Mine stockpile design to minimise

environmental impact
J. E. Everett
Faculty of Economics and Commerce,
The University of Western Australia

Abstract
In Western Australia, iron ore is extracted from inland mines and railed to a port,
where it is loaded onto ships for export. Quality depends upon uniform
composition (“grade control”), not only in iron, but also in several contaminant
minerals. To achieve grade control, and to provide a buffer between production
and demand, crushed ore is stored on to large stockpiles and then reclaimed,
either at the mine or at the port. Environmental impact (and cost) is reduced if
the land area devoted to stockpiles can be reduced without loss of grade control.
There may also be environmental benefits in building the stockpiles at the mine
rather than at the port. The stockpile array can be considered as a low-pass filter,
filtering out short-term fluctuations in composition. Techniques considered
include the use of multiple build and/or reclaim stockpiles, with and without
intelligent stacking and reclaiming. Because of the complex correlations between
the minerals and across time, design of the stockpile array is not readily
amenable to mathematical analysis. Simulation using historic data provides a
useful insight for comparing available stockpile policies. A simulation model has
been constructed to aid the mine planner choose between alternative stockpile
policies. The model has been constructed using VBA in Excel, to provide ease of
use and portability.
Keywords: quality control, decision support, mining, stockpile design, intelligent
stacking/reclaiming, forecasting, simulation.

1 Introduction
Iron ore from inland mines is railed to port, where it is crushed, separated into
lump and fines products and stored onto large stockpiles. Ships are loaded for
export with ore reclaimed from the stockpiles.

Development and Application of Computer Techniques to Environmental Studies X

G. Latini, G. Passerini & C. A. Brebbia (Eds) © 2004 WIT Press, www.witpress.com, ISBN 1-85312-718-3
54 Development and Application of Computer Techniques to Environmental Studies X

The exported ore is destined for blast furnaces, whose efficient operation
depends upon the ore being of consistent composition, not only in iron but also
in the major contaminants, including silica, alumina and phosphorus. The quality
of the ore therefore depends upon each shipment closely matching the target
composition vector.
Putting the ore onto stockpiles before shipment serves a double purpose.
The stockpiles provide a quantity buffer between the arrival of ore trains from
the mine and the more irregular departure of ships. But of just as much
importance is the effect of stockpiles in blending out variations in ore
composition.
Typically, stockpiles are of the order of 100 kilotonnes when full, with a
footprint about 30 metres wide and 100 metres long. Stockpiles are built in a
“chevron” pattern, with the stacker running back and forth the whole length of
the pile. The stockpiles are built to completion, so that reclaiming does not start
until stacking is finished. Reclaiming is from one end, in three “pilgrim steps”,
so any shipment contains closely equal portions taken from the top, middle and
bottom of the pile.
The method of stacking and reclaiming is designed to ensure that the ore
stacked to and then reclaimed from a stockpile is effectively uniform in
composition. Studies comparing assays of stacked ore and reclaimed ore have
shown that this goal is achieved, at least to within the errors of assay
measurement. Further improvement in quality control therefore depends on
controlling the variability in composition between successive stockpiles.
The port yard comprises an array of stockpile sites. If ore is stacked on
arrival to a single stockpile, which is built to completion, then the larger the
stockpile the more composition variability is smoothed out.
There may also be a quality benefit in having multiple completed stockpiles
available at any one time, so further quality improvement can be achieved by
loading ships with ore blended from multiple stockpiles, but the present paper
does not consider this.
Increasing the size and number of stockpile sites comes at considerable
economic and environmental cost. Coastal land in the vicinity of the port is
mainly mangrove swamp, part of a complex and sensitive ecosystem. It is
desirable to minimise the disturbance to the ecosystem by designing a stockpile
system able to give adequate quality control with as small a stockpile footprint as
possible.
Pitard [1] provides a thorough study of stockpile procedures for mining.
Everett [2,3] discusses methods of iron ore quality control. Kamperman et al [4]
describe application of these methods to a West Australian iron ore mining
operation.

2 Scope
This paper describes a computer-based simulation model to aid the design of a
stockpile system, based upon real data, to give improved quality control while
minimising the stockpile footprint. The model design and development fully

Development and Application of Computer Techniques to Environmental Studies X

G. Latini, G. Passerini & C. A. Brebbia (Eds) © 2004 WIT Press, www.witpress.com, ISBN 1-85312-718-3
 Development and Application of Computer Techniques to Environmental Studies X 55

involved the mine operators and planners, both to incorporate their domain
knowledge and to gain the cooperation and enthusiasm. To provide them with a
“fishing line rather than a fish”, a simulation model was built which they could
use to explore alternative scenarios. To avoid software licensing fees and the
costs of user acceptance and training, a dedicated simulation package was not
used. Instead, the simulation model was built using Visual Basic (VBA) macros
in Excel. Since staff were accustomed to using Excel spreadsheets, they quickly
felt comfortable using a simulation model that had the appearance of an Excel
workbook, with menus and buttons to drive the macros.

3 Quality control
Iron ore quality requires each shipment of ore to match target composition as
closely as possible. The target composition is a vector of four components {iron,
silica, alumina, phosphorus}, so there are four objective functions, one for each
mineral. The customer’s objective function may be skewed: for example, they
may tolerate iron above target or silica and alumina below target more happily
than the opposite. But, for the producer, to deliver ore richer than target is an
opportunity cost. So we must treat the objective function as symmetric.
For each mineral, we can identify a tolerance level. If the composition
departs from target by one tolerance unit in any mineral, the customer is equally
unhappy. We can therefore define a dimensionless stress in each mineral, as the
departure from target divided by the tolerance. Squaring the component stresses
and adding then together defines a total stress, which provides a combined
objective function to be minimised. This quadratic objective function, together
with a set of linear constraints, is used to select the ore mined from the available
sources within the open-cut mine pit. It is a classic quadratic programming
problem (see Taha [5]), although for convenience we use the heuristic optimiser
“Solver”, supplied with Excel, for making the daily mining decisions.
Variations in the composition of the mined ore have long, medium and
short-term frequency components, which we need to filter out. Monthly mine
plans are designed to remove long-term variations. The daily mine plan should
smooth medium-term variations. Building and reclaiming stockpiles, the subject
of this study, serves to filter out short-term variations, with period up to the
capacity of the stockpile yard.
The monthly mine plan, daily schedule and stockpile building decisions are
all further complicated by the problem that we do not know the “true”
composition of any ore until it has been mined, railed to port, crushed, sampled
and assayed. All the operational decisions as to the mining and stockpiling of ore
have to be made before its assay results are known. The mine planning decisions
have to be made using coarsely spaced data from exploratory drill holes. Daily
scheduling decisions can use more detailed estimates from assays of material
collected during the drilling of blast holes prior to the ore blocks being broken up
ready for mining. Even the blast hole estimates have considerable random error
and bias compared to the final assays from the mined and crushed ore.

Development and Application of Computer Techniques to Environmental Studies X

G. Latini, G. Passerini & C. A. Brebbia (Eds) © 2004 WIT Press, www.witpress.com, ISBN 1-85312-718-3
56 Development and Application of Computer Techniques to Environmental Studies X

To improve the information for daily scheduling and stockpiling decisions,

we adjust the blast-hole estimates by a regression model, and add a bias
estimated from a forecast based on the exponentially smoothed difference
between the port assays and the blast hole estimates of recently mined ore.
Hanke and Reitsch [6] discuss forecasting methods using exponential smoothing.

4 Building stockpiles
4.1 Assay Variability

We have seen that, within assay measurement error, ore stacked to and recovered
from a stockpile can be considered uniform. Our goal is to minimise the
variation in composition between stockpiles.
The assay data unit is a trainload (or more strictly a rake, or half-train) of
about 10 kilotonnes. Trainloads are stacked to build stockpiles that can be from
about 50 to 200 kilotonnes. If the composition of successive trainloads were
uncorrelated, and stockpiles were built to completion from successive trainloads,
then the variation in stockpile composition would be inversely proportional to
the square root of the stockpile size.
However, not only do the assay data show strong correlations between
minerals (silica and alumina are positively correlated with each other, and
negatively with iron), but they also have strong serial correlations, usually
positive. Because of the serial correlation, variability between stockpiles falls off
more slowly than the inverse square root of stockpile size.
The complicated correlations, between minerals and across time, make it
unwise to do any planning based upon synthetic data. Any modelling we do is
based on historical data. Planning for a new operation uses data from some other
operation that is geologically matched as closely as possible.

4.2 Stacking policies

Our purpose is to design a stockpile system that gives adequate control of

variability while using as small a total stockpile footprint as possible.
Building larger stockpiles can improve the variability, but at the cost of
increasing the stockpile footprint. For example, to halve the assay variability
would require that the stockpile footprint be increased more than four-fold.
As an alternative to increasing stockpile size, we can stack each trainload to
a destination chosen from multiple current stockpiles, as shown in Figure 1. The
choice of stockpile can be random, sequential or “intelligent”.
The incoming trainload has weight w and composition stress vector x. The
stress vector has components {xk}, for minerals k=1 .. K. Each component of the
stress vector is defined as the departure from target divided by the tolerance for
that mineral.
The nth stockpile (n=1,2, ... N) has weight Wn, and the stress vector
Xn={Xnk}.

Development and Application of Computer Techniques to Environmental Studies X

G. Latini, G. Passerini & C. A. Brebbia (Eds) © 2004 WIT Press, www.witpress.com, ISBN 1-85312-718-3
 Development and Application of Computer Techniques to Environmental Studies X 57

Candidate Stockpiles Stress

Incoming X1 = {X1k}
W1
Trainload

w W2 X2 = {X2k}

stress x = {xk} X3 = {X3k}

W3

Figure 1: Stacking to multiple candidate stockpiles.

4.3 Sequential or random stacking

With sequential stacking, each trainload is sent in turn to each of the stockpiles
currently being built. Random stacking is similar, but the destination choice is
either truly random, or based on operational convenience (such as interaction of
the stacker with other equipment).
If the composition of successive trainloads has no serial correlation,
sequential or random stacking provides no benefit over stacking a single
stockpile to completion. However, if they are serially correlated (equivalent to
periodic variation), then sequential or random stacking removes some of the
periodic variation by spreading each stockpile over a longer production period.

4.4 Intelligent stacking

Intelligent stacking uses the composition of the incoming ore, or an estimate of

its composition, to decide the destination stockpile.
If we know the tonnages and estimated compositions of the incoming
trainload and of the candidate stockpiles being built, we can send the trainload to
the destination chosen to minimise the total stress of the stockpile yard.
As in Figure 1, the incoming trainload has weight w and composition stress
vector x. The stress vector has components {xk}, for minerals k=1 .. K. The total
weighted stress “s” of the trainload is given by:

s = wx.wx= w2x.x = w2∑xk2 (1)

Similarly, the nth stockpile, with weight Wn and stress vector Xn, has weighted
stress Sn = Wn2Xn.Xn. So the current total weighted stress of the stockpiles is:

∑Sn = ∑Wn2Xn.Xn = ∑Wn2∑Xnk2 (2)

If we add the trainload to the nth stockpile, its weighted stress will increase by:

Development and Application of Computer Techniques to Environmental Studies X

G. Latini, G. Passerini & C. A. Brebbia (Eds) © 2004 WIT Press, www.witpress.com, ISBN 1-85312-718-3
58 Development and Application of Computer Techniques to Environmental Studies X

δSn = (w+Wn)2∑[(wxk+WnXnk)/(w+Wn)]2 - Wn2∑Xnk2

= 2wWn∑xkXnk+w2∑xk2 = 2wWnx.Xn +w2∑xk2 (3)

Thus, to minimise the resulting total weighted stress, the trainload should be sent
to the stockpile giving the smallest value for Wnx.Xn (the stockpile weight
multiplied by the scalar product of the trainload and stockpile stress vectors).

4.5 Estimating train composition

Intelligent stacking, as described above, depends upon knowledge of the

composition of the incoming ore. As discussed, at the time the stacking decision
has to be made the assay results for the trainload are not yet available.
So in place of the true stress vector x we have to use an estimated x*. For
intelligent stacking, we now send the trainload sent to the stockpile giving the
smallest value for Wnx*.Xn
To estimate x*, we take the blast hole estimates of composition for the
mined ore that went on to the train, regress these estimates towards the mean and
then add a bias. The bias is the exponentially smoothed error of recent assays.
The regression and exponential smoothing coefficients are established by
analysis of previous blast hole and port assays, and reviewed every few months.
In the limit, if the estimated stress vector x* is uncorrelated with the true
stress x, intelligent stacking becomes random stacking. In the simulation studies
to be described, we investigated the effect of progressively blurring the estimated
stress, from perfect knowledge, through the accuracy of available estimates,
down to purely random assignment.

4.6 Stockpile size, number and footprint

Our goal is to limit the area devoted to stockpiles, while still filtering out
composition variation for ore stacked and reclaimed through the stockpile array.
Stockpiles are built to a specific height and width, constrained by the
stacking and reclaiming equipment. The length of a stockpile footprint is
therefore proportionate to the stockpile capacity, and a partly built stockpile still
needs to have the full footprint dedicated to it. Ignoring the comparatively small
space lost between stockpiles, the area devoted to stockpiles is proportional to
the number of stockpile sites, multiplied by the tonnage capacity of each
stockpile.
Because stockpiles are built to completion before reclaiming, there will
always be one stockpile currently being reclaimed. To provide a buffer between
ship demand and trainload supply, a further stockpile site is needed, either full,
waiting to be reclaimed, or empty, waiting to be stacked. So, if stacking is to N
multiple sites at any time, then N+2 stockpile sites need to be made available,
and the total stockpile footprint will be approximately proportional to (N+2)C,
where “C” is the capacity of each stockpile site.
If intelligent reclaiming is being considered, the number of stockpile sites
must be further increased to allow for multiple reclaim piles. However,
intelligent reclaiming is beyond the scope of the present paper.

Development and Application of Computer Techniques to Environmental Studies X

G. Latini, G. Passerini & C. A. Brebbia (Eds) © 2004 WIT Press, www.witpress.com, ISBN 1-85312-718-3
 Development and Application of Computer Techniques to Environmental Studies X 59

5 The simulation model

A simulation model was built, to help the operators and planners to explore
implications of the alternative stacking policies. The model was designed to help
them develop stacking policies to give adequate control of variability while
minimising the footprint of the stockpile array.
To make the model accessible and easily used by the staff, it was designed
as an Excel workbook, with the simulation driven by macros. The specification
sheet is shown in Figure 2. It allows the user to specify parameters for the
desired simulation run.

Run Name 10 kt Rakes (Mine 1, based on Mine 2 7/9/01-13/11/03)

Tolerance Modify Input Rakes to Correln Rakes
Data Specification
Stress Ship Mean Std Devn Port/Blast Stress
Rake kt 10 Fe 1.00 0.40 62.15 0.96 0.67 0.96
Wt Index 1 P .0200 .0050 .0549 .0054 0.64 0.27
Replications 1 SiO2 0.33 0.31 3.41 0.64 0.57 1.93
Al2O3 0.16 0.14 1.92 0.45 0.77 2.83

Single Run Simulation Multiple Run

Stack Piles 3 Pile kt 90 Choose up to Stack Piles
Rakes/Pile 9 Footprint kt 450 Five Plots 1 1 2 3 4 5 6
3 Piles Pile/ To Completion
Stack Build Policy S.D. Stress Rake Sequential
Fe 0.16 0.16 17% Random
Sequential P .0014 0.07 27% Intelligent
Random SiO2 0.15 0.45 23% Perf. Knowledge
Intelligent Al2O3 0.08 0.52 18% X-Axis, kt X-Axis Range (kt)
Perf. Knowl. Pile Min Max Step
Total 0.71 20%
Footprint 90 210 30

Figure 2: Simulation model specification sheet.

The simulation uses data, stored in a sheet of the workbook. In the example
here, the data are assays for 1800 trainloads, as actually dumped at the port. The
statistical properties of the historic data can be adjusted to match a desired
scenario, and blast hole estimates are generated, of accuracy determined by the
specified correlation between port and blast.

Development and Application of Computer Techniques to Environmental Studies X

G. Latini, G. Passerini & C. A. Brebbia (Eds) © 2004 WIT Press, www.witpress.com, ISBN 1-85312-718-3
60 Development and Application of Computer Techniques to Environmental Studies X

Pressing either the “Single Run Simulation” or the “Multiple Run” button
initiates the VBA macros that run the simulation.

5.1 Single Run

The Single Run option simulates a single scenario and reports the composition of
the resulting completed stockpiles, in tabular and graphic form.
As an example, Figure 3 shows the silica composition of completed
stockpiles, using intelligent stacking to three stockpile destinations. We see that
six of the 200 completed stockpiles lie outside the tolerance range for silica.

4.0 SiO2
3.8

3.6

3.4

3.2

3.0
0 50 100 150 Pile # 200

Figure 3: Silica composition for completed stockpiles.

Stacking and reclaiming ore through a stockpile is a low-pass filter,
filtering out short-period variations in composition.
The simulation reports Fourier analyses to explore the filtering. The
variance is cumulated from zero frequency, and expressed as a cumulative
standard deviation, for input trainloads and for completed stockpiles.
The frequency analysis graphs for silica are shown in Figure 4. The graph
for each cumulated standard deviation stops at the relevant Nyquist (or folding)
frequency.
The Nyquist frequency has period twice the data interval, so has period 20
kt for 10 kt trainloads and 180 kt for 90 kt stockpiles.
If stacking to a single stockpile, the cumulated standard deviations would
overlay, and the reduction (from 0.64 to 0.30) achieved by stockpiling would be
equivalent to stopping the graph at the stockpile Nyquist frequency.
The further reduction in standard deviation (from 0.30 to 0.15) is achieved
by intelligent stacking.

5.2 Multiple Run

A Multiple Run simulation explores and compares a set of scenarios. Each

scenario is specified as to the number of stack piles, and whether they are to be
built in sequence, at random, or with intelligent stacking. The unattainable
scenario of perfect knowledge, with assay values known accurately, may also be
simulated, to provide a comparison.

Development and Application of Computer Techniques to Environmental Studies X

G. Latini, G. Passerini & C. A. Brebbia (Eds) © 2004 WIT Press, www.witpress.com, ISBN 1-85312-718-3
 Development and Application of Computer Techniques to Environmental Studies X 61

0.6 SiO2
Cumulative
0.5 Standard Deviation
0.4
0.3
Input Trainloads
0.2 Stockpiles
0.1 Ship Tolerance
Stockpile Nyquist
0.0
10,000 1,000 100 Kilotonnes/Cycle 10

Figure 4: Frequency analysis.

For each scenario, the simulation is repeated for a series of stockpile sizes.
The standard deviations are reported, tabulated and as graphs, plotted against
either stockpile size or total footprint. Figure 5 shows the standard deviation for
silica plotted against stockpile size, for stacking to a single stockpile destination,
and for intelligent stacking to two and to three stockpile destinations.

SiO2 Standard Deviation 1 Pile Complete

0.40 2 Piles Intelligent
3 Piles Intelligent
0.35 Ship Tolerance

0.30

0.25

0.20

0.15
30 35 40 45 50 55 60 Stockpile
65 kt 70

Figure 5: Intelligent stacking plotted against stockpile size.

Figure 5 suggests intelligent stacking to two and to three stockpiles yields

increasing benefit. But increasing the number of stockpiles increases the
footprint. Figure 6 plots the results against footprint. We now see intelligent
stacking to two stockpiles gives improved control for a given footprint size, but
no further benefit is achieved by intelligent stacking to three stockpiles.

Development and Application of Computer Techniques to Environmental Studies X

G. Latini, G. Passerini & C. A. Brebbia (Eds) © 2004 WIT Press, www.witpress.com, ISBN 1-85312-718-3
62 Development and Application of Computer Techniques to Environmental Studies X

SiO2 Standard Deviation 1 Pile Complete

0.40 2 Piles Intelligent
3 Piles Intelligent
0.35 Ship Tolerance

0.30

0.25

0.20

0.15
100 120 140 160 Footprint
180 kt 200

Figure 6: Intelligent stacking plotted against stockpile footprint.

6 Conclusions
We have seen that an Excel-based simulation model can help practitioners
explore scenarios for stacking ore, so as to reduce the area used for stockpiles
while maintaining control over variability in composition of the shipped ore. The
modelling can be readily extended to explore alternative reclaiming scenarios.

References
[1] Pitard, F.P. Pierre Gy’s Sampling Theory and Sampling Practice, 2nd ed.,
CRC Press: Boca Raton, Florida, 1993.
[2] Everett, J.E. Iron ore handling procedures enhance export quality,
Interfaces, 26(6), pp. 82-94, 1996.
[3] Everett J.E. Iron ore production scheduling to improve product quality,
European Journal of Operational Research, 129, pp. 355-361, 2001.
[4] Kamperman, M., Howard, T. & Everett, J.E., Controlling product quality at
high production rates. Proc. of the Iron Ore 2002 Conference, ed. R.
Holmes, Australasian Institute of Mining and Metallurgy: Carlton, Victoria,
pp. 255-260, 2002.
[5] Taha, H.A., Operations Research, 4th ed., Collier MacMillan: London, 1987.
[6] Hanke, J.E. and Reitsch, A.G., Business Forecasting, 6th ed., Prentice-Hall:
Englewood Cliffs, 1998.

Development and Application of Computer Techniques to Environmental Studies X

G. Latini, G. Passerini & C. A. Brebbia (Eds) © 2004 WIT Press, www.witpress.com, ISBN 1-85312-718-3