VANCOUVER 2015

UPDATE ON THE
MODELING OF SEMIAUTOGENOUS GRINDING PROCESSES
IN A MOLY-COP TOOLS ENVIRONMENT

*J. E. Sepúlveda1, R. Muranda2 and J. Jofré2

1
Moly-Cop Group
Avda. Pedro de Valdivia 0168
Santiago, Chile
(*Corresponding author: [email protected] )
2
Moly-Cop Chile
Avda. Pedro de Valdivia 0168
Santiago, Chile

1|P age
VANCOUVER 2015

UPDATE ON THE
MODELING OF SEMIAUTOGENOUS GRINDING PROCESSES
IN A MOLY-COP TOOLS ENVIRONMENT

ABSTRACT

An extension of the well known steady-state phenomenological model for conventional ball mill
grinding processes to describe the more complex SAG grinding applications was first presented at the SAG
2001 Conference as well as included in the Moly-Cop Tools© software package. Over the last 15 years,
much knowledge has been gained as to the capabilities of the originally proposed model to describe actual
operating SAG processes, allowing for an improved understanding of the basic underlying mechanisms and
controlling factors affecting the performance of such systems. The purpose of the current publication is to
share these learnings with the international comminution community by presenting and discussing a series
of illustrative simulations and examples of applications of the revised model.

KEYWORDS

AG/SAG grinding, apparent charge density, breakage function, grinding media, mill power, model
©
calibration, Moly-Cop Tools , pebbles, rock competency, specific selection function, volumetric ball-to-
rock ratio.
INTRODUCTION

No doubt, the operational behavior of SAG milling systems is much more complex than that of
conventional ball milling circuits. To a great extent, this is due to the presence of two distinct types of
grinding media (rocks and balls) acting simultaneously in the mill charge. In these systems, a broad
diversity of critical conditions (both ore properties and operating variables) may influence the performance
of the production line. Due to the complex interactions between such conditions, it is nearly impossible to
empirically determine the independent, decoupled impact of each condition on the observed mill
performance; a highly critical step when searching for optimal processing conditions. An effective
approach to better understand the basics of such processes is via the intense use of computer simulations
based on a sufficiently detailed and representative mathematical model of the specific process under
evaluation, so best design and operational practices could be readily identified.

A mathematical model of such characteristics was first presented at the SAG 2001 Conference
(Sepulveda, 2001). The primary purpose of the current publication is to share the accumulated experience –
over the last 15 years – in the calibration and application of the model to actual systems, the necessary
model revisions which arose from these analyses, as well as to present and discuss a series of exploratory
‘what-if?’ simulations suggesting specific guidelines in order to optimize existing operations.

THEORETICAL FRAMEWORK

Since the details of the proposed mathematical model structure were already presented in the
referred previous publication (Sepulveda, 2001) only its main revisions will be discussed at this time.

SAG Model Structure

The SAG model in reference was derived as an extension of the widely accepted Linear Theory of
Comminution (Herbst et al., 1973; Gutierrez & Sepulveda, 1986). This theory introduced two new sets of
parameters: the Selection Function S and the Breakage Function B. The first set – also referred to as

2|P age
VANCOUVER 2015

Fractional Breakage Velocity or Grindability – relates to the grinding kinetics of each independent particle
and the second set – also referred to as Distribution of Primary Fragments – characterizes the size
distribution of the fragments produced as a result of breakage events.

In its differential form, with respect to the grinding time ‘t’, the model consists of a set of ‘n’
simultaneous equations of the type:
1
∂(fi)/∂t = - Si fi + Σ bij Sj fj (1)
j=i-1

where the subindex ‘i’ identifies a specific particle size fraction ‘fi’ amongst the ‘n’ fractions being
considered; i = 1 being the coarsest fraction. By definition, the size of all particles retained in fraction ‘i’
ranges from di to di+1.

The analytical solution of this complex system of differential equations is known, under the
simplifying assumption that parameters S and B are invariant with time; so giving rise to the particular
solution of the general system denoted the ‘linear solution’, which in its matrix form may be expressed as:

f = ( T J T-1 ) f° (2)
where:
f = {fi | i = 1, 2, …, n} = vector containing the size distribution of the mill discharge (ground
product)
f° = {fi° | i = 1, 2, …, n} = vector containing the size distribution of the mill feed
T = {Tij | i,j = 1, 2, …, n} = lower triangular matrix of values Tij recursively defined as:

Tij = 0 ; when i < j

Tij = 1 ; when i = j
i-1
Tij = Σ ( bik Sk Tkj )/( Si – Sj ) ; when i > j
k=j

J = {Jij | i,j = 1, 2, …, n} = diagonal matrix of values Jij defined as:

Jij = exp (-Sit) ; when i = j (batch grinding)
Jij = (1 + Siτ/N)-N ; when i = j (continuous grinding)
Jij = 0 ; in all other cases
where τ – the average mean residence time in the mill charge – and N are parameters
characteristic of the Residence Time Distribution (RTD) of the mineral slurry in the mill,
represented by (Herbst et al., 1973):

E(t) = [ NN (t/τ)N-1 / τ Γ(N) ] exp (-Nt/τ) (3)

referred to as the ‘N-Mixers in Series Model’ and where the parameter N may be approximated by
a fairly simple empirical correlation (Rajamani & Lo, 1990).
The critical role of the specific energy input (kWh/ton) in comminution processes is duly reflected
by the introduction of the Specific Selection Function parameter, defined as:

SiE = Si (H/P) ; i = 1, 2, ..., n (4)

where H represents the total hold-up of ore in the mill charge and P represents either the power drawn by
the ball charge (Pballs) or the power drawn by the rocks (Procks) as derived from the modified Hogg &
Fuerstenau Model (Sepulveda, 2001; Hogg & Fuerstenau, 1972) for the estimation of the total power draw
of a given tumbling mill:

3|P age
VANCOUVER 2015

Pnet = η Pgross = 0.238 D3.5 (L/D) Nc ρap (J - 1.065 J2) sinα (5)

where:
Pgross = Gross power draw (kW) = Pnet / η
η = Overall electrical and mechanical transmission efficiency, °/1
D = Effective mill diameter, ft
L = Effective mill length, ft
Nc = Rotational speed; expressed as a fraction (°/1) of the critical speed : Ncrit = 76.6/D0.5
J = Apparent charge level, °/1 (including balls, rocks, slurry and the interstitial voids in
between the balls and the rocks)
α = Lifting angle [defines dynamic position of the center of gravity of the load (‘kidney’)
with respect to the vertical reference].
and where ρap denotes the Apparent Charge Density (that is, the tons of charge in the mill for each m3 of
volume apparently occupied), which may be evaluated on the basis of its 3 components (balls, rocks and
slurry):

ρap = { (1-fv) ρb Jb + (1-fv) ρm (J - Jb) + ρp Jp fv J } / J (6)

with:
fv = Volumetric fraction (°/1) of interstitial voids in between the balls and the rocks
Jb = Apparent ball filling (°/1)
Jp = Interstitial slurry filling (°/1)
ρm = Ore density, ton/m3
ρp = Slurry density, ton/m3; directly related to the weight fraction of solids (fs) in the slurry,
by the expression: 1/[(fs/ρm) + (1-fs)].

Equation 6 above reveals the contribution of each mill charge constituent to the Apparent Charge
Density and hence, to the total Mill Power Draw, characterized by Equation 5, as follows:

Pballs = [(1-fv) ρb Jb / ρap J] · (Pnet) (7)

Procks = [(1-fv) ρm (J – Jb) / ρap J] · (Pnet) (8)

Rearranging Equation 4 above, in terms of the Specific Energy Input ‘E’ (kWh/ton), it is demonstrated
that:
SiE E = Si t (batch grinding)
SiE E = Si τ (continuous grinding) (9)

Therefore, by replacing the product (Si t) or (Si τ) by (SiE E) in Equations 2 above, the solution of
the Linear Grinding Model is then expressed in terms of the Specific Selection Function, SiE, so
recognizing the critical role of the Specific Energy Input in comminution processes.

A commonly accepted assumption is that not all particles are broken at the same rate but all
particles in the same size fraction (di to di+1) do break at practically equal rates. The particle size
dependence of the SiE grindability rates may be characterized by expressions of the type (Austin et al.,
1984):

SiE = α0 (di*)α1 / [ 1 + (di*/dcrit)α2] (10)

where:

4|P age
VANCOUVER 2015

di* = Average size of the particles retained in size fraction ‘i’.

dcrit = Adjustable parameter representative of the critical particle size above which
grindability rates gradually decrease for coarser and coarser particles.
α0 = Adjustable parameter representative of the grindability rate of the very fine particles
(di* approaching 1 µm)
α1 = Adjustable parameter representative of the slope of the log(SiE) vs. log(di*) plot for
particle sizes well below the dcrit size.
α2 = Adjustable parameter representative of the slope of the log(SiE) vs. log(di*) plot for
particle sizes well above the dcrit size.

Primary Breakage Mechanisms in SAG Systems

The key SAG model building assumption is that at least 3 different size reduction mechanisms are
simultaneously and independently acting in the mill charge:

• Balls acting on Particles, similar to conventional ball mills, where the size dependence
of the Selection Function, SiE, may be characterized by:

(SiE)balls = α0balls (di*)α1balls / [ 1 + (di*/dcritballs)α2balls] (11)

• Rocks acting on Particles, serving their role as autogenous grinding media,

characterized by:

(SiE)rocks = α0rocks (di*)α1rocks / [ 1 + (di*/dcritrocks)α2rocks] (12)

• Self-breakage of the Rocks, as a result of their own tumbling action in the mill,
characterized by:

(SiE)self = α0self (di*)α1self (13)

In recognition to the controlling role of the Specific Energy Input (kWh/ton) in grinding kinetics,
the current development postulates that the effectiveness of the balls is consequently proportional to the
power they absorb (Equation 7), the effectiveness of the rocks as grinding media is similarly proportional
to the power they absorb (Equation 8) and also that, these same rocks self-break as a function of the power
they absorb (Equation 8). Hence, there would be an Overall Selection Function defined as the weighted
average of the independent contributions of each of the indicated 3 basic breakage mechanisms, according
to the expression:

[SiE]overallE = [SiE]balls Eballs + [SiE]rocks Erocks + [SiE]self-breakage Erocks (14)

Figure 1 provides an illustration of the relative magnitude of the 3 components of the Overall
Specific Selection Function. It can be observed that, in comparison to balls, the rocks are expected to show
much lower energy efficiency as grinding media (α0rocks < α0balls and dcritrocks <dcritballs) and that these rocks
self-break faster for larger and larger sizes (dcritself → ∞) since it is reasonable to assume that larger rocks
would contain a larger population of critical internal cracks. The continuous red curve represents the
weighted average resulting from the application of Equation 14 and reveals the existence of the so called
‘critical particle size’ (pebbles) in the range of 2” to 4”, characterized by slower grinding rates that create
an accumulation of these sizes in the mill load, at steady state.

5|P age
VANCOUVER 2015

1
Balls on Particles
Rocks on Particles

Specific Selection Function, ton/kWh

Self Breakage
Overall

Rocks as Grinding Media

0.1
α 0 (rocks)/α 0 (balls) Rock Competency
α 0 (self)/α 0 (balls)

0.01
100 1000 10000 100000 1000000

Particle Size, microns

Figure 1 - Example of an Overall Specific Selection Function, obtained as the energy-weighted average of
the respective grindability rates associated to each of the 3 independent breakage mechanisms

As part of the efforts to reduce the number of independent parameters affecting the response of
the model, actual data analyses – as those discussed in the next section – have shown it is possible to
assume that all slopes in the finer particle range (well below dcrit) would be of similar magnitude; that is:

α1balls ≅ α1rocks ≅ α1self (15)

Under this simplifying assumption, the lines above become parallel in the finer particle range
(well below dcrit) and the following two relevant parameter ratios, regarding the actual role of rocks in SAG
mills, can be observed:
• Rocks as Grinding Media. A lower value of the α0rocks/α0balls ratio indicates that the
rocks are not as effective as balls in their role as grinding media. As a general rule, this
ratio should be proportional to the ore/ball density ratio and the characteristic feed size of
ore, hence:

α0rocks/α0balls = dcritrocks/dcritballs = ρm/ρb * (F80/127000) (16)

indicating that lighter and smaller feed rocks would be significantly less effective as
grinding media.

• Rock Competency. A lower value of the α0self/α0balls ratio indicates that the rocks are
more capable of sustaining impact breakage (more competent) in the mill charge; that is,
rocks ‘self-break’ at reduced rates. This ratio has been observed to vary for different ore
types in the range of 0.05 (high competency) to 0.15 (low competency).

With regard to the Breakage Function, Bij, it has been customarily assumed to conform to a family
of non-normalizable breakage functions described by:

Bij = β0j (di /dchips,j)β1 + (1 - β0j)(di /dj+1)β2 , if di < dchips,j

β2
Bij = β0j + (1 - β0j)(di /dj+1) , if di > dchips,j (17)

6|P age
VANCOUVER 2015

with:

dchips,j = β3j (dj+1)

β0j = β00 (dj+1 /d1)-β01 , always ≤ 1
β3j = β30 (dj+1 /d1) -β31
, always ≤ 1

The motivation for originally proposing this expanded, more complex form of the Breakage
Function arose from the empirical observations that the larger rocks break into just a few large pieces and a
minor proportion of smaller fragments or ‘chips’, without generating much intermediate sizes. Figure 2
illustrates an example of computations with Equations 17. The absence of intermediate size fragments can
be clearly observed for the coarser rocks (Fraction 1), while the smaller particles (Fraction 10) break
normally, as in conventional ball mills.

1.0
Breakage Function, B ij

0.1

Fraction 1
Fraction 3
Fraction 6
Fraction 10
0.0
10 100 1000 10000 100000
Particle Size, microns

Figure 2 - Primary fragment size distributions (Breakage Function), for various size fractions ‘j’

Further, it has been conveniently assumed that all particles are broken according to the same
breakage pattern depending only on the particular ore properties; therefore, the same Bij is assumed to
apply for all breakage mechanisms originating the fragments.

MATHEMATICAL MODEL CALIBRATION

The mathematical model described above has been incorporated to the Moly-Cop Tools© 3.0
software package (Sepulveda, 2012). But, before any given mathematical model may be considered
‘useful’, its applicability to actual systems must be repeatedly demonstrated by ‘tuning’ or ‘calibrating’ the
model responses to actual experimental data, ideally obtained at both pilot and full industrial scales.
Unfortunately, the complex nature of AG/SAG milling applications precludes any testing arrangements
that could be implemented at a more convenient, smaller laboratory scale.

The Parameter Estimation Problem Statement

In those applications where experimental SAG milling data – at any operational scale – is made
available, it is then possible to identify the best set of intrinsic adjustable parameters defined above (α’s,
β’s, ...) that can best provide a mathematical description of the observed mill response. For such purpose,

7|P age
VANCOUVER 2015

non-linear regression algorithms, such as those included in the Excel’s SOLVER subroutine (Moly-Cop
Tools’ SAGParam applications), may be conveniently applied to minimize the objective function:
min φ = Σi (Fi - Fi*)2 (18)

In this expression, Fi represents the observed or measured ‘% Passing the ith Screen’ in the mill
discharge stream and Fi* represents the corresponding model fitted value.

Like all search algorithms, calculations must start from a set initial guesses for each relevant
model parameter. Figure 3 provides an example of the Control Panel sheet of the SAGParam application
which – as expected – shows that the initial guesses provided were not ideal and the objective function is
unacceptably large.

Figure 3 - Definition of the set of initial guesses for all relevant adjustable parameters that determine the
mathematical model response

After the SOLVER subroutine is properly executed, the minimization algorithm finds a new set of
parameter values which minimize the established objective function, as illustrated in Figure 4.

8|P age
VANCOUVER 2015

Figure 4 - Final adjustable parameter values, obtained after executing subroutine SOLVER, that best
described the reference experimental data

Parameter Reduction Options

An important feature that greatly contributes to make a mathematical model ‘useful’ is that the
total number of independent adjustable parameters is as low as possible.

As indicated in Equations 15 and 16 above, the following parameter reduction assumptions are
considered acceptable in most actual applications:

α1balls ≅ α1rocks ≅ α1self (15)

α0rocks/α0balls = dcritrocks/dcritballs = ρm/ρb * (F80/127000) (16)

With respect to the assumed Breakage Function (Bij) form, most model calibration exercises have
been satisfactorily accomplished under the special case:

β01 = 0
β30 = β31 = 1 (19)

In this particular case, the set of Equations 17 above reduce to the more familiar normalizable Breakage
Function (Herbst et al., 1973; Austin et al., 1984; Gutierrez & Sepulveda, 1986):

Bij = β00 (di /dj+1)β1 + (1 - β00)(di /dj+1)β2 , for all di < dj+1 (20)

typical of conventional ball milling applications.

All of these simplifying assumptions allow for the reduction of truly significant parameters down
to a maximum of 6 parameters, namely: α0balls, α1balls, dcritballs, β00, β1 and α0self, which is to be compared
with 5 parameters for ball mill grinding.

9|P age
VANCOUVER 2015

Mathematical Model Capability

Over the last 15 years, since the first publication of the current model (Sepulveda, 2001), a large
number of actual case studies have been made available to these authors, mostly from pilot scale
evaluations, considering the inheriting difficulties of conducting representative sampling campaigns at full
industrial scale.

These various data sets have allowed gaining trust on the capability of the model to satisfactorily
describe actual operating systems. As an example, Figure 5 illustrates the comparison of actual vs. model
fitted values for SAG mill throughput and product size distributions, based on pilot scale data obtained
with ore samples from three major SAG milling projects.

100
12.0 Concentrator A 90 Concentrator A

% Passing in Mill Discharge, Fitted

Concentrator B 80 Concentrator B
10.0
Concentrator C 70 Concentrator C
tph, Model Fitted

8.0
60

50
6.0
40

4.0 30

20
2.0 ± 10 % error bands
10 ± 10 % error bands
0.0 0
0.0 2.0 4.0 6.0 8.0 10.0 12.0 0 10 20 30 40 50 60 70 80 90 100
tons/hr, Experimental % Passing in Mill Discharge, Experimental

Figure 5 - Comparison of actual vs. model fitted values characterizing the SAG mill response of various
ores tested in a typical 6’ x 2’ pilot SAG mill

Important to highlight that all model fitted values presented in Figure 5 above were generated
from a single set of parameters correspondingly representative of each the various ore samples tested.

EXPLORATORY ‘WHAT-IF?’ SIMULATIONS

As indicated previously, due to the complex interactions between ore parameters and operating
conditions, it is nearly impossible to empirically determine the independent, decoupled impact of each
process variable or ore parameter on the observed mill performance. The availability of a reliable
mathematical model, duly incorporated into a functional simulator, opens valuable opportunities to
systematically explore for improved process performance.

For the purpose of illustration of the proposed optimization methodology, a nominal reference
case was defined, as shown in Figure 6 which corresponds to a typical 36’f x 17’ SAG mill in close circuit
with pebble removal, crushing and recycling (SABC-1 Configuration), operating at close to 920 ton/hr.

10 | P a g e
VANCOUVER 2015

Figure 6 - Basic design and operating conditions defined for exploratory simulation purposes

Ball-to-Rock Ratio Effect

Figure 7 provides an example of the anticipated effect of the Ball-to-Rock ratio (Apparent Charge
Density, see Equation 6) on circuit capacity, indicating there would be an optimal Apparent Charge
Density that would maximize the capacity developed by the SAG mill, in this case 4.18 ton/m3, resulting
from operating with 17.4% Ball Filling and 26% Total Charge, both apparent.

1000 6.0
950

Apparent Charge Density, ton/m3

900 5.5
850
Grinding Capacity, ton/hr

800 5.0
750
700 4.5
650
600 4.0
550
500 3.5
450
400 3.0
6.0 8.0 10.0 12.0 14.0 16.0 18.0 20.0 22.0 24.0

Ball Filling, %

Figure 7 - There exist an optimal Apparent Charge Density that maximizes mill throughput

Worth noting that in some cases, specially for fine feed ores, such optimal % Ball Filling could
well exceed maximum acceptable levels of 20% or less.

Once the optimal Apparent Charge Density has been identified, Mill Throughput could be
maximized by increasing the Total Mill Charge level, as shown in Figure 8.

11 | P a g e
VANCOUVER 2015

1100 14000
1050

Grinding Capacity, ton/hr

1000 13000
950

Mill Power, kW (net)

900 12000
850
800 11000
750
700 10000
650
600 9000
22.0 24.0 26.0 28.0 30.0 32.0

Total Mill Filling, %

Figure 8 - Grinding Capacity maximization at constant, optimal Apparent Charge Density

Feed Size Effect

Paradoxically, from a fully autogenous perspective, grinding capacity increases as finer particles
(i. e. less rocks = autogenous media) are fed to the mill, in due reflection of the poor performance of the
rocks as autogenous grinding media, as illustrated in Figure 9.

2000
1800 F80 Feed Size
1600 2.0 in
Grinding Capacity, ton/hr

1400 2.5 in
1200
3.0 in
1000
4.0 in
800
600 6.0 in

400
200
0
6.0 8.0 10.0 12.0 14.0 16.0 18.0 20.0 22.0 24.0 26.0 28.0

Ball Filling, %

Figure 9 - Expected Grinding Capacity improvement as a result of reducing the presence of large rocks in
the circuit feed

Also notice that, if finer and finer feed materials are processed, the model suggests that ball filling
should be gradually increased, as shown by the dotted line. There would be an inverse link between the
Feed F80 and the optimal charge density which maximizes mill capacity; indicating that higher Ball-to-
Rock ratios should be maintained for finer feed ores, as shown in Figure 10. This dependence is a function
of the Rock Competency (i.e. their ability to sustain impact without breakage) characterized by the
α0self/α0balls ratio.

12 | P a g e
VANCOUVER 2015

200

180 Rock Competency >>> Actual Operations

160

F80 Feed Size, mm

140 Not enough
balls!
120 Too many
balls!
100

80

60

40
3.4 3.6 3.8 4.0 4.2 4.4 4.6 4.8

Charge Density, ton/m3

Figure 10 - Finer feed demand for higher Apparent Charge Densities; that is, higher Ball Filling levels

The previous simulations are consistent with actual plant data collected via informal interviews to
various site contacts, as represented by the open circles in Figure 10 above. If they were to obey the trend
indicated by the model, some of these operations (those shown at the lower-left in the chart) would benefit
from increasing the ball level in the mill.

From a different perspective, if large rocks needed to be present in the charge, then enough room
should be provided for them to absorb energy and grind themselves; hence, lower ball filling and vice versa.

Discharge Grate Opening Effect

Another option to improve mill capacity is to increase the opening in the discharge grate slots,
whenever possible, as illustrated in Figure 11, where benefits in the range of 5-20% would be attainable,
depending on grate opening and feed fineness.

1600 5200
3.0" Grate 3.0" Grate
2.5" Grate 2.5" Grate
1400
Transfer Size, T80, microns

2.0" Grate 5000 2.0" Grate

Grinding Capacity, tph

1/2" Grate 1/2" Grate

1200
4800
1000

4600
800
16% Ball Filling 16% Ball Filling
600 4400
20 30 40 50 60 70 80 90 20 30 40 50 60 70 80 90

Feed Fineness, % -2" Feed Fineness, % -2"

Figure 11 - Anticipated effect of the Grate Discharge Opening on SAG mill response

It should not be overlooked that opening the grates may allow for not only higher grinding
capacity but also a coarser T80 transfer size to the secondary ball mills. A larger proportion of the overall
grinding task is left for the ball mills – more tons of coarser feed – to the extent that many operations have
been forced to add a second, a third and even a fourth ball mill in order to cope with the additional circuit
capacity so generated. Also, the proportion of pebbles being crushed and recycled increases as the grate
opening increases, imposing additional demand on the crusher(s).

13 | P a g e
VANCOUVER 2015

Screen Opening Effect

Similarly, increasing the opening of the SAG Discharge screen increases SAG mill capacity at the
expense of a further overload of the secondary ball mill(s), as illustrated in Figure12.

1600 8000
3/8" Screen
7000
1/2" Screen
1400

Transfer Size, T80, microns

3/4" Screen 6000
Grinding Capacity, tph

1200 5000

4000
1000 3000

2000 3/4" Screen

800 1/2" Screen
16% Ball Filling 1000 16% Ball Filling 3/8" Screen
600 0
20 30 40 50 60 70 80 90 20 30 40 50 60 70 80 90

Feed Fineness, % -2" Feed Fineness, % -2"

Figure 11 - Anticipated effect of the Grate Discharge Opening on SAG mill response

The observations arising from the set of simulations here discussed are consistent with global
trends in the operation of modern SAG milling circuits. In fact, since AG grinding technologies were first
proposed, early last century, there has been a continuous evolution in operational practices with regard to:

• The addition of increasing amounts of balls as ancillary grinding media,

• The sustained increment in the diameter of such balls,
• The removal and crushing of the critical sizes (pebbles) that otherwise would accumulate in the load
and …
• The pre-crushing (elimination) of either the larger rocks or the intermediate particle size fractions
contained in the fresh feed ore.

Consequently, little is left today of the original intention of using the larger rocks as autogenous
grinding media for the smaller particles.

FINAL REMARKS

The mathematical model here proposed – updated from its original publication at the SAG 2001
©
Conference – has been incorporated to a variety of Moly-Cop Tools 3.0 applications covering single and
multistage flowsheets in combination with ball mills, screens and crushers.

The model is capable of providing descriptions of actual systems in great detail. However, its
scalability from pilot to full industrial operations needs to be further demonstrated, keeping in mind that
gathering accurate operating data around large scale SAG mills has become a serious, almost
unsurpassable, challenge.

REFERENCES

Austin, L. G., Klimpel, R. R. and Luckie, P. T. (1984). Process Engineering of Size Reduction: Ball
Milling. SME of AIME Publication.

Gutiérrez, L. & Sepúlveda, J.E. (1986). Dimensionamiento y Optimización de Plantas Concentradoras

Mediante Técnicas de Modelación Matemática. CIMM-Chile Publication.

Herbst, J. A. et al. (1973). Population Balance Models for the Design of Continuous Grinding Mills.
Proceedings of the 10th IMPC, p. 23, London

14 | P a g e
VANCOUVER 2015

Hogg, R. & Fuerstenau, D. W. (1972). Power Relations for Tumbling Mills. Trans. SME-AIME, Vol. 252,
pp. 418-432.

Rajamani, K. & Lo, Y. C. (1990) Personal communication.

Sepúlveda, J. E. (2001). A Phenomenological Model of Semiautogenous Grinding Processes in a Moly-

Cop Tools Environment. Proceedings of the SAG2001 Conference, Vancouver, Canada.
©
Sepulveda, J. E. (2012). Moly-Cop Tools , Version 3.0: Software for the assessment and optimization of
grinding circuit performance. Available upon request at [email protected]

15 | P a g e