Professional Documents
Culture Documents
To cite this article: F. R. Albor Consuegra & R. Dimitrakopoulos (2010) Algorithmic approach to
pushback design based on stochastic programming: method, application and comparisons, Mining
Technology, 119:2, 88-101, DOI: 10.1179/037178410X12780655704761
To link to this article: https://doi.org/10.1179/037178410X12780655704761
Introduction starting pushbacks which contain y10 and y22 million ore
tonnes respectively. The conventionally generated LOM
A pushback (or cutback or phase) is an aggregation of production schedules show a 5% (y$10 million) difference
mining blocks. A pushback design is used to guide the in the achieved NPV due to the location, tonnage and grade
sequence of extraction of an orebody from the point where distribution of subsequent pushbacks exclusively as a
the mining operation begins and where it stops. The design consequence of the design of the starting pushback.
of pushbacks is essential to life-of-mine scheduling Conventional production schedules are gen-erated through
because it discretises the pit space into individual pit units optimisers that do not account for geological uncertainty.
with their own working face, whichwhen extracted, control With regard to pit limits, a mining operation should be
ore and waste production. In addition, pushbacks assist in stopped before negative cash flows are obtained. Ultimate
meeting ore production targets, deferring waste production, pit limits are also dependent on the pushback design.
providing a mini-mum mining width to accommodate Figure 2 shows an example where ultimate pit limits are
access and mobility of equipment, and ensuring safe pit extended to assess the risk associated to the uncertainty in
slopes. As a result, different pushback designs yield pit limits. The extension of pit limits is carried out through
different annual cash flows resulting in different net the addition of pushbacks at the bottom of the pit. Figure
present value (NPV) assessments.1 A starting pushback has 2a shows the NPV of LOM production schedules based on
a significant impact over the distribution of cash flows different pit limits and Fig. 2b shows the physical
throughout the life of the project. Figure 1a shows a difference in scheduling patterns. A 5% (y$10 million)
comparison of NPV forecasts of two lives of mine (LOM) NPV difference between the schedules is shown. These
production schedules based on different pushback designs production schedules were generated through the Milawa
in an application at a porphyry copper deposit, and Fig. 1b NPV Algorithm of the Whittle software. Owing to the
shows the physical differences of the pushback designs deterministic nature of conventional production schedules,
(left to right). Designs a and b differ on the design of there is a risk of not meeting production targets and
therefore NPV forecasts.2 A solution to such an issue is
available through stochastic optimisers, which have the
COSMO – Stochastic Mine Planning Laboratory, Department of Mining ability to integrate uncer-tainty into the scheduling process.
and Materials Engineering, McGill University, FDA Building, 3450 Net present value increments in the order of 10–25% due to
University Street, Montreal, Que. H3A 2A7, Canada
the use of
*Corresponding author, email roussos.dimitrakopoulos@mcgill.ca
2010
VOL 119 NO 2
Mining Technology
Albor Consuegra and Dimitrakopoulos Algorithmic approach to pushback design based on stochastic programming
1 a difference of NPV forecasts due to difference in design of starting pushbacks and b cross-sections of pushback designs
stochastic optimisers have been reported;3–5 in such cases, through a search for nested pits at incremental depths, in
production targets are met. Relevant topics to the other words, maximal closures of a graph that lie on the
methodology proposed in this paper are briefly described convex hull. The reader is referred to Seymour11 for
below. further details on this topic.
The problem of defining optimum pit limits was first Although the design of nested pits may be used to define
addressed in optimisation as finding the maximum closure a sequence of extraction, it is not necessarily optimal for
of a graph.6 Later, the problem was formulated as that of the problem at hand. This is due to:
(i) not considering grade blending or metal pro-
finding a minimum cut7 which may be solved by maximum duction requirements
flow algorithms.8 However, the design of an open pit is not (ii) large variations in the size of nested pits
complete because one encounters the problem of defining (iii) ignoring uncertainty of the economic value of a
the best way to reach the final contour. Considering that block
there are many ways to do so, it is of interest to define the (iv) nested pits not being sufficiently spaced apart to
sequence of extraction to reach the final contour that provide a minimum mining width that allows the
maximises the value of the project. The design of access to mining equipment
intermediate pits is traditionally used to guide the sequence
(v) the lack of economic discounting in the objective
of extraction. To generate such pits it is a common practice
function.
to find maximal closures on a graph with resource or
capacity con-straints, each closure representing a different An objective function that maximises the discounted value
intermedi-ate pit. Solving this problem requires the of the project leads to the definition of a production
dualisation of the constraints, which yields a Lagrangian schedule. Ore, waste, metal, grade blending and stockpiling
requirements are easily integrated into production
relaxation problem.9 Now, the problem changes into
scheduling approaches. Nested pits may undergo a
finding the value of the Lagrangian multiplier which
conversion into pushbacks; they can be naively grouped
reduces the gap of optimality. A number of methods are
into pushbacks by analysing a pit by pit graph, a graph that
readily available to solve this problem.10 Another approach plots the cumulative economic value and tonnage versus
to designing intermediate pits relates to what is known as each pit. Such an approach relies on the subjectivity of the
parameterisation6 which consists of generating closures on planning engineer and is based on the assumption that each
a graph as a function of another parameter related to the pushback is mined out completely before moving on to the
properties of the blocks of the orebody model, e.g. finding next. In mining operations, different pushbacks may be
pits that maximise the economic value for different desired mined out simultaneously.
volumes. The process is carried out
2 a difference of NPV forecast due to different pit limits and b cross-sections of corresponding scheduling patterns
J(TPB{1) N
E b t{
Max t~1 " ~ ðNPVÞlt l s~1 cot0 dsot0zcut0 dsut0 #
4 Number of pushback designs as function of target num-ber
where l is the block identifier, t is the time period, t0 is the of pushbacks and number of available nested pits
ore production target type, u is the lower bound, o is the
upper bound, s is the simulation number, T is the maximum
number of scheduling periods, L is the total number of
‘selective’ binary definition not only reduces the solution
blocks to be scheduled, btl is a variable time but also does not affect the quality of the solution by
representing the portion of block l to be mined in period t, keeping it optimal. The formulation can be easily extended
if it is defined as a binary variable, it is equal to 1 if the
block l is to be mined out in period t and equal to 0 if to generate production schedules that minimise the risk of
deviating from metal and grade targets; the reader is
otherwise; E ð NPVÞtl is the expected NPV to be referred to an example in Dimitrakopoulos and Ramazan.14
generated if block l is mined in period t considering
simulation s, cto0 and ctu0 are the unit costs for excess and
Case study at porphyry copper deposit
deficient ore production respectively, and dsot0 and dsut0 are
the excess and deficient amount of ore production in period The approach described in the previous section is tested on
t considering simulation s. The objective function is subject a porphyry copper deposit. The technical and economic
to reserve, mining, processing, slope and grade constraints. information related to the case study are shown in Table 2.
For purposes of completeness, the proces-sing constraints
are defined below. Consider Osl to be the ore tonnage of a
given block l conditioned to simulation s, dummy variables
Stage I
atsu0 and atso0 to balance the equality, and the maximum To discretise the pit space, the industry common practice is
and minimum ore produc-tion expected per production to make use of an implementation of the Lerchs–
period of the LOM; then, ore tonnage production must lie Grossman algorithm available commercially in Whittle
within lower and upper software. Their idea is to parameterise the space through
bounds, Omin and Omax. The variables and parameters the variation of the ratio of metal price to extraction cost.17
described above lead to define the processing constraint
P1
l
P1
l
In this case, a wide range of values of the parameter
L L
as t
Oslb zd t0 t
~Omin and Oslbtzdt0 t provides enough information to analyse the sensitivity of
{a 0 {a 0
~
lsu su
~
lso so
the pit size to the price of the commodity. The information
~Omax. Risk management is accomplished by the in Table 2 and a conventionally estimated orebody block
introduction of a geological risk discounting rate into the model created through ordin-ary kriging are provided to the
calculation of the costs for excess and deficient Whittle software in order to generate a set of 17 nested pits.
production.14 Aside from a grade cut-off, a probability cut-
off is used to classify the blocks as ore or waste. Stage II
Furthermore, this SIP implementation takes into con- The total number of possible pushback designs is
sideration a different way of defining binary variables. 15 216565536. Figure 4 shows the possible number of designs
This consists of reducing the amount of binary variables by as a function of the target number of pushbacks
setting the waste blocks to linear variables and the
ore blocks to binary variables. Ramazan and
Dimitrakopoulos15,16 also show a case study where such Table 3 Grouping of pits based on given target number of
pushbacks: grouping criterion is maximum economic
value
Table 2 Economic and technical parameters
. Target number
Copper price, US$/lb 19
. of pushbacks TPB Resulting grouping of pits
Selling cost, US$/lb 04
.
Mining cost, $/t 10
. 3 [3 6 17]
Processing cost, $/t 90
Slope angle, u 45 5 [2 5 7 13 17]
. 6 [2 4 6 7 12 17]
Processing recovery 09
7 [1 3 4 6 8 12 17]
Block dimensions, m 20620610
Ore production target, Mt/year .
75
9 [1 3 4 5 6 7 12 16 17]
. 10 [1 3 4 5 6 7 12 14 16 17]
Waste production target, Mt/year 20 5
5 Performance of LOM production schedule based on 3 pushback design in terms of a ore production, b waste produc-tion, c
cumulative metal production and d cumulative discounted cash flows
for the case where the number of available nested pits is the 17 different pushback designs available from the set of
17. For the case study, the nested pits are grouped based on 17 nested pits. Nevertheless, as it will be explained the
the maximisation of the economic value of each design. approach is tested only for six different pushback
The value of each pushback is discounted based on its life designs, TPB53, 5, 6, 7, 9, 10. The case of TPB51 is ignored
and the time required for depletion. Note that since there because it equals to not considering pushback
are 17 nested pits available, at most there are design. Although the cases of TPB52, 4, 8 were initially
17 different pushback designs that may be tested in stage part of the case study, solutions in stage III were not
III. In other words, there exists a unique pushback design, obtained in a feasible amount of time. Regarding a TPB.10,
mj ~maxj~1...Jff (mj)g, for each TPB that yields the highest recall that defining a sequence of extraction based on the
economic value. Then, in this case study, the number of available set of nested pits is neither necessary optimal nor
possible groupings of nested pits reduces to
feasible for the case at hand. As TPB approaches the
17.
number of available nested pits, the production schedule
Stage III presents greater deviations from production targets and
leads to smaller ultimate pits, which is illustrated in the
Twenty simulated orebody models are fed to the SIP model
described in the previous section to generate a LOM cases of TPB59, 10. Therefore, the approach is tested until.
production schedule for each one of the pushback designs
in Table 3. The orebody models were generated using the Analysis of solutions
direct block simulation algorithm.3 The SIP model For the case of three pushbacks TPB53 the grouping that
described above could be solved for each one of yields the highest economic value is shown in Fig. 5.
8 Performance of LOM production schedule based on five pushback design in terms of a ore production, b waste pro-duction, c
cumulative metal production and d cumulative discounted cash flows
The starting pushback contains the 1st, 2nd and 3rd nested during the 3rd and 4th periods there is a y5% deviation
pits. The starting pushback allows the scheduling of the 1st (over production) from the ore production target and a y2%
and part of the 2nd production period. Regarding the deviation (under production) during the 5th period, as
performance of the production schedule, shown in Fig. 5. The risk of not meeting the
10 Performance of LOM production schedule based on six pushback design in terms of a ore production, b waste pro-duction, c
cumulative metal production and d cumulative discounted cash flows
12 Performance of LOM production schedule based on seven pushback design in terms of a ore production, b waste production, c
cumulative metal production and d cumulative discounted cash flows
Table 4 Global proportions of ore–waste tonnage extraction, NPV, maximum deviation from ore production targets and maximum
stripping ratio of LOM production schedule based on different pushback designs
14 Performance of LOM production schedule based on nine pushback design in terms of a ore production, b waste pro-duction, c
cumulative metal production and d cumulative discounted cash flows
production target during the 6th period is higher than that When the third pushback is scheduled, the optimiser is not
observed in the other production periods; note that the risk able to meet the ore production target for the 4th period
profile is wider. without mining the huge amounts of waste left behind from
The design of five pushbacks TPB55 is shown in Fig. 6. the previous pushbacks, leading to an infeasible waste
The starting pushback contains the 1st and 2nd nested pits; production rate.
there are clear physical differences, regard-ing the starting The design of six pushbacks TPB56 is shown in Fig. 9.
pushback, between this case and the previous design The starting pushback is designed as in the previous case
leading to different scheduling patterns as shown in Fig. 7. leading to the same scheduling patterns during the 1st
The risk analysis of the LOM production schedule in Fig. 8 production period. The intermediate pushbacks contain
shows that there are no major deviations of the ore different nested pits from the previous cases as shown in
production target. Nevertheless, the schedule is not Table 3. This leads to different scheduling patterns in the
necessarily feasible due to the increment of the waste subsequent production periods. Figure 10 shows a 3 and
mining rate during the 4th production period; the waste 4% deviation (over production) from the ore production
tonnage extracted in the 3rd period increases from 11?3 to target during the 4th and 5th periods and a 3% deviation
29?2 Mt in the 4th period. This result relates to the (under production) during the 6th period which results in a
grouping of pits selected for the five pushback design; y15% reduction in the metal production during the 6th
recall that the grouping criteria is the maximisation of the period, when compared to previous periods. Table 4 shows
economic value, and different criteria would have that the overall ore– waste tonnage extracted in this case is
suggested a different design. Furthermore, for the case the same as the one in previous cases; this is reflected in
study under consid-eration the waste production rate is not the identical ultimate pit contours as shown in Fig. 5.
constrained, allowing it a free variation. Scheduling the
first two pushbacks satisfied the ore production target of The design of seven pushbacks is shown in Fig. 11.
the 1st, 2nd and 3rd periods; the schedule defers waste Figure 12 shows that its corresponding LOM production
mining. schedule presents a y6% deviation (over production)
16 Performance of LOM production schedule based on 10 pushback design in terms of a ore production, b waste pro-duction, c
cumulative metal production and d cumulative discounted cash flows
from the production target during the 3rd and 4th periods. the cash flows become negative. This leads to a smaller
The average grade of the ore fed to the mill is constant ultimate pit than that obtained in the previous case. Figure
throughout the LOM, resulting in a constant metal 15 shows the ultimate pit contour and Fig. 16 shows the
production rate. Table 4 shows that the overall tonnage differences in the performance of the LOM production
extracted in this case is higher than the previous cases schedule.
resulting in a different ultimate pit contour as shown in Fig.
11. This is due to the difference in the pushback design Selecting pushback design
which yields a different cash flow pattern and therefore a
different physical point in which the cash flows become Table 4 shows the overall tonnage extracted, NPV, the
negative. maximum deviation from the ore production target, and the
The design of nine pushbacks is shown in Fig. 13. The maximum stripping ratio presented throughout the LOM
total tonnage extracted with this schedule is 4% lower than for the production schedules discussed above. The positive
that in the previous cases, which implies that the operation (z) and negative (2) signs used in the column of the
is stopped at different times. A proof of this result is the maximum deviations from ore production targets express
physical difference in the pit limits between this case and over and under production respectively. The overall
the previous cases (see Fig. 13). The waste mining rate of tonnages may be used as a basis for comparison of
the 3rd production period (Fig. 14) affects the performance different pit limits presented in the cases above. The cases
of the schedule by restricting the availability of ore during of nine and ten pushback designs extract considerably less
the 5th production period. Owing to the pushback design tonnage than the previous cases, with NPV differences of
which causes the ore shortage during the 5th production the order of y3%. To analyse these results one must
period, the optimiser extracted the ore within the 17 nested consider that deviations from ore production targets are in
pits in eight periods. The 8th production period yielded the order of 13%, leading to an erroneous NPV forecast.
negative cash flows; therefore, mining stops at the 7th Similar observations apply for the design with five
period. This highlights the importance of pushback design pushbacks; erroneous NPV forecast caused by the
and its effect on the definition of ultimate pit limits. operational infeasibility is due to the high stripping ratio,
3?89%.
The design of 10 pushbacks is shown in Fig. 15. The
main difference between this and the previous design lies Design of starting, intermediate and
on the pushbacks that discretise the bottom of the pit. In
this case, the bottom of the pit is broken up into three
bottom pit pushbacks
pushbacks, defined by the nested pits [12 14 16 17]. Such The design of starting pushbacks in this case study consists
discretisation provides more information about when of three possible groupings: 1st, 2nd and 3rd
19 Performance of LOM production schedule based on design with seven pushbacks in terms of a ore production, b waste
production, c cumulative metal production and d cumulative discounted cash flows
design. The 5th period is scheduled mainly on the 4th design is considered through the following grouping of
pushback defined by the nested pits [6 7], and the 6th nested pits: [1 4 6 8 14 16 17]. The design is based on
period is scheduled mainly in the 5th pushback defined by seven pushbacks and it is shown in Fig. 18. The physical
the nested pits [7 12]. To avoid ore shortage issues, the 4th patterns of the LOM production schedule are shown in Fig.
and 5th pushbacks are enlarged to envelope the 8th and 18. Figure 19 shows the performance of the LOM
14th nested pit respectively, which thus forces the production schedule and that the ore production rate meets
optimiser to mine additional waste and ore tonnage during the ore production target except for the 4th and 6th
the 5th production period in order to facilitate the access to production periods which present a y4% deviation (over
ore blocks during the 6th production period. Note that the production). The waste and overall rate of production
extension of the 4th pushback provides ore and waste alike, fluctuates between time periods; however, the percentage
hence avoiding ore shortage issues as in the cases of nine of idle capacity of the mining fleet is 3– 4% at the most,
and ten pushbacks. Following the previous discussion, which is at this stage of planning practically insignificant.
another pushback
20 Performance of conventional LOM PS based on ordinary kriging model in terms of a ore production and b cumulative discounted
cash flows
21 Performance of conventional LOM PS based on E-type model in terms of a ore production and b cumulative dis-counted cash
flows
Conventional versus stochastic approach: scheduler that does not account for uncertainty, the
performance and comparison performance of its key parameters will be deficient. To
Conventional practices do not integrate geological illustrate this, Fig. 22 shows the performance of a LOM
uncertainty into the process of pushback design. In Figs. 20 production schedule based on the pushback design shown
and 21 the broken line shows the expected performance of in Fig. 18. The schedule is generated by the Milawa NPV
a conventional LOM production sche-dule, generated by Algorithm in the Whittle Software. The ore production
the Milawa NPV Algorithm in the Whittle software, based targets are never met. In comparison to the production
on an estimated orebody model and an E-type model schedule shown in Fig. 18, the conven-tionally generated
respectively. The risk analysis shows that ore production schedules shown in Figs. 20–22 present the following
targets are never met and that grades are underestimated at issues: they require one extra year to mine the deposit, they
the bottom of the pit resulting in a misleading NPV do not meet ore production targets and the NPV forecasts
forecast as shown in Figs. 20 and 21. Furthermore, are misleading because the extra maintenance and
consider a pushback design obtained through the operational costs incurred through idle mineral processing
methodology presented herein, if a LOM production capacity are not considered. When compared to the
schedule is generated by a conventional
22 Performance of conventional LOM PS based on design with seven pushbacks, shown in Fig. 17, in terms of a ore production, b
waste production, c cumulative metal production and d cumulative discounted cash flows
10. B. Tachefine and F. Soumis: ‘Maximal closure on a graph with 14. R. Dimitrakopoulos and S. Ramazan: ‘Uncertainty based produc-tion
resource constraints’, Comput. Oper. Res., 1997, 24, (10), 981–990. scheduling in open pit mining’, SME Trans., 2004, 316, 106–112.
11. F. Seymour: ‘Pit limit parameterization from modified 3D Lerchs- 15. S. Ramazan and R. Dimitrakopoulos: ‘Recent applications of
Grossman algorithm’, SME Paper, 1994, 11. operations research and efficient MIP formulations in open pit
12. P. J. Ravenscroft: ‘Risk analysis for mine scheduling by conditional mining’, SME Trans., 2004, 316, 73–78.
simulation’, Trans. IMM A, Min. Ind., 1992, 101A, A104–A108. 16. S. Ramazan and R. Dimitrakopoulos: ‘Production scheduling with
13. S. Ramazan and R. Dimitrakopoulos: ‘Stochastic optimization of long uncertain supply: a new solution to the open pit mining problem’,
term production scheduling for open pit mines with a new integer Technical Report No. 1, COSMO Laboratory, 2006, 257–294, McGill
programming formulation’, ‘Orebody modelling and strategic mine University, Montreal.
planning’, Spectrum Series, 2nd edn, Vol. 14, 353– 360; 2007, The 17. J. Whittle: ‘Beyond optimisation in open pit design’, Proc. Canadian
Australasian Institute of Mining and Metallurgy, Melbourne. Conf. on ‘Computer applications in the mineral industries’, 331–337;
1988, Rotterdam, Balkema.