Surveying Comminution Circuits and Ore Characterisation

JULIUS KRUTTSCHNITT MINERAL RESEARCH CENTRE
Isles Road, Indooroopilly Telephone: (07) 3365 5888

Qld 4068, Australia Fax: (07) 3365 5999
E-Mail: [email protected]
SURVEYING COMMINUTION CIRCUITS AND ORE CHARACTERISATION

T Kojovic and S Morrell
SUMMARY
The ability to assess the performance of comminution circuits depends largely on the quality and nature of information
collected from the circuit. This is particularly true of any modelling and simulation study.
A survey consists of collecting data and samples from the circuit over a particular operating period, which are
representative of the operation of the circuit during that period. The representative character of the survey is
important, as circuit analysis and simulation is based on the ability to build a model that is representative of a real
system. There are several issues which need to be addressed, including the nature of the data to be collected, sampling
points, the method of sampling, the equipment used, and the sample processing which follows.
This paper presents a general practical methodology for sampling and data collection in conventional crusher,
SAG/AG and/or ball mill-cyclone circuits (an example of this kind of circuit is shown in Figure 1). Additional
information is included on data mass balancing techniques, sample analysis (particularly coarse size analysis), the
selection of sample size, examples of the types of errors that occur in the sampling of pulp streams within a grinding
circuit, and a review on ore characterisation.
Figure 1: Schematic diagram of typical comminution circuit,

showing some alternative sub-circuits
1
Surveying Comminution Circuits and Ore Characterisation
INTRODUCTION
Many types of comminution machine and circuit are encountered in practice. In broad terms, however,
three main types of circuit are found:
• Crushing and screening - e.g. iron ore processing, coal processing, quarry processing, diamond
recovery, and preparation for grinding.
• Coarse grinding, using AG/SAG, rod and/or ball mills - e.g. metalliferous processing, as preparation
for beneficiation processes such as flotation or leaching; an example of this kind of circuit is shown in
Figure 1, including typical sampling points.
• Fine grinding, using ball, tower or stirred mills - e.g. regrind in metalliferous processing, or industrial
mineral processing.
Each of them has particular features which must be addressed in any sampling or data collection exercise;
sampling a crusher feed is very different in a practical sense from sampling a hydrocyclone overflow.
However, there are some guiding principles and techniques which apply to all such campaigns, and which
must be adhered to for best results.
SOME SAMPLING PRINCIPLES

The objective of all sampling is to obtain a representative sample. This is an ideal concept which is rarely
realised in practice. The problem is really a statistical one, since it depends on the nature and magnitude of
the errors which accumulate when collecting and processing a sample. Some of the errors and disturbances
which can contribute to overall error in determining some quantity such as a solids concentration or
particle size distribution are:
1. Plant dynamics. Processes are rarely in steady state, and sampling policy must be determined
accordingly. A „snapshot‟ sample may be appropriate for a single device with essentially no
dynamics, such as a crusher or hydrocyclone, but a circuit is usually sampled by accumulating a
number of incremental samples taken over 1-2 hours to „smooth out‟ disturbances in the process.
2. Sample cutter design.
3. Sub-sampling a primary sample.
4. Analytical errors, e.g. weighing, screens with worn or incorrect apertures, inadequate screening time,
incorrect calibration or selection of constants, etc. Such errors are more common than generally
admitted.
5. The propagation of error when calculating quantities.
6. The fundamental statistical uncertainty (error) involved in choosing a small, finite sample to represent
the properties of a large (effectively infinitely large) population.
The surveyor has some control over the first five of these, but essentially no control over the fundamental
error (FE) which is a statistical property of the particulate system being studied. The object of any
sampling and analysis exercise should therefore be to minimise the effects of items 1 - 5 so that their
contribution to overall error is small relative to FE. The size of sample is then chosen to achieve the
required confidence in the light of the prevailing FE.
Sampling statistics is a complex topic. Gy has developed a widely-used theory of particulate sampling,
which has been adapted and described by Pitard (1993), who also discusses many other aspects of sampling
such as cutter design.
2
Size of Sample for Size Analysis

How large a sample of solids from a process stream should be presented for size analysis? One answer is:
as large as possible, since this will maximise the reliability of the final result. In practice, however, the
selection of sample size is a pragmatic compromise between economics (the cost of collecting and
processing the sample) and the confidence needed in the answer.
Chapter 18 of Pitard‟s book deals specifically with sampling for size analysis. However, to give a simple
estimate of the order of sample size required to accommodate the FE, Barbery (1972) derived an expression
based on Gy‟s theory, which is easy to use:
f d 3m
M = 2P
(1)
where M = mass of sample required (g)

f = shape factor for material (0 < f < 1)
= density of material (g/cm3)
dm = mean size in size range of interest (cm)
P = expected proportion of material in size range of interest (to be measured)
= standard deviation of the number of particles in that size range.
(The variance 2 is essentially the FE referred to earlier).
These terms require some explanation. The shape factor of a single particle is
f = m/ d3. f = 0.1 for flat, plate-like particles and approaches 1 for spheroidal particles. For most natural
ores and coal, 0.3 < f < 0.7, and 0.6 is not a bad guess in many cases.
d 3m can be calculated as (Barbery, 1972):
3 3
3 d1 + d2
dm =
2 (2)
where d1 and d2 are the limiting sizes of the size range of interest. The size range of interest is that which
is likely to have the least number of particles in it, which is nearly always the coarsest size interval. This
ensures that the error on the proportions estimated for the other size intervals will always be less than this,
thus ensuring a conservative choice. However, it is pointless to choose an interval with practically no
material in it. A good rule of thumb is that the coarsest size interval should be chosen to give P 5%. is
determined from the precision of estimation and confidence required:
=
z (3)
where f = chosen precision (relative proportion)
z = normal ordinate at the chosen confidence level.
Table 1 gives values of z for different confidence levels.
Table 1: Normal ordinates (from the normal distribution)
Confidence Level (%) z

50 0.6745
80 1.2816
90 1.6449
95 1.9600
99 2.5758
99.9 3.2905
3
A confidence level of 90% (z = 1.64) is usually adequate. The definition of is interpreted as follows: if the
proportion of material expected to lie in the coarsest size range is 5% (P = 0.05), and we want to estimate
this to a precision of 10% relative with 90% confidence, then
10/100
= = 0.061, and P 5% 0.5%, with 90% confidence .
1.64
Propagation of Error in Calculated Quantities

Some quantities in a survey will be calculated rather than measured. A good example is the simple method
for estimating the solids flow through a mill which has significant water addition at the feed, as illustrated
in Figure 2.
M M
% s olidsd % s olidsf
Ball Mill
Figure 2: Calculation of ball mill throughput from the water addition at the feed
The mass flow is given by:
W
M (4)
100 100
%SOL d %SOL f
where M is the mass flow (tph), W is the water addition (m 3/h), and %SOLd and %SOLf are the mill
discharge and feed slurry densities respectively.
The definition of variance suggests that
2 2
2 M 2 M M 2
( M) ( %SOL d ) ( %SOLf ) 2 ( W )2
%SOL d %SOLf W
where signifies „maximum error in‟ and ( M)2 is the variance of M.
The %SOLd, %SOLf and W „errors‟ are usually taken to be the precision in reading an instrument (e.g.
the flow meter) or the maximum error in measuring the quantity (e.g. the percent solids density).
Two points are worth noting:
1. The calculated error can be quite large, even though its components might be regarded as „reasonable‟
in terms of normal practice.
2. Errors often propagate in unexpected or counter-intuitive ways.
A similar exercise should be conducted for all such calculated quantities, so that the surveyor can identify
those quantities which need to be measured with particular care.
4
Sample Cutters
Sampling coarse material on a conveyor belt is usually done by stopping the belt and removing an
appropriate length of material. Proprietary automatic belt and transfer point samplers are also available.
The usual sampling device used in grinding surveys is the manual sample cutter. Unfortunately in many
cases a Marcy can or bucket is used. This practice is not appropriate. It is worth investing some time in
appropriate cutter designs, for example, one complete survey of a SAG/BM circuit grinding circuit costs
around $40000, excluding production time lost.
A design of the type shown in Figure 3 is suitable for fine slurries. Ingenuity is often necessary due to
difficult local conditions. The major problem with using sample cutters is their bulk and weight. Sample
point accessibility in grinding circuits can be atrocious, particularly SAG/AG and ball mill discharge
samples (from underneath trommels). Another major problem is the sampling of large pulp flowrate
streams (>100 tph).
Figure 3: Recommended cutter design for cyclone underflow and overflow sampling
The cutter edges should be horizontal and parallel. For streams where the largest particle size d > 3mm, the
minimum width of the cutter should be at least 3d. For smaller particles (d < 3 mm), the width should not
be less than 10mm. The depth of the cutter should be large enough to prevent the sample splashing out or
overflowing.
The cutter should be passed through the total stream cross-section at a uniform speed. If this is not feasible
for large flowing streams, a number of successive cuts moving across the total stream cross-section is
recommended. Care should be taken that material does not fall into the cutter between cuts (eg. dust or
5
slurry). The cutter should be „washed‟ in the slurry stream and tapped to ensure that it is empty before
sample cuts are taken. Dilution with water would alter the percent solids obtained from the sample.
It is almost impossible to make screens or crushers operating in parallel behave in an identical manner.
Therefore it is much better to test one unit over a range of operating conditions.
SAMPLING PRACTICE
The objectives of the survey should be clearly established before planning and undertaking it. These may
include the establishment of the performance of a single unit (before or after optimisation), the audit of an
entire section or circuit, or the collection of data for model parameter estimation.
Note that in all sampling surveys, the proper safety procedures must be followed at all times. These may
include, where appropriate, the electrical isolation of equipment, and the use of safety harness when taking
awkward samples. For major sampling campaigns, a written procedure should be prepared for each
participant
The set point tonnage should be held as constant as possible during the sampling campaign. To achieve
this in grinding circuits, the grinding control loops may have to be de-activated. Once set up, the grinding
circuit illustrated in Figure 1 should run at reasonably steady state conditions providing feed sizing,
flowrate and ore hardness do not change significantly. Once a set of operating conditions has been chosen,
at least an hour (i.e. several residence times) should be allowed for steady state conditions to occur. It
should be noted that steady state condition is an ideal state and is never fully realised in practice; however
an indication of the degree of steady state can usually be obtained from the consistency of the cyclone feed
pump amperage draw and speed (for VS drives).
With the advent of DCS monitoring software, it is relatively straight forward to prepare and view a group
of relevant operating variables prior to commencing the survey. The same data, including the period of the
survey, are then stored on disk which can be easily accessed once the survey is complete.
The circuit should be sampled over a period of 1-2 hours (either 8 cuts every fifteen minutes, or 4 cuts
every thirty minute period). Multiple samplers (people) make the sampling more efficient, as more time is
made available for the cutting of the stream samples. It also allows the samples to be taken simultaneously
(or in a logical sequence appropriate to the residence times in each section), which is highly advantageous.
If this is not possible the order of samples should be varied. Ensure that sample containers are clearly
labelled; where multiple containers are used for the single sample, each container should be marked with
container number and total number of containers.
If a significant disturbance occurs during a survey, either terminate sampling if close to the end of
campaign, or otherwise abandon sampling and return another time. Back-up samples should if possible be
collected in case of mishaps in the lab during sample processing. As snapshot samples are the usual
method for sampling crushers and screens, it is important to stop all the necessary conveyors at the correct
time. To ensure that the samples taken from product conveyors are representative of the measured
operating conditions, the feed belt can be marked at the start of shutdown by dropping an appropriate
marker onto the belt or processing unit (e.g. flour, paper, etc.), and the product conveyors stopped when
the marker appears.
It is useful to know the ball charge filling of the mills after each sampling campaign. Therefore, after
collecting the slurry samples, the mills should be crash stopped to allow feed belt samples to be collected
and the mill charge level to be measured via physical inspection of mill contents. This key step in the
survey is discussed in more detail in the sections which follow.
Feed size analysis requires special consideration. It is important to obtain an accurate measurement of the
feed size distribution particularly at the coarse end. It is therefore necessary to collect a belt cut sample and
record the belt speed and length of belt over which this sample was collected. Feed samples should also be
taken for ore characterisation tests (see later).
6
The samples collected need to be processed on site to determine size distribution and percent solids. Where
possible, consideration should be given to collecting a separate sample solely for percent solids
determination.
The plant dry ore feed rate (and other belt flowrates where possible) should be determined from the
weightometer readings and feed sample moisture measurements. The weightometer calibration should be
checked using measured belt speeds and a suitable belt sample.
Readings of the fresh feed rate to the circuit, power draws, cyclone or screen feed rate and density, and
water additions to the circuit should be recorded periodically, from which averages over the survey period
can be determined. Cyclone pressures should be recorded manually. If possible power readings should be
cross-checked using local power meters or kWh meters (see below). A full list of data to be collected
follows.
Data to be Collected
Apart from the stream samples and circuit operating condition, the information required from a full
comminution circuit survey is listed in Table 2 below.
Note that all internal cyclone dimensions should be measured, not assumed. Normal wear, and maintenance
errors, can lead to significant discrepancies in cyclone dimensions.
For cone crushers, lead the crusher (to determine closed-side setting) in at least three positions around the
mantle immediately before or after the survey. Lead thickness should preferably be measured with a
micrometer or callipers, not a ruler. Take the average of the measured readings. For crushers with
automatically adjusted gaps, record gap reading and calibrate against measurements with leads.
Table 2: Data to be collected from particular unit operations
Crushers Hydrocyclones
Type Type
Closed side setting Number of cyclones
Eccentric throw Cyclone pressure (kPa)
No-Load power draw (kW) Feed flowrate and density
Operating power draw (kW) Diameter
Wear condition (liner age) Inlet diameter
Feedrate weightometer reading (tph) Cylinder length
Vortex finder diameter
Mills Spigot diameter
Type Cone angle
Number of mills
Diameter (inside liners) Vibrating Screens
Cylinder length (inside liners) Type
Cone length (inside liners) Number of decks
Speed (rpm, or % critical) Length x width
Discharge mechanism Aperture (length and width)
Grate aperture/layout dimensions Hole type (square, round, slot etc)
Pan lifter depth (grate mills only) % open area
Trommel dimensions and aperture Angle of inclination
Lifter height Deck material
Trunnion Diameter
Ball Filling (% mill volume) Conveyors
Total charge filling (% mill volume) Belt speed
Ore SG Ore type
No-Load power (kW) Historical data (e.g. WI, UCS, etc.)
Operating power draw (kW)
Feedrate weightometer reading (tph)
Bearing load and pressure (kPa)
7
Power Measurement
Experience shows that crusher or mill power draw is often one of the least reliable of the quantities
collected during a circuit survey. The surveyor should ensure that all power records are legitimate and
correct.
Power draw is measured in several ways. The best way is to use an in-line power transducer to measure
in-phase power, which does not therefore have to be corrected for power factor. Most modern plants will
incorporate such measurement, with a readout in the plant control room or motor control centre (MCC).
Even here, signal transmission or calibration problems can lead to an erroneous readout in the control
room, so the readings should be cross-checked wherever possible.
In older plants, one sometimes has to rely on an analogue ammeter (often with a wildly swinging needle).
In such a case, for AC 3-phase motors, total power delivered to the load, PT , is given by
PT VLI L 3 cos kW (5)
where VL = line voltage (rms value, V; varies with site and machine)
IL = line current (rms value, A; read from the ammeter)
= phase angle between the phase voltage and phase current.
Cos is the power factor for the circuit, which will vary with the motor‟s inductive load. For normal loads,
cos 0.8 - 0.95. For „no-load‟ power draw, the p.f. will be lower. The plant electrical engineer should be
consulted as to the correct power factor to use in the calculation. A plant (average) p.f. is not necessarily an
appropriate value, as the individual machine‟s load may differ significantly from others.
The golden rule is to cross-check all power readings and calculations. If possible take a portable power
meter to site and make independent readings. Where more than one plant reading is available, all should
be taken and cross-checked. Finally, the JKMRC crusher and mill power prediction procedures are now
good enough to provide an independent check of plant data (Napier-Munn et al, 1996).
Coarse Feed Sampling
As noted earlier, the size of sample required from statistical considerations for coarse can be impractical.
To tackle this problem, the JKMRC has adopted an approach based on a combination of coarse and fine
sampling. After crash-stopping the crusher or mill, and isolating the feed conveyor drive, the feed sample
can be taken. This should comprise a complete 2-5m belt plus at least 50 +75mm rocks. This typically is a
total sample of at least 500/800 kg. Figure 4 illustrates this process at the Pasminco Elura Mine, sampling
the AG mill ROM feed.
8
Figure 4: Example of 7.11 x 3.78 m AG Mill ROM fine feed sampling at Pasminco Elura Mine
The size distribution (mass per m in each coarse size fraction) is then combined with the feed size
distribution determined from the 2-5m belt cut of fines (mass per m in each -75mm size fraction) to give a
complete feed size distribution.
Mill Contents
Charge Distribution
The JKMRC SAG/AG mill model relies on the calculation of mill load mass and its size distribution. This
information is often made available from pilot plant studies. Should the ball and charge size distribution
for a given mill not be available, it is back calculated from the surveyed discharge size distribution and
estimate of load mass and filling. In this instance, the mill breakage rates may not be unique in the sense
that a range of breakage regimes may produce a similar discharge and mill filling, yet show quite a
difference in load size distribution. To provide the necessary data to test the model on full scale mills, the
JKMRC has undertaken the arduous task of sizing the complete mill contents on three separate occasions,
namely at Warrego, Alcoa Pinjarra and Red Dome.
An alternative procedure that may provide a reasonable estimate of the coarse end of the load size
distribution is to remove a few tonnes from the mill after inching the mill for several revolutions. The
inching is required to mix the contents of the mill and avoid segregation.
Mill Filling
The mill filling is the volume of charge in the mill which is the principal determinant of power draw. In the
case of ball mills the filling remains fairly steady over time, as it comprises mostly steel balls. In AG/SAG
mills, however, the feed ore contributes significant quantities of rock to the grinding media. As such,
variations in feed ore hardness and size distribution will affect the quantity of ore in the mill and hence the
power draw. Knowledge of the mill filling is therefore a key parameter in a grinding survey, as it quantifies
the response of a given mill to the prevailing feed ore characteristics.
A direct measurement of the load entails the crash-stopping of the AG/SAG mill under load whilst the mill
is running under steady state. Once access to the mill is gained, typically after the feed chute is removed via
the feed end trunnion, the width of the charge in three places should be taken, together with inside-liner
dimensions. From these measurements the load volume may be calculated using simple geometry, as
shown in Figure 5.
9
Figure 5: Example of mill load geometry parameters used to calculate volume
S 2r sin (6)
2
h r 1 cos (7)
2
h S2
r (8)
2 8h
h
Ac 3h 2 4S2 (9)
6S
Ac
V 100 (10)
A
2
A r (11)
where S = chord length (m)

r = mill radius (m, inside liners)
h = depth of charge (m)
= angle that chord subtends at mill centre (°)
Ac = cross-sectional area of mill charge (m2)
A = cross-sectional area of mill (m2)
V = estimate of mill filling (%)
The Allis Chalmers mill filling approximation formula, based on the measured distance between the mill
roof and charge surface, is as follows:
H
V 113 126 (12)
D
where H = vertical distance between the charge surface and mill roof, inside liners
D = mill diameter, inside liners
Morrell (1994) has shown that equation 12 gives relatively high errors for fillings less than 20%.
10
Estimation of the ball filling which is mixed with the ore charge is difficult, and a grind-out is therefore
required. The time required for the grind-out depends on the mill, ball load and feed ore hardness.
Typically less than 30 minutes is sufficient to achieve less than a few percent of ore charge. However, due
to the damage that this procedure can cause to the liners/lifters, this is not always possible or desirable.
Instead an alternative solution is to use the JKMRC power model and measured mill filling and power
draw under reduced (say 50%) load. By trial and error, it is possible to determine a ball load that would
give the measured power draw.
Controlled Mill Crash-Stop

Having decided to crash-stop the AG or SAG mill, it is worth noting that an indirect indication of the
nature of the charge in the mill can be inferred from the response of the mill just prior to the mill being
stopped. It has been found from experience that if the mill is kept running after simultaneously stopping all
the feed ore and water, scats and slurry recycle to the mill, the power and load in the mill should fall as it
grinds out. This condition would be reflected in a charge which has no visible slurry on the surface, as
shown in Figure 6.
However, should the mill not be able to effectively discharge the slurry from the mill, a pool will form at
the toe of the charge. If the mill is crash-stopped under steady-state load, slurry should be evident above
the rock charge. During a controlled crash-stop, however, once all the feed to the mill is stopped, the slurry
has a chance to drain, the charge becomes less fluid and starts to draw more power (ie. the torque required
to overcome the increased friction in lifting the charge). This condition typically occurs at high recycle loads
in closed-circuit AG/SAG mills, but may even be a problem for high throughput open circuit mills.
Figure 6: Example of a ‘dry’ AG mill load after a routine crash-stop under load
Evidence of such a condition is illustrated in Figure 7 which shows the mill power and load data extracted
from a circuit survey on the WMC LNO full closed-circuit AG mill. A power surge of approximately 600-
800kW with a corresponding load reduction of 20t occurred when the feed and recycle streams were
stopped.
In overflow mills, of course, excess slurry must be present to allow discharge through the trunnion. It is
therefore expected that the less able a grate discharge mechanism is in removing slurry from within the
mill, the lower will be its power draw. In the extreme case, where slurry builds up to the level of the feed
trunnion, the power of a grate discharge mill will equal that of an equivalent sized overflow mill.
11
5000 300
4000 260
pow er
Mill Power (kW)
Mill Load (t )
3000 220
load
2000 180
1000 140
0 100
0 10 20 30 40
Tim e (m in)
Figure 7: Example of mill power and load response during a

controlled crash-stop test at WMC LNO
SAMPLE ANALYSIS PROCEDURES
Sample Splitting
The most reliable, convenient and inexpensive method is riffling and rotary dividing. However, for large
samples the fractional shovelling approach is best.
Ore Specific Gravity

The specific gravity of the feed ore affects the charge density in the mill and volume flow calculations for
slurry streams in the circuit. In AG and SAG mills in particular, the ore specific gravity has a significant
effect on charge density and hence power draw. However where blends of ore are being treated, a harder
component may be present which has a different SG to that of the rest of the ore, in which case the SG of
the ore in the mill will be different to that of the feed ore. To determine if this may be an issue, as part of
the drop-weight test procedure, the JKMRC measures the SG of 100 individual particles of approximately
25mm size using a He gas pycnometer. The SG of a sub-sample of the -6.7mm fraction is also measured to
identify variations in coarse liberation.
Particle Sizing
The main analytical procedure in comminution circuit performance measurement is the determination of
the size distribution of the solids in the samples taken during the survey. Many techniques exist for particle
size analysis; Wills (1988) gives a good summary, and Allen (1996) covers the subject in depth.
The commonest method is sieving. However, there are several alternative methods of fine (sub-sieve) sizing
now available, and on-line size analysis instruments are also used in some concentrators. It is important to
ensure that
1. Whatever method is chosen is applied carefully and correctly.

2. The surveyor is aware of the differences in result which the different methods will produce.
DATA ANALYSIS
Initial Data Inspection
12
It is important to mass balance the data to obtain a rational picture across the circuit, and to obtain the best
estimate of each quantity through data redundancy.
The accuracy of survey data required depends on the objectives of the surveys. Comminution circuit
surveys are normally made to generate reliable data which may be assessed to determine how to make a
relative improvement in circuit performance, not necessarily an absolute outcome.
There are a number of simple data analysis procedures that can be used to evaluate your data before using
a computer mass balancing package such as JKMBal (included in JKSimMet). It is recommended that these
are applied to the survey data before attempting mass balancing, as there is much insight to be gained from
a quick review of the overall consistency. The most common mass balance is conducted around a classifier
or sump. Two simple manual analyses can be used to evaluate the data:
• a water balance around the cyclone, and

• a calculation of the mass split in each size fraction around the cyclone.
Both of these calculations are based on the two product formula.
Let solids mass flows be denoted by F, O and U, percent weight in each size fraction by f i, oi, and ui, and
percent solids in the feed, underflow and overflow by %SOL f, %SOLu and %SOLo respectively (Figure 8).
O, oi , %SOL o
F, fi , %SOL f
U, u i , %SOL u
Figure 8: Quantities around a hydrocyclone
The mass balance around the node for solids is given by
F=O+U (13)
and for each size fraction by
F. fi = O.oi + U.ui (14)
and for water by
100 100 100

1 F 1 O 1 U (15)
%SOLf %SOLo %SOL u
Then the mass solids split to underflow can be estimated from the size distribution and/or percent solids
data. Each size fraction yields an estimate of the total solids mass split, whereas the percent solids gives just
one estimate.
Hence, from each size fraction
13
U fi o i
total mass split to underflow, (16)
F u i oi
or, from the percent solids measurements

1 1
U %SOLf %SOL o
total mass split to underflow, 1 1
(17)
F
%SOL u %SOL o
Remember that in all cases the feed values must lie between those of the overflow and underflow.
Mass Balancing
Mass balancing is perhaps most useful as a method of data assessment. There is, however, a common
perception that mass balancing can fix bad data. This is simply not true. Mass balancing helps to identify
bad data in order to refine the measurement techniques so as to collect better data next time.
The general rule is that one should not be in too much of a hurry to mass balance. Make sure that data
have been carefully assessed first, and that some feeling has been developed for what the values of the
streams in the circuit should be.
A prerequisite for mass balancing is data redundancy (more data than are required to calculate a balance by
difference). Fortunately, collection of grinding circuit data usually involves a degree of data redundancy
which enables cross checks to be applied to the data, particularly the percent solids and sizing results. If it
is found that the water balance and sizing balance disagree, the samples(s) are unlikely to be representative,
or some error has occurred.
To mass balance the comminution circuit data, one reference stream is required to be assumed accurate (ie.
contain zero error). In grinding circuits this figure should be taken from the weightometer reading of the
new feed after adjustment for moisture content. It is usual to assume steady state conditions (ie. input tph =
outputtph across the circuit, or cyclone o/ftph = new feedtph). The concept of weighting the measured data
in JKMBal can be regarded as the inclusion of previous experience in the calculation procedure. The
mathematical equivalent of previous experience is the estimate of variance. As such, JKMBal minimises the
sum of weighted squared errors, that is
2 2
N L Xij xij L Ai ai
SSQ = (18)
j 1i 1 ij i 1 i
with respect to xij and ai and subject to prevailing material balance constraints. Here N is the number of
measurements, L the number of streams, X is a measurement (e.g. solids, assay, size), x is the adjusted
measurement, A is the measured flow and a the adjusted flow. ij and i are the weights or standard
deviations for the measurements and flows respectively.
Weights for Sizing Data

Choice of the weights is very important, particularly for sizing data. In practice, measured values of are
rarely available. In such cases the combination of sampling and measurement errors is most conveniently
considered as a combination of a fixed and a relative error when sufficient numbers of particles are present
(i.e. more than 10,000 in the size range of interest). The JKMRC approach uses a simple heuristic model as
follows:
• For more than 9% by weight in a 2 size fraction the error is assumed to be 1% absolute.
• For less than 9%, the error is taken as 0.1 plus the actual percentage divided by 10.
Table 3 shows that the resulting relative error model has about the same shape as 1/x. This shape is typical
of measured reproducibility curves.
14
Table 3: Errors assumed for sizing data
Weight % Retained Absolute s.d. Relative s.d.(%)

0.1 0.11 110
0.5 0.15 30
1.0 0.2 20
5.0 0.6 12
10.0 1.0 10
20.0 1.0 5
50.0 1.0 2
The absolute errors must be symmetric about 50%, i.e. similar for x% and 100-x%. These are referred to as
the „Whiten‟ errors or s.d., and are built into JKSimMet. They represent good sampling and sizing
procedures at slurry stream sizes finer than 80% -2mm.
Simple Mass Balance Solution for Multicomponent Data
For a single node like the cyclone discussed earlier, it is quite straight forward to derive a solution for the
flowrate based on size or assay data. Consider a cyclone as single node, as shown in Figure 9:
oi
(1- )
1
fi
ui
Figure 9: Mass balance with multi-size data on each stream (fi, oi, ui)
The objective of the mass balance is to find the best fit mass split to underflow, , and adjustments to the
measured percent weight retained values fi, oi, and ui, since the estimates of calculated from equation (16)
will rarely be consistent. The solution is to find the adjusted mass fractions in each size fraction, fi , oi and
u i such that the sum of squared errors between measured and adjusted masses in each fraction is
minimised, that is
2 2 2
Minimise ui ui oi (1 )o i fi fi (19)
i
with respect to , oi , u i , and fi , subject to the constraint that u i oi fi .
At the minimum the residual or mass balance error for each size is
i fi ui (1 )o i (20)
A number of variations of the solution to this problem exist. The general case of allowing the accuracy of
measurement of each variable to be considered is beyond the scope of this text and is best dealt with
software such as JKMBal. The solution offered here is based on the method of Lagrange multipliers, which
is used to simplify minimisation or maximisation problems which are subject to conditions or constraints.
The constraints are expressed in such a way that they are equal to zero. In this example the constraint
would be written as
ui oi fi 0 (21)
Then the sum of squares to be minimised is modified by adding each constraint equation multiplied by a
Lagrange multiplier (the +2 is for convenience), as follows:
15
2 2 2
ui ui ui (1 )o i fi fi + 2 i (u i oi fi ) (22)
i i
By differentiating the above expression with respect to the Lagrange multipliers i, and each of the
unknowns , oi , u i , and fi , and making appropriate substitutions, the solution for the best fit mass split to
underflow can be shown to be
(o i u i )(fi oi )
i (23)
(o i u i )2
i
This solution assumes that no data variances were measured, and as such are set to one. The usual way of
distributing the mass balance errors i between the measured values to give adjusted values, which are
numerically consistent at the calculated flow rates, is as follows:
fi fi i ui ui i oi oi i (24)
2 2 2
which assumes the measurement errors are proportional to mass flow rates in each stream.
In general the accuracy of measurement of the components will not be equal. For example, to incorporate
an estimate of accuracy for each size fraction it is necessary to weight the squared component error with
2
the appropriate variance i , in which case the solution becomes:
(o i u i )(fi oi )
2
i i (25)
(o i u i )2
2
i i
The Whiten weightings mentioned earlier are an alternative to the experimental variances if the latter are
not available.
ORE CHARACTERISATION
In the work related to comminution, the main objective has been to derive reliable ways of assessing how a
particular material breaks in a comminution machine such as a crusher or mill. Clearly some ores, coals or
quarry rocks are harder than others, and the appropriate machine, its size and its consumption of energy
will depend not only upon the duty (feedrate and desired product size) but also upon the „hardness‟ of the
material to be broken. Hardness itself is a difficult property to define, and its relationship with the machine
will not necessarily be straightforward: an autogenous mill treating a soft ore may have a low capacity
because of the lack of hard ore media to promote grinding; a cone crusher of a given set will crush a wide
range of ores to a similar product size, but will draw different powers in doing so.
Standard rock mechanics tests of strength, such as fracture toughness and uniaxial compressive strength,
are not usually seen as appropriate in comminution studies. They provide information on the stress
required to cause failure under particular modes of loading, in the form of a single hardness or strength
parameter. In comminution, it is also important to identify the product size distribution resulting from
applying a particular breakage mechanism to a given feed size, and the energy required to generate that
product size. In particular it is the energy-size reduction relationship which is the main focus of the
laboratory tests developed to assist in comminution equipment specification, and circuit design and
optimisation. Such tests include the Bond tests, batch grinding, and single particle testing. However
evidence is accumulating that some rock mechanics fracture tests can also be usefully interpreted in
comminution terms (Bearman et al, 1991).
16
This section briefly describes the commonly available laboratory breakage characterisation tests, with
particular respect to comminution energy and size reduction, and the use of these tests in predicting the
performance of industrial comminution machines. It commences with a brief review of the industry
standard tests such as the widely used Bond tests and concludes with a detailed description of the single
particle breakage tests developed at the JKMRC, and their application to the determination of material-
specific parameters in the comminution process models.
The Bond Test
In order to size crushing and grinding machinery, and also to specify motor sizes, a method is required for
determining the energy requirements of a comminution process. Rittinger‟s and Kick‟s 19th century
theories had severe practical limitations and were consequently of little use in real applications. It was not
until 1952 that Fred Bond published the approach that has continued to be the major comminution design
tool used by industry, particularly the equipment manufacturers, to the present day. It has also served as a
useful operating tool to evaluate and optimise crushing and grinding circuits.
According to Bond‟s „third theory‟ of comminution, the work input is proportional to the new crack tip
length produced in particle breakage, and equals the work represented by the product minus that
represented by the feed. He expressed this relationship as follows:
1 1
W = 10 WI (26)
P80 F80
and
P = T*W (27)
where W = work input (kWh/t)
WI = work index - a material-specific constant (kWh/t)
P80 = size at which 80% of the product passes (µm)
F80 = size at which 80% of the feed passes (µm)
T = throughput of new feed (t/h)
P = power draw (kW)
The work index expresses the resistance of the material to crushing and grinding. Numerically, it is the
kWh per tonne required to reduce the material from theoretically infinite feed size to 80% passing 100
microns.
Grinding power calculated using work indices obtained from Bond grindability tests applies reasonably
well in the range of conventional rod mill and ball mill grinding conditions, and can be corrected for other
conditions. Equation 26 is also useful for determining an „operating work index‟ for an existing
comminution operation, which can be used to compare feed ore types, or assess the crushing or grinding
performance in relative terms.
Although the Bond theory is very simple, it is not always successful when applied to real circuits because
deviations in transport and classification characteristics of the commercial circuit from those of the
'standard' circuit. In this regard it is a poor predictor of what happens in a real closed circuit when
throughput is increased, unless classifier performance is „adjusted‟ to cope with the new conditions. It is
not appropriate in systems where size distribution slopes change.
How to Use the Bond Work Index
For a given ore work index WI (kWh/t) and required new feed capacity T (tph), the required power P (kW)
to grind the feed F80 (µm) to product P80 (µm) is given by:
1 1
P = T x WI x (EF1 x EF2 x EF3 .... ) x (28)
P80 F80
where EF1 ... are Bond‟s efficiency factors.
17
Mill dimensions for an industrial unit are then chosen which are predicted to draw the required power,
using an appropriate mill size-power correlation.
Where these dimensions and/or proposed operating conditions deviate from Bond‟s standard,
correction factors recommended by Bond are applied to the work index.
Application of Bond Theory to AG/SAG Mills
The Bond calculation is not a good predictor of the grinding of large rocks, and specifically for AG/SAG
mill behaviour. As such standard Bond grindability tests cannot be used . It is necessary to calculate a work
index from operating data (e.g. from a pilot plant) so that power draw and mill sizing can be calculated.
For open circuit AG/SAG mills, the discharge can be hypothetically classified as per performance of the
ball mill cyclones, as illustrated in Figure 10. This „phantom‟ cyclone gives a more realistic F80 and quantity
of unfinished material (tph) for Bond calculations.
Phantom Final Product

Cyclone Cyclones P80
(Bond)
sam e
cl assi fication
curve
F80, TPH
(Bond)
New Feed
F80 Ball Mill
AG/SAG Mill
Figure 10: Example of phantom cyclone used to correct the F80 for Bond calculations
Single Particle Breakage Tests
The simulation-based approach that the JKMRC uses for comminution circuit optimisation and design
relies on laboratory breakage tests (Napier-Munn et al, 1996). The success of the circuit optimisation study
ultimately rests on the representativeness of the ore samples that are obtained for the testwork programme.
For the laboratory tests used at the JKMRC 50-100 kg samples of mill or crusher feed (of appropriate
particle size range) are used in the evaluation. For design 50mm or larger drill core samples are adequate.
The JKMRC comminution models rely on rock breakage characterisation data which are obtained using a
drop-weight device (crushers, ag/sag mills and ball mills) and tumbling tests (ag/sag mill only). In this
way the JKMRC research has aimed to improve the decoupling of material and machine in the simulation
of comminution. Figure 11 shows a schematic of the JKMRC drop-weight device.
18
Figure 11 : The JKMRC drop weight testing device
The tests involve breaking a wide range of particle sizes (-63+13.2mm) at a variety of energy levels. These
data are then used to “map” the breakage characteristics of the ore in terms of specific breakage energy
(kWh/t) and resulting product size distribution. The product size distribution data are condensed into a
single parameter called the t10 (Narayanan and Whiten, 1983). This is the percent of product passing
1/10th of the original particle size (Figure 12). A family of t n parameters can be generated in a similar
manner to the t10 eg. t2 is the percent passing 1/2 of the original particle size etc. It has been found that t 10
is related to the t family of parameters in a very similar manner for a wide range of rock types. It is
therefore possible to generate an entire product size distribution curve if the t 10 is known. From the
laboratory testwork the relationship between t 10 and kWh/t is measured and typically takes the shape
shown in Figure 13. This relationship is described using the equation:
t10 = A [1 - e(-b. Ecs)] (29)
where t10 is the % passing 1/10th of the initial mean particle size, Ecs is the specific comminution energy
(kWh/t), and A and b are the ore impact breakage parameters required by the model. A lower value of „b‟
indicates a harder ore. Parameter „A‟ represents the maximum level of breakage for a particular hard-rock
ore. As there is some interaction between A and b in the impact breakage equation (29), the JKMRC uses
A*b and t10 (at 1 kWh/t) for comparison because these numbers are better defined. A*b is the slope of the
curve at „zero‟ input energy.
19
t10 (%)
Y = Test par ticle siz e A = 49.1

Cumulat ive percent passing %
50
t 40
2
30
b = 0.87
20
t4
10 -b.Ecs
t10 = A [1-e ]
t10
0
0.0 0.5 1.0 1.5 2.0 2.5
Y/10 Y/4 Y/2 Y
Particle s ize m m Specific Comminution Ener gy Ecs (kW h/t)
Figure 12: Relationship between t10 and Figure 13: Example of relationship between
product size distribution t10 and specific energy
In AG/SAG milling, size reduction is believed to occur both by high energy impact (crushing) and low
energy abrasion or chipping. Abrasion and chipping breakage events leave the original particle largely
intact, the products being relatively fine particles. Impact, however, typically breaks entire particle into
fragments of a range of sizes. The appearance functions are size-energy dependent, and the prevailing
energy levels in the mill are estimated from the mill load (determined iteratively) and mill size.
The resistance of the ore to low energy breakage or abrasion is determined by tumbling a 3kg charge of
-55+38mm material in a 300mm by 300mm laboratory mill for 10 minutes at 70% of the critical speed and
sizing the broken product. The test provides the parameter „t a‟, taken as 1/10th of the t10 of the sized
product, which is related to the proportion of fines generated in a tumbling test. A low value of t a indicates
a high degree of resistance to abrasion. Ore types with t a values around 0.3 or below represent good
competent AG/SAG grinding media. Higher values of t a typically reflect softer ores. An increase in ta will
therefore be expected to translate into reduction in the specific power consumption in an operational mill.
Correlations with Bond Work Index

There is some correlation between the ore impact (A*b) and abrasion ( t a ) parameters and the Bond ball
mill work index. However, the work index measures the reaction of a -3.35mm feed to the mix of abrasion
and impact present in a dry ball mill. The JKMRC breakage tests try to separate impact and abrasion
breakage and, as such, would not be expected to correlate strongly, as supported by the data on 47
different ore types presented in Figure 14.
20
A*b = -3.5WI + 117 -1.34

t = 19.7WI
a
250 3.0
2.5
200
Abrasion Parameter ta
Impact Par ameter A*b
2.0
150
1.5
100
1.0
50
0.5
0 0.0
5 10 15 20 25 30 5 10 15 20 25 30
Bond Work Index, WI (kWh/t) Bond Work Index, WI (kWh/t)
Figure 14: Correlations between A*b and ta parameters, and Bond work index
Crusher Model Breakage Function

The current crusher model (Andersen, 1988), which is a development of Whiten‟s original crusher model
(1974), uses the energy-size reduction results derived from the drop-weight test on coarse particles to
predict both breakage and crusher power consumption. In this case the drop weight test results are used to
determine two sets of crusher ore-specific parameters.
The first part is an appearance function which relates the degree of breakage, t 10 , to the remainder of the
size distribution (Table 4). Unlike the AG/SAG model, the crusher model uses actual ore appearance data,
not the standard mapping.
Table 4: Example of ore parameters used in JKMRC crusher model
(a) Appearance Function Data, tn (%)
t10 (%) t75 t50 t25 t4 t2

10 2.92 3.89 5.97 20.74 51.58
20 5.54 7.46 11.73 40.16 76.04
30 8.21 10.97 17.26 58.35 88.35
(b) Size reduction/specific comminution energy, Ecs (kWh/t)
t10 (%) Initial Particle Size (mm)

12.2 17.5 29.0
10 0.32 0.27 0.25
20 0.65 0.53 0.50
30 0.98 0.79 0.76
The second part of the crusher ore breakage description relates the size-specific energy required to achieve
a particular degree of breakage (Table 4b). The crusher model uses splines (or mathematical curves) to
relate t 10 to Ecs, instead of the exponential relationship (equation 29). These splines are described at
nominal t 10 levels of 10, 20 and 30%, for each of three particle sizes used in the drop weight or pendulum
test.
The latter information is also used to determine how much power (per unit feed mass) would be required
to crush feed size to product size if all the crushing was carried out in the pendulum or drop weight tester.
This 'pendulum power' is closely correlated with net crusher power consumption. This method of power
correlation is much more flexible than the traditional F80 / P 80 approach (Morrell et al, 1992).
21
In applying the JKMRC crusher model when the feed is known and the product is the desired output, the
problem resolves itself into obtaining estimates of the breakage function and classification parameters for a
particular machine and feed material. Whilst the classification parameters are directly related to the crusher
closed and open side settings, the ore-specific energy-size-reduction relationship for the crusher is usually
unknown. This is summarised in a parameter T10 which is related to the size distribution of the breakage
products. Typically this parameter needs to be fitted to survey data from the crusher in question. However,
in vertical shaft impact crushers it may be possible to determine the applied specific comminution energy
and hence T10 parameter from first principles. The basis of this information comes from the drop-weight or
pendulum tests on the rock sample, rotor speed and diameter.
The drop-weight or pendulum test should be conducted on representative ore particles over the range of
the crusher feed size. Where a specific mathematical performance model has been developed from plant
surveys, the power draw may then be predicted for different operating conditions.
In a design situation, given the feed and the desired product size distributions, the t 10-size-Ecs relationship
(equation 29) for the ore to be processed must be obtained from the drop-weight or pendulum test and this
information used in the model to calculate the total comminution energy required. The crushing power
requirements can then be determined for a similar crusher from a power correlation obtained from another
site.
The effect of particle size on the t 10-Ecs relationship is usually negligible in hard ores, but the A and b
parameters may both be dependent on particle size, shape and ore type.
Ball Mill and Rod Mill Appearance Function

The JKMRC mill models are based on the assumption that the contents of the mill are perfectly mixed
(Napier-Munn et al, 1996). At steady state, a material balance can be determined for each size fraction:
feed + appearance via breakage = breakage to smaller +

product (30)
of larger particles particles
If the contents of the mill are known, the breakage rates can be computed from equation 30 knowing the
feed and product size distributions and the appearance function. In practice, this is rarely if ever the case
and the equation is modified to replace the mill contents with the term product/discharge rate. Since the
ore-specific appearance function can be independently determined from the pendulum breakage test, the
parameters of the model then become the breakage rate/discharge rate (r/d) terms, which are determined
by non-linear least squares fitting of the modified equation to operating data. Normally r/d is found to be a
regular function of particle size, and this dependency is represented in the model by spline functions, the
estimation of the spline knots being part of the model fitting procedure. In practice r/d terms are also
normalised through mill dimension which allows the model to scale.
The ball mill and rod mill models use an ore-specific appearance function determined from the ore-specific
t10 calculated using equation 29, and the standard 't' family of curves using a specific comminution energy
Ecs for a standard input energy level (Ei) of 41.788 kg.cm, applied to the -5.6+4.75 mm size fraction
(Narayanan, 1985). Hence the test Ecs will depend largely on the ore sg and particle shape, that is the
average particle mass in the -5.6+4.75 mm size fraction. This has been traditionally determined using a
smaller pendulum device than that used for the SAG/crusher tests, and more recently the lower drop
weight tester energy range.
Once the ore-specific t 10 -Ecs relationship and standard Ecs have been determined, the appearance function
values are obtained using a computer program which reconstitutes the product size distribution from the
predicted t 10 and the given t-curves. The results of this program represent a relative size distribution (mass
fraction retained), as illustrated in Table 5.
The appearance function starts with the value 0 since the model assumes that every particle once broken
will no longer belong to the initial size fraction. Hence the daughter particles will be distributed in a 2
sieve series, starting with the next fraction down. To illustrate the range of values seen between hard and
22
soft ores, Figure 15 shows the appearance functions or breakage distributions for a primary and oxide gold
bearing ores. Note the shift in the peak to the left with the softer oxide ore.
Table 5: Typical ball or rod mill appearance function
Internal Appearance
Sizing (mm) Values
16.0 0
11.3 0.043
8.00 0.084
5.66 0.114
4.00 0.119
2.83 0.111
2.00 0.100
1.41 0.088
1.00 0.074
0.707 0.061
0.500 0.050
0.354 0.041
0.250 0.032
0.177 0.025
0.125 0.019
0.088 0.014
0.063 0.010
0.044 0.007
0.031 0.004
0.022 0.003
For ball mill modelling, and fine grinding devices like tower or stirred mills, the single particle approach as
well as the range of particle sizes may in some applications be inappropriate. Accordingly attention is being
given to alternative methods of obtaining the breakage function for fine particles. Work is also being done
on determining the breakage function for the comminution of particle beds, such as occurs in the high
pressure grinding rolls.
15
10
Wt .% Retained
Primary
Oxide
5
0
.01 .1 1 10 100
Particle Size (m m )
Figure 15: Example of ball mill appearance function for a hard (primary)
and soft (oxide) gold ore
CONCLUSIONS
23
The acquisition of reliable data from an operating circuit lies at the heart of any optimisation exercise.
Great care must be devoted to the task, as conditions will usually conspire to frustrate the surveyor at
every turn, and the need for good data is absolute; no compromise should be entertained if at all possible.
The sequence of events should be:
1. Define the objectives, and identify the units to be surveyed.

2. Plan the survey, taking into account the
sample points
size of sample required at each point
data to be collected
any likely difficulties, e.g. accessibility, production interruptions, non steady-state conditions,
missing data etc.
3. Conduct the survey according to the tenets of good practice.
4. Analyse the samples, again with care.
5. Analyse and mass balance the data; reject poor or doubtful data, and re-survey if necessary.
6. Use the data as defined by objectives.
Various laboratory rock breakage characterisation tests have been developed for studying breakage
phenomena and for designing and predicting the performance of industrial comminution equipment.
These tests fall into three main classes: conventional rock and fracture mechanics measurements, standard
grindability tests, and the single particle tests.
The JKMRC simulation models aspire to a clear goal, that of separating ore characteristics from those of the
processing machine. In this context, neither conventional rock and fracture mechanics measurements, nor
standard grindability tests, provide information on the breakage function, which is necessary in the
prediction of the product size distributions from crushers and tumbling mills.
The conventional grindability test is based on Bond's „third theory‟ of comminution (1952). This method
continues to be the main tool for designing comminution machines, and its long use has engendered
considerable confidence in its application. The particular value of the Bond grinding test is its simulation of
the recycle elements of closed circuit grinding systems, including the behaviour of components of different
„hardness‟. However, the Bond test also has some disadvantages. For example it is not a good predictor of
AG/SAG mill behaviour (for which it was not designed), nor of situations in which it is desired to
investigate changes (usually increases) in feedrate.
Single particle breakage tests are now in routine use at the JKMRC and in a number of licensed testing
laboratories around the world.
The breakage size distributions resulting from breakage of hard-rock ores in a pendulum or drop weight
device have been shown to form a one-parameter „t‟ family of curves independent of ore type and
normalised for particle size. In addition an empirical equation with ore-specific parameters is used to relate
the t10 parameter to the specific comminution energy, Ecs, measured in the pendulum or drop weight tests.
This approach covers the range of shapes of breakage distribution functions commonly used in crusher and
tumbling mill modelling, depending on the Ecs level of the single particle breakage event.
In summary, the energy-size reduction results from pendulum and drop-weight breakage tests are now
routinely used to provide ore parameters for use in JKSimMet simulation models of ball and rod mills,
autogenous/semi-autogenous mills and crushers. The unit parameters (eg. breakage rates) are calculated
from plant survey data. Once the models have been customised to an existing circuit, the behaviour of the
comminution circuit over a wide range of operational conditions can be accurately predicted, as evidenced
in numerous optimisation studies carried out to date. In design of new plants, the extensive JKMRC
database on ore types and related circuit performance is used to select appropriate ore and unit parameters
for the simulation studies.
Current experience suggests that the single particle test, when combined with process models, is superior
to other methods for evaluating coarse breakage machines such as crushers and AG/SAG mills. Although
24
it is also used successfully in describing ball mills, it is likely that the breakage characteristics of relatively
large rocks (say down to 5mm) may not describe well the same material at say 100mm. Further work is
therefore required to improve the independent measurement of the breakage characteristics of fine
particles, and this research is underway.
REFERENCES
ALLEN T. 1996. Particle size measurement. 5th ed. Chapman and Hall.
ANDERSEN J.S. 1988. Development of a cone crusher model. M.Eng.Sc. Thesis, University of Queensland
(JKMRC).
BARBERY G. 1972. Derivation of a formula to estimate the mass of a sample for size analysis, Trans
AusIMM, 81, (784), March, C49-C51.
BEARMAN R.A., BARLEY R.W. and HITCHCOCK A. 1991. Prediction of power consumption and
product size in cone crushing. Minerals Eng, 4(2), 1243-1256.
BOND F.C. 1952. The third theory of comminution. Trans SME/AIME, 193, 484-494.
MORRELL S., NAPIER-MUNN T.J. and ANDERSEN J. 1992. The prediction of power draw for
comminution machines. Comminution: Theory and Practice, Ed: Kawatra, 405-426, 233-248, (AIME).
MORRELL S. 1994. Power draw of grinding mills - its measurement and prediction. 5th Mill Ops Conf,
Roxby Downs, Oct, 109-114 (AusIMM).
NAPIER-MUNN T.J., MORRELL S., MORRISON R.D. and KOJOVIC T., 1996. Mineral Comminution
Circuits: Their Operation and Optimisation. JKMRC, University of Queensland, Brisbane, 413pp.
NARAYANAN S.S. and WHITEN W.J. 1983. Breakage characteristics of ores for ball mill modelling.
Proceedings AusIMM, 286, June, 31-39.
NARAYANAN S.S. 1985. Development of a laboratory single particle breakage technique and its application to
ball mill modelling and scale-up. PhD Thesis, University of Queensland (JKMRC).
PITARD F.F. 1993. Pierre Gy’s sampling theory and sampling practice. CRC Press, 2nd ed, 488pp.
WHITEN W.J. 1974. A matrix theory of comminution machines. Chem Eng Sci, 29, 588-599.
WILLS B.A. 1992. Mineral Processing Technology. 5th Ed., Pergamon Press, Oxford, 855pp.
25

Surveying Comminution Circuits and Ore Characterisation

Uploaded by

Copyright:

Available Formats

Surveying Comminution Circuits and Ore Characterisation

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Surveying Comminution Circuits and Ore Characterisation

Uploaded by

Copyright:

Available Formats

JULIUS KRUTTSCHNITT MINERAL RESEARCH CENTRE

Isles Road, Indooroopilly Telephone: (07) 3365 5888

SURVEYING COMMINUTION CIRCUITS AND ORE CHARACTERISATION

Figure 1: Schematic diagram of typical comminution circuit,

SOME SAMPLING PRINCIPLES

2. Sample cutter design.

3. Sub-sampling a primary sample.

5. The propagation of error when calculating quantities.

Size of Sample for Size Analysis

where M = mass of sample required (g)

(The variance 2 is essentially the FE referred to earlier).

d 3m can be calculated as (Barbery, 1972):

Table 1 gives values of z for different confidence levels.

Table 1: Normal ordinates (from the normal distribution)

Confidence Level (%) z

Propagation of Error in Calculated Quantities

The mass flow is given by:

The definition of variance suggests that

where signifies „maximum error in‟ and ( M)2 is the variance of M.

Two points are worth noting:

2. Errors often propagate in unexpected or counter-intuitive ways.

Table 2: Data to be collected from particular unit operations

PT VLI L 3 cos kW (5)

Coarse Feed Sampling

Figure 5: Example of mill load geometry parameters used to calculate volume

where S = chord length (m)

Controlled Mill Crash-Stop

Figure 7: Example of mill power and load response during a

SAMPLE ANALYSIS PROCEDURES

Ore Specific Gravity

1. Whatever method is chosen is applied carefully and correctly.

Initial Data Inspection

• a water balance around the cyclone, and

Both of these calculations are based on the two product formula.

Figure 8: Quantities around a hydrocyclone

The mass balance around the node for solids is given by

and for each size fraction by

F. fi = O.oi + U.ui (14)

and for water by

100 100 100

Hence, from each size fraction

or, from the percent solids measurements

Weights for Sizing Data

Table 3: Errors assumed for sizing data

Weight % Retained Absolute s.d. Relative s.d.(%)

Simple Mass Balance Solution for Multicomponent Data

with respect to , oi , u i , and fi , subject to the constraint that u i oi fi .

The Bond Test

How to Use the Bond Work Index

where EF1 ... are Bond‟s efficiency factors.

Application of Bond Theory to AG/SAG Mills

Phantom Final Product

F80 Ball Mill

Single Particle Breakage Tests

Figure 11 : The JKMRC drop weight testing device

t10 = A [1 - e(-b. Ecs)] (29)

Y = Test par ticle siz e A = 49.1

Correlations with Bond Work Index

A*b = -3.5WI + 117 -1.34