used were finely ground quartz sand and magnetite concentrate. Experiments
were first conducted using magnetite alone and quartz alone, followed by experiments
with mixtures of magnetite and quartz. All samples were collected in triplicate at each
hydrocyclone operating condition, and each of the triplicate samples were analyzed sep
arately to determine the random variations from test to test. In all experiments, the random
variation in the fraction of each size reporting to the underflow was less than r0.01 units.
This is smaller than the magnitude of the inflections that were observed, demonstrating
that the inflections were actually present and not simply due to experimental error.
Experiments with Single Minerals. For experiments using magnetite alone and using
quartz alone, the size distributions of the hydrocyclone feeds are shown in Figure 10. For
these experiments, the size distributions of magnetite and quartz were reasonably simi
lar, and in particular they had similar quantities of particles finer than 10 ȝm. This made
it possible to determine whether there were differences in the fine inflections for the two
minerals that could be attributed to differences in the mineral properties.
Experiments were run at three different percent solids for each mineral (2.5%,
5.5%, and 16.5% solids for the magnetite; 2.3%, 8.4%, and 16.9% solids for the quartz).
At each percent solids, the cyclone inlet pressure was varied over the range of 5 to 20 psi
to alter the hydrocyclone flow rate.
Experiments with Magnetite/Quartz Mixtures. Additional experiments were con
ducted with 50:50 mixtures by weight of magnetite and quartz. Two sets of these experi
ments were conducted, with the size distributions of the quartz and magnetite being
very different. This was done in order to produce the maximum coarse inflection. In the
first set of experiments, the quartz was the underflow (coarse) product from the cyclone
experiments using quartz alone, while the magnetite was the overflow (fine) product
from the cyclone experiments using magnetite alone. This resulted in a moderate amount
of overlap between the magnetite and quartz size distributions, as can be seen in Figure 11.
The second set of 50:50 magnetite–quartz experiments used quartz that had been
further processed by sedimentation to remove the finest particles, and magnetite that
had been processed by sedimentation to remove the coarsest particles, with the coarse
magnetite particles then being ground using an attrition mill and added back to produce
a finer magnetite size distribution. This resulted in much less overlap between the mag
netite and quartz size distributions, and the magnetite being significantly finer than in
the first set of experiments (as shown in Figure 11).
For these experiments, the percent solids was varied from as high as 30% solids to as
low as 2.5% solids, and the cyclone operating pressure was varied from 7 to 20 psi to
alter the flow rates.
142 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
100
10
20
30
40
50
60
70
80
90
0
C
u
m
u
l
a
t
i
v
e
%
P
a
s
s
i
n
g
10 100 10 1,000
Magnetite
Quartz
FIGURE 10 Magnetite and quartz size distributions used in laboratory hydrocyclone experiments
using magnetite alone and quartz alone
100
10
20
30
40
50
60
70
80
90
0
C
u
m
u
l
a
t
i
v
e
%
P
a
s
s
i
n
g
10 100 10 1,000
Quartz Set 1
Quartz Set 2
Magnetite Set 1
Magnetite Set 2
Overlap between
Magnetite and Quartz
Size Distributions
FIGURE 11 Magnetite and quartz size distributions used for experiments with a 50:50 mixture
by weight. Two sets of experiments were run, with the second set having less material in the
“overlap” region than did the first set.
INFLECTIONS IN HYDROCYCLONE EFFICIENCY CURVES 143
Results and Discussion
Individual Mineral Results. The results for the individual minerals did not show a
coarse inflection, due to the feed being composed of minerals of a single density. How
ever, both the quartz and magnetite did show a fine inflection.
For quartz, the fine inflection was relatively small, as shown by the example curves
in Figure 12, but was still much larger than the random variations between triplicate
samples, demonstrating that it was a real effect and not due to random error. The fine
inflection was only slightly affected by percent solids, and was completely unaffected
by flow rate, although the d50 size for the remainder of the curve varied, as predicted by
Equation (3).
For pure magnetite, however, the fine inflection shown in Figure 13 was much more
pronounced than the fine inflection for silica shown in Figure 12. The magnetite feed had a
size distribution similar to the quartz feed, and so the magnitude of the fine inflection
would be expected to be similar in both cases if the same mechanism was responsible for
the fine inflection. This very large difference between the quartz and the magnetite, and
the fact that the magnetite is much more sensitive to flow rate and percent solids, clearly
indicates that different mechanisms are responsible in each case.
Mineral Mixture Results. Experiments with mixtures of minerals were able to
show both the coarse inflection and the fine inflection. In the first set of experiments,
with a relatively large overlap of the magnetite and quartz particle sizes, the coarse
inflection in the efficiency curves was not observed, although the fine inflection was
present, as shown in Figure 14. The second set of experiments, with very little overlap
between the magnetite and quartz size distributions, showed a very visible coarse inflec
tion at the lower values of the percent solids, as can be seen in Figure 15. This coarse
inflection was clearly visible in all three of the triplicate samples in each of the experi
ments where it appeared, and was significantly larger than the random variations
between triplicates, indicating that it is a real effect and not due to random errors. In
these experiments, it can be seen that as the percent solids increased, the d50 also increased
as would be expected from Equation (3). This caused the efficiency curve to sweep
through the size where the feed changes from being entirely quartz to being entirely
magnetite. As a result, the inflection occurs near the top of the curve at the lowest per
cent solids, and the inflection moves down the curve as the percent solids increases and
the d50 size coarsens. Note that the coarse inflection is most easily visible near the top of
the curve; however, it is difficult to distinguish at all when it is near the middle or bottom
of the curve, as it simply appears to be a decrease in the sharpness of separation.
The fine inflection also can be seen in the results shown in both Figure 14 and
Figure 15. It is very interesting to note that, even though all of the material producing
the fine inflection in these experiments is magnetite, the behavior is much more similar
to that of the pure quartz curve (Figure 12) than the pure magnetite curve (Figure 13).
This shows that the magnitude of the fine inflection is controlled by the nature of the coarse
particles, not by the nature of the fine particles. This is not consistent with the entrain
ment model for the fine inflection (Finch 1983), as in that case the fine inflection would
be controlled entirely by the characteristics of the fine particles.
Also, the behavior of the fine inflection is not entirely consistent with the hydrody
namic model (Neesse, Dueck, and Minkov 2004), as this model would not predict such a
large alteration in behavior simply by changing the density of the coarse particles. The result
is much more consistent with agglomeration of fine particles onto the coarse surfaces. In this
particular system, there is potential for magnetic agglomeration of fine magnetite particles
onto the coarse magnetite, which would account for the much higher value of the fine
inflection for pure magnetite than for pure quartz. When there is no coarse magnetite
144 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
1.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
F
r
a
c
t
i
o
n
o
f
S
i
z
e
t
o
U
n
d
e
r
f
l
o
w
1 10 100
17.5% Solids
8.1% Solids
2.5% Solids
Fine Inflection
FIGURE 12 Efficiency curves for pure silica as a function of percent solids at a fixed flow rate
(110 L/min). At each percent solids, samples were collected in triplicate and analyzed
independently to ensure reproducible results.
1.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
F
r
a
c
t
i
o
n
o
f
S
i
z
e
t
o
U
n
d
e
r
f
l
o
w
1 10 100
202 L/m
157 L/m
109 L/m
NOTES:
Flow rate was varied from 109 L/m to 202 L/m, at a constant percent solids of 5.5%.
The fine inflection for pure magnetite was significantly affected by flow rate.
FIGURE 13 Efficiency curves for pure magnetite as a function of flow rate
INFLECTIONS IN HYDROCYCLONE EFFICIENCY CURVES 145
present to carry the fine magnetite particles and only nonmagnetic quartz particles are
present, then the behavior of the fine magnetite becomes much more similar to that of
the fine quartz.
An additional phenomenon that was noted with the mixedmineral experiments is
that when the magnetite was made finer and the quantity of material in the particle size
range from 1 to 5 Pm increased, the magnitude of the fine inflection was reduced. This
can be seen by comparing Figure 14 (which had a smaller amount of material in the fine
fraction and exhibited a large fine inflection) with Figure 15 (which had a larger amount
of material in the fine fraction and exhibited a small fine inflection). This indicates that
only a portion of the fine material can follow the fine inflection, with the remainder
behaving as would be predicted by conventional classification theory. If the fine inflec
tion is being caused by the fine particles agglomerating onto the coarser particles, then
there is a limited amount of fine material that can be recovered before all of the available
coarseparticle surface area is used up. Similarly, if the fine inflection is caused by fine
particles being carried in the wakes of the coarse particles, there is a limited amount of
wake volume available to carry the particles. As a result, the fine inflection is only signif
icant when there is relatively little material present in the finest size range. As the
material in the finest sizes increases, the behavior of the fines that are not being trans
ported by coarse particles swamps the effects of the fines being carried by the coarse par
ticles, and the fine inflection becomes negligible. In the results reported here, even when
there is a very large fine inflection, the size distributions shown in Figures 10 and 11
reveal that less than 10% of the total weight of the feed was affected, and so the fine
inflection has a minimal effect on the actual composition of the bulk cyclone products.
1.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
F
r
a
c
t
i
o
n
o
f
S
i
z
e
t
o
U
n
d
e
r
f
l
o
w
1 10 100
31% Solids
14% Solids
5.1% Solids
2.4% Solids
NOTES:
Inlet pressure was 12 psi, providing a flow rate of approximately 200 L/m, and the
percent solids was varied as indicated. For each experiment, samples were collected
and analyzed in triplicate to ensure reproducible results. The coarse inflection is
difficult to see because of the relatively large overlap in the quartz and magnetite size
distributions, but the fine inflection is prominent.
FIGURE 14 Representative efficiency curves for the first set of mixedmineral experiments
146 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
CONCLUSI ONS
There are two distinct phenomena that lead to inflections in hydrocyclone efficiency
curves. The coarse inflection is caused by the presence of fine, highdensity particles
combined with coarser, lowdensity particles in the cyclone feed. This results in the effi
ciency curve being dominated by lowdensity materials at the coarser sizes and highdensity
materials at the finer sizes, with the inflection occurring when the curve switches from
one to the other. In a comminution circuit, such an inflection is an indicator that high
density particles are being preferentially retained in the grinding circuit and are being
overground. As overgrinding is a significant waste of energy and can cause losses of valu
able minerals, the appearance of a coarse inflection in a hydrocyclone efficiency curve in
a plant is a sign that there is a problem that needs to be corrected.
The fine inflection, on the other hand, has been explained by a number of theories,
many of which do not fully account for the behavior seen in the results reported here. The
observed behavior is most consistent with an agglomeration mechanism, where fine par
ticles agglomerate onto coarse particles and are carried into the cyclone underflow. This
mechanism can only carry a limited quantity of fines, and so when there are very large
quantities of fine particles present, the majority of them follow the theoretical curve
because there is insufficient coarse particle surface to carry them all. As a result, the fine
inflection does not appear likely to be of much industrial significance, as it is only clearly
seen in cases where there is very little of the cyclone feed at the affected sizes.
1.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.0
F
r
a
c
t
i
o
n
o
f
S
i
z
e
t
o
U
n
d
e
r
f
l
o
w
1 10 100
27% Solids
21.7% Solids
9.7% Solids
5.8% Solids
NOTES:
For each experiment, samples were collected and analyzed in triplicate to ensure
reproducible results. The coarse inflection is only easily seen at the two lowest
percent solids. The increased quantity of magnetite in the fine fraction also resulted
in a smaller fine inflection than was seen in the experimental results shown in
Figure 14.
FIGURE 15 Variations in efficiency curves for a quartz/magnetite mixture with very little overlap
between the magnetite and quartz size distributions as the percent solids in the cyclone feed changes
INFLECTIONS IN HYDROCYCLONE EFFICIENCY CURVES 147
ACKNOWLEDGMENTS
This project was partially supported by the U.S. Department of Energy under Grant
No. DEFC2601NT41062. The support of the ClevelandCliffs Iron Co. is also gratefully
acknowledged. The authors would also like to thank Ted Weldum, Todd Davis, Gary
Rajala, and Ron Mariani for their considerable advice and assistance in carrying out this
work. Laboratory experiments were conducted by H.J. Walqui and J.G. Jelsma.
REFERENCES
Banisi, S., Laplante, A.R., and Marois, J. 1991. The behavior of gold in Hemlo Mines Ltd.
grinding circuit. CIM Bulletin 84(955):72–78.
Coelho, M.A.Z., and Medronho, R.A. 2001. A model for performance prediction of
hydrocyclones. Chemical Engineering Journal 84(1):7–14.
Del Villar, R., and Finch, J.A. 1992. Modelling the cyclone performance with a size
dependent entrainment factor. Minerals Engineering 5(6):661–669.
Finch, J.A. 1983. Modelling a fishhook in hydrocyclone selectivity curves. Powder Tech
nology 36:127–129.
Heiskanen, K. 1993. Pages 59–60, 87, 264–272 in Particle Classification. 1st edition. Lon
don: Chapman and Hall.
Kelly, E.G. 1991. The significance of bypass in mineral separators. Minerals Engineering
4(1):1–7.
Kraipech, W., Chen, W., Parma, F.J., and Dyakowski, T. 2002. Modelling the fishhook
effect of the flow within hydrocyclones. International Journal of Mineral Processing
66:49–65.
Laplante, A.R., and Finch, J.A. 1984. The origin of unusual cyclone performance curves.
International Journal of Mineral Processing 13:1–11.
Luckie, P.T., and Austin, L.G. 1975. Mathematical analysis of mechanical air separator
selectivity curves. Transactions of the Institute of Mining and Metallurgy 84:C253–C255.
Lynch, A.J., and Rao, T.C. 1968. The operating characteristics of hydrocyclone classifiers.
Indian Journal of Technology 6:106.
NapierMunn, T.J., Morrell, S., Morrison, R.D., and Kojovic, T. 1996. Pages 326–327 in
Mineral Comminution Circuits: Their Operation and Optimisation. Brisbane, Australia:
JKMRC, University of Queensland.
Neesse, Th., Dueck, J., and Minkov, L. 2004. Separation of finest particles in hydrocy
clones. Minerals Engineering 17:689–696.
Pasquier, S., and Cilliers, J.J. 2000. Submicron particle dewatering using hydrocyclones.
Chemical Engineering Journal 80(1–3):283–288.
Plitt, A.J. 1976. A mathematical model of the hydrocyclone classifier. CIM Bulletin
69(776):115–123.
Rouse, B.D., Clayton, J.S., and Brookes, G.F. 1987. Confirmation of modelling techniques
for small diameter cyclones. Pages 7–20 in Proceedings of the 3rd International Con
ference on Hydrocyclones. Edited by P. Wood. Amsterdam: Elsevier.
149
SimulationBased Performance
Improvements in the Ispat Inland
Minorca Plant Grinding Circuit
S. Ersayin,
*
W.M. Bond,
†
R. Strukel,
†
J. Arola,
†
and B. Kettunen
‡
ABSTRACT
This study was initiated by plant sampling. Raw data generated by sample analysis were
mass balanced and used for performance assessment and model fitting. Evaluation of mass
balanced data indicated that the ball mill circuit at the Ispat Inland Minorca plant was the
bottleneck limiting throughput. A team of engineers developed performance improvement
ideas to alleviate the load around the circuit. These included modifications that have a
direct effect on the circuit, such as makeup ball size, volumetric ball charge, critical speed,
percent solids in the mill, more efficient hydrocyclones, and replacing of hydrocyclones with
stack sizers. Other modifications that have indirect effects—namely, dry cobbing, improved
fine screening, and separate grinding of fine screen oversize—were also simulated. Simula
tions were carried out using an improved version of Usim Pac mineral processing simulation
software. Improvements included incorporation of magnetic separator, hydroseparator, and
fine screen models into the software. Although the simulations were aimed at reducing the
load around the ball mill, complete plant simulations were carried out to determine the
effects of modifications on downstream flows and to simulate the upstream effects of
increased feed rates.
Results of simulations indicated that all the modifications would provide some degree of
benefit to the circuit/plant performance. However, the most promising and feasible alterna
tive was the use of finer makeup balls with more efficient hydrocyclones. The finer makeup
ball size modification was immediately implemented in the plant, and substantial improve
ments in throughput were obtained. Plant sampling was repeated to quantify benefits from
this modification. Massbalanced data indicated a very good fit between actual and pre
dicted performances. As a result of simulationbased modifications at the plant, throughput
was increased more than 10%. Further improvements are expected following replacement of
existing 15in. cyclones by more efficient 20in. units.
I NTRODUCTI ON
The Ispat Inland plant annually processes approximately 9 Mt of crude magnetitebearing
iron ore (taconite) to produce 2.8 Mt of final concentrate containing less than 4% silica.
* University of Minnesota, Coleraine Minerals Research Laboratories, Coleraine, Minnesota
† Mittal Steel, Minorca Mine (formerly Ispat Inland Mining Co.), Virginia, Minnesota
‡ Noramco Engineering, Hibbing, Minnesota
150 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
The plant consists of magnetic separation and flotation circuits. The magnetic separation
circuit has three parallel lines and produces magnetic concentrate containing 6%–7% silica.
The flowsheet of the magnetic separation circuit is shown in Figure 1. Fine ore (–25 mm)
coming from the crushing plant goes through two stages of grinding and three stages of
magnetic separation. Magnetic concentrate from all three lines is combined and fed to a
flotation circuit floating silica using amine as a collector. Flotation is carried out in two
stages: rougher cells produce a final magnetite concentrate, whereas rougher floats
(tails) are further treated in scavenger cells to recover magnetite carried into the tail
stream (Ersayin et al. 2005).
The final product from the plant is fired iron ore pellets. Final concentrate from con
centration circuits goes through filtering, balling, and induration steps to produce the
final product. The pellet plant has ample capacity. Therefore, pellet production is con
centrate limited. It is estimated that the pellet plant could easily process 3 Mt of concen
trate annually. This created an incentive for plant engineers to search for improved
concentrator performance that could increase the throughput by 10%.
The integrated size reduction and concentration nature of iron ore processing and
lack of reliable models for magnetic separation and other unit operations specific to iron
ore processing had prevented iron ore processing plants from making full use of mineral
processing simulation techniques. Triggered by impressive performance improvements
in the National Steel Pellet Company’s plant through a combination of modeling and a
pilotscale testing study (Wennen, Nordstrom, and Murr 1995), the iron ore producers
on the Iron Range in northern Minnesota decided to establish the Concentrator Model
ing Center to develop mathematical models needed for reliable simulation of their plants
and to provide simulationbased services. The Iron Ore Cooperative Research Program of
the Minnesota Department of Natural Resources provided funding for the center, which
was established within the Coleraine Minerals Research Laboratory (CMRL), University
of Minnesota. Following the development of basic models for magnetic separators,
hydroseparators, and fine screens, the center was ready to carry out reliable simulations
of taconite plants. In 2002, the U.S. Department of Energy provided major funding for
the project to demonstrate how mineral processing simulation can reliably be used for
improving performance of iron ore processing plants. This paper summarizes a large
portion of this study aimed at improving ball mill grinding circuit efficiency.
SI MULATI ON BASED STUDY
The study was initiated by plant sampling and followed typical steps of efficiency
improvement plans, including sample analysis, mass balancing, performance evaluation,
development of alternatives for improved performance, modeling and plant simulation,
selection of most feasible alternatives, plant
implementation, and validation of simula
tion results. Details of these steps are described in the following subsections.
Plant Sampling and Sample Analysis
Prior to sampling, operating conditions were checked to ensure that the plant was oper
ating under steadystate conditions. All the streams in one of the magnetic separation
circuit lines from rod mill feed to magnetic concentrate were sampled. Stream sampling
was repeated at hourly intervals during an entire shift. Although plant sampling also
included the flotation circuit to assess potential effects of modifications in the magnetic
circuit on flotation performance, this portion of the study was excluded from this paper
due to its focus on the grinding circuit. Details of flotation circuit study can be found
elsewhere (Ersayin et al. 2005). Plant sampling was repeated for the second ore blend
that is processed during a different period in the year. However, only the data for the first
SIMULATIONBASED PERFORMANCE IMPROVEMENTS 151
blend are presented in this paper because plant data indicated that the circuit was oper
ating at its limiting (maximum) capacity during sampling of this particular blend. Such
plant data formed a good basis for comparison with simulations aimed at alleviating
bottlenecks.
Stream samples were immediately filtered and dried at the plant. Then they were
transferred to the CMRL for sample analysis work, which included size analysis, sizeby
size total iron, Satmagan iron, and silica analysis. Bond ball mill grindability tests were
also carried out to assess the efficiency of the ball mill grinding circuit and to have a ref
erence point for future plant sampling data. In these tests, cobber magnetic separator
concentrates, which were the fresh feed to ball mill in the circuit, were used.
Mass Balancing and Performance Evaluation
Raw size and chemistry data generated by sample analysis were mass balanced using the
massbalancing algorithm of Usim Pac (BRGM, Orleans, France) mineral processing soft
ware. Based on a measured rod mill feed rate of 350 tph, all the flow rates within the cir
cuit were calculated. In general, massbalanced data provided a very good fit to raw
data, indicating quality of sampling. The only node that created a minor massbalancing
problem was the cobber magnetic separators, due to difficulty in obtaining a representative
sample from a large volume and coarsesized stream of rod mill discharge. Massbalanced
data for major streams are summarized in Table 1. From massbalanced data and flow
rates, magnetic iron losses in tailing streams and recovery in magnetic concentrate were
calculated (Table 2). The circuit had a high magnetic iron recovery of 96.2%, with much
of the losses occurring in cobber and rougher tails. The circulating load ratio (hydrocy
clone underflow to overflow) around the ball mill circuit was 380%. Plant data also
showed that existing hydrocyclones were performing poorly (i.e., they had a bypass of
more than 40%; Figure 2). A smaller fraction of the circulating loads was coming from
the fine screens. Due to density effect, bypassing fines were lowsilica fractions unneces
sarily circulated back to the ball mill. Plant data showed that hydrocyclone underflow
contained approximately 25% concentrate quality material (Table 3).
As shown in Figure 2, fine screening was not particularly efficient. These devices
also act as a concentration device separating coarse silica, so that screening efficiency
Rod Mill Feed
Ball Mill
Roughers Cyclones
Hydroseparator
Fine Screens
Magnetic Concentrate
Finishers
Tails Thickener
Fine Tails
O/F
Coarse Tails
Cobbers
FIGURE 1 Existing flowsheet of the Ispat Inland plant magnetic circuit
152 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
was compromised to obtain low silica in the magnetic concentrate. Operating them at
high feed percentsolids results in a finer product containing lower silica, which also cre
ates high bypass.
Identifying Bottlenecks. The objective was to increase plant throughput, there
fore, identification of circuit bottlenecks that limit capacity was crucial. Discussion
involving team members and control room operators identified the existence of three cri
teria that control the rod mill feed rate: (1) pumping capacity of hydrocyclone feed
pump; (2) processing capacity of fine screens; and (3) flotation concentrate silica. Mass
balanced data corresponded to the maximum limits for the first two items, which were
TABLE 1 Massbalanced data for major streams
Stream Flow Rate, tph Magnetic Iron, % Silica, %
Feed 350 25.8 45.5
Cobber concentrate 237 37.5 33.9
Ball mill discharge 1,536 48.0 23.9
Cyclone feed 1,454 50.6 21.3
Cyclone overflow 303 54.0 16.9
Hydroseparator underflow 290 56.4 14.6
Fine screen undersize 142 63.6 9.6
Magnetic concentrate 135 64.5 6.96
TABLE 2 Relative flow rates and magnetic iron recovery in tailing streams and magnetic
concentrate
Stream Flow Rate, % Recovery, %
Cobber tails 32.5 1.9
Rougher tails 23.4 1.3
Hydroseparator overflow 3.7 0.2
Finisher tails 1.9 0.4
Magnetic concentrate 38.6 96.2
100
90
80
70
60
50
40
30
20
10
0
P
a
r
t
i
t
i
o
n
C
o
e
f
f
i
c
i
e
n
t
,
%
100 1,000 10
Hydrocyclones
Fine Screens
FIGURE 2 Partition curves of existing hydrocyclones and fine screens. The curves represent the
actual operation with no correction for water bypass to underflow.
SIMULATIONBASED PERFORMANCE IMPROVEMENTS 153
1,450–1,500 tph at 45% solids and 300 tph (or 1,600 gpm) at 55% solids, respectively.
Flotation concentrate silica is primarily controlled by adjusting amine rates. However,
when this type of control fails to provide the desired level of silica, the rod mill feed rate
is reduced.
Alternatives for Improved Performance
Performance evaluation indicated that the ball mill grinding circuit was the major bottle
neck limiting plant throughput. In order to increase throughput, the circulating load
needed to be reduced. This could be achieved through improved grinding, hydrocyclone
classification, and/or, to a lesser degree, fine screening.
Several alternatives were considered for improved grinding efficiency. Existing elec
tric motors driving ball mills had ample power. This could create an opportunity to
increase power draw of the mills by increasing ball load in the mill and/or critical speed.
Analysis of ball mill data indicated that the existing makeup ball charge was too coarse.
Use of finer balls could increase the rate of fines production, thereby reducing circulat
ing loads. Another option for grinding efficiency was to increase feed percentsolids.
Increased feed percentsolids would increase retention time in the mill and result in
improved grinding efficiency. However, this was a variable that was difficult to control
because it required control of percentsolids in all streams feeding the ball mill (i.e., cob
ber concentrate, hydrocyclone underflow, and fine screen oversize). Nevertheless, plant
operators could try to keep feed percentsolids high, if substantial benefits could be
obtained by such a strategy.
For improved classification efficiency, the primary option was double hydrocyclon
ing (Figure 3), which implied a secondary separation of fines in the existing hydrocy
clone underflow by a secondary set of hydrocyclones. Later, several other alternatives
emerged: retrofitting existing 15in. cyclones to improve their efficiencies; use of 20in.
moreefficient cyclones; and replacing hydrocyclones with moreefficient sizeseparation
devices, known as stack sizers. One of the hydrocyclone manufacturers claimed that the
efficiency of the existing cyclones could be improved by retrofitting, which involved con
verting the existing constant angle conical part to two conical sections with two different
angles. It also was suggested that a new set of largerdiameter cyclones with two conical
parts could provide further improvements in terms of efficiency. A radical solution to the
inefficiency problem would be the use of stack sizers, which are essentially highcapacity
screens with durable screen panels (Valine and Wennen 2002). In recent years, they have
emerged as an alternative to hydrocyclones and are very efficient sizeseparation devices.
Other alternatives for improved efficiency were dry cobbing, separate grinding of fine
screen oversize, and fine screen feed dilution. Dry cobbing involves magnetic separation
of lowgrade, nearbarren material from the ore before it is fed to a plant. Typically, rod
mill feed would be treated by dry magnetic drums, and only the magnetic fraction would
TABLE 3 Silica content of size fractions in hydrocyclone underflow stream
Size, μm Weight, % Silica, %
150 42.4 36.9
105 15.2 23.5
75 14.2 12.2
53 12.6 5.0
38 5.5 4.1
25 2.5 3.0
–25 7.6 4.8
154 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
be fed to the plant (Wu 1997). This would not have a direct impact on the grinding cir
cuit, but it had a potential to increase concentrate production by eliminating a substan
tial portion of silicabearing particles from the plant feed. As a result, feed grade would
be higher and grinding energy would be better spent on particles that could easily be
beneficiated. As fine screen oversize is relatively fine material, it is not expected to go
through an efficient grinding and liberation process when it is circulated back to the ball
mill, which is designed for a much coarser feed. An alternative is to have a separate
grinding circuit for this stream (Figure 4). Vertical mills are successfully used in these
types of applications, and substantial improvements in throughput have been reported
(Benner 1998). Such a modification would directly reduce the load on the ball mill cir
cuit. Although it is known that diluting fine screen feed would increase the magnetic
concentrate silica, this could decrease the load on the ball mill by reducing the fine
screen oversize rate as a result of lower bypass and increased cut size. A small increase in
silica could be handled by the subsequent flotation process, if benefits prove to be rea
sonably high.
Modeling and Plant Simulation
An enhanced version of Usim Pac software was used for simulations. Improvements
included magnetic separator, hydroseparator, and fine screen models developed at the
CMRL (Ersayin 2003, 2004; Pletka 2004). For simulation of dry cobbing, data from a
pilotscale test were used to calculate rod mill feed characteristics for this option (Wu 1997).
For double cycloning, hydrocyclone retrofitting, and 20in. cyclones, expected perfor
mance and equipment data provided by a hydrocyclone vendor are used to modify
model parameters of the Plitt model available in Usim Pac (BRGM 2003). A similar
approach was used for stack sizer modeling; Derrick Corporation provided test data for
the screen mesh (0.15 mm) to be used in the study. Test data were converted to partition
curves for each component. These curves formed the mathematical basis for the simula
tions. For rod and ball mill modeling, Usim Pac uses a kinetic model combined with an
energetic approach. The model adjusts grinding rates in line with the variations in power
draw, which could arise due to changes in operating conditions (BRGM 2003). However,
this model does not have the capability to simulate the effect of makeup ball size. To
Rod Mill Feed
Ball Mill
Roughers
Primary
Cyclones
Secondary Cyclones
Hydroseparator
Fine Screens
Magnetic Concentrate
Finishers
Tails Thickener
Fine Tails
O/F
Coarse Tails
Cobbers
FIGURE 3 Modified flowsheet for the simulated alternative of double hydrocycloning
SIMULATIONBASED PERFORMANCE IMPROVEMENTS 155
overcome this deficiency, size distributions for different makeup ball sizes generated by
the JKTech ball mill model (NapierMunn et al. 1996) were used to devise a coupling for
such an effect. For the screen oversize grinding, a ball mill model with similar grinding
parameters as the (primary) ball mill was used. The objective of this particular simula
tion was to have a size distribution from this separate circuit similar to the magnetic cir
cuit. The number of hydrocyclones and their geometry were adjusted until the objective
was achieved.
For simulation purposes, it was assumed that the ore consisted of two components,
namely, magnetite and gangue. Massbalanced magnetic iron grades were converted to
magnetite on the basis of atomic weights, dividing by 0.7236. The remainder was con
sidered gangue. Eventually, empirical equations developed using massbalanced data
were employed to calculate silica in each stream after magnetite–ganguebased simulations
were performed.
As a first step, the current operation was simulated. Initially, the bestfit model
parameters for each unit were calculated individually. Then, model parameters were
finetuned to obtain a satisfactory fit between simulated and actual flow rates, grades,
and size distributions. Fine tuning was a major task around the ball mill to match operat
ing data with simulated data, due to a number of circulating streams. Finally, an excel
lent fit to all three types of data was obtained.
As a summary, the list of simulations carried out is presented:
Dry Cobbing
Hydrocyclone Efficiency Improvements
– Double Cycloning
– Retrofitting the Existing Cyclones
– 20in. Cyclones
– Stack Sizers Replacing Hydrocyclones
Ball Mill Efficiency Improvements
– Makeup Ball Size—1.75 and 1.5 in.
Rod Mill Feed
Ball Mill
Roughers Cyclones
Cyclones
Vertmill
Hydroseparator
Fine Screens
Magnetic Concentrate
Finishers
Tails Thickener
Fine Tails
O/F
Coarse Tails
Cobbers
FIGURE 4 Modified flowsheet for the simulated alternative of separate grinding of fine screen
oversize
156 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
– Increased Ball Charge
– Increased Critical Speed
– Feed PercentSolids
Fine Screen Feed Dilution
Fine Screen Oversize Grinding
For each alternative, complete plant simulations were carried out. To simplify compari
sons, existing operating conditions were kept constant for the rest of the plant. The effect
was then measured by three criteria: ball mill discharge rate, magnetic iron recovery,
and silica in magnetic concentrate.
Results of Simulations, Selection of MostFeasible Alternatives, and Plant Implementation
Results of the simulation study are summarized in Table 4. Several simulations were car
ried out to quantify the effects of increased ball charge, critical speed and feed percent
solids in ball milling, and feed dilution in fine screens, which had existing values of 34%,
0.667, 68.5%, and 56.5%, respectively. For these variables, only one set of simulation
results is presented in Table 4. They correspond to 38% ball charge, 0.75 critical speed,
72% solids in feed to ball mill, and 52% solids in feed to fine screens.
In general, all alternatives had a certain degree of potential to reduce the circulating
loads around the ball mill, thus creating room for increased throughput. For the dry cob
bing alternative, rod mill feed rate had to be reduced to 330 tph to maintain the existing
level of circulating load, due to a highergrade feed. As the bottleneck was in the ball milling
process, separation of silica gangue by dry cobbing did not substantially alleviate the loads
around the ball mill. Nevertheless, simulations showed that dry cobbing would increase
the rate of concentrate production as a result of increased feed grade to the rod mill.
Increasing ball mill power draw by increased ball load or critical speed would pro
duce similar benefit. However, increased ball load would require narrowing the diameter
of the discharge ring because the ball mill had a tendency to discharge balls when ball
charge exceeded 35%. The other option required replacement of the pinion shaft. Bene
fits of increased feed percentsolids would be relatively small, with the risk of increasing
viscosity beyond a point that could create grinding inefficiency. The most significant
benefits would be obtained by simply changing ball size from 2 in. to 1.5 in.
Of the three hydrocyclone efficiency improvement alternatives, double hydrocyclon
ing produced a benefit similar to retrofitting. It was found that 20in. cyclones would
produce moreefficient hydrocycloning. It would reduce the ball mill load appreciably
and create enough room for increased throughput.
Although stack sizers showed a very large decrease in the ball mill load, detailed
data indicated that downstream flow rates would be almost doubled even with a rod mill
feed rate of 350 tph. A cut size coarser than the existing meant higher downstream flow
rates. Implications were that this alternative requires not only the replacement of hydro
cyclones with stack sizers, but also doubling of the downstream equipment sizes.
Diluting fine screen feed generated relatively small benefits with increased silica in
the magnetic concentrate, indicating that it might be used as a relief when the circuit is
overloaded. Separate grinding of fine screen oversize appeared to have a large potential
to increase plant throughput.
Use of 1.5in. makeup balls appeared to be the most feasible alternative to imple
ment at the plant. It had a large potential to increase throughput. Although finer balls
were more expensive, benefits would easily pay for the additional cost. This alternative
also had several advantages: it did not require a capital investment; it would not increase
the power draw; no downstream problem was expected; and it was easy to implement.
SIMULATIONBASED PERFORMANCE IMPROVEMENTS 157
Consequently, this option was primarily selected for plant implementation. It was also
concluded that the current hydrocyclones should be replaced by moreefficient ones.
After the alternative was selected, a number of simulations were carried out to
examine the sensitivity of the modified plant to feed grade, feed size distributions, and
increased feed rate to flotation circuit. Simulation data indicated that a 10% increase in
throughput was an achievable target, even with moderate variations observed in feed
grade, feed rate, and coarseness of the feed.
Validation of Simulation Results
As a precautionary move, initially one of the lines was partially (30%) converted to
1.5in. makeup ball size. After running for a period of 4 months without any upsets in the
line, it was decided to carry out limited plant sampling to examine if such a modification
were producing expected benefits. Massbalanced flow rates and size distributions were
used to simulate the ball mill with a 30% 1.5in. makeup ball charge. A very good fit
between actual and simulated size distributions was observed (Figure 5). Bond tests
indicated that there has not been any significant change in ball mill grindability of the ore
despite a 3year difference between the two sampling periods. Encouraged by the
observed throughput increase in this particular line, it was decided to convert all three
lines at the plant to use 1.5in. makeup balls. Final plant sampling was carried out after
the plant had operated with 1.5in. makeup balls for approximately 1 year. During this
period, plant throughput measurements clearly indicated that more than a 10% increase
in throughput was achieved. This increase resulted in a decrease in energy used per ton
of concentrate production. However, it also became apparent that upstream unit opera
tion would have problems in supplying the increased ore demand by the plant. Unfortu
nately, plant sampling had to be carried out during such a period. The line was running
with a rod mill feed rate of 360 tph. Nevertheless, plant data could still validate the sim
ulation no matter what the feed rate was.
Plant sampling and sample analysis followed the same procedures as the initial sam
pling. Operating conditions were recorded and raw data were mass balanced. Plant sampling
conditions were simulated using model parameters determined from the original data.
The following data representing the new conditions were modified: rod mill feed rate,
rod mill feed size distribution, rod and ball mill charge levels, and hydrocyclone and fine
TABLE 4 Summary of simulation results
Performance Improvement Alternative
Ball Mill Discharge
Rate, tph
Magnetic Concentrate
Recovery, % Silica, %
Current 1,536 96.2 6.96
Dry cobbing 1,542 96.6 6.92
Double hydrocycloning 1,245 96.1 7.35
Hydrocyclone retrofit 1,296 96.1 7.31
20in. hydrocyclones 1,149 96.0 7.65
Stack sizers (0.15 mm) 1,039 96.2 7.24
1.75in. makeup balls 1,299 96.2 6.93
1.5in. makeup balls 1,085 96.2 7.00
Increased ball charge (38%) 1,395 96.3 6.99
Increased critical speed (0.75) 1,282 96.3 7.00
Ball mill feed % solids (72%) 1,386 96.3 6.99
Fine screen feed dilution (52%) 1,382 96.2 7.28
Separate grinding of fine screen oversize 835 96.2 6.53
158 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
screen feed percentsolids. Mill charge levels were not directly measured. Instead, they
were adjusted for the recorded power draws. As shown in Table 5, this simulation pro
vided a very good fit to massbalanced data in terms of flow rates, grades, and recovery,
thereby validating the findings of the original simulations.
Comparison of simulated and actual data was also carried out for individual pieces
of equipment. Only the magnetic separators showed unsatisfactory deviations from the
simulated data, probably due to the variations in liberation characteristics and/or the
way that separators are currently operated. There had been a 4year difference between
the two plant sampling surveys. Each piece of equipment was individually simulated and
simulated size distributions of major streams were compared to actual streams
(Figure 6). Particularly, the simulated size distribution of the ball mill discharge pro
vided an excellent fit to the actual distribution. Other simulated and actual parameters
corresponded well, all validating the simulation results.
CURRENT STATUS
Replacement of 15in. cyclones with moreefficient 20in. cyclones has been completed.
Optimization of plant operation with the new cyclones is being carried out. As pointed
out previously, after the modifications, several upstream problems became apparent.
Decisions already have been made to solve these problems. Replacement of existing cob
ber magnetic separators by highercapacity 4ft magnetic separators has been initiated to
enable the circuit to process increased throughputs. Modifications in the crushing plant
are being made. The capacity of mining equipment to deliver the increased ore demand
is being examined.
FUTURE PLANS
The next step is to develop a control strategy to optimize circuit efficiency. Steadystate
simulations will be used to develop such a strategy. Preliminary simulations prior to
hydrocyclone replacement showed that the simulator had excellent capability to mimic
circuit operation. However, the existing hydrocyclone model failed to simulate the newly
designed 20in. cyclones after their installation. Therefore, it became necessary to modify
100
80
60
40
20
0
100 1,000 10,000 10
Simulated
Actual
C
u
m
u
l
a
t
i
v
e
P
a
s
s
i
n
g
,
%
FIGURE 5 Ball mill discharge size distributions: Simulated versus actual after 30% 1.5in.
makeup ball charge
SIMULATIONBASED PERFORMANCE IMPROVEMENTS 159
the current hydrocyclone model to gain the capability of simulating these new hydrocy
clones. After such work is completed, a simulationbased control strategy will be devel
oped by carrying out a number of offline simulations. Eventually, this strategy will be
tested in the plant.
CONCLUSI ON
Simulation has been successfully used to improve performance of the grinding circuit at
the Ispat Inland plant. A large number of simulations were carried out to quantify the
potential benefits from performance improvement ideas. Eventually, it was found that
replacement of current 2in. makeup balls with 1.5in. balls would provide moreefficient
grinding and would increase plant throughput by more than 10%. This appeared to be
the most feasible option, and it was also easy to implement. Further improvements were
TABLE 5 Comparison of simulated and actual performance of magnetic circuit after
modifications were implemented at the plant
Performance Criteria Simulated Actual
Feed rate, tph 360 360
Ball mill discharge rate, tph 1,228 1,239
Hydrocyclone pressure, psi 20.2 20
Fine screen feed rate, gpm 1,363 1,351
Magnetic concentrate
80% passing size, μm 44 45
Magnetic iron, % 65.0 65.3
Recovery, % 96.0 95.2
100
80
90
70
50
30
10
60
40
20
0
C
u
m
u
l
a
t
i
v
e
P
a
s
s
i
n
g
,
%
100 1,000 10,000 100,000 10
Rod Mill Discharge: Simulated
Rod Mill Discharge: Actual
Ball Mill Discharge: Simulated
Ball Mill Discharge: Actual
Cyclone O/F: Simulated
Cyclone O/F: Actual
Magnetic Concentrate: Simulated
Magnetic Concentrate: Actual
FIGURE 6 Simulated and actual size distributions of major streams in the circuit after
implementation of 100% 1.5in. makeup ball size modification
160 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
expected by additional replacement of existing hydrocyclones with moreefficient ones.
As a result of simulationassisted modifications that have been implemented at the plant,
a throughput increase of more than 10% has been achieved. Other performance
improvement options are still being pursued.
ACKNOWLEDGMENTS
The authors would like to express their thanks to the U.S. Department of Energy, the Iron
Ore Cooperative Research Program of the Minnesota Department of Natural Resources,
and the Permanent University Trust Fund of the University of Minnesota for their fund
ing of this project.
REFERENCES
Benner, R.B. 1998. The Application of Vertimill to the Fine Grinding of Taconite. Technical
Report CMRL/TR9828. Coleraine, MN: Coleraine Minerals Research Laboratory,
University of Minnesota.
BRGM. 2003. Usim Pac 3.0: Unit Operation Model Guide. Orleans, France: BRGM.
Ersayin, S. 2003. Iron Ore Processing Improvements through Process Modeling and Computer
Simulation—2003. Technical Report CMRL/TR0308. Coleraine, MN: Coleraine
Minerals Research Laboratory, University of Minnesota.
———. 2004. Low intensity magnetic separator modelling: A pseudo liberation approach.
Mineral Processing and Extractive Metallurgy 113:C167–C174.
Ersayin, S., W.M. Bond, J. Arola, and B. Kettunen. 2005. Simulation of flotation feed pre
classification. In Proceedings of the Centenary of Flotation 2005 Symposium.
Brisbane, Australia: AUSIMM.
NapierMunn, T.J., S. Morrell, R.D. Morrison, and T. Kojovic. 1996. Page 413 in Mineral
Comminution Circuits: Their Operation and Optimization. Indooroopilly, Australia:
JKMRC.
Pletka, J. 2004. Development of a Mathematical Model for Fine Screening. Technical
Report NRRI/TR2004/15. Coleraine, MN: Coleraine Minerals Research Laboratory,
University of Minnesota.
Valine, S.B., and J.E. Wennen. 2002. Fine screening in mineral processing operations.
Pages 225–236 in Mineral Processing Plant Design, Practice, and Control. Edited by
A.L. Mular, D.N. Halbe, and D.J. Barratt. Littleton, CO: SME.
Wennen, J.E., W.J. Nordstrom, and D.L. Murr. 1995. National Steel Pellet Company’s
secondary grinding circuit modifications. Pages 19–25 in Comminution Practices.
Edited by S.K. Kawatra. Littleton, CO: SME.
Wu, C. 1997. Dry Cobbing Test on Rod Mill Feeds Inland Steel Minorca Mine Crushed Fine
Ores. Technical Report CMRL/TR9704. Coleraine, MN: Coleraine Minerals
Research Laboratory, University of Minnesota.
161
Determining Relevant Inputs for SAG
Mill Power Draw Modeling
Rajive Ganguli,
*
Sridhar Dutta,
†
and Sukumar Bandopadhyay
‡
ABSTRACT
Recurrent neural network models were developed to model semiautogenous grinding (SAG)
mill power draw. This paper demonstrates how sometimes seemingly relevant inputs have
no effect on the parameter being modeled. Initially, six inputs (SAG density, bearing pres
sure, revolutions per minute, noise, recycle, and feed rate) were used to model power draw.
Model predictions had a high R
2
(0.87), with 90% of the predictions being within 5%. How
ever, a closer inspection revealed certain deficiencies. When the number of inputs was
reduced to three (revolutions per minute, recycle, and feed rate), the prediction R
2
improved
to 0.91, and the number of predictions within 5% increased to 96.2%.
I NTRODUCTI ON
SAG mills are a major component within mining. Given their impact on important mining
parameters such as throughput and power consumption, it is not surprising that they
receive a lot of attention. In order to improve efficiencies, efforts are constantly being
made to understand and control them better (Austin 1990; Galan, Barton, and Romagnoli
2002; McCaffery, Katom, and Craven 2002).
PROBLEM
Effective control requires that the various factors that influence a process be identified
and their impact on the process quantified. Identifying factors that affect a process may
not be simple necessarily. Sometimes factors that should logically impact a process do
not have any significant bearing, thereby making them a poor choice as a control param
eter. Quantifying the impact of the various factors is even more difficult. Many choose to
use empirical equations to relate/obtain process values. Among those are Galan, Barton,
and Romagnoli (2002), who, for example, modeled the SAG mill power draw using the
following equation:
power draw (kW) = 25.9(W
E
+ W
L
) – 2064.5 (EQ 1)
where W
E
and W
L
represent the empty and loaded mill weights in tons.
* Associate Professor of Mining Engineering, University of Alaska Fairbanks
† Doctoral Candidate, University of Alaska Fairbanks
‡ Professor of Mining Engineering, University of Alaska Fairbanks
162 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
A possible complaint about Equation (1) is that it imposes a fixed linear relationship
between the mill power draw and mill weights. This may not be reflective of realtime
operating characteristics of a SAG mill. This is demonstrated in Figure 1, which plots the
relationship between the bearing pressure and horsepower for the SAG mill considered
in this paper. A SAG mill’s bearing pressure is dependent on the weight and, therefore,
Figure 1 plots a relationship between factors similar to those given in Equation (1). How
ever, as can be seen, within a short period of time (the two were plotted a few minutes
apart), the relationship between the two factors reverses, rather than remaining con
stant, as implied by Equation (1).
Additionally, the recurrent network model does not consider other factors, such as
rotation of the SAG mill, in determining the power draw. Yet, SAG rotation does affect
power draw. For example, after a certain point, when the rotation is increased, more
material inside the mill remains midair, thereby reducing power draw.
BACKGROUND TO THE CURRENT WORK
Many SAG mills operate at the limit of their power draws; therefore, this study was
designed around power draw (i.e., to understand the various factors that impact power
768.0 768.5 769.0 769.5 770.0 770.5 771.0 771.5 772.0
Bearing Pressure, psi
(a) Horsepower Decreasing with Bearing Pressure
13,700.0
13,680.0
13,660.0
13,640.0
13,620.0
13,600.0
13,580.0
13,560.0
13,540.0
13,520.0
H
o
r
s
e
p
o
w
e
r
770.0 770.5 771.0 771.5 772.0 772.5 773.0
Bearing Pressure, psi
(b) Horsepower Increasing with Bearing Pressure
13,950.0
13,900.0
13,850.0
13,800.0
13,750.0
13,700.0
13,650.0
H
o
r
s
e
p
o
w
e
r
y = –35.932x + 41263
R
2
= 0.8036
y = 58.849x – 31574
R
2
= 0.5622
FIGURE 1 Contrasting relationship between horsepower and bearing pressure of the SAG mill,
within minutes of each other
INPUTS FOR SAG MILL POWER DRAW MODELING 163
draw and to quantify their impact). The intent was to develop realtime models that
reflect the dynamic nature of SAG mills, with relationships among factors changing, such
as ore type and other operating parameters.
Some researchers have used expert systems for SAG control (McCaffery, Katom, and
Craven 2002). Essentially, rules (fuzzy or otherwise) based on human experiences are
tied to the controller, who applies them as situations change. As a limitation of this
approach, only relationships that are easily understood and captured by humans become
incorporated. Additionally, trends often are not caught early.
This paper builds a recurrent neural network (RNN) model (rather than imposes
rules) for horsepower based on the various factors that affect it. Initially, six relevant fac
tors to modeling the horsepower are used as inputs. Subsequently, an effort is made to
reduce the number of inputs. As the ultimate intent of the modeling exercise was to
determine the control parameters for horsepower, reducing the number of inputs (and
hence the number of control parameters) was worthwhile. Control parameters are those
factors which, when manipulated, impact the power draw (i.e., horsepower) in a known
way. This paper does not present work on horsepower control.
RECURRENT NEURAL NETWORK
RNNs are a type of neural network where outputs depend not only upon the current
inputs but also upon the previous inputs. This is unlike regular neural networks where
outputs depend only upon the current inputs. RNNs are preferred when process parame
ters have temporal relationships.
State–space concepts are often used in describing RNNs. The dynamic behavior of a
state–space model is described by the following equations:
x(n + 1) = M (x(n), u(n)) (EQ 2)
y(n) = C x(n) (EQ 3)
where x(n) is the current state vector at time n; u(n) is the input vector; y(n) is the
desired output vector; C is a matrix of weights characterizing the output layer; and M is a
nonlinear function. In an RNN (Figure 2), the hidden layer defines the state.
In Figure 2:
SAG MI LL MODELI NG
The SAG mill (at a surface gold mine) that was modeled is typically run at 2,000 st/hr,
with a peak power draw of 13 MW. Information from OSIsoft’s PI System (www.osisoft.com,
San Leandro, California) was used for the modeling. Table 1 shows a snapshot of the
SAGrelated data from the plant.
Initially, six parameters that affected the power draw were considered: SAG revolu
tions per minute, discharge end bearing pressure, SAG density (percent solids), noise,
recycle (tons per hour), and feed rate (tons per hour). The model, therefore, had six
inputs and one output, and 20,120 minutes of data (20,120 rows) was used for the mod
eling. The first 10,000 rows were used for training, while the next 5,000 rows were used
for calibration. The final 5,120 rows were used to test the model. Note that for an RNN,
selection for training, calibration, and prediction subsets cannot be random because the
x
x
1
x
2
x
3
= u
u
1
u
2
=
164 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
z
–1
z
–1
z
–1
z
–1
u
1
(n)
x
1
(n)
x
1
(n + 1)
x
2
(n + 1)
x
3
(n + 1)
x
2
(n)
x
3
(n)
u
2
(n)
Bias
z
–1
= Unit Delay
y(n)
FIGURE 2 A fully connected twoinput RNN, with two hidden neurons and one output neuron
TABLE 1 A snapshot of the realtime database (May 14, 2004)
Approx.
1min.
Intervals
SAG,
rpm
SAG
Discharge
End Bearing
Pressure,
psig
SAG
Density,
%solids
Average
Noise,
Knox units
Recycle,
tph
New Feed,
tph
SAG
Power,
hp
Pebble
Crusher
Motor
Current,
amps
SAG Ball
Charge,
% volume
Mill
Tonnage
Constraint
6:30 10.4 768.6 73.9 40.7 206.5 1848.8 13692.3 29.8 8.7 SAG
6:31 10.4 768.4 73.9 40.8 203.0 1843.5 13665.9 29.8 8.7 SAG
6:32 10.4 768.6 73.9 40.9 200.0 1907.3 13643.0 28.9 8.7 SAG
6:33 10.4 768.9 73.9 40.9 207.0 1833.7 13626.3 29.4 8.7 SAG
6:34 10.4 769.2 73.9 41.0 206.4 1827.7 13610.8 29.8 8.7 SAG
6:35 10.4 769.4 73.9 41.1 203.8 1850.9 13592.8 29.7 8.7 SAG
6:36 10.4 769.7 73.9 41.2 206.7 1804.6 13584.9 29.9 8.7 SAG
6:37 10.4 770.0 73.9 41.3 202.4 1844.8 13595.6 29.8 8.7 SAG
6:38 10.4 770.2 73.9 41.4 196.0 1865.3 13591.9 29.1 8.7 SAG
6:39 10.4 770.5 73.8 41.5 189.8 1841.5 13577.5 28.7 8.7 SAG
6:40 10.4 770.7 73.8 41.6 192.2 1843.1 13562.9 28.7 8.7 SAG
6:41 10.4 770.8 73.8 41.6 195.0 1865.0 13548.4 28.9 8.7 SAG
6:42 10.4 770.9 73.7 41.7 197.2 1841.7 13539.4 29.1 8.7 Ball mill
6:43 10.4 771.0 73.7 41.8 199.3 1823.8 13541.5 29.2 8.7 Ball mill
6:44 10.4 771.2 73.7 41.9 199.0 1779.2 13544.0 28.7 8.7 Ball mill
6:45 10.4 771.3 73.7 42.0 198.0 1852.1 13548.4 29.1 8.7 Ball mill
6:46 10.4 771.4 73.7 42.2 196.3 1862.8 13559.6 29.2 8.7 Ball mill
6:47 10.4 771.4 73.7 42.8 194.5 1781.0 13571.5 29.2 8.7 Ball mill
6:48 10.4 771.3 73.8 43.4 192.4 1848.2 13577.3 29.2 8.7 Ball mill
6:49 10.4 771.2 73.9 44.0 189.1 1769.8 13567.2 28.6 8.7 Ball mill
6:50 10.4 771.0 74.0 44.7 191.5 1823.4 13556.3 28.4 8.7 Ball mill
6:51 10.4 770.9 74.1 45.0 193.6 1764.7 13553.2 28.7 8.7 Ball mill
6:52 10.4 770.8 74.1 44.6 192.5 1766.8 13559.1 28.6 8.7 Ball mill
6:53 10.4 770.8 74.1 44.3 191.2 1791.7 13569.9 28.6 8.7 Ball mill
INPUTS FOR SAG MILL POWER DRAW MODELING 165
model has to be exposed to and depends upon the temporal relationships within the data.
All subsets have to be continuous in time domain. Quick stop training (Yu et al. 2004) was
followed to ensure network generalization.
RESULTS
Neuroshell (Ward Systems, Inc., Frederick, Maryland) was used for RNN modeling. The
R
2
for the prediction subset was 0.87, with 90% of the predictions being within 5% of the
true values. At first glance, the results seem impressive. However, a closer inspection
(Figure 3) highlights the deficiencies (lending credence to those who consider R
2
to be a
poor judge of performance).
The model shown in Figure 3 seems to be treading the middle in the first half of the
plot, and has difficulty adjusting when the horsepower changes to a different level (on
the righthand side). Additionally, it never entirely tracks the extreme values in horse
power. Therefore, the mean absolute error is relatively high, 348 hp (another model
deficiency). Model residual analysis reveals that the R
2
between the residuals (true
minus predicted) and true values is a somewhat high (0.23). However, this is simply an
artifact of the model not adjusting quickly to the swings in horsepower, especially on the
right side of plot (Figure 3), where it never quite caught up to the true horsepower.
Note that many RNN models fit the data and the model reported here is simply rep
resentative. The above performance could be thought of as the general performance of
RNNs on the data set.
Reduced Number of Inputs
The next part of the project explored if all the inputs were relevant to the model. Noise
and SAG density were removed as inputs either because their measurement quality or
their relationship to horsepower consumption was deemed questionable. The bearing
pressure is primarily dependent on feed rate, recycle, and SAG empty weight. The empty
weight of SAG is essentially constant, therefore, any trend in bearing pressure is entirely
dependent on feed rate and recycle. Bearing pressure becomes redundant once the feed
rate and recycle are used as inputs. Therefore, in the next phase, the model only had
three inputs (feed rate, recycle, and revolutions per minute). The rest of the modeling
procedure remained the same.
The R
2
of the prediction subset increased to 0.91, with 96.5% of the predictions
being within 5%. The mean absolute error was much better at 189 hp. The residual R
2
14,000
13,500
13,000
12,500
12,000
11,500
11,000
10,500
10,000
H
o
r
s
e
p
o
w
e
r
1
2
6
7
5
3
3
7
9
9
1
0
6
5
1
3
3
1
1
5
9
7
1
8
6
3
2
1
2
9
2
3
9
5
2
6
6
1
2
9
2
7
3
1
9
3
3
4
5
9
3
7
2
5
3
9
9
1
4
2
5
7
4
5
2
3
4
7
8
9
5
0
5
5
Time, min
True
Predicted
FIGURE 3 True versus predicted horsepower values for RNN modeled with six inputs
166 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
was lower (i.e., better) at 0.1. Thus, elimination of some inputs seemed to have only
improved the model.
Using the threeinput model shown in Figure 4, the R
2
for the entire data set
(including the prediction subset) was 0.97, with 96.2% of the predictions within 5%. The
mean absolute error was 188 hp. Figure 5 gives the line plot of predictions for the entire
data set. Further reduction in inputs hurt the model (not presented here).
From Figure 4, it is apparent that the threeinput model tracks the true horsepower
significantly better than the earlier model. However, as seen also in Figure 5, it fails to
adequately model the steady regions. For example, in the right half of Figure 4, while the
horsepower oscillates in the 13,500–14,000 range, the model predictions simply stay in
the middle, with changes in predictions being minor. Ultimately, the steadystate regions
are the most important.
DI SCUSSI ONS
The results indicate that predictability did not suffer when the number of inputs was
reduced. This could be due to
Lack of quality of certain inputs (not all sensors have robust first principles of
measurement)
The impact of certain inputs on the output being subtle compared to the scale of
observation
Interrelationship between inputs, deeming some of them irrelevant
Some inputs simply not having an impact on the output
Some inputs distracting the relationship between other inputs and the output;
this results in model improvement when the distracting inputs are removed
In any modeling exercise, parsimony of parameters, especially inputs, is desired
because it reduces the errors of estimation. When the goal of the modeling exercise is
control (such as here, even though power draw control is not presented here), reduction
in inputs implies reduction in control parameters. Control can only be effective if the
selected control parameters truly have an impact on the output.
In this paper, it became evident that despite many inputs initially deemed relevant,
only three were ultimately sufficient to model power draw. The failure of recurrent mod
els to accurately model the power draw (the high R
2
notwithstanding) is partly due to
the lack of all information necessary to model the power draw. For example, no informa
tion was present on ore hardness. Although determining hardness is difficult in real
time, an indirect measure of hardness, such as particle size analysis, could be used. Cam
eras (and image processing software) installed above the feed belt could provide this
information.
A criticism of the work presented here would be that the model was not adaptive.
The training subset (10,000 rows) consisted of almost 7 days of data. The final predic
tion subset was 3.5 days long. In other words, during the prediction phase, while the
SAG underwent various stages of operation, the model did not change. An improvement
may exist in incorporating very short term or adaptive modeling (spanning minutes
rather than days), where numerical models are automatically fit to a recent history of
process information. This may also solve the problem of the model not being able to pre
dict steady/oscillating horsepower.
INPUTS FOR SAG MILL POWER DRAW MODELING 167
CONCLUSI ONS
The power draw of a SAG mill was modeled using RNNs. Initially, power draw estima
tion was based on six inputs. It was found that when three of the inputs were eliminated,
power draw estimation did not suffer. In other words, fewer inputs—SAG density, noise,
and bearing pressure—were as adequate for evaluating power draw as more inputs (SAG
revolutions per minute, feed rate, and recycle).
RNN modeling yielded a high R
2
of prediction. However, closer inspection revealed
that they did not model the steady/oscillating regions very well. Possible improvements
to modeling power draw could include additional relevant data (especially on particle
size distribution) and adaptive modeling.
min
14,000
13,500
13,000
12,500
12,000
11,500
11,000
10,500
10,000
H
o
r
s
e
p
o
w
e
r
1
2
3
2
4
6
3
6
9
4
9
2
5
1
1
5
6
1
3
8
7
1
6
1
8
1
8
4
9
2
0
8
0
2
3
1
1
2
5
4
2
2
7
7
3
3
0
0
4
3
2
3
5
3
4
6
6
3
6
9
7
3
9
2
8
4
1
5
9
4
3
9
0
4
6
2
1
4
5
8
2
5
0
8
3
Time,
True
Predicted
FIGURE 4 True versus predicted horsepower values for RNN modeled with three inputs
min
10,000
H
o
r
s
e
p
o
w
e
r
1
Time,
True
Predicted
10,500
11,000
11,500
12,000
12,500
13,000
13,500
14,000
14,500
15,000
7
5
0
1
4
9
9
2
2
4
8
2
9
9
7
3
7
4
6
4
4
9
5
5
2
4
4
5
9
9
3
6
7
4
2
7
4
9
1
8
2
4
0
8
9
8
9
9
7
3
8
1
0
4
8
7
1
1
2
3
6
1
1
9
8
5
1
2
7
3
4
1
3
4
8
3
1
4
2
3
2
1
4
9
8
1
1
5
7
3
0
1
6
4
7
9
1
7
2
2
8
1
7
9
7
7
1
8
7
2
6
1
9
4
7
5
2
0
2
2
4
FIGURE 5 Plot comparing the true and predicted values for the final model for the entire data set
168 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
REFERENCES
Austin, L. 1990. A mill power equation for SAG mills. Minerals and Metallurgical
Processing (February): 57–62.
Galan, O., Barton, G.W., and Romagnoli, J.A. 2002. Robust control of a SAG mill. Powder
Technology 124:264–271.
Haykin, S. 1999. Neural Networks: A Comprehensive Foundation. 2nd edition. Englewood
Cliffs, NJ: Prentice Hall.
McCaffery, K.M., Katom, M., and Craven, J.W. 2002. Ongoing evolution of advanced
SAG mill control at Ok Tedi. Minerals and Metallurgical Processing (May): 72–80.
Yu, S., Ganguli, R., Walsh, D.E., Bandopadhyay, S., and Patil, S.L. 2004. Calibration of
online analyzers using neural networks. Mining Engineering 56(9):99–102.
169
Cement Clinker Grinding Practice
and Technology
Hakan Benzer,
*
Alex Jankovic,
†
and Levent Ergun
*
ABSTRACT
The current world consumption of cement is close to 2 billion tpa and is increasing by about
1% per annum. Conventional cement grinding circuits consist of twocompartment tube
mills and air separators. Alternative mills such as highpressure grinding rolls (HPRGs), vertical
roller mills, and Horomills have been applied in recent times to improve grinding efficiency.
Air separators play a crucial role in improving the overall energy efficiency of a cement
grinding circuit, and their design has been improving continuously over the decades.
The introduction of a clinker precrushing stage can significantly improve cement grind
ing energy efficiency. Due to the relatively low capital cost associated with installation of a
Barmac crusher, it is proving an attractive upgrade option. Hybrid grinding circuits with
HPGRs are being widely used, primarily to increase energy efficiency, with specific energy
consumption reduced to almost 50% compared to some conventional circuits.
I NTRODUCTI ON
The current world consumption of cement is close to 2 billion tpa. During the last 10 years,
cement production has increased by 38%. Different types of portland cement are manu
factured to meet different physical and chemical specifications. The American Society for
Testing and Materials (ASTM) has designated five types of portland cement, the charac
teristics of which are outlined in Table 1.
Portland cement is made from exact proportions of materials containing calcium,
silica, alumina, and iron. Approximately 1.5 t of raw materials are required to produce 1 t
of finished cement. Grinding is an important operation in the cement making process,
occurring at the beginning and end of the production cycle. The last stage in the process
of manufacturing portland cement is the finish grinding of clinker together with small
amounts of gypsum and some admixtures. The principal objectives of clinker grinding
are to promote the hydration of cement and to ensure complete coating of inert aggre
gates. The fineness of the cement affects the placeability, strength, and permeability of
the concrete properties. The finer the grind, the more reactive the finished cement.
Therefore, every type of cement must exhibit a particular degree of fineness to meet its
quality specification. In Figure 1, the particlesize distribution of the different cement
types are presented.
* Hacettepe University, Ankara, Turkey
† Metso Minerals Process Technology (Asia Pacific), Brisbane, Australia
170 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
The electrical energy consumed in the conventional cement making process is in the
order of 110 kWh/t, about 30% of which is used for raw materials preparation and 40%
of which is used during final cement production by cement clinker grinding. Figure 2
shows the consumption of electrical energy by the different processes in a typical cement
production plant (Fujimoto 1993). Minimizing production costs and increasing environ
mental concerns have emphasized the need to use less energy and therefore promoted
the development of moreenergyefficient machines for grinding and classification.
EQUI PMENT USED FOR CLI NKER GRI NDI NG
Tube Ball Mill
The continuous ball mill has been used for more than 100 years and is still the most
widely installed grinding equipment for this application. Cement is ground in tube ball
mills operating either in open or closed circuit. The tube mills are characterized by their
length/diameter (L/D) ratio with a ratio of 3 found to be best to minimize energy expen
diture (Schnatz and Knobloch 2000). The tube ball mills can be operated with one, two,
or three compartments, and the length of each compartment should be designed to
achieve optimum size distribution variation from feed to the discharge end.
Special diaphragms divide the cylinders of multicompartment mills. The dia
phragms are primarily designed to prevent loss of the balls to the next compartment
TABLE 1 Portland cement classification with its constituents and fineness
Types Clinker, % Admixture, % Minor Component, % Fineness +45 μm, %
CEM I 95–100 — 0–5 11.4
CEM II 80–94 6–20 0–5 14.2
CEM III 35–64 36–65 0–5 5.9
CEM IV 65–89 11–35 0–5 11.6
CEM V 40–64 18–30 0–5 17.0
CEM I
CEM II
CEM III
CEM IV
CEM V
0.001 0.01 0.1 1
Particle Size, mm
100
90
80
70
60
50
40
30
20
10
0
C
u
m
u
l
a
t
i
v
e
%
P
a
s
s
i
n
g
FIGURE 1 Size distribution of different cement types
CEMENT CLINKER GRINDING PRACTICE AND TECHNOLOGY 171
while allowing the flow of ground material through the mill. The design of the dia
phragm influences the fineness of the ground material (Duda 1985).
Various shapes of mill liners have been developed for cement mills (see Figure 3).
The classifying liners for clinker grinding have a specific design. This lining causes a clas
sification of the grinding ball sizes down the length of the mill. The grooved liner is usually
used in the second or third compartment of the cement mill to produce a cascading
motion which promotes abrasion breakage.
Operation of the tube ball mills is relatively well understood with several design and
operating parameters of the ball milling operation affecting the mill efficiency and the
quality of the cement produced (Gouda 1981).
Vertical Roller Mill
Vertical roller mills (VRMs) have been used for limestone and coal grinding in the
cement industry for many years due to their high drying capacity, low energy consump
tion, compactness, and reliability in operation. The largest mill in operation has an
installed power of 6 MW and grinds 840 t/h of lump feed down to 85% passing 90 Pm.
Cement grinding by a VRM has found applications in pregrinding systems, advanced pre
grinding systems, and finish grinding systems (Shimoide 1996).
Quarry Crushing and
PreHomogenization,
5%
Raw Material
Grinding, 24%
Feed Homogenization,
6%
Burning and
Cooling, 22%
Conveying, Packing,
and Loading, 5%
Finish Cement
Grinding, 38%
FIGURE 2 Energy consumption for different stages of cement production
FIGURE 3 Example of mill liners in the first and second compartment of a cement ball mill
172 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
In a VRM, the interparticle comminution takes place in a material filled gap between
the rotating table and the grinding rollers. The mill feed is charged to the center of the
table and moves, affected by centrifugal forces and friction, toward the table’s edge. On
its way, it is nipped by two, three, four, or six conical rollers installed at the outside rim
of the table. The rollers are attached to hydraulic cylinders that provide the grinding
force for comminution of the material. The ground particles leave with the airstream and
are taken up by the separator incorporated into the casing of the mill. The fine product
reports to the mill discharge, and the coarse reject of the separator falls back onto the
table as a recirculating load.
The VRM was first used in a commercial operation to finishgrind cement in 1984
(Shimoide 1996). Since then, however, further applications of this technology in the
industry have been relatively limited. One reason is that a portion of the power savings
achieved in the VRM (because of higher grinding efficiency) is lost due to additional
power consumption by the fan. In addition, the VRM suffers from roller wear problems.
Recent plant trials, however, have indicated that the problem can be reduced with new
roller designs. The wear rate and the throughput of the system depends very heavily on
the consistency of the materials being ground (Nobis 2001). Effective comminution
largely depends upon the formation of a stable grinding bed between the rollers and the
grinding table.
The main operational bottleneck of the VRM is its high circulating load from the sepa
rator back to the table. This causes inefficient grinding operation because of the high load
accumulation inside the mill. To overcome this problem, the roller mills can be operated
with external material circulation. It has been reported that the specific power consump
tion involved in producing portland cement with external material circulation was 30%
less than for producing these cements in tube mills (Feige 1981). The Kawasaki Inc. CKP
mill is an example of this type of machine and was developed based on the proven tech
nology of VRMs (Sutoh et al. 1992). In CKP systems, material is fed through a central
chute. A centrifugal force, produced by rotation of the table, distributes the product over
the table surface. After grinding, which is carried out between the table and rollers, the
material is extracted from the CKP by gravity with the assistance of scrapers (Miranda
et al. 1998). CKP mills are generally used as pregrinders, and the grinding energy effi
ciency of these mills as a pregrinder has resulted in grinding energy savings of 17%
(Dupuis and Rhin 2003).
Horizontal Roller Mill
The horizontal roller mill (Horomill) consists of a horizontal cylinder supported on slide
shoe bearings and driven through an open gear train. Terms used to describe the principles
of operation of a Horomill include a bed material compression mill, a multicompression
mill, and a highcapacity mill (Cornille 1999). A simplified diagram outlining the principles
of operation is shown in Figure 4.
The material passes into the mill at one end of the cylinder and, because of the cen
trifugal effect caused by operating the cylinder above the critical speed, is carried as a
uniformly distributed layer of material on its inner surface. The finished product is col
lected in a dust filter, while the coarse particles are recycled to the mill. The grinding
force is transmitted to the roller by hydraulic cylinders. Internal fittings are provided to
control the material recirculation. It’s been reported that the grinding process based on
multiple compressions gives the machine a high stability, and also the recirculating load
can be adjusted to suit the quality target (Cordonnier 1994).
Compared to a ball mill, the Horomill operates with a larger grinding bed thickness
and moderate pressures that lead to energy savings of 35% to 40% when used for
cement grinding. In operation, the specific costs related to the liner and wear parts are
CEMENT CLINKER GRINDING PRACTICE AND TECHNOLOGY 173
higher than in an equivalent ball mill (Brunelli 2001). Mechanical problems with a
Horomill have been reported in a Konya cement plant in Turkey (Fochardiere 1999).
HighPressure Grinding Roll
The highpressure grinding rolls (HPGRs) developed by Professor Schoenert have been
offered as a comminution technology with claims of improved performance when com
pared to conventional grinding technology. In particular, it has been claimed that the
HPRG has a lower specific energy consumption (Schoenert 1979).
The material to be ground in an HPGR is compressed in a gap between two counter
rotating grinding rolls (see Figure 5) with circumferential speed of 1 to 1.8 m/sec. The
product from the HPGR is a compacted cake that contains fine particles and coarser par
ticles with a large number of incipient cracks and weak points that greatly reduce the
energy expenditure during further comminution (Ellerbrock 1994).
An HPGR can be used at different stages in the cement grinding process: pre
crushed, finish grinding, hybrid grinding, and semifinish grinding. When an HPGR has
been used in the precrushed stage, 20% reductions in overall energy consumption have
been achieved (Kellerwessel 1996). Hybrid grinding involves splitting the coarse fraction
from the air classifier to the HPRGs and ball mill, respectively. In the semifinish grinding
application, the HPRG is operated in closed circuit with the air classifier, and the fines
from the separator are finally ground in a tube mill circuit. In the finishgrinding applica
tion, the HPRGs operate with an air classifier in closed circuit. Using this finishgrinding
configuration, the potential energy savings can be as high as 50% (Kellerwessel 1996), but
the water requirements in the subsequent mortar production process are significantly
higher due to the narrow size distribution produced (Roseman 1989; Odler and Chen 1995).
Air Classifier
Classification in the clinker grinding circuits is achieved using the air classifiers. Devel
opment of the air classifier was based on the operating principles of two devices, the sim
ple expansion chamber and the Mumford and Mood separator, patented in 1885
(Klumpar, Currier, and Ring 1986).
There are two types of air classifiers, dynamic and static. Static air classifiers are an
old technology without moving parts. Classification is achieved by changes in air velocity
and direction. The principle of operation is shown in Figure 6a. The airstream carrying
the particles is converted from a directional flow through the outer cone into a rotating flow
by guide vanes. The particles are subject to a centrifugal force—the coarse particles moving
to the outer wall of the inner cone and collected in a bin, while the fine particles leave
with the air and are sent to a dust collector. The product size can be altered to some
F
Horomill Tube
Scraper
Forward
Plate
Roller
Rotation of
Horomill Tube
Shoe Bearing Grinding Force
Material
Outlet
Material Inlet
F F
Inlet
Outlet
FIGURE 4 Comminution principle in a Horomill
174 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
extent by changing the angle of the vanes, but the efficiency is low and static classifiers
can be regarded more as grit separators than efficient classifiers.
Dynamic classifiers have both moving and fixed internal parts. The dynamic air clas
sifiers utilize a distribution plate to disperse the feed material into the separation zone.
Thus a particle of material is subjected to three forces: centrifugal force from the distri
bution plate, uplift from the air current, and gravity. Figure 6b indicates the forces acting
on a particle in a dynamic air classifier.
Dynamic classifiers have evolved through three generations, each being significantly
better than its predecessor. The firstgeneration classifier had a distributor plate, and the
air circulation in the classifier was provided by a vertically supported rotor. The main
problems with the firstgeneration classifiers were that the circulating air became very
hot, fine particles were not removed from the recycling air, and the control of the product
was very difficult. Figure 7a shows a simplified sketch of a firstgeneration air separator.
The secondgeneration classifier (see Figure 7b) is based on the same operating
principles as the first but an external fan is used to circulate the air and a cyclone is used
Fixed Roll
Feed
Moveable Roll Oil Cylinders
Product
Nitrogen cylinder
FIGURE 5 The principle of operation of HPGRs
Adjustable
Blades
Immersion Tube
Fines
Tailings
(a) (b)
Particle Feed
Direction of Rotation
To Coarse
Particle Cone
To FineParticle Chamber
f
C
f
D
f
G
R
T
FIGURE 6 (a) Schematic of the static air classifier; and (b) separation mechanism in a dynamic
air classifier
CEMENT CLINKER GRINDING PRACTICE AND TECHNOLOGY 175
to remove fine particles. Greater product control is possible due to the ability to adjust
the rotor speed and air velocity separately.
The thirdgeneration separators are highly efficient separator devices (see Figure 8).
The feed material to the separator is delivered as a dispersed curtain of particles, and the
horizontal air flow to the separator results in uniform separation performance across
the unit. The fine particles pass through a rotating cage before going to the fine product.
The bars of the cage assist in the performance of the separator.
CI RCUI T CONFI GURATI ON FOR I MPROVED ENERGY EFFI CI ENCY
For most of the twentieth century, the common drygrinding circuits for the production
of finished cement from cement clinker consisted of twocompartment tube mills with or
without the air separators. The advantage of this circuit is its simplicity and ease of oper
ation; however, the energy consumption is high, especially for opencircuit operation.
One of the reasons that the twocompartment tube mill circuit has limited energy effi
ciency is due to the high reduction ratio that must be achieved in the single comminution/
classification step. Clinker feed size can vary from F
80
= 10–40 mm and the final product
size from P
80
= 35–40 ȝm with the size reduction ratio being in the order of 250–1,000.
Large balls (up to 100 mm) are required in the first compartment of the tube mill to
crush the coarse clinker. Ball mill grinding efficiency for feed sizes larger than F
80
= 2–3 mm
is particularly poor, and it should therefore be more energy efficient to precrush the clinker.
Recent work indicates that introduction of the Barmac crusher for clinker precrushing
can increase the cement circuit throughput on the order of 10%–20%. Alternatively, the
total energy consumption of the circuit can be reduced on the order of 5%–10% (Jankovic,
Valery, and Davis 2004). This is an attractive upgrade option due to the relatively low
capital investment involved in the installation of a Barmac crusher.
Clinker precrushing can be carried out with a variety of different crushers. Figure 9
shows the product size distributions from a Barmac and, alternatively, a highperformance
(HP) cone crusher in closed circuit with a 4.75mm screen at 2.3 kWh/t specific energy
input. Although the 80% passing size for the HP cone crusher is finer, the Barmac product
is potentially more favorable due to its higher content of fines. This advantage, however, is
Fresh Air
Exhaust
Feed
Cyclone
Fines
Rejects
Return Air Vanes
Recycling Fan
Distribution Table
a
b
c
d
e
F
G
A
(a) (b)
Closed
Circuiting
FIGURE 7 (a) Firstgeneration dynamic air separator; and (b) secondgeneration dynamic air
separator
176 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
not crucial for the selection, as the clinker feed size, hardness, and abrasivity, as well as
the required capacity, will have an effect on the particular crusher best suited to the par
ticular application.
In order to obtain the most efficient breakage in the first compartment of the ball
mill after introduction of the precrushing stage, the ball size distribution should be
changed to suit the new particlesize distribution of the material fed to the mill. An example
of the measured particlesize distribution of the combined ball mill feed (new feed + 150%
recycle) with raw and precrushed clinker is shown in Figure 10. A significant fraction of
the material in the feed containing the raw clinker is coarser than 5 mm. To effectively
Shaft
Fines Plus Air
Blade
Vane
Feed Plus
Air Inlet
Spoke
Coarse
Tertiary Air
Volute
Annular Space
Plate
Section
FIGURE 8 Schematic of a thirdgeneration dynamic air separator
0.1 0.01 1 10 100
0
10
20
30
40
50
60
70
80
90
100
HP Cone, 2.3 kWh/t, Product
Barmac, 2.3 kWh/t, Product
HP Cone Feed
Barmac Feed
Size, mm
C
u
m
u
l
a
t
i
v
e
%
P
a
s
s
i
n
g
FIGURE 9 Product size distribution from the closed Barmac and HP cone crusher circuit
CEMENT CLINKER GRINDING PRACTICE AND TECHNOLOGY 177
grind this sized feed, the calculated top ball size required (using the Bond formula)
would be 90–100 mm. For precrushed feed, the top ball size would be 35–40 mm due to the
absence of coarse particles.
With precrushed feed, the optimum ratio in length between the first and second
compartment would also be affected. Design of the transfer grate, mill liners, and sweep
air velocity should also be reviewed to suit the new reduced ball size and provide effi
cient removal of fine particles.
In the last 20 years, HPGRs have been used extensively in cement grinding circuits
due, primarily, to their higher grinding efficiency compared to the conventional two
compartment tube mills. HPGRs can be used for precrushing, finish grinding, hybrid grinding,
and semifinish grinding. Table 2 shows the energy consumption of five cementgrinding
circuits employing HPGR units in different applications (Aydo÷an, Ergün, and Benzer
2004). It can be observed that the overall circuit specific energy consumption decreases
when a large portion of the size reduction (higher HPGR kWh/t) is performed by the
HPGR. Circuits that employ HPGR mills can achieve in excess of 40% improvements in
grinding energy efficiency, providing that the circuit is optimized and automated process
control is employed.
In order to assess the performance of a particular cementgrinding circuit and to
compare efficiency of different circuit configurations, complete audits are required. The
audit includes monitoring and sampling of different circuit streams during steadystate
operation, as well as mill inspection and sampling after a crashstop. Based on informa
tion obtained from the audit, a mass balance can be carried out to determine material
Particle Size,
mm
37.5
25
19
13.7
9.5
4.75
2.36
1.18
0.6
0.3
0.15
0.075
0.053
0.038
0.032
0.025
0.01
0
Optimum
Ball Size,
mm
142
116
101
90
71
50
36
25
18
13
9
6
/
/
/
/
/
/
Raw Clinker,
% Retained
0.20
0.53
3.71
6.30
4.15
10.28
7.50
2.86
0.94
4.63
3.821
13.92
9.82
8.17
3.41
3.97
8.27
7.49
Precrushed
Clinker,
% Retained
0
0
0
0
0
0.40
9.65
11.26
4.74
4.40
6.83
16.12
9.89
8.63
3.84
4.60
9.53
10.10
0.1 0.01 1 10 100
18
16
14
12
10
8
6
4
2
0
%
R
e
t
a
i
n
e
d
Size, mm
Raw Clinker
Precrushed Clinker
FIGURE 10 Combined ball mill feedsize distribution when processing raw and precrushed clinker
TABLE 2 Specific energy consumption in different cement grinding circuits utilizing HPGRs
Cement Grinding Circuit Description
HPGR Specific Energy
Consumption, kWh/t
Circuit Overall
Specific Energy
Consumption, kWh/t
Opencircuit HPGR, closedcircuit ball mill 4.05 34.2
Opencircuit HPGR with partial recycling, closedcircuit
ball mill
8.9 29.6
Hybrid grinding — 29.9
Closedcircuit HPGR, closedcircuit ball mill 8.0 21.7
Semifinish grinding 9.8 23.0
178 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
flows (solids and air) around the circuit. Using this information, the performance of the
circuit and individual pieces of equipment can be assessed and potential bottlenecks
identified. To assist with circuit optimization, site and equipmentspecific models are cal
ibrated based on the results from the audit. Models can be then used to simulate differ
ent operating conditions and circuit scenarios (Benzer et al. 2001, 2003).
CONCLUSI ON
The current world consumption of cement is close to 2 billion tpa and is increasing at
about 1% per annum. The electrical energy consumed in the conventional cementmaking
process is approximately 110 kWh/t, and around 40% of this energy is consumed for
clinker grinding.
For most of the twentieth century, the drygrinding circuits for the production of fin
ished cement from cement clinker consisted of twocompartment tube mills and air separa
tors. Alternative mills such as the HPGRs, VRMs, and the Horomill have been applied in
recent times to improve the grinding efficiency. Significant energy savings are reported
in applications that utilize these mills, the HPGR being the most widely used. During this
period, the design of the air separators has evolved from the very inefficient static sepa
rators to the highly efficient dynamic separators. These separators play a crucial role in
improving the overall energy efficiency of the cementgrinding circuits.
The increasing demand for “finer cement” products, and the need for a reduction in
energy consumption and greenhouse gas emissions, increases the importance of grind
ing optimization. In the last two decades, significant progress has been achieved through
improved equipment design and the use of new circuit configurations. Introduction of a
clinker precrushing stage can significantly improve the energy efficiency. Barmac crusher
installation, due to its relatively low capital cost, is an attractive upgrade option. Hybrid
grinding circuits incorporating HPGR units are being used widely, primarily to achieve
higher energy efficiency, with specific energy consumption reduced by almost 50% com
pared to that achieved in some conventional circuits.
In order to optimize a grinding circuit, a detailed knowledge of circuit operation is
required. Modeling and simulation techniques can be effectively utilized to assist in this
process optimization exercise.
BI BLI OGRAPHY
Aydo÷an, N., Ergün, ù.L., and Benzer, H. 2004. HPGR applications in the cement industry.
Pages 33–48 in JKMRC International Student Conference 2004, 6–7 September. Bris
bane, Australia: JKMRC.
Benzer, H., Ergün, L., Lynch, A.J., and Öner, M. 2003. Case studies of models of tube mill
and air separator grinding circuits. Pages 1524–1533 in Proceedings: XXII Interna
tional Mineral Processing Congress. Edited by L. Lorenzen and D.J. Bradshaw. Cape
Town, South Africa: African Institute of Mining & Metallurgy.
Benzer, H., Ergün, L., Oner, M., and Lynch, A.J. 2001. Simulation of open circuit clinker
grinding. Minerals Engineering 14(7):701–710.
Brunelli, G. 2001. A proven partnership. International Cement Review (February): 37–40.
Buzzi, S. 1997. The Horomill. ZKG International 3:127–138.
Cordonnier, A. 1994. The Horomill—a new finish grinding mill. ZKG 11:643–647.
Cornille, J.P. 1999. Pages 21.1–21.2 in Horomill: Latest Developments and Results. Euro
pean Cement Conference Proceedings. Surrey, UK: Pro Publications International.
CEMENT CLINKER GRINDING PRACTICE AND TECHNOLOGY 179
Duda, W.H. 1985. Cement Data Book—International Process Engineering in the Cement
Industry. 3rd edition. Bauwerlag: GMBH.
Dupuis, J., and Rhin, C. 2003. Increased grinding capacity at R.A.K. World Cement (Feb
ruary): 79–83.
Ellerbrock, H.G. 1994. High pressure grinding rolls. ZKG 4:1047–1100.
Feige, F. 1981. Cement grinding in a roller mill with external material circulation. ZKG
11(81):560–562.
Fochardiere, R. 1999. Horomill: One year’s operating experience. World Cement (Sep
tember): 98–104.
Fujimoto, S. 1993. Reducing specific power usage in cement plants. World Cement 7:25–35.
Gouda, G.R. 1981. Technical aspects of comminution in the cement industry. Part 1.
World Cement Technology (April): 112–122.
Jankovic, A., Valery, W. Jr., and Davis, E. 2004. Cement grinding optimisation. Minerals
Engineering Journal 17(11–12).
Kellerwessel, H.A.M. 1996. High pressure particle bed comminution. State of the art, appli
cation, recent developments. Engineering and Mining Journal (February): 45–52.
Klumpar, I., Currier, F., and Ring, T.A. 1986. Air classifiers. Chemical Engineering
(March): 77–92.
Marchal, G. 1995. FCB breaks into Asian market with Horomill. World Cement (Septem
ber): 23–25.
Miranda, R.F., Minas, I., Yamana, W.T., Pirapora, S., Cimentos, V., and Tete, P. 1998. Brazilian
progress in grinding. World Cement (May): 40–42.
Namik, A.A., Levent, E., and Benzer, H. 2004. High pressure grinding rolls (HPGR) appli
cation in the cement industry. Presented at JKMRC International Conference, Brisbane,
Australia.
Nobis, E. 2001. Experience with grinding slag and clinker in a Loesche mill. ZKG
54(4):196–204.
Odler, I., and Chen, Y. 1995. Influence of the method of comminution on the properties
of the cement. ZKG 48(9):496–500.
Roseman, H. 1989. Investigations on a high pressure grinding roll mill used for cement
grinding. ZKG 42(6):142–144.
Schnatz, R., and Knobloch, O. 2000. Influence of the Ball Filling Factor on the Power
Consumption and Throughput of Ball Mills in Combined Grinding Plants. ZKG
8(53):438.
Schoenert, K. 1979. Verfahren zur Fein und Feinstzerkleinerung von Materialien sproden
Stoffverhaltens. German Patent DP 2708053.
Shimoide, K. 1996. Cement grinding by vertical roller mill. World Cement (September):
68–74.
Sutoh, K., Murata, M., Hashimoto, S., Hashimoto, I., Sawamura, S., and Ueda, H. 1992.
Current report on preliminary grinding of clinker and raw material using the CKP
system. ZKG 3:94–96.
181
Extended Semiautogenous Milling:
Smooth Operations and Extended
Availability at C.M. Doña Ines de
Collahuasi SCM, Chile
F. Romero,
*
L. Yacher,
†
and O.A. Bascur
‡
ABSTRACT
Comminution circuits represent one of the largest operating costs in mineral processing. The
current stateoftheart plant information systems enable the use of a wide range of supervi
sion techniques. The use of a realtime plant information system has derived major economic
benefits. Semiautogenous milling operations are supervised in real time for any substandard
or abnormal conditions. Furthermore, the use of advanced statistical technologies allows for
an increased potential of plant performance. The use of semiautogenous milling operations
have been made easier, and major economic benefits can be obtained, by extending mill
operating time and reducing maintenance costs.
I NTRODUCTI ON
Comminution circuits represent one of the largest operating costs in mineral processing
plants. These circuits are highly multivariable and their analysis and optimization are diffi
cult. Compañia Minera Doña Ines de Collahuasi SCM, Chile, has implemented a Real
Time Performance Management (RtPM) system powered by OSIsoft’s PI System to integrate
and capture data and events of their processes and metallurgical information. This RtPM
system is used to extract, analyze, provide context, distribute, and display information.
Operators and managers can act quickly and with confidence in their decisionmaking.
The RtPM infrastructure provides an environment for transforming realtime and
historical information into action. Under RtPM integration infrastructure, many valuable
applications can be designed by customers and third parties (Bascur and Kennedy 2000,
2004). This paper presents an example of adding an advanced statistical analysis tool to
a basic platform.
The operation of semiautogenous mill grinding circuits is multivariable and dynamic.
The process variables involved in mineral processing plants show strong interaction. It is
difficult to identify the cause and effect between process variables and ore mineralogy.
There are nonlinearities and delays in the interaction of many different process variables,
equipment wear characteristics, slurry rheology, mineral composition, and operator behavior.
* C.M. Doña Ines de Collahuasi SCM, Iquique, Chile
† Contac Ingenieros LTDA., Santiago, Chile
‡ OSIsoft, Inc., Houston, Texas
182 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
Analysis of the process history allows for defining critical process patterns (models)
that can be put to work online, using the RtPM calculation engine. Specific models can
be developed to help operations with early alerts of pattern deviations, estimation of
projected process variables, and their influence on specific targets, among others. The
RtPM tools simplify the implementation of continuous improvement strategies.
It can be challenging to estimate the liner wear in large machines. Both increasing
the availability of the equipment and improving metallurgical performance are key for
optimal operations.
Mineral composition plays a very important role in overall metallurgical performance,
flotation recovery and thickener, and pipeline performance. Changes in specific gravity
of the ore can increase viscosity, and these changes affect the hydrodynamic conditions of
the process.
To deal with this complex scenario, advanced multivariable statistical techniques
have demonstrated a great application potential for characterizing variable relationships
and the development of performance models in operations.
This paper highlights some of the findings in applying multivariate analysis to semi
autogenous grinding (SAG) mill operations. It will discuss the basic findings in predicting
liner wear and detection, the effect of stockpile in metallurgical performance, and the
effect of iron content in feed.
RESULTS FOR PATTERN RECOGNI TI ON I N SAG MI LL OPERATI ON
To analyze the behavior of complex systems, one can add known facts to the data, such
as complying with mass balance constraints. This is a typical use of data reconciliation to
identify how close the data are to satisfying mass balance constraints. Romero, Suarez,
and Bascur (2004) have presented such a case (Bascur and Linares 2005).
When no structure is available to relate the patterns of a complex system, principal
component analysis (PCA) is now a wellknown method for multivariate data analysis.
PCA has been discussed by several authors in the mineral processing and metals industry
(MacGregor et al. 1991; MacGregor and Kourti 1998; Hodouin et al. 1993; Dudzic 1998;
Vaculik and Smyth 2003; Romero, Orchard, and Yacher 2003).
The multivariable statistical analyses use two basic concepts. One is the variability
factor analysis or “VFA,” also known as PCA. Information technology evaluates the pre
dominant sources of variability in a data set. The procedure consists of a data reduction
of the original dataset space by selecting the vectors where the variability is maximized.
Thus, it is possible to represent the process by only analyzing these new projections
structures called variability factors (VFs), a reduced set of calculated variables that rep
resent the variability of the operations (Rodriguez and Tobias 2001).
The second most important concept is the projection to latent structures (PLS), a
valuable tool for model identification around an operating point. PLS generates latent
vectors for the statistically normalized input and output variables for which simplified
output versus input models can be generated for realtime relationships evaluation.
PLS models can be used in several ways; the most common are variable estimators
and soft sensors. Recent works (mostly in machine surveillance) have demonstrated
potential for early detection of pattern deviations (a pattern being defined as “normal,”
“desired operating region,” among others, depending upon the model), and for the
offline and online evaluation of the relative influence of model variables.
In the case of SAG mill circuits, the VFA approach detects time instances when the
SAG mill’s operation adjusts to a recommended performance, generating an operation
pattern for the process. The process is captured by the PI System and analyzed using PI
DataLink (OSIsoft). Afterward, the pattern is analyzed with the statistical tests included
EXTENDED SEMIAUTOGENOUS MILLING 183
in SCAN software (Contac Ingenieros, L. Yacher). SCAN is used to obtain the reduction
in the representation of the data and evaluates the contributions of the data. It also pro
vides the tools to develop a predictive model and the necessary code to run it online in
OSIsoft’s PI ACE (advanced computing engine). As a result, the main sources for process
variability are identified, and the principal components for the data set are determined
(Figure 1).
Figure 2 shows a typical set of realtime information that is used to further analyze
the data to find structure or relationships for detecting pattern changes based on normal
condition model characterizations. PI DataLink is used to extract filtered data to con
struct the basic data set.
Figure 2 shows the offline data analysis using Excel with PI Datalink and SCAN add
ins. All the process variables are extracted to the spreadsheet for further multivariable
analysis using SCAN offline. As the process consists of two mill lines, first a comparison
between each line performance is needed to define a single normal operation pattern,
which is shown in Figure 3. From Figures 3 and 4, it is clear that each SAG mill (operat
ing at the same conditions) was affected by an unmeasured disturbance.
A conclusion of the analysis, taking into account the maintenance records, the pro
cess history, and the operator’s experience, estimates the unmeasured variable affecting
Pebbles
Crusher (2)
To Stockpile
From Stockpile
SAG Mill (2)
32 ft × 15 ft,
8 [MW]
Ball Mill (2)
22 ft × 36 ft,
9.7 [MW]
Hydrocyclones
Battery (4)
12xDS26
To Flotation
Hydrocyclones
Pump (4)
Sump (2)
FIGURE 1 SAG mill circuit process diagram
FIGURE 2 Raw data set in Excel using PI Datalink and SCAN addins
184 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
20
15
10
5
0
–5
–10
–15
–20
–20 –15 –10 –5 0 5 10 15 20
VF1
V
F
2
FIGURE 3 Influences of liner wear in SAG mill 1 data variance
8
6
4
2
0
–2
–4
–6
–8
–10 –8 –6 –4 –2 0 2 4 6 8 10
VF3
V
F
4
FIGURE 4 Influence of liner wear in SAG mill 2 data variance
EXTENDED SEMIAUTOGENOUS MILLING 185
the SAG mill’s performance was the liner wear. The final analysis considered the exist
ence of two main operation patterns, one of them when the SAG mill’s liners were new
and the other when they were worn. The process variables included in the analysis are
ore and water feed, mill speed, motor power (two motors for each SAG mill), pressures
(feed and discharge), and pebble mass flow.
SAG Mill Variability Patterns (New Liners)
From the VFA loading maps (Hodouin et al. 1993), the following conclusions can be stated:
The main sources of variability are the mill speed, bearing oil pressures, and the
motor’s power.
Ore feed, water feed, and pebble flow give additional and complementary infor
mation about the mill’s operation.
There is more residual variability in the discharge pressure than in the feed pressure.
A powerful correlation is seen between the ore and water feed loops.
Any unusual increment of variability in the mill speed or in pebble flow can be
considered an important disturbance.
SAG Mill Variability Patterns (Worn Liners)
Four variability factors need adequate supervision for each SAG mill. The factors are
approximately 97% of the total dispersion of the pattern data set. In this case, only 3% of
the remaining variance is related to the typical noise of the process.
Ore feed, water feed, and pebble flow still give additional and complementary
information about the mill operation.
There is more residual variability in pebble flow than in the previous condition.
There is less correlation between the ore and water feed loops than that appreci
ated in the previous condition.
There exists less residual variance in the mill’s power.
As in the previous case, the adequate variability range for both SAG mill lines 1 and
2 have been characterized through the same control ellipses. The axis of these ellipses
has been calculated to obtain a 95% and 99% of confidence for normal operation, as
long as the scores are inside them (Figure 5).
The created geometric areas inside the score map for the mentioned VFs ease the
visualization of the actual mill’s behavior and the earlier detection of abnormalities
according to the position of the score cloud versus time. Figure 6 shows the regions
which have been identified from the data analysis to detect when liners become ineffec
tive. The online identification of this state has importance for the online and offline pro
cess performance monitoring and analysis of the grinding circuit. As such, the circuit has
been able to extend the availability of the circuit for more than 2 months.
After the inferential model is defined, an online calculation can be programmed
using PI ACE for online alert detection of the state of the SAG mill circuits. This new
inferred variable or soft sensor can be used for additional knowledge monitoring of the
system.
Figure 7 shows a simplified data flow as the PI System collects the data, the analysis
step used to identify the process patterns, and generation of the online calculations using
SCAN. SCAN provides the necessary contribution plots for analysis of the data. Online
calculations are processed by the PI ACE to generate soft sensors and alerts to avoid deteri
oration of the SAG mill equipment and also prevent overloading the mill.
186 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
Figure 8 shows a PI ProcessBook with the weight of the variables as it is calculated
online by the SCANinferred multivariable data set.
Several additional historical data analyses have been done to evaluate the effect
other disturbances in the SAG mill circuits.
Example: Iron Content in SAG Mill Feed
The analysis of the influence of the size distribution and the effect of iron content on the
SAG mill total feed was made. The PLS model used the different size fraction as inputs to
check the effect on total tonnage. The results show the model weight factors are shifted
from low iron content to higher iron content. This is due to the change in specific gravity
FIGURE 5 Variability factor results
–25 –20 –15 –10 –5 0 5 10 15 20 25
20
15
10
5
0
–5
–10
–15
–20
Pressure’s
Increment
Possible Typical
Overwhelm
Power and Pressure
Increments
Pressure’s
Decrement
Strong Power’s
Increment
Strong Power’s
Decrement
VF1
V
F
3
FIGURE 6 SAG mill’s nonsecure operation areas
EXTENDED SEMIAUTOGENOUS MILLING 187
PI Visualization
Tools
Process
Models
Calculation
Engine
Embedding
Offline
Analysis
Online
Analysis
SCAN
PI System
FIGURE 7 Data analysis flow of offline and online processes
FIGURE 8 PI ProcessBook with the derived multivariate weights of all mills
188 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
of the ore, which changes the rheology of the pulp. The slurry transport in the mill is
affected and the tonnage is reduced. Figure 9 shows a shift of the weight factors reveal
ing the effect of the iron content in the size distribution.
Example: Effect of Stockpile Level
The stockpile minimum level has been a source of discussion for some time. In the test
plant, the typical limits are 100,000 to 600,000 t. A long period of SAG mill operational
time was analyzed and a PLS model was developed, as shown in Figure 10. The model
weight for stockpile was particularly high, and a detailed analysis of the process PI his
tory showed a decay of the plant throughput whenever the stockpile was below 150,000 t.
There is a penalty of 12 t in reduction of feed due to segregation of the material when
the stockpile is below 150,000 t. It was advised to maintain a minimum level to increase
overall production and minimize SAG mill disturbances due to changes in feed size dis
tribution. Figure 10 shows the PLS model for SAG mill behavior.
Up to this point, the stockpile level influence was “embedded” in the rest of the pro
cess variable values. The ability to isolate individual influences was key to obtaining
these results. As a conclusion of the study, a process operation policy was made to reflect
this restriction.
(a) (b)
< 0.25
0.25–0.38
0.38–0.5 0.75–1.0
0.5–0.75 1.0–1.5 2.0–2.5
2.5–3.0
3.0–4.0
4.0–8.0
< 0.25
0.25–0.38
0.38–0.5 0.75–1.0
0.5–0.75 1.0–1.5
1.5–2.0
2.0–2.5
2.5–3.0
4.0–8.0
Desplazamiento de distribucion
FIGURE 9 Contributions plots of (a) low and (b) high iron content at several size fractions
26 36 46 56 66 76 86 96 106 116 126 136 146 156 166 176
3,680.5
3,480.5
3,280.5
3,080.5
2,880.5
2,680.5
2,480.5
2,280.5
2,080.5
1,880.5
1,680.5
Predicted Data
Real Data
Time
FIGURE 10 PLS model for SAG mill behavior
EXTENDED SEMIAUTOGENOUS MILLING 189
FUTURE APPLI CATI ONS
The obtained results for the multivariate SAG mill analysis promoted further investiga
tion for other process areas where statistical testing and modeling procedures available
in SCAN software can be applied.
A broader application includes the consideration of the flotation plant where
dynamic linear and nonlinear modeling modules are applied in order to obtain a predic
tion for the copper concentrate grade as a function of process measurable variables. A
close adjustment of the model allows for the online estimation of the individual
“weights” or relative influences of each of the manipulated variables, giving the operator
a guideline for the most effective ones at given times. The analysis is based on latent
structures in order to reduce model complexity. Thus, a NIPALS (Nonlinear Iterative Par
tial Least Squares) algorithm is used to find the structures of the process variable matrix
X, which are more correlated with the actual concentrate grade.
However, inner linear and nonlinear relationships are being used to explain the vari
ations of the predicted variable (copper concentrate grade) around an operation point.
Figure 11 shows the obtained results for variable prediction over 750 minutes consider
ing a 5minute sample time.
CONCLUSI ONS
The availability of RtPM infrastructure simplifies the addition of specialized data analysis
tools. The ease of collection adds context and helps filter the data to simplify the data recon
ciliation, multivariate data analysis, and the development of online pattern detection models.
Multivariate analysis has demonstrated the ability to identify variability sources that
were not initially considered in the characterization of the operation’s behavior. Specifically,
0 50 100 150
2.5
2.0
1.5
1.0
0.5
0
–0.5
–1.0
–1.5
–2.0
FIGURE 11 Copper concentrate grade prediction
190 ADVANCES IN COMMINUTION COMMINUTION PRACTICES
the software detected the influence of the SAG mill liner’s state in its operation and iden
tified the established relationships between the main variability sources.
The knowledge acquired in the pattern identification procedures can be successfully
incorporated in the online process supervision in order to detect and predict plant distur
bances. In the presented cases, the implemented methodology helped define the refer
ence operation’s patterns, and main variability sources aided in the identification and
definition of the existent correlation structures for the controlled process.
Future applications should demonstrate a decisive potential in operation improve
ment, not only for the SAG mill liner but also in other important production areas,
including the flotation plant. Therefore, the use of information obtained through statistical
multivariate analysis and from the operator’s knowledge for the continuous improvement
of process control and maintenance are shown to be important factors for future evaluation.
SCAN, which is developed by Contac Ingenieros, uses PI Datalink and PI ACE from
OSISoft as the basic tools for implementation.
BI BLI OGRAPHY
Bascur, O.A., and Kennedy, J.P. 2000. Pages 115–121 in Web Enabled Industrial Desktop
to Increase Overall Process Effectiveness in Metallurgical Plants. IFAC Workshop, Fin
land, August 22–24.
———. 2004. Are you really using your information to increase the effectiveness of your
assets and people? Improving and optimizing operations: Things that actually work!
In Proceedings Plant Operators Forum. Edited by E.C. Dowling and J. Marsden. Little
ton, CO: SME.
Bascur, O.A., and Linares, R. 2005. Grade Recovery Optimization Using Data Unification and
Realtime Gross Error Detection. Centenary Flotation Symposium, Brisbane Australia,
June 6–9. Edited by G. Jameson and R.H. Yoon. Littleton, CO: SME.
Dudzic, M. 1998. The use of advanced multivariate statistical tecnologies (chemometrics)
at Dofasco. AISE Conference, MIT, Cambridge, MA, July.
Hodouin, D., MacGregor, J.F., Hou, M., and Franklin, M. 1993. Multivariate statistical
analysis of mineral processing plant data. CIM Bulletin 23–34.
MacGregor, J.F., and Kourti, T. 1998. Pages 31–41 in Multivariable Statistical Treatment of
Historical Data for Productivity and Quality Improvements. FOCAP0 98. Volume 94.
AIChE Symposium Series 320. Edited by J.F. Pekny and G.E. Blau. Chelsea, MI: Ann
Arbor Press.
MacGregor, J.F., Marlin, T.E., Kresta, J., and Skagerberg, B. 1991. Multivariate statistical
methods in process analysis and control. AIChE Publication P67. Pages 17–22 in
Proceedings of the CPCIV. AIChE Symposium. Edited by Y. Arkun and W.H. Ray.
Rodriguez, R., and Tobias, R. 2001. Multivariate methods for process knowledge discov
ery: The power to know your process. Pages 252–26 in Statistics, Data Analysis, and
Data Mining.
Romero, F., Orchard, M., and Yacher, L. 2003. Statistical multivariate analysis for
improved plant control at C.M. Doña Ines de Collahuasi SCM, Chile. 2003 SME
Annual Meeting, Cincinnati, OH.
Romero, F., Suarez, M., and Bascur, O.A. 2004. Improving metallurgical performance at
Collahuasi. SME Annual Meeting, Denver, CO.
Vaculik, V., and Smyth, A. 2003. Dofasco’s enhanced monitoring system. OSIsoft PI Users
Conference, San Francisco, May 13.
191
PART 3
Liberation and Breakage
193
Shell and Pulp Lifter Study at the Cortez
Gold Mines SAG Mill
Raj. K. Rajamani,
*
Sanjeeva Latchireddi,
*
and Julius Stieger
†
ABSTRACT
The energy efficiency of certain highthroughput grinding mills can be attributed to the field of
breakage and slurry transport. The charge motion and breakage of particles inside the mill
depends upon the shell lifter design, whereas the discharge of ground particles is controlled
by the grate and the pulp lifters. The design of these mill components has been largely based
on trial and error and hence varies considerably among manufacturers. This paper discusses
the use of two stateoftheart design tools—MillSoft and FlowMod. MillSoft is an effective
tool to design the shell lifters and optimize charge motion, whereas FlowMod simulates the
slurry discharge system to design grate and pulp lifters. A study done at the Cortez Gold
Mines semiautogenous grinding (SAG) mill shows how the redesign of the shell lifter brings
about a reduction in energy consumption when slurry transport through the mill is adequate.
I NTRODUCTI ON
There are a number of SAG mills in operation around the world with diameters reaching
up to 40 ft. These operations continually invest in new technologies to improve their
energy efficiency and capacity in their SAG circuit. Commercial SAG mill performance is
determined by a large number of variables, both minesite variables and mill variables.
In many cases these variables dictate production capacity seemingly randomly. There
fore, a number of operating philosophies, each specific to a plant, have arisen. In almost
all concentrators, the SAG operation is continually evolving. Every year, ways and means
are sought to increase capacity, decrease energy consumption, and prolong lifter and
liner life. Ore blending, newer designs of lifters, recycle crushing, and redesign of grates
and trommel screens are a few routes taken at considerable expense.
Operation of SAG Mills
The processing capability of a SAG mill is greatly affected by ore geology and operating
variables within the mill. The key issues broadly can be classified into two categories:
field of breakage and charge motion, and flow through the grate and pulp lifters. The
field of breakage and charge motion are affected primarily by the design of shell lifters
and mill speed. Once the ore is ground to a size that can pass through the grate holes, the
slurry flows into the pulp lifter chamber that transports it into the discharge trunnion.
These components of the SAG mill are schematically shown in Figure 1.
* Department of Metallurgical Engineering, University of Utah, Salt Lake City, Utah
† Cortez Gold Mines, Crescent Valley, Nevada
194 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
Once the slurry has made its way via the grinding media charge, its first stage of dis
charge is via the grate. Hence, in the absence of any subsequent restriction, the maximum
flow capacity that can be obtained for a given mill is determined by the grate design.
Here, the design variables are the open area and radial distribution of slots. The driving
force for slurry transport from the mill shell through the grate holes is the difference in
pressure head across the grate.
Field of Breakage. The motion of charge or rocks and balls in SAG mills can be
viewed as a field of breakage generated as a result of the internal profile of the lifters and
the rotational speed of the mill shell. The ore entering through the feed port is ground by
this field and, after being sufficiently ground to the grate slot size, the slurry leaves
through the slots in the grate. The field of breakage influences the rock mass in the SAG
mill. Should the incoming ore be harder and the field of breakage insufficient to reduce
the size, the ore stays in the mill longer because it is unable to pass through the grate.
The net effect is an increase in rock mass, and the feed rate to the mill must be decreased
appropriately to maintain rocktoball ratio. On the other hand, when the ore is soft, the
field of breakage reduces the ore size rapidly, and hence the rock mass decreases. To sus
tain a set rock mass, the feed rate must be increased. The complicating factor is that the
incoming ore feed itself determines the breakage field.
Flow through the Grate and Pulp Lifters. Discharge grates and pulp lifters play an
important role in performance of the autogenous and semiautogenous mills (Latchireddi
2002). The performance of the pulp lifters in conjunction with grate design determines
the flow capacity of these mills. The function of the pulp lifters is simply to transport the
slurry passing through the discharge grate into the discharge trunnion. Its performance
depends on the mill size and design, the grate design, and mill operating conditions,
such as mill speed and charge level. The difficulties associated with slurry transportation
from SAG mills have become more apparent in recent years with the increasing trend to
build largerdiameter mills for grinding high tonnages. This is particularly noticeable when
SAG mills are run in closed circuit with classifiers such as fine screens or hydrocyclones.
The performance analysis of conventional pulplifter designs shows that a large
amount of slurry flows back from the pulp lifter into the mill (Latchireddi and Morrell
1997, 2003a,b; Rajamani et al. 2002). The backflow depends on the size and design of
the pulp lifters. The back face of the pulp lifter is the grate itself, so that the slurry readily
flows back into the mill. Subsequently, the field of breakage diminishes when excessive
slurry builds in the mill.
Feed
Trunnion
Mill
Shell
Grate Pulp
Lifter
Discharge
Trunnion
FIGURE 1 Schematic of a typical SAG mill
SHELL AND PULP LIFTER STUDY 195
Charge Motion. In a concentrator, all of the auxiliary equipment (i.e., pumps,
conveyers, screens, and hydrocyclones) and two primary resources (i.e., steel and electric
ity) primarily serve to maintain grinding action in the belly of the SAG mill. It is this
action that dictates capacity. Therefore, this grinding action should be observed continu
ously from the control room so that the necessary steps can be taken to keep the grinding
field at its highest potential. Unfortunately, the grinding environment within the mill
shell is very severe, and none of the online instrumentation developed so far is able to
survive the continuous impact of large steel balls. Direct observation is impractical,
therefore, the next available option is a simulation of the grinding field to gauge the
intensity of grinding or the lack of intensity of grinding.
Mill Power Draft. The field of breakage and flow through the grate and pulp lifter
influence each other, and the net effect is the buildup of a holdup level in the mill, which
draws a certain power, and this power draft is clearly linked with mill throughput. If the
interaction can be understood, then mill capacity can be determined much more clearly.
Then the expectation of increasing capacity at the same level of power draft by one
means or another can be safely evaluated.
The power draft of a SAG mill and its consequences are illustrated in Figure 2,
wherein 5 days of operating data in a 32u14ft SAG mill is plotted. The power draft of the
mill is held between 6 and 7 MW, whereas tons per hour (tph) of ore feed to the mill
shows wide variations between 1,000 and 1,600 tph. Figure 2 shows that the feed rate
drops whenever the power draft shows an increasing trend, whereas intuitive reasoning
would suggest that feed rate should be proportional to power draft.
The data show that the specific energy consumption (kWh/t) of ore is not steady, as
one would expect for a typical ore body. Even within a 24hour time frame, where the
feed ore hardness may be assumed constant, the variation in feed rate is dramatic. The
internal dynamics within the SAG mill, as exemplified by the three broad concepts, are
causing wide fluctuations in grinding rate, which in turn is reflected as capacity.
SAG Mill Efficiency
The energy efficiency of tumbling mills can be examined directly by looking at the
motion of ore and grinding balls inside the mill. The makeup of the charge and the lifter
bars attached to the inside of the mill shell can be designed particularly to maximize the
mass of ore fractured per unit of energy spent. At the same time, the unnecessary colli
sions of steel balls against the mill shell can be reduced. Furthermore, the cascading
charge flow can be altered in such a way as to maximize grinding efficiency. First, the
shell lifters are designed in such a way that the motion is fully cascading and that part of
cataracting motion is made to strike in the vicinity of the toe. In such a charge motion
regime, both shearing action and impacts are fully utilized in grinding the ore.
The shell lifters are usually replaced once or twice a year. Pulp lifters survive for 2 or
3 years. The design of these two important mill components has been based largely on
trial and error and hence varies considerably among manufacturers. However, over the
years, important tools such as MillSoft and FlowMod have begun to help the designers
analyze and understand the influence of the internal components of SAG mills—shell lifters,
grates, and pulp lifters, respectively. The following sections discuss the basic principles
and application of these two simulators.
MillSoft—Discrete Element Simulation of SAG Mills. The SAG mill is made up of a
cylindrical shell with two conical shells attached on both ends. Lifter elements are
attached in both the cylindrical and conical shell sections. As the SAG mill rotates, typi
cally around 10 rpm, the internal flat walls of the lifter and shell impart momentum to
the balls and rocks. The momentum is transferred primarily to particles in direct contact
with the plate elements (Villouta 2001). These particles, in turn, impart their momentum
196 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
to adjacent layers of particles. In this manner the motion of the entire charge evolves,
resulting in what is commonly referred to as cascading and cataracting charge.
The discrete element method embedded in MillSoft replicates the evolution of charge
motion as described above. In the simulation, the exact dimensions of lifters, plates, and
balls are used. MillSoft can be used effectively to show the effects of different lifter con
figurations, mill loading, mill speed, and other operating conditions. Unlike singleparticle
trajectory programs that show only the outermost particle path, MillSoft takes into
account the entire charge; one can effectively see the “kidney” shape of the charge, the
dead zone, toe and shoulder of the charge, and areas of high impact on the mill lifters.
MillSoft also can be used to follow lifter wear, shell plate wear, and particle breakage.
Most importantly, the location and the intensity of impacts on lifter bars can be recorded
computationally, and a corresponding metal abrasion at that location can be worked out.
In a like manner, the energy of impact can be used to fragment the rock particles in the
simulation. However, the distribution and number of fragments produced overwhelm
the computational task, and hence, this computational path is rarely followed.
Shell Lifter Design and Charge Motion Analysis with MillSoft. A typical example
of charge motion analysis with MillSoft is described in this section. The SAG mill under
consideration is a 38u24ft mill drilled for 60 rows of shell lifters. Therefore, the mill can
be fitted with a set of 30 high and 30 low lifters or a total of 60 high lifters. The total
charge is 27% with 15% balls. The mill is expected to draw 15–18 MW power. The mill
speed is set at 76% critical speed. The snapshots of the charge motion at different condi
tions are shown in Figure 3.
Figure 3a shows the SAG mill fitted with traditional lifters. The high lifters are the
tophat type with a 7˚ release angle. The velocity vectors are superimposed on the balls
and the rock particles. The central region reveals where grinding action is minimum.
Due to the small release angle, a considerable amount of rock and ball particles are
thrown against the mill shell. The balltoliner strike zone extends as high as the nine
16Mar01 17Mar01 18Mar01 19Mar01 20Mar01 21Mar01 22Mar01 23Mar01
18
16
14
12
10
8
6
4
2
0
tph × 100
rpm
MW × 1
FIGURE 2 Five days of plant operating data: 32×14ft SAG circuit (fresh feed rate in tons per
hour, mill power in megawatts)
SHELL AND PULP LIFTER STUDY 197
o’clock mark. The mill may not reach design capacity, especially with a hard ore type.
Furthermore, considerable damage to liners is imminent within 4 months of operation.
Next, we consider the same mill fitted with 30 high lifters and 30 low lifters of a 25˚
release angle. Figure 3b shows the snapshot of charge animation. Here, ball strikes on
liners are seen nearly up to the eight o’clock position, a considerable improvement over
that shown in Figure 3a. This lifter is suitable for maintaining a moderately aggressive
charge motion at the expense of shorter lifter wear life.
Next, the release angle is increased to 35˚ with 30 high lifters, as shown in Figure 3c.
The cataracting charge lands within the toe of the charge. This type of charge motion is
ideal for SAG mills in order to preserve the life of shell lifters. The mill speed may be
increased without fear of damaging liners. The Alumbrera mines exploited this concept
to increase production. With such lifters, Alumbrera even used 150mm top ball size.
SAG Mill Study at Cortez Gold Mines
The aforementioned analysis of shell and pulp lifters is illustrated with the work done at
Cortez Gold Mines, Crescent Valley, Nevada (Stieger et al. 2005). The grinding circuit
consists of a 26u13ft SAG mill in closed circuit with a pebble crusher. The discharge of
the SAG mill is screened on a 0.75in. screen, and the oversize material is fed to the cone
crusher. The undersize is sent to the ball milling circuit. The typical SAG mill feed is
400 stph, which varies anywhere between 250 and 550 stph, depending on the ore type.
At least five different ore types are encountered at this mine site.
During the course of many years, the SAG shell lifter has evolved to a high–low pattern
with a typical highlifter dimension of a 7in. height, 5in. top width, and 17˚ face angles
on both sides. The lowlifter dimensions are 5in. height, 5in. top width, and 17˚ face
angles on both sides. The mill shell has been drilled for 52 rows of shell lifters. The open
area of the grate is 7% with the typical 2.75in.
2
opening.
The plant operating work index (Wio) depicted in Figure 4 shows an average of 15 kWh/st
in the year 2004 until the liner change, since then decreasing to about 13 kWh/st. This
reflects the efficient usage of energy in the SAG mill as well as the ball mill operation,
which is probably getting a relatively finer feed.
Grate and Pulp Lifters. First, a review of the grate plate and pulp lifter showed
that an open area of 7% was adequate for handling the dailytargeted tonnage. In fact,
the grate openings were found to be free of balls or rocks during many inspections. The
discharge capacity of the grate is 482 m
3
/h of slurry. However, FlowMod calculations
indicated that the pulp lifter diminished this flow to 382 m
3
/h as a result of backflow
phenomena. However, this flow rate is adequate to handle current daily tonnage. Also,
(a) Top Hat (7
˚
) (b) 25
˚
Release Angle (c) 35
˚
Release Angle
FIGURE 3 Snapshots of charge motion
198 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
the radial distribution of grate opening in the mill periphery indicated that some advan
tages could be gained by redistributing the open area in the most optimal flow regime.
The recommendation was implemented in a subsequent grate redesign. In summary, the
grate and pulp lifter combination was operating more than satisfactorily, although there
is always room for further increase in the pulp discharge capability of the mill.
Figure 5 shows discharge flow rate as a function of fractional slurry holdup. At the
current operating conditions, increasing the open area to 9% and redistributing the slots
radially may increase the discharge flow rate to 450 m
3
/h.
It is very critical that the pulp transport capacity of the mill be set at its maximum
value before changing shell lifters. The shell lifters may increase the production of fines,
but there must be the capacity to discharge these fines.
High–Low ShellLifter Experience. The high–low shelllifter design leaves a gap
of 10 in. between the lifters. As a result, caking between lifters was very severe, as shown
in Figure 6. Due to cake buildup, the effective height of the high lifter over the base is a
mere 2 in. Figure 6 shows the MillSoft simulation of charge motion with the high lifters.
The 17˚ lead face angle causes cataracting between the eight o’ clock and nine o’ clock
positions of the mill circle.
With the use of 5in. grinding balls and exposed shell plates, the cataracting caused
consistent and moderatelevel damage to the mill shell. Some of the lifters were broken,
and in other places there was severe peening. As a result, the mine experienced unsched
uled SAG mill–related downtime every month. Figure 7 shows the unscheduled down
time for a 2year period.
It can be seen that though the lifter is in the last 4 months of operation leading up to
September 2004, there is downtime due to lifter damage. A crash stop done during this
period shows cake buildup between lifters and a fair amount of slurry retention within
the mill (see Figure 6).
ShellLifter Redesign. A decision was made to reduce downtime and increase
energy efficiency with a new design of lifters. In particular, it was decided to bring the 5in.
ball trajectory to the toe of the charge by correct choice of the leading face angle. Fur
thermore, it was decided to eliminate every other lifter row to minimize packing and
10
11
12
13
14
15
16
17
18
Jan
04
Feb
04
Mar
04
Apr
04
May
04
Jun
04
Jul
04
Aug
04
Sep
04
Sep
04
Oct
04
Nov
04
Dec
04
Jan
05
Feb
05
Mar
05
Apr
05
May
05
Jun
05
New Design
0
5
10
15
20
25
30
35
40
45
50
W
i
o
,
k
W
h
/
s
t
Time
%
P
o
w
e
r
t
o
S
A
G
Wio kWh/st
% Power to SAG
FIGURE 4 Monthly record of plant operating work index
SHELL AND PULP LIFTER STUDY 199
maximize lift as well as to increase mill volume. Figure 8 shows a MillSoft simulation of
the new design (9in. height, 5in. top width, with 28˚ leading and trailing face angles).
As anticipated, the cataracting charge lands near the toe of the charge at around the
seven o’clock position of the mill circle. The simulated power draw was consistent with
operating power draw. This type of liner with the leading face inclined at a steep angle
(22˚–35˚) has been well documented in the literature. A number of mine sites, including
Alumbrera (Sherman 2001), Collahuasi (Villouta 2001), Candelaria, Los Pelambres,
and others (Bird et al. 2001), have had success with these types of lifters. In addition to
the design criteria for optimal trajectories for 5in. grinding balls, a number of other
5
6
7
8
9
10
11
12
13
14
15
0 100 200 300 400 500 600
Slurry Flow Rate, m
3
/h
%
S
l
u
r
r
y
H
o
l
d
u
p
,
v
/
v
Slurry Removal Efficiency of Pump Lifters = 382.65/481.88 = 79.4%
Grate Only
Radial Pulp Lifter
420 stph
FIGURE 5 FlowMod analysis of slurry flow rate through the SAG mill
FIGURE 6 Charge motion simulation and crashstop picture with high–low shell lifters
Packing
Slurry Pooling
200 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
issues—such as safety of liner handling, safety of mill noise, inching drive capability, load
on the mill motor, and mill startup—were addressed and taken care off.
Slurry Transport and Load Buildup in the Mill. Slurry transport out of the mill
plays an important role in determining SAG capacity. Figure 9 shows the cyclical behav
ior of the SAG circuit 3 weeks after changeover to the new lifter design. In particular, the
feed to the SAG mill cycles up and down every 2 hours. Mill bearing pressure exhibits
similar behavior. The cyclical behavior is primarily due to the pulp lifter returning part of
the slurry passing through the grate back into the mill. In other words, backflow in the
pulp lifter returns part of the slurry to the mill. As a result, the mill slurry holdup
increases and the controller cuts the feed to the SAG mill. This cyclical behavior points
out that the circuit capacity can be improved by a proper choice of pulp lifters.
Impact on Power Draw and Energy Consumption. The energy efficiency of the
SAG mill is the main focus of this section. Figure 10 shows the SAG throughput before
and after installation of a new shell lifter. The SAG circuit maintains more or less the
12.0
10.0
8.0
6.0
4.0
2.0
0.0
1,400,000
1,200,000
1,000,000
800,000
600,000
400,000
200,000
0
J
a
n

0
3
F
e
b

0
3
M
a
r

0
3
A
p
r

0
3
M
a
y

0
3
J
u
n

0
3
J
u
l

0
3
A
u
g

0
3
S
e
p

0
3
O
c
t

0
3
N
o
v

0
3
D
e
c

0
3
J
a
n

0
4
F
e
b

0
4
M
a
r

0
4
A
p
r

0
4
M
a
y

0
4
J
u
n

0
4
J
u
l

0
4
A
u
g

0
4
S
e
p

0
4
O
c
t

0
4
N
o
v

0
4
D
e
c

0
4
J
a
n

0
5
F
e
b

0
5
M
a
r

0
5
A
p
r

0
5
M
a
y

0
5
J
u
n

0
5
Date
D
o
w
n
t
i
m
e
,
h
R
e
v
e
n
u
e
,
$
9/2003:
Implemented
maximum SAG
speed limit
Downtime still
due to old design
high–low lifters
New Design
1/2004:
Replaced old design
SAG shell lifters with
new design
Cumulative $
SAG Liner Unscheduled Downtime, h
FIGURE 7 Summary of SAG mill unscheduled downtime
FIGURE 8 Charge motion simulation and crashstop picture with new shell lifters
SHELL AND PULP LIFTER STUDY 201
250
300
350
400
450
500
550
600
650
700 100
90
80
70
60
50
40
30
20
10
0
%
S
o
u
n
d
1
2
:
0
0
:
0
4
1
2
:
3
5
:
2
6
1
3
:
1
0
:
4
7
1
3
:
4
6
:
0
9
1
4
:
2
1
:
3
0
1
4
:
5
6
:
5
2
1
5
:
3
2
:
1
3
1
6
:
0
7
:
3
5
1
6
:
4
2
:
5
6
1
7
:
1
8
:
1
8
1
7
:
5
3
:
3
9
1
8
:
2
9
:
0
1
1
9
:
0
4
:
2
2
1
9
:
3
9
:
4
4
2
0
:
1
5
:
0
6
2
0
:
5
0
:
2
7
2
1
:
2
5
:
4
9
2
2
:
0
1
:
1
0
2
2
:
3
6
:
3
2
2
3
:
1
1
:
5
3
2
3
:
4
7
:
1
5
Time
B
e
a
r
i
n
g
P
r
e
s
s
u
r
e
/
C
u
r
r
e
n
t
,
s
t
p
h
Feed
Bearing Pressure
Current
Sound
FIGURE 9 Cyclical behavior of SAG mill
Time
M
i
l
l
F
e
e
d
,
/
M
i
l
l
B
e
a
r
i
n
g
P
r
e
s
s
u
r
e
,
s
t
p
h
p
s
i
510
490
470
450
430
410
390
370
350
J
a
n

0
4
F
e
b

0
4
M
a
r

0
4
A
p
r

0
4
M
a
y

0
4
J
u
n

0
4
J
u
l

0
4
A
u
g

0
4
S
e
p

0
4
S
e
p

0
4
O
c
t

0
4
N
o
v

0
4
D
e
c

0
4
J
a
n

0
5
F
e
b

0
5
M
a
r

0
5
A
p
r

0
5
M
a
y

0
5
J
u
n

0
5
New Design
Learning Curve
SAG Mill,
SAG Mill Feed,
tph
psi
FIGURE 10 SAG mill operation before and after installation of new shell lifters
202 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
same throughput after lifter installation. Because ore type is changing from day to day, it
will take more than 4 months to encounter all the different ore types. The major advan
tage gained with the new lifter design is shown in Figure 10. The fact that the mill bear
ing pressure stays at a steady value both before and after the lifter change implies that
the mill load is unaffected by the design of the new lifters. However, the mill feed rate
steadily increased after the lifter change. These two observations imply that the new
lifter design is bringing about an efficient breakage of ore particles. Thus, the mill
throughput increases while maintaining the same load.
The critical impact of the new lifter design is illustrated in Figure 11. The SAG mill
exhibits a very definite reduction in power draft. It is estimated that the power decreases
in the 230–370kW range. Hence, energy consumption per ton of ore milled decreases by
0.3–1.3 kWh/t. This energy savings is in the 10% range. Furthermore, a 1%–10% reduc
tion in recirculation to the cone crusher was noticed due to efficient impact breakage of
critical size material. All of these results amount to a significant reduction in operating
costs.
A more efficient ball trajectory or charge motion means that there is less direct
impact of grinding balls on shell plates and lifters. As a result, grinding ball consumption
and steel loss in lifter wear must be impacted. At the Cortez Gold Mines operation, grind
ing ball consumption could not be tracked via a digital control system; however, it was
noticed that during 16 weeks of operation, only 2.5 in. of lifter height was lost due to
wear. It was estimated that the new design results in 57 kg/day in steel loss compared to
84 kg/day for the previously installed lifter, a 47% savings in steel loss. Furthermore, in
the 9 months of operation leading up to June 2005, the mill did not experience any
downtime due to cracked shell plates, severely peened lifters, broken lifters, or leaky
bolts. Thus, we find that proper design of shell lifters leads to a decrease in energy con
sumption per ton of ore.
CONCLUSI ONS
The design of both the shell lifters and the grate–pulp lifter assembly are crucial for optimal
performance of a SAG mill. The design of shell lifters, which control the charge motion
and thus the breakage field, can be optimized using MillSoft—a discrete element numeri
cal method. FlowMod, a steadystate simulator, can be used to optimize the design of
grate and pulp lifters to handle the given flow through the mill. It estimates the slurry
holdup inside the mill and shows its dynamic surface at any mill operating condition.
These two tools were employed in the study of the SAG mill at Cortez Gold Mines. First,
the flow or discharge capacity of the grate and pulp lifter were analyzed, and it was
found that the capacity was adequate for meeting daily tonnage. Next, the redesign of
shell lifters readily resulted in a 230–370kW reduction in mill power draw while main
taining the same throughput level. The SAG mill circuit exhibited cyclic loading behav
ior, indicating that there was room for further increase in capacity via pulp lifter
redesign.
ACKNOWLEDGMENTS
The authors from the University of Utah would like to thank the U.S. Department of
Energy, Industries of the Future Program, for support of this study through contract
DEFC2603NT41786. The authors also thank Cortez Gold Mines management for par
ticipating in this study.
SHELL AND PULP LIFTER STUDY 203
BI BLI OGRAPHY
Bird, S., A.E. Lamb, W. Lamb, and D.W. Partridge. 2001. Evolution of SAG mill liner design
at Kennecott Utah Copper Concentrator. Pages 256–259 in International Autogenous
and Semiautogenous Grinding Technology. Volume III. Edited by D.J. Barratt, M.J. Allan,
and A.L. Mular. Vancouver, BC: University of British Columbia, Department of Min
ing and Mineral Process Engineering.
Denis, N., S. Morrell, B. Chapman, and S. Latchireddi. 2001. The development and
installation of the twin chamber pulp lifter at Alcoa. In Proceedings SAG ’01, Vancou
ver, BC.
Latchireddi, S.R. 2002. Modeling the performance of grates and pulp lifters in autoge
nous and semiautogenous mills. Ph.D. thesis. Brisbane, Australia: University of
Queensland.
Latchireddi, S.R., and S. Morrell. 1997. A new design of pulp lifter for grate discharge
mills. Pages 57–61 in 6th Mill Operators Conference, Madang, Papua New Guinea,
October.
———. 2003a. Slurry flow in mills: Grateonly discharge mechanism, part 1. Minerals Engi
neering 16(7):625–633.
———. 2003b. Slurry flow in mills: Grate–Pulp lifter discharge mechanism, part 2. Miner
als Engineering 16(7):635–642.
Mishra, B.K., and R.K. Rajamani. 1994a. Simulation of charge motion in ball mills. Part 1:
Experimental verifications. International Journal of Mineral Processing 40:171–176.
———. 1994b. Simulation of charge motion in ball mills. Part 2: Numerical simulations.
International Journal of Mineral Processing 40:187–197.
2,200
2,300
2,400
2,500
2,600
2,700
2,800
2,900 10.0
9.0
8.0
7.0
6.0
5.0
4.0
3.0
2.0
1.0
0.0
Jan
04
Feb
04
Mar
04
Apr
04
May
04
Jun
04
Jul
04
Aug
04
Sep
04
Sep
04
Oct
04
Nov
04
Dec
04
Jan
05
Time
S
A
G
,
k
W
/
a
m
p
s
SAG Mill,
SAG Mill,
kW
kWh/t
FIGURE 11 SAG mill power data
204 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
Moys, M.H. 1986. The effect of grate design on the behavior of grate discharge mills.
International Journal of Mineral Processing 18:85–105.
Rajamani, R.K. 1997. Slurry transport through the dynamic porosity of the ball mill.
Transactions of the Indian Institute of Metals 50(5):337–347.
Rajamani, R.K., and B.K. Mishra. 1996. Dynamics of ball and rock charge in SAG mills.
Pages 700–712 in International Autogenous and Semiautogenous Grinding Technol
ogy. Volume 2. Edited by A.L. Mular, D.J. Barratt, and D.A. Knight. Vancouver, BC:
University of British Columbia, Department of Mining and Mineral Process Engineering.
Rajamani, R.K., B.K. Mishra, A. Joshi, and J. Park. 2002. Two and three dimensional sim
ulation of ball and rock charge motion in large tumbling mills. In DEM Numerical
Modelling of Discontinua. Geotechnical Special Publication 117. Edited by B.K. Cook
and R.P. Jensen. Reno, VA: American Society of Civil Engineers.
Rajamani, R.K., B.K. Mishra, and P. Songfack. 1997. The modeling of rock and ball
charge motion in SAG mills. Pages 195–200 in Comminution Practices. Edited by S.K.
Kawatra. Littleton, CO: SME.
Sherman, M. 2001. Optimisation of the Alumbrera SAG mills. Pages I59–I75 in SAG
2001, Vancouver. Edited by D.J. Barratt, M.J. Allan, and A.L. Mular. Vancouver, BC:
University of British Columbia, Department of Mining and Mineral Process Engineering.
Stieger, J., D. Plummer, S. Latchireddi, and R. Rajamani. 2005. SAG mill operation at
Cortez: Evolution of liner design. SME Annual Conference, Salt Lake City, UT, Feb
ruary 28–March 2.
Villouta, R.M. 2001. Collahuasi: After two years of operation. Pages I31–I42 in SAG 2001,
Vancouver. Edited by D.J. Barratt, M.J. Allan, and A.L. Mular. Vancouver, BC: Uni
versity of British Columbia, Department of Mining and Mineral Process Engineering.
205
Breakage and Damage of Particles by Impact
L.M. Tavares,
*
L.G. Austin,
†
and R.P. King
‡
ABSTRACT
Measurements of forces resulting from impacts of 3×3mm cylindrical pellets fired at a con
stant velocity against a target mounted on a rapidresponse load cell showed that the maxi
mum force generated at each impact was not constant. The measured distribution of
maximum force was used to calculate a distribution of energy utilization factors. The results
indicated that measurements of the distribution of strengths calculated, assuming that the
specific impact energy defines strength, are in error for nonspherical particles. Dropweight
impact tests on roughly spherical mineral particles mounted on an impact load cell showed
that the distribution of force required to produce fracture was dependent on the distribution
of existing damage as indexed by initial stiffness factor, that of the damage accumulation
coefficient, and that of the critical deformation at fracture. Knowledge of the distribution of
strength alone is not sufficient to define the fracture behavior of the particles under known
impact conditions, especially for repeated lowlevel impacts where damage accumulates.
I NTRODUCTI ON
The two most important types of grinding mills in terms of the tons ground per year
around the world are tumbling media mills (mainly for rocks and ores) and ballrace/
rollrace mills (mainly for pulverized coal for electricity generation). The traditional
laboratoryscale tests used to determine the required size of these mills for different feed
materials are the Bond test and the Hardgrove test, respectively. These tests use stan
dardized test mills that are smallscale versions of the fullscale mills, and each test delivers
a single number that compares the “ease of grindability” of different materials: the Bond
Work Index (Bond 1960) and the Hardgrove Grindability Index. Although the breakage
actions are quite different between the two mill types, the two indices have a reasonable
empirical intercorrelation (Austin and Aplan 1998): a material that is predicted by the
Bond test to be easy (or hard) to grind is also predicted to be easy (or hard) to grind by
the Hardgrove test. The actual use of these empirical indices for mill design is based on
many years of industrial experience with each type of mill on a range of different feed
materials.
These two indices are not useful generally either for the design of the many other
types of grinding mills of importance or for predicting performance of new types of
mills. Thus, there is a clear need for a more fundamental approach to describe those
properties of a given material that make it easy or difficult to comminute in any given
* Universidade Federal do Rio de Janeiro, Rio de Janeiro, Brazil
† The Pennsylvania State University, University Park, Pennsylvania
‡ University of Utah, Salt Lake City, Utah
206 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
type of device. Recognition of this need in the past led to a substantial amount of effort
in this area, with little apparent success. However, recent years have seen a renewed
interest in characterizing the breakage properties of particles under specified stressing
conditions, using new techniques and ideas.
It is hardly worth the effort to try to extract data from the older literature because
the reported test results—from dropweight tests, for example—lack important informa
tion. On the other hand, certainly a number of important concepts were discovered. For
example, the cracking and eventual collapse of cubes of coal subjected to repeated light
impacts was described (Bond 1954), which is an example of the damage accumulation
that will be discussed later in the paper. Similarly, Austin and coworkers (Austin, Shoji,
and Everell 1973) postulated that if the distribution of “strengths” of a given size of par
ticle were known, and the distribution and frequency of applied forces were known, then
the two distributions could be convoluted to give rates of breakage. The problems of
applying such ideas, of course, included the need for the development of suitable test
procedures for (1) the determination of strength, and (2) the determination of the suite
of forces in an operating machine. In addition, there was no formal structure to describe
and measure the process of damage accumulation.
This chapter will present briefly the background relevant to the recent work done in
this field, with special reference to solving the problem of describing and measuring
damage accumulation.
I DEALI ZED STRESS DI STRI BUTI ONS AND FRACTURE MECHANI CS
I N I MPACTED SPHERES
Figure 1 shows idealized diagrams of how a sphere of homogenous material appears
when it is impacted between smooth, rigid anvils or fired to impact against an anvil.
The Hertzian solutions for these impacts are known (Goldsmith 1960), and the result
ing equations are summarized in Appendix A. For example, the equation for the maximum
force produced by the perpendicular impact of a sphere (of diameter d, material of den
sity ȡ, and stiffness factor k
p
—also called particle stiffness) on a flat surface (of stiffness
k
s
) at velocity v is (Equation A9)
(EQ 1)
where K
e
= k
p
k
s
/(k
p
+ k
s
), k = Y/(1 – µ
2
), Y is the Young’s modulus, and µ is the Poisson
ratio.
This maximum occurs at the instant when the kinetic energy is converted completely
to the strain energy of compression and when the relative velocity of sphere to surface is
zero (before rebound). The solutions also show that the compression caused by the impact
produces a maximum in tensile stress acting around the perimeters of the contact circles.
Griffith (1921) showed that a small defect (a Griffith flaw) in a brittle solid leads to con
centration of the stress field around the flaw, and that a large flaw suitably aligned in a
region of high tensile stress can act to initiate an expanding crack. If the stressing condi
tions are appropriate, the crack can propagate rapidly, bifurcate at other flaws, and lead
to a disintegrative failure. For spheres, then, the cracks would be expected to propagate
from flaws in the rim of the contact circle, and the rest of the stress field would drive the
cracks to produce fracture into rough segments. This type of behavior has been observed
(Schönert 1986). Appendix A shows that the tensile stress ı
T
in the perimeter of the con
tact circle is proportional to the (mass) specific impact energy E
1/5
(Equation A11):
(EQ 2)
F
m
0.757d
2
K
e
2 5
µv
2
( )
3 5
=
o
T
0.32 1 2µ – ( )K
e
4 5
µE ( )
1 5
=
BREAKAGE AND DAMAGE OF PARTICLES BY IMPACT 207
Thus, there will be a critical, specific impact energy that will make ı
T
sufficient to cause
disintegrative fracture of the sphere. Similarly, another method of defining the strength
is given by (Hiramatsu and Oka 1966)
(EQ 3)
where o
p
was called “particle fracture strength” (Tavares and King 1998). F
c
is the critical
value of F
m
at which failure occurs. It can be shown that this is proportional to the spe
cific impact energy E
3/5
for the case of onepoint contact (n = 1 in Equation A12):
(EQ 4)
Equations 2 and 4 show that “strength” can be defined as the critical ı
T
, or as ı
p
, or
by the critical, specific impact energy. The strength is a function of the basic bond
strength of the material and of the size, density, and alignment of the flaws.
The application of the definition of strength as the “critical, specific impact energy
required to produce fracture” requires the measurement of the mass fraction of spheres
that are broken when a reasonably large number of the spheres (of a chosen size) are
impacted with a known specific impact energy. Repeating (with a fresh sample) at
higher and higher specific impact energies gives larger and larger fractions broken. In
this way, a curve or vector of “cumulative fraction broken versus specific impact energy”
is constructed. This is the “distribution of strengths” of the spheres of that size.
With respect to the variation of strength with sphere size, Weichert (1992) and
Vogel and Peukert (2003) have treated the length of the perimeter of the contact circle
as if it were a chain length in the “weakest link” concept (Weibull 1939). It can be shown
that smaller spheres have a lower probability of having weak flaws in the contact circle
perimeter and thus will be stronger than larger spheres. This is in agreement with exper
imental evidence. However, if it is considered that smaller particles have been produced
by breakage of larger particles, then it is possible that the population of flaws for smaller
particles has a different strength distribution than that for larger particles, having a
smaller fraction of the larger, weaker flaws. Whether this effect is significant or negligible
(a) (b)
v
v
Contact
Circle
Contact
Circle
FIGURE 1 Spheres impacted (a) against an anvil and (b) between anvils
o
P
2.8F
c
td
2
 =
o
P
1.56K
e
2 5
µE ( )
3 5
=
208 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
depends on the flaw density; if the density is high, many steps of breakage can occur
without significant reduction of the density of larger, weaker flaws, and vice versa.
A MORE REALI STI C PI CTURE OF THE I MPACT BREAKAGE
OF NONSPHERI CAL PARTI CLES
Examination of the shapes of particles of interest, such as crushed or ground rocks, ores,
coals, and so forth, shows a wide variety of shapes for a given material and size, includ
ing sharp angles, flat faces, and projections, with very few particles approximating a
spherical shape. It cannot be expected that the equations for ideal spheres can be applied
directly to irregularshaped particles, except as a guide to the form of the descriptive
equations.
A major advance was made (Vervoorn and Scarlett 1990) with the construction of a
device to fire cylindrical pellets against a rigid target fitted with a force transducer that
produced force–time curves for the impact. The pellets (catalytic cracker pellets of sin
tered Al
2
O
3
) were 3 mm in diameter by 3 mm in length, 30 to 39 mg in weight, and were
tested over impact velocities of 6 to 24 m/sec.
It was found (Vervoorn 1986) that, contrary to expectations, the maximum impact
force measured from the graphical output of the transducer was not constant at a given
impact velocity but varied over a range with a ratio of about 1:5. The assumption that
the mass and velocity of the impact determined the maximum force of impact was not
correct. This finding invalidates much of the earlier work on the measurement of
strength distributions for nonspherical particles.
The impact process was found to be much more complex than that expected from
spheres. In some impacts, a single major impact peak occurred over a time period of
about 10 µsec, which is consistent with the expectation from Hertzian theory. Smaller
peaks followed, presumably due to fragments broken from the pellet also striking the
transducer. However, in many cases a double impact occurs as the pellet hits, twists, and
strikes again, giving a much smaller maximum force.
Later analysis (Vervoorn and Austin 1990) showed that the range of measured max
imum force F
m
at a given velocity could be normalized to the median value F
m50
and fitted
to a loglogistic function:
(EQ 5)
where G(F
m
) is the fraction of impacts that produces a maximum force s F
m
, G(F
m50
) = 0.5.
This curve fitted the data reasonably well for all the test impact velocities. The values of
F
m50
varied with impact velocity according to
F
m50
= 3.03v
6/5
Newtons, v in m/sec (EQ 6)
A cylinder of equal diameter and length is not much different than a sphere in geometry,
but it is different enough to give maximum impact forces that vary considerably from the
idealized treatment for spheres given previously. It must be expected that the same
result would be found for irregular particles that can also twist and dissipate impact
energy as several small impacts instead of one large one. The result will be a proportion
of impacts that do not break at a given impact velocity because the force is not as high as
calculated from the mass and velocity. This will make the strengths in the measured
strength distribution versus velocity (or versus calculated specific impact energy) appear
stronger than the real particle strengths.
G F
m
( )
1
1 F
m50
F
m
( )
5
+
 =
BREAKAGE AND DAMAGE OF PARTICLES BY IMPACT 209
To proceed further, it was necessary to have an independent estimate of the distribu
tion of strengths and an estimate of the modulus of rigidity of the pellets. This was done
(Austin, Trubelja, and Scarlett 1993) by fracturing 100 of the pellets, one by one, using
slow compression in a universal testing machine, with the pellet orientation in the Bra
zilian mode of test. With this type of test there can be no twisting of the pellets, and the
initial region of the output curve of the tester can be used to estimate the Young’s modu
lus during the pellet compression. The “strength” in this test is the load P required to pro
duce a disintegrative fracture. The distribution of strengths by this definition could be
fitted with reasonable accuracy by a loglogistic function:
cumulative fraction broken at loads s P = (EQ 7)
where P
50
was 48 N.
However, it also was found that the measured Young’s modulus was not constant
but had a wide distribution. The variation of the modulus did not correlate with the
strength. This was an important finding because the modulus of rigidity affects the max
imum force obtained by an impact of a given specific energy. Thus, part of the distribu
tion of impact forces seen in the impact tests (Equation 5) would be due to the variation
in modulus in addition to the effect of variation in the orientation of the impact. The dis
tribution of the modulus could be fitted approximately by
cumulative fraction of pellets with modulus below Y = (EQ 8)
where Y
50
was 1.3 GPa.
The equation relating the maximum impact force to the impact velocity for perpendic
ular impact of a sphere against a (much more) rigid target is (Equation 1, with )
(EQ 9)
Equation 9 predicts that the maximum impact force varies with the 6/5 power of the
impact velocity (as found in Equation 6 for F
m50
), and that a higher modulus of rigidity
increases the maximum force at a given velocity.
Equation 9 can be used to estimate the expected maximum forces for the pellet
impacts. At a given set of conditions, the only variable is k
p
, so all the other values can be
combined into a new constant, . Substituting k
p
for Y in Equation 8 and
using k
p
= (F
m
/const.)
5/2
gives
(EQ 10)
where . Thus, the distribution of the impact forces predicted solely from
the distribution of Young’s modulus is also a loglogistic function, but with an exponent
of 8.5. This can be compared with Equation 5, where the exponent was found to be 5.
This means that the measured force distribution is much wider about F
m50
than expected
from the variation of Young’s modulus alone (a higher value of the exponent in the log
logistic function gives a narrower distribution), showing that the effect of pellet orienta
tion on impact force is significant. It is clearly invalid to assume that an impact at velocity v
produces a unique force on the pellet.
1
1 P
50
P ( )
4
+

1
1 Y
50
Y ( )
3.4
+

K
e
k
p
~
F
m
0.757d
2
k
p
2 5
µv
2
( )
3 5
=
F
m
const.k
p
2 5
=
G F
m
( )
1
1 F
m50
F
m
( )
3.4 5 2 ( )
+
 =
F
m50
const.k
p50
2 5
=
210 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
This can be illustrated by using Equation 9, taking d as the equivalent spherical
diameter corresponding to the pellet volume, (2.73)10
–3
m, a median value of Young’s
modulus of (1.3)10
9
Pa, and a pellet density of (1.4)10
3
kg/m
3
corresponding to a pellet
porosity of about 55%. At this porosity, the Poisson ratio can be taken as 0.25 (Jaeger
and Cook 1969). Then
F
m50
= 4.5v
6/5
Newtons, v in m/sec (EQ 11)
Figure 2 shows the values of G(F
m
) predicted from Equations 10 and 11, at an impact
velocity of 10 m/sec, for comparison with the experimentally determined values of
Equation 5.
It can be seen that the upper limit of impact forces predicted from treating the
impacts of the pellets as if the pellets were spheres agrees with those found experimen
tally, providing the variation in the Young’s modulus found in compressive tests is
included. However, the experimentally determined impact forces are otherwise much
lower; compare (a) and (b) in Figure 2. This is in agreement with the concept that a few
of the impacts will be full impacts, but many will give lower force values due to the
impact energy being spread over two or more partial impacts. The theoretical estimates
of maximum force are too high; therefore, the measured values of “cumulative fraction
broken at force F
m
(or specific impact energy E)” are for a much lower force, so the parti
cles are actually much weaker.
An energy utilization factor U can be defined as the fraction of specific impact
energy that is actually used in the production of the largest F
m
in a multipleimpact
sequence. This factor will generally lie between 0 and 1. Equation 9 is now modified to
(EQ 12)
where and U
*
are values normalized to the median values, and
.
It was assumed that the distribution of U, F(U), can be fitted with a modified log
logistic function:
(EQ 13)
where U
*
= U/U
50
.
Appendix B shows how the effect of orientation can be extracted from the data,
using convolution of the distributions (Gardner and Austin 1975). A search was made
for the values of U
50
and ì that made the predicted distribution of forces agree as closely
as possible with the experimental values of Equation 5, and reasonable agreement was
obtained with U
50
= 0.51 and ì = 3.65. This result is also shown in Figure 2.
Figure 3 shows the cumulative fraction of impacts with an energy utilization factor
less than or equal to U. It is seen that the percentage of impacts with U = 1 is about 7 and
that more than 50% of the impacts have a U value of less than half a percent, meaning
that calculation of impact forces (and strengths) from the specific energy of impact,
assuming that U is 1, will give substantial errors. In fact, the values of U also contain the
errors generated by assuming that the impacts are comparable to those of equivalent
spheres.
F
m
0.757d
2
k
p
2 5
Uµv
2
( )
3 5
Ck
p
*2 5
U
*3 5
= =
k
p
*
C 0.757d
2
µv
2
( )
3 5
=
k
p50
2 5
U
50
3 5
F U
*
( )
1
1 1 U
*
( )
ì
+
, = U 1 <
1, = U 1 =
BREAKAGE AND DAMAGE OF PARTICLES BY IMPACT 211
THE ORI ENTATI ON EFFECT I N REPEATED I MPACTS
The discussion in the previous section described the impact forces, but the test results
also included the fraction of pellets broken after 1 impact and after 10 impacts, at several
impact velocities (Vervoorn and Austin 1990). In the traditional viewpoint, a strong par
ticle is always strong (except for damage accumulation), and the strength distribution is
of the strengths between particles. Vervoorn and Austin tried a different type of analysis
by assuming the opposite extreme; that is, the distribution of strengths was within each
particle due to different strengths in different impact orientations. In this viewpoint, a
particle that does not break in one impact at a given velocity has the chance to break in
the next impact of the same velocity. The equation that results (for a sufficiently large
number of impacted particles of a given size) is
(EQ 14)
where w(N) is the fraction left unbroken after N impacts, and S is the equivalent to a spe
cific rate of breakage, with units of fraction broken per impact event. The decrease in the
amount remaining unbroken at each step is given by the factor exp(–S) applied to the
previous value.
It seems much more likely that there is a distribution of strengths between the pel
lets plus a distribution of strengths within each pellet due to the orientation effect. This
would apply to any system in which random orientations of irregularly shaped particles
can occur in repeated impacts. Thus, the theoretical investigation (Austin 2004) of
grinding kinetics in a batch system was invalid because it did not include the orientation
effect. Further work is required in this area.
0.0
0.2
0.4
0.6
0.8
1.0
0 50 100 150
(a)
(b)
(c)
Maximum Impact Force F
m
, N
C
u
m
u
l
a
t
i
v
e
N
u
m
b
e
r
o
f
I
m
p
a
c
t
s
≤
F
o
r
c
e
F
m
FIGURE 2 Comparison of distributions of force
w N ( ) w 0 ( ) SN – ( ) w 0 ( ) , exp 1 N , 1 2 3 . , , , = = =
212 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
The important points to note are (1) that the usual methods of determining the “dis
tribution of strengths” do not differentiate between the two distributions but give a com
bined effect; and (2) the two different distributions will give entirely different results in
a mill model, so it is necessary to determine which one is applicable and to measure both
if they are relevant.
I MPACT TESTS AND THE KI NG TAVARES TEST
There are two major types of tests that have been used in recent years to measure the dis
tribution of strengths of particles to impact. The first type consists of firing a stream of
the particles at a target at known velocity, or of hitting suspended (sized) particles or a
stream of falling particles by a hammer moving at a known velocity (Vogel and Peukert
2003). The test is repeated at higher velocities to generate a distribution of strength
defined by the calculated specific impact energy. This type of test has again the advantage
of rapid testing of a large number of particles, but if the particles are not spheres, the cal
culated impact energies are not correct due to the orientation effect discussed above.
The second type of test is impact by a drop weight on a single particle using an
instrumented apparatus called a rapidresponse impact load cell (King and Bourgeois
1993). The test particle is laid resting flat on the end of a vertical steel rod (that acts as
the anvil) and can be held in place by a thin film of grease in order to limit relocation
during contact. The theory of wave propagation along a rigid cylinder is used to get a
measure of the force–time curve produced by the impact and, hence, the force–deformation
curve (Tavares and King 1998). This type of tester is capable of recording impacts of
duration in the range of tenths and hundredths of microseconds. The impact produces a
0.0 0.2 0.4 0.6 0.8 1.0
0.0
0.2
0.4
0.6
0.8
1.0
Energy Utilization Factor U – (–)
C
u
m
u
l
a
t
i
v
e
F
r
a
c
t
i
o
n
o
f
I
m
p
a
c
t
s
≤
U
FIGURE 3 The distribution of the energy utilization factor, as predicted by Equation 13
BREAKAGE AND DAMAGE OF PARTICLES BY IMPACT 213
planecompressive wave that moves down the rod at high speed and is detected by fast
response strain gauges placed far enough away from the contact surface to allow the
development of the plane wave. The output of the strain gauges is taken to a digital stor
age oscilloscope and later processed in order to give the force–deformation curve
(Tavares 1997).
Rounded or very regular particles of a given material and particle size were hand
selected for the test, as these shapes of particles conform closest to the assumptions in
data analysis, enabling the most accurate measurements of rigidity modulus and dam
age accumulation coefficients (discussed later in this paper).
An advantage of a singleparticle test of this kind is that the impact energy can be
made low enough to produce deformation without fracture, and then the particle can be
impacted again. Figure 4 presents the simulated results of such a test, which show that
the force–deformation curve is not reversible but returns to give a lower starting stiffness
modulus for the curve produced by the next impact.
Two important conclusions can be made. First, the stiffness factor decreases as the
particle is compressed by the impact. This is attributed to microcracking in the solid that
leaves the outline of the solid unchanged after the stress is removed. Second, the ability
of the solid to compress more readily (lower stiffness) is permanent. These two factors
require a reexamination of the theories of deformation and fracture.
The equation relating force and deformation for a sphere between anvils is, for a
constant stiffness modulus (particle stiffness) k
p
(Equation A6 with n = 2),
0 100 200 300 400 500
0
20
40
60
80
100
120
140
F
o
r
c
e
,
N
Starting κ
0
(Deformation, μm)
3/2
FIGURE 4 Force–deformation curves produced by repeated impacts on a particle, without fracture.
Model parameters: d = 4.4 mm, k
0
= 17 GPa, o
c
= 70 μm, ¸ = 9, and impact energy = 1.4 mJ.
214 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
(EQ 15)
where k
s
is the stiffness of the anvil, which in the case of steel is about 230 GPa.
The force–deformation curves of Figure 4 show that k
p
is not constant, so an appar
ent mean value of k
p
for the rising part of a curve, k, can be defined in Equation 15 by
replacing k
p
with k. The degree of “damage” was defined as a measure of the degrada
tion of particle stiffness (Kachanov 1958; Tavares and King 2002) as
(EQ 16)
where k
0
is the initial value at the start of the curve. The variation of k with deformation
o was fitted by the empirical equation
(EQ 17)
where o
C
is the critical deformation at the point of fracture, D
c
is the critical damage, and
¸ is the “damage accumulation coefficient.” A large value of this coefficient means that
damage accumulation is small until the deformation approaches the critical value,
whereas low values give large increases of damage as the deformation increases.
Rewriting Equation 15 as to allow for the variation of the stiffness factor (Equation 16)
and also considering that the stiffness of the anvil is typically reasonably higher than that
of the particle ( ) gives
(EQ 18)
where D
c
corresponds to the value of damage for which the derivative of Equation 18 is
equal to zero, giving D
c
= 3/(2¸ + 3).
The critical force to produce fracture is approximately given by
(EQ 19)
Figure 5 shows the force–deformation curves predicted by this equation up to the
point of fracture, for a single impact of sufficient energy to produce fracture.
Several important conclusions can be made from Figure 5. If the response curve of
the impact of a weight falling on a particle follows this pattern, then the values of the
three parameters k
0
, o
c
, and ¸ can be determined from the curve. The k
0
is not necessar
ily an exact value because it is a spherical equivalent value, but it is the variation of the
value that is important. The experimental value of the force required to produce disinte
grative fracture (when the value of stiffness factor tends to zero) can be predicted by the
parameters. It is not necessary to make the assumption (generally erroneous) that an
impact of a known specific energy can be used to predict the force. Thus, the “strength”
of the particle can be defined explicitly by this critical force. Alternatively, it is common
to use “strength” as the “specific impact energy required to produce fracture,” which
appears to be due to much of the work being done by mineral process engineers who are
familiar with kilowatthours per ton as an index of the difficulty of size reduction, or to
F
d
1 2
3

k
p
k
s
k
p
k
s
+

\ .

 
o
3 2
=
D 1 k k
0
– =
k k
0
1 D
c
o o
c
( )
¸
–   =
k
0
1 D – ( ) k
s
+ k
0
k
s
+ ~
F
d
1 2
3

k
s
k
0
k
s
k
0
+

\ .

 
1 D
c
o
o
c

\ .
 
¸
– o
3 2
=
F
c
d
1 2
k
s
k
0
1 D
c
–  
3 k
s
k
0
+ ( )
o
c
3 2
=
BREAKAGE AND DAMAGE OF PARTICLES BY IMPACT 215
the use of experimental techniques where the actual force cannot be measured directly.
The relationship between the deformation and the specific strain energy E is obtained by
integrating Equation 18, which gives
(EQ 20)
where m
p
is the particle weight, and n the number of contact points (1 or 2).
In this test there is no orientation effect because the particle cannot twist and reimpact
during impact. It is only necessary to test enough particles to encompass different shapes
and measure the actual fracture force each time. Of course, the equipment is complex,
and testing particles one at a time is very time consuming, but the results are far more
informative than those from other types of tests.
It will be noted from Equation 18 that the influence of the damage accumulation
coefficient on the critical force (strength) becomes small for large values of the coeffi
cient as 3/(2¸ + 3) becomes small compared to 1. However, if an impact is not sufficient
to cause disintegrative fracture, then the particle is damaged and will be weaker for
another impact. Tavares and King (2002) show that the calculation of the decreasing
strength with each impact depends on the value of ¸, and it is essential to know this
value, even if each amount of damage is small, in order to estimate how many impacts
will cause disintegrative fracture.
0 100 200 300 400 500 600 700 800
0
40
80
120
160
200
240
α
c
3/2
Fracture
γ = 9
Fracture
γ = 90
F
o
r
c
e
,
N
Starting κ
0
(Deformation, μm)
3/2
FIGURE 5 The force–deformation curves predicted for impact on a sphere of 4.4mm diameter,
k
0
= 17 GPa, o
c
= 70 μm
E
2
3 1–n 2 ( )
d
1 2
k
s
k
0
3m
p
k
s
k
0
+ ( )

2
5
 D
c
o
o
c

\ .
 
¸
– a
5 2
=
216 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
EXPERI MENTAL RESULTS
Table 1 shows a set of data for a material that is damaged relatively easily (low ¸), with
the data presented in the order of decreasing strength. There are several important con
clusions. It might be argued that the distribution of stiffness factor seen in the pellet tests
discussed previously is due to variability in the manufacture of the pellets by pressing
and sintering. However, the particle data also show a wide distribution of the factor. This
can be interpreted as a combination of natural material variability and the damage sus
tained as the test particles are produced (usually by crushing). A given size of particle
can originate from a single breakage or it might result from a sequence of breakages, so
the damage prehistory will be different. A particle that starts with a low value of k will
reach k tending to zero sooner in a compression, other factors being equal. On the other
hand, particles with a high value of ¸ will resist further damage and appear stronger. Any
“theoretical” treatment that assumes that particles have a single value for the mechanical
properties is clearly incorrect.
The data also show that there is little correlation between the initial k values and the
¸ values. This implies that the prehistory of the particles controls the starting stiffness
factor, whereas further damage occurs independently of the existing damage. For this set
of data, it can be shown that there is a weak and scattered relation between the critical
deformation and the starting stiffness factor. It appears that lower stiffness is associated
with a larger critical deformation, so that some degree of damage enables the material to
deform more before it breaks.
Equation 20 shows that the strength is affected by each of the three parameters.
There is an important finding that results when a particular value of strength can arise
from different combinations of the three parameters (if they are not strongly correlated
and are almost independent of one another). When such particles are repeatedly
impacted, they will behave differently, and a given number of impacts at a given specific
impact energy may break some of the particles and not others, even though they have
the same strength to start (Tavares and King 2002). This invalidates treatments (e.g.,
Austin 2004) that assume that particles of the same strength behave identically when
repeatedly impacted.
The damage accumulation coefficient acts in two ways. If it is relatively small, the
damage accumulates gradually during the impact and the critical force (or specific
impact energy) that is required to produce fracture depends on the value. If the value is
large, there is little effect until the deformation approaches the critical value, and the
value of the critical force has a negligible dependence on the damage accumulation coef
ficient. Thus the most accurate way of determining ¸ when it is large is by analyzing
results from repeated lowenergy impacts (Tavares and King 2002), although this is a
timeconsuming test compared to the singleimpact test.
The results from data similar to those in Table 1 are shown in Table 2 for different
mineral particles. The cumulative distribution of ¸ was fitted approximately to the log
logistic function:
(EQ 21)
where ¸
50
is the median value.
Although there is no strong correlation between ¸ and k
0
seen in the individual data
sets, because each particle has its own starting stiffness factor, Figure 6 shows that ¸
50
is
strongly correlated with F
m50
. Material that is strong on the average is more difficult to
damage on the average. However, it must be remembered that individual particles have
their own ¸ values, to be calculated from ¸
50
and F
m50
.
F ¸ ( )
1
1 ¸
50
¸ ( )
ì
ì
+
 =
BREAKAGE AND DAMAGE OF PARTICLES BY IMPACT 217
Additionally, a comparison can be made between the distribution of strengths from
impact load cell tests (n = 2) (Tavares 1997) and impact on a target (n = 1) (Vogel and
Peukert 2003) on irregularly shaped particles. It shows that the target impact data give
distributions that have consistently higher maximum strengths than the impact load cell
tests, which is in agreement with what we have argued previously.
FRAGMENT DI STRI BUTI ONS
It is well understood that the description of the strength behavior of a collection of parti
cles is not generally sufficient for the purposes of predicting mill performance because it
is also necessary to describe the distribution of fragments formed on the breakage. This
is easily done in those tests where many particles are broken and the collected fines can
be sized. However, it is clear that the result will be the mean of many different types of
breakage and it will be difficult to predict the values for other breakage processes. It is
possible to collect and analyze the fragments from (automatically wellcharacterized)
TABLE 1 KingTavares test data for marble (4.0–4.75 mm)
Maximum Impact Force
F
m
, N
Initial Stiffness
Factor k
0
, GPa
Damage Accumulation
Coefficient ¸ (–)
Critical Deformation
o
c
, μm
190 34 12 47.6
186 26 20 53.4
171 31 10 47.3
170 28 11 49.5
164 26 6.5 54.9
161 21 11 57.7
160 21 10 58.7
152 21 9.0 55.6
144 25 8.0 49.2
129 23 6.5 51.8
125 20 7.0 51.5
123 18 8.5 54.4
123 20 9.0 49.9
110 16 7.0 54.6
89 15 6.0 50.5
87 14 9.0 49.9
85 11 9.5 57.6
78 9.5 6.0 63.5
67 7.5 5.0 69.4
60 8.2 7.7 57.7
TABLE 2 Results from the KingTavares test on particles with size 4.00–4.75 mm
Material
Median Impact
Force F
m50
, N
Median Damage Accumulation
Coefficient ¸
50
(–)
Damage Accumulation
Distribution Factor ì
¸
(–)
Heattreated quartz 105 10.0 2.6
Marble 125 8.5 6.0
Iron ore 180 15.0 2.6
Limestone 185 31.5 2.4
Cement clinker 360 45.0 2.35
Quartz 540 68.0 1.65
218 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
individual particle breakage in the KingTavares test, but only for large particles. The test
work is tedious and has not yet been done in sufficient detail to describe the laws con
trolling the fragment distribution.
There is also an additional complication. When the impact energy is greater than
that required to cause the primary disintegrative fracture, then the remaining energy
may, under some circumstances, be used to break the resulting fragments. It has been
suggested (Austin 2002) that this can be treated as a repetitive breakage process, and
the appropriate algorithm has been formulated, but the assumptions involved have not
been rigorously tested.
CONCLUSI ONS AND FUTURE WORK
The brittle breakage model based on the damage mechanical theory has two extremes.
At low values of the damage accumulation coefficient, a particle undergoes irreversible
damage by microcracking (the stiffness factor decreases) as the particle is compressed by
the impact. At a critical value of deformation, the accumulated damage makes the parti
cle so weak (the stiffness factor tends to zero) that the particle disintegrates. On the
other hand, at high values of the damage accumulation constant, the particle suffers lit
tle damage until the critical deformation is approached and the stiffness factor drops
suddenly to zero. This second extreme is close to the conventional theory of brittle frac
ture, with the critical deformation producing crack propagation from a Griffith flaw with
a critical tensile stress. In both cases, repeated impact at levels too low to give fracture
will produce weaker and weaker particles until fracture occurs.
In general, it is necessary to know the initial stiffness factor, the critical deformation,
and the damage accumulation coefficient in order to define the failure properties of any
given particle. In an assemblage of closely sized particles of natural materials such as
Median Damage Accumulation Coefficient
0
100
200
300
400
500
600
M
e
d
i
a
n
I
m
p
a
c
t
F
o
r
c
e
,
N
0 10 20 30 40 50 60 70 80
FIGURE 6 Correlation of median damage coefficient with median force required to cause fracture
(data from Table 2)
BREAKAGE AND DAMAGE OF PARTICLES BY IMPACT 219
mineral ores, coals, and so forth, there will be distributions of these three controlling
parameters, which will vary with particle size. Knowing these parameters enables the
distribution of strengths to be calculated and the failure behavior of the assemblage
under specified impact conditions, including repeated impacts, to be predicted. There
are no mathematical difficulties in the calculations.
Tests that measure only the distribution of strengths are of limited use because the
distributions of the three controlling parameters cannot be deduced from that informa
tion alone. Future advances in understanding will come only from tests that use appro
priate load cells to measure the fundamental material properties.
It is not yet clear whether the damage accumulation coefficient derived from a single
destructive impact is the same as that derived from repeated small impacts. In repeated
impacts, the compressive stress is relieved after each impact, and this could lead to
increased damage by microcrack extension.
The Hertzian solutions for ideal impact of spheres can be used to guide the research,
but they must be used with caution. Impacts of nonspherical particles can occur with an
inefficient transfer of the impact energy to the primary breakage, so calculations based
on the specific energy of impact must allow for the energy utilization factors described in
this paper. It is difficult to imagine the concept of contact circle of impact applying with
exactitude to the variety of impacts of irregularshaped mineral particles.
This paper has attempted to define the problems to be solved rather than to present
detailed investigations of specific systems. It is clear that a great deal of work remains to
be done, involving accurate and detailed investigations of different sizes of different
materials.
REFERENCES
Austin, L.G. 2002. A treatment of impact breakage of particles. Powder Technology 126:
85–90.
———. 2004. The effect of damage on breakage kinetics. Powder Technology 143–144:
151–159.
Austin, L.G., and F.F. Aplan. 1998. The powder technology of standard grinding tests:
Part I, ball milling and rollerrace milling. In Fine Powder Processing Technology.
Edited by R. Hogg, R. Cornwall, and C.C. Huang. University Park, PA: Pennsylvania
State University.
Austin, L.G., K. Shoji, and M.D. Everell. 1973. An explanation of abnormal breakage of
large particle sizes in laboratory mills. Powder Technology 7(1):3–7.
Austin, L.G., P.M. Trubelja, and B. Scarlett. 1993. A study of the fracture of pellets fired
against a target. Particle and Particle Systems Characterization 10:347–352.
Bond, F.C. 1954. A twodimensional phenomenon in the breakage of coal. Fuel 33(2):249.
———. 1960. Crushing and grinding calculations. British Chemical Engineering 6:378–391,
543–548.
Gardner, R.P., and L.G. Austin. 1975. The applicability of the firstorder grinding law to
particles having a distribution of strengths. Powder Technology 12(1):65–69.
Goldsmith, W. 1960. Impact: The Theory and Physical Behavior of Colliding Solids. London:
Edward Arnold.
Griffith, A.A. 1921. The phenomena of rupture and flow in solids. Series A. Philosophical
Transactions of the Royal Society 221:163–198.
220 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
Hiramatsu, Y., and Y. Oka. 1966. Determination of the tensile strength of rock by a com
pression test of an irregular test piece. International Journal of Rock Mechanics and
Mining Sciences 3:89–99.
Jaeger, J.C., and N.G.W. Cook. 1969. Fundamentals of Rock Mechanics. 1st edition. London:
Methuen.
Kachanov, L.M. 1958. Time of the rupture process under creep conditions (in Russian).
Izv. Akad. Nauk AN SSSR 8:26–31.
King, R.P., and F.S. Bourgeois. 1993. Measurement of fracture energy during single
particle fracture. Minerals Engineering 6:353–368.
Schönert, K. 1986. Advances in the physical fundamentals of comminution. Pages 19–32
in Advances in Mineral Processing. Edited by P. Somasundaran. Littleton, CO: SME.
Tavares, L.M. 1997. Microscale investigation of particle breakage applied to the study of
thermal and mechanical predamage. Ph.D. dissertation, Salt Lake City, UT: Univer
sity of Utah.
Tavares, L.M., and R.P. King. 1998. Singleparticle fracture under impact loading. Inter
national Journal of Mineral Processing 54:1–28.
———. 2002. Modeling of particle breakage by repeated impacts using continuum damage
mechanics. Powder Technology 123:138–146.
Vervoorn, P.M.M. 1986. Particle Attrition. International Fine Particle Research Institute
Final Report FRR 13.01. Netherlands: Delft University of Technology.
Vervoorn, P.M.M., and L.G. Austin. 1990. The analysis of repeated breakage events as an
equivalent rate process. Powder Technology 63:141–147.
Vervoorn, P.M.M., and B. Scarlett. 1990. Particle impact testing. Pages 195–204 in Pro
ceedings 7th European Symposium Comminution. Ljubljana, Yugoslavia.
Vogel, L., and W. Peukert. 2003. Breakage behavior of different materials—construction
of a mastercurve for the breakage probability. Powder Technology 129:101–110.
Weibull, W.A. 1939. A statistical theory of the strength of materials. Ingvetenskakad
Handl 151:5–45.
Weichert, R. 1992. Application of defect statistics and fracture mechanics for describing
comminution processes. ZementKalkGips 45:51–57.
BREAKAGE AND DAMAGE OF PARTICLES BY IMPACT 221
APPENDIX A
Hertzian Solutions for Impact of Spheres
From the potential theory (Goldsmith 1960) the pressure distribution in the surface of
contact of two curved bodies is given by
(EQ A1)
where o represents the decrease under pressure of the distance along the normal from
the center of the sphere through the center of the contact circle to a point in the anvil dis
tant from the contact, and r is the radial distance from the center of the contact circle.
The value of o is
(EQ A2)
Substituting Equation A2 in Equation A1 gives
(EQ A3)
The radius of the contact circle, a, is obtained from
(EQ A4)
where F is the applied load. Substituting Equation A2 in Equation A4
(EQ A5)
This is the constitutive equation describing onepoint contact of curved surfaces.
Note that Hertzian contact theory predicts a nonlinear relationship between load and
deformation, which is the result of the fact that the contact area changes continuously
with the applied load. For the case of two contact points (such as that of a sphere com
pressed between two platens), o must be halved in Equation A5. Therefore, the general
constitutive equation from the Hertz theory that describes the impact of a sphere against
a target (n = 1) or the impact of a sphere between anvils (n = 2) is
(EQ A6)
with K
e
= k
p
k
s
(k
p
+ k
s
) and K
g
= d/2.
p r ( )
K
e
t

o
a
2
r
2
–
 –
r
2
K
g
a
2
r
2
–

a
2
r
2
–
K
g
 – +
\ .


 
=
o
a
2
K
g
 =
p r ( )
2K
e
tK
g
 a
2
r
2
– – =
F
4a
3
K
e
3K
g
 =
F
4
3
 K
e
K
g
1 2
o
3 2
=
F
2
3 1–n 2 ( )
3
 d
1 2
K
e
o
3 2
=
222 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
Integrating Equation A6 up to a given deformation, o, gives the elastic energy stored
(EQ A7)
The relationship between the massspecific elastic strain energy stored in the spheri
cal particle and the force is given by substituting Equation A6 in Equation A7 is
(EQ A8)
Further, the relationship between the maximum force and the impact velocity at
onepoint contact (n = 1 in Figure 1) is given by
(EQ A9)
given that E = v
2
/2.
The maximum tensile stress occurs at the circular boundary of the surface of contact
(r/a = 1)
(EQ A10)
This can be rewritten as a function of specific impact energy (E) as
(EQ A11)
From the definition of particle strength (Equation 3), it is possible to show that
(EQ A12)
elastic strain energy
2
3 1–n 2 ( )+1
15
 d
1 2
K
e
o
5 2
=
F 0.90 2
3 1–n ( ) 5
K
e
2 5
E
3 5
d
2
=
F
m
0.757d
2
K
e
2 5
µv
2
( )
3 5
=
o
T
2 4µ –
t

F
9

K
e
d

\ .
 
2
1 3
=
o
T
0.32 1 2µ – ( ) 2
1 n – ( ) 5
( )K
e
4 5
µE ( )
1 5
=
o
p
0.678 2
3 2 n – ( ) 5
( )K
e
2 5
µE ( )
3 5
=
BREAKAGE AND DAMAGE OF PARTICLES BY IMPACT 223
APPENDIX B
Allowance for Energy Utilization Factors
The distribution of stiffness can be described by (from Equation 8)
(EQ B1)
Differentiating with respect to
(EQ B2)
Consider a particular value of F
m
. For in the range to all impacts with
(EQ B3)
will give maximum force s F
m
. As U
*
has a maximum of 1/U
50
when U = 1, the maximum
for this condition is
(EQ B4)
Then the total fraction of impacts with maximum force s F
m
is
(EQ B5)
Equation 14 gives the distribution of U
*
. Using this and Equations B2, B3, and B4 in
Equation B5 becomes,
(EQ B6)
This integral can be calculated numerically for any value of F
m
, with U
50
and ì as
adjustable constants, since k
p50
and ì
1
are known and values of d, µ, and v are necessary
to define the impact conditions.
F
1
k
p
*
( )
1
1 1 k
p
*
( )
ì
1
+
 =
k
p
*
dF
1
k
p
*
( )
dk
p
*

ì
1
1 k
p
* ì
1
–
+ ( )
2
k
p
*1 ì
1
+
 =
k
p
*
k
p
*
k
p
*
dk
p
*
+
U
*
k
p
*2 5
C
F
m

\ .

 
5 3
s
k
p
*
k
m
*
F
m
C

\ .
 
5 2
1
U
50

\ .
 
=
G F
m
( )
dF
1
k
p
*
( )
dk
p
*

0
k
m
*
)
F
2
U
*
( )dk
p
*
=
G F
m
( )
ì
1
1 k
p
* ì
1
–
+ ( )
2

0
k
m
*
)
1
1 k
p
*1 ì
1
+
+

\ .

 
1
1
k
p
*2 3
F
m
C

\ .

 
ì
+
dk
p
*
=
225
The Rationale behind the Development
of One Model Describing the Size
Reduction/Liberation of Ores
Ronald L. Wiegel
*
ABSTRACT
A.M. Gaudin espoused the use of the Rubik’scube arrangement of mineral grains to concep
tually describe binary mineral liberation. That idealized concept has been examined, modi
fied, tested, and extended over the past 40 years, resulting in a useful quantitative model for
describing the liberation of mediumgrade ores. This paper describes the background and
rationale for accepting this approach for modeling mineral liberation as part of an overall
mineral process simulation capability. As much of the size reduction used for liberation is
aimed at lowergrade mineral deposits, comments have been included concerning extending
the approach to lowgrade ores.
I NTRODUCTI ON
A Visualization of Mineral Liberation Due to Size Reduction
In his 1939 text, Principles of Mineral Dressing, A.M. Gaudin described a conceptual
model for mineral liberation caused by the threedimensional geometric effects of size
reduction in a binary mineral system (Gaudin 1939). This idealized conceptual model is
admittedly an extreme simplification of a real mineral system, as it assumed that the
mineral grains of both mineral species were of the same uniform cubic shape and were
aligned adjacent to each other, such that the eight corners of the cube were touching
other cubes’ corners. This then provides intimate surfacetosurface, edgetoedge, and
cornertocorner contacts, similar to what has more recently become known as a Rubik’s
cube arrangement. Further, when Gaudin’s binary mineral grain assemblage was broken
in a size reduction event, the resultant fracture plains ran parallel to the mineral grain
surfaces and produced uniformly sized cubic particles. The composition of the individual
resultant particles depends upon whether the grain fragments that constitute it are all of
the same mineral species or are of different species. In the first case, there would be lib
erated particles of either waste or value, and in the second, there would be locked mid
dling particles of a wide range of compositions.
The author believes the original purpose of Gaudin’s conceptual model was to convey
the ideas that help one to understand qualitatively how the composition of the particles
produced by size reduction relate to the original mineral grain size, the particle size, and
* Mineral Processing Consultant, Lakeland, Florida
226 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
the location of similar or dissimilar grains adjacent to one another in the original grain
assemblage being broken. There were some quantitative relationships provided with
Gaudin’s description, but because special situations were required to cause these rela
tionships to be valid, little attention was paid to their possible use.
REVI EW OF PREVI OUS WORK ON THE GAUDI N MODEL
Converting a Qualitative Model to a Quantitative One
The Fully Liberated Minerals. The Gaudin liberation model was reviewed and
revived in the 1960s (Wiegel 1964, 1965; Wiegel and Li 1967) with the purpose of devel
oping more general quantitative relationships to describe the liberation effects of size
reduction. Although the same idealized conceptual model was used, the idea of a ran
dom location of the two ordered mineral grain species was introduced and was defined
by binary probability concepts. It also was assumed that the uniformly spaced fracture
planes, which run parallel to the mineral grain surfaces, are equally likely to occur at any
position in the mineral grain. This then permitted the derivation of the following fairly
complex relationships, which are valid over the entire size spectrum, for the proportion
of liberated waste particles (Equation 1), liberated values particles (Equation 2), and
locked waste and values particles (Equation 3) as a function of the mineral grain size,
particle size, and the original ore’s volumetric fraction of values (feed grade). At that
time, the newly termed Gaudin Random Liberation Model (GRLM) was still a mathemat
ical curiosity, with no proven usefulness. It defined the analytical relationships for an
idealized binary mineral system among the quantities of liberated and locked minerals
and two liberation parameters, the volumetric proportion of value in the ore and the
ratio of mineral grain size to particle size. A plot of these liberation relationships for the
case where the original ore contains 0.25 volume fraction of values is shown in Figure 1.
(EQ 1)
(EQ 2)
PAB = 1 – PA – PB (EQ 3)
/o = t + e = 1/K (EQ 4)
where
PA = fraction of particles by volume of liberated waste mineral
c = fractional remainder in ratio /o
VA = volume fraction of waste mineral in original feed
t = largest integer in ratio /o
PB = fraction of particles by volume of liberated values mineral
VB = volume fraction of values in original feed
PAB = fraction of particles by volume of locked values and waste
 = particle size (linear dimension)
o = mineral grain size (linear dimension)
K = grain size–particle size ratio
Conceptually, when carrying out the GRLM calculations, there is only one particle size
into which the mineral grain aggregate is broken. In reality, there is an entire distribution
PA 1 c – ( )
3
VA
t+1 ( )
3
3c 1 c – ( )
2
VA
t+1 ( )
2
t+2 ( )
3c
2
1 c – ( )VA
t+1 ( ) t+2 ( )
2
c
3
VA
t+2 ( )
3
+
+ +
=
PA 1 c – ( )
3
VB
t+1 ( )
3
3c 1 c – ( )
2
VB
t+1 ( )
2
t+2 ( )
3c
2
1 c – ( )VB
t+1 ( ) t+2 ( )
2
c
3
VB
t+2 ( )
3
+
+ +
=
RATIONALE FOR ONE MODEL DESCRIBING THE SIZE REDUCTION/LIBERATION OF ORES 227
of particle sizes resulting from size reduction. The way the GRLM has been used in simu
lations is to calculate liberation results for the log mean particle size for each individual
screen fraction. One is therefore making the assumption that the particles in a resultant
narrow size range have a composition distribution spectrum similar to the uniform, cubic
particles generated by the GRLM calculations.
Comparison of the Model with Reality—Magnetic Iron Formations. The underlying
reason for developing more quantitative liberation relationships was due to an interest
in better explanations and possible control of the rejection of waste by magnetic separa
tion following the several stages of size reduction in the magnetic taconite process,
which was receiving a great deal of process engineering interest at that time. To see if
the GRLM had a potential quantitative use, despite its many idealized assumptions, a
suite of crude magnetic iron ore samples were collected of the feed and product from the
initial sizereduction step in 12 magnetite concentrators operating around the world
(Wiegel 1975; Lynch 1977). These individual product samples were screened into closely
sized fractions and concentrated in laboratoryuse Davis tube separations, with the
weighed concentrate and tail products subjected to Satmagan measurements for magne
tite, specific gravity measurements, and wet chemical analyses for gangue components.
It should be pointed out that the Davis tube, when used on small quantities (10–20 g)
of closely sized particles for reasonable testing times (10–20 min), is an extremely effi
cient separator of particles containing magnetite from those that are nonmagnetic. The
relatively close particle size all but eliminates the tendency for strings of magnetic parti
cles to entrap nonmagnetic particles, such as that which happens routinely in commercial
magnetic separators, where particlesize distributions are wide and separation time is
minimal. The Davis tube results are therefore a very good indication of the quantity of
particles containing magnetite (liberated magnetite and locked magnetite and waste)
versus those that have no magnetite (liberated waste).
1.0
0.8
0.6
0.4
0.2
0.0
0.1 1 10 100
Grain to Particle Size Ratio
C
u
m
u
l
a
t
i
v
e
F
r
a
c
t
i
o
n
b
y
V
o
l
u
m
e
Liberated Waste
Locked Values and Waste
Liberated Values
FIGURE 1 Fraction locked and liberated particles versus grain size–particle size ratio for an ore
with 0.25 fraction values
228 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
A comparison of the shape of the concentrategrade versus particlesize results for
the Davis tube—with GRLM calculations assuming total capture of magnetitecontaining
particles in the concentrate and total rejection of nonmagnetite particles in the tailing
when converted to weight units—provided support for the potential use of the GRLM as a
quantitative tool for the explanation of liberation, at least for magnetic taconites. It also
was possible to obtain an estimate of the “effective” mineral grain size for each magne
tite iron ore source by best fitting the calculated GRLM plots to the experimental separa
tion data. Concentrategrade data typical of these samples are shown in Figure 2, versus
those calculated using the GRLM. The concentrate quantity versus particlesize results
for these tests were not as encouraging; they were somewhat distorted by the tendency
for the magnetite and the gangue minerals to grind at different rates.
Gradual Liberation by Batch Grinding. An examination of the use of the GRLM in
describing the gradual size reduction/liberation of an ore was made (Wiegel 1976a)
when a series of laboratorybatch grinds were carried out on a coarsegrained magnetic
iron formation sample, with the product size fractions subjected to Davis tube separa
tions and analyses. A portion of these concentrategrade and quantity results are shown
in Figures 3 and 4, for grinds of 0, 1, 2, 5, 10, and 20 minutes’ duration. In this case,
there was an indication that some differences existed in the grinding rates of the waste
and values species, as evidenced by trends in magnetite content increasing with smaller
particle size, indicating that the magnetite was ground more rapidly than the waste. A
useful concept that also came out of this batchgrinding study was the definition of a
directional coefficient, which described the proportion of material that started in a spe
cific size/composition location and proceeded as a result of size reduction to a finer size/
composition location. This concept is useful in tracing what happens to particles when
modeling simultaneous size reduction and mineral liberation.
(EQ 5)
(EQ 6)
where
V(I, J) = quantity of particles of size J, composition I
Q(I, II, J) = directional coefficient, going from composition I to II
I = composition index in beginning particle size range, J
II = composition index in next finer particle size range, J + 1
The GRLM was incorporated into an early simulation program for a magnetic taco
nite concentrator (Wiegel 1976b, 1979), which included models for hydrocycloning, fine
screening, closedcircuit grinding, and multidrum magnetic separation.
Expansion of the Gaudin Random Liberation Model
For some 20 years (1980 to 2000), there was little interest shown in the use of the
GRLM, and no attempts were made to pull more information from the potential quanti
tative aspects of the model. There were also no further efforts to develop liberation infor
mation from the Davis tube separation of magnetite ores, despite the advantage of an
almost perfect separation of magnetic from nonmagnetic particles. Rather, the general
emphasis of liberation studies shifted to an interest in ores, where liberation information
was determined from imageanalysisrelated techniques (Barbery 1991; Schneider 1995).
V I J , ( )Q I II J , , ( )
I
_
V II J 1 + , ( ) =
Q I II J , , ( )
II
_
1 =
RATIONALE FOR ONE MODEL DESCRIBING THE SIZE REDUCTION/LIBERATION OF ORES 229
In the mid1990s, a renewed interest in iron ore mineral process modeling and sim
ulation was underscored by the success of one magnetic taconite concentrator in achiev
ing a 34% increase in capacity and a 26% reduction in grinding energy consumption per
unit of production (Wennen, Nordstrom, and Murr 1997). This improvement was attrib
uted to significant, but not extraordinary, process changes resulting from a combination
of smallscale pilot testing, computer modeling of portions of the concentrator flow
scheme, and critical evaluation and analysis of the test results.
20
25
30
35
40
45
50
55
60
65
70
75
1 10 100 1,000 10,000 100,000
D
a
v
i
s
T
u
b
e
C
o
n
c
e
n
t
r
a
t
e
G
r
a
d
e
,
%
m
a
g
F
e
b
y
w
t
GRLM Plot
GRLM Plot
Sample 8
Effective Grain Size: 32 μm
Feed Grade: 30%
Sample 6
Effective Grain Size: 1,200 μm
Feed Grade: 45%
Particle Size, μm
FIGURE 2 Davis tube concentrate grade versus particle size
0
10
20
30
40
50
60
70
80
90
100
10 100 1,000 10,000
0 min
1 min
2 min
5 min
10 min
20 min
GRLM
D
a
v
i
s
T
u
b
e
C
o
n
c
e
n
t
r
a
t
e
G
r
a
d
e
,
%
F
e
3
O
4
b
y
v
o
l
.
CoarseGrained Magnetite
Liberation Parameters:
Effective Grain Size: 1,200 μm
Mineralized Ore: 30% Fe
3
O
4
Additional: 25% Barren Dilution
Particle Size, μm
FIGURE 3 Davis tube concentrate grade for ground samples
230 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
Measuring Liberation Parameters. The mineral process modeling and simulation
software packages, which had entered the marketplace in the late 1980s, did not have
either a mineral liberation or magnetic separation modeling capability, so efforts began
at the University of Minnesota’s Coleraine Minerals Research Laboratory to determine
what might fill these gaps. As a result of the earlier encouragement provided by the
GRLM, a new and more detailed look was taken at what additional information could be
gleaned from the idealized system. As mentioned above, with an appropriate choice of
“effective” mineral grain size, the GRLM provided a quantitative estimate of the amount
of either liberated waste or liberated values, with the remainder being locked particles
covering the entire range of compositions. As a step towards obtaining better quantita
tive measures of those parameters directly affecting mineral liberation, a leastsquares
computer program was written to obtain a best fit of the already recognized variables of
“effective mineral grain size” and original mineralized crude ore feed grade. It also was
realized that an additional parameter of barren waste dilution provided a significant
reduction in the residual errors. This dilution, which was assumed to be of the same
composition as the waste mineral, could be visualized either as a true dilution that takes
place during mining or as a mathematical adjustment for the fact that—at least in the
case of many iron formations—the iron oxide mineral is layered into higher and lower
grade bands (Wiegel 1999a). The GRLM information displayed in Figures 3 and 4 use
three liberation parameters: volume fraction of values in the mineralized ore, effective
mineral grain size, and barren waste dilution.
Quantifying Locked Particles. It was found that a computer simulation of the
breakage of the assemblage of cubic grains of the binary mineral aggregate in the GRLM,
with a proper randomized manipulation of the spacing of the breakage grid, could be
used to generate information on the quantity and composition of the locked and liber
ated particles for a specific mineral grain size–particle size ratio and original ore grade
(Wiegel 1999b). These breakage simulations for a particular product size were carried
out with a minimum of 200,000 particles being created, to ensure reasonable statistical
0
10
20
30
40
50
60
70
80
90
10 100 1,000 10,000
D
a
v
i
s
T
u
b
e
C
o
n
c
e
n
t
r
a
t
e
Q
u
a
n
t
i
t
y
,
%
F
e
e
d
v
o
l
.
Particle Size, μm
0 min
1 min
2 min
5 min
10 min
20 min
GRLM
FIGURE 4 Davis tube concentrate quantity versus particle size
RATIONALE FOR ONE MODEL DESCRIBING THE SIZE REDUCTION/LIBERATION OF ORES 231
confidence while not requiring excessive simulation time. The particle composition spec
trum was divided into 12 volume classes: liberated waste at 0% values, 10 locked particle
classes with ranges increasing in 10% increments, and liberated values at 100% values.
Each particle created by the simulation was placed in its appropriate composition slot.
For a given volumetric feed grade, separate simulations were carried out for grain size–
particle size ratios in the range of 0.125 to 32, with size ratios increasing by a factor of
the fourth root of 2 (1.189), providing roughly 30 sets of observations for each feed
grade. Similar sets of simulations were carried out for feed grades of 5%, 15%, 25%,
35%, and 45% values by volume. The higher feedgrade results can be calculated from
these due to the symmetry introduced by having only two mineral components, values
and waste. The composition range midpoint was chosen as a reasonable approximation
to the mean particle composition for each of the ranges resulting from the breakage
simulations.
The detailed simulation results for the GRLM provide an improved mathematical or
statistical insight into how quantities of locked particles behave as they experience size
reductioninduced mineral liberation. It has hitherto proven difficult to visualize, let
alone quantify, the gradual change that takes place in composition as the particle’s
dimensions are broken to finer and finer sizes. Although some appreciation can be
gained by examining microscopic images before and after breakage, there are so many
variables that can affect imaging results for a specific sample that it is impossible to gain
a quantitative understanding of the overall subject. As an example of the additional
information provided by these locked particle breakage simulations, the GRLMcalculated
results related to complete liberation of values and waste, as presented in Figure 1, are
complemented by the inclusion of the comparable locked particle quantity and composi
tion information as shown in Figure 5. The mineral grain aggregate starts out containing
0.25 volume fraction of values at a particle size, which contains in excess of 700 grains
and grain fragments of values and waste. As the linear size dimension is reduced from
8 grains to 4, to 2, there is a very gradual disappearance of the original feed particles in
the 0.2–0.3 composition range and a gradual appearance of 0.1–0.2 and 0.3–0.4 particles.
It is not until the particles approach the size of a single grain that a measurable quantity
of liberated waste appears, and it is not until particles reach about one tenth of the grain
size that there is appreciable liberation of both values and waste. This particle size
approximates the situation of one grain being broken into about one thousand particles.
As an initial approximation, it seemed reasonable to assume that the same distribution
of particle compositions would be generated by breakage from a specific size/composition
range to a finer specific size, regardless of how the particles had entered into that starting
size/composition range. That, in essence, states that the particles’ progeny is only depen
dent on the particles’ size and composition and not on the particles’ previous history.
GRLM lockedparticle simulation results have provided a way for one to approach
quantitatively, an explanation of the process that takes place when achieving gradual lib
eration as a result of size reduction. Although volumetric balances dictate the require
ment for conservation of overall volume and of the two individual mineral components’
volumes, in passing from one particle size into the next smaller particle size, it is still not
possible to calculate directly the portion of the volume in a specific composition range
that moves to each composition range in the next finer particle size. Only in the case of
the breakage of liberated values and waste particles do we know which composition slots
they fit into, as particles that are once liberated stay liberated.
One Approach to Solving the Distribution Problem for Locked Particle Breakage. The
use of the totalvolume and componentvolume balances in Equations 5 and 8, together
with the use of the earlier mentioned directional coefficients of Equation 6, permit one
to create a linear programming (LP) problem of this distribution phenomena (Wiegel
232 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
2000). Although, in LP there are many more unknowns (directional coefficients) than
linear equations, it is frequently possible to obtain a solution that satisfies a stated objec
tive criteria, which in this case was the maximization of the directional coefficients that
retain material in the same composition class as particle size is reduced, shown by Equa
tion 7. The relationships for the LP are as follows:
Objective criteria:
(EQ 7)
From Equation 5, total volume balances for II = 1 to 12
From Equation 6, directional coefficient balances for I = 1 to 12:
Values volume balances for I = 1 to 12:
where MV(I) = mean value of a composition range I.
1.0
0.8
0.6
0.4
0.2
0.0
0.1 1 10 100
GraintoParticleSize Ratio
C
u
m
u
l
a
t
i
v
e
F
r
a
c
t
i
o
n
b
y
V
o
l
u
m
e
Liberated Waste
Liberated Values
1.0
0.4
0.2
0.3
0.1
0.0
FIGURE 5 Fraction liberated and locked particles with compositions versus grain size–particle
size ratio for an ore with 0.25 fraction values
Q I I J , , ( )
I
_
maximum =
V I J , ( )Q I II J , , ( )
I
_
V II J 1 + , ( ) =
Q I II J , , ( )
II
_
1 =
V I J , ( ) Q I II J , , ( )MV II ( )
II
_
V I J , ( )MV I ( ) =
RATIONALE FOR ONE MODEL DESCRIBING THE SIZE REDUCTION/LIBERATION OF ORES 233
Or simplified:
(EQ 8)
In addition, there are 24 directional coefficients with values that are restricted to 1
or 0, by the fact that once a particle is liberated, it will remain liberated through all sub
sequent sizereduction steps. The remaining directional coefficients must have values
between 0 and 1. In summary, for each sizereduction step, this provides a total of 36
equations with 120 unknown directional coefficients to be evaluated. Fortuitously, the
solution to this particular series of LP problems indicated that only three nonzero direc
tional coefficients were required, out of a possible twelve, for each lockedparticle com
position range in a particlesizereduction step. This then gives a total of 30 nonzero
directional coefficients for each size step problem. All of the variables in Equations 5,
through 8, except the values of the directional coefficients, were known and obtained
from the lockedparticle breakage simulations for the GRLM. It was also noted that there
was a pattern in the occurrence of the nonzero directional coefficients that permitted
the distribution relationships to be grouped into five regions of the grain size–particle
size ratio: K <
1
/8,
1
/8 < K <
1
/2,
1
/2 < K < 1, 1 < K < 22.6, and K > 22.6. Additional detail is pre
sented in other literature (Wiegel 2000).
Breakage of a particular narrow lockedparticle composition range from one specific
particlesize range to a smaller specific size range, with the GRLM, is treated as a series
of sizereduction steps. A demonstration of the calculation is shown in Table 1, how the
distribution of composition classes is handled by the GRLM simulation. In this case,
the demo starts with an ore with a 0.25 volume fraction of values, and for the sake of
simplicity, the directional coefficients that describe the retention of material in their ini
tial composition ranges are assumed to be 0.50 for all but the totally liberated composi
tion ranges, which are by definition unity. Again for simplicity, the demo is based on that
material which does change from one composition class to another, moving to the near
est neighbors, both higher and lower in composition. The quantity that moves in each
direction is calculated by the one feed–two product formula, based on the mean compo
sition of the composition ranges involved. In this example, the barren waste dilution is 0,
but if it were other than 0, it would appear as additional liberated waste, passing from
size to size as portions of it were broken in sizereduction steps. If the original mineral
ized ore had a 0.23 volume fraction of values, this would be simulated by starting with
80% of the feed volume in the 0.20–0.30 values range (0.25 mean) and 20% in the 0.10–
0.20 values range (0.15 mean). The actual GRLM simulation results for a crude ore with
a 0.25 volume fraction of values are shown in Table 2 for the grain size–particle size
ratio range of
1
/8 to 8 in a progression of the square root of 2.
Incorporation into a Size Reduction/Mineral Liberation Simulation Program. The
directional coefficient approach and the values obtained from the LP solutions for a
range of original ore feed grades were then converted to regression relationships and
used as a basis to construct a computer program to simultaneously simulate tumbling
mill size reduction and the resultant progressive liberation for a binary mineral system
(Wiegel 2000). This BASIC program’s use was demonstrated for the simulation of the
batch grinding of the coarsegrained magnetite ore, described previously, and the
closedcircuit grinding, hydrocycloning, and magnetic separation of a magnetic taconite
ore (Wiegel 2002). One interesting aspect of this simulation capability is the possibility
of recognizing and incorporating different grinding rates for the values and waste constitu
ents, and likewise, by volumetric weighting of the locked particles. The size reduction/
mineral liberation portion of the program has since been translated into FORTRAN for
Q I II J , , ( )MV II ( )
II
_
MV I ( ) =
2
3
4
A
D
V
A
N
C
E
S
I
N
C
O
M
M
I
N
U
T
I
O
N
L
I
B
E
R
A
T
I
O
N
A
N
D
B
R
E
A
K
A
G
E
TABLE 1 Demonstration of calculation of quantity in each composition range when simulating three individual sizereduction steps
Composition
Index
Composition
Mean
K = 0.125
Original
K = 0.149
Step 1/3
K = 0.177 K = 0.210
Step 2/2 Step 2/3 Step 2/4
Step 2
Total Step 3/1 Step 3/2 Step 3/3 Step 3/4 Step 3/5
Step 3
Total
0 0 2.08 2.08
1 0.05 6.25 6.25 3.12 6.25 9.38
2 0.15 25.00 12.50 12.50 25.00 1.04 12.50 9.38 22.92
3 0.25 100.00 50.00 6.25 25.00 6.25 37.50 6.25 18.75 6.25 31.25
4 0.35 25.00 12.50 12.50 25.00 9.38 12.50 1.56 23.44
5 0.45 6.25 6.25 6.25 3.12 9.38
6 0.55 1.56 1.56
7 0.65
8 0.75
9 0.85
10 0.95
11 1
Total quantity 100.00 100.00 25.00 50.00 25.00 100.00 6.25 25.00 37.50 25.00 6.25 100.00
Total composition 0.25 0.25 0.15 0.25 0.35 0.25 0.05 0.15 0.25 0.35 0.45 0.25
R
A
T
I
O
N
A
L
E
F
O
R
O
N
E
M
O
D
E
L
D
E
S
C
R
I
B
I
N
G
T
H
E
S
I
Z
E
R
E
D
U
C
T
I
O
N
/
L
I
B
E
R
A
T
I
O
N
O
F
O
R
E
S
2
3
5
TABLE 2 GRLM simulation results for an ore grade with a 0.25 volume fraction of values for grain size–particle size ratio range of
1
/
8 to 8
Composition
Index
Composition
Mean K = 0.125 K = 0.177 K = 0.250 K = 0.354 K = 0.500 K = 0.707 K = 1.00 K = 1.41 K = 2.00 K = 2.83 K = 4.00 K = 5.66 K = 8.00
0 0 0.04 4.31 13.02 31.75 44.91 54.00 60.28 64.63 67.67
1 0.05 1.76 11.46 20.76 25.86 16.32 10.02 6.25 4.00 2.63 1.76
2 0.15 2.96 15.65 24.93 25.32 22.83 14.13 9.51 6.54 4.46 3.03 2.07 1.43
3 0.25 100.00 94.08 68.70 48.42 32.15 17.82 12.67 7.96 5.59 3.98 2.84 2.03 1.46
4 0.35 2.96 15.65 21.47 18.64 11.06 8.83 6.93 4.99 3.58 2.57 1.84 1.32
5 0.45 3.28 8.37 11.58 8.09 6.54 4.90 3.65 2.70 1.99 1.45
6 0.55 0.14 3.22 7.65 6.55 5.28 4.18 3.21 2.41 1.78 1.30
7 0.65 0.73 2.72 4.44 4.96 4.21 3.35 2.57 1.93 1.42
8 0.75 0.07 0.72 3.59 3.66 3.36 2.81 2.22 1.69 1.25
9 0.85 0.48 1.79 3.05 3.64 3.33 2.69 2.03 1.48
10 0.95 0.07 0.89 2.24 2.75 2.70 2.32 1.84 1.38
11 1 0.14 1.80 4.91 8.68 12.37 15.54 18.08
Total quantity 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00
Total composition 0.25 0.25 0.25 0.25 0.25 0.25 0.25 0.25 0.25 0.25 0.25 0.25 0.25
236 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
eventual inclusion in one of the mineral process modeling and simulation packages now
on the market. This approach appears to be adequate for its original intended target—the
simulation of magnetic taconite size reduction/liberation—but the author’s opinion is
that it is limited to original ore feed grades in the range of 15% to 85% values by volume.
The magnetic taconite is usually in the 20% to 40% by volume range and therefore fits
the suitability criteria.
EXTENSI ON OF THE DI RECTI ONAL COEFFI CI ENT APPROACH
TO LOW GRADE ORE
Much of the interest in size reduction/liberation, however, is related to the processing of
lowgrade, hardrock ores in the onehalf to several percent values range, for which this
current program is not suitable. The limitation relates to the choice of ten evenly spaced
ranges to represent the locked particle composition spectrum for mediumgrade ores.
This results in the lowest composition range of 0% to 10% values having a mean of 5%,
which serves no purpose when attempting to represent an ore with, say, 2% values by
volume. There is no reason to believe, however, that the approach used for medium
grade ores cannot be applied to the lowgrade ores. That is, the quantitative information
on composition distribution could be generated by locked particle breakage simulations
using the GRLM, and the directional coefficients could be obtained by setting up and
solving a similar set of LP problems.
At this point, serious thought has been given to the differences between the two
problems, and several items become obvious:
It will be necessary to expand the number of composition ranges significantly to
adequately cover lowgrade ore.
It is probably appropriate to look at something akin to logarithmic composition
ranges in the low fraction values region, and in the high values region as well,
because of the symmetry due to a binary mineral system.
When the number of composition ranges is increased, it may be necessary to look
at smaller graintoparticlesize ratios than the
1
/8 value limit used with the
mediumgrade simulations.
The number of directional coefficients is related to the square of the number of
composition ranges, and the problem of solving for their values becomes signifi
cantly more difficult—already for the mediumgrade ore there were 120 unknown
values.
Although an increase in the number of composition ranges will increase the quan
tity of variable storage and the program execution time in the simulation pro
gram, this is probably of minor concern.
The interest in lowgrade ores frequently coincides with an interest in an ore
more complex than binary, but the GRLM can be used to generate locked particle
breakage information for such an ore, recognizing that the underlying mineral
grain size, shape, and breakage assumptions for the GRLM are now being
extended even further.
Expanding the Number of Composition Ranges
As a starting point, a series of locked particle breakage simulations for the GRLM was
carried out for a 0.015 fraction values ore, in which the number of composition ranges
was increased from the current 12 to 30. The single 0–0.1 range was expanded to ten:
0–0.0001, 0.0001–0.0002, 0.0002–0.0005, 0.0005–0.001, 0.001–0.002, 0.002–0.005,
0.005–0.01, 0.01–0.02, 0.02–0.05, and 0.05–0.1. There was a complementary expansion
RATIONALE FOR ONE MODEL DESCRIBING THE SIZE REDUCTION/LIBERATION OF ORES 237
at the upper end as well. A summary of the results is presented in Table 3, expressed as
the number of particles out of a total of one million that were broken. It is interesting to note
how very few particles occur in the composition region of 0.9995–1.0 fraction values
over the entire size ratio range of
1
/16 to 64. In the lowercomposition region of 0–0.0005
fraction values, a maximum of about 2.5% of the total number of particles occur. The
author’s opinion is that the increased complexity and difficulty in obtaining directional
coefficients for these additional six composition classes, three at both the low and high
composition ends, would outweigh the marginal benefit that might be provided. The
plan is therefore to proceed with the development of information based on 24 composi
tion ranges, recognizing that the calculation of even these directional coefficients may in
itself be a Herculean task.
Solitary Grain Model Representation of a LowGrade Ore
As a part of the original work aimed at deriving quantitative liberation relationships
from Gaudin’s conceptual model, it was recognized that a set of analytical equations
could be obtained to describe the distribution of locked particles for a lowgrade ore,
where some complexities of the GRLM become negligible (Wiegel and Li 1967). This
model was termed the Solitary Grain Model (SGM) and can be visualized as a solitary
cubic grain of valuable mineral completely surrounded by grains of waste mineral. The
same form of cubic fracture lattice is imposed on this assemblage of grains, but because
there is no chance of having adjacent valuablemineral grains, many of the terms in
Equations 1 and 2 disappear or can be simplified. When one considers the breakage of
the solitary grain of values to particles smaller than the grain size (K > 1), the number
of liberated particles of values is given by Equation 9, whereas the locked particles are of
three types as given by Equation 10: those located along the six surfaces of the cubic
valuablemineral grain, which are composed of two grain fragments; those located along
the 12 edges of the valuablemineral grain, which are composed of four grain fragments;
and those located at the eight corners of the valuablemineral grain, which are com
posed of eight grain fragments. The remaining particles are liberated waste mineral
(Equation 11).
PB = VB(K – 1)
3
/K
3
(EQ 9)
PAB = VB[6(K – 1)
2
+ 12(K – 1) + 8]/K
3
(EQ 10)
PA = 1 – VB(K + 1)
3
/K
3
(EQ 11)
For this model, with particle size larger than grain size (K < 1), there are no liber
ated values, but there is a maximum values particle composition, which occurs when a
full grain of valuable mineral is contained in a particle, and is given by Equation 13. The
liberated waste continues to follow Equation 11, but there is a size ratio at which the
approximation to the amount of liberated waste mineral becomes zero, as shown in
Equation 14. The quantity of locked particles is now given by Equation 15.
PB = 0 (EQ 12)
VBMAX = (o/)
3
= K
3
(EQ 13)
K(@ PA = 0) = 1/((1/VB)
1/3
– 1) (EQ 14)
PAB = VB(K + 1)
3
/K
3
(EQ 15)
2
3
8
A
D
V
A
N
C
E
S
I
N
C
O
M
M
I
N
U
T
I
O
N
L
I
B
E
R
A
T
I
O
N
A
N
D
B
R
E
A
K
A
G
E
TABLE 3 Summary of particle breakage simulation for liberation of ore with a 0.015 volume fraction of values for size ratio of 1/16 to 64
VB Average/K 0.0625 0.125 0.25 0.5 1 2 4 8 16 32 64
1 0 0 0 0 0 1959 6423 10113 12397 13657 14318
0.99995 0 0 0 0 0 2 4 0 0 0 0
0.99985 0 0 0 0 0 1 1 3 0 1 0
0.99965 0 0 0 0 0 4 1 2 1 0 1
0.99925 0 0 0 0 0 6 6 1 1 1 1
0.9985 0 0 0 0 0 10 13 7 5 3 1
0.9965 0 0 0 0 0 33 36 23 16 8 3
0.9925 0 0 0 0 0 58 67 45 26 11 7
0.985 0 0 0 0 1 106 126 92 48 30 14
0.965 0 0 0 0 5 395 392 256 146 72 40
0.925 0 0 0 0 28 647 656 436 248 135 68
0.85 0 0 0 0 198 1518 1393 891 499 263 134
0.75 0 0 0 0 580 1846 1510 922 506 265 136
0.65 0 0 0 0 1228 2254 1631 960 516 268 136
0.55 0 0 0 0 2269 2755 1787 999 527 270 137
0.45 0 0 0 0 4033 3411 1986 1054 542 274 138
0.35 0 0 0 40 6836 4241 2228 1118 556 277 138
0.25 0 0 0 1693 11667 5612 2568 1201 578 284 140
0.15 0 0 20 43826 21231 7927 3151 1344 614 292 143
0.075 0 0 16301 74265 19204 5778 1960 772 334 156 72
0.035 4580 160127 278479 82920 18904 4982 1499 525 214 89 45
0.015 993275 678329 296679 44884 9858 2261 617 216 76 37 16
0.0075 2145 148912 111079 31673 6581 1391 368 118 43 16 8
0.0035 0 11921 74322 25900 5423 1060 263 72 28 11 3
0.0015 0 480 29873 12149 2363 433 102 27 10 3 2
0.00075 0 125 17581 7280 1388 255 55 13 4 2 1
0.00035 0 50 12607 5368 1135 179 38 9 2 1 1
0.00015 0 20 4929 2222 362 68 15 4 1 0 0
0.00005 0 20 6938 2846 591 83 22 3 1 1 0
0 0 16 151192 664934 886115 950725 971082 978774 982061 983573 984297
Calculated VB 0.01508 0.01694 0.01658 0.01651 0.01531 0.015079 0.01502 0.01500 0.01500 0.01500 0.01500
Calculated volume 1000000 1000000 1000000 1000000 1000000 1000000 1000000 1000000 1000000 1000000 1000000
RATIONALE FOR ONE MODEL DESCRIBING THE SIZE REDUCTION/LIBERATION OF ORES 239
There are SGM analytical relationships for the quantity of locked particles expressed
as a function of the size ratio and linear and logarithmic functions of locked particle
composition (Wiegel and Li 1967). Although repeating these relationships is not war
ranted here, they are quantitative approximations for lowgrade ore which, because they
are in an analytical form, may prove of value in extending the modeling and simulation
capability of size reduction and liberation. A comparison of data obtained from the SGM
and the GRLM are shown in Figure 6 for an ore containing 0.015 volume fraction of valu
able mineral.
SUMMARY AND CONCLUSI ONS
Mineral liberation is an extremely important phenomenon to understand and to be able
to quantitatively model, as it is this liberation that permits subsequent mineral separa
tions to be carried out and simulated effectively. The development of the GRLM began
with the work of Gaudin in the 1930s, as a conceptual understanding of how size reduc
tion affects the liberation of an idealized binary mineral system. In the 1960s, the Gaudin
conceptual model was modified by the application of binary probability concepts, and as
a result it could be quantified for the entire range of ore feed grades and particle sizes.
Since then, over a 40year period, there has been a gradual collection of experimental
data to provide support for the quantitative model. Additional efforts have been made
more recently to extend both its qualitative and quantitative usefulness by tapping some
of its more intricate features, such as information on the gradual change in the composi
tional distribution of locked particles as particle size is reduced and mineral liberation
proceeds. With the expected inclusion of the GRLM into a popular mineral processing
simulation package, it is now ready for a rigorous test of its applicability to the simula
tion of real mineral systems. With the capability of incorporating mineral liberation into
the suite of unit operation models available to the process engineer, it will no longer be
0.00001
0.0001
0.001
0.01
0.1
1
0.1 1 10 100
C
u
m
u
l
a
t
i
v
e
F
r
a
c
t
i
o
n
b
y
V
o
l
u
m
e
Grain to Particle Size Ratio
0.00 Values
0.01 Values
0.10 Values
0.50 Values
0.90 Values
0.99 and 1.00 Values
Liberated Waste
Liberated Values
Solitary Grain Model
Open Symbols and Curve
GRLM Filled Symbols
FIGURE 6 Comparison of GRLM results with SGM calculations for ore with 0.015 volume fraction
of values
240 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
necessary to assume either complete homogeneity or complete liberation in flowscheme
simulations, but rather it should be possible to better simulate the “entire real process.”
Unfortunately, one current restriction on the use of the model is its applicability only
to intermediate or mediumgrade ores, in the range of 15% to 85% by volume of valu
able mineral. However, there is no reason to believe that the GRLM concept and the
approach used to apply it to the mediumgrade binary ores cannot be extended to cover
the lowgrade, multicomponent ores as well. To that end, work has begun on extending
the current 12 composition ranges of the GRLM to 24. The author invites anyone inter
ested to apply their own ingenuity to this task, and would be happy to share with such
industrious individuals any unpublished ideas and special programs that have been
developed for generating the locked particle composition distributions from particle
breakage simulations.
Virtually all of the published experimental verification of the applicability of the
GRLM to real mineral systems has been for magnetic iron ores. The GRLM is not limited
to this application; but rather this situation is due first to the author’s professional career
being focused on the Minnesota and Northern Michigan iron ore industry and the Florida
phosphate rock industry where there are no mineral liberation problems; and second, to
the usefulness of the Davis tube as an efficient, simple, and inexpensive laboratory tech
nique that ensures an almost perfect separation of magnetics from nonmagnetics. Informa
tion comparable to the Davis tube could be collected from flotation or gravity separation
samples, where the individual concentrate and tailing samples from the plant or laboratory
test are screened into narrow size fractions and subjected to analytical measurements to
obtain the required mineral liberation information, recognizing that some separation
inefficiencies will necessarily be present.
REFERENCES
Barbery, G. 1991. Mineral Liberation—Measurement, Simulation and Practical Use in Mineral
Processing. Quebec, Canada: Editions GB.
Gaudin, A.M. 1939. Pages 70–91 in Principles of Mineral Dressing. New York: McGrawHill.
Lynch, A.J. 1977. Mathematical model of mineral liberation. Pages 187–202 in Mineral
Crushing and Grinding Circuits, Their Simulation, Optimisation, Design and Control.
New York: Elsevier Scientific Publishing.
Schneider, C.L. 1995. Measurement and calculation of liberation in continuous milling
circuits. Ph.D. dissertation. Salt Lake City, UT: University of Utah.
Wennen, J.E., W.J. Nordstrom, and D.L. Murr. 1997. National Steel Pellet Company’s sec
ondary grinding circuit modifications. Pages 19–25 in Comminution Practices. Edited
by S.K. Kawatra. Littleton, CO: SME.
Wiegel, R.L. 1964. A mathematical model for mineral liberation by size reduction. Master’s
paper, Pittsburgh, PA: Carnegie Mellon University, Chemical Engineering Department.
———. 1965. A quantitative approach to mineral liberation. Pages 3–9 in VII International
Mineral Processing Congress. Volume I. Edited by N. Arbiter. New York: Gordon and
Breach.
———. 1975. Liberation of magnetite iron formations. AIME Transactions 258(3):247–256.
———. 1976a. Integrated size reductionmineral liberation model. AIME Transactions
260(2):147–152.
———. 1976b. Simulation of magnetic taconite concentration processes. Ph.D. dissertation.
Brisbane, Australia: University of Queensland.
RATIONALE FOR ONE MODEL DESCRIBING THE SIZE REDUCTION/LIBERATION OF ORES 241
———. 1979. Application of process modeling to taconite. AIME Transactions 266:1863–1876.
———. 1999a. Fitting of Liberation Model Parameters to Davis Tube Test Data. Technical
Report CMRL/TR9913. Duluth, MN: University of Minnesota–Duluth, Coleraine
Minerals Research Laboratory.
———. 1999b. Magnetic Taconite Concentration Modeling. Technical Report CMRL/TR9912.
Duluth, MN: University of Minnesota–Duluth, Coleraine Minerals Research Laboratory.
———. 2000. Development of an Approach to the Simulation of Size Reduction/Mineral
Liberation for Magnetic Taconite Ore in Tumbling Mills, and Its Implementation in a
Basic Computer Program. Technical Report. CMRL/TR0016. Duluth, MN: University
of Minnesota–Duluth, Coleraine Minerals Research Laboratory.
———. 2002. Size reduction/mineral liberation simulation for a magnetic taconite con
centrator. Mineral and Metallurgical Processing 19(3):113–122.
Wiegel, R.L., and K. Li. 1967. A random model for mineral liberation by size reduction.
AIME Transactions 238:179–189.
243
Influence of Slurry Rheology on Stirred
Media Milling of Limestone
Mingzhao He
*
and Eric Forssberg
*
ABSTRACT
This paper reviews the influences of solids concentrations and dispersants with a range of
molecular weights on the flowability of limestone slurries as well as the effects on wet
ultrafine grinding in order to reduce energy cost and increase the fineness of a product.
Sodium polyacrylate with a molecular weight of 5,500 (i.e., Dispersant S40) appears most
effective for the grinding due to the effective reduction of apparent viscosities and the main
tenance of steady flowability. Optimal solids concentration exists at a certain beads load for
the effective wet ultrafine grinding of limestone, and a rheological explanation is presented.
I NTRODUCTI ON
Due to some advanced properties of ultrafine powders—such as surface chemistry, pack
ing characteristics, strength, optical properties and reaction kinetics, and an increasing
demand for ultrafine powders for industries—wet ultrafine grinding has found increased
use in many fields (He, Wang, and Forssberg 2005). Most of the mills used in wet
ultrafine grinding are stirred media mills due to their high unit outputs and highenergy
efficiencies (Bernhardt, Reinsch, and Husemann 1999; Kapur et al. 1996; Blecher,
Kwade, and Schwedes 1996). The stirred media mills are equipped with a stationary
grinding chamber and a highspeed stirrer (disks or pins) fixed on a drive shaft. The
grinding chamber is filled with small grinding media (normally spherical annealed glass,
steel, or ceramic beads) at a high beads load. By stirring a slurry–bead mixture at a high
stirring speed, a characteristic flow pattern and a grinding action are generated in the
chamber. The respective kind of flow determines the spatial distribution of zones with
high grinding intensities and the predominant types of grinding mechanisms as well as
their composition (Blecher, Kwade, and Schwedes 1996; Kwade, Blecher, and Schwedes
1996). Thus, the predominant grinding mechanisms in stirred media mills are depen
dent upon compressional, shear, and torsional stresses, which are invoked by stirring the
slurry–bead mixture at a very high velocity (Blecher, Kwade, and Schwedes 1996;
Kwade, Blecher, and Schwedes 1996; Kwade 1999a,b). The effective grinding motions of
the mixture are correlated to its flowability in the grinding chamber.
From a diagnostic point of view, the rheological behavior of a mineral slurry is
indicative of the level of interparticle interaction or aggregation in the slurry. Therefore,
it is a useful variable to be controlled in industrial processes such as transportation of
slurries, dewatering, and wet grinding (Muster and Prestidge 1995). Physical and chemical
* Division of Mineral Processing, Luleå University of Technology, Luleå, Sweden
244 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
properties of a slurry—such as solids concentration, use of dispersants, particle size and
distribution, particle shape, pH value, shear rate, and slurry temperature—have a significant
influence on the slurry rheology due to the change or modification in surface property
(He, Wang, and Forssberg 2004).
Product fineness significantly increases with grinding time in a wet ultrafine grind
ing operation characterized by a high solids concentration and the presence of excessive
fines; therefore, the surface properties tend to predominate in the system (Bernhardt,
Reinsch, and Husemann 1999; Klimpel 1999; Gao and Forssberg 1993b). Interparticle
forces such as van der Waals forces (Greenwood et al. 2002; Reinisch, Bernhardt, and
Husemann 2001) and electrostatic forces (Bernhardt, Reinsch, and Husemann 1999;
Muster and Prestidge 1995) lead to the formation of agglomeration and aggregation.
This results in changes in rheological property in wet ultrafine grinding operations. The
effect of slurry flowability or slurry rheology in wet ultrafine grinding becomes particu
larly important. The optimization of the rheological behavior of a ground slurry can
enhance the energy efficiency and throughput in wet ultrafine grinding operations. For
instance, the addition of an optimum dispersant to a given feed slurry can result in a
drastic reduction or even the elimination of yield stress and permits a higher solids con
centration of the ground slurry (Kapur et al. 1996; Klimpel 1999; Greenwood et al.
2002; Reinisch, Bernhardt, and Husemann 2001). In the absence of any dispersant, the
typical maximum percentage solids by weight in a slurry is approximately 50% for the
feed of ultrafine grinding in stirred media mills, whereas an upper limitation of solids
concentration is up to 80 wt % in the presence of an optimal dispersant (Greenwood et al.
2002). Therefore, the improvement of rheological behaviors of a feed slurry with the
addition of a suitable dispersant can enhance the productivity and throughput for wet
ultrafine grinding.
Many studies related to slurry rheology in conventional tumbling ball mills have
been published, but there is still little understanding of slurry rheology relevant to wet
ultrafine grinding characterized by a very fine product size and a high slurry concentra
tion; this is due to the complex slurry rheological behaviors in stirred media mills (Gao
and Forssberg 1993b; Blecher and Schwedes 1996). Also, findings from tumbling ball
mills involving the role of slurry rheology on the grinding results cannot be completely
applicable to the stirred media milling case due to their different breakage mechanisms
(Kwade 1999a,b; Gao and Weller 1993; Austin, Klimpel, and Luckie 1984). Therefore, it
is necessary for scientific understanding and industrial application to systematically
investigate slurry rheology and its effect on wet ultrafine grinding performance. The
objective of this paper is to investigate the influences of solids concentration and dispers
ants on the rheological behavior of limestone slurry, and its effect on wet ultrafine
grinding performance in order to reduce the energy cost and increase the throughput
and the fineness of a product. The grinding results are evaluated by energy efficiency
and the median size of a ground product with respect to specific energy consumption.
MATERI ALS AND EXPERI MENTAL METHODS
Materials
A limestone powder provided by SMA Karbonater AB, Sweden, was used for experiments
in this study. Figure 1 presents the particlesize distribution of the powder. The chemical
analysis and physical characteristics of the limestone powder are listed in Tables 1 and 2,
respectively. Sodium polyacrylates with a range of molecular weights such as BCX476,
Dispersant S40, and BCX552 obtained from CDM AB, Sweden, were selected as dispers
ants. Table 3 shows the physical and chemical properties of the sodium polyacrylates,
INFLUENCE OF SLURRY RHEOLOGY ON STIRRED MEDIA MILLING OF LIMESTONE 245
100
80
60
40
20
0
1 10 100 1,000
C
u
m
u
l
a
t
i
v
e
M
a
s
s
F
i
n
e
r
,
%
FIGURE 1 Particlesize distribution of limestone powder
TABLE 1 Chemical analysis of limestone composition
Main Chemical Composition Percent Standard Deviation
CaO
total
50 1.0
CaO
ASTM C602
52.5 0.7
CaCO
3 ASTM C602
93.7 1.3
MgO 4.50 1.0
Fe
2
O
3
0.70 0.2
SiO
2
3.50 1.0
Al
2
O
3
0.85 0.2
MnO 0.15 0.03
P
2
O
5
0.01
Loss on ignition 40 0.6
Moisture 0.2
TABLE 2 Physical characteristics of limestone
Real Density,
kg/m
3
Volume Density,
kg/m
3
Mohs
Hardness
Whiteness
(ISO 457), % pH
Particle
Shape
Specific surface
area, m
2
/g
2,700 1,000 3 77 9 Nodular 1.174
* Molecular weight determined by gel permeation chromatography (GPC).
TABLE 3 Physical and chemical properties of three sodium polyacrylates
Physical and Chemical Properties BCX476 Dispersant S40 BCX552
Solid content, wt % 45 45 30
Active content, wt % 40 40 26
pH 7.5 7.5 7.5
Density at 20°C 1.30 1.30 1.315
Molecular weight
*
2,000 5,500 85,000
Sodium polyacrylate content, wt % 40 40 26
Water content, wt % 55 55 70
Solubility (in water) Very soluble Very soluble Very soluble
246 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
including their molecular weights. In this investigation, the additive amount of a dispers
ant is the percentage of the pure dispersant (i.e., active content) by weight relative to the
weight of solids in a limestone slurry.
Experimental Methods
Mixup. A limestone–water slurry was prepared at a predetermined solids con
centration by the addition of 40 to 50 kg of limestone powder into a certain amount of
water to make up about 35 L of the slurry. The slurry was stirred as the limestone pow
der was added to the water, and the stirring continued for 15 min after the completion of
adding the limestone powder to make the slurry uniform. If a given dispersant was nec
essary, the slurry was stirred again for 15 min after its addition to make the chemical dis
perse uniformly in the slurry. About 200 mL of slurry was taken for each sample, which
was used to analyze the particle size and distribution and to measure the rheological
behavior of the slurry. The rest of the limestone slurry was used for grinding experiments.
Stirred Media Milling. A stirred media mill utilized for the grinding experiments is
the PMH 5 TEX Drais mill (Draiswerk GmbH, Germany). It consists of a )150u420mm
stainlesssteel cylinder chamber (5.6 L of net grinding chamber volume) and a stirrer
with six )120u10 perforated discs installed on a horizontaldriven shaft. The grinding
chamber is equipped with a water jacket for cooling. The discs rotate at 1,808 rpm (cor
responding to a peripheral speed of 11.36 m/sec). The flow rate of a feed slurry in this
study was controlled by a feeding pump at 1.5 L/min. The discharge of a product was
facilitated by means of two specially designed screen cartridges inserted at the end of the
cylinder. Zircon beads with diameters of 1.6–2.0 mm and a density of 3,700 kg/m
3
were
used as grinding media. The grinding operation in the PMH 5 TEX is the mode of circula
tion pass by pass. Fifteen of the 35 L of slurry were fed through the mill first and then
thrown away in order to attain a steady milling state with respect to solids concentra
tion. The remaining
~
20 L of slurry was circulated through the mill for several passes.
About 200 mL of slurry sample was taken after each pass. The samples were then analyzed
for particle size, rheological property, and specific surface area, respectively.
Viscometer. A rotational viscometer called a Bohlin Visco 88 BV (Bohlin Reologi
UK Ltd., United Kingdom) was used for the determination of slurry viscosities and shear
stress–shear rate curves. It employs a concentric cylinder geometry with a rotating inner
cylinder and a stationary outer cylinder. In this study, all samples were measured by the
use of C30 system (C30 DIN), which has a gap width of 1.5 mm between the inner and
outer cylinders and can provide a viscosity range of 0.007 to 6.18 Pa·sec. The inner cylin
der has 8 different rotation speeds, from 20 to 1,000 rpm, corresponding to a shearrate
range of 4 to 1,200 sec
–1
. The torque developed on the inner cylinder due to a sample is
directly related to the sample viscosity and should be in the range of 0.5 to 9.5 mN·m for
accuracy. A thermal jacket allows the use of an external fluid circulator to control or reg
ulate the temperature of a sample measured. The viscometer employs a socalled “vis
cosoft” computation program so the digital readings, or measurement parameters, such
as shear rate, shear stress, viscosity, and torque, are directly displayed on the screen of
the viscometer. The viscometer exhibits good reproducibility with a rather small stan
dard deviation for all the measurement parameters and accurately gives a resolution of
0.001 Pa·sec at a torque >0.5 (He, Wang, and Forssberg 2005).
Slurry samples were aged for 4 hours and then were shaken at an intensity of
225 min
–1
(TH30 shaker from Edmund Bühler, Germany) for 1 hr to redisperse the sam
ples prior to the rheological measurement. Each slurry sample was first presheared for
3 min at the highest shear rate of the viscometer (i.e., 1,200 sec
–1
) prior to measure
ment; then the measurement started from this highest shear rate. The shear rate was
stepped down one by one until the torque reading was less than 0.5 mN·m. The digital
INFLUENCE OF SLURRY RHEOLOGY ON STIRRED MEDIA MILLING OF LIMESTONE 247
readings (viscosity, shear rate, shear stress, and torque) were recorded at each shear
rate. About 25 mL of slurry sample was required for each measurement. Each sample
was measured three times, and the mean values of shear stress, shear rate, and viscosity
were used for analysis. In addition, the extrapolated Bingham yield stress, W
B
, was uti
lized, which was obtained by fitting the experimental data into the Bingham model (Gao
and Forssberg 1993b; Prestidge 1997; Muster and Prestidge 1995):
(EQ 1)
where W is the shear stress, K
pl
is the plastic viscosity, and J is the shear rate.
Energy Consumption. The energy consumed by the mill was measured by an elec
trical meter called a Micro VIP (Elcontrol Co., Italy). In this study, only the active power
(kilowatts) was recorded and used by considering the power factor. The active power of
the mill is sensitive to the current change at all levels up to the rated power of the motor.
An active power reading was recorded every minute during each grinding pass, and
about 10 readings were taken for each pass. The mean active power of each pass is
regarded as its real one. The mean active power, P
mn
(kW), of the nth pass for a grinding
experiment was determined by
(EQ 2)
where P
ni
is the ith discrete power reading of the nth pass, and m is the number of read
ings of the nth pass. The milling energy of the nth pass was calculated by taking away the
idle power draw of the mill, P
0
(kW) (without grinding media or ground material) from
the mean active power, P
mn
(kW) at the given rotation speed of 1,808 rpm. Only the
power adsorbed by the mill chamber was accounted for in all the tests in this study. In
order to evaluate the net energy consumption of the Drais mill, the mass specific energy
consumption, E
m
(kWh/t), was determined by (Stehr and Weyand 1990):
(EQ 3)
where M
p
(kg/sec) is the mass flow rate of a slurry suspension fed to a mill, and C is the
solids concentration by weight. Due to the volume flow rate of a limestone slurry being
controlled and measured in this study, Equation 3 is modified as
(EQ 4)
where M
v
(m
3
/sec) is the volume flow rate of the slurry suspension fed to the mill, C
v
is
the volume concentration of the slurry suspension, and ȡ (kg/m
3
) is the density of solid.
Specific Surface Area. The specific surface area of a sample was measured by a
Flow Sorb ÉÉ 2300 (Micromeritics Co., Ltd., United States), which is an instrument
designed to take the measurements on bonedried powders by N2 gas adsorption and
desorption in liquid nitrogen temperature and room temperature (BrunauerEmmet
Teller [BET] method), respectively. A representative amount of sample was taken
from each sample, and then was dried in an oven at 110˚C for 24 hours in order to
remove the residual moisture of the sample prior to the measurement. The mean value
W W
B
K
pl
J + =
P
mn
P
ni
i=1
m
¦
m
 =
E
m
P
mn
P
0
–
3.6M
p
C
 =
E
m
P
mn
P
0
–
3,600M
v
C
v
U
 =
248 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
of the adsorption and desorption specific surface areas of a sample was regarded as its
real one.
Calculation of Energy Efficiency. Energy efficiency or energy utilization, E
f
(m
2
/Wh),
which is defined as the increment of specific surface area per unit of specific energy con
sumption (Gao and Forssberg 1993a), was calculated by
(EQ 5)
where 'S = S – S
0
; S and S
0
are the specific surface areas (m
2
/g) of a ground product and
its feed by BET, respectively.
Particle Size and Distribution. An xray sedimentometer SediGraph 5100D (Micro
meritics) was used to analyze the particle size and distribution of samples. It measures
particle size ranges from 0.1 to 300 Pm, which is suitable for the samples in this study. A
representative amount of sample was directly dispersed in aqueous solution of 0.1 wt %
of Calgon for the measurement.
RESULTS AND DI SCUSSI ON
Effect of Molecular Weight of a Dispersant
The effect of the concentration of sodium polyacrylates with various molecular weights
on the rheological behavior of the limestone slurry at 75 wt % of solids concentration
appears similar. Figure 2 shows the effect of the concentration of the Dispersant S40
sodium polyacrylate on the flowability of the slurry.
The apparent viscosity of the slurry first increases, then decreases, and reverts after
reaching a minimum. The slurry with less than 0.01 wt % of BCX476 or Dispersant S40,
or 0.015 wt % of BCX552 shows a pseudoplastic characteristic with a yield stress and
possesses a higher apparent viscosity than that without dispersant at a given shear rate.
The reason is that an insufficient dispersant causes the flocculation of particles in the
slurry by bridging attraction forces, leading to larger flow units (Zhou, Scales, and Boger
2001; Johnson et al. 2000). At the additive amount of a dispersant up to 0.015 wt % for
BCX476 or Dispersant S40, and 0.02 wt % for BCX552, the apparent viscosity of the
slurry is lower than that without dispersant, and the slurry still exhibits pseudoplastic
flowabilities with an evident yield stress. At the additive level of 0.03 wt %, the slurry
shows different rheological properties with different molecular weights of dispersants.
For Dispersant S40, the slurry is transformed into a dilatant flow (Figure 2), whereas the
slurry presents a pseudoplastic property with an insignificant yield stress for BCX476 or
BCX552. The gradual increase in the additive amount of a dispersant leads to a com
plete transition to a weakly dilatant flow and a further slight decrease of the apparent
viscosity in the range of shear rates investigated, and to a minimum at an additive level
of 0.2 wt % for BCX476 or 0.1 wt % for Dispersant S40, or 0.04 wt % for BCX552. In
these cases, the saturation adsorption of the dispersants on the particle surface has been
attained, and electrostatic and steric stabilizations (electrosteric stabilizations) occur. By
further adding the dispersants, the apparent viscosity at a given shear rate insignificantly
reverts for BCX476 and Dispersant S40, but evidently returns for BCX552. In the case of
BCX476 or Dispersant S40, the excessive dispersant above the adsorption saturation
exists in the slurry but does not adsorb onto the suspended particles, and causes a deple
tion flocculation (Papo, Piani, and Ricceri 2002; Zhou, Scales, and Boger 2001; Johnson
et al. 2000). In addition, the excessive dispersant can increase the ion strength of the
slurry, resulting in a compression of the electrical double layers around the particles and
a reduction of the range and magnitude of the electrostatic repulsive force between the
E
f
1,000'S
E
m
 =
INFLUENCE OF SLURRY RHEOLOGY ON STIRRED MEDIA MILLING OF LIMESTONE 249
particles, that is, the electrosteric forces are decreased (Papo, Piani, and Ricceri 2002;
Zhou, Scales, and Boger 2001; Banash and Croll 1999; Ewais, Zaman, and Sigmund
2002). In the case of BCX552, it possesses a larger molecular weight (i.e., 85,000) with
a longer molecular chain, so the depletion flocculation is more significant, in addition to
the reduction of electrosteric force (Banash and Croll 1999). Therefore, above the adsorp
tion saturation, a further addition of BCX552 evidently raises the apparent viscosity of
the slurry at a given shear rate. Figure 3 shows the results. Clearly, the yield stress is
eliminated when the additive amount of dispersants with three different molecular
weights exceeds 0.03 wt %. In the case of 75 wt % of the limestone slurry, the apparent
0
20
40
60
80
100
120
0 500 1,000 1,500
0
0.01 wt %
0.02 wt %
0.04 wt %
0.2 wt %
0.6 wt %
0.004 wt %
0.015 wt %
0.03 wt %
0.1 wt %
0.4 wt %
S
h
e
a
r
S
t
r
e
s
s
,
P
a
Shear Rate, L/sec
(a)
0
0.01 wt %
0.02 wt %
0.04 wt %
0.2 wt %
0.6 wt %
0.004 wt %
0.015 wt %
0.03 wt %
0.1 wt %
0.4 wt %
0.01
0.1
1
0 500 1,000 1,500
A
p
p
a
r
e
n
t
V
i
s
c
o
s
i
t
y
,
P
a
·
s
e
c
Shear Rate, L/sec
(b)
FIGURE 2 Flowability of 75 wt % of solids concentration with various dosages of Dispersant S40
at 25 ± 0.2°C
250 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
viscosity of the slurry at a given shear rate stays constant in a wider range of the concen
tration of dispersant (about 0.1–0.6 wt %) for BCX476 than that (0.1–0.2 wt %) for Dis
persant S40. Similar phenomena were observed for kaolin suspensions with sodium
tripolyphosphate and sodium polyphosphate (Papo, Piani, and Ricceri 2002) and zirco
nia slurries with triammonium citrate (Ewais, Zaman, and Sigmund 2002).
For 70 wt % of the limestone slurry, a similar phenomenon was also observed. The
apparent viscosity of 70 wt % of the limestone slurry reaches a minimum at an additive
level of 0.1 wt % for BCX476 or Dispersant S40, and 0.04 wt % for BCX552. Figure 4
shows the accessible minimum viscosity for 70 wt % of the limestone slurry with a given
additive amount for each dispersant. In the case of the reduction of viscosity of the lime
stone slurry, Dispersant S40 and BCX476 are better than BCX552. Dispersant S40 and
BCX476 present almost the same effect for 70 wt % of the limestone slurry (Figure 4),
while the former is better than the latter for 75 wt % of the slurry (Figure 3). This is
because Dispersant S40 gives a better steric stabilization than BCX476 in a denser lime
stone slurry due to its relatively larger molecular weight with respect to BCX476.
0
0.02
0.04
0.06
0.08
0.10
0.12
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Addition of Dispersants with Various Molecular Weights, wt %
BCX476
Dispersant S40
BCX552
(b)
A
p
p
a
r
e
n
t
V
i
s
c
o
s
i
t
y
a
t
t
h
e
S
h
e
a
r
R
a
t
e
o
f
6
6
3
s
e
c
–
1
,
P
a
·
s
e
c
0
5
10
15
20
25
30
0.0 0.01 0.02 0.03 0.04 0.05
Addition of Dispersants with Various Molecular Weights, wt %
E
x
t
r
a
p
o
l
a
t
e
d
B
i
n
g
h
a
m
Y
i
e
l
d
S
t
r
e
s
s
,
P
a
BCX476
Dispersant S40
BCX552
(a)
FIGURE 3 Change in rheological behaviors of 75 wt % of solids concentration with the addition of
dispersants with different molecular weights at 25 ± 0.2°C: (a) extrapolated Bingham yield
stress; (b) apparent viscosity at the shear rate of 663 sec
–1
INFLUENCE OF SLURRY RHEOLOGY ON STIRRED MEDIA MILLING OF LIMESTONE 251
Thus, the use of sodium polyacrylates with a range of molecular weights as grinding
aids can change the surface nature of particles in a ground limestone slurry, resulting in
interparticle forces being entirely repulsive to improve the slurry flowability by decreas
ing the viscosity and by eliminating the Bingham yield stress. This can increase the energy
efficiency and the fineness of a product for wet ultrafine grinding. To clarify the effect of
the molecular weights of sodium polyacrylates used as grinding aids on the wet ultrafine
grinding of limestone through the improvement of slurry rheology, Figure 5 shows the
grinding results for three sodium polyacrylates with different molecular weights at 75 wt %
of solids concentration with 0.2 wt % of each dispersant.
Clearly, Dispersant S40 with a molecular weight of 5,500 gives the better grinding
results (i.e., a higher energy efficiency and a smaller median size) in the wet ultrafine
grinding of limestone when other operation conditions are kept constant. The reason is
that Dispersant S40 maintains a lower viscosity during the grinding (Figure 6). Clearly,
the feed of 75 wt % of solids concentration with BCX552 exhibits higher viscosities in
the shear rate range studied (Figure 6a). This gives a lower energy efficiency and a larger
median size during pass 1. The feeds of 75 wt % of solids concentration with BCX476
and Dispersant S40 display almost the same rheological property in the shear rate range
and produce an insignificant difference in grinding results during pass 1. After pass 1,
the better grinding results are obtained for 75 wt % of solids concentration with Dispers
ant S40 due to its lower viscosities in the shear rate range. However, whatever dispersants
exist at 75 wt % of solids concentration, the grinding operation is forced to automatically
stop due to high pressure inside the mill chamber by a safety control device when a
ground slurry presents a pseudoplastic flow with a definite extrapolated Bingham yield
stress (Figures 6d and 7). The pseudoplastic slurry with a yield stress could be the reason
because the more viscous the slurry, the more power is needed to make the slurry flow
when the slurry is circulated through the mill by a pump. This increases the pressure in
the mill up to the limit. In addition, for 75 wt % of solids concentration with 0.2 wt % of
BCX552, the grinding process can only run for one pass but can run for three passes for
BCX476 or Dispersant S40. Consistent results were also obtained in the case of 70 wt %
of solids concentration with the three dispersants.
0 500 1,000 1,500
Shear Rate, L/sec
0
0.01
0.02
0.03
A
p
p
a
r
e
n
t
V
i
s
c
o
s
i
t
y
,
P
a
·
s
e
c
BCX476: 0.1 wt %
Dispersant S40: 0.1 wt %
BCX552: 0.04 wt %
FIGURE 4 Accessible minimum viscosity for 70 wt % of solids concentration in the presence of
three chemicals with a certain addition at 25 ± 0.2°C
252 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
Effect of Solids Concentration
The influence of solids concentration on slurry rheology is significant because various
ranges of solids concentrations can lead to different types of flows, as shown in Figure 8.
The results show the rheological properties of seven solids concentrations of lime
stone slurries without dispersant at 25 r 0.2˚C. The slurry rheological behavior is trans
formed from a weakly dilatant characteristic to an evidently pseudoplastic one with a
yield stress when the solids concentration is increased from 60 wt % (35.71 vol. %) to
78.5 wt % (57.49 vol. %). At a solids concentration d65 wt % (40.75 vol. %), the slurry
appears to be a weakly dilatant flow. This is because in a dilute slurry (i.e., d65 wt % or
40.75 vol. %), the interparticle distance is so large that the limestone particles in the
slurry are not subjected to the attractive forces between the particles but are free to
move as individuals. At lower shear rates, the particles have enough opportunities to slip
over each other. At higher shear rates, the shearing process becomes increasingly rapid,
causing restricted movement of the particles. The local accumulation of solid particles
causes the slurry to behave like a solid system. In a dilatant system, however, this state is
not stable. As long as the external force is removed, the particles without any tendency
to adhere are distributed more uniformly in the suspension. At solids concentrations up
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
5.5
0 50 100 150
BCX476
Dispersant S40
BCX552
(b)
Cumulative Specific Energy Input, kWh/t
45
50
55
60
65
70
BCX476
Dispersant S40
BCX552
(a)
0 50 100 150
C
u
m
u
l
a
t
i
v
e
E
n
e
r
g
y
E
f
f
i
c
i
e
n
c
y
,
m
2
/
W
h
Cumulative Specific Energy Input, kWh/t
FIGURE 5 Effect of chemicals with various molecular weights at the additive level of 0.2 wt % on
the grinding results for 75 wt % of solids concentration at 74 vol. % of beads load
INFLUENCE OF SLURRY RHEOLOGY ON STIRRED MEDIA MILLING OF LIMESTONE 253
0 500 1,000 1,500
0
0.05
0.10
0.15
(a)
BCX476
Dispersant S40
BCX552
0 500 1,000 1,500
0
0.04
0.02
0.06
0.08
0
0.04
0.02
0.06
0.08
0
0.8
0.4
1.2
1.6
(b)
BCX476
Dispersant S40
BCX552
0 500 1,000 1,500
(c)
BCX476
Dispersant S40
0 500 1,000 1,500
(d)
BCX476
Dispersant S40
BCX552
Shear Rate, L/sec
A
p
p
a
r
e
n
t
V
i
s
c
o
s
i
t
y
,
P
a
·
s
e
c
FIGURE 6 Change of rheological behaviors for 75 wt % of solids concentration with various
molecularweight chemicals at the additive level of 0.2 wt % with pass number (grinding time)
increasing at 74 vol. % of beads load: (a) feed; (b) after pass 1; (c) after pass 2; and (d) where
grinding is forced to stop automatically
0
10
20
30
40
50
60
70
BCX476 Dispersant S40 BCX552
E
x
t
r
a
p
o
l
a
t
e
d
B
i
n
g
h
a
m
Y
i
e
l
d
S
t
r
e
s
s
,
P
a
Chemicals with Different Molecular Weights
FIGURE 7 Extrapolated Bingham yield stress where the grinding process is forced to automatically
stop for 75 wt % of solids concentration with various molecularweight chemicals at the additive
level of 0.2 wt % at 74 vol. % of beads load
254 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
to 67 wt % or 42.92 vol. %, the flowability of the slurry exhibits a pseudoplastic charac
teristic without a definite yield stress at shear rates <663 sec
–1
and a weakly dilatant one
at shear rates >663 sec
–1
. This indicates that in a lower range of shear rates, the interpar
ticle attractive forces are predominant over the hydrodynamic forces exerted by a flow
field at 67 wt % of solids concentration, as opposed to shear rates >663 sec
–1
. With fur
ther increases in the solids concentration, to 70 wt % (i.e., 46.36 vol. %) or more, the
slurry rheology is changed into a pseudoplastic flow with an evident shear yield stress at
shear rates <362 sec
–1
, followed by a transition to a Bingham plastic flow (with a higher
extrapolated Bingham yield stress) at shear rates >362 sec
–1
. Furthermore, the degree of
pseudoplasticity and the shear yield stress increase with increasing solids concentration
when the solids concentration is >70 wt % (Figure 8a). Similar phenomena were
observed for various slurries of materials such as coal and quartz (Tangsathitkulchai and
Austin 1988), galena (Prestidge 1997), and sphalerite (Muster and Prestidge 1995). The
effect of solids concentration on the apparent viscosity at a given shear rate and the
extrapolated Bingham yield stress for the slurries is shown in Figure 9. The viscosity and
the yield stress increase rather sharply in exponential and powerlaw forms with increas
ing solids concentration when the solids concentration is >70 wt % (i.e., 46.36 vol. %),
0.01
0.1
1
10
0 500 1,000 1,500
(b)
Shear Rate, L/sec
A
p
p
a
r
e
n
t
V
i
s
c
o
s
i
t
y
,
P
a
·
s
e
c
60 wt % (35.71 vol. %)
65 wt % (40.75 vol. %)
67 wt % (42.92 vol. %)
70 wt % (46.36 vol. %)
75 wt % (52.63 vol. %)
77 wt % (55.36 vol. %)
78.5 wt % (57.49 vol. %)
0
10
20
30
40
50
60
70
80
90
100
(a)
0 500 1,000 1,500
Shear Rate, L/sec
S
h
e
a
r
S
t
r
e
s
s
,
P
a
60 wt %
65 wt %
67 wt %
70 wt %
75 wt %
77 wt %
78.5 wt %
FIGURE 8 Effect of solids concentration on the rheological properties of the limestone slurry in
the absence of chemicals at 25 ± 0.2°C
INFLUENCE OF SLURRY RHEOLOGY ON STIRRED MEDIA MILLING OF LIMESTONE 255
respectively. This is similar to a previous conclusion (Tseng and Chen 2003); indicating
that there are strong interactions between the particles to hold the particles together in
the slurry and form loosely packed flocs, immobilizing some water within them, which is
indispensable to flow. This is due to a smaller interparticle distance in a denser slurry,
producing an increased attraction potential and a larger probability of collisions between
the particles, resulting in more particles attracting each other. A shearing force over a
certain shear yield stress has to be exerted on the slurry to overcome the internal friction
among the particles constituting the flocs and to make it flow again. Once the slurry
flows, the flocs are broken down into smaller flow units, and the water entrapped within
them is gradually released with increasing shear rate. This facilitates the slurry to flow
and leads to a decrease of the slurry viscosity.
A small addition of a suitable dispersant is indispensable for the wet ultrafine grind
ing of limestone at a higher solids concentration (>70 wt %) due to the rapid increase in
the viscosity and the yield stress, which results in an increase in the energy consumption
for wet ultrafine grinding and even makes the operation impossible. In order to investigate
the effect of the difference in slurry rheology resulting from different solids concentra
tions on the wet ultrafine grinding, Dispersant S40 (molecular weight 5,500) is selected
as a grinding aid on the basis of the previous results. Figure 10 shows the grinding
0
5
10
15
20
(b)
55 60 65 70 75 80 85
Solids Concentration, wt %
E
x
t
r
a
p
o
l
a
t
e
d
B
i
n
g
h
a
m
Y
i
e
l
d
S
t
r
e
s
s
,
P
a
A
p
p
a
r
e
n
t
V
i
s
c
o
s
i
t
y
,
P
a
·
s
e
c
0
0.05
0.10
0.15
0.20
0.25
0.30
0.35
(a)
55 60 65 70 75 80 85
663 L/sec
362 L/sec
194 L/sec
Solids Concentration, wt %
FIGURE 9 Effect of solids concentration on (a) the apparent viscosity and (b) the extrapolated
yield stress at 25 ± 0.2°C
256 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
results of limestone slurries with three different solids concentrations with 0.2 wt % of
Dispersant S40 at 74 vol. % of beads load.
The cumulative energy efficiency increases first and then decreases with increasing
solids concentration from 65 wt % to 75 wt % at a given specific energy input. The median
size of the product varies with solids concentration in an opposite way from the cumula
tive energy efficiency. This is in agreement with a previous conclusion (Bernhardt, Reinsch,
and Husemann 1999). The best grinding results are obtained at 70 wt % of solids con
centration with 0.2 wt % of Dispersant S40 at 74 vol. % of beads load. The reason for this
is that the slurry of 70 wt % of solids concentration with 0.2 wt % of the dispersant
exhibits proper viscosities in a wide range of shear rate (Figure 11) and produces better
stress conditions (i.e., a higher stress intensity and a larger average number of stress
events for each particle). In the case of 65 wt % of solids concentration with 0.2 wt % of
the dispersant, a lower viscosity and a larger average interparticle distance make grind
ing beads difficult to effectively capture particles. This increases the possibility of the
direct collision between the beads, resulting in higher energy loss. A higher solids con
centration (i.e., 75 wt %) gives a smaller average interparticle distance, which leads to a
larger average number of stress events of each particle. However, at 75 wt % of solids
concentration, the larger viscosity damps the motion of the grinding beads in the mill
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
5.5
(b)
0 50 100 150 200
Cumulative Specific Energy Input, kWh/t
65 wt %
70 wt %
75 wt %
35
40
45
50
55
60
65
70
75
(a)
65 wt %
70 wt %
75 wt %
0 50 100 150 200
Cumulative Specific Energy Input, kWh/t
C
u
m
u
l
a
t
i
v
e
E
n
e
r
g
y
E
f
f
i
c
i
e
n
c
y
,
m
2
/
W
h
FIGURE 10 Effect of various solids concentrations with 0.2 wt % of Dispersant S40 on the
grinding results at 74 vol. % of beads load
INFLUENCE OF SLURRY RHEOLOGY ON STIRRED MEDIA MILLING OF LIMESTONE 257
and significantly increases the attenuation of the velocity and kinetic energy of the
beads, which bring about lower stress intensities of collisions among the beads, particles,
and chamber inner wall. The decrease in stress intensity is dominant, compared with the
increase in the number of stress events. Therefore, the captured particles cannot be
effectively ground, which causes an ineffective milling operation. However, the grinding
operation at a solids concentration from 65 to 75 wt % with a given amount of Dispersant
S40 (i.e., 0.2 wt %) automatically ceases when a ground slurry exhibits a pseudoplastic
flow with an evident extrapolated Bingham yield stress (Figures 11d and 12).
CONCLUSI ONS
The influences of dispersants with different molecular weights and solids concentration
on the flowability of limestone slurries as well as their effect on wet ultrafine grinding
have been investigated in order to reduce energy cost and increase the fineness of a
product. The effect of sodium polyacrylate salts with a range of molecular weights on the
rheological behaviors appears similar for a given solids concentration of limestone
slurry. With the additive amount increasing, the apparent viscosity of the slurry at a
given shear rate first increases, then decreases, and reverts after reaching a minimum.
Also, the yield stress is eliminated when the additive amount exceeds a certain value.
Sodium polyacrylate with a molecular weight of 5,500 (i.e., Dispersant S40) appears to
be more effective for the reduction of apparent viscosity and the maintenance of steady
flowability. Thus, Dispersant S40 as a grinding aid gives better grinding results (i.e., a
0 500 1,000 1,500
0
0.01
0.02
0.03
0.04
0.05
0
0.2
0.4
0.6
0.8
1
(a)
65 wt %
70 wt %
75 wt %
65 wt %
70 wt %
75 wt %
65 wt %
70 wt %
75 wt %
65 wt %
70 wt %
75 wt %
0 500 1,000 1,500
0
0.01
0.02
0.03
0.04
0.05
0.06
0
0.04
0.02
0.06
0.08
(b)
0 500 1,000 1,500
(c)
0 500 1,000 1,500
(d)
Shear Rate, L/sec
A
p
p
a
r
e
n
t
V
i
s
c
o
s
i
t
y
,
P
a
·
s
e
c
FIGURE 11 Change of rheological behaviors of three different solids concentrations with 0.2 wt %
of Dispersant S40 with pass number (grinding time) increasing at 74 vol. % of beads load: (a) feed;
(b) after pass 1; (c) after pass 3; and (d) where grinding is forced to stop
258 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
higher energy efficiency and a smaller median size). The rheological behavior of lime
stone slurry is transformed from a weakly dilatant characteristic to a pseudoplastic one
with a yield stress when the solids concentration is increased from 60 wt % (35.71 vol. %)
to 78.5 wt % (57.49 vol. %). The apparent viscosity and the extrapolated Bingham yield
stress increase in exponential and powerlaw forms with increasing solids concentration,
respectively, when the solids concentration of the slurry is greater than 70 wt % (i.e.,
46.36 vol. %). A small addition of a suitable dispersant is indispensable for the wet
ultrafine grinding of limestone at a higher solids concentration (>70 wt %) due to the
rapid increases in the viscosity and the yield stress. An optimal solids concentration
exists at a certain beads load for the effective wet ultrafine grinding of limestone.
ACKNOWLEDGMENTS
The financial support from the Swedish National Energy Administration and Hesselman
ska Stiftelsen in Sweden (MinFo’s Process Technology Program) is acknowledged. Pro
fessor Anders Sellgren and Yanmin Wang at the Luleå University of Technology are
appreciated for their helpful comments. The authors are grateful to BrittMarie Antti at
the paper mill in Piteå (Kappa Kraftliner Piteå, Sweden) for her kind help in the slurry
viscosity measurements.
REFERENCES
ASTM (American Society for Testing and Materials). 1996. C60295a. Standard specifi
cation for agricultural liming materials. Pages 306–308 in Annual Book of ASTM
Standards. Volume 04.01. West Conshohocken, PA: ASTM International.
Austin, L.G., R.R. Klimpel, and P.T. Luckie. 1984. Process engineering of size reduction:
Ball milling. New York: SMEAIME.
Banash, M.A., and S.G. Croll. 1999. A quantitative study of polymeric dispersant adsorp
tion onto oxidecoated titania pigments. Progress in Organic Coatings 35:37–44.
Bernhardt, C., E. Reinsch, and K. Husemann. 1999. The influence of suspension properties
on ultrafine grinding in stirred ball mills. Powder Technology 105:357–361.
0
5
10
15
20
25
30
35
65 wt % 70 wt % 75 wt %
E
x
t
r
a
p
o
l
a
t
e
d
B
i
n
g
h
a
m
Y
i
e
l
d
S
t
r
e
s
s
,
P
a
Solids Concentration by Weight, wt %
FIGURE 12 Extrapolated Bingham yield stress where the grinding process is forced to
automatically stop for three solids concentrations with 0.2 wt % of Dispersant S40 at 74 vol. % of
beads load
INFLUENCE OF SLURRY RHEOLOGY ON STIRRED MEDIA MILLING OF LIMESTONE 259
Blecher, L., A. Kwade, and J. Schwedes. 1996. Motion and stress intensity of grinding
beads in a stirred media mill. Part 1: Energy density distribution and motion of single
grinding beads. Powder Technology 86:59–68.
Blecher, L., and J. Schwedes. 1996. Energy distribution and particle trajectories in a
grinding chamber of a stirred ball mill. International Journal of Mineral Processing
44–45:617–627.
Ewais, E., A.A. Zaman, and W. Sigmund. 2002. Temperature induced forming of zirconia
from aqueous slurries: Mechanism and rheology. Journal of the European Ceramic
Society 22:2805–2812.
Gao, M., and E. Forssberg. 1993a. A study on the effect of parameters in stirred ball milling.
International Journal of Mineral Processing 37:45–59.
———. 1993b. The influence of slurry rheology on ultrafine grinding in a stirred ball mill.
Pages 237–244 in Proceedings of XVIII International Mineral Processing Congress. Vol
ume 2. Sydney, Australia.
Gao, M., and K. Weller. 1993. Fine grinding in mineral processing using stirred ball mills.
Chemical Engineering in Australia 18(2):8–12.
Greenwood, R., N. Rowson, S. Kingman, and G. Brown. 2002. A new method for deter
mining the optimum dispersant concentration in aqueous grinding. Powder Technol
ogy 123:199–207.
He, M., Y. Wang, and E. Forssberg. 2004. Slurry rheology in wet ultrafine grinding of
industrial minerals: A review. Powder Technology 147:94–112.
———. 2005. Parameter studies on rheology of limestone slurries. International Journal of
Mineral Processing. In press.
ISO (International Organization for Standardization). 1983. ISO 457. Soaps—Determination
of Chloride Content—Titrimetric Method. Geneva, Switzerland: ISO.
Johnson, S.B., G.V. Franks, P.J. Scales, D.V. Boger, and T.W. Headly. 2000. Surface
chemistry—rheology relationships in concentrated mineral suspensions. Interna
tional Journal of Mineral Processing 58:267–304.
Kapur, P.C., T.W. Healy, P.J. Scales, D.V. Boger, and D. Wilson. 1996. Role of dispersants
in kinetics and energetics of stirred ball mill grinding. International Journal of Mineral
Processing 47:141–152.
Klimpel, R.R. 1999. The selection of wet grinding chemical additives based on slurry
rheology control. Powder Technology 105:430–435.
Kwade, A. 1999a. Determination of the most important grinding mechanism in stirred
media mills by calculating stress intensity and stress number. Powder Technology
105:382–388.
———. 1999b. Wet comminution in stirred media mills—research and its practical applica
tion. Powder Technology 105:14–20.
Kwade, A., L. Blecher, and J. Schwedes. 1996. Motion and stress intensity of grinding
beads in a stirred media mill. Part 2: Stress intensity and its effect on comminution.
Powder Technology 86:69–76.
Muster, T.H., and C.A. Prestidge. 1995. Rheological investigations of sulphide mineral
slurries. Minerals Engineering 8:1541–1555.
Papo, A., L. Piani, and R. Ricceri. 2002. Sodium tripolyphosphate and polyphosphate as
dispersing agents for kaolin suspensions: Rheological characterization. Colloids and
Surfaces A: Physicochemical Engineering Aspects 201:219–230.
260 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
Prestidge, C.A. 1997. Rheological investigations of ultrafine galena particle slurries
under flotationrelated conditions. International Journal of Mineral Processing
51:241–254.
Reinisch, E., C. Bernhardt, and K. Husemann. 2001. The influence of additives during
wet ultrafine grinding in agitator bead mills. Part 1: General principles and experi
mental. Ceramic Forum International: Berichte der Deutschen Keramischen Gesell
schaft 78(3):E38–E42.
Stehr, N., and C. Weyand. 1990. Control System for Agitated Media Mills. Pages 681–695
in 7th European Symposium of Comminution, Ljubljana, Yugoslavia. Volume 2.
Edited by K. Schönert.
Tangsathitkulchai, C., and L.G. Austin. 1988. Rheology of concentrated slurries of particles
of natural size distribution produced by grinding. Powder Technology 56:293–299.
Tseng, W.J., and C.N. Chen. 2003. Effect of polymeric dispersant on rheological behaviour
of nickelterpineol suspensions. Materials Science and Engineering A347:145–153.
Zhou, Z., P.J. Scales, and D.V. Boger. 2001. Chemical and physical control of the rheology
of concentrated metal oxide suspensions. Chemical Engineering Science 56:2901–2920.
261
Experimental Evaluation of a Mineral
Exposure Model for Crushed Copper Ores
D. Garcia,
*
C.L. Lin,
*
and J.D. Miller
*
ABSTRACT
Copper mineral inclusions dispersed in crushed ore particles have a certain size distribution
(grainsize distribution). For efficient heap leaching processes, the crushing plant should be
designed and operated to crush the ore to an appropriate particlesize distribution so that
copper mineral grains are exposed and can be leached. In this regard, based on the approach
of Hsih, Wen, and Kuan (1995), a mineral exposure model has been evaluated to describe
the extent of grain exposure as a function of particle size. Experimental evaluation of the
mineral exposure model for different copper ores has been accomplished by 3D analysis of
crushed ore particles using cone beam xray microtomography. The model evaluation with
microCT data suggests that the extent of preferential grain boundary breakage varies both
with ore type and with particle size for a given ore type.
I NTRODUCTI ON
In the copper heap leaching process, inclusions of copper mineral grains (copperbearing
minerals) are to be dissolved from ore particles. The copperbearing minerals have some
unknown grainsize distribution, texture/exposure, and spatial distribution in the ore
particles. The procedure is to crush the ore so that the copper mineral grains are exposed
and can be dissolved during the heap leaching process. If the relationship between the
percentage of the exposed copper mineral grains and the particle size for a given ore
type can be determined, then the practical recovery in the heap leaching process can be
predicted for a specific particlesize distribution. It is, therefore, extremely important to
determine the percentage of exposed copper mineral grains for a given ore as a function
of particle size. Xray microtomography (XMT) is currently the only direct measurement
technique available for such mineral exposure analysis. In this regard, XMT has been
used to determine the copper mineral grainsize distribution and extent of exposure for a
given particlesize distribution.
For the approach used in traditional mineral processing, comminuted particles can
be classified as either free particles or locked particles. In the case of hydrometallurgy,
the fraction of grain exposure determines the extent of leaching. Thus, both liberated
and exposed grains will respond to chemical attack during leaching. The unexposed
grains that remain as inclusions in the gangue particles are not dissolved easily during
the leaching operation.
* Department of Metallurgical Engineering, University of Utah, Salt Lake City, Utah
262 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
According to generally accepted terms, “grain” pertains to the crystalline and inti
mately mixed mineral phases with welldefined boundaries, composed of distinctive
crystal chemistry and microstructure, whereas “particle” refers to crushed single and
multiphase particles, generally composed of heterogeneous crystal chemistry and tex
ture/structure. “Free particles” are particles of ore consisting of a single mineral.
“Locked particles” are particles of ore consisting of two or more minerals. “Degree of lib
eration” is the fraction of a specific mineral in the form of free particles relative to that of
both free and locked particles. “Exposed grains” are grains exposed at the surface of
locked particles, whereas “unexposed grains” are grains enclosed within other host min
eral, the gangue. The ratio of the volume of exposed valuable grains to the total volume
of both exposed and unexposed grains in the comminuted particle is defined as “the
degree of exposure” and is analogous to “the degree of liberation” for comminuted parti
cles as used in particle separation processes. Thus, as the degree of liberation defines the
limits to recovery in particle separation processes, so the degree of exposure defines
the extent of leaching that may be expected in a reasonable time.
THEORY
The concept of Hsih, Wen, and Kuan’s (HWK’s; 1995) exposure model is based on a liber
ation theory developed by Gaudin (1939). In this model, the mineral grainsize distribu
tion is related to the crushed particlesize distribution, using an intergranular fracturing
factor, P. The intergranular fracturing factor is primarily controlled by the mineralogical
characteristics of the particular feed material (P = 1 for pure intergranular fracture, pref
erential grain boundary fracture, and P = 0 for pure transgranular fracture).
The derivation of the exposure model follows the approach used by Gaudin. In the
case of a fixed grain size (d), the fractional exposure (F
E
) is given for a specific particle
size (D) as follows:
In order to derive the model, the following assumptions are made:
The ore consists of a scarce phase of valuable mineral and an abundant phase of
gangue.
Both mineral and gangue have the same uniform size of cubic grains d.
The grains are aligned in the ore so that the grain surfaces form continuous
planes.
Grains of the two species are randomly located throughout the ore.
The ore is broken into uniformly sized particles D, according to a cubic fracture
lattice either randomly or nonrandomly superimposed on the ore parallel to the
grain surfaces.
P is the probability of fractures occurring at the grain boundary, which is a real
number between 0 and 1; (1 – P) is the probability of fracture occurring as random,
transgranular breakage events during which event interfacial area is conserved.
The crushed particle is invariably larger than the size of the grain.
Although these assumptions represent no actual conditions, they provide a reasonable
foundation for initial study.
F
E
K P
K
3
K 2 –
3
–
K
3
 1 P –
K
3
K 1 –
3
–
K
3
 + = K
D
d
 =
MINERAL EXPOSURE MODEL FOR CRUSHED COPPER ORES 263
MI NERAL EXPOSURE ANALYSI S
The advanced XMT system at the University of Utah was designed and assembled to
obtain 2,048 u 2,048 pixel reconstruction over a 10mm diameter, while allowing for the
3D imaging of somewhat larger (40 mm) objects (Lin and Miller 2002). Specifically, the
specimenpositioning stage system can be manually mounted at one of three different
locations, providing system magnifications of 5, 2, or 1.25, and spheres of reconstruction
with respective diameters of 10, 25, or 40 mm. Also, the system has been designed to be
capable of handling highdensity materials, even materials having a density as high as
8.0 g/cm
3
. A photograph of the cone beam XMT system is shown in Figure 1.
ParticleSize Distribution for Heap Leaching
Copper mineral inclusions have a certain size distribution (grainsize distribution), N(X),
where N is the weight fraction of grains for particle size X (Miller et al. 2003). The heap
leaching process should be designed to crush the ore so that the copper mineral grains
are exposed and can be leached. Figures 2 to 4 present the grainsize distribution for var
ious particle size classes for Composite 2 (rhyolite/sulfide), Composite 4 (andesite/sulfide),
and Composite 6 (andesite/oxide) samples. In this regard, a 3D connected components
labeling technique was used to label and classify each individual grain volume (number
of volume elements, voxels). In this way, then, the grain size is defined as the cube root
of the grain volume. There is evidence that the copper mineral grainsize distribution is
bimodal for particle sizes greater than 1.7 mm, as revealed in the grainsizes distribution
presented in Figures 2, 3, and 4.
RESULTS AND DI SCUSSI ON
In the case of Composites 2 and 6 (Figures 5 and 6, respectively) it seems that for parti
cles greater than 10 mm in size the predominant mechanism of breakage is intergranular
fracture. In contrast, for Composite 4 (Figure 7), the predominant mechanism of break
age appears to be transgranular fracture for particles greater than 10 mm in size. One
possible explanation for this behavior, in the case of Composite 4, is that the grain sizes
FIGURE 1 The cone beam XMT system at the University of Utah
264 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
are much larger than the grain sizes for Composites 2 and 6, hence a possible difference
in fracture mechanism.
For particles smaller than 5 mm in size, there is no predominant mechanism of fracture
for Composite 2; it appears that both transgranular and intergranular mechanisms are
present. Composite 4 appears to exhibit mainly intergranular fracture and Composite 6
transgranular fracture. There is no evident explanation for this difference in behavior
between these three samples. Analysis of the fracture mechanism requires more detailed
0.15 × 0.075
0.425 × 0.15
1.7 × 0.425
3.18 × 1.7
6.3 × 3.18
9.5 × 6.3
12.7 × 9.5
19.0 × 12.7
25.4 × 19.0
37
105
420
1,680
6,300
25,400
10
20
30
40
50
60
70
P
a
r
t
ic
le
S
iz
e
C
la
s
s
,
m
m
W
e
i
g
h
t
,
%
FIGURE 2 Overall grainsize distributions of Composite 2 for different particle size classes
0.15 × 0.075
0.425 × 0.15
1.7 × 0.425
3.18 × 1.7
6.3 × 3.18
9.5 × 6.3
12.7 × 9.5
19.0 × 12.7
25.4 × 19.0
37
105
420
1,680
6,300
25,400
10
20
30
40
50
60
70
P
a
r
t
ic
le
S
iz
e
C
la
s
s
,
m
m
W
e
i
g
h
t
,
%
FIGURE 3 Overall grainsize distributions of Composite 4 for different particle size classes
MINERAL EXPOSURE MODEL FOR CRUSHED COPPER ORES 265
0.15 × 0.075
0.425 × 0.15
1.7 × 0.425
3.18 × 1.7
6.3 × 3.18
9.5 × 6.3
12.7 × 9.5
19.0 × 12.7
25.4 × 19.0
37
105
420
1,680
6,300
25,400
10
20
30
40
50
60
70
P
a
r
t
ic
le
S
iz
e
C
la
s
s
,
m
m
W
e
i
g
h
t
,
%
FIGURE 4 Overall grainsize distributions of Composite 6 for different particle size classes
0
20
40
60
80
100
0 5 10 15 20 25
Particle Size, mm
E
x
p
o
s
u
r
e
,
%
XMT
P = 0.9
P = 0.5
P = 0.1
FIGURE 5 Exposure for Composite 2 given by XMT, model using P = 0.1 (mainly transgranular
fracture), P = 0.9 (mainly intergranular fracture) and P = 0.5 (equal probability for transgranular
and intergranular fracture)
266 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
0
20
40
60
80
100
0 5 10 15 20 25
Particle Size, mm
E
x
p
o
s
u
r
e
,
%
XMT
P = 0.9
P = 0.5
P = 0.1
FIGURE 6 Exposure for Composite 6 given by XMT, model using P = 0.1 (mainly transgranular
fracture), P = 0.9 (mainly intergranular fracture) and P = 0.5 (equal probability for transgranular
and intergranular fracture)
0
20
40
60
80
100
0 5 10 15 20 25
Particle Size, mm
E
x
p
o
s
u
r
e
,
%
XMT
P = 0.9
P = 0.5
P = 0.1
FIGURE 7 Exposure for Composite 4 given by XMT, model using P = 0.1 (mainly transgranular
fracture), P = 0.9 (mainly intergranular fracture) and P = 0.5 (equal probability for transgranular
and intergranular fracture)
MINERAL EXPOSURE MODEL FOR CRUSHED COPPER ORES 267
measurements of interfacial area to determine the extent of preferential breakage of
multiphase particles for different sizes. With these data, further arguments can be devel
oped regarding the breakage mechanism. Such research is now in progress.
Thus, the mineral exposure model is limited to a fixed fraction of grain boundary
breakage independent of particle size. It seems from the results of these crushing experi
ments that the significance of grain boundary fracture (preferential breakage) varies not
only with ore type but also with particle size. For example, in the case of Composites 2
and 6 (coarse grainsize distributions as described in Figures 2 and 4), the data best fit
the model with a grain boundary fracture probability of 0.9 for coarse particle sizes. On
the other hand at finer particle sizes, the data best fit the model with a grain boundary
fracture probability of 0.1, indicating that random transgranular fracture predominates.
These results are particularly interesting because they substantiate earlier research that
suggested that for a given ore type, there may be a critical size at which the breakage
mechanism for multiphase particles changes from transgranluar fracture to intergranular
fracture (Bradt et al. 1995). Specifically, the previous results for the breakage of single,
multiphase particles defined this critical size concept as revealed in Figure 8. This figure
shows the case for which a pure particle of mineral B is stronger than a pure particle of
mineral A for all sizes (volumes) considered. Both pure minerals increase in strength
A B
Interface
Mineral “A” Mineral “B”
Binary AB
Case 2
Binary AB
Case 1
M
i
n
e
r
a
l
B
M
i
n
e
r
a
l
A
V
1
V
i
V
2
Source: Bradt et al. 1995.
FIGURE 8 Top: Schematic representation of binary mineral particle, AB; Bottom: Schematic
representation of conditions for liberation by preferential interfacial breakage of a hypothetical
binary particle
268 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
with decreasing particle size as observed by statistical analysis of experimental results
from the fracture of brittle solids and as expected from Weibull fracture statistics.
In Figure 8, case 1, the strength of the interface of the binary particle is weaker than
either of its components at a volume size (particle size) of less than V
1
. As shown in Fig
ure 7 for Composite 4, preferential fracture occurs for a particle size of less than 3.2 mm.
On the other hand, for Composite 6 (Figure 6), most of the fracture mechanism is trans
granular for a particle size of less than 6.3 mm.
SUMMARY
Based on XMTmeasured grainsize distributions, the percentage of exposure was calcu
lated from HWK’s exposure model and the calculated results compared with experimental
XMT exposure data. In general, the exposure measurements from XMT data can be
explained from the HWK exposure model.
These results are particularly interesting because they substantiate earlier research
that suggested that for a given ore type there may be a critical size at which the breakage
mechanism for multiphase particles changes from transgranular fracture to intergranular
fracture (Bradt et al. 1995).
REFERENCES
Bradt, R.C., C.L. Lin, J.D. Miller, and G. Chi. 1995. Interfacial fracture of multiphase particles
and its influence on liberation phenomena. Minerals Engineering 8(4–5):359–366.
Gaudin, A.M. 1939. Pages 70–91 in Principles of Mineral Dressing. New York: McGrawHill.
Hsih, C.S., S.B. Wen, and C.C. Kuan. 1995. An exposure model for valuable components
in comminuted particles. International Journal of Mineral Processing 43:145–165.
Lin, C.L., and J.D. Miller. 2002. Cone beam xray microtomography—a new facility for
threedimensional analysis of multiphase materials. Minerals and Metallurgical Pro
cessing 19:65–71.
Miller, J.D., C.L. Lin, C. Garcia, and H. Arias. 2003. Ultimate recovery in heap leaching
operations as established from mineral exposure analysis by xray microtomography.
International Journal of Mineral Processing 72:331–340.
269
Linking Discrete Element Modeling to
Breakage in a PilotScale AG/SAG Mill
R. Morrison,
*
B. Loveday,
†
N. Djordjevic,
*
P. Cleary,
‡
and P. Owen
‡
ABSTRACT
One of the more challenging areas in comminution is to link ore characterisation techniques
with computationally intensive modeling techniques, such as discrete element modeling
(DEM). Julius Kruttschnitt Mineral Research Centre (JKMRC) dropweight tester character
istics have been shown to be suitable for the prediction of singleevent breakage in impact
and conventional crushers. However, the prediction of particle breakage in a tumbling mill
environment is a continuing challenge because it seems very likely that breakage occurs as a
result of several mechanisms and often involves multiple events.
A range of small, wellinstrumented mills have been developed at the University of
KwaZuluNatal. These mills allow the rate of generation of fine material to be measured in
close to real time for autogenous grinding (AG) and semiautogenous grinding (SAG) mill
charges. This paper reports on the application of discrete element modeling techniques to the
power draw and charge motion within a 1.2m diameter u 300mm long pilotscale mill.
Data from a series of experiments were analysed, in which rocks ranging in size from 70 to
150 mm were tumbled at different charge levels.
An earlier regression model for the rate of attrition of the rocks is compared with a
DEMbased model, which relates the attrition rate to the rockonrock impacts. The DEM
outputs are also used to predict charge motion within the pilot mill. The DEMmodel predic
tions support the idea of an approximately constant wear rate for 20–180mm particles for
any particular load configuration after the initial particle rounding process is completed.
I NTRODUCTI ON
The purpose of this work is to compare threedimensional (3D) DEM results with experi
mental measurements of the power draw and the selfabrasion of wellworn rocks in a
pilotscale AG mill (Loveday 2004) in order to enhance the interpretation of the experi
mental results. Earlier work (Loveday and Naidoo 1997; Loveday and Whiten 2002) con
cluded that the selfabrasion rate could be considered as an approximately constant
surface wear rate over a range of particle sizes. The pilotmill experimental work (Luckan
and Pillay 2004) was undertaken to further test this idea using a largerdiameter test
mill. The particle wear rates were related to operating conditions by regression modeling
(Luckan and Pillay 2004).
* Julius Kruttschnitt Mineral Research Centre, University of Queensland, Australia
† School of Chemical Engineering, University of KwaZuluNatal, Durban, South Africa
‡ CSIRO Division of Mathematics and Information Sciences, Clayton, Victoria, Australia
270 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
The DEM work reported in this paper was undertaken to see if a simple link could be
made between the DEM collision histories for each particle size and the measured rates
of wear. In other words, did the DEM model provide any insight into why the surface
wear rate should be approximately constant? The results from two of the measured data
sets were apparently anomalous, and it was hoped that the DEM model might offer some
insight into possible reasons.
PI LOT MI LL EXPERI MENTAL WORK AND RESULTS
The pilotmill experiments were performed on quartzite rock from a quarry in the Durban
area, South Africa. These experiments were performed using a rubberlined mill with an
inside diameter of 1.19 m and length of 0.31 m. The mill was fitted with 14 square, metal
lifters with heights of 40 mm. The rotational velocity of the mill was 3.14 rad/sec (77%
of critical speed).
Two sizes of rocks were used in these tests: “large” defined as 100–130 mm; and
“small” defined as 70–100 mm. The small and large rock categories have one important
difference—relative to the mill lifters, the centre of mass of the small rocks should be less
than the mill lifter height while the centre of mass of the large rocks will be greater.
Figures 1 and 2 show realisations of the pilot mill and its rock charge made using
Particle Flow Code 3D (PFC3D; Itasca 1999) and CSIROMIS DEM codes (Cleary 2001;
Morrison and Cleary 2004), respectively.
Size reduction within AG mills is based on progressive wear of the larger rocks in the
charge. This rock wear, defined in general terms, could be due to impacts, compression,
shearing, chipping, or attrition. Initially, size reduction is rapid as asperities are removed
and the rocks become rounded. Further size reduction is much slower.
The rocks used in the test program had already been subjected to the period of rapid
wear. Each rock was weighed before and after each test and the total weight loss com
pared with the total weight of recovered fines.
The program of AG test work and the results are summarised in Table 1. Tests 9–15
were for SAG mill operation and will be reported in a later paper.
The energy consumption per ton of product is a number that is well known to opera
tors of AG/SAG mills. The energy per ton of “fines” produced from rocks is an indication
of the efficiency of size reduction by abrasion/chipping in these mills. This is directly
linked to the specific wear rate (Rs) and the holdup of rocks in the mill as follows:
Pu = P/(Rs u M) (EQ 1)
where
Pu = power utilisation
P = net power (gross power less empty running power)
Rs = measured specific wear rate of rock
M = mass of rock in the mill
Pu can be calculated from the wear rate, gross power consumption, and difference
in fragment mass between the start and finish of each test and average power utilisation
(Loveday 2004). Table 1 shows that the experimental program covered a reasonable
range of operating conditions in terms of Pu while the Rs is approximately constant over
a reasonable range of mill load masses and size distributions. However, the average val
ues for small rocks alone and large rocks alone are given in Table 2 and do appear to be
significantly different. The detailed data for each rock also support at least some degree
of size dependence.
LINKING DISCRETE ELEMENT MODELING TO BREAKAGE IN A PILOTSCALE AG/SAG MILL 271
FIGURE 1 3D DEM model of the AG test mill created by PFC3D code
FIGURE 2 3D DEM model of the AG test mill created by CSIROMIS DEM code
272 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
If the rock wear rate is a material “constant” that can be determined from testing, it
potentially offers a very simple model that might relate mill feed size distribution and
feed rate to mill load size distribution and mass (from calculated or modeled power
draw) and the mass of material leaving the mill. Obviously, a reasonable balance will be
required for stable model or mill operation.
Loveday and Whiten (2002) tested this model against several data sets from a much
smaller test mill with encouraging results.
MODELI NG APPROACH
The computing power of modern desktop computers presents an opportunity to use DEM
for mill modeling (Mishra and Rajamani 1994). In this case, the modeling has been per
formed using PFC3D (Itasca 1999) and CSIROMIS DEM (Cleary 2001; Morrison and
Cleary 2004) codes. Both codes in this case simulated the behaviour of spherical parti
cles. The particles were enclosed by the mill, which was modeled by finite volume ele
ments (PFC3D) or triangular surface mesh elements (CSIROMIS DEM). Both codes
keep a record of individual particles and update any contact with other particles or walls.
Each calculation step includes application of the laws of motion to a particle, a force
displacement law to each contact, and constant updating of wall position (Cundall and
Strack 1979).
The modeling is based on the assumption that the individual particles (balls) are
treated as stiff bodies. At contacts, rigid particles can overlap. The magnitude of the overlap
is related to the contact force. The overlaps are small relative to the size of the particles.
Individual particles can also be bonded to form clusters to mimic rock shape and strength.
During contact, the behaviour of a material is simulated using a linear contact
model. The contact force vector between two balls or ball and wall is composed of normal
and shear components. The normal contact force vector is calculated using the following
formula:
Source: After Luckan and Pillay 2004.
* The specific wear rate (Rs) of the rocks was defined as the rate of loss of mass (per unit of time) per unit of initial mass of
the particle. The dimension of this specific wear rate is inverse time (1/time).
TABLE 1 Pilotmill AG test runs and results
Run No. Mill Loading, % Rock Size
Specific Wear
Rate Rs, 1/h*
Gross
Power, W 'M, kg Pu, kWh/t
1 30 Large 0.204 833 7.149 571.5
2 20 Large 0.223 684 7.856 390.0
3 20 Small 0.241 709 3.317 888.2
4 30 Small 0.227 895 5.539 712.4
5 20 Large 0.223 701 9.477 330.8
6 30 Mixture 0.202
7 20 Mixture 0.219 694 3.531 897.9
8 25 Mixture 0.211 801 6.712 564.8
16 25 Large 0.215
17 25 Small 0.235 798 7.476 453.9
TABLE 2 Average rock wear rates for large and small particles
Rock Size Specific Wear Rate, 1/h Average Power Utilisation, kWh/t
Small 0.234 ± 0.007 684.8 ± 218.5
Large 0.216 ± 0.009 430.8 ± 125.9
LINKING DISCRETE ELEMENT MODELING TO BREAKAGE IN A PILOTSCALE AG/SAG MILL 273
Fn = Kn u Un u ni (EQ 2)
where
Fn = normal force at contact
Kn = normal stiffness at the contact
Un = relative contact displacement in the normal direction
ni = unit normal vector
The incremental shear force is calculated using the following formula:
'Fs = –Ks u 'Us (EQ 3)
where
'Fs = incremental shear force
Ks = shear stiffness at contact
'Us = incremental shear displacement at contact
Each DEM code also includes a slip model. The slip model is defined by the friction
coefficient at the contact, where the active relevant friction coefficient is taken to be the
minimum friction coefficient of the two contacting entities. Each contact is checked for
slip conditions by calculating the maximum allowable shear contact force:
Fs (max) = P u abs (Fn) (EQ 4)
where
Fs = shear component of the contact force
P = friction coefficient
abs = absolute value of (the normal force)
Fn = normal force
The energy state of the entire set of particles can be examined by recording various
forms of energy. Frictional work is defined as the total cumulative energy dissipated by
frictional sliding at all contacts. Considering that the shape of the model particles is
spherical, the calculated frictional energy is probably more similar to the slower size
reduction of rounded particles.
The model of the mill is composed of a number of weightless walls that represent
liner and lifters as well as rock and steel “balls” which represent the mill charge. The
power of the mill is calculated for each instant of time by summing products of the moments
that are applied to the mill liner and lifters and the rotational velocity of the mill.
A critical aspect of DEM is the selection of material parameters. Parameters such
as material stiffness, coefficient of friction, and damping ratio may not only affect the
value of the power draw but also the required computational time. Considering that
realistic full3D modeling of a tumbling mill may require many thousands of particles,
the ability to produce a modeling result within a reasonable time is of great practical
significance.
Previous DEM tumbling mill studies (Djordjevic 2003; Djordjevic, Shi, and Morrison
2004; Morrison, Cleary, and Valery 2001) have shown that a fullsized mill can be reli
ably represented by a vertical slice of sufficient thickness. These studies also showed that
in terms of power draw, the DEM code model power is consistent with the power draw
predicted using empirical models, even when the stiffness of the modeled material is
much lower than the measured stiffness of the rock or the steel media and mill liners.
The material parameters used for PFC3D modeling are provided in Table 3.
274 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
COMPARI SON OF MEASURED AND MODELED PI LOT MI LL POWER DRAW
Before comparing modeled power with measured power draw, it is necessary to distin
guish between the measured gross power draw and net power draw consumed by the
motion of the charge. The pilot mill was provided with a torque arm for accurate mea
surement of power consumed. However, from Figure 3, the power consumed still has a
significant zero offset of about 105 W at an extrapolated charge mass of zero. The net
mill power draw (measured – zero offset) has been used for comparison with the power
draw predicted by DEM modelling as shown in Figure 4.
The results calculated by PFC3D show good agreement with the measured net
power of the small AG mill, as shown in Figure 4. The average relative error for all of the
AG tests is 3.65%. CSIROMIS DEM performed similar modeling and achieved a slightly
better prediction (average error of 3.08%), as shown in Figure 5.
PREDI CTI ON OF ROCK WEAR RATES
As the milling environment in this case seems likely to provide predominantly frictional
interaction, modeled frictional power (the power that corresponds with the energy con
sumed as friction between particles or between particles and mill liners and lifters) was
compared with the empirical parameter Pu devised for average power utilisation (Love
day 2004; Loveday and Naidoo 1997) as defined by Equation 4.
TABLE 3 Material parameters used for PFC3D models
Density 2,650 kg/m
3
Normal contact stiffness Kn = 4e5 N/m
Shear contact stiffness Ks = 3e5 N/m
Coefficient of friction 0.5
Damping coefficient (normal and shear direction) 0.5
0
200
400
600
800
1,000
1,200
1,400
1,600
1,800
0 50 100 150 200 250
Mass, kg
P
o
w
e
r
D
r
a
w
,
W
y = 6.8606x + 102.18
R
2
= 0.9887
From data reported by Luckan and Pillay 2004.
FIGURE 3 Measured power draw of the pilot experimental mill as a function of the measured
charge mass
LINKING DISCRETE ELEMENT MODELING TO BREAKAGE IN A PILOTSCALE AG/SAG MILL 275
This comparison is presented in Figure 6 and shows a good correlation between mod
eled frictional power and measured Pu. For the two measurements, which are not on the
trend line, it is possible that either the rock type varied or there might have been an exper
imental error. However, these two results were also anomalies in the regression model.
The result shown in Figure 6 suggests that an increase in the energy consumed
through friction will result in an increase of particle abrasion in a predictable manner. As
the frictional power can be estimated for each individual particle, it offers a much more
detailed correlation than the empirical parameter, which depends on the measured
power of the mill—that is, the power drawn by all of the rocks. However, it should be
remembered that this frictional calculation is based on a simple model for collisions
between spheres.
Loveday and Naidoo (1997) monitored the mass of individual rocks within a small test
mill (0.6 m diameter) and noted that rapid rounding of the rocks occurred initially,
including failure along obvious planes of weakness. This is followed by slow but reproducible
1 2 3 4 5 7 8 17
Run No.
0
100
200
300
400
500
600
700
800
900
Measured Net
Modeled Net
P
o
w
e
r
,
W
FIGURE 4 Modeled power (PFC3D) and measured net power draw of the experimental AG mill
0
100
200
300
400
500
600
700
800
900
1 2 3 4 5 7 8 17
Run No.
P
o
w
e
r
D
r
a
w
,
W
Measured Net
Modeled Net
FIGURE 5 Modeled (CSIRO) and measured net power draw of the experimental AG mill
276 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
removal of the surface material. Loveday and Whiten (2002) have shown that if the
actual rate of wear of the rocks, per unit mass, remains relatively constant over the size
range 20 mm to 180 mm, a simple mathematical model can be applied to estimate mill load
and size distribution of coarse particles based on a mill feed rate and size distribution.
Given that the mass of a particle is proportional to the cube of its diameter, the spe
cific rate of abrasion can be expressed as
Rs = 3K/D (EQ 5)
where
Rs = specific rate of abrasion (loss of rock mass per unit of mass)
K = rate of diameter reduction (m/h)
D = particle diameter (m)
Equation 5 assumes that all particles will undergo a similar degree of relative size
reduction, regardless of their size. This implies that the rate of diameter reduction is a
rockspecific “constant,” not a function of the particle diameter. However, a more
detailed examination of the DEM results shows that smaller particles exhibit a different
pattern of motion from that exhibited by larger particles. As noted earlier, the key differ
ence is most likely the relationship between particle size and lifter height. For a constant
lifter size and shape, smaller particles will be lifted more efficiently than larger ones,
which will result in higher potential/kinetic energy and impact velocities for smaller par
ticles. Previous work at JKMRC (Djordjevic, Shi, and Morrison 2004) has indicated that
the effect of particle diameter on the modeled power draw is insignificant when the ratio
of the lifter height to particle diameter is kept constant under otherwise identical simula
tion conditions.
Therefore, the effectively constant surface wear rate observed by Loveday and
Whiten (2002) may occur as a result of two opposite trends: First, that the smaller parti
cles will be exposed to higher impact forces, which will increase their propensity for size
reduction; and second, that smaller particles (with smaller actual surface areas) will be
less prone than larger particles to size reduction due to abrasion. These two trends will
tend to cancel each other, resulting in a wear rate of rock that is relatively constant over
particle size range of 20 to 180 mm.
The DEM results also indicate that the energy consumed through shear loading at
particle contacts (opposite from normal loading) is a function of the particle size. The
smaller particles experienced an increase in the amount of net power consumed through
0 200 400 600 800 1,000
0
200
400
600
800
1,000
Modeled Frictional Power, W
M
e
a
s
u
r
e
d
A
v
e
r
a
g
e
P
o
w
e
r
U
t
i
l
i
s
a
t
i
o
n
,
k
W
h
/
t
FIGURE 6 Modeled frictional power versus power utilisation parameter Pu
LINKING DISCRETE ELEMENT MODELING TO BREAKAGE IN A PILOTSCALE AG/SAG MILL 277
shear motion at particle contacts—as shown in Figure 7. This result is in qualitative
agreement with experimental results (Table 2), which show that small particles lose
mass a little faster than large particles.
Table 4 shows the energy consumed, predicted using the CSIROMIS DEM code, for
each collision type (all collisions, rock–rock collisions, and rock–liner collisions) for various
mill volume loads. This table indicates that for increasing mill volume loading: (1) the
shear energy consumed in rock–liner collisions remains steady at about 21%; (2) the nor
mal energy consumed in rock–liner collisions decreases from 36% to 29%; (3) the shear
energy consumed in rock–rock collisions increases from 21% to 25%; and (4) the normal
energy consumed in rock–rock collisions increases from 22% to 25%. At higher mill
loadings there are more rocks in the mill, and, as expected, the normal and shear energy
consumed in rock–rock collisions make up a higher proportion of the total energy con
sumed. At 20% mill loading, only a few rocks cascade from the shoulder; most are cata
racted and impact the liner near the toe region. At higher mill loadings, more rocks
cascade down the mill, producing a thicker protective layer for the liner. So, as expected,
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
0.0 0 2 0.4 0 6 0.8 1.0
Time, , 10
1
sec
S
p
e
c
i
f
i
c
S
h
e
a
r
P
o
w
e
r
,
,
1
0
2
W
/
k
g
FIGURE 7 PFC3D estimates of specific shear power for small particles (upper curve) and large
particles (lower curve) versus time from mill startup
TABLE 4 Energy consumed by collision type at various mill volume loads for CSIROMIS DEM
models
Mill Volume
Load, %
Energy Consumed by Collision Type, % of total input energy
All Collisions Rock–Rock Collisions Rock–Liner Collisions
Normal Shear Total Normal Shear Total Normal Shear Total
20 58 42 100 22 21 43 36 22 57
25 54 46 100 24 24 48 31 21 52
30 54 46 100 25 25 50 29 21 50
278 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
the normal energy consumed in rock–liner collisions makes up a reduced proportion of
the total energy consumed.
Table 5 shows the energy consumed, which was predicted using the CSIROMIS
DEM code, for each collision type (all collisions, rock–rock collisions, and rock–liner col
lisions) for the various rock size classes. This tables indicates that for increasing rock
sizes in the mill load: (1) the shear energy consumed in rock–liner collisions increases
slightly from 20% to 22%; (2) the normal energy consumed in rock–liner collisions
increases from 29% to 35%; (3) the shear energy consumed in rock–rock collisions decreases
from 26% to 21%; and (4) the normal energy consumed in rock–rock collisions is
approximately constant at about 24%. The CSIROMIS DEM results also show that the
shear energy consumed in rock–rock collisions is a function of the particle size. The
smaller particles that are involved in rock–rock collisions show a higher proportion of
the total energy. For increasing particle size, the increase in the normal energy con
sumed in rock–liner collisions is directly related to the increase in the kinetic energy of
the cataracting stream of particles.
DI SCUSSI ON OF DEM PREDI CTI ONS OF DETAI LED CHARGE MOTI ON
Considering the detailed estimates of the motion of one large and one small (randomly
selected) particle within a mixedsize mill charge illustrates plausible reasons for the
variations in behaviour.
As shown in Figure 8, the translational velocity of the smaller particle (mass of 1.155 kg)
is significantly higher than that of the larger particle (mass of 4.478 kg). As mentioned
previously, this is most likely due to the centre of mass of the larger particle being signif
icantly larger than lifter height. Hence, the lifters will be unable to lift it as far as the
smaller particle. In contrast, the rotational velocities of the large and small particles are
approximately the same, as shown in Figure 9.
The larger particle will not be lifted as efficiently as the smaller one, which is clearly
visible from the particle pattern of motion in the vertical plane of the mill, as shown in Fig
ures 10 and 11. Due to its higher dropoff point, the smaller particle has a higher maximum
velocity at each moment of impact, as well as a higher residual velocity after impact.
This discussion also can be considered in terms of the empirical model developed by
Luckan and Pillay (2004). The overall particle wear rate was regressed against feed siz
ing, mill loading, and percentage of steel in the charge. This regression model suggested
slightly higher specific wear rates for the smaller particles than the larger—which is con
sistent with the previous discussion. This empirical approach is not well suited to the
mixedsize mill charges.
This difference in trajectories provides a plausible rationale for the quite different lev
els of specific shear power per unit of mass and per unit of surface area to be 98 W/kg and
4,071 W/m
2
for the small particle and 18.9 W/kg and 1,260 W/m
2
for the large particle.
The spatial distribution of shear forces was divided into three ranges—10 to 100 N,
100 to 1,000 N, and >1,000 N—and plotted in Figures 12 to 14. These figures show that
TABLE 5 Energy consumed by collision type for various rock sizes for CSIROMIS DEM models
Rock Size
Class
Energy Consumed by Collision Type, % of total input energy
All Collisions Rock–Rock Collisions Rock–Liner Collisions
Normal Shear Total Normal Shear Total Normal Shear Total
Small 54 46 100 24 26 50 29 20 50
Mix 55 45 100 24 24 47 31 21 53
Large 57 43 100 22 21 43 35 22 57
LINKING DISCRETE ELEMENT MODELING TO BREAKAGE IN A PILOTSCALE AG/SAG MILL 279
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
0.0 0.2 0.4 0.6 0.8 1.0
Time, , 10
1
sec
T
r
a
n
s
l
a
t
i
o
n
a
l
V
e
l
o
c
i
t
y
,
m
/
s
FIGURE 8 PFC3D estimates of the time history of the translational velocity of a small particle
(black—higher maxima) and a large particle (gray)
0.5
1.0
0.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
5.5
6.0
6.5
7.0
7.5
8.0
0.0 0.2 0.4 0.6 0.8 1.0
Time, , 10
1
sec
R
o
t
a
t
i
o
n
a
l
V
e
l
o
c
i
t
y
,
,
1
0
1
r
a
d
/
s
FIGURE 9 PFC3D estimates of the time history of the rotational velocity of a small particle
(black) and large particle (gray) showing minimal difference
280 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
–6.0
–6.0
–4.0 –2.0 0.0 2.0 4.0 6.0
–5.0
–4.0
–3.0
–2.0
–1.0
0.0
1.0
2.0
3.0
4.0
5.0
6.0
Horizontal Coordinate, m, ×10
–1
V
e
r
t
i
c
a
l
C
o
o
r
d
i
n
a
t
e
,
m
,
×
1
0
–
1
FIGURE 10 Spatial pattern of motion in the vertical plane of a randomly selected single small
particle in a mixed size charge in the PFC3D model
–6.0
–6.0
–4.0 –2.0 0.0 2.0 4.0 6.0
–5.0
–4.0
–3.0
–2.0
–1.0
0.0
1.0
2.0
3.0
4.0
5.0
6.0
Horizontal Coordinate, m, ×10
–1
V
e
r
t
i
c
a
l
C
o
o
r
d
i
n
a
t
e
,
,
m
,
×
1
0
–
1
FIGURE 11 Spatial pattern of motion in the vertical plane of a randomly selected single large
particle in a mixed size charge in the PFC3D model
LINKING DISCRETE ELEMENT MODELING TO BREAKAGE IN A PILOTSCALE AG/SAG MILL 281
–6.0
–6.0
–4.0 –2.0 0.0 2.0 4.0 6.0
–5.0
–4.0
–3.0
–2.0
–1.0
0.0
1.0
2.0
3.0
4.0
5.0
6.0
Horizontal Coordinate, m, ×10
–1
V
e
r
t
i
c
a
l
C
o
o
r
d
i
n
a
t
e
,
m
,
×
1
0
–
1
FIGURE 12 Spatial pattern in the vertical plane of shear contact forces of 10 to 100 N
–6.0
–6.0
–4.0 –2.0 0.0 2.0 4.0 6.0
–5.0
–4.0
–3.0
–2.0
–1.0
0.0
1.0
2.0
3.0
4.0
5.0
6.0
Horizontal Coordinate, m, ×10
–1
V
e
r
t
i
c
a
l
C
o
o
r
d
i
n
a
t
e
,
m
,
×
1
0
–
1
FIGURE 13 Spatial pattern in the vertical plane of shear contact forces with intensities of 100 to
1,000 N
282 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
low and mediumintensity shear forces occur predominantly within the charge, whereas
strong shear forces occur during freefall and at the resultant impact. Significantly higher
intensity shear forces also occur in the base of the mill charge at the interface between
the mill liners and the particles.
CONCLUSI ONS
Based on the results presented, 3D DEM can accurately predict net power draw of the
pilot experimental AG mill. This success does not necessarily imply similar accuracy in
prediction of power draw in operating mills, because only a small quantity of ore fines
and water are present in this pilot mill.
As the noload power of the larger AG mills can be predicted with sufficient accuracy
using established empirical models (NapierMunn et al. 1996), the DEM model may well be
capable of accurate prediction of total mill power draw if slurry behaviour can be included.
A strong correlation exists between the observed power utilisation and modeled net
frictional power consumed within the mill charge. Therefore, subject to further valida
tion of DEM, the modeled netfrictional power may be able to be used in a similar manner
to the empirical power utilisation parameter for optimisation of performance of industrial
scale AG mills.
The insight offered into the interactions between lifter height and particle size goes
well beyond liner damage caused by excessive liner height and offers the possibility of
liner design for maximum production.
The DEM results provide a rationale for the approximately constant rockwear rate
reported in the literature—even though there are significant differences due to the inter
action with the mill lifters by different sized particles. Hence, there is also a strong possi
bility that a single, measurable characteristic may be used to model this type of rock
wear when combined with a DEM model.
–6.0
–6.0
–4.0 –2.0 0.0 2.0 4.0 6.0
–5.0
–4.0
–3.0
–2.0
–1.0
0.0
1.0
2.0
3.0
4.0
5.0
6.0
Horizontal Coordinate, m, ×10
–1
V
e
r
t
i
c
a
l
C
o
o
r
d
i
n
a
t
e
,
m
,
×
1
0
–
1
FIGURE 14 Spatial pattern in the vertical plane of shear contact forces >1,000 N
LINKING DISCRETE ELEMENT MODELING TO BREAKAGE IN A PILOTSCALE AG/SAG MILL 283
ACKNOWLEDGMENTS
The DEM modeling work reported in this paper has been partially funded by the Centre
for Sustainable Resource Processing.
REFERENCES
Cleary, P.W. 2001. Recent Advances in DEM Modelling of tumbling mills. Minerals Engi
neering 14:1295–1319.
Cundall, P.A., and O.D.L. Strack. 1979. A discrete numerical model for granular assemblies.
Geotechnique 29(1):47–65.
Djordjevic, N. 2003. Discrete element modelling of the influence of lifters on power draw
of tumbling mills. Minerals Engineering 16(4):331–336.
Djordjevic, N., F.N. Shi, and R.D. Morrison. 2004. Determination of lifter design, speed and
filling effects in AG mills by 3D DEM. Minerals Engineering 17(11–12):1135–1142.
Itasca Consulting Group. 1999. PFC3D (Particle Flow Code in 3 Dimensions). Minneapolis,
MN: Itasca Consulting Group.
Loveday, B.K. 2004. The use of FAG and SAG batch tests for measurement of abrasion
rates of fullsize rocks. Minerals Engineering 17(11–12):1093–1098.
Loveday, B.K., and D. Naidoo. 1997. Rock abrasion in autogenous milling. Minerals Engi
neering 10(6):603–612.
Loveday, B.K., and W.J. Whiten. 2002. Application of a rock abrasion model to pilotplant
and plant data for fully and semiautogenous grinding. Transactions of the Institution
of Mining and Metallurgy 111:C39–C43.
Luckan, P.I., and K. Pillay. 2004. The development of an autogenous model for quartzite
by using semibatch laboratoryscale milling. Laboratory Project 2004 DNC4IP1.
Durban, South Africa: University of KwaZuluNatal. Unpublished.
Mishra, B.K., and R.K. Rajamani. 1994. Simulation of charge motion in ball mills. Part 1:
Experimental verifications. International Journal of Mineral Processing 40(3–4):
171–186.
Morrison, R., and P.W. Cleary. 2004. Using DEM to model ore breakage within a pilot
scale SAG mill. Mineral Engineering 17:1117–1124.
Morrison, R., P.W. Cleary, and W. Valery. 2001. Comparing power and performance
trends from DEM and JK modelling. Pages 284–300 in SAG 2001. Volume IV. Vancou
ver, BC: University of British Columbia, Department of Mining and Mineral Process
Engineering.
NapierMunn, T.J., S. Morrell, R.D. Morrison, and T. Kojovic. 1996. Mineral Comminution
Circuits: Their Operation and Optimisation. Brisbane, Australia: Julius Kruttschnitt
Mineral Research Centre.
285
Significance of the ParticleSize
Distribution in the Quality of Cements
with Fly Ash Additive
Viktória Gável
*
and Ludmilla Opoczky
*
ABSTRACT
The fineness of fly ash used as a cement additive influences the quality of the cement. The
cement industry characterizes fineness based on the Blaine surface area. However, according
to our investigations, the Blaine surface does not reflect exactly the actual fineness and particle
size distribution of fly ash or cements with fly ash added.
Cements with fly ash added are characterized by “coarser” and “narrower” particlesize
distributions than cements without fly ash added, but both types of cement typically have
approximately the same Blaine surface area.
The value of specific surface area calculated from the particlesize distribution of fly
ash, using the exponential approximation method, gives a better estimate of its real fine
ness. Given this knowledge, the fineness and, therefore, the quality of cements with fly ash
additive can be influenced favorably during cement production.
I NTRODUCTI ON
Fly ash—the byproduct of coal powder–fired thermal power stations—has been utilized
as an additive to cement for several decades.
Recently, several problems concerning the production of composite cements that
contain fly ash were raised regarding the technology of grinding and the analysis of par
ticle size. The most important issues were (1) the role of the fineness (specific surface
and particlesize distribution) of fly ash in the development of the cement quality; and
(2) the revision and improvement of methods used for characterization and testing of
the fineness (Opoczky and Juhász 1990). This paper presents the principal results of our
investigations that were carried out in this field.
EXPERI MENTAL MATERI ALS AND METHODS
To conduct the experiments, we used various types of cements produced in Hungarian
cement plants, as well as fly ash from two Hungarian power stations. The quality of the
materials investigated (strength, water demand, pozzolanic activity, etc.) was tested
according to the related European and Hungarian standards.
Particle composition of the ground products was determined by a Cilas (Marcoussis,
France) Model 715 laser granulometer. For the characterization of the particlesize
* Research & Development Ltd. for the Cement Industry (CEMKUT Ltd.), Budapest, Hungary
286 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
distribution, we used two parameters from the RosinRammlerSperlingBennett (RRSB)
equation: the fineness number ( ) and the uniformity coefficient (n). Uniformity coeffi
cient n characterizes the dispersion (width) of the particlesize distribution curve (i.e.,
the lower the value of n, the “wider” [more disperse] the particlesize distribution). Fine
ness number characterizes the fineness of the ground product (i.e., the smaller the
value of , the finer the ground product; Beke 1981).
(EQ 1)
where
R(x) = oversize
x = particle size
= fineness number
Fineness and degree of dispersion of the initial materials and the ground products
were characterized by their specific surface, as determined by the widely used Blaine
apparatus (based on measuring the permeability of a packed bed of powder). This is
commonly referred to as the “Blaine surface” (S), as it is generally taken to be related to
the specific surface area of the particles. Fineness was also calculated from the particle
size distribution data using an exponential approximation method referred to as “calcu
lated surface.” The essence of this calculation is that the individual particles are assumed
to be spherical, which is similar to the laser granulometric analysis where the apparatus
expresses the size of the particles through the diameter values of the equivalent spheres.
In this way, the specific surface area, S
mg
, for the assembly consisting of different contin
uous spheres of various sizes can be expressed as
(EQ 2)
where
x = particle size (i.e., the diameter of the equivalent sphere)
f(x) = function describing the particlesize distribution (frequency curve)
A single function seldom describes the size distribution exactly; therefore, approxi
mations were made for the individual sections of the curve using either the same type of
functions with various parameters or using different types of functions. Additionally, the
definite integrals were summarized by sections.
Knowing that the smallest particles play a decisive role in constituting the specific
surface, and that the individual particle classes are usually not sufficiently narrow as
compared to the size of the particles, a more exact result was achieved when using an
exponential approximation to the distribution function.
The specific surface by particle classes could be calculated using the following formula:
(EQ 3)
The exponent of the power function by particle classes also could be calculated as
follows:
x
x
x
R x e
x
x

© ¹
§ ·
–
=
x
S
mg
6
U
S

1
x

x
min
x
max
³
f x dx =
'S
mgi
m
i
m
i
1 –

F
i
x
i

F
i–1
x
i–1
 –
© ¹
§ ·
=
PARTICLESIZE DISTRIBUTION IN THE QUALITY OF CEMENTS WITH FLY ASH ADDITIVE 287
(EQ 4)
where
m
i
= exponent of the power function for the given particle class
F
i
= cumulative distribution function for the given particle class
RESULTS AND DI SCUSSI ON
The degree of dispersion, or fineness, of fly ash used as an additive to cement has a sig
nificant influence on the quality of the fly ash and that of the composite cements with fly
ash additive (Opoczky 1996; Opoczky and Tamás 2002).
The fineness of fly ash is characterized in the cement industry by the Blaine surface,
and the fly ash is also qualified by this value. According to our investigations, assessment
by the Blaine surface value does not provide adequate information on the actual particle
composition or fineness of the fly ash (Opoczky 2001; Gável 2003). This is shown in Fig
ure 1 and Table 1 where the particlesize distribution of two products of about the same
Blaine surface, ground clinker and fly ash, are illustrated in the RRSB system of coordi
nates (DIN [German Institute for Standardization] 66145), accepted in both the Euro
pean and Hungarian practice.
Particlesize distribution of the fly ash that had about the same Blaine surface
(
~
3,500 cm
2
/g) when characterized by the fineness number ( = 60 Pm) proved to be
much more coarse than that of the ground clinker ( = 18 Pm). In order to approach the
particlesize distribution of the clinker, fly ash had to be ground to a Blaine surface of
~
6,000 cm
2
/g.
A similar conclusion also resulted in the case of fly ash and cement(s) when the
Blaine surface values were compared with those calculated from particlesize distribu
tion measurements data using the exponential approximation method (calculated sur
faces) (Table 2).
In the case of cement(s) with no additive, the calculated surface values (arrived at
from the particlesize distribution measurements data using the exponential approximation
method) do not differ significantly from the Blaine surface determined by the permeabil
ity method. However, in the cases of fly ash and of cements with fly ash additive, the dif
ference between the Blaine surface value and the calculated surface value is significant.
The difference between the Blaine surface values of cement and fly ash can be
explained, on the one hand, by a significant difference in their particlesize distribution
and, on the other hand, by the fact that the fly ash always contains—in addition to rela
tively coarse particles—very fine particles of elementary carbon, the presence of which
significantly increase the Blaine surface value.
The particlesize distribution of fly ash plays an essential role in the quality of com
posite cements containing fly ash additive.
According to our investigations, there is a definite relationship between the unifor
mity coefficient, n, of composite cements and their water demand. Cements of higher
uniformity coefficient (n)—that is, of more “narrow” particlesize distribution—usually
require more water. The more narrow the particlesize distribution, the less tightly the
particles can pack together, and so more water is required to fill the pores and gaps
(Opoczky and Tamás 2002).
As the fly ash has more narrow particlesize distribution than the cements, it
increases the uniformity coefficient (n) of the cements and simultaneously increases its
water demand when being added to the cement (Figure 2). This adversely affects the
m
i
lg F
i
F
i–1
e
lg x
i
x
i–1
e
 =
x
x
288 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
strength, workability, and other application properties of the cements and mortars or
concretes made of such cements.
By applying an adequate fine grinding, the particlesize distribution of the fly ash can
be influenced favorably.
Interrelations between the fineness number ( ) and Blaine surface value of fly ash are
shown in Figures 3 and 4. From Figures 3 and 4 one can determine what Blaine surface
value the fly ash should be ground to in order for its fineness and particlesize distribution
( – fineness number, calculated surface) to approach the fineness characteristics of the
cement without fly ash additive. For example, in order to achieve a fineness number ,
~
25 Pm the fly ash should be ground to a Blaine surface value of at least 5,500 cm
2
/g.
1 10 100 1,000
1
10
36.8
75
90
95
99
99
90
75
50
25
10
5
1
Clinker, 3,500 cm
2
/g
Fly ash, 6,600 cm
2
/g
Fly ash, 3,500 cm
2
/g
2
/g
Particle Size, µm
O
v
e
r
s
i
z
e
,
%
U
n
d
e
r
s
i
z
e
,
%
FIGURE 1 Particlesize distribution
TABLE 1 Particlesize distribution and Blaine surface
Denomination
of Materials
Investigated
Blaine
Surface,
cm
2
/g
Parameters of the RRSB Equation Fraction Composition, %
Fineness
Number, μm
n
Uniformity
Coefficient 0–3 μm 3–32 μm 32–192 μm
Ground clinker ~3,500 ~18 0.9013 15.90 64.00 20.10
Original fly ash ~3,500 ~60 1.0870 3.70 34.70 61.60
Ground fly ash ~6,000 ~19 1.0835 13.20 67.70 19.10
x
* European Standard: EN 1971.
TABLE 2 Fineness characteristics of fly ash and cements with fly ash additive
Denomination of
Materials Investigated
Fly Ash
Content,
m/m %
Parameters of the RRSB Equation
Blaine
Surface,
cm
2
/g
Calculated
Surface, cm
2
/g
Fineness
Number, μm
n
Uniformity
Coefficient
Original fly ash
*
100 95 1.0237 3,460 1,940
Ground fly ash 100 48 1.0342 3,840 3,230
CEM I 42.5N
*
0 19 0.9953 3,570 3,620
CEM II/AV 42.5N
*
20 21 0.9555 3,750 3,640
CEM II/AV 32.5R
*
20 25 0.9575 3,590 3,420
CEM II/BV 32.5N
*
35 24 1.0249 3,340 3,190
x
x
x
x
PARTICLESIZE DISTRIBUTION IN THE QUALITY OF CEMENTS WITH FLY ASH ADDITIVE 289
n
Wd
0.90
0.95
1.00
1.05 36
34
32
30
28
26
24
0
100
10
90
20
80
35
65
Fly Ash
Clinker
m
/
m
%
U
n
i
f
o
r
m
i
t
y
C
o
e
f
f
i
c
i
e
n
t
(
n
)
W
a
t
e
r
D
e
m
a
n
d
(
W
d
)
,
m
/
m
%
FIGURE 2 Change of the uniformity coefficient (n) and water demand (Wd) of the cement,
depending on the fly ash content
0
10
20
30
40
50
60
70
80
2,000 3,000 4,000 5,000 6,000 7,000
Blaine Surface (S), cm
2
/g
F
i
n
e
n
e
s
s
N
u
m
b
e
r
(
x
)
,
µ
m
FIGURE 3 Connection between the fineness number and Blaine surface of ground fly ash
0
500
1,000
1,500
2,000
2,500
3,000
3,500
4,000
4,500
2,000 3,000 4,000 5,000 6,000 7,000
Blaine Surface (S), cm
2
/g
C
a
l
c
u
l
a
t
e
d
S
u
r
f
a
c
e
,
c
m
2
/
g
FIGURE 4 Connection between the calculated surface and Blaine surface of ground fly ash
290 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
Other examples of this connection are
It is possible to reduce the adverse effect of the fly ash additive on the strength—
particularly the initial strength—of the cement, as shown in Figure 5.
It is possible to reduce the adverse effect of the fly ash additive on the water
retaining capability of the cement (Figure 6).
It is possible to reduce sulfatecaused expansion of the cement or to produce
cements of increased sulfate resistance (Figure 7).
CONCLUSI ONS
The fineness of fly ash used as a cement additive has a significant influence on the qual
ity of composite cements containing fly ash. In the cement industry, the fineness of fly
ash is characterized through the Blaine surface, which is determined by a permeability
method.
According to our investigations, the Blaine surface value does not provide adequate
information on the actual particle composition of the fly ash. Namely, in the case of hav
ing approximately the same Blainespecific surface value, fly ash is usually characterized
by coarser and narrower particlesize distribution ( = fineness number, n = uniformity
coefficient) than cements without additives.
We arrived at a similar conclusion when comparing the Blainesurface and calculated
surface values of fly ash and cements.
Knowing the correlations between particlesize distribution, calculated surface, and
Blaine surface of the fly ash, one can establish to what Blaine surface value the fly ash
should be ground in order for its fineness and particlesize distribution to approach the
fineness characteristics of the cement.
0
0
10
20
30
40
50
60
10
20
35
3
7
28
Fly Ash Content,
m
/
m
%
C
o
m
p
r
e
s
s
i
v
e
S
t
r
e
n
g
t
h
,
N
/
m
m
2
Blaine surface (S) of
cement with fly ash
content in cm
2
/g
Days
3,000
3,500
4,000
FIGURE 5 Change of compressive strength of cement, depending on fly ash content and fineness
of cement (cogrinding)
x
PARTICLESIZE DISTRIBUTION IN THE QUALITY OF CEMENTS WITH FLY ASH ADDITIVE 291
Through the proper adjustment of the particlesize distribution of the fly ash, the
properties (water demand, water retaining capability, strength, sulfate resistance, etc.)
of fly ash containing composite cements can be influenced favorably.
REFERENCES
Beke, B. 1981. The Process of Fine Grinding. Volume 1. Developments in Mineral Science
and Engineering series. Budapest, Hungary: Akadémiai Kiadó and Martinus Nijhoff/
Dr. W. Junk Publishers.
Gável, V. 2003. Description of grinding fineness of flyash and cements with flyash (in
Hungarian). Paper presented at the 20th Cementipari Konferencia, Hortobágy
Máta, Hungary, October 13–15.
50
60
70
80
90
100
3,200
Cement
2,200
Fly Ash
3,000
Fly Ash
4,000
Fly Ash
Blaine Surface (S), cm
2
/g
W
a
t
e
r

R
e
t
a
i
n
i
n
g
C
a
p
a
b
i
l
i
t
y
,
m
/
m
%
FIGURE 6 Waterretaining capability of cements with 20% different fineness of fly ash
CEM
I
42.5
10 15
Fly Ash Content,
m
/
m
%
0.4
0.8
1.2
0
E
x
p
a
n
s
i
o
n
,
m
m
/
m
Fly Ash S
Blaine
, cm
2
/g
~3,000 ~4,000
S54
Cement
10 15
Fly Ash Content,
m
/
m
%
0.3
0.4
0.5
0.2
E
x
p
a
n
s
i
o
n
,
m
m
/
m
Fly Ash S
Blaine
, ~6,000 cm
2
/g
FIGURE 7 Effect of different fineness of fly ash on the sulfate resistance of cements in 28 days
292 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
Opoczky, L. 1996. Grinding technical questions of producing composite cement. Interna
tional Journal of Mineral Processing 44–45:395–404.
———. 2001. Silicatechemical properties of flyashes. Pages 255–262 in Oilfield Chemistry.
Volume 3. Edited by I. Lakatos. Budapest, Hungary: Akadémiai Kiadó.
Opoczky, L., and A.Z. Juhász. 1990. Mechanical Activation of Minerals by Grinding: Pul
verizing and Morphology of Particles. Budapest, Hungary: Akadémiai Kiadó and Ellis
Horwood Publishers.
Opoczky, L. and F. Tamás. 2002. Multicomponent composite cements. Pages 559–594 in
Advances in Cement Technology: Chemistry, Manufacture and Testing. 2nd edition.
Edited by S.N. Gosh. New Delhi, India: Technical Books International.
293
Modeling Attrition in Stirred Mills
Applying Statistical Physics
Thomas Neesse,
*
Johann Dück,
*
and Friedrich Schaaff
*
ABSTRACT
Grinding attrition is a new purification process that can be used for mineral residues. It can
be applied to efficiently remove surfaceadsorbed contaminants in the particle size range
<100 µm. The process is a special case in comminution technology to be performed in con
ventional stirred mills. The operating parameters are selected so as to avoid particle break
age. For this process, a mathematical model was developed, which describes the kinetics of
the grinding attrition on the basis of the microprocess (i.e., the attrition at an individual
particle). In a next step, the model was extended to the macroprocess, applying principles
of statistical physics for the entire milling chamber. The model was validated on the basis of
experimental data.
I NTRODUCTI ON
Grinding attrition is a newly developed purification process for fine particles of contami
nated soils and mineral residues, where the contamination is fixed adsorptively at the
particle surface (Schricker, Dueck, and Neesse 1998). Prior to grinding attrition, the sandy
feed material is deslimed at about 20 µm. Subsequently, the particle outer layer is
abraded and collected in a highly contaminated fines fraction <10 µm, which is then
removed in hydrocyclones. In grinding attrition, normally conventional stirred mills, as
used in comminution technology, are utilized. To prevent particle breakage, special process
conditions must be selected. Typical process conditions are grinding media size/maximum
particle size
~
1.5–2.0; solids concentration of the feed suspension 0.5–0.56 vol. %; and
specific energy input
~
20–30 kWh/t. Specific knowledge of grinding attrition from labo
ratory up to the industrial scale (Tiefel, Schricker, and Neesse 1999) is now available.
This paper deals with the physical modeling of grinding attrition, which can be under
stood as a special case of comminution in stirred mills.
LAB SCALE EXPERI MENTS
Labscale grinding attrition tests were performed with the test setup shown in Figure 1.
It consisted of a 1.5L and a 3.0L laboratory mill with a vertical attritor shaft. The test
material was quartz sand with a mean particle diameter of 92 µm. Selected experimental
conditions to be used to validate the model are listed in Table 1.
During the discontinuous tests, both the temporal change of the torque of the attri
tor shaft and the change of the particlesize distribution of the feed suspension (i.e., fines
* FriedrichAlexander University, ErlangenNuremberg, Germany
294 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
generation as a result of attrition) were determined. The total mill volume was almost
ideally mixed and samples were taken at different positions, showing fluctuations of the
particlesize distribution of <1%. The total error was determined by fluctuations of the
feed size distribution and amounted to
~
3%. The current status of the process was eval
uated by taking two samples of
~
3 g after interrupting the process.
MODELI NG OF THE GRI NDI NG ATTRI TI ON MI CROPROCESS
Within comminution, the modeling of the grinding attrition is comparatively simple
because only one fine fraction is produced, as demonstrated in Figure 2.
Variable n
Measured Value: M
PC
SpeedControlled Drive
RingType Stirrer
Sampling
ParticleSize
Analysis
Attrition Container
Water Cooling
Water Cooling
Water, Attrition Material,
Grinding Media
FIGURE 1 Test rig for grinding attrition
TABLE 1 Experimental conditions
Parameters and Experimental Conditions Case 1 Case 2 Case 3
Number of shaft revolutions (min
–1
) 1,400 1,600 1,200
Grinding media (size d
GM
= 200–400 μm) Steel spheres Glass spheres Steel spheres
Volume ratio of grinding media to the contaminated
feed particle (–)
1 1 1
Solids volume concentration (grinding media + attrition
material) of suspension (–)
0.54 0.56 0.56
Initial value of the torque (Nm) 0.704 0.434 0.678
MODELING ATTRITION IN STIRRED MILLS APPLYING STATISTICAL PHYSICS 295
The following simplifying assumptions are made for process modeling:
1. All grinding media are of the same size.
2. The particlesize distribution of the attrition material can be described by a mean
particle diameter.
3. The particlesize distribution of the attrition material remains approximately
constant during attrition.
4. The fine particles produced have a narrow size distribution and also are charac
terized by a mean particle diameter.
5. The fines result only from friction impact between the grinding media and the
particles of the attrition material.
6. The mass ratio between the grinding media and the attrited particles remains
approximately constant during the process because the produced fines fraction is
less than 0.1.
The starting point for the analysis of the grinding attrition is the statement (Joost
and Schwedes 1996) that, in microgrinding, two parameters are the deciding factors: the
stress intensity and the stress number. For single particles of the attrition material, anal
ogous proceedings should be applied:
(EQ 1)
where
e
ges
= entire energy transfer to a particle in respect to a grinding body (J/particle)
t = attrition time (sec)
h = stress frequency of a particle (–)
e = stress intensity (J/particle)
e = stress number of a particle respectively a grinding ball (L/sec)
M = torque of the shaft, (J)
N
tot
= total number of the particles and grinding media, (–)
More simply stated, it is assumed that the energy input is evenly distributed between all
impact partners N
tot
. This is justified because grinding media and the attrition material are
close in size.
P
a
r
t
i
c
l
e

S
i
z
e
D
i
s
t
r
i
b
u
t
i
o
n
ParticleSize
Distribution of
the Generated
Fines
ParticleSize
Distribution after
the Attrition
ParticleSize
Distribution before
the Attrition
ParticleSize
Distribution of the
Grinding Media
d
AM
d
r
d
GM
Particle Diameter
FIGURE 2 Changes of the particlesize distribution during attrition (schematically)
e
ges
t ( ) h e et
M t ( )
N
tot
 = =
296 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
The special conditions of grinding attrition are characterized by the facts that only
one fines fraction is produced by abrasion and that the stress intensity of the attrition
material decreases with the increasing fraction of the fines generated (Schricker 1999).
This decrease in friction between the particles due to fine particle layers is shown
schematically in Figure 3. The fines form a cushion between the impact partners, which
attenuates the stress intensity. The reduction of the measured torque of the agitator with
increasing fines fraction m
f
is shown in Figure 4.
The obvious proportionality can be written as follows:
(EQ 2)
where
M
0
= torque at t = 0 (Nm)
m
f
= mass ratio of the fines at time t (–)
e
0
= stress intensity of the particles at t = 0 on condition that m
f
= 0 (J)
o is a measure for the attenuation of the particle impact due to the produced fines. It can
be taken from the slope of the straight line from Figure 4. On the basis of Equation 1,
Equation 3 then applies:
(EQ 3)
In the following discussion, the microprocess of the friction impact between a grind
ing ball and a particle is studied.
In the attrition mill, the grinding media/particle filling exhibits a pseudoturbulent
movement, obeying statistic laws. Thus, the stress intensity of a particle is a fluctuating
value. This means that the necessary energy barrier sufficient for abrasion can only be
exceeded at a limited number of impacts. This is schematically shown in Figure 5. Sim
plifying, it was assumed that there is no stochastic oscillation of the critical intensity.
The probability that a particle is stressed so much during grinding attrition by a fric
tion impact that a fine particle of the size d
f
is formed can be formulated according to
statistic physics as follows (Levich 1997):
(EQ 4)
where
w = probability for a successful friction impact (–)
e
A
= specific surface energy formed by attrition (J/m
2
)
d
f
= size of the fine particle produced (m)
Equation 4 expresses that the attrition result is connected with the increase of the specific
surface energy by the value e
A
d
f
2
. The kinetics of fines production by grinding attrition
can now be formulated by the following kinetic equation:
(EQ 5)
where
m
f
= mass ratio of fines at time t
e = stress number (L/sec)
M t ( )
M
0

e t ( )
e
0
 1 om
f
t ( ) – ( ) = =
e t ( ) 1 om
f
– ( )e
0
=
w e >> e
A
d
f
2
( ) ( )
e
A
d
f
2
e
 –
\ .
 
exp =
dm
f
dt
 e w e
e
A
d
f
2
1 om
f
– ( ) e
0
 – exp = =
MODELING ATTRITION IN STIRRED MILLS APPLYING STATISTICAL PHYSICS 297
It should be noted that in accordance with the assumption in Equation 5, the mass of
the stressed particles is not considered (which is not usual in comminution kinetics).
Equation 5 applies initially to the attrition process of one particle of the attrition material.
Assuming the same initial conditions for all particles, however, this relationship also
describes the overall process.
The approximate integration of the Equation 5 leads to the following relationship:
(EQ 6)
Grinding
Media
Grinding
Media
Grinding
Media
Attrition
Material
Attrition
Material
Attrition
Material
One Impact
Full Attrition
ith Impact
Reduced Attritiion
nth Impact
No Attrition
FIGURE 3 Microprocess of the grinding attrition
1.000
0.875
0.750
0.625
0.500
M
/
M
0
(
–
)
0 0.05 0.10 0.15 0.20 0.25 0.30
m
f
(–)
Case 1: M/M
0
= –1.57 m + 1.03
Case 2: M/M
0
= –2.02 m + 1.05
Case 3: M/M
0
= –1.67 m + 1.03
Case 1
Case 2
Case 3
Regression
FIGURE 4 Relation between torque of the attritor agitator and produced fines m
f
m
f
m
f0
1 om
f0
– ( )
2
o
 1
t
t
c
 +
\ .
 
ln + =
298 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
with
(EQ 7)
and
(EQ 8)
where  is a measure for the free surface energy produced under certain stress intensities,
and t
c
is a characteristic attrition time.
Furthermore, on the condition t > t
c,
the following is obtained:
(EQ 9)
With this equation, the parameters o, , and t
c
can be determined using experimental
data. The total quantity of the produced fines Am
f
during attrition can be determined on
the basis of Equations 5 and 8, according to which the increase in the fine particles
diminishes over time. The production rate can be calculated according to the following
relationship:
(EQ 10)
Here a process duration t
p
is to be defined, at which the production rate of the fines
decreases on 0.01. This corresponds to a time interval of t
p
= 99 t
c
and is at the same time
the interpretation of the parameter t
c
.
Stress Intensity
Critical Intensity for Abrasion
O
s
c
i
l
l
a
t
i
n
g
o
f
S
t
r
e
s
s
I
n
t
e
n
s
i
t
y
Time
FIGURE 5 Oscillating stress intensity of a particle in the attrition mill (schematic)

e
A
d
f
2
e
0
 =
t
c
1 om
f0
– ( )
2
o  e


1 om
f0
–

\ .
 
exp =
m
f
m
f0
1 om
f0
– ( )
2
o

t
t
c

\ .
 
ln + =
dm
f
dt

dm
f
dt

t=0
1
1 t t
c
+
 =
MODELING ATTRITION IN STIRRED MILLS APPLYING STATISTICAL PHYSICS 299
For the total fines production Am
f
, the following results from Equation 6:
(EQ 11)
Figure 6 shows the measured fines fraction over ln t for three experimental data
sets. The linear progression of this dependence confirms the validity of Equation 9. The
parameters of the process equation for the three experiments are listed in Table 2.
Equation 5 is thus the result of a physically based model of the attrition process, pro
ceeding from a statistic distribution of the stress intensity in the milling chamber. With
this model, it is possible to present the fines production in the form of a dimensionless
equation with three parameters. These parameters o, , and t
c
must be experimentally
determined.
MACROPROCESS OF THE GRI NDI NG ATTRI TI ON
In the previous section, grinding attrition was studied as a microprocess. Figure 7 illus
trates the model of the macroprocess referring to the entire milling chamber and consid
ering the main process parameters. These are
Size and density of the grinding media d
GM
, µ
GM
, and the particles of the attrition
material d
AM
, µ
AM
Volume ratio of grinding media/attrition material V
GM
/V
AM
Volume fraction of water c
v
in the milling chamber
Specific mechanical energy input E
m
referred to as the mass of the attrition material
The modeling of the macroprocess is again based on the assumption of a stochastic
movement of the grinding media and the particles. Blecher (1993) showed that energy
dissipates in two main portions, one portion at the tip of the stirrer and another one at
the inner surface of the mill. Contrary to real comminution in stirred mills, for grinding
attrition, there are not any zones with a high local energy dissipation. Due to a high sol
ids content and a moderate energy input, the energy dissipation of the grinding media
occurs in an equidistributed manner in the mill chamber. The Reynolds number of the
movement in the mill is very high and indicates a turbulent regime. Turbulence in the
stirred mill is also the basis of the investigations of Theuerkauf and Schwedes (1999)
modeling the movement of the filling of stirred media mills. The turbulent movement is
statistically recordable as a random walk of the turbulence elements in all three direc
tions in space. Thus, turbulence is responsible for the impacts between particles and
grinding media. In the following discussion, grinding attrition is characterized by a
homogeneous turbulence with the mean velocity .
To model the macroprocess, the relations of the mean free path length ì with
respect to the mean impact time t in the milling chamber must be found.
STRESS NUMBER
The macromodel is based on determining the stress numbers for one particle and one
grinding body per unit of time.
According to Figure 8, a cylindrical cell with the crosssectional area (r
GM
+ r
AM
)
2
and the length ì = v At is observed, where a particle moves through with the velocity v
during At. The mean impact number in this volume results from the contact number of
the particle and the grinding media, whose central point is located in the cell observed.
Here, it should be noted that the impact partners (particle and grinding media) have dif
ferent sizes.
Am
f
4.6
1 om
f0
– ( )
2
o
 =
v'
2
v =
300 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
The time t between two impacts is
(EQ 12)
and the impact frequency e is
(EQ 13)
The mean free path length of one particle can now be determined in analogy to gas mol
ecules, considering different sizes of particle and grinding media:
(EQ 14)
where
r
AM
= radius of attrited material
r
GM
= radius of grinding media
c
GM
= volume concentration of grinding media
1 2 3 4 5
0.24
0.20
0.16
0.12
0.28
Δ
m
f
In t
Case 1: Δm
f
= 0.0399 In t + 0.1498
Case 2: Δm
f
= 0.0542 In t + 0.0591
Case 3: Δm
f
= 0.0438 In t + 0.1153
Case 1
Case 2
Case 3
Regression
FIGURE 6 Comparison of the nonlinear model with the experimental data
TABLE 2 Values of the parameters according to Equation 9 for the experimental data
Case 1 Case 2 Case 3
o = 1.57 o = 2.02 o = 1.67
 = 14.8  = 8.22  = 12.6
t
c
= 0.024 min t
c
= 0.34 min t
c
= 0.072 min
t
ì
v
 =
e
1
t
 =
ì
1
t r
AM
r
GM
+ ( )
2
c
GM
 =
MODELING ATTRITION IN STIRRED MILLS APPLYING STATISTICAL PHYSICS 301
Solids Concentration
(attrition material +
grinding media + water)
Diameter of the
RingType Stirrer
Filling Volume
Revolution
Number
Volume Element
for Determining the
Stress Number
Size and Density of
Particles
and
Grinding Media
Volume Ratio of
Particles and
Grinding Media
FIGURE 7 Scheme of the macroprocess, including the main process parameters
r = r
GM
+ r
AM
r
AM
r
GM
Particle Does Not Hit Grinding Medium
Particle Hits Grinding Media
< r
< r
> r
= r
According to Atkins 2001.
FIGURE 8 Observed volume element to calculate the mean free path
302 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
The impact frequency of one particle with the grinding media yields
(EQ 15)
Further, the impact frequency O of all particles in the above cylindrical volume is
(EQ 16)
For the volume concentration c
GM
of the grinding media, one may assume the
following:
(EQ 17)
N
GM
= number of grinding media particles
V
GM
= volume of grinding media
V = volume of milling chamber
d
GM
= size of grinding media
V
S
= volume of solids
c
V
= volume fraction of water in the milling chamber
Analogously, for the volume concentration c
AM
of the particles of the attrited material,
the following equation is valid:
(EQ 18)
Assuming v
~
nD, the impact frequency O
A
for the total volume of the apparatus can
be determined as follows:
(EQ 19)
where
V
A
= mill volume (m
3
)
c
v
= solids concentration of the suspension containing the attrition material and
the grinding media (–)
Specific Energy Input
The specific energy input transferred from the attritor shaft to the grinding material with
the mass m
g
is computed according to the following:
(EQ 20)
where
= turbulent velocity of the grinding media (m/sec)
µ
GM
= density of the grinding media (kg/m
3
)
e vt r
AM
r
GM
+ ( )
2
c
GM
=
O vt r
AM
r
GM
+ ( )
2
c
AM
c
GM
=
c
GM
N
GM
V

V
GM
V

1
d
GM
3

V
GM
V
S

V
S
V

1
d
GM
3

V
GM
V
S
c
V
1
d
GM
3
 = = = =
c
AM
V
AM
V
S
c
v
1
d
GM
3
 =
O
A
V
A
nD
d
AM
d
GM
+ ( )
2
d
AM
3
d
GM
3
c
v
2
V
GM
V
AM

1
V
GM
V
AM
 +
\ .
 
2
 ~
E
0
N
GM
µ
GM
d
GM
3
N
GM
N
AM
+
 v
GM
2
( )
N
AM
µ
AM
d
AM
3
N
GM
N
AM
+
 v
AM
2
( ) +
\ .

 
=
v
GM
( )
MODELING ATTRITION IN STIRRED MILLS APPLYING STATISTICAL PHYSICS 303
= turbulent velocity of the attrition material (m/sec)
µ
AM
= density of the attrition material (kg/m
3
)
Equations 19 and 20 apply, however, only to the initial phase of the grinding attri
tion, provided that no attenuation has begun as a result of the fines production. As a fur
ther simplification, it is to be assumed that the mean velocity of all particles and grinding
media is equal: . Thus, Equation 20 becomes
(EQ 21)
Considering the appropriate volumes of the grinding media V
GM
and the attrition
material V
AM
, depending upon the reference values, the following is obtained:
(EQ 22)
where k is a numerical constant.
Implementing now the microprocess from Equation 11 and Equation 7, the follow
ing is obtained with E
0
instead of e
0
:
(EQ 23)
Further, the energy input E
0
will be related to the produced fines m
f
which results in
(EQ 24)
For a better survey and to enable scaleup considerations, the process Equation 24 is
modified, introducing the following dimensionless numbers:
v
AM
( )
v
2
( ) v
GM
2
( ) v
AM
2
( ) = =
E
0
N
GM
µ
GM
d
GM
3
N
GM
N
AM
+

N
AM
µ
AM
d
AM
3
N
GM
N
AM
+
 +
\ .

 
n
2
D
2
m
G
1 –
~
E
0
kn
2
D
2
µ
GM
d
GM
3
1
µ
AM
µ
GM

V
AM
V
GM
 +
\ .

 
1
d
GM
3
d
AM
3

V
AM
V
GM
 +
\ .

 
 m
G
1 –
kn
2
D
2
µ
GM
d
GM
3
1
µ
GM
µ
AM

V
GM
V
AM
 +
\ .

 
1
d
AM
3
d
GM
3

V
GM
V
AM
 +
\ .

 
 m
G
1 –
=
=
Am
f
4.6k
1 om
f0
– ( )
2
o

n
2
D
2
µ
AM
d
AM
3
E
A
d
f
2

1
µ
GM
µ
AM

V
GM
V
AM
 +
\ .

 
1
d
AM
3
d
GM
3

V
GM
V
AM
 +
\ .

 
 =
E
mf
2tn 1 om
0
– ( )
M 0 ( )
µ
AM
V
AM
O
A

d
f
2
E
A
1 om
0
– ( )n
2
D
2
µ
AM
d
AM
3

1
µ
GM
µ
AM

V
GM
V
AM
 +
\ .

 
1
d
AM
3
d
GM
3

V
GM
V
AM
 +
\ .

 
 exp =
304 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
; ; ; ;
With these dimensionless numbers, Equations 23 and 24 can be transformed to
(EQ 25)
and
(EQ 26)
The correction constant k serves to adapt the model equation to the experimental values.
MODEL VALI DATI ON
To validate Equations 25 and 26, attrition experiments in a 1.5L cell were conducted.
The attrition material was hard limestone in the particle size range between 100 and
150 µm. Figures 9–12 present the comparison of experimental data and curves calcu
lated using the model equations. Obviously, the model satisfactorily reflects the influence
of grinding body size and density on fines production. The same can be concluded for
the dependency of specific energy input on the circumferential stirrer speed. The con
stant k was determined as k ~ 550.
Especially remarkable is the dependency of fines production on circumferential stir
rer speed measured in three mills of different sizes—1, 3, and 6 L. According to Figure 12,
making a scaleup of the attrition in stirred mills seems to be justified.
u
V
GM
V
AM
 = A
d
AM
d
GM
 = u
µ
GM
µ
AM
 = H
d
f
2
E
A
1 om
0
– ( )n
2
D
2
µ
AM
d
AM
3
 = +
V
a
D
d
GM
4

1
A
4
 =
Am
f
4.6k
1 om
f0
– ( )
o

1
H

1 uu + ( )
1 A
3
u + ( )
 =
E
ˆ
mf
1 u + ( )
2
k+ A 1 A + ( )
2
c
v
u
 H
1 uu + ( )
1 A
3
u + ( )
 exp =
0
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
0.18
0 1,000 2,000 3,000 4,000 5,000 6,000 7,000 8,000
Density of Grinding Media, kg/m
3
Calculated n = 1,600
n = 1,600 1/min
Calculated n = 1,200 1/min
n = 1,200 1/min
Calculated n = 1,000
n = 1,000 1/min
t = 16 min
c
v
= 0.56
V
GM
/V
AM
= 1
V
A
= 1.51
FIGURE 9 Influence of density of grinding media on fines production
MODELING ATTRITION IN STIRRED MILLS APPLYING STATISTICAL PHYSICS 305
CONCLUSI ONS
The presented macromodel of grinding attrition can substantially reduce experimental
expenditure for the optimization of the attrition conditions. First, it is necessary for the
parameters of the kinetic equation to be determined in very few tests. The influence of
fluctuating process conditions can then be simulated with Equations 9 and 5. Compari
sons with experimental data show that the developed kinetic model can be used for a
prediction of the fines production under certain process conditions. Further investiga
tions are required to improve the model’s applicability to ensure that experimental
expenditures to define process conditions can be minimized.
0
0.02
0.04
0.06
0.08
0.10
0.12
0 200 400 600 800 1,000 1,200 1,400
Diameter of Grinding Media d
GM
,
Calculated n = 1,600
n = 1,600 (1/min)
Calculated n = 1,000
n = 1,000 (1/min)
Calculated n = 800
n = 800 (1/min)
t = 16 min
c
v
= 0.56
ρ
GM
= 2.5 kg/L
V
GM
/V
AM
= 1
V
A
= 1.51
FIGURE 10 Influence of grinding media size on fines production at different revolution numbers of
the agitator
0
1,000
2,000
3,000
4,000
5,000
6,000
7,000
8,000
0 500 1,000 1,500 2,000
Revolution Number n, 1/min
S
p
e
c
i
f
i
c
E
n
e
r
g
y
I
n
p
u
t
E
m
f
,
J
/
k
g
t = 16 min
c
v
= 0.56
ρ
GM
= 2.5 kg/L
V
GM
/V
AM
= 1
V
A
= 1.51
d
GM
= 750 μm
Calculated d
GM
= 750 μm
d
GM
= 350 μm
Calculated d
GM
= 350 μm
d
GM
= 250 μm
Calculated d
GM
= 250 μm
FIGURE 11 Influence of agitator revolution number on specific energy input at different grinding
media sizes
306 ADVANCES IN COMMINUTION LIBERATION AND BREAKAGE
REFERENCES
Atkins, P.W. 2001. Physikalische Chemie. Weinheim: WileyVCH.
Blecher, L. 1993. Strömungsvorgänge in Rührwerkskugelmühlen (Flow pattern in stirred
mills). Ph.D. dissertation. Braunschweig: Technische Universitat.
Joost, B., and J. Schwedes. 1996. Comminution of white fused alumina and wear of
grinding beads in stirred media mills. Part II: Effect of operational parameters,
grinding chamber geometry and beads hardness on the wear of grinding beads.
Ceramic Forum International 73:69–76.
Levich, B.G. 1997. Theoretical Physics. Amsterdam: NorthHolland.
Schricker, B. 1999. Intensivierung der Attrition bei der physikalischchemischen Sanierung
kontaminierter mineralischer Abfallstoffe (Intensifying the attrition with physical
chemical cleanup of contaminated, mineral waste material). Ph.D. dissertation.
Nurnberg: Universität Erlangen.
Schricker, B., J. Dueck, and T. Neesse. 1998. Grindingattrition for treatment of contami
nated mineral residues. In Proceedings of the Third World Congress on Particle Tech
nology, Brighton, Great Britain. Warwickshire: Institution of Chemical Engineers.
Theuerkauf, J., and J. Schwedes. 1999. Theoretical and experimental investigation on
particle and fluid motion in stirred media mills. Powder Technology 105:406–412.
Tiefel, H., B. Schricker, and Th. Neesse. 1999. Hochleistungsattrition zur mechanischen
Reinigung von mineralischen Roh und Reststoffen (High performance attrition for
mechanical cleaning of mineral raw materials and residues). Aufbereitungs Technik
40(4):160–164.
0.02
0.00
0.04
0.06
0.08
0.10
0.12
0.14
0.16
0.18
0 500 1,000 1,500 2,000
Revolution Number n, 1/min
Δ
m
f
<
1
0
μ
m
(
–
)
Calculated Horizontal Mill 31
Horizontal Mill 31
Calculated Vertical Mill 11
Vertical Mill 11
Calculated Vertical Mill 61
Vertical Mill 61
FIGURE 12 Fines production for three different mill sizes at different agitator revolution numbers
307
PART 4
Mill Design
309
Design of Iron Ore Comminution
Circuits to Minimize Overgrinding
T.C. Eisele
*
and S.K. Kawatra
*
ABSTRACT
Overgrinding is a major concern in grinding circuits because it (1) wastes energy, and (2) can
result in valuable minerals being ground to such fine sizes that they are difficult to process.
Overgrinding can be particularly severe in grinding circuits that are processing mixtures of
minerals of different densities while using classifiers for size control. This results in the high
density minerals that have already reached the target size being retained in the circuit and
ground to a finer size than the lowdensity minerals. While this problem could, in principle,
be solved by the use of screens for size control, in practice, screening costs become excessive
for fine grinding due to high maintenance and low screening efficiency. A novel approach is
to redesign the circuit so that the particles that are very close to the target size can be sepa
rated out and handled differently than particles that are definitely coarser than the target
size. Simulations of this type of approach are presented. For the circuit simulated in this
paper, the circulating load was reduced from 250% to only 42.5% without sacrificing qual
ity, resulting in a 50% increase in grinding circuit throughput.
I NTRODUCTI ON
In grinding circuits, energy is wasted when particles are ground to a size finer than is
necessary. Such overgrinding can be caused when particles are retained in the circuit for
too long and continue to be ground even after they have reached the target product size
of the circuit. There are two factors that contribute heavily to overgrinding: (1) ineffi
cient classification returning a large fraction of the finest particles to the circuit; and
(2) excessive mill retention times causing particles to be broken multiple times before
they are discharged.
It is normally expected that there is a tradeoff between overgrinding and the pres
ence of locked particles. If the circuit product size is made coarser to reduce the over
grinding, this typically leads to an increase in the top size of the product as well, with the
top size particles being poorly liberated. This reduces the grade that can be produced in
the processing that follows comminution. In order to reduce this tradeoff, it is necessary
to determine how to change the grinding circuit to produce a narrower size distribution,
which will reduce overgrinding while simultaneously controlling the quantity of coarse
locked particles leaving the circuit.
A significant cause of overgrinding is when an ore contains minerals of different
densities, and a hydrocyclone or other classifier is used to control product size from the
* Department of Chemical Engineering, Michigan Technological University, Houghton, Michigan
310 ADVANCES IN COMMINUTION MILL DESIGN
circuit grinding the ore. The most important case of this is iron ore grinding, where the
iron oxides have densities of approximately 5 g/cm
3
whereas the gangue minerals are
less than 3 g/cm
3
, and the iron oxides make up a large fraction of the mass of the raw
ore. As classifiers separate based on both size and density, the higherdensity iron oxides
continue to report to the classifier coarse product even when they are already finer than
the lowerdensity gangue particles that report to the classifier fine product. This results
in the iron oxides being ground to a much finer size than necessary. To illustrate the
extent of this problem, the results of plant sampling studies are presented in the next
section.
PLANT SAMPLI NG STUDI ES
The plant examined was a magnetite concentrator located in the Lake Superior iron ore
district. The flowsheet was a pebble mill circuit, as shown in Figure 1. This type of plant
was selected because iron ore is one of the most highvolume metallic ores produced and
is therefore of considerable practical importance. Sampling and sample analysis were
conducted by plant personnel.
Size analyses were carried out by three methods:
1. Wet sieving at 25 Pm using a wovenwire test sieve, followed by dry sieving of
the +25Pm particles using wovenwire test sieves in a RoTap sieve shaker
2. Microsieving of dry powders using electroformed nickelfoil sieves in a Sonic
Sifter apparatus, which allowed sieving down to 10Pm particle size
3. Microtrac laser diffraction particle size analysis to measure particle sizes down to
1 Pm
For the sieved samples, each individual size fraction was assayed using a dichromate
titration method (ASTM 2001) by plant personnel to determine the iron assay of each
sieved size fraction for each stream mentioned above. The size distribution and assay
data were then mass balanced, and the magnetite concentration in each size fraction was
calculated from the iron assays.
The results for the pebble mill circuit were as shown in Table 1. It should be noted
that the –25Pm fraction of the cyclone underflow/mill feed is 30.8% of the material
being returned to the pebble mill for grinding, even though it is already finer than the
target grind size of 25 Pm. Further, this fraction of the pebble mill feed is unusually
enriched in magnetite and consists of 92.9% magnetic material. From this information, it
is clear that a great deal of the pebble mill feed consists of fine magnetite that is being
retained in the circuit by the hydrocyclones and, as a result, is being overground. This is
a very large contribution to the circulating load, which is 250% of the circuit new feed.
The Bond equation can be used to estimate the energy wasted in overgrinding the
fine magnetite. The target size was 80% passing 25 Pm, but size analysis of the magne
tite product from the circuit showed that it was 80% passing 20 Pm. Given that the aver
age work index of the ore was 11.6 kWhr/t, the energy wasted can be calculated as
(EQ 1)
As the overall circuit product was 57.4% magnetite, thus, the energy wasted per ton
of total circuit feed was (0.57)(2.74) = 1.57 kWhr/t. The plant grinds approximately 15 Mt
of ore per year, and so the overgrinding losses from this source are 23 million kWhr/
year, or 2.4 u 10
11
Btu/year for a single plant.
W 11.6
10
20

10
25
 –
© ¹
¨ ¸
§ ·
2.74 kWhr/t = =
DESIGN OF IRON ORE COMMINUTION CIRCUITS TO MINIMIZE OVERGRINDING 311
Chips
–½ in. + 1 mm
Back to Primary Mill
Primary
Mill (1)
32 ft × 16.5 ft
8,500 hp
Excess Pebble
Crusher
Nordber 200 hp
ShortHead
Cone 350 hp
Vibrating
Screen Top
Deck: ½ in.
Bottom Deck:
1 mm
Crushed
Pebbles
Roll Press
KHD RP 7.0
1,400 mm (d) × 800 mm (l)
Feed: 67% – ½ in.
Production: 84% – ¼ in.
Cobbers Magnetic Separator (3)
36 in. (d) × 10 ft (l)
Cobbers Magnetic
Separator (3)
36 in. (d) × 10 ft (l)
Cobber Tails
to Tailings Dam
Cobber Tails
to Tailings Dam
Cobbers Feed
and Sump Cyclones Feed
and Sump
Cyclones Feed
and Sump
Pebbles
(–2½ in. + ½ in.)
Pebbles
(–2½ in. + ½ in.)
CNF
CNF
COF
COF
Krebs 15 in. Cyclone
16 Total, 2 Standby
Vortex: 5¼ in.
Apex: 3 in.
Pressure: 20–30 psi
Krebs 15 in. Cyclone
16 Total, 2 Standby
Vortex: 5¼ in.
Apex: 3 in.
Pressure: 20–30 psi
CUF
CUF
Chips
Chips
Pebble Mill (1)
15.5 ft × 32.5 ft
2,650 hp
Pebble Mill (1)
15.5 ft × 32.5 ft
2,650 hp
Pebble Mill Discharge
Pebble Mill Discharge
NOTES: CNF = circuit new feed; COF = cyclone overflow; CUF = cyclone underflow.
FIGURE 1 Configuration of the base grinding circuit sampled for this study. The paired pebble
mills were the primary area of interest.
* Overall percentages are the percentage of each size fraction in the stream, with the sum of + 25μm and –25μm values
equaling 100%.
† Magnetic and nonmagnetic percentages are the assays for each size fraction.
TABLE 1 Results of plant sampling
Stream Size Fraction Overall
*
Magnetic
†
Nonmagnetic
†
Circuit new feed—122.5 metric
tons per hour (tph)
+25 μm 89.3 tph
72.9%
46.1 tph
51.7%
43.2 tph
48.3%
–25 μm 33.2 tph
27.1%
29.0 tph
87.4%
4.2 tph
12.6%
Cyclone underflow/mill feed—
304.1 tph
+25 μm 210.4 tph
69.2%
117.6 tph
55.9%
92.8 tph
44.1%
–25 μm 93.7 tph
30.8%
87.0 tph
92.9%
6.7 tph
7.1%
Pebble mill discharge—
311.8 tph
+25 μm 141.2 tph
45.3%
77.8 tph
55.1%
63.4 tph
44.9%
–25 μm 170.6 tph
54.7%
108.8 tph
63.8%
61.8 tph
36.2%
Cyclone overflow/circuit
product—130.1 tph
+25 μm 16.4 tph
12.6%
7.1 tph
43.5%
9.2 tph
56.5%
–25 μm 113.8 tph
87.4%
67.6 tph
59.4%
46.2 tph
40.6%
312 ADVANCES IN COMMINUTION MILL DESIGN
These results show that there is a significant benefit to be gained if the plant can be
redesigned to prevent overgrinding. This is an ideal situation for simulation studies, as it
is necessary to consider radical rearrangements of the circuit that would not be feasible
to conduct on a speculative basis in actual plant studies.
CI RCUI T SI MULATI ON
Base Circuit Simulations
In order to simulate the pebble mill portion of the grinding circuit, a pebble mill model
and hydrocyclone model were needed. Both of these models were implemented using
USIMPAC 3.0 simulation software (BRGM 2004).
The pebble mill model used was based on the following mill characteristics.
The forms of the selection and breakage functions used in the model were based on work
by Austin and Herbst (BRGM 2004; Austin, Manacho, and Pearcy 1987; Kinneberg and
Herbst 1984).
Breakage function:
(EQ 2)
where B
ij
is the fraction of the mass of broken particles from size fraction i that reports to
size fraction j; and x
i
is the top size limit of size fraction i.
The breakage function parameters that were found to give the best fit to plant data
were
Selection function:
(EQ 3)
where S
i
is the fraction of particles in size fraction i that are broken; d
i
is the geometric
mean particle diameter of size fraction i; and d
i(ref
)
is the reference particle size class.
Number of mills in parallel: 1
Mill diameter inside shell (m): 4
Length–diameter ratio: 2.1
Fraction of critical speed: 0.8785
Mill discharge: Overflow
Filling of the mill (%): 43
Reference size for the wear function (mm): 15.875
Wear coefficient (0 = surface; 1 = volume): 0
Wear rate of pebbles (1/hr): 3.45
Reference size class for the selection function: 10
) 0.096
E 3.93
J 0.608
B
ij
) x
i–1
x
j
e
J
1 ) – x
i–1
x
j
e
E
+ =
S
i
S
1
E
a
1
d
i
d
i ref
e ln a
2
d
i
d
i ref
e ln
2
+ > @ exp =
DESIGN OF IRON ORE COMMINUTION CIRCUITS TO MINIMIZE OVERGRINDING 313
The selection function parameters that were found to give the best fit to plant data were
The simulations did not incorporate a liberation model. The proportions of magne
tite and quartz by size were determined by chemical analysis of sieve fractions collected
from the plant, and it was assumed that as the target grind size was being held constant,
the degree of liberation at each size would also be essentially constant and would there
fore not have a major effect on the predictions of the model.
By using these parameters to simulate the pebble mill, the results shown in Figure 2
were obtained, which show an excellent match between the predicted and measured mill
discharge. Several different sets of data from the same plant at different flow rates were
used to validate the model parameters, with similar results.
The hydrocyclone model used was that developed by Plitt (1976), which performed
adequately in predicting the cyclone performance in this circuit.
(EQ 4)
where
C(d) = classification function for particles of size d after correcting for particles that
bypass classification
d = particle size, in micrometers
d50c = particle size that has equal probability of reporting to the overflow or the
underflow, after correcting for the particles that bypass classification
m = measure of the sharpness of separation
The corrected d50 size was calculated using the following formula:
(EQ 5)
where
d50c = corrected d50 (Pm)
I = volumetric fraction of solids in feed
D
c
= cyclone diameter (cm)
h = free vortex height (cm)
D
i
= inlet diameter (cm)
Q = volumetric flow rate of feed (L/min)
D
o
= overflow diameter (cm)
U
s
= solid density (g/cm
3
)
D
u
= underflow diameter (cm)
U = liquid density (g/cm
3
)
Simulations of the overall pebble mill/hydrocyclone circuit reproduced the tendency
of the circuit to retain fine magnetite in the pebble mill feed, as can be seen in Figure 3.
Here, it can be seen that the nonmagnetic material was much coarser than the magnetics,
0.75
a
1
–1.5
a
2
–0.5
S
1
E
C d 1 e
0.693
d
d50c

© ¹
§ ·
m
–
– =
d50c
50.5D
c
0.46
D
i
0.6
D
o
1.21
0.063I exp
D
u
0.71
h
0.38
Q
0.45
U
s
U –
0.5
 =
314 ADVANCES IN COMMINUTION MILL DESIGN
and the overall size distribution of the mill feed was dominated by the presence of fine
magnetite that was being returned to the mill and overground.
It is particularly noteworthy that the magnetics are concentrated in the size range of
25 to 50 Pm. If this size fraction could be segregated into a separate process stream to be
managed separately, it would allow the particles coarser than 50 Pm to be ground with
out excessive production of overground material.
Modified Circuit Simulations
The results for the unmodified circuit suggested a novel approach to the problem of over
grinding the magnetite fines, shown in Figure 4. If the particles that are near the target
size are separated from the rest of the pebble mill feed by using twostage cycloning,
then they will be prevented from being trapped in the circulating load, allowing the first
pebble mill to concentrate on grinding the particles that are definitely coarse enough to
require vigorous grinding.
Twostage classification has been used in the past to improve the performance of
classifiers (Dahlstrom and Kam 1987; Hukki, Karhunen, and Lindsberg 1977). The most
effective applications of twostage classification have been in cases where the bypass
fraction of a singlestage classifier is large, allowing a large quantity of the fine particles
to bypass classification and be returned to the grinding circuit. With a twostage cyclone
circuit of proper configuration and water additions, the fineparticle bypass can be
decreased, resulting in lower circulating loads and the possibility of higher feed capacity
(Peterson and Herbst 1984). However, this approach is of little effect if the hydrocy
clones are already being operated with a very low bypass fraction, as is often the case in
iron ore concentrators (T. Weldum, personal communication, 2003).
When fine, highdensity particles are, in fact, undergoing classification and not
bypassing classification, they are then classified into the coarse product due to their density.
The settling rate of fine, dense particles is the same as that of coarser, less dense parti
cles, so that the fine magnetite tends to concentrate in the hydrocyclone underflow as a
direct result of correct operation of the hydrocyclone. In this situation, the reduction of
100
90
80
70
60
50
40
30
20
10
0
1 10 100 1,000 10,000
C
u
m
u
l
a
t
i
v
e
%
P
a
s
s
i
n
g
Feed
Measured Mill Discharge
Predicted Mill Discharge
FIGURE 2 Simulation results for the pebble mill model as compared with actual grinding mill
performance
DESIGN OF IRON ORE COMMINUTION CIRCUITS TO MINIMIZE OVERGRINDING 315
100
90
80
70
60
50
40
30
20
10
0
1 10 100 1,000 10,000
C
u
m
u
l
a
t
i
v
e
%
P
a
s
s
i
n
g
Overall
Magnetics
Nonmagnetics
FIGURE 3 Size distributions of the magnetic and nonmagnetic material in the hydrocyclone
underflow/mill feed, as calculated by simulations. These results were consistent with the
observed performance of the grinding circuit.
–½ in. + 1 mm
Back to Primary Mill
Primary
Mill (1)
32 ft × 16.5 ft
8,500 hp
Excess Pebble
Crusher
Nordber 200 hp
ShortHead
Cone 350 hp
Vibrating
Screen Top
Deck: ½ in.
Bottom Deck:
1 mm
Crushed
Pebbles
Roll Press
KHD RP 7.0
1,400 mm (d) × 800 mm (l)
Feed: 67% – ½ in.
Production: 84% – ¼ in.
Cobbers Magnetic
Separator (3)
36 in. (d) × 10 ft (l)
Cobbers Magnetic
Separator (3)
36 in. (d) × 10 ft (l)
Cobber Tails
to Tailings Dam
Cobber Tails
to Tailings Dam
Cobbers Feed
and Sump
Cyclones Feed
and Sump
Pebbles
(–2½ in. + ½ in.)
Pebbles
(–2½ in. + ½ in.)
CNF
COF
Chips
Chips
Chips
Mill Product
CUF
CUF
Dewatering Cobber
ClosedCircuit
Pebble Mill (1)
15.5 ft × 32.5 ft
OpenCircuit
Pebble Mill (1)
15.5 ft × 32.5 ft
Pebble Mill Discharge
FIGURE 4 Modified circuit that was produced from the circuit shown in Figure 1. This is largely a
rearrangement of existing equipment, with the only major added unit being the dewatering cobber
for the opencircuit pebble mill.
316 ADVANCES IN COMMINUTION MILL DESIGN
the bypass fraction will not cause these particles to be removed from the coarse product,
and an attempt to use twostage classification will simply result in a stream of trapped,
narrowly sized highdensity particles continuously circulating between the two hydrocy
clone classifiers. This stream of nearsize particles then continuously increases in volume
until the hydrocyclones become overloaded (T. Weldum, personal communication, 2003).
If the nearsize particles can be separated as a sufficiently narrow size distribution,
the possibility arises of grinding this stream from twostage cycloning in an opencircuit
mill. This would eliminate the possibility of nearsize particles being trapped in the circu
lating load and overground. Normally, opencircuit grinding will produce a broad size
distribution due to the lack of particle size control (Kinneberg and Herbst 1984). How
ever, it is also normal for the feed to an opencircuit mill to itself have a broad size distri
bution, which would broaden still further during opencircuit grinding.
In previous work in this plant (T. Weldum, personal communication, 2003), it had
been determined that the use of twostage cycloning could produce one stream with a
very narrow size distribution. If such a closely sized feed were ground in open circuit, the
size distribution would be expected to remain moderately narrow. In addition, it is normal
for the coarsest particles to have a higher probability of breaking than the finer particles
in a tumblingmedia mill (Teke et al. 2002), and so a pass through an opencircuit mill
would be expected to preferentially break the coarsest particles.
Based on these considerations, a new circuit was developed and simulated, as
shown in Figure 4. The primary considerations in this circuit were
Use twostage cycloning to produce three product streams: (1) coarse particles for
grinding in a closedcircuit mill; (2) fine particles that are definitely fine enough
to be removed from the grinding circuit as product; and (3) nearsize particles
that are only slightly coarser than the target product size, and that have an
extremely narrow size distribution.
Use opencircuit grinding to reduce the top size of the nearsize particle stream.
Material only makes a single pass through this mill, so there is no opportunity to
continuously recycle and regrind highdensity particles. The narrow size distribu
tion of the nearsize particle stream will allow the opencircuit mill to produce a
moderately closely sized product.
As far as possible, use only equipment that is already present in the existing circuit.
By reconfiguring the mills and existing cyclones, it would be possible to imple
ment the circuit shown in Figure 4 directly from the circuit shown in Figure 1,
with the only major equipment addition being the dewatering cobber added to
remove excess water from the opencircuit mill feed.
The product size of an opencircuit mill is quite sensitive to the feed rate (Teke et al.
2002); therefore, it was first necessary to ensure that an appropriate amount of material
was sent to this mill by the two hydrocyclones. It was determined that, in order to pro
duce the target feed size distribution, the opencircuit mill needed to receive 81 tph of
feed. In order to produce this quantity of material, the hydrocyclones were operated with
the efficiency curves shown in Figure 5.
The two mill feed products produced by the twostage cyclone processing are shown
in Figure 6. The feed to the closedcircuit mill is much coarser than it had been for the
unmodified circuit (compare with the size distributions shown in Figure 3 for the
unmodified circuit). In addition, the difference between the magnetite and quartz size
distributions is greatly reduced, so that there is much less tendency to overgrind the
magnetite while reducing the size of the nonmagnetic quartz. For the feed to the open
circuit mill, the hydrocyclones are successfully producing a very narrowly sized stream
DESIGN OF IRON ORE COMMINUTION CIRCUITS TO MINIMIZE OVERGRINDING 317
where the size distributions for the magnetite and quartz are very similar. Again, this
similarity of the size distribution will help to prevent overgrinding of the magnetite.
Two products are produced by the twostage circuit: a primary cyclone overflow
stream and an open circuit mill discharge stream. The size distributions predicted by the
simulation for these streams are shown in Figure 7, along with the sum of the two product
streams, and the size distribution of the final product of the original singlestage circuit
100
90
80
70
60
50
40
30
20
10
0
1 10 100 1,000
%
o
f
S
i
z
e
t
o
U
n
d
e
r
f
l
o
w
TwoStage Primary Cyclone
TwoStage Secondary Cyclone
SingleStage Cyclone
FIGURE 5 Efficiency curves for the twostage cyclones used in the combined closedcircuit/
opencircuit flowsheet, including a comparison with the efficiency curve for the single stage of
cycloning used in the original, unmodified circuit. The close spacing between the efficiency
curves ensured a very sharp size distribution in the feed to the opencircuit mill, while still
providing the 81 tph of solids needed to produce the desired size distribution.
100
90
80
70
60
50
40
30
20
10
0
1 10 100 1,000 10,000
C
u
m
u
l
a
t
i
v
e
%
P
a
s
s
i
n
g
CC Feed Overall
CC Feed Magnetics
CC Feed Nonmagnetics
OC Feed Overall
OC Feed Magnetics
OC Feed Nonmagnetics
FIGURE 6 Size distributions of the feed streams produced by the twostage cyclone arrangement
for the closedcircuit (CC) mill and the opencircuit (OC) mill shown in Figure 4
318 ADVANCES IN COMMINUTION MILL DESIGN
for comparison. The opencircuit mill discharge stream can be seen to be maintaining a
narrow size distribution, due to the fact that the mill feed consisted of closely sized parti
cles. It should be noted that the size distribution of the opencircuit mill feed was consid
erably finer than the plant feeds that were originally available for use in determining and
validating the model parameters. The model is therefore not fully validated for the open
circuit mill conditions, and it is possible that the predicted size distribution from the sim
ulation will not be identical to the actual size distribution that would be produced in the
plant. However, as there is no circulating load from the opencircuit mill, any errors from
this portion of the circuit simulation will not propagate back to the rest of the circuit.
When the two products from the twostage mill are added together, the size distribu
tion predicted is nearly identical to that produced by the original singlestage grinding
circuit. This indicates that the twostage grinding circuit is fully capable of reaching the
target grind size. The main benefit is seen when the relative flow rates of material
through the circuit are compared, as shown in Table 2. The use of the twostage circuit
allows the circulating load to be drastically reduced, from a 253% circulating load to
only 42.5%. This makes available a great deal of extra hydrocyclone and grinding mill
capacity, allowing the capacity of the circuit to be increased by 50%. This reduction in
circulating load is a direct result of the large reduction in the recirculation and over
grinding of the fine, highdensity magnetite particles. It should be noted that the “circu
lating load” and “primary cyclone feed” streams for the twostage circuit contained,
respectively, only 66.0% and 64.5% magnetite, compared to 89.2% and 82.0% magne
tite for the corresponding streams in the singlestage closed grinding circuit. This shows
that the twostage open/closed grinding circuit was removing the nearsize magnetite
from the circulating load and diverting it to the opencircuit mill feed for removal from
the circuit.
120
100
80
60
40
20
0
1 10 100 1,000 10,000
C
u
m
u
l
a
t
i
v
e
%
P
a
s
s
i
n
g
Primary Cyclone Overflow
OpenCircuit Mill Discharge
Total TwoStage Product
SingleStage Product
FIGURE 7 Size distributions for the two products produced by the twostage open/closedcircuit
grinding simulation, compared with the product from the original singlestage closedcircuit
grinding process
DESIGN OF IRON ORE COMMINUTION CIRCUITS TO MINIMIZE OVERGRINDING 319
CONCLUSI ONS
Overgrinding of valuable minerals is often caused by the tendency of hydrocyclones to
retain dense minerals in a closed grinding circuit until they are ground to an excessively
fine size. This is a particular problem for iron ore concentrators, where a large fraction of
the mass of the ore consists of the higherdensity iron oxide minerals. The classical solu
tion to this problem is to use screens for product size control rather than classifiers, but
this is not practical when the target size is very fine due to limited screen capacity and
high maintenance costs.
Simulations of a magnetite ore grinding circuit indicated that a reconfiguration of
the circuit could greatly reduce the overgrinding problem. The use of twostage hydrocy
cloning can concentrate the nearsize highdensity magnetite particles into a closely
sized single stream that requires only a very small amount of grinding to reach the target
size. Opencircuit grinding of this stream is predicted to preferentially grind the coarsest
particles, leaving a product that has the necessary size distribution to be a finished
project. This introduction of opencircuit grinding greatly reduced the circulating load of
the grinding circuit (from 253% to only 42.5%) because the fine magnetite was no
longer being continuously returned to the mill for repeated regrinding. The simulations
indicated that this change would increase the circuit capacity by as much as 50% while
still making the target grind specification.
ACKNOWLEDGMENTS
This project was partially supported by the U.S. Department of Energy under Grant
No. DEFC2601NT41062. The support of the ClevelandCliffs Iron Co. is also gratefully
acknowledged. The authors would also like to thank Ted Weldum, Todd Davis, Gary
Rajala, and Ron Mariani for their considerable advice and assistance in carrying out this
work, and B.K. Mishra of IIT for his useful suggestions. Parameters for the pebble mill
simulation model were determined by H.J. Walqui and J.G. Jelsma.
REFERENCES
ASTM (American Society for Testing and Materials). 2001. Standard test methods for
determination of iron in iron ores and related materials by dichromate titration. In
Annual Book of ASTM Standards. Standard Designation E24601. West Consho
hocken, PA: ASTM International.
TABLE 2 Solids flow rates and percent magnetite of process streams in the simulated twostage
open/closed grinding circuit, as compared to the two singlestage closed grinding circuits that it
would replace
Process Stream
Pair of SingleStage Closed
Grinding Circuits
TwoStage Open/Closed
Grinding Circuit
Circuit feed 240 tph solids
63.9% magnetite
360 tph solids
63.9% magnetite
Circulating load 606 tph solids (253%)
89.2% magnetite
153 tph solids (42.5%)
66.0% magnetite
Primary cyclone feed 846 tph solids
82.0% magnetite
513 tph solids
64.5% magnetite
Opencircuit mill feed — 81 tph solids
86.5% magnetite
Primary cyclone overflow 240 tph solids
63.8% magnetite
273 tph solids
58.4% magnetite
320 ADVANCES IN COMMINUTION MILL DESIGN
Austin, L.G., J.M. Menacho, and F. Pearcy. 1987. A general model for semiautogenous
and autogenous milling. Pages 107–126 in Proceedings of APCOM 1987. Volume 2.
Johannesburg: South African Institute of Mining and Metallurgy.
BRGM. 2004. USIM PAC 3.0 Unit Operation Model Guide. Version 3.0.7.0. Caspeo, 3 Avenue
Claude Guillemin—BP 6009, 45060 Orleans Cedex 2, France.
Dahlstrom, D.A., and W.P. Kam. 1987. Potential energy savings in comminution by two
stage classification. International Journal of Mineral Processing 22(1–4):239–250.
Hukki, R.T., J. Karhunen, and R. Lindsberg. 1977. Research on twostage classification.
ZementKalkGips 30(7):314–22.
Kinneberg, D.J., and J.A. Herbst. 1984. A comparison of linear and nonlinear models for
opencircuit ball mill grinding. International Journal of Mineral Processing 13:143–165.
Peterson, R.D., and J.A. Herbst. 1984. Effects of twostage hydrocyclone classification on
mineral processing plant performance. Canadian Metallurgical Quarterly 23(4):
383–391.
Plitt, A.J. 1976. A mathematical model of the hydrocyclone classifier. CIM Bulletin 69
776:115–123.
Teke, E., M. Yekler, U. Ulusoy, and M. Canbazoglu. 2002. Kinetics of dry grinding of
industrial minerals: Calcite and barite. International Journal of Mineral Processing
67:29–42.
321
Evaluation of LargerDiameter
Hydrocyclone Performance in a
Desliming Application
J.J. Campbell,
*
R. Zhu,
*
J.M. Young,
*
and P.T. Nielsen
*
ABSTRACT
The drive for continuous improvement and lower operating and capital costs for mineral
processing operations is of increasing importance to industry. Therefore, an assessment of
the feasibility of using largerdiameter hydrocyclones in fine ore desliming applications was
conducted.
The work focused on developing and implementing a systematic methodology for com
paring the performance of a standarddiameter (100 mm) and a largerdiameter (250 mm)
hydrocyclone in a desliming application. The results show that in terms of the performance
criteria, similar metallurgical performance was obtained from the largerdiameter hydrocy
clone compared with a standarddiameter unit but at significantly higher throughput. The
results indicate that the largerdiameter units would be suitable in this application and
should lead to simpler and more costeffective desliming circuit performance.
I NTRODUCTI ON
Previous investigations of hydrocyclones in classification applications looked at the influ
ence of vortex finders and spigot sizes (Chu, Chen, and Lee 2000, 2001), cylindrical
height (Kraipech et al. 2002), viscosity and pulp density (Kawatra, Bakshi, and Rusesky
1996), and hydrocyclone diameter coupled with other parameter changes (Lynch and
Rao 1975). In this work, a wideranging test program was conducted to establish the
optimum conditions for primary and secondary hydrocyclone desliming of a fine iron
ore. This ore contained high proportions of gangue components in the finer fractions
(<10 Pm). Desliming processes separate the fine fractions to the overflow and provide
products with higher iron and lower gangue content at minimal loss of iron units.
A range of hydrocyclone types were assessed at the standard diameter (100 mm),
and tests were also conducted to compare the performance of a largerdiameter (250 mm)
unit against the optimum standarddiameter performance. The standarddiameter unit
is in current industrial use for the desliming processes. There was interest in whether the
diameter of the hydrocyclone could be enlarged to increase throughput without signfi
cant detriment to performance. This paper presents the performance comparisons
between standard and largerdiameter units for both primary and secondary hydrocy
clone desliming. The purpose of the program was to obtain actual operating data, as
* CSIRO Minerals, Kenmore, Queensland, Australia
322 ADVANCES IN COMMINUTION MILL DESIGN
opposed to results from simulations, to allow an objective assessment of both opera
tional and metallurgical performances. The ultimate outcome of the work was to assess
whether the largerdiameter unit was appropriate in this application.
EXPERI MENTAL APPROACH
Feed Preparation
A bulk sample of –0.5mm iron ore was used for the testwork. Representative splits
(from a rotary splitter) of the feed were used to conduct the various test runs.
Hydrocyclone Rig and Sampling System
The key to accurate assessment of hydrocyclone performance is highquality sampling.
To ensure this, CSIRO designed a purposebuilt, fully transportable, computercontrolled
hydrocyclone rig (Figure 1) with online monitoring and control of process parameters.
A notable feature of the rig is a special splitter box that collects simultaneous cuts of the
hydrocyclone underflow (U/F) and overflow (O/F) streams (Figure 2).
Test Conditions
The objective of the hydrocyclone testwork was to determine appropriate operating con
ditions that achieved target performance, which were defined as
Corrected cut point of d10 Pm
Highquality separation, as evidenced by a low partition imperfection (<0.5)
Hydrocyclone U/F density of >55% solids by weight
High mass recovery to underflow (>85%)
High throughput
The experimental approach was to test each variable in turn while the others were
held constant. The variables tested were
Hydrocyclone type and design
Vortex finder and spigot sizes
Hydrocylone pressure
Feed density
Hydrocyclone Type and Design. A key aim of the testwork was to quantify the
comparative performance of different types of hydrocyclones (Types 1 and 2). Type 1 is
currently used at a number of Australian iron ore operations and was nominally a short
cone design of the standard 100mm diameter (Figure 3). However, it did have the
option of adding an extra cylindrical section to increase body length, which was in place
for all test results reported in this paper. The Type 2 unit was a standarddiameter long
cone design. For comparison to the larger scale, the two different types were available
with a 250mm diameter (Figure 4), and a similar program of tests were conducted on
these units.
Vortex Finder and Spigot Sizes. The project assessed the performance of various
vortex finder and spigot combinations to quantify the effect of the changes on cutpoint,
imperfection, and capacity. However, in this paper, the only results examined are from
tests conducted when the vortexfindertospigotarea ratio was kept constant at approx
imately 2.5, because this ratio corresponded most closely to the performance criteria
(particularly that of the d
50c
). This ratio was determined from the Plitt (1976) formula:
(EQ 1) d
50c
14.8D
c
0.46
D
i
0.6
D
o
1.21
0.063V exp D
u
0.71
h
0.38
Q
0.45
S L –
0.5
e =
LARGERDIAMETER HYDROCYCLONE PERFORMANCE IN A DESLIMING APPLICATION 323
(EQ 2)
where
d
50c
= corrected d
50
(Pm)
D
c
, D
i
, D
o
, D
u
= internal diameters of the hydrocyclone, inlet, vortex finder, and spigot,
respectively (cm)
V = volumetric percentage solids in the feed
h = distance from the bottom of the vortex finder to the top of the spigot (cm)
Q = flow rate of the feed slurry (m
3
/hr)
S = density of solids (t/m
3
)
L = density of liquid (t/m
3
)
P = pressure (kPa)
Hydrocyclone Pressure. Previous testing and commissioning testwork at CSIRO
on the hydrocyclone rig on similar samples indicated hydrocyclone pressures around
180 kPa would be required to achieve target performance in this application. In this pro
gram, three operating pressures were used from 120 to 240 kPa. The wide pressure
Sample
Cutters
Control
Room
Feed Tank
Hydrocyclone
FIGURE 1 CSIRO hydrocyclone test rig
Q 0.021P
0.56
D
c
0.21
D
i
0.53
h
0.16
D
u
2
D
o
2
+
0.49
0.0031V exp e =
324 ADVANCES IN COMMINUTION MILL DESIGN
range was chosen to quantify performance over a range that simulated a plant being
required to operate at significantly variable tonnages.
Feed Density. Following a range of preliminary tests, it became evident that at
feed pulp densities of less than 20% solids, the U/F density could only reach approxi
mately 40% solids. Hence, the feed pulp densities of 25%, 30%, and 35% solids were
tested so that an acceptable U/F density could be achieved and an optimum determined
by interpolation. Feed densities higher than 35% solids would achieve the target U/F
density; however, the Plitt calculation indicated that performance in terms of cut size
and imperfection would fail to achieve target levels. The feed density was measured online
using an Amdel nucleonic density gauge with manual crosschecks conducted using a
Marcy scale. Measurements were also made on all samples by calculating the feed den
sity from wet and dry sample weights.
Analysis Procedures
Samples collected during the test program were sized and prepared for further analysis
according to the following processing procedures.
Feed and U/F samples:
Wet screening at 38 Pm
Dry screening of the +38Pm fractions into a roottwo size series using a RoTap
sieve shaker
Laser sizing of the –38Pm fractions
Laser sizing was used for each O/F sample.
In order to speed up the sizing analysis process, laser sizings were used for the –38Pm
fractions of all test products. Cyclosizing of –38Pm fractions was only used when sam
ples were required for assay and image analysis.
The laser sizer made the assumption that all particles were spherical, and therefore
calculated the particlesize distribution on that basis. When the material was examined
under a light microscope, the majority of the particles had elongated shapes. Therefore,
Hydrocyclone
Discharge
Sample
Containers
Sample
Cutters
Pneumatic
Rams
FIGURE 2 Sampling rams and cutters
LARGERDIAMETER HYDROCYCLONE PERFORMANCE IN A DESLIMING APPLICATION 325
the spherical size distribution obtained from the laser sizer was “converted” to a cylindri
cal shape with an aspect ratio of 2:1. This correction also ensured that when the sizing
results from screening and laser sizer were combined for the feed and U/F samples, no
discontinuity in the curve was present. This correction was used for all lasersized
results. The sizing correction of a typical result is shown in Figure 5.
RESULTS
Quality of Separation
The basis for comparing the separation efficiency in this work is the separation imperfec
tion (I) given by
I = (d
75c
– d
25c
) / 2 u d
50c
(EQ 3)
where the d
75c
, d
25c
, and d
50c
are the corrected d
75
, d
25
, and d
50
(in Pm), respectively.
145
320
200
405
Ø40
Ø98
Ø98
Ø47
Ø25
Ø62
FIGURE 3 Schematic of the Type 1 100mm
hydrocyclone (not to scale; dimensions in mm)
FIGURE 4 Schematic of the Type 2 250mm
hydrocyclone (not to scale; dimensions in mm)
260
255
250
360
360
360
Ø61
Ø240
Ø256
Ø143
Ø114
Ø75
Ø38
Ø100
326 ADVANCES IN COMMINUTION MILL DESIGN
Table 1 shows that the data from the initial tests on each hydrocyclone revealed that
the Type 2 hydrocyclone would satisfy the separation efficiency criteria, which were a
partition imperfection of less than 0.5 and a cutpoint d
50c
of approximately 10 Pm. The
largerdiameter Type 1 hydrocyclone performed well in terms of separation efficiency
but at some cost to capacity, as will be shown later. However, the smallerdiameter
Type 1 unit struggled to achieve the criteria in terms of cutpoint. Lower feed density
lowered the cutpoint to the desirable range; however, this resulted in borderline U/F
density (55.6% solids) and unacceptably high imperfections (0.56).
Hydrocyclone U/F Density
The intentional similarity of vortex finder–spigot area ratios across the various hydrocy
clone types allowed feed density to be the primary influence on hydrocyclone U/F density.
Figure 6 shows a typical result of the effect of feed density on U/F density. Feed densities
>30% solids were required to produce the target U/F density of >55% solids. Pressure
variation had little influence on U/F density.
Table 2 shows the typical U/F densities produced by different hydrocyclones at the
same feed density. Note that both of the Type 1 hydrocyclones produced higher U/F den
sities than the Type 2s at equivalent conditions.
Mass Recovery to Underflow
Table 2 also shows that mass recoveries to underflow were somewhat different for the dif
ferent units, indicating the influence of hydrocyclone geometry (long cylinder versus long
cone). Across the pressure range tested, only the standarddiameter Type 1 hydrocyclone
failed to achieve the target recovery. Also notable was the performance of the standard
diameter Type 2 unit, which exhibited mass recoveries to underflow of around 90%.
Screen
Laser
Corrected
0.0
0.1
1
10
100
C
u
m
u
l
a
t
i
v
e
%
P
a
s
s
i
n
g
0.1 1 10 100 1,000
Size, mm
FIGURE 5 Comparison of original and corrected sizing results
LARGERDIAMETER HYDROCYCLONE PERFORMANCE IN A DESLIMING APPLICATION 327
* VF = vortex finder; Sp = spigot diameter.
TABLE 1 Typical separation performance
Type (diameter) VF,
*
mm Sp,
*
mm Press, kPa Density, % I d
50c
1 (100 mm) 40 25 180 30 0.50 12.57
2 (100 mm) 41 26 180 30 0.38 9.74
1 (250 mm) 60 40 180 30 0.28 10.61
2 (250 mm) 61 38 180 30 0.41 10.73
40
45
50
55
60
65
70
U
/
F
D
e
n
s
i
t
y
,
%
S
o
l
i
d
s
100 150 200 250
Pressure, kPa
35% Solids Primary
30% Solids Primary
25% Solids Primary
FIGURE 6 Hydrocyclone pressure versus U/F density
TABLE 2 U/F density and mass recoveries to underflow for each make (feed density 30% w/w)
Type
(Diameter)
Pressure,
kPa VF, mm Sp, mm
U/F density,
% w/w
Mass Recovery to Underflow,
%
1 (100 mm) 120 40 25 56.59 83.46
180 40 25 60.62 79.74
240 40 25 60.99 82.91
2 (100 mm) 120 41 26 55.08 90.33
180 41 26 53.74 90.83
240 41 26 52.38 91.04
1 (250 mm) 120 60 40 60.42 85.55
180 60 40 60.48 84.05
240 60 40 60.38 84.32
2 (250 mm) 120 61 38 56.10 84.77
180 61 38 54.77 82.61
240 61 38 56.89 85.09
328 ADVANCES IN COMMINUTION MILL DESIGN
Cutpoint d
50c
As with tests manipulating the feed density, the vortex finder–spigot area ratio was held
as constant as possible to highlight the effects of different pressures on the measured
cutpoint. In all cases, the corrected cutpoint decreased as pressure increased. However,
the impact of the shorter cone length for the 100mmdiameter Type 1 was evident in
that it failed to achieve the target cutpoint of 10 Pm at any of the pressure settings
tested. Both makes of 250mmdiameter hydrocyclones achieved the target within the
pressure range tested. Table 3 shows typical results for each unit.
Hydrocyclone Throughput/Flow Rate
It was expected that the hydrocyclone flow rate and, hence, throughput would increase
with increased hydrocyclone diameter. Table 4 shows that this was true. However, there
are higher reported flow rates for the Type 2 units compared with the Type 1 units of the
same diameter. This result quantifies the extra capacity of the longcone units.
As expected, Figure 7 shows that the differences in flow rate translated into similar
differences in dry throughput as a function of pressure.
The results presented in Table 4 and Figure 7 highlight some interesting differences
among the hydrocyclones types. As noted earlier, the 100mm Type 1 hydrocyclone was a
longcylinder unit, whereas the 100mm Type 2 unit was a longcone unit. Note that
while the intention with the additional cylinder length in the Type 1 unit was to provide
similar capacity to a longcone unit, the results show that the longcone unit achieved
consistently higher flow rate and throughputs at equivalent conditions. The contrast is
even more dramatic for the largerdiameter units. It is important to note that the 250mm
Type 1 unit was not fitted with the extra cylinder, so it was of a conventional shortcone
design. Clearly, the results on the 250mmdiameter units show the substantially higher
capacity for the longcone unit.
Secondary hydrocycloning tests were also conducted for both types of hydrocy
clones using primary hydrocyclone underflow as feed, and the trends in the results were
consistent with those described above.
A final comparison was conducted at optimum conditions for secondary hydrocy
cloning. In terms of the cone length, given that only the Type 2 hydrocyclones were
directly comparable, it is this comparison that is shown in Table 5.
Clearly, both units satisfy all criteria for this assessment; however, in all measures, it
is the largerdiameter unit that provided superior performance.
TABLE 3 Change in cutpoint with pressure (30 wt % solids)
Type (Diameter) Pressure, kPa VF, mm Sp, mm Area Ratio d
50c
1 (100 mm) 120 40 25 2.56 12.90
180 40 25 2.56 12.57
240 40 25 2.56 10.83
2 (100 mm) 120 41 26 2.49 9.98
180 41 26 2.49 9.74
240 41 26 2.49 9.26
1 (250 mm) 120 60 40 2.25 11.74
180 60 40 2.25 10.61
240 60 40 2.25 9.52
2 (250 mm) 120 61 38 2.58 11.23
180 61 38 2.58 10.73
240 61 38 2.58 9.46
LARGERDIAMETER HYDROCYCLONE PERFORMANCE IN A DESLIMING APPLICATION 329
TABLE 4 Flow rates for different operating pressures (30 wt % solids)
Type (Diameter) VF, mm Sp, mm Pressure, kPa Flow Rate, L/sec
1 (100 mm) 40 25 120 3.94
40 25 180 5.04
40 25 240 6.19
2 (100 mm) 41 26 120 4.34
41 26 180 6.56
41 26 240 7.79
1 (250 mm) 60 40 120 11.30
60 40 180 15.50
60 40 240 19.12
2 (250 mm) 61 38 120 14.40
61 38 180 18.20
61 38 240 22.74
0
5
10
15
20
25
30
35
40
100 120 140 160 180 200 220 240 260
T
h
r
o
u
g
h
p
u
t
,
t
p
h
250mm Type 2
250mm Type 1
100mm Type 2
100mm Type 1
Pressure, kPa
FIGURE 7 Throughput performance for the units tested
TABLE 5 Comparison of the performance of the Type 2 hydrocyclones (secondary hydrocyclone
optimum)
Performance Characteristics
Type 2 (100 mm)
VF 41 mm, Sp 26 mm
Type 2 (250 mm)
VF 61 mm, Sp 38 mm
Pressure, kPa 180 180
Feed density, % 33 33
I 0.37 0.34
Dry throughput, tph 10.2 32.4
U/F density, % 56 65
Recovery to underflow, % 88 93
d
50c
, mm 10.1 8.1
330 ADVANCES IN COMMINUTION MILL DESIGN
CONCLUSI ONS
The results achieved in this assessment demonstrate that largerdiameter hydrocyclones
may be used in fine desliming applications provided that careful attention is paid to set
ting appropriate operating conditions. In this work, the performance of largerdiameter
(250 mm) units at least equaled—and in the optimum secondary hydrocycloning case,
exceeded—that of the smallerdiameter (100 mm) standard unit in terms of the perfor
mance criteria, namely corrected cutpoint of d10 Pm; highquality separation, as evi
denced by a low partition imperfection (<0.5); hydrocyclone U/F density of >55% solids
by weight; high mass recovery to underflow (>85%); and high throughput/capacity.
These results illustrate the opportunity to realize significant potential gains in circuit
capacity, lower circuit footprint, and simpler operating configuration, and hence, lower
costs with no loss of metallurgical performance by the use of larger and fewer hydrocy
clones in this application.
REFERENCES
Chu, LY., MW. Chen, and XL. Lee. 2000. Effect of structural modification on hydrocy
clone performance. Separation and Purification Technology 21:71–86.
———. 2001. Effects of geometric and operating parameters and feed characters on the
motion of solid particles in hydrocyclones. Separation and Purification Technology
26:237–246.
Kawatra, S.K., A.K. Bakshi, and M.T. Rusesky. 1996. Effect of viscosity on the cut (d
50
)
size of hydrocyclone classifiers. Minerals Engineering 9(8):881–891.
Kraipech, W., W. Chen, F.J. Parma, and T. Dyakowski. 2002. Modelling the fishhook
effect of the flow within hydrocyclones. International Journal of Mineral Processing
66:49–65.
Lynch, A.J., and T.C. Rao. 1975. Modelling and scaleup of hydrocyclone classifiers.
Pages 9–25 in Proceedings 11th International Mineral Processing Congress, Cagliari,
Italy.
Plitt, L.R. 1976. A mathematical model of the hydrocyclone classifier. CIM Bulletin
69:114.
331
Selection and Design of Mill Liners
Malcolm Powell,
*
Ian Smit,
†
Peter Radziszewski,
‡
Paul Cleary,
§
Bruce Rattray,
**
KlasGoran Eriksson,
††
and Leon Schaeffer
‡‡
ABSTRACT
Dramatic shortcomings of mill liner designs, especially of large semiautogenous grinding
(SAG) mills—such as rapid failure, mill shell damage arising from the charge impacting
directly on the liner, and unsuitable spacing of lifter bars yielding unfavourable compro
mises between lifter bar height and liner life — highlight the significance of correct mill liner
selection. Liners protect the mill shell from wear and transfer energy to the grinding charge.
A careful balance is required to optimise these conflicting requirements. This review serves to
highlight these problems and addresses logical and often inexpensive resolutions by consid
ering charge trajectories and liner spacing criteria, in conjunction with liner wear monitoring.
An overview of the principal types and materials of liner construction is given, with a
focus on liner design based on the best technology available, combined with experience and
logical engineering practice.
Methods of monitoring the progressive wear of liners and their relation to the perfor
mance of the mill are presented. The value of wear monitoring in ongoing liner optimisation
and cost saving, through balancing the longevity of the lifters and shell plates and providing
reliable comparative data for testing different liner materials and designs, is explained.
Weartesting techniques and their drawbacks and limitations are discussed, along with new
tests that are under development.
The contribution of advanced computation techniques, such as the Discrete Element
Method (DEM), to predict the wear profiles of liners and integrate this information into
optimising the overall performance of the mill from a production and cost perspective are
considered in some detail. This takes into account the change of the charge trajectories,
energy transfer, and milling efficiency as the mill liner wears and the profile changes.
It is hoped that this review will enable mill operators to select suitable mill liners, with a view
toward decreasing production costs while maintaining mill performance near optimal levels.
I NTRODUCTI ON
Poor liner design has a detrimental effect on milling performance and on liner life (Powell
1991a), which can result in a loss of revenue and increased operational costs. Reduced
* Mineral Processing Research Unit, University of Cape Town, Cape Town, South Africa
† Anglo Research, Johannesburg, South Africa
‡ Department of Mechanical Engineering, McGill University, Montreal, Canada
§ CSIRO Mathematics and Information Sciences, Clayton, Victoria, Australia
** Castech Solutions, South Fremantle, Perth, Australia
†† Mill Linings Systems/Technical Support, Metso Minerals, Ersmark, Sweden
‡‡ Mill Linings, Weir Rubber Engineering, Salt Lake City, Utah
332 ADVANCES IN COMMINUTION MILL DESIGN
milling efficiency can cause excess power usage and decreased recovery of valuable min
erals. Excess liner wear results in exorbitant liner materials costs and excessive down
time, which reduces mill availability and impacts on plant throughput. For a plant with a
number of mills, this also entails the employment of extra mill relining staff and the risks
and costs associated with frequent relining. Optimised liner design can be used to strike
the best economic balance between liner life and mill grinding performance, thus
enhancing the profitability of a mining operation.
Protection of the mill shell from the aggressive impacting and abrasive environment
inside a mill is the primary purpose of mill liners. Generally, the care of liners came
under the maintenance and engineering department, where the objective was to utilise a
liner that lasted as long as possible, or was as inexpensive as possible, or, preferably
both. Liners were treated merely as a cost overhead and a cause of downtime, and the
maintenance approach has been to reduce the cost while remaining within acceptable
downtime constraints. Cost savings led to the development of profile liners and lifter
bars, as these dramatically increase the life of the liner. The downtime constraints and
high stresses in large SAG mills helped to drive the development of greatly improved
liner materials. However, this costengineering approach ignored the mill performance
and overlooked the other key function of mill liners.
The second critical function of a liner is to transfer rotary motion of the mill to the
grinding media and charge. After all, the liner is the interface between the mill and
the grinding charge. Although literature on the grinding action in mills has been pub
lished for 100 years (White 1905; Davis 1919), the first publication on the influence of
liner design on the charge motion appeared about 70 years later (McIvor 1983). With the
advent in the 1980s of larger SAG mills running in singlestream circuits, it became
apparent to the operating staff that the liner was having a significant influence on mill
performance. This had been hidden previously by the regular changing of liners over a
number of mills in the older plants that had many mills in parallel function. In fact, this
is generally still the case in the multistream plants, where mill liner design and selection
is only tackled on a costconsumables basis. However, the gains to be had through good
liner design and selection are just as great as on the large SAG mills.
This paper discusses recognising problems in liner design and selection in existing
operations and reviews liner selection for new applications.
TYPES OF LI NERS
Design and Structure
The design of a liner is driven by the material of construction and the application, and is
limited by casting, moulding, and handling constraints. For large mills with wide inlet
trunnions in excess of 1.5 m, liner handling machines are now in common use, and this
has allowed the evolution of large integral liner blocks, each weighing up to 1.5 t
(Figure 1). This holds great advantage for minimising relining time, as there are fewer
blocks to handle. For example, at the Kalgoorlie Consolidated Gold Mines (Western Aus
tralia), the number of outer head liners in a 36ft SAG mill was reduced from 36 to 18 pieces,
and in doing so, the time to replace them was reduced by 9 hours at a costdowntime
savings of about US$25,000/hr. In this case, the liners sections are 3.5 t each.
For smaller mills, the liners have to be handled and installed manually, so smaller
blocks with removable lifter bars are generally favoured (Figure 2). Following is a list of
the primary types of liners, including comments on their application, advantages, and
disadvantages:
SELECTION AND DESIGN OF MILL LINERS 333
1. Solid liners: These types of liners have an integral lifter and liner, as shown in
Figure 1. They have fewer pieces and are easier to install, but they tend to have a
high scrap weight, as once the lifter section is worn down, liner performance
drops and necessitates changeout.
2. Removable lifter: In a liner with a removable lifter, the lifter can be changed
rather than the complete liner (Figure 2), thereby maximising liner life and
assisting in manually relined mills. The drawback is that there are more pieces to
be installed, and the liners can move during relining. If they are not well secured
against the backing liner, the lifter can shift and work loose; this is especially a
problem if the bolts begin to stretch.
3. Grid liners: Pocketed grid liners is a system (that appears to be unique to South
ern Africa) where the grinding media packs in the grid structure and forms an
integral part of the liner (Figure 3). Often the liners have a flat profile, suited to
the high speeds (85% to 90% of critical) at which most of the older mills operate.
These liners have been demonstrated to be economically unbeatable for highly
abrasive ores in small to mediumsize mills (Powell 1991a). They are lightweight
and make use of the grinding media hardness to provide an effective wear material.
They must be manufactured in manganese steel to wedge the steel balls, but the
manganese steel spreads on impact and can make removal difficult. Safety
FIGURE 1 Solid steel liners with integral lifter bars
FIGURE 2 Removable lifter bars
334 ADVANCES IN COMMINUTION MILL DESIGN
aspects should be considered because of the risk of balls dislodging when the
mill is entered for inspection or relining. The liners require a thorough hosing
down to prevent this.
4. Wedged liners: Wedged liners were common in the first half of the last century
but are dangerous to install and no longer used. Liner blocks are now wedged in
by bolted lifter bars, which allows simple castings of the liner blocks.
5. Integral wave blocks: These are commonly used in ball mills, and the profile of
the liners has become sufficiently sophisticated to enable the liner profile to be
maintained as the liner wears.
6. Unidirection profiled liners: The lifter has different leading and trailing profiles
(Figure 4). The profile can be better customised to suit mill speed and filling and
therefore optimise performance. It allows more material in the lifter for a given
base width, but the mill must only run in one direction.
7. High–low doublewave ball mill liners: These liners are a refinement of the wave
liner (Figure 5). This was applied to Cadia Hill gold mine through evaluation of
their existing wear profile and wear rate, and it provided a more consistent wear
profile through the liner’s working life. The correct wave face angle needs to be
calculated and applied because an incorrect angle can lead to ball segregation
and loss of grind.
An indication of our limited ability to accurately “design” liner profiles is that few
liners are optimal at original installation or in the postcommissioning set, and it is
imperative as a user to vigorously pursue improvement of the design to get the most out
of the liners.
FIGURE 3 Austenitic manganese steel grid liners
FIGURE 4 Singledirection tophat liners: an integral liner (left), and bidirectional liners with
removable lifters (right)
SELECTION AND DESIGN OF MILL LINERS 335
Materials
The selection of the construction material is a function of the application, abrasivity of
ore, size of mill, corrosion environment, size of balls, mill speed, and so forth. Liner
design and material of construction are integral and cannot be chosen out of context.
Following is a list of the primary materials of construction, including particular uses and
strengths.
1. Austenitic manganese steel (AMS): AMS is used for grid liners generally, in
smaller mills. Its great advantage is that it hardens under stress, yet the substrate
remains tough and can withstand extreme impacting without fracture. Its pri
mary disadvantage is that it spreads with impact, so solid liners begin to squeeze
together and become extremely difficult to remove, and can damage a mill shell
if the stress is allowed to build to an extreme level.
2. Lowcarbon chrome moly steel (300 to 370 BHN [Brinell hardness number]):
Generally used for mill liners (autogenous grinding [AG], SAG, and ball) prior to
the movement to highercarboncontent steels. It has excellent wear characteris
tics with some impact resistance and is generally used for discharge grates where
slightly better impact resistance is required, compared to the higher carbon
chrome moly steels or for thinner section liners.
3. High carbon chrome moly steel (325 to 380 BHN): This steel is now considered
the main material used for SAG mill liners. There are a number of variations with
either different carbon or chrome contents. The variations tend to have a bearing
on the size of the liner and its section thickness. There is ongoing development
within this area as the size of the liners are outstripping the properties provided
by the standard high chrome moly steels.
4. Nihard iron (550 BHN): The use of this type of material generally began with rod
mills and ball mills, where impacts were considered low enough for this brittle
yet highly abrasiveresistant wear material to perform well. However, it is consid
ered obsolete in light of the use of high chrome irons and chrome moly white iron.
FIGURE 5 High–low wave ball mill liner
336 ADVANCES IN COMMINUTION MILL DESIGN
5. High chrome irons (+600 BHN) chromium iron: This iron is considered to have
superior wearabrasion characteristics, and it is generally used in rod and ball
mills. It is more costcompetitive and more brittle than chrome moly white irons.
6. Chrome moly white irons (600 to 700 BHN): This cast material is considered to
be the ultimate and was developed and used to date for abrasion resistance in
milling. It is commonly used in cement mills and in some of the largest ball mills
in the world, and where performance has not been improved to date.
Rubber Liners
The interplay of a material and its configuration are especially significant in rubber lin
ers. During the last half century, rubber mill linings have been used successfully in sec
ondary and regrind milling applications and are specified today for these new
applications (Figure 6). However, now with improved materials and computeraided
design programs, they are being used more and more in primary grinding applications.
In addition to their abrasion resistance, they also are resistant to most chemicals
(Schnarr, Schaeffer, and Weinand 2002).
The more technical term for rubber is “elastomer.” A good elastomer for a mill liner
would have an elongation of 500% to 600%, which means that it can be stretched five to
six times its length without damaging it. In addition, the tensile strength should be
around 20.68 MPa (3,000 psi). The third important physical characteristic is hardness,
and this should be between 55 and 70 durometer on the A scale.
The material used for a rubber mill liner usually consists of a blend of a natural and
synthetic rubber. In some applications, the material may be all synthetic. The mixture of
the rubber and synthetic materials plus various chemicals and fillers is called a “compound.”
Each rubber milllining manufacturer has their own recipes for their compounds, as well
as their own designation.
In designing a rubber lining, the same computer tools as described elsewhere in this
paper are used. Whether the lining material is metal or rubber, the same type of commi
nution is required in the charge; therefore, the same simulation tools can be used with
some adjustments for the lining material. For maximum life, rubber performs best with a
90˚ impact, so this is taken into consideration when designing. Many improvements in
FIGURE 6 Rubber lining in a ball mill and feed head metalcapped lifter
SELECTION AND DESIGN OF MILL LINERS 337
rubber compounds have been made over the years, but current research and develop
ment is providing nanotechnology, which should further improve wear life in the future.
Rubber and Steel Composites. Rubber and steel have been used successfully in
many applications. In some cases, the rubber and metal can be separate components
(i.e., metal lifter bars and rubber plates). During the last two decades more emphasis has
been placed on metalcapped rubber lifter bars (Figures 6 and 7). The material used for a
metal cap is similar to that used for a metal lining, but a hardened steel plate can also be
used. The joining of the metal and rubber has to be with a chemical bond plus a mechan
ical type of attachment to ensure a positive fastening of the two materials for the life of
the component.
With the use of computer simulations and careful inspection of the existing wear
profile, a greatly improved liner design can be generated. Thus, all high, or high–low, or
lowersloped lifters are recommended for different applications, as illustrated in Figure 7.
Rubber is oneseventh the weight of metal, and in many countries the cost is less, so it is
very beneficial to utilise rubber wherever possible. By strategically placing the metal cap
material with the minimal amount of metal, the best economy can be obtained. Some
applications require a metal leading face only. Others require a metal face and top pro
tection (Figure 7).
An important feature of combination linings for ball mills is the configuration of the
lining, where the lifting action is transferred to the charge, and therefore will remain
constant throughout the life of the lining; whereas solid linings will wear more on the
lifting portion and become smoother with less lifting action as the lining wears down.
The different wear characteristics of the two materials in Skega PolyMet (Mill Linings
Systems, Metso Minerals, Ersmark, Sweden) make it possible to design a lining that will
maintain its profile throughout its life, as illustrated in Figure 8.
Magnetic Liners
The lining system in magnetic liners consists of permanent magnets embedded in a rub
ber moulding. The powerful magnets keep the lining in place without liner bolts and
ensure that the lining attracts magnetically susceptible material available in the mill
(Figure 9). The particles attracted to the surface of the magnetic lining form a thin, con
tinuous layer in a wave profile.
The total thickness of the magnetic lining, including the wear layer, is much less
than that of a conventional lining. The mill will thus have a larger effective diameter. The
lining configuration is ideal for fine grinding, giving an efficient grinding performance in
FIGURE 7 Configurations of metalcapped rubber liners
338 ADVANCES IN COMMINUTION MILL DESIGN
these applications. A combination of the previously described features has resulted in
higher throughput (or lower energy consumption) and, in several cases, a lowering of
media consumption by at least 10%.
Because of the complicated manufacturing process, the magnetic liner elements are
much more expensive than conventional rubber lining, but in an ideal application, the
wear on the lining is almost negligible and therefore can give years of troublefree
operation.
The limitation for this lining concept is that the magnets are not very resistant to
impact because they are brittle. They are suitable in mills of >12 ft diameter using maxi
mum 1in. balls, and in mills s12 ft in diameter using maximum 1
1
/2in. balls. Magnetic
liners have been utilised successfully in vertical stirred mills.
An example of the application of the Metso OreBed lining is at the LKAB Kiruna iron
ore operation in Sweden (K. Tano, personal communication, 2005). A test on the pri
mary ball mill showed that the conventional PolyMet lining slightly outperformed the
OreBed lining in terms of a 0.5tph higher throughput, so the plant remained with Poly
Met. However, they installed the OreBed lining in all their pebble mills, where they have
successfully operated for more than 10 years without any maintenance or replacement.
Figure 10 shows the liner with the coating of magnetite ore and slurry. It was concluded
at the site that the magnetic lining works well for secondary grinding, where abrasion is
more important than impact.
6
0
9
0
FIGURE 8 Designing steelcapped liners for even wear
Homogeneous Bed of Fine Magnetic Material
Coarser Bed of Small Pieces of Magnetic Material
Fluid Bed of Fine and Coarse Magnetic Material
Rubber Steel Shell Permanent Magnet
S
S
N
N
FIGURE 9 Magnetic liner (Metso OreBed)
SELECTION AND DESIGN OF MILL LINERS 339
HOW GOOD I S CURRENT DESI GN?
Liner installations have resulted in variable performance, from outstanding to disas
trous. This range indicates the potential of good design and application, as well as the
potential for poor installations. In this section, a few case studies will highlight this
range, what was identified as the cause, and the lessons learned.
Good Examples
Two examples of successful liner design applications that highlight key design aspects
are as follows:
Cadia Hill gold mine 40ft SAG mill—reduced rows: The primary objective was to
reduce the packing of material between the existing high–low liner design. A two
thirds row design was installed, which allowed for increased spacing between lift
ers and thus the use of a larger release angle. Packing was eliminated and the mill
was able to run at a higher speed, thereby increasing grinding performance and
reducing ball breakage.
Codelco Andina 20ft ball mill—high–low ball mill liners (white iron): The objec
tive was to increase liner life, thereby increasing plant availability and reducing
running costs. The high–low wave profile liner developed at Cadia was installed
in chrome moly white iron. The mill operating performance was not hampered,
yet liner life increased by more than 50% over the previous sets of doublewave
liners that had been in use.
Bad Examples
Some disasters with the emphasis on cause, and the lessons learned, are presented:
Largediameter SAG mill in South America—pulp lifter design: The outer pulp
lifter had not been designed correctly to allow for ease of removal without the
need to remove the shell liners adjacent to it. This arose from a lack of knowledge
of how relines are carried out and the importance of timely removal during shut
downs. This highlights the need to discuss liner design changes or new concepts
with the maintenance crews and reliners so as to detect design retailing flaws.
FIGURE 10 OreBed liner in secondary ball mill, and a single panel
340 ADVANCES IN COMMINUTION MILL DESIGN
Large ball mill—white iron liners: The existing liner design had always been con
structed from high carbon chrome moly steel; however, to increase liner life, the
move was made to use chrome moly white iron. However, the liners cracked
severely during installation. Bolthole details had remained unchanged, but on
close examination, the bolthole profile was found to not be a tight fit with the
liner bolt. This allowed pinpoint loading to occur, which acted like a guillotine,
cracking the liners down the bolthole centre line, as illustrated in Figure 11. This
demonstrates the need for a full reevaluation of the liner design, with close atten
tion to fit faces (curves, angles, etc.), clamping face, lifting lugs, and bolthole
shape when changing liner material.
Dimpled, peened, and cracked SAG liner: Examples of dimpled, peened, and
cracked SAG liners in a 24ft SAG mill are shown in Figure 12. The mill was found
to have the correct lifter profile and filling, but the feed was excessively diluted
(to assist in flushing feed through a poorly designed feed chute). From listening
carefully to the mill (in the absence of a proper microphone system), it was con
cluded that the angle of repose of the charge was abnormally steep, resulting in
the toe being very low in the mill. The dilute charge also significantly removed the
padding influence of the charge contents. The feed chute had to be reconstructed
before this problem could be resolved.
FIGURE 11 Bolthole cracking
FIGURE 12 Heavily dimpled, peened, and cracked liners
SELECTION AND DESIGN OF MILL LINERS 341
It is difficult to comprehend, given the number of SAG mills in operation globally,
that the design of SAG mill liners is still largely troublesome. A number of papers from
the SAG 2001 conference (CIM 2001) refer to the following problems:
Candelaria: “Liner design progressed from 72 lifter rail type design, with aggres
sive face angle, to 36 lifter design with 35 degree face angle. Mill throughput
increased by 15%.”
Alumbrera: “Three years worth of trials have been conducted to optimise lifter
geometry. Liner progression from 72 to 48 to 36 rows of lifter bars. Could origi
nally not operate SAG mill at speeds in excess of 70% of critical because of impact
on the shell. Operation at reduced speed resulted in low power draw and reduced
throughput.”
Los Pelambres: “SAG mill liner progression from 72 rows with 8 degree face angle
to 36 rows with less aggressive 30 degree face angle. Changes allowed for the
mills to be operated safely at higher steel loads without increased risk of liner
damage. Increased power draw resulted in increased primary mill throughput.”
Collahuasi: “The original SAG mill liner design, HiHi with a 6 degree contact
angle, was changed to a 17 degree angle and later to 30 degree angle. An 11%
increase in mill throughput was achieved.”
It is common practice to rely on the mill vendor to supply the liner design. This is
quite often a contractual requirement with respect to the vendor guaranteeing the per
formance of the mill. When the mill does not achieve the required throughput rate—
because it has to be operated differently from that originally intended in order to prevent
liner damage—most vendors appear to adopt a trialanderror approach, leading to the
iterations such as those referred to previously. The net result is a lengthy rampup time
and loss of production.
SYMPTOMS OF POOR LI NER DESI GN
When liners are performing to expectations, they are usually left as is. It is usually only
when they suffer from premature failure, or come under the cost spotlight, that they are
assessed and their performance scrutinised. This section of the paper provides basic
guidelines to assessing whether the liners may require reevaluation and identifies possi
ble causes of problems in liners that are known to be problematic.
Noisy Mill
A distinct impact rattling indicates that the grinding media is impacting directly on the
liner, rather than on the toe of the grinding charge.
The consequences of balls impacting on the liner are
1. Greatly accelerated liner wear due to highenergy impacts on the liner. For a 5m
diameter mill, these will be in excess of 8 m·s
–1
; for a 10mdiameter mill, these
exceed 12 m·s
–1
. This causes high chrome liners to spall and crack, and they may
even fracture. This can reduce liner life from a year to less than a few months.
Rubber liners can split and tear under these excessive stresses, especially when
worn to partial thickness.
2. Reduced milling efficiency from the highest impact collisions occurring against
the liner instead of on the toe of the charge—where effective work can take place
3. Lowered power draw from dilation of the mill charge and the balls returning
energy to the mill shell rather than to the mill charge
342 ADVANCES IN COMMINUTION MILL DESIGN
4. Ball fracture from the highvelocity impacts directly onto steel—an impulse force
that many balls cannot withstand. This results in a loss of topsize grinding
media and reduced milling efficiency, plus increased ball addition costs.
5. Loosening of liner bolts arising from high stress on the liners stretching the bolts.
The damage consequences of rocks impacting the liner are essentially the same as
for ball impact, but the impact forces are lower and less detrimental. Autogenous milling
takes place when the larger rocks land on smaller rocks and transfer their large energy to
the smaller rocks, which can then be sufficient to fracture them (NapierMunn et al.
1996). As the highenergy collisions are occurring on the liner rather than on the rocks in
the toe of the charge, the milling efficiency is reduced.
Mill Listening Devices. It is of great advantage to monitor the sound of impacts
on the mill shell to warn of direct impacting. This has recently taken a step beyond sim
ple decibel monitoring to full Fourier analysis of the frequency spectrum in research
work conducted at the Julius Kruttschnitt Mineral Research Centre (JKMRC; Pax 2001).
Broken Liners
Broken liners can result from media impacting directly on the liner, and this is particu
larly severe for large AG/SAG mills. A high incidence of fracture of lifter bars without
corresponding evidence of porosity or casting faults generally indicates impact breakage.
Excessive Liner Wear
If the liners have a low or flat profile, this generally indicates excessive slip of the grind
ing media on the liner.
Consequences of excessive slip are as follows:
Liner wear increases substantially and can show evidence of circumferential
grooving of the lining.
There is a substantial loss of energy transfer to the mill. Although a small fraction
of grinding may take place at the liner–ball interface, a 10% loss in energy due to
slip results in about a 10% loss of energy transfer to the grinding media.
Reduced mill throughput results from reduced milling efficiency.
Often a lining wears through a favourable profile regime. This can occur in a num
ber of different manners:
Mill throughput drops markedly when new liners are installed. This can indicate
an unfavourable new profile, or an excessively thick new lining—to counteract
excessive wear from poor liner design.
Mill throughput peaks during the liner life, usually at the end of the life of the
liner. This is often a symptom of oversized lifter bars, or lifters with too vertical a
face angle.
Mill throughput decreases towards the end of the liner life, and the liner has a
flattened profile—a sign of excessive slip, and the liner should be replaced sooner.
An incorrect mill product size can result from the incorrect tumbling action within
the mill:
Primary mills require a vigorous action with highenergy impacts to fracture the
ore. If the action is primarily gentle cascading, then a fine product and low
throughput would result.
SELECTION AND DESIGN OF MILL LINERS 343
Regrind mills require a cascading action to maximise the frequency of abrasion
interactions. Highenergy impacting wastes energy, reduces the rate of abrasion
interactions, and reduces the grinding pressure, thus reducing the ability of the
mill to produce fines. This can result in a high recirculating load of oversized par
ticles and a reduced mill throughput, as limited by the required product size.
PI LOT TESTS OF THE I NFLUENCE OF LI NER DESI GN
It can be difficult to assess the influence of liner design upon mill performance in a pro
duction environment, as it tends to be a small influence superimposed upon a number of
operational variations, especially in semiautogenous milling. Standard pilot tests do not
account for the liner design; they use a standard liner profile for all testwork, generally
designed to give adequate lift to the charge at standard mill operating conditions—75% of
critical speed and 25% filling for autogenous/semiautogenous milling. In this section, a
test procedure is presented that can be applied to batch pilot milling.
To assess the direct influence of liner design on mill efficiency, a 1.8mdiameter
pilot mill was utilised (Powell and Vermeulen 1994). The use of batch milling meant that
reasonable size samples (<2 t) of the ore under consideration could be used to provide a
rapid test (a few days) of a number of liner profiles.
The mill was operated in batch mode, and the rate of production of final product—
minus 75 µm, in this instance—was measured when the charge size distribution was sea
soned and the slurry percent solids was at the equilibrium conditions found in a continuous
mill. The results showed a substantial influence of liner profile on the performance of the
mill, as illustrated in Figure 13. Differences of up to 30% in the rate of production of
fines were measured.
S
m
o
o
t
h
0.57
G
r
i
d
s
0.64
4
0
m
m
,
7
0
˚
0.71
7
0
m
m
,
7
0
˚
0.55
S
m
o
o
t
h
0.56
G
r
i
d
s
4
0
m
m
,
5
0
˚
7
0
m
m
,
5
0
˚
7
0
m
m
,
9
0
˚
0.78
0.64
0.60 0.60
0.80
0.75
0.70
0.65
0.60
0.55
0.50
0.45
0.40
R
a
t
e
,
k
g
/
m
i
n
Rate of Production of Fines for Different
Liner Configurations at Two Mill Speeds
80%
90%
FIGURE 13 Influence of liner profile on mill performance
344 ADVANCES IN COMMINUTION MILL DESIGN
The test results were sensitive to when the sample was taken. The rate of production
of fines dropped in time as a normal charge size distribution and slurry density were
attained, and this occurred at a different time for each type of liner. In followup work
(Powell 1993), the technique was refined to sample during the period when a desired
slurry density was reached. The particular exploratory technique described in the works
by Powell (1993) can be improved upon in a number of aspects but, nevertheless,
showed the clear influence of liner profile on milling efficiency. The test requires direct
correlation studies with production mills to provide a quantitative correlation with the
influence in a production mill.
TESTI NG LI NER WEAR RATES
Tests used to predict the wear rate of production liners are notoriously inaccurate, to
such an extent that they are misleading rather than informative. The fundamental prob
lem besetting all these tests is reproducing the wear modes present in production mills.
The contact pressure, rates of relative movement, and abrasive material properties all
play a large role in determining the resultant wear rate. A poorly conceived test can easily
be in error by an order of magnitude and, additionally, will not rank a range of materials in
the correct order. Considering one extreme, a lowpressure sliding abrasion test can
invert the ranking of tested materials relative to an application under high impact. A pre
dictive wear test must be tuned to the application; a single standard test cannot possibly
predict the liner (or ball) wear for a whole host of very different types and sizes of mills.
This is a fundamental problem of the Bond wear test, also known as the Pennsylvania
abrasion test. It was established for, and correlated to, a limited range of similar mills yet
is now expected to provide meaningful results for large SAG mills, well beyond its scope
of application.
NearFieldCondition Testing
Nearfieldcondition testing attempts to reproduce the overall action and forces encoun
tered in the application. A test developed by Powell (Powell and Cornelius 1992) utilised
the same 1.8m batch mill as used for liner profile testing. This could reasonably simu
late the wear modes observed on liners in the 5mdiameter mills being tested.
Small blocks of the material, 200 × 40 × 80 mm, were clamped onto the top of custom
built lifter bars (Figure 14). A limited, and accurately controlled, surface area was
allowed to protrude and be exposed to wear. The rate of wear could then be precisely
monitored by progressive mass loss over a period of a test lasting a few days and requir
ing less than 2 t of ore. The test was not designed to produce an absolute wear rate but
Clamping
Wedge
Clamping
Bolt
Protective
Strip
Sample Hard Facing
Mill Shell
3.00
2.50
2.00
1.50
1.00
0.50
0.00
R
e
l
a
t
i
v
e
L
i
f
e
M
i
l
d
S
t
e
e
l
A
u
s
t
e
n
i
t
i
c
M
a
n
g
a
n
e
s
e
S
t
e
e
l
H
i
g
h
C
r
W
h
i
t
e
I
r
o
n
M
a
r
t
e
n
s
i
t
i
c
M
a
n
g
a
n
e
s
e
S
t
e
e
l
B
a
i
n
i
t
i
c
M
a
n
g
a
n
e
s
e
S
t
e
e
l
FIGURE 14 Wear testing mill and results of relative wear rates
SELECTION AND DESIGN OF MILL LINERS 345
rather a relative wear rate against a standard sample—usually the current production
liner—used in the test. Once this sample had been calibrated in a production environ
ment, the results could be converted to absolute wear rates.
Advantages of this test are
Small samples of test material are required (<1 kg).
Up to six different materials can be tested simultaneously.
Short test period (a few days)
Simple sample geometry (rectangles)
Accurate wear rate figures
Specific to the ore under consideration
Ball load determines the degree and force of impact wear.
Laboratory Tests
As mentioned previously, laboratory tests tend to give misleading results. However, a
new test is being developed as part of the AMIRA P9 collaborative research project
(Radziszewski 2001). The advance of this work over previous techniques lies in the
investigator’s emphasis on duplicating the forces and wear modes found in a production
mill. The measurement and isolation of the forces in a mill have long been a stumbling
block to this approach, but with the advent of DEM techniques and their application to
milling (Mishra and Rajamani 1994a, b; Cleary 1998; Inoue and Okaya 1996; Radziszewski
1986; Radziszewski and Tarasiewicz 1989; Radziszewski and Morrell 1998; Herbst and
Nordell 2001; Zhang and Whiten 1998), the tools are now available to mathematically
derive the required forces. Radziszewski has utilised this computational capability to
guide the design of a comprehensive test that accounts for impact and abrasion, utilising
the forces derived from computational mathematical derivations. He also incorporates
corrosion, thus addressing the three main known causes of material wear (Rajagopal and
et Iwasaki 1992). This test is still in the developmental stage and is currently being
expanded from grinding media to liner wear testing. Initial outcomes should be available
from Radziszewski by the end of 2006.
Background. Charge motion models also can determine the forces acting on mill
liners (Figure 15a), which can be used together with wear models to determine liner
wear (Figure 15b; Radziszewski and Tarasiewicz 1993; Radziszewski 1997).
In the case of liner wear, the main contribution comes from abrasion wear. Abrasive
wear, or the volume of material removed, is a function of the applied force, F, and the
F
g2
F
g1
F
g3
F
gq
F
cp
F
cp–1
F
gq–1
F
c1
F
c2
F
c3
(a) (b)
FIGURE 15 (a) Ball charge forces acting on mill liner, and (b) bevelled liner profile wear simulation
346 ADVANCES IN COMMINUTION MILL DESIGN
distance slid, x, and is described using an abrasive grain represented by angle u, as illus
trated in Figure 16. An abrasive wheel is used to simulate this mode of wear.
The relationship between wear and energy can be described by Equation 1, where
m
abr
represents metal loss per liner; the energy E
abr
is dissipated in grinding on one liner;
H
r
and µ are metal hardness and density, respectively; and u is the abrasion grain angle
(Rabinowicz 1996).
(EQ 1)
Coupling an abrasive wear model to the abrasive forces and energies acting on a mill
liner is the basis for all DEM liner wear models. The model parameters are backcalculated
from operating data of observed liner wear behaviour. These calibrated liner wear mod
els can then be used to simulate the effect of wear on modified liner profiles where the
liner material is of the same material as the original. The challenge is to determine these
wear parameters from laboratory tests and correlate them to operating data. This will
develop the capacity to predict the effect of changes of liner materials on liner wear as
well as predict the liner wear in greenfield applications.
Abrasion Test Development. The standard abrasion wheel (Figure 17; Misra and
Finnie 1980) has been extensively used to investigate abrasive wear under varying con
ditions, materials, and abrasives.
Gore and Gates (1997) first investigated the substitution of the rubberlined wheel
with a steel wheel that used an abrasion force between 45 and 130 N. Radziszewski
(1997, 2001) introduced the use of mineral ore as the abrasive. In these initial tests, it
was shown that the abrasion grain angle was a function of the applied force. These tests
led to the development of a test capable of applying forces up to 1,000 N. The evolving
design of the abrasion experiment has incorporated measurement of the torque acting
on the abrasion wheel. The net power consumed by the abrasion wheel is the product of
the torque acting on the abrasion wheel and the wheel rotation speed expressed in radi
ans per second. The torque can be used to determine the friction coefficient µ.
Modifying the relationship of Equation 1 to include both the abrasion grain angle
function and friction leads to Equation 2. The product µFx describes the energy lost in
abrasion or net energy consumed by the abrasion wheel.
(EQ 2)
Initial test results confirm the need to generate forces applied in abrasion that
approach those found in real mills. Figure 18 shows that the wear rates can be significantly
h
Abrasive Grain
2r
x
F
Volume
Removed
θ
Bearing
Surface
Source: Rabinowicz 1996.
FIGURE 16 Abrasion wear description
m
abr
µ
u ( ) tan
tH
r
E
abr
=
m
abr
µ
u F ( ) ( ) tan
tH
r
µFx =
SELECTION AND DESIGN OF MILL LINERS 347
different for tests at similar energies—force multiplied by the sliding distance for a given
time—but with a range of applied forces that give rise to different energy rates. As the
objective is to predict the abrasive wear contributions to liner wear, it becomes impor
tant to apply forces representing those found in real mills; hence, the critical importance
of linking this to realistic DEM simulations of mill forces.
Future Directions. With the evolution of the abrasion test, promising results
should contribute to a better understanding of the parameters affecting wear in tumbling
mills. Future work should address the contribution of impact wear, both to work harden
ing of mill liners and media as well as the effects of ore breakage on liner wear. The
sophisticated outputs of DEM simulations (discussed elsewhere in this paper), when
Abrasive
Test Specimen
Weight
Abrasive
Hopper
½ in. × ½ in.
Chlorobutyl
Tire
8 in.
Dia. Wheel
(a) General SetUp (b) Initial Test Apparatus
Source: Misra and Finnie 1980.
FIGURE 17 Abrasion wheel setup
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0 500 1,000 1,500 2,000
Energy Rate, J/s
M
a
s
s
W
o
r
n
,
m
g
/
s
aa
bb
cc
dd
ee
FIGURE 18 Mass worn differences for similar energy rates (1045 steel, Ottawa foundry sand)
348 ADVANCES IN COMMINUTION MILL DESIGN
linked with wear experiments that can reproduce the simulated wear modes, is undoubt
edly the best available technique for development of a truly predictive weartesting
methodology.
Plant Trials
Plant trials can be used effectively to select the liner material and design most suited to
an application. They do, however, suffer from a number of drawbacks:
They take a long time—usually more than a year—to yield results.
The tests are expensive—the cost of a set of liners that may or may not perform
well is no trivial matter.
There is downtime involved in installing and monitoring the liners.
Results can be inconclusive as plant conditions vary with time.
Monitoring of the installation and removal masses, and recording dates of individu
ally marked and located liner blocks comprise the best approach for obtaining reliable
results. Monitoring liner wear throughout the test is also a vital component to meaning
ful comparison. Monitoring two identical mills in tandem before and during the test is a
useful technique, as used in assessing a mill with a new liner design (Powell 1991a) in
which a 5% change in throughput was measured.
An alternative method of conducting field trials is to test a number of materials
simultaneously in one mill. This gives a direct comparison but requires sensible test
design and wear monitoring to provide useful data. Such a trial was conducted by Powell
(1991a) and provided conclusive data on the wear life, wear rate, and liner cost per ton
milled for six very disparate materials—white iron, high chrome white iron, solid manga
nese steel, manganese grid liners, rubber, and mild steel.
Historic Data. In the absence of a reliable test, production mill wear data from
similar applications are possibly the most useful data to use for assessing the wear rates
of liners in a new mill. Keep in mind the following points when “looking over the fence”:
The ore must be similar.
The applications must be similar: AG, SAG, total mill load, mill speed.
Obtain data on new and fully worn masses and thicknesses to obtain the actual
wear rates, and compare different materials and sites, which have different fully
worn dimensions.
I NFLUENCE OF LI FTER BAR HEI GHT ON LI NER LI FE
It is well known that the height of the lifter bar directly influences the life of the liner
block as a whole—the higher the lifter bar, the lower the wear rate of the liner. The trade
off is in mill production, which drops as the lifter bar height is increased above an opti
mal height. In assessing the wear of a liner, it is useful to monitor the wear of the backing
block relative to the height of the lifter bar. This can be used to assess the useful operat
ing height of the lifter bar. In work conducted on monitoring the wear of liners, this mea
sure provided a direct correlation between lifter height and liner wear (Powell 1991a).
This is illustrated in Figure 19, which shows a drop of >40% in liner wear rate when the
lifters were renewed from 35 mm to 80 mm in height. This dramatic change clearly
showed that the lifters were being left in for too long, and the economy of extending the
life of the lifter bars was imposing a more expensive penalty of accelerated wear of the back
ing liner. The data also show how different materials respond differently to the degree of
protection.
SELECTION AND DESIGN OF MILL LINERS 349
The onset of accelerated liner wear also can be used to give a reliable indication of
when slip of the charge on the liner begins to occur, with the resultant reduction in mill
ing efficiency. For this testwork, the wear rate had risen rapidly during the wear of the
first set of lifter bars, levelling out when the lifters were about 50 mm high (0.17 Mt point).
A further output of thorough liner wear monitoring is the ability to balance the life
of the lifter bar with that of the backing liner. Whether the lifter bar is removable or inte
gral to the liner, this is an essential piece of information in optimising the life and cost of
liners. Ideally the lifter bar and backing plate should wear to minimum productivity
height and safe thickness simultaneously. This prevents the scrapping of excess unworn
liner and losses in mill throughput while the maintenance manager tries to maximise
liner life. The technique also can be used to allow for uneven wear along a mill by
designing lifters of different heights along the length of the mill, so as to enable a single
changeout routine.
Monitoring the full profile of a liner as it wears is useful in providing the input for cal
culating the changing charge trajectories as the profile wears. This information can be used
to change the operating window of a mill, such as minimum filling and maximum speed.
The trajectories and safe operating window are illustrated, and the monitoring techniques
are presented, in the Utilising Charge Trajectory Predictions subsection in this paper.
OPTI MI SI NG LI NER DESI GN
It is not necessary to select either a long liner life or a high mill throughput; the two can
be optimised simultaneously with suitable liner selection. There are a number of tools
available for guiding the design of liners so as to help optimise the profile to suit a partic
ular application. The benefits can be substantial, with dramatic changes in liner life,
downtime, and liner cost. At the same time, mill productivity can be held around the
optimum if the interplay of liner design and mill operation is understood. Tuning liners
to an application has many potential benefits:
1. AG/SAG maximises media drop height to maximise impact grinding
2. Ensures a cascading action for regrind mills
40
60
80
100
120
140
160
180
200
0.0 0.1 0.2 0.3 0.4 0.5
Mass of Feed Treated, Mt
W
e
a
r
R
a
t
e
,
m
m
/
M
t
o
f
F
e
e
d
Mild Steel
HighChromium White Iron
White Iron
Rubber
AMS Grids
Source: Powell 1991a.
FIGURE 19 Liner wear rate after lifter bar renewal
350 ADVANCES IN COMMINUTION MILL DESIGN
3. Avoids impact on the mill shell
4. Maximises liner life by providing sufficient protection with lifter bars
5. Prevents ball breakage by avoiding impacts directly onto the shell
6. Maximises mill throughput with the correct spacing and height of lifter bars—an
essential factor in liner design
7. Balances liner life and mill throughput with a protective profile of the lining
while retaining the correct charge motion of the grinding media
Successful Applications of Improved Liner Design
There are many cases of successful changes to liner designs that have greatly increased
the liner life and economy; a few examples follow. At Deelkraal mine of Goldfields Ltd.
(Powell 1991a), a wave profile liner in a primary singlestage runofmine mill had been
modified by placing a 100mm lifter bar in the recess.
This was a poor design option, as the lifter protruded too little and left the main por
tion of the liner exposed to maximum wear. The liners were replaced with simple flat
profile grids, with 100mm lifter bars. This increased liner life sevenfold, equating to savings
in liner costs of more than R300,000 Rands and downtime of 100 hours per annum. At
Kloof gold mine, liner monitoring (Powell 1991a) indicated that the shell backing plate
thickness could be reduced to match one liner plate to two sets of lifter bars, and this
yielded a 5% improvement in mill throughput. At Rustenburg Platinum Amandelbult
Merensky section, closely spaced rows of rectangular lifters in the primary SAG mills
were resulting in a loss of mill throughput. The correct lifter face angle was calculated,
and the spacing of the lifters was corrected by installing alternating high and low lifters;
this design is still in use since installation in 1993. Lifter bars with a low 50˚ face angle
were installed over flat grid liners in a primary runofmine mill operating at 90% of crit
ical speed at Lindum Reefs gold mine (Powell 1994). This increased liner life more than
fivefold while maintaining mill production. At Freeport Indonesia, the new 34ft SAG
mill suffered severe liner wear and obvious impact on the liners. Morrison (R.D. Morri
son, personal communication, 2001) examined the problem by applying the MillTraj tra
jectory equations were used to verify the source of the problem—overly high lifters with
almost no relief angle—and to assess a range of alternative configurations. In the short
term, Freeport staff overcame the problem by fitting an available set of low–low lifters
(Coleman and Veloo 1996).
In notable cases (generally unpublished), liner life and mill performance have been
dramatically improved once inappropriate liner design has been corrected.
UTI LI SI NG OUTER CHARGE TRAJ ECTORI ES TO DESI GN LI NER PROFI LES
A detailed study of the trajectories of the outermost layer of charge in a mill was con
ducted and used to test a mathematical model of the trajectories relative to lifter bar
height and angle (Powell 1991b). Based on this validated theory, the MillTraj software
program (supplied by Liner Design Services) was developed for use in conducting design
simulations (Powell 2000). The program predicts the trajectory of the outermost layer of
charge, which forms the envelope within which the rest of the charge cascades or cata
racts downwards. The primary design criterion is to ensure that the media impacts on
the toe of the charge rather than on the mill lining. As discussed earlier, direct impacts
on the liner are wasted energy and cause accelerated liner wear. By ensuring that the
grinding media lands on the toe of the charge, the drop height and energy transfer can
be maximised, which is ideal for a SAG mill.
SELECTION AND DESIGN OF MILL LINERS 351
MillTraj software predicts the likely position of the toe of the charge, even though it
is not possible to predict this with great accuracy, as it is heavily dependent upon the
type of ore, slurry percent solids, ball filling, and other bulk media properties. Therefore,
a 10˚ range is given that forms a safe area to impact upon. The position of the toe is most
strongly dependent upon the mill filling, so selection of a suitable liner design must be
according to the operating filling range of the mill.
The other important design criterion for mill liners is the spacing to height (S/H)
ratio of the lifter bars. This is to ensure sufficient wear resistance, while at the same time
not allow packing of the charge into a dead area between lifters. This is applicable to
AG/SAG mills and primary ball mills; it is less important for secondary ball mills and
does not apply to wave linings. The S/H ratio is strongly dependent upon mill speed.
Skega developed an empirical formula for this ratio (called A/B), which is still widely
used (Moller and Brough 1989). The relationship of mill capacity and power draw to liner
spacing is illustrated in Figure 20. The S/H ratio varies as the liner wears, so it should start
with a low value and finish with a high value when the liner is replaced; a range of ideal
S/H ±1 is considered reasonable. If a suitable S/H ratio cannot be achieved with the
number of rows of liners, it is sometimes necessary to resort to a high–low liner profile,
so as to achieve a quasi wider lifter bar spacing.
Utilising Charge Trajectory Predictions
Presenting the results in graphical form best illustrates the influence of liner design. The
trajectories are shown as a stream of ball positions at equal time intervals. The trajecto
ries for a range of different liner designs can be simulated and shown as separate ball
paths on one plot for direct comparison. The calculated toe position is used as a refer
ence from which to select a suitable liner design.
A
B
Relative
Energy
Consumption
Relative
Capacity
Capacity
kWh
0 1 2 3 4 5 6
Valid for 75% of Nc Range Proportion A/B
FIGURE 20 The Skega A/B ratio related to mill capacity and throughput
352 ADVANCES IN COMMINUTION MILL DESIGN
Lifter Bar Face Angle. The lifter bar face angle is a crucial aspect of liner design
and, if neglected, can lead to premature liner failure and loss of mill throughput. The
angle is quoted from the base of the lifter, so a rectangular lifter has a face angle of 90˚.
Figure 21 shows how trajectories vary with lifter face angle for an 8mdiameter mill,
with 30% charge filling, running at 75% of critical speed. Each trajectory is for a progres
sively lower angle from 90˚ (rectangular) down to 60˚. The impact point is sensitive to
the lifter angle, indicating that this is an important liner design criterion. The position of
the toe is shown, and trajectories that impact above this are considered unsuitable. For
this application the ideal angle is 65˚, shallower than may otherwise be selected.
Lifter Bar Height. The ideal S/H ratio at 75% of the critical speed is 4.0, giving an
ideal lifter height of 95 mm. For a ball size of 125 mm, the lifter should not be allowed to
wear <60 mm, otherwise slip and accelerated wear begin to occur. If a 170mm lifter is
chosen, the midpoint of the lifter is about 115 mm. The lifter has an increasing rate of
wear as it is worn down, so the halflife is likely to be >120 mm. The 95mm height will
be reached after three fourths of the liner life. So for the first half of the liner life, the lift
ers will be unsuitably high for this speed and spacing. This leads to packing between the
lifters in the region from toe to shoulder of the charge, the degree of which is dependent
upon the stickiness of the ore, resulting in reduced mill throughput. To span the ideal
spacing range, the new height of the lifter should be no more than 125 mm, so alternat
ing rows of high and low lifters would be recommended to ensure a lifter height that is
adequate for liner life.
High–Low Lifters. In the given example, the new lifters could be 170 mm, and
worn low lifters, 90 mm. When the lifters have worn to about 120 mm and 60 mm, the
spacing will be in the ideal range. When the high lifters are worn down to 90 mm, the low
ones will be down in the sub40mm range. If the lifters are removable, then just the
alternating rows of worn low lifters can be replaced with high lifters to repeat the cycle.
Although this sounds attractive, relining teams generally opt for the easier routine of
replacing all the lifters less frequently. For integral liners it is not feasible to replace alter
nate rows, so a larger scrap mass will result from removing the higher liners that still
have a lot of life left in them.
The authors contend that the use of high–low lifters is not ideal. Although necessary
for many existing applications, new mills should not be designed to require this high–
low system but should require even rows of lifters.
Mill Drilling. Mill drilling is generally determined by the mill manufacturer on
the following conventional formula
rows of drill holes = mill diameter (in ft) × 2 (EQ 3)
or the Skega modification of subtracting 2 from the conventional formula. These rela
tionships were developed more than 20 years ago on much smaller mills than those
being installed currently, and the Skega formula already indicated a nonlinear relation
ship. As mill size has increased, it has been found from practical experience that these
formulae are no longer applicable. A far wider spacing is required to ensure optimal mill
performance.
It is proposed that it is more rigorous to select mill bolthole spacing based on liner
requirements than upon a mill size formula. The following example illustrates this
approach. A lifter bar height and liner shell plate thickness are selected to yield a desired
liner life. These height calculations are based on manufacturers’ databases and plant
data, or (in the future) proven pilot and laboratory tests. For the current example, the
liner supplier may recommend that a lifter wear thickness of 100 mm is required to last a
year, with a backing plate wear of 50 mm being sufficient to last the same period of time.
SELECTION AND DESIGN OF MILL LINERS 353
For a minimum lifter height of 70 mm, this yields a new height of 170 mm. The base
width of the lifter may be recommended as 160 mm to support this height. The average
halflife height of the lifters may be about 125 mm, so this is used as the ideal height for
calculating lifter spacing. Allowing for the sloping face of the lifter increasing the effec
tive spacing, the ideal number of rows is 36. This is considerably fewer than the 52 rows
that would be recommended by the conventional formula.
The issue of hole spacing has received considerable attention, and a number of mills
have undergone dramatic changes in liner spacing during the past 5 years. For practical
purposes of installation without having to redrill the mills, these have gone to fitting two
lifters over three bolt holes (i.e., two thirds of the original spacing). Thus, at Cadia the
spacing in this 40ft mill was reduced from 78 rows to 52 (Rattray 2000) after simulation
work that included the use of the MillTraj software (Radziszewski and Valery 1999). This
has been operating successfully since December 1998, with two liner changes that have
included adjustments to the lifter bar angle, from 78˚ down to 60˚ (Hart et al. 2001).
These changes have eliminated packing and ball breakage, and have allowed an increase
in power draw and throughput. Similar circumstances hold for KCGM, Alcoa, and MIM
operations, with all the liners successfully in place for a few years. At the Alumbrera
mine, the spacing was increased to every second row, which resulted in a dramatic
reduction in liner life, showing that this was too far apart. They then moved to the 2/3
configuration, which has been in place since 2000. These instances illustrate that the
simple conventional formula is inadequate. These mills now have a number of lifter rows
of 1.3 times the diameter, in feet. These were by force of circumstance, and multipliers in
the range of 1.4 to 1.6 seem to be a more appropriate guideline.
The wider spacing between lifter bars has given rise to a new problem in some
installations. The backing plates have been suffering from peening and spreading, arising
from less protection from the lifters, which leaves them exposed to increased impacting.
Thus the materials of construction may have to advance with these design changes.
Mill Speed. The trajectory of the media is strongly influenced by the mill speed
(Figure 22). If this mill had a variablespeed drive, then operating at speeds of more than
75% of critical for the given load of 30% filling would result in direct impacts on the mill
FIGURE 21 Trajectories for different angled lifter bars
354 ADVANCES IN COMMINUTION MILL DESIGN
liner. The liner profile can be selected to suit the mill speed, or provide the best compro
mise for a variablespeed mill.
Final Liner Selection. The final liner profile and height selection can be based
around the desired operating window of the mill. Figure 23 shows how this was used for
a gold mine SAG mill. The lines give the safe upper operating mill speed as a function of
lifter bar face angle, for new (215 mm) and fully worn (62 mm) lifters, and for the upper
and lower range of mill filling (25% to 30%). Figure 23 shows how the mill operating
speed can be increased from 70% up to 80% of critical as the lifter angle wears down
from 72˚ to 55˚.
Mill Control. Understanding the charge trajectories in a mill can provide control
guidelines to mill operators, to keep the mill in the correct operating regime and, possi
bly, more importantly, out of the undesirable regimes. A control window was set up for
an openpit gold mine (Figure 24). A control rule is especially useful in preventing liner
damage in the early stages of mining if soft oxidised ore is received, which results in low
loading conditions. A control rule can ensure that the mill speed is lowered, feed rate
pushed up, and if necessary the mill stopped to protect the liners. For new high and
aggressive liners, it was determined that the maximum speed was 75% of critical for the
maximum mill filling. Only when the liners were worn in could the speed be increased,
while remaining in a safe operating window. Such a control strategy if implemented at
startup of a SAG mill, can prevent the severe liner pounding that the mill liners tend to
receive with subsequent liner failures, and even mill damage and major shutdown and
repair periods. As absolute load is a function of ball filling for a given mill, filling the
derived curves is subject to recalibration, which should be a standard function of mill
maintenance.
65%
70%
75%
80%
90%
SAG Mill
8m Diameter
46 Rows
30% Filling
125mm Balls
FIGURE 22 Influence of mill speed on ball trajectory
SELECTION AND DESIGN OF MILL LINERS 355
Summary of Optimising Liner Design
Liners can be tuned to suit the application of each particular mill. A primary input to this
is the profile of the liners, which can be selected to provide a suitable grinding action
throughout the life of the liner and to enhance liner life. One does not have to select
either a good liner life or a high mill throughput; the two can be optimised simulta
neously with suitable liner selection. The direct influence of liner design on mill effi
ciency was studied in a 1.8mdiameter pilot mill (Powell and Vermeulen 1994), in
which a difference of up to 10% in the rate of production of fines was measured. Design
guidelines are summarised as follows:
65
70
75
80
85
50 55 60 65 70 75
Lifter Angle
M
i
l
l
S
p
e
e
d
,
%
C
r
i
t
i
c
a
l
62 mm, 30% Fill
62 mm, 25% Fill
215 mm, 30% Fill
215 mm, 25% Fill
FIGURE 23 Safe operating window of a mill as a function of speed, lifter angle, lifter height, and
mill filling
76
74
73
72
71
70
69
68
67
75
200 220 240 260 280 300
Line calibrated at 2% ball charge.
5,385 kW, 267 t, 30.3% filling
update calibration if ball load changes
indicated by power, load cell, and
filling correlation.
72
˚
Lifter
Safe
Maximum
M
i
l
l
S
p
e
e
d
,
%
C
r
i
t
i
c
a
l
Load Cell, t
FIGURE 24 Safe operating range as a function of mill speed and filling for an example mill with
new high lifters with an aggressive face angle
356 ADVANCES IN COMMINUTION MILL DESIGN
1. Maximise media drop height for primary AG and SAG to maximise impact grinding.
2. Ensure a cascading action for regrind mills.
3. Avoid impact on the mill shell by ensuring that the balls and rocks land on the
toe of the grinding media.
4. Maximise liner life by providing sufficient protection with lifter bars or an inte
gral liner profile.
5. Prevent ball breakage by promoting the correct cascading action and avoiding
impacts directly onto the shell.
6. Maximise mill throughput with the correct spacing and height of lifter bars, an
essential factor in liner design.
7. Balance liner life and mill throughput by maximising the protective profile of the
lining while retaining the correct charge motion of the grinding media.
FULL CHARGE TRAJ ECTORI ES—DEM MODELI NG
The Discrete Element Method (DEM) is a numerical tool for modeling the behaviour of
discontinuous and particulate systems. The recent advent of affordable highspeed desk
top computers has made the simulation of complex systems feasible, and hence the
increased interest shown by the mining industry to simulate various applications. The
general DEM methodology and its variants are well established and are described in
review articles (Campbell 1990; Barker 1994; Walton 1994), and its uses in mineral pro
cessing are described by a number of authors (Cleary 1998b; Inoue and Okaya 1996;
Mishra and Rajamani 1994b; Powell, McBride, and Govender 2003). The behaviour of
the rock and ball charge within a tumbling mill is of express interest, as a fundamental
understanding of these systems is hard to gain purely from operational experience and
smallscale laboratory experiments. An important aspect is prediction of equipment wear
and its effect on process performance. In this section we describe the process of predic
tion of liner wear using the DEM.
Of particular interest for the holistic design of mills is the effect of lifter wear on the
performance of a mill and the rate at which this wear occurs. Lifter wear results in the
charge motion changing over time. The lifter design needs to take the wear factor into
account to ensure that the worn lifter profile produces a reasonable charge motion for
the purpose of grinding while still minimising damage to the shell over the lifetime of the
lifter. The effect of wear on charge motion can be directly observed using DEM simula
tions. Consider the series of DEM simulations of a laboratory mill shown in Figure 25;
the first three images show the effect of increasing the mill speed on the charge motion.
The change in ball trajectories, for a fixed mill speed, from the third image to the fourth
image shows the dramatic effect of varying the face angle of the lifter on the charge
motion.
A 0.5m slice of a fullsize industrial mill is simulated next to demonstrate the capa
bility of the DEM in contributing to greatly improved liner design. An example of a fullmill
simulation has been published by Cleary (2004). The mill modeled here is a traditional
36ft SAG mill running at 78% of critical speed with 72 rows of symmetric, closepacked,
steep faceangle lifters (7˚ from the face), loaded to 30% by volume with a 10% ball
charge. Representative SAG mill ball and rock size distributions are used—modeled as
spheres—and total 185,000 particles of +25 mm.
*
* The coefficient of restitution used 0.3 for rock–rock collisions, 0.5 for rock–steel collisions, and 0.8 for
steel–steel collisions. The friction coefficient used was 0.5 for all materials. Using a standard spring
stiffness of 10
6
gives a timestep of 3.4 × 10
–5
s.
SELECTION AND DESIGN OF MILL LINERS 357
Charge Motion
Figure 26 shows two snapshots of the 36ft SAG mill, with the particles shaded by diam
eter and velocity. The particles near the mill shell rotate with the liner rotation and at
moderate speed from the toe position around and up to the shoulder position. The parti
cles between the lifters are thrown in a high dilute cataracting stream, which impacts on
the liner at around the three o’clock position. The cataracting particles accelerate to
more than 13 m/s. The impact region of the cataracting stream is well above the toe,
even though the mill speed is only 78% of critical. This occurs because the lifters have
steep face angles. The majority of the charge avalanches down the steep free surface.
The effects of radial segregation are clear with a concentration of fine particles against
the mill shell and in the upper part of the cataracting stream. The bulk of the lower part
of the cataracting stream consists of larger rocks and balls. Segregation happens to be
useful in this scenario, as the balls in the cataracting stream are at least on the shallower
trajectories and lead to less damaging impacts with the liner. It should be noted that mill
speed is a strong driver in determining the type and severity of segregation.
Face Angle = 45
˚
Speed = 50% Critical
Face Angle = 45
˚
Speed = 70% Critical
Face Angle = 45
˚
Speed = 90% Critical
Face Angle = 90
˚
Speed = 90% Critical
Rendered
DEM
Snapshot
FIGURE 25 Example of DEM charge motion predictions
(a) Particle Sizes (dark = fines, light = coarse) (b) Particle Velocity (dark = fast, light = slow)
FIGURE 26 Slice model of a 36ft SAG mill
358 ADVANCES IN COMMINUTION MILL DESIGN
Liner Stresses
Stress and wear data are collected on a highresolution triangular mesh that covers the
mill liner. The average element edge length is 20 mm, which is half the size of the small
est particles in the simulation, giving good spatial information about the stress and wear
distributions. The contribution of every particle collision with the mill liner is accumu
lated in the triangular elements in which the collisions occur. These raw data are then
aggregated across the depth of the mill slice and spatially smoothed to remove noise on
the scale of the particle size. Figure 27 shows a section of the liner shaded by the stress
components. These distributions are easier to analyse when presented as a line plot fol
lowing the surface of the lifter/liner starting from the bottom of the front face, moving
up across the top surface and down the back of the lifter, and then finally across the liner
plate. Figure 28 shows the normal and shear stress distributions along the liner for the
SAG mill simulated in Figure 27. In the plot, each section of the lifter/liner stress is sepa
rated by a vertical line.
The normal and shear stresses have very similar spatial distributions, but the normal
stress is around three times higher than the shear stress. The stress is zero at the base of
the lifter (because no particles can make contact with this part of the lifter) and rises
steadily along the front of the lifter to a peak occurring near the top corner. This reflects
the force transmitted to the front of the lifter as it pushes into and then lifts up the
charge. The highest stresses for this new liner actually occur exactly at the corners, but
these are not shown in Figure 28 because the magnitudes are much higher and any
amount of rounding of the corner by wear will substantially reduce these peaks. The
stress along the top of the lifter is relatively constant and is about half the level of that at
the top of the face. The stress on the back of the lifter is low, at about half the level of the
top of the lifter. This reflects the fact that only a small proportion of the charge is sup
ported by the backs of the lifters. The stress on the liner plate is higher and increases as
the front of the next lifter is approached. Note that there are small, welldefined peaks
near the front and back of each lifter. These occur 18 mm from the corners and corre
spond closely to the average radius of the particle when trapped against the lifter. The
ability to capture such features demonstrates the spatial accuracy possible using DEM
simulation.
Liner Wear Distributions
To predict the impact damage on the liner, two measures were used (Cleary 1998a, b).
The first was the energy dissipated in the normal direction during collisions between the
particles and the liner. The second was a measure of excess kinetic energy of impact.
Lowspeed collisions (<0.1 m/s), which are large in number but of limited importance
for damage, make no contribution to the damage estimateʊbut highspeed collisions do
much more damage because of the quadratic dependence on speed. The impact damage
measure is shown on the surface of the liner in Figure 29a, and as a line plot in Figure 30
for both impact measures. The distributions are quite different than those of the normal
stress distributions. In particular, the wear is substantially higher across the entire top of
the lifter, with peaks near the corners. There is little impact damage on either front or
back faces because they are protected from the cataracting stream by the steep face
angles and close lifter spacing. There is some moderately higher wear on the upper part
of the leading face produced when the lifters crash into the charge in the toe region. The
wear on the liner plate is peaked in the middle in the plate. This damage is produced by
the penetration of the cataracting stream (and particularly balls) between the lifters
where they impact the middle of the liner, which is clearly more exposed to impact and
so has the highest predicted impact wear rate.
SELECTION AND DESIGN OF MILL LINERS 359
The abrasive wear is also estimated using two measures (Cleary 1998a, b). The first
is the shear work, which is the energy dissipated by sliding (tangential) interactions
between particles and the liner. The second uses the kinetic energy of each collision with
the inclusion of a strong angular dependence. This takes into account the fact that colli
sions at around 22˚ produce significantly more scouring/abrasion damage than particles
sliding directly along the boundary or impacting in the normal direction. Figure 29b
Normal Stress, N/m
2
50,000.0
40,000.0
30,000.0
20,000.0
10,000.0
Shear Stress, N/m
2
5,000.0
4,000.0
3,000.0
2,000.0
1,000.0
(a) Normal Stress (b) Shear Stress
FIGURE 27 Stress distributions on the SAG liner (dark = high magnitudes; light = low)
(a) Normal Stress, N/m
2
0 200 400 600 800
60,000
50,000
40,000
30,000
20,000
10,000
0
Distance along Surface, mm
Distance along Surface, mm
(b) Shear Stress, N/m
2
0 200 400 600 800
20,000
15,000
10,000
5,000
0
N
/
m
2
N
/
m
2
FIGURE 28 Normal and shear stresses along the lifter and liner plate. The distance is measured
from the base of the lifter front. Each region (lifter front, top, and rear; and finally the liner plate)
is separated by a vertical line.
360 ADVANCES IN COMMINUTION MILL DESIGN
shows the abrasion damage prediction on the surface of the liner and Figure 31 shows
both the abrasive wear distributions as line plots. These are, again, both quite different
from the stress and impact damage distributions shown previously. The highest abrasive
wear occurs on the front face of the lifter with the wear rate increasing with height up
the lifter. This wear will lead to steadily increasing lifter face angle (as one would expect).
There is also significant abrasive erosion from the top surface of the lifter, which one
would expect to lead to steadily decreasing lifter height. The abrasion on the back of the
Impact Damage
2,000.0
1,500.0
1,000.0
500.0
0.0
Abrasion
600.0
450.0
300.0
150.0
0.0
(a) Impact Damage (b) Abrasion Damage
FIGURE 29 Wear distributions on the SAG liner (dark = high magnitudes; light = low)
(a) Normal Work, J
0 200 400 600 800
1,500
1,000
500
0
Distance along Surface, mm
Distance along Surface, mm
(b) Impact Damage
0 200 400 600 800
400
300
200
100
0
J
o
u
l
e
s
J
o
u
l
e
s
FIGURE 30 Impact wear rate along the lifter and liner plate as given by the normal work rate and
an impact damage measure
SELECTION AND DESIGN OF MILL LINERS 361
lifter and the liner plate predicted by the shear work is relatively even and is approximately
one third of the magnitude on the lifter top. The second abrasion measure gives similar
predictions to the shear work for the front and top faces but predicts much lower wear
on the liner and the rear face. The lower prediction arises from the much lower weight
ing given to the many lowspeed contacts sliding directly along these surfaces. They dis
sipate a reasonable amount of energy, but this is likely to be an overestimate, as shown
by the second measure. The second measure suggests that the wear will decrease with
height down the back face, which is reasonable because most oblique impacts will occur
during the filling of the space between lifters as they pass through the toe region. There
is a peak in the abrasion in the middle of the liner for similar reasons.
The rate of normal work is around double the rate of shear work. This might cause one
to conclude that the dominant erosion mechanism is from impact rather than abrasion.
The actual erosion rates, however, also must be dependent upon the material properties
of the liner and its resistance to impact and abrasion damage. Highquality steel might be
expected to be very resistant to impact and erode predominantly by abrasion. The Cleary
CSIRO DEM code currently has the capability to evolve the shape of the liner in accor
dance with the wear rates predicted. However, it is not clear which of these wear rates or
which combinations should be used in order to obtain the best quantitative predictions
of the wear behaviour. Ultimately, it will be a combination of an impact and an abrasion
measure weighted by the resistance of the liner material to each damage mechanism.
Effect of Liner Height on Abrasive Wear
Figure 32 shows the rate of abrasive wear (using the shear energy absorption) on the
liner of a 36ft SAG mill as the lifter height is decreased in 50mm increments. Initially
0 200 400 600 800
800
400
600
200
0
0 200 400 600 800
140
60
80
100
120
40
20
0
J (a) Shear Work, J
Distance along Surface, mm
Distance along Surface, mm
(b) Abrasion
FIGURE 31 Abrasive wear rate along the lifter and liner plate as given by the shear work rate and
an angulardependent abrasion measure
362 ADVANCES IN COMMINUTION MILL DESIGN
(for the original 200mm lifter shown in Figure 32a), the wear is high across the top sur
face of the lifter with the peak abrasion occurring on the outer parts of the top surface.
There is reasonable abrasion on the front face, which decreases with distance down the
front face. There is little sign of abrasion on the liner plate, as it is protected by the lifter.
This distribution is consistent with the line plot shown in Figure 31.
When the lifter height is decreased to 150 mm (Figure 32b) (either as a design
change or as a crude representation of the liner wear), there is a significant reduction in
the abrasive wear rate on its top surface. The wear is now more concentrated on the
front half of the top surface, and the peak abrasion rate is reduced by approximately
25%. There is also a marked reduction in the rate of wear on the front face. The distribu
tion along the front face, though, remains similar with decreasing magnitude as the liner
plate is approached. There is little change to the wear on the liner plate. This indicates
that the wear rate of the lifters are higher for new 200mm lifters and decrease once
their height has been reduced.
When the lifter height is decreased to 100 mm (Figure 32c), there is little change to
the abrasive wear pattern or the magnitudes compared to the 150mm case. The wear on
the top surface is still weighted towards the front corner of the lifter, and the magnitude
is similar or perhaps slightly lower than for the 150mm lifter. There is now a small
amount of wear observed on the liner plate, with the higher wear concentrated on the
right side, just in front of the next lifter. The similarity of the wear for the 100 and 150mm
Shear Power
1,500.0
1,125.0
750.0
375.0
0.0
(a) 200 mm
Shear Power
1,500.0
1,125.0
750.0
375.0
0.0
(b) 150 mm
Shear Power
1,500.0
1,125.0
750.0
375.0
0.0
(c) 100 mm
Shear Power
1,500.0
1,125.0
750.0
375.0
0.0
(d) 50 mm
FIGURE 32 Rate of shear energy (abrasive wear) on the liner as the lifter height is decreased in
50mm increments
SELECTION AND DESIGN OF MILL LINERS 363
cases indicates that the wear behaviour of a liner can be fairly constant throughout the
middle parts of the lifter life span.
If the lifter height is decreased further to 50 mm (Figure 32d), then there is a sharp
change in the abrasive wear patterns and the rate of wear. The abrasion remains highest
on the top surface of the lifter and is strongest on the front half of the top surface. The
magnitude of the peak wear has risen back to the original level observed for the 200mm
liner. The wear on the liner plate has also increased sharply, with the peak values found
immediately before the next advancing lifter at around 40% of the overall maximum
wear on the top surface. This means that the 50mm lifter experiences significantly
higher wear than the 100 and 150mm lifters. This can be understood geometrically
because the median size of the rocks in the mill is 50 mm, meaning that half the rocks
are now taller than the lifter and the lifter is no longer able to properly lock into and lift
the charge. The charge now slides much more readily over the liner, leading to much
enhanced abrasive wear on the front face and the front half of the top of the lifter. This
behaviour of accelerating wear near the end of the lifter life span is often reported anec
dotally and is demonstrated in Figure 19. The effect is both reproducible and explainable
using DEM simulation.
Summary of DEM Potential
The DEM can be used to predict the charge behaviour and performance of the mill over
the life span of the lifter. This allows the liner profile and material to be optimised in a
more holistic manner by taking into consideration mill performance and liner cost as an
integrated objective function. When linked with a meaningful weartesting technique
that can reproduce the wear modes found in the mill, the DEM can be used to predict
liner wear and profile evolution. This can be used for fasttracking of liner selection,
assessing new designs and materials, and deriving an optimal balance of liner life and
mill performance. Additionally, the time between relining can be calculated.
MI LL LI NER MANAGEMENT
The lining in a mill serves a dual purpose. It not only protects the shell of the mill from
impact and abrasiverelated wear, but it also transfers the energy to the charge where it
is required for breakage. The performance and the costeffectiveness of a mill are thus
largely dependant on both the design and care of the mill lining.
Given that the lining normally constitutes a large portion of the operating cost of the
mill, it is not commonplace to keep a spare set of liners on site, as these are seen as a
form of “dead capital.” The lead time on procuring a set of liners is also normally in the
order of months. Damaging a set of liners can thus be a very expensive exercise when the
downtime that could result is taken into consideration.
Despite these risks, the importance of proper liner management is often underesti
mated. The lining in a mill can be damaged severely in the time period between even
weekly liner inspections. Most operations only conduct liner inspections on a biweekly to
monthly cycle. It is thus very important that mill operators understand and manage risks
by conducting proper mill liner inspections and by placing sufficient emphasis on the
safe operation of the mill.
Liner Wear Measurement
As with most processes, optimisation does not come without measurement. By monitor
ing the change in the liner profile with time, valuable information regarding the wear
rates of the different facets of the lining allow for the refinement of the liner design.
Given that the lining in a SAG mill could last for 12 to 18 months, the optimisation of the
364 ADVANCES IN COMMINUTION MILL DESIGN
liner profile for a particular mill could take a number of years. A dedicated and ongoing
measurement effort is thus required in order to achieve success in this regard.
The challenge of obtaining a measure of the profile of a liner is easily overcome with
the use of the simple yet highly effective mechanical gauge (Figure 33). By measuring
the length of the rods that are displaced when an imprint of the liner profile is taken, it is
possible to present the liner wear data graphically. The change to the profile of a lifter
bar with time, in a large SAG mill, is shown in Figure 34.
An electronic gauge has been developed at the University of Cape Town, capable of
measuring six positions along three liners in less than 10 minutes. The data are automat
ically logged and transferred to a liner profile and wearmonitoring program. PERI Asso
ciates supplies an automated electronic gauge driven by a wormgear motor. Metso has a
full mill liner profile gauge, which is inserted down the centre of the mill and within
hours can collect a detailed profile along the entire length of the mill. Tools of this nature
can be used to collect accurate, reproducible, and reliable data on a regular basis, so as
to build up an accurate history of the wear profile of liners.
A critical measurement—in addition to the profile—is the liner thickness at each end
of the profile, which is used to calculate the absolute liner thickness along the profiles.
The measurement of the thickness of the liner plate can present considerable difficulties.
One of the authors has been on site where the mill reliners overpredicted the liner life by
months, based on incorrect liner thickness estimates. This mistake led to catastrophic
failure and extensive mill downtime. For rubber liners, the accepted technique is to ham
mer a nail through the liner and measure the length of the protruding section. This can
be done at the joint between steel liners, but it is slow; it is difficult to find a clear spot in
the joints, which are packed with steel slivers and slurry, through which the nail must
penetrate. It is all too easy to take incorrect readings, as it is not obvious whether the nail
has penetrated to the mill shell. Even careful measurement of internal mill diameter can
lead to inaccurate liner thickness calculations, as a small number is being calculated
from the difference between two large numbers (diameters inside the shell and inside
the liner). An ultrasonic thickness gauge can be used, but its effectiveness is dependent
upon the structure of the steel. The signal is seriously attenuated by any porosity, which
can result in plausible but incorrect readings, and an expensive highpenetration gauge
is required to penetrate 100 mm of steel. Based on the expense and unreliability of this
route, it has been deemed unsuitable. In response to this situation, a measurement indi
cator that is installed with the liners, and is accurate to 1 mm, is being developed at the
University of Cape Town.
It is best to collect data from a consistent set of liners, and in the same place each
time, so as to obtain accurate and consistent wear data. Selecting three rows of liners
and taking five to six profiles along each row should be sufficient.
Observing the change in the lifter profile over a period of time not only provides an
indication as to when to change the direction of rotation of the mill, in the case of a bidirec
tional mill, but also provides information as to the relative wear rates of the different
liner components. It can be seen in Figure 34, for example, that the base section of the
liner is wearing at a much lower rate than that of the lifter bar itself. It may therefore be
possible to design the next set of liners with a slightly thinner base section that would
effectively lead to an increase in mill diameter, which could possibly have a positive
effect on mill throughput. Also, by optimising the distribution of the steel across the pro
file of a liner, by making observations such as those mentioned previously, it is possible
to derive the maximum wear benefit out of a casting and to minimise scrap at the end of
the life of the lining. This can only be made possible through the implementation of a
liner wear measurement program.
SELECTION AND DESIGN OF MILL LINERS 365
Mill Liner Inspections
It is good practice to schedule mill liner inspections rather than to conduct them on an
adhoc basis. Other routine mill maintenance, such as cyclone maintenance for example,
also can be conducted during this period. When conducting mill liner inspections, it is
also important to look for the following:
Signs of cracks in the castings resulting from ball impacts
Raceways or abnormal wear patterns
Signs of damage around liner bolt holes
Edge spreading as a result of impact
Pegging of the discharge grates
FIGURE 33 Mill liner profile measurement
16012001
05042001
12042001
18042001
09052001
07062001
16062001
21062001
19072001
26072001
–2.0
–
2
.
0
0.0
2.0
4.0
6.0
8.0
10.0
12.0
14.0
16.0
18.0
20.0
22.0
24.0
26.0
28.0
30.0
32.0
0
.
0
2
.
0
4
.
0
6
.
0
8
.
0
1
0
.
0
1
2
.
0
1
4
.
0
1
6
.
0
1
8
.
0
2
0
.
0
2
2
.
0
2
4
.
0
2
6
.
0
2
8
.
0
3
0
.
0
3
2
.
0
3
4
.
0
3
6
.
0
3
8
.
0
4
0
.
0
4
2
.
0
4
4
.
0
4
6
.
0
4
8
.
0
5
0
.
0
c
e
n
t
i
m
e
t
e
r
s
centimeters
FIGURE 34 The changing profile of a lifter bar in a 24ftdiameter SAG mill
366 ADVANCES IN COMMINUTION MILL DESIGN
Presence of freshly broken steel balls
Smallest surviving ball diameter
Shape of the grinding media
Presence of excess steel scats
Loose discharge grates
Loose liner bolts
Cracks in the castings, dimpled impact marks, and peening of liner edges indicate
excessive impacting on the liner, as shown in Figure 12. Freshly broken balls are nor
mally indicative of severe ballonliner impact. If any of these conditions are observed, it
is imperative that the operation of the mill be addressed so as to prevent any further liner
damage. Loose liner bolts, normally identified by wet patches on the shell of the mill
while it is in operation, are also indicative of shell impacts.
Abnormal wear patterns in liners normally suggest abnormal mill operating condi
tions. The operation of a mill with too dilute a slurry charge, for example, can result in
severe slippage between the charge and the lining and lead to accelerated wear. An
example of this phenomenon is shown in Figure 35 for a wavetype liner design in a 6MW
ball mill. This mill had been in operation for only a number of weeks.
It is good practice during liner inspections to inch the mill periodically in order to
obtain a view of all of the liners in the mill. It is not necessary to inspect each liner with a
great deal of detail, but it is important to glance over each casting to identify any obvious
defects. When measuring liner wear profiles, it is important that the same liners are
measured each time that the mill is stopped.
LI NER DESI GN DETAI LI NG
There are many practical issues that should be taken into consideration when detailing
liner design. Seemingly small details in design can be the difference between durability
and reliability, and a disastrous liner. Some guidelines for consideration in liner detailing
are listed:
1. Understand clearly what the mill operator wants from the mill performance,
durability versus performance considerations, downtime intervals and extent,
variability in ore, mill filling, mill speed, ball loading, etc.
2. Discuss potential liner designs with the relining and maintenance crews. Their
input and buyin can make a significant difference to the performance of a liner.
3. The liner designs must be tailored to the materials liners from which they are to
be manufactured. For example, a high carbon chrome moly steel has different
liner design requirements than those of a chrome moly white iron.
4. Proper calculation of lifter wear height versus liner plate thickness is required.
This wear ratio needs to be either known or estimated so as to prevent premature
failure of either the lifter or the backing plate; failure to do so would leave a lot
of expensive scrap and a reduced liner life.
5. Liner handler capacity and clamping capability must be taken into accountʊthese
control maximum liner mass, size, and the positioning of handling lugs.
6. Consider designing the mill drill pattern to be flexible so as to allow the row spacing
to be varied over a limited range, so that the optimal row spacing can be selected.
7. Allow sufficient (if not more) pulp lifter depth in the liner assembly of AG/SAG mills.
Many of these mills have been found to have inadequate discharge capacity over
their full range of operation, especially as throughput is forced up over the years.
SELECTION AND DESIGN OF MILL LINERS 367
8. Design criteria should include ease of removal of worn liners. It should be
assumed that liners could flow together if not protected, and this should be taken
into account.
9. It is generally prudent to base designs and materials on known current working
trends. Although it is not necessary to be too conservative, any changes should
be properly assessed in light of current experience with operational liner designs
and materials.
Further insights into design detailing are given by Rattray (Rattray 2000).
COMMI SSI ONI NG
The commissioning period of a liner often presents abnormal operating conditions for a
liner, and this should be taken into account when designing the first set of liners for a new
mill, especially a SAG mill. It is often not appreciated by the mill operators that the mill
should be kept within a certain window of reasonably normal operation during the com
missioning period, so as to prevent damage to, and possible premature failure of, the liners.
The challenges facing the commissioning metallurgist often include softer than normal
runofmine ores, aggressive mill liner profiles, uncalibrated instrumentation/control
systems, and inexperienced operators. In addition, the time frame allowed to overcome
some of these issues is normally limited, owing to an eagerness/pressure to start produc
ing. The commissioning of SAG mills is thus not without risk, but with proper planning
and preparation, the risk can be managed.
Feed Preparation
The early stages of the life of an openpit mine are normally characterised by large vol
umes of soft overburden or oxide ore. This material is often grade bearing and cannot be
discarded. It is important that this is taken into consideration during the design phase of
the milling circuit. If it is not possible during the commissioning of the operation to
obtain material that is competent enough to establish a load in the SAG mill, it may be
necessary to bypass the SAG mill altogether and process this material in the ball mill
until such time that the ore competency increases.
AngloGold Ashanti’s Morila circuit was started up in this manner with the feed
diverted from the SAG mill feed conveyor head pulley to a temporary ball mill feed hop
per located some short distance away. Converting back to a twostage circuit required a
few chute changes and was completed in a matter of hours. The circuit was operated in
FIGURE 35 Raceways and accelerated “pocketing” resulting from the operation of a ball mill
with too dilute a slurry charge
368 ADVANCES IN COMMINUTION MILL DESIGN
this manner for a number of weeks until the competency of the ore increased sufficiently
in order to warrant semiautogenous milling. This approach is considered to be good
practice in terms of reducing the risk of SAG mill liner damage during commissioning.
Safe Operating Window
Identifying the safe operating window with respect to mill filling and speed is a vital part
of the commissioning of largediameter SAG mills. The same applies to the operation of
any SAG mill with a new liner profile.
It is often assumed that the newly installed liner is capable of catering to the entire
speed range of the mill. Experience has shown that often this is not the case, particularly
because it is difficult to design a liner profile to accommodate wide speed ranges. Most
SAG mills, on commissioning, need to be operated at reduced speeds and higher volu
metric fillings than those stipulated in the design specifications. From a charge motion
point of view, mill liners are at their most aggressive when newly installed. In most
cases, it is not possible to operate the mill within the upper speed region of the mill
speed range until such time that the lifters have been subjected to a certain amount of
wear. Trajectory studies are very useful in defining the safe operating window of an
AG/SAG mill with respect to mill filling and volumetric loading.
For example, one such study demonstrated that the speed of a particular 6mdiameter
SAG mill should initially be limited to the 65%–70% of critical speed (Nc) range and that
the mill filling should be in the range of 30% to 33%. It was predicted that the operation
of the mill at fillings of below 25%, even at the lowest speed setting, would result in
direct impact of the charge on the mill shell. This was converted into chord measurement
figures, so that the operating staff could relate the operating window directly to their
raw measured data (Figure 36).
Load Calibration
To control the load in a SAG mill to a known volumetric filling, it is necessary to calibrate
some measure of the load mass to that of volumetric filling. The majority of mills are
either mounted on load cells or have some measure of slipper pad hydraulic pressure
that can be related to the mass of the charge in the mill.
The volumetric filling in the mill is measured by “crashstopping” the mill, locking it
out, and measuring the width of the chord along the charge surface, or the height from
the charge surface to the “roof” of the mill. The height measurement is a more accurate
method of determining the mill filling, and it has been found that an industrial surveying
laser distance meter with a rugged, waterresistant casing and 100m range accurate to
1 mm, is eminently suited to this, allowing quick and accurate height measurements to be
taken in any size mill (Figure 37).
By observing the indicated loadcell reading, or bearing backpressure, it is possible
to determine a relationship between mill volumetric filling and mass or pressure. This
relationship is obviously dependant on the bulk density of the material in the mill. As the
steel loading in the mill fluctuates over time, it is necessary to calibrate the mass–volume
relationship fairly frequently by recording the indicated mass reading and measuring the
charge volume during routine mill stoppages.
The load calibration curves measured during the commissioning of a 30ft mill oper
ating without a steel load are shown in Figure 38. The shaded rectangular area indicates
the safe operating window for the operation of the mill at 68% of critical speed.
The trends shown in Figure 38a contain many data points. It is normally not neces
sary to obtain as many points to determine the mass–pressure versus volumetric filling
relationship. In this particular example, a great deal of signal noise was observed, and
hence many points were obtained in order to determine a better average. Figure 38b
SELECTION AND DESIGN OF MILL LINERS 369
shows the relationship for a nonlinear bearingpressure response, and Figure 38c shows
the linear relationship between a loadcell reading and mill filling. Once the mass–pressure
versus filling relationship has been obtained, limits should be placed on these to define
the safe operating window of the mill. It is important that mill operators gain an under
standing of this process, as they will be relied upon to conduct the necessary calibrations
in the future.
650
770
660
670
680
690
700
710
720
730
740
750
760
780
790
20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
Volumetric Filling, %
S
,
c
m
S
Chord Width
along Charge
Surface (S)
Liner Damage
Even at Lowest
Speed
Safe Operating
Speed 65% Nc
Safe
Operating
Range
65%–70% Nc
Safe
70%–75%
Nc
Load Running
in Trunnion
FIGURE 36 Example of defined safe operating window for a 6m SAG mill with new liners
Laser Tape Measure
Support Pole Resting on Charge
FIGURE 37 Measuring the mill filling with a laser tape measure
370 ADVANCES IN COMMINUTION MILL DESIGN
A set point for the mill mass is normally then chosen for the operation of the mill. It
is good practice to vary the feed rate in accordance with the indicated mill mass in order
to maintain this mass set point. This can be accomplished with the use of a properly
tuned proportional/integral/derivative (PID) controller, or a system if it is already oper
ational. Operator vigilance is still required to ensure that the mill load is maintained
within the safe operating window. It is good practice to define a lower mass limit and to
incorporate this into an alarm function in the control software. Some operators apply a
rule/interlock in the software that automatically stops the mill should the