New approach to flexible open pit

optimisation and scheduling

M. Armstrong* and A. Galli
This paper proposes a dynamic procedure for pit optimisation as the resources become better
known and as commodity prices and costs evolve over time. Mining sequences are defined as the
quantities to be extracted from each macroblock (rather than small blocks) in each time period.
Feasible mining sequences are mining sequences that respect the accessibility constraints, and
also the geometric constraints on the amounts that can be extracted from any one block over time,
or from all the active blocks at a given time. Having defined feasible mining sequences, their
mathematical properties (closure and convexity under certain operations) were studied. Feasible
mining sequences are generated randomly. Then a two-step procedure is used to find the best
one (or the best few) using three economic criteria. The procedure was tested on a synthetic gold
mine based on the characteristics of the Essakane mine in Burkina Faso. The results were
encouraging.
Keywords: Dynamic pit optimisation, Feasible mining sequences, Sequencing

Introduction different (but fixed) values of the commodity price and

the costs.
Traditionally pit optimisation and scheduling are carried In practice neither the grades nor the economic
out assuming that the resources, the commodity prices variables are known. Two broad approaches have been
and production costs are all known. Once the pit proposed for using geostatistical simulations to take
contour and the mining schedule have been determined, account of the uncertainty on the resources. Dowd
geostatistical simulations can be used to assess the (1994), Rossi and van Brunt (1997) and Thwaites (1998)
impact of the uncertainty in the resources on the pit generated several conditional simulations of the orebody
design and the schedule, and likewise stochastic simula- and optimised the pit for each one. When it became
tions can be used to test its robustness to fluctuations in possible to generate far more simulations, Dimitrako-
costs and prices. However, it would be more interesting poulos et al. (2002) proposed generating the pit contour
to have a dynamic procedure for optimising the pit as using the kriged orebody model, then running the
the resources become better known and as economic economic analysis for each simulation (see also Kent
factors such as commodity prices and costs evolve over et al., 2007). The next step was to include the uncertainty
time. This paper presents a novel approach for doing on the economic variables by stochastic simulations of
this. prices and costs (Nicolas et al., 2007; Armstrong et al.,
In the standard approach the deposit is divided into 2009; Abdel Sabour and Dimitrakopoulos, 2009).
millions of small blocks. The Lerchs–Grossman algo- Recently two approaches for strategic pit optimisation
rithm is used to construct a nested set of pit shells; the have been developed based on large blocks/panels rather
one which optimises the net present value (NPV) is than millions of small blocks. Whittle and Whittle (2007)
usually selected and the scheduling is carried out on this explained that while small pit optimisation problems can
pit, usually by mixed integer programming. An alter- be solved by mixed integer programming, large ones defy
native to the Lerchs–Grossman algorithm was proposed solution by that means. The Whittles’ new proce-
by Matheron (1975a, 1975b, 1975c) and implemented by dure consists of randomly generating feasible mining
Dagdalen and Francois-Bongarçon (1982), François- sequences and then using a search algorithm to find one
Bongarçon and Guibal (1982, 1984) and Coleou (1987). that optimises the NPV. Whittle (2009) explained that in
Matheron tackled the problem from a different point of their new methodology, a sequence consists of panels to
view. He considered that the economic parameters be mined in a specified order. Each panel consists of
should be kept separate from the grades. The family of parcels which have a mine material type, a tonnage, a cost
nested pits generated by his method corresponds to of mining per tonne and one or more grade attributes, but
no specific location. In a similar vein, Menabde et al.
(2007) divided the deposit into large clumps of relatively
homogeneous material. They then optimised the mine
Cerna, Mines-Paristech, 60 Boulevard Saint-Michel, 75272 Paris, France schedule (and the cut-off grade). Geostatistical simula-
*Corresponding author, email [email protected] tions were used to take account of the uncertainty on the

ß 2012 The Australasian Institute of Mining and Metallurgy

Published by Maney on behalf of the Institute of Materials, Minerals and Mining and The AusIMM
Received 23 January 2012; accepted 13 May 2012
132 DOI 10.1179/1743286312Y.0000000008 Mining Technology 2012 VOL 121 NO 3
 Armstrong and Galli New approach to flexible open pit optimisation and scheduling

grades but the economic parameters were considered as the size of the stockpile. In multistage programming care
being known. They concluded that an interesting general- is taken to separate the variables into state variables
isation would be to incorporate stochastic models for such as commodity prices and in situ grades over which
prices. the decision makers have no control*, and the others
This paper presents a new approach for optimising which are under their control. This distinction is
and scheduling open pit mines which builds on previous maintained when defining mining sequences. These give
work, but which is radically different from it. Following the quantity of material that the mine planner decides to
Menabde and the Whittles, it was decided to work with extract from each large block in each time period. In
large blocks called macroblocks. A random search contrast to the Whittles no assumptions are made about
procedure is used to find feasible mining sequences that the grades.
achieve a close-to-optimal NPV as the Whittles do, but It is assumed that the deposit has been divided into
in contrast to them grade attributes are not included in NB large blocks called macroblocks. These could
the information associated with each macroblock. represent push-backs, for example. The material in each
The first step in this approach is to set up a population macroblock can be divided into two categories (possibly
of feasible sequences by a random construction proce- more): pure waste which is sent to the waste dump, and
dure. In the second stage uncertainty is introduced: potential ore which becomes ore if its grade is above the
stochastic processes are used to simulate market variables cut-off grade at the time of mining. Let Tot(j) be the
and geostatistical simulations are used to generate total tonnage of material in the jth block. Macroblocks
possible realisations of the orebody. At that point the are composed of a finite number of selective mining
destination of different macroblocks (treatment plant/ units.
stockpile/waste dump) can be determined as a function of
the simulated grade attributes and the simulated prices Definition of mining sequence Q
and a value criterion can be associated with each Let Q be a matrix of size NT6NB where NT is the
sequence. This could be NPV, or the measure of upside number of time periods. The quantity Q(k,j) is the
potential/downside risk proposed by Dimitrakopoulos amount to be mined from the jth block in the kth time
et al. (2007). A two-stage procedure is then used to select period. Following Whittle and Whittle (2007), it is
firstly a subpopulation of more promising sequences assumed that if part of a block is extracted in a time
using simplified economic criteria and then the best period, pure waste and potential ore are extracted pro
sequence is selected from that subpopulation using more rata. Let the cumulative quantity extracted from block j
detailed criteria. As the second stage in the selection up to and including time k be SQ(k,j)
procedure is more computer intensive than the first, the
two-stage procedure saves time. X
k
SQ (k,j)~ Q(l,j)
The mathematical properties of these feasible mining l~1
sequences turn out to be very interesting. First, they can
be characterised in a very succinct way by identifying the As the most important constraints on both open pit and
break points in the sequences, that is, the times when the underground mining are accessibility (precedence) con-
various geometric accessibility constraints are first met straints, the next step is to define the accessibility matrix
thereby allowing mining to start on other macroblocks. A. This is a square NB6NB matrix which specifies the
Sequences with the same breakpoints can be improved order in which macroblocks may be mined
by operations similar to cross-over and mutation in 
1 if block j2 can only be mined after block j1
genetic algorithms. Second, it is shown that these A(j1 ,j2 )~
sequences form branching families. This opens up the 0 otherwise
possibility of using optimisation techniques such as Level 0 blocks are those which can be mined immedi-
multistage programming with recourse to give decision ately, that is, for which
makers a dynamic procedure for optimising the pit as
the resources become better known and as economic X
NB
A(j1 ,j2 )~0
factors such as commodity prices and costs evolve over
j1 ~1
time. An example of how multistage programming with
recourse can be applied in mining is given by Armstrong The remaining blocks can only be mined after the
et al. (2012). geometric accessibility constraints have been met. If
This paper is structured as follows. In the next section there are NL0 Level 0 blocks then there are (NB2NL0)
feasible mining sequences are defined and their proper- higher level blocks.
ties are presented. The following section describes the The array AC takes the value zero only if the jth block
iterative procedure developed for generating them and is accessible at time k
the two-step procedure for selecting the best one. The 8
>
> P
NB
next section presents tests that were carried out on a > A(l,j) if k~1
<
synthetic gold mine. The discussion and conclusions l~1
AC(k,j)~
follow in the last section. In the Appendix a simple >
> P ½Tot(l){SQ (k{1,l)Aðl,j Þ if kw1
NB
>
example with only six macroblocks is presented to :
l~1
illustrate the key concepts in the new method.

Mining sequences * While some very large mining companies have enough market power to
influence the prices of the commodities that they produce, most are price
The long term objective is to use multistage program- takers. Therefore, in this paper it is assumed that prices are outside the
company’s control. Having said that, as one of the reviewers pointed out,
ming with recourse to optimise the mine contour and the iron ore mines can influence prices by the way that ore is blended into
mine plan, and later the choice of the cut-off grade, and products.

Mining Technology 2012 VOL 121 NO 3 133

Armstrong and Galli New approach to flexible open pit optimisation and scheduling

Feasible mining sequences These are put into a random order. Taking each block
~
In order to be a feasible mining sequence, a sequence Q j in turn in the randomised order, the maximum amount
must satisfy four constraints. The first ensures that only that can be extracted from that block at that time k is
accessible blocks are mined. The second limits the total computed
amount that can be extracted from a given block while Max extractable~
the third limits the amount that can be extracted in a 2 3
given time period. The fourth constraint limits the     X
~ ~ ~
amount that can be extracted from one block in a single min4Qmax j ,Tot j {SQ (k{1, j ),QM{ Q(k,jj)5
time period ~
jjv j

X
NB The first term is the maximum that can be extracted
AC(k,j)Qðk,j Þ~0 Vk from that block at any time, the second is the amount
j~1 left in the block and the third is the maximum that can
X
NT still be extracted at that time given what has been
Qðk,j ÞƒTot(j) allocated to be extracted from other blocks at that time.
k~1 A random number u is drawn between 0 and 1. Set
X
NB  
~
0ƒ Q(k,j)ƒQM Vk k~1, . . . NT Q k, j ~max½u|Totð j Þ,Max extractable
j~1
and proceed to the next block in the randomised order.
0ƒQ(k,j)ƒQmaxð j Þ Vj j~1, . . . NB & Vk k~1, . . . NT When all accessible blocks have been considered or
when the maximum possible QM has been extracted
where Qmax(j) is the maximum that could be mined
from that row, move on to the next one. If only a small
from the jth block per time period and QM is the
amount is left in a block (say 20% of the total), all of it is
maximum that can be extracted per time period. These
extracted.
two parameters could vary over time. To illustrate the
new concepts developed here, a simple example with Selecting best sequence
only six macroblocks is presented in the Appendix.
A two-step procedure is used to select the best sequence,
Break points that is, the one that maximises one or more economic
The key times in a sequence are when one or more new criteria (usually the NPV) subject to the uncertainty on
blocks can start to be mined because the accessibility both the resources and the prices and costs. Typically
constraints are met for the first time. These times are 100 geostatistical simulations and 1000 stochastic
called break points (because mining in one set of blocks simulations of the commodity price are generated. It is
breaks off, allowing access to others or ending the supposed that 5000 feasible mining sequences have been
extraction). As there are (NB–NL0) higher level blocks, generated. In the first step the best 500 sequences are
there are NBP((NB–NL0) break points. Let BP be the selected using a flat commodity price and a representa-
column vector containing the NBP break points. Let tive subset of the geostatistical simulations. The latter
SQS be an array of size NBP6NB containing the can be selected by applying the scenario reduction
cumulative quantities mined from each block at the procedure developed by Armstrong et al. (2010),
break points. This is called the sequence summary. Armstrong et al. (2011) and Armstrong et al. (2012b).
In the second step, the subset of the best 500 sequences is
Properties of feasible mining sequences tested more thoroughly. Three criteria are used: the
Let F (BP,SQS) be the set of feasible sequences having expected NPV; the probability that the NPV is negative;
the break points BP and the sequence summary SQS. It and the percentage of years when the cashflow was
is easy to prove that F (BP,SQS) is convex. Second if negative. These are computed for each of the top 500
Qmax(j) and QM are constant over time, F (BP,SQS) is sequences for all 100 geostatistical simulations and all
closed under certain types of row permutations. Let P 1000 stochastic simulations of prices, i.e. 100 000
be a permutation that affects only the rows in Q up to combinations.
and including the first break point, or only the rows One reason for using a flat commodity price in the
between two break points. Let the QP be the array first step was that most mining companies use a constant
consisting of the permuted rows and the rest of the price in feasibility studies, so this makes it possible to
array. QP is also a feasible mining sequence and so compare the value obtained there with that in the
F (BP,SQS) is closed under this type of row permuta- feasibility study.
tions.
Subsequences of rows after one break point and up to Testing the method
and including the next one can be handled indepen-
dently of previous subsequences provided that the A synthetic example was constructed to test the
sequence summary SQS remains the same. This means proposed method. The key characteristics of the ore-
that subsequences can be mixed and matched. It is body and the economic parameters were set up to mimic
planned to use this in future work to improve feasible those of the Essakane Gold Mine in Burkina Faso,
mining sequences. because the updated feasibility study (IAMGOLD,
2009) for the mine was available on the internet. Over
Generating feasible mining sequences the 9?5 year mine life its owners plan to mine 330 000 oz
An iterative procedure is used to generate a large of gold per year at a cut-off of 0?5 g t21. The capital
number of feasible mining sequences. Starting with the expenditure was planned to be about $450 million. After
first time period, the accessible blocks are identified. analysing their figures, the fixed cost was set at $50

134 Mining Technology 2012 VOL 121 NO 3

 Armstrong and Galli New approach to flexible open pit optimisation and scheduling

1 Drill hole layout

2 Bird’s eye view of lowest level of pit (above), and verti-
cal section show mineralised blocks plus waste that
million per year, and the variable costs at $1?20/t of must be extracted to ensure 45u slope on walls (below)
material extracted from the pit and $10/t for mining and
processing ore.
simplified approach: a flat gold price of $600/oz and a
Orebody model subset of 12 simulations that is representative of the 100
The mineralised zone is a rectangular block 10506 geostatistical simulations. These were selected by apply-
5006120 m deep. The grid consists of 110 drill holes on ing the scenario reduction procedure developed by
a 100650 m grid as shown in Fig. 1. Each of the vertical Armstrong et al. (2010) and Armstrong et al. (2011b).
drill holes contains 12 sections 10 m long which have been The user specifies how many simulations to retain (here
analysed for gold content. The ore in the central part has 12) and the procedure selects them together with their
higher grades. The data have an overall log normal type probabilities. Armstrong et al. (2011b) demonstrated that
distribution with a few high values. the expected NPV computed using the representative
Figure 2 presents a bird’s eye view of the lowest level subset together with the weighting coefficients is very
of the orebody (above) and a vertical cross-section close to the average NPV computed from the full set of
(below) showing the mineralised material and the 100 simulations. Figure 4 shows the 5000 values of the
additional waste that has to be extracted to maintain expected NPV obtained using a gold price of $600 and
a 45u pit slope. The macroblocks are 1506100640 m averaging over the 12 representative simulations. These
and contain 600 selective mining units (smus), each were arranged in descending order from left to right. The
10610610 m. The orebody contains a total of 197 top 500 sequences were selected for more detailed study.
macroblocks, 99 in the top level, 63 in the middle level In the second step three criteria were considered: the
and 35 in the lowest level. expected NPV; the probability that the NPV was
negative; and the number of years when the cash flow
Stochastic model for price of gold was negative. These were computed based on all 100
For simplicity’s sake a geometric Brown motion with an geostatistical simulations and all 1000 simulated price
initial price of $600 and a volatility of 20% was used. paths for the gold price (that is, 100 geostatistical
Figure 3 shows 25 of the 1000 simulations of the gold simulations61000 price paths6500 feasible sequences).
price. Table 1 gives the values of these three criteria for the
best three sequences. In this case one sequence is better
Generating 5000 feasible mining sequences than the others according to all three criteria, especially
The first step was to generate 5000 feasible mining the percentage of years with negative cash flows which
sequences. As there are a total of 197 macroblocks to be gives the management an idea of the risk involved.
mined over 10 years, the matrices Q and S will be
106197 in size. As the mining equipment is designed to Discussion and conclusions
extract the equivalent of 12 000 smus per year, the value
of QM is set to 12 000. Because of the geometry of the The long term objective is to use multistage programming
pit the 197 macroblocks are not all the same size. As with recourse in order to evaluate and optimise mining
blocks are 40 m high, waste panels have to be 40 m wide projects subject to technical and financial uncertainty, that
in order to respect the 45u stability angle. Three waste is, uncertainty on the resources and the commodity prices
block sizes need to be considered depending on whether and costs. One of the underlying principles in multistage
they are oriented EW or NS, or whether they are corner programming is to separate state variables such as prices
blocks (see Fig. 2). The EW waste blocks are and grades (which the decision maker cannot change)
150640640 m and have a volume equivalent to from the control variables such as the amount to mine
240 smus, while the NS ones are 100640640 m and from each macroblock in each time period, which are
are equivalent to 160 smus. Last, corner blocks are under the management’s control. This is why mining
40640640 m and contain 64 smus. These give us the sequences were defined as being the quantities to be
values of Tot(j) for the different macroblocks. For extracted without including grade attributes. Feasible
simplicity Qmax(j) was set to be equal to Tot(j). mining sequences are mining sequences that respect the
accessibility constraints, and also the geometric con-
Selecting best of 5000 feasible mining straints on the amounts that can be extracted from any
sequences one block over time, or from all the active blocks at a
A two-stage procedure was used to select the best feasible given time.
mining sequence. In the first step, the expected NPV was Having defined feasible mining sequences, their
computed for each of the 5000 feasible sequences using a mathematical properties were defined. It was found that

Mining Technology 2012 VOL 121 NO 3 135

Armstrong and Galli New approach to flexible open pit optimisation and scheduling

3 Twenty of 100 simulations of gold price over 10 year on weekly basis: model was geometric Brownian motion with
initial price of $600 (as in Essakane feasibility study) and volatility of 20%

trinomial trees used to model prices in finance and in

real options.
Further work is in progress to optimise the cut-off
grade as well as the order in which the macroblocks are
mined. Preliminary results indicate that mining should
stop if the price drops too low for too long. In that case
the macroblocks at the end of the sequence would
simply not be mined. In other words, the pit would be
smaller. Therefore, these new feasible mining sequences
provide a way of optimising the mining sequence and the
pit contour.
While this new method builds on existing work by
others, it is radically different in several regards and
notably because it separates the state variables from the
control variables. In our opinion this will open up the
way to optimising mining projects (underground as well
as open pit) while taking account of the inherent
4 Expected NPV computed in first step of selection pro-
uncertainty on the resources and on the economic
cess using flat gold price of $600 and subset of 12 of
variables.
100 geostatistical simulations: 5000 values have been
arranged in descending order from left to right Acknowledgement
This research was carried out as part of a Consortium on
natural breakpoints occur in the sequences when one set the Application of Real Options in Mining. The authors
of blocks have been completely mined out, thereby would like to thank the sponsors, Codelco (Santiago,
opening up access to others. It was discovered that Chile) and two multinational mining groups, for their
families of sequences exist, and that they are convex and support and encouragement during the consortium.
closed under certain types of row permutation opera-
tions. Convexity is an advantage when optimising. The
closure property opens up the possibility of mixing and
References
matching subsequences of rows in the sequences. Abdel Sabour, S. A. and Dimitrakopoulos, R. 2009. Accounting for joint
Another interesting property is that the sequences form ore supply, metal price and exchange rate uncertainties in mine
design, in Orebody mining and strategic mine planning, 21–
branching families, rather like the binomial and 26, Perth, WA, The Australasian Institute of Mining and
Metallurgy.
Armstrong, M., Galli, A. and Ndiaye, A. A. 2009. A case study on the
Table 1 Values of three criteria (expected NPV, pro- impact of hedging against foreign exchange risk and energy price
bability of negative cash flow and proportion of risk, in Orebody modelling and strategic mine planning, 95–
years with negative cash flow) for three top 99, Perth, WA, The Australasian Institute of Mining and
Metallurgy.
ranking feasible mining sequences
Armstrong, M., Galli, A. and Razanatsimba, R. 2011. Multistage
Probability of Proportion of years stochastic optimisation for managing major production incidents,
Expected a negative with negative cash Proc. 35th APCOM Conf., 491–500, Wollongong, NSW,
Australia, September 2011, AusIMM.
Rank NPV NPV/% flow/%
Armstrong, M., Ndiaye, A. A. and Galli, A. 2010. Scenario reduction
in mining, Proc. IAMG Annual Meet., xxxx, Budapest, Hungary,
1 $322.57 M 28.3 12.74
August 2010, IAMG.
2 $321.65 M 28.8 15.68
Armstrong, M., Ndiaye, A. A., Razanatsimba, R. and Galli, A. 2012a.
3 $321.52 M 28.5 14.46 Scenario reduction in mining, Math. Geosci., to be published.

136 Mining Technology 2012 VOL 121 NO 3

 Armstrong and Galli New approach to flexible open pit optimisation and scheduling

Armstrong, M., Vincent, A., Galli, A. and Méheut, C. 2012b. Genetic 2 3

algorithms and scenario reduction, Proc. Geostat 2012 Conf., 0 0 0 0 1 0
Oslo Norway, June 2012, Congress-Conference As. 60 0 0 0 1 07
Coleou, T. 1987. The two dimensional dynamic programming 6 7
6 7
algorithms as used in the technical parametrisation of reserves, 60 0 0 0 0 17
A~6
60
7
0 17
Internal Note CGMM, Ecole des Mines de Paris, 17.
Dagdalen, K. and François-Bongarçon, D. 1982. Towards the complete 6 0 0 0 7
6 7
double parameterization of recovered reserves in open pit mining, 40 0 0 0 0 05
Proc. 17th Int. APCOM Symp., 288–296, Littleton CO, Society
for Mining, Mettalurgy and Exploration. 0 0 0 0 0 0
Dimitrakopoulos, R., Farelly, C. T. and Godoy, M. 2002. Moving Table 2 gives the total amount Tot(j) in each macro-
forward from traditional optimisation: grade uncertainty and risk
effects in open pit design, Trans. IMM, 111, A82–A88. block together with the maximum amount Qmax(j) that
Dimitrakopoulos, R., Martinez, L. and Ramazan, S. 2007. Optimising can be mined from the jth block per time period. The
open pit design with simulated orebodies and Whittle four-X – a matrix Q below is a feasible mining sequence for
maximum upside/minimum downside approach, in Orebody producing all the material in 10 time periods. An
modelling and strategic mine planning, Spectrum series, 2nd equivalent way of expressing the mining sequence is by
edn, Vol. 14, 201–206, Melbourne, NSW, The Australasian
Institute of Mining and Metallurgy. computing the cumulative quantities extracted up to and
Dowd, P. A. 1994. Risk assessment in reserve estimation and open- including any time period. The matrix S gives this
pit planning, Trans. Inst. Min. Metall. A: Min. Ind., 103, A148–
A154. Time Block no: Block no:
François-Bongarçon, D. and Guibal, D. 1982. Algorithms for
parameterizing reserves under different geometrical constraints, Period ½1 2 3 4 5 6 ½1 2 3 4 5 6
Proc. 17th Int. APCOM Symp., 297–309, Littleton CO, Society 2 3 2 3 2 3
for Mining, Mettalurgy and Exploration.
1 2 2 1 0 0 0 2 2 1 0 0 0
6 7 6 7 6 7
François-Bongarçon, D. and Guibal, D. 1984. Parameterization of 6 2 7 61 2 1 2 0 07 6 2 0 07
optimal designs of an open-pit – beginning a new phase of 6 7 6 7 63 4 2 7
6 7 6 7 6 7
research, Trans. Soc. Min. Eng. AIME, 274, 1801–1805. 6 3 7 63 2 0 1 0 07 6 3 0 07
IAMGOLD. 2009. Updated feasibility study on the Essakane 6 7 6 7 66 4 2 7
6 7 6 7 6 7
6 4 7 62 2 1 1 0 07 6 4 0 07
Gold Deposit (Burkina Faso), http://www.iamgold.com/pdf/
6 7 6 7 68 8 3 7
Operations/43-101%20Technical%20Report%20Essakane%20Project 6 7 6 7 6 7
6 5 7 60 0 2 2 1 07 6 6 1 07
%20March%202009.pdf 6 7 6 7 68 8 5 7
Kent, M., Peattie, R. and Chamberlain, V. 2007. Incorporating grade 6 7Q~6 7S~6 7
6 6 7 60 0 2 1 2 07 6 7 3 07
uncertainty in the decision to expand the main pit at the 6 7 6 7 68 8 7 7
Navachab Gold Mine, Namibia, through the use of stochastic 6 7 6 7 6 7
6 7 7 60 0 2 2 1 07 6 9 4 07
simulation, in Orebody mining and strategic mine planning, 6 7 6 7 68 8 9 7
spectrum series, Vol. 14, 207–216, Melbourne, NSW, AusIMM. 6 7 6 7 6 7
6 8 7 60 0 0 0 2 37 6 9 6 37
Matheron, G. 1975a. Le parametrage technique des réserves, Internal 6 7 6 7 68 8 9 7
6 7 6 7 6 7
Note CGMM, Ecole des Mines de Paris, 54. 6 9 7 60 0 0 0 2 37 6 9 8 67
Matheron, G. 1975b. Parametrage des contours optimaux, Internal 4 5 4 5 48 8 9 5
Note CGMM, Ecole des Mines de Paris, 54. 10 0 0 0 0 2 3 8 8 9 9 10 9
Matheron, G. 1975c. Compléments sur le parametrage des contours
optimaux, Internal Note CGMM, Ecole des Mines de Paris, 19.
Menabde, M., Froyland, G., Stone, P. and Yeates, G. 2007. Mining
schedule optimisation for conditionally simulated orebodies, in It is supposed that QM56, that is, at most six units can be
Orebody mining and strategic mine planning, Spectrum series, produced from the pit in any time period. Computing the
Vol. 14, 353–357, Melbourne, NSW, AusIMM.
row sums shows that the sequence respects this constraint.
Nicolas, G. D., Coward, S. J., Armstrong, M. and Galli, A. 2007.
Integrated mine evaluation – implications for mine management, As all the material from blocks 1 and 2 has been extracted
in Orebody mining and strategic mine planning, Spectrum series, by time period 4, mining could start on block 5 in the
Vol. 14, 9–19, Melbourne, NSW, AusIMM. following period, but as the extraction of blocks 3 and 4
Rossi, M. E. and van Brunt, B. H. 1997. Optimising conditionally was only completed in the 7th time period, work on block
simulated orebodies with Whittle 4D, Proc. Conf. on ‘Optimising
6 could only start in time period 8. In this case the
with Whittle’, 119–128, 1997, Perth, WA, Whittle Programming Pty.
Thwaites, A. M. 1998. Assessment of geological uncertainty for a breakpoints are 4, 7 and 10 (when all the material has been
mining project, Proc. 27th Int. APCOM Symp., London, UK, removed). The vector of the break points (BP), together
April 1998, IMM, 396–404. with the sequence summary matrix SQS are given below
Whittle, G. 2007. Global asset optimisation orebody modelling and 2 3 2 3
strategic mine planning, Spectrum series, 2nd edn, Vol. 14, 331– 4 8 8 3 4 0 0
336, Melbourne, NSW, AusIMM. 6 7 6 7
BP~4 7 5 SQS~4 8 8 9 9 4 0 5
Whittle, J. 2009. The global optimiser works – what next?, in Orebody
modelling and strategic mine planning, 3–5, Perth, WA, AusIMM. 10 8 8 9 9 10 9
Whittle, J. and Whittle, G. 2007. Global long-term optimisation of very
large mining complexes, Proc. 33rd Int. APCOM Symp.,
Santiago, Chile, the University of Chile, 253–260.
Row permutations
Appendix Two more mining sequences, Q2 and Q3, are considered,
obtained by permuting rows 2 and 3 and then rows 4
and 5 respectively. Q2 is a feasible mining sequence but
Simplified example
Table 2 Values of parameters in simplified example
A case where there are six macroblocks is considered.
Blocks 1–4 are accessible immediately. Block 5 can only Macroblock j 1 2 3 4 5 6
be exploited after blocks 1 and 2 have been mined, while
block 6 can only be exploited after blocks 3 and 4. The Tot(j) 8 8 9 9 10 9
Qmax(j) 3 3 3 3 3 2
accessibility matrix A is

Mining Technology 2012 VOL 121 NO 3 137

Armstrong and Galli New approach to flexible open pit optimisation and scheduling

Q3 is not because it attempts to mine block 5 before is [8 8 3 4 0 0]. This would be the first row in the
blocks 1 and 2 have been fully mined out. Q2 permutes sequence summary SQS in both cases
rows within the same subsequence; Q3 attempts to 2 3 2 3
permute them across a break point. 2 2 1 0 0 0 2 2 1 1 0 0
61 2 1 2 0 07 62 2 1 1 0 07
6 7 6 7
Time Block no: Block no: 6 7 6 7
43 2 0 1 0 05 42 2 0 1 0 05
Period ½1 2 3 4 5 6 ½1 2 3 4 5 6 2 2 1 1 0 0 2 2 1 1 0 0
2 3 2 3 2 3
1 2 2 1 0 0 0 2 2 1 0 0 0
6 7 6 7 6 7 Next take a linear combination of these two sets of rows
6 2 7 63 2 0 1 0 07 61 2 1 2 0 07 (i.e. weighted average with weights a and b such that
6 7 6 7 6 7
6 7 6 7 6 7 azb51). The sequence summary corresponding to row
6 3 7 61 2 1 2 0 077 63 2 0 1 0 07
6 7 6 6 7 4 is still [8 8 3 4 0 0]. Therefore, the linear combination
6 7 6 7 6 7
6 4 7 62 2 1 1 0 07 60 0 2 2 1 07 belongs to the same family
6 7 6 7 6 7
6 7 6 7 6 7
6 5 7 60 0 2 2 1 07 62 2 1 1 0 07
6 7 6 7 6 7
6 7Q2 ~6 7Q3 ~6 7 2 3 2 3
6 6 7 60 0 2 1 2 07 60 0 2 1 2 07
6 7 6 7 6 7 2 2 1 0 0 0 2 2 1 1 0 0
6 7 6 7 6 7 6 7 6 7
6 7 7 60 0 2 2 1 077 60 0 2 2 1 07 61
6 7 6 6 7 6 2 1 2 0 077
62
6 2 1 1 0 077
6 7 6 7 6 7 a6 7zb6 7
6 8 7 60 0 0 0 2 377 60 0 0 0 2 37 63 7 62
6 7 6 6 7 4 2 0 1 0 05 4 2 0 1 0 075
6 7 6 7 6 7
6 9 7 60 0 0 0 2 37 60 0 0 0 2 37
4 5 4 5 4 5 2 2 1 1 0 0 2 2 1 1 0 0
10 0 0 0 0 2 3 0 0 0 0 2 3
2 3
2az2b 2az2b azb b 0 0
6 7
6 az2b 2az2b azb 2azb 0 07
Linear combinations 6 7
~6 7
This time only the first four rows are considered (i.e. 6 3az2b 2az2b 0 azb 0 07
4 5
rows in a subsequence between breakpoints) from two
2az2b 2az2b azb azb 0 0
feasible mining sequences from the same family
F (BP,SQS). In both cases the cumulative amount to This illustrates some of the interesting properties of
be extracted up to and including the fourth time period these families of feasible mining sequences.

138 Mining Technology 2012 VOL 121 NO 3