You are on page 1of 20

UPGRADING THE JK SAG MILL MODEL

* T. Kojovic1, M.S. Powell1, C. Bailey2 and D. Drinkwater2

1
University of Queensland, JKMRC - SMI
40 Isles Rd
Indooroopilly, Australia, 4068
(*Corresponding author: t.kojovic@jkmrc.uq.edu.au)
2
JKtech Pty Ltd
40 Isles Rd
Indooroopilly, Australia, 4068

1
UPGRADING THE JK SAG MILL MODEL

ABSTRACT

The original JKMRC AG/SAG mill model was developed over 30 years ago and has been
upgraded a number of times since then, with the current published version being the Variable Rates model
that is now over 15 years old. Since this model was developed further enhancements have been derived,
some of which are applied in-house and some which have yet to be incorporated. The published and the
commercially available model in JKSimMet is currently undergoing a comprehensive overhaul with a
vastly extended data base for model evolution and validation. Improvements developed within the AMIRA
P9 project, in-house development and a number of collaborative projects, are being incorporated into the
upgraded model structure. Enhanced capabilities will include: power-based breakage rates scaling; a far
wider mill filling range (from 10% to 40%) that exhibits peak in throughput as mill filling increases; higher
ball loads; improved pebble discharge function; improved slurry discharge function incorporating
important mill design and operational influences; and the new size dependent breakage function. The
model will also be able to accommodate some influence of liner design and multi-component ores (blends
of hard and soft components). This upgrade is still a work in progress but substantial changes have been
made to date. This paper presents some of the new improvements and their influence on the predictive
capability of the model.

KEYWORDS

AG/SAG mill model, JKSimMet, mill filling, breakage modelling, discharge function

INTRODUCTION

The research activities of the Julius Kruttschnitt Mineral Research Centre (JKMRC) have
recognized the continued popularity of autogeneous (AG) and semi-autogeneous (SAG) mills, and the need
for improved models to predict the performance of these machines under a wide range of conditions.
These models are relevant at all stages of the development of a comminution circuit; from the conceptual
design stage and geometallurgical modelling, to scale-up from pilot test work, and finally optimisation of
the industrial circuit.

The mathematical models of AG and SAG mills developed by the JKMRC over the past 30 years,
incorporated into their process simulator JKSimMet, have been successfully used in a variety of
applications. However, in some situations, the models do not reflect reality, forcing the user to limit the
range of operating conditions that can be reliably explored. These limitations have become well known
amongst the expert users, resulting in a range of proprietary compensating techniques that are used in
parallel with the JKSimMet model. Recognizing that these techniques are derived from the additional
knowledge and expertise accumulated over the past decade, the JKMRC embarked on a comprehensive
upgrade of the underlying model framework to incorporate as much of this new knowledge as possible.
The improvements earmarked for inclusion will address the key mechanisms of the AG/SAG process:
Breakage energy related to applied pinion power,
Sensitivity of throughput to mill filling and ball/rock charge ratio,
Effect of operational and design variables on slurry and pebble discharge, and
Effect of particle size on breakage.

The resulting model needs to be tested using industrial data under a range of conditions. To this
end the JKMRC has accumulated over 200 pilot and full scale AG/SAG mill data sets with which to

2
develop, test and refine its mathematical models. This paper describes the progress to date, and examples
of the new predictive capability of the model.

BRIEF OVERVIEW OF CURRENT JK AG/SAG MODEL STRUCTURE

This section summarizes the main components of the current JKMRC AG/SAG mill model, which
is based on the initial framework developed by Leung in 1987. A full description of the model can be
found in Napier-Munn, Morrell, Morrison & Kojovic (1996).

The size reduction processes inside an AG/SAG mill can be represented in a simplified manner as
shown in
Figure 1. Feed enters the mill and is subjected to breakage from collision with other particles
and/or the mill shell. The products from breakage either exit via the grate or remain to undergo further
collisions. This process has essentially three components, namely

Collision frequency (breakage rate, R),


Description of ore size distribution after collision (appearance function, A),
Description of particle transport out of the mill (discharge rate, D).

Classification

Feed Product
Breakage

high Low
energy energy Grate
impact impact

Figure 1– Schematic of AG/SAG Mill Operation

At steady state these components are combined in the perfect mixing model (Whiten, 1974) which
can be simplified as follows:

f R.s A.R.s D.s 0 (1)


where
f = feed rate
p = product rate
R = breakage rate
s = mill contents
D = discharge rate
A = appearance or breakage distribution function

Data from 65 pilot plant and full scale mills, covering a wide range of ore types, feed sizes and
operating conditions were used to fit the breakage rates in Equation 1. The resulting relationships between
the prevailing operating conditions and the breakage rates were used to develop the current Variable Rates
AG/SAG model in JKSimMet (Morrell and Morrison, 1996).

3
Appearance Function

The appearance or breakage distribution function is a description of how an ore particle breaks
when subjected to impact energy. In AG/SAG milling both crushing (high energy) and abrasion/chipping
(low energy) size reduction are believed to occur. Laboratory techniques have been developed at the
JKMRC which are used to characterise an ore with respect to both these breakage modes. High energy
breakage is characterised using the Drop Weight Test (DWT) or Rotary Breakage Test (RBT) device,
whilst low energy breakage is tested using a laboratory mill tumbling test (Napier-Munn et al., 1996). In
the standard data reduction procedures, the DWT results from testing five size fractions and a wide specific
energy range (0.1 to 2.5 kWh/t), are used to calibrate two parameters in the JKMRC breakage model:

t10 A 1 e b.Ecs (2)

where t10 is a size distribution ‘fineness’ index (defined as the progeny percent passing one tenth of the
initial mean particle size), Ecs is the specific comminution energy (kWh/t), and A and b are the ore impact
breakage parameters determined from DWT results.

The index A*b has become well known in the mining industry as a reliable indicator of impact ore
hardness, and essentially describes the rate at which fines are produced (t10) for a set amount of specific
energy (Ecs).

The value of Equation 2 is embedded in the JKSimMet comminution models, which rely on the t 10
to generate a full size distribution given the relationships between t 10 and tn-family curves established from
the drop weight test database (Narayanan and Whiten, 1988). That is, the model only needs to know the
Ecs and the ore parameters A and b to generate the product size distribution for a given breakage event.

However, since Equation 2 is used to fit the DWT data with one set of A and b parameters for all
particle sizes, this typically results in a scattered plot due to the particle size effect, as illustrated in Figure
2. This ‘average’ set of A and b parameters used in the AG/SAG model assumes that particles of different
sizes would be broken in the same way when subjected to the same impact energy. However, this
assumption may bias the fitted breakage rates using Equation 1, affecting the simulation accuracy when the
feed type and/or size distribution significantly changes. Hence this component of the original model was
modified in the upgrade, as outlined in the New Model Structure section.

60 10.3 mm

50 14.5 mm

40 20.6 mm
t10 (%)

28.9 mm
30
41.1 mm
20
57.8 mm
10
Calculated
0
0.0 2.0 4.0 6.0
Ecs (kWh/t)

Figure 2 – The JK breakage model (Eqn. 2) fitted to DWT data with one set of model parameters A and b
for all particle sizes

4
Discharge Rate

The discharge rate is considered to be the product of two mechanisms:

1. transport to the grate, and


2. classification by the grate.

and is modelled using the following equation:

di Dci (3)
where
di = discharge rate of size class i
ci = classification function value for size class i
D = maximum discharge rate

The maximum discharge rate (D) is determined iteratively within the model using an empirical
relationship which relates the slurry hold-up to the normalized volumetric flowrate of slurry discharged
from the mill:
L m1( F V) m 2 (4)
where
L = fraction of mill volume occupied by below grate size material
F = normalized flowrate through mill (mill fulls per minute)
m1 = constant (linked to grate design, grate open area and mill speed)
m2 = 0.5

The classification function (ci) at the grate is based on a simple model as shown in Figure 3.

Inflection point

Xm Xg Xp

Figure 3 – Example of current model grate classification function, superimposed over actual data for pilot
mill with 8mm grate and 50mm ports

The value of Xg is the effective grate aperture, whilst Xm is the maximum particle size which
behaves ‘like water’, ie. is not subject to classification. Xp is the effective pebble port aperture and the
model input includes fp, the fraction of total open area that the pebble ports represent. When pebble ports
are used, the current JKSimMet AG/SAG model simplifies the discharge curve with 3 straight lines as
opposed to the two that the grate-only model uses. Xg, which in theory should have an x co-ordinate equal

5
to the grate aperture, defines an inflection point in the line from Xm to Xp. The Y co-ordinate for Xg is
calculated from the pebble port fraction of total open area, fp. For example, if the actual fp was 0.5 then the
X and Y co-ordinates of the inflection point would be 8 mm and 31 mm (ie. 0.5 times the maximum
discharge function value D). The pebble port fraction of open area for the data shown in Figure 3 was
approximately 0.05, and hence, according to the model, the discharge function value d i at Xg should have
been 3. Instead its actual value is around 20. The misplacement of this point has two effects, 1) it causes
the model to predict the wrong pebble product size distribution, and 2) it also causes the model to predict
the incorrect pebble flowrate. The latter more serious problem results in the JKSimMet model predicting
the incorrect AG/SAG mill throughput. Hence this part of the model was recognized as one of the critical
areas requiring upgrade, as discussed in the New Model Structure section. In recent years, this shortcoming
has been compensated for by artificially adjusting Xg as described in Bailey, Lane, Morrell & Staples
(2009).

The Variable Rates model includes an additional function which links Equation 4 to the grate
design, grate open area and mill speed (Morrell and Stephenson, 1996) via the m1 constant in Equation 4:

J kF0.5 1.25A 0.5 0.67D 0.25 (5)


where
J = fractional slurry hold-up
k = constant (depends on grate size and discharge coefficient)
F = flowrate of slurry discharged by mill (m3/h)
A = total area of grate apertures (m2)
D = mill diameter (m)
= fraction critical speed
= mean relative radial position of grate apertures

Breakage Rate

Given a feed and product size distribution from an AG/SAG mill, and an appearance function of
the ore being treated, the breakage rate distribution can be back-calculated using Equation 1. The
distribution is represented using cubic splines (Ahlberg, Nilson and Walsh, 1967) at five spline knots
situated at 0.25, 4, 16, 44, 128 mm. The rates at these sizes are then used to represent the entire breakage
rate distribution across the full particle size range. The breakage rate is related to particle size and typically
takes the form shown in Figure 4.

1000
Breakage Rate (hr-1)

100

10

R1 R2 R3 R4 R5
1
0.1 1.0 10.0 100.0 1000.0

Size (mm)

Figure 4 – Example of an AG/SAG Mill Breakage Rate Function

6
Of particular relevance are the maximum and minimum points on the curve. These points are
believed to indicate the limits, with respect to size, of the various breakage mechanisms that occur within
the mill. By studying how the shape of the breakage rate curve changes, the effect of key operating
variables can be evaluated and explained.

The shape of the breakage rate distribution is believed to be governed by the charge particle size
distribution and the frequency of charge rotation. It is therefore dependent on the ball charge, mill filling,
feed size distribution and mill speed. These relationships are captured in a set of 5 regression equations
which form the current Variable Rates model (Morrell and Morrison, 1996). The equations were calibrated
using 52 sets of pilot data on 16 ore types covering a wide range of operating conditions (feed size, ball
load and recycle load). Since the pilot database had little variation in mill speed and ball size, the
equations were corrected to take into account the effect of these two critical operating variables.

Breakage Energy

The current model assumes the nominal specific energy (EL) in the mill to be a function of the
mill diameter and average size of the coarsest 20% of particles in the mill charge (S20). This includes the
balls which are converted to equivalent rock particles using a simple density scale-up approach. The
nominal EL is then used to estimate the specific comminution energy for each particle size, using a
relationship developed by Austin, Klimpel and Luckie (1984):

EL
Ecsi (6)
x i1.5
where
Ecsi = specific comminution energy (kWh/t)
EL = nominal specific energy level in mill (based on mill diameter and S20 mean size in charge)
xi = means size of class i

The Ecs calculated for each size, using Equation 6, is used to determine the high energy
appearance function, given the DWT test ore-specific A and b parameters and Equation 2. The high
energy t10 (the) is then combined with the low energy t10 (tle ), determined from the JK Abrasion Tumbling
test, to provide a combined appearance function for each breakage event:

t he A he t le A le
A (7)
t he t le
where
A = combined appearance function for a given Ecs
Ahe = crushing appearance function (high energy)
Ale = abrasion/attrition appearance function (low energy)
the = t10 parameter for crushing breakage (high energy)
tle = t10 parameter for abrasion/attrition breakage (low energy)

The weighting system represented by Equation 7 has no direct link to the actual energy
distribution in an AG/SAG mill and more specifically, to the proportion of the applied energy used in
crushing and abrasion. Similarly, there is no relationship between the nominal energy level EL in Equation
6 and specific energy applied, other than the fact that EL approximates the potential energy of the coarsest
20% of material in the mill. Figure 5 shows the relationship between the EL and specific energy for 20
industrial scale AG/SAG mills, which confirms the weak correlation. These deficiencies warrant further
refinement if the model is to correctly predict the effect of changing the mill filling, speed and rock/ball
charge distribution.

The current model algorithm is shown in Figure 6 which shows the main steps outlined above and
the iterative solution controlled by the mass transfer model represented by Equation 4. The ability of the

7
model to reproduce experimental data largely depends on the interaction of the mill load and breakage rate
function, which is assumed to be exclusively machine related. However, experience has shown that the
breakage in the mill is also affected by the mill charge environment which changes in response to ore and
mill operating conditions. As such, the complex and dynamic nature of the AG/SAG mill process has not
completely been described in the current algorithm, despite the improvements made in the Variable Rates
model released in 1996.

3.0E-04

2.5E-04
Model EL (kWh/t)

2.0E-04

1.5E-04

1.0E-04

5.0E-05

0.0E+00
0 50 100 150

Mill Specific Power (kWh/t)

Figure 5 – Relationship between model energy level EL and actual specific power
(Feed Rate/Gross Power)

8
INPUTS
Mill Design
Mill Ball Load
Feed Rate and Size Distribution
A, b and ta
Ore SG
Breakage Rates

Estimate volume of
below Xg size material
in mill, L using mass
transfer model (Eqn. 4)

Estimate discharge function


Estimate breakage rate Estimate the Ecs for each
from current estimate of
distribution from 5 rates particle size given X20 and D
maximum discharge rate Dmax
and using Splines (Eqn. 6, Eqn. 2)
Xg, Xm and Xp & fp (Eqn. 3)

Calculate mill load and product


using Perfect Mixing Model
(Eqn. 1 and Eqn. 7)

Compare predicted
amount of below Xg size
material with estimate L

Is difference
Adjust Dmax
acceptable?
NO

YES

EXIT

Figure 6 – Flowchart of current JK AG/SAG mill model algorithm

NEW MODEL UPGRADES

In embarking on the JK AG/SAG model upgrade, five changes were identified for the initial
revision to address some of the main deficiencies noted above:
1. Introduce the new JKMRC breakage model (Shi and Kojovic, 2007) to take into account the
size effect on impact breakage behaviour. This will incorporate the size-by-size nature of
rock strength that is often apparent from the results of DWT tests. Across the standard test
range of 13.2 to 63mm, the slope of normalised A*b vs particle size can vary by a factor of 2.
2. Upgrade the structure to 6 knots instead to the current 5, to handle finer grinding mills. The
6th knot was set at 75 microns. This has been particularly absent in analysis of platinum ore
grinding circuits targeting very fine products in single stage grinding.
3. Replace the current S20/EL concept with a power based approach, distributing the net applied
energy with respect to the surface area in the mill load, from the coarsest to the finest size
class (excluding sub-mesh).

9
a. The power draw would be estimated from the load and mill operating conditions
using the JKMRC model (Napier-Munn et al, 1996).
b. The surface area per particle will be estimated using the relationships developed
by Michaux and Djordjevic (2010).
4. Introduce a mill filling sensitivity to the breakage rates, based on the approach developed
within the Amira P9M project. The final implementation of this sensitivity will depend on
the impact of the first three modifications.
5. Introduce a new pebble discharge model into the mill load iteration loop following the mass
transfer law, adjusting the effective fp to ensure the expected pebble rates fall within 10% of
the expected rate.
The upgrade was applied to the original Leung model structure, since the Variable Rates model is
also based on the same structure, as noted previously. Similarly, the form of the regression equations was
not expected to remain the same once the structure was modified. The only compromise was the
introduction of the Variable Rates m1 relationship since the original model kept m1 constant. The m2 was
set to 0.5, as per the Variable Rates model. Future refinements will consider incorporating the slurry limit
relationship developed by Powell (2001) which incorporates the ball load and pulp lifter depth.

New Appearance Function

The JKMRC comminution research team has developed a new breakage model incorporating the
effect of particle size, based on a theoretical approach described in Vogel and Peukert, (2004), considering
a generalised dimensional analysis proposed by Rumpf (1973), and a detailed fracture mechanical model
based on Weibull statistics (1951). The new breakage model has the same form as Equation 2, but with
particle size and breakage properties incorporated explicitly in the model (Shi and Kojovic, 2007):

t10 M 1 exp Fmat XEcs (8)


where
M = maximum possible value of t10 in impact (%)
Fmat = material property (kgJ-1m-1) = f(X,p,q) and p and q being ore specific constants
q = ore size effect parameter
X = particle size (mm)
Ecs = impact breakage energy (J/kg)

A*b can be determined from M and Fmat at the standard DWT size as follows:

A*b 3600.M.Fmat .X (9)

where Fmat is calculated at X=32.6mm, and the constant 3600 is used for unit conversion.

This direct relationship means that the upgraded model will work with the existing ore
characterisation data, since the independent variables incorporated in the new breakage model can all be
extracted from the existing DWT database.

The validity of the new breakage model is well borne out by Figure 7, which compares the fitting
results of the new model to the same data previously shown in Figure 2. Clearly, the test data for all six
particle size fractions now fall on the similar trend line. This comparison suggests that the present
breakage model has a fundamentally better structure for describing the effect of particle size on the
breakage distribution function. So far, over 100 sets of DWT data have been used to validate the new
breakage model, including single impact, repetitive impacts and bed breakage of mill feed ore, drill cores
and coal samples. In all cases, the new model produced excellent fitting results.

10
70
10.3 mm
60
14.5 mm
50
20.6 mm
40
t10 (%) 28.9 mm
30
41.1 mm
20
57.8 mm
10
Calculated
0
0.0 1.0 2.0 3.0 4.0
fmat .x.k(Ecs -Em in )

Figure 7 – The new breakage model (Eqn. 8) fitted to 42 data points from DWT on same ore

The new breakage model has been incorporated in the upgraded AG/SAG model structure.
Comparisons between the fitted breakage rates using the size-dependant and the size-averaged breakage
appearance functions show that the size-averaged appearance function may over-predict the breakage rates
at large particle sizes, but under-predict at medium to fine particle sizes (see Figure 8). The magnitude of
this bias would be expected to be ore specific. The results of a validation case study presented in a latter
section suggest the new breakage model provides a more accurate basis for AG/SAG mill circuit design
and optimisation.

100
Breakage Rate (hr-1 )

10

R1 R2 R3 R4 R5
1
0.1 1.0 10.0 100.0 1000.0

new model
Size (mm)old model

Figure 8 – Comparison of fitted breakage rates using the original (Eqn.2) and new breakage model (Eqn. 8)
fitted to Cannington AG mill survey data

Breakage Energy Distribution

One of the key drivers that is not used in the current JK AG/SAG model is the power consumed.
Instead the model uses the S20/EL concept which was shown to be poorly correlated with the actual
specific power. If this concept is to be replaced, the alternative needs to offer a sound basis particularly in
the way the applied energy is distributed across the mill charge. The initial approach was to assume the
energy would be consumed in direct proportion to the available surface area of the particles in the load,
excluding the sub-mesh fraction. To facilitate this calculation, the particle volume and surface area
equations of Michaux and Djordjevic (2010) were used to estimate the total surface area in each size class.

11
The resulting distribution of surface area by class was then compared with the current model expectation.
The two curves have the same form, as shown in Figure 9, which suggests the underlying surface area basis
is similar to the model proposed by the Austin et al (1984).

100

90

80
Cumulative %

70

60

50

40

30

20 Sur.Area
10 Ecs (model)
0
0.0 0.1 1.0 10.0 100.0 1000.0

Size (mm)

Figure 9 – Comparison of new energy distribution versus current model based on Austin et al (1984)

The significance of the surface area approach is that it provides a convenient way to distribute the
applied specific energy by class. Hence the new model no longer requires the S20 calculation which has
caused innumerable JKSimMet warning messages, when the feed rock top size is finer than the equivalent
ball size. The surface area approach considers the balls in the charge as additional surface area, which is
more robust than the S20 approach. The new breakage energy distribution is calculated using the
expression:

%SA i L.Mass _ Total Net _ kW TPH


Ecs i (10)
100 L.Mass i
where
Ecsi = specific comminution energy (kWh/t) for size class i
%SAi = percent of total surface area in mill load (including rocks and balls) in size class i
L.Mass_Total = total mass of mill load (rocks only)
L.Massi = mass of rocks in size class i
Net_kW = net power drawn by mill
TPH = fresh feed rate

The proposed change to the AG/SAG model algorithm means the iterative solution would replace
the S20/EL calculation with the surface area and power draw calculation. The end result would be a
specific energy distribution across the full particle size range.

Mill Filling Sensitivity

In the current JKMRC AG/SAG model, the perfect mixing structure (Equation 1) predicts that
throughput is linearly related to the load mass (S) since the appearance function (A) and breakage rate (R)
are constant:
TPH RS(1 A ) (11)

12
The sensitivity to mill filling is not addressed by the Variable Rates regression equations which
are independent of the rock load mass and volume. The current linear throughput versus load level
relationship can give reasonable predictions up to the load level, where throughput is a maximum. Beyond
that load level the model predictions are progressively over-optimistic and misleading. Typically the load
level which maximises throughput falls in the range 20-40%, depending on the ore hardness, ball charge,
mill speed, and whether the circuit is run in open or closed circuit. Work in the Amira P9 project has
explored the load-throughput relationship and offered ways to model the inherent non-linearity. The
AG/SAG model upgrade is based on this work, focusing on how the energy might be utilized by the load as
the mill filling and ball charge is varied. These concepts will be verified using DEM.

The algorithm in the revised AG/SAG model uses two factors determined using the relationship
between the thickness of the bed at the toe of the charge and the volume occupied by the total charge. The
factors related to the influence of impact and abrasion/attrition are:
H U
Impact Adjustment Factor (I) (12)
B
Abrasion Adjustment Factor (A) B (13)
where
H = relative impact height in mill (distance between toe and shoulder)
U = charge energy utilization ratio (related to the ball:rock ratio)
B = relative charge bed thickness (based on JKMRC power model concept, Napier-Munn et al,
1996)

Since the JKMRC power model has a relationship for the charge thickness as well as the drop
height, these factors were simple to extract. The adjustment factors were applied to the breakage rate
distribution in Equation 1, using cubic splines at the 6 size knots, as illustrated in Figure 10. Note that the
new fine 6th knot has been incorporated at R0. The impact factor (I) was applied to rates R3-R5, the
attrition factor (A) was applied to rates R0 - R2.

1000
R0 R1 R2 R3 R4 R5
Breakage Rate (hr-1)

100

10

1
0.01 0.10 1.00 10.00 100.00 1000.00

low filling
Size (mm)
high filling

Figure 10 – Effect of mill filling on the breakage rate distribution in SAG mill (4% BL)

Pebble Discharge Model

Until a more comprehensive engineering model is available, the pebble discharge model will
follow the approach developed in the Amira P9M project. This relationship, shown in Figure 11, suggests
the pebble discharge rate, expressed in m3/h per m2 of port area, is linearly related to the proportion of

13
material in the load, which in theory could report as trommel screen oversize. The trend is based on pilot
data across a wide range of pebble port sizes and feed ore types. The strong correlation suggests the load
size distribution has a considerable influence on the pebble production rate; the percentage of pebble-sized
rocks in the load varying in the range 30 – 55% in the pilot mill.

40

Pebble Disch. Rate (m3/h per m2) 35 y = 0.5375x


R² = 0.82
30

25

20

15

10

0
0 10 20 30 40 50 60 70

% Pebbles in Mill Load

Figure 11 – Relationship between pebble discharge rate and mill load pebble content for pilot data
The upgraded relationship will include the mill filling, ball charge and mill speed; variables which
would be also expected to control the final pebble discharge rate. The proposed form of the model is as
follows:
Peb.Disch.Rate k L%( Xp Xt ) J Jo Cs o Cs 1 a (BL BL o (14)

where
L%(-Xp+Xt) = percentage of mill load between pebble port size and screen aperture
J, Jo = fractional mill filling (% of mill volume, actual and baseline)
Cs, Cso = mill critical speed (%, actual and baseline)
BL, BLo = ball charge (% of mill volume, actual and baseline)
k = constant (to be calibrated)

The baseline factors in Equation 14 were initially set to be 25% filling, 75% critical speed and
10% ball load; typical conditions for a SAG mill. However, more work is required to identify the most
appropriate model form.

VALIDATION OF NEW MODEL

For the validation, three data sets were used to demonstrate the capability of the new model, using
conditions which were known to cause issues with the current model. These data were based on JKMRC
surveys at the following operations:

Newcrest Cadia Hill (Cu/Au) – SABC, open circuit with pebble crushing (A*b ~ 41)
o 12.19 m dia. x 6.71 m EGL (Variable Speed)
Alumbrera Mine (Cu/Au) – SAG, open circuit (A*b ~ 55)
o 10.97 m dia. x 5.87 m EGL (Variable Speed)
BHP Billiton Cannington (Ag/Pb) – AG, closed circuit (A*b ~ 88)
o 8.53 m dia. x 4.57 m EGL (75% Cs)

14
The Cannington data were collected as part of the AMIRA P9 programme, whilst the Cadia Hill
data were collected during the JKMRC post-commissioning studies. The Alumbrera data sets were
collected during an optimisation study and were donated to the JKMRC by MIM.

The model upgrades were programmed in Fortran, the original source code used in the JKSimMet
AG/SAG model and supporting subroutines. Though this approach was necessary to fast track the changes,
it was less flexible in terms of testing alternative circuit configurations. The code was set-up to allow both
calibration of breakage rates and simulation of alternative feed and mill operating conditions. The mill
filling sensitivity was embedded in the simulation mode, with the program adjusting the rates according to
the prevailing filling, which should follow the non-linear throughput-load curve for the mill conditions in
question. If the mill motor power or milling filling is exceeded, the user is warned and asked to reduce the
throughput accordingly. Preliminary testing of this sensitivity needs further refinement but the initial
structure appears to be heading in the right direction.

It appears the new energy/surface area calculation, combined with the new breakage model, will
impact on the shape of the familiar AG/SAG mill breakage rate curves, as illustrated in Figure 12. The
open circuit SABC circuit treating competent ore shows a curve more aligned with the ball mill, yet the
soft ore, open circuit SAB circuit shows a very different shape, suggesting the coarse end is very friable.
The lowest rates stem from the closed circuit Cannington AG mill treating a medium-soft ore.
.

1000000 R4
R0 R1 R2 R3 R5

100000
Breakage Rate (hr-1)

10000

1000

100

10

1
0.01 0.10 1.00 10.00 100.00 1000.00
CDH4% Size (mm)CAN0%
ALB12% CDH13%

Figure 12 – Comparison of Cadia Hill, Cannington and Alumbrera breakage rate distributions using
upgraded AG/SAG model algorithm

To validate the upgraded model, the data from one of the surveys were used to fit the breakage
rates of the revised AG/SAG mill model at a particular mill load level. The model was then used to predict
the throughput and product size distribution of the mill at the other mill filling levels surveyed. This
exercise was then repeated but using the current JK AG/SAG model, which assumes the rates are fixed
regardless of the load level. This generated two sets of predictions called “old model” and “new model”.
The results comparing the response of the old and new models with the actual throughput and product P80
are given in Figure 13 to Figure 15 for the three data sets from Cannington, Cadia Hill and Alumbrera.

For Cannington, a further validation was carried out at the same mill load, by varying the ore
hardness and feed size distribution, to compare the new breakage model with the original average A and b
approach.

15
2500 2.5

2000 2.0

Product P80 (mm)


Feed Rate (TPH)

1500 1.5

1000 1.0

500 0.5

0 0.0
0 10 20 30 40 50 20 30 40 50
Mill Filling (% Vol) Mill Filling (% Vol)

New Model Old Model Survey New Model Old Model Survey

Figure 13 – Comparison of new and old model predictions against survey data for Cadia Hill SAG mill

450 0.140

400 0.130

0.120
350
Product P80 (mm)
Feed Rate (TPH)

0.110
300
0.100
250
0.090

200 0.080

150 0.070
20 25 30 35 40 45 20 25 30 35 40 45
Mill Filling (% Vol) Mill Filling (% Vol)

New Model Old Model Survey New Model Old Model Survey

Figure 14 – Comparison of new and old model predictions against survey data for Cannington AG mill

4500 1.5

4000
1.3
3500
1.1
3000
Product P80 (mm)
Feed Rate (TPH)

2500 0.9

2000 0.7
1500
0.5
1000
0.3
500

0 0.1
20 25 30 35 40 20 25 30 35 40
Mill Filling (% Vol) Mill Filling (% Vol)

New Model Old Model Survey New Model Old Model Survey

Figure 15 – Comparison of new and old model predictions against survey data for Alumbrera SAG mill

16
Cannington Case Study

The final validation presented considers the Cannington circuit treating a hypothetical coarser and
more competent feed. To determine whether the new breakage model gives reasonable results, the
Cannington mill was simulated with a hypothetical coarser and more competent ore at the same mill
filling/power. The simulation results are compared with the current model in Table 1. It can be seen that
both models predict a reduced throughput, due to the coarser and harder feed ore. However, the new size-
dependant breakage function suggests 16% more throughput than that predicted by the old model. The
question of which model prediction was more realistic, was addressed by comparing the historical
operating data, focusing on the final grind size versus specific power trend during the same period as the
surveys were conducted. Data published by Leung et al (1999) and information from JKMRC surveys in
2001 support the new model is a lot closer to reality, as shown in Figure 16. The simulation with the size-
averaged breakage function is off the trend line and likely to be in error.

Table 1 – Comparison of Cannington Simulation Results

Scenario Feed Mill Power F80 P80 Spec. Power Wio


TPH (kW) (mm) ( m) (kWh/t) (kWh/t)
Survey 280 4863 118 83.4 17.4 16.3
Old Model 205 4857 178 71.3 23.7 20.4
New Model 240 4858 178 78.2 19.6 17.7

140
130
Circuit P80 (microns)

120
110
100
90
80
70
60
10 15 20 25

Mill Specific Energy (kWh/t)

new model old model plant data

Figure 16 – Comparison of new and old model response against plant data for the Cannington AG mill

DISCUSSION

Though far from being finalized, the upgraded AG/SAG model has demonstrated some interesting
responses in the preliminary testing completed so far. The key findings include:

1. The new surface area based energy distribution, combined with the size sensitive appearance
function, has influenced the shape of the breakage rate distribution. As illustrated in Figure
12 previously, the familiar JKSimMet breakage rate curve shape doesn’t necessarily stay

17
constant for all configurations, which is not surprising given the findings from DEM, that
suggest low energy repetitive impact is the dominant breakage mechanism in AG/SAG mills
(Djordjevic, Shi and Morrison, 2004).

2. The slurry discharge vs hold-up relation interacts significantly with the new load/energy
calculation, which points to the need for an improved transport model that may avoid the
iterative solution currently employed. The revised transport model has to be sensitive to
both the mill load contents and design/operating conditions.

3. The proposed pebble discharge model is at best an interim fix until a more robust model can
be developed to describe this behaviour. An alternative model is being concurrently
developed and will be tested within the upgraded AG/SAG model algorithm as soon as it
becomes available.

4. The initial implementation of the mill filling sensitivity appears to be a significant


improvement over the current status; predicting different responses for each configuration. It
suggests that peak throughput occurs at a higher load, if the mill is close-circuited with a fine
classifier. However, more testing is required to fully understand the interaction of filling and
rates, and how the corrections should be implemented. The Impact and Attrition factors
extracted from the power model are on the right track, but more work is required to confirm
the most appropriate way to incorporate them in the upgraded model framework.

5. The impact of the new model structure is expected to affect the variable rates model format,
so the original regression equations developed by Morrell and Morrison (1996) are not
expected to survive in their current form into the new framework. A lot of the drivers
including speed, ball load and feed size are already directly incorporated in the new model.
The effect of ore hardness and circuit configuration may become significant.

CONCLUSIONS

The JKMRC has embarked on a comprehensive overhaul of the JK AG/SAG model, focusing on
several key changes designed to overcome long standing deficiencies and limitations. The work in
progress has commenced with the underlying key aspect of breakage, seeking to improve the way the
energy in the mill is handled and how the ore behaviour is described. Other improvements related to
breakage include the addition of a 6th rate knot at 75 microns, which should enhance simulation of fine
grinding AG/SAG mills, particularly with multi-component ores.

The upgraded model also includes a mill filling sensitivity which enables the throughput and size
reduction changes in AG/SAG mill performance to be predicted as the load volume in the mill is altered.
These changes are non-linear and are not well predicted by the current SAG mill model in certain cases.

An engineering model has been proposed to correct the discharge rate function, which currently
struggles to correctly simulate the change in pebble discharge when the pebble port open area is increased.
However, this model requires further testing before being fully deployed within the new framework.

Preliminary testing using survey data from Cannington, Cadia Hill and Alumbrera suggests the
upgraded model is working well, but mill filling sensitivity and pebble discharge models still need more
work. It appears the new energy/surface area approach is more robust than the current S20/EL approach,
but the final form of the rates correction, required to handle the throughput-load non-linearity, has yet to be
confirmed. The pebble discharge rate correction looks fine, but the implementation within the current
iterative solution can be unstable and hence needs more work.

18
The new model framework will need to be completed with the variable rates regression equations,
to take into account key operating conditions like ball charge, feed size, mill speed, etc. The form of these
regression equations is expected to change in light of the re-engineering of the power and breakage
components in the upgrade.

ACKNOWLEDGEMENTS

The research and development of the new AG/SAG model has been funded by The University of
Queensland, JKMRC and JKTech. The authors also wish to acknowledge the sponsors of the AMIRA
International P9 Project, particularly BHP-Billiton’s Cannington mine, Newcrest Cadia Hill and MIM
(Xstrata) Alumbrera for permission to use data.

REFERENCES

Ahlberg, J.H., Nilson, E.N. & Walsh, J.L. (1967). The Theory of Splines and their Applications,
Mathematics in Science and Engineering, Vol 38, Academic Press, New York and London.

Austin L.G., Klimpel R.R. & Luckie P.T. (1984). The process engineering of size reduction: ball milling.
AIME, New York

Djordjevic, N., Shi, F. & Morrison, R. (2004). Determination of lifter design, speed and filling effects in
AG mills by 3D DEM. Minerals Engineering, 17, 1135-1142.

Bailey, C. L., Lane, G., Morrell, S. & Staples, P. (2009). What Can Go Wrong in Comminution Circuit
Design? 10th Mill Operators Conference, AusIMM, Adelaide, South Australia, 143-149.

Leung K., Duffy M. & Andreatidis J. (1999). Autogenous grinding at Cannington. Proceeding of AMF
Conference Mineral Processing and Hydrometallurgy Plant Design, 61 – 70.

Michaux, S. P. & Djordjevic, N. (2010). Approximation of surface area of fines in blast induced
fragmentation, XXV International Mineral Processing Congress, Brisbane, Qld, Australia, (1029-1037). 6-
10 September, 2010.

Morrell S. & Morrison R. (1996). AG and SAG mill circuit selection and design by simulation.
International Conference on Autogenous and Semi-autogenous Grinding Technology, Vol. 2, Vancouver,
Canada, 769 - 790.

Morrell, S. & Stephenson, I. (1996). Slurry discharge capacity of autogenous and semi-autogenous mills
and the effect of grate design, Int Journal of Mineral Processing, 46(1-2): 53-72

Napier-Munn, T.J., Morrell, S., Morrison, R.D., & Kojovic, T. (1996). Mineral comminution circuits:
their operation and optimisation. ISBN 0 646 28861 x. Julius Kruttschnitt Mineral Research Centre.

Narayanan, S.S., & Whiten, W.J. (1988). Determination of comminution characteristics from single
particle breakage tests and its application to ball mill scale-up. Trans. Inst. Miner. Metall., 97, C115-C124.

Rumpf, H. (1973). Physical aspects of comminution and a new formulation of a Law of Comminution.
Powder Technology, 7, 145-159.

Shi, F. & Kojovic, T. (2007). Validation of a model for impact breakage incorporating particle size effect,
Int. Journal of Mineral Processing, 82, 156-163.

19
Vogel, L. & Peukert, W. (2004). Determination of material properties relevant to grinding by practicable
labscale milling tests, Int. Journal of Mineral Processing, 74S, 329-338.

Weibull, W. (1951). A statistical distribution function of wide applicability. Journal of Applied


Mechanics, 9, 293-297.

Whiten, W.J. (1974). A matrix theory of comminution machines. Chem Eng Sci, 29, 588-599

20

You might also like